THE CLASSIFICATION OF PURELY INFINITE C*-ALGEBRAS

USING KASPAROV’S THEORY

(PRELIMINARY VERSION: 27.05.2022)

After Eberhard Kirchberg’s death in August, 2022, his wife Sommai agreed to make this
manuscript accessible to the mathematical public. The manuscript contains a wealth of
deep results, ideas, and techniques, but it is not completely finished. The manuscript is
in the form in which we found it on his computer and we did not make any changes.
The plan is that Jamie Gabe and Mikael Rørdam will add an explanatory text and that
the manuscript will be published via the University Library of the University of Münster.
In the meantime it will be available at:
https://ivv5hpp.uni-muenster.de/u/echters/ekneu1.pdf
For the time being, please refer to this manuscript as
Eberhard Kirchberg, The Classification of Purely Infinite C*-Algebras Using Kasparov’s
Theory, manuscript available at https://ivv5hpp.uni-muenster.de/u/echters/ekneu1.pdf

Münster, July 18, 2023

Joachim Cuntz, Siegfried Echterhoff, Jamie Gabe, Mikael Røradam

1
The Classification of Purely Infinite C*-Algebras
Using Kasparov’s Theory
(preliminary version: 27.05.2022)

Eberhard Kirchberg
 Contents

Attempt to give a very short overview (abstract) 9

Chapter 1. Introduction and main results 13

1. The case of simple purely infinite algebras 25
2. Towards a classification of non-simple purely infinite algebras 44
3. Background and some basic ideas in proofs 73
4. On J. Elliott’s Classification Programm and Conjectures 75
5. The way to here. Acknowledgements 84
6. To do: Unify NOTATION in all Chapters! 87

Chapter 2. Basics on purely infinite C *-algebras 89

1. Infinite elements, factorization and transitivity 104
2. Characterizations of simple purely infinite algebras 143
3. Proof of the Dichotomy Theorem E for simple C *-algebras 187
4. On absence of infinitesimal sequences 189
5. Properly infinite elements in non-simple C *-algebras 202
6. Cases: many projections, linear ordered ideal-lattice ... 224
7. Residually antiliminary C *-algebras 232
8. Basics on quasi-traces 264
9. Pure infiniteness versus (2-quasi-) traceless. 268
10. Local stability and properly infinite full projections 287
11. Weak pure infiniteness for non-simple C *-algebras 291
12. Non-simple purely infinite algebras 298
13. Singly generated ideals of coronas 313
14. Point-wise approximately inner maps 318
15. More on pi(n)-algebras 323
16. Characterizations of non-simple strongly p.i. algebras 331
17. Permanence properties of strongly p.i. algebras 339
18. Strongly p.i. crossed products and generalized Toeplitz algebras 343

Chapter 3. Nuclear c.p. maps and operator-convex cones 349

1. Exact C *-algebras and nuclear maps 352
2. On approximate inner nuclear maps 374
3. From m.o.c. cones to morphisms in general position 397
4. M.o.c. cones and Hilbert bi-modules 398
5. Operator-convex cones of c.p. maps (2) 400
6. Approximate decomposition of residually nuclear maps 411
3
4 CONTENTS

7. Operator-convex Cones versus Actions by topological Spaces 429

8. Separation of m.o.c. cones by actions 431
9. Temporary demands from other places 441
10. Non-simple algebras and the WvN-property 442
11. More on strongly purely infinite algebras 454
12. Collection of some topics 461
13. Graded m.o.c. cones 468

Chapter 4. Comparison and Addition of some C *-morphisms 471

1. Two projections in C *-algebras 475
2. On K-theory of properly infinite C *-algebras 485
3. Addition of C *-morphisms 527
4. Groups defined by absorbing C *-morphisms 535
5. Generalizations and limitations 596
6. On K-theory of corona C *-algebras 604
7. Additional remarks and further plans for Chp.4 615

Chapter 5. Generalized Weyl–von Neumann Theorems 621

1. Strict convergence in Multiplier algebras 627
2. Some properties of corona C *-algebras 664
3. Functions that respect quasi-central approximate units 670
4. A “tautologic” Weyl–von Neumann type result 678
5. Ψ-equivariant Stability of extensions 709
6. W-vN type results for weakly nuclear maps 729
7. W-vN type results for simple p.i. algebras 734
8. Extensions and Ext(C; A, B) 738
9. The Ψ-residually nuclear case and Ext(C; A, B) 753
10. The case of purely large extensions 777

Chapter 6. Exact subalgebras of Purely infinite algebras 781

1. Normalizers of hereditary subalgebras (Part 1) 782
2. The Proof of Theorem A 789
3. The residually nuclear case 795

Chapter 7. Groups of asymptotic morphisms R(C; A, B) 819

1. The semigroups SR(C; A, B) 819
2. The asymptotic corona Q(X, A) 824
3. Remarks on the case of group actions 840
4. Collection of needed results. 841
5. Related questions 849

Chapter 8. The isomorphism of KK(C; A, B) and Ext(SC; A, SB) 851

1. Some basics on KK and Ext 854
2. Isomorphism of Ext(C; A, B) and KK(C; A, B(1) ) 862
3. Ext(SC; A, SB) and KK(C; A, B) considered as K1 -groups 886
4. The residually nuclear case 902
 CONTENTS 5

5. Verifying Condition (DC) of the basic Theorem 4.4.6 910

Chapter 9. Scale-invariant maps (i+ii of Thm’s B,M) 917

1. Scale-invariant elements of SR(X; A, B) 919
2. Mapping-cone construction defines isomorphism 926
3. Unsuspended stable E-theory versus Ext-groups 938
4. Proofs of Parts (i) and (ii) of Theorems B and M 951

Chapter 10. Unitarily homotopic algebras and Theorem B(iv) 957

1. Unitary homotopies and Proof of Theorem B(iv) 957
2. Proof of property (III) 959
3. Some corollaries 963

Chapter 11. PI-SUN algebras represent the nuclear KK-classes 969

1. Proof of (α) 977
2. Preliminaries for the proof of (β) 977
3. Proof of (β) 983
4. Examples via Cuntz-Pimsner algebras 987

Chapter 12. Non-commutative Selection and Proof of Thm. K 989

1. Non-commutative asymptotic Selection 990
2. Selections and Ψ-equivariant embedding 1001
3. Interjection: Non-commutative Selection Conjecture 1028
4. Proof of (old !!!) Theorem O 1040

Appendix A. Cuntz equivalence, multiplicative domains, ?related topics? 1045

1. (*) The algebras On , En and O∞ 1046
2. Semi-Projectivity of the En and On 1071
3. On the Jiang-Su algebra Z 1073
4. Fix-point algebras of compact group actions 1074
5. Comparison of elements in C *-algebras 1078
6. Basic properties of Cuntz semigroups 1086
7. Basics on Quasi-traces 1098
8. Projectivity of some C *-algebras 1103
9. Projectivity and c.p.c. order zero maps 1107
10. The “socle” of a C *-algebra. 1121
11. C *-version of Cohen factorization 1123
12. On centralizers in asymptotic coronas 1124
13. Order unit spaces and pre-ordered semigroups 1128
14. Approximate intertwining of inductive limits 1142
15. Intersection with sums of left and right ideals 1145
16. Surjectivity and control of perturbations 1148
17. On perturbation of unitaries 1152
18. Open projections, C *-spaces and Kadison transitivity 1156
19. On non-unital C *-systems 1168
20. Open projections and Kadison transitivity 1179
6 CONTENTS

21. On excision of pure states 1181

22. Some technical functions on [0, 1] 1187
23. K1 -injectivity and generalized Kuiper theorem 1189
24. Tensor Intersection lemma 1193
25. Characterization of closed ideals by inner automorphisms 1194
26. On the Corona Factorization Property (CFP) 1195
27. On dense algebraic Ideals 1204
28. Dimension-Functions in Sense of J.Cuntz 1205

Appendix B. Exact C*-algebras and examples 1207

1. Operations on Multiplier algebras (1) 1208
2. Actions and monotone maps 1209
3. Topological actions 1211
4. Some properties of exact C *-algebras 1213
5. Approximate divisible algebras 1218
6. Positions of C *-subalgebras in O2 1219
7. Quotients of nuclear maps 1222
8. Questions about pure infiniteness 1229
9. On the properties “IR” and “stable rank one” 1234
10. Uniform global Glimm halving 1242
11. Examples that exhaust the UCT-class 1242
12. S.p.i. subalgebras B with Hausdorff Prim(B) 1244
13. Approximate locally liftable maps 1252
14. Reductions to separable subspaces (sep 1) 1253
15. Passage to separable subalgebras (sep 2) 1257
16. Reductions to separable cases (sep 3) 1259
17. Singly generated separable C *-algebras (sep. 4) 1263
18. On asymptotically estimating functions 1266
19. An elementary matrix norm bound 1267
20. On flip-invariant subalgebras of A ⊗ A 1268
21. On ideal structure of crossed products 1270
22. On semi-projective relations 1274
23. Non-2-sub-additive quasi-traces exist. 1274
24. Central sequence algebras, the ε-test and its application. 1281

Appendix C. Temporary (!!) monitor and To-Do List 1287

1. Circulation of Changes 1287
2. Proposed changes (1) 1292
3. List of Changes: labels and names 1301
4. Ref’s and Cite to be fixed — not up to date 1302
5. Temporary references / compare with my way 1302
6. Collection of Definitions of pure infinite C*-algebras 1304
7. On passage to suitable separable (!) C*-subalgebras 1307
8. MORE on infinite elements 1311
9. On ideals generated by n-homogenous elements 1311
 CONTENTS 7

10. On order in A+ for C*-algebras A: 1312

11. On dense (algebraic) ideals of C*-algebras: 1313

Bibliography 1315
 Attempt to give a very short overview (abstract)

First we consider a simple special case and show that KK-equivalent simple
purely infinite separable unital nuclear C *-algebras A and B are isomorphic if
the group isomorphism from K0 (A) onto K0 (B), that are induced by the KK-
equivalence, maps the K0 -class [1A ] of the unit element of A in K0 (A) to the class
[1B ] ∈ K0 (B).
We call them pi-sun algebras, because they are automatically simple by the
requirement of the algebraic property that for each non-zero a ∈ A there are
elements c, d ∈ A with cad = 1A . It yields that pi-sun algebras are isomorphic
if their K∗ -groups are isomorphic by an isomorphism that maps the K0 -classes
of their unit elements into each other and if both are KK-equivalent to Abelian
C *-algebras, i.e., if the Universal Coefficient Theorem applies to both of them.
Here are the first question:
Comment on this. Is KK-equivalence to commutative C *-algebras really needed?
Is each p.i.s.u.n. algebra A isomorphic to A ⊗ O∞ ?
Is A ⊗ O2 isomorphic to O2 if A is a separarable unital nuclear
simple C*-algebra?
Give references for or against it (with cite or refs)
Where are examples or counter-examples?
The considerations are based on the here applicable Cuntz–Kasparov isomor-
phism KK(A, B) ∼ = Ext(A, SB) and the homotopy invariance of KK(A, B), that
will be used to show that the natural group-morphism from the Rørdam groups
R(A, B) – a sort of “unsuspended” and “nuclear” E-theory – into Ext(A, SB) is
an isomorphism. The last step for this proof of the functorial equivalence will be
done in Chapter 8.
M. Rørdam was/sounded not happy with my interpretation
of -- from him defined groups R(A, B).
Perhaps my interpretation is different
and I should call them GK(A, B).
A similar partly classification is given later for non-simple separable stable nu-
clear strongly purely infinite C *-algebras. Those algebras absorb the Cuntz-algebra
O∞ tensorial, i.e., a separable nuclear C *-algebra A is isomorphic to A ⊗ O∞ , if
and only if, A strongly purely infinite (cf. Definition 1.2.2), that is, if A satisfies
a generalized Weyl–von-Neumann–Voiculescu type theorem for weakly residually
nuclear maps (cf. Chapters 5 and 10).

9
10 ATTEMPT TO GIVE A VERY SHORT OVERVIEW (ABSTRACT)

The basic study of properties of tensorial self-absorbing unital separable C *-

algebras will be postponed to Chapter 11 because it has another nature, e.g. it is
not really related to pure infiniteness.
We remove in Chapter 12 by study of some properties of ultrapowers some
technical additional assumptions that we added in the other chapters to make the
some of the proofs, e.g. in Chapter 6, more transparent: We contain a proof that
Prim(A) for separable non-simple C *-algebras A has always a natural property that
we call “Abelian factorization” property 1.2.4. It leads to a rather long study over
the interrelations between general Dini spaces (i.e., second countable locally quasi-
compact point-complete T0 -spaces), “coherent” Dini spaces and second countable
locally compact Hausdorff spaces and is not only related to functional analysis.
The in Chapter 6 made additional assumption – that the considered separable
stable σ-unital nuclear C *-algebras have some property that is (formally) weaker
than “regular” factorization through a l.c. Hausdorff space for their prime ideal
space – can here (in this book) only be reduced in Chapter 12 where we use that
every separable C *-subalgebra of an ultra-power (or more generally in that what
we call “corona algebras”) is contained in a bigger separable C *-algebras that
contains a nice Abelian C *-algebra with the much stronger property that it not
only separates the ideals but is also “regular” in the sense of our Definition 1.2.9.
(or B.4.1 ????)..
This is a much stronger property than regular factorization. And regular fac-
torization is usually stronger than having an Abelian C *-subalgebra that separates
the ideals of a stable separable C *-algebra (with the additional property that it can
be done in a way that the norm functions of its elements exhausts all Dini functions
on Prim(A)).
A special consequence of our classification results is the following:
If A and B are separable stable nuclear C *-algebras and if there is a homeomor-
phism γ from the T0 -space X := Prim(A) of primitive ideals in A onto Prim(B),
then we prove the existence of an isomorphism ψ from A ⊗ O2 onto B ⊗ O2 , such
that ψ induces γ in the sense ψ(J ⊗ O2 ) = γ(J) ⊗ O2 for J ∈ Prim(A). The ψ with
this property is unique up to approximate unitary equivalence.
We generalize Kasparov’s equivariant functor R KKG (X; ., .) (here only in the
case where the group G is trivial) to some kind of generalized “actions” of locally
quasi-compact T0 -spaces X on C *-algebras (cf. Definitions 1.2.6 and 1.2.10). We
denote this functor by KK(X; A, B). In fact, we show that Kasparov’s KK-functor
extends in a natural way to a functor KK(C; A, B) from the category of matrix
operator convex cones (m.o.c. cone) C ⊂ CP(A, B) into the abelian groups. In this
technical approach the algebras itself play then only the role of indices and we have
to study only the category of m.o.c. cones, and then to translate it back to the
category of C *-algebras ...
The main result of this considerations is the following:
Suppose that A1 and A2 are strongly purely infinite, separable, stable and nuclear,
 ATTEMPT TO GIVE A VERY SHORT OVERVIEW (ABSTRACT) 11

and that there is given an homeomorphism ψ from X := Prim(A1 ) onto Prim(A2 )

and that A and B are KK(X; ., .)-equivalent in a ψ-equivariant manner. Then there
exist ψ-equivariant *-monomorphisms h1 : A1 → A2 and h2 : A2 → A1 that induces
the given KK(X, ., .)-equivalence, and that h2 ◦h1 and h1 ◦h2 unitarily homotopic to
the identity maps of A1 respectively A2 . It implies the existence of an isomorphism
γ : A1 → A2 from A1 onto A2 , such that γ is approximately unitary equivalent to
h1 and γ −1 is approximately unitary equivalent to h2 . (But it is at present still not
clear if γ itself can be chosen such that γ is unitarily homotopic to the given h1 .)
On the way of the proof we obtain also several results that are basic for our
main results, e.g. the following:

1. Every exact separable C *-algebra is a C *-subalgebra of O2 (Theorem A).

More generally we show the following “embedding theorem”:
If a (not necessarily simple) C *-algebra D is separable, stable and is “strongly
purely infinite” in the sense of Definition 1.2.2, and if X := Prim(D) acts by a
map ΨA : O(X) → I(A) on a separable stable exact C *-algebra A lower semi-
continuously and upper monotone continuously, then there is a *-monomorphism
h0 : A → D such that h0 ⊕ h0 is unitarily equivalent to h0 and that h0 induces
the given action ΨA . The h0 is unique up to unitary homotopy (Theorem K),
moreover, h0 can be realized by a non-degenerate *-monomorphism, if and only if,
the action ΨA of X on A is non-degenerate.
If D is a non-degenerate stable C *-subalgebra of a C *-algebra B, then B is
stable, D defines an upper semi-continuous action ΨB of X on B, and h0 defines
an element of the semigroup of unitary equivalence classes [Homnuc (X; A, B)] of
ΨA –ΨB –residually nuclear *-homomorphisms from A to B.
We use the above h0 (a generator of some m.o.c. cone C ⊆ CP(A, B) to de-
scribe the nuclear KK-groups KKnuc (X; A, B) as classes of unitarily homotopic
Ψ-residually nuclear homomorphisms modulo stabilization with [h0 ], cf. Theorems
B and M:
[Homnuc (X; A, B)] + [h0 ] ∼
= KKnuc (X; A, B) .

2. The spatial tensor product of two simple C *-algebras is purely infinite if

one of them is not stably finite and the other is non-elementary, cf. Theorem E.
More generally the minimal C *-tensor products A ⊗ B of any C *-algebra A with
a strongly purely infinite C *-algebra B is strongly purely infinite.
3. The UCT holds for all separable nuclear C *-algebras, if and only if, up to
isomorphisms there is only one pi-sun algebra A with K∗ (A) = 0, cf. Corollary J
(1).
??? The question is here: Is every separable nuclear C *-algebra B KK-
equivalent to some pi-sun C *-algebra A. One can first tensor A with O∞ ⊗ K
... ( Seems to be that ??: A ⊗ O∞ ⊗ K is KK–equivalent to A – but here should no
1This is a very important observation, that should direct all the further research to the study

of all pi-sun algebras A with K∗ (A) = 0. Nobody started it yet (January 2022). Why?
12 ATTEMPT TO GIVE A VERY SHORT OVERVIEW (ABSTRACT)

commutative C*-algebra be involved by constructing the KK-equivalence ... The

KK-equivalence of O2 with C0 ((0, 1]) should be directly constructed. Same with
O∞ with C([0, 1]) or with C ... )
If A is stable, i.e., A ∼
= A ⊗ K then one can repeat the constructions given by
using a suitable unital mono-morphism
M(A ⊗ K) 7→ 1M(A) ⊗ L(K) ⊆ M(A ⊗ K) .

This operation will be infinitely repeated and the resulting inductive limit (re-
stricted to the tower of images of A ⊗ c0 (K) should be KK-equivalent to A ⊗ c0 (K)
and the obvious endomorphism given by the forward shift (k1 , k2 , . . .) 7→ (0, k1 , k2 )
on c0 respectively by the natural endomorphism of the above described inductive
limit.
This should be KK-equivalent to A⊗c0 (K), and the natural shift endomorphism
S on c0 should produce the isomorphism S o c0 ∼ = Kp using c0 (K) ∼ = c0 ⊗ K and
∼ ∼
K ⊗ K = K KK-equivalence of A = A ⊗ K with the above defined inductive limit
that is a stable pi-sun C*-algebra.
4. Hausdorff distances of C *-convex hulls of elements a, b and distance between
the generalized Gelfand-transformations of a and b are the same, cf. Corollary
3.10.12 and Remark 3.11.3.
5. Results on ideal spaces and approximate 1-step-innerness of residually nu-
clear maps are given in Chapters 2, 3 and 12.
6. Generalized Weyl–von-Neumann–Voiculescu type theorems and asymptotic
analogs of them are proven in Chapters 5 and 7.
7. The non-negative Dini functions on Prim(A), cf. Definition 12.2.5, are the
generalized Gelfand transforms of elements of A if A is separable and A ∼
= A ⊗ O∞
(Proposition 12.2.6 – a very special case of [447] –).
 CHAPTER 1

Introduction and main results

Following topics should be extracted and published in more detail (also as

beamer presentations, and with lists of its definitions of properties in an elementary
way).
1.) “Squeezing Property” implies K1 -injectivity. (See Chp. 4. for definitions
of this properties! )
Here a question remains:
When K1 -injectivity implies the Squeezing Property?
At least in unital separable nuclear case ?? (Would be nice ! )
(But have no idea until now: Jan 2022 !!!)
2.) Precise version of a quasi-trace state on a type-I C *-algebra that is not a
2-quasi-trace.
See end of Chapter 1 (= Introduction), and the more details in a (still in work)
section of the appendices.
Here following questions remain:
What happens if a separable nuclear C *-algebra has no type-I sub-quotient?
Are then all quasi-traces “trivial” or 2-quasi-traces?
3.) Forgotten!!! Try to remember!
Something important ...? Also related to topics in this book, e.g.:
When (all?) pi-sun algebras A satisfy the UCT?
( Has it to do with the existence/non-existence of ?? Cartan sub-algebras ?? )
Seems to be studied in new papers (since 2016 –2021)!
For example: Is every separable nuclear C*-algebra B KK-equivalent to a
pi-sun C*-algebra?

Not clear what he is Possible ideas? Check it!: B ⊗O∞ and B ⊗K are KK-equivalent to B ? (Seems
thinking here. to be.)
Here K ( – also denoted by K(`2 ) – ) means the algebra of compact operators
on `2 (N).
Then the stable separable nuclear C*-algebra C := B ⊗ O∞ ⊗ K is strongly
purely infinite and absorbs O∞ tensorial. It should be KK-equivalent to B. (Seems
to work!)

13
14 1. INTRODUCTION AND MAIN RESULTS

Notice that L(`2 ) = M(K) is unitally contained in M(C), because M(C)

contains a natural unital copy of
M(B) ⊗ O∞ ⊗ M(K) .
Therefore, there exists an injective non-degenerate endomorphism ρ of C into 1 ⊗
1⊗M(K) ⊂ M(C). Here ”non-degenerate” means that each of the products ρ(C)C
and Cρ(C) are dense in C.
This implies that ρ extends naturally to a unital endomorphism
M(ρ) : M(C) 7→ M(C)
.
The inductive limit with help of iterations of this unital endomorphism M(ρ)
produces a ( huge ) unital C*-algebra E with an ”distinguished” endomorphism
λ : E 7→ E.
Moreover, there is an injective C*-algebra morphism η : C →??? ⊂ E such that
η(C) is an ideal of E with the property η(C) + λ(E) = E
(??? check it ! It is very important)
and η(C) ∩ λ(E) = {0}. ???
Now define inductive D1 := η(C), D2 := η(C) + ρ(C) = C ⊆ M(C),
. . . , D3 := C + ...??
Is it at least the case if B satisfies the UCT? It asks then the weaker question
only:
Is at least every separable nuclear C*-algebra B K-equivalent to a pi-sun alge-
bra (in general)?
Can a positive answer for the special case K∗ (A) = 0 give also a positive answer
for the general case?
E.g. that K∗ (A) = 0 implies A ∼
= O2 ???
Case of K∗ (A) = (Z, 0), as A ∼
= O∞ ???
Are there relations to the QWEP conjecture?
By definition, a C*-algebra B has the QWEP if B is a quotient C*-algebra
∼
B = C/J of a C*-algebra C that has the ”weak expectation property” (WEP).
A C*-algebra C ⊆ L(H) has the property WEP, if and only if, there exists a
completely positive contraction V : L(H) → C ∗∗ into the bi-dual C ∗∗ of C such
that V (c) = c for all c ∈ C.
Perhaps it could follow from (????):
C ∗ (F∞ ) ⊗max L(H) = C ∗ (F∞ ) ⊗min L(H)
where F∞ is the free group on countably many (or more!) generators.
In particular, for every C*-representation d : C → L(K) on a Hilbert space K
there exists completely positive contraction W : L(K) → d(C)00 ⊆ L(K) with the
 1. INTRODUCTION AND MAIN RESULTS 15

property that W (d(c)) = d(c) for all c ∈ C. This can be used also as definition,
because one can take here a sufficiently universal representation d of C.
(So far the answer to the QWEP question is open for exact C*-algebras.)
All nuclear separable C*-algebras B are WEP algebras, because their bi-duals
B are always injective von-Neumann algebras B ∗∗ ∼
∗∗
= N = N 00 ⊆ L(H) (because
L(H) is injective for each Hilbert space H).
The QWEP conjecture and its possible answer is very important, because is
could be that we need some very selective and constructive definition of ultra-filters
and approximation methods to get a coherent picture of the desired results with
sensible methods.
One can reduce this question a bit if every simple, separable, nuclear, unital,
strongly purely infinite C*-algebra A contains a UCT-class member 1A ∈ B ⊆ A
with all other properties the same as A.
This is clearly the case. Are all separable typ I C*-algebras A in the UCT-class? (Seems to be !)

This must be Is the (separable nuclear) UCT-class closed under extensions, tensor-products
known. and inductive limits? (Seems to be! But with which necessary restrictions?)
Also under crossed products – by actions of amenable groups G on A ?
What about crossed products by an endomorphism? I.e., by an injective action
of the natural numbers N
I am not sure what Look to the Rørdam - Property !! ??Which one?
he refers to here.
Preview.
Later add (???) on weak semi-projectivity
of pi-sun (??) algebras and
kernel of η : Aut(A) → Aut(K∗ (A)) in UCT case (important!)
and
classification of tensorial self-absorbing algebras in UCT class

Def. of ‘‘pi-sun’’ is on 3 places -- 2 with motivation

Is ‘‘pi-sun’’ good notation? ??
A long time ago George Elliott [258] did conjecture that simple separable
amenable C *-algebras can be classified by invariants that are, in some sense, K-
theoretic in nature (1).
In fact the status of the conjecture of G. Elliott can now described very roughly
as a classification up to isomorphisms of those simple separable nuclear C *-algebras
A with following properties:
1 A C *-algebra A is “amenable”, if and only if, A is “nuclear”, cf. [340] and [159]. We

rarely use the terminology “amenable” for nuclear C *-algebras, because we work with matrix
operator-convex cones C of nuclear maps, but not directly with the (co-) homological properties
of amenability. But it would be interesting to see if some direct (!) applications of the (co-)
homological definition of amenability give additional new insight for answers to the questions and
problems that we study here, give new insides of open questions ...
 16 1. INTRODUCTION AND MAIN RESULTS

∼ A⊗Z
(a) that absorb the Jiang-Su algebra Z tensorial, in the sense that A =
(see Definition A.3.1 of the Jiang-Su algebra in Section 3) of Appendix A,
and
(b) where A is KK-equivalent to some separable commutative C *-algebra C.

The properties (a,b) are satisfied e.g. for the (Joachim) Cuntz algebra O2 : for (a)
because, O2 absorbs any unital simple nuclear separable C *-algebra B tensorial, i.e.
by isomorphisms O2 ∼ = O2 ⊗ B for any unital simple nuclear separable C *-algebra
B, – especially if we take the Jiang-Su algebra Z as B. The property (b) is satisfied
with C := C0 (0, 1], because K∗ (C0 (0, 1]) = {0} = K∗ (O2 )., – an observation of J.
Cuntz. That will be shown later in this book in any detail as partial result of some
more general observations.
A basic desire is an explicit answer to the following question: Let A a pi-
sun C *-algebra, i.e., a purely infinite separable unital nuclear C *-algebra with the
property that {0} = K∗ (A). Is A isomorphic to O2 ?
(Because all separable exact C *-algebras are C *-subalgebras of O2 , we can
suppose the additional assumption that A is a unital C *-subalgebra of O2 .)
The algebra C := C[0, 1] of continuous functions on the interval [0, 1], and
the unital simple C*-algebras Z (the Jiang-Su algebra) and O∞ (the Cuntz algebra
generated by an infinite sequence s1 , s2 , . . . of isometries with mutually orthogonal
ranges) are all KK-equivalent to the algebra complex numbers C (considered as a
unital C*-algebra).

This is Kirchberg's Give here precise references for the observation that O2 ∼
= B ⊗ O2
fundamental for unital separable simple and nuclear C *-algebras B etc. !!!
theorem. Where is the KK-equivalence to commutative C *-algebras shown??
Give reference!!!
We say in Section 4 of this introduction more about his formulations and the
now (almost) complete state of the proof of the Conjecture of George Elliott. Except
in the until now only partially understood case of stably projection-less algebras
there is a still remaining questions that is equivalent to the below stated question
(Q2) that is equivalent to the question if all separable amenable C *-algebras are
KK-equivalent to Abelian C *-algebras and, to some more general questions that
are “in spirit similar” to our question (Q1), namely what additional structure
has to be imposed to make sure that a given amenable simple separable stably
infinite C *-algebra can be classified by the so far accepted invariants of simple
C *-algebras, among them e.g. the corona factorization property (CFP) and real
rank zero together, or absence of certain types of “infinitesimal” sequences in its
Cuntz-semigroup.
Give ref.s to next??
if invariants of K∗ -theoretic nature (including the pairing with trace cones) together
and with “local approximation” allow to detect if a given separable simple amenable
C *-algebras A tensorial absorbs the Jiang-Su algebra Z. It leads for all separable,
 1. INTRODUCTION AND MAIN RESULTS 17

purely infinite and nuclear C *-algebras A to the important conclusion that A is

isomorphic to A ⊗ O∞ if and only if A is isomorphic to A ⊗ Z.
This ????? second ?????
open question is for the stably infinite amenable simple separable C *-algebras
if they are KK-equivalent to an Abelian C *-algebra C0 (X):
Give a direct construction of a pi-sun C *-algebra that is KK-equivalent to C0 (X)
for a given finite polyhedron X. They should meet transparently in the ?????
in the study of the generators and relations of there K∗ -theories ?????
Is there any example of ?????
Here are some special questions:
Let A denote a separable C*-algebra ...
The History of this circle of questions:
All of the work on this book was mainly motivated and inspired by the original
conjectures of G. Elliott and pioneering ideas of J. Cuntz on this circle of questions.
The “easy” beginning sections of the chapters in this book are concerned with an
alternative proof of the Elliott conjecture but only in the case of simple purely
infinite separable nuclear C*-algebras A in the UCT class. The original definition
of purely infinite (simple) algebras given by J. Cuntz [172, p. 186] – stated there
only in the unital case –, can be reformulated for general unital rings A as: For
each non-zero a ∈ A, there are c, d ∈ A with cad = 1. (See proof of [172, prop. 1.6]
and [169, 1.3, 3.4].) Clearly, this definition implies automatically that A is simple.
But one has also to require then that A not ∼ = C · 1A .
It is easy to see that K∗ (L(`2 )) = 0, because B := L(`2 ) has real rank zero
Mn (B) ∼= B, (1B ⊕p) ∼= 1B for each projection p ∈ B, and for each unitary in u ∈ B
one finds q ∈ B and ξ ∈ C such that |ξ| = 1 and k(1 − q)u(1 − q) + ξ · p − uk < 1/9.
(Moreover the unitary group of B := L(`2 ) is ”globally” contractible to 1, in
the sense that ???? WHAT ????)
The elements are all exponentials u = exp(ih) of self-adjoint operators h∗ =
h ∈ B, because each of them is contained in commutative W*-subalgebra of B.
It follows local contractibility.
(But gives it global contractibility??)
But the Calkin-Algebra Q(`2 ) := L(`2 )/K(`2 ) with `2 := `2 (N) is also a simple
C*-algebra with K-theory, that is K0 (Q) = 0 and K1 (Q) = Z K-theory, because
again Q(`2 ) ∼
= Mn (Q(`2 )) and non-zero projections are equivalent (thus all are sta-
bly equivalent), numbers in K1 (Q) are the ”indices” (e.g. of the Toeplitz operators
on `2 .
Moreover, Q(`2 ) contains infinitely many ... pair-wise non-isomorphic ... sep-
arable simple unital C*-subalgebras with trivial K-theory. (Can they be exact ?
18 1. INTRODUCTION AND MAIN RESULTS

Are there continuously many – or at least countably many – unital, non-nuclear

and simple C*-subalgebras of O2 not pairwise isomorphic? )
If one tensors the Cuntz algebra O2 with any separable simple exact and unital
C*-algebra B, then A := B ⊗ O2 has again trivial K-theory K∗ (A) = 0.
Thus, only in case of pi-sun C*-algebras A (i.e. purely infinite, separable,
unital and nuclear C*-algebras) there is a chance that K∗ (A) = 0 implies that A is
isomorphic to the Cuntz algebra O2 . )
The below given Definition 1.2.1 of purely infinite C *-algebras is equivalent
to this algebraic definition of J. Cuntz if A is a unital and simple C*-algebra,
cf. Proposition 2.2.1: It turns out that a not necessarily simple or unital C *-algebra
A is purely infinite (more precisely: is 1-purely infinite) in the sense of the below
given Definition 1.2.1 if every non-zero element a ∈ A+ is properly infinite, i.e., if
there exists a sequence dn ∈ M2 (A) (depending on a, and the dn can be chosen with
zeros on the second line: (dn )21 = 0 = (dn )22 ) with limn d∗n (a ⊕ 0)dn = a ⊕ a , (2).
This is property pi-1, that is a special case of property pi-n defined in Definition
??
and the, – in a non-trivial manner equivalent – Definition 1.2.1 that is a special
case of 1-purely infinite algebras, (i.e., pi(1)-algebras) of the weakly purely infinite
algebras defined in Definition 2.0.4.
See Proposition 2.12.12 for the general equivalences ??.
The name and definition of nuclear C *-algebras is “inspired” – beside is very
different – by the description of “nuclear” locally convex vector spaces V by unique-
ness of the topology given by any compatible locally convex structure on alge-
braic tensor products V W with other locally convex vector spaces W , given by
A. Grothendieck [336]. The C *-algebra A is nuclear, if and only if, there is only one
C *-norm on the algebraic tensor product A B for every C *-algebra B, cf. [766]
(where it was called “ property T ” in 1964), and [509].
It has been later observed that the C *-algebra A is nuclear if and only if there
exists a net {Vτ }τ ∈T of completely positive maps Vτ : A → A of finite linear rank
that converges in point-norm topology to idA , (see [426] as outlined in our Remark
3.1.2(iii,f) that does not factorize through the work of A. Connes on von Neumann
algebras, – or consult the paper of Choi and Effros [145, thm. 3.1] that applies
results of A. Connes on von Neumann algebras).
A. Connes has shown that separable C *-algebras A are nuclear if they are
amenable in the sense of Banach algebra theory, cf. [159, cor. 2]. Later, U. Haagerup
proved that each nuclear C *-algebra A is amenable in this sense, cf. [340, thm.3.1].
We do never use the amenability of nuclear C *-algebras in our considerations, and
read throughout this book “amenable” as synonym of “nuclear”.

2Here and some other places the notation x ⊕ y denotes the ”diagonal” 2 × 2-matrix [x ]
jk
with entries x11 := x, x12 := 0, x21 := 0 and x22 := y.
 1. INTRODUCTION AND MAIN RESULTS 19

We say that a separable C *-algebra A is in the UCT class, if A satisfies the

conditions of the Universal Coefficient Theorem of Rosenberg and Schochet [699],
cf. [73, thm. 23.1.1]. An observation of G. Skandalis says that the UCT for a
separable C *-algebra A is equivalent to: There exists a locally compact metric
space X such that A is KK-equivalent to C0 (X) ( 3 ).
If a separable C *-algebra A is KK-equivalent to an Abelian C *-algebra, then
A is also KK-equivalent to an algebra C0 (X), with X a locally compact Polish
space that is locally (!) a finite CW-complex of dimension ≤ 3. This is because for
every countably generated Z2 -graded Abelian group G = G0 ⊕ G1 there is a locally
compact Polish space X that is locally an at most 3-dimensional CW-complex with
K∗ (X) ∼ = G for those X, cf. Lemma B.11.1 or proof of [73, cor.23.10.3]. If G
is finitely generated then one can take for X a usual (= global) CW-complex of
dimension ≤ 3.
CHECK (and replace if necessary) ”equivariant” by some accepted expression
...
It seems likely that every separable nuclear C *-algebra C is KK-equivalent to
a pi-sun algebra A. Then only 2 questions remain: Does there exists a separable,
simple, and nuclear C that is not in the UCT class ...
Other question ???
What about K∗ (A) = 0 for general separable nuclear A ?
Can the UCT-class – in a sense – distinguish between O∞ and the Jiang-Su
algebra? ...
The proof of the Elliott Conjecture that is given in this book for the very special
class of purely infinite, simple, nuclear and separable algebras uses some methods
that will be stepwise generalized later to the case of non-simple strongly purely
infinite (s.p.i.) C *-algebras cf. Definition 1.2.2. One of the remaining problems
for the applicability of the later described ideal-system equivariant classification of
separable nuclear s.p.i. C *-algebras has – among others – to do with the question
under which conditional properties a purely infinite nuclear C *-algebras A becomes
strongly purely infinite (which is then in case of separable nuclear A equivalent to
A∼ = A ⊗ O∞ or to A ∼ = A ⊗ Z, where Z denotes the Jiang-Su algebra).
Give reference!! where this is proved here.
Conj.: ?????
(OK!)
If A is separable then Ac := (A0 ∩ Aω )/Ann(A) is always a unital C *-algebra.
Here A ⊆ Aω is natural embedded via a ∈ A 7→ πω (a, a, . . .) into Aω :=
πω (`∞ (A)) for some ultra-filter ω (= non-trivial character of the commutative C *-
algebra `∞ (C) ).
(seems that there are counterexamples for next !!)
3 See [73, thm. 23.10.5], cf. also below reasoning for Corollary J.
20 1. INTRODUCTION AND MAIN RESULTS

If Ac has no character then Ac contains a unital copy of the Jiang-Su algebra

Z and A ∼
= A ⊗ Z ??? ???? It seems one needs much stronger assumptions ...
At least if A is s.p.i., separable and exact, or ?????
The question is:
Let B a separable unital C *-algebra without characters. Does B contain a nuclear
C *-subalgebra D with 1B ∈ D and without any character?

Moreover: Does C := B ⊗max B ⊗max · · · contain a copy of Z ?

Seems not to be the case ? Need stronger assumptions.
But does C contain unitally a copy of a unital nuclear D without characters?
Let D nuclear and unital, but without characters, does D ⊗min D ⊗min · · ·
contain the Jiang-Su algebra Z ?
Seems to have a negative answer ...?
This generalization explores all aspects of the technics that we introduce for
the study of simple nuclear C *-algebras and requires some additional study of the
m.o.c. cones ...
Give reference res.nuc. c.p. maps HERE !! of the later in Definition
????
See the definition of residually nuclear c.p. maps in Definition 3.10.1 defined
“residually nuclear” completely positive maps.
The Definition 3.5.9 – here considered in the special case of the ideal lattices
of the considered algebras.
The residually nuclear maps generalize fiber-wise nuclear C(X)-modular maps
between C(X)-C *-algebras and generalize asymptotic C(X)-algebras. But a suit-
able equivariant sort of the UCT and related families of representatives for suf-
ficiently general UCT classes for the classification of non-simple amenable C *-
algebras is not in sight yet. But in case of separable nuclear C *-algebras A with
non-Hausdorff primitive ideal space Prim(A) of finite decomposition dimension ( 4 )
there could be a chance to find classifying representatives, cf. more in Chapter 8.
This monograph is essentially self-contained and requires only very basis knowl-
edge on C *-algebras, very elementary facts of K∗ -theory and the classical results
of J. Cuntz [169], [172] on his algebras On ( 5 ).
The proofs of Theorems A and B use that every unital separable exact C *-
algebra is a sub-quotient of the CAR-algebra M2∞ (first shown in cf. [437] and

4 Notice that the decomposition dimension and the covering dimension are very different for

non-Hausdorff topological spaces in general!

5 See Sect. 1 of Appendix A. We need only the defining relations and that O and O
2 ∞
are simple, purely infinite, separable, unital and nuclear, with K∗ (O2 ) = 0, K0 (O∞ ) = Z and
K1 (O∞ ) = 0. And that they really are KK-equivalent to the commutative C *-algebras C0 (0, 1]
and C[0, 1]), or the algebra of one point: C in case of O∞ .
 1. INTRODUCTION AND MAIN RESULTS 21

[438]). The ideal-equivariant generalizations in Theorem 6.3.1 is more general

than Theorem A but uses similar ideas in its proof. But the point is that those
ideas have to be generalized before in an ideal system equivariant manner. Its proof
contains implicitly the old result on sub-quotients of M2∞ in [438], that is a kind of
“inverse” of Glimm’s characterization of non-type-I C *-algebras as a very special
case.
We refer the reader to [73], [692] and [816] for the needed basic K-theory, and
to [73] for some of the later used ideas of KK-theory, but we give in Chapter 4 proofs
of those facts in K-theory that will be used for an interpretation of KK-groups as
K∗ -groups of generated by certain relative commutators.
We call a simple, purely infinite, separable, unital and nuclear C *-algebra A a
pi-sun algebra ( 6 ), cf. Chapters 2 and 3 for equivalent definitions and properties,
for example, Part(vi) of Proposition 2.2.1 is another reason for our decision not to
mention the simplicity of A in “pi-sun”.
The class of simple purely infinite, separable, unital and nuclear C *-
algebras (in short: pi-sun algebras) include the simple Cuntz-Krieger algebras
associated to Markov chains [183], [174] or – more generally – some simple graph-
algebras and the Ruelle algebras associated to hyperbolic homeomorphisms of com-
pact spaces [653] as well as some C *-algebras associated to boundary actions of
certain groups and to a class of groupoids [19], [507]. The crossed product C *-
algebra C(X) o Γ defined by the action of a lattice Γ ⊂ G on the Fuerstenberg
boundary X of G is a pi-sun algebra if G is a real connected semi-simple Lie group
G without compact factors and with trivial center, [19]. Sometimes invariants of
pi-sun algebras reflect geometric properties of the underlying dynamical systems,
for example the K-theory groups of the Cuntz-Krieger algebras OA are the Bowen–
Franks invariants of flow equivalence for the related matrix A ([102], [183]). Many,
but not all, Cuntz-Pimsner algebras of Hilbert bi-modules over commutative C *-
algebras, and certain classes of crossed products by “properly outer” actions are
strongly purely infinite (7).
All pi-sun algebras A in the UCT-class are stably isomorphic to (generalized)
Toeplitz C *-algebras of a suitable Hilbert C0 (X, K)-bi-module H where X is a
locally compact Polish space with K∗ (X) = K∗ (A). This is one of the many pos-
sible ways to show that the K-theory invariants are really exhausted by the pi-sun
algebras in the UCT-class up to stable equivalence.
To provide the reader just here with at least one example in each stable iso-
morphism class we give here a description of suitable bi-modules H for given locally
compact X (see Chapter 11 for more details):
6 We do not mention the “simplicity” in the name “pi-sun” because the – very first – original

algebraic definition of pure infiniteness of unital algebras A was: For 0 6= a ∈ A there are b, c ∈ A
with bac = 1. Non-simple p.i. algebras require anyway a much more elaborate classification theory!
7 Results of [359] show that the example of Rørdam [687] of a unital simple finite, but

stably infinite (!), algebra R is a full corner of a Cuntz-Pimsner algebra given by a suitable
Hilbert bi-module H(A, h) over A := C(S2 × S2 × · · · , K) . Thus, R is KK-equivalent to A.
22 1. INTRODUCTION AND MAIN RESULTS

Let D : C0 (X, K) → L(`2 ) = M(K) be a faithful non-degenerate *-representation

that is unitarily equivalent to its infinite repeat δ∞ ◦ D ∼
= D ⊕ D ⊕ · · · . Then H :=
C0 (X, K) becomes a Hilbert C0 (X, K)-bi-module with left action φ of C0 (X, K) on
H given by
φ(f )g := (1 ⊗ D(f ))g for f, g ∈ C0 (X, K) .

The corresponding generalized Toeplitz-Pimsner C *-algebra T (H) is simple, sepa-

rable, stable and nuclear, cf. Corollary 2.18.7 ( 8 ). It is KK-equivalent to C0 (X, K)
(and, hence, to A) by results of Pimsner [633, thm.4.4, cor.3.14]. Using Voiculescu’s
generalization of the Weyl–von-Neumann theorem, cf. [43] – or our further going
generalizations of it in Proposition 5.4.1, one can show that T (H) is moreover purely
infinite (cf. Corollary 2.18.14). Our Theorem B implies that A ⊗ K is isomorphic to
T (H) if A is a pi-sun algebra that is KK-equivalent to C0 (X). Since T (H) is stable
and simple, it is the same as the Cuntz-Pimsner algebra OH . The latter implies
that A ⊗ K is the crossed product B o Z of some type-I algebra B by a suitable
automorphism of B.
In this way, we can see that the possible K∗ -invariants are exhausted by the
above examples, because they are exhausted by the K∗ (X)-groups of locally com-
pact Polish spaces X, cf. Lemma B.11.1. Other constructions of pi-sun algebras
with given K∗ -invariants have been used by Elliott and Rørdam [269]. But all this
examples do not say that it is their only appearance. Usually it is not easy to verify
from some given informations whether an algebra is a (simple) pi-sun algebra in
the UCT class or not. The present status is that there is no general method to
detect from defining relations on generators if the universal C *-algebra defined by
this relations is KK-equivalent to a commutative C *-algebra or not.
See more about the relations between our interpretation of the classification
conjecture and its formulations by G. Elliott in Section 4, cf. also [686, sec. 2.2].
We consider here at first the special case where A is simple, separable, nuclear,
T (A) = {0} and the real rank of A is zero. Here we denote by T + (A) the cone of
+

non-negative semi-finite lower semi-continuous traces τ : (A ⊗ K)+ → [0, ∞].

It turns out that our considerations of the special case of (simple) pi-sun C *-
algebras imply that the general Conjecture of Elliott – here restricted to the par-
ticular case of pi-sun algebras – is equivalent to the below stated questions (Q1)
and (Q2). The notion of simple purely infinite algebras was introduced in [172,
p. 186] and will be discussed in Chapter 2 in detail. We prefer to work with the
below given Definition 1.2.1.
By [269], or by Theorem I below, the possible Elliott invariants are exhausted
by C *-algebras in the class of those simple purely infinite separable nuclear unital
C *-algebras (here called pi-sun algebra) that are KK-equivalent to commutative
C *-algebras, i.e., that are in the ”UCT-class”.

8 Since D is unitarily equivalent to its infinite repeat δ ◦ D, the algebra T (H) is equal to
∞
the Cuntz-Pimsner algebra O(H).
 1. INTRODUCTION AND MAIN RESULTS 23

Summing up we get: A complete proof of the Conjecture of Elliott in the

case of simple separable nuclear C *-algebras A and B of real rank zero and with
T + (A) = T + (B) = 0 would imply in particular that the following two long-standing
open Questions (Q1) and (Q2) could be answered affirmative:

(Q1) Is every simple stably infinite unital nuclear C*-algebra A of real rank
zero also purely infinite?
This means: Is every non-zero projection p ∈ A infinite in A if A has
”real rank zero”, A is simple, is nuclear, is separable, and A contains at
least one infinite projection q 6= 0 ?
(Q2) Is O2 up to isomorphisms the only (simple) pi-sun C*-algebra A with
K∗ (A) = 0 ?

Here O2 denotes the Cuntz algebra on two isometries as generators (i.e. s1 , s2 ∈ O2 ,

with s∗1 s1 = s∗2 s2 = 1 = s1 s∗1 + s2 s∗2 cf. Section 1 of Appendix A or [169].
By Theorem E(ii) below, the Question (Q1) has a positive answer (also if A
has not real rank zero) in the special case where A ∼ = B ⊗C and where the algebras
B and C both are not “elementary”, i.e., both are not isomorphic to the compact
operators K(H) on a Hilbert space H of infinite or finite dimension. Implicitly, our
results imply also that a simple unital separable nuclear C *-algebra A is purely
infinite, if and only if, the relative commutant A0 ∩ Aω of A in its ultrapower Aω is
not stably finite (cf. [448, rem. 2.13] for details), and this is the case, if and only if,
this relative commutant is purely infinite and simple (9). Then, one has A ∼ = O2 , if
0
and only if, 0 = [1] ∈ K0 (A ∩ Aω ) (cf. Corollary G).
But even if it turns out that Question (Q1) has a negative answer, we could
add to the invariants the pre-ordered “local” Cuntz semigroup CS(A), cf. Definition
2.4.3, or [171], [175] for the unital case, as an almost K-theoretic invariant. Then
we can describe the purely infinite algebras in the following way (cf. Corollary
2.4.6):
A simple C *-algebra A is purely infinite – or is isomorphic to K(H) for some
Hilbert space H –, if and only if, the “local” Cuntz semigroup CS(A) of A does not
contain an “infinitesimal sequence” [an ] ∈ CS(A) in the sense of Definition 2.4.4 of
infinitesimal sequences in pre-ordered semi-groups.
Another possible “weak additional assumption” for an answer of question (Q1)
is the assumption that A has the corona factorization property (CFP) – that we
discuss in Section 26 of Appendix A.
If a simple separable C *-algebra A has real rank zero and satisfies the (CFP)
then A is stably finite or is purely infinite.
If B := A ⊗ K(`2 ) contains an infinite projection P , then this projection is
properly infinite by simplicity of B. More generally if P ∈ B is a projection and

9 More generally: If a C *-algebra A has the property that F (A) := (A0 ∩ A )/ Ann(A, A )
ω ω
and A are both separable, then A is simple and nuclear, and then only the three cases can occur:
A = 0, or A and F (A) are both purely infinite, or where A ⊗ K ∼
= K and F (A) ∼= C, cf. [448].
24 1. INTRODUCTION AND MAIN RESULTS

0 6= b = P bP is an infinite element, then b and P are properly infinite in B by

simplicity of B
⇐ Where is this proved?
How the (CFP) should manage that this strong alternative happens?
If one of the (non-zero) projections q is infinite, then q is automatic properly
infinite by simplicity of A.
It is sufficient then to consider qAq because then there exists a unital morphism
of the Cuntz algebra O∞ into qAq. To get that A is purely infinite, one needs that
every non-zero projection 0 6= p ≤ q is infinite (otherwise no projection in pAp is
infinite!).
It needs that each non-zero projection p is infinite, because then p is automatic
properly infinite by simplicity of A. This follows from the general property that
the set of elements a ∈ A with a ⊕ p - p is a closed ideal of A.
But the assumptions give only that all non-zero projections p have the property
that there exists an individual number n(p) ∈ (N ) such that 1n(p) ⊗ p is infinite in
Mn(p) ⊗ A. – It is then there in Mn(p) ⊗ A properly infinite.
Until here we have not used the ... ?????
The nuclearity of A is not considered in all combinations of above listed addi-
tional assumptions to get A purely infinite: There could be some weaker additional
assumption in case of nuclear A, not found or conjectured yet (2022).
The below given Corollary J shows that Question (Q2) is equivalent to the
following important open question (Q2*), and that positive answers to (Q1) and
(Q2) together imply a proof of the Conjecture of Elliott in the case T + (A) = {0}
and amenable separable A with real rank zero .

(Q2*) Does every separable nuclear C*-algebra satisfy the Universal Coefficient
Theorem?

Here the Universal Coefficient Theorem (=: UCT) in the sense of Rosenberg and
Schochet is considered, [699]. We refer the reader to [73, sec. 23]. A separable
C *-algebra satisfies the UCT, if and only if, it is KK-equivalent to a commutative
C *-algebra (cf. [73, thm: 23.10.5], [726, prop. 3.5], [699, cor. 7.5]). It suffices
to consider in question (Q2) simple purely infinite A with K∗ (A) = 0 that are
fixpoint-algebras A of circle actions on O2 , cf. Corollary J.

We use a careful explained and conceptual approach for proofs in case of simple
algebras, that is not the shortest possible way in this special case, but illustrate
more involved generalizations that allow later to generalize some of the results to
non-simple algebras: the results allow to classify all separable stable nuclear C *-
algebras up to tensor products with the Cuntz algebra O∞ by its KK-equivalence
classes for some generalized ideal-system equivariant KK-theory. We consider the
non-simple case and the needed definitions in the second part of this introduction.
 1. THE CASE OF SIMPLE PURELY INFINITE ALGEBRAS 25

In Chapter 12 we give the proofs of lemmas and propositions which are used to
extend the ideas for simple algebras to the non-simple case.
If we consider all from this general viewpoint, then, in Chapters 2 and 3, we
establish the implication { purely infinite ⇒ “strongly” purely infinite } in the
particular case where Prim(A) consists only of one point. In Chapters 5 and 7 we
study the consequences of this implication. Generalizations of results of Chapters 2
and 3 to the case of non-simple algebras are in parts joint works with E. Blanchard
and M. Rørdam. Detailed proofs did appear in separate papers, ([462], [463], [92],
[93]), cf. remarks in Chapters 2 and 3.
Our exposition is completely self-contained and needs only basic knowledge on
C *-algebras, as e.g. functional calculus and some basic observations of KK-theory
from textbooks, e.g. [73]. The use of some ideas from [172] will be explained in
Chapter 4 and Appendix A.
Notice that this very detailed introduction contains all the proofs of the Corol-
laries C, D, F, G, H, J, L and N.
Only the very key Theorems A, B, E, I, K, M and O will be proved in the
chapters of this book, together with some other independent results that could be
initial for some future research.

1. The case of simple purely infinite algebras

We consider first the case of simple C *-algebras. The spatial tensor product
of C *-algebras A and B by A ⊗ B it is the completion of the algebraic tensor
product A B with respect to the minimal C *-norm on it, cf. [766]. C *-algebras
A are called exact if the functor B 7→ A ⊗ B (here with spatial = minimal tensor
product) is an exact functor on the category of C *-algebras (see [428]-[438], [810]
and Chapters 3 and 6).
Obviously every nuclear C *-algebra A is exact, by the definition of nuclearity
via uniqueness of the C *-norm on the algebraic tensor product A B for each
C *-algebra B ([766], [509]).
The following Theorems A, B and E in the case of simple algebras (and The-
orems K and M in the non-simple cases) are the fundamental results of this book.
...

Theorem A. Let A denote a unital separable C*-algebra.

(i) A is exact, if and only if, A is isomorphic to a C*-subalgebra of O2 .

(ii) A is nuclear, if and only if, A is isomorphic to the range of a unital
conditional expectation from O2 onto a C*-subalgebra of O2 .

Clearly Part (i) remains true if A is not unital because A is exact if and only
if its unitization A
e is exact.

The embedding of unital separable exact A into O2 as expressed in Part (i) can
be always taken as unital *-monomorphism from A into O2 , because K∗ (O2 ) = 0
26 1. INTRODUCTION AND MAIN RESULTS

implies that every non-zero projection p 6= 1 of O2 is the range of an isometry,

cf. [172] and Lemma 4.2.6(ii).
A slight modification of the arguments in the proof of Corollary C (below) shows
that Theorem B(i)-(iii) implies that any two unital monomorphisms h, k : A → O2
are unitarily homotopic in the sense of the below given definition. In particular,
they are approximately unitarily equivalent.
We prove Theorem A in Chapter 6. The idea of the proof is the following: Our
Weyl–von-Neumann–Voiculescu type theorem for p.i. algebras (see Chapter 5)
give exact cite for our WvNV-theorem in Chp.5 !!!
implies that sub-quotients of O2 are isomorphic to subalgebras of O2 . We apply
Glimm’s theorem to the anti-liminal algebra O2 , and then we use the characteriza-
tion of separable exact algebras as sub-quotients of the CAR algebra M2∞ , [438].
Certainly, one could also use directly the natural inclusion of M2∞ into O2 . But
then one does not get the additional result that the embedding of unital nuclear A
into O2 can be chosen such that the conditional expectation in Part (ii) of Theorem
A, is an extreme point of the linear contractions of O2 , cf. Remark 6.2.2. It implies
e.g. that there are unital separable nuclear C *-algebras A and unital embeddings
of A into O2 ∼= O2 ⊗ 1 ⊂ O2 ⊗ O2 ∼ = O2 that can not be “conjugate” in O2 ⊗ O2 ⊗ K
by using automorphisms of A ⊗ O2 ⊗ K and O2 ⊗ O2 ⊗ K, though even all unital
C *-monomorphisms of A into O2 are unitarily homotopic (in the sense of Definition
5.0.1 in Chapter 5).
Theorem A will be used to formulate and prove the below stated Theorem B,
which we start to explain now:
Suppose that A is a separable unital exact C *-algebra. Theorem A shows
the existence of unital *-monomorphisms h1 : A → O2 and h2 : O2 ⊗ O2 → O2 ,
because O2 ⊗ O2 is nuclear. We define a unital *-monomorphism hu0 : A → O2 by
hu0 (a) := h2 (h1 (a) ⊗ 1). If B is a unital C *-algebra which contains a (fixed) unital
copy of O2 , then hu0 defines a unital nuclear *-monomorphism from A into B (which
we fix from now on).
We define h0 : A ⊗ K → B ⊗ K by h0 := hu0 ⊗ idK .
Let D, E be C *-algebras and h : D → E, k : D → E C *-morphisms. We
say that h and k are unitarily equivalent if there is a unitary u in the mul-
tiplier algebra M(E) of E such that u∗ k(·)u = h, and we say that h and k
are unitarily homotopic ( 10 ) if there exists a strongly continuous map t 7→
U (t) from R+ = [0, ∞) into the unitary operators in M(E) ⊆ L(E) such that
limt→∞ U (t)∗ k(b)U (t) = h(b) for every b ∈ D. (Note that this does not imply that
h and k are homotopic: unitarily equivalent morphisms are unitarily homotopic,
but unital inner automorphisms are not necessarily homotopic to the identity map.)
If h and k are unitarily homotopic, then h is homotopic to U (0)∗ k(·)U (0) in the
usual sense. But we can conclude homotopy from unitary homotopy in the case

10 Compare with the more general Definition 5.0.1 in the case where h, k : D → M(E).
 1. THE CASE OF SIMPLE PURELY INFINITE ALGEBRAS 27

where the unitary group of M(E) is connected, e.g., if E is stable and σ-unital, if
E is an AF-algebra, or if E is purely infinite and simple and K1 (E) = 0, cf. [172],
e.g. in case E = On , n = 2, 3, . . . , ∞. Under this very special circumstance the
unitary homotopy becomes considerably stronger than homotopy. But homotopy
implies unitary homotopy in some cases, see Corollary D and the remarks below
Corollary L and Theorem M.
Let [Homnuc (D, E)] denotes the classes of unitarily equivalent nuclear *-homo-
morphisms from D into E. If M(E) contains a unital copy C ∗ (s, t) ⊆ E of O2 , then
one can define the Cuntz addition [h] + [k] := [h ⊕O2 k] of unitary equivalence
classes of morphisms h, k ∈ Homnuc (D, E) by:

h ⊕O2 k(d) := sh(d)s∗ + tk(d)t∗ for d ∈ D ,

where s and t denote the canonical generators of the copy of O2 in E. The definition
is independent – up to unitary equivalence – of the choice of the representatives h
and k of the unitary equivalence classes [h] and [k] and of the chosen unital copy
of O2 . Homnuc (D, E) becomes with Cuntz addition a commutative semigroup. All
P ∗
that follows from the observation of Cuntz [172] that ti = usi where u = ti si is
unitary if si and ti for (i = 1, ..., n) are canonical generators of unital copies of On
(i.e., t∗i ti = s∗i si = 1,
P ∗ P ∗
ti ti = si si = 1).
The Cuntz addition coincides with the natural addition of morphisms if E is
stable, i.e., if E ∼
= F ⊗ K for some C *-algebra F , because the “natural” embedding
K ⊕ K ⊂ M2 (K) ∼ = K is defined by a selection of a unital copy of O2 in L(H) =
M(K), and this copy is so natural or artificial as the Cuntz addition is in general:
It becomes “natural” or “well-defined” only after passage to unitary equivalence
classes. It is useful to remind this phenomenon if one seeks for generalizations of
this constructions to the case where one wants to get some sort of “equivariant”
isomorphisms up to conjugacy for actions of groups, quantum-groups, (or specific
sub-systems of ideals) by suitable generalizations of our approach, because then
one can only expect some sort of outer co-cycle conjugacy, sometimes only by 2-co-
cycles.
The natural map Hom(A, B) 3 h 7→ [h − 0] ∈ KK(A, B) is given
by the Kasparov module (h, B, 0), i.e., φ := h, E := B and F := 0.
Or some other equivalent constructions,
as e.g. by the natural isomorphism KK(A, B) ∼ = Ext(SA, B) and the
mapping cone construction.
We compare the semi-group of unitary equivalence classes of nuclear morphisms
in Homnuc (A ⊗ K, B ⊗ K) with nuclear KK-theory and nuclear Ext-theory in Chap-
ters 8 and 9, using Theorem 4.4.6 and other basics from Chapter 4. A natural
semigroup morphism

α : Homnuc (A ⊗ K, B ⊗ K) → Extnuc (A ⊗ K, SB) ∼

= KKnuc (A, B)

is defined as follows:
28 1. INTRODUCTION AND MAIN RESULTS

We define a nuclear *-monomorphism

ψh : A ⊗ K → M(SB ⊗ K)/(SB ⊗ K)

(i.e., a representative of an element of Extnuc (A⊗K, SB)) for a given representative

h : A⊗K → B ⊗K of [h] ∈ Homnuc (A⊗K, B ⊗K). It gives the Busby invariant of an
extension [h−0] in Extnuc (A⊗K, SB), which is just the mapping cone construction.
It can be described with help of the natural monomorphisms and isomorphisms

B ⊗ K ⊂ Q(R+ , B ⊗ K) := Cb ([0, ∞), B ⊗ K)/ C0 ([0, ∞), B ⊗ K)

and ???
Is this the correct definition?

Or have we to take here 0 ⊕ (B ⊗ K) ? ??

Or can we use here that B ⊗ K ⊂ Cb ([0, ∞), B ⊗ K)?
Compare with image of [(h, B, 0)] ∈ KK(A, B)
by the map KK(A, B) → Ext(A, SB) !!!
Or explore here that B ⊗ K ⊂ Cb ([0, ∞), B ⊗ K) ?

(B ⊗ K) ⊕ (B ⊗ K) ⊂ Q(R− , B ⊗ K) ⊕ Q(R+ , B ⊗ K) ∼
= Q(R, B ⊗ K)

and Q(R, B ⊗ K) ⊂ M(SB ⊗ K)/SB ⊗ K , where we use the notation

Q(X, D) := Cb (X, D)/ C0 (X, D)

for a locally compact space X and a C *-algebra D. Then ψh is defined by

ψh (a) := (0, h(a)) ∈ (B ⊗ K) ⊕ (B ⊗ K) ⊂ Q(R, B ⊗ K).

The corresponding extension of A ⊗ K by SB ⊗ K is the well-known mapping cone

Ch of h, see [73, p. 236/237].
We denote by [h − 0] the class of h in Extnuc (A ⊗ K, SB).
Next red nearly a black-out??
One has to add the universal trivial extension
H0 : A ⊗ K → M(SB)/SB and gets [h − 0] = [k − 0]
if and only if
H0 ⊕ h and H0 ⊕ k are unitary equivalent mod SB !!
We have that [h − 0] = [k − 0] in Extnuc (A ⊗ K, SB) if h and k are stably
unitarily equivalent by unitaries in M(C0 (R+ , B ⊗ K))/ C0 (R+ , B ⊗ K), because,
for unital B, every unitary of M(C0 (R+ , B ⊗ K))/ C0 (R+ , B ⊗ K) lifts to a unitary
in M(C0 (R+ , B ⊗ K)), but those unitaries are given by strictly continuous maps
from R+ into the unitaries of M(B ⊗ K) (11).

11 In this particular case the unitary equivalence can be given by an operator-norm contin-

uous path t 7→ exp(ih1 (t)) · . . . · exp(ihn (t)) with continuous maps t 7→ hk (t) = hk (t)∗ ∈ B ⊗ K.
 1. THE CASE OF SIMPLE PURELY INFINITE ALGEBRAS 29

In particular, α([h]) := [h − 0] is well-defined, and the natural isomorphism

Extnuc (A ⊗ K, SB) ∼ = KKnuc (A, B) maps the element [h − 0] into the usual KK-
difference of h and the zero morphism, which is given by the KKnuc -class of the
Kasparov module (B, h, 0).
The key result for the classification of pi-sun algebras is the following Theorem
B, which combines three essentially different “theorems” as Parts (i)+(ii), (iii) and
(iv).
Theorem B. Let A be a separable unital exact C*-algebra and suppose that
B is unital and contains a copy of O2 unitally. Let h0 : A ⊗ K → B ⊗ K the up to
unitary homotopy unique *-monomorphism that is defined by Theorem A.

(i) The map α : [h] 7→ [h − 0] (given by the mapping cone construction by

h) is a semigroup epimorphism from the semigroup of unitary equivalence
classes [h] of nuclear C*-morphisms h ∈ Homnuc (A ⊗ K, B ⊗ K) onto
KKnuc (A, B), and
(ii) Needs to prove first homotopy invariance of R(h0 ; A, B)?
The decomposition argument works now well ! ??
[h − 0] = [k − 0] in KKnuc (A, B) if and only if h ⊕ h0 and k ⊕ h0 are
unitarily homotopic.
(iii) If moreover B is purely infinite and simple, then h ⊕ h0 is unitarily homo-
topic to h for all *-monomorphisms h ∈ Hom(A ⊗ K, B ⊗ K).
(iv) Check again proof of part (iv)!
Does it work for stable D and E ? ??
Let D and E separable C*-algebras and h : D → E, k : E → D *-
monomorphisms such that h ◦ k is unitarily homotopic to idE and k ◦ h is
unitarily homotopic to idD .
Then there exists an isomorphism ϕ from D onto E that is unitarily
homotopic to h.
Can only prove -- so far -- that:
Then there exists an isomorphism ϕ from D onto E that is approxi-
mately unitary equivalent to h.

Notice here that KK(A, B) = KKnuc (A, B) if one of A or B is nuclear.

The proof of Parts (i) and (ii) of Theorem B is given in Chapter 9. We prove
Part (iii) in Chapter 7 and Part (iv) in Chapter 10.
One can use the result of Theorem B to relax the above given very special
definition of the “zero-element” h0 as follows.
A C *-algebra morphism h : A ⊗ K → B ⊗ K is unitarily homotopic to h0 , if and
only if, there exist *-monomorphisms k1 : A⊗K → O2 ⊗K and k2 : O2 ⊗K → B ⊗K
such that
h = k2 k1 ?????? (or only h unitarily homotopic to k2 k1 ??) ??
and B ⊗ K is the closed ideal generated by k2 (O2 ⊗ K). This can be seen from
Theorem B and following general observations (1)-(3) using [172]:
30 1. INTRODUCTION AND MAIN RESULTS

(1) If O2 ∼= C ∗ (s, t) ⊆ B is a given copy of O2 in B with 1B ∈ O2 , then every

(non-zero) C *-morphism h : O2 ⊗ K → B ⊗ K that generates B ⊗ K as the
closed ideal of B ⊗ K generated by the image of h, i.e., the C *-morphism
h satisfies (B ⊗ K)h(O2 ⊗ K)(B ⊗ K) = B ⊗ K , is “unitary homotopic”
in the multiplier algebra M(B ⊗ K) – in the sense of Definition 5.0.1 – to
a C *-morphism k ⊗ idK : O2 ⊗ K → B ⊗ K, where k : O2 → B is a unital
C *-morphism.
This new copy k(O2 ) of O2 can be different from the given copy
∗
C (s, t) and k is not necessarily homotopic in Hom(O2 , B) to the C *-
morphism h : O2 → B defined by h(s) := s and h(t) := t, because it can
happen that the class [k(s)s∗ + k(t)t∗ ] is non-zero in K1 (B) ( 12 ).
(2) The statements (i) and (ii) of Theorem B are independent of a particular
selection of a unital copy of O2 in B.
(3) h0 : A ⊗ K → O2 ⊗ K is unitarily homotopic to k ⊗ idK by Parts (i,ii,iii)
of Theorem B because KK(A, O2 ) = 0, cf. [172].
It requires that Thm. B is correct with ????
But this can be shown also by elementary considerations form
K∗ (O2 ) = 0, the K1 -injectivity of O2 and of D2 := O2 ⊗ O2 ⊗ · · · , if one
uses both of [172] and the Elliott-Rørdam isomorphism O2 ∼ = D2 .

The proof of Parts (i) and (ii) of Theorem B is given in Chapter 9, using
Chapters 7, 8 and ????. It can be outlined as follows:
We extend α to a semigroup morphism from Homnuc (A ⊗ K, Q(R+ , B ⊗ K))
into Extnuc (A ⊗ K, SB) ∼
= KKnuc (A, B) in the obvious way.
(It is formally almost the same definition as for the mapping cone construction!)
We use basic observations in Chapter 4 to recognize that

[h0 ] + Homnuc (A ⊗ K, Q(R+ , B ⊗ K))

is naturally isomorphic to the Grothendieck group R(A, B) of the semi-group of

unitary equivalence classes of morphisms in Homnuc (A ⊗ K, Q(R+ , B ⊗ K)) .
in between: there is a new way to apply directly Theorem 4.4.6,
using Section 5 of Chapter 8!
proof property (CD) first?
It requires some careful technical observations and in essence an independent
proof for homotopy invariance of R(A, B) with respect to the second variable B
In fact is not clear if the older variant does that,
except by N.Ch. Phillips argument,
seems not to work by application of Theorem 4.4.6,
because [k ⊕ H0 ] = [H0 ] in Ext(A, SB)
does (perhaps) not automatically imply that k ⊕ h0
and h0 are stably asymptotically homotopic.

12 This is one of the reasons that we pass to stable algebras

 1. THE CASE OF SIMPLE PURELY INFINITE ALGEBRAS 31

--> This is now at the end of Chp. 8 fixed.

Is it still open if the ‘‘algebraic’’ KK

is really homotopy invariant?.
Needs that evaluations at t ∈ [0, 1] define
isomorphisms KK([0, 1]C; A, C([0, 1], B)) → KK(C; A, B) .
Here, the operator homotopy seems to be enough?
to show that α induces an isomorphism of R(A, B) and Extnuc (A ⊗ K, SB).
This yields, in particular, a proof of Theorem B(ii).
??? Next needs first to show the injectivity of G(h0 ; A, B) → G(H0 ; A, B) to
get the homotopy invariance !!!
The natural isomorphism Extnuc (A⊗K, SB) ∼ = KKnuc (A, B) and the homotopy
invariance of KKnuc (A, B) imply that, for a nuclear asymptotic C *-morphism k
from A ⊗ K into B ⊗ K, i.e., , for k ∈ Homnuc (A ⊗ K, Q(R+ , B ⊗ K)) , h0 ⊕ k is
unitarily equivalent to all its “up/down-scales”, given by homeomorphisms of R+ .
But we show this homotopy invariance directly and use the resulting scaling
invariance much earlier in the way of proving that α induces an isomorphism from
the continuous Rørdam groups R(A, B) onto KKnuc (A, B).
next? Senseless? Correction?
It means that the powerful Theorem B can be used
(and are used here ?) as a reminder of several results that can be deduced from
Theorem B but can be also proved independently by other methods.
The homotopy invariance implies, for

k : A ⊗ K → Q(R+ , B ⊗ K) ,

the existence of h ∈ Homnuc (A ⊗ K, B ⊗ K) that is unitary equivalent to h0 ⊕ k in

Homnuc (A ⊗ K, Q(R+ , B ⊗ K)) .

Does it prove finally Part (i) of Theorem B ???

The absorption result (iii) of Theorem B is an asymptotic version (given in
Chapter 9) of the generalized Weyl–von-Neumann–Voiculescu theorem (cf. Chapter
5), and Part (iv) of Theorem B is a “continuous” version of a result of Elliott,
cf. [256], [678], and Chapter 10.
Notice that our proof of Part (iv) is more general, because it needs only as-
sumptions on D and E that are independent and weaker than all those required for
A and B in Theorem B. In particular, exactness, amenability (nuclearity), stability
or pure infiniteness are all not used.
Theorem B leads to the below given classification results Corollaries C and H.
32 1. INTRODUCTION AND MAIN RESULTS

If A and B are pi-sun algebras, then we find pi-sun algebras C and D, such that
C and D both contain O2 unitally, and A⊗K ∼ = C ⊗K, B ⊗K ∼ = D ⊗K, cf. J. Cuntz
[172]. If E or F is nuclear and separable, we have Hom(E, F ) = Homnuc (E, F )
and KK(E, F ) = KKnuc (E, F ). Thus, by Theorem B, for z ∈ KK(A, B), there is a
*-monomorphism h ∈ Hom(A ⊗ K, B ⊗ K) such that z = [h − 0] and that h(A ⊗ K)
contains a strictly positive element of B ⊗ K.
If moreover z = [h − 0] is a KK-equivalence and w = [k − 0] ∈ KK(B, A) is the
KK-inverse of z then [idA −0] = z ⊗B w = [kh − 0] in KK(A, A) and [idB − 0] =
[hk − 0] in KK(B, B). By Theorem B(ii)+(iii), hk is unitarily homotopic to idB⊗K
and kh is unitarily homotopic to idA⊗K . Depends from validity of OLD B(iv)
By Theorem B(iv), there exists an isomorphism λ from A⊗K onto B ⊗K which
is unitarily homotopic to h.
Let γ : KK(A, B) → Hom(K∗ (A), K∗ (B)) be the natural functor from KK(·, ·)
into Hom(K∗ (·), K∗ (·)) given by γ(z) : [x] ∈ K∗ (A) 7→ [x] ⊗A z ∈ K∗ (B), see [73,
sec. 23.1]. Then γ([h − 0]) = K∗ (h) if h ∈ Hom(A ⊗ K, B ⊗ K), cf. Chapter 8 for
more details.
But [1A ] = [1A ⊗ p11 ] ∈ K0 (A) and K0 (h)([1A ⊗ p11 ]) = [h(1A ⊗ p11 )] ∈
K0 (B), where pjk here denotes the natural basis of n Mn ⊆ K with p∗jk = pkj and
S

pjk p`m = δk,` pjm i.e., p11 is the “upper left corner minimal non-zero projection” in
K. Thus, if z = [h − 0] and γ(z) maps [1A ] to [1B ], then [h(1A ⊗ p11 )] = [1B ⊗ p11 ],
and, by [172], there exist a unitary v ∈ M(B ⊗ K) with v ∗ h(1A ⊗ p11 )v = 1B ⊗ p11 .
b ∈ B 7→ b ⊗ p11 ∈ B ⊗ K is an isomorphism from B onto (1 ⊗ p11 )(B ⊗ K)(1 ⊗ p11 ).
Let µ be the inverse of this isomorphism. Then ϕ(a) := µ(v ∗ h(a ⊗ p11 )v) is a unital
*-monomorphism from A into B with [ϕ − 0] = [h − 0] = z in KK(A, B), because
composition with a 7→ a ⊗ p11 (respectively with µ and v ∗ (·)v) induce the identity
on KK(A, B), cf. [73, chap. VIII].
If h is moreover unitarily homotopic to λ by u(t) ∈ M(B ⊗ K) and λ is an
isomorphism from A ⊗ K onto B ⊗ K, then we find as above a unitary w ∈ M(B ⊗
K) such that w∗ λ(1A ⊗ p11 )w = 1B ⊗ p11 and ψ(a) := µ(w∗ λ(a ⊗ p11 )w) is an
isomorphism from A onto B. t → u1 (t) := v ∗ u(t)w defines a unitary homotopy from
v ∗ h(·)v to w∗ λ(·)w. In particular, limt→∞ u1 (t)∗ (1B ⊗ p11 )u1 (t) = 1B ⊗ p11 . Thus
there exists t0 > 0 such that a small perturbation of b(t) = µ(1B ⊗p11 u1 (t)1B ⊗p11 )
for t > t0 defines a unitary homotopy from ϕ to the isomorphism ψ.
The above arguments prove that Theorem B has the following corollary:

Corollary C. Suppose that A and B are pi-sun algebras.

(i) Every z ∈ KK(A, B) such that γ(z) maps [1A ] ∈ K0 (A) to [1B ] ∈ K0 (B)
is given by a unital C*-morphism ϕ : A → B, such that z = [ϕ − 0].
Also here is to check: z − [ϕ − 0] is perhaps only modulo the kernel
KK(A, B) → Hom(K0 (A), K0 (B)) determined!
(ii) Check proof of C(ii) again! ??
 1. THE CASE OF SIMPLE PURELY INFINITE ALGEBRAS 33

If z is moreover a KK-equivalence and ϕ : A → B is a unital *-

monomorphism with z = [ϕ − 0],
then there are an isomorphism ψ from A onto B and a continuous
map t 7→ u(t) from R+ = [0, ∞) into the unitaries of B, such that
ψ(a) = lim u(t)∗ ϕ(a)u(t) for all a ∈ A .
t→∞

Thus, z = [ψ − 0].

In particular, Aut(A) 3 ϕ 7→ [ϕ − 0] ∈ KK(A, A) maps onto the invertible

elements z ∈ KK(A, A) with γ(z)([1A ]) = [1A ].
Two automorphisms of A have the same image in KK(A, A), if and only if,
they are unitarily homotopic.

The KK-equivalence of unital endomorphisms ϕ and ψ gives

that ψ(1) = ϕ(1) and that
φ0 ⊕ h0 and ψ0 ⊕ h0 on Ast are unitarily homotopic.
Now ‘‘absorption’’ , i.e., φ0 ⊕ h0 unitary homotopic to φ0 .

Corollary C implies the following Corollary D, because KK is homotopy invari-

ant (without assuming UCT ! ), and homotopy equivalences fixes the K0 -class of
the unit element. Here we consider general homotopy (in the usual sense), and not
our above defined unitary homotopy (cf. also Definition 5.0.1).

Corollary D. Homotopy equivalent pi-sun algebras are isomorphic.

Next depends from old version of Cor. C(ii) ??
The isomorphism can be chosen such that it is unitarily homotopic to the C*-
morphism that defines the homotopy equivalence.

One of the problems is:

Let α : A → A and β : A → A such that
α ◦ β and β ◦ α are unitarily homotopic to idA .
Is α unitarily homotopic to an automorphism of A ?
Really? Is that enough?
Notice that here A is “stably infinite” if A ⊗ K contains a full properly infinite
projection. In case of simple A this is equivalent to the existence of a (non-zero)
infinite projection in A ⊗ K. The following theorem shows that finite but not stably
finite simple C *-algebras must be necessarily “tensorial prime”. Its proof is given
in Section 3 of Chapter 2.

Theorem E. Suppose that A and B are simple C*-algebras, that are not
isomorphic to the algebra of compact operators on a Hilbert space.

(i) If A or B is stably infinite, then the spatial tensor product A ⊗ B is purely

infinite (and simple).
(ii) If A ⊗ B is nuclear and is stably infinite, i.e., A ⊗ B ⊗ K contains an
infinite projection, then A ⊗ B is purely infinite.
34 1. INTRODUCTION AND MAIN RESULTS

In particular, A⊗B is purely infinite if B is pi-sun and A is simple, e.g. A⊗On is

pi-sun for n = 2, . . . , ∞ and each simple, separable, unital and nuclear C *-algebra
A ( 13).
Notice here, that the identity map id of A ⊗ O2 ⊗ K is homotopic to id ⊕ id,
because this is true for O2 , cf. [172]. Thus, KK(A ⊗ O2 , A ⊗ O2 ) = 0 and λ : a ∈
O2 7→ 1A ⊗ a ∈ A ⊗ O2 must necessarily define a KK-equivalence z = [λ − 0] ∈
KK(O2 , A ⊗ O2 ). By Theorem E and Corollary C, λ is unitarily homotopic to an
isomorphism from O2 onto A ⊗ O2 . In the same way one can see that for pi-sun A
with KK(A, A) = 0 the only element of KK(O2 , A) = 0 is given by an isomorphism
from O2 onto A. (Here we have used that O2 is pi-sun, cf. [172].)
The map ϕ(a) = a ⊗ 1 is a unital monomorphism from A into A ⊗ O∞ , and
z = [ϕ − 0] is just the Kasparov tensor product of w = [idA − 0] with the KK-
equivalence [ϕ1 − 0] given by ϕ1 : t ∈ C 7→ t · 1 ∈ O∞ . Thus, z is a KK-equivalence
(– or use [73, 17.8.5 and 17.8.6] twice, or see the related arguments used in Chapter
8 for the generalization to KK(X; ·, ·) and KK(C; ·, ·) –). So we get from ϕ(1A ) =
1A ⊗ 1O∞ and from Corollary C that ϕ is unitarily homotopic to an isomorphism
from A onto A ⊗ O∞ . Thus, we obtain the following corollary of Theorems B, E
and Corollary C :

Corollary F. If A is a separable unital nuclear simple C*-algebra, then:

(i) A ⊗ O∞ is a pi-sun algebra.

(ii) A ⊗ O∞ ∼ = A if and only if A is pi-sun.
(iii) A ⊗ O2 ∼ = O2 .
(iv) If A is pi-sun, then KK(A, A) = 0 if and only if A ∼
= O2 .

Perhaps proof more generally:

Give first ref’s to: On is pi-sun!
(ii,1) The C *-algebra A ⊗ D is strongly p.i. in sense of Definition 1.2.2 for every
non-zero C *-algebra A and for every (non-zero) strongly p.i. C *-algebra D.
In particular this applies to D := O∞ .
(ii,2) For every nuclear separable strongly p.i. C *-algebra A there exists an
isomorphism ϕ from A ⊗ O∞ onto A.
(ii,3) If a C *-algebra A is isomorphic to A ⊗ O∞ , then there exists an (other)
isomorphism ϕ from A ⊗ O∞ onto A such that the endomorphism a ∈ A 7→ ϕ(a ⊗
1) ∈ A is unitarily homotopic to the identity map idA on A.
(In particular, the isomorphism ϕ−1 : A → A ⊗ O∞ has the property that
−1
ϕ (J) = J ⊗ O∞ for each closed ideal J of A.)
Way of proof:
It is enough to show:
13 If B = O then one can allow here also A = M , k = 1, 2, . . ., or A = K by [172], cf. refhere
n k
in book appendix A???
 1. THE CASE OF SIMPLE PURELY INFINITE ALGEBRAS 35

(1) A ⊗ D is strongly p.i.

[should be in Chp. 2 ready?]
(2) O∞ is strongly p.i., because pi-sun, and O∞ ⊗ O∞ ∼
= O∞ ,
[should use Chp. 2,4, 11 ???]
(3) Separable nuclear s.p.i. C *-algebra A has property: 1F (A) is properly
infinite in F (A) := A0 ∩ Aω .
[should be first in Chp. 2 ?]
(4) O∞ ⊗ O∞ ⊗ · · · ⊆ F (A) (unitally) implies for separable A that A ∼
= A ⊗ O∞
[Should be first in Chp. 11 complete?
Needs that O∞ ⊗ O∞ has approximately inner flip and that x ∈ O∞ → x ⊗ 1 ∈
O∞ ⊗ O∞ is approximately unitary equivalent to an isomorphism from O∞ onto
O∞ ⊗ O∞ .]
(5) Prove all of (ii,3) for A = O∞ and then for A ⊗ O∞ (– In place of A ∼
=
A ⊗ O∞ ). Then use the isomorphism of A ⊗ O∞ onto A.
M. Rørdam [678] gave a proof for the isomorphism A ∼ = A ⊗ O2 if A is unital,
separable and contains an approximately central sequence of unital copies of O2 .
A modification of an idea of G. Elliott leads us to a more general result from [438,
cor. 2.4(II)] (see also [786]) at the end of Chapter 10:
There is an isomorphism A ∼ = A⊗B⊗B⊗· · · if A contains an approximately central
sequence of unital copies of B in A, and if the C *-monomorphisms b 7→ b ⊗ 1B and
b 7→ 1B ⊗ b are approximately unitarily equivalent in B ⊗ B. Notice that then B is
automatically nuclear and simple by an argument from Effros and Rosenberg [243],
cf. Corollary 10.3.5. In the case of non-unital separable A, one has to require that
B has an approximately inner flip, and that a copy of B is unitally contained in
F (A) := (A0 ∩ Aω )/ Ann(A, Aω ), then one gets A ∼= A ⊗ B if A is stable or contains
an approximate unit consisting of projections. We get the equivalence of B ⊆ F (A)
and of A ∼= A ⊗ B for (general) separable A if B has the additional property that
B is unitally contained in the ultrapower E(B, B)ω of tensorial join algebra

E(B, B) := {f ∈ C([0, 1], B ⊗ B) ; f (0) ∈ B ⊗ 1 , f (1) ∈ B ⊗ B } .

See [448, thm. 4.5(3)]. The property follows from the K1 -injectivity of B, and is
therefore satisfied for all C *-algebras B that are purely infinite or that absorb the
Jiang-Su algebra tensorial by
Rørdam ?????? citation ? ... Def. of K1 -injectivity of B ?
But the latter property is now known for all separable C *-algebras B that
absorb tensorial, – i.e., in sense B ⊗ E ∼
= B –, a tensorial self-absorbing separable
C *-algebra E, because all this E absorb the Jiang-Su algebra Z tensorial by a
result of W. Winter in [831].
36 1. INTRODUCTION AND MAIN RESULTS

Is the K1 -bijectivity not secure by Z absorption from work of

M. Rørdam?
Check Ref’s again:
This property is now established for all tensorial self-absorbing separable E
with approximately inner flip that allow in addition to apply the Künneth Theorem,
e.g. if they are in the UCT-class. This is because W. Winter [831] observed that
all those absorb the Jiang-Su algebra Z tensorial.
The tensorial absorption B ⊗ Z ∼
= B of Z implies K1 -injectivity, and allows to
deduce that U0 (B) = U(B) if B is tensorial self-absorbing, separable and is in the
Künneth Theorem class.
On the other hand, – by Corollary F(iii) –, we have that O2 ∼ = O2 ⊗ O 2 ⊗ · · · .
Thus O2 contains an approximately central sequence of unital copies of O2 . Again
by Corollary F(iii), this together implies the following corollary.
Corollary G. Let A be a simple, separable, unital and nuclear C*-algebra.
Then A ∼ = O2 , if and only if, A contains an approximately central sequence
ψn : O2 → A (of copies of O2 ) unitally.

With other words: A =∼ O2 , if and only if, A is simple, separable, unital and
0
nuclear, and A ∩ Aω contains a copy of O2 unitally.

Recall that the natural group morphism γ : KK(A, B) → Hom(K∗ (A), K∗ (B))
is an epimorphism if A satisfies the UCT. Moreover, if A and B both satisfy the UCT
and γ(z) : K∗ (A) → K∗ (B) is bijective, then z ∈ KK(A, B) is a KK-equivalence,
cf. [73, prop. 23.10.1]. Therefore, we obtain from Corollary C the following.
Corollary H. Suppose that A and B are pi-sun algebras, and that A satisfies
the universal coefficient theorem (UCT) for KK(A, ·).

(i) If σ∗ = (σ0 , σ1 ) : K∗ (A) → K∗ (B) is an element of Hom(K∗ (A), K∗ (B))

such that σ0 ([1A ]) = [1B ] in K(B), then there is a unital *-monomorphism
τ : A → B such that K∗ (τ ) = σ∗ .
(ii) If B also is in the UCT-class and σ∗ is an isomorphism from K∗ (A) onto
K∗ (B) and τ : A → B is a unital *-morphism with K∗ (τ ) = σ∗ , then there
is a *-isomorphism ψ : A → B from A onto B
NEXT has to be checked again: ??
such that ψ is unitarily homotopic to τ .

Check proof again!

At least ψ and τ are approximately unitary equivalent!?
In particular, K∗ (ψ) = σ∗ .
(iii) If Ext1 (K∗ (A), K∗ (B)) = 0 , then all unital *-monomorphisms τ : A → B
with K∗ (τ ) = σ∗ are unitarily homotopic.

We list here some immediate consequences of Corollary H:

O∞ ∼
= O∞ ⊗ O∞ ⊗ · · ·
 1. THE CASE OF SIMPLE PURELY INFINITE ALGEBRAS 37

by an isomorphism that is unitarily homotopic to

τ : a ∈ O∞ 7→ a ⊗ 1 ⊗ 1 ⊗ · · · ∈ O∞ ⊗ O∞ ⊗ · · · ,

because both algebras are pi-sun with UCT and K∗ -groups Z[1] and 0 (i.e., here
K∗ (τ ) is a K∗ -isomorphism and Ext1 = 0).
But we outline also alternative
proofs of the latter conclusions, by proving first
the homotopy invariance of a general version of Rørdam groups.
In the beginning we gave explicit examples of pi-sun algebras A in the UCT-
class with prescribed K∗ (A) using Cuntz-Pimsner algebras, and we give some al-
ternative construction further below. It follows that Corollary H(i) implies:
(n)
If B is pi-sun and G∗ ⊆ K∗ (B) is an increasing sequence of finitely generated
(n) S (n)
graded subgroups with [1B ] ∈ G0 and n G∗ = K∗ (B) , then there is a sequence
(An ) of pi-sun algebras in the UCT-class, unital *-morphisms τn : An → An+1 and
(n)
a unital *-morphism η : A → B for A := indlimn An with K∗ (An ) = G∗ , K∗ (τn )
(n) (n+1) (n)
the inclusion G∗ ,→ G∗ and K∗ (η ◦ τn ) is the inclusion G∗ ,→ K∗ (B) In
particular, K∗ (η) is an isomorphism from K∗ (A) onto K∗ (B).
Since A is in the UCT-class, Corollary H(ii) now implies:

Next works also if Corollary H(ii)

has to be corrected to weaker version,
where KK-equivalences are not all expressible by isomorphisms.
If B is in the UCT-class, then the above *-morphism η : A → B can be chosen
as an isomorphism from A onto B.
We denote by Aut(K∗ (A), [1]) the automorphisms of the Z2 -graded Abelian
group K∗ (A) that fix the element [1A ] ∈ K0 (A).
Then we get from Corollary H:
Suppose that A is pi-sun and satisfies the UCT.
This is partly still in question ?? Or oK?
The composition γ ◦ α of the map α given in Theorem B(i) with the natural
map γ : KK(A, A) → End(K∗ (A)) defines an epimorphism ???????

γ ◦ α : Aut(A) → Aut(K∗ (A), [1]) .

If Ext1 (K∗ (A), K∗ (A)) = 0, then the kernel of this epimorphism ????????
Must check how the kernel of

KK(·, ·) → Hom∗ (K∗ (·), K∗ (·))

looks like?
is equal to the set of those automorphisms of A that are unitarily homotopic to
id of A.
38 1. INTRODUCTION AND MAIN RESULTS

Every pi-sun algebra A in the UCT-class is anti-isomorphic to itself. Moreover,

if [1A ] = 0 in K0 (A), then there is a kind of “co-multiplication”-like unital mono-
morphism A ⊗ A → A. It comes from a – not necessarily unital – commutative
algebra C ⊂ A that is KK-equivalent to A (via this embedding), and from the
diagonal map C b 3 χ 7→ (χ, χ) ∈ Cb × C.
b It is co-commutative and co-associative.

A first application of the above general results is the following:

Let Am be finite direct sums of algebras of form C(X, Mk (On )), where X are
compact metric spaces and k = 1, 2, . . ., n = 2, 3, . . .. Let hm : Am → Am+1 be
unital C *-morphisms and B := indlim(hm : Am → Am+1 ). It is easy to see that
B ∼= B ⊗ O∞ , because above we have seen that An ∼ = An ⊗ O∞ and O∞ ∼ =
O∞ ⊗ O∞ ⊗ · · · . In particular B is purely infinite if B is simple (see Corollary
3.2.23 for a more general result). The algebras An and therefore B are in the UCT
class. Thus Corollary H classifies the simple one’s of those inductive limits up to
isomorphism.
To see which invariants are exhausted in given classes of metric spaces X and
values k, m, we need to calculate the K∗ -invariants, because the corresponding
realizations by C *-morphisms can be obtained from Theorem B using the UCT
( 14 ). The below described construction (before Theorem I stated below), gives a
simpler way to produce pi-sun examples representing given KK-classes of nuclear
separable C *-algebras. This construction can be generalized to the category of
C(X)-algebras (see end of Chapter 11).

An separable C *-algebra A satisfies the UCT if and only if, A is in the “boot-
strap category” or “bootstrap class” N (see [699] or [73, def. 22.3.4] for a def-
inition), if and only if, A is KK-equivalent to a commutative C *-algebra by an
observation of G. Skandalis [724].
G. Elliott and M. Rørdam have shown that, given any pair of countable com-
mutative groups G0 and G1 and a distinguished element g0 ∈ G0 , there exists a
simple purely infinite separable unital nuclear C *-algebra A in the (little) bootstrap
class N` such that (K0 (A), [1A ]) ∼
= (G0 , g0 ) and K1 (A) ∼
= G1 , cf. [269, thm. 5.6],
and [679, thms. 3.1, 3.6]. Our Corollary H says that such A is unique if we consider
simple pi-sun algebras in the UCT class. In particular, simple pi-sun algebras in
the bootstrap class N are automatically in “small” bootstrap class N` .
A more general result by a rather simple method is given by the following
Theorem I if we combine this with [73, cor.23.10.3] or Lemma B.11.1.

14 The K -groups and its group homomorphisms for certain “elementary” CW-complexes X
∗
can be calculated in good cases with help of the Künneth theorem, 6-term exact sequence, Bott pe-
riodicity and Mayer-Vietoris sequence. The calculation of K∗ (X) for a arbitrary CW complex X is
not straight-forward, except one knows explicitly the involved elements of Ext(K∗+1 (Xn ), K∗ (Rn ))
and of Hom(K∗ (Xn ), K∗ (Rn )) that correspond to attaching n-cells to the (n−1)-dimensional CW-
subcomplex Xn , . . . etc.
 1. THE CASE OF SIMPLE PURELY INFINITE ALGEBRAS 39

We describe now an almost functorial construction of simple separable exact

purely infinite algebras P (A), such that P (A) is KK-equivalent to a given separable
exact C *-algebra A
Check: Now on new places?!
Check again the exact case! Check references!!! (15).
Let A be a separable exact C *-algebra and let O∞ st := (1−s1 s∗1 )O∞ (1−s1 s∗1 ),
where s1 , s2 , . . . are the canonical generators of O∞ . Then O∞ st contains a copy
of O2 unitally, cf. [172] or Lemma 4.2.6(iii). We let At := A ⊗ O∞ st if A is unital,
and define At as the natural unital split extension of O2 ∼ = 1M(A) ⊗ O2 by A ⊗ O∞ st
if A is not unital, i.e., we have then a unital split-exact sequence

0 → A ⊗ O∞ st → At → 1 ⊗ O2 → 0 .
e ⊗ O∞ st and contains
The algebra At is a unital separable exact C *-subalgebra of A
a copy of O2 unitally. The injective C *-morphism η0 : a ∈ A 7→ (a ⊗ (s2 s∗2 )) ∈ At
defines a KK-equivalence of A with At . Clearly, the algebra At is nuclear if and
only if A is nuclear.
The Theorem A provides a unital *-monomorphism hu0 : At → O2 ⊂ At , e.g. as
described before Theorem B, and then define a unital *-monomorphism for a ∈ At
with help of isometries t1 , t2 ∈ O2 ⊂ O∞ st ⊆ At by

h(a) := t1 a t∗1 + t2 hu0 (a) t∗2 ,

that is here expressed using the Cuntz-addition – discussed in Chapter 4 ( 16 ) – of

the identity map and h0

idAt ⊕t1 ,t2 hu0 : At → At .

Now simply let P (A) := indlim(h : At → At ) the “stationary” inductive limit

with natural defining monomorphisms h∞ ∞ ∞
n : At → P (A) given by hn+1 h = hn , and
let η(a) = h∞
1 (η0 (a)) ∈ P (A) the associated natural unital embedding from At into
P (A).
The author has outlined the proof of the following easy theorem at the
Check again: ??
Conference on Operator Algebras and Quantum Field Theory (in Rome, Italy)
July 1996.

Theorem I. Let A be a separable exact C*-algebra. Then:

(i) The algebra P (A) is a simple, separable and exact C*-algebra that con-
tains a copy of O2 unitally.
(ii) P (A) ∼
= P (A) ⊗ O∞ . In particular, P (A) is purely infinite.

15 See [810] or Chapters 3 and 6 on the here needed definition and basic properties of exact

C *-algebras.
16 The “Cuntz sum” defined in Chapter 4 is the base of our there discussed version of KK-

groups that is “constructive” in some sense.

40 1. INTRODUCTION AND MAIN RESULTS

(iii) The monomorphism η : A → P (A) defines a KK-equivalence [η − 0] be-

tween A and P (A).
(iv) There is a conditional expectation from the hereditary C*-subalgebra D ⊆
P (A), that is generated by η(A), onto η(A) ∼= A.
(v) P (A) is nuclear, if and only if, A is nuclear.

A proof of a more general result for continuous fields is given in Chapter

11. We surround the discontinuity of the functor KK(X ; P (A), ( · )) with respect
to inductive limits, and use the continuity of KK(X; ( · ), P (A)) in the sense of
[73, thms. 21.5.2, 19.7.1] and some considerations of E-theoretic nature, the X-
equivariant continuous version of the
??? Effros-Rørdam ??? or Elliott-Rørdam ??
groups considered in Chapter 7.
Check next red and over-next again!
check correct refs to Chapter 11
Are so many footnotes necessary? ??
In fact, we prove in Chapter 11 a more general result for continuous fields
(Ax )x∈X of exact C *-algebras Ax on a compact metric space X with the additional
property that the C *-algebra A of continuous sections is exact ( 17 ):
We take again := A ⊗ O∞ st if A is unital, and define At as the natural unital split
extension of
C(X, O2 ) ∼
= [C(X)1M(A) ] ⊗ O2
by A ⊗ O∞ st if A is not unital.
Not clear if the C(X)-algebra At is again continuous
and has unital fibers that are the minimal unitizations of fibers of
A
Check construction and its continuity.
Compare with fibre-wise construction!
Then we can use the sub-triviality theorem of E. Blanchard [89] ( 18 ), to get
a unital C(X)-module *-monomorphism hu0 : A → C(X, O2 ) ⊆ A . We define again
the Cuntz sum h := idAt ⊕hu0 and let then P(A) := indlim(h : At → At ) – the
stationary inductive limit. All the operations are compatible with the above defined
(fiber-wise) constructions of (Ax )t and P (Ax ) for the fibers Ax of (Ax )x∈X . It
turns out that (P (Ax ))x∈X is a continuous field of C *-algebras, and that P(A) is
the algebra of continuous sections of this field. By construction, P(A) is exact,
and is C(X)-module isomorphic to P(A) ⊗ O∞ . Further, the *-monomorphisms
ηx : Ax → P (Ax ) define a C(X)-module monomorphism η from A into P(A).

17 By [471], this is equivalent to exactness of all A and the continuity of the field (A ⊗
x x
B)x∈X for each separable C *-algebra B. It follows e.g. from nuclearity of the fibers Ax and the
continuity of the field.
18 Theorem A implies a proof of the sub-triviality. The sub-triviality itself is a special case

of Theorem K below.
 1. THE CASE OF SIMPLE PURELY INFINITE ALGEBRAS 41

We show in Chapter 11 a more general result:

The monomorphism η defines a KK(X; ·, ·)-equivalence between A and P(A) . ( 19 ).
At the end of Chapter 11 we outline a different construction that leads to
KK(X; ·, ·)-equivalent exact (respectively nuclear) stable algebras P (A) with prim-
itive ideal space isomorphic to a T0 space X that acts monotone-continuously on
A (cf. the definitions below and Theorems 11.4.1 and O):
One starts with a suitable “universal” Hilbert bimodule and builds the correspond-
ing Cuntz-Pimsner algebra. The universality of the construction has the advantage
that G-actions of locally compact groups G on (Ax )x∈X lead in a natural way to
G-actions on (P (Ax ) ⊗ K)x∈X ( 20 ).
Furthermore, we modify in Chapter 11 the above described (non-universal !)
construction of A ⇒ P (A) in a way that we get (cf. Theorem 11.0.7):
Suppose that A is a unital separable nuclear C*-algebra, G is a discrete countable
exact group, and that α : G → Aut(A) is group-morphism. Then there exist a
pi-sun algebra B, a group-morphism β : G → Aut(B) and a G-equivariant unital
monomorphism η : A → B that defines a KKG -equivalence [η] ∈ KKG (A, B).
The latter construction plays nicely together with permutation actions of G on
multiple tensor products.
Next is wrong! ??
P∞ is isomorphic to the crossed product (O∞ st ⊗ O∞ st ) o Z2 by the flip auto-
morphism a ⊗ b 7→ b ⊗ a on O∞ st ⊗ O∞ st .
Here P∞ denotes the unique pi-sun algebra in the UCT-class with K0 (P∞ ) =
{0} and K1 (P∞ ) = Z .
(But it is likely that the Z2 crossed product by the flip on P∞ ⊗P∞ is isomorphic
to O∞ st .)
The flip automorphism is approximately inner on O∞ st ⊗ O∞ st , because the
flip on O∞ ⊗ O∞ is approximately inner:
(Un ) a sequence of unitaries that approximately implement the flip.
Then [Un , (1 − s1 s∗1 ) ⊗ (1 − s1 s∗1 )] → 0 in O∞ ⊗ O∞ . Small perturbations of
the Un produces unitaries Vn that commute with (1 − s1 s∗1 ) ⊗ (1 − s1 s∗1 ) .
The Wn := Vn · ((1 − s1 s∗1 ) ⊗ (1 − s1 s∗1 )) are unitaries in O∞ st ⊗ O∞ st that
approximately implement the flip on O∞ st ⊗ O∞ st .

Let us now consider an other application of Theorem I :

If A is a separable nuclear C *-algebra, then there is a pi-sun algebra B = P (A)
which is KK-equivalent to A and contains a unital copy of O2 .

19 The bi-functor KK(X; A, B) is, for compact metric spaces X, the same as the Kasparov

functor R KKG (X; ·, ·), with G = the trivial group.

20 At least in the case where the fibers A are nuclear, because then, by Theorem M(iii),
x
there are C(X)-module isomorphisms of P(A) ⊗ K onto any other p.i. stable separable nuclear
C(X)-algebra B that is KK(X; ·, ·)-equivalent to A and has simple fibers Bx .
42 1. INTRODUCTION AND MAIN RESULTS

∼ K∗ (A). We can find a separable commutative C *-algebra

In particular K∗ (B) =
∼
D with K∗ (D) = K∗ (A), cf. [73, cor. 23.10.3], [715] or Lemma B.11.1.
We replace D by a KK-equivalent pi-sun algebra E, which contains a unital
copy of O2 . Then E satisfies the UCT and there is an isomorphism λ from K∗ (E)
onto K∗ (B). By the UCT for E, there exists z ∈ KK(E, B) with γ(z) = λ, and
by Corollary C there exists a unital monomorphism ϕ from E into B such that
z = [ϕ − 0], and thus λ = K∗ (ϕ). But then

λ : K∗ (E) → K(∗−1)mod2 (SB) = K∗ (B)

is the connecting map for the six term exact sequence for the mapping cone Cϕ
and therefore K∗ (Cϕ ) = 0. Cϕ is nuclear because Cϕ is an extension of nuclear
C *-algebras.
If KK(Cϕ , Cϕ ) = 0, then KK(Cϕ , F ) = 0 for every (σ-unital) C *-algebra F .
By the six term exact sequence for KK one gets that B and E are KK-equivalent,
i.e., that A is KK-equivalent to a commutative C *-algebra.
Hence, if the nuclear separable C*-algebra A is not KK-equivalent to a com-
mutative C*-algebra, then KK(Cϕ , Cϕ ) is not trivial, and the pi-sun algebra N :=
P (Cϕ ) satisfies K∗ (N ) = 0 but KK(N, N ) is non-zero.
We can go a bit further to a more special class of algebras that is sufficient to
decide the UCT-problem by following observations (21):
There is a circle action on O2 with simple and purely infinite fix-point algebra B0
in the UCT-class with K0 (B0 ) = Z2 , K1 (B0 ) = 0 and 0 = [1] ∈ K0 (B0 ).
Indeed: Take B0 := P (C ⊕ C) in Theorem I. It is a pi-sun algebra that is
KK-equivalent to C ⊕ C and contains a copy of O2 unitally. By Corollary F(ii),
B0 ∼
= B0 ⊗ O∞ by an isomorphism that is unitarily homotopic to b 7→ b ⊗ 1, and,
– by Corollary H(ii) –, there is an automorphism ψ of B0 such that

(i) ψ0 (m, n) = (n, n + m) for (n, m) ∈ Z2 ∼ = K0 (B0 ) (hence id −ψ0 has

determinant = −1),
(ii) ψ n is strictly outer (in the sense of [578]) for every n 6= 0, because ψ0n 6= 0
for all n 6= 0.
(iii) B0 oψ Z ∼ = (B0 oψ Z) ⊗ O∞

In part (iii) we could also use that B0 ∼ = B0 ⊗ O∞ and then change ψ to idO∞ ⊗ψ
to omit the proof of pure infiniteness for B0 oψ Z. One can see, e.g. by [578] or
by Proposition 2.18.1, that C := B0 oψ Z is simple, because B0 is simple and ψ
satisfies (ii). It is known that C is nuclear, because B0 is nuclear and Z is amenable,
and that C is in the UCT-class because B0 is in the UCT-class (cf. [73, 22.3.5g]).
It follows that C is a pi-sun algebra in the UCT-class with K∗ (C) = 0, by Corollary
F(ii) and by the Pimsner-Voiculescu sequence [73, thm.10.2.1]. Then B0 is the
fixpoint algebra of the (dual) R/Z-action on C, and C ∼ = O2 by Corollary H(ii).

21 Build the tensor product of the example [73, 23.15.12] of Blackadar with O
∞ to get a
version of N .
 1. THE CASE OF SIMPLE PURELY INFINITE ALGEBRAS 43

Now we consider any separable nuclear C *-algebra D and let B0 := P (C ⊕ C)

and A := P (D)⊗B0 . The natural embedding η of D⊕D ∼ = D⊗(C⊕C) into P (D)⊗
P (C ⊕ C) defines a KK-equivalence η : (D ⊕ D) → A, because KK-equivalences are
compatible with tensor products if one of the factors are nuclear, cf. [389, lem.
2.1.27]. The algebra A is the fixpoint algebra of a circle action on P (D) ⊗ C ∼
= O2 ,
by Corollary F(iii).
Together with Corollary F(iv) and [73, cor.23.10.8] the above reductions (to
the case of pi-sun A with K∗ (A) = 0 and to the fix-point algebra of a circle action)
on O2 prove the following.

Corollary J. For every separable nuclear C*-algebra D there is a pi-sun

algebra A with [1A ] = 0 in K0 (A) and a *-monomorphism η : D ⊕ D ,→ A such that

(i) [η] is a KK-equivalence of D ⊕ D and A .

(ii) A is a fix-point algebra of a circle action on O2 .

In particular, every separable nuclear C*-algebra is KK-equivalent to a commutative

C*-algebra, if and only if, every pi-sun algebra A with K∗ (A) = 0, that is isomorphic
to a fixpoint algebra of a circle action on O2 , is also isomorphic to O2 itself.

Thus, Questions (Q2) and (Q2*) have the same answer, and we can limit (Q2)
to the case where A is the fix-point algebra of a circle action on O2 .
revise/check next text blue/red text
One can even step a bit further: If we consider P (Aop ⊕ A) ⊗ B0 in place of A,
then we can see by the above arguments that the possible existence of a separable
and nuclear example B that does not satisfy the UCT implies the existence of
pi-sun A with K∗ (A) = 0 that is anti-isomorphic to itself, that does not satisfy the
UCT, and A is the fix-point algebra of a circle action on O2 . The reason is that
Thus, the non-existence of an anti-automorphism does not necessarily produce a
counterexample for the UCT in the class of amenable separable C *-algebras.
If one considers the class of arbitrary (not necessary simple) separable nuclear
stable C *-algebras that have trivial KK-theory (e.g. contractible algebras) then
every separable nuclear C *-algebra is KK-equivalent to a fixed point algebra of a
Z2 -action of one of them. But one has to consider also inner actions, e.g. given by
conjugation by a symmetry in the multiplier algebra.
On the other hand one can say: If the fix-point algebra of the flip automorphism
Π(a ⊗ b) := b ⊗ a on O2 ⊗ O2 (respectively on O∞ st ⊗ O∞ st ) is isomorphic to O2
(respectively to O∞ st ) then the Cuntz standard form Ast is isomorphic to the fixed
point-algebra of a Z2 -action on O2 .
The reason is, that this conjectures together imply the conjecture that the fix-
point algebra of of the flip on P∞ ⊗ P∞ ∼= O∞ st is isomorphic to O2 . The flip is an
order-2 isomorphism of O∞ st that changes the signs of elements in Z ∼ = K(O∞ ).
Similar arguments show that the Künneth Theorem on tensor products holds
for all separable nuclear C *-algebras, if and only if, K∗ (A ⊗ A) = 0 for all pi-sun
44 1. INTRODUCTION AND MAIN RESULTS

A with K∗ (A) = 0 that is the fix-point algebra of a circle action on O2 and is

anti-isomorphic to itself (22).
If one has such a C *-algebra A, then one can consider C := P (A ⊗ O∞ ) and
ask if ?????? C := O∞ st .
In particular, let D := pCp for a projection p ∈ C with 1 = [p] ∈ K0 (C) ∼
= Z.
Is D tensorial self-absorbing ?
Here we say that a unital separable D is a tensorial self-absorbing algebra
if D ∼
6 C and the unital monomorphism D 3 d 7→ d ⊗ 1 ∈ D ⊗ D is approximately
=
unitarily equivalent to an isomorphism from D onto D⊗D (see [448, sec.4.def.4.1]).

We mention this, because it shows that the UCT-problem and related prob-
lems have connections with the question of the classification of the tensorial self-
absorbing algebras A – without assuming before that A is in the UCT-class !
A special case of the UCT problem is the following interesting and long standing
open question:
Let A a separable unital tensorial self-absorbing algebra that has the property that
Mn is not unitally contained in A for every n > 1. Is A ∼ = O∞ or A ∼ = Z? Here
Z denotes the Jiang-Su algebra. The answer is positive in the UCT-class. But so
far one has no idea what happens without UCT.
Is Mn also not unitally (!) contained in A ⊗ O∞ ? There exist no projection in
A ⊗ O∞ with n[p] = [1] in K0 (A ⊗ O∞ ) ∼= K0 (A) ...

2. Towards a classification of non-simple purely infinite algebras

In the period from 1995-2015 the author has studied methods that allow to gen-
eralize the classification of pi-sun algebras to a classification of non-simple nuclear
strongly purely infinite stable separable C *-algebras.
Compare the 2 below given blue discussions of [442, lem.3.8]
A report on results and outline of the proofs has been published in [442].
But the important “abstract” Lemma [442, lem.3.8] was not proved there and
the author has to regret for his claim “It is a straight-forward proof” (Es ist ein
Gradeaus-Beweis). Indeed, the author had overseen that the list of conditions in
the technical Lemma [442, lem. 3.8] is incomplete:
The surjectivity of the there considered natural group morphism of in Chapter 4
defined groups G(h0 ; A, E) to G(H0 ; A, E) is easy to derive from the listed prop-
erties, but the there proposed injectivity is unclear. Therefore we have revised the
Lemma [442, lem. 3.8] in a very detailed technical Chapter 4 and have reformu-
lated it as Theorem 4.4.6. It is in a sense a general tool for comparing sorts of
unsuspended but stable E-theories with naturally corresponding KK-theories. This
Theorem lists additional conditions, each of them equivalent to the injectivity of

22 It suffices to check K (A ⊗ A) = 0, because one can replace A by P (A ⊕ (A ⊗ P )) ⊗ B .

0 ∞ 0
 2. TOWARDS A CLASSIFICATION OF NON-SIMPLE PURELY INFINITE ALGEBRAS 45

the canonical epimorphism from G(h0 ; A, E) onto G(H0 ; A, E). An explicit (cer-
tainly very abstract !) example where the natural map is not injective only under
the assumptions of the old Lemma [442, lem. 3.8] has not been found so far – but
still could exist for some example that fits the general “axiomatic” assumptions of
[442, lem. 3.8] ( 23 ).
The Lemma [442, lem.3.8] is here worked out in more detail and extended to
Theorem 4.4.6 by adding a condition (DC), and we give a detailed proof of it in
Chapter 4. But we have used/misused Chapter 4 to give also a very detailed account
of sorts of calculations in K-theory of properly infinite unital C *-algebras and of
K∗ (h0 (A)0 ∩ E) for “KK-theory-defining” morphism h0 : A → E, where h0 (A)0 ∩ E
contains a copy of O2 unitally as our minimal permanent assumption. It leads to
groups G(h0 ; A, E) that must be understood in detail by the reader before he can
understand the proofs in several other parts of this book entirely, e.g. the groups
Ext(C ; A, B) in Chapter 5.
Fortunately we can generalize Kasparov’s proof (and the way over the cobor-
dism variant defined by J. Cuntz and G. Skandalis) of the identities KKp (A, B) =
KKc (A, B) = KKh (A, B) , i.e., the proof of the homotopy invariance of KK-theory,
to get that our cone-depending groups KK(C; A, B) are homotopy invariant. This
C-related result will be used to prove general decomposition theorems for strictly
continuous paths of “deriving” unitaries, compare Section 5 in Chapter 8 for details.
This decomposition results allow to proof that Ext(SC ; A, SB) – which is natu-
rally isomorphic to KK(C ; A, B) – satisfies the assumptions of Theorem 4.4.6 in
Chapter 4, in particular the – long time disputed – decomposition property (DC). It
implies that the natural group epimorphism from the m.o.c. cone C-equivariant un-
suspended E-theory – the generalized Rørdam groups R(C; A, B) – onto the group
KK(C; A, B) is an isomorphism. This is our base for isomorphism results, including
the case of pi-sun A, B and C = CP(A, B).
Has next blue part to be changed? Technics works?
Phillips method works?
Relation to KK-theory clear?
Independently from this result that completes of [442, lem. 3.8] we have in Sec-
tion 3 of Chapter 9 also outlined a generalization of the arguments of N. Ch. Phillips
in a way that they allow to show the homotopy invariance of m.o.c. cone C-related
unsuspended stable E theories – like our continuous variants R(C; A, B) of a kind
groups that M. Rørdam and G. Elliott used in their proofs for the classification of
certain Cuntz-Krieger algebras.
It is based on the homotopy invariance of the Grothendieck group of those
kinds of semigroups – but that are not homotopy invariant as semigroups itself
(sic !) – defined as unsuspended, but stable, variants of E-theory that satisfy
asymptotically some additional relations, who’s proof is essentially inspired by ideas

23 A specific counter-example would be interesting for finding the minimal necessary assump-

tions for the axioms that make the surjective map from G(h0 ; A, E) onto G(H0 ; A, E) injective.
46 1. INTRODUCTION AND MAIN RESULTS

of N.Ch. Phillips in [627]. But we do not know if this can be used to prove in this
special case the injectivity of G(h0 ; A, E) → G(H0 ; A, E), i.e., if two asymptotic
morphisms h1 : A → Q(R+ , B) and h2 : A → Q(R+ , B) are “stably homotopic”
if h1 and h2 define the same element of Ext(SC ; A, SB) – with C ⊆ CP(A, B)
generated by h0 .
Above and below should be solved now at the end of Chapter 8 !!!
It is also not clear how to produce an asymptotic homotopy between k ⊕ h0
and h0 if there is strictly continuous path of unitaries t ∈ [0, ∞) 7→ U (t) ∈ M(B)
with U (0) = 1 and

t ∈ [0, ∞) → U (t)∗ (k ⊕ H0 (a))U (t) − H0 (a) ∈ B ,

(where H0 := δ∞ ◦ h0 = h0 ⊕ h0 ⊕ · · · ) and

lim kU (t)∗ (k ⊕ H0 (a))U (t) − H0 (a)k = 0 .

t→∞

But please without (sic !) using the decomposition argument (DC) that we
prove in Section 5 of Chapter 8, because this proves anyway the full Theorem 4.4.6
in Chapter 4, and factorize through the whole – most abstract – study of C-related
KK-theory in Chapter 8 to get the general (unrestricted) homotopy invariance of
all this functors. It is a generalization of non-trivial observation of G. Kasparov
(plus arguments of J. Cuntz and G. Skandalis).
But above needs the still not proven facts that KK(C ; A, B) is homotopy in-
variant with respect to B, and that it is isomorphic to the kernel of

K1 πB (HC (A))0 ∩ Qs (B) → K1 Qs (B) .

If one wants to conclude as in [442] then one needs as an additional information,

e.g. the homotopy invariance of the Rørdam groups – a very special case of unsus-
pended but stable ideal-equivariant E-theory introduced and studied in Chapters 7
and 9.
But need also that the kernel of R-groups to KK
is E-homotopy equivalent of h0 . ?????
But there is then needed an extra condition, ????? related to Kasparov’s argu-
ments ????, for the homotopy invariance of the groups KK(C; ·, ·) and Ext(C; ·, ·).
But finally the classification is given by a generalized Kasparov theory. In
particular, A ⊗ O2 ∼
= B ⊗ O2 if A and B are separable stable nuclear C *-algebras
and have topologically isomorphic primitive ideal spaces.
Certainly, there is the still open question about the explicit calculation of our
“generalized KK-invariants” given by

KKnuc (X; ·, ·) := KK(CX ; ·, ·)

for the m.o.c. cone CX := CPXnuc of the X-equivariant residually nuclear c.p. maps,
i.e., nuclear maps that are residually nuclear with respect to a given “action” of
X. Notice that this is the open Question (Q2) in the special case where X is a
 2. TOWARDS A CLASSIFICATION OF NON-SIMPLE PURELY INFINITE ALGEBRAS 47

point. But one can define generalizations of the UCT-class, e.g. with help of the
“tautological” epimorphism and the 6-term exact sequence, cf. the proof of Theorem
O in Chapter 12 for some simple example. Again, the non-trivial question is, which
nuclear C *-algebras are in this class.
Compare Theorem O for an example of some related difficulties in question.
Now for applications of the classification the most natural question is:
When does a separable nuclear C *-algebra A absorb O∞ tensorial, i.e., when
A ⊗ O∞ ∼= A ? (If such an isomorphism exist then there exists an other that is
moreover an ideal-system preserving *-isomorphism.)
There are several necessary and sufficient conditions for A absorbing O∞ . We
show – finally at the end of Chapter 10 and independently from the classification
results and only from general properties of strong pure infiniteness and from the
observation that O∞ ∼ = D∞ := O∞ ⊗ O∞ ⊗ · · · by an isomorphism that is unitarily
homotopic to the embedding a ∈ O∞ 7→ a⊗1⊗· · · ∈ D∞ –, that a separable nuclear
A is isomorphic to A ⊗ O∞ if and only if A is “strongly” purely infinite in the sense
of the below given Definition 1.2.2 (24). The following definition is essentially the
same as [462, def. 4.1].

Definition 1.2.1. We call a C *-algebra A purely infinite if A has following

properties (i) and (ii):
(i) A has no character, i.e., every irreducible representation of A has dimension
> 1, and
(ii) For every a ∈ A+ and every positive c in the closed ideal generated by a, and
every ε > 0, there exists an element d := d(a, c, ε) ∈ A such that

k c − d∗ ad k < ε . (2.1)

([462, def. 2.3] is a Defination of MvN-equivalence, with reference to G.K.

Pedersen [621] that it is an equivalence relation on A+ .)
A C *-algebra A is p.i. if and only if A has the – formally stronger – property
that every nonzero c ∈ A+ is properly infinite, cf. [462, lem. 4.2], or Corollary 2.5.6.
To be checked: or more directly:
because an element a ∈ A is properly infinite in A, if and only if, πJ (a) is
infinite in A/J for every closed ideal J / A with a 6∈ J, cf. Part (v) of Lemma 2.5.3,
and since properties (i) and (ii) of the definition pass to πJ (a) for all quotients A/J
of A with closed ideals J 63 a .
Implies ????
Definition 1.2.1 has not much to do with that for “pi-1” or “pi(1)” because Def.
1.2.1 includes simplicity ????

24 In the case of stable or unital A we get this also from Theorem M in the same way as we

can deduce A ∼
= A ⊗ O∞ from Theorem B in the case of pi-sun algebras A.
48 1. INTRODUCTION AND MAIN RESULTS

NO! So as it is now, it is exactly the same as pi(1) (up to that it is not required
that `∞ (A) has no character!! But this can be shown in this case, because pi(1)
and pi-1 are equivalent properties.
It is still an open question – also for separable nuclear non-simple C *-algebras
A – if purely infinite C *-algebras are “strongly” purely infinite in sense of the
following Definition 1.2.2. Our ideal-system equivariant KK(X; ·, ·)-classification
works only for separable nuclear C *-algebras that are “strongly” purely infinite
C *-algebras in the sense of the following Definition 1.2.2.

Definition 1.2.2. The non-zero C *-algebra A is strongly purely infinite if

A satisfies the following condition:
For every a, b ∈ A+ and ε > 0 there exist elements d := d(ε, a, b), e := e(ε, a, b) ∈ A
such that

k a2 − d∗ a2 d k < ε , kb2 − e∗ b2 ek < ε and kd∗ abek < ε . (2.2)

(We abridge ‘purely infinite’ by ‘p.i.’ and ‘strongly purely infinite’ by ‘s.p.i.’)

It is easy to see that every non-zero strongly p.i. algebra is a purely infinite C *-
algebras in sense of Definition 1.2.1, indeed: Let a := b := c1/2 in Definition 1.2.2
for c ∈ A+ , then Definition 1.2.2 shows that each non-zero c ∈ A+ is a properly
infinite element of A in the sense of [462, def. 3.2], cf. also Section 1 of Chapter 2.
If we use the isomorphism of C and C ⊗ O∞ ⊗ O∞ · · · , then we can see that
A := B ⊗C is strongly p.i. for every (simple) pi-sun algebra C and every (non-zero)
C *-algebra B. We show moreover that A := B ⊗ Z is s.p.i. if A (2-quasi-) traceless
and Z denotes the Jiang-Su algebra, cf. also [690] in case of exact B.
If a C *-algebra A has the WvN-property of Definition 1.2.3, then A has the
property that for every self-adjoint h∗ = h ∈ M(A⊗K)ω there is a *-monomorphism
λh : C ∗ (h) ⊗ O∞ → M(A ⊗ K)ω with λ(h ⊗ 1) = h. The latter implies that M(A) is
purely infinite. (It is still an open question if the pure infiniteness of M(A) implies
that A is strongly p.i., except some special cases as
e.g. ??? Even in case that ???? .)
M. Rørdam and the author have shown in [463] that strongly purely infinite
C *-algebras have the WvN-property, see Remarks 2.15.12 and 3.11.4 in Chapters
2 and 3 for an outline of the proof (including correction of typos in [463]).
It is still an open question, whether or not p.i. (or at least the existence of
the above described *-monomorphisms λh in sufficiently general position) implies
“strongly” p.i. in general.
By [93, thm.4.17], a sufficient condition for A being strongly p.i. is that every
non-zero quotient A/J of A is p.i. in the sense of the definition of J. Cuntz [172,
p. 186] (cf. Introduction to Chapter 2 and compare also [169, thm. 1.13, thm. 1.4]
in the special cases of A = On for n = 2, . . . , ∞), i.e., for every closed ideal J ⊆ A
holds that every non-zero hereditary C *-subalgebra D of A/J contains a (non-zero)
properly infinite projection p ∈ D.
 2. TOWARDS A CLASSIFICATION OF NON-SIMPLE PURELY INFINITE ALGEBRAS 49

In particular, if A has real rank zero, then A is strongly p.i., if and only if, each
projection of A is infinite.
The algebra C0 ((0, 1], O2 ) is strongly p.i. but is not purely infinite in the sense
of J. Cuntz. The unitization of O2 ⊗ K has real rank zero and is purely infinite in
the sense of J. Cuntz but is not p.i. in the sense of our Definition.

We list the needed results on purely infinite and strongly purely infinite non-
simple C *-algebras in Chapters 2 and 3 and outline the ideas for their proofs.
This results are mainly joint work with M. Rørdam and E. Blanchard and we refer
sometimes for details of the proofs to [462], [463] and [93].
Begin of discussion of ‘‘WvN-property’’:
1) Has to be revised at several places.

2) Sort mentioned results for chapters, and formulate them there.

3) Where are Defs. of: ‘‘residually nuclear maps’’ ?.

In case of A ⊆ B and V : A → B residually nuclear c.p. contraction
?.
V is ‘‘1-step approximately inner’’ c.f. Definition 3.10.1? Not a
good place!

We prefer to work in this book with the WvN-property of the following Defi-
nition 1.2.3 and, sometimes, even with additional assumptions as listed in Remark
3.11.1, because the WvN-property is, in a conceptional sense, a starting point for
most of the proofs, and the strong pure infiniteness is a basic property for all of our
constructions and its applications. It turns out in Chapter 12 as a result of ideal-
system equivariant KK(X; ·, ·)-classification that this two properties are equivalent.

Definition 1.2.3. A C *-algebra B has the Weyl–von-Neumann property

– in short WvN-property – if for every σ-unital hereditary C *-subalgebra D ⊆ B and
every C *-subalgebra C of the multiplier algebra M(D⊗K) of D⊗K, every residually
nuclear completely positive contraction V : C → D ⊗ K can be approximated in
point-norm by maps C 3 c 7→ d∗ cd with contractions d ∈ D ⊗ K ( 25 ).
We say that a C *-algebra B has residually nuclear separation if, for every
separable C *-subalgebra C ⊆ B, every a ∈ C+ and every ε > 0, there exists a
residually nuclear contraction V : C → B such that kV (a) − ak < ε.

Notice that we can modify the residually nuclear contraction V : C → B in

Definition 1.2.3 that it has in addition the property V (C) ⊆ CBC.
Next should be part of a Proposition/ Theorem -- to be cited
The WvN-property implies that B is purely infinite in the sense of Definition
1.2.1, that equivalently says that all elements of B+ are properly infinite in B, i.e.,
25 It implies that the c.p.c. V is “1-step approximately inner” in the sense of Definition 3.10.1.
50 1. INTRODUCTION AND MAIN RESULTS

B satisfies the equivalent properties pi(1) of Definition 2.0.4 and pi-1 of Definition
??, cf. also [462, def. 3.2, def. 4.1, thm. 4.16].
Give ref.s/not.cite for def.s and proofs

Shift details to other places?!!

Moreover the WvN-property implies that for every separable C *-subalgebra
D ⊆ B with the property that the inclusion map ιD : d → d is nuclear (which implies
automatically that D is exact), there is a C *-morphism h : (D⊗K)⊗O∞ → (B⊗K)ω
such that h((d ⊗ k) ⊗ 1) = d ⊗ k for d ∈ D, k ∈ K under the natural inclusion
B ⊗ K ⊂ (B ⊗ K)ω of B ⊗ K into the ultrapower (B ⊗ K)ω . (26).
Indeed, let C := D ⊗ K, s1 , s2 ∈ M(D ⊗ K) isometries with s∗j sk = δj,k 1 the
map c 7→ s1 cs∗1 + s2 cs∗2 is residually nuclear, thus is 1-step approximately inner in
B ⊗ K if B has the WvN-property.
And it shows that each element of (B ⊗ K)+ is properly infinite if B has the
WvN-property.
It seems (! open question?) that the WvN property also implies
that
πB (D)0 ∩ M(B)/B contains a unital copy of O∞ .
Try to factorize the cone of D over `∞ (B ⊗ K)/c0 (B ⊗ K) ???
One can show (27) that a C *-algebra B is strongly p.i. if B has both of the
WvN-property and the residually nuclear separation property of Definition 1.2.3.
It is obvious that the WvN-property passes to hereditary C *-subalgebras and
is invariant under stabilization, hence is an invariant of Morita equivalence. The
definition of the WvN-property says only something that one can limit the algebras
C in the definition of the WvN-property to the case of separable C.
Moreover it is not difficult to see
BY DEFINITION ??
that B has the WvN-property if Bω has the WvN-property.
Clearly the identity map idB =: V on nuclear C *-algebras B defines residually
nuclear separation for each separable C *-subalgebra C of B. Thus, for nuclear
C *-algebras B the WvN-property and strong pure infiniteness are the same.
Studies of primitive ideal spaces of nuclear C *-algebras in combination with
[359] and [456] lead to the observation that all stable separable C *-algebras A have
the (formally) stronger property of Abelian factorization, which obviously implies
residually nuclear separation:
26 We define the ultrapower A of A by A := ` (A)/c (A) where ω ∈ γ(N) := β(N) \ N
ω ω ∞ ω
is a point in the non-elementary part of the Stone-Čech compactification of N.
Other authors use the notation Aω for this algebra and then Aω for the central sequence algebra
A0 ∩ Aω .
27 E.g. by ideas in the proof of [463, cor. 7.22] – that contains unfortunately some typos,

see Chapters 2/3?? for corrected versions.

 2. TOWARDS A CLASSIFICATION OF NON-SIMPLE PURELY INFINITE ALGEBRAS 51

Definition 1.2.4. A stable separable C *-algebra A has the Abelian fac-

torization property if there exists a locally compact Polish space P and non-
degenerate *-monomorphisms h : C0 (P, K) → M(A) and k : A → M(C0 (P, K))
such that the “C *-correspondence” γ := M(h) ◦ k : A → M(A) (– a special kind
of Hilbert A-module with a left-sided action by A – ) has the property
J = γ −1 M(A, J) for all J ∈ I(A) .

All simple stable C *-algebras A have Abelian factorization by the trivial reason
that one can take here for P := {p} a point, i.e., C0 (P, K) = K and use that A⊗K ∼ =
A if A is stable, h(T ) := 1⊗T ∈ M(A⊗K) ∼ = M(A) and let k : A → L(`2 ) ∼ = M(K)
any non-degenerate *-representation of A on `2 .
It turns out that, if A has the Abelian factorization property, then the l.c. Polish
space P and h can be chosen always such that P is 1-dimensional ( 28 ).
Therefore strong pure infiniteness and the WvN-property ( ?????? ref? ) are
now known to be equivalent for all C *-algebras, because both have to be checked
only on suitable sufficiently large separable C *-subalgebras with one of this prop-
erties.
A proof of the Abelian factorization (see ??) Definition in all generality uses a
detailed study of the topology of second countable locally quasi-compact sober (=
“point-complete”) T0 spaces (that we call “Dini spaces”) and their “Dini functions”,
using the very last conclusions in Chapter 12. Therefore we have not included a
proof here in the book. And we avoid to use Abelian factorization here directly.
Instead we use methods that are still fairly elaborate but use formally weaker
assumptions than the “residual factorization property” of A, (cf. Definition 1.2.3).
But in the special case of separable exact C *-algebras, one can derive residu-
ally nuclear separation, defined in Definition 1.2.3 with help of Theorem K by the
following construction:
Use Theorem K with A ⊗ O2 ⊗ K in place of both A and B and with the natural
“action” of Prim(A) = Prim(B) on A for the proof of residually nuclear separation
of separable exact A :
The resulting residually nuclear endomorphism h0 has the property that its
restrictions to commutative C *-subalgebras C of A approximately dominates the
inclusion map C ,→ A in the sense of Definition 3.10.1. Thus, the point-norm
closed matricial operator-convex cone generated by h0 defines “residually nuclear
separation” for A as defined in Definition 1.2.3. This argument works so simple
only for separable exact C *-algebras A. And, unfortunately, needs ultra-power
constructions that we want not to bore the reader at the main parts of this books.
So we make sometimes additional assumptions that we all replace in Chapter 12
by embeddings into suitable separable C *-algebras in the ultrapower that have the
required properties. Then we observe that this what we can do in the ultrapower
28 But P is not something like a one-dimensional polyhedron. It is only some projective limit

of such polyhedra.
52 1. INTRODUCTION AND MAIN RESULTS

“in one step” can be done approximately in the given algebra itself. Such a “fitting
together” and approximation arguments in the earlier chapters can be applied to
all studied cases in a controlled way.
Alternatively, mone can use a deeper result for separable exact C *-algebras
A: It says that A ⊗ O2 contains a regular Abelian C *-subalgebra C in the sense
of Definition B.4.1. It implies that A has Abelian separation (in sense of: give
reference here!) and follows also from Corollary 12.3.1 and applications of Theorem
K and Corollary L in the remark following Corollary 12.3.1, and the corresponding
result for nuclear C *-algebras in [464] and [359]. But we should not get lazy and
put non-trivial stuff simply in the assumptions: Applicability or Non-applicability
depends mainly from the difficulty to observe the assumptions of some derived
useful applications and conclusions (in the theorems). The study of assumptions
and its interrelations, weaker properties and inducing properties is what a result
makes a true invention.
Thus, we get finally that an exact C*-algebra A is strongly p.i., if and only if,
A has the WvN-property. But we can not use this simple (but not obvious) identity
of two properties before the very end of Chapter 12!
But the reader should not forget that one first has to give a proof of Theo-
rem K to get a base e.g. for the deduction of the permanently used and needed
WvN-property, that we must derive from strong pure infiniteness to reach our ap-
plications.
We go along an alternative way as follows:
A C *-algebra B is s.p.i. if and only if its ultra-power Bω is s.p.i., cf. Proposition
2.16.8. A C *-algebra B is weakly purely infinite if and only its ultra-power Bω has
“residually nuclear separation” in the sense of Definition 1.2.3 (and in a very “local”
way), see also [463, prop. 7.13] ( 29 ).
Next should be a Prop./Thm. in Chp.3 ?? Which section?
More precisely:
If B is weakly purely infinite (i.e., B is n-pi for some n ∈ N) and A ⊆ Bω is a
separable C *-subalgebra of Bω , C ⊆ A an Abelian C *-subalgebra of A then there
exists a separable C *-subalgebra D ⊆ Bω , an Abelian C *-subalgebra C1 ⊆ D and
a conditional expectation E : D → C1 such that A ⊆ D, C ⊆ C1 , and E(D ∩ J) ⊆
C1 ∩ J for each closed ideal of J / Bω .
Moreover, one can find such a system (D, E : D → C1 ⊂ D, D) for each given
separable A ⊂ Bω with this properties and with the additional property that E
is an – inside D – approximately inner c.p. map. In particular, C1 is a “regular”
C *-subalgebra of D.

29 This property of B is “natural” because B is completely characterless – i.e., no σ-unital

ω
hereditary C *-algebra of Bω has a non-zero character – if B weakly p.i.
 2. TOWARDS A CLASSIFICATION OF NON-SIMPLE PURELY INFINITE ALGEBRAS 53

Conversely, if Bω has no irreducible representations that contain the compact

operators in its image, and if in Bω such systems (D, E : D → C1 ) exists for each
given C ⊆ A ⊂ Bω , then B is weakly purely infinite.
Since B is strongly p.i. if and only if Bω is strongly p.i., it follows that Bω has
the WvN-property if B is strongly p.i.
In this way we obtain that a C*-algebra A is s.p.i. in the sense of Definition
1.2.2, if and only if, Aω has the WvN-property of Definition 1.2.3.
That the WvN-property of Aω induces the WvN-property of A, can be shown
also directly from the definition of the WvN-property. ???? Really ???? Idea
forgotten? ??
Thus, every s.p.i. C*-algebra A has the WvN-property.
The converse direction – that the WvN-property of A implies that A is s.p.i.–
is for the final proofs of theTheorems ?? and ????? important. ????
We need “nuclear separation” to establish the realizations of actions Ψ by the
related cones CΨ of Ψ-equivariant residually nuclear maps.
Which proofs use “nuclear separation” and where is it defined?
Chp.2: residual nuclear separation and WvN-property imply strong p.i.
But for simple C *-algebras there are many formally weaker – but for them
equivalent – properties.
Chp.3: residually nuclear maps build the “minimal” ideal system invariant cone
(the later needed property is that this cone is non-degenerate for separable stable
C *-algebras).
Chp.6: “nuclear separation” is – needed / required – to proof the “embedding
theorem” (give references to def’s and theorems !!!)
(Starting alone from any non-degenerate l.s.c. action of Prim(B) on separable
exact A that is monotone upper s.c., it can be proved only fully after one has proven
with help of this special case in chp.12 that the nuclear separation of the action
can be proved with help of results in Chp. 7, 9 and 12).
Chp.12: // B separable and B stable, σ-unital, and B ⊗ O∞ ∼ = B, show
existence of separable stable C *-subalgebras D with B ⊆ D ⊂ Q(R+ , B), D stable,
σ-unital, and D ⊗ O∞ ∼ = D, with a regular Abelian subalgebra C (that separates
the ideals of D).
Then finding for A :=??? c.p. maps V : A → C ⊂ D (that realize the action of
Prim(B) = piB (Prim(Q(R+ , B))) on A and with their help a nuclear embedding
k1 : A ⊗ O2 → D suitable (existing by the main embedding theorem) and realizing
the action of Prim(D) on A. But h1 := k1 ((·)⊗1) is re-scaling invariant by Chapter
7. Thus, unitary equivalent to a “defining” nuclear embedding A → O2 → B ⊗ O∞ .
In particular this implies that the given action of Prim(B) on A has nuclear
separation.
54 1. INTRODUCTION AND MAIN RESULTS

This argument works only for separable exact C *-algebras. Because for non-
separable or non-exact C *-algebras there does not exist an “embedding theorem”.
(Into O2 ?)
The WvN-property of A implies always that A is p.i.
Since WvN-property implies p.i., it is also part (??) of the open problem if p.i.
implies s.p.i.
It is unknown if p.i. implies s.p.i.
It is not unlikely that an elementary observation exists that shows that A with
WvN-property is always s.p.i.
The results of this book allow to proof that for all separable stable A the natural
action of X := Prim(A) on A – given by the natural isomorphism OX ∼ = I(A) – has
“residually nuclear” separation (i.e., this action is “induced” from the m.o.c. cone
CPrn (A, A) of residually nuclear maps from A to A), but to carry the proof out one
needs the possibly stronger – and not in this book proved – “Abelian separation”
property for separable C *-algebras.
important???// This can be used for ????????
The QUESTION is/was:
Is there a simple way to derive directly
for separable *-algebras A from WvN-property of A
the STRONG pure infiniteness of A?
Or, -- equivalently -- from WvN-property of A
the WvN-property of Aω .
The residual nuclear separation follows in case of separable exact C *-algebras
from Corollary L, it implies directly that every separable exact C *-algebra A has
the stronger property that A ⊗ O2 contains a “regular” Abelian C *-subalgebra that
separates the ideals of A.
The existence of a regular Abelian C *-subalgebra implies in particular the
existence of an “Abelian factorization” for A ⊗ K, which is (formally) stronger than
“residually nuclear separation”.
A more easy way to prove that the WvN-property implies strong pure infinite-
ness is not in sight yet.
Despite the rather involved result of existence of “Abelian factorization” it
remains the more special and interesting question, whether A ⊗ O2 contains a
regular Abelian C *-subalgebra in the sense of Definition B.4.1 for every separable
C *-algebra A. By Corollary 12.3.1 and [464, thm. 6.11] the latter is the case, if
and only if, A ⊗ O2 contains a regular exact C *-subalgebra.
End discussion of ‘‘WvN-property’’. Still not well explained !
In special cases we obtain that “p.i.” implies “strongly p.i.”, e.g., for algebras
of real rank zero, for algebras with Hausdorff primitive ideal space, or for separable
 2. TOWARDS A CLASSIFICATION OF NON-SIMPLE PURELY INFINITE ALGEBRAS 55

C *-algebras with linearly ordered lattice of closed ideals. In particular, all simple
purely infinite algebras are strongly purely infinite.
Why we consider different definitions of pure infiniteness?
One can only work with the property of strong pure infiniteness, but the others
are usually easier to verify. We can then try to check additional properties of the
studied algebras that allow to conclude strong pure infiniteness e.g. from (1-) pure
infiniteness. But this is also an open problem for separable nuclear C *-algebras B.
There are only some sufficient conditions as e.g. B has Hausdorff primitive ideal
space,
(of finite dimension ?),
I(B) is linearly ordered ideal lattice, and to “s.p.i.” equivalent properties, as
e.g. B ∼
= B ⊗O∞ if B is nuclear. Simple and exact strongly purely infinite separable
C *-algebras B do not absorb O∞ tensorial in general, as e.g. the example of the
∗
(strongly) purely infinite algebra B := Creg (F2 ) ⊗ R shows, where R is the finite
∗
but not stably finite simple nuclear C *-algebra of M. Rørdam [687] and Creg (F2 )
∗
is the (exact) regular group C *-algebra of Creg (F2 ). M. Rørdam itself denoted
his algebra R by W .

At first we give some definitions concerning non-Hausdorff spaces, which are

needed for suitable generalizations of the main results for simple p.i. algebras to
non-simple p.i. algebras.

Definition 1.2.5. A topological T0 -space X is sober if every prime closed

subset of X is the closure of a point ( 30 ).
Here a closed subset F ⊆ X is prime, if F is not the union F1 ∩ F2 of two
closed subsets F1 , F2 of F that are both different from F .

Since, by results of J. Dixmier

citation? See also book of Pedersen ...
for a separable C *-algebras A, closed prime ideals of A are primitive, the
primitive ideal space Prim(A) of a separable C *-algebra is a “sober” T0 -space (we
call it “point-complete”) with a countable base of its topology. The countable
base can be described by the supports of the lower semi-continuous functions J ∈
Prim(A) 7→ kπJ (bn,k )k, where bn,k = (an − 1/k)+ for a sequence a1 , a2 , . . . which is
dense in {a ∈ A+ : kak = 1} and n, k ∈ N.
In 2001 N. Weaver [815] published an example of non-separable prime C *-
algebras that is not non-primitive. Later T. Katsura [411] constructed examples
of non-separable AF-algebras that are prime but not non-primitive.

30It is also called point-complete or tidy. Sober l.c. T -spaces are sometimes also called
0
spectral space , but some authors define spectral spaces as point-complete T0 -spaces that have a
base of the topology consisting of compact open subsets.
56 1. INTRODUCTION AND MAIN RESULTS

But it seems that one needs some set theory axiom that is
stronger than the Axiom of choice?
But weaker than the Continuum Hypothesis??

Definition 1.2.6. Let X be a sober T0 -space that is second countable, i.e.,

has a topology with a countable base.
We say that X acts on a C *-algebra A, if there is given a monotone map
Ψ from the lattice O(X) of open subsets Z of X to the closed ideals I(A) of a
C *-algebra A with Ψ(∅) = 0 and Ψ(X) = A .
The action Ψ of X on A will be called upper semi-continuous , if, moreover,

(i) Ψ(Z1 ∪ Z2 ) = Ψ(Z1 ) + Ψ(Z2 ) for every pair of open subsets Z1 and Z2 of
X, and
S S
(ii) Ψ( Zn ) is the closure of Ψ(Zn ) if Z1 ⊆ Z2 ⊆ · · · is an increasing
sequence of open subsets of X, i.e., Ψ is monotone.

The action Ψ is called lower semi-continuous if (instead of (i) and (ii)):

(iii) Ψ(Z1 ∩ Z2 ) = Ψ(Z1 ) ∩ Ψ(Z2 ) for every pair of open subsets Z1 and Z2 of
X, and
T T
(iv) Ψ(K) = Ψ(Zn ), where K is the interior of Zn in X, if Z1 ⊃ Z2 ⊃ · · ·
is a decreasing sequence of open subsets of X.

If the monotone map Ψ satisfies (i)–(iv), then we say that X acts on A con-
tinuously. (Or: Ψ is continuous).
It is monotone continuous if it satisfies only (ii)–(iv).

An example of a continuous action is the action of Y := Prim(A) on A given

by the one-to-one correspondence between the open subsets of Prim(A) and closed
ideals I(A) of A:
\
Z ∈ O(Y ) 7→ ΨA (Z) := {J : J ∈ Prim(A) \ Z} .

Note that ΨA (∅) = 0 and that here we define ΨA (Y ) := A.

If nothing is said about an action of Prim(A) on A, then this natural corre-
spondence is used.
Therefore, the monotonous maps Ψ from X to the closed ideals of A are in
one-to-one correspondence to the monotonous maps from the open subsets of X
into the open subsets of the primitive ideal space Y = Prim(A) of A.
Check cite of Fell60 [296]!
By [559] and [296], if A is separable, then I(A) is naturally order anti-
isomorphic (respectively the family F(Prim(A)) of closed subsets of Prim(A) is order
isomorphic) to a closed subset ZA of the Hilbert cube [0, 1]∞ with the coordinate-
wise order: Take the map

J ∈ I(A) 7→ (kπJ (cn )k)∞

n=1 ,
 2. TOWARDS A CLASSIFICATION OF NON-SIMPLE PURELY INFINITE ALGEBRAS 57

where c1 , c2 , . . . is a dense sequence in {a ∈ A+ : kak = 1}. The induced Haus-

dorff topology on F(Prim(A)) is the Fell-Vietoris topology. The one-to-one map
F ∈ F(Prim(A)) 7→ Prim(A) \ F ∈ O(Prim(A)) ∼ = I(A) defines on O(Prim(A)) a
topology that is equal to the so-called Lawson topology.
Since the Dini-functions ϕ : X → [0, 1] on the Dini spaces X
(= point-complete second countable l.c. T0 -spaces)
are naturally related to the ‘‘way-below’’ (= ‘‘well-below’’)
continuous monotone functions on
the continuous lattice O(X) into [−∞, ∞]
if we take e.g. the strictly order-monotone homeomorphism
ψ : t ∈ [0, 1] 7→ [−∞, +∞] given by ψ(t) := tan(π(t − 1/2)).

Countably generated Continuous lattices Ω are in 1-1-correspondence

with O(X) with -- up to homeomorphisms unique -- Dini space X.

There is a Theorem in lattice theory that gives the same embedding

in a different formulation (Graetzer [332], Lattice Theory , Thm.
47 (of Larson),
see also [321, thm. IV-3.20].
The set ZA ∼ = I(A) contains zero (0, 0, . . .) ↔ A and ZA is preserved under
forming of component-wise maxima: If α, β ∈ ZA , then max(α, β) ∈ ZA . The
points of Prim(A) can be found by considering the prime elements of ZA with
respect to the continuous “multiplication” given by (α, β) 7→ max(α, β) .
If we identify Prim(A) with this subset of prime elements in ZA ∼
= I(A), then
the closure of a set S in Prim(A) ⊆ ZA is given by the hull-kernel operation
hk(S) := {α ∈ Prim(A) : α ≤ max S} .
A simple set-theoretical argument shows, that, for second countable sober
(point-complete) T0 -spaces X, the above defined upper semicontinuity of a faithful
and monotonous map Ψ : O(X) → I(A) which satisfies moreover (iii) and the non-
degeneracy Ψ(∅) = 0 and Ψ(X) = A, is equivalent to the existence of a (unique)
continuous map p from Prim(A) into X such that Ψ(Z) is the ideal corresponding
to p−1 (Z).
If Ψ is given by p−1 , then properties (i)–(iii), Ψ(X) = A and Ψ(∅) = 0 hold.
The action Ψ is faithful, if and only if, p(Prim(A)) ∩ (U \ V ) 6= ∅ for all open subsets
V ⊂ U ⊆ X with V 6= U . Then, furthermore, (iv) holds for Ψ, if and only if, p is
“pseudo-open” (which is the same as “open” only if X is Hausdorff, see [359] for
details).
Let us consider the special case where X is a locally compact Hausdorff space
and Ψ is faithful and satisfies (i)–(iii) and Ψ(∅) = 0, Ψ(X) = A. By Dauns-Hofmann
theorem ([402, exercise 10.5.84], [616, cor. 4.4.8]) one gets that there is a unique
C *-morphism α from C0 (X) into the center of M(A) such that Ψ(Z) = α(C0 (Z))A
58 1. INTRODUCTION AND MAIN RESULTS

for every open subset Z of X. It is then easy to see, that A is C *-bundle with
base space X in the sense of [471] (31), if and only if, Ψ is non-degenerate and
satisfies conditions (i)–(iv). Thus, the faithful continuous actions of locally compact
Hausdorff spaces X on a C *-algebra A are in natural one-to-one correspondence to
the structures of C *-bundles on A in the sense of [471]. It says that our general
theory contains those results for C*-algebra bundles.
In the case of continuous bundles, E. Blanchard [89] has found a special case
of the below given “trivialization” Theorem K. The remaining points are, to give
necessary and sufficient conditions on the bundle under which it is exact (this was
done in [471]), and to give other conditions under which it is (strongly) purely
infinite. See Chapter 2 for partial answers and related problems.

A C *-algebra B defines on its multiplier algebra M(B) or, more generally,

on the C *-algebra L(E) of B-linear operators with bounded adjoints on a Hilbert
B-module E, an action Ψup of X = O(Prim(B)) ∼ = I(B) as follows:

Ψup : J ∈ I(B) 7→ M(E, J) ∈ I(L(E))

where M(E, J) := {T ∈ L(E) ; T E ⊂ EJ} . In the case where E = B (considered

as left B-module), we have M(B) = L(E) and M(B, J) is the kernel of the natural
C *-morphism M(B) → M(B/J). Ψup satisfies (i), (iii) and (iv) if B is σ-unital
and E is countably generated over B. But Ψup does not satisfy (ii) in general,
e.g. in the case B = c0 ⊗ K. If λ is an isometric B-module isomorphism from E1
onto E2 , then λ(M(E1 , J)) = M(E2 , J) for every closed ideal J of B.
How could the realization of a given action of Prim(B) on A look like?
Above we have defined the action Ψup : I(B) → I(M(B)) . In a more general
setting we have:

Definition 1.2.7. Let A ⊆ M(B). One has two natural maps ΨA,B up
down and ΨB,A
from the open subsets O(Prim(A)) ∼
= I(A) of Prim(A) into I(B) (respectively from
O(Prim(B)) ∼= I(B) into I(A)), given by:

ΨA,B
down : K ∈ I(A) 7→ BKB ∈ I(B) ,

respectively by
Ψup
B,A : J ∈ I(B) 7→ A ∩ M(B, J) ∈ I(A) .

Here BKB means the closure of the linear span of elements bac, where a ∈ K,
b, c ∈ B

The corresponding actions of Prim(A) on B and of Prim(B) on A are increasing

maps, the one corresponding to ΨA,B down satisfies (i) and (ii) (but it is in general not
faithful and has neither (iii) nor (iv)), and the one corresponding to Ψup B,A fulfills
(iii) and (iv) (but it is in general not faithful and has neither property (i) nor (ii)).

31 I.e., A is the C *-algebra of continuous sections in a continuous field of C *-algebras

{Ax }x∈X , equipped with its C0 (X)-module structure.

 2. TOWARDS A CLASSIFICATION OF NON-SIMPLE PURELY INFINITE ALGEBRAS 59

If A ⊆ M(B) satisfies A ∩ (I + J) = A ∩ I + A ∩ J for all closed ideals I, J of M(B)

and if B is σ-unital, then Ψup
B,A satisfies (i).

In view of Theorem K, it is important that Ψup

B,A satisfies (ii) if A ⊆ B.

There are examples which show that

J ∈ I(B) 7→ M(B, J) + B ∈ I(M(B)/B)
is in general neither lower semi-continuous nor upper semi-continuous, but satisfies
(i) and (iii) if B is σ-unital. This causes a striking technical difficulty for proofs of
generalized Weyl-von Neumann-Voiculescu theorems, cf. Chapter 5.
The results and methods of this monograph allow to prove
(but not here in this book because it requires in detail several
another hundert pages)
that each separable stable C *-algebra A has
‘‘Abelian factorization’’.
This implies a proof of the n.c. Michael selection Conj.
and a proof that the M vN -property implies strong pure infiniteness

It remains the question if, for every separable C *-algebra A,

the O2 -stabilization A ⊗ O2 contains a
‘‘regular abelian C *-algebra’’ C in sense of Definition 1.2.9
The non-commutative Michael selection Conjecture 12.3.2 says that for separa-
ble stable C *-algebras A and B and lower semi-continuous actions Ψ : I(B) → I(A)
of Prim(B) on A there is a C *-morphism h : A → M(B) such that Ψ = h−1 Ψup B,h(A) ,
−1
i.e., that Ψ(J) = h M(B, J) for J ∈ I(B). Moreover, h should be weakly Ψ-
residually nuclear in the sense of Definition 1.2.8. A proof of Conjecture 12.3.2
would imply that C *-algebras with WvN-property in Definition 1.2.3 are strongly
purely infinite, because Conjecture 12.3.2 implies the conjecture that every C *-
algebra has residually nuclear separation in sense of Definition 1.2.3.
Theorem 12.1.8 and the below stated Theorem K are partial results in the
direction of the non-commutative Michael selection Conjecture 12.3.2. Theorem
12.1.8 (and Proposition 12.2.15) will be used for the proof of Theorem K.

Definition 1.2.8. Suppose that X is a point-complete (also called “spectral”,

“sober” etc.) T0 -space with a countable base of its topology. Let ΨA and ΨB
monotonous actions of X on C *-algebras A and B. A completely positive map
V : A → B is called Ψ-equivariant (or, more precisely, ΨA –ΨB -equivariant) if
V (ΨA (Z)) ⊆ ΨB (Z) for every open subset Z of X.
V is Ψ-residually nuclear (or, more precisely ΨA –ΨB -residually nuclear) if
it is Ψ-equivariant and the class maps [V ] : A/ΨA (Z) → B/ΨB (Z) are nuclear for
every open subset Z of X.
A completely positive map T from A into M(B) will be called weakly Ψ-
equivariant (respectively weakly Ψ-residually nuclear) if Vb := b∗ T (·)b is Ψ-
equivariant (respectively is Ψ-residually nuclear) for every b ∈ B.
60 1. INTRODUCTION AND MAIN RESULTS

We say that V is residually nuclear (respectively that T is weakly resid-

ually nuclear) in the special situation where X = Prim(B), A ⊆ M(B), ΨB is
natural (i.e., ΨB := idI(B) ), ΨA = Ψup
B,A and V is Ψ-residually nuclear (respectively
T is weakly Ψ-residually nuclear).

If A is exact, then V is Ψ-residually nuclear if V is Ψ-equivariant and is nuclear.

But this is not true for arbitrary separable A (and arbitrary actions). If A ⊆ B
and V : A → B
In Chapter 12 we finish the proof of the following embedding (or “sub-
triviality”) Theorem K which generalizes Theorem A and E. Blanchard’s sub-
triviality theorem for exact C *-algebra bundles in [89].

Theorem K (Embedding Theorem). Suppose that A and B are separable and

stable C*-algebras, such that A is exact and B is strongly purely infinite.
Let ΨA : O(X) ∼ = I(B) → I(A) be a lower semi-continuous action of X :=
Prim(B) on A which satisfies the “monotone upper semi-continuity” condition (ii)
of Definition 1.2.6, together with the non-degeneracy condition
ΨA (B) = ΨA (X) = A and ΨA (∅) = 0 .

Then there exists a Ψ-residually nuclear *-monomorphism h : A⊗O2 ,→ B such

that, for h0 := h((·) ⊗ 1) and every closed ideal J of B,
h0 (ΨA (J)) = J ∩ h0 (A) .
Moreover, h0 is unitarily homotopic in the sense of Definition 5.0.1 to h0 ⊕ h0 .
Every nuclear *-monomorphism h1 from A into B, which is unitarily homotopic
to h1 ⊕h1 and satisfies h1 (ΨA (J)) = J ∩h1 (A) for J ∈ I(B), is unitarily homotopic
to h0 .
If ΨA (J) = A always implies that J = B, then h can be found such that h0 (A)B
is dense in B.

A part of the proof of this Embedding Theorem is given in Chapter 6. It uses an

additional assumption of a property of B: The existence of a “regular Abelian C *-
subalgebra” in B ⊗O2 in sense of Definition 1.2.9. From this property we derive the
“Abelian factorization” property and the (for some arguments in our proof needed)
“minimal” assumption of the “residual nuclear separation” property of B that we
must require for the proof in Chapter 6. After that we can free the proof of the
embedding theorem from the extra assumption of existence of “residually nuclear
separation” by showing in Chapter 12 that we can find in the asymptotic corona
Q(R+ , B) of B a separable strongly purely infinite subalgebra F that contains B,
contains a regular abelian C *-subalgebra and has a monotone continuous action and
upper s.c. action of I(A) on F that is compatible under rescaling automorphisms
of Q(R+ , B). The uniqueness up to unitary equivalence of the embedding of A ⊗ O2
in Q(R+ , B) gives that this is unitarily equivalent to an embedding from A ⊗ O2 in
B that fits into the given action of Prim(B) on A.
 2. TOWARDS A CLASSIFICATION OF NON-SIMPLE PURELY INFINITE ALGEBRAS 61

The results of this book (all together) lead to a proof that any separable stable
C *-algebra B has an Abelian factorization. The existence of regular Abelian C *-
subalgebras in B ⊗ O2 follows for separable exact B from Corollary L given below.
But despite of the not in this book proven existence of “Abelian factorization”
for separable stable B, the Question about the existence of a regular Abelian C *-
∗
subalgebra in Cmax (F∞ ) ⊗ O2 is still open. (If it exists, then B ⊗ O2 contains a
regular Abelian C *-subalgebra AB ⊆ B ⊗ O2 for every separable C *-algebra B.)
Why that? Give reason here, or reference to reasons given on other places!

Definition 1.2.9. We call an Abelian C *-subalgebra C ⊆ B of a C *-algebra

B regular in B if

(i) (J + I) ∩ C = (J ∩ C) + (I ∩ C) for all I, J ∈ I(B), and

(ii) C separates the lattice of ideals of B, i.e., I ∩ C = J ∩ C implies I = J.

The additional assumption of the existence of a regular Abelian C *-subalgebra

C in B ⊗ O2 will be removed finally in Chapter 12 by showing that the asymptotic
corona Q(R+ , B) contains a separable strongly purely infinite C *-subalgebra B1 ⊂
Q(R+ , B) with the properties that B ⊂ B1 and that B1 ⊗ O2 contains a regular
Abelian C *-subalgebra C ⊂ B1 ⊗ O2 . The action Ψ of Prim(B) on A can be
extended to an action Ψ1 of Prim B1 on A by letting Ψ1 (K) := Ψ(K ∩ B) for
K ∈ I(B1 ). This extended action is invariant under “rescaling” (i.e., parameter
change).
The uniqueness of the Ψ-preserving nuclear *-monomorphisms from A ⊗ O2 in
Q(R+ , B) up to unitary homotopy (cf. Chapters 5, 7 and 9) leads to a “constant”
nuclear embedding of A ⊗ O2 into B itself that realizes the action Ψ.
With other words, under the above assumptions on A, B and ΨA , the action ΨA
up
of Prim(B) on A can be realized as h−1
0 ΨB,h0 (A) for some non-degenerate nuclear *-
monomorphism h0 : A → B with h0 ⊕ h0 unitarily homotopic to h0 . Moreover, then
the nuclear *-monomorphism h0 is uniquely determined, up to unitary homotopy,
by ΨA and the property that h0 ⊕ h0 is unitarily homotopic to h0 .
The existence and uniqueness of h0 up unitary homotopy immediately implies
the following Corollary L of Theorem K :

Corollary L. Suppose that A and B are separable stable nuclear C*-algebras

and that γ is a topological isomorphism from Prim(A) onto Prim(B).
Then there is a *-isomorphism ϕ from A ⊗ O2 onto B ⊗ O2 that “realizes” γ ,
in the sense that ϕ(J ⊗ O2 ) = γ(J) ⊗ O2 for primitive ideals J of A.
The isomorphism ϕ with this property is unique up to unitary homotopy.

MOVE TO PROOF of?

Notice, that the uniqueness up to unitary homotopy quoted in Corollary L is
an “elementary” fact because [ϕ1 ⊕ ϕ2 ] = [ϕk ] in R(Prim(A); A ⊗ O2 , B ⊗ O2 ) for
k ∈ {1, 2} for any two γ-equivariant ϕ1 , ϕ2 .
62 1. INTRODUCTION AND MAIN RESULTS

Another easy consequence of Theorem K is Corollary 12.3.1, that implies that

for every separable exact C*-algebra A there exists a unique separable stable nu-
clear C*-algebra B with the same primitive ideal space Prim(A) ∼ = Prim(B) and
∼
with B = B ⊗ O2 .
Such an algebra B and the homeomorphism γ from Prim(A) onto Prim(B) are
unique up to algebraical and topological isomorphisms.
By Theorem K and with this properties the C *-algebra B is unique up to
isomorphisms that fix the given homeomorphism γ from Prim(B) onto Prim(A).
There are γ-equivariant non-degenerate C *-morphisms from B into A and from B
into A.

We are now going to use the Theorem K and the there defined h0 : A → B in
the same way as we have used above Theorem A to obtain Theorem B.
For that we need to generalize Kasparov’s R KKG (X; ·, ·) functor in some di-
rection.

Definition 1.2.10. Let X be a point-complete ( 32 ) T0 -space with countable

basis, and suppose that ΨA and ΨB are monotonous actions of X on A and B.
By Hom(X; A, B) – or more precisely by Hom(Ψ; A, B) – we denote the Ψ-
equivariant C *-morphisms from A to B. If the multiplier algebra of B con-
tains a unital copy of O2 , then the unitary equivalence classes [h] of elements
h ∈ Hom(X; A, B) are preserved under Cuntz addition and they constitute in a
natural way an abelian semigroup.
By Homnuc (X; A, B) we denote the Ψ-residually nuclear C *-morphisms.
We define a Ψ-equivariant Kasparov module as a Kasparov (A, B)-module
E = (E, φ, F ) (see e.g. Definition 8.2.1) which has the additional property that φ is
ΨA -Ψup ◦ ΨB -equivariant. Or equivalently, for each e ∈ E, the completely positive
map
Ve : a ∈ A 7→ hφ(a)e, ei ∈ B
satisfies Ve (ΨA (U )) ⊆ ΨB (U ) for all open subsets U ∈ O(X). The class of Ψ-
equivariant Kasparov (A, B)-modules is closed under direct sums of Kasparov mod-
ules and under unitary equivalence.
Let ∼scp denote the equivalence relation in the (A, B)-Kasparov modules given
by unitary isomorphisms and compact perturbations (of F , cf. Section 1 of Chapter
8). The equivalence classes build a commutative semigroup E(A, B)/ ∼scp , such
that KK(A, B) := Gr(E(A, B)/ ∼scp ).
If (E1 , φ1 , F1 ) ∼scp (E, φ, F ) and (E, φ, F ) is Ψ-equivariant, then (E1 , φ1 , F1 )
is so.
We define the Ψ-equivariant KK-group KK(X; A, B) as the Grothendieck
group of the sub-semigroup E(X; A, B)/ ∼scp of the ∼scp classes of Ψ-equivariant

32 Also called “spectral” or “sober ”.

 2. TOWARDS A CLASSIFICATION OF NON-SIMPLE PURELY INFINITE ALGEBRAS 63

(A, B)-Kasparov modules in E(A, B)/ ∼scp . The induced equivalence-relation on

the Ψ-equivariant (A, B)-Kasparov modules implies the equivalence ∼c of “cobor-
dism” or “homology” (but is build inside the semigroup of unitary equivalence
classes of Ψ-equivariant Kasparov modules, cf. Chapter 8).
Give precise references and def’s !!!
A Ψ-equivariant Kasparov module E = (E, φ, F ) is called Ψ-nuclear if for
every e ∈ E the completely positive map a 7→ hφ(a)e, ei from A to B is Ψ-residually
nuclear. If a Kasparov module is ∼scp equivalent to a Ψ-nuclear Kasparov module,
then it is itself Ψ-nuclear. The Ψ-nuclear ∼scp classes are closed (i.e. ”invariant”)
under direct sums and unitary equivalence.
Thus, the ∼scp equivalence classes of Ψ-nuclear Kasparov (A, B)-modules build
a sub-semigroup Enuc (X; A, B)/ ∼scp of the semigroup E(A, B)/ ∼scp of ∼scp
equivalence classes of Kasparov modules.
We define the Ψ-nuclear KK-group KKnuc (X; A, B) as the Grothendieck group
of the sub-semigroup Enuc (X; A, B)/ ∼scp of the semi-group E(A, B)/ ∼scp .

More generally, we consider in Chapters 8 and 9 the groups KK(C; A, B) :=

Gr(E(C; A, B)/ ∼scp ) where C ⊆ CP(A, B) is a point-norm closed operator-convex
cone of completely positive maps, and E(C; A, B)/ ∼scp ⊆ E(A, B)/ ∼scp denotes
the semigroup of ∼scp -equivalence classes of C-compatible Kasparov modules
(E, φ, F ) (i.e., of the modules with the property that for every e ∈ E the com-
pletely positive map a 7→ hφ(a)e, ei is in C). Notice that almost nothing is changed
in Kasparov’s theory except that we restrict the constructions to more special
classes of A-B-modules that produce automatically the desired equi-variances.
In this more systematic approach the old non-equivariant theory is still con-
tained and can be seen by considering KK(A, B) = KK(C; A, B) for C := CP(A, B),
and KK(X; A, B) = KK(C; A, B) for the m.o.c. cone C of with respect to O(X)-
action equivariant elements in CP(A, B) ...
An adaptation of the proofs in usual KK-theory shows that KK(X; ·, ·),
KKnuc (X; ·, ·) and the most general KK(C; A, B) behave analogously to KK:
There is a natural semigroup morphism

[h] ∈ [Hom(X; A, B)] 7→ [h − 0] := [Eh ] ∈ KK(X; A, B),

for stable separable A and B, where the Kasparov B-module Eh is given by Eh :=

(B, h, 0) and [h] denotes the unitary equivalence classes by unitaries in M(B).
The usual KK-groups can be rediscovered from our general approach by
KK(A, B) = KK({point}; A, B) .
One can prove the existence of Kasparov products x ⊗B y ∈ KK(X; A, C) for
x ∈ KK(X; A, B) and y ∈ KK(X; B, C) in the same way as for ordinary KK-theory.
There is a natural bi-additive map

KK(X; A, B) × KK(C, D) → KK(X; A ⊗ C, B ⊗ D),

64 1. INTRODUCTION AND MAIN RESULTS

that is induced by tensor products of modules (if C and D are nuclear), where the
action of X on the tensor products, e.g. on A ⊗ C, is given by
O(X) 3 Z 7→ ΨA (Z) ⊗ C ∈ I(A ⊗ C) .

If we fix X, then KK(X; A, B) is a bi-functor in the category of Ψ-equivariant

C *-morphisms. The bi-functor KK(X; ·, ·) is homotopy invariant for Ψ-equivariant
homotopy, has Bott periodicity in each variable, and has six-term exact sequences in
each variable on Ψ-equivariant (and Ψ-equivariant semi-split) short exact sequences
of separable algebras,
where we suppose that the actions Ψ of X on A and B satisfy some regularity
conditions expressed by the related m.o.c. cone C(X, ΨA , ΨB ; A, B).
Exactly which ‘‘regularity conditions’’,
Where mentioned?
If X is Hausdorff and the actions on A and B are continuous, then KK(X; A, B)
is the same as Kasparov’s R KKG (X; A, B) for the trivial group G = {e}, even if
differently defined.
The functor KKnuc (X; A, B) has similar properties, but KK(X; ·, ·)-equivalence
does not necessarily imply KKnuc (X; ·, ·)-equivalence. If A or B is nuclear, then
the natural morphism from KKnuc (X; A, B) into KK(X; A, B) is an isomorphism.
Consider e.g. X = {0, 1} with topology O(X) = {∅, {0}, {0, 1}}. Then the
elements of KK(X; A, B) give invariants for transformations between extension
0 → I → A → A/I → 0 into 0 → J → B → B/J → 0, if the Busby invariants of
the extensions dominate the zero representation (cf. Section 8 of Chapter 5).
We define in Chapter 5 weakly nuclear extension groups Extnuc (X; A, B) –
and more general Ext(C ; A, B) for countably generated operator convex cones
C ⊆ CP(A, B). The analog of Kasparov’s proof for the isomorphism Extnuc (A, SB)
and KKnuc (A, B) shows also that Extnuc (X; A, SB) is naturally isomorphic to
KKnuc (X; A, B), see Chapter 8.
If B is σ-unital and stable, then a kind of Weyl–von-Neumann–Voiculescu the-
orem for the class of weakly residually nuclear maps V : C ⊆ M(B) → M(B) holds
if and only if B has the WvN-property of Definition 1.2.3.
Moreover, Bω (respectively Q(R+ , B)) has the WvN-property, if and only if, B
is “strongly purely infinite” in the sense of Definition 1.2.2. Therefore all proofs,
mapping cone constructions and difference constructions carry over to this new
category, cf. Chapters 5, 7, 8 and the proofs of Theorems B and M in Chapter 9.
Now we make, for the next Theorem M, the following assumptions (1)-(3) on
A, B, N ⊆ B, X := Prim(N ) and the actions of X on A and B :

(1) Suppose that A is separable, stable and exact, B is σ-unital, and N is a

strongly purely infinite separable stable C*-subalgebra of B such that N B
is dense in B.
(2) Let X := Prim(N ), and let X act on B by ΨB := ΨN,B down .
 2. TOWARDS A CLASSIFICATION OF NON-SIMPLE PURELY INFINITE ALGEBRAS 65

(3) Further, suppose that X acts lower semi-continuously on A, such that,

moreover, ΨA (X) = A, ΨA (∅) = 0, ΨA (Z) = A implies Z = X, and such
that ΨA satisfies (ii) of Definition 1.2.6, i.e., the action ΨA of A on A is
non-degenerate and monotone continuous.

We can apply Theorem K to the action of Prim(N ) on A, and get a non-

degenerate nuclear *-monomorphism h0 : A → N ⊆ B, such that h0 is unitarily
homotopic to h0 ⊕ h0 , and h0 (ΨA (J)) = h0 (A) ∩ J for J ∈ I(N ). Note, that h0 is
determined by this properties up to unitary homotopy as a nuclear C *-morphism
from A into N .
Is the following intermediate action not contained
in the general type of C- or Ψ-equivariant actions?

Theorem M. Suppose that A, B, N ⊆ B, X, and the actions ΨB , ΨA satisfy

the above assumptions (1)-(3).
Let h0 : A → N ⊆ B be the *-monomorphism given by Theorem K, let [h] ∈
[Homnuc (X; A, B)] denote the unitary equivalence class of h, and let [h − 0] ∈
KKnuc (X; A, B) be the difference construction for h ∈ Homnuc (X; A, B).

(i) The map

α : [Homnuc (X; A, B)] 3 [h] 7→ [h − 0] ∈ KKnuc (X; A, B)

is a semigroup epimorphism.
(ii) [h − 0] = [k − 0] in KKnuc (X; A, B) if and only if h ⊕ h0 and k ⊕ h0 are
unitarily homotopic.
(iii) If, moreover, B is itself strongly purely infinite and ΨB is an isomorphism,
i.e., comes from an isomorphism of Prim(N ) onto Prim(B), then h ⊕ h0
is unitarily homotopic to h for every *-monomorphisms h : A ,→ B which
satisfies h(ΨA (Z)) = h(A) ∩ ΨB (Z) for open subsets Z ⊆ X.

Notice that ΨB is an isomorphism, if and only if, ΨB defines an isomorphism

of Prim(N ) onto Prim(B) – and is defined by such an isomorphism.
The proof of Parts (i) and (ii) will be given together with the proof of Parts
(i) and (ii) of Theorem B. But we assume there the existence of the nuclear *-
monomorphism h : A ⊗ O2 ,→ N with the properties listed in Theorem K for N (in
place of B). Theorem K will be proven independently in Chapter 12, based on a
partial result in Chapter 6. The important absorption result (iii) of Theorem M is
shown in Chapter 7.
KK(X; ·, ·) classifies all separable stable nuclear C *-algebras up to O∞ -stable
isomorphisms by the following generalization of Corollary C.

Corollary N. Suppose that A and B are separable, stable and nuclear C*-
algebras, and that there is a topological isomorphism γ from Prim(A) onto Prim(B).
Let ΨB denote the continuous action of X := Prim(A) on B defined by γ.
66 1. INTRODUCTION AND MAIN RESULTS

If z ∈ KK(X; A, B) is a KK(X; ·, ·)-equivalence between A and B, then there

is a Ψ-equivariant isomorphism ϕ from A ⊗ O∞ onto B ⊗ O∞
It seems only true that there are X-equivariant *-monomorphisms
h : A → B and k : B → A such that k ◦ h is unitarily homotopic to idA
and h ◦ k is unitarily homotopic to idB such that [ϕ − 0] = z.
There is only one isomorphism ϕ with this property up to unitary homotopy.

The proof follows from Theorem M in the same way as the proof of Corollary
C follows from Theorem B (see the arguments above Corollary C).
As in the case of simple nuclear algebras, the C *-algebras A and B absorb O∞
tensorial if A and B are moreover strongly p.i.
BEGIN: discussion of homotopy invariance.
The example in detail should be moved to some suitable chapter.
It should be noted that homotopy invariance of the isomorphism classes of
strongly purely infinite separable stable nuclear C *-algebras does not follow, even
if we assume in addition that the ideal structure is preserved on the way of the ho-
motopy and can be “corrected” to an ideal system fixing path of full endomorphisms
by adding the non-degenerate “zero” endomorphism h0 to this path.
This can be illustrated transparently by separable nuclear s.p.i. algebras that
??????
For example, consider the half-open interval X0 := (0, 1]lsc with the “lower
semi-continuity”-describing T0 -topology given by the intervals (r, 1] (r ∈ [0, 1))
and ∅ as open sets, see also [359, thm. 1.4].
By [559] (see also [689]), there is a canonical way to produce a separable nuclear
C *-algebra C with Prim(C) = X0 = (0, 1]lsc that is “locally purely infinite” in a
rather weak sense of Definition 2.0.3.
It implies by Corollary 2.6.6 that C is strongly p.i. , which implies C ∼
= C ⊗ O∞
by Theorem ??.
There exist separable nuclear C *-algebras D with Prim(D) := (0, 1]lsc that are
not purely infinite. Since C is separable, purely infinite and has no unital quotient
it is stable. Thus, there is a unital copy of C ∗ (s1 , s2 ) ∼
= O2 in its multiplier algebra
∼ ∼
M(C) . This shows C = C ⊗ O∞ = C ⊗ O∞ ⊗ K, and that there exists an ideal
system preserving non-degenerate *-monomorphism h0 : C ⊗O2 ∼ = C ⊗O2 ⊗K → C,
by using a non-degenerate C *-morphism O2 ⊗ K → O∞ ⊗ K .
Question: Are all approximate to id unitary equivalent isomorphisms of sepa-
rable K ⊗ A ⊗ O2 unitary homotopic to id?
Here the question could be how near the unitaries Un∗ Un+1 of such approxi-
mate unitary equivalence defining unitaries can be taken to the unitary group
U(A ⊗ O2 + 1M(A) ⊗ O2 ).

This group of isomorphisms is not studied well until yet.

 2. TOWARDS A CLASSIFICATION OF NON-SIMPLE PURELY INFINITE ALGEBRAS 67

We can build the Cuntz sum idC ⊕s1 ,s2 h0 : C → C that is unitarily homotopic
to idC because idC asymptotically dominates h0 by Corollary ??.
This unitary homotopy – in sense of Definition 5.0.1 – defines a point-norm
continuous path ψt : C → C with ψ0 = id and ψt (c) = u(t)∗ cu(t) and ψ1 (c) =
idC ⊕h0 . Moreover, since C is σ-unital and stable, there is a norm-continuous path
of unitaries t ∈ [0, 1) 7→ u(t) ∈ M(C) such that limt→1 u(t)∗ s2 cs∗2 u(t) = c for all
c ∈ C, i.e., idC and 0 ⊕s1 ,s2 idC are unitarily homotopic, in particular 0 ⊕s1 ,s2 h0 is
ideal system preserving homotopic to h0 .
If we combine the above defined paths of endomorphisms then we find a *-
monomorphism from ϕ : C → C([0, 1], C) that is ideal system preserving, thus
defines an element of KK(Prim(C) ; C, C([0, 1], C)) and satisfies π0 ◦ ϕ = idC and
π1 ◦ ϕ = h0 . If we apply the above arguments in a bit more general situation (with
obvious generalization), we get the following corollary of the classification:

Corollary 1.2.11. If A is a separable nuclear strongly purely infinite C*-

algebra that admits a ideal system preserving zero homotopy, then A is stably iso-
morphic to A ⊗ O2 ⊗ K, i.e., A ⊗ K ∼ = A ⊗ O2 ⊗ K.

It follows that the homeomorphism class of the primitive ideal space is the only
invariant of the isomorphism class of A ⊗ O2 ⊗ K. And it is not difficult to see that
Prim(A) must by homeomorphic to (0, 1]lsc × Prim(A).
Indeed, let a1 , a2 , . . . a dense sequence in the positive contractions of A and
P −n
let b := n2 an . Then, clearly, kπI (b)k ≤ kπJ (b)k for J ⊆ I. The equation
kπI (b)k = kπJ (b)k implies that J = I.
Give a transparent proof:
The latter, because in case J 6= I (and J ⊂ I) there exists a pure state ρ on A
with ρ(J) = {0} but ρ(I) 6= {0} ...
(All non-empty open sets are prime and there is a countable sequence of open
subsets Vn of Prim(A) that build a base of the topology of Prim(A). If we take for
each of this a strictly positive contraction an ∈ Jn for the corresponding ideal and
let b := n 2−1 an . If kπJk (b)k = kπJ` (b)k and Jk ⊆ J` ...????
P

Check cite/ref and explain

A modification of the construction in ???
[816, ????] ?? ????
It seems not to be the right class
Perhaps Rørdam’s paper [689] seems to be nearer to the idea.
reference to his Israel J. Math. paper?
allows to construct C ∼ = C ⊗ O∞ in a manner, that it has an explicit ideal-
preserving zero-homotopy, this is a C *-morphism γ : C → C((0, 1], C) with γt ∼=
πt ◦ γ : J(s, 1] → J(ts, 1],
where we ?????????????? and ???????????
68 1. INTRODUCTION AND MAIN RESULTS

T
Notice that J(t, 1] ⊂ J(s, 1] and t∈(0,1] J(t, 1] = {0}, i.e., there is a point-norm
continuous path {γt ; t ∈ [0, 1]} of endomorphisms γt : C → C such that γ1 = id,
γ0 = 0 and γt (J) ⊆ J for every closed ideal J of C and every t ∈ [0, 1], cf. [464,
prop. 6.1].
Give the path explicit ?!!
Check cite !! ??
It follows that the (non-zero) strongly purely infinite algebra A := C ⊕ (O2 ⊗K)
is homotopic to B := O2 ⊗ K in a Ψ-equivariant manner,
but the algebras have primitive ideal spaces that are not homotopic.
Check: Are Z := X0 ] {−1} with ‘‘open’’ {−1} and the space {−1}
NOT homotopic??
Let ψ(t)(s) := t · s and ψ(t)(−1) = −1.
Is each ψ(t) continuous on Z?
Is ψ(t) globally continuous?
Is X0 via ψ(t) homotopic to ∅?
Clearly, this phenomenon can not appear if A and B are replaced by simple
(p.i.) C *-algebras.
Now let, more generally, C an arbitrary strongly purely infinite stable separable
nuclear C *-algebra (in particular C ∼ = C ⊗ O∞ ⊗ K), and let γt : C → C an ideal
system preserving homotopy between id and 0 (like in the special case above). Then
we can (Cuntz-)add to the morphisms γt the mono-morphism h0 , and get that γt ⊕
h0 : C → C defines a Ψ-equivariant homotopy from idC ⊕h0 to the monomorphism
0⊕h0 . In this special case it is equivalent to the property that (γt ⊕h0 )(J) generates
J for each J ∈ I(C).
They must define the same element of KK(X; C, C) by homotopy invariance
of KK(X; ·, ·). Both satisfy the assumptions of Part (iii) of Theorem M. It follows,
that h0 : C → C is unitarily homotopic to idC , which implies that

C 3 c 7→ c ⊗ 1 ∈ C ⊗ O2

is an KK(X0 , ·, ·)-equivalence. Thus, C 3 c 7→ c ⊗ 1 ∈ C ⊗ O2 is approximately

unitary equivalent to an isomorphism from C onto C ⊗ O2 . This shows:
If a strongly purely infinite separable nuclear C*-algebra C admits an ideal-
system preserving zero homotopy, then C absorbs O2 tensorial, i.e., C ∼
= C ⊗ O2 .
See the paper [464] for a more detailed study of this class of nuclear C *-
algebras:
It turns out that they are all AH-algebras with building blocks Mn (C0 (P, p∞ ))
where (P, p∞ ) are certain pointed one-dimensional finite CW-complexes P with
each point connected to p∞ ∈ P .
Give precise conditions on P !
 2. TOWARDS A CLASSIFICATION OF NON-SIMPLE PURELY INFINITE ALGEBRAS 69

The converse does not hold, e.g. O2 ⊗ K tensorial absorbs O2 but is not homo-
topic to zero: No (non-zero) simple C *-algebra A is zero-homotopic, [175], because
A ⊗ O2 has real rank zero for simple A, cf. Theorem E.
It follows that one has necessarily to consider a sort of more special Ψ-
equivariant homotopy, that ensure that also the corresponding primitive ideal
spaces are homeomorphic.
Definition 1.2.12. Suppose that ΨA : O(X) → I(A) and ΨB : O(X) → I(B)
are actions of a T0 space X on A respectively B.
Let k0 , k1 : A → B be ΨA -ΨB –equivariant C *-morphisms (respectively Ψ-
residually nuclear C *-morphisms, respectively k0 , k1 ∈ C ⊆ CP(A, B)).
We say that k0 and k1 are Ψ-homotopic ,
if ????? and ??? how to avoid singularities?,
if there is is a point-norm continuous path [0, 1] 3 t 7→ γt in the ΨA -ΨB –
equivariant C *-morphisms (respectively Ψ-residually nuclear C *-morphisms, re-
spectively γt ∈ C is C-compatible for C ⊆ CP(A, B) ) γt : A → B such that γ0 = k0
and γ1 = k1 .
The family {γt } is an equivariant homotopy, respectively is a C-compatible
homotopy.
The Ψ-equivariant homotopy γt between k0 and k1 will be called full if, for
each U ∈ O(X), γt (ΨA (U )) generates for every t ∈ [0, 1] the same ideal Φ(U ) :=
Bk0 (ΨA (U ))B ⊆ ΨB (U ) of B .

Check homotopy statements

above and from here to next label ! : ??
The homotopy invariance of KK(X; A, B) and of KKnuc (X; A, B) ensures
that Ψ-homotopic k0 and k1 have the same KK(X; A, B)-class, respectively same
KKnuc (X; A, B)-class.
On the other hand, we can find (by Corollary N) a Ψ-equivariant isomorphism
from A ⊗ O∞ onto B ⊗ O∞ , if we consider only those homotopy equivalences
which are given by h : A → B, k : B → A, p(t) : A → A and q(t) : B → B, which
satisfy kh = p(0), p(1) = idA , p(t)(A) ∩ J = p(t)(J) and Ap(t)(J)A dense in J for
every J ∈ IA and every t ∈ [0, 1], and hk = q(0), q(1) = idB , q(t)(B) ∩ J = q(t)(J)
and Bq(t)(J)B is dense in J for every J ∈ IB and every t ∈ [0, 1]. It means that
we require that p(t) induces the identity map of the lattice O(Prim(A)) ∼ = I(A) for
each t ∈ [0, 1] (and same for q(t) and O(Prim(B))).
The latter (strong) non-degeneracy condition for an Ψ-equivariant homotopy
can be arrived by replacing h by h ⊕ hA,B
0 (respectively k by k ⊕ hB,A
0 , pt by pt ⊕ h0
and qt by qt ⊕ h0 ).
This strong conditions on the homotopy imply in particular that Prim(A) and
Prim(B) are homeomorphic. If then h : A → B and k : B → A define an equivariant
homotopy with respect to the natural actions of Prim(B) (respectively of Prim(A))
70 1. INTRODUCTION AND MAIN RESULTS

of the above stronger type, then [h] ∈ KK(Prim B; A, B) is a KK(Prim B; ·, ·)-

equivalence, and h ⊕ h0 is unitarily homotopic to an isomorphism from A onto B
(33).
Check above formula! And next statement on ??????
Thus, a strongly p.i. separable stable nuclear C*-algebra A is isomorphic to
A ⊗ O2 if it is homotopic to zero in a Prim(A)-equivariant manner.
(For example, C ∼= C ⊗ O2 if C is the above considered algebra with ideal
system preserving zero homotopy). The converse does not hold as the example
A := O2 ⊗ K shows. But one has, for s.p.i. separable nuclear stable algebras A,
that A ⊗ O2 ∼
= A if and only if KK(Prim(A); A, A) = 0. (Use almost verbatim the
same arguments as for Part (iv) of Corollary F.)
Partly mentioned further above? What? ??
END: discussion of homotopy invariance.

Corollary L tells us nothing about the structure of the primitive ideal spaces of
separable nuclear C *-algebras A. A pure topological description of the primitive
ideal spaces of nuclear C *-algebras A is given in [359], that uses results of [464]
and of Chapters 3 and 12.
It would be desirable to know under which conditions on A the algebra D =
A ⊗ O2 ⊗ K is the inductive limit of C *-algebras C0 (Yn , O2 ⊗ K) for Polish l.c.
spaces Yn . This is the case, e.g. , if I(A) is linearly ordered (by inclusion of ideals).
Unfortunately there are also examples where D can not be expressed as such
an inductive limit, e.g. , if A := {f ∈ C([0, 1], M2 ) : f (1) ∈ ∆} where ∆ denotes
the subalgebra of diagonal matrices in M2 . (In fact, for the last example A, the
algebra D can not be any inductive limit of C *-algebras with Hausdorff primitive
ideal spaces, cf. Remark B.12.1.)
All strongly purely infinite separable nuclear C *-algebras A that admit an ideal-
system preserving homotopy beween idA and 0 satisfy A ∼ = A ⊗ O2 ⊗ K and are
inductive limits of algebras A1 ⊆ A2 ⊆ · · · with An isomorphic to C0 (Γ, p) ⊗ Mkn ,
where Γ is a finite connected graph and C0 (Γ, p) denotes the algebra of continuous
functions on Γ that vanish at a distinguished point p ∈ Γ, cf. [464].

It is still an open question whether, for every separable stable nuclear C *-

algebra B, the C *-algebra D := B ⊗ O2 is an inductive limit of C *-subalgebras of
D that are isomorphic to C *-algebras A ⊗ O2 , where A is of type I.
One can show that D is a crossed product of an inductive limit of algebras of the
form C0 (Pn \{qn }, Mn ) with Z, where Pn denotes a connected finite one-dimensional
polyhedron and qn is a point of Pn , cf. [464].
It implies that for every nuclear separable C *-algebras A the algebra A ⊗ O2
contains an Abelian C *-subalgebra C ⊂ A ⊗ O2 that is “regular” in the sense of
Definition 1.2.9.
33 Notice here that hA,B ◦ hB,A is unitarily homotopic to hA,A .
0 0 0
 2. TOWARDS A CLASSIFICATION OF NON-SIMPLE PURELY INFINITE ALGEBRAS 71

It follows that for the T0 space Y := Prim(A) of a separable exact C *-algebra A

there exist a locally compact Polish space X ( 34 ) and a continuous map ψ : X → Y
that is “pseudo-open” and “pseudo-surjective” in the sense of Definitions ?? and
??. In fact, then C := C(X) can be embedded by Theorem ?? into A ⊗ O2 such
that ϕ : C → A ⊗ O2 induces ψ −1 .
Conversely, for any locally quasi-compact sober T0 space Y with the property
that such a continuous map ψ : X → Y exists for some locally compact Polish space
X there exists a separable nuclear C *-algebra A such that Prim(A) ∼
= Y by [359,
thm. 1.4].

Next must be changed drastically, because of New results ??

Which New results?? Give Ref’s or Cite !!!
Let us consider an other aspect of the the “UCT”:
M. Dadarlat published in [193, sec. 3] an example (based on an observation
in [369]) of a continuous field (Ax )x∈Q of C *-algebras Ax over the Hilbert cube
Q := [0, 1]∞ that has fibers Ax ∼= O2 and that its C *-algebra of continuous sections
A has non-trivial K1 (A) 6= 0. If one tensors with P∞ then one gets an example
with Ax ∼= O2 but K0 (A) 6= 0.
In particular, (Ax )x∈Q can’t be locally trivial. One can see (e.g. from the
construction of Dadarlat) that the nuclear algebra A is strongly purely infinite
(i.e., absorbs O∞ tensorial) and satisfies the UCT.
Let (Ax )x∈Q an arbitrary continuous field over Q := [0, 1]∞ with all fibers
isomorphic to O2 .
Is the C *-algebra A of continuous sections in the field (Ax )x∈Q always in the
UCT-class?
This kind of algebras A are all “locally” purely infinite in the sense of Definition
2.0.3. Are this algebras A purely infinite (weakly purely infinite, strongly purely
infinite)? Counterexamples? What are reasonable invariants to distinguish them?
It leads to the question of classification of the isomorphisms from O2 ⊗ O2 onto O2
by a suitable “classifying space”.
Next comes some old stuff. Partly to be removed!! ??
Let Y a (not-necessarily Hausdorff) sober T0 -space. We use for an action
Ψ : O(Y ) → I(A) of a topological space Y on A the notation A|Z := Ψ(U )/Ψ(U ) ∩
Ψ(U \ Z) if Z is a closed subset of an open subset U ⊆ X. We say that X acts on A
continuously if Ψ is a lattice monomorphism and is both “upper semi-continuous”
and “lower semi-continuous” in the sense of Definitions ??.

34 i.e., X is a locally compact second countable Hausdorff space, or equivalently expressed:

X is homeomorphic to a separable complete metric space that is in addition locally compact. The
complete metric is not uniquely determined by this property of X.
72 1. INTRODUCTION AND MAIN RESULTS

Theorem O. Let X a second countable locally quasi-compact point-complete

T0 space, and let G a second countable l.c. group that acts continuously on X by
µ : G → Homeo(X).
Then holds:
If X acts continuously on a stable separable nuclear C*-algebra A and α : G →
Aut(A) is an action on A with α(g)(Ψ(U )) = Ψ(µ(g)(U )) for all g ∈ G, then
there exists a unique separable stable purely infinite nuclear C*-algebra B with a
continuous action β : G → Aut(B), an isomorphism λ : X → Prim(B) from X
onto Prim(B), an X-equivariant monomorphism ϕ : A → B and a new action
α0 : G → Aut(A), such that
α0 is (exterior equivalent ?? or weakly?? outer) 1-cocycle conjugate to α ?
only 2-cocycle conjugate ????
??
ϕ ◦ α0 (g) = β(g) ◦ β(g)
β(g)(B|λ(U )) = B|λ(µ(g)(U ))
[ϕ] ∈ KK(X; A, B) is a KK(X; ·, ·)-equivalence.
B is determined up to unitary homotopy and X-equivariant isomorphisms.
Moreover, TFAE:

= X, and A/J ∼
(i) If A is nuclear and separable, Prim(A) ∼ = (A/J) ⊗ O2 for
every primitive ideal, then A ∼
= A ⊗ O2 .
(ii) If X acts on separable nuclear C*-algebras A and B continuously,
and if ψ : A → B is an X-equivariant *-monomorphism such that,
for each x ∈ X, the induced morphism A|{x} → B|{x} defines a
KK({x}; A|{x}, B|{x}) equivalence,
then ψ defines a KK(X; A, B) equivalence.
(iii) more? dim(X) < ∞?
(iv) more?

In the meantime Dadarlat [192] has shown that A ∼ = C0 (X, O2 ) (respec-

∼
tively A = C0 (X, O∞ )) if A is separable and unital, X := Prim(A) is a a finite-
dimensional Hausdorff space and if every primitive quotient of A is isomorphic to
O2 (respectively O∞ ).
Is some local triviality needed?
He shows also that O2 and O∞ are the only pi-sun algebras B with the property
∼
A = C(X, B) for all compact metric spaces X of finite dimension, and all C(X)-
algebras A with fibers Ax ∼
= B.
(We did not check if here only the UCT class has been studied !!!)
It follows that X satisfies (i) of Theorem O if X has a decomposition series Xγ
with Xγ+1 \ Xγ is a finite-dimensional Hausdorff space,
 3. BACKGROUND AND SOME BASIC IDEAS IN PROOFS 73

Def. of strong UCT-class for KK(X; ·, ·) given by suitable actions of X

on Polish l.c. spaces Y or on separable type one C *-algebras.

HERE we should also list the new defined basic properties for the classification
and its terminology.
Riesz decomposition property, ”strongly self-absorbing” C*-algebras, Com-
pletely positive maps of order zero, QDQ vs. UCT, Decomposition rank (of
subhomogenuous C*-algs), pure C*-algebras, Nuclear dimension, Covering di-
mension for nuclear C*-algebras, locally finite decomposition rank, Controlled
KK-theory, property A, Watatani Index for C*-algebras, slice maps, nonc. Weyl -
von Neumann theorem, Elliott invariant, Flat dimension growth for C*-algebras,
Nonstable K-Theory, corona factorisation property,

3. Background and some basic ideas in proofs

The conceptional background of our proofs of Parts (i) and (ii) of Theorems B
and M can be described as follows:
If we have “actions” ΨA : O(X) → I(A) and ΨB : O(X) → I(B) of spaces X
on C *-algebras A and B, then we can select suitable matrix operator-convex cones
C ⊆ CP(A, B) of c.p. maps V that are equivariant with respect to the actions, i.e.,
V (ΨA (U )) ⊆ ΨB (U ) for U ∈ O(X), and satisfy like-wise additional conditions,
e.g. that V is residually nuclear for the ideal systems selected via the actions ΨA
and ΨB .
This procedure is almost equivalent to a functorial selections of subsemigroups
of the semigroups of unitary equivalence classes of Kasparov A-B-modules.
We define in Chapters 5 and 8 groups Ext(C; A, B) and KK(C; A, B),
depending on non-degenerate operator-convex cones C ⊆ CP(A, B) in a func-
torial way, such that e.g. KK(A, B) = KK(CP(A, B); A, B) or KKnuc (A, B) =
KK(CPnuc (A, B); A, B) if A is separable and B is σ-unital (see Chapters 5, 8, 9).
Our extension groups Ext(C; A, B) will be defined by relations that are similar
to relations given for KK1 (A, B) ∼= Ext(A, B) for trivially graded stable σ-unital
A and B, the difference is that the generalized morphisms (ψ, P ) of the Cuntz-
Kasparov picture – cf. [73, 17.6.4] – are required to fulfill the extra condition that
b∗ ψ(·)b ∈ C for each b ∈ B. We prove and apply that

KK(C; A, B) ∼
= Ext(C(R); A, C0 (R, B))

to bring C-compatible asymptotic morphisms and KK-groups together.

The functorial construction (C; A, B) 7→ KK(C; A, B) is homotopy in-
variant inside the – chosen/allowed – matricial operator-convex cones (de-
fined and partly studied in Chapter 3) and there is a natural isomorphism
Ext(C(R); A, C0 (R, B)) ∼
= KK(C; A, B), where C(R) ⊆ CP(A, C0 (R, B)) denotes
the operator-convex cone of c.p. maps V : A → C0 (R, B) with V (·)(t) ∈ C for all
74 1. INTRODUCTION AND MAIN RESULTS

t ∈ R. Moreover, Ext(C(R); A, C0 (R, B)) is naturally isomorphic to the kernel of

K1 (πB (H0 (A))0 ∩ Qs (B)) → K1 (Qs (B)) .

(Here H0 : A → M(B) is a non-degenerate *-morphism with infinite repeat δ∞ ◦ H0

unitarily equivalent H0 , such that the c.p. maps A 3 a 7→ b∗ H0 (a)b ∈ B are dense
in C.)
The results will be used in Chapter 9 to prove with help of Theorem A (respec-
tively Theorem K) the following result that is more general than Parts (i) and (ii)
of Theorems B and M :
Let A and B stable C *-algebras, where A is separable and B is σ-unital, and
let h0 : A ,→ B a non-degenerate *-monomorphism such that h0 ⊕ h0 = h0 . Let
C ⊆ CP(A, B) denote the point-norm closed matrix operator-convex cone generated
by h0 . Then,
the difference construction h 7→ [(B, h, 0)] =: [h − 0] defines an additive map from
Hom(A, B) ∩ C onto KK(C; A, B) such that [h − 0] = [k − 0] if and only if h ⊕ h0
and k ⊕ h0 are unitarily homotopic.
Here the addition on Hom(A, B) is given (up unitary equivalence in M(B)) by the
Cuntz addition with a copy of O2 that unitally contained in M(B).
If we use the non-degenerate monomorphism h0 : A ⊗ K → B ⊗ K, from The-
orems A and K, then we get Parts (i) and (ii) of Theorems B and M. With this
general result in hand, the hard work reduces to the proof of Theorems A and K.
On the way to the proofs of Theorems A and K we obtain and use (among
others) a sufficient criteria for existence of an X-equivariant lift:

Next must be a Prop/Thm in Chap. 5. Check! ?? Suppose that A and

B are stable C *-algebras, where A is separable B is σ-unital, and that H0 : A ,→
M(B) and ϕ : A ,→ Q(B) = M(B)/B are nuclear *-monomorphisms with the
properties that H0 (A)B is dense in B and, for every a ∈ A, πB (H0 (a)) and ϕ(a)
generate the same closed ideal of Q(B). Then, there is a unitary U ∈ M(B) such
that ϕ = πB (U ∗ H0 (·)U ), if one of the following conditions (i) or (ii) is satisfied:

(i) There exists a C *-algebra C such that B ∼

= C ⊗ O2 ⊗ O2 ⊗ · · · , or
−1
(ii) πB (ϕ(A)) is stable, and there are C *-morphisms H : A ⊗ O2 ,→ M(B)
and Φ : A ⊗ O2 ,→ Q(B) = M(B)/B with H0 (a) = H(a ⊗ 1) and ϕ(a) =
Φ(a ⊗ 1) for a ∈ A.

It has to be emphasized that much of the machinery needed for this work was
already present in the literature (at least implicitly), especially done in the works
of J. Cuntz, J. Glimm, G. Kasparov and D. Voiculescu until mid of the eighties of
last century (cf. Chapters 2, 4. 5 and 8). Our viewpoints are inspired by early work
of Elliott and Rørdam on the classification of Cuntz-Krieger algebras.
Our proofs of Theorem A and of Theorem B need old results of J. Cuntz
concerning the realization of K0 and K1 for purely infinite C *-algebras in [172],
 4. ON J. ELLIOTT’S CLASSIFICATION PROGRAMM AND CONJECTURES 75

our characterization of exact C *-algebras [438], some basic results of Kasparov

theory [405], and improvements of Kasparov’s Proof by Cuntz and Skandalis and
others, see [73] and the “nuclear” variant of Kasparov theory given by Skandalis
[726].
Additional background basic material can be found in textbooks [73], [616],
[810], and in the articles [172] and [726].
Generalizations of results from Glimm’s early work are important technical
tools. This gives a way of approximating nuclear maps by maps of the form
x 7→ c∗ xc (under certain additional assumptions). It leads us in Chapter 5 to a
generalized Weyl–von-Neumann–Voiculescu theorem, by modifying and using ideas
of Kasparov [404].
The results of Chapters 2, 3, 4, 5 and 7 are fundamental and each is crucial
for the proofs in the following Chapters. Chapter 6 contains the proof of Theorem
A. Its method will be explored to prove Theorem 6.3.1, which is a special case of
Theorem K, but Theorem K without any extra technical assumption will be finally
proved in Chapter 12 with help of a sort non-commutative variant of the Michael
selection theorem.

4. On J. Elliott’s Classification Programm and Conjectures

Section to be rewritten/shorten:
Short cut, and outline some new results.
But only as summary e.g. following ICM lectures?
Don’t forget the stable projection-less case
My (in-adequate?) remarks could be moved in Appendix A or B?
The below formulated Elliott classification conjecture is now verified for all
simple separable nuclear C *-algebras A that tensorial absorb the Jiang-Su algebra
Z and satisfy the Universal Coefficient Theorem (UCT) for its KK-theory. The
latter means that there exists a separable commutative C *-algebras C that is KK-
equivalent to A.
The classification of the ”elementary” cases, e.g. K∗ (A) = 0 could perhaps also
replace the UCT by other asymptotic equivalence..???
But it gives not an essentially simpler proof for the above considered case of
(simple) pi-sun C *-algebras. And the methods used in the case where e.g. A is
stably projection-less is a bit away from the methods developed and used here. In
particular, stably projection-less simple nuclear C *-algebras have automatic one or
more quasi-traces (cf. E.K. paper cited ???), and are in a sense the opposite of the
pi-sun C *-algebras.
The complete data are the ordered K-theory, – with K0 pointed and scaled by
a possible unit element –, and the cone of l.s.c. traces and its pairing with K0 .
76 1. INTRODUCTION AND MAIN RESULTS

It seems that the case of stably projection-less C *-algebras is not completely

studied, e.g. it could be that most of them are mostly crossed products of stable
purely infinite algebras by some R-action.
An inspiring role for this development of this final results was the idea of
“decomposition rank” and to use the “balancing” role of tensorial absorption of the
Jiang-Su algebra Z, that spells out many in early study unexpected difficulties. The
cases of AF-algebras, special study on AH-algebras and ASH-algebras did pave the
way, also the in this book considered special case of nuclear purely infinite algebras
related to graphs did play an inspiring role on the way to the success until now.
Partial results on stably projection-less nuclear C *-algebras do also exist, but
have until now no systematic description of all possible cases with given trace cone,
or has not appeared until now. For example, stably projection-less C *-algebras
A remain stably projection-less if we tensor them with the Jiang-Su algebra Z
(or other ASH-algebras). All simple separable nuclear stably projection-less C *-
algebras have densely defined (e.g. on the positive part of the Pedersen algebra)
additive traces that play a role for the classification.
It should be emphasized that the, up to tensorial Z stably isomorphic, classes
of simple nuclear C *-algebras A can have different K0 (A)+ , e.g. the embedding
a 7→ a ⊗ 1 defines a natural isomorphism K∗ (A) ∼ = K∗ (A ⊗ Z) but the ordered
“cones” K∗ (A)+ and K∗ (A⊗Z)+ of the stable equivalence classes of idempotents in
(A⊗K)+ (respectively in (A⊗Z ⊗K)+ ) are in some cases different. This can happen
also for simple ASH-algebras A, because there are examples of unital simple ASH-
algebras where the ordered semigroup K0 (A ⊗ K)+ is not “weakly unperforated”,
but it is known (!!!!! find citation / reference !!!!!) that K0 (A ⊗ Z ⊗ K)+ is weakly
unperforated for every C *-algebra A.
It has to do – and in the “locally” purely infinite case that is our minimal
assumption – with properties of the “large” version of the Cuntz semi-group Cu(A),
that reduces to the set of Dini-functions on Prim(A) in the cases that A ∼
= A ⊗ O∞
as it happens in the case that we consider.
We restate a special case of the original Classification Conjecture of G. El-
liott in a way that allows the reader to see, how our viewpoints and the results
of N.Ch. Phillips and the author in the special case of separable simple unital
amenable C *-algebras are related to it, and which questions – namely (Q1) and
(Q2) concerning the classification of simple purely infinite algebras remain open.
It is possible that one has answers if one requires additional properties, as
e.g. the CFT property or infinitesimally properties.
Before we begin with this we want to acknowledge with congratulation the
important almost final result for the classification of unital simple C *-algebras:
If A1 , A2 are a separable unital simple nuclear C *-algebra that are KK-
equivalent to separable Abelian C *-algebra C1 , C2 and tensorial absorb the
 4. ON J. ELLIOTT’S CLASSIFICATION PROGRAMM AND CONJECTURES 77

Jiang-Su algebra Z in the sense Ak ⊗ Z ∼ = Ak , then A1 and A2 are isomorphic if

and only if A1 and A2 the same Elliott invariant (as defined below ??).
Since A and A⊗O∞ are isomorphic for all strongly purely infinite separable nu-
clear algebras (and since for simple A pure infiniteness and strong pure infiniteness
are the same) A ∼= A ⊗ Z follows from O∞ ⊗ Z = O∞ .
(But I do not know if also for the stably projection-less separable simple nuclear
C *-algebras A with A ∼= A ⊗ Z the classification in the UCT class becomes now
complete.
An interesting question is if they all come from crossed products AoR of stable
pi-sun algebras A by an action of the real numbers R, or as crossed product by an
action of the semigroup R+ = [0, ∞), or by (0, ∞).
Next is a bit out of discussion and perhaps not related.
We define a Z2 -graded group K∗ (A) := K0 (A) ⊕ K1 (A), and a pre-order on
K0 (A) by the subsemigroup K0 (A)+ := {[p] ∈ K0 (A) : p ∈ Proj(A ⊗ K)} .
Let A be a C *-algebra and let T + (A) denote the locally compact cone of non-
negative (additive) traces on the Pedersen ideal of A⊗K. There is a natural pairing
h ·, · i between K0 (A) and T + (A), defined by hz, τ i := τ (p − q) ∈ R for z ∈ K0 (A)
and τ ∈ T + (A).
Is next correct? Here p, q are projections in A e ⊗ K such that z = [p] − [q],
p − q is in the Pedersen ideal of A ⊗ K, and A is the unitization of A.
e

Suppose that A and B are simple, separable and nuclear C *-algebras, and
that σ0 : K0 (A) → K0 (B) and σ1 : K1 (A) → K1 (B) are group isomorphisms from
the K-groups of A onto the K-groups of B such that σ0 (K0 (A)+ ) = K0 (B)+ . Let
σ T : T + (B) → T + (A) be a continuous additive isomorphism from T + (B) onto
T + (A), such that they have the coherence properties hz, σ T (τ )i = hσ0 (z), τ i for all
z ∈ K0 (A) and τ ∈ T + (B).
If σ1 , σ2 and σT are given, then a variant of the Conjecture of Elliott Cite
source for Elliott invariant !!! can be reformulated in an equivalent way
as follows:
There exists an isomorphism ψ from A ⊗ K onto B ⊗ K such that

σ∗ = K∗ (ψ) and σ T (τ ) = τ ◦ ψ ∀ τ ∈ T + (B) .

If A and B are unital and σ0 ([1]) = [1] then one requires in addition that the
isomorphism σ can be chosen such that, ψ(1 ⊗ p11 ) = 1 ⊗ p11 .
If one of the algebras A or B is not known to be of real rank zero, then sometimes
additional invariants have been introduced to verify the Elliott conjecture even only
in very special cases. The now most prominent one is tensorial absorption of the
Jiang-Su algebra Z, cf. [391], because A ⊗ Z and A have “almost” the same Elliott
invariants. This excludes also the cases where one of A or B is “elementary”, i.e., is
78 1. INTRODUCTION AND MAIN RESULTS

stably isomorphic to the compact operators K, e.g. to distinguish C and the Jiang-
Su algebra Z, or where one of the algebras is unital and finite but is not stably
finite, or [687].
And ??????????????
??
(35),
extra requirements on the stable rank, on the real rank, or the decomposition
rank ... weakly unperforated perforated K0 (A)+ etc. ??? .
It happens if one tensors by Jiang-Su algebra as shown by Rørdam
‘‘strongly K1 -surjective’’ [690, def. 6.1]????.
An example of a property that is perhaps not encoded in the Elliott invariants
is the local K1 -surjectivity:
We say that a simple C *-algebra A is locally K1 -surjective , if for every non-zero
projection p ∈ A ⊗ K and every x ∈ K1 (A) there exists a projection q ∈ A ⊗ K such
that

(i) [p] = [q] in K0 (A), and

(ii) x is in the image of the natural C *-morphism from the unitary group of
q(A ⊗ K)q into K1 (A).

This condition is trivially satisfied for all stably projection-less C *-algebras A.

The property of local K1 -surjectivity has an explanation with help of the graded
K∗ -groups: There is a natural isomorphism

K∗ (A) := K0 (A) ⊕ K1 (A) ∼

= K0 (C(S 1 , A)) ,

cf. [73, sec.9.4.1], and one can use this isomorphism to introduce an pre-order
structure K∗ (A)+ on K∗ (A) by the sub-semigroup K0 (C(S 1 , A))+ of K0 (C(S 1 , A)).
One can see that this semigroup consists of the elements ([p], [u]) ∈ K∗ (A),
where p ∈ A ⊗ K is a projection and u ∈ U(p(A ⊗ K)p) is a unitary (that has to be
extended to a unitary in the unitization (A ⊗ K)e by u 7→ u + (1 − p) ).
Indeed, if P = {p(e2πit ) ∈ Mn (A) ; t ∈ [0, 1]} is a projection in C(S 1 , Mn (A)) ⊆
C(S 1 , A⊗K) then there exists a continuous path t ∈ [0, 1] → v(t) ∈ U(Mn (A+C·1))
with v(0) = 1 and v(t)∗ p(1)v(t) = p(e2πit ).
Then u := pv(1)p ∈ U(p(A ⊗ K)p) for p := p(1). The element [p] ∈ K0 (A) is
the image of [P ] under the map [λ]0 : K0 (C(S 1 , A)) → K0 (A) induced by λ : f ∈
C(S 1 , A ⊗ K)) 7→ f (1) ∈ A ⊗ K. This attaches to P a pair (p, u) with u ∈ U(p(A ⊗
K)p) and p = λ(P ).

35 The semigroup of all Murray–von-Neumann equivalence classes of projections in M(A⊗K),

or the infimum of the dimensions of the maximal ideal spaces of the family of Abelian Kadison-
Singer-type C *-subalgebras of A (i.e., with unique pure states extension property), or comparison
properties (certain sub-semigroups of) the Cuntz semigroup of A ⊗ K.
 4. ON J. ELLIOTT’S CLASSIFICATION PROGRAMM AND CONJECTURES 79

Let ([q], [w]) ∈ K∗ (A) with q ∈ A⊗K and w ∈ U(q(A⊗K)q). Then w⊕s,t (w∗ ) ∈
U0 ((q ⊕s,t q)(A ⊗ K)(q ⊕s,t q)) and if w(t) ∈ U0 ((q ⊕s,t q)(A ⊗ K)(q ⊕s,t q)) is a path
with w(0) = q ⊕s,t q and w(1) = w ⊕s,t w∗ , then P (e2πit ) := w(t)(q ⊕s,t 0)w(t)∗ is
a projection in (q ⊕s,t q)(A ⊗ K)(q ⊕s,t q) with p := λ(P ) = q ⊕s,t 0 and pw(1)p =
w ⊕s,t 0 ∈ U(p(A ⊗ K)p). more details???? ??
All simple stably infinite C *-algebras A are locally K1 -surjective, and for them
holds K0 (A)+ = K0 (A) and K∗ (A)+ = K∗ (A), because for each projection p ∈ A⊗K
there exists a full and properly infinite projection q ∈ A⊗K with [q] = [p] in K0 (A),
cf. Lemma 4.2.6(ii) (compare also classical work of J.Cuntz [172] ! ). Then there
are natural isomorphisms K1 (A) ∼ = K1 (A ⊗ K) ∼ = K1 (E) for the unital C *-algebra
E := q(A ⊗ K)q, and E has a properly infinite unit q, which implies that E is
K1 -surjective, i.e., that u ∈ U(E) → [u] ∈ K1 (E) is surjective, cf. Lemma 4.2.6(v).
If A is simple and stably finite, then it is easy to see that A is locally K1 -
surjective, if and only if,

K∗ (A)+ = {(0, 0)} ∪ (K0 (A)+ \ {0}) ⊕ K1 (A) ,

where ∅⊕K1 (A) := ∅, e.g. stably projection-less A are locally K1 -surjective because
K∗ (A)+ = 0 ⊕ 0 = {(0, 0)}.
Simple unital locally K1 -surjective algebras A with cancellation property (for
the semi-group V (A) = [P∞ (A)]) are necessarily K1 -surjective in the sense that
u ∈ U(A) → [u] ∈ K1 (A) is a surjective map.
Unital simple C *-algebras A with stable rank one are locally K1 -surjective (36).
Unital simple C *-algebras that are inductive limits of homogeneous algebras have
in general stable rank > 1, cf. Villardsen [797], [795].

Question 1.4.1. Can methods of [797] and [795] help to construct a unital
simple ASH algebra that is not locally K1 -surjective?

Even for simple nuclear C *-algebras of real rank zero it is not known if they
are locally K1 -surjective (but they are K1 -injective, cf. [525, cor.4.2.10]).
The Conjecture of Elliott has been confirmed in many cases, but always under
strong restrictions on A and B, e.g. that they are inductive limits of certain special
sub-homogenous algebras An and Bn (37), or, in some very special cases, where A
and B are stably isomorphic to (cocycle) crossed products of such inductive limits
by outer actions of Z, Zn or by R. Until now, on has not found a method to
deduce this required properties of A and B from conditions of K-theoretic nature
alone, or from “general defining relations” of A and B given by non-commutative
polynomials.

36 Apply Gauss algorithm to invertible matrices in M (pAp) that have invertible entries.
n
37 The primitive ideal spaces Prim(A ) and Prim(B ) contain only Hausdorff subspaces X
n n
with “small” dimension in comparison with the minimal dimension of its irreducible representa-
tions
80 1. INTRODUCTION AND MAIN RESULTS

Stably projection-less simple exact C *-algebras A have always non-zero additive

traces (cf. e.g. [441]), i.e., T + (A) 6= 0. There exist inductive limits A of sub-
homogenous algebras such that A is stably projection-less and K∗ (A) = 0. It is
then not difficult to see that they satisfy T + (A) 6= 0. The existence of such algebras
A can be seen from [259, thm. 3.2, thm. 5.2], that shows that the possible Elliott
invariants (G+ +
0 , G0 , G1 , T , h·, ·i), with the additional conditions T
+
6= 0 and that
+ +
hz, τ i > 0 (for all τ ∈ T ) implies z ∈ G0 , are exhausted by the invariants of
simple inductive limits A of sub-homogenous algebras An that are extensions of
sums of sub-homogenous algebras with at most 2-dimensional Hausdorff subspaces
of primitive ideal spaces (38).

Let us consider from now on only the the Elliott invariant (K∗ (A)+ ⊆ K∗ (A))
of simple separable nuclear C*-algebras A with T + (A) = 0:
A combination of works of Blackadar, Cuntz [78] and Haagerup [342] shows
that, for simple and exact C *-algebras A, the equation T + (A) = 0 holds, if and
only if, A is not stably finite (cf. B.4.5 or [441]). If, in addition, A is separable
then there is a simple unital C *-algebra B with properly infinite unit (in fact we
find then B that contains a copy of O2 unitally) such that A ⊗ K ∼ = B ⊗ K.
Thus, if T + (A) = 0, then K0 (A)+ = K0 (A) by [172], and A ⊗ K contains a
properly infinite projection, which implies e.g. that A ⊗ K can not be the inductive
limit of type-I C *-algebras (39). Every simple inductive limit A of type-I C *-
algebras has a non-zero positive trace on its minimal dense ideal Ped(A). More
generally, a simple C *-algebra A without a densely defined 2-quasi-trace can’t be
a subalgebra of an ultra-product
Q
ω (B1 , B2 , . . .) of a sequence B1 , B2 , . . . of type-I algebras, which involves a
striking difference between the class of stable infinite nuclear simple C *-algebras
and all classes of algebras that can be described as inductive limits of type-I C *-
algebras. This sorts of inductive limits contain strictly all other sorts of algebras
that have been successfully classified up to isomorphisms by its Elliott invariant
until so far.
If K0 (A) 6= 0 and K0 (A)+ = K0 (A), or if K0 (A) = 0 and A ⊗ K contains
a non-zero projection, then A is stably infinite and T + (A) = 0. So we see that
the equation T + (A) = 0 for simple nuclear A describes the stably infinite algebras
among the simple nuclear C *-algebras A. For them, the invariants of Elliott reduce
only to K0 and K1 . M. Rørdam has constructed in [687] a simple, separable, unital,
and nuclear C *-algebra R that is finite but is stably infinite.

38 They are defined by a generalized mapping torus construction A (a special difference

n
construction) for pairs of morphisms between sums of sub-homogenous algebras with primitive
ideal spaces such that the Hausdorff parts have dimension ≤ 2. See [273, appendix] for an
alternative proof, where the building blocks An are sub-homogenous algebras with at most 1-
dimensional Hausdorff subspaces of its primitive ideal spaces.
39 Moreover, there is a finite subset X of A such that 1/2 ≤ max{dist(x, B) ; x ∈ X} for

every type-I C *-subalgebra B of A∗∗ .

 4. ON J. ELLIOTT’S CLASSIFICATION PROGRAMM AND CONJECTURES 81

In particular, R is not purely infinite in the sense of Definition 1.2.1, given

below. The algebra R is stably isomorphic to the crossed product of a certain type
I C *-algebra C by an endomorphism λ of C, thus, R is stably isomorphic to a
crossed product D oµ Z by an automorphism µ of D := limn→∞ λn : Cn → Cn+1
with Cn := C, and λn := λ. The algebra R has not real rank zero, but contains
“small” projections, cf. [691] and K∗ (R) = K∗ (C(X)) for X := S 2 × S 2 × · · ·
where S 2 means here the 2-dimensional sphere. In fact it is stably isomorphic to
the generalized Fock-Toeplitz algebra (and to the Cuntz-Pimsner algebra) of some
Hilbert C(X, K)–bi-module.
Its K∗ -groups are both isomorphic to Z⊕Z⊕· · · , i.e., K∗ (R) ∼
= K∗ (S 1 ×S 1 ×· · · ).
Let us say some words about the general classification problem for non-p.i. al-
gebras and its connection with the Rørdam groups R(C; A, B) that we introduce in
Chapter 7.
to be filled in: overview! and Rordam groups ??

update: (Now 2021 ???): The classification covers now all – unital? – separable
simple nuclear C *-algebras that absorb the Jiang-Su algebra tensorial and are in
the UCT-class?)
Some of the proofs of Elliott’s Conjecture work well for inductive limits A :=
indlimn An of sub-homogenous algebras An , e.g. provided that A is of real rank
zero and that there is some bound ∞ > Γ ≥ Dn /Rn for the ratio between the
supremum Dn of the dimensions of the Hausdorff subspaces of Prim(An ) and the
infimum Rn of the dimensions of the irreducible representations of the An ’s, or
that A is simple and the An ’s are extensions of finite dimensional algebras Cn by
suspensions SBn of finite dimensional algebras Bn , i.e., if there are exact sequences
0 → SBn → An → Cn → 0.
No description of this additional assumptions in terms of the K-theory has been
found yet.
E.g. it is even an open problem whether a simple separable unital nuclear C *-
algebra A of real rank zero with unique trace state and with the pre-ordered K- and
KK-groups of the CAR algebra M2∞ is stably isomorphic to M2∞ . In particular
suppose that A⊗O∞ ∼ = M2∞ ⊗O∞ and that (K0 (A), [1A ], K0 (A)+ ) and the Elliott
invariant of the UHF-algebra M2∞ are order isomorphic by an isomorphism that
is compatible with the – by O∞ -stable isomorphism – given isomorphism of K0 (A)
with K0 (M2∞ ).
Question: Is A stably isomorphic to M2∞ ?
This should be the case if A is in the UCT-class ... ???
The true problem is here if such A must automatic tensorial absorb the Jiang-Su
algebra Z. Or at least if F (A) := A0 ∩ Aω has no character ... ??????
82 1. INTRODUCTION AND MAIN RESULTS

It is even not clear if there exists a simple unital nuclear C *-algebra C of real
rank zero with the Elliott invariant (Z, Z+ , 1) and unique trace state, but (!) such
that F (C) has a character.
(I want to see how it is constructed as an inductive limit of finite-dimensional
C *-algebras An by completely positive contractions, because all separable nuclear
C *-algebras A are such inductive limits up to completely positive and completely
isometric isomorphisms.)
Forgive me, if I want to see pi-sun algebra with trivial K-theory that is not
isomorphic to O2 , which is simply generated by two isometries s1 , s2 ∈ L(`2 (N)
with

1 = s∗1 s1 = s∗2 s2 = s1 s∗1 + s2 s∗2 .

Perhaps it is e.g. the case if C has not the “corona factorization property”
(CFP).
One can see here that ??????? CFP is very necessary and excludes cases that
have not CFP (or satisfy not the UCT ?). Therefore we write here all at a construc-
tive and elementary level that shows where we suddenly this additional properties
UCT or CFP need.
Moreover it is even unknown whether pre-ordered KK-invariants and trace
invariants are enough to characterize the CAR algebra among the simple separable
nuclear quasi-diagonal C *-algebras in the UCT-class with real rank zero and stable
rank one. (Please don’t misunderstand me: Here we should start with data from
the ordered K∗ -theory and not from the methods to construct them from inductive
limits of (many step) extensions of stabilized sub-homogenous algebras combined
with semi-direct products by semigroups or quantum groups. The question should
be: how many pairwise stably non-isomorphic simple nuclear C *-algebras have the
same invariants, and how look the amenable C *-algebras in those classes of partially
ordered K∗ -theory that are NOT in (some sort of) the UCT-class.
Is ”seems” that the (non-zero) separable, unital, nuclear (= amenable), purely
infinite, and simple C *-algebras A with A ∼
= A ⊗ Z are all contained in the UCT-
class.
But no proof or counter examples have been found yet (end of 2021?). Notice
here that all pi-sun A satisfy A ∼
= A ⊗ O∞ (give references here) and that O∞ ∼
=
O∞ ⊗ Z.
Next has to be edited !!! update:
The latter follows (not simply) from the fact that O∞ , Z and O∞ ⊗ Z are
all nuclear, simple, separable, unital and are in the UCT class, and that O∞ and
O∞ ⊗ Z are strongly purely infinite (because are p.i. and simple).
In the case of simple separable nuclear projection-less stable algebras there are
only some interesting examples studied so far.
 4. ON J. ELLIOTT’S CLASSIFICATION PROGRAMM AND CONJECTURES 83

One of it is given by a crossed product of O2 with an irreducible circle action,

????? In the case of a unital C *-algebra without non-trivial projections there is
e.g. the open question if the Jiang-Su algebra Z, [391], is the only tensorial self-
absorbing separable unital C *-algebra A 6= C in the UCT-class that has the same
Elliott invariant as C.
Since W. Winter has shown that all separable tensorial self-absorbing unital
C *-algebras with approximately inner flip absorb the Jiang-Su algebra Z, and
since the class of Z-absorbing unital simple nuclear stably finite C *-algebras in the
UCT-class are classified by its Elliott invariant it follows that tensorial Z-absorbing
and a kind of ordered (K0 (A), K0 (A)+ , 1A ) is then a complete invariant.
The big open problem is: Can we escape from the very serious additional
assumption that A is in the UCT-class for the classification of pi-sun algebras A?
More generally is A ⊗ O∞ (or A ⊗ Z) for all amenable separable A in the UCT
class?
Is the UCT property equivalent to the Corona factorization property?
“Sudden” Question:
Consider the universal C *-algebra An = C ∗ (a, dk , 1, s, t) given for a ≥ 0,
d1 , . . . , dn , 1, s, t by the relations k d∗k adk = 1, s∗ t = 0, s∗ s = 1, t∗ t = 1.
P

(It implies that An contains a copy of O∞ . Let t1 , t2 , ... natural generators

build from s and t.)
X := k tk a1/2 dk satisfies X ∗ X = 1.
P

Is a ⊗ 1 properly infinite in An ⊗ Z?
I.e., does there exist g ∈ An ⊗ Z with g ∗ (a ⊗ 1)g = 1 ⊗ 1?
It is enough to find h ∈ A ⊗ Z, and non-zero b ∈ Z+ with h∗ (a ⊗ 1)h = 1 ⊗ b,
because Z is simple, O∞ ⊗ Z ∼ = O∞ and 1A ∈ O∞ ⊆ A.
Take elements b and x1 , . . . , xn ∈ Z with x∗k x` = δk,` b (e.g. coming from an
n-homogenous element ψ : C0 (0, 1] ⊗ Mn → Z b = ψ(f0 ⊗ p11 )).
P
Then g := k dk ⊗ xk has this property.
In [467] it is shown that the (CFP) is valid if F (A) = (A0 ∩ Aω )/ Ann(A, Aω )
has no character.
Theorem 4.3. (of [467]):
Let A be a unital separable C *-algebra such that the central sequence algebra
F (A) has no characters. Then A has the Strong Corona Factorization Property (s
CFP ? SCFP ?).
“Strong” Corona Factorization Property for semigroups, cf. citation [18, Defi-
nition 2.12] (in [467] ??? what is 18??):
84 1. INTRODUCTION AND MAIN RESULTS

For every x0 , x, y1 , y2 , y3 , . . . in Cu(A) and m ∈ N such that x0 << x and

x ≤ myn in Cu(A) for all n ≥ 1, there exists k ≥ 1 such that x0 ≤ y1 + y2 + . . . + yk
in Cu(A).
The (strong) Corona Factorization Property can therefore be viewed as a weak
comparability property for Cu(A).
It is not known how the Rørdam semi-groups SR(A, B) of Chapter 7 play
together with (versions of) the Cuntz semigroups, even not in the important case,
where A and B tensorial absorb the Jiang-Su algebra Z.
This shows that we can’t present here an idea for answering classification ques-
tions in full generality if other types of algebras or additional structure like group
actions have to be studied. We develop in this book a method to reduce some
classification problems to questions on the realization of KK-theory with help of
generalized Weyl–von-Neumann–Voiculescu theorems. The main tools are contin-
uous modifications of the Rørdam groups R(A, B), as defined in Chapter 7. They
play a role in several generalizations, e.g. to the non-simple case as described above,
or others, as mentioned in the remarks of Chapter 7. Some of the known stable in-
variants for simple separable nuclear C *-algebras are the same for R(·, ·)-equivalent
C *-algebras A and B (e.g. they are KK-equivalent, F (A) and F (B) contain, up to
isomorphisms, the same simple separable C *-subalgebras, the stable rank, the real
rank and decomposition rank are the same). A part of further study on classifica-
tion could be the study of the consequences of R(·, ·)-equivalence. Partial results
in this direction are Theorems B and M and its Corollaries C and N above. Above
where? Above where? Give references !!

5. The way to here. Acknowledgements

This book is a very expanded and generalized version of the preprint of Decem-
ber 1994 (3rd draft) with the same title “Classification of purely infinite C*-algebras
using Kasparov’s theory”. In the original 1994-preprint a simpler “discrete” asymp-
totic theory using EKω (A, B) has been taken to eliminate some technical difficulties
arising in the continuous case. Now we have replaced the discrete asymptotic meth-
ods (working with sequences) by their technically more involved continuous versions
(working with paths going to infinity). This more elaborate theory allows to see
more clearly where we have homotopy-invariant results. Now they are all integrated
into our newest version that rigorously uses matricial operator-convex cones, be-
cause, up to the – not here – discussed group- or quantum-group equivariance.
A pre-version (of some sections concerning basics on purely infinite simple al-
gebras) partly has been written in 1998 by postdoc students of the Fields Institute
year on C *-algebras. It was based on a seminar which was organized and held by
postdoc students, but was supported by several lectures, consultations, e-mails and
outlines from the author. The sections concerning simple algebras in the new pre-
sentation reflects a compromise between their detailed and elementary approach and
the authors new insights, that changed during more then three decades considerably.
 5. THE WAY TO HERE. ACKNOWLEDGEMENTS 85

More details and background are given. In fact, the exposition is sometimes very
detailed. But here we are running into a problem: half of the technical lemmas are
folklore trivialities for those readers coming from the theory of exact C *-algebras,
tensor products and operator spaces. The other half is trivial for readers who have
a clear understanding how different variants of KK-theories and K-theories can be
constructed, and how they can be compared. But both is presented here at an
elementary level, – so far as it is really used for the proof of our theorems. Only
textbook material is omitted or is only sketched ( 40 ).
The old and sketchy proof of the natural isomorphism of the Grothendieck
group R(A, B) of [Homnuc (A ⊗ K, Q(R+ , B ⊗ K))]u with KKnuc (A, B) from the
Appendix A of the Dec 94 preprint is now explained in a more transparent manner,
by an axiomatic study of the semigroups of morphisms which are dominated by a
fixed morphism (see Theorem 4.4.6, Proposition 7.2.13 and Chapter 9). But, up
to using a technical framework that divide the arguments into smaller steps and
offer more details, we use the same idea as in the proof outlined in the Appendix
A of the Dec 94 preprint. Now it becomes visible that the ideas in the preprint
apply also to the semigroups SR(X; A, B) := [Homnuc (X; A, Q(R+ , B))] and to the
even more general semigroups SR(C; A, B) for (non-simple) stable separable exact
A and arbitrary σ-unital stable strongly purely infinite B, and, more generally,
to the semigroups SR(C; A, B) for in a “functorial way” selected matrix operator-
convex cones C ⊆ CP(A, B). But one needs to use the homotopy invariance of
the generalized m.o.c. cone equivariant Kasparov groups KK(C; A, B) to get the
basic ingredient for the proof that the natural group epimorphism from the C-
controlled unsuspended-but-stable E-theory groups R(C; A, B) onto KK(C; A, B)
is also injective.
Notice that E. Blanchard [89] has extended Theorem A to C *-algebra bundles.
He uses Theorem A and Rørdam’s proof of O2 ∼ = O2 ⊗ O2 ⊗ · · · (cf. [678]). This
was done independent by N. Ch. Phillips [?] and the author with other methods
and applications.
Also, there is an alternative proof of Part (i) of Theorem A and Parts (ii) and
(iii) of Corollary F, given by N. Ch. Phillips and the author [?], that does not use
a generalized Weyl–von-Neumann–Voiculescu theorem, that we, here in this book,
prove and use.
N. Ch. Phillips [627] gave a shorter proof of Corollary C based on the observa-
tions given here in Parts (ii) and (iii) of Corollary F above (as in the beginning of the
authors Dec 94 preprint). His proof uses E-theory and deep results of M. Dadarlat,
N. Higson, A. Loring, H. Lin and M. Rørdam.
One can proceed in the non-simple case along a way similar to the approach
of N.Ch. Phillips with a generalized ideal lattice-equivariant version of E-theory
in the non-simple case, after one has proved Theorem K (= non-simple variant of

40 Some knowledge on Cuntz-Krieger algebras is useful for considering examples. But the

most basic examples are discussed here in the Appendix A.

86 1. INTRODUCTION AND MAIN RESULTS

Theorem A). In fact, some results of N. Higson, J. Cuntz and others indicate that
every sort of an KK-theory is functorial equivalent to a “liftable” (or “semi-split”)
variant of some E-theory. We are quite happy to find a convincing proof in our case
of KK(C; A, B) and the related stable but unsuspended version of E-theory, here
realized by the generalization R(C; A, B) of Rørdam’s sequence groups.
This can be made precise in analogy with our natural isomorphisms

R(C; A, B) ∼
= Ext(C; A, SB) ∼
= KK(C; A, B) ,

where A and B are stable and separable and C = C(h0 ) for some non-degenerate
*-morphism h0 : A → B that is unitarily homotopic to h0 ⊕ h0 . That means: We
show that in the considered cases there is a bijection between certain m.o.c. cones
and such morphisms h0 .
If one imposes into the definitions directly some sort of homotopy-equivalence
then the above considered isomorphisms are “almost evident”. Then the iso-
morphisms are much easier to derive than our “constructive” results in Chap-
ters 8 and 9 that give really unitary homotopy rather than general homotopy
equivalence. But general homotopy equivalences are useless for our purpose. In
fact, – at least in view of our applications –, the great invention of Kasparov
was the observation that the groups Ext(A, B) with its fairly algebraic defini-
tion – see Chapter 5 – are naturally isomorphic to the homotopy invariant groups
KK(A, SB), cf. Chapters 5 and 8 for more details and Section 5 for the most im-
portant implication of this homotopy invariance. We need a more “constructive”
proof of the isomorphism Ext(C; A, SB) ∼ = KK(C; A, B) than those proofs given
usually, because it becomes an important ingredient of our proof of the isomorphism
R(C; A, B) ∼ = Ext(C; A, SB) at the very end of Chapter 9. The reader also should
observe that some technical preparation for the proofs of above isomorphisms con-
tain new results, that we have formulated and established in much more generality
than actually used here in the book for proofs of the main topics.
The author has outlined proofs of Theorem A(i), Corollary F(iii) and Corollary
G (and of the result that a pi-sun algebra has a central sequence of unital copies
of O∞ ) in his talk on the ICM-satellite conference in Geneva in August 1994. The
idea of the proofs of Theorem I and Corollary J has been outlined in the authors
talk in Rome, July 1996, at the conference on Operator Algebras and Quantum
Field Theory.
In 1995-1998 we have extended the results, except Corollary H, to continuous
bundles of C *-algebras with finite dimensional metrizable compact base, in the
sense of [471]. One only has to replace KKnuc (·, ·) by a “nuclear” version of the
Kasparov functor R KKG (X; ·, ·) for the base space X (where here G := {e} is
the trivial group!), and the sub-triviality theorem of E. Blanchard [89] replaces
Theorem A.
 6. TO DO: UNIFY NOTATION IN ALL CHAPTERS! 87

In the period from March to September 1998 we arrived at a complete classifi-

cation of all stable separable nuclear C *-algebras that tensorial absorb O∞ up to
ideal system preserving isomorphisms by their KK(X; ·, ·)-class.
Theorem M has been presented in the authors talk in Copenhagen, August
1998. A talk about Theorem K was given in Münster, March 1999. An overview of
the main results of this book was given in [442], including outlines of some proofs.
In the meantime many new results have appeared, and our new exposition should
allow the reader to see that several fundamental observations can be derived also by
our approach. The more general viewpoint using operator-convex cones in Chapters
3, 5, 7, 8 and 9 has been worked out in 2001–2006, and was outlined in conference
and workshop talks in 2007 and 2008. Among the applications are ??????
list some applications of classification ??????? ??
The author likes to thank R. Bhat, S. Friese, P. Friis, T. Häberlen, H. Harnisch,
H. Osaka and G. Vaillant. They have helped to prepare versions of parts of the
early 1994 preprint and/or have worked out a more detailed version of parts of it.
Inspiring, helpful and sometimes (with any right !) critical remarks came from
C. Anantharaman-Delaroche, D. Bisch, B. Blackadar, E. Blanchard, U. Bunke,
J. Cuntz, M. Dadarlat, G. Elliott, U. Haag, G. Kasparov, M. Rørdam, V. Voicu-
lescu, A. Wassermann, S. Wassermann, J. Weidner and W. Werner. They also have
suggested improvements that I tried to follow on.
J. Cuntz, J. Elliott and M. Rørdam have supported the work of the author on
this monograph several times.

6. To do: Unify NOTATION in all Chapters!

Use ≈ (Cuntz equiv.) and ∼M vN (Murray–von-Neumann equiv.)

Adjust in all chapters of the book the use of ∼ and ≈
And the same with [a]∼ , hai∼ , [a]≈ , hai≈ .
Put it in the Index list. Decide where a reminder is necessary... ??.
 CHAPTER 2

Basics on purely infinite C *-algebras

We consider here several properties that cause pure infiniteness of simple or non-
simple C *-algebras. It requires to discuss first the relations between the different
definitions of infiniteness and the related terminology for C *-algebras. The three
Sections 2, 3 and 4 consider mainly the case of simple algebras. But the later
considered pure infiniteness of non-simple algebras play also for the classification of
simple C *-algebras an important role. It causes a partly study of the huge variety
of definitions of sorts of “infiniteness” for non-simple C *-algebras or of elements in
them. Then some of the later used permanence properties will be studied here in
this chapter.
The last three Sections 16, 17 and 18 study properties of strong purely infinite
C *-algebras in sense of Definition 1.2.2, and some constructions that lead to strong
p.i. C *-algebras. Before and in-between there is the study of general observations
needed in the proofs of results in later chapters, where sometimes only locally some
sort of infiniteness plays a role for the applications.
The inventor – and pioneer of the study – of “purely infinite simple C *-
algebras” was Joachim Cuntz. He introduced in [172, p. 186] a notion of pure in-
finiteness in case of simple C *-algebras using his observation in below given Lemma
2.1.6 for simple C *-algebras:
A C *-algebra A is said to be purely infinite – in the sense of J. Cuntz – if every
non-zero hereditary C*-subalgebra D of A contains an infinite projection p ∈ D,
p 6= 0, i.e., a non-zero projection p that is Murray–von-Neumann equivalent to a
proper sub-projection of itself in the sense of following definition:

Definition 2.0.1. Elements a, b ∈ A+ are Murray–von-Neumann equiv-

alent (short: MvN-equivalent) in A if there exists d ∈ A with d∗ d = a and
dd∗ = b . The relation will be denoted by a ∼M vN b or simply by a ∼ b .

The MvN-equivalence class of a ∈ A+ will be denoted by [a]M vN – or simply

by [a] if there is no danger to be mixed up with the considerably “bigger” class
[a]≈ of all b ∈ A+ that are Cuntz–equivalent to a (cf. Definition 2.5.1 ).
Basic examples of purely infinite simple algebras are the Cuntz algebras On , n =
2, 3, . . . , that are the universal unital C *-algebras generated by elements s1 , . . . , sn
Pn
with defining relations s∗j sk = δj,k 1 for j, k ∈ {1, 2, 3, . . . , n} and k=1 sk s∗k = 1,
respectively the Cuntz algebra O∞ generated by a sequence of elements s1 , s2 , . . .
with relations s∗j sk = δj,k 1 for j, k ∈ N, cf. [169]. In particular, J. Cuntz studied
89
90 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

and confirmed in early work some related conjectures of J. Dixmier (1964) in [214]
concerning O2 .
Some properties of O∞ and O2 did play a fundamental role for the proofs of the
classification results for pi-sun algebras both in the approach of N. Ch. Phillips and
in the the approach of the author of this book. Therefore, we list some properties of
Cuntz-algebras On and O∞ in Section 1 of Appendix A and outline there elementary
proofs for this properties, – but only so far as they do not follow immediately from
some of the very basic observations in Chapters 2, 3 and 4. Moreover, we give in
Section 1 of Appendix A some additional informations on the algebras On that will
be used often in proofs, applications and examples.
Our “long” Proposition 2.2.1 says that on the class of simple C *-algebras pure
infiniteness in the sense of J. Cuntz is equivalent to each of the local, weak or
strong pure infiniteness defined below, and to many other properties that are useful
for several applications useful. But there are examples of amenable (= nuclear)
separable unital simple C *-algebras that are (quasi-) trace-less but are not purely
infinite in the sense of any here given and considered definitions, cf. the example of
M. Rørdam [687] ( 1 ).
It has been shown in [462, prop. 4.7], [463, cor. 6.9] and [93, thm. 4.17], –
cf. also our Proposition 2.6.5 –, that in case of C *-algebras A of real rank zero all
of our below given definitions of pure infiniteness of simple C *-algebras A coincide.
Unfortunately, this is not the case for non-simple A. ????
Next blue/red text has to be verified, is ‘‘under observation":
More generally, equivalence of all definitions of pure infiniteness ( – except possibly
strong pure infiniteness! – ) for C *-algebras holds for all C *-algebras with the
property that
for every closed ideal J 6= A of A and every non-zero hereditary C*-subalgebra D
of A/J there exists a non-zero projection p ∈ D,
i.e., where A is “rich of projections” in the sense of our Definition 2.6.1: If C *-
algebras A with this property has our “weakest” property of infiniteness, – i.e. the
“local pure infiniteness” (l.p.i) of Definition 2.0.3 – then it turns out that A is
purely infinite in the sense of Definition 1.2.1.
This class of C *-algebras A contains the class given by the property that
every non-zero hereditary C*-subalgebra D ⊆ A/J of any quotient A/J of A con-
tains a non-zero infinite projection p ∈ D,
– i.e., those where all quotients are purely infinite in the sense of the above quoted

1 Since a long time it seems to be still (January 2022) an open question if there exists a

simple nuclear C *-algebra of “real rank zero” that is stably infinite but is not purely infinite,
cf. Question (Q1) in Chapter 1. This cited famous example of M. Rørdam has not this properties.
Such an example (of real rank zero), – if it exists –, can be modified to produce a unital simple
algebra A with real rank zero, that contains no infinite projection, but M2 (A) contains a copy of
O2 unitally. Even if this exists, then the question is: Can such kind of A be a nuclear C*-algebra
with all this properties?
 2. BASICS ON PURELY INFINITE C *-ALGEBRAS 91

definition of J. Cuntz in [172] (2) –,

are purely infinite in sense of Definition 1.2.1.
We do not know if purely infinite C *-algebras A, that are “rich of projections”,
in sense of our Definition 2.6.1, are moreover strongly purely infinite in sense of
Definition 1.2.2, – our so far strongest infiniteness property of a C *-algebra that is
”rich of protections”.
We need only a kind of semi-splitting
(in cases where we can reduce all arguments to the separable case
by considering suitable C∗-subalgebras E if possible):
An element b ∈ E+ with bEb ⊆ A with b ≈ (b ⊕ b) and a non-zero element c ∈ E+
orthogonal to b, i.e., cb = 0 and πA (c) 6= 0.
Can only work if A(??) is not “essential”???...
But it could be that all non-zero projections of cEc are contained in cAc ...?
No progress ... !!!
To find b, c one could take a strictly positive element of a ∈ A+ , i.e., A = aEa.
Try to use the projectivity of C0 ((0, 1], M2 ), and consider M2 ⊆ E/A ...
It gives positive contractions c, d ∈ E+ with cd = 0, πA (c) = p11 , πA (d) = p22 .
Then try to modify a + d ...

NO PROGRESS !!!
One question is, if in case of separable σ-unital A, where A is an ideal of
E, that is “rich of projections” and is purely infinite (⇐ HERE: A or E ?), has
the property that an exact sequence 0 → A →η E →π O2 → 0 must split, i.e.,
that there exists a C *-morphism φ : O2 → E with π ◦ φ = idO2 for the quotient
epimorphism π : E → O2 with kernel η(A) for the inclusion morphism η : A → E
onto an ideal of E if E has real rank zero.
(Questions: Is this splitting necessary for E being s.p.i.? Or is ”semi-splitness”
enough?)
Because then it would be the same as our Definition 2.6.1 (but with property
l.p.i. added as additional property) if we could show that each copy of O2 can be
lifted for a strongly p.i. ideal with many projections. Here we can reduce all to the
separable case, and build later inductive limits of them.
In general the question is:
Does there exists a separable p.i. C *-algebra A such that every non-zero here-
ditary C *-subalgebra D of A is not strongly p.i.?
There exists D that contains may nonzero n-homogenous C *-subalgebra En :=
Fn ⊗ Mn with En+1 ⊆ En .
(The injections ??????? .)
If such example not exists, then every p.i. algebra is s.p.i. !!!

2... defined there mainly for the study of simple Cuntz-Krieger algebras!
92 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

(Where are the proofs of this?)

Thus, it suffices to consider suitable separable C *-subalgebras of ultrapowers
of p.i. algebras ... ???
But if A has real rank zero then local pure infiniteness of A implies strong pure
infiniteness of A, because it is a special case of the following result:
If A satisfies the stronger condition that, – for every hereditary C *-subalgebra
D ⊆ A and pure state ρ on A with ρ|D 6= 0 – , there exists a projection p ∈ D with
ρ(p) 6= 0 ( 3 ), then A is strongly purely infinite in sense of Definition 1.2.2 if A is
locally purely infinite in sense of Definition 2.0.3 .
So far we don’t have seen any example of a purely infinite separable C *-algebra
A that satisfies the properties in Definition 2.6.1 of C *-algebras with “many pro-
jections”, but is not strongly purely infinite.
Check, compare and relate this blue part to above to the written
text. Remove repeats in blue parts!!!
The two next discussed cases seem not to be equivalent! Check
again!
Merge below and above text about special cases!
Look more to applications than to
problems with very weak definitions!
THIS is somewhere in the proof:
It implies immediately that every πJ (a) ∈ A/J is infinite for each a ∈ A+ and
closed ideal J of A with a 6∈ J, if one takes D := πJ (a)(A/J)πJ (a). Thus, each
non-zero element of A is infinite.
Suppose that each non-zero element of A+ is infinite, is then each non-zero
element of A+ properly infinite?
??? lem:2.infinite.elem ??? cf. Lemma ??(v) and discussion of case of algebras
that are “rich of projections” ...
Definition 2.6.1
Propositions on generalized Cuntz def. ?? and ??
[prop:2.Cases.where..Cuntz.pi..implies..l.p.i]
[prop:2.cases.where..l.p.i..implies..p.i.?]
Refs.: proof of cases where l.p.i implies p.i.
e.g. tensor ‘‘q-traceless’’ with Jiang-Su algebra
linear ordered lattice of ideals leads to s.p.i.
being ‘‘rich of projections’’ ... ???????
... then A is moreover strongly purely infinite in sense of Definition 1.2.2.
(Really???)
3 Equivalently expressed: For every a ∈ A with kak = 1 and pure state ρ on A with ρ(a) = 1
+
there exists a projection p ∈ aAa with ρ(p) > 1/2.
 2. BASICS ON PURELY INFINITE C *-ALGEBRAS 93

It seems that this class of C *-algebras (with real rank zero or at least with
many projections) and the class of C *-algebras A with the property that the ideal-
lattice of A is linear ordered, are the only cases where we can show so-far that the
properties (l.p.i.) and (s.p.i) coincide.
Then the permanence properties for the class of strong purely infinite C *-
algebras allows to define from this classes other classes of C *-algebras where (l.p.i.)
and (s.p.i) coincide.
But until now (Jan 2022) we have seen no example of a C *-algebra A that is
locally or weakly purely infinite but is not strongly purely infinite ... but, unfortu-
nately, some proofs and most interesting applications require strong pure infinite-
ness. END OF BLUE DISPUT TEXT. ????
On the other hand, the algebra O2 ⊗ K + C · 1 (with unit 1 := 1O2 ⊗ id`2 )
has real rank zero and is purely infinite in the sense of the above mentioned defi-
nition of J. Cuntz [172, p. 186], but is even not locally p.i. in the weakest sense:
the “local pure infiniteness” of Definition 2.0.3, because its unit element 1 is not
properly infinite (inside the multiplier algebra M(O2 ⊗ K) of O2 ⊗ K. The algebra
C0 ((0, 1], O2 ) is strongly p.i. in the sense of our Definition 1.2.2, but is not purely
infinite in sense of the definition in [172], because it is stably projection-less.
Therefore, it is necessary to underline that we use or consider the original
definition of J. Cuntz only in case of simple C *-algebras A :
The Definition 1.2.2 of strongly purely infinite algebras is better applicable to
non-simple C *-algebras, and has nice functorial properties, e.g. passes to quotients,
extensions (see Section 17) and to tensorproducts with exact C *-algebras ( 4 ).
One can even work with the below given (possibly weaker) definitions, that are
both equivalent to strong pure infiniteness in particular cases, e.g. where A has real
rank zero, or is simple, or has a Hausdorff primitive ideal space Prim(A) of finite
dimension, or where the lattice I(A) of closed ideals of A is linearly ordered.
What about examples or counter examples?
But here is good news for readers that feel disturbed by the “zoological gar-
den” of definitions of pure infiniteness of non-simple C *-algebras A and their still
incomplete list of permanence properties:
If one tensors a C *-algebra A 6= {0} with the Jiang-Su algebra Z then all
our definitions of pure infiniteness become the same, i.e., A ⊗ Z is “locally purely
infinite” in sense of Definition 2.0.3 if and only if A ⊗ Z is strongly purely infinite
as defined in Definition 1.2.2.
It implies immediately that A ⊗ O∞ is s.p.i. for every C *-algebra A 6= {0},
because O∞ ∼ = O∞ ⊗ Z.

4 Despite we haven’t seen so far a, – necessarily non-simple –, example of a weakly p.i.

C *-algebra that is not strongly p.i. It is related to the more general open question if non-
zero elements have good approximate Glimm halving in non-simple residually anti-liminary C *-
algebras, cf. Section 7.
94 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Give cite or ref for isomorphism O∞ ∼

= O∞ ⊗ Z !
(Must show KK-equivalence ??? of Z to C or O∞ )
Moreover this is even the case if A ⊗ Z has no non-trivial 2-quasi-trace. Here
“trivial” means that the 2-quasi-trace takes only the values {0, +∞} on (A ⊗ Z)+ .
(All quasi-traces on C*-algebras A are automatically 2-quasi-traces if A is both
simple and exact ... ?
”Prime” C*-algebras with non-trivial l.s.c. 1-quasi-traces can exist. The full
free group C*-algebra C ∗ (F∞ ) has all sorts of quasi-traces, among them those that
are not 2-quasi-traces.)
Moreover A ⊗ Z is strongly purely infinite if each lower semi-continuous
2-quasi-trace τ : A+ → [0, ∞] takes only the values 0 and +∞, i.e., if each τ is
“trivial” in the following sense:

Compare and choose from the kernel descriptions!!

A lower semi-continuous 2-quasi-trace τ : A+ → [0, +∞] is called trivial if τ
takes only values in {0, +∞} , and is defined by its closed kernel ideal τ −1 (0) =:
J ⊆ A.
“Trivial” lower semi-continuous 2-quasi-traces τ : A+ → {0, +∞} correspond
naturally to closed ideals Jτ defined by (Jτ )+ := τ −1 (0), and, conversely, τJ is
defined from a closed ideals J / A by τJ (a) := 0 if a ∈ J+ and by τJ (a) := +∞ for
a ∈ A+ \ J.
If A is an exact C *-algebra, then A ⊗ Z is s.p.i. , if and only if, A+ has no
non-trivial additive trace.
Why we need here that A is exact? Has to do with the additivity of of 2-quasi-
traces (are they natural extendable to M2 (A)+ etc. ...) ???
Give exact citation !!!
Corollary 3.12 of the authors Abel Proc. paper – submitted in 2005! (or 2004?)
– says moreover:
A ⊗ Z is s.p.i. if every l.s.c. 2-quasi-trace on A+ is trivial.
(It is (= seems to be) clear that A ⊗ Z has no non-trivial 2-quasi-trace if A has
no non-trivial 2-quasi-trace.
Thus, the point in the proof is: If A ⊗ Z has no non-trivial 2-quasi-trace, then
A ⊗ Z is s.p.i.
(All s.p.i. C*-algebras have no traces ... A bit imprecise expressed:
The traces on B+ a C*-algebra B can be extended to and calculated from the
”open” projections in W*-algebra (B ⊗ K)∗∗ . The natural order and equivalence
will be respected ... If they are approximated from below by stable projections then
only the values 0 and +∞ can appear.)
 2. BASICS ON PURELY INFINITE C *-ALGEBRAS 95

The point seems to be that we have to consider the corresponding ”dimension

functions” on the open projections. They must be trivial in sense that only 0 and
+∞ can appear.
Definition: Pedersen Ideal := The ideal of B that is generated by the elements
(b − ε · 1)+ with b ∈ B+ and ε ∈ (0, kbk) and 1 ∈ M(B).
Does this arguments use that Z ∼
= Z ⊗ Z ?)
This has been proven by M. Rørdam, first in the special case of nuclear C *-
algebras A in [690, thm.5.2], – where one can use the result of U. Haagerup
[342] that all 2-quasi-traces are additive on exact C *-algebras – and later in [690,
thm.5.?2?] for all A . We give a different proof in Section ??, that is a modification
of the authors original proof in ?? Abel proc...??? ... ??????
?????? Kirchberg ???
More important for our more special applications is that the tensor products
A⊗O∞ of any non-zero C *-algebra A with the Cuntz algebra O∞ is strongly purely
infinite (= s.p.i.) in sense of Definition 1.2.2.
Only in the very special case of separable nuclear C *-algebras A we have al-
ways the strong conclusion: Separable nuclear C *-algebras A are strongly purely
infinite, if and only if, A ∼
= A ⊗ O∞ . For general exact and separable C *-algebras
A this seems to be Give reference to this isomorphism, it is likely in
Chapter 11?

Example 2.0.2. !!! Adjust my notation for R by notion W ,

because M. Rørdam called it W in [467].
Then change it also on other places, e.g. in the Introduction Chapter
1
Also in the Introduction where it is mentioned!!
Where is written that this A has the property
(CFP) (= Corona Factorization property ) and where that
F (A) has a character ?
Where is F (A) defined ???? Give Ref’s !!!
∗
No-Where! But W and Cred (F2 ) have the properties
∗
that F (W ) and F (Cred (F2 )) have characters,
It is not known if F (A) has a character.
There is a claim that some conjectures related to (CFP) would be
destroyed if F (A) has a character.
Let W := R denote the unital simple separable nuclear C *-algebra W with
K∗ (W ) ∼= K∗ (S 2 × S 2 × · · · ) constructed by M. Rørdam in [687] that is finite but
is not stable finite, where S 2 denotes here the 2-dimensional unit sphere in R3 .
Where is the existence of a character on F (W ) proven? Or a general theorem
that says: F (A) without character implies that A is not stably finite ...
∗
The simple, separable, unital and exact C *-algebra A := Cred (F2 )⊗R is purely
infinite by Theorem E. Hence, it is also strongly purely infinite by simplicity of this
96 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

A. But this C *-algebra A does not absorb tensorial any amenable C *-algebra 6= C.
( 5 ).
∗
Moreover, the algebras B := Cred (F2 ) and C := R (= W ) have both the
property that the unital C *-algebra F (B) and F (C) have a character.
(Where are W and R defined, and described. In a paper of Rørdam. The W
seems to be )
∗
In case of Cred (F2 ) this follows from the classical observation that the commu-
0
tant vN (F2 ) ∩ vN (F2 )ω of the von-Neumann algebra vN (F2 ) in its von-Neumann
algebra ultrapower vN (F2 )ω has a character, because it is simply C · 1.
Notice that vN (F2 )ω is the adjoint of the Banach space (vN (F2 )∗ )ω (in our
∗
terminology). There is a natural unital C *-morphism from Cred (F2 )ω into the von-
Neumann algebra vN (F2 ) . It maps Cred (F2 ) ∩Cred (F2 )ω into vN (F2 )0 ∩vN (F2 )ω ,
ω ∗ 0 ∗

∗
because k[a, u]k+k[a, v]k ≥ k[a, u]k2 +k[a, v]k2 and the unit ball of Cred (F2 ) is dense
∗
in the unit ball of vN (F2 ), – with respect to the `2 -norm on Cred (F2 ) ⊂ `2 (F2 ).
This seems TO BE WRONG and has not been claimed in [467]!!: A has
perhaps the (then rather surprising) property that the central sequence algebra
F (A) := A0 ∩ Aω has a character, see [467, ex. 4.6].
NO? that above is still not proven! But is a question how near an operator
P
g∈X Lg ⊗ cg is to 1 ⊗ R (now denoted by 1 ⊗ W ?) if it commutes almost with
the generators of F2 .
??? Next also still to check:
This example could show also that simple separable unital exact C *-algebras A
with the Corona Factorization Property (CFP)
Give reference to Def. of (CFP) !!!
exist ??? with the property that the invariant

F (A) := A0 ∩ Aω / Ann(A, Aω )

has a character ???

Seems not clear if this example has a character!, All simple purely infinite exact
C *-algebras should have always the (CFP)? Or not?? Some answer ???
All σ-unital simple purely infinite C *-algebras A are unital or stable.
Citation of Prop./Cor.?
This will be shown for separable A, but carries automatically over to σ-unital
simple purely infinite C *-algebras A, because they can be considered as an limit of
an directed net {Aτ } of separable simple purely infinite C *-subalgebras of A that
contain a strictly positive element of A.

5 It is not shown there that A 6∼ A ⊗ M , but that follows for this A from the fact that
= n
there exist non-zero [p] ∈ K0 (A) with the property that there does not exist [q] ∈ K0 (A) with
[p] = n[q], i.e., [p] is not “divisible” inside K0 (A) by some n ∈ N.
 2. BASICS ON PURELY INFINITE C *-ALGEBRAS 97

If A is then stable, then M(A)/A is simple and purely infinite. This

could/should imply the corona factorization property (CFP) ???
Above was not proven! Ref. to Def?
If a simple separable and exact C *-algebra A has the property that F (A) has
not a character then one has always that (simple ?) A has property (CFP).
?? Above and below the same ??
But it is also known that if A is separable and F (A) has no character then A
has corona factorization property (CFP), cf. [467, thm. 4.3].

Our study starts with a variety of helpful weaker properties that are related to
pure infiniteness but are often easier to verify than pure infiniteness itself, e.g. for
C *-bundles (i.e., C *-algebras of bounded continuous sections in continuous fields
of C *-algebras), or for C *-algebras with linearly ordered ideal lattice.
The property that an element a ∈ A with a 6= 0 is “infinite” in A has been
defined by J. Cuntz by the requirement that there exists non-zero element b ∈ A
such that
b ⊕ a - a.

See Definitions 2.0.3 ?? and ??.

Among the almost “weakest” definition of pure infiniteness of C*-algebras is
that of “local pure infiniteness”:
Check blue text, refs and labels of l.p.i
Check/COMPARE other places of Def. ?? ??? with next blue.
Move equivalent formulations to places where they are used.
The local stability is perhaps a very rare property! Have to find other types of
elements to work with.

Definition 2.0.3. A non-zero C *-algebra A is locally purely infinite (l.p.i.)

if, for every non-zero a ∈ A+ and pure state ρ on A with ρ(a) > 0, there exists a
stable C *-subalgebra D ⊆ aAa with ρ(D) 6= {0}.

Later the Lemma 2.7.15 says that we can replace in Definition 2.0.3 the “pure
state” simply by “state”, and get therefore a formally stronger definition than that
formulated in Definition 2.0.3.
(That definition could be ”incorrect” for a suitable version of ”local pure in-
finiteness”, because ”infinite” is not necessarily ”stable” ... Moreover, it should be
called “locally stable” !!! But a stable C *-algebra D is isomorphic to D ⊗ K(`2 ), i.e.
D∼ = D ⊗ K and contains a σ-unital stable separable C *-subalgebra E ⊆ D with
the property that there exists a non-zero epimorphism from F := C0 (0, 1] ⊗ K onto
E. But the diagonal compact operator f := (1, 1/2, 1/4, . . . , 2−n . . .) is properly
infinite operator in K, in the sense that there exist sequences of elements (an ) and
(bn ) such that
f ⊗ 12 = lim[an , bn ]∗ [f an , f bn ]
n
98 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

in M2 (K) ∼= K This implies that F and E contain a strictly positive contraction

that is properly infinite.
Or is it somewhere shown that for p.i. elements a ∈ A+ the algebra aAa
contains a non-zero stable C *-subalgebra D ⊆ aAa?
If a is ”infinite”, then (perhaps ??????) some element e ∈ aAa+ with kek = 1
exists that is infinite in sense (e − t)+ ⊕ (e − t)+ - (e − t)+ for suitable t ∈ (0, 1).
Perhaps ... this implies that aAa+ contains a (non-zero) stable C *-algebra?
⇐ Would be important and interesting to know!
But stability D ∼
= D ⊗ K of D implies natural pure infiniteness of each strictly
positive elements a ∈ D+ , i.e. with D = aDa.
Suppose that B is ”purely infinite” ... (says that b ⊕ b - b for all b ∈ B+ )
... or suppose something weaker ???, this could be that every element of B
is ”infinite” in the sense, that – for each non-zero b ∈ B+ – there exits a non-
zero cb ∈ B+ with cb ⊕ b - b ... ??? It is known that those elements cb build the
positive part J(b)+ of some closed ideal J(b) of B. And one knows that the element
πJ(b) (b) := b + J(b) of B/J(b) is finite in B/J(b) ... ( Give References ! the strict
finiteness of ).
It implies: If every non-zero positive element of every quotient of B/J is infinite,
then each element b ∈ B+ is properly infinite, i.e. b ⊕ b - b.
Moreover, suppose that B is σ-unital, non-unital, and each (non-zero) quotient
B/J has no unit element.
Is B then itself stable? (Stability Question: B ⊗ K ∼
= B ?)
Look to Rordam paper ... ???
Ref.? Big question: to what kind of ”purely infinite” C *-algebras.!! )
” Zhang’s Dichotomy ”:
A σ-unital, purely infinite (!!!), simple C*-algebra is either unital or stable.
(Reference/Cite ???).
What kind of definition of ”pure infiniteness” is used for ”Zhang’s Dichotomy”?

[688, prop. 5.4] Rordam, Japan 2004 (or 2003),

Definition 6.10.? of this Rordam paper (defines Property: ”regular”)
A C*-algebra I is called ”regular” if every full, hereditary sub-C*-algebra of I,
that has no unital quotients and no bounded traces, is stable.
(E.K.: I := K – of `2 (N) – is not regular ????) Every non-zero hereditary
C*-subalgebra D of K is ”full” in K? because each nonzero element x of K+
generates K as the closed ideal of K that is generated by x. The D is given by a
positive compact operator (by taking one of its strictly positive contractions in it).
Alternatively it is given as P KP for some orthogonal projection P on `2 (N). P KP
 2. BASICS ON PURELY INFINITE C *-ALGEBRAS 99

has finite dimension, if and only if, P `2 (N) is finite-dimensional (and has a faithful
trace). Otherwise P KP is stable.
Thus, each D := P K(`2 )P satisfies the definition of a regular hereditary sub-
C*-algebra of K(`2 ) (given by M. Rørdam for all hereditary C*-subalgebras of
K(`2 ).
(Check this again !!!)
[688] Proposition 6.12. (of Rordam [688] ???) Let 0 − − > I − − > A − − >
B − − > 0 be a short exact sequence of separable C*-algebras and suppose that I
is regular. Then A is stable, if and only if, I and B are stable.
Question 6.5. of Rordam [688]:
Does every (separable) C*-algebra A have a greatest stable ideal (i.e., a stable
ideal that contains all other stable ideals)?
(Is a the closure of an upward directed family of stable ideals of A a stable ideal
of A?)
Equivalently: Let A be the sum of two stable ideals A = J + I. Is A stable?
Is C0 (0, 1] ⊗ K ( = C0 ((0, 1], K) ??) semi-projective ?.
compare next blue with Lemma 2.7.15!
If each stable C *-subalgebra D ⊆ aAa satisfies ρ(D) = {0}, then ρ is zero on
the hereditary C *-subalgebra E of aAa that is generated by all stable elements in
aAa .
DEFINITION (cite/ref: where first def.?) of ”stable” elements b ∈ A+ : bAb is
stable, i.e. bAb ∼
= bAb ⊗ K .
Equivalently: b ∈ A+ is a stable element, if and only if, there exists a
C *-homomorphism ψ : C0 (0, 1] ⊗ K → A such that b = ψ(f0 ⊗ (t1 , t2 , . . .)), for
(t1 , t2 , . . .) ∈ c0 ⊂ K with 1 ≥ t1 ≥ t2 ≥ · · · , and all tn > 0.
(First one has to find an isomorphism bAb ∼
= bAb ⊗ K and then ???? use ????
that b is there approximately equivalent to b ⊗ c0 , ( or simply replace b by b ⊗ c0
...).
But E is invariant under automorphisms of D, in particular E is a closed ideal
of D, i.e., in general the hereditary C *-subalgebra E of D generated by the convex
combinations of all positive ”stable” elements in D is an ideal of D. The E must
be equal to D if A is locally purely infinite in sense of Definition 2.0.3.
If a hereditary C *-subalgebra E of D is invariant under conjugation by certain
unitary elements in the multiplier algebra M(D) ... ?
It follows that ??????
The following Definition is equivalent to Definition [93, def. 1.2].

Definition 2.0.4. We call a C *-algebra A n-purely infinite (or: A is pi(n))

for some fixed n ∈ N if A satisfies that
100 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

(i) for every nonzero a ∈ A+ , positive b in the closed ideal J(a) of A

generated by a, and every ε > 0, there exists d1 , . . . , dn ∈ A such that

X
kb − d∗k adk k < ε ,
1≤k≤n

and
(ii) every quotient C *-algebra 6= 0 of `∞ (A) has (linear) dimension > n2 .

The Part (ii) says here that `∞ (A) has no irreducible *-representation on a
Hilbert space H of dimension k ≤ n.
Why we use here in Part (ii) of Definition 2.0.4 the huge `∞ (A) in place of A?
Is there an example where (ii) for A (in place of `∞ (A)) is not enough?
Is not clear if we can prove that `∞ (A) has no irreducible representation on a
Hilbert space of dimension k ≤ n if A has no irreducible representation on a Hilbert
space of Dimension k ≤ n ...
Use projectivity of Mk (C0 (0, 1]) ...
But there are more general and precise observations needed ...
... because this ”proof” gives only a sequence of (needed) positive functions

fm : (0, 1] → (0, 1]

with fm (1/m, 1] = 1 and fm+1 fm = fm , and that have the property that a convex
combination ak of n + 1-homogenous contractions in `∞ (A) exist that are ... ( !!!
check hand-written papers for an idea of proof ... but seems not complete ...).
If we can use only the following weaker requirement in Part (ii*) of Definition
2.0.5 in place of Part (ii) of Definition 2.0.4 then it is not clear if the two Definitions
2.0.4 and 2.0.5 are really equivalent, or if Definition 2.0.4 is strictly stronger than
Definition 2.0.5 ...

Definition 2.0.5. Let A a C*-algebra. Then A is called (n-step) ”locally

purely infinite” (or ”A is pi(n)”) if

(i*) Each element a ∈ A+ is n-step infinite in the sense:

For each elements c1 , . . . , cn+1 ∈ A and ε > 0 there exist d1 , . . . , dn ∈ A
such that

X X
k c∗j acj − d∗k adk k < ε ,
1≤j≤n+1 1≤k≤n

and
(ii*) A has no irreducible representation on a Hilbert space H of dimension
k ≤ n.

We say that A is weakly purely infinite (for short w.p.i.) if A is pi(n) for
some n ∈ N = {1, 2, . . .}.
 2. BASICS ON PURELY INFINITE C *-ALGEBRAS 101

The property (i*) in Definition 2.0.5 implies that for a C*-algebra A with
this property implies that the property (ii*) is equivalent to each of the following
properties (ii*,1)–(ii*,4?) for its elements and irreducible representations:
(ii*,1) No hereditary C*-subalgebra D of A has a (non-zero) character.
(ii*,2) No hereditary C*-subalgebra D of A has a (non-zero) finite-dimensional
quotient D/J. (Here one can take a closed ideal I of A and consider J := I ∩ D,
because for each closed ideal J of a hereditary C *-subalgebra D of A the closed
ideal J of A generated by I, i.e., the closure I of the linear span of A · J · A, satisfies
J = I ∩ D, because DAJAD ⊆ I.)
(ii*,3) No irreducible representation ρ(A) of A contains a (non-zero) compact
operator in its image.
(ii*,4) No closed ideal J of A has an irreducible representation ρ : J 7→ L(H)
with ρ(J) = K(H).
It is not difficult to see that the Part (i) of Definition 2.0.4 and
??????
are equivalent to Part (i*) of Definition 2.0.5.
It follows that A is purely infinite in the sense of Definition 1.2.1, if and only
if, A is pi(1) in the sense of Definition 2.0.4, but pi(1) follows from the formally
weaker requirements (i) for n := 1 and (ii*) that A has no character (instead to
require moreover that `∞ (A) has no character !).
The conjecture (!) is that the same holds for the requirement (ii) for pi(n), i.e.,
that it is enough to require that

(ii*) A (itself) has no irreducible representation of dimension ≤ n.

But until now (2022) no general argument for a proof of this has been found.
But properties (i) and (ii*) imply together that all (non-zero) n-homogenous ele-
ments of A are properly infinite, cf. Proposition ??? ??. And this implies that all
non-zero n-homogenous elements of `∞ (A) and of Aω are properly infinite...
But this argument does not exclude the existence of irreducible representations
of dimension < n on Aω or M(A)... ??? Examples ?? for this phenomenon ??
Part (ii) says that `∞ (A) has no irreducible representation d : `∞ (A) → L(H)
with Dim(H) ≤ n. Since M(A) is a unital C *-subalgebra of `∞ (M(A)) and `∞ (A)
is an essential ideal of `∞ (M(A)), it would be stronger to require that M(A) has
no irreducible representation of dimension ≤ n. It is not known, even in the case
where A is separable, if M(A) can have an irreducible representation of dimension
≤ n, but its sequence algebra `∞ (A) has no irreducible representation of dimension
≤ n. (But `∞ (M(A)) has then a C *-epimorphism onto `∞ (Mk ) for some k ≤ n.)
It has to do with the more general question if `∞ (A) is again residual antilimi-
nary if a C *-algebra A is residually antiliminary, cf. the discussions below Definition
2.7.2 of residual antiliminary C *-algebras.
102 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

The Definition 2.0.4 is an equivalent formulation of [93, def. 1.2]. It was used in
[93] for some results on pure infiniteness of C *-bundles ( := algebras of continuous
sections vanishing at ∞ in continuous fields of C *-algebra ).
An alternative description of weak pure infiniteness is given in the following
definition of a perhaps stronger property “pi-n” defined in [463, def. 4.3].
This is mentioned below a second time (with footnote?). Compare
it and remove one of them:
The pi(n) C *-algebras have the formally stronger property pi-m of Definition
2.0.6 for suitable (but still unknown) m ∈ N with n ≤ m.

Definition 2.0.6. We say that a C *-algebra A has property pi-n (or: A is

pi-n) if, for each non-zero a ∈ A, the n-fold direct sum a ⊕ a ⊕ · · · ⊕ a = a ⊗ 1n is
properly infinite in Mn (A), if considered as a diagonal matrix in Mn (A).

If A has the property of Definition ?? then no (non-zero) hereditary C *-

subalgebra D of A has a character and any n-homogenous positive contraction
a ∈ A is properly infinite in A.
The Definition 1.2.1 and the above given Definitions 2.0.3, 2.0.4, and ?? of local
or weak pure infiniteness coincide in case of simple C *-algebras, cf. Proposition
2.2.1.
In Section 12 of this chapter it is shown that property pi-n implies property
pi(n) on non-simple A.
Conversely, if A has property pi(m) for some m then there exists n ≥ m
such that A has property pi-n, but no general function f : N → N is known yet
(Febr.2022) with the property that n ≤ f (m) . Moreover, there exists, for every
factorial state ρ on A, a closed ideal J of A with the properties

(i) ρ|J 6= 0,
(ii) J is “essential” in A, i.e., a ∈ A and a · J = {0} implies a = 0, and
(iii) the ideal J has property pi-n (if A has property pi(n)).

Moreover, if A has property pi(m) then there exists n ≥ m such that A has
property pi-n. But until now (Febr 2022 ?) no estimating function f (m) ∈ N
is known with the property that A is pi-f (m) and f (m) ≥ m if A has property
pi(m). It seems that such estimating functions m → f (m) depend from topological
properties of Prim(A).
We give an overview about the known results and open questions concerning
non-simple (locally, weakly or strongly) purely infinite algebras in Section 11 of
this chapter – with some parts of the needed proofs postponed to later sections and
chapters, because there are then the needed observations available for the reader.
By Definition 1.2.1 in the Introduction Chapter 1, a C *-algebra A is purely
infinite, if and only if, A has a property that is formally weaker than the Property
pi(1) of Definition 2.0.4: But it suffices to require that A has no non-zero character
 2. BASICS ON PURELY INFINITE C *-ALGEBRAS 103

– instead to require the additional property that `∞ (A) has no character as it is

extra required in Definition 2.0.4.
We give later in Proposition ?? a list of order between infiniteness.
In fact, – without (!) some sort of infiniteness –, there could be examples of
separable (non-simple?) C *-algebras A that have no irreducible representations of
finite dimension but that `∞ (A) has a non-zero character. Then M(A)/A has a
character ??? and, for all b ∈ A, the sequence (b, b, . . .) ∈ `∞ (A) is in the kernel
ideal of this representation.
It is related to questions about the ultrapowers of states on A and finite di-
mensional quotients of M(A)/A.
What about A := M2∞ ⊗ K?
Why M(A)/A should have a "character"? Reference ?? The question if this
A has no character is equivalent to the questions: Has M(A) a character ?? Has
M(A) a trace?
It turns out that A is purely infinite in sense of Definition 1.2.1, if and only
if, each non-zero element a ∈ A is properly infinite in the sense of Definition 2.1.1.
But this says that A has Property pi-1 of Definition 2.0.6. See Corollary 2.5.6 for
a proof of this equivalences based on a more engaged discussion of the variety of
notions of “infinity” of elements and of pure infiniteness of C *-algebras. It is still
not known (2022 ?) what happens in case n = 2, e.g. even in case of C([0, 1], A)
for non-simple properly infinite C *-algebras A. Up to some special cases, it is not
known (Jan 2022 ?) if Property pi-1 implies that A is strongly purely infinite in
sense of Definition 1.2.2, except in some very particular cases, e.g. if A contains
“many projections” in a sense perhaps equivalent to A having “real rank zero”, or
where the lattice of closed ideals of A is linear ordered ...
Recall that a non-zero element a ∈ A+ is called n-homogenous if there exists
a (non-zero) C*-algebra morphism ψ : C0 ((0, 1]) ⊗ Mn 7→ A with a = ψ(f0 ⊗ 1n ),
where f0 (t) := t ∈ (0, 1]. With other words: There exists a non-zero positive
contraction b ∈ A+ (e.g. with b := ψ(f0 ⊗ e1,1 ) and elements c2 , . . . , cn ∈ A with
c∗j cj = b, c∗j b = 0 and c∗j ck = 0 for j 6= k, j, k ∈ {2, . . . , n}. Then the n-homogenous
a ∈ A becomes a = b + c2 (c∗2 ) + · · · + cn (c∗n ).
The comparison between the definitions should be reached with the formula-
tions with help of the J. Cuntz semigroup and by absence of traces ! No (non-zero)
hereditary quasi-traces
Let us finish all this possible ideas concerning pure infiniteness of C*-algebras
A by the following (until here weakest) definition of pure infiniteness:
We say that a C *-algebra A has property pi-n (or: A is pi-n) if, for each
non-zero a ∈ A, the n-fold direct sum a ⊕ a ⊕ · · · ⊕ a = a ⊗ 1n is properly infinite
in Mn (A), if considered as a diagonal matrix in Mn (A).
There is a relation between the two definitions:
104 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Above pi-n says that a ⊗ 1n is s.p.i. in Mn (A) for all a ∈ A. It is not obvious
that this implies that a = b+c2 (c∗2 )+· · ·+cn (c∗n ) (like in Definition 2.0.7) is properly
infinite. But the diagonal matrices with diagonals (a, 0, 0, ..., 0), (b, c2 c∗2 , ..., cn (c∗n ))
and (b, b, ..., b) are equivalent in Mn (A). The diagonal matrix (b, b, ..., b) = b ⊕ b ⊕
· · · ⊕ b is p.i. in Mn (A) if A has the property pi-n of Definition 2.0.6. This matrix
is MvN-equivalent to (a, 0, ..., 0) = a ⊕ 0 ⊕ · · · ⊕ 0 in Mn (A). Thus, the Property
pi-n of Definition 2.0.6 implies the formally weaker Property ”pi-n” of Definition
2.0.7.
It is now the question: What happens with the others: pi-n, pi(n), ...?

Definition 2.0.7. We say that a C *-algebra A is ”locally n-purely infinite” if

(i) No (non-zero) hereditary C *-subalgebra D of A has a (non-zero) charac-

ter.
(ii) Each (non-zero) n-homogenous positive contraction a ∈ A, a := b +
c2 (c∗2 ) + · · · + cn (c∗n ) with c∗k cj = δj,k b, has the property that a ⊕ c - a for
each c ∈ J+ , where J ⊆ A is the closed ideal generated by b (that contains
the element a).

In particular Part(ii) implies that any n-homogenous positive contraction a ∈

A is properly infinite in A. And Part(i) is equivalent to the property that no
irreducible representation of A has a non-zero compact operator in its image.
Below part of old text ?? Begin of change?
property pi-n (or: A is pi-n) if, for each non-zero a ∈ A, the n-fold direct
sum a ⊕ a ⊕ · · · ⊕ a = a ⊗ 1n is properly infinite in Mn (A).
We define in the next section a formally (much?) weaker property of pure
infiniteness on C *-algebras in Definition ??. weak.p.i. ? Each ?????

1. Infinite elements, factorization and transitivity

This section collects basic definitions concerning infiniteness and some needed
technical lemmata that are interesting in itself and play a role in several poofs in
the other chapters. But some of those proofs will be postponed and then given or
outlined in the Appendices A and B, simply to run not too far away from the main
topic of this chapter: A small portion of many ideas for non-commutative versions
of “infiniteness”.
We recall first our later often used definitions of infinite, properly infinite, stable,
n-infinite, n-properly infinite and n-homogenous positive elements of a C *-algebra
A, and some equivalents of it. Notice that we have changed and refined them for our
applications, because those given by other authors have only applications to very
special types of C *-algebras. See e.g. some remarks below and in other chapters
on this differences.
New way:
All the variants of infiniteness imply the last n-homogenous variant.
 1. INFINITE ELEMENTS, FACTORIZATION AND TRANSITIVITY 105

(THAT is not clear now! And, if it is so, then the reachability in n-steps shows
only that all (n + 1)-homogenous elements are properly infinite. It requires the
additional assumption that no hereditary C*-subalgebra D of A has a character.
But it is, in case of non-simple separable C*-algebras A, not enough to show
that all non-zero elements a ∈ A+ are infinite – but for each pure state ρ on
D := aAa there exists an in D non-zero ”infinite” element d ∈ D+ with ρ(d) > 0.)
It is important to consider those definitions that carry over to suitable separable
C*-sub-algebras of A ...
The minimal requirement should be that for every hereditary C*-subalgebra
D ⊆ A and every pure state ρ on D there exists an infinite positive contraction
d ∈ D with ρ(d) > 2/3. ”Infinite” means here that there exists non-zero e ∈ D+
with e ⊕ d -D d ...
No!!! We must require that the ideal J of D generated by the elements e ∈ D+
with the property e ⊕ d -D d contains d, i.e., d ⊕ d -D d and d is properly infinite.
This means that we have to find e ∈ D+ with e ⊕ d -D d and ρ(e) > 0.
How is this related to:
1) No hereditary C*-subalgebra D ⊆ A has a character
(i.e. equivalently (!): no irreducible representation ρ of A contains the compact
operators in its image.),
and
2) There exists (general and fixed) n ∈ N with the property that, for each
a ∈ A+ , b ∈ A+ with b in the closed ideal Ja of A generated by a, and ε ∈ (0, kak),
Pn
there exists d1 , . . . , dn ∈ A with k=1 d∗k adk = (b − ε)+ .
(Can we here replace this by n := f (m) for a suitable function f and the
requirement that a ⊗ 1m is properly infinite in Mm (A) ?)
!!! But it is not clear if we must require (below, later or here) that there exists
suitable e ∈ D+ with ρ(e) > 1/2 and e ⊕ d -D d. !!!
If we require here that there exists e ∈ D+ with ρ(e) 6= 0 and e ⊕ d -D d, then
this would show that d itself must satisfy d ⊕ d -D d, i.e. that d is p.i. (and not
only infinite).
The point in the case of separable C*-algebras A is that in each non-zero here-
ditary C*-subalgebra D of a quotient A/J there exists a properly infinite element in
D. This generates ”maximal” properly infinite elements in the quotients hereditary
C*-sub-algebras of A.
If one has then an extension theorem (that is yet not available) for (rather
special) extensions of separable C*-algebras with p.i. strictly positive elements.
Then one can try to find out if the extension has also a p.i. strictly positive
element.
106 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

The problem is then to show that every separable subset X of A is contained

in a separable C*-sub-algebra B of A, that is again suitably generated by infinite
elements ..., has again this (given) local pure infiniteness ... :
In particular it says: c ⊕ b -B b if b, c ∈ B ⊆ A and c ⊕ b -A b.
Requirements:
(1) No non-zero hereditary C*-sub-algebra D of A has a character. (This is
equal to the property that for every irreducible representation φ of A does not
intersect the algebra of compact operators.)
(2) if D is a non-zero σ-unital hereditary C*-sub-algebra of A, and J 6= A a
closed ideal of A, such that D is not contained in J, then (the requirement is that)
E := D/(D ∩ J) ⊆ A/J contains an infinite element, i.e. there exists non-zero
elements f, g ∈ E+ with g ⊕ f -E f .
But we need the stronger property: For every a ∈ A+ and every closed ideal J of
A with a 6∈ J there exists a non-zero element b ∈ (A/J)+ with b⊕πJ (a) -A/J πJ (a).
It says a is contained in the closed ideal Ja of A defined by: e ∈ Ja , if and only
if, e ⊕ a -A a. The element πJa (a) is always finite in A/Ja .
Thus a ∈ A is properly infinite
(or is ”purely infinite” ??)
in A, i.e., a ⊕ a -A a) if and only if, for every closed ideal J of A πJ (a) is
infinite in A/J or a ∈ J (i.e., that πJ (a) = 0).
(This relation in M2 (E) for E := D/(D ∩ J) is here the same as the relation
g ⊕ f -A/J f in M2 (A/J). One should require more precise that ?????
The point is:
So far, we can all reduce to the separable case (– depending much from the
chosen definition of ”locally infinite” elements! –), then we can reduce our study to
separable A, and then for given d ∈ A+ and let D := dAd there exists a ”maximal”
closed ideal J := eAe = eDe of D with the property that e ∈ D+ is
(1.) strictly positive in J and
(2.) e is properly infinite in D,
This maximality comes in the case of separable D from the fact that there
exists an extremal closed ideal J of D such that J the strictly positive elements
e ∈ J of J are all properly infinite, i.e. e ⊕ e -J e.
This existing (≈-) maximal p.i. contraction e ∈ D+ should turn out (by a
suitable proof) to be strictly positive in D, i.e. J = eAe and e ⊕ e -J e.
The proof is likely indirect:
(1.) We have to show that for each closed ideal of J of D with D 6= J that
each non-zero σ-unital hereditary sub-algebra E of D/J exists a (non-zero) infinite
element e ∈ E+ .
 1. INFINITE ELEMENTS, FACTORIZATION AND TRANSITIVITY 107

that is ????
We could pass to F := {a ∈ D; πJ (a) ∈ E}. Then J ⊆ F ⊆ D and F is
hereditary in D and in A. Moreover, J ⊆ F is a closed ideal of F .
If J is σ-unital then F is again σ-unital (as D by definition).
Since one can all reduce to the separable case (by preserving - with some care
!) we can always suppose and use that all is σ-unital.
It is necessary and sufficient to consider the case of hereditary C*-sub-algebras
D of separable C*-algebras A ... Each closed ideal J of D is the intersection
J = I ∩ D of a closed ideal I of A with D.
This implies that ?????.
and D/J contains again a properly infinite element f ∈ D/J ... ...)
All can be reduced to a suitable separable C*-sub-algebra B of A that contains
a given separable subspace X of A and satisfies that b1 -Mn (B) b2 for b1 , b2 inMn (B)
if and only if b1 -Mn (A) b2 .
(THAT IS NOT obvious ...)
All are compatible with inductive limits (of this separable subspaces).
In this sense:
It is likely that many p.i.-elements appear in the separable case:
There, in hereditary D ⊆ A, are (inside D) extremal (!) positive p.i. elements
d ∈ D+ , in the sense that for every p.i. element e ∈ D+ with d ∈ eDe holds that
e -D d (and automatically d -D e). The separability of A (and then that of D)
secure that the closed ideal J ⊇ D of D generated by d has the property that a
strictly positive contraction of f ∈ J+ has the property that d ≈ f . It implies that
there are elements an , bn ∈ A with a∗n dan = (f − 2−n )+ (by proper infiniteness of
d) and b∗n f bn = (d − 2−n )+ . It shows that for the corresponding open projections
p, q ∈ A∗∗ that define D and J are equal, i.e. we get that D = J.
(In fact then J := dDd is for such d ∈ D+ is always an ideal of D by the
maximality and this could give then also uniqueness! ???).
(NEXT: try to use this to show that ”p.i.” implies ”s.p.i.” ??? Find it in the
appendix ??? !!!!!!)
SOME TECHNICAL CONSIDERATIONS:
Let a, b ∈ A+ with kak ≤ 1 and kbk ≤ 1, then straight calculation shows that
p
2(a − (a − b)+ ) = (a + b) − (a − b)2 .
p
Notice here that (a − b)2 = |a − b|.
a − (a − b)+ and b − (b − a)+ have norms ≤ 2kabk
Pedersen book: Proposition 1.3.8.: If 0 < β ≤ 1 the function t 7→ tβ is operator
monotone for each t ∈ [0, ∞).
108 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

I.e. in a C*-algebra A, then a, b ∈ A+ , 0 < β ≤ 1, a ≤ b implies aβ ≤ bβ .

(we need case β = 1/2).
Does it follow that kbβ − aβ k ≤ kb − akβ , at least for β = 1/2 ?
(cite here Pedersen book ...?)
The relevant formula for the needed decompositions by the estimate for con-
tractions a, b ∈ A+ is given by some estimate like (?? upper next is not really
checked !)
k(a + b) − |a − b|k ≤ 2kabk ,
or ≤ (2kabk)1/2 and the (correct !) equation:
p
2(a − (a − b)+ ) = (a + b) − (a − b)2 .
p
Notice here that (a − b)2 = |a − b|.
Needs study of operator-monotone function:
Something like kx − yk ≤ kx2 − y 2 k1/2 for x, y ∈ A+ ??? or equivalently
1/2
ka − b1/2 k ≤ ka − bk1/2 for a, b ∈ A+ by operator-monotony of the map a ∈
A+ 7→ a1/2 .
SEE APPENDIX ”C” !!! for more.

Definition 2.1.1. An element a ∈ A+ is infinite if there exists non-zero

b ∈ A+ such that, for every ε > 0, there exist d1 , d2 ∈ A (depending on ε) with
kd∗1 ad1 − ak < ε, kd∗2 ad2 − bk < ε and kd∗1 ad2 k < ε . (1.1)

Expressed by use of the Cuntz semigroup as: a ⊕ b -A a in Cu(A).

A non-positive element a in Ped(A) is called “infinite” if a∗ a is infinite.
The element a ∈ A+ is properly infinite if we can take b := a in the second
of the Inequalities (1.1).
Thus a ∈ A and a ⊕ a -A a says that a is a properly infinite element of A.
We call a non-zero element a ∈ A stable if the σ-unital hereditary C *-
subalgebra a∗ Aa is a stable C *-subalgebra of A.

Notice here that a∗ Aa is stable, if and only if, aAa∗ is stable. Following
Lemma shows that Property pi-n of Definition ?? implies Property pi(n) of Defi-
nition 2.0.4.

Lemma 2.1.2. Let A denote a C*-algebra and let a ∈ A a non-zero element in

A. Suppose that there exist ` ∈ N sequences of elements D` , E` ∈ Mn+1 (A) with
lim D` diag(a, . . . , a, 0)E` = diag(a, . . . , a, a) . (1.2)
`→∞

Then 1n ⊗ a is properly infinite in Mn (A) ∼ = Mn ⊗ A, and, for each element

b ∈ J(a) in the closed ideal J(a) of A generated by a and ε > 0, there exists
Pn
f1 , . . . , fn , g1 , . . . , gn with kb − k=1 fk agk k < ε . If a and b are positive then one
can find here the elements fk , gk ∈ A with fk = gk∗ .
 1. INFINITE ELEMENTS, FACTORIZATION AND TRANSITIVITY 109

Let a, b1 , b2 , . . . ∈ A+ contractions with the properties that (a − ε)⊗ 1n and

(bk − ε)+ ⊗ 1n are properly infinite in Mn (A) or zero for each ε ∈ (0, 1].
Then b := (b1 , b2 , . . .) ∈ `∞ (A), (respectively b := (a, a, . . .)) has the property
that h(b) 6∈ K(H) for every irreducible representation h : `∞ (A) → L(H), i.e., h(b)
is not compact if h(b) 6= 0.
In particular, if a ⊗ 1n is properly infinite in Mn (A) for each non-zero ele-
ment a ∈ A+ , then `∞ (A) has no irreducible representations that contains compact
operators in its image.
Especially a ∈ A, respectively (a, a, . . .) and (b1 , b2 , . . .) in `∞ (A) are contained
in the kernel of all irreducible representations of finite dimension of A, respectively
of `∞ (A) .

Proof. The equations 1.2 say that 1n+1 ⊗a - 1n ⊗a in the --terminology of J.

Cuntz, or, even more abstractly expressed, that (n+1)[a] ≤ n[a], in the pre-ordered
“large” Cuntz semi-group Cu(A), cf. Section 5 in Appendix A, our definitions in
Section
section 2.infinitesimal ??
and Definition 2.5.1.
We can repeat the approximation by suitable choice of new operators
D`,m , E`,m ∈ Mn+m (A) beginning with D`,1 := D` , E`,1 := E` and the suit-
able products D`,m := diag(1m , D` ) ∈ Mm+k (A) and E`,m := diag(1m , D` ) ∈
Mm+k (A) to get that 1n+m+1 ⊗ a is the limit of D`,m ((1n+m ⊗ a) ⊕ 0)E`,m for
` → ∞.
The following uses that generally diag(h(b), . . . , h(b)) ∼
= (h ⊗ idn )(b ⊗ 1n ) is
properly infinite in C ⊗ Mn if h(b) 6= 0 and b ⊗ 1n is properly infinite in B ⊗ Mn ,
for each given C *-morphism h : B → C.
Notice here that positive compact operators c ∈ K(H)+ ⊆ L(H) have the
property that ((c ⊗ 1n ) − ε)+ = (c − ε)+ ⊗ 1n is finite or zero for each ε > 0, –
because they have finite rank in K(H) ⊗ Mn ∼
= K(H ⊗ `2 (n)) . In particular, c = 0
if c ≥ 0, c ⊗ 1n ∈ K(H) ⊗ Mn and ((c ⊗ 1n ) − ε)+ = (c − ε)+ ⊗ 1n is properly
infinite or zero for each ε > 0 .
Let B := `∞ (A) and b := (b1 , b2 , . . .) or b := (a, a, . . .) in B+ , respectively
B := A and b := a ∈ A, with the property that (b − ε)+ ⊗ 1n is zero or properly
infinite in B ⊗ Mn .
Suppose that there exists irreducible representation h : B → L(H) such that
h(b) is a non-zero compact operator on H.
Let h : B → L(H) an irreducible representation, and suppose that h(b) is
compact, i.e., h(b) ∈ K(H). Then, for ι := idMn , (h ⊗ ι)(b ⊗ 1n ) ∈ K(H) ⊗
1n ⊆ K(H ⊗2 `2 (n)) and (h ⊗ ι)((b − ε)+ ⊗ 1n ) = ((h(b) ⊗ 1n ) − ε)+ . Thus,
(h ⊗ ι)((b − ε)+ ⊗ 1n ) has non-zero finite rank in the Hilbert space H ⊗2 `2 (n) for
ε ∈ (0, kbk), or ((h(b) ⊗ 1n ) − ε)+ = 0. But, by assumption, (b − ε)+ ⊗ 1n is properly
110 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

infinite if ε ∈ (0, kbk). It follows that ((h(b) ⊗ 1n ) − ε)+ is properly infinite or zero,
and that ((h(b) ⊗ 1n ) − ε)+ has finite rank. Thus, ((h(b) ⊗ 1n ) − ε)+ = 0 for every
ε > 0, i.e., h(b) = 0. It says h(b) = 0 if h(b) ∈ K(H).
Thus, the images of irreducible representations of A and `∞ (A) can not contain
compact operators. In particular, they have no irreducible representations that
contain non-zero compact images in its images. 

If we use Lemma 2.1.9, then we can see that an element a ∈ A+ is infinite if

there exists non-zero b ∈ A+ with the property that for every ε > 0 there exists
δ ∈ (0, ε) and d := d(a, b, ε, δ) ∈ M2 (A) such that

d∗ (a − δ)+ ⊕ 0 d = (a − ε)+ ⊕ (b − ε)+ .

Clearly the element a ∈ A+ is properly infinite if we find for each ε > 0 some δ > 0
and d := d(a, ε) ∈ M2 (A) such that

d∗ (a − δ)+ ⊕ 0 d = (a − ε)+ ⊕ (a − ε)+ .

It is not difficult to obtain from Definition 2.1.1 that positive a ∈ A+ is properly

infinite inside the hereditary C *-subalgebra D := aAa of A if and only if a is
properly infinite in A, cf. the arguments in the proof of Proposition 2.2.5(i).
It implies that every stable element is properly infinite. But (non-zero) properly
infinite projections are obviously not stable.
Sometimes it is useful to allow also a = 0 to be “properly infinite” (respectively
“stable”) in the formulation of some results, but notice that 0 is not an infinite
element by our Definition 2.1.1.
But the Definition 2.1.1 of (non-properly infinite but) infinite elements allows
to conclude only that a is infinite in A if (and only if) a∗ a is infinite inside the
hereditary C *-subalgebra D := a∗ Aa = (a∗ a)A(a∗ a). Therefore a ∈ A is infinite in
A if and only a is infinite inside the closed ideal J(a) := span(AaA) of A generated
by a, cf. the “compression” arguments in the proof of Proposition 2.2.5(i).
By Lemma 2.5.3(v), a non-zero a ∈ A+ is properly infinite if (and only if)
πJ (a) is zero or infinite in A/J for every closed ideal J of A. It implies e.g. that
all non-zero elements in A+ are properly infinite in A if A is purely infinite in the
sense of Definition 1.2.1, which is, almost verbatim, equal to the below considered
property pi(1) in Definition 2.0.4, and this turns later out to be equal to property
pi-1 of Definition 2.0.6, and says that each non-zero a ∈ A+ is properly infinite.
But the ”property pi-1” defined in Definition 2.0.7 seems to be different. be-
cause it requires that a is 2-homogenous. ???
We ask here for some care, because different definitions of infiniteness of ele-
ments in some publications do not imply infiniteness in the sense of our Definition
2.1.1, e.g. H. Lin and Sh. Zhang gave in [532, def. 1.1] the following definition for
positive elements a in the Pedersen ideal Ped(A) :
 1. INFINITE ELEMENTS, FACTORIZATION AND TRANSITIVITY 111

An element a ∈ Ped(A)+ is called infinite in [532], if there are nonzero positive

elements b and c in Ped(A)+ such that bc = cb = 0, and there exist sequences (xn )
and (yn ) in A such that b+c = limn x∗n cxn , and c = limn yn∗ ayn . In our terminology,
this property of a ∈ Ped(A)+ is equivalent to c - a and is slightly stronger than
b ∈ IA (c), i.e., c is infinite and a majorize c. And, in our terminology, this implies
that [a] ≥ [c] in Cu(A) and that c is infinite. Lemma 2.1.6 shows that in case of
simple C *-algebras A the definition [532, def. 1.1] of infinite elements is equivalent
to our notion of infinite elements given in our Definition 2.1.1. The existence of a
nonzero element b ∈ A+ with bc = 0 and b + c - c inside A is not equivalent to
infiniteness of c if the C *-algebra A is not simple. But it is a sufficient condition
for being infinite in non-simple C *-algebras, cf. Lemma 2.5.4.
This definitions of proper infiniteness are not the same in case of non-simple
C *-algebras A :
Let A here denote the unitization of C0 (0, 1] ⊗ O2 : Then 1 ⊗ 1 ≥ f0 ⊗ 1 and the
element f0 ⊗ 1 ∼M vN f0 ⊗ (s1 s∗1 ) is properly infinite in A in the sense of definition
[532, def. 1.1], because f0 ⊗ (s1 s∗1 ) is orthogonal to the element f0 ⊗ (s2 s∗2 ). But
1 ⊗ 1 is finite, because a non-unitary isometry can not be homotopic to 1 in O2
inside the class of isometries. We don’t know if the above cited definition of Lin and
Zhang of infinite elements are equivalent to our definition at least for non-simple A
with real rank zero, (but didn’t see a counterexample). Consider the unitization of
C0 (X, O2 ) for X := C ∩ (0, 1], where C ⊂ [0, 1] is the Cantor subset. The unit is
infinite but is not properly infinite: It can’t be properly infinite because restriction
to a zero-sequence in X shows this by observing that (1, 1, . . .) is not properly
infinite in c0 (O2 ) + C · (1, 1, . . .) ⊂ `∞ (O2 ).
Let A ⊂ `∞ (O2 ) denote the C *-algebra generated by c0 (O2 ) and the unit
element (1, 1, . . .) ∈ `∞ (O2 ), then each element of A is infinite, but (1, 1, . . .) is not
properly infinite in A. This A has real rank zero and each projection is infinite, but
only the elements of c0 (O2 ) are properly infinite.
But for projections a, b, c in a C *-algebra A of real rank zero the projection
a is infinite in sense of our Definition 2.1.1 if one restricts the definition of Lin
and Zhang only to projections a, b, c ∈ A. It says then only that each non-zero
projection a that satisfies their conditions has the property that a = p + q with p
an infinite projection.
We have to consider also several non-simple C *-algebras with non-zero real
rank if we study general purely infinite C *-algebras.
Some conditions allow only to see that diagonal elements diag(a, a, . . . , a) =
a ⊗ 1n ∈ Mn (A) ∼ = A ⊗ Mn are infinite or properly infinite for some unknown
(!) fixed n ∈ N, e.g. this is the case where one can show only that `∞ (A) has no
non-trivial quasi-trace. We are forced to variate and refine the Definition 2.1.1 in
the following manner:
112 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Definition 2.1.3. An element a ∈ A+ is n-infinite (respectively is n-

properly infinite), if a ⊗ 1n is infinite (respectively is properly infinite) in
A ⊗ Mn .
An element a ∈ A+ is n-stable if the hereditary C *-subalgebra D :=
(a ⊗ 1n )(A ⊗ Mn )(a ⊗ 1n ) of A ⊗ Mn contains a stable hereditary C *-subalgebra
E that is full in D.

Notice that here n-stability of a ∈ A+ only means that the stable E ⊆ D

generates the same ideal of A ⊗ Mn as D does.
Important warning: 1-stable elements a ∈ A+ are not necessarily stable elements
in the sense of Definition 2.1.1.
If each element a ∈ A+ is n-infinite (or n-properly infinite), then at least every
n-homogenous element a ∈ A+ is infinite (respectively n-properly infinite) in A,
and it is always contained in the kernel of every irreducible representation of A of
dimension < n. It is also contained in the kernel of Of all irreducible representations
od dimension n if a is n-properly infinite.
We remind the definition of n-homogenous positive elements:

Definition 2.1.4. An element a ∈ A+ is called n-homogenous, if there exists

a C *-morphism ρ : C0 (0, 1] ⊗ Mn → A with kak · ρ(f0 ⊗ 1n ) = a for the generator
f0 (t) = t of C0 (0, 1].

The proof of Proposition A.8.4. or [538, prop. 2.7], give the following equiva-
lent description of an n-homogenous element a ∈ A+ : There exist – not uniquely
determined – contractions z2 , . . . , zn ∈ A with properties zk zj∗ = 0 for j 6= k,
zj zk = 0 for j, k = 2, . . . , n and zj∗ zj = z2∗ z2 for j = 3, . . . , n, such that a =
Pn
kak · (z2∗ z2 + k=2 zk zk∗ ) .
A C *-morphism ρ : C0 (0, 1]⊗Mn → A with kak·ρ(f0 ⊗1n ) = a is then defined
by {z2 , . . . , zn } and the assignments

ρ : f0 ⊗ pk,1 7→ zk .

By Remark 2.1.16, a C *-algebra A contains an n-homogenous element a ∈ A+ with

kak = 1 if A is not a k-sub-homogenous C *-algebra for some k < n.
Recall that C *-algebra A is k-sub-homogenous if A is isomorphic to a C *-
subalgebra of Mk (C) for some commutative C *-algebra C. This happens, if and
only, if all irreducible representations of A have dimension ≤ k. All separable k-
sub-homogenous C *-algebras are C *-subalgebras of Mk ⊗ `∞ = `∞ (Mk ).
Our notions of n-homogenous and of n-stable positive elements have nothing
in common, because n-stable elements a ∈ A+ have the property that a ⊗ 1n is
properly infinite in Mn (A), but n-homogenous a ∈ A+ have not necessarily this
property, as e.g. a := 1n ∈ Mn .
 1. INFINITE ELEMENTS, FACTORIZATION AND TRANSITIVITY 113

Sometimes we use following important property of n-homogenous elements:

If a ∈ A+ is n-homogenous, then a ⊗ p11 ∈ A ⊗ Mn is MvN-equivalent to b ⊗ 1n in
Mn (A) ∼= A ⊗ Mn for suitable b ∈ aAa+ that commutes with a in A.
We allow only in some exceptional cases for all the above given definitions of
properties of elements a ∈ Mn (A) that a = 0, then just to simplify the formulation
of conditions or of results.

Remark 2.1.5. Some authors prefer to use (completely positive) “order-zero

morphisms” ψ : C → A from C to A instead of C *-algebra homomorphisms
h : C0 ((0, 1], C) → A. Those ψ are often used in cases where C = Mn (and more
generally if C is of finite dimension).
The order-zero morphisms ψ are defined as completely positive linear maps
ψ : C → A with kψk ≤ 1 and the property that ψ(c)ψ(d) = 0 for all c, d ∈ C+
with cd = 0. In particular, ψ(c) := h(f0 ⊗ c) is an order-zero morphism if
h : C0 ((0, 1], C) → A is a C *-algebra homomorphism.
We modify in Section 9 of Appendix A arguments in the proofs of [838, thm.2.3]
of M. Wolff and of W. Winter and J. Zacharias in [836], where they show that for
each completely positive order-zero map ψ : C → A there exists a unique C *-algebra
homomorphism h : C0 ((0, 1], C) → A with h(f0 ⊗ c) = ψ(c) for all c ∈ C.
In fact we need for our recovery of h from ψ only the 2-positivity of the linear
map ψ and the orthogonality property of ψ on C+ , i.e., that idM2 ⊗ψ is positive
and that ψ(c)ψ(d) = 0 for c, d ∈ C+ for cd = 0. to obtain the latter, cf. Section 9
of Appendix A.

A crucial starting point for the study of simple purely infinite C *-algebras was
observed first by J. Cuntz. It is the nice observation stated in the following Lemma
2.1.6. Its proof should be an exercise, – e.g. by using an argument that is a bit
more elementary than using properties of the general Cuntz relation - on elements
of A ⊗ K(`2 ).

Lemma 2.1.6. Suppose that A is a simple C*-algebra. Let a, b ∈ A+ . If a ≤ b

and a is infinite in A, then a and b are both properly infinite in A, and a is properly
infinite inside the hereditary C*-subalgebra aAa of A.

This Lemma does not apply to non-simple C *-algebras, cf. Example 2.5.13,
but it is also a very special case of Lemma 2.5.3(i) on the Cuntz majorization - in
arbitrary C *-algebras.
Another often used fact is Part (ii) of the following Lemma 2.1.7 concerning
stably generated ideals, and ideals generated by ideals of C *-subalgebras. The first
statement in each item plays a particular role Consider blue changes! Give
new proofs?// Consider also the remark before proof!// Compare with
Lemma 2.7.3!!
114 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Lemma 2.1.7. Let A a C*-algebra, B ⊆ A a C*-subalgebra, and let J the

closed ideal of A that is generated by B, i.e., J is the closed linear span of the set
of products A · B · A.

(o) The set B · A · B of products b1 ab2 with b1 , b2 ∈ B and a ∈ A is the

same as the hereditary C*-subalgebra E := span(BAB) of A generated
by B, where span( · ) denotes the linear span, i.e., no addition operations
or approximations are needed to get all elements of the hereditary C*-
subalgebra E from the set of products b1 ab2 .
In particular, B is hereditary in A, if and only if, the set B · A · B is
equal to B.
(i) If the unit 1M(B) is properly infinite in M(B), then the hereditary C*-
subalgebra E = B ·A·B of A contains a stable hereditary C*-subalgebra D
of A that generates the same closed ideal J of A as generated by B (also
generated by E).
(ii) If the unit 1M(B) is properly infinite in M(B), and a1 , a2 , . . . ∈ J+ is
any sequence in J, then there exists a sequence d1 , d2 , . . . ∈ A with the
properties d∗j dk = δjk ak and dk d∗k ∈ D ⊆ E .
(iii) In particular, if p ∈ M(A) is a non-zero properly infinite projection,
then pAp contains a stable hereditary C*-algebra D such that the ideal J
generated by D contains pAp.
(iv) If, moreover, B is stable then 1M(B) is properly infinite and the hereditary
C*-algebra E = BAB itself is stable.
(v) (Ideal-separation and ideals of hereditary C*-subalgebras)
Let B ⊆ A a C*-subalgebra of A and denote by JB := Span(ABA) the
closed ideal of A generated by B.
Consider the maps Φ : Y 7→ Span(AY A) from the family of all subsets
Y ⊆ A of A into the family closed ideals of A, and the map Ψ : I 7→ I ∩ B,
from the family of closed ideals I of A into the closed ideals J of B.
Then, Φ(Ψ(I)) = I ∩ JB for each closed ideal I of A, if and only if,
the elements of B separate the closed ideals of the hereditary C*-subalgebra
BAB of A.
If,– in addition –, each closed ideal of J of B is the intersection J =
I ∩ B of a closed ideal I ⊆ A with B then the restriction Φ|I(B) of Φ to
the family of closed ideals J of B is the inverse map of the map I 7→ B ∩I,
i.e., Ψ(Φ(J)) = J each for each closed ideal J of B.
Thus, then J = B ∩ Span(AJA) for all ideals J ∈ B and I =
Span(A(B ∩ I)A) for all closed ideals | of A.
In particular, for all hereditary C*-subalgebras D of A and every
closed ideal I of A, the closed ideal D ∩ I of D generates the closed ideal
I ∩ JD of A and each closed ideal K of D is the intersection K = D ∩ I
of the closed ideal I of A generated by K.

Proof. (o): Let E := span(B · A · B) the closed linear span of the elements
b1 ab2 = (b∗2 a∗ b∗1 )∗ with a ∈ A and b1 , b2 ∈ B.
 1. INFINITE ELEMENTS, FACTORIZATION AND TRANSITIVITY 115

Clearly, E ⊆ A is a hereditary C *-subalgebra of A, and E is a non-degenerate

left and right B-module. Moreover for every countable subset X of E there exists a
sequence of positive contractions b1 , b2 , . . . ∈ B+ such that x = limn bn x = limn xbn
for all x ∈ X. If we let d := n 2−n bn , then
P

X ⊆ d·E·d ⊆ d·A·d ⊆E.

We can apply the factorization theorem of G.K. Pedersen [621, thm. 4.1] to the non-
degenerate C ∗ (d)-bimodule d · E · d ⊆ E, cf. also our special version in Theorem
A.11.1. This allows to see that for each finite sequence e1 , . . . , en ∈ d · E · d ⊆ E
there exists a positive contraction d0 ∈ C ∗ (d)+ ⊆ B+ and elements f1 , . . . , fn ∈ E
such that ek = d0 fk d0 for k = 1, . . . , n. In particular, E is identical with the set
B · A · B of products b1 ab2 with a ∈ A and b1 , b2 ∈ B.
(i): The non-degenerate *-monomorphism defined by the inclusion B ,→ E :=
B · A · B defines a unital *-monomorphism M(B) ,→ M(E) .
By assumptions, there are isometries s, t ∈ M(B) ⊆ M(E) with orthogonal
ranges. Then s1 := s, sn := tn−1 s for n = 2, 3, . . . correspond to a unital C *-
morphism from O∞ into M(E).
Thus, M(E) has a properly infinite unit, i.e., contains a copy of O∞ :=
C (s1 , s2 , . . . ; s∗j sk = δjk 1) unitally. The C *-subalgebra D of E generated by
∗

{sk es∗j ; j, k ∈ N , e ∈ E} is a stable C *-subalgebra of E because D is the closure

of m,n sm Es∗n , which implies that D ∼
S
= E ⊗ K, e.g. via the isomorphism ψ given
∗
by ψ(e ⊗ pkj ) = sk esj for the matrix units pjk of K.
The D is full in E and is hereditary in E, because it is the closure of the union
of the increasing family of split corners Pn EPn of E with Pn := s1 s∗1 + · · · + sn s∗n ,
s1 Es∗1 ⊂ Pn EPn ⊂ E, and s1 es∗1 ∼M vN e inside E for each e ∈ E+ .
Hence, D, E and B generate the same closed ideal J := span(ABA) of A.
(ii): By Part (i), D is isomorphic to E ⊗K via natural transformation with help
of the isometries s1 , s2 , . . . ∈ M(E). (But the sn are not necessarily in M(D).)
If we use that K ⊗ K ∼ = K we get isometries Tn ∈ M(K) with Tj∗ Tk = δjk 1 such
that k Tk Tk∗ strictly converges to the unit 1M(K) . It follows that the isometries
P

Sk := 1 ⊗ Tk are isometries in M(E) ⊗ M(K) ⊆ M(E ⊗ K) such that k Sk Sk∗

P

converges strictly to 1 in M(E ⊗K). The isomorphism ψ from D onto E ⊗K extends

to a strictly continuous *-isomorphism M(ψ) from M(D) onto M(B ⊗ K). Then
the sequence t1 , t2 , . . . ∈ M(D) of isometries tk := M(ψ)−1 (Sk ) has the property
that t∗j tk = δjk 1 and that k tk t∗k converges in M(D) strictly to 1.
P

It is enough to show that for each a ∈ J+ there exists an element h ∈ J with

h h = a and hh∗ ∈ D to obtain the requested elements dn ∈ D :
∗

Indeed, we find for the given sequence a1 , a2 , . . . ∈ J+ a sequence h1 , h2 , . . . ∈ J+

with h∗k hk = ak and hk h∗k ∈ D. Then the elements dk := tk hk satisfy d∗j dk = δjk ak
and dk d∗k ∈ D.
Let a ∈ J+ where J := span(A · D · A) , and let ε > 0.
116 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Then D is full in J, and there are e1 , . . . , em ∈ D+ and f1 , . . . , fm ∈ J with

ka − bk < ε for b := j fj∗ ej fj .
P

If we use Lemma 2.1.9, then we find c ∈ A such that g1 := f1 c, . . . , gm :=

fn c ∈ A with j gj∗ ej gj = (a − ε)+ . The element x := j tj (ej )1/2 gj ∈ A satisfies
P P

x∗ x = (a − ε)+ and xx∗ ∈ D . The latter uses that D is hereditary and tj ej t∗j ∈ D.
In particular x ∈ J. Let ν(t) := (t−ε−δ)+ /(t−ε)+ , the element y := x(1−ν(a))1/2
satisfies y ∗ y = (a − ε)+ − (a − ε − δ)+ .
In this way, we find y1 , y2 , . . . ∈ J with yn yn∗ ∈ D, y1∗ y1 = (a − 1/2)+ , yn∗ yn =
(a − 2−n )+ − (a − 22−n )+ for n > 1. Then h :=
P
n tn yn is an element of the
requested type, because the series is absolutely convergent by

ktn yn k ≤ kyn k ≤ k(a − 2−n )+ − (a − 2−(n−1) )+ k1/2 ≤ 2−n/2 for n > 1,

h∗ h = yn∗ yn = a and hh∗ = limn→∞ ( j,k≤n tj yj yk∗ t∗k ) ∈ D .

P P
n

(iii): There are partial isometries u, v ∈ A with u∗ u = v ∗ v = p and uu∗ +vv ∗ ≤

p . Let B = C ∗ (u, v). Then M(B) = B and 1M(B) = p is properly infinite, E = pAp
and D is the closure of m,n sm Es∗n , where here sn := un−1 v for n = 1, 2, . . . and
S

u0 := p. Notice that psn p = sn and s∗j sk = δjk p .

(iv): The C *-algebra B is stable, if and only if, its multiplier algebra M(B)
contains a sequence t1 , t2 , . . . of isometries with mutually orthogonal ranges, such
that the sum n tn t∗n converges to 1M(B) in the strict topology of M(B), cf. Re-
P

mark 5.1.1(8) for more about “stability”.

The natural map ι : M(B) → M(E) is injective, unital and strictly continuous
for E = BAB. Therefore s1 := ι(t1 ), s2 := ι(t2 ), . . . ∈ M(E) satisfy again
∗
P
n sn sn = 1. Thus, E is stable if B is a stable.

(v): The subsets Φ(Y ) := Span(AY A) of A are the closed ideals of A generated
by non-empty subsets Y of A, in particular JB := Φ(B) is the smallest closed ideal
of A with B ⊆ JB . Notice that Φ(I ∩ B) ⊆ I ∩ JB for all closed ideals of I of A,
and Φ(Φ(Y )) = Φ(Y ) for all non-empty subsets Y ⊆ A of A.
Let I a closed ideal of A. Recall that Ψ(B ∩I) is the closed ideal of A generated
by B ∩ I. Since B ∩ I is contained in the ideal JB ∩ I, it follows that

Ψ(B ∩ I) ⊆ JB ∩ I .

And it implies that B ∩ Ψ(B ∩ I) = B ∩ I, because B ∩ I ⊆ Ψ(B ∩ I) and

B ∩ (JB ∩ I) = B ∩ I by B ⊆ JB .
The above equations show that B ∩(JB ∩I) = B ∩Ψ(B ∩I), which is equivalent
to B ∩(D ∩I) = B ∩(D ∩Ψ(B ∩I)) for the hereditary C *-subalgebra D := B ·A·B,
cf. Part (i).
It follows: The C *-subalgebra B of A separates the closed ideals of D :=
B · A · B, if and only if, B separates the ideals of JB , if and only if, the ideal JB ∩ I
of JB that is generated by B ∩ I for all closed ideals I of JB , i.e., each closed ideal
of JB is determined by its intersection with B.
 1. INFINITE ELEMENTS, FACTORIZATION AND TRANSITIVITY 117

Thus, always B ∩ Ψ(I ∩ B) = B ∩ I, or B does not separate the closed ideals

of BAB, respectively of JB .
It follows then that I ∩ B = Ψ(I ∩ B) ∩ B for all closed ideals I of A. Thus, if B
is separating for the ideals of JB (or – equivalently – for the ideals of D := B · A · B)
then this says that Ψ(I ∩ B) = I ∩ JB for all closed ideals I of A.
But the ideals I ∩B are not all ideals of B if B is not hereditary in A. There are
examples of finite-dimensional C *-algebras A where B can have much more closed
ideals than the special ideals I ∩ B for ideals I of A. 

Example 2.1.8. The stable C *-subalgebra D defined in the proof of Lemma

2.1.7 is not necessarily equal to E, because B := A := O∞ satisfy our assumptions,
but E := O∞ and J = O∞ are not stable.
The behaviour of the isometries and elements constructed during the proof has
to be considered with some care because they are usually not in M(A):
Consider the C *-algebra A := O∞ ⊗ K, and a pure state ρ on O2 . Let L := {c ∈
O2 ; ρ(c∗ c) = 0} the closed left-ideal corresponding to ρ and B := (L∗ ∩ L) ⊗ p11 ⊂
A. Then B is a full stable hereditary C *-subalgebra of A, and the algebras B ⊆ A
satisfy with E = B = D and J = A all the assumptions of Lemma 2.1.7, where here
B itself is full, stable and hereditary, but the in the proof of Lemma 2.1.7(i) defined
and used isometries and projections sn , Pn ∈ M(B) ⊆ M(E) are not elements of
M(J) = M(A). But there exists an element X ∈ A with X ∗ X strictly positive
in A and XX ∗ strictly positive in B that creates an almost canonical isomorphism
from M(B) ⊂ B ∗∗ ⊂ A∗∗ onto M(A) ⊆ A∗∗ by the isometry V ∈ A∗∗ given by the
polar decomposition X = V (X ∗ X)1/2 .

Beginning of transport back from Appendix A:

Replace e.g. all ‘‘Lemma lem:A.old.2.5 (iv)’’ by Remark 2.1.16
The following often used Lemma is [463, lem. 2.4(iii)]. We give here a different
proof that is short, elementary and does not use results about operator monotone
functions. The self-adjoint operator a+ − a− is the polar decomposition of a∗ = a.

Lemma 2.1.9. Suppose that a ∈ M(A), b ∈ A and γ ∈ [0, ∞) satisfy in M(A)

that
0 ≤ b ≤ a + γ · 1.
Then, for every ε > γ, there exists d := d(ε) ∈ bC ∗ (a, b)b ⊆ A with

kdk ≤ 1 and d∗ a+ d = (b − ε)+ .

In particular, there exists d ∈ A with kdk ≤ 1 and d∗ a+ d = (b − ε)+ if a∗ =

a ∈ A and b ∈ A+ satisfy kb − ak < ε .

1/2
Proof. Let ϕ(t) := (t − γ)−1
+ (t − (ε + γ)/2)+ and g := ϕ(b)a+ ϕ(b) .
The function ϕ(t) is zero on [0, µ] for µ := (ε + γ)/2 > γ and is a non-negative
continuous function on [µ, ∞). It is increasing because ϕ(t)2 = 1 − (µ − γ)(t − γ)−1
118 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

for t ∈ [µ, ∞). This and the inequalities 0 < γ < µ < ε imply (t − γ)− · ϕ(t) = 0,
ϕ(t) ∈ [0, 1), 0 ≤ ϕ(b) ≤ ϕ(kbk) · 1. Recall that (b − ε)+ ≤ (b − µ)+ for 0 ≤ µ ≤ ε.
Then kϕ(b)k < 1, ϕ(b) · (b − γ)− = 0 and

(b − ε)+ ≤ (b − µ)+ = ϕ(b)(b − γ)ϕ(b) ≤ g . (1.3)

1/2
We define d as the norm-limit of the elements dn := ϕ(b)(g+1/n)−1/2 (b−ε)+ :
The dn ∈ A are contractions and the sequence (d1 , d2 , . . .) converges in A,
because kϕ(b)k ≤ 1 and Inequality (1.3) imply kd∗n dn k ≤ 1 and, by monotony of
the C *-norms on the cone of positive operators, that

kdn − dm k2 ≤ k g(g + 1/n)−1 − g(g + 1/m)−1 k ≤ 2/ min(n, m) .

It is obvious from the definition of the dn that (b − ε)+ = limn d∗n a+ dn , – and this
convergence is independent from the above shown convergence for the sequence
(dn ).
If a∗ = a ∈ A and b ∈ A+ satisfy γ := kb − ak < ε , then b ≤ a + γ . 

Remark 2.1.10. It seems to be well-known folklore that matrices a = [aj,k ] ∈

Mn (A) have operator norm estimates

kak ≤ n · max{kaj,k k ; 1 ≤ j, k ≤ n} .

Moreover, one gets the better estimate

kak ≤ (n − 1) · max{kaj,k k ; 1 ≤ j, k ≤ n} .

if the “diagonal” sub-matrix LR(n) = diag(a1,1 , a22 . . . , an,n ) of [aj,k ] has only zero
entries,.
The estimate follows from the choice of n sub-matrices LR(`), ` ∈ {1, . . . , n},
each with at most n non-zero suitably chosen entries aj,k , e.g. with (k − j)mod` for
` ∈ {1, . . . , n} in Z/nZ , and the property that LR(`)∗ LR(`) is a diagonal matrix.
Thus, the norm of LR(`) is given exactly by the maximum of the norms of the
entries of LR(`) . Then kak ≤ the sum of the norms of this sub-matrices.
See Section 18 in Appendix B for more details on such sub-matrices.

An immediate consequence of Lemma 2.1.9 is the following observation:

Lemma 2.1.11. Let e1 , e2 , . . . , en ∈ A and γ > (n − 1) · ke∗j ek k for all j > k.

Then there exists a contraction D = [dj,k ] ∈ Mn (A) such that fj∗ fk = δj,k (e∗k ek −γ)+
for the elements fk := e1 d1,k + e2 d2,k + . . . + en dn,k with k ∈ {1, 2, . . . , n} .

Proof. Consider the matrices a, b ∈ Mn (A)+ defined by

a := [e1 , . . . , en ]∗ · [e1 , . . . , en ] and b := diag(e∗1 e1 , . . . , e∗n en ) .

The matrix a − b ∈ Mn (A) has zero diagonal elements, Remark 2.1.10 applies and
says that
ka − bk ≤ (n − 1) · max{ke∗j ek k ; 1 ≤ j < k ≤ n } .
 1. INFINITE ELEMENTS, FACTORIZATION AND TRANSITIVITY 119

Lemma 2.1.9 provides a matrix D = [djk ] ∈ Mn (A) that satisfies D∗ aD = (b − γ)+

and hast norm kDk ≤ 1 . 

We need in Chapter 3 and other places the equivalences for a given element
h ∈ A+ listed in the following Lemma:

Lemma 2.1.12. Let A a C*-algebra, h ∈ A+ and 2 ≤ q ∈ N .

The following properties (a, i)–(a,iii) of h are equivalent:

(a, i) For every ε > 0, there exist e1 , e2 , . . . , eq ∈ A with ke∗j ek − δj,k hk < ε for
j, k ∈ {1, 2, . . . , q} .
(a, ii) For each γ > 0 and ψ ∈ C0 (0, khk]+ with ψ|[0, γ] = 0 there are elements
e1 , . . . , eq ∈ A with e∗j ek = δjk · ψ(h) for j, k ∈ {1, . . . , q} .
(a,iii) Each element g ∈ hAh + satisfies the property in part (a, i) – with g in
place of h .

The following, considerable stronger, properties (b, i)–(b,iv) of h ∈ A+ and the

closed ideal J := J(h) ⊆ A, generated by h, are equivalent to each other:

(b, i) For every a ∈ J+ and ε > 0 there exist elements e1 , e2 ∈ A with

ke∗j ek − δjk ak < ε for j, k ∈ {1, 2} .

(b, ii) For each n > 1 and ε > 0 there exist elements e1 , . . . , en ∈ A with

e∗j ek = δjk (h − ε)+ for j, k ∈ {1, 2, . . . , n}.

(b,iii) For every a ∈ J+ , n ∈ N and ε > 0 there exist d1 , . . . , dn ∈ J such that

d∗j dk = δjk (a − ε)+ for j, k ∈ {1, 2, . . . , n} .

(b,iv) A∞ := `∞ (A)/c0 (A) contains a stable hereditary C*-subalgebra D such

that the element ∆(h) := (h, h, . . .)+c0 (A) is contained in the closed ideal
of A∞ generated by D.

We do not require in part (a, i) that ej e∗j ∈ hAh , e.g. the case A = M2 , q = 2
and h = p11 shows that the properties (a, i) and (b, i) are very different.
The property (b, i) in Lemma 2.1.12 applies to all a ∈ A+ e.g. if the unit element
of M(A) is properly infinite, or if – more general – A contains an approximate unit
consisting of positive contractions eσ that are M-vN equivalent to two orthogonal
pairs of contractions, e.g. there are fσ , gσ ∈ A with fσ∗ gσ = 0, fσ∗ fσ = eσ and
gσ∗ gσ = eσ .

Proof. All items are equivalent if h = 0. It suffices to consider on all places

the cases where khk = 1, kgk = 1 respectively kak = 1, because otherwise we can
rescale the ε > 0, γ > 0, and change the function ψ ∈ C0 (0, khk]+ in a suitable
manner.
Equivalence of (a, i), (a,ii) and (a,iii):
(a,iii) ⇒ (a, i): h ∈ C ∗ (h3 )+ ⊆ hAh+ .
120 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

(a, i) ⇒ (a,iii): Let khk = 1 = kak , g ∈ hAh+ with kgk = 1, ε ∈ (0, 1/2) and
γ := ε/7 . There exists a ∈ A with a∗ = a and kg 1/2 − h1/2 ah1/2 k < γ, because
hAh = h1/2 Ah1/2 . Then kg − h1/2 ahah1/2 k < (2 + γ)γ < ε/2 .
Let µ := ε/(1 + 2kak2 ) , then kah1/2 k2 · µ ≤ µkak2 < ε/2.
By assumption there exists elements e1 , e2 , . . . , eq ∈ A with ke∗j ek − δj,k hk < µ
for j, k ∈ {1, . . . , q} . Let fk := ek ah1/2 . Then kfj∗ fk − δj,k h1/2 ahah1/2 k <
µ · kah1/2 k2 ≤ µkak2 < ε/2 . Hence, kfj∗ fk − δj,k gk < ε .
(a,ii) ⇒ (a, i): Let ε > 0 and take δ := ε/2 and ψ(t) := max(0, t − δ)+ .
(a, i) ⇒ (a,ii): It follows from the special case where ψ := ϕγ for γ > 0 and
ϕγ (t) := max(0, t − γ). Indeed, if ψ ∈ C0 (0, khk]+ with and ψ|[0, 2γ] = 0 is given,
and d1 , . . . , dq ∈ A exists with d∗j dk = δjk · ϕγ (h) , then the elements ek := dk a

1/2
satisfy e∗j ek = δjk · ψ(h) for a := λ(h) with λ(t) := max(0, t − γ)−1 ψ(t) .
In the special case ϕγ (h) = (h−γ)+ we can apply Lemma 2.1.11 to the situation
in Part (a, i):
Let µ := γ/(q + 1) , and find f1 , . . . , fq ∈ A with kfj∗ fk − δj,k hk < µ . By Lemma
2.1.9 there exist contractions gk ∈ A with gk∗ fk∗ fk gk = (h − µ)+ . It follows that
still kgj∗ fj∗ fk gk k ≤ kfj∗ fk k < µ for j 6= k . The inequality q · µ < γ allows
to apply Lemma 2.1.11 and obtain e1 , . . . , eq ∈ A with e∗j ek = δjk · (h − γ)+ for
j, k ∈ {1, . . . , q} .
Equivalence of (b, i)–(b, iv):
We use the equivalence of (a, i)–(a,iii) to prove equivalence of (b, i)–(b,iv).
The implications (b,iii) ⇒ (b, i) and (b,iii) ⇒ (b, ii) are obvious. We show the
implications (b, i) ⇒ (b,iii), (b,ii) ⇒ (b,iv) and (b,iv) ⇒ (b, i).
(b, i)⇒(b,iii): The equivalence of (a, i) and (a, ii), applied to a ∈ J+ in place
of h and q := 2, shows that the property in Part (b, i) is equivalent to the property
that, for every a ∈ J+ and every ε > 0, there exist elements e1 , e2 ∈ A with
e∗j ek = δj,k (a − ε)+ , for j, k ∈ {1, 2}. Notice that all e1 , e2 ∈ A with this property
are automatically in J.
We proceed by induction over n ≥ 2 and suppose that we have shown that for
every a ∈ A+ and every ε > 0 there exist elements e1 , e2 , . . . , en ∈ A – depending
from a and ε – that satisfy the following equations 1.4:

e∗j ek = δj,k (a − ε)+ for j, k ∈ {1, 2, . . . , n} . (1.4)

The assumptions in (b, i) allow to find n + 1 elements e1 , . . . , en , en+1 ∈ A that
satisfy Equation (1.4):
Let ε > 0, a ∈ J+ and γ := ε/3 and 2 ≤ n ∈ N . There exists d1 , d2 , . . . , dn ∈ A
with with d∗j dk = δj,k (a − γ)+ by induction assumption. This implies that dk ∈ J,
Pn
dk d∗k ∈ J+ and d∗j dk = 0 for j, k ∈ {1, 2, . . . , n} . Let b := ∗
k=1 dk dk ∈ J+ .
Pn ∗ ∗
(b − γ)+ = k=1 (dk dk − γ)+ because dj dk = 0 for j 6= k .

By assumptions in Part (b, i), there exists f1 , f2 ∈ A with fj∗ fk = δj,k (b − γ)+ .
 1. INFINITE ELEMENTS, FACTORIZATION AND TRANSITIVITY 121

The orthogonality of the fk fk∗ , uniqueness of polar decomposition allow to show

1/2 1/2 1/2
with functional calculus that (b − γ)+ dk = (dk d∗k − γ)+ dk = dk (d∗k dk − γ)+ =
1/2
dk (a − 2γ)+ . Thus, (f` dj )∗ f` dk = δj,k (a − 2γ)+ (a − γ)+ . Recall that ε = 3γ and
(well-) define a function ϕ ∈ C0 (0, kak]+ by

1/2
ϕ(t) := (t − ε)+ / ((t − 2γ)(t − γ)) .
Then ek := f1 dk ϕ(d∗k dk ) for k ∈ {1, . . . , n} and en+1 := f2 d1 ϕ(b) satisfy e∗j ek =
δj,k (b − ε)+ for j, k ∈ {1, . . . , n, n + 1} .
(b,ii)⇒(b,iv): By assumptions of Part (b,ii) we can find ek,n ∈ A , k ∈
{1, . . . , n} with the property
e∗j,n ek,n = δj,k (h − 2−n )+ for j, k ∈ { 1, . . . , n } .
Define ek,n := 0 for k > n, and elements Rk ∈ A∞ := `∞ (A)/c0 (A) by
Rk := (ek,1 , ek,2 , . . . , ek,n , ek,n+1 , . . .) + c0 (A) .
Then Rj∗ Rk = δj,k · ∆(h) for all j, k ∈ N. Thus, the closed linear span of the
elements Rj Rk∗ ∈ A∞ for j, k ∈ N generate a stable hereditary C *-subalgebra D of
A∞ ( 6 ). The ∆(h) is contained in the closed ideal of A∞ that is generated by D .
(b,iv)⇒(b, i): Let D ⊆ A∞ a stable hereditary C *-subalgebra of A∞ :=
`∞ (A)/c0 (A) and let I ⊆ A∞ the closed ideal of A∞ that is generated by D. If
(h, h, . . .) + c0 (A) =: ∆(h) ∈ I then ∆(a) ∈ I for each a ∈ J = span(AhA) .
By Lemma 2.1.7(ii), there exist for each element b ∈ I+ elements c1 , c2 ∈ I with
c∗1 c2
= 0 and c∗1 c1 = c∗2 c2 = b. If we apply this to b := ∆(a) and take, for k ∈ {1, 2},
(k) (k) (k) (k)
elements (d1 , d2 , . . .) ∈ `∞ (A) with norms = kck k and (d1 , d2 , . . .)+c0 (A) = ck
then
∗ (k)
lim k(d(j)n ) dn − δj,k bk = 0 for j, k ∈ {1, 2} .
n→∞
(1) (2)
Thus, the inequality in Part (b, i) is satisfied if we let e1 := dn and e2 := dn for
suitable n ∈ N. 

Remark 2.1.13. It is not clear if the equivalent properties in Parts (b,i)–(b,iv)

of Lemma 2.1.12 are also equivalent to the following version of “local stability”:
For every ε > 0 there exists a stable hereditary C*-subalgebra Dε of A such that
(h − ε)+ is in the closed ideal of A generated by Dε . ( 7 )

Remark 2.1.14. The implication (b, ii) ⇒ (b,iii) in Lemma 2.1.12 was very
indirectly obtained in the proof of Lemma 2.1.12 by the row of implications:
(b, ii) → (b,iv) → (b, i) → (b,iii).
Here is an alternative direct proof for the implication (b, ii) ⇒ (b,iii) :
Let a ∈ J+ , n ∈ N and ε > 0 . Since a is a positive element in the ideal
generated by h, there exist m ∈ N and elements g1 , . . . , gm ∈ A and γ ∈ (0, ε/2)
6 This D is isomorphic to K ⊗ C ∗ (h)
7 It could be that this can be translated into a question about the existence of a full stable

hereditary C *-subalgebra in a certain hereditary C *-subalgebra of an infinite amalgamated free

product.
122 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Pm
such that ka − j=1 gj∗ (h − γ)+ gj k < ε/2 . Let q := mn. By Part (b, ii) there
exists q elements ej,k ∈ A with e∗j,k ej,k = (h − γ)+ (1 ≤ j ≤ m, 1 ≤ k ≤ n).
Pm ∗
The elements fk := 6 k and
j=1 ej,k gj have the properties f` fk = 0 for ` =
∗
ka − fk fk k < γ for `, k ∈ {1, . . . , n} .
By Lemma 2.1.9 there exists contractions dk ∈ A with d∗k fk∗ fk dk = (a − ε)+ .
Thus, e∗j ek = δj,k (a − ε)+ for ek := fk dk and j, k ∈ {1, . . . , n}.

Some old citation changes/compare again:

Proposition A.21.4 (A.1.24 or: A.old.2.4)
Proposition A.21.4 (or: K.2.4, A.old.2.4 also in old: A.8.1)
Only 1-cited in proof of Lemma 2.2.3.
Lemma A.6.1 (old A.1.9.)
Proposition A.21.4 and/or Lemma lem:A.old.3.4c = ?? Lemma 2.1.22
(old A.1.24 (= lem:A.old.2.4) - old A.1.28)
Lemma 2.1.15 covert by new lem:2.advance.Kadison.transit exhausting
old: lem:A.old.3.4b or lem:2.6-1
The Part (i) of the following Lemma 2.1.15 is one of the possible characteriza-
tion of projections in the “socle” of bi-duals A∗∗ ( 8 ). The observation that they
are “closed” projections imply a very important and often used “excision” result.

Lemma 2.1.15 (Generalized Kadison transitivity and Excision).

Let A a C*-algebra and p ∈ A∗∗ a non-zero projection.

(i) The W*-subalgebra pA∗∗ p of A∗∗ has finite dimension, if and only if, there
are pairwise inequivalent irreducible representations Dk : A → L(Hk ) and
projections qk ∈ L(Hk ) of finite rank such that for the natural normal
(and unital) extensions of Dk to A∗∗ satisfy
(D1 ⊕ . . . ⊕ Dn )(pap) = q1 D1 (a)q1 ⊕ . . . ⊕ qn Dn (a)qn
for a ∈ A and that (D1 ⊕ · · · ⊕ Dn )|pA∗∗ p is faithful on pA∗∗ p ( 9 ).
(ii) (Generalized Kadison transitivity.) If p ∈ A∗∗ is a projection such that
M := pA∗∗ p is finite-dimensional, then the natural C*-morphism π from
{p}0 ∩A into pA∗∗ p, given by a 7→ ap , is an epimorphism, and there exists
a C*-morphism
ψ : C0 ((0, 1], M ) → {p}0 ∩ A
such that π(ψ(f )) = ψ(f )p = f (1) for all f ∈ C0 ((0, 1], M ).
(iii) (Generalized Excision Lemma.) Suppose that M := pA∗∗ p is finite-
dimensional, Ω ⊆ A is a compact subset of A and
ψ : C0 ((0, 1], M ) → {p}0 ∩ A
8 “socle” is french, but the english translation “base” does not fit here.
9 Our notation does not distinguish between a representation D of A and its normal extension

to A∗∗ , given by Π ◦ D∗∗ via the natural normal *-epimorphism Π : L(H)∗∗ → L(H).
 1. INFINITE ELEMENTS, FACTORIZATION AND TRANSITIVITY 123

is any C*-morphism such that

π(ψ(f )) = ψ(f )p = f (1) for all f ∈ C0 ((0, 1], M ) . (1.5)

Then there exists a positive contraction g ∈ A+ with gp = p such that

lim kg n a − ψ(f0 ⊗ pap) g n k = 0 for all a ∈ Ω ,

(1.6)
n→∞

where f0 (t) := t, i.e., (f0 ⊗ pap)(t) = t · pap ∈ pA∗∗ p for t ∈ [0, 1] .

We can Ω replace by the compact set Ω ∪ {ψ(f0 ⊗ c) ; c ∈ M, kck ≤ 1}, without

change of other assumptions or conclusions in Part (iii).
Notice that b − ψ(f0 ⊗ pbp) ∈ L∗ + L for all b ∈ Ω , because L∗ + L is the kernel
of A 3 a 7→ pap, where L ⊆ A denotes the closed left-ideal

L := {a ∈ A ; p(a∗ a)p = 0 } .

A suitable decomposition of Ω could help / can help / to give a proof of Parts (ii)
and (iii). See the arguments for the special case where L is a left-kernel of a pure
state on A.

Proof. (i): If one does not go the “rigorous” way mentioned in our Remark
2.1.19 and use instead e.g. G.K. Pedersen’s improved version [616, thm. 2.7.5] of
the Kadison transitivity theorem, then we have to show before at least the (almost
obvious?) existence of some normal unital representation H : A∗∗ → L(H) with the
property that the restriction to pA∗∗ p is faithful and that H(p) has finite rank in
L(H), i.e., it is necessary to show first that H can be selected with the additional
property H(p) ∈ K(H). The following is such an elementary argument that shows
the existence of H with this for the application of [616, thm. 2.7.5] very needed
properties:
Let p∗ p = p ∈ A∗∗ nonzero. Suppose that pA∗∗ p has finite (linear) dimension.
Then there are projections q1 , . . . , qn in the center of pA∗∗ p such that qk A∗∗ qk =
qk pA∗∗ p ∼
= Mnk and q1 + . . . + qn = p. We find “minimal” projections 0 6= ek ≤ qk
with ek A∗∗ ek = C · ek . It implies that central projections Q ∈ Z(A∗∗ ) can only
satisfy one of Qqk = qk or Qqk = 0. Moreover, Qek = 0 implies Qqk = 0. Let Rk the
maximal central projection with Rk ek = 0 and let Qk := 1 − Rk . The Qk ∈ Z(A∗∗ )
is the central support projection of ek and A∗∗ Qk is a type-I-factor. By definition of
the Qk holds pQk = qk , Qk Qj = 0 for k 6= j and p(Q1 + . . . + Qn ) = p. Thus, there
are Hilbert spaces Hk and normal *-isomorphisms dk : A∗∗ Qk → L(Hk ) from A∗∗ Qk
onto L(Hk ). Let Dk (a) := dk (aQk ) for a ∈ A. Here we identify A with its natural
image in the W*-algebra A∗∗ . Clearly the restrictions Dk |A to A of the normal
*-representations Dk of A∗∗ are pairwise non-equivalent irreducible representations
of A and the normalization D of the direct sum (D1 |A) ⊕ · · · ⊕ (Dn |A) of this
irreducible representations is identical with

D1 ⊕ · · · ⊕ Dn : A∗∗ → L(H1 ⊕2 · · · ⊕2 Hn ) .

Moreover D(p) is a projection of finite rank Rk(p) ≤ Dim(pA∗∗ p).

124 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Conversely, suppose that p ∈ A∗∗ is a projection and that pairwise inequivalent

irreducible representations Dk : A → L(Hk ), k = 1, . . . , n and projections qk ∈
L(Hk ) of finite rank exist such that, for a ∈ A and the normal extensions of Dk to
A∗∗ (again denoted by Dk ), holds

(D1 ⊕ · · · ⊕ Dn )(pap) = q1 D1 (a)q1 ⊕ · · · ⊕ qn Dn (a)qn

and that (D1 ⊕ · · · ⊕ Dn )|pA∗∗ p is faithful on pA∗∗ p. Now use that irreducible
*-representations are always non-degenerate, direct sums of non-degenerate *-
representations are non-degenerate, and that normalizations of non-degenerate
*-representations are unital. It follows that the direct sum D := D1 ⊕ · · · ⊕ Dn
of the normal extensions of Dk to A∗∗ is unital, satisfies D(p) = q1 ⊕ . . . ⊕ qn , –
i.e., D(p) has finite rank – and D|pA∗∗ p is faithful. Thus, pA∗∗ p has finite linear
2
P
dimension (equal to sum of squares of the ranks k Rk(qk ) ). This properties
show that the above selected representation D : A → L(H) satisfies all assumptions
of [616, thm. 2.7.5].
(ii): With Part(i) in hand, we are in position to use [616, thm. 2.7.5] and
get the desired surjectivity of the natural C *-morphism π : {p}0 ∩ A 7→ pA∗∗ p with
π(a) := ap = pap for a ∈ {p}0 ∩ A.
Alternatively, – but with some in Remark 2.1.18 given necessary additional
observations –, one could use also [400, thm. 5.4.5] and its generalization by S. Sakai
[704, thm. 1.21.16] to derive this surjectivity.
In fact we decide here to use Part(i) to obtain – from the assumption that pA∗∗ p
is of finite dimension – the following special kind of *-representation D exists:
Let D a normal unital *-representation of A∗∗ onto a von-Neumann algebra in
some L(H) and p ∈ A∗∗ is a projection such that D|pA∗∗ p is faithful and D(p) is
of finite rank, i.e., Dim(D(p)L(H)) < ∞ . Then there exists a central projection
Q ∈ Z(A∗∗ ) with Qp = p such that D|QA∗∗ Q is faithful (and again unital).
Thus, the natural C *-morphism ({p}0 ∩ A) 3 a 7→ pa ∈ pA∗∗ p is surjective by
this observation and [616, thm. 2.7.5].
(See Remark 2.1.18 or Remark 2.1.19 for alternative proofs of this surjectivity.)
Let π denote the natural *-epimorphism from A0 := {p}0 ∩A onto M := pA∗∗ p,
and let J := ker(π), and consider the evaluation map η : CM → M given by
f 7→ f (1) =: η(f ) from CM := C0 ((0, 1], M ) onto M . By Proposition A.8.4,
the cone CM := C0 ((0, 1], M ) is projective. Thus, there exist a C *-morphism
ψ : CM → A0 ⊆ A with π ◦ ψ = η.
(iii): Recall that M := pA∗∗ p and CM := C0 ((0, 1], M ) . By Part(ii), there
exists a C *-morphism ψ : CM → A0 := {p}0 ∩ A that satisfies π(ψ(f )) = f (1) =
pf (1)p for all f ∈ CM .
The separable case has to do with Proposition A.15.2 and the determination
of the multiplicative domain of πL∗ +L : A → A/(L∗ + L) considered in Theorem
A.16.5(iii). ?????
 1. INFINITE ELEMENTS, FACTORIZATION AND TRANSITIVITY 125

The excision result needs Lemma A.21.1 on peak elements.

The non-separable case can then be considered by selecting a suitable separable
C *-subalgebra B of A that contains ψ(CM ) and the given separable subset. This
can be done with help of the separable selection Lemma B.14.1 and Proposition
A.15.2.
Later ??? bring ??? Proposition A.21.4 back to Chapter 2.???
Or refers in place of his proof to here???
To get more precise in calculations we discuss some topics ... with help of listed
below/above.. ???
Here, p ∈ A∗∗ is a projection such that M := pA∗∗ p is a C *-algebra of fi-
nite (linear) dimension, π denotes the restriction to A0 of the completely positive
contraction Vp (a) := pap for a ∈ A∗∗ ⊇ A.
The C *-subalgebra B0 of A0 generated by ψ(CM ) is separable and B0 /(B0 ∩
J) = A0 /J ∼
= M is unital, where J ⊂ A0 is the kernel of π.
Obviously (why ???) J = D := A ∩ ((1 − p)A∗∗ (1 − p)) is a hereditary C *-
subalgebra of A and is equal to the kernel ideal of π : a ∈ A0 7→ pa = pap ∈ A∗∗ .
L := A ∩ (A∗∗ (1 − p)), A =? = L∗ + L + A0 , D = L∗ ∩ L = J, A0 = B0 + D =
N (A, D).
By Proposition A.15.2, L∗ + L is a closed linear subspace of A and the natural
map A/(L∗ +L) → pA∗∗ p induced by π : A 3 a 7→ pap ∈ pA∗∗ p defines a completely
isometric isomorphism from (A/(L∗ + L))∗∗ onto pA∗∗ p =: M . Since π(ψ(CM )) =
M It follows that A = L∗ +L+ψ(CM ). Let A1 denote the separable C *-subalgebra
of A generated by Ω ∪ ψ(CM ). By Lemma B.14.1, there exists a separable C *-
subalgebras B of A with the property that A1 ⊆ B and B ∩ (L∗ + L) = (B ∩
L)∗ + (B ∩ L) and the natural map from B/ (B ∩ L)∗ + (B ∩ L) to A/(L∗ + L) is

completely isometric and completely positive.

Since A/(L∗ + L) ∼ = M and πL∗ +L ◦ ψ = idM it follows that we can repeat the
above consideration in case for separable A now for B and Ω ∪ ψ(CM ).
Here possible end of proof,
if separable case is before completely shown.
Or is it below?
The crucial points of the proof are to show that the projection p is closed, and
that the open support projection qD ∈ A∗∗ of D is equal to 1−p and J = D = L∗ ∩L.
Then e ∈ A+ with kek = 1, e(1 − e) ∈ D, ep = p and e(1 − e) strictly positive in
D exists. By separability of D (if A is separable) and p is the unit of M = pA∗∗ p.
Apply the Lemma A.21.1 to E := π −1 (C · p) ⊆ A0 and J to get a positive
contraction g ∈ E with the properties that π(g) = p and g − g 2 ∈ J+ is strictly
positive in J.
Then g is strictly positive in A and has the excision property of Equation (1.6).
126 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

(1) in case where A is separable, that A = A0 + DA + AD and that there exists

g ∈ (A0 )+ with kgk = 1, π(g) = p and g − g 2 strictly positive in D. (Because then
g has obviously the in (iii) proposed properties ????
A0 = N (A, D) = D + ψ(CM ) ???
(2) Reduction of the general case to the separable case if A is non-separable:
Find (!!!) separable C *-subalgebra B of A with ψ(CM ) ∪ Ω ⊆ B such that
F := B ∩ D satisfies B = BF + F B + ψ(CM ) and B/(BF + F B) ∼ = M naturally.
(The point is to manage that N (B, F ) ⊆ A0 and BF = B ∩ (AD).???)
Could it be that dist(b, BF ) = dist(b, AD) for all b ∈ B is sufficient?
Notice that ψ(CM ) ∩ D is an ideal of ψ(CM ).
Then apply above arguments to separable B (in place of A).
The argument works well if D = J is σ-unital. Reduce to this case !!! ???
Next to be filled in / corrected !?
It is necessary to use characterizations of open and closed
projections -- as considered in Appendix A.
Take better all to there.
And 1 − p ∈ A∗∗ is the “open” support projection of D, i.e., D∗∗ ∼
= wcl(D) =
∗∗
(1 − p)A (1 − p).
We consider first the case where A is separable:
Cite here the Prop. ?? on closed/open projection!!!
Existence of contraction e ∈ A+ with pep = p has to be deduced!!!
There exists e ∈ A+ with pep = p and kek ≤ 1 by Part (ii), e.g. e := ψ(f0 ⊗1M ).
Let f ∈ D+ a strictly positive contraction for D := L∗ ∩ L, where L :=
A ∩ (A∗∗ (1 − p)).
Then D = A ∩ (1 − p)A∗∗ (1 − p) is the kernel of the C *-morphism A0 = {p}0 ∩
A → pA∗∗ p, because a ∈ {p}0 ∩A, if and only if, a ∈ A∩(pA∗∗ p+(1−p)A∗∗ (1−p)).
And pap = 0 if and only if a ∈ (1 − p)A∗∗ (1 − p)) .
We get for the above taken e, f ∈ A+ that e, f ∈ {p}0 ∩ A = N (A, D), and that
there exists g ∈ ({p}0 ∈ A)+ with kgk ≤ 1, pgp = p and g(1 − g) is strictly positive
in D .
Build g from e and f . !!!!!!
a 7→ pap = V (a) has kernel A · D + D · A ???? V (a − ψ(f0 ⊗ V (a))) =
V (a) − pV (a) = 0 ???
g n d, dg n → 0 for all d ∈ D, because g(1 − g) is strictly positive in D.
Way of arguments:
1) p is closed,
 1. INFINITE ELEMENTS, FACTORIZATION AND TRANSITIVITY 127

2) the open 1 − p is the right support projection of the closed left ideal L :=
A ∩ A∗∗ (1 − p). (Thus L∗∗ = A∗∗ (1 − p).)
3) pA∗∗ p = pAp ∼
= A/(R + L) for R := L∗ := {a∗ ; a ∈ L}
4) {p}0 ∩ A = N (A, R ∩ L)
5) The completely positive map V (a) := pap from A onto pA∗∗ p has kernel
L + L ?, i.e., ((1 − p)A∗∗ + A∗∗ (1 − p)) ∩ A = L∗ + L, for L := A ∩ (A∗∗ (1 − p)) =
∗

{a ∈ A ; ap = 0} by p closed?
6) Cite here definition of N ?????
A = L∗ + L + N (A, L∗ ∩ L)
7) D := L∗ ∩ L is ideal of N (A, D), N (A, D) = {p}0 ∩ A.
8) (M := pA∗∗ p 6= {0} is finite-dimensional) Ω ⊆ A The compact subset Ω of
A and the image of ψ : C0 ((0, 1], M ) → A generate a separable C *-subalgebra B
of A.
9) Can enlarge B to a separable C *-subalgebra C of A such that B ⊆ C,
C ∩(R+L) = (C ∩R)+(C ∩L), (C ∩D)·C = L∩C, N (C, C ∩D)+C ∩(R+L) = C,
{p}0 ∩ C = N (C, C ∩ D)
(C ∩ R) + (C ∩ L) is always closed. See Appendices ??? behind Reduction to
separable.
More ??
Now proceed as in case of separable A.
Fact list:
1. p is a closed projection with

p ∈ wcl({p}0 ∩ A) ∼
= {p}0 ∩ A∗∗ = (1 − p)A∗∗ (1 − p) + pA∗∗ p

Needs that pa∗ ap = 0 if and only if a ∈ A∗∗ (1 − p) .

π extends to the c.p. contraction V : A 3 a 7→ pap ∈ pA∗∗ p
i.e., {p}0 ∩ A is the multiplicative domain of V .
1a.
J := kernel of π is the same as D := A ∩ (1 − p)A∗∗ (1 − p), because D ⊆ A0
and D ⊆ J. a ∈ J ⊆ A0 implies pa = ap = 0, thus a ∈ A and a = (1 − p)a(1 − p).
and
1b.
1 − p ∈ A∗∗ is the open support projection of D,
???? Needs pA∗ p ∼
= (pA∗∗ p)∗ is σ(A∗ , A)-closed ???
Unit ball of pA∗ p is equal to unit ball of X > , for X := {a ∈ A ; v(a) = 0, v ∈
∗
pA } ???
128 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

because J = D ⊆ (1 − p)A∗∗ (1 − p) and the open support projection qD ∈ A∗∗

of D is contained in the weak closure of of A0 in A∗∗ by J = D and satisfies
qD + p ≤ 1, i.e., (1 − qD ) ≥ p. If (1 − qD ) − p 6= 0, then
A0 /J ∼
= M = pA∗∗ p via implies that wcl(A0 )/(qD A∗∗ qD ) ∼
= M.
Needs qD A∗∗ qD = wcl(D) ⊆ (1 − p)A∗∗ (1 − p)
The map (1 − qD )A∗∗ (1 − qD ) → pA∗∗ p given by (1 − qD )a(1 − qD ) 7→ pap is
surjective, and is multiplicative for a in the weak closure of A0 in A∗∗ .
because p is a closed projection ????
why it is closed??? pAp = pA∗∗ p – confirmed in Part (ii).
1c.
The set L := A·D of products a·d is a closed left ideal of A, because the closed
linear span of L is obviously contained in A and L = L · D by Cohen factorization.
(The arguments for this are similar to those in Part (o) in proof of Lemma
2.1.7.)
with A/(L∗ + L) ∼
= M ???
To be shown.
D := L∗ ∩ L = L∗ · L is kernel of π : {p}0 ∩ A → M . D = ((1 − p)A∗∗ (1 − p)) ∩ A
? by L = (A∗∗ (1 − p)) ∩ A
Needs: p closed, and 1 − p support of D.
q ∈ A∗∗ open then qAq ⊆ A
??? (If A is unital, then this is wrong in general, because then this implies
pinA
L= A·D ?
The C *-morphism φ defines a linear map λ from A into L∗ + L given by
λ(a) := a − φ(V (a)).
Let {dn } ∈ D+ an approximate unit of D given by contractions
The rest is reduction to separable case.
We enlarge B0 to the still separable C *-algebra B1 := C ∗ (Ω∪B0 ) . The natural
map b 7→ pbp is a completely positive contraction that maps the unit-ball of B1 onto
the unit ball of pA∗∗ p and maps φ(f0 ⊗ x) to x ∈ pA∗∗ p.
But this is not enough to find in B1 the desired positive contraction g. ....????
Let J the kernel ideal of {p}0 ∩ A → M := pA∗∗ p = pAp. Then one can show
that L := A·J is a closed left ideal of A such that L∗∗ = A∗∗ (1−p) (by noticing that
p is a closed projection and 1 − p is an open projection), that φ : A 3 a 7→ pap ∈ M
satisfies φ−1 (0) = L∗ +L and defines a completely positive and completely isometric
isomorphism from the C *-system A/(L∗ + L) onto M .
 1. INFINITE ELEMENTS, FACTORIZATION AND TRANSITIVITY 129

If B is some separable C *-subalgebra of A such that φ maps the closed unit-ball

of B onto the closed unit-ball of M , then this does not imply that B/(φ−1 (0) ∩ B)
is completely isometric to M .
We must do a careful selection of a bigger separable C *-algebra Be ⊇ B of
A such that the kernel ideal I of π|Be : {p}0 ∩ Be → M defines a closed left-ideal
Le := Be · I with the property that the natural map from Be /(L∗e + Le ) onto M is
completely positive and completely isometric.
The key for this inductive selection is the Lemma B.14.1.
We get from Lemma B.14.1 a separable C *-subalgebra Be ⊇ B such that
Be ⊆ BAB, (Be ∩ L)∗ + (Be ∩ L) is closed in Be , that it is the same as Be ∩ (L∗ + L)
and that the natural map from Be /(Be ∩ (L∗ + L)) to A/(L∗ + L) ∼ = pA∗∗ p defines
a surjective completely positive complete isometry from ({p}0 ∩ Be )/(J ∩ Be ) onto
pA∗∗ p.
NOTICE 31.12. 2017:
If B ⊆ A is a C *-subalgebra of A and L ⊆ A is a closed left ideal and R ⊆ A is
a closed left ideal, then R + L is a closed subspace of A by Part (iv) of Proposition
A.15.2.
This implies that B ∩ (R + L), B ∩ R and B ∩ L are closed linear subspaces of
A. Thus, B ∩ R and B ∩ L are closed right and left ideals of B. Again, by Part(iv)
of Proposition A.15.2, (B ∩ R) + (B ∩ L) is a closed linear subspace of B, obviously
contained in the closed linear subspace B ∩ (R + L) of B.
By our pre-assumption that (B ∩ R) + (B ∩ L) is dense in B ∩ (R + L), we get
the equality
B ∩ (R + L) = (B ∩ R) + (B ∩ L) .
(Without this “density assumption” the formula would be wrong: A : C ⊕ C,
B := C · (1, 1), L := C ⊕ 0, R := 0 ⊕ C. A = L + R, B ∩ L = 0, B ∩ R = 0 and
B ∩ (L + R) = B.)
But requires an additional selection process that also shows that

B/(B ∩ (R + L)) → A/(R + L)

is isometric, respectively, is completely isometric, by tensoring all with Mn ...

Needs to show that dist(b, R + L) = dist(b, B ∩ (R + L)) ...
This seems to work only with a suitable iteration process, until B ⊗ K ⊆ A ⊗ K
is relatively weakly injective.
But above procedure has to be suitably changed to an inductive procedure that
allows to construct a suitable separable C *-subalgebra B ⊆ C ⊆ A by Corollary
A.15.3.
Moreover it hold then that the natural C *-monomorphism {p}0 ∩Be → {p}0 ∩A
is surjective by separability of A/(R + L).
???? WHERE it is SHOWN ????
130 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

L ∩ Be = Be · (J ∩ Be ).
Recall here that J ⊆ {p}0 ∩ A is the kernel of the surjective C *-morphism
{p}0 ∩ A → pA∗∗ p.
This implies that Be · (J ∩ Be ) = L ∩ Be , Be /(L ∩ Be + L∗ ∩ Be ) ∼
= M . and
∼
Be /(J ∩ Be ) = M , with all isomorphisms canonical.
It follows that the separable C *-algebra J ∩ Be contains a strictly positive
contraction and the C *-subalgebra π −1 (C · 1M ) ⊆ ({p}0 ∩ Be ) contains a positive
contraction a0 with π(a0 ) = 1M and (1 − a0 )a0 strictly positive in J ∩ Be .
Here the set Ω ⊆ A is not considered.
Consider e.g. the separable C *-subalgebra A1 := C ∗ (Ω, B0 ) ⊆ A.
It it a question if A1 has to be suitably enlarged (still separable and “suitable”).
This bigger separable C *-algebra Be should have the property that Be plays
the same role as A.
Say, A itself is separable. M := A/(L∗ + L), D := L∗ ∩ L, N (A, D)/D ∼
= M,
{p}0 ∩ A? =? N (A, D) ??
Then L1 := A1 ∩ (A∗∗ (1 − p)) is a closed left-ideal of A1 , D1 := L∗1 ∩ L1 is a
hereditary C *-subalgebra and
????
A1 /(L∗1 + L1 ) ∼
= pA1 p ∼
= M.
????
N (D1 , A1 ) ⊇ A1 ∩ {p}0 contains φ(CM ) and is ( in ???) the multiplicative
domain of a ∈ A1 7→ pap ∈ pA∗∗ p.
N (D1 , A1 )/D1 ∼
= pA∗∗ p ? p(a − φ(f0 ⊗ pap))p = 0 for all a ∈ A.
Need g ∈ A1 ∩ {p}0 with g ≥ 0, kgk = 1, pg = p, g(1 − g) strictly positive in
D1 .
Then g n (X − φ(f0 ⊗ pXp))g n → 0 for X ∈ A1 and n → ∞.
Needs that kernel of A1 → pA∗∗ p is equal to sum L∗1 + L1 ????
Needs to construct a “regular” separable enlargement of A1 by “reduction to
separable case”.
???? By Lemma A.21.1 there exists ????
Next is OLD version?????
The proof refines the proofs of Lemmas 2.1.7 or ?? and Remark 2.1.16
Let M := pA∗∗ p of finite linear dimension, C := {p}0 ∩ A and π : f ∈ C →
f p ∈ M the C *-morphism from C into M considered in Part (ii). Then π is
an epimorphism by (ii). We define ϕ : C0 ((0, 1], M ) → M by ϕ(f ) := f (1) for
f ∈ C0 ((0, 1], M ).
 1. INFINITE ELEMENTS, FACTORIZATION AND TRANSITIVITY 131

By Proposition A.8.4, C0 ((0, 1], M ) is projective. It says that there exists a

C *-morphism ψ0 : C0 ((0, 1], M ) → C with π ◦ ψ = ϕ, i.e., ψ(f0 ⊗ a)p = a for all
a ∈ M.
The convex set Sp (A) ( = Cp (A∗∗ ) ) of positive functionals ρ ∈ A∗ = (A∗∗ )∗
with ρ(p) = kρk is σ(A∗ , A)-closed, because this set is naturally isomorphic to the
compact quasi-state space of M = pA∗∗ p. Thus, the projection p ∈ A∗∗ is closed,
and there is a closed left-ideal L ⊆ A such that q := 1 − p is the open support
projection of L in A∗∗ , i.e., A∗∗ q is the weak*-closure of L in A∗∗ .
It follows, that kpapk = dist(a, qA∗∗ + A∗∗ q) = dist(a, L∗ + L) . In particular,
L∗ + L is the kernel of the c.p. map a ∈ A → pap ∈ M . We get a − ψ0 (f0 ⊗ pap) ∈
L∗ + L , because a − ψ0 (f0 ⊗ pap) ∈ A and p(a − ψ0 (f0 ⊗ pap))p = 0 for all a ∈ A.
More ????????
Since Ω is a compact subset of A, there is a separable C *-subalgebra E0 of A
with Ω ∪ ψ0 (C0 ((0, 1], M )) ⊆ E0 .
Since E0 is separable, there exist a separable C *-subalgebra E1 ⊆ A with E0 ⊆
E1 , a−ψ0 (f0 ⊗pap) ∈ (E1 ∩L)∗ +(E1 ∩L) and kpapk = dist(a, (E1 ∩L)∗ +(E1 ∩L))
for all a ∈ E0 .
Next (commented?) comes also from Lemma B.14.1.
to be filled in
Part(iii) of Lemma 2.1.15, ??
If γ : Mnk ∼
= M qk is a *-isomorphism from Mnk onto M qk , then there is a
natural *-isomorphism τ from Fk onto Fk,1 ⊗ Mnk such that Fk,1 = yEy = y(Fk )y
for y := ψ0 (f0 ⊗ γ(p11 )) and τ (ψ0 (f0 ⊗ γ(α))) = y ⊗ α for α ∈ Mnk .
Step ?:
(Recall here that q := 1 − p is the above defined open support projection of a
hereditary C*-subalgebra of A.)
Text/Statement:
For every separable C *-subalgebra C ⊆ A, there exists a positive contraction
g ∈ {p}0 ∩ A with gp = p and limn kg n cg n k = kpcpk for all c ∈ C .
It is important that p ∈ A∗∗ is “closed”, – in the sense that 1 − p ∈ A∗∗ is the
“open” support projection of a hereditary C *-subalgebra D ⊆ A of A.
Step ?:
More general case:
If J / B contains a strictly positive element e ∈ J+ , and if the C *-morphism
ϕ : C0 ((0, 1], M ) → B/J
has kernel ker(ϕ) = C0 ((0, 1), M ), – i.e., there is a *-monomorphism λ : M →
B/J with ϕ(f ) = λ(f (1)) for f ∈ C0 ((0, 1], M ) –, then the homomorphic lift
ψ : C0 ((0, 1], M ) → B of ϕ can be chosen such that lim kg n ek = 0 for g := ψ(f0 ⊗
1M ), where f0 ∈ C0 (0, 1] is given by f0 (t) := t. 
132 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Remark 2.1.16. The Parts (i,ii) of Lemma 2.1.15 show that there exist a non-
zero C *-morphism ψ : C0 ((0, 1], Mn ) → A if a C *-algebra A has an irreducible
1/2
representation ρ : A → L(H) of dimension ≥ n . If we let fk := ψ(f0 ⊗ pk,1 ),
e := ψ(f0 ⊗ p11 ) with f0 (t) = t for t ∈ [0, 1], we get the following Part (i) and its
consequences in Parts (ii), (iii) and (iv) :

(i) If A has an irreducible representation of dimension ≥ n then there exist

elements e ∈ A+ and f1 , . . . , fn ∈ A that satisfy the relations kek = 1 and
fj∗ fk = δj,k e for j, k = 1, . . . , n .
(ii) In particular, if a ∈ A and a∗ Aa is not a commutative algebra, then there
exists b ∈ a∗ Aa with b2 = 0 and kbk = 1. (Local Glimm halving Lemma,
cf. [616, lem. 6.7.1]).
(iii) If each non-zero hereditary C *-subalgebra D of A has no character, then
for every n ∈ N and non-zero hereditary C *-subalgebra D of A we find
e, f1 , . . . , fn ∈ D with the relations listed in Part (i).
(In particular, if A is antiliminary, then each non-zero hereditary C *-
subalgebra D of A is also antiliminary, and the D contain a copy of Mn ⊗
C0 (0, 1] for each n ∈ N.)
(iv) If, moreover, A is strictly antiliminary (also called residually antiliminary,
cf. Definition 2.7.2) in the sense that each non-zero quotient A/J of A
is antiliminary in the sense of [616, sec. 6.1.1], then for every non-zero
hereditary C *-subalgebra D of A, every pure state ρ on D and every
n ∈ N there exists a C *-morphism ψ : C0 ((0, 1], Mn ) → D with

ρ(ψ(f0 ⊗ p11 )) = 1 .

(And this is equivalent to the property that no hereditary C *-subalgebra

of A has a non-zero character.)

Indeed, the Parts (i,ii) of Lemma 2.1.15 apply to all non-zero hereditary C *-
subalgebras D ⊆ A, because the property that each D has no character implies that
all non-zero hereditary C *-subalgebras D 6= {0} of A can not have any irreducible
representation of finite dimension (even can not contain non-zero compact operators
in its images).

Lemma 2.1.17. Let ρ a pure state on a C*-algebra A. We denote the corre-

sponding irreducible representation with cyclic vector ξρ by Dρ : A → L(L2 (A, ρ)).
Then there exists for each a ∈ A+ and ε > 0 a contraction g := g(a, ε) ∈ A
such that
kDρ (a)k < ε + ρ(g ∗ ag) . (1.7)

In particular, for each (fixed) pure state ρ on a simple C*-algebra A,

kak2 = sup { ρ(g ∗ (a∗ a)g) ; g ∈ A, kgk ≤ 1} .

Here L2 (A, ρ) := A/Lρ with Lρ := {a ∈ A ; ρ(a∗ a) := 0 }. It satisfies for all a ∈ A

that ρ(a) = hDρ (a)ξρ , ξρ i .
 1. INFINITE ELEMENTS, FACTORIZATION AND TRANSITIVITY 133

More precisely there exists h∗ = h ∈ A with khk ≤ π such that exp(ih) ∈

A + C · 1 ∈ A∗∗ satisfies Inequality (1.7) with exp(ih) in place of g. One can find
a contraction f ∈ A+ such that Dρ (f )ξρ = ξρ . Then g := exp(ih)f ∈ A satisfies
Inequality (1.7).

Proof. The connected component U0 (A)

b of 1 in the unitary group U(A)
b of
the unitization
Ab := A + C · 1 ⊆ M(A) ⊆ A∗∗

operates transitive on the unit sphere of the Hilbert space H := L2 (A, ρ) =

Dρ (A)xρ , because the (reformulated) Kadison transitivity in Lemma 2.1.15(ii,iii)
implies that for every y ∈ H the natural c.p. contraction

V : a + ξ1 7→ p(Dρ (a) + ξ1)p

maps the multiplicative domain of V onto pL(H)p, where p ∈ L(H) is the orthogo-
nal projection onto the complex linear span of {y, ξρ }, i.e., pH is 1-dimensional or
is 2-dimensional. If it is one-dimensional, then y = α · ξρ with α = exp(iγ) for some
γ ∈ [0, 2π) .
If kyk = 1 and U ∈ pL(H)p ∼ = M2 (C) is (partial) unitary with U ∗ U = p = U U ∗
and U (ξρ ) = y, then U is an exponential in pL(H)p, because all unitaries in M2 (C)
are exponentials, i.e., U = p exp(ik)p = exp(ik)p for some k ∗ = k ∈ pL(H)p with
kkk ≤ π.
By Part (ii) of Lemma 2.1.15 there exists h∗ = h ∈ {p}0 ∩ A with pDρ (h) = k.
It follows for T := Dρ (h) in L(H) that

exp(iT )p = p exp(iT ) = exp(ik)p = exp(ik) − (1 − p) .

In particular, exp(iT )(ξρ ) = exp(ik)(ξρ ) = y.

If we extend Dρ naturally to a *-representation of A + C · 1 ⊆ M(A) then
this says that Dρ (exp(ih))ξρ = y . If we use that b exp(ih) ∈ A, then we get that
Dρ (b exp(ih))(ξρ ) = Dρ (b)(y), and – therefore – that

ρ(exp(ih)∗ b∗ b exp(ih)) = kDρ (b)yk2 .

We apply this precise general observation to get the desired more relaxed esti-
mate:
Given 0 6= a ∈ A+ and ε ∈ (0, 1/2) we let δ := ε/µ with µ := 2(1 + kak1/2 ) .
Then δ 2 + 2δkak1/2 < ε .
There exists y ∈ H with kyk = 1 and kDρ (a1/2 )k ≤ δ + kDρ (a1/2 )yk . It
implies
kDρ (a)k < ε + hDρ (a)y, yi .
As we have seen above, there exists h∗ = h ∈ A with ρ(exp(ih)∗ a exp(ih)) =
kDρ (a1/2 )yk2 . Thus,

kDρ (a)k < ε + ρ(exp(ih)∗ a exp(ih)) .

134 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

We get the “relaxed” Inequality (1.7) for g := fδ (a) exp(ih), where fδ (t) :=
min(1, δ −1 max(t − δ, 0)) with sufficiently small δ ∈ (0, 1/4). 

Remark 2.1.18. There are several other methods to prove the surjectivity of
π : ({p}0 ∩ A) 3 a 7→ pa ∈ pA∗∗ p for “closed” projections p ∈ A∗∗ , e.g. by modifying
the original Kadison transitivity theorem [400, thm. 5.4.3, thm. 5.4.5], or – more
comfortable – its generalization by S. Sakai [704, thm. 1.21.16], but one can directly
use the far more general results mentioned in below given Remark 2.1.19.
Give here also reference to Appendix A!!!
Here is one of the very elementary ways: A closer inspection of there original
proofs shows that both cited original results (of Kadison or Sakai) did prove only
that the map
({p}0 ∩ (A + C · 1)) 3 a 7→ pa ∈ pA∗∗ p
is surjective for A + C · 1 ⊆ A∗∗ .
Check this critics on the citations to Kadison or Sakai again!
Simply to avoid a mistake!
But a simple trick shows that one can replace here A + C · 1 by A itself ( 10 ):
Let Dk : A∗∗ → L(Hk ) normal and surjective with supports Qk ∈ Z(A∗∗ ) and
let pk := Dk (p) = Dk (Qk p), – as considered in Lemma 2.1.15(i).
Further let D0 denote the restriction of D1 ⊕ · · · ⊕ Dn to pA∗∗ p . Consider a
unitary w ∈ pA∗∗ p of pA∗∗ p and define uk := Dk (w) ∈ pk L(Hk )pk .
The advanced forms of the Kadison transitivity theorem [704, thm. 1.21.16],
(that follows also directly from [400, thm. 5.4.5] or [616, thm. 2.7.5]) applied to the
unitization Ab = A + C · 1 ⊆ A∗∗ of A and to the pairwise inequivalent irreducible
representations Dk : A b → L(Hk ) show the following: There is a unitary U ∈ A b
with pk Dk (U )pk = uk for k = 1, . . . , n .
Thus D0 (pU p − w) = 0 for the above defined D0 . Since D0 |pA∗∗ p is faithful,
we get pU p = w. It implies pU ∗ pU p = p = pU pU ∗ p, pU (1 − p) = 0 = (1 − p)U p
and (pU − U p) = 0 . It shows that the C *-morphism ψ : a ∈ C1 7→ ap ∈ pA∗∗ p is a
*-epimorphism from C1 := {p}0 ∩ (A + C1) onto pA∗∗ p .
On the other hand, by 1 · p = p = p · 1 , C1 = C + C1 for C := {p}0 ∩ A and
C = C1 or C1 /C ∼= C. Obviously, in the second case ψ(C) is a closed ideal of pA∗∗ p
of co-dimension at most one!
If we use that pA∗∗ p has finite linear dimension, then the latter implies that
pA p has a character, given by a central projection q of pA∗∗ p.
∗∗

Now comes the “dirty trick”: We replace A by A ⊗ M2 and p by

p ⊗ 12 ∈ A∗∗ ⊗ M2 ∼
= (A ⊗ M2 )∗∗ .

Then (p ⊗ 12 )(A ⊗ M2 )∗∗ (p ⊗ 12 ) ∼

= pA∗∗ p ⊗ M2 has no character anymore.

10 This trick was not observed or precisely mentioned in any of the 3 above cited books.
 1. INFINITE ELEMENTS, FACTORIZATION AND TRANSITIVITY 135

Above considerations (applied to A ⊗ M2 instead on A) show that the image

of the natural C *-morphism
η : {p ⊗ 12 }0 ∩ (A ⊗ M2 ) → pM ∗∗ p ⊗ M2
is an ideal of pA∗∗ p ⊗ M2 of co-dimension ≤ 1. But pM ∗∗ p ⊗ M2 has no character.
Thus, η must be surjective.
The surjectivity of η implies the surjectivity of the map π defined by π(a) := ap
for a ∈ {p}0 ∩ A. Indeed: Let x ∈ pA∗∗ p. There exists b ∈ {p ⊗ 12 }0 ∩ A ⊗ M2 with
b · (p ⊗ 12 ) = x ⊗ e11 . It follows that there is a ∈ A with (1 ⊗ e11 )b(1 ⊗ e11 ) = a ⊗ e11
and ap = x, where e11 ∈ M2 is the upper-left matrix unit. Moreover,
[a, p] ⊗ e11 = (1 ⊗ e11 )[b, p ⊗ 12 ](1 ⊗ e11 ) = 0 .
This proves that the C *-morphism π : {p}0 ∩ A → pA∗∗ p is surjective. It will be
used in the proof of Lemma 2.1.15(ii), by appying first one of the cited (older)
textbook versions of the Kadison transitivity theorem.
HERE ends Remark 2.1.18

Remark 2.1.19. One can shorten the proof of all Parts of Lemma 2.1.15 consid-
erably by using more general results that we anyway have to use for our embedding
results in Chapter 5.
First check that each projection p ∈ A∗∗ of finite rank (in the sense that pA∗∗ p
has finite dimension) is a “closed” projection in A∗∗ , i.e., (1 − p) is the supremum
of an upward directed net of positive contractions in A, or equivalently:
The closed left ideal L := A∩(A∗∗ (1−p)) of A has an open right support projection
that is equal to 1 − p. This means that the weakly closed left ideal A∗∗ (1 − p) of
A∗∗ is the bi-polar of a closed ideal L := A ∩ (A∗∗ (1 − p)) of A (that is, in our here
considered special case, the intersection of finitely many left-kernels Lρ := {a ∈
A ; ρ(a∗ a) = 0} of pure states ρ on A).
This idea has been used in the proof of [704, thm. 1.21.16] ( 11 ), but in the there
considered special case it was not completely proved that the natural C *-morphism
N (A, L∗ ∩ L) → (A/(L∗ + L)) ∩ M(A/(L∗ + L)) ⊆ pA∗∗ p
by using the natural isomorphism pA∗∗ p ∼
= A∗∗ /((1 − p)A∗∗ + A∗∗ (1 − p)) , is
a surjective C*-morphism with kernel Ann(A, L∗ ∩ L) – that is a closed ideal of
N (A, L∗ ∩ L) .
The reason for the surjectivity of the map from normalizers onto the multiplier
algebra of the quotient C *-space is that the closed unit ball of A maps onto the
closed unit ball of A/(R+L) if R is a closed right-ideal of A and L is closed left-ideal
of A.
It should be shortened by reference to the appendices
It comes by using the bipolar theorem for
?????
11 Rewritten here in our notation / terminology – and not as in [704]
136 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

those convex subsets of Banach spaces that are closures of its open (again
convex) interiors and there bi-duals from the more universal (and almost trivial)
perturbation property of maps a ∈ M 7→ paq ∈ pM q for projections p, q in W*-
algebras M :
If a, b ∈ M satisfy kak, kbk < 1 and kpaq − pbqk < ε then there is c ∈ M with
√
kck < 1, pbq = pcq, and ka − ck ≤ ε + 2ε .
(A study of kak u1 and kbk u2 in place of a and b defined by suitably “iterated”
construction of “Halmos” unitaries ( 12 ) u1 , u2 ∈ M3 (M ) leads to an idea of proof
of this non-obvious fairly sharp inequality.)
In the very special case considered in Lemma 2.1.15(iii) we have the particular
situation that M(A/(L∗ + L)) = pA∗∗ p + (1 − p)A∗∗ (1 − p) where L := A ∩
(A∗∗ (1 − p)) is nothing else the intersection of the “left”-kernels of finitely many
pure states on A. The C *-subalgebra B := {p}0 ∩ A of A maps always onto
M(A/(L∗ + L)) ∩ pAp, which is in this special situation equal to pA∗∗ p because p
is a closed projection.
We can then simply consider the multiplier algebra M(A//E) = ∼ pA∗∗ p of the
unital operator system A/(L∗ + L) =: A//E for E := L∗ ∩ L to get the desired
“precise” variant of an “advanced” Kadison transitivity theorem by the projectivity
of cones over C *-algebras of finite dimension.
HERE ends Remark 2.1.19
More precise and short references to the Appendix Section would be useful.

In the three following lemmata Cn denotes the complex Hilbert space `2 (n) of
dimension n ∈ N. Next one uses an extension that reduce all to the until case:

Lemma 2.1.20. Let C a unital C*-algebra, B ⊆ C a C*-subalgebra and

V : B → Mn = L(Cn ) a c.p. contraction.
There exists a unital c.p. map U : C → Mn and a positive contraction d ∈
(Mn )+ with V (b) = dU (b)d for all b ∈ B.
Moreover, d = V ∗∗ (1)1/2 for 1 ∈ B ∗∗ .

Proof. The “Stinespring” dilation of the c.p. map V is given by some Hilbert
space H0 , a *-representation ρ : B → L(H0 ) and a linear map g : Cn → H0 such
that V (b) = g ∗ ρ(b)g .
We can suppose here that ρ is non-degenerate, i.e., H0 = ρ(B)H0 , because
otherwise we can replace g by q · g and ρ by ρ(b)|qH0 , where q is the orthogonal
projection from H0 onto the closure of the linear span of ρ(B)H0 . Then the nor-
malization ρ : B ∗∗ → L(H0 ) satisfies ρ(1) = idH0 and the second conjugate V ∗∗ of
V : B → Mn is given by V ∗∗ = g ∗ ρ( · )g. It implies V ∗∗ (1B ∗∗ ) = g ∗ g.
There exist a unital *-representation D : C → L(H) and an isometry I : H0 →
H such that ρ(b) = I ∗ D(b)I for all b ∈ B. This can be shown e.g. by using extension

12 See Remark 4.2.4 for the definition of Halmos unitaries.

 1. INFINITE ELEMENTS, FACTORIZATION AND TRANSITIVITY 137

of positive functionals, cf. [616, prop. 3.1.6]. The map W (c) := I ∗ D(c)I is a unital
c.p. map W : C → L(H0 ) with W |B = ρ .
Let p ∈ Mn denote the support projection of g ∗ g ∈ Mn . There is partial
isometry z ∈ L(`2 (n), H0 ) with z(g ∗ g)1/2 = g and z ∗ z = p. Let d := (g ∗ g)1/2 . We
can define a c.p. contraction U0 : C → Mn by U0 (c) := z ∗ W (c)z with U0 (1) = p .
Then V (b) = g ∗ ρ(b)g = dz ∗ W (b)zd = dU0 (b)d . Let µ : C → C a state on C and
define U1 (c) := µ(c) · (1 − p). The sum U := U0 + U1 is a c.p. map from C into
Mn with U (1C ) = 1n and dU (b)d = V (b) .
If 1C ∈ B then V ∗∗ (1C ) = dU ∗∗ (1C )d = dU (1C )d = d2 .
If 1C 6∈ B then the C *-subalgebra B + C · 1C of C has a non-zero character χ
with χ(B) = 0. It extends to a state µ on C with µ(B) = {0}. If we define U1 with
this state µ, then we get U (b) = U0 (b) for b ∈ B and dU (b)d = V (b) = g ∗ ρ(b)g .
Now observe that V ∗∗ (x) = g ∗ ρ(x)g for x ∈ B ∗∗ . It follows V ∗∗ (1B ∗∗ ) = g ∗ g =
d , because V ∗∗ (x) = dU0∗∗ (x)d for x ∈ B ∗∗ .
2


The following “existence” Lemma 2.1.21 is needed, because we consider e.g. in

Proposition 2.2.1 non-separable purely infinite simple C *-algebras, and then non-
separable ultra-powers and corona algebras. It requires to check some cases for
non-separable behavior.
Recall that the projections P = P ∗ = P 2 with separable P H generate (– in a
pure algebraic manner! –) a non-unital simple closed ideal J of L(H)/K(H) if the
Hilbert space H is not separable. In case n = 1 the assumptions of next Lemma
−1
allow that λ is a state on L(H) with λ|πK (J) = 0, then λ is zero on P L(H)P . It
is one of the reasons for our “local redefining” of λ on separable C *-subalgebras B
of L(H) like in next Lemma ...

Lemma 2.1.21. Let H a non-separable Hilbert space, B ⊆ L(H) a separable

C*-subalgebra, Q ∈ L(H) an orthogonal projection onto a separable subspace of H
and λ : L(H) → Mn a c.p. map with λ(K(H)) = 0 .
Then there exists an orthogonal projection P ∈ L(H), a (new) unital c.p. map
ρ : P L(H)P → Mn and an element g ∈ (Mn )+ , such that P H is separable, P Q = Q,
P b = bP for all b ∈ B, and with λ(b) = gρ(P b)g, for all b ∈ B, and ρ(P KP ) = {0}.

Proof. Still to be checked again:

Consider the separable C *-subalgebra C of L(H) generated by {1, Q, R} ∪ B,
where R denotes the orthogonal projection onto (B ∩ K) · H and let 1 := idH . (13)
The subspace RH is separable because B ∩ K is separable and, therefore, con-
tains an in B ∩ K strictly positive compact operator b0 . It has has above defined
R as its support projection in L(H). The C *-algebra C is unital and 1 ∈ C .
Take a sequence c1 , c2 , . . . in C that is dense in the unit-ball of C, and find finite-
dimensional subspaces Xn ⊆ H with the property that kck |Xn k + 2−n ≥ kck k for
13 Here the notation is different from that in the proof of Lemma 2.1.20.
138 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

k ≤ n. Let X∞ denote the separable linear span of the Xn and of QH. Now take
the smallest C-invariant closed subspace H0 of H containing X∞ . It is separable by
separability of C and X∞ . Then QH ⊆ H0 and the restriction c|H0 of the elements
c ∈ C is a faithful unital C *-algebra morphism D from C into L(H0 ) given by
D(c) := c|H0 = (c · P )|H0 , where we denote by P the orthogonal projection from
H onto H0 .
Then K(H0 ) = P K(H)P , and

kπK(H0 ) (D(c))k = kπK(H) (c)k ∀ c∈C,

because D(c) ∈ K(H0 ), if and only if, c ∈ K(H).

Since C is a separable C *-algebra, the closed ideal C ∩K of C contains a strictly
positive contraction e ∈ C+ that is a strictly positive element of C ∩ K, and the
support projection of e is equal to P and commutes with C.
It holds
D(C) ∩ K(H0 ) = D(C ∩ K) ,
and, for c ∈ C, D(c) = (c ◦ P ) ∈ P KP ∼ = K(H0 ) if and only if c ∈ K . And this
implies that kπK (c)k = kπK(H0 ) (D(c))k for all c ∈ C .
Thus, C/(C ∩ K) ∼
= (C + K)/K ⊆ L(H)/K(H) and D(C)/(D(C) ∩ K(H0 )) ∼ =
(D(C) + K(H0 ))/K(H0 ) are isomorphic C *-algebras, and up to natural isomor-
phisms, C/(C ∩ K) ⊆ L(H)/K(H) and D(C)/(D(C) ∩ K(H0 )) contained in
L(H0 )/K(H0 ).
Because λ : L(H) → Mn is a c.p. map with λ(K(H)) = 0 , the restriction to
C +K(H) defines a c.p. map from C +K to Mn with kernel K. Let [λ] : (C +K)/K →
Mn the c.p. map with λ = [λ] ◦ πK The above discussed isomorphism [D] from
(D(C) + K(H0 ))/K(H0 ) onto (C + K)/K allows to define a c.p. map ??????
C/(C ∩ K) ∼
= D(C)/(D(C) ∩ K) to Mn and then from ?????
....The ρ : L(H0 ) → Mn comes ... later ... as unital c.p. map ρ(T ) :=????? by
use of Lemma ?? extension ... of the
Then P commutes with the elements of b ∈ C and dist(K, b) = dist(K, P · b) for
K = K(H). This distance is also identical with dist(P KP, b) and with kπK (b)k =
kπC∩K (b)k for b ∈ C. Moreover C ∩ K = C ∩ P KP , C/C ∩ K is natural isomorphic
to (C + K)/K and is isomorphic to π(D(C) + K(H0 ))/K(H0 ).
The point for the latter is: 1) b → bP is a C *-morphism, that is faithful on B?
and K ∩ (B · P ) = (B ∩ K) · P ?
??? We can now the restriction λ to b ∈ B of λ given by ρ(P b) =: λ(b), extend
to L(P H) with ρ(K ∩ L(P H)) = {0} because K ∩ L(P H) = P KP ∼ = K(P H) and
λ(B) ∩ (P · K · P ) = λ(B ∩ K)
λ|(B ∩ P · K · P )
????
Or extend ρ(P b) to P L(H)P ∼
= L(P H) with ρ(P K(H)P ) = {0} ???
 1. INFINITE ELEMENTS, FACTORIZATION AND TRANSITIVITY 139

Here is used that ρ(K) = 0 ... ???

To be filled in !! ?? 

The below given Lemma 2.1.22 and its proof shows that the generalizations
of the classical Weyl–von-Neumann Theorem along the lines of D. Voiculescu is in
fact a very special case of Proposition 5.4.1 that generalizes the Weyl–von-Neumann
Theorem in an m.o.c. cone equivariant manner, – in particular in an ideal-system
preserving manner if those m.o.c. cones are suitably chosen.
The space Cn means the n-dimensional complex Hilbert space `2 (n) in next
Lemma 2.1.22. Similar results can be shown also in the real case with almost
obvious modifications here and in Lemma 2.1.20.

Lemma 2.1.22 (J. Glimm, D. Voiculescu). Suppose that A ⊆ L(H) is a C*-

algebra and that S : A → Mn = L(Cn ) is a completely positive contraction that
annihilates A ∩ K(H) , i.e., satisfies S(A ∩ K(H)) = {0}.
Then for every compact subset Ω1 ⊆ A, every compact subset Ω2 ⊆ H and
every ε > 0, there is a linear isometry ?? perhaps only: contraction ?? v : Cn → H
such that, for all a ∈ Ω1 , y ∈ Ω2 and x ∈ Cn holds:

kS(a) − v ∗ avk < ε and |hv(x), yi| < εkxk . (1.8)

If, moreover, idH ∈ A+K(H) and S(a) = 1n for all a ∈ A with a ∈ K(H)+idH ,
or if idH is not contained in A + K(H), then there exists an isometry v : Cn ∼
=
`2 (n) → H that satisfies the inequalities (1.8).

What are the differences between this isometries? check the proof again!
Notice that idH 6∈ A + K(H) if A + K(H) is not unital. In particular A + K(H)
is not unital if A is stable or if A ⊆ K(H) .
Here Cn is the n-dimensional complex Hilbert space `2 (n), and 1n ∈ Mn ∼ =
L(`2 (n)) is then the identity map of Cn . It is crucial for applications that the
map v can be chosen always as a contraction, respectively as an isometry in the
mentioned two special cases.

Proof of Lemma 2.1.22. We use Lemma 2.1.20 for the reductions to the
unital case and (later) reduction to the unital separable case for a unital c.p. map
T : (A + K + C · idH )/K → Mn . Here H is not necessarily separable, but restriction
to separable C *-subalgebras of A and then to separable subspaces of H allows a
reduction to separable closed subspaces of H. This reductions to the separable case
allows to use (and cite) then several places where this cases are considered (with
slightly different proofs).
The original S is then obtained by “compression” as in the Lemma 2.1.20, that
leads to the case where v : Cn → H is then only a contraction.
140 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

In the following, K means K(H) and 1 := idH . If 1 is not contained in A + K

then A + K is a closed ideal of the C *-algebra A + K + C · 1 ⊆ L(H), and A ∩ K is a
closed ideal of A and is also a closed ideal of A + K with (A + K)/K ∼= A/(A ∩ K).
Thus, S : A → Mn with S(A ∩ K) = {0} can be considered as a completely
positive contraction T : (A + K) → Mn with T (K) = 0 .
If 1 ∈ A + K, i.e., there exists a ∈ A with a ∈ K + 1 and S(a) = 1n for all
a ∈ A with a ∈ K + 1 , then T is unital.
If 1 is not contained in A + K, then A + K is an essential ideal of A + K + C · 1,
and the minimal extension Te of T : A+K → Mn to a positive map from A+K+C·1
to Mn is given by the restriction to A + K + C · 1 ⊆ (A + K)∗∗ of the bi-adjoint
T ∗∗ : (A + K)∗∗ → Mn of T .
??? Compare Lemma 2.1.20 !!!
1 ∈ A + K(H) and S(a) = 1n for all a ∈ A with a ∈ K(H) + C · 1
???
that annihilates A ∩ K(H) , i.e., satisfies S(A ∩ K(H)) = {0} .
The Stinespring dilation of S : A → Mn ∼
= L(`2 (n)) with ?????
We use here the notation Cn := `2 (n) , i.e., Cn is equipped with the norm
k(α1 , . . . , αn )k2 := ( k |αk |2 )1/2 . And “c.p.” means “completely positive”.
P

The estimates in the inequalities (1.8) have to be shown only for the compact
subsets Ω1 ⊆ A and Ω2 ⊆ H, and for given ε > 0. It implies that it suffices
to consider in place of A and H the separable C *-subalgebra C ∗ (Ω1 ) ⊆ A and
the closed separable subspace H0 ⊆ H that contains Ω2 and is invariant under
C ∗ (Ω1 ) . They satisfy C ∗ (Ω1 ) ∩ K(H0 ) ⊆ A ∩ K(H), i.e., we can suppose that
A and H are separable from now on. But there are critical points here: Possibly
we have to enlarge the separable closed subspaces H0 ⊆ H and the separable C *-
algebra A0 ⊆ A with A0 H0 ⊆ H0 in the cases where idH ∈ A + K(H) to get that
idH0 ∈ A0 + K(H0 ). We find k ∈ K(H) and a1 ∈ A with k + a1 = idH ...
Then we can ...
In case that ...
and to check that we can select the new separable C *-algebra A0 (in place of
A) and , such that the new Hilbert space H0 and the C *-algebra A0 is chosen big
enough ...
that in the cases where idH ∈ A + K(H) and S(a) = 1 for all a ∈ A with
a ∈ K(H) + idH , or where idH is not contained in A + K(H),
the new A0 |H0 ⊆ L(H0 ) given now by
again satisfies idH0 ∈ A0 + K(H0 ) and S(a) = 1 for all a ∈ A0 with a ∈
K(H0 ) + idH0 , respectively again satisfies
????????????
 1. INFINITE ELEMENTS, FACTORIZATION AND TRANSITIVITY 141

and
Details ?????.
Reduction of the general case to the case where A and H are separable :
The compact sets Ω1 ⊂ A and Ω2 ⊂ H generate a separable C *-subalgebra B of
A and a B-invariant separable closed subspace H1 of H such that Ω2 ⊂ H1 and
ψ : b ∈ B 7→ b|H1 is a faithful C *-morphism and ψ(B ∩ K(H)) = ψ(B) ∩ K(H1 ) .
If idH ∈ A + K(H) and S(a) = 1 for all a ∈ A with a ∈ K(H) + idH ,
or if idH is not contained in A + K(H), ...
Notice that Ω1 and Ω2 contain for each ε finite subsets X1 and X2 such that
the ε/3 balls around them cover all of Ω1 , respectively Ω2 . Then all predicted
conclusions
(still to be proved !)
are reduced to the study of the case where the algebra A and the Hilbert space
H are separable and the subsets Ω1 ⊂ A and Ω2 ⊂ H are finite sets.
Further reductions of the general case with separable A and H :

We start with a reduction of all cases to the case where idH ∈ A + K(H) = A,
S(idH ) = 1n and S(K(H)) = {0} .
Next step not done yet !!!
but cited 3 places where it was done for separable C *-subalgebras of
the Calkin algebra.
Below ????
Then we can give references to different nice proofs of this very special case,
i.e., where A is separable, idH = 1A and S(idH ) = 1 , i.e., where n = 1 and S is the
restriction to A of a state on the Calkin algebra L(H)/K(H).
It says essentially that the restriction of a state S on the Calkin algebra to a
separable C *-subalgebra of A ⊆ L(H) is a point-wise limit on A of a net of vector
states that converges on K(H) point-wise to zero.
The reduction to the special separable unital case uses Lemma 2.1.20.
Refer to above Lemma ???
The following observation is easy to verify:
If D and E are (arbitrary) C *-algebras, T : D → E a c.p. map and J ⊆ D closed
ideal with T (J) = {0}, the the class-map [T ]J : D/J → E is a c.p. map.
We apply this observation and Lemma 2.1.20 to C := L(H)/K, B := (A +
K)/K ∼
= A/(A ∩ K) and the quotient map V := [S] : B → Mn .
This reduces all cases to the special case where 1 := idH ∈ A+K and S(a) = 1n
for all a ∈ 1+K, and then to H ∼ = `2 (N) and separable A. Now we can cite textbooks
and papers for the ???? remaining two steps: ???? Which steps???
142 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

The proof for this special case reduces to the well-known more special case
where A and H are separable and n = 1:
Each of [798](1976), [43](1977) or [207, lem. II.5.2](1996) give a reduction of
the unital separable case with n > 1 to the case n = 1 .
Then all is reduced to the case of unital separable A and n = 1, i.e., where
Mn = C and S is a state on A that annihilates A ∩ K(H). This is the basic case,
that has been shown in each of the references Glimm [323](1960), Dixmier [217,
lem. 11.2.1](1969), [123] (1976), or [207, lem. II.5.1](1996) with slightly different
proofs requiring different basic knowledge. 

Next: Still to be checked:

Remark 2.1.23. The Part(iii) of Lemma 2.1.15 and Lemma 2.1.22 and 2.1.15
imply together in the special case of a non-elementary simple C *-algebra A the
following observation:
Let V : A → Mn a completely positive contraction. Then, for every compact
subset Ω ⊆ A, pure state ρ on A and ε > 0, there exist contractions e, b1 , . . . , bn ∈ A
with the properties b∗j bk = δj,k e , ρ(e) = 1 and

kb∗j abk − V (a)j,k ek < ε for j, k ∈ {1, . . . , n} and all a ∈ Ω .

!!! Wanted result:

There exists contractions b1 , . . . , bn ∈ A and e ∈ A+ with kek = ρ(e) = 1,
b∗j bk = δjk e such that k∆(a)k < ε for a ∈ Ω where

∆(a) := [b1 , . . . , bn ]∗ a[b1 , . . . , bn ] − V (a) ⊗ e ∈ Mn (A) .

It is equivalent to:

kV (a)j,k e − b∗j abk k < ε/(n + 1) for all j, k ∈ {1, . . . , n}, a ∈ Ω .

(Here one can take partial isometries bk and a projection e ∈ A if A is simple and
has real rank zero.)
Indeed:

Let dρ : A → L(H) the irreducible representation corresponding to some pure

state ρ on A.
The simplicity of A allows us to identify the elements a ∈ A with dρ (a) ∈ L(H)
for H := A/Lρ . Then dρ (A) ∩ K(H) = {0} , because A is non-elementary by
assumption.
We ?? get ?? by Lemma ?? ???
Let V be a completely positive contraction gW (·)g from L(H) to Mn , where
W is unital and 0 ≤ g ≤ 1 in g ∈ (Mn )+ , kgk ≤ 1.
The Lemma 2.1.22 implies the existence of an isometry I : Cn → H with the
property that kI ∗ dρ (a)I − V (a)k < ε/3 for all a ∈ Ω.
 2. CHARACTERIZATIONS OF SIMPLE PURELY INFINITE ALGEBRAS 143

Let xj := Iej ∈ A/Lρ for the canonical ONB of Cn = `2 (n) with norm
k(α1 , . . . , αn )k2 = 2
P
j |αj | . The Lemma 2.1.15 gives for the irreducible represen-
tation dρ the existence of contractions b1 , . . . , bn ∈ A with bj + Lρ = xj , kb1 k = 1
and b∗j bk = δj,k b∗1 b1 , j, k ∈ {1, . . . , n} , such that ρ(b∗1 b1 ) = 1.
The quotient map A → A/Lρ has the property that the closed unit-ball of A
maps onto the closed unit-ball of the Hilbert space A/Lρ , cf. Remark ??.
The point is here:
We find a C *-morphism ψ : C0 ((0, 1], Mn ) → A such that dρ (ψ(f0 ⊗ 1n )) = Pn for
the orthogonal projection onto the linear span of x1 , . . . , xn , and

ρ(f0 ⊗ pj,j ) = Iej = xj ???

1/2
Take bk := f0 ⊗ pk,1 . Could replace f0 by f0n with the property that

ρ(ψ(f0m ⊗ 1n )a(f0m ⊗ 1n )) − ρ(a) ⊗ 1n

becomes small ?? ??????

2. Characterizations of simple purely infinite algebras

Among the simple C *-algebras, the purely infinite C *-algebras can be char-
acterized in terms of fairly different criteria but those turn out to be equivalent
in this special case. The most useful and applicable criteria are the more compli-
cate looking once. We provide a selection of those criteria in the Proposition 2.2.1
without claiming completeness or originality. The list of this equivalent properties
is the work of many hands. We tried to find the most simple explanations for
this equivalences. But notice that we give later several other different criteria that
also characterize pure infiniteness of simple C *-algebras A, see e.g. Proposition
2.2.5(iv), Corollary 2.4.6, Corollary 2.2.11, Corollary 3.2.16(ii) in conjunction with
Remark 3.2.17, and, last but not least, the most important for our applications:
a simple non-elementary C *-algebra A has the WvN-property – given in Definition
1.2.3 –, if and only if, A is purely infinite, cf. Chapters 3 and 5 for basic applications
of the WvN-property (= Weyl – von Neumann Property).
There exist fairly different properties that also characterize separable non-
elementary simple purely infinite nuclear C *-algebras A by simplicity of F (A) :=
(A0 ∩ Aω )/ Ann(A, Aω ) in [448, thm. 2.12], or by the “unrestricted” validity of an
analog of the Weyl–von-Neumann–Voiculescu theorem, cf. Corollary 5.7.3.
Most of the criteria in the parts of Proposition 2.2.1 itself do not imply simplic-
ity, only the properties in Parts (ii), (iv), (vi), (vii), (xv) and (xvi) of Proposition
2.2.1 imply also that A is simple. Therefore we prefer to add here the requirement
that A is simple and non-zero.
What about Part (ix)? (Same number now???)
The properties in Parts (iv), (vii) and (viii) are only equivalent if A is simple,
because weakly purely infinite C *-algebras are not necessarily simple.
144 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

But (iv) and (vii) both imply simplicity ... Are (iv) and (vii) “formally” equiv-
alent? Formally (iv) implies (vii), and (vii) implies (viii). ...
They are almost obviously the same in case of simple C *-algebras, but (vii)
(and (iv)?) imply the simplicity of A. ?????
Several of those properties are not equivalent to each other if the C *-algebra
A is not simple as e.g. those in Parts (i) and (ii).
Part(ii) of Proposition 2.2.1 expresses only the Definition 1.2.1 of not necessarily
simple purely infinite C *-algebras in the special case of simple C *-algebras – in a
way that this new definition of a “simple purely infinite” C *-algebras A contains
also the simplicity of A and is clearly not equivalent to the property in Part(x) that
works also for non-simple C *-algebras A .
This many equivalent properties allow to say: If A is simple and has one of
the properties listed in Proposition 2.2.1, then A has each of the listed properties,
e.g. if A is p.i. in sense of J. Cuntz, then it has also all the properties listed e.g. in
(vi), (ix), (x) and (xvi). For example, by Parts (ix) and (x), a non-zero simple
C *-algebra A is p.i. if and only if every σ-unital hereditary C *-subalgebra D of A
is stable or has a properly infinite unit 1D := p , i.e., D = pAp with in A properly
infinite projection p.
We say that a simple C *-algebra A is non-elementary if A is not isomorphic
to the algebra of compact operators K(H) on a Hilbert space H of finite or infinite
dimension, in particular A 6∼
= C.
Needed for Citation ?: Which parts show directly that
Property pi(1) [ pi(n) ⇐ (iv)] is equal to Property pi-1 ⇐ (ii);
pi-n ⇐ (xi), (xiv)?, ??
What about (x), (xii), (xiii)?

Proposition 2.2.1. Suppose that A is a (non-zero) simple C*-algebra. Then

the following possible properties (i)–(xvii) of A are equivalent to each other:

(i) A is purely infinite in the sense of J. Cuntz [172, p. 186], i.e., every
non-zero hereditary C*-subalgebra D of A contains an infinite projection
p ∈ D (14).
(ii) A is purely infinite in the sense of Definition 1.2.1, i.e., A 6= C, and for
each a, b ∈ A+ with kak = kbk = 1 and ε > 0, there exists c ∈ A with
kb − c∗ ack < ε (15).
(iii) A is locally purely infinite in the sense of Definition 2.0.3, i.e., every
non-zero hereditary C*-subalgebra E of A contains a non-zero stable C*-
subalgebra B ⊆ E.

14 The projection p is then properly infinite by simplicity of A, cf. Lemma 2.1.6. In par-

ticular, A has the ”small projections” property (SP) that says that every non-zero hereditary
C *-subalgebra D of A contains a non-zero projection.
15 There is c ∈ A with kck = 1 and kb−c∗ ack < ε , cf. Proof of (iv)⇒(ii) or Proof of (iii)⇒(ii).
 2. CHARACTERIZATIONS OF SIMPLE PURELY INFINITE ALGEBRAS 145

(iv) The algebra A is non-elementary and, for every b ∈ A+ with kbk = 1

and every ε > 0, there exists a number m := m(b, ε) ∈ N (allowed here
to depend on both of b and ε) such that, for every a ∈ A+ with kak =
1, there exist elements c1 , . . . , cm ∈ A (depending on a, b, ε) that satisfy
Pm
kb − j=1 c∗j acj k < ε . (16)
(v) A is strongly purely infinite in the sense of Definition 1.2.2, i.e., for every
a1 , a2 ∈ A+ and ε > 0 there exist d1 , d2 ∈ A with kd∗i ai aj dj − δij ai aj k < ε
for i, j = 1, 2.
(vi) A is algebraical simple in the sense that A 6= C and for every a, b ∈ A
with a 6= 0 there are x, y ∈ A with xay = b .
(vii) There exists (general, – not depending on a, b, ε –) n ∈ N such that A is not
isomorphic to Mk (C) for k = 1, . . . , n, and, for every non-zero a, b ∈ A+
Pn
and ε > 0, there exists d1 , . . . , dn ∈ A such that kb − j=1 d∗j adj k < ε.
(17)
(viii) A is simple and is weakly purely infinite in the sense of Definition 2.0.4.
(18)
(ix) A is non-elementary and every non-zero non-unital σ-unital hereditary
C*-subalgebra D of A is stable.
(x) A has real rank zero and every non-zero projection in A is properly infinite.
(xi) Every non-zero element a ∈ A+ is infinite (and A is non-zero and simple).
(19)
(xii) There exists n ∈ N such that A is not isomorphic to Mk (C) for each k < n
and each n-homogenous non-zero element a ∈ A+ is infinite.
(xiii) There exists n ∈ N such that a ⊗ 1n is infinite in A ⊗ K for each non-zero
a ∈ A+ .
(xiv) A has property pi-n of Definition ?? for some n ∈ N.
(xv) A is non-elementary and, for every a, b ∈ A+ with kak = kbk = 1 and
every ε > 0, there exists ν := ν(b, a, ε) ∈ N (depending on each of a, b, ε)
such that, for every k = 1, 2, . . ., there exist contractions c1 , c2 , . . . , cν ∈ A

16check again, and give reference to W (A): Essentially, the assumptions of Part (iv)

say that [(b − ε)+ ] ≤ m(b, ε)[a] in the Cuntz semigroup W (A), and are fulfilled with a fixed
constant number m(b, ε) := n if A is simple and has Property pi(n) of Definition 2.0.4 for some
n ∈ N.
17This property follows from Property pi(n) in Definition 2.0.4 in the special case that A is

simple. It is formally NOT a special case of Part (iv).

18This Part is almost the same as in Part (vii). It is not a special case of Part (iv), because

it implies no simplicity.
19Notice here our overall pre-assumption that A is simple and non-zero. It does not directly

imply that A has Property pi(1) of Definition 2.0.4 or Property pi-1 of Definition ??.
Property pi(1) says in case of simple A, that A 6= C, and for every elements a ∈ A+ , b ∈ A
and ε > 0, that there exists d1 , d2 ∈ A with kd1 ad2 − bk < ε.
pi-1 says: Every non-zero element is properly infinite. But since A has no non-zero non-trivial
quotients it is equivalent to require only that each 0 6= a ∈ A is infinite.
But how to formulate the latter understandable?
146 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

(depending on k, a, b, ε, but with same number ν(b, a, ε) ∈ N) such that

Pν
kb − j=1 c∗j ak cj k < ε . (20).
(xvi) For each n ∈ N holds: If V : A → Mn and W : Mn → A are completely
positive contractions, then for each finite subset F ⊂ A and ε > 0 there
exists contractions d1 , d2 ∈ A (depending on W ◦ V, F, ε) with d∗1 d2 = 0
and

k(W ◦ V )(a) − d∗k adk k < ε for all a ∈ F, k ∈ { 1, 2 } .

(xvii) For every non-zero e ∈ A+ there exists in C ∗ ( eAe , 1) ⊆ M(A) an element

that is left-invertible in A + C1 but is not right-invertible.

We shorten the proof of the equivalences of conditions in (ix, x) with the other
conditions in Parts (i)–(viii), (xi)–(xiv) by using a stability criterium for σ-unital
C *-algebras of Hjelmborg and Rørdam [373] – cf. also our Corollary 5.5.1 with
an alternative proof of this criterium –, and we use the characterizations of real
rank zero C *-algebras given by L.G. Brown and G.K. Pedersen in [113] as those
C *-algebras that have the property that every σ-unital hereditary C *-subalgebra
contains an approximate unit consisting of projections. See Remark 2.2.2 concern-
ing applications of the (almost ridiculous) Part (xvii) and is applications.

Proof. All in Part (o) has to be discussed and/or shifted to

other places.
But remove or change references from (o) to more relevant!!
Needed/used? is reduction to separable case (Perhaps in Appendix A
or B)
(o): We start with this preliminary observation (o) that is related to the condi-
tions in Parts (iv) and (vii) but uses formally weaker assumptions. It has (implicit)
applications in proofs of Parts (i)–(xvi) and in the proof of Corollary 2.4.6.
TO BE DONE !!!
Define the numbers m(a, b, ε), µ(a, b, ε), n(a, b, ε) ν(b, a, ε)
and others carefully and relate them
to the reductions to separable cases ...
Do we really need for (xv)⇒(iv) the case
where the ck are required to be contractions?
Those numbers should be also discussed in Appendices !!!! ????
Let A a simple and non-elementary C *-algebra, b ∈ A+ with kbk = 1, fixed
from now on, and define maps (0, 1] 3 ε 7→ m(b, ε) ∈ N ∪ {∞} with the following
property:
Let a, b ∈ A+ and ε > 0.

20 In general, e.g. for non-simple A, the numbers ν(b, a, ε) ∈ N ∪ {+∞}, can be bigger than

the m(b, ε) ∈ N ∪ {∞} in Part (iv): The cj in (iv) are not necessarily contractions.
 2. CHARACTERIZATIONS OF SIMPLE PURELY INFINITE ALGEBRAS 147

We say that b can be ε-accessed from a in m := m(a, b, ε) steps if there exists

Pm
c1 , . . . , cm ∈ A with kb − k=1 c∗k ack k < ε . If A is not simple, then we allow here
m := +∞ if b is not in the closed ideal span(AaA) generated by a ∈ A+ .
(It should be almost the smallest number m with (b − ε)+ ⊗ p1,1 - a ⊗ 1m in
Mn (A) ???????????)
Compare (b − ε)+ - a ⊗ 1m , i.e., [(b − ε)+ ] ≤ m[a] in Cu(A), carefully
with m := m(a, b, ε). The point is that we allow that m variates with
ε considerably. So it is not at point of start the same as b - a ⊗ 1m
with fixed m ∈ N
What is this/next ? Put it to right place!
For each a, b ∈ A+ there exists a simple separable C *-subalgebra B ⊆ A
with a, b ∈ B and n(c, d, ε; B) = n(c, d, ε; A) and ν(c, d, ε; B) = ν(c, d, ε; A) for all
c, d ∈ B+ .
In case of non-simple A we allow m(a, b, ε) := ∞ if b is not in the closed ideal
of A generated by a, and can define m(a, b, ε) := m ∈ N if (b − ε/2)+ is in the
closed ideal of A generated by a.
?? with no restrictions on the values (except m(b, ε) 6= ∞ in case of simple A).
Let X ⊂ A any subset.
We say that b is approximately accessible in m(b, ε)-steps from X, if for each
a ∈ X with kak = 1 and each ε > 0 there exist c1 , . . . , cm ∈ A – with m := m(b, ε)
Pm
– such that kb − k=1 c∗k ack k < ε . We require here no bound for the norms kck k
(k = 1, . . . , m).
Is it then not simply, -- in case b ≥ 0 --, [(b − ε)+ ] ≤ m(b, ε)[a] in
Cu(A ⊗ K) ????
Equivalent is that there exists γ ∈ (0, ε) and d1 , . . . , dm with (b − γ)+ =
Pm ∗
k=1 dk adk .

Here m := m(b, ε) depends on both of b and ε. But by our assumption this

number m is independent from a ∈ A+ \ {0}. Clearly, the c1 , . . . , cm depend also
on a. It implies, by Lemma 2.1.9, in the notation of Definition 2.4.2 that, for all
0 6= a ∈ A, [(b − ε)+ ] ≤ m(b, ε)[a] in the large Cuntz semi-group Cu(A). Then the
below given arguments can be roughly outlined by saying that for 0 6= a ∈ A+ and
n ∈ N there exists 0 6= g ∈ A+ with n[g] ≤ [a] in case of non-elementary simple A.
The latter is wrong for “elementary” A ∼ = K(H) for any Hilbert space H. There
m(b, ε) behaves almost like the rank of (b − ε)+ ∈ K(`2 (N)) .
Suppose that A is a non-elementary simple C *-algebra and that a given element
b ∈ A+ with kbk = 1 is approximately accessible in m(b, ε)-steps (from every a ∈ A+
with kak = 1).
Then we find for each a ∈ A+ with kak = 1 and each ε ∈ (0, 1), a contraction
d := d(a, ε) ∈ A with d∗ ad = (b − ε)+ .
148 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

In particular, kb − d∗ adk ≤ ε , and we can take m = 1 for the numbers m(b, ε),
i.e., m(b, ε) := 1 for all ε > 0.
Compare next blue with proof of (iv).
What is shorter or more transparent?
Indeed: Let a ∈ A+ with kak = 1 and ε ∈ (0, 1). Define δ := ε/2, f0 (t) := t,
f1 (t) := max(0, t − (1 − δ)) and f2 (t) := (1 − δ)−1 min(t, 1 − δ), for t ∈ [0, 1]+ .
Notice kf0 − f2 k = δ if 1 − δ is in the spectrum of a. Thus if ka − f2 (a)k ≤ δ.
Then g := (a − (1 − δ))+ = f1 (a) has norm kgk = δ. It implies D := gAg 6= 0.
The element a1 := (1 − δ)−1 a − (a − (1 − δ))+ = f2 (a) satisfies ga1 = g = a1 g

and a ≤ a1 ≤ (1 − δ)−1 a .
Let n := m(b, δ) ∈ N from now on in this part of proof. (Notice that we require
for the given map ε 7→ m(b, ε) no kind of regularity or general bound and require
only that m(b, ε) < ∞ for all ε ∈ (0, 1).)
The hereditary C *-subalgebra D = gAg of A is simple and is non-elementary by
assumption on A and because g 6= 0. Thus, D admits an irreducible representation
ρ : D → L(H) such that ρ(D) ∩ K(H) = {0}. It implies that H is not finite-
dimensional. In particular, there exists a projection Q ∈ K(H) of rank = n. Let
λ : Mn → QK(H)Q an isomorphism from Mn onto QK(H)Q, and denote by µ : ϕ 7→
ϕ(1) the evaluation map from C0 ((0, 1], Mn ) onto Mn .
By Lemma 2.1.15(ii), there exists a C *-morphism ψ : C0 ((0, 1], Mn ) → D
with (ρ ◦ ψ)(ϕ) = λ(ϕ(1)) = (λ ◦ µ)(ϕ) for all ϕ ∈ C0 ((0, 1], Mn ) → D. Let
h := ψ(f0 ⊗ p11 ) ∈ D. Then khk = 1 and n = m(b, ε/2) imply that there are
elements c1 , . . . , cn ∈ A with k b − k c∗k hck k < δ – by the assumption that b is
P

approximately accessible in n := m(b, δ)-steps. The Lemma 2.1.9 shows the exis-
P ∗
tence of a contraction e0 ∈ A such that (b − δ)+ = k dk hdk for dk := ck e0 ,
1/2
k ∈ {1, . . . , m}. Notice k dk hdk = e∗ e with e := k ψ(f0 ⊗ pk,1 )dk , and where
P ∗ P

pk,` denote the matrix units of Mn .

The element e is contained in the closed left-ideal L := D · A by Lemma
2.1.7(o) because ψ(C0 (0, 1] ⊗ Mn ) ⊆ D, and the equation a1 g = g implies that
a1 x = x = xa1 for all x ∈ D = gAg, and that a1 y = y for all y ∈ L . In particular,
(b − δ)+ = e∗ e = e∗ a1 e and kek2 = kbk − δ = 1 − δ.
Thus, we obtain that k(b − δ)+ − e∗ aek ≤ (1 − δ)ka1 − ak ≤ (1 − δ)δ . By
Lemma 2.1.9, there exists a contraction f ∈ A such that b − (1 − ε/8)ε + = d∗ ad

for d := ef . In particular, kb − d∗ adk < ε.

We consider in Proposition 2.2.1 also non-separable simple C *-algebras A.
Therefore we “relax” the definition of “approximately accessible” elements b ∈ A+
with kbk = 1 by considering certain separable simple C *-subalgebras B of A with
b ∈ B that we define in the following manner:
We define numbers n(b, a, ε; B) for C *-algebras B and non-zero a, b ∈ B with
b in the closed ideal J(a) of B generated by a, by letting n := n(b, a, ε; B) the
 2. CHARACTERIZATIONS OF SIMPLE PURELY INFINITE ALGEBRAS 149

minimal number n ∈ N such that there exist c1 , . . . , cn , d1 , . . . dn ∈ B with

n
X
kb − d∗j acj k < ε .
j=1

Here we can take dj = cj if a, b ∈ B+ by Remark B.15.1. We can define m(b, ε; B) ∈

N ∪ {+∞} as the least upper bound of the numbers n(b, a, ε; B) if B is simple.
Give ref.s to def’s of n(b, ...)! The definitions of n(b, a, ε; B) ∈ N and
m(b, ε; B) ∈ N for non-zero a, b ∈ B+ says that supa∈B+ \{0} n(b, a, ε; B) =
m(b, ε; B) if B is simple, the m(b, ε; B) is defined as above Where exactly? and if
the set of numbers n(b, a, ε; B) is bounded.
Let A a simple C *-algebra. We say that b ∈ A+ with kbk = 1 is local ap-
proximately accessible if, – for every simple separable C *-subalgebra B ⊆ A with
b ∈ B and with the property that n(x, y, ε; B) = n(x, y, ε; A) for all x, y ∈ B+
– there exists a number m := m(b, ε; B) ∈ N such that, for each a ∈ B+ with
kak = 1 there exist f1 , . . . , fm ∈ A with kb − k fk∗ afk k < ε . This implies that
P

n(b, a, ε; A) ≤ m(b, ε; B).

Now we use that n(b, a, ε; B) = n(b, a, ε; A) for the selected B and obtain that b
is approximately accessible in m := m(b, ε; B)-steps from each a ∈ B+ with kak = 1
(with the c1 , . . . , cm inside B).
It says in particular that b is approximately accessible in B by m(b, ε, B) steps.
We have seen above that this implies that b is approximately accessible by one
step from each a ∈ B+ with kak = 1. By Proposition B.15.2(iii,iv), there exists
for each a, b ∈ A+ a simple separable C *-subalgebra B ⊆ A with a, b ∈ B and
n(b, a, ε; B) = n(b, a, ε; A). Thus, n(b, a, ε; A) = 1 for all non-zero a ∈ A+ . Where
we use that n(b, a, ε; B) ≤ m(b, ε; B) for all non-zero a ∈ B+ .

(i)⇒(iii): Let D ⊆ A be a non-zero hereditary C *-subalgebra. The definition

of J. Cuntz [172, p. 186] says that D contains an infinite projection p 6= 0. It
means that pAp contains a C *-subalgebra that is isomorphic to the Toeplitz algebra
T := C ∗ (s ; s∗ s = 1). Thus, D contains an isomorphic copy of the compact
operators K ⊆ T .
(iii)⇒(ii): Obviously A 6= C.
Let a, b ∈ A+ with kak = kbk = 1 and 1 > ε > 0 be given, and let η := ε/3.
Then e := (a − (1 − η))+ ∈ A+ and D := eAe are non-zero. By assumption, D
contains a non-zero stable C *-subalgebra E ⊆ D.
The multiplier algebra M(E) of the stable C *-algebra E contains a copy
C (s1 , s2 , . . . ; s∗i sj = δij 1) ⊆ M(E) of O∞ unitally, because L(`2 (N)) is unitally
∗

contained in M(E), cf. Remark 5.1.1(8) or Lemma 2.1.7(iv).

If d ∈ E+ ⊆ D+ with norm kdk = 1, then the elements fj := sj d1/2 (j =
1, 2, . . .) are contractions in E ⊆ D that satisfy fi∗ fj = δi,j d .
P ∗
Since A is simple, we find g1 , . . . , gn ∈ A with kb − gk dgk k < η . Let
∗
P
h := 1≤k≤n fk gk ∈ D · A, cf. Lemma 2.1.7(ii). Then kh h − bk < η and, thus,
150 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

0 < 1 − η < khk2 < 1 + η. On the other hand, (a − e)h = (1 − η)h because
h = lim e1/n h and (a − e)e1/n = (1 − η)e1/n . Let c := khk−1 · h, then kck = 1,
h = khk · c and
c∗ ac − b = c∗ a − (1 − η)−1 (a − e) c + (c∗ c − h∗ h) + (h∗ h − b) .

This gives the estimate

kc∗ ac − bk ≤ ka − (1 − η)−1 (a − e)k + |1 − khk2 | + kh∗ h − bk < 3η = ε .

(ii)⇒(i): The compact operators on a Hilbert space H of dimension > 1 do not

satisfy the criteria listed under (ii), because a rank-one projection is not equivalent
to a rank-two projection.
We postpone for a moment the below given proof of the existence of a non-zero
projection p ∈ eAe ⊆ E ⊆ A for any given non-zero hereditary C *-subalgebra E of
A and a suitable non-zero e ∈ E+ with kek = 1.
Let p ∈ A any non-zero projection. The unital C *-subalgebra D := pAp (with
unit p) contains f ∈ D+ with kf k = 1 and 0 ∈ SpecD (f ), because otherwise
pAp ∼= C · p by the Gelfand-Mazur theorem, and A must be isomorphic to the
algebra K(H) of the compact operators on some Hilbert space H. But this has
been excluded above.
By assumption, we find c ∈ A with kc∗ f c − pk < 1/2 . Then f 1/2 cp ∈
pAp is left-invertible in pAp, but is not right-invertible in pAp, because kck2 b ≥
(b1/2 cp)(pcb1/2 ) . It shows that every non-zero projection p ∈ A is infinite. (And is
then properly infinite inside pAp by Lemma 2.1.6 because A is simple.)
It remains to show that every non-zero hereditary C *-subalgebra E of A con-
tains a non-zero projection. We find a “scaling element” in E and modify the
method of J. Cuntz and B. Blackadar in [78] to produce from this element a non-
zero projection in E:
Take e ∈ E+ with kek = 1 and define the commuting positive contractions
a := (4e − 3)+ , b := 4e − (4e − 1)+ and g := 2e − (2e − 1)+ in C ∗ (e)+ ⊂ E+ . Notice
that kak = kbk = kgk = 1, ga = a, bg = g and, therefore, (b − t)+ g = (1 − t)g and
(b − t)+ a = (g − t)+ a = a for all t ∈ [0, 1).
If g is a projection then this is a non-zero projection in E, and nothing has to
be shown anymore. Otherwise, 0 ≥ g − g 2 = g(1 − g) 6= 0, and we can proceed
further in the following way:
By Part (ii) there exists c ∈ A with kc∗ a2 c − bk < (3/5)2 . Then Lemma 2.1.9
gives a contraction d ∈ A that satisfies
d∗ c∗ a2 cd = (b − (3/5)2 )+ .
It follows, by using 52 − 32 = 42 , that the element z := (5/4)acd satisfies
z ∗ z = (1 − (3/5)2 )−1 (b − (3/5)2 )+ .
The element z is in the hereditary C *-subalgebra E because a, b ∈ E. Moreover
kzk = 1, gz = z (by ga = a), and z ∗ zg = g = gz ∗ z (by (b − t)+ g = (1 − t)g for
 2. CHARACTERIZATIONS OF SIMPLE PURELY INFINITE ALGEBRAS 151

t := 9/25). It implies (g − g 2 )z = 0 and (g − g 2 )(1 − z ∗ z)1/2 = 0. Notice that

(z ∗ z)z = z (by z = gz and z ∗ zg = g). Therefore, (1 − z ∗ z)1/2 z = 0. Let

u := z + (1 − z ∗ z)1/2 ∈ E
b := E + C · 1 ⊆ M(E) .

Then (g − g 2 )uu∗ = 0 and

u∗ u = 1 + z ∗ (1 − z ∗ z)1/2 + (1 − z ∗ z)1/2 z + 1 − z ∗ z = 1 .

This shows that u is a non-unitary isometry in E, b and that p := 1 − uu∗ ∈ E is a

projection in E with (g − g 2 ) ≤ p. Thus, p∗ p = p 6= 0.
(iv)⇒(ii): The conditions in Part(iv) on A require that A is non-elementary
and they obviously imply that A is simple. Simple non-elementary C *-algebras are
antiliminary in the sense that each non-zero hereditary C *-subalgebra contains a
non-zero 2-homogenous element, cf. Remark 2.1.16(ii,iii).
Let a, b ∈ A+ with kak = kbk = 1 and 1 > ε > 0 , define η := ε/3 . Then


e := a − (1 − η) + ∈ A+ is non-zero. Let m := m(b, η). By Remark 2.1.16(iii), we
find in the non-zero hereditary C *-subalgebra D := eAe an element d ∈ D+ and
f1 , f2 , . . . fm ∈ D with kdk = 1 and fi∗ fj = δi,j d for i, j = 1, . . . , m, because A is
antiliminary.
P ∗
By assumption, we find c1 , . . . , cm ∈ A with kb − ck dck k < η . Let h :=
P
1≤k≤m k k f c .
Then hh∗ ∈ D, kh∗ h − bk < η , 0 < 1 − η < khk2 < 1 + η and (a − e)h =
(1 − η)h . We get kc∗ ac − bk < 3η = ε for c := khk−1 · h , by the same calculation
as in the proof of the implication (iii)⇒(ii).
If A is simple and has Property pi(n) of Definition 2.0.4 for some n ∈ N then
we can take m(b, ε) := n for each b ∈ A+ and the Property pi(n) excludes the cases
A∼ = Mk for k ≤ n in its definition. It follows that A with Property pi(n) can not
be an elementary C *-algebra, because if A ∼ = K(H) with Dim(H) > n and a, b ∈ A
are contractions with ranks (Rk)(a) = 1 and (Rk)(b) > n then there can not exist
c1 , . . . , cn ∈ A with kb − c∗k ack k < 1/2.
P

(ii)⇒(iv): The simple algebra A can not be elementary, because if p, q ∈ A are

non-zero projections with rank Rk(q) = 1 then there exists c ∈ A with kp − c∗ qck <
1/2, and this shows that all projections in A \ {0} have rank one, i.e., A ∼
= C. The
case A = C has been excluded in Definition 1.2.1.
Thus A satisfies the assumptions of Part (iv) with m(b, ε) = 1 .
(i)⇒(v): We show moreover that for each a, b ∈ A+ and ε > 0 there exist
d1 , d2 ∈ A with d∗1 abd2 = 0 , ka2 − d∗1 a2 d1 k ≤ ε , and kb2 − d∗2 b2 d2 k ≤ ε .
We may suppose that max(kak, kbk) ≤ 1 and that ε ∈ (0, 1), because we can
start here with smaller ε > 0 if necessary.
If ab = 0, then let β := ε/2, d1 := aβ and d2 := bβ . We get kd∗1 abd2 k = 0 ,
ka2 − d∗1 a2 d1 k ≤ ε , and kb2 − d∗2 b2 d2 k ≤ ε by functional calculus, because 0 ≤
t2 − t2+ε ≤ ε for t ∈ [0, 1].
152 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

If ab 6= 0, then there is δ ∈ (0, kabk2 ) with c := (ab2 a − δ)+ 6= 0. The non-zero

hereditary C *-subalgebra D := cAc contains a non-zero infinite projection p by
assumption (i).
The elements pcp and pab2 ap are strictly positive in pAp, because p ∈ cAc and
c ≤ ab2 a. Thus, pab2 ap and pa2 p ≥ pab2 ap are invertible inside the C *-algebra
pAp with unit p.
Let f := (pab2 ap)−1/2 ∈ (pAp)+ . It satisfies f pab2 apf = p. Since f = pf p,
and p ∈ cAc we get f ab2 af = p ≤ δ −1 ab2 a ≤ δ −1 a2 .
The element v := baf is a partial isometry in A and q := vv ∗ = baf 2 ab is a
projection in bAb. The element b2 is strictly positive in bAb, thus qb2 q must be
invertible in qAq, i.e., there is η > 0 with ηq ≤ qb2 q. It follows that ηp = ηv ∗ qv ≤
v ∗ qb2 qv = v ∗ b2 v , and that v ∗ b2 v is invertible in pAp.
All infinite projections in the simple algebra pAp are properly infinite by Lemma
2.1.6. Thus, partial isometries s, t ∈ A with s∗ s = p = t∗ t and ss∗ + tt∗ ≤ p exist.
Let g1 := f s and g2 := baf t = vt. Then g2∗ bag1 = t∗ f ab2 af s = t∗ ps = 0 .
Since f , pa2 p and v ∗ b2 v are invertible positive elements of pAp, we get that g1∗ a2 g2 =
s∗ f pa2 pf s and g2∗ b2 g2 = t∗ v ∗ b2 vt are positive elements in pAp that are invertible
inside pAp. It implies the existence of h1 , h2 ∈ (pAp)+ with h1 (g1∗ a2 g1 )h1 = p
and h2 (g2∗ b2 g2 )h2 = p. The elements ek := gk hk satisfy e∗1 a2 e1 = p = e∗2 b2 e2 and
e∗2 bae1 = 0 .
Lemma 2.1.7(i) applies to pAp with properly infinite p. Thus, there are x1 , x2 ∈
A with x∗1 x1 = a2 , x∗2 x2 = b2 , and x1 x∗1 , x2 x∗2 ∈ pAp. The elements d1 := e1 x1
and d2 := e2 x2 in A satisfy d∗2 bad1 = x∗2 e2 bae1 x1 = 0, d∗1 a2 d1 = a2 and d∗2 b2 d2 = b2 .
(v)⇒(xi): Let a ∈ A+ . If we take a1 := a2 := a1/2 and ε > 0 in (v), then we
get d1 , d2 ∈ A with kd∗i adj − δij ak < ε . Thus, a is properly infinite.
(xi)⇒(vii): The algebra A is simple and non-zero – by our overall pre-
assumption. If every non-zero element a ∈ A+ is infinite and if A is simple then
each non-zero a is also properly infinite by Lemma 2.1.6. In particular, A must
be non-elementary, because rank-one projections are not infinite in the algebra of
compact operators on a Hilbert space.
We show that A satisfies the assumptions of Part (vii) with n := 1:
A is non-zero and is not one-dimensional because A is non-elementary. Let
a, b ∈ A+ non-zero and ε > 0. Since A is simple there exist a minimal m ∈ N such
Pm
that there exist elements d1 , . . . , dm ∈ A with kb − k=1 d∗k adk k < ε . We show
that necessarily m = 1 if a is properly infinite:
Suppose that m > 1 . Then there is δ > 0 with δ · 2(kdm−1 k + kdm k)2 < ε −
kb − k d∗k adk k . Since a ∈ A+ is properly infinite, there are e1 , e2 ∈ A with
P

ka − e∗k aek k < δ and ke∗1 ae2 k < δ . Let fm−1 := e1 dm−1 + e2 dm and fk := dk
Pm−1
for j ≤ m − 2 . Then kb − k=1 fk∗ afk k < ε, which contradicts the minimality of
m > 1 with this property.
 2. CHARACTERIZATIONS OF SIMPLE PURELY INFINITE ALGEBRAS 153

(vii)⇒(iv): Let n ∈ N such that A is not isomorphic to Mk (C) for each k ≤ n

and that, for every non-zero a, b ∈ A+ and ε > 0, there exists d1 , . . . , dn ∈ A such
Pn
that kb − k=1 d∗k adk k < ε .
We show that A satisfies the conditions of Part (iv) with m(b, ε) := n for all
b ∈ A+ and ε > 0 with kbk = 1 :
∼ K(H) for some Hilbert space H, then each projection p ∈
Suppose that A =
K(H) has rank ≤ n. Thus, H has dimension ≤ n, and A is isomorphic to Mk
for some k ≤ n . Since the latter is excluded by the assumptions, A must be
non-elementary.
Let a, b ∈ A+ with kak = 1 = kbk and ε > 0 . We let δ := ε/4 .


Consider f := a − (1 − δ) + ∈ A+ . Then f 6= 0 and (1 − a)f ≤ δf . By
Pn
conditions in (vii), there are d1 , . . . , dn ∈ A with kb − k=1 d∗k f dk k ≤ δ . The
elements ek := f 1/2 dk satisfy k k e∗k ek k ≤ 1 + δ and e∗k (1 − a)ek ≤ δe∗k ek . Thus,
P

k k ek (1 − a)ek k ≤ (1 + δ)δ . We let ck := (1 + δ)−1/2 ek . Then kck k ≤ 1 ,

P ∗

k k c∗k (1 − a)ck k ≤ δ , and kb − k c∗k ck k ≤ 2δ .

P P
Pn
It follows kb − k=1 c∗k ack k ≤ 3δ < ε .
(viii)⇒(vii): Our overall assumption is that A is simple. For simple A the
Definition 2.0.4 of weak pure infiniteness reformulates obviously as follows:
There exists (general) n ∈ N such that for each a, b ∈ A+ with a 6= 0 and every
ε > 0, there exists d1 , . . . , dn ∈ A – depending on (a, b, ε) – such that
n
X
kb − d∗j adj k < ε
j=1

and the algebra `∞ (A) has no irreducible representation of dimension ≤ n.

The latter implies for simple A that the algebra A can not be isomorphic to
Mk (C) for some k ≤ n, because otherwise then `∞ (A) ∼ = Mk (`∞ ) would have
irreducible representations of dimension ≤ n, which is forbidden by (viii).
(i)⇒(vi): We may suppose that kak = 1. Then the hereditary C *-algebra
E generated by (a∗ a − 1/2)+ contains an infinite projection p ∈ E. It follows
that pa∗ ap is invertible in pAp, and the positive element v := (pa∗ ap)−1/2 ∈ pAp
satisfies (va∗ )av = p . Since p is infinite, there exists a (non-zero) stable hereditary
C *-subalgebra D ⊆ pAp.
For every (full) stable hereditary C *-subalgebra D ⊆ A and every b ∈ A there
exists d ∈ A with dd∗ ∈ D and d∗ d = (b∗ b)1/4 by Lemma 2.1.7. Let b = w(b∗ b)1/2
the polar decomposition of b in A∗∗ . The element z := w(b∗ b)1/4 ∈ A∗∗ is in A,
because
z = lim b(1/n + (b∗ b)1/4 )−1
n

∗ 1/4 −1
in norm and (1/n + (b b) ) is a multiplier of A in A∗∗ . Clearly zd∗ d =
z(b∗ b)1/4 = b.
Let x := zd∗ va∗ and y := vd. Then xay = zd∗ va∗ avd = zd∗ pd = b.
154 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

(Alternatively one could combine Lemma 2.1.7 and [616, prop. 1.4.5] and gets
almost the same decomposition.)
(vi)⇒(ii): Let a, b ∈ A+ with kak = kbk = 1 and ε ∈ (0, 1), and let δ := ε/3.
There are x, y ∈ A with x(a − (1 − δ))+ y = b1/2 . It follows that x 6= 0 and e∗ e ≥ b
1/2
for e := kxk(a − 1 + δ)+ y. By Lemma 2.1.9 there exists a contraction f ∈ A
with (ef )∗ (ef ) = (b − δ)+ . The element d := ef is a contraction in A, and dd∗ is
contained in the hereditary C *-subalgebra of A that is generated by (a − (1 − δ))+ ,
because
dd∗ ≤ ee∗ ≤ kxk2 kyk2 (a − (1 − δ))+ .
Thus, d∗ h(a)d = d∗ d and kd∗ d − d∗ adk ≤ kh(a) − ak ≤ δ for the function h(t) :=
min((1 − δ)−1 t, 1) with property (t − (1 − δ))+ h(t) = (t − (1 − δ))+ for t ∈ [0, 1].
Since kd∗ d − bk ≤ δ, we get kd∗ ad − bk < ε .
(ii)⇒(viii): Clearly, if A satisfies (ii) then A satisfies Condition (i) of Definition
2.0.4 of 1-purely infinite algebras.
Let a ∈ A+ with kak = 1, b := (b1 , b2 , . . .) a positive contraction in `∞ (A),
and ε ∈ (0, 1). Define g := η −1 (a − (1 − η))+ for η := ε/8 and let ∆(a) :=
(a, a, . . .) ∈ `∞ (A) for a ∈ A. Then kgk = 1 and, by Part (ii), there exists elements
hk with kbk − h∗k g 2 hk k < η for k ∈ N. It implies kghk k < (1 + η)1/2 and with
dk := (1 + η)−1/2 ghk , that kdk k < 1 and kbk − d∗k adk k ≤ ε/2 for k ∈ N, because

b − d∗k adk = (b − h∗k g 2 hk ) + (1 + η)d∗k (1 − a)dk + ηd∗k adk .

This implies kb − d∗ ∆(a)dk < ε in `∞ (A) for the contraction d := (d1 , d2 , . . .) ∈

`∞ (A).
Since A has no non-zero character by assumption in (ii) and since ∆(A) gen-
erates `∞ (A) as a closed ideal of `∞ (A), it follows that `∞ (A) does not have a
non-zero character. Thus, condition (ii) of the Definition 2.0.4 of 1-purely infinite
algebras is also satisfied.
(ix)⇒(iii): Let D a non-zero hereditary C *-subalgebra of A and a ∈ D+ with
kak = 1. Then the non-zero σ-unital hereditary C *-subalgebra E := aDa ⊆ D is
non-unital – or – zero is isolated in the spectrum of a. In the first case E is stable
by assumption of (ix). In the second case E = pDp is simple and unital with unit
given by the projection p := ϕ(a) using the function

ϕ(t) := min(1, max(0, αt − γ/2)) ,

where γ denotes the minimum of Spec(a) \ {0} and α := (γ + 2)/2γ. In the second
case we can consider a maximal commutative C *-subalgebra C of the unital C *-
algebra pDp. It is not difficult to check that a unital commutative C *-algebra
C with the property that every c ∈ C+ with kck = 1 has all spectral values t ∈
(0, 1) ∩ Spec(c) isolated from 0 and 1 must be of finite (linear) dimension. But this
would cause that E must be isomorphic to Mn for some n ∈ N. And this would
imply that also D and A are elementary, but is excluded by the assumptions on A
in Part (ix).
 2. CHARACTERIZATIONS OF SIMPLE PURELY INFINITE ALGEBRAS 155

Thus, C ⊆ E contains at least one a ∈ C+ with kak = 1, 0 not isolated from

the spectrum of a ∈ C and F := aAa ⊆ D . Then F is a non-zero stable hereditary
C *-subalgebra of D by assumption of Part (ix). Hence, the algebra D contains in
both cases a non-zero stable C *-subalgebra.
(x)⇒(i): Our overall pre-condition on A requires that A is simple and is non-
zero.
Suppose that A has real rank zero and every non-zero projection p ∈ A is
infinite. Then the simplicity of A implies that each projection p is properly infinite
in A by Lemma 2.1.6.
Since every non-zero projection p ∈ A is properly infinite, it suffices to show
that a given non-zero σ-unital hereditary C *-subalgebra D of A contains a non-zero
projection.
By a result of L.G. Brown and G.K. Pedersen [113], A has real rank zero
(RR(A) = 0), if and only if, every non-zero σ-unital hereditary C *-subalgebra D
of A contains an approximate unit consisting of projections. In particular, each
nonzero hereditary D contains a projection p 6= 0. That is properly infinite by our
assumptions.
(Notice here that the definition of the “real rank” in [113] – specialized to the
case RR(A) = 0 – says that A has real rank zero, if and only if, the self-adjoint
invertible elements of A + C · 1 ⊆ M(A) are dense in Asa + R · 1 . It is shown in
[113] that this implies the property (FS), i.e., that the self-adjoint elements with
finite spectrum are dense in Asa . It is proven there that this is equivalent to the
property that every nonzero hereditary C *-subalgebra D ⊆ A has an approximate
unit of D consisting of projections.)
(i)⇒(x): Let p ∈ A a non-zero projection. Then E := pAp is a non-zero
hereditary C *-subalgebra of A. Since A is purely infinite in the sense of J. Cuntz,
E contains a non-zero infinite projection q ≤ p. Thus p is also infinite. Moreover p
is properly infinite because A is simple, cf. Lemma 2.1.6.
By [113], A has real rank zero, if and only if, each non-zero σ-unital hereditary
C *-subalgebra D of A contains an approximate unit consisting of projections.
Recall that D is σ-unital, if and only if, D = eAe for some (inside D) strictly
positive contraction e ∈ D+ . D is then unital, if and only if, 0 is isolated in the
spectrum of e.
Thus, to conclude, it suffices to show that every non-unital σ-unital D is stable,
because then D ∼ = D⊗K ∼ = pAp ⊗ K for each non-zero projection p ∈ D by
L.G. Brown’s stable isomorphism theorem [107], and D has an approximate unit
consisting of projections Pn ∈ D given by Pn := p ⊗ (e11 + . . . + enn ), using the
natural isomorphism D ∼ = pAp ⊗ K.
The stability of D = eAe follows from the fact that for each ε > 0 there exists
d ∈ D with kd∗ edk ≤ ε and d∗ d = (e − ε)+ , because the (two-sided) annihilator
F := Ann((e − ε)+ , D) of (e − ε)+ in D is a non-zero hereditary C *-subalgebra of
156 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

D, and contains a non-zero infinite projection q ∈ F by assumptions in Part (i).

Thus, there exists d ∈ D with d∗ (e − ε)+ d = 0 and d∗ d = e, by Lemma 2.1.7. The
stability of D follows now from [373] – or alternatively from our Corollary 5.5.1 of
our generalized version of the WvN-theorem.
(i)⇒(ix): The proof of the implication (i)⇒(x) applies almost verbatim. It
shows
Can delete next ‘‘red/blue’’ to get ‘‘new version of (ix)’’
that every non-zero projection in A is infinite (even properly infinite by Lemma
2.1.6), and
that every non-zero non-unital σ-unital hereditary C *-subalgebra D of A is
stable.
Alternatively, the stability of non-zero non-unital σ-unital hereditary C *-sub-
algebras D of purely infinite C *-algebra follows also directly from [462, thm. 4.24]:
If D is σ-unital, purely infinite, and has no unital quotient, then D stable. It uses
[373]. Here we can use instead also Corollary 5.5.1 that does not use [373] for
its proof and is not circular intertwined with proofs given here in Chapter 2 or on
other places.
[Check/decide Alternative:]
A is non-elementary and every non-zero non-unital σ-unital hereditary C *-
subalgebra D of A is stable.
Still applicable for new Part (ix)?
Proof of alternative formulation of (ix) out of given (old?) formulation of Part
(ix):
By assumption A is non-elementary, i.e., A ∼
6= K(H) for any Hilbert space H .
In particular pAp is not of finite (linear) dimension for each non-zero projection
∗
p p = p ∈ A. It follows that every maximal commutative C *-subalgebra C of
pAp contains a positive element b ∈ C ⊆ pAp with the property that 0 is not
isolated in the spectrum of b. This implies that the non-zero σ-unital hereditary
C *-subalgebra D := bAb of A is not unital. Hence, D is stable by the assumptions
in Part (ix). By our ”overall assumption”: A is simple. Thus D generates A as
ideal of A. Then Part (ii) of Lemma 2.1.7 shows that there exists c, d ∈ A with
d∗ d = p = c∗ c, cc∗ = dd∗ ∈ D ⊆ pAp and c∗ d = 0. Thus, p is properly infinite in
A, by assumptions.
(xii)⇒(xi): Suppose that A is simple and that there exists n ∈ N such that each
n-homogenous non-zero element a ∈ A+ is infinite, and that A is not isomorphic
to Mk (C) for k = 1, . . . , n − 1.
Since A is simple, it follows that each n-homogenous a ∈ A+ \ {0} is properly
infinite in A. The algebra A is non-elementary, because the projections of rank
equal to n in the algebra of compact operators on Hilbert spaces H of dimension
≥ n are n-homogenous, but – obviously – are not properly infinite.
 2. CHARACTERIZATIONS OF SIMPLE PURELY INFINITE ALGEBRAS 157

It follows that the non-zero closed hereditary C *-subalgebras D of A are also

non-elementary. Hence, each non-zero D contains non-zero n-homogenous elements.
It implies that every non-zero element a ∈ A+ majorizes a non-zero properly infinite
element. Thus, each non-zero a ∈ A+ is itself properly infinite by Lemma 2.1.6.
(xi)⇒(xiii): It is obvious with n := 1.
(xiii)⇒(xii): Suppose that A is simple and there exists n ∈ N with the property
that a⊗1n is infinite in A⊗K for each non-zero a ∈ A+ . Then A is non-elementary
because this can not happen for a ∈ A+ with non-zero a = a2 and aAa = C · a.
Apply Lemma 2.1.6 to the infinite elements a⊗1n and get that a⊗1n is properly
infinite in the simple C *-algebra A ⊗ K for each a ∈ A+ .
Let a ∈ A+ an n-homogenous contraction and ϕ : C0 ((0, 1], Mn ) → A a C *-
morphism with a = ϕ(f0 ⊗ 1n ) for f0 (t) := t (t ∈ [0, 1]). If pjk denote the matrix
units, then 1n := p11 +. . .+pnn is the unit element of Mn . Let c := ϕ(f0 ⊗p11 ) ∈ A+ .
The elements a⊗p11 = ϕ(f0 ⊗1n )⊗p11 and c⊗1n = ϕ(f0 ⊗p11 )⊗1n are Murray–
von-Neumann equivalent in A ⊗ Mn ∼ = Mn (A), because f0 ⊗ 1n ⊗ p11 is unitarily
equivalent to f0 ⊗ p11 ⊗ 1n in C0 (0, 1] ⊗ Mn ⊗ Mn by a unitary in 1 ⊗ Mn ⊗ Mn .
It is easy to check that proper infiniteness of positive elements is invariant under
Murray–von-Neumann equivalence, because the proper infiniteness of an element a
means (a ⊕ a) ≈ a with Cuntz equivalence ≈. Therefore proper infiniteness passes
to all elements b ≈ a. Use that ∼M vN implies the (weaker) equivalence ≈ to obtain
that a⊗p11 is properly infinite in A⊗K. Thus, each positive n-homogenous element
a in A ∼= A ⊗ p11 is properly infinite in A ⊗ K. The definition of proper infiniteness
shows that the positive element a is also properly infinite in A itself, because the
definition of properly infinite elements at the beginning of Section 1 shows that
an inside A properly infinite element a is always in the hereditary C *-subalgebra
D := aAa ∼ = D ⊗ p11 properly infinite if a ⊗ p11 is properly infinite in A ⊗ K, cf. the
argument in proof of Proposition 2.2.5(i).
(xiii)⇔(xiv): Suppose that there exists n ∈ N such that a ⊗ 1n is infinite in
Mn (A) for each non-zero a ∈ A+ . Since Mn (A) is simple, a ⊗ 1n is properly infinite
by Lemma 2.1.6, i.e., A has property pi-n of Definition ??.
The implication (xiv)⇒(xiii) is obvious by the Definition of property pi-n.
(ii)⇒(xv): If A satisfies (ii), then A is non-elementary and the additional
condition kck ≤ 1 can be derived like in the proof of the implication (vi)⇒(ii).
It says that A satisfies the conditions in Part (xv) with ν(a, b, ε) := 1 for all
a, b ∈ A+ and ε > 0.
(xv)⇒(iv): The conditions on the algebra A in Part (xv) imply that A is simple,
because the existence of the numbers ν(a, b, ε) < ∞ implies that the element b is
contained in the ideal generated by a for each a, b ∈ A+ with kak = 1.
Only non-elementary A are allowed in Part (xv) by assumptions. Thus, to
verify that the assumptions of Part (xv) imply the assumptions in Part (iv), we
158 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

are going to show that there exists, for each b ∈ A+ with kbk = 1 and each ε > 0,
a number m := m(b, ε) ∈ N that satisfies the conditions in Part (iv), i.e., has the
property that for any a ∈ A+ with kak = 1 there exists an “m-step inner” c.p. map
V := f1∗ (·)f1 + · · · + fm
∗
(·)fm with kb − V (a)k < ε. (If this is the case, then the
above proven equivalence of Parts (iv) and (ii) shows that, moreover, m(b, ε) = 1
for all b ∈ A+ with kbk = 1 and ε > 0.)
Notice that the definition of the numbers m := m(b, ε) ∈ N in Part (iv) is given
by the upper bound (over all a ∈ A+ ) of the numbers m := m(b, a, ε) ∈ N that are
defined as follows:
Given non-zero a ∈ A+ and ε > 0, then there exists a minimal m ∈ N such that
Pm
there exist elements f1 , . . . , fm ∈ A that satisfy the inequality kb− `=1 f`∗ af` k < ε.
Here without any proposed bound for the norms kf` k.
The reader should observe here that the number ν := ν(b, a, ε) ∈ N is differently
defined. It is the minimal number ν that satisfies the following condition:
For each power ` ∈ N there exist ν(b, a, ε) contractions c1 , . . . , cν ∈ A – depending
Pν
on ε > 0, b, a ∈ A+ and ` ∈ N – such that k b − k=1 c∗k a` ck k < ε . The crucial
point is here that we require here that this number ν is independent from the ` ∈ N.
We consider first the case of separable A, and discuss later the reduction to
the study of suitable simple separable C *-subalgebras B of A that have the same
values ν(b, a, ε; B) = ν(b, a, ε; A) for all a, b ∈ B .
We show for separable simple A that if ν(a, b, ε) < ∞ for all a ∈ A and ε > 0
then m(b, ε) < ∞ for all ε > 0. The latter causes that Part (iv) applies to A and
shows that m(b, ε) = 1, i.e., that A is purely infinite by the implication (iv)⇒(ii).
We use in the separable case Part (iii) of Lemma 2.1.15, Lemma 2.1.17 and the
classical “excision” Proposition A.21.4 for the proof.
Move the Excision Lemma back to Chp.2. !!?
Let b ∈ A+ with kbk = 1 and let ρ denote a pure state on A with ρ(b) = 1.
Any pure state on A would do the job!
Only the below defined c plays a role.
By the above cited Lemmata
(???? Cite more precise !!!)
there exists a positive contraction c ∈ A+ with kck = 1 such that ρ(c) = 1, and
for all a ∈ A,

lim kc2n ρ(a) − cn acn k = 0 .

n

By assumptions of Part (xv), there are numbers ν(b, c2 ; ε) < ∞, for every
ε > 0 with the in Part (xv) quoted properties. But for fixed b and ε > 0 they could
depend possibly on the contraction c ∈ B+ .
Let ε > 0 given and let δ := ε/(2ν(b, c2 ; ε/2)).
 2. CHARACTERIZATIONS OF SIMPLE PURELY INFINITE ALGEBRAS 159

Let a ∈ A+ non-zero, w.l.o.g. , kak = 1 . By Inequality (1.7) in Lemma 2.1.17

there exists a contraction g ∈ A with 1 = kak ≤ δ + ρ(g ∗ ag) ≤ δ + 1. Thus, there
exists n0 ∈ N with kρ(g ∗ ag)c2n − cn g ∗ agcn k < δ for all n ≥ n0 . Hence,

kc2n − cn g ∗ agcn k < 2δ for all n ∈ N, n ≥ n0 .

By assumptions of Part (xv), we find contractions d1 , . . . , dν ∈ A with ν :=

ν(b, c2 ; ε/2) a fixed number, but with the dk also depending from the power n ∈ N
of (c2 )n , such that
ν
X
kb − d∗k c2n dk k < ε/2 .
k=1

Then kfk∗ afk − d∗k c2n dk k < δ for the contractions fk := gcn dk (k = 1, . . . , ν), and
therefore
ν
X
kb − fk∗ afk k < ε/2 + 2ν · δ ≤ ε .
k=1

It shows that for a, b ∈ A+ and ε > 0 and kak = kbk = 1 holds n(a, b, ε) ≤
ν(b, c2 , ε/2) < ∞ , with suitable fixed c ∈ A+ and kck = 1 . Thus m(b, ε) < ∞ for
each b ∈ A+ with kbk = 1 and ε > 0 .
This means that – in case that A is separable – the conditions in Part (xv)
imply the conditions in Part (iv).
Case with separable A is ready here.
Now we discuss how to reduce to the separable case.
By Proposition B.15.2(iii,iv), there exists for each a, b ∈ A+ a separable C *-
subalgebra B ⊆ A with a, b ∈ B and the property that n(e, d, ε; B) = n(e, d, ε; A)
and m(e, d, ε; B) = m(e, d, ε; A) for all ε > 0 and d, e ∈ B+ , where for given C *-
subalgebras B ⊆ A the numbers n := n(e, d, ε; B) are defined for each d, e ∈ B+
and ε > 0 as the minimal number n ∈ N with the property that there exists n con-
Pn
tractions c1 , . . . , cn ∈ B with ke − k=1 c∗k dck k < ε. The above defined (possibly
infinite) numbers ν(b, a, ε; B) ∈ N ∪ {+∞} is then equal to sup`∈N n(b, a` , ε; B).
In particular this causes that B is simple if A is simple, and that ν(a, b, ε; B) =
ν(a, b, ε; A) for a, b ∈ B+ .
Since we can find for each a, b ∈ A+ a separable C *-subalgebra B with
this property, we get finally that A satisfies the assumptions of Part (iv), i.e.,
n(b, a, ε; A) = 1 for all non-zero a ∈ A+ . Where we use n(b, a, ε; B) ≤ m(b, ε; B)
for all non-zero a ∈ B+ .
We start now the study of the above proposed separable case:

We reduce the general case to the separable case by Proposition B.15.2(iii,iv).

It says that, for each a, b ∈ A+ , there exists a simple separable C *-subalgebra
B ⊆ A with a, b ∈ B and n(c, d, ε; B) = n(c, d, ε; A) and ν(c, d, ε; B) = ν(c, d, ε; A)
for all c, d ∈ B+ .
160 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

In particular, n(b, a, ε; A) = n(b, a, ε, B) = 1 . Thus, n(b, a, ε) = 1 for all

a, b ∈ A+ with kak = kbk = 1 . It implies finally that Part (xv) implies Part (iv)
with n(a, b; ε) = 1 .
(xvi)⇒(ii): Let a, b ∈ A+ with kak = kbk = 1 and ε > 0. There is a character χ
on C ∗ (a) with χ(a) = 1. The character χ extends to a linear functional ρ on A with
ρ(a) = 1 = kρk = kχk by Hahn-Banach extension. Consider the maps V : A → C
and W : C → A defined by V (x) := ρ(x) for x ∈ A and W (ξ) := ξ · b. The maps
V and W are completely positive, have norms = 1 and satisfy W (V (a)) = b. By
assumptions of Part (xvi), there exist d1 , d2 ∈ A with d∗1 d2 = 0 and kb − d∗k adk k =
kW (V (a)) − d∗k adk k < ε for k ∈ {1, 2}. Thus, A satisfies the condition in Part (ii).
TEXT-Info (xvi):
For each finite subset F ⊂ A, each n ∈ N, ε > 0, each completely positive con-
tractions V : A → Mn , W : Mn → A there exists contractions d1 , d2 ∈ A (depending
from F, V, W and ε) with d∗1 d2 = 0 and

k(W ◦ V )(x) − d∗k xdk k < ε for all x ∈ F, k ∈ {1, 2} .

From Proposition 3.1.9(ii) :

Suppose that A has the property, that for every positive contraction e ∈ A+ and
ε > 0 there exist contractions e1 , e2 ∈ A with ke∗j ek − δjk ek < ε for j, k ∈ {1, 2}.
This property of e ∈ A+ has been studied also further above in Chp.2...???
Why refer to Chapter 3? Find it and compare the ref.s !!!
Then W can be “orthogonal approximated in norm” – in the sense that for
each γ > 0 there exist c.p. maps U0 , U1 : Mn → A such that Uk (1n ) ≤ V (1n ),
kUk − W k < γ and that U0 and U1 are “elementary” and “orthogonal” in the sense
that there are column-matrices dk ∈ Mn,1 (A) with Uk (β) = d∗k βdk and d∗0 βd1 = 0
for all β ∈ Mn ∼
= Mn (C · 1) ⊆ Mn (M(A)) and k ∈ {0, 1} .
(xi)⇒(xvi): Let F ⊂ A a finite subset, ε > 0, V : A → Mn , and W : Mn → A
completely positive contractions. Let b := W (1n ).
Let 0 < δ := ε/????? , ρ a pure states on A and dρ : A → L(H) with H :=
L2 (A, ρ) := A/Lρ the corresponding irreducible representation with cyclic vector
x0 satisfying ρ(a) = hdρ (a)x0 , x0 i . Here Lρ := {a ∈ A ; ρ(a∗ a) = 0}.
By Lemma 2.1.22 there exist an isometry I from `2 (n) into L(H) with
kI ∗ dρ (a)I − V (a)k < δ for a ∈ F . By simplicity of A we can take any irreducible
representation ρ : A → L(`2 ) and find a suitable isometry T from Cn into `2 such
that kV (a) − T ∗ ρ(a)T k < δ for all a ∈ F .
By assumption of Part (xi), for each a ∈ A+ there exists d, e ∈ A with d∗ e = 0
and d∗ d = (a − γ)+ = e∗ e . This allows to apply Proposition 3.1.9(ii).
Is it better to use an elementary tool? Not from Chp.3??
 2. CHARACTERIZATIONS OF SIMPLE PURELY INFINITE ALGEBRAS 161

It says that each c.p. contraction W : Mn → A can be approximated in norm

by c.p. maps U : Mn → A with U (1n ) ≤ W (1n ), that are “elementary” in the sense
that there are column-matrices d ∈ Mn,1 (A) with U (α) = d∗ αd for α ∈ Mn .
This means that we can find d ∈ Mn,1 (A) with kW (Y ) − d∗ Y dk < δkY k for all
Y ∈ Mn and d∗ d = k d∗k dk ≤ W (1n ) . In particular kd∗ dk ≤ 1 , because W is a
P

c.p. contraction.
Give reference for next ‘‘blue’’
Seems to be better to factorize over a fixed copy of Mn sitting
in A itself ?
By Part (i) and Lemma 2.1.6 there exists a properly infinite projection 0 6=
p∗ p = p ∈ A. Let s, t ∈ pAp partial isometries with s∗ t = 0 and s∗ s = t∗ t = p.
Consider the partial isometry vk := tk s
(i)⇒(xvii): Let D a non-zero hereditary C *-subalgebra of A and p ∈ D an
infinite projection. This means that pDp = pAp contains a non-unitary isometry,
i.e., there exists partial isometry v ∈ D with v ∗ v = p, vv ∗ ≤ p and vv ∗ 6= p. Let 1
denote the unit of M(A). Then T := (1 − p) + v ∈ D + 1M(A) is a non-unitary
isometry in the C *-algebra D + C · 1 ⊆ M(A).
(xvii)⇒(iii): Let D ⊆ A a non-zero hereditary C *-subalgebra of A and 0 6= e ∈
D+ . The hereditary C *-subalgebra E := eAe is non-zero. Let 1 denote the unit-
element of A if A is unital, otherwise let it denote the unit 1M(A) of the multiplier
algebra of A.
By assumptions of Part (xvii), there exists an element T ∈ E + C · 1 ⊆ M(A)
that is left-invertible in A + C · 1 but is not right-invertible in A + C · 1.
Let L ∈ A+C·1 a left inverse for T , i.e., LT = 1. Then kLk2 T ∗ T ≥ T ∗ L∗ LT =
1. It implies that (T ∗ T )−1/2 ∈ E +C·1 exists and that S := T (T ∗ T )−1/2 ∈ E +C·1
is a non-unitary isometry in E + C · 1. The S is non-unitary, because otherwise T
would be also right-invertible.
It follows that p := 1 − SS ∗ is a non-zero projection in E and the hereditary
C *-subalgebra F of E generated by p and S n p(S ∗ )n for n ∈ N is a non-zero stable
C *-subalgebra of E ⊆ D. 

Remark 2.2.2. The Part (xvii) of Proposition 2.2.1 is a “poor-man variant”

of a criterium of L.G. Brown, P. Friis and M. Rørdam. See [109] and [305]:
If A is simple and satisfies the Friis-Rørdam “Property IR”, then there is a
strict alternative that excludes each other:
A is purely infinite or A has “stable rank one”.
More precisely, the alternative for simple A with Property IR is: A is purely
infinite or each element of A + C · 1M(A) ⊆ M(A) is contained in the norm closures
of the invertible (!) elements of A + C · 1M(A) .
See Section 9 in Appendix B for the definitions of Friies-Rørdam “Property
IR” and “stable rank one” and some explanations, in particular why Part (xvii) of
162 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Proposition 2.2.1 is one of the observations for the proof of the Brown-Friis-Rørdam
alternative for simple C *-algebras with Property IR between pure infiniteness and
“stable rank one”.

Next lemma will be used later to prove in some cases pure infiniteness of tensor
products of C *-algebras.

Lemma 2.2.3 ([93], lem. 2.15). Let D be a non-zero hereditary C*-subalgebra

of the minimal C*-algebra tensor product A ⊗ B of C*-algebras A and B.
Then there exists 0 6= z ∈ A ⊗ B with zz ∗ ∈ D and z ∗ z = e ⊗ f for some
non-zero e ∈ A+ and f ∈ B+ .
If d ∈ D+ and pure states ϕ ∈ A∗ , ψ ∈ B ∗ are given with (ϕ ⊗ ψ)(d) > 0, then
such element z ∈ A ⊗ B can be found such that, moreover, ϕ(e)ψ(f ) > 0.

Proof. Let d ∈ D+ with kdk = 1, and let C := A ⊗ B . The minimal C *-

algebra tensor product is the same as spatial tensor product with respect to the
direct sum of irreducible representations of A and B (cf. [704, prop. 1.22.9], [766]).
Thus, there are pure states ϕ on A and ψ on B such that (ϕ ⊗ ψ)(d) > 0.
We can assume from now on, that fixed pure states ϕ and ψ and a fixed
contraction d ∈ D+ with (ϕ ⊗ ψ)(d) > 0 are given, i.e., we prove also the second
part of this Lemma (2.2.3) at the same time.
Then a := (ϕ ⊗ idB )(d) ∈ B+ is a non-zero contraction and 0 < ψ(a) ≤ kak.
Let δ := ψ(a)/2 and f := (a − δ)2+ . Thus 0 < δ ≤ 1/2, f ∈ B+ and ψ(f ) > 0,
because ψ(f )1/2 ≥ ψ(f 1/2 ) ≥ ψ(a) − δ > 0.
There exists a separable C *-subalgebra G of A such that d is in the closure of
the algebraic tensor product G B, because d is the limit of a sequence in A B.
By Proposition A.21.4 and Lemma A.21.3 there exists b ∈ A+ with kbk = 1 = ϕ(b)
(for all m ∈ N) such that

lim kbn gbn − ϕ(g)b2n k = 0 for all g ∈ G .

n

The state ϕ extends to a state ϕe on the unitization A + C1 ⊂ M(A) of A with

ϕe (1 − b) = 0. It gives ϕe (a(1 − b)) = 0 by Cauchy inequality. In particular,
ϕ(a) = ϕ(ab) for all a ∈ A and ϕ(bn ) = 1 for all n ∈ N. Thus, the restriction of ϕ
to C ∗ (b) ⊆ A is a character on C ∗ (b).
The maps

Tn : y ∈ A ⊗ B 7→ (bn ⊗ 1)y(bn ⊗ 1) − (b2n ⊗ (ϕ ⊗ id)(y))

converge on G ⊗ B point-wise to zero, because Tn is a difference of two completely

positive contractions on C and tends on G B point-wise to zero.
Thus, there exists n with kTn (d)k < δ 2 , i.e.,

(b2n ⊗ a) − δ 2 ≤ (bn ⊗ 1)d(bn ⊗ 1)

 2. CHARACTERIZATIONS OF SIMPLE PURELY INFINITE ALGEBRAS 163

1/2
in the unitization of C. Let g := b2n and t := d1/2 (bn ⊗ 1)((g ⊗ a) − δ 2 )+ . Then
we get ((g ⊗ a) − δ 2 )2+ ≤ t∗ t and t C t∗ ⊆ d1/2 C d1/2 ⊆ D .
2
Now let e := (g − δ)+ ∈ A+ . Using that ϕ(g) = kgk = 1 and g ≥ 0, i.e., that
ϕ is a character on C ∗ (g), and that 0 < δ ≤ 1/2, we get ϕ(e) = (1 − δ)2 > 0 .
Recall that f := (a − δ)2+ . Functional calculus in C ∗ (g) ⊗ C ∗ (a) shows that
e ⊗ f ≤ ((g ⊗ a) − δ 2 )+ )2 .
If t = (tt∗ )1/2 v is the polar decomposition of t in the second conjugate of C,
then vxv ∗ ∈ t C t∗ ⊆ D and vx1/2 ∈ C for every x ∈ C with 0 ≤ x ≤ t∗ t, because
x1/2 is in the norm closure of t∗ C t and vt∗ = (tt∗ )1/2 .
Since e ⊗ f ≤ t∗ t we get that z := v((g − δ)+ ⊗ (a − δ)+ ) is in A ⊗ B, and e, f, z
satisfy z ∗ z = e ⊗ f , zz ∗ ∈ D and ϕ(e)ψ(f ) > 0. 

Next Lemma 2.2.4 allows later applications by using suitable central sequences.

Lemma 2.2.4. Let F ⊆ A \ {0} a finite subset of non-zero elements in a C*-

algebra A.
Suppose that f ∗ ∈ F for each f ∈ F and that, for every µ ∈ (0, 1), there exist
contractions s1 , s2 ∈ M(A) (depending on µ) with the properties that for all f ∈ F
and j ∈ {1, 2} holds:

ksj f − f sj k + k(1 − s∗j sj )f k + k(s1 s∗1 s2 s∗2 )f k < µ kf k . (2.1)

Then, for each finite sequence d1 , d2 , . . . , dn ∈ M(A) and ε > 0, there exists
Pn
elements g1 , g2 ∈ A with g1∗ g2 = 0 , gk∗ gk ≤ ∗
`=1 d` d` for k ∈ {1, 2} and

n
X
kgj∗ f gk − δj,k · d∗` f d` k < ε · k f k , ∀ 0 6= f ∈ F, j, k ∈ {1, 2} . (2.2)
`=1

Proof. We show below that the existence of contractions s1 , s2 ∈ M(A) sat-

isfying the inequalities (2.1) for each µ > 0 and fixed given F implies that we can
find for the given f ∈ F the following, – formally stronger –, existence property
(*):

(*) For each γ ∈ (0, 1) there exist contractions t1 , . . . , t2n ∈ A with the
property that t∗j tk = 0 for j 6= k and kt∗j f tk − δj,k f k < γ for all j, k ∈
{1, . . . , 2n} and f ∈ F .

If we have shown the existence (*) – for each given γ > 0 – then we can use the
Pn Pn
contractions t1 , . . . , t2n to define g1 := `=1 t` d` and g2 := `=1 t`+n d` . Clearly,
∗ ∗
Pn ∗
g1 g2 = 0 and gk gk ≤ `=1 d` d` for k ∈ {1, 2}. This elements g1 , g2 ∈ A satisfy
then the Inequalities (2.2):
Indeed, consider the matrices S[f ] := [δj,k f ] ∈ M2n (A) and T [f ] := [t∗j f tk ] ∈
M2n (A), and the columns Z1 := [d1 , . . . , dn , 0, . . . , 0]> ∈ M2n,1 (M(A)) and Z2 :=
[0, . . . , 0, d1 , . . . , dn ]> ∈ M2n,1 (M(A)) . Then, for f ∈ F , gj∗ f gk = Zj∗ T [f ]Zk and
164 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

d∗` f d` = Zj∗ S[f ]Zk . Thus,

P
δj,k

X n
X
kgj∗ f gk − δj,k d∗` f d` k ≤ kS[f ] − T [f ]k · k d∗` d` k .
` `=1

The entries of the self-adjoint matrix S[f ] − T [f ] ∈ M2n (A) have norms ≤ γ by
Property (*). A simple matrix-norm estimate says that an operator m × m matrix
[xjk ] ∈ Mm (A) has operator norm ≤ m max{kxj,k k ; 1 ≤ j, k ≤ m} in Mm (A),
cf. Section 19 of Appendix B and Remark 2.1.10, thus: kS[f ] − T [f ]k ≤ 2n · γ .
Hence, we get elements g1 , g2 ∈ A that satisfy all desired properties if we
can show – for given ε ∈ (0, 1) – the existence of the contractions tk with above
properties (*) for suitable γ ∈ (0, 1) – that satisfies e.g.
X
kd` k2 ≤ ε .


2nγ · 1 +
`

Now we are going to derive the existence of contractions t1 , . . . , t` , t2n ∈ A for given
γ ∈ (0, 1) with the above listed property (*) for given γ ∈ (0, 1) fixed given finite
subset F ⊂ A, with f ∗ ∈ F for f ∈ F .
We avoid complicate iterated estimates for commutators of polynomials and use
instead its translation into precise algebraic relations in corona spaces – together
with a sort of approximate semi-projectivity for relations like x∗j xk f = δj,k f and
kxk k ≤ 1 for j, k ∈ {1, . . . , 2n} and f in a finite set.
Indeed, we derive from the existence of s1 , s2 that satisfy the inequalities (2.1)
for µ = 2−m (m = 1, 2, . . .) the following observation ( 21 ):
Above are used other/same letters, give new notation!
For each n ∈ N there exists of 2n contractions T` = (t`,1 , t`,2 , . . .) ∈ `∞ (A)
(` ∈ {1, 2, . . . , 2n}) that have the following properties :
T`∗ Tk = 0 for k 6= ` , T` f − f T` ∈ c0 (A) and f − T`∗ T` f ∈ c0 (A) for all f ∈ F ,
– where we identify here f ∈ F with (f, f, . . .) ∈ `∞ (A) .
We can find finally the desired contractions t1 , . . . , t` , . . . t2n ∈ A – with the
above discussed desired properties – by picking up entries t` := t`,n0 for suitable
(fixed) n0 ∈ N from the elements T` = (t`,1 , . . . , t`,n0 −1 , t`,n0 , . . .) .
Thus, the proof is complete if we have shown the existence of the contractions
T` ∈ `∞ (A) (` = 1, . . . , 2n) with the above proposed property,
To get such T` ∈ `∞ (A), we modify the elements s1 , s2 ∈ M(A) that satisfy the
inequalities (2.2) with help of a quasi-central approximate unit of A in the sense of
[616, thm. 3.12.14], cf. also Part (3) of Remarks 5.1.1:
If τ ∈ (0, 1) is given, then we use that F ⊆ A is a finite set and find – in a
quasi-central approximate unit of A – a positive contraction e ∈ A+ such that, for

21 Here n ∈ N can not be replaced by “∞” because the used projectivity property does not

work for infinitely many generators! Says that the estimates are delicate and need some care.
 2. CHARACTERIZATIONS OF SIMPLE PURELY INFINITE ALGEBRAS 165

j ∈ {1, 2} and f ∈ F ,

ksj e − esj k < τ and kef − f k + kf e − f k < τ kf k .

It follows for 0 6= f ∈ F that kef − f ek < τ kf k, kef e − f k < τ kf k , ke2 f − f k ≤

2τ kf k and that ks1 e2 s∗1 s2 e2 s∗2 − e2 s1 s∗1 s2 s∗2 e2 k ≤ 4τ . It allows to calculate that the
contractions s1 e, s2 e ∈ A satisfy, for f ∈ F and j ∈ {1, 2}, that the sum of norms

k(sj e)f − f (sj e)k + kf − (sj e)∗ (sj e)f k + k(s1 e)(s1 e)∗ (s2 e)(s2 e)∗ f k

is less or equal to the sum

ksj f − f sj k + kf − s∗j sj f k + ks1 s∗1 s2 s∗2 f k + 7τ · kf k .

Thus, if the contraction e ∈ A+ is chosen as above for sufficiently small τ > 0, then
we obtain that the contractions s1 e and s2 e satisfy again the Inequality (2.1) in
place of s1 and s2 . It follows that our assumptions imply that for each µ ∈ (0, 1)
there are contractions s1 , s2 ∈ A that satisfy the inequalities (2.1). Thus, we can
find the s1 , s2 with this properties in A itself.
Now let A∞ := `∞ (A)/c0 (A), consider A as C *-subalgebra of A∞ via the
natural C *-monomorphism A 3 a 7→ (a, a, . . .) + c0 (A), and let C := C ∗ (F ) ⊆ A ⊆
A∞ the separable C *-subalgebra of A generated by the finite set F .
Let s1,n , s2,n ∈ A contractions that satisfy the inequalities (2.1) for sj,n (j ∈
{1, 2}) in place of sj and with µ := 2−n (n = 1, 2, . . .). Define contractions Sj ∈ A∞
(j ∈ {1, 2}) by Sj := (sj,1 , sj,2 , . . . , sj,n , . . .).
Here is ... ???
Compare above with below def.s and notations !!! ???
We use now that the universal C *-algebra

C ∗ (x1 , . . . , xn ; kxk k ≤ 1 , x∗j xk = 0 , j 6= k , j, k = 1, . . . , n )

is projective, cf. Corollary A.8.5. It shows the existence of the above predicted
contractions T1 , . . . , T2n ∈ `∞ (A) with orthogonal ranges ( 22 ).
This method allows to translate the question, if contractions t1 , . . . , t2n with
the above quoted properties exist, into the question, if we can find isometries

D1 , . . . , D2n ∈ (C 0 ∩ A∞ )/ Ann(C, A∞ )

with mutually orthogonal ranges, i.e., with Dj∗ Dk = δjk 1 for j, k ∈ {1, . . . , 2n} .
It works well because we can lift the isometries Dj by Corollary A.8.5 to con-
tractions
HERE are the Ej defined. Somewhere below called Tj ???

22 Instead using the projectivity of this universal C *-algebra we could use – only for this part

of the proof – that the below considered subalgebras and sub-quotients of A∞ are all “locally”
sub-Stonean to get the desired lifts with pairwise orthogonal ranges. But it isn’t much simpler all
together.
166 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Ej := (ej,1 , ej,2 , . . .) + c0 (A) ∈ C 0 ∩ A∞ with Ej∗ Ek = 0 for j 6= k and

Ej f − f Ej = 0 and f − Ej∗ Ej f = 0, where we identify here the f ∈ F with
(f, f, . . .) + c0 (A).
Then we can use the Corollary A.8.5 a second time to lift the contractions
Ej ∈ A∞ = `∞ (A)/c0 (A) to contractions Tj ∈ `∞ (A) with mutually orthogonal
ranges, i.e., kTj k = 1, Tk∗ Tj = 0, Tj f − f Tj , f − Tj∗ Tj f ∈ c0 (A), where we identify
f ∈ F with (f, f, . . .) ∈ `∞ (A).
Recall that C ⊆ A ⊂ A∞ denotes the separable C *-subalgebra of A generated
by F ⊆ A∞ . The two-sided annihilator of C in A∞ is defined by

Ann(C, A∞ ) := X ∈ A∞ ; X · C = {0} = C · X .

It is easy to see that Ann(C, A∞ ) is a closed ideal of the relative commutant C 0 ∩A∞
of C ⊆ A∞ , cf. [448] for similar definitions and considerations in the – perhaps for
some readers more delicate – case of ultra-powers Aω := `∞ (A)/cω (A), in place of
A∞ := `∞ (A)/c0 (A) .
The contractions t1 , . . . , t2n ∈ A, with the property that

t∗j tk = 0 and kt∗j f tk − δj,k f k < γ for all j, k ∈ {1, . . . , 2n} and f ∈ F ,

can be obtained by lifting the

Where Ek !!!, Tk was defined?
to contractions Tk with pairwise orthogonal ranges in `∞ (A), where we use
that F is a finite subset of A, because then we can select suitable “coordinates”
tk := Tk,n0 (n0 sufficiently large and then fixed) of Tk = (Tk,1 , Tk,2 , . . .) + c0 (A) in
an obvious manner such that the considered inequalities are fulfilled.
Here !!! and a bit above ⇑ starts notation disorder.
New attempts below? E and T and S same?:
The assumptions allow to find for µn = 2−n , n = 2, 3, . . . sequences
(sj,1 , sj,2 , . . .) of contractions in A such that sj,n , j ∈ {1, 2}, define an element
Sj := (sj,1 , sj,2 , . . .) ∈ `∞ (A) (j ∈ {1, 2}) such that, for f ∈ F and k ∈ {1, 2},

Sj f − f Sj , f − (Sj∗ Sj )f , (S1 S1∗ S2 S2∗ )f ∈ c0 (A) .

Let A∞ := `∞ (A)/c0 (A) ⊃ A, and consider the bounded sequences in A

(1) (2)
modulo c0 (A) as elements of A∞ , i.e., Sk := (sk , sk , . . .)+c0 (A) where we embed
here A into A∞ by a 7→ (a, a, . . .) + c0 (A).
Then it happens in the algebra A∞ that kSk k ≤ 1, S1∗ S2 = 0, Sk f = f Sk and
(1 − Sk∗ Sk )f = 0 for all f ∈ F and k ∈ {1, 2}.
It implies (1 − Sk∗ Sk )(f ∗ f + f f ∗ ) = 0 for all f ∈ F = F ∗ . Since 1 − Sk∗ Sk ≥ 0
it follows that F ∪ F · F is contained in the two-sided annihilator of the positive
element 2 − S1∗ S1 − S2∗ S2 of A∞ . Thus, the C *-algebra C := C ∗ (F ) ⊆ A ⊆ A∞ is
contained in the two-sided annihilator of 2 − S1∗ S1 − S2∗ S2 .
Using that f ∗ ∈ F for all f ∈ F we get that F ⊆ {S1 , S1∗ , S2 , S2∗ }0 .
 2. CHARACTERIZATIONS OF SIMPLE PURELY INFINITE ALGEBRAS 167

It follows that the elements of the C *-subalgebra C of B := A∞ ???? generated

by F commute with S1 and S2 . Thus S1 , S2 ∈ C 0 ∩B, i.e., the elements Sk commute
with all elements in the C *-subalgebra C ⊆ B generated by F .
Since F ∗ = F , we get that C is the closure of the linear span of products of
elements f ∗ + f , f ∗ f and f f ∗ with f ∈ F .
This implies that (1−Sk∗ Sk )C = {0}, Sk c−cSk = 0 for all c ∈ C, ... (k ∈ {1, 2}).
???? CHECK !!! ???
Y := Ann(1 − Sk∗ Sk , B) ⊆ B := `∞ (M(A))/c0 (A) ⊇ A∞ is a hereditary C *-
subalgebra of B with F ⊆ Y . It follows C := C ∗ (F ) ⊆ Y . Thus (1−Sk∗ Sk ) ∈ C 0 ∩B.
Thus we can consider first the two-sided annihilator Ann(C, B) of C in B and
the commutator C 0 ∩ B of C in B.
Then Ann(C, B) is a closed ideal of C 0 ∩ B. If we find isometries T1 , . . . , Tn ∈
(C 0 ∩ B)/ Ann(C, B) with pairwise orthogonal ranges. Then we can lift them to
contractions t1 , . . . tn in C 0 ∩ B with pairwise orthogonal ranges such that t∗j ctk =
δj,k c for all 1 ≤ j, k ≤ n and t∗j tk = 0 for j 6= k .
If d1 , . . . , dn ∈ M(A) ⊆ B are given, then g := k tk dk satisfies g ∗ g ≤ k d∗k dk
P P

and g ∗ f g = d∗k f dk .
P

Now one can take take ??or?? find mutually orthogonal representing sequences
for the element g ∈ B and of the d1 , . . . , dn .
By Corollary A.8.5, we get that the universal C *-algebra generated by n con-
tractions with orthogonal ranges is projective.
We consider first the case of isometries s1 , s2 ∈ M(A) with the quoted prop-
erties, to make the idea of the proof transparent. The additional study for the
approximate version will be added at the end of the proof.
Notice that the isometries s1 , s2 with s∗k s` = δk,` (k, ` ∈ {1, 2}) define isometries
by t1 := s1 and tn+1 := sn2 s1 for n = 1, 2, . . . that satisfy t∗m tn = δm,n 1 for n, m ∈ N.
If a ∈ F satisfies ksk a − ask k < γ (k = 1, 2) for some γ ∈ (0, ∞) and k ∈ {1, 2}
then, using that t∗m (atn − tn a) = t∗m atn − δm,n a, we get

kt∗m atn − δm,n ak ≤ katn − tn ak < nγ .

Follows from

ka(b1 · b2 ) − (b1 · b2 )ak ≤ kb2 kk[a, b1 ]k + kb1 kk[a, b2 ]k .

What about working in M(A)∞

and replacing then γ by zero, e.g.
s1 , s2 ∈ M(A)∞ with s∗1 s2 = 0, (1 − s∗k sk )a = 0 and [a, sk ] = 0
for a ∈ F , k ∈ {1, 2} ...
It implies (sk2 s1 )∗ a(s`2 s1 ) = δk,` a for a ∈ F , k, ` ∈ {0, 1, . . . , n}, with
s02 := 1.
168 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

The problem is:

Does there exist a ‘‘local unit’’ e ∈ M(A)∞ with 0 ≤ e ≤ 1 for F
such that ea = a = ae for a ∈ F
and (sk2 s1 )∗ e(s`2 s1 ) = δk,` e ??
Let s1 , s2 ∈ A contractions with s∗1 s2 = 0 , ksk a − ask k < γ · kak and k(1 −
s∗k sk )ak < γ · kak (k ∈ {1, 2}) for all a ∈ F and some γ ∈ (0, 1). We can define
contractions tn ∈ M(A) by t1 := s1 and tn+1 := sn2 s1 for n = 1, 2, . . ., and show
that, for each a ∈ F ,
ktn a − atn k ≤ n · γkak and kt∗m atn − δm,n ak ≤ 2 min(m, n) · γkak .
It implies that k(1 − t∗n tn )ak ≤ 3n · γkak.
We use the convention s02 := 1 ∈ M(A). By assumptions, kt1 a − at1 k ≤ γkak
for t1 := s1 and a ∈ F . Then [a, xy] = x[a, y] + [a, x]y shows that tn+1 a − atn+1 =
sn2 (s1 a − as1 ) + (sn2 a − asn2 )s1 and
ksn2 a − asn2 k ≤ nks2 kn−1 ks2 a − as2 k ≤ n · γkak .
Since s1 and s2 are contractions we get ktn a − atn k ≤ n · γkak for all n = 1, 2, . . . .
An estimate for kt∗m atn − δm,n ak :
Let s0k := 1 for k ∈ {1, 2}. It suffices to consider the case where 1 ≤ m ≤ n because
(t∗m atn )∗ = t∗n a∗ tm and a∗ ∈ F for a ∈ F . If 1 ≤ m ≤ n then tm = s2m−1 s1 ,
tn = sm−1
2 sn−m
2 s1 and δn,m = 1 if and only if n = m.
X := (s∗2 )m−1 asm−1
2 − a , Y := s∗1 Xsn−m
2 s1 and Z := s∗1 asn−m
2 s1 − δm,n a ,
∗ ∗
satisfy Y + Z = tm atn − δm,n a . Thus ktm atn − δm,n ak ≤ kXk + kZk .
If δm,n = 1 then m = n and Z = s∗1 as1 − a. If δm,n = 0 then m < n,
Z = s∗1 (s2 a − as2 )(sn−m−1
2 s1 ). The latter implies kZk ≤ γkak.
The equations s∗k ask − a = s∗k (ask − sk a) − (1 − s∗k sk )a imply ks∗k ask − ak ≤
2 · γkak for k ∈ {1, 2}. In particular, kZk ≤ 2 · γkak if δn,m = 1. We show that
kXk ≤ 2(m − 1) · γkak:
Pp
Let p > 1. Then (s∗2 )p a(s2 )p − a = `=1 (s∗2 )p−` (s∗2 as2 − a)(s2 )p−` .
Hence, k(s∗2 )p a(s2 )p − ak ≤ 2p · γkak .
Summing up, our calculations show that
kt∗m atn − δm,n ak ≤ kXk + kZk ≤ 2 min(m, n) · γkak .

Consider here the easy case of isometries s1 , s2 .

Let a ∈ F , ε > 0 and d1 , . . . , dn ∈ A given, take µ := k k d∗k dk k.
P

We find for each γ > 0 isometries s1 , s2 ∈ M(A) with ksk a − ask k < γ???? for
k = 1, 2 and a ∈ F .
Let ?????
γ := ε / 1 + 2n (kd1 k + . . . + kdn k)2 .


 2. CHARACTERIZATIONS OF SIMPLE PURELY INFINITE ALGEBRAS 169

By assumptions, we find isometries s1 , s2 ∈ M(A) with s∗1 s2 = 0 and ksk a − ask k <
γ for k ∈ {1, 2}. The above from the s1 , s2 defined isometries t1 , . . . tn satisfy

kd∗i (t∗i atj − δij a)dj k < nγkdi kkdj k .

Pn
Let c := ti di . Then c∗ c = i d∗i di , c∗ ac − i d∗i adi = ∗ ∗
P P P
i,j=1 di (ti atj − δij a)dj
and
Xn X n
kc∗ ac − d∗i adi k ≤ kd∗i (t∗i atj − δij a)dj k < ε .
i=1 i,j=1

Detailed calculations:
c∗ ac = i,j x∗i t∗i atj xj .
P

kc∗ ac − bk ≤ kc∗ ac − x∗i axi k + kb − x∗i axi k .

P P

kb − x∗i axi k < ε/2 by property of the chosen xi .

P
P ∗ P ∗
i xi axi = i,j xi δij axj .

n
X n
X
kx∗i (t∗i atj − δij a)xj k ≤ kxi k · kxj k · kt∗i atj − δij ak .
i,j=1 i,j=1

Using that k[a, sk ]k ≤ γ := ε/2n(1 + ( i kxi k)2 ), and n maxk=1,2 k[a, sk ]k ≥

P
Pn
katj −tj ak ≥ kt∗i atj −δij a)k for i, j ≤ n, we get for i,j=1 kxi k·kxj k·kt∗i atj −δij a)k
the estimate n(kx1 k + . . . + kxn k)2 · γ ≤ ε/2 .


Recall here that γ := ε/ 2n(1 + ( i kxi k)2 ) . END OF THE EASY CASE
P

is here !!!???
Let a ∈ F , ε > 0 and d1 , . . . , dn ∈ A given, take µ := k k d∗k dk k.
P

By assumptions, we find for each δ > 0 contractions s1 , s2 ∈ M(A) with

s∗2 s1
= 0 and ksk a − ask k < δ and k(1 − s∗k sk )ak < δ for k = 1, 2 and a ∈ F .
Pn
Attempt (?? !!!): g := k=1 tk dk for sufficiently small γ.
Calculate kg ∗ ag − k d∗k adk k for a ∈ F . Uses matrix-norm of [t∗j atk − δjk a] for
P

contractions a ∈ F and k d∗k dk k with suitable bound.

P


We list some sufficient criteria for a C *-algebra A to be simple and purely

infinite:

Proposition 2.2.5. Let A be a non-zero C*-algebra. Each one of the following

permanences or properties (i)–(v) implies that A is simple and purely infinite.

(i) A is a hereditary C*-subalgebra of a simple purely infinite C*-algebra.

(ii) A ⊗ K ∼= B ⊗ K for a simple purely infinite C*-algebra B and the compact
operators K on some Hilbert space H.
(iii) A is isomorphic to the inductive limit indlimn (hn : An → An+1 ) of non-
zero C*-algebras An 6= C · 1 with non-zero C*-morphisms hn that satisfy:
(*) For every positive integer n, every ε > 0 and every a, b ∈ (An )+
with kbk ≤ kak = 1, there exists a contraction c ∈ An+1 such that
kc∗ hn (a)c − hn (b)k < ε .
170 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

(In particular, the property (*) is satisfied if all An are simple and
purely infinite.)
(iv) The C*-algebra A is simple and has the following property (∆):
(∆) For every a ∈ A+ and ε > 0, the multiplier algebra M(A) contains
isometries s1 , s2 , depending on (a, ε), with

s∗1 s2 = 0 and ksj a − asj k < ε for j ∈ {1, 2} .

(v) The C*-algebra A is isomorphic to the relative commutant D0 ∩ B of D

in B, where D is a separable C*-subalgebra of a unital C*-algebra B that
satisfies following properties (α) and (β):
(α) D0 ∩ B 6= C1.
(β) For every a ∈ (D0 ∩ B)+ with kak = 1, there exists a contraction
s ∈ B with d = s∗ ads for every d ∈ D.

Remark 2.2.6. It is evident that Proposition 2.2.5(iii) and Proposition 2.2.1(ii)

together imply that inductive limits of simple purely infinite C*-algebras are again
simple purely infinite C*-algebras (23).
Parts (i) and (ii) of Proposition 2.2.5 together show invariance of the class of
simple p.i. C *-algebras under Morita equivalence.
Notice also that every simple purely infinite C *-algebra A is isomorphic to each
of the types of algebras considered in Parts (i)–(iii):
Indeed, take B := A ⊗ K to fulfill the conditions in Parts (i) and (ii), and
take An := A and hn := idA for the conditions in Part (iii). The Condition (*)
in Part (iii) is satisfied for hn = idA , because if a, b ∈ A+ are contractions with
kak = 1 and kbk = 1 then there exists c ∈ A with kc∗ ac − bk < ε by Part (ii) of
Proposition 2.2.1. The arguments in the proof of the implication (iv)⇒(ii) or of
the implication (iii)⇒(ii) for Proposition 2.2.1(ii,iii,iv) show that we can take here
c ∈ A with kck ≤ 1.
That the above examples satisfy the assumptions (i–iii) can be seen from
Parts (??? to be listed !!!) of Proposition 2.2.1. (Must show for (i,ii) that A⊗K
is p.i. if A p.i. and that pi passes to non-zero hereditary C *-subalgebras. The
latter is done in the proof.)
For the general general property pi(1) (which is equal to “purely infinite” in
sense of Definition 1.2.1) the passage to M2 (A) is equivalent to the (not obvious)
proof that Property pi(1) is equal to Property pi-1.
But for simple A the equivalence of pi(1) and pi-1 is given by of Parts (???)
and (???) of Proposition 2.2.1 ??? Condition (*) in Part (iii) implies injectivity

of the hn :
If the hn are non-zero (as required !) then Condition (*) causes that each hn is
injective.
23If they are “non-zero” in the sense that the all C *-morphisms ψ
m,n : Am → An are non-
zero!
 2. CHARACTERIZATIONS OF SIMPLE PURELY INFINITE ALGEBRAS 171

Indeed, since hn is non-zero there exists b ∈ (An )+ with kbk = 1 and khn (b)k = 1.
Let a ∈ (An )+ with kak = 1. Then condition (*) on hn provides c ∈ An+1 with
kc∗ hn (a)c − hn (b)k < 1/2 . It implies hn (a) 6= 0, and that each hn is faithful.

The Property (∆) of Part (iv) holds for all σ-unital purely infinite simple C *-
algebras A.
Indeed, by Proposition 2.2.1(v) all simple purely infinite C *-algebras are
strongly purely infinite, Theorem ?? (A.spi.implies.M(A).spi???.is where?) says
that the multiplier algebra M(A) of every σ-unital strongly purely infinite C *-
algebra A is again strongly purely infinite and Theorem ?? (nuclear.embed.of.exact)
says that each nuclear C *-morphism h : C → E of a separable exact C *-
algebra into a strongly purely infinite “corona” C *-algebra E – e.g. as it is the
case for E = M(A)∞ := `∞ (M(A))/c0 (M(A)) – extends to a C *-morphism
hext : C ⊗ O∞ → E with hext (c ⊗ 1) = h(c). Then apply this to
ι : C ∗ (a, 1) → M(A) ⊆ M(A)∞ .
It gives that all (not necessary simple) σ-unital strongly purely infinite C *-algebras
A satisfy Property (∆) of Part (iv) of Proposition 2.2.5.
The approximate decomposition condition (∆) in Part (iv) of Proposition
2.2.5 is satisfied for every not necessarily simple but strongly purely infinite
σ-unital C *-algebra A, because for them C 0 ∩ M(A)ω is purely infinite for ev-
ery nuclear embedded separable exact C *-subalgebra C of Aω . This applies to
C := C ∗ (πω (a, a, . . .)) ∈ Aω and yields the existence of the s1 , s2 ∈ M(A) with
the property in Part (iv).
Simple purely infinite C *-algebras are strongly purely infinite by Proposition
2.2.1(v). Thus, the condition in Proposition 2.2.5(iv) is moreover equivalent to pure
infiniteness of σ-unital simple C *-algebras A.

Proof of Proposition 2.2.5. The assumptions in Parts (i), (ii) or (iv) as-
sume or imply the simplicity of A. But we prove also the similar statement for
non-simple C *-algebras if possible, i.e., if the condition itself implies the simplicity.
???
It is not difficult to see that each of the properties (i)–(iv) of A in Proposition
2.2.5 implies that A is simple and that A is not isomorphic to the compact operators
on some Hilbert space. Why it is easy to see from (iv)? i.e., where A is
simple and has Property (∆): The K(H) do not satisfy (∆).
(i): Part (i) follows straight from the definition of J. Cuntz, cf. [172, p. 186]
or Part (i) of Proposition 2.2.1, in case of simple C *-algebras.
But we use here the general Definition 1.2.1, because it is also suitable for non-
simple C *-algebras B. The proof that non-zero hereditary C *-subalgebras D ⊆ B
satisfy Part (ii) of Definition 1.2.1 is easy:
If a, b ∈ D+ and ε > 0 are given with a 6= 0 and b in the closed ideal of B generated
by a, then, by Definition 1.2.1, there exists g ∈ B with kg ∗ ag − bk < ε/3 , because
172 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

b ∈ D+ is in the closed ideal of D generated by a, then it is also in the closed ideal

of B generated by a.
Let γ := kgk, δ := ε/(3γ 2 + 3) and ψ(t) := min 1, δ −1 · max(0, t − δ) .

Define g1 ∈ D by g1 := ψ(a)1/2 gψ(b)1/2 . Then kg1∗ ag1 − bk < ε .

But we must prove that quotients B/J of B can not contain non-zero projec-
tions p ∈ B/J with p(B/J)p = C · p. This requires to show first that quotients
B/J are also purely infinite in sense of Definition 1.2.1.
It is obvious from the Definition 1.2.1 that non-zero ideals J of B (respectively
non-zero quotients B/J of B) can not have a non-zero character, because otherwise
such a character extends to, respectively defines, a non-zero character of B.
If a, b ∈ B+ and πJ (b) is in the closed ideal of B/J that is generated by πJ (a) the
Pn
for ε > 0 there exists δ > 0 and c1 , . . . , cn ∈ B with (b−ε/3)+ − k=1 c∗k (a−δ)+ ck ∈
J . By Part (ii) of Definition 1.2.1 there exists d ∈ B with
n
X
kd∗ ad − c∗k (a − δ)+ ck k < ε/3 .
k=1

This implies together that

kπJ (b) − πJ (d)∗ πJ (a)πJ (d)k < ε .

The map πJ : B → B/J is surjective. Hence B/J is again purely infinite in the
sense of Definition 1.2.1 for each closed ideal J 6= B of B.
Let D ⊆ B is a non-zero hereditary C *-subalgebra of purely infinite C *-algebra
B. We show that D can not have a non-zero character (and is therefore also purely
infinite in sense of Definition 1.2.1):
Suppose that D has a non-zero character ξ : D → C. The kernel K ⊂ D of ξ is
a closed ideal of D. The closed ideal J of B generated by K has the property that
J ∩ D = K, cf. ?????
Cite here the intersection Lemma ????
The argument (for the in Definition 1.2.1 required non-existence of characters
on D) is the following – that works also in the non-simple case:
THE NEXT HERE is an application of the study of
residually anti-liminariy C *-algebras. Refere to there!!!
By assumption (i) of Definition 1.2.1, the algebra B has no characters. But the
Property (ii) of Definition 1.2.1 implies then moreover that no irreducible represen-
tation λ of B contains a non-zero compact operator in its image, because otherwise
λ must be necessarily one-dimensional representation, by Property 1.2.1(ii) and
the combination of the Kadison transitivity theorem and the semi-projectivity of
C0 ((0, 1], Mn ) applied in Lemma 2.1.15, i.e., λ would be a character on B. (No-
tice here that the Properties 1.2.1(i,ii) pass to quotients B/J of B and hereditary
C *-subalgebras of B also in case where B is non-simple.)
 2. CHARACTERIZATIONS OF SIMPLE PURELY INFINITE ALGEBRAS 173

The property of a C *-algebra B having only irreducible representations that

do not contain non-zero compact operators passes as a property to all non-zero
quotients of B.
It passes also to all non-zero hereditary C *-subalgebras of B, because for here-
ditary D ⊆ B each irreducible representation ρ : D → L(H1 ) extends, up to unitary
equivalence, uniquely to an irreducible representation λ : B → L(H1 ⊕2 H2 ) such
that λ(d) = ρ(d) ⊕ 0 for d ∈ D.
(ii): By Part (i), the Definition 1.2.1 of pure infiniteness passes to non-zero
hereditary C *-subalgebras. In case of simple C *-algebras it is equivalent to all
other definitions of pure infiniteness (except being only stably properly infinite !).
It remains to observe that the simple C *-algebra B ⊗ K is purely infinite in the
sense of J. Cuntz if B is purely infinite:
Let D ⊆ B ⊗ K be any non-zero hereditary C *-subalgebra.
By Lemma 2.2.3 there exists non-zero z ∈ B ⊗ K with zz ∗ ∈ D and z ∗ z = b ⊗ a
for some (non-zero) b ∈ B+ and a ∈ K+ . Then w = z(1 ⊗ v) satisfies w∗ w = b ⊗ p11
and ww∗ ∈ D for a suitable element v ∈ K . Thus, D contains the closure of
w(B ⊗ K)w∗ which is isomorphic to the (non-zero) closure of bBb (cf. Remark
2.3.1). Now use that b 6= 0 and B is purely infinite in the sense of J. Cuntz. It says
that bBb and D contain both (non-zero) infinite projections p ∈ bBb and q ∈ D,
that are Murray–von-Neumann equivalent in B ⊗ K. Thus, B ⊗ K is purely infinite
in the sense of Cuntz.
(iii): We establish here a more general necessary and sufficient condition for
inductive limits of C *-algebras to be simple and purely infinite. It is easy to see
that the assumptions of Part (iii) satisfy this condition:
Let hn : An → An+1 be a sequence of C *-morphisms. Then for m < n

hm,n := hn−1 · . . . · hm+1 hm : Am → An

is a C *-morphism. We embed the inductive limit B := indlim (hn : An → An+1 )

Q∞ L∞
naturally in ( n=1 An )/( n=1 An ) in the below described way.
We explain sometimes our viewpoint by use of the different notations
∞
Y ∞
M
( An )/( An ) = `∞ (A1 , A2 , . . .)/c0 (A1 , A2 , . . .) ⊆ L(H)∞
n=1 n=1

if the An are considered as C *-subalgebras of L(H).

The canonical morphisms h∞
n : An → B are then given by

h∞ ∞
n (a) := (0, . . . , 0, a, hn,n+1 (a), hn,n+2 (a), . . .) + ⊕k Ak ∀ a ∈ An .

Thus ker(hn,k ) ⊆ ker(hn,k+1 ) ⊆ ker(h∞ ∞

n ) and the kernel of hn is the closure of the
union of the kernels of hn,k for k = n + 1, n + 2, . . .. If the hn are injective, then
the hm,n : Am → An and the h∞ n : An → B are injective. In this case it follows
dim(B) > 1 if dim(An0 ) > 1 for some n0 .
174 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

In the inductive limit B holds h∞ ∞

m (Am ) ⊆ hn (An ) for m < n, and B is the
S ∞
closure of hn (An ). We show that
B is simple and purely infinite, if and only if,

(1) Dim(B) > 1 , and

(2) for every m ∈ N, ε > 0, and e, f ∈ (Am )+ with kf k ≤ 1 = kek =
khm,m+k (e)k for k = 1, 2, . . ., there exist p > n > m and a contraction
g ∈ An such that the distance of hm,n (f ) − g ∗ hm,n (e)g from ker(hn,p ) is
less than ε.

Note that the distance from c := hm,n (f ) − g ∗ hm,n (e)g in An to ker(h∞ n ) is less
∞
than ε, if and only if, khn (c)k < ε, if and only if, there exists p > n such that c
has distance from ker(hn,p ) less than ε.
For the proof of Proposition 2.2.1(ii) it suffices to consider the only elements a
and b in n h∞
S
n (An ) with 0 ≤ a , 0 ≤ b and kak = kbk = 1 . Let such a and b given
and let m ∈ N such that a, b ∈ h∞
m (Am ). Then there exist contractions e, f ∈ (Am )+
∞ ∞
with hm (e) = a and hm (f ) = b. It implies 1 ≤ khm,m+k (e)k ≤ kek ≤ 1 and
kf k = 1.
By (2) there exits, for given ε > 0. indices n > m and a contraction g ∈ Am
such that kb − d∗ adk < ε for the contraction d := h∞
m (g). Thus, (1) and (2) imply
together that B is simple and purely infinite.
Conversely, suppose that B is simple and purely infinite. Then Dim(B) > 1 by
Proposition 2.2.1. Let ε > 0 and e, f ∈ (Am )+ with kf k ≤ 1 = khm,m+k (a)k =
kak .
Then a := h∞ ∞
m (e) and b := hm (f ) are in B+ and γ := kbk ≤ 1 = kak . By
Proposition 2.2.1(ii), there exist a contraction d ∈ B such that kb − γd∗ adk < ε . If
we replace d by a small perturbation of d it satisfies the same inequality. Therefore
we can assume that d ∈ h∞ n (An ) for some n > m. If g ∈ An is a contraction with
∞ 1/2
hn (g) = γ d, then there exists p > n such that (2) is satisfied for m, e, f, ε, g, n, p.
Thus (1) and (2) is satisfied if B is simple and purely infinite.
(iv): The algebra A is supposed to be simple by assumption in Part (iv). But
we show that all – not necessarily simple – non-zero C *-algebras A with Property
(∆) are purely infinite in sense of Definition 1.2.1:
If the algebra A 6= {0} satisfies Property (∆), then A can not have a character
because M(A) contains two isometries s1 , s2 with orthogonal ranges. (In particular
A is not commutative, especially A 6= C.)
Let a, b ∈ A with kak = 1 and let ε > 0 .
We can consider here also non-simple A with property (∆). But then we must
require in addition that b is in the closed ideal generated by a, i.e., that kπJ (b)k = 0
if kπJ (a)k = 0 for all ideals J of A. It is enough to know this for all primitive ideals
J of A.
 2. CHARACTERIZATIONS OF SIMPLE PURELY INFINITE ALGEBRAS 175

Since b is in the ideal generated by a (that is automatic if A is simple and a is

non-zero), there are x1 , x2 , . . . , xn in A such that kb − k x∗k axk k < ε/2 .
P

The set X := {a} ⊂ A satisfies the assumptions of Lemma 2.2.4. Thus, there
exists for x1 , . . . , xn an element d ∈ A with
X X
kdk2 = k x∗k xk k and kd∗ ad − x∗k axk k < ε/2 .
k k

In particular, kb − d∗ adk < ε.

Thus, Property (∆) implies that A satisfies all conditions of the Definition
1.2.1 of purely infinite C *-algebras. If A is simple, then A satisfies the criterium in
Proposition 2.2.1(ii) for A being purely infinite. It causes that A is strongly p.i. in
sense of Definition 1.2.2 if A is simple, cf. Proposition 2.2.1(v).
(v): We use criterion (ii) of Proposition 2.2.1. Notice first that A := D0 ∩B 6= C
by our assumptions. For a ∈ (D0 ∩B)+ and kak = 1 there are unital *-epimorphisms
h1 : C(Spec(a))⊗D → C ∗ (a, D) and h2 : C(Spec(a))⊗D → D given by h1 (f ⊗d) =
f (a)d and h2 (f ⊗ d) = f (1)d for f ∈ C(Spec(a)) and d ∈ D. If f ⊗ d is in the
kernel of h1 , and d 6= 0, then f (a) = 0, because D is simple. Thus f = 0 on
Spec(a). Therefore, by Lemma 2.2.3, the kernel of h1 is zero. By assumption, there
exists an isometry s in B with s∗ f (a)ds = h(f (a)d) = h2 ((h−1 1 )(f (a)d)) = f (1)d
for f ∈ C(Spec(a)) and b ∈ D . This yields s ∈ D0 ∩ B and s∗ as = 1. 

Corollary 2.2.7. The algebra O∞ is simple and purely infinite.

If C is a unital C*-algebra with properly infinite unit element and B is any C*-
algebra, then B ⊗max C ⊗max C ⊗max · · · and its quotient B ⊗min C ⊗min C ⊗min · · ·
are purely infinite.
In particular, B ⊗ O∞ (and thus B ⊗ O∞ ⊗ O∞ ⊗ · · · ) are simple and purely
infinite if B is simple.

We remind our convention that ⊗ always denotes the minimal (= spatial) C *-

algebra tensor product ⊗min .
Check consistency of notations:
⊗ , ⊗min , ⊗, ⊗max , ⊗max , ⊗
min b
Where used? For what?
Other tensor products will be denoted by (= algebraic tensor product) and
⊗max (= maximal tensor product = universal in both variables short exact tensor
product functor). Notice that there exist (at least) continuously may tensor prod-
uct functors on the category of separable C *-algebra with very different partial
injectivity or (short-) exactness properties, see Ozawa and Pisier [602] if combined
with considerations in [431].

Proof. Check and sort proof again or move it or parts of it to the appendices.
Give here exact ref’s to En in Appendices !!! ??
176 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

In proof of property (sq) for En in Chp. 4

is already said something about the ‘‘algebraic’’ version of En .
Give ref’s to all places with En and On discussion.
The C *-algebra En is the universal C *-algebra generated by n isometries with
orthogonal ranges, i.e.,

En := C ∗ (s1 , . . . , sn ; s∗k s` = δk` 1) ,

and the C *-algebra O∞ is the universal unital C *-algebra generated by a (count-

able) sequence of isometries with mutually orthogonal ranges:

O∞ := C ∗ (s1 , s2 , . . . ; s∗k s` = δk` 1 , k, ` = 1, 2, . . .) .

It implies immediately that the C *-algebra O∞ is isomorphic to the inductive limit

of the sequence of the natural unital C *-morphisms hn : En → En+1 defined by
hn (sk ) := sk for k = 1, . . . , n.
We are going to show that the C *-morphisms hn are injective and satisfy the
condition (*) of Proposition 2.2.5(iii). It implies that O∞ is simple and purely
infinite.
Let Tq (w) := sk1 sk2 · . . . · skq for the “word” w = k1 k2 . . . kq of length q ∈ N
with k1 , k2 , . . . , kq ∈ {1, 2, . . . , n}.
Denote by An the (algebraic) linear span of elements 1, Tq (w), Tq (w)∗ and
Tp (v)Tq (w)∗ with p, q ∈ N, v a word of length = p and w a word of length = q.
Calculation shows that An is a *-subalgebra of En . Since An contains the canonical
generators of En the vector space An is a dense *-subalgebra of En .
Let pn := 1 − (s1 s∗1 + . . . + sn s∗n ). Then pn is a projection in En that satisfies
pn Tq (w) = 0 for “words” w in the “alphabet” {1, . . . , n} if w is not the “empty”
word (i.e., if the “lengths” q of w is not zero).
It implies that pn An pn = C · pn and – therefore – that pn · En · pn = C · pn , i.e.,
pn is a minimal projection in En that generates a closed ideal J(pn ) of En that is
isomorphic to an elementary simple C *-algebra, i.e., J(pn ) ∼= K(`2 (N)) .
Clearly, Tq (w)pn 6= 0 because Tq (w)∗ Tq (w) = 1. But pn Tr (v)∗ Tq (w)pn = 0 for
w ∈ {1, . . . , n}q and v ∈ {1, . . . , n}r if r 6= q, or if r = q and v 6= w.
This shows that J(pn ) ∼
= K(`2 (N)), i.e., the closed ideal J(pn ) of the C *-algebra
En generated by pn := 1 − (s1 s∗1 + . . . + sn s∗n ) is isomorphic to K(`2 (N)), and pn is
a minimal projection of J(pn ) ∼
= K.
The results of J. Cuntz in [169, 172] show that the ideal J(pn ) is an essential
ideal of En and that On ∼
= En /J(pn ) is simple.
We explore her a different method to show that J(pn ) is an essential ideal of
En by using the “gauge action” action of S 1 on En defined by γ(z)(sn ) := zsn on
En (calculated modulo J(pn ) in On ) for all z ∈ S 1 ⊆ C
But it shows not that On := En /J(pn ) is simple, instead only the – here
sufficient – result that On contains no nontrivial closed ideal that is invariant under
 2. CHARACTERIZATIONS OF SIMPLE PURELY INFINITE ALGEBRAS 177

the “gauge action” of S 1 defined by γ(z)(sn ) := zsn on En (calculated modulo

J(pn ) in On ) for all z ∈ S 1 ⊆ C and, of course, ( 24 ).
( 25 ).
It satisfies γ(z)(en ) = en for all z ∈ S 1 . Therefore, the closed ideal J(en ) of En
generated by pn is invariant under the action of the gauge group. To distinguish
between the gauge action on En and On we denote by [γ](z) the action on On .
The projection pn ∈ En is fixed by the natural/canonical circle action given by
γ(z)(sk ) := z · sk on En for z ∈ C with |z| = 1. This action is called “gauge action”
and is well-defined by the universality of the defining relations for En . Clearly,
the ideal J(pn ) generated by pn is invariant under the circle action γ(z),because
γ(z)(pn ) = pn for z ∈ S 1 ⊆ C. It follows immediately that also the set K of
all elements b ∈ En with b · J(pn ) = {0} is a closed ideal of En that is invariant
under the gauge action, i.e., that γ(S 1 )(K) ⊆ K. This annihilator K of the ideal
J(pn ) is a closed ideal K of En and must be γ(S 1 )-invariant – by γ(S 1 )-invariance
of J(pn ). Therefore πJ(pn ) (K) must be a closed ideal of On that is invariant under
the gauge action [γ] (modulo J(pn )) on On . Moreover, πJ(pn ) |K is faithful on K
and is equivariant with respect to the gauge actions γ on En and [γ] on On .
The fix-point algebra of [γ] on On is isomorphic to the UHF-algebra Mn∞ as
a study of the algebraic dense subalgebra πJ(pn ) (An ) of On shows. In particular it
contains no non-trivial ideals. It allows to apply the following general elementary
observation: Each continuous action of a compact group G on a C *-algebra B has
a fix-point algebra that contains an approximate unit of B, this happens also for
every G-invariant closed ideal I of B, and the G-fix-point algebra of I is a closed
ideal of the fix-point subalgebra of G in B.
If we apply this to I := πJ(pn ) (K) and the circle action [γ] : S 1 → Aut(On )
then we get that the intersection of I with the fixed point-algebra of [γ] is an ideal
of the fix-point algebra of [γ] that contains a strictly positive element of πJ(pn ) (K).
Straight forward calculation shows that the fix-point algebra of the gauge action
on On is isomorphic to the simple UHF algebra Mn∞ . It proves that the ideal
πJ(pn )(K) ∼ = K of On can only be zero or all of On , i.e., K = {0} or 1 ∈ πJ(pn )(K).
Suppose that 1 ∈ πJ(pn ) (K). The orthogonality K · J(pn ) = 0 implies that
the quotient map πJ(pn ) is faithful on K. Hence, K contains a projection Q ∈ K
with πJ(pn ) (Q) = 1, and it follows that Q · J(pn ) ⊆ K · J(pn ) = {0} and that

24 This arguments are elementary but do not prove the simplicity of O .

n
25
It proves the simplicity of On not completely. A complete proof of the simplicity and
nuclearity of On can be obtained by showing that the canonical conditional expectation onto
the fix-point algebra =∼ Mn∞ of the gauge action is an approximately inner c.p. map on On .
Equivalently, by nuclearity and simplicity of Mn∞ , that the natural conditional expectation Pn
is “faithful” as c.p. map and must preserve all closed ideals, i.e., for each positive a ∈ On the
element Pn (a) ∈ Mn∞ ⊂ On is contained in the ideal generated by by a. This can be seen by
approximating a by elementary elements b and then show that Pn (b) can be approximated up to
ε > 0 by X ∗ bX for a suitable elementary element X. It shows that also that all positive elements
of On are purely infinite.
178 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

πJ(pn ) (1 − Q) = 0, i.e., 1 − Q ∈ J(pn ). Since Q is a projection with Qx = 0 = xQ

for all x ∈ J(pn ). It follow that P := 1 − Q is a unit element for J(pn ). But we
have seen above that J(pn ) ∼ = K(`2 (N)) is not unital. This shows that only the
case K = {0} remains, and this implies that J(pn ) is an essential ideal of En (as
discussed above).
It gives us the here needed injectivity of the natural map from En into En+1
????
The [γ]-simplicity and nuclearity of On follows from the fact that Mn∞ is the
fixed point algebra of the circle action (and that S 1 is a compact group), and that
the by this action define conditional expectation Pn from On onto Mn∞ defined by
the integral over this action is an approximately inner u.c.p. map. There exists an
explicit approximation by inner c.p. maps. It is a 1-step innerness that proves also
the pure infiniteness if the elements of On . It can be seen by considering “algebraic”
elements.
But we do not us it this here. See ... ... give Citation ???? to Appendix
Since, as above shown, a ∈ En 7→ L(a) ∈ M(J(pn )) with L(a)b = ab for
all b ∈ J(pn ) is a faithful unital C *-morphism, we obtain for a ∈ (En )+ with
kak = 1 that there exists a partial isometry z ∈ J(pn ) ∼
= K with z ∗ z = pn and
∗
kpn − z azk < ε .
We define c := hn (z)sn+1 hn (b1/2 ) ∈ En+1 . Notice that hn (pn ) = pn+1 +
sn+1 s∗n+1 ∈ En+1 and pn+1 sn+1 = 0 . This shows that

kc∗ hn (a)c − hn (b)k < ε .

It means that the canonical morphisms hn : En → En+1 satisfies the criterium (*)
of Proposition 2.2.5(iii). Thus O∞ = indlimn→∞ (hn : En → En+1 ) is simple and
purely infinite.
Part (iv) of Proposition 2.2.5 implies that B ⊗ C ⊗ C ⊗ · · · is simple and purely
infinite if B is simple and C is a simple C *-algebra with properly infinite unit
element 1C , because then 1C ∈ O∞ ⊆ E2 ⊆ C (with *-monomorphism defined by
O∞ 3 sn 7→ tn−12 t1 ∈ E2 for O∞ = C ∗ (s1 , s2 , . . .) and E2 = C ∗ (t1 , t2 ) ⊆ C.
For each a ∈ B ⊗ C ⊗ C ⊗ · · · and ε > there exists a0 ∈ B ⊗ C ⊗ · · · ⊗
C ⊗ 1C ⊗ 1C ⊗ · · · with ka0 − ak < ε/2 and a0 commutes with a copy of O∞ ∼=
1B ⊗ 1C ⊗ · · · ⊗ 1c ⊗ O∞ ⊗ 1C ⊗ · · · .
The proof shows also for every (not necessarily simple) C *-algebra B that
B ⊗C ⊗C ⊗· · · (with ⊗ := ⊗min ) is strongly purely infinite if C is a unital properly
infinite C *-algebra and the argument works also for finitely many a1 , . . . , an ∈
B ⊗ C ⊗ C ⊗ ···.
Can also use dichotomy for tensor prod.s of simple C *-alg.s
if B and C are simple !!!
Then it is only needed that both are non-elementary and that one of
B or C is stably infinite. 
 2. CHARACTERIZATIONS OF SIMPLE PURELY INFINITE ALGEBRAS 179

Lemma 2.2.8. Let ε ∈ (0, 1/4) and a, b ∈ A+ contractions that satisfy

kab1/2 − b1/2 ak < ε and kabak > 1 − ε .

Then kb1/2 a2 b1/2 − abak < 2ε and the non-zero hereditary C*-subalgebra E :=
abaAaba of aAa has the property that
√
k(1 − b)1/2 dk ≤ 3ε · kdk for all d ∈ E . (2.3)

Notice here for later applications that (1 − b)1/2 is a (positive) contraction.

Proof. Suppose that e ∈ A+ is a contraction with 3/4 < 1 − ε < kek ≤ 1 ,

and let f := (e − (1 − ε))+ . Then 0 ≤ f 6= 0, kf k ≤ ε and E := f Af is a non-zero
hereditary C *-subalgebra of A. We use that

(1 + 2ε − t)(t − 1 + ε)1/n ≤ 3 · ε for all t ∈ [1 − ε, 1] and n ∈ N ,

to obtain that k(1 + 2ε − e)f 1/n k ≤ 3ε for all n ∈ N . Since the f 1/n commute with
the positive element (1 + 2ε − e) and f is a strictly positive element in E, we get
that kd∗ (1 + 2ε − e)dk ≤ 3ε · kdk2 for all d ∈ E. It shows that
√
k(1 + 2ε − e)1/2 dk ≤ 3εkdk for all d ∈ E . (2.4)

We apply this observation to our special case:

Let a, b ∈ A+ elements that satisfy the in Lemma 2.2.8 listed pre-assumptions
with 0 < ε < 1/4, and let x := b1/2 a. Then kxk ≤ 1 and kx∗ − xk = kab1/2 −
b1/2 ak < ε. It gives

kb1/2 a2 b1/2 − abak = kxx∗ − x∗ xk < 2ε .

Thus, b1/2 a2 b1/2 − aba ≤ 2ε · 1 . Together with 0 ≤ a2 ≤ 1 it implies

0 ≥ 1 − b ≤ 1 − b1/2 a2 b1/2 ≤ 1 + 2ε − aba .

Now let e := aba and f := (e − (1 − ε))+ build from a and b with the
listed properties. This e, f ∈ A+ have the previously considered properties 3/4 <
1 − ε < kek ≤ 1 , because a, b ∈ A+ are contractions that satisfy kabak > 1 − ε by
assumptions on a, b and ε .
The monotony of the C *-norm on positive parts of C *-algebras implies that
for all elements d ∈ E := f Af holds

kd∗ (1 − b)dk ≤ kd∗ (1 + 2ε − aba)dk .

The above shown Inequality 2.4 says that kd∗ (1 + 2ε − e)dk ≤ 3ε · kdk2 for all
d ∈ E := f Af , where we let here e := aba and f := (aba − (1 − ε))+ .
Using again monotony of the C *-norm on A+ we get that

kd∗ (1 − b)dk ≤ kd∗ (1 + 2ε − aba)dk ≤ 3ε · kdk2 ,

for all d ∈ E := f Af . This is equal to the Inequality (2.4). 

180 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Lemma 2.2.9. Let A a non-zero σ-unital C*-algebra, and S ∈ M(A)+ a positive

contraction with norm kπA (S)k = 1 for the image πA (S) of S in the corona Q(A) :=
M(A)/A of A.

(i) There exists a positive contraction T ∈ M(A)+ with kπA (T )k = 1 and

(1 − S)T ∈ A.
(ii) If A is antiliminary in the sense of [616, sec. 6.6.1], cf. our Definition
2.7.2, i.e., if a∗ Aa is non-commutative for each non-zero a ∈ A, then
there exists a contraction T ∈ M(A) with T 2 = 0, kπA (T )k = 1 and
(1 − S)T, T (1 − S) ∈ A.
(iii) If A is simple and non-elementary then for each a ∈ A+ with kak = 2 there
exists a sequence of contractions d1 , d2 , . . . ∈ A such that (a−(a−1)+ )dn =
dn , kdn k = 1, d∗m dn = 0 and dn (dm )∗ = 0 for m, n ∈ N, and such
P ∗
that the sum n dn dn is strictly convergent in M(A) to a contraction
P ∗
T := n dn dn ∈ M(A)+ that satisfies kπA (T )k = 1 and (1 − S)T ∈ A.
(26).

Proof. The corona algebra Q(A) := M(A)/A is non-zero (and therefore is

unital) by existence of a contraction S ∈ M(A)+ with kπA (S)k = 1.
(i) and preparations for (ii): Let e ∈ A+ with kek = 1 strictly positive in A, i.e.,
eAe = A. For every separable C *-subalgebra B ⊆ M(A), e.g. for B := C ∗ (e, S)
with the considered contraction S ∈ M(A)+ , we find a linear filtration Xn ⊆ B
with S, e ∈ X1 , dim(Xn ) < ∞ and Xn∗ = Xn ⊆ Xn+1 – as considered in Remark
5.1.1(3) of Chapter 5 – and an approximate quasi-central unit (en )n∈N of form
en := fn (e) with continuous functions fn ∈ C0 (0, 1] with f0 (t) := t, fn fn+1 = fn
for n ≥ 1, kf0 − fn f0 k → 0, and, for b ∈ Xn ,

k [fn (e), b] k + k[(fm (e) − fn (e))1/2 , b] k < 4−n kbk .

Then, for every sub-selection gk := fnk (e), nk+1 > nk ≥ k > 0 (and g0 := 0), we
can build the completely positive unital map V : `∞ (M(A)) → M(A) with help of
the strictly convergence of sums
∞
X
V (a1 , a2 , . . .) := (gk − gk−1 )1/2 ak (gk − gk−1 )1/2
k=1

for bounded sequences a1 , a2 , . . . in M(A). Notice that b − V (b, b, . . .) ∈ A for all

b ∈ B and that V (ba1 , ba2 , . . .) − bV (a1 , a2 , . . .) ∈ A for each sequence (a1 , a2 , . . .) ∈
`∞ (M(A)) and every b ∈ B. This is the case because V is a unital completely
positive map from `∞ (M(A)) into M(A) with V (b, b, . . .) = b for all b ∈ B ⊆ M(A),
i.e., the images (b, b, . . .) by the diagonal embedding of b ∈ B in `∞ (M(A)) are
contained in the multiplicative domain of V .
(See Remarks 5.1.1(2,...,5) for more details on this standard constructions.)
If c1 , c2 , . . . ∈ A is a bounded sequence that satisfies the condition ck =
ck (gk − gk−1 ) = (gk − gk−1 )ck (e.g. if ck gk = ck and ck gk−1 = 0) for k ∈ N then

26 But the sum d + d + · · · is in general not strictly convergent in M(A).

1 2
 2. CHARACTERIZATIONS OF SIMPLE PURELY INFINITE ALGEBRAS 181

kπA (V (c1 , c2 , . . .))k = lim supn kcn k , ck ck+1 = 0 = c∗k ck+1 and the series
P
k ck is
P
strictly convergent in M(A) and k ck = V (c1 , c2 , . . .).
If moreover limk kck S−ck k = 0, then V (c1 , c2 , . . .)S−V (c1 , c2 , . . .) ∈ A, because
of V (a1 , a2 , . . .) ∈ A if lim kan k = 0.
It follows T − T S ∈ A and kπA (T )k = 1 for the contraction T := V (c1 , c2 , . . .)
with c1 , c2 , . . . ∈ A contractions that satisfy lim supn kcn k = 1 and limn kcn S−cn k =
0 and cn (gn − gn−1 ) = cn = (gn − gn−1 )cn . Similarly also T − ST ∈ A if here in
addition the contractions cn satisfy lim kcn − Scn k = 0.
If, moreover, those cn can be fund such that c2n = 0 in addition, then T 2 = 0,
i.e., in the hereditary C *-subalgebra Ann(1 − πA (S)) of elements X ∈ Q(A) with
X(1 − πA (S)) = 0 and X ∗ (1 − πA (S)) = 0 is not Abelian.
Notice that Ann(1 − πA (S)) is automatically contained in πA (S) Q(A)πA (S) ⊆
D if S ∈ M(A)+ is a positive contraction with πA (S) ∈ D.
The above outlined method requires a careful inductive choice of the gk :=
enk := fnk (e) for suitable n1 , . . . , nk , . . . and then of non-zero hereditary C *-
subalgebras Ek with the property that for all d ∈ Ek holds d(gk+1 − gk ) = d
and kdS − dk < 1/2k .
(Those nk and Ek can be selected suitably with help of Lemma 2.2.8.)
Then all elements X = πA (V (d1 , d2 , . . .)) are in Ann(1 − πA (S)) if dk ∈ Ek
for k = 1, 2, . . . and lim supn kdk k = kXk. If d2k = 0 for all k ∈ N it follows that
X 2 = 0.
Compare below text with above. Give precise selection rule.
Let S ∈ M(A)+ with kSk = kπA (S)k = 1 and B := C ∗ (S, e). Choose en :=
fn (e) for B as above described for n = 1, 2, . . ..
We select gk := enk := fnk (e) such that there exists a non-zero hereditary
C *-subalgebra Ek of A with kd − Sdk < 2−n and d(gk − gk−1 ) = d for all d ∈ Ek .
Q
Then V (d1 , d2 , . . .) defines a C *-morphism from k Ek := `∞ (E1 , E2 , . . .) into
M(A) with V (c0 (E1 , E2 , . . .)) ⊆ A, and (1 − S)V (d1 , d2 , . . .), V (d1 , d2 , . . .)(1 − S) ∈
A for all (d1 , d2 , . . .) ∈ `∞ (E1 , E2 , . . .).
We find m1 ∈ N with kem1 Sem1 k ≥ 1 − 2−2 and k[S, em1 ]k < 2−2 .
by Lemma 2.2.8 ??
Let E1 the hereditary C *-algebra generated by ((em1 Sem1 ) − 1/2)+ . Then
em1 +1 d = d for all d ∈ E1 , and kd − Sdk ≤?????.
Take ????
???? with the above listed properties such that kck k = 1 in addition. Then
T := V (c1 , c2 , . . .) ∈ M(A) has the desired property T − ST ∈ A.
Define contractions c1 , c2 , . . . ∈ A+ and nk ∈ N step-wise:
182 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Recall that k(en − em )Sk ≥????. Take g1 := e2 , i.e., k1 := 2, and find in the
hereditary C *-algebra generated by e1 Se1 let c1 := e1 = e1 e2 for Part (i).
Find for Part (ii) in the hereditary C *-subalgebra E1 generated by e1 Se1 an
element with d21 = 0 and kd1 k = 1 if E1 6= {0}
and ????.
Then kc1 k = 1 and c1 (e3 − e0 ) = c1 .
Then k(1 − e4 )S(1 − e4 )k = 1 in B ⊂ M(A) and we find nk ∈ N, nk+1 ≥ 2 + nk
with k(enk+1 −1 − enk +1 )1/2 S 1/2 k > 1 − 2−nk . It follows that in the closed right ideal
Lk generated by (enk+1 −1 − e1+nk )1/2 S 1/2 A contains a positive contraction ck with
kck Sk ≥ 1 − 2−k1 .
The c1 , c2 , . . . and g1 , g2 , . . . have the desired properties.
(ii): The pairwise orthogonal non-zero hereditary C *-subalgebras Ek consid-
ered in the proof Part (i) contain elements ck ∈ Ek with c2k = 0, kck k = 1.
2
P
Then T = k ck = V (c1 , c2 , . . .) satisfies T = 0, kT k = kπA (T )k = 1 and
T S − S, ST − S ∈ A.
(iii): With the hereditary C *-subalgebras Ek considered in Parts (i,ii) we need
only to find in the hereditary C *-subalgebra F := (a − 1)+ A(a − 1)+ a sequence of
elements a1 , a2 , . . . ∈ F+ with kan k = 1, am an = 0 for m 6= n. Then the simplicity
of A allows to find d1 , d2 , . . . ∈ A with d∗n dn ∈ En , dn (dn )∗ ∈ an Aan and kdn k = 1.
This is possible for simple non-elementary A. 

Lemma 2.2.10. Let A a σ-unital C*-algebra and let D ⊆ Q(A) a non-zero

simple hereditary C*-subalgebra of Q(A). Then D has the following properties:

(i) D is purely infinite or is “elementary” ( 27 ).

(ii) If A is antiliminary, in the sense of Definition 2.7.2, then D is non-
elementary, i.e., D can not contain a non-zero minimal projection.
(iii) The corona Q(A) of A is purely infinite if Q(A) is simple (and non-zero).
If Q(A) is simple then, for each S ∈ M(A)+ with kπA (S)k = 1, there
exists a contraction d ∈ M(A) with 1 − d∗ Sd ∈ A.

Proof. (i): The non-zero simple C *-subalgebra D ⊆ Q(A) satisfies the con-
dition (xv) of Proposition 2.2.1 if D is non-elementary:
Let a, b ∈ D+ with kak = kbk = 1 and ε > 0. There exist R, S ∈ M(A)+ with
πA (R) = b, πA (S) = a, kRk = 1 and kSk = 1.
By Lemma 2.2.9(i) there exists T ∈ M(A)+ with kT k = kπA (T )k = 1 and
T S − T ∈ A.
Let g := πA (T ). We get g = aga ∈ D , because g = ga = ag and D is
hereditary in M(A). Since D is simple and g ∈ D+ has norm kgk = 1 we find for
every ε > 0 a number ν ∈ N such that there exists contractions d1 , . . . , dν ∈ D with
27 The simple D is “elementary”, if and only if, D contains a non-zero projection p ∈ D with

p Q(A)p = C · p . Then D ∼
= K(H) for some Hilbert space H .
 2. CHARACTERIZATIONS OF SIMPLE PURELY INFINITE ALGEBRAS 183

Pν
kb − d∗j g 2 dj k < ε. Define ν(b, a, ε) as this ν ∈ N. Then kb − j=1 cj ak cj k < ε
P

with contractions cj := gdj for each k ∈ N.

Thus, D is purely infinite by Proposition 2.2.1(xv).
(ii): Let p ∈ D ⊆ Q(A) a non-zero projection. There exists a positive contrac-
tion S ∈ M(A)+ with πA (S) = p.
By Lemma 2.2.9(ii) there exists T ∈ M(A) such that b := πA (T ) ∈ Q(A)
satisfies kbk = 1, b2 = 0 and pb = b = bp. Thus b ∈ p Q(A)p ⊆ D and b 6∈ C · p.
(iii): Suppose that 0 6= p is a projection in Q(A) with p Q(A)p = C · p and that
Q(A) is simple. Then there exists d1 , . . . , dn ∈ Q(A) with k d∗k pdk = 1. It follows
P

that Q(A) ∼ = Mm (C) for some m ≤ n.

The following argument shows that σ-unital A is unital (i.e., Q(A) = {0}) if
every Abelian C *-subalgebra of Q(A) is finite dimensional.
If a ∈ A+ is a strictly positive contraction then M(C ∗ (a))/C ∗ (a) ⊆ Q(A) by a
natural unital *-monomorphism.
Let X ⊆ (0, 1] denote the non-zero values in the spectrum of a in C ∗ (a) ∼=
C0 (X). If 0 is in the closure of X, then Cb (X)/ C0 (X) infinite dimensional (as
vector space) because then `∞ (N)/c0 (N) is a quotient of it.
Thus, A is unital if Q(A) has only finite dimensional abelian C *-subalgebras
(which is equal to Q(A) finite-dimensional). 

Corollary 2.2.11. Let A a non-zero C*-algebra, and K := K(`2 (N)).

(i) The stable corona Qs (A) := M(A ⊗ K)/(A ⊗ K) is simple, if and only if,
(a) A is σ-unital and simple, and
(b) A is either purely infinite or is “elementary” (i.e., A ⊗ K ∼
= K).
s s
If Q (A) is simple then Q (A) is purely infinite.
(ii) The ultrapower Aω is simple, if and only if, either A is simple and purely
infinite, or A ∼
= Mn for some n ∈ N.
(iii) The hereditary C*-subalgebra of Aω generated by A ⊆ Aω is simple, if and
only if, either A is simple and purely infinite or A is isomorphic to the
compact operators K(H) on some Hilbert space H.

Remarks 2.2.12. To be re-read again and sorted !!!

(1) There is an interesting observation of the author that is related to Part (iii)
of Corollary 2.2.11, but with a technically much more involved proof (cf. [448, thm.
2.12]).
HERE check "cite" (of article and of thm.) of new version !!!
It says:
If A is separable, then the algebra F (A) := (A0 ∩ Aω )/ Ann(A, Aω ) is simple,
if and only if, A is nuclear (sic !) and simple and, either A is purely infinite, or
A⊗K∼ = K. The isomorphism A ⊗ K ∼ = K is equivalent to F (A) ∼
= C, because only
A = Mn or A = K are isomorphic to hereditary C*-subalgebras of K.
184 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Here ω ∈ β(N) \ N, is a “free” ultrafilter on N, Aω := `∞ (A)/Jω is the (ω-)

ultrapower of A with

Jω := {(a1 , a2 , . . .) ∈ `∞ (A) ; lim kan k = 0 } ,

n→ω

and Ann(A, Aω ) is the two-sided annihilator of A in Aω , that is obviously an ideal

in A0 ∩ Aω .
(Hint: The main reason for this result is that the elements d := πω (d1 , d2 , . . .) ∈
Aω with the property that there exists a bounded sequence of nuclear maps Vn : A →
A such that ω − lim kd∗n adn − Vn (a)k = 0 for all a ∈ A, build a non-zero closed right
ideal of Aω . It is a left module for A0 ∩ Aω , is not contained in Ann(A, Aω ) and it
contains an element in A0 ∩ Aω that is not in the two-sided annihilator Ann(A, Aω ).
Now the reader can start here/his exercise by combining this with (ii).)
No other single property of separable non-elementary C *-algebras characterizes
the pi-sun C *-algebras by only one condition!
(2) ????
Compare next ‘‘blue’’ with further below
Part(i) of Corollary 2.2.11 was proved by M. Rørdam [674, thm. 3.2] for the
special case of non-elementary unital A.
(2??) The Parts (i,ii) of Corollary 2.2.11 has been proved for unital A by M.
Rørdam ([674, thm. 3.2], [681]).
Then H. Lin [517, thm. 3.8] studied the case of non-elementary σ-unital A and
observed that A must be algebraically simple. It implies that A ⊗ K contains a
scaling element.
The latter is equivalent by [78, thm. 1.2] to the existence of a projection in
A ⊗ K . This proves the general result by reduction to the unital case considered in
[674, thm. 3.2] by M. Rørdam.
Notice that we use a different and more elementary idea in the proof of Part
(i).
Part (i) of Corollary 2.2.11 has also been obtained by S. Zhang [842] under the
additional pre-assumptions that A is σ-unital and has real rank zero. Then H. Lin
[517, thm. 3.8] extended the result [674, thm. 3.2] of M. Rørdam by replacing the
requirement that A is unital by the requirement that A σ-unital.
The new argument of
S. Zhang ... or was it Lin? ...
is that stable σ-unital simple A with simple M(A)/A contains a “scaling”
element in the sense of B. Blackadar and J. Cuntz [78].
They show in [78] that it allows to define a non-zero projection in A, cf. proof of
the implication (iii)⇒(ii) for Proposition 2.2.1 for a similar argument. This allows
to apply [674, thm. 3.2] also for the σ-unital case without the extra assumption
that A is unital.
 2. CHARACTERIZATIONS OF SIMPLE PURELY INFINITE ALGEBRAS 185

Later H. Lin proved in [526] that, if A is simple, σ-unital and non-unital then
simplicity of M(A)/A is equivalent to M(A)/A being purely infinite and simple.
But he contributed also examples of
separable ??? or non-separable?? non-stable
simple non-unital C *-algebras A with M(A)/A simple but where A is not
purely infinite (but is then at least “algebraically simple”).
It follows that a non-stable version of Part (i) of Corollary 2.2.11(i) can not
exist!
Our proofs of Propositions 2.2.1 and 2.2.5 and Corollary 2.2.11 are more gen-
eral and elementary in a sense, because Remark 2.2.6, Lemma 2.2.3 and the used
technical Lemmata A.21.3 and 2.1.22 and the passage to the non-separable case in
Proposition A.21.4 are elementary basic observations in C *-algebra theory (that
we here often apply).
Our Corollary 5.7.3 characterizes the nuclear simple unital p.i. C *-algebras
A 6= C by the property that for each separable unital C *-subalgebra C ⊆ M(A⊗K)
and each unital C *-morphism

h : C → M(A ⊗ K) with h(C ∩ (A ⊗ K)) = {0}

there exists a unitary U ∈ M(A ⊗ K) such that

c ⊕ h(c) − U ∗ c U ∈ A ⊗ K for all c ∈ C .

The proof uses Proposition 2.2.5(vi) and a Weyl–von-Neumann type theorem, that
is related to an absorption result of G. Elliott and D. Kucerovsky, cf. [264] and a
correction (of what?) observed by J. Gabe [309].
Recall that the Calkin algebra C(`2 ) = Q(K) = L(`2 )/K is simple and purely
infinite.
Proposition 2.2.5(iii) shows that any unital simple C*-algebra A that contains
an approximately central sequence of copies of O∞ is purely infinite.
We can replace O∞ in the statement of Proposition 2.2.5(iii) by E2 :=
C (s, t : s∗ s = t∗ t = 1, s∗ t = 0) or any C *-algebra with properly infinite unit
∗

because
E2 ⊆ O∞ ⊆ E2 ⊆ On
(via suitable unital monomorphisms) for every n > 1. See Chapter 10 on related
topics.
(3??) F (L(`2 )/K(`2 )) ∼
= C. Mention here also the result of Farah and


Phillips on F L(`2 (N)) . It is simple if and only if one uses Axiom of Choice?

Proof of Corollary 2.2.11. (i): Let Qs (A) := M(A ⊗ K)/(A ⊗ K), and
a ∈ A+ with kak = 1.
If the unital C *-algebra Qs (A) is simple then there exist c1 , . . . , cn ∈ M(A⊗K)
with c∗1 (a ⊗ 1)c1 + . . . + c∗n (a ⊗ 1)cn = 1 + d for some d ∈ A ⊗ K, – here 1 ∈ M(K) ∼
=
186 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

L(`2 (N)), by using that M(A) ⊗ M(K) is naturally contained in M(A ⊗ K) and
identifying M(K) with 1M (A) ⊗ M(K) naturally. We find a projection pn ∈ K
with k(1 ⊗ pn )d(1 ⊗ pn ) − dk < 1/2 . Thus, there is an isometry s ∈ M(K) with
k(1 ⊗ s)∗ d(1 ⊗ s)k < 1/2 . If we let gk := ck (1 ⊗ s), then the positive element
T := g1∗ (a ⊗ 1)g1 + . . . + gn∗ (a ⊗ 1)gn satisfies k1 − T k < 1/2 in M(A ⊗ K),
and the elements e1 , . . . , en ∈ M(A ⊗ K) given by ek := gk T −1/2 satisfy with
P ∗
k ek (a ⊗ 1)ek = 1 .

It implies that fn := k e∗k (a ⊗ pn )ek ∈ A ⊗ K is a countable approximate unit

P

of A ⊗ K if pn is an approximate unit of K (consisting of projections). In particular,

each non-zero element a ∈ A+ is not contained in a non-trivial ideal of A. It proves
that A is σ-unital and simple if Qs (A) is simple.
In case where A is the algebra of compact operators on a Hilbert space H, it
implies that H must be separable. Hence A ⊗ K ∼ = K, if A is isomorphic to the
compact operators on a Hilbert space, i.e., A = Mn or A ∼
∼ = K(`2 (N)).
It remains to consider the case where A is not isomorphic to the compact
operators on a Hilbert space.
We show that the non-elementary simple stable σ-unital C *-algebra B := A⊗K
satisfies in this case the criteria in Part (vx) of Proposition 2.2.1 for the pure
infiniteness of B (thus also of A) if A is non-elementary:
Let a, b ∈ B+ with kak = kbk = 1 and ε ∈ (0, 1/2), and define e := (a−(1−ε))+
and c := a − e.
Since B is again non-elementary, the non-zero hereditary C *-subalgebra
D := eBe of B contains a sequence of mutually orthogonal positive contractions
f1 , f2 , . . . ∈ D+ with kfn k = 1. Notice that dc = cd = (1 − ε)d for all d ∈ D.
Let s1 , s2 , . . . ∈ M(K) ⊂ M(B) a sequence of isometries with the property that
∗ 2 ∗
P P
n sn sn converges strictly to 1M(K) = 1M(B) . Then k sn fn sn converges strictly
to an element S ∈ M(B) with norms kSk = kπB (S)k = 1.
By Lemma 2.2.10(i) the simplicity of M(B)/B implies that M(B)/B is purely
infinite. Thus, there exists a contraction Z ∈ M(B) with 1 − Z ∗ SZ ∈ B by using
Parts (ii) or(vi) of Proposition 2.2.1.
It follows that there exist m ∈ N with k1 − s∗m Z ∗ SZsm k < 1/2 and 1 −
s∗m Z ∗ SZsm ∈ B. It implies there exists G ∈ M(B) with G∗ SG = 1 . Then
b = y ∗ Sy with y := Gb1/2 ∈ B. The strict convergence of k sn fn2 s∗n to S implies
P

that
p
X
b = lim y ∗ sn fn2 s∗n y
p
n=1

in the norm of B. It means that b is the limit of the increasing sequence

Pp
x∗1 x1 , x∗2 x2 , . . . in B, where xp := ∗
n=1 fn sn y ∈ B, and where we use that
fn fm = 0 for m 6= n. (The sequence x1 , x2 , . . . itself is not necessarily convergent
in B.)
 3. PROOF OF THE DICHOTOMY THEOREM E FOR SIMPLE C *-ALGEBRAS 187

Now we use that fn ∈ D and get that cxp = (1 − ε)xp , x∗p cxp = (1 − ε)x∗p xp and
c ≤ a ≤ 1. Thus, (1 − ε)x∗p xp ≤ x∗p axp ≤ x∗p xp , and kb − x∗p axp k ≤ kb − x∗p xp k + ε.
Since b = limp x∗p xp , we get that limp kxp k = kbk1/2 . Thus, B is simple and is
purely infinite if B is not elementary.
(ii,iii): If J is a non-zero closed ideal of A, then Jω is a non-zero closed ideal
of Aω that has non-trivial intersection with the hereditary C *-subalgebra D :=
D(A) := A · Aω · A of Aω generated by A. Thus J = A and A is simple if D is
simple.
If A = Mn , then Aω = Mn .
If A ∼
= K(H) and if the Hilbert space H has infinite dimension, then D(A) is a
hereditary C *-subalgebra of the closed ideal of Aω that is generated by a rank-one
projection in A ∼= K(H) (in fact the closed ideal of Aω generated by A = K(H)
is naturally isomorphic to the compact operators K(Hω ) on Hω ). But the whole
ultrapower K(H)ω of the compact operators K(H) on H (with Dim(H) = ∞) is
not simple, because K(Hω ) 6= K(H)ω . The latter can be seen directly or by the
fact that K(H)ω contains the CAR-algebra as a subalgebra (and, therefore, is not
of type I).
If A is not isomorphic to the compact operators on a Hilbert space, and if
D(A) = A(Aω )A is simple, then A must satisfy the criteria in Proposition 2.2.5(iv):
The element that is represented by the constant sequence (b, b, . . .) ∈ A is in the
closed ideal generated by the element with representing sequence (a3 , a4 , a5 , . . .) ∈
A(Aω )A ⊆ Aω if a ∈ A+ , kak = 1 and there is c ∈ A+ with c a = a , kck ≤ 1,
compare Remark 2.2.6.
Conversely, if A is simple and purely infinite, then Proposition 2.2.1(ii) implies
that for positive contractions a, b ∈ Aω with kak = kbk = 1 there is a contraction
d ∈ Aω with d∗ ad = b. This can be seen with help of representing sequences
a = πω (a1 , a2 , . . .), b = πω (b1 , b2 , . . .) and d = πω (d1 , d2 , . . .), because one can find
the representing sequences such that kan k = kbn k = 1 for n = 1, 2, . . .. 

3. Proof of the Dichotomy Theorem E for simple C *-algebras

The proof of the in the Introduction 1 stated Theorem E uses some elementary
observations, and from our general study of p.i. algebras only that for simple C *-
algebras pure infiniteness and local pure infiniteness are the same.
Among the needed observations is that, for every n ∈ N, every non-elementary
simple C*-algebra A and every non-zero hereditary C*-subalgebra D of A, there
exists a non-zero n-homogenous element a ∈ D+ , cf. Remark 2.1.16.
We call a simple C *-algebra A non-elementary if A is simple (and non-zero)
and is not isomorphic to the algebra of compact operators on some Hilbert space
(of finite or infinite dimension). An element a ∈ D+ ⊆ A is n-homogenous if
there exists an isomorphism λ : C ∗ (a) ⊗ Mn → D with λ(a ⊗ 1n ) = a.
Notice that then aDa ∼
= D0 ⊗ Mn for the closure D0 of λ(a ⊗ e11 )Dλ(a ⊗ e11 ).
188 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

The following remark will be used in the proof of Theorem E. It is stated here
more general than needed in the proof of Theorem E, because the more general
statement will be used in Chapters 4, 5 and 7.

Remark 2.3.1. Let A ⊆ E C *-algebras, x ∈ A and v ∈ E with kvk ≤ 1,

xv = (xx∗ )1/2 and v ∗ x = (x∗ x)1/2 .
∗

If D1 , D2 ⊆ A denote the hereditary C*-subalgebras of A that are generated by

x x respectively by xx∗ , then ϕ(a) := vav ∗ defines an isomorphism ϕ from D1 onto
∗

D2 .
For example, if we let E := A∗∗ then such an element v ∈ A∗∗ is given by the
partial isometry v arising in the polar decomposition x = v(x∗ x)1/2 of x in A∗∗ .

Details for Remark 2.3.1. The mappings v(·)v ∗ and v ∗ (·)v are completely
positive contractions on E, and map D1 into D2 respectively D2 into D1 . Indeed:

vD1 v ∗ = v(x∗ Ax)v ∗ ⊆ vx∗ Axv ∗ = (xx∗ )1/2 A(xx∗ )1/2 = D2 ,

v ∗ D2 v = v ∗ (xAx∗ )v ⊆ v ∗ xAx∗ v = (x∗ x)1/2 A(x∗ x)1/2 = D1 .

The equations xv ∗ vx∗ = xx∗ and x∗ vv ∗ x = x∗ x imply, that ϕ(ab) = ϕ(a)ϕ(b), for
a, b ∈ x∗ Ax, and that xv ∗ v = x, vv ∗ x = x, i.e., that ϕ and v ∗ (.)v| D2 are inverse
to each other. 

Proof of Theorem E.. (i): By a result of M. Takesaki [767, cor. IV.4.21],

A ⊗ B is simple if A and B are simple, because the C *-tensor product ⊗ := ⊗min is
the completion of the algebraic tensor product A B with respect to the minimal
C *-seminorm N (·) on A B with the property that N (a ⊗ b) 6= 0 for all non-zero
a ∈ A+ and b ∈ B+ , [767, thm. IV.4.19]. It turns out that this norm is the same as
the “spatial” tensor product norm on A B ⊆ L(H1 ⊗2 H2 ) given by any faithful
representations A ⊆ L(H1 ) and B ⊆ L(H2 ), cf. [767, lem. IV.4.11].
It suffices to consider the case where A is not stably finite, because for all
C *-algebras A and B holds A ⊗ B ∼ = B ⊗ A.
If A is not stably finite then for every non-zero hereditary C *-subalgebra F
of A there is a positive integer n = n(F ) such that Mn (F ) contains an infinite
projection p ∈ Mn (F ). It follows that p(Mn (F ))p contains an isomorphic copy
of the Toeplitz algebra C ∗ (s ; s∗ s = 1) and, therefore, Mn (F ) contains a copy of
algebra of compact operators K ⊂ C ∗ (s ; s∗ s = 1), cf. [172].
Now let D ⊆ A ⊗ B be a non-zero hereditary C *-subalgebra of A ⊗ B . By
Lemma 2.2.3 and Remark 2.3.1, there exist non-zero hereditary C *-subalgebras
E ⊆ D, F ⊆ A and G ⊆ B such that E ∼ = (F ⊗ G). Here F , G and E denote
the hereditary C *-subalgebras that are generated by the elements e, f and zz ∗
of Lemma 2.2.3, and the natural isomorphism is given by the partial isometry v
in the second conjugate of A ⊗ B which appears in the polar decomposition of z,
cf. Remark 2.3.1.
We show that F ⊗ G contains a non-zero stable C *-subalgebra :
 4. ON ABSENCE OF INFINITESIMAL SEQUENCES 189

The algebra F ⊗ Mn ∼ = Mn (F ), for the number n := n(F ) as above defined,

contains a hereditary C *-subalgebra D1 that is stable and non-zero.
By assumption, the algebra B is non-elementary, i.e., B is simple and is not
isomorphic to the compact operators on some Hilbert space. Hence, B is necessarily
antiliminary, i.e., every non-zero hereditary C *-subalgebra G of B has no finite-
dimensional irreducible representations. This applies to our above defined G:
By Remark 2.1.16, the non-zero hereditary C *-subalgebra G of B contains
a non-zero hereditary C *-subalgebra G1 which is isomorphic to Mn ⊗ G2 for a
suitable non-zero hereditary C *-subalgebra G2 of G1 .
We get that

D1 ⊗ G2 ⊆ F ⊗ Mn ⊗ G2 ∼
= F ⊗ G1 ⊆ F ⊗ G ∼
= E ⊆ D,

and the C *-algebra D1 ⊗ G2 is non-zero and stable. Thus, the given non-zero
hereditary C *-subalgebra D ⊇ E contains a non-zero stable C *-subalgebra.
In conclusion, every non-zero hereditary C *-subalgebra D ⊆ A ⊗ B contains a
non-zero stable C *-subalgebra. By Proposition 2.2.1(iii), this is equivalent to the
pure infiniteness of the simple C *-algebra A ⊗ B .
(ii): The algebras A and B are simple and nuclear, because A⊗B is simple and
nuclear by assumption in Part (i): E.g. A is nuclear, because θ ◦ idA⊗B ◦η = idA ,
for the completely positive maps

η : A 3 a 7→ a ⊗ b ∈ A ⊗ B and θ := (idA ⊗ψ) : A ⊗ B → A

for some b ∈ B+ with kbk = 1 and a state ψ with ψ(b) = 1, and idA⊗B is nuclear.
The simplicity of A ⊗min B implies that A and B must be simple, because the
minimal tensor product is a bi-functor with respect to A and B.
Suppose that A ⊗ B is not purely infinite. Then A and B are both stably finite
by Part (i). By [342] (see [441] for the non-unital case) there are non-zero lower
semi-continuous semi-finite traces on A+ and B+ . Thus, there is a non-zero lower
semi-continuous trace on (A ⊗ B)+ , which contradicts T + (A ⊗ B) = 0 (for the
simple stably infinite A ⊗ B). 

4. On absence of infinitesimal sequences

The – not so deep – results of this section are mostly known to experts (since
about 1970?) and have been considered as “folklore”, but we could not find always
elementary and precise references for them.
Check here, and sort, definitions and blue discussions on
Cuntz semi-groups! e.g. ‘‘-’’ and ‘‘≈’’ given in Def. 2.5.1.
Go into opposite direction to restrict to minimal necessary pre-info
...
Move then all further details to an overview section,
e.g. to a Section 5 in Appendix A,
190 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

or to places where they will used first time.

Check again the def.’s for CS(A).
The here considered “small” Cuntz semigroup CS(A) has not been used ever
before, despite it reflects the original version of the semigroups introduced and
considered in [170, 171] by J. Cuntz, but here we generalize it to the cases of non-
simple and non-unital C *-algebras. The different generalization of the definition of
J. Cuntz require some explanations at the beginning, because we use here the – not
so common – “smallest possible” variant of the original definition of J. Cuntz. It is
the “small Cuntz semi-group” CS(A) – near to the original definition of J. Cuntz,
??? even if in that time there was not the Pedersen ideal invented ???:
We explain it with help of the Pedersen ideal Ped(B) of a C *-algebra B: It
is the smallest dense ideal of a C *-algebra B. It is algebraical generated by the set
of all ε-cut-downs (b − ε)+ of elements b ∈ B+ with ε ∈ (0, kbk).
Check next blue again!
It considers only classes of elements in the Ped(A ⊗ K) of A ⊗ K, i.e., the
minimal dense ideal of A ⊗ K. If we identify Mn (A) in the natural way with
S
(1 ⊗ pn )(A ⊗ K)(1 ⊗ pn ) then Ped(A ⊗ K) is in general not contained in n Mn (A)
and Mn (A) 6= Mn (Ped(A)),
[
Mn (A) 6⊃ Ped(A ⊗ K) ⊂ A ⊗ K .
n
S
On the other hand n Mn (Ped(A)) is contained in Ped(A ⊗ K) and contains all
representatives, but is not identical to Ped(A ⊗ K)...
S
For each element a ∈ Ped(A⊗K) there exists some element b ∈ n Mn (Ped(A))
with the property that b∗ b is MvN-equivalent to a∗ a, i.e., b∗ b ∼M vN a∗ a .
There are natural semigroup morphisms

V(A) → CS(A) → W(A) → Cu(A) . (4.1)

They are given by the map that respects the ≈-relation between elements, cf. Def-
inition 2.4.1, for the below defined or “recalled” semigroups V(A), CS(A), W(A)
and Cu(A).
Here V(A) denote the Murray–von-Neumann equivalence ( ∼M vN ) classes of
projections in A ⊗ K, cf. Definition 2.0.1, that is in general considerably stronger
relation than the ( ≈ ) relation.
Beginning with CS(A), the natural morphisms of pre-ordered semigroups in
(4.1) are injective, but the natural semi-group morphism V(A) → CS(A) is in
general not necessarily injective, e.g. . CS(O2 ) = CS(O∞ ) = {[0], [1]} but ??????
E.g. V(O2 ) = {[0], [1]} and V(O∞ )\{[0]} is naturally isomorphic to Z, CS(A) =
W(A) = Cu(A) = {[0], [1]} for all purely infinite simple unital C *-algebras A.
In particular their equivalence classes have nothing to do with with its K∗ -
theory!
 4. ON ABSENCE OF INFINITESIMAL SEQUENCES 191

More generally, for all simple purely infinite C *-algebras A holds CS(A) =
{[0], [p]} = Cu(A), with any non-zero projection p ∈ A ⊗ K, i.e., for each a, b ∈
(A ⊗ K) \ {0} there exist c, d, e, f ∈ A with cad = b and ebf = a.
If non-zero A is simple and stably projection-less then V(A) = {0} , but
CS(A) 6= {0} .
But one can adjoin a unit to A and then consider the kernel of a + z · 1 → z · 1
...
Next ‘‘filtration’’ has to be explained -- if true
There are natural filtrations (indexed by sub-semigroups of R+ ∪ {+∞}) and
certain order topologies that allow to see that the right ones are completions of the
left ones (except V(A), where this happens only in the case where A has real rank
zero).
This does not say that the set of elements a ∈ Xb ⊆ A ⊗ K that represent
S
the class of b ∈ n Mn (A) in A ⊗ K with respect to the below defined equivalence
relation a ≈ b (given by a - b and b - a in A ⊗ K ) is the same as the class of
S
c ∈ n Mn (A) with c ≈ a
?????
This causes that it is in general not clear under what circumstances c ∈ A ⊗ K
and c - a ∈ Mn (A) implies that c ≈ b for some b ∈ n Mn (A) ( 28 ).
S

The point is that a ≈ b in A ⊗ K and a, b ∈ Mn (A)+ (respectively a, b ∈

Ped(A⊗K)+ ) implies that a ≈ b also inside Mn (A) (respectively inside Ped(A⊗K) ).
The “≈” classes become in A ⊗ K much bigger. For example, in case A :=
C0 (0, 1], we can take the increasing pice-wise linear functions hn ∈ C0 (0, 1]+ given
by hn (t) := min(1, max(2n t − 1, 0)) for t ∈ [0, 1], (n = 0, 1, . . .). (We use them for
some constructions, see Section 22 in Appendix A.)

1/2 P∞
If we define fn (t) := 2−n (hn+1 (t) − hn (t)) and let T := n=1 fn ⊗ p1,n ,
then T ∈ C0 (0, 1] ⊗ K and T T ∗ = ( fn2 ) ⊗ p11 = e0 ⊗ p11 for e0 (t) := t, but
P

T ∗ T 6∈ n Mn (A).
S

It is not clear if in general the natural image of W(A) in Cu(A), i.e., the image
of the map that sends classes in W(A) into the corresponding – usually much
bigger classes (of ≈-equivalent elements in A ⊗ K) which are elements of Cu(A)
is a hereditary sub-semigroup in Cu(A) with respect the natural order in Abelian
semi-groups. (It has to do with variants of the so-called “radius of comparison”.)
I.e. it is unknown if W(A) “is” a “hereditary” sub-semigroup of Cu(A), i.e., it
is not known if c ∈ A ⊗ K and c - a for some a ∈ Mm (A) and m ∈ N implies the
existence of some n ∈ N and b ∈ Mn (A) such that c ≈ b. See [85] for an additional
property on A (that is inspired by the “bounded radius of comparison” for vector
bundles) which implies that W(A) is “hereditary” in Cu(A).

28 It has to do with the classical idea of a “radius of comparison” for vector bundles on

compact manifolds.
192 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

It is not difficult to see that for each element a ∈ Ped(A ⊗ K)+ there exists
n ∈ N and b ∈ (Ped(A) ⊗ Mn )+ such that a ∼M vN b. But there are elements in
a ∈ Ped(A⊗K)+ that are not itself contained in the algebraic tensor product A F ,
where here F ⊂ K means here the *-algebra of finite rank operators on `2 (N).
It shows that CS(A) is isomorphic to a --hereditary sub-semigroup both of
W(A) and of Cu(A), that W(A) is in a natural manner a sub-semigroup of Cu(A)
and that CS(A) and Cu(A) are stable invariants of C *-algebras A, i.e., there are
natural isomorphisms CS(A) ∼ = CS(A ⊗ K) and Cu(A) ∼
= Cu(A ⊗ K). Moreover, it
is not difficult to check that W(A ⊗ K) = Cu(A) in a natural way (using the flip
on K ⊗ K).
This natural isomorphisms are induced by the embedding T ∈ K 7→ T ⊗ p11 ∈
K ⊗ K that extends to a *-monomorphism of A ⊗ K onto the full stable corner
A ⊗ K) ⊗ p11 of A ⊗ K ⊗ K. The *-endomorphism is unitarily homotopic to an
isomorphism from A ⊗ K onto A ⊗ K ⊗ K, because the *-monomorphism T ∈
K 7→ T ⊗ p11 ∈ K ⊗ K is unitarily homotopic in sense of Definition 5.0.1 to an
isomorphism from K onto K ⊗ K by a norm-continuous path t ∈ [0, ∞) → U (t) of
unitary elements in M(K ⊗ K). (See the comments following Definition 5.0.1 for
more details.)
?? Verification discussion?:
Take an isometry S from `2 (N × N) onto `2 (N), that is defined by a suitable
bijection from N × N onto N × N. It defines an isomorphism from K onto K ⊗ K by
a 7→ S ∗ aS. Let p11 the upper-left minimal projection in K. There is an isometry R
from `2 onto `2 ⊗e1 given by R(x) = x⊗e1 . It defines the map a → RaR∗ = a⊗p11 .
Then a → SRaR∗ S ∗ = T aT ∗ is a C *-morphism from K into K given by an isometry
T ∈ M(K) = L(`2 ). If we find a norm-continuous path t 7→ U (t) in the unitaries
of M(K) with U (t)∗ T aT ∗ U (t) → a for all a ∈ K then

S ∗ aS = lim (S ∗ U (t)S)(a ⊗ p11 )(S ∗ U (t)S)∗ .

t→∞

Consider projections in K with q1 ≤ q2 ≤ · · · and limn kqn aqn −ak = 0 for all a ∈ K
and then find t 7→ U (t) norm-continuous with U (t)∗ T qn T ∗ U (t) = qn for t ≥ n = 1.
Compare next with comments below Def. 5.0.1
The latter holds because M(K ⊗ K) ∼ = L(`2 ), U(L(`2 )) = U0 (L(`2 )) and all
isometries s ∈ L(`2 ) with Dim((1 − ss∗ )`2 ) = ∞
Are (???) unitary equivalent. Perhaps only in case that n (1 − sn (sn )∗ )`2 is
S

dense in `2 ???
Needs to study the unitary part of s for the unitary equivalence? Does the
standard part for Wold decomposition U ⊕ (T ⊗ idH3 ) – up to unitary equivalence
of H with H1 ⊕ (`2 ⊗ H3 ), where T ∈ L(`2 ) is the Toeplitz shift (unilateral shift)
and U is a unitary on H1 – always “absorb” the unitary part?
Or try to show that there is a unitary V from `2 (N) ⊗ `2 (N) onto `2 (N) such
that V ◦ (1 ⊗ S) = s ◦ V ?
 4. ON ABSENCE OF INFINITESIMAL SEQUENCES 193

Give Ref for precise unitary equivalence!!!

But there exists also something explicit ?????
Alternatively, one can take the isometry R := 1 ⊗ s ∈ M(A ⊗ K ⊗ K) with
RR∗ = 1 ⊗ p11 . It gives RbR∗ ∼M vN b (in A ⊗ K ⊗ K itself) and RbR∗ = h(b) ⊗ p11
for well-defined h(b) ∈ A ⊗ K for all b ∈ (A ⊗ K ⊗ K) . Then check that h is an
isomorphism from A ⊗ K ⊗ K onto A ⊗ K and that h(·) ⊗ p11 is unitary homotopic
to id on A ⊗ K ⊗ K.
This monomorphism defines isomorphisms from CS(A) onto CS(A ⊗ K) and
from Cu(A) onto Cu(A⊗K), because each positive element of Ped(A⊗K) is Murray–
S
von-Neumann equivalent (∼M vN equivalent) to an element of n Mn (Ped(A)) by
Pedersen’s characterization of sets of generators of the minimal dense ideal Ped(B)
of a C *-algebra B, as the ε-cut-down’s (a − ε)+ for a ∈ G in any – fixed given –
subset G ⊆ B+ that generates a dense two-sided ideal of B.
The semigroup V(A) consists of the Murray–von-Neumann equivalence classes
of projections in A ⊗ K (It is zero if A is stably projection less). Projections are
always in the Pedersen ideal Ped(A ⊗ K) of A ⊗ K and p - q implies that there
exists a partial isometry v ∈ A ⊗ K with v ∗ v = p and vv ∗ ≤ q. Thus, v ∗ qv = p.
The map from V(A) to CS(A ⊗ K) is the natural one and is not necessarily
injective, because properly infinite full projections in A ⊗ K (if exist) are all ≈-
equivalent but need not to be ∼M vN equivalent in A ⊗ K :
The existence of partial isometries v, w ∈ A⊗K with v ∗ v = p, vv ∗ ≤ q, w∗ w = q
and ww∗ ≤ p does not imply that p and q are MvN-equivalent. The existence of
such v, w is equivalent to the existence of projections p1 ≤ q1 ≤ p with q1 ∼M vN q
and p1 ∼M vN p. The latter does not imply q ∼M vN p because p might be an
infinite projection (e.g. p1 6= p).
The equation CS(A) = CS(A ⊗ K) can be seen as follows:
Notice that Ped(A ⊗ K) is the same as the algebraic ideal of A ⊗ K generated by
Ped(A) ⊗ p11 . Here Ped(A) denotes the Pedersen ideal of A. If a, b ∈ Ped(A) ⊗ Mn
with a - b in A ⊗ K are given, then a - b in Ped(A) ⊗ Mn , i.e., if a = limn d∗n bcn
for suitable sequences cn , dn ∈ M(A ⊗ K). We can replace the cn and dn by
1/n 1/n 1/n 1/n
en := (b − 1/n)+ cn (a − 1/n)+ and fn := (b − 1/n)+ dn (a − 1/n)+ . Then
en , fn ∈ Ped(A ⊗ Mn ) = Ped(A) ⊗ Mn and a = limn fn∗ ben .
The ideal Ped(A) is algebraically generated by the ε-cut-downs (a − ε)+ of
positive contractions a ∈ A+ , and the ideal Ped(A) ⊗ p11 of A ⊗ p11 generates
(algebraically) a minimal dense ideal ideal of A ⊗ K – as one can see from Lemma
2.1.9. This implies that for each element b in the Pedersen ideal Ped(A ⊗ K)
of A ⊗ K there exists n = n(b) ∈ N, contractions e1 , f1 , . . . , en , fn ∈ A+ and
d1 , d2 ∈ A ⊗ K with fk ek = ek for k = 1, . . . , n, b∗ b = d∗1 diag(e1 , . . . , en )d2 . It
follows that b ≈ b∗ b ∼W vN cc∗ with 2cc∗ diag((f1 − 1/2)+ , . . . , (fn − 1/2)+ ) = cc∗
for suitable c ∈ A ⊗ K. Thus, cc∗ ∈ Ped(A ⊗ Mn ) = Mn (Ped(A)), and each
S
b ∈ Ped(A ⊗ K) is ≈-equivalent to some element in n Mn (Ped(A)) ⊆ Ped(A ⊗ K) .
194 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

S
(But notice here that Ped(A ⊗ K) 6= n Ped(A) ⊗ Mn in general, if the Mn are
S
considered here as Pn KPn with Pn of rank = n, and Pn ≤ Pn+1 and n Pn KPn
dense in K, e.g. this happens even for A := C or A := C0 (0, 1] .)
We use here the (in literature not common) small “local” hereditary sub-
semigroup CS(A) of the “larger” Cuntz-semigroups W(A) ⊆ Cu(A). for an “in-
finitesimal” characterization of simple p.i. C *-algebras. Therefore we discuss the
differences and different applications before we return to the study of p.i. simple
C *-algebras. The definition of J. Cuntz in [171] itself did not care about the pos-
sibly different definitions of the related semi-group, and that they have different
properties – even for simple ASH C *-algebras or Cuntz-Pimsner algebras.
The below displayed elements and semigroups show that ????? what ???
Z+ ∼= V(C) = CS(C) = W(C) = CS(A) 6= W(A) if A = K because then
W(K) = Cu(K) ∼ = Z+ ∪ {+∞} . The isomorphism Cu(K) → Z+ ∪ {+∞} is given
by [a] 7→ Rk(a) the rank of a.
Moreover W(A) 6= Cu(A) if A is unital and stably finite.
There exists a C *-algebra A with CS(A) 6= W(A)
Really? A = c0 ?, Cite ???
Above examples correct? Below considerations?
The “small” Cuntz algebras CS(A) have in particular the following extra prop-
erty:
The C *-algebra A is simple if and only if for each non-zero a, b ∈ Ped(A) there
exists m, n ∈ N with [a] ≤ m[b] and [b] ≤ n[a] in CS(A).
This is not the case for the semigroup W(A).
Example of simple A and elements a ∈ Mm (A) and b ∈ Mn (A) with the
property that [b] 6≤ k[a] for all k ∈ N : A = K, b := diag(1, 1/2, 1/3, . . .) a :=
diag(1, 0, 0, . . .).
Compare Definition 2.5.1 !!
Definition 2.4.1. Let B denote a non-zero (not necessarily simple) C *-algebra
and let b, c ∈ B. We write b - c if there exist sequences dn , en ∈ B such that
b = limn dn c en .
We say that b and c are Cuntz equivalent if b - c and c - b and denote this
by b ≈ c.

It is easy to see that a - b, and b - c implies a - c, and that the relation b ≈ c

is an equivalence relation on B. It will be also denoted sometimes by b ∼ c. The
corresponding equivalence classes, say of b ∈ B will be denoted by [b]. Thus b ≈ c
or [b] = [c] are equivalent to b - c and c - b.
In some special (notational) situation b ≈ c will be also denoted by b ∼ c or by
[b] = [c]. We write later also [b] ≤ [c] for b - c.
Decide here for ≈ or ∼ !!! ?? ??
 4. ON ABSENCE OF INFINITESIMAL SEQUENCES 195

Then ≈ becomes an equivalence relation on B. The ≈-class of b ∈ B will be

denoted by [b] (and sometimes more precisely by [b]B ).
It is easy to see that - is a transitive relation on B × B with b - b and 0 - b
for all b ∈ B. And it is not difficult to see that b - b∗ b - b, and b - c for 0 ≤ b ≤ c.
There is an in some cases different equivalence relation [a] = [b] that considers
the isomorphism classes of the right Hilbert B-modules aB and bB.
Find reference for above and next red!?
The source is a bit semi-philosophic ... (Elliott and Co.)
I.e., if there exists a linear isometry T from the closed right ideal Ra := aB
generated by a onto Rb := bB that satisfies with (T x)c = T (xc) for all c ∈ B??
Gives that T (aa∗ )1/m = limn bxm,n ????? Not really understood!!!
(??? What happens with ”open” support projections P and Q in B ∗∗ of aa∗
and bb∗ . What gets lost in the exact or amenable case?
Seems to be that T is given by a partial isometry in B ∗∗ with T = QT P and
We can here always suppose that the sequences (dn ) and (en ) satisfy dn ∈
1/n 1/n 1/n 1/n
(bb − 1/n)+ · B · (cc∗ − 1/n)+ and en ∈ (c∗ c − 1/n)+ · B · (b∗ b − 1/n)+ . It
∗

does not change the definition of “ - ” in Definition 2.4.1.

Sometimes we write more precisely b -B c (or later [b]B ≤ [c]B for its equiv-
alence classes) if b, c ∈ A ⊆ B ⊆ D, a - b in B but not b - c in A. Clearly
[b]D ≤ [c]D if [b]B ≤ [c]B and B ⊆ D.
The definition of ≈ implies that b - c, d ≈ b and c ≈ e imply d - e. Therefore
we can define on the classes [b] and [c] a pre-order [b] ≤ [c] if b - c. (29)
The MvN-equivalence b ∼M vN c of elements b, c ∈ B+ – defined by the exis-
tence of d ∈ B with d∗ d = b and dd∗ = c – implies that b ≈ c, i.e., [b] = [c]. It is
also easy to see that a, b ∈ B+ and a ≤ b imply that [0] ≤ [a] ≤ [b].
If M(B) is properly infinite – as it is e.g. the case for B := A ⊗ K –, then we
can define on the ≈-classes [b] a commutative and associative addition by [b]+[c] :=
[sbs∗ + tct∗ ], where s, t ∈ M(B) are isometries with orthogonal ranges: s∗ s = 1,
t∗ t = 1 and s∗ t = 0. It is easy to see that it is well-defined on the family of ≈-classes
[b] and that it is independent from the chosen isometries s and t (with s∗ t = 0). See
Lemma 4.2.6 and its proof for the transformation of those generalized Cuntz-sums.
It studies the invariance for unitary equivalence classes if ss∗ + tt∗ = 1 or in
case that 1 − ss∗ + tt∗ is full and infinite. But that can be carried over to ≈-classes
in case where 1 − ss∗ + tt∗ is full and infinite e.g. by passage to st, t ...
Need Remark (!) to Lemma 4.2.6, that explains that in the sufficiently de-
generated case this shows also the ≈-invariance, a1 ≈ a2 , b1 ≈ b2 implies that
a1 ⊕s,t b1 ≈ a1 ⊕q,r b1 if s, t, q, r ∈ M(A) are isometries with s∗ t = 0 and q ∗ r = 0.

29The classes will be sometimes also denoted by [b], hbi or [b] , simply because for the later
≈
used several equivalence classes similar notations have to be used.
196 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

We define now a large (or maximal ) version of the Cuntz semigroup that con-
tains the usual used semigroup W (A) and our very small version
Definition 2.4.2. We define here a large Cuntz-semigroup Cu(A) by tak-
ing above B := A ⊗ K, i.e.,
Cu(A) := {[a]≈ ; a ∈ A ⊗ K}
consisting of the ≈-classes [a] with a ∈ A ⊗ K and with addition defined by
[a] + [b] := [a ⊕ b]
the direct sum of representatives of the classes: [a] + [b] := [a ⊕s,t b] with help of
any unital copy of E2 = C ∗ (s, t) in M(A ⊗ K) .
We define W(A) as the sub-semigroup of Cu(A) given by the classes [a] ∈ Cu(A)
S
with a ≈ b for some b ∈ n Mn (A) ⊆ A ⊗ K.

The usual definition of W (A), e.g. in ?????? is given on the disjoint union of
the sets Mn (A) and use of “canonical” unital C *-morphisms from Mm ⊕ Mn into
Mm+n . This definitions of W(A) are equivalent on ≈-classes, because the cartesian
sum
Mm (A) ⊕ Mn (A) → Mm+n (A)
defined by
(b1 , b2 ) 7→ b1 ⊕ b2 ∈ Mm+n (A) ⊆ A ⊗ K for b1 ∈ Mm (A) , b2 ∈ Mn (A) ,
– using the natural C *-morphisms Mn (A) ∼
= A ⊗ Mn and Mm ⊕ Mn → Mm+n –,
is unitary equivalent to the map
(b1 , b2 ) 7→ b1 ⊕s,t b2 ∈ sMm (A)s∗ + tMn (A)t∗ ,
by a suitable unitaries Un,m ∈ M(A ⊗ K).
S
But notice here that b1 ⊕s,t b2 is not necessarily in n Mn (A) – interpreted as
S
n A ⊗ Mn ⊂ A ⊗ K anymore. But there exists for given m.n ∈ N isomorphisms
of A ⊗ K that are approximately inner inside M(A ⊗ K) and define a “natural”
isomorphism from sMm (A)s∗ + tMn (A)t∗ ⊂ A ⊗ K onto (A ⊗ Mm ) ⊕ (A ⊗ Mn ) ⊆
A ⊗ Mm+n ⊂ A ⊗ K.
The “large” semi-group Cu(A) is often not useful for our application, because it
contains in the case of σ-unital A always an additively absorbing “infinite” idempo-
tent [e], e.g. let f ∈ A+ a strictly positive contraction and define a strictly positive
contraction e ∈ A ⊗ K by e := n n−1/4 f 1/n ⊗ pn,n . Then [b] ≤ [b] + [e] = [e]
P

in Cu(A) for all b ∈ A ⊗ K . In particular, the Grothendieck group Gr(Cu(A)) is

zero for each σ-unital C *-algebra A . The reasons for [b] + [e] = [e] are that for
all C *-algebras B with strictly positive element e ∈ B holds b - e for all b ∈ B,
e ⊕s,t e is again strictly positive in A ⊗ K, e ≈ 0 ⊕ e - b ⊕ e, b - e and e ⊕ e ≈ e.
Definition 2.4.3. We define the local Cuntz semigroup as the sub-
semigroup CS(A) of Cu(A) given by
[
CS(A) := {[a] ∈ Cu(A) ; a ∈ Mn (Ped(A)) } .
n
 4. ON ABSENCE OF INFINITESIMAL SEQUENCES 197

S
Notice here that for the Pedersen ideals always holds n Mn (Ped(A)) ⊆
Ped(A ⊗ K), but also that for each positive element b in the Pedersen ideal
Ped(A ⊗ K) of A ⊗ K there exists n ∈ N and a positive element c ∈ Mn (Ped(A))+
that is Murray–von Neumann equivalent to b. Therefore CS(A) and Cu(A) are
both invariants of the Morita equivalence classes of σ-unital C *-algebras A.
The most common sub-semigroup of the “large” Cuntz semigroup Cu(A) is its
S
sub-semigroup W(A) consisting of the ≈-classes of elements in n Mn (A) ⊆ A ⊗ K.
S
It does not matter if we consider in the definition n Mn (A) as “disjoint” union
of the sets Mn (A), or if we consider Mn (A) in a natural way as Mn (A) ⊕ 0p ⊂
Mn+p (A) ⊂ M(A ⊗ K) and “identify” Cuntz-equivalent elements, because (it is
easy to see that) inside A ⊗ K holds

SaS ∗ + T bT ∗ ≈ a ⊕ b ≈ diag(a, b)

for a ∈ Mn (A), b ∈ Mp (A) and any isometries S, T ∈ M(A ⊗ K) with S ∗ T = 0.

Further notice that W(A) = CS(A) = CS(A ⊗ K) if A is unital, but that
W(A ⊗ K) = Cu(A) is not isomorphic to W(A) if A is stably finite.
An alternative way – inspired by [171] – to define W(A) and our version CS(A)
is the following (that is near to the original definition of J. Cuntz in [171]):
Take a unital copy of O2 given by isometries s, t ∈ M(K) with ss∗ + tt∗ = 1.
Define a copy of O2 in M(A ⊗ K) by S, T ∈ M(A) ⊗ M(K) ⊆ M(A ⊗ K) by
S := 1M(A) ⊗ s and T := 1M(A) ⊗ t (recall that ⊗ denotes here the minimal
C *-algebra tensor product, that is also called spatial tensor product).
Let F ⊂ K(`2 ) denote the the set of all operators of finite rank on `2 (N).
It is an ideal of K(`2 ), because it is just the Pedersen ideal of K(`2 ) because it
is algebraically generated by all elements (T − ε)+ for compact positive operators
T ∈ K(`2 )+ . Notice that the algebraic tensor product F F is a dense *-subalgebra
of the C*-algebra tensor product of K(`2 ) ⊗ K(`2 ) ∼ = K(`2 ⊗2 `2 ). But is not an
P −n
ideal of K(`2 ) ⊗ K(`2 ), because, e.g., the operator X := n2 Pn ⊗ Pn (with
Pn (v) :=< v, en > ·en for v ∈ `2 and {e1 , e2 , ..., en , ...} an orthogonal basis of `2 )
is the orthogonal projection of vectors w ∈ `2 ⊗2 `2 ∼ = `2 on the vector vX :=
P −n
n 2 e n ⊗ e n . Thus X is a projection in K(` 2 ⊗ 2 ` 2 ) but is not in the algebraic
tensor product F ⊗ F ⊆ K(`2 ⊗2 `2 ). But the algebraic ideal (!) of K(`2 ⊗2 `2 )
generated by the algebra F ⊗ F is again the minimal dense ideal of K(`2 ⊗2 `2 ).
The set F is identical with the norm-dense algebraic ideal of the closed ideal
S
K := K(`2 (N)) generated by its subset n Mn – where we identify here Mn nat-
urally with pn Kpn and pn is the orthogonal projection onto the linear span of
{e1 , . . . , en } for the canonical basis of `2 (N).
Obviously, F is a *-ideal of M(K) ∼ = L(`2 ), that is dense in K, and is minimal
with this properties. Thus, F is exactly the Pedersen ideal Ped(K) of K, i.e., is the
minimal dense ideal of K.
But the algebraic tensor product ???? ???
198 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

The algebraic tensor products A F and Ped(A) F ⊆ A F are dense in

the algebra A ⊗ K and are algebraic *-ideals of the algebraic (!) tensor product
M(A) ⊗alg M(K) that is in general strictly contained in M(A ⊗ K).
By the definition of the Pedersen ideal, the algebraic ideal of A ⊗ K generated
by Ped(A) F contains the Pedersen ideal Ped(A ⊗ K) ( := minimal dense ideal
of A ⊗ K).
This follows from the general observation (that is easy to see) that Ped(A) ⊗
Ped(B) ⊆ Ped(A ⊗ B), because for positive contractions a ∈ A+ and b ∈ B+ the
products (a − ε)+ ⊗ (b − δ)+ ≤ ((a ⊗ b) − γ)+ for a ∈ A+ , b ∈ B+ , ε > 0, δ > 0,
and 0 < γ < (ε · δ)/2
A⊗K ???? and generates ????. Above we have seen that sometimes Ped(A) F
is not an (algebraic) ideal of A ⊗ K.
Thus are in general not algebraic *-ideals of the C *-algebra M(A) ⊗ M(K).
The “sum” X ⊕ Y := SXS ∗ + T Y T ∗ for given isometries, e.g. S := 1 ⊗ s and
T := 1 ⊗ t is well defined and preserve ≈ (and ∼M vN ) classes of the elements in
A F , respectively in Ped(A) F .
The classes of elements in A F build W(A) and the classes of elements in
Ped(A) ⊗ F build CS(A).
The definition of the Pedersen ideal of a C *-algebra shows that Ped(A ⊗ K) is
identical with the algebraic ideal of A ⊗ K generated by Ped(A) ⊗ F , in particular
F = Ped(K).
????
The set F · F of products f1 · f2 is in in general not identical with F , i.e., F F
(perhaps) generates F of A ⊗ B even in case A = K B = K.
Here is something wrong!!!
F = Ped(K) seems to be OK! This are all elements in K with finite ranks.
But F F is not the Pedersen ideal of K ⊗ K but is contained in the Pedersen
ideal and generates the Pedersen ideal of K ⊗ K ∼
= K.
Consider it in `2 ⊗ `2 ????
Thus, CS(A ⊗ K) ∼
= CS(A) in a natural way. ???? Checked ???
??? Moreover (A ⊗ K) F is “locally” isomorphic to A ⊗ K ⊗ K ????
The only stably invariant versions of the original Cuntz semigroups are CS(A) ∼
=
CS(A ⊗ K) and Cu(A) ∼ = Cu(A ⊗ K) ∼ = W(A ⊗ K), where we take for Cu(A) “our”
– above given – Definitions.
That W(A ⊗ K) ∼ = Cu(A ⊗ K) follows from A ⊗ K ⊗ Mn ∼
= A ⊗ K ⊗ e11 by
*-conjugating with an isometry in 1M(A) ⊗ M(K ⊗ Mn ).
The “large” Cuntz semigroup can give some sort of informations about the
countably generated Hilbert-modules over A.
 4. ON ABSENCE OF INFINITESIMAL SEQUENCES 199

But notice that there are examples of C *-algebras A where CS(A), W(A) and
Cu(A) are all different, e.g. for A := c0 (N) ⊗ C([0, 1]∞ ), the zero sequences in
C([0, 1]∞ ).
last example A OK ?? Check it??
But, since they are all contained in Cu(A), one can carry out all calcula-
tions inside Cu(A), – except some relations in their corresponding (pre-ordered)
Grothendieck groups, i.e. given by additional absorbing elements.
Check Def. Cuntz semi-group. Compare with that in Sec. ??? in
App. A
The difference between Cu(A) and its hereditary sub-semigroup CS(A) becomes
visible in case of simple C *-algebras A, because then CS(A) is a simple semi-group,
i.e., for any two non-zero elements [a], [b] ∈ CS(A) there exist m, n ∈ N such that
[b] ≤ m[a] (i.e., b - a ⊗ 1m ) and [a] ≤ n[b]. But in general, for

a ∈ Ped(A)+ ∼
= Ped(A)+ ⊗ p11 ⊆ Ped(A ⊗ K)

in the Pedersen ideal of A there does not exist a positive integer m such that
[e] ≤ m[a], i.e., e - a ⊗ 1m , for the above defined “infinite” element [e] ∈ Cu(A).
It would imply that 3m[a] ≤ m[a] and causes the existence of a scaling element
b ∈ A ⊗ K. But the existence of non-zero scaling elements causes the existence of
a non-zero projection p ∈ A ⊗ K and k ≥ 1 with 2k[p] ≤ k[p].
Exactly as ‘‘absorbing’’ is defined? Check this:
Thus, the “large” semi-groups Cu(A) of σ-unital A contain always a kind of
“absorbing” infinite elements by this trivial reason (and is therefore not suitable
for our purpose).
If A is simple and σ-unital then Cu(A) = CS(A) if and only if A is stably
infinite, i.e., if and only if A ⊗ K contains an infinite projection p 6= 0 (which is
then a properly infinite projection by Lemma 2.1.6) ( 30 ).
A better ‘‘defining’’ name could be:
‘‘compatibly pre-ordered additive semi-group’’

Definition 2.4.4. Let S = (S, +, 0, ≤) a compatibly pre-ordered additive

semigroup (shortly named as c.p.a. semigroup) with order-minimal zero-element
0, i.e., (S, +) is a commutative semigroup, and for x, y ∈ S holds: x + 0 = x ,
y ≤ x + y , and that x ≤ 0 implies x = 0 .
We say that S is simple if for every for every x, y ∈ S \ {0} there exist n ∈ N
with y ≤ nx.
A sequence (x1 , x2 , . . .) with xn ∈ S is infinitesimal (with respect to given y ∈
S \ {0}) if kn → ∞ (by n → ∞) for each sequence of natural numbers k1 , k2 , . . . ∈ N
with the property that y ≤ kn xn for all n ∈ N.

30 We have introduced here the “small” variant CS(A) of a Cuntz semi-group for a transparent

formulation of the proof and statement of Corollary 2.4.6.

200 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Clearly every infinitesimal sequence (x1 , x2 , . . .) satisfies that the set X :=

{x1 , x2 , . . .} ⊆ S is an infinite subset of S.
Indeed, consider – more general – any finite subset X of S and let y ∈ S \{0}. Then
– alternatively – there exists xn0 ∈ X such that there is no k ∈ N with y ≤ kxn0
or there exists k(y) ∈ N with y ≤ k(y)xn for all xn ∈ X. Hence, the sequence
x1 , x2 , . . . ∈ X is not infinitesimal (with respect to y).
Thus, finite pre-ordered additive semigroups S = (S, +, 0, ≤) with order-
minimal zero-element can not contain an infinitesimal sequence.
In particular, the semi-group S = {0, y} with relations 0 ≤ y and 2y = y has
no infinitesimal sequence.
The pre-ordered semi-group (Z+ , +) is simple and does not contain any infini-
tesimal sequence (x1 , x2 , . . .) ∈ Z∞
+ , because for 0 < y ∈ Z holds that the relation
y ≤ m · x implies that m, x ∈ N = Z+ \ {0}, i.e., 1 ≤ x and y ≤ y · x. Thus, with
k(y) := y we can see that (Z+ , +) contains no infinitesimal sequence.

Lemma 2.4.5. Let S = (S, +, 0, ≤) a simple pre-ordered additive semi-group.

Then S \ {0} contains no infinitesimal sequence, if and only if, for each y ∈
S \ {0} there exists k(y) ∈ N such that y ≤ k(y)x for all x ∈ S \ {0}.

Proof. Suppose that for each y ∈ S \ {0} there exists k(y) ∈ N – depending
only on y – such that y ≤ k(y)x for all those x ∈ S \ {0} that have the property
that there exists `(x, y) ∈ N with y ≤ `(x, y)x.
Let (x1 , x2 , . . .) is a sequence in S and k1 , k2 , . . . a sequence in N. If y ∈ S \ {0}
satisfies y ≤ kn xn then xn 6= 0 and y ≤ k(y)xn for all n ∈ N, i.e., (x1 , x2 , . . .) is
not an infinitesimal sequence. Thus, S contains no infinitesimal sequences (with
respect to any y).
Now suppose that S contains no infinitesimal sequence (with respect to y ∈
S \ {0}) and that S is simple. We show that this implies the existence of k(y) ∈ N
with y ≤ k(y)x for all x ∈ S \ {0} .
Notice that if z ∈ S \ {0}, then there are m, n ∈ N with y ≤ mz, z ≤ ny
for suitable m, n ∈ N, and that if (x1 , x2 , . . .) is a sequence in S \ {0}, then this
sequence is infinitesimal with respect to z, if and only if, (x1 , x2 , . . .) is infinitesimal
with respect to y. This is because, e.g. z ≤ nkn xn if y ≤ kn xn .
The simplicity of S implies that, for each x ∈ S \ {0}, there exists n ∈ N with
y ≤ nx. Let `(y, x) ∈ N denote the minimal number n ∈ N with this property. We
show that the subset My := {`(y, x) ; 0 6= x ∈ S} is bounded:
Otherwise there exists a sequence (x1 , x2 , . . .) in S \ {0} with `(y, xn ) → ∞ for
n → ∞.
Let k1 , k2 , . . . ∈ N with y ≤ kn xn . Then kn ≥ `(y, xn ) by definition of `(y, xn ).
Thus, kn → ∞. It means that (x1 , x2 , . . .) is an infinitesimal sequence in S \ {0}.
Its existence contradicts our assumption on infinitesimal sequences.
 4. ON ABSENCE OF INFINITESIMAL SEQUENCES 201

It implies that the number k(y) := max My ∈ N exists and satisfies y ≤ k(y)x
for all x ∈ S \ {0}. 

... Where is the definition of ”non-elementary” ?

Corollary 2.4.6. A non-elementary simple C*-algebras A is purely infinite,

if and only if, the “small” Cuntz semigroup CS(A) of A does not contain an infin-
itesimal sequence, in the sense of Definition 2.4.4.
If a non-elementary simple C*-algebra A has real rank zero and the natural
image of V(A) in CS(A) does not contain an infinitesimal sequence for CS(A) then
A is purely infinite.
If A is an elementary simple C*-algebra then CS(A) ∼= (Z+ , 0, +), and
(Z+ , 0, +) does not contain an infinitesimal sequence.

Proof. If A is elementary, then the pre-ordered semi-group CS(A) is isomor-

phic to the ordered additive semi-group of non-negative integers Z+ = { 0, 1, . . . },
given by semi-group isomorphism [a] 7→ Rk(a) for a ∈ Ped(K) (where Rk(a) denotes
the rank of a ∈ K, it is on other places also denoted by: rank(a)).
Above we have seen that the semigroup (Z+ , 0, +) contains no infinitesimal
sequence – in sense of our Definition 2.4.4.
Recall that [a] = [a∗ ] = [a∗ a] in CS(A) for a ∈ Ped(A ⊗ K), cf. [175], – or use
our Lemma 2.5.3(xi) and Lemma A.6.1.
If A is purely infinite and simple, then Definition 2.4.1 of Cuntz equivalence
classes [a] and Proposition 2.2.1(ii) implies that [a] = [a∗ a] = [b∗ b] = [b] for any non-
zero a, b ∈ A ⊗ K , cf. Lemma 2.5.3(xi), and that A is non-elementary (i.e., A is not
isomorphic to the compact operators on a Hilbert space). Thus, CS(A) = { [0], [a] }
for any non-zero element a ∈ A, [0] is the zero of CS(A), and 2[a] = [a] > [0] .
By the considerations on finite additive semi-groups (above here and in Lemma
2.4.5) this finite pre-ordered semi-group with zero element does not contain any
infinitesimal sequence in the sense of Definition 2.4.4.
To show the opposite direction we suppose that A is simple and non-
elementary, and that CS(A) does not contain an infinitesimal sequence.
It is easy to see that for all simple A 6= {0} the “small” Cuntz semigroup CS(A)
is simple (and vice versa).
By Lemma 2.4.5, the non-existence of infinitesimal sequences in simple semi-
groups is equivalent to the existence of – “universal” = only from y depending –
numbers ν(y) ∈ N with the property that y ≤ ν(y)x for all non-zero elements x.
Thus, for every non-zero b ∈ Ped(A) there exist ν(b) ∈ N such that [b] ≤ ν(b)·[a]
for every non-zero a ∈ Ped(A). But this implies the following:
Let a, b ∈ A+ with kak = 1 = kbk and ε ∈ (0, 1). If we let δ := ε/3 then (a−δ)+
and (b−δ)+ are in Ped(A). Thus, there exists numbers n := n(b, ε) := ν((b−δ)+ ) ∈
N, – depending only from b and ε > 0 – with the property (b − δ)+ ⊗ e11 -
202 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

(a − δ)+ ⊗ 1n in A ⊗ Mn . In particular, there exists c1 , . . . , cn ∈ A with

kb − (c∗1 ac1 + . . . + c∗n acn )k < ε .

It says that A satisfies the criteria in Part (iv) of Proposition 2.2.1. Therefore A is
purely infinite.
Now we consider the more special case where A satisfies the additional pre-
assumptions that A is a simple, non-elementary C *-algebra, that has real rank
zero, again with the additional non-degeneracy property that there exists 0 6= y ∈
Ped(A ⊗ K)+
with what ????
Then there exists a non-zero projection q ∈ A⊗K and m, n ∈ N with [q] ≤ m[y]
and [y] ≤ n[q].
Suppose that there are a1 , a2 , . . . ∈ Ped(A ⊗ K)+ with kak k = 1 that represent
an infinitesimal sequence [a1 ], [a2 ], . . . ∈ CS(A), then nonzero projections

pk ∈ (ak − 1/2)+ (A ⊗ K)+ (ak − 1/2)+

build an infinitesimal sequence in [p1 ], [p2 ], . . . ∈ CS(A) with respect to [q].

By assumptions, we have excluded the existence of infinitesimal sequences de-
fined by projections in the Cuntz semigroup CS(A). Thus, CS(A) does not contain
any infinitesimal sequence, and the above considered cases apply here also to A and
prove purely infiniteness of A. 

5. Properly infinite elements in non-simple C *-algebras

The study of non-simple purely infinite algebras requires basics on majorization

in the sense of J. Cuntz and properly infinite elements as defined by M. Rørdam.
Some needed properties of the majorization relation b - c are considered here.
Others are listed and studied in Appendix A, see e.g. the Lemma A.6.1.
Our Definition 1.2.1 becomes more transparent and applicable by [462,
thm. 4.16]. It says that non-zero C *-algebras A are purely infinite in the sense of
Definition 1.2.1, if and only if, each element a ∈ A+ is properly infinite in the sense
of following Definition 2.5.1. We outline a proof of this result, because the behavior
of properly infinite elements play a fundamental role in proofs of our main results.
Versions of Cu(A) have been defined also before!!? Refer to
this Defs.!!
Compare also Definition 2.4.1 for the relation -, ...
There exist other Defs of infinite or properly infinite elements
of C *-algebras.
Even on the level elements are sometimes diferent.
Only in case of elements in s imple C *-algebras
all the definitions are equivalent ...

Definition 2.5.1. Let a, b, c elements of a C *-algebra A.

 5. PROPERLY INFINITE ELEMENTS IN NON-SIMPLE C *-ALGEBRAS 203

We say that c majorizes b (denoted by b - c ) if there exist sequences of

elements dn , en ∈ A with b = limn en c dn ( 31 ). It is easy to check that for b ∈ A+
and c ∈ A holds b - c if and only if (b − εn )+ - c for a zero-sequence (εn ) in
(0, kbk).
Refer here to Appendix Section for - (continuity) properties.
The relation - is transitive and becomes “reflexive” with respect to the as
follows defined equivalence relation ” ≈ ” :

S
We write a ≈ b if a - b and b - a. The ≈-class of a ∈ n Mn (A) ⊂ M∞ (A) will
be denoted also by [a]≈ , or simply by [a] if it can not mixed up in some places with
S
the considerably smaller class [a]M vN of elements in n Mn (A) that are Murray–
von-Neumann equivalent to a. The definition shows that the ≈-classes of elements
in M∞ (A) are (relatively) closed subsets, and that the set of b ∈ M∞ (A) with b ≈ a
are closed sets (relatively to M∞ (A)).
Then a partial order [b] ≤ [a] can be defined by b - a.
There are natural embeddings a ∈ A 7→ a ⊕ 0 ∈ M2 (A) and – more generally
S
– Mm (A) ⊕ 0n ⊂ Mm+n (A), consider them as a common ∗-subalgebra n Mn (A)
of M∞ (A) ⊇ A ⊗ K(`2 ). In this way the ≈-classes become an pre-ordered abelian
semi-group with addition [a] + [b] := [a ⊕ b] and the relation [a] ≤ [b] is compatible
with this addition.
We denote the corresponding pre-ordered semi-group by W(A), and call it the
Cuntz semigroup of A. In case of non-unital A the semigroup W(A) can be
bigger than the “small” local Cuntz semigroup CS(A) of Definition 2.4.3 and is
for all stably finite σ-unital C *-algebras A smaller than the in Definition 2.4.2
defined (large) general Cuntz semi-group Cu(A), because Cu(A) contains always
an “infinite” absorbing idempotent, cf. the remarks below Definition 2.4.3. If A is
unital, then obviously W(A) = CS(A).
The element a ∈ A absorbs the element b ∈ A, if a majorizes a ⊕ b, i.e., if
a ⊕ b - a ⊕ 0 in M2 (A) (The matrix a ⊕ b := diag(a, b) ∈ M2 (A)+ denotes the
diagonal 2 × 2-matrix with diagonal entries a and b).
Let I(a) ⊆ A denote the set of b ∈ A that are absorbed by a , i.e.,

I(a) := IA (a) := {b ∈ A ; [b] + [a] ≤ [a] } .

We write I(a) if it should be clear which C *-algebra A is considered in that moment.

Notice that the natural semigroup morphism IB (a) → IA (a) for a ∈ B ⊂ A is not
necessarily injective. It was denoted by J(a) in [462, def. 3.11].
Check if notation is consistent to other use
of I(a), J(a), IA (a), ... etc.
Give an example where IB (a) → IA (a) for some a ∈ B ⊂ A is not injective.

31 The sequences (d ) and (e ) are here not necessarily required to be bounded.

n n
204 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Then a ∈ A is called finite if I(a) = {0}, and infinite if I(a) 6= {0}.

We say that a ∈ A is properly infinite, if a ∈ I(a), i.e., if a ⊕ a - a ⊕ 0 in
M2 (A), here we allow only a 6= 0 ( 32 ), i.e., a ∈ A is properly infinite if and only
if 2[a] ≤ [a] in the pre-ordered Abelian semi-group Cu(A) of Cuntz equivalence
classes in A ⊗ K with [a] corresponding naturally to the class of a ⊗ p11 in Cu(A),
cf. Definition 2.4.2.

There exist several other formulations of infiniteness of elements or of Next

Remark 2.5.2 is used in the proofs of Part(viii) of Lemma 2.5.3 and of Lemma
2.5.4.

Remark 2.5.2. If d ∈ A+ and b ∈ D := dAd then b - d in D and A. But

from b - d in A one gets only that b ∈ J(d), where J(d) denotes the closed ideal
of A generated by d.
Indeed, the existence of elements en , fn ∈ D with b = limn en d fn follows
from the equation D = D · d · D, that is implied by the equations D = d · D and
D = D · D ( 33 ).
Moreover, this shows that b - c if c ∈ A is properly infinite in A and b is
contained in the closed ideal of A generated by c ∈ A .

Obviously, the Murray–von-Neumann equivalence a ∼M vN b of Definition 2.0.1

implies a ≈ b.
Recall that a positive element a ∈ B+ is stable by Definition 2.1.1 if the here-
ditary C *-subalgebra aBa is stable (or zero). By Part (viii) of the following Lemma
2.5.3, every non-zero stable element is properly infinite. But properly infinite ele-
ments, e.g. projections, that are not stable.
Concerning several changes of references:
Observe changes in order of items of Lemma 2.5.3:
(ix) to (vii), (x) to (viii), (xi) to (ix), (xii) to (x), (xiii) to
(xi), ... ,
and then in (vii) to (xii) to (i), (viii) to (xiii) to (ii),
(xiv) to (iii), and (xv) to (iv), ...

Go to all places where Lemma 2.5.3 is cited, and check if ref’s are
corrected !!!
Give ref. to Def. of properly infinite "element"s
and infinite C*-algebras.
Ref.s and Cite bring ORDER !!! Is it 2.5.1 for elements ??

32We allow 0 to be a properly infinite element on some places to simplify notations!

33 Here D · D it the set of products {ab ; a, b ∈ D}, but is equal to the closed linear span of

this products, cf. Lemma 2.1.7(o).

 5. PROPERLY INFINITE ELEMENTS IN NON-SIMPLE C *-ALGEBRAS 205

Lemma 2.5.3. Let I(a) := IA (a) denote the set of elements b ∈ A that are
“absorbed” by a ∈ A, i.e., b ∈ IA (a) if and only if b ⊕ a - a inside M2 (A), i.e.,
[b] + [a] ≤ [a] in W(A) .
Have we here only to consider W(A) or also the bigger Cu(A)?

(i) IA (a) is a closed ideal of A that satisfies IA (a) = IA (c) for all c ∈ A with
a ≈ c.
In particular, if a ∈ A+ and b∗ = b ∈ M2 (IA (a)) then |b| ∈ M2 (IA (a))
and, for all d, e ∈ A, |b| + [d, e]∗ a[d, e] - a in M2 (A) ( 34 ).
(ii) If D ⊆ A is a hereditary C*-subalgebra of A and if a ∈ D, then ID (a) =
IA (a) ∩ D.
(iii) The element πIA (a) (a) = a + IA (a) is always finite in A/IA (a) .
(iv) Every C*-morphism ϕ : A → B maps IA (a) into IB (ϕ(a)).
In particular, if J / A is a closed ideal and if πJ (a) is finite in A/J,
then IA (a) ⊆ J , ( 35 ).
(v) If J is a closed ideal of A and J ⊆ IA (a), then πJ (IA (a)) = IA/J (πJ (a)).
In particular, a ∈ A is properly infinite in A, if and only if, πJ (a) is
infinite in A/J for every closed ideal J / A with a 6∈ J.
(vi) If a, b ∈ A+ are orthogonal, then I(a) + I(b) ⊆ I(a + b).
(vii) If a ∈ A+ , then a ≈ a + b for all b ∈ I(a)+ .
In particular, if a ∈ A+ is properly infinite then the elements a + x
are properly infinite for each positive element x ∈ span(AaA) .
(viii) If d ∈ A+ is non-zero and the hereditary C*-subalgebra D := dAd of
A has the property that M(D) contains isometries s1 , s2 ∈ M(D) with
orthogonal ranges, i.e., s∗j sk = δj,k 1, then d is properly infinite in A.
In particular, if d ∈ A+ \{0} is stable in the sense of Definition 2.1.1,
i.e., D := dAd is stable, then d is properly infinite in A.
(ix) An element a ∈ A+ is properly infinite, if and only if, for every ε > 0,
there exists x = x(ε) ∈ M1,2 (A) such that x∗ ax = (diag(a, a) − ε)+ =
(a − ε)+ ⊗ 12 in M2 (A) .
(x) If 0 6= a ∈ A+ and there exists δ > 0 such that (a − ν)+ is properly infinite
for every ν ∈ (0, δ), then a is properly infinite in A.
(xi) If a ≈ b and a is properly infinite, then b is properly infinite.
In particular, the elements a, a∗ a, aa∗ and (a∗ a)1/2 are all properly
infinite if one of them is properly infinite.

Proof. We use in the proof of Part (i) the precise notation X ⊕0m ∈ Mn+m (A)
if X ∈ Mn (A), but in the proofs of other parts we write simply X for X ⊕ 0m .
(i): More generally we define In := In (a) for a ∈ A as the set of x ∈ Mn (A)
with a⊕x - a⊕0n in Mn+1 (A). In particular I1 (a) = IA (a). Below given arguments
show that the In are closed ideals of Mn (A) and In = Mn (IA (a)) for n ∈ N. Clearly

34 Identify here a with a ⊗ p

11 in A ⊗ M2 , or with the matrix [aj,k ] given by a1,1 := a and
aj,k = 0 if j 6= 1 or k 6= 1 .
35 Case I (a) = {0} and I
A A/J (πJ (a)) = A/J can appear, cf. Examples 2.5.12 and 2.5.13.
206 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

we get the same subset of Mn (A) if we consider the elements x ∈ Mn (A) with the
property x ⊕ a - 0n ⊕ a in Mn+1 (A) .
In the case where x ∈ Mn (A)+ and a ∈ A+ this is equivalent to (x − γ − ε)+ ⊕
(a − ε)+ - a for all ε ∈ (0, γ) and γ ∈ (0, kxk), because it implies (x − γ)+ ⊕ a - a
for every γ ∈ (0, kxk) and then the relation x ⊕ a - a. Both implications follow
from the continuity properties of the relation - on M2 (A), respective from the fact
that In (a) is closed in Mn (A), as shown now in more detail:
Temporarily and only here, we denote by ξ(a ⊕ 0n ) the set of elements z ∈
Mn+1 (A) with z - (a ⊕ 0n ). We write this as z - a via identifying A with
A ⊗ e11 ∼
= A ⊕ 0n . Likewise, we can use here instead the identification of A with
A ⊗ en+1,n+1 ∼
= 0n ⊕ A in Mn+1 (A).
By Lemma A.6.1(iv,viii), the set ξ(a ⊕ 0n ) is a closed subset of Mn+1 (A) that
satisfies Xξ(a ⊕ 0n )Y ⊂ ξ(a ⊕ 0n ) for all X, Y ∈ M(Mn+1 (A)) ∼ = Mn+1 (M(A)).
More precisely, ξ(a⊕0n ) is the closure in Mn+1 (A) of the set of elements X(a⊕0n )Y
with X, Y ∈ M(Mn+1 (A)) ∼ = Mn+1 (M(A)).
The sets In (a) are closed subsets of Mn (A) because

a ⊕ In (a) = (a ⊕ Mn (A)) ∩ ξ(a ⊕ 0n ) .

This equation shows also that In (a) invariant under right and left multiplications
by operators in M(Mn (A)).
In particular, IA (a) = I1 (a) ⊆ A and the sets In (a) ⊆ Mn (A) are closed in
Mn (A) and satisfy cIn (a)d ⊆ In (a) for all c, d ∈ M(Mn (A)).
We derive that if x ∈ Im (a) and y ∈ In (a), then x ⊕ y ∈ Im+n (a) :
Notice that in Mm+n+1 (A) for all v ∈ Mm (A) and z ∈ Mn (A) holds

(a ⊕ v) ⊕ z ≈ a ⊕ (v ⊕ z) ≈ a ⊕ (z ⊕ v) ≈ (a ⊕ z) ⊕ v ,

where the middle equivalence is given by 1M (A) ⊕ S for some suitable inner auto-
morphism S of Mn+n . The elements x ∈ Im (a) ⊆ Mm (A) and y ∈ In (a) ⊆ Mn (A)
satisfy a ⊕ x - a ⊕ 0m in Mm+1 (A) and a ⊕ y - a ⊕ 0n in Mn+1 (A).
It follows that

a ⊕ (x ⊕ y) ≈ (a ⊕ x) ⊕ y - (a ⊕ 0m ) ⊕ y

in Mm+n+1 (A). Then

(a ⊕ 0m ) ⊕ y ≈ (a ⊕ y) ⊕ 0m - (a ⊕ 0n ) ⊕ 0m ≈ a ⊕ 0m+n .

Together it gives that

a ⊕ (x ⊕ y) - a ⊕ 0m+n
in Mm+n+1 (A), i.e., x ⊕ y ∈ Im+n (A). We can also use the relations

a ⊕ x - a ⊕ 0m , a ⊕ (0m ⊕ y) - a ⊕ 0m+n

and Lemma A.6.1(ii) to get the relation:

a ⊕ (x ⊕ y) ≈ (a ⊕ x) ⊕ y - (a ⊕ 0m ) ⊕ y - a ⊕ 0m+n .
 5. PROPERLY INFINITE ELEMENTS IN NON-SIMPLE C *-ALGEBRAS 207

Since Mm+1 (A) is naturally isomorphic to the hereditary C *-subalgebra D of b ∈

Mm+n+1 (A) with b = c ⊕ 0n for some c ∈ Mm+1 (A), and since a ⊕ 0m ⊕ 0n and
a ⊕ x ⊕ 0n are both in D, we get that x ∈ Mm (A) is in Im (a), if and only if,
x ⊕ 0n ∈ Im+n (a), cf. Lemma A.6.1(viii).
If x, y ∈ In , then a ⊕ (x + y) ⊕ 0n - a ⊕ (x ⊕ y) - a ⊕ 02n in M2n+1 (A) , because
(x + y) ⊕ 0n - x ⊕ y in M2n (A), cf. Lemma A.6.1(x).
Thus, In (a) is also additively closed, i.e., is a closed ideal of Mn (A). In par-
ticular there is a unique ideal J of A such that In (a) = J ⊗ Mn ∼ = Mn (J) and this
ideal is determined by the equation J ⊗ e11 = (A ⊗ e11 ) ∩ In (a).
Then I(a) ⊕ 0n−1 = In (a) ∩ (A ⊕ 0n−1 ) by the above gives definitions of In (a) ⊆
Mn (A) and I(a) ⊆ A . Combined with the natural isomorphism A ⊕ 0n−1 ∼ = A ⊗ e11
it gives that In (a) = Mn (I(a)).
Alternatively one could use here that I(a) ⊕ 0n−1 is a full hereditary C *-
subalgebra of the ideal In (a) ⊆ Mn (A), to obtain that In (a) = Mn (I(a)).
For all elements X ∈ In (a) = Mn (I(a)) holds X ⊕ a ≈ a ⊕ X - a ⊕ 0n inside
Mn+1 (A).
If a ≈ c and x ∈ I(a), then x ∈ I(c), because

c ⊕ x - a ⊕ x - a ⊕ 0 - c ⊕ 0.

It implies then In (a) = In (c) for all n ∈ N.

It holds X + Y - X ⊕ Y for all X, Y ∈ Mn (A) by Lemma A.6.1(x), and
|Z| = Z 2 ≈ Z for all selfadjoint Z ∈ In (a). It implies that |Z| ⊕a - a if Z ⊕a - a.
2

Thus, |b| ⊕ [d, e]∗ a[d, e] - |b| ⊕ a - a for each a ∈ A+ and b∗ = b ∈ M2 (I(a)).
(ii): Clearly, ID (a) ⊆ IA (a) ∩ D by definitions of IA (a) and ID (a) in case of
a ∈ D. If x ∈ IA (a) ∩ D, then a ⊕ x - a ⊕ 0 in M2 (A). Since M2 (D) is hereditary
in M2 (A) and a ⊕ x, a ⊕ 0 ∈ M2 (D), it follows x ∈ ID (a) by Lemma A.6.1(viii).
(iii): The observation in Part (iii) is equivalent to [462, lem. 3.13]. We give an
alternative proof:
Let a ∈ A+ and denote by π := πI(a) : A → A/I(a) the quotient map, where
here I(a) := IA (a) .
Let x ∈ A+ with π(x) ⊕ π(a) - π(a), i.e., with π(x) ∈ I(π(a))+ . Here the
short notation I(π(a)) means – more precisely – the ideal IA/I(a) (π(a)) of A/I(a),
and π(x) denotes the element πI(a) (x) of A/I(a).
We show below that this implies π(x) = 0, i.e., show that x ∈ IA (a). It follows
then that I(π(a)) = 0, which means that the element π(a) is finite in A/I(a).
Since I(a) is a closed ideal of A, it suffices to show that (x − γ)+ ∈ I(a) for
each γ ∈ (0, kxk), i.e., that (x − γ)+ ⊕ a - a .
208 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

By definition of - in A/I(a), for each ε ∈ (0, γ) there exist δ := δ(ε) > 0 and
d, e ∈ A with

[π(d), π(e)]∗ π((a − δ)+ )[π(d), π(e)] = (π(x) − γ − ε)+ ⊕ (π(a) − ε)+ .

We can rewrite this equivalently as “Y ∈ M2 (I(a))” for the selfadjoint element

Y ∈ M2 (A) defined by

Y := (x − γ − ε)+ ⊕ (a − ε)+ − [d, e]∗ (a − δ)+ [d, e] ∈ M2 (I(a)) .

It follows that

(x−γ −ε)+ ⊕(a−ε)+ = Y +[d, e]∗ (a−δ)+ [d, e] ≤ |Y |+[d, e]∗ (a−δ)+ [d, e] ∈ M2 (A) .

Since Y ∈ M2 (I(a)) we get |Y | ∈ M2 (I(a)) and by Part (i), this implies that
|Y | + [d, e]∗ (a − δ)+ [d, e] - a .
Thus (x − γ − ε)+ ⊕ (a − ε)+ - a in M2 (A). This happens for every ε ∈ (0, γ),
and implies (x − γ)+ ⊕ a - a, i.e., (x − γ)+ ∈ I(a) for each γ ∈ (0, kxk). Thus
x ∈ I(a), because IA (a) is closed.
(iv): By Lemma A.6.1(iii), if a ⊕ x - a ⊕ 0 and

ψ := id ⊗ϕ : M2 ⊗ A → M2 ⊗ B ,

then ψ(a ⊕ x) - ψ(a ⊕ 0) in M2 (B). Thus, ϕ(IA (a)) ⊆ IB (ϕ(a)) ⊆ B.

(v): It is – up to suitable notation – an almost trivial consequence of parts (iii)
and (iv). Let J ⊆ I(a) a closed ideal. Define K := πJ (IA (a)) = IA (a)/J / A/J ,
L := IA/J (a + J) / A/J, B := A/IA (a) and b := πIA (a) (a) ∈ B (To inflate not
iterated indices).
Notice that IB (b) = 0 by Part (iii), i.e., b := πIA (a) (a) is finite in B. By Part
(iv) we get K ⊆ L in A/J.
Now, under natural identification of (A/J)/K with B := A/IA (a), we get
πK (πJ (x)) = πIA (a) (x) for all x ∈ A. Thus, Part (iv) says also that

πK : A/J → (A/J)/K ∼
=B

maps L := IA/J (πJ (a)) into

IB πK (πJ (a)) = IB πIA (a) (a) = IB (b) = {0} .

It implies that also L ⊆ K. Now L = K rewrites as πJ (IA (a)) = IA/J (πJ (a)).
(vi): If a, b ∈ A+ are orthogonal, and x ∈ I(a), y ∈ I(b), then (a+b)⊕0 ∼ a⊕b,
and a ⊕ b ⊕ x ⊕ y - a ⊕ b, i.e., x ⊕ y ∈ I(a ⊕ b) = I2 (a + b) – with notations in proof
of Part(i). Then x + y ∈ I(a + b) follows finally from (x + y) - x ⊕ y.
(vii): a ⊕ a - a implies IA (a) ⊇ span AaA . Thus, (a + x) ⊕ 0 - a ⊕ x - a ⊕ 0
for positive x ∈ span AaA . This and a ≤ a + x imply a ∼ a + x.
(viii): Let 0 6= d ∈ A+ a stable element in A. Then D ∼= D ⊗ K for D := dAd
by Definition 2.1.1. Then M(D) contains isometries s1 , s2 ∈ M(K) ⊆ M(D) with
orthogonal ranges, i.e., s∗1 s2 = 0.
 5. PROPERLY INFINITE ELEMENTS IN NON-SIMPLE C *-ALGEBRAS 209

Let d ∈ A+ a non-zero positive element, D := dAd and suppose that s1 , s2 ∈

M(D) are isometries with s∗1 s2 = 0. Thus, d ∼M vN dk := sk ds∗k = gk∗ gk for
gk := d1/2 sk and k ∈ {1, 2}, because gk gk∗ = d .
The element d1 + d2 = d ⊕s1 ,s2 d ∈ D is Cuntz-equivalent in M2 (A) to
diag(d, d) = d ⊕ d. Hence diag(d, d) - d1 + d2 ∈ D in M2 (D). Since d is a
strictly positive element of D we get that d ⊕s1 ,s2 d - d by Remark 2.5.2.
Thus, d ⊕ d - d ⊕ 0 in M2 (D). Hence d is properly infinite in D and in A in
the sense of Definition 2.5.1.
(ix): If a ∈ A+ , then a ⊕ a - a, if and only if, for every ε > 0, there exists
x ∈ M1,2 (A) such that x∗ ax = ((a ⊕ a) − ε)+ , cf. Lemma A.6.1(vi).
(x): ((a ⊕ a) − ν)+ = (a − ν)+ ⊕ (a − ν)+ - (a − ν)+ ≤ a for all ν ∈ (0, δ)
implies a ⊕ a - a by Lemma A.6.1(vii).
(xi): b ⊕ b - a ⊕ a - a - b if a ≈ b and a is properly infinite. Thus b ∈ I(b)
by Lemma A.6.1(ii,i). It applies to |a| = (a∗ a)1/2 ≈ a ≈ a∗ a ∼ aa∗ by Lemma
A.6.1(v). 

MORE of ABOVE? OR LESS?

A rather practical elementary sufficient criteria of – not necessarily proper –
infiniteness is contained in the following lemma:

Lemma 2.5.4. An element 0 6= a ∈ A+ is infinite in A

Where is "infinite" defined? Give Reference !!
if there exists non-zero positive c ∈ D := aAa with the property that, for each
ε > 0, there exists b ∈ D+ and d ∈ A with
More BETTER And Clear DETAILS !!!
c ⊕ a - a (depending on ε) with bc = 0 and d∗ bd = (a − ε)+ .
In particular, a ≥ 0 is infinite in A if there exists a non-zero projection p ∈ aAa
that is infinite in A .

Notice that the element c ≥ 0 is untouched here – in the sense that the d can
be chosen that cd = 0 –, because we can replace b by b1/3 and d by b1/3 d. Recall
that the notation x ⊕ y denotes the the 2 × 2 matrix aij with entries a1,2 = 0 = a2,1 ,
a1,1 = x and a2,2 = y.
MORE details !!!!! ???

Proof. Let D := aAa and non-zero c ∈ D+ with proposed property: For each
ε ∈ (0, kak) there exist b ∈ D+ with bc = 0 and d ∈ A with d∗ bd = (a − ε)+ .
Then c+b ∈ D+ implies that c+b - a, cf. Remark 2.5.2, and c⊕b = g ∗ g ∼M vN
gg ∗ = c + b ⊕ 0 in M2 (D) ⊆ M2 (A), by the row matrix g ∈ M2 (D) with entries
g11 := c1/2 , g12 := b1/2 and g2,1 = 0 = g22 .
210 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

The equation d∗ bd = (a − ε)+ implies then, that c ⊕ (a − ε)+ - c ⊕ b. Using

that ∼M vN implies ≈, we get that c ⊕ (a − ε)+ - a ⊕ 0 in M2 (A) for each ε > 0.
But this says that c ⊕ a - a ⊕ 0 in M2 (A) by lower semi-continuity of -. Thus,
c ∈ IA (a)+ and a is infinite in A by Definition 2.5.1 and Lemma 2.5.3(i).
Now we consider the case of a ∈ A+ with the property that there exists an
inside A infinite projection 0 6= p ∈ D := aAa . We show below that this implies
the existence a non-unitary isometry T ∈ M(D), with 1 − T T ∗ ∈ D. It follows
that c := (1 − T T ∗ )a(1 − T T ∗ ) 6= 0 and b := T a1/3 T ∗ , d := T a1/3 satisfy bc = 0,
d∗ c = 0 = cd and d∗ bd = a ( 36 ).
By Definition 2.5.1 and Lemma 2.5.3 the projection p ∈ D is infinite if ID (p) 6=
0, i.e., there exists nonzero b ∈ ID (p)+ with b ⊕ p - p. We can suppose here that
kbk = 1 and b ∈ pAp ⊆ D. Then b ⊕ p - p implies that there exists G ∈ M2 (D)
with
kG∗ diag(p, 0)G − diag(p, b)k < 1/2 .
We apply Lemma 2.1.9 to this inequality and get a contraction d ∈ M2 (D) with
(Gd)∗ diag(p, 0)(Gd) = diag((p − 1/2)+ , (c − 1/2)+ ) . This rewrites as a row f :=
[f1 , f2 ] := diag(p, 0)Gd ∈ M1,2 (D), and the equations f1∗ pf1 = (1/2)p, f1∗ pf2 = 0
√
and f2∗ pf2 = (b − 1/2)+ . In particular, v := 2pf1 ∈ D is a partial isometry with
v ∗ v = p and g := pf2 f2∗ p 6= 0. Then gv = 0 = v ∗ g, vv ∗ ≤ p and c ≤ q := p − vv ∗ ≤
p.
It gives that q 6= 0 and that p − q = vv ∗ , v ∗ v = p. The operator T :=
(1M(D) − p) + v ∈ 1 + D is an isometry with T ∗ T = 1 and 0 6= 1 − T T ∗ = q ∈ D. 

Next: General criteria of infiniteness. Not urgent!

Question 2.5.5. What about the necessary direction of Lemma 2.5.4 for in-
finiteness?
Let a ∈ A+ with kak = 1 and D := aAa. Consider D as a C *-subalgebra of
D∞ := `∞ (D)/c0 (D) by the map b ∈ D 7→ (b, b, . . .) + c0 (D).
Is a infinite in A if a is infinite in D∞ ?
Is a infinite in A if a is infinite in the ultrapower Aω ?
Suppose that a ∈ A+ is infinite in A. Does there exist in D∞ := `∞ (D)/c0 (D)
a (fixed) positive c ∈ D∞ with kck = 1 such that for each ε > 0 there exists d ∈ D∞
with d∗ c = 0 and d∗ d = (a − ε)+ ?
Can we find in (D∞ )+ positive contractions b, c ∈ A∞ := `∞ (A)/c0 (A) with
bc = 0, and for every ε > 0 some (bounded) dε ∈ A∞ with d∗ bd = ψ((a − ε)+ ) for
ψ : A → A∞ defined by ψ(a) := (a, a, . . .) + c0 (A)?
An attempt is the following: Suppose that non-zero a ∈ A+ is infinite in A, i.e.,
IA (a) 6= 0 by Definition 2.5.1 and Lemma 2.5.3(i). Take g ∈ IA (a)+ with kgk = 1.
36 It shows also that there exists q = (1 − T T ∗ ) ∈ pDp with a - (1 − q)a(1 − q) in A. So q can

play the role of c – instead the inside qAq invertible qaq –, but needs an approximation argument
... an exercise for the reader.
 5. PROPERLY INFINITE ELEMENTS IN NON-SIMPLE C *-ALGEBRAS 211

Then g ⊕ a - a by definition of - in Definition 2.5.1, i.e., for every sufficiently

small ε > 0 there exist δ ∈ (0, ε) and d1 , d2 ∈ A with d∗1 (a − δ)+ d1 = (a − ε)+ ,
d∗2 (a − δ)+ d2 = (g − 1/2)+ and d∗2 (a − δ)+ d1 = 0. We could consider cε := (a −
1/2 1/2
δ)+ d2 d∗2 (a − δ)+ , and bε := ((a − δ)+ d1 d∗1 (a − δ)+ )1/4
1/3
?? Perhaps power better?? Check it!
1/2
The polar decomposition of (a − δ)+ d1 is equal to BZ with a partial isometry
Z ∈ A∗∗ and B := b2ε . We can take dε := bε · Z . Get bε · cε = 0, d∗ε bε dε = (a − ε)+ ,
cε ≤ kd2 k2 · (a − δ)+ and
kbε k2 = kak − ε ???,
kcε k = k(g − 1/2)+ k.
We only know that δ ∈ (0, ε) ... no reasonable functional dependence. ?????

The cases of pi(n) are similar to pi(1)?

Notice that for all positive elements X, Y in a C *-algebra A with kXk = kY k =
1 and X ∈ (Y − 1/2)+ A(Y − 1/2)+ holds that X(Y − (Y − 1/2)+ ) = (1/2)X and
for Z := 2(Y −(Y −1/2)+ )−2(X −(X −1/2)+ ) that kZk = 1 and Z(X −1/2)+ = 0.
If (X − 1/2)+ is in the ideal generated by Z, and if each element of this ideal is in
the closure of the set of products A · Z · A (not its linear span), then Z is infinite.

Corollary 2.5.6. For (non-zero) C*-algebras A following properties (1)-(4)

are equivalent:

(1) A is purely infinite in sense of Definition 1.2.1.

(2) A satisfies property pi (1) of Definition 2.0.4.
(3) A satisfies property pi-1 of Definition ??.
(4) Every non-zero element of A+ is properly infinite in A in sense of Defi-
nition 2.1.1.

Proof. (3)⇔(4): The definitions of property pi-n in Definition ?? imply in

case n = 1 that A satisfies pi-1, if and only if, each element of A is properly infinite
in the sense of Definition 2.1.1.
(4)⇒(2): (Compare Lemma 2.1.2 for more details.) Suppose that every non-
zero element a ∈ A+ is properly infinite in A in sense of Definition 2.1.1.
Then it is easy to see that each non-zero element of `∞ (A)+ is properly infinite,
and that this property pass to all non-zero quotients of `∞ (A). In particular, `∞ (A)
can not have a non-zero character, because 1 ∈ C is not infinite in C.
If a is properly infinite and b is in the closed ideal generated by a then there
Pm
exist for ε > 0 elements c1 , d1 , . . . , cm , dm ∈ A with kb − `=1 ck adk k < ε/2 . The
proper infiniteness of a causes that there exists g, h ∈ M1,m (A) with

k1m ⊗ a − g ∗ (a ⊕ 0m−1 )hk < ε/(2m2 (max(kck k + kdk k))) .

k

Thus, there exist e, f ∈ A with kb − eaf k < ε. It shows that A satisfies property
pi(1) of Definition 2.0.4.
212 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

(2)⇒(1): The property pi(1) in Definition 2.0.4 requires from A that `∞ (A) has
no non-zero character, and that, for each element a ∈ A+ and c ∈ A+ in the closed
ideal generated by a and for each ε > 0, there exists d ∈ A with d∗ ad = (c − ε)+ .
It implies immediately that A is purely infinite in sense of Definition 1.2.1,
because this requires only that A has no non-zero character and that for every
a ∈ A+ and every positive c in the closed ideal generated by a, and every ε > 0,
there exists an element d ∈ A such that kc − d∗ adk < ε .
(1)⇒(4): Suppose that A has no character and that for every non-zero a ∈ A,
every b ∈ A in the closed ideal J(a) ⊆ A generated by a, and every ε > 0 there
exists d, e ∈ A (depending from a, b, ε) with kdae − bk < ε. It it easy to reformulate
this property with help of polar decomposition and Lemma 2.1.9 as follows:
The C *-algebra A has no character and, for every positive contractions in a, b ∈ A+ ,
with b in the closed ideal J(a) ⊆ A generated by a and ε > 0, there exists δ > 0
and d ∈ A with d∗ (a − δ)+ d = (b − ε)+ .
The properties of A carry over to each non-zero quotient A/K of A by a closed
ideal K of A, because if A has no non-zero character then A/K can not have a
non-zero character, and if e, f ∈ (A/K)+ , f ∈ J(e) in A/K and ε ∈ (0, kf k) given,
then there exist a ∈ A+ with e = πK (a) and elements g1 , . . . , gn ∈ A such that
πK (b) = (f − ε/2)+ for b := g1∗ ag1 + . . . + gn∗ agn ∈ J(a). There exists d ∈ A and
δ > 0 with d∗ (a − δ)+ d = (b − ε/2)+ . It says that (f − ε)+ = πK (d)∗ (e − δ)+ πK (d)
for given ε ∈ (0, kf k) and suitable δ ∈ (0, kek).
By Part (v) of Lemma 2.5.3, non-zero elements a ∈ A are properly infinite, if
and only if, πJ (a) is infinite for each closed ideal J of A with a 6∈ J.
Above we have seen that non-zero quotients A/K satisfy the same properties
as A . Therefore it suffices to show in general that each non-zero element of A+ is
infinite if A is purely infinite in sense of Definition 1.2.1.
If D ⊆ A is a non-zero hereditary C *-subalgebra of A then D can not have a
character χ : D → C:
Indeed: Suppose that a non-zero character χ on D exists and let I ⊆ D the
kernel ideal of χ.
Let J ⊆ A the closed ideal J of A generated by the kernel I of χ. Then
the ideal J has the property that C ∼
= D/I ⊆ A/J is a 1-dimensional hereditary
C *-subalgebra of A/J, because J ∩ D = I by Lemma ??.
Let p ∈ D/I ⊆ A/J the unique projection with C·p = D/I, then every element
in the closed ideal K of A/J generated by p is itself in C · p. Thus, J must be the
kernel of a character on A. But this is forbidden for A in Definition 1.2.1.
It follows that each non-zero hereditary C *-subalgebra D of A has no character.
It implies by Lemma 2.1.15(ii) that the D contains a 2-homogenous element g :=
ψ(f0 ⊗ 12 ) for some non-zero *-morphism ψ : C0 ((0, 1], M2 ) → D with kgk = 1.
 5. PROPERLY INFINITE ELEMENTS IN NON-SIMPLE C *-ALGEBRAS 213

Now let a ∈ A+ non-zero and kak = 1. The hereditary C *-subalgebra D :=

(a − 1/2)+ A(a − 1/2)+ contains a 2-homogenous element g := ψ(f0 ⊗ 12 ) with
kgk = 1.
Let c := ψ((f0 − 1/2)+ ⊗ p22 ) ∈ D, d := 2ψ((f0 − (f0 − 1/2)+ ) ⊗ p22 , and
e := 2(a − (a − 1/2)+ ) − d. Then dc = c = cd, da = d = ad, a is in the ideal of aAa
generated by e, and ec = 0.
There exists a pure state ρ : A → C with ρ(a) = 1. Let dρ : A → L(H) the
corresponding irreducible representation and x ∈ L(H) with kxk = 1 and
ρ(a) = hdρ x, xi
Perhaps it is easier to show first that (a − 1/2)+ A(a − 1/2)+ can not have a
character
By Lemma ??
NO-NO! need here a new criteria for infiniteness! Which one? ???
Part (v) of Lemma 2.5.3
to show that a (respectively πJ (a)) is infinite.
Something like:
?????????????? PROOF? 

There are places, e.g. Lemma 2.6.8(i) - or ?? -, with results

partly similar to the in Lemma 2.5.7 considered!

Lemma 2.5.7. Let p, q ∈ B projections and b ∈ B+ . If (b ⊕ q) - (p ⊕ 0) in

M2 (B), then for each 0 < ε < min(kbk, 1) there exists an element d = d(ε) ∈ B and
a partial isometry z = z(ε) ∈ B – depending both on ε –, such that for r := z(z ∗ )
holds r ≤ p, z ∗ z = q, (p − r)d = d and d∗ d = (b − ε)+ .
Special cases are:

( i) If, moreover, b is a projection then this implies that b ⊕ q is in M2 (B)

MvN-equivalent to a sub-projection of p ⊕ 0, i.e., there are projections
p1 , p2 ∈ B such that p1 p2 = 0, p1 + p2 ≤ p, p1 ∼M vN b and p2 ∼M vN q .
( ii) If p, q ∈ B are projections that satisfy p - q then there exists a projection
r ≤ q that is Murray–von-Neumann equivalent to p, i.e., there exists a
partial isometry v ∈ B with v ∗ v = p and vv ∗ =: r ≤ q.
In particular, p ≈ q, if and only if, there exist partial isometries
v, w ∈ B such that v ∗ v = q, vv ∗ ≤ p, w∗ w = p and ww∗ ≤ q ( 37 ).
(iii) If B is unital and 1 := 1B , then there exists 0 6= c ∈ B with c ⊕ 1 - 1, if
and only if, B contains a non-unitary isometry.
And this is equivalent to I(1) 6= {0} by the definition of the ideal I(1)
given in Lemma 2.5.3.

37Compare Remark 2.5.8.

214 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

(iv) The unit 1 := 1B is properly infinite in B, – in the sense that there exists
isometries S, T ∈ B with S ∗ T = 0 –, if and only if, 1 ⊕ 1 - 1 ⊕ 0 in
M2 (B), if and only if, 1 ∈ I(1), if and only if, I(1) = B .

Proof. Recall that b ≈ b∗ b ≈ (b∗ b)1/2 . It implies that the relation b ⊕ q - p is

equivalent to (b∗ b)1/2 ⊕ q - p. Thus, we can suppose that b is positive in all later
considered cases.
The relation b ⊕ q - p ⊕ 0 implies by Lemma A.6.1(iv) that for each ε > 0 there
exists δ = δ(ε) > 0 and a matrix T ∈ M2 (B) such that
(b − ε)+ ⊕ (q − ε)+ = ((b ⊕ q) − ε)+ = T ∗ ((p ⊕ 0) − δ)+ T .

Since ε < min(kbk, 1), it follows that the left side is non-zero, and therefore
δ < 1. Thus,
(b − ε)+ ⊕ (1 − ε)p = (1 − δ)S ∗ (p ⊕ 0)S ,
where S := [x, y] is the first row of T (and we can suppose that px = x and
py = y). We get that x∗ py = 0 and S ∗ (p ⊕ 0)S = diag(x∗ px, y ∗ py). It implies that
(b − ε)+ = (1 − δ)x∗ px and (1 − ε)p = (1 − δ)y ∗ py We define a partial isometry by
z := z(ε) := (1 − ε)−1/2 (1 − δ)1/2 py
and an element d := d(ε) ∈ B by
d := (1 − δ)1/2 px .
Straight calculation shows that z and d have the proposed properties.
Proof of the special cases:
(i): If b is a projection and ε < 1/2 then let p1 := r = z(ε) · z(ε)∗ and
p2 := (1 − ε)−1 d(ε) · d(ε)∗ . This works because b = (1 − ε)−1 d(ε)∗ d(ε) .
(ii): It if moreover b = 0, then (i) says p1 = 0 and that r := p2 ⊆ p and
v := (1 − ε)−1/2 d(ε) satisfy v ∗ v = q ≤ p and vv ∗ = r.
(iii): The definition of the ideal I(1) ⊆ B with 1 := 1B , given in Lemma 2.5.3,
says that c ∈ I(1) if and only if c ⊕ 1 - 1. Thus, I(1) 6= {0}, if and only if, there
exists non-zero c ∈ B with c ⊕ 1 - 1 .
If there exists 0 6= c ∈ B with c ⊕ 1 - 1, then b ⊕ 1 - 1 for b := c∗ c, because
b ≈ c.
Let q := 1 , p := 1 , b := c∗ c and ε := kbk/2 in the general case. Then
there exists elements d ∈ B and z ∈ B with z ∗ z = 1, d∗ d = (b − kbk/2)+ and
(1 − zz ∗ )d = d. It follows that z is a non-unitary isometry in B.
If B contains a non-unitary isometry z ∈ B, then c ⊕ 1 - 1 for c := 1 − zz ∗ 6= 0,
because c⊕1 ≈ c⊕zz ∗ and 1−zz ∗ ⊕zz ∗ = R∗ (1⊕0)R for the row R := [1−zz ∗ , zz ∗ ] ,.
Thus, (1 − zz ∗ ) ⊕ 1 - 1 .
(iv): The subset I(1) ⊆ B is a closed ideal of B by Lemma 2.5.3(i). Thus,
1 ∈ I(1) , if and only if, B = I(1) . By Definition of the ideal I(1) of B, 1 ∈ I(1) ,
if and only if, 1 ⊕ 1 - 1 ⊕ 0 in M2 (B).
 5. PROPERLY INFINITE ELEMENTS IN NON-SIMPLE C *-ALGEBRAS 215

The latter 1 ⊕ 1 - 1 is the case of Part (i) with b := 1, q := 1 and p := 1.

By Part (i), there are projections p1 , p2 ∈ B such that p1 p2 = 0, p1 + p2 ≤ 1,
p1 ∼M vN 1 and p2 ∼M vN 1 . It says that there exists elements S, T ∈ B with
S ∗ S = 1 = T ∗ T with S(S ∗ ) = p1 and T (T ∗ ) = p2 . This are isometries with
S ∗ T = S ∗ (SS ∗ )(T T ∗ )T = S ∗ p1 p2 T = 0. 
Remark 2.5.8. The relations v ∗ v = q, vv ∗ ≤ p, w∗ w = p and ww∗ ≤ q in
Part(ii) of Lemma 2.5.7 do not imply W-vN equivalence: This can be seen in stable
simple purely infinite C *-algebras: The W-vN equivalence classes of projections in
B ⊗ K correspond bijective to the elements of K0 (B) but all non-zero projections
in p, q ∈ B ⊗ K satisfy p ≈ q.
Corollary 2.5.9. Let A a unital C*-algebra. Then 1A is properly infinite in
A, if and only if, A/J contains a non-unitary isometry for each closed ideal J 6= A
of A.

Proof. Part (v) of Lemma 2.5.3 says that 1 is properly infinite in A, if and
only if, πJ (1) = 1A/J ∈ A/J is infinite in A/J for each closed ideal J 6= {0} of A.
By Part (iii) of Lemma 2.5.7, 1A/J is infinite in A/J, if and only if, there exists a
non-unitary isometry in A/J. 
Remark 2.5.10. The closed ideal IA (a) is always contained in the closed ideal
J(a) of A generated by a, and J(a) = IA (a) if and only if a is properly infinite.
Then J(a) is equal to the closure of n A · (a∗ a − 1/n)+ · A , i.e., of the set of
S

products b · (a∗ a − 1/n) · c , with b, c ∈ A ( 38 ).

The Definition 2.0.4 of property pi(n) immediately allows to see that it implies
that each non-zero (n + 1)-homogenous element is properly infinite in A if A has
property pi(n). But the Corollary 2.7.18 of Proposition 2.7.16 gives the better result
that all non-zero n-homogenous elements are properly infinite in A with property
pi(n). It implies that `∞ (A) has no irreducible representation on a Hilbert space of
dimension ≥ n, and that property pi(n) passes to `∞ (A) if A has property pi(n).
Property pi-n in Definition ?? implies property pi(m) for some m ≤ n, but
no explicit lower bound for m is known. The non-trivial direction is, that each
C *-algebra with property pi(m) has property pi-n for some n ≥ m, but it seems
that some property of the ideal system of the multiplier algebra plays a certain role.
For real-rank zero C *-algebras they are all the same and say that each non-zero
projection is infinite.
Some sensible property of the definitions of infiniteness of elements can be seen
in the C *-subalgebra A ⊆ C([0, 1], O2 ) of operator-valued functions
f : [0, 1] → O2 := C ∗ (sk ; s∗k sk = 1 = s1 s∗1 + s2 s∗2 )
with f (1) ∈ M2∞ ⊆ O2 . There f ∈ A+ is infinite if and only if f (1) = 0. The –
inside A – properly infinite element f given by the function f (t) := (1 − t)s1 s∗1 ∈
38 But it seems that the latter property does not imply always that a is properly infinite.

Examples for that?

216 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

C0 ([0, 1), O2 ) has the property that no element g ∈ A+ with kgk ≤ 1 and gf = f
can be infinite in A.
Pm
Lemma 2.5.11. Let b := k=1 ak with a1 , . . . , am ∈ A+ .

If the elements ak ⊗ 1n are all properly infinite in Mn (A), i.e., if ak ⊗ 1n+1 -

ak ⊗ 1n for k = 1, . . . , n, then b ⊗ 1m·n is properly infinite in Mm·n (A).

Proof. Notice that b - (a1 ⊕ . . . ⊕ am ) ≤ b ⊗ 1m in Mm (A), and that (a1 ⊕

. . .⊕am )⊗1n+1 - (a1 ⊕. . .⊕am )⊗1n in Mm·n (A), because the ak ⊗1n are properly
infinite in Mn (A).
Induction over k ∈ N gives that (a1 ⊕ . . . ⊕ am ) ⊗ 1k+n - (a1 ⊕ . . . ⊕ am ) ⊗ 1n .
Gives b ⊗ 1k+n - b ⊗ 1mn . for all k ∈ N. In particular, b ⊗ 11+(mn) - b ⊗ 1mn ,
i.e., b ⊗ 1mn is properly infinite. 

Example 2.5.12. There exist a closed ideal J / A and a ∈ A+ with J ⊆ D :=

aAa, such that D is stably finite and I(d) = {0} for all d ∈ D (hence A is stably
finite), but πJ (a) = a + J is properly infinite in A/J.
Indeed: By an “opposite version” of a theorem of Glimm in [438, cor.1.4(v)]
there exists a hereditary C *-subalgebra D of the CAR-algebra B := M2∞ such
that the unital two-sided normalizer algebra A := N (D, B) ⊆ B of D in B satisfies
A/D = O2 . Then a := 1 ∈ A satisfies that the C *-subalgebra aAa = A ⊆ B is
stably finite, i.e., I(a ⊗ 1n ) = {0} for all n ∈ N . But πJ (a) is the properly infinite
unit of O2 for J := D / A .
e of C0 ((0, 1], O2 ) is a
Example 2.5.13. The unit element 1 of the unitization A
stably finite element of A.
e

More generally, the unit element 1A of the “join” algebras

A := Emax (B, C) ⊆ C([0, 1], B ⊗max C)
of unital algebras B and C, or its quotient
A := E(B, C) ⊆ C([0, 1], B ⊗min C)
is stably finite if 1n ⊗ 1B finite in Mn ⊗ B for all n ∈ N.
Even more generally, 1A is infinite in A if and only if 1A = T ∗ T for some
non-unitary isometry T ∈ A, cf. Lemma 2.6.8(i).

In the next lemma we can replace the pair (ε, 2ε) with 0 < ε < kak/2 obviously
by any pair of positive reals 0 < δ < γ < kak , simply by passing from a to f (a) with
a suitable strictly increasing continuous real function f ∈ C0 (0, kak]+ on [0, kak]
with f (0) = 0 .

Lemma 2.5.14. Let B a C*-algebra, a ∈ B+ non-zero, ε ∈ (0, kak /2) and

denote by J the closed ideal that is generated by the two-sided annihilator
Ann((a − 2ε)+ , B) := { b ∈ B ; b (a − 2ε)+ = 0 = (a − 2ε)+ b }
of (a − 2ε)+ in B.
 5. PROPERLY INFINITE ELEMENTS IN NON-SIMPLE C *-ALGEBRAS 217

(i) If (a − ε)+ is in J then J = B, i.e., then Ann((a − 2ε)+ , B) is a full

hereditary C*-subalgebra of B.
(ii) The hereditary C*-subalgebra Ann((a − 2ε)+ , B) is full in B if there exists
b ∈ B that satisfies the equations:

b∗ (a − 2ε)+ = 0 and b∗ b = (a − ε)+ . (5.1)

If, moreover, the elements of Ann((a − 2ε)+ , B) are properly infinite, then
conversely the fullness of Ann((a − 2ε)+ , B) implies the existence of b ∈ B
that satisfies Equations (5.1).
More generally, Ann((a − 2ε)+ , B) is full in B if there exist elements
u, v ∈ B and δ > 0 such that (vv ∗ − δ)+ = (a − 2ε)+ , u∗ (vv ∗ − δ)+ = 0,
and [u] ≥ [v] in Cu(B).
(iii) If J 6= B then B/J is unital, and πJ ((a − ε)+ ) is invertible in B/J with
spectrum in [ ε, kak − ε ].
In particular, πJ (a) is invertible in B/J, if and only if, πJ (a) 6= 0.

Proof. Let ε, µ > 0. Define fµ (t) := min{ 1, (t/µ − 1)+ } for t ∈ [0, ∞), i.e.,
fµ (t) := 0 for t ≤ µ, fµ (t) := t/µ − 1 for t ∈ [µ, 2µ] and fµ (t) := 1 for t ∈ [2µ, ∞).
Notice that µfµ (t) ≤ max(t − µ, 0) and (1 − fµ (t)) max(t − 2µ, 0) = 0.
(i,iii): Let a 6= 0, 0 < ε < kak/2 and let J denote the closed ideal J of B,
generated by Ann((a − 2ε)+ , B).
The ideal C ∗ (a) ∩ J of the C *-subalgebra C ∗ (a) ⊆ B contains the closed ideal
K of C ∗ (a) that is generated by (1 − fε (a)) · C ∗ (a) because fε (t) = 1 for t ≥ 2ε.
It follows that the non-zero values of the spectrum of πJ ((a−ε)+ ) are contained
in the spectrum of πK ((a − ε)+ ) in C ∗ (a)/K. The latter is contained in [ε, kak − ε]
if 2ε < kak.
The elements (1 − fε (a))a and fε (a) − fε (a)2 are contained in K ⊆ J. It follows
that P := πJ (fε (a)) is a projection in B/J.
The equation (1 − fε (a))(a − 2ε)+ = 0 implies that b − fε (a))b ∈ J for all b ∈ B,
because

(1 − fε (a))B(1 − fε (a)) ⊆ Ann((a − 2ε)+ , B) ⊆ J .

Thus, P is the unit element of B/J if J 6= B, and then εP ≤ πJ ((a − ε)+ ) ≤

(kak − ε)P . We get that πJ ((a − ε)+ ) and πJ (a) are invertible in B/J with inverses
of norms ≤ ε−1 , respectively ≤ (2ε)−1 .
The equation J = B is equivalent to (a − ε)+ ∈ J because that latter causes
P = 0.
(ii): If a ∈ J = B then there exist b1 , . . . , bn ∈ B with bk b∗k (a − 2ε)+ = 0 and
Pn ∗
k=1 bk bk = (a − ε/2)+ .
218 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

If the elements of Ann((a − 2ε)+ , B) are properly infinite, then it follows that
there exists γ > 0 and d1 , d2 ∈ B such that
n
X
(d1 bk d2 )∗ (f − γ)+ (d1 bk d2 ) = (a − ε)+
k=1
Pn ∗
for the element f := k=1 bk bk of Ann((a − 2ε)+ , B).
By assumption on Ann((a − 2ε)+ , B), the element f is properly infinite. It
implies that there exist g1 , . . . , gn ∈ B such that gj∗ f gk = δjk (f − γ)+ . The element
Pn
b := k=1 f 1/2 gk d1 bk d2 satisfies bb∗ (a − 2ε)+ = 0 and b∗ b = (a − ε)+ .
If there exists b ∈ B with bb∗ (a−2ε)+ = 0 and b∗ b = (a−ε)+ , then (a−ε)+ ∈ J.
It implies J = B by Part (i). (This particular observation do not use an assumption
on infiniteness of elements in Ann((a − 2ε)+ , B).)
Suppose, more generally, that there exist elements u, v ∈ B and δ > 0 such
that (vv ∗ − δ)+ = (a − 2ε)+ , u∗ (vv ∗ − δ)+ = 0, and [u] ≥ [v] in Cu(B).
Then vv ∗ - u∗ u, and there exist d ∈ B with d∗ u∗ ud = (vv ∗ − δ/2)+ ∼M vN
(v ∗ v −δ/2) . The element b := ud satisfies b∗ (vv ∗ −δ)+ = 0 and b∗ b = (vv ∗ −δ/2)+ .
Thus Ann((vv ∗ − δ)+ , B) = Ann((a − 2ε)+ , B) is full in B, because we can take
(vv ∗ , δ/2) in place of (a, ε) in our former considerations.
In case where Ann((a − 2ε)+ , B) is full and purely infinite, we can find the
above considered element b and can define u := b, v := b∗ , δ := ε. They satisfy the
requirements on v, w and δ, because b∗ (b∗ b − ε)+ = b∗ (a − 2ε) = 0 and [b] ≥ [b]. 

Does there exists a ‘‘relative’’ variant

for some spectral p.i. elements in e ∈ J??

Lemma 2.5.15 ([462]). Let a ∈ A+ with A = aAa , and let J / A a closed

ideal, such that πJ (a) = a + J is properly infinite in A/J. Then:

(i) For b, c1 , c2 , c3 ∈ A+ holds c1 ⊕ c2 ⊕ c3 - b ⊕ 0 ⊕ 0 in M3 (A), if and only

if, for every ε > 0, there are f1 , f2 , f3 ∈ bA and γ > 0 such that

fj∗ (b − γ)+ fk = δj,k (ck − ε)+ for j, k = 1, 2, 3 .

(ii) For every ε > 0,there exist elements a1 , a2 ∈ A+ , e ∈ J+ and n ∈ N, such

that a1 a2 = 0, (a1 + a2 )e = 0, and (a − ε)+ - ak ⊕ (e ⊗ 1n ) for k = 1, 2.
(iii) If each 0 6= e ∈ J+ is properly infinite, then a is properly infinite.

Proof. We give here a proof that contains more details than in [462].
(i): Let c := c1 ⊕ c2 ⊕ c3 and ε > 0. By Lemma A.6.1(vi), there are F ∈ M3 (A)
and δ > 0 with (c − ε)+ = F ∗ ((b − δ)+ ⊕ 0 ⊕ 0)F .
We define fk := ξ(b)F1,k for k ∈ {1, 2, 3}, where ξ ∈ C0 (0, 1 + kbk] is given by
ξ(t) := min(1, (2/δ) max(0, t − δ/2)) .
Then fk ∈ bA and the row G := [f1 , f2 , f3 ] ∈ M1,3 (A) ⊂ M3 (A) satisfies

(c − ε)+ = F ∗ ((b − δ)+ ⊕ 02 )F = G∗ ((b − δ)+ ⊕ 02 )G .

 5. PROPERLY INFINITE ELEMENTS IN NON-SIMPLE C *-ALGEBRAS 219

(ii): Let ε ∈ (0, kπJ (a)k) and µ := ε/3. Since πJ (a) ⊗ 13 - πJ (a) by proper
infiniteness of πJ (a), and since always a ≈ a2 , there exist g1 , g2 , g3 ∈ A/J and δ > 0
with gi∗ πJ ((a − δ)+ )gj = δi,j πJ ((a − µ)+ ) . If we let xk := πJ ((a − δ)+ )1/2 gk , then

(xk x∗k − µ)+ ∼ (x∗k xk − µ)+ = πJ ((a − 2µ)+ ) ,

in A/J, and there is a (unique) C *-morphism ϕ : C0 (0, 1] ⊗ M3 → A/J with

ϕ(f0 ⊗ pij ) = xi (x∗j ) . The C *-algebras C0 (0, 1] ⊗ Mn are projective by Proposition
A.8.4 in the sense of Definition A.8.1 in Appendix A . It implies that there exists a
C *-morphism ψ : C0 (0, 1] ⊗ M3 → A with πJ ◦ ψ = ϕ , cf. also Lemma 2.1.15 and
Remark 2.1.16 concerning the projectivity of C0 ((0, 1], M ), for C *-algebras M of
finite dimension.
Let bk := ψ(f0 ⊗ pkk ), for k = 1, 2, 3 and take yk ∈ A with πJ (yk ) = xk . Then
b1 ∼ b2 ∼ b3 , bj bk = 0 for j 6= k, and define

zk := yk∗ (bk − µ)+ yk − (a − 2µ)+ · (a − µ)+ ∈ J .

Thus, (a − 2µ)+ - (bk − µ)+ ⊕ v for k = 1, 2 and v := |z1 | + |z2 | . It follows, that
there exists γ > 0 such that, for k = 1 and k = 2, holds

(a − ε)+ - (bk − µ)+ ⊕ (v − γ)+ .

The annihilator D := Ann((b1 + b2 − µ)+ , A) is a hereditary C *-subalgebra of

A that contains b3 . It implies that b1 + b2 is in the closed ideal generated by D,
and that D is full by Lemma 2.5.14. Since D is full, also J ∩ D is full in J , i.e.,
span(J(D ∩ J)+ J) is dense in J .
It follows that there exist d1 , . . . , dn ∈ J and e ∈ (D ∩ J)+ with j d∗j edj =
P

(v − γ)+ . In particular, (v − γ)+ - e ⊗ 1n .

(iii): Let ε > 0, a1 , a2 , e ∈ A+ and n ∈ N as in Part (ii). Then

(a − ε)+ ⊕ (a − ε)+ - a1 ⊕ a2 ⊕ (e ⊗ 12n ) .

Since e is properly infinite, and since a1 , a2 , e are pairwise orthogonal, we get:

((a ⊗ 12 ) − ε)+ - a1 ⊕ a2 ⊕ e ∼ a1 + a2 + e - a ,

because a is a strictly positive element of A. Since ε > 0 is arbitrary, we obtain

a ⊗ 12 - a. 

Definition 2.5.16. We say that 0 6= b ∈ A+ is spectral properly infinite if

(b − t)+ is properly infinite in A for each t ∈ (0, kbk). (See Section 1 or Definition
2.5.1 of properly infinite elements.)

Lemma 2.5.17. Let A a C*-algebra and ε > 0.

(o) Each b ∈ A+ \ {0} is spectral properly infinite, if and only if, A is a purely
infinite C*-algebra.
(i) If b ∈ A+ is spectral properly infinite in sense of Definition 2.5.16, then
elements d1 , d2 ∈ A with d∗j bdk = δj,k (b − ε)+ can be found with norm-
bound kdj k2 ≤ 2 kbk /ε .
220 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

(ii) If the positive part A+ of a C*-algebra A contains a dense subset X ⊆ A+

such that each b ∈ X is spectral properly infinite, then A is purely infinite,
i.e., each non-zero a ∈ A+ is properly infinite.

Proof. (o): By definition, A is purely infinite if each non-zero element a ∈ A+

is properly infinite. In particular (b − t)+ is properly infinite in A for each b ∈ A+
and t ∈ (0, kbk).
(i): Let b ∈ A+ spectral properly infinite, and ε = 2δ ∈ (0, kbk). There
exist g1 , g2 ∈ A with gj∗ (b − δ)+ gk = δjk · (b − 2δ)+ because (b − δ)+ is properly
1/2
infinite. Then dk := b−1/2 (b − δ)+ gk is in A and kdk k2 ≤ δ −1 k(b − 2δ)+ k , because
δ(t − δ)+ ≤ t(t − δ)+ for t ∈ (0, ∞), and A+ 3 a 7→ kgk∗ a gk k ∈ R+ is order
monotone.
(ii): Let a ∈ A+ with kak = 1 and ε > 0. There exists b ∈ X with kb − ak <
γ := ε/3 . It implies k(b − 2γ)+ − ak < 3γ, because k(b − 2γ)+ − bk ≤ 2γ.
By Lemma 2.1.9 there exists contractions d1 , d2 ∈ A with d∗1 ad1 = (b − γ)+
and d∗2 (b − 2γ)d2 = (a − 3γ)+ .
By assumption, each b ∈ X is spectral properly infinite, thus (b−γ)+ is properly
infinite for each γ ∈ (0, kbk), i.e., (b − γ)+ ⊕ (b − γ)+ - (b − γ)+ , and we find
e1 , e2 ∈ A with e∗j (b − γ)+ ek = δjk (b − 2γ)+ . It follows that ck := d1 ek d2 satisfies
c∗j ack = δjk (a − 3γ)+ . Thus, each a ∈ A+ is properly infinite and A is purely
infinite. 

Example 2.5.18. Let a1 , a2 , . . . ∈ K := K(`2 (N))+ defined by an :=

diag(1, (n + 1)−1 , (n + 1)−2 , . . .) then each an is properly infinite, and the sequence
(a1 , a2 , . . .) is decreasing with norm-limit (1, 0, 0, . . .) in K. Thus, convergent
sequence of of properly infinite contractions have in general no properly infinite
limit.

Are they really properly infinite?

Does there exist dk ∈ K with

diag(1, 1, 1/n, 1, n, 1/n2 , 1/n2 , . . .) = lim d∗k diag(1, 1/n, 1/n2 , . . .)dk
k

in norm?
(1, 1/n, 1/n2 , . . .) - (1, 1/n2 , 1/n4 , 1/n6 , . . .) ⊕ (1/n, 1/n3 , 1/n5 , 1/n7 . . .)
≈ (1, 1/n, 1/n2 , 1/n3 , . . .) ⊕ (1, 1/n, 1/n2 , 1/n3 , . . .) via dk := ak ⊕ bk ak :=
(1, n1/2 , n, 1/n6 , ...)
The above technical lemmata on the properties of the Cuntz-relation - and ≈
yield almost immediately the following permanence properties of the class of purely
infinite C *-algebras.
Notice that A ⊗ C0 (0, 1] is not in this list.

Proposition 2.5.19 (Permanence of pure infiniteness). The class of (not neces-

sarily simple) purely infinite C*-algebras A is invariant under following operations:
 5. PROPERLY INFINITE ELEMENTS IN NON-SIMPLE C *-ALGEBRAS 221

(i) passage to quotients A/J,

(ii) inductive limits,
(iii) infinite direct products `∞ (A1 , A2 , . . .) and ultrapowers Aω ,
(iv) passage to hereditary C*-subalgebras D = DAD,
(v) Morita equivalence, and
(vi) extensions.

Proof. Recall that A is purely infinite, if and only if, for each a ∈ A,

a ∈ I(a) := {b ∈ A ; b ⊕ a - a} ,

and recall that I(a) is a closed ideal of A by Lemma 2.5.3(i).

(i): If a ∈ A \ J is properly infinite in A, i.e., a ∈ I(a), then πJ (a) = a + J
is properly infinite in A/J, i.e., πJ (a) ∈ I(πJ (a)) because πJ (I(a)) ⊆ I(πJ (a)) by
Lemma 2.5.3(v). Thus, each quotient A/J by a closed ideal J 6= A of A is purely
infinite if the C *-algebra A is purely infinite.
(ii): If { hσ,τ : Aσ → Aτ }τ,σ∈T is a directed net of C *-morphisms,

A := indlimτ →T (hσ,τ : Aσ → Aτ )

is the (canonical) inductive limit of C *-algebras, then A is generated by the upward

directed net of the images of hσ,∞ : Aσ → A. Thus, A contains an upward directed
net of C *-subalgebras hσ,∞ (Aσ ) that are quotients of purely infinite C *-algebras.
By Part (i), this C *-subalgebras are purely infinite. The elements of the images
build together a dense *-subalgebra of A. The elements (a∗ a − t)+ in the positive
part of this *-subalgebra are properly infinite. By Lemma 2.5.17(ii) this implies
that A is purely infinite.
(iii): Let A1 , A2 , . . . a sequence of purely infinite C *-algebras, an ∈ An positive
contractions (possibly an = 0), and ε > 0 .
(n) (n)
For n ∈ N with an 6= 0 there exist by Lemma 2.5.17(i) elements d1 , d2 ∈ An
with
(n)
kdk k2 ≤ 2kan k /ε
and
(n) (n)
(dj )∗ an dk = δj,k (an − ε)+ .
(1) (2) (n)
Let a := (a1 , a2 , . . .) and dk := (dk , dk , . . .) (k = 1, 2), where we put dk = 0 if
an = 0. Then dk ∈ `∞ (A1 , A2 , . . .) and d∗j adk = δj,k (a − ε)+ . Hence, each positive
contraction in `∞ (A1 , A2 , . . .) is properly infinite.
The ultrapower Aω is purely infinite by Part (i), because it is a quotient of
`∞ (A).
(iv): If D is a hereditary C *-subalgebra of A, a ∈ D and a is properly infinite
in A, i.e., a ∈ IA (a), then a is properly infinite in D, i.e., a ∈ ID (a), because
ID (a) = IA (a) ∩ D by Lemma 2.5.3(ii). Hence, D is purely infinite if A is purely
infinite.
222 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

(v): Clearly, a C *-algebra B is purely infinite if there exists a purely infinite

C *-algebra D and a *-isomorphism from D onto B,
Two C *-algebras B1 and B2 are Morita equivalent (in the category of C *-
algebras), if there exists a C *-algebra A and *-monomorphisms ϕk : Bk → A with
k ∈ {1, 2} such that Dk := ϕk (Bk ) are full hereditary C *-subalgebras of A. Here
“full ” means that A is the closure of the ideal in A generated by Dk .
We know from Part (iv) that a hereditary C *-subalgebra D of A is purely
infinite if A is purely infinite.
It follows that the invariance of pure infiniteness under Morita equivalence is
equivalent to the following statement:
C *-algebras A are purely infinite if they contain a full hereditary C *-subalgebra D
that is purely infinite.
We show more generally that a purely infinite hereditary C *-subalgebra D of
A generates a purely infinite closed ideal J of A.
Let D a purely infinite hereditary C *-subalgebra of A and let J := span(ADA)
the ideal J of A generated by D.
Let a ∈ J+ ⊆ A+ and ε > 0. There exist b1 , . . . , bn ∈ D+ and f1 , . . . , fn ∈ A
with
n
X
k fk∗ bk fk − ak < ε/2 .
k=1

Pn
It follows for b := b1 + · · · + bn ∈ D+ that a ≤ ε/2 + k=1 fk∗ bfk , calculated in
Pn
A + C1 ⊆ M(A). Let γ := k k=1 fk∗ fk k and ν := ε/(2 + 3γ).
By assumption on D, b ∈ D+ is properly infinite in D. Thus, b ⊗ 1n - b
inside D, and there exist e1 , . . . , en ∈ A with e∗j bek = δjk (b − ν)+ . The element
Pn
g := k ek dk satisfies g ∗ bg = k=1 d∗k (b − ν)+ dk and a ≤ ε/2 + g ∗ bg.
P

By Lemma 2.1.9, there exists a contraction h ∈ A such that (a − ε)+ = z ∗ z for

P 1/2
z := ( k (b − ν)+ dk )h.
Since zz ∗ ∈ D and z ∗ z = (a − ε)+ we get [zz ∗ ] = [z ∗ z] = [(a − ε)+ ] ≤ [a] and
[zz ] + [zz ∗ ] ≤ [zz ∗ ] ≤ [a]. It follows that [(a − ε)+ ] + [(a − ε)+ ] ≤ [a] for all ε > 0.
∗

But this is equivalent to [a] + [a] ≤ [a], i.e., to a ⊕ a - a in J, and a is properly

infinite in J. This shows that for each a ∈ J+ \ {0}, and shows that J is purely
infinite.
Alternatively expressed: Take a ∈ J+ and verify that for each ε ∈ (0, kak) the
element (a − ε)+ is M-vN-equivalent to some element in D+ . The use that proper
infiniteness is invariant under M-vN equivalence.
(vi): Let J / A a purely infinite closed ideal with purely infinite quotient A/J,
and let a ∈ A+ .
We can replace A by its hereditary C *-subalgebra D := aAa . Then still D ∩ J
and D/(D ∩ J) ∼= πJ (a)(B/J)πJ (a) are purely infinite by Parts (i) and (iv). Then
 5. PROPERLY INFINITE ELEMENTS IN NON-SIMPLE C *-ALGEBRAS 223

aDa is dense in D and πJ (a) is properly infinite in D/(D ∩ J). Thus, Lemma
2.5.15(iii) applies and a is properly infinite in A .
This applies to each a ∈ A+ and shows that A is purely infinite. 

We underline that some proofs of similar permanence properties for the classes
of weakly purely infinite and of strongly purely infinite C *-algebras are much more
involved. The latter class of C *-algebras is relevant for our classification. Up to
now (2019) it is unknown if this three infiniteness properties (locally, weakly, pi(1))
are different or not, even in the case of separable nuclear C *-algebras. It is also
unknown if C([0, 1], A) p.i. if A is p.i.

Question 2.5.20. Suppose that A is a C *-algebra with properly infinite unit

element and let a ∈ A a contraction.
Does there exist isometries s, t ∈ A with 3ks∗ atk < 2 ?
(The latter is the “squeezing property” of Definition 4.2.14.)

Compare next with Proof of Prop. 4.2.15(b) and with Prop. ??!!
New references give different results??
The property that for each contraction a ∈ A there exist isometries s, t ∈ A
with 3ks∗ atk < 2 is equivalent to the “squeezing” Property (sq) for A, cf. Definition
4.2.14.
It is easy to see that it holds for all unital strongly purely infinite C *-algebras,
because it is true for all contractions a ∈ A, if and only if, the positive 2 × 2-matrix
M := [bjk ] with entries b11 := b22 := 1 and b∗21 = b12 := a can be approximately
diagonalized with help of a diagonal matrix D := diag(d1 , d2 ) up to ε ∈ (0, 1/2),
i.e.,
k 12 − D∗ M D k < ε .

The Lemma 4.2.13 allows to show that all unital purely infinite (not necessarily
simple) C *-algebras A have the squeezing property, cf. Part (b) of Proposition
4.2.15.
The Property (sq) alone does not imply that a properly infinite unital C *-
algebra A is purely infinite, e.g. take A := M(B) for B := C([0, 1], K) or B :=
M2∞ ⊗ K. In the case B := M2∞ ⊗ K the corona Q(B) has Property (sq) and is
residually antiliminary, but is not weakly p.i..
Question concerning Property (sq):
What happens in case of “properly” infinite non- p.i. simple unital A ? Examples?
Relations to (Q.1) in real rank zero case?
Optional one could attempt to use
--e.g. in case of linearly ordered ideal lattice --
following Definition 2.5.21.
But it could be that it is equivalent to ‘‘residually antiliminary’’
224 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Definition 2.5.21. We call a C *-algebra B “locally infinite” if for every

pure state ρ on B and each hereditary C *-subalgebra D of B with kρ|Dk = 1
there exists a contraction d := d(D, ρ) ∈ D with ρ(d∗ d) = 1, ρ(dd∗ ) = 0 and
(dd∗ − 1/3)+ ≤ d∗ d.

Where is here really ”infiniteness” ??? What is it in case of simple C*-algebras

with RR=0?
Definition 2.5.21 implies:
A “locally infinite” C *-algebra B is residual antiliminary in the sense of
Definition 2.7.2, because the property of “local infiniteness”

(i) passes to each non-zero quotient and every non-zero hereditary C *-

subalgebra, and
(ii) implies that a pure state ρ on a non-zero hereditary C *-algebra D of B
can not be a non-zero character on D.

It is not difficult to see that for pure states ρ on D ⊆ B ...

Does “l.p.i.” or “pi(n)” imply this? (Both pass to hereditary C *-subalgebras
and quotients!)
If E ⊆ D is hereditary, stable and σ-unital with ρ(E) 6= {0} then there exists
an element d := d(ρ, D) ∈ E with the properties in Definition 2.5.21...

6. Cases: many projections, linear ordered ideal-lattice ...

Still to be worked out in detail!!!

Desired Main Result:
If A has the property that for each closed ideal J 6= A and any non-zero
hereditary C *-subalgebra D of A/J contains a (non-zero) projection p ∈ D, then
A is strongly purely infinite, if and only if, A is locally purely infinite, ...

Definition 2.6.1. We say that a C *-algebra A is rich of projections if, for

each closed ideal J 6= A of A and non-zero hereditary C *-subalgebra D of A/J,
there exists a non-zero projection p ∈ D.

In particular, A has then the “small projections” property (SP) that requires
that every non-zero hereditary C *-subalgebra D of A contains a non-zero projec-
tion.
Among the C *-algebras that are “rich of projections” are the C *-algebras
A with real rank zero, a property that is equivalent to the property that each
hereditary C *-subalgebra D of A contains an approximate unit of D consisting of
projections in D, cf. [73, thm. 6.5.6].
(We do not know if C *-algebras that are “rich of projections” have real rank
zero, because projections usually do not lift to projections.
 6. CASES: MANY PROJECTIONS, LINEAR ORDERED IDEAL-LATTICE ... 225

It is also not clear if every non-zero hereditary C *-subalgebra D ⊆ A contains

an approximate unit for D consisting of convex combinations of projections ... )

Lemma 2.6.2. Let A a unital C*-algebra. If for each pure state ρ of A there
exits a stable C*-subalgebra D ⊆ A with ρ(D) 6= {0} then there exists n ∈ N such
that 1A ⊗ 1n ∈ A ⊗ Mn is properly infinite in A ⊗ Mn .
If there exists a stable C*-subalgebra D ⊆ A that generates A (in the sense that
the linear span of A · D · A is dense in A), then 1A itself is properly infinite.

Proof. The algebraic union V of the upward directed family of finite sums of
stably generated ideals of A is an ideal of A that is dense in A by its definition,
because V can not be in the kernel of any pure state. It implies that 1 ∈ V .
Thus, 1 is contained in an ideal that is a finite sum of stably generated ideals,
i.e., there exists – by assumptions and Part (ii) of Lemma 2.1.7 – stable hereditary
C *-subalgebras D1 , . . . , Dn ⊂ A and elements dj ∈ A (j = 1, . . . , n) with d∗1 d1 +
. . . + d∗n dn = 1 and dj d∗j ∈ Dj .
The n ∈ N with this property can be chosen minimal with the property that
the sum of ideals generated by the Dj contains 1, because if dj d∗j and dk d∗k both
are in a closed ideal J that is generated by some stable C *-subalgebra D, then one
can find d ∈ A with dd∗ ∈ D and d∗ d = d∗j dj + d∗k dk , because M(D) contains
isometries S1 , S2 ∈ M(D) with S1∗ S2 = 0.
The stability of the Dj causes that its two-sided multiplier algebra M(Dj )
contains a copy Bj of E2n generated by 2n isometries Sjk ∈ Bj , k ∈ {1, . . . , 2n},
with mutually orthogonal ranges:
∗
Sjk Sj,` = δk,` 1j , where 1j denoted the unit element of M(Dj ).
Consider the two n × n matrices T0 := [Tj,k,0 ] and T1 := [Tj,k,1 ] in Mn (A)
with entries Tj,k,0 := Sj,k dk and Tj,k,1 := Sj,(k+n) dk for j, k ∈ {1, . . . , n} then
Tx∗ Ty = δx,y · 1n for x, y ∈ {0, 1}.
Thus, 1n ∈ Mn (A) is properly infinite.
In case that a single hereditary C *-subalgebra D generates A then there exist
d1 , d2 ∈ A with d1 d∗1 + d2 d∗2 ∈ D, d∗1 d2 = 0 and d∗1 d1 = d∗2 d2 = 1. 

Lemma 2.6.3. The class of C*-algebras A, B, . . . that are rich of projections is

invariant under following operations:

(i) passage to non-zero hereditary C*-subalgebras D ⊆ A,

(ii) passage to non-zero quotients A/J,
(iii) extensions 0 → A → E → B → 0,
(v) building inductive limits,
(iv) passage to a Morita equivalent algebras. In particular A is rich of projec-
tions, if and only if, A ⊗ K is rich of projections.

Each C*-algebra A that is rich of projections is the inductive limit of the net
of separable C*-subalgebras that are rich of projections and are relatively weakly
injective in A.
226 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Proof. to be filled in ?? 

Proposition 2.6.4. Suppose that a C*-algebra A is rich of projections in the

sense of Definition 2.6.1. Then the following properties (i)–(iii) are equivalent:

(i) For each closed ideal J 6= A and nonzero hereditary C*-subalgebra D ⊆

A/J the algebra D contains a (non-zero) infinite projection q ∈ D.
(ii) A is locally purely infinite in sense of Definition 2.0.3.
(iii) A is purely infinite, cf. Definition 1.2.1.

The C*-algebra A is strongly purely infinite in sense of Definition 1.2.2, if A

is purely infinite and has the stronger property that, for each closed ideal J of A
and each hereditary C*-subalgebra D ⊆ A/J there exists a projection p ∈ A \ J with
πJ (p) ∈ D.

We do not know if for separable C *-algebras that are “rich of projection” in

the fairly weak sense of ?????
Compare with Proposition 2.6.5!
Notice that Part (i) of Proposition 2.6.4 is just the the definition of J. Cuntz
for pure infiniteness.

Proof. Obviously (iv) implies (iii). Part (iii) is equivalent to properties pi(1)
and pi-1 on A by Corollary 2.5.6, and property pi(n) implies that A is locally purely
infinite, i.e., Part (ii).
⇐ References?
(ii)⇒(i):
(i)⇒(iii):
(iii)⇒(ii):
Still not understood... (iii)⇒(iv): Perhaps wrong ?? ?? 

From (ii) to (i):

First step:
Suppose that A is is locally purely infinite, i.e., for each non-zero b ∈ A+ and
pure state λ on E := bAb there exists a C *-morphism ψ : C0 (0, 1] ⊗ K → E with
λ ◦ ψ 6= 0.
We show that it implies that, for each closed ideal J 6= A of A, every nonzero
hereditary C *-subalgebra D ⊆ A/J and every non-zero projection p ∈ D, there
exists n ∈ N such that p ⊗ 1n is properly infinite in D ⊗ Mn . (The multiplicity n
could depend here from p, D and J.)
Let J, D ⊆ A/J given and p ∈ D ⊆ A/J a non-zero projection. We are going
to show that for each pure state ρ on the unital C *-algebra pDp = p(A/J)p, – i.e.,
with ρ(p) = 1 –, there exist a stable C *-subalgebra G ⊆ pDp with ρ(G) 6= {0}.
 6. CASES: MANY PROJECTIONS, LINEAR ORDERED IDEAL-LATTICE ... 227

By Lemma 2.6.2 it follows that the upward directed net of finite sums of stably
generated ideals of pDp must contain p and this implies that p ⊗ 1n is properly
infinite in pDp ⊗ Mn for suitable n ∈ N .
There is a nonzero contraction b ∈ A+ with πJ (b) = p and a pure state and
ρ : A/J → C with ρ(p) = 1. The state ξ := ρ ◦ πJ satisfies ξ(b) = 1 and is a pure
state on F := bAb . Since A is w.p.i., there exists a C *-morphism ψ : C0 (0, 1]⊗K →
F with ξ ◦ ψ 6= 0. Thus ρ ◦ (πJ ◦ ψ) 6= 0. Now the Lemma 2.6.2 applies, and shows
that there exists n ∈ N such that p ⊗ 1n is properly infinite in pDp ⊗ Mn .
We combine this with the assumption that A is “rich of projections” in sense
of Definition 2.6.1, ...
Let 0 6= a ∈ A+ and J ⊆ A a closed ideal of A with a 6∈ J, δ ∈ (0, kπJ (a)k),
then let E := (a − δ)+ A(a − δ)+ and D := πJ (E) ∼ = E/(E ∩ J). Then D is a
non-zero hereditary C *-subalgebra of A/J. By the general pre-assumption that A
is “rich of projections”, there exists a non-zero projection q ∈ D ⊆ A/J.
The above given arguments show that, for a pure state ρ on A/J with ρ(q) = 1
there exists a nonzero stable C *-subalgebra of qDq given by the image of a non-zero
morphism ψ : C0 (0, 1] ⊗ K → qDq = q(A/J)q with ρ ◦ ψ 6= 0. Now we can consider
the non-zero hereditary C *-subalgebra F := cAc for some positive contraction
c ∈ A+ with πJ (c) = ψ(f0 ⊗ p11 ). Then G := πJ (F ) is a hereditary C *-subalgebra
of A/J.
By assumption that ... 12.11.2018, not clear which one...
Let b ∈ E+ a contraction with πJ (b) = p and ρ a pure state on A/J with
ρ(p) = 1 and ρ(b) = kbk = 1, i.e., there is a pure state ρ1 on A/J with ρ := ρ1 ◦ πJ
and ρ1 (p) = 1.
sort best from below:
By Part (ii), ... there exists C *-morphism ψ : C0 (0, 1] ⊗ K → F := bAb ⊆ E
with ρ ◦ ψ 6= 0.
Let F := bAb ⊆ A, then πJ (F ) = pDp. By the assumption that A is locally
purely infinite there exist a non-zero C *-morphism ψ : C0 ((0, 1], K) → F with
ρ ◦ ψ 6= 0 .
Thus, πJ ◦ ψ is non-zero, and its image is a non-zero stable hereditary C *-
subalgebra of πJ (F ) ⊆ pDp.
Define f0 ∈ C0 (0, 1] by f0 (t) := t, and let H := ψ(f0 ⊗ p11 ). Then G :=
H · A · H) ⊂ F is a hereditary C *-subalgebra of A with πJ (G) = HAH ...
check again
Then the hereditary C *-subalgebras of pDp = πJ (F ) defined by πJ (G)DπJ (G)
is a stable hereditary C *-subalgebra of pDp.
It contains the nonzero hereditary C *-subalgebra generates a non-zero stable
C *-subalgebra of pDp.
228 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Has then A then the additional property that for each closed ideal J of A and
hereditary C *-subalgebra D of A with πJ (D) 6= {0} there exists a projection p ∈ D
with πJ (p) infinite or zero?
Answer unknown! The assumptions say only that that there is a non-zero
projection q ∈ πJ (D). But it is not clear under which circumstances it can be lifted
to a projection in p ∈ D ...
Then A has the property that for each pure state ρ on πJ (D) ⊂ A/J there
exists a properly infinite projection p ∈ D with ρ(p) 6= 0?
For separable C *-algebras A the prime and primitive ideals are the same,
cf. [616, Prop. 4.3.6]. Thus, if one can reduce a property to (suitably selected)
separable C *-subalgebras and pure states, then separating pure states (from zero)
is the same as separating factorial states ...
It seems that one must require also that ρ(p) 6= 0 for a given nonzero pure state
ρ on D ??
Or that p = πJ (q) comes from a projection q
No! It is inclusive in case of pure state!:
One get this, if ρ is any nonzero state on A with ρ(J) = {0} and with kernel
I ≥ J of the cyclic representation defined by ρ. If ρ|D 6= 0, then πI (D) 6= 0,
and πI (D) is a non-zero hereditary C *-subalgebra of A/I. If q ∈ A is a nonzero
projection with
with by passing to E := πI (D)

Proposition 2.6.5. If a C*-algebra A has real rank zero, then following prop-
erties of A are equivalent:

(i) A is p.i.
(ii) For every non-zero projection p ∈ A there exist partial isometries u, v ∈ A
with u∗ u = v ∗ v = p and uu∗ + vv ∗ ≤ p.
(iii) Every quotient of A is purely infinite in the sense of Cuntz, i.e., for every
hereditary C*-subalgebra D and every closed ideal J of A which does not
contain D, there is a (non-zero) infinite projection in D/(D ∩ J).
(iv) A is locally purely infinite.
(v) A∞ has no non-trivial l.s.c. quasi-trace.

Compare with Proposition 2.6.4!

See [93, thm. 4.17] for the equivalence of (iii) and (iv). The equivalence of
(i), (ii) and (iii) follows from Corollaries 2.15.7(ii) and 2.15.8. Recall here that
an infinite projection p is not necessarily properly infinite. It can be shown to be
properly infinite only if all πJ (p) are infinite or zero for all closed ideals J of A.

Proposition 2.6.6. If A is a C*-algebra with linear ordered lattice I(A) of

closed ideals and A is locally purely infinite, then A strongly purely infinite.
 6. CASES: MANY PROJECTIONS, LINEAR ORDERED IDEAL-LATTICE ... 229

There exist at most one non-zero simple quotient A/J and at most one non-zero
simple closed ideal I ⊆ A .
If, in addition, A is separable (or σ-unital = A has a strictly positive element)
and has no unital quotient, then A is stable.

There are/can exist non-simple unital or stable simple quotients, and there can
be non-simple unital quotients:
Consider L(`2 (N )) or any extension of a simple separable C *-algebra by the
compact operators.

Definition 2.6.7. A C *-algebra B has the small projection property if

every non-zero hereditary C *-subalgebra D of B contains a non-zero projection.
(If this happens also for all quotients, i.e., for non-zero hereditary D ⊆ B/J,
then we say that B is rich of projections, cf. Definition 2.6.1.)
Suppose that a C *-algebra B has the small projection property. Then we say
that B “has sufficiently many locally bounded dimension functions” if for each
non-zero contraction a ∈ B+ and each non-zero projection p ∈ aBa there exists a
dimension function D : B → [0, ∞] with

0 < D(p) < ∞ and sup D((a − 1/n)+ ) < ∞ .

n∈N

We do not require that the dimension function D is lower semi-continuous on

a B+ a or bounded. In particular it may happen that D(a) = ∞.
Are above Lemma and next two Lemmas 2.6.8 and 2.6.9 used? cited? Otherwise
put it into Appendix A/B. Eg. with title “Pure infiniteness for algebras that are
rich of projections”?.
There are in next? section some study of projections in locally p.i. algebras.
Compare!!!
Recall that I(a) is the closed ideal of A defined for a ∈ A in Definition 2.5.1
and Lemma 2.5.3(i).

Lemma 2.6.8. Let A a C*-algebra.

(i) If p ∈ A is a projection, then I(p) 6= 0, if and only if, pAp contains a non-
unitary isometry, i.e., if and only if p is Murray–von-Neumann equivalent
to a proper sub-projection q ≤ p, q 6= p in A.
In particular, a projection p is infinite with respect to the family of
projections of A, i.e., there exists a non-zero projection r ∈ A such that
p ⊕ r is MvN-equivalent in M2 (A) to p ⊕ 0, if and only if, p is infinite as
an element of A+ , i.e., if there exists non-zero a ∈ A with [p] + [a] ≤ [p]
in Cu(A).
(ii) Exists also in other lemmas: If q ∈ cAc is a properly infinite projec-
tion, then c ⊕ q - c, i.e., q ∈ I(c).
230 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

(iii) Exists also in other lemmas: If p is a properly infinite projection in

A and p - c ∈ A+ , then p ∈ I(c), i.e., c ⊕ p - c. In particular c is infinite
in A.
(iv) Where is it used?: If A has the “small projection property” of Defini-
tion 2.6.7 and has “sufficiently many locally bounded dimension functions”
in the sense of Definition 2.6.7, then each non-zero element a ∈ A+ is fi-
nite, i.e., I(a) = {0}, which says for b ∈ A+ that b ⊕ a - a implies
b = 0.

Proof. (i): Let p⊕x - p with 0 6= x ∈ A+ and 0 < ε < min(kxk, 1), δ ∈ (0, ε)
and f = [f1 , f2 ] ∈ M1,2 (A) ⊆ M2 (A) with f ∗ (p − δ)+ f = (p − ε)+ ⊕ (x − ε)+ . Notice
that (p − t)+ = (1 − t)p for t ∈ [0, 1]. Then (1 − δ)f1∗ pf1 = (1 − ε)p and f1∗ pf2 = 0,
but (1 − δ)f2∗ pf2 = (x − ε)+ 6= 0. Thus, V := (1 − ε)−1/2 (1 − δ)1/2 pf1 p is a
non-unitary isometry in pAp.
Conversely, if V ∈ pAp is a non-unitary isometry, then p ⊕ (p − V V ∗ ) = f ∗ pf
for f := [V, (p − V V ∗ )].
(ii): Let c ∈ A+ and q ∈ cAc a properly infinite projection. Then the projection
q ∈ A is also properly infinite in the hereditary C *-subalgebra qAq ⊆ cAc by Part
(ii) of Lemma 2.5.3. Thus, we may suppose that c is a strictly positive element
of A. Let d := (1 − q)c(1 − q). Then d + q is also a strictly positive element of
A = cAc, and therefore, (d + q) ≈ c.
Take u, v ∈ qAq that satisfy u∗ v = 0, u∗ u = v ∗ v = q and define

f := [(d1/2 + u), v] ∈ M1,2 (A) ,

then in M2 (A) holds

(d + q) ⊕ q ≈ (d + q) ⊕ q = f ∗ f ≈ f f ∗ = d + uu∗ + vv ∗ ≤ d + q .

Thus, c ⊕ q ≈ (d + q) ⊕ q - d + q ≈ c.
(iii): If p∗ = p2 = p - c for c ∈ A+ , then there is d ∈ A with d∗ cd = (1/2)p,
and q = 2c1/2 dd∗ c1/2 is a projection in cAc with q ∼ p. If p is properly infinite (in
pAp), then q is properly infinite in cAc. Thus p ⊕ c ≈ q ⊕ c - c by Part (ii), i.e.,
p ∈ I(c).
(iv): Let a ∈ A+ and suppose that I(a) 6= {0}, i.e., that there exists non-zero
b ∈ A+ with b ⊕ a - a.
Then a ⊕ b - a implies b - a. The latter says that there exists, for each
ε ∈ (0, kbk), some γ ∈ (0, ε) and d ∈ A with d∗ (a − γ)+ d = (b − ε)+ .
Since A has the “small projection” property by assumptions in (iv), there exists
a projection q ∈ A with

0 6= q ∈ (b − ε)+ A(b − ε)+ .

In particular
q - (a ⊕ q) - (a ⊕ b) - a .
 6. CASES: MANY PROJECTIONS, LINEAR ORDERED IDEAL-LATTICE ... 231

Since q ∈ A is a projection with q - a, there are δ ∈ (0, kak), a partial isometry

v ∈ A such that vv ∗ = q and p := v ∗ v ∈ (a − δ)+ A(a − δ)+ . It follows that
a ⊕ p - a inside aAa , because
(a ⊕ p) ∼ (a ⊕ q) - (a ⊕ (b − ε)+ ) - a .
It implies that, for each ε ∈ (0, kak), there exists γ ∈ (0, kak) and d = [d1 , d2 ] ∈
M1,2 (A) with
d∗ ((a − γ)+ ⊕ 0)d = (a − ε)+ ⊕ p .
Now we apply the assumption on the (local) existence of locally bounded dimension
S
functions: There is a dimension function D : n Mn (aAa) → [0, ∞) with D(p) > 0
and



θ := sup D (a − γ)+ = sup D (a − 1/k)+ < ∞ .
γ>0 k
We get, for each ε ∈ (0, kak), that




θ ≥ D (a − γ)+ ≥ D p + D (a − ε)+ .
Since θ = sup{D((a − ε)+ ) ; ε ∈ (0, kak) } this contradicts D(p) > 0, and – by
assumptions on A – that a ⊕ b - a. 

Lemma 2.6.9. Let A denote an AW*-algebra, a Rickart algebra , or let A the

C*-subalgebra of B ∗∗ generated by the elements in the σ-up-hull of a C*-algebra B
in B ∗∗ .
We need “finiteness” of the projections and that each element a ∈ B+ has a
“support projection” pa := 1 − P , where P B = Ra for Ra = {b ∈ B ; ab = 0}.
W
Moreover we need that pa = {p(a−t)+ ; t ∈ (0, kak]} .

(i) There is no (non-zero) properly infinite element in a finite Rickart C*-

algebra.
(ii) If A is an AW*-algebra and if b ∈ A+ is properly infinite, then the support
projection pb of b is a properly infinite projection.
(iii) If, for every ε > 0, there exists δ ∈ (0, ε) and a ∈ A+ with properly infinite
support projection pa such that ??????? and (b − ε)+ ≤ a ≤ (b − δ)+ , then
b itself is properly infinite (in the AW*-algebra A).

Proof. (i): We use only that the Rickart C *-algebras and AW*-algebras A
have the “small projection” property because they have real rank zero as C *-
algebras.
And we claim that A has sufficiently many “locally bounded” dimension func-
tions if A is a finite Rickart algebra or finite AW*-algebras A. Then Part (iv) of
Lemma 2.6.8 applies.
Are the finite 2-q-traces on M2 (A)+ separating from 0
for the non-zero projections p in
finite Rickart C *-algebra or AW*-algebras A??
If A is a finite AW*-algebra, then there is a unique center-valued 2-quasi-trace
T : A+ → Z(A) on a A+ (cf. [79, p. 320]).
232 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

It extends to T2 : M2 (A)+ → Z(M2 (A)) ∼

= Z(A) .
The restrictions to (the positive parts of maximal) commutative C *-subalgebras
C of M2 (A) are faithful, because T2 (p) 6= 0 for non-zero projections p ∈ C, see [64,
Chap. 6]. Thus, 0 ≤ T2 (a) 6= 0 for 0 6= a ∈ A+ .
Let diag(a, p) - diag(a, 0) ?????
Let χ : Z(A) → C a character with χ(T (a)) = kT (a)k. The dimension function
D : M∞ (A) → [0, ∞) corresponding to the 2-quasi-trace χ ◦ T : A+ → [0, ∞) must
satisfy D(a) 6= 0. Hence, a can not be properly infinite.
Check above and below arguments for Rickart algebras again.
The Z-bimodule A is not a Z-bundle ??!!!
The fibers are not simple in general.
And [χ ◦ T ] need not to be faithful on the fibers.
(ii): The element b is properly infinite inside the Rickart C *-algebra or AW*-
algebra B := pb Apb .
Now let q ∈ Z(B) a central projection that is finite in B.
Then every element of qB is finite, because otherwise there exists
– by the consideration in proof of Part (i) ??? –
an infinite projection r in qB and q = r + (q − r) is infinite in rBr + C · (q − r).
The element qb is zero or properly infinite in qB if b is properly infinite, by
Lemma A.6.1(iii), ?????
Since qB is finite, it follows that qb = 0. Only 0 is orthogonal to b in pb Bpb .
This implies that the support projection pb is properly infinite in the AW*-algebra
A.
Same argument applies for Rickart algebras etc.
(iii): Suppose that, for every ε > 0, there exists δ > 0 and a ∈ A+ with
properly infinite support projection pa such that
?????? and (b − ε)+ ≤ a ≤ (b − δ)+ . Then

(b − ε)+ ⊗ 12 ≤ kbk(pa ⊗ 12 ) - pa ≤ p(b−δ)+ ≤ δ −1 b .

Thus, b is properly infinite, by Part (ix) of Lemma 2.5.3. 

End of Rickart-algebra considerations!!!

Cited below? ??
HERE ENDS PART 1 of Chp. 2

7. Residually antiliminary C *-algebras

We define here the class of “residual antiliminary” C *-algebras A and obtain

for this particular class of (not necessarily simple) C *-algebras A some relations
between the there formally different definitions (and classes) of pure infiniteness.
 7. RESIDUALLY ANTILIMINARY C *-ALGEBRAS 233

In a not obvious sense the class of residually antiliminary C *-algebras is the

“strict opposite” of the class of C *-algebras of type I, – that are also called “post-
liminal” C *-algebras ( 39 ).

Lemma 2.7.1. Let K(H) denote the algebra of compact operators on a Hilbert
space H, and let A ⊆ K(H) a C*-subalgebra of of K(H).
If A is irreducible on H, – in the sense that Ax is dense in H for every non-zero
x ∈ H –, then it follows that A = K(H).

Proof. The C *-subalgebras A of K(H) are generated by projections p ∈ A

of finite rank, because each T ∗ = T ∈ A ⊆ K(H) has discrete spectrum that
related to operators P ∈ A of finite rank. The transitivity of A on H of P AP in
P K(H)P shows that P AP = P K(H)P . In particular, this shows that A contains
projections of rank = 1. The transitivity of A in H implies that A contains all rank-
one projections in K(H). Since K(H) is generated as C*-algebra from its rank-one
projection, this shows that A = K(H). 

Definition 2.7.2. A (non-zero) C *-algebra A is residual antiliminary if,

for each closed ideal I 6= A of A and each non-zero b ∈ A/I the hereditary C *-
subalgebra E := b∗ (A/I)b of A/I is not commutative.
Is it not identical to the property that each non-zero hereditary
C *-subalgebra D ⊆ A of A has not a character?
(= D has not a 1-dimensional quotient-algebra)
What is perhaps different?
Here there is also a good place for my separate notes on the ultra-power prop-
erty: If c0 (A1 , A2 , ...) has no quotient of dimension = n, then `∞ (A1 , A2 , . . .) has
no quotient of dimension = n. (Same with ultra -powers.)
A C *-algebra B (as e.g. B := A/I) is antiliminary if B contains no non-
zero hereditary Abelian C *-subalgebra, i.e., b∗ Bb is not Abelian for every non-zero
b ∈ B, compare [616, sec. 6.1.1].

Notice that Definition 2.7.2 equivalently says that a C *-algebra B is (only)

antiliminary, if and only if, the hereditary C *-subalgebra b∗ Bb of B is not Abelian
for every non-zero b ∈ B, and the Definition says that a C *-algebra A is residual
antiliminary, if and only if, for every non-zero a ∈ A the hereditary C *-subalgebra
D := a∗ Aa of A has no non-zero character.
List of topics to be considered:
(0.1) Each irreducible representation of A comes fro a pure state of ρ of A. If A
is residually anti-liminary then the image dρ (A) of every irreducible representation
dρ : A → L(Hρ ) does not contain a non-zero compact operator. Here dρ is defined
by (some suitable) pure state ρ on A.

39Could be called also “residual liminal” by some intuition based on old history of math ...
234 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

(Proof: A non-zero intersection of dρ (A) with the algebra K(Hρ ) of compact

operators on Hρ must necessarily be the image dρ (I) of a non-zero closed ideal I
of A, e.g. I := (dρ )−1 (dρ (A) ∩ K(Hρ ).
If the C *-subalgebra J := K∩dρ (A) of K acts irreducible on Hρ then necessarily
J = K. Every
again an irreducible C*-subalgebra of the compact operators, because otherwise
?????????
This can only happen if dρ (A) contains all compact operators.)
(0.2) If A has the property that dρ (A) for every irreducible representation dρ
of A does not contain a non-zero compact operator, then this property passes to
all non-zero hereditary C *-subalgebras D ⊆ A of A.
In particular, then all non-zero hereditary C *-subalgebras D of A have no
characters.
And this implies that no non-zero hereditary C *-subalgebra D of A has a non-
zero quotient C *-algebra D/J with finite dimension.
(????? Fist attempt of proof ... : ????)
Suppose that d0 : D 7→ L(H) is an irreducible *-representation such that d0 (D)
contains a non-zero compact operator.
Let x0 ∈ H with kx0 k = 1, and let ρ0 denote the pure state of D given by
ρ0 (a) :=< ax0 , x0 >, and suppose that the corresponding irreducible representation
d0 : D 7→ L(H) contains nonzero compact operators in its image.
The inverse image I := (d0 )−1 (d0 (D) ∩ K(H)) of the ideal of compact operators
defines a closed ideal I of D.
The closed ideal J of A generated by I and I itself have the following properties:
(0.2. i) The set I ⊆ A is also a hereditary closed C *-subalgebra of A, because
I is hereditary in D as every closed ideal of D. (And the hereditary C*-subalgebras
have a lattice order given by its open support projections in A∗∗ .)
(0.2. ii) We show that I = J ∩D, that the support projections of the hereditary
C *-subalgebras J, D and I of A are open projections in A∗∗ , and that the support
projection of J is in the centre of A∗∗ and the support projections of I in the centre
of D∗∗ – here considered as hereditary W*-subalgebra of A∗∗ .
(We obtain implicit that pJ · pD = pI for the corresponding open questions in
∗∗
A .)
Proof of I = J ∩ D:
The ideal J is the closed linear span of AIA. The intersection J ∩ D is an
ideal of D, obviously with I ⊆ J ∩ D. The latter because I = I · I · I (as for every
C *-algebra) and I 3 ⊆ AIA ⊆ J.
 7. RESIDUALLY ANTILIMINARY C *-ALGEBRAS 235

The closed linear span L of A · I is a left ideal of A with bi-polar (here = second
conjugate) L∗∗ = A∗∗ pL , where pL is an open projection in A∗∗ (with respect to
σ(A∗∗ , A∗ ) - topology on A∗∗ ).
and more ???????? what is needed below ????
The element d ∈ D is in J ∩D if there exists for each ε > 0 elements a1 , . . . , an ∈
A, b1 , . . . , bn ∈ A and c1 , . . . , cn ∈ I with the property
X
kd − ak ck bk k < ε .
k

If we use an approximate quasi-central unit (eτ ) of D and an approximate quasi-

P
central unit (fτ ) of I, then the sum k ak ck bk can be replaced by suitable sum-
mands
X
(eτ ak fτ ck fτ bk eτ )
k

with suitable index τ such that also eτ deτ − d is sufficiently small. The elements
eτ ak fτ and ck fτ bk eτ are then in DAI respectively in IAD.
(0.3) Question:
Does for every separable C *-subalgebra B ⊆ A of A exists a separable C *-
subalgebra C ⊆ A, such that C has the property in (0.1) – for irreducible d(C) –,
if A has the property in (0.1)?
(0.4) Suppose that A is separable and satisfies the conditions (0.1). Can then
M(A) have an irreducible representation of finite dimension? e.g. a character?
(0.5a) ?????????
(0.5b) If M(A) has no irreducible representation of dimension ≤ n then, A,
`∞ (A) and `∞ (M(A)) (and all non-zero quotient C*-algebras and ideals of them)
have no irreducible representation of dimension ≤ n.
Suppose that one of A, `∞ (A) or `∞ (M(A)) have an irreducible representation
ρ of dimension k ≤ n (here k = dimension of the corresponding Hilbert space).
Thus, suppose that one of the cases ρ(A) = Mk , ρ(`∞ (M(A))) = Mk , or
ρ(`∞ (A)) = Mk with some k ≤ n appears.
In the first case ρ extend a normal unital morphism from A∗∗ onto Mk with
k ≤ n. Since A ⊆ M(A) ⊆ A∗∗ , this shows that, in the case ρ(A) = Mk with
k ≤ n, there must exists an irreducible representation of M(A) of some dimension
k ≤ n.
In the case of an irreducible representation ρ : `∞ (M(A)) → Mk it is always
ρ(1, 1, . . .) = 1k in Mk . The natural diagonal embedding ∆ given by y ∈ M(A) 7→
(y, y, . . .) ∈ `∞ (M(A)) is unital and ρ ◦ ∆ : M(A) → Mk is a unital C*-algebra
morphism from M(A) into Mk .
(0.6) Question:
Suppose that A is separable and M(A) has an irreducible representation of finite
dimension. Has then A∞ (or Aω ) an irreducible representation of finite dimension?
236 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Hope? Question: Are elements of M(A)/A expressible as sum of two commut-

ing elements of A∞ ? (perhaps then with a shift of one those 2 elements?)
Are separable parts of M(A)/A contained in sums of suitable separable ?????
Using a quasi-diagonal approximate unit of A, we can find 2 hereditary C *-
subalgebras of `∞ (A) that have irreducible ????
If M(A) has no irreducible representation of finite dimension, then M(A)/A
(??? and ???? `∞ (A) ????) have no irreducible representation of finite dimension.
Suppose that `∞ (A) has no irreducible representation of finite dimension. Does
it imply that M(A) has no irreducible representation of finite dimension (at least
in case that A is σ-unital, or separable)?
(0.7) If any irreducible representation of A does not contain non-zero compact
operators in its image, then each hereditary C *-subalgebra of A has this property.
(1) l.p.i, weakly p.i., p.i., and strongly p.i. C *-algebras are all residually antil-
iminary.
(1.0) It suffices to show that locally p.i. C *-algebras are residual antiliminary.
This can be seen as follows:
Foe each non-zero quotients D/(D ∩ J), where J is a closed ideal of A and
D ⊆ A is a hereditary C *-subalgebra of A, there exist a pure state ρ on D with
ρ(D ∩ J) = {0} and a positive contraction d ∈ D+ with ρ(d) = 1 (by using
Kadison transitivity theorem, or Lemma ??). By Definition 2.0.3 of locally purely
infinite C *-algebras, there exists a C *-morphism h : C0 ((0, 1], K) → dAd ⊆ D with
ρ · h 6= 0. It shows that D/(D ∩ J) can not be one-dimensional (i.e., is ∼6 C). This
=
shows that A must be residual antiliminary. )
The implications “strongly p.i. ⇒ p.i. ⇒ weakly p.i. ⇒ locally p.i. ” show
then that the all sorts of “purely infinite” C *-algebras are residually antiliminary.
(2) ??? The free product A of countably many algebras An , An ∼ =
C0 ((0, 1], K(`2 (N)) has no finite-dimensional irreducible representation, but `∞ (A)
has a character (and moreover irreducible representations of every dimension n for
each n ∈ N).
This is not proven completely, because we have only an example where represen-
tations ρn : A → K “converge” to a character on some *-subalgebra of `∞ (A)/c0 (A)
that contains in its kernel an ideal J of A such that A/J is commutative.
But A itself has no character. Thus, ??? ρ∞ (A) ⊆ c0 (A) ???...
It is not clear if ρω itself has abelian image on all elements of `∞ (A)/c0 (A).
It is not clear if this A is residually antiliminary, or can be used to construct a
residually antiliminary C *-algebra B with the property that `∞ (B) has a character.
(3) Mn has the n × n-matrix [ajk ] with 1 = a12 = a23 = · · · = an−1,n (and zero
at all other places) as generator.
 7. RESIDUALLY ANTILIMINARY C *-ALGEBRAS 237

(4) Let In denote the ideal of A that is generated by all homomorphisms of

C0 ((0, 1], Mn ) into A,
Then A/In has only irreducible representations of dimension ≤ n − 1 :
Suppose that A/In has an irreducible representation d : A/In → L(H) with
dimension of H being ≥ n . Let P ∈ K(H) a projection of rank = n, then, by the
generalized Kadison transitivity Lemma ?? and semi-projectivity of C0 ((0, 1], Mn ),
cf. by Proposition ??, there exist a C *-morphism h : C0 ((0, 1], Mn ) → A with
d ◦ πIn (f0 ⊗ 1n ) = P . But h(C0 ((0, 1], Mn )) ⊆ In by definition of In and therefore
d ◦ πIn (f0 ⊗ 1n ) = 0 (with f0 (t) = t on (0, 1]). This contradicts the existence of
P ∈ K(H) of rank = n and shows that the dimension of H is ≤ n − 1 .
Moreover the ideal In has no irreducible representations of dimension < n :
Indeed, there would be an irreducible representation d : A → L(H) with k :=
dim(H) < n and d|In 6= 0. But, by definition of In , there exists a C *-morphism
h : C0 ((0, 1], Mn ) → In such that d ◦ h : C0 ((0, 1], Mn ) → L(H) is not = 0. But
there does not exist a C *-morphism from C0 ((0, 1], Mn ) onto onto Mk for k < n .
What is with the (perhaps bigger) ideal Jn generated by contractions a in A
with kan−1 k = 1 and an = 0? (Notice that Jn contains In !)
The universal C *-algebra An := C ∗ (a∗ , a ; kan−1 k = 1 ≥ kak, an = 0) has no
irreducible representations of dimension ≤ n − 1.
A2 := C ∗ (b∗ , b ; kbk = 1, b2 = 0) has a C *-morphism ϕ into An give by
ϕ(b) := an−1 , or by ϕ(b) := a` with ` ≥ n/2.
There must be an irreducible representation on a Hilbert space H and x ∈
H with kxk = 1, kan−1 xk = 1 , because kan−1 k = 1. Consider the vectors
{x, ax, a2 x, . . . , an−1 x} . They have all the norm = 1. It generates an a-invariant
(i.e., with aL ⊆ L) linear subspace L ⊆ H of dimension ≤ n. It contains the a-
invariant subspaces ak L generated by {ak x, ak+1 x, . . . , an−1 x} of dimension ≤ n−k:
{C · an−1 x} = an−1 L ⊆ an−2 L ⊆ . . . ⊆ . . . ⊆ a2 L ⊆ aL .
If k ∈ {1, . . . , n−2} exists with Dim(ak L) = Dim(ak−1 L) then a|ak−1 L is bijec-
tive on ak−1 L. Hence, ak−1 L = ak L and a|ak L is again bijective. This contradicts
an L = {0}, and shows that linear span of {x, ax, a2 x, . . . , an−1 x} is vector space
of dimension = n.
In particular, Mn−1 (C) can not contain a contraction a with kan−1 k = 1 and
an = 0.
Beginning from here, above considerations are not (!) integrated
so far.
Let H of dimension m < n, a ∈ L(H), with kak ≤ 1, kan−1 k = 1 and an =
0. Can suppose that C ∗ (a∗ , a ; kan−1 k = 1 ≥ kak, an = 0) is there irreducible
represented. The above considerations show that H must have dimension ≥ n.
Find x ∈ H with kxk = 1, kan−1 xk = 1. It must exists because kan−1 k = 1.
It follows that the vectors x, ax, a2 x, . . . , an−1 x necessarily have all the norm = 1.
238 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

They can not be linearly independent, because the dimension of H is = m < n.

Let H0 denote the linear span of {x, ax, a2 x, . . . , an−1 x}. It has some dimension
=: m < n.
There is a linear map T : Cn → H defined by T (e1 ) := x and T (ek ) := ak−1 x
for k = 2, . . . , n . The linear map T satisfies T Sn = aT for the linear map a ∈ L(H)
and the map Sn ∈ L(Cn ) defined by Sn (ek ) := ek+1 (k = 1, . . . , n − 1) and
Sn (en ) := 0 . The map Sn satisfies Sn` (ek ) = ek+` for k + ` ≤ n and Sn` (ek ) = 0 for
k + ` > n. In particular, Snn−1 (e1 ) = en , Snn−1 (ek ) = 0 for 1 < k ≤ n and Snn = 0.
It implies T (Sn (ek )) = T (ek+1 ) = ak x = aT (ek ), for k < n and T (Sn (en )) =
0 = an x = aT (en ), because T (en ) = an−1 x and an = 0. Thus T Sn = aT and
T : Cn → H0 is surjective.
More?
Move it to some Appendix ??.
(5) The ideal In is identical with the hereditary C *-subalgebra Dn of A that is
build by convex combinations of all positive n-homogenous elements of A. (Because
the Dn is invariant under inner automorphisms of A induced by unitaries in U0 (A +
C1).)
(6) Conjecture: If A is σ-unital and its asymptotic central sequence algebra
F (A) := (A0 ∩Aω )/Ann(A, Aω ) has no characters, then Aω is residual anti-liminary.
Attempt:
Let h ∈ Aω a positive contraction. D := hAω h. Then there is a unital C *-
morphism from F (A) into M(D) (? To be checked !!!?). The D can not have a
character.
Thus Aω is residual antiliminary.
(7) If A has only l.s.c. 2-quasi-traces that take the values 0 and +∞ then A is
residually antiliminary.
(8) All simple non-elementary C *-algebras are residually antiliminary.
Equivalently expressed, cf. [616, sec. 6.1.1], B is “antiliminary” if B does not
contain a non-zero “abelian element” b ∈ B+ . The element b ∈ B+ is called
“abelian” if bBb is a commutative subalgebra of B, cf. [616, sec. 5.5.1].
The following Proposition 2.7.7 and the Lemmata 2.7.3 and 2.7.13 will be used
for the recognition and applications of residual antiliminary C *-algebras.
Compare Remark A.5.8 and Lemma 2.1.7(o,v) with next Lemma and its proof.

Lemma 2.7.3. Let D a hereditary C*-subalgebra of a C*-algebra A . Then the

map J 7→ J ∩ D is a surjective map from the lattice I(A) of closed ideals of A onto
I(D) . In particular, I = D ∩ Span(AIA) for each closed ideal of D .

Proof. The latter identity can be seen easily with help of an approximate unit
{eτ } of D, by using that DAD = D , cf. Lemma 2.1.7(o). It shows the surjectivity
 7. RESIDUALLY ANTILIMINARY C *-ALGEBRAS 239

of the map J 7→ J ∩ D, because D ∩ J = I for J := Span(AIA) if I is a closed

ideal I of D ⊆ A . 

Why should this non-zero character exist???

Lemma 2.7.4. Let D a C*-algebra and c ∈ D+ with kck = 1 and suppose that

χ : E := cDc → C

is a non-zero character.
Then 0 < χ(c) ≤ 1 , and, for each non-zero positive contraction b ∈ D+
with kc − bk < χ(c)/3 and δ ∈ (kc − bk, χ(c)/3), the hereditary C*-subalgebra
(b − δ)+ D(b − δ)+ of D has a non-zero character.

Proof. By assumptions, the hereditary C *-subalgebra E := cDc of D has

non-zero character χ : E → C with kernel ideal I ⊂ cDc, and kck = 1. Since c is a
strictly positive element of E it follows that 0 < χ(c) ≤ 1.
Let I ⊆ E the kernel ideal of the character χ, and let J ⊆ D the closed ideal
of D generated by I, i.e., J := Span(D · I · D). It can be shown that J ∩ E = I by
using an approximate unit of I, cf. Lemma 2.7.3.
It implies that πJ (I) = C·p for some non-zero projection p ∈ D/J that satisfies
p(D/J)p = C · p and πJ (c) = χ(c) · p. In particular, kπJ (c)k = χ(c).
The ideal K of D/J generated by the 1-dimensional hereditary C *-subalgebra

πJ (E) = E/(J ∩ E) = E/I = C · p = p(D/J)p

is equal to πJ (Span(DcD)). The ideal K is also equal to the closure of the linear
span Span((D/J)p(D/J)). Thus, K is isomorphic to the algebra of compact oper-
ators K(H) on some Hilbert space H. The projection p ∈ K has rank equal to one
if considered as element in K(H) ∼
= K because pKp = C · p.
Let b ∈ D+ a with kbk ≤ 1, kb − ck < χ(c)/3 and δ ∈ (kb − ck, χ(c)/3).
HERE XXX check again !!!
There exists a contraction d ∈ D with d∗ cd = (b − δ)+ by Lemma 2.1.9. It
implies that in the ideal K of D/J the element πJ ((b − δ)+ ) = χ(c)πJ (d)∗ pπJ (d)
has at most the rank one. It means that, if πJ ((b − δ)+ ) 6= 0, then there exists a
projection q ∈ K ⊆ D/J with q(D/J)q = C · q and

πJ ((b − δ)+ ) = kπJ ((b − δ)+ )kq .

Thus, if πJ ((b − δ)+ ) 6= 0 then the πJ defines a C *-algebra epimorphism from the
hereditary C *-subalgebra (b − δ)+ D(b − δ)+ of D onto the algebra C · q in D/J .
This shows that (b − δ)+ D(b − δ)+ has a non-zero character if I(πJ ((b − δ))k > 0 .
We finish the proof by showing that I(πJ ((b − δ))k > 0 :
We have k(πJ ((b − δ)+ )k = k(πJ (b) − δ)+ k ≥ kπJ (b)k − δ and kπJ (b)k ≥ kπJ (c)k −
kπJ (c − b)k ≥ χ(c) − kc − bk. Thus,

k(πJ ((b − δ)+ )k ≥ χ(c) − kc − bk − δ > χ(c)/3 .

240 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

In particular, k(πJ ((b − δ)+ )k > 0. 

The next Lemma 2.7.5 could be improved with respect to the possible estimates,
but it is good enough for our applications.

Lemma 2.7.5. Let A a C*-algebra and B ⊆ A a C*-subalgebra. If A contains

a non-zero projection p with the properties pAp = C · p and that the distance of p
to B is less than 1/9, i.e., dist(p, B) < 1/9, then there exists a projection q ∈ B
with kp − qk < 1 and qBq = Cq.
Moreover, there is a unitary U ∈ U0 (M(A)) ∩ (1 + A) with q := U ∗ pU ∈ B.

Proof. Recall here that U0 (M(A)) denotes the in operator-norm connected

component of 1 in the unitaries U(M(A)). There is a positive linear functional ψ
on A with ψ(a)p = pap for all a ∈ A, because pAp = C · p. It shows that ψ must be
a pure state on A. Notice that there exists only one state on A with this property,
and the property that the closed ideal J ⊆ A of A generated by p is isomorphic to
the compact operators K(H) of some Hilbert space H, i.e., J ∼ = K(H).
By assumption, there exist b ∈ B with kb − pk < 1/9 . The selfadjoint f :=
(1/2)(b∗ + b) ∈ B is the real part of b and satisfies again kf − pk < 1/9 . It says,
for δ ∈ (0, 1/9 − ks − pk) and the unit 1 ∈ B ∗∗ , that

−(1/9 − δ)1 ≤ p + f− − f+ ≤ (1/9 − δ)1 .

Let χ a character on C ∗ (f ) ⊆ B ⊆ A with χ(f− ) = kf− k and χ(f+ ) = 0. It extends

to a pure state on B and then, – further on –, to a pure state on ρ : A → C with
the properties that ρ(f− ) = kf− k and ρ(f+ ) = 0. If we apply it to above inequality,
then we get ρ(p) + kf− k ≤ (1/9 − δ)+ and 0 ≤ ρ(p) ≤ 1 . Thus, kf− k < 1/9 and
positive part c := f+ satisfies kc − pk < 2/9, and c ≥ 0 .
Let ξ ∈ (kc − pk, 2/9) . It satisfies the inequalities kc − pk < ξ and, therefore,
1 − ξ < kck < 1 + ξ. In particular, 0 < ξ < kck because 9 ξ < 2 .
Lemma 2.1.9 gives a contraction d ∈ A with d∗ pd = (c − ξ)+ 6= 0, because
kc − pk < ξ and kck > ξ .
It implies that q := (kck − ξ)−1 (c − ξ)+ ∈ B is a rank-one projection in the
ideal J ∼
= K of A generated by p.
They imply following identities and estimates: (kck − ξ) · q = (c − ξ)+ , 1 − ξ <
kck < 1 + ξ,

kq − (c − ξ)+ k = |1 − (kck − ξ)| = 1 + ξ − kck < 2ξ ,

and
k(c − ξ)+ − pk ≤ kc − pk + kc − (c − ξ)+ k < 2ξ .
It gives the desired estimate:

kq − pk ≤ kq − (c − ξ)+ k + k(c − ξ)+ − pk < 4ξ < 1 .

 7. RESIDUALLY ANTILIMINARY C *-ALGEBRAS 241

The inequality kp − qk < 1 implies for the two rank-one projections p, q ∈ J ∼

=
K(H) that they are identical if they are linear dependent. But if they are not
linear dependent then the support project Q of p + q in J has rank two, i.e.,
p, q ∈ QJQ ∼
= M2 . Then there exists a unitary V ∈ U0 (QJQ) with V ∗ pV = q. The
unitary U := V + (1 − Q) ∈ 1M(A) + A has the proposed properties. 

Corollary 2.7.6. Let A a C*-algebra and b1 , b2 , . . . ∈ A+ a sequence contrac-

tions that converge to a contraction c ∈ A+ with kck = 1, i.e., limn kc − bn k = 0.
If the hereditary C*-subalgebras (bn − ρ)+ A(bn − ρ)+ have no non-zero charac-
ters for every n ∈ N and every rational number ρ ∈ (0, kbn k), then the hereditary
C*-subalgebra (cAc)+ can not have a non-zero character.

Proof. If cAc has a non-zero character then infinitely many of the hereditary
C *-subalgebras (bn − ρ)+ A(bn − ρ)+ must have a character, by Lemma 2.7.4. It
would contradict the assumptions. 

Proposition 2.7.7. Following properties of a C*-algebras A are equivalent:

(i) A is residual antiliminary in the sense of Definition 2.7.2.

(ii) Each hereditary C*-subalgebra D of A does not have a non-zero character.
(iii) Each non-zero hereditary C*-subalgebra D of A does not have irreducible
representations that contain non-zero compact operators in its image.
(iv) No irreducible representation ρ : A → L(H) of A contains a non-zero com-
pact operator in its image.
(v) There exists a norm-dense subset S ⊆ {a ∈ A+ ; kak = 1} with the prop-
erty that for each b ∈ S and each rational numbers ρ ∈ (0, 1) the hereditary
C*-subalgebra (b − ρ)+ A(b − ρ)+ of A has no non-zero character.

Proof. We give indirect proofs for all implications, except for the trivial im-
plications (ii)⇒(v) and (iii)⇒(iv).
(i)⇒(ii): Suppose that D is a hereditary C *-subalgebra of A that has a non-
zero character character χ : D → C and let I denote the kernel of χ. By Lemma 2.7.3
there exists a closed ideal J of A that satisfies J ∩ D = I. Then E := πJ (D) ∼
= C is
a hereditary C *-subalgebra of A/J and there is a unique non-zero projection p ∈ E
with E = C · p. Then E = p∗ (A/J)p is a non-zero commutative and hereditary C *-
subalgebra of the quotient A/J of A. It contradicts that A is residual antiliminary.
(ii)⇒(iii): Suppose that there exists a hereditary C *-subalgebra D of A that
admits an irreducible ρ : D → L(H) with the property that ρ(D) contains a non-
zero compact operator T ∈ ρ(D). Let I ⊂ D denote the kernel of ρ.
Since irreducible representations of C *-algebras are cyclic with respect to each
non-zero vector, one can show that the non-zero C *-subalgebra ρ(D) ∩ K(H) of
K(H) is an irreducible C *-subalgebra of K(H). It implies that K(H) ⊆ ρ(D).
Thus, there exist a projection p ∈ K(H) of rank one with the property p ∈ ρ(D) . It
implies pρ(D)p = C · p . The C *-subalgebra E := ρ−1 (C · p) = ρ−1 (pρ(D)p) of D is
a hereditary C *-subalgebra of D, – hence is also hereditary in A –, and E/(I ∩ E)
242 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

is one-dimensional, i.e., E is a hereditary C *-subalgebra of A that has a character.

It contradicts that E can not have a character by (ii).
(iii)⇒(i): Suppose that A is not residual antiliminary in in the sense of Defi-
nition 2.7.2.
Then there exists a closed ideal I 6= A of A and a non-zero element b ∈ A/I
such that the hereditary C *-subalgebra E := b∗ (A/I)b of A/I is commutative. Let
D := πI−1 (E) and ξ : E → C a non-zero character of E. Then D is a non-zero
hereditary C *-subalgebra of A, and χ := ξ ◦ (πI |D) is a non-zero character on D,
contradicting Property (iii) on A.
(iii)⇒(iv): Consider D := A.
(iv)⇒(iii): Suppose that D is a hereditary C *-subalgebra of A, ρ0 : D → L(H0 )
an irreducible representation and that T ∈ ρ0 (D) is a non-zero compact operator
on H0 . Take e ∈ D with ρ0 (e) = T . We can suppose that e ≥ 0 and T ≥ 0 by
replacing them otherwise by e∗ e and T ∗ T .
The irreducible representation ρ0 of D “extends” to an irreducible representa-
tion ρ1 : A → L(H1 ) in the sense that there exists an isometry I : H0 → H1 such
that ρ1 (d) ◦ I = I ◦ ρ0 (d) and ρ1 (d)x = 0 for all x ∈ H1 >I(H0 ) and for all d ∈ D.
Thus, e ∈ D ⊆ A satisfies that ρ1 (e) is a non-zero compact operator on H1 . This
contradicts (iv).
(v)⇒(ii): The assumptions of Part (v) imply by Corollary 2.7.6 that bAb has
no non-zero character for all b ∈ A+ with kbk = 1.
It follows that each hereditary C *-subalgebra D of A does not have a non-zero
character:
Suppose that there exists a non-zero character ξ on D. Then there exists some
a ∈ D with ξ(a) = 1. Let b := a∗ a − (a∗ a − 1)+ , then b ∈ D+ , kbk = 1 and ξ(b) = 1.
Thus, the restriction of ξ to bAb is a non-zero character. But this impossible. 

Lemma 2.7.8. A C*-algebra B has no irreducible representation of dimension

≤ n, if and only if, the closed ideal Jn of B generated by all elements b ∈ B with
the property bm−1 6= 0 and bm = 0 and m ≥ n, is equal to B.

If B has an irreducible representation of dimension ≥ n then B contains an

element with bn−1 6= 0 and bn = 0.
If B contains an element with b ∈ B with bn−1 6= 0 and bn = 0, then B has an
irreducible representation with dimension ≥ n.
Thus, if B has only irreducible representations of dimension < n, then B can
not contain b with bn−1 6= 0 and bn = 0.
If B does not contain elements b ∈ B with bn−1 6= 0 and bn = 0 then every
irreducible representation of B has dimension < n.
Let Jn denote the closed ideal of B generated by all elements b ∈ B with
bn−1 6= 0 and bn = 0.
 7. RESIDUALLY ANTILIMINARY C *-ALGEBRAS 243

(1) B/Jn has only irreducible representation on Hilbert spaces of dimension

≤ n.
It is equivalent to the property that for every irreducible representation ρ : B →
L(H) of dimension dim(H) > n there exists a contraction b ∈ B with bn+1 = 0,
bn 6= 0 and ρ(b) 6= 0. (Because this implies that ρ|Jn is an irreducible representation
of the Jn .)
Proof of (1):
Let ρ : B → L(H) an irreducible representation and dim(H) > n.
By (sharpened form of) the generalized Kadison transitivity theorem there
exists C *-algebra homomorphism ϕ : C0 ((0, 1], Mn+1 ) → B such that f → ρ ◦ ϕ(f )
has kernel C0 ((0, 1), Mn+1 ), and defines a (non-zero) C *-morphism η : Mn+1 →
K(H) of Mn+1 into K(H). Let d := e1,2 +e2,3 +· · ·+en−1,n Then b := ϕ(f0 ⊗d) ∈ B
satisfies bn−1 6= 0 and bn = 0.
(2) Every irreducible representation ρ : B → L(H) of dimension dim(H) ≤ n
contains Jn in its kernel ideal.
This is equivalent to the property of Jn that says that Jn is contained in the
intersection of the kernels of all irreducible representations ρ : B → L(H) with
Dim(H) ≤ n.
To show this, we have to prove that there does not exist an irreducible repre-
sentation ρ : B → L(H) on a Hilbert space H such that ρ(b) 6= 0 for some b ∈ B in
the given set of generators for Jn .
The generators of Jn are the elements b ∈ B with the property bn−1 6= 0 and
bn = 0.
Thus we have to determine: What is the dimension of a Hilbert space H if
L(H) contains a contraction T with T n = 0 and T n−1 6= 0.
We show that for a Hilbert space H holds:
Dim(H) ≥ n, if and only if, L(H) contains an operator of finite rank T ∈ L(H)
with the property T n = 0 and T n−1 6= 0.
Equivalently this says that if Dim(H) ≤ n then there does not exists an operator
T ∈ L(H) with T m−1 6= 0 and T m = 0 if m > n.
(There exists T with T n−1 6= 0 and T n = 0 if Dim(H) = n, but no T with
m−1
T 6= 0 and T m = 0 if Dim(H) < m < ∞)
The images Hk := T k H are linear subspaces for k ∈ {0, . . . m}, where we let
T 0 := idH . They are closed subspaces of H, because Dim(H) ≤ n. Obviously
Hk+1 ⊆ Hk , and the Hk are invariant subspaces of T , i.e., T Hk ⊆ Hk .
If there is k ∈ {0, . . . , m − 1} with Hk = Hk+1 and Hk then is T |Hk is a
bijective linear map on Hk . It implies that T ` 6= 0 for all ` ∈ N, if Hk = Hk+1 for
some k ∈ {0, 1, . . . , m − 1}. It would contradict the assumption that T m = 0.
244 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Thus, Dim(Hk+1 ) < Dim(Hk ) 6= {0} for k ∈ {0, 1, . . . , m − 1} and Hm =

T H = {0}, because T m−1 6= 0 and T m = 0.
m

It follows that ????????????

Since T m = 0, by assumption, this can not happen. Thus, Dim(T k+1 H) <
Dim(T k H) for all k ∈ {0, . . . , m}. {0} = Hm and Dim(Hk+1 ) < Dim(Hk ) for
k ∈ {0, 1, . . . , m}.
It follows that Dim(Hi) ≥ m − 1 ???
(2) If B has an irreducible representation ρ of dimension m > n then there
exists an element
????????
If B/Jn has an irreducible representation ρ : B/Jn → Mn
of dimension m or n here ?????
then the semi-projectivity of C0 ((0, 1], Mm ) shows that there is a C *-morphism
ψ : C0 ((0, 1], Mm ) → B with (ρ ◦ ψ)(f ) = f (1) for f ∈ C0 ((0, 1], Mm ) . Then
b := ψ(f0 ⊗ (p1,2 + p2,3 + . . . + pn1 ,n ) satisfies bm−1 6= 0 and bm = 0 . Thus, m < n
if b 6∈ Jn .
Need also to show that Jn has no irreducible representation of dimension < n
B/Jn has only irreducible representation of dimension < n. It is NOT clear if it is
so!
We know only that if B has an irreducible representation D : B → Mk of
dimension k < n then B contains a contraction b ∈ B with kb` k = 1 for ` < k
and bk = 0, because B contains then an image ψ(Mk (C0 (0, 1])) of Mk (C0 (0, 1]) by
some C *-morphism ψ with D ◦ ψ(f0 ⊗ 1n ) = 1k . In this (very special) situation
one has moreover that D(b)` 6= 0 for 1 ≤ ` < k. The b is given by ψ(f0 ⊗ Q) for
Pk−1
Q := `=1 p`,1+` .
But b ∈ B with this properties can be found if B has any irreducible representa-
tion D of dimension ≥ k. Simply by considering the the hereditary C *-subalgebra
that maps onto Mk ⊕ 0n−k ⊆ Mk ⊕ Mn−k ⊆ Mn .
If B does not contain an element b ∈ B with bn−1 6= 0 and bn = 0 then B can
not have any irreducible representation of dimension ≥ n.
The problem is, if the relations are “liftable”, i.e., if B/J does not contain an
element c ∈ B/J with cn−1 6= 0 and cn = 0
(in particular B/J can not have an irreducible representation of dimension
≥ n),
is then every element b ∈ B with bn−1 6= 0 and bn = 0 contained in J?
The problem is, that it can be that πJ (b) 6= 0, but also πJ (b` ) = 0 for some
1 < ` < n − 1. This can happen in case B = B1 ⊕ B2 if b = b1 ⊕ b2 and
????. bn = 0 for n := max(n1 , n2 ) if
 7. RESIDUALLY ANTILIMINARY C *-ALGEBRAS 245

?????

Remark 2.7.9. The property that the closed ideal Jn (B), generated by all
elements b ∈ B with bn−1 6= 0 and bn = 0, is equal to B, like considered in Lemma
2.7.8 does not pass in general to `∞ (B) or its quotients, as e.g. the ultrapowers Bω .
It is still not clear if this is a true counter examples!
Can a sequence of surjective (!) C *-morphisms hn : B → K
produce a morphism from Bω into Kω that is not surjective?
The ultrapower Bω of the free product B of countably many copies of
C0 ((0, 1], K(`2 )) could have a character, even if B has no irreducible representation
of finite dimension?
It is only known that the restriction to B ⊂ Bω is a character.

Is the ideal Jn different from the ideal generated by all n-homogenous positive
elements?
Case n = 2, i.e., general C ∗ (b∗ , b) with b2 = 0?
Then C ∗ (b∗ , b) ∼
= M2 ⊗ C ∗ (b∗ b) by letting b correspond to p1,2 ⊗ (b∗ b)1/2 :
The polar decomposition b = v(b∗ b)1/2 in C ∗ (b∗ , b)∗∗ has the property that
b = (bb∗ )1/2 v. In particular, b∗ = v ∗ (bb∗ )1/2 is the polar decomposition of b∗ ,
v(b∗ b)v ∗ = bb∗ and v ∗ (bb∗ )v = b∗ b.
Let f := (b∗ b + bb∗ )1/2 = (b∗ b)1/2 + (bb∗ )1/2 in C ∗ (b∗ , b). The partial isometry
v commutes with f : vf = v(v ∗ v)(b∗ b)1/2 = b and f v = (bb∗ )1/2 v = b . Notice that
the non-zero values in the spectra of f and (b∗ b)1/2 are the same.
Thus, there is a C *-morphisms ρ from C ∗ (v ∗ , v) ⊗ C ∗ (f ) onto C ∗ (b∗ , b) with
the property that ρ(v ⊗ f ) = b and ρ((vv ∗ + v ∗ v) ⊗ f ) = f . Since C ∗ (v ∗ , v) ∼
= M2 ,
∗ ∗ ∗
we get an isomorphism from M2 (C (f )) onto C (b , b).

Proof. Let Dn ∈ Mn denote the matrix with entries dk,k+1 = 1 for k =

1, . . . , n − 1 and dk,` = 0 on all other places. Then Dnn−1 6= 0 but Dnn = 0.
Let Jn ⊆ B the closed ideal of B that is generated by all elements b ∈ B with
b = 0 and bn−1 6= 0.
n

We show that B/Jn is sub-homogenous with irreducible *-representations of

dimension < n, and that Jn has only irreducible representation of dimension ≥ n.
Let ρ : B → Mm a *-morphism, and b ∈ B with bn−1 6= 0 but bn = 0 with
n−1
ρ(b ) 6= 0...
Need: C ∗ (b∗ , b) is n-homogenous (or n-sub-homogenous C *-subalgebra) if bn =
0 and bn−1 6= 0???
Suppose that ρ : Jn → L(H) has an irreducible representation of dimension
m < n, then exists an element b ∈ B with bn = 0 and bn−1 6= 0, and ρ(b) 6= 0 ...
HERE IS SOMETHING wrong??
246 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

We need also to show that each irreducible representation with ρ|Jn 6= 0 has
dimension ≥ n .
This could be shown by proving directly that Jn is also generated by all n-
homogenous elements of B.
At least, it is easy to see that the image of all non-zero C *-morphisms
ψ : C0 ((0, 1], Mn ) → B is contained in Jn , because ψ(f0 ⊗ Dn ) is contained in Jn ,
(n)
where Dn means the matrix with entries dj,k = 1 for k := j + 1 and j = 1, . . . , n − 1
(n)
and let dj,` := 0 if ` − j 6= 1 . The element f0 ⊗ Dn generates C0 ((0, 1], Mn ) as
a C *-algebra, because t · Dn generates Mn as C *-algebra for each t ∈ (0, 1] (Use
here the variant of the classical Stone-Weierstrass theorem for type-I C *-algebras,
e.g. [704, cor. 4.7.8]).
The question is, if b ∈ B with bn = 0 and bn−1 6= 0 generates an n-homogenous
C *-subalgebra?
What about b − bn−1 ? Has it a good polar decomposition?
(1 − b)(b + b2 + . . . + bn−1 ) = b − bn and bn = 0 if b is as above.
How about explicit construction?
b(bn−1 H) = {0}.
Let ρ : B → L(H) and irreducible with ρ(Jn ) = {0}. Then the dimension of H
is < n :
Suppose that the dimension of H is ≥ n. Then there exists a subspace L ⊆ H of
dimension n, a C *-morphism λ from Mn into L(H) with λ(1n ) = PL the orthogonal
projection onto L, and a C *-morphism ψ from C0 ((0, 1], Mn ) into B with the
property ρ ◦ ψ = λ (by Kadison transitivity theorem and projectivity of cones over
finite-dimensional C *-algebras).
We take in B the element C := ψ(f0 ⊗ Dn ). It satisfies C n−1 6= 0 but C n = 0.
Thus, C ∈ Jn and ρ(C) 6= 0, in contradiction to the requirement ρ(Jn ) = {0}.
The family of elements b ∈ B with bn = 0 and bn−1 6= 0 is invariant under all
automorphisms of B ... Thus, the hereditary C *-algebra generated by them is an
ideal. The same happens with n-homogenous elements.
Use Lemma 2.1.15(iii)
The case n = 1, i.e., b2 = 0 is the interesting case.
Is the C *-subalgebra generated by them an ideal of B? 

Lemma 2.7.10. The following properties of C*-algebras A are equivalent:

( i) For every irreducible representation d : A → L(H) the image d(A) does

not contain a nonzero compact operator on H, i.e., d(A) ∩ (C)(H) = {0}.
( ii) Every non-zero hereditary C*-subalgebra D of A has no character.
(iii) For every non-zero hereditary C*-subalgebra D of A, k ∈ N and state
ρ on D there exists a C*-algebra morphism h : C0 ((0, 1], Mk ) → D with
ρ(h((0, 1] ⊗ 1k )) = 1.
 7. RESIDUALLY ANTILIMINARY C *-ALGEBRAS 247

Lemma 2.7.11. Let A a C*-algebra. Following observations are nearly obvious:

( i) If B ⊆ C ⊆ A are C*-subalgebras of A and if an element b ∈ B+ has the

property that bBb has no non-zero character, then bCb and bAb have no
non-zero characters.
( ii) If c ∈ A+ and cAc has no non-zero character, C ⊆ A is a separable C*-
subalgebra of A with c ∈ C, then there exists a separable C*-subalgebra
B ⊆ A such that C ⊆ B and cBc has no non-zero character.
(iii) Let X ⊆ A+ be a countable subset of elements with the property that xAx
has no non-zero character for each x ∈ X. Then there exists a separable
C*-subalgebra C ⊆ A with X ⊆ C and the property that xCx has no
non-zero characters for each x ∈ X.
(iv) Let C a C*-algebra and let X ⊆ C+ a countable subset of contractions
that is norm-dense in the set of positive contractions of C. If all elements
x ∈ X have the property that (x − ρ)+ C(x − ρ)+ has no nonzero character
for all ρ ∈ (0, kxk), then C*-algebra C is residually antiliminary (and
separable).

Proof. Part (i) is obvious, because the restriction to bBb of a character on

bCb or bAb would be a (non-zero) character on bBb.
(ii): TEXT(ii): If c ∈ A+ and cAc has no non-zero character, C ⊆ A is a
separable C *-subalgebra of A with c ∈ C, then there exists a separable C *-sub-
algebra B ⊆ A such that C ⊆ B and cBc has no non-zero character.
If c ∈ A+ and D := cAc has no non-zero character, then cAc is – as a C *-
algebra – generated by its 2-homogenous elements. They are determined by the
elements a ∈ cAc with a2 = 0 and kak = 1. This is, because C ∗ (a∗ , a) ⊆ A is then
a 2-homogenous C *-algebra. And for every pure state ρ on cAc there is contraction
a ∈ cAc with ρ(a∗ a) = 1 and a2 = 0. Indeed, let ρ ∈ A∗ a pure state with ρ(c) 6= 0
then the irreducible representation dρ : A → L(H) with cyclic vector x ∈ L(H),
i.e., dρ (A)x = H, kxk = 1 and hdρ (a)x, xi = ρ(a) is not a character if an only if
dim(H) > 1. Thus there exists y ∈ H with kyk = 1 and hx, yi = 0.
By ???? some of above lemmata ???
Let E ⊆ A a separable C *-subalgebra of A that contains c, let C := cEc ⊆ D .
If ????
C *-subalgebra with c ∈ C+ ???
(iii): remember TEXT(iii): Let X ⊆ A+ be a countable subset of elements
with the property that xAx has no non-zero character for each x ∈ X. Then there
exists a separable C *-subalgebra C ⊆ A with X ⊆ C and the property that xCx
has no non-zero characters for each x ∈ X.
(iv): Let c ∈ C+ with kck = 1. Then there exists a sequence x1 , x2 , . . . ∈ X
such that limn kxn − ck = 0. The assumption that (xn − ρ)+ C(xn − ρ)+ has no
248 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

non-zero character for all rational numbers ρ ∈ (0, kxn k) implies that cCc has no
non-zero character, cf. Corollary 2.7.6.
Complete proof be filled in ... ??


Proposition 2.7.12. The class RAL of the ‘ residual anti-liminary’ C*-

algebras, defined in Definition 2.7.2, is invariant under each of following opera-
tions:

( i)
Passage to non-zero quotients,
( ii)
Passage to non-zero hereditary C*-subalgebras (including closed ideals),
( iii)
Forming of finite direct sums,
( iv)Inductive limits,
( v)Maximal tensor products A ⊗max B, ????,
( vi)Infinite direct c0 -sums A1 ⊕ A2 ⊕ · · · ,
(Follows from (iii) and (iv))
( v) Passage from A to `∞ (A).

If A is residual anti-liminary, then for each countable subset S of A there exists

a separable C*-subalgebra B ⊆ A such that S ⊆ B and B is residual anti-liminary.
In particular, each residual anti-liminary C*-algebra A is the inductive limit
of its separable residual anti-liminary C*-subalgebras (in its natural containment
order as indices of the order).

WHAT ABOUT: tensor products?,

Proof. The proposed permanence properties of the class of residual anti-

liminary C *-algebras A can be seen by following indirect arguments:
If A/J is a non-zero quotient of A and E ⊆ A/J a closed hereditary C *-
subalgebra of A/J and has a non-zero character χ : E → C (i.e., if A/J does not
satisfy (ii) in place of A), then D := πJ−1 (E) is a non-zero hereditary C *-subalgebra
of A, and D has the non-zero character χ ◦ πJ . But this is impossible by property
(ii) of A.
Thus, E must satisfies Part(ii) if A satisfies Part(ii), i.e., E must be residual
anti-liminary if A is residually anti-liminary. 

Very important open Question:

Is `∞ (A) residual anti-liminary if A is residually anti-liminary?
(It seems to be very difficult to decide, despite it can be reduced to the case of
separable A.)
Ques: Is every separable subset of a residual anti-liminary C *-algebra A con-
tained in a residual anti-liminary separable C *-subalgebra of A?
Is a C *-subalgebra of a residual anti-liminary separable C *-algebra again resid-
ual anti-liminary?
 7. RESIDUALLY ANTILIMINARY C *-ALGEBRAS 249

Is the inductive limit of a sequence of residual antiliminary separable C *-

algebras again residual anti-liminary?
Has positive answer, using following obvious fact:
(The C can be generated by a countable family of – inside C – 2-homogenous
contractions ????)
Combine this with following observation:
If b ∈ A+ is an element such that bAb has no non-zero character, then there
exists a separable C *-subalgebra B ⊆ A such that b ∈ B and bBb has no character.
It shows then by countability of a dense subset of the positive contractions in
a separable C *-algebra C:
If C is a separable C *-subalgebra of A and X ⊂ C+ is a countable subset of
the positive contractions that is dense in {c ∈ C+ ; kck = 1} then there exists a
separable C *-subalgebra D of A such that C ⊆ D and for each x ∈ X and each
rational number ρ ∈ (0, kxk) the hereditary C *-subalgebra (x − ρ)+ D(x − ρ)+ has
no non-zero character.
??? To be shown !!! ???
One can show that this implies that cDc has no non-zero character for every
c ∈ C+ .
In this way we find a sequence C1 := C ⊆ C2 ⊆ C3 . . . of separable C *-
subalgebras Cn of A, and a countable subset S of the positive contractions D :=
T
n Cn ⊆ A that is dense in {d ∈ D+ ; kdk = 1} and has the property that cDc has
no non-zero character for each c ∈ S.
Ques: Is the inductive limit of residual antiliminary C *-algebras again residu-
ally anti-liminary?
S
(Seems to be: D ⊂ E hereditary, E = τ Eτ with Eτ residual anti-liminary, τ
in upward directed net, Eτ ≤ Eσ if τ < σ... D with character ξ.
Then by Remark/Lemma ?? ???: If D ⊆ E is a hereditary C *-subalgebra of E
and J is a closed ideal of D, the the ideal I of E generated by J satisfies D ∩ I = J.
We consider ideal I of E generated by kernel of ξ (on D), and then the ideal
J generated by the image of D in E/I. It would show that kπI (f )k = kπJ (f )k for
every f ∈ Eτ . This leads to a contradiction. Thus, D can not have a character.
Thus, we can consider the quotient E/I ⊇ D/J. But D/J is one-dimensional:
there is a projection p ∈ D/J ⊆ E/I with C · p = D/J. and p(E/I)p = D/J
Thus E/I contains a minimal projection p with p(E/I)p = C · p.
The ideal K of E/I generated by p is necessarily isomorphic to K(H) of some
Hilbert space.
The C *-algebras πI (Eτ ) ∼
= Eτ /(Eτ ∩ I) are again residual anti-liminary by
Prop. ????(??).
250 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Therefore there intersections with K must be = {0}. It follows that

kπK (πI (e))k = kπI (e)k for all e ∈ Eτ for each τ .
S
We use now that τ πI (Eτ ) is dense in πI (E):
There exists τ and e ∈ πI (Eτ )+ with kek ≤ 1 and kp − ek < 1/4 . This implies
that kek = kπK (e)k < 1/4 but kek ≥ kpk − kp − ek ≥ 1 − 1/4 = 3/4 (because
1/2 = kek + kp − ek ≥ kpk = 1) a contradiction.
Compare Remark 2.1.16(iv):
Item(iv):
If, moreover, A is strictly antiliminary (also called residually antiliminary, cf. Def-
inition 2.7.2) in the sense that each non-zero quotient A/J of A is antiliminary in
the sense of [616, sec. 6.1.1]
(Here footnote:
It is equivalent to the formally stronger assumption that no hereditary C *-
subalgebra of A has a non-zero character.),
then for every non-zero hereditary C *-subalgebra D of A, every pure state ρ
on D and every n ∈ N there exists a C *-morphism ψ : C0 ((0, 1], Mn ) → D with
ρ(ψ(f0 ⊗ p11 )) = 1.
Big big QUESTION:
Does here exists a non-zero element e ∈ A+ such that the hereditary C *-subalgebra
Ann(e) := {a ∈ A ; ae = 0} of A generates A as an ideal of A if A is residually
antiliminary ????
In the case of separable A one could look to e := (ϕ(f0 ⊗ p11 ) − ε) with ϕ and
ε suitable...?

Lemma 2.7.13. Let ρ : D → L(H) an irreducible representation that does not

contain a non-zero compact operator in its image ρ(D), and (p1 , p2 , . . . , pm ) a se-
quence of orthogonal projections pn ∈ L(H) of finite rank kn ∈ N, 1 ≤ n ≤ m.
Then D contains kn -homogenous positive contraction en ∈ D+ with pn ρ(en ) =
ρ(en )pn = pn .
If the p1 , . . . , pm are mutually orthogonal, then the en can be found mutually
orthogonal.

Proof. Use the “advanced version” of Kadison transitivity in Lemma

2.1.15(ii).
The case of mutually orthogonal p1 , . . . , pm can be studied by considering the
projection p1 + · · · + pm . 

The following Lemma 2.7.14 lists some later used reformulations of the Defini-
tion 2.7.2.
It could be that some items are only one-sided implications... Check it!
 7. RESIDUALLY ANTILIMINARY C *-ALGEBRAS 251

Lemma 2.7.14. Following properties of a C*-algebra A 6= {0} are equivalent

(???):

(i) Any irreducible representation of A does not contain a non-zero compact

operator in its image, i.e., for every irreducible representation ρ : A →
L(H) holds that ρ(A) ∩ K(H) = {0} .
(ii) A is residual antiliminary in sense of Definition 2.7.2.
(iii) Each hereditary C*-subalgebra D of A has no (non-zero) character.
(iv*) [Definition 2.7.2 of “residual antiliminary” !!!] Every non-zero quotient
A/I of A is antiliminary for each closed ideal I 6= A of A .
[??? Follows immediately from [616, prop. 6.2.8]. ????
???? Take J := π −1 (K(H)). If x ∈ J then how to find an other
representation ???] .
(v) For each closed ideal I 6= A and x ∈ A \ I there exists an irreducible
representation π : A → L(H) with π(I) = {0} and π(x) 6∈ K(H) .
(vi) All non-zero quotients A/I of A do not contain a “minimal” projection
p ∈ A/I (i.e., with the property p(A/I)p = Cp 6= {0} ).
(viii) Each non-zero hereditary C*-subalgebra D of A has no non-zero quotient
of Type I.
What means Type I here?
(It implies (iii): D has no character.)
(Implies: Each non-zero quotient of D contains a non-zero n-
homogenous positive contraction.)
(ix) For each non-zero hereditary C*-subalgebra D of A and each pure state
λ on A with λ(D) 6= 0 and each n ∈ N there exists an n-homogenous
element e ∈ D+ with λ(e) > 0 .
(x) For every n ∈ N, every closed ideal I of A, and every hereditary C*-
subalgebra D of A with πI (D) 6= {0} there exists an n-homogenous positive
contraction a ∈ D with kπI (a)k = 1.
(xi ???) More needed ???

Proof. To be filled in ??
The problems are ?: Given a non-zero hereditary Abelian C *-subalgebra D
of A. Does there exists an irreducible representation ρ of A such that ρ(D) is
1-dimensional ????
(extend a character ξ of D to a pure state on A/I and use that D ∩ I the the
ideal I of A generated by the kernel of the character ???????
Let D any hereditary C *-subalgebra of A and ξ a character of D, K ⊆ D the
kernel of ξ.
Citation of Lemma for next is where ???:
The closed ideal of ideal J of A generated by K has the property that J ∩D = K.
Thus, πJ (D) ∼
= D/K ∼
= C, and therefore, πJ (D) = C · p for some projection
p ∈ A/K with p(A/K)p = C · p.
252 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

?(x)⇒(ix): trivial.
?(ix)⇒(x): indirect.
The equivalence of (?v) and (?vi) (= old ???) follows immediately from [616,
prop. 6.2.8].
The other implications follow from the fact that for each closed ideal J of a
hereditary C *-subalgebra D of A the closed ideal I := Span(AJA) of A satisfies
I ∩ D ⊇ J.
latter OK ????? 

Lemma 2.7.15. Let A a C*-algebra, Cn := C0 ((0, 1], Mn ) = Mn (C0 ((0, 1]) for
n > 1, C∞ := C0 ((0, 1], K(`2 (N)) and D a hereditary C*-subalgebra of A.
Let κ ∈ {2, 3, . . .} or κ = ∞. Then the following properties (i) and (ii) of A
and D are equivalent (for the same fixed κ).

(i) For every pure state ρ on A with ρ|D 6= 0 there exists a C*-morphism
ϕ : Cκ → D with ρ(ϕ(Cκ )) 6= {0}.
(ii) For every non-zero positive functional ρ on A (or on D) with ρ|D 6= 0
there exists a C*-morphism ϕ : Cκ → D with ρ(ϕ(Cκ )) 6= {0}.

Moreover, if Part (i) holds then D has no irreducible representation π : D →

L(H) with dimension Dim(H) < κ.
In case of κ ∈ N, i.e., with κ < ∞, this property of D is moreover equivalent
to Part (i).

Proof. For the following we use that for each closed ideal J of D the ideal I
of A generated by J satisfies J = I ∩ D, cf. Lemma ??.
Consider the closed ideal Jκ of D generated by the images ϕ(Cκ ) of all possible
C *-algebra morphisms ϕ : Cκ → D.
If J = D then clearly Part(ii) holds, – and implies Part(i).
Conversely, if Part (i) holds then Jκ = D because, otherwise, there would exists
a pure state λ on D with λ(Jκ ) = {0}. The pure state λ has a unique extension to
a pure state ρ on A.
This shows the equivalence of Parts (i) and (ii).
If Part (i) holds, then D has no irreducible representation π : D → L(H) with
dimension Dim(H) < κ :
Indeed, if π : D → L(H) is an irreducible representation, then the vector state
λ(d) := hπ(d)x, xi, for given x ∈ H with kxk = 1, is a pure state on D that uniquely
extends to a pure state ρ on A. By Part (i) there exists a C *-morphism ϕ : Cκ → D
with ρ(ϕ(Cκ )) 6= {0}.
The positive functionals ρ and λ coincide on ϕ(Cκ ) ⊆ D. Thus π(ϕ(Cκ )) ⊆
L(H) is isomorphic to a non-zero quotient C *-algebra of C0 ((0, 1], Mκ ) and has a
 7. RESIDUALLY ANTILIMINARY C *-ALGEBRAS 253

dimension ≥ κ2 , considered as vector space. Thus, L(H) has vector-space dimension

≥ κ2 . But this me means that Dim(H) ≥ κ .
In case κ = ∞ this should be read as: H must be of infinite dimension, because
then ρ(ϕ(Cκ ))x is as a vector space infinite dimensional for each x ∈ H with
ρ(ϕ(Cκ ))x 6= {0} ( 40 ).
Above we have seen that the property of D ⊆ A and 2 ≤ κ ≤ ∞, as stated in
Part (i), implies that D has no quotient that is isomorphic to Mκ if 2 ≤ κ < ∞,
i.e., D has no irreducible representation on a Hilbert space of dimension < κ.
The opposite direction of this implication holds in case of 1 < κ ∈ N, i.e., if D
has no irreducible representation π : D → L(H) with dimension Dim(H) < κ, then
the hereditary subalgebra D ⊆ A satisfies Part (i):
Let ρ a pure state on A with ρ|D 6= 0. Then there is an irreducible representa-
tion π : A → L(H) with cyclic vector x such that H = π(A)x and ρ(a) = hπ(a)x, xi
for a ∈ A. The closed linear subspace H0 := π(D)x is invariant under π(D) and
the restrictions ψ(d) := π(d)|H0 of π(d) to H0 for d ∈ D define an irreducible
*-representation ot D on H0 . We denote by P the orthogonal projection from H to
H0 and let y := kP xk−1/2 · P x. Then kP xk = kρ|Dk > 0 and λ(d) := hψ(d)y, yi
is pure state on D of norm = 1.
Now we use the assumption that D has no irreducible representation on a
Hilbert space of dimension < κ, i.e., that D has no quotient D/I that is isomorphic
to Mm for some m < κ. It follows that Dim(H0 ) ≥ κ. Let P ∈ L(H0 ) ⊆ L(H) a
projection with Dim(P H) = κ. By Lemma ??
there exists a C *-morphism ϕ : Cκ → D with ϕ(Cκ ) = P L(H0 )P = P L(H)P
and ψ ◦ ϕ(f0 ⊗ P ) = P . Thus, ρ(ϕ(Cκ )) 6= {0} and D ⊆ A fulfill Part (i).
Thus, in case where 2 ≤ κ < ∞, the non-existence of irreducible representations
ρ : D → L(H) with Dim(H) < κ is equivalent to Part (i) with this κ. 

The following equivalent formulations l.p.i.(1) and l.p.i(2) of local pure in-
finiteness imply that l.p.i. C *-algebras A are residually antiliminary in the sense of
Definition 2.7.2 ????? where ????? :
l.p.i.(1):
The C*-algebra A is locally purely infinite if, for every non-zero hereditary C*-sub-
algebra B of A and every pure state ρ of B, there exists a σ-unital stable C*-sub-
algebra D ⊆ B with ρ|D 6= 0 .
Indeed, if B ⊆ A is a hereditary C *-subalgebra of A and ρ is a pure state of
B, then ρ can be uniquely extended to a pure state ρext on A with ρext |B = ρ and
there exists a positive contraction e ∈ B+ with ρ(e) = 1.
Let D := eAe. Then D ⊆ B and ρext (a) = ρ(eae) for all a ∈ A.
By assumption there exists a a stable C *-subalgebra D ⊆ E with ρ(D) 6= {0}.
40 Notice here that ρ ◦ ϕ is non-zero, but is not necessarily cyclic or even irreducible, because

ϕ(Cκ ) is in general not necessarily hereditary in D.

254 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

It follows that ????

for every non-zero a ∈ A+ and pure state ρ on A with ρ(a) > 0, there exists a
stable C *-subalgebra D ⊆ aAa with ρ(D) 6= {0}.
Thus, ?????
Conclusion:
This property implies in particular that every nonzero hereditary C *-
subalgebra B of A can not have a character ρ, because the non-zero restriction ρ|D
to any stable non-zero D ⊆ B can not be character.
Thus, locally purely infinite C *-algebras are residual antiliminary in sense of
Definition 2.7.2.
l.p.i.(2): The C *-algebra A is locally purely infinite if, for every d ∈ A+ with
kdk = 1 and every state ρ of A with ρ(d) 6= 0, there exists b ∈ D := dAd with the
properties that 0 ≤ b , ρ(b) > kρ|Dk/2 , kbk = 1 and bAb is stable.
Here: One of the (old! last) estimate is not clear ?!
Question: Carries the latter property over to A∞ := `∞ (A)/c0 (A) ?
Can one at least show that A∞ is residually antiliminary if A is residual antil-
iminary?
It can be reduced to the separable case, because one can consider finite free pro-
ducts of C0 ((0, 1], Mn ), n = 2, 3, . . .. And build suitable inductive limits ...? Look
if ultrapowers of this inductive limits have ultra-powers with non-zero characters.
One could try to construct an inductive limit of free products of C0 ((0, 1], Mn )
that is “universal” for residual antiliminary separable C *-algebras.
Proof of l.p.i. ⇒ l.p.i.(1):
Definition 2.0.3 implies l.p.i.(1): If ρ is a pure state on E then there exists,
e.g. by cf. Lemma 2.1.15(ii), a contraction a ∈ E+ with ρ(a) = 1. We can extend
ρ to a pure state on A. Thus, there exists a stable C *-subalgebra D ⊆ aAa ⊆ E
with ρ|D 6= 0 by Definition 2.0.3.
Next ques should be somewhere discussed
Question:
Let A a residually antiliminary C *-algebra and b ∈ A+ with kbk = 1 and ρ a
state on A with ρ(b) > 2/3. Does there exist a general continuous function f (t) > 0
with limt→0 f (t) > 0 such that there exist always an n-homogenous element a ∈ A+
with a min((1 + 1/n)b, 1) = a and ρ(a) > f (1/n) ?????
Sort above Lemmata and its Proofs!
Until here some are on wrong place ??

The following Proposition 2.7.16 shows that possible differences between the
definitions of Properties pi-n and pi(n) disappear if the multiplier algebra M(A)
of A has a properly infinite unit.
 7. RESIDUALLY ANTILIMINARY C *-ALGEBRAS 255

Proposition 2.7.16.
Suppose that A satisfies the following properties (i) and (ii) :

(i) For each n-homogenous element a ∈ A+ and each b ≥ 0 in the closed ideal
J(a) of A generated by a there exists a sequence of elements d1 , d2 , . . . ∈ A
with a + b = limn d∗n adn .
(ii) A has no irreducible representations ρ : A → L(H) of dimension ≤ n (of
the Hilbert space H).

Then every non-zero n-homogenous positive element is properly infinite inside

the C*-algebra A .

We do not require in Part (i) that ab = 0 for the in Part (i) considered sums
a + b.
The property in Part (i) holds for A if each n-homogenous element of A is
properly infinite, but Part (ii) does not follow from Part (i), e.g. for the C *-algebra
A := O2 ⊕Mn−1 . (Each non-zero n-homogenous element a = x⊕y satisfies a = x⊕0
with x 6= 0 and J(a) = O2 ⊕ 0. Then there exists d with d∗ ad = 1 ⊕ 0.)
A trivial example is A := Mn−1 , it satisfies condition (i) trivially, because only
0n−1 is n-homogenous in Mn−1 , but A satisfies not (ii).

Proof. The idea of the proof are the following:

(α) The assumptions (i) and (ii) pass to all non-zero quotients of all non-zero
hereditary C *-subalgebras of A.
(In particular, A is residually antiliminary in sense of Definition 2.7.2, because
this permanence implies that non-zero hereditary C *-subalgebras can not have
contractions.)
(β) It follows from this permanence properties that it suffices to show that
each non-zero n-homogenous positive contraction X ∈ A+ is not finite in A (i.e., is
infinite in A).
(γ) To show (β) we shall find for the given X non-zero n-homogenous positive
contractions T ≤ S ≤ X and R ≤ X such that R + S = X, T S = T 6= 0, S in the
ideal generated by R and T R = 0.
Then it follows that X - R, because X = R + S is in the ideal generated by
the n-homogenous R. This implies together with R + T ≤ R + S = X, that X ≈ R
and then that R ≈ T + R.
The relations R ≈ T + R, T 6= 0 and T R = 0 imply that R is infinite in A.
Thus X is infinite in A.
Now we prove this claims and carry out this constructions explicitly:
The Part(i) says that if ϕ : C0 (0, 1] ⊗ Mn → A is a C *-morphism, then for
c := ϕ(f0 ⊗ 1n ) holds: If b is in the closed ideal generated by a := ϕ(f0 ⊗ p11 )
then b - c. Notice that c is MvN-equivalent in Mn (A) ⊂ M∞ (A) to the n-fold sum
diag(a, a, . . . , a) = a ⊕ a ⊕ · · · ⊕ a .
256 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

If expressed in W (A) it says:

If b ∈ A satisfies [b] ≤ m[a] then this implies [b] ≤ min(m, n)[a].
It implies that for every nonzero n + 1-homogenous element b := ψ(f0 ⊗ 1n+1 )
the n-homogenous element c := ψ(f0 ⊗ 1n ) is properly infinite in A, because it says
in W (A) that [b] = n[a] + [a] = [c] + [a] ≤ n[a] = [c] and then inductively (over k)
that [c] + k[a] ≤ [c] and finally that [c] + [c] ≤ [c], i.e., that c is properly infinite in
A.
The arguments show also that A has the property that all n-homogenous ele-
ments c = ψ(f0 ⊗1n ) of A “absorb” all elements b ∈ A+ – in the sense that b⊕c - c
- with bc = 0 and and b is in the closed ideal of A that is generated by c ( 41 ).
It is easy to see that this property passes to all non-zero hereditary C *-
subalgebras D ⊆ A of A.
The passage to the quotient goes as follows: Every selfadjoint contraction b−c ∈
A/J (with bc = 0, and 0 ≤ b ≤ 1 and 0 ≤ c ≤ 1) can be lifted to a selfadjoint
contraction in A, and every n-homogenous positive contraction in A/J can be lifted
to an n-homogenous contraction in A, cf. Proposition A.8.4 (or [540, cor. 3.8]).
Now we show by an indirect argument that this implies that the C *-algebra A
is residually antiliminary in the sense of Definition 2.7.2 using the characterization
in Part (iv) of Proposition 2.7.7:
Let ρ : A → L(H) an irreducible representation of A. Suppose that there exists
a positive contraction in a ∈ A+ with 0 6= ρ(a) ∈ K(H). Then the dimension
of H is ≥ n + 1 by assumption in Part (ii) of Proposition 2.7.16. Moreover then
ρ(A)∩K(H) is a non-zero C *-subalgebra of K(H) that acts irreducibly on H. Thus,
ρ(A) ∩ K(H) = K(H), and we find an projection P = P ∗ P ∈ K(H) of rank n + 1.
It follows that there exists a C *-subalgebra B of A such that ρ|B is a epi-
morphism from B onto P K(H)P ∼ = Mn+1 . The projectivity of C0 ((0, 1], Mn+1 ),
cf. Proposition A.8.4 (or [540, cor. 3.8]), shows that there exists an (n + 1)-
homogenous element b := ψ(f0 ⊗ 1n+1 ) ∈ B ⊆ A with ρ(b) = P . But above
we have seen that [b] = (n + 1)[ψ(f0 ⊗ p11 )] satisfies 2[b] ≤ [b], i.e., the b is properly
infinite in A. It is easy to see that non-zero image P of the properly infinite element
b ∈ A are properly infinite. Since P is not properly infinite this contradicts the
existence of P and that of non-zero elements in ρ(A) ∩ K(H).
Hence, ρ(A) ∩ K(H) = {0} for every irreducible representation ρ : A → L(H)
of A, i.e., A is residually antiliminary.
Change notation, use only two functions f1 , f2 .!?
Remove blue repeat of assumptions?
Suppose that each n-homogenous element a ∈ A+ has the following property:

41 It seems that this property is not equivalent to the stated assumptions on A.

 7. RESIDUALLY ANTILIMINARY C *-ALGEBRAS 257

For every positive element b ∈ span(AaA) =: J(a) there exist d1 , d2 , . . . ∈ A

with limk d∗k adk = a + b, and that A has no irreducible representation of dimension
≤ n.
We show the following observations (1.0)-(1.3), (2.1) and (2.2), and get finally
in (3.0) the desired conclusion.
(1.0) The C *-algebra A is residual antiliminary in sense of Definition 2.7.2.
(1.1) The properties (i) and (ii) of A pass to each non-zero hereditary C *-
subalgebra D of A, and to all quotients A/I for closed ideals I 6= A .
(It is done at the beginning ?)
(1.2) All non-zero (n + 1)-homogenous elements of (A/I)+ are infinite in A/I .
(It is done at the beginning ?)
(1.3) Non-zero quotients A/I of A can not contain a “minimal” projection p ∈
A/I with p(A/I)p = Cp. (Should be equivalent to “A is residual antiliminary”???)
In particular, any irreducible representation of A does not contain a non-zero
compact operator in its image, and each non-zero hereditary C *-subalgebra D of
A has no character.
(1.3) has been discussed further above or below. Synchronize!
(2.1) Every non-zero n-homogenous a ∈ A+ is infinite in A.
(2.2) Every non-zero n-homogenous a ∈ (A/I)+ is infinite in A/I for every
closed ideal I 6= A of A. [Follows immediately from (2.1) and (1.1) by replacing A
by A/I.]
(3) Final conclusion from (2.1) and (2.2): Every non-zero n-homogenous ele-
ment a ∈ A+ is properly infinite in A, because it says equivalently – by Lemma
2.5.3(v) – that πI (a) is infinite in A/I or a ∈ I for every closed ideal I of A.
Ad(1.1): Let {0} 6= D ⊆ A a hereditary C *-subalgebra of A and 0 6= a ∈ D+ ,
b ∈ D+ in the closed ideal span(DaD) of D generated by a. Then b ∈ span(AaA)
and there exist d1 , d2 , . . . ∈ A with limk d∗k adk = a + b. The elements ek :=
a1/mk dk (a + b)1/k are in D and satisfy limk e∗k aek = a + b if the mk > k are
chosen such that ka1/2 a1/mk − a1/2 k < 2−n · (kdk k2 + 1)−1 .
If ρ : D → L(H) is an irreducible representation ??????????
properties (i) and (ii) of A pass to each non-zero
?? hereditary ??
C *-subalgebras D of A :
Let I 6= A a closed ideal of A. Then A/I can not have an irreducible represen-
tation ρ of dimension ≤ n, because ρ ◦ πI would be an irreducible representation of
A on a Hilbert space of dimension ≤ n, that we have excluded by assumption (ii).
258 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Now let 0 6= c ∈ (A/I)+ an n-homogenous element and g ∈ (A/I)+ in the ideal

generated by c. We can suppose that kck = 1, otherwise replace c by kck−1 c. Then
there exists a C *-morphism ψ : C0 ((0, 1], Mn ) → A/I with c := ψ(f0 ⊗ 1n ).
The C *-algebra C0 ((0, 1], Mn ) is projective, cf. Proposition A.8.4. Thus there
exists a C *-morphism ϕ : C0 ((0, 1], Mn ) → A with πI ◦ ϕ = ψ. The element
a := ϕ(f0 ⊗ 1n ) ∈ A+ satisfies πI (a) = c.
Let ε ∈ (0, 1). There are h1 , h2 , . . . , hm ∈ A with kg − k πI (hk )∗ cπI (hk )k <
P

ε/2 , because the element g is contained in the ideal generated by c.

P ∗
Property (i) of n-homogenous a ∈ A+ and that b := k hk ahk give the exis-
tence of d ∈ A with k(a + b) − d adk < ε/2 . Thus k(c + g) − πI (d)∗ cπI (d)k < ε .
∗

This shows Part (1.1), i.e., that A/I satisfies the assumptions (i) and (ii) – with
c and A/I in place of a and A.
Ad(1.2): Let c ∈ A+ an (n + 1)-homogenous element. By Lemma 2.5.3(v) it
suffices to show that πI (c) is infinite in A/I for all closed ideals I of A with c 6∈ I.
By (1.1) the non-zero quotients A/I of A satisfy again conditions (i) and (ii) with
0 6= πI (a) ∈ (A/I)+ in place of 0 6= a ∈ A+ . Therefore, it suffices to show that
each non-zero (n + 1)-homogenous element in A+ is infinite in A:
If ψ : C0 (0, 1]⊗Mn+1 → A is a non-zero C *-morphism and c := ψ(f0 ⊗1), then
a := ψ(f0 ⊗(1−p11 ) is an n-homogenous element and b := ψ(f0 ⊗p11 ) is in the ideal
1/2
generated by a, because b = zz ∗ and z ∗ z = ψ(f0 ⊗ p22 ) ≤ a for z := ψ(f0 ⊗ p12 ).
Thus, b + a = limn d∗n adn for a suitable sequence d1 , d2 , . . . ∈ A by assumptions on
A and a. In particular c = b + a - a, ab = ba = 0 and b 6= 0, i.e., a is infinite in
sense of Definition 2.5.1. Then a ≤ c - a shows that c ≈ a, and c is infinite in A.
Ad(1.3): Suppose that p ∈ A/I is a projection with p(A/I)p = C · p. Then the
closed ideal K of A/I generated by p is isomorphic to K(H) ∼
= K for some Hilbert
space H .
Suppose that Dim(H) > n . Then there exists an n + 1-homogenous element
in K (given by any projection p ∈ K of rank = n + 1). But this is not possible,
because all n + 1-homogenous elements in A/I are properly infinite by Step (1.2).
It follows that Dim(H) ≤ n. Then K(H) ∼ = K is unital and its unit is in
the center of A/I. It would define an irreducible representation of A of dimension
≤ n. But this has been excluded by the assumption (ii) on A. Hence, a minimal
projection p can not exist in any quotient of A.
If D is a non-zero hereditary C *-subalgebra of A ?????
Ad(1.4): Let {0} 6= D ⊆ A a hereditary C *-subalgebra of A and 0 6= Y ∈ D+
an n-homogenous element and b ∈ D+ in the ideal J(Y ) generated by Y .
(1.4.a) An element b ∈ D is in the ideal of D generated by Y inside D, if and
only if, b is in the ideal J(Y ) of A generated by Y , i.e., D ∩ J(Y ) = I(Y ) for the
ideal of I(Y ) of D generated by Y .
Compare here also Remark A.5.8 or Lemma 2.7.3!
 7. RESIDUALLY ANTILIMINARY C *-ALGEBRAS 259

This is because if f1 , . . . , fm ∈ A satisfy kb − k fk∗ Y fk k < ε then for suitable

P

β ∈ (0, 1/2) and gk := Y β fk bβ holds kb − k gk∗ Y gk k < ε .

P

Moreover then there exists a sequence of elements dn ∈ A with limn d∗n Y dn =

Y +b. Again we can suppose that the dn are in D, by replacing the dn by Y βn dn (Y +
b)βn for a suitable zero sequence βn ∈ (0, 1/2) .
(1.4.b)
D has no irreducible representation of dimension ≤ n:
An irreducible representation of D of finite dimension extends to an irreducible
representation ρ of A that contains the compact operators in its image. Let I denote
the kernel of ρ. It would imply that a quotient A/I of A contains a projection p
with p(A/I)p = C · p. We have seen in
Ad(1.2)
that this is impossible.
Thus, D has no irreducible representation of finite dimension. In particular, D
has no character.
Ad(2): Every nonzero n-homogenous Y ∈ A+ is infinite in A:
Let Y ∈ A+ non-zero and n-homogenous. Since Y ≈ kY k−1 Y , we can suppose
that kY k = 1. Then there exists ψ : C0 (0, 1] ⊗ Mn → A with ψ(f0 ⊗ 1n ) = Y ,
where f0 (t) = t on [0, 1].
Consider the functions f1 (t) := min(3t, 1), f2 (t) := min(max(3t − 1, 0), 1), and
f3 (t) := max(3t−2, 0) in C0 (0, 1]+ . They are increasing with fk (1) = 1 and fk (0) =
0 (k = 1, 2, 3) and satisfy f0 ≤ f1 ≤ 3f0 , f2 f1 = f2 , f3 f2 = f3 , f1 + f2 + f3 = 3f0 ,
f3 = (3f0 − 2)+ , f2 = (3f0 − 1)+ − (3f0 − 2)+ and f1 = 3f0 − (3f0 − 1)+ .
Let X1 := 3Y −(3Y −1)+ = ψ(f1 ⊗1n ), X2 := (3Y −1)+ −(3X−2)+ = ψ(f2 ⊗1n )
and X3 := (3Y − 2)+ = ψ(f3 ⊗ 1n ). Then the Xk are all n-homogenous, have
norms kXk k = 1, and Xk+1 = Xk+1 Xk (for k = 1, 2). Moreover Y ≤ X1 ≤ 3Y , in
particular Y ≈ X1 .
The element Z := ψ(f3 ⊗p11 ) has norm kZk = kX3 k = 1, because the elements
ψ(f3 ⊗ pjj ) (j = 1, . . . , n) are pairwise orthogonal, are MvN-equivalent and their
sum is X3 = ψ(f3 ⊗ 1n ). It implies kψ(f3 ⊗ pjj )k = kX3 k = 1 for 1 ≤ j ≤ n.
Moreover we get that Zψ(f2 ⊗ p11 ) = Z = Zψ(f1 ⊗ p11 ) because f3 f2 = f3 =
f3 f1 . Let D := ZAZ the hereditary C *-subalgebra of A generated by Z.
In Ad(2) we have seen that D has no character. In particular D is not Abelian.
A non-abelian C *-algebra D contains a nonzero element x ∈ D with kxk = 1
and x2 = 0. I.e., D contains a 2-homogenous element ϕ : C0 (0, 1] ⊗ M2 → D with
ϕ(f0 ⊗ p12 ) = x. In particular kϕ(f0 ⊗ 12 )k = 1.
(It can be seen from the existence of an irreducible representation of dimension
≥ 2 and a strong variant of the Kadison transitivity theorem, - Lemma 2.1.15(ii)-,
and projectivity of C0 (0, 1] ⊗ M2 .
260 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Give ref’s or cite!?)

Now we use the natural isomorphism X3 AX3 ∼ = D ⊗ Mn for X3 := ψ(f3 ⊗ 1n )
in a way that ψ(f3 ⊗ pjk ) corresponds to to Z ⊗ pjk ∈ D ⊗ Mn . Notice that
X2 (Z ⊗ pjk ) = (Z ⊗ pjk ) = (Z ⊗ pjk )X2 under this natural identification.
We build the elements g1 , g2 ∈ C0 (0, 1] by g1 := 2f0 − (2f0 − 1)+ = min(2f0 , 1)
and g2 := 2f0 − min(2f0 , 1) = max(2f0 − 1, 0). Notice that g1 + g2 = 2f0 , g2 =
(2f0 − 1)+ , g1 g2 = g2 and g1 (1) = g2 (1) = 1. In particular kϕ(g1 ⊗ ejk )k = 1 =
kϕ(g1 ⊗ 12 )k, kϕ(g2 ⊗ 12 )k = 1 and ϕ(g1 ⊗ 12 )ϕ(g2 ⊗ 12 ) = ϕ(g1 ⊗ 12 ).
Let E ⊆ A the hereditary C *-subalgebra of A defined by

E := ψ(f1 ⊗ 1n )Aψ(f1 ⊗ 1n ) .

There is a natural isomorphism γ from E onto the C *-algebra F ⊗ Mn , where F is

the hereditary C *-subalgebra of A given by

F := ψ(f1 ⊗ p11 )Aψ(f1 ⊗ p11 ) .

Notice that ψ(f1 ⊗p11 )ϕ(h) = ϕ(h) for all h ∈ C0 (0, 1]⊗M2 , because ψ(f1 ⊗p11 )d =
d for all d ∈ D.
We define S := ϕ(g2 ⊗ e11 ) ⊗ 1n , T1 := ϕ(g1 ⊗ e11 ) ⊗ 1n , T2 := ϕ(g1 ⊗ e22 ) ⊗ 1n ,
R := X1 − T1 = [ψ(f1 ⊗ p11 ) − ϕ(g1 ⊗ e11 )] ⊗ 1n . Notice that T1 S = S implies
RS = 0, that S ≤ T1 ∼M vN T2 and T2 ≤ R. It follows that T1 and S are in closed
ideal generated by R. Since ψ(f1 ⊗ p11 )ϕ(g1 ⊗ ejj ) = ϕ(g1 ⊗ ejj ) it follows that R
is a positive n-homogenous element in A and that RT2 = T2 = T2 R implies that
T2 ≤ R and kRk = 1.
Clearly R ≤ X1 and X1 = R + T1 . Thus R - X1 . Since R is n-homogenous
and T1 ≈ T2 ≤ R it follows by assumption on n-homogenous elements in A that
X1 = R + T1 - R.
Thus, X1 ≈ R. It implies that the X1 is infinite in A if and only if R is infinite
in A. Since Y ≈ X1 (by Y ≤ X1 = 3Y − (3Y − 1)+ ≤ 3Y ) this would cause also
that Y is infinite in A.
Since RS = 0, kSk = 1 and S ≤ T1 ≈ T2 ≤ R we have by assumption on
n-homogenous elements R ∈ A+ that there exist d1 , d2 , . . . ∈ A with S + R =
limn d∗n Rdn . This implies S + R - R. On the other hand (S + R) ⊕ 0 =
[S 1/2 , R1/2 ][S 1/2 , R1/2 ]∗ and S ⊕ R = diag(S, R) = [S 1/2 , R1/2 ]∗ [S 1/2 , R1/2 ] in
M2 (A), i.e., (S + R) ⊕ 0 ∼M vN S ⊕ R in M2 (A). It implies that S ⊕ R - R ⊕ 0 in
M2 (A). Thus R, X1 and Y are infinite elements of A+ .
Since the considered property of Y pass to each quotient A/I of A with Y ∈
6 I,
it follows finally that Y is properly infinite in A by Lemma 2.5.3(v). 

The following Corollaries 2.7.18 and 2.7.19 are consequences of Proposition

2.7.16.
But we have here inserted also other permanence properties.
Move this to other places.
 7. RESIDUALLY ANTILIMINARY C *-ALGEBRAS 261

Proposition 2.7.17. The Property pi-n of a C*-algebra A implies that A has

Property pi(n).
Both properties imply that all non-zero n-homogenous elements in A are prop-
erly infinite.
Both properties pass to non-zero hereditary C*-subalgebras and to non-zero quo-
tients.
If, for each a ∈ A+ and ε ∈ (0, kak), there exists elements d1 , . . . , dn ∈ A with
d∗j dk= δjk (a − ε)+ , then Property pi(n) implies also Property pi-n.
Not verified, needs special topology of prime(A)!:
??? And then both are equivalent to the property that A has no irreducible
representation of dimension ≤ n and all n-homogenous elements are properly infi-
nite.???
The Property pi(n) (and therefore also the Property pi-n) implies that A is
residually antiliminary in sense of Definition ??.
The algebra K(H) of compact operators in any Hilbert space H can not have
one of the properties pi-n or pi(n). In particular, C*-algebras with this properties
have no irreducible representations of finite dimension.

Proof. The Property pi-n implies Property pi(n):

Let a ∈ A+ , b ∈ A+ in the closed ideal generated by a, and ε > 0, then there
Pm
exists δ > 0 and d1 , . . . , dm ∈ A with k=1 dk (a − 2δ)+ (dk )∗ = (b − ε)+ . Let ` ∈ N
with n ≤ (` − 1) · n ≤ m ≤ ` · n. Then the row-matrix D ∈ M1,m with entries
[d1 , . . . , dm ] in M1,m satisfies D((a − 2δ)+ ⊗ 1m )D∗ = (b − ε)+ .
By assumption of Property pi-n, the diagonal matrix

diag((a − γ)+ , . . . , (a − γ)+ ) = (a − γ)+ ⊗ 1n

is properly infinite inside Mn (A) for each γ ∈ (0, kak). Thus, there exists an
operator matrix R ∈ Mm,n (A) with

R((a − δ)+ ⊗ 1n )R∗ = (a − 2δ)+ ⊗ 1m .

The row matrix D · R ∈ M1,n (A) satisfies (D · R)((a − δ)+ ⊗ 1n )(D · R)∗ = (b − ε)+ .
This shows that an element b in the in the in closed ideal generated by a non-zero
element a ∈ A can be approximated by sums of n products cad ∈ A · a · A.
The Properties pi-n and pi(n) pass to non-zero hereditary C *-subalgebras D,
in particular to non-zero closed ideals J, and they pass to all non-zero quotients.
What happens with stabilizations? A ⊗ Mn ?
The algebra K(H) of compact operators in any Hilbert space H can not have
one of the properties pi-n or pi(n). In particular, C *-algebras with this properties
have no irreducible representations of finite dimension. 
262 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Corollary 2.7.18. If the C*-algebra A has Property pi-n of Definition ?? or

Property pi(n) of Definition 2.0.4, then every non-zero n-homogenous element of
A+ is properly infinite.

Proof. The Property pi-n implies Property pi(n) by Lemma 2.1.2

Let a ∈ A+ , b ∈ A+ in the closed ideal generated by a, and ε > 0, then there
Pm
exists δ > 0 and d1 , . . . , dm ∈ A with k=1 dk (a − 2δ)+ (dk )∗ = (b − ε)+ . Let ` ∈ N
with n ≤ (` − 1) · n ≤ m ≤ ` · n. Then the row-matrix D ∈ M1,m with entries
[d1 , . . . , dm ] in M1,m satisfies D((a − 2δ)+ ⊗ 1m )D∗ = (b − ε)+ .
By assumption of Property pi-n, the diagonal matrix

diag((a − γ)+ , . . . , (a − γ)+ ) = (a − γ)+ ⊗ 1n

is properly infinite inside Mn (A) for each γ ∈ (0, kak). Thus, there exists an
operator matrix R ∈ Mm,n (A) with

R((a − δ)+ ⊗ 1n )R∗ = (a − 2δ)+ ⊗ 1m .

The row matrix D · R ∈ M1,n (A) satisfies (D · R)((a − δ)+ ⊗ 1n )(D · R)∗ = (b − ε)+ .
This shows that an element b in the closed ideal generated by a non-zero element
a ∈ A can be approximated by sums of n elements cad in A · a · A.
The Properties pi-n and pi(n) pass to non-zero hereditary C *-subalgebras D,
in particular to non-zero closed ideals J, and they pass to all non-zero quotients.
What happens with stabilizations? A ⊗ Mn ?
The algebra K(H) of compact operators in any Hilbert space H can not have
one of the properties pi-n or pi(n). In particular, C *-algebras with this properties
have no irreducible representations of finite dimension.
Let ρ a pure state on A, L := {a ∈ A ; ρ(a∗ a) = 0}, H := A/L, and D : A →
L(H) the corresponding irreducible representation. Then there exists a contraction
e ∈ A+ with ρ(e) = 1 = kρk. By Property pi-n, the diagonal matrix diag(e, . . . , e)
is infinite in Mn (A).
In case of Property pi-n, the Definition ?? rather directly implies that all n-
homogenous non-zero elements in A with Property pi-n are properly infinite in
A:
If a ∈ A+ is n-homogenous and kak = 1, then there exists a C *-morphism
ψ : C0 (0, 1] ⊗ Mn → A with ψ(f0 ⊗ 1n ) = a, where f0 (t) = t for t ∈ [0, 1].
Let F := f0 ⊗ 1n G := f0 ⊗ e11 in C0 (0, 1] ⊗ Mn , the equivalence ψ(F ) ⊗
e11 ∼M vN ψ(G) ⊗ 1n in A ⊗ Mn and an isomorphism

A ⊇ ψ(F )Aψ(F ) ∼
= Mn ( ψ(G)Aψ(G) ) ⊆ Mn (A)

that maps the n-homogenous element ψ(F ) ∈ ψ(F )Aψ(F ) ⊆ A to the element
ψ(G) ⊗ 1n ∈ A ⊗ Mn . The element ψ(G) ⊗ 1n is properly infinite in A ⊗ Mn
by Definition ??. It is then automatic also properly infinite inside the heredi-
tary C *-subalgebra ψ(G)Aψ(G) ⊗ Mn of A ⊗ Mn generated by ψ(G) ⊗ 1n . But
 7. RESIDUALLY ANTILIMINARY C *-ALGEBRAS 263

there is an isomorphism from the C *-algebra Mn ( ψ(G)Aψ(G) onto the heredi-

tary C *-subalgebra aAa = ψ(F )Aψ(F ) with the n-homogenous element ψ(F ) = a
corresponding to the in ψ(G)Aψ(G) ⊗ Mn properly infinite ψ(G) ⊗ 1n . Thus, the
n-homogenous a ∈ A+ is properly infinite in A, and this shows that the property
pi-n of A implies that each non-zero n-homogenous element ψ(F ) ∈ A is properly
infinite in A.
In the case of Property pi(n) the proper infiniteness of non-zero n-homogenous
elements can be derived with help of the (not so trivial) Proposition 2.7.16:
The Definition 2.0.4 requires (!) that A has no irreducible representation on a
Hilbert space of dimension ≤ n and that A has the property that each element c
in the closed ideal J(a) generated by a ∈ A satisfies [c] ≤ n[a] in the (large) Cuntz
algebra Cu(A) .
If moreover a, c ∈ A+ are non-zero then the latter means that, for each ε > 0
there exists δ ∈ (0, kak) and d1 , . . . , dn such that

(c − ε)+ = d∗1 (a − δ)+ d1 + . . . + d∗n (a − δ)+ dn .

In special case where a ∈ A+ is n-homogenous and kak = 1, there ex-

ists ψ : C0 (0, 1] ⊗ Mn → A such that a = ψ(F ) for F := f0 ⊗ 1n . Let
b := ψ(f0 ⊗ p11 ) ∈ A+ . Then the element c ∈ J(a) = J(b) satisfies also that
c ∈ J(b) and now [c] ≤ n[b] = [a].
Thus, we can in this special case the above defined n-homogenous element a :=
ψ(F ), the element b := ψ(G) and get for every c ∈ J(a)+ , that c + a ∈ J(a) = J(b)
and [c + a] ≤ n[b] = [a]. It gives that c + a - a in A for all n-homogenous
a ∈ A+ and c ∈ J(a)+ . The additional requirement that A has no irreducible
representation of dimension ≤ n allows to apply Proposition 2.7.16 to get that all
non-zero n-homogenous elements of A are properly infinite in A. 

The Corollary 2.7.18 implies immediately the following Corollary:

Corollary 2.7.19. If the unit element 1M(A) of the multiplier algebra is prop-
erly infinite, i.e., if there exists isometries S, T ∈ M(A) with S ∗ T = 0, then A has
Property pi-n of Definition ??, if and only if, A has Property pi(n) of Definition
2.0.4.

Check next proof again. Something not clear?

Proof. Since M(A) is unitally contained in M(`∞ (A)), M(A∞ ) and M(Aω ),
it follow that this algebras have no finite-dimensional quotient C *-algebras.
In fact, the properties pi-n and pi(n) are equivalent if for every element a ∈ A+
and ε ∈ (0, kak) there exist elements b1 , . . . , bn ∈ A with b∗j bk = δj,k (a − ε)+ . In
1/2
case of Corollary 2.7.19 one can use bk := S k T · (a − ε)+ , for k = 1, . . . , n .
Pn ∗
The b := k=1 bk bk ∈ A+ is a properly infinite n-homogenous element and
b ⊗ p11 is approximately equivalent in Mn (A) to the diagonal n × n-matrix C :=
diag((a − ε)+ , . . . , (a − ε)+ ) ∈ Mn (A).
264 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

But the question is if C is properly infinite in Mn (A), i.e., if C ⊕0 ≈ C ⊕(a−ε)+

in Mn+1 (A).
Pm
If c = k=1 d∗k adk , then for every ε > 0 there exists δ > 0 and ej ∈ A such
Pn ∗
that (c − ε)+ = j=1 ej (a − δ)+ ej .

Thus ??????
Then k (bk )∗ bk = n · (a − δ)+ and k bk (bk )∗ is an n-homogenous element
P P

that is equivalent to (a − ε)+ ⊗ 1n in Mn (A).

???? 

Next BLUE is not proven yet. Perhaps it is wrong ...

The free product A := C1 ∗ C2 ∗ · · · of a sequence of C1 , C2 , . . ., where the Cn
are copies Cn ∼= C0 ((0, 1], K(`2 (N))) of the cones over the compact operators, has
no irreducible representation of finite dimension but the ultrapower Aω of A has
a character (and has irreducible representation of each finite dimension). But this
example A is not residually antiliminary:
A has e.g. the compact operators K on `2 (N) as its quotient, because K is the
quotient of one of the freely generating copies Cn of C0 ((0, 1], K).
If there is an example of a separable C *-algebra A that has no characters but
A∞ has (non-zero) characters then the free product F of countably many copies
of M2 (C0 (0, 1]) must have this property: The algebra A must be a quotient of
F , because the C *-subalgebra B of A generated by a dense sequence in the set
of all 2-homogenous contractions contains all 2-homogenous elements in A+ . It is
then not difficult to see that for each a ∈ A+ the closed hereditary C *-subalgebra
aAa contains a positive contraction b ∈ B+ such that B ∩ aAa = bBb and that
alternatively bAb = aAa or bAb is an ideal of aAa with the property that aAa/bAb
is commutative.
The convex cone K2 in A+ generated by all 2-homogenous positive contractions
in A should be identical with A+ , and F has no characters.
It follows from the invariance of K2 under automorphisms of A and that the
(norm-) closure of K2 is hereditary and K2 separates the pure states of A (by
Kadison transitivity theorem, cf. Lemma ??).
It needs a Lemma that says if aAa has no characters and 0 ≤ a ≤ b1 + . . . + bn ,
with bk all 2-homogenous and positive, than a can be approximated by convex
combinations of positive 2-homogenous elements in aAa (In case that A is residually
antiliminary this happens for every non-zero a ∈ A+ ).

8. Basics on quasi-traces

Begin of eksec2-Part2.tex !!!

The main destinations are:
 8. BASICS ON QUASI-TRACES 265

∗ ∗
1) If A ⊗ Cred (F ) has no non-trivial l.s.c. (additive) trace, then A ⊗ Cred (F ) is
s.p.i., if F is one of the free groups Fn , n ∈ {2, 3, . . .} ∪ {∞} .
∗
(The exactness of the reduced group C *-algebra Cred (F ) ensures that the ques-
∗
tions are reduced to the observation that all quasi-ideals of M ⊗ Cred (F ) are ideals
∗
and to the non-existence of central states on M if M ⊗ Cred (F ) has no non-zero
central state.)
2) If Aω has no non-trivial l.s.c. 2-quasi-trace then Aω is weakly purely infinite
(it implies that A is weakly p.i.).
3) If A ⊗ Z has no non-trivial l.s.c. quasi-trace, then A ⊗ Z is s.p.i.
Work here with Rørdam condition on the Cuntz semi-group Cu(A) and using
central sequences of the join algebra E(Mp , Mq ) in Z .
4) Permanence properties for s.p.i. algebras (including extensions, coronas etc.)
tensor with exact C *-algebras ...
5) Special cases of continuous fields with s.p.i. fibres ...
6) A ⊗ O∞ ∼
= A if A is separable, nuclear and s.p.i.
More ?????
Before we give in the next section some results concerning the relations between
2-quasi-traceless C *-algebras and purely infinite C *-algebras, we report some basic
knowledge on 2-quasi-traces, give necessary definitions and explain results that we
apply here.
We have all the Definitions in the first section
of Chapter 2 or in the Introduction

Definition 2.8.1. A map τ : A+ → [0, ∞] is a quasi-trace if τ (0) = 0 and

τ (a + b) = τ (a) + τ (b) for all commuting a, b ∈ A+ , and τ (a∗ a) = τ (aa∗ ) for all
a ∈ A . It is lower semi-continuous (l.s.c.) if τ (a) = supδ>0 τ ((a − δ)+ ) for each
a ∈ A+ ( 42 ). If µ : A+ → [0, ∞] is any quasi-trace then µ∗ (a) := supδ>0 µ (a−δ)+

is an l.s.c. quasi-trace. It satisfies µ∗ ≤ µ

Check this: and µ(a) = µ∗ (a) for all a ∈ A+ with µ(a) < ∞ .
The quasi-trace τ is a 2-quasi-trace if there is a quasi-trace

τ2 : M2 (A)+ → [0, ∞]

such that τ (a) = τ2 (a ⊗ p11 ) for all a ∈ A.

We say that A is (2-quasi-) traceless (respectively A is traceless) if all lower
semi-continuous 2-quasi-traces (respectively all l.s.c. and additive τ ) on A+ take
only the values 0 and +∞.

42 Then τ is an l.s.c. function from A into [0, +∞] in the ordinary sense with respect to the
+
norm-topology on A+ and the Hausdorff topology on [0, +∞] ∼
= [0, 1].
266 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Hold next ready for later needed reference!

But shift all further explanations to Appendix B.
Including all hints for proofs.

Proposition 2.8.2. Let τ : A+ → [0, ∞] a lower semi-continuous 2-quasi-trace.

Then

(1) τ : A+ → [0, ∞] is an order-monotone map on A+ ,

(2) τ (a + b) ≤ 2(τ (a) + τ (b)) for all a, b ∈ A+ ,
(3) {a ∈ A ; τ (a∗ a) = 0} is a closed ideal of A,
(4) {a ∈ A ; τ (a∗ a) < ∞} is an algebraic *-ideal of A.
(5) If U := M2 ⊗M3 ⊗· · · denotes the universal UHF algebra then there exists
a unique l.s.c. “local” quasi-trace µ : (U ⊗ A)+ → [0, ∞] with µ(1 ⊗ a) =
τ (a) for all a ∈ A+ .
(6) For each n ∈ N there exists a unique lower s.c. quasi-trace
τn : (A ⊗ Mn )+ → [0, ∞]
with τ (a) = τn (a ⊗ e11 ) for all a ∈ A.

Every l.s.c. “local” quasi-trace on a C *-algebra A of real rank zero, e.g. on an

AW*-algebra or a Rickard algebra, is a 2-quasi-trace. (And it can be interpreted
as some sort of evaluations on the J. von Neumann “perspectives” for the related
affine geometry.)
It is (2019?) a still open question if unital 2-quasi-traces on AW*-algebras are
always additive, i.e., are unital traces.
The minimal requirements on a C *-algebra A for having the property that all
quasi-traces on A are 2-quasi-traces has to do with the question if all quasi-ideals
of `∞ (A)/c0 (A) are ideals ... (Consider here also the quasi-traces that take only
the values 0 and ∞ ...
But there exists a unital l.s.c. quasi-traces on C([0, 1]) ∗ C([0, 1]) and on
∗
C (F2 ) = C(S1 ) ∗ C(S1 ) that are not a 2-quasi-traces, – do not mix it up here with
∗
the reduced group C *-algebra Cred (F2 ), which is an exact C *-algebra.
Notice here that e.g. t ∈ [0, 1] → eit ∈ S1 defines a unital epimorphism from
C(S1 ) onto C([0, 1]). Thus we have only to give a ???
The τ2 -existence condition in Definition 2.8.1 of 2-quasi-trace τ is equivalent
to each of the two following inequalities for all a, b ∈ A+ :

τ (a + b)1/2 ≤ τ (a)1/2 + τ (b)1/2 ,

(which was 1991 observed by U. Haagerup [342]), and the 2-sub-additivity:
τ (a + b) ≤ 2(τ (a) + τ (b)) .

The 2-sub-additivity follows from Haagerup’s inequality, because for x, y, z ∈

[0, +∞] the inequality x ≤ y + z implies x2 ≤ 2(y 2 + z 2 ). But it can be seen also
directly from the τ2 -existence by considering the matrix C := a1/2 ⊗ p11 + b1/2 ⊗ p21
 8. BASICS ON QUASI-TRACES 267

by using that C ∗ C = (a + b) ⊗ p11 , CC ∗ ≤ 2(a ⊗ p11 + b ⊗ p22 ) and that τ2 (a ⊗

p11 + a ⊗ p22 ) = τ (a) + τ (b) for all a, b ∈ A+ .
A now classical result of U. Haagerup [342], says that all 2-sub-additive states
on exact C *-algebra are additive.
The question if every l.s.c. 2-quasi-trace on a C *-algebra A is additive is an
open question. A possible positive answer is equivalent to the likely more than 65
(?) years old conjecture of I. Kaplansky that every finite AW*-factor is a W*-factor.
Next requires details ...
J.F. Aarnes [3] has described a rather natural quasi-state τ on C([0, 1]2 )+ that
has the property that τ |C ∗ (1, f )+ is additive if all winding numbers w(σ, f, z) of
f ∈ C([0, 1]2 ) are zero for all Jordan curves σ : S 1 → [0, 1]2 and all z ∈ C \ f (S 1 ).
There exists unital type-I C *-algebras A, given by a suitable extension 0 →
c0 (K) → A → C([0, 1]2 ) → 0, and a unital map µ : A+ → [0, ∞), defined by the
Aarnes quasi-state on C([0, 1]2 ), that has the property that the restriction µ|C+ for
each commutative C *-subalgebra C ⊂ A is additive on C+ , µ(a∗ a) = µ(aa∗ ) for
all a ∈ A, but there exists a, b ∈ A+ with µ(a) = 0, µ(b) = 0 and µ(a + b) > 2. In
particular this implies that µ is a quasi-trace state that is not a 2-quasi-trace.
An other example is the free unital product B := C([0, 1]) ? C([0, 1]) and the
unital quasi-trace given by τA ◦ π, π denotes here the canonical unital epimorphism
π : B → C([0, 1]2 ) and τA is the Aarnes quasi-state on C([0, 1]2 ), cf. [3].
(Compare Section ?? for more information on the existence of non-2-sub-
additive quasi-traces!)
In general an l.s.c. quasi-trace τ : A+ → [0, ∞] is sub-additive ( = 1-sub-
∗
additive), if and only if, τ is additive, if and only if, τ “extends” to A ⊗ Cred (F2 )
∗
in the sense that there exists an l.s.c. quasi-trace ρ on (A ⊗ Cred (F2 ))+ with
ρ(a ⊗ b) = τ (a)tr(b), where tr is the unique trace state on the reduced group
∗
C *-algebra Cred (F2 )+ of the free group F2 on two generators.
∗ ∗
(Here it is important to remind that Cred (Fn ), for n = 2, 3, . . . and Cred (F∞ )
contain the Jiang-Su algebra Z unitally, cf. [667, section 6.3].)
Give reference for above
Every l.s.c. 2-quasi-trace τ : A+ → [0, ∞] on an exact C *-algebras A is additive,
∗
because it “extends” to an l.s.c. quasi-trace ρ on (A ⊗ Cred (F2 ))+ with ρ(a ⊗ b) =
∗
τ (a)tr(b). (Here we use only that Cred (F2 ) is simple, has a unique trace state and
is unitally contained in Mω .)
Where is Mω defined?
Give example for next blue text??
Notice that on traceless C *-algebras A there can still exist non-zero 2-quasi-
traces τ on the positive part D+ of some hereditary C *-subalgebras D of A that
have also non-zero finite values on D+ , but those 2-quasi-traces can not have lower
semi-continuous extensions to A+ as 2-quasi-traces.
268 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Examples for above green text? Case of ideal?

IS IT REALLY TRUE, e.g. for simple C *-algebras and degenerate quasi-traces?
If D is a hereditary C *-subalgebra of A with l.s.c. 2-quasi-trace τ on D+ ∩ A,
then τ should extend to A.
First extend to E := I(D)+ for the ideal E := I(D) generated by D by
extending to D ⊗ K and then using that D ⊗ K ∼ = E ⊗ K for σ-unital hereditary
C *-subalgebras D and E that generate the same closed ideal of A.
Is it really on D ∼
= D ⊗ p11 the same 2-q-trace?
We recall here that not necessarily bounded lower semi-continuous 2-quasi-
traces are in one-to-one correspondence to lower semi-continuous dimension func-
tions D (= sub additive rank functions), and, therefore, are always order monotone
maps.
(what we can see directly from the definition???) See Section 7 in Appendix
A, ????? e.g. [342] and [93, sec. 2.9] more precise in BlanKir2 ???? or Appendix
A??? for more details.

Remark 2.8.3. The C *-algebra A is (2-quasi-) traceless, if and only if, A has
the property that

(∗ ) for every a ∈ A+ and ε > 0 there exists n := n(a, ε) ∈ N with

(a − ε)+ ⊗ 12n - a ⊗ 1n for all n ≥ n(a, ε) .

Compare Corollaries A.13.10 and A.7.3, or [462, prop. 5.7] for a proof.
Notice that this can be expressed equivalently in the Cuntz semigroup W (A)
as 2n · [(a − ε)+ ] ≤ n · [a] for all n ≥ n(a, ε).
Variation of ε and using that ((a − ε)+ − δ)+ = (a − (ε + δ))+ shows that

(∗∗ ) for each a ∈ A+ , ε > 0 and k > 1 there exist a number n(a, ε, k) such that
kn · [(a − ε)+ ] ≤ n[a] for all n ≥ n(a, ε, k).

In particular there exists, for n ≥ n(a, ε, k), an element d ∈ M1,k (A ⊗ Mn ) with

d∗ (a ⊗ 1n )d = ((a − 2ε)+ ⊗ 1n ) ⊗ 1k .

9. Pure infiniteness versus (2-quasi-) traceless.

Lemma 2.9.1. Let A and B C*-algebras, and denote by A ⊗ B the minimal C*-
algebra tensor product, and by P × the natural continuous map from the cartesian
product of T0 -spaces prime(A) × prime(B) into the T0 -space prime(A ⊗ B), that is
defined by
P × (I, J) := (I ⊗ A) + (B ⊗ J) .

(i) If A or B is exact, then P × is a (surjective) homeomorphism.

 9. PURE INFINITENESS VERSUS (2-QUASI-) TRACELESS. 269

(ii) Suppose that P × is surjective, D is a hereditary C*-subalgebra of A ⊗ B

and I ∈ I(A ⊗ B) is a prime ideal
Or I must it be primitive ?? Check original!
that does not contain D.
Then there exist a ∈ A+ , b ∈ B+ , c ∈ A ⊗ B, pure states λ on A, and
ρ on B such that
(α) λ ⊗ ρ(I) = {0} ,
(β) c c∗ ∈ D and c∗ c = a ⊗ b ,
(γ) λ(a) = kak = 1 and ρ(b) = kbk = 1 .
(iii) Suppose that the C*-algebras A and B have the property that P × is sur-
jective.
If A or B is (2-quasi-)-traceless then A ⊗ B is (2-quasi-)-traceless.

Notice that Part (ii) implies that – in case where P × is surjective – the set of
elements F := {a ⊗ b ; a ∈ A+ , b ∈ B+ } is a filling family for A ⊗ B in sense of
[469, def. 4.2].
Not.published? No!: It is now published, see [469]. But one
has to pay 30 British pound.
But arXiv:1503.08519v2 is fairly readable... Check typos again?
Check new references !!!

Proof. (i): See [93, prop. 2.16 and prop.2.17(2)].

(ii): See Lemma 2.2.3 (cf. also [93, lem. 2.18]),
Lemma 2.2:
Let D be a non-zero hereditary C *-subalgebra of the minimal C *-algebra tensor
product A ⊗ B of C *-algebras A and B.
Then there exists 0 6= z ∈ A ⊗ B with zz ∗ ∈ D and z ∗ z = e ⊗ f for some
non-zero e ∈ A+ and f ∈ B+ .
If d ∈ D+ and pure states ϕ ∈ A∗ , ψ ∈ B ∗ are given with (ϕ ⊗ ψ)(d) > 0, then
the element z ∈ A ⊗ B can be found such that, moreover, ϕ(e)ψ(f ) > 0.
(iii): By symmetry it suffices to consider the case where B is (2-quasi-) trace-
less. For each fixed a ∈ A+ , the map

τa : b ∈ B+ 7→ τ (a ⊗ b) ∈ [0, ∞]

is an l.s.c. 2-quasi-trace on B+ . The τa can only take the values 0 and +∞, because
B is (2-quasi-) traceless. We show that this implies that τ can take only the values
0 and +∞ by lower semi-continuity of τ :
The set of g ∈ (A ⊗ B)+ with τ (g) = 0 is the positive part J+ of a closed ideal
J of A ⊗ B by Proposition 2.8.2.
Let e ∈ (A ⊗ B)+ with τ (e) < ∞. To prove that τ (e) = 0 it suffices by lower
semi-continuity of τ to show that, for each ε > 0, the hereditary C *-subalgebra D
generated by (e − ε)+ is contained in J.
270 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Suppose that D is not contained in J. Then there exists a primitive ideal of I

A ⊗ B such that J ⊆ I and D 6⊆ I.
Let a ∈ A+ , b ∈ B+ , c ∈ A ⊗ B, λ ∈ A∗ , and ρ ∈ B ∗ as in Part (ii). Then
a ⊗ b = c∗ c 6∈ I but cc∗ ∈ D, in particular a ⊗ b 6∈ J, i.e., 0 < τ (a ⊗ b) ≤ ∞ and
τa : x ∈ B+ 7→ τ (a ⊗ x) is an l.s.c. 2-quasi-trace on B+ .
The latter contradicts the assumption that B is traceless. Thus, D ⊆ J and
τ (e) = 0 for each e ∈ (A ⊗ B)+ with τ (e) < ∞. 

The Remark 2.8.3 has the following consequences for tensor products of A with
U HF algebras, and for algebras A with (2-quasi-) traceless ultra-powers of A:

Corollary 2.9.2. Let A a ( 2-quasi- ) traceless C*-algebra, E an exact C*-

algebra and B an UHF algebra (of infinite dimension) then E ⊗ A ⊗ B is a purely
infinite C*-algebra.

In fact E ⊗ A ⊗ B is also strongly purely infinite, because one can decompose

B into a tensor product B = B1 ⊗ B2 with B1 and B2 UHF-algebras of infinite
dimensions, cf. Proposition ??.

Proof. By Lemma 2.9.1(iii) E ⊗A is (2-quasi-) traceless. Thus we can rename

E ⊗ A by A for the following.
We express B as a tensor product B = C1 ⊗ C2 ⊗ · · · with Ck = Mnk such that
nk+1 > 2 · n1 · n2 · . . . · nk . The union of the increasing sequence of C *-subalgebras

Am := A ⊗ C1 ⊗ C2 ⊗ · · · ⊗ Cm ⊗ 1nm+1 ⊗ 1nm+2 ⊗ · · ·

of A ⊗ B is dense in A ⊗ B. To simplify notation we write simply Cm for

1M(A) ⊗ 1n1 ⊗ · · · 1nm−1 ⊗ Cm ⊗ 1nm+1 ⊗ · · · .

The algebras Am are matrix algebras over A and are q-trace-less because A is q-
trace-less. Thus A ⊗ B must be q-trace-less. (We can use here also exactness of B
and Part (iii) of Lemma 2.9.1.)
By Remark 2.8.3 we get that for each positive a ∈ Am and ε > 0 there exists
n(a, ε) ∈ N with (a − ε)+ ⊗ 12n - a ⊗ 1n for all n ≥ n(a, ε).
It implies 2n[(a − ε)+ ] ≤ n[a] for all n ≥ n(a, ε).
One needs here that the numbers n(a, ε) for given ε > 0 can be chosen in
the list of finite products of “remaining” numbers nk = Dim(Ck )1/2 starting from
k := m + 1.
???? Next not correct !!! ???
For each a ∈ Am there exists n ≥ m such that (n + 1)[a] ≤ n[a] in Cu(Am ).
Then for all k ≥ n and ` > 1 holds k[a] ≥ k`[a].
Indeed: We get by induction (n + r)[a] ≤ n[a] for all r ∈ N. Let r := (k · `) − n.
Then k`[a] ≤ n[a] ≤ k[a].
 9. PURE INFINITENESS VERSUS (2-QUASI-) TRACELESS. 271

In particular, it follows that for each ε > 0 there exist sufficiently big k :=
k(a, ε) ∈ N, δ > 0 and d1 , d2 ∈ Ak ⊂ A ⊗ B such that d∗i adj = δij (a − ε)+ .
This is the case because the relative commutant

(Am )0 ∩ M(A) ⊗ B ⊆ (Am )0 ∩ M(A ⊗ B)

of Am ⊆ A ⊗ B in M(A) ⊗ B contains sufficiently big unital matrix algebras Cm ,

and there is n ∈ N with a ⊗ 12k - a ⊗ 1k for all k ≥ n.
It follows that A⊗B is purely infinite by Lemma 2.5.17(ii), because the positive
S
elements in m∈N Am are dense in (A ⊗ B)+ and are spectral properly infinite in
A ⊗ B. 

M. Rørdam [690, prop. 2.2] gets the following observation using results in the
Jiang-Su paper [391]: There is a unital embedding of E(m, n) := E(Mm , Mn ) into
Z for every pair of natural numbers m < n that are relatively prime.
Recall that a C *-algebra B is called “(2-quasi)-traceless” if every l.s.c. 2-
quasitrace τ : B+ → [0, ∞] takes only the values 0 or +∞ . The lower semi-
continuity of the 2-quasitrace and the 2-sub-additivity are important ingredients
of this terminology, because the l.s.c. 2-sub-additive maps τ on A+ with values in
{0, ∞} and τ (a∗ a) = τ (aa∗ ) correspond one-to-one to the lattice of closed ideals
J /A. The bijection is given by τJ (a) := kπJ (a)k·∞ and Jτ := {a ∈ A ; τ (a∗ a) = 0}.

Proposition 2.9.3 ([463]). The ultra-power Aω of a C*-algebra A is (2-quasi )-

traceless, if and only if, there exists general n ∈ N such that a ⊗ 1n is properly
infinite in A ⊗ Mn for every non-zero a ∈ A+ , i.e., A has property pi-n in sense of
Definition ??.
In this case A is also pi(m) in sense of Definition 2.0.4, with m ≤ n (cf. Propo-
sition ?? NO other !).

??? Strange formulation:

Recall that A has property pi-n if and only if A is m-purely infinite for some
m≤n
and that pi-n implies pi(n),
but nothing is known over the converse ...
or m ≤ n2 ?
by Proposition ??, i.e., A is pi(m) in sense of Definitions 2.0.4 with m ≤ n.

Proof. Suppose that Aω has no non-zero 2-quasitrace. By Remark 2.8.3 we

get that for each positive b ∈ Aω and ε > 0 there exists n(b, ε) ∈ N with (b − ε)+ ⊗
12n - b ⊗ 1n for all n ≥ n(b, ε).
Give reference
Suppose that there does not exist an n ∈ N such that a ⊗ 1n is properly infinite
in A ⊗ Mn for every a ∈ A+ .
272 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Then there exists a sequence of positive contractions ak ∈ A+ and and nk ∈ N

such that nk < nk+1 , and ak ⊗ 1nk is not properly infinite in A ⊗ Mnk for each
k ∈ N.
If we use observation (x) of Lemma 2.5.3 then we find εk ∈ (0, 1) with εk+1 < εk
such that
(ak ⊗ 12nk − εk )+ 6- ak ⊗ 1nk .
1/mk
There are mk ∈ N such that (εk )1/mk > 1/2 for each k ∈ N. Let bk := ak ∈A
and b := πω (b1 , b2 , . . .) in Aω . It satisfies
(b − 1/4)+ ⊗ 12n 6- b ⊗ 1n .
This contradicts that Aω admits no l.s.c. 2-quasi-trace. 

It is not clear if the following Definition 2.9.4 is useful for general study:

Definition 2.9.4. Let C a unital C *-algebra and suppose that there exist
contractions e1 , e2 ∈ C+ and d1,1 , . . . , d1,m , d2,1 , . . . , d2,m ∈ C such that e1 e2 = 0
Pm
and k=1 d∗j,k ej dj,k = 1 for j = 1, 2 .
The smallest number m ∈ N with the property that such (ej , dj,k ) exist will be
denoted by m(C) ∈ N ∪ {+∞}. We call m(C) the “minimal orthogonality” of
C.
And we define a universal unital C *-algebra
Gm := C ∗ (ej , dj,1 , . . . , dj,m ; j = 1, 2, hRi )
with above considered relations (R) on the ej and dj,k .

Next Lemma 2.9.5 and its proof are wrong because we can take V := idMn ,
but there does not exist a unitary U ∈ Mn with U ∗ p11 U = (1/n)1n , because the of
norms (1, 1/n) and ranks (1, n) are different.

Lemma 2.9.5. Let V : Mn → A a completely positive map.

Then there exists a unitary U ∈ Mn such that V (U ∗ p11 U ) = 1/nV (1n ) ,
V (U ∗ p22 U ) = 1/(n − 1)V (1n − p11 ) for 1 ≤ k ≤ n.

Lemma 2.9.6. Let A a unital C*-algebra. Then A has no character, if and

only if, there exists m ∈ N and contractions ej,k ∈ A, j = 1, . . . , m, k ∈ {1, 2}, that
P ∗ ∗ ∗
satisfy the relations: j ej,k ej,k = 1 , ej,1 ej,2 = 0 = ej,2 ej,1 and

????? Look also to centalseq. paper with M.Rordam!// Perhaps

this can not be realized?:
e∗j,1 ej,1 = e∗j,2 ej,2 ????? because it is likely that
T −1/2 p1,1 T −1/2 6= T −1/2 p2,2 T −1/2
It is even not an estimate available !!!
( 43 ).
43 The relations do not imply – or require – that e∗ e
i,k j,k = 0 for i 6= j.
 9. PURE INFINITENESS VERSUS (2-QUASI-) TRACELESS. 273

Proof. To be filled in ! ??
Pn
Let S denote the set of all finite sums n−1 k=1 ak ∈ A+ of ak := ψk (f0 ⊗ 12 )
where ψk denote any non-zero *-homomorphism from C0 ((0, 1], M2 ) ∼ = M2 (C0 (0, 1])
into A. The set S ⊆ A+ is a convex subset of the contractions in A that is invariant
under automorphisms of A. In particular the hereditary convex sub-cone of A+ ,
i.e., the set of a ∈ A+ with the property that there exists b ∈ A+ and γ ∈ [0, ∞)
with a + b ∈ γ · S, i.e., is generated by S, has as its norm-closure a hereditary closed
convex cone that invariant under all automorphisms of A and is generated by all
2-homogenous elements of A+ .
By Lemma ??
Says/should say:
(1) the closures C of a hereditary sub-cone of A+ , defined by a subset X ⊂ A+
of contractions in A+ , is a hereditary (!!) sub-cone of A+ . It is the positive part
D+ = C of a hereditary C *-subalgebra D of A.
The D is invariant under conjugation by all unitaries of A (= inner automor-
phisms) if C it is invariant under conjugation by all exp(ih) with h∗ = h ∈ A and
khk < 1.
(2) If a hereditary C *-subalgebra D is invariant under conjugation by all
exp(ih) with h∗ = h ∈ A and khk < 1 then D is a closed ideal of A.
(3) For every element d ∈ D+ and ε > 0 there exists f in the convex span of
X and λ ∈ (0, ∞) with (d − ε)+ ≤ λx. ???? (at least if X is invariant under inner
automorphisms of A.)
It is the positive part of the closed ideal of A generated by all images ψ(f0 ⊗12 ).
The quotient A/J can not contain a non-zero 2-homogenous element, because
C0 ((0, 1], Mk ) is a projective C *-algebra for all k ∈ N, cf. Proposition A.8.4 or [540,
cor. 3.8].
Thus, J is also the closed ideal of A generated by all commutators ab − ba, i.e.,
is the intersection of the kernels of all characters on A.
By assumption, A has no character. It shows that A = J and 1 ∈ J.
Pr
By the above study of J+ there exists r ∈ N and a finite sum q=1 aq ∈ A+ of
2-homogenous positive elements aq := ψq (f0 ⊗ 12 ) = ψq (f0 ⊗ p11 ) + ψq (f0 ⊗ p22 )
such that
Pr Pr
1A ≤ q=1 aq =: T . Then q=1 T −1/2 aq T −1/2 = 1 .
But this T is imbalanced with respect to the entries ψq (f0 ⊗ pj,k ) if aq =
ψq (f0 ⊗ p1,1 ).
1/2 1/2 1/2
ψq (f0 ⊗ p11 ) + ψq (f0 ⊗ p12 ) = ψq (f0 , p11 + p12 )
√
Let x := (1/ 2)(p11 + p12 ) gives x∗ 6= x x∗ x = 1/2(p11 + p12 + p2,1 + p2,2 ), and
√
for y := (1/ 2)(p22 − p21 ) gives y ∗ y = 1/2(p11 − p12 − p2,1 + p2,2 ) and y ∗ x = 0 .
274 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Can we find some estimate for x∗ x ? x∗ x ≤ p11 + p22 because x∗ x + y ∗ y =

p11 + p22
The elements T −1/2 ψq (f0 ⊗12 )T −1/2 ∈ A with properties ψq : C0 ((0, 1], M2 ) →
A, 1 ≤ q ≤ r imply the statement ... ???
1/2
(Likely ???: m := 2r, ???? ej,1 := ψj (f0 ⊗ p11 )T −1/2 for 1 ≤ j ≤ n and
en+j,1 := ψj−n (f0 ⊗ p22 )T −1/2 for n + 1 ≤ j ≤ 2r ????)
But, what can we show ??? It is:
If unital A has no non-zero character, then there exist m contractions
z1 , . . . , zm ∈ A with the properties zk2 = 0 and
m
X
(zk (zk )∗ + zk∗ zk ) ≥ 1 .
k=1

In case of only σ-unital A, with strictly positive element e ∈ A+ that satisfies

kek = 1 and has no non-zero character, this should have similar properties:
Likewise this:
For every ε > 0 there exists δ ∈ (0, ε), and m ∈ N such that there are exist m
contractions z1 , . . . , zm ∈ (e − δ)+ A(e − δ)+ with the properties zk2 = 0 , and
m
X
(zk (zk )∗ + zk∗ zk ) ≥ (e − ε)+ .
k=1

There exists m ∈ N and contractions ej,k ∈ A, j = 1, . . . , m, k ∈ {1, 2}, that

P ∗ ∗ ∗ ∗
satisfy the relations: j ej,k ej,k ≥ 1 , ej,k ej,` = 0 and ej,k ej,k = ej,` ej,` for k 6= `,
(The relations do not imply or require that e∗i,k ej,k = 0 for i 6= j). 

Remarks: The Lemma 2.9.6 implies immediately the following ??????

If B is a σ-unital C *-algebra, then the unital C *-algebra

F (B) := (B 0 ∩ Bω )/ Ann(B, Bω )

has no character, if and only if there exist dj,k ∈ B 0 ∩ Bω j = 1, . . . , m, k ∈ {1, 2},

d∗j,k dj,` = 0 and d∗j,k dj,k = d∗j,` dj,` for k 6= ` and j d∗j,k dj,k ∈ 1 + Ann(B, Bω ).
P

Notice that for E(M2 , M3 ) ⊆ Z ⊆ O∞ we get m(E(M2 , M3 )) ≤ 3.

Or m(...) ≤ 2 ???
One can see that the infinite unital free product C ∗1 C ∗1 C ∗1 · · · of the unital
cones C := {f ∈ C([0, 1], M2 ) ; f (0) = 12 } (with unit element 1C (t) := 12 for all
t ∈ [0, 1]) has only one character χ given by f 7→ f (0) ∈ C · 12 .
It has to be considered in a natural way as the C *-algebra inductive limit by
the injective maps

C 7→ C ∗1 1 ⊂ C ∗ C 7→ C ∗1 C ∗1 1 ⊂ C ∗1 C ∗1 C...
 9. PURE INFINITENESS VERSUS (2-QUASI-) TRACELESS. 275

The unital character χ then extends step by step to a character of

C ∗1 C ∗1 C ∗ · · · .

This should be the only character on it ???

Let D denote the kernel ideal of the character χ. Then D ∗1 1C and 1C ∗1 D
together generate C ∗1 C .....????

Proposition 2.9.7. Let A a purely infinite C*-algebra. If there exists m ∈ N

such that for every finite subset F ⊆ A and δ > 0 there exists a unital C*-morphism
φ : Gm → M(A) with k[x, φ(ej )]k < δ and k[x, φ(dj,k )]k < δ for x ∈ F , j = 1, 2
and k = 1, . . . , m. Then A is strongly p.i.

Proof. In the general case, e.g. where A is only pi-n, we get at least that all
matrices with n-homogenous diagonal entries are approximately diagonalizable by
n-homogenous entries in the diagonal.
The assumptions imply that, for every separable C *-subalgebra B of A, there
exist a unital C *-morphism ψ : Gm → M(A)ω with zero commutators [b, ψ(ej )] = 0
and [b, ψ(dj,k )] = 0 for all b ∈ B.
Let a, b ∈ A+ properly infinite elements and consider the C *-subalgebra
∗
C (a, b) of A generated by a and b . Find in A a separable C *-subalgebra B with
the following properties:
(1) a, b ∈ B+ and a, b are inside B properly infinite.
(2) B contains sequences that represent the generating elements of ψ(Gm ) such
that ψ(Gm ) commutes with B and is in Bω .
Now we consider the 2×2-matrix C := [cp,q ] = [a, b]> [a, b] with entries c11 = a2 ,
c22 = b2 , c1,2 = ab and c2,1 = ba. If a and b are properly infinite in A, then they
are also properly infinite in Aω .
This happens also for elements be1 and ae2 in Aω if be1 = e1 b and ae2 = e2 a if
?????.
?????? IS THE LATTER TRUE ??

More generally let dj,k ∈ B 0 ∩ Bω j = 1, . . . , m, k ∈ {1, 2}, d∗j,k dj,` = 0 and

d∗j,k dj,k = d∗j,` dj,` for k 6= ` and j d∗j,k dj,k ∈ 1 + Ann(B, Bω ).
P

Define properly infinite a, b ∈ A+ by elements f1 , . . . , fm , g1 , . . . gm ∈ A with

fj∗ a2 fk= δj,k (a2 − ε)+ and gj∗ b2 gk = δj,k (b2 − ε)+ .
Pm Pm
Then define h1 := j=1 fj dj,1 ∈ Bω and h2 := j=1 gj dj,2 ∈ Bω
????
Notice that h∗1 a2 h1 = (a2 − ε)+ and h∗2 b2 h2 = (b2 − ε)+ , but to get h∗1 abh2 = 0,
one needs in addition that d∗j,k di,` = 0 for all j 6= i and all k, ` ∈ {1, 2}.
To be filled in ?? 
276 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Corollary 2.9.8. Let Z denote the Jiang-Su algebra. If A is a ( 2-quasi-)

traceless C*-algebra, then A ⊗ Z is strongly purely infinite.

Proof. By Proposition n?? A ⊗ Z is purely infinite.

By [391], the algebra Z contains E(M2 , M3 ) and Z ∼ = Z ⊗ Z ⊗ · · · . Since
m(E(M2 , M3 )) ≤ 3 it follows that there is a unital C *-morphism from G3 into Z.
The isomorphism Z ∼ = Z ⊗Z ⊗· · · implies that there exists a “central” sequence
of unital C *-morphisms from G3 into M(A) ⊗ Z ⊂ M(A ⊗ Z). Now Proposition
2.9.7 applies and shows that A ⊗ Z is strongly purely infinite.
Alternatively one can Z represent as inductive limit of E(M2∞ , M3∞ ) given by a
suitable endomorphism of E(M2∞ , M3∞ ). The latter can be described as inductive
limit of E(M2n , M3n ). But E(M2n , M3n ) contains obviously a central sequence of
copies of E(M2 , M3 ).


Definition 2.9.9. Let A, B unital C *-algebras. The tensorial joint algebra

(also called joining algebra , or winding around algebra) of A and B is defined
in [448, following Prop. 1.17] as the C *-subalgebra of C([0, 1], A ⊗max B) defined
by:
E(A, B) := { f ∈ C([0, 1], A ⊗max B) ; f (0) ∈ A ⊗ 1 , f (1) ∈ 1 ⊗ B } (9.1)
In cases where A := Mm and B := Mn , this algebras also are called dimension-drop
C *-algebras.

Corollary 2.9.10. If A is a (2-quasi-) traceless C*-algebra, and if B := Mp

and C := Mq are UHF-algebras with infinite supernatural numbers p and q, then
the tensor product
A ⊗ E(B, C) ⊆ A ⊗ C([0, 1], B ⊗ C)
of A with the “tensorial joint” algebra E(B, C) of Definition 2.9.9 is strongly purely
infinite.

Proof. Consider unital C *-algebras B and C. Let H1 denote the sub-algebra

of C[0, 1/2] ⊗ C defined by the maps f ∈ C([0, 1/2], C) with f (0) = 1, and H2
the sub-algebra of maps g ∈ C[1/2, 1] ⊗ B with g(1) = 1 . Then there are natural
C *-epimorphisms φ1 from B ⊗ H1 onto B ⊗ C with φ1 (b ⊗ f ) = b ⊗ f (1/2) and
φ2 : H2 ⊗ C → B ⊗ C with φ2 (g ⊗ c) = g(1/2) ⊗ c.
It is easy to see that E(B, C) ⊆ C([0, 1], B ⊗ C) is naturally isomorphic to the
pull-back of φ1 and φ2 .
It follows that A ⊗ E(B, C) is the pull-back of the epimorphisms idA ⊗φ1 and
idA ⊗φ2 onto A ⊗ B ⊗ C. In particular, it is an extension of A ⊗ B ⊗ C by the ideal
(A ⊗ φ−1 −1
1 (0)) ⊕ (A ⊗ φ2 (0)) of A ⊗ E(B, C).

Notice that E(C, B) is the C *-subalgebra of C([0, 1], B) of continuos functions

f : [0, 1] → B wit f (0) = 1B . Notice that φ−1 ∼ −1 ∼
1 (0) = B ⊗ E(C, C) and φ2 (0) =
E(C, B) ⊗ C.
 9. PURE INFINITENESS VERSUS (2-QUASI-) TRACELESS. 277

For this ideals and for the A ⊗ B ⊗ C the Lemma ?? applies if B := M2∞
and C := M3∞ , because all tensorial absorb UHF algebras. Thus they are purely
infinite infinite if A is (2-quasi-) traceless.
By Proposition 2.9.12 they are moreover strongly purely infinite because they
have central sequences of matrix algebras.
Thus, A ⊗ E(B, C) is an extension of strongly p.i. C *-algebras if A is (2-quasi-)
traceless. By Proposition ?? it follows that A ⊗ E(B, C) is p.i., because the class
of purely infinite C *-algebras is invariant under extensions.
next Ext.part of Thm. is not present here. Only cited.
Since extensions of s.p.i. algebras are s.p.i. by Theorem 2.9.13, we get moreover
that A ⊗ E(B, C) is strongly p.i.


Corollary 2.9.11. Let A a (2-quasi-) traceless C*-algebra and Z the Jiang-Su

algebra, then A ⊗ Z is strongly purely infinite.

Proof. If B is a unital C *-algebra and ψ : B → B a unital C *-morphism.

The map ψ is called “trace-collapsing” if τ1 ◦ ψ = τ2 ◦ ψ for any trace state on B,
i.e., ψ ∗ (T (A)) consists of a fix-point of T (A).
By [693, thm. 3.4], Z is the inductive limit


indlimn φn : E(M2∞ , M3∞ ) → E(M2∞ , M3∞ ) ,

where φn := φ is a suitable fixed “trace-collapsing” unital endomorphism φ

of E(M2∞ , M3∞ ). “Trace-collapsing” means here that for each trace state
ρ on E(M2∞ , M3∞ ) the state ρ ◦ φ is identical with the state induced on
R1
E(M2∞ , M3∞ ) ⊆ C([0, 1], M2∞ ⊗ M3∞ ) by µ(f ) = 0 τ (f (t)) dt, where τ is
the unique trace state on M2∞ ⊗ M3∞ .
We have seen above (in a more general case) that A ⊗ E(M2∞ , M3∞ ) is s.p.i.
because there exists a “central” sequence of unital C *-morphisms ψk : E(Mp , Mq ) →
E(M2∞ , M3∞ ), where p := 2k and q := 3` for suitable k, ` ∈ N.
?????????
The classes of the purely infinite C *-algebras and of the strongly purely infinite
C *-algebras are both invariant under inductive limits.
By Theorem ??(????) the class of s.p.i. algebras is invariant under inductive
limits. 

Proposition 2.9.12. If A is (2-quasi-) traceless and B is a UHF algebra (of

infinite dimension), then A ⊗ B is s.p.i.

Proof. We can split B into a tensor-product B = B1 ⊗ B2 where B1 and

B2 are UHF-algebras of infinite dimension. Then A ⊗ B1 is purely infinite by
Lemma ??. Thus we can suppose that A is purely infinite. Let a, b ∈ (A ⊗ B)+
278 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

contractions and ε > 0 There exist contractions c, d ∈ (A ⊗ B)+ with ka − ck < ε/2
and kb − dk < ε/2
???? ???? 

Theorem 2.9.13. The class of s.p.i. C*-algebras is permanent under exten-

sions.
(But only citation. The known proof is complicate.)
Other operations.?

Proof. ??? 

Corollary 2.9.14. If A is (2-quasi )-trace-less then A ⊗ Z is strongly p.i.

Proof. Use that Z has central sequences of copies of E(M2 , M3 ). 

Below we give later some results on tensor products that are farer going than
those in Corollaries 2.9.2 and 2.9.10, e.g. we show that the tensor products A ⊗ B
and A ⊗ E(M2∞ , M3∞ ) considered in this corollaries are moreover strongly p.i.

Proof. to be filled in ?? 

Synchronize below given Def’s and statements with further above!!

Some from Paper with Blanchard.
Give extra Definitions (!) for all sorts of pure infiniteness!!!

Remarks 2.9.15.
(i) If a ∈ A+ , b ∈ Mn (A)+ and suppose that there is a matrix e ∈ Mm,n (A)
with kb − e∗ (a ⊗ 1m )ek < ε for some ε > 0, then
(b − ε)+ = f ∗ ((a − 2η)+ ⊗ 1m )f
for some matrix f ∈ Mm,n (A) with kf k ≤ kek and some η ∈ (0, ε/2) .
The def’s (ii) of p.i. etc. are given earlier in Section 1,
later again !!! where???

(ii) An element a ∈ A+ \ {0} in a (not necessarily purely infinite) C *-algebra

A is properly infinite if, for every ε > 0, there exists a row d = [d1 , d2 ] ∈ M1,2 (A)
(depending on a and ε) such that kd∗ ad − a ⊗ 12 k < ε, cf. [462, def. 3.2].
This property of a is equivalent to [a] + [a] ≤ [a] in Cu(A), i.e., a ⊕ a - a, or
to a ∈ I(a) by definition of I(a) in Definition 2.5.1, cf. also Lemma 2.5.3(i).
An element a ∈ A+ is properly infinite if for every closed ideal J of A – that
does not contain a – there is an element h 6= 0 in (A/J)+ such that for every δ > 0
there exists a row matrix d = [d1 , d2 ] ∈ M1,2 (A/J) with kd∗ πJ (a)d − (πJ (a) ⊕ h)k <
δ, cf. Part (v) of Lemma 2.5.3 or [462, prop. 3.14].
The latter says that [h] + [πJ (a)] ≤ [πJ (a)] in Cu(A/J), i.e., I(πJ (a)) 6= 0 for
all J ∈ I(A) with a 6∈ J, where I(b) for b ∈ πJ (A) is defined by Definition 2.5.1.
 9. PURE INFINITENESS VERSUS (2-QUASI-) TRACELESS. 279

This definition of properly infinite infinite (positive) elements coincides with

Definition 2.5.1 of a properly element a ∈ A by a ⊕ a - a.

(iii) A C *-algebra A is purely infinite in the sense of Definition 1.2.1, if and

only if, every element a ∈ A+ \ {0} is properly infinite, [462, thm. 4.16].
there is also Ref. in ???? to place here!!

(iv) All purely infinite C *-algebras A have the global Glimm halving property
of Definition 2.15.9.

Proof. (i): We find η ∈ (0, ε/2) such that we have kb−e∗ ((a−η)+ ⊗1m )ek < ε .
By Lemma 2.1.9 there is a contraction d ∈ Mn (A) such that f := ed is as desired.
(ii?): ??
The element πI(a) (a) = a + I(a) is always finite in A/I(a) by Part (iii) of
Lemma 2.5.3. Thus above property of a ∈ A+ implies that a ∈ I(a), i.e., a ⊕ a - a.
(ii?): ?? If one applies (i) with m = 1, n = 2 then one finds u, v ∈ aAa with
u∗ u = v ∗ v = (a − ε)+ and u∗ v = 0 ([462, prop. 3.3(v)]), i.e., there exists a row
w = [u, v] ∈ M1,2 (aAa) satisfying
w∗ w = (a − ε)+ ⊗ 12 in A ⊗ M2 ,
cf. also Lemma 2.5.3(ix,xi)
(iv): ??
For each a ∈ A+ \ {0} and ε > 0, then b = vu∗ ∈ aAa with u, v from (ii??)

2
verifies b2 = 0 and (a − ε)+ = v ∗ bu , so that (a − ε)+ ∈ AbA . 

How often is fε defined and mentioned??

In the following, recall that fε (t) := min(1, max(0, 2t/ε − 1)) for t ≥ 0 and
ε > 0. Those special pice-wise linear functions have been used fist in Glimm’s
paper [324] for his pioneering study of C *-algebras of Type I.
Notice that the Part (i) of the next Lemma 2.9.16 applies in particular to all
non-zero elements a ∈ A+ that are spectral properly infinite in sense of Definition
2.5.16.
Later appears a similar remark!!!

Lemma 2.9.16. Let a ∈ A+ with kak = 1, 0 < δ < ε < 1 , and let B := aAa .
Suppose that for every ν ∈ (δ, ε) , the element (a − ν)+ is properly infinite in
the sense of Definition 2.5.1.
1/3
(i) There exists an infinite sequence w1 , w2 , . . . ∈ (a − δ)+ A(a − ε)+ such
that
wn∗ wm = δn,m (a − ε)+ for all m, n ∈ N .
The element d := n∈N 2−n wn wn∗ ∈ B generates a stable hereditary C*-
P

subalgebra Eε := dAd of (a − δ)+ A(a − δ)+ that is naturally isomorphic

to D ⊗ K for D := (a − ε)+ A(a − ε)+ .
280 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

In particular the algebra Eε is a full hereditary C*-subalgebra of the

closed ideal of A that is generated by (a − ε)+ .
(ii) If, moreover, every non-zero element of B is properly infinite, then there
are elements wn ∈ B such that wn∗ wm = δn,m · (a − 2−n )+ for all n ≥
m ∈ N.
−n
wn wn∗ ∈ B := a A a where the elements w1 , w2 , . . . ∈
P
(iii) Let d := n∈N 2
B are the elements from Part (ii).
The hereditary C*-subalgebra D := d A d is contained in B, and there
exist elements z1 , z2 , . . . ∈ B with zj∗ zk = δj,k a and zk zj∗ ∈ D.
P −n
In particular, the contraction e := n 2 zn zn∗ generates a heredi-
tary C*-subalgebra E := e D e = e A e of D that is full in B.
(iv) The hereditary C*-subalgebra B := a A a is itself stable if each element of
B is properly infinite and the annihilators Ann((a−1/n)+ , B) of (a−1/n)+
are full in B for each n ∈ N.
(v) The annihilators Ann((a − 1/n)+ , B) are full in B for each n ∈ N if B
has no unital quotient.
(vi) The hereditary C*-subalgebra B := a A a is stable if B has no unital
quotient and a is “spectral” properly infinite, i.e., ψ(a) is properly infinite
or zero for every function ψ ∈ C0 (0, kak]+ .

Proof. Recall that an element a itself is properly infinite if (a − ν)+ is prop-

erly infinite for every ν ∈ (0, δ) for some δ > 0, and that the elements b∗ b, bb∗
and (b∗ b)1/2 are all properly infinite if one of its is properly infinite, cf. Lemma
2.5.3(xi,xii).
by Remark 2.9.15(ii) ?
Find precise Ref.s for next.
(i): We can take δ < δ 0 < ε0 < ε, replace a by (a−δ 0 )+ and then show moreover
that there are elements c1 , c2 , . . . in the closure of (a − δ 0 )+ A(a − ε0 )+ such that
c∗n cm = δn,m (a−ε0 )+ for m, n ∈ N . We get the desired wn as wn := cn ·ϕ(c∗n cn )1/2
for the function ϕ(t) := (t − ε)/(t − ε0 ) for t ≥ ε and ϕ(t) = 0 for t < ε .
Hence, we can rename (a − δ 0 )+ by a and ε0 − δ 0 by ε and show only the
existence of wk ∈ aA(a − ε)+ with wk∗ wj = δj,k (a − ε)+ .
For n ∈ N , let εn := 2−n−1 ε and µn := 0≤k≤n εk = (1 − 2−n−1 ) · ε < ε .
P

Recall that, if (a − δ)+ is properly infinite, then, for each µ > δ, there exist
τ ∈ (δ, µ) and u, v ∈ A with

u∗ v = 0 , u∗ u = v ∗ v = (a − µ)+ and uu∗ + vv ∗ ∈ (a − τ )+ A(a − τ )+ . (9.2)

Compare Remark 2.9.15(ii) or Lemma 2.5.3(ix,xi). or other reference ?

If we let v−1 := a1/2 , then we can find for each n ∈ N inductively – by repeated
use of Equation (9.2) – elements un , vn ∈ aAa such that
∗
(1) un , vn in vn−1 Avn−1 ⊆ aAa (n ∈ N)
 9. PURE INFINITENESS VERSUS (2-QUASI-) TRACELESS. 281

(2) for every ν ∈ (0, ε−νn ) , the element (vn∗ vn −ν)+ is either zero or properly
infinite,
(3) u∗n un = vn∗ vn = (vn−1
∗
vn−1 − εn )+ = (a − µn )+ ≥ (a − ε)+ and
∗
(4) un vn = 0 ,
∗
Induction over m shows that condition (1) implies that vn−1 Avn−1 ⊆
∗ ∗
vm+1 Avm+1 ⊆ vm Avm for m = 1, . . . , n − 2.
Thus, u∗m vn−1 Avn−1
∗
= 0 and u∗m un = 0 for m < n by (1) and (4). Hence,
u∗m un = δm,n (a − µn )+ for m, n ∈ N by (3).
Notice that (1) and (3) imply un , vn ∈ vn−1 A(a − µn−1 )+ ⊆ aAa.
For n ∈ N , let ϕn : R+ → [0, 1] be the continuous function with ϕn (t) := 0 for
t < ε and ϕn (t) := (t − ε)/(t − µn ) for t ≥ ε.
The elements wn = un fn (a)1/2 ∈ aA(a − ε)+ satisfy the requested relations,
because (t − µn )+ ϕn (t) = (t − ε)+ .
Let B := (a − ε)+ A(a − ε)+ . The relations wk∗ w` = δk,` (a − ε)+ imply that
the polar decomposition of the wk define a natural isomorphism B ∼ = w1 Aw1∗ and
that the C *-subalgebra D := C ∗ (wm bwn∗ ; n, m ∈ N, b ∈ A) of aAa is naturally
isomorphic to B ⊗ K. In particular D is stable.
The algebra D contains d := n∈N 2−n wn wn∗ and is identical with the heredi-
P

tary C *-subalgebra dAd ⊆ B, with d as strictly positive element.

The hereditary subalgebra D generates the same closed ideal of A as the element
(a − ε)+ because 2w1∗ dw1 = (w1∗ w1 )2 = (a − ε)2+ .

(ii): Let a ∈ A+ with kak = 1, and suppose that every non-zero element of
B := aAa is properly infinite and let D ⊆ B (any) full hereditary C *-subalgebra
of B.

(α) Suppose that D ⊆ B := aAa (with a ∈ A+ , kak = 1) is a full hereditary

C *-subalgebra of B and that each non-zero e ∈ D+ is properly infinite.
Then for each γ ∈ (0, 1) there exists elements u, v ∈ B with u∗ u =
v ∗ v = (a − γ/2)+ , v ∗ u = 0 and vv ∗ , uu∗ ∈ D.
(β) If D is any full hereditary C *-subalgebra of B := aAa , where a ∈ A+
with kak = 1, and if
check here if a weaker property also works!
every element of D+ is properly infinite, then for each γ ∈ (0, 1) there
exists an element w ∈ B that satisfies w∗ w = (a − γ)+ , ww∗ ∈ D, and
that the annihilator Ann(ww∗ , D) of ww∗ in D is again a full hereditary
C *-subalgebra of B.

We postpone the proof of (α) and (β) and show that (β) allows to construct
the in Part (ii) requested elements w1 , w2 , . . . ∈ B := aAa with the relations
wn∗ wm = δn,m · (a − 2n )+ for n ≥ m.
We proceed by induction using (β):
282 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

We find by (β) an element w1 ∈ D0 := B with w1∗ w1 = (a − 1/2)+ such that

D1 := Ann(w1 w1∗ , B) is a full hereditary C *-subalgebra of B.
Suppose we have selected w1 , . . . , wn ∈ B that satisfy

wk∗ w` = δk,` · (a − 2−` )+

for k ≤ ` ≤ n, and that

Dn := Ann(w1 w1∗ + · · · + wn wn∗ , B) = Ann(wn wn∗ , Dn−1 )

is full in B.
Recall here that Ann(e, D) = Ann(e, B) ∩ D for hereditary D ⊆ B and e ∈ D+ .
Thus, we can apply (β) to D := Dn and γ := 1/2n+1 and get wn+1 ∈ Dn with
∗
wn+1 wn+1 = (a − 2−(n+1) )+ and

Dn+1 := Ann(wn+1 wn+1 , Dn ) = Ann(w1 w1∗ + · · · + wn+1 wn+1

∗
, B)

is full in B. It implies that

∗
wn+1 wk = δn+1,k (a − 2−(n+1) )+ for k = 1, . . . , n + 1 .

Proof of observation (α): Let γ ∈ (0, 1). Since D is full in B and a ∈ B+ there
exists d1 , · · · , dn ∈ B with dk d∗k ∈ D for k = 1, . . . n and

d∗1 d1 + · · · + d∗n dn = (a − γ/4)+ .

It follows n[e] ≥ [(a − γ/4)+ ] in Cu(B), for e := d1 d∗1 + · · · + dn d∗ ∈ D.

Since, by assumptions, e ∈ D+ is a properly infinite element of B, we get
[e] ≥ 2n[e] ≥ 2[(a − γ/4)+ ], i.e.,

(a − γ/4)+ ⊕ (a − γ/4)+ - e ⊕ 0 .

In particular there exists f1 , f2 ∈ B and µ > 0 with

[f1∗ , f2∗ ](e − µ)+ [f1 , f2 ]> = diag(a − γ/2)+ ⊕ (a − γ/2)+ .

1/2 1/2
The elements u := (e − µ)+ f1 and v := (e − µ)+ f2 satisfy uu∗ , vv ∗ ∈ D, u∗ v = 0
and u∗ u = v ∗ v = (a − γ/2)+ .
Proof of observation (β): Each hereditary C *-subalgebra D of B := aAa that
is full in B and is purely infinite fulfills the assumptions for Part (α).
Let γ ∈ (0, 1) and let u, v ∈ B the elements from Part (α), i.e., u∗ u = v ∗ v =
(a − γ/2)+ , u∗ v = 0 and uu∗ , vv ∗ ∈ D.
The function ψγ defined by ψγ (t) := (t − γ)/(t − γ/2) for t ≥ γ and ψγ (t) := 0
for t ≤ γ is in C0 (0, 1]+ and satisfies ψγ (t) ≤ 1. Clearly

ψγ (a)(a − γ/2)+ = (a − γ)+ = ((a − γ/2)+ − γ/2)+ .

Define w := vψγ (a)1/2 ∈ B. It satisfies w∗ w = (a − γ)+ = (v ∗ v − γ/2)+

and ww∗ = (vv ∗ − γ/2)+ . The latter equation can be seen with help of the polar
decomposition v = zv (v ∗ v)1/2 = zv (a − γ/2)+ using that f (vv ∗ )zv = zv f (v ∗ v) for
f ∈ C0 (0, kvk2 ].
 9. PURE INFINITENESS VERSUS (2-QUASI-) TRACELESS. 283

The elements ww∗ , vv ∗ and d := uv ∗ are in D and satisfy ww∗ = (vv ∗ − γ/2)+ ,
d∗ (vv ∗ − γ/2)+ = vu∗ ww∗ = 0 and [d] = [vv ∗ ] in Cu(D). The latter because
d∗ d = vu∗ uv ∗ = v(a − γ/2)+ v ∗ = (vv ∗ )2 . Thus, the hereditary C *-subalgebra
Ann(ww∗ , D) is full in D by the last observation in Part (ii) of Lemma 2.5.14.
Since D is full in B it implies that Ann(ww∗ , D) is a hereditary C *-subalgebra of
B that is full in B.
(iii): TEXT:
−n
wn wn∗ ∈ B := a A a where the elements w1 , w2 , . . . ∈ B are
P
Let d := n∈N 2
the elements found in Part (ii).
The hereditary C *-subalgebra D := d A d is contained in B, and there exist
elements z1 , z2 , . . . ∈ B with zj∗ zk = δj,k a and zk zj∗ ∈ D.
P −n
In particular, the contraction e := n 2 zn zn∗ ∈ B generates a hereditary
C *-subalgebra E := e D e = e A e of D that is full in B.
We define for t ≥ 0 and m, n ∈ N, m < n, the increasing function ϕ(m, n; ·) ∈
C0 (0, 1] with ϕ(m, n; t) := 0 on [0, 2−m ] and ϕ(m, n; t) := max(t − 2−m , 0)/ max(t −
2−n , 0) for t ≥ 2−m .
Let λ : N → N×N a bijective map from N onto N → N and let Mk := λ−1 ({k}×
N).
Notation Mk is not good,
NEXT property HERE necessary?
We can re-index the infinite subsets Mk of N such that they have the property
that min(Mk ) < min(Mk+1 ) for k ∈ N.
The map λ defines a unique increasing injective map µk : N → N with µk (N) =
Mk .
We let a[k, 1] := 1 ∈ C ∗ (a, 1) ⊆ M(B),

a[k, n + 1] := 1 − ϕ(µk (n), µk (n + 1); a) ∈ C ∗ (a, 1)+ ,

and define elements vk ∈ B for k ∈ N by

X
vk := wµk (n) · a[k, n]1/2 .
n∈N

The sum converges in B, because wµ∗ k (n) wµk (m) = δm,n (a − 2−n )+ and, for n > 1,

a[k, n]1/2 · wµ∗ k (n) wµk (n) · a[k, n]1/2 = (a − 2−µk (n) )+ − (a − 2−µk (n−1) )+ .

It has norm ≤ 2−µk (n−1) − 2−µk (n) < 2−µk (n−1) . The series n>1 (2−µk (n−1) −
P

2−µk (n) ) is absolutely convergent with sum = 2−µk (1) .

One gets from w`∗ wm = 0 for ` 6= m that

vk [n]∗ vk [n] = (a − 2−µk (n) )+

for the partial sums

n
X
vk [n] := wµk (m) · a[k, m]1/2 .
m=1
284 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

It follows that
X
vk∗ vk = (a − 2−µk (n) )+ a[k, n] = a .
n
∗ ∗
Since wm wn = 0 and wn wm ∈ D for m 6= n and µk (N) ∩ µ` (N) = ∅ for k 6= `, it
follows that vk v` = δk,` a. and v` vk∗ ∈ D.
∗

(iv):
TEXT:
The hereditary C *-subalgebra B := a A a is itself stable if each element of B
is properly infinite and the annihilators Ann((a − 1/n)+ , B) of (a − 1/n)+ are full
in B for each n ∈ N.
to be filled in ??
We show that it implies that there exists a zero sequence kak > γ1 > γ2 >
· · · > 0 and elements un ∈ B with

(1) u∗n un = (a − γn )+ ,
(2) u∗n+1 (a − γn )+ = 0, and
1/n
(3) un = limn→∞ un (a − γn+1 )+ .

What can be done with it ?

Since a is a strictly positive element of B it follows that the criteria for stability
of B is satisfied for B. Alone from (1) and (2)?
Indeed we can define for the exhausting family of order preserving injective
maps ???k : N → Mk := λ−1 ({k} × N) from a partition λ : N → N × N
(v): Let J denote the closed ideal of B generated by Ann((a − 1/n)+ , B). By
Lemma 2.5.14, B/J is unital and πJ (a) is invertible in B/J if J 6= B. Thus, J = B
if B has no unital quotients. 

Remark 2.9.17. Let a ∈ A+ with kak = 1. The hereditary C *-subalgebra

aAa contains a full stable hereditary C *-subalgebra D, if and only if, there exists
a sequence of elements bn ∈ aAa with b∗m bn = δm,n (a − 2−n )+ for m, n ∈ N.
Here “full” – equivalently – means that a is in the closed ideal generated by D,
i.e., that a ∈ A · D · A = span(ADA) by stability of D.
∗ ∗
It implies that there are g1 , g2 , . . . ∈ A with gm gn = δm,n a and gm gm ∈ D.

It is not clear when C ∗ (bn b∗m ; m, n ∈ N) itself is stable.

Proof. The existence of the quoted sequence d1 , d2 , . . . is necessary:

If D is a stable and full hereditary C *-subalgebra of B := aAa then there exists
d ∈ B with d∗ d = a and dd∗ ∈ D, cf. Lemma ??(ii).
Since D is stable, its multiplier algebra M(D) contains a unital copy of L(`2 ) ⊃
O∞ = C ∗ (s1 , s2 , . . . ; s∗m sn = δm,n 1), cf. Remark 5.1.1(8). We can take bn :=
sn dgn (a), where gn (0) := 0 and gn (t) := ((t − 2−n )+ /t)1/2 with (t − µ)+ :=
max(0, t − µ) for t > 0.
 9. PURE INFINITENESS VERSUS (2-QUASI-) TRACELESS. 285

Conversely, the existence of elements bn ∈ aAa with b∗m bn = δm,n (a − 2−n )

(for m ≤ n) is sufficient for the existence of a full stable hereditary C *-subalgebra
D ⊆ aAa that contains a in the ideal ADA of A, because by the below given
calculations C ∗ (bm b∗n ; m, n ∈ N ) is isomorphic to the hereditary C *-subalgebra
E := d(C ∗ (a) ⊗ K)d of C ∗ (a) ⊗ K where

d := diag((a − 1/2)+ , (a − 1/4)+ , . . .)1/2 ∈ `∞ (C ∗ (a)) ⊂ M(C ∗ (a) ⊗ K) ,

i.e., e := diag(1/2(a − 1/2)+ , . . . , 2−n (a − 2−n )+ , . . .) is a strictly positive element

of C ∗ (a) ⊗ K.
An isomorphism from E onto C ∗ (bm b∗n ; m, n ∈ N ) can be defined as follows:
First define λ : C ∗ (a) ⊗ K → B ⊗ K as the natural embedding. The λ
is non-degenerate because B = a B a . Therefore the strictly continuous
M(λ) : M(C ∗ (a) ⊗ K) → M(B ⊗ K) exists and is injective natural inclu-
sion of M(C ∗ (a) ⊗ K) into M(B ⊗ K). Let g := M(λ)(d) and h := λ(e) =
1/2(a − 1/2)+ ⊗ p1,1 + 1/4(a − 1/4)+ ⊗ p2,2 + · · · . Then λ(E) is a non-degenerate
C *-subalgebra of the hereditary C *-subalgebra g(B ⊗ K)g = h(B ⊗ K)h =: F . It
is easy to see that λ(E) := F ∩λ(C ∗ (a)⊗K), because it is generated by the elements
1/2
(a − 2−m )1/2 (a − 2−n )+ ⊗ pm,n , and, even by definition of h, h = λ(e) ∈ F+ is a
strictly positive contraction of F and λ(E).
P −n ∗
The element P := n2 bn bn ∈ B is a strictly positive contraction of
∗ ∗
C (bm bn ; n, m ∈ N).
Let T := n 2−n/2 bn ⊗ p1,n ∈ B ⊗ K. Then T T ∗ = P ⊗ p1,1 and
P
X
T ∗T = 2−n (a − 2−n )+ ⊗ pn,n = h .
n

The polar decomposition T = U (T ∗ T )1/2 of T is given by the partial isometry

vn ⊗ p1,n ∈ B ∗∗ ⊗ L(`2 ) ∼
X
U := = (B ⊗ K)∗∗ ,
n

∗
where (T T ) 1/2
=h 1/2
and (bn b∗n )1/2 vn = vn (b∗n bn )1/2 = bn . Notice that U ∗ U =
P ∗ ∗
P ∗ ∗ ∗∗
n vn vn ⊗ pn,n , and U U = n vn vn ⊗ p1,1 , the element vn vn ∈ B is equal
−n ∗∗ −n 1/2
to the open support projection Pn of (a − 2 )+ in B , vn (a − 2 )+ = bn and
∗ 1/2 1/2
vm vn = δm,n Pn . In particular U ∗ (bn b∗m ⊗ p1,1 )U = (a − 2−n )+ (a − 2−m )+ ⊗ pn,m .
Thus
U ∗ (C ∗ (bn b∗m ; m, n ∈ N) ⊗ p1,1 )U = λ(E) .
The C *-algebra E is isomorphic to a C *-subalgebra of C ∗ (a)⊗K of the sort consid-
ered in Lemma 2.9.18 and contains therefore a full stable C *-subalgebra by Lemma
2.9.18. 

Lemma 2.9.18. Let µ1 > µ2 > · · · a strictly decreasing zero sequence in (0, 1)
and define a positive contraction d ∈ C0 (0, 1] ⊗ K by

d := diag(2−1 (f0 − µ1 )+ , 2−2 (f0 − µ2 )+ , . . .)1/2 ∈ C0 (0, 1] ⊗ c0 ⊂ C0 (0, 1] ⊗ K ,

where f0 ∈ C0 (0, 1] is given by f0 (t) := t.

286 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

The hereditary C*-subalgebra E := d (C0 (0, 1] ⊗ K) d of C0 (0, 1] ⊗ K contains

a full stable C*-subalgebra D of E.

Proof. Let λ : N → N × N a bijective map from N onto N × N.

Then λ defines a decomposition of N into countably many pairwise disjoint
infinite subsets Mn := λ−1 ({n} × N). The natural order in Mn leads to bijective
order-preserving surjective maps ψn : N → Mn from N onto Mn . Let n0 , n1 ∈ N with
1 ∈ Mn0 and with min(N \ Mn0 ) ∈ Mn1 . Then ψn0 (k) = k for k = 1, . . . , ψn1 (1) − 1
and ψn (k) > k for all other pairs (n, k).
We define isometries Tn ∈ L(`2 ) ∼ = M(K) by Tn (ηk ) := ηψn (k) for the canonical
∗
Tn = δm,n 1 and n Tn Tn∗
P
ONB {η1 , η2 , . . .} of `2 (N). The Tn satisfy the relations Tm
converges strictly to 1 ∈ M(K) .
We define dk := (1 ⊗ Tk )d, then d∗j dk = δj,k d2 ∈ E and Check next
X
dk d∗k = 2−n (f0 − µn )+ ⊗ eψk (n),ψk (n) .
n

Since 2−n (f0 − µn )+ ≤ 2−ψk (n) (f0 − µψk (n) ) ???

dk d∗k =?????
We get dk d∗k ∈ E and dk ∈ E, because always n ≤ ψk (n), E is hereditary and
d∗k dk = d2 ∈ E.
Thus D := C ∗ (dj d∗k ; j, k ∈ N) is a full and stable C *-subalgebra of E. 

Remark 2.9.19. The algebra E := d(C0 (0, 1] ⊗ K)d of Lemma 2.9.18 is the
inductive limit of the hereditary C *-subalgebras
En := gn (C0 (0, 1] ⊗ K)gn with positive contractions gn ∈ `∞ (C0 (0, 1] ⊗ K) ⊂
M(C0 (0, 1] ⊗ K) given by the sequences

gn := diag(2−1 (f0 − µ1 )+ , . . . , 2−n (f0 − µn )+ , 2−n (f0 − µn )+ , . . .)1/2 ,

that are stationary beginning from n-th entry.

It is likely that the En are stable because each M(En ) is properly infinite and
En+1 is a extensions of K by

En ⊕ (0n ⊕ C0 (µn+1 , µn ) ⊗ K) .

Remark 2.9.20. The algebra D := d A d of Part (iii) of Lemma 2.9.16

give reason for
‘‘HAS NO UNITAL QUOTIENT’’
has no unital quotient and, for each element b ∈ D+ and ε > 0, there exist
c := c(b, ε) ∈ D such that kc∗ bck < ε and kc∗ c − bk < ε. By a criterium of J.
Hjelmborg and M. Rørdam in [373], this properties of D imply that D itself is a
stable σ-unital full C *-subalgebra of a A a . (Compare also Corollary 5.5.2 for an
alternative proof of this stability criterium.)
 10. LOCAL STABILITY AND PROPERLY INFINITE FULL PROJECTIONS 287

It follows that every σ-unital purely infinite C*-algebra A contains a stable σ-

unital hereditary C*-subalgebra D that is full in A. (cf. J. Hjelmborg, M. Rørdam,
[373], [688, prop. 5.4]).

10. Local stability and properly infinite full projections

Recall that a non-zero element a ∈ A is stable if a∗ Aa is stable, cf. Definition

2.1.1 of stable elements and Definition 2.0.3 of locally purely infinite C *-algebras.
(Perhaps one can here moreover take 1-stable elements or even n-stable ele-
ments in sense of Definition 2.1.3?)

Lemma 2.10.1. A C*-algebra B is locally purely infinite in the sense of Def-

inition 2.0.3, if and only if, for every non-zero b ∈ B+ and ε ∈ (0, kbk), there
exist n = n(b, ε) ∈ N and stable positive elements a1 , . . . , an in the hereditary
C*-subalgebra bBb such that (b − ε)+ is in the ideal of B generated by {a1 , . . . , an }.

Proof. Consider the algebraic ideal J0 that is generated by all stable positive
element of bBb. The closure J of the *-ideal J0 is an ideal of B. We show that
b∈J:
Suppose γ := kπJ (b)k > 0. Then there is a pure state ρ on B/J with ρ(πJ (b)) =
γ, simply by extending a character χ on C ∗ (πJ (b)) with χ(πJ (b)) = kπJ (b)k to a
pure state ρ on B/J.
Then λ := ρ◦πJ is a pure state on B that satisfies λ(b) = γ > 0 and λ(J) = {0}.
Let K the kernel of the irreducible representation Dλ : B → L(L2 (B, λ)) cor-
responding to λ. Clearly J ⊆ K follows from λ(J) = 0.
By Definition 2.0.3 there is a non-zero stable hereditary C *-subalgebra D ⊆
bBb which is not contained in K. Therefore, D is not contained in J. The iso-
morphism D ∼ = F ⊗ K (for some C *-algebra F ) shows that there is a separable
stable C *-subalgebra D1 ⊆ D with D1 6⊆ J. Then a strictly positive contraction
a ∈ (D1 )+ is not in J, but a is stable, because aBa = D1 . This contradicts the
assumption b 6∈ J.
The algebraic ideal J0 generated by the positive stable elements a ∈ bBb is
dense in J ⊃ bBb and (therefore) contains the Pedersen ideal of bBb. In particular,
(b − ε)+ ∈ J0 for all ε > 0. 

Lemma 2.10.2. Each non-zero locally purely infinite C*-algebra A (in the sense
of Definition 2.0.3) is ( 2-quasi-) traceless.

Proof. Let τ : A+ → [0, ∞] a lower s.c. 2-quasi-trace. If a ∈ A+ is a non-zero

contraction with τ (a) 6= ∞ and if δ ∈ (0, kak) then D := (a − δ)+ A(a − δ)+ is a
hereditary C *-subalgebra of A with (a − δ)+ ∈ D+ , dϕδ (a) = ϕδ (a)d = d for all
d ∈ D, where
ϕδ (t) := min(1, max(0, 2t/δ − 1)) ≤ 2t/δ ,
288 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

and
τ (d) ≤ kdk · 2τ (a)/δ for all d ∈ D+ . (10.1)
We show below that the Inequalities (10.1) yield that τ (d) = 0 for all d ∈ D+ ,
in particular that τ ((a − δ)+ ) = 0. The lower semi-continuity of τ shows then in
general that τ (a) = 0 if τ (a) < ∞, i.e., that τ takes only the values 0 and +∞ on
A+ . This holds for all l.s.c. 2-quasi-traces τ on A+ and means that A is traceless.
The Inequalities (10.1) show that τ is bounded on the set of contractions in
D+ 6= {0}.
Next similar to above said!!?
The Inequalities 10.1 hold because d ≤ kdkϕδ (a) ≤ kdk(2/δ)a for ϕδ (t) :=
min((2/δ)(t − δ/2)+ , 1), and C ∗ (d, ϕδ (a)) and C ∗ (a) are commutative C *-
subalgebras of A.
Compare next with above
and Lemma on stably generated C *-alg’s.
If b ∈ D+ is a non-zero stable contraction, then the multiplier algebra of bDb
contains a copy of O∞ unitally. Thus, there are d1 , d2 , . . . ∈ bDb with d∗j di = δi,j b.
Pn
It implies that nτ (d) = τ ( j=1 (dj )(dj )∗ ) ≤ (2/δ)τ (a) for each n ∈ N. Thus
τ (d) = 0.
Since the set of d ∈ D+ with τ (d) = 0 is the positive part of some closed ideal
of D, and since the ideal generated by the stable elements in D+ is dense in D
by Lemma 2.10.1, we get that τ (D+ ) = {0}. It shows that, τ ((a − δ)+ ) = 0 for
every δ > 0. Since τ is lower semi-continuous, we get τ (a) = 0 for all a ∈ A+ with
τ (a) 6= ∞. 

Lemma 2.10.3. The primitive ideal space Prim(A) can be covered by a countable
family of quasi-compact subsets – i.e., Prim(A) is σ-compact –, if and only if, there
exists a ∈ A+ with J(a) := span(AaA) = A (i.e., if and only if, A contains a full
σ-unital hereditary C*-subalgebra D := aAa ).
The space Prim(A) is quasi-compact, if and only if, there exists a ∈ A+ with
ka + Jk = 1 for all primitive ideals J ∈ Prim(A)

Proof. For b ∈ A+ with kbk ≤ 1, the generalized Gelfand transformation

bb : Prim(A) → R+ is defined by bb(J) := kb + Jk for all J ∈ Prim(A). (It is called
b̌ and Ǎ in [616].)
Notice here that the function bb is lower semi-continuous on the topological space
Prim(A), i.e., the J ∈ Prim(A) with bb(J) > s ≥ 0 build an open subset of Prim(A).
Let b ∈ A and t ∈ (0, ∞). We denote by Ct (b) ⊆ Prim(A) the set of J ∈
Prim(A) with (kb + Jk =)kπJ (b)k ≥ t.
Relate next together:
It is known that Ct (b) is a (quasi-) compact subset of Prim(A), cf. [217,
prop. 3.3.7], [616, prop. 4.4.4].
 10. LOCAL STABILITY AND PROPERLY INFINITE FULL PROJECTIONS 289

Then bb−1 [t, kbk] = Ct (b), and, therefore, Ct (b) is a quasi-compact subset of
Prim(A).
Let a ∈ A with I(a) = A, then k a + J k := k πJ (a) k > 0 for any closed ideal
J of A with J 6= A.
S
It follows Prim(A) = n C1/n (a). Now use that the sets C1/n (a) are quasi-
compact. Thus, Prim(A) is σ-quasi-compact.
Conversely: Let Kn quasi-compact subsets of Prim(A) such that Prim(A) =
S
n Kn . The supports in Prim(A) of the functions b with b ∈ A+ and kbk < 1 build
b
an upward directed system of open subsets of Prim(A), because for b1 , b2 ∈ A+ with
kbi k < 1 there is b3 ∈ A+ with kb3 k < 1 and bi ≤ b3 (i = 1, 2), cf. proof of [616,
thm. 1.4.2], and the generalized Gelfand transformation b 7→ bb is monotone on A+ .
For each primitive ideal J of A there is b ∈ A+ with kbk < 1 and bb(J) = kb+Jk > 0.
Thus, by quasi-compactness of Kn , there are bn ∈ A+ with kbn k < 1 such that Kn
bn . It follows that a := n 2−n bn ∈ A+ satisfies
P
is contained in the support of c
ka + Jk > 0 for all J ∈ Prim(A), i.e., J(a) is not contained in any primitive ideal
of A. Thus J(a) = A.
Notation overlap:
former notation I(a) had two different meanings!!!
Here we have changed to J(a) := span(A a A) .
If, moreover, Prim(A) is quasi-compact, then the open subsets Un :=
a−1 (1/n, ∞) of Prim(A) cover Prim(A). It follows that Up = Prim(A) for
b
some p ∈ N. We can replace a by f (a) for the function f (t) := min(p · t, 1). Then
a = 1.
b 

Is here the right place to show semi-projectivity of En ?

Better put it in the appendices. Near to other En stuff?
For later applications of the next result, recall that locally p.i. algebra are
traceless, cf. Lemma 2.10.2.

Proposition 2.10.4. Suppose that A is traceless (respectively that A is purely

infinite) and that Prim(A) is quasi-compact.
Give ref. to Def. of "quasi-compact" (T0 spaces)
Then there exists n ∈ N and a full properly infinite projection p ∈ Mn (A)
(respectively a full properly infinite projection p ∈ A).

Proof. At first, we consider the case where A is purely infinite, i.e., we con-
sider case where the element a is properly infinite for all non-zero a ∈ A. Later we
reduce the case of traceless A to the case of p.i. A.
Since Prim(A) is quasi-compact (by assumption), there exists a positive con-
traction a ∈ A+ with ka + Ik = 1 for all primitive ideals I ∈ Prim(A) (cf. Lemma
2.10.3).
290 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

It follows that (a − t)+ is a full positive contraction in A for all t ∈ [0, 1), i.e.,
J((a − t)+ ) := span(A(a − t)+ A) = A.
Since every (a − t)+ is properly infinite and full, there are d1 , d2 ∈ A with
d∗k (a − 2/3)+ dk = (a − 1/4)+ (k = 0, 1) and d∗1 (a − 2/3)+ d2 = 0.
1/2 1/2 1/2
Let e := (a − 2/3)+ d2 d∗2 (a − 2/3)+ and z := (a − 2/3)+ d1 h(a)1/2 for the
continuous function h(t) with h(t) := 8t for 0 ≤ t ≤ 1/2, and h(t) =: (t − 1/4)−1
for 1/2 ≤ t ≤ 1. It follows, (z ∗ z)(zz ∗ ) = zz ∗ , e ≥ 0, zz ∗ + e ≤ 1, and J(e) =
J((a − 1/4)+ ) = A. Let V := z + (1 − z ∗ z)1/2 . Then V is an isometry in the
unitization fA of A with V ∗ eV = 0, i.e., e ≤ p := 1 − V V ∗ ∈ A. We get that p is
a full projection in A.
Since p is properly infinite, there exists a C *-morphism ψ : E2 → A with ψ(1) =
p.
We consider now the general case where A itself is only traceless:
The algebra A ⊗ M2∞ is purely infinite if A is traceless (cf. [462, thm. 5.9], or
Theorem ??).
Thm. on p.i. tensor products ?!
Above we have seen that there exists a C *-morphism ψ : E2 → A ⊗ M2∞ such
that ψ(1) =: p is a full projection of A ⊗ M2∞ .
Since the union of the C *-subalgebras A ⊗ M2n is dense in A ⊗ M2∞ , we get
from Lemma A.2.1 that there are n ∈ N, a C *-morphism ψn : E2 → A ⊗ M2n and
a partial isometry vn ∈ A ⊗ M2∞ with vn vn∗ = ψ(1) and vn∗ vn = ψn (1).
Let I denote the closed ideal of A ⊗ M2n generated by vn∗ vn . Then I = J ⊗ M2n
for some closed ideal J of A, and ψ(1) = vn vn∗ is in the closed ideal J ⊗ M2∞ of
A ⊗ M2∞ . Since ψ(1) is a full projection of A ⊗ M2∞ , it follows J = A and that
ψn (1) = vn∗ vn must be a full projection of A ⊗ M2n . 

Corollary 2.10.5. If A is locally purely infinite and unital, then there is n ∈ N

such that the unit element 1A ⊗ 1n of Mn (A) is properly infinite.

Proof. Locally p.i. algebras are traceless by Lemma 2.10.2. If A is unital, then
Prim(A) is quasi-compact. Thus, by Proposition 2.10.4, there exists n ∈ N such
that Mn (A) contains a properly infinite full projection p. It follows p ≤ 1A ⊗1n - p,
i.e., 1A ⊗ 1n ≈ p is properly infinite in Mn (A). 

Lemma 2.10.6. If A is locally purely infinite, then for a ∈ A+ and ε > 0 there
exist n := n(ε) ∈ N and a σ-unital stable hereditary C*-subalgebra D ⊆ A⊗Mn with
ed = d for all d ∈ D and e := fε (a)⊗1n , such that (a−2ε)+ ∈ (A⊗Mn )D(A⊗Mn ).

If A has property pi-m, then one can find the n(ε) such that n(ε) ≤ m.
In particular, if A is purely infinite, then there is a stable σ-unital hereditary
C*-subalgebra D ⊆ A with (a − 2ε)+ ∈ ADA and fε (a)d = d for d ∈ D.
 11. WEAK PURE INFINITENESS FOR NON-SIMPLE C *-ALGEBRAS 291

Proof. Let B := Bε := (a − ε)+ A(a − ε)+ . By definition of l.p.i. algebras,

there are n ∈ N and stable hereditary C *-subalgebras Dk ⊆ B, k = 1, . . . , n, such
that (a − (3/2)ε)+ is in the closed ideal J generated by D1 ∪ D2 ∪ · · · ∪ Dn .
The direct sum F := D1 ⊕ D2 ⊕ · · · ⊕ Dn is stable and is naturally contained
in Mn (B) ∼
= B ⊗ Mn ⊆ A ⊗ Mn .
The algebra D := F (A ⊗ Mn )F ⊆ B ⊗ Mn is a stable hereditary C *-subalgebra
of A ⊗ Mn , and D generates the closed ideal J ⊗ Mn of A ⊗ Mn . The element
(a−(3/2)ε)+ ⊗1n is contained in J ⊗Mn , and (fε (a)⊗1n )x = x for all x ∈ B ⊗Mn .
If A has property pi-m, then g(a) ⊗ 1m = g(a ⊗ 1m ) is properly infinite for
all non-zero g ∈ C0 (0, kak]. Then Lemma ?? says that there is a stable hereditary
C *-subalgebra D ⊆ B ⊗ Mm such that the ideal J ⊗ Mm generated by D contains
(a − (3/2)ε)+ ⊗ 1n .
to be filled in ?? 

Lemma 2.10.7. If A is p.i. then for each a ∈ A+ , ε > 0, n ∈ N there exists a

row d ∈ M1,n (A) with dd∗ fε (a) = dd∗ and d∗ d = (a − ε)+ ⊗ 1n .
Here fε (t) is the continuous function on [0, ∞) with fε (t) = 1 on [ε, ∞), fε (t) =
0 on [0, ε/2] and fε (t) = 2t/ε − 1 on [ε/2, ε].

Proof. Recall fε (t) := min 1, (2/ε) max(0, t − ε/2) for t ≥ 0 and ε > 0.
Then (a − ε)+ is contained in the annihilator Ann(D) of the hereditary C *-
subalgebra D := (1 − fε (a))A(1 − fε (a)) of A.
?? By the observation of M. Rørdam [688, prop. 5.4], see also Proposition
??, there is a σ-unital stable hereditary C *-subalgebra E in (a − ε)+ A(a − ε)+ ⊆
Ann(D), such that (a − ε)+ is contained in the closed ideal of Ann(D) generated
by E.
It follows from Lemma ?? that there exists d1 , d2 ∈ Ann(D) with d1 d∗1 + d2 d∗2 ∈
E, d∗1 d2 = 0 and d∗1 d1 = (a − ε)+ = d∗2 d2 . 

Move to suitable places !!!

11. Weak pure infiniteness for non-simple C *-algebras

We have seen in Part ( ?????) ???? The properties pi(n) and pi-n are the same
for C *-algebras A with the additional property that 1M(A) is properly infinite in
M(A), e.g. as it isthe case if A is stable. Moreover, then both properties coincide
with the property that all non-zero n-homogenous elements a ∈ A+ are properly
infinite. This is because,
a) the algebra `∞ (A) is an ideal of `∞ (M(A)) and – therefore – has no quotient
of finite dimension if O∞ is unitally contained in M(A) (– more general if M(A)
has no quotient of finite dimension –) and
292 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

b) the property pi(n) always implies that an element a ∈ A+ is properly infinite

(as defined in Section ?? ) if a is n-homogenous.
Give ref’s to ‘‘new’’ places where it (= which ??) is proved !
There could be a considerable shift between the numbers n and the m in general.
Notice here that the numbers m and n are only upper estimates. Even in case of
separable amenable (non-simple) C *-algebras we do not know the possible minimal
values, except in special cases, e.g. if the algebras absorb the Jiang-Su algebra
tensorial.
An estimate for the n ∈ N of property pi-n for A ⊗ K for A with property pi-m
was given by n ≤ m2 in [463, prop. 4.5(vi)]. But it is unknown for which n ∈ N
the stabilization A ⊗ K is pi(n) (and is then there equal to pi-n) if A has property
pi(m). It seems to be related to topological properties of the space prime(A) of
prime ideals of A.
If A is pi(k) or pi-` for some k, ` ∈ N then there exist a number n ∈ N with
n ≤ min(k, `) such that every quasi-compact subset X of the space of prime ideals
prime(A) is covered by (finitely many) open subsets that are dense in prime(A)
and correspond to (stable) closed ideals J / A ⊗ K that are generated by an n2 -
homogenous element of A such that J has property pi(n) and pi-n at the same
time as stated in Proposition 2.11.1. It indicates that the relation between between
pi(n) and pi-n has something to do with the topology of prime(A) (= prim-ideal
space = point-wise completion of Prim(A)).
A partial converse and the passage to stabilization is given in Proposition 2.11.1.
Unfortunately we get only “local” equality, despite of the different focus of the
definitions.
Next is in parts only a conjecture!!

Proposition 2.11.1. Let A a non-zero C*-algebra.

Consider the following properties

(i,a) A is a pi(n) algebra in sense of Definition 2.0.4.

(i,b) A ⊗ K is “locally” a pi(n) algebra, i.e., for each n-homogenous element
b ∈ A+ the ideal J(b) := span(AbA) of A generated by b has the property
that J(b) ⊗ K has property pi-n.
It is known that it is true if b is instead n2 -homogenous in A.
Suppose that A has property pi(n). And that b ∈ A+ is n-homogenous.
Then (b − δ)+ is n-homogenous for every δ ∈ [0, kck) and is properly infi-
nite. Thus, for every ε ∈ (0, kck) there exists in D := (b − ε)+ A(b − ε)+
an n2 -homogenous element c ∈ D+ such that (b − 2ε)+ is contained in the
ideal J(c)AbA of A. Thus J((b−2ε)+ ) ⊆ J(c) ⊆ J(b) and J((b−2ε)+ )⊗K
has property pi-n. (The latter because property pi-n passes to hereditary
C*-subalgebras.) Property pi-n is preserved under inductive limits and
J(b) ⊗ K is the inductive limit of J((b − 1/n)+ ) ⊗ K, n ∈ N.
Thus:
 11. WEAK PURE INFINITENESS FOR NON-SIMPLE C *-ALGEBRAS 293

If A has property pi(n) and b ∈ A+ is n-homogenous then than J(b) ⊗

K has property pi-n.
(ii) A ⊗ K has property pi(n).
(iii) A ⊗ K has property pi-n in sense of Definition ??.
(iv) A has property pi-n.
(v) `∞ (A) has property pi-n.
(vi) `∞ (A) is locally p.i. in sense of Definition 2.0.3.
(vii) If J / A is a closed ideal of A, then A has pi-n (respectively pi(n)) with
n ≤ k + ` if J and A/J have pi-k (respectively pi-k) and pi-` (respectively
pi-`).

Then (ii) and (iii) are equivalent and imply both of (i,a) and (iv).
If A has property pi-n then `∞ (A), all non-zero quotients of A and all non-zero
hereditary C*-subalgebra of A have Property pi-n.
In particular, A is purely infinite in the sense of Definition 1.2.1, if and only
if, each non-zero element 0 6= a ∈ A is properly infinite, i.e., a ⊕ a - a for each
a ∈ A (and means that A has property pi-1).

Ideas of a proof of Proposition 2.11.1 will be given farer below after we have
stated and proven the needed lemmata.
Clearly ???? the trivial implications are ??? (ii)⇔(iii), ??? ??? (ii)⇒(i), and
??? (iii)⇒(iv), the implication (iv)⇒(i) requires only some simple and obvious
matrix reformulation of the property pi(n) in Definition 2.0.4.
The nontrivial implication ??? (i)⇒(ii)? need some more work and preparation.

Lemma 2.11.2. Suppose that A has no irreducible representation of dimension

≤ n.
If A satisfies Part ( i ) of property pi(n), then each non-zero element a ∈ A+
in the ideal J generated by some (n + 1)-homogenous element b ∈ A+ satisfies that
a ⊗ 1n is properly infinite in A ⊗ Mn .
(Alternatively, we could require here that b ∈ A+ is n-homogenous and that for
each ε > 0 the ideal generated by Ann((b − ε)+ , A)+ := {c ∈ A+ ; c(b − ε)+ = 0 }
contains (b − 2ε)+ .)
There is (somewhere farer below) a new result that says every non-zero n-
homogenous element b ∈ A+ is properly infinite in A if A is pi(n). I.e., the latter
annihilator condition on (b − ε)+ is not needed.
If A is pi-n, then for each element a ∈ (A ⊗ K)+ and ε > 0 there exists b ∈ A+ ,
d ∈ A ⊗ K such that d∗ (b ⊗ 1n )d = (a − ε)+ . It follows that c := a ⊗ 1n2 is
“spectrally properly infinite” for each a ∈ A ⊗ K, i.e., ϕ(c∗ c) is properly infinite for
each ϕ ∈ C0 ((0, kck2 ]).
Properties pi(n) and pi-n coincide on stable C*-algebras.
294 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Holds in general for all C*-algebras

A with properly infinite 1 in M(A).
If A has Property pi(n) ϕ : B := C0 ((0, 1], Mn2 +1 ) → A is a C*-morphism,
then the closed ideal J of A generated by ϕ(B) has the property that J ⊗ K has
property pi-n and this property coincide there with property pi(n), as for all stable
C*-algebras.

In Proposition ?? we list some permanence properties of the class of weakly

purely infinite C *-algebras. We consider weak versions of pure infiniteness because
it is usually easier to recognize if a C *-algebra A is locally or weakly purely infinite
in the sense of Definitions 2.0.3 and 2.0.4 e.g. if one considers tensor products,
crossed products or continuous fields. Only strongly purely infinite algebras have
properties that in good cases allow to classify them by functors into Abelian types
of categories. Usually one gets the needed stronger assumptions by combination
– say of weak pure infiniteness with corona factorization properties or with the
existence of reasonable central sequences.
A basic open question is if property pi(n) or even property l.p.i. for A im-
plies that A has property pi(1) or moreover that A has the in applications needed
property that A is strongly purely infinite in sense of Definition 1.2.2.
We do not know any example where A is weakly purely infinite but is not
pi(1) or where A is l.p.i. but is not weakly purely infinite, and we do not know an
example where A is purely infinite in sense of Definition 1.2.1, i.e., is pi(1) in sense
of Definition 2.0.4 but is not strongly purely infinite.
It is not difficult to see that C([0, 1], A) is pi(2) if A is purely infinite. More
generally, C0 (X, A) is pi(n) with n ≤ m + dim(X) if A is pi(m) and X is a finite-
dimensional and locally compact.
Reference for above? Formula OK?
For the following we need a definition and two lemmata:

Definition 2.11.3. An element a ∈ A is spectral properly infinite inside A if

for each ε > 0 the cut down (a∗ a − ε)+ of a∗ a is properly infinite in A.
ESpec (A) ⊆ A (or simply ESp ) will denote the set of all spectral properly infinite
elements in A.
EFunct (A) ⊆ A (or simply EFunct ) denotes the set of all elements a ∈ A with the
property ψ(a∗ a) ∈ ESpec (A) for all functions ψ ∈ C0 (0, kak2 )+ with ψ(a∗ a) 6= 0 .

Lemma 2.11.4. Let Espec (A) ⊆ A, (respectively Efunct (A) ⊆ A) the sets of all
“spectral” (respectively “functional”) properly infinite elements in sense of Defini-
tion 2.11.3.
Then ESpec (A) and EFunct (A) are closed subsets of A \ {0}.

Proof. To be filled in ?? 
 11. WEAK PURE INFINITENESS FOR NON-SIMPLE C *-ALGEBRAS 295

Next is still a conjecture ... !!!

Good for what? ... replace by other idea !!!

Lemma 2.11.5. Let a, b ∈ A elements such that ψ(a∗ a) is properly infinite

in A for every ψ ∈ C0 (0, kak2 ]+ with ψ(a∗ a) 6= 0. Let a = v(a∗ a)1/2 the polar
decomposition of a in A∗∗ . Then there exists for each δ ∈ (0, kak2 ) a continuous
path of elements t ∈ [0, 1] 7→ b(t) ∈ A with b(t)∗ b(t) = ((a∗ a) − δ)+ for all t ∈ [0, 1],
1/2 1/2
b(0) = v((a∗ a) − δ)+ and b(1) = ((a∗ a) − δ)+ .

It is wrong in this formulation:

Let A := O2 = C ∗ (s, t ; s∗ t = 0, ss∗ + tt∗ = 1, s∗ s = 1 = t∗ t)
1/2
and a := s, δ = 1/2. Then v((a∗ a − 1/2)+ = 2−1/2 s. There is no continuous
path t → b(t) with b(t)∗ b(t) = 1/2, b(0) = 2−1/2 s and b(1) = 2−1/2 , because there
is no continuous path from s to 1 inside the the set of isometries of O2 .
But if there enough “stable room” orthogonal to (a∗ a − δ)+ and (aa∗ − δ)+
(perhaps moving with the parameter) then there is some hope.

Proof. (The definition of gδ has to made precise.) The idea is: For each µ, γ ∈
(0, kak2 ) there exist contractions s1 , s2 ∈ A with s∗1 s2 = 0, s∗1 s1 = g2δ (a∗ a) = s∗2 s2
and s1 s∗1 + s2 s∗2 ≤ gδ (a∗ a) (a “soft” version of E2 with gδ (t) = 1 for t ≥ δ and
gδ (t) = 0 for t ≤ δ/2 linear on [δ/2, δ].) and

(a∗ a − γ)µ+ s∗k sk (a∗ a − γ)µ+ = (a∗ a − γ)2µ ∗

+ g2δ (a a) .
1/2 1/2
Now consider (with suitable δ), X := v ·((a∗ a)1/2 −δ)+ s∗1 , Y := ((a∗ a)1/2 −δ)+ s∗2
1/2
and Z1 := s2 s∗1 − s1 s∗2 (or + ?) XZ1 Y ∗ = v((a∗ a)1/2 − δ)+ s∗1 s1 s∗2 s2 ((a∗ a)1/2 −
1/2
δ)+ =
= v((a∗ a)1/2 − δ)+ gδ/2 (a∗ a)2 (“nearly” XY ∗ )
Z2 := s1 s∗1 + s2 s∗2
Has to study (X + Y )Zk X ∗ =??????
Idea does not work??
To be filled in ?? 

We do not know (2019) if A is purely infinite for separable A if A has the

property that there exists a unital C *-morphism ψ : C[0, 1] → M(A) into the
center of M(A) such that the “fibers” At := A/It are all purely infinite. Here the
the ideal It of A is the closed linear span of ψ(C0 ([0, 1] \ {t})) · A).
(Perhaps, one gets an answer if A has the “corona factorization” property
(CFP) in addition? But what have they to do with (CFP)?)
But we know that C0 (X, A) (∼= A ⊗ C0 (X)) is pi-n if A is pi-n and X is any
locally compact Hausdorff space:
It reduces to the case of C([0, 1], A), because then one can step up by induc-
tion over m ∈ N to C([0, 1]2m+1 , A). C0 (X, A) with dim(X) ≤ m is an ideal of a
296 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

quotient of C([0, 1]2m+1 , A), and C0 (X, A) of an arbitrary (not necessarily metriz-
able) l.c. Hausdorff spaces X is an inductive limit of an upward directed net of
C *-subalgebras isomorphic to C0 (Xλ , A) with Xλ of finite dimension.
Thus, it reduces to the case of C([0, 1], A), and it is easy to see that is suffices to
consider only those elements of C([0, 1], A) that are given by pice-wise linear map
φ from [0, 1] into A+ with kφ(t)k ≤ 1, i.e., have to check pice-wise linear maps φ
from [0, 1] into positive contractions in A. In fact it suffices to consider those maps
φ : [0, 1] → A+ for given δ > 0 with break points t` ∈ [0, 1] with t` < t`+1 , t1 = 0,
tm = 1 and kφ(t` ) − φ(t`+1 )k < δ, but it suffices to consider those paths that satisfy
moreover φ(t2k ) = φ(t2k+1 ) for n = 2k, i.e., with φ(t) = φ(t2k ) for t ∈ [t2k , t2k+1 ]
and only φ(t2k−1 ) 6= φ(t2k ) . This is because any continuous map ψ from [0, 1] into
the contractions in A+ can be approximated by such continuous maps φ arbitrarily
well.
We can go with φ(t) to the next break point with a constant pice of “path”
between.
Let Ik := [t2k−1 , t2k ] ⊆ [0, 1]. Then we can “compress” the path φ(Ik ) by a
continuous map c : Ik → A, given by a path {c(t) ; t ∈ Ik } ⊆ A for Ik of contractions
in A to
Xk := 2−1 (φ(t2k−1 ) + φ(t2k )) − δ/2 + ,

i.e., find a contraction ck ∈ C(Ik , A) such that ck (t)∗ φ(t)ck (t) = Xk for t ∈ Ik . Here
we can apply Lemma 2.1.9 to the elements a := φ|Ik and b := 2−1 (φ(t2k−1 )+φ(t2k ))
in the C *-algebra C(Ik , A) because

kφ(t) − 2−1 (φ(t2k−1 ) + φ(t2k ))k < δ/2 for all t ∈ Ik .

Then kXk − φ(t)k < δ for t ∈ Ik = [t2k−1 , t2k ].

Then we can go back “up to δ” from the “constant” path (Xk − δ)+ on Ik
to φ(Ik ) ∈ C(Ik , A), i.e., find by a contraction in e = {e(t)} ∈ C(Ik , A) with
e(t)∗ (Xk − δ)+ e(t) = (φ(t) − 2δ)+ for t ∈ [t2k−1 , t2k ]. This is possible by Lemma
2.1.9 because k(Xk − δ)+ − φ(t)k < 2δ.
We find for the properly infinite element 1n ⊗ Xk ∈ Mn (A) elements dk ∈
M2n (A) with d∗k ((1n ⊗ Xk ) ⊕ 0n )dk = 12n ⊗ (Xk − δ)+ , because A has property
pi-n.
We define Dk (t) := ((1n ⊗ c(t)) ⊕ 0)dk (12n ⊗ ek (t)) .
Then Dk ∈ C([t2k−1 , t2k ], M2n (A)) and

Dk (t)∗ (1n ⊗ φ(t) ⊕ 0n )Dk (t) = 12n ⊗ (φ(t) − 2δ)+ .

But we do not know if C([0, 1], A) is pi(1) if A is pi(1). By Proposition ?? for

every compact space X the algebra C(X, A) is pi-n (respectively is s.p.i.) if and
only if A is pi-n (respectively is s.p.i.). For simple A all up to now given definitions
of pure infiniteness coincide with strong pure infiniteness.
 11. WEAK PURE INFINITENESS FOR NON-SIMPLE C *-ALGEBRAS 297

But for A with property pi(n) we get only that C(X, A) has pi(m) with estimate
m ≤ n · (Dim(X) + 1). ???? (If X = [0, 1] then m ≤ 2n.)
It would be sufficient to prove that C([0, 1], A) is pi-n if A is pi-n, cf. proof of
Proposition ??.
OLD or NEW or REMAINING?: The question, if C([0, 1], A) is pi(1) for
purely infinite A, is equivalent to the following:

Let A a purely infinite C *-algebra, a, b ∈ A+ positive contraction, ε > 0, τ > 0

and d1 , d2 ∈ A with d∗j adk = δjk (a − ε)+ (j, k ∈ {1, 2}).
Try to find e1 , e2 ∈ C([0, 1], A) with kdj − ej (0)k < τ for j ∈ {1, 2} and
ej (t)∗ ((1 − t)a + tb)ek (t) = δjk ((1 − t)a + tb) − ε +


for t ∈ [0, 1] .

One can separate the question into the intervals [0, 1/2] and [1/2, 1]. a ≤ a + b,
there exist contractions T, S ∈ C([0, 1/2], A) with T (t)∗ ((1 − t)a + tb)T (t) = (1 − t)a
and S(t)∗ ((1 − t)a + tb)S(t) = tb, as e.g. the
Not ready?
strict ??? limit in ??? Cb ([0, 1/2], A)
T (t) := lim((1 − t)a + tb + 1/n)−1/2 (1 − t)1/2 a1/2
n

More likely is that one can find a contraction T ∈ C([0, 1/2], A) with T (t)∗ (a +
t/(1 − t)b)T (t) = (a − ε)+ and use this locally ???
It suffices to consider the case of small ka − bk.
Given f ∈ C([0, 1], A)+ with kf k < 1, we find for each ε > 0 and t ∈ [0, 1],
some δ ∈ (0, 1) such that kf (t) − f (s)k < ε for all s ∈ [t − δ, t + δ] ∩ [0, 1]. There we
take γ ∈ (0, min(δ, ε)) and a contractions d1 , d2 ∈ C([0, 1], A) with
d1 (s)∗ (f (s) − γ)+ d1 (s) = (f (t) − ε)+
and
d2 (s)∗ (f (t) − 2ε)+ d2 (s) = (f (s) − 3ε)+ .

One gets at the end-points f (t − δ) and f (t + δ) different rows D, e.g. with

D∗ f (t − δ)D = diag((f (t − δ) − 3ε)+ , (f (t − δ) − 3ε)+ ) . ????
For simple locally purely infinite A the algebra C([0, 1], A) is strongly purely
infinite because simple l.p.i. algebras are strongly p.i. by Proposition 2.2.1(v). The
later considered applications requires strong pure infiniteness in sense of Definition
1.2.2 for the algebras in question.
We apply permanence properties of the considered class of strongly purely
infinite algebras, and sometimes additional properties of A ( 44 ), to obtain that
a particular class of C *-algebras A has the property that the locally or weakly
purely infinite algebras in this particular class are automatically strongly purely
infinite. This unsolved verification problem restricts the applications of the notion
44E.g., that A ∼ A ⊗ D for some tensorial self-absorbing unital algebra D 6= C.
=
298 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

of purely infinite algebras considerably. It is still unknown if all the definitions of

pure infiniteness coincide at least in case of (non-simple) nuclear C *-algebras A.
In case of non-elementary simple C *-algebras A, i.e., A 6∼ = K(H) for any Hilbert
space H, the algebra A is locally p.i. (i.e., is l.p.i. in sense of Definition 2.0.3) if and
only if A is strongly p.i. in sense of Definition 1.2.2.
It could be that topological properties of the primitive ideal space Prim(A) of
A plays a role in a final answer. It could also be that non-existence or existence
of characters for the C *-algebras F (B, A) := (B 0 ∩ Aω )/ Ann(B, Aω ) for B any
character-less sub-homogenous C *-subalgebra of Aω plays a role in a final answer.
Unfortunately only rather partial results exist and are not obvious. Possibly related
properties could be ideal system equivariant versions of the corona factorization
property (CFP), but that is not clear.
In principle, all what we must require for the classification works only for
“target” algebras B that are strongly purely infinite, e.g. the construction of suitable
nuclear *-monomorphisms A ,→ B of separable exact A into a C *-algebra B needs
that B is strongly purely infinite. One can do really nothing with the algebra B if
one can only prove pure infiniteness, weak pure infiniteness or even only local pure
infiniteness. But sometimes additional properties of B allow to conclude later that
B or “sufficiently big” C *-subalgebras of B are strongly purely infinite. This is the
true reason for our study of the “non-strong” versions of pure infiniteness, and that
this “weak” properties are easier to check.

12. Non-simple purely infinite algebras

There are several definitions of “pure infiniteness” for C *-algebras between

“locally purely infinite” and “strongly purely infinite”. Their interrelations and
permanence properties have been studied in [462], [463], [92], [93] and [443]. The
properties are equivalent in the case of algebras of real rank zero or in the case
of algebras with finite-dimensional Hausdorff primitive ideal spaces. It is still an
open question, whether or not all this properties are equivalent in general. We
get some partial results in this direction (cf. also [443]). Open problems will be
listed in Section ??. We generalize some methods of Section 2 for the study of pure
infiniteness of non-simple algebras.
HERE comes imported text from other places!
Move next blue discussion to later place!
Replace citations by similarities ??? here!
Property pi-n implies property pi(n) on A. Conversely if A has property pi(m)
for some m then there exists n ≥ m such that A has property pi-n, but no general
function f : N → N is known yet with the property that n ≤ f (m) , such that A has
property pi-f (m) if A has property pi(m). Moreover, there exists for every factorial
state ρ on A a closed ideal J of A with the properties

(i) ρ|J 6= 0,
 12. NON-SIMPLE PURELY INFINITE ALGEBRAS 299

(ii) J is “essential” in A, i.e., a ∈ A and a · J = {0} implies a = 0, and

(iii) the ideal J has property pi-n (if A has property pi(n)).

If A has property pi(m) then there exists n ≥ m such that A has property pi-n.
But until now no estimating function function n(t) is known with the property that
n(m) ≥ m if A has property pi(m). It seems to depend from topological properties
of Prim(A).
We give an overview about the known results and open questions concerning
non-simple (locally, weakly or strongly) purely infinite algebras in Section ?? – with
some parts of the needed proofs postponed to later sections and chapters.
By [463, lem. 4.4]):
Let a be a non-zero positive element in a C *-algebra A and let n be a natural
number. The following conditions are equivalent:
(i) a ⊗ 1n is properly infinite,
(ii) a ⊗ 1m - a ⊗ 1n for all (natural numbers) m ∈ N,
(iii) a ⊗ 1m - a ⊗ 1r for all natural numbers r, m with r ≥ n,
(iv) a ⊗ 1n+1 - a ⊗ 1n ,
(v) for each ε > 0 and for each m in N there is x in Mm,n (A) such that x∗ x belongs
to Mn (aAa) and xx∗ = (a − ε)+ ⊗ 1m .
Permanence properties of property pi-n:
The property pi-n on A passes to non-zero hereditary C *-subalgebras D ⊆ A,
non-zero quotients A/J, and to A∞ := `∞ (A)/c0 (A) and the ultrapowers Aω and
(that are special quotients of `∞ (A)).
If A and B have property pi-n, then A ⊕ B has property pi-n, `∞ (A1 , A2 , . . .)
is pi-n for sequences A1 , A2 , . . . of C *-algebras.
Both of pi-n and pi(n) pass from A to ultra-powers Aω .
All n-homogenous elements of A are properly infinite in A itself, if A has
property pi-n or pi(n).
The property pi(n) is equal to property pi-n on each closed ideal J of A that
is generated by a hereditary C *-subalgebra D of A with the property that D is
n-homogenous in the sense that there exists a hereditary C *-subalgebra E ⊆ D
such that Mn (E) ∼ = D. The maximal closed ideals J with this property have in
Prim(A) and prime(A) open supports that are there dense, i.e. there is no non-zero
closed ideal orthogonal to J.
Thus, all is reduced to an extension problem for the sum of two ideals ...
The Property pi-n will be preserved by inductive limits,
passes to M2 (A), – and then to stabilizations of A, – ...???
Pm ∗
If A+ 3 b = k=1 dk adk for a ∈ A+ , then for each ε > 0, there exists
Pn ∗
e1 , . . . , en ∈ A with the property (b − ε)+ = j=1 ej aej . The algebra A has
no non-zero quotient A/J of finite dimension, because property pi-n carries over to
300 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

non-zero quotients, and all non-zero hereditary C *-subalgebras of A/J. Certainly

C is not pi-n.
More generally, all C *-algebras with property pi(n) or pi-n are residually an-
tiliminary.
??? Why???
Part (iv) shows only that an n-homogenous element in A is infinite in A itself
if there exist something orthogonal to this element...
Needs to to show that the infiniteness in A itself is the same as infiniteness in
Mn (A).
A ∈ a ≈ a ⊗ e11 in A ⊗ Mn and a ⊗ e11 ≈ b ⊗ 1n and (b ⊗ 1n ) ⊕ 0 ≈ (b ⊗ 1n ) ⊕ c
in A ⊗ Mn+1 , then there must be shown that

a ⊗ e11 ≈ a ⊕ c = a ⊗ e11 + c ⊗ e22

in A ⊗ M2 ∼
= M2 (A). Here A ⊕ A is identified with the diagonal matrices in M2 (A).
It is not difficult to see that `∞ (A) is pi-n if A is pi-n. Therefore it implies
that `∞ (A) can not have a finite dimensional quotient, i.e., that property pi-n
implies property pi(m) for some m ≤ n. This unknown m ∈ N could depend from
the topology of Prim(A). This could be because for T0 (non-Hausdorff) spaces the
covering dimension and the decomposition dimension can be very different. Perhaps
also the lattice structure of the family of hereditary C *-subalgebras (or closed left
ideals) could play some role ...
It is easy to see that property pi(n) implies that each non-zero (n + 1)-
homogenous element is properly infinite in A. tion ?? provides the non-trivial
fact that in general all n-homogenous elements in A are properly infinite if A has
property pi(n).
It implies properties pi(n) and pi-n are the same for C *-algebras A with the
property that 1M(A) is properly infinite in M(A), e.g. in case where A is stable.
Moreover, then both properties coincide with the property that all non-zero n-
homogenous elements a ∈ A+ are properly infinite, cf. Proposition 2.7.16.
The above given Definitions 2.0.4, ?? of local, weak pure infiniteness coincide
for simple C *-algebras, cf. Proposition 2.2.1.
(Was mentioned earlier)

Definition 2.12.1. We call a C *-algebra A hyper-antiliminary if each non-

zero quotient C *-algebra `∞ (A)/J of `∞ (A) is antiliminary (= NGCR).

Is it equivalent to the property that no hereditary C*-subalgebra D of `∞ (A)

has a character?
It should be equivalent to the property that `∞ (A) is “residual antiliminary”
in the sense of Definition 2.7.2
FIND or Give general Definition
 12. NON-SIMPLE PURELY INFINITE ALGEBRAS 301

The in next Lemma used term “antiliminary” should be recalled “residual an-
tiliminary” ??

Lemma 2.12.2. The C*-algebra A is hyper-antiliminary, if and only if, there

exist an universal constant n := n(A, 2) ∈ N such that for each contractions
a1 , a2 , a3 ∈ A+ with a1 a2 = a1 and a2 a3 = a2 there are n contractions d1 , . . . , dn ∈
a3 Aa3 and 2-homogenous positive contractions e1 , . . . , en ∈ a2 Aa2 such that a1 =
Pn ∗
j=1 dj ej dj .

If A is hyper-antiliminary then there exist numbers n(A, k) ∈ N, k = 2, 3, . . .,

n(A, k+1) ≥ n(A, k), such that the same holds with n := n(A, k) and k-homogenous
contractions ej ∈ A+ .
The C*-algebra A is hyper-antiliminary if no non-zero hereditary C*-subalgebra
of `∞ (A) has a character.
The latter is the case if and only if D(`∞ (A))∩K(H) = {0} for every irreducible
representation D : `∞ (A) → L(H) of `∞ (A).

Proof. It is not difficult to see that a C *-algebra B is antiliminary if and only

if b B b has no non-zero character for each positive contraction b ∈ B.
Indeed: If b B b has a character, then this character extends to a pure state ρ
on B that defines an irreducible representation D of B that contains the compact
operators in the image of B.
Conversely, if D is an irreducible representation and D(B) contains the compact
operators, then there exists a contraction b ∈ B+ such that D(b) is a projection
with rank one. Then b B b has a non-zero character.
Moreover, if J / B is a closed ideal such that B/J contains a non-zero Abelian
hereditary C *-subalgebra E, then πJ−1 (E) is a hereditary C *-subalgebra of B that
has a non-zero character.
Take now B := `∞ (A) and let b := (b1 , b2 , . . .) with contractions bn ∈ A+ .
Consider the Abelian C *-subalgebras C ∗ (cn ) ⊆ A generated by cn :=
−k
P
k2 bn+k .
to be filled in... ??
lem:Char.A.hyper.antilim 

Proposition 2.12.3. Suppose that A locally purely infinite (l.p.i.). Then A

has following properties:

(1) Every hereditary C*-subalgebra D ⊆ A is l.p.i.

(2) Every quotient A/J is l.p.i.
(3) A ⊗ B is l.p.i. for every exact B.
(4) A is (2-quasi-) traceless, cf. Lemma 2.10.2.
(5) If A is l.p.i. then A is residually antiliminary.

Moreover:
302 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

All inductive limits of l.p.i. algebras are l.p.i.

It is not known if `∞ (A) is l.p.i. if A is l.p.i. It would show that “l.p.i.” implies
“w.p.i.” (in some sense).

Proof. ?? Where is the Def. of ”residually anti-liminary”

(Is it the same as: Every hereditary 

Lemma 2.12.4. Suppose that A satisfies condition (i) of Definition 2.0.4 of the
property pi(n). Then:

(i) Property (i) of Definition 2.0.4 for pi(n), i.e., that each element b ∈ J(a)
P
can be approximated by n-term sums 1≤k≤n ek adk , passes to quotients
A/J of A, hereditary C*-subalgebras of A and to `∞ (A).
(ii) This has been improved in between (time?):

Each non-zero n-homogenous a ∈ A+ is properly infinite inside D :=

aAa if A satisfies Property (i) of Definition 2.0.4 for pi(n) and A has no
irreducible representation of dimension ≤ n. –
It carries over to n-homogenous elements of `∞ (A) but excludes not
irreducible representations of `∞ (A) with dimension ≤ (n − 1)... ????
One point here is that we do not know if characters χ on h`∞ (A)h
with χ(h) = 1 for h = (h1 , h2 , . . .) ∈ `∞ (A)+ with khk = 1 (can moreover
take here h with khn k = 1 for all n ∈ N). can be represented on all
separable C*-subalgebras C ⊆ `∞ (A) with h ∈ C by a sequence ρn of pure
(!) states on A such that V : `∞ (A) → `∞ (C) given by
V ((a1 , a2 , . . .)) := (ρ1 (a1 ), ρ2 (a2 ), . . .)
and a character ψ on `∞ (C) have the property that for all c ∈ C holds
χ(c) = ψ(V (c)).
The point is that all bigger separable C*-sub-algebras in h`∞ (A)h must
have the same representation of χ.
Then still all is rather complicate: Have to show that there exists
1/2 1/2
2-homogenous elements gn ∈ A+ with kgn k = 1 and ρn (hn gn hn ) ≥
2/3ρn (hn ).
(Perhaps even a stable element gn if in question for other topics?)
Then χ can not be a character on h`∞ (A)h, provided that χ = ψ ◦ V
holds also on the hereditary C*-algebra generated by the element
(h1 , h2 , . . .) ...
A lot to check!
Suppose that non-zero a ∈ A+ is n-homogenous ( 45 ).
Let J := J(a) := Span(AaA) denote the closed ideal of A that is
generated by {a}, and let Ann(a, J) := {b ∈ J ; ab = 0 = ba} , i.e.,
Ann(a, J) = J ∩ Ann(a, A).

45i.e., there exists a C *-morphism ψ : C (0, kak]⊗M → A with a = ψ(f ⊗1 ) for f (t) := t
0 n 0 n 0
 12. NON-SIMPLE PURELY INFINITE ALGEBRAS 303

If the closed two-sided ideal of A generated by Ann(a, J) contains a,

then a is properly infinite.
In particular, if c ∈ A+ with kck = 1 is an (n+1)-homogenous element
of A, then (c − δ)+ is properly infinite for each δ ∈ [ 0 , kck ).
The dimension of H is ≤ n, if there exists an irreducible representa-
tion D : A → L(H) with D(A) ∩ K(H) 6= {0}.
(iii) If A has no non-zero quotient A/J of dimension ≤ n2 , then a∗ Aa is pi(n)
for each a ∈ Ped(A) (:= Pedersen ideal, the minimal dense ideal of A).
(iv) If M(A) has no non-zero quotient of dimension ≤ n2 , then A is pi(n).
(v) If A is σ-unital, then, for every S, T ∈ Q(A) := M(A)/A (respectively
S, T ∈ M(A)) with S in the closed ideal generated by T , and every ε > 0,
there exist X1 , . . . , X2n ∈ Q(A) (respectively X1 , . . . , X3n ∈ M(A)) with
kS − j Xj∗ T Xj k < ε.
P

(vi) If A is σ-unital and is pi(n), then M(A) has no finite-dimensional quotient

and is pi(k) (for some k ≤ 3n).
(vii) If A (itself ) has no irreducible representations of dimension ≤ n, then A
is locally purely infinite (cf. Def. 2.0.3).
In particular, then A is traceless (in sense of Def. 2.8.1).

Proof. (i): The passage to hereditary C *-subalgebras and to quotients A/J

is almost obvious from the definition.
The passage of property pi(n) to `∞ (A) requires an uniform estimate for prop-
erty (i) of Definition 2.0.4:
If bk := Sk ak Sk∗ , for ak , bk ∈ A+ , Sk ∈ M1,m (A) (for some fixed m ∈ N) with
kak k ≤ 1 and γ := supk kSk k < ∞ , then kbk − Sk (ak − δ)+ Sk∗ k ≤ γ 2 δ . This allows
√
to find Tk ∈ M1,n (A) with supk kTk k ≤ 2/ ε, such that kbk − Tk ak Tk∗ k < ε for all
k ∈ N. Now use that M1,n (`∞ (A)) = `∞ (M1,n (A)).
(ii): Let b := ψ(f0 ⊗ p1,1 ). The elements a and b generate the same closed
ideal J, and b ⊗ 1n ∼ a ⊕ 0n−1 in Mn (A). For x ∈ Ann(a, J) holds a + x ∈ J.
Thus (a + x) ⊕ 0n−1 - b ⊗ 1n , . Since ax = xa = 0, also a ⊕ x ∼ (a + x) ⊕ 0.
Together: a ⊕ x ⊕ 0n−2 - a ⊕ 0n−1 , i.e., x is in I(a) := {x ∈ A ; a ⊕ x - a} for all
x ∈ Ann(a, J). By Lemma 2.5.3(i), I(a) is a closed ideal of A. By assumptions, a
is in the closed ideal of A that is generated by Ann(a, J)+ . Thus, a ⊕ a - a ⊕ 0
follows from Ann(a, J) ⊆ I(a).
If ψ : C0 (0, γ] ⊗ Mm → A is given with γ := kck , m > n and ψ(f0 ⊗ 1m ) = c
for f0 (t) := t, then, for each 0 ≤ δ < γ , the element (a − δ)+ = ψ((f0 − δ)+ ⊗ en )
(where en := p11 + p22 + · · · + pnn ) satisfies the above considered assumptions on a.
Thus, (a − δ)+ is properly infinite. The element (c − δ)+ is in the ideal generated
by {(a − δ)+ }. We get (c − δ)+ - (a − δ)+ ≤ (c − δ)+ .
Thus (c − δ)+ is properly infinite for all δ ∈ [0, kck) ...????
If D : A → L(H) is an irreducible representation, then the compact operators
K(H) are contained in the image of D, and K(H) becomes isomorphic to a quotient
304 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

of an ideal of A. K(H) satisfies property (i) of Definition 2.0.4, if and only if H has
dimension ≤ n, because K(H) contains no properly infinite element, but contains
n + 1-homogenous elements if the dimension of H is > n.
(iii): All finite-dimensional irreducible representations of A are of dimension
≤ n, by our assumptions on A. But this sort of irreducible representations can not
exist by our assumption that A has no non-zero quotient of dimension ≤ n2 . It
implies that f Mn (A)f has no finite-dimensional quotient if 0 6= f ∈ Mn (A)+ .
Let a ∈ Ped(A). Then b := a∗ a ∈ Ped(A) and B := a∗ Aa = bAb. The
(minimal dense) Pedersen ideal of A is algebraically generated by the elements
{(e − ε)+ ; e ∈ A+ , ε > 0} . Hence, there are e ∈ A+ , ε > δ > 0 and d1 , . . . , dn with
b = d∗1 (e − δ)+ d1 + · · · + d∗n (e − δ)+ dn (the n comes from property (i) of Definition
2.0.4).
There are contractions f, x ∈ Mn (A) and x ∈ Mn (A) such that f xx∗ = xx∗
and f ≥ 0 and B ⊗ p11 = xMn (A)x∗ .
It follows that `∞ (B) is isomorphic to a hereditary C *-subalgebra of
Mn (`∞ (A)) that is contained in gMn (Aω )g for g := (f, f, . . .), where we use
`∞ (Mn (A)) ∼
= Mn (`∞ (A)).
Since, by Part (i), B, `∞ (B) and `∞ (A) satisfy the condition (i) of Defini-
tion 2.0.4, all irreducible representation of `∞ (B) that contain non-zero compact
operators in its image are of dimension ≤ n, and all irreducible representation of
Mn (`∞ (B)) that contain non-zero compact operators in its image are of dimen-
sion ≤ n2 . Therefore, each irreducible representation of `∞ (B) of finite dimension
extends to an irreducible representation of Mn (`∞ (A)), that contains compact op-
erators in its image, thus is of dimension ≤ n2 .
But gMn (Aω )g can not have an irreducible representation of finite dimension,
because otherwise, f Mn (A)f has a non-zero quotient of finite dimension.
Thus, `∞ (B) has no quotient of dimension ≤ n2 , i.e., satisfies also condition
(ii) of Definition 2.0.4, i.e., B is pi(n).
(iv): If M(A) has no non-zero quotient of dimension ≤ n2 , then `∞ (M(A))
can not have a non-zero quotient C *-algebra of dimension ≤ n2 , because M(A) is
unitally contained in `∞ (M(A)).
The same happens with the (essential) ideal `∞ (A) of `∞ (M(A)) .
(v): like in [463] ??
Next considers relations between separable C *-subalgebras in Q(A) and in A∞
?

Lemma 2.12.5. Suppose that A is σ-unital. Then, for every separable C*-
subalgebra D ⊆ M(A), there exist a sequence a0 , a1 , . . . ∈ A+ of commuting con-
P
tractions such that an = 1 (strictly in M(A)) an am = 0 for |n − m| > 1,
P 1/2 1/2 P
T − n an T an ∈ A, and M T − T M ∈ A for M := n∈X an , for all T ∈ D
and all X ⊆ N.
 12. NON-SIMPLE PURELY INFINITE ALGEBRAS 305

??????????????
to be filled in
In particular, for each separable C*-subalgebra C ∈ Q(A), there exist hereditary
C*-subalgebras E1 , E2 ⊆ `∞ (A) and mono-morphisms ψj : Ej → M(A) such that
C ⊆ πA (ψ1 (E1 )) + πA (ψ2 (E2 )). ??????

Proof. ?? 

(???): If A is σ-unital, then, for each separable C *-subalgebra C ∈ Q(A), there

exist hereditary C *-subalgebras E1 , E2 ⊆ `∞ (A) and monomorphisms ψj : Ej →
M(A) such that C ⊆ πA (ψ1 (E1 )) + πA (ψ2 (E2 )). See 2.12.5.
If A is pi(n), then every hereditary C *-subalgebra E of `∞ (A) does not have
non-zero quotient algebras of finite dimension.
Suppose now that M(A) has an irreducible representation D : M(A) → L(H)
of finite dimension, then necessarily D(A) = {0} , i.e., D = d ◦ πA for an irreducible
representation d of Q(A). Let C ⊆ Q(A) a separable C *-subalgebra with d(C) =
L(H), and E1 , E2 as above.
?????????????
Then D◦ψ1 or D◦ψ2 is non-zero. Thus, there exists a hereditary C *-subalgebra
E of `∞ (A) that has a non-zero quotient of finite dimension. This contradicts the
existence
of
???????????????????????????
a non-zero hereditary C *-algebra E of `∞ (A) and a C *-morphism ψ : E →
M(A) with D(ψ(E)) 6= 0.
(???): to be filled in
?? 

Recall that a C *-algebra A Check definitions!! has property pi-n (or A is

pi-n) if, for each a ∈ A, the n-fold sum a ⊕ a ⊕ · · · ⊕ a = a ⊗ 1n is properly infinite
in Mn (A), cf. Definition ??.
There are other places where this result is discussed and partly proven. Com-
pare and unite them!

Lemma 2.12.6. The C*-algebra A is pi-n, if and only if, `∞ (A) is pi-n, if and
only if, Aω is pi-n.
If A is pi-n, then all non-zero quotients and non-zero hereditary C*-subalgebra
are pi-n.
If A is pi-n then A is traceless. (In particular, A contains no non-zero here-
ditary C*-subalgebra D that has a non-zero character. In particular, A admits no
irreducible representation of finite dimension.)
306 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

If A is pi-n, then A has property pi(m) for some m ≤ n.

If A is stable then A has property pi(n), if and only if, A is pi-n. More general,
this holds if the unit element of M(A) is properly infinite.
More general:
A has property pi-n if A has property pi(n) and for each element a ∈ A+ and ε > 0
there exists an n-homogenous element b ∈ A+ with the property that b is the closed
ideal generated by a and (a − ε)+ is contained in the closed ideal of A generated by
b.

In between, question for further studies:

If a, b ∈ A+ are residually properly infinite in A, is then a + b infinite in A?
Give Ref. to residually p.i. elements in A+ !!!
(What about counterexamples?
A can be supposed separable?
Where is the Definition of residually properly infinite elements in A+ ?)
If this is the case, then all pi(n) and pi-n are are equivalent to pi-1.!!! Why
that ???
If J ⊆ A is a closed ideal with property pi-n and suppose that A/J has pi-n.
It is then known that A has property pi-m for some m ≤ 2n. can it be improved
to m = n?

Proof. to be filled in ??
If A is stable, then its multiplier algebra M(A) contains isometries S, T with
∗
S T = 0. Suppose from now on that A has the property that M(A) contains
isometries S, T ∈ M(A) with S ∗ T = 0. The isometries Tk := T k S have mutually
orthogonal ranges: S ∗ (T ∗ )` T k S = δk,` 1 .
Let a, b ∈ Mn (A). Clearly a - b in Mn (A) if and only if a - b in Mn (M(A)).
????????? 

It is sometimes useful and easier, to consider local properties of the ultrapower

Aω of A, instead of local properties of the algebra A itself.
Let a ∈ A+ , t ∈ (0, ∞) and V : A → A an inner completely positive contraction
46
( ). We define µ(a, V, t) as the smallest m ∈ N of contractions d1 , . . . , dm ∈ A such
Pm Pm
that j=1 d∗j adj = (V (a) − t)+ . Here we do not require that k j=1 d∗j dj k ≤ 1.
Add remark/def concerning control of CPin (A) from notes (which
one ????). ??

Lemma 2.12.7. A C*-algebra A satisfies condition (i) of the Definition 2.0.4

of property pi(n) for some n ∈ N, if and only if, µ(t) := supa,V µ(a, V, t) < ∞ for

46 A c.p. map V is inner, if there exist d , . . . , d ∈ M(A) with V (a) := P d∗ ad .

1 m j j
 12. NON-SIMPLE PURELY INFINITE ALGEBRAS 307

each t ∈ (0, 1), where (a, V ) runs over all inner c.p. contractions V ∈ CP(A, A)
and over all positive contractions a ∈ A+ .

Proof. Suppose that n := µ(1/4) < ∞. Let 0 6= b ∈ A+ a contraction and

ε ∈ (0, 1/2) such that b ∈ span(AaA) =: Ja . We show that there are contractions
d1 , . . . , dn ∈ A and δ > 0 such that
n
X
(2/δ) d∗j adj ≥ (b − ε)+ .
j=1

It implies then condition (i) of Definition 2.0.4 by Lemma 2.1.9.

Let c := bα for 0 < α := − log(2)/ log(ε) < 1 , i.e., for α ∈ (0, 1) with εα = 1/2.
Then (c − 1/2)+ = (bα − 1/2)+ ≥ (1/2)(b − ε)+ and c ∈ Ja . Thus, there exists
e1 , . . . , em ∈ A and δ > 0 with
m
X
e∗k (a − δ)+ ek = (c − 1/4)+ .
k=1

It holds a ≥ δfδ (a) for the fδ (t) := min(2/δ(t − δ/2)+ , 1) for t ∈ [0, ∞). Define
V : A → A for x ∈ A by
m
1/2 1/2
X
V (x) := e∗k (a − δ)+ x(a − δ)+ ek .
k=1

Then kV k ≤ 3/4 and V (fδ (a)) = (c−1/4)+ . Thus (V (fδ (a))−1/4)+ = (c−1/2)+ .
It follows, that there are contractions d1 , . . . , dn ∈ A with n ≤ µ(1/4) and
n
X
d∗j fδ (a)dj = (c − 1/2)+ ≥ (1/2)(b − ε)+ .
j=1

Proposition 2.12.8. Let A denote a non-zero C*-algebra and Aω its ultra-

power with respect to some free ultrafilter ω ∈ γ(N) := β(N)\N. Then the following
properties of A are equivalent:

(i) A is weakly purely infinite (i.e., A has property pi(n) for some n ∈ N ).
(ii) Aω is locally p.i.
(iii) Every lower semi-continuous 2-quasi-trace of (Aω )+ is trivial, i.e., takes
only the values 0 and ∞.
(iv) There exists m ∈ N, such that a ⊗ 1m is a properly infinite element of
Mm (A) for every non-zero a ∈ A+ , i.e., A is pi-m.
(v) The ultrapower Aω has no finite-dimensional irreducible representation,
and, for every sequence of approximately inner completely positive con-
Q
tractions Vn : A → A, the ultrapower Vω := ω (Vn ) of (V1 , V2 , . . .) is an
ideal-system preserving c.p. map on Aω , i.e., Vω satisfies Vω (J) ⊆ J for
every closed ideal J of Aω .
(vi) Aω is pi-m for some m ∈ N.

Question: What happens if no closed hereditary C *-subalgebra of Aω has a

character?
308 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

The part (v) says equivalently that ?????

(v): The ultrapower Aω has no finite-dimensional irreducible representation,
and, for every sequence of approximately inner completely positive contractions
Q
Vn : A → A, the ultrapower Vω := ω (Vn ) of (V1 , V2 , . . .) is an ideal-system equi-
variant c.p. map on Aω , i.e., Vω satisfies Vω (J) ⊆ J for every closed ideal J of
Aω .
(vi): Aω is pi-m for some m ∈ N.
The new result says: If A is stable then pi(n) = pi-n and pi(n)
requires only
that span(AaA) can be approximated in n steps and that
A has no irreducible representations of dimension ≤ n.

Proof. (iv)⇒(i): By Lemma ??, A is pi(n) for some n ≤ m.

(i)⇒(ii): By Lemma ??, `∞ (A) and its quotient Aω are pi(n). By Lemma ??,
pi(n) implies local p.i.
(ii)⇒(iii): local p.i. implies traceless
(iii)⇒(iv): By Remark 2.8.3, Aω is traceless, if and only if,

(*) for every a ∈ (Aω )+ and ε > 0 there is n(a, ε) ∈ N such that (a − ε)+ ⊗
12n - a ⊗ 1n for all n ≥ n(a, ε).

For b, c ∈ Mk (A)+ holds b - c in Mk (A), if and only if, b - c in Mk (Aω ) =

(Mk (A))ω . Thus, for every a ∈ A+ and ε > 0 there is n ∈ N with (a − ε)+ ⊗ 12n -
a ⊗ 1n in M2n (A). We denote from now on by n(a, ε) the minimal n ∈ N with this
property.
Suppose that there exists ε > 0 and a sequence of contractions a1 , a2 , . . . ∈
A+ such that n(ak , ε/2) → ∞ if k → ∞. Then the positive contraction a :=
πω (a1 , a2 , . . .) ∈ Aω satisfies
check: ????????????
(a − ε/4)+ ⊗ 12n 6- a ⊗ 1n for all n ∈ N. The latter contradicts property (*) of
Aω .
Thus, there is “universal” m ∈ N with the property that (a−1/2)+ ⊗12n - a⊗1n
in M2n (A) for each contractions a ∈ A+ for some n = n(a, 1/2) ≤ m.
????????????????
(iv)⇒(v): Property (ii) passes to `∞ (A), to its quotient Aω , and to any non-
zero quotients of Aω . In particular, Aω cannot have any finite dimensional irre-
ducible representation.
If A satisfies (ii) and if a = (a1 , a2 , . . .) ∈ `∞ (A)+ is a representative of b ∈
(Aω )+ , then (Vω (b) − 2ε)+ = πω (c1 , c2 , . . .) with cn = (Vn (an ) − 2ε)+ .
Since an ⊗ 1m is properly infinite for some fixed m ∈ N, and since Vn (an ) is in
the ideal generated by an , we find, as in the last step of the proof of Proposition
 12. NON-SIMPLE PURELY INFINITE ALGEBRAS 309

2.15.11, dn,1 , . . . , dn,m ∈ A with k(dn,1 )∗ dn,1 + · · · + (dn,m )∗ dn,m k ≤ 2kaεn k , such
that cn = (dn,1 )∗ an dn,1 + · · · + (dn,m )∗ an dn,m . Thus (Vω (b) − 2ε)+ is in the ideal
generated by b.
(v)⇒(iv): An indirect argument shows that (v) implies the existence of (uni-
versal) m ∈ N, such that, for contractions a, b ∈ A+ with b in the ideal generated by
P ∗
a, there are d1 , . . . , dm ∈ A with dj adj = (b − 1/2)+ . Then 1/2 can be replaced
α
by any ε > 0 if we replace b by b with α := −2 log 2/ log ε.
Moreover, this implies that for every ε > 0 and b ∈ A+ in the closed ideal
generated by a ∈ A+ there is δ > 0 such that there are
??????????
If such m ∈ N exists, then
??????????
passes to Aω with the same number m.
It implies that a lower semi-continuous 2-quasi-trace τ on Aω annihilates the
closed ideal J(B) of Aω , which is generated by the image B := h(C0 ((0, 1], M2m ))
of a C *-morphism h from C0 ((0, 1], M2m ) into Aω , so far as τ |B+ is semi-finite.
Since Aω has no irreducible representation of finite dimension, the Pedersen ideal
of every closed ideal of Aω is contained in the union of the direct sums of ideals
J(B). Now we use again, that τ is lower semi-continuous, and get that τ can only
take the values 0 and ∞. 

It is obvious, that (iii), equivalently, means, that all singly generated ideals of
Aω are closed unions of ultrapowers of sequences of singly generated closed ideals
of A and that those ultrapowers are singly generated, and that Aω has no finite-
dimensional quotient.

Corollary 2.12.9. Let n ∈ N. Suppose that A satisfies condition (i) in the

Definition 2.0.4 of the property pi(n), i.e., for each a ∈ A+ , b ∈ J(a) := span(AaA)
and ε > 0 there exist d1 , . . . , dn ∈ A with kb − k d∗k adk k < ε.
P

This property passes to quotients A/J of A and to `∞ (A).

The following properties (i)–(v) of A are equivalent under condition (i) of Def-
inition 2.0.4:

(i) A is pi(n) .
(ii) There exists some free ultrafilter ω such that Aω has no irreducible repre-
sentation of dimension ≤ n.
(iii) `∞ (A) is pi(n) .
(iv) Aω is pi(n) for any free ultrafilter ω.
(v) `∞ (A) has no irreducible representation of dimension ≤ n.

A sufficient condition for Part (v) is that M(A) has no irreducible representa-
tion of dimension ≤ n.
And this condition is equivalent to Part (v) if A is σ-unital.
310 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Proof. (i) ⇔ (v): By Definition 2.0.4 A is pi(n) if and only if it satisfies

Condition (i) of Definition 2.0.4 and its Condition (ii), which says that `∞ (A) has
no irreducible representation of dimension ≤ n.
(v)⇒(iv): Each Aω is a quotient of `∞ (A).
(v)⇒(i): By Lemma 2.12.4(i), the quotient Aω of `∞ (A) has also property (i)
of Definition 2.0.4.
If ω ∈ β(N) \ N is given with the property that Aω has no non-zero irreducible
representation of dimension ≤ n, then Aω is pi(n) by Part (iii) of
?????????????, to be filled in ??
because, for each positive a ∈ Aω , there is a positive contraction e ∈ Aω with
ea = a, in particular, Ped(Aω ) = Aω .
By Proposition 2.12.8, there exists m ∈ N such that A is pi-m (for some m ≥ n).
Then `∞ (A) is pi-m by Lemma 2.12.6.
In particular, `∞ (A) can not have a finite-dimensional quotient.


Remark 2.12.10. If σ-unital A satisfies condition (i) of Definition 2.0.4, then

condition 2.0.4(ii) of `∞ (A)/c0 (A) is equivalent to:
M(A) has no irreducible representation of dimension ≤ n.
This happens, because – otherwise – a suitable hereditary C *-subalgebra D of
`∞ (A)/c0 (A) has a C *-morphism into Q(A) := M(A)/A with image that is not in
the kernel of that irreducible representation.
One can see:
If Aω has no irreducible representation of dimension ≤ n, then, for each δ > 0
there is m = m(δ) ∈ N such that, for every contraction a ∈ A+ , there are n + 1-
homogenous contractions e1 , . . . , em ∈ A+ and contractions d1 , . . . , dm ∈ A with
P ∗
k dk ek fk = (a − δ)+ .

Question: Is each element b ∈ Q(A)+ a sum of 4 commuting positive elements

that are in the closed ideal generated by the n + 1-homogenous elements?

The closed ideal Jn+1 of Q(A) generated by the n + 1-homogenous elements

of Q(A) is the intersection of all kernels of the C*-morphisms from Q(A) into Mk
with k < n + 2. The minimal requirement is therefore that Q(A) has no irreducible
representation into Mk for k ≤ n + 1. ...
The possible differences between properties pi(n) and pi-n can be seen by the
following Proposition 2.12.11 (see also Question ??). Here Ped(A) denotes the
minimal dense algebraic ideal of A, the “Pedersen-ideal”.

Proposition 2.12.11. Let A an inductive limit of algebras A1 , A2 , . . ..

(i) If each Ak is pi(n), then, for every non-zero a ∈ Ped(A), the algebra a∗ Aa
is pi(n).
 12. NON-SIMPLE PURELY INFINITE ALGEBRAS 311

(ii) If each Ak has property pi(n) and if M(A) has no non-zero finite-
dimensional quotient C*-algebra, then A is pi(n).
(iii) If each Ak has property pi-n, then A has property pi-n.
(iv) If each Ak is locally p.i., then A is locally p.i.

The next proposition lists some elementary implications of the properties Xpi
of C *-algebras.

Proposition 2.12.12. If A has Property pi-n, then A is pi(n) (thus is weakly

purely infinite).
If A is pi(n) (respectively has Property pi-n), then `∞ (A), Aω , every hereditary
C*-subalgebra D of A and every quotient of A are pi(n) (respectively have Property
pi-n).
This is somewhere below also shown?!
The C*-algebra A is purely infinite (in sense of Definition 1.2.1, i.e., A is
pi(1)), if and only if, A has property pi-1, i.e., each non-zero a ∈ A+ is properly
infinite.
If A is weakly purely infinite, then A is locally purely infinite.
Locally purely infinite algebras are traceless.

Proof. To be filled in ?? 

Proposition 2.12.13. We have the implications: A ∼= A ⊗ O∞ implies A s.p.i.

0
implies C ∩ Aω p.i. for all separable commutative subalgebras C ⊆ Aω
implies A pi(1) = A pi-1 = A p.i. = Aω p.i. implies A pi-n (“ has property
p.i.-n ”) = Aω is p.i.-n implies A ⊗ K p.i.-n2 and A is pi(n) (including absence of
quotients of Aω of dimension ≤ n2 ). = Aω pi(n) implies Aω has property p.i.-m
for some m ≥ n implies Aω “traceless” = Aω w.p.i. (:= pi(n) for some n ∈ N)
= Aω locally p.i.
= Aω satisfies: No non-trivial hereditary C*-subalgebra of Aω has a non-zero
character, and for each sequence of inner c.p. contractions Vn : A → A holds
Vω (I) ⊆ I for each closed ideal I of Aω .
= Aω satisfies:
No non-trivial hereditary C*-subalgebra of Aω has a non-zero character, and,
for each separable C*-subalgebra B of Aω and each commutative C*-subalgebra
C ⊆ B, there is a separable commutative C*-algebra D with C ⊆ D ⊆ Aω and
an (in Aω ) approximately inner c.p. contraction T : B → D with T |C = idC .
=
C 0 ∩ Aω is w.p.i. for every finitely generated Abelian C*-subalgebra C of Aω
=
{a}0 ∩ Aω is “traceless” for every positive a ∈ Aω .
312 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

implies
A l.p.i.
implies
A is “traces-less”
implies
A ⊗ B is locally p.i. for every non-elementary simple exact C*-algebra B.

Proposition 2.12.14. Every Xpi-algebra A is the inductive limit of its sep-

arable relative weakly injective C*-subalgebras B ⊆ A with property Xpi, and the
property that for each closed ideal I of B there exists a closed ideal J of A with
I = B ∩ J. Moreover, for every non-zero b ∈ A+ and separable C*-subalgebra
C ⊆ bAb of A with b ∈ C there exists a separable C*-subalgebra B ⊆ bAb with the
properties C ⊆ B, b ∈ B and c -B d for c, d ∈ B, if and only if, c -A d.
Here ”Xpi” stands for l.p.i., pi(n), pi-n, p.i., semi-s.p.i.( concerning commu-
tants of special??) or s.p.i.
and the W-vN-property?

Proof. We have to verify that all the definitions of pure infiniteness are “local”
in nature. It requires to generalize the properties as relative properties given by
certain countable families of semi-metrics on pairs of elements in C *-subalgebras.
The metrics are: to be filled in. See Appendix A or B? give Ref’s
??

Use uniform local-global m-almost-halving for n-p.i. permanences.

Use description of l.p.i. by morphisms h : C0 ((0, 1], K) → A. 

Give here references of the xpi-definitions !!!

Proposition 2.12.15. Permanences:

If A is weakly purely infinite and σ-unital then M(A) is weakly purely infinite.
(More precisely, M(A) is pi(2n) or pi(3n)???? if A is pi(n) and is σ-unital.)
Passage to hereditary subalgebras.
Stabilization (by K).
Quotients.
Extensions.
But do here not forget that
extensions of stable C*-algebras are
in general not (!) stable again.
`∞ (A), Aω , C 0 ∩ Aω for C ⊆ Aω commutative with character space of finite
dimension.
continuous fields
 13. SINGLY GENERATED IDEALS OF CORONAS 313

C(X)-algebras
C0 (X, A) w.p.i. for A w.p.i. and Dim(X) < ∞
tensor products (min?)

Proof. to be filled in ?? 

13. Singly generated ideals of coronas

Collect here all about ideals of M(B) and Q(B).

Also from Chps. 5, 6, 12!!! ?? The notation K for closed ideals of
M (B) is irritating ...
Important! Check ideal characterization carefully again!!!

Lemma 2.13.1. Let B a σ-unital stable C*-algebra, and b ∈ M(B)+ . Then

there exists a sequence (a1 , a2 , . . .) ∈ `∞ (B)+ such that
X
(1/2)k b + B + K k ≤ k ( sn an s∗n ) + B + K k ≤ k b + B + K k ,
n

for every norm-closed ideal K of M(B) ( 47 ). In particular,

X
I(b) + B = I( sn an s∗n ) + B
n
∗
P
for the norm-closed ideals I(b) and I( n sn an sn ) of M(B) generated by b respec-
tively by n sn an s∗n .
P

Proof. Recall that kb + B + Kk = kπB+K (b)k in M(B)/(B + K) ∼

=
Q(B)/πB (K) .
If we use a suitable, approximately with b commuting, commutative approx-
imate unit e1 , e2 , . . . ∈ B+ with en en+1 = en , then we can manage that for
g0 := 0, gn := (en − en−1 )1/2 and b0 :=
P P
n g2n bg2n , b1 = n g2n−1 bg2n−1 ,
P 2 P 2 P 2
holds gn = 1, b0 − ( n g2n )b ∈ B, b1 − ( n g2n−1 )b ∈ B, b0 + b1 − b ∈ B
and kbj + K + Bk ≤ kb + K + Bk for every closed ideal K of M(B). Now let
g2n b1/2 s2n , Y := g2n−1 b1/2 s2n−1 . Then XX ∗ = b0 , Y Y ∗ = b1 and
P P
X :=
Y X ∗ = 0. Let A := X ∗ X + Y ∗ Y . It follows,
kA + Kk = max(kX ∗ X + Kk, kY ∗ Y + Kk) = max(kb0 + Kk, kb1 + Kk)
and kb0 + b1 + Kk ≤ kb0 + Kk + kb1 + Kk ≤ 2kA + Kk for every closed ideal
K of M(B). In particular, (1/2)kb + K + Bk = kA + K + Bk ≤ kb + K + Bk,
because b + B = b0 + b1 + B. Thus, the sequence an := b1/2 gn2 b1/2 has the desired
properties. 

The following Proposition 2.13.2 has applications to ideal-equivariant extension

and lifting problems. The stable corona Q(B) of (general) σ-unital and stable B has
usually many ideals that are very different from the special ideals πB (M(B, J)), –
47 Here, the s , s , . . . ∈ M(B) are isometries with the property that P s (s )∗ converges
1 2 n n n
strictly to 1, cf. Remark 5.1.1(8).
314 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

even if B is stable and separable. Notice that the corona algebra Q(B) of the stable
(not w.p.i.) C *-algebra B := C0 ((0, 1] × [0, 1]∞ , K) does not have the property
derived in Proposition 2.13.2 for s.p.i. algebras.

Proposition 2.13.2. Suppose that B is a σ-unital and weakly purely infinite

C*-algebra. Let a, b ∈ Q(B) := M(B)/B. Then

ka + πB (M(B, J))k ≤ kb + πB (M(B, J))k , for every J ∈ I(B) ,

implies that the element a is in the closed ideal I(b) of Q(B) generated by b.

General ideals should be J, K, ..., because I(b) is reserved for the by b absorbed
ideal...

Proof. We may suppose that B is stable, because B ⊗ K is again weakly

purely infinite and σ-unital, and there is an isomorphism ϕ from Q(B) onto the
corner πQ⊗K (M(B) ⊗ p11 ) of Q(B ⊗ K), that satisfies

ϕ(πB (M(B, J))) = ϕ(Q(B)) ∩ πQ⊗K (M(B ⊗ K, J ⊗ K)) .

Now suppose that B is stable, σ-unital and is weakly p.i.

Why B ⊗ K is again weakly purely infinite? Answer: pi-n of B implies pi-n2
of B ⊗ K, see paper with Rørdam .
With pi(n) could be difficulties !!!.
Needs: If X ∈ M2 (B)+ is given. Then all entries of diag(X, . . . , X) ∈ M2n (B) can
be considered as properly infinite elements in M2 (Mn (B)), i.e., it needs some sort
of “almost” strong pure infiniteness!
Give Ref for next!: By Definition ?? and Proposition ?? there exists n ∈ N
such that B is pi-n in the sense of Definition ??, i.e., b ⊗ 1n is properly infinite (in
Mn (B)) for each b ∈ B, cf. Proposition 2.12.8(iv).
Since B is stable, M(B) contains a copy of On , i.e., there are isometries
T1 , . . . , Tn ∈ M(B) such that T1 T1∗ + · · · + Tn Tn∗ = 1. Let δn (b) := T1 bT1∗ +
· · · + Tn bTn∗ for b ∈ M(B). Then δn (J) = J ∩ δn (M(B)) and kδn (b) + Jk =
kb + Jk for every closed ideal J of M(B) and every b ∈ M(B). Moreover,
δn (b) ⊗ p11 = Z(b ⊗ 1n )Z ∗ ∼MvN b ⊗ 1n in Mn (M(B)), by the isometry Z =
T1 ⊗ p11 + · · · + Tn ⊗ p1n ∈ M(B) ⊗ Mn . Thus, δn (b) is properly infinite for every
0 6= b ∈ B.
Let s1 , s2 , . . . ∈ M(B) a sequence of isometries such that n sn (sn )∗ converges
P

strictly and unconditionally to 1 ∈ M(B), cf. Remark 5.1.1(8). This applies also to
the sequences of isometries tn := Tj(n) sk(n) and tn := sk(n) Tj(n) . Therefore U :=
∗ ∗
P
j,k Tj sk Tj sk converges strictly and unconditionally, cf. Remark 5.1.1(2). The sum
U is a unitary in M(B) by Lemma 5.1.2(i) and satisfies U k sk (δn (bk ))s∗k U ∗ =
P

sk bk s∗ ) for every sequence (b1 , b2 , . . .) ∈ M(B). In summary, we get kJ +

P
δn (
P k ∗ k P ∗
k sk bk sk k = kJ + k sk δn (bk )sk k for every (norm-)closed ideal J of B and every
sequence (b1 , b2 , . . .) ∈ `∞ (B).
 13. SINGLY GENERATED IDEALS OF CORONAS 315

By Lemma 2.13.1, for each Y ∈ M(B)+ there is a sequence (y1 , y2 , . . .) ∈

`∞ (B)+ such that and
X
(1/2) dist(Y, B + J) ≤ dist( sn yn s∗n , B + J) ≤ dist(Y, B + J)
n

for each closed ideal J of M(B).

It implies, that I( n sn yn s∗n ) + B = I(Y ) + B. The new sequence,
P

(δn (y1 ), δn (y2 ), . . .) ∈ `∞ (B) has the same properties as (y1 , y2 , . . .), but sat-
isfies in addition that f (δn (yn )) = δn (f (yn )) is properly infinite or zero for every
continuous function f : R+ → R+ with f (0) = 0.
We have seen above:
There are sequences (c1 , c2 , . . .), (d1 , d2 , . . .) ∈ `∞ (B)+ such that f (cn ) and f (dn )
are properly infinite or zero for each n ∈ N and every f ∈ C0 (0, 1], and such that,
for c := πB ( n sn cn s∗n ) and d := πB ( n sn dn s∗n ), holds I(c) = I(a), I(d) = I(b),
P P

kc + Kk ≤ ka + Kk and kb + Kk ≤ 2kd + Kk for all closed ideals K of Q(B).

∗
P
Let e := 2d, en := 2dn and E := n sn en sn . Notice that e = π(E) and
πB ( n sn an s∗n ) generate the same ideal of Q(B) if there is a surjective map λ : N →
P

N such that an = eλ(n) and λ−1 (k) is finite for all k ∈ N. Indeed: If σ is a
permutation N, then n sn aσ−1 (n) s∗n = n sσ(n) an (sσ(n) )∗ is unitarily equivalent
P P
∗ ∗
P P
to n sn an sn in M(B) by the unitary n sσ(n) sn , cf. Lemma 5.1.2(i). After
a suitable permutation (of indices) of the sequence (a1 , a2 , . . .) we may suppose
that there is a strictly increasing sequence 1 = m1 < m2 < · · · ∈ N such that
ek = amk = amk +1 = · · · = amk+1 −1 for k = 1, 2, . . ., i.e., that the sets λ−1 (k) are
now disjoint intervals with j < i for all j ∈ λ−1 (k) and i ∈ λ−1 (`) if k < `. Let
Pmk+1 −1 Pmk+1 −1
sj ek s∗j , R := k smk ek (smk )∗ and Pk := j=m sj s∗j . Then
P
Ak := j=m k k
∗
P P P
k Pk = 1, Ak ∈ B, A := n sn an sn = k Ak ≥ R, and there is an isometry
T ∈ M(B) with T T ∗ = k smk s∗mk and T ∗ AT = T ∗ RT = E := n sn en s∗n . Let
P P

δ > 0. Since (ek − δ)+ is zero or properly infinite, there is gk ∈ BPk such that
kgk k2 ≤ 2/δ and gk∗ ek gk = (Ak − δ)+ . The sequence k sk gk is strictly convergent
P

to an element G ∈ M(B) with G∗ EG = (A − δ)+ , cf. Remark 5.1.1(2).

Further notice, that kf (e) + πB (M(B, J))k = f (ke + πB (M(B, J))k)
for increasing continuous functions f : R+ → R+ with f (0) = 0, because
kf (e) + πB (M(B, J))k is the norm of the natural image of e ∈ Q(B) in Q(B/J).
Now we show that, if kc+πB (M(B, J))k ≤ ke+πB (M(B, J))k for all J ∈ I(B),
then C := n sn cn s∗n is contained in I(E) + B :
P

We have, for δ > 0,

k(c−δ)+ +πB (M(B, J))k = (kc+πB (M(B, J))k−δ)+ ≤ k(e−δ)+ +πB (M(B, J))k .

It follows, that (C − δ)+ = n sn (cn − δ)+ s∗n ∈ B + M(B, Jk ) for each k ∈ N,

P

where Jk denotes the closed ideal of B generated by {(en − δ)+ ; n = k, k + 1, . . . }.

Let n sn (cn −δ)+ s∗n = x+y with x ∈ B and y ∈ M(B, Jk ), then lim s∗n xsn = 0
P

implies that there exists m ∈ N (depending on k) such that k(cn − δ)+ − s∗n ysn k <
316 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

δ/2 for all n ≥ m(k). Thus, there exist contractions zn ∈ B with zn∗ s∗n ysn zn =
(cn − 2δ)+ . It follows that (cn − 2δ)+ ∈ Jk for all n ≥ m.
We define m(k) ∈ N inductively: Let m(0) = 1, and let m(k) denote the
smallest m ≥ max(k + 2, m(k − 1) + 2) with the property (cn − 2δ)+ ∈ Jk for all
n ≥ m.
For each n ≥ m(k), there are `(n, k) ≥ k and hn,k ∈ B such that
`(n,k)
X
(cn − 3δ)+ = h∗n,k sn sj (ej − δ)+ s∗j s∗n hn,k .

j=k

P` ∗
This `(n, k) and hn,k exists, because the sums j=k sj (ej − δ)+ sj are properly
infinite, generate Jk , and (cn − 3δ)+ is in the Pedersen ideal Ped(Jk ) of Jk . Now
P`(n,k)
let Yn := sn ( j=k sj ej s∗j )s∗n for n ≥ m1 , ψ(t) := (max(t − δ, 0)/t)1/2 , and Hn :=
ψ(Yn )hn,k s∗n . The operators Hn satisfy Hn∗ Yn Hn = sn (cn − 3δ)+ s∗n and kHn k2 ≤
∗ ∗
sn (s∗n ψ(Yn )hn,k )s∗n
P P P P
2/δ. The sums n Yn = n sn (sn Yn sn )sn and Hn =
converge strictly to elements Y ∈ M(B)+ , H ∈ M(B) and satisfy H ∗ Y H =
∗ ∗
P
n≥m1 sn (cn − 3δ)+ sn , cf. Remark 5.1.1(2). Thus, (C − 3δ)+ − H Y H ∈ B+ . The
construction of Y and H depends on δ > 0.
We are going to show that Y - A in M(B) for some suitable A = m sm am s∗m
P

(of the above considered type). It implies Y ∈ I(A) = I(E) and, finally, C ∈
I(E) + B.
Consider the subsets

Xk := {(n, j) ; mk ≤ n < mk+1 , k ≤ j ≤ `(n, k) } ⊆ N × N .

S
The sets Xk are pairwise disjoint and X := k Xk ⊆ N × N has the property, that
the map p2 : X 3 (n, j) 7→ j ∈ N is surjective with finite (p2 )−1 (n) for each n ∈ N.
The map p1 : X 3 (n, j) 7→ n ∈ N maps X onto N \ {1, . . . , m1 − 1}, and λ−1 1 (n)
is finite for each n ≥ m1 . We find a bijective map ξ : N → X from N onto X.
Now let λ := p1 ◦ ξ and consider the sequence (a1 , a2 , . . .) with am = eλ(m) , and let
A := m sm am s∗m Furthermore, consider the map (n, j) 7→ T(n,j) := sn sj . Then
P

X X X X
∗
Y = Yn = T(n,j) cj T(n,j) = Tξ(m) am Tξ(m) .
n≥m1 n≥m1 j∈p−1 (n) m≥1
1

Thus Y = Z ∗ AZ for the isometry Z := ∗

P
sn Tξ(m) . The latter sum converges
∗ ∗
P
strictly by Remark 5.1.1(2), because Tξ(m) Tξ(n) = δm,n 1 and m Tξ(m) Tξ(m) =
P∞ ∗
n,j=1 sn sj χn,j (sn sj ) for (χn,j ) ∈ `∞ (N × N) ⊆ `∞ (M(N )) the characteristic
function χ := χ(X) of X. 

Lemma 2.13.3. Suppose that S is a family of closed ideals of a C*-algebra A,

and that a, b ∈ A+ are positive contractions. The following properties (i) and (ii)
are equivalent.

(i) For every ε > 0 there exists δ = δ(ε, S) > 0 such that (a − ε)+ ∈ J for
every J ∈ S with (b − δ)+ ∈ J.
 13. SINGLY GENERATED IDEALS OF CORONAS 317

(ii) There exists a continuous strictly increasing function f : [0, 1] → [0, 2] with
f (0) = 0 such that kf (b) + Jk ≥ ka + Jk for all J ∈ S.

The property (i) is satisfied for all closed ideals J of A if the element a ∈ A+ is
contained in the closed ideal I(b) of A that is generated by b.

Proof. If a ∈ span(AbA), then, for every ε > 0, there are d1 , . . . , dn ∈ A and

δ := δ(ε) > 0, depending on a, b, ε, such that j d∗j (b − δ)+ dj = (a − ε)+ . It
P

implies that (a − ε)+ ∈ J if (b − δ)+ ∈ J.

Now we can proof the equivalence of (i) and (ii):
(ii)⇒(i): Let ε > 0 and

δ := δ(ε, S) := f −1 (min(ε, f (1))) ,

then (f (b)−ε)+ ≈ (b−δ)+ . In particular, (b−δ)+ ∈ J, if and only if, (f (b)−ε)+ ∈ J,

if and only if, ε ≥ kπJ (f (b)k = kf (b) + Jk .
Since kf (b) + Jk ≥ ka + Jk, we get ε ≥ kπJ (a)k, i.e., (a − ε)+ ∈ J.
(i)⇒(ii): Let ε ∈ (0, 1], consider the non-empty set X(ε) ⊆ (0, 1] of δ ∈ (0, 1]
with the property that J ∈ S and (b − δ)+ ∈ J together imply (a − ε)+ ∈ J.
Clearly, X(ε) is an interval (0, ξ(ε)) or (0, ξ(ε)], and X(ε) ⊆ X(ε1 ) if ε ≤ ε1 .
Define g(ε) := sup X(ε) = ξ(ε). Then g : (0, 1] → (0, 1] is a positive and increasing
function with g(1) = 1 and with following property:
If g(ε) > kb + Jk and J ∈ S, then ε ≥ ka + Jk. (Because t ≥ kb + Jk is equivalent
to (b − t)+ ∈ J.)
Since g is increasing and positive, there exists a strictly increasing continuous
function h : [0, 1] → [0, 1] with h(0) = 0 and h(ε) < g(ε) for all ε ∈ (0, 1]. If
h(1) < 1, then we extend h to [0, 2 − h(1)] by letting h(t) := h(1) + (t − 1) for
t ∈ (1, 2 − h(1)], and get h([0, 2 − h(1)]) = [0, 1]. Now define f : [0, 1] → [0, 2] by
f := h−1 .
Then, ε ≥ kf (b) + Jk = f (kb + Jk), if and only if, h(ε) ≥ kb + Jk, which
implies g(ε) > kb + Jk. It follows that ε ≥ kf (b) + Jk and J ∈ S together imply
ε ≥ ka + Jk. In particular, there is no ideal J ∈ S with ka + Jk > kf (b) + Jk. 

Corollary 2.13.4. Suppose that B is a σ-unital, stable and weakly purely

infinite C*-algebra, and that b ∈ Q(B)+ .
Let δ > 0, and let J (b, δ) denote the family of ideals πB (M(B, J)) with (b −
δ)+ ∈ πB (M(B, J)). Define I(b, δ) as the intersection of all ideals in J (b, δ).
Then the closed ideal of Q(B) generated by b is the same as the closure of the
S
union δ>0 I(b, δ).
Especially all countably generated ideals of Q(B) are of this type.
If, in addition, B is simple, then Q(B) is simple and purely infinite.

Notice that the particular final conclusion says:

Q(B) is simple and p.i. if B is simple stable and p.i.
318 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

The more difficult opposite direction, namely that simplicity of Q(B) for stable
σ-unital B implies that B is simple and p.i., is shown in
reference or cite ??????

Proof. We combine Proposition 2.13.2 and Lemma 2.13.3, where the system
of all ideals πB (M(B, J)) plays the role of S.
All countably generated ideals are singly generated.
If, in addition, B is simple (and is weakly p.i. by assumptions !), then Q(B) is
simple, because every singly generated ideal coincides with Q(B), cf. ?????
By Lemma 2.12.4(vi), Q(B) is again weakly purely infinite. The simplicity and
the weak pure infiniteness of Q(B) implies the strong pure infiniteness of Q(B) by
Proposition 2.2.1. 

14. Point-wise approximately inner maps

The technicalities of our proves have to do with the different infiniteness proper-
ties of the prime quotients of the considered C *-algebras, or the properties of their
multiplier algebras and corona algebras. Only in very comfortable situations exists
a sort of local triviality with the respect to prime quotients that have reasonable
infiniteness, factorial or prime properties. We try to describe such problems.
Definition 2.14.1. Let X a locally compact space. We call a c.p. map
V : Cb (X, B) → Cb (X, B) point-wise approximately inner, if there is point-
norm continuous map x ∈ X 7→ Vx from X into the approximately inner c.p. maps
on C *-algebras B such that V (f )(x) = Vx (f (x)) for all x ∈ X and f ∈ Cb (X, B).

Notice here that the positive map V is defined on all f ∈ Cb (X, B), if and only
if, there is a general bound given by supx∈X kVx k < ∞ on the “fibres” Vx . (We
start here with some very elementary stuff!)
Definition 2.14.2. A c.p. map V : A → A is called ideal system preserv-
ing (or I(A)-equivariant) if V (J) ⊆ J for all closed ideals J of A.

One of the critical points is: Let a, b ∈ A+ contractions with ab = a and

suppose that a is infinite (in some suitable sense). Is b infinite? Is it the case if
(a − δ)+ is infinite for each δ ∈ [0, kak)?
( Notice that ab = a for a, b ∈ A+ implies that a = (c − 1)+ for c := (b − (b −
1)+ ) + a. And then a(c − a) = a. But b can not been reconstructed from c and a,
because (b − 1)+ is then lost be this construction of c.)
What means ”infinite” here? One definition is: There exists non-zero 0 ≤ eδ ∈
(a − δ)+ A+ (a − δ)+ with
eδ ⊕ (a − δ/2)+ - (a − δ/2)+ .

It depends here from the answer to the question if (a−δ/2)+ ⊕(b−(a−δ/2)+ ) -

b . This don’t happen if A is commutative!
 14. POINT-WISE APPROXIMATELY INNER MAPS 319

But something like (a − δ/2)+ ⊕ (b − (a − δ)+ ) - b in M2 (C0 (a, b)) ??? and
(a − δ/2)+ + (b − (a − δ)+ ) ≥?????

Theorem 2.14.3. Let X is a σ-compact l.c. space, and suppose that the C*-
algebra C0 (X, B) is weakly p.i.
Then each point-wise approximately inner c.p. map

V : Cb (X, B) → Cb (X, B)

in sense of Definition 2.14.1 is ideal system preserving (in sense of Definition

2.14.2) completely positive map.

Proof. It is not difficult to see that positivity preserving linear maps T : C →

D from C *-algebras C into C *-algebras D that are defined on all elements of
C *-algebras C are always norm-continuous.
Thus, our map V is bounded and supx∈X kVx k ≤ kV k < ∞. The assumption of
point-norm continuity of x 7→ Vx and the obvious estimate kVx k ≤ kV k implies that
the map x 7→ Vx (b(x)) ∈ B is continuous on X for each continuous map b : X → B.
Since X is σ-compact, there is a function γ ∈ C0 (X) with 0 < γ(x) ≤ 1 for all
x ∈ X and maxx∈X γ(x) = 1.
◦
S
We define compact sets Yn ⊆ X with the properties Yn ⊆ Yn+1 and X = n Yn
and functions αn ∈ C0 (X) by

αn := 2n+1 (γ − 2−(n+1) )+ − 2n (γ − 2−n )+

and Yn := αn−1 (1). Notice αn+1 αn = αn and that the support of αn is contained
in Yn .
Then β0 := 0 and βn := αn − αn−1 for n ∈ N satisfy βn+k βn = 0 for k > 1
P
and n βn = 1 on X. The support of βn is contained Yn+1 .

By assumption, there is k ∈ N such that C0 (X, B) is pi(k). This passes to the

quotients C(Y, B) of C0 (X, B) for compact subsets Y ⊆ X. Let a ∈ Cb (X, B)+ ,
ε > 0 and y ∈ Y ⊆ X. There exists n(y) ∈ N, b1 , . . . , bn ∈ B such that
X
kVy (a(y)) − b∗j a(y)bj k < ε ,

because Vy : B → B is approximately inner. The continuity of the function

X
x 7→ kVx (a(x)) − b∗j a(x)bj k

implies that there is an open neighborhood U (y) of y such that

X
kVx (a(x)) − b∗j a(x)bj k < ε ∀ x ∈ U (y) .

By compactness of Y , thus implies the existence of finitely many U1 , . . . , Um and

(i)
n(i) ∈ N, bj ∈ B, i = 1, . . . , m, j = 1, . . . , n(m) such that the Ui cover Y and
P (i) (i)
kVx (a(x)) − j (bj )∗ a(x)bj k < ε for x ∈ Ui . If we take a partition of unit
320 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

P
f1 , . . . , fm ∈ C(Y )+ with supports of fi in Ui and i fi = 1, and let c` ∈ C(Y, B)
1/2 (i)
for ` = 1, . . . , n(1) + · · · + n(m) denote one and only one of the fi · bj , then
X
kVx (a(x)) − c` (x)∗ a(x)c` (x)k < ε ∀ x ∈ Y .
`

This shows that the restriction of x 7→ Vx (a(x)) to Y is in the closed ideal generated
by a|Y , for each a ∈ Cb (X). It follows that for δ > 0 and compact Y ⊆ X there
exists d1 , . . . , dk with
k
X
kVx ((a(x) − δ)+ ) − dj (x)∗ (a(x) − δ)+ dj (x)k < δ for x ∈ Y .
j=1

(n) (n)
We get for each of the above constructed Yn elements d1 , . . . , dk ∈ C(Yn , B)
(n)
such that the last inequality holds for x ∈ Yn with dj replaced by dj . We define
elements ej , fj ∈ Cb (X, B) for j = 1, . . . , k by
∞
1/2 (2n)
X
ej (x) := β2n−1 (x)1/2 (a(x) − δ)+ dj ,
n=1
∞
1/2 (2n+1)
X
fj (x) := β2n (x)1/2 (a(x) − δ)+ dj .
n=1
Pk Pk
Then k j=1 e∗j ej k ≤ kakkV k + δ, k j=1 fj∗ fj k ≤ kakkV k + δ, gδ (a)ej = ej , and
gδ (a)fj = fj for j = 1, . . . , k, where gδ (t) := min(δ −1 t, 1). Since the functions β2n
P
(respectively the β2n−1 ) have mutually orthogonal supports and βn = 1, one can
see that
X k
kV ((a − δ)+ ) − e∗j gδ (a)ej + fj∗ gδ (a)fj k ≤ δ .
j=1
−1
Since gδ (a) ≤ δ a, the element (V ((a − δ)+ ) − 2δ)+ is in the closed ideal I(a)
of B that is generated by a. The δ > 0 can be taken arbitrarily small with ej , fj
(j = 1, . . . , k) depending on δ. Hence, V (a) ∈ I(a). 

The Theorem 2.14.3 is needed, because even the answer to following questions
are still open.

Questions 2.14.4. Is C([0, 1]∞ , A) weakly purely infinite if A is purely infinite?

A positive answer would follow from a positive answer of the question: Is
C([0, 1], A) purely infinite if A is purely infinite? (In fact a positive answer would
imply that C0 (X) ⊗ A is purely infinite if A is purely infinite and X is a locally
compact Hausdorff space.)
To understand the non-triviality of this question recall here:
A is purely infinite ⇔ A is pi(1) ⇔ A is 1-purely infinite (i.e., A has property pi-1).
We know only that C([0, 1], A) has property pi-2 if A is p.i. So the zoological
garden of really different pi-n or pi(n) could be bigger than we hope for. What is
the secret behind such possible differences?
 14. POINT-WISE APPROXIMATELY INNER MAPS 321

Corollary 2.14.5. Suppose that X is a σ-compact, non-compact, locally com-

pact Hausdorff space, and B is a C*-algebra. Let Y ⊆ β(X) \ X a closed subset of
the corona of X, A ⊆ Cb (X, B)|Y a separable C*-subalgebra of Cb (X, B)|Y and let
C ⊆ A a commutative C*-subalgebra of A.
If the C*-algebra C0 (X, B) is weakly p.i., then there exists a continuous map

V : X 3 x → Vx ∈ CP(B)

from X into the approximately inner c.p. contractions CPin (B), such that the cor-
responding c.p. map VY := V |Y : Cb (X, B)|Y → Cb (X, B)|Y maps A into a com-
mutative C*-subalgebra C ∗ ((VY (A)) of Cb (X, B)|Y , fixes the elements of C, and
has the property, that VY (b) is in the closed ideal of Cb (X, B)|Y that is generated
by b for each positive b ∈ Cb (X, B) |Y .

Proof. to be filled in, use w.p.i. of ???? ?? 

Recall that Sin (B) is the set of all inner c.p. maps V defined by V (b) :=
Pn ∗
k=1 dk bdk for some n := n(V ) ∈ N and some finite sequence d1 , . . . , dn ∈ B. The
elements of Sin (B) is an algebraic matrix operator-convex cone. Its point-norm
closure will be denoted by CPin (B).

Corollary 2.14.6. Let X be a non-compact locally compact σ-compact Haus-

dorff space, and let ∅ 6= Y ⊆ (βX) \ X denote a (non-empty) closed subset of the
corona of X.
Suppose that, for every point-norm continuous family x ∈ X 7→ Vx ∈ CPin (B)
with k Vx k ≤ 1, the completely positive contractions

V |Y : Cb (X, B) |Y → Cb (X, B) |Y

are ideal-system equivariant, i.e., V |Y (J) ⊆ J, for each closed ideal of J of

Cb (X, B) |Y .
Then B has the following property:
There exists ω ∈ β(R+ ) \ R+ such that, for every continuous path t ∈ R+ 7→
a(t) ∈ B+ with ka(t)k ≤ 1 and for every ε > 0, there exists an open subset U =
U (ε) ⊆ R+ , in the ultrafilter of O(R+ ) defined by ω, with the property that the
values µ(t) := µ(a(t), Sin (B), ε) remain bounded on U (i.e., there is n(U ) ∈ N such
that µ(t) ≤ n(U ) for all t ∈ U ).
In particular, there exists a sequence n < tn < tn+1 such that supn µ(tn ) < ∞.
S
If, in addition, X is “disconnected at ∞”, i.e., X = Un for compact open
subsets U1 ⊆ U2 ⊆ · · · ⊆ X, then B has the stronger property (i) of Definition 2.0.4
for some n ∈ N.

Is the existence of the sequence (tn )n equivalent to Property (i) of Definition

2.0.4 for pi(n) in ???? for some n ∈ N?
322 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Proof. Recall that the notation a |Y for a ∈ Cb (X, B) refers to the fact that
Cb (X, B) is a Cb (X)-algebra, i.e., Cb (X) is unitally contained in Cb (X, M(B)) ⊆
M(Cb (X, B)), and that C(βX) ∼ = Cb (X) (both in a natural way).
By Lemma ??, it sufficed to show that the convex set Sin (B) ⊆ CPin (B) of
inner c.p. contraction is controlled in the sense of Definition ??, because this implies
the existence of n ∈ N such that B satisfies the property (i) of Definition 2.0.4.
Suppose that Sin is not controlled.
Then there is ε > 0 and a sequence of contractions a1 , a2 , · · · ∈ B+ such that
Pm
µ(an , Sin (B), 3ε) > 2n , i.e., there are Vn ∈ Sin such that the equation j=1 d∗j adj =
(Vn (an ) − 3ε)+ with contraction d1 , . . . , dm ∈ B implies m ≥ 2n .
??
Change text and proof of cor. 2.14.6 !!!
Reduction to case: B separable, X = R+ , Y = {ω}, ω ∈ γ(R+ ) .
???
?? proof has to be revised: Consider the case X = [1, ∞).
Need following stronger result:
There exist ε ∈ (0, 1/4), a continuous path t ∈ [1, ∞) 7→ a(t) ∈ B+ with
ka(t)k ≤ 1, and a point-norm continuous path t ∈ [1, ∞) 7→ St ∈ Sin (B), such that
m
X
St (a(t)) − 3ε ≤ d∗j adj ,
j=1

with contractions d1 , . . . , dm ∈ B, implies m ≥ t (for each fixed t ∈ [1, ∞)).

Remark: If B is separable and all non-zero σ-unital hereditary ????
Following seems NOT to work:
We define continuous gn (t) on [1, ∞) by gn (t) := 0 for t ∈ [1, n] and t ≥ n + 2,
gn (n+1) := 1, and gn linear on [n, n+1] and on [n+1, n+2]. Then we define a(t) :=
P∞ P∞
n=1 gn (t)an ∈ B+ and St (a) := n=1 gn (t)Sn (a) for t ∈ [1, ∞). The function
a(t) defines a positive contraction a ∈ Cb ([1, ∞), B)+ , the map t ∈ [1, ∞) 7→ St (a)
is continuous for each a ∈ B and kSt k ≤ 1. Therefore, S(f )(t) := St (f (t)) for
f ∈ Cb ([1, ∞), B) defines a positive contraction S ∈ CP(Cb ([1, ∞), B)).
We proceed with better definition of a(x) and Vx ,
given further above. (or below?)
End: suggestion for revision.
Since X is σ-compact and locally compact, there is a continuous function
h : X → (0, 1] such that h−1 [δ, 1] is compact for each δ ∈ (0, 1]. We let Vx (a) :=
St (a) and b(x) := a(t) for t := h(x)−1 .
By definition of V |Y , the equation (V |Y )(b |Y ) = c |Y holds for the map
x ∈ X 7→ c(x) := Vx (b(x)) in Cb (X, B)+ with kck ≤ 1. Notice that ((c |Y ) − ε)+ =
(c − ε)+ |Y , because a ∈ Cb (X, B) 7→ a |Y ∈ Cb (X, B) |Y is a *-epimorphism.
 15. MORE ON PI(n)-ALGEBRAS 323

By our assumption, (c |Y ) is in the ideal generated by b |Y .

It follows, that there are m ∈ N and contractions e1 , . . . , em ∈ Cb (X, B) |Y ,
such that j e∗j (b |Y )ej = (c−ε)+ |Y . Let fj ∈ Cb (X, B) contractions with fj |Y =
P

ej , then
 
Xm
r(x) :=  fj (x)∗ b(x)fj (x) − (Vx (b(x)) − ε)+
j=1

defines a selfadjoint element r ∈ Cb (X, B) with r |Y = 0.This means for the

bounded continuous non-negative function λ ∈ Cb (X) ∼ = C(βX) with λ(x) :=
kr(x)k for x ∈ X that λ |Y = 0.
The subset Z := {x ∈ βX ; λ(x) < ε} is an open subset of βX with Z ⊃ Y . In
particular, Z 6= ∅. Since X is a dense open subset of βX, it follows that U := X ∩ Z
is a non-empty open subset of X. The set U can not be contained in a compact
subset K ⊆ X, because, otherwise, λ(x) ≥ ε for all x ∈ (βX) \ X, which is
impossible by ∅ =6 Y ⊆ (βX) \ X and λ |Y = 0.
Thus, inf{ h(x) ; x ∈ U } = 0, and there exists a sequence x1 , x2 , . . . ∈ U with
lim h(xn ) = 0, such that λ(xn ) = kr(xn )k < ε .
Let x ∈ U with t := h(x)−1 > 2m. Then a(t) = b(x) and, with dj := fj (x),
m
X
St (a(t)) − 2ε = Vx (b(x)) − 2ε ≤ d∗j a(t)dj .
j=1

The latter inequality contradicts the property m ≥ t for a(t), St and contractions
d1 , . . . , dn ∈ B. 

15. More on pi(n)-algebras

Recall here for the following Lemma 2.15.1, that an element ... ????

Lemma 2.15.1. Let A a C*-algebra, D ⊆ A a hereditary C*-subalgebra of A,

λ : A → L(H) an irreducible representation such that λ(D) 6= {0} and such that
λ(A) does not contain a non-zero operator of finite rank.
Let x ∈ H with kxk = 1 and λ(D)x 6= 0, and denote by Q the orthogonal
projection from H onto λ(D)x.
Then, for each n ∈ N , there exists an n-homogenous positive contraction an ∈
D+ with 2kλ(an )xk ≥ kQxk ≥ kλ(an )xk.

Proof. The restriction of λ to D and to λ(D)x ⊆ H is an irreducible repre-

sentation of D that does not contain any non-zero compact operators in its image,
because otherwise it would contain also non-zero operators of finite rank in its image
λ(D) ⊆ L(H).
Since λ(D)x is a dense vector space in QH, we find for each given n ∈ N a linear
subspace Ln ⊆ Ln+1 of λ(D)x of dimension = n such that dist(Qx, Ln ) ≤ kQxk/2.
324 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Let Pn denote the orthogonal projection onto Ln . The variant of the Kadison
transitivity theorem given in Lemma 2.1.15 gives that there is a closed projection
qn ∈ D∗∗ of rank = n such that the normal surjective extension λ|D : D∗∗ → L(QH)
of the restriction λ|D : D → L(QH) defines an isomorphism Ψn from qn D∗∗ qn onto
L(Pn H) with Ψn (Pn dPn ) = qn D(d)qn for d ∈ D. In particular Mn ∼ = qn D∗∗ qn
∗∗
for some isomorphism γ of Mn with qn D qn . By Lemma 2.1.15 there exists a
C *-morphism h from Mn ⊗ C0 (0, 1] into D with qn h(Mn ⊗ C0 (0, 1))qn = 0 and
qn h(T ⊗f0 )qn = γ(T ) for T ∈ Mn . In particular a := h(1n ⊗f0 ) is an n-homogenous
element in D+ with λ(a) = Ψn (γ(1n )) = Pn .
Check it again! :
We get λ(an )x = Ψn (qn )x = Pn x, Ln = Pn H = Pn QH, and Qx = Pn x +
(Q − Pn )x is an orthogonal decomposition for Qx. It follows dist(Qx, Ln ) = kQx −
Pn xk ≤ kQxk/2 by construction of Ln = Pn H.
The orthogonal decomposition Qx = Pn x + (Q − Pn )x leads to

kQxk2 = kPn xk2 + kQx − Pn xk2 ≤ kPn xk2 + (kQxk/2)2

and gives that kQxk2 (3/4) ≤ kPn xk2 . It implies that kQxk ≤ 2kPn xk. 

Proposition 2.15.2. If A is a pi(n) C*-algebra then, for each non-zero here-

ditary C*-subalgebra D ⊆ A and each pure state λ on A with λ(D) 6= {0}, there
exists a hereditary C*-subalgebra E ⊆ D with λ(E) 6= {0} such that all non-zero
positive elements b ∈ JE in the closed ideal JE := span(AEA) of A generated by
E have the property that b ⊗ 1n is properly infinite in JE ⊗ Mn .
Moreover there exist hereditary C*-subalgebras E ⊆ D with the above properties
that satisfy moreover that the ideal JE ⊗ K of A ⊗ K generated by E has property
pi-n.
The primitive ideal space Prim(A) ∼ = Prim(A ⊗ K) has a base of its topology
that is generated the open subsets that correspond to closed ideals J of A such that
J ⊗ K has property pi-n.
Maximal open subsets with this property have the property that ... ????

Idea of proof: Take a maximal family of pairwise orthogonal n2 -homogenous

contraction in A+ . The ideal generated by them has this property.
Recall here that the positive integer n for pi(n)-algebras or for the C *-algebras
with property pi-n is only an upper bound. We do not know if there exists for
each n ∈ N a pi(n)-algebra that is not a pi(n − 1)-algebra. And it is only for pi(1)-
algebras (i.e., for the ordinary “purely infinite” C *-algebras) known that this class
of algebras remains invariant under extensions.
Where it is shown that extensions of pi(1) algebras gives pi(1) algebras. Give
cite or ref !!!
But we know only that C([0, 1], A) is pi-2 if A is p.i. ( = pi-1). This, and the
above Proposition 2.15.2 indicates that some sort of “decomposition dimension”
 15. MORE ON PI(n)-ALGEBRAS 325

of the saturated quasi-compact subsets of Prim(A) plays a role (it means not the
covering dimension which is only for Hausdorff spaces the same as the decomposition
dimension).
Corollary 2.15.3. The following properties (i), (ii), (iii) and (iv) are equiv-
alent for a (non-zero) C*-algebra A.

(i) A is purely infinite in sense of Definition 1.2.1 – which is equal to the

property pi(1) in Definition 2.0.4 –, i.e., A has no character, and, for
each a, b ∈ A+ , with b in the closed ideal of A generated by a, and given
ε > 0 there exist d ∈ A with d∗ ad = (b − ε)+ .
(ii) A has property pi-1 as defined in Definition ??, i.e., each non-zero element
of A+ is properly infinite, i.e., for each nonzero a ∈ A+ and δ > 0 there
exist b, c ∈ A with kb∗ ack < δ, kb∗ ab − ak < ε and kc∗ ac − ak < ε.
(iii) For each a ∈ A+ and ε > 0 there exist d, e ∈ aAa with e∗ d = 0 and
d∗ d = e∗ e = (a − ε)+ .
(iv) For each a ∈ A+ and each closed ideal J of A with a 6∈ J the element
πJ (a) ∈ A/J is infinite in A/J, i.e., there exists non-zero b ∈ (A/J)+
with πJ (a) ⊕ b - πJ (a).

Refer below to general pi(n) and - rules, where it is possible?!

The part (iv) seems to be the suitable definition of for all non-simple C*-algebras.

Proof. Notice that the definition of -A for C *-algebras A is a local property

in the sense that that the relation a -A b holds for a, b ∈ A+ , if and only if, a -D b
in the hereditary C *-subalgebra D := (a + b)A(a + b) of A.
(i)⇒(ii): If a ∈ A+ is a non-zero 2-homogenous element then each element in
the hereditary C *-subalgebra D := aAa is properly infinite in D.
Since a is 2-homogenous there exits ????
Then D ∼= M2 (E), where E is a hereditary C *-subalgebra of A that has the
property that all elements ???????
Indeed, by definition of 2-homogenous elements there is C *-morphism ψ : M2 ⊗
1/3
C0 (0, 1] → D with ψ(1 ⊗ f0 ) = a. The element b := ψ(p11 ⊗ f0 ) has the property
1/3
that b ≤ a1/3 and and d∗1 bd1 + d∗2 bd2 = a for d1 := ψ(p11 ⊗ f0 ) ∈ A and
1/3
d2 := ψ(p12 ⊗ f0 ) ∈ A. By Definition 1.2.1 of purely infinite algebras we get that
there exists a sequence cn ∈ A with a = lim c∗n bcn .
We define E := bAb. If e ∈ E+ then e ⊕ e - e, because kgn egn − 12 ⊗ ek → 0 in
1/n 1/n
M2 ⊗ E for n → ∞, where gn := ψ(p11 ⊗ f0 )h1,n + ψ(p12 ⊗ f0 )h2,n with ??????

It follows that E is a non-zero hereditary C *-subalgebra of D ⊆ A with the

property that each element of E is properly infinite in A (an thus also in D).
Let D ⊆ A a hereditary C *-subalgebra such that every non-zero elements
d ∈ D+ is properly infinite in A. Then each d ∈ D+ is also properly infinite in D
by Remark ??.
326 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

We show that each c ∈ J+ := span ADA+ is properly infinite. Indeed, for

each ε > 0 there exist d1 , . . . , dn ∈ D+ and f1 , . . . , fn ∈ A with (c − ε/2)+ =
Pn ∗ ∗
k=1 fk dk fk . It follows that there is g ∈ A with g dg = (c − ε)+ for d :=
1/2 ∗ 1/2
d1 + · · · + dn . Thus, the element (c − ε)+ satisfies d gg d ∼M vN (c − ε)2+ , and,
hence, is Murray–von-Neumann equivalent to a properly infinite element of D+ . It
follows that c is properly infinite in A by Lemma ??.
If a ∈ A+ is the limit of elements an ∈ A+ , and there exists γ > 0 such
that for each δ ∈ (0, γ) and n ∈ N the elements (an − δ)+ are properly infinite
in A+ , then a is properly infinite in A : Given ε ∈ (0, γ) then there exist n0 ∈ N
with kan − ak < ε/3 for all n ≥ n0 . By Lemma 2.1.9 there exist contractions
dn , en ∈ A with d∗n adn = (an − ε/3)+ and e∗n (an − 2ε/3)+ en = (a − ε)+ . Since
(an − 2ε/3)+ ⊕ (an − 2ε/3)+ - (an − ε/3) it follows that (a − ε)+ ⊕ (a − ε)+ - a.
It implies that a is properly infinite in aAa.
Thus, if Jτ is an upward directed family of closed ideals of A with the property
S
that all non-zero positive elements of τ Jτ are properly infinite, then all non-zero
S
positive elements in the closed ideal J := τ Jτ are properly infinite.
It follows that A contains a maximal ideal J with the property that all nonzero
positive elements a ∈ J+ are properly infinite.
The maximality implies that J is closed in A.
Suppose that J 6= A. Since A has no characters, the quotient A/J can not
have a character. Thus, there exists a 2-homogenous element b = ϕ(12 ⊗ f0 ) for
some non-zero C *-morphism ϕ : M2 ⊗ C0 (0, 1] → A/J. Since M2 ⊗ C0 (0, 1] is a
projective C *-algebra, there exists a C *-morphism ψ : M2 ⊗ C0 (0, 1] → A with
πJ ψ(12 ⊗ f0 ) = b. The element a := ψ(12 ⊗ f0 ) is a 2-homogenous element in A
that is not in J.
Let h : A → L(H) an irreducible representation of A, ?????
(ii)⇒(iii):
(iii)⇒(i):
(ii)⇒(iv): The property (ii) says a ⊕ a - a for each a ∈ A. If J is a closed
ideal of A with a 6∈ J, then this implies that πJ (a) ⊕ πJ (a) - πJ (a)
??????
(iv)⇒(???): The part (iv) says that for each a ∈ A+ and closed ideal J of A
with πJ (a) 6= 0 there exists non-zero b ∈ (A/J)+ with πJ (a) ⊕ b - πJ (a).
This implies that a is properly infinite in A, i.e., a ⊕ a - a in A by Lemma ??
???
Let J := {b ∈ A; a ⊕ b - a}. The set J is by an observation of J. Cuntz [?] a
closed ideal of A with the property that πJ (a) ”finite” in A/J, i.e. the only element
c ∈ A/J with with πJ (a) ⊕ c - πJ (a) is

 15. MORE ON PI(n)-ALGEBRAS 327

Compare next Rem. with Def. 1.2.1 or Def. of pi-1 or pi in 1st

Sect. in Chp. 2!!

Remark 2.15.4. We recall that a (not necessarily simple) C *-algebra A is

purely infinite in the sense of Definition 1.2.1, that is equal to property pi(1) in
Definition 2.0.4, if

(i) A is non-zero and has no character, i.e., every irreducible representation

has dimension > 1, and
(ii) for every positive a in A, every positive b in the ideal generated by a, and
every ε > 0 there exists d ∈ A such that kb − d∗ adk < ε.

We write sometimes ‘p.i.’ for ‘purely infinite’.

From the definition it is easy to see that non-zero quotients of p.i. algebras are
p.i.
A hereditary C *-subalgebra D ⊆ A can not have a (non-zero) character χ.
Indeed, otherwise a character χ would extend to a pure state ρ on A such that the
GNS construction defines an irreducible representation dρ : A → L(L2 (ρ)) with the
property that the algebra of compact operators K(L2 (ρ)) is contained in dρ (A). But
this is impossible, because the quotient A/d−1 ∼
ρ (0) = dρ (A) is again purely infinite.
But a rank-one projection can not be properly infinite.
Since every positive element in the ideal generated by a is in the norm-closure
of elements n c∗n acn , we can express (ii) equivalently:
P

For a ∈ A+ , c1 , c2 ∈ A and ε > 0, there is d ∈ A with kc∗1 ac1 + c∗2 ac2 − d∗ adk < ε.
One can see that it suffices to consider here c1 , c2 in a dense subset of the unit-
ball of A and a in a dense subset S of A+ with the property that (a − 1/n)+ ∈ S
for all a ∈ S and n ∈ N.
There is no essentially better reduction to dense subsets of A+ , because e.g. for
any unital A the set A+ + (0, 1) · 1A is dense in A+ and always j c∗j acj = d∗ ad
P

for given c1 , . . . cn ∈ A if a ∈ A+ + (0, 1) · 1A and d := a−1/2 ( c∗j acj )1/2 . And this
P

is certainly not equivalent to pure infiniteness.

The reader can see from Definition 1.2.1, e.g. that the algebras of compact
operators on a Hilbert space can not be purely infinite, and, more generally, that any
quotient A/J of a p.i. algebra A can not contain a (non-zero) Abelian hereditary C *-
subalgebra, i.e., A is strictly antiliminary (equivalently expressed: no irreducible
representation of A contains a non-zero compact operator in its image).

We list in Proposition 2.15.5 some elementary facts on purely infinite non-simple

C *-algebras. The proofs follow in the given order step by step almost trivially
from the foregoing result. The reader can find the proofs of the propositions and
corollaries 2.15.5–2.6.5 also in [462], [463] and [93]. The results on extensions and
passage to stabilizations are not trivial.
Is it good to use ‘‘stably p.i.’’ ?
328 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Proposition 2.15.5. Let A 6= {0} a C*-algebra.

(o) A is p.i. , if and only if, Mn (A) is p.i. , if and only if, A ⊗ K is p.i.
Compare with other places e.g. Part (iv)
(i) Let J is a closed ideal of A, with {0} = 6 J 6= A. Then A is p.i. if and only
if, J and A/J are p.i.
(ii) Every p.i. algebra A has no irreducible representation that contains a (no-
zero) compact operator in its image.
In particular, A is strictly antiliminary (i.e., each non-zero quotient
of A is NGCR).
(iii) Non-zero hereditary C*-subalgebras of a p.i. algebra are p.i.
(iv) The inductive limit of p.i. algebras is p.i.
Q
(v) The C*-algebra n Bn of bounded sequences (a1 , a2 , . . .) with an ∈ Bn is
p.i. if Bn is p.i. for every n = 1, 2, . . ..
In particular, the ultrapowers Bω of B are p.i. if M2 (B) is p.i.
(vi) M2 (A) is p.i. if and only if A is p.i. and, for every a ∈ A+ and ε > 0,
there exist b, c ∈ A with b∗ c = 0 and b∗ b = c∗ c = (a − ε)+ .
(vii) If A is p.i. and σ-unital, and has no unital quotient, then A is stable.
(viii) Zero is the only semi-finite lower semi-continuous 2-quasi-trace on
Ped(A)+ if A is p.i.

Part (vii) generalizes Zhang’s dichotomy for simple p.i. algebras. Its proof
reduces to a special case of [373, prop. 5.1], cf. our Corollaries 5.5.1 and 5.5.3.
The next result shows that the class of p.i. algebras is closed under extensions.
Temporary we call A stably p.i. if M2 (A) is p.i. This is equivalent to the
property that A ⊗ K is p.i. by Proposition 2.15.5(vi).

Proposition 2.15.6. Suppose that A is a C*-algebra, and J is a closed ideal

of A, such that J and A/J are both stably p.i. Then A is stably p.i.

Corollary 2.15.7. Every p.i. C*-algebra is stably p.i. This means:

(i) A ⊗ K is p.i. , if and only if, A is p.i.

(ii) A is p.i. if and only if for every b ∈ A+ and ε > 0 there exists c, d ∈ bAb
with c∗ d = 0 and c∗ c = d∗ d = (b − ε)+ .

Corollary 2.15.8. Let A a unital C*-algebra. Every non-zero quotient A/J

of A contains a non-unitary isometry, if and only if, 1A is properly infinite, i.e., if
and only if, there exists a unital C*-morphism ϕ : O∞ → A.

Proof. Easy direction: Suppose that 1A is properly infinite, – i.e., that there
is a unital C *-morphism ψ : E2 = C ∗ (s, t ; s∗ t = 0, s∗ s = 1 = t∗ t) → A –, then the
compositions πJ ◦ ψ with the C *-epimorphisms πJ deliver non-unitary isometries
in A/J for all closed ideals J 6= A of A.
The less trivial opposite direction is an evident consequence of Part (iii) of
Lemma 2.5.3: Let a := 1 and J := I(a). Then πJ (a) is finite in A/J by 2.5.3(iii).
 15. MORE ON PI(n)-ALGEBRAS 329

By assumption on the quotients, it means that A/J = {0} and J = I(a) = A, i.e.,
there exist d ∈ M2 (A) with d∗ (1 ⊕ 0)d = 1 ⊕ 1. This gives that 1 is properly infinite
in A: s∗ s = 1, s∗ t = 0 and t∗ t = 1 for s := d1,1 , t := d1,2 .
There is also a different looking “amazing” elementary argument for a proof of
the non-trivial direction. We carry it out in any detail (!) – otherwise it would take
only three lines. It goes as follows:
Since A itself contains at least one non-unitary isometry t by our assumptions,
the set of the projections pt := 1−tt∗ for all isometries t ∈ A is not empty. Consider
now the well-defined non-zero closed ideal J of A that is generated by all (co-range)
projections pt , where t runs through all non-unitary isometries in A. All positive
elements of the ideal J can be approximated by elements of the form d∗ pt d, because
the family of elements of this type is hereditary and d∗ ps d + e∗ pt e = f ∗ ps·t f for
f := ps d + se. (Use here that ps·t = ps + spt s∗ .)
If J = A then we are ready, because then there exist t, d ∈ A with t∗ t = 1
and kd∗ pt d − 1k < 1/2 . The isometries s := pt d(d∗ pt d)−1/2 and t have orthogonal
ranges s∗ t = 0 .
We show now that A/J does not contain a non-unitary isometry (in every
generality for all unital C *-algebras A that contain a non-unitary isometry). It
implies that J = A, i.e., A/J = {0} if each non-zero quotient of A contains a
non-unitary isometry.
Notice for the following, that a unital C *-algebra B does not contain any
non-unitary isometry, if and only if, every left-invertible element of B is also right-
invertible (and vice versa).
By definition of J, the image πJ (t) ∈ A/J of each isometry t ∈ A is a unitary
element in A/J because pt := 1 − tt∗ ∈ J . It implies that also the image πJ (b) of
each right-invertible or left-invertible element b ∈ A is invertible in A/J, because
t := b(b∗ b)−1/2 (respectively t := b∗ (bb∗ )−1/2 ) is an isometry.
We show now that – moreover – for every contraction a ∈ A with 1 − a∗ a ∈ J
there exists an isometry t ∈ A and d ∈ A such that b := ta−pt d satisfies k1−b∗ bk <
1/2 . Then pt ∈ J and πJ (t)∗ πJ (b) = πJ (a) is invertible in A/J by above remarks.
Indeed, as shown above, there exists d ∈ A and an isometry t ∈ A with

k(1 − a∗ a) − d∗ (1 − tt∗ )dk < 1/2

but this means for b := ta − pt d that k1 − b∗ bk < 1/2 .

Thus, for every unital C *-algebra A, the quotient A/J has the property that
each isometry in A/J is unitary, if J denotes the closed ideal of A that is generated
by {1 − tt∗ ; t ∈ A, t∗ t = 1}. Moreover, above considerations show that this ideal
J is equal to the ideal I(1) for 1 ∈ A defined in Lemma 2.5.3(iii). 

Definition 2.15.9. We say that A has the Global Glimm halving property
if, for each b ∈ A+ and δ > 0, there exist e ∈ A with e2 = 0 and e∗ e + ee∗ ∈ bAb
such that (b − δ)+ is in the ideal generated by e.
330 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Equivalently expressed:
For each b ≥ 0 and δ > 0, there is a C *-morphism

ϕ : M2 (C0 (0, 1]) → bAb

such that (b − δ)+ is contained in the closed ideal of A generated by the image of
ϕ.

Here seems to be a wrong definition?: The original “halving property”

says that, for given X ∈ A+ and ε > 0, there exist d1 , d2 ∈ A such that kd∗1 Xd2 k < ε
and (X − ε)+ is in the closed ideal J1 ∩ J2 of A, where Jk := J(d∗k Xdk ) is the closed
ideal generated by d∗k Xdk .
Is it the same? CHECK!

Remark 2.15.10. Every purely infinite C *-algebra has the global Glimm halv-
ing property by Proposition 2.15.11.
The other sorts of infinity of C *-algebras are in a sense of local nature that
does not shows any understandable relation to “Glimm having” ...

Proposition 2.15.11. For every non-zero C*-algebra A the following proper-

ties (i) and (ii) are equivalent:

(i) The algebra A has the Global Glimm halving property (Def. 2.15.9), and,
for every b ∈ A+ and every ε > 0, there is an m = m(b, ε) ∈ N, –
depending only on b and ε (and not on a) –, such that:
If b is in the closed ideal generated by a ∈ A+ , then there are d1 , . . . , dm ∈
A with
X
kb − d∗j adj k < ε .
1≤j≤m

(ii) The algebra A is purely infinite.

Proof. (ii)⇒(i): If A is purely infinite, then we can take m(b, ε) := 1 for all
b ∈ A+ and ε > 0. We use Lemma 2.5.15
Corollary 2.15.7 to check the global Glimm halving property:
There are τ ∈ (0, δ) and d1 , d2 ∈ A with d∗1 (a − τ )+ d2 = 0 and (a − δ)+ = d∗j (a −
1/2 1/2
τ )+ dj (j = 1, 2). Let e := (a − τ )+ d2 d∗1 (a − τ )+ , then e2 = 0, ee∗ + e∗ e ∈ aAa,
1/2
and d∗2 (a − τ )+ e(a − τ )1/2 d1 = (a − δ)2+ .
(i)⇒(ii): The proof is an adaptation of the proof of Proposition 2.2.1(iv):
By induction, we see from condition (ii), that, for every a ∈ A+ , δ > 0, and
m ∈ N, there exist f1 , . . . , fm in the closure of (a − 2δ)+ A(a − 2δ)+ , such that
fj∗ fi = δij f1∗ f1 , and (a − 3δ)+ is in the closed ideal of A which is generated by
f0 := f1∗ f1 . In particular, A can not have a non-zero character.
Now let a, b ∈ A+ , such that b is in the closed ideal generated by a. Let ε > 0
be given. We define ρ := ε/3, and let m := m((b − ρ)+ , ρ) .
 16. CHARACTERIZATIONS OF NON-SIMPLE STRONGLY P.I. ALGEBRAS 331

There exist δ > 0 and c1 , . . . , cn ∈ A, such that

X
c∗k (a − 3δ)+ ck = (b − ρ)+ .
1≤k≤n

Let f1 , . . . , fm as above and f0 := (f1 )∗ f1 . Then (b − ρ)+ is in the ideal generated

by f0 . By definition of m, there exist d1 , . . . , dm ∈ A, such that
X
k d∗j f0 dj − (b − ρ)+ k < ρ .
1≤j≤m

We define g ∈ C0 ((0, kak]) by g(t) := 0 for t ∈ [0, δ], g(t) := t/δ − 1 for
t ∈ [δ, 2δ], and g(t) = 1 for t ∈ [2δ, kak]. Let h(t) := (g(t)/t)1/2 .
Then d := h(a) · 1≤j≤m fj dj satisfies kd∗ ad − bk < ε .
P


Remark 2.15.12. In joint works with Étienne Blanchard and Mikael Rørdam
the above observation was used to show the following:
Suppose that A is a separable C*-algebra and that the primitive ideal space
Prim(A) is Hausdorff, i.e., A is the algebra of continuous sections vanishing at ∞
of a continuous field of simple C ∗ -algebras in the sense of [471] (48).
If, in addition, dim(Prim(A)) < ∞, then A is p.i. if and only if every simple
quotient of A is p.i. (i.e., if every fiber Ax of the continuous field on Prim(A)
associated to A is p.i.)
The criteria (i) and (ii) for p.i. algebras applies, because there is a “global”
version of the “local ” Glimm halving lemma, cf. Remark 2.1.16(ii).
The condition (i) comes from the finite dimension of Prim(A) if the fibers are
simple and p.i. A “global” version of the “halving” in Remark 2.1.16(ii) probably
does not extend to C *-bundles with non-simple fibers.

It could be conjectured that also continuous fields of C *-algebras over a Haus-

dorff space, and with strongly purely infinite fibers are strongly purely infinite,
in particular if the algebra of continuous sections again has a Hausdorff space as
primitive ideal space.

16. Characterizations of non-simple strongly p.i. algebras

The strongly purely infinite C *-algebras A, as defined in Definition 1.2.2, are

the C *-algebras that are later used for our variant of the classification program.
We need several equivalent characterizations that allow in special cases to verify
the strong pure infiniteness of given C *-algebras. So far it seems that the study
of the zoological garden of possibly different versions of pure infiniteness becomes
not easier to study in case of separable nuclear C *-algebras. And this is obvious,
because we have to work also with its ultra-powers and corona algebras that are in
general not nuclear.
48 I.e., A is the algebra of continuous sections vanishing at infinity of a continuous field of

simple C *-algebras over a Polish l.c. space.

332 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

We do not use here the later proven results of Chapters 9 or 10 right now (just
to underline logical consistency), instead we work formally with the “decoy” version
D∞ := O∞ ⊗ O∞ ⊗ · · · of O∞ , to emphasis that we use here only above proved
material and results from Appendices A and B.
!!! compare notations in the following with new names
coming from notation in extra papers (started at 2002, now 2018--2021)
We recall here the inequalities (2.2) for the definition of strongly pure infinite
C *-algebras A:
For any given a, b ∈ A+ and ε > 0 there exist d, e ∈ A that satisfy following
inequalities 16.1.

k a2 − d∗ a2 d k < ε , kb2 − e∗ b2 ek < ε and kd∗ abek < ε . (16.1)

If we find for given non-zero a, b ∈ A+ , δ ∈ (0, max(kak, kbk)) and ε ∈ (0, δ 2 ),

some elements d, e ∈ A that satisfy the inequality (16.1) with (a − δ)+ and (b − δ)+
in place of a and b, then we can find “solutions” d, e of the original inequality
(16.1) for a, b itself – but now with the additional upper estimate of the norms
max{kdk , kek} ≤ (2/ε) · max(kak, kbk) for d and e.
Improvement ??:
Take in M2 (A) the diagonal matrix G with diagonal entries d and e and then the
matrix F := [fjk ] with entries f11 = a, f12 = b and f21 = 0 = f22 and diagonal H
in M2 (A) with h11 = a2 , h22 = b2 and h12 = 0 = h21 then kG∗ (F ∗ F )G − Hk < 2ε.
One has here to find an estimate of the norm kGk of G !!!
We need a better estimate to get the formaly stronger matrix diagonalisation
property:
For each positive matrix T = [tjk ] ∈ Mn (A)+ and ε > 0 there exists a diagonal
matrix G with kG ∗ T G − Dk < ε where D is the diagonal of T , i.e. D is the
diagonal matrix with entries djk := δjk · tjk .
To find an estimate of kGk for G, one could use here that, for example,

kd∗ (a − ε)2+ d − (a − ε)2+ k < γ 2

with ε ≤ kak and γ ∈ (0, ε) implies

k(a − ε)+ dk ≤ (γ 2 + (kak − ε)2 )1/2 ≤ kak .

Then notice that ka−1 (a − ε)+ dk ≤ ε−1 k(a − ε)+ dk .

Remark 2.16.1. If A is a strongly purely infinite, J a closed ideal of A and D

a non-zero hereditary C *-subalgebra of A, then A/J and D are again strongly p.i.
To see this, take for a1 , b1 ∈ (A/J)+ elements a, b ∈ A+ with πJ (a) = a1 and
πJ (b) = b1 . Find d, e ∈ A that satisfy inequality (2.2) for a and b then d1 := πJ (d)
and e1 := πJ (e) fulfill the Inequality (2.2) for a1 and b1 (in place of a, b and d, e).
 16. CHARACTERIZATIONS OF NON-SIMPLE STRONGLY P.I. ALGEBRAS 333

Let a, b ∈ D+ and d, e ∈ A that satisfy the Inequality (2.2). Let c := a2 + b2

and γ := kak2 + kbk2 , then, for sufficiently large n ∈ N, en := c1/n dc1/n and
fn := c1/n ec1/n satisfy the Inequality (2.2) with en , fn ∈ D in place of e, f . It is
clear that kck ≤ γ and it is easy to see that
max(ka − c1/n ak , kb − c1/n bk) ≤ k(1 − c1/n )ck ≤ max{ 1/(n + 1) , γ · |1 − γ 1/n | } .
It follows that D is again strongly purely infinite.

Uses: lim c1/n a = a for each 0 ≤ a ≤ c: Indeed, it implies 0 ≤ a1/n ≤ c

a2 ≤ kaka ≤ kakc, and therefore
k(1 − c1/n )ak2 = k(1 − c1/n )a2 (1 − c1/n )k ≤ kak · k(1 − c1/n )2 ck ≤ sup{(t − t1+1/n )2 ,
and
sup{|t − t1+1/n | ; t ∈ [0, γ] } = max{ 1/(n + 1) , γ · |1 − γ 1/n | } .

Lemma 2.16.2. Suppose that a C*-algebra A contains a dense *-subalgebra B

with the properties that (b∗ b − ε)+ ∈ B for each b ∈ B and ε > 0.
Then A is strongly purely infinite, if and only if, for each f, g ∈ B and ε > 0,
there exist d, e ∈ A with
max{ k (f ∗ f )2 − d∗ (f ∗ f )2 d k , k(g ∗ g)2 − e∗ (g ∗ g)2 ek , kd∗ (g ∗ g)(f ∗ f )ek } < ε .
(16.2)
The elements d, e ∈ A in the inequalities (16.2) can be chosen with norms
−1
≤ ε · 2 max{ kf k4 , kgk4 } .
In particular, inductive limits of strongly purely infinite algebras are strongly
purely infinite.

Proof. If we take in Definition 1.2.2 of s.p.i. C *-algebras A the elements

a := f ∗ f and b := g ∗ g then we find d, e ∈ A that satisfy inequalities (16.2).
Conversely, if a, b ∈ A+ with kak = kbk = 1 and ε ∈ (0, 1) are given, then let
δ := ??? · ε???. We find f = f ∗ , g = g ∗ ∈ B with kf − a1/2 k < δ and kg − b1/2 k < δ.
There exist d, e ∈ A with kdk ≤ (2/ε)????, .... kd∗ f 4 d − f 4 k < ε/2, ke∗ g 4 e −
g k < ε/2, ... and kd∗ f 4 g 4 e − f 4 k < ε/2, ... It follows existence of contractions
4

x, y ∈ A with x∗ f 2 x = (a − γ)+ and y ∗ g 2 y = (b − γ)+ ...

By Lemma 2.1.9, there exists contractions x, y ∈ A with x∗ f+ x = (a1/2 −ε2 /4)+
By assumption, there exist exist d, e ∈ A that satisfy Inequalities (16.2). 

Definition 2.16.3. We say that a C *-algebra A has the (m.d.) matrix di-
agonalization property if for every matrix M = [bij ] ∈ M2 (A)+ and ε > 0 there
exists a diagonal matrix D := diag(d1 , d2 ) ∈ M2 (A) with
kD∗ M D − diag(b11 , b22 )k < ε .

Proposition 2.16.4. The C*-algebra A is strongly purely infinite, if and only

if, A has the matrix diagonalization property (m.d.), cf. Definition 2.16.3.
334 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Proof. ??
Suppose that A has the matrix diagonalization property of Definition 2.16.3.
We show that it implies that A is strongly purely infinite in sense of Definition
1.2.2, i.e., that, for every a, b ∈ A+ and ε > 0, there exist d, e ∈ A such that

k a2 − d∗ a2 d k < ε , kb2 − e∗ b2 ek < ε and kd∗ abek < ε . (16.3)

Here we can apply this formula with non-zero a, b ∈ A+ replaced by (a − γ)+

and (b − γ)+ for some γ ∈ (0, min(kak, kbk)) we can find a solution (d, e) of the
inequalities (16.3) for ε := γ 2 . Then k(a − γ)+ dk2 ≤ γ 2 + (kak − γ)2 ≤ kak2 ,
and similarly k(b − γ)+ ek ≤ kbk.
If a, b ∈ A+ satisfy kak = 1 = kbk and if ε ∈ (0, 1), then this gives the general
estimate for a solution (d, e) of the inequalities (16.3) with max(kek, kdk) ≤ 1/ε,
where one uses that e.g. ka−1 (a − γ)+ dk ≤ k(a − γ)+ dk /γ .
Let a, b ∈ A+ and consider the 2 × 2-matrix [aik ] = [a, b]> [a, b] in M2 (A)+ with
entries a11 := a2 , a22 := b2 , a∗21 = a12 = ab, and let ε > 0.
By assumption we find d1 , d2 ∈ A with

k diag(d1 , d2 )∗ [aik ] diag(d1 , d2 ) − diag(a, b)k < ε .

But this implies the Inequalities (16.3) with d := d1 and e := d2 .

Conversely suppose that A is strongly purely infinite. Let c ∈ A+ and take
a := b := c1/2 in (16.3) for given ε > 0. We observe that the there proposed
d, e ∈ A satisfy kd∗ cd − ck < ε, ke∗ ce − ck < ε and kd∗ cek < ε. Which says
k[d, e]∗ diag(c, 0)[d, e] − diag(c, c)k < ε in M2 (A) with [d, e] := d ⊗ p11 + e ⊗ p12 .
This can be rewritten also as

k diag(d, e)∗ (c ⊗ E) diag(d, e) − diag(c, c)k < ε ,

where E denotes the 2 × 2-matrix with all entries = 1.

Now let M = [bij ] ∈ M2 (A)+ a positive 2 × 2-matrix and f := b11 + b22 . Then

M ≤ 2 diag(b11 , b22 ) ≤ 2 diag(f, f )

By operator monotony of the function t1/2 this implies that

M 1/2 ≤ 21/2 diag(f 1/2 , f 1/2 )

It follow that
M = lim M 1/2 diag(f 1/n , f 1/n )M 1/2 ,
n→∞
thus, for δ > 0 there exists n ∈ N and d, e ∈ A such that for c := f 1/n (with suitable
n ∈ N) the row matrix
[g, h] := [c1/2 d, c1/2 e]M 1/2
satisfies
kM − [g, h]∗ [g, h]k < δ
That means, that a given positive matrix M ∈ M2 (A)+ can be arbitrary well
approximated by matrices that are products [g, h]∗ [g, h].
 16. CHARACTERIZATIONS OF NON-SIMPLE STRONGLY P.I. ALGEBRAS 335

Let u(g ∗ g)1/2 = g and v(h∗ h)1/2 = h the polar decompositions and α ∈
(0, 1/2) sufficiently small. Define S := (g ∗ g)α u∗ = u∗ (gg ∗ )α , T := (h∗ h)α v ∗ ,
a := (gg ∗ )1/2+α and b := (hh∗ )1/2+α . Then

diag(S, T )∗ [g, h]∗ [g, h] diag(S, T ) = [a, b]∗ [a, b]

and there exist d, e ∈ A that satisfy the Inequalities (16.3) for a, b with
max(kdk, kek) < 4/ε .
Thus, if we have for fixed ε the given 2 × 2-matrix M = [bij ] ∈ M2 (A)+
approximated well enough by a matrix [g, h]∗ [g, h] and take the α in the above
formula small enough then we get

k diag(d, e)∗ M diag(d, e) − diag(b11 , b22 )k < ε .

Remark 2.16.5. Property (m.d.) of a C *-algebra A implies the (formally

stronger) property of A that, for every m, n ∈ N with n > 1, every positive ele-
ments a1 , . . . , am ∈ A+ and arbitrary elements x` ∈ A (` = 1, . . . , n) there exist
contractions d1 , . . . , dm ∈ A such that kd∗j aj dj − aj k < ε for j = 1, . . . , m and
kd∗j x` dk k < ε for ` = 1, . . . , n and j 6= k, 1 ≤ j, k ≤ m.
?? [the cite is KirRor2] Compare [463, ????], but there are misleading typos
that should be read: ?????

Definition 2.16.6. The following properties of C *-algebras A are between

pure infiniteness and strong pure infiniteness.

(lsp) The algebra A is purely infinite and for every non-zero contraction a ∈
A there exist isometries s, t in the multiplier algebra M(D) of D :=
(a a∗ + a∗ a) A (a a∗ + a∗ a) with s∗ t = 0 and 3 · ks∗ atk < 2 .
(Then we call A sq-pi.)
Why not ”(lsp)” ? Anyway, it seems to be a bit difficult to use or to
compare with others ...
(labs) For every positive contraction a ∈ (Aω )+ , there exists a *-monomorphism
ψ : C ∗ (a) ⊗ O∞ → Aω with ψ(a ⊗ 1) = a. (Then we say that A is locally
O∞ -absorbing.)
(lcpi) {a}0 ∩ Aω is p.i. for every positive element a ∈ (Aω )+ .
(A is locally commutant-p.i..)
(cpi) For every separable commutative C *-subalgebra C of Aω , there exists a
*-monomorphism ψ : C ⊗ O∞ → Aω with ψ(a ⊗ 1) = a for a ∈ C.
Then A is commutant purely infinite or c.p.i..

Remark 2.16.7. We have following two rows of implications:

s.p.i. ⇔ m.d.p. ⇒ (lsp) ⇒ p.i.

s.p.i. ⇒ (cpi) ⇒ (lcpi) ⇒ (labs) ⇒ p.i.

336 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Here is an equivalent formulations of (c.p.i):

There exist increasing continuous functions fn : [0, 2] → [0, 3] with fn (0) = 0

with the property that for n contractions a1 , . . . , an ∈ A+ , there are contractions
d1 , d2 ∈ A such that

kdi ak − ak di k + kd∗i ak dj − δi,j ak k ≤ fn (γ(a1 , . . . , an ))

for 1 ≤ k ≤ n, i, j ∈ {1, 2}, where we let γ(a1 , . . . , an ) := sup1≤p,q≤ k ap aq − aq ap k .

For every separable and commutative C *-subalgebra C ⊆ Aω , the relative com-
mutant C 0 ∩ Aω is purely infinite.
(Equivalently: F (C, A) := (C 0 ∩ Aω )/ Ann(C, Aω ) is p.i. for every separable com-
mutative C *-subalgebra C of Aω .)

Proposition 2.16.8. We have the implications:

A∼
= A ⊗ D∞
implies that
A is s.p.i. , if and only if, Aω is s.p.i. ,
and this implies that
C 0 ∩ Aω is p.i. for all separable commutative subalgebras C ⊆ Aω .
The latter implies that A is pi(1), iff A pi-1, iff A is p.i. Aω is p.i.

Remark 2.16.9. “Strongly” p.i. algebras in the sense of Definition 1.2.2 are
p.i.
But it needs some work to see that strongly p.i. algebras in the sense of Defi-
nition 1.2.2 have the WvN-property of Definition 1.2.3.
The WvN-property is needed for our classification results. We give in Chapter
3 a proof (based on [443] and the joint paper with M. Rørdam [463]) that strongly
p.i. algebras have the WvN-property.
Compare the remarks at the end of Chapter 3.
A ⊗ F is strongly p.i. for F := O∞ ⊗ O∞ ⊗ . . . and for every C *-algebra A.
Notice that F ∼
= O∞ because O∞ ⊗ O∞ ∼ = O∞ and
Give precise citations (!!!) of the work with E. Blanchard and M. Rørdam:
In joint works with E. Blanchard and M. Rørdam about p.i. algebras there are
some sufficient criteria for p.i. algebras to be strongly p.i. and the study of some
cases where p.i. implies “strongly” p.i. in some cases, e.g.:
If A is p.i. and Prim(A) is Hausdorff, then A is strongly p.i.
If A is locally p.i. and Prim(A) is a Hausdorff space of finite dimension, then
A is p.i. (and therefore is s.p.i.).
If A is exact and “approximately divisible”, ...
 16. CHARACTERIZATIONS OF NON-SIMPLE STRONGLY P.I. ALGEBRAS 337

Where is the definition of “approximately divisible”? Are they criteria con-

cerning F (A) ???
... then A is strongly purely infinite if every (additive) lower semi-continuous
traces τ : A+ → [0, ∞] takes only the values {0, ∞}. The same happens if A ∼ =
A ⊗ Z, where Z is the Jiang-Su algebra.
The algebras A ⊗ O∞ are always s.p.i. Separable nuclear s.p.i. algebras A are
isomorphic to A ⊗ O∞ ( 49 )
Since O∞ is approximately divisible, cf. Remark B.5.1, this implies that a
separable nuclear C *-algebra A is s.p.i., if and only if, A is approximately divisible
and has only trivial lower semi-continuous traces.
give Refs or Cites:
An other equivalent to s.p.i. for separable nuclear A is that A tensorial absorbs the
Jiang-Su algebra Z and that A has only trivial l.s.c. traces. (The proof is more
engaged then the case of ”approximate divisibility”.)
If A has real rank zero and is locally p.i., then A is strongly p.i. (More generally,
for each projection p in a locally p.i. algebra A there is n = n(p, A) ∈ N such that
p ⊗ 1n is infinite in Mn (A). This property holds also for each projection r in
each quotient A/J of A – with n(πJ (a), A/J) possibly different from n(p, A) –,
because A/J is locally p.i. too. Since p ∈ J for projection p ∈ A, or there is stable
hereditary C *-subalgebra D of pAp that is not contained in J. Then πJ (D) is
non-zero and stable and contains a non-zero projection r ∈ πJ (D) ⊆ A/J. Now
let n := n(r, A/J). The stability of π(D) allows to define a monomorphism from
r(A/J)r⊗Mn into π(D). Thus πJ (D) contains an infinite projection. The argument
works also if each non-zero hereditary C *-subalgebra E of each quotient A/J of A
contains a non-zero projection r. This projection must be properly infinite by the
above arguments (applied to r and the quotients A/I for J ≤ I ≤ A). It again
follows that every non-zero a ∈ A+ is properly infinite, i.e., that A is purely infinite,
One gets that A is strongly p.i. if A has real rank zero case, because algebras of
real rank zero contain sufficiently many locally central elements: namely just the
projections.)
If A is weakly p.i. or if M(A) is (itself) locally p.i., then there is n ∈ N such that
Mn (M(A)) has a properly infinite unit. (The first case has an estimate n ≤ 1 + 2k
if A is pi(k).)

Remark 2.16.10. If X is any (non-empty) locally compact Hausdorff space

and ω ∈ β(X).
Where is β(X) defined?

49 The stable and the unital case follow from Theorem M or from [463, thm. 8.6], this passes

to the general case by the extension property of s.p.i. algebras, which implies that the unit of
M(A) is properly infinite for s.p.i. A and that E(O∞ , O∞ ) is s.p.i. Alternatively one can use the
arguments in Chapter 10.
338 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Let Iω the kernel of the character χω associated to ω. It is given by

f ∈ Cb (X) ∼
= C(β(X)) 7→ f (ω) := χω (f ) .

We define (similar to the case X = N) the C *-algebra Aω := Cb (X, A)/Jω , where

Jω := Iω · Cb (X, B)
It is not immediate but similar to the arguments used in case ω ∈ β(N) \ N that
we have that A is strongly p.i. if Aω is strongly p.i.
Clearly the case ω ∈ X is trivial, because s.p.i. passes to quotients, cf. Propo-
sition ?? or Proposition 2.17.1.
The proof that strong pure infiniteness of A implies strong pure infiniteness
of Cb (X, A) (at least for σ-compact X) and of its quotient Aω follows from the
the strong pure infiniteness of A and factorizes over the strong pure infiniteness of
C0 (X, A) and its multiplier algebra M(C0 (X, A)) = Cb,st (X, M(A)) with σ-unital
A and σ-compact X.
(Here, Cb,st (X, M(A)) denotes the C *-algebra of bounded maps f : X 7→
M(A) with the property that x 7→ f (x)a and x 7→ af (x) are in Cb (X, A) for
all a ∈ A.)

Proof. to be filled in ??
We use the norm-function N : Cb (X, A) → Cb (X)+ , that is defined as
N (f )(x) := kf (x)k. The norm of πω (f ) ∈ Aω is then given by χω (N (f )) = kπω (f )k.
If χω (N (f )) = 0 then N (f ) ∈ (Iω )+ . We get inf x∈X G(x) = 0 for all G ∈ (Iω )+ if
we use now that Iω ∼ = C0 (β(X) \ {ω}), and that X is dense in β(X).
Let a1 , a2 ∈ A positive contractions and ε > 0. Since Aω is s.p.i., there exist
contractions f1 , f2 ∈ Cb (X, A) such that the functions

gj,k := N (fj∗ a∗j ak fk − δj,k a∗j ak ) ∈ Cb (X)+

satisfy χω (gj,k ) < ε/2 for j, k ∈ {1, 2}.

P
Let G(x) := j,k (gj,k (x) − ε/2)+ . Then χω (G) = 0, i.e., G ∈ Iω . Thus,
inf x∈X G(x) = 0. This implies the existence of x0 ∈ X with G(x0 ) < ε/2. But this
yields gj,k (x0 ) < ε for j, k ∈ {1, 2}, and d1 := f1 (x0 ) and d2 := f2 (x0 ) satisfy for
j, k ∈ {1, 2} the inequalities

kd∗j a∗j ak dk − δj,k a∗j ak k < ε .

Question 2.16.11. Let A := C0 ((1/2, 1], K) and F ⊆ C ∗ ((0, 1], M(A)) the
C *-subalgebra that is generated by A and f : t ∈ (0, 1] 7→ ψ(t) · 1M(A) with ψ(t) :=
min(2t, 1).
Is f an infinite element in F ? (certainly not properly infinite).
( It seems not to be such ???, even if one tensors again...??? )
 17. PERMANENCE PROPERTIES OF STRONGLY P.I. ALGEBRAS 339

17. Permanence properties of strongly p.i. algebras

We list some permanence properties of the class of strongly p.i. C *-algebras A

(i.e., where A is ”s.p.i.”, as defined in Definition ????? ):
Passage to non-zero hereditary C *-subalgebras D of A
A is s.p.i. if A contains a full hereditary C *-subalgebra D such that D is s.p.i.
Morita-equivalence,
Extensions,
passage to quotients
inductive limits
Each s.p.i. A is the inductive limit of the net of separable C *-subalgebras that
are s.p.i. and are relatively weakly injective in A.
Infinite direct products, Π∞ (A1 , A2 , . . .), (and passage to its quotients, e.g. Aω
if An := A)
Commutator algebras of separable amenable C *-subalgebras of Aω if A is s.p.i.
spatial tensor product with arbitrary C *-algebras in case that prime-ideals are
cartesian. (Is exactness necessary?)
Some crossed products by quantum groups??

Proposition 2.17.1. If A is strongly p.i., then every hereditary C*-subalgebra

and every non-zero quotient A/J is strongly p.i.

Proof. Let D ⊆ A a hereditary C *-subalgebra, a1 , a2 ∈ D+ positive contrac-

tions and ε > 0.
Since ka4γ a2 − a2 k ≤ 2γ for positive contractions a ∈ A+ , we get, for γ := ε/4
and k, j ∈ {1, 2}, that

δj,k kaj ak − a2γ 2γ

j aj ak ak k ≤ ε/2 .

There exist e1 , e2 ∈ A such that, for j, k ∈ {1, 2} ,

ke∗j aγj aj ak aγk ek − δjk aγj aj ak aγk k < ε/2 .

Because aγ1 and aγ2 are positive contractions in D, we obtain that the elements d1 :=
aγ1 e1 aγ1 and d2 := aγ2 e2 aγ2 are in D and satisfy, for j, k ∈ {1, 2}, the inequalities

kd∗j aj ak dk − δjk aj ak k < ε/2 + ε/2 .

If J / A is a closed ideal, b1 , b2 ∈ (A/J)+ and ε > 0, then there are a1 , a2 ∈ A+ and

e1 , e2 ∈ A such that πJ (ak ) = bk and ke∗j aj ak ek − δjk aj ak k < ε for j, k ∈ {1, 2} .
Thus, dj := πJ (ej ), j = 1, 2 satisfy kd∗j bj bk dk − δjk bj bk k < ε for j, k ∈ {1, 2} . 

Proposition 2.17.2. If A is a purely infinite C*-algebra and B is a simple

non-elementary C*-algebra and at least one of A or B is exact, then A ⊗ B is
strongly purely infinite.
340 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Proof. Where is the general definition of ” non-elementary C *-algebra ” ???

Does it work well if B is UHF?
What is the bound if for every finite sequence y1 , . . . , yn ∈ B, b1 , b2 ∈ B+
and δ > 0 there exist contractions z1 , z2 , g1 , g2 , h1 , h2 ∈ B such that kz1∗ z2 k ≤ δ,
k[zk , y1 ]k ≤ δ for k = 1, . . . , n gk∗ (zk∗ bk zk )gk + h∗k (zk∗ bk zk )hk = (bk − δ)+ .
Since one of A or B is exact, we get from Lemma ?? that

F := {a ⊗ b ; 0 6= a ∈ A+ , 0 6= b ∈ B+ }

is a “filling family” for A ⊗ B.

Definition of “filling family” for a C *-algebra ???
The non-zero elements a ⊗ b ∈ F are properly infinite in A ⊗ B and a ⊗ b ∼ a ⊗ c
in A ⊗ B for b, c ∈ B+ \ {0}.
P ∗
Moreover we find for δ > 0 elements g1 , . . . , gn ∈ B with g` bg` = (c − δ)+ .
h1 , . . . , hn with

diag((a − δ)+ , . . . , (a − δ)+ ) = [h1 , . . . , hn ]∗ a[h1 , . . . , hn ] .

Then D := ` h` ⊗ g` ∈ satisfies D∗ (a ⊗ b)D = (a − δ)+ ⊗ (c − δ)+ .

P

Give controlling bound !!

If follows that it is enough to prove that for each δ > 0, elements
Pn
a1 , a2 , c1 , . . . , cn ∈ A+ , b1 , b2 , d1 , . . . , dn ∈ B+ , y := k=1 ck ⊗ dk there exist
non-zero contractions x1 , x2 ∈ B+ with kxj bj xj k ≥ kbj k − δ and x1 b1 dk b2 x2 = 0
for k = 1, . . . , n.
Provided one has a controlling bound for the xj and/or D depending on δ > 0.
... 

Theorem 2.17.3. The minimal (= spatial) C*-algebra tensor product A ⊗ B

is strongly p.i. if A and B satisfy at least one of the following conditions:

(1) A is strongly p.i. and the natural map (I, J) 7→ A ⊗ J + I ⊗ B from

prime(A) × prime(B) into prime(A ⊗ B) is surjective.
(2) A is weakly p.i, and B is exact, simple and non-elementary.
(??) ?????????????? If A is (quasi-)traceless, B is exact and tensorial absorbs
a tensorial self-absorbing separable unital C*-algebra D ∼
6 C.
=

Proof. to be filled in ?? 

Corollary 2.17.4. Let A and B C*-algebras. The algebra A ⊗ B is strongly

p.i. if A is strongly p.i. and at least one of the C*-algebras A or B is exact.

Proof. Recall that, by our convention, A ⊗ B denotes the completion of the

algebraic tensor product A B with respect to the spatial C *-norm. This is the
minimal C *-norm on the algebraic tensor product A B of C *-algebras A and B,
 17. PERMANENCE PROPERTIES OF STRONGLY P.I. ALGEBRAS 341

cf. [766], [767, thm. IV.4.19]. In particular each non-zero closed ideal of A ⊗ B
has non-zero intersection with A B.
More precisely, it is shown in [766], but is not explicitly stated there, that if a
C *-semi-norm N on A B has the property that N (a ⊗ b) > 0 for each non-zero
a ∈ A+ and non-zero b ∈ B+ , then kxkmax ≥ N (x) ≥ kxk∗ for each x ∈ A B,
where we here (temporary and only here) denote by kxk∗ the spatial norm. This
shows that kxk∗ is the minimal possible C *-norm on A B, from now on denoted
by kxkmin := kxk∗ or simply by kxk.
But this shows even more: If K is a non-zero closed ideal of A ⊗ B then there
exist non-zero positive contractions a ∈ A+ and b ∈ B+ such that a ⊗ b ∈ K,
because otherwise for the quotient map A ⊗ B → (A ⊗ B)/K would hold kxkmin ≥
kπK (x)k ≥ kxkmin for all x ∈ A B, which implies K = {0}. Compare also the
stronger result of Lemma A.24.1, in Section 24 of Appendix A.
Thus, every non-zero closed ideal K of the spatial tensor product A⊗B contains
a tensor product I ⊗ J of non-zero ideals I / A and J / B.
Exactness of a C *-algebra A means that for every C *-algebra B and each
closed ideal J / B the sequence

0 → A ⊗ J → A ⊗ B → A ⊗ (B/J) → 0

is short exact. The exactness property of A induces the exactness of I and A/I for
each closed ideal I / A of A. Moreover the exactness of A yields the local lifting
property for its quotients A/I. The latter yields also the short exactness of the
sequences
0 → I ⊗ B → A ⊗ B → (A/I) ⊗ B → 0 ,
and all together imply – now by the 3 × 3-lemma (also called “5 of 6 lemma”) of
category theory – the short exactness of

0 → A ⊗ J + B ⊗ I → A ⊗ B → (A/I) ⊗ (B/J) → 0 ,

i.e., if A or B is exact then for each I / A and J / B there are natural isomorphisms

(A ⊗ B)/(A ⊗ J + B ⊗ I) ∼
= (A/I) ⊗ (B/J) .

Recall that a closed ideal K / C of a C *-algebra is prime, K ∈ prime(C), if

K1 ∩ K2 ⊆ K implies always that at least one of K1 and K2 is contained in K
for ideals K1 / C and K2 / C. The original definition says only that K1 ∩ K2 =
K implies K1 = K or K1 = K, but one can change this into the above more
flexible formulation if we replace Kj by K + Kj for j = 1, 2. Then it means that
πK (K1 ) ∩ πK (K2 ) = 0 implies that πK (K1 ) = 0 or πK (K2 ) = 0.
Note that all primitive ideals are prime [616, prop. 3.13.10], and in case of
separable C *-algebras all prime ideals are primitive, cf. [616, prop. 4.3.6.]. For
non-separable A the space of prime ideals prime(A) is the “sobrification” or “point-
wise completion” of the space Prim(A) of primitive ideals. It means that the lattice
of open subsets of Prim(A) is the same as the lattice of open subsets of prime(A).
342 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

If one of A or B is exact, then the map (I, J) 7→ A ⊗ J + I ⊗ B from prime(A) ×

prime(B) into prime(A⊗B) is surjective. To see this, observe that, for a prime ideal
K of A⊗B, the maximal possible ideals I ∈ I(A) and J ∈ I(B) with A⊗J +I ⊗B ⊆
K are prime, and that the exactness implies then that K = A ⊗ J + I ⊗ B. The
maximal I and J with the properties I ⊗ B ⊆ K and A ⊗ J ⊆ K are well-defined,
because, e.g. the set of all algebraic *-ideals I ∈ A with I ⊗ B ⊆ K is closed under
finite and infinite algebraic sums and under taking the closure.
Indeed, the epimorphism A⊗B → A/I ⊗B/J has, – by exactness of A or of B –,
precisely the kernel A ⊗J +I ⊗B, as we have seen above. If π : A⊗B → A/I ⊗B/J
denotes the quotient map, then π(K) becomes an ideal of A/I ⊗B/J that must have
trivial intersection π(K) ∩ (A/I B/J) = {0} with the algebraic tensor product
A/I B/J, because otherwise our above considerations on the minimal (= spatial)
C *-norm (here on A/I B/J) show that there would be elements c ∈ A+ and
d ∈ B+ with c ⊗ d 6∈ A ⊗ J + I ⊗ B and c ⊗ d ∈ K. But the latter would imply, by
primeness of K, that also the one of the ideals A ⊗ J1 + I ⊗ B or A ⊗ J + I1 ⊗ B
is contained in K, where I1 := I + span(AcA) and J1 := J + span(BdB). This
contradicts the maximality of the above defined ideals I and J with the property
A ⊗ J + I ⊗ B ⊆ K.
In fact, the map

(I, J) ∈ prime(A) × prime(B) 7→ A ⊗ J + I ⊗ B ∈ prime(A ⊗ B)

is moreover a homeomorphism of (sober) locally quasi-compact T0 spaces if one of

A or B is exact.
Thus, Theorem 2.17.3(i) applies. 

Since C0 (X, A) ∼= C0 (X) ⊗ A and C0 (X) is exact, we get from Theorem 2.17.3,
Corollary 2.17.4 and Proposition ??:

Corollary 2.17.5. For every locally compact Hausdorff space X, C0 (X, A) is

strongly purely infinite, if and only if, A is strongly purely infinite.

Theorem 2.17.6. Suppose that A is σ-unital. Then M(A) is strongly p.i., if

and only if, A is strongly p.i.

Proof. By Proposition 2.17.1 the ideal A of M(A) is strongly p.i., if M(A)

is strongly p.i.
to be filled in, see extra SPI-paper ?? 

Corollary 2.17.7. Let X a non-empty σ-compact locally compact Hausdorff

space. The algebra Cb (X, A) is strongly p.i., if and only if, A is strongly p.i., if and
only if, Q(X, A) := Cb (X, A)/ C0 (X, A) is strongly p.i., if and only if, Q(X, A)|F
is strongly p.i. for a non-empty closed subset F of β(X) \ X.

Proof. The algebra A is s.p.i. if and only if A ⊗ C0 (X) is s.p.i., by Corollary

2.17.5.
 18. STRONGLY P.I. CROSSED PRODUCTS AND GENERALIZED TOEPLITZ ALGEBRAS343

If X is σ-compact, A is s.p.i. and D ⊆ A is a σ-unital hereditary C *-subalgebra,

the algebra C0 (X, D) ∼ = D ⊗ C0 (X) is σ-unital and s.p.i. By Theorem 2.17.6, it
follows that M(D ⊗ C0 (X)) is s.p.i. if and only if D ⊗ C0 (X) is s.p.i.
The algebra Cb (X, D) is an ideal of the algebra M(C0 (X, D)). It implies that
Cb (X, D) is s.p.i. if C0 (X, D) is s.p.i. (and σ-unital).
Since X is σ-compact, the C *-algebra Cb (X, A) is the inductive limit of its
C *-subalgebras Cb (X, D) with D ⊆ A σ-unital hereditary C *-subalgebras of A,
we obtain that Cb (X, A) is s.p.i. if A is s.p.i.,
Now use that Q(X, A)|F is a quotient of Cb (X, A).
Thus, Q(X, A) (case F = β(X) \ X), Q(X, A)|F and, in particular,
Q(X, A)|{ω} =: Aω for a point ω ∈ F ⊆ β(X) \ X are strongly p.i. by Proposition
2.17.1.
If Aω is s.p.i. for some ω ∈ β(X) then A is s.p.i. by Remark 2.16.10. 

We do not know if it is necessary to suppose that X is σ-compact in Corollary

2.17.7.
If X is not σ-compact then in general Cb (X, A) is not anymore the inductive
limit of its C *-subalgebras Cb (X, D) with hereditary σ-unital D ⊆ E, and M(D ⊗
C0 (X)) is possibly not s.p.i.
The class of strongly purely infinite C *-algebras is closed under extension by
the following:

Theorem 2.17.8. Let J / A a closed ideal. The C*-algebra A is strongly purely

infinite if and only if J and A/J are strongly purely infinite.

Proof. We know that A is purely infinite by Theorem ??.

The proof can easily reduced to the case, where A is separable and unital
(thus contains a copy of O∞ unitally), and where J is stable. The proof uses the
K1 -injectivity of A as an important tool.
See [443] for a proof of Theorem 2.17.8. 

18. Strongly p.i. crossed products and generalized Toeplitz algebras

change whole section.

Proposition 2.18.1. Suppose that A is a C*-algebra and that α ∈ Aut(A) has

the property that {0} and A are the only α-invariant ideals.
simple and p.i. crossed products appear in following cases:
If, for every a ∈ A+ \ {0} and n ∈ N there exist b ∈ A+ \ {0} such that b ≤ a
and αk (b) are pairwise orthogonal, then A oα Z is simple and purely infinite.
above this is wrongly stated
Full-corner endomorphisms, ...????
344 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

There is a projection p ∈ Z(A∗∗ ) such that αn (p) is faithful for A

W
n∈Z

(or: is not contained in any α-invariant open central projection of A∗∗ ????)
and αm (p) ⊥ αn (p) for m 6= n.
For every non-zero α-invariant closed central projection p and every n ∈ N,
a ∈ A+ , ε > 0 there is a non-zero central projection q ≤ p with αj (q)αk (q) = 0 for
−n ≤ j, k ≤ n and krak + ε ≥ kpak. Then A oα Z is simple.
A simple and p.i., for every n ∈ N there is a unitary Un ∈ F (A) such that
α(Un ) = e2πi/n Un . Then A oα Z is simple.
The action F (α) on F (A) has full spectrum.

Proof. ??
(Reduction to separable case?) 

Corollary 2.18.2. Suppose that A is purely infinite and simple, and α ∈

Aut(A) is such that α∗ has infinite order in Aut(K∗ (A)). Then A oα Z is purely
infinite and simple.

Proof. to be filled in ?? 

Corollary 2.18.3. Suppose that B is a simple C*-algebra and ψ : B → B is

an endomorphism from B into B, such that for each a, b ∈ B+ with kbk = 1 ≥ kak
and for each ε > 0 there are k ∈ N and d ∈ B with kψ k (a) − d∗ bdk < ε (respectively
ka − d∗ ψ k (b)dk < ε).
Then the semi-crossed product B oψ N is purely infinite.

Proof. Consider the natural circle action on B oψ N ...

to be filled in ??


Corollary 2.18.4. Suppose that B is stable and σ-unital and that

h : B ,→ M(B)

is a non-degenerate *-monomorphism.
Let s, t ∈ M(B) isometries that are canonical generators of a copy of O2 .
Suppose that the following conditions (i)–(iii) are satisfied:

(i) h(B) ∩ B = {0} .

(ii) For each 0 6= b ∈ B the closed ideal of B generated by h(b)B is equal to
B.
(iii) The C*-morphism h is approximately unitarily equivalent to the C*-
morphism δ2 ◦ h := sh(·)s∗ + th(·)t∗ .

Then the Toeplitz algebra TE of the Hilbert B-B bi-module E defined by the
left-action of B on B – given by h : B → M(B) – is KK-equivalent to B.
 18. STRONGLY P.I. CROSSED PRODUCTS AND GENERALIZED TOEPLITZ ALGEBRAS345

?????????????

Proof. Consider the endomorphism β : C → C on C := S for S :=

span( n M(h)n (B)), given by β := M(h)|C. Then (indlim(β : S → S)) o Z is
S

simple and purely infinite, because Corollary 2.18.3 applies to (C, β). 

Next Def.s should be quoted... give them precise !...

Definition 2.18.5. Let H denote a Hilbert A-B bi-module.
Let A a separable C *-algebra and H a Hilbert A-A bi-module.
T (H) (generalized) Toeplitz-Fock C *-algebra corresponding to H.
Let OH denote the Cuntz-Pimsner C *-algebra corresponding to T (H).
The gauge group is available ...
(where is Def. of ”gauge group” ???)
The natural action of bi-module automorphisms and its generalizations. ...

Remark 2.18.6. Automorphisms of H act on T (H).

Special case: circle action. (Def. of ”Circle action” ?)
Moreover: epimorphisms and other constructions.
On , A = C, H ∼
= Cn
O∞ , A = C, H = separable Hilbert space of countable dimension.
Compare Appendix A.
Corollary 2.18.7. Let A a separable C*-algebra and H a
countably generated
Hilbert A-A bi-module. Then T (H) is separable.
Quotients of T (H) can be found by invariant ideals of H.
If A is nuclear (respectively is exact, has WEP = is weakly injective, etc. ...)
then T (H) is nuclear (respectively exact, has WEP = is weakly injective ???,
etc....).
If d(A) ∩ K(H) = {0} and d is injective then T (H) = OH
OH is simple if ??????.
T (H) is simple, separable, stable and nuclear if A is separable, stable and nu-
clear, d(A) ∩ K(H) = {0} and there are no-invariant ideals (modular quotients).

Proof. Use Lemma ??

(ref. is:: ” lem:App.A.fix-algebra.of.compact.group”)
and the gauge circle action on T (H) for nuclearity, exactness and WEP ( =
weak injectivity ).
(here WEP means ”weak expectation property” ??) 
346 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

Definition 2.18.8. An automorphism α of a C *-algebra A is properly outer,

if for every a ∈ A
e = A + C1 and every non-zero hereditary C *-subalgebra B of A
holds that
inf{ kxaα(x)k ; 0 ≤ x ∈ B, kxk = 1 } = 0 .

Remark 2.18.9. Suppose that A is separable.

(1) By [578, thm. 6.6], the automorphism α is properly outer, if and only if, for
every non-zero α-invariant closed ideal I / A and every unitary u in the multiplier
algebra M(I) holds kα|I − Adu k = 2.
This is also equivalent to:
There is no non-zero α-invariant I / A such that there is a *-derivation δ of I and
a unitary u ∈ M(I), with α|I = Adu ◦ exp δ.
Check next again
(2) Proper outer-ness of α is implied by the property that for each primitive
ideal J of A oα Z the group T(α,
e J) of λ ∈ T (= S 1 ) with α
bλ (J) = J is non-trivial,
where αb : T → Aut(A oα Z) means the dual action of the Z-action n ∈ Z 7→ αn .
(See proof of [478, lem.1.1].)
(3) If A is separable and simple, then
\
{1} =
6 T(α)
e := T(α,
e J)
J

implies that α is inner, cf. [577].

(4) The subgroup T(α)
e ⊆ T is called the strong Connes spectrum of α, cf. [477].

Theorem 2.18.10. Suppose that G 6= {e} is a countable discrete group, that

A is a separable C*-algebra A of infinite dimension, and that α : G → Aut(A) is
an action of G on A such that the automorphism α(g) of A is a properly outer for
each g ∈ G \ {e} (cf. Definition 2.18.8).
If there is (fixed) n ∈ N such that, for each a, b ∈ A+ with kak ≤ 1 = kbk and
ε > 0, there are g1 , . . . , gn ∈ G and d1 , . . . , dn ∈ A with ka− k d∗k α(gk )(b)dk k < ε ,
P

then the reduced crossed product A oα,r G is simple and purely infinite.

Proof. Let {0} 6= I / A invariant under α(G), and let b ∈ I+ with kbk =
1. Furthermore, let a ∈ A+ with kak = 1 and ε > 0. By assumption, we find
g1 , . . . , gn ∈ G and d1 , . . . , dn ∈ A with ka + Ik ≤ ka − k d∗k α(gn )(b)dk k < ε.
P

Thus a ∈ I and I = A. It follows now from the proof of [478, thm. 3.1] or
by [578, thm. 7.2] that C := A oα,r G is simple. It can not be (linearly) one-
dimensional, because A is (isomorphic to) a C *-subalgebra of C. We can consider
A as a subalgebra of C. Suppose that C is of finite dimension, then A is unital and
the reduced group C *-algebra of G is contained in C
...??? further?
We show that the simple C *-algebra A oα,r G is locally purely infinite.
To be filled in ?? 
 18. STRONGLY P.I. CROSSED PRODUCTS AND GENERALIZED TOEPLITZ ALGEBRAS347

Definition 2.18.11. Let α : G → Aut(A) an action of a discrete group G on

a C *-algebra A. We write also α(g) for the natural extension α(g)∗∗ to the second
conjugate A∗∗ of A.
The action α has the weak Rokhlin property if the center of A∗∗ contains
a projection p with the properties

(i) pα(g)(p) = 0 for all g ∈ G \ {e}, and

(ii) supg∈G kp · α(g)(a)k > 0 for all nonzero a ∈ A.

The action α has the residual weak Rokhlin property if the projection p ∈
Z(A∗∗ ) satisfies (i) and the following stronger property (ii*) instead of (ii):

(ii*) For each a ∈ A and every α(G)-invariant closed ideal J / A (with central
support qJ in A∗∗ ) ka + Jk = supg∈G k(1 − qJ )p · α(g)(a)k.

The action α of a locally compact group G on A is exact if J oα|,r G is the kernel

of the natural epimorphism A oα,r G → (A/J) o[α],r G for every α(G)-invariant
ideal J / A.
An l.c. group G is exact if every action of G on C *-algebras is exact.

Lemma 2.18.12. The weak Rokhlin property passes to subgroups H of G.

The weak Rokhlin property for G = Z or G = Zn implies that α(g) is properly
outer for each g ∈ G \ {e}.
The residual Rokhlin property passes to quotients.

Proof. to be filled in ?? 

Theorem 2.18.13. Suppose that A is a separable C*-algebra, G a countable

discrete exact group, and α : G → Aut(A) is an action that satisfies (I) and (II):

(I) α satisfies the residual Rohklin∗ property, i.e., the center Z(A∗∗ ) of the
second conjugate of A contains a projection P with the following properties
(a) P α(g)(P ) = 0 for all g ∈ G \ {eG }
(b) If R ≤ S are α(G)-invariant A-open central projections with R 6= S,
then there is g ∈ G with (S − R)α(g)(P ) 6= 0.
(II) For any a, b ∈ A+ , c ∈ A and ε > 0 there ared1 , d2 ∈ A and g1 , g2 ∈ G
with kd∗1 ad1 − α(g1 )(a)k < ε, kd∗2 bd2 − α(g2 )(b)k < ε, and kd∗1 cd2 k < ε.

Then A oα,r G is strongly purely infinite.

compare the following with the new observations on HarKir and

KirSira

Corollary 2.18.14. Suppose that A is stable and separable, that

h : A → M(A) is non-degenerate, faithful, and satisfies

(i) h and its infinite repeat δ∞ ◦ h are approximately unitarily equivalent in

M(A). (In particular, h(A) ∩ A = {0}.)
(iii) a ∈ span(Ah(a)A) for all a ∈ A+ .
348 2. BASICS ON PURELY INFINITE C *-ALGEBRAS

(ii) if J ∈ I(A) and h(J)A ⊆ J, then h(J) = h(A) ∩ M(A, J).

(iv) M(h) ◦ h is approximately unitarily equivalent to h in M(A). Check!!!

Then T (H) = OH for the Hilbert A–A bi-module H := A defined by h : A → M(A),

and T (H) is strongly purely infinite.
check: Moreover, h : A → M(A) defines a KK(C ; A, M(A)) equivalence of A
and T (H) = OH with respect to the matrix operator convex cone C generated by h.
This seems strange? Because M(A) is often KK-trivial?
 CHAPTER 3

Nuclear c.p. maps and operator-convex cones

We have seen in Chapter 2, among others, that simple purely infinite algebras in
the sense of J. Cuntz are purely infinite in the sense of Definition 1.2.1. Moreover we
have shown that locally purely infinite (i.e., l..p.i.) simple C *-algebras are strongly
purely infinite in the sense of Definition 1.2.2. Now we study the effect of the
property of strong pure infiniteness for C *-algebras on residual nuclear c.p. maps
between them. (Notice that strong pure infiniteness of C *-algebras and residual
nuclearity of C *-morphisms are in general completely unrelated properties.)
We prove in this Chapter 3, among others, that the ultrapower Dω and asymp-
totic corona Q(R+ , D) of strongly purely infinite algebras D (cf. Definition 1.2.2)
satisfy a local version of Weyl–von-Neumann property (WvN-property of Definition
1.2.3), cf. Proposition 3.2.15. It implies that the algebras Q(R+ , D) are strongly
purely infinite in the sense of Definition 1.2.2, because they are rich of approximately
inner completely positive maps from separable C *-subalgebras A into commutative
C *-subalgebras C that fix given elements of A ∩ C. It is a basic observation for the
later applications with help of our cone-related KK-theory and to the proof of its
equivalence with cone-related un-suspended E-theory.
The WvN-property allows to prove in Chapter 5 variants of generalized Weyl-
von Neumann theorems. We use the results of Chapters 3 and 5 in Chapter 6 for
the proof of Theorem A and for the proof of Theorem 6.3.1 that is a special case
of the, in the last Chapter 12 finally proven, Theorem K.

Remark 3.0.1. We get later results of the following form, that one can consider
as an approximate factorization over the cones of sums of matrix algebras.
Let A and B C *-algebras, U : A → B a nuclear map, X ⊂ A+ a finite subset
of the positive contractions in A and ε > 0.
Then there exist following

(a) a finite dimensional C *-algebra F ,

(b) a c.p. map V : F → B with kV k = kU k,
(c) a C *-morphism φ : C0 ((0, 1], F ) → A
(d) a contraction d ∈ A+ (almost commuting with φ(C0 ((0, 1], F )) ?)

such that kU (dad)−U (a)k < ε for all a ∈ X, and dist(dad, φ(C0 ((0, 1], F ))) < ε
for all a ∈ X, and kV (g(1)) − U (dn φ(g)dn )k < ε for all g ∈ C0 ((0, 1], F ).
349
350 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

This means: We can locally “compress” X ⊂ A – by some approximately inner

c.p. map – into a suitable C *-subalgebra φ(C0 ((0, 1], F )) of A and then apply the
given nuclear map U to this compressed subset of φ(C0 ((0, 1], F )).
This is then in a sense “almost” the restriction U to X.

HERE IS NEEDED a list of “corona” type algebras:

Q(R+ , B) (!!!), Q(X, B) (X with finite dimension??), Qs (B) (!!!), some quo-
tients of those, e.g. ultrapowers ...
FIND list of defining properties! Certainly more than sub-Stonean!?
(At present state, in the terminology of Ilijas Farah ... ).
B∞ and Bω among them?,
Desire: Should be σ-sub-Stonean.
Moreover, should satisfy Kasparov’s Technical Lemma (KTL).
It should also require that they all have a property that there are sufficiently
many “good” Abelian C *-subalgebras that allows to get approximate factorization
of separable C *-subalgebras through algebras with regular abelian C *-algebras.
Would be interesting to see how is the interplay... σ-sub-Stonean versus (KTL).
Where is the Definition ”def:res.nuc.cp.map” ?? ???: of residual nuclear (or
”residually nuclear”) c.p. maps V : A → B that are “ideal system preserving”, i.e.,
V (A ∩ J) ⊆ J for all J ∈ I(B) (in case A ⊆ B).
Or that (more generally) an map λ : J ∈ I(B) → λ(J) ∈ I(A) from the system
of ideals J ∈ I(B) of B to the system of ideals I(A) is given and we require that
V (λ(J)) ⊆ J. ????
The proof of next Theorem needs:

Lemma 3.0.2. Let A ⊆ B C*-algebras, and suppose that A or B strongly p.i.

Then each residually nuclear c.p. contraction V : A → B ( 1 ) is approximately
1-step inner, i.e., for each finite subset F ⊂ A and ε > 0 there exists a contraction
d ∈ B with kV (a) − d∗ adk < ε for all a ∈ F .

Proof. WHERE is the Def. of to be filled in ?? 

Need for later applications the following 1-step inner-ness

(in blue, to be moved to suitable place).
But perhaps we must suppose that ϕ is itself nuclear?
Define the notion of ‘‘corona type C *-algebra’’

Theorem 3.0.3. Let A an exact C*-algebra and ϕ : A ⊗ O∞ → B a C*-

morphism, and suppose that V : A → B is a nuclear c.p. map that has the following
Property (*):
1 The Definition ?? of residual nuclear c.p. maps includes that V is “ideal system preserving”,

i.e., V (A ∩ J) ⊆ J for all J ∈ I(B).

 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES 351

(*) For every a ∈ A+ , its image V (a) is contained in the closure of the alge-
braic ideal span(Bϕ(a ⊗ 1)B) of B that is generated by ϕ(a ⊗ 1).

Then there exist for every finite subset X ⊆ A and ε > 0 an element b ∈ B with
kbk2 ≤ kV k and
kb∗ ϕ(a ⊗ 1)b − V (a)k < ε for all a ∈ X .

If, in addition, A is separable and B is a “corona” type algebra, then V is

one-step “inner”, i.e., V (a) = b∗ ϕ(a ⊗ 1)b for all a ∈ A for some b ∈ B with
kbk2 ≤ kV k.
Proposition 3.0.4. If A is separable and exact, B is strongly purely infinite
(or if B is not necessarily s.p.i., but A is strongly p.i.) and ϕ : A → B is a nuclear
C*-morphism, then ϕ “extends” to a C*-morphism
ϕe : A ⊗ O∞ → Bω := `∞ (B)/cω (B)
with ϕe (a ⊗ 1) = ι(ϕ(a)) for a ∈ A and ι(b) = (b, b, . . .) + Jω for b ∈ B.
If B is a strongly purely infinite “corona” algebra
? Precise Def. of ‘‘corona C*-algebra’’ C?
Perhaps in Chp. 5, or extra section in Appendices?
There should be σ-unital D such that C is a non-zero hereditary
subalgebra of M(D)/D,
Characterize coronas by ‘‘better than sub-Stonean’’? ,
then ϕ itself extends to a C*-morphism ψ : A ⊗ O∞ → B with ψ(a ⊗ 1) = ϕ(a)
for a ∈ A.

Transfer Remark 3.0.5 to suitable place?!

Remark 3.0.5. Operator-matrix calculations appear in most chapters. We
identify often the ternary rings Mp,q (A) ⊆ Mn (A) with max(p, q) ≤ n with the
tensor products Mp,q ⊗A. The results of calculations in both of them look sometimes
different if we do not care about different interpretations of this calculations.
This difference between this two conventions appears because the correct mul-
tiplication r · c ∈ e11 ⊗ A of r ∈ M1,n ⊗ A and c ∈ Mn,1 ⊗ A comes from the iden-
tification of columns in Mn,1 (A) with tensors in Mn,1 ⊗ A and of rows in M1,n (A)
with tensors in M1,n ⊗ A.
Then M1,n (A) · Mn,1 (A) = M1,1 (A) = A but (M1,n ⊗ A) · (Mn,1 ⊗ A) = e1,1 ⊗ A
considered in Mn ⊗ A. If we use in applications both of them, then we have in some
cases to observe that we must identify a ∈ A with e11 ⊗ a ∈ e11 ⊗ A.
In other calculation we have to identify a ∈ A with 1n ⊗ a in Mn ⊗ A. and Mn
with Mn (C · 1M(A) ) ⊆ M(Mn (A)) .
If the multiplier algebra M(A) contains a copy of O2 with 1M(A) ∈ O2 , and all
elements have to be considered only up to MvN-equivalence, then such differences
are usually not important.
352 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

1. Exact C *-algebras and nuclear maps

We recall some definitions and basic facts concerning nuclear maps, exact C *-
algebras and nuclear C *-algebras. A C *-algebra B is called nuclear if its algebraic
tensor B C product with any other C *-algebra C admits only one C *-algebra
norm ( 2 ).
U. Haagerup showed in [340] with help of results of A. Connes in [159] that
a C *-algebra is nuclear if and only if it is amenable ( 3 ). We don’t use the (co-
homological) amenability here, because we work here with special classes of nuclear
maps, and therefore, we use characterizations of nuclear maps and nuclear C *-
algebras that we recall in Remark 3.1.2(i).
A C *-algebra B is called exact if, for every short exact sequence of C *-
algebras,
0 → J → A → A/J → 0 ,
the spatial (or “minimal”) tensor products with B of this sequence,
0 → J ⊗ B → A ⊗ B → (A/J) ⊗ B → 0
is again exact, i.e., if the closure J ⊗B of the algebraic tensor product J B in A⊗B
is the same as the kernel of the natural C *-algebra epimorphism A⊗B → (A/J)⊗B.
With other words: The algebra B is exact if the functor A 7→ A ⊗ B is a short
exact functor in the category of C*-algebras.
Every nuclear C *-algebra B is exact, because the functor A 7→ A ⊗max B is
short-exact for every C *-algebra B by the “universality” of the maximal C *-tensor
product A ⊗max B ( 4 ).

Definition 3.1.1. Let A and B be C *-algebras and let V : A → B be a

completely positive map. The map V will be called factorable if it factorizes
through Mn for some n, i.e., if there exist completely positive maps S : A → Mn
and T : Mn → B such that the diagram

A
V
/B
>

S T
Mn
commutes.
By definition, a map V : A → B is nuclear if V can be approximated in point-
norm topology by factorable maps (i.e., is the point-wise limit of a net of factorable
maps).
2 It was called property (T) by Takesaki [766].
3 The C *-algebra B is amenable if every derivation of an dual Banach B-bi-module X ∗ is

inner, i.e., if X is a Banach B-bi-module and if ∂ : B → X ∗ satisfies ∂(ab)(x) = ∂(b)(xa)+∂(a)(bx)

for all a, b ∈ B and x ∈ X, then there is ρ ∈ X ∗ with ∂(a)(x) = ρ(ax − xa) for all a ∈ B, x ∈ X.
4 The algebraic sequence 0 → J B → A B → (A/J) B → 0 of algebraic tensor products

is always exact, and the same holds then automatic for the maximal C *-algebra tensor product
⊗max – in place of – by universality and uniqueness of bi-functor .
 1. EXACT C *-ALGEBRAS AND NUCLEAR MAPS 353

We say that a completely positive map V : A → M from a C *-algebra into a

von-Neumann algebra M is weakly nuclear if it can be approximated in point-
σ(M, M∗ ) topology by factorable maps.

The next remark collects a few key observations concerning nuclearity and
exactness.

Remark 3.1.2. In the following denotes CP(A, B) the (point-norm) closed

convex cone of completely positive maps from A into B. The algebraic tensor
product and the minimal (respectively maximal) C *-algebra tensor product of C *-
algebras will be denoted respectively by A B and A⊗B (respectively by A⊗max B),
corresponding to the minimal C *-norm k · kmin (respectively to the maximal C *-
norm k · kmax on A B).
(o) The set CPf (A, B) of factorable c.p. maps V : A → B is a convex cone in
the space L(A, B), – the bounded linear maps from A to B with finite-dimensional
image in B –, but with the important additional property that the map a 7→ c∗ V ⊗
idk (r∗ (a)r) c is in CPf (A, B) for every k ∈ N, every row matrix r ∈ M1,k (M(A))

and every column matrix c ∈ Mk,1 (M(B)) :

Indeed: Let k = 1, 2, Vk = Tk Sk with c.p. maps Sk : A → Mnk and Tk : Mnk → B.
Then V1 + V2 = T3 S3 with
S3 : A 3 a 7→ S1 (a) ⊕ S2 (a) ∈ Mn1 ⊕ Mn2 ⊆ Mn3 ,
and suitable c.p. maps T3 : Mn3 → B for n3 := n1 + n2 , e.g. for T3 := R ◦ P
where P denotes the conditional expectation from Mn3 onto Mn1 ⊕ Mn2 ⊂ Mn3 ,
and R(c ⊕ d) := V1 (c) + V2 (d) for c ⊕ d ∈ Mn1 ⊕ Mn2 ).
The set CPf (A, B) is a matrix operator-convex cone in sense of Definition 3.2.2
and can be described as follows:
If V = T S, then V ⊗ idm = (T ⊗ idm )(S ⊗ idm ) is in CPf (A ⊗ Mm , B ⊗ Mm ) ,
(S ⊗ idm )(r∗ (·)r) is in CP(A, Mnm ) for r ∈ A ⊗ Mn and c(T ⊗ idm )(·)c∗ is in
CP(Mnm , B) for c ∈ B.
Notice that CPf (A, B) is contained in the set CPf.r. (A, B) ⊆ CP(A, B) of
c.p. maps of finite rank inside CP(A, B). (But it is still not clear if all c.p. maps
V : A → B of finite vector space dimension dim(V (A)) < ∞ are contained in
CPf (A, B)).
The operator-norm closure of CPf (A, B) in the Banach space of general
bounded linear maps L(A, B) ( 5 ) contains all c.p. maps V : A 7→ B of finite rank
cf. Lemma 3.1.11(iv).
The finite rank self-adjoint linear maps from A to B are differences of c.p. maps
of finite rank (using Lemma 3.1.4).
(i) Simple Hahn-Banach separation arguments (cf. proof of [426, lem. 2] or similar
arguments in Section 7) show, that a completely positive map V : A → B is nuclear,
5 This “uniform” closure of CP (A, B) is usually smaller than the point-norm closure, and
f
therefore does not contain all nuclear maps in general.
354 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

if and only if, for every C *-algebra D,

V ⊗max idD : A ⊗max D → B ⊗max D

annihilates the kernel of the natural epimorphism A⊗max D → A⊗min D, (cf. [140],
[726, prop. 1.2]). I.e., V ∈ CP(A, B) is nuclear, if and only if, V ⊗max idD naturally
factorizes over the spatial (= minmal) C*-algebra tensor product A ⊗ D.
It suffices to consider unital singly generated D (e.g. by taking stabilization
and unitization). If we let G := C ∗ (F2 ), the full group C *-algebra of the free group
F2 on two generators. Then D ∼ = G/J for a suitable closed ideal J . G.
Every C *-semi-norm µ on the algebraic tensor product A G with µ(a⊗1) > 0
for a 6= 0 is given on c ∈ A G by µ(c) = k(id ⊗πJ )(c)kβ for suitable J . G and
some C *-norm k · kβ on A (G/J) with k · kmin ≤ k · kβ ≤ k · kmax , i.e., the
kernel of A ⊗max D → A ⊗β D is contained in the kernel of A ⊗max D → A ⊗min D
for D = G/J ( 6 ). Now notice that B ⊗max (·) is always a short exact functor, i.e.,
that B ⊗max (J) is the kernel of B ⊗max G → B ⊗max (G/J). If we combine the
above observations using a diagram check, then we can reformulate the tensorial
nuclearity criterion equivalently by the following more flexible criterion:
A map V ∈ CP(A, B) is nuclear, if and only if, (V ⊗max id)(I) is contained in
B⊗ max
J for every closed ideal I of A ⊗max C ∗ (F2 ) with I ∩ (A ⊗ 1) = {0} and the
largest ideal J of C ∗ (F2 ) with A ⊗max J ⊆ I. ( 7 ).
In particular, the identity map idA on a C*-algebra A is nuclear if and only if
A is nuclear ([140], [426], [145]).
(ii) A C *-algebra A ⊆ L(H) is exact, if and only if, the inclusion map A ,→ L(H)
is nuclear. We gave two different proofs in [432] and in [438]. The proof in [432]
uses part(i), a simple intersection result (less general than the “distance” lemma
[438, lem. 3.6]), an inspection of kernels in certain commutative diagrams, and the
fact that on the algebraic tensor product L(H) C ∗ (F ) of L(H) with the full group
C *-algebra C ∗ (F ) of the free groups F on countably many generators there is only
one C *-norm. The authors proof in [432] of the uniqueness of the C *-norm on
L(H) C ∗ (F ) has been simplified considerably by G. Pisier [640].
(iii) The composition of a nuclear map with any other completely positive map is
again nuclear. In particular:

(a) W = b∗ V (·)b is nuclear if V is nuclear.

(b) The restriction V |D : D → C is nuclear if D ⊆ A is a C *-subalgebra and
V : A → C is nuclear.
Therefore from (i) and (ii) we see, that nuclear C*-algebras are exact
and that C*-subalgebras of exact C*-algebras are again exact.

6 It follows from [766]: A C *-seminorm ν on A D satisfies ν(c) ≥ kck

min for all c ∈ A D ,
if and only if, ν(a ⊗ d) > 0 for each non-zero a ∈ A+ and d ∈ D+ . See also [767, thm. IV.4.19].
7 It is a special case of a characterization of point-norm closed m.o.c. cones C ⊆ CP(A, B)

by its associated action ΨC : A ⊗max C ∗ (F2 ) 7→ A ⊗max C ∗ (F2 ) cf. Section 7.

 1. EXACT C *-ALGEBRAS AND NUCLEAR MAPS 355

(c) Every completely positive map from A to B is nuclear if A or B is nuclear,

cf. Part (i).
(d) Every completely positive map from A to B is nuclear if A is exact and B
has the weak expectation property (i.e., equivalently, B is weakly injective
in the sense: ψ⊗max idC : B⊗max C → D⊗max C for every *-monomorphism
ψ : B → D and every C *-algebra C).
Indeed, let A ⊆ L(H) and B ⊆ L(H) (for sufficiently large H and with
embedding maps ηA and ηB ), then a c.p. contractions V : A → B extends
to a c.p. contraction T : L(H) → L(H) by the Arveson extension theorem
(complete injectivity of L(H), cf. [42]), i.e., T ◦ ηA = ηB ◦ V for the
inclusion maps ηA and ηB . Then T ◦ ηA is nuclear by (ii). The nuclearity
criteria in part (i) and the above given criteria for the weak injectivity of
B show that V : A → B is nuclear.
(e) In general a completely positive map V : A → B ⊆ C is not nuclear as a
map from A to B if V ∈ CPnuc (A, C), i.e., where V becomes nuclear if
considered as a map from A to C.
(Use part (ii) and the existence of non-nuclear exact C *-algebras A, as
∗
e.g. V := idA , A = B = Cred (G) ⊆ L(`2 (G)) =: C for G := SL2 (Z).)
(f) Every c.p. map V : A → B of finite rank, i.e., with Dim(V (A)) < ∞, is
the limit in operator-norm of factorable maps, i.e., there exists a sequence
of factorable maps Tn : A → B with limn kTn − V k = 0 in the norm of
L(A, B), cf. [426], see also Lemma 3.1.11(iv).
(g) It follows from part (f) that the definition of nuclear c.p. maps given in
[426] and that of M.D. Choi and E.G. Effros in [140] define the same
class of maps.

(iv) There exist c.p. maps W : A/J → B such that V := W ◦πJ : A → B is nuclear,
but the map W itself is not nuclear.
This follows from the existence of W*-algebras that are not weakly exact (compare
Remark B.7.4, [436], [597]).
It is a problem to find non-trivial sufficient criteria on (A, J / A, B) that allow
to conclude the nuclearity of W : A/J → B from the nuclearity of W ◦ πJ : A → B.
We use non-exact subalgebras of the asymptotic corona B = Q(R+ , D), – e.g. in our
proofs on the classification of nuclear algebras. Those algebras are often not locally
reflexive, because Q(R+ , D) is locally reflexive, if and only if, D is sub-homogenous.
Local reflexivity is formally weaker than exactness of a C *-algebra A, i.e.,
exactness implies that A is locally reflexive. But it was still unknown (June 2019)
if local reflexivity implies exactness.
Therefore, we must bypass the problem concerning the nuclearity of [V ]J =
W : A/J → B for nuclear V : A → B with V (J) = {0}, and have introduced
the notion of Ψ-residually nuclear maps, cf. Definition 1.2.8, and recognize that
it is of interest for us (!) to know when Ψ-equivariant and nuclear c.p. maps are
automatically Ψ-residually nuclear maps.
356 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

(v) Here is a very special case, where one has a positive answer to the problem
discussed in Part(iv):
Let J / A a closed ideal and suppose that J ⊗ D → A ⊗ D → (A/J) ⊗ D is exact
for every unital separable C *-algebra D ( 8 ), then it holds:

(∗) If W ∈ CP(A/J, B) and W ◦ πJ is nuclear – i.e., W ◦ πJ ∈ CPnuc (A, B),

then W is nuclear.

Local lifting results of [238] for locally reflexive C *-algebras A show that the
sequence
J ⊗ D → A ⊗ D → (A/J) ⊗ D
is exact for every locally reflexive C *-algebra A, every ideal J . A and every C *-
algebra D. Hence:
W ∈ CP(A/J, B) is nuclear if A is locally reflexive and W ◦ πJ : A → B is
nuclear.
One can see that exact C *-algebras are locally reflexive, because the separable
exact C *-algebras are quotients of subalgebras of the nuclear CAR-algebra M2∞
by [438, cor. 1.4(v)], and sub-quotients of locally reflexive C *-algebras are locally
reflexive by [238].
In the mean time the author has obtained the following ?????

Proof of (∗). If W ◦ πJ is nuclear, then, by Part(i),

(W ◦ πJ ) ⊗max id : A ⊗max D → B ⊗max D
(naturally) factorizes over A⊗D, in the sense that (W ◦πJ )⊗max id maps the kernel
ideal of A⊗max D → A⊗D to zero. Thus, there is a c.p. map V : A⊗D → B ⊗max D
with V ◦λA,D = (W ◦πJ )⊗max id for the natural *-epimorphism λA,D : A⊗max D →
A ⊗ D . It follows that V (a ⊗ d) = W (a + J) ⊗ d for a ∈ A and d ∈ D. It yields
V (J ⊗ D) = {0}. The exactness of J ⊗ D → A ⊗ D → (A/J) ⊗ D implies that
there is a C *-morphism β : (A/J) ⊗ D → B ⊗max D with β ◦ (πJ ⊗ idD ) = α, i.e.,
β((a + J) ⊗ d) = W (a + J) ⊗ d for a ∈ A, d ∈ D. Thus, the natural c.p. map
W ⊗max id : (A/J) ⊗max D → B ⊗max D factorizes over (A/J) ⊗ D.
It implies that W : A/J → B is nuclear by Part(i). 

(vi) Quotients A/J of exact C *-algebras A are exact.

Indeed: If A/J ⊆ L(K) and A ⊆ L(H), then, by the Arveson extension theorem,
cf. [42], there is a complete contraction T : L(H) → L(K) with T (a) = πJ (a), i.e.,
T ◦ηA = ηA/J ◦πJ , where η indicates the inclusion map. The map ηA : A ,→ L(H) is
nuclear by (ii), the maps ηA/J ◦ πJ and ηA/J : A/J ,→ L(K) are nuclear by Remark
(v). See [436] for a different proof of exactness for quotients of exact C *-algebras.
(vii) A C *-algebra B is nuclear, if and only if, A is exact and the natural *-
epimorphism A ⊗max C ∗ (F2 ) → A ⊗min C ∗ (F2 ) is an isomorphism.
8 The sequence J ⊗ D → A ⊗ D → (A/J) ⊗ D is exact for D := L(` ) if and only if id
2 A/J is
locally completely contractive liftable, cf. [238].
 1. EXACT C *-ALGEBRAS AND NUCLEAR MAPS 357

Indeed: It induces that A ⊗max D → A ⊗min D is an isomorphism for for every

hereditary C *-subalgebra D of C ∗ (F2 ). The exactness of A ⊗min (·) and the general
exactness of (·) ⊗max (·) imply that the natural C *-morphisms A ⊗min D/J →
A ⊗max D/J are isomorphisms for each closed ideal of D. It is not difficult to see
that each separable C *-algebra is isomorphic to a quotient of a hereditary C *-
subalgebra C ∗ (F2 ). Thus, A ⊗max B → A ⊗min B is an isomorphism for every
separable C *-algebra B. This carries over to all C *-algebras B, i.e., A is nuclear.

Remark 3.1.3. In in late 2015 the author has obtained the following character-
ization of locally reflexive separable C *-algebras B by the following “co-exactness
criterium” (unpublished but presented in talks of the author):
A separable C*-algebra B is locally reflexive, if and only if, for every closed left
ideal L ⊆ K ⊗ B and every separable C*-algebra C the sequence

L ⊗min C → (K ⊗ B) ⊗min C → ((K ⊗ B)/L) ⊗min C

is exact. That means equivalently that the natural linear contraction

((K ⊗ B) ⊗min C)/(L ⊗min C) 7→ ((K ⊗ B)/L) ⊗min C

is an isomorphism of Banach spaces.

So far no example of a non-exact separable locally reflexive C *-algebra has
been found. But it is important to check this.

For completeness we give the almost obvious separation on finite sets of linear
functionals on C *-algebras.

Lemma 3.1.4 (Separation lemma). If ρ1 , . . . , ρn are continuous linear function-

als on a C*-algebra A, then there is a cyclic *-representation d : A → L(H) with
cyclic vector ξ ∈ H of norm kξk = 1, and operators t1 , . . . , tn ∈ d(A)0 , such that
ρk (a) = hd(a)ξ, tk ξi for a ∈ A and k = 1, . . . , n .
The operators tk ∈ d(A)0 are uniquely determined by (d, ξ, ρk ).
If A is a W*-algebra and if ρ1 , . . . , ρn are normal, then d : A → L(H) and can
be taken normal, i.e., such that the functionals hd(·)x, yi are in the predual A∗ of
A.

Proof. Since A∗ is the linear span of the positive linear functionals on A,

P
we may suppose that the ρk are positive and ρk 6= 0. Let λ := k ρk . The
cyclic GNS-representation d : A → L(H) of A with respect to λ has cyclic vector
y ∈ H with kyk2 = kλk and satisfies λ(a) = hd(a)y, yi for a ∈ A. In particular,
ρk (a∗ a) ≤ kd(a)yk2 . It follows, that there are unique positive contractions Ck ∈
L(H) with ρk (c∗ b) = hd(b)y, Ck d(c)yi, and the uniqueness of the Ck implies that
Ck d(a) = d(a)Ck for a ∈ A, (cf. [217, prop. 2.5.1]). Now let ξ := kλk−1/2 y and
tk := kλkCk .
If A is a W*-algebra and the ρj are normal, then we can apply the same
arguments to the ρ1 , . . . , ρn inside the predual A∗ of A. 
358 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

The next lemma contains a useful criteria for the nuclearity of a c.p. map
V : A → B, that will be used later (e.g. in Chapter 6).
Say precisely where is it used and cited ? in chp.6 ? Give, or cite, or define
here what means ”essential ” here.

Lemma 3.1.5. Suppose that D ⊆ B is an essential hereditary C*-subalgebra of

B with (open) support projection pD ∈ B ∗∗ , and that V : A → B is a completely
positive map.
If (1 − pD )B ∗∗ (1 − pD ) is an injective W*-algebra and if a ∈ A 7→ Vd (a) :=
dV (a)d ∈ D is a nuclear map for all d ∈ D+ , then V : A → B is nuclear.

Proof. We use that every c.p. map W : A → M from a C *-algebra A to an

injective W*-algebra M is weakly nuclear, because
??? the identity map idM is weakly nuclear for all injective W*-algebras M ,
??? only for all ???
and that a c.p. map V : A → B is nuclear if V : A → B ∗∗ is weakly nuclear.
The inclusion map for A ⊆ L(H) is always weakly nuclear ??? Here is some
misunderstanding ...
A better argument should be: If (1 − pD )B ∗∗ (1 − pD ) is injective (as vN-
algebra), then there exists a central projection Q ≥ (1 − pD ) in B ∗∗ such that
B ∗∗ Q is injective. Then the kernel of b 7→ bQ in a closed ideal J of B with J ⊆ D.
But it is not clear if ?????
Indeed, by Definition 3.1.1, each weakly nuclear map is in the point-σ(B ∗∗ , B ∗ )
topology the point-wise limit of factorable maps from A to B ∗∗ and those can be
again point-wise approximated in σ(B ∗∗ , B ∗ ) topology by factorable maps from A
to B . Thus, we can then use a Hahn-Banach separation argument to see that
V : A → B can be approximated in point-norm by factorable maps from A to B.
We show the weak nuclearity of V : A → B ∗∗ :
Let Q denote the smallest central projection of B ∗∗ with (1 − pD ) ≤ Q. Then
(1 − Q) ≤ pD and B ∗∗ Q is an injective W*-algebra.
Thus, V (a) = (1 − Q)V (a) + QV (a) is the sum of the weakly nuclear maps
a 7→ QV (a) (by injectivity of B ∗∗ Q) and of (1 − Q)V (a) = (1 − Q)pD V (a)pD (that
is an weak limit of a 7→ (1 − Q)dV (a)d with contractions d ∈ D+ ).
Hence, V is weakly nuclear – if considered as a c.p. map from A to B ∗∗ . 

Remark 3.1.6. Any positive linear map V from a unital complex C *-algebra
B into a C *-algebra C ⊆ L(H) has norm kV k = kV (1B )k, cf. [402, exercise
10.5.10] (using [402, exercise 10.5.4]), or use the original Russo-Dye theorem [222]
for complex C *-algebras to reduce the calculation of the norm kV k to the case of
the special complex C *-algebra B := C(S 1 ), where all positive V are automatically
completely positive. The proof of [207, thm. I.8.4] shows the stronger result that
 1. EXACT C *-ALGEBRAS AND NUCLEAR MAPS 359

set of all convex combinations of finitely many unitaries in a unital C *-algebra A

contains all a ∈ A with kak < 1. This is also shown in [402, exercise 10.5.92].
In the case of real C *-algebras C acting on real Hilbert spaces one has at
least kV k ≤ kV2 (1B ⊗ 12 )k in the case that V2 := V ⊗ idM2 (R) is positive, because
V is “real” 2-positive, and it suffices for real C *-algebras B to consider the case
C := M2 (R) .
If a map V is moreover completely positive then this implies that kV (1)k =
kV k = k idn ⊗V k = kV kcb := supn k idn ⊗V k . Here idn denotes the identity map
on Mn and the norm on Mn ⊗ C ⊆ Mn ⊗ L(H) ∼ = L(`2 (n) ⊗ H) is the operator
norm.
The complete positivity of a map V ∈ CP(Mn , A) implies that (idn ⊗V )(X) ≥ 0
for all X ∈ (Mn ⊗ A)+ , where we naturally identify Mn (A) with Mn ⊗ A.
The generalized Schwarz inequality (of R.V. Kadison [399]) says that, if a
linear map η : A → B is positive (i.e., if η(A+ ) ⊆ B+ ) and kη ∗∗ (1A )k ≤ 1, then
η(a)2 ≤ η(a2 ) for all a ∈ As.a. .

!! Compare with other remarks above/below !! e.g next.

Remark 3.1.7. We have often to do with positive maps T : C → A between

C *-algebras. This maps T are bounded:
Indeed, otherwise for each n ∈ N there would exist cn ∈ C+ with kcn k = 1 and
P −n
kT (cn )k ≥ 4n , and the well-defined element S := n2 cn ∈ C+ must have the
−n n
– impossible – property kT (S)k ≥ 2 kT (cn )k ≥ 2 for all n ∈ N.
Another often used elementary fact for approximations of positive maps
T : C → A used special case is the following property:
Let C ⊆ M(A) a C *-subalgebra, S a set of positive maps from C into A with
the property that S(e(·)e) ∈ S for all S ∈ S and e ∈ C+ . If T : C → A can
be approximated in point-norm topology by S ∈ S, then T is positive and one
finds also a point-norm approximation of T by S ∈ S with the additional property
kSk ≤ kT k, cf. Lemma 3.1.8.

The following Lemma 3.1.8 applies to to the particular case of the family of n-
step approximately inner c.p. maps V : C → A and to the approximation of nuclear
maps by more special factorable maps.
See also Lemma B.16.5(i) in Appendix B.

Lemma 3.1.8. Suppose that C is a C*-algebra and B is a C*-algebra (respec-

tively that B is a von-Neumann algebra). Denote by S a set of positive linear maps
from a C*-algebra C into B.
If V : C → B is a contraction in the closure of S with respect to the point-
norm topology (respectively with respect to the point-strong topology), then V is in
the closure of the set of positive maps
S1 := {S(e(·)e) ; S ∈ S, e ∈ C+ , kek < 1 , kS(e2 )k ≤ 1 }
360 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

with respect to the point-norm topology

(respectively then V is in the closure of the set of positive maps

S2 := {f S(e(·)e)f ; S ∈ S, e ∈ C+ , kek < 1, f ∈ B+ , kf k ≤ 1 kf S(e2 )f k ≤ 1 }

with respect to the point-*strong topology).

Notice for later applications of Lemma 3.1.8 that the maps in S1 and S2 are
contractions, but that we do not assume any sort convexity for the given set S.
But in cases where we can consider convex sets S, the following general fact in
Functional analysis applies and improves Lemma 3.1.8 a bit:
If the point-norm closure of S ⊆ L(C, B) is convex, then the point-σ(B, B ∗ ) closure
of S coincides with its point-norm closure, (respectively the point-σ(M, M∗ ) closure
of bounded convex S ⊆ L(C, M ) coincides with its point-*strong closure of S).

Proof. Case, where B is a C *-algebra and V : C → B is in the point-norm

closure of S:
Let c1 , . . . , cn ∈ C and ε > 0, M := max(kc1 k, . . . , kcn k), and η := ε/(2 + M ).
There exists d ∈ C+ with kdk < 1 such that kdcj d − cj k < η for j = 1, . . . , n.
There is S ∈ S with kS(d2 )k ≤ kV (d2 )k + η and kS(dcj d) − V (dcj d)k < η. Let
t := (1 + η)−1/2 , e := td and T (b) := S(ebe) for b ∈ C + C · 1. Then kT k =
kT (1)k = t2 kS(e2 )k ≤ 1, because T is positive. The map T |C is in S1 and

kT (cj ) − V (cj )k ≤ 2t2 η + (1 − t2 )M ≤ ε for j = 1, . . . , n .

Case, where B ⊆ L(H) is a von-Neumann algebra and V : C → B is in the

point-strong closure of S:
Suppose that B ⊆ L(H), x1 , . . . , xm ∈ H, c1 , . . . , cn ∈ C, and ε > 0. Let µ :=
2 max(1, kx1 k, . . . , kxm k), and c0 := 1 ∈ B. There are e ∈ C+ with kek < 1 and
kck − eck ek < ε/µ. By assumption, there is a directed net (Sγ ) of elements of S
such that Sγ (eck e) converges strongly to V (eck e) for k = 0, 1, . . . , n.
The function h : t ∈ R → (0, ∞) with h(0) := 1 and h(t) := min(|t|−1/2 , 1) for
t 6= 0 is in C0 (R)+ . Thus, fγ := h(Sγ (e2 )) ≤ 1 converges strongly to h(Vγ (e2 )) =
1 by [400, thm. 5.3.4]. Moreover, fγ Sγ (e2 )fγ ≤ 1 . It follows that the map
Tγ (c) := fγ Sγ (ece)fγ (for c ∈ C + C · 1) is in the set S2 , and has the property
that Tγ (ck ) converges strongly to V (eck e) for k = 1, . . . , m. Thus there is T ∈ S2
with kT (ck )xj − V (ck )xj k ≤ ε/2 + kck − eck ekkxj k < ε for j = 1, . . . , m and
k = 1, . . . , n. 

Proposition 3.1.9. Let ejk ∈ Mn denote the matrix units of Mn for n ∈ N

and let A denote a C*-algebra.
There is a bijective and additive relation between the cone of positive matrices
[ajk ] ∈ Mn (A)+ and the matrix operator-convex cone of all completely positive maps
V : Mn → A :
 1. EXACT C *-ALGEBRAS AND NUCLEAR MAPS 361

If [ajk ] ∈ Mn (A)+ is given then a completely positive map V ∈ CP(Mn , A) is

defined by
X n
V (β) := βjk ajk for all β ∈ Mn .
j,k=1
If a completely positive map V ∈ CP(Mn , A) is given, then a positive matrix
[aj,k ] ∈ Mn (A)+ is defined by its entries ajk := V (ejk ). Their norms estimate
each other by
kV kcb = kV k = ka11 + a22 + · · · + ann k and k[ajk ]k ≤ nkV k .
Following Parts (o)-(iv) show more details:

(o) If ejk ∈ Mn denote the n2 canonical matrix-units and if V : Mn → A is

a completely positive map from Mn into a C*-algebra A , then the linear
map idn ⊗V ∈ L(Mn ⊗ Mn , Mn ⊗ A) is a positive map from Mn ⊗ Mn to
Mn ⊗ A . In particular, the matrix [V (ejk )] = (idn ⊗V )([ejk ]) =: [ajk ] is
positive in the C*-algebra Mn (A).
(i) If [ajk ] ∈ Mn (A) is a positive element in the C*-algebra Mn (A), then
there exists column matrices c1 , . . . , cn ∈ Mn,1 (A) such that the completely
positive map V ∈ CP(Mn , A) defined by
n
X
V (β) := c∗k βck for all β ∈ Mn
k=1
P
satisfies V (ejk ) = ajk for j, k ∈ {1, . . . , n} . Thus, V (β) = j,k βj,k ajk
and kV k = kV kcb = ka11 + · · · + ann k .
Pn
It implies that k k=1 c∗k ck k = kV k ( 9 ).
(ii) Suppose that A has the property that for every positive contraction a ∈ A+
and ε > 0 there exist elements b1 , b2 ∈ A with kb∗j bk − δjk ak < ε for
j, k ∈ {1, 2} ( 10 ).
Then each V ∈ CP(Mn , A) can be “orthogonal and elementary ap-
proximated in norm” – in the sense that for each γ ∈ (0, 1) there exist
columns d0 , d1 ∈ Mn,1 (A) with d∗0 βd1 = 0 and d∗0 βd0 = d∗1 βd1 for all
β ∈ Mn , and with the property that the completely positive map U (β) :=
d∗0 βd0 (β ∈ Mn ) satisfies that V − U is a completely positive map with
kV − U k < γ . In particular, U (1n ) ≤ V (1n ) and kd0 k = kd1 k ≤ kV k1/2 .
(iii) If A = D/J for a C*-algebra D and a closed ideal J of D, then there
exists a c.p. map T : Mn → D with πJ ◦ T = V and kT k = kV k.
(iv) Let A ⊆ N a C*-subalgebra of a W*-algebra N that is weakly dense in N
and let T : Mn → N a c.p. contraction.
Then there exists a net {Vρ } of c.p. contractions Vρ : Mn → A that
converges in point-*ultrastrong topology to T .

Part (ii) applies e.g. if the unit element of M(A) is properly infinite or if A
is purely infinite, cf. Lemma 2.1.7(ii). Equivalent formulations of the additional
9 Here we identify M naturally with M (C · 1) ⊆ M (M(A)).
n n n
10 We don’t require here that b b∗ ∈ aAa .
j j
362 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

assumption in Part (ii) can be found in Parts (b, i) and (b,iii) of Lemma 2.1.12.
We use in our proof also the equivalence of Parts (a, i) and (a,ii) of Lemma 2.1.12.

Proof. Let M(A) denote the C *-algebra of two-sided multipliers of A. We

identify naturally Mn (A) ⊆ Mn (M(A)) with Mn ⊗ A ⊆ Mn ⊗ M(A) and Mn =
Mn (C) with Mn ⊗ 1 ⊆ Mn ⊗ M(A). ( 11 ).
(o): We identify sometimes Mn (A) with Mn ⊗ A naturally. By Remark 3.1.6,
the complete positivity of V ∈ CP(Mn , A) implies that kV (1n )k = kV k = k idn ⊗V k
and (idn ⊗V )(X) ≥ 0 for all X ∈ (Mn ⊗ A)+ .
Recall that the matrix units eij := e∗i ej ∈ Mn , for i, j = 1, . . . , n, are build
from row-matrices
e1 := [1, 0, 0, . . . , 0] , e2 := [0, 1, 0, . . . , 0] , . . . , en := [0, 0, . . . , 0, 1]
that are the elements of the canonical basis of M1,n (C) ∼
= Cn .
The n2 ×n2 matrix E ∈ Mn (Mn ) with the n×n matrices eij ∈ Mn as its entries
P
is given by E := [eij ] ∈ Mn (Mn ) and can be expressed as E = ij eij ⊗ eij ∈
Mn ⊗ Mn ∼ = Mn (Mn ). It is positive in Mn2 ∼ = Mn ⊗ Mn because E = R∗ R for
the row-matrix R := [e1 , e2 , . . . , en ] ∈ M1,n2 , the norm of E is kEk = n because
P := n−1 E is an orthogonal projection.
We define a matrix F = [fjk ] ∈ Mn (A) ∼= Mn ⊗ A by taking its entries fij :=
V (eij ), i.e., define
X
F := (idn ⊗V )(E) = eij ⊗ V (eij ) ∈ Mn ⊗ A .
i,j

Then F ∈ Mn (A)+ is a positive matrix by the assumption of complete positivity of

Pn
V , and V (β) = j,k=1 βjk fj,k for β ∈ Mn , because P := n−1 E is an orthogonal
projection in Mn ⊗ Mn . The norm of F can be estimated roughly by
kF k ≤ nk idn ⊗V k = nkV (1n )k = nkf11 + · · · + fnn k .

(i): Let F any positive matrix in Mn (A)+ with entries fjk ∈ A, and let
G := F 1/2 ∈ Mn (A)+ with entries gij ∈ A. We define a map V : Mn → A by
Pn
V (ejk ) := fj,k , i.e., V (β) := j,k=1 βj,k fj,k . The complete positivity of the below
shown alternative expression for the above defined map V : Mn → A, – as sum of
elementary c.p. maps β 7→ c∗ βc with some c ∈ Mn,1 (A) –, implies then that V must
be completely positive.
In conjunction with Part (i), this shows moreover:
Each completely positive map V ∈ CP(Mn , A) can be expressed as sum of n maps
of the form β 7→ c∗ βc with columns c ∈ Mn,1 (A).
(The ck ∈ Mn,1 (A) are not uniquely determined by this property.)
We can rewrite the matrix G ∈ Mn (A)+ in the notation for elements of Mn ⊗ A
as G = i,j eij ⊗ gij ∈ Mn ⊗ A . The equations G∗ = G and G2 = F imply that
P

11Matrices b = [b ] ∈ M (M(A)) with entries b

jk n jk ∈ C · 1 can be written as b = β ⊗ 1 with
unique β ∈ Mn .
 1. EXACT C *-ALGEBRAS AND NUCLEAR MAPS 363

∗
Pn
gij = gji and k=1 eij ⊗ gik gkj = eij ⊗ V (eij ), i.e.,
n
X
∗
fij = V (eij ) = gki gkj .
k=1

We define columns c1 , c2 , . . . , cn ∈ Mn,1 (A) ∼

= Mn,1 ⊗ A ⊆ Mn ⊗ A by

ck := [gk,1 , gk,2 , . . . , gk,n ]> ∈ Mn,1 (A) ,

i.e., take here the transposes of the rows rk = [gk,1 , gk,2 , . . . , gk,n ] in M1,n (A) of the
n × n-matrix G ∈ Mn (A) ( 12 ).
In the tensorial terminology with a ∈ A identified with e11 ⊗a it can be rewritten
P
as ck = j ej1 ⊗ gk,j .
It gives that c∗k (eij ⊗ 1)ck = e11 ⊗ (gk,i
∗
gk,j ) = e11 ⊗ gik gk,j for k ∈ {1, . . . , n},
Pn ∗
and implies e11 ⊗ V (eij ) = k=1 ck (eij ⊗ 1)ck for i, j ∈ {1, . . . , n} .
Notice that the products c∗j Xck are in e11 ⊗A for all X ∈ Mn (A) ∼
= Mn ⊗A if we
P
consider columns cj as element of Mn,1 ⊗A, i.e., as the element j ej1 ⊗gk,j ∈ Mn .
Then c∗k eij ⊗ 1M(A) ck = e11 ⊗ gi,k gk,j and, therefore,


X
c∗k β ⊗ 1M(A) ck = e11 ⊗ V (β) .

k
Pn
Hence e11 ⊗ V (β) = k=1 c∗k (β ⊗ 1M(A) )ck for β ∈ Mn .
If we identify Mn with Mn (C1) ⊆ Mn (M(A)) = M(Mn (A)), then c∗k eij ck =
∗
gi,k gk,j . Thus, k c∗k eij ck = k gi,k gk,j = V (eij ), which implies that
P P
gk,i gk,j =
P ∗
k ck βck= V (β) for all β ∈ Mn , i.e., the conventional interpretation says that
there are columns c1 , . . . , cn ∈ Mn,1 (A) that satisfy:
n
X
V (β) = c∗k βck for all β ∈ Mn .
k=1

(ii): We identify Mn with Mn (C1) = Mn ⊗1 ⊆ Mn (M(A)), where 1 := 1M(A) .

By Part (i) there exists columns c1 , . . . , cn ∈ Mn,1 (A) ∼
= Mn,1 ⊗ A that satisfy
P ∗ (k)
V (β) = k ck βck . The entries of ck are given by c`,1 := gk,` i.e.,

ck := [gk,1 , gk,2 , . . . , gk,n ]> ,

and the gj,k are entries of a positive matrix G = [gj,k ]2 = [V (ejk )]. It suffices to
consider the case where kV k = kV (1n )k = 1, because in case V = 0 nothing is to
prove and non-zero V can be replaced by kV k−1 V if necessary.
We get norm estimates for the gk,` (k, ` ∈ {1, . . . , n}) by
∗ ∗
kgk,` gk,` k = kgk,` gk,` k ≤ kc∗k ck k ≤ kV (1n )k ≤ 1 .

Let h := n−2 k,` gk,` gk,` ∗ ∗

≤ n2 · h implies that
P
∈ A+ . Then khk ≤ 1, and gk,` gk,`
gk,` ∈ hA for k, ` ∈ {1, . . . , n} .

12 This are not the columns of the adjoint matrix. Even for positive complex 2 × 2-matrices,

e.g. [αj,k ] with αj,k := ij−k . Only for matrices with self-adjoint entries is this the same as the
columns of the adjoint matrix.
364 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

Consider the functions ψm (t) := min 1, max((m + 1)t − 1/m, 0) for m ∈ N
and t ∈ [0, ∞]. Then ψm (t) = 0 for t ∈ [0, 1/(m2 + m)], ψm (t) = 1 for all t ≥ 1/m
and ψm is linear in between. Thus, limm→∞ kx∗ (1 − ψm (h))xk = 0 for all x ∈ hA .
In particular, there exists m ∈ N with

k(1 − ψm (h))1/2 gk,` k2 < γ/(1 + n2 ) for all k, ` ∈ {1, . . . , n} .

Let ϕ := ψm and Q := 1 − ϕ(h) ≥ 0 from now on.

We write 1n ⊗ Q for the diagonal matrix in Mn (M(A)) with entries Q in the
main diagonal.
The function ϕ is an increasing non-negative function ϕ ∈ C0 (0, 1] with
ϕ|[0, δ] = 0 for δ = (m2 + m)−1 and ϕ has the properties that kϕk ≤ 1 and

kQ1/2 gk,` k2 = k(1 − ϕ(h))1/2 gk,` k2 < γ/(1 + n2 ) for all k, ` ∈ {1, . . . , n} .

Since kc∗k (1n ⊗ Q)ck k ≤ ` kgk,`

∗
P
Qgk,` k it implies that
n
X
k c∗k (1n ⊗ Q)ck k < γ . (1.1)
k=1

By assumptions in Part (ii), the C *-algebra A has the property that for every
positive contraction h ∈ A+ and ε > 0 there exist contractions b1 , b2 ∈ A with
kb∗j bk − δj,k hk < ε for j, k ∈ {1, 2} . It causes that Lemma 2.1.12 applies to h ∈
A+ . The equivalences (b,i) ⇐⇒ (b,ii) and (a,i) ⇐⇒ (a,ii) in Lemma 2.1.12 imply
together immediately that there exist b1 , b2 , . . . , b2n−1 , b2n ∈ A with

b∗i bj = δi,j ϕ(h) for all i, j ∈ {1, . . . , 2n} .

The diagonal matrices 1n ⊗ bj ∈ Mn (A) (j ∈ {1, . . . , 2n}) and all β ∈ Mn

satisfy (1n ⊗ bj )∗ β(1n ⊗ bk ) = δjk · β ⊗ ϕ(h) and

(1n ⊗ Q)1/2 β(1n ⊗ Q)1/2 = β ⊗ Q = β ⊗ (1 − ϕ(h)) .

Thus, β = (1n ⊗ bj )∗ β(1n ⊗ bj ) + (1n ⊗ Q)1/2 β(1n ⊗ Q)1/2 for β ∈ Mn and

j ∈ {1, . . . , 2n} . If we conjugate this with the columns ck ∈ Mn,1 (A) then this
shows that the completely positive maps Vk,j and Wk defined by

Vk,j : Mn 3 β 7→ ((1n ⊗ bj )ck )∗ β((1n ⊗ bj )ck ) ∈ A

and
Wk : Mn 3 β 7→ ((1n ⊗ Q)1/2 ck )∗ β((1n ⊗ Q)1/2 ck ) ∈ A
have the property that Vk,j (β)+Wk (β) = c∗k βck for all β ∈ Mn and j ∈ {1, . . . , 2n}
It implies Vk,i (β) = Vk,j (β) for all i, j ∈ {1, 2, . . . , 2n − 1, 2n}, and the estimate
Pn
(1.1) implies for the completely positive map W := k=1 Wk the estimate
n
X
kW k = kW (1n )k = k c∗k (1n ⊗ Q)ck k < γ .
k=1

Calculation shows for µ, ν ∈ {1, . . . , 2n} and i, j, k, ` ∈ {1, . . . , n} that

((1n ⊗ bµ )ck )∗ (ei,j ⊗ 1)((1n ⊗ bν )c` ) = δµ,ν · c∗k (ei,j ⊗ ϕ(h))c` .

 1. EXACT C *-ALGEBRAS AND NUCLEAR MAPS 365

We define for ` ∈ {0, 1} the columns

n
X n
X
d` := (1n ⊗ bk+n` )ck = ej,1 ⊗ (bk+n` gk,j ) ∈ Mn,1 (A) .
k=1 k,j
Pn
Above considerations show that the columns dj , j ∈ {0, 1} satisfy ( k=1 c∗k βck ) −
n
d∗j βdj = ∗ 1/2
β(1n ⊗ Q)1/2 ck , and that the c.b.-norm on the right
P
k=1 ck (1n ⊗ Q)
side is ≤ γ · kβk .
The orthogonality (1 ⊗ bj )∗ β(1 ⊗ bk ) = δjk β ⊗ ϕ(h) causes that d∗1 βd0 = 0 for
β ∈ Mn .
It follows that dj (j ∈ {0, 1}) are column matrices in Mn,1 (A) with d∗0 βd1 = 0
and d∗0 βd0 = d∗1 βd1 such that the c.p. map U (α) := d∗0 αd0 satisfies that W := V −U
is c.p. and kW k < γ and U (1n ) ≤ V (1n ) .
(iii): It suffices to consider the case where kV k = 1. Let [ajk ] ∈ Mn (A)+
associated to V ∈ CP(Mn , A) by ajk = V (ej,k ). Denote by πJ : D → A = D/J
the quotient map. By Parts (o) and (i) it suffices to find in Mn (D)+ a matrix [djk ]
with πJ (dj,k ) = aj,k and kd1,1 + d2,2 + · · · + dn,n k ≤ 1 .
It is possible to find [cj,k ] ∈ Mn (D)+ with πJ (cj,k ) = aj,k because the map
Mn (D) 3 [dj,k ] 7→ [πJ (dj,k )] ∈ Mn (A) is an C *-algebra epimorphism onto Mn (A).
Then let e := c1,1 + · · · + cn,n and define f := (1 + (e − 1)+ )−1/2 ∈ M(D)
and the desired matrix [dj,k ] ∈ Mn (D) by dj,k := f cj,k f . The corresponding map
T : Mn → D has norm kT k = kf ef k ≤ 1 and satisfies πJ ◦ T = V .
P
(iv): It suffices to consider the case where kV k = k k bk,k k = 1 for completely
positive V : Mn → N and the corresponding matrix B := [bjk ] ∈ Mn (N )+ with
bj,k := V (ej,k ). Then kBk = k[bj,k ]k ≤ n . Let C = [cj,k ] ∈ Mn (N )+ the square
root of B. Then kCk ≤ n1/2 .
Since A ⊆ N is weakly dense in N , the C *-subalgebra Mn (A) is weakly dense
in the W*-algebra Mn (N ). By the Kaplansky density theorem [616, thm. 2.3.3],
(or [704, thm. 1.9.1], [400, thm. 5.3.5]) the unit-ball of the C *-algebra Mn (A) is
dense in the unit-ball of Mn (N ) with respect to the *-ultra-strong topology (i.e.,
is τ (Mn (N ), Mn (N )∗ )-dense in the unit-ball of Mn (N )), because Mn (A) is weakly
dense (i.e., is σ(Mn (N ), Mn (N )∗ )-dense) in Mn (N ) if A is weakly dense in N . It
follows that there exists a net of elements {Cρ } ⊆ Mn (A) with kCρ k ≤ n1/2 that
converges to C with respect to the *-ultra-strong topology on Mn (N ).
Let eρ := j,k c∗k,j ck,j and let f (ρ) ∈ Mn (M(A))+ the positive diagonal matrix
P

with entries (f (ρ) )j,k := δj,k (1 + (eρ − 1)+ )−1/2 ∈ M(D) .

The obvious *-ultra-strong continuity of addition, multiplication and involution
on bounded parts of Mn (N ), cf. [616, sec. 2], shows that the directed net {Dρ }
defined by
D(ρ) := f (ρ) Cρ∗ Cρ f (ρ) ∈ Mn (A)+
Pn (ρ)
satisfies that k k=1 Dk,k k ≤ 1 and converges to [bjk ] in Mn (N ) with respect to
the *-ultra-strong topology. It implies for the related c.p. maps Vρ ∈ CP(Mn , A)
366 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

that kVρ k ≤ 1 and that the net {Vρ } converges to V : Mn → N with respect to the
point–*-ultra-strong topology. 

Remarks 3.1.10. The last here mentioned criteria seems to be the

relevant issue! But check it again!
The arguments in the proof of Part(ii) of Proposition 3.1.9 show that we could
find moreover elements b1 , b2 , . . . , b2n−1 , bkn ∈ A with b∗i bj = δi,j ϕ(h) and the other
properties of the bj ∈ A. Then we obtain from them columns d0 , d1 , . . . , dk ∈ Mn,1
with d∗j βdk = δj,k U (β) with U and V − U c.p. with norm kV − U k < γ.
But the proof of Proposition 3.1.9 shows a bit more:
If A is a C *-subalgebra of a C *-algebra B such that for each element a ∈ A+
and ε > 0 there exists d1 , d2 ∈ B with kd∗i dj − δi,j ak < ε, then the columns
d0 , d1 , . . . , dk ∈ B can be found with above properties and such that the c.p. con-
traction U ∈ CP(Mn , B) – defined by the dj via δi,j · U (β) = d∗i βdj – is also
contained in CP(Mn , A) . It does not require that M(A) contains any properly in-
finite elements, and it does not require that all elements in B are properly infinite.
One could suppose that the elements of A are “large” in B in the sense that
the multiplier algebra of the hereditary C *-subalgebra ABω A of Bω has a properly
infinite unit.
But the weakened condition on A and B simply is equivalent to the property
that A is contained in a stably generated ideal of Bω or likewise of B∞ .

Lemma 3.1.11. Let V : A → B a completely positive map, and A ⊆ D where

D is a unital C*-algebra.

(i) If V is factorable, then there is always a factorization V = T ◦ S through

some matrix algebra Mn such that S ∗∗ : A∗∗ → Mn is unital, and (there-
fore) kT k = kT (1)k = kV k . There exists a u.c.p. map Se : D → Mn with
Se |A = S ( 13 ).
(ii) If V is nuclear, then V can be approximated in point norm by factorable
maps T S where S : A → Mn and T : Mn → B are completely positive
maps such that S ∗∗ : A∗∗ → Mn is unital and kT k ≤ kV k.
If A, B and V are unital then S is unital and T can be chosen with
T (1) = 1.
(iii) The map V : A → B is nuclear, if and only if, V : A → B ∗∗ is weakly
nuclear – considered as a map into the W*-algebra B ∗∗ .
(iv) Every completely positive map V : A → B of finite rank can be approxi-
mated in operator-norm on CP(A, B) by factorable maps.

Proof. (i): This is evident if x = S ∗∗ (1) is invertible, because we can replace

S with x−1/2 S(·)x−1/2 and T with T (x1/2 (·)x1/2 ). If S ∗∗ (1) is not invertible then,

13 The extension S is an special case of the Arveson extension theorem [42]. Since the
e
identity map of L(`2 ) factorizes over `∞ (M2 , M3 , . . .) this special cases imply conversely the
general Arveson extension.
 1. EXACT C *-ALGEBRAS AND NUCLEAR MAPS 367

since 0 ≤ a ≤ 1 implies 0 ≤ S ∗∗ (a) ≤ S ∗∗ (1), S ∗∗ must take its values in a corner

of Mn . Replacing Mn with this corner and adjusting S, T accordingly, we may
assume that S ∗∗ (1) is invertible, and use the above transformations.
Since T (S ∗∗ ) : A∗∗ → B ⊆ B ∗∗ is normal, (T S)∗∗ = T (S ∗∗ ), and thus kT (1)k =
k(T S)∗∗ (1)k. But kT Sk = k(T S)∗∗ k = k(T S)∗∗ (1)k and kT k = kT (1)k, because T
and T S are positive.
It is easy to verify that a map R : D → Mn is completely positive, if and only
if, the linear functional ρe : D ⊗ Mn → C with ρe (d ⊗ α) = (1/n)Tr(α> R(d)) is
positive (Consider R as a map from D into Mn ⊆ L(L2 (Mn , (1/n)Tr)) ∼
2
= L(Cn )).
Since A ⊆ D, we have A∗∗ ⊆ D∗∗ and D is a unital C *-subalgebra of D∗∗ . Let
p ∈ D∗∗ denote the unit element of A∗∗ . We extend S : A∗∗ → Mn to a unital c.p.
map S0 from C := A∗∗ + (1 − p)D∗∗ (1 − p) into Mn by S0 (c) := S ∗∗ (pcp) for c ∈ C.
The n2 -positivity of S0 implies that the (well-defined) unital linear functional
ρ on C ⊗ Mn with ρ(c ⊗ α) = (1/n)Tr(α> S0 (c)) is positive, i.e., is a state on
C ⊗Mn . By Hahn–Banach extension theorem, it extends to a state ρe on D∗∗ ⊗Mn
with ρe (1 ⊗ α) = (1/n)Tr(α). Thus, the unique linear map Se : D → Mn with
ρe (d ⊗ α) = (1/n)Tr(α> Se (d)) for all d ∈ D and α ∈ Mn must be unital and
extends S0 .
(ii): Let A
e the unitization of A. V is the point norm limit of completely
e → B is defined by Ve (a + γ1) := V (eae + γe2 ) for
positive maps Ve |A where Ve : A
e ∈ A+ , kek ≤ 1 (Take an approximate unit {e} of A). Ve is completely positive
and kVe k ≤ kV k. Every point-norm approximation of V by a net of factorable
maps Wλ = Tλ Sλ defines a point-norm approximation of Ve by the factorable
maps (Wλ )e . Thus it suffices to consider the case where A is unital. But then
the approximation satisfies kV (1) − Wλ (1)k → 0. Thus, eventually we find a small
perturbation of the Tλ with the desired properties.
(iii): By Remark 3.1.2(o), the set CPf (A, B) of factorable c.p. maps W : A → B
is an operator-convex cone in CP(A, B). It follows that V : A → B is in the point-
norm closure of CPf (A, B) if and only if V is in the point-σ(B, B ∗ ) closure of
CPf (A, B).
Every factorable c.p. map U = T ◦ S : A → B ∗∗ , T : Mn → B ∗∗ , S : A → Mn
with S ∗∗ (1) = 1 and kT k = kT (1)k = kU k is the point–*-strong limit of a net of
maps Uµ = Tµ ◦ S by part (i) and Proposition 3.1.9(iv), where Tµ : Mn → B is a
contraction with kTµ k ≤ kT k .
Thus, V : A → B is in the point-σ(B, B ∗ ) closure of CPf (A, B), if and only if,
V is in the point-σ(B ∗∗ , B ∗ ) closure of CPf (A, B ∗∗ ).
The latter means that V : A → B ⊆ B ∗∗ is a weakly nuclear map from A into
B ∗∗ .
(An alternative proof can be given by means of Remark 3.1.2(i) and its counterpart
for weakly nuclear maps.)
368 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

(iv): We consider B as a C *-subalgebra of L(H1 ) for some Hilbert space

H1 . Since V : A → B has finite rank, there are functionals ρ1 , . . . , ρn ∈ A∗ and
P
operators b1 , . . . , bn ∈ B with V (a) = k ρk (a)bk . By Lemma 3.1.4 there exists a
cyclic representation d : A → L(H0 ), with cyclic vector ξ ∈ H0 of norm kξk = 1,
and operators tk ∈ d(A)0 with ρk (a) = hd(a)ξ, tk ξi for a ∈ A.
Let C := d(A)0 , T :=
P ∗
k t k ⊗ bk ∈ C B ⊆ C ⊗ B and let λa denote the
linear functional on C given by λa (c) := hcd(a)ξ, ξi for a ∈ A and c ∈ C. It has
norm kλa k ≤ kak. Then V (a) = (λa ⊗ id)(T ) for a ∈ A. Since the algebraic tensor
product d(A)ξ H1 is dense in the Hilbert-space tensor product H0 ⊗H1 , a straight
calculation shows that the complete positivity of V implies that T is positive in
L(H0 ⊗ H1 ). It follows that T is also positive in C ⊗ B (by spectral permanence
on the class of C *-algebras).
Let ε > 0, then there is S ∈ C B with kT − S ∗ Sk < ε in C ⊗ B. Let
W (a) := (λa ⊗ id)(S ∗ S) then, for all a ∈ A,

kW (a) − V (a)k = k(λa ⊗ id)(T − S ∗ S)k ≤ kλa kkT − S ∗ Sk < εkak ,

i.e., kW − V k < ε in L(A, B). Since d : A → L(H0 ) has a cyclic vector ξ, the map
γ : C = d(A)0 → H0 given by γ : c 7→ cξ is injective, and we can write S = j cj ⊗dj
P

(j = 1, . . . , m) with hγ(ci ), γ(cj )i = δi,j . Let un : Cm → H0 the isometry with

un (ej ) = γ(cj ), W1 : A → Mm the c.p. contraction W1 (a) := u∗n d(a)un ∈ L(Cm ),
and D ∈ M1,m (B) the row matrix D = [d1 , . . . , dm ] . We define W2 : Mm → B
by W2 (α) := DαD∗ , where Mm is considered as subalgebra of Mm (M(B)) ∼ =
M(Mm (B)). Straight calculation shows that W = W2 ◦ W1 . Thus, V can be
approximated in operator-norm by factorable maps. 

Compare below Proof with that of Part(iv) above! Take the better
one! Notations have then to be adapted/changed.
Details to Part (iii,f ) of Remarks 3.1.2, see also Lemma 3.1.11(iv):
∗ 0
P
Consider V (a) := j fj (a)bj ∈ B ⊆ L(H) and find ϕ ∈ A+ , Tj ∈ Dϕ (A) , such
P
that fj (a) = hDϕ (a)x0 , Tj x0 i , cf. Lemma 3.1.4. Then S := Tj ⊗ bj ≥ 0 in the
C *-algebra Dϕ (A)0 ⊗ B. Thus, S 1/2 can be approximated in operator norm by
elements dn in the algebraic tensor product Dϕ (A)0 B . The maps

a 7→ (ϕ ⊗ idB ) d∗n (Dϕ (a) ⊗ 1)dn

are factorable c.p. maps from A into B that approximate V with respect to the
operator norm of L(A, B). 
We denote by socle(A) the algebraic ideal of A generated by the projections
p ∈ A with pAp = Cp. See [210], [388, p. 64] and [661, pp. 46,261–267] for notation
and properties.
Compare
 1. EXACT C *-ALGEBRAS AND NUCLEAR MAPS 369

Lemma 3.1.12. Suppose that W : A → Mm is a completely positive map with

∗∗
W (1) = 1m , and that Γ is a set of pure states on A, such that the irreducible rep-
resentations ρj : A → L2 (A, µj ) corresponding to µj ∈ Γ are pairwise inequivalent
and build a separating family for A. Then:

(i) For every compact subset Ω of A and every ε > 0, there exist pure states
µ1 , . . . µk ∈ Γ, k, m1 , . . . , mk ∈ N, isometries Ij : Cmj → L2 (A, µj ), and a
unital completely positive map T : F := Mm1 ⊕ · · · ⊕ Mmk → Mm , such
that
kT ◦ S(a) − W (a)k ≤ ε ∀ a ∈ Ω,
where S : A → Mm1 ⊕· · ·⊕Mmk denotes the completely positive contraction
S(a) := I1∗ ρ1 (a)I1 ⊕ · · · ⊕ Ik∗ ρk (a)Ik for a ∈ A.
(ii) If, in addition, W (socle(A)) = 0, then one can manage that m1 , . . . , mk ≤
m, and that T (f ) = j d∗j fj dj for f = f1 ⊕ · · · ⊕ fk ∈ F , for suitable
P
P ∗
dj ∈ Mmj ,m , with dj dj = 1m .
(iii) If, in addition, W (socle(A)) = 0 and the irreducible representations dµ
with µ ∈ Γ and dµ (socle(A)) = 0 build a separating family for A/socle(A),
then in Part(i) the µ1 , . . . , µk ∈ Γ can be taken such that µj (socle(A)) = 0
and S(socle(A)) = 0.

Note that S satisfies S ∗∗ (1) = 1. Part (iii) does not say that F , T and S can be
found such that the properties of µj , Ij and T in (i), (ii) and (iii) hold at the same
time (but this can be managed if socle(A) = 0 or, at least, if socle(A/socle(A)) = 0
and W (socle(A)) = 0).

L
Proof. (i): Let H := µ∈Γ L2 (A, µ) and ρ : A → L(H) be the direct sum
of irreducible representations which are defined by pure states in Γ. Then ρ is
faithful. We denote by R : A → L(`2 (H)) the infinite repeat of ρ (i.e., R(a) :=
ρ(a) ⊕ ρ(a) ⊕ · · · ). Then R(A) ∩ K(H) = 0, and, by Lemma 2.1.22, there exists an
isometry v : Cm → H with kW (a) − v ∗ R(a)vk < ε/3 for a ∈ Ω.
By definition of R, a small perturbation of v yields q ∈ N, µ1 , . . . , µq ∈ Γ (not
necessarily different) and an isometry t : Cm → L2 (A, µ1 ) ⊕ · · · ⊕ L2 (A, µq ) with
k v ∗ R(a)v − t∗ (ρ1 (a) ⊕ · · · ⊕ ρq (a))t k < ε/3 for a ∈ Ω, where ρj : A → L2 (A, µj )
are the representations corresponding to µj . We can arrange that µ1 , . . . , µk are
different and {µk+1 , . . . , µq } ⊆ {µ1 , . . . , µk } if k ∈ N denotes the maximal number
of different elements of {µ1 , . . . , µq }. There are contractions sj : Cm → L2 (A, µj )
P ∗
(j = 1, . . . , q) such that t(x) = s1 (x) ⊕ · · · ⊕ sq (x) for x ∈ Cn , sj sj = 1m , and
q
X
k v ∗ R(a)v − s∗j ρj (a)sj k < ε/3 ∀ a ∈ Ω.
j=1

For 1 ≤ j ≤ k, we denote by Lj the linear subspace of L2 (A, µj ) generated by

the image of sj and the images si (Cm ) for i ∈ {k + 1, . . . , q} with µi = µj . Let
mj := Dim(Lj ) and Ij : Cmj → Lj ⊆ L2 (A, µj ) an isometry (j = 1, . . . , 0). Let
1 ≤ j ≤ k and X(j) := {1 ≤ i ≤ q ; µi = µj } . We define Tj : Mmj → Mm by
370 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

∗ ∗
P
Tj (g) := i∈X(j) si Ij gIj si for g ∈ Mmj and 1 ≤ j ≤ k, and then define S : A →
F := Mm1 ⊕ · · · ⊕ Mmk and T : F → Mm by S(a) := I1∗ ρ1 (a)I1 ⊕ · · · ⊕ Ik∗ ρk (a)Ik
respectively T (f1 , . . . , fk ) := T1 (f1 ) + · · · + Tk (fk ) for a ∈ A and fj ∈ Mmj .
Pq
Straight calculation shows that j=1 s∗j ρj (a)sj = T (S(a)) for all a ∈ A.
(ii): Let H and ρ : A → L(H) as above. Then J := ρ−1 (ρ(A) ∩ K(H)) is the
closed ideal generated by the projections p ∈ A with pAp of finite dimension, i.e., J
is the closure of socle(A). Since W (socle(A)) = 0 by assumption, we get W (J) = 0.
Now we can repeat the arguments for part (i) with ρ in place of R. Then, k = q
and Dim(Lj ) ≤ m for 1 ≤ j ≤ k. It follows that the map t(x) := s1 (x) ⊕ · · · ⊕ sk (x)
is an isometry from Cm into I(Cm1 ⊕ . . . ⊕ Cmk ) and that Tj (g) = s∗j Ij gIj∗ sj for
all g ∈ Mmj . Let dj := Ij∗ sj ∈ L(Cm , Cmj ) ∼ = Mmj ,m . Let pj : Cn → Cmj
P ∗
denote the orthogonal projection onto Cmj ⊆ Cn . Then T (f ) = dj fj dj for
P ∗
f := f1 ⊕ · · · ⊕ fk ∈ F , in particular, dj dj = 1m .
(iii): Apply part (i) to A/socle(A) and [W ] : A/socle(A) → Mm . 

Lemma 3.1.13. Let A a C*-algebra and W : A → Mm a completely positive

contraction with W ∗∗ (1) = 1, Ω ⊆ A+ a norm-compact subset, f0 (t) := t (t ∈ [0, 1])
and let ε > 0.

(I) There exist

(i) a finite-dimensional C*-algebra F := Mn1 ⊕ · · · ⊕ Mnk
(ii) a C*-morphism ψ : C0 ((0, 1], F ) → A ,
(iii) a completely positive contraction S : A → F with S(ψ(f )) = f (1) for
all f ∈ C0 ((0, 1], F ), and
(iv) a unital completely positive map T : F → Mm ,
such, that for g := ψ(f0 ⊗ 1) and each a ∈ Ω,

kT ◦ S(a) − W (a)k < ε and lim sup kg n+1 ag n+1 − g n ψ f02 ⊗ S(a) g n k < ε .


n

(II) If W (socle(A)) = {0} then F , ψ, S and T can be found such that, more-
over, for every η ∈ (0, 1) the hereditary algebras

(fj − (1 − η))+ · A · (fj − (1 − η))+

contain n2j non-zero pairwise orthogonal positive elements for j = 1, . . . , k

(j)
(where fj := ψ(f0 ⊗ p11 )).
(III) If W (socle(A)) = {0} then one can find F , ψ, S and T such that nj ≤ m
and that S : F → Mn is given by S(f1 , . . . , fn ) = j s∗j fj sj for suitable
P

sj ∈ Mmj ,m .

Proof. Apply Lemma 2.1.15. To be filled in ?? 

Question 3.1.14.
Let A ⊆ B a separable C *-subalgebra of a unital C *-algebra B.
Under what conditions is every nuclear c.p. map T : A → B approximately
1-step inner inside B?
 1. EXACT C *-ALGEBRAS AND NUCLEAR MAPS 371

I.e., when does there exists for each c.p. contractions T ∈ CP(Mn , B) and
S ∈ CP(A, Mn ) a sequence s1 , s2 , . . . of contractions in B such that T ◦ S(a) =
limk→∞ s∗k ask for all a ∈ A?
A necessary condition (for the case n = 1) is that for each nonzero a ∈ A+
with kak = 1 and every c ∈ B+ there exists d ∈ B – depending on c – with
d∗ ad = (c − 1/2)+ .
This implies that A is “relatively to B simple” (or A = {0}) and that the right
ideal R := AB satisfies R∗ R = B.
Is this also sufficient?
Compare the result of Elliott and Kucerovsky [264] and its improvements in
[309] in the case where B is a corona algebra M(C)/C.

Following proposition is a non-commutative version of the Tietze extension

theorem.

Proposition 3.1.15. Let A a σ-unital C*-algebra and J ⊆ A a closed ideal.

Then the natural unital C*-morphism M(πJ ) : M(A) → M(A/J) has kernel
M(A, J) := {T ∈ M(A) ; AT ∪ T A ⊆ J}.
This natural C*-morphism is surjective if J is σ-unital.

Proof. Let h : A → M(B), – e.g. for B := A/J –, a C *-morphism with the

property that the hereditary C *-subalgebra h(A)Bh(A) of B contains an approxi-
mate unit of B .
This is equivalent to B = h(A)B . Thus, there is a unique unital W*-morphism
H : A∗∗ → B ∗∗ that maps M(A) ⊆ A∗∗ into M(B) ⊆ B ∗∗ and satisfies H(a)b =
h(a)b for all a ∈ A and b ∈ B. The kernel of H|M(A) is easily seen as the operators
T ∈ M(A) with T A ∪ T ∗ A ⊆ J, where J is here the kernel of h : A → M(B).
This covers the special case B := A/J and h := πJ , where here simply H =
∗∗
h .
To get surjectivity for M(h) = h∗∗ |M(A) one can reduce the situation to
suitable separable C *-subalgebras if J and B = A/J are σ-unital: We can find
for given positive T ∈ M(B) a separable C *-subalgebra C ⊆ B that contains a
strictly positive element f of B and 1M(B) , T ∈ M(C) ⊆ M(B) . Then we can find
a separable C *-subalgebra D of A that contains a strictly positive element of J and
satisfies πJ (D) = C. Thus, D contains a strictly positive element of A, satisfies
M(D) ⊆ M(A), and M(D) contains the unit element of M(A). Notice that under
this conditions on D and C we get that

M(πJ )|M(D) = M(πJ∩D ) : M(D) → M(C) ⊆ M(B) .

This reduces in the case where J and B = A/J are σ-unital the question on
the surjectivity of M(πJ ) to the case where B is separable. There the answer is
affirmative, cf. [616, prop. 3.12.10]. 
372 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

Proposition 3.1.15 could be used to prove the following proposition:

Proposition 3.1.16. If A is a σ-unital C*-algebra, then Q(A) := M(A)/A

is σ-sub-Stonean, i.e., for each separable C*-subalgebra B ⊂ Q(A) and each c, d ∈
(B 0 ∩ Q(A))+ with cd = 0 there exist contractions e, f ∈ (B 0 ∩ Q(A))+ with ef = 0,
ec = c and f d = d.
I.e., Q(A) is σ-sub-Stonean in the sense of [448, def. 1.4].

Compare Chapter 5 for more general study

and compare with the -- possibly stronger -- Kasparov’s technical
theorem:
M(A ⊗ K)/(A ⊗ K) = Qs (A) satisfies Kasparov’s technical theorem. It implies
that Qs (A) and Q(A) ∼
= P Qs (A)P are sub-Stonean. Here P := πA⊗K (1M(A) ⊗p11 ).
Check below proof again. Compare this with Chapter 5.

Proof. We can suppose that 1 ∈ B and kc − dk = 1 without loss of generality.

Let g ∗ = g ∈ M(A) with πA (g) = c − d and kgk = 1, and let a0 ∈ A+ a strictly
positive contraction in A. We find a separable unital C *-subalgebra D ⊂ M(A)
with πA (D) = B. Consider the separable unital C *-subalgebra E := C ∗ (a0 , g, D)
of M(A) that is generated by a0 , g ∈ A and the C *-subalgebra D ⊆ E.
Take dense sequence e1 , e2 , . . . in the unit ball of E containing e1 := g+ and
e2 := g− in the first places.
Build from a0 via functional calculus with increasing functions f1 , f2 , . . . in
C0 (0, 1]+ (with properties kfn k = 1, fn+1 fn = fn and fn ([0, αn ]) = {0} for some
suitable zero sequence αn > αn+1 > 0) a sequence of positive contractions an =
fn (a0 ) ∈ A+ that have the property

−n
kan ek − ek an k + ka1/2 1/2
n ek − ek an k ≤ 8 .

1/kn
Parallel selection of k[g+ , ek ]k and its commutator norms is
needed:
1/k
We know that [g+ n , ek ] ∈ A. The next kn+1 can be found after the [(an+1 −
an )1/2 , ek ] and ??? becomes small enough and ????
Then we define a map V on T ∈ M(A) by
∞
1/2 1/2
X
V (T ) := a1 T a1 + (an+1 − an )1/2 T (an+1 − an )1/2 .
n=1

The right side is convergent with respect to the strict convergence in M(A). Then
V is a unital completely positive unital map from M(A) into M(A).
It has the property that V (en ) − en ∈ A for all n ∈ N. Thus V (x) − x ∈ A for
all x ∈ E ⊆ M(A).
 1. EXACT C *-ALGEBRAS AND NUCLEAR MAPS 373

We define a positive contraction Y ∈ M(A)+ by

∞ ∞
1/2 1/2 1/n
X X
Y := := a1 g+ a1 + (an+1 − an )1/2 g+ (an+1 − an )1/2 .
n=1 n=1

Here could be a problem with Y x − xY ∈ A

It could be that one has to select some
1/`
of the akn and g+ n more careful !! :

1/n
We have [g+ , ek ] ∈ A for all n, k ∈ N, but need further control of
the selection!!
The n has to go slowly enough to ∞??
The positive contraction Y has the properties Y g+ − Y ∈ A, Y g− ∈ A and
Y x − xY ∈ A for all x ∈ E = C ∗ (a0 , g, D).
Thus, e := πA (Y ) ∈ Q(A) is a positive contraction with ec = e, ed = 0 and
e ∈ B 0 ∩ Q(A).
Now we can repeat the arguments, but starting now with e in place of former
d and with d in place of former c and get a positive contraction f ∈ B 0 ∩ Q(A) with
f e = 0 and f d = d.


Remarks/Conjectures:
It is known that Aω is σ-sub-Stonean for every non-zero C *-algebra A.
Is A∞ σ-sub-Stonean for non-zero A?
Suppose we could prove the following (1) and (2) then Qs (A) is σ-sub-Stonean
for each non-zero σ-unital C *-algebra :
(1) A∞ is σ-sub-Stonean.
(2) If A is σ-unital and stable, then, for every separable C *-subalgebra B of
Q(A) there exist a c.p. contraction V : Q(A) → A∞ that is orthogonality preserving
on B and a c.p. contraction W : A∞ → Q(A) that is multiplicative on C ∗ (V (B))
such that W ◦ V (b) = b for all b ∈ B.
Indeed: Suppose that A is σ-unital and stable, C ⊂ Q(A) separable, and that
e, f ∈ C 0 ∩ Q(A) are positive contractions with ef = 0 .
Suppose (2) is valid:
Let B := C ∗ (C ∪ {e, f }), V : Q(A) → A∞ a c.p. contraction that is orthogonal-
ity preserving on B, and let W : A∞ → Q(A) that is multiplicative on C ∗ (V (B))
such that W ◦ V (b) = b for all b ∈ B.
Since V is orthogonality preserving, we get that V (e) · V (f ) = 0 .
Perhaps V preserves commutativity? Leads to question (Let T : C0 (X) → D
an orthogonality preserving c.p. contraction with D a C *-algebra, X a closed
subset of R2 , and T is “completely” isometric. Is C ∗ (T (C0 (X))) a commutative
374 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

C *-algebra? We can not expect that T is multiplicative because f 7→ f ⊗ g is c.i.

and c.p. if kgk = 1 and g ≥ 0.)
Now suppose that (2) implies that V respects commutativity and is orthogo-
nality preserving.
Then V (e)V (f ) = 0 and V (e), V (f ) commute with all elements from C ∗ (V (B)).
(If comm. is respected!)
Now, if (1) is valid, then there exist g, h ∈ C ∗ (V (B))0 ∩ A∞ with g · h = 0,
gV (f ) = 0 gV (e) = V (e) and hV (f ) = V (f ).
If we apply W : A∞ → Q(A), then we get that W (g)e = e, W (g)f = 0 W (g)b =
bW (g) for all b ∈ C, and W (g)f = 0. This implies then finally that Q(A) is σ-sub-
Stonean.

2. On approximate inner nuclear maps

The following Lemma 3.2.1 gives a necessary and sufficient criteria for the
containment CPnuc (A, B) ⊆ C for point-norm closed subsets C ⊆ CP(A, B) of the
c.p. maps from a C *-algebra A into a C *-algebra B.

Lemma 3.2.1. Let C a point-norm closed subset of CP(A, B) with the property
that S(a∗ (·)a) ∈ C for S ∈ C and a ∈ A, and denote by f0 ∈ C0 (0, 1] the generator
f0 (t) := t.
Then the following are equivalent:

(i) The set C contains all nuclear c.p. maps U : A → B.

(ii) For every C*-morphism ψ : C0 ((0, 1], F ) → A of the cone C0 ((0, 1], F )
over a finite-dimensional C*-algebra F with the property kψ(f0 ⊗ p)k = 1
for all non-zero projections p ∈ F and for every completely positive map
V : F → B, there exists a sequence Sn ∈ C with limn Sn (ψ(f )) = V (f (1))
for all f ∈ C0 ((0, 1], F ).

Proof. The condition on C implies that tC ⊆ C for t ∈ R+ (use an approximate

unit of A). Thus, it suffices to show that every factorable contraction U from A
into B can be approximated in point-norm topology by elements of C. Suppose that
U = V ◦W with c.p. contractions W : A → Mm and V : Mm → B with W ∗∗ (1) = 1,
cf. Lemma 3.1.11(i).
Let ε > 0 and Ω ⊆ A a compact subset. We find F , ψ : C0 ((0, 1], F ) → A,
T : F → Mm , and S : A → F as in Lemma 3.1.13(I). Then 1 ≥ kψ(f0 ⊗ p)k ≥ kpk,
because p = T (ψ(f0 ⊗ p)). Thus, by assumptions, there is a sequence Sn ∈ C with
limn Sn (ψ(f )) = V ◦T (f (1)) for all f ∈ C0 ((0, 1], F ). Let G := ψ(C0 ((0, 1], F )) ⊆ A
and let D denote the hereditary C *-subalgebra of A that is generated by G. Lemma
3.1.8 applies to the set C|G, because ψ is a C *-morphism and V (a∗ (·)a) ∈ C for
V ∈ C and a ∈ A. Hence, we can select the sequence Sn ∈ C such that, in addition,
kSn |Gk ≤ kV ◦ T k ≤ 1 . But this implies that kSn |Dk ≤ 1, because G contains an
 2. ON APPROXIMATE INNER NUCLEAR MAPS 375

approximate unit of D. It follows limn Sn (g k ψ(f0 ⊗ S(a))g k ) = V (T (S(a))) and

inf kSn (g k+1 ag k+1 ) − V ◦ W (a)k < 3ε ∀ a ∈ Ω.

n,k

Conversely, suppose that CPnuc (A, B) ⊆ C. Let F a finite-dimensional algebra

and ψ : C0 ((0, 1], F ) → A a *-morphism with kψ(f0 ⊗ p)k = 1 for all non-zero


projections p ∈ F . The image ψ C0 ((0, 1], F ) is contained in the normalizer al-
gebra N (D) ⊆ A of the hereditary C *-subalgebra D ⊆ A that is generated by
ψ(f0 (1 − f0 ) ⊗ 1). The condition kψ(f0 ⊗ p)k = 1 for non-zero projections p ∈ F
is equivalent to the injectivity of [πD ◦ ψ] : F → N (D)/D where πD : N (D) →
N (D)/D is the quotient map. Since F is an injective algebra, we can use the Arve-
son extension to obtain completely positive contractions W1 : N (D) → F with
???????
and then W : A → F with W (ψ(f )) = W1 (ψ(f )) = f (1) for all f ∈
C0 ((0, 1], F ). If V : F → B is completely positive, then Sn := V ◦ W is nuclear
and is contained in C by assumption. 

Definition 3.2.2. Let A and B C *-algebras and CP(A, B) the cone of com-
pletely positive maps from A into B.
A subset C of CP(A, B) is a matrix operator-convex cone of c.p. maps, if
C has the following properties (i) and (ii):

(OC1) d∗1 V1 (.)d1 + d∗2 V2 (.)d2 ∈ C for V1 , V2 ∈ C, d1 , d2 ∈ B ( 14 )

(OC2) The map a ∈ A 7→ c∗ V ⊗ idn (r∗ ar) c is in C for every V ∈ C, every

row-matrix r ∈ M1,n (A) and every column-matrix c ∈ Mn,1 (B).

(I.e., for every V ∈ C, n ∈ N, r1 , . . . , rn ∈ A and c1 , . . . , cn ∈ B, the
completely positive map W (a) := ij c∗i V (ri∗ arj )cj is again in C.)
P

We abridge “matrix operator convex cone” as “m.o.c. cone” .

The m.o.c. cone C closed if C ⊆ CP(A, B) ⊆ L(A, B) is closed with respect
to topology of point-norm convergence on L(A, B) (= strong operator topology on
L(A, B)).
An m.o.c. cone C is non-degenerate if, for every b ∈ B+ and ε > 0, there are
V ∈ C and a ∈ A+ with kV (a) − bk < ε .
The cone C is faithful if a ∈ A+ and V (a) = 0 for all V ∈ C implies a = 0.
Let S be a subset of CP(A, B). We denote by Calg (S) the smallest subset of
CP(A, B) that is invariant under the operationsin (OC1) and (OC2), and by C(S)
the point-norm closure of Calg (S) (i.e., the closure of Calg (S) in L(A, B) w.r.t. the
strong operator topology). Then Calg (S) and C(S) are operator-convex cones of
completely positive maps.
We say that S generates a closed m.o.c. cone C if C = C(S).

14 Thus P b∗ V (.)b ∈ C if V , . . . , V ∈ C and b , . . . , b ∈ B.

i i i 1 n 1 n
376 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

One can see, with help of approximate units in A or B, that every point-norm
closed subset V of of CP(A, B) is a convex cone in the usual sense if it satisfies
(OC1), and that V satisfies tV ⊆ V for t ∈ [0, ∞) if V satisfies (OC2).

Remark 3.2.3. Let S ⊆ CP(A, B) any set of c.p. maps. We define the (OC2)–
hull hSi2 of S as smallest set hSi2 ⊂ CP(A, B) of c.p. maps containing S and being
invariant under the operations in (OC2) of Definition 3.2.2.
The point norm-closure C2 := hSi2 in CP(A, B) of the (OC2)-hull of S is in
general not convex.
In particular such C2 does not satisfy the (operator convexity) property (OC1)
of Definition 3.2.2 in general, and this even if S itself is convex :
Consider e.g. S := R+ · η for η : A := C ∗ (p11 − p22 ) ,→ B := M2 . Then
hSi2 = {X ∗ η(·)X ; X ∈ M2 } has non-convex closure in CP(A, B), because the
set {X ∗ p11 X ; X ∈ M2 } is the closed set that contains 02 and all positive matrices
in M2 of rank one. It is certainly not convex.

Definition 3.2.4. A Hilbert (A, B)-module (E, φ : A → L(E)), with right

Hilbert B-module E, generates the closed m.o.c. cone C, if C := C(S) for the set
S of completely positive maps Vx : a ∈ A 7→ hφ(a)x, xi ∈ B.
We say that (E, φ : A → L(E)) defines a closed m.o.c. cone C if (E, φ) generates
C and, moreover, (E, φ) is unitarily equivalent to (E ⊕B E, φ ⊕ φ).

(Does the latter mean that φ(A)0 ∩ L(E) contains a copy of O2 unitally?)
Need but have not the following:
Let V, W ∈ CP(A, B) with separable C *-algebra A.
To be shown: Then V and W are in the m.o.c. cone generated by V + W .
(Could work with the tensor V ⊗max idC ∗ (F∞ ) criterium ?)
Direct way ??
Question: Are V and W “coefficients” of the Hilbert A-B–module generated by
V + W ? Need an absorption theorem for Hilbert A-B-bi-module, like for Hilbert
B-modules [73, thm. 13.6.2].
Suppose that E is a Hilbert B-module and that φ : A → L(E) a C *-morphism.
Let C ⊆ CP(A, B) denote the point-norm closed m.o.c. cone generated by the
“coefficients” Vx : a ∈ A 7→ hφ(a)x, xi ∈ B for x ∈ E.
Let E ∞ the Hilbert-B-module given by the sequences (x1 , x2 , . . .) with xk ∈ E
P∞
and k=1 hxk , xk i ∈ B .
Define φ∞ : A → L(E ∞ ) by φ∞ (a)(x1 , x2 , . . .) := (φ(a)x1 , φ(a)x2 , . . .)
Then the set of “coefficient” maps A 3 a 7→ hφ∞ (a)y, yi ∈ B for y ∈ E ∞ is an
m.o.c. cone C(φ∞ ) ⊆ CP(A, B).
This cone is in particular σ-additively and “homogenous ??” closed.
 2. ON APPROXIMATE INNER NUCLEAR MAPS 377

(not so likely) CONJECTURE (1):

The family of “coefficients” is hereditary in the following sense:
If V : A → B is a c.p. map and has the property that there exists a c.p. map
W : A → B such that V +W is a “coefficient” of a representation ϕ : A → L(HB ) ∼
=
M(B ⊗ K) , then V and W are “coefficients” of ϕ.
(very likely) CONJECTURE (2):
Suppose that A and B are stable C *-algebras, where A is separable and B
is σ-unital. φ1 , φ2 : A → M(B) C *-morphisms such that there exists a unitary
u ∈ M(B) such that δ∞ (φ1 (a))u − uδ∞ (φ1 (a)) ∈ B for all a ∈ A.
Then the “coefficients” b∗ φ1 (·)b and b∗ φ2 (·)b (for all b ∈ B together!) generate
the same point norm closed m.o.c. cone C ⊆ CP(A, B).
(By a result in Chp. 5 one gets – conversely ??? – that δ∞ ◦ φ1 and δ∞ ◦ φ2
are unitarily homotopic in M(B).)
Urgently needed / speculative Conjectures:
If A is σ-unital then C(φ∞ ) is closed with respect to the point-norm topology.
This is not true? Counterexample: Let A := C([0, 1]), B := C and φ : A →
P
L(`2 ) = M(B ⊗ K) the representation generated by φ(f ) = n f (ρ(n))en for some
map ρ from N onto the rational numbers in [0, 1]. Then the point-norm closure
of of the algebraic C(φ∞ ) defined by the set of “coefficients” does not contain the
R1
c.p. map f ∈ A 7→ 0 f (t)dt ∈ C. But this map is in the point norm closure of
C(φ∞ ) ⊆ CP(A, C).
Need here more precise definitions !!
Perhaps use the following 3 lemmata (if true !):
1. Let K a convex cone in C(Cn , B)+ that satisfies that b∗ Kb ⊆ K for all
b ∈ B and that K is σ-additive in the sense that if V1 , V2 , . . . ∈ K and W (a) :=
n
P
n Vn (a) ∈ B for all a ∈ C then W ∈ K. Then K is point-norm closed.

2. Let A a separable C *-algebra, C ⊆ CP(A, B) an m.o.c. cone with the

additional property that C is σ-additive – in the sense that if V1 , V2 , . . . ∈ C and
X
W ⊗ id2 (a) := Vn ⊗ id2 (a) ∈ B ⊗ M2 for all a ∈ A ⊗ M2 ,
n

then W ∈ C.
Then C is point-norm closed.
(Unfortunately it is not true in case A = C([0, 1]), B := C ???)
3. Something like this being “almost hereditary”:
Let C ⊆ CP(A, B) a point-norm closed m.o.c. cone, W ∈ CP(A, B) and V ∈ C
such that for each T ∈ (A ⊗ Mn )+ there exists γT,n ∈ (0, ∞) with (W ⊗ idn )(T ) ≤
γT,n V (T )
and ???? then W ∈ C ???? or?
378 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

Remark 3.2.5. Suppose that A is separable and stable, B is σ-unital and stable
and that S ⊆ CP(A, B) ⊆ L(A, B) is
a separable subset of L(A, B) with respect to point-norm topology ????, ????
or is “countably generated as m.o.c. cone”
in CP(A, B) ????
in the strong operator topology on L(A, B) ( 15 ).
Then, by Corollary 5.4.4, there is a C *-morphism H0 : A → M(B) such that

(i) H0 is unitarily equivalent to its infinite repeat δ∞ ◦ H0 ,

(ii) the “coefficients” A 3 a 7→ b∗ H0 (a)b ∈ B are all in the point-norm closed
m.o.c. cone C(S) ⊆ CP(A, B) generated by S, and
(iii) for each V ∈ S there exists x ∈ B with V (a) = hH0 (a)x, xi for all a ∈ A.
more ????:
Need: (But have only weaker results !)
All V ∈ C(S) have this property !
Means:
The set of “coefficients” is closed in the topology of point-norm con-
vergence.
(But this is not satisfied in general ... e.g. in case S ⊆ CP(A, C) is
generated by algebraical by a σ(A∗ , A) dense subset of the pure states of
A.)

In particular, H0 (A) ∩ B = {0}, and V ∈ C(S) if and only if there is a sequence

b1 , b2 , . . . in B with kbn k2 ≤ kV k such that V (a) = limn b∗n H0 (a)bn for all a ∈ A.
(This is a better property !!! But it does not say that V is itself a coefficient of of
∗
P
δ∞ ◦H0 . (The infinite repeat δ∞ on M(B) is given by δ∞ (a) := n sn a(sn ) where
∗
P
s1 , s2 , . . . is a sequence of isometries in M(B) with n sn (sn ) strictly convergent
to 1M(B) , cf. Remark 5.1.1(8).)
H0 is determined up to unitary homotopy in the sense of Definition 5.0.1,
cf. Corollary 5.4.4.
(One has H0 = 0 if S = {0}.)
If A and B are separable then every m.o.c. cone C ⊆ CP(A, B) ⊂ L(A, B) is
separable with respect to the point-norm operator topology (– given by the semi-
norms V 7→ kV (a)bk on L(A, B) –), because L(A, B) is separable in the point-norm
operator topology.
Question: Is the point-norm closure identical with the cone defined by the
related kernel ideal of A ⊗max C ∗ (F∞ ) → B ⊗max C ∗ (F∞ ) ?

Lemma 3.2.6. Suppose that m1 , . . . , ms ∈ N and X1 , . . . , Xs are closed subsets

of (0, 1] with 1 ∈ Xj , j = 1, . . . , s .

15 It is equal to the topology of point-norm convergence of L(A, B).

 2. ON APPROXIMATE INNER NUCLEAR MAPS 379

Let A := C0 (X1 , Mm1 ) ⊕ · · · ⊕ C0 (Xn , Mms ) , F := Mm1 ⊕ · · · ⊕ Mms , let

λ : C0 ((0, 1] ⊗ F ) → A the natural epimorphism given by

λ(f ⊗ (α1 ⊕ · · · ⊕ αs )) := ((f |X1 ) ⊗ α1 ) ⊕ · · · ⊕ ((f |Xn ) ⊗ αs ) ,

let π1 : A → F the epimorphism given by π1 (g1 ⊕ · · · ⊕ gs ) := g1 (1) ⊕ · · · ⊕ gs (1),

and let S ⊆ CP(A, B) a subset with the property that
for mutually orthogonal a1 , . . . , ak ∈ A+ with kai k = 1, b ∈ B+ with kbk = 1 and
ε > 0, there are d1 , . . . , dk ∈ B and S ∈ S with

kd∗i S(aj )di − bk < ε for i = 1, . . . , k .

Let C denote the point-norm closure of the set of c.p. maps T : A → B such that
there are S ∈ S, n ∈ N, a row r ∈ M1,n (A) and a column c ∈ Mn,1 (B) with
T (·) = c∗ S(r∗ (·)r)c.

(o) C contains the c.p. map T = V ◦ π1 if V : F → B is of elementary type,

i.e., V ◦ ηj : Mj → B is given by a column b ∈ Mn,1 (B) in the sense
V ◦ ηj (α) = b∗ αb, where ηj (α) := 0 ⊕ · · · ⊕ α ⊕ · · · ⊕ 0 with α at position
j.
(i) If, for every positive contraction b ∈ B+ and ε > 0, there are contractions
b1 , b2 ∈ B with b∗2 b1 = 0 and kb∗j bj b − bk < ε, then CP(F, B) ◦ π1 ⊆ C,
i.e., C contains all c.p. maps T : A → B with T (λ(f0 (1 − f0 ))) = 0, i.e.,
V ◦ π1 ∈ C for all V ∈ CP(F, B) .
(ii) If C is convex, then CP(F, B) ◦ π1 ⊆ C.
(iii) If 1 is not isolated in Xj for every j = 1, . . . , k, then CP(F, B) ◦ π1 ⊆ C.

For the parts (o)–(ii) one needs only the assumption on S with k = 1. The
assumption with k = max(m1 , . . . , ms ) is needed for the proof of part (iii).

Proof. We consider the point-norm closed subset C1 of CP(F, B), consisting

of the maps V ∈ CP(F, B) with the property that there is a sequence Tn ∈ C such
that limn Tn (f ) = V (f (1)) for all f ∈ C0 ((0, 1], F ) .
The set C1 is invariant under the operations (OC2) of Definition 3.2.2, because C
is invariant under the operations (OC2) and π1 : f 7→ f (1) is an epimorphism. With
other words, V = c∗ W (r∗ (·)r)c ∈ C1 if W ∈ C1 and c ∈ Mn,1 (B) and r ∈ M1,n (F ) .
In particular, we have that V (p(·)) ∈ C1 and V1 (p1 (·)) + V2 (p2 (·)) ∈ C1 for
central projections p, p1 , p2 ∈ F with p1 p2 = 0, and for V, V1 ; V2 ∈ C1 . (This
reduces also the proof to the case where F is a full matrix algebra, i.e., to the case
s = 1.)
(o): We denote the (upper left corner) minimal projection of ηj (Mmj ) ∼ = Mmj
by qj ∈ F , and let ρj : F → C denote the (unique) pure state with ρj (qj ) = 1.
Recall that ρj (g)qj = qj gqj for g ∈ F . Let ρ = ρ1 + . . . + ρs . (The qj have pairwise
orthogonal central supports pj .)
If V : F → B is an
380 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

Def. of ‘‘elementary’’ in Part(o):

“ V ◦ ηj : Mj → B is given by a column b ∈ Mn,1 (B) in the sense V ◦ ηj (α) = b∗ αb,
where ηj (α) := 0 ⊕ · · · ⊕ α ⊕ · · · ⊕ 0 with α at position j.”
elementary c.p. map and δ > 0, then there exists a contraction e ∈ B+ , such
that V = c∗ W (r∗ (·)r)c for W = ρ(·)e some c ∈ Mt,1 (B) and r ∈ M1,t (F )
Why? More precise? Part? (cf. Prop. 3.1.9).
Prop. 3.1.9 has been changed!!!
Thus, it suffices to show that, for b ∈ B+ with kbk = 1, ε > 0 and j ∈ {1, . . . , s},
there exists Vj ∈ C1 and cj ∈ B with kc∗j Vj (pj )cj − ek < ε. Indeed, this implies that
the point-norm closed C1 actually “contains” ρj (·)b, and then W, V ∈ C1 , because
the central supports pj of the qj are orthogonal.
Define a c.p. map W : F → C0 ((0, 1], F ) by W (g) := f0 ⊗ g for g ∈ F , where
f0 (t) = t for t ∈ [0, 1]. Let q := qj and define

hδ := δ −1 ((f0 − (1 − δ))+ ) ⊗ 1 ∈ C0 ((0, 1], F )+ .

Then aδ := W (q)hδ is positive and kaδ k = 1 , thus, there is Sδ ∈ S and dδ ∈ B

with kd∗δ Sδ (a2δ )dδ − bk < δ . Note that

Tδ (g) := (dδ )∗ Sδ (aδ gaδ )dδ for g ∈ C0 ((0, 1], F )

defines a map in C and that limδ→0 kTδ (g) − ρj (g(1))bk = 0 for all g ∈ C0 ((0, 1], F ).
Hence ρj (·)b ∈ C1 .
(i): By part (o), all
Why the Lemma applies? ??? Definition? of :
“elementary”
c.p. maps V : F → B are contained in C1 . The supposed property of B implies
that the elementary c.p. maps are dense in CP(F, B) by Proposition 3.1.9(ii). ???
Prop. 3.1.9(ii) = Old lem:3.3(ii) has been changed
(ii): This follows from part (o), because the set of V ∈ CP(F, B) with V ◦π1 ∈ C
contains all elementary maps V and is convex by the proposed convexity of C.
(iii): If 1 is not isolated in Xj (for j ∈ {1, . . . , s}), then we can find a strictly
increasing sequence 0 =: t0 < t1 < t2 < · · · in Xj such that lim tn = 1. Let
q := qj ∈ F , m := mj and p := pj the central support of q.
Revise proof!
Prop. 3.1.9(i) has been changed!!
We show that our general assumptions (with k ≥ m) implies that C1 contains
V (p(·)) = V ◦ πj for every V ∈ CP(Mm , B) :

By Proposition 3.1.9(i) there are columns c1 = [c11 , c12 , . . . , c1n ]> , . . . , cn ∈

Pn ∗
Mn,1 (B) such that V (α) = i=1 ci βci . There are contractions en ∈ B+ with
kci − (en ⊗ 1n )ci k < 1/n.
 2. ON APPROXIMATE INNER NUCLEAR MAPS 381

We identify C0 (0, 1] naturally with the center C0 (0, 1] ⊗ 1 of C0 ((0, 1], Mm ) ⊆

C0 ((0, 1], F ). We find (pice-wise linear) non-negative functions fn ∈ C0 (0, 1]+ with
support in [(tn−1 + tn )/2, (tn + tn+1 )/2] such that kfn k ≤ 1 and f (tn ) = 1.
Further there are functions bn ∈ C0 (0, 1]+ with support in [tn−1 , 1] such that
0 ≤ bn ≤ 1 and bn |[(tn−1 + tn )/2, 1] = 1. Then bn bn+1 = bn+1 and bn fn+j = fn+j
for all j ≥ 0.
Let W (g) := f0 ⊗ g
Where/ What was Part (o) ???
(as in proof of (o))
and an := W (q)fn ∈ C0 (0, 1] ⊗ q. Then kan k = tn and the an are mutu-
ally orthogonal. Therefore (and by assumptions on S), there exist Sn ∈ S and
dn,0 , . . . , dn,m−1 ∈ B with

k(dn,j )∗ Sn (an+j )2 dn,j − (tn+j )2 en k < 1/n

∀ j ∈ {0, . . . , n − 1} .

Consider the c.p. maps Tn (g) := Zn∗ Sn (Rn∗ gRn )Zn with column Zn ∈
Mm2 ,1 (B) with transposed row (Zn )> given by

(Zn )> := [dn,0 · (c1 )> , . . . , dn,m−1 · (cm )> ] = [dn,0 c11 , dn,0 c21 , . . . , dn,m−1 cmm ]

and row Rn ∈ M1,m2 (C0 ((0, 1], Mm )) given by

(n) (n)
Rn := [r1 , . . . , rm ],
(n)
where rj := [(1 ⊗ e1,1 ) · an+j−1 , . . . , (1 ⊗ em,1 ) · an+j−1 ] which is equal to
fn+j−1 · [(f0 ⊗ e1,1 ), . . . , f0 ⊗ en,1 ] .
Here ei,k denote the matrix units of Mm . Note that q = e1,1 and [αij q] =
x∗ αx ∈ Mm (Mm ) for α ∈ Mm and x := [e1,1 , . . . , em,1 ] ∈ M1,m (Mm ). It follows
XX
Tn (f ⊗ α) = ( αik c∗ij d∗n+j−1 Sn (an+j−1 (f ⊗ q)an+j−1 )dn+j−1 ckj )
j i,k

for f ∈ C0 (0, 1] and α ∈ Mm . check? to be filled in

??
The c.p. maps d∗n Sn (an (·)an )dn have norms ≤ 1 + 1/n ≤ 2. Since 0 ≤ g ≤
f ⊗ 1m ≤ kgk for f (t) = kg(t)k if g ∈ C0 ((0, 1], F )+ , we can see that kTn k ≤
2mkV k. For ε > 0 there is n0 with kf bn − f (1)bn k < ε/(2m)3 for n ≥ n0 . Since
Tn ((f bn ) ⊗ α) = Tn (f ⊗ α) and Sn (an+j−1 (bn ⊗ q)an+j−1 ) = Sn (a2n+j−1 ), it follows
that, for n ≥ n0 ,
XX
kTn (f ⊗ α) − f (1) αik c∗ij en+j−1 ckj k ≤ max(|αik |) · m3 (1/n + 2ε/(2m)3 )
j ik

Thus limn kTn (f ⊗ α) − f (1)V (α)k = 0 for all f ∈ C0 (0, 1] and α ∈ Mm . Since
sup kTn k < ∞, it follows limn kTn (g) − V (g(1))k = 0 for every g ∈ C0 ((0, 1], F ) .
By our construction, Tn ∈ C. Hence V ∈ C1 . 
382 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

Lemma 3.2.7. Suppose that S ⊆ CP(A, B) is a set of completely positive maps

such that, for every k ∈ N, a1 , . . . , ak ∈ A+ with kaj k = 1 and ai aj = 0 for i 6= j,
every b ∈ B+ with kbk ≤ 1 and every ε > 0, there exist d1 , . . . , dk ∈ B and S ∈ S
with kd∗j S(aj )dj − bk < ε for j = 1, . . . k .
Let C ⊆ CP(A, B) denote the point-norm closure of the (OC2)-hull hSi2 of S,
i.e., let C denote the point-norm closure of the set of c.p. maps T : A → B such
that there are S ∈ S, n ∈ N, a row r ∈ M1,n (A) and a column c ∈ Mn,1 (B) with
T (·) = c∗ S(r∗ (·)r)c .
Then we get that ????
CPnuc (A, B) ⊆ C, i.e., each nuclear c.p. map V : A → B can be approximated
in point-norm by maps T := c∗ S(r∗ (·)r)c with S ∈ S .
in each of the following cases (i), (ii) or (iii):

(i) If, for every b ∈ B and ε > 0, there exist e1 , e2 ∈ B with e∗1 e2 = 0 and
kb − e∗j ej k < ε for j = 1, 2.
(ii) If C is convex.
(iii) If socle(A) = {0} .

The condition (i) is e.g. satisfied if B stable, or if B contains an approximate

unit {eα } of properly infinite contractions.
The proofs of the cases (i), (ii), and (iii) are rather different, e.g. in the proof
of (i) and (ii), we use only the case k = 1. In a sense the assumption on S
is almost necessary, because if b ∈ B+ and orthogonal a1 , . . . , ak ∈ A+ of norm
one are given, then there are pure states ρ1 , . . . , ρk on A with ρj (aj ) = 1. Then
S(a) := ( j ρj (a))b1/3 and dj := b1/3 satisfy the assumptions on S, e.g. in case
P

S := {f (·)b ; f ∈ A∗+ , b ∈ B+ }.

Proof. (i, ii): We check the criteria for CPnuc (A, B) ⊆ C of Lemma 3.2.1.
Let F finite-dimensional and ψ : C0 ((0, 1], F ) → A a *-morphism such that kψ(f0 ⊗
p)k = 1 for all non-zero projections p ∈ F . Then the quotient ψ(C0 ((0, 1], F )) of
C0 ((0, 1], F ) and the restrictions of V ∈ S satisfy the assumptions of Lemma 3.2.6,
because (OC2)-hull of S ◦ ψ is point-norm dense in C ◦ ψ for the (OC2)-hull of
S ⊆ CP(A, B).
Thus the criteria is satisfied in the cases (i) and (ii) by Lemma 3.2.6(i,ii).
Hence, CPnuc (A, B) ⊆ C under the additional assumptions of parts (i) and (ii) by
This proves (i) and (ii).
(iii): We use Lemma 3.2.6(iii) to show that the criteria of Lemma 3.2.1 is
satisfied. Again let F finite-dimensional and ψ : C0 ((0, 1], F ) → A a *-morphism
such that kψ(f0 ⊗p)k = 1 for all non-zero projections p ∈ F . Unfortunately, 1 could
be isolated in the spectrum of ψ(f0 ⊗p) for some minimal non-zero projection p ∈ F .
 2. ON APPROXIMATE INNER NUCLEAR MAPS 383

∼ Mm ⊕ · · · ⊕ Mm and let ψj : C0 ((0, 1], Mm ) → A the *-morphisms

Let F = 1 k j

with ψ(g) = ψj ◦ πj (g) for j = 1, . . . , k, where πj denotes the natural epimorphism

πj : F → Mm j .
The hereditary C *-subalgebras Dj of A generated by ψj (C0 ((0, 1], Mmj )) are
mutually orthogonal. Since socle(A) = 0 (by assumption), socle(Dj ) = 0 and
Lemma A.10.3 says that there exist increasing continuous maps λj : [0, 1] → [0, 1]
with λj (0) = 0 and λj (1) = 1, *-morphisms ϕj : C0 ((0, 1], Mmj ) → Dj with ϕj (f ◦
λj ) = ψj (f ) for f ∈ C0 ((0, 1], Mmj ), such that 1 is not isolated in the spectrum
of ϕj (f0 ⊗ 1) and ϕj (f0 ⊗ 1)ϕi (f0 ⊗ 1) = 0 for i 6= j. Then the image of ϕ(g) :=
P
ϕj (πj (g)) satisfies the assumptions of Lemma 3.2.6(iii), i.e., 1 is not isolated in
X1 , . . . , Xn ⊆ (0, 1] with ϕ(C0 ((0, 1] ⊗ F )) ∼ = C0 (X1 , Mm1 ) ⊕ · · · ⊕ C0 (Xk , Mmk ).
Thus Lemma 3.2.6(iii) applies to ϕ, and for every V ∈ CP(F, B) there is a sequence
Sn ∈ C with limn Sn (ϕ(g)) = V (g(1)) for all g ∈ C0 ((0, 1], F ). By Lemma 3.1.8,
it follows from the (OC2)-invariance of C that sequence Sn can be chosen with
kSn k ≤ kV k . Thus Sn |D has norm ≤ kV k, where D denotes the hereditary C *-
algebra generated by ϕ(f0 ⊗ 1). In particular, limn Sn (ϕ((f0 (1 − f0 ) ⊗ 1)) = 0.
Since Sn is positive, it follows limn Sn (a) = 0 for all a ∈ E, where E denote the
hereditary C *-subalgebra of A generated by ϕ((f0 (1 − f0 ) ⊗ 1). By Lemma A.10.3,
P
ψ(g) − ϕ(g) = j ϕj (πj (g ◦ λj ) − πj (g)) ∈ E for g ∈ C0 ((0, 1], F ). It follows
limn Sn (ψ(g)) = limn Sn (ϕ(g) + e) = V (g(1)). Thus, the criteria of Lemma 3.2.1 is
satisfied. 

Corollary 3.2.8. The set CPnuc (A, B) of nuclear c.p. maps is a point-norm
closed m.o.c. cone in the sense of Definition 3.2.2.
It is the point-norm closure of the smallest subset S ⊆ CP(A, B), that
Pn
(a) contains all maps a 7→ j=1 ρj (a)bj with pure sates ρj on A (j = 1, . . . , n)
and bj ∈ B+ , and
(b) is closed under the operation (OC2) in Definition 3.2.2.

If socle(A) = {0}, then it suffices to consider only pure states ρj on A that have
pairwise non-equivalent irreducible representations Dj : A → L2 (A, ρj ).

Pn
Proof. The set of the maps a 7→ j=1 ρj (a)bj is convex and satisfies the
assumptions of Lemma 3.2.7(ii). 

Remark 3.2.9. A first application of Lemma 3.2.7(ii) is the following observa-

tion:
Suppose that C, D ⊆ M(B) are C*-subalgebras of the multiplier algebra M(B) of
B, such that DB is dense in B and CD ⊆ D, and that D is simple. Then every
nuclear map V : C → B is approximately inner, i.e., can be approximated in point-
norm topology by inner c.p. maps c ∈ C 7→ j b∗j cbj ∈ B (with b1 , . . . , bn ∈ B).
P

(Indeed, the cone CPin (C, B) of inner c.p. maps c 7→ j b∗j cbj satisfies (OC1)
P

and (OC2). The assumptions of Lemma 3.2.7(ii) are satisfied, because D is a simple
non-degenerate C *-subalgebra of M(B) and C ⊆ M(D) ⊆ M(B):
384 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

Indeed, if a1 , . . . , ak ∈ C+ are non-zero, pair-wise orthogonal and have norms

kaj k = 1, b ∈ B+ , then, for ε > 0, there are e0 , e1 , . . . , ek ∈ D with e∗i aj ei =
P 1/2 ∗
δij e0 , ke0 k = 1 and f1 , . . . , fm ∈ C with kb − b fj e0 fj b1/2 k < ε (because
C ⊆ M(D) ⊆ M(B), D is simple and DB is dense in B). Thus, T (c) :=
1/4 ∗
) c(ei fj b1/2 ) and di = b1/4 satisfy the assumptions of Lemma 3.2.7.)
P
i,j (ei fj b

Next related to WvN-Thm. form Elliott and Kucerovsky .

Give precise reference to it!!!

Lemma 3.2.10. Suppose that C ⊆ M(B) is a C*-subalgebra, J / C is a closed

ideal and that for every c ∈ C+ with kc + Jk = 1, every b ∈ B+ and ε > 0 there
exists d = d(c, b, ε) ∈ B with d∗ cd ≥ (b − ε)+ .
Let V : C → B a c.p. map with V (J) = 0 such that [V ]J : C/J → B is nuclear.
Then V is approximately one-step inner in M(B).
?...in the sense that there exists a net of elements {dτ } ⊆ B with V (c) =
limτ d∗τ cdτ for each c ∈ C. In particular, limτ d∗τ cdτ = 0 for all c ∈ J....?

Proof. We let A := C/J and let C ⊆ CP(A, B) denote the set of c.p. maps
V : A → B with the property that V ◦ πJ : C → B is one-step approximately inner,
i.e., that for c1 , . . . , ck ∈ C+ and ε > 0 there is d ∈ B with kdk2 ≤ kV k and
kV (πJ (cj )) − d∗ cj dk < ε for j = 1, . . . , k. Clearly, C is closed under the operations
(OC2) of Definition 3.2.2.
If A := C/J ∼ = C then A = C · q. Let V : A → B completely positive, W :=
V ◦ πJ , b := V (q), and f ∈ C+ a contraction with πJ (f ) = q. For c1 , . . . , cn ∈ C+
and δ > 0 there is a contraction e ∈ J+ with k(1 − e)(f ck f − αk f 2 )(1 − e)k < δ if
αk p = πJ (ck ), because f ck f −αk f 2 is in J. Let d ∈ B with kb−d∗ (1−e)f 2 (1−e)dk <
δ, then kW (ck ) − g ∗ ck gk < 2δ for k = 1, . . . , n.

Suppose now that A := C/J has (linear) dimension ≥ 2, then a maximal

commutative C *-subalgebra F of A has dimension ≥ 2. Thus there are a1 , a2 ∈ A+
with a1 a2 = 0 and kai k = 1 for i = 1, 2. There is c ∈ C with c∗ = c, kck = 1 and
π(c) = a1 − a2 . Let c1 := c+ and c2 := c− . Then πJ (ci ) = ai for i = 1, 2.
Let b ∈ B and ε > 0, then there are (by assumptions) di ∈ B with d∗i (ci )2 di ≥
(b − δ)+ for δ := ε/3. By Lemma 2.1.9 there are contractions fi ∈ B such that
e∗i ei = (b − 2δ)+ for ei = ci di fi ∈ B (i = 1, 2), and e∗1 e2 = 0.
Thus suffices to show, that A, B and S satisfy the assumptions of Lemma
3.2.7(i) (where the case k = 1 is sufficient as the proof of Lemmas 3.2.6(i) and
3.2.7(i) shows), i.e., it suffices to show that, for (given) a0 ∈ A+ with ka0 k = 1,
b ∈ B+ with kbk ≤ 1 and ε > 0, there is S ∈ C with kS(a0 ) − bk < ε.
Let c0 ∈ C+ a contraction with πJ (c0 ) = a0 , and let ϕ ∈ A∗ a pure state on
A with ϕ(a0 ) = 1. Then ψ := ϕ ◦ πJ is a pure state on C with ψ(J) = 0. Let
V (a) := ψ(a)b . It suffices to show that W := V ◦ πJ is approximately inner. Note
that W (c) = ψ(c)b .
 2. ON APPROXIMATE INNER NUCLEAR MAPS 385

By Lemma ??, there exists a directed net G ⊆ C of positive contractions

g ∈ G with ψ(g) = 1 such that limg→G kgcg − ψ(c)g 2 k = 0 for all c ∈ C. Thus, for
c1 , c2 , . . . , cn ∈ C and ε > 0 (and δ := ε/M for M := (2+2 maxj kcj k)) there exists
a contraction g ∈ C+ with ψ(g) = 1 and kgcj g − ψ(cj )g 2 k < δ for j = 1, . . . , n.
Since kπJ (g)k ≥ ψ(g) = 1, there is d ∈ B with δ −1 d∗ (g 2 − (1 − δ))+ d = (b − δ)+ (by
assumption and by Lemma 2.1.9). Let f := δ −1/2 (g 2 − (1 − δ))+ d. Then kf k ≤ 1,
and for j = 1, . . . , n,

k(gf )∗ cj (gf ) − ψ(cj )bk ≤ δ(1 + 2 max kcj k) < ε .

j

HERE in next lem.: ‘‘compact’’ and ‘‘nuclear’’ in what sense?

The point is the question what kind of approximation by factorable
maps is required,
i.e., what kind of topology in the set of c.p. contractions has to be
taken,
and what kind of c.p. maps S into Mn are accepted?
Strictly continuous S?,
uniformly continuous on which kind of subsets?
all compact subsets, ... norm-continuous anyway ...
Minimal: approx on finite subsets ...,
but ‘‘uniformly’’ continuous on the base sets of the topology.
Classic: point-wise uniform on finite sets.

Lemma 3.2.11. Let C be a separable C*-subalgebra of the multiplier algebra

M(D) of a C*-algebra D (respectively of a von-Neumann algebra M ) and V : C →
B a nuclear completely positive contraction into a C*-algebra B.
Then for every compact subset Ω ⊆ C, every ε > 0 there exist n ∈ N and
completely positive contractions S : M(D) → Mn (respectively S : M → Mn ),
T : Mn → B, such that

kV (c) − T S(c)k < ε for c ∈ Ω,

and S is unital and continuous w.r.t. the strict topology on M(D) (respectively S
is unital and σ(M, M∗ )-norm continuous).

Proof. The multiplier algebra M(D) is unitally contained in the W*-algebra

D with pre-dual D∗ = (D∗∗ )∗ . The natural embedding is continuous w.r.t. to the
∗∗

strict topology on M(D) on bounded parts and w.r.t. the σ(D∗∗ , D∗ )-topology on
D∗∗ . In particular, S|M(D) is unital and strictly continuous for every ultra-weakly
continuous u.c.p. map S : D∗∗ → Mn . Thus, we can restrict to the case of C ⊆ M ,
where M is a von-Neumann subalgebra of L(H) for some Hilbert space H. Notice
that Su (a) := u∗ (a ⊗ 1)u ∈ L(Cn ) ∼
= Mn defines an ultra-weakly continuous u.c.p.
map Su : M → Mn if u : Cn → H ⊗ `2 (N) is an isometry.
386 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

Let {c1 , . . . , cm } ⊆ Ω a δ-dense subset of Ω for δ := ε/4. By Definition 3.1.1,

Lemma 3.1.8 and Lemma 3.1.11(ii), there are m ∈ N and completely positive con-
tractions S0 : C → Mm and T0 : Mm → B, such that kV (cj ) − T0 (S0 (cj ))k < δ for
j = 1, . . . , m.
Since C ⊆ M ⊆ L(H) we can extend S0 to a c.p. contraction W : L(H) → Mn
(e.g. by the Arveson extension theorem [42]).
The natural extensions of the irreducible representations of D to irreducible
representations of M(D) are strictly continuous if one considers them with the
*strong operator topology. The family of this particular irreducible representations
of M(D) is separating for M(D). If we apply Lemma 3.1.12(i) to this class of
irreducible representations of M(D), then we get a strictly continuous unital c.p.
map S1 : M(D) → Mm1 ⊕ · · · ⊕ Mmk =: F and a c.p. contraction T1 : F → Mn
with kW (cj ) − T1 (S1 (cj ))k < δ for j = 1, . . . , n. Thus kV (c) − T (S(c))k < ε for
c ∈ Ω and S := T1 ◦ S1 . 

The following Lemma 3.2.12 is only 1× really cited and used in

the proof of Lemma ??. Formulations and proofs have to be improved
in both cases.
In the next proofs we use the property of non-elementary simple C *-algebras
stated in the following Lemma 3.2.12. Here [ · ] denotes matrices of operators,
and k · k means their norms. In part (iv) we put two consecutive rows [fik ]i
(i = 1, . . . , kn , k = n, n + 1) together to a row [hin ]i of length kn + kn+1 .

Recall that a simple C *-algebra A is “non-elementary” if A is not isomorphic to

the algebra of compact operators on a Hilbert space (and is non-zero). The point
of the following Lemma 3.2.12 is that the irreducible representation Dϕ : A →
L(A/Lϕ ) is faithful on A but has the property that A ∩ K(A/Lϕ ) = {0}.

Lemma 3.2.12. Let A be a non-elementary simple C*-algebra, ϕ be a pure state

on A and Y1 ⊆ Y2 ⊆ . . . a sequence of subsets of the contractions in the multiplier
algebra M(A) such that each Yn is compact with respect to the strict topology on
M(A).
(This topology on M(A) is given by the semi-norms T ∈ M(A) 7→ kT ak for
a ∈ A?
Check by Def.’s given in Chp. 5.
If A s unital then we could take Xn = {1} ??? What is here Xn ? Is it Yn ?)
Where the simplicity plays a role?
Are ‘‘strictly’’ compact subsets of the unit ball of M(A)
(i.e., compact with respect to the strict topology)
separable subsets of M(A) in norm topology?
It seems that this is supposed implicit below.
Have only that this sets are separable w.r.t. strict topology on
M(D) if D isσ-unital.
 2. ON APPROXIMATE INNER NUCLEAR MAPS 387

Is the unit ball of L(`2 ) in the strong* operator topology compact?

It is at least weakly compact ...
Let Sn : M(A) → Mkn a sequence of strictly continuous unital completely pos-
itive maps from M(A) into Mkn .
Does it say that the Sn are ‘‘supported’’ in A?
And that the Sn are the unique strictly continuous extensions of
Sn |A?
If ‘‘yes’’ then we find positive contractions em,n ∈ A+ with
k1kn − Sn (em,n )k < 2−(m+n) ,
This could the lead to the a reduction to the case
where A is σ-unital?

Then there exist contractions fin ∈ A (n = 1, 2, . . . , i = 1, . . . , kn ), b ∈ A+ and

a sequence m1 < m2 < . . . of positive integers such that

∗
(i) k[ϕ(fin afjn )]ij − Sn (a)k < 4−n for a ∈ Yn ,
∗
(ii) k[ϕ(fim afjn )]ij k < 4−m for a ∈ Yn and m < n, i = 1, . . . , km , j =
1, . . . , kn ,
(iii) kbk = 1 and ϕ(b) = 1,
(iv) k[b` h∗in ahjn b` ]ij − [b2` ϕ(h∗in ahjn )]ij k < 4−n for a ∈ Yn and mn ≤ ` where
hjn = fim with i = j, m = n if j ≤ kn , and with i = j − kn , m = n + 1 if
kn < j ≤ kn+1 .

Proof. (i) and (ii): We can suppose that 1 ∈ Yn .

Since the unital completely positive maps Sn are strictly continuous, there are
sequences d`,n ∈ A+ such that kd`,n k = 1 and lim`→∞ Sn (d`,n ) = 1kn .
This implies that all Sn are supported on a σ-unital C *-subalgebra eAe of A
with {d`,n ; `, n ∈ N} ⊆ eAe.
Moreover, the Sn are restrictions to M(A) ⊆ A∗∗ of its unique normalizations
(i.e., weakly continuous extensions to A∗∗ ) of its restrictions Sn |A to A.
The compactness of the sets Yn with respect to the strict topology on M(A)
S
implies that the set Z := {e} ∪ {d`,n ; n, ` ∈ N} and the union n,m∈N Yn ZYm
generates a separable C *-subalgebra E of A, such that Yn E + EYn ⊆ E for n =
1, 2, . . . .
Above we claimed that there is and invariant separable C *-subalgebra E of A,
S
such that n Yn contained in the two-sided normalizer N (E, M(A)) ⊆ M(A) of E
in M(A).
At least we could find a σ-unital C *-subalgebra E of A such that the Sn are
supported on E, i.e., lim` kSn (e1/` ) − 1kn k = 0 for each n ∈ N if e ∈ E+ is a strictly
positive contraction in E.
(That would be sufficiant for the rest of the proof.)
388 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

If e ∈ E+ is a strictly positive contraction, then necessarily lim`→∞ Sn (e1/` ) =

1kn for n = 1, 2, . . . by strict continuity of the Sn : A → Mkn , i.e., lim` kSn (e1/` ) −
1kn )k = 0 in Mkn for each n ∈ N.
It follows that limk→∞ Sn (e1/k ae1/k ) = Sn (a) for a ∈ M(A) and n = 1, 2, . . . ,
because
kS(f bf ) − S(b)k2 ≤ 8kbk2 k1C − S(f )k
for every unital completely positive map S : B → C and f ∈ B+ with kf k ≤ 1.
The verification of the latter inequality can be reduced to the case C := C:
For every state ρ on (general unital) C and contraction c ∈ C+ , holds |1 −
ρ(c)| ≤ k1 − ck and there exists a pure state ρ on C with |ρ(S(f bf ) − S(b))| =
kS(f bf ) − S(b)k and such that the restriction of ρ to the Abelian C *-subalgebra
generated by S(f bf ) − S(b) is a character.
But if λ (e.g. λ := ρ ◦ S) is a positive functional on B with norm kλk ≤ 1 and
b, f ∈ B+ with kf k ≤ 1 then
|λ(f bf ) − λ(b)|2 ≤ 8kbk2 |1 − λ(f )| .

Proof of latter inequality:

We can suppose that B ⊆ L(H) and λ is defined by x ∈ H as λ(b) :=< bx, x >.
We let y := f 1/2 x and z := f x = f 1/2 y. Have kxk ≤ 1 and kzk ≤ 1. λ(f ) =<
y, y >, x − z = (1 − f )x
| < bx, x > − < bz, z > | = |λ(f bf ) − λ(b)|
|(< bx, x > − < bx, z >) + (< x, bz > − < z, bz >)| ≤ kbkkxkkx − zk +
kbkkzkkx − zk
≤ 2kbk(k(1 − f )xk) .
kx − f xk2 =< x, x > + < f x, f x > −2 < f x, x > = k(1 − f )xk2 =< (1 −
f ) x, x >=< x, x > −2 < f x, x > + < f 2 x, x >≤ 2(kxk2 − < f x, x >).
2

≤ 2(1− < f x, x >) if 0 ≤ f ≤ 1 and kxk ≤ 1.

Obtain estimate:

|λ(f bf ) − λ(b)| ≤ 2kbk(2(1 − λ(f )))1/2

From the Banach-Steinhaus (uniform boundedness) theorem we get the exis-

tence of constants Cn < ∞ with kak ≤ Cn for all a ∈ Yn . Thus we find gn ∈ N
such that kSn (en aen ) − Sn (a)k ≤ 8−n for en = e1/gn , a ∈ Yn .
It seems that here a quasi-central approximate unit en
would do on all places a much better job!
Ωn := e1 Y1 e1 ∪ · · · ∪ en Yn en is a compact subset of E ⊆ A.
We consider Mkn as naturally realized on Ckn . The irreducible representation
Rϕ defined by ϕ on H := L2 (A, ϕ) (with canonical cyclic vector x) is faithful and
 2. ON APPROXIMATE INNER NUCLEAR MAPS 389

Rϕ (A) ∩ K = 0, because A is simple. By Lemma 2.1.22, there is a sequence of

isometries vn : Cn → H such that kSn (a) − vn∗ avn k ≤ 8−n and kvn∗ avm k < 8−n
for a ∈ Ωn , m < n. By the Kadison transitivity theorem, we get contractions
pin ∈ A such that Rϕ (pin )x = vn (ei,kn ). Thus [ϕ(p∗jn apin )]ij = vn∗ avn in Mkn and
[ϕ(p∗jm apin )]ij = vn∗ avm in Mkn ,km = L(Ckm , Ckn ) for a ∈ M(A) and m < n,
where i = 1, . . . , kn , j = 1, . . . , km . The elements fin := en pin are as desired.
(iii) and (iv): The existence of b ∈ A+ with (iii) and (iv) follows from Lemma
∗
??, because the union of the sets fim Yk fjn generate a separable C *-subalgebra of
A by compactness of the sets Yk with respect to the strict topology. 

next Prop. and others overlap with chp. 5?

Proposition 3.2.13. Suppose that B is a C*-algebra, and that D ⊆ M(B)

is a simple, purely infinite, and non-degenerate C*-subalgebra of the multiplier
algebra M(B) of B, (i.e., DB = B).

Let C a separable C*-subalgebra of M(B) that satisfies CD ⊆ D.

If V : C → B is a nuclear completely positive contraction, then V can be approx-
imated by 1-step inner c.p. maps in the sense of following property (i), respectively
properties (i,ii) if B is unital:
(i)
(i) There exist sequences of contractions dn ∈ B (i = 1, 2, n = 1, 2, . . .) with
(i) (i) (i) (j)
limn→∞ (dn )∗ adn = V (a) for i = 1, 2, a ∈ C, and limn→∞ (dn )∗ adn =
0 , for i 6= j, a ∈ C + C1 .
(ii) If, moreover, B is unital, 1B ∈ C and V (1) = 1, then the sequences
(i)
(dn ) (i = 1, 2) in (i) can be chosen as isometries with orthogonal ranges:
(i) (j)
(dn )∗ dn = δij 1 .

Proof. The crucial role plays here that we have the intermediate simple purely
infinite C *-subalgebra D ⊆ M(B).
(i): We prove a more precise and general result: Let C ⊆ M(D), D ⊆ M(B)
with DB = B (i.e., 1M(B) ∈ M(D) ⊆ M(B)), D simple and purely infinite,
and let Ω1 ⊆ Ω2 ⊆ . . . ⊆ C be a sequence of norm compact subsets of C and
Wn : C → B nuclear completely positive contractions, then there exist contractions
d1 , d2 , . . . ∈ B such that kd∗n dn+1 k ≤ 2−n and kd∗n bdn − Wn (b)k < 2−n for b ∈ Ωn
and kd∗n bdn+1 k < 2−n for b ∈ Ωn , n = 1, 2, . . ..
Let V1 = . . . = V8 := W1 , Vn+8 := Wn . At first we find by Lemma 3.2.11
completely positive contractions Sn : M(D) → Mkn and Tn : Mkn → B such that
Sn is unital and strictly continuous, and kTn (Sn (b)) − Vn (b)k < 8−n for b ∈ Ωn .
Prop. 3.1.9(i) (was old a Lemma) has been changed!!
Moreover we can assume that Tn is of the form described in Proposition 3.1.9(i)
(by passing, if necessary, to Mkn ⊗ Mkn and Sn (.) ⊗ 1). That is, there exist
390 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

(n) (n)
e1 . . . ekn ∈ B with Tn (α) = En αEn∗ for α ∈ Mkn where En is the row matrix
(n) (n) Pkn (n) ∗ (n)
[(e1 )∗ , . . . , (ekn )∗ ] and En En∗ = j=1 (ej ) ej is a contraction.
The Lemma 3.2.12 is imprecise formulated and is used only here!!
Find a better source or an other proof here or there?
(n) (n)
By Lemma 3.2.12 there exist row matrices Fn = [f1 , . . . , fkn ] with elements
in D, b ∈ D+ and a sequence m1 < m2 < . . . of positive integers such that (with
bn := bmn )
k(1kn ⊗ bn )Fn∗ aFn (1kn ⊗ bn ) − Sn (a) ⊗ b2n k < 4−n 2
and
k(1kn ⊗ bn )Fn∗ aFn+1 (1kn+1 ⊗ bn+1 )k < 4−n 2 ∀ a ∈ Ωn ∪ {1}.

Let δn := (sup{kSn (a)k : a ∈ Ωn ∪ {1}})−1 and γn := δn /(5kn ). Consider

(n)
the separable C *-subalgebra B1 of B which is generated by the ej , n = 1, 2, . . .,
j = 1, . . . , kn . B1 contains a strictly positive element e ≥ 0 with kek = 1. We find
a sequence of positive integers pn < pn+1 < . . . such that, with en := e1/pn , we get
(n) (n)
ken ej − ej k < 16−n γn for j = 1, . . . , kn , n = 1, 2, . . .. Then from B = DB we
get contractions gn ∈ D with kgn en − en k < 16−n γn .
The separable C *-subalgebra D1 of D generated by gn contains a strictly pos-
itive element h ∈ D+ with khk = 1, and we find positive integers rn < rn+1 < . . .
such that (with hn := h1/rn ) khn gn − gn k < 16−n γn .
It follows
(n) (n)
khn ej − ej k < 16−n 5γn = 16−n δn /kn and kEn∗ − (1kn ⊗ hn )En∗ k < 16−n δn .
Since D is purely infinite and simple, bn , hn ∈ D+ and kbn k = khn k = 1, we
find (by Proposition 2.2.1(ii)) cn ∈ D with kcn k = 1 and kc∗n b2n cn − hn k < 16−n δn .
Then
kEn (α ⊗ c∗n b2n cn )En∗ − En αEn∗ k < 4−n kαkδn
for α ∈ Mkn , because En is a contraction.
Here Mkn is embedded in Mkn (M(D)) ∼ = Mkn ⊗ M(D) by α 7→ α ⊗ 1.
∗
Pkn (n) (n)
Now let zn := Fn (1kn ⊗ (bn cn ))En ∈ B, i.e. zn := j=1 fj bn cn ej . Then
kzn∗ azn − Tn Sn (a)k < 4−n 2 + kEn (Sn (a) ⊗ c∗n b2n cn )En∗ − En Sn (a)En∗ k.

kzn∗ azn+1 k ≤ k(1kn ⊗ bn )Fn∗ aFn+1 (1kn+1 ⊗ bn+1 )k,

because k(1 ⊗ cn )En∗ k = kcn k · kEn k ≤ 1. It follows that
kzn∗ azn+1 k < 4−n 2 and kzn∗ azn − Tn Sn (a)k < 4−n 3 ∀ a ∈ Ωn ∪ {1}.
In particular kzn k2 ≤ 1 + 4−n 3.
Let dn := (max(1, kzn k))−1 zn , then dn is as desired.
(i) (j) (i) (i)
(ii): We have limn→∞ (dn )∗ dn = 0 and limn→∞ (dn )∗ dn = 1 (i 6= j and
(i)
i, j = 1, 2) from part (i). Thus, eventually small perturbations of the dn ’s (i=1,2)
are as desired. 

Next remark sort of pre-version of the WvN-theorem in chp.5?

 2. ON APPROXIMATE INNER NUCLEAR MAPS 391

Remark 3.2.14. There is a more delicate counterpart of Proposition 3.2.13 for

the case D ∼
= K (which is implicitly contained in [?]):
Suppose that B is a C *-algebra, and that C, D ⊆ M(B) are C *-subalgebras of
the multiplier algebra M(B) of B, such that D is non-degenerate, i.e., D · B = B,
and D ∼ = K . Let C be a separable C *-subalgebra of M(B) such that CD ⊆ D, and
let V : C → B a completely positive map such that V (C ∩ D) = {0} and suppose
that the c.p. map [V ] : C/(D ∩ C) → B on the “classes”
[c] := c mod (D ∩ C) = c + D ∩ C
is a nuclear completely positive map.
Then V is approximately 1-step inner. More precisely, there exists a sequences
of contractions dn ∈ B with limn→∞ d∗n adn+k = δ0,k V (a) for a ∈ C.
The proof uses Lemma 2.2.3 and modifications of arguments from the proof of
Proposition 3.2.13.

In the following proposition Cb (X, B) means the C *-algebra of continuous func-

tions on a locally compact space X with values in B, and C0 (X, B) is the closed
ideal in Cb (X, B) formed by the functions f ∈ Cb (X, B) with x ∈ X 7→ kf (x)k
vanishing at infinity.

Proposition 3.2.15. Let D ⊆ B be C*-algebras, where D is simple and purely

infinite and DB = B . Let X be a locally compact and σ-compact space, C a sep-
arable C*-subalgebra of Cb (X, D), and x ∈ X 7→ (Vx : D → B) ∈ CP(D, B) a
point-norm continuous map from X into the nuclear completely positive contrac-
tions from D to B, i.e., Vx ∈ CPnuc (D, B) for all x ∈ X.
Consider the c.p. contraction V : Cb (X, D) → Cb (X, B) defined by V (d)(x) :=
Vx (d(x)) for x ∈ X and d ∈ Cb (X, D) .

(i) There exists a contraction d ∈ Cb (X, B) such that d∗ cd−V (c) ∈ C0 (X, B)
for c ∈ C .
(ii) If moreover B and D are unital, then 1D = 1B . If then 1D ∈ C, i.e., if
C contains the unit element of the C*-algebra Cb (X, D), and nuclear c.p.
map Vx : D → B is unital for every x ∈ X, then the element d in Part(i)
can be chosen such that d is moreover an isometry in Cb (X, B).

Notice that the Cb (X)-modular c.p. map c ∈ Cb (X, D) → V (c) ∈ Cb (X, B)

is in general not nuclear, but it is easy to see with help of the “tensor product
criterium” that the c.p. maps c 7→ f ∗ V (c)f ∈ C0 (X, B) with f ∈ C0 (X, B) are all
nuclear.

Proof. We reduce the proof essentially to the proof of Part(i) by adding to C

a special constant positive contraction e ∈ D+ ⊆ Cb (X, D)+ that makes the larger
C ∗ (C, e) ⊇ C a separable σ-unital C *-subalgebra of Cb (X, D), and then select the
proposed contraction d ∈ Cb (X, B) such that it satisfies in addition the inequality
kd∗ ed − V (e)k < 1/3 .
392 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

(i): By local compactness and σ-compactness of X, there exist an increasing

sequence X1 ⊆ X2 . . . ⊆ X of open subsets of X with compact closures Xn ⊆ Xn+1
S
such that X = Xn .
If we use the Tietze-Urysohn extension theorem then we find continuous func-
tions fn : X → [0, 1] with fn (x) = 1 for x ∈ Xn and fn (x) = 0 for x ∈ X \ Xn+1 .
1/2
Let g1 := f1 , gn := (fn − fn−1 )1/2 for n > 1 . Then fn fn+1 = fn , and therefore
gn gm = 0 if |n − m| > 1 .
Since C is separable by our assumptions, we find separable C *-subalgebras
Dn ⊆ D such that the evaluation epimorphisms πx : Cb (Xn , D) → D for x ∈ Xn
map the separable C *-algebra C|Xn into the separable C *-subalgebra Dn , i.e.,
S
πx (C) ⊆ Dn . Thus, the separable C *-subalgebra D∞ := n Dn has the property
that C ⊆ Cb (X, D∞ ). Let e ∈ D∞ a strictly positive contraction in D∞ . Consider
it as constant element of Cb (X, D) and replace C by C ∗ (e, C) ⊆ Cb (X, D). Then
e ∈ C ∗ (e, C) is a strictly positive element of C ∗ (e, C). If D is unital then we can
likewise also take here 1D in place of e. From now on we suppose that C contains a
“distinguished” strictly positive contraction e. Likewise this is the unit of Cb (X, D)
if D is unital.
Now let Y1 ⊆ Y2 ⊆ . . . ⊆ C be finite-dimensional linear subspaces of C, such
that C is the closure of the union of the Yn . Since C ⊆ Cb (X, D), we find compact
subsets Ω1 ⊆ Ω2 ⊆ . . . ⊆ D such that y(x) ∈ Ωn for points x ∈ Xn+1 and elements
y ∈ Yn with kyk ≤ n + 1. Consider the nuclear completely positive contractions
Vn : D → C0 (X, B) given by Vn (b)(x) := (fn+1 V (b)fn+1 )(x) for x ∈ X and b ∈ D,
where the fn ∈ C0 (X)+ are as above defined.
Recall that D ⊆ B ⊆ Cb (X, B) and Cb (X, B) ⊆ M(C0 (X, B)) naturally.
C0 (X, B) is the closure of D C0 (X, B), because is the closure of B C0 (X, B)
and B is the closure of DB by the assumptions.
We have shown – under this assumptions of non-degeneracy – in the proof of
Proposition 3.2.13 that there exist contractions d1 , d2 , · · · ∈ C0 (X, B) with kd∗n bdn −
Vn (b)k < 2−n−1 , kd∗n bdn+1 k < 2−n−1 for b ∈ Ωn and kd∗n dn+1 k < 2−n−1 .
P∞
Let h := n=1 gn+1 dn . Now use the above listed properties of gn , dn , and the
definitions of Ωn and Yn .
Then a simple calculation shows, that h ∈ Cb (X, B), kh(x)k ≤ 1 + 2−n for
x ∈ X \ Xn , kh(x)∗ b(x)h(x) − Vx (b(x))k ≤ 2−n kbk for b ∈ Yn and x ∈ X \ Xn and
n > 1, or n = 1 and y ∈ X.
Need that kd(x)∗ ed(x) − ek < 1/4 ... ????
Thus d(x) := (max(1, kh(x)k))−1 h(x) is a contraction d in Cb (X, B) with the
desired properties.
(ii): If B is unital, then DB = B implies 1B ∈ D.
If then 1B ∈ C and the Vx (·) are unital, then the element we can assume that
 2. ON APPROXIMATE INNER NUCLEAR MAPS 393

1 ∈ Y1 and have d(x)∗ d(x) ≤ 1 and 1 − d∗ d ∈ C0 (X, B). Thus we can re-
place d(x) := h(x)(h(x)∗ h(x))−1/2 is an isometry in Cb (X, B) with the desired
properties. 

Corollary 3.2.16. Suppose that B is purely infinite and simple.

(i) Let C be a C*-subalgebra of M(B). Then every nuclear completely positive

contraction V from C to B can be approximated point-wise by maps of the
form c ∈ C 7→ d∗ cd ∈ B, where d ∈ B and kdk ≤ 1.
(ii) If C is a separable C*-subalgebra of `∞ (B) and if Vn : B → B (n =
1, 2, . . .) is a sequence of nuclear completely positive contractions, then
there exist sequences d1 , d2 , . . . and e1 , e2 , . . . of contractions in B such
that lim e∗n dn = 0 for n = 1, 2, . . ., lim kd∗n πn (c)dn − Vn (πn (c))k = 0 and
lim ke∗n πn (c)en − Vn (πn (c))k = 0.

Proof. (i): Let Y ⊆ C a finite subset and ε > 0. By Proposition 3.2.13 (with
B = D) there exists a contraction d ∈ B with kd∗ ad − Vn (a)k < ε for a ∈ Z.
(ii): Let Y1 ⊆ Y2 ⊆ . . . ⊆ C be a linear filtration of C. By Proposition
3.2.13 (with B = D) there exist contractions dn , en ∈ B with ke∗n dn k ≤ 1/n,
kd∗n adn − Vn (a)k < 1/n and ke∗n aen − Vn (a)k < 1/n for every a in the compact set
{πn (c) : c ∈ Yn , kck ≤ n}. 

Remark 3.2.17. Corollary 3.2.16(ii) has a converse, which is easily established

(using Proposition 2.2.1(ii) and the nuclear contractions a 7→ ψ(a)b for suitable
states ψ):
If every nuclear contraction from an arbitrary subalgebra B of A 6= {0} into A can
be approximated by maps b 7→ d∗ bd, d ∈ A and kdk ≤ 1, then either A is simple
and purely infinite or A ∼
= C.

Remark 3.2.18. As noted in Remark 3.2.17, the assumption that A is purely

infinite is quite essential to approximate nuclear maps on subalgebras by maps
b 7→ d∗ bd .
However, Haagerup and Zsido [351] published a proof of the Dixmier conjecture
that every simple unital C *-algebra A with a unique trace state ρ has the so-called
“Dixmier property”: There exists a net {Vτ } in the convex combination of the inner
automorphisms on A that converges point-wise to the map ρ(·)1.

Remark 3.2.19. For contractions a and b in a C *-algebra B and ε ∈ [0, 1)

with ka∗ bk < ε2 there exist contractions c, d ∈ B such that c∗ d = 0, ka−ck < ε1/2
and kb − dk < ε1/2 .
In the case of positive contractions a, b ∈ C0 (X, M2 )+ with kabk < ε2 one can
find contractions c, d ∈ C0 (X, M2 )+ with c∗ d = 0 and kc−ak < 2ε and kd−bk < 2ε.
This is also the general conjecture for contractions a, b ∈ C ∗ (a, b) ⊆ L(`2 ) .
It is easy to see that in case of any functions a, b ∈ C0 (X) with kabk ≤ ε there
exists functions c, d ∈ C0 (X) with cd = 0, kc − ak ≤ ε and kd − bk ≤ ε.
394 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

In case of projections a = p and b = q we can take c := p and d := (1 − p)q

with ka − ck = 0 and kb − dk = kabk.
The estimates allow us to show that the proofs of Proposition 3.2.13(i) and
(i) (j)
Corollary 3.2.16(i) can be arranged such that, moreover, (dn )∗ dn = 0 for i 6= j
(respectively that e∗n dn = 0).

More details for Remark 3.2.19 : We can replace the contractions a and
b by the positive contractions e := (aa∗ )1/2 and f := (bb∗ )1/2 . Then a = ev =
v(a∗ a)1/2 and b = f w = w(b∗ b)1/2 for the polar decompositions. They satisfy
kef k = ka∗ bk. Let γ := kef k.
An estimate of the infimum of the distances max(ke−ge1/2 k, kf −hf 1/2 k) to all
contractions g, h ∈ A with g ∗ h = 0 then we get also the desired estimate, because
c := ge1/2 v and d := hf 1/2 w satisfy kc−ak = ke−ge1/2 k and kd−bk = kf −hf 1/2 k.
Thus it suffices to consider the special case where ke − ge1/2 k and kf − hf 1/2 k
1/2 1/2 1/2
with g := (e − f )+ and h := (f − e)+ = (e − f )− . Then (f − e)+ = (e − f )−
implies g ∗ h = 0.
Recall 2(f −e)+ = |f −e| +(f −e) and 2(e−f )+ = |f −e| −(f −e) = 2(f −e)− .
It implies e − (e − f )+ = 2−1 (e + f − |e − f |) = f − (f − e)+ .
Since kek ≤ 1 and the square root is operator convex, we get
ke−ge1/2 k ≤ ke1/2 −gk ≤ ke−g 2 k1/2 = ke−(e−f )+ k1/2 = 2−1/2 ·ke+f −|e−f |k1/2 .
Now we use again that square root is operator convex and and that |e−f |2 = (e−f )2
to obtain ke + f − |e − f |k ≤ k(e + f )2 − (e − f )2 k1/2 . Using (e + f )2 − (e − f )2 =
2(ef + f e) and kef k = kf ek we obtain ke + f − |e − f |k ≤ 21/2 kef k1/2 and finally
that
ke − ge1/2 k ≤ kef k1/4 .
Similar calculation shows that
kf − hf 1/2 k ≤ kef k1/4 .

Corollary 3.2.20. Suppose that B is a simple C*-algebra and A is a C*-

subalgebra of M(B).
Then for every nuclear contraction V : A → B, every compact subset Ω of A
and every ε > 0 there are b1 , . . . , bn in B such that k b∗i bi k ≤ 1 and
P
X
V (a) − b∗i abi < ε for all a ∈ Ω .

It is an idea from ‘‘3rd draft’’.

It is true for all ideal-system preserving residually nuclear maps
V : A → B from A ⊆ M(B) to B, i.e., if
V (A ∩ M(B, J)) ⊆ J and [V ]J : A/(A ∩ M(B, J)) → B/J is nuclear for all
closed ideals J ⊆ B.
Thus, residually nuclear maps are approximately sums of inner c.p.
maps ...
 2. ON APPROXIMATE INNER NUCLEAR MAPS 395

It comes from the generalized separation theorem for m.o.c. cones.

Derive it as a Corollary from the general result for those

C *-algebras where every ideal-respecting nuclear map is automatic
residual nuclear??
E.g. where all ideals are ‘‘locally semi-split’’.
???Are all exact C *-algebras locally semi-split ???

Proof. Consider the nuclear map V ⊗ id from A ⊗ O∞ ⊆ M(B ⊗ O∞ ) into

B ⊗ O∞ (that is again nuclear because O∞ is a nuclear C *-algebra; i.e., V ⊗ id
is nuclear). By Theorem E, the C *-algebra B ⊗ O∞ is simple and purely infinite,
because B is simple and O∞ is simple and purely infinite, cf. [169, 172]. By
Corollary 3.2.16, V ⊗id can be approximated by maps of the form a⊗ξ 7→ x∗ (a⊗ξ)x
for a suitable contraction x ∈ B ⊗ O∞ . In particular, for a given ε > 0, we can
find x ∈ B ⊗ O∞ , kxk ≤ 1 such that kV (b) ⊗ 1 − x∗ (b ⊗ 1)xk < ε/2 for b ∈ Ω .
Approximating x by elementary tensors, we can achieve
X X
kV (b) ⊗ 1 − ( ai ⊗ di )∗ (b ⊗ 1)( ai ⊗ di )k < ε

and
X
k ai ⊗ di k ≤ 1
for suitable elements ai ∈ B and di ∈ O∞ . Now let ρ be a pure state on O∞ .
Applying the Gram-Schmidt orthogonalization procedure we may suppose that
ρ(d∗i dj ) = δij for all i, j, and obtain suitable new
P
ai ⊗ di . But then (apply-
ing id ⊗ρ) we get
X
kV (b) − a∗i bai k < ε
for b ∈ Ω. Now k a∗i ai k ≤ 1 because k ai ⊗ di k ≤ 1.
P P


Next follows from s.p.i. permanences:

A s.p.i. and B exact
(e.g. B nuclear, B = C0 (X) for l.c. space X)
then A ⊗ B s.p.i.
Result and Proof should be given somewhere in Chp. 2.

Corollary 3.2.21. If D is simple and purely infinite, Y a compact space,

ε > 0 and a, b ∈ C(Y, D)+ such that kbk ≤ 1 and ka(y)k = 1 for every y ∈ Y then
there exists a contraction d ∈ C(Y, D) such that kb − d∗ adk < ε.

Proof. For every y ∈ Y we find a state λy on D such that λy (a(y)) = 1. In

an open neighborhood U (y) of y we have |1 − λy (a(x))| < ε/2 for x ∈ U (y). Since
Y is compact, we find a finite sequence y1 , . . . , yn ∈ Y such that the union of the
U (yk ) equals Y . We may assume that the system is minimal with this property.
Let ek : Y → [0, 1] (k = 1, . . . , n) a finite decomposition of 1 such that the support
of ek is contained in U (yk ).
P
Let λk := λyk and Ty (c) := ( ek (y)λk (c))b(y).
396 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

Then kTy (a(y)) − b(y)k ≤ ε/2 for every y ∈ Y and y 7→ Ty is a strongly

continuous map from Y into the nuclear completely positive contractions from D
into D. Since {a(y) : y ∈ Y } is norm compact, there exists a contraction d ∈
C(Y, D) with kTy (a(x)) − d(y)∗ a(x)d(y)k < ε/2 for all x, y ∈ Y , by Proposition
3.2.15. To obtain this estimate from Proposition 3.2.15, let B := D, X := Y × N,
i.e. X is the countable disjoint union of copies of Y , define V by V (a)(y, n) := Ty (a),
and let C be the separable C *-subalgebra of D ⊆ Cb (X, B) = `∞ (C(Y, D)) which
is generated by {a(x) : x ∈ Y }.
Thus kb − d∗ adk < ε . 

Lemma 3.2.22. Let A := C(X1 , D1 ) ⊕ . . . ⊕ C(Xn , Dn ) , where the Xk (k =

1, . . . , n) are compact metric spaces and Dk (k = 1, . . . , n) are simple purely infinite
C*-algebras and ε > 0.
If a, b ∈ A, b∗ = b, 0 ≤ a, kak ≤ 1 and there are contractions c1 , . . . , cm ∈ A
such that kb − c∗j acj k < ε and k c∗j cj k ≤ 1, then there is a contraction d ∈ A
P P

with kb − d∗ adk < 3ε .

Better use here:
Direct sums of s.p.i. algebras Ak are s.p.i.
and C(X, D) = C(X) ⊗ D is s.p.i. if D is s.p.i.
⇐ follows from s.p.i. permanences in Chapter 2.
Find -- give -- references!
In other words: A is a non-simple strongly purely infinite C*-algebra in the
sense of Definition 1.2.2.

Proof. We have b = b1 ⊕ . . . ⊕ bn , a = a1 ⊕ . . . ⊕ an . Thus, it suffices to

consider the case A = C(X, D) for simple purely infinite D and compact X. But
then
X
kb − c∗j acj k < ε
implies kb − b+ k < ε and
kb(y)k < ε + ka(y)k
c∗j acj
P
for y ∈ X, because a → is a completely positive contraction.
Let Y := {y ∈ X : 2ε ≤ kb(y)k } . By functional calculus, for y ∈ Y , we have
kb(y)k = kb(y)+ k, ε ≤ k(b(y) − ε)+ k = kb(y)k − ε and k(b(y) − ε)− k ≤ 2ε, i.e.,
kb(y) − (b(y) − ε)+ k ≤ 2ε.
On the other hand ε < ka(y)k and 0 < ka(y)k−1 (kb(y)k − ε) < 1 for y ∈ Y .
Let c(y) := (kb(y)k − ε)−1 (b(y) − ε)+
By Corollary 3.2.21, there exists a contraction d1 ∈ C(Y, D) such that
kc(y) − d1 (y)∗ (ka(y)k−1 a(y))d1 (y)k < ε ∀y∈Y .

g(y) := ((kb(y)k − ε)/ka(y)k)1/2 has norm ≤ 1 on Y . The contraction d2 (y) =

g(y)d1 (y) in C(Y, D) satisfies k((b|Y )−ε)+ −d∗2 (a|Y )d2 k < ε and therefore k(b|Y )−
d∗2 (a|Y )d2 k < 3ε.
 3. FROM M.O.C. CONES TO MORPHISMS IN GENERAL POSITION 397

Now let d3 ∈ C(X, D) a contraction with d3 (y) = d2 (y) for y ∈ Y . Then there
is an open neighborhood U of Y with kb(x) − d3 (x)∗ a(x)d3 (x)k < 3ε for x ∈ U . We
find h : X → [0, 1] continuous with h(y) = 1 for y ∈ Y and h(x) = 0 for x ∈ X \ U .
The map d(x) := h(x)1/2 d2 (x) defines a contraction d ∈ C(X, D) such that

kb(x) − d(x)∗ a(x)d(x)k ≤ (1 − h(x))kb(x)k + h(x)kb(x) − d3 (x)∗ a(x)d3 (x)k < 3ε .

B in next Corollary is s.p.i. because the An are s.p.i.

and inductive limits of s.p.i. C *-algebras are s.p.i.

Corollary 3.2.23. Let hn : An → An+1 a sequence of C*-algebra morphisms.

Suppose that

(i) B := indlim(hn : An → An+1 ) is non-zero, and

(ii) each An is a finite direct sum of algebras C(Xj , Dj ) of continuous func-
tions from compact spaces Xj into a simple purely infinite C*-algebra Dj
(j = 1, . . . , kn ).

Then B is strongly purely infinite.

Proof. By Lemma 3.2.22, the algebras An are strongly purely infinite in the
sense of Definition 1.2.2. Therefore, B is strongly purely infinite by Proposition
??(??) (See Proposition 2.15.5(iv) for the p.i. case.) 

3. From m.o.c. cones to morphisms in general position

We need a good metric on m.o.c. cones,

that gives the point-norm convergence topology
in case of separable A and countably generated C ⊆ CP(A, B).
And have to describe the natural transformations and extensions with
respect to Morita equivalence.
Need
KK(C; A, B) ∼
= KK(C ⊗ CP(K, K); A ⊗ K, B ⊗ K) .
Need
Ext(C; A, B) ∼
= Ext(C ⊗ CP(K, K); A ⊗ K, B ⊗ K) .

Compare with related sections on m.o.c.c’s versus bi-modules !!

Definition 3.3.1. Let B a stable σ-unital C *-algebra and A a σ-unital C *-

algebra. A C *-morphism H : A → M(B) is in general position if H is unitarily
homotopic to its infinite repeat δ∞ ◦ H
398 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

By Definition 5.0.1 and Remark 5.1.1(8) this means that there exists a norm-
continuous path of unitaries t ∈ [0, ∞) → U (t) ∈ M(B) such that U (t)∗ H(a)U (t)−
δ∞ (H(a)) ∈ B and, for all a ∈ A ,

lim kU (t)∗ H(a)U (t) − δ∞ (H(a))k = 0 .

t→∞

It is not clear if the condition U (t)∗ H(a)U (t) − δ∞ (H(a)) ∈ B (for all a ∈ A) in
Definition 3.3.1 implies that for the norm-continuous map t ∈ [0, ∞) → X(t) given
by X(t) := U (t)∗ H(a)U (t) − δ∞ (H(a)) ∈ B holds that X(t) converges strictly to
0 if and only if t → kX(t)k converges to 0 for t → ∞. If it is true, then it does
not matter if we require that limt→∞ X(t) = 0 in the strict topology on M(B)
or in norm topology on M(B). Here “strictly” says that limt kX(t)bk = 0 for
each b ∈ B, and “in norm-topology” means that limt kX(t)k2 = 0 for the bounded
norm-continuous path t 7→ X(t) ∈ B.
But it is not clear if one can here replace the strictly continuous path t ∈ R+ →
U (t) ∈ U(M(B)) by a norm-continuous path.
It is known ???? by definition ???? that each C *-morphism H : A → M(B)
“ in sufficiently general position”, i.e., where H is unitary equivalent modulo B to
δ∞ ◦ H, is unitarily homotopic to its infinite repeat – in the sense of Definition
5.0.1, cf. Remark 5.1.1(8).

4. M.o.c. cones and Hilbert bi-modules

Remark 3.4.1. For every point-norm closed m.o.c. cone C ⊆ CP(A, B) there
is a (right) Hilbert B-module H and a *-representation d : A → L(H) such that for
every x ∈ H the map a ∈ A 7→ hd(a)x, xi ∈ B is in C and for every V ∈ C there
exists x ∈ H such that V (a) = hd(a)x, xi for a ∈ A.
But even for separable A and B the needed “multiplicity” of H over B can be
much bigger than the cardinality of a subset of C that is dense in C with respect to
the matricial point-norm topology on CP(A, B). This can be seen even in the – for
us not important – easy case where A := C([0, 1]), B := C and C := CP(A, B) ∼ =
C([0, 1])∗+ .
Suppose, in addition, that A is separable, B is σ-unital and stable and that
C ⊆ CP(A, B) ⊆ L(A, B) is countably generated, faithful and non-degenerate.
Then an other module is given by a C *-morphism HC : A → M(B) such that HC
is unitarily equivalent to its infinite repeat δ∞ ◦ HC , that b∗ HC (·)b ∈ C for each
b ∈ B, and that for each V ∈ C with kV k ≤ 1 there exist a sequence of contractions
b1 , b2 , . . . ∈ B such that V (a) := limn b∗n HC (a)bn for all a ∈ A.
The HC with this properties is uniquely determined by C up to unitary homo-
topy.
(Uses general W-vN-theorem: Sums of two of them will be asymptotic unitarily
absorbed by each of them ...?).
(See Corollary 5.4.4 and Remark 5.4.5.)
 4. M.O.C. CONES AND HILBERT BI-MODULES 399

Compare next with Parts (xvi) of Proposition 2.2.1 in Chp. 2 !

There it is the non-unital case.

Lemma 3.4.2. Suppose that 1E ∈ A ⊆ E are C*-algebras, where A is sepa-

rable and for every a ∈ A+ with kak = 1 there exists a sequence of contractions
T1 , T2 , . . . ∈ E with limn Tn∗ aTn = 1E . (In particular, each nonzero a ∈ A+ is
properly infinite inside E and generates E as closed ideal.)
Then every factorable c.p. contraction U ◦V with V : A → Mn and U : Mn → E
is approximately 1-step inner in E, i.e., there exists a sequence of contractions
S1 , S2 , . . . ∈ E such that limn Sn∗ aSn = U (V (a)) for all a ∈ A.

Proof. Should be part of Chapter 2 ?? 

More general alternative formulation for non-unital and not necessarily simple
E:

Lemma 3.4.3. Suppose that A ⊆ E are C*-algebras, where A is separable and

for every a ∈ A+ with kak = 1 and every contraction b ∈ E+ in the closed ideal
of E generated by a there exists a sequence of contractions T1 , T2 , . . . ∈ E with
limn Tn∗ aTn = b.
Then every c.p. contraction U ◦ V given by c.p. contractions V : A → Mn and
U : Mn → E such that U ◦ V is ideal-system preserving
Should work only for E with finitely many ideals???
We need case of simple A or simple E!
is approximately 1-step inner in E, i.e., there exists a sequence of contractions
S1 , S2 , . . . ∈ E such that limn Sn∗ aSn = U (V (a)) for all a ∈ A.

Proof. ?? 

Remarks 3.4.4. Let A a separable C *-algebra and B a σ-unital stable C *-

algebra. Take any faithful non-degenerate *-representation ϕ : A → M(K) ∼ =
L(`2 (N)) and any non-degenerate *-representation λ : K → M(B).
Then the H0 for CPnuc (A, B) is up to unitary homotopy given by H0 := δ∞ ◦
M(λ) ◦ M(ϕ).
If D is non-unital, then πB ◦ H0 dominates zero.
If a C *-morphism ψ : A → Q(B) := M(B)/B has the property that for every
a ∈ A+ and ε > 0 there exists a contraction Ta ∈ Q(B) with Ta∗ ψ(a)Ta = kak(1 −
ε)1, then ψ and ψ ⊕ πB ◦ H0 are unitarily equivalent by a unitary in U0 (Q(B)) if
B is non-unital. If A and ψ are both unital then ψ and ψ ⊕ πB ◦ H0 are equivalent
by unitary in U(Q(B)).
The existence of the Ta is equivalent to the property that for each a ∈ A+ ,
δ ∈ (0, kak) and c ∈ M(B) with πB (c∗ c) = ψ((a − δ)+ ) the hereditary *-subalgebra
cBc∗ of B contains a full stable C *-subalgebra of B.
(Compare also [264] and [310].)
400 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

Definition 3.4.5. Let C ⊆ CP(A, B) a matrix o.c. cone. A Hilbert A-B-

module (H, φ) is C-compatible if the c.p. maps

A 3 a 7→ hx, φ(a)xi ∈ B

are in C for all x ∈ H.

Definition 3.4.6. A point-norm closed matrix o.c. cone C ⊆ CP(A, B) is

countably generated if C = C(S) for some countable subset S ⊆ C. We say that
C is singly generated if C = C({V }) for some V ∈ C.

Remark 3.4.7. If a point-norm closed m.o.c.c. C ⊆ CP(A, B) is countably

generated, say by c.p. contractions {V1 , V2 , . . .} ⊆ C, then C is also singly generated,
e.g. by V := n 2−n Vn , because C is hereditary in CP(A, B) by Corollary ??.
P

If B is σ-unital (with strictly positive element b0 ∈ B+ ) and if A admits a

faithful positive functional f0 ∈ A∗+ , then CPnuc (A, B) is singly generated by V =
f0 (·)b0 .
If A and B are separable, then each point-norm closed m.o.c. cone C ⊆
CP(A, B) is singly generated.

Proposition 3.4.8. Suppose that a compact metric group G acts on A and

B (by α and β), where A is separable and B is σ-unital. Let C ⊆ CP(A, B) a
countably generated m.o.c. cone with β(g) ◦ C ◦ α(g −1 ) = C for all g ∈ G.
Then there is a “universal” C-compatible Hilbert A-B-module (H, φ) with a G-
action γ : G → Iso(H), such that γ(g)(φ(a)xb) = φ(α(g)(a))γ(x)β(g)(b) for a ∈ A,
b ∈ B, x ∈ H, and that G 3 g 7→ hy, γ(g)(x)i ∈ B is continuous for all x, y ∈ H.

Proof. Use G-invariant V ∈ C for construction. Then use that m.o.c. cones
are hereditary in CP(A, B).
to be filled in ?? 

Question 3.4.9. Remains Proposition 3.4.8 true if G is a second countable l.c.

group?

5. Operator-convex cones of c.p. maps (2)

This section explains the reason for our study of commutative subalgebras in ul-
trapowers of w.p.i. and s.p.i. C *-algebras, i.e., it shows that asymptotic and approx-
imate versions of generalized Weyl–von-Neumann theorems for “residually nuclear”
maps can be obtained from the corresponding results on c.p. maps from separable
C *-subalgebras of ultrapowers (or, more generally, corona algebras) into commu-
tative C *-subalgebras of the ultrapower or corona algebra in question. One has to
apply the below outlined theory of (matrix) operator-convex cones of c.p. maps (in
conjunction with Property (3) in Remark 3.11.3).
 5. OPERATOR-CONVEX CONES OF C.P. MAPS (2) 401

We introduce a useful duality between point-norm closed operator-convex cones

C of completely positive maps in CP(A, B) and intersection-preserving maps (“ac-
tions”) of the Hausdorff lattice I(B ⊗ C ∗ (F∞ )) of closed ideals of B ⊗ C ∗ (F∞ ) on
A ⊗ C ∗ (F∞ ).
This in an operator theoretic version of the Hahn-Banach separation theorem
for m.o.c. cones.
It implies a duality between m.o.c. cones C of “residually nuclear” maps and
actions Ψ of Prim(B) on A. Global operations with residually nuclear maps are
functorial in a natural manner.

Definition 3.5.1. Suppose that A and B are C *-algebras. Let CP(A, B)

denote the cone of completely positive maps from A into B. A subset C of CP(A, B)
is an operator-convex cone of c.p. maps if C has the following properties (i) and (ii):

(i) d∗1 V1 (.)d1 + d∗2 V2 (.)d2 ∈ C for V1 , V2 ∈ C, d1 , d2 ∈ B.

(ii) a ∈ A 7→ c∗ (V ⊗ idn )(r∗ ar)c is in C for every V ∈ C, every row-matrix
r ∈ M1,n (A) and every column-matrix c ∈ Mn,1 (B).

Let S be a subset of CP(A, B). We denote by Calg (S) the smallest subset of
CP(A, B) which is invariant under the operations in (i) and (ii), and by C(S) the
point-norm closure of Calg (S) (i.e., the closure of Calg (S) in L(A, B) w.r.t. the strong
operator topology). Then Calg (S) and C(S) are operator-convex cones of completely
positive maps.
We call S the generating set for the operator convex cone C if C = C(S).

One can see with help of approximate units in A and B that every point-norm
closed subset C of CP(A, B) is a convex cone in the usual sense if it satisfies Part(i)
of Definition 3.5.1, and that C satisfies tC ⊆ C for t ∈ [0, ∞) if C satisfies Definition
3.5.1(ii).
We introduce a useful duality between closed m.o.c. cones C of completely pos-
itive maps in CP(A, B) and intersection-preserving maps (“actions”) of the Haus-
dorff lattice I(B ⊗C ∗ (F∞ )) of closed ideals of B ⊗C ∗ (F∞ ) on A⊗C ∗ (F∞ ). It leads
to a duality of m.o.c. cones of “residually nuclear” maps and actions of Prim(B)
on A. Operations with residually nuclear maps are functorial in a natural manner.
First we characterize the c.p. maps V in C(S) by the values V ⊗max id(c) of
elements c ∈ A ⊗max C ∗ (F∞ ). Here C ∗ (F∞ ) denotes the full group C *-algebra of
the free group F∞ on infinitely many generators, and ⊗max denotes the maximal
C *-algebra tensor product. In the following theorem the ideal-kernel of a positive
functional λ on D := B ⊗max C ∗ (F∞ ) means the maximal ideal of D in the kernel
of λ, i.e., the set of elements d ∈ D with λ(edf ) = 0 for all e, f ∈ D. This is also
the kernel of the cyclic representation ρ : D → L(L2 (D, λ)) defined by λ on D.
Here is some relation between Part(v) of Theorem 3.5.2 to and the
later considered Separation Theorem.
402 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

Theorem 3.5.2. Suppose that S is a subset of CP(A, B) and that V : A → B

is completely positive. Then the following properties (i)–(v) of V are equivalent.

(i) V ∈ C(S) := the point-norm closed m.o.c. cone generated by S.

(ii) For every c ∈ A ⊗max C ∗ (F∞ ), the element V ⊗max id(c) is in the closed
ideal of B ⊗max C ∗ (F∞ ) that is generated by the c.p. maps

{ W ⊗max id((a∗ ⊗ 1)c(a ⊗ 1)) ; W ∈ S, a ∈ A } .

(iii) For every factorial representation ρ : B → L(H) of B, ρ ◦ V is in the

point-weak closure of C( { ρ ◦ W ; W ∈ S } ).
(iv) For every pure state λ of B ⊗max C ∗ (F∞ ), the kernel of λ ◦ (V ⊗max id)
contains the intersection of all ideal-kernels of λ ◦ (b∗ W b ⊗max id) for
W ∈ S, b ∈ B.
(v) For every factorial cyclic representation ρ : B → L(H) (with cyclic vector
η ∈ H) and every unital C*-morphism h : C ∗ (F∞ ) → ρ(B)0 , the kernel of
the positive linear functional λ ◦ (V ⊗max id) contains the intersection of
all ideal-kernels of λ ◦ ((b∗ W (·)b) ⊗max id) for W ∈ S, b ∈ B.
(Here λ is the positive functional on B ⊗max C ∗ (F∞ ) with λ(b ⊗ f ) =
hρ(b)h(f )η, ηi.)

If one wants to apply the result easier for some sort of cone-related KK-theory,
then one could replace C ∗ (F∞ ) ⊆ C ∗ (F2 ) = C(S 1 ) ∗ C(S 1 ) by the unital free
product C *-algebra C([0, 1]) ∗ C([0, 1]) ⊃ C ∗ (F2 ) . The proofs are almost verbatim
the same as the below given, but with C ∗ (F∞ ) replaced by C([0, 1]) ∗ C([0, 1]).
The point is that this shows that no extra homotopy invariant is involved by this
reduction to the study of ideals of tensor products.
Alternatively we can formulate the for applications important equivalence be-
tween Parts (i) and (ii) as follows:
Let T ⊆ CP A ⊗max C ∗ (F2 ), B ⊗max C ∗ (F2 ) denote the set of c.p. maps

T := { W ⊗max id ; W ∈ S } .

And define a lower semi-continuous action

Ψ := Ψ[T ] : I B ⊗max C ∗ (F2 ) → I A ⊗max C ∗ (F2 )

by

Ψ(J) := biggest I ∈ I(A ⊗max C ∗ (F2 )) with W ⊗max id(I) ⊆ J ∀ W ∈ S .

I.e., Ψ is the lower semi-continuous action of Prim B⊗max C ∗ (F2 ) on A⊗max C ∗ (F2 )

that attaches to a closed ideal J / B ⊗max C ∗ (F2 ) the largest ideal I / A ⊗max C ∗ (F2 )
with the property that (W ⊗ id)(I) ⊆ J for all W ∈ S.
This is just the natural l.s.c. action ΨC(T ) defined by the point-norm closed
m.o.c. cone C(T ) that is generated by T := S ⊗max id.
The property of V in part (ii) is equivalent to:

V ⊗max id(ΨC(T ) (J)) ⊆ J for all J ∈ I(B ⊗max C ∗ (F2 )) .

 5. OPERATOR-CONVEX CONES OF C.P. MAPS (2) 403

Thus the equivalence of (i) and (ii) can be expressed as

V ∈ C(S) ⇐⇒ V ⊗max id(ΨC(S⊗max id) (J)) ⊆ J ∀ J ∈ I(B ⊗max C ∗ (F2 )) .

If we let Ψ := ΨC(S⊗max id) , where “id” means the identity automorphism

of C ∗ (F∞ ), and if we define, more generally, for C *-algebras E, F and actions
Ψ : I(F ) → I(E) the point-norm closed m.o.c. cone CΨ ⊆ CP(E, F ) by

CΨ := {V ∈ CP(E, F ) ; V (Ψ(J)) ⊆ J ∀ J ∈ I(F )} ,

then we can express the equivalence of (i) and (ii) by

C(S) ⊗max id = (CP(A, B) ⊗max id) ∩ CΨ . (5.1)

Proof. (i)⇒(ii): Let c ∈ B ⊗max C ∗ (F∞ ) positive and let I(S, c) denote the
closed ideal of B ⊗max C ∗ (F∞ ) which is generated by

I(S, c) := { W ⊗max id((a∗ ⊗ 1)c(a ⊗ 1)) ; W ∈ S, a ∈ A } .

If T : A → B is a completely positive map such that T ⊗max id((a∗ ⊗ 1)c(a ⊗ 1))

is in I(S, c) for all a ∈ A, then T ⊗max id((b∗ ⊗ 1)c(a ⊗ 1)) is in I(S, c) for all
a, b ∈ A, as polar-decomposition of Hermitian forms shows. It follows that the set
of all c.p. maps T : A → B with T ⊗max id((a∗ ⊗ 1)c(a ⊗ 1)) ∈ I(S, c) for all a ∈ A
contains S and is closed under the operations (OC1) and (OC2) of Definition 3.2.2,
and is point-norm closed. Thus it contains C(S).
(ii)⇒(i): Suppose that V is not in C(S) . Then there are

(α) a cyclic representation ρ : B → L(H) with cyclic vector η ∈ H ,

(β) elements e1 , . . . , en ∈ ρ(B)0 ,
(γ) elements a1 , . . . , an ∈ A , and
(δ) a *-epimorphism h from C ∗ (F∞ ) onto C ∗ (1, e1 , . . . , en ) ∈ ρ(B)0 and ele-
ments f1 , . . . , fn ∈ C ∗ (F∞ ) with h(fk ) = ek ,

such that the complex number λ((V ⊗max id)(c)) is not in the closure of the set of
complex numbers { λ((W ⊗max id)(c)) ; W ∈ C(S) }, where

c := a1 ⊗ f1 + . . . + an ⊗ fn ∈ A ⊗max C ∗ (F∞ )

and λ is the positive functional on B ⊗max C ∗ (F∞ ) with λ(b ⊗ f ) := hρ(b)h(f )η, ηi .
(See proof of [463, lem. 7.18], and notice that the Part (δ) is obvious.)
Thus, the positive functional λV := λ ◦ (V ⊗max id) is not in the weak closure
of the convex cone κ of positive functionals λW := λ ◦ (W ⊗max id) for W ∈ C(S).
Since η is cyclic for ρ, one can see (as in the proof of [463, lem. 7.18]) that
the functionals d∗ (λW )d is in the weak closure of κ for W ∈ C(S) and d ∈ A ⊗max
C ∗ (F∞ ).
Let J denote the closed ideal of d ∈ A ⊗max C ∗ (F∞ ) with λW (d∗ d) = 0 for all
W ∈ C(S).
404 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

It follows from [463, lem. 7.17(ii)] that there is d0 ∈ J with λV (d∗0 d0 ) > 0. On
the other hand, I(S, d∗0 d0 ) is contained in the kernel of λ by definition of J. Hence
V ⊗max id(d∗0 d0 ) is not in I(S, d∗0 d0 ) .
(i)⇒(v): Let J(S) ⊆ A⊗max C ∗ (F∞ ) denote the intersection of all ideal-kernels
of all functionals λ ◦ ((b∗ W b) ⊗max id) on A ⊗max C ∗ (F∞ ) with W ∈ S, b ∈ B. We
show that J(S) is the intersection of all ideal-kernels of all functionals λ◦(V ⊗max id)
with V ∈ C(S) :
Let T ∈ CP(A, B) with J(S) in the kernel of λ ◦ ((b∗ T b) ⊗max id) for all b ∈ B,
then, using the polar-formula for Hermitian forms,

λ((b∗2 ⊗ 1)T ⊗max id((a∗2 ⊗ 1)c(a1 ⊗ 1))(b1 ⊗ 1)) = 0

for c ∈ J, a1 , a2 ∈ A and b1 , b2 ∈ B. It follows that the set of c.p. maps V : A → B

with J(S) in the kernel of λ ◦ ((b∗ T b) ⊗max id) for all b ∈ B is closed under the
operations (OC1) and (OC2) of Definition 3.2.2 and under pont-norm convergence,
i.e., is a closed operator-convex cone, which contains the set S. Note that b∗ V (.)b
is in C(S) for V ∈ C(S) and b ∈ B.
(v)⇒(iii): Suppose that ρ ◦ V : A → L(H) is not in the point-weak closure of
the operator-convex cone C({ρ ◦ W ; W ∈ S}).
Then there is n ∈ N and η ∈ Hn , such that ρ1 ◦ V : A → L(H1 ) is not in the
point-weak closure of C({ρ1 ◦ W ; W ∈ S}) for the (again factorial) cyclic sub-
representation ρ1 of

ρ ⊗ 1n : b ∈ B 7→ ρ(b) ⊗ 1n ∈ L(Hn ) ∼
= L(H) ⊗ Mn

obtained by restriction of ρ ⊗ 1n to H1 := (ρ(B) ⊗ 1n )η. The rest of the proof is

similar to the proof of the implication (ii)⇒(i).
(iii)⇒(iv): Let ι : B ⊗max C ∗ (F∞ ) → L(H) the irreducible representation with
cyclic state η corresponding to λ. Let ρ(b) := ι(b ⊗ 1). Then ρ : B → L(H) is
factorial, and there is a representation h : C ∗ (F∞ ) → ρ(B)0 such that ι(b ⊗ f ) =
ρ(b)h(f ).
The elements c in the intersection J of the ideal-kernels of λ ◦ (b∗ W b) ⊗max id

with W ∈ S, b ∈ B satisfy ι W ⊗max id(dce) = 0 for all d, e ∈ A ⊗max C ∗ (F∞ ) and

W ∈ S.
If c ∈ A ⊗max C ∗ (F∞ ) satisfies ι W ⊗max id(dce) = 0 for all d, e ∈ A ⊗max

C ∗ (F∞ ) and W ∈ S, then ι T ⊗max id(c) = 0 for all T ∈ C(S) . Since ρ◦V is in the

point-weak closure of the maps { ρ◦W ; W ∈ C(S) }, we get that ι◦(T ⊗max id) is in


the point weak closure of the cone of maps ι◦(W ⊗max id). Thus λ W ⊗max id(c) =
0.
(iv)⇒(ii): Suppose that there is c ∈ A ⊗max C ∗ (F∞ ), such that V ⊗max id(c) is
not in I(S, c). Then there is a pure state λ on B ⊗max C ∗ (F∞ ) with λ(I(S, c)) = 0
and λ(V ⊗max id(c)) 6= 0.
(b∗ ⊗ 1)W ⊗max id(gch)(b ⊗ 1) ∈ I(S, c) for W ∈ S, g, h ∈ A ⊗max C ∗ (F∞ )
and b ∈ B, because W ⊗max id (a∗ ⊗ 1)c(a ⊗ 1) is in the ideal I(S, c) for a ∈ A.


 5. OPERATOR-CONVEX CONES OF C.P. MAPS (2) 405

Thus, the closed ideal J of A ⊗max C ∗ (F∞ ) generated by c is contained in all

kernels of λ ◦ ((b∗ W b) ⊗max id) for W ∈ S, b ∈ B. But c is not in the kernel of
λ ◦ (V ⊗max id). 

We derive in good cases 1-step approximately inner approximation for the mem-
bers of C(S) if the elements of S are 1-step-approximately inner.

Proposition 3.5.3. Suppose that A and B are C*-algebras and that S is a set
of c.p. maps V : A → B.

(i) Suppose that, for V1 , V2 ∈ S, the multiplier algebra M(B) contains

contractions e1 , e2 with e∗1 e2 = 0 such that e1 V1 (.)e∗1 + e2 V2 (.)e∗2 ∈ S,
e∗k ek Vk (a) = Vk (a) for all a ∈ A and k = 1, 2.
(a local property of J.Cuntz)??
Then C(S) is the point-norm closure of the set of all maps

V : a ∈ A 7→ c∗ (W ⊗ idn )(r∗ ar) c ∈ B

with W ∈ S, rows r ∈ M1,n (A) and columns c ∈ Mn,1 (B), n ∈ N.

(ii) If, in addition to part (i), A ⊆ M(B) and there is (fixed) m ∈ N such that
every W ∈ S is approximately m-step-inner, then every c.p. contraction
V ∈ C(S) can be approximated in point-norm by maps

T : a ∈ A 7→ d∗1 ad1 + d∗2 ad2 + . . . + d∗m adm ∈ B

with d1 , . . . , dm ∈ B and kd∗1 d1 + . . . + d∗m dm k ≤ 1.

(iii) If, in addition to (ii), A is separable and for every W ∈ S there exists a
bounded sequence of elements d1 , d2 , · · · ∈ B such that limk→∞ d∗k adk+n =
δ0,n W (a) for n = 0, 1, . . ., a ∈ A, then, for every V ∈ C(S), there exists a
sequence e1 , e2 , . . . ∈ B with ke∗k ek k ≤ kV k, limk→∞ e∗k aek+n = δ0,n V (a)
for n = 0, 1, . . ., a ∈ A, and limk→∞ ke∗k ek+n k = 0 for n > 0.

Proof. (i): For V1 , V2 ∈ S, r1 ∈ M1,n (A), r2 ∈ M1,m (A), c1 ∈ Mn,1 (B),

c2 ∈ Mm,1 (B) and d1 , d2 ∈ B, let V (a) := e1 V1 (a)e∗1 + e2 V2 (a)e∗2 (a ∈ A), r :=
(r1 , r2 ) ∈ M1,n+m (A), and d := (d∗1 c∗1 (e1 ⊗ 1n ), d∗2 c∗2 (e2 ⊗ 1m ))∗ ∈ Mn+m,1 (A),
where e1 , e2 ∈ M(B) are the elements as descibed in (i).
Then d∗1 (c∗1 V1 ⊗ idn (r1∗ ar1 )c1 )d1 + d∗2 (c∗2 V2 ⊗ idm (r2∗ ar2 )c2 )d2 = c∗ V ⊗
idn+m (r∗ ar)c. Thus the minimal matrix operator-convex hull Calg (S) of S is
identical with the set of maps c∗ V ⊗ idn (r∗ (.)r)c with V ∈ S, r ∈ M1,n (A),
c ∈ Mn,1 (A), n = 1, 2, . . .. C(S) is then the point-norm closure of Calg (S).
(ii): By [463, lem. 7.2], a suitable modification of the maps T (with help
of an approximate unit of A) leads to a net of m-step inner c.p. maps T with
kd∗1 d1 + . . . + d∗m dm k ≤ 1 (in addition) such that the net converges in point-norm to
V , if V ∈ CB(A, B) can be approximated in point-norm by m-step inner c.p. maps
T.
406 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

If V ∈ S can be approximated in point-norm by a net of m-step inner c.p. maps

T = d∗1 (.)d1 + . . . + d∗m (.)dm , then c∗ V ⊗ idn (r∗ (.)r)c can be approximated in point-
norm by the net with elements T 0 = c∗ T ⊗ idn (r∗ (.)r)c = e∗1 (.)e1 + . . . + e∗m (.)em
where ej = r(dj ⊗ 1n )c for j = 1, . . . , m.
(iii): If V = c∗ W ⊗ idn (r∗ ar)c and d1 , d2 , . . . satisfies limk→∞ d∗k adk+n =
δ0,n W (a) for n = 0, 1, . . ., then fk := r(dk ⊗ 1n )c satisfies limk→∞ fk∗ afk+n =
δ0,n V (a) for n = 0, 1, . . ..
If we multiply the sequence fk from the left by the elements in a suitable
countable approximate unit a1 , a2 , . . . of A then ek := ak fk satisfies ke∗k ek k < kV k,
limk→∞ e∗k aek+n = δ0,n V (a) for n = 0, 1, . . ., and limk→∞ ke∗k ek+n k = 0 for n >
0. 

Remark 3.5.4. Suppose that A is separable and A ⊆ B, and that D ⊆ M(B)

is a unital C *-subalgebra of M(B) with B ⊆ D such that for every b ∈ B there is
an isometry t ∈ D with bt = 0, and that D contains two isometries with orthogonal
ranges.
Then one gets for (ii) and (iii) of Proposition 3.5.3:

(α) for T in (i) holds T (a) = f1∗ af1 + . . . + fm ∗

afm for all a ∈ A with
∗ ∗
f1 , . . . , fm ∈ D such that f1 f1 + . . . + fm fm = kV k · 1 (in particular,
f1 is an isometry in case m = 1 and V contractive),
(β) the elements e1 , e2 , . . . in (iii) can be replaced by isometries t1 , t2 , . . . ∈ D
with t∗j tk = δj,k 1 and limk kV kt∗k atn+k = δ0,n V (a) for a ∈ A.

Proof. (α): Let a0 ∈ A+ a strictly positive element of A, g := a0 + d1 d∗1 +

. . . dm d∗m , h := (kV k · 1 − (d∗1 d1 + . . . + d∗m dm ))1/2 ∈ D, t an isometry in D with
gt = 0. Then fk := dk + (1/m)th ∈ D satisfies d∗k adk = fk∗ afk for a ∈ A and
f1∗ f1 + . . . + fm∗
fm = kV k.1.
(β): Let g := a0 + k 2−k ek e∗k and t ∈ D an isometry with gt = 0. If
P

D contains two isometries with orthogonal ranges, then D contains a countable

sequence s1 , s2 , . . . ∈ D of isometries with pair-wise orthogonal ranges: s∗j sk = δj,k .
If V = 0 then tk = tsk is as desired. If V 6= 0, let rk := ek + tsk (kV k − e∗k ek )1/2 ,
then rk ∈ D, rj∗ rk = e∗j ek + δj,k (kV k − e∗k ek ) and rj∗ ark = e∗j aek for a ∈ A,
j, k = 1, 2, . . .. It follows that there is a sequence u1 , u2 , . . . of unitaries in D such
that limk kuk − 1k = 0 and sk := kV k−1/2 uk rk is a sequence of isometries in D
with s∗j sk = δi,k 1. It must satisfy limk kV ks∗k ask+n = δ0,n V (a) for a ∈ A. 

Definition 3.5.5. Let Y a T0 space (e.g. Y = ∼ Prim(D) for some algebra D),
∼
and O(Y ) its lattice of closed ideals (e.g. O(Y ) = I(D)).
We call a map Ψ : O(Y ) → O(Prim(A)) ∼= I(A) an action or Ψ-action of Y on
a C *-algebra A, if Ψ(U ) ⊆ Ψ(V ) for U ⊆ V . (More generally, one can consider
increasing maps from down-ward directed partially ordered sets in place of O(Y )
into the set I(A) of closed ideals of A.)
 5. OPERATOR-CONVEX CONES OF C.P. MAPS (2) 407

Ψ is by definition lower semi-continuous if Ψ(( Uτ )◦ ) = ( τ Ψ(Uτ ))◦ for every

T T

family of open subsets of Y . (We use this name because the intersection property
implies that Ψ transforms l.s.c. functions on Prim(A) into l.s.c. functions on Y in
a natural manner.)
If ΨA : O(Y ) → I(A) and ΨB : O(Y ) → I(B) are two actions, then a c.p. map
V : A → B is Ψ-equivariant if V (ΨA (U )) ⊆ ΨB (U ) for all U ∈ O(Y ). The set of
Ψ-equivariant maps will be denoted by C(ΨA , ΨB ). Clearly, C(ΨA , ΨB ) is a point-
norm closed operator-convex cone of c.p. maps, i.e., C(ΨA , ΨB )) = Calg (ΨA , ΨB ).
If Y = Prim(B) and ΨB is the identity map idI(B) of O(Y ), then we write
C(ΨA ) for C(ΨA , ΨB ). Clearly V ∈ C(ΨA ) if and only if V (a) ∈ J for J ∈ I(B)
and a ∈ ΨA (UJ ).
If S is a subset of CP(A, B) let
ΨS (J) := {a ∈ A ; V (b∗ ab) ∈ J ∀ V ∈ S, ∀ b ∈ A }
It is a lower semi-continuous action of Prim(B) on A, because O(Prim(B)) ∼
= I(B)
naturally, and, obviously,
\ \
ΨS ( Jτ ) = ΨS (Jτ )
τ τ

for every family {Jτ }τ of closed ideals of B. The action ΨS is possibly degenerate,
e.g. if S = {0}.

Obviously, ΨS = ΨC(S) and C(S) ⊆ C(ΨS ). Clearly V ∈ C(ΨS ) if and only if,
for every a ∈ A, V (a) is contained in the closed ideal of B generated by W (a) for
all W ∈ S.

Example 3.5.6. If B, C ⊆ Aω , then Y := Prim(Aω ) acts on B and C by

the maps from ΨB and ΨC from O(Y ) ∼ = I(Aω ) into I(B) (respectively I(C)) by
ΨB (J) := B ∩ J (respectively ΨC (J) := C ∩ J). The corresponding action of
Prim(C) on B is in general not l.s.c.

The equivalence of (i) and (ii) of Theorem 3.5.2 gives the following characteri-
zation of m.o.c. cones by actions:

Corollary 3.5.7. Let S ⊆ CP(A, B), V : A → B completely positive, and

consider the set T of completely positive maps W ⊗max id from A ⊗max C ∗ (F∞ )
into B ⊗max C ∗ (F∞ ) for W ∈ S.
Then V ∈ C(S) if and only if V ⊗max id ∈ C(ΨS ).

there is a similar result later on

compare results and notations ??

Remark 3.5.8. An action ΨB naturally extends to stabilizations or to tensor-

products: ΨB⊗E (Z) = ΨB (Z) ⊗ E.
If E is simple and nuclear then this extension is unique (as e.g. in the often
considered cases where E is one of O∞ , O2 , K, Mn or the Jiang-Su algebra Z).
408 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

A natural extension of ΨB to C0 (X, B), e.g. for X = [0, 1] or X = (0, 1], is

given by ΨC0 (X,B) (Z) := C0 (X, ΨB (Z)).
Also it extends naturally to Hilbert–B-modules H and to L(H) by ΨH (Z) :=
HΨB (Z),
Next Def. o.k.? ΨL(H) (Z) := { T ∈ L(H) ; T H ⊆ ΨH (Z) } and so on ...
In particular, ΨM (B) (Z) = M (B, ΨB (Z)) = strict closure of ΨB (Z), that is
T ∈ ΨM (B) (Z) if and only if d∗ T d ∈ ΨB (Z). (Note that ΨB (Z) is an ideal of B.)

Definition 3.5.9. If ΨA and ΨB are actions of Y on A and B, then a c.p. map

V ∈ CP(A, B) is called Ψ-residually nuclear if

(i) V is Ψ-equivariant, i.e., V (ΨA (U )) ⊆ ΨB (U ) for every U ∈ O(Y ), and

(ii) the completely positive class map

[V ] : A/ΨA (U ) → B/ΨB (U )

is nuclear for every U ∈ O(Y ).

Crn (ΨA , ΨB ) denotes the cone of Ψ-residually nuclear maps . It is a point-norm

closed (matrix) operator-convex cone of completely positive maps.
If A ⊆ M(B), then Y = Prim(B) acts naturally on A by
Same notation Ψ for ΨA and ΨB ???
ΨA (J) := A ∩ M(B, J) and ΨB (J) = J
(????)
for J ∈ I(B) ∼ = O(Prim(B)), then we write “residually nuclear” instead of
“Ψ-residually nuclear”.
We use the notation Crn (A ⊆ M(B)) or Crn (Ψ) for the operator-convex cone
of residually nuclear maps V : A → B.
That means V (Ψ(J) ∩ A) ⊂ J and [V ]J : A/Ψ(J) → B/J is nuclear for each
J / B.
Note, that, for the natural action of Prim(B) on A ⊆ M(B), a c.p. map
V : A → B is equivariant if and only if V (a) is contained in the closed ideal of B
generated by BaB for every a ∈ A+ .

Remark 3.5.10. A Ψ-equivariant map V : A → B is Ψ-residually nuclear, if

and only if, for every U ∈ O(Y ) and every separable C *-algebra F , the completely
positive map

[V ] ⊗max id : (A/ΨA (U )) ⊗max F → (B/ΨB (U )) ⊗max F

factorizes over A/ΨA (U ) ⊗min F , i.e., the kernel of

(A/ΨA (U )) ⊗max F → (A/ΨA (U )) ⊗min F

is contained in the kernel of [V ] ⊗max id.

 5. OPERATOR-CONVEX CONES OF C.P. MAPS (2) 409

The reason is that a c.p. map T : D → E is nuclear if and only if

T ⊗max id : D ⊗max F → E ⊗max F

factorizes over D ⊗min F for every separable C *-algebra F . The latter can be seen
by a simple Hahn-Banach separation argument using [463, lem. 7.17(i)], the proof
is implicitly contained in the proof of a less general result in [426].
Since the containment of ideals can be checked with help of primitive ideals
and since the maximal tensor product on C *-algebras is short-exact, we get the
following equivalent characterizations of Ψ-residually nuclear maps:
V ∈ CB(A, B) is Ψ-residually nuclear if and only if for every irreducible repre-
sentation
ρ : B ⊗max C ∗ (F∞ ) → L(H) ,
every U ∈ O(Y ) with ρ(ΨB (U ) ⊗max C ∗ (F∞ )) = 0 and every closed ideal J of
C ∗ (F∞ ) with ρ(B ⊗max J) = 0, the completely positive map

ρ ◦ (V ⊗max id) : A ⊗max C ∗ (F∞ ) → L(H)

factorizes over (A/ΨA (U )) ⊗min (C ∗ (F∞ )/J).

By the above cited separation arguments this can be easily seen to be equivalent
to the following:
V ∈ CP(A, B) is Ψ-residually nuclear, if and only if, for every C *-morphism h
from B into a von-Neumann factor N with h(B)00 = N and every U ∈ O(Y ) with
h(ΨB (U )) = 0, holds h(V (ΨA (U ))) = (h ◦ V )(ΨA (U )) = 0 and that the c.p. map
[h ◦ V ] : A/ΨA (U ) → N is weakly nuclear.

Question 3.5.11. Suppose that A is separable. Is Crn (ΨS ) = C(S)rn for

S = C(ΨA , ΨB ) ? I.e., Crn (ΨC(ΨA ,ΨB ) ) = Crn (ΨA , ΨB ) ? This would show that
all operator-convex cones of Ψ-residually nuclear maps from A to B are cones of
residually nuclear maps for suitable lower semi-continuous actions Ψ of Prim(B) on
A.

Question has now (since 2015, unpublished) a positive answer

for separable A and B,
because each stable separable C *-algebra B has
‘‘Abelian’’ factorization.

Lemma 3.5.12. Suppose that A is a C*-algebra, N a von-Neumann factor and

C ⊆ CP(A, N ) an operator-convex cone of completely positive maps.
If C is separating for A, i.e., if W (a) = 0 for all W ∈ C implies a = 0, then
every nuclear c.p. map V : A → N is contained in the point-weak closure of C.

Proof. One can reduce all to the particular case where N is a von-Neumann
subalgebra of L(H), and H contains a cyclic vector η for N , in a way that it suffices
to show that the positive linear functional λ ◦ (V ⊗max id) on A ⊗max M is in the
410 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

point-weak closure of the convex cone κ of linear functionals λ ◦ (W ⊗max id) with
W ∈ C (compare the arguments in proofs of [463, lem. 7.17, 7.18]).
Here M := N 0 ⊆ L(H) is the commutant of N , and let λ the partially normal
positive functional on N ⊗max M given by λ(n ⊗ m) = hmnη, ηi.
The functional λ ◦ (V ⊗max id) factorizes over A ⊗min M , i.e., annihilates the
elements in the kernel of the epimorphism A⊗max M → A⊗min M , because V : A →
N is weakly nuclear.
Since η is an cyclic vector for N , one gets (as in the proof of the implication
(ii)⇒(i) for Theorem 3.5.2) that [463, lem. 7.17(ii)] applies to weak closure of κ, i.e.,
every positive linear functional ϕ on A ⊗max M which annihilates the intersection J
of the ideal-kernels of the functionals in κ is in the point-weak closure of κ. J does
not contain a non-zero elementary element a ⊗ m, because M and N are factors
and C is separating for A.
It follows that J is contained in in the kernel of A ⊗max M → A ⊗min M . Hence
λ ◦ (V ⊗max id) is in the point-weak closure of κ. 

Remark 3.5.13. Let A ⊆ M (D) ⊆ D∗∗ such that A ,→ D∗∗ is weakly nuclear.
Then V : A → D is residually nuclear if and only if V is approximately inner.
In particular, the identity map of every nuclear C *-subalgebra A ⊆ M(D) is
residually nuclear for the natural action of Prim(D) on A.

The following corollaries are combinations of special cases of the above results
combined with some of the results of Sections 4–8.

Corollary 3.5.14. Suppose that A and B are C*-algebras and that S is a set
of c.p. maps V : A → B.

(i) For every J ∈ I(B) holds

\
ΨS (J) = V −1 (J)
V ∈C(S)

where ΨS is the lower semi-continuous action of Prim(B) on A as defined

in Definition 3.5.5.
(ii) Every ΨS -residually nuclear map W : A → B is in C(S).

Proof. (i) follows from b∗ V (a∗ (.)a)b ∈ C(S) for V ∈ C(S).

(ii): Combine Theorem 3.5.2(iii), Remark 3.5.10 and Lemma 3.5.12. 

Corollary 3.5.15. Suppose that B, C ⊆ A are C*-subalgebras of a C*-algebra

A and that V : B → A is a c.p. map with V (B) ⊆ C.

(i) V is a.i. in A, – i.e., V can be approximated inside CP(B, A) in point

norm by restrictions to B of inside A inner completely positive maps b ∈
A 7→ j (aj )∗ baj –, if and only if, V (B ∩ J) ⊆ J for every closed ideal J
P

of A and [h ◦ V ] : B/(B ∩ J) ∼
= h(B) → h(C) ⊆ N is weakly a.i. in N for
every von-Neumann factor representation h : A → N ⊆ L(H) of A.
 6. APPROXIMATE DECOMPOSITION OF RESIDUALLY NUCLEAR MAPS 411

(ii) If B or C is nuclear then V is a.i. in A, if and only if, for every b ∈ B+ ,

V (b) is in the closed ideal span(AbA) of A generated by b.

Corollary 3.5.16. Suppose that A is an exact C*-subalgebra of a C*-algebra

B, and that V : A → B is a c.p. map. We consider the natural action Ψ of Prim(B)
on A given by

Ψ : J ∈ I(B) ∼
= O(Prim(B)) 7→ J ∩ A ∈ I(A) .

(i) V is residually nuclear, if and only if, V is nuclear and Ψ-equivariant with
respect to the natural action of Prim(B) on A, if and only if, V is nuclear
and (finite-step-) a.i. in B.
(ii) Moreover, if B is s.p.i. and , V is 1-step-approximately inner, if V is
residually nuclear.
(iii) Suppose that S ⊆ CP(A, B) satisfies that, for every a ∈ A and ε > 0,
there exists W ∈ S, u ∈ A and v ∈ B such that kv ∗ W (u∗ au)v − ak < ε.
Then every Ψ-residually nuclear map is in C(S). In particular, every
nuclear Ψ-equivariant map V : A → B is approximately one-step inner if
the elements of S are approximately one-step inner.
(iv) If B = Dω for some s.p.i. algebra D, V is a residually nuclear contraction
and A is separable, then there is a contraction d in B with V (b) = d∗ bd
for all b ∈ A.
(v) If, in addition to (iv), D is stable, then there is an isometry S ∈
(M (D))ω ⊆ M (B) such that V (b) = S ∗ bS for b ∈ A.

Proof. To be filled in ?? 

6. Approximate decomposition of residually nuclear maps

We derive now an important later used sufficient condition for approximate

decompositions and for one-step innerness of residually nuclear maps inside ultra-
powers Bω . Recall that D ⊆ M(B) is “non-degenerate” if D · B = B.

Proposition 3.6.1. Suppose that D is a strongly purely infinite non-degenerate

C*-subalgebra of M(B) and that C ⊆ M(D) ⊆ M(B) is a separable C*-subalgebra.
Let X := Prim(D), and define actions of X on C and B by

Ψup ∼
ΨC := D,C : I(D) = O(X) → I(C)

and
ΨD,B ∼
ΨB := down : I(D) = O(X) → I(B) ,

as in Definition 1.2.7. Then:

(i) A map V : C → B is Ψup B,C - idI(B) –residually nuclear, if and only if, V is
ΨC -ΨB –residually nuclear, if and only if, V is idI(C) - ΨC,B
down –residually
nuclear.
412 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

(ii) Every ΨC -ΨB –residually nuclear contraction V : C → B ⊆ Bω can be ap-

proximated in point-norm topology by compositions T ◦ S, where T : Dω →
Bω is a one-step-approximately inner completely positive contraction, and
S : C → Dω is a residually nuclear contraction with respect to the action
Ψup
Dω ,C of Prim(Dω ) on C, where we consider the embeddings

C ⊆ M(D) ⊆ M(D)ω ⊆ M(Dω ) .

(iii) The maps S◦T of Part (ii) can be approximated by 1-step-inner completely
positive contractions.

In particular, V : C → B is approximately inner in M(B) if V is ΨC -ΨB –residually

nuclear.

We use the following Lemma 3.6.2 for the proof of Proposition 3.6.1.

Lemma 3.6.2. Let A ⊆ M(D) and D ⊆ M(B) non-degenerate. Then:

(i) M(B, J) is the strict closure of J / B, and, for every strictly closed ideal
I of M(B) holds I = M(B, J) for J := I ∩ B / B.
(ii) Ψup up up
D,A (ΨB,D (J)) = ΨB,A (J) for all J ∈ I(B).
(iii) ΨA,B up D,B
down (ΨD,A (K)) ⊆ Ψdown (K) for all K ∈ I(D).
(iv) ΨA,B up D,B up
down (ΨB,A (J)) ⊆ Ψdown (ΨB,D (J)) ⊆ J for all J ∈ I(B).
up A,B
(v) L ⊆ ΨB,A (Ψdown (L)) for all L ∈ I(A).

Proof. (i): If s ∈ M(B) and sB ⊆ J, then bs ∈ B and bse ∈ J for all e ∈ B+ ,

which implies that Bs ⊆ J, i.e., s ∈ M(B, J). For t ∈ M(B), s ∈ M(B, J) holds
tsB ⊆ tJ = tBJ ⊆ J and stB ⊆ sB ⊆ J. Thus M(B, J) an ideal of M(B).
If (tα ) ⊆ M(B, J) converges strictly to s ∈ M(B) then sb = lim tα b ∈ J for
all b ∈ B, i.e., s ∈ M(B, J). Thus M(B, J) is a strictly closed ideal of M(B). In
particular, M(B, J) is norm-closed in M(B). Therefore it is a *-ideal of M(B).
For s ∈ M(B, J) and a bounded approximate unit {eα } of B holds that seα ∈ J
converges strictly to s in M(B).
It holds IB ⊆ B ∩ I and J := B ∩ I is a closed ideal of B, thus I ⊆ M(B, J)
and J ⊆ I. It follows M(B, J) = I because M(B, J) is the strict closure of J in
M(B).
(ii): Since D ⊆ M(B) is non-degenerate, there is a unique strictly continuous
and unital embedding ι : M(D) ,→ M(B) with ι(d) = d for all d ∈ D ( 16 ), and we
can identify M(D) with ι(M(D)).
Let K := Ψup B,D (J) = D ∩ M(B, J) . Then M(D, K) ⊆ M(D) ∩ M(B, J),
because M(D, K) is the strict closure of K in M(D) by (i), M(D) ,→ M(B) is
strictly continuous and M(D) ∩ M(B, J) is strictly closed in M(D). The same
arguments show that M(D) ∩ M(B, J) is a strictly closed ideal of M(D). By
(i), there is a closed ideal L of D with M(D, L) = M(D) ∩ M(B, J). It follows

16In general, ι(M(D)) is not strictly closed in M(B), even if D ,→ M(B) is non-degenerate.
 6. APPROXIMATE DECOMPOSITION OF RESIDUALLY NUCLEAR MAPS 413

L ⊆ K and M(D, K) ⊆ M(D, L), ie K = KD ⊆ L. Thus, A ∩ M(B, J) =

A ∩ M(D) ∩ M(B, J) = A ∩ M(D, K), i.e., Ψup up up
B,A (J) = ΨD,A (K) for K = ΨB,D (J).

(iii): Let K / D. Then I := Ψup

D,A (K) ⊆ M(D, K) Thus BDIDB ⊆
BDM(D, K)DB ⊆ BKB . Since DB is dense in B, we get ΨA,B up
down (ΨD,A (K)) ⊆
ΨD,B
down (K).

(vi): Let J / B. Consider K := Ψup up

B,D (J) and L := ΨD,A (K). We have
ΨD,B up
down (K) ⊆ J, because K = ΨB,D (J) ⊆ M(B, J) and BM(B, J)B ⊆ J. On
the other hand, Ψup
B,A (J) = L by (ii), and DLD ⊆ DM(D, K)D ⊆ K. Thus
BDLDB ⊆ BKB. Since DB = B, it follows ΨA,B D,B
down (L) ⊆ Ψdown (K).

(v): Let L / A, and let J := span(BLB) = ΨA,B ∗

down (L)). For x ∈ L, bx xb is in
J := span(BLB), which implies xb ∈ J. Thus LB ⊆ J and BL = (LB)∗ ⊆ J, i.e.,
L ⊆ Ψup
B,A (J). 

Proof of Proposition 3.6.1. Recall that ΨC (K) := Ψup D,C (K) := C ∩

D,B
M(D, K) and that ΨB (K) = Ψdown (K) is defined as the closure of span(BKB)
for K ∈ I(D).
(i): Suppose that V : C → B is Ψup B,C -idI(B) – residually nuclear, and let I
a closed ideal of D. Then M(D, I) ⊆ M(B, BIB) (because IB + BI ⊆ BIB
and DB = B = BD), and C ∩ M(D, I) = ΨC (I) ⊆ Ψup B,C (BIB). It follows
V (ΨC (I)) ⊆ BIB = ΨB (I). Furthermore, [V ]I : C/ΨC (I) → B/BIB satisfies
[V ]I ((C ∩ M(B, BIB))/ΨC (I)) = 0, because V (C ∩ M(B, BIB)) ⊆ BIB. Thus,
[V ]I factorizes [V ]I = [V ]J ◦ π for the nuclear map [V ]J : C/ΨupB,C (J) → B/J where
J := BIB. It follows that V is ΨC -ΨB –residually nuclear.
Suppose that V is ΨC -ΨB –residually nuclear, and let K a closed ideal of C.
Then ΨC,B
down (K) = span(BKB) = BDKDB = BIB for I := DCD. Then K ⊆
C ∩ M(D, I) = ΨC (I) and

V (K) ⊆ V (ΨC (I)) ⊆ ΨB (I) = ΨC,B

down (K) ,

and [V ]K : C/K → B/ span(BKB) = B/ΨB (I) factorizes through the nuclear map
[V ]I : C/ΨC (I) → B/ΨB (I) because K ⊆ ΨC (I). It follows that V is idI(C) -ΨC,B
down –
residually nuclear.
Suppose that V is idI(C) -ΨC,B
down – residually nuclear, and let J a closed ideal of
B. Then K := C ∩ M(B, J) = Ψup B,C (J) is a closed ideal of C with

ΨC,B
down (K) = span(BKB) ⊆ J ,

because cB +Bc ⊆ J for all c ∈ M(B, J). It follows that V (K) ⊆ span(BKB) ⊆ J,
and that [V ] : C/K → C/span(BKB) is nuclear. Thus [V ] : C/K → C/J is also
nuclear and K = ΨupB,C (J) .

Thus, V : C → B is Ψup
B,C -idI(B) – residually nuclear.

(ii)+(iii): First we consider all maps S : C → Dω with the property that the
C *-subalgebra AS := C ∗ (S(C)) of Dω is abelian and that there exist contractions
414 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

s1 , s2 , . . . ∈ Dω with s∗n (C + C1)sm = {0} for n 6= m and s∗n csn = S(c) for all
c ∈ C and n ∈ N. Clearly, the maps S are Ψup Dω ,C -residually nuclear, and, for all
contraction b ∈ Bω , the maps c ∈ C 7→ b∗ S(c)b ∈ Bω are Ψup Bω ,C -residually nuclear
in C ⊆ M(Dω ) ⊆ M(Bω ). By definition, the maps c 7→ b∗ S(c)b are one-step
approximately inner in M(B)ω ⊆ M(Bω ). (Here we use that Dω ⊆ M(B)ω ⊆
M(Bω ) is non-degenerate, i.e., Dω Bω = Bω .)
We are going to show:

(α) The above considered maps c 7→ b∗ S(c)b are point-norm dense in the set
of contractions of a matrix operator-convex cone C ⊆ CPin (C, Bω ).
(β) For every c0 ∈ C+ and b1 , . . . , bn ∈ Bω with k j b∗j bj k ≤ 1 there exist
P

S : C → Dω as above descibed and a contraction b ∈ Bω with b∗ S(c)b =

P ∗
j bj cbk .

By (α) and (β), the operator-convex cone C is a sub-cone of the cone of approxi-
mately inner c.p. maps from C ⊆ M(Bω ) to Bω which has the property that the
action ΨC of I(Bω ) ∼ up
= O(Prim(Bω )) on C defined by C is the same as ΨBω ,C (which
is the action defined by the cone of approximately inner c.p. maps). It follows from
Corollary 3.5.14(ii) that all Ψup
Bω ,C -residually nuclear contractions V from C into
Bω are in the point-norm closure of the set of maps c 7→ b∗ S(c)b as described above.
The Ψup up
B,C -residually nuclear contractions V : C → B define ΨBω ,C -residually nu-
clear contractions V : C → B ⊆ Bω . By (i) this shows that (α) and (β) imply (ii)
and (iii).
For the proof of (α) and (β) we list the following facts (1)-(4) on ultrapowers
Dω of strongly purely infinite C *-algebras D (taken from [463] and [443]):
(1) If X is a separable subset of Dω + Bω , then there exists positive contraction
e ∈ Dω with ex = xe = x for all x ∈ X, and with ec = ce and kcek = kck for all
c ∈ C.
(2) For every separable C *-subalgebra A ⊆ Dω and every commutative sep-
arable C *-subalgebra F ⊆ Dω there exists a contraction d ∈ Dω with df = f d,
f d∗ d = f , d∗ adf = f d∗ ad, d∗ a∗ dd∗ ad = d∗ add∗ a∗ d for all f ∈ F and a ∈ A.
(3) For every separable commutative C *-subalgebra F ⊆ Dω there exists a
*-morphism h : F ⊗ O∞ → Dω with h(f ⊗ 1) = f for all f ∈ F .
(4) For every separable C *-subalgebra A ⊆ Dω and every approximately inner
completely positive map T : A → Dω with commutative C ∗ (T (A)) there exists
d ∈ Dω with kdk2 ≤ kT k and d∗ ad = T (a) for a ∈ A.
(α): to be filled in ??
(β): There are positive contractions e, f ∈ C 0 ∩Dω with ebj = bj and f e = e for
j = 1, . . . , n by (1), used twice. By (2), there is a contraction d ∈ {c0 e, e, f }0 ∩ Dω
with H := C ∗ (d∗ f Cf d ∪ {c0 e, e, f }) commutative and d∗ df = f . Note ed∗ f cf de =
d∗ eced = ed∗ cde for c ∈ C and d∗ e(ec0 )ed = c0 e3 . There is *-monomorphism
k : H ⊗ O∞ → Dω with k(h ⊗ 1) = h for h ∈ H by (3). Let t1 , t2 , . . . denote
 6. APPROXIMATE DECOMPOSITION OF RESIDUALLY NUCLEAR MAPS 415

the canonical generators of O∞ and let S(c) := d∗ eced, sn := edk(f ⊗ tn ), b :=

∗ 3 ∗
P
j k(f ⊗ tj )bj ∈ Bω . Then sn csm = δn,m S(c), S(c0 ) = c0 e , C (S(C)) ⊆ H is
commutative, b∗ b = j b∗j bj and b∗ S(c)b = b∗j S(c)bj . Thus S and b ∈ Bω are as
P P

stipulated. 
Definition 3.6.3. Suppose that C ⊆ CP(A, B) a point-norm closed matrix
operator-convex sub-cone of CP(A, B), and that E, F are nuclear C *-algebras.
We define CP(E, F ) ⊗ C ⊆ CP(E ⊗ A, F ⊗ B) as the point-norm closure C(S)
of Calg (S), where S := {V ⊗ S ; S ∈ C, V ∈ CP(E, F )} .

Notice the m.o.c. cone CP(E, F ) ⊗ C is not the completion of some algebraic
tensor product.
Corollary 3.6.4. Suppose that C ⊆ CP(A, B) is a point-norm closed operator-
convex cone of completely positive maps, that E and F are nuclear C*-algebras, and
T ∈ CP(E ⊗ A, F ⊗ B). Then:

(i) If S2 is a generating set for C and S1 is a generating set for CP(E, F ),

then S := {U ⊗V ; U ∈ S1 , V ∈ S2 } is a generating set for CP(E, F )⊗C.
In particular, CP(Mn , Mn ) ⊗ C is the point-norm closure of Calg (S) for
S := {id ⊗V ; V ∈ C}.
(ii) T1 T2 ∈ CP(E, F ) ⊗ C, if T1 ∈ CP(G, F ) ⊗ C1 and T2 ∈ CP(E, G) ⊗ C2
for a nuclear C*-algebra G and operator-convex cones C1 ⊆ CP(A, C) and
C2 ⊆ CP(C, B) with S2 ◦ S1 ⊆ C for generating subsets S1 ⊆ C1 and
S2 ⊆ C2 .
(iii) The union (over n = 1, 2, . . .) of the compositions S1 S2 S3 with S2 ∈
CP(Mn , Mn ) ⊗ C, S1 ∈ CP(Mn , F ) ⊗ CPin (B, B), and S3 ∈ CP(E, Mn ) ⊗
CPin (A, A) is point-norm dense in CP(E, F ) ⊗ C .
(iv) T ∈ CP(E, F ) ⊗ C, if and only if, the c.p. maps
a ∈ A 7→ (χ ⊗ idB )(T (e ⊗ a)) ∈ B
are in C for every e ∈ E+ and for every pure state χ on F .
(v) The action Ψ : I(F ⊗ B) ∼
= O(Prim(F ⊗ B)) of Prim(F ⊗ B) ∼ = Prim(F ) ⊗
Prim(B) on E ⊗ A satisfies
Ψ(J) = E ⊗ ΨC (JB )
for J / F ⊗ B in I(F ⊗ B), where JB is the biggest of the ideals I / B with
F ⊗ I ⊆ J.
(vi) The natural map Φ from I((F ⊗ B) ⊗max C ∗ (F2 )) into I((F ⊗ A) ⊗max
C ∗ (F2 )) defined by CP(E, F ) ⊗ C satisfies Φ(J) = F ⊗ Φ1 (I(J)), where Φ1
denotes the natural map from I(B ⊗max C ∗ (F2 )) into I(A ⊗max C ∗ (F2 ))
defined by C and I(J) is the biggest ideal I of B ⊗max C ∗ (F2 ) with F ⊗ I ⊆
J.

Recall that the operator-convex cone CPin (A, A) of approximately inner

c.p. maps from A into A is nothing else C(idA ), and that every nuclear map
V ∈ CPnuc (C, C) is approximately inner if C is simple, e.g. CP(Mn , Mn ) = Calg (id).
416 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

Proof. Let H2 a Hilbert A-B–module with infinite repeats, that is naturally

related to C in the sense of Proposition ??, i.e., C is the point-norm closure of
the c.p. coefficient maps a ∈ A 7→ hx, axi ∈ B for x ∈ H2 . Further let H1 a
Hilbert E-F –module (with infinite repeats) that corresponds to CP(E, F ) in the
same way. Then H1 ⊗ H2 is an Hilbert E ⊗ A-F ⊗ B–module (with infinite repeats),
corresponding to the m.o.c. cone CP(E, F ) ⊗ C, cf. Proposition ??.
(i): If S2 ⊆ C is a generating subset of C than a suitable Hilbert A-B–module
H2 can be constructed by infinitely often repeated Hilbert A-B–module sums of
the Stinespring dilations of the V ∈ S1 . We can elaborate a suitable Hilbert E-F -
module H1 from S2 in the same way. Then H1 ⊗H2 is just the Hilbert E ⊗A-F ⊗B–
module sum of the dilations of the maps U ⊗ V with U ∈ S1 and V ∈ S2 (infinitely
often repeated). Thus CP(E, F ) ⊗ C is generated by {U ⊗ V ; U ∈ S1 , V ∈ S2 }.
Here end of proof of (i)? ??
Let J /(F ⊗B)⊗max C ∗ (F2 ) a closed ideal and let Ψ(J) := I/(E⊗A)⊗max C ∗ (F2 )
the maximal closed ideal with T ⊗max id(I) ⊆ J for all generators T = V ⊗ S of
CP(E, F )⊗C, where S ∈ C and V ∈ CP(E, F ) = CPnuc (E, F ). By Proposition ?? ,
then T is in the operator-convex cone C⊗CP(E, F ), if and only if, T ⊗max id(Ψ(J)) ⊆
J for all J /(F ⊗B)⊗max C ∗ (F2 ). Recall here, that it is enough to consider primitive
ideals J of (F ⊗ B) ⊗max C ∗ (F2 ) = F ⊗ (B ⊗max C ∗ (F2 )). Since F is nuclear, they
are sums
J = JF ⊗ (B ⊗max C ∗ (F2 )) + F ⊗ J1 ,
where J1 is a primitive ideal of B ⊗max C ∗ (F2 ) and JF is a primitive ideal of F .
Now note that CP(E, F ) = CPnuc (E, F ) is generated as an operator-convex
point-norm closed cone by maps e ∈ E 7→ ξ(e)f , for f ∈ F+ and ξ is a pure state
on E. Thus, ΨCP(E,F ) (J) = 0 if J / F and J 6= F and, trivially, ΨCP(E,F ) (F ) = E.
(iv): Let e ∈ E+ and χ : F → C a pure state on F . Let C3 ⊆ CP(A, A)
and C1 ⊆ CP(B, B) denote the approximately inner c.p. maps. Then C ◦ C3 = C
and C1 ◦ C = C. Consider the maps V3 : z ∈ C → ze ∈ E and V1 := χ : F →
C . Then V3 ⊗ idA ∈ CP(C, E) ⊗ C3 and V1 ⊗ idB ∈ CP(F, C) ⊗ C1 . It follows
(V1 ⊗ idB ) ◦ T ◦ (V3 ⊗ idA ) ∈ C = C ⊗ CP(C, C) by (ii).
If U ∈ CP(E, F ), V ∈ C, h ∈ M1,m (E A), and k ∈ Mm,1 (F B), then
∗ ∗


S0 (a) := χ ⊗ idB k (U ⊗ V ⊗ idm (h (e ⊗ a)h))k defines a completely positive map
that is in C :
orthogonalize the ei,j -components of hi
and the fi,j -components of ki
Now let T ∈ CP(E, F ) ⊗ C, a1 , . . . , am ∈ A, and ε > 0. Then there are n ∈ N,
Uj ∈ CP(E, F ), Vj ∈ C, fj ∈ M1,n (E A), gj ∈ Mn,1 (F B), j = 1, . . . , n with
X
kT (e ⊗ ak ) − gj∗ (Uj ⊗ Vj ) ⊗ idn (fj∗ (e ⊗ ak )fj )gj k < ε .
j

It follows that there is S ∈ C with

kχ ⊗ idB (T (e ⊗ ak )) − S(ak )k < ε for k = 1, . . . , m .
 6. APPROXIMATE DECOMPOSITION OF RESIDUALLY NUCLEAR MAPS 417

Thus (χ ⊗ id)T (e ⊗ (·)) ∈ C.

try again?
(idB ⊗χ) ◦ T ◦ ((·) ⊗ e) ∈ C, because this happens if T is one of the generators
where S ∈ C and V ∈ CP(E, F ) = CPnuc (E, F ), and because
???????????????????
The set of T ∈ CP(A ⊗ E, B ⊗ F ) with (idB ⊗χ) ◦ T ◦ ((·) ⊗ e) ∈ C for all e ∈ E+
and pure states χ on F does not change the ???????
??
Conversely, suppose that (idB ⊗χ) ◦ T ◦ ((·) ⊗ e) ∈ C for every pure state χ on
F and e ∈ E+ . Since idB ⊗χ is in the point-norm closure of the cone generated by
???? It suffices to show that (id ⊗U W ) ◦ V ∈ C ⊗ CP(C, F ) for c.p. contractions
W : Mn → F and U : F → Mn .
Now, (id ⊗U ) ◦ C ⊗ CP(C, Mn ) ⊆ C ⊗ CP(C, F ) as one can see on the generating
sets.
to be filled in ?? 

Next question has a positive answer:

Question 3.6.5. Suppose that Cj ⊆ CP(Bj , Bj+1 ) (j = 1, 2) and C3 ⊆

CP(B1 , B3 ) are point-norm closed matrix-operator convex cones, and that Sj ⊆ Cj
a generating set for Cj (j = 1, 2) in sense of Definition ??.
Let V ∈ CP(B2 , B3 ) such that V ◦ (b∗ W (·)b) ∈ C3 for all W ∈ S1 and b ∈ B2 .
Is V ◦ C2 ⊆ C3 ?
Let W ∈ CP(B1 , B2 ) and V ◦ (b∗ W (·)b) ∈ C3 for all V ∈ S2 . Is C1 ◦ W ⊆ C3 ?

What about situation of ⊕-closed generating system?

Remark 3.6.6. ?? Suppose that X and Y are T0 -spaces, Ψ0 : O(Y ) → O(X)

lower semi-continuous and Ψ0 : O(X) → O(Y ) upper semi-continuous.
There are maps Φ0 : O(X) → O(Y ) and Φ0 : O(Y ) → O(X), determined by

Φ0 (U ) ⊆ V ⇔ Ψ0 (V ) ⊆ U ,

respectively
Φ0 (V ) ⊆ U ⇔ Ψ0 (U ) ⊆ V .
The map Φ0 is upper semi-continuous, and Φ0 is lower semi-continuous.

Question 3.6.7. Suppose that A is separable, J / A is a closed ideal and that

idA/J is not locally liftable (i.e., that J ⊗ L(`2 ) → A ⊗ L(`2 ) → (A/J) ⊗ L(`2 ) is
not exact), and let B := L(`2 ) (or at least B := L(`2 )/K).
Does there exist a non-nuclear c.p. contraction W : A/J → B such that W ◦πJ : A →
B is nuclear?
The question for B := L(`2 )/K seems to be less difficult.
418 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

Question 3.6.8. Let C ⊆ CP(A, B) a point-norm closed operator-convex cone,

A and B separable. Let ΨA,C denote the action of Prim(B) on A induced by C.
Recall, that ΨA,C (J) := V ∈C V −1 (J) for J ∈ I(B).
T

Denote by Crn ⊆ CP(A, B) the point-norm closed operator-convex cone of Ψ-

residually nuclear maps. (Then Crn ⊆ C.)
Is Crn ⊆ (Crn )rn ?
(Note here that ΨA,C (J) ⊆ ΨA,Crn (J) for J ∈ I(B).)
Is the answer positive if A locally reflexive?

Question 3.6.9. Suppose that S1 ⊆ CP(A, B) is point-norm closed and is

invariant under the operations (OC1) of Definition 3.2.2. Let S2 denote the point-
norm closure of the smallest subset of CP(A, B) that contains S1 and is invariant
under the operations (OC2).
Is S2 konvex ?
(Examples show that the convexity of S1 alone is not enough to get convexity of
S2 .)

Lemma 3.6.10. Suppose that C ⊆ CP (A, B) is the matrix operator-convex cone

of completely positive maps that is generated by a subset S ⊆ C .
Then every contraction T ∈ C can be approximated in point norm by maps
Pn
S := j=1 c∗j Vj (rj∗ (·)rj )cj with suitable m, n ∈ N, Vj ∈ S, rows rj ∈ M1,m (A), and
Pn
columns cj ∈ Mm,1 (B), such that k j=1 c∗j Vj (rj∗ rj )cj k ≤ 1 .

Proof. to be filled in ?? 

Lemma 3.6.11. Separation from cone C ??

Definition 3.6.12. A subset S ⊆ CP(A, B) is invariant under ε-

generalized Cuntz addition, if d∗1 V (e(·)e)d1 + d∗2 W (e(·)e)d2 ∈ S for ev-
ery V, W ∈ CP(A, B), e ∈ A+ , ε > 0, and d1 , d2 ∈ B with d∗1 d2 = 0 and
d∗j dj = (V (e2 ) + W (e2 ) − ε)+ .

Lemma 3.6.13. Suppose that for every b ∈ B+ and ε > 0 there exist d1 , d2 ∈ B
with d∗1 d2 = 0 and kb − d∗j dj k < ε for j = 1, 2 (17).
If S ⊆ CP(A, B) is invariant under ε-generalized Cuntz addition, then the
operator-convex cone generated by S is contained in the point-norm closure of the
smallest subset of CP(A, B) that invariant under the operations (OC2) of Definition
3.2.2 and contains S .

Proof. ?? 

Definition 3.6.14. Let S ⊆ CP(A, B) a subset. Then the smallest subset

Calg (S) of CP(A, B) that contains S and is invariant under the operations (OC1)
and (OC2) of Definition 3.2.2 will be called the operator-convex cone alge-
braically generated by S.
17 It does not say that b itself is infinite, e.g. every stable B has this property.
 6. APPROXIMATE DECOMPOSITION OF RESIDUALLY NUCLEAR MAPS 419

The point-norm closure C(S) of Calg (S) is called the point-norm closed
operator-convex cone generated by S.

Lemma 3.6.15. Let S ⊆ CP(A, B) a set of completely positive maps, and let
Calg (S) denote the the smallest subset of CP(A, B) that contains S and is invariant
under the operations (OC1) and (OC2) of Definition 3.2.2.

(i) The point-norm closure C of Calg (S) is again a matrix operator-convex

cone.
(ii) Every contraction in C can be approximated in point-norm topology
Pn ∗ ∗
by maps of the form k=1 ck (Vk ⊗ idnk )(rk ( · )rk )ck with Vk ∈ S,
row-matrices rk ∈ M1,nk (A), and column-matrices ck ∈ Mnk ,1 (B),
k = 1, . . . , n, that satisfy
n
X
k c∗k (Vk ⊗ idnk )(rk∗ rk )ck k ≤ 1 .
k=1

(iii) In particular, C is the point-norm closure of the convex hull of the set
of maps c∗ (V ⊗ idn )(r∗ ( · )r)c with V ∈ S, row-matrix r ∈ M1,n (A) and
column-matrix c ∈ Mn,1 (B).

Proof. (i): The operations (x, y) ∈ B ⊕ B 7→ b∗1 xb1 + b∗2 yb2 , a ∈ A 7→ r∗ ar ∈

A ⊗ Mn , x ∈ B ⊗ Mn 7→ c∗ xc ∈ B are bounded and V ∈ CP(A, B) 7→ V ⊗ idn ∈
CP(A ⊗ Mn , B ⊗ Mn ) is continuous with respect to point-norm topology. Thus,
the point-norm closure of a subset K ⊆ CP(A, B) that is invariant under the
operations (OC1) and (OC2) is again invariant under the operations (OC1) and
(OC2) in Definition 3.2.2.
(iii): Let X denote the set of maps c∗ (V ⊗ idn )(r∗ (·)r)c with V ∈ S, r ∈
M1,n (A), c ∈ Mn,1 (B), n ∈ N. Then X is invariant under the operations (OC2),
because, if W := c∗1 (V ⊗idm )(r1∗ (·)r1 )c1 , then c∗2 (W ⊗idn )(r2∗ (·)r2 )c2 for is equal the
map c∗3 (V ⊗ idmn )(r3∗ (·)r3 )c3 , with r3 ∈ M1,mn (A) given by (r3 )1,` := (r1 )1,j (r2 )1,k
for ` = (j − 1)n + k ∈ {1 . . . mn} and the entries (ri )1,p of ri (i = 1, 2, 3), and,
similar, with c3 ∈ M1,mn (B) given by (c3 )`,1 := (c1 )j,1 (c2 )k,1 for ` = (j − 1)m + k.
If a subset X ⊆ CP(A, B) is invariant under the operations (OC2), then (in
particular) f ∗ b∗ V (a∗ e∗ (·)ea)bf ∈ X for every a ∈ M(A), e ∈ A, b ∈ M(B) and
f ∈ B. If we use approximate units {eα } of A and {fβ } of B, then this shows
that the point-norm closure Y := X of X contains all maps b∗ V (·)b with b ∈ B
and V ∈ X . With the arguments in proof of part(i), we obtain that S ⊆ Y,
R+ · Y ⊆ Y and that Y contains c∗ (W ⊗ idn )(r∗ (·)r)c if W ∈ Y, r ∈ M1,n (M(A)),
c ∈ Mn,1 (M(B)), n ∈ N.
Thus, the set Z of finite sums W1 + · · · + Wm with Wj ∈ Y is a cone that is
invariant under the operations (OC1) and contains S ∪ X ⊆ Y. Since Y is invariant
under (OC1) and since c∗ ((W1 + · · · + Wm ) ⊗ idn )(r∗ (·)r)c = U1 + · · · + Um for
Uj := c∗ (Wj ⊗ idn )(r∗ (·)r)c, we get moreover that Z is also invariant under the
operations (OC2).
420 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

It follows that Calg (S) ⊆ Z. On the other hand, X ⊆ Calg (S) and Z is contained
in the point-norm closure of the set W of finite sums V1 + · · · + Vm with Vj ∈ X .
It follows that every element of the point-norm closure C(S) of Calg (S) can
Pn ∗
be approximated in point-norm by maps W ∈ W, where W := k=1 ck (Vk ⊗
idnk )(rk∗ ( · )rk )ck with Vk ∈ S, row-matrices rk ∈ M1,nk (A), column-matrices ck ∈
Mnk ,1 (B), k = 1, . . . , n. Obviously, W ⊆ Calg (S) ⊆ C.
(ii): The norm kW k of W ∈ W is given by sup{kW (e)k ; e ∈ A+ , kek ≤ 1}
Pn
which is equal to k k=1 c∗k (Vk ⊗ idnk )(rk∗ rk )ck k . Since W (e(·)e) ∈ W for W ∈ W,
we get from Lemma 3.1.8 that the contractions in the point-norm closure C of W can
be approximated in point-norm by contractions in W. It proves also part(ii). 

Definition 3.6.16. Let C1 ⊆ CP(A, B) and C2 ⊆ CP(E, F ) point-norm closed

matrix operator-convex cones.
We define the point-norm closed m.o.c. cone C1 ⊗ C2 ⊆ CP(A ⊗ E, B ⊗ F )
as the smallest point-norm closed m.o.c. cone that contains all tensor products
S ⊗ T ∈ CP(A ⊗ E, B ⊗ F ) of S ∈ C1 and T ∈ C2 .
The m.o.c. cone C1 ⊗ C2 ⊆ CP(A ⊗ E, B ⊗ F ) is called the “minimal tensor
product” of the m.o.c. cones C1 and C2 .
If A = C0 (X) and F = C, respectively E = C and B = C0 (Y ), then we write
sometimes also [X]C, respectively C[Y ], to simplify notation.

The m.o.c. cone C1 ⊗ C2 is not a tensor product in the usual sense because it
contains usually elements that can not be approximated in point-norm by convex
combinations of elementary tensors S ⊗ T .

Corollary 3.6.17. Suppose that S1 ⊆ CP(A, B), S2 ⊆ CP(E, F ) are subsets

and that C1 ⊆ CP(A, B), C2 ⊆ CP(E, F ) are the point-norm closed matrix operator-
convex cones generated by S1 respectively S2 .
Then the point-norm closed matrix operator-convex cone C1 ⊗ C2 ⊆ CP(A ⊗
E, B ⊗ F ) that is generated by {V ⊗ W ; V ∈ C1 , W ∈ C2 } is also generated by
{S ⊗ T ; S ∈ S1 , T ∈ S2 }.

Proof. Let C3 ⊆ CP(A ⊗ E, B ⊗ F ) denote the point-norm closure of the

operator-convex cone generated by {S ⊗ T ; S ∈ S1 , T ∈ S2 }. Then C3 ⊆ C1 ⊗
C2 and C3 is point-norm closed, is convex and is invariant under the operations
(OC1) and (OC2) by Lemma 3.6.15(i). Thus, C1 ⊗ C2 = C3 if C3 contains every
tensor product V ⊗ W with V ∈ C1 and W ∈ C2 . By Lemma 3.6.15(ii,iii) and
the convexity of the cone C3 is suffices to show that (c∗1 (S ⊗ idm )(r1∗ (·)r1 )c1 ) ⊗
(c∗2 (T ⊗ idn )(r2∗ (·)r2 )c2 ) is in C3 if S ∈ S1 , T ∈ S2 , c1 ∈ Mm,1 (B), r1 ∈ M1,m (A),
c2 ∈ Mn,1 (F ) and r2 ∈ M1,m (E). We have

(c∗1 (S ⊗ idm )(r1∗ (·)r1 )c1 ) ⊗ (c∗2 (T ⊗ idn )(r2∗ (·)r2 )c2 ) = c∗3 ((S ⊗ T ) ⊗ idmn )(r3∗ (·)r3 )c3

for c3 = c1 ⊗ c2 ∈ Mmn,1 (B ⊗ F ) and r3 := r1 ⊗ r2 ∈ M1,mn (A ⊗ E) where we use

the natural isomorphisms Mmn,1 (B ⊗ F ) ∼ = Mm,1 (B) ⊗ Mn,1 (F ), M1,mn (A ⊗ E) ∼
=
 6. APPROXIMATE DECOMPOSITION OF RESIDUALLY NUCLEAR MAPS 421

M1,m (A) ⊗ M1,n (E), (A ⊗ E) ⊗ Mmn ∼

= (A ⊗ Mm ) ⊗ (E ⊗ Mn ) and (A ⊗ E) ⊗ Mmn ∼
=
18
(A ⊗ Mm ) ⊗ (E ⊗ Mn ) ( ). Thus, C3 = C1 ⊗ C2 . 

Definition 3.6.18. Let A and B C *-algebras. We define a matrix-order

S ≤ T on the cone of completely positive maps S, T ∈ CP(A, B) by

S≤T if and only if T − S ∈ CP(A, B) .

Obviously, this definition says that: S ≤ T if and only if there exists R ∈

CP(A, B) with T = S + R. It is not enough to have S(a) ≤ T (a) for all a ∈ A+ .
But S ≤ T is equivalent to the property that for each n ∈ N and all positive
matrices a ∈ Mn (A)+ holds Sn (a) ≤ Tn (a), where Sn and Tn apply to a = [ajk ] as
Sn (a) = [S(ajk )].

Corollary 3.6.19. Every, with respect to the point-norm convergence topology

closed, matrix operator-convex cone C ⊆ CP(A, B) is hereditary with respect to the
matrix-order on CP(A, B) in sense of Definition 3.6.18, i.e., if S, T ∈ CP(A, B)
and S + T ∈ C then S, T ∈ C.
If C ⊆ CP(A, B) is a point-norm closed hereditary convex sub-cone of CP(A, B),
then C is matrix operator-convex, if and only if, the maps U2∗ V (U1∗ (·)U1 )U2 are in C
for every V ∈ C, and for all unitaries U1 := exp(ih) ∈ A + C1 and U2 := exp(ik) ∈
B + C1 with h∗ = h ∈ A, khk < π, k ∗ = k ∈ B and kkk < π .

Proof. If I ⊆ A ⊗max C ∗ (F2 ) and J ⊆ B ⊗max C ∗ (F2 ) are closed ideals with
V ⊗max id(I) ⊆ J for all V ∈ C, then ((S + T ) ⊗max id)(I) ⊆ J for S, T ∈ CP(A, B)
with S + T ∈ C. Let e ∈ I+ , f := S ⊗max id(e) and g := T ⊗max id(e) in
(B ⊗max C ∗ (F2 ))+ . Since f + g ∈ J and f, g ≥ 0, it follows that f, g ∈ J+ ,i.e.,
S ⊗max id(I) ⊆ J and T ⊗max id(I) ⊆ J for all closed ideals I ⊆ A ⊗max C ∗ (F2 ) and
J ⊆ B ⊗max C ∗ (F2 ) with the property V ⊗max id(I) ⊆ J for all V ∈ C. It implies
S, T ∈ C by Theorem 3.8.4.
Every point-norm closed matrix operator-convex cone C is convex, because
V1 + V2 and λV1 are in C by condition (OC1), as one can see with help of an
approximate unit of B. The maps b∗ V (a∗ (·)a)b are in C for every V ∈ C, a ∈ A
and b ∈ B, by condition (OC2).
Suppose now that C is a hereditary point-norm closed convex sub-cone of
CP (A, B) such that the maps U2∗ V (U1∗ (·)U1 )U2 are in C for every V ∈ C, and for
all unitaries U1 := exp(ih) ∈ A + C1 and U2 := exp(ik) ∈ B + C1 with h∗ = h ∈ A,
khk < π, k ∗ = k ∈ B and kkk < π. Since c + i(1 − c2 )1/2 = exp(ih) with h∗ = h of
norm khk ≤ 2π/3 for c∗ = c with kck ≤ 1/2, and since d∗ xd + dxd∗ = 2(hxh + kxk)
for d = h + ik with h∗ = h, k ∗ = k in every C *-algebra, we can use that the convex
cone C is hereditary and that the c.p. maps b∗ V (a∗ (·)a)b are in C for every V ∈ C,
a ∈ A and b ∈ B.
18The isomorphisms are ( – up to the associativity transformation maps – ) induced by the

isomorphism Cmn ∼ = Cm ⊗ Cn given by writing the rows of the tensors (on on the right side) as
consecutive row (on the left side).
422 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

If V ∈ C, r = [a1 , . . . , an ] ∈ M1,n (A) and c = [b1 , . . . , bn ]> ∈ Mn,1 (B), then the
map Z(a) := (n( j a∗j aaj ) ⊗ 1n ) − r∗ ar from A into Mn (A) is completely positive
P

by Lemma A.5.6. The maps c∗ (V ⊗ idn )(a∗j (·)aj ⊗ 1n )c =

P ∗
k bk V (aj (· · · )aj )bk
P ∗
are in C (by additivity of C). Since a 7→ n( j aj aaj ) ⊗ 1n is the sum of the c.p.
maps Z and a 7→ r∗ ar, we get that c∗ (V ⊗ idn )(r∗ (·)r)c ∈ C , i.e., C is invariant
under the operations (OC2) of Definition 3.2.2. It is also closed with respect to the
operations (OC1), because C is point-norm closed, is convex and b∗ V (·)b is in C for
every b ∈ B. 

Corollary 3.6.20. Suppose that C1 ⊆ CP(A, B) and C2 ⊆ CP(B, C) are point-

norm closed operator convex cones, that S1 ⊆ C1 and S2 ⊆ C2 are generating sets
and that X ⊆ B is a subset of B with dense linear span.

(i) Then the set

{W (b∗ V (·)b) ; V ∈ S1 , W ∈ S1 , b ∈ X}

of c.p. maps generates the m.o.c. cone C2 ◦ C1 ⊆ CP(A, C).

(ii) CPin (B) ◦ C1 ◦ CPin (A) = C1 .
(iii) If ???????????????????

Proof. to be filled in ?? 

Lemma 3.6.21. Let S ⊆ CP(A, B) and let C denote the closure of the smallest
subset of CP(A, B) that contains S and is invariant under the operations (OC2) of
Definition 3.2.2.
Suppose that, for every V1 , V2 ∈ S and b1 , b2 ∈ B, there is a subset X ∈ S, such
that

(i) the set of maps c∗ (V ⊗ idn )(r∗ (.)r)c with V ∈ X , r ∈ M1,n (A) and c ∈
Mn,1 (B) has convex point-norm closure Y := X ,
(ii) there is W ∈ CP(A, B) such that b∗1 V1 (·)b1 + b∗2 V2 (·)b2 + W is in Y.

Then C satisfies automatic also (OC1) of Definition 3.2.2.

Proof. to be filled in ?? 

Proposition 3.6.22. Suppose that C ⊆ CP(A, B) is point-norm closed matrix

operator-convex cone, and that E and F are nuclear C*-algebras.

(i) If S ⊆ CP(A, B) generates C, then the set of maps S ⊗ T for S ∈ S and

T := ρ(·)f with f ∈ F+ and ρ a pure state on E generate C ⊗ CP(E, F ).
Check proof and statement of Part (i) !? ??????????? ??
(ii) If R ∈ C ⊗ CP(Mm , Mn ) W1 ∈ CP(E, Mm ), and W2 ∈ CP(Mn , F ), then
V := (idB ⊗W2 ) ◦ R ◦ (idA ⊗W1 ) is contained in C ⊗ CP(E, F ).
(iii) T ∈ CP(A ⊗ E, B ⊗ F ) is in C ⊗ CP(E, F ), if and only if, the map
a ∈ A 7→ (id ⊗ρ)(T (a ⊗ e)) ∈ B is in C for all e ∈ E+ and all pure states
ρ on F .
 6. APPROXIMATE DECOMPOSITION OF RESIDUALLY NUCLEAR MAPS 423

Proof. (i): The set of c.p. maps ρ(·)f with f ∈ F+ and ρ a pure state on E
generate CPnuc (E, F ) = CP(E, F ), by Corollary ??. Thus, the S ⊗ T with S ∈ S
and those T = ρ(·)f generate C ⊗ CP(E, F ) as an m.o.c. cone by Corollary 3.6.17.
(ii,iii): It suffices to compare the corresponding actions of Prim((B⊗F )⊗max C)
on (A ⊗ E) ⊗max C.
By part (i) and Lemma 3.6.15(ii), it suffices to consider the case where R =
c (S ⊗ T ⊗ idp )(r∗ (·)r)c with S ∈ C, T := ψ(·)f0 , r ∈ M1,p (A ⊗ Mm ) and c ∈
∗

Mp,1 (B ⊗ Mn ), f0 ∈ (Mn )+ , ψ pure state on Mm . If U is a suitable unitary in Mm ,

then T (U ∗ αU ) = α1,1 f0
????????????????????
By Lemma 3.1.9(i), there are columns c1 , . . . , cn ∈ Mn,1 (F ) with W2 (β) =
Pn ∗
j=1 cj βcj for all β ∈ Mn .

Let r = (r1 , . . . , rp ) with ????

P
It follows that id ⊗ W2 (R(a ⊗ d)) = ????
and ???????????
to be filled in ?? 

Lemma 3.6.23. Let C ⊆ CP(A, B) a point-norm closed matrix operator-convex

cone, A, B, E, F C*-algebras.

(i) Every W ∈ C ⊗ CPnuc (E, F ) can be approximated in point-norm by sums

of maps

W 0 = (idB ⊗U2 ) ◦ I2 ◦ (V ⊗ idn ) ◦ I1 ◦ (idA ⊗U1 )

where V ∈ C, U1 ∈ CP(E, Mn ), U2 ∈ CP(Mn , F ), and the maps

I1 : A ⊗ Mn → A ⊗ Mn and I2 : B ⊗ Mn → B ⊗ Mn
are both 1-step-inner c.p. maps.
(ii) If W 0 ∈ CP(A ⊗ E, B ⊗ F ) is as in (i) then W 0 ∈ C ⊗ CPnuc (E, F ).

Proof. (i): Let T = S2 ◦ S1 : E → F with S1 : E → C S2 : C → F given by

S1 (e) := ρ(e) for some pure state ρ on E, and S2 : z ∈ C → zf0 ∈ F for some
f0 ∈ F . We know from Remark ??// from Corollary ??// that CPnuc (E, F ) is
generated by maps T of this kind. It implies by Corollary 3.6.17 that the maps
V ⊗ T with V ∈ C and T = S2 ◦ S1 ∈ CP(E, F ) generate C ⊗ CP(E, F ).
Thus, by Lemma ??, every W ∈ C ⊗CPnuc (E, F ) can be approximated by finite
sums of maps
X
W 0 (a) := (cj )∗ (V ⊗ T )((rj )∗ ark ) ck
1≤j,k≤m

where a ∈ A ⊗ Mm , V ∈ C, T = S2 ◦ S2 is of the above type, r1 , . . . , rm ∈ A E,

c1 , . . . , c m ∈ B F .
P pj
We have rj = i=1 xij ⊗ eij and

S1 ((eij )∗ eei0 k ) = ρ((eij )∗ eei0 k ) = hD(e)(D(ei0 k )x), D(eij ))xi

424 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

for the irreducible representation D : E → L(L2 (E, ρ)) with cyclic vector x ∈
L2 (E, ρ) corresponding to ρ. Thus we may suppose that pj = p, eij = ei and
that {D(ei )x}1≤i≤p is an orthonormal basis of of the linear span of {D(eij )x}, i.e.,
Pp
it suffices to consider the case, where rj = i=1 aij ⊗ ei (some of the aij can be
zero). Then
(idA ⊗S1 ⊗ idm )(r∗ (·)r) = J1 ◦ (idA ⊗R1 )(·) ,

where r := [r1 , . . . , rm ] ∈ M1,m (A ⊗ E) ⊆ A ⊗ E ⊗ Mm , J1 : A ⊗ Mp →

A ⊗ Mm given by J1 (·) := X ∗ (·)X with X := [aij ] ∈ Mp,n (A), and R1 (e) :=
[ρ(e∗i eei0 )]i,i0 = κ∗ D(e)κ is defined by the isometry κ : Cp ,→ L2 (E, ρ) with
P
κ(z1 , . . . , zp ) := zi ei x. The map J1 ◦ (idA ⊗R1 ) maps A ⊗ E to A ⊗ Mm .
Pqj
Consider cj := i=1 yji ⊗ fij ∈ B F for j = 1, . . . , m. Let f1 , . . . , fq a basis
Pq
of the linear span of {fij }, then there are unique bji ∈ B with cj = i=1 bji ⊗ bi .
Let Y := [bji ]j,i ∈ Mm,q (B), then J2 (·) := Y ∗ (·)Y maps B ⊗ Mm into B ⊗
Mq . Now we define R2 : Mq → E by R2 (·) := g ∗ (·)g for the column-matrix g :=
1/2 1/2
[f0 f1 , . . . , f0 fq ]> ∈ Mq,1 (F ) .
We get
c∗ ((idB ⊗S2 ) ⊗ idp (·))c = (idB ⊗R2 ) ◦ J2 ,

thus

W 0 = c∗ ((V ⊗ S2 ◦ S1 ) ⊗ idm (r∗ (·)r)c = (idB ⊗R2 ) ◦ J2 ◦ (V ⊗ idm ) ◦ J1 ◦ (idA ⊗R1 ) .

If we let n := max(m, p, q) and consider Mm , Mp and Mq as corners of Mn ,

then X ∈ A ⊗ Mn , Y ∈ B ⊗ Mn , J1 (respectively J2 ) can be considered as 1-
step-inner c.p. map of A ⊗ Mn (respectively B ⊗ Mn ) R1 becomes a c.p. map
U1 : E → Mn , and R2 extends naturally to a c.p. map U2 : Mn → F . Hence,
W 0 = (idB ⊗U2 ) ◦ I2 ◦ (V ⊗ idn ) ◦ I1 ◦ (idA ⊗U2 ) is as desired.
(ii): We show that W 0 ∈ C ⊗ CPnuc (E, F ):
If I1 = idA ⊗ idn and I2 = idB ⊗ idn then W0 = V ⊗ (U2 ◦ U1 ) ∈ C ⊗ CPnuc (E, F ) .
??????????????
Since, W1 := I2 ◦ (V ⊗ idn ) ◦ I1 ∈ C ⊗ CP(Mn ) is suffices to show that W2 =
W1 ◦(idA ⊗U1 ) ∈ (C ⊗CP(A, Mn ))◦(CPin (A)⊗CP(E, Mn )) is in C ⊗CPnuc (E, Mn ),
and that W 0 = (idB ⊗U2 ) ◦ W2 ∈ CPin (B) ⊗ CP(Mn , F ) ◦ (C ⊗ CPnuc (E, Mn )) is in
C ⊗ CPnuc (E, F ).
to be filled in ?? 

Next needed in Chapter 8: ??

Suppose that C1 ⊆ CP(A, B) and C2 ⊆ CP(B, C) are point-norm closed matrix
operator-convex cones, that (E1 , φ1 ) is a C1 -compatible Hilbert (A, B)-module and
(E2 , φ2 ) is a C2 -compatible Hilbert (B, C)-module.
Then the Hilbert (A, C)-module (E3 , φ3 ) := (E1 ⊗B E2 , φ1 (·) ⊗B 1) is C2 ◦ C1 -
compatible.
 6. APPROXIMATE DECOMPOSITION OF RESIDUALLY NUCLEAR MAPS 425

The module (E3 , φ3 ) is generating for C2 ◦ C1 if – moreover – (E1 , φ1 ) and

(E2 , φ2 ) are generating for C1 respectively C2 .
next needed in chp:5 and chp:8,
Stinespring-Kasparov dilation inside C
perhaps shift to the appendix:

Lemma 3.6.24. Suppose that E is a (right) Hilbert B-module and V : A → L(E)

is a completely positive contraction.
Let C ⊆ CP(A, B) denote the point-norm closed m.o.c. cone that is generated
by the c.p. maps Ve : a ∈ A 7→ hV (a)e, ei ∈ B with e ∈ E.
Then there is a (left) Hilbert B-module F and a *-morphism φ : A → L(E⊕B F )
such that for the natural projection π1 : E ⊕B F → E holds

(i) V (a)e = π1 (φ(a)(e, 0)) for all e ∈ E and a ∈ A.

(ii) hφ(·)(e, f ), (e, f )i ∈ C for all (e, f ) ∈ E ⊕B F .

If A is separable, B is σ-unital and E is countably generated over B then F

and φ : A → L(E ⊕B F ) can be found such that F is again countably generated over
B.
If, furthermore, B is stable and E = B (as right Hilbert B-module), then
there is a *-morphisms ψ : A → M(B) and isometries s, t ∈ M(B) such that
ss∗ + tt∗ = 1, V = s∗ φ(·)s, and b∗ φ(·)b ∈ C for all b ∈ B.

Proof. to be filled in ?? 

Definition 3.6.25. Let A and B graded C *-algebras with gradings βA ∈

Aut(A) and βB ∈ Aut(B) and C ⊆ CP(A, B) a m.o.c.cone with C ◦ βA ⊆ C and
βB ◦ C ⊆ C.
We say that a (graded) Hilbert (A, B)-module (E, φ) generates C if (E, φ) is
C-compatible and the vector states

Vx : a ∈ A 7→ hφ(a)x, xi ∈ B

generate C.

Next: What about (OC1)? ??

Remark 3.6.26. One can show that a set V of c.p. maps that is closed under
operations (OC2) and under “local” generalized Cuntz addition has a point-norm
closure that also satisfies (OC1).

Corollary 3.6.27. Suppose that A 6= C is simple and purely infinite and that
X is locally compact.
Then for every a, b ∈ C0 (X, A)+ with kb(x)k ≤ ka(x)k for all x ∈ X and
ε ∈ (0, 1) there exists a contraction d ∈ C0 (X, A) with kb − d∗ adk < ε.
426 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

Proof. Since A is simple, it follows that b(x) is in the ideal generated by c(x),
where c(x) := (a(x) − δka(x)k)+ 6= 0 for δ := ε/2.
Since A 6= C is simple and purely infinite the algebra A is strongly purely
infinite, by Proposition 2.2.1(v). The strong pure infiniteness of A implies that
C0 (X, A) is strongly purely infinite (cf. Corollary 2.17.5), in particular C0 (X, A) is
purely infinite. Thus, there is e ∈ C0 (X, A) with e∗ ce = (b − ε)+ . 

Corollary 3.6.28. A point-norm closed convex subcone C of CP(A, B) is

matrix operator-convex, if and only if, C is invariant under compositions with inner
automorphisms Ad(U ) for unitary elements U ∈ M(A) or U ∈ M(B), and C is a
hereditary sub-cone of CP(A, B) , i.e., T1 + T2 ∈ C implies T1 , T2 ∈ C .

Proof. ?? Use Hahn-Banach separation for C

and for the point-norm closed m.o.c. cone
generated by C. 

Lemma 3.6.29. Suppose that J is a closed ideal of B, (X, ρ) a separable metric

space, and S is a set of continuous maps f : X → B that has the following properties
(i) and (ii).

(i) There exists a continuous function µ(t) on R+ with µ(0) = 0 such that
kf (x) − f (y)k ≤ µ(ρ(x, y)) for all x, y ∈ X and f ∈ S.
(ii) For every f ∈ S, every finite subset Y ⊆ X, every ε > 0 and g ∈ S,
there exists h ∈ S with kπJ (h(y)) − πJ (g(y))k < ε and kf (y) − h(y)k <
ε + kπJ (f (y)) − πJ (g(y))k for all y ∈ Y .

Then for every map F : X → B/J in the point-norm closure of πJ ◦ S there is

point-wise convergent sequence fn ∈ S such that F (x) = πJ (limn fn (x)).
The condition (ii) is automatically satisfied if the map
h(x) := e1/2 f (x)e1/2 + (1 − e)1/2 g(x)(1 − e)1/2
is in S, for every positive contraction e ∈ J+ with kek < 1 and for every f, g ∈ S.

Proof. Let (x1 , x2 , . . .) a dense sequence in X. Suppose that F : X → B/J

is in the point-norm closure of πJ ◦ S . Then, for fixed x, y ∈ X , kF (x) − F (y)k ≤
supf ∈S kf (x) − f (y)k ≤ µ(ρ(x, y)), and there is a sequence fn ∈ S such that
kF (xk ) − πJ (fn (xk ))k < 2−n−2 for k ≤ n. By property (ii) and induction, we find
hn ∈ S with h1 := f1 ,
khn (xk ) − hn+1 (xk )k ≤ 2−n−2 + kπJ (hn (xk )) − πJ (fn+1 (xk ))k
and kπJ (hn+1 (xk ))−πJ (fn+1 (xk ))k < 2−n−3 for k ≤ n+1. Then limn πJ (hn (xk )) =
F (xk ) for all k ∈ N, and khn (xk ) − hn+1 (xk )k ≤ 2−n for k ≤ n. Thus, H(xk ) =
limn hn (xk ) exists for all k ∈ N. It follows, kH(xk ) − H(xj )k ≤ µ(ρ(xk , xj )) for
k, j ∈ N. Since {x1 , x2 , . . .} is dense in X, we get that H extends uniquely to a
continuous map G from X into B with kG(x) − G(y)k ≤ µ(ρ(x, y)). For x ∈ X
and ε > 0 there exist xk with µ(ρ(x, xk )) < ε/3. Then kG(x) − hn (x)k ≤ kH(xk ) −
hn (xk )k + 2ε/3. Thus G(x) = limn hn (x) for all x ∈ X. Since πJ (G(xk )) = F (xk )
 6. APPROXIMATE DECOMPOSITION OF RESIDUALLY NUCLEAR MAPS 427

and F is continuous, it follows πJ ◦ G = F . In particular, F (x) = πJ (limn hn (x))

for all x ∈ X, and (hn ) is a sequence in S that converges point-wise to G.
We show that the condition (ii) is satisfied if the map

h(x) := e1/2 f (x)e1/2 + (1 − e)1/2 g(x)(1 − e)1/2

is in S, for every positive contraction e ∈ J+ with kek < 1 and for every f, g ∈ S :
Consider the continuous function ψ(ξ) := (1 − ξ 2 )1/2 for ξ ∈ [0, 1]. Let Y ⊂ X
a finite subset, ε ∈ (0, 1) and δ := ε/4.
If we let e := d2 for d ∈ J+ with kdk < 1 then h(·) becomes

h(·) := hd (·) := df (·)d + ψ(d)g(·)ψ(d)

depending on d. Then πJ (h(x)) = πJ (g(x)) for all x ∈ X. Notice here that

πJ (d) = 1 because ψ(d) ∈ C ∗ (d, 1). It follows

kf (y) − h(y)k ≤ k[f (y), d]k + k[f (y), ψ(d)]k + kψ(d)(f (y) − g(y))ψ(d)k .

Since B/J is a Banach quotient of B by J, there exists for each y ∈ Y an element

c(y) ∈ J with

kf (y) − g(y) + c(y)k ≤ δ + kπJ (f (y)) − πJ (g(y))k .

Consider the finite sets Q := {c(y) ; y ∈ Y } ⊂ J and R := {f (y) ; y ∈ Y } ⊂ B.

The ideal J contains sequence of elements dn ∈ J+ with kdn k < 1 and

lim(kq − dn qk + kdn q − qk) = 0

n

for all q ∈ Q and limn k[r, dn ]k = 0 for all r ∈ R by the existence of a quasi-
central approximate unit in J+ for B, cf. [616][thms. 1.4.2, 3.12.14]. It follows that
limn kψ(dn )qψ(dn )k = 0 for q ∈ Q and limn k[r, ψ(dn )]k = 0 for r ∈ R. Thus, we
find n ∈ N such that for d := dn the inequalities

kψ(d)(f (y) − g(y))ψ(d)k ≤ δ + kψ(d)(f (y) − g(y) + c(y))ψ(d)k .

and
k[f (y), d]k + k[f (y), ψ(d)]k < 2δ .
Notice that

kψ(d)(f (y) − g(y) + c(y))ψ(d)k ≤ 2δ + kπJ (f (y)) − πJ (g(y))k .

Summing up gives kf (y) − h(y)k < ε + kπJ (f (y)) − πJ (g(y))k for all y ∈ Y . 

Proposition 3.6.30. Suppose that A is separable, that C ⊆ CP(A, B) is a

matrix operator-convex cone in sense of Definition 3.2.2, and that J is a closed
ideal of B.
Then πJ ◦ C := {πJ ◦ V ; V ∈ C} ⊆ CP(A, B/J) is a matrix operator-convex
cone.
The m.o.c. cone πJ ◦ C is closed with respect to the point-norm topology in
L(A, B/J) if C is closed with respect to the point-norm topology in L(A, B).
428 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

Moreover, for every W ∈ πJ ◦ C there exists V ∈ C with kV k = kW k and

W = πJ ◦ V if C is point-norm closed.

Proof. Clearly, πJ ◦ C ⊆ CP(A, B/J). By Definition 3.2.2 of m.o.c. cones C

we have to check that for V1 , V2 , V ∈ C, elements s, t ∈ B/J, rows r = [r1 , . . . , rn ] ∈
M1,n (A) and columns c = [c1 , . . . , cn ]> ∈ Mn,1 (B/J) there exist W0 , W ∈ C such
that πJ (W0 (a)) = s∗ πJ (V1 (a))s + t∗ πJ (V2 (a))t and πJ (W (a)) = c∗ ((πJ ◦ V ) ⊗
idn )(r∗ ar)c for a ∈ A.
Since πJ is surjective, we find b1 , b2 , dk ∈ B with πJ (b1 ) = s, πJ (b2 ) = t
and πJ (dk ) = ck for k = 1, . . . , n. Let W0 (a) := b∗1 V1 (a)b1 + b∗2 V1 (a)b2 and
W (a) := d∗ (V ⊗ idn )(r∗ ar)d for the column d := [d1 , . . . , dn ]> ∈ Mn,1 (B) with
(πJ ⊗ idn )(d) = c.
Then W0 and W are in in C and have the desired properties, i.e., πJ ◦ C is a
(not necessarily closed) m.o.c. cone.
It follows that the point-norm closure C 0 of πJ ◦ C is again an m.o.c. cone, and
that every element of of T ∈ C 0 is the limit in point-norm convergence of a sequence
πJ ◦ Vn ∈ πJ ◦ C with kπJ ◦ Vn k ≤ kT k, cf. Lemma 3.1.8.
We show that πJ ◦ C itself is closed in point-norm topology if C is closed in
point norm topology, i.e., that the set πJ ◦ C of images of c.p. maps V ∈ C under
the map C 3 V → πJ ◦ V ∈ CP(A, B/J) is closed under point-wise convergence.
Notice first that V := e1/2 V1 (·)e1/2 + (1 − e)1/2 V2 (·)(1 − e)1/2 is in C for
contractions e ∈ J+ with kek ≤ 1 and V1 , V2 ∈ C, and that kV k ≤ max(kV1 k, kV2 k).
To see that V is in the point-norm closure of C it suffices to observe that for an
approximative unit {cτ } ⊂ B+ of B the maps a ∈ A 7→ (1−e)1/2 cτ V2 (a)cτ (1−e)1/2
converge point-wise to (1 − e)1/2 V2 (·)(1 − e)1/2 .
Thus the condition (ii) of Lemma 3.6.29 is satisfied for X := A, ρ(x, y) =
kx − yk, the set C of c.p. maps V ∈ C ⊂ CP(A, B) (in place of S there) and the map
f : V 7→ π ◦ V . The condition (i) of Lemma 3.6.29 is also satisfied, with µ(t) := t.
Thus, πJ ◦ C ⊆ CP(A, B/J) is point-norm closed by Lemma 3.6.29.
The same arguments apply if we apply Lemma 3.6.29 to the set S of V ∈ C ⊂
with kV k ≤ 1: This S is closed in the point-norm topology (if C is closed in point-
norm topology), X := A is separable and

V := e1/2 V1 (·)e1/2 + (1 − e)1/2 V2 (·)(1 − e)1/2 ∈ S

for V1 , V2 ∈ S and e ∈ J+ with kek ≤ 1.

Thus the image πJ ◦ S is closed in the topology of point-norm convergence on
A.
Since πJ ◦ S is closed under point-wise convergence, it suffices to show that
πJ ◦ S is dense in the set of W ∈ πJ ◦ C with kW k ≤ 1, to get that each W ∈ πJ ◦ C
with kW k ≤ 1 is in πJ ◦ S itself.
 7. OPERATOR-CONVEX CONES VERSUS ACTIONS BY TOPOLOGICAL SPACES 429

Let W ∈ πJ ◦ S with kW k ≤ 1. Above we have shown that there exists V ∈ C

with π ◦ V = W . Since A is separable, there exists a strictly positive contraction
c ∈ A+ . It satisfies for all n ∈ N that
dist(V (c2/n ), J) = kπJ (V (c2/n ))k = kW (c2/n )k ≤ kck2/n ≤ 1 .
The sequence of c.p. contractions a ∈ A 7→ W (c1/n ac1/n ) converge point-wise to
W , because a = limn→∞ c1/n ac1/n for each a ∈ A.
Thus, it suffices to find for each contraction c ∈ A+ a sequence of positive
contractions bm ∈ B+ such that kbm V (c2 )bm k ≤ 1 and
lim kπJ (bm )W (cac)πJ (bm ) − W (cac)k = 0 for each a ∈ A .
m

Since W (cAc) is contained in the hereditary C *-subalgebra of B/J generated by

W (c2 ), it suffices to find contractions bm ∈ C ∗ (V (c2 ))+ with πJ (bm ) = W (c2 )1/m
and kb2m V (c2 )k ≤ 1, e.g. we can take bm := ϕ(V (c2 )1/m ) for the function ϕ(ξ) :=
min(ξ, 1). 

Proposition 3.6.30 and the surjectivity in Proposition 3.1.9(iii) together imply

the following equivalent formulation of the Choi-Effros lifting theorem for nuclear
maps [140], cf. also [43].

Corollary 3.6.31. If A is separable, and J is a closed ideal of B, then

πJ ◦ CPnuc (A, B) = CPnuc (A, B/J) .

Proof. Clearly, πJ ◦ CPnuc (A, B) ⊆ CPnuc (A, B/J) . Proposition 3.1.9(iii)

says that πJ ◦ CPf (A, B) = CPf (A, B/J). The m.o.c. cones CPnuc (A, B)
and CPnuc (A, B/J) are the point-norm closures of CPf (A, B) respectively of
CPf (A, B/J). It follows
CPf (A, B/J) ⊆ πJ ◦ CPnuc (A, B) ⊆ CPnuc (A, B/J) .
But πJ ◦ CPnuc (A, B) is point-norm closed by Proposition 3.6.30. 

7. Operator-convex Cones versus Actions by topological Spaces

Lemma 3.7.1. Suppose that C ⊆ CP(B, A) is an m.o.c. cone and that

Φ : I(B) → I(A) is the related upper s.c. action defined by C (i.e., Φ(J) := the
closed ideal of A, that is generated by {V (b) ; b ∈ J+ , V ∈ C }).
Let Jb := span(BbB) . Then {V (b) ; b ∈ J+ , V ∈ C } = Φ(J)+ , and, for every
b ∈ B+ , a ∈ Φ(Jb )+ and ε > 0, there exist V ∈ C with kV (b) − ak < ε.
Conversely:
Let Φ : I(B) → I(A) an upper semi-continuous action, and Ψ : I(A) → I(B) its
lower semi-continuous Galois adjoint. Suppose that C := CPrn (Ψ; B, A) realizes
Ψ, in the sense that Ψ = ΨC . Then ΦC = Φ.

Proof. ?? 
430 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

Give (or refer to) definitions of CPrn (Ψ; B, A), CPrn (Prim(B), Ψ, id; A, B),
CPrn (Prim(A), id, Φ; A, B), CPrn (Prim(B), id, id; B, B), CPin (B, B) = C(idB ),
CP(Prim(B); B, B).

Lemma 3.7.2. Let Ψ : I(B) → I(A) a lower s.c. action and Φ : I(A) → I(B)
its upper s.c. Galois adjoint. Then

CPrn (Prim(B), Ψ, id; A, B) = CPrn (Prim(A), id, Φ; A, B)

and
CP(Prim(B), Ψ, id; A, B) = CP(Prim(A), id, Φ; A, B) .

Proof. ?? 

Remark 3.7.3. Let CPrn (B) := CPrn (B, B) := CPrn (Prim(B), id, id; B, B).
Then
CPrn (B) ⊆ CPin (B, B) = C(idB ) ⊆ CP(Prim(B); B, B) .

If B is separable and nuclear, then all this 3 m.o.c. cones are identical. (For
the proof one can use that then B ⊗ O2 contains a regular Abelian C *-subalgebra,
cf. [359], and – therefore – can apply Proposition ?? about non-commutative se-
lection.)
The content of ref. to missing Prop. should be:
If B is separable and nuclear, and B ⊗ O2 contains a “regular Abelian C *-
subalgebra” (here only cited) then this implies the existence of sufficient many
“n.c. selections” that generate CP(Prim(B); B, B) ???
If separable B 6= {0} is simple and is not nuclear, then CPrn (B), CPin (B, B)
and CP(B, B) are all different m.o.c. cones, but define the same lower s.c. action
of Prim(B) = {{0}} on B. This happens e.g. for the exact simple C *-algebra
∗
B := Cred (SL(2, Z)).

Proposition 3.7.4. If B is separable and B⊗O2 contains a regular abelian C*-

subalgebra (cf. Definition B.4.1) C ⊆ B ⊗O2 , then, for every lower semi-continuous
action Ψ : I(B) → I(A) of Prim(B) on a separable C*-algebra A is realizable by
CPrn (Ψ; A, B), i.e., Ψ = ΨC for C := CPrn (Prim(B), Ψ, id; A, B).

Proof. to be filled in ??
transfer proof from chp. 12 to here ! 

Proposition 3.7.4 implies immediately:

Corollary 3.7.5. If B is separable and B ⊗ O2 contains a regular abelian

C*-subalgebra (cf. Definition B.4.1) C ⊆ B ⊗ O2 , then B has residual nuclear
separation (cf. Definition 1.2.3).

Proposition 3.7.6. Suppose that B has residually nuclear separation (cf. Def-
inition 1.2.3). Then B has the Weyl–von-Neumann property (cf. Definition 1.2.3),
if and only if, B is strongly purely infinite (cf. Definition 1.2.2).
 8. SEPARATION OF M.O.C. CONES BY ACTIONS 431

Proof. To be filled in ??
Proof was somewhere outlined, perhaps in Chp. 1 ? 

8. Separation of m.o.c. cones by actions

There are proofs of some of the following corollaries, that are more elementary
than given below. But we want to invite the reader to use (and think about) the
conceptional ideas presented here. For possible future classification of some classes
of stably finite algebras with non-trivial traces, one has to produce more refined
tools, that give a “applicable” answers e.g. to the following more difficult question:
When is a given completely positive map V in a given cone C ⊆ CP(A, B) of
completely positive maps compatible with a given map T (B) → T (A)?
Here T (B) means the set of lower semi-continuous 2-quasi-traces τ : B+ →
[0, ∞] – possibly with values only in {0, ∞}. Our results together contain an an-
swer in the special case where τ (B+ ) ⊆ {0, ∞} for all τ ∈ T (B), because then
the problem reduces to to the determination of CPnuc (Ψ; A, B) for a lower semi-
continuous action Ψ : I(B) ∼= O(Prim(B)) → I(A), if we identify a {0, ∞}-valued
lower semi-continuous 2-quasi-trace τ : B+ → [0, ∞] by its kernel J := span{a ∈
B+ ; τ (a) = 0}. Notice that we do not require operator convexity in the following
definition.
Compare for next Def. also Def. 3.12.1

Definition 3.8.1. Let S ⊆ CP(A, B) any subset and J ∈ I(B).

We define the ideal ΨS (J) ∈ I(A) by its positive part ΨS (J)+ as the set of all
a ∈ A+ with V (exp(−ih)a exp(ih)) ∈ J for all h∗ = h ∈ A with khk < π and for all
V ∈ S. Then ΨS (J)+ is the positive part of a closed ideal ΨS (J) of A, and

ΨS : I(B) ∼
= O(Prim(B)) ∼
= O(prime(B)) → I(A)

is a lower semi-continuous action of Prim(B) on A, cf. Lemma 3.12.2(iv).

Let I any closed ideal of A. We denote by ΨS (I) ∈ I(B) the smallest closed
ideal of B that contains {V (a) ; a ∈ I, V ∈ S}. The map

ΨS : I(A) ∼
= O(Prim(A)) ∼
= O(prime(B)) → I(B)

is an upper semi-continuous action, cf. Lemma 3.12.2 (xii).

Let X a topological space, and let ΨA : O(X) → I(A) and ΨB : O(X) →
I(B) increasing maps, i.e., actions of X on A and B. We define the cone of Ψ-
equivariant c.p. maps as the set CΨ := CP(ΨA , ΨB ; A, B) ⊆ CP(A, B) of maps
V ∈ CP(A, B) with V (ΨA (U )) ⊆ ΨB (U ) for all open subsets U ⊆ X. In the special
case, where X := Prim(B) and ΨB : O(Prim(B)) → I(B) is given by the natural
identification of the open subsets U of Prim(B) with the corresponding closed ideal
JU of B, where JU denotes the intersection of all primitive ideals I ∈ Prim(B) \ U ,
we write Ψ := ΨA and drop ΨB , i.e., write CΨ := CP(Ψ; A, B) ⊆ CP(A, B) where
Ψ : O(Prim(B)) → I(A) is a lower semi-continuous action.
432 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

The use of the notation CΨ , in place of CΨA ,ΨB , becomes justified in a different
way in part (i) of Lemma 3.12.2.

Compare next with Lemma 3.12.2 in eksec3-Part2 .tex

Lemma 3.8.2. Let A and B C*-algebras, S ⊆ CP(A, B) a set of completely

positive maps, C ⊆ CP(A, B) a m.o.c. cone, X a topological space, ΨA : O(X) →
I(A) and ΨB : O(X) → I(B) actions of X on A respectively B,
Suppose that ?????????????
List of necessary assumptions? ??
Look for minimal m.o.c. cone assumptions!!!
Part (i) is only lattice stuff ?

(i) There exists a lower semi-continuous action Ψ0 : O(Prim(B)) ∼ = I(B) →

0
I(A) with Ψ (ΨB (U )) ⊇ ΨA (U )), which is minimal in the sense that
Φ(J) ⊃ Ψ0 (J) for every J ∈ I(B) if Φ : I(B) → I(A) is lower semi-
continuous and Φ(ΨB (U )) ⊃ ΨA (U ).
This (minimal) action satisfies CΨ0 = CΨ .
(ii) The point-norm closure of an m.o.c. cone (in the algebraic sense) is a
matrix operator-convex cone.
The intersection of a family of m.o.c. cones is am m.o.c. cone.
(iii) Let Calg (S) denote the smallest “algebraic” m.o.c. cone that contains a
subset S ⊆ CP(A, B).
Every contraction V in the point-norm closure C(S) of can be
approximated by maps W ∈ Calg (S) of the particular form W :=
P ∗ ∗
k ck (Vk ⊗ idn )(rk (·)rk )ck with Vk ∈ S, rk ∈ M1,n (A), ck ∈ Mn,1 (B) and
k k c∗k Vk (rk∗ rk )ck k ≤ 1.
P

(iv) ΨS (J)+ is the positive part of the closed ideal ΨS (J) of A defined in
Definition 3.8.1, and

ΨS : O(Prim(B)) ∼
= I(B) 3 J 7→ ΨS (J) ∈ I(A)

is a lower semi-continuous action of Prim(B) on A.

The action ΨS satisfies ΨS (B) = A and V (ΨS (J)) ⊆ J for all J ∈ I(B)
and V ∈ S.
(v) ΨS = ΨC for C := C(S).
(vi) If S1 ⊆ S then ΨS (J) ⊆ ΨS1 (J) for all J ∈ I(B).
(vii) The set CΨ is a point norm-closed m.o.c. cone of completely positive maps.
(viii) CΨ1 ⊆ CΨ if Ψ(J) ⊆ Ψ1 (J) for all J ∈ I(B).
(ix) C ⊆ CΨC .
(x) ΨCΨ (J) ⊃ Ψ(J) for J ∈ I(B) ∼ = O(Prim(B)).
0 max max
(xi) If C ⊆ CP(A ⊗ D, B ⊗ D) is a point-norm closed m.o.c. cone,
then the set C ⊆ CP(A, B) of T ∈ CP(A, B) with T ⊗max idD ∈ C 0 is a
point-norm closed m.o.c. cone.
 8. SEPARATION OF M.O.C. CONES BY ACTIONS 433

(xii) Let M a W*-algebra and B ⊆ M a σ(M, M∗ )-dense C*-subalgebra of M ,

and let C ⊆ CP(A, B) ⊆ CP(A, M ) an m.o.c. cone. Then the point-*ultra-
strong closure C ⊆ CP(A, M ) of C is an m.o.c. cone that is also closed in
the point-ultra-weak topology (σ(M, M∗ )-topology on M ).
(xiii) ΨS : I(A) → I(B) is an upper semi-continuous action of Prim(A) on
B. I ⊆ ΨS (ΨS (I)) for every closed ideal I of A and every subset S ⊆
CP(A, B). It says that ΨS is the Galois adjoint of ΨS and vice-versa
(ix) What about ΨS ◦ ΨS ?

Proof. (iv): Let J a closed ideal of B and T : A → B a positive map. The

set of a ∈ A+ with (πJ ◦ T )(a) = 0 is a hereditary closed convex sub-cone CT,J of
A+ . Thus, the set CS,J of a ∈ A+ with T (a) ∈ J for all T ∈ S is a hereditary
closed convex sub-cone of A+ , and ΨS (J)+ ⊆ A+ of Definition 3.8.1 is the set of
a ∈ A+ with exp(ih)a exp(−ih) ∈ CS,J .
By Lemma A.25.1, ΨS (J)+ is the positive part of a closed ideal ΨS (J) of A.
Clearly, ΨS (B) = A, and V (ΨS (J)+ ) ⊆ J for J ∈ I(B) and V ∈ S, by definition
of ΨS .
T
If {Jσ }σ∈Σ is a family of closed ideals and a ∈ A+ and let J := σ Jσ . then
V (exp(ih)a exp(−ih)) ∈ J for all V ∈ S and h∗ = h ∈ A with khk < π, if and only
if, V (exp(ih)a exp(−ih)) ∈ Jσ for all V ∈ S and h∗ = h ∈ A with khk < π, i.e.,
T T
a ∈ ΨS (J)+ if and only if a ∈ σ ΨS (Jσ )+ = ( σ ΨS (Jσ ))+ . Thus,

ΨS : O(Prim(B)) ∼
= I(B) 3 J 7→ ΨS (J) ∈ I(A)

is a lower semi-continuous map. The lower semi-continuity of the map ΨS implies

that ΨS is monotone. Thus, ΨS is a lower s.c. action of Prim(B) on A.
(vii): Since ΨB (U ) is a closed ideal, the set of maps V ∈ CP(A, B) with
V (a) ∈ ΨB (U ) for all a ∈ ΨA (U ) is point-norm closed. Thus, CΨ is point-norm
closed. Similar arguments show:
Let V1 , V2 ∈ CP(A, B) with Vk (ΨA (U )) ⊆ ΨB (U ) for all U ∈ O(X) (k = 1, 2),
t ∈ [0, ∞), and r ∈ M1,n (A), c ∈ Mn,1 (B). Then (V1 + tV2 )(a) ∈ ΨB (U ) and
c∗ (V1 ⊗ idn )(r∗ ar)c ∈ ΨB (U ) for U ∈ O(X) and a ∈ ΨA (U ) . Thus CΨ is a point-
norm closed m.o.c. cone of c.p. maps.
(i): If we use the natural isomorphisms I(A) ∼ = O(Prim(A)) and I(B) ∼ =
O(Prim(B)), and Lemma ??, then we get that there is a unique minimal lower semi-
continuous action Ψ0 : I(B) → I(A) of O(Prim(B)) on I(A) with Ψ0 (ΨB (U )) ⊃
ΨA (U ).
If V ∈ CΨ0 , then V (ΨA (U )) ⊆ V (Ψ0 (ΨB (U ))) ⊆ ΨB (U ) for U ∈ O(X) , because
ΨA (U ) ⊆ Ψ0 (ΨB (U )) . Thus, CΨ0 ⊆ CΨ = CΨA ,ΨB .
Let S := CΨ , then ΨS : I(B) =∼ O(Prim(B)) → I(A) is a lower semi-continuous
action by part (iv) with V (ΨS (J)) ⊆ J for all V ∈ S .
Since V (exp(ih)a exp(−ih)) ∈ ΨB (U ) for all a ∈ ΨA (U )+ and h∗ = h ∈ A,
ΨS (ΨB (U ))+ ⊃ ΨA (U )+ for U ∈ O(X). Thus ΨS (J) ⊃ Ψ0 (J) for every J ∈ I(B)
434 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

by minimality of Ψ0 . It implies that V (Ψ0 (J)) ⊆ V (ΨS (J)) ⊆ J, i.e., V ∈ CΨ0 , for
all V ∈ CΨ = CΨA ,ΨB .
(ii): The point-norm closure C of an m.o.c. cone C0 is an m.o.c. cone, because
the topology of point-norm convergence on L(A, B) coincides with the strong oper-
ator topology, and because Vα ⊗ idn converges in point-norm to V ⊗ idn if Vα ∈ C0
converges to V .
(iii): Let n ∈ N and let C (n) denote the set of completely positive maps W :=
∗ ∗
P
k ck (Vk ⊗ idn )(rk (·)rk )ck with Vk ∈ S, rk ∈ M1,n (A) and ck ∈ Mn,1 (B), and
let C0 := n C . Clearly, C (n) ⊆ C for every matrix operator-convex cone C ⊆
S (n)

CP(A, B) with S ⊆ C, and the set C (n) is closed under multiplication with positive
scalars and under addition.
Moreover, c∗ (W ⊗ idm )(r∗ (·)r)c ∈ C (mn) if W ∈ C (n) , r ∈ M1,m (A) and c ∈
Mn,1 (B), because

c∗ (Wk ⊗ idm )(r∗ (·)r)c = C ∗ Vk ⊗ idmn (R∗ (·)R)C

for Wk := c∗k (Vk ⊗ idn )(rk∗ (·)rk )ck , R = [R1 , . . . , Rmn ] ∈ M1,mn (A) and C =
(k) (k)
[C1 , . . . , Cmn ]> ∈ Mmn,1 (B), where Rjm+y := ry rj and Cjm+y := cj cy with
(k) (k) (k) (k)
rk = [r1 , . . . , rn ] ∈ M1,n (A), r = [r1 , . . . , rm ] ∈ M1,m (A), ck = [c1 , . . . , cn ]> ∈
Mn,1 (B), and c = [c1 , . . . , cm ]> ∈ Mm,1 (B).
If we fill the rows rk ∈ M1,n (A) and columns ck ∈ Mn,1 (B) with zeros, then
we see that C (n) ⊆ C (n+k) for all n, k ∈ N. Thus C0 = n C (n) is a (not necessarily
S

closed) m.o.c. cone that is contained in every matrix operator convex cone C with
S ⊆ C.
Since A has an approximate unit {eτ } of positive contractions, we get kW k =
k k c∗k Vk (rk∗ rk )ck k for W ∈ C0 , and that every element V ∈ S is in the point-norm
P

closure of C (1) ⊆ C0 .
By part (ii), the point-norm closure C1 of C0 is an m.o.c. cone that contains S.
Since C0 is contained in every point-norm closed m.o.c. cone C with S ⊆ C, we get
C(S) = C1 . The contractions T in the point-norm closure of C0 can be approximated
by contractions W ∈ C0 with kW k ≤ 1 by Lemma 3.1.8.
(v): Since S ⊆ C(S) =: C, it holds ΨC (J) ⊆ ΨS (J) for all J ∈ I(B), cf. part
(vi).
If a ∈ ΨS (J)+ then V (exp(ih)a exp(−ih)) ∈ J for all h∗ = h ∈ A with khk < π
and all V ∈ S .
By part (iv), this implies that c∗` V (r`∗ exp(ih)a exp(−ih)rk )ck ∈ J for V ∈ S,
∗
h = h ∈ A, r1 , . . . , rn ∈ A, c1 , . . . , cn ∈ B. Thus, W (exp(ih)a exp(−ih)) ∈ J for
all W ∈ C := C(S) and h∗ = h by part (iii), i.e., a ∈ ΨC (J)+ , and ΨS = ΨC for
C := C(S).
(vi): Straight from definition of ΨS .
(viii): If V ∈ CΨ1 and Ψ(J) ⊆ Ψ1 (J), then V (Ψ(J)) ⊆ V (Ψ1 (J)) ⊆ J. Thus,
V ∈ CΨ if Ψ(J) ⊆ Ψ1 (J) for all J ∈ I(B).
 8. SEPARATION OF M.O.C. CONES BY ACTIONS 435

(ix): For J ∈ I(B) and a ∈ ΨC (J) holds V (a) ∈ J for all V ∈ C by definition
of ΨC .
(x): Let J ∈ I(B), and let a ∈ Ψ(J)+ ⊆ A+ , V ∈ CΨ and h∗ = h ∈ A. Then
V (exp(ih)a exp(−ih)) ∈ J by definition of CΨ , i.e., a ∈ ΨCΨ (J)+ by definition of
ΨC . Thus, Ψ(J) ⊆ ΨCΨ (J).
(xi): If {Tγ }γ∈Γ ⊆ CP(A, B) is a (norm-)bounded net that converges in point-
norm to T ∈ L(A, B), then T ∈ CP(A, B) and the net
{Tγ ⊗max id}γ∈Γ ⊆ CP(A ⊗max D, B ⊗max D)
converges point-wise to Tγ ⊗max id, because A D is dense in A ⊗max D. Thus, if
Tγ ⊗max id ∈ C 0 for all γ ∈ Γ, then T ⊗max id ∈ C 0 . Clearly, the set C of T ∈ CP(A, B)
with T ⊗max id ∈ C 0 is convex. If T ⊗max id ∈ C 0 , r ∈ M1n (A), c ∈ Mn1 (B), e ∈ D+ ,
then R := r ⊗ e ∈ M1n (A ⊗ D), C := c ⊗ e ∈ Mn1 (B ⊗ D), and
R∗ ((T ⊗max id) ⊗ (idn ))(C ∗ (·)C)R = T 0 ⊗max Se
for T 0 := r∗ (T ⊗ idn )(c(·)c∗ )r Se (d) := e2 de2 (d ∈ D). Since C 0 is matrix operator
convex and is point-norm closed,it follows that T 0 ⊗max id ∈ C 0 . Thus, the set
C ⊆ CP(A, B) of T ∈ CP(A, B) with T ⊗max idD ∈ C 0 is a point-norm closed m.o.c.
cone.
norm
(xii): The point-norm closure C0 := C of C is a convex cone with c∗ (V ⊗
∗
idn )(r (·)r)c ∈ C0 for r ∈ Mn,1 (A), c ∈ M1,n (B) and V ∈ C0 . Since the point-
*strong closures of C and C0 in L(A, M ) coincide, and since Vγ ⊗ idn → W ⊗
idn point-*strongly if Vγ → W point-*strongly , we obtain that the point-*strong
closure C ⊆ L(A, M ) is a convex cone that satisfies c∗ (W ⊗ idn )(r∗ (·)r)c ∈ C0 for
r ∈ Mn,1 (A), c ∈ M1,n (B) and W ∈ C. The unit-ball of Mn (B) is *-ultra-strongly
dense in the unit-ball of Mn (M ) by Kaplansky density theorem ([616, Thm. 2.3.3]).
Thus, for c ∈ M1,n (M ), there exists a net cγ ∈ M1,n (B) with kcγ k ≤ kck that
*strongly converges to c. Hence, the point-*strong closure C satisfies (OC2).
If {Vγ } ⊆ C is a directed net that converges in point-σ(M, M∗ ) topology to
W ∈ Lin(A, M ), then W (e2 ) ≥ 0 for every e ∈ A+ . In particular, W is positive
and bounded, thus lim supγ kVγ (e)k =: µ(e) < ∞ for every e ∈ A+ , and µ :=
lim supγ kVγ k < ∞ by two-fold application of the Banach-Steinhaus theorem. It
implies that W is also in the point-σ(M, M∗ ) closure of the convex set of V ∈ C
with kV k ≤ 1 + ν. Now a Hahn-Banach separation argument shows that W ∈ C.
(xiii): ΨS ( γ Iγ ) = S
P P S
γ Ψ (Iγ ) and a ∈ ΨS (Ψ (I)) for a ∈ I / A follow
straight from the definitions of ΨS and ΨS . 

Compare also Example 3.12.3!!!

Example 3.8.3. We consider the action Ψnuc : I(B) → I(A) defined by
CPnuc (A, B) and examine some possible generating subsets S0 ⊆ CP(A, B) for
CPnuc (A, B).
Let S ⊆ A∗+ denote an “almost separating” set of positive functionals on A
that is invariant under inner automorphisms of A, more precisely:
436 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

for each a ∈ A+ there is ϕ ∈ S and h∗ = h ∈ A with khk < π such that with
ϕ(exp(ih)a exp(−ih)) > 0.
For example we can take as S the set of pure states of A, or any set S of
pure-states ϕ of A such that the family {dϕ }ϕ∈S of irreducible representations of
A is separating for A.
Furthermore, let P ⊆ B+ a set of positive elements of B that generates B as a closed
ideal of B, i.e., for each c ∈ B+ and ε > 0, there is n ∈ N and are b1 , . . . , bn ∈ P
and d1 , . . . , dn ∈ B with kc − k d∗k bk dk k < ε.
P

We consider the set of S0 := SS,P ⊆ CP(A, B) of c.p. maps Vϕ,b (a) := ϕ(a)b,
where ϕ ∈ S and b ∈ P , and calculate the action Ψ0 : I(B) → I(A) defined by S0 :
Lemma 3.12.2(iv) shows that ΨS0 (B) = A. Let J 6= B a closed ideal of B. Then
there is b ∈ P that is not in J. If 0 6= a ∈ A+ then there is h∗ = h ∈ A with khk < π
and ϕ ∈ S with ϕ(exp(ih)a exp(−ih)) > 0. It follows that Vϕ,b (exp(ih)a exp(−ih))
is not contained in J.
Thus ΨS0 (J) = {0} for every closed ideal J 6= B and ΨS0 (B) = A.
Since S0 ⊆ CPf (A, B) ⊆ CPnuc (A, B) ⊆ CP(A, B) we get

ΨCP(A,B) = ΨCPnuc (A,B) (J) ⊆ ΨC0 (J)

for all closed ideals J of B. It says that the possibly different m.o.c. cones induce
the same action
ΨCP(A,B) = ΨCPnuc (A,B) = ΨS0 .
The special case where A = B is not nuclear shows that,
in general, the lower semi-continuous action ΨC : I(B) ∼= O(Prim(B)) → I(A)
defined by an m.o.c. cone C ⊆ CP(A, B) does not identify C itself.
In particular, different non-degenerate m.o.c. cones can define the same corre-
sponding l.s.c. action.
It implies that there is not a unique way to define KK-theory corresponding to
a given lattice action, e.g. as defined for O(X) on C(X)-algebras.

The situation of example 3.8.3 changes considerably if we consider, instead of

ΨC , the action Ψ0 : I(B ⊗max C) → I(A ⊗max C) that is defined by the point-
norm closed m.o.c.c. C ⊗max CPin (C, C) ⊆ CP(A ⊗max C, B ⊗max C), where C :=
C ∗ (F∞ ) denotes the full group C *-algebra over the free group F∞ on countably
many generators and where C ⊗max CPin (C, C) is generated by the tensor products
V ⊗max idC with V ∈ C. Then the following separation theorem of “Hahn-Banach
type” holds for all point-norm closed m.o.c. cones.
Compare Thm. 3.8.4

Theorem 3.8.4 (Separation of m.o.c. cones). Let C := C ∗ (F∞ ) and S ⊆

CP(A, B) a set of c.p. maps, and denote by Ψ0 : I(B ⊗max C) → I(A ⊗max C) the
action on the ideal that is defined by the set S 0 of c.p. maps

{ V ⊗max idC : A ⊗max C → B ⊗max C ; V ∈ S } .

 8. SEPARATION OF M.O.C. CONES BY ACTIONS 437

Then a c.p. map T : A → B is in the point-norm closed matrix operator-convex

cone C(S) generated by S, if and only if, T ⊗max idC is Ψ0 -equivariant.

It implies that we can “essentially” define point-norm-closed m.o.c. cones C ⊆

CP(A, B) with help of suitable lower semi-continuous actions Ψ of prime(B ⊗max
∗ ∗
Cmax (F2 )) on A ⊗max Cmax (F2 ). But this does not say (and is wrong) that a point-
norm-closed m.o.c. cone C is defined by the action ΨC : I(B) → I(A) defined by
C. Only for the later considered “residual nuclear” case there is really some sort of
bijective duality.

Proof. Let Ψ0 := ΨS 0 : I(B ⊗max C) → I(A ⊗max C) be the l.s.c. action of

Prim(B ⊗max C) on A ⊗max C defined by the set S 0 of c.p. maps
V ⊗max idC : A ⊗max C → B ⊗max C
with V ∈ S, cf. Lemma 3.12.2(iv). We denote by
C 0 ⊆ CP(A ⊗max C , B ⊗max C)
the m.o.c. cone
C 0 := CΨ0 = CP(Ψ0 ; A ⊗max C, B ⊗max C)
of Ψ0 -equivariant maps.
Then C 0 is a point-norm closed matrix operator convex cone by Lemma
3.12.2(vii).
The set C ⊆ CP(A, B) of all T ∈ CP(A, B) with T ⊗max idD ∈ C 0 is a point-
norm closed matrix operator convex cone that contains S, cf. Lemma 3.12.2(xi),
i.e., C(S) ⊆ C and
(V ⊗max idC )(Ψ0 (J)) ⊆ J
for all closed ideals J of B ⊗max C.
Let T ∈ C. We want to show that T ∈ C(S). Since C(S) ⊆ L(A, B) is
point-norm closed (by definition of C(S)), it suffices to show that for (contractions)
a01 , . . . , a0n ∈ A and ε > 0 there is V ∈ C(S) with maxk kV (a0k ) − T (a0k )k < ε.
Since C(S) is convex, the Hahn-Banach separation applied to the convex cone K :=
{(V (a01 ), . . . , V (a0n )) ; V ∈ C(S) } in B ⊗ `∞ (1, . . . , n) says that (T (a01 ), . . . , T (a0n ))
is in the closure of K, if and only if,
X
< fk (T (a0k )) ≥ 0
k

for every linear functionals f1 , . . . , fn ∈ B ∗ on B with < k fk (V (a0k )) ≥ 0 for all

P

V ∈ C(S).
Now let f1 , . . . , fn ∈ B ∗ such that 0 ≤ <( k fk (V (a0k ))) for all V ∈ C(S).
P

By Lemma 3.1.4 there is a cyclic representation d : B → L(H) with cyclic

vector ξ ∈ H (we can suppose that kξk = 1) and r1 , . . . , rn ∈ d(B)0 with fk (b) =
hd(b)ξ, rk∗ ξi . We find unitary operators in v1 , . . . , vm ∈ C ∗ (1, r1 , . . . , rm ) ⊆ d(B)0
such that {r1 , . . . , rn } is contained in the linear span of {v1 , . . . , vm }. Let G :=
438 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

C ∗ (v1 , . . . vm ) = C ∗ (1, r1 , . . . , rm ) ⊆ d(B)0 . It implies, that there are a1 , . . . , am ∈

A such that k a0k ⊗ rk = j aj ⊗ vj in the algebraic tensor product A G. In
P P

particular,
X X X
hd(L(aj ))ξ, vj∗ ξi = hd(L(a0k ))ξ, rk∗ ξi = fk (L(a0k ))
j k k

for every linear map L : A → B.

Consider the unital *-epimorphism λ : C → G with λ(uj ) = vj for j = 1, . . . , m
and λ(uj ) = 1 for j > m. (Here the unitaries u1 , u2 , . . . ∈ F∞ are canonical
generators of F∞ .)
It defines a C *-morphism D : B ⊗max C → L(H) with D(b ⊗ c) = d(b)λ(c) for
b ∈ B, c ∈ C. We let J ⊆ B ⊗max C denote the kernel of D. The cyclic vector
ξ ∈ H defines a positive state ρ(y) := hD(y)ξ, ξi with ρ(J) = {0} and
X X
ρ((W ⊗max id)(x)) = hd(W (aj ))ξ, vj∗ ξi = fk (W (a0k ))
j k
P
for x := j aj ⊗ uj ∈ A ⊗ C and every completely positive map W : A → B. In
particular, we get for the real part < the inequalities
<(ρ((V ⊗max id)(x)) ≥ 0 for all V ∈ C(S) .

Next we show that the property <(ρ((V ⊗max id)(x)) ≥ 0 for all V ∈ C(S)
implies that (x∗ + x) + Ψ0 (J) is positive in (A ⊗max C)/Ψ0 (J). It is the crucial point
of the proof.
Let P denote the set of positive functionals gV ∈ (A ⊗max C)∗ with gV :=
ρ((V ⊗max idC )(·)) (V ∈ C(S)). The set P of the positive linear functionals gV has
the following properties:

(α) gV (y) = 0 for all gV ∈ P (i.e., for V ∈ C(S)), if and only if, y ∈ Ψ0 (J).
(β) gV (z ∗ (·)z) is in the point-σ(A∗ , A) closure of P for every z ∈ A ⊗max C
(γ) (y ∗ +y)+Ψ0 (J) is positive in (A⊗max C)/Ψ0 (J), if and only if, <(gV (y)) ≥ 0
for all V ∈ C(S).

(α): If V ∈ C(S), y ∈ Ψ0 (J) then V ⊗max id(y) ∈ J and gV (y) = ρ(V ⊗max
id(y)) = 0, i.e., gV (Ψ0 (J)) = {0} for all gV ∈ P .
If y ∈ A ⊗max C, V ∈ CP(A, B) and ρ (b∗ ⊗ 1)(V ⊗max id)(y)(b ⊗ 1) = 0 for

all b ∈ B, then (V ⊗max id)(y) ∈ J. Indeed, ρ((b ⊗ 1)∗ z(b ⊗ 1)) = hD(z)d(b)ξ, d(b)ξi
for z ∈ B ⊗max C, d(B)ξ is dense in H and J is the kernel of D : B ⊗max C → L(H).
We have that W := b∗ V (·)b ∈ C(S) and gW (y) = ρ((b∗ ⊗ 1)(V ⊗max id)(y)(b ⊗
1) for V ∈ C(S) and b ∈ B. Thus, gV (y) = 0 for all V ∈ C(S) implies that
(V ⊗max id)(y) ∈ J for all V ∈ C(S). Conversely, gV (y) = ρ((V ⊗max id)(y)) = 0 if
(V ⊗max id)(y) ∈ J. Thus, gV (y) = 0 for all gV ∈ P , if and only if, (V ⊗max id)(y) ∈ J
for all V ∈ C(S).
For y ∈ A ⊗max C, a ∈ A, c ∈ C and V ∈ CP(A, B) holds (V ⊗max id)((a ⊗
c)∗ y(a ⊗ c)) = (1 ⊗ c)∗ W ⊗max id(y)(1 ⊗ c) where W = V (a∗ (·)a). Since V (a∗ (·)a) ∈
C(S) for V ∈ C(S), it follows that the closed linear subspace K of y ∈ A ⊗max C
 8. SEPARATION OF M.O.C. CONES BY ACTIONS 439

with (V ⊗max id)(y) ∈ J for all V ∈ C(S) (i.e., with gV (y) = 0 for all gV ∈ P )
satisfies (a⊗c)∗ K(a⊗c) ⊆ K. If we use the polar formula for bilinear maps (several
times) and the linearity of K, we get that that the closed subspace K is a two-sided
ideal of A ⊗max C. Since (V ⊗max id)(K) ⊆ J for all V ∈ S ⊆ C(S) and since
Ψ0 (J) ⊆ K, we get K = Ψ0 (J) (cf. definition of Ψ0 (J)).
(β): Since the algebraic tensor product A C is dense in A ⊗max C, it suffices
Pm
to consider z = k=1 ak ⊗ ck ∈ A C ( 19 ). Then
m
X
gV (z ∗ yz) = hD(V ⊗max id((a∗j ⊗ 1)y(ak ⊗ 1)))λ(ck )ξ, λ(cj )ξi .
k,j=1

(n) (n) (n)

Let Fn := [b1 , . . . , bm ]> ∈ Mm1 (B) such that limn d(bk )ξ = λ(ck )ξ in H for
every k = 1, . . . , m. If we let E := [a1 , . . . , am ] ∈ M1m (A) and take Vn := Fn∗ V ⊗
id` )(E ∗ (·)E Fn ∈ C(S) , then gV (z ∗ yz) = limn gVn (y) for every y ∈ A ⊗max C.

(γ): Let y = (` − k) + ih with h∗ = h, k ≥ 0, ` ≥ 0, and k` = 0, W ∈ C(S) .

Then <gW (kyk) = −gV (k 3 ) and y ∗ + y = ` − k.
By (β), we get that gW (k(·)k) is a point-wise limit of gV (V ∈ C(S)). Thus,
<gV (y) ≥ 0 for all V ∈ C(S) implies gW (k 3 ) = 0 for all W ∈ C(S), which means
k 3 ∈ Ψ0 (J) by (α). It follows that (y ∗ + y) + Ψ0 (J) = ` + Ψ0 (J) is positive in
(A ⊗max C)/Ψ0 (J).
Conversely, we have <(gV (y)) = gV (y ∗ + y) ≥ 0 if (y ∗ + y) + Ψ0 (J) is positive
in (A ⊗max C)/Ψ0 (J), because gV (Ψ0 (J)) = {0} by (α).
Above we have shown that there exist a1 , . . . , am ∈ A, a closed ideal J of
max
C, and a state ρ on B⊗max C, such that ρ and the element x := k ak ⊗uk ∈
P
B⊗
A ⊗max C satisfy the following conditions (i)–(iii):

(i) ρ((W ⊗max idC )(x)) = k fk (W (ak )) for all W ∈ CP(A, B),
P

(ii) ρ(J) = {0}, and

(iii) (x∗ + x) + Ψ0 (J) is positive in (A ⊗max C)/Ψ0 (J).

Notice that the property in (iii) follows from property (γ), because

<(gV (x)) = <(ρ((V ⊗max id)(x))) ≥ 0 for all V ∈ C(S) .

The properties (i)–(iii) imply, that T ∈ CP(A, B) with T ⊗max id(Ψ0 (J)) ⊆ J
satisfies
X
2< fk (T (a0k )) = ρ((T ⊗max idC )(x∗ + x)) ≥ 0 ,
k

because then (T ⊗ id)(x∗ + x) + J is positive in (B ⊗max C)/J. (Notice here,

max

that if E and F are C *-algebras, I / E and J / F are closed ideals, and if L : E → F

a is positive map with L(I) ⊆ J, then L(y) + J is positive in F/J if y + I is positive
in E/I.) 

19 Here m ∈ N and a ∈ A, c ∈ C are arbitrary, i.e., are not the specific ones for the
k k
definition of the element x above.
440 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

Example 3.8.5. Let A, B and C denote C *-algebras. We describe the action

Ψ0 := ΨC 0 : I(B ⊗max C) → I(A ⊗max C)

that is induced from C 0 := CPnuc (A, B) ⊗max CPin (C, C) :

More generally, we consider Ψ0 := ΨC 0 induced from C 0 = C(S 0 ), where S 0 :=
{V ⊗max idC ; V ∈ S0 } with S0 := SS,P ⊆ CP(A, B) the set SS,P of c.p. maps
Vϕ,b (a) := ϕ(a)b, where ϕ ∈ S and b ∈ P , S ⊆ A∗+ and P ⊆ B+ as in Example
3.8.3.
Let πA,C : A ⊗max C → A ⊗ C := A ⊗min C the natural epimorphism from the
maximal C *-tensor product to the minimal C *-tensor product of A with C, let IA,C
denote its kernel, and let Sϕ : A ⊗max C → C = C ⊗ C the map Sϕ := (ϕ ⊗ idC ) ◦
πA,C . We denote by K(J) the biggest closed ideal of C with B ⊗max K(J) ⊆ J.
(It can happen that K(J) = {0}.)
Let Y ⊆ C a closed linear subspace. The space F (A, Y ; A ⊗ C) ⊆ A ⊗ C
is defined as the set of z ∈ A ⊗ C such that ϕ ⊗ id(z) ∈ Y for all ϕ ∈ A∗ .
Since M ⊗N1 = Fσ (M, N1 ; M ⊗N ) (by the tensorial bi-commutation theorem for
vN-algebras) for the weakly continuous W*-algebra analog and all W*-subalgebras
N1 ⊆ N , one has F (A, Y ; A ⊗ C) = (A ⊗ C) ∩ (A∗∗ ⊗Y ∗∗ ) in A∗∗ ⊗C ∗∗ if Y is a
C *-subalgebra of C. Thus F (A, Y ; A ⊗ C) is a C *-subalgebra (respectively is a
closed ideal) of A ⊗ C if Y is a C *-subalgebra (respectively a closed ideal) of C. It
is not difficult to see that z ∈ F (A, Y ; A ⊗ C) if ϕ ⊗ id((a∗ ⊗ 1)z(a ⊗ 1)) ∈ Y for all
a ∈ A and ϕ ∈ S ⊆ A∗+ , where span(a∗ Sa) is σ(A∗ , A) dense in A∗ (which is the
case if S is separates all non-zero positive elements of A from zero).
An element x ∈ (A ⊗max C)+ is in Ψ0 (J)+ , if and only if, (Vϕ,b ⊗ id)(y) =
b ⊗ Sϕ (y) ∈ J for all y := exp(ih)x exp(−ih) (h∗ = h ∈ A C), ϕ ∈ S and b ∈ P .
Since Ψ0 (J)+ is the positive part of a closed ideal of A ⊗max C, this happens also
for (a ⊗ c)∗ x(a ⊗ c) (in place of x, and for all a ∈ A and c ∈ C). If we apply
the property that P ⊆ B+ generates B as a two-sided ideal, then we obtain – for
x ∈ (A ⊗max C)+ – that x ∈ Ψ0 (J)+ if and only if Sϕ ((a ⊗ 1)∗ x(a ⊗ 1)) ∈ K(J) for
all ϕ ∈ S, a ∈ A and , i.e., πA,C (x) ∈ F (A, K(J); A ⊗ C).
−1
We get Ψ0 (J) = πA,B (F (A, K(J); A ⊗ C)) for every closed ideal J of B ⊗max C.
We see that Ψ0 (J) does not depend from from the special choice of P ⊆ B+ and
S ⊆ A∗+ .
If I is a closed ideal of C, then K(B ⊗max I) = I, in particular, Ψ0 (J) =
Ψ0 (B ⊗max K(J)) for all closed ideals J of B ⊗max C.
It follows that T ⊗max idC is Ψ0 -residually equivariant, i.e., T ⊗max id ∈ CΨ0 =
CP(Ψ0 , A ⊗max C, B ⊗max C), if and only if, for every closed ideal I of C,

−1
T ⊗max id(πA,B (F (A, I; A ⊗ C))) ⊆ B ⊗max I .

Every linear map T : A → B of finite rank has this property, because T =

∗ ∗
P
k αk ϕk (·)bk with ϕk ∈ A+ , bk ∈ B+ , αk ∈ C. If we let C = C (F∞ ), then
 9. TEMPORARY DEMANDS FROM OTHER PLACES 441

Theorem 3.8.4 implies

C(SS,P ) = CPf (A, B) =: CPnuc (A, B) ,
for all S ⊆ A∗+ and P ⊆ B+ as in Example 3.8.3.
One can also say something in the case of C := CP(A, B) ⊗max CPin (C, C):
We obtain from the above considered nuclear case, and from the trivial inclusions
V · id(A · K(J)) ⊆ B · K(J) ⊆ J that – for all closed ideals J of B ⊗max C –,
−1
A ⊗max K(J) ⊆ ΨC (J) ⊆ πA,B (F (A, K(J); A ⊗ C)) .

The general study of actions J ∈ I(B) 7→ ΨC (J) ∈ I(A) defined by C ⊆

CP(A, B) seems to be difficult, because related slice map problems for non-exact
C *-algebras have no suitable answer.

Corollary 3.8.6. Each point-norm closed m.o.c. cone C ⊆ CP(A, B) is here-

ditary inside CP(A, B), i.e., if V, W ∈ CP(A, B) and V + W ∈ C then V, W ∈ C.

Proof. By Theorem 3.8.4 the action

Ψ0 : I(B ⊗max C ∗ (F2 )) → I(A ⊗max C ∗ (F2 ))
determines the elements of C by being Ψ0 -equivariant in the sense that T ∈ C if
and only if (T ⊗ id)(Ψ0 (J)) ⊆ J for all J ∈ I(B ⊗max C ∗ (F2 )). If we let here
T := V + W then this implies that the c.p. maps V and W satisfy this condition,
because ((V + W ) ⊗ id) = (V ⊗ id) + (W ⊗ id) and all are completely positive maps
on A ⊗max C ∗ (F2 ). Use here that x + y ∈ J+ for x ≥ 0 and y ≥ 0 implies that
x, y ∈ J+ . 

9. Temporary demands from other places

Prove and Move next blue to reasonable place !!!

Corollary 3.9.1. Let A and B stable and separable C*-algebras and let
Ψ : I(B) → I(A) a lower semi-continuous action of Prim(B) on A, such that
Ψ(0) = 0 and Ψ−1 (A) = {B} .
And let C(Ψ) ⊆ CP(A, B) the point-norm closed m.o.c. cone of “all Ψ-
equivariant c.p. maps V ∈ CP(A, B)”, i.e., of all that satisfy V (Ψ(J)) ⊆ J for
all J ∈ I(B) and V ∈ C, and suppose that C(Ψ) is separating for the action
Ψ : I(B) → I(A), – i.e., that for each J ∈ I(B),
Ψ(J) = { a ∈ A ; V (a) ∈ J, for all V ∈ C } .

If C ⊆ C(Ψ) is a point-norm closed m.o.c. cone that defines Ψ.

Then C contains all Ψ-residually nuclear c.p. maps V : A → B, i.e., all V ∈
Cb (A, B) with the property that V (Ψ(J)) ⊆ J for all J ∈ I(B), and that all the
quotient maps [V ] : A/Ψ(J) → B/J are again nuclear.
I.e., C contains the m.o.c. cone Cnuc (Ψ) of all Ψ-residually nuclear c.p. maps
V : A → B with V (Ψ(J)) ⊆ J for all J ∈ I(B) and [V ] : A/Ψ(J) → B/J is nuclear.
442 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

Proof. ??
Idea of proof:
Consider the closed ideal J of A⊗max C ∗ (F∞ ) that is defined by the intersection
of the kernels of the c.p. maps V ⊗max idC ∗ (F∞ ) into B ⊗max C ∗ (F∞ ).
?????????? 

Citation: Corollary 3.9.2

For the following Corollary 3.9.2 we refere to Remark 5.1.1(8) for the existence
– and uniqueness up to unitary equivalence – of the “infinite repeat” endomorphism
δ∞ : M(B) → M(B) for stable C *-algebras B.

Corollary 3.9.2. Let A and B stable σ-unital C*-algebras and let Hk : A →

M(B), k ∈ {1, 2}, two non-degenerate C*-morphisms, i.e., Hk (A)B is dense in B.
If A is separable and if C(H1 ) = C(H2 ) for the point-norm closed m.o.c. cones
C(Hk ) ⊆ CP(A, B), generated by the coefficient maps A 3 a 7→ b∗ Hk (a)b ∈ B, then
there exists a norm-continuous path t ∈ [0, ∞) → U (t) of unitaries in M(B) such
that, for all a ∈ A, δ∞ ◦ H1 (a)U (t) − U (t)δ∞ ◦ H2 (a) ∈ B for all t ∈ [0, ∞) and
limt→∞ kδ∞ ◦ H1 (a)U (t) − U (t)δ∞ ◦ H2 (a)k = 0 , i.e., δ∞ ◦ H1 and δ∞ ◦ H2 are
unitary homotopic in the sense of Definition 5.0.1.

Proof. ?? 

HERE ENDS eksec3-Part.1.tex

10. Non-simple algebras and the WvN-property

HERE STARTS eksec3-Part2.tex !!

We generalize some parts of the above results on simple C *-algebras to non-
simple C *-algebras. In fact, we replace some of the above considered general prop-
erties of simple p.i. algebras by more useful suitable definitions in the non-simple
situation. See, for example, the WvN-property in Definition 1.2.3.

Definition 3.10.1. Let C be a C *-subalgebra of the multiplier algebra M(B)

of a C *-algebra B and V : C → B a completely positive map. We call the map V
inner if there are m ∈ N and di ∈ B, i = 1, . . . , m, such that
m
X
V (a) = d∗i adi .
i=1

The c.p. map V is 1-step inner if we can take here m = 1, i.e., V := d∗ (·)d.
A completely positive map V : C ⊆ M (B) → B is approximately inner if V
is the point-norm limit of inner completely positive maps. The map V is 1-step
approximately inner if V can be approximated by 1-step inner c.p. maps.
Let J denote a closed ideal of a C *-algebra B, then define
M(B, J) := {T ∈ M(B) : T B + BT ⊆ J} .
 10. NON-SIMPLE ALGEBRAS AND THE WVN-PROPERTY 443

Let C ⊆ M(B) a C *-subalgebra of the multiplier algebra of a C *-algebra B.

We call a completely positive map V : C → B residually nuclear if
V (C ∩ M(B, J)) ⊆ J and [V ] : C/(C ∩ M(B, J)) → B/J is nuclear for every closed
ideal J of B.
Let h : C → M a C *-morphism into a von-Neumann algebra. The morphism
h weak-approximately majorizes a c.p. contraction W : C → M if there is a
net (Tγ ) of inner completely positive maps from M into M such that W is the
point-weak limit of the net (Tγ ◦ h).
A C *-morphism h : D → M(B) approximately majorizes a completely pos-
itive map V : D → B if there exists a net (dγ ) of elements dγ ∈ B, such that V is
the point-norm limit of the net of maps d∗γ h(·)dγ .
The morphism h majorizes a completely positive contraction V : D → B if
there is a contraction d ∈ B with V = d∗ h(·)d.
The homomorphism h : D → M(B) approximately-inner majorizes a com-
pletely positive contraction V : D → B if h weak-approximately majorizes V in the
von-Neumann algebra M := B ∗∗ ⊃ M(B).

If one uses an approximate unit of C, then one can see that one can choose the
dj and Tγ in Definition 3.10.1 such that k d∗j dj k ≤ kV k respectively kTγ k ≤ kW k,
P

cf. Lemma 3.1.8. Recall Definition 3.1.1

Lemma 3.10.2. Suppose that A ⊆ M a C*-subalgebra of a von-Neumann factor

M , and that W : A → M a weakly nuclear contraction, i.e., kW k ≤ 1 and W can
be approximated in the point-σ(M, M∗ ) topology by a net of maps

(Wγ : A → M )γ∈Γ ,

that are factorable maps in the sense of Definition 3.1.1. Then there is a net
(Vλ : M → M )λ∈L of inner completely positive contractions on M , such that W is
the point-σ(M, M∗ )-limit of the net (Vλ |A : A → M ).

Proof. There is even a net (Wγ : A → M ) of factorable maps that converges

in point-strong topology to W , because the set of factorable maps from A into M
is convex by Remark 3.1.2(o). The Lemma 3.1.8 shows that we can suppose that
the net (Wγ : A → M ) consists of contractions. Therefore, again by Lemma 3.1.8,
it suffices to show that every factorable contraction W = T ◦ S : A → M can be
approximated in point-σ(M, M∗ ) topology by inner c.p. maps on M .
If M is of type In , II1 or if M is countably decomposable and of type III,
then M is simple, and Corollary 3.2.20 gives that W can be approximated by inner
completely positive contractions (even in point norm topology).
If M is a semi-finite II∞ -factor, then the Breuer ideal D of M is simple. (Recall
that the Breuer ideal of M is generated as C *-algebra by the finite projections in
M .) If M is of type III, then the ideal D generated by the countably decomposable
projections is simple.
444 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

In both cases M is the multiplier algebra M(D) of a simple C *-algebra D

which contains an approximating unit consisting of projections.
For every projection q ∈ D, the factorable c.p. contraction Wq : a ∈ A ⊆
M(D) 7→ qW (a)q ∈ D is the point-norm limit of a net of inner completely positive
contractions by Corollary 3.2.20, i.e., Wq is approximately inner on D in the sense
of Definition 3.10.1.
Since Wq converges point-strongly to W for q % 1 strongly, also W is the
point-strong limit of a net of inner completely positive contractions. 

The following Lemma 3.10.3 shows that we can replace in the above definition
of the residually nuclear completely positive maps the set I(B) of all closed ideals
J of B by the factorial ideals I of B.
Recall that I /B is factorial if it is the kernel of a factorial representation of B.
All primitive ideals are factorial ideals, and factorial ideals are prime (with respect
to intersection in the lattice of closed ideals), as one can easily see, cf. e.g. proof of
[616, prop. 3.13.10]. Since Prim(B) has the Baire property for C *-algebras B, all
prime ideals of separable C *-algebras B are primitive if B is separable, cf. [616,
prop. 4.3.6]. (This does not hold for non-separable C *-algebras B, cf. [815].)

Lemma 3.10.3. Let C a C*-subalgebra of M(B).

(i) For every family {Jγ } of closed ideals of B holds

T T
M(B, Jγ ) = M(B, Jγ ) and
T
(ii) dist(a, M(B, Jγ )) = supγ dist(a, M(B, Jγ )) for a ∈ M(B).
(iii) If B is σ-unital then M(B, J1 ) + M(B, J2 ) = M(B, J1 + J2 ) for every
pair of closed ideals J1 and J2 of B.
(iv) Suppose that J0 ⊆ B is a closed ideal and that V : C → B is a completely
positive map, such that V (C ∩ M(B, J)) ⊆ J and
[V ]J := [πJ ◦ V ] : C/(C ∩ M(B, J)) → B/J
is nuclear for every factorial ideal J of B with J0 ⊆ J, then
[V ]J0 : C/(C ∩ M(B, J0 )) → B/J0
is nuclear.
(v) If V : C → B is residually nuclear, then for every non-degenerate factorial
*-representation ρ : B → N of B holds:
M(B, ker(ρ)) = ker M(ρ), V (C ∩ ker M(ρ)) ⊆ ker(ρ) and the restricted
map M(ρ)|C : C → N majorizes ρ ◦ V approximately with respect to the
weak topology.

Proof. The easy proofs of (i)-(iii) are left to the reader: Use an approximate
unit of B for the straight calculations.
(iv): We have V (C ∩ M(B, J0 )) ⊆ V (C ∩ M(B, J)) ⊆ J for all primitive ideals
J of B with J ⊇ J0 . The closed ideal J0 is the intersection of the primitive ideals
J that contain J0 (cf. [616, cor. 3.13.8]). Hence V (C ∩ M(B, J0 )) ⊆ J0 .
 10. NON-SIMPLE ALGEBRAS AND THE WVN-PROPERTY 445

By Remark 3.1.2(i), a map T : D → E is nuclear, if and only if,

T ⊗max id : D ⊗max C ∗ (F ) → E ⊗max C ∗ (F )

annihilates the kernel I of the natural epimorphism D ⊗max C ∗ (F ) → D ⊗ C ∗ (F ) .

It is easy to see that an irreducible representations ρ of E ⊗max C ∗ (F ) factorizes
over (E/K) ⊗max C ∗ (F ) for some factorial ideal K of E.
In particular, the family of quotient maps πK ⊗max C ∗ (F ) from E ⊗max C ∗ (F )
to (E/K) ⊗max C ∗ (F ) is separating for E ⊗max C ∗ (F ) ( 20 ). Hence, T is nuclear,
if and only if, (πK T ) ⊗max id(I) = 0 for all factorial ideals K of E, if and only if,
πK ◦ T : D → E/K is nuclear for all factorial ideals J of E.
Now apply this to D := C/(C ∩ M(B, J0 )), E := B/J0 , T := [V ]J0 , and use
that πJ/J0 ◦ [V ]J0 = [V ]J ◦ ηJ for ηJ : D → C/(C ∩ M(B, J0 )), and that K := J/J0
runs through all factorial ideals of E if J runs through all factorial ideals J ⊃ J0
of B.
(v): Let E := ρ(B) ⊆ N ⊆ L(H). Since ρ : B → L(H) is non-degenerate, the
normalizer algebra N (E) := {t ∈ L(H) ; tE + Et ⊆ E} of E in L(H) is unital and
can be identified with the multiplier algebra M(E) of E in a natural way. It is
easy to see that N (E) ⊆ E 00 ⊆ N . The representation ρ defines a *-epimorphism
from B onto E. Thus, ρ uniquely extends to a unital strictly continuous map
M(ρ) : M(B) → M(E) = N (E) ⊆ N such that M(ρ)(c)ρ(b) = ρ(cb) for c ∈ M(B)
and b ∈ B. It follows that the kernel ker(M(ρ)) of M(ρ) is M(B, J) := {c ∈
M(B) ; cB +Bc ⊆ J} for J := ker(ρ). Let Ψ(J) := C ∩M(B, J) = C ∩ker(M(ρ)).
If we let A := M(ρ)(C) ⊆ N , then M(ρ) defines a natural isomorphism γ from
C/Ψ(J) onto A with h := M(ρ)|C = γ ◦ πΨ(J) , i.e., γ(c + Ψ(J)) = M(ρ)(c) for
c∈C.
We define W : A → N by W := [ρ]J ◦ [V ]J ◦ γ −1 : A → E ⊆ N . Here [ρ]J
denotes the natural isomorphism from B/ ker(ρ) = B/J onto E = ρ(B) ⊆ N with
[ρ]J (b+J) = ρ(b) for b ∈ B, i.e., [ρ]J ◦πJ = ρ. The map [V ]J : C/Ψ(J) → B/J with
[V ]J ◦πΨ(J) = πJ ◦V is nuclear, because V is residually nuclear. Therefore W : A →
N is nuclear. By Lemma 3.10.2, there is a net of inner c.p. maps (Vλ : N → N )
with kVλ k ≤ 1 such that (Vλ |A) converges in point-σ(N, N∗ ) topology to W .
Thus the net Vλ ◦ h tends in point-σ(N, N∗ ) topology to W ◦ h = [ρ]J ◦ [V ]J ◦
πΨ(J) = ρ ◦ V . It means that h weak-approximately majorizes ρ ◦ V (in the sense
of Definition 3.10.1). 

Proposition 3.10.4. Suppose that C is a C*-subalgebra of M(B) and that

V : C → B is a completely positive contraction. Then the following are equivalent:

(i) V is approximately inner (in the sense of Def. 3.10.1).

20 Moreover, by an reduction to certain separable C *-subalgebras of E with an argument in

Chapter 12, one can show that the family of maps πK ⊗max id is separating for E ⊗max C ∗ (F ) if
K runs through all primitive ideals of E. Therefore one can replace the factorial or prime ideals
by primitive ideals in Lemma 3.10.3(iv) and in Proposition 3.10.4(ii).
446 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

(ii) For every prime ideal J of B, V (Ψ(J)) ⊆ J and [V ] : C/Ψ(J) → B/J is

approximately inner (cf. Def. 3.10.1), where Ψ(J) := C ∩ M(B, J) and
C/Ψ(J) ⊆ M(B)/M(B, J) ⊆ M(B/J) . ( 21 )
(iii) For every non-degenerate factorial representation ρ : B → N , the natural
extension h := M(ρ)|C : C → N of ρ to C weak-approximately majorizes
W := ρ ◦ V , where N := ρ(B)00 in the sense of Definition 3.10.1.

Proof. From Definition 3.10.1 we can see that (i) implies (ii) and that (ii)
implies (iii). The latter, because the kernel of each factorial representation of B is
a prime ideal J of B.
(iii)⇒(i): Suppose that (iii) holds, but that V is not approximately inner.
Let η : C ,→ M(B) denote the inclusion map. The Hahn-Banach separation
argument in the proofs of [426, lem.2,lem.3] shows that there exist c1 , . . . , cn ∈ C
and a cyclic representation d : B → L(H) with cyclic vector x ∈ H and f1 , . . . , fn ∈
d(B)0 , such that:

(*) The positive linear functional ψ ◦d1 ◦(V ⊗max idF ) is not weakly contained
in the representation k := d1 ◦ (η ⊗max idF ) of C ⊗max F on H.

Here F := C ∗ (1, f1 , . . . , fn ) ⊆ L(H), d1 denotes the natural extension of d to a

unital *-representation of M(B)⊗max F into L(H) such that d1 (c⊗f )d(b) = d(cb)f
for c ∈ M(B), b ∈ B, f ∈ F , and where ψ is the vector state on L(H) given by x.
Since there is p ∈ (C ⊗max F )+ with 0 = k(p) = d1 ((η ⊗max idF )(p)) and
0 6= d((V ⊗max idF )(p)) ∈ d(B ⊗max F ) , there is a pure state ϕ on d(B ⊗max F )
with ϕ(d((V ⊗max idF )(p))) > 0, and we obtain:

(**) There is a pure state ϕ on d(B ⊗max F ) which is weakly contained in

d : B ⊗max F → H such that the positive functional ϕ ◦ d ◦ (V ⊗max idF )
on C ⊗max F is not weakly contained in k.

There is a unique extension λ of the pure state ϕ to a pure state of d1 (M(B)⊗max F ),

because d(B⊗max F ) is an ideal of d1 (M(B)⊗max F ). Thus, λ◦k is weakly contained
in k.
But, ϕ is the cyclic vector of an irreducible representation d2 of B ⊗max F over
a Hilbert space H1 with corresponding cyclic vector y such that ϕd1 = hd2 (.)y, yi.
The irreducible representation d2 decomposes into a commuting pair of factorial
representations ρ and d3 of B and F respectively.
If d4 denotes the natural extension of d2 to M(B)⊗max F then λ(.) = hd4 (.)y, yi,
and c 7→ d4 (c ⊗ 1) is the natural extension M(ρ) of ρ to M(B).
The representation d4 is weakly contained in d1 , because d2 is weakly contained
in d and is non-degenerate. Thus d4 (η ⊗max idF ) is weakly contained in k.

21 The inclusions are defined by the natural monomorphisms.

 10. NON-SIMPLE ALGEBRAS AND THE WVN-PROPERTY 447

By the assumptions of (iii), h := M(ρ)|C weak-approximately majorizes ρV

in N := ρ(B)00 . Thus, ϕ ◦ d ◦ V ⊗max idF ) is weakly contained in d4 (η ⊗max idF ),
and, therefore, it is weakly contained in k.
We have derived a contradiction to (**), which was implied from the assumption
that V is not approximately inner. 

Suppose that h1 , h2 : A → B are C *-morphisms and that s, t ∈ M(B) are

two isometries with orthogonal ranges. Then we define for this chapter the Cuntz
addition by
(h1 ⊕ h2 )(a) := (h1 ⊕s,t h2 )(a) := sh1 (a)s∗ + th2 (a)t∗ , for a∈A
defines a C *-morphism from A into B ( 22 ). The C *-algebra generated by s and t
contains a unital copy of O∞ . Therefore one can define in a similar way the k-fold
sum h ⊕ h ⊕ · · · ⊕ h of k copies of h, k = 2, 3, . . .. (Compare Chapter 4 for more
details in the particular situation where B contains a copy of O2 unitally.)
Lemma 3.10.5. Suppose that A is a separable C*-algebra, E is a C*-algebra such
that M(E) contains isometries s, t with ortogonal ranges (i.e., s∗ t = 0), h : A → E
is a C*-morphism, and that (Vk : E → E)k is a sequence of approximately inner
c.p. contraction.

(i) If h majorizes (respectively approximately majorizes) the C*-morphism

h ⊕ h := sh(·)s∗ + th(·)t∗ ,
then there exists a C*-morphism g from A ⊗ O∞ into E, – respectively
into E∞ := `∞ (E)/c0 (E) –, with g(a ⊗ 1) = h(a) for a ∈ A.
(ii) If there is a C*-morphism g from A ⊗ O∞ into E (respectively into
E∞ ) with g(a ⊗ 1) = h(a) for a ∈ A, then there exists a sequence
dn of contractions in E such with the properties d∗n bdn+m = 0 (re-
spectively limn→∞ d∗n bdn+m = 0) for b ∈ h(A) + C1, m > 0, and
limn→∞ kd∗n h(a)dn − Vn (h(a))k = 0 for a ∈ A.

Note that in general (i) does not imply the existence of isometries s1 , s2 , . . . ∈
h(A)0 ∩ E with orthogonal ranges. But this could happen modulo the two-sided
annihilator of h(A) in E.

Proof. (i): If there is a contraction d ∈ E with d∗ h(·)d = h⊕h, then the proof
of Proposition 4.3.5(v) works and gives that there is a C *-morphism g : A⊗O∞ → E
with g(a ⊗ 1) = h(a) for a ∈ A.
Suppose that h approximately majorizes h ⊕ h. The C *-morphism h ⊕ h is a
contraction. By Lemma 3.1.8 h ⊕ h is in the point-norm closure of the set of maps
d∗ h(·)d with contractions d ∈ E. Since A is separable, we can select a sequence
(d1 , d2 , . . .) of contractions in E such that h(a) ⊕ h(a) = limn d∗n h(a)dn for a ∈ A .
Thus h⊕h = d∗∞ h(·)d∞ in E∞ for the contraction d∞ := (d1 , d2 , . . .)+c0 (E) ∈ E∞ .
22 The here given definition of h ⊕ h is a bit more general than the definition of the Cuntz
1 2
Addition in Chapters 1 and 4, where we require that moreover ss∗ + tt∗ = 1.
448 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

(ii): We fix a strictly positive contraction c ∈ A+ . Let X1 ⊆ X2 ⊆ · · · ⊆ A

an approximate filtration of A by finite-dimensional linear subspaces Xn , and let
s1 , s2 , . . . the canonical generators of O∞ . By Lemma 3.1.8, for each n ∈ N we find
(n) (n)
k(n) ∈ N and f1 , . . . , fk(n) ∈ E with
k(n)
(n) (n)
X
k (fj )∗ fj k ≤ 1
j=1

and kVn (h(a)) − Wn (h(a))k ≤ 2−n kak for a ∈ Xn , where Wn is the inner c.p. con-
traction on E given by
X (n) (n)
Wn (b) := (fj )∗ b(fj ) for b ∈ E .
j

We find p(n) ∈ N such that kc1/p(n) ac1/p(n) − ak ≤ 2−n kak for a ∈ Xn . Now let
cn := c1/p(n) , r(n) := k(1) + k(2) + · · · + k(n − 1) for n > 1 and r(1) := 0. We
(n)
define en := j g(cn ⊗ sr(n)+j )fj for n = 1, 2, . . .. Then en ∈ E∞ , e∗n bem = 0 for
P

m 6= n and e∗n ben = Wn (cn bcn ) for b ∈ h(A) + C · 1. In particular, ken k ≤ 1 and,
for a ∈ Xn ,
ke∗n h(a)en − Vn (h(a))k ≤ 3 · 2−n kak .
If g(A ⊗ O∞ ) ⊆ E, then the elements dn := en are contractions in E and have the
desired properties.
In the case where g(A ⊗ O∞ ) ⊆ E∞ , we can select from the representing
sequences of contractions for en suitable elements dn that have the in part (ii)
desired properties. 

Corollary 3.10.6.

(I) If C ⊆ M(B) and T : C → B is residually nuclear, then T is approxi-

mately inner in the sense of Definition 3.10.1.
(II) Suppose h : A → M(B) is a C*-morphism and V : A → B is a completely
positive contraction that satisfy the following conditions:
(α) h(A) commutes elementwise with a unital copy of O∞ in M(B),
(β) V (Ψ(J)) ⊆ J, and
(γ) [V ] : A/Ψ(J) → B/J is nuclear for every factorial ideal J ⊆ B, where
Ψ(J) := h−1 (h(A) ∩ M(B, J)).
Then h approximately majorizes V in the sense of Definition 3.10.1.

Proof. (I) follows from Proposition 3.10.4(iii) and Lemma 3.10.3(v).

(II): Let C := h(A). For J = 0, JA is the kernel of h. Thus there is a
completely positive map T : C → B such that T h = V . Then (β) and (γ) say that
T is residually nuclear. By (I), T is approximately inner. But since C commutes
with a copy of O∞ , we get that T is moreover one-step approximately inner. This
means that h approximately majorizes V . 

By Remark 3.1.2(iv), Corollary 3.10.6 implies:

 10. NON-SIMPLE ALGEBRAS AND THE WVN-PROPERTY 449

Corollary 3.10.7. Suppose that C ⊆ M(B), C is exact, V : C → B is

nuclear, and V (C ∩ M(B, J)) ⊆ J for every closed ideal J of B, then V is approx-
imately inner in the sense of Definition 3.10.1.

Corollary 3.10.8. Suppose that M(B) contains two isometries with orthog-
onal ranges, A is separable, h : A → B is a C*-morphism that approximately ma-
jorizes h ⊕ h and that T : A → B is a completely positive contraction.
Assume that T is nuclear and A is exact, or that [T ] : A/J → B/JT is nuclear
for every closed ideal J of A, where JT denotes the closed ideal of B which is
generated by T (J).
Then the following are equivalent:

(i) h approximately majorizes T .

(ii) T (J) is contained in the closure of Bh(J)B for every closed ideal J of A.
(iii) T (h−1 (h(A) ∩ K)) ⊆ K, i.e., a 7→ V (a) + K annihilates the kernel of
a 7→ h(a) + K, for every primitive ideal K of B.

Proof. The implication (i)⇒(ii) is trivial. (ii)⇒(iii), because, with J :=

−1
h (h(A) ∩ K), h(J) ⊆ K and, therefore, T (J) ⊆ K.
(iii) ⇒ (i): By induction, h approximately majorizes the k-fold sum of h for
k = 1, 2, . . .. Therefore, and since A is separable, it suffices to show that T is in the
point norm closure of the convex set of maps S ◦ h, where the S runs through the
inner complete contractions on B. For the kernel J of h we have T (J) = 0, because
J ⊆ h−1 (h(A) ∩ K and, therefore, T (J) ⊆ K for every primitive ideal K of B.
Let C := h(A) and let V be the unique completely positive contraction with
V h = T . Then, for every primitive ideal K of B, J := C ∩ K is a closed ideal of C
and V (J) is contained in K. If K0 is any closed ideal, then V (C ∩K0 ) ⊆ V (C ∩K) ⊆
K for every primitive ideal K of B with K0 ⊆ K. Thus V (C ∩ K0 ) ⊆ K0 .
If A is exact and T is nuclear, then C is exact and V is nuclear by Remark
3.1.2(iv). Thus V is approximately inner by Corollary 3.10.7.
If A is not exact, but [T ] : A/J → B/JT is nuclear for every closed ideal J
of A, then with J := h−1 (C ∩ K) we get V (C ∩ K) ⊆ T (J) ⊆ K and, therefore,
T (J) ⊆ JT ⊆ K, [T ] : A/J → B/K is nuclear. Under the natural isomorphism
A/J ∼= C/(C ∩ K), [V ] : C ∩ K → B/K is the same as [T ]. Thus V is residually
nuclear, and is approximately inner by Corollary 3.10.6(I).
Since V is a contraction, it is then, moreover, in the point-norm closure of the
convex set of restrictions S|C of inner complete contractions on B. 

Let us recall here the Definition 1.2.3 of : ??????

Remark 3.10.9.
(i) We shall see in Chapter 5 that σ-unital C *-algebras B with the WvN–property
satisfy a generalized Weyl–von-Neumann–Voiculescu type theorem for weakly resid-
ually nuclear completely positive contractions.
450 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

(ii) Proposition 3.2.13 says that simple purely infinite C *-algebras have the WvN–
property.
(iii) C *-algebras B with the WvN-property are p.i. in the sense of Definition 1.2.1:
Let a ∈ B+ . We consider C = C ∗ (diag(a, 0)) and the *-monomorphism V
from C into M2 (B) which sends diag(a, 0) to diag(a, a). V is residually nuclear.
Therefore, by the WvN-property, for every ε > 0, there exists a contraction d ∈
M2 (B) with d∗ diag(a, 0)d = (diag(a, a) − ε)+ . This implies that B is p.i.
(iv) It follows from joint work with M. Rørdam and from [443] that all strongly
p.i. algebras (in the sense of Definition 1.2.2) have the WvN–property, and that
in several particular cases p.i. algebras are strongly p.i., see the Remarks 2.15.12,
3.11.1, 3.11.6 and 12.2.9.
(v) Residually nuclear separation passes to hereditary C *-subalgebras.
(vi) It is easy to see that the non-commutative Michael selection principle for B ⊗K
in Conjecture 12.3.6 implies residually nuclear separation for B in the sense of
Definition 1.2.3, cf. Remarks 12.3.7, 12.3.8.
(Every simple stable C *-algebra trivially satisfies the non-commutative Michael
selection principle.)
(vii) It follows from [463] and [443] that C *-algebras with the WvN–property and
residually nuclear separation are strongly p.i. in the sense of Definition 1.2.2.
(viii) Since all separable C *-algebras have Abelian factorization, each C *-algebra
has residually nuclear separation.
Since every separable C *-algebra has abelian factorization,
we get that they all also have residually nuclear separation
??

Corollary 3.10.10. Suppose that B has the WvN–property.

(i) If C is an exact subalgebra of M(B) such that that V : C → B is nuclear

and V (C ∩ M(B, J)) ⊆ J for every primitive ideal of B, then V is 1-step
approximately inner, i.e. if V is approximately majorized by idC in the
sense of Definition 3.10.1.
(ii) If A is exact, h : A → B is a nuclear C*-morphism, and M(B) contains
two isometries s, t with orthogonal ranges, then h approximately majorizes
h ⊕s,t h .

Proof. (i) follows from Corollary 3.10.7 and Definition 1.2.3.

(ii): Let C := h(A). C is exact and εC : C ,→ B is nuclear by Remark 3.1.2(iv).
Since h and h ⊕ h have the same kernel, there is a C *-morphism V : C → B such
that V h = h⊕h, in fact V = εC ⊕εC , which is nuclear with εC . Thus (i) applies. 

Corollary 3.10.11. Let c ∈ (O2 )+ be an element with Spec(c) = [0, 1]. If

a, b ∈ A+ and % ≥ 0 satisfy kb + Jk ≤ ka + Jk + % for all primitive ideals J
 10. NON-SIMPLE ALGEBRAS AND THE WVN-PROPERTY 451

of A, then, for every δ > 0 there exist an isometry s ∈ M(A ⊗ O2 ) such that
ks∗ (a ⊗ c)k s − ((b ⊗ c) − %)k+ k < δ for k = 1, 2.
Thus ks∗ (a ⊗ c)k s − (b ⊗ c)k k < δ + % max(1, 2kbk) for k = 1, 2.

Proof. The assumptions imply that kb ⊗ ck = kbk and, for every closed ideal
I of A,

kb + Ik = sup{ kb + Jk : J ∈ Prim(A), I ⊆ J } ≤ ka + Ik + % .

In particular, kdk ≤ kak where d := ((b ⊗ c) − %)+ .

Let h0 : C([0, 1]) → O2 denote the unital *-monomorphism with h0 (f ) := f (c)
for f ∈ C([0, 1]). By Corollary 3.10.10, h0 approximately majorizes h0 ⊕h0 , because
C([0, 1]) is nuclear, O2 is a simple p.i. algebra and, therefore, by Proposition 3.2.13,
O2 has the WvN–property.
Let C := C([0, kak]), and γ : C → C ⊗ C([0, 1]) the unital *-monomorphism
given by γ(f )(x, y) := f (xy). We define C *-morphisms h and k from C into the
unitization of A ⊗ O2 , and h1 from C into the unitization of A, by h(f ) := f (a ⊗ c),
k(f ) := f (d) and h1 (f ) := f (a).
Then h1 ⊗ h0 and, therefore, h = (h1 ⊗ h0 )γ approximately majorizes their
two-fold sums (h1 ⊗ h0 ) ⊕ (h1 ⊕ h0 ) = h1 ⊗ (h0 ⊕ h0 ) and h ⊕ h.
Note that, for every closed ideal J of A and e ∈ A+ , the spectrum of πJ (e) ⊗ c
is the closed interval [0, ke + Jk]. By Proposition B.4.2, for every closed ideal K of
A ⊗ O2 there exist a closed ideal J of A with K = J ⊗ O2 .
Thus, for every factorial ideal K = J ⊗ O2 of A ⊗ O2 , the spectrum of πK (d)
is [0, (kb + Jk − %)+ ], which is contained in the spectrum [0, ka + Jk] of πK (a ⊗ c).
Hence, by Corollary 3.10.8, h approximately majorizes k. This implies easily
the stated result. 
P ∗
Corollary 3.10.12. Let κ(a) denote the convex set of elements c = dk adk
P ∗
in A+ , where the dk are in the unitization of A with dk dk = 1 in M(A), k =
1, . . . , n, n ∈ N.
Then, for a, b ∈ A+ ,

dist(b, κ(a)) = max(0, sup{kb + Jk − ka + Jk : J ∈ Prim(A)}) .

Proof. Let δ > 0.

P ∗
There exists V = dk ( · )dk , such that kb − V (a)k < δ + dist(b, κ(a)).
For every closed ideal J, kV (a)+Jk ≤ ka+Jk and kb−V (a)+Jk ≤ kb−V (a)k.
Thus, kb + Jk − ka + Jk ≤ δ + dist(b, κ(a)).
The other direction, i.e.,
dist(b, κ(a)) ≤ % for every % > 0 with kb + Jk − ka + Jk ≤ % for all primitive ideals J
of A, follows from Corollary 3.10.11 by the method in the proof of Corollary 3.2.20,
452 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

because there exists by Proposition 3.2.13 a sequence of isometries dn ∈ O2 with

d∗n cdn → 1. 

Remark 3.10.13. There are strictly continuous versions of Corollaries 3.10.11

and 3.10.12 for a, b ∈ M(A)+ and weakly continuous versions for a, b in the positive
part of a von Neumann algebra N . Of course, one has to modify all what is needed,
beginning with Definition 3.10.1, but in a very straightforward way. The outcome
is, e.g., the following modifications of Corollary 3.10.12:
Let κ(a)st and κ(a)s denote the strict closure of κ(a) in M(A) and the strong
closure of κ(a) in N , respectively. Then, for a, b ∈ M(A)+ ,


dist(b, κ(a)st ) = max 0, sup{kb + M(A, J)k − ka + M(A, J)k ; J ∈ Prim(A)} .

Let PZ denote the set of projections in the center of N . Then, for a, b ∈ N+ ,

dist(b, κ(a)s ) = max(0, sup{kbpk − kapk ; p ∈ PZ}).

Corollary 3.10.14. Suppose that B and D ⊆ M(B) are σ-unital and stable,
where D has the WvN–property and DB is dense in B.
Let A ⊆ M(D) ⊆ M(B) a separable C*-subalgebra. Suppose that there exists
a *-monomorphism h : A ,→ M(D) such that, for every closed ideal J of D,

h(A ∩ M(D, J)) = h(A) ∩ M(D, J)

and, for every d ∈ D, the c.p. map a ∈ A 7→ d∗ h(a)d is residually nuclear.

with respect to which action? l.s.c. from I(B) to I(D) and then from I(D) to
A???
Furthermore, let V : A → B a completely positive contraction such that

(i) V (A ∩ M(D, J)) is contained in the closed linear span of BJB, and
(ii) [V ] : A/(A ∩ M(D, J)) → B/JB is nuclear for every closed ideal J of D.

Then V is 1-step approximately inner in the sense of Definition 3.10.1, i.e., there
exists a sequence of contractions bn ∈ B such that V (a) = lim b∗n abn for every
a ∈ A.

Proof. First we show, that V is approximately majorized by an infinite repeat

of h, – cf. Remark 5.1.1(8) and Lemma 5.1.2(i,ii) in Chapter 5 concerning infinite
repeats δ∞ ◦ h .
Let s1 , s2 , . . . be a sequence of isometries in M(D) such that s∗n sm = δnm and
sn s∗n strictly converges to the unit element of M(D). We define an infinite
P

sn as∗n .
P
repeat δ∞ on M(B) by δ∞ (a) :=
Since δ∞ (M(D, J)) = δ∞ (M(D)) ∩ M(D, J) for every closed ideal J of D,
the infinite repeat δ∞ ◦ h of h satisfies that (δ∞ ◦ h)(A ∩ M(D, J)) coincides with
((δ∞ ◦ h)(A)) ∩ M(D, J) h : A ,→ M(D) for every closed ideal J of D,
and that δ∞ ◦ h is again residually nuclear ???? For which action of what
lattice?
 10. NON-SIMPLE ALGEBRAS AND THE WVN-PROPERTY 453

Therefore, we can assume that the image C := h(A) of A in M(D) commutes

elementwise with a unital copy of O∞ in M(D). (It is not the copy that is defined
by the isometries s1 , s2 , . . ..)
Let us denote by ΨD,B
down (J) the closed linear span of BJB in B for ideals J of
D.
We also use the notations Ψup up
D,A (J) := A ∩ M(D, J), and, similarly ΨD,C (J),
for closed ideals J ⊆ D, and Ψup up
B,C (I) := C ∩ M(B, I), ΨB,D (I), for closed ideals
I ⊆ B, as introduced at the end of Chapter 1.
Let IA := h−1 (Ψup
B,C (I)) and JA := h
−1
(Ψup
D,C (J)).

Our assumptions say that, in particular, Ψup up

D,A (J) =: JA , V maps ΨD,A (J)
D,B D,B
into Ψdown (J) and that [V ] : A/JA → B/Ψdown (J) is nuclear.
Suppose that I is an ideal of B, then

Ψup up up
B,C (I) ⊆ ΨD,C (ΨB,D (I)) ,

because t ∈ M(D) and tB ⊆ I imply tDB ⊆ I and, therefore, tD ⊆ M(B, I) ∩ D.

Further ΨD,B up
down (ΨB,D (I)) ⊆ I, because B(D ∩ M(B, I))B ⊆ I.

Thus we get h−1 (Ψup D,B

B,C (I)) ⊆ JA and Ψdown (J) ⊆ I for J := D ∩ M(B, I).

Therefore, the conditions (α)-(γ) of Corollary 3.10.6(ii) are satisfied for A, h

and V . Thus, there exists a sequence of contractions dn in B such that d∗n h(.)dn
converges in point-norm to V .
Since DB is dense in B, we find a net en of contractions in D+ such that
en dn − dn converges to zero.
But, by assumption, en h(·)en is residually nuclear on A ⊆ M(D), and D has
the WvN–property. Thus, there is a sequence fn of contractions in D such that
fn∗ afn − en h(a)en tends to zero for every a ∈ A.
Let bn := fn dn , then V (a) = lim b∗n abn for every a ∈ A. The bn are contractions
in B. 

Proposition 3.10.15. For every σ-unital C*-algebra B, F := B ⊗O∞ ⊗O∞ ⊗

. . . and Cb (R+ , F )/ C0 (R+ , F ) have both the WvN–property.

Proof. Let D be a stable hereditary σ-unital C *-subalgebra of F ⊗ K (respec-

tively of E ⊗ K).
It is not difficult to see that such D have the property that we can find, for given
η > 0, f1 , . . . , fn ∈ D and m ∈ N, isometries s1 , . . . , sm in the multiplier algebra
M(D) of D such that k[si , fj ]k < η and s∗k si = δi,k for j = 1, . . . , n, i, k = 1, . . . , m.
To see this, e.g. for D ⊆ E ⊗ K, one can at first show that the unital *-
monomorphisms b 7→ 1 ⊗ b and b 7→ 1 ⊗ b from O∞ to O∞ ⊗ O∞ are homotopic
in O∞ ⊗ O∞ . This can be seen by examination of finite subsets t1 , . . . , tn of the
generators of O∞ by Theorem E, [172] and Künneth theorem applied to K∗ (O∞ ⊗
O∞ ).
454 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

This can be used to show that there is a unital copy of O∞ in the multiplier
algebra of (F ⊗ K)ω (respectively of Q(R+ , F ⊗ K)) that commutes with D. Since
D is hereditary, we get such a copy of O∞ also in M(Dω ).
Let C ⊆ M(D) and suppose that V : C → D is residually nuclear. By Corollary
3.10.6, for 1 > ε > 0 and c1 , . . . , cn ∈ C, there exist d1 , . . . , dm ∈ D such that
kV (cj ) − d∗i cj di k < ε/3 for j = 1, . . . , n .
P

Let γ := 1 + k d∗i di k + max{kcj k}2 and η := ε/(3m2 γ).

P

There is a contraction e ∈ D+ with kedi − di k < η for i = 1, . . . , m. We

find isometries s1 , . . . , sm in the multiplier algebra of D such that k[si , ecj e]k < η
and s∗k si = δi,k for j = 1, . . . , n, i, k = 1, . . . , m . Then kV (cj ) − d∗ cj dk < ε for
P
d := e si di , j = 1, . . . , n. 

11. More on strongly purely infinite algebras

The following remarks give an outline of some of the results of joint work with
M. Rørdam and E. Blanchard on purely infinite algebras. See also end of Chapter
2.

Remark 3.11.1. Let (An ) be a sequence of unital C *-algebras that contain a

copy of O∞ unitally.
Let F := B ⊗ A1 ⊗ A2 ⊗ . . ., then the same arguments as in the proof of
Proposition 3.10.15 show that F , Fω and Q(R+ , F ) have the WvN–property. An
approximately commuting copy of O∞ can also be used to show that F , Fω and
Q(R+ , F ) are strongly purely infinite.
Since in the middle of Chapter 10 we finish the proof of Theorem B, we can
then use Corollary H and see from Proposition 3.10.15 and its proof that B ⊗ O∞
always has the WvN–property and is strongly purely infinite. This yields a proof
that B ⊗ C has the WvN–property and is strongly purely infinite if C is a simple
purely infinite separable nuclear C *-algebra.
Conversely, B ∼
= B ⊗ O∞ if B has the WvN–property and is separable, stable
and nuclear, by Corollary 10.3.8.
It is an open question whether or not (non-simple) purely infinite separable
stable nuclear C *-algebras B tensorial absorb O∞ . (It is the case if and only if B
is s.p.i.).
Lemma 3.2.22 shows, that C(X, D) has the WvN–property for compact X and
simple purely infinite D.

Remark 3.11.2. A stable C*-algebra D has the WvN–property, if for every

separable subalgebra A of the ultrapower Dω of D, every approximately inner com-
pletely positive contraction V : A → Dω into a commutative subalgebra C ∗ (V (A))
of Dω is 1-step approximately inner.
 11. MORE ON STRONGLY PURELY INFINITE ALGEBRAS 455

Proof. The following observations (1)–(6) together prove the sufficiency

criterium for the WvN-property:

(1) The assumption immediately implies that Dω is purely infinite, by the

argument in Definition 1.2.3.
(2) This implies that D is purely infinite.
(3) For every separable C *-subalgebra A of the ultrapower Dω of a p.i. algebra
D and every separable commutative C *-subalgebra C of A there is a completely
positive residually nuclear contraction V from A into a commutative C *-subalgebra
C ∗ (V (A)) of Dω , such that V |C = id |C. (see [443], compare also Lemma 12.1.6
for the first step in this direction, and use the semigroup structure argument of the
proof of Lemma 12.1.6).
(4) The set CA of approximately inner completely positive contractions V : A →
Dω with commutative C ∗ (V (A)) is closed under Cuntz addition (see Chapter 4 or
the remark before Corollary 3.10.8), and, therefore, by the observation (3) and the
below given generalization of Lemma 3.2.7, CA generates a dense subset of the
convex set X of the residually nuclear completely positive contractions from A into
Dω , where one gets X if one applies to the set CA only Cuntz addition and the
operations in part (OC2) of Definition 3.2.2.
(5) The set of 1-step approximately inner completely positive maps is stable
under point-norm limits and under the operations in part (OC2) of Definition 3.2.2,
and, therefore, the assumptions and observation (4) together imply that all resid-
ually nuclear completely positive maps from A into Dω are 1-step approximately
inner.
(6) Stable D (and moreover Dω ) has the WvN–property if, for every separable
C *-subalgebra A of Dω , every residually nuclear completely positive contraction V
from A into Dω is 1-step approximately inner:
if C is a separable C *-subalgebra of M(D) and V : C → D is a residually nu-
clear completely positive contraction, then V is approximately inner by Corollary
3.10.6(I). If we use a suitable, approximately with C commuting, approximate unit
en of D, we get a contraction e ∈ (Dω )+ with ec = ce for c ∈ C and an approxi-
mately inner completely positive contraction W : A → Dω from the C *-subalgebra
A of Dω , which is generated by {en c : c ∈ C, n ∈ N}, such that V (c) = W (an c) for
c ∈ C and n ∈ N. The residual nuclearity of V then implies the residual nuclear-
ity of W , see Proposition 2.12.8 for a step in this direction. If (dk ) is a sequence
of contractions in Dω such that d∗k (.)dk converges to W in point norm, then one
can pick up from the representing sequences (en dkn ) of edk elements fm such that
∗
fm cfm converges to V (c) for every c ∈ C. 

A generalization of the sixth observation shows that Dω has the WvN–property

if every approximately inner map is
locally 1-step approximately inner.
456 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

Since Dω has residually nuclear separation in the sense of Definition 1.2.3, it

follows that Dω is strongly purely infinite.
Therefore, D is strongly purely infinite if, for every separable C *-subalgebra
A of Dω , every residually nuclear contraction V : A → Dω is 1-step approximately
inner.

We replace the assumption of Lemma 3.2.7 and its conclusion in the following
manner to get a generalization of Lemma 3.2.7 :
We suppose that A ⊆ M(B) and replace the assumption of Lemma 3.2.7 by the
following strict non-degeneracy criteria where C ⊆ CP(A, B) denotes a point-norm
closed m.o.c. cone:
Property (OC-snd):
Let A ⊆ M(B). Consider the following property of a subset C ⊆ CP(A, B):
For every a ∈ A+ ⊆ M(B)+ , and every b ∈ B+ with the property that

kb + Jk ≤ ka + M(B, J)k

holds for every closed ideal J of B, there exists for every ε > 0 a c.p. contraction
T ∈ C, that satisfies kT (a) − bk < ε and has the property that T (A ∩ M(B, J)) ⊆ J
for every closed ideal J of B.
If A · B is dense in B and C contains all inner c.p. maps a 7→ c∗ ac with c ∈ B,
then condition (OC-snd) is satisfied:
Indeed, if kb+Jk ≤ ka+M(B, J)k for all J ∈ I(B), then b ∈ J0 := span(BaB),
because a ∈ M(B, J0 ).
Thus, for given ε ∈ (0, 1), let γ := γ(ε) := (2kbk1/2 + 1)−1 ε there are
c1 , . . . , cn , d1 , . . . , dn ∈ B with kb1/2 − ek < γ for e := 1/2
P
k dk a ck . This im-
∗
plies kb − e ek < ε.
Notice that

e = [d1 , . . . , dn ] · diag(a, . . . , a)1/2 · [c1 , . . . , cn ]> .

Therefore,
X X
e∗ e ≤ k dk d∗k k · c∗k ack .
k k
∗
P
It follows that there is f ∈ B such that V (a) := k (ck f ) a(ck f ) satisfies (b−ε)+ =
V (a).
The question is, if property (OC-snd) also implies that C contains all c.p. maps
V (a) := c∗ ac with c ∈ B. This seems to be the case if A ⊆ M(B), A is exact and
B is weakly injective.
But it would be sufficient that B has residual nuclear separation and
CPnuc (A, B) ⊆ C.
We would get that C ⊆ CP(A, B) satisfies property (OC-snd) in case A ⊆
M(B), if and only if, CPnuc (A, B) ⊆ C.
 11. MORE ON STRONGLY PURELY INFINITE ALGEBRAS 457

Modified version of Lemma 3.2.7:

If A ⊆ M(B) satisfies (OC-snd), and if ΨA (J) = A∩M(B, J) denotes the l.s.c. ac-
tion of Prim(B) on A, then every Ψ-residually nuclear contraction V : A → B can
be approximated in point norm by contractions in C := CPin (A, B).

The proof is similar to the proof of Lemma 3.2.7. But one has to use the
arguments in the proof of Lemma 3.10.3 and of Proposition 3.10.4. In fact, it is
easier to use Stinespring-Kasparov dilation directly. Then, by (OC1), (OC2) and
(OC-snd), for an residually nuclear contraction V : A → B, one finds a σ-unital
hereditary subalgebra E ⊆ B and a *-monomorphism h : A ,→ M(E ⊗ K) such
that A ∪ V (A) ⊆ E, and, for every contraction d ∈ E ⊗ K, (1 ⊗ p11 )d∗ h(.)d(1 ⊗ p11 )
is of form T (·) ⊗ p11 for some
approximately inner???
T
in the point norm closure of K =?????,
and h(A∩J) = h(A)∩M(E ⊗K, J ⊗K) for every closed ideal J of E. Moreover,
we can replace h by its infinite repeat δ∞ ◦ h. Thus, we may assume that h(A)0 ∩
M(E ⊗ K) contains a copy of O2 unitally.
Now we are in position to apply Corollary 3.10.6(II) on A, h and a ∈ A 7→
V (a) ⊗ p11 to get a proof of the above given generalization of Lemma 3.2.7.

The general idea of the application of the above modified Lemma 3.2.7 is the
following observation:
Suppose that, for every separable C *-subalgebra A of the ultrapower B = Dω
of a stable σ-unital C *-algebra D, there exists a set S of 1-step approximately inner
completely positive contractions V : A → B into commutative C *-subalgebras of
B, such that S is closed under Cuntz addition, and for every a ∈ A+ there exist V
in S and d ∈ B such that ka − d∗ V (a)dk < 1.
Then, by the modified version of Lemma 3.2.7, every residually nuclear T : A →
B is 1-step approximately inner.
As we have seen above, this implies that D and its ultrapower Dω have the
WvN–property and are strongly p.i.
The above given third, fourth and fifth observations also show that the modified
version of Lemma 3.2.7 yields the following result:
Suppose that A is a separable C*-subalgebra of Dω , and that for every residually
nuclear contraction V : A → Dω with commutative image C ∗ (V (A)), there is a
sequence of contractions dn ∈ Dω such that, for a ∈ A+ and k > 0,

lim d∗n adn = V (a), and

lim d∗n adn+k = 0.

Then such a sequence (dn ) exists also for every residually nuclear contractions
V : A → Dω with non-commutative image V (A).
458 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

Note that, with help of a countable approximate unit of A, we can modify the
sequence dn such that it also satisfies lim d∗n dn+k = 0 for k > 0.

Proof. Consider the set IA of all contractions V : A → Dω such that there

exist a sequence dn with the above properties.
It is easy to see, that the operations (OC2) in Definition 3.2.2 let IA invariant.
The assumption says that CA of the above fourth observation is contained in
IA .
By the third and fourth observation, the smallest subset of IA , that contains
CA and is invariant under the operations (OC2) in Definition 3.2.2, satisfies the
assumptions of the above given modified version of Lemma 3.2.7.
It is not very hard to check that IA is closed under point norm limits. Thus, by
the modified version of Lemma 3.2.7, every residually nuclear contraction V from
A into Dω is in IA . 

FIND END OF NEXT REMARK !!!

Remark 3.11.3. We list here some later needed properties of strongly purely
infinite C *–algebras. (See [463] for details of the here outlined proofs.)
In the following let [x, y] denote the commutator xy − yx.
Suppose that A is strongly purely infinite.
Let b, a1 , . . . , an ∈ A+ commuting elements, c ∈ A+ , F := [fjk ] ∈ Mm (A)+ a
positive m × m matrix, and let ε > 0.

(i) Aω and A ⊗ K are strongly purely infinite.

Every quotient and every hereditary C *-subalgebra of A is strongly
purely infinite.
(ii) There exist contractions d1 , . . . , dm ∈ A, such that

k D∗ F D − diag(f11 , . . . , fmm ) k < ε ,

where D := diag(d1 , . . . , dm ).
(iii) There exist contractions v1 , v2 ∈ A with
k[vi , ak ]k < ε and kvi∗ bvj − δi,j bk < ε for i, j ∈ {1, 2}, k = 1, . . . , n.
(iv) There exist contractions s1 , s2 ∈ A with
ks∗i ak si − ak k < ε, k[si , ak ]k < ε, ks∗1 s2 k < ε, ks∗1 cs2 k < ε, and
k[s∗i csi , ak ]k < ε, for i = 1, 2, k = 1, . . . , n.

Property (ii) is an inductive repeat of the defining property of ‘strongly p.i.’ in

Definition 1.2.2. (ii) implies (iii) and (iv). The proof of this implication is not trivial.
It uses appropriate decompositions of the unit of the unitization of C ∗ (a1 , . . . , an , b)
respectively of C ∗ (a1 , . . . , an ), to translate (iii) and (iv) to problems, which are
solvable by property (ii). The proofs can be found in [463].
It follows from [463] that p.i. algebras with (iii) and (iv) have the WvN-property
of Definition 1.2.3.
 11. MORE ON STRONGLY PURELY INFINITE ALGEBRAS 459

In fact we show that properties (iii) and (iv), applied to the strongly p.i. algebra
Aω , yield the following stronger results (v)-(vii).
Suppose that A is strongly p.i., that C is a separable commutative C *-
subalgebra of Aω , that B is a separable C *-subalgebra of Aω , and that V is a
residually nuclear contraction from B into C. Then:

(v) There is a *-monomorphism h from C ⊗O∞ into Aω with h((.)⊗1) = id |C,

(vi) For pairwise commuting a1 , . . . , an ∈ (Aω )+ , arbitrary y1 , . . . , ym ∈ Aω
and ε > 0, there is a contraction d ∈ Aω , such that kd∗ ai d − ai k < ε
for 1 ≤ i ≤ n and that {d∗ a1 d, . . . , d∗ an d, d∗ y1 d, . . . , d∗ ym d} generates a
commutative C *-subalgebra of Aω .
(vii) There is a sequence of contractions dn ∈ Aω with lim d∗n bdn = V (b) and
lim d∗n bdn+k = {0} for b ∈ B, k > 0.

Here (vii) follows from (v) and (vi).

Now we can apply the ideas of Remark 3.11.2:
In [463] it is shown that for a C *-algebra A the following properties (1)–(4)
and (8) are equivalent. The equivalence of (1) and (5)–(7) follows from [443] and
the results of Chapter 7.

(1) A is strongly purely infinite in the sense of Definition 1.2.2.

(2) The ultrapower Aω of A is strongly purely infinite.
(3) A admits no non-zero character, and, for every separable C*-subalgebra
B of Aω and every approximately inner completely positive contraction
V : B → Aω into a commutative C*-subalgebra C of Aω there is a con-
traction d ∈ Aω with V (b) = d∗ bd for b ∈ B.
(4) For every residually nuclear contraction V from a separable C*-subalgebra
C of Aω into Aω there is a sequence of contractions dn ∈ Aω with d∗n cdn =
V (c) for c ∈ C, d∗m dn = 0, and d∗m Cdn = {0} for m 6= n, m, n = 1, 2, . . . .
(5) The asymptotic corona Q(R+ , A) of A is strongly purely infinite.
(6) Q(R+ , A) has the WvN-property of Definition 1.2.3.
(7) A admits no non-zero character, and for every separable C*-subalgebra
B of Q(R+ , A) the following property holds:
Let t ∈ R+ 7→ V (t) a strongly continuous map from R+ into the
approximately inner completely positive contractions from A into A and
let T the completely positive contraction from Q(R+ , A) into Q(R+ , A)
given by
T (a + C0 (R+ , A)) := V (a) + C0 (R+ , A) ,
where a ∈ Cb (R+ , A) and V (a)(t) := V (t)(a(t)) .
The property of A in question is:
If T |B is residually nuclear, then there exists a contraction d ∈ Q(R+ , A)
such that T (b) = d∗ bd for b ∈ B.
(8) For every a, b ∈ A+ and ε > 0 there exist contractions s, t ∈ A such that
ka2 − s∗ a2 sk, kb2 − t∗ b2 tk and ks∗ abtk are all less than ε.
460 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

The property in (8) implies that the class of strongly purely infinite algebras is
closed under inductive limits.

Remark 3.11.4. Propositions 3.2.13 and 3.2.15 are used to prove a Weyl-von
Neumann–Voiculescu type theorem in Chapter 5 and its asymptotic version in
Chapter 7.
The reader can observe that our proofs of Propositions 3.2.13 and 3.2.15 works
also in the following more general and non-simple case:
Suppose that D is σ-unital, stable and strongly purely infinite, and that for
every separable C *-subalgebra A of D there exists a *-monomorphism h : A ,→
M(D), such that:

(1) h(A ∩ J) = h(A) ∩ M(D, J), and

(2) [h]J : A/(A ∩ J) → M(D/J) is weakly nuclear for every closed ideal J of
D.
(3) The closures of DKD and Dh(K)D in D coincide for every closed ideal
K of A.
(4) For every sequence Ωn of compact subsets of A, and for every approxi-
mately commutative unit of D for h(A) with en en+1 = en , there exists a
sequence fn of contractions in D such that kfn∗ afn − en h(a)en k < 2−n ,
kfn∗ afn+k k < 2−n and kfn∗ fn+k k < 2−n for a ∈ Ωn , k > 0.

From Corollary 3.10.6(I) we get, that if we assume (1) and (2), then (3) is
equivalent to:
(3’) Dh(A)D is dense in DAD.
It follows from a result with M. Rørdam [463] that the property (4) in Remark
3.11.3 implies, that a σ-unital stable purely infinite C *-algebra D has residually
nuclear separation in the sense of definition 1.2.3, if and only if, for every separa-
ble C *-subalgebra A ⊆ D, there exists a *-monomorphism h : A ,→ M(D) with
properties (1)-(4) of Remark 3.11.4.
The modifications in the assumptions of Propositions 3.2.13 and 3.2.15 are then:
B is σ-unital, D ⊆ M(B) with D · B dense in B, and D has the above considered
property.
In the general version of Proposition 3.2.13, C has to be a separable C *-
subalgebra of M(D) ⊆ M(B), and V : C → B is a completely positive contraction
such that V (Ψup D,B up D,B
D,C (J)) ⊆ Ψdown (J) and [V ] : C/ΨD,C (J) → B/Ψdown (J) is nuclear
for every closed ideal J ⊆ D. See the last part of the introduction or the proof of
Corollary 3.10.14 for the definition of the Ψ’s.
In the modified version of Proposition 3.2.15, C has to be a separable C *-
subalgebra of Cb (X, D) and the point-norm continuous family of complete contrac-
tions Vy from D into B must consist of residually nuclear maps.

Note that (1)-(4) imply also the assumptions of Corollary 3.10.14.

 12. COLLECTION OF SOME TOPICS 461

Proposition 3.11.5. Let B a stable σ-unital strongly p.i. algebra, A a separable

exact algebra, and h : A → Q(B) := M(B)/B a nuclear C*-morphism. Then
A0 ∩ Q(B) is strongly p.i.

Proof. ?? to be filled in 

Remark 3.11.6. In [462], [463], [92], [93] and [443] it is shown, that p.i. al-
gebras in several cases are strongly p.i. (and thus have the WvN–property). We
list some related results:

(i) Locally purely infinite C *-algebras of real rank zero are strongly purely
infinite, cf. [463] and [93].
(ii) The algebra A of sections of a continuous field of C *-algebras over a locally
compact space is strongly purely infinite, if it is purely infinite and if the
field has simple fibers. This is just the case where A is purely infinite and
Prim(A) is Hausdorff. (Joint work with E. Blanchard, [92], [93].)
In particular, C0 (X, D) is strongly purely infinite, if D is simple and
purely infinite.
(iii) C0 (X, D) is strongly purely infinite if D is strongly purely infinite and X
is locally compact ([443]).
(iv) M(D) is strongly purely infinite if D is strongly purely infinite and σ-
unital ([443]).
(v) Extension of strongly purely infinite algebras are strongly purely infinite
([443]).

It follows that Q(R+ , B) is strongly purely infinite if B is strongly purely infi-

nite.

12. Collection of some topics

Next: from old proof of Proposition 3.6.1

????? (i):
Let J / B and C ⊆ M(B) a C *-subalgebra of M(B). Recall that Ψup B,C (J) :=
C ∩ M(B, J), that M(B, J) := {t ∈ M(B) ; tB + Bt ⊆ J} is the same as the
closure of J / B in the strict topology on M(B), and that every strictly closed ideal
I of M(D) coincides with M(D, I ∩ D).
Since DB and BD are dense in B (by assumption), we get that the natural
unital *-monomorphism M(D) ,→ M(B) is strictly continuous. Thus, M(B, J) ∩
M(D) = M(D, D ∩ M(B, J)) and Ψup up
B,C (J) = ΨC,D (K) =: ΨC (K) for K :=
D ∩ M(B, J). The ideal J1 := ΨB (K) := ΨD,B
down (K) of B is defined as the closure
of span(BKB) and is contained in B ∩ M(B, J) = J. Indeed, span(BKB) ⊆
M(B, J), because K ⊆ M(B, J).
Ψup
D,B (ΨD (I)) =??????
462 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

something missing ???

Suppose that V is ΨC -ΨB –residually nuclear, and J ∈ I(B). Let K :=

Ψup
B,D (J). Then L := ΨC (K) = ΨupB,C (J) by Lemma 3.6.2(ii).

Thus, V (L) = V (ΨC (K)) ⊆ ΨB (K), and [V ] : C/ΨC (K) → B/ΨB (K) is nu-
clear. Since ΨB (K) ⊆ J (by Lemma 3.6.2(iv)), it follows that V (Ψup B,C (J)) ⊆ J,
and that [V ]J : C/Ψup
B,C (J) → B/J is nuclear. Hence, V is Ψup
B,C -residually nuclear.
Conversely, suppose that V is Ψup
B,C –residually nuclear, and let N /D. Consider
the ideal J := ΨB (N ) := span(BN B). We get N ⊆ K := D ∩ M(B, J) and
J = BN B ⊆ BKB ⊆ J. Thus M(D, N ) ⊆ M(D, K) = M(D) ∩ M(B, J) and
ΨC (N ) ⊆ Ψup
B,C (J).

Since V (Ψup up
B,C (J)) ⊆ J and [V ]J : C/ΨB,C (J) → B/J is nuclear, we get that
V (ΨC (N )) ⊆ J = ΨB (N ) and [V ] = [V ]J ◦ π : C/ΨC (N ) → B/J is nuclear, where
π : C/ΨC (N ) → C/Ψup B,C (J) is the natural epimorphism. Hence, V is ΨC -ΨB –
residually nuclear.
End of old proof-parts. Missing somewhere? Remove above??

Used in Remark ?? and quoted to Chapter 3:

Suppose that B is strongly purely infinite. Then, for every finitely generated
Abelian C*-subalgebra A ⊆ Bω with at most 1-dimensional maximal ideal space
Prim(A), there exists a C*-morphism

h : A ⊗ O ∞ → Bω with h(a ⊗ 1) = a ∀ a ∈ B .

Overlap with Chp.3 Part 2 beginning here

Compare below with Section 7
See also begin of Section 8 for next:
There are proofs of some the following corollaries, that are more elementary
than given below. But we want to invite the reader to use (and think about)
the more conceptional ideas presented here. For the classification of stably finite
algebras with non-trivial traces, one has to produce refined tools, that answer
e.g. the more difficult question (by a non-trivial and applicable answer):

When there is a given completely positive map V in a given cone C of completely

positive maps C ⊆ CP(A, B) that are compatible with a given map T (B) → T (A)?
Here T (B) means the set of lower semi-continuous (unbounded) 2-quasi-traces
τ : B+ → [0∞]. Our results together contain an answer in a special case where
τ (B+ ) ⊆ {0, ∞} for all τ ∈ T (B) , because then the problem reduces to the study
of CPnuc (Ψ; A, B) for a lower semi-continuous action Ψ : I(B) ∼ = O(Prim)(B) →
I(A), if we identify a {0, ∞}-valued lower semi-continuous 2-quasi-trace τ : B+ →
[0, ∞] by its kernel J := span{a ∈ B+ ; τ (a) = 0}.
Compare for next Def. also Def. 3.8.1
 12. COLLECTION OF SOME TOPICS 463

Definition 3.12.1. Let S ⊆ CP(A, B).

We define ΨS (J)+ as the set of a ∈ A+ with V (exp(−ih)a exp(ih)) ∈ J for all
h∗ = h ∈ A with khk < π and for all V ∈ S. Then ΨS (J)+ is the positive part of
a closed ideal ΨS (J) of A, and

ΨS : I(B) ∼
= O(Prim(B)) → I(A)

is a lower semi-continuous action of Prim(B) on A, cf. Lemma 3.12.2(iv).

Let I a closed ideal of A. We denote by ΨS (I)) the smallest closed ideal of B
that is contains {V (a) ; a ∈ I, V ∈ S}. The map

ΨS : I(A) ∼
= O(Prim(A)) → I(B)

is an upper semi-continuous action, cf. Lemma 3.12.2(xii).

Let X a topological space, and let ΨA : O(X) → I(A) and ΨB : O(X) →
I(B) increasing maps, i.e., actions of X on A and B. We define the cone of Ψ-
equivariant c.p. maps as the set CΨ := CP(ΨA , ΨB ; A, B) ⊆ CP(A, B) of maps
V ∈ CP(A, B) with V (ΨA (U )) ⊆ ΨB (U ) for all open subsets U ⊆ X. In the special
case, where X := Prim(B) and ΨB : O(Prim(B)) → I(B) is given by the natural
identification of the open subsets U of Prim(B) with the corresponding closed ideal
JU of B, where JU denotes the intersection of all primitive ideals I ∈ Prim(B) \ U ,
we write Ψ := ΨA and drop ΨB , i.e., write CΨ := CP(Ψ; A, B) ⊆ CP(A, B) for
Ψ : O(Prim(B)) → I(A) .
The use of the notation CΨ , in place of CΨA ,ΨB , becomes justified in a different
way in part (i) of Lemma 3.12.2.

Compare with Lemma 3.8.2

Lemma 3.12.2. Let A and B C*–algebras, S ⊆ CP(A, B) a set of completely

positive maps, C ⊆ CP(A, B) a matrix operator convex cone, X a topological space,
ΨA : O(X) → I(A) and ΨB : O(X) → I(B) actions of X on A respectively B,
Suppose that ???????????? ??

(i) There exists a lower semi-continuous action Ψ0 : O(Prim(B)) ∼ = I(B) →

I(A) with Ψ0 (ΨB (U )) ⊃ ΨA (U )), which is minimal in the sense that
Φ(J) ⊃ Ψ0 (J) for every J ∈ I(B) if Φ : I(B) → I(A) is lower semi-
continuous and Φ(ΨB (U )) ⊃ ΨA (U ).
It holds CΨ0 = CΨ .
(ii) The point-norm closure of a matrix operator-convex cone (in the algebraic
sense) is a matrix operator-convex cone.
The intersection of a family of matrix operator-convex cones is a matrix
operator-convex cone.
(iii) Let C(S) denote the smallest point-norm closed matrix operator-convex
cone that contains a subset S ⊆ CP(A, B). Then every contraction V in
the point-norm closure C(S) of can be approximated by maps W ∈ Calg (S)
464 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

P ∗ ∗
of the particular form W := k ck (Vk ⊗ idn )(rk (·)rk )ck with Vk ∈ S,
rk ∈ M1,n (A), ck ∈ Mn,1 (B) and k k ck Vk (rk∗ rk )ck k ≤ 1 .
P ∗

(iv) ΨS (J)+ is the positive part of a closed ideal ΨS (J) of A, and

ΨS : O(Prim(B)) ∼
= I(B) 3 J 7→ ΨS (J) ∈ I(A)

is a lower semi-continuous action of Prim(B) on A.

The action ΨS satisfies ΨS (B) = A and V (ΨS (J)) ⊆ J for all J ∈ I(B)
and V ∈ S.
(v) ΨS = ΨC for C := C(S).
(vi) ΨS (J) ⊆ ΨS1 (J) for all J ∈ I(B), if S1 ⊆ S.
(vii) The set CΨ is a point norm-closed matrix operator convex cone of com-
pletely positive maps.
(viii) CΨ1 ⊆ CΨ if Ψ(J) ⊆ Ψ1 (J) for all J ∈ I(B).
(ix) C ⊆ CΨC .
(x) ΨCΨ (J) ⊃ Ψ(J) for J ∈ I(B) ∼ = O(Prim(B)).
0 max max
(xi) If C ⊆ CP(A ⊗ D, B ⊗ D) is a point-norm closed matrix operator
convex cone, then the set C ⊆ CP(A, B) of T ∈ CP(A, B) with T ⊗max
idD ∈ C 0 is a point-norm closed matrix operator convex cone.
(xii) Let M a W*-algebra and B ⊆ M a σ(M, M∗ )-dense C*-subalgebra of M ,
and let C ⊆ CP(A, B) ⊆ CP(A, M ) a matrix operator-convex cone. Then
the point-*ultra-strong closure C ⊆ CP(A, M ) of C is a point-σ(M, M∗ )
closed matrix operator convex cone.
(xiii) ΨS : I(A) → I(B) is an upper semi-continuous action of Prim(A) on
B. I ⊆ ΨS (ΨS (I)) for every closed ideal I of A and every subset S ⊆
CP(A, B).

Proof. (iv): Let J a closed ideal of B and T : A → B a positive map. The set
of a ∈ A+ with (πJ ◦ T )(a) = 0 is a hereditary closed convex sub-cone CT,J of A+ .
Thus, the set CS,J of a ∈ A+ with T (a) ∈ J for all T ∈ S is a hereditary closed
convex sub-cone of A+ , and ΨS (J)+ ⊆ A+ of Definition 3.12.1 is the set of a ∈ A+
with exp(ih)a exp(−ih) ∈ CS,J . By Lemma ??, ΨS (J)+ is the positive part of a
closed ideal ΨS (J) of A. Clearly, ΨS (B) = A, and V (ΨS (J)+ ) ⊆ J for J ∈ I(B)
and V ∈ S, by definition of ΨS .
T
If {Jσ }σ∈Σ is a family of closed ideals and a ∈ A+ and let J := σ Jσ . then
V (exp(ih)a exp(−ih)) ∈ J for all V ∈ S and h∗ = h ∈ A with khk < π, if and only
if, V (exp(ih)a exp(−ih)) ∈ Jσ for all V ∈ S and h∗ = h ∈ A with khk < π, i.e.,
T T
a ∈ ΨS (J)+ if and only if a ∈ σ ΨS (Jσ )+ = ( σ ΨS (Jσ ))+ . Thus,

ΨS : O(Prim(B)) ∼
= I(B) 3 J 7→ ΨS (J) ∈ I(A)

is a lower semi-continuous map. The lower semi-continuity of the map ΨS implies

that ΨS is monotone. Thus, ΨS is a lower s.c. action of Prim(B) on A.
(vii): Since ΨB (U ) is a closed ideal, the set of maps V ∈ CP(A, B) with
V (a) ∈ ΨB (U ) for all a ∈ ΨA (U ) is point-norm closed. Thus, CΨ is point-norm
closed. Similar arguments show:
 12. COLLECTION OF SOME TOPICS 465

Let V1 , V2 ∈ CP(A, B) with Vk (ΨA (U )) ⊆ ΨB (U ) for all U ∈ O(X) (k = 1, 2),

t ∈ [0, ∞), and r ∈ M1,n (A), c ∈ Mn,1 (B). Then (V1 + tV2 )(a) ∈ ΨB (U ) and
c∗ (V1 ⊗ idn )(r∗ ar)c ∈ ΨB (U ) for U ∈ O(X) and a ∈ ΨA (U ). Thus CΨ is a point-
norm closed matrix operator convex cone of c.p. maps.
(i): If we use the natural isomorphisms I(A) ∼ = O(Prim(A)) and I(B) ∼
=
O(Prim(B)), and Lemma ??, then we get that there is a minimal lower semi-
continuous action Ψ0 : O(Prim(B)) → I(A) with Ψ0 (ΨB (U )) ⊃ ΨA (U ).
If V ∈ CΨ0 , then V (ΨA (U )) ⊆ V (Ψ0 (ΨB (U ))) ⊆ ΨB (U ) for U ∈ O(X), because
ΨA (U ) ⊆ Ψ0 (ΨB (U )). Thus, CΨ0 ⊆ CΨ = CΨA ,ΨB .
Let S := CΨ , then ΨS : I(B) ∼
= O(Prim(B)) → I(A) is a lower semi-continuous
action by part (iv) with V (ΨS (J)) ⊆ J for all V ∈ S.
We have ΨS (ΨB (U ))+ ⊃ ΨA (U )+ for U ∈ O(X), because V (e−ihaeih ) ∈
ΨB (U ) for a ∈ ΨA (U )+ and h∗ = h ∈ A. Thus, ΨS (J) ⊃ Ψ0 (J) for every J ∈ I(B)
by minimality of Ψ0 . It implies that V (Ψ0 (J)) ⊆ V (ΨS (J)) ⊆ J, i.e., V ∈ CΨ0 , for
all V ∈ CΨ = CΨA ,ΨB .
(ii): The point-norm closure C of a matrix operator-convex cone C0 is a matrix
operator-convex cone, because the topology of point-norm convergence on L(A, B)
coincides with the strong operator topology, and because Vα ⊗ idn converges in
point-norm to V ⊗ idn if Vα ∈ C0 converges to V .
(iii): Let n ∈ N and let C (n) denote the set of completely positive maps W :=
∗ ∗
P
k ck (Vk ⊗ idn )(rk (·)rk )ck with Vk ∈ S, rk ∈ M1,n (A) and ck ∈ Mn,1 (B), and
let C0 = n C . Clearly, C (n) ⊆ C for every matrix operator-convex cone C ⊆
T (n)

CP(A, B) with S ⊆ C, and the set C (n) is closed under multiplication with positive
scalars and under addition.
Moreover, c∗ (W ⊗ idm )(r∗ (·)r)c ∈ C (mn) if W ∈ C (n) , r ∈ M1,m (A) and c ∈
Mn,1 (B), because

c∗ (Wk ⊗ idm )(r∗ (·)r)c = C ∗ Vk ⊗ idmn (R∗ (·)R)C

for Wk := c∗k (Vk ⊗ idn )(rk∗ (·)rk )ck , R = [R1 , . . . , Rmn ] ∈ M1,mn (A) and C =
(k) (k)
[C1 , . . . , Cmn ]> ∈ Mmn,1 (B), where Rjm+y := ry rj and Cjm+y := cj cy with
(k) (k) (k) (k)
rk = [r1 , . . . , rn ] ∈ M1,n (A), r = [r1 , . . . , rm ] ∈ M1,m (A), ck = [c1 , . . . , cn ]> ∈
Mn,1 (B), and c = [c1 , . . . , cm ]> ∈ Mm,1 (B).
If we fill the rows rk ∈ M1,n (A) and columns ck ∈ Mn,1 (B) with zeros, then
we see that C (n) ⊆ C (n+k) for all n, k ∈ N. Thus C0 := n C (n) is a (not necessarily
S

closed) matrix operator convex cone that is contained in every matrix operator
convex cone C with S ⊆ C.
Since A has an approximate unit {eτ } of positive contractions, we get kW k =
k k c∗k Vk (rk∗ rk )ck k for W ∈ C0 , and that every element V ∈ S is in the point-norm
P

closure of C (1) ⊆ C0 .
By part (ii), the point-norm closure C1 of C0 is a m.o.c. cone that contains
S. Since C0 is contained in every point-norm closed m.o.c. cone C with S ⊆ C,
466 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

we get C(S) = C1 . The contractions T in the point-norm closure of C0 can be

approximated by contractions W ∈ C0 with kW k ≤ 1 by Lemma 3.1.8.
(v): Since S ⊆ C(S) =: C, it holds ΨC (J) ⊆ ΨS (J) for all J ∈ I(B), cf. part
(vi).
If a ∈ ΨS (J)+ then V (exp(ih)a exp(−ih)) ∈ J for all h∗ = h ∈ A with khk < π
and for all V ∈ S.
By part (iv), this implies that c∗` V (r`∗ exp(ih)a exp(−ih)rk )ck ∈ J for V ∈ S,
h∗ = h ∈ A, r1 , . . . , rn ∈ A and c1 , . . . , cn ∈ B. Thus, W (exp(ih)a exp(−ih)) ∈ J
for all W ∈ C := C(S) and h∗ = h by part (iii), i.e., a ∈ ΨC (J)+ , and ΨS = ΨC for
C := C(S).
(vi): Straight from definition of ΨS .
(viii): If V ∈ CΨ1 and Ψ(J) ⊆ Ψ1 (J), then V (Ψ(J)) ⊆ V (Ψ1 (J)) ⊆ J. Thus,
V ∈ CΨ if Ψ(J) ⊆ Ψ1 (J) for all J ∈ I(B).
(ix): For J ∈ I(B) and a ∈ ΨC (J) holds V (a) ∈ J for all V ∈ C by definition
of ΨC .
(x): Let J ∈ I(B), and let a ∈ Ψ(J)+ ⊆ A+ , V ∈ CΨ and h∗ = h ∈ A. Then
V (exp(ih)a exp(−ih)) ∈ J by definition of CΨ , i.e., a ∈ ΨCΨ (J)+ by definition of
ΨC . Thus, Ψ(J) ⊆ ΨCΨ (J).
(xi): If {Tγ }γ∈Γ ⊆ CP(A, B) is a (norm-)bounded net that converges in point-
norm to T ∈ L(A, B), then T ∈ CP(A, B) and the net

{Tγ ⊗max id}γ∈Γ ⊆ CP(A ⊗max D, B ⊗max D)

converges point-wise to Tγ ⊗max id, because A D is dense in A ⊗max D. Thus, if

Tγ ⊗max id ∈ C 0 for all γ ∈ Γ, then T ⊗max id ∈ C 0 . Clearly, the set C of T ∈ CP(A, B)
with T ⊗max id ∈ C 0 is convex. If T ⊗max id ∈ C 0 , r ∈ M1n (A), c ∈ Mn1 (B), e ∈ D+ ,
then R := r ⊗ e ∈ M1n (A ⊗ D), C := c ⊗ e ∈ Mn1 (B ⊗ D), and

R∗ ((T ⊗max id) ⊗ (idn ))(C ∗ (·)C)R = T 0 ⊗max Se

for T 0 := r∗ (T ⊗ idn )(c(·)c∗ )r and Se (d) := e2 de2 (d ∈ D). Since C 0 is matrix

operator convex and is point-norm closed,it follows that T 0 ⊗max id ∈ C 0 . Thus, the
set C ⊆ CP(A, B) of T ∈ CP(A, B) with T ⊗max idD ∈ C 0 is a point-norm closed
matrix operator convex cone.
norm
(xii): The point-norm closure C0 := C closure of C is a convex cone with
∗ ∗
c (V ⊗ idn )(r (·)r)c ∈ C0 for r ∈ Mn,1 (A), c ∈ M1,n (B) and V ∈ C0 . Since the
point-*strong closures of C and C0 in L(A, M ) coincide, and since Vγ ⊗ idn →
W ⊗ idn point-*strongly if Vγ → W point-*strongly , we obtain that the point-
*strong closure C ⊆ L(A, M ) is a convex cone that satisfies c∗ (W ⊗idn )(r∗ (·)r)c ∈ C0
for r ∈ Mn,1 (A), c ∈ M1,n (B) and W ∈ C. The unit-ball of Mn (B) is *-ultrastrongly
dense in the unit-ball of Mn (M ) by Kaplansky density theorem ([616, Thm. 2.3.3]).
Thus, for c ∈ M1,n (M ) there exists a net cγ ∈ M1,n (B) with kcγ k ≤ kck that
*strongly converges to c. Thus, the point-*strong closure C satisfies (OC2).
 12. COLLECTION OF SOME TOPICS 467

check next argument ??(xii) If Vγ ∈ C converges in point-σ(M, M∗ ) topology

to W ∈ Lin(A, M ), then W (e2 ) ≥ 0 for every e ∈ A+ . In particular, W is positive
and bounded, thus lim supγ kVγ (e2 )k =: µ < ∞ by uniform boundedness theorem
(Banach-Steinhaus). It implies that W is also in the point-σ(M, M∗ ) closure of the
convex set of V ∈ C with kV k ≤ µ. An Hahn-Banach separation argument shows
now that W ∈ C.
(xiii): ΨS ( γ Iγ ) = S
P P S
γ Ψ (Iγ ) and a ∈ ΨS (Ψ (I)) for a ∈ I / A follow
straight from the definitions of ΨS and ΨS . 

Compare also Example 3.8.3!!!

Example 3.12.3. We consider the action Ψnuc : I(B) → I(A) defined by

CPnuc (A, B) and examine some possible generating subsets S0 ⊆ CP(A, B) for
CPnuc (A, B).
Let S ⊆ A∗+ denote an “almost separating” set of positive functionals on A
that is invariant under inner automorphisms of A, more precisely:
for each a ∈ A+ there is ϕ ∈ S and h∗ = h ∈ A with khk < π such that with
ϕ(exp(ih)a exp(−ih)) > 0. For example we can take as S the set of pure states of
A, or any set S of pure-states ϕ of A such that the family {dϕ }ϕ∈S of irreducible
representations of A is separating for A.
Furthermore, let P ⊆ B+ a set of positive elements of B that generates B as a
closed ideal of B, i.e., for each c ∈ B+ and ε > 0, there are b1 , . . . , bn ∈ P and
d1 , . . . , dn ∈ B with kc − k d∗k bk dk k < ε .
P

We consider the set of S0 := SS,P ⊆ CP(A, B) of c.p. maps Vϕ,b (a) := ϕ(a)b,
where ϕ ∈ S and b ∈ P , and calculate the action Ψ0 : I(B) → I(A) defined by S0 :
We have Ψ( S0 )(B) = A by Lemma 3.12.2(iv). Let J 6= B a close ideal of B. Then
there is b ∈ P that is not in J. If 0 6= a ∈ A+ then there is h∗ = h ∈ A with khk < π
and ϕ ∈ S with ϕ(exp(ih)a exp(−ih)) > 0. It follows that Vϕ,b (exp(ih)a exp(−ih))
is not contained in J.
Thus ΨS0 (J) = {0} for every closed ideal J 6= B and ΨS0 (B) = A.
Since S0 ⊆ CPf (A, B) ⊆ CPnuc (A, B) ⊆ CP(A, B) we get ΨCP(A,B) =
ΨCPnuc (A,B) (J) ⊆ ΨC0 (J) for all closed ideals J of B. Thus,

ΨCP(A,B) = ΨCPnuc (A,B) = ΨS0 .

The special case where A = B is not nuclear shows that,

in general, the lower semi-continuous action ΨC : O(Prim(B)) → I(A) defined by
a m.o.c.c. C ⊆ CP(A, B) does identify C itself.

The situation of example 3.12.3 changes considerably if we consider, instead

of ΨC , the action Ψ0 : I(B ⊗max C) → I(A ⊗max C) that is defined by the point-
norm closed m.o.c.c. C ⊗max CPin (C, C) ⊆ CP(A ⊗max C, B ⊗max C), where C :=
C ∗ (F∞ ) denotes the full group C *-algebra over the free group F∞ on countably
many generators and where C ⊗max CPin (C, C) is generated by the tensor products
V ⊗max idC with V ∈ C. Then the following theorem holds:
468 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

Theorem 3.12.4 (Separation of m.o.c.c.’s). Let S ⊆ CP(A, B) a set of c.p.

maps, and denote by Ψ0 : I(B ⊗max C) → I(A ⊗max C) the action that is defined
by the set S 0 of c.p. maps V ⊗max idC : A ⊗max C → B ⊗max C.
Then a c.p. map T : A → B is in the point-norm closed matrix operator-convex
cone C(S) generated by S, if and only if, T ⊗max idC is Ψ0 -equivariant.

Compare also Thm. 3.8.4

Proof. Let Ψ0 := ΨS 0 : I(B ⊗max C) → I(A ⊗max C) be the l.s.c. action of

Prim(B ⊗max C) on A⊗max C defined by the set S 0 of c.p. maps V ⊗max idC : A⊗max
C → B ⊗max C for V ∈ S, cf. Lemma 3.12.2(iv).
We denote by C 0 ⊆ CP(A⊗max C, B⊗max C) the cone C 0 := CΨ0 = CP(Ψ0 ; A⊗max
C, B ⊗max C) of Ψ0 -equivariant maps.
Then C 0 is a point-norm closed matrix operator convex cone by Lemma
3.12.2(vii).
The set C ⊆ CP(A, B) of all T ∈ CP(A, B) with T ⊗max idD ∈ C 0 is a point-
norm closed matrix operator convex cone that contains S, cf. Lemma 3.12.2(xi),
i.e., C(S) ⊆ C and (V ⊗max idC )(Ψ0 (J)) ⊆ J for all closed ideals J of B ⊗max C.
?????????


13. Graded m.o.c. cones

Needed topics:
Graded Hilbert B-modules E and interplay with grading on B.
Opposite E op module: Interchange of the Z2 “eigen” spaces.
Opposite grading on B for E op ?
Gradings on E op . Induced grading on L(E) is given by conjugation with the
grading operator α on E :
T ∈ L(E)(0) if and only if T commutes with α.
T ∈ L(E)(1) if and only if T anti-commutes with αT α = −T , i.e., αT +T α = 0.
Notation on graded C *-algebras A :
α or αA is a C *-automorphism of order 2.
A(0) := {a ∈ A ; αA (a) = a}, A(1) := {a ∈ A ; αA (a) = −a},
(also denoted by A0 , A1 )
a ∈ A(0) ∪ A(1) are “homogeneous”. Depending if a ∈ A(0) or a ∈ A(1) for
homogeneous a ∈ A is defined a degree deg(a) := j =: ∂a if a ∈ A(j) .
The graded commutator [a, b]gr in a graded algebra A with grading automor-
phism α ∈ Aut(A) is defined by [a, b] = ab − ba if α(a) = a or α(b) = b i.e., if
 13. GRADED M.O.C. CONES 469

a, b ∈ A0 . Defined in general by:

[a, b]gr := ab − (−1)deg(a)·deg(b) ba

It gives [a, b]gr = ab − ba if at least one of a, b has degree zero. if both have
degree one, i.e., deg(a) = 1 = deg(b), then [a, b]gr := ab + ba ∈ A(0) .
Gradings on m.o.c. cones C. H
d B for graded B. Effect to passage to opposite
op
grading on B, HB and Hd B := H B ⊕ HB .
Related grading on HC : A → L(HB ) ???
Interplay of grading with on A and B with CP(A, B) ⊂ L(A, B)? What kind
of compatibility of m.o.c. cones C ⊆ CP(A, B) is necessary to get that HC : A →
L(Hd B ) becomes grading preserving?

Change on E the “eigen” spaces but not the grading on L(E)?

E = E (0) ⊕ E (1) (Is this really the orthogonal sum ?)
The action of graded B should be (all calculated modulo 2):
E (m) B (n) ⊆ E (m+n)
and
E (m) , E (n) ⊆ B (m+n)
E op = (E op )(0) ⊕ (E op )(1) is same E, but with (E op )(0) := E (1) and (E op )(1) :=
(0)
E .
Means replacing the order-2 grading isomorphism α by (−1) · α.
(Don’t know what happens if E (1) = 0, then only 0 is fixed?)
See sec. 14.2 of Blackadar K-theory.
Standard odd grading on A ⊕ A is given by (A ⊕ A)(0) = {(a, a) ; a ∈ A} and
(A ⊕ A)(1) = {(a, −a) ; a ∈ A} .
If A = C then C1 is a Clifford algebra, isomorphic to the group C *-algebra of
Z2 .
Standard even grading on M2 (A) is given by conjugation with diag(1, −1),
i.e., M2 (A)(0) are the diagonal matrices, and M2 (A)(1) are the matrices with zero
diagonal.
If α is grading on A and β grading on Hilbert B-module E and φ : A → L(B)
is grading preserving. Then:
(0.) β := −α is the opposite grading on E.
(1.) ψ := φ ◦ β : A → L(E) is grading preserving for E op if φ was grading
preserving for for the grading on L(E).
(2.) If (E, φ, F ) is a Kasparov module then (E op , ψ, −F ), (E ⊕E op , φ⊕ψ, G(t))
are Kasparov module for the 2 × 2-matrix G(t) with entries G(t)11 := cos(tπ/2)F ,
G(t)22 := − cos(tπ/2)F , G(t)12 := sin(tπ/2) and G(t)12 := sin(tπ/2) .
470 3. NUCLEAR C.P. MAPS AND OPERATOR-CONVEX CONES

The G(t) is a linear combination of diag(F, −F ) = G(1) and G(0) . Both have
degree 1 under the grading diag(α, −α)(·) diag(α, −α) in Lin(E ⊕ E op ). The G(0)
commutes with φ(a) ⊕ ψ(a) = diag(φ(a), ψ(a)) for all a ∈ A with αA (a) = a, and
“anti-commutes” for all a ∈ A with αA (a) = −a . This is because deg(G(0)) = 1,
A ∈ a 7→ diag(φ(a), ψ(a)) is grading preserving, i.e., deg(diag(φ(a), ψ(a))) = 1 in
M2 (L(E)) if deg(diag(φ(a), ψ(a))) = 1 and

[ G(0), diag(φ(a), ψ(a)) ]gr = G(0) · diag(φ(a), ψ(a)) + diag(φ(a), ψ(a)) · G(0) = 0

In a the graded C *-algebra L(E ⊕ E op ) we have the grading automorphism

α(·)α for the grading α := αE ⊕ (−αE ) coming from the grading on E ⊕ E op :
And α · G(1) · α = −G(1) ???
 CHAPTER 4

Comparison and Addition of some C *-morphisms

Actual plan:
(A) Remove 2-fold explanations/definitions
(B) Finish proof of Thm. 4.4.6.
(C) Outline ‘‘germ’’ of un-suspended but stable variant of E-theory.
In particular:
(C1) Derive central sequences of copies of O2 for O2
(C2) Derive approximate unitary equivalence of all unital
endomorphisms of O2 and of O∞ .
(C3) Derive needed central sequences from this.
(C4) Also quasi-central asymptotic paths of copies of O2 and of O∞
... ??
(C?? from Into 1.)
“Squeezing Property” implies K1 -injectivity. See Chp. 4. for definitions!
Here a question remains:
When K1 -injectivity implies the Squeezing Property?
At least in unital separable nuclear case ??
(Would be very nice ! But have no idea about this)

We study here some properties of Cuntz addition, that is, a “sum” of mor-
phisms in the category of C *-algebras, more generally we consider “sums” of general
maps between C *-algebras. This idea of an addition has been used implicitly in the
topological K-theory of C *-algebras since a long time, but without precise descrip-
tion e.g. of the “almost ridiculous ” relationship between the isomorphisms of K
with M2 (K) and unital representations of the Cuntz algebra O2 in M(K) = L(H),
needed for precise algebraic definition of equivalence classes of asymptotic and
in some sense “equivariant” C *-morphisms. It was J. Cuntz ([172], [169]) who
pointed out in a more general context that a unital copy of O2 or at least of O∞ is
needed to adjust them carefully. The natural transformations between those copies
of O2 are present for our definitions of sorts of equivariant theories by actions of
groups, ideal-lattices or categories of matrix-convex cones.
The point is that we use them later to describe the constructive aspects of the
used Ext-, KK-theories and unsuspended variants of cone-depending E-theory that
corresponds to them.
471
472 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

The general notion of Cuntz addition is related to K∗ -theory (cf. Lemma 4.2.6
and Proposition 4.4.3). It is taylor-made for our study of nuclear asymptotic C *-
morphisms into non-simple purely infinite algebras in Chapters 3, 5 and 7. And
it helps to describe the construction of our more general KK(C ; ·, ·)-theory for
operator-convex cones C in Chapter 8. The Cuntz addition is neither a direct
product nor a direct sum in the usual sense of category theory – here on the category
of C *-algebras. This restricts “straight borrowing” from category theory analogies.
The considered situations and constructed objects appear only modulo various
ideals in its applications in Chapters 5, 7, 8 and 9. We apply the here proved
results, after dividing by suitable ideals in question. But this has to be done with
some care:
Two C *-morphisms h1 , h2 : D → E into a unital C *-algebra E can become
unitarily equivalent after dividing out an ideal J of E. In general, then there is no
unitary u in the original algebra with h1 (a) − u∗ h2 (a)u ∈ J for all a ∈ D . But in
the particular situation where this unitary equivalence in E/J can be realized by a
unitary in the connected component U0 (E/J) of 1 in U(E/J) , then this unitary can
be lifted to a unitary in E. This is, for example, the case if there exists an isometry
r ∈ E/J with r∗ πJ (h1 (d) + h2 (d))r = 0 for d ∈ D, cf. Proposition 4.3.6(iv). It
applies e.g. to equivalence of stable extensions, cf. Chapter 5, and to asymptotic
C *-morphisms into stable σ-unital algebras, cf. Chapter 7 and Corollary 4.6.7.
Recall here that U(E) denotes the unitary group of a unital C *-algebra E and
U0 (E) ⊆ U(E) the connected component of 1 ∈ U(E).
We consider also some real C *-algebras and real versions E = ER of C *-
algebras E, e.g. the real versions (O2 )R , (E2 )R and C ∗ (P, Q, 1)R of O2 , E2 and
the universal unital (complex) C *-algebra C ∗ (P, Q, 1)R ⊗R C generated by two
projections.
We do not discuss KK-theory of real C *-algebras here, but it seems natural
to use the real basic algebras, because the complex versions are often just natural
complexification of the real version. In the real cases E = ER , (O2 )R and (E2 )R we
use also the notations “unitary”, U(E) and U0 (E) instead of “orthogonal operator”,
O(E) and SO(E), for the orthogonal operators u ∈ E, respectively for u ∈ E in
the connected component of 1 in O(E).
Formulas that cover also the “real” case look sometimes more complicate.
If the reader wants simpler formulas for complex C *-algebras E then he can
replace e.g. the order 4 unitary

(1 − zz ∗ − z ∗ z) + (z ∗ − z) = exp((π/2)(z ∗ − z)) ∈ U0 (E)

for partial isometries z with z 2 = 0 by the symmetry (1 − zz ∗ − z ∗ z) + (z ∗ + z).

Symmetries are not necessarily in U0 (E) if E is a real C *-algebra. But there are
also real C *-algebras where every symmetry is connected inside the orthogonal
operators with 1 by a path of orthogonal operators. Examples are all real unital
 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS 473

C *-algebras that have a central sequence of unital copies of M2 (R), i.e., that absorb
the real UHF algebra M2 (R) ⊗ M2 (R) ⊗ · · · tensorial.
This is one of the reasons that we try to express canonical transformations as
products of Halmos unitaries (defined in Remark 4.2.4), because those are also for
real unital C *-algebras A connected to 1 in U(A).
We need the verification of the homotopy invariance for definitions and con-
structions, those “soft” homotopy invariances will be later used to prove “hard”
unitary homotopy in the sense of Definition 5.0.1. The passage to unitary homo-
topy is a basic tool in our proofs of classification results from stable homotopy
coming from KKX -equivalences or from the more general KK(C; ·, ·)-equivalences.
The results of Lemmata 4.2.6, 4.2.10 and Propositions 4.2.11, 4.2.15, 4.3.6 and
4.4.3 will be used in the proofs in Chapters 5, 7, 8, 9 and 11. Theorem 4.4.6 plays
a crucial role in the proof of the first parts of Theorems B and M, that are given in
Chapters 6 and 9 – in case of Theorem M with an additional separation assumption
that will be removed later in Chapter 12 by the last step in the proof of Theorem
K that uses the very special case of Theorem M proven before in Chapter 6.

Definition 4.0.1. Let E a C *-algebra such that its multiplier algebra M(E)
contains a copy of O2 unitally. For a fixed pair of generators s1 , s2 of O2 in M(E),
we define the Cuntz addition ⊕s1 ,s2 in E as follows:

a ⊕s1 ,s2 b = s1 as∗1 + s2 bs∗2 ; a, b ∈ E .

We define and use in the “overview” Lemma 4.2.6 also a generalization of this
addition, that is more natural, up to certain transformations and up to unique-
ness depending from K1 -injectivity. A careful study of all transformations is in
particular necessary if one wants to go into further research concerning equivariant
actions of lattices, groups, quantum groups, groupoids, etc. After discussing some
basic material on absorbing elements in Abelian semigroups and compatibility of
projections, we study some less trivial properties of the generalized Cuntz addi-
tion of C *-morphisms. This is not new, but gives a constructive “algebraic” base
for our in applications used versions of K-theory, cone-depending KK-theory and
unsuspended “continuous” E-theory.

Remark 4.0.2. The above defined Cuntz addition is a unital *-monomorphism

from E ⊕ E into E. It is, up to unitary equivalence in E, independent of the choice
of copies of O2 : If s1 , s2 and t1 , t2 are generators of two copies of O2 in E, then

a ⊕s1 ,s2 b := u(a ⊕t1 ,t2 b)u∗ for all a, b ∈ E

where u := s1 t∗1 + s2 t∗2 ∈ E is a unitary.

More generally, one can add n elements of E along generators of a unitally

contained copy of On in E with generators s1 , . . . , sn in the natural way:
X
(a1 , a2 , . . . , an ) 7→ si ai s∗i .
474 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

As before the sum is independent of the choice of generators – up to unitary equiv-

alence by a fixed unitary. Notice that O2 contains all Cuntz algebras On , n ∈ N,
unitally, e.g. O3 ∼
= C ∗ (s1 , s2 s1 , s22 ). (In fact every separable unital exact C *-algebra
is isomorphic to a unital C *-subalgebra of On by – cite: Kirchberg-Phillips.)
Computation shows that adding along On is the same as successively adding
along O2 – up to unitary equivalence. In particular, Cuntz addition is associative
and commutative up to (global) unitary equivalence.
The uniqueness up to unitary equivalence allows us to write sometimes simply
⊕ instead of ⊕s1 ,s2 if we consider unitary equivalence classes and different unital
copies of O2 in a given C *-subalgebra of E.
The generators of a copy of O2 define an “almost inner” *-isomorphism ψ from
M2 (E) onto E, by
X
ψ : [aik ] ∈ M2 (E) 7→ v[aik ]v > = si aik s∗k ∈ E ,
ik

where v is the row matrix v := [s1 , s2 ] ∈ M1,2 (E) .

There exists a unitary matrix V ∈ M4 (E) (a Halmos unitary as in Remark
4.2.4) such that V ∗ diag(p; 0, 0)V = diag(ψ(p), 0; 0, 0) for all projections p ∈ M2 (E)
and V ∗ diag(u; 1, 1)V = diag(ψ(u), 1; 1, 1) for all unitaries u ∈ M2 (E), cf. proof of
Lemma 4.2.6(v,2).
Therefore, equivalence and homotopy in M2n (E) can be over-carried to equiv-
alence and homotopy in E itself.
In some calculations it plays a role if this can be made only up to unitary
equivalence or, moreover, up to unitary homotopy. This difference has usually to
do with problems related to K1 -injectivity and has forced us to show at least all (not
necessarily simple) purely infinite unital C *-algebras are K1 -injective, cf. Proposi-
tion 4.2.15(b), respectively Lemma 4.2.13 for the “squeezing property” of Definition
4.2.14 that induces K1 -injectivity by Proposition 4.2.15(?).
discover what???
!!! HIER BIN ICH !!!!
KUERZE DAS ZEUG ODER GEBE GENAUE DEF’S, FORMELN AN ...

A detailed discussion of the required transformation and precise conditions for

uniqueness – up to unitary homotopy – will be given in the proofs of Lemma 4.2.6,
that considers a generalized version of Cuntz addition on A in the case where here
E := M(A) is properly infinite.
The below given Lemmata 4.1.3, 4.2.3 and 4.2.6 collect some basic facts con-
cerning the description of K∗ (E) for properly infinite unital C *-algebras E by
Murray–von Neumann equivalence classes of properly infinite projections in E (re-
spectively by homotopy classes of unitaries in E), and they rephrase some technics
appearing in proofs of [172] and of [180]. In particular the notions of absorption
 1. TWO PROJECTIONS IN C *-ALGEBRAS 475

and domination will be discussed in connection with some technics to surround a

possibly missing K1 -injectivity for some properly infinite C *-algebra.
We show that all purely infinite unital C *-algebras have the more general
“squeezing property” of Definition 4.2.14 that induces K1 -injectivity by ???????

1. Two projections in C *-algebras

We have often, but sometimes only “implicit”, to do with two projections and
their relative positions, e.g. with the open support projections of two closed left
ideals, or with positions of two hereditary C *-subalgebras of a C *-algebra A. We
consider in this Section 4.1 only selfadjoint projections, i.e. in the following a
“projection” p in a C *-algebra A ⊆ L(H) means always an orthogonal projection
on a real or complex Hilbert space H, i.e., p = p∗ p. Then 0 ≤ p = p2 and we
denote the set of projections in A by Proj(A). The group of unitary elements in
the unitization Ae will be denoted by

Or if we study orthogonality of positive elements of A ...

This ??????
Projections p, q ∈ A (respectively unitaries u, v in the unitization Ae of A) are
homotopic — denoted by p ∼h q (respectively u ∼h v ) —, if there is a (continuous)
path t ∈ [0, 1] 7→ p(t) ∈ Proj(A) (respectively t ∈ [0, 1] 7→ u(t) ∈ U(A)
e ) such that
p(0) = p and p(1) = q (respectively u(0) = u and u(1) = v).
Here a ∼M vN b for a, b ∈ A+ means Murray–von-Neumann equivalence
– also denoted by [a]M vN = [b]M vN or simply by a ∼ b – and is defined as the
existence of d ∈ A with d∗ d = a and dd∗ = b.
The notation p ∼ q (more precisely p ∼M vN q) for projections p, q ∈ A denotes
Murray–von-Neumann equivalence, i.e., the existence of v ∈ A with v ∗ v = p and
vv ∗ = q. Recall that this implies [p] = [q] in K0 (A).
Beginning from HERE some text of old Appendix has to be integrated
!!!
Ends up ????
Integrate next into the further below considerations!!!

Lemma 4.1.1. Suppose that A is a C*-algebra, and let p, q ∈ A projections.

(i) If kpqk < 1, then 0 is isolated in Spec(p + q). And q ∼h (1B − p) inside
the unital C*-subalgebra B := C ∗ (p, q) ⊆ (p + q)A(p + q).
(Is identical with (p + q) − ξ1M(A) invertible in M(A) for all non-zero
ξ ∈ C with 0 < |ξ| <???? for small ????. Explicit bound ???? )
(ii ???) There is a (norm-continuous) path [0, 1] 3 t 7→ u(t) ∈ U(A) e with u(0) = 1
∗
and u(1) pu(1) = q, if and only if, p ∼h q in A.
(If A is unital, then a path t 7→ u(t) with u(0) = 1 and u(1)∗ pu(1) = q.
can be found in U(A) itself.)
476 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

(ii,iii): See [692, prop. 2.2.6], or [73, prop. 4.3.3, prop. 4.6.3], or
[207, prop. IV.1.2, lem.IV.1.4].
(iii ???) If kpqk < 1, and if there exists a partial isometry z ∈ A with z ∗ z = p and
zz ∗ = q, then p ∼h q.

Proof: (ii,iii): See [692, prop. 2.2.6], or [73, prop. 4.3.3, prop. 4.6.3], or [207,
prop. IV.1.2, lem.IV.1.4].
(iii): By (i), there is a path t 7→ q(t) ∈ Proj(A) with q(0) = q and q(1)p = 0,
e with u∗ qu = q(1). Then w := u∗ z satisfies
and by (ii), there is a unitary u ∈ U0 (A)
w∗ w = p and ww∗ = q(1). Thus, s := w∗ +w+(1−p−q(1)) is a self-adjoint unitary
in Ae with sps = q(1). Hence, (us)p(us)∗ = q and us ∈ U0 (A). e If t 7→ U (t) ∈ U(A)e
is a path with U (0) = 1 and U (1) = (us)∗ , then t → p(t) := U (t)∗ pU (t) is the
desired path, and p ∼h q .
Recall that a hereditary C *-subalgebra D ⊆ A is full in A if the ideal of A
generated by D is dense in A. An element b ∈ A is full in A if the linear span of
AbA is dense in A, i.e., if A = J(b) for the closed ideal J(b) of A generated by {b}.
In particular, a projection p∗ p = p ∈ A is a full projection of A if the hereditary
C *-subalgebra pAp of A is full in A.

Remark 4.1.2. Let D ⊆ A a full hereditary C *-subalgebra of A, then the

inclusion map η : a ∈ D 7→ a ∈ A defines an isomorphism η∗ from K∗ (D) onto
K∗ (A) .
In particular, [q1 ] = [q2 ] in K0 (D) if q1 , q2 ∈ Mn (D) and there are projections
p1 , p2 ∈ A with [p1 ] = [p2 ] in K0 (A) and pk ∼M vN qk in Mn (A) for k = 1, 2.

The study of sufficient and necessary criteria for K1 -injectivity of a given prop-
erly infinite unital C *-algebra A requires some basics on pairs of projections p, q ∈ A
that are “almost compatible” in the sense that min(k(1 − q)pk, k(1 − p)qk) < 1. The
notation M2 means in the following lemma the algebras M2 (R) or M2 (C) – depend-
ing if we consider a real or complex C *-algebra C ∗ (p, q, 1) generated by projections
p, q. The Lemma itself should be “folklore”, but since we use it in several proofs
for estimates and check of “controlled” homotopy we give a careful detailed and
elementary proof. Some of it can be also obtained by observing first that the
“universal” unital C *-algebra C ∗ (P, Q, 1) := C ∗ (P, Q, 1 ; P ∗ P = P, Q∗ Q = Q) is
naturally isomorphic to the group C *-algebra C ∗ (ZoZ2 ). The viewpoint of Lemma
4.1.3 is to derive algebraic and geometric informations from the behaviour of the
norms (i.e., the “distance”) about the kind of algebra they generate and how looks
a unitary that transforms the symmetries into each other. We extract rules for
moving near elements with help of a path in the inner automorphisms. A mainly
operator theoretic and algebraic big overview is contained in [101], but there with
some different terminology.

Lemma 4.1.3. Let A a real or complex unital C*-algebra and p, q ∈ A projec-

tions.
 1. TWO PROJECTIONS IN C *-ALGEBRAS 477

(i) The C*-algebra C ∗ (p, q, 1) has a character χ with |χ(p − q)| = 1 and
χ(pq) = 0, if and only if, kp − qk = 1, if and only if, k(1 − q)pk = 1 or
k(1 − p)qk = 1.
If kp − qk < 1 then kp − qk = k(1 − q)pk = k(1 − p)qk . In particular,


always kp − qk = max k(1 − q)pk , k(1 − p)qk .
In more detail: If k(1 − q)pk < 1 then the following properties (a)–(d)
are equivalent:
(a) k(1 − p)qk = 1.
(b) kp − qk = 1.
(c) There exist a projection p0 6= 0 in C ∗ (p, q) with q0 ≤ 1 − p, q0 ≤ q.
(d) There exists a character χ on C ∗ (p, q, 1) with χ(q) = 1 and χ(p) = 0.
(ii) Always Spec(p + q) ⊆ {0} ∪ [1 − kpqk, 1 + kpqk] .
In particular, if kpqk < 1 then 0 is isolated in Spec(p + q), and q ∼h
(1B −p) inside the unital C*-subalgebra B := C ∗ (p, q) ⊆ (p + q)A(p + q).
The hereditary C*-subalgebra (p + q)A(p + q) of A – generated by p + q –
is unital and coincides with the algebraic *-algebra (p + q)A(p + q) with
unit gδ (p + q) ≤ δ −1 · (p + q), where δ := (1 − kpqk)/2 and gδ (t) :=
min(1, max(0, t − δ)).
(iii) The universal unital real (or complex) C*-algebra C ∗ (P, Q, 1) generated
by two projections P, Q is naturally isomorphic to the C*-subalgebra of
C([0, π/2], M2 ), that is generated by 1 := 12 ∈ M2 , P =: e11 = [ηjk ] ∈ M2
with η11 = 1 and ηjk = 0 for (j, k) 6= (1, 1), and Q := exp(−H)P exp(H),
where H ∈ C([0, π/2], M2 ) is given by H(ϕ) := ϕ · Z (ϕ ∈ [0, π/2]) for
Z := [ζjk ] ∈ M2 with ζjk := j − k, j, k ∈ {1, 2}.
(iv) If A is unital and k(1 − q)pk < 1, then exist h ∈ C ∗ (p, q) ⊆ A and
a projection q0 ∈ C ∗ (p, q) with q0 p = 0, php = 0 = (1 − p)h(1 − p),
h∗ = −h ( 1 ), and khk = arcsin k(1 − q)pk < π/2, such that hq0 = q0 h
and
exp(−h)p exp(h) + q0 = exp(−h)(p + q0 ) exp(h) = q .
The projection q0 is non-zero if and only if k(1 − p)qk = 1.
(v) If A is unital and kp − qk < 1 then there exists h ∈ C ∗ (p, q) ⊆
(p + q)A(p + q) with h∗ = −h such that exp(−h)p exp(h) = q, khk =
arcsin kp − qk < π/2 and php = 0 = (1 − p)h(1 − p).
(vi) If kpqk < 1 and kp − qk < 1, then there exists h ∈ (p + q)A(p + q) with
h∗ = −h, php = 0 = (1 − p)h(1 − p) and khk = arcsin kp − qk such that
exp(h) ∈ (p + q)A(p + q) and q = exp(−h)p exp(h).
(old i) If kpqk < 1, then 0 is isolated in Spec(p + q). And q ∼h (1B − p) inside
the unital C*-subalgebra B := C ∗ (p, q) ⊆ (p + q)A(p + q).
(Is identical with (p + q) − ξ1M(A) invertible in M(A) for all non-zero
ξ ∈ C with 0 < |ξ| <???? for small ????. Explicit bound???? )
(old i and ii ???) There is a (norm-continuous) path [0, 1] 3 t 7→ u(t) ∈ U(A)
e with u(0) = 1
∗
and u(1) pu(1) = q, if and only if, p ∼h q in A.
1 Thus, if A is complex then (ih)∗ = ih.
478 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

(If A is unital, then a path t 7→ u(t) with u(0) = 1 and u(1)∗ pu(1) = q.
can be found in U(A) itself.)

The elementary realization C ∗ (P, Q, 1) ⊂ C([0, π/2], M2 ) of the universal unital

C *-algebra generated by two projections – as defined in Part (iii) – and its proper-
ties will be used in the proofs of Parts (iv) and (vi) and on other places, e.g. Remark
4.5.7. Compare Remark 4.5.8 for an alternative approach if A is complex.

Proof of Lemma 4.1.3. (i): The C *-subalgebra C ∗ (p, q) ⊆ A generated by

p and q is an ideal of C ∗ (p, q, 1). If p + q is not invertible in C ∗ (p, q, 1) then C ∗ (p, q)
is the kernel of the character χ0 on C ∗ (p, q, 1) with χ0 (p) = χ0 (q) = 0. Notice that
(p + q)/2 is a strictly positive contraction of the C *-algebra C ∗ (p, q).
The W*-algebras generated by two projections p∗ = p = p2 and q ∗ = q = q 2
are described in [767, chp.V, thm. 1.41]. It implies in particular the below listed
observations concerning the characters and the irreducible representations of any
unital C *-algebras C ∗ (p, q, 1) generated by two projections p, q.
It does not matter here if we consider here real or complex C *-algebras.
Real or complex unital C *-algebras C ∗ (p, q, 1) generated by (self-adjoint) pro-
jections p, q have only 1- or 2-dimensional irreducible *-representations.
The algebra C ∗ (p, q, 1) admits at most 4 different characters.
Alternatively, one can obtain the above and below listed properties of C ∗ (p, q, 1)
by straight calculation from the identities (p − q)2 p = p(p − q)2 and (p − q)2 q =
q(p − q)2 , i.e., from the fact that the positive self-adjoint contraction (p − q)2 is
always in the center of C ∗ (p, q, 1).
The selfadjoint contraction (D(p) − D(q))2 is necessarily a non-negative scalar
for each irreducible representation D : C ∗ (p, q, 1) → L(H), which implies immedi-
ately that the Hilbert space H has dimension ≤ 2. It has dimension = 2 if and only
if the projections D(p) and D(q) do not commute. The latter can only happen if the
projections D(p) and D(p) have rank one and D(pqp) = αD(p) for some α ∈ (0, 1).
The scalar α defines our parameter ϕ := arccos(α1/2 ) = arcsin kD(p − q)k that
determines D up to unitary equivalence (respectively orthogonal equivalence in the
real case).
The parameter

ϕ := arcsin kD(p(1 − q))k = arccos kD(pq)k ∈ (0, π/2)

determines all unitary equivalence classes of 2-dimensional irreducible *-represent-

ations D : C ∗ (p, q, 1) → M2 uniquely, because arccos krsk determines all pairs (r, s)
of non-commuting orthogonal projections r, s ∈ M2 (necessarily both of rank one)
up to unitary equivalence in M2 :
If r, s ∈ M2 are projections that do not commute, there exists u ∈ O(2) (re-
spectively u ∈ U(2)) such that u∗ ru = P := e11 := [ηjk ] with η11 = 1 and ηjk = 0
for (j, k) 6= (1, 1) and that u∗ su = [αjk ] with real α12 < 0.
 1. TWO PROJECTIONS IN C *-ALGEBRAS 479

It follows that there is unique ϕ ∈ (0, π/2) such that

u∗ su = Q(ϕ) := [cos(ϕ), − sin(ϕ)]> · [cos(ϕ), − sin(ϕ)] .

Thus, P − Q(ϕ) = sin(ϕ) · T where T ∗ = T = [tjk ] with t11 = sin(ϕ) = −t22 , t1,2 =
t2,1 = cos(ϕ) for some ϕ ∈ (0, π/2). It follows T 2 = 12 and (P −Q(ϕ))2 = sin(ϕ)2 12 ,
u∗ (r − rsr)u = P − P Q(ϕ)P = sin(ϕ)2 P and u∗ (1 − r)s(1 − r)u = sin(ϕ)2 (1 − P ) .
Thus, if r, s ∈ M2 are projections with rs 6= sr then 0 < kr − sk < 1 and ϕ :=
arcsin(kr − sk) ∈ (0, π/2) satisfies

sin(ϕ) = kr − sk = k(1 − s)rk = k(1 − r)sk .

Let p, q ∈ A projections with p 6= q. Since e.g. (1 − q)(p − q) = (1 − q)p, we get

that kp − qk ≥ max k(1 − q)pk , k(1 − p)qk . If k(1 − q)pk = 0 = k(1 − p)qk, then
p ≤ q and q ≤ p, i.e., kp − qk = 0.
If 0 < kp−qk < 1, then there exists a pure state ρ on C ∗ (p, q, 1) with |ρ(p−q)| =
kp − qk.
Since 0 < |ρ(p − q)| < 1, the state ρ can not be multiplicative on C ∗ (p, q), i.e.,
the corresponding irreducible representation is necessarily 2-dimensional.
Let D : C ∗ (p, q, 1) → M2 the irreducible GNS representation of C ∗ (p, q, 1) de-
fined by ρ and let λ a state on M2 with ρ = λ ◦ D. Then we get that r := D(p)
and s := D(q) satisfy

|λ(r − s)| ≤ kr − sk ≤ kp − qk = |ρ(p − q)| = |λ(r − s)| ,

i.e., kr − sk = kp − qk < 1. It leads to

k(1 − q)pk ≤ kp − qk = kr − sk = k(1 − s)rk ≤ k(1 − q)pk

and
k(1 − p)qk ≤ kp − qk = kr − sk = k(1 − r)sk ≤ k(1 − p)qk
by the above mentioned properties of non-commuting projections (r, s) in M2 .
If kp − qk = 1 then C ∗ (p − q, 1) has a character ρ with |ρ(p − q)| = 1 . Since
χ(1) = 1 and kχk ≤ 1, the character ρ extends to a state χ on C ∗ (p, q, 1) by Hahn-
Banach extension. The state χ satisfies χ(p), χ(q) ∈ [0, 1] and |χ(p) − χ(q)| = 1. It
follows that χ(p), χ(q) ∈ {0, 1} and χ(p)χ(q) = 0. In particular, 1, p and q are in
the multiplicative domain of the unital completely positive map χ from C ∗ (p, q, 1)
into R (respectively into C), because in both cases χ(p)2 = χ(p) = χ(p2 ) and
χ(q)2 = χ(q 2 ).
It follows that χ is a character on C ∗ (p, q, 1) with |χ(p) − χ(q)| = 1.
Thus, χ is a character on C ∗ (p, q, 1) with χ(p − q) ∈ {+1, −1}. This implies
that (χ(p), χ(q)) is equal to (1, 0) or to (0, 1). In particular, 1 = χ(p − pqp) or
1 = χ(p − qpq). The latter implies 1 = kp − pqpk = k(1 − q)pk2 or 1 = kq − qpqk =
k(1 − p)qk2 .
It yields that kp − qk = max(k(1 − p)qk, k(1 − q)pk) and that k(1 − p)qk =
k(1 − q)pk if kp − qk < 1.
480 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

We show the equivalence of the properties (a)–(d) in Part (i) under the pre-
assumption k(1 − q)pk < 1, in row (a)→(b)→(d)→(a)→(c)→(a), where (b)→(d) is
a consequence of the pre-assumption, and only (a)→(c) is non-trivial.
(a)⇒(b): Since (1 − p)q = −(1 − p)(p − q), 1 ≥ kp − qk ≥ k(1 − p)qk. Thus
k(1 − p)qk = 1 implies kp − qk = 1.
(b)⇒(d): Above we have seen that kp − qk = 1 implies the existence of a
character χ on C ∗ (p, q, 1) with |χ(p) − χ(q)| = 1. It follows that (χ(p), χ(q)) is
equal to (0, 1) or to (1, 0) and implies that max(χ(p), χ(q)) = 1 and χ(qp) = 0.
Since
1 > k(1 − q)pk ≥ |χ((1 − q)p)| = |χ(p)| ,
only the cases χ(p) = 0 and χ(q) = 1 remain.
(d)⇒(a): Suppose that there exists a character χ on C ∗ (p, q) with χ(p) = 0
and χ(q) = 1. Then k(1 − p)qk = 1, because

1 = 1 − χ(p) = χ(q − qpq) ≤ kq − qpqk = k(1 − p)qk2 ≤ 1 .

(a)⇒(c): Suppose that k(1 − p)qk = 1. Then kq − qpqk = k(1 − p)qk2 = 1

and the positive element qpq of the C *-algebra qC ∗ (p, q, 1)q has necessarily 0 in its
spectrum:
0 ∈ Spec(qpq, qC ∗ (p, q, 1)q) .
Since our pre-assumption is equivalent to k1 − (pqp + (1 − p))k = k(1 − q)pk2 < 1, it
follows that the positive elements T := pqp + (1 − p) and T p = pT p are invertible
in C ∗ (p, q, 1) respectively in pC ∗ (p, q, 1)p. Then S := T −1/2 ∈ C ∗ (p, q, 1) is a
(bounded) positive invertible element with

kT k−1 · 1 ≤ S 2 ≤ kT −1 k · 1 .

Since T (1 − p) = (1 − p) = (1 − p)T , the positive operator S commutes with p and

S(1 − p) = 1 − p.
The element v := qpS satisfies v ∗ v = p and r := vv ∗ ≤ q, i.e., v is a partial
isometry and r a projection in qC ∗ (p, q, 1)q.
Moreover, using that kT k−1 ≤ S 2 ≤ kT −1 k by definition of S,

kT k−1 qpq ≤ vv ∗ ≤ kT −1 kqpq .

Thus qpq ≤ r and qpq is invertible in rC ∗ (p, q, 0)r and 0 is isolated in the spectrum
of qpq in qC ∗ (p, q, 1)q with non-zero support projection q0 := q − r. The projection
q0 satisfies q0 p = 0, because T −1 pqp = p and v = qpS satisfy

pq0 p = p(q − vv ∗ )p = pqp − pqpS 2 pqp = pqp − pqpT −1 pqp = 0 .

(c)⇒(a): If p0 ∈ C ∗ (p, q) is a non-zero projection with q0 ≤ 1 − p and q0 ≤ q,

then q0 (1 − p)q = q0 . It follows that 1 ≥ k(1 − p)qk ≥ kq0 k = 1.
(ii): Spec(p + q) ⊆ {0} ∪ [1 − kpqk, 1 + kpqk] can be seen from the spectrum
of the Murray–von-Neumann equivalent 2 × 2-matrix X := [p, q]> · [p, q], because
[p, q] · [p, q]> = (p + q) ⊕ 0 in M2 (C ∗ (p, q)) . The projection Y := diag(1 − p, 1 − q)
 1. TWO PROJECTIONS IN C *-ALGEBRAS 481

satisfies Y X = 0 = XY and X + Y = 12 + T for the self-adjoint 2 × 2-matrix

T := [βjk ] with β11 = β22 = 0 and off-diagonal entries β12 := pq, β21 := qp. Thus
p + q has – up to the possible spectral values {0 , 1} – the same spectrum as 12 + T .
Clearly Spec(12 + T ) ⊆ [1 − kT k, 1 + kT k] and kT k = kpqk.
(iii): Each unital C *-algebra C ∗ (p, q, 1) generated by projections p, q is in
a natural way a quotient of the universal unital C *-algebra C ∗ (Pu , Qu , 1) with
defining relations Pu∗ = Pu = Pu2 and Q∗u = Qu = Q2u .
In particular, there is a unique unital C *-morphism Φ with Φ(Pu ) = P and
Φ(Qu ) = Q from the universal unital C *-algebra C ∗ (Pu , Qu , 1) onto the unital C *-
subalgebra C ∗ (P, Q, 1) of C([0, π/2], M2 ) generated by 1, P ∈ M2 ⊂ C([0, π/2], M2 )
and Q := exp(−H)P exp(H) ∈ C([0, π/2], M2 ), with H ∈ C([0, π/2], M2 ) as defined
in Part (iii).
The natural *-epimorphism Φ from C ∗ (Pu , Qu , 1) onto C ∗ (P, Q, 1) =
Φ(C ∗ (Pu , Qu , 1)) is faithful because C ∗ (P, Q, 1) has 4 different characters and
every irreducible 2-dimensional representation D factorizes over Φ. The factoriza-
tion property for the irreducible representations D of C ∗ (Pu , Qu , 1) can be seen by
using the parameter ϕ of the unitary equivalence class of D:

arcsin k(P − Q)|{ϕ} k = arcsin kP − Q(ϕ)k = ϕ := arcsin kD(Pu ) − D(Qu )k

for Q(ϕ) := exp(−H(ϕ))P exp(H(ϕ)) = exp(−ϕZ)P exp(ϕZ). Notice that

exp(ϕZ) =: [αjk (ϕ)] is an orientation preserving rotation with angle ϕ, i.e.,
α11 (ϕ) = α22 (ϕ) = cos(ϕ) and α21 (ϕ) = sin(ϕ) = −α12 (ϕ) .
Since C ∗ (P, Q, 1) is naturally isomorphic to the universal unital C *-algebra
C ∗ (Pu , Qu , 1) generated by two projections, each unital C *-algebra C ∗ (p, q, 1)
generated by two self-adjoint projections p, q is in a natural way a quotient of
C ∗ (P, Q, 1) by the unique unital *-epimorphism η : C ∗ (P, Q, 1) → C ∗ (p, q, 1) that
satisfies η(P ) = p and η(Q) = q. This completes the proof of Part (iii).

We add here some additional information on the realisation C ∗ (P, Q, 1) ⊂

C([0, π/2], M2 ) that can be useful to get alternative and less abstract proofs of
Parts (iv)–(vi):
There is no essential difference in the properties of the real or complex version of
the universal algebra C ∗ (P, Q, 1), i.e., we can define it as the real C *-subalgebra of
C([0, π/2], M2 (R)) or as the complex C *-subalgebra of C([0, π/2], M2 (C)) generated
by {P, Q, 1} : Simply replace R by C and the in Part (iii) defined order 4 orthogonal
operator Z with Z = −Z ∗ by the self-adjoint (unitary) symmetry −iZ = (−iZ)∗ ∈
M2 (C) to get the complex case.
The definition of (P, Q, 1) shows that C ∗ (P, Q, 1) is the C *-subalgebra of
C([0, π/2], M2 ) given by all continuous sections of the continuous field on [0, π/2]
with full fibers M2 at ϕ ∈ (0, π/2), “diagonal” fibers C ∗ (12 , P ) = C ∗ (P, 12 − P ) ⊂
M2 at ϕ = 0 and the fiber generated by {P, Q(π/2)} = {P, 12 − P } ⊂ M2 at
ϕ = π/2. I.e., the C *-algebra C ∗ (P, Q, 1) consists of all continuous matrix-valued
482 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

functions f : [0, π/2] → M2 in C([0, π/2], M2 ) that take values f (0) and f (π/2) in
the diagonal matrices of M2
An element f ∈ C([0, π/2], M2 ) is in C ∗ (P, Q), if and only if, f ∈ C ∗ (P, Q, 1)
and f (0) = αP for some α ∈ R (respectively α ∈ C). Therefore the in Part (iii)
above defined elements Z ∈ M2 , H, exp(H) ∈ C([0, π/2], M2 ) are not contained in
C ∗ (P, Q, 1), but the restrictions H|[0, ψ] are in C ∗ (P, Q)|[0, ψ] for each ψ ∈ [0, π/2).
Moreover, exp(H)|[ψ1 , ψ2 ] is in C ∗ (P, Q)|[ψ1 , ψ2 ] for 0 < ψ1 ≤ ψ2 < π/2.
The quotient C ∗ (P, Q)|[ψ, π/2] ⊂ C([ψ, π/2], M2 ) of C ∗ (P, Q) (given by restric-
tion to [ψ, π/2]) is unital for 0 < ψ ≤ π/2, because (P + Q)|[ψ, π/2] is invertible in
C([ψ, π/2], M2 ).
The ideal C0 ((0, π/2), M2 ) of C([0, π/2], M2 ) is the intersection of the kernels
of the 4 characters of C ∗ (P, Q, 1) and is strictly contained in C ∗ (P, Q), because the
C *-algebra C ∗ (P, Q) has 3 characters.
Let ψ ∈ (0, π/2) and define a function
fψ (ϕ) := (ϕ − ψ)+ := max(0, ϕ − ψ) ∈ C0 (0, π/2]+ .
The element fψ · P ∈ C ∗ (P, Q) generates the closed ideal Jψ of C ∗ (P, Q) ⊂
C([0, π/2], M2 ). The closed subspace Jψ is also an ideal of C ∗ (P, Q, 1), because
C ∗ (P, Q) is an ideal of C ∗ (P, Q, 1).
If 0 < ψ < π/2, then the quotient C ∗ (P, Q, 1)/Jψ of C ∗ (P, Q, 1) is naturally
isomorphic to
C ∗ (P, Q, 1)|[0, ψ] ⊕ (R · Q|{π/2}) ⊂ C([0, ψ] ∪ {π/2} , M2 ) ,
and the algebra C ∗ (P, Q, 1)|[0, ψ] is equal to
R · P + R · (1 − P ) + C0 ((0, ψ], M2 ) ⊂ C([0, ψ], M2 ) .
The parameter ϕ for Q(ϕ) is natural in the sense that the restrictions P = P |{ϕ} =
P |{0} and Q(ϕ) := Q|{ϕ} satisfy k(1 − Q(ϕ))P k = sin(ϕ), k(1 − P )Q(ϕ)k = sin(ϕ),
kQ(ϕ)P k = cos(ϕ) and kP − Q(ϕ)k = sin(ϕ). Thus, for all ϕ ∈ [0, π/2],
kP (1 − Q(ϕ))P k + kP Q(ϕ)P k = 1 .
E.g., the matrix (P −Q)|{ϕ} ∈ M2 is given by [αjk ] := P −exp(−ϕZ)P exp(ϕZ) and
has entries α11 := sin2 (ϕ), α12 := α21 := − sin(ϕ) cos(ϕ) and α22 := − sin2 (ϕ). It
is sin(ϕ)-times an orthogonal matrix, thus k (P − Q)|{ϕ} k = sin(ϕ) .
Since the norms of (1 − Q(ϕ))P , (1 − P )Q(ϕ) and P − Q(ϕ) are increasing and
since the norms of Q(ϕ)P are decreasing with respect to the parameter ϕ, we get
the equal value sin(ϕ) for the norms of restrictions of (1 − Q)P , (1 − P )Q and P − Q
to [0, ϕ], and that kP Q|[ϕ, π/2]k = cos(ϕ).
e := A + R · 1 ⊆ M(A) (respectively A
(iv): Recall that A e := A + C · 1) if A is
not unital and Ae := A if A is unital.

Let γ : C ∗ (P, Q, 1) → A
e denote the canonical C *-morphism from the universal
C *-algebra C ∗ (P, Q, 1) onto C ∗ (p, q, 1) ⊆ A.
e The kernel of γ is the intersection of
the kernels of ρ◦γ, where ρ runs through all irreducible representations of C ∗ (p, q, 1).
 1. TWO PROJECTIONS IN C *-ALGEBRAS 483

The natural C *-morphism

γ : C ∗ (P, Q, 1) → C ∗ (p, q, 1) ⊆ A

satisfies γ((P (1−Q)P −k(1−q)pk2 )+ ) = 0 if k(1−q)pk < 1. Since P (1−Q(ϕ))P =

sin(ϕ)2 P , this implies that the element (P (1 − Q)P − k(1 − q)pk2 )+ ∈ C ∗ (P, Q)
and the in proof of Part (iii) defined element fψ · P ∈ C ∗ (P, Q) ⊂ C([0, π/2], M2 )
generate the same ideal of C ∗ (P, Q, 1) for ψ := arcsin k(1 − q)pk .
Thus, γ factorizes over the restriction C ∗ (P, Q, 1)|([0, ϕ1 ] ∪ {π/2}) with ϕ1 :=
arcsin k(1 − q)pk < π/2 and maps P |{π/2} = P (1 − Q)P |{π/2} to zero.
The projection Q0 ∈ C ∗ (P, Q, 1)|([0, ϕ1 ] ∪ {π/2}) defined by Q0 (ϕ) = 0 on
[0, ϕ1 ] and by Q0 (π/2) = 1 − P = Q|{π/2} at {π/2} is not in the kernel of γ, if
and only if, C ∗ (p, q, 1) has a character χ with χ(p) = 0 and χ(q) = 1 . Compare
the equivalence of (c) and (d) in Part(i).
This character χ is “isolated” in the sense that the image C ∗ (p, q) of the C *-
morphism γ : C ∗ (P, Q) → A is isomorphic to a direct sum C ∗ (p, q) ∼ = B ⊕ C such
that q corresponds to q1 ⊕ 1, q0 to 0 ⊕ 1 and p to p ⊕ 0. It shows that the character
χ is given by the projection of B ⊕ C onto 0 ⊕ C ∼ = C.
The restriction H 0 := H|[0, ϕ1 ] is contained in the non-unital C *-subalgebra
C ∗ (P, Q)|[0, ϕ1 ] , the restriction exp(H 0 )|[0, ϕ1 ] of the unitary exponential exp(H 0 )
to [0, ϕ1 ] is contained in C ∗ (P, Q, 1)|[0, ϕ1 ] and

exp(−H 0 )P |[0, ϕ1 ] exp(H 0 ) = Q|[0, ϕ1 ] ≤ Q|([0, ϕ1 ] ∪ {π/2}) .

If C ∗ (p, q, 1) has a character χ with χ(p) = 0 and χ(q) = 1, then the character χ ◦ γ
on C ∗ (P, Q, 1)|[0, ϕ1 ] ∪ {π/2} is necessarily supported at the isolated one-point set
{π/2}, and is given on C ∗ (P, Q, 1)|{π/2} ∼ = diag(M2 ) by χ ◦ γ(diag(1, 0)) = 0 and
χ ◦ γ(diag(0, 1)) = 1. It follows that in this case, and only in this case, γ contains
in its support ⊆ [0, π/2] also the isolated point π/2, and P |{π/2} is in the kernel
of γ .
Thus h := γ(H 0 ) is a skew adjoint element that is contained in C ∗ (p, q).
The elements h and q0 := γ(Q0 ) ≤ q := γ(Q) fulfill the quoted properties in
Part (iv), because q0 exp(h) = exp(h)q0 = q0 , q0 ≤ q and q0 p = 0.
This completes the proof of Part (iv).
(v): By Part (i), kp−qk < 1 implies that 1 > kp−qk = k(1−p)qk = k(1−q)pk.
Thus, Part (iv) applies with khk = arcsin kp − qk and q0 = 0.
(vi): By Part (ii), the inequality kpqk < 1 implies that the algebraic *-algebra
(p + q)A(p + q) is a unital C *-algebra. Since we require also that kp − qk < 1, the
Part (v) applies to p, q ∈ (p + q)A(p + q) and the, in this case unital, C *-algebra
(p + q)A(p + q) in place of A. Then h (= −h∗ ) and the unitary exp(h) can be
defined inside the unital C *-algebra (p + q)A(p + q).
(old ?? ii,iii ??): See [692, prop. 2.2.6], or [73, prop. 4.3.3, prop. 4.6.3], or
[207, prop. IV.1.2, lem.IV.1.4].
484 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Remarks 4.1.4. Following remarks about the natural Z2 -grading α of the

universal algebra C ∗ (P, Q, 1) can be easily deduced from Lemma 4.1.3:
There is a unique grading automorphism β : C ∗ (P, Q, 1) → C ∗ (P, Q, 1) with
β(P ) := 1−P and β(Q) := 1−Q that is well-defined by universality of C ∗ (P, Q, 1).
If we use the canonical embedding C ∗ (P, Q, 1) ⊆ C([0, 1], M2 ) then β extends
to C([0, 1], M2 ) by the “odd”, not “inner” in C ∗ (P, Q, 1), grading of M2 , here by the
order 2 outer automorphism β(a)(t) := −Za(t)Z = Z −1 a(t)Z for the orthogonal
matrix Z = exp((π/2)Z) in M2 (R) of order 4, as defined in Part (iii) of Lemma
4.1.3, i.e., with Z := [ζj,k ] ∈ M2 where ζj,k := j − k, j, k ∈ {1, 2}.
The elements of degree = 0 are given by f · 12 + g · Z (f, g ∈ C[0, 1]) and the
elements of degree = 1 by f · X + g · ZX (f, g ∈ C[0, 1]) with X := diag(1, −1).
Thus, in the case of real M2 (R) and f, g ∈ C([0, 1], R) the elements of degree = 0
are just the natural images of complex-valued continuous functions t 7→ f1 (t)+if2 (t)
in its “real” interpretation.
Let B a unital C *-algebra, α ∈ Aut(B) a grading of B, i.e., α2 = idB and let
p, q ∈ B projections. Then 1 − 2p is of degree = 1, i.e., α(1 − 2p) = 2p − 1, if and
only if, α(p) = 1 − p.
Notice that k(1 − 2q) − (1 − 2p)k < 2 if and only if kp − qk < 1.
If A := C ∗ (p, q, 1) ⊂ B is the unital C *-subalgebra of B generated by p, q
that satisfy α(1 − 2p) = −(1 − 2p) and α(1 − 2q) = −(1 − 2q), then the natural
C *-morphism ϕ from C ∗ (P, Q, 1) onto C ∗ (p, q, 1) satisfies ϕ ◦ β = α ◦ ϕ.
In particular, α(C ∗ (p, q, 1)) = C ∗ (p, q, 1).
If k(1 − 2q) − (1 − 2p)k < 2 then kp − qk < 1 and there exists H ∈ C ∗ (p, q)
with α(H) = H, H ∗ = −H, kHk < π/2 and q = exp(−H)p exp(H).
The grading α|C ∗ (p, q, 1) itself is given by an inner automorphism of C ∗ (p, q, 1),
if and only if, kp − qk < 1 and kpqk < 1, i.e., both inequalities must be valid.

Remark 4.1.5. Let A a not-necessarily unital C *-algebra. Consider in the non-

unital case A + C · 1M(A) ⊆ M(A). Each u ∈ U0 (A) with πA (u) = 1 is a product
u = exp(h1 ) · exp(h2 ) · . . . exp(hn ) of exponentials of elements −h∗k = hk ∈ A. We
can here suppose that khk k < π/2. If p, q ∈ A projections, then p ∼h q in A, i.e.,
p and q are (norm-) homotopic inside the metric space of of projections in A, then
Lemma 4.1.3(v) shows that there exists a unitary u ∈ U0 (A + C · 1) ∩ (A + 1) with
u∗ pu = q. Clearly, for each u ∈ U0 (A + C · 1) the projection q := u∗ pu ∼h p in the
projections of A.
 2. ON K-THEORY OF PROPERLY INFINITE C *-ALGEBRAS 485

2. On K-theory of properly infinite C *-algebras

Recall that a projection p 6= 0 in a (not necessarily unital) C *-algebra A is

properly infinite if there are projections r ≤ p and s ≤ p such that r + s ≤ p and
r, s are both Murray–von-Neumann equivalent to p.
A unital C *-algebra E is called properly infinite if its unit element 1E is
properly infinite.
We can reduce our calculations related to K∗ -theory and KK(C; ·, ·) to the case
of properly infinite unital C *-algebras E or to C *-algebras A with properly infinite
E := M(A) by using – among others – the following lemmata and remarks.
Compare other places with next topic !?
Rewrite shorter on places with repeats. Observations (C0), (Cu) ?
Better explanation? ...
This observation ???? has a partial converse in the class of full properly infinite
projections, as the following lemma of J. Cuntz [172] shows.
Recall that projections p ∈ A are full in A if the linear span of ApA is dense
in A. Some elementary observations on properly infinite projections – considered
first by J. Cuntz [172] – are collected in the following Lemma:

Lemma 4.2.1. Let p ∈ A a properly infinite projection in a C*-algebra A, i.e.,

there exist partial isometries u, v ∈ A with

u∗ u = v ∗ v = p and uu∗ + vv ∗ ≤ p .

Let B := span(ApA) (the closed ideal of A generated by p) and q ∈ B ⊗ Mn ∼=

Mn (B) a projection, then there exist a partial isometry z ∈ Mn (B) ⊆ Mn (A) with
zz ∗ ≤ p ⊗ e11 and z ∗ z = q.
Let r := p − vv ∗ . There exist partial isometries s, t ∈ rAr with s∗ s = t∗ t = r
and s∗ t = 0.
The projection r := p − vv ∗ is a full projection in B and the elements of K0 (B)
are the classes [q] ∈ K0 (B) given by properly infinite projections q ≤ r that are full
in B and have the property that r − q is also properly infinite and full in B.
The addition can be carried out inside the classes of Murray–von-Neumann
equivalent projections of rBr = rAr itself via the operation [q1 ] + [q2 ] = [q1 ⊕ q2 ]
(where ⊕ = ⊕s,t , i.e., q1 ⊕ q2 := sq1 s∗ + tq2 t∗ ).
If, in addition, q ∈ A is also properly infinite, [q] = [p] ∈ K0 (A) and p ∈
span(AqA), then there exists a partial isometry w ∈ span(ApA) with w∗ w = p and
ww∗ = q.
In particular, full properly infinite projections p, q ∈ A satisfy [p] = [q] ∈ K0 (A)
if and only if p and q are M-vN-equivalent, i.e., if there exists a partial isometry
v ∈ A with v ∗ v = p and vv ∗ = q.
486 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

If A contains a full properly infinite projection p ∈ A then for each x ∈ K0 (A)

there exists a properly infinite full projection q ∈ A with q ≤ p and x = [q] ∈ K0 (A).

We omit the proof of Lemma 4.2.1. It is easy to see, except that for each unital
C *-algebra A with properly infinite unit element 1 ∈ A and [1] = 0 in K0 (A) there
exist isometries s, t ∈ A with s∗ t = 0 and ss∗ + tt∗ = 1A , see the “basic property”
(Cu) of full properly infinite projections in the proof of Lemma 4.2.6(i,ii). Combined
with Lemma 4.2.3 it delivers all other statements. Plus+ some extra service for the
reader concerning of pairs of near odd graded isometries and others.
If A is unital (with properly infinite unit), then a projection p ∈ A is (called)
splitting if p and 1 − p are both full and properly infinite.
Recall that A is K1 -injective, if u ∈ U(A)e and 0 = [u] ∈ K1 (A) e together
always imply that u ∈ U0 (A) , i.e., u ∼h 1 in U(A). It implies that unitaries u1
e e
e are homotopic inside U(A),
and u2 in A e if and only if, [u1 ] = [u2 ] in K1 (A).
e

There is a natural group-morphism U(E) 3 u 7→ [u] ∈ K1 (E) from the unitary

group U(E) of E into K1 (E) if E is unital. Its kernel contains the connected
component U0 (E) of the identity element 1 ∈ E.

Definition 4.2.2. A unital C *-algebra E is called K1 -surjective if the natural

group-morphism u 7→ [u] is a surjective map from U(E) onto K1 (E).
The C *-algebra E is K1 -injective if the connected component U0 (E) of 1E
in U(E) is equal to the kernel of U(E) → K1 (E), i.e., the induced map from
U(E)/U0 (E) to K1 (E) is faithful (= injective).
The C *-algebra E is K1 -bijective if E is K1 -surjective and K1 -injective, i.e.,
U(E)/U0 (E) ∼
= K1 (E) by the natural group morphism.

The K1 -bijectivity of stable coronas obtained in Proposition 4.2.15 with help

of the “squeezing” property defined in Definition 4.2.14 plays later a role in some
of our computations.
It is (Feb 2021) not known if the unital (full) free product C *-algebras O∞ ∗O∞ ,
C(S 1 ) ∗ O2 ∼
= O2 ∗ O2 or C(S 1 ) ∗ E2 are K1 -injective.
Where is a proof of C(S 1 ) ∗ O2 ∼
= O2 ∗ O2 ?
The proof of Lemma 4.2.10(i) shows that the question “Is every unital C*-
algebra with properly infinite unit K1 -bijective?” is equivalent to the question – if
the unitary U ∗ · (1 − s1 s∗1 + s1 U s∗1 ) in the unital full free product A := C(S 1 ) ∗ E2
is in U0 (A), where U denotes the canonical unitary generator of C(S 1 ) and the
s1 , s2 ∈ E2 denote the generating isometries of E2 ( 2 )
If S is an Abelian semigroup then we denote by Gr(S) the Grothendieck group
of S, i.e., the formally defined group of “formal differences” of elements of the
cancellation semi-group build by stable equivalence classes of elements in S. The
2 This was also mentioned in a talk of B. Blackadar in Barcelona (2007). Compare also

Lemma 4.2.10(iii).
 2. ON K-THEORY OF PROPERLY INFINITE C *-ALGEBRAS 487

natural semigroup morphism from S into Gr(S) is given by x 7→ [x] and maps
onto the cancellation semi-group [S] ⊆ Gr(S), where [x] ⊆ S denotes the stable
equivalence class of x in S given by
[x] := { y ∈ S ; ∃ s ∈ S with x + s = y + s } .
In general the natural injective additive morphism from the Abelian cancellation
semigroup [S] of S into the Grothendieck group Gr(S) is not surjective, [S] is not
identical with Gr(S), i.e., [S] is not itself a group.
The proof of the following Lemma 4.2.3 is straight calculation and is left to the
reader. It gives some sufficient conditions for cases where [S] = Gr(S), i.e., where
[S] is itself is a group .

Lemma 4.2.3. Let S a commutative semigroup and denote, for z ∈ S, by

S(z) ⊆ S the set of all elements x ∈ S with the property that there exists y ∈ S
with x + y = z .

(i) If e ∈ S satisfies e + e = e, and that for every x ∈ S there exists some

y ∈ S with x + y = e, i.e., S(e) = S, then e is the only element of S with
this property and the sub-semigroup G := S + e is a subgroup of S.
An element x ∈ S is in G, if and only if, x absorbs e, i.e., if x = x+e.
The map x ∈ S 7→ x + e ∈ S is a semigroup homomorphism that
extends to a natural isomorphism from G onto the Grothendieck group
Gr(S) of S in the following manner:
By universality of Gr(S) there is a well-defined natural group homo-
morphism ϕ from Gr(S) into G, given by ϕ([x]) := x + e on the classes
[x] ∈ Gr(S) for x ∈ S.
This homomorphism is a group isomorphism from Gr(S) onto G with
inverse isomorphism x + e 7→ [x + e] ∈ Gr(S).
(ii) More generally, if S contains a element z0 ∈ S such that for each x ∈ S
there exists y ∈ S with x + y = z0 , i.e., S = S(z0 ), then there exists an
element y0 ∈ S with 2z0 + y0 = z0 .
The element e := z0 +y0 and pair (S, e) have the in Part (i) considered
properties.
(iii) If z0 ∈ S is an element that satisfies only 2z0 ∈ S(z0 ), i.e., if there
exists y0 ∈ S with 2z0 + y0 = z0 , then e := e(z0 ) := z0 + y0 satisfies
S(e(z0 )) = S(z0 ) and e + e = e.
The subset S(z0 ) of S is a sub-semigroup of S such that G(e(z0 )) :=
S(z0 ) + e(z0 ) ⊆ S(z0 ) and G(e(z0 )) is a subgroup of S, that is natural
isomorphic to the Grothendieck group of S(z0 ).
(iv) If z0 , z1 ∈ S satisfy z0 ∈ S(z1 ) and 2zj ∈ S(zj ) for j = 0, 1, then
e(z1 ) = e(z0 ) + e(z1 ) and the mapping
y ∈ S(z0 ) + e(z0 ) 7→ y + e(z1 ) ∈ S(z1 ) + e(z1 )
is a group homomorphism from G(e(z0 )) into G(e(z1 )).


488 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

There are examples of group homomorphism of the type described in Part (iv)
of Lemma 4.2.3 that are neither injective nor surjective.

An element a ∈ A+ of a (not-necessarily unital) C *-algebra A will be called

full in A if span(AaA) is dense in A, i.e., if aAa is full hereditary C *-subalgebra
of A. (For example, all strictly positive elements of A are full in A.)
Recall that the unit element 1E of a unital C *-algebra E is properly infinite in
E, if and only if, there are isometries s1 , s2 ∈ E with s∗1 s2 = 0. Then we say that
E is a properly infinite unital C *-algebra.
We let p[−1] := 1−s1 s∗1 −s2 s∗2 . This projection becomes zero if and only if s1 , s2
are the canonical generators of a unital copy of O2 in E. It holds [p[−1] ] = −[1] in
K0 (E).
Then a generalized variant of the Cuntz addition using isometries s1 , s2
with s∗1 s2 = 0 is defined by

a ⊕ b := a ⊕s1 ,s2 b := s1 as∗1 + s2 bs∗2 for a, b ∈ E .

If we consider unitaries u, v ∈ U(E) then we can use the corrected version of

generalized Cuntz addition given with p[−1] := 1 − s1 s∗1 − s2 s∗2 by

u ⊕0 v := u ⊕0s1 ,s2 v := s1 us∗1 + s2 vs∗2 + p[−1] .

We say that a projection p ∈ E in a unital C *-algebra E is splitting if p and 1 − p

are both properly infinite projections and are full elements of E.
The definitions of proper infiniteness and fullness yield that this property is
equivalent to the property that there exist isometries t1 , t2 ∈ E with t1 t∗1 ≤ p and
t∗2 t2 ≤ 1 − p.
The notion of splitting projections generalizes the notion of a proper projec-
tion p defined by the property that p and 1 − p are both MvN-equivalent to 1.
Most of the following results on splitting projections have been obtained before by
J. Cuntz and N. Higson for the special case of proper projections in place of the
more general “splitting” projections (introduced by B. Blackadar, 2007).
It is easy to see, e.g. using Lemma 4.1.3(v), that projections p, q ∈ E are
homotopic inside the subset of projections in E, if and only if, there is a unitary
u ∈ U0 (E) with u∗ pu = q.

Remark 4.2.4. The Halmos unitary U (c) in M2 (E), build from a contraction
c ∈ E, is an “orientation-preserving rotation” given by the unitary matrix (respec-
tively by the “special orthogonal” matrix if E is a real C *-algebra) U (c) := [ajk (c)]
with entries a11 (c) := c, a12 (c) := −(1 − cc∗ )1/2 , a21 (c) := (1 − c∗ c)1/2 and
a22 (c) := c∗ .
Each Halmos unitary U (c) is in U0 (M2 (E)) because U (1) = 12 and c 7→ U (c) is
uniformly Hölder continuous on the closed unit ball of E with very rough estimate
(?????? see commented lines !!!?)
 2. ON K-THEORY OF PROPERLY INFINITE C *-ALGEBRAS 489

kU (c1 ) − U (c2 )k ≤ (2kc1 − c2 k)1/2 + kc1 − c2 k ≤ 3kc1 − c2 k1/2 (2.1)

It shows that each Halmos unitary U (c) is in U0 M2 (C ∗ (c, 1)) and has an

exponential length cel(U (c)), – i.e., the “geodesic” distance of U (c) from 12 inside
the metric space U0 (M2 (E)) –, of value cel(U (c)) < 3π/2 .
If one considers the very special case E := C([0, 1]) then it cuts down to the
case of E := C. And in this case one can use the estimates of cel(U (c)) for c ∈ C
with |c| ≤ 1.
check this estimate 3π/2 again!!
How long is the path t ∈ [0, 1] 7→ U (tc + (1 − t)) in U0 (E)
for fixed c ∈ E with c ∈ E with kck < 1?
How good is the estimate for positive contraction?
Very important:
Give precise citation/reference to definition of cel of a unitary
or change and DEFINE notation !!!
In the special case of U (T ) for an isometry T ∈ E we can take a new
parametrization ϕ ∈ [0, π/2] 7→ U (cos(ϕ) · T ) ∈ M2 (E) and get a continuously
differentiable curve of length ≤ π/2.
For h∗ = h ∈ E with 0 ≤ h ≤ π · 1, the matrix Z = [zij ] ∈ M2 with zij := i − j,
and Xh := h ⊗ Z ∈ M2 (E) holds exp(Xh ) = U (cos(h)). Therefore we have changed
the original definition of P. Halmos [352] into our formula for U (c) to make it fitting
to other terminology used for the terms in our estimates.
The original definition given by Halmos was diag(1, −1) · U (c) · diag(1, −1), if
we express it with our definition of U (c). Some authors define the matrix U (c) ·
diag(1, −1) as Halmos unitary.

Compare above text about estimates and arc-length

with below blue - including blue text in proof of Lemma 4.2.6!!!
The estimate (2.1) follows from

1/2
kU (c1 )−U (c2 )k ≤ kc1 −c2 k+ max{kc∗1 c1 −c∗2 c2 k, kc1 c∗1 −c2 c∗2 k} ≤ 3kc1 −c2 k1/2 ,

where we have used that ka1/2 − b1/2 k ≤ ka − bk1/2 for a, b ∈ E+ . More precise
is kU (c1 ) − U (c2 )k ≤ γ(kc1 − c2 k) with the function γ(ξ) := ξ + (2ξ)1/2 for ξ ∈
[0, 2]. This estimate can be applied to the special pairs of contractions (c1 , c2 ) :=
((1/2)c, 0) and (c, (1/2)c). It shows that U (c) = U (0) exp(H1 ) exp(H2 ) for H1 , H2 ∈
E with Hk∗ = −Hk and kHk k < π/2. Since U (0) is a 90◦ rotation, it implies that
U (c) has exponential length bounded by 3π/2.
But this estimate (with this parametrization) does not imply that the arc
[0, 1] 3 t 7→ U (tc) is a rectifiable path (of finite length), that requires Lips-
chitz continuity (and not only this obtained Hölder continuity). In more detail:
kU (tc) − U (sc)k ≤ (t − s) + (t2 − s2 )1/2 for 0 ≤ s < t ≤ 1.
490 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Recall here | exp(iϕ)−1| ≤ x < 2 for ϕ ∈ [−π, π] if and only if 2−x2 ≤ 2 cos(ϕ),
i.e., |ϕ| ≤ arccos(1 − x2 /2) < π.
Thus, kU (tc) − U (0)k ≤ 2t < 2 for t ∈ [0, 1), implies that there is a unique
H ∈ E with U (tc) = U (0) exp(H), H ∗ = −H, and kHk ≤ arccos(1 − 2t2 ) < π
and U (tc) = U (0) exp(H) if t ∈ [0, 1), i.e., kHk ≤ arccos(1/2) < π/2 if t = 1/2. If
δ ∈ [0, 1), then

kU (c) − U ((1 − δ)c)k ≤ δ + (1 − (1 − δ)2 )1/2 ≤ 3δ < π ,

√
and we get for δ := 1/2, that kU (c) − U ((1/2)c)k ≤ x with 1 < x := (1 + 3)/2 <
√ √
2 and 1 − x2 /2 = (2 − 3)/4 > 0 and there exists K ∈ E with K ∗ = −K,
√
U (c) = U ((1/2)c) exp(K) and kKk ≤ arccos((2 − 3)/4) < π/2.
(It seems that the Hölder continuous map t ∈ [0, 1] → U (t · c) ∈ U0 (M2 (E))
itself is not always rectifiable, because it is not Lipschitz continuous for arbitrary
contractions c ∈ E.)
In case of U (T ) for an isometry T ∈ E we can take a parametrization ϕ ∈
[0, π/2] 7→ U (cos(ϕ) · T ) ∈ M2 (E) and get a continuously differentiable curve of
length ≤ π/2, because, for s ∈ [0, 1],

U (s · T ) = diag(T, 1)U (s · 1) diag(1, T ∗ ) + X ,

where X = [xjk ] is given by x12 := 1 − T T ∗ and x11 = x21 = x22 := 0.

We use often the following general equation for isometries t ∈ E and unitaries
v ∈ U(E) :
[(1 − tt∗ ) + tvt∗ ] = [v] in K1 (E) . (2.2)
Indeed, the diagonal matrices diag(v, 1) and diag (1 − tt∗ ) + tvt∗ , 1 are unitary

and are unitary equivalent in M2 (E) by the Halmos unitary U (t) ∈ U0 (M2 (E)),
i.e.,
U (t) diag(v, 1) = diag (1 − tt∗ ) + tvt∗ , 1 U (t) .

Remark 4.2.5. Recall that C *-algebra E has a properly infinite unit, if and
only if, E contains two isometries t1 , t2 with orthogonal ranges, i.e., with relations
t∗i tj = δi,j 1E .
Obviously the elements sn := (t1 )n−1 t2 satisfy s∗k s` = δk,` 1E , i.e., generate a
copy of O∞ that is unitally contained in E , cf. the proof of Proposition 4.2.11.
It implies that for every contraction a ∈ E there exists a unitary u ∈ U0 (E)
such that s∗1 us2 = a with the isometries s1 , s2 are taken from this fixed copy of E3 .
It shows that any separable C *-algebra E with a properly infinite unit element
is a quotient of the unital free product C(S 1 )∗E2 , because all separable C *-algebras
with properly infinite unit are singly generated by Proposition B.17.2.
Indeed, we can use the Halmos unitary U (a) ∈ M2 (a) and the projection
p := 1 − s1 s∗1 − s2 s∗2 ∈ E. Then the unitary u := p + [s1 , s2 ]U (a)[s2 , −s1 ]∗ ∈ E
has this properties, because U (a) ∈ U0 (A) is connected in U0 (A) to U (0) and
[s1 , s2 ]U (0)[s2 , −s1 ]∗ = 1 − p.
 2. ON K-THEORY OF PROPERLY INFINITE C *-ALGEBRAS 491

Lemma 4.2.6. Let E a unital C*-algebra with properly infinite unit 1E . Then
there exist isometries s1 , s2 ∈ E with orthogonal ranges, i.e., with s∗1 s2 = 0, such
that s1 and s2 satisfy the following non-degeneracy condition (ND):

(ND) The projection

p[−1] := 1 − s1 s∗1 − s2 s∗2 = 1 − (1 ⊕ 1)

is either full and properly infinite or p[−1] = 0.

The s1 , s2 can be always found such that p[−1] is full and properly infinite, s1 , s2
with p[−1] = 0 exists if and only if [1] = 0 in K0 (E).
The (generalized) Cuntz addition ⊕s1 ,s2 defined by isometries s1 , s2 ∈ E, that
satisfy s∗1 s2 = 0 and condition (ND), has following properties and connections to
unitary operators and splitting projections:

(o) Commutativity and Associativity of ⊕ up to unitary homotopy:

There exist Uc , Ud ∈ U0 (C ∗ (s1 , s2 )) ⊆ U0 (E) with exponential length
cel(Uc ) ≤ π/2 and cel(Ud ) ≤ π
Look up definition of ‘‘cel’’ in published papers, change
notation?
Give Citation!!!
that satisfy p[−1] Uc = p[−1] = p[−1] Ud , Uc∗ (a ⊕ b)Uc = (b ⊕ a) , for
a, b ∈ E, and

Ud∗ a ⊕s1 ,s2 (b ⊕s1 ,s2 c) Ud = (a ⊕s1 ,s2 b) ⊕s1 ,s2 c

for a, b, c ∈ E .

(i) The map (p, q) 7→ p ⊕s1 ,s2 q is a commutative and associative operation
up to unitary equivalence by unitary elements u ∈ U0 (E), i.e., up to ho-
motopy in the projections of E. It is compatible with unitary equivalence,
homotopy and Murray–von Neumann equivalence.
If p, q ∈ E are projections then diag(p ⊕ q, 0) is MvN-equivalent to
diag(p, q) in M2 (E). The projection p ⊕ q is MvN-equivalent to p + q if
pq = 0.
In particular, [p ⊕ q] = [p] + [q] ∈ K0 (E) for projections p, q ∈ E, and
[p ⊕ q] = [p + q] if pq = 0.
(ii) Suppose that p, q ∈ E are projections and that there exist isometries s, t ∈
E with ss∗ ≤ p and tt∗ ≤ q, i.e., p and q are full and properly infinite.
Then p and q are Murray–von Neumann equivalent in E if and only
if [p] = [q] ∈ K0 (E).
If p and q are splitting projections, – i.e., if each of the projections p,
1 − p, q and 1 − q are full and properly infinite –, then p and q are unitary
equivalent in E if and only if [p] = [q] ∈ K0 (E).
If p, q ∈ E are unitary equivalent projections and p is full and properly
infinite then there exists u ∈ U(E) with 0 = [u] ∈ K1 (E) and u∗ pu = q.
492 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Any projection q ∈ Mn (E) is MvN-equivalent to a projection q 0 ∈ E

such that q 0 ≤ p for some splitting projection p ∈ E ( 3 ).
(iii) The unitary equivalence classes [p]u of splitting projections p ∈ E build
a group with generalized Cuntz addition [p]u + [q]u := [p ⊕ q]u , and the
natural map [p]u → [p] ∈ K0 (E) defines an isomorphism from this group
onto K0 (E).

(iv) (a) (cf. [94, prop. 2.5]) If p and q are full and properly infinite projections
with [p] = [q] ∈ K0 (E), and if there exists an isometry t ∈ E with
t∗ (p + q)t = 0 , then there exists a unitary u ∈ U0 (E) such that u∗ pu = q,
i.e., p and q are homotopic in the projections of E.
(b) Special case: Two projections p, q ∈ E represent the same element
[p] = [q] of K0 (E), if and only if, there exists a unitary u ∈ U0 (E) such
that u∗ (p ⊕ 1 ⊕ 0)u = q ⊕ 1 ⊕ 0 . (4)
(v) Let u ∼h v denote homotopy of u, v ∈ U(E) inside the unitary group U(E)
of E, and let U0 (E) the connected component of 1 in U(E).
The natural map u 7→ [u] from the group of unitary operators U(E)
of E into K1 (E) is an epimorphism, i.e., E is K1 -surjective. It has
following properties – expressed with the “corrected” Cuntz sum u ⊕0 v :=
(u ⊕ v) + p[−1] on the unitary group:
(1) [u ⊕0 v] = [u] + [v] = [uv] in K1 (E),
(2) [u] = [v] ∈ K1 (E), if and only if, u ⊕0 1 ∼h v ⊕0 1.
In particular, u ⊕0 1 ∈ U0 (E) if [u] = 0 ∈ K1 (E).
(Recall u ⊕0 1 := s1 us∗1 + (1 − s1 s∗1 ) .)
(3) The unitary operators u∗ ⊕0 u are in U0 (E).
(4) u ⊕0 v ∼h vu ⊕0 1 ∼h v ⊕0 u .
(5) Let p ∈ E a projection such that 1 − p is full and properly infinite.
If u ∈ U(pEp) is a unitary with [u + (1 − p)] = 0 in K1 (U(E)) then
u + (1 − p) ∈ U0 (E).
If p ∈ E is moreover a splitting projection, then η : pEp ,→ E defines
an isomorphism η∗ from K∗ (pEp) onto K∗ (E).
It satisfies η1 ([u]) = [u + (1 − p)] for u ∈ U(pEp) and η0 ([q]) = [q]
for projections q ∈ pEp.
In particular, for splitting p ∈ E and u ∈ U(pEp) holds:

0 = [u] ∈ K1 (pEp) ⇐⇒ u + (1 − p) ∈ U0 (E) .

(6) If z ∈ E is a partial isometry with z 2 = 0 then the “rotation” unitary

R(z) := z − z ∗ + (1 − z ∗ z − zz ∗ ) = exp((π/2)(z − z ∗ ))

is in U0 (E). If E is a complex C*-algebra, then the “symmetry”

S(z) := z + z ∗ + (1 − z ∗ z − zz ∗ )

3 Identify p ∈ E with p ⊗ e ∈ E ⊗ M ∼ M (E).

11 n = n
4 (1 ⊕ 1) ⊕ 0 and 1 ⊕ 1 are MvN-equivalent but are not unitarily equivalent.
 2. ON K-THEORY OF PROPERLY INFINITE C *-ALGEBRAS 493

is in U0 (E).
It holds R(z)∗ aR(z) = zaz ∗ = S(z)aS(z) for all a ∈ z ∗ zEz ∗ z.
(vi) (cf. [94, lem. 2.4(ii)]. ) Let p ∈ E a splitting projection in E.
If u1 , u2 ∈ U(E) satisfy [u1 ] = [u2 ] ∈ K1 (E) and kuk p − puk k < 1 for
k = 1, 2, then u1 ∼h u2 in U(E).
Special cases are:
(α) u ∈ U0 (E) if kup − puk < 1 and 0 = [u] ∈ K1 (E).
Thus, u ∈ U0 (E) if up = pu and [u] = 0 in K1 (E) .
COMPARE, also clever:
(iv,old): If u ∈ U(A), p ∈ A is a splitting (!!!) projection, and
kup − puk < 1, then there exists a unitary v ∈ U(pAp) such that u is
homotopic to v + (1 − p) in U(A).
(Proof: iv,old): If w is unitary with wp = pw, and z ∈ A with
zz ∗ = (1 − p) and z ∗ z ≤ p, then p + (1 − p)w is unitary and is
homotopic to z ∗ wz + (p − z ∗ z) + (1 − p) (using that C ∗ (z, 1 − p) ∼
= M2
∗
has unit z z + (1 − p)).
Thus, w ∼h v + (1 − p) with v := pw(z ∗ uz + (p − z ∗ z)) ∈ U(pAp).
If kup − puk < 1, then ku − ak < 1 for a := pup + (1 − p)u(1 − p).
The polar decomposition a = w(a∗ a)1/2 in pAp + (1 − p)A(1 − p)
gives a unitary w ∈ A with −1 6∈ Spec(w∗ u) and wp = pw. Thus,
u ∼h w ∼h v + (1 − p) for suitable v ∈ U(pAp) .
(β) If t1 , t2 ∈ E are isometries with kt∗1 t2 k < 1 and if u ∈ U(E) satisfies
kt∗1 ut2 k < 1 and [u] = 0 ∈ K1 (E), then u ∈ U0 (E).
(v,old) FROM: lem:old.A.K-basic1:
If u ∈ U(A) with 0 = [u] ∈ K1 (A), and if there exists a splitting
projection p ∈ A with kup − puk < 1, then u ∈ U0 (A).
In particular, A is K1 -injective if this is the case for every u ∈ U(A)
with 0 = [u] ∈ K1 (A).
[Above is now contained in in new (vi,β)!!!]
(v,old): Let u ∈ U(A) with 0 = [u] ∈ K1 (A), and suppose that
there is a splitting projection p ∈ A with kpu − upk < 1. By (iv),
u ∼h v+(1−p) for some unitary v ∈ U(pAp). It follows [v+(1−p)] =
[u] = 0 in K1 (A). Since [v + (1 − p)] = 0, there exists n ∈ N such
that (v + (1 − p)) ⊕ 1n ∈ U0 (Mn+1 (A)), i.e., there is a path W (t) ∈
U(Mn+1 (A)) with W (0) = 1n+1 and W (1) = (v + (1 − p)) ⊕ 1n . Since
1−p is full and properly infinite in A, the projection (1−p)⊕0n is full
and properly infinite in Mn+1 (A). Thus, there is a partial isometry
Z ∈ Mn+1 (A) with Z ∗ Z = (1 − p) ⊕ 1n and ZZ ∗ ≤ (1 − p) ⊕ 0n . Let
w(t) := (p+Z)W (t)(Z ∗ +p)+((1−p)⊕0n −ZZ ∗ ). Then w(t)∗ w(t) =
1 ⊕ 0n = w(t)w(t)∗ , w(0) = 1 ⊕ 0n and w(1) = (v + (1 − p)) ⊕ 0n .
If we naturally identify A with the corner A ⊕ 0n of Mn+1 (A), then
w(t) becomes a continuous path in U(A) that connects v +(1−p) with
1. Thus, u ∼h 1.
494 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

(vii) Independence of ⊕ from the isometries s1 , s2 ∈ E with ⊕ := ⊕s1 ,s2 and

non-degeneracy property (NP) up to unitary equivalence:
Suppose that t1 , t2 ∈ E are isometries with orthogonal ranges such
that q[−1] := 1 − t1 t∗1 − t2 t∗2 is full and properly infinite if p[−1] 6= 0 and is
q[−1] = 0 if p[−1] = 0.
Then there exists a unitary U ∈ U(E) such that U sk = tk (k = 1, 2),
and U (a ⊕s1 ,s2 b)U ∗ = a ⊕t1 ,t2 b and U p[−1] U ∗ = q[−1] for a, b ∈ E.
If p[−1] and q[−1] are full and properly infinite, then the unitary U ∈
U(E) can be chosen such that [U ] = 0 in K1 (E).
In the cases where p[−1] 6= 0 and E is K1 -injective, or where there
exists a full and properly infinite projection r ∈ E with r ≤ p[−1] and
r ≤ q[−1] , one can find U ∈ U0 (E) with U sk = tk .

Before we start with the proofs of the part ???? of Lemma ?? here hint of one
of its limited applicability: ?????
This comes from old appendix!!! Transfer old Lemma partly into
other Lemmata. Trash remains.
Lemma 4.2.7. (v) FROM: lem:old.A.K-basic1:
If u ∈ U(A) with 0 = [u] ∈ K1 (A), and if there exists a splitting
projection p ∈ A with kup − puk < 1, then u ∈ U0 (A).
In particular, A is K1 -injective if for every u ∈ U(A) with 0 = [u] ∈
K1 (A) there exists a splitting projection p ∈ A with kup − puk < 1, then
u ∈ U0 (A).
(v): Let u ∈ U(A) with 0 = [u] ∈ K1 (A), and suppose that there is a
splitting projection p ∈ A with kpu − upk < 1. By (iv), u ∼h v + (1 − p)
for some unitary v ∈ U(pAp). It follows [v + (1 − p)] = [u] = 0 in K1 (A).
Since [v + (1 − p)] = 0, there exists n ∈ N such that (v + (1 − p)) ⊕ 1n ∈
U0 (Mn+1 (A)), i.e., there is a path W (t) ∈ U(Mn+1 (A)) with W (0) = 1n+1
and W (1) = (v + (1 − p)) ⊕ 1n . Since 1 − p is full and properly infinite
in A, the projection (1 − p) ⊕ 0n is full and properly infinite in Mn+1 (A).
Thus, there is a partial isometry Z ∈ Mn+1 (A) with Z ∗ Z = (1 − p) ⊕ 1n
and ZZ ∗ ≤ (1 − p) ⊕ 0n . Let w(t) := (p + Z)W (t)(Z ∗ + p) + ((1 − p) ⊕
0n − ZZ ∗ ). Then w(t)∗ w(t) = 1 ⊕ 0n = w(t)w(t)∗ , w(0) = 1 ⊕ 0n and
w(1) = (v +(1−p))⊕0n . If we naturally identify A with the corner A⊕0n
of Mn+1 (A), then w(t) becomes a continuous path in U(A) that connects
v + (1 − p) with 1. Thus, u ∼h 1.
(vi) If, for any two splitting projection p, q ∈ A, there are splitting projections
r, s ∈ A with r ≤ p, s ≤ q and krsk < 1, then all splitting projections with
same class [p] = [q] in K0 (A) are homotopic.
(vi): If A does not contain splitting projections then (vi) is true.
Suppose that A contains splitting projections. (Then the unit of A is
properly infinite, and if S, T ∈ A are isometries with S ∗ T = 0, then SS ∗
and T T ∗ are splitting projections.)
Let p, q ∈ A splitting projections with [p] = [q] in K0 (A).
 2. ON K-THEORY OF PROPERLY INFINITE C *-ALGEBRAS 495

By assumptions, there are splitting projections r ≤ p and s ≤ q with

krsk < 1. Since, r and s are properly infinite and full, there exists partial
isometries y, z ∈ A with y ∗ y = p, r1 := yy ∗ ≤ r, z ∗ z = q, and s1 :=
zz ∗ ≤ s. It implies, that r1 and s1 are splitting projections, and that
kr1 s1 k ≤ krsk < 1.
Since [p] = [q], we get [r1 ] = [s1 ]. Thus, the properly infinite projec-
tions r1 and s1 are MvN-equivalent (cf. [172]), i.e., there is x ∈ A with
x∗ x = r1 and xx∗ = s1 . Now parts (iii) gives that r1 ∼h s1 .
Since p is splitting, 1 − p is full and properly infinite, and there is
w ∈ A with w∗ w = p and ww∗ ≤ 1 − p. It follows ww∗ ∼h p by part (iii).
Part (iii) applies also to ww∗ = (yw∗ )∗ (yw∗ ) and r1 = (yw∗ )(yw∗ )∗ , and
gives ww∗ ∼h r1 . Thus, p ∼h r1 . Similar arguments show q ∼h s1 . It
follows p ∼h q.
(vii) If A has a properly infinite unit, then A is K1 -injective, if and only if, all
splitting projections with same class [p] = [q] in K0 (A) are homotopic.
(viii) ????? For every isometries s and t with orthogonal ranges (i.e., s∗ t = 0)
and every unitary u ∈ U(A) the projections tt∗ and u∗ tt∗ u are homotopic
inside the projections of A, noted by tt∗ ∼h u∗ tt∗ u.
(Requires K1 -injectivity !!!???)
(ix) For every isometries s and t with orthogonal ranges (i.e., s∗ t = 0) and
every unitary u ∈ U(A) the projections tt∗ and u∗ tt∗ u are homotopic
inside the projections of A, – noted by tt∗ ∼h u∗ tt∗ u –, for every unitary
u ∈ A , if and only if, A is K1 -injective.

Remark 4.2.8. The Part (v,5) of Lemma 4.2.6 has to be considered with some
care, because it can happen for some u ∈ U(pEp) that u + (1 − p) ∈ U0 (E) but
0 6= [u] ∈ K1 (pEp) if the projection p is not full in E, e.g. :
Let E = M(C(S 1 ) ⊗ K), u0 ∈ U(C(S 1 )) the canonical generator of C(S 1 ),
p = 1 ⊗ e11 ∈ C(S 1 ) ⊗ K. Then pEp ∼
= C(S 1 ). If u := u0 ⊗ e11 , then u + (1 − p) ∈
U0 (E) = U(E) by the generalized Kuiper theorem [180], but 0 6= [u] ∈ K1 (pEp).

Proof. (vii): By assumption, the unit 1 is properly infinite in A, i.e., there

are isometries s, t ∈ A with s∗ t = 0. Thus A contains a splitting projection p,
e.g. p := ss∗ .
Suppose that all splitting projections with same class in K0 (A) are homotopic,
and let u ∈ U(A) with [u] = 0 in K1 (A). The projection u∗ pu is again splitting, and
is MvN-equivalent to p. Thus, [p] = [u∗ pu] in K0 (A), and p ∼h u∗ pu. By part(ii),
there is v ∈ U0 (A) with v ∗ u∗ puv = p. We have [uv] = [u] + [v] = 0. The arguments
in the proof of part (vi) show that uv ∈ U0 (A). Thus u ∼h 1.
Suppose, conversely, that A is K1 -injective. Let p, q ∈ A splitting projections
with [p] = [q] in K0 (A). It follows [1−p] = [1]−[p] = [1−q]. Since p, q, 1−p and 1−q
are all full and properly infinite (by definition of splitting projections), it follows
from “an observation in [172]”, that p ∼ q and 1 − p ∼ 1 − q (WvN-equivalence).
496 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Thus, there exists a unitary W ∈ A with W ∗ pW = q. Since 1 − p is full and

properly infinite, there exists an isometry s ∈ A with ss∗ ≤ 1 − p. The projection
1 − ss∗ satisfies [1 − ss∗ ] = 0. Thus, the properly infinite full projections 0 ⊕ 1
and (1 − ss∗ ) ⊕ 1 are MvN-equivalent in M2 (A), and there is a partial isometry
Z ∈ M2 (A) with Z ∗ Z = 0 ⊕ 1 and ZZ ∗ = (1 − ss∗ ) ⊕ 1. Let V := (s ⊕ 0) + Z,
then V is unitary and ((1 − ss∗ ) + sW s∗ ) ⊕ 1 = V (W ⊕ 1)V ∗ . In particular,
[(1 − ss∗ ) + sW s∗ ] = [W ] in K1 (A). The operator u := ((1 − ss∗ ) + sW ∗ s∗ )W is
unitary, satisfies u∗ pu = q, and [u] = [W ] − [(1 − ss∗ ) + sW s∗ ] = 0 in K1 (A). Since
A is K1 -injective, we get u ∈ U0 (A), i.e., p ∼h q.
(ix): If s, t ∈ A are isometries with s∗ t = 0, then p := tt∗ is a splitting
projection. Now suppose that u ∈ U(A) satisfies 0 = [u] ∈ K1 (A), and that
p ∼h u∗ pu. Then, by part (ii), there is w ∈ U0 (A) with p = w∗ (u∗ pu)w, i.e.,
vp = pv for the unitary v = uw. We get [v] = [u] + [w] = 0, because, [u] = 0 = [w].
It follows v ∈ U0 (A) by part (iv). Hence, u = vw∗ ∈ U0 (A).
If A is K1 -injective and u ∈ A unitary, then u∗ tt∗ u = v ∗ tt∗ v ∼h tt∗ for v :=
((1 − ss∗ ) + su∗ s∗ )u, because [v] = 0 in K1 (A), which implies v ∈ U0 (A) . 

Proof. The Lemmata 4.2.1 and 4.2.6 are also valid for real C *-algebras E.
Some definitions have to be done with some care, see e.g. proof of Part (v,6) or
proofs in preliminary Observation (o).
Suppose that 1E is properly infinite, then there are isometries t1 , t2 ∈ E with
t∗1 t2 = 0. We can take the isometries s1 := t1 and s2 := t2 t1 , then s∗1 s2 = 0 and

p[−1] := 1 − (s1 s∗1 + s2 s∗2 ) ≥ t22 (t22 )∗

is full and properly infinite with [p[−1] ] = [1] − 2[1] = −[1] in K0 (E). Thus here
s1 := ψ(T1 ) and s2 := ψ(T2 ) come from from a unital C *-morphism from E3 :=
C ∗ ( T1 , T2 , T3 ; Tj∗ Tk = δj,k 1 ) into E.
The more special case is where we can find isometries t1 , t2 ∈ E with t1 t∗1 +
t2 t∗2= 1 then [1] = [t1 t∗1 ] + [t2 t∗2 ] = 2[1] in K0 (E), i.e., [1] = 0 in K0 (E), and the
s1 := t1 , s2 := t2 come from a unital C *-morphism of O2 := C ∗ ( T1 , T2 ; Tj∗ Tk =
δj,k 1, T1 T1∗ + T2 T2∗ = 1 ) . This is exactly the case where [1] = 0 in K0 (E): Then
1 − t2 t∗2 is a properly infinite and full projection in E with 0 = [1] = [1 − t2 t∗2 ] ∈
K0 (E). By the below proved Part (ii), or by [172, thm. 1.4], there is an isometry
s1 ∈ E with s1 s∗1 + t∗2 t2 = 1. Let s2 := t2 and get p[−1] = 0 for this s1 , s2 , that
generate copy of O2 .
Recall that MvN-equivalence of projections p, q ∈ Mn (A) means that there is
a partial isometry v ∈ Mn (A) such that v ∗ v = p and vv ∗ = q. The equivalence
definition does not require that A is unital.
Part (i) can be seen from the definitions regarding the arguments in preliminary
Part (o) of our proof and Lemmata 4.1.3 and 4.2.1.
Part (ii) is contained essentially in [172, thm. 1.4], which says – in our termi-
nology – that the Murray–von-Neumann equivalence classes of full properly infinite
 2. ON K-THEORY OF PROPERLY INFINITE C *-ALGEBRAS 497

projections in a properly infinite unital C *-algebra E build a group with (general-

ized) Cuntz addition and that this group is naturally isomorphic to K0 (E).
If we combine this with the below considered observation (C0) and Remark
4.1.2 then we get the following basic property (Cu) that generalizes [172, thm. 1.4]
(for simple unital C *-algebras A) to the case of general C *-algebras A:
Basic property of full properly infinite projections:

(Cu) Full and properly infinite projections p, q in a C *-algebra A are Murray–

von-Neumann equivalent if and only if [p] = [q] ∈ K0 (A) . (Here A is not
necessarily unital.)
If A contains a full and properly infinite projection e ∈ A, then each
element x ∈ K0 (A) is the class [p] = x of a properly infinite and full
projection p ∈ A with p ≤ e.

The observation (Cu) of J. Cuntz is almost equivalent to Part (ii). Our proof
uses the Parts (i) and (ii) of the “innocent” abstract Lemma 4.2.3. It shows at
first that the sub-semigroup of MvN-equivalence classes of full and properly infinite
projections build a group under Cuntz sum and that the natural morphism of this
group into K0 (A) is surjective. Then we compare the defining relations and obtain
from Lemma 4.2.3(ii) that this morphism must be a group isomorphism. This proof
of observation (Cu) is only slightly different from that of [172, thm. 1.4]. It is useful
to prove the below considered Observation (o) and Parts (i)–(iii) together.
The below stated basic observation (C0) in the proof of (i,ii,iii) reduces the
proof of observation (Cu) to the case of the unital properly infinite E = eAe, using
the natural isomorphism K0 (E) ∼ = K0 (A) of Remark 4.1.2, cf. also Part (v,5).
(o): Let
Uc := s1 s∗2 − s2 s∗1 + p[−1] = exp (π/2)(s1 s∗2 − s2 s∗1 ) .


(2.3)
Then Uc ∈ U0 (CR∗ (s1 , s2 )), Uc2 = p[−1] − (s1 s∗1 + s2 s∗2 ), Uc has “geodesic” distance
cel(Uc ) ≤ (π/2)ks1 s∗2 − s2 s∗1 k = π/2
to 1 inside the metric space U0 (E), and
Uc s1 = −s2 , Uc s2 = s1 , Uc p[−1] = p[−1] .
Hence, Uc∗ (a⊕s1 ,s2 b)Uc = b⊕s1 ,s2 a for a, b ∈ E and Uc ∈ U0 (E) for real or complex
C *-algebras E.
We define a partial unitary Z with Z ∗ Z = 1 − p[−1] = ZZ ∗ by
Z := s21 s∗1 + s2 (s∗2 )2 + (s1 s2 )(s2 s1 )∗ − s1 p[−1] s∗2 ∈ C ∗ (s1 , s2 ) ⊆ E .



Straight calculation gives Z a ⊕ (b ⊕ c) = (a ⊕ b) ⊕ c Z for a, b, c ∈ E, and
Zs2 p[−1] s∗2 = s1 p[−1] s∗1 Z.
If U (s1 ) ∈ U0 M2 (C ∗ (s1 , s2 )) ⊆ U0 (M2 (E)) denotes the Halmos unitary of

the isometry s1 (cf. Remark 4.2.4), then

Z = [s1 , s2 ] U (s2 ) · diag(1, Uc ) [s1 , s2 ]∗ .


498 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Thus Z ∈ U0 (1 − p[−1] )C ∗ (s1 , s2 )(1 − p[−1] ) , and the unitary

Ud := Z + p[−1] = s21 s∗1 + (s1 s2 )(s2 s1 )∗ + s2 (s∗2 )2 − s1 p[−1] s∗2 + p[−1] (2.4)

is in U0 (C ∗ (s1 , s2 )) ⊆ U0 (E), and that Ud “realizes” the distributive law for ⊕s1 ,s2
as unitary equivalence via Ud .
By Remark 4.2.4, cel(U (s2 )) ≤ π/2. If we apply this to the above given decom-
positions of Z and Ud , then we get that the unitary Ud has “exponential length”,
i.e., “geodesic” distance

cel(Ud ) ≤ cel(U (s2 )) + cel(Uc ) ≤ π

from 1 inside the metric space U0 (E).

Calculation shows that the unitary equivalence by Uc and Ud includes also the
“un-truncated” or “corrected” version ⊕0 of ⊕s1 ,s2 given by

u ⊕0 v := s1 us∗1 + s2 vs∗2 + p[−1]

on the group of unitary operators or the semi-group of isometries u, v, w ∈ E :

Uc∗ (u ⊕0 v)Uc = v ⊕0 u

and
Ud∗ ((u ⊕0 v) ⊕0 w))Ud = u ⊕0 (v ⊕0 w) .
This can be seen from Uc p[−1] = p[−1] and Zs2 p[−1] s∗2 = s1 p[−1] s∗1 Z.

(i,ii,iii): The commutative and associative law for ⊕ = ⊕s1 ,s2 , up to unitary
equivalence with a unitary in U0 (E), follows from Part (o). The other formulae by
simple calculations, e.g. u∗ (a ⊕ b)u = (u∗1 au1 ) ⊕ (u∗2 au2 ) for u := (u1 ⊕ u2 ) + p[−1] ,
and is in U0 (E) if u1 , u2 ∈ U0 (E), ...
Old Lemma form Appendix A, now somewhere above contained:
Suppose that p, q ∈ A are full properly infinite projections.
If [p] = [q] in K0 (A) then there exists v ∈ A with v ∗ v = p and vv ∗ = q.
If A contains a full and properly infinite projection p then for each x ∈ K0 (A)
there exists a full properly infinite projection q ≤ p with x = [q].
If p, r ∈ A are full and properly infinite projections that have the same class
[p] = [q] ∈ K0 (A) then there exists a partial isometry z ∈ A with z ∗ z = p and
zz ∗ = q.
The proof of the observation (Cu) of J. Cuntz uses following elementary observa-
tion (C0), given below, but the proofs of (Cu), (ii) and (iii) follows not immediately
from (C0):

(C0) If A is a (not necessarily unital) C *-algebra and e ∈ A is a full and

properly infinite projection, then for every n ∈ N and every projection
p ∈ Mn (A) there is a projection p0 ∈ A such that p and p0 ⊕ 0n−1 =
diag(p0 , 0, . . . , 0) are MvN-equivalent in Mn (A), p0 ≤ e and e − p0 is full
and properly infinite in A.
 2. ON K-THEORY OF PROPERLY INFINITE C *-ALGEBRAS 499

The observation (C0) reduces the proof of observation (Cu) to the case of
properly infinite unital algebras.
In fact it is formally more general, and is easy to see, that the following holds:
If e, p are projections in a C *-algebra B such that e is properly infinite and p is
in the closed ideal BeB generated by e, then there exists a partial isometry z ∈ B
with z ∗ z = p and zz ∗ ≤ e. Below we outline this in case B = Mn (eAe).
The proof of (C0) goes as follows: E := eAe has a properly infinite unit
1E := e, i.e., there are isometries s1 , s2 ∈ E with s∗1 s2 = 0. Define isometries
tk := sk−12 s1 ∈ E for k = 1, 2, . . . with t∗k tj = δj,k 1E . Let zk := tk ⊕ 0n−1 =
diag(tk , 0, . . . , 0) ∈ Mn (E) ⊆ Mn (A). Then zj∗ zk = δjk diag(e, 0, . . . , 0).
Since the projection e ∈ A is full in A and is properly infinite and since the
direct sum A ⊕ {0n−1 } ∼ = A ⊗ p11 is full in Mn (A) ∼ = A ⊗ Mn , the projection
p := diag(e, 0, . . . , 0) = e ⊕ 0n−1 is full and properly infinite in Mn (A).
It follows that, for each projection q ∈ Mn (A) and ε ∈ (0, 1/4], there are m ∈ N
P
and elements a1 , . . . , am , b1 , . . . , bm ∈ A such that kq − k ak pbk k ≤ ε . We get
kq − x∗ pyk ≤ ε for x :=
P P
j zj aj and y := k zk bk . It implies the existence of an
element c ∈ Mn (A) with c pc = q. The projection r = pcc∗ p ≤ p = (e, 0, . . . , 0) has
∗

form r = (p00 , 0, . . . , 0) for a projection p00 ≤ e. The projection p0 := s1 p00 s∗1 ≤ e

has the quoted property that e − p0 ≥ s2 s∗2 is full and properly infinite in E = eAe
and the given projection r ∈ Mn (A) is Murray–von-Neumann equivalent to p0 ∈
E∼ = E ⊕ 0n−1 inside Mn (A).
Therefore it suffices to consider only the MvN-equivalence classes [p] of pro-
jections p ∈ E := eAe, because each MvN-equivalence class [q] of a projection q
S
in n Mn (A) is identical with the MvN-class [p] of some projection p ∈ E, and
we restrict later our considerations to C *-algebras E with properly infinite unit
element.
The projections in E build a commutative semi-group S under Cuntz addi-
tion [p]M vN + [q]M vN := [p ⊕s1 ,s2 q]M vN . See Parts (o) and (i) for the additive
and associative law modulo unitary homotopy, i.e., modulo unitary equivalence by
unitaries in U0 (E).
If we consider E as a full corner of Mn (E) by a ∈ E 7→ diag(a, 0, . . . , 0), then
the sum p ⊕ q is MvN-equivalent to p0 + q 0 if p0 , q 0 ∈ Mn (E) are projections with
p0 q 0 = 0, and are MvN-equivalent in Mn (E) to p respectively q. It follows that the
natural map [p]M vN 7→ [p] ∈ K0 (E) becomes a semi-group epimorphism from the
semi-group S of the MvN-equivalence classes of projections p ∈ E onto K0 (E).
Since, by definition of K0 , [p] = [q] ∈ K0 (E) if and only if there exists n ∈ N
and a projection r0 ∈ Mn (E) and such that diag(p, r0 ) and diag(q, r0 ) are MvN-
equivalent in Mn+1 (E), we can find a projection r ∈ E that is MvN-equivalent to
r0 and such that p ⊕s1 ,s2 r and q ⊕s1 ,s2 r are MvN-equivalent. It follows that the
map from the semi-group S of MvN-classes of projections in E into K0 (E) defines
500 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

just the same equivalence relations on S as the canonical semi-group morphism

from S into its Grothendieck group Gr(S).
In conclusion, the natural group epimorphism from Gr(S) onto K0 (E) is an
isomorphism.
By observations (C0), the element z0 := [1E ]M vN ∈ S can play the role of z0 in
Lemma 4.2.3(ii) for the semi-group S of MvN-equivalence classes of projections in
E with Cuntz addition, because for every p ∈ E, p⊕(1−p) is MvN-equivalent to 1E ,
cf. rules in Part (i), i.e., [p] + [1 − p] = [1] = z0 in S. If we let y0 := [1 − (1 ⊕ 1)] =
[p[−1] ], then 2z0 + y0 = z0 in S, and e := z0 + y0 = [1 ⊕ p[−1] ] is a dominate
zero element of S by Lemma 4.2.3. Thus, the “splitting” projection 1 ⊕ p[−1] ⊕ 0
represents the “absorbing” zero element of e + S ∼ = Gr(S) via e + s 7→ [s]Gr by
Lemma 4.2.3. Since e + e = e in S, the natural map from S onto K0 (E) maps e
into the zero element and defines an isomorphism from the subgroup e + S of S
onto K0 (E), because it induces an isomorphism of the Grothendieck group Gr(S)
with K0 (E) by Lemma 4.2.3(i).
It implies that p and q are MvN-equivalent, if p, q ∈ E if [p] = [q] in K0 (E)
and are MvN-equivalent to projections in E ⊕ (1 ⊕ p[−1] ), i.e., have MvN-class in
S + e. Since 1 ⊕ p[−1] ≥ s1 s∗1 is full and properly infinite, each projection that is
MvN-equivalent to projections in E ⊕ (1 ⊕ p[−1] ) is necessarily full and properly
infinite.
The observation (C0) gives that full and properly infinite projections p ∈ E
are MvN-equivalent to projections in E ⊕ (1 ⊕ p[−1] ), because there exist a partial
isometry z ∈ E with z ∗ z = 1 ⊕ p[−1] and zz ∗ ≤ p, thus p is MvN-equivalent to
(p − zz ∗ ) ⊕ (1 ⊕ p[−1] ). The projections in E ⊕ (1 ⊕ p[−1] ) itself are all splitting
because its complements majorize s2 (1 − p[−1] )s∗2 + p[−1] ∼M vN 1 . This finishes
the proof of observation (Cu), and shows that the elements of S + e can be realized
all as MvN-equivalence classes of splitting projections in E.
It is clear that the full and properly infinite projections p are all MvN-equivalent
to splitting projections, e.g. take s1 ps∗1 in the MvN-equivalence class of p.
Sums p ⊕ q with full and properly infinite p are again full and properly infinite,
and if at least one of 1 − p or 1 − q is full and properly infinite then p ⊕ q becomes
a splitting projection.
The observation (Cu) implies that full and properly infinite projections p, q ∈ E
are MvN-equivalent if and only if [p] = [q] in K0 (E).
If [p] = [q] then also [1 − p] = [1] − [p] = [1] − [q] = [1 − q]. If p and q are
splitting then 1 − p and 1 − q are again full and properly infinite. Thus, there exists
a partial isometry z with z ∗ z = 1 − p and zz ∗ = 1 − q. The unitary u := v + z
satisfies up = qu if vv ∗ = q and v ∗ v = p.
If p, q ∈ E are properly infinite full projections then there exists an isometry
t ∈ E with tt∗ ≤ p. With this isometry it holds: If v ∈ E is a unitary with v ∗ pv = q
 2. ON K-THEORY OF PROPERLY INFINITE C *-ALGEBRAS 501

then u := (tv ∗ t∗ + (1 − tt∗ ))v is a unitary with 0 = [u] ∈ K1 (E) and u∗ pu = q,

because tv ∗ t∗ + (1 − tt∗ ) commutes with p.

(iv,a): Let p, q ∈ E full and properly infinite projections with [p] = [q] ∈ K0 (E),
and suppose that there exists an isometry t ∈ E with t∗ (p + q)t = 0. Then p, q
and r := tpt∗ are splitting projections and [p] = [q] = [r] in K0 (E). Since this
projections are full and properly infinite, there exist by Part (ii) partial isometries
z1 , z2 ∈ U(E) with z1∗ z1 = r = z2∗ z2 and z1 z1∗ = p and z2 z2∗ = q. We get from
t∗ (p + q)t = 0 that t∗ (p + q) = 0 and r(p + q) = 0. Thus zk2 = 0 for k = 1, 2.
The unitary elements

uk := (1 − zk∗ zk − zk zk∗ ) + (zk∗ − zk ) = exp((π/2)(zk∗ − zk )) ,

are in U0 (E) and satisfy u∗k (zk zk∗ )uk = zk∗ zk = r for k = 1, 2 by Part (v,6). Hence
u∗1 pu1 = r = u∗2 qu2 , and the unitary u := u1 u∗2 with u∗ pu = q is in U0 (E).
(iv,b): Let p, q ∈ E projections, and define P := (p ⊕ 1) ⊕ 0 = s21 p(s∗1 )2 +
s1 s2 (s1 s2 )∗ and Q := (q ⊕ 1) ⊕ 0 = s21 q(s∗1 )2 + s1 s2 (s1 s2 )∗ . Then [p] = [q] in
K0 (E), if and only if,

[P ] = [(p ⊕ 1) ⊕ 0] = [p] + [1] = [q] + [1] = [(q ⊕ 1) ⊕ 0] = [Q] .

The projections P and Q are full and properly infinite because s1 s2 (s1 s2 )∗ ≤ P and
s1 s2 (s1 s2 )∗ ≤ Q. They are also splitting projections and satisfy the assumptions
of Part (iv,a), because s∗2 (P + Q)s2 = 0.
Thus, [p] = [q], if and only if, [P ] = [Q], if and only if, there exist u ∈ U0 (E)
with u∗ P u = Q.

(v): We show first (v,2) and that u ∈ U(E) 7→ [u] ∈ K1 (E) is surjective. Since
(v,6) is obvious, this allows to verify (v,1) and (v,4) easily by using (v,3) and (v,6).
Therefore we display only the proofs of (v,2), (v,5) and (v,3) in more detail.
(v,2): Recall that K1 (E) for unital C *-algebras E can be described as the in-
ductive limit of the discrete groups U(Mn (E))/U0 (Mn (E)) by the group morphisms

u ∈ U(Mn (E)) 7→ u ⊕ 1 ∈ U(Mn+1 (E)) ,

where on this place ⊕ means the direct sum (despite that the original definition of
the topological K1 (B) for unital Banach algebras is formally different).
Let s1 , s2 ∈ E isometries with s∗1 s2 = 0 (e.g. as above chosen). We define
isometries tk := sk−1
2 s1 ∈ E for k = 1, 2, . . ..
The tn satisfy t∗n tm = δn,m 1. Let T (more precisely Tn ) denote the row
[t1 , t2 , . . . , tn , tn ] ∈ M1,n (E). The row T defines a partial isometry in Mn (E) with
T ∗ T = 1n and T T ∗ ≤ diag(1, 0, . . . , 0). We define a C *-morphism
X
ϕ : Mn (E) 3 [ajk ] 7→ T [ajk ]nn T ∗ = tj ajk t∗k
jk

from Mn (E) onto en Een , where

en := T T ∗ = t1 t∗1 + · · · + tn t∗n ≤ 1 .
502 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

The restriction to diagonal elements is given by

X
ϕ diag(a1 , a2 , . . . , an ) = Tn diag(a1 , a2 , . . . , an ) Tn∗ = tj aj t∗j .

1≤j≤n

That u ∈ U(E) 7→ [u] ∈ K1 (E) is surjective follows from the fact that V ⊕ 1n
and
diag(ϕ(V ) + 1 − ϕ(1n ) , 1, . . . , 1) := (ϕ(V ) + 1 − ϕ(1n )) ⊕ 12n−1
are unitarily equivalent in M2n (E) for each unitary V ∈ Mn (E) by the Halmos
unitary U (Tn ) ∈ M2 (Mn (E)) ∼ = M2n (E) of the contraction Tn (cf. Remark 4.2.4),
i.e., U (Tn ) = [ajk ] is given by a11 := Tn (considered as matrix in Mn (E) with 2-nd
to n-th rows equal to zero), a21 := 0, a22 := a∗11 and a12 := −(1 − a11 a∗11 ). The
matrix U (Tn ) is a Halmos unitary in U0 (M2n (E)), and

U (Tn )( V ⊕ 1n )U (Tn )∗ = a11 V a∗11 + 1n − a11 a∗11

The right side of the latter formula is the same as diag( ϕ(V ) + 1 − ϕ(1) , 1, . . . , 1) .
Thus, [ϕ(V ) + 1 − ϕ(1)] = [V ] in K1 (E) and the natural map u ∈ U(E) 7→ [u] ∈
K1 (E) is surjective.

Let u0 , u1 ∈ U(E) and u ∈ U(pEp) with [u0 ] = [u1 ] in K1 (E).

By (topological) definition of K1 -groups, cf. [692, def. 8.1.3], [u0 ] = [u1 ] in
K1 (E), if and only if, there is n ∈ N such that the unitaries diag(u0 , 1, . . . , 1) and
diag(u1 , 1, . . . , 1) are homotopic in U(Mn (E)), i.e., W := diag(u∗1 u0 , 1, . . . , 1) ∈
U0 (Mn (E)) for sufficiently big n ∈ N. It follows that

s1 (u∗1 u0 )s∗1 + (1 − s1 s∗1 ) = ϕ(W ) + (1 − en ) ∈ U0 (E) .

It gives the desired homotopy in U(E):

(u1 ⊕ 1) + p[−1] = s1 u1 s∗1 + (1 − s1 s∗1 ) ∼h s1 u0 s∗1 + (1 − s1 s∗1 ) = (u0 ⊕ 1) + p[−1] .

(v,5): The difference between the case where p is not full and the here consid-
ered case where p is full can be seen in Example 4.2.9.
Let p ∈ E a projection such that 1 − p is full and properly infinite in E and let
u ∈ U(pEp) a unitary that satisfies [u + (1 − p)] = 0 in K1 (E).
Since 1 − p is full and properly infinite, the unit of E must be properly infinite.
Thus, there are isometries s1 , s2 ∈ E that satisfy the non-degeneracy property (ND)
on p[−1] .
We define below unitary operators W, v ∈ E that satisfy

(v ⊕s1 ,s2 1) + p[−1] = s1 vs∗1 + (1 − s1 s∗1 ) = W ∗ (u + (1 − p)W .

It follows [(v ⊕1)+p[−1] ] = [u+(1−p)] = 0 in K1 (E). We get (v ⊕1)+p[−1] ∈ U0 (E)

by Part (v,2), and can conclude that u + (1 − p) = W ((v ⊕ 1) + p[−1] )W ∗ ∈ U0 (E).
There is an isometry t ∈ E with tt∗ ≤ (1−p), because (1−p) is full and properly
infinite. The projection r := t(s2 s∗2 + p[−1] )t∗ is splitting, because ts2 (ts2 )∗ ≤ r
and ts1 (ts1 )∗ ≤ 1 − r. Obviously, s2 s∗2 + p[−1] is a splitting projection.
 2. ON K-THEORY OF PROPERLY INFINITE C *-ALGEBRAS 503

The Weyl–von-Neumann equivalence r ∼W vN s2 s∗2 + p[−1] , and the unitary

equivalence of splitting projections with same class in K0 (E) – as established in Part
(iii) – imply the existence of a unitary W ∈ U(E) with W ∗ rW = s2 s∗2 + p[−1] =
1 − s1 s∗1 .
Since r ≤ (1 − p), the element e := 1 − (p + r) ≤ 1 − p is a projection in E with
p + e = 1 − r and u + e is a unitary in (p + e)E(p + e) = (1 − r)E(1 − r).
For every u ∈ U(pEp), we can define an element v ∈ E by

v := s∗1 W ∗ (u + e)W s1 = s∗1 W ∗ (u + (1 − p))W s1 .

The element v ∈ E is unitary, because u + e ∈ U((p + e)E(p + e)), W s1 s∗1 =

(1 − r)W = (p + e)W and s∗1 W ∗ W s1 = 1. Then 1 − p = e + r and

(v ⊕ 1) + p[−1] = s1 vs∗1 + (1 − s1 s∗1 ) = W ∗ ((u + e) + r)W = W ∗ (u + (1 − p))W .

We consider now the case where the projection p ∈ E is splitting in E:

If p ∈ E is a splitting projection, then the inclusion map η : pEp ,→ E defines
an isomorphism η∗ from K∗ (pEp) onto K∗ (E), because pEp and E are in a natural
way stably equivalent: There exists an isometry T in M(E ⊗ K) with T T ∗ = p ⊗ 1.
Explicitly the group homomorphisms η∗ are given for the classes of splitting
projections e ∈ pEp by η0 ([e]) := [e] and by η1 ([u]) := [u + (1 − p)] for unitary
u ∈ U(pEp).
The (generalized) Cuntz addition ⊕s1 ,s2 can be chosen such that η∗ is compati-
ble with the splitting projection, e.g. such that s1 s∗1 +s2 s∗2 ≤ p and and p−s1 s∗1 −s2 s∗2
is a splitting projection in pEp.
This isometries s1 , s2 can be used to describe the inverse θ∗ := (η∗ )−1 of the
η∗ on classes [e] ∈ K0 (E) and [u] ∈ K1 (E) by θ0 ([e]) := [s1 es∗1 ] and θ1 ([u]) :=
[s1 us∗1 + (p − s1 s∗1 )]. Elementary verifications of η1 ◦ θ1 = id and θ1 ◦ η1 = id follow
then from the identities

s1 us∗1 + (p − s1 s∗1 ) + (1 − p) = s1 us∗1 + (1 − s1 s∗1 )

and [u] = [s1 us∗1 + (1 − s1 s∗1 )] in K1 (E), and of [s1 (u + (1 − p))s∗1 + (p − s1 s∗1 )] = [u]
in K1 (pEp). The identities η0 ◦ θ0 = id and θ0 ◦ η0 = id come from [s1 es∗1 ] = [e] for
e ∈ E in K0 (E) and for e ∈ pEp inside K0 (pEp).
One of the consequences for splitting p ∈ E and u ∈ U(pEp) is:
[u + (1 − p)] = 0 in K1 (E), if and only if, [u] = 0 in K1 (pEp), if and only if,
u + (1 − p) ∈ U0 (E).
(v,3): (u∗ ⊕ u) + p[−1] = V ∗ Uc∗ V Uc , for V := s1 us∗1 + (1 − s1 s1 ) = u ⊕0 1 and
with Uc ∈ U0 (E) defined by Equation (2.3) in the proof of Part (o).
(vi): We prove first the result for the special case in sub-part (vi, α), where u
commutes with p:
If p is splitting, i.e., if p and 1 − p are both full and properly infinite projections in
504 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

E then there exist isometries t1 , t2 ∈ E with t1 t∗1 ≤ p and t2 t∗2 ≤ 1 − p. The unitary

R := t1 t∗2 − t2 t∗1 + (1 − t1 t∗1 − t2 t∗2 ) = exp((π/2)(t1 t∗2 − t2 t∗1 ))

is in U0 (E). Obviously t∗1 R = t∗2 , pt1 t∗1 = t1 t∗1 , (1−p)t2 t∗2 = t2 t∗2 and R∗ t1 t∗1 R = t2 t∗2 .
Thus,
R∗ pRt2 t∗2 = R∗ pt1 t∗1 R = R∗ t1 t∗1 R = t2 t∗2 .
It implies that R∗ (1 − p)R t2 t∗2 = 0, u(1 − p)R t2 t∗2 = 0 and that

R∗ (p + u(1 − p))R t2 t∗2 = t2 t∗2

Since (1 − p)t2 t∗2 = t2 t∗2 , we get finally that

(up + (1 − p))R∗ (p + u(1 − p))Rt2 t∗2 = t2 t∗2 .

We let q := 1 − t2 t∗2 and V := (up + (1 − p))R∗ (p + u(1 − p))R. Then 1 − q = t2 t∗2

is full and properly infinite in E and V (1 − q) = (1 − q). Hence, there is a unitary
w ∈ qEq such that w + (1 − q) = V . Since R ∈ U0 (E), the unitary V is homotopic
in U(E) to u = (up + (1 − p))(p + u(1 − p)). By assumption, [u] = 0 in K1 (E). It
follows [w + (1 − q)] = [V ] = [u] = 0.
We can apply Part (v,5) to w + (1 − q) (in place of u + (1 − p) of Part (v,5)),
because (1 − q) = s2 s∗2 is full and properly infinite in E, and w ∈ U(qEq) satisfies
[w + (1 − q)] = 0 in K1 (E). Part (v,5) implies that V = w + (1 − q) is in U0 (E).
Since u ∼h V , it follows that u ∈ U0 (E). This proves case (α).

Now suppose, more generally, that u1 , u2 ∈ U(E) and a splitting projection

p ∈ E satisfy kuk p − puk k < 1, for k = 1, 2 and that [u1 ] = [u2 ] ∈ K1 (E). Then
qk := uk pk u∗k satisfy kqk − pk < 1 for k = 1, 2.
By Lemma 4.1.3(v), if p, q ∈ E are two projections with kp − qk < 1, then there
exists a unitary v ∈ U0 (A) with v ∗ pv = q in the connected component U0 (A) of 1
in U(A), where A := C ∗ (1, p, q) ⊆ E.
This applies to the projections p and qk := uk pu∗k (k = 1, 2), and gives
v1 , v2 ∈ U0 (E) with vk qk vk∗ = p. Thus, v1 u1 pu∗1 v1∗ = v2 u2 pu∗2 v2∗ and u := u∗2 v2∗ v1 u1
commutes with p = t1 t∗1 . Since [u1 ] = [u2 ] and [vk ] = 0 we get [u] = [v1 ] + [u1 ] −
[v2 ] − [u2 ] = 0.
By the above considered special case (α), we get that u ∈ U0 (E). The unitary
u is homotopic to u∗2 u1 in U(E) because v2∗ v1 ∈ U0 (E). Hence, u∗2 u1 ∈ U0 (E) and
u1 ∼h u2 in U(E).
(vi,β): Let t1 , t2 ∈ E isometries with kt∗1 t2 k < 1 and u ∈ U (E) such that
kt∗1 ut2 k < 1 and [u] = 0 ∈ K1 (E).
We consider the projection q := 1 − t2 t∗2 and unitarily equivalent properly
infinite full projections p1 := t1 t∗1 and p2 := u∗ p1 u. Then p1 and p2 are unitarily
equivalent and, for k = 1, 2,

k(1 − q)pk k = kpk t2 k ≤ max kt∗1 t2 k, kt∗1 ut2 k < 1 .

 2. ON K-THEORY OF PROPERLY INFINITE C *-ALGEBRAS 505

Hence, Lemma 4.1.3(iv) applies and gives skew-adjoint operators hk = −h∗k ∈ E

with khk k ≤ arcsin k(1 − q)pk k < π/2 such that rk := exp(−hk )pk exp(hk ) ≤ q .
The definition of the projections pk shows that p1 and p2 are unitarily equivalent
properly infinite and full projections in E. The same follows for the rk . Moreover
the rk satisfy t∗2 (r1 + r2 )t2 = 0 because rk ≤ q = 1 − tt∗ . By Part (iv,a), there
exists a unitary v ∈ U0 (E) such that v ∗ r1 v = r2 . It follows that

v ∗ exp(−h1 )p1 exp(h1 )v = v ∗ r1 v = r2 = exp(−h2 )u∗ p1 u exp(h2 ) ,

i.e., the unitary w = u exp(h2 )v ∗ exp(−h1 ) commutes with the projection p1 . The
projection p1 = exp(h1 )r1 exp(−h1 ) is splitting, because p1 ≤ 1 − SS ∗ for the
isometry S := exp(h1 )t2 .
The K1 (E)-class [w] of w is zero, because [u] = 0 by assumption, v, exp(hk ) ∈
U0 (E) and [w] = [u] + [exp(h1 )] − [v] − [exp(h2 )] = 0 . Thus, Part (vi,α) applies
to w and p1 . It proves that w ∈ U0 (E), and that u = w exp(h1 )v exp(−h2 ) is in
U0 (E).

(vii): The observation (Cu), applied to p := 1 − s1 s∗1 and q := 1 − t1 t∗1 ,

gives immediately that for two pairs s1 , s2 ∈ E and t1 , t2 ∈ E of isometries with
orthogonal ranges there exist a unitary u ∈ E such that us1 = t1 and [u] = 0 ∈
K1 (E). It can appear that there does not exist any unitary u with us1 = t1 that
satisfies also us2 = t2 . This is e.g. the case if p[−1] and q[−1] are not both zero or
are not both properly infinite and full projections.
If p[−1] = q[−1] = 0, then u := t2 s∗2 + t1 s∗1 is the unique unitary in E with
usk = tk . Therefore, e.g. if K1 (E) 6= {0} and 0 = [1] ∈ K0 (E), it can happen that
u ∈ U(E) and usk = tk imply [u] 6= 0.
But one can always find a unitary v that satisfies vs1 = t1 and 0 = [u] ∈ K1 (E)
(but not necessarily vs2 = t2 ), e.g. let v := u(1 ⊕s1 ,s2 u∗ ) in case p[−1] = 0.
In the general case of given pairs of isometries (s1 , s2 ) and (t1 , t2 ) with orthog-
onal ranges let v := t1 s∗1 + z for some partial isometry z with z ∗ z = p := 1 − s1 s∗1
and zz ∗ = q := 1 − t1 t∗1 , existing by Part (ii), because p and q are full and properly
infinite. The unitary u := w∗ v with w := (1 − t2 t∗2 ) + t2 vt∗2 is a unitary with
us1 = t1 and [u] = 0 ∈ K1 (E).

We can proceed as in the above consideration for the finding of a unitary if p[−1]
and q[−1] are both full and properly infinite: Use that [p[−1] ] = −[1E ] = [q[−1] ] and
get from Part (ii) that there is a partial isometry z with z ∗ z = p[−1] and zz ∗ = q[−1] .
Then v := t1 s∗1 + t2 s∗2 + z is a unitary with vsk = tk .
Since p[−1] is full and properly infinite, we get that there exists an isometry
S ∈ E with SS ∗ ≤ p[−1] . Then u := v · (1 − SS ∗ ) + Sv ∗ S is a unitary with

usk = tk and [u] = 0 in K1 (E).

The general observation is the following: Let s1 , . . . , sn and t1 , . . . , tn isometries
in E with s∗j sk = δjk 1 = t∗j tk , i.e., the sk and tk correspond to unital C *-morphisms
506 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Pn Pn
h0 and h1 from En into E. Let p[1−n] := 1− k=1 sk s∗k and q[1−n] := 1− k=1 tk t∗k .
Clearly [p[1−n] ] = (1 − n)[1] = [q[1−n] ] in K0 (E).
Pn ∗
If p[1−n] = q[1−n] = 0 then u := k=1 tk sk is the unique unitary with usk = tk
and it can happen that [u] 6= 0.
The projections p[1−n] and q[1−n] are MvN-equivalent in E, if and only if, there
exists a u ∈ U(E) with usk = tk . This unitary induces a unitary equivalence of
p[1−n] and q[1−n] by up[1−n] u∗ = q[1−n] . In particular, p[1−n] ∼M vN q[1−n] .
The latter is always the case if p[1−n] and q[1−n] are both full and properly
infinite:
If p[1−n] and q[1−n] are both full and properly infinite, then they are MvN-
equivalent by Part (ii), and there exist isometries sn+k (k = 1, 2, . . .) with
s∗n+k sn+` = 0 for k 6= ` and sn+k s∗n+k ≤ p[1−n] .
The unitary u ∈ U(E) with usk = tk is determined up to multiplication with
elements in G := (1 − p[1−n] ) + U(p[1−n] Ep[1−n] ) from the right side. It is the
same as the multiplication with elements in (1 − q[1−n] ) + U(q[1−n] Eq[1−n] ) from
the left side – because up[1−n] = q[1−n] u. Since v := sn+1 u∗ s∗n+1 + (1 − sn+1 s∗n+1 )
is in the group G, it follows that uG contains unitaries w with [w] = 0 in K1 (E),
e.g. w := uv. Our below explained method and Part (vi,β) indicates that it suffices
to find an isometry t ∈ E such that k(1 − p[1−n] )tk < 1 and k(1 − p[1−n] )u∗ tk < 1
to show that uG ∩ U0 (E) is not empty.
The consideration of the case n = 2 explains the ideas and appearing problems
in the general case of missing K1 -injectivity. We discuss here only the case n = 2
in any detail:
The case – where a properly infinite and full projection r ∈ E with r ≤ p[−1]
and r ≤ q[−1] exists – uses Part (v,5), or likewise Part (vi,α). (Notice that the
above given proofs of Parts (v) and (vi) are independent from our considerations
given here.)
Suppose that a properly infinite and full projection r ∈ E exists with r ≤ p[−1]
and r ≤ q[−1] . We find an isometry T ∈ E such that T T ∗ ≤ r. Define s3 := t3 :=
T s1 , and the partial isometry V := t1 s∗1 + t2 s∗2 + t3 s∗3 . Then 1 − V ∗ V =: p[−2] =
p[−1] − T s1 s∗1 T and 1 − V V ∗ =: q[−2] = q[−1] − T s1 s∗1 T are full and properly infinite
projections, because each of it dominates the range T s2 s∗2 T ∗ of the isometry T s2
and E is properly infinite. The projections p[−2] and q[−2] are properly infinite and
[p[−2] ] = [1] − [V ∗ V ] = [1] − [V V ∗ ] = [q[−2] ]. Thus, there exists a partial isometry
Z with Z ∗ Z = p[−2] and ZZ ∗ = q[−2] . We obtain that

U := (T + Z)(1 − T s2 s∗2 T ∗ + T s2 (T + Z)∗ s∗2 T ∗ )

satisfies U sk = tk (k = 1, 2, 3) and [U ] = 0 in K1 (E). Since s3 = t3 , we get

U = W + (1 − P ) with P := 1 − T s1 s∗1 T ∗ = 1 − s3 s∗3 and W := U P ∈ U(P EP ).
The facts that (1 − P ) is full and properly infinite and [W + (1 − P )] = 0 in K1 (E)
imply that U = W +(1−P ) ∈ U0 (E) by Part (v,5) or Part (vi,α). Clearly U sk = tk
and U p[−1] = q[−1] U . 
 2. ON K-THEORY OF PROPERLY INFINITE C *-ALGEBRAS 507

Example 4.2.9. The difference between the use of splitting or non-splitting

properly infinite projection p in the formulation of Lemma 4.2.6(v,5) can be seen
from the following example of a strongly purely infinite C *-algebra E with U0 (E) =
U(E) and a non-splitting properly infinite projection p ∈ E that satisfies the first
assumption in Lemma 4.2.6(v,5) such that 1 − p properly infinite and full in E, but
K1 (pEp) 6= {0}:
Consider E := M(P∞ ⊗ K) and p := 1 ⊗ p11 , where P∞ denotes here the (up
to isomorphisms) unique unital pi-sun algebra in the UCT-class with K0 (P∞ ) = 0
and K1 (P∞ ) ∼= Z . Thus, [u] = 0 implies in this case [u + (1 − p)] = 0, but the
opposite direction is wrong.

Recall that all properly infinite unital C *-algebras are K1 -surjective by Lemma
4.2.6(ii). The following Lemma 4.2.10 collects necessary and sufficient conditions
for the K1 -injectivity of properly infinite unital C *-algebras. (But the squizing
property of Definition allowes more sufficiant cases for K1 -injectivity.)

Lemma 4.2.10. Let E a properly infinite unital C*-algebra. Each of the fol-
lowing conditions (i)–(iv) is equivalent to the K1 -injectivity of E :

(i) For every unitary u ∈ E with [u] = 0 in K1 (E) there exist isometries
t1 , t2 ∈ E and a unitary v ∈ U0 (E) with kt∗1 ut2 k < 1 and kt∗1 vt2 k < 1.
(ii) For every unitary u ∈ E with [u] = 0 in K0 (E), there exist isometries
r1 , r2 ∈ E and a unitary v ∈ U0 (E) with r1∗ vr2 = 0 and r1∗ ur2 = 0.
(iii) For every isometry s ∈ E with full and properly infinite 1 − ss∗ and for
every unitary u ∈ E one can find isometries t1 , t2 ∈ E and a unitary
v ∈ U0 (E) with ust1 = vst2 .
(iv) Splitting projections p, q ∈ E with [p] = [q] ∈ K0 (E) are unitarily equiva-
lent by a unitary in U0 (E).
(v) For every isometries s and t with orthogonal ranges (i.e., s∗ t = 0) and
every unitary u ∈ U(E) the projections tt∗ and u∗ tt∗ u are homotopic
inside the projections of E (denoted by tt∗ ∼h u∗ tt∗ u).

A sufficient condition for the K1 -injectivity of E is the following (possibly

stronger) property (vi):

(vi) For every u ∈ U(E) with 0 = [u] ∈ K1 (E) there exists a splitting projection
p ∈ E with kup − puk < 1 .
For example, this is the case for each fixed given u ∈ U(E), if there
exists a C*-morphism

Hu : C ∗ (u) ⊗ O∞ → E ∞ := `∞ (E)/c0 (E)

with Hu (u ⊗ 1) = (u, u, . . .) + c0 (E) ∈ E ∞ .

Those C *-morphisms Hu : C ∗ (u)⊗O∞ → E ∞ exist for example if E is strongly

purely infinite, cf. Chapter 3. It is not known if such morphisms Hu exist for all uni-
tal purely infinite C *-algebras E and u ∈ U(E). An answer to this question could
508 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

be interesting for the open question if every purely infinite C *-algebra is strongly
purely infinite. But notice here that all unital p.i. C *-algebras are K1 -injective
because they satisfy the formally stronger “squeezing” property (sq) defined in
Definition 4.2.14.

Proof. The equivalence of each of (i)–(v) with K1 -injectivity:

K1 -injectivity ⇒ (iii): Let s ∈ E an isometry with full and properly infinite
1 − ss∗ and u ∈ E unitary. There exists an isometry t ∈ E with t∗ s = 0. Let
v := u(tu∗ t∗ + 1 − tt∗ ) and t1 := t2 := t. Then ust1 = vst2 and v is a unitary with
[v] = 0 in K1 (E). The K1 -injectivity implies v ∈ U0 (E).
(iii) ⇒ (ii): Let u ∈ U(E) with [u] = 0 in K0 (E). Since 1E is properly infinite,
there exists an isometry s ∈ E with full and properly infinite 1 − ss∗ . By assump-
tions of (iii), there exist isometries t1 , t2 ∈ E and a unitary u0 ∈ U0 (E) such that
ust1 = u0 st2 .
There exists an isometry t ∈ E with t∗ s = 0, because 1−ss∗ is full and properly
infinite by assumptions on s.
The element

w := 1 − ss∗ − tt∗ + st2 s∗ + tt∗2 t∗ − s(1 − t2 t∗2 )t∗ = (1 − ss∗ − tt∗ ) + [s, t]U (t2 )[s, t]∗

is a unitary in U0 (E), because the Halmos unitary U (t2 ) is in U0 (M2 (E)), cf. Remark
4.2.4.
Obviously, ws = st2 . Since ust1 = u0 st2 , we obtain that (u0 t)∗ (u0 w)st1 =
t∗ st2 t1 = 0 and (u0 t)∗ u(st1 ) = (u0 t)∗ u0 st2 = 0.
If we take r1 := u0 t, r2 := st1 and v := u0 w then v ∈ U0 (E), and r1 , r2 are
isometries with r1∗ vr2 = 0 and r1∗ ur2 = 0, i.e., the elements v, r1 , r2 fulfill condition
(ii) for the given unitary u with [u] = 0.

The implication (ii) ⇒ (i) is obvious.

(i) ⇒ K1 -injectivity: Let u ∈ U(E) with 0 = [u] ∈ K1 (E) and suppose that
(i) holds, i.e., that there exist isometries t1 , t2 ∈ E and a unitary v ∈ U0 (E) with
kt∗1 vt2 k < 1 and kt∗1 ut2 k < 1.
We can apply Part (vi,β) of Lemma 4.2.6 to (t1 , vt2 , uv ∗ ) (in place of the triple
(t1 , t2 , u) in Part (vi,β) of Lemma 4.2.6), because [uv ∗ ] = [u] − [v] and [u] = 0,
v ∈ U0 (E), and get uv ∗ ∈ U0 (E) and finally u ∈ U0 (E). Thus every unitary u ∈ E
with 0 = [u] ∈ K1 (E) is in U0 (E).
K1 -injectivity ⇒(iv): Let p, q ∈ E splitting projections with [p] = [q] ∈ K0 (E).
By Part (ii) of Lemma 4.2.6, there exists a unitary u ∈ E with u∗ pu = q and
0 = [u] ∈ K1 (E). If E is K1 -injective this implies u ∈ U0 (E).
(iv)⇒ K1 -injectivity: Let v ∈ E a unitary with 0 = [v] ∈ K1 (E). The projec-
tion p := s1 s∗1 ∈ E is splitting, because s2 s∗2 ≤ 1 − p. Clearly q := v ∗ pv is also a
splitting projection that has same class [q] = [p] in K0 (E).
 2. ON K-THEORY OF PROPERLY INFINITE C *-ALGEBRAS 509

By assumption of (iv) there exists a unitary u ∈ U0 (E) with p = u∗ qu. Then

vu commutes with the splitting projection p = s1 s∗1 and 0 = [v] + [u] = [vu]. The
Part (vi,α) of Lemma 4.2.6 shows that vu ∈ U0 (E). Thus v ∈ U0 (E).
K1 -injectivity ⇒(v): Suppose that E is K1 -injective. Let s, t ∈ E isometries
with t∗ s = 0 and u ∈ U(E). Define v := ((1 − ss∗ ) + su∗ s∗ )u, then [v] = 0 in K1 (A).
It implies v ∈ U0 (A), because E is K1 -injective. It follows u∗ tt∗ u = v ∗ tt∗ v ∼h tt∗ .
(v)⇒ K1 -injectivity: If s, t ∈ E are isometries with s∗ t = 0, then p := tt∗ is a
splitting projection. Let u ∈ U(A) with 0 = [u] ∈ K1 (E). By assumption of (v),
p ∼h u∗ pu. Remark 4.1.5 says that there exists w ∈ U0 (A) with p = w∗ (u∗ pu)w, i.e.,
vp = pv for the unitary v = uw. We have [v] = [u] + [w] = 0, because [u] = 0 = [w].
It follows v ∈ U0 (A) by Lemma 4.2.6(vi,α). Hence, u = vw∗ ∈ U0 (A).

This finish the proof of the equivalences of (i)–(v).

(vi)⇒(i): Suppose that for each u ∈ U(E) there exists a C *-morphisms
Hu : C ∗ (u) ⊗ O∞ → E ∞ with Hu (u ⊗ 1) = u ∈ E ⊆ E ∞ .
Let u ∈ U(E) with [u] = 0 in K1 (E) and

Hu : C ∗ (u) ⊗ O∞ → E ∞ := `∞ (E)/c0 (E)

a C *-morphism with Hu (u ⊗ 1) = u ∈ E ⊆ E ∞ .
Let s1 , s2 , . . . canonical generators of O∞ . Hu (1 ⊗ sk ) ⊆ E ∞ for k ∈ {1, 2} lift
to a sequence of contractions (sk,1 , sk,2 , . . .) ∈ `∞ (E) that satisfy, for j, k ∈ {1, 2},
the equations limn ks∗j,n sk,n − δj,k 1k = 0 and limn k[u, sk,n ]k = 0 .
Replace, for sufficiently big n ∈ N, the sk,n (k ∈ {1, 2}) by its small perturba-
tions tk,n ∈ E that are isometries and satisfy k[u, tk,n ]k < 1/4 and kt∗1,n t2,n k < 1/4.
If we let v := 1, tk := tk,n , then u, v, t1 and t2 satisfy the condition in Part (i).
Thus, E is K1 -injective if it has the property in Part (v). 

In the following Proposition 4.2.11 the elements s1 , s2 , . . . denote the canonical

generators of O∞ := C ∗ (s1 , s2 , . . . ; s∗j sk = δjk 1) . Recall that C *-morphisms
h0 , h1 : A → E are homotopic, if there exists a point-norm continuous path ξ ∈
[0, 1] 7→ Hξ ∈ Hom(A, E) of C *-morphisms with h0 = H0 and h1 = H1 .

Proposition 4.2.11. Suppose that E is a C*-algebra with properly infinite unit

element.
Then, for any unital C*-morphisms h0 , h2 : O∞ → E, there exists a unitary
u0 ∈ U(E) with u0 h2 (s1 ) = h0 (s1 ) and 0 = [u0 ] ∈ K1 (E) such that h0 is homotopic
to h1 in Hom(O∞ , E) with point-norm topology, where we build h1 : O∞ → E from
h2 by h1 (sn ) := u0 h2 (sn ) for n ∈ N.
The algebra E is K1 -injective if and only if any two unital C*-morphisms from
O∞ into E are homotopic.
510 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Proof. The unital C *-morphisms h0 , h2 : O∞ → E are injective because O∞

is simple (and purely infinite) by Corollary 2.2.7, cf. also [169, thm. 1.13, thm. 3.4]
and [172, prop. 1.6].
We use this to simplify notations, because we can now suppose that O∞ is
unitally contained in E (via h0 ) by identifying the canonical generators sn of O∞
with its images h0 (sn ), and get a reference copy of O∞ inside E. Hence, we can
suppose without loss of generality that O∞ ⊆ E and that h0 (sn ) = sn , i.e., that h0
is the identity map of O∞ .
Let Z := h2 (s1 )s∗1 . Then Zs1 = h2 (s1 ), and the projections Z ∗ Z = s1 s∗1 ,
ZZ ∗ = h2 (s1 s∗1 ), 1−Z ∗ Z = 1−s1 s∗1 and 1−ZZ ∗ = h2 (1−s1 s∗1 ) are full and properly
infinite projections with [1−Z ∗ Z] = [1−ZZ ∗ ] in K0 (E). By Lemma 4.2.6(ii), cf. also
[172, thm. 1.4], there exists a partial isometry Y ∈ E with Y ∗ Y = 1 − Z ∗ Z and
Y Y ∗ = 1 − ZZ ∗ . Since Y ∗ Z = 0 = ZY ∗ , the sum W := Y + Z is a unitary with
W s1 = h2 (s1 ). The element V := W ((1 − s2 s∗2 ) + s2 W ∗ s∗2 ) is again a unitary with
V s1 = h2 (s1 ), but now with class [V ] = 0 in K1 (E).
Let u0 := V ∗ . It satisfies [u0 ] = 0 and u0 h2 (s1 ) = s1 = h0 (s1 ). We define a
new C *-morphism h1 : O∞ → E by h1 (sn ) := u0 h2 (sn ) for n = 1, 2, . . .. It has the
property h1 (s1 ) = s1 .
The arguments used above (for h2 , Z and construction of V ) work also for h1
Pn
in place of h2 and the partial isometries Zn := k=1 h1 (sk )s∗k :
Pn
The Zn satisfy Pn := Zn∗ Zn = k=1 sk s∗k and
n
X
Zn Zn∗ = h1 ( sk s∗k ) = h1 (Pn ) .
k=1

The full and properly infinite projections 1 − Zn∗ Zn and 1 − Zn Zn∗ = h1 (1 − Zn∗ Zn )
are MvN-equivalent because they have have same K0 (E)-class. Let Yn ∈ E a partial
isometry with Yn∗ Yn = 1 − Zn∗ Zn and Yn Yn∗ = 1 − Zn Zn∗ . The sum Wn := Yn + Zn
is unitary and the unitary

Vn := Wn (1 − sn+1 s∗n+1 + sn+1 Wn∗ s∗n+1 )

satisfies 0 = [Vn ] ∈ K1 (E) and Vn sk = h1 (sk ) for k ≤ n. In particular, V1 P1 = P1

and V1 (1 − s1 s∗1 ) is a unitary in U((1 − s1 s∗1 )E(1 − s1 s∗1 )). Since s1 s∗1 is a splitting
projection and [V1 ] = 0, this implies by Lemma 4.2.6(v,5) that V1 (1 − s1 s∗1 ) has
class [V1 (1 − s1 s∗1 )] = 0 in K1 ((1 − s1 s∗1 )E(1 − s1 s∗1 )), and that V1 ∈ U0 (E).
It follows that there is a continuous path λ ∈ [0, 1] 7→ Vλ ∈ U0 (E) that connects
V0 := 1 and the above defined unitary V1 ∈ U(E).
Similar arguments apply for n ≥ 1 to the unitary Qn := Vn∗ Vn+1 and the
projection Pn = Zn∗ Zn :
We get that Qn Pn = Pn and that Pn is splitting. Hence, [Qn ] = [Vn+1 ]−[Vn ] =
0 in K1 (E) implies that [Qn (1 − Pn )] = 0 in K1 ((1 − Pn )E(1 − Pn )) because 1 − Pn
is a full projection in E, see Lemma 4.2.6(v,5) with Qn (1 − Pn ) and 1 − Pn in place
of u and p.
 2. ON K-THEORY OF PROPERLY INFINITE C *-ALGEBRAS 511

If we let P0 := 0 and P1 = s1 s∗1 , then by Lemma 4.2.6(v,5),

Qn (1 − Pn−1 ) = Qn (1 − Pn ) + sn s∗n ∈ U0 ((1 − Pn−1 )E(1 − Pn−1 )) .

(n)
We obtain a continuous path λ ∈ [0, 1] 7→ Rλ ∈ U((1 − Pn−1 )E(1 − Pn−1 )) with
(n)
R0 = (1 − Pn−1 ) and

R1n+1 = Qn (1 − Pn ) + sn s∗n .
(n)
Notice that Vn+1 = Vn Qn = Vn · (R1 + Pn−1 ).
We combine the above defined maps λ ∈ [0, 1] 7→ Vλ and n ∈ N 7→ Vn to a
continuos path λ ∈ [0, ∞) 7→ Vλ ∈ U0 (E) for λ ∈ [n, n + 1] (n ≥ 1) by
(n)
Vλ := Vn · (Rλ−n + Pn−1 ) .

Since Vn sk = h1 (sk ) for k ≤ n and Pn−1 sk = sk for k ≤ n − 1, it follows that

Vλ sk = h1 (sk ) for all k ≤ λ − 1.
The above defined path λ ∈ [0, ∞) → Vλ ∈ U0 (E) defines a point-norm contin-
uous path ξ ∈ [0, 1] 7→ Hξ in the unital C *-morphisms from O∞ into E by assigning
to the generators s1 , s2 , . . . ∈ h0 (O∞ ) and ξ ∈ [0, 1) the values

Hξ (sn ) := V(1−ξ)−1 ξ · sn = uξ · sn

where we let uξ := V(1−ξ)−1 ξ and H1 := h1 . Then H0 = h0 , and the path is

continuous at ξ = 1, because

lim uξ sn = h1 (sn ) for all n ∈ N .

ξ→1

Moreover, Hξ (sk ) = uξ sk = sk if k + 1 ≤ (1 − ξ)−1 by our construction.

Suppose that E is K1 -injective, i.e., that [u] = 0 ∈ K1 (E) implies u ∈ U0 (E).

Then above considered u0 is in U0 (E) and we find a path ξ ∈ [1, 2] 7→ u(ξ) ∈ U(E)
with u(1) = u0 and u(2) = 1. We can define a point-norm continuous path

ξ ∈ [0, 2] 7→ Hξ : O∞ → E

by Hξ := hξ for ξ ∈ [0, 1], and by Hξ (sn ) := u(ξ)h1 (sn ) for ξ ∈ [1, 2]. This path
satisfies H0 = h0 and H2 = h2 .

Suppose that 1E is properly infinite and that any two unital C *-morphisms
h1 , h2 : O∞ → E are homotopic. Let u ∈ U(E) with 0 = [u] in K1 (E).
Since 1 is properly infinite there exist isometries t1 , t2 ∈ E with orthogonal
ranges, i.e., t∗j tk = δjk 1 for j, k ∈ {1, 2}. The sequence sn := tn−1
2 t1 with t02 := 1
defines a unital C *-morphism H (0) from O∞ into E, because s∗j sk = δjk 1 for
j, k ∈ N, and O∞ is the universal C *-algebra with this defining relations. Since
O∞ is simple, cf. Corollary 2.2.7, the C *-morphisms are injective.
We get another *-monomorphism H (1) from O∞ into E that maps the canonical
generators s1 , s2 , . . . of O∞ ∼
= H (0) (O∞ ) to H (1) (sn ) := usn for n ∈ N.
By assumption there exists a point-norm continuous path ξ ∈ [0, 1] 7→ Hξ into
the unital C *-morphisms Hξ : O∞ → E, such that Hk = H (k) for k ∈ {0, 1}.
512 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Then pξ := Hξ (s1 s∗1 ) is a continuous path in the projections with p0 = s1 s∗1

and p1 = H (1) (p0 ). Thus, there exists V ∈ U0 (E) with V ∗ p0 V = p1 = up0 u∗ ,
cf. Lemma 4.1.3(v).
We get V up0 = p0 V u for the splitting projection p0 = s1 s∗1 . Now V ∈ U0 (E)
implies that [V u] = [u] = 0 in K1 (E). It follows that V u ∈ U0 (E) by Part (vi,β) of
Lemma 4.2.6. Using again V ∈ U0 (E), we obtain u ∈ V ∗ U0 (E) = U0 (E). 
Remark 4.2.12. The proof of Proposition 4.2.11 shows that two unital C *-
morphisms h1 , h2 : O∞ → E are homotopic if h1 (s1 ) = h2 (s1 ).

Notice that following Lemma 4.2.13 is not trivial, because it does not require
that A is simple.
Lemma 4.2.13. Let A a purely infinite unital C*-algebra, let e, f ∈ A+ with
0 ≤ f ≤ e ≤ 1 and S ∈ A an isometry such that S ∗ f S is contained in the closed
ideal J := I(1 − e) generated by (1 − e).
Then there exists an isometry T ∈ A and δ ∈ (0, 1] that satisfy kT ∗ f T k =
1 − δ < 1.

Proof. The inequality 0 ≤ (1 − e) ≤ (1 − f ) implies that I(1 − e) ⊆ I(1 − f ).

Thus, S ∗ f S ∈ I(1 − f ), and for each ε > 0 there exist g1 , . . . , gn ∈ A such that
n
X
kS ∗ f S − gk∗ (1 − f )gk k < ε
k=1

Combined with the equation S f S = 1 − S ∗ (1 − f )S it shows that 1A is contained

∗

in the closed ideal I(1 − f ) generated by 1 − f .

If A is purely infinite then this is equivalent to the existence of dε ∈ A with
k1 − d∗ε (1 − f )dε k < ε for each ε ∈ (0, 1/2) .
CHECK AGAIN THE ESTIMATES!! :
There exists δ ∈ (0, 1), – depending on the norm kd1/4 k –, with
k1 − d∗1/4 (1 − δ − f )+ d1/4 k < 1/4 .
1/2
The element T := C(C ∗ C)−1/2 with C := (1 − δ − f )+ d1/4 is an isometry in A
1/n
with kT ∗ f T k ≤ 1 − δ < 1, because kT ∗ f T k ≤ supn k(1 − δ − f )+ f k . 

Definition 4.2.14. We say that a unital C *-algebra A has the “squeezing”

property if A satisfies following Property (sq):

(sq) For each a ∈ A there exist isometries r1 , r2 ∈ A (depending on a) with

kr1∗ ar2 k ≤ 2kak/3.

It is still not known if C *-algebras with properly infinite units are K1 -injective,
even in case of weakly purely infinite C *-algebras. Almost all cases where one could
prove ( 5 ) the K1 -injectivity have implicitly to do with the now defined “squeezing”
Property (sq) that implies K1 -bijectivity.
5 – up to August 2018 –
 2. ON K-THEORY OF PROPERLY INFINITE C *-ALGEBRAS 513

Proposition 4.2.15. If a unital C*-algebra A has the “squeezing” Property

(sq) of Definition 4.2.14 then A is K1 -bijective in sense of Definition 4.2.2.
Following examples (a - c) of C*-algebras satisfy Property (sq):

(a) En := C ∗ (s1 , . . . , sn ; s∗i sj = δij 1 ) , for n = 2, . . . , ∞.

(b) each unital purely infinite C*-algebra A, and
(c) the multiplier algebra A := M(B) of every stable σ-unital C*-algebra B.

The class of C*-algebras A, A1 , A2 , . . . with Property (sq) is invariant under

following constructions (i-vii):

(i) Infinite direct sums `∞ (A1 , A2 , · · · ),

(ii) tensor products A ⊗max B with any unital C*-algebra B,
(iii) Cb (X, A) for locally compact σ-compact Hausdorff spaces X,
(iv) quotients A/J, for closed ideals J 6= A of A,
(v) inductive limits with unital C*-morphisms,
(vi) fix-point algebras of permutation actions on the n-times tensor product
n
with n ≥ 2 : A⊗ := A ⊗ A ⊗ · · · ⊗ A , and
(vii) B + ϕ(O∞ ), for every unital C*-morphism ϕ : O∞ → M(B) if B is a σ-
unital C*-algebra and its multiplier algebra M(B) has the Property (sq)
(e.g. if B is stable – by Part (c) ).

In particular it follows:
If a unital C*-algebra A satisfies Property (sq) then the asymptotic corona
Cb ([0, ∞), A)/ C0 ([0, ∞), A) and the C*-algebras E := (A ⊗max C)/J, for every
unital C*-algebra C and every closed ideal J 63 1 of A ⊗max C, have Property (sq)
and are all K1 -bijective.
Stable coronas Qs (B) of σ-unital C*-algebras B have Property (sq) and are
K1 -bijective.
Examples (a) and permanence property (v) show that O∞ = indlimn En has
Property (sq) and is K1 -bijective.

Proof. We divide the proof into 3 groups.

(1st) shows that unital A with Property (sq) of Definition 4.2.14 contains isome-
tries s, t ∈ A with s∗ t = 0 – which implies K1 -surjectivity by Part (v) of Lemma
4.2.6 –, and that Property (sq) implies that A satisfies the necessary and sufficient
condition (i) of Lemma 4.2.10 for K1 -bijectivity. Moreover we reformulate Property
(sq) in a more applicable form.
(2nd) the Property (sq) has the quoted permanence properties (i-vii), and
(3rd) the under (a,b,c) listed classes of C *-algebras have Property (sq).

Ad(1st): The Property (sq) implies that A contains two isometries s1 , s2 with
“orthogonal ranges”, i.e., with s∗1 s2 = 0 . Indeed, for a := 1 exist by Property (sq)
isometries r1 , r2 ∈ A with kr1∗ r2 k ≤ 2/3. Let p := r1 r1∗ and q := 1 − r2 r2∗ . The
projections p, q satisfy kp(1 − q)k = kr1∗ r2 k ≤ 2/3. By Lemma 4.1.3(iv) there exists
514 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

a unitary w ∈ U0 (A) with p ≤ wqw∗ = 1 − (wr2 )(wr2 )∗ , i.e., r1∗ (wr2 ) = 0. Thus,
s∗1 s2 = 0 for the isometries s1 := r1 and s2 := wr2 , and A has a properly infinite
unit element. It implies that A is K1 -surjective by Lemma 4.2.6(v).
We derive a stronger and more flexible formulation of Property (sq):
If there exists δ0 ∈ (0, 1) with the property that, for each a ∈ A, there exist
isometries r1 , r2 ∈ A – depending on a – with kr1∗ ar2 k ≤ (1 − δ0 ) · kak , then we
can replace a by r1∗ ar2 and get isometries t1 , t2 ∈ A – depending on r1∗ ar2 – with
k(r1 t1 )∗ a(r2 t2 )k ≤ kak · (1 − δ0 )2 .
Repeats of this argument show that the Property (sq) is equivalent to the
property that for each element a ∈ A and given ε > 0 there exists isometries
r1 , r2 ∈ A with kr1∗ ar2 k < ε .
If we replace in Definition 4.2.14 the element a by s∗1 as2 with isometries s1 , s2 ∈
A with orthogonal ranges, i.e., s∗j sk = δjk 1, then we find isometries r1 , r2 ∈ A with
kr1∗ (s∗1 as2 )r2 k < ε. The element t1 := s1 r1 and t2 := s2 r2 are isometries in A that
satisfy
kt∗1 at2 k < ε, and t∗j tk = δjk 1 . (2.5)
It implies immediately that A fulfills the sufficient condition for K1 -injectivity in
Lemma 4.2.10(i) with v := 1 for each u ∈ U(A) . Indeed, take a := u and tk := rk
(k ∈ {1, 2}) for some isometries r1 , r2 ∈ A with kr1∗ ar2 k < 1/2 and r1∗ r2 = 0.
Induction over n in Inequality (2.5) shows that, for any finite subset F =
{a1 , . . . , an } ⊂ A and ε > 0, there exist isometries r1 , r2 ∈ A with kr1∗ xr2 k < ε for
all x ∈ F , and r1∗ r2 = 0.
The induction step from n to n + 1 uses that isometries s, t ∈ A exist with
ks∗ (r1∗ an+1 r2 )tk < ε. Then (r1 s)∗ (r2 t) = 0 and k(r1 s)∗ ak (r2 t)k < ε for k =
1, . . . , n.

Ad(2nd): C *-algebras A with the above derived formally stronger property, –

but that is equivalent to Property (sq) –, have the permanence properties (i-vi):
(i): If A1 , A2 , . . . are unital C *-algebras with Property (sq), then the algebra
`∞ (A1 , A2 , · · · ) of bounded sequences a := (a1 , a2 , . . .) with ak ∈ Ak is a unital
C *-algebra. If ε > 0 is given then we find isometries j ∈ {1, 2}, rk,j ∈ Ak such that
∗
krn,1 an rn,2 k < ε/2. The rj := (r1,j , r2,j , . . .) are isometries in `∞ (A1 , A2 , . . .) and
satisfy kr1∗ ar2 k < ε.
(ii): If A has Property (sq) and B is a unital C *-algebra, then the tensor
product A ⊗max B has Property (sq).
Indeed, let c ∈ A ⊗max B and ε > 0. There exist n ∈ N, a1 , . . . , an ∈ A and
b1 , . . . , bn ∈ B with kbk k ≤ 1 such that in A ⊗max B holds

k c − (a1 ⊗ b1 + . . . + an ⊗ bn ) k < ε/2 .

By the (2nd) observation there exist isometries t1 , t2 ∈ A with t∗1 t2 = 0 and

kt∗1 ak t2 k < ε/2n for k = 1, . . . , n . Then r` := t` ⊗ 1 ∈ A ⊗max B (` = 1, 2)
are isometries that satisfy kr1∗ cr2 k < ε .
 2. ON K-THEORY OF PROPERLY INFINITE C *-ALGEBRAS 515

Induction shows that moreover for every finite sequence c1 , . . . , cn ∈ A ⊗max B

and ε > 0 there exists isometries t1 , t2 ∈ A with t∗1 t2 = 0 that satisfy the inequalities

k(t∗1 ⊗ 1)ck (t2 ⊗ 1)k < ε for 1 ≤ k ≤ n .

(iii): Suppose that A is unital and has property (sq), and that X is a locally
compact σ-compact Hausdorff space. The σ-compactness of X is equivalent to the
property that C0 (X)+ contains a function x 7→ e(x) ∈ [0, 1] that defines a strictly
positive element e of the C *-algebra C0 (X) with kek = 1. We fix such a function
e(x).
Let X 3 x 7→ a(x) ∈ A a continuous map with ka(x)k ≤ 1 for all x ∈ X. We
show the existence of isometries S, T ∈ Cb (X, A) with T ∗ S = 0, and

kT (x)∗ a(x)S(x)k ≤ e(x)/2 for all x ∈ X .

In particular, kT ∗ aSk ≤ 1/2 and T ∗ aS ∈ C0 (X, A).

We consider A as the C *-subalgebra of Cb (X, A) consisting of constant A-
valued functions from X to A. The subsets

Kn := e−1 [2−n , 1] = {x ∈ X ; e(x) ≥ 2−n }

build an increasing family of compact “reference” subsets of X with the properties:

[
Kn−1 ⊆ Kn◦ and Kn = X ,
n

that follow from

Kn−1 = e−1 ([2−(n−1) , 1]) ⊆ e−1 (2−n , 1] ⊆ Kn◦ ⊆ Kn = e−1 ([2−n , 1]) .

The arguments need some care because we can not suppose that the interior of
e−1 (2−n ) is empty, – different to the case e.g. of the function f0 (τ ) := τ on (0, 1] –
in place of e for (0, 1] in place of X.
We use the functions hn ∈ C0 (0, 1]+ considered in Section 22 of Appendix A
to build a suitable decomposition of 1 ∈ Cb (X, A) inside Cb (X)+ :
The increasing functions hn ∈ C0 (0, 1]+ are given by

hn (τ ) := min 1, max(2n τ − 1, 0) , for τ ∈ [0, 1] , n = 0, 1, . . . .

The set (2−n , 1] is the open support of hn , contained in h−1

n+1 (1) = [2
−n
, 1] . Thus,
hn hn+1 = hn and the non-negative function hn+1 − hn has support in the open
interval (2−(n+1) , 2−(n−1) ). It implies that the below used functions ψn := (hn −
hn−1 )1/2 (applied to the values e(x) of e) are well-defined and have the property
that ψn ψm = 0 for |n − m| > 1. Notice that h0 := 0 , h1 = (2f0 − 1)+ , . . ., and in
general hn = (2n f0 − 1)+ − (2n−1 f0 − 1)+ in the algebra C0 (0, 1] and for n ∈ N.
It follows that for each γ ∈ (0, 1] only finitely many functions hn − hn−1 have
P
support in [γ, 1]. In particular, n≥1 (hn − hn−1 ) converges on each interval [γ, 1]
uniformly to 1. Notice that

(hn ◦ e)−1 (0, 1] = e−1 (h−1 −1

(2−n , 1] ⊆ Kn◦ ⊆ Kn .


n (0, 1]) = e
516 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

It shows that (hn+1 ◦ e)(x) = 1 if and only if x ∈ e−1 ([2−n , 1]) = Kn . The support
of hn ◦ e is contained in the interior Kn◦ of Kn because hn+1 hn = hn .
Hence, the open support of the non-negative function hn (e(x)) − hn−1 (e(x)) is
contained in Vn := Kn◦ \ Kn−2 . It says that the support of (hn ◦ e) − (hn−1 ◦ e) is
an open subset of X that is contained in the open subset Vn of X. Obviously Vn is
◦
contained in the closed subset Fn := Kn \ Kn−2 of X. Summing up, the functions

1/2
ϕn (x) := hn (e(x)) − hn−1 (e(x)) := ψn (e(x)) with ψn := (hn − hn−1 )1/2 for
n = 1, 2, . . . earn from the hn − hn−1 ≥ 0 the properties (I), (II) and (III) :

(I) ϕn ϕm = 0 if |n − m| > 1 ,
(II) The open supports Un := ϕ−1 n ((0, 1]) of ϕn are contained in the open
subsets Vn := Kn◦ \ Kn−2 . And Vn is contained in the closed subset
◦
Fn := Kn \ (Kn−2 ).
P∞ 2
(III) ϕ
n=1 n (x) converges on each compact subset Y of X uniformly to
1. More precisely: There exists ν(Y ) ∈ N such that ϕn |Y = 0 for all
n > ν(Y ).
P
Parts (II) of this property list shows that n ϕn (x)an is a well-defined element
in Cb (X, A) of norm ≤ 2 if a1 , a2 , . . . ∈ A is an arbitrary sequence of contractions.
We define isometries S, T ∈ Cb (X, A) by the sums
∞
X ∞
X
S(x) := ϕn (x)sn and T (x) := ϕn (x)tn ,
n=1 n=1

Here sn , tn are the below constructed isometries in A with mutually orthogonal

ranges, i.e., with s∗m tn = 0 and s∗m sn = δm,n 1 = s∗m sn for all m, n ∈ N . The in
this way defined S and T are always in Cb (X, A), and satisfy S ∗ S = 1 = T ∗ T ,
S ∗ T = 0.
We complete the proof by an inductive selection of isometries sn , tn ∈ A with
above orthogonality properties and such that the corresponding isometries T and
S satisfy
kT (x)∗ a(x)S(x)k ≤ 4−n for all x ∈ Kn◦ \ Kn−2 .
This implies kT (x)∗ a(x)S(x)k ≤ e(x)/2 for all x ∈ X because e(x) ∈ [2−n , 4 · 2−n )
on Kn \ Kn−2 .
In the (1st) step it was shown that A has a properly infinite unit if A has
Property (sq). Let r1 , r2 ∈ A isometries in A with r1∗ r2 = 0, and let yn := r1n r2
∗ ∗ ∗
and zn := r2n r1 . Then ym yn = δm,n 1 = zm zn and ym zn = 0 for all m, n ∈ N .
We define by induction sequences of isometries (g1 , g2 , . . .) and (h1 , h2 , . . .) with
gn , hn ∈ A, and build from them and from yn , zn new sequences (s1 , s2 , . . .) and
(t1 , t2 , . . .) of isometries in A with the desired property by the following rules:

(α) s1 := r1 r2 · g1 and t1 := r2 r1 · h1 ,
(β) sn+1 := r1 sn gn+1 and tn+1 := r2 tn hn+1 .

Then s1 = y1 g1 , s2 = y2 g1 g2 , sn = yn g1 · . . . · gn and tn = zn h1 · . . . · hn . To
get the desired estimate we have to require that the isometries gn , hn ∈ A satisfy
 2. ON K-THEORY OF PROPERLY INFINITE C *-ALGEBRAS 517

the following inequalities, where an := a|Kn . We must consider also the overlap for
the x ∈ X with ϕn (x)ϕn+1 (x) 6= 0 .
It implies that we must find in each step pairs of isometries (g1 , h1 ), respectively
(gn+1 , hn+1 ), in A that fulfill the three inequalities kh∗1 c1,k g1 k < 4−1 (k = 1, 2, 3)
for given elements c1,1 , c1,2 , c1,3 ∈ C(K2 , A), respectively kh∗n+1 cn,k gn+1 k < 4−1 for
– in induction process new defined – given elements cn,1 , cn,2 , cn,3 ∈ C(Kn+2 , A).
In our case here we have to take in C(K2 , A) the 3 elements

c1,1 := r1∗ r2∗ a2 r1 r2 , c1,2 := r1∗ (r2∗ )2 a2 r1 r2 and c1,3 := r1∗ r2∗ a2 (r1 )2 r2 .

In the induction step we have to consider in C(Kn+2 , A) the elements

cn,1 := t∗n r2∗ an+2 r1 sn , cn,2 := t∗n (r2∗ )2 an+2 r1 sn and cn,3 := t∗n r2∗ an+2 (r1 )2 sn .

If A has Property (sq) and B := C(K) for a compact space K then the proof
of Part (2nd,ii) shows moreover that – for c1 , . . . , cn ∈ C(K, A) and ε > 0 – there
exist isometries s, t ∈ A ⊆ C(K, A) with t∗ s = 0 and kt∗ ck sk < ε for k = 1, . . . , n .
This implies the existence of the desired gn+1 , hn+1 ∈ A :
Simply put here ε := 4−n and take for c1 , c2 and c3 the above considered elements
in C(Kn+2 , A). Then we can take gn+1 := s and hn+1 := t. This finishes the proof
of the existence of the proposed isometries gn+1 and hn+1 in A, that define together
with r1 , r2 the desired new isometries sn+1 and tn+1 in A by composition rules (α)
and (β).
(iv): If A is unital and has property (sq) and J 6= A is a closed ideal of A,
then A/J is unital. Let b ∈ A/J and ε > 0. There exists a ∈ A with πJ (a) = b and
isometries r1 , r2 ∈ A with kr1∗ ar2 k < ε . The isometries tj := πJ (rj ) ∈ A/J satisfy
kt∗1 bt2 k < ε .
(v): If A is the inductive limit of a directed net of unital morphisms between
C *-algebras, then A has the “local” property that for each contraction a ∈ A and
ν ∈ (0, 1/4) there exists a unital C *-algebra B with property (sq), an element b ∈ B
with kbk ≤ 1, isometries r1 , r2 ∈ B that satisfy r1∗ r2 = 0 and kr1∗ br2 k ≤ ν, and a
unital C *-morphism ψ : B → A with kψ(b) − ak < ν. Then t1 := ψ(s1 ) and t2 :=
ψ(s2 ) are isometries in A with orthogonal ranges that satisfy kt∗1 at2 k ≤ 2ν < 1/2.
(vi): Suppose that the unital C *-algebra A has property (sq), and let
n n
φ : A⊗ → A⊗ a permutation *-automorphism that is defined via

φ(a1 ⊗ a2 ⊗ · · · ⊗ an ) := aπ(1) ⊗ aπ(2) ⊗ · · · ⊗ aπ(n)

by a permutation π of {1, . . . , n}.

n
As we have seen in (1st) Part and Part (ii) that for every b ∈ A⊗ and ε > 0
there exist isometries t1 , t2 ∈ A such that t∗1 t2 = 0 and

k (t∗1 ⊗ 1 ⊗ · · · ⊗ 1)b(t2 ⊗ 1 ⊗ · · · ⊗ 1) k < ε .

518 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Thus sk := tk ⊗ tk ⊗ · · · ⊗ tk , k ∈ {1, 2} are isometries with φ(sk ) = sk , s∗1 s2 = 0

and ks∗1 bs2 k < ε . The s1 and s2 are contained in the fix-point algebra of all actions
n
of permutation automorphisms of A⊗ .
(vii): The C *-algebra M(B) has a properly infinite 1, i.e., contains isometries
S, T ∈ M(B) with T ∗ S = 0, and M(B) is K1 -bijective by (1st) Part. For each
b ∈ B and ε ∈ (0, 1) there exists isometries r1 , r2 ∈ M(B) with r2∗ r1 = 0 and
kr2∗ br1 k < ε by proof of (1st) Part. We may suppose here that 1 − r1 r1∗ − r2 r2∗ is a
full and properly infinite projection, otherwise replace r2 by r2 r1 .
Let ϕ : O∞ → M(B) a unital C *-morphism, X ∈ B + ϕ(O∞ ) and ε ∈ (0, 1/8).
Then X = c + ϕ(d) with c ∈ B and d ∈ O∞ . O∞ has Property (sq) because
it is the inductive limit of the En in Part (a) and Part (v) applies (or use Part
(b) because O∞ is purely infinite by Corollary 2.2.7). If follows that there exists
isometries t1 , t2 ∈ O∞ with t∗2 t1 = 0 and kt∗2 dt1 k < ε. Here 1 − (t1 t∗1 + t2 t∗2 )
is a full and properly infinite projection ( 6 ). Define b := ϕ(t2 )∗ dϕ(t1 ) ∈ B.
Then k(ϕ(t2 )S2 )∗ X(ϕ(t1 )S1 )k < ε + kS2∗ bS1 k for all isometries S1 , S2 ∈ M(B).
The Property (sq) of M(B) implies that we find isometries S1 , S2 ∈ M(B) with
kS2∗ bS1 k < ε and S2∗ S1 = 0. Here we can replace S2 by S2 S1 to make sure that
1 − S1 S1∗ − S2 S2∗ is full and properly infinite. Then we find a unitary v ∈ U(M(B))
with vϕ(t1 ) = S1 , vϕ(t2 ) = S2 and [v] = 0 ∈ K1 (M(B)). The Property (sq) of
M(B) causes the K1 -injectivity of M(B), by (1st) Part. Thus, v ∈ U0 (M(B))
and there exist f1 , . . . , fn ∈ M(B) with fk∗ = −fk and v = exp(f1 ) · . . . · exp(fn ).
Let {ej } ⊂ B+ (j ∈ N) a quasi-central approximate unit in B, build e.g. by func-
tional calculus from a strictly positive contraction in B+ . The unitaries Uj :=
exp(ej f1 ej ) · . . . · exp(ej fn ej )) are in 1M(B) + B and limj→∞ kUj∗ xUj − v x vk = 0 for
each x ∈ B. Thus, there exists U ∈ U(B + C · 1) ∩ (1 + B) with kU ∗ bU − v ∗ bvk < ε.
We get k(U ϕ(t1 ))∗ b(U ϕ(t2 ))k ≤ 2ε. Then Tk := ϕ(tk )U ϕ(tk ) (k = 1, 2) are isome-
tries in B + ϕ(O∞ ) with kT2∗ XT1 k < 3ε < 1/2.

(Ad 3rd) Some algebras with Property (sq):

The argument at the beginning of the proof of (1st) shows also that is suffices
to require for all elements a in a dense subset X of A that there exists a constant
γ ∈ (0, 1) such that for each a ∈ X there exist isometries r1 , r2 (depending from
a) with kr1∗ ar2 k < γkak, e.g. we could start with the property that there exist for
each a ∈ X isometries r1 , r2 ∈ A with 3kr1∗ ar2 k ≤ 2kak.
(a): En := C ∗ ( s1 , . . . , sn ; s∗i sj = δij ) , n = 2, . . . , ∞ , satisfy Property (sq):
Indeed, let w := k1 k2 . . . kp , v := `1 `2 . . . `q , “words” of length p, q ∈ N with
“letters” kj , `i ∈ {1, . . . , n} from the “alphabet” {1, . . . , n}. Define T0 := 1 and an
isometry by the product

Tp (w) := sk1 sk2 · . . . · skp .

6 Use here that [1] = 1 by K (O ) ∼ Z, or simply replace t by t t to make sure that

0 ∞ = 2 2 1
1 − t1 t∗1 − t2 t∗2 is full and properly infinite.
 2. ON K-THEORY OF PROPERLY INFINITE C *-ALGEBRAS 519

It is easy to see that the linear span of the element 1 and of the “elementary”
products Tq (w), Tq (w)∗ and Tq (w)(Tr (v)∗ ) is dense in the C *-algebra En (in both
of real or complex case), because the product Tq (v)∗ Tp (w) is equal to 1 if v = w, is
equal to zero if there exists g ≤ r := min(p, q) with kg 6= `g , is equal to T(p−q) (u1 )
if q < p and w = vu1 and is equal to T(q−p) (u2 )∗ if p < q and v = wu2 .
We denote by G one of this elementary products or let G = 1. A sort of “length”
L(G) is defined by L(1) := 0, L(Tq ) := q, L((Tq )∗ ) := q and L(Tq · (Tr )∗ ) := q + r.
Clearly, s∗k Tp (w) = δk,k1 Tp−1 (v) for all k = 1, . . . , n with word v := k2 k3 . . . kn for
w = k1 k2 . . . kn .
Let G = Tp (w)Tq (v)∗ with w = k1 . . . kp , v = `1 . . . `q (p, q ≥ 1), then,
L(s∗k Gs` )
= L(G) − 2 if and only if k1 = k and `1 = `, otherwise only s∗k Gs` = 0
can happen.
In case G = Tp (w), w = k1 . . . kp , we get s∗k Gs` 6= 0 if and only if k = k1
and then L(s∗k Gs` ) = L(G), because then G = Tp (u) with u = k2 . . . kp `. Since
L(G∗ ) = L(G) the same holds for G = Tp (w)∗ with the role of k and ` interchanged.
Pm
Therefore, one can find, for each linear combination C = α0 1 + j=1 αj Gj
(with scalars αj ), an index 1 ≤ j0 ≤ m and suitable k, ` ∈ {1, . . . , n} such that
k 6= `, s∗k Gj0 s` = 0 for at least one j0 ∈ {1, . . . , m} and L(s∗k Gj s` ) ≤ L(Gj ) for
j 6= j0 . Thus s∗k Cs` = j6=0,j6=j0 αj s∗k Gj s` with elementary products s∗k Gj s` (it
P

can be equal to 1).

Iteration – say m-times – of this operation leads to words w and v in the alpha-
bet {1, . . . , n} such that for the isometries Tm (w) and Tm (v) holds Tm (w)∗ Tm (v) =
0 and Tm (w)∗ CTm (v) = 0. It says that En has Property (sq).
(b): The proof for purely infinite unital C *-algebras is much more engaged
than the proof for strongly purely infinite unital C *-algebras. The latter class
contains all simple unital purely infinite C *-algebras by Proposition 2.2.1(v).
The question whether or not pure infiniteness implies the – for our considera-
tions important – strong pure infiniteness is still unsolved (June 2017).
Let us give first, for the readers convenience, the obvious and easy proof for
strongly purely infinite C *-algebras. This algebras have the “matrix diagonaliza-
tion” property, cf. Definition 2.16.3 and Proposition 2.16.4:
If a ∈ A is a contraction then we can build the positive 2 × 2-matrix b = [bjk ] ∈
M2 (A)+ with diagonal entries b11 := 1 =: b22 and b∗21 := b12 := a. By the matrix
diagonalization property of A there exists a diagonal matrix D := diag(d1 , d2 ) with
kD∗ bD − 12 k < 1/4 , i.e., kd∗1 ad2 k < 1/4 and kd∗j dj − 1k < 1/4 for j = 1, 2. Thus,
the positive elements d∗j dj are invertible with spectrum Spec(d∗j dj ) ⊆ (3/4, 5/4)
and sj := dj (d∗j dj )−1/2 are isometries with ks∗1 as2 k ≤ (1/4)(4/3) = 1/3.
The different and more engaged proof of Part (b) for the general case of purely
infinite unital C *-algebras A goes as follows:
520 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Since 1 ∈ A is properly infinite for purely infinite A there exists isometries

S1 , S2 ∈ A with orthogonal ranges, i.e., Sj∗ Sk = δjk 1:
The isometries Tk := S2k S1 ∈ A satisfy Tk∗ T` = δjk 1, and C ∗ (T1 , T2 , . . . ) ⊂ A is
a copy O∞ in A .
Let a ∈ A an element of norm kak = 1 . We are going to show that there exist
isometries s, t ∈ A such that kt∗ ask < 1 :
If kT2∗ aT1 k < 1 , then we can take s := T1 and t := T2 . If kT2∗ aT1 k = 1, then
we consider (from now on) the closed ideal J := I(1−T1∗ a∗ T2 T2∗ aT1 ) of A generated
by 1 − T1∗ a∗ T2 T2∗ aT1 ∈ A+ :
If 1A is contained in J, then Lemma 4.2.13 applies with e := f := T1∗ a∗ T2 T2∗ aT1
and S := 1, and implies that there exists an isometry T ∈ A and δ ∈ (0, 1] with
kT ∗ f T k = 1 − δ < 1 . Thus, kT2∗ a(T1 T )k = (1 − δ)1/2 < 1 , and we can take
s := T1 T and t := T2 .
If 1A is not contained in J, then πJ (T1∗ a∗ T2 T2∗ aT1 ) = 1. Thus πJ (T2∗ aT1 ) is
an isometry in A/J. Then πJ (T2∗ aT1 T3 ) = πJ (T2∗ aT1 )πJ (T3 ) is again an isometry
in A/J. The contraction Z := T2∗ aT1 T2 ∈ A satisfies πJ (Z ∗ )πJ (T2∗ aT1 T3 ) = 0 and
πJ (Z ∗ Z) = 1.
We define a closed left ideal of A by

L := { b ∈ A ; b · T2∗ aT1 T3 ∈ J } .

It contains Z ∗ and all elements of J , i.e., J ∪ {Z ∗ } ⊆ L.

Let K denote the closed ideal of A that is generated by the hereditary C *-
subalgebra D := L∗ ∩ L. Then L∗ · L ⊆ D, and D contains J and the contraction
ZZ ∗ . It follows that the closed ideal K of A contains J and Z ∗ (ZZ ∗ )Z. The
equation πJ (Z ∗ )πJ (ZZ ∗ )πJ (Z) = πJ (1) shows that 1 − Z ∗ (ZZ ∗ )Z ∈ J. Hence,
1 ∈ K and K = A.
The C *-algebra A and its full hereditary C *-subalgebra D are purely infinite,
and the ideal generated by D contains 1A . It implies the existence of elements
g ∈ A and d ∈ D+ such that g ∗ dg = 1.
The isometry V := d1/2 g ∈ A satisfies V ∗ ∈ L because V V ∗ ∈ D. Thus,
V ∗ T2∗ aT1 T3 ∈ J by definitions of L and D = L∗ ∩ L.
The positive contractions f := T1∗ a∗ T2 V V ∗ T2∗ aT1 , e := T1∗ a∗ T2 T2∗ aT1 and
isometry S := T3 satisfy the assumptions of Lemma 4.2.13, i.e., 0 ≤ f ≤ e and
S ∗ f S ∈ I(1 − e) = J .
By Lemma 4.2.13 there exists an isometry T ∈ A with kT ∗ f T k < 1. Thus,
kV ∗ T2∗ aT1 T k = kT ∗ f T k1/2 < 1 . It says that the isometries s := T1 T and t := T2 V
satisfy kt∗ ask < 1.
We have seen above that for each unital purely infinite A and a ∈ A with
kak = 1 there exists isometries s, t ∈ A such that kt∗ ask < 1 and t∗ s = 0. This
argument does not give immediately a fixed general constant γ0 ∈ (0, 1) with the
 2. ON K-THEORY OF PROPERLY INFINITE C *-ALGEBRAS 521

property that we for every a ∈ A there exist isometries s, t ∈ A (depending on a)

with kt∗ ask ≤ (1 − γ0 ) · kak . So that the iteration argument in the (1st) Part of
the proof does not apply here.
But we can give an indirect argument that proves that for each a ∈ A and
ε > 0 there are isometries s, t ∈ A (depending on a and ε) with kt∗ ask < ε:
Define for a ∈ A the number

γ(a) := inf{ kt∗2 at1 k ; t1 , t2 ∈ A , t∗j tk = δj,k 1 , } .

Obviously 0 ≤ γ(a) ≤ kak and γ(ξa) = |ξ| · γ(a). We show that γ(a) = 0 for all
a ∈ A:
Suppose – to derive a contradiction – that γ(a) > 0 for some a ∈ A with
kak = 1. Then we can find sequences tk,1 , tk,2 , . . . , tk,n , . . . ∈ A (k ∈ {1, 2}) with
t∗j,n tk,n = δj,k 1 and limn kt∗2,n at1,n k = γ(a). Let

bn := kt∗2,n at1,n k−1 t∗2,n at1,n .

Then b := (b1 , b2 , . . .) is a contraction in `∞ (A) that has the property that for every
isometries S, T ∈ `∞ (A) holds kT ∗ bSk = 1.
It is easy to see that C *-algebra `∞ (A) is again purely infinite if A is purely
infinite in sense of Definition 1.2.1, cf. the permanence properties of purely infinite
C *-algebras listed in Proposition 2.5.19. Hence, our above “individual” squeezing
result is also true for elements b ∈ `∞ (A).
But this general fact contradicts the existence of a contraction b ∈ `∞ (A) with
the property that kT ∗ bSk ≥ 1 for all isometries S, T ∈ `∞ (A).
It follows that the above defined number γ(a) ∈ [0, kak] is zero for all a ∈ A if
A is a unital purely infinite C *-algebra. In particular, we find then for each a ∈ A
isometries s, t ∈ A with 3kt∗ ask ≤ 2.
– (c): Let B a stable σ-unital C *-algebra and A := M(B). By Lemma A.23.1,
for every separable C *-subalgebra C ⊂ M(B) there exist isometries S, T ∈ M(B)
with the property that T ∗ CS ⊆ B and T ∗ S = 0 .
By Remark 5.1.1(8) we find a sequence of isometries sk ∈ M(B) with the
property that k sk s∗k converges strictly to 1 with respect of the strict topology on
P

M(B). If we select suitable skn from the sequence s1 , s2 , . . . then we can manage
that Tn := T skn and Tn := T skn satisfy limn kTn∗ cSn k = 0 and Tn∗ Sn = 0 for all
c ∈ C. Thus, A := M(B) has Property (sq). 

Now we turn to the generalization of K1 -injectivity and K1 -surjectivity to the

case of non-unital σ-unital C *-algebras :

Definition 4.2.16. If A is a not necessarily unital C *-algebra, then we can

consider the natural unitization A e := A + C1 ⊆ M(A) of A, and then use the
definitions of K1 -surjectivity and K1 -injectivity for A
e in place of E in Definition
4.2.2, where we notice that K1 (A) e = K1 (A) and each unitary u ∈ A + C1 is
522 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

homotopic in U(A)e to a unitary v ∈ U(A)

e ∩ (1 + A). Notice that our definition of
A entails that A = A if A is unital.
e e

I.e., A is K1 -surjective (respectively is K1 -injective) if the natural map

u = 1 + a ∈ U(A)
e 7→ [u] ∈ K1 (A)

from (1 + A) ∩ U(A + C · 1) to K1 (A) is surjective (respectively has kernel equal to

(1 + A) ∩ U0 (A)).
e

The C *-algebra A is K1 -bijective if A is K1 -surjective and K1 -injective.

Following Remark 4.2.17 lists some sufficient criteria for K1 -surjectivity and K1 -
injectivity of some special non-unital C *-algebras. It will be used for our “construc-
tive” version of K∗ -theory and the study of Cuntz semi-groups of C *-morphisms.

Remark 4.2.17. C *-algebras A with the property that the unit element of
M(A) is properly infinite appear often in our considerations.
Suppose that A is a σ-unital non-unital C *-algebra. The multiplier algebra
M(A) has a properly infinite unit, if and only if, there exists a unital C *-morphism
O∞ → M(A), and this is the case if and only if A is isomorphic to an ideal of some
C *-algebra E with properly infinite unit.
The class of C *-algebras A with properly infinite 1 ∈ M(A) is invariant under
passage to non-zero ideals and non-zero quotients.
If M(A) has a properly infinite unit, then A is K1 -surjective in the sense of
Definition 4.2.16.
If the multiplier algebra M(A) has a properly infinite unit, then the algebra A
is K1 -injective in the sense of Definition 4.2.16, if and only if, A + O∞ ⊆ M(A) is
a K1 -injective C *-algebra.
Every σ-unital stable C *-algebra A is K1 -bijective, because M(A) has the
“squeezing” Property (sq), and this property carries over to A + ψ(O∞ ) for every
unital C *-morphism ψ : O∞ → M(A), cf. Parts (c) and (vii) of Proposition 4.2.15.

Details for Remark 4.2.17. Let E a C *-algebra, J a closed ideal of E and

ψ : J → A an isomorphism from J onto A. Then M(ψ) is an isomorphism from
M(J) onto M(A) and there exists a unital C *-morphism ϕ : E → M(A). If the
unit of E is properly infinite then the unit 1M(A) = ϕ(1E ) of M(A) is properly
infinite. Conversely, if 1M(A) is properly infinite in M(A), then one can take
E := M(A) and J = A.
If A is a non-zero ideal of a C *-algebra E with properly infinite unit 1E then
every closed ideal {0} =
6 J of A is a closed ideal E. If J 6= A, then A/J is a non-zero
closed ideal of E/J and E/J has again a properly infinite unit.
If B is unital and O∞ ⊆ M(B) = B, then B = 1 + B is K1 -surjective, by
Lemma 4.2.6(v).
 2. ON K-THEORY OF PROPERLY INFINITE C *-ALGEBRAS 523

If A is non-unital and 1 ∈ M(A) is properly infinite, then there is copy of O∞

unitally contained in M(A) with O∞ ∩ A = {0}. With this copy there are natural
isomorphisms K1 (A) = K1 (A+C·1) ∼= K1 (A+O∞ ), and U(A+C·1)/U0 (A+C·1) ∼ =
U(A + O∞ )/U0 (A + O∞ ) and (A + 1) ∩ U(A + C · 1) / (A + 1) ∩ U0 (A + C · 1) ∼



=
U(A + C · 1)/U0 (A + C · 1) .
The proposed K1 -injectivity of M(A) implies that unital C *-morphisms
ϕ1 , ϕ2 : O∞ → M(A) are point-norm homotopic, cf. Proposition 4.2.11 i.e., there is
a norm-continuous path of unitaries u(ξ) ∈ U(M(A)) ξ ∈ [0, ∞) with u(1) = 1 and
limξ→∞ u(ξ)ϕ1 (sn ) = ϕ2 (sn ) for all n ∈ N. Such unitaries can be in strict topology
approximated by products of of exponentials with exponents −x∗ = x ∈ A. That
shows that the interaction with elements in A can be done by correction with
products by u1 , . . . , un ∈ (A + 1) ∩ U0 (A + C · 1). This gives hope for over carry of
Property (sq).
If A is stable and σ-unital, then the multiplier algebra M(A) has Property (sq)
and K∗ (M(A)) = 0, which implies U(M(A)) = U0 (M(A)) by K1 -bijectivity.
If A is not unital, but 1M(A) is properly infinite in M(A), then A + O∞ ⊂
M(A) is unital and K1 -surjective and the inclusion A + C · 1 ,→ A + O∞ defines
a natural isomorphism K1 (A) = K1 (A + C · 1) ∼ = K1 (A + O∞ ) that coincides with
∼
the isomorphism K1 (A) = K1 (A + O∞ ) given by splitting of the 6-term exact K∗ -
sequence for the (right-)split exact sequence 0 → A → A+O∞ → O∞ → 0, because
A ∩ O∞ = {0} if A is not unital, and K1 (O∞ ) = K1 (C) = 0.
Since A + O∞ is K1 -surjective by Lemma 4.2.6(v), for each x ∈ K1 (A + O∞ ) ∼ =
K1 (A + C · 1) there is a unitary u = a + v ∈ U(A + O∞ ) with a ∈ A and v ∈ U(O∞ )
such that x = [u] = [v] + [v ∗ a + 1]. But U(O∞ ) = U0 (O∞ ) ⊂ U0 (A + O∞ ) by [172,
cor. 3.12], cf. also Section 1 of Appendix A. Thus [v] = [1] = 0 ∈ K1 (A + O∞ ) and
x = [v ∗ a+1] ∈ K1 (A+O∞ ) ∼ = K1 (A+C·1) is the class of the unitary v ∗ a+1 ∈ A+1.
This establishes the K1 -surjectivity of A, i.e., shows the surjectivity of the map

u := a + 1 ∈ U(A + C · 1) ∩ (A + 1) 7→ [u] ∈ K1 (A) .

Suppose that A + O∞ (⊆ M(A)) is K1 -injective. If 0 = [a + 1] ∈ K1 (A) for

a + 1 ∈ U(A + C · 1), then 0 = [a + 1] ∈ K1 (A + O∞ ) and a + 1 ∈ U0 (A + O∞ ). Let
π : A + O∞ → O∞ the natural *-epimorphism π(a + b) = b for a ∈ A and b ∈ O∞ .
By the proposed K1 -injectivity of A + O∞ , there exist a continuous path ξ ∈
[0, 1] 7→ v(ξ) ∈ U(A + O∞ ) such that v(0) = 1 and v(1) = a + 1. Then u(ξ) :=
π(v(ξ)∗ ) · v(ξ) is a continuous path in U(A + C · 1) ∩ (A + 1) with u(0) = 1 and
u(1) = a + 1, i.e., a + 1 ∈ U0 (A + C · 1). Thus, A + C · 1 is K1 -injective.
Now suppose conversely that A+C·1 is K1 -injective and u = a+v ∈ U(A+O∞ )
satisfies 0 = [u] ∈ K1 (A + O∞ ). Notice π(u) = v ∈ U0 (O∞ ) = U(O∞ ). Then
π(u∗ )u = v ∗ a + 1 satisfies 0 = [v ∗ a + 1] ∈ K1 (A + C · 1) ∼
= K1 (A + O∞ ). Thus,
∗ ∗
v a + 1 ∈ U0 (A + C1) ⊂ U0 (A + O∞ ) and u = v · (v a + 1) ∈ U0 (A + O∞ ).
Every σ-unital stable C *-algebra A is K1 -bijective:
524 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

If A is stable, then there is a non-degenerate C *-morphism ψ : K → M(A)

with ψ(K) ∩ A = {0}. It extends to a strictly continuous unital *-monomorphism
M(ψ) : L(`2 ) ∼= M(K) → M(A). Let T1 , T2 , . . . ∈ L(`2 ) isometries such that
∗
P
n Tn Tn converge to id`2 in L(`2 ) with respect to the *strong operator topology.
Then sn := M(ψ)(Tn ) have the property that n sn s∗n converges strictly to 1 in
P

M(A).
If O∞ = C ∗ (s1 , s2 , . . .) ⊆ M(A) is a copy of O∞ with the additional property
that n sn s∗n converges strictly to 1 in M(A), then A + O∞ ⊆ M(A) satisfies
P

the squeezing condition (sq) of Definition 4.2.14 and is K1 -injective by Proposition

4.2.15 :
Indeed, let a + X ∈ A + O∞ with a ∈ A and X ∈ O∞ , and ε > 0. We find
snk ∈ O∞ with ks∗nk asnk+1 k < ε/2, by strict convergence to 1M(A) of
P ∗
sn sn .
Proposition 4.2.15(a,iii) provide isometries S, T ∈ O∞ that satisfy

kS ∗ (s∗nk Xsnk+1 )T k < ε/2 .

The isometries (snk S, snk+1 T ) are ε-squeezing for a + X.

Thus if ψ : O∞ → M(A) is a unital C *-morphism with the additional property
that n sn s∗n converges strictly to 1 in M(A), then A+ψ(O∞ ) has the “squeezing”
P

Property (sq) of Definition 4.2.14. It implies by Proposition 4.2.15 that A + ψ(O∞ )

is K1 -bijective.
If ϕ : O∞ → M(A) is any unital C *-morphism with the property that there
exists a unitary u ∈ M(A such that u∗ ϕ(b)u − ψ(b) ∈ A for all b ∈ O∞ , then the
C *-subalgebra A + ϕ(O∞ ) ⊆ M(A) satisfies again the “squeezing” Property (sq)
of Definition 4.2.14 and is therefore K1 -bijective. 

Notice here that the proof of Proposition 4.2.15(vii) shows that A+ϕ1 (O∞ ) has
Property (sq), if and only if, A + ϕ2 (O∞ ) has Property (sq) under the assumption
that M(A) is K1 -injective.
If ϕ : O∞ → M(A) is any unital C *-morphism then the C *-subalgebra A +
ϕ(O∞ ) ⊆ M(A) satisfies again the “squeezing” Property (sq) of Definition 4.2.14.

Remark 4.2.18. Let A be a σ-unital non-unital C *-algebra such that M(A) is

K1 -injective, as e.g. in case where A is stable, and let ϕk : O∞ → M(A), k ∈ {1, 2},
unital C *-morphisms.

( i) A + ϕ1 (O∞ ) has Property (sq), if and only if, A + ϕ2 (O∞ ) has Property
(sq).
(ii) If ϕ : O∞ → M(A) is any unital C *-morphism then the C *-subalgebra
A + ϕ(O∞ ) ⊆ M(A) satisfies again the “squeezing” Property (sq) of
Definition 4.2.14.

Details for Remark 4.2.18. Then ϕ1 and ϕ2 are homotopic by Proposition

4.2.11.
 2. ON K-THEORY OF PROPERLY INFINITE C *-ALGEBRAS 525

(i): Is true and provable, because arguments in the proof of Proposition

4.2.15(vii) apply here.
(ii): It follows from Proposition 4.2.15(vii), because M(A) satisfies the “squeez-
ing” Property (sq) of Definition 4.2.14 if A is σ-unital and stable. 

It would be useful for the above considerations if the following Conjectures

4.2.19 can be verified. Are there similar results for O2 in place of O∞ ?

Conjecture 4.2.19. Let A a separable stable C*-algebra and ϕk : O∞ →

M(A) (k = 1, 2) unital C*-morphisms.

(C. 1) The C*-subalgebras A + ϕ1 (O∞ ) and A + ϕ2 (O∞ ) are isomorphic.

(C. 2) There exist a unitary U ∈ M(A) such that U ∗ ϕ1 (b)U − ϕ2 (b) ∈ A for all
b ∈ O∞ .
(C. 3) The C*-morphisms ϕ1 , ϕ2 : O∞ → M(A) are approximately unitary
equivalent by a sequence of unitaries u1 , u2 , . . . ∈ M(A) with the addi-
tional property u∗k ϕ2 (b)uk − ϕ1 (b) ∈ A for all b ∈ O∞ .

Clearly (C. 3) implies (C. 2), and (C. 2) implies (C. 1).
Is (C. 1) also true for non-stable σ-unital non-unital separable A if it is true
for all stable separable A?

The following lemma explains the equivalence of rather different looking defin-
ing relations that are used for definitions of Ext-groups or KK-groups (in particular
that here in this book versus those given by other authors).

Lemma 4.2.20. Suppose that E is unital, D is a full hereditary C*-subalgebra

of E, and that A is a unital C*-subalgebra of E such that D is an ideal of A.

(o) There are isomorphisms

K∗ (A/D) ∼
= kernel(K∗ (A) → K∗ (E)) ,

which split the 6-term exact sequence into split exact sequences

0 → K∗ (D) → K∗ (A) → K∗ (A/D) → 0 .

(i) Let p1 and p2 projections in A, such that p2 − p1 ∈ D and [p1 ]E = [p2 ]E

in K0 (E). Then [p1 ]A = [p2 ]A in K0 (A).
(ii) Let u1 and u2 unitaries in A, such that u2 − u1 ∈ D and [u1 ]E = [u2 ]E in
K1 (E). Then [u1 ]A = [u2 ]A in K1 (A).
(iii) If A is K1 -surjective, then A/D is K1 -surjective.

The equations [p1 ]A = [p2 ]A in (i), or [u1 ]A = [u2 ]A in (ii), can fail if D / A is
not hereditary in E or is hereditary in E but is not full in E.
Give example!
526 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

This obstruction plays a role for the classifications of non-stable and non-unital
extensions, or unital extensions that do not dominate a unital “zero” extension. See
the ????
Give example for the latter !! Gabe’s remarks ???
Is A/D K1 -bijective if A and E (hence D ??) are K1 -bijective?

Proof. (o): The inclusion maps ηD : D ,→ A and ηA : A ,→ E define group

homomorphisms [ηD ]∗ : K∗ (D) → K∗ (A) and [ηA ]∗ : K∗ (A) → K∗ (E). The compo-
sition α := ηA ◦ ηD : d ∈ D 7→ d ∈ E of this homomorphisms defines isomorphisms
[α]∗ = [ηA ]∗ ◦ [ηD ]∗ from K∗ (D) onto K∗ (E), because D is full and hereditary in E.
It follows that [ηD ]∗ must be injective and that [ηA ]∗ is a surjective group
homomorphism from K∗ (A) onto K∗ (E). Thus, the boundary maps

∂ : K(∗ +1) mod 2 (A/D) → K∗ (D)

of the natural 6-term exact sequence must be trivial, i.e.,

0 → K∗ (D) → K∗ (A) → K∗ (A/D) → 0

are exact sequences with group morphisms [ηD ]∗ and [πD ]∗ : K∗ (A) → K∗ (A/D)
defined by the quotient map πD : A → A/D.
In particular, the maps [πD ]∗ are epimorphisms with kernels [ηD ]∗ (K∗ (D)).
Natural “splitting” epimorphisms from K∗ (A) onto K∗ (D) can be defined by

β∗ := [α]−1
∗ ◦ [ηA ]∗ : K∗ (A) → K∗ (D) .

We define a group endomorphisms γ∗ : K∗ (A) → K∗ (A) with (γ∗ )2 = γ∗ by

γ∗ := [idA ]∗ − [ηD ]∗ ◦ β∗ .

The endomorphism γ∗ maps K∗ (A) onto the kernel of [ηA ]∗ : K∗ (A) → K∗ (E). The
“orthogonal” idempotent group endomorphism [idA ]∗ − γ∗ = [ηD ]∗ ◦ [α]−1
∗ ◦ [ηA ]∗ of
K∗ (A) maps K∗ (A) onto [ηD ]∗ (K∗ (D)).
The restriction of [πD ]∗ onto the kernel γ∗ (K∗ (A)) of [ηA ]∗ : K∗ (A) → K∗ (E)
defines an isomorphism [πD ]∗ ◦ γ∗ from the kernel of [ηA ]∗ : K∗ (A) → K∗ (E) onto
K∗ (A/D).
Thus, the kernel of [ηA ]∗ : K∗ (A) → K∗ (E) is isomorphic to K∗ (A/D) in a
natural way by [πD ]∗ ◦ γ∗ .
We get in particular that

[ηA ]∗ ⊕ [πD ]∗ : K∗ (A) → K∗ (E) ⊕ K∗ (A/D)

are group isomorphisms from K∗ (A) onto K∗ (E) ⊕ K∗ (A/D) that realize the split-
ting of K∗ (D) → K∗ (A) → K∗ (A/D).
(i,ii): Let p1 , p2 ∈ A projections (respectively u1 , u2 ∈ U(A)). Consider x0 :=
[p2 ] − [p1 ] ∈ K0 (A) (respectively x1 := [u2 ] − [u1 ] ∈ K1 (A)).
 3. ADDITION OF C *-MORPHISMS 527

If p2 −p1 ∈ D (respectively u2 −u1 ∈ D) then [πD ]∗ (x∗ ) = 0. If [p1 ]E = [p2 ]E in

K0 (E) (respectively [u1 ]E = [u2 ]E in K1 (E)), then [ηA ]∗ (x∗ ) = 0 in K∗ (E). Above
we have seen that [ηA ]∗ ⊕ [πD ]∗ is faithful on K∗ (A). Thus, the assumptions of
Parts (ii) or (iii) imply that x∗ = 0 in K∗ (A).
(iii): If u ∈ U(A) 7→ [u] ∈ K1 (A) and [πD ]1 : K1 (A) → K1 (A/D) are surjective,
then for each x ∈ K1 (A/D) there exist u ∈ U(A) with x = [πD ]1 ([u]) = [πD (u)].
Thus, Part (iii) follows from πD (U(A)) ⊆ U(A/D). 

3. Addition of C *-morphisms

We apply the above definition of ⊕s1 ,s2 to C *-morphisms and completely posi-
tive maps, and study later associated commutative semigroups and the correspond-
ing Grothendieck groups. Throughout this section we consider two C *-algebras D
and E such that E is unital and contains a copy of O2 unitally.

Definition 4.3.1. Let h1 , h2 : D → E be two C *-morphisms (or completely

positive maps) and let s1 , s2 be generators of a unital copy of O2 in E. Then the
Cuntz addition of h1 and h2 (with respect to s1 , s2 ) is defined by:

(h1 ⊕s1 ,s2 h2 )(b) := s1 h1 (b)s∗1 + s2 h2 (b)s∗2 , ∀b ∈ D.

Two C *-morphisms h, k : D → E are said to be unitarily equivalent (nota-

tion: [h] = [k]) if there exists a unitary u in E such that k(b) = u∗ h(b)u for all
b ∈ D (notation: k = u∗ h(·)u).
Sometimes one has to consider only unitary equivalences k = u∗ h(·)u with
unitaries u in a subgroup G of the unitary group U(E) of E, where G often contains
the connected component U0 (E) of the unit element 1E in U(E) (notation: h ≈G k).
We write h ≈ k or [h]≈ = [k]≈ if here G := U0 (E), i.e.,

h≈k ⇔ k ∈ [h]≈ ⇔ ∃ u ∈ U0 (E) with k = u∗ h(·)u .

Cuntz addition defines the structure of a commutative semigroup on the set

[Hom(D, E)] of unitary equivalence classes of C *-morphisms from D into E, by
setting
[h1 ] + [h2 ] := [h1 ⊕s1 ,s2 h2 ]

for arbitrary choice of generators {s1 , s2 } of O2 in E. The correctness of this

definition can be verified using the following Proposition 4.3.2.
It shows moreover that the more refined addition

[h1 ]G + [h2 ]G := [h1 ⊕s1 ,s2 h2 ]G

is well-defined if U0 (E) ⊆ G ⊆ U(E) and defines the structure of a commutative

semigroup on the set of all equivalence classes [h]G . We write [Hom(D, E)]≈ in case
G = U0 (E).
528 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Proposition 4.3.2. Let hi : D → E, i = 1, 2, 3 be C*-morphisms and let

{s1 , s2 } and {t1 , t2 } be canonical generators of copies of O2 in E. The Cuntz-
addition [h1 ] + [h2 ] := n[h1 ⊕s1 ,s1 h2 ] on unitary equivalence classes [h] of C*-
morphisms h : D → E is well-defined, independent from the chosen canonical gen-
erators s1 , s2 of O2 ⊆ E, commutative and associative:

(i) (⊕ is well-defined) For any unitary operators v and w in E,

vh1 v ∗ ⊕s1 ,s2 wh2 w∗ = u(h1 ⊕s1 ,s2 h2 )u∗ ,

where u := s1 vs∗1 + s2 ws∗2 is a unitary operator in E.

(ii) (Consistency)

u∗ ((h1 ⊕s1 ,s2 h2 )(b))u = (h1 ⊕t1 ,t2 h2 )(b), b ∈ D,

where u := s1 t∗1 + s2 t∗2 is a unitary operator in E;

(iii) (Commutativity)

Uc (h1 ⊕s1 ,s2 h2 )Uc∗ = h2 ⊕s1 ,s2 h1 ,

where Uc := s2 s∗1 − s1 s∗2 is a unitary operator in the connected component

U0 (E) of 1E in U(E);
(iv) (Associativity)

Ud (h1 ⊕ (h2 ⊕ h3 ))Ud∗ = (h1 ⊕ h2 ) ⊕ h3 ,

where ⊕ means ⊕s1 ,s2 and Ud := s21 s∗1 + s2 (s∗2 )2 + s1 s2 (s2 s1 )∗ is in U0 (E).

Proof. Each of Parts (i)–(iv) can be seen by straight calculation – compare

introductory remarks in the
proof of Lemma 4.2.6 and the
proofs of Parts (o) and (vii) of Lemma 4.2.6. The unitaries Uc and Ud are in
U0 (C ∗ (s1 , s2 )), cf. proof of Part (o) of Lemma 4.2.6. 

The observation that Uc , Ud ∈ U0 (C ∗ (s1 , s2 )) follows also from the less ele-
mentary observation that U(O2 ) = U0 (O2 ), cf. [172, thm. 1.9, thm. 3.7] in case
of complex C *-algebras. It is possible to show it also for the “real” version (O2 )R
of O2 that each “orthogonal” operator in (O2 )R is inside the orthogonal operators
connected to 1 by a continuous path, but the proofs of [172, thms. 1.9, 3.7] have
to be modified slightly, cf. Section 1 of Appendix A. We say this here because it
implies that some parts of our results remain valid in the case of real C *-algebras.

Definition 4.3.3. Let h1 , h2 : D → E be C *-morphisms. Then h1 will be said

to n-dominate h2 if there exist a1 , . . . , an in E such that a∗1 a1 +. . .+a∗n an = 1 and
h2 (b) = a∗1 h1 (b)a1 + . . . + a∗n h1 (b)an for all b ∈ D . We shall say that h1 dominates
h2 if h1 1-dominates h2 , i.e., if h2 (b) = s∗ h1 (b)s, b ∈ D, for some isometry s in E.

Note that for unital D a non-unital h1 can dominate a unital h2 . But a unital
h1 can not dominate a non-unital h2 by (iii) of the following Lemma 4.3.4.
 3. ADDITION OF C *-MORPHISMS 529

Lemma 4.3.4. Suppose that h1 : D → E is a C*-morphism, that s, t ∈ E are

isometries and that g ∈ E is a contraction. Define completely positive contractions
h2 , h3 , T : D → E by h2 := s∗ h1 (·)s, h3 := t∗ h1 (·)t and T := g ∗ h1 (·)g.

(i) The c.p. map T : D → E is a C*-morphism if and only if gg ∗ ∈ h1 (D)0 ∩E

and (gg ∗ − gg ∗ gg ∗ )h1 (D) = {0}.
Moreover, T = h1 if and only if g ∈ h1 (D)0 ∩E and (1−g ∗ g)h1 (D) = {0}.
In particular, h2 is a C*-morphism if and only if ss∗ ∈ h1 (D)0 ∩ E. It
holds h2 = h1 if and only if s ∈ h1 (D)0 ∩ E.
(ii) If h2 is a C*-morphism, then h3 is unitarily equivalent to h2 , if and only
if, tt∗ is Murray–von Neumann equivalent to ss∗ in h1 (D)0 ∩ E.
Moreover, h2 = h3 if and only if ts∗ ∈ h1 (D)0 ∩ E.
(iii) h2 is unital if h1 is unital.
(iv) h1 dominates zero if and only if there is an isometry v ∈ E which range
vv ∗ orthogonal to the image of h1 , i.e., vv ∗ h1 (D) = 0.
(v) Suppose that there exists a unital C*-morphism from O2 into E that is
given by isometries s1 , s2 ∈ E as canonical generators, that h1 ⊕s1 ,s2 h1 =
r∗ h1 (·)r for an isometry r ∈ E, and that h2 := s∗ h1 (·)s and h3 := t∗ h1 (·)t.
Then h2 ⊕s1 ,s2 h3 = q ∗ h1 (·)q for the isometry q := r(s ⊕s1 ,s2 t).

Proof. (i): Let g ∈ E with kgk ≤ 1 and a∗ = a ∈ E (likewise a = h1 (d) for

some d∗ = d ∈ D).
Clearly, g ∗ agg ∗ bg = g ∗ abg and g ∗ bgg ∗ ag = g ∗ bag for all b ∈ E, if gg ∗ commutes
with a and if x = 0 for x := g ∗ (1 − gg ∗ )a. That means gg ∗ a = agg ∗ and 0 = x∗ x =
(gg ∗ − gg ∗ gg ∗ )a(gg ∗ a) = 0.
Conversely, if g ∗ a2 g = g ∗ agg ∗ ag, then (1 − gg ∗ )1/2 ag = 0. It implies gg ∗ a =
gg agg ∗ = agg ∗ , because a∗ = a and (1 − gg ∗ )agg ∗ = 0. It follows that gg ∗
∗

commutes with a and (gg ∗ − gg ∗ gg ∗ )a = 0.

Hence, gg ∗ ∈ h1 (D)0 ∩ E and (gg ∗ − gg ∗ gg ∗ )h1 (D) = {0}, if and only if, the
map T := g ∗ h1 (·)g from D into E satisfies T (dc) = T (d)T (c) for all d∗ = d, c ∈ D,
i.e., if and only if T is a C *-morphism.
If g ∗ ag = a and g ∗ a2 g = a2 , then 0 ≤ (ag − ga)∗ (ag − ga) = −a(1 − g ∗ g)a ≤ 0.
Thus, ga = ag and (1 − g ∗ g)a = 0 for all a∗ = a ∈ h0 (D) if h1 = g ∗ h1 (·)g =: T .
The converse is obvious.
If g is an isometry g = s then only the necessary and sufficient condition
ss∗ ∈ h0 (D)0 ∩ E remains, and the condition s ∈ h0 (D)0 ∩ E for the case h0 = h2 .
(ii): If u ∈ E is a unitary in E with u∗ h2 (·)u = h3 (·), then v = sut∗ is a partial
isometry in E with vv ∗ = ss∗ , v ∗ v = tt∗ and v ∗ h1 (·)v = tt∗ h1 (·)tt∗ = tt∗ h1 (·).
Thus h1 (b)v = ss∗ h1 (b)v = vv ∗ h1 (b)v = vtt∗ h1 (b) = vh1 (b) for every b ∈ D.
Conversely, if there is a partial isometry v ∈ h1 (D)0 ∩ E with vv ∗ = ss∗ and
v ∗ v = tt∗ , then u = s∗ vt is a unitary in E such that u∗ h2 (·)u = h3 .
(iii): s∗ h1 (1)s = s∗ s = 1.
530 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

(iv): Suppose that h1 dominates zero. Then there is an isometry v ∈ E

with v ∗ h1 (a)v = 0 for all a ∈ D. Thus vv ∗ h1 (a)(vv ∗ h1 (a))∗ = 0 for all a ∈ D.
Conversely, v ∗ h1 (a)v = v ∗ (vv ∗ h1 (a))v = 0 if v ∈ E is an isometry with range
orthogonal to h1 (D).
(v): by straight calculation. 

Proposition 4.3.5. Let D and E be C*-algebras and suppose that E contains

a copy of O2 = C ∗ (s1 , s2 ) unitally. Let hi : D → E, i = 1, 2 be C*-morphisms.

(i) If h1 dominates h2 and h2 (D)0 ∩ E contains a copy of O2 unitally, then

h1 ⊕ h2 and h1 are unitarily equivalent.
(ii) If h1 n-dominates h2 and h1 (D)0 ∩ E contains two isometries with orthog-
onal range, then h1 dominates h2 .
(iii) [h1 ] + [h1 ] = [h1 ] if and only if h1 (D)0 ∩ E contains a copy of O2 unitally.
(iv) h1 dominates h1 ⊕ h1 if and only if h1 (D)0 ∩ E contains a copy of O∞
unitally.
For every non-zero projection p in this copy of O∞ in h1 (D)0 ∩ E with
[p] = 0 in K0 (O∞ ) there exists an isometry s ∈ E with ss∗ = p. The c.p.
map h0 := s∗ h1 (·)s is a C*-morphism from D into E with [h0 ] + [h0 ] =
[h0 ], and h0 dominates h1 .
(v) If there is a contraction f ∈ E such that h1 ⊕ h1 = f ∗ h1 (·)f , then there
exists a C*-morphism k from D ⊗ O∞ into E such that h1 = k((·) ⊗ 1).

The copies of O2 considered in the assumptions and in Parts (i) and (iii), or
the copies of O∞ appearing in Parts (ii), (iv) and (v) can be different, are even
not necessarily homotopic. In the situation of Part (v) it can happen that there is
no copy of O∞ unitally contained in E itself that commutes with the elements of
h1 (D).

Proof. (i): Suppose h2 (·) = t∗ h1 (·)t, for t ∈ E with t∗ t = 1. Then tt∗

commutes with the images of h1 by Lemma 4.3.4(i). If r1 , r2 ∈ h2 (D)0 ∩ E are
isometrics with r1 r1∗ + r2 r2∗ = 1, then t1 := (1 − tt∗ ) + tr1 t∗ and t2 := tr2 are
isometries that satisfy t1 t∗1 + t2 t∗2 = 1. Thus, V := t1 r1∗ + t2 r2∗ is a unitary. The
unitary V realizes the unitary equivalence V (h1 ⊕r1 ,r2 h2 )V ∗ = h1 ⊕t1 ,t2 h2 . Then
th2 (·)t∗ = t(t∗ h1 (·)t)t∗ = tt∗ h1 (·) implies h1 ⊕t1 ,t2 h2 = h1 because

(1 − tt∗ )h1 (·) + tr1 h2 (·)r1∗ t∗ + tr2 h2 (·)r2∗ t∗ = (1 − tt∗ )h1 (·) + th2 (·)t∗ = h1 .

The unitary equivalence of h1 and h1 ⊕s1 ,s2 h2 with the before given isometries
s1 , s2 can be realized by u∗ h1 (·)u = h1 ⊕s1 ,s2 h2 where u := s1 t∗1 + s2 t∗2 . This
unitary u can be replaced here by any unitary v0 · u · (v1 ⊕s1 ,s2 v2 ) with unitaries
v0 , v1 ∈ h1 (D)0 ∩ E and v2 ∈ h2 (D)0 ∩ E.
(ii): By assumption, there are isometries t1 , t2 ∈ h1 (D)0 ∩E satisfying t∗1 t2 = 0.
The isometries sj := tj2 t1 , j = 1, . . . , n , are in h1 (D)0 ∩ E and satisfy s∗i sj = δij ,
Pn
1 ≤ i, j ≤ n. Now suppose h2 = j=1 a∗j h1 (·)aj , with aj ∈ E and
P ∗
aj aj = 1.
 3. ADDITION OF C *-MORPHISMS 531

P
Then, by taking b := j sj aj , we have
X X
h2 = a∗j h1 (·)aj = a∗i s∗i h1 (·)sk ak = b∗ h1 (·)b.
i,k

Moreover,
X X
b∗ b = a∗i s∗i sk ak = a∗j aj = 1.
i,k

(iii): By definition [h1 ] + [h1 ] = [h1 ] means that there is a unitary u ∈ E

such that u∗ h1 (·)u = s1 h1 (·)s∗1 + s2 h1 (·)s∗2 for isometries s1 , s2 ∈ E which are the
canonical generators of a copy of O2 . Then ti := usi , i = 1, 2, are generators of O2
in h1 (D)0 ∩ E, because, for a ∈ D and i ∈ {1, 2},

h1 (a)ti = t1 h1 (a)t∗1 ti + t2 h1 (a)t∗2 ti = ti h(a) .

Conversely, h1 ⊕t1 ,t2 h1 = h1 if t1 , t2 ∈ h1 (D)0 ∩ E are isometries and generators of

a copy of O2 . Now apply Proposition 4.3.2(i).
(iv): If h1 dominates h1 ⊕ h1 , then there is an isometry t ∈ E such that
t h1 (·)t = s1 h1 (·)s∗1 +s2 h1 (·)s∗2 . The elements ti := tsi , i = 1, 2, are isometries with
∗

orthogonal ranges such that h1 = t∗i h1 (·)ti for i = 1, 2. We get t1 , t2 ∈ h(D)0 ∩ E by

Lemma 4.3.4(i). Thus C ∗ (t1 , t2 ) is unitally contained in h1 (D)0 ∩ E and contains
the copy C ∗ (tn2 t1 ; n = 1, 2, . . .) of O∞ unitally.
Conversely, suppose that h1 (D)0 ∩ E contains a copy of O∞ unitally. Then
there are isometries t1 and t2 in h1 (D)0 ∩ E with orthogonal ranges. The element
t := t1 s∗1 + t2 s∗2 ∈ E is an isometry and satisfies t∗ h1 (·)t = s1 h1 (·)s∗1 + s2 h1 (·)s∗2 .
Suppose that p 6= 0 is a projection in a given copy of O∞ ⊆ h1 (D)0 ∩ E with
[p] = 0 in K0 (O∞ ) and 1E ∈ O∞ , e.g. p = 1 − (t2 t1 )(t2 t1 )∗ if O∞ ∼
= C ∗ (tn2 t1 ; n =
1, 2, . . .) ⊆ E. It implies that also [p] = 0 in K0 (E) and that p is properly infinite
in O∞ and E, because O∞ is simple and purely infinite by Corollary 2.2.7, (cf. also
[169, thm. 3.4]).
Thus, by Lemma 4.2.6(ii), p is Murray–von-Neumann equivalent to 1 in E, i.e.,
there exists an isometry s ∈ E with ss∗ = p. The map h0 := s∗ h1 (·)s is a C *-
morphism from D into E, because p = ss∗ commutes with h1 (D). The commutant
h0 (D)0 ∩ E of h0 (D) contains s∗ pO∞ ps unitally, and s∗ pO∞ ps is isomorphic to
pO∞ p. By [172], O∞ is a simple purely infinite C *-algebra. It follows for non-zero
projections p ∈ O∞ with [p] = 0 in K0 (O∞ ) that the hereditary C *-subalgebra
pO∞ p of O∞ (with unit p) contains a copy of O2 unitally. Thus [h0 ] + [h0 ] = [h0 ]
by Part (iii).
The C *-morphism h0 dominates h1 , because there exists an isometry t ∈ O∞ ⊆
h1 (D)0 ∩E with tt∗ ≤ p = ss∗ . Then ss∗ t = t, s∗ t is an isometry in E, h1 = t∗ h1 (·)t
and t∗ sh0 (·)s∗ t = t∗ ss∗ h1 (·)ss∗ t.
(v): Let {s1 , s2 } denote the canonical generators of the given copy of O2 in E.
We define C *-subalgebras C, N ⊆ E by

N := Ann(h1 (D)) := { e ∈ E ; h1 (D)e ∪ eh1 (D) = {0} }

532 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

and by C := h1 (D)0 ∩E. Then N is an ideal of C and C/N is a unital C *-algebra. It

is not difficult to see that there exists a unique C *-morphism λ : D ⊗max C/N → E
with the property

λ(d ⊗ (g + N )) = dg for all d ∈ D and all g ∈ C .

If f ∗ h1 (·)f = h1 ⊕s1 ,s2 h1 for a contraction f ∈ E, then the elements f s1 and

f s2 are in C and tj := f sj + N (j = 1, 2) are isometries in C/N with t∗1 t2 = 0,
i.e., h1 (a)f sk = f sk h1 (a), (1 − s∗k f ∗ f sk )h1 (a) = 0 for a ∈ D for k = 1, 2, and
s∗1 f ∗ f s1 + s∗2 f ∗ f s2 = f ∗ f ⊕ f ∗ f ≤ 1. The latter is obvious, and the first two
equations follow from s∗k f h1 (·)f sk = h1 by Lemma 4.3.4(i).
The isometries Tn := tn2 t1 for n ∈ N build a sequence of isometries in C/N with
Tn∗ Tm = 0 for m 6= n. Thus, there is a unital *-monomorphism γ : O∞ → C/N .
Then k := λ◦(idD ⊗γ) is a C *-morphism from D⊗O∞ into E with k(d⊗1) = h1 (d)
for d ∈ D. 

The Cuntz addition [h] + [k] of unitary equivalence classes of morphisms

h, k : D → E is far away from the sum of linear maps, e.g. [h] + [0] 6= [h] if h is
unital.
If we consider the zero-absorbing maps h⊕0 then the following proposition gives
additional later useful sufficient conditions on domination and unitary equivalence
by unitaries in U0 (E). Here U0 (E) denotes the connected component of 1E in the
group U(E) of unitaries in E with operator-norm topology.

Proposition 4.3.6. Suppose that E is unital and contains a copy of O2 uni-

tally with canonical generators s1 , s2 ∈ O2 ⊆ E. Let h1 : D → E be a C*-morphism
and h2 : D → E a contractive linear map.

(i) h1 dominates zero, if and only if, [h1 ] + [0] = [h1 ].

This is the case, if and only if, there exists u ∈ U(E) such that t1 :=
us1 , t2 := us2 and u satisfy t1 h1 (a) = h1 (a) = h1 (a)t1 , h1 (a)t2 = 0 for
a ∈ D.
The unitary u can be taken such that [u] = 0 ∈ K1 (E) in addition.
(ii) If h1 dominates zero and x ∈ E is a contraction, then the element s :=
t1 x + t2 (1 − x∗ x)1/2 – with t1 , t2 as in Part(i) – is an isometry in E that
satisfies h1 (·)x = h1 (·)s and x∗ h1 (·) = s∗ h1 (·).
In particular, s∗ h1 (·)s = h2 for an isometry s ∈ E if h1 dominates
zero and x∗ h1 (·)x = h2 for some contraction x ∈ E.
(iii) Suppose that
(α) h1 dominates zero,
(β) the inclusion map C ∗ (h2 (D)) ,→ E dominates zero, and
(γ) there are contractions x, y ∈ E such that h2 (·) = x∗ h1 (·)y and
xx∗ h1 (·)yy ∗ = h1 , i.e., also xh2 (·)y ∗ = h1 .
Then there exist U, V ∈ U(E) such that h2 (·) = U ∗ h1 (·)V .
If, moreover, x = 1, then one can find U and V such that U = 1.
 3. ADDITION OF C *-MORPHISMS 533

In particular, h1 is unitarily equivalent to h2 , if h1 and h2 dominate

both zero and there exists x ∈ E with kxk ≤ 1 , x∗ h1 (·)x = h2 and
xh2 (·)x∗ = h1 .
(iv) Suppose that h1 dominates zero and that h2 is unitarily equivalent to h1
in E.
(a) There exists v ∈ U(E) with h2 = v ∗ h1 (·)v and [v] = 0 in K1 (E).
(b) If E is K1 -injective, then there is a unitary v ∈ U0 (E) with h2 =
v ∗ h1 (·)v.
(c) If there exists an isometry r ∈ E with r∗ (h1 (a) + h2 (a))r = 0 for all
a ∈ D, then there exists v ∈ U0 (E) with h2 = v ∗ h1 (·)v.
(d) If E has moreover the property that for every unitary u ∈ E and
every isometry t ∈ E there are isometries r, s ∈ E and w ∈ U0 (E)
with utr = wts then there is v ∈ U0 (E) with h2 = v ∗ h1 (·)v .
(v) If R ∈ E is an isometry with R∗ h1 (·)R = 0 and p ∈ E is a projection
with pR = 0, then the u ∈ U(E) in Part(i) can be chosen such that
0 = [u] ∈ K1 (E) and t1 = us1 satisfies t1 h1 (a) = h1 (a) = h1 (a)t1 for
a ∈ D and p t1 = p = t1 p.

Proof. (i): If h1 dominates zero, then [h1 ]+[0] = [h1 ] by Proposition 4.3.5(i),
because the copy of O2 in E commutes with the zero morphism 0(a) := 0.
Conversely, if [h1 ] + [0] = [h1 ], then, by definition of Cuntz Addition, there is a
unitary v ∈ E with vs1 h1 (·)s∗1 v ∗ = h1 . Then rj := vsj (j = 1, 2) are isometries with
r1 h1 (·)r1∗ = h1 and r1 r1∗ + r2 r2∗ = 1. In particular, h1 (·)r2 = 0, i.e., h1 dominates
zero. It follows h1 (·)r1 r1∗ = h1 (·) and h1 (D)r2 Er2∗ = {0}.
Let q2 := r2 r1 and q1 := r1 r1∗ + r2 r1 r2∗ , then q1 and q2 are isometries in E
with q1 q1∗ + q2 q2∗ = 1, h1 (·)q2 = 0 and q1 h1 (·) = h1 (·) = h1 (·)q1 . The unitary
v := q1 s∗1 + q2 s∗2 ∈ U(E) satisfies vsj = qj . Let w := q1 q1∗ + q2 v ∗ q2∗ , u := wv,
t1 := us1 = q1 and t2 := us2 = q2 v ∗ . Then u, t1 , t2 have the desired properties,
because h1 (·)q2 = 0 and [u] = [w] + [v] = [1] + [v ∗ ] + [v] = [1] = 0 in K1 (E).
(ii): Obviously s := t1 x + t2 (1 − x∗ x)1/2 is an isometry and h1 (·)s = h1 (·)t1 x =
h1 (·)x by Part (i).
If h2 = x∗ h1 (·)x then h2 = s∗ h1 (·)s, because x∗ h1 (a) = (h1 (a∗ )x)∗ and D2 =
D.
(iii): Let a ∈ D+ , d1 := xx∗ , d2 := yy ∗ and b := h1 (a).
Then b ≥ 0, 0 ≤ di ≤ 1 (i = 1, 2) and d1 b1/n d2 = b1/n (n = 1, 2, . . .). This
implies d1 pd2 = p for the support p of b in E ∗∗ , because p is the weak limit of b1/n .
Thus pd1 pd2 p = p. Since 0 ≤ di ≤ 1, we get pdi p = p, p(1 − di )p = 0,
(1 − di )1/2 p = 0 = p(1 − di )1/2 , and (1 − d1 )1/2 b = 0 = b(1 − d2 )1/2 , because
pb = b = bp. for i = 1, 2. That means

(1 − d1 )1/2 h1 (·) = 0 = h1 (·)(1 − d2 )1/2 .

534 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Let k : C ∗ (h2 (D)) → E the inclusion map. By Part (i) there exist isometries
τ1 , τ2 ∈ E such that 1 = τ1 τ1∗ + τ2 τ2∗ , k(·)τ1 = τ1 k(·) = k(·) and k(·)τ2 = 0. Let
t1 , t2 ∈ E for h1 as in (i). It follows from Part (i) and from the above equations for
d1 , d2 and h1 that

U := t1 xτ1∗ − t2 x∗ τ2∗ + t1 (1 − d1 )1/2 τ2∗ + t2 (1 − x∗ x)1/2 τ1∗

and
V := t1 yτ1∗ − t2 y ∗ τ2∗ + t1 (1 − d2 )1/2 τ2∗ + t2 (1 − y ∗ y)1/2 τ1∗
are unitaries in E with

U ∗ h1 (·)V = τ1 (x∗ h1 (·)y)τ1∗ = τ1 h2 (·)τ1∗ = h2 .

If x = 1, then, with a := c1/2 for c ∈ D+ ,

U ∗ h1 (c)U = h2 (a)(h2 (a)∗ ) = h1 (a)d2 h1 (a) = h1 (a2 ) = h1 (c).

Thus, U commutes with the elements of h1 (D), and h2 (·) = h1 (·)U ∗ V .

Suppose that h1 and h2 both dominate zero and that there is a contraction
x ∈ E with x∗ h1 (·)x = h2 and xh2 (·)x∗ = h1 . Then there is an isometry t ∈ E
with t∗ h2 (·)t = 0, i.e., h2 (D) = span(h2 (D+ )) is contained (1 − tt∗ )E(1 − tt∗ ).
Thus, t∗ bt = 0 for all b ∈ C ∗ (h2 (D)), and the above construction of U and V works
for x and y := x. It gives d2 = d1 and V = U . Hence, h2 = U ∗ h1 (·)U .

(iv): Let U ∈ U(E) with U ∗ h1 (·)U = h2 . In particular, [h2 ] = [h1 ] = [h1 ] +

[0] = [h2 ] + [0], i.e., h2 dominates zero.
(iv,a): Let t1 , t2 ∈ E the isometries with the properties given in (i), then
V := (1 ⊕t1 ,t2 U ∗ )U is unitary and satisfies V ∗ h1 (·)V = U ∗ h1 (·)U = h2 . [V ] =
[1] + [U ∗ ] + [U ] = 0 in K1 (E), cf. Lemma 4.2.6(v,1).
(iv,b): If U0 (E) is the kernel of U(E) → K1 (E) (i.e., if E is K1 -injective) then
V ∈ U0 (E) for the V as in Part (iv, a).
(iv,c): Suppose that there is an isometry r ∈ E with r∗ (h1 (a) + h2 (a))r = 0
for all a ∈ D. Then hk (D) ⊆ (1 − rr∗ )E(1 − rr∗ ) for k = 1, 2. We consider the
C *-subalgebra F := C ∗ (h1 (D) + h2 (D)) ⊆ E generated by

h1 (D) + h2 (D) ⊆ (1 − rr∗ )E(1 − rr∗ ) .

The inclusion morphism f ∈ F 7→ f ∈ E dominates zero, because r∗ F r = 0.

By Part (i) there are canonical generators t1 , t2 of a copy of O2 in E such that
t1 f = f = f t1 and f t2 = 0 = t∗2 f for f ∈ F . Let V = t1 U t∗1 + t2 U ∗ t∗2 . Then V is
in U0 (E) by Lemma 4.2.6(v,3), and V ∗ h1 (·)V = t1 U ∗ h1 (·)U t∗1 = t1 h2 (·)t∗1 = h2 .
(iv,d): The property in Part (iv,d) implies that U0 (E) is the kernel of U(E) 3
u 7→ [u] ∈ K1 (E), cf. Lemma 4.2.10(iii). Thus Part (iv, b) applies.
(v): Since pR = 0 and R∗ h1 (D)R = {0}, the C *-subalgebra F :=
C ∗ (h1 (D), p) ⊆ E is contained in (1 − RR∗ )E(1 − RR∗ ). The identity map
f ∈ F 7→ f ∈ E dominates zero because R∗ f R = 0 for all f ∈ (1−RR∗ )E(1−RR∗ ).
 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 535

Application of Part (i) to (F, idF ) – in place of (D, h1 ) – gives that there is a
unitary u ∈ U(E) with 0 = [u] ∈ K0 (E) and t1 f = f = f t1 for t1 := us1 and f ∈ F .
In particular t1 p = p = pt1 and t1 h1 (a) = h1 (a) = h1 (a)t1 for a ∈ D. 

Corollary 4.3.7. Suppose that A and E are unital C*-algebras, h1 , h2 : A →

E, unital C*-morphisms, and that unital C*-morphisms ψ1 , ψ2 : O2 → E exist with
ψk (O2 ) ∈ hk (A)0 ∩ E for k = 1, 2.
If h1 and h2 one-step dominate each other, i.e., if there exist contractions
S, T ∈ E with h2 := S ∗ h1 (·)S and h1 := T ∗ h2 (·)T , then h1 and h2 are unitary
equivalent in E.

Proof. By Proposition 4.3.5(i), h1 is unitarily equivalent to h1 ⊕ h2 and h2 is

unitarily equivalent to h2 ⊕ h1 .
The symmetry T := s2 s∗1 + s1 s∗2 satisfies T ∗ (h1 ⊕s1 ,s2 h2 )T = h2 ⊕s1 ,s2 h1 . 

Corollary 4.3.8. Two unital C*-morphisms h0 , h1 : O2 → E are approxi-

mately unitary equivalent.
They are homotopic if and only if the unitary u := h1 (s1 )h0 (s∗1 ) + h1 (s2 )h0 (s∗2 )
is in U0 (E).

Proof. In fact h1 and h2 are unitary equivalent in E∞ := `∞ (E)/c0 (E) if we

use here the later in Chapter 5 proven result that O2 ⊗ O2 is unitally contained in
O2 , (or could use here that O2 ∼
= O2 ⊗ O2 by the later proven Corollary F(iii)), and
that this unital imbedding ψ : O2 ⊗ O2 → O2 has the property that a 7→ ψ(a ⊗ 1) is
approximately unitarily equivalent to the identity map of O2 by the approximate
innerness of δ2 : O2 → O2 . gives the ψ (In fact, the morphisms δ2 : O2 → O2 and
a 7→ ψ(a⊗1) are moreover unitarily homotopic to the identity map of O2 .) This and
the nuclearity of O2 ensures that that any two unital C *-morphisms from O2 into
E∞ 1-step dominate each other. Hence Corollary 4.3.7 applies to h1 , h2 : O2 → E∞ .
The homotopy of h1 and h2 with h1 (sk ) = u(1)h0 (sk ) for a continuous map
[0, 1] 3 t 7→ u(t) ∈ U(A) with u(0) = 1 is given by ht : O2 → U(A) with ht (sk ) :=
u(t)sk . 

Remark 4.3.9. If h0 , h1 : O2 → E are homotopic, then one can show that they
are moreover “unitarily homotopic”, cf. Corollary 4.6.4, i.e., are unitary equivalent
in Cb ([0, ∞), E)/ C0 ([0, ∞), E). (It does not imply that, conversely, h0 and h1 itself
are unitarily equivalent or are homotopic in E.)

4. Groups defined by absorbing C *-morphisms

Suppose that h0 : D → E is a C *-morphism such that h0 (D)0 ∩ E contains

O2 unitally. Proposition 4.3.5(iii) says that this is equivalent to the existence of a
unital C *-morphism from O2 into E and the equation [h0 ] + [h0 ] = [h0 ].

Definition 4.4.1. We define then the semigroup

S(h0 ; D, E) := { [h] ; h ∈ Hom(D, E) , h is dominated by h0 }
536 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

and
G(h0 ; D, E) := { [h] + [h0 ] ; [h] ∈ S(h0 ; D, E) } .
Notice that G(h0 ; D, E) = [h0 ] + S(h0 ; D, E) ⊆ S(h0 ; D, E) .
Due to the terminology used in the abstract Lemma 4.2.3, we call sometimes
the morphisms h ∈ Hom(D, E) with the property [h] ∈ G(h0 ; D, E) the absorbing
morphisms in the class of those morphisms k ∈ Hom(D, E), that are dominated by
h0 .

G. Kasparov has introduced a similar notion of absorption in his early paper

[405] on Ext-groups. By Proposition 4.3.5(i), a unitary equivalence class [h] ∈
S(h0 ; D, E) is in G(h0 ; D, E) if and only if h dominates h0 .
The definitions, the above listed properties and Remark 3.1.2(iii) together im-
mediately imply the following consequences of “[k] ∈ S(h0 ; D, E)”:

(i) k(D) ⊆ J if [k] ∈ S(h0 ; D, E), h0 (D) ⊆ J and J is an ideal of E.

(ii) k is nuclear if [k] ∈ S(h0 ; D, E) and h0 is nuclear.

More general, it is obvious that k ∈ C if h0 ∈ C for some matrix operator-convex

cone C ⊆ CP(D, E). This applies in particular to the m.o.c. cone C(h0 ) that is
generated by h0 .

Proposition 4.4.2. Suppose that E contains a copy of O2 unitally and

h0 : D → E is a C*-morphism with [h0 ] + [h0 ] = [h0 ].

(i) S(h0 ; D, E) is a commutative semigroup under Cuntz addition of unitary

equivalence classes, and
(ii) G(h0 ; D, E) is a subgroup of S(h0 ; D, E), and [h] → [h] + [h0 ] is a
semigroup epimorphism from S(h0 ; D, E) onto G(h0 ; D, E). It identi-
fies G(h0 ; D, E) naturally with the Grothendieck group of S(h0 ; D, E),
in the sense that every semi-group morphism ϕ : S(h0 ; D, E) → Γ into a
group Γ satisfies ϕ([h]) = ϕ([h] + [h0 ]) for all [h] ∈ S(h0 ; D, E).

Proof. Let s1 , s2 ∈ h0 (D)0 ∩ E isometries with s1 s∗1 + s2 s∗2 = 1, as given by

Proposition 4.3.5(iii).
(i): If h1 is dominated by h0 then the same property holds for every map in
the unitary equivalence class [h1 ]. Suppose that [h1 ] and [h2 ] are in S(h0 ; D, E).
Then [h1 ] + [h2 ] ∈ S(h0 ; D, E), because h1 ⊕s1 ,s2 h2 is dominated by h0 , cf. Lemma
4.3.4(v). The semigroup structure follows now from Proposition 4.3.2(i).
(ii): Since [h0 ] + [h0 ] = [h0 ], the subset G(h0 ; D, E) is a subgroup of the semi-
group S(h0 ; D, E), and [h0 ] is the neutral element (zero element) of G(h0 ; D, E).
The inverse element of [h] in G(h0 ; D, E) can be detected as follows: Since h ∈
S(h0 ; D, E), there is an isometry t ∈ E with h = t∗ h0 t. Then p = tt∗ is a projection
in h0 (D)0 ∩ E by Lemma 4.3.4(i), and (1 − p)t = 0 = t∗ (1 − p). Hence,

k := (1 − p)h0 (·) + th0 (·)t∗

 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 537

is a C *-morphism. The morphism h0 dominates k by Proposition 4.3.5(ii) because

h0 2-dominates k.
Now it is easy to verify that k ⊕s1 ,s2 h = u∗ h0 (·)u, where {s1 , s2 } are generators
of O2 in h0 (D)0 ∩ E, and u is the unitary operator s1 ts∗2 + s1 (1 − p)s∗1 + s2 t∗ s∗1 .
Then [(h ⊕ h0 ) ⊕ (k ⊕ h0 )] = [h0 ⊕ h0 ⊕ h0 ] = [h0 ]. Hence [k ⊕ h0 ] is the inverse
of [h ⊕ h0 ] in G(h0 ; D, E). This shows that Lemma 4.2.3 applies to the semigroup
S(h0 ; D, E) and G(h0 ; D, E) = [h0 ] + S(h0 ; D, E). 

The following Proposition 4.4.3 identifies G(h0 ; D, E) with a subgroup of the

K0 group K0 (h0 (D)0 ∩ E) and gives the later used characterizations of the elements
of this subgroup, and it describes the K∗ -theoretic picture of our Ext(C; ·, ·)- and
R(C; ·, ·)-groups studied in Chapters 5 and 7. Here the Cuntz addition ⊕ in Parts
(ii) and (iii) has to be defined with help of the canonical generators of a fixed copy
of O2 that is unitally contained in h0 (D)0 ∩ E.
Proposition 4.4.3. Let h0 : D → E a C*-morphism, such that h0 (D)0 ∩ E
contains a copy of O2 unitally, and let p and p0 projections in h0 (D)0 ∩ E.

(i) The group G(h0 ; D, E) is naturally isomorphic to the kernel of the natural
K∗ -theory map
i0 : K0 (h0 (D)0 ∩ E) → K0 (E),
where i0 := K0 (i) for the inclusion map i : h0 (D)0 ∩ E ,→ E .
(ii) The element [p] ∈ K0 (h0 (D)0 ∩ E) is in the kernel of i0 , if and only if,
there is a unitary v in the connected component U0 (E) of 1E in U(E) such
that v ∗ (p ⊕ 1 ⊕ 0)v = 1 ⊕ 0.
The equation [p] = [p0 ] holds in K0 (h0 (D)0 ∩ E), if and only if, there
is a unitary u ∈ U0 (h0 (D)0 ∩ E) such that u∗ (p ⊕ 1 ⊕ 0)u = p0 ⊕ 1 ⊕ 0 .
(iii) If h0 dominates zero, then projections p and p0 in h0 (D)0 ∩ E define the
same element of K0 (h0 (D)0 ∩ E), if and only if, [p] = [p0 ] in K0 (E) and
there is a unitary u ∈ h0 (D)0 ∩E such that ∆ := (p0 ⊕1⊕0)−u∗ (p⊕1⊕0)u
is in the (two-sided) annihilator of h0 (D), i.e., h0 (a)∆ = 0 for all a ∈ D.

Proof. We let C := h0 (D)0 ∩E and define the Cuntz addition ⊕ by isometries

s1 , s2 ∈ C with s1 s∗1 + s2 s∗2 = 1E , i.e., ⊕ = ⊕s1 ,s2 .
(i): A C *-morphism h : D → E has unitary equivalence class [h] in S(h0 ; D, E)
if and only if there is an isometry t ∈ E with h = t∗ h0 (·)t (cf. Definition 4.3.3).
Then tt∗ ∈ C by Lemma 4.3.4(i). If s ∈ E is an other isometry with s∗ h0 (·)s = h,
then z = ts∗ is a partial isometry in C by Lemma 4.3.4(ii). We have tt∗ = zz ∗
and z ∗ z = ss∗ , i.e., tt∗ and ss∗ are Murray–von-Neumann equivalent in C. Let
k : D → E a C *-morphism in the unitary equivalence class [h] of h. Then there is
a unitary u ∈ E with k = u∗ h(·)u = u∗ t∗ h0 (·)tu . Thus k is a C *-morphism that is
dominated by h0 , and [k] = [h]. Note that (tu)(tu)∗ = tt∗ .
It follows that the map θ from S(h0 ; D, E) into K0 (C), defined by
θ([h]) := [tt∗ ] ∈ K0 (C)
538 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

for h = t∗ h0 (·)t, with t∗ t = 1 is well-defined, because [s∗ h0 (·)s] = [t∗ h0 (·)t] = [h]
implies that the projections tt∗ and ss∗ are Murray–von Neumann equivalent in C.
By Lemma 4.3.4(v), h1 ⊕ h2 = r∗ h0 (·)r with r = t1 ⊕ t2 where hi (·) = t∗i h0 (·)ti
for i = 1, 2. Thus, θ([h1 ] + [h2 ]) = [t1 t∗1 ⊕ t2 t∗2 ] ∈ K0 (C) . This is the same as
[t1 t∗1 ] + [t2 t∗2 ] = θ([h1 ]) + θ([h2 ]) by Lemma 4.2.6(i). In particular θ([h0 ]) = 0,
because 2θ([h0 ]) = θ([h0 ]). Since K0 (C) is a group, [h] + [h0 ] 7→ θ([h]) is a well-
defined group homomorphism from G(h0 ; D, E) into K0 (C).
As 1E ∈ O2 ⊆ E, we get for h = t∗ h0 (·)t with t∗ t = 1,
[0]E = [1]E = [t∗ t]E = [tt∗ ]E ∈ K0 (E),
so that θ maps S(h0 ; D, E) into the kernel of i0 .
Let h1 = s∗ h0 (·)s, h2 = t∗ h0 (·)t and θ([h1 ]) = θ([h2 ]), i.e., [ss∗ ] = [tt∗ ] in
K0 (C). The projections 1 ⊕ ss∗ (= (1 ⊕ s)(1 ⊕ s)∗ ) and 1 ⊕ tt∗ are full and properly
infinite in C, thus are MvN-equivalent in C by Lemma 4.2.6(ii). But this means that
h0 ⊕ h1 and h0 ⊕ h2 are unitarily equivalent, by Lemma 4.3.4(ii). Thus [h1 ] + [h0 ] =
[h2 ] + [h0 ] if θ([h1 ]) = θ([h2 ]), and θ|G(h0 ; D, E) is a monomorphism.
Now let x ∈ K0 (C) with i0 (x) = 0. Since C is properly infinite, it follows from
Lemma 4.2.6(iii), that there exists a projection p ∈ C, and isometries s, t ∈ C ⊆ E,
such that x = [p], ss∗ ≤ p and tt∗ ≤ 1−p. Then i0 (x) = [p] = 0 = [1] in K0 (E), and,
by Lemma 4.2.6(ii), there is r ∈ E with rr∗ = p ∈ C and r∗ r = 1. The completely
positive map h := r∗ h0 (·)r is a C *-morphism from D into E by Lemma 4.3.4(i) and
h0 dominates h, i.e., [h] ∈ S(h0 ; D, E). Moreover, θ([h] + [h0 ]) = θ([h]) = [rr∗ ] =
[p] = x. Hence, [h] + [h0 ] 7→ θ([h]) is a group isomorphism from G(h0 ; D, E) onto
ker(i0 ) ⊆ K0 (C).
(ii): Recall that 1 ⊕ 0 = s1 s∗1 is unitarily equivalent to 1 ⊕ 1 ⊕ 0 by a unitary
in U(O2 ) = U0 (O2 ), cf. Proposition 4.3.2. Thus, (ii) is a special case of Lemma
4.2.6(iv,b) (with C in place of E for the equation [p] = [p0 ] ∈ K0 (C)).
(iii): Let p, p0 are projections in C. If [p] = [p0 ] in K0 (C), then [p] = [p0 ] in
K0 (E) and, by (ii), there is a unitary u ∈ C with u∗ (p ⊕ 1 ⊕ 0)u = p0 ⊕ 1 ⊕ 0.
Conversely, suppose that [p] = [p0 ] in K0 (E) and that there is a unitary u ∈ C
such that ∆ := (p0 ⊕ 1 ⊕ 0) − u∗ (p ⊕ 1 ⊕ 0)u satisfies ∆ · h0 (D) = {0}.
Let p1 := u∗ (p ⊕ 1 ⊕ 0)u and p2 := p0 ⊕ 1 ⊕ 0. Then [p1 ] = [p] + [1] + [0] = [p]
and [p2 ] = [p0 ] as well in K0 (C) as in K0 (E). Therefore, p1 and p2 are projections in
C with [p1 ] = [p2 ] in K0 (E) and p2 − p1 ∈ N , where N ⊆ E denotes the (two-sided)
annihilator of h0 (D) in E.
The two-sided annihilator N of h0 (D) is a hereditary C *-subalgebra of E and
is an ideal of C = h0 (D)0 ∩ E. The hereditary C *-subalgebra N of E is full in E,
because h0 dominates zero by assumptions of Part (iii), i.e., there is an isometry
t ∈ E with tt∗ ∈ N . By Lemma 4.2.20(i), we get [p1 ] = [p2 ] ∈ K0 (C) from
p2 − p1 ∈ N and [p1 ] = [p2 ] ∈ K0 (E). 

REVISE HISTORIC REMARKS:

 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 539

The assumptions in the following Theorem 4.4.6 are the same as of Lemma [442,
lem. 3.8], but with assumption [442, lem. 3.8(vi)] removed, because it can be derived
from the other assumptions, cf. Lemma 4.4.7(vii). An old outline of the proof of
Lemma [442, lem. 3.8] – displayed in beamer presentations – incorrectly claimed
that the natural surjective group homomorphism ϕ : G(h0 ; D, E) → G(H0 ; D, E)
is always injective. This is still unknown, and Professor Claire Anantharaman-
Delaroche mentioned her concerns about the injectivity claim for the natural group
homomorphism ϕ to the author (likely around 1996). And she suggested that the
homotopy invariance of the “unsuspended” but stable E-theory (respectively the
slightly different O2 -unital E-theory in the approach of N. Ch. Phillips) should
show the equivalence to KK-theory.
It turns out that the Grothendieck groups of some kinds of unsuspended but
stable E-theories are homotopy invariant, but that gives not a proof for the co-
incidence with the related sort of KK-theories if equipped with certain additional
“equivariant” behavior.
At the actual state we have two different proofs of the homotopy invariance of
the continuous Rørdam groups R(C ; A, B) that are defined in Chapter 7 for stable
separable A and B and non-degenerate m.o.c. cone C := C(h0 ), – which is in this
special case isomorphic to the below considered group G(h0 ; D, E):
One way is to prove first independently that the groups R(C ; A, B) are
homotopy invariant itself by a generalization and modification of the ideas of
N.Ch. Phillips in his proof of the homotopy invariance of the generalized unsus-
pended E-theory for pi-sun algebras in [627].
See Section ?? concerning a similar but more general sort of stable, but
unsuspended, E-semigroups and an outline of the homotopy invariance of its
Grothendieck group. It is not clear if one can use a “controlled” homotopy
invariance of R(C ; A, B) to establish the injectivity of the natural epimor-
phism ϕ : G(h0 ; D, E) → G(H0 ; D, E), in the special case where naturally
R(C ; A, B) = G(h0 ; D, E) with D := A ⊗ K, E := M(C0 (R, B))/ C0 (R, B) and

h0 : D → B ⊂ J := Cb (R+ , B)/ C0 (R+ , B) ⊂ E ,

a generating homomorphism for the given m.o.c. cone C ⊂ CP(A, B), cf. Chp. 7.
Notice that G(h0 ; D, E) is in this particular case the same as Ext(C ; A, SB) ∼
=
KK(C ; A, B) .
Until now:
We have only that (DC) is valid for our special applications,
that (DC) implies functorial equivalence of R(C; ., .) with KK(C; ., .).
The injectivity criteria of the following Theorem 4.4.6
and arguments in Chapters 7,
8 and 9 show that the injectivity of the
group epimorphism ϕ implies also homotopy invariance of R(C ; A, B).
540 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

An ‘‘independent proof’’ from the natural isomorphism

between R(C ; A, B[0, 1]) and R(C ; A, B)
to the injectivity of ϕ is not found or written up!!!
So far, until now.
An isomorphism R(CPnuc ; A, B[0, 1]) ∼
= R(CPnuc ; A, B))
was shown first by N.Ch. Phillips in [627] (for A and B in Cuntz
standard form?).
Then he used the characterization of N. Higson [364].
We do not know if we can also go this way
but the general problem is that it is not known if U (0) = 1 and

U ∗ (k ⊕ H0 )(a)U − H0 (a) ∈ C0 ([0, ∞), B) for all a ∈ A

always implies that h0 ⊗ k and h0
are asymptotical homotopic,
i.e., if there exists gs,t ∈ S(h0 ; A, B) with
s, t ∈ [0, 1] × [0, ∞) point-norm continuous
and g0,t = h0 ⊕ k and g1,t = h0 .
We generalize in Section ?? an idea of N.Ch. Phillips that shows that the
Grothendieck groups of certain semigroups of asymptotic morphisms are homotopy
invariant, in a way that covers also the case of our (!) more general types of Rørdam
groups.
But we underline that in our special case the injectivity criteria (DC) of The-
orem 4.4.6 follows from the homotopy invariance of KK(C ; A, B), where it is very
important that this holds for arbitrary m.o.c. cones. It allows a precise explanation
for the sketched and a bit imprecise arguments outlined at the end of the Preprint
[434], cf. also [442].
We describe in Theorem 4.4.6 the hypothetical elements in the kernel of the
homomorphism ϕ, and give an applicable sufficient criteria (DC) for the injectivity
of ϕ.
We show in Section 3 of Chapter 9 that in the later used special case of the
canonical surjection R(C ; A, B) → Ext(C ; A, SB) ∼
= KK(C ; A, B) the decomposi-
tion condition (DC) of Theorem 4.4.6 is a consequence of
KK(C(0, 1] ; A; C0 ((0, 1], B)) = 0 ,
i.e., the homotopy invariance of KK(C , A, B) allows to proof for our applications
the property (DC) that implies finally the needed bijectivity.
( 7 ).
It shows that the homotopy invariance of the Kasparov groups alone imply
that of C-related Rørdam groups and KK-groups are isomorphic. As claimed in the
appendix of the Preprint [434], there not really proved.
7 The author regrets for his remarks “Es ist ein Gradeaus-Beweis.” below [442, lem. 3.8],

and ”... should be able to work out the details” in the appendix of [434].
 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 541

We need for the description of the kernel of the group homomorphism

ϕ : G(h0 ; D, E) → G(H0 ; D, E) in the following Theorem 4.4.6 and in Lemma
4.4.15, some basic notations and definitions of algebraic nature:

Definition 4.4.4. Let A ⊆ E a C *-subalgebra, I / E an closed ideal of E,

C := A0 ∩ E the commutant (or “centralizer”) of A in E, and define by
Ann(A) := Ann(A, E) := { e ∈ E ; eA ∪ Ae = {0} } . (4.1)
the annihilator of A in E. We consider following C *-subalgebras of E:
Der(A, I) := { e ∈ E ; ea − ae ∈ I, ∀ a ∈ A } = πI−1 πI (A)0 ∩ (E/I) ,


(4.2)
N (A, I) := { e ∈ E ; ea, ae ∈ I, ∀ a ∈ A } = πI−1 Ann(πI (A), E/I) .


(4.3)

Compare Remarks 6.1.2 for further details on Ann( · ) and N ( · , · ). Notice that
C = Der(A, {0}) ⊆ Der(A, I) and that N (A, I) is a closed ideal of Der(A, I).


Definition 4.4.5. We define a discrete abelian group Γ A, I, E by




Γ A, I, E := K1 Der(A, I) / µ K1 (C) (4.4)
where the group morphism µ := [η]1 : K1 (C) → K1 (Der(A, I)) comes from the
inclusion map η : C ,→ Der(A, I) .

The formulation “later specified” used in Parts (iii) and (iv) of the following
Theorem means that different possible choices of the (non-unital) C *-subalgebra
F ⊆ C with the properties in Part (iii) and of isometries s0 , t0 ∈ E with the
properties in Part (iv) are not uniquely determined, but have to exist and can
be be chosen/adjusted such that they interplay in the manner that is described
in Parts (iii,iv). Their existence with the required interplay is a property of the
system (J ⊆ E, β, h0 , H0 ) .

Theorem 4.4.6. Suppose that h0 : D → E and H0 : D → E are *-monomor-

phisms, where D is a separable C*-algebra and E is a unital C*-algebra. Further
let J be a closed ideal of E and β a *-automorphism of E, such that h0 (D) ⊆ J,
β 2 = id, βH0 = H0 and β(J) ∩ J = 0.
We require that E, J, D, H0 , h0 , and β satisfy the following assumptions
(i)-(vi), with C := H0 (D)0 ∩ E.

(i) The algebra h0 (D)0 ∩ H0 (D)0 ∩ E ⊆ C contains β-invariant isometries s, t

with ss∗ + tt∗ = 1.
(ii) If H0 dominates k : D → E, then there exists a unitary u ∈ E such that
u∗ (k ⊕ H0 )(a)u − H0 (a) ∈ J for all a ∈ D.
(iii) There exists a (later specified) C*-subalgebra F ⊆ C with following prop-
erties:
(a) F · H0 (D) ⊆ J + β(J),
(b) β(F ) = F , and
(c) for every f ∈ J + β(J) and every projection q ∈ F , there is a projec-
tion p ∈ F with pf = f p = f , q ≤ p and q 6= p.
542 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

(iv) There are (later specified) β-invariant isometries s0 , t0 ∈ E that interplay

with F , C, H0 and h0 in the following manner
p0 := s0 s∗0 ∈ F , t0 ∈ C , s0 s∗0 + t0 t∗0 = 1
and
s∗0 H0 (a)s0 = h0 (a) + β(h0 (a)) for all a ∈ D .
(v) Every non-zero projection in F is unitarily equivalent to p0 by a unitary
in 1 + F .
(vi) There exists a unitary u1 in J 0 ∩ E such that u1 βh0 (·)ss∗ = tt∗ βh0 (·)u1 .

If assumptions (i)-(vi) are satisfied, then h0 ⊕ H0 is unitarily equivalent to H0

and the mapping ϕ : [k] + [h0 ] 7→ [k] + [H0 ] is a (well-defined) group epimorphism
from G(h0 ; D, E) onto G(H0 ; D, E).
The kernel of this epimorphism is naturally isomorphic to Γ(H0 (D), J, E).

The natural group epimorphism ϕ : [k] + [h0 ] 7→ [k] + [H0 ] from G(h0 ; D, E)
onto G(H0 ; D, E) is injective, if and only if, E, J, β, D, and H0 satisfy the
following decomposition condition (DC):

(DC) For each u ∈ U0 (E) ∩ Der(H0 (D), J) there exists v ∈ U(C) such that
v(u ⊕s,t 1) ∈ N (H0 (D), J + β(J)) + C · 1 .

The condition (DC) is equivalent to the following (formally weaker) property

(wDC):

(wDC) If u ∈ U0 (E) ∩ Der(H0 (D), J) and k ∈ S(h0 ; D, E) satisfy

(k + βh0 ) ⊕s0 ,t0 H0 = u∗ H0 (·)u ,
then
[k] + [h0 ] = [h0 ] .

Compare below with above concerning R(...) to KK(...).

The group homomorphism G(h0 ; D, E) → G(H0 ; D, E) is a “minimal”
abstract version of the natural group epimorphism from the generalized Rør-
dam groups R(C ; D, B), cf. Chapter 7, to the C-equivariant extension groups
Ext(C(R); D, SB) ∼= KK(C; D, B) introduced in Chapter 5 and compared with
KK-groups in Chapter 8.
The criteria (wDC) shows that the injectivity of this group epimorphism is
equivalent ???
to a suitable version of homotopy invariance for the groups R(C ; D, B).
Not proved until now !!! Imprecise idea exists?
The Rørdam groups are unsuspended – but stable – versions of, from a cone
C depending, E-theory groups. The definition of R(C ; D, B) is entirely algebraic
and does not require any sort of homotopy invariance in its formulation, even not
implicitly.
 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 543

A cited condition (vwDC) is not defined somewhere!!!

In fact we show that a below given formally weaker condition (vwDC) implies that
Γ(H0 (D), J, E) = 0.
We have used in the formulation of Parts (iii) and (iv) in Theorem 4.4.6 the
terminus “specified”, that should say the following (together with Part (v)): We
demand that our “given” natural system (E, J, D, H0 , h0 , β) has the property that
F , s0 and t0 exist with the properties listed in Parts (iii). But then we fix one of the
possible systems (F, s0 , t0 ) and add it to the structure given by (E, J, D, H0 , h0 , β)
simply for further references in the proof. The additional (F, s0 , t0 ) are not uniquely
determined by (E, J, D, H0 , h0 , β). The F in most of our applications satisfies in
addition that K1 (F ) = 0, that can not be derived from the assumptions of Theorem
4.4.6.
The difference to Part (v) is that this is an additional requirement on F , and
we need from Part (vi) only that such u0 exists in later proofs it not necessary to
use always the same u0 with the in (vi) quoted property.

The Parts (i)–(v) of following Lemma 4.4.7 are not obvious because [ϕ] =
[ϕ + 0] 6= [ϕ] + [0] = [ϕ ⊕ 0] for C *-morphisms ϕ that do not dominate zero. The
used isometry s0 and p0 := s0 s∗0 are given by assumption (iv) of Theorem 4.4.6.
Recall that C := H0 (D)0 ∩ E in Theorem 4.4.6.

Lemma 4.4.7. The assumptions (i)–(vi) on h0 and H0 in Theorem 4.4.6 imply

the following properties (i)-(xii):

(i) Let Bj , j = 1, 2, separable C*-algebras, kj : Bj → J ⊆ E C*-morphisms,

and use u1 ∈ U(E) from assumption (vi) in Theorem 4.4.6. Then, for all
b ∈ B1 , c ∈ B2 and d ∈ D,

u1 (k1 (b) + βh0 (d)) ⊕s,t k2 (c) u∗1 = k1 (b) ⊕s,t (k2 (c) + βh0 (d)) .

In particular, for all C*-morphisms k1 , k2 : D → J ⊆ E ,

[k1 + βh0 ] + [k2 ] = [k1 ] + [k2 + βh0 ] .

For each σ-unital C*-algebra B and C*-morphism k : B → J + βJ

holds [k] = [k] + [0]. Moreover, for every isometry T ∈ E there exists a
unitary u ∈ U0 (E) such that k(b)u = k(b)T ∗ for all b ∈ B. In particular,
u∗ k(·)u = T k(·)T ∗ .
Special cases are [k1 + βk3 ] = [k1 + βk3 ] + [0], for all C*-morphisms
k1 , k3 : D → J ⊆ E, and – with k2 := 0, k3 := h0 –,

[k1 + βh0 ] = [k1 + βh0 ] + [0] = [k1 ] + [βh0 ] .

(ii) [h0 ] + [βh0 ] = [h0 + βh0 ] = [h0 + βh0 ] + [h0 ] = [h0 + βh0 ] + [0].
(iii) [H0 ] + [h0 + βh0 ] = [H0 ].
(iv) [H0 ] + [h0 ] = [H0 ] = [H0 ] + [βh0 ].
(v) Each of H0 , h0 + βh0 , βh0 and h0 dominates zero.
544 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

(vi) There exist isometries s1 , t1 ∈ E with s1 s∗1 + t1 t∗1 = 1 , p0 t1 = p0 = t1 p0

and t1 H0 (·) = H0 (·) = H0 (·)t1 . In particular, t1 ∈ C, s1 s∗1 ≤ 1 − p0 ,
s1 s∗1 H0 (·) = 0 and s1 s∗1 Es1 s∗1 = s1 Es∗1 ⊆ Ann(H0 (D), E) ⊆ C.
(vii) If [h] ∈ S(H0 ; D, E) and h(D) ⊆ J + βJ, then h has a unique decom-
position h = k1 + βk2 with C*-morphisms k1 , k2 : D → J that satisfy
[k1 ], [k2 ] ∈ S(h0 ; D, E).
In particular, [k] ∈ S(h0 ; D, E) if [k + βh0 ] ∈ S(H0 ; D, E) and
k(D) ⊆ J.
If x ∈ E is a contraction, such that (1 − xx∗ )H0 (D)x = {0} and
x∗ H0 (D)x ⊂ J + βJ, then there are unique k1 , k2 ∈ S(h0 ; D, E) with
k1 + βk2 = x∗ H0 (·)x.
This applies to projections x := P ∈ C with H0 (D)P ⊆ J + βJ.
(viii) Let B a σ-unital C*-algebra and k1 , k2 : B → J C*-morphisms. Then
[k1 ] = [k2 ], if and only if, there exists a unitary u ∈ U0 (βh0 (D)0 ∩ E) with
βh0 (d)u = βh0 (d) for all d ∈ D and and k2 (b) = u∗ k1 (b)u for all b ∈ B.
In particular, then u ∈ U0 (E) and β(u) ∈ h0 (D)0 ∩ E.
If B := D then [k1 + βh0 ] = [k2 + βh0 ] implies [k1 ] = [k2 ].
If B is a σ-unital C*-algebra, k : B → J a C*-morphism and T ∈
E is an isometry, then there exists u ∈ U0 (βh0 (D)0 ∩ E) that satisfies
u∗ k(·)u = T k(·)T ∗ and βh0 (d)u = βh0 (d) for all d ∈ D.
(ix) Let q1 , q2 ∈ F denote non-zero commuting projections in F . Then there
exists a unitary u ∈ (F + 1) ∩ U0 (F + C · 1) ⊆ C with u∗ q1 u = q2 .
For every separable subset X ⊆ J + βJ there exists a projection p ∈ F
and an isometry R ∈ C with pR = 0 and pxp = x for all x ∈ X.
For each isometry V ∈ E (or unitary V ) there exists U ∈ U0 (E) with
∗
xV = xU for all x ∈ X.
(x) There exist isometries s2 , s3 ∈ E that satisfy

s∗2 H0 (·)s2 = h0 , s∗3 H0 (·)s3 = βh0 , and s2 s∗2 + s3 s∗3 = p0 .

The projection p2 := s2 s∗2 is in C, and and p2 , s2 and s3 satisfy s2 h0 s∗2 =

p2 H0 (·) = s0 h0 s∗0 and

s3 βh0 s∗3 = (p0 − p2 )H0 (·) = s0 βh0 s∗0 = β(s0 h0 s∗0 ) .

In particular, p2 H0 (D) ⊆ J and (p0 − p2 )H0 (D) ⊆ βJ.

If s, t ∈ C are the isometries in Theorem 4.4.6(i), then the partial
isometries s2 ss∗2 , s2 ts∗2 , s3 ss∗3 , s3 ts∗3 , s0 ss∗0 and s0 ts∗0 are in C.
In particular, the projections p2 = s2 s∗2 , p0 − p2 = s3 s∗3 and p0 are
properly infinite in C, satisfy 0 = [p2 ] ∈ K0 (p2 Cp2 ),
??? 0 = [p0 − p2 ] ∈ K0 ((p0 − p2 )C(p0 − p2 )) ??? or = 0 in K0 (p0 Cp0 )
???
and [p0 ] = 0 in K0 (p0 Cp0 ).
There exists an isometry t2 ∈ C with t2 t∗2 = 1 − p2 and unitary
elements u, v ∈ U(C) such that

u∗ p2 u = p2 + t2 p2 t∗2 and v ∗ p2 v = p2 ⊕s,t p2 .

 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 545

The elements s∗0 s2 , s∗0 s3 ∈ E are isometries with (s∗2 s0 )(s∗0 s3 ) = 0.

If B is a σ-unital C*-algebra and h : B → J a C*-morphism then there
exists uh ∈ U0 ((βh0 )(D)0 ∩ E) such that βh0 (·)uh = βh0 and eh(b)uh =
eh(b)s∗2 s0 for all b ∈ B and e ∈ E.
(xi) For every non-zero projection p ∈ F there exists properly infinite pro-
jections q, r ∈ C with q + r = p, qr = 0, q ∈ N (H0 (D), J) and
r ∈ N (H0 (D), βJ). Each decomposition p = q + r of this kind satisfies
[qH0 (·)] = [h0 ] and [rH0 (·)] = [βh0 ].
(xii) If p ∈ F is a projection with p ≥ p0 and p 6= p0 , then there exist V ∈ U(C)
and W ∈ U0 (E) such that W ∗ (h0 + βh0 )W = p0 H0 (·), V p0 = p0 = p0 V
and
V ∗ (p − p0 )H0 (·)V = t0 W ∗ (h0 + βh0 )W t∗0 .

Proof. We give a pre-information in Point (o) and prove Part (ix) before the
other Parts of Lemma 4.4.7.
(o): Assumptions (iii) and (iv) of Theorem 4.4.6 together imply p0 H0 (d) =
s0 h0 (d)s∗0 + s0 βh0 (d)s∗0 and t0 H0 (d)t∗0 = (1 − p0 )H0 (d) for d ∈ D. Thus

H0 = (h0 + βh0 ) ⊕s0 ,t0 H0 .

(ix): Let q1 , q2 ∈ F non-zero commuting projections. We show the existence

of u ∈ U0 (F + C · 1) ∩ (1 + F ) with u∗ q1 u = q2 :
If q1 , q2 ∈ F are projections with q1 q2 = q2 q1 , then

qj ≤ q3 := (q1 + q2 ) − (q1 q2 ) ∈ F for j ∈ {1, 2} ,

and q3 is a projection. The unitaries in (1 + F ) ∩ U0 (F + C · 1) build a group. Thus,

it suffices to consider the special cases where 0 6= q1 ≤ q2 :
By assumption (iii) of Theorem 4.4.6 with f = 0, there is a projection p ∈ F ⊆
C := H0 (D)0 ∩ E such that p ≥ q2 and p 6= q2 .
Any non-zero projection q ∈ F is unitary equivalent to p0 by a unitary w ∈
1 + F by assumption (v) of Theorem 4.4.6. Thus, there are unitaries wk , w ∈
1 + F , k ∈ {1, 2}, with wk∗ p0 wk = qk and w∗ p0 w = p − q2 . The partial isometries
vk := wk∗ p0 w ∈ F , satisfy vk vk∗ = qk , vk∗ vk = p − q2 and vk2 = 0. It follows that
qk + (p − q2 ) = vk vk∗ + vk∗ vk and

Uk := vk∗ + vk + (1 − vk vk∗ + vk∗ vk )

are selfadjoint unitaries in U0 (F + C · 1) with Uk qk = (p − q2 )Uk for k = 1, 2. Hence,

u∗ q2 u = q1 and u ∈ (F + 1) ∩ U0 (F + C · 1) for u := U2∗ U1 .
(If E is a real C *-algebra E then we can take here u := V2∗ V1 with unitaries

Vk := exp((π/2)(vk∗ − vk )) = vk∗ − vk + (1 − vk vk∗ − vk∗ vk )

because also Vk qk = (p − q2 )Vk .)

546 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Let p ∈ F is a projection with p ≥ p0 . We show the existence of isometries

R, T ∈ C, that satisfy pR = 0, pT = T p = p and T T ∗ + SS ∗ = 1 for S := Rs0 ∈ E
(not necessarily in C):
Let u ∈ U0 (F + C · 1) ⊆ U0 (C) with u∗ p0 u = p ≥ p0 .
If p := p0 , then we can take the isometries R0 , T0 ∈ C and S0 ∈ E defined by
R0 := t0 , T0 := p0 + t20 t∗0 and S0 := R0 s0 = t0 s0 .
The assumptions s0 s∗0 = p0 , s0 s∗0 + t0 t∗0 = 1, p0 ∈ F ⊆ C, t0 ∈ C in parts
(iii,iv) of Theorem 4.4.6, show that S0 ∈ E and R0 , T0 ∈ C are isometries with
R0 R0∗ = 1 − p0 , p0 R0 = 0, p0 S0 = 0, and T0 T0∗ + S0 S0∗ = 1.
It follows that R := u∗ R0 , S := Rs0 = u∗ S0 and T := u∗ T0 u have the desired
properties with respect to p = u∗ p0 u.

For every separable subset X ⊆ J + βJ there exists a projection p ∈ F and an

isometry R ∈ C with pxp = x and xR = 0 for all x ∈ X:
Let A := C ∗ (X) ⊆ J + βJ the separable C *-subalgebra generated by X and
f ∈ A+ a strictly positive contraction for A. By assumptions (iii,c) and (iv) of
Theorem 4.4.6 there exists a projection p ∈ F with pf = f = f p and p ≥ p0 =
s0 s∗0 ∈ F . Since f Af is dense in A, we get pxp = x for all x ∈ X ⊆ A.
Let p ∈ F with p ≥ p0 and pxp = x for all x ∈ X.
There exists u ∈ U0 (C) with u∗ p0 u = p as shown above. The isometry R = u∗ t0
is in C and satisfies pR = 0. It follows that xR = xpR = 0 for all x ∈ X.

Let X ⊆ J + βJ a separable subset, and V ∈ E an isometry. We show the

existence of U ∈ U0 (E) with xV ∗ = xU for all x ∈ X :
If we replace the above considered separable subset X ⊆ J +βJ by the separable
subset XV ∗ ∪ X ⊆ J + βJ, then the above observations lead to a projection p ∈ F
with p ≥ p0 , xp = x and xV ∗ p = xV ∗ for all x ∈ X.
Take a unitary u ∈ U0 (F + C · 1) ∩ (1 + F ) with u∗ p0 u = p. Let R := u∗ t0 ,
S := Rs0 and T := p + Rt0 R∗ as above considered with pS = 0, pT = p = T p and
SS ∗ + T T ∗ = 1.
Thus, for x ∈ X, xS = xpS = 0, xT = xpT = xp = x and

xV ∗ T ∗ = xV ∗ pT ∗ = xV ∗ p = xV ∗ .

It implies that xT V ∗ T ∗ = xV ∗ and xSe = 0 for all x ∈ X and e ∈ E. We define

U ∈ E by
U := T V ∗ T ∗ + SV S ∗ + S(1 − V V ∗ )T ∗ .

Since xS = 0 we get

xU = xT V ∗ T = xV ∗ for all x ∈ X .

It remains to prove that U ∈ U0 (E) :

 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 547

The operator Z := [T, S] ∈ M1,2 (E) ⊂ M2 (E) satisfies ZZ ∗ = 1 and Z ∗ Z =

12 := diag(1, 1), i.e.,
ψ : a ∈ M2 (E) 7→ ZaZ ∗ ∈ E
is a *-algebra isomorphism ψ from M2 (E) onto E. Obviously, the above defined
unitary U ∈ U(E) is the image U = ψ(U (V ∗ )) of the Halmos unitary U (V ∗ ) ∈
M2 (E) built from the contraction V ∗ , as defined in Remark 4.2.4. The Halmos
unitaries are all in U0 (M2 (E)). Hence, U = ψ(U (V ∗ )) ∈ U0 (E).

(i): Let kj : Bj → J C *-morphisms of C *-algebras Bj , j ∈ {1, 2, 3, 4}. The

property β(J) ∩ J = {0} shows that the linear maps (a, b) 7→ ki (b) + βkj (c) from
Bi ⊕ Bj into J + βJ are C *-morphisms.
If Bj := B for j ∈ {1, 2, 3, 4} then this implies that the linear maps b ∈ B 7→
kj (b) + βkj+2 (b) (j ∈ {1, 2}) are C *-morphism form B into J + βJ. The equality
[k1 + βk3 ] = [k2 + βk4 ] in [Hom(B, J + βJ)] implies [k1 ] = [k2 ] and [k3 ] = [k4 ],
because there exists a unitary u ∈ E that satisfies for v := β(u) and all b ∈ B that

J 3 (u∗ k1 (b)u − k2 (b)) = −β(vk3 (b)v ∗ − k4 (b)) ∈ βJ ,

which implies that u∗ k1 (·)u = k2 and v ∗ k3 (·)v = k4 by assumption J ∩ βJ = {0}

of Theorem 4.4.6.
In particular kj + βh0 for j = 1, 2 are C *-morphisms from D into J + βJ if
B1 := B2 := D and k3 := h0 , because h0 (D) ⊆ J.
If we use the Cuntz sum ⊕s,t with s, t in Theorem 4.4.6(i), then

u1 (e + βh0 (d)) ⊕s,t f u∗1 = e ⊕s,t (f + βh0 (d)) for all d ∈ D, e, f ∈ J , (4.5)

where u1 ∈ U(E) comes from assumption (vi) of Theorem 4.4.6. We use here that
u1 ∈ U(J 0 ∩E) by assumption 4.4.6(vi), that β(s) = s and β(t) = t are in h0 (D)0 ∩E
by assumption (i) of Theorem 4.4.6, and that – therefore – assumption 4.4.6(vi)
implies moreover
u1 sβh0 (·)s∗ u∗1 = tβh0 (·)t∗ . (4.6)

Equation (4.5) says that (k1 (b) + βh0 (d)) ⊕s,t k2 (c) is unitarily equivalent to
k1 (b) ⊕s,t (βh0 (d) + k2 (c)) by the unitary u1 for all b ∈ B1 , c ∈ B2 and d ∈ D.
If B1 = B2 = D we get especially

[k1 + βh0 ] + [k2 ] = [k1 ] + [k2 + βh0 ] .

An other special case is k2 := 0, k1 := k and B1 := B for σ-unital C *-

algebras B and C *-morphisms k : B → J. We get that the C *-morphisms (b, d) 7→
(k(b) + βh0 (d)) ⊕s,t 0 and (b, d) 7→ k(b) ⊕s,t βh0 (d) are unitary equivalent.
It implies [k + βh0 ] + [0] = [k] + [βh0 ] for k : D → J.

Let B any σ-unital C *-algebra, k : B → J + βJ a C *-morphism and T ∈ E an

isometry. There exists a strictly positive contraction b0 ∈ B+ for B if B is σ-unital.
If we apply the above proven Part (ix) to the subset X := {k(b0 )} of J + βJ,
then we see that there exists u ∈ U0 (E) with k(b)T ∗ = k(b)u for all b ∈ Bb0 . We get
548 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

k(b)T = k(b)u for all b ∈ B = Bb0 . Thus T k(b∗ a)T ∗ = u∗ k(b∗ a)u for all a, b ∈ B.
It yields T k(·)T ∗ = u∗ k(·)u, because both sides are bounded linear maps and the
linear span of B ∗ · B is dense in B (actually B · B = B by [616, prop. 1.4.5]).
If we take T := s ∈ C with s from assumption (i) of Theorem 4.4.6, then we
get [k] + [0] = [k ⊕s,t 0] = [T k(·)T ∗ ] and [sk(·)s∗ ] = [k].

(ii): We get [h0 ] + [βh0 ] = [h0 + βh0 ] + [0] = [h0 + βh0 ], if we let k1 := h0 ,
k2 := 0 and k3 := h0 in Part (i). By Theorem 4.4.6(i), s, t ∈ h0 (D)0 ∩ E and
ss∗ + tt∗ = 1. It implies that [h0 ] + [h0 ] = [h0 ]. We obtain:

[h0 + βh0 ] + [h0 ] = [h0 ] + [βh0 ] + [h0 ] = [h0 ] + [βh0 ] .

(iii): Info (o) shows that [H0 ] = [h0 + βh0 ] + [H0 ] as elements of S(H0 ; D, E).

(iv): [H0 ] + [h0 ] = [H0 ] + [h0 + βh0 ] + [h0 ] = [H0 ] + [h0 + βh0 ] = [H0 ] in
S(H0 ; D, E), by Parts (iii) and (ii).
[H0 ] + [βh0 ] = [H0 ] + [h0 ] + [βh0 ] = [H0 ] + [h0 + βh0 ] = [H0 ] by Parts (ii, iii).

(v): By Part (i), with k3 := h0 and k1 = 0, we get [βh0 ] = [βh0 ] + [0], and,
with k3 := 0 and k1 = h0 , [h0 ] = [h0 ] + [0], i.e., that each of βh0 and h0 dominate
zero. Part (ii) implies that h0 + βh0 dominates zero. This and Part (iv) together
imply that H0 dominates zero, i.e., that [H0 ] = [H0 ] + [0].

(vi): By Proposition 4.3.6(v), it suffices to show that there exists an isometry

R ∈ E with p0 R = 0 and R∗ H0 (·)R = 0.
Part (v), – i.e., [H0 ] = [0] + [H0 ] by Proposition 4.3.5(i) –, and assumptions
(i,iv) of Theorem 4.4.6 imply the existence of a unitary u ∈ E with

u∗ H0 (·)u = 0 ⊕s0 ,t0 H0 (·) = t0 H0 (·)t∗0 = (1 − p0 )H0 (·) .

Let R := t0 us0 . Then R∗ R = 1, p0 R = 0 and t∗0 H0 (·)R = H0 (·)us0 , because s0

and t0 are isometries with p0 = s0 s∗0 , s∗0 t0 = 0 and t0 ∈ C – by assumptions (iv) in
Theorem 4.4.6. It follows for a ∈ D that

R∗ H0 (a)R = s∗0 u∗ H0 (a)us0 = s∗0 t0 H0 (a)t∗0 s0 = 0 .

(vii): If [h] ∈ S(H0 ; D, E) then there exists an isometry T ∈ E with h =

∗
T H0 (·)T by definition of S(H0 ; D, E).
If h(D) ⊆ J + βJ, then the bounded linear map λ(d) := H0 (d)T (d ∈ D)
maps D into J + βJ. The C *-subalgebra A = C ∗ (λ(D)) of J + βJ generated by
the image λ(D) of the separable D by the linear contraction λ is separable. Let
f ∈ A+ a strictly positive contraction. By assumptions (iii) and (iv) of Theorem
4.4.6, there is a projection p ∈ F ⊆ C := H0 (D)0 ∩ E such that pf = f and p ≥ p0 .
In particular T ∗ pH0 (d)T = T ∗ H0 (d)T = h(a) for d ∈ D. The above proven Part
(ix) provides a unitary u ∈ (F + 1) ∩ U0 (F + C · 1) ⊆ U0 (C) with upu∗ = p0 . It
follows that h = T0∗ p0 H0 (·)T0 for the isometry T0 := uT . Let k1 := T0∗ s0 h0 (·)s∗0 T0
and k2 := β(T0 )∗ s0 h0 (·)s∗0 β(T0 ) . Then kj (D) ⊆ J for j = 1, 2 and k1 , k2 are
 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 549

C *-morphisms from D into J with h = k1 + βk2 , because p0 H0 = s0 (h0 + βh0 )s∗0 ,

h0 (D) ⊆ J, β(s0 ) = s0 and J ∩ βJ = {0}.
By Part (v), h0 dominates zero, i.e., [h0 ] = [0] + [h0 ] and there exists a unitary
v ∈ E such that such that v ∗ h0 (·)v = s0 h0 (·)s∗0 . Then T1 := vT0 = vuT and
T2 := vβ(T0 ) are isometries in E such that h = k1 + βk2 for kj := Tj∗ h0 (·)Tj
(j = 1, 2). In particular, [k1 ], [k2 ] ∈ S(h0 ; D, E).
The property J · β(J) = J ∩ β(J) = {0} of J in Theorem 4.4.6 implies that
arbitrary maps k1 , k2 : B → J of any C *-algebra B must be automatically C *-
morphisms if h := k1 + βk2 is a C *-morphism and that k1 + βk2 is the unique
decomposition of h as sum h = k1 + βk2 by maps k1 , k2 : B → J.
Thus, h = k1 + βk2 with [k1 ], [k2 ] ∈ S(h0 ; D, E) is the unique decomposition
of h := T ∗ H0 (·)T with h(D) ⊆ J + βJ.
The C *-morphism H0 dominates zero by Part (v). Hence Proposition 4.3.6(ii)
applies and gives that [h] ∈ S(H0 ; D, E) for h := x∗ H0 (·)x if x ∈ E is a contraction
with (1 − xx∗ )H0 (D)x = {0}.
It happens in particular for all projections x := P ∈ C.
If, in addition, x∗ H0 (D)x ⊂ J + βJ, then above arguments show that h =
k1 + βk2 for unique k1 , k2 : D → J, that satisfy [k1 ], [k2 ] ∈ S(h0 ; D, E).

(viii): Let k, ` : B → J + βJ C *-morphisms with [k] = [`] for a σ-unital C *-

algebra B, i.e., suppose that there exists V ∈ U(E) with V k(·)V ∗ = `(·).
If we apply the before proven Part (ix) to X := {k(b0 )} then we get u ∈ U0 (E)
with k(b)V ∗ = k(b)u for all b ∈ B, as in the proof of Part (i). Thus u∗ k(·)u =
V k(·)V ∗ = `(·).
This shows that if B is σ-unital and k1 , k2 : B → J satisfy [k1 ] = [k2 ] then
there exists a (norm-) continuous path ξ ∈ [0, 1] 7→ Uξ ∈ U(E) with U0 = 1 and
U1∗ k1 (·)U1 = k2 .
Let b0 ∈ B a strictly positive contraction in B, and let A0 denote the separable
C *-subalgebra of J + βJ generated by βh0 (D) and by the elements k1 (b0 )Uξ ∈ J
for n ∈ N and ξ ∈ [0, 1]. Notice that A0 = (A0 ∩J)+βh0 (D), because J ∩βJ = {0}.
Let a0 a strictly positive contraction of A0 ∩ J and d0 a strictly positive contraction
of D. It follows that the C *-subalgebra A of J + βJ generated by βh0 (D) and
k1 (b)Uξ ∈ J for b ∈ B and ξ ∈ [0, 1] is σ-unital with strictly positive contraction
a0 + βh0 (d0 ), because again A = (A ∩ J) + βh0 (D), A0 ⊂ A, and each k1 (b)Uξ is
contained in the closed left-ideal generated by k1 (b0 )Uξ , i.e., a0 ∈ A ∩ J ⊆ J · a0 .
If we let h : A → J + βJ denote the identity map on the σ-unital C *-algebra
A, then Part (i) applies to A and h. The Part (i) shows that there exists a unitary
W ∈ U0 (E) with aW ∗ = as∗ for all a ∈ A, where we consider the isometries s, t ∈ E
from assumption (i) of Theorem 4.4.6. It implies W a = sa and W aW ∗ = sas∗ for
a ∈ A, because A = A · A as set.
550 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Equation (4.5) yields

(u1 W )(e + βh0 (d))(u1 W )∗ = ses∗ + tβh0 (d)t∗ for all d ∈ D, e ∈ A ∩ J . (4.7)

In particular, tβh0 (·)t∗ (u1 W ) = u1 W βh0 (·) . Equation (4.7) holds more generally
for all d ∈ D and all b ∈ J with the property that W bW ∗ = sbs∗ .
We define a norm-continuous group-morphism λW : U(E) → U(E) by

λW (U ) := W ∗ u∗1 (U ⊕s,t 1)u1 W for U ∈ U(E) .

Clearly λW (U ∗ ) = λW (U )∗ . Straight calculation shows that λW (U ) · βh0 (d) =

βh0 (d) for d ∈ D and all U ∈ U(E). The property u1 ∈ J 0 ∩ E implies that
λW (U ∗ )aλW (U ) = U ∗ aU if a ∈ J and U ∈ U(E) satisfy that W aW ∗ = sas∗ and
W U ∗ aU W ∗ = sU ∗ aU s∗ .
By our above construction of A = (A ∩ J) + βh0 (D) we get a norm-continuous
path ξ ∈ [0, 1] 7→ Vξ ∈ U(E) with V0 = 1, Vξ∗ k1 (·)Vξ = Uξ∗ k1 (·)Uξ and

Vξ · βh0 (·) = h0 (·) = βh0 (·) · Vξ

if we define Vξ by
Vξ := λW (Uξ ) .
In particular, Vξ ∈ βh0 (D)0 ∩ E for each ξ ∈ [0, 1], because βh0 (d)Vξ = βh0 (d) for
all d ∈ D. Thus, u := V1 ∈ U0 (βh0 (D)0 ∩ E) satisfies u∗ k1 (b)u = k2 (b) for b ∈ B
and uβh0 (d) = h0 (d) = βh0 (d)u.
If k : B → J is a C *-morphism, B is separable and T ∈ E an isometry then
there exists a unitary v ∈ U0 (E) with v ∗ k(·)v = T k(·)T ∗ by Part (i) (proven via
Part (ix)).
If we apply above observations to k1 := k and k2 := T k(·)T ∗ with k2 =
v k1 (·)v, then we obtain the existence of a unitary u ∈ U0 (β(h0 (D))0 ∩ E) that
∗

satisfies βh0 (·)u = βh0 and u∗ k(·)u = T k(·)T ∗ .

(x): Recall that s∗0 H0 (·)s0 = h0 + βh0 . By Part (ii), [h0 + βh0 ] = [h0 ] + [βh0 ].
Thus, there is a unitary V ∈ U(E) such that

V (h0 + βh0 )V ∗ = sh0 s∗ + tβh0 t∗ ,

where the isometries s, t come from Theorem 4.4.6(i,vi). If we apply the assumption
(i) of Theorem 4.4.6 to k := h0 + βh0 and T := V , we get some u ∈ U0 (E) with

u∗ s∗0 H0 (·)s0 u = sh0 s∗ + tβh0 t∗ .

Thus, s2 := s0 us and s3 := s0 ut are isometries in E that satisfy s2 s∗2 +s3 s∗3 = p0 ,

s∗2 H0 (·)s2 = h0 and s∗3 H0 (·)s3 = βh0 .
By Lemma 4.3.4(i), the equation s∗2 H0 (·)s2 = h0 implies that p2 := s2 s∗2 ∈ C
and p0 − p2 = s3 s∗3 ∈ C because p0 ∈ F ⊆ C by Theorem 4.4.6(iv,iii).
Since p0 , p2 ∈ C, we get p2 H0 (a) = s2 h0 (a)s∗2 ∈ J, and (p0 − p2 )H0 (a) =
s3 βh0 (a)s∗3 ∈ βJ for a ∈ D.
 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 551

To compare e.g. s2 h0 s∗2 with s0 h0 s∗0 , we define linear maps L1 , L2 : D → E by

L1 (a) := (p2 H0 (a)) − s0 h0 (a)s∗0 and L2 (a) := ((p0 − p2 )H0 (a)) − s0 βh0 (a)s∗0 .

Then L1 + L2 = 0, i.e., L1 (a) = −L2 (a) for a ∈ D, L1 (D) ⊆ J and L2 (D) ⊆ βJ.
The orthogonality J · βJ = J ∩ βJ = {0} yields L1 (a) = 0 and L2 (a) = 0 for
a ∈ D. Thus, p2 H0 (·) = s0 h0 s∗0 and (p0 − p2 )H0 (·) = s0 (βh0 )s∗0 , i.e., s2 h0 s∗2 =
s0 h0 s∗0 and s3 (βh0 )s∗3 = s0 (βh0 )s∗0 .

We let G1 := h0 (D)0 ∩ E and G2 := βG1 , then s0 Gj s∗0 ⊆ C for j = 1, 2,

s2 G1 s∗2 ⊆ C and s3 G2 s∗3 ⊆ C as the following calculations show:
The elements p0 = s0 s∗0 , p2 := s2 s∗2 and p0 − p2 := s3 s∗3 are in C, i.e., commute
with the elements of H0 (D). The definitions of s0 , s2 and s3 give us that

s∗0 H0 s0 = h0 + βh0 , s∗2 H0 s2 = h0 and s∗3 H0 s3 = βh0 .

Thus,
H0 s0 = s0 (h0 + βh0 ) , H0 s2 = s2 h0 , and H0 s3 = s3 βh0 .
It implies also

(h0 + βh0 )s∗0 = s∗0 H0 , h0 s∗2 = s∗2 H0 , and βh0 s∗3 = s∗3 H0 .

Let x ∈ G1 , d ∈ D, then

H0 (d)s2 xs∗2 = s2 h0 (d)xs∗2 = s2 xh0 (d)s∗2 = s2 xs∗2 H0 (d) .

The verifications of s3 G2 s∗3 ⊆ C and s0 Gj s∗0 ⊆ C (j = 1, 2) are similar.

The isometries s, t ∈ E commute element-wise with h0 (D) ∪ H0 (D) and are
β-invariant by Theorem 4.4.6(i), and since t0 ∈ C by Theorem 4.4.6(iv), the partial
isometries s2 ss∗2 , s2 ts∗2 , s3 ss∗3 , s3 ts∗3 , s0 ss∗0 , s0 ts∗0 , t0 st∗0 and t0 tt∗0 . are all in C.
It shows that the projections p2 = s2 s∗2 ≤ p0 , p0 −p2 and p0 are properly infinite
in p2 Cp2 , (p0 − p2 )C(p0 − p2 ) and p0 Cp0 and there (!) [p2 ] = 0, [p0 − p2 ] = 0 and
[p0 ] = 0, i.e., this hereditary C *-subalgebras of C contain copies of O2 unital.
Since 1 − p2 ≥ 1 − p0 = t0 t∗0 and p2 , t0 ∈ C it follows that 1 − p2 , 1 − p0 and
1 = 1C are full and properly infinite projections in C with same class in K0 (C).
Thus, by Lemma 4.2.6(ii), there exists an isometry t2 ∈ C with t2 t∗2 = 1 − p2 .

There exist u2 , v2 ∈ U(C) such that

u∗2 p2 u2 = p2 + t2 p2 t∗2 and v2∗ p2 v2 = p2 ⊕s,t p2 .

This happens because p2 is properly infinite with [p2 ] = 0 in K0 (p2 Cp2 ), because
the projections p2 + t2 p2 t∗2 and p2 ⊕s,t p2 are in the ideal of C generated by p2 , and
the complementary projections 1 − p2 , 1 − (p2 + t2 p2 t∗2 ) and 1 − (p2 ⊕s,t p2 ) are
respectively the ranges of the isometries t2 , t22 and t2 ⊕ t2 .
The Murray–von-Neumann equivalences

q0 := p0 + t0 p0 t∗0 ∼ p0 ∼ p0 ⊕s,t p0 =: r0
552 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

and
q2 := p2 + t2 p2 t∗2 ∼ p2 ∼ p2 ⊕s,t p2 =: r2

can be given in C explicitly by the partial isometries z0 := s0 ss∗0 +s0 ts∗0 t∗0 , and z2 :=
s2 ss∗2 + s2 ts∗2 t∗2 , respectively by y0 := (s0 ⊕s,t s0 )s∗0 and y2 := (s2 ⊕s,t s2 )s∗2 . Thus,
below listed unitary elements u0 , u2 , v0 , v2 ∈ U(C) satisfy u0 q0 = p0 u0 , u2 q2 = p2 u2 ,
v0 p0 = r0 v0 and v2 p2 = r2 v2 .

u0 := s0 ss∗0 + s0 ts∗0 t∗0 + t0 (t∗0 )2 , (4.8)

u2 := s2 ss∗2 + s2 ts∗2 t∗2 + t2 (t∗2 )2 , (4.9)
v0 := ss0 s∗ s∗0 + ts0 t∗ s∗0 + st0 s∗ t∗0 + tt0 t∗ t∗0 , (4.10)
v2 := ss2 s∗ s∗2 + ts2 t∗ s∗2 + st2 s∗ t∗2 + tt2 t∗ t∗2 . (4.11)

The MvN-equivalences can be seen also from

p0 = (1 ⊕s,t 1) ⊕s0 ,t0 0 ,

p0 + t0 p0 t∗0 = 1 ⊕s0 ,t0 (1 ⊕s0 ,t0 0)
p0 ⊕s,t p0 = (1 ⊕s0 ,t0 0) ⊕s,t (1 ⊕s0 ,t0 0) ,

because the below given Lemma 4.4.8 recognizes that u0 , u2 v0 and v2 conjugate
the used iterations of Cuntz addition.

The elements S := s∗0 s2 and T := s∗0 s3 are isometries in E with S ∗ T = 0,

because s2 and s3 are isometries with s2 s∗2 + s3 s∗3 = p0 = s0 s∗0 .
If B is σ-unital and h : B → J is a C *-morphism then there exists a unitary
U ∈ U0 (βh0 (D)0 ∩ E) with U ∗ k(·)U = Sk(·)S ∗ and βh0 (d)U = h0 (d) for d ∈ D by
Part (viii). Take uh := U .
(xi): By assumption (v) of Theorem 4.4.6, there exists a unitary u ∈ 1+F ⊆ C
such that u∗ p0 u = p. Let q := u∗ p2 u and r := u∗ (p0 − p2 )u with p0 ≥ p2 ∈ C the
projection from Part (x).
Since F ⊆ C and p0 , p2 ∈ C, also q, r ∈ C and q + r = p.
The projections satisfy q ∈ N (H0 (D), J) and r ∈ N (H0 (D), βJ), because
u ∈ C and p2 H0 (·) = s0 h0 s∗0 ∈ J and (p0 − p2 )H0 (·) = β(s0 h0 s∗0 ) ∈ J.
More generally, suppose that are given a non-zero projection p ∈ F and pro-
jections p, r ∈ C with p = q + r with qH0 (D) ⊆ J and rH0 (D) ⊆ βJ.
By assumption 4.4.6(v), there exists u ∈ U(1 + F ) ⊆ C with u∗ p0 u = p. Let
p3 := uqu∗ ∈ C. Then p3 H0 (a) = u(qH0 (a))u∗ ∈ J and (p0 − p3 )H0 (a) ∈ βJ for
a ∈ D.
Thus, [p3 H0 (·)] = [qH0 (·)] and [(p0 − p3 )H0 (·)] = [rH0 (·)].
The same calculation as in proof of Part (x) with p2 replaced by p3 shows that
p3 H0 (·) = s0 h0 s∗0 and (p0 − p3 )H0 (·) = s0 βh0 s0 . Thus [qH0 (·)] = [h0 ] + [0] = [h0 ]
and [rH0 (·)] = [βh0 ] + [0] = [βh0 ] by Part (i).
 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 553

(xii): Let p ∈ F a projection with p ≥ p0 and p 6= p0 , where p0 = s0 s∗0 for

the isometries s0 , t0 in assumption (iv) of Theorem 4.4.6 with p0 ∈ F , t0 ∈ C and
t0 t∗0 = 1 − p0 .
If we let B := D, k := h0 + βh0 and T := s0 in Part (i), we get a unitary
W ∈ U0 (E) with

W ∗ (h0 + βh0 )W = s0 (h0 + βh0 )s∗0 = p0 H0 (·) .

The elements t0 , p0 and p − p0 are in C. Thus, the proof of Part (xii) reduces
to to the construction of V ∈ U(C) that satisfies

p0 V = p0 = V p0 and V ∗ (p − p0 )V = t0 p0 t∗0 (4.12)

Since p0 , p − p0 and p are non-zero commuting projections in F , we find by Part

(ix) unitaries
V1 , V2 ∈ U0 (F + C · 1) ⊆ U0 (C)
with V1∗ p0 V1 = p − p0 and V2∗ p0 V2 = p, and define V ∈ E by

V := p0 + V1∗ p0 t∗0 + V2∗ (1 − p0 )t∗0 .

Straight calculation shows that V is unitary and fulfills Equations (4.12). The
unitary V is in C because p0 , t0 , V1 and V2 are in C. 

Proof of Theorem 4.4.6 ( Part 1: ϕ is surjective ).

The C *-morphism H0 dominates h0 by the equation [h0 ] + [H0 ] = [H0 ] in Lemma
4.4.7(iv). Thus
S(h0 ; D, E) ⊆ S(H0 ; D, E) ,
and [k] + [h0 ] = [h] + [h0 ] implies [k] + [H0 ] = [h] + [H0 ]. Since [h0 ] = [h0 ] + [h0 ]
and [H0 ] = [H0 ] + [H0 ], the mapping

ϕ : [k] + [h0 ] 7→ [k] + [H0 ]

is well-defined and additive, i.e., is a group homomorphism from G(h0 ; D, E) =

S(h0 ; D, E) + [h0 ] into G(H0 ; D, E) = S(H0 ; D, E) + [H0 ]. It is useful to consider
G(h0 ; D, E) and G(H0 ; D, E) as subgroups of the big commutative semigroup
[Hom(D, E)] of all unitary equivalence classes of C *-morphisms from D into E
with Cuntz addition, as introduced below Definition 4.3.1. This kind of subgroups
of [Hom(D, E)] have natural morphism that are considered in parts (iii) and (iv)
of the Lemma 4.2.3 in more generality on semi-groups.

We show that the group homomorphism ϕ is surjective:

Let [g] ∈ S(H0 ; D, E). By assumption (ii) of Theorem 4.4.6, there exists a unitary
u ∈ E with

T (a) := u∗ (g(a) ⊕s0 ,t0 H0 (a))u − H0 (a) ∈ J for all a ∈ D .

554 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

We use here the isometries s0 , t0 from assumption (iv) of Theorem 4.4.6 to realize
the Cuntz sum ⊕ := ⊕s0 ,t0 , simply to make calculations more transparent. Com-
pare Proposition 4.3.2(ii) for the change of u if one takes another realization of ⊕
by other copies of O2 in E.
The image T (D) of the linear map T generates a separable C *-subalgebra
B ⊆ J, because D is separable. Let f ∈ J be a strictly positive element of B.
By assumption 4.4.6(iii), there exists a non-zero projection p ∈ F ⊆ C with
pf = f p = f . It follows that (1 − p)c = c(1 − p) = 0 for every c ∈ B, and that p
commutes element-wise with H0 (D). In particular, (1 − p)T (·) = 0 = T (·)(1 − p),
i.e., pT (a) = T (a) = T (a)p for all a ∈ D.
Since F · H0 (D) ⊆ J + βJ and F ⊆ C by assumption (iii) of Theorem 4.4.6,
we get for a ∈ D that

pu∗ (g(a) ⊕s0 ,t0 H0 (a))u = pT (a) + pH0 (a) ∈ J + βJ ,

pu∗ (g(a) ⊕s0 ,t0 H0 (a))u = u∗ (g(a) ⊕s0 ,t0 H0 (a))up

and
(1 − p)u∗ (g ⊕s0 ,t0 H0 )u = (1 − p)H0 .

Let s0 , t0 the isometries from Theorem 4.4.6(iv) with t0 ∈ C := H0 (D)0 ∩ E

and s0 s∗0 = p0 = 1 − t0 t∗0 with p0 ∈ F ⊆ C. By assumption (v) of Theorem 4.4.6,
there is a unitary v ∈ 1 + F ⊆ C with v ∗ pv = p0 .
We get pv = vp0 and that W := uvs0 is an isometry in E such that the
projection W W ∗ = uvp0 v ∗ u∗ = upu∗ commutes element-wise with (g⊕s0 ,t0 H0 )(D).
It implies that p0 = s0 s∗0 commutes element-wise with the image of
(uv)∗ (g ⊕s0 ,t0 H0 ) uv.
We define a C *-morphism h : D → J + βJ by

h := W ∗ (g ⊕ H0 )W = s∗0 (uv)∗ (g ⊕s0 ,t0 H0 ) uvs0 .

The commutation properties of p0 = s0 s∗0 implies that

s0 h(·)s∗0 = p0 (uv)∗ (g ⊕s0 ,t0 H0 ) uv = v ∗ (pV (·) + pH(·))v ∗ .

The equations

(1 − p0 )(uv)∗ (g ⊕s0 ,t0 H0 ) uv = v ∗ (1 − p)(T (·) + H0 (·))v = (1 − p0 )H0 (·) = t0 H0 (·)t∗0

and
s0 h(·)s∗0 = p0 (uv)∗ (g ⊕s0 ,t0 H0 ) uv ,
together lead to
v(h ⊕s0 ,t0 H0 )v ∗ = u∗ (g ⊕s0 ,t0 H0 )u .

If we use that s0 s∗0 = p0 ∈ F ⊆ C, t0 ∈ C and s0 s∗0 + t0 t∗0 = 1 by assumptions

(iii) and (iv) of Theorem 4.4.6, we get

t0 H0 (·)t∗0 = (1 − p0 )H0 (·) == (1 − p0 )(uv)∗ (g ⊕ H0 )uv

 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 555

and
h ⊕s0 ,t0 H0 = s0 h(·)s∗0 + (1 − p0 )H0 (·) = (uv)∗ (g ⊕s0 ,t0 H0 )uv .
Thus,
[h] + [H0 ] = [g] + [H0 ] ∈ S(H0 ; D, E) .
It says that H0 dominates h ⊕s0 ,t0 H0 by Definition 4.4.1. (Compare also Definition
4.3.3, Lemma 4.3.4(iv) and Proposition 4.3.6.)
The equation h = s∗0 (h ⊕s0 ,t0 H0 )s0 shows that H0 dominates also h, i.e.,
[h] ∈ S(H0 ; D, E).
Since v ∈ F + 1 ⊆ C is unitary, we get that v ∗ H0 (·)v = H0 , and, by definition
of T and v, that p0 v ∗ T (a)v = v ∗ T (a)v ∈ J for a ∈ D. It implies

s0 h(a)s∗0 − p0 H0 (a) = p0 (vu∗ (g(a) ⊕s1 ,t1 H0 (a))uv) − H0 (a) = p0 v ∗ T (a)v .

By assumption 4.4.6(iv), s∗0 H0 (a)s0 = h0 (a) + βh0 (a) and s∗0 s0 = 1. Thus,

h(a) − h0 (a) − β(h0 (a)) = s∗0 v ∗ T (a)vs0 ∈ J for a ∈ D .

Since h0 (D) ⊆ J it follows that h(D) ⊆ J + β(J) and h = k + βh0 for a unique
C *-morphism k : D → J, because h0 (D) ⊆ J and J ∩ β(J) = {0}.
Now [h] = [k + βh0 ] ∈ S(H0 ; D, E) and h(D) ⊆ J + βJ implies that the
homomorphism h0 dominates k by Lemma 4.4.7(vii), i.e., [k] ∈ S(h0 ; D, E).
By Lemma 4.4.7(iii,i,iv),

[k] + [H0 ] = [k] + [h0 + βh0 ] + [H0 ] = [h] + [h0 ] + [H0 ] = [h] + [H0 ] .

It follows [k] + [H0 ] = [h] + [H0 ] = [g] + [H0 ] . Thus, the group morphism

ϕ : [k] + [h0 ] ∈ G(h0 ; D, E) 7→ [k] + [H0 ] ∈ G(H0 ; D, E)

is surjective. 

We start here the collection of observations that will be used in the proof of the
necessity and sufficiency of the Condition (DC) for the injectivity of the natural
epimorphism from G(h0 ; D, E) onto G(H0 ; D, E). It is essentially a detailed study
of its possibly existing kernel.

Lemma 4.4.8. Let (s, t) and (s0 , t0 ) the isometries from assumptions (i) respec-
tively (iv) of Theorem 4.4.6 and (s2 , t2 ) the isometries in Lemma 4.4.7(x).
The elements u0 , u2 , v0 and v2 – as defined by equations (4.8), (4.9), (4.10)
and (4.11) in the proof of Lemma 4.4.7(x) – are in U(C) and have the properties
that, for a, b, c, d ∈ E and k ∈ {0, 1},

u∗k (a ⊕s,t b) ⊕sk ,tk c uk = a ⊕sk ,tk (b ⊕sk ,tk c) ,

and
vk∗ (a ⊕sk ,tk c) ⊕s,t (b ⊕sk ,tk d) vk = (a ⊕s,t b) ⊕sk ,tk (c ⊕s,t d) .

In particular, for all C*-morphisms k, ` : D → J and h := k + β`,

u∗0 (h ⊕s,t (h0 + βh0 )) ⊕s0 ,t0 H0 u0 = h ⊕s0 ,t0 H0 ,

556 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

u∗2 (k ⊕s,t h0 ) ⊕s2 ,t2 H0 u2 = k ⊕s2 ,t2 H0 ,

and

v0∗ ((k + βh0 ) ⊕s0 ,t0 H0 ) ⊕s,t ((` + βh0 ) ⊕s0 ,t0 H0 ) v0 = ((k ⊕s,t `) + βh0 ) ⊕s0 ,t0 H0 .

Proof. The elements u0 , u2 , v0 and v1 of E are in C because s, t ∈ C by

assumption (i) of Theorem 4.4.6, t0 , t2 , s0 ss∗0 , s0 ts∗0 , s2 ss∗2 and s2 ts∗2 are in C by
assumption (iv) of Theorem 4.4.6 and Lemma 4.4.7(x).
Straight calculation shows that uk and vk are unitary for k ∈ {0, 2} and have
the quoted transformation properties.
Use (h0 + βh0 ) ⊕s0 ,t0 H0 = H0 and h0 ⊕s2 ,t2 H0 = H0 in the last equations. 

Lemma 4.4.9. Let A := H0 (D), U ∈ U N (A, J + βJ) + C · 1 and q ∗ q = q ∈ F .

There exists a projection p ∈ F with p ≥ q, p 6= q and

U ∗ (H0 (·)p)U = U ∗ H0 (·)U p , (4.13)

∗
U (H0 (·)(1 − p))U = H0 (·)(1 − p) . (4.14)

More generally we find for each separable subset X ⊆ N (A, J + βJ) and each
projection q ∈ F a projection p ∈ F with p ≥ q and p 6= q and H0 (·)x = H0 (·)pxp
for all x ∈ X.

Proof. Multiplying U by suitable ξ ∈ C we may suppose that U = 1 + y with

y ∈ N (A, J + βJ)).
More generally, let Y a separable C *-subalgebra of N (A, J + βJ)). Then the
separable subset X := (A·Y)∪(Y ·A) of J +βJ, generates a separable C *-subalgebra
B of J + βJ.
Let f ∈ B+ ∈ β(J) + J a strictly positive contraction for B. By assumption
(iii,c) of Theorem 4.4.6 there exists a projection p ∈ F with f = pf p, p ≥ q and
p 6= q. Thus pb = b = bp for all b ∈ B. In particular, (1 − p)B = {0} = B(1 − p).
Since A = H0 (D) and p ∈ F ⊆ C, it follows all y ∈ Y that H0 (·)py = H0 (·)y =
H0 (·)yp, pyH0 (·) = yH0 (·) = ypH0 (·) and (1 − p)H0 (·)y = 0 = H0 (·)y(1 − p).
H0 (·)y = H0 (·)pyp
If we let U := 1 + y then we get H0 (·)pU = H0 (·)U p and

H0 (·)(1 − p)U = (1 − p)H0 (·)U = (1 − p)H0 (·) .

In particular, U ∗ pH0 (·)U = U ∗ H0 (·)U p.

Notice that (1 − p)2 = 1 − p ∈ C. It implies for all c, b ∈ D that

U ∗ H0 (c∗ b)(1 − p)U = ((1 − p)H0 (c)U )∗ (1 − p)H0 (b)U = H0 (c∗ b)(1 − p) .

Lemma 4.4.10. Let A := H0 (D). The maps

λ0 : U(βh0 (D)0 ∩ E) 3 u 7→ s0 us∗0 + (1 − p0 ) ∈ U(E)

 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 557

and
λ2 : U(E) 3 u 7→ s2 us∗2 + (1 − p2 ) ∈ U(E)
are norm-continuous group morphisms into U(Der(A, J)) ∩ (p0 Ep0 + (1 − p0 )) .
If moreover βh0 (·)u = βh0 , then 1 − λ0 (u) ∈ N (A, J) ∩ p0 Ep0 .
For each u ∈ U(E) , 1 − λ2 (u) ∈ p2 N (A, J)p2 = p2 Ep2 ⊆ N (A, J) .
For every σ-unital hereditary C*-subalgebra B of J there exists a unitary wB ∈
U0 (βh0 (D)0 ∩ E) that satisfies βh0 (·)wB = βh0 and

bwB = bs∗2 s0 for all b ∈ B .

In particular, for each C*-morphism k : D → J with k(D) ⊆ B and the unitary

W := wB ⊕s0 ,t0 1 ∈ U0 (Der(A, J)) holds (1 − W ) ∈ N (A, J) ∩ p0 Ep0 and



(k + βh0 ) ⊕s0 ,t0 H0 W = W k ⊕s2 ,t2 H0 .

Proof. Recall that s0 (βh0 )s∗0 = (p0 − p2 )H0 = s3 (βh0 )s∗3 and (1 − p2 )H0 =
t2 H0 t∗2 by Lemma 4.4.7(x). By assumption (iv) of Theorem 4.4.6, t0 ∈ C and
t0 t∗0 = 1 − p0 . We obtain

s0 (βh0 )s∗0 + t0 H0 t∗0 = t2 H0 t∗2 = (1 − p2 )H0 .

It follows that

λ0 (u) := s0 us∗0 + (1 − p0 ) ∈ p0 Ep0 + (1 − p0 )

is a unitary in E.
The λ0 (u) commute elementwise with s0 βh0 (D)s∗0 ∪ (1 − p0 )H0 (D) if u ∈
U(βh0 (D)0 ∩ E).
In particular, λ0 (u) commutes with the elements of (1 − p2 )H0 (D) if u ∈
U(βh0 (D)0 ∩ E). It follows that λ0 (u) − 1 ∈ Der(A, J) ∩ p0 Ep0
If u satisfies moreover that βh0 (·) · u = βh0 then (1 − λ0 (u))H0 (·)(1 − p2 ) = 0
and (1 − λ0 (u)∗ )H0 (·)(1 − p2 ) = 0. It implies

1 − λ0 (u) ∈ N (A, J) ,

because H0 (D)p2 = s2 h0 (D)s∗2 ⊆ p2 Jp2 .

The element 1 − λ2 (u) is in p2 N (A, J)p2 for every u ∈ U(E) because p2 =
s2 s∗2 ∈ N (A, J) implies s2 Es∗2 = p2 N (A, J)p2 .
If only u ∈ U(βh0 (D)0 ∩ E) we see at least that the unitaries 1 − λ0 (u) and
1 − λ2 (u) are in Der(A, J) ∩ p0 Ep0 , because they commute with the elements in
(1 − p2 )H0 (D) and p2 H0 = s2 h0 s∗2 takes values in J.
The map λ0 satisfies λ0 (1) = 1 and is a norm-continuous group homomorphism
from U(βh0 (D)0 ∩ E) into the intersection of U(Der(A, J)) with p0 Ep0 + (1 − p0 ).
In particular, λ0 maps U0 (βh0 (D)0 ∩ E) into the intersection of U0 (Der(A, J)) and
p0 U0 (E)p0 + (1 − p0 ).
558 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Since p2 H0 = s2 h0 s∗2 has image in J we get that p2 Ep2 = s2 Es∗2 ⊂ N (A, J) ⊆

Der(A, J), and that λ2 maps U(E) into the unitaries of p2 Ep2 + (1 − p2 ). Thus,
λ2 (U0 (E)) is contained in U0 (N (A, J) + C · 1) ⊆ U0 (Der(A, J)).
Notice that p2 Ep2 + (p0 − p2 ) ⊂ p0 Ep0 .

Let B a σ-unital hereditary C *-subalgebra of J, and h : B ,→ J the inclusion

map h(b) = b.
By Lemma 4.4.7(x), there exists a unitary uh ∈ U0 (βh0 (D)0 ∩ E) with
uh βh0 (·) = βh0 = βh0 (·)uh and buh = h(b)uh = h(b)s∗2 s0 = bs∗2 s0 for b ∈ B .
Thus u∗h buh = s∗0 s2 bs∗2 s0 for b ∈ B, and we get that wB := uh has the claimed
additional properties, i.e., bwB = bs∗2 s0 for b ∈ B, wB ∈ U0 (βh0 (D)0 ∩ E) and
βh0 (·)wB = βh0 . Then W := λ0 (wB ) ∈ U0 (Der(A, J)), (1 − W ) ∈ N (A, J) ∩ p0 Ep0
and (k ⊕s2 ,t2 H0 )W = W (k ⊕s0 ,t0 H0 ) for each k : D → J with k(D) ⊆ B:
By definition of wB , we have bwB = bs∗2 s0 for b ∈ B and wB ∈ U0 (βh0 (D)0 ∩ E)
and βh0 (·)wB = βh0 .
Let W := λ0 (wB ) = wB ⊕s0 ,t0 1. Then W ∈ U0 (Der(A, J)), (1 − W ) ∈
N (A, J) ∩ p0 Ep0 and
s∗0 (b ⊕s0 ,t0 0)W = bs∗0 s0 wB s∗0 = bwB s∗0 = bs∗2 s0 s∗0 = bs∗2
for b ∈ B, because s0 s∗0 = p0 and s2 s∗2 ≤ p0 by Lemma 4.4.7(x).
We combine this with
p0 · (bj ⊕s0 ,t0 0) = bj ⊕s0 ,t0 0 ,
for j ∈ {1, 2}, and obtain for b := b∗2 b1 that
W ∗ (b ⊕s0 ,t0 0)W = b ⊕s2 ,t2 0 .
Since λ0 (wB )(D) commutes element-wise with elements in the image of the map
s0 βh0 s∗0 + t0 H0 t∗0 = t2 H0 t∗2 , we get for C *-morphisms k : D → E with the property


k(D) ⊆ B ⊆ J that W (k ⊕s2 ,t2 H0 ) = (k + βh0 ) ⊕s0 ,t0 H0 W . 

Lemma 4.4.11. Let A := H0 (D) and u ∈ U(Der(A, J + βJ)).

There exist v ∈ U0 (F + C · 1) ⊆ U0 (C) and C*-morphisms k1 , k2 : D → J with
(1 − v) ∈ F ⊂ C ∩ N (A, J + βJ) , [k1 ], [k2 ] ∈ S(h0 ; D, J) and
(k1 + βk2 ) ⊕s0 ,t0 H0 = vu∗ H0 (·)uv ∗ .

If moreover u ∈ U(Der(A, J)), then necessarily k2 = h0 , and there exists W ∈

U0 (Der(A, J)) such that (1 − W ) ∈ N (A, J) ∩ p0 Ep0 and
k1 ⊕s2 ,t2 H0 = W ∗ (k1 + βh0 ) ⊕s0 ,t0 H0 W = W ∗ vu∗ H0 (·)uv ∗ W .

Proof. Let u ∈ U(Der(A, J + βJ)). The image of the corresponding H0 -

derivation
δu : d ∈ D 7→ uH0 (d) − H0 (d)u
is contained in a separable C *-subalgebra B ⊂ J +βJ. Let f ∈ B a strictly positive
contraction in B.
 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 559

By assumptions (iii) of Theorem 4.4.6, there exists a projection p ∈ F ⊂ C

with p ≥ p0 , pf = f and p 6= p0 . It follows that pb = b = bp for all b ∈ B.
In particular, u∗ δu (d) = u∗ δu (d)p for d ∈ D. It implies for d ∈ D that

u∗ H0 (d)up = H0 (d)p − u∗ (uH0 (d) − H0 (d)u) ∈ J + βJ

because u ∈ Der(A, J + βJ), F · H0 (D) ⊂ J + βJ and F ⊆ C by assumptions (iii)

of Theorem 4.4.6.
Moreover, (1 − p)δu (d) = 0 and δu (d)(1 − p) = 0 imply (1 − p)uH0 (d) =
(1 − p)H0 (d)u, and uH0 (d)(1 − p) = H0 (d)u(1 − p).
Thus, H0 (d)(1 − p) = u∗ H0 (d)u(1 − p) and u∗ H0 (d)up ∈ J + βJ for all d ∈ D.
It follows that 1 − p and p commute with the elements of u∗ H0 (D)u and that

k : D 3 d 7→ u∗ H0 (d)up

is a C *-morphism from D into J + βJ.

Part (ix) of Lemma 4.4.7 provides a unitary v ∈ (1 + F ) ∩ U0 (F + C · 1) with
∗
vpv = p0 .
Since F ⊂ C by assumption (iii) of Theorem 4.4.6, we get that

t0 H0 (d)t∗0 = H0 (d)(1 − p0 ) = (uv ∗ )∗ H0 (d)uv ∗ (1 − p0 ) .

In particular, 1 − p0 and p0 commute element-wise with (uv ∗ )∗ H0 (D)uv ∗ . Recall

that pv ∗ = v ∗ p0 . We use that p0 = s0 s∗0 by assumption (iv) of Theorem 4.4.6 and
define a C *-morphism h : D → E by

h(d) := s∗0 (uv ∗ )∗ H0 (d)uv ∗ s0 = (v ∗ s0 )∗ k(d)(v ∗ s0 ) .

Then h(D) ⊆ J + βJ and

h ⊕s0 ,t0 H0 = (uv ∗ )∗ H0 (·)uv ∗ .

Hence, h = T ∗ H0 (·)T for the isometry T := uv ∗ s0 and [h] ∈ S(H0 ; D, E).

Moreover, h(D) ⊆ J + βJ because k(D) ⊆ J + βJ. Thus Lemma 4.4.7(vii)
applies to h and shows that there are unique C *-morphisms k1 , k2 : D → J with
[k1 ], [k2 ] ∈ S(h0 ; D, E) and h = k1 + βk2 .
Finally, we obtain that

(k1 + βk2 ) ⊕s0 ,t0 H0 = v(u∗ H0 (·)u)v ∗ .

Suppose now that u ∈ U(Der(A, J)). Then, uv ∗ ∈ U(Der(A, J)) for v ∈ U(C),
because C ⊆ Der(A, J), and, for d ∈ D holds

s0 k1 (d) − h0 (d) + β(k2 (d) − h0 (d)) s∗0 = v ∗ u∗ H0 (d)uv − H0 (d) ,

(4.15)

because 0 ⊕s0 ,t0 H0 = (1 − p0 )H0 (·) and p0 H0 = s0 (h0 + βh0 )s∗0 .

The element s0 (k1 (d) − h0 (d))s∗0 and the right side of Equation (4.15) are in J,
but the element s0 β(k2 (d) − h0 (d))s∗0 is in βJ for all d ∈ D. Thus, k2 = h0 by the
assumption J ∩ βJ = {0} in Theorem 4.4.6.
560 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Suppose now that k2 = h0 . By Lemma 4.4.10 there exists W ∈ U0 (Der(A, J))

such that W ∈ p0 Ep0 + (1 − p0 ) and

W (k1 ⊕s2 ,t2 H0 ) = ((k1 + βh0 ) ⊕s0 ,t0 H0 )W .

Hence, k1 ⊕s2 ,t2 H0 = W ∗ vu∗ H0 (·)uv ∗ W . 

We do not suppose that k1 , k2 ∈ S(h0 ; D, E) in the following lemma.

Lemma 4.4.12. Let A := H0 (D) and k1 , k2 : D → J C*-morphisms – not

necessarily in S(H0 ; D, E). The following are equivalent:

(i) [k1 ] + [h0 ] = [k2 ] + [h0 ] in the semi-group [Hom(D, J)].

(ii) There exists U0 ∈ U N (A, J + βJ) + C · 1 such that



(k1 + βh0 ) ⊕s0 ,t0 H0 U0 = U0 (k2 + βh0 ) ⊕s0 ,t0 H0 . (4.16)



(iii) There exists U1 ∈ U N (A, J) + C · 1 such that U1 ∈ U0 Der(A, J) and



(k1 + βh0 ) ⊕s0 ,t0 H0 U1 = U1 (k2 + βh0 ) ⊕s0 ,t0 H0 . (4.17)


(iv) There exists U2 ∈ U0 N (A, J) + C · 1 such that

(k1 ⊕s2 ,t2 H0 )U2 = U2 (k2 ⊕s2 ,t2 H0 ) . (4.18)

In particular, a C*-morphism k : D → E satisfies [k] + [h0 ] = [h0 ] in

[Hom(D, E)], if and only if, k(D) ⊆ J and there exists u ∈ U N (A, J + βJ) + C · 1
and v ∈ U(C) such that

(k + βh0 ) ⊕s0 ,t0 H0 = vu∗ H0 (·)uv ∗ .

If [k0 ] + [h0 ] = [h0 ] then there exists U3 ∈ U0 Der(A, J) with (1 − U3 ) ∈ N (A, J)
and
(k + βh0 ) ⊕s0 ,t0 H0 = U3∗ H0 (·)U3 .

Proof. Let A := H0 (D). Part (iii) implies Part (ii) with U0 := U1 because
N (A, J) ⊆ N (A, J + βJ).
(iv)⇒(iii): If we apply Lemma 4.4.10 to the separable C *-subalgebra B of J
that is generated by k1 (D) ∪ k2 (D), then we get a unitary W ∈ U0 (Der(A, J)) such
that for i ∈ {0, 1} holds



(ki + βh0 ) ⊕s0 ,t0 H0 W = W ki ⊕s2 ,t2 H0 .

Let U1 := W U2 W ∗ . The C *-algebra N (A, J) is a closed ideal of the C *-subalgebra

Der(A, J) ⊂ E with 1E ∈ Der(A, J). It implies that U0 N (A, J) + C · 1 is a normal



subgroup of U Der(A, J) . Thus U2 ∈ U0 N (A, J) + C · 1 . This implies together



with W ∈ U Der(A, J) that U1 ∈ U0 N (A, J) + C · 1 .
Straight calculation shows that U1 fulfills the equation (4.17).


(ii)⇒(i): Suppose that U0 ∈ U N (H0 (D), J + βJ) + C · 1 satisfies Equation
(4.16).
 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 561

By Lemma 4.4.9 with U := U0 and q := p0 , there exists a projection p ∈ F

with p ≥ p0 and p 6= p0 such that U0∗ H0 (·)(1 − p) U0 = H0 (·)(1 − p). If we use

that t0 H0 t∗0 = (1 − p0 )H0 we get:

U0∗ s0 (k1 + βh0 )s∗0 + (p − p0 )H0 (·) U0 = s0 (k2 + βh0 )s∗0 + (p − p0 )H0 (·)

By Part (xii) of Lemma 4.4.7 there exists unitaries V ∈ U(C) and W ∈ U0 (E) such
that W ∗ (h0 + βh0 )W = p0 H0 , V p0 = p0 = p0 V and

V ∗ (p − p0 )H0 V = t0 W ∗ (h0 + βh0 )W t∗0 .

We define a unitaries

V0 := s0 s∗ + t0 W ∗ t∗ and V1 := V0∗ V ∗ U0 V V0 .

Straight calculation, using assumptions (i) of Theorem 4.4.6, shows that

V1∗ (k1 ⊕s,t h0 ) + β(h0 ) V1 = (k2 ⊕s,t h0 ) + β(h0 ) .

If we use that J · βJ = {0} then we get finally

V1∗ (k1 ⊕s,t h0 )V1 = (k2 ⊕s,t h0 ) ,

that yields [k1 ] + [h0 ] = [k2 ] + [h0 ].

(i)⇒(iv): Suppose that [k1 ]+[h0 ] = [k2 ]+[h0 ]. Let hi := ki ⊕s,t h0 for i ∈ {1, 2}.
Then [h1 ] = [h2 ] in [Hom(D, J)] and there exists v ∈ U(E) with v ∗ h1 v = h2 .
Since hi (D) ⊆ J, we can find even a unitary u ∈ U0 (E) with u∗ h1 u = h2 by
Lemma 4.4.7(viii). We define a unitary W := u ⊕s2 ,t2 1 in s2 U0 (E)s∗2 + (1 − p2 ).
Then 1 − W ∈ N (A, J) and W is contained in U0 (N (A, J) + C · 1) ⊂ U0 (Der(A, J)),
because p2 H0 (·) = s2 h0 s∗2 by Lemma 4.4.7(x) and because x∗ H0 (D) ⊂ J implies
xEx∗ ⊆ N (A, J). Obviously, W ∗ (h1 ⊕s2 ,t2 H0 )W = h2 ⊕s2 ,t2 H0 .
By Lemma 4.4.8, the element u2 := s2 ss∗2 + s2 ts∗2 t∗2 + t2 (t22 )∗ from equation
(4.9) is in U(C) and satisfies



(a ⊕s,t b) ⊕s2 ,t2 c) u2 = u2 a ⊕s2 ,t2 (b ⊕s2 ,t2 c) for a, b, c ∈ E .

Part (x) of Lemma 4.4.7 shows that h0 ⊕s2 ,t2 H0 = p2 H0 + (1 − p2 )H0 = H0 . Thus,
ki ⊕s2 ,t2 H0 = u∗2 (hi ⊕s2 ,t2 H0 )u2 for i ∈ {1, 2}, and with U2 := u∗2 W u2 we get

U2∗ (k1 ⊕s2 ,t2 H0 )U2 = k2 ⊕s2 ,t2 H0 .

The algebra N (A, J) is an ideal of Der(A, J). This implies that the group
U0 (N (A, J) + C · 1) is a normal subgroup of U(Der(A, J)) that is contained in
U0 (Der(A, J)).
If we use that W is contained in U0 (N (A, J) + C · 1) and C ⊆ Der(A, J) then
we can see that the above defined U2 is an element of U0 (N (A, J) + C · 1).
Moreover (1 − W )H0 (·)(1 − p2 ) = 0 implies that also (1 − W ∗ )H0 (·)(1 − p2 ) = 0,
and thus (1 − W ) ∈ N (A, J). It follows (1 − U2 ) ∈ N (A, J).

Finally we consider the special case k1 := k and k2 := h0 :

562 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

If a C *-morphism k : D → E satisfies [k] + [h0 ] = [h0 ] in [Hom(D, E)], then

there exists V ∈ U(E) with k ⊕s,t h0 = V ∗ h0 (·)V . Thus k = T ∗ h0 (·)T for the
isometry T := V s, i.e., [k] ∈ S(h0 ; D, E) and k(D) ⊆ J.
Recall that [h0 ] = [h0 ] + [h0 ] and H0 = (h0 + βh0 ) ⊕s0 ,t0 H0 by assumptions
(i,iv) of Theorem 4.4.6.
Therefore, the equivalence of Parts (i), (ii) and (iii) shows that [k] + [h0 ] = [h0 ]


is equivalent to k(D) ⊆ J and the existence of U0 ∈ U N (A, J + βJ) + C · 1 ,



respectively of U1 ∈ U N (A, J) + C · 1 with U1 ∈ U0 Der(A, J) , such that, for
j ∈ {0, 1}:
(k + βh0 ) ⊕s0 ,t0 H0 = Uj∗ H0 Uj .
(Notice that necessarily U0∗ U1 ∈ C.)
We find ξ ∈ C with |ξ| = 1 such that 1 − ξU1 N (A, J) and can replace U1 by
ξU1 in Part (iii).


If u ∈ U N (A, J + βJ) + C · 1 and v ∈ U(C) are given such that

(k + βh0 ) ⊕s0 ,t0 H0 = vu∗ H0 (·)uv ∗ ,

then we can take U0 := vuv ∗ to satisfy Formula (4.16) in Part (ii). 

Next Lemma 4.4.13 true? To be checked: ??

But it could work with kind of “stabilization”?

Lemma 4.4.13. Let A := H0 (D) ⊆ E, C := A0 ∩ E and I := J + βJ as in

Theorem 4.4.6.
For each separable C*-subalgebra B ⊆ N (A, I) ⊂ Der(A, I) there exist u ∈
(F + 1) ∩ U0 (F + C · 1) ⊆ U0 (C) and an isometry T ∈ E such that

u∗ bu − s0 T ∗ bT s∗0 ∈ Ann(A, E) for all b ∈ B .

In particular, for each contraction e ∈ 1 + N (A, I) with 1 − e∗ e ∈ Ann(A) and

1 − ee∗ ∈ Ann(A) there exists a unitary in u ∈ N (A, I) + 1 with u − e ∈ Ann(A).

We do not know if this also holds with I := J / E (replacing J + βJ) and

u bu − s2 T ∗ bT s∗2 ∈ Ann(A, E) for b ∈ B (in place of the known weak conclusion
∗

u∗ bu − s0 T ∗ bT s∗0 ∈ Ann(A, E)).

Is it used or needed somewhere?
It needs certainly a more elaborate proof if it is true.

Proof. In other words, we have to show that for each separable C *-subalgebra
B ⊆ E with the property B · H0 (D) ⊆ I := J + βJ there exists an isometry T ∈ E
and a unitary u ∈ 1+F with u ∈ U0 (F +C·1) such that (u∗ bu−s0 T ∗ bT s∗0 )·H0 (D) =
{0} for every b ∈ B.
Let B ⊆ N (A, I) a separable C *-subalgebra and b0 ∈ B a strictly positive
contraction in B.
 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 563

Then let e ∈ G+ ⊆ I+ a strictly positive contraction for the separable C *-sub-

algebra G of I generated by B · A (recall A := H0 (D)).
Notice that the linear span of all elements in (BA)n , (AB)n , A(BA)n =
(ABBA)n and B(AB)n = (BAAB)n for n = 1, 2, . . . is dense in G. It follows
that BG ∪ GB ⊆ G and AG ∪ GA ⊆ G, i.e., that A ∪ B ⊂ N (G, E).
The elements x := b1 a1 + a2 b2 with a1 , a2 ∈ A and b1 , b2 ∈ B build a set of
generators of G, that is invariant under involution.
Check next:
b0 x, a0 x ∈ G for b0 ∈ B and the linear span
AB ∪ BA ∪ AB 2 A ∪ BA2 B ∪ (AB)n ∪ (BA)n ∪ A(BA)n generates C ∗ (A ∪ B)
????
Let fn ∈ I+ a sequence of positive contractions such that fn fn+1 = fn ,
limn fn e = e = limn efn , limn k(1 − fn )b(1 − fn )ak = 0, limn k[b, fn ]k = 0 and
limn k[a, fn ]k = 0 for b ∈ B and a ∈ A
????
Let g := n 2−n fn . There exists a projection p ∈ F ⊆ C with p ≥ p0 and
P

p 6= p0 such that pg = g = gp by assumption (iii,c) of Theorem 4.4.6.

(And there exist projections q, r ∈ E, but not in F , with p = q + r and
q, β(r) ∈ N (A, J) by Lemma 4.4.7(xi).)
Write below new proof
fitting to above!!
If BA ⊆ J, then e, fn , g ∈ J and
??? rg, gr ∈ Ann(A, E),
because Ar = 0 = rA and rgA = 0 ??? Using an with respect to C ∗ (A, r)
approximately central approximate unit (cτ ) ∈ J+ of J, we get cτ g → g ...
Thus, (1 − p)BA = {0} and BA(1 − p) = B(1 − p)A = {0}. It follows also that
AB(1 − p) = {0} and A(1 − p)B = {0}. Thus (b − pbp) ∈ Ann(A) for all b ∈ B,
because b − pbp = b(1 − p) + (1 − p)bp.
Moreover pb∗ (1 − p)cp ∈ Ann(A) for all b, c ∈ B, because (pb∗ (1 − p)cp)∗ =
(pc (1 − p)bp) and (pb∗ (1 − p)cp)∗ (pb∗ (1 − p)cp)A = {0}. We get pBp ⊆ N (A, I)
∗

because p ∈ C ⊆ Der(A, I) and N (A, I) is an ideal of Der(A, I).

There exists u ∈ (1 + F ) ∩ U0 (F + C · 1) with u∗ pu = p0 .
Let γ0 (b) := p0 u∗ bup0 . Then u∗ γ0 (b)u = pbp.
Is b − qbq = (b − up2 u∗ bup2 u∗ )H0 (d) = 0 for all b ∈ B? Likely not! More
general open:
Is there in case I = J a decomposition
u bu − s2 γ(b)s∗2 ∈ Ann(A, E)???
∗
564 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Let T := us0 and γ(b) := T ∗ bT . Then pbp = us0 γ(b)s∗0 u∗ . Thus u∗ bu −

s0 γ(b)s∗0 ∈ Ann(A) for all b ∈ B.
H0 (d)T = us0 s∗0 H0 (d)s0 = us0 (h0 (d) + βh0 (d)).
There exist u ∈ (F + 1) ∩ U0 (F + C · 1) ⊆ U0 (C) and an isometry T ∈ E
such that, for all b ∈ B, u∗ bu − s2 T ∗ bT s∗2 ∈ Ann(A) in case I = J (respectively
u∗ bu − s0 T ∗ bT s∗0 ∈ Ann(A) in case I = J + βJ):
To be filled in
There exists an isometry T ∈ E with T ∗ H0 (·)T = 0 and T T ∗ + p0 ≤ 1 :
By Lemma 4.4.7(vi), there exists an isometry s1 ∈ E with s1 s∗1 ≤ 1 − p0 and
s1 s∗1 H0 (·) = 0. Thus T := s1 is as desired. (One could also use directly T := R for
R in the proof of Lemma 4.4.7(vi).)
To be filled in. Check. ?? The “naive” definition should be the following
definition:
Let U (e) the Halmos unitary of e defined in Remark 4.2.4 and s1 , t1 ∈ E the
isometries from Lemma 4.4.7(vi), and define

u := [t1 , s1 ]U (e)[t1 , s1 ]∗ = t1 et∗1 + s1 e∗ s∗1 − t1 (1 − ee∗ )1/2 s∗1 + s1 (1 − e∗ e)1/2 t∗1

Then s1 e∗ s∗1 , t1 (1 − ee∗ )1/2 s∗1 , s1 (1 − e∗ e)1/2 t∗1 ∈ Ann(A) and t1 a = a = at1 = t∗1 a
for all a ∈ A.
Thus ua = t1 et∗1 a = t1 ea and au = aet∗1 for a ∈ A. Need modification of s1 , t1
such that t1 ea = ea for a ∈ A and aet∗1 = ae for a ∈ A.
Notice that ea − ae = g is contained in a separable C *-subalgebra G of I,
t1 ea = t1 (g + ae) = t1 g + ae = ea + (t1 − 1)g . aet∗1 = (t1 ea)∗ = ((t1 − 1)g)∗ + ae
for suitable g ∈ G
Thus we have to find a unitary V in C := A0 ∩ E such that s∗1 V g = 0 for all
g ∈ G. Then still V ∗ s1 s∗1 V ≤ 1 − p0 and V ∗ s1 s∗1 V a = V ∗ s1 s∗1 aV = 0 for a ∈ A.
We can rename V ∗ s1 by R and use it to construct a suitable correction of s1 , t1 :
To find V ∈ F , we take q ∈ F with q ≥ p0 , q 6= p0 and qg = g = gq for all
g ∈ G (is possible by Parts (iii,iv) of Theorem 4.4.6). Then s1 V has the property
V ∗ s1 s∗1 V ≤ (1 − q) = V ∗ (1 − p0 )V ≤ (1 − p0 ) .
Let R := V ∗ s1 . Then R∗ A = {0} and R∗ G = {0} for V ∈ U0 (F ) with
∗
V p0 V = q as defined above. Since E is properly infinite there exists isometries S, T
with SS ∗ ≤ RR∗ , SS ∗ + T T ∗ = 1 and T a = a = aT = aT ∗ , T g = g = gT = gT ∗
for a ∈ A and g ∈ G.
Now define u := [T, S]U (e)[T, S]∗ . Then we get that (e−u)A = {0} = A(e−u).

A construction of u in the special case that e := s2 f s∗2 + 1 − p2 in case I = J

(respectively e := s0 f s∗0 +1−p0 in case I = J+βJ) where f ∈ E is a contraction such
that 2p2 − s2 f ∗ f s∗2 − s2 f f ∗ s∗2 ∈ Ann(A, E) (respectively 2p0 − s0 f ∗ f s∗0 − s0 f f ∗ s∗0 ∈
Ann(A, E)).
 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 565

Let Z := [s2 , T, 1 − p2 − T T ∗ ] ∈ M1,3 (E) if I = J (respectively let Z :=

[s0 , T, 1 − p0 − T T ∗ ] if I = J + βJ) and U (f ) ∈ M2 (E) the Halmos unitary of f as
defined in Remark 4.2.4.
Then ZZ ∗ = 1 and u := Z(U (f )⊕1)Z ∗ is a unitary that satisfies u ∈ N (A, I)+1
and u − e ∈ Ann(A, E). 

Remark 4.4.14. Let A := H0 (D), I := J + βJ and u ∈ U(Der(A, I)). By

Lemma 4.4.11 there exist v ∈ U(C) and a C *-morphism k : D → I, i.e., k = k1 +βk2
with C *-morphisms k1 , k2 : D → J, that satisfy

k ⊕s0 ,t0 H0 = v ∗ u∗ H0 (·)uv .

The following are equivalent:

(i) [k] + [h0 + βh0 ] = [h0 + βh0 ] .

(ii) There are w1 , w2 ∈ U(C) such that 1 − w1 uw2 ∈ N (A, I).
(iii) There exist w ∈ U(C) such that wuv ∈ U0 (N (A, I) + C · 1).

Proof. The proof of the equivalences in Parts (i)–(iii) of Remark 4.4.14 is

similar to the proof of Lemma 4.4.12 with ⊕s2 ,t2 and V := u2 replaced by ⊕s0 ,t0
and V := u0 in Lemma 4.4.8 and in proof of Lemma 4.4.7(x). 

Let C, Ann(A, E), Der(A, I), N (A, I), µ : K1 (C) → K1 (Der(A, I)) and


Γ A, I, E as introduced in Definitions 4.4.4 and 4.4.5.
Notice that Der(A, I0 ) ⊆ Der(A, I) if I0 ⊆ I for an ideal I0 / E. In particular,
C := A0 ∩ E = Der(A, {0}) ⊆ Der(A, I).
Recall that N (A, I) is an ideal of Der(A, I) and that N (A, I0 ) ⊆ N (A, I) if
I0 ⊆ I. In particular, Ann(A, E) := N (A, {0}) ⊆ C ∩ N (A, I).
We use below the group morphisms

µ : K1 (C) → K1 (Der(A, I)) and µ0 : K1 (N (A, I)) → K1 (Der(A, I))

given by the inclusion maps C ,→ Der(A, I) and N (A, I) ,→ Der(A, I). Recall that

Γ(A, J, E) := K1 (Der(A, J))/µ(K1 (C)) .

Lemma 4.4.15. Let A := H0 (D), C := A0 ∩ E and I := J + βJ (respectively

I := J). Let Ann(A, E) := N (A, {0}), Der(A, I) and N (A, I) as in Definition
4.4.4.

(i) N (A, I) is a closed ideal of Der(A, I) and a full hereditary C*-subalgebra

of E. It holds
Der(A, I) = C + N (A, I) .
In particular there is a natural isomorphism

C/(C ∩ N (A, I)) ∼

= Der(A, I)/ N (A, I) ,
566 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

(ii) The annihilator Ann(A, E) of A in E is a closed ideal of C and is a full

hereditary C*-subalgebra of E.
It satisfies

Ann(A, E) = N (A, J) ∩ β(N (A, J)) ⊆ N (A, I) .

In particular, the quotient maps π : C → C/ Ann(A, E) and

Der(A, I) → C/(C ∩ N (A, I)) induce surjective maps on the K∗ -groups,
e.g. [π]1 (K1 (C)) = K1 (C/ Ann(A, E)) :
For every contraction c ∈ C with 1 − c∗ c, 1 − cc∗ ∈ Ann(A, E) there
exists a unitary u ∈ C such that

1 − u(c ⊕s,t 1) ∈ Ann(A, E) .

(iii) The C*-algebras Der(A, I), C, N (A, I)+C·1 and all its non-zero quotients
and unifications of closed non-zero ideals are K1 -surjective C*-algebras.
(iv) U( N (A, I)+C·1 ) is a closed normal subgroup of U(Der(A, I)) and U(C)
is a closed subgroup of U(Der(A, I)) .
(v) Let G(I) denote the set U(C) · U(N (A, I) + C · 1) of products u · v with
u ∈ U(C) and v ∈ U(N (A, I) + C · 1) .
The set G(I) is an open subgroup of U(Der(A, I)) that satisfies

G(I) ⊕s,t 1 ⊆ G(I) .

In particular, U0 (Der(A, I)) ⊆ G(I) .

(vi) For each u ∈ U(E) there exists a unitary v ∈ 1 + Ann(A, E) ⊂ C such
that [v · u] = 0 in K1 (E).
For each unitary in u ∈ U(N (A, J + βJ) + C · 1) with the additional
property u ∈ Der(A, J) there exists unitaries v, w ∈ U(C) such that

v u w ∈ U0 (N (A, J) + C · 1) ⊆ U0 (Der(A, J)) .

In particular, the image µ0 (K1 (N (A, J))) ⊆ K1 (Der(A, J)) of

K1 (N (A, J)) is contained in the image µ(K1 (C)) in K1 (Der(A, J)),
and the mapping

G(J) 3 u 7→ [u] ∈ K1 (Der(A, J))

maps G(J) onto µ(K1 (C)) ⊆ K1 (Der(A, J)).

(vii) The natural group morphisms K1 (C) → K1 (C/ Ann(A, E)) and



K1 Der(A, I) → K1 Der(A, I)/ N (A, I)

are surjective.
(viii) The natural group epimorphism ϕ : [k]+[h0 ] 7→ [k]+[H0 ] from G(h0 ; D, E)
onto G(H0 ; D, E) is injective, if and only if, the natural group morphism


K1 (C/ Ann(A, E)) → K1 C/(C ∩ N (A, J))

is surjective.
(ix) This map ϕ : [k] + [h0 ] 7→ [k] + [H0 ] is injective, if and only if,

u ⊕s,t 1 ∈ G(J) for all u ∈ U(Der(A, I)) ∩ U0 (E) . (4.19)

 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 567

(x) A sufficient condition for the injectivity of ϕ is that the natural group
morphism



U C/ Ann(A, E) → U C/(C ∩ N (A, I))


contains 1 ⊕ U C/(C ∩ N (A, I)) in its image.

Proof. Recall that A := H0 (D), and that we let alternatively I := J or

I := J + βJ to avoid repeats. We need sometimes additional considerations for the
case I := J.
(i): Clearly, Der(A, I) = πI−1 πI (A)0 ∩ (E/I) . Thus, Der(A, I) is a C *-

subalgebra of E and

C := A0 ∩ E = Der(A, {0}) ⊆ Der(A, I) .

In the same way N (A, I) = πI−1 (Ann(πI (A), E/I)), where Ann(πI (A), E/I) means
the two-sided annihilators in E/I of the elements of πI (A).
Since Ann(πI (A), E/I) is a closed ideal of πI (A)0 ∩ (E/I) , the C *-algebra
N (A, I) is a closed ideal of Der(A, I).
In particular, C + N (A, I) is a C *-subalgebra of Der(A, I) in general for all
closed ideals I of E.
Let e ∈ E+ with eH0 (d) − H0 (d)e ∈ I := J + βJ for all d ∈ D, i.e., e ∈
Der(A, I). Since D is separable, the image T (D) of T (d) := eH0 (d) − H0 (d)e is
contained in a separable C *-subalgebra B of J + βJ. Let f ∈ B+ a strictly positive
contraction for B. By assumption (iii) of Theorem 4.4.6, there is a projection p ∈ F
with pf = f = f p, p0 ≤ p and p0 6= p.
Since F ⊆ C by assumption (iii) of Theorem 4.4.6, it follows for d ∈ D that

0 = (1 − p)T (d) = (1 − p)eH0 (d) − H0 (d)(1 − p)e and 0 = T (d)(1 − p) .

Thus, (1 − p)e, pe(1 − p) ∈ C and e − pep = (1 − p)e + pe(1 − p) ∈ C. By

assumptions (iii,v) of Theorem 4.4.6, pH0 (D) = H0 (D)p ⊂ J + βJ. Thus, pep ∈
N (H0 (D), J + βJ).
It also shows that Der(A, I) = C + N (A, I) in the case where I = J + βJ:
We can use that Der(A, J) ⊂ Der(A, J + βJ) in the case where I = J:
If e ∈ Der(A, J)+ , then e ∈ Der(A, J + βJ) and there exists a projection p ∈ F
with e − pep ∈ C by our above considerations. Since C = Der(A, {0}) ⊂ Der(A, J)
and F ⊆ N (A, J + βJ) it follows that

pep = e − (e − pep) ∈ Der(A, J) ∩ N (A, J + βJ) .

It turns out that the latter implies that pep is in N (A, J) :

By Part (xi) of Lemma 4.4.7 there exist projections q, r ∈ C with the properties
p = q + r, [qH0 (·)] = [h0 ] and [rH0 (·)] = [βh0 ]. In particular q ∈ C ∩ N (A, J) and
r ∈ C ∩ N (A, βJ).
568 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

For d ∈ D holds rH0 (d) = H0 (d)r ∈ βJ. Thus,

β(J) 3 rerH0 (d) − H0 (d)rer = r(eH0 (d) − H0 (d)e)r ∈ J .

Since J ∩ βJ = {0} it follows that rer ∈ C. Hence, pep − rer ∈ Der(A, J) and

e − (pep − rer) = (e − pep) + rer ∈ C .

We show that the element pep − rer = qer + req + qeq is in N (A, J), and get that
e ∈ C + N (A, J):
Since q, r ∈ C, we get e.g. from qerH0 (d) − H0 (d)qer = q(eH0 (d) − H0 (d)e)r
and e ∈ Der(A, J)+ that qer, req, rer ∈ Der(A, J). It follows from J ∩ βJ = {0},
H0 (d)r ∈ βJ, qH0 (d) ∈ J and reqH0 (d)−H0 (d)req ∈ J that necessarily H0 (d)req =
0 for all d ∈ D. Since (qer)∗ = req ∈ Der(A, J) it follows qerH0 (d) = 0 for all
d ∈ D.
(re + qe)(qH0 (d)) ∈ J and qer H0 (d) = 0 imply (qer + req + qeq)H0 (d) ∈ J.
Since pep − rer is self-adjoint it follows that pep − rer ∈ N (A, J).
Summing up we get e = x+y with x = (e+rer)−(pep) ∈ C and y = pep−rer ∈
N (A, J).
Thus, Der(A, I)+ ⊆ C + N (A, I). Since C + N (A, I) is a C *-subalgebra of
the C *-algebra Der(A, I) it follows Der(A, I) = C + N (A, I) for A := H0 (D) and
I := J + βJ (respectively A := H0 (D) and I := J).
The ideal C ∩ N (A, I) of C is the kernel of the restriction to C of the epi-
morphism Der(A, I) → Der(A, I)/ N (A, I). Since Der(A, I) = C + N (A, I), this
quotient map defines an isomorphism

Der(A, I)/ N (A, I) ∼

= C/(C ∩ N (A, I)) .

(ii): The (two-sided) annihilator Ann(A, E) is a closed ideal of C, because

c1 ec2 a = c1 eac2 = 0 and ac1 ec2 = c1 aec2 = 0

if e ∈ Ann(A, E), a ∈ A and c1 , c2 ∈ C.

The C *-subalgebra Ann(A, E) is hereditary in E, because e1 Ee2 ⊆ Ann(A, E)
if e1 , e2 ∈ Ann(A, E).
If e ∈ Ann(A, E), i.e., if ea = 0 = ae for all a ∈ A ⊆ E, then e ∈ N (A, J) ⊆
N (A, I) and e ∈ C = A0 ∩ E. Thus Ann(A, E)
Since βH0 = H0 , it follows βA = A and β(Ann(A, E)) = Ann(A, E).
Thus, Ann(A, E) ⊆ N (A, J) ∩ β N (A, J).
We get β N (A, J) = N (A, βJ) from β(A) = A.
If e ∈ E satisfies eH0 (D) ⊆ J and eH0 (D) ⊆ βJ, then eH0 (D) ⊆ J ∩βJ = {0}.
Thus, N (A, J) ∩ β N (A, J) = Ann(A, E).
 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 569

The *-monomorphism H0 dominates zero by Lemma 4.4.7(vi), i.e., there exists

an isometry s1 ∈ E with s∗1 H0 (·)s1 = {0}. It implies that s1 s∗1 ∈ Ann(A, E) if
A := H0 (D), i.e., Ann(A, E) is full in E.
(iii): The unital algebra C := A0 ∩ E ⊆ Der(A, I) contains the isometries
s, t with orthogonal ranges by assumption (i) of Theorem 4.4.6, and N (A, I) is a
non-zero ideal of Der(A, I) by Part (i).
All non-zero quotients of C and of Der(A, I) and all (unitization of) non-zero
ideals of this quotients are K1 -surjective by Remark 4.2.17.
(iv): The groups U(N (A, I) + C · 1) and U(C) are norm-closed subgroups of
U(Der(A, I)), because unitary groups of C *-algebras are norm-closed.
The group U(N (A, I) + C · 1) is normal in U(Der(A, I)) because N (A, I) is an
ideal of Der(A, I) = C + N (A, I).
(v): The set G(I) of products U(C) · U(N (A, I) + C · 1) is a subgroup of
U(Der(A, I)) because U(N (A, I) + C · 1) is a normal subgroup of U(Der(A, I)) by
Part (iv).
If u ∈ U0 (Der(A, I)) then u is a product of exponentials, i.e., there exist tm ∈
Der(A, I) (m = 1, · · · , n) such that t∗m = −tm and u = exp(t1 )·. . .·exp(tn ). We find
cm ∈ C (m = 1, . . . , n) with c∗m = −cm and tm −cm ∈ N (A, I), because Der(A, I) =
C + N (A, I) by Part (i). If we define v ∈ U0 (C) by v := exp(c1 ) · . . . · exp(cn ), then
v ∗ u ∈ 1 + N (A, I), because U(N (A, I) + C · 1) is a normal subgroup of U(Der(A, I))
and exp(−cm ) exp(tm ) ∈ 1 + N (A, I) for m = 1, . . . , n.
It follows that G(I) is an open subgroup of U(Der(A, I)).
Recall that every open subgroup G2 of a topological group G1 is also closed
in G1 , because G2 is the complement in G1 of the union of open sets G2 · g with
g ∈ G1 \ G2 .
If u, v ∈ U(C) then u ⊕s,t v = sus∗ + tvt∗ ∈ U(C) because s, t ∈ C.
If w ∈ U(N (A, I) + C · 1) then w = (ξ1) · (e + 1) for some ξ ∈ C with |ξ| = 1
and e ∈ N (A, I).
The unitary s∗ ws+tt∗ = (ξ ·ss∗ +tt∗ )(ses∗ +1) is in G(I), because ξ ·ss∗ +tt∗ ∈
U0 (C) and ses∗ ∈ N (A, I).
Thus G(I) ⊕s,t 1 ⊆ G(I).
(vi): By Lemma 4.4.7(vi) there exist isometries s1 , t1 ∈ E with s1 s∗1 + t1 t∗1 = 1
and s1 satisfies s∗1 A = {0} = As1 . It implies that s1 E s∗1 ⊆ Ann(A, E). If u ∈ U(E)
then [u] = [u ⊕0 1] = [(1 − s1 s∗1 ) + s1 us∗1 ] in K1 (E) by Lemma 4.2.6(v,2).
Thus v := (1 − s1 s∗1 ) + s1 u∗ s∗1 is a unitary in 1 + Ann(A, E) with [vu] = 0 in
K1 (E).
Remaining TEXT:
570 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

For each unitary in u ∈ U(N (A, I) + C · 1) with the additional property u ∈

Der(A, J) there exists unitaries v, w ∈ U(C) such that

v u w ∈ U0 (N (A, J) + C · 1) ⊆ U0 (Der(A, J)) .

In particular, the image µ0 ( K1 (N (A, J)) ) ⊆ K1 ( Der(A, J) ) of K1 ( N (A, J) )

is contained in the image µ( K1 (C) ) in K1 ( Der(A, J) ), and the mapping

G(J) 3 u 7→ [u] ∈ K1 (Der(A, J))

maps G(J) onto µ(K1 (C)) ⊆ K1 (Der(A, J)).

??
(old ?) TEXT:
This ref part is only granted modulo Ann(A, I)
or only in K1 (N (A, I))?

Next text is different to ORIGINAL.

For each unitary in u ∈ U(N (A, I) + C · 1) (respectively u ∈ U(E)) there exists
a unitary v ∈ U(C) ∩ (1 + N (A, I)) (respectively only v ∈ U(C))
OR ?: v ∈ 1 + Ann(A, E) ???
such that v u ∈ U0 (N (A, I) + C · 1)
Respectively only (minimal possible): vu ∈ U0 (Der(A, I)) · U(C).
(respectively v ∈ U(C) such that vu ∈ U0 (E) and a unitary w ∈ 1 +
Ann(A, E) ⊆ C with [wu] = 0 in K1 (E)).
In particular, the image µ0 (K1 (N (A, I))) in K1 (Der(A, I)) is contained in the
image µ(K1 (C)) of K1 (C) in K1 (Der(A, I)), and the image of the map

G(I) 3 u 7→ [u] ∈ K1 (Der(A, I))

coincides with µ(K1 (C)) ⊆ K1 (Der(A, I)).

END TEXT
Use in next proof lem. 4.4.7(vi)? in place of 4.4.7(iv)??
By Lemma 4.4.7(vi), there exist isometries s1 , t1 ∈ E with s1 s∗1 + t1 t∗1 = 1 and
s∗1 H0 (D)= {0}, i.e., s1 Es∗1 ⊆ Ann(A, E).
CASE u ∈ U(E) :
Let u ∈ U(E). Define v := (u⊕R,T 1)∗ ∈ Ann(A, E). Then v ∈ C and [v] = −[u]
in K1 (E), i.e., [vu] = 0 in K1 (E) and (vu)∗ H0 (·)vu = u∗ H0 (·)u.
NEXT ??????
There exists a unitary v ∈ U(C) ∩ (1 + Ann(A, I)) such that vu ∈ U0 (N (A, I) +
C · 1).
????
(vi): New approach:
 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 571

Next to PROOF of a Lemma ??:

Let u ∈ U(N (A, I) + C · 1). Then u ∈ Der(A, I) and by Lemma 4.4.11 there
exist k : D → I and v ∈ U(C) such that

k ⊕s0 ,t0 H0 = v ∗ u∗ H0 (·)uv ,

where k = k1 + βk2 for C *-morphisms k1 , k2 : D → J. Moreover, k2 = h0 in case

I = J.
Lemma 4.4.8 (in case I = J) and Remark 4.4.14 (in case I = J + βJ) show
that v ∗ uv ∈ U(N (A, I) + C · 1) implies that [k] + [h0 + βh0 ] = [h0 + βh0 ] (which is
equivalent to [k1 ]+[h0 ] = [h0 ] in case I = J), that this is equivalent to the existence
of a unitary U ∈ U0 (N (A, I) + C · 1) with

k ⊕s0 ,t0 H0 = U ∗ H0 (·)U .

It follows that w := uv · U ∗ ∈ U(C), and w∗ uv = U ∗ ∈ U0 (N (A, I) + C · 1).

(vii): Clearly Ann(A, E) is a closed ideal of C and is a hereditary C *-subalgebra
Ann(A, E) of E. It is full in E because there exist an isometry s1 ∈ E with
s∗1 H0 (·)s1 = 0 by Lemma 4.4.7(vi).
Since N (A, I) = πJ−1 (Ann(πJ (A), E/J)), it is a hereditary C *-subalgebra of
E that contains Ann(A, E). Thus N (A, I) is full in E. By Part (i), N (A, I) is a
closed ideal of Der(A, I) that is hereditary and full in E.
Since Ann(A, E) is a closed ideal of C and N (A, E) is a closed ideal of
Der(A, E), and both are full and hereditary in E, Part (vii) follows from Lemma
4.2.20(o).
END Proofs (i)-(vii).
TEXT OF GENERAL CONCLUSIONS:
Finally, the K∗ -theory 6-term exact sequence

0 → (C ∩ N (A, I)) → C → C/(C ∩ N (A, I)) → 0 .

shows that the group morhism K1 (C) → K1 (C/(C ∩ N (A, I))) is surjective, if and
only if, K0 (C ∩ N (A, I))) → K0 (C) is injective.
The latter is related to the injectivity of ϕ : [k] + [h0 ] 7→ [k] + [H0 ].
???????
END TEXT
BEGIN TEXT OF GENERAL Conclusion:
The group morphism ϕ : [k] + [h0 ] 7→ [k] + [H0 ] is injective, if and only if, the
natural group morphism



K1 C/ Ann(A, E) → K1 C/(C ∩ N (A, I))

is surjective.
This is the case if and only if u ⊕s,t 1 ∈ G for all u ∈ U(Der(A, I)).
572 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

A sufficient condition for the latter is that the natural group morphism



U C/ Ann(A, E) → U C/(C ∩ N (A, I))

is surjective.
Notice that Ann(A, E) ⊆ C ∩ N (A, I) are ideals of C and that Ann(A, I) is a
full hereditary C *-subalgebra of E. By Lemma 4.2.20, K∗ (C/ Ann(A, I)) is natural
isomorphic to the kernel of K∗ (C) → K∗ (E), and the natural sequences

K∗ (Ann(A, I)) → K∗ (C) → K∗ (C/ Ann(A, I))

are split-exact sequences.

Recall that G(H0 ; D, E) (respectively G(h0 ; D, E)) is isomorphic to the kernel
of K0 (C) → K0 (E) (respectively of K0 (h0 (D)0 ∩ E) → K0 (E)).


Proof of Theorem 4.4.6 (Part 2: kernel of ϕ in G(h0 , D, E)).

Let A (temporary) denote the set of k ∈ Hom(D, E) with k(D) ⊆ J and the
property that k ⊕ H0 is unitary equivalent to H0 by a unitary in E, i.e., with
[k] + [H0 ] = [H0 ] in [Hom(D, E)]u . Recall that [k] + [H0 ] is the unitary equivalence
class of k ⊕H0 in the elements in Hom(D, E), which is independent from the chosen
isometries S, T ∈ E with SS ∗ + T T ∗ = 1 for the sum ⊕ := ⊕S,T .
If k ∈ Hom(D, E) is in A, i.e., k(D) ⊆ J and and [k] + [H0 ] = [H0 ], then k
satisfies

(1a) [k] ∈ S(h0 ; D, E), (by Lemma 4.4.7(vii)), and

(1b) [k] + [h0 ] is in the kernel of ϕ : G(h0 ; D, E) → G(H0 ; D, E) . (Because
([k] + [h0 ]) + [H0 ] = [k] + ([h0 ] + [H0 ]) = [k] + [H0 ] by Lemma 4.4.7(iv).)
(2) [k+βh0 ]+[H0 ] = [H0 ]. (Because [k+βh0 ] = [k]+[βh0 ] for k ∈ Hom(D, E)
with k(D) ⊆ J and [βh0 ] + [H0 ] = [H0 ] by Lemma 4.4.7(i, iv).)

Conversely properties (1a) and (1b) or properties (1a) and (2) imply for k ∈
Hom(D, E) that k ∈ A.
Indeed: (1a) implies k(D) = T ∗ h0 (D)T ∈ J for some isometry T ∈ E and
ϕ([k] + [h0 ]) = [H0 ] says by definition of ϕ that ([k] + [h0 ]) + [H0 ] = [H0 ], i.e., that
[k] + [H0 ] = [k] + ([h0 ] + [H0 ]) = ([k] + [h0 ]) + [H0 ] = [H0 ] by Lemma 4.4.7(iv).
Thus, the set of unitary equivalence classes [k] ∈ [Hom(D, E)] of the elements
k ∈ A is identical with the set of elements [k] in the semi-group S(h0 ; D, E) that
satisfy ϕ([k] + [h0 ]) = [H0 ].
The set A contains h0 by Lemma 4.4.7(iv). If k ∈ A and [h] = [k] then h ∈ A
(here [·] denotes the unitary equivalence classes of elements of Hom(D, E)). If
k1 , k2 ∈ A then k1 ⊕ k2 ∈ A.
This is because [??????] =?.
 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 573

Let k ∈ Hom(D, J) ⊂ Hom(D, E), then

[k] + [H0 ] = [H0 ] ⇔ [k + βh0 ] + [H0 ] = [H0 ] ,

because, by Lemma 4.4.7(iv,i,iii),

[k + βh0 ] + [H0 ] = [k + βh0 ] + [0] + [H0 ] = [k] + [βh0 ] + [H0 ] = [k] + [H0 ] .

Thus, if we take s0 and t0 as in assumption (iv) of Theorem 4.4.6, then [H0 ] =

[k] + [H0 ] becomes equivalent to [H0 ] = [(k + βh0 ) ⊕s0 ,t0 H0 ] , i.e., is equivalent to
the existence u ∈ U(E) that satisfies following equation:

s0 (k(·) + βh0 (·))s∗0 + (1 − p0 )H0 (·) = (k + βh0 ) ⊕s0 ,t0 H0 = u∗ H0 (·)u . (4.20)

The equation shows that u ∈ Der(H0 (D), J) and that the unitary u is deter-
mined by k in the right side of the equation up to multiplications V u by unitaries
V ∈ U(C).
We can also use the isometries s2 , t2 of Lemma 4.4.7(x) and get directly v ∈
U(E) with
k ⊕s2 ,t2 H0 = v ∗ H0 (·)v (4.21)

This unitary v is again in Der(H0 (D), J) and is determined by the left side, i.e., by
k, only up to multiplications U v with U ∈ U(C).
By Lemma 4.4.10 there exists a unitary W ∈ U0 (Der(H0 (D), J)) ∩ (p0 Ep0 +
(1 − p0 )) with ((k + βh0 ) ⊕s0 ,t0 H0 )W = W (k ⊕s2 ,t2 H0 ) .
It follows that W ∗ u∗ H0 (·)uW = v ∗ H0 (·)v, i.e., U(C)uW = U(C)v. In partic-
ular, µ(K1 (C)) + [u] = µ(K1 (C)) + [v] in K1 (Der(H0 (D), J)), because [W ] = 0 in


K1 Der(H0 (D), J) .


Recall here, that we have defined µ : K1 (C) → K1 Der(H0 (D), J) by the
natural *-monomorphism C ,→ Der(H0 (D), J).
The above considerations show that the class of the unitaries u and v in Equa-
tions (4.20) and (4.21) define for k ∈ A the same class

γ0 (k) := [u] + µ(K1 (C)) = [v] + µ(K1 (C)) (4.22)

in K1 (Der(H0 (D), J))/µ(K1 (C)).

We get that that γ0 (h0 ) = µ(K1 (C)), because s0 (h0 + βh0 )s∗0 = p0 H0 and
t0 H0 t∗0 = (1 − p0 )H0 by Assumptions (iii) and (iv) of Theorem 4.4.6.
If we use the definition of the unitary v in Equation (4.21) then we obtain that
γ0 (k) = γ0 (U ∗ kU ) for all unitaries U ∈ E:
Indeed, the property k(D) ⊆ J implies that there exists a unitary V ∈ U0 (E) with
U ∗ k(·)U = V ∗ k(·)V , cf. Lemma ??(????).
Let v ∈ U(Der(H0 (D), J) as in Equation (4.21) and W := v(V ⊕s2 ,t2 1) then

(U ∗ k(·)U ) ⊕s2 ,t2 H0 = W ∗ H0 (·)W .

574 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

It implies that U(C)W = U(C)v(V ⊕s2 ,t2 1). The unitary V ⊕s2 ,t2 1 is in s2 U0 (E)s∗2 +
t2 t∗2 . The group s2 U0 (E)s∗2 + t2 t∗2 is contained in the connected component of
U(Ann(H0 (D), J) + C1) ⊆ U(Der(H0 (D), J)).
It implies [W ] = [v] ∈ U(Der(H0 (D), J)) and γ0 (U ∗ k(·)U ) = γ0 (k) , by the
defining equation (4.22) of γ0 .
It follows that there is a unique mapping γ from the set [A] of elements [k] ∈
S(h0 ; D, E) with the property that [k] + [h0 ] is in the kernel of

S(h0 ; D, E) → G(h0 ; D, E) → G(H0 ; D, E)

to K1 Der(H0 (D), J) /µ K1 (C) , such that γ satisfies γ0 (k) = γ([k]) for k ∈ A
and γ([h0 ]) = µ(K1 (C)).
We show now that γ is additive:
Let k1 , k2 ∈ A and take u1 , u2 ∈ U(Der(H0 (D), J)) with kj ⊕s2 ,t2 H0 = u∗j H0 uj for
j ∈ {1, 2}.
By Lemma 4.4.8, if (s2 , t2 ) are the isometries in Lemma 4.4.7(x) and the unitary
v2 ∈ C is defined by Equation (4.11) given in proof of Lemma 4.4.7(x), then for
a, b, c, d ∈ E holds:

v2∗ (a ⊕s2 ,t2 c) ⊕s,t (b ⊕s2 ,t2 d) v2 = (a ⊕s,t b) ⊕s2 ,t2 (c ⊕s,t d) .

If we let a := k1 (d), b := k2 (d), c = d := H0 (d) then we get for kj ⊕s2 ,t2 H0 =

u∗j H0 uj and the unitary U := (u1 ⊕s,t u2 ) · v2 the equation

U ∗ H0 U = v2∗ (u∗1 H0 u1 ) ⊕s,t (u∗2 H0 u2 ) v2 = (k1 ⊕s,t k2 ) ⊕s2 ,t2 H0 .

It follows that γ0 (k1 ⊕s,t k2 ) = µ(K1 (C)) + [U ] Since [U ] = [u1 ] + [u2 ] + [v2 ] and
v2 ∈ C, it follows that

γ0 (k1 ⊕s,t k2 ) = γ0 (k1 ) + γ0 (k2 )

in K1 (Der(H0 (D)), J)/µ(K1 (C)) =: Γ H0 (D), J, E .
Thus, γ0 is additive with respect to Cuntz addition on A with γ0 (h0 ) =
µ(K1 (C)) and defines an additive map γ1 from the sub-semigroup of those elements
[k] of S(h0 ; D, E) with [k] + [h0 ] in the kernel of ϕ.
NEXT steps:
Step (next 1):
TEXT:
Let k ∈ A then γ0 (k) = µ(K1 (C))
if and only if [k] + [h0 ] = [h0 ]:
It shows that γ1 ([k]) = µ(K1 (C)) and k ⊕s2 ,t2 H0 = u∗ H0 u if and only if
[u] ∈ µ(K1 (C)).
Notice that [A] + [h0 ] is a subgroup of G(h0 ; D, E) and is identical with the
kernel of ϕ.
 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 575

It will be shown below that k ∈ A and γ0 (k) = 0 together imply that [k]+[h0 ] =
[h0 ]. It follows that γ1 |[A] + [h0 ] “restricted” to the kernel of ϕ in G(h0 ; D, E) is
faithful on the kernel ϕ−1 ([H0 ]) of ϕ.
The neutral element “0” of K1 (Der(H0 (D), J))/µ(K1 (C)) is γ0 (h0 ) = µ(K1 (C))
and the invariance under unitary equivalence that [k] + [h0 ] = [h0 ] implies
γ0 (k) = γ1 ([k]) = µ(K1 (C)) .

Suppose that k ∈ A and γ0 (k) = µ(K1 (C)). It follows that there exists
v ∈ U(C) and w ∈ U0 (Der(H0 (D), J)) with (v ⊕ s, t1)w = (u ⊕s,t 1) (cf. Lemma
4.2.6(v,2)).
Each w ∈ U0 (Der(H0 (D), J)) decomposes by Lemma 4.4.15(v) into a product
w = w1 w2 with w1 ∈ U(C) and w2 ∈ U(C · 1 + N (H0 (D), J)).
Is it not U(C1 + N (H0 (D), J)) ???
It follows that (k ⊕s,t h0 ) ⊕s2 ,t2 H0 = w2∗ H0 w2 .
This causes by Lemma ?? that [k ⊕h0 ] = [h0 ] provided that in addition ???????
Step (next 2): γ1 is surjective:
(a) Der(H0 (D), J) and C are K1 -surjective.
i.e., For x ∈ K1 (Der(H0 (D), J)) exists u ∈ U(Der(H0 (D), J)) with x = [u].
compare next with places above!
We define for k ∈ A a class U(k) = U(C)u ⊂ U(Der(H0 (D), J)) of
U(C)\ U(Der(H0 (D), J)) by u ∈ U(k) if and only if u satisfies Equation (4.20).
If k ∈ A, then the set U(k) ⊆ U(Der(H0 (D), J)) is an element of the right
homogenous space
U(C)\ U(Der(H0 (D), J))
given by right multiplication of elements of U(Der(H0 (D), J)) by elements of
U(C) ⊆ U(Der(H0 (D), J)) .
Clearly, the unitary u in Equation (??) is determined by k only up to multipli-
cations vu with v ∈ U(C). We denote this subset of U(E) by
U(k) := U(C)u
if a C *-morphism k : D → J satisfies [k] + [H0 ] = [H0 ]. Thus, U(k) = U(C) · U(k).
U(k) ⊆ U(Der(H(D), J)):
The elements of U(k) are contained in Der(H(D), J), because
p0 H(·) = s0 (h0 + βh0 )s∗0 ,
by assumption 4.4.6(iv), and this together with Equation (??) implies
H0 (a)u − uH0 (a) = us0 (k(a) − h0 (a))s∗0 ∈ J for all a ∈ D .
Thus u ∈ Der(H0 (D), J). Since C ⊆ Der(H0 (D), J) it follows that
U(C)u =: U(k) ⊆ U(Der(H0 (D), J))
576 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

The class
U(k) ∈ U(C) \U(Der(H0 (D), J))
defines a class
γ0 (k) := [u] + µ(K0 (C)) ∈ Γ := Γ(H0 (D), J, E) := K0 (Der(H0 (D), J))/µ(K0 (C)) .

Next steps:
(next 1)
γ0 (k1 ) = γ0 (k2 ) if [k1 ] = [k2 ], i.e., γ0 (k) is invariant under unitary equivalence.
Let w ∈ U(E) with w∗ k2 (·)w = k1 .
If [k1 ] = [k2 ] then there exists by Lemma 4.4.7(viii) a unitary w ∈ U0 (βh0 (D)0 ∩
E) such k1 + βh0 = w∗ (k2 + βh0 )w.
Let ξ ∈ [0, 1] 7→ w(ξ) ∈ U(βh0 (D)0 ∩ E) a continuous path with w(0) = 1 and
w(1) = w. The continuous path
v(ξ) := s0 w(ξ)s∗0 + t0 t∗0 ∈ U(E)
satisfies v(ξ) = 1 and for d ∈ D that
v(ξ)H0 (d) − H0 (d)v(ξ) = s0 (w(ξ)h0 (d) − h0 (d)w(ξ))s∗0 ∈ J ,
i.e., v(ξ) ∈ Der(H0 (D), J). It follows that v(1) ∈ U0 (Der(H0 (D), J)). Thus,
[v(1)] = 0 in K1 (Der(H0 (D), J)) and
γ0 (k2 ) = γ0 (k1 ) + (µ(K1 (C)) + [v(1)]) = γ0 (k1 ) .


Notice that w ∈ U0 (E) also implies that v ∈ U0 N (H0 (D), J + βJ) + C · 1 .
An alternative proof of the equivalence of γ0 with respect to unitary equivalence
can be given directly by Lemma 4.4.7(i):
The equality [k1 ] = [k2 ] yields by ???????????
Consequence of the invariance:
We can define
γ1 ([k]) := γ0 (k)
for a representative k ∈ Hom(D, J) of the class [k] ∈ S(h0 ; D, E).

(next 2)
γ1 ([k1 ] + [k2 ]) = γ1 ([k1 ]) + γ1 ([k2 ]), i.e., the mapping
γ1 : S(h0 ; D, E) → Γ := Γ(H0 (D), J, E) := K1 (Der(H0 (D), J))/µ(K1 (C))
is an additive map from the semi-group S(h0 ; D, E) into the group Γ.
Use Lemma 4.4.8 and that s = β(s), t = β(t) ∈ (h0 (D) ∪ H0 (D))0 ∩ E by
assumptions of Theorem 4.4.6, both cases of definitions with ⊕s0 ,t0 or with ⊕s2 ,t2 .

(next 3)
U(h0 ) = U(C), i.e., γ1 ([h0 ]) = γ0 (h0 ) = 0.
 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 577

In particular, γ1 ([k] + [h0 ]) = γ1 ([k]) for all k : D → J with [k] + [H0 ] = [H0 ].
Thus, γ([k]) := γ1 ([k]) defines a group homomorphism from the ker-
nel ϕ−1 ([H0 ]) ⊆ G(h0 ; D, E) of ϕ : G(h0 ; D, E) → G(H0 ; D, E) into Γ :=
Γ(H0 (D), J, E).

(next 4)
The homomorphism γ is surjective:
For every x ∈ K1 (Der(H0 (D), J)) there exists u ∈ U(Der(H0 (D), J)) with
[u] = x, because C ⊆ Der(H0 (D), J) is properly infinite and this implies that
Der(H0 (D), J) (respectively C) is K1 -surjective by Lemma 4.2.6(v).
It suffices to find k : D → J, v ∈ U(C) and w ∈ U0 (Der(H0 (D), J)) with
(k + βh0 ) ⊕s0 ,t0 H0 = w∗ v ∗ u∗ H0 (·)uvw .

Give Ref’s for next blue!!

In fact, there exists v ∈ U0 (F + C · 1) ∩ (1 + F ) such that uv ∈ U(k) for some
C *-morphism k : D → J with [k] + [H0 ] = [H0 ].
Thus, [uv] = [u] + [v] = [u] in K1 (Der(H0 (D), J)), and γ([k] + [h0 ]) =
µ(K1 (C)) + [u] in Γ. It shows that γ is surjective.

(next 5)
If k : D → J is a C *-morphism, y ∈ N (J + βJ), 1 + y ∈ U(E) and
s0 (k + βh0 )s∗0 + (1 − p0 )H0 (·) = (1 + y ∗ )H0 (·)(1 + y) ,
then [k] + [h0 ] = [h0 ]:
Indeed, by Lemma 4.4.9 there exists a projection p ∈ F with p ≥ p0 , p 6= p0
and
(1 + y ∗ )(1 − p)H0 (·)(1 + y) = (1 − p)H0 (·) ,
because it implies that
(1 + y ∗ )pH0 (·)(1 + y) = s0 (k + βh0 )s∗0 + (p − p0 )H0 (·) .

By Part (xii) of Lemma 4.4.7, there exists a unitary V ∈ C and u2 ∈ U0 (E)

with up0 = p0 = p0 u and
V ∗ (p − p0 )H0 (·)V = t0 u∗2 (h0 + βh0 )u2 t∗0 .
It follows for W := V (p0 + t0 u∗2 t∗0 ) that
(k + βh0 ) ⊕s0 ,t0 (h0 + βh0 ) = W ∗ (1 + y ∗ )pH0 (·)(1 + y)W .
By part (ix) of Lemma 4.4.7 there exists a unitary u3 ∈ U0 (F + C · 1) ⊂ C with
u∗3 pu3 = p0 = s0 s∗0 and, by assumption (iv) of Theorem 4.4.6, p0 H0 (·) = s0 (h0 +
βh0 )s∗0 . We obtain with the unitary U := u3 (1 + y)W ∈ U(E) that
(k + βh0 ) ⊕s0 ,t0 (h0 + βh0 ) = U ∗ (h0 + βh0 ) ⊕s0 ,t0 0) U .

If we use now that J ∩ βJ = 0 and h0 (D) ∪ k(D) ⊂ J then we get

578 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

k ⊕s0 ,t0 h0 = U ∗ (h0 ⊕s0 ,t0 0) U ,

which is up to multiplication of U by unitaries in (s0 h0 (D)s∗0 )0 ∩ E equivalent to

[k] + [h0 ] = [h0 ] + [0]. But we know from Part (ii) of Lemma 4.4.7 that [h0 ] =
[h0 ] + [0].

(next 6)
The kernel of γ is {[h0 ]}:
If γ([k]) = µ(K1 (C)), and u ∈ U(k) ⊆ Der(H0 (D), J), then there exist V ∈
U(C) and W ∈ U0 (Der(H0 (D), J)) such that u ⊕S,T 1 = V W .
Definition of S, T ??
It follows that
((k ⊕s,t h0 ) + βh0 ) ⊕s0 ,t0 H0 = (u ⊕S,T 1)∗ H0 (·)(u ⊕S,T 1) = W ∗ H0 (·)W .

By Lemma 4.4.15(iv ????) ????, W = V1 W1 with V1 ∈ C and W1 = y + 1 ∈

U(N (H0 (D), J) + C · 1) with y ∈ N (H0 (D), J). Thus, for W1 := y + 1,
s0 ((k ⊕s,t h0 ) + βh0 )s∗0 + (1 − p0 )H0 (·) = W1∗ H0 (·)W1 .

Since
C = H0 (D)0 ∩ E = Der(H0 (D), {0}) ⊆ Der(H0 (D), J) ,
we can define for k ∈ Hom(D, J) with [k] ∈ S(h0 ; D, E) in the kernel of ϕ an
element U(k) of the (right-sided) homogenous space U(C)\ U(Der(H0 (D), J)) by
U(k) := {vu ; v ∈ U(C) }
where is u some unitary satisfying Equation (??).
Obvious examples are U(h0 ) = U(C) and U(0) = U(C)uℵ , where uℵ := s1 s∗0 +
t1 t∗0 with the isometries s1 , t1 of Lemma 4.4.7(vi).
If [k] = [k 0 ], then [k + βh0 ] = [k 0 + βh0 ] by Lemma 4.4.7(viii), and the unitary
equivalence can be realized by a unitary w ∈ U0 (E) with w∗ (k + βh0 )w = k 0 + βh0 .
It implies that
U(k 0 ) = U(k) · (w ⊕s0 ,t0 1) .
Notice that w ⊕s0 ,t0 1 is a unitary in 1 + N (H0 (D), J) and is in U0 (N (H0 (D), J) +
C · 1). In particular, U(k) and U(k 0 ) are contained in the same two-sided class of
U(C)\ U(Der(H0 (D), J))/U0 (N (H0 (D), J) + C · 1) .
It follows that U(k) ⊆ U(C) · U(N (H0 (D), J)) if [k] = [h0 ].
The isometries s1 , t1 of Lemma 4.4.7(vi) satisfy p0 t1 = t1 p0 = p0 , t1 ∈ C,
H0 (·)s1 = 0, s∗1 H0 (·) = 0, H0 (·)t1 = H0 , and t∗1 H0 (·) = H0 . It yields t1 at∗1 = a for
all a ∈ s0 Es∗0 , and t1 (1 − p0 )H0 (·)t∗1 = (1 − p0 )H0 (·).
Let u ∈ U(E), then
u1 := u∗ ⊕s1 ,t1 u := s1 u∗ s∗1 + t1 ut∗1 (4.23)
 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 579

has the properties

u∗1 H0 (·)u1 = t1 u∗ H0 (·)ut∗1 and u1 ∈ U0 (E) .

The above equations together imply for u ∈ U(k) ⊆ U(E) in Equation (??), that
the unitary u1 ∈ U0 (E) defined by (4.23) again satisfies Equation (??):

(k + βh0 ) ⊕s0 ,t0 H0 = u∗1 H0 (·)u1 .

It follows that u1 ∈ U(k) ∩ U0 (E) 6= ∅ if [k] ∈ S(h0 ; D, E) and [k] + [H0 ] = [H0 ].
We seek for an estimate of the set U(k1 ⊕s,t k2 ) ⊆ U(Der(H0 (D), J)) for kj : D →
E with [kj ] ∈ S(h0 ; D, E) and [kj ] + [H0 ] = [H0 ], j = 1, 2.
New generators S, T of O2 come from the generators (s, t) and (s0 , t0 ) in The-
orem 4.4.6(i,iv) by

S := s0 ss∗0 + t0 st∗0 and T := s0 ts∗0 + t0 tt∗0 (4.24)

Then S, T ∈ C: use β(s) = s, β(t) = t, s, t ∈ h0 (D)0 ∩ H0 (D)0 ∩ E,

SH0 = Ss0 (h0 + βh0 )s∗0 + St0 H0 t∗0 = s0 (sh0 + sβh0 )s∗0 + t0 sH0 t∗0 = s0 (h0 s +
(βh0 )s)s∗0 + t0 H0 st∗0 = H0 S . Similarly, T H0 = H0 T .
Since Ss0 = s0 s, T s0 = s0 t, Sp0 = p0 S, T p0 = p0 T and

(βh0 ) ⊕s,t (βh0 ) = β(h0 ⊕s,t h0 ) = βh0 ,

we get for k1 , k2 : D → J that

(k1 + βh0 ) ⊕s0 ,t0 H0 ⊕S,T (k2 + βh0 ) ⊕s0 ,t0 H0 = (k1 ⊕s,t k2 ) + βh0 ⊕s0 ,t0 H0 .

If k1 , k2 ∈ S(h0 ; D, E) with [k] + [H0 ] = [H0 ] and [k 0 ] + [H0 ] = [H0 ], then for
k3 := k1 ⊕s,t k2 holds [k3 ] = [k1 ] + [k2 ] and [k3 ] + [H0 ] = [H0 ].
The class U(k3 ) is given by

U(k1 ⊕s,t k2 ) = U(C) · (U(k1 ) ⊕S,T U(k2 )) ,

because, for u1 ∈ U(k1 ) and u2 ∈ U(k2 )

k1 ⊕s,t k2 ) + βh0 ⊕s0 ,t0 H0 = (u1 ⊕S,T u2 )∗ H0 (·)(u1 ⊕S,T u2 ) .

It follows that
U(k ⊕s,t h0 ) = U(C) · (U(k) ⊕S,T 1) ,
and, if [k] + [h0 ] = [k 0 ] + [h0 ], that there is a unitary u0 ∈ U0 (E) with

u∗0 (k ⊕s,t h0 ) + βh0 )u0 = k 0 ⊕s,t h0 ) + βh0 ,

i.e., then
U(k 0 ⊕s,t h0 ) = U(C) · (U(k) ⊕S,T 1) · (u0 ⊕s0 ,t0 1) .
Since
u0 ⊕s0 ,t0 1 ∈ U0 (E) ⊕s0 ,t0 1 ⊆ U0 (N (H0 (D), J) + C · 1) ,
this implies in particular

U(k 0 ⊕s,t h0 ) ⊆ U(C) · (U(k) ⊕S,T 1) · U0 (N (H0 (D), J) + C · 1) .

580 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

We show below that the weaker condition

U(k 0 ) ⊆ U(C) · (U(k) ⊕S,T 1) · U0 (N (H0 (D), J + βJ) + C · 1) ,
implies conversely that
[k] + [h0 ] = [k 0 ] + [h0 ] .

It follows the equivalence of

[k] + [h0 ] = [h0 ] = [h0 ] + [h0 ]
and
U(k) ⊆ U(C) · U0 (N (H0 (D), J + βJ) + C · 1) .

If we let ????????, then we can find a unitary u ∈ E that satisfies Equation

(??) with the additional condition
1 ⊕S,T u ∈ U0 (E) .

Next also mentioned above ?

Let (k + βh0 ) ⊕s0 ,t0 H0 = u∗ H0 (·)u. Since p0 H0 (·) = s0 (h0 (·) + βh0 (·))s∗0 by
assumption (iv) of Theorem 4.4.6, we get that
u∗ H0 (a)u − H0 (a) = s0 (k(a) − h0 (a))s∗0 ∈ J for all a ∈ D.

It follows that u ∈ U(Der(H0 (D), J)), u(1−p0 ) ∈ C and up0 u∗ H0 (D) ⊆ J +βJ.
Since 1 − up0 u∗ = u(1 − p0 )2 u∗ ∈ C, the map
a ∈ D 7→ up0 u∗ H0 (a)
is a C *-morphism from D into J + β(J), because p0 u∗ H0 (a)u − p0 H0 (a) ∈ J and
p0 H0 (a) ∈ J + β(J) for all a ∈ D.
Let (k + βh0 ) ⊕s0 ,t0 H0 = u∗ H0 (·)u. It implies u ∈ Der(H0 (D), J).
We get (see above)
((k ⊕s,t h0 )?????βh0 ) ⊕s0 ,t0 H0 = U ∗ H0 (·)U
with U := u ⊕S,T 1.
Suppose now more generally that u ∈ Der(H0 (D), J + βJ) is given with the
additional property and u ⊕s,t 1 = wv with unitaries w ∈ C and v = 1 + x with
x ∈ N (H0 (D), J + βJ).
Let W := sS ∗ + tT ∗ . Since s, t, S, T ∈ C it follows that W ∈ C and u ⊕S,T 1 =
∗
W (u ⊕s,t 1)W .
We get u ⊕S,T 1 = W ∗ wW · W ∗ vW , and W ∗ vW = 1 + y with W ∗ wW ∈ C and
y = W ∗ xW ∈ N (H0 (D), J + βJ).
Thus, U ∗ H0 (a)U = (1 + y ∗ )H0 (a)(1 + y). Since H0 (D)y ⊆ J + βJ and D is
separable, there is a separable C *-subalgebra B ⊆ J + βJ that contains H0 (D)y.
Let f ∈ B+ is a strictly positive contraction B. By assumptions (iii), (iv) and (v) of
 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 581

Theorem 4.4.6 there exists a projection q ∈ F ⊆ C with qf = f , q ≥ p0 and q 6= p0 .

It follows that qH0 (a)y = H0 (a)y = H0 (a)yq. Thus qH0 (a)(1 + y) = H0 (a)(1 + y)q
and qU ∗ H0 (a)U = (W ∗ vW )∗ qH0 (a)W ∗ vW .
Since q = p0 + r with r = q − p0 = q(1 − p0 ) ≤ t0 t∗0 , r ∈ E, it follows for
???????
that

s0 ((k ⊕s,t h0 ) + βh0 )s∗0 + rH0 (·) = (W ∗ vW )∗ qH0 (a)W ∗ vW .

There are unitaries u1 , u2 ∈ 1 + F such that u1 ru∗1 = p0 and u2 qu∗2 = p0 . We get

rH0 (·) = u∗1 p0 H0 (·)u1 and qH0 (·) = u∗2 p0 H0 (·)u2 . Since p0 H0 (·) = s0 (h0 +βh0 )s∗0 =
(h0 + βh0 ) ⊕s0 ,t0 0 and J · βJ = 0 = J ∩ βJ, we obtain for the summands with
values in J that

s0 (k ⊕s,t h0 )s∗0 + u∗1 s0 h0 (·)s∗0 u1 = (W ∗ vW )∗ u∗2 s0 h0 (a)s∗0 u2 W ∗ vW ,

with p0 u∗1 s0 = 0. Can define V1 ∈ U(E) such that

t1 V1∗ s0 h0 (·)s∗0 V1 t∗1 = u∗1 s0 h0 (·)s∗0 u1 = rH0 (·)

It follows that
[k ⊕s,t h0 ] + [h0 ] + [0] = [h0 ] + [0] .

From u ∈ U(Der(H0 (D), J)) to (k, uw):

Recall that F ⊆ C ∩ N (H0 (D), J) in the following calculations.
SURJECTIVITY:
Let u ∈ U(Der(H0 (D), J)), then there exists an element g ∈ F and a C *-
morphism k : D → J with [k] ∈ S(h0 ; D, E), such that w := 1 + g ∈ U0 (F + C · 1)
and
(uw)∗ H0 (·)uw = (k + βh0 ) ⊕s0 ,t0 H0 .

Indeed: Let ∂u (x) := ux−xu for x ∈ E. Then u∗ ∂u (H0 (D)) ⊆ J. and generates
a separable C *-subalgebra B of J. Let b0 ∈ B+ a strictly positive contraction of B
and f := b0 + β(b0 ). By assumptions 4.4.6(iii,iv,v) there are projections p, q ∈ F
with p0 ≤ q ≤ p with qf = f , p0 6= q, q 6= p. By Lemma 4.4.7(ix) there exists
w ∈ U0 (F + C1) such that q = wp0 w∗ . Since w ∈ U0 (C) ⊆ Der(H0 (D), J), we get
uw ∈ Der(H0 (D), J) and [uw] = [u] ∈ K1 (Der(H0 (D), J)).
Then (1 − q)u∗ H0 (a)u = (1 − q)H0 (a) and w ∈ U0 (C). Thus (1 −
p0 )(uw)∗ H0 (a)(uw) = (1 − p0 )H0 (a) and

h(a) := s∗0 p0 (uw)∗ H0 (a)(uw)s0

is a C *-morphism. Since uws0 is an isometry we get [h] ∈ S(H0 ; D, E). The

C *-morphism h satisfies

h ⊕s0 ,t0 H0 = (uw)∗ H0 (·)(uw) ,

582 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

and h(a)−s∗0 H0 (a)s0 ∈ J for a ∈ D. Since s∗0 H0 (a)s0 = h0 (a)+βh0 (a) by Theorem
4.4.6(iv), h = k + βh0 with [k] ∈ S(H0 ; D, E) and k(D) ⊆ J, and by Lemma
4.4.7(vii) that [k] ∈ S(h0 ; D, E), uw ∈ U(k), and [uw] = [u] ∈ K1 (Der(H0 (D), J)).
In particular,
(k + βh0 ) ⊕s0 ,t0 H0 = (uw)∗ H0 (·)(uw) .

Where is next needed??

The C *-morphism h0 : a ∈ D 7→ h0 (a) := upu∗ H0 (a) = us0 h(·)s∗0 u∗ maps D
into J, therefore it satisfies [h0 ] = [h] + [0] = [h], [h0 ] = [k + βh0 ] and h0 ⊕s0 ,t0 H0 =
u∗1 H0 (·)u1 for some unitary u1 ∈ Der(H0 (D), J + β(J)) such that u1 := v1 uw1 for
some v1 ∈ C and w1 ∈ N (H0 (D), J + β(J)).
Next: From (DC) to (wDC)?
We consider now the special case, where u ∈ U0 (E) and k ∈ S(h0 ; D, E) satisfy

(k + βh0 ) ⊕s0 ,t0 H0 = u∗ H0 (·)u ,

and suppose in addition that u ∈ U(C) · U(N (H0 (D), J + βJ)).

Let B0 the separable C *-algebra of J that is generated by k(D)∪h0 (D), g ∈ B0
a strictly positive contraction of B0 . By assumption (iii) of Theorem 4.4.6 there
exists q1 ∈ F , q1 ≥ p0 with q1 (g + βg) = g + βg and q1 6= p0 .
Above we have seen that (k + βh0 ) ⊕s0 ,t0 H0 = u∗ H0 (·)u . implies u ∈
Der(H0 (D), J).
We suppose now that a given u ∈ U(Der(H0 (D), J)) satisfies the assumption
of the decomposition condition (DC), i.e., that there is v ∈ C such that w :=
v(u ⊕s,t 1) ∈ 1 + N (H0 (D), J + β(J)).
This implies (u ⊕s,t 1) ∈ U(C) · U(N (H0 (D), J + βJ)).
We show that the latter implies that [k] + [h0 ] = [h0 ]:
Recall:

(DC) If u ∈ U0 (E) satisfies uH0 (a) − H0 (a)u ∈ J for all a ∈ D, then there exists
a unitary v ∈ C := H0 (D)0 ∩ E such that

v(u ⊕s,t 1) ∈ N (H0 (D), J + β(J)) + C · 1 .

THIS IMPLIES:
If u ∈ U0 (E) and u ∈ Der(H0 (D), J) then
[u] = [u ⊕ 1] = [w] − [v] with v ∈ U(C) and w ∈ U(N (H0 (D), J)).
Thus, the kernel of K1 (Der(H0 (D), J)) → K1 (E) is contained in γ(K1 (C)).
We know that K1 (C) → K1 (E) is surjective, cf. . ?????????, because H0
dominates (absorbs) zero.
Let q1 ≥ p0 as above. Then w = 1 + y for some y ∈ N (H0 (D), J + β(J)). There
is a separable C *-algebra B ⊆ J such that B + β(B) contains H0 (D)y ∪ yH0 (D). If
 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 583

b0 ∈ B+ is a strictly positive contraction for B, then there exists a projection q2 ∈

F ⊆ C with q2 ≥ q1 , q2 6= q1 and q2 (b0 + β(b0 )) = b0 + β(b0 ) by Theorem 4.4.6(iii).
It follows q2 H0 (a)y = H0 (a)y = H0 (a)yq2 , q2 H0 (a)w = H0 (a)q2 + q2 H0 (a)y =
H0 (a)wq2 and q2 wH0 (a) = wH0 (a)q2 in the same way. Moreover q2 − q1 ∈ F is
a non-zero projection and there is a unitary z3 ∈ (1 + F ) ∩ U0 (F + C · 1) with
z3∗ (q2 − q1 )z3 = p0 .
Let z1 , z2 ∈ 1 + F unitaries with z1∗ q1 z1 = p0 and z2∗ q2 z2 = p0 , cf. assumption
(v) of Theorem 4.4.6.
We get:
??? Then p0 H0 (·)z ∗ wz = H0 (·)z ∗ wzp0 .
New approach:
The crucial points are:

q2 u∗ H0 u = q2 w∗ H0 w = w∗ q2 H0 (·)w

and
q2 ((k + βh0 ) ⊕s0 ,t0 H0 ) = s0 (k + βh0 )s∗0 + (q2 − p0 )H0
and that q2 H0 and (q2 − p0 )H0 are unitarily equivalent to p0 H0 = s0 (h0 + βh0 )s∗0
by unitaries v1 , v2 ∈ C: Say q2 H0 = v1∗ p0 Hv1 and (q2 − p0 )H0 = v2∗ p0 Hv2 .
Since h0 dominates zero, we get that [q2 u∗ H0 u] = [h0 + βh0 ]. Let

h1 (a) := t∗0 (q2 − p0 )H0 (a)t0 = t∗0 v1 p0 H0 (a)v1∗ t0 .

Then t0 h1 (a)t∗0 = (q2 − p0 )H0 (a). Thus [k + βh0 ] + [h1 ] = [h0 + βh0 ].
We show that h1 dominates zero:
By Theorem 4.4.6(iii) – with f := 0 – there is a projection q3 ∈ F with q2 ≤ q3
and 0 6= q4 = q3 − q2 ≤ 1 − p0 = t0 t∗0 . It follows that the projection q5 := t∗0 q4 t0
is Murray–von-Neumann equivalent to q4 = t0 q5 t∗0 and satisfies h1 (·)q5 = 0. The
assumption (v) of Theorem 4.4.6 allows to find a unitary u4 ∈ 1 + F ⊆ C with
u∗ q4 u = q0 = s0 s∗0 . The element T := s∗0 u∗ t0 q5 ∈ E satisfies T ∗ T = 1 and
T T ∗ = q5 . Thus, T ∗ h1 (·)T = 0, and h1 dominates zero.
It follows [0] + [h1 ] = [h1 ], i.e.,

t0 h1 (·)t∗0 = (q2 − p0 )H0 (·) = v2∗ p0 H0 (·)v2

is unitarily equivalent to h1 by Proposition 4.3.5(i). We obtain

[h1 ] = [p0 H0 (·)] = [0] + [h0 + βh0 ] = [h0 + βh0 ]

by Lemma 4.4.7(ii). It yields [k + βh0 ] + [h0 + βh0 ] = [h0 + βh0 ]. Summing up, we
get the existence of a unitary U ∈ E with

s0 (k + βh0 )s∗0 + t0 (h0 + βh0 )t∗0 = U ∗ (h0 + βh0 )U .

Since k(D) ∪ h0 (D) ⊆ J and J ∩ β(J) = {0}, it follows that s0 k(a)s∗0 + t0 h0 (a)t∗0 =
U ∗ h0 (a)U for all a ∈ D, i.e., [k] + [h0 ] = [h0 ].
Next Part of proof not complete !!! ??
584 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

The plan (!!!) is to apply assumption 4.4.6(iii) to suitable separable C *-

subalgebras M ⊆ J + βJ to get projection p, q ∈ F with p0 ≤ p, p0 ≤ q and
p0 6= p, such that there exists a unitary V ∈ E with the (needed) property

V ∗ H0 (·)qV = s0 (k(·) + βh0 (·))s∗0 + (p − p0 )H0 (·) .

By Part (v), there are unitaries U1 , U2 and U3 in 1 + F ⊆ C with U1∗ p0 U1 = q,

U2∗ p0 U2 = p − p0 and U3∗ p0 U3 = p. We get that

U4∗ (H0 (·)p0 )U4 = (k + βh0 ) ⊕s0 ,R ((U2∗ (H0 (·)p0 )U2 ⊕R,s0 0)

where U4 := U1 V , R ∈ C is an isometry with with RR∗ = 1 − p0 and R(p − p0 ) =

(p − p0 )R = (p − p0 ), e.g. R := U3∗ (1 − p0 )U2 + (p − p0 ). It implies

[h0 + βh0 ] + [0] = [k + βh0 ] + [h0 + βh0 ] + [0] .

Since h0 + βh0 absorbs zero, we get [h0 + βh0 ] = [k + βh0 ] + [h0 + βh0 ] on uni-
taries equivalence classes (with Cuntz addition outside the brackets [.] and ordi-
nary addition of linear maps inside [.]). The unitary equivalence of h0 + βh0 and
(k + βh0 ) ⊕ (h0 + βh0 ) = (k ⊕ h0 ) + (βh0 ⊕ βh0 ) induces a unitary equivalence of
h0 and k ⊕ h0 because J ∩ β(J) = 0. This means [k] + [h0 ] = [h0 ].


Lemma 4.4.16. The assumptions (i)–(vi) in Theorem 4.4.6 imply the following
properties (i)-(iv).

(i) There exist isometries s2 , t2 ∈ E with s2 s∗2 +t2 t∗2 = 1, t2 ∈ C, s∗2 H0 (·)s2 =
h0 , p2 := s2 s∗2 ≤ p0 .
Moreover, p2 is a properly infinite projection in the ideal C ∩
N (H0 (D), J) of C with 0 = [p2 ] in K0 (C ∩ N (H0 (D), J)), i.e., there
exists a unital C*-morphism from O2 into p2 Cp2 .
The isometry s2 satisfies also β(s2 )∗ H0 (·)β(s2 ) = βh0 .
(ii) The projection p0 is the sum p0 = q + r of properly infinite projections
q, r ∈ C with q ∈ N (H0 (D), J) and r ∈ N (H0 (D), βJ). For each decom-
position p0 = q + r of this kind holds [qH0 (·)] = [h0 ] and [rH0 (·)] = [βh0 ].
(iii) If e ∈ E satisfies e∗ H0 (D)e ⊆ J + βJ then there exist elements x, y, z ∈ E
with max{kxk, kyk, kzk} ≤ kek and a unitary w ∈ 1 + F such that e =
wqx + wry + z , H0 (D)z = {0} , z ∗ (wqx) = 0 and z ∗ (wqy) = 0, where
q, r ∈ C are projections with p0 = q + r and q, β(r) ∈ N (H0 (D), J).
Pn ∗
(iv) Let T1 , . . . , Tn ∈ E such that k(a) := j=1 Tj H0 (a)Tj ∈ J + βJ for
all a ∈ D. Then there exist S1 , S2 ∈ E, such that k = k1 + βk2 with
k` (a) = S`∗ h0 (a)S` for a ∈ D, ` ∈ {1, 2}.
In particular, if [k] ∈ S(H0 ; D, E) and k(D) ⊆ J + βJ, then k =
k1 + βk2 with unique k1 , k2 ∈ S(h0 ; D, E).

New numbers X-6

 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 585

Proof. (i): By Lemma 4.4.7(ii), there exists a unitary w ∈ E with

w∗ (h0 + βh0 )w = sh0 s∗ + tβh0 t∗ .

The isometry T := ws ∈ E satisfies T ∗ (h0 + βh0 )T = h0 . Let s2 := s0 T . Then

p2 := s2 s∗2 ≤ p0 = s0 s∗0 and s∗2 H0 (·)s2 = h0 . Thus p2 H0 (D) = s2 h0 (D)s∗2 ⊆ J,
p2 ∈ C, p2 Cp2 = s2 (h0 (D)0 ∩ E)s∗2 and p2 ∈ N (H0 (D), J).
We get p2 Cp2 ⊆ C ∩ N (H0 (D), J), because p2 cp2 ≤ kckp2 for c ∈ C+ , p2 ∈
C ∩ N (H0 (D), J) and because N (H0 (D), J) = (πJ )−1 (Ann(πJ (H0 (D)), J)) is a
hereditary C *-subalgebra of E.
In particular, s2 ss∗2 and s2 ts∗2 generate a unital copy s2 C ∗ (s, t)s∗2 of O2 that is
unitally contained in the corner p2 Cp2 of the closed ideal C ∩ N (H0 (D), J) of C.
Since there is a unital C *-morphism from O2 into p2 Cp2 – with unit p2 –, p2
is properly infinite in C ∩ N (H0 (D), J) and [p2 ] = 0 in K0 (C ∩ N (H0 (D), J)).
It follows that [1 − p2 ] = [1] = 0 in K0 (C).
Since t0 t∗0 ≤ 1 − p2 and t0 ∈ C, we get that 1 − p2 is full and properly infinite
in C. By Lemma 4.2.6(ii), 1 and 1 − p2 are MvN-equivalent in C, i.e., there exists
an isometry t2 ∈ C with t2 t∗2 = 1 − s2 s∗2 .
The equation β(s2 )∗ H0 (·)β(s2 ) = βh0 holds, because βH0 = H0 . The corner
β(p2 )Cβ(p2 ) of C is a corner of the ideal C ∩ N (H0 (D), β(J)) of C.
(ii): Let s2 as in Part (i). Since

[h0 ] + [βh0 ] = [h0 ⊕s,t βh0 ] = [h0 + βh0 ]

by Lemma 4.4.7(ii), and since s∗0 H0 (·)s0 = h0 + βh0 by assumption (iv) of Theorem
4.4.6, there exists a unitary w ∈ E such that z = (s2 ⊕s,t β(s2 ))ws∗0 is a partial
isometry in C with z ∗ z = p0 and zz ∗ = sp2 s∗ +tβ(p2 )t∗ – compare Lemma 4.3.4(ii)
and proof of Proposition 4.4.3(i).
The inequality ss∗ + tβ(p2 )t∗ ≤ 1, implies s(1 − p2 )s∗ ≤ 1 − zz ∗ . Using that
p2 ≤ p0 , and that 1 − p0 = t0 t∗0 and s(1 − p0 )s∗ = st0 t∗0 s∗ are properly infinite
projections in C, we obtain that 1 − zz ∗ is properly infinite in C. Since 1 − zz ∗ is
properly infinite and [1 − zz ∗ ] = [1 − p0 ] = [t0 t∗0 ] = [1] in K0 (C), Lemma 4.2.6(ii)
applies to 1 − zz ∗ , 1 ∈ C, and there exists an isometry T ∈ C with T T ∗ = 1 − zz ∗ .
The element W := z + T t∗0 is a unitary in C with

p0 = (z ∗ z)2 = W ∗ zz ∗ W = W ∗ (sp2 s∗ + tβ(p2 )t∗ )W .

Hence q := W ∗ sp2 s∗ W = (W ∗ ss2 )(W ∗ ss2 )∗ and r := W ∗ tβ(p2 )t∗ W are properly
infinite projections in C that are MvN-equivalent in C to p2 and β(p2 ), and satisfy
p0 = q + r.
Since βH0 = H0 and s∗ W, t∗ W ∈ C, we get q ∈ N (H0 (D), J) and r ∈
N (H0 (D), βJ), because p2 ∈ N (H0 (D), J).
If p0 = q + r with q ∈ N (H0 (D), J) and r ∈ C ∩ N (H0 (D), βJ) are given, then
k1 (a) := qH0 (a) and k2 (a) := rH0 (a) satisfy, by assumption (iv) of Theorem 4.4.6,
586 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

that

k1 (a) + k2 (a) = p0 H0 (a) = s0 (h0 (a) + βh0 (a))s∗0 .

Since J ∩ βJ = {0} and β(s0 ) = s0 it follows that k1 = s0 h0 (·)s∗0 = βk2 , i.e.,

[k1 ] = [βk2 ] = [h0 ] + [0] = [h0 ] .
(iii): Let e ∈ E and e∗ H0 (a∗ a)e ∈ J +βJ for all a ∈ D. Then H0 (D)e ⊆ J +βJ.
Let f ∈ J +βJ a strictly positive element of the separable C *-subalgebra A of J +βJ
that is generated by H0 (D)e.
There exists a projection p ∈ F with p0 ≤ p and pf = f by assumption (iii) of
Theorem 4.4.6. In particular H0 (D)(1 − p)e = {0}.
By assumption (v) of Theorem 4.4.6, there is a unitary w ∈ 1+F with wp0 w∗ =
p. Let z := (1−p)e, x := qw∗ e, and y := rw∗ e for projections q, r ∈ C with p0 = q+
r, q, β(r) ∈ N (H0 (D), J). Then max{ kxk, kyk, kzk } ≤ kek and wqx+wry +z = e,
and z ∗ wq = 0 = z ∗ wr follows from (1 − p)wp0 = 0.
(iv): Let k : D → E with [k] ∈ S(H0 ; D, E) and k(D) ⊆ J + βJ. By definition
of S(H0 ; D, E), there exists an isometry T ∈ E with T ∗ H0 (·)T = k .
Pn ∗
More generally, if T1 , . . . , Tn ∈ E such that k(a) := j=1 Tj H0 (a)Tj ∈ J +
n
βJ for all a ∈ D, then k(a) = T ∗ H0 (a)T for T :=
P
j=1 Sj Tj with isometries
∗ 0 ∗
S1 , . . . , Sn ∈ C (s, t) ⊆ H0 (D) ∩ E. In particular, T H0 (D)T ⊆ J + βJ . By Part
(ix), there exists a unitary w ∈ 1 + F ⊆ C, and x, y, z ∈ E with norms ≤ kT k such
that T = wqx + wry + z and H0 (D)z = {0}, where q, r ∈ C are projections with
q, β(r) ∈ N (H(D), J). It follows T ∗ H0 (·)T = x∗ qH0 (·)x + y ∗ rH0 (·)y .
The c.p. maps k1 := x∗ qH0 (·)x and k0 := y ∗ rH0 (·)y satisfy k = k0 + k1 . Let
k2 := βk0 , then k = k1 + βk2 . Since J ∩ β(J) = {0} and β 2 = id, the maps k1 and
k2 are uniquely determined by the map k = k1 + βk2 .
Now use that [qH0 (·)] = [h0 ], [rH0 (·)] = [βh0 ] by Lemma 4.4.16(ii).
It means that there are unitaries u1 , u2 ∈ E with qH0 (·) = u∗1 h0 (·)u1 and
rH0 (·) = u∗2 βh0 (·)u2 .
It follows that k1 = S1∗ h0 (·)S1 with S1 := u1 x and k2 = S2∗ h0 (·)S2 with
S2 = β(u2 y).
If kT k ≤ 1 then kSj k ≤ 1 for j = 1, 2, and k is a c.p. contraction. If k is
moreover multiplicative then k1 and k2 must be multiplicative, because J · βJ =
J ∩ βJ = {0}. If k is multiplicative, then we can use that h0 dominates zero
by Lemma 4.4.7(v), and find, by Proposition 4.3.6(ii), isometries T1 , T2 ∈ E with
T1∗ h0 (·)T1 = k1 and T2∗ h0 (·)T2 = k2 . This means that k1 , k2 ∈ S(h0 ; D, E), if,
moreover, kT k ≤ 1 and k is multiplicative. 

Lemma 4.4.17. The following properties of Der(H0 (D), J), C = H0 (D)0 ∩ E

and N (H0 (D), J) follow from assumptions (i)-(vi) of Theorem 4.4.6.

(o) If Q ∈ C is a projection and k(a) := Q · H0 (a) for a ∈ D, then [k] ∈

S(H0 ; D, E).
 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 587

If, in addition, k(D) ⊆ J + β(J), then k = k1 + βk2 for unique

C*-morphisms kj : D → E with [kj ] ∈ S(h0 ; D, E).
(i) The two-sided normalizer algebra N (H0 (D), J) is a full hereditary C*-
subalgebra of E and is an ideal of the C*-algebra Der(H0 (D), J) of deriva-
tions of H0 (D) into J. It holds Der(H0 (D), J) = C + N (H0 (D), J) .
All this is also true for the ideal J + β(J) in place of J.
(ii) If k : D → J is given with [k] + [H0 ] = [H0 ], then there exists u ∈ U(E)
with
u∗ H0 (·)u = s2 k(·)s∗2 + (1 − p2 )H0 (·) ,
u − p2 up2 ∈ C and u ∈ Der(H0 (D), J).
With v := u(s2 s∗0 + t2 t∗0 ) holds v ∗ H0 (a)v = s0 k(a)s∗0 + (1 − p0 )H0 (a)
for all a ∈ D, and v − p0 vp0 ∈ C and v ∈ Der(H0 (D), J + βJ), i.e.,

v ∗ H0 (a)v − H0 (a) ∈ J + β(J) ∀a ∈ D .

The above described u and v in U(E) are uniquely defined by k : D → J

up to left-multiplication by unitaries in C.
Compare with (xv) ???
If [k] + [h0 ] = [h0 ] in S(h0 ; D, E), then u = w1 u1 with u1 ∈ 1 +
N (H0 (D), J) and w1 ∈ C.
(iii) If G ⊆ U(Der(H0 (D), J + βJ)) is a separable subgroup of the unitaries
in Der(H0 (D), J + βJ), then there exist a projection p ∈ F ⊆ C and a
unitary w ∈ 1 + F , such that p ≥ p0 , p 6= p0 , w∗ pw = p0 and (1 − p)u ∈ C
for all u ∈ G.
(u) (u)
For each u ∈ G there are C*-morphisms k1 , k2 ∈ S(h1 ; D, E) such
that
(u) (u)
w∗ u∗ H0 (·)uw = (k1 + βk2 ) ⊕s0 ,t0 H0 .
(u)
If u ∈ Der(H0 (D), J) then k2 = h0 .
(iv) If G is a separable subgroup of U(Der(H0 (D), J)) then there exist isome-
tries S, T ∈ E with T T ∗ + SS ∗ = 1, S ∗ u∗ H0 (·)uS = 0 and T u∗ H0 (·)u =
u∗ H0 (·)u = u∗ H0 (·)uT for all u ∈ G.
(v) If u ∈ Der(H0 (D), J) is unitary, and S, T ∈ E are the isometries from
Part (iv) for a separable subgroup G of U(Der(H0 (D), J)) that contains u,
then the unitary W := Su∗ S ∗ + T uT ∗ ∈ U0 (E) satisfies (W ∗ )k H0 (·)W k =
(u∗ )k H0 (·)uk for all k ∈ Z.
In particular, W k (u∗ )k ∈ C for all k ∈ Z and [W ]E = 0 in K1 (E).
(vi) Next calculation needed?
For u ∈ Der(H0 (D), J) and S, T from (iii) – depending on a chosen
closed separable subgroup G of U(Der(H0 (D), J)) that contains u – holds
v[u]∗ H0 (·)v[u] = u∗ H0 (·)u, where

v[u] := (T uT ∗ + SS ∗ ) .

Two unitaries u1 and u2 in G ⊆ Der(H0 (D), J) have same class [u1 ] =

[u2 ] in K1 (Der(H0 (D), J)), if and only if, v[u1 ] and v[u2 ] are homotopic
in U(Der(H0 (D), J))
588 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

????
Is it at least true modulo the closed ideal of Der(H0 (D), J) generated
by the annihilators of H0 (D)?
Homotopy not in U(Der(H0 (D), J)) ??
Because the S is not in Der(H0 (D), J).
I.e., the (multiple-valued) map u 7→ v[u] ∈ U(E) has the prop-
erty that v[u1 ]∗ v[u2 ] ∈ U0 (Der(H0 (D), J)), if and only if, u1 , u2 ∈
U(Der(H0 (D), J)) have the same K1 -class in K1 (Der(H0 (D), J)).
(vii) If u ∈ U(Der(H0 (D), J)), then there exists a unitary w ∈ 1 + F ⊆ C and
k ∈ S(h0 ; D, E), such that (uw)∗ H0 (·)uw = k ⊕s0 ,t0 H0 .
Why not s2 , t2 in place of s0 , t0 ?
So far it is only shown that

(uw)∗ H0 (·)uv = (k + βh0 ) ⊕s0 ,t0 H0

for suitable v ∈ U(C) and k ∈ S(h0 ; D, E).

Notice that then one can replace uv by v ∗ uv. Is u∗ v ∗ uv ∈ U(C) ·
U(N (H0 (D), J)).
Is 1 ⊕ w ∈ U(C) · U(N (H0 (D), J)) if [w] = 0 ∈ K1 (Der(H0 (D), J))?
What about next? Typo?
(viii) For every u ∈ U(Der(H0 (D), J)) there exist w ∈ U(C) and k ∈
S(h0 ; D, E) with w∗ (u∗ H0 (·)u)w = k ⊕s0 ,t0 H0 , such that 1 − uw ∈
N (H0 (D), J)u
?????
?? What is required above? Typos?
(ix) The unitary group of Der(H0 (D), J) is invariant under forming u1 ⊕s,t u2
(modulo multiplication with unitaries in N (H0 (D), J) + 1 or in C).
The sum u1 ⊕s,t u2 corresponds to

(k1 ⊕s,t k2 ) + βh0

up to multiplication by a unitary v ∈ U(C) · U(N (H0 (D), J)) if k1 , k2 ∈

S(h0 ; D, E) are morphisms with

u∗j H0 (·)uj = (kj + βh0 ) ⊕s0 ,t0 H0 .

(x) If [v1 ] = [v2 ] ∈ K1 (Der(H0 (D), J)), then [k1 +βh0 ]+[h0 ] = [k2 +βh0 ]+[h0 ].
(xi) A C*-morphism k : D → J satisfies [k + βh0 ] + [h0 + βh0 ] = [h0 + βh0 ] if
and only if there exists a unitary v in 1 + N (H0 (D), J) such that

s2 k(·)s∗2 + (1 − p2 )H0 (·) = v ∗ H0 (·)v .

It implies [k] + [H0 ] = [H0 ].

(xii) πJ (u) ∈ U0 πJ (H0 (D))0 ∩ (E/J) , if and only if, there is path v(t) in

U0 (Der(H0 (D), J)) with v(0) ∈ 1 + J ⊆ 1 + N (H0 (D), J) and v(1) = u.

(xiii) If u = u1 u2 with u1 a unitary in C and u2 a unitary in the unitization
of N (H0 (D), J) then u∗ H0 (·)u = s2 k(·)s∗2 + (1 − p2 )H0 (·) for k : D → J
implies that [k + βh0 ] + [h0 ] = [h0 + βh0 ] .
(xiv) If k : D → J and [k + βh0 ] + [h0 ] = [h0 + βh0 ] , then [k] + [h0 ] = [h0 ] .
 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 589

(xv) If k : D → J and u ∈ U(E) are given with u∗ H0 (·)u = k(·) ⊕s2 ,t2 H0 (·),
then u ∈ Der(H0 (D), J).
What about the unitary W0 := s∗2 s0 + t∗2 t0 that transforms

k(·) ⊕s2 ,t2 H0 (·)

into k(·) ⊕s0 ,t0 H0 (·) ??

Is NOT useful as transformation
If u ∈ Der(H0 (D), J) then there exists a unitary w ∈ C := H0 (D)0 ∩E
with 0 = [wu] ∈ K1 (E) ( 8 ).
Then [k] + [h0 ] = [h0 ] if and only if there exist v ∈ C with vu ∈
N (H0 (D), J) + 1.
Part (xv) is shown in the below ????? given proof of Theorem 4.4.6.
What about the general relations between u, u0 ∈ U(E) for k and k 0
with [k 0 ] + [h0 ] = [k] + [h0 ]?
What about the unitary for [k1 + βh0 ] + [k2 + βh0 ]?
(xvi) The set P := U(C) · U(N (H0 (D), J) + C · 1) is an open subgroup of
U(Der(H0 (D), J)). U(N (H0 (D), J) + C · 1) is a closed normal subgroup
of U(Der(H0 (D), J)).
The set of u ∈ U(Der(H0 (D), J)) with the property that u ⊕s,t 1 ∈ P
is an open subgroup of U(Der(H0 (D), J)).
Each element of Der(H0 (D), J), N (H0 (D), J) + C · 1 and C are K1 -
surjective and satisfy that [u1 ] = [u2 ] if and only if (u∗1 u2 ) ⊕s,t 1 is in
U0 (Der(H0 (D), J)) (respectively in U0 (N (H0 (D), J)), or U0 (C)).
In particular:
If x = [u] ∈ K1 (Der(H0 (D), J)) for some u ∈ U(Der(H0 (D), J))
and if x = µ1 (y) + µ2 (z) for some y = [v1 ] ∈ K1 (C) and z =
[v2 ] ∈ K1 (N (H0 (D), J)), where µ1 : K1 (C) → K1 (Der(H0 (D), J))
and µ2 : K1 (N (H0 (D), J)) → K1 (Der(H0 (D), J)) are induced by the
canonical inclusion. Then (u∗ v1 v2 ) ⊕s,t 1 ∈ U0 (Der(H0 (D), J)).

more details? others? ??

Since N (H0 (D), J) is an ideal of Der(H0 (D), J) and is a full hereditary C *-
subalgebra of E and since C := H0 (D)0 ∩ E ⊆ Der(H0 (D), J) contains the copy
of C ∗ (s, t) ⊆ C of O2 , the natural maps from U(Der(H0 (D), J)) and U(E) into
K1 (Der(H0 (D), J)) and K1 (E) are surjective. We can use Lemma 4.2.20(ii) to
obtain that [u1 ] = [u2 ] ∈ K1 (Der(H0 (D), J)) if [u1 ] = [u2 ] ∈ K1 (E) and u∗2 u1 ∈
1 + N (H0 (D), J).

Proof. (o): Let T := s1 (1 − Q) + t1 Q for the isometries s1 , t1 as in Lemma

4.4.7(vi). Then T ∗ T = 1 and T ∗ H0 (·)T = QH0 (·) = k. Thus, [k] ∈ S(H0 ; D, E).
If moreover k(D) ⊆ J + βJ, then Lemma 4.4.16(iv) applies and k = k1 + βk2
with unique k1 , k2 ∈ S(h0 ; D, E).

8 Notice here that (wu)∗ H (·)wu = u∗ H (·)u.

0 0
590 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

(i): Let A := H0 (D) ⊆ E. Clearly Der(A, J) = πJ−1 πJ (A)0 ∩ (E/J) . Thus,

it is a C *-subalgebra of E and C := A0 ∩ E ⊃ Der(A, J). In the same way

N (A, J) = πJ−1 (Ann(πJ (A), E/J)), where Ann(πJ (A), E/J) means the two-sided
annihilators of the elements of πJ (A). Since Ann(πJ (A), E/J) is a closed ideal of
πJ (A)0 ∩ (E/J) , the C *-algebra N (A, J) is a closed ideal of Der(A, J).
Similar arguments work with J + β(J) in place of J. Thus, N (A, J + β(J)) is
a closed ideal of Der(A, J + β(J)).
Let e ∈ E with ea − ae ∈ J for all a ∈ A, i.e., e ∈ Der(A, J). Since A := H0 (D)
is separable, the C *-algebra M ⊆ J generated by the image of a ∈ A 7→ ea − ae
is separable, and M contains a strictly positive element f . Assumption (iii) of
Theorem 4.4.6, gives a projection p ∈ F ⊆ C with p0 ≤ p and pf = f = f p. Thus
px = x = xp for all x ∈ M , and it follows that e(1 − p), (1 − p)e ∈ C. Since p ∈ C
we get that e − pep ∈ C, and pepa − apep ∈ J for all a ∈ A.
Let u ∈ 1 + F ⊆ C with u∗ pu = p0 . Then pep = ugu∗ for g := p0 u∗ eup0
and g satisfies ∂g (a) = p0 u∗ (ea − ae)up0 ∈ J for ∂g (a) := ga − ag and a ∈ A, i.e.,
g ∈ Der(A, J).
By Lemma 4.4.16(ii), p0 is the sum p0 = q + r of properly infinite projections
q, r ∈ C with q ∈ N (A, J) and r ∈ N (A, βJ).
Since N (A, J) is an ideal of Der(A, J) ⊃ C and q ∈ N (A, J), we get that

gq = g(1 − r), qg = (1 − r)g, (1 − r)gr = qgr ∈ N (A, J) .

It follows g − rgr ∈ N (A, J) and rgr = g − (g − rgr) ∈ Der(A, J).

The C *-algebra N (A, β(J)) is a hereditary C *-subalgebra of E, and r ∈
N (A, β(J)). Thus, rgr ∈ N (A, β(J)). Since rgra − argr ∈ J ∩ βJ = {0} for
all a ∈ A, we get rgr ∈ C. Since u ∈ 1 + F ⊆ C, it follows urgru∗ ∈ C.
Let e1 := (e − pep) + urgru∗ and e2 := u(g − rgr)u∗ . Since pep = ugu∗ , we
get e = e1 + e2 . Above we have seen that e1 ∈ C and e2 ∈ N (A, J).

We consider Der(A, J + βJ) and – more generally – the case where X is a

countable subset of Der(A, J + βJ): Every e ∈ X satisfies ∂e (a) := ea − ae ∈ J + βJ
for all a ∈ A, and the linear map ∂e is the sum ∂e = Te + βSe , where Te , Se : A → J
are bounded linear maps, because J ∩ βJ = {0}. Let G denote a separable C *-sub-
algebra of J ⊆ βJ that contains the separable subsets Te (A) ∪ Se (A) for all e ∈ X
and satisfies β(G) = G. Let f ∈ G+ a strictly positive contraction of G. There
is a projection p ∈ F ⊆ C with p ≥ p0 and pf = f by 4.4.6(iii). It follows that
ea − ae = pepa − apep and e − pep ∈ C for all e ∈ X. By 4.4.6(v), there is a unitary
w ∈ 1 + F ⊆ C = A0 ∩ E with w∗ pw = p0 .
It follows that p∂e (a) = ∂e (a) = ∂e (a)p = p∂e (a)p = ∂pep for a ∈ A and e ∈ X.
In particular, (1 − p)∂e (A) = {0} = ∂e (A)(1 − p) for all e ∈ X.
Recall that F ⊆ N (A, J + βJ) ∩ C and that N (A, J + βJ) is a hereditary
C *-subalgebra of E. Thus pep ∈ N (A, J + βJ) for each orthogonal projection
 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 591

p ∈ N (A, J + βJ) and e ∈ E. Since p ∈ F ⊆ N (A, J + βJ) and w ∈ 1 + F ⊆ C,

we get that e − pep ∈ C and pep ∈ N (A, J + βJ) for all e ∈ X. In particular,
Der(A, J + βJ) = C + N (A, J + βJ).
(ii): If [k] + [H0 ] = [H0 ] then [k ⊕s2 ,t2 H0 ] = [H0 ]. There exists u ∈ U(E) with

u∗ H0 (·)u = k ⊕s2 ,t2 H0 = s2 k(·)s∗2 + (1 − p2 )H0 (·) ,

where p2 , s2 and t2 are as in Lemma 4.4.16(i). It follows ut2 ∈ C, because t2 ∈ C,

(1 − p2 ) = t2 t∗2 and H0 (·)u(1 − p2 ) = u(1 − p2 )H(·) . Furthermore, up2 u∗ commutes
with H0 (D) and, for a ∈ D, up2 u∗ H0 (a) = us2 k(a)s∗2 u∗ ⊆ J. In particular,
k(a) = s∗2 u∗ H0 (a)us2 . Recall that p2 H0 (·) = s2 h0 (·)s∗2 .
We get u∗ H0 (a)u − H0 (a) = s2 (k(a) − h0 (a))s∗2 ∈ J for all a ∈ D, because
k(D) ⊆ J. Thus, u ∈ Der(H0 (D), J).
Clearly v ∗ H0 (·)v = s0 k(·)s∗0 + (1 − p0 )H0 (·) with v := u(s2 s∗0 + t2 t∗0 ) . Then
vt0 ∈ C and v ∈ Der(H0 (D), J + β(J)).
Since p2 , p0 ∈ N (H0 (D), J + βJ) and s2 s∗0 = p2 (s2 s∗0 )p0 , the partial isometry
∗
s2 s0 is in Der(H0 (D), J + β(J)) but is not in Der(H0 (D), J), because p0 = s0 s∗0 =
(s2 s∗0 )∗ s2 s∗0 is not in Der(H0 (D), J).
Next also (xv):
If [k] + [h0 ] = [h0 ] in S(h0 ; D, E), then u = w1 u1 with u1 ∈ 1 + N (H0 (D), J)
and w1 ∈ C.
(iii): Let Y ⊆ U(E) an at most countable subset of the unitaries in E, and
suppose that u ∈ Der(H0 (D), J + βJ) for all u ∈ Y . If G is the (norm-)closure
of the subgroup of U(E) generated by Y , then G ⊆ Der(H0 (D), J + βJ), because
Der(H0 (D), J + βJ) is a C *-algebra by Part (i).
If we take for X in the proof of Part (i) the sub-group of G that is algebraical
generated by Y , then we get a projection p ∈ F ⊆ C and a unitary w ∈ 1 + F ,
such that p ≥ p0 , u − pup ∈ C for all u ∈ G, and w∗ pw = p0 . Moreover pup ∈
N (H0 (D), J + βJ) for all u ∈ G, because w ∈ C, p0 ∈ N (H0 (D), J + βJ) and
N (H0 (D), J + βJ) is an ideal of Der(H0 (D), J + βJ).
It follows that (1 − p)u∗ H0 (·)u = (1 − p)H0 (·) for all u ∈ G. This implies
that (1 − p0 )(uw)∗ H0 (·)uw = (1 − p0 )H0 (·), and Q := uwp0 (uw)∗ is a projection
in C, i.e., p0 commutes with (uw)∗ H0 (·)uw . Let k(a) := s∗0 (uw)∗ H0 (a)uws0 , i.e.,
s0 k(a)s∗0 = p0 (uw)∗ H0 (a)uw, then k ∈ S(H0 ; D, E) by Part (o) and

(uw)∗ H0 (·)uw = k ⊕s0 ,t0 H0 .

Since

(1 − p0 ) · (uw)∗ H0 (·)uw − H0 (·) = (1 − p0 )(wu)∗ ∂wu H0 (·) = 0 ,

p0 H0 (·) = s0 h0 (·) + βh0 (·) s∗0 and uw ∈ Der(H0 (D), J + βJ) it follows that, for

a ∈ D,

s0 k(a) − h0 (a) − βh0 (a) s∗0 = (uw)∗ ∂uw (H0 (a)) ∈ J + βJ .

592 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Thus, k(a) ∈ J + βJ for a ∈ D. Since k ∈ S(H0 ; D, E) it implies k = k1 + βk2 for

unique k1 , k2 ∈ S(h0 ; D, E) by Lemma 4.4.7(x).
If moreover u ∈ Der(H0 (D), J), then the above arguments show that k(a) −
h0 (a) − βh0 (a) ∈ J for each a ∈ D. This implies k2 = h0 because J ∩ βJ = {0}.
(iv): Let G a separable (norm-) closed subgroup of U(Der(H0 (D), J + βJ)).
By the proof of Part (iii) there exist a projection p ∈ F ⊆ C, and a unitary
w ∈ 1 + F such that p ≥ p0 , u − pup ∈ C for all u ∈ G, and w∗ pw = p0 . Moreover
pup ∈ N (A, J + βJ) for all u ∈ G. It follows that (1 − p)u∗ H0 (·)u = (1 − p)H0 (·)
for all u ∈ G. It follows that there exists an isometry R with R∗ (1 − p)R = 1, i.e.,
RR∗ ≤ 1 − p, and with H0 (D)RR∗ = {0}:
Indeed, by Lemma 4.4.7(vi), there exist isometries s1 , t1 ∈ E with t1 ∈ C,
s1 s∗1
+ t1 t∗1 = 1, p0 t1 = p0 = t1 p0 , s1 s∗1 H0 (·) = 0 and t1 H0 (·) = H0 = H0 (·)t1 .
It implies s1 s∗1 = 1 − t1 t∗1 ≤ 1 − p0 . Since w ∈ C and wp0 w∗ = p, the isometry
R := ws1 satisfies RR∗ ≤ 1 − p = w(1 − p0 )w∗ , and R∗ H0 (·)R = s∗1 H0 (·)s1 = 0.
Then RER∗ ⊥ H0 (D), and for all u ∈ G,

R∗ u∗ H0 (·)uR = R∗ (1 − p)u∗ H0 (·)uR = R∗ (1 − p)H0 (·)R = 0 .

Let S := Rs and T := (1 − RR∗ ) + RtR∗ . Then S and T are isometries in

E, T T ∗ = (1 − RR∗ ) + Rtt∗ R∗ , and SS ∗ ≤ RR∗ ⊥ u∗ H0 (D)u, T u∗ H0 (·) = (1 −
RR∗ )u∗ H0 (·) = u∗ H0 (·), H0 (·)uT = H0 (·)u for all u ∈ G, and SS ∗ + T T ∗ =
(1 − RR∗ ) + Rss∗ R∗ + Rtt∗ R∗ = 1.
(v): Let S, T ∈ E the isometries from Part (iv). It is well-known that Wu :=
Su S ∗ + T uT ∗ is in U0 (E), and straight calculation shows Wu∗ H0 (·)Wu = u∗ H0 (·)u
∗

for all u ∈ G. It follows uWu∗ ∈ C and [Wu ]E = 0 in K1 (E).

(vi): For any u ∈ E holds v ∗ H0 (a)v = u∗ H0 (a)u for v := (SuS ∗ + T T ∗ ).
Then w := vu∗ = SuS ∗ u∗ + T T ∗ u∗ is in C.
TYPO ????????
Unitaries u1 and u2 in Der(H0 (D), J) have same class in K1 (Der(H0 (D), J)),
if and only if, the corresponding – in this way defined – v1 and v2 are homotopic
in U(Der(H0 (D), J)).
(I.e., the new sort of v ∈ U0 (E) has the property that, any two of
them are homotopic in U(Der(H0 (D), J)) if they have the same K1 -class in
K1 (Der(H0 (D), J)).)
This implies ????
(vii): To be shown: If u ∈ U(Der(H0 (D), J)), then there exists a unitary
v ∈ 1 + F ⊆ C and k ∈ S(h0 ; D, E), such that with w = uv holds

w∗ H0 (·)w = (k + βh0 ) ⊕s0 ,t0 H0 .

Compare with k ⊕s2 ,t2 H0 ??

(viii): To be shown: ??????
 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 593

For every u ∈ U(Der(H0 (D), J)) there exist w ∈ U(C) and k ∈ S(h0 ; D, E)
with w∗ (u∗ H0 (·)u)w = k ⊕s0 ,t0 H0 , such that 1 − uw ∈ N (H0 (D), J)u ?????.
It would imply: u∗ − w ∈ u∗ N (H0 (D), J)u = N (H0 (D), J). It is desirable,
but still not prove-able. ????
(xiv): [k + βh0 ] + [h0 ] = [h0 + βh0 ], is the same as [k] + [h0 + βh0 ] = [h0 + βh0 ]
by Lemma 4.4.7(i). Since [k]+[h0 +βh0 ] = [k ⊕s,t (h0 +βh0 )], there exists a unitary
u ∈ E with
(u∗ sk(a)s∗ u + u∗ th0 (a)t∗ u − h0 (a)) + (u∗ tβh0 (a)tu − βh0 (a)) = 0
for all a ∈ D. Since J ∩ β(J) = Jβ(J) = {0} it follows u∗ sk(·)s∗ + th0 (·)t∗ u = h0 ,

i.e., [k] + [h0 ] = [h0 ] .

(xv): Let k : D → J and u ∈ U(E) with u∗ H0 (·)u = k(·) ⊕s2 ,t2 H0 (·) .
Then k = s∗2 u∗ H0 (·)us2 ∈ S(H0 ; D, E) and k ∈ S(h0 ; D, E) by Lemma
4.4.7(x).
Suppose [k] + [h0 ] = [h0 ], ....?????
Since s, t ∈ h0 (D)0 ∩ E for the isometries s, t in Theorem 4.4.6(i), we get
W0 (h0 ⊕s2 ,t2 h0 )W0∗ = h0 for the unitary W0 := ss∗2 + tt∗2 .
If [k] + [h0 ] = [h0 ], then there is a unitary V0 ∈ E with V0∗ (k ⊕s2 ,t2 h0 )V0 =
h0 = p2 H0 (·) . The unitary V1 := V0 W0 satisfies
V1∗ (k ⊕s2 ,t2 h0 )V1 = h0 ⊕s2 ,t2 h0 .
The unitary T := (s2 )2 s∗2 + s2 t2 s∗2 t∗2 + t2 (t∗2 )2 satisfies for all X, Y, Z ∈ E that



T X ⊕s2 ,t2 (Y ⊕s2 ,t2 Z) = (X ⊕s2 ,t2 Y ) ⊕s2 ,t2 Z T .
Notice that H0 = h0 ⊕s2 ,t2 H0 . It implies
H0 = h0 ⊕ (h0 ⊕ H0 ) = T ∗ ((h0 ⊕ h0 ) ⊕ H0 )T
Let W1 := V1 ⊕ 1 and v := T ∗ W1∗ T = T ∗ (V1∗ ⊕ 1)T . Then rv = r = vr for
r := T ∗ t2 t∗2 T = t22 (t22 )∗ = t2 (1 − p2 )t∗2 , 1 − r = p2 + t2 p2 t∗2 is in C ∩ N (H0 (D), J),
because the latter is an ideal of C and t2 ∈ C. It implies t2 H0 (D)p2 t∗2 ⊆ J. Thus
v ∈ 1 + (1 − r)E(1 − r) ⊆ 1 + N (H0 (D), J), and satisfies:
v ∗ H0 v = v ∗ T ∗ ((h0 ⊕ h0 ) ⊕ H0 )T v = T ∗ W1 ((h0 ⊕ h0 ) ⊕ H0 )W1∗ T =
T ∗ (k ⊕ h0 ) ⊕ H0 T = k ⊕ (h0 ⊕ H0 ) = k ⊕s2 ,t2 H0 .

It follows that uv ∗ = w ∈ C, i.e., that u is the product wv of a unitary v ∈

1 + N (H0 (D), J) and a unitary w ∈ C.
Check again:
Now we consider the opposite direction: Suppose that u = wv with v a unitary
in 1 + N (H0 (D), J) and w ∈ U(C). Then
v ∗ H0 (·)v = k(·) ⊕s2 ,t2 H0 (·) .
and there is normal g ∈ N (H0 (D), J) with v = g + 1. The sets H0 (D)g and
∂v (H0 (D)) = ∂g (H0 (D)) together generate a separable C *-subalgebra of J. By
594 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Theorem 4.4.6(ii) there exists p ∈ F with p ≥ p0 , p 6= p0 such that p(H0 (a)g −

gH0 (a))p = H0 (a)g −gH0 (a) and H0 (·)gp = H0 (·)g = H0 (·)pg. If we add to the last
equation H0 (·)p, then we get H0 (·)vp = H0 (·)pv. It follows v ∗ H0 (·)vp = v ∗ H0 (·)pv,
H0 (·)pvp − pvpH0 (·) = H0 (·)v − vH0 (·) and v ∗ H0 (·)v(1 − p) = H0 (·)(1 − p). Since
p = p0 + r for some r 6= 0, it follows r = rt0 t∗0 , v ∗ H0 (·)pv = v ∗ H0 (·)vp = s2 k(·)s∗2 +
rH0 (·). By Theorem 4.4.6(v), there exist unitaries W1 , W2 ∈ 1 + F with W1∗ pW1 =
p0 and W2∗ rW2 = p0 . Then H0 (·)p = W1 H0 (·)p0 W1∗ and H0 (·)r = W2 (H0 (·)p0 )W2∗ .
Recall that H0 (·)p0 = s0 (h0 + βh0 )s∗0 . It gives

[h0 + βh0 ] + [0] = [k] + [h0 + βh0 ] + [0] .W2 H0 (·)p0 W2∗ t2 )t∗2

By Lemma 4.4.7(ii,i), this is equivalent to [h0 + βh0 ] = [k + βh0 ] + [h0 ] It follows

[h0 ] = [k] + [h0 ] by Part (xiv).
?? to be filled in 

We study at first the case, where the OLD !!!! assumptions (I) and (II)
are satisfied (in addition to (i)-(vii)):
If u ∈ Der(H0 (D), J) is unitary, then, by Lemma 4.4.17(v), there exists w ∈ C
such that u∗ H0 (a)u = (wu)∗ H0 (a)(wu) for all a ∈ A and wu ∈ U0 (E).
Then by assumption (I) there exists a unitary v ∈ C ∩ U0 (E) such that for
u0 = wuv ∈ U0 (E)

(u0 − 1)β(J) = {0} .

to be filled in ?? ??????????????????????
Start:
towards isomorphism of kernel and Γ
Now we take a more K-theoretic approach for the study of the general situ-
ation:
Let C2 := h0 (D)0 ∩ E. Then G(h0 ; D, E) ∼
= ker(K0 (C2 ) → K0 (E)) by
??????????? ?? from Lemma 4.2.20 ???
It holds C2 ∼
= p2 Cp2 via the *-isomorphism b ∈ C2 7→ s2 bs∗2 .
The subalgebras Ann(H0 (D), E) and C ∩ N (H0 (D), J) are ideals of C, and
Ann(H0 (D), E) ⊆ C ∩ N (H0 (D), J).
The projection p2 + Ann(H0 (D), E) is properly infinite in C/ Ann(H0 (D), E),
which is a full C *-subalgebra of

J2 := (C ∩ N (H0 (D), J))/ Ann(H0 (D), E) .

Since H0 dominates zero, we get that Ann(H0 (D), E) is a full hereditary C *-

subalgebra of E. Now Lemma 4.2.20 shows that

G(H0 ; D, E) ∼
= K0 (C/ Ann(H0 (D))) ∼
= ker(K0 (C) → K0 (E)) .
 4. GROUPS DEFINED BY ABSORBING C *-MORPHISMS 595

The same happens with Ann(h0 (D), E) ⊆ C2 in E. Moreover, s2 bs∗2 ∈

Ann(H0 (D), E), if and only if, b ∈ Ann(h0 (D), E). Let

A := C/ Ann(H0 (D), E) .

The algebra J2 is a closed ideal of A with A/J2 ∼

= C/(C ∩ N (H0 (D), J)) and the
natural map G(h0 ; D, E) → G(H0 ; D, E) transforms to

K0 (C2 / Ann(h0 (D), E)) ∼

= K0 (J2 ) → K0 (A) .

The surjectivity of G(h0 ; D, E) → G(H0 ; D, E) is equivalent to the exactness

of the sequence

K1 (A) → K1 (A/J2 ) →∂ K0 (J2 ) → K0 (A) → 0 .

One can use the identity Der(H0 (D), J) = C + N (H0 (D), J), i.e., that

A/J2 ∼
= Der(H0 (D), J)/ N (H0 (D), J) ,
the natural map K1 (C) → K1 (Der(H0 (D), J)) and the splitting (by Lemma 4.2.20)
of the exact sequence

0 → K1 (N (H0 (D), J)) → K1 (Der(H0 (D), J)) → K1 (A/J2 ) → 0 ,

to verify that the quotient K1 (A/J2 )/∂(K1 (A)) of K1 (A/J2 ) by the image of
∂ : K1 (A) → K1 (A/J2 ) is isomorphic to Γ(H0 (D), J, E) as defined before Theorem
4.4.6.
It seems likely that the following argument gives that the natural map K1 (A) →
K1 (A/J2 ) is zero, i.e., that K1 (J2 ) → K1 (A) is surjective too. The only difficult
point in such an attempt is the verification of assumption (ii) of Theorem 4.4.6 for
the “new” system.
Let P∞ denote the unique pi-sun algebra in the UCT-class with K0 (P∞ ) ∼ = {0}
∼
and K1 (P∞ ) = Z If we repeat all arguments above with E, J, β, F , H0 , h0 , s,
t, s0 , t0 , and u1 , but replaced by E ⊗ P∞ , J ⊗ P∞ , F ⊗ 1, β ⊗ id, H0 (·) ⊗ 1,
h0 (·) ⊗ 1, s ⊗ 1, . . ., then one gets also that K1 (J2 ) → K1 (A) is surjective, provided
assumption (ii) of Theorem 4.4.6 can be verified by the new system, at least in a
suitable approximate or asymptotic sense.

Remark 4.4.18. One can show that the following conditions (I) and (II) to-
gether imply that [k ⊕ h0 ] 7→ [k ⊕ H0 ] is injective.

(I) If u ∈ U0 (E) is in C + N (H0 (D), J) then there exists a unitary w ∈

C ∩ U0 (E) such that (wu − 1)β(J) = {0}.
Alternatively(?): ... such that vu ∈ 1 + N (H0 (D), J +
β(J)).
(II) If v ∈ U0 (E) is in C + N (H0 (D), J), vp0 v ∗ ∈ C ∩ N (H0 (D), J) and
(v − 1)β(J) = {0}, then k(·) := (vp0 v ∗ )H0 (·) satisfies

[k] + [h0 + βh0 ] = [h0 + βh0 ] .

596 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

5. Generalizations and limitations

The following remarks and corollaries discuss some limitations of the arguments
in above given proofs and generalize them partly for “soft” asymptotic considera-
tions.
We need in later chapters local estimates for an approximate version of Propo-
sition 4.3.5(i). The following remark and corollary give estimates in very special
situations, well applicable e.g. to singly generated C *-algebras D.

Remark 4.5.1. Let h1 : D → E a C *-morphism and s1 , s2 , t ∈ E isometries

with s1 s∗1 + s2 s∗2 = 1.
Define a c.p. contraction h2 : D → E by h2 (a) := t∗ h1 (a)t for a ∈ D and
isometries t1 := (1 − tt∗ ) + ts1 t∗ , t2 := ts2 in E.
Let C(s1 , s2 ; y) := k [s1 , y] k + k [s2 , y] k denote the sum of norms of the
commutators of y ∈ E with s1 and s2 , and define by

1/2
curv(h2 , a) := max k h2 (a∗ a) − h2 (a∗ )h2 (a) k , k h2 (aa∗ ) − h2 (a)h2 (a∗ ) k

the local curvature of the completely positive contraction h2 at a ∈ D, i.e., the

failure of h2 to be multiplicative at the elements a and a∗ .
Then the isometries t1 , t2 satisfy t1 t∗1 + t2 t∗2 = 1 and for the Cuntz sum holds


kh1 (a) − (h1 ⊕t1 ,t2 h2 )(a)k ≤ C s1 , s2 ; h2 (a) + curv(h2 , a) .

If h3 : D → E is another C *-morphism, then we denote by

var(h2 , h3 ; a) := max{ kh2 (x) − h3 (x)k ; x = a, a∗ a, aa∗ } ,

the local variation of h2 and h3 at a ∈ D, i.e., the distance of h2 and h3 on

{a, a∗ , a∗ a, aa∗ } .
The C *-morphism h1 ⊕t1 ,t2 h3 : D → E satisfies, for a contraction a ∈ E, that

kh1 (a) − (h1 ⊕t1 ,t2 h3 )(a)k

≤ C s1 , s2 ; h3 (a) + C s1 , s2 ; h3 (a) − h2 (a) + 4 · var(h2 , h3 ; a)1/2

≤ C s1 , s2 ; h3 (a) + 10 · var(h2 , h3 ; a)1/2 .

Proof. Obviously, the elements t1 := (1 − tt∗ ) + ts1 t∗ and t2 := ts2 of E are

isometries with t1 t∗1 + t2 t∗2 = 1. Let p := tt∗ ,


X(a) := h1 (a) − h1 (a) ⊕t1 ,t2 h2 (a)

and recall that h2 := t∗ h1 (·)t.

Thus, ph1 (a)p = th2 (a)t∗ , pt1 h1 (a)t∗1 p = ts1 h2 (a)s∗1 t∗ and

pX(a)p = th2 (a)t∗ − ts1 h2 (a)s∗1 t∗ − ts2 h2 (a)s∗2 t = t∆(a)t∗

for ∆(a) := h2 (a) − (h2 ⊕s1 ,s2 h2 )(a). Clearly, for y ∈ E and qk := sk s∗k ,

ky−(y⊕s1 ,s2 y)k ≤ kyq1 −s1 ys∗1 k+kyq2 −s1 ys∗1 k = kys1 −s1 yk+kys2 −s2 yk ≤ 4kyk ,
 5. GENERALIZATIONS AND LIMITATIONS 597

because q1 + q2 = 1, s∗j qk = δj,k sj and kys∗j k = kyk. It follows that

kpX(a)pk ≤ k∆(a)k ≤ C s1 , s2 ; h2 (a) .

The equations (1 − p)t1 = (1 − p) = t∗1 (1 − p) and pt1 = ts1 , pt2 = t2 = ts2 show
that
p · (h1 (a) − X(a)) = t(h2 (a) − ∆(a))t∗ = (h1 (a) − X(a)) · p ,
(1 − p)X(a)(1 − p) = (1 − p)(h1 (a) − t1 h1 (a)t∗1 )(1 − p) = 0, and that

pX(a)(1 − p) + (1 − p)X(a)p = ph1 (a)(1 − p) + (1 − p)h1 (a)p .

If we use that kx ± yk = max(kxk, kyk) if x∗ y = 0 = xy ∗ , then we get for z ∈ E

that

k[z, p]k = kpz(1 − p) − (1 − p)zpk =

kpz(1 − p) + (1 − p)zpk = max( kpz(1 − p)k , k(1 − p)zpk ) .

It follows that above estimates sum up to

kX(a)k ≤ C s1 , s2 ; h2 (a) + max kph1 (a)(1 − p)k, k(1 − p)h1 (a)pk .

Use now that for p := tt∗ and h2 := t∗ h1 (·)t holds

kh2 (a∗ a) − h2 (a)∗ h2 (a)k = k(1 − p)h1 (a)pk2 .

We obtain

curv(h2 , a) = max k ph1 (a)(1 − p) k , k (1 − p)h1 (a)p k = k[h1 (a), tt∗ ]k .

This implies the proposed estimate C s1 , s2 ; h2 (a) + curv(h2 , a) of kh1 (a) −
(h1 ⊕t1 ,t2 h2 )(a)k .
To verify the estimate of the norm of h1 (a) − (h1 ⊕t1 ,t2 h3 )(a) consider following
inequalities:

kh1 (a) − (h1 ⊕t1 ,t2 h3 )(a)k ≤ kh3 (a) − h2 (a)k + kh1 (a) − (h1 ⊕t1 ,t2 h2 )(a)k ,




C s1 , s2 ; h2 (a) ≤ C s1 , s2 ; h3 (a) + C s1 , s2 ; h2 (a) − h3 (a)


and C s1 , s2 ; h2 (a) − h3 (a) ≤ 4kh2 (a) − h3 (a)k .

A rough estimate of curv(h2 , a)2 can be seen from

kh2 (a∗ a) − h2 (a)∗ h2 (a)k ≤ kh3 (a∗ a) − h2 (a∗ a)k + kh3 (a)∗ h3 (a) − h2 (a)∗ h2 (a)k
≤ (2kak + 1) · var(h2 , h3 ; a) .

Since var(h2 , h3 ; a∗ ) = var(h2 , h3 ; a), we get curv(h2 , a)2 ≤ 3 · var(h2 , h3 ; a) for

kak ≤ 1.
Notice that
√ √
kh3 (a) − h2 (a)k ≤ 2 · kh3 (a) − h2 (a)k1/2 ≤ 2 · var(h2 , h3 ; a)1/2 ,

for all contractions a ∈ E, in particular, 4kh3 (a) − h2 (a)k ≤ 6 · var(h2 , h3 ; a)1/2 .

It leads in case kak ≤ 1 to the proposed estimates

C s1 , s2 ; h3 (a) + C s1 , s2 ; h3 (a) − h2 (a) + 4 · var(h2 , h3 ; a)1/2

598 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

and the bigger upper estimate

C s1 , s2 ; h3 (a) + 10 · var(h2 , h3 ; a)1/2

for the norm of h1 (a) − (h1 ⊕t1 ,t2 h3 )(a) . 

The proof of the following corollary is based on the estimates in Remark 4.5.1.

Corollary 4.5.2. If T, S ∈ E are isometries and h, k : D → E are C*-

morphisms, then there exists a unitary U ∈ E ⊗ O2 such that

kU ∗ (h(a) ⊗ 1)U − k(a) ⊗ 1k ≤ 8 · µ(S, T ; h, k; a)1/2 , ∀ a ∈ D, (5.1)

where µ(S, T ; h, k; a) denotes – for fixed a ∈ D – the maximum of

{kS ∗ h(x)S − k(x)k , kT ∗ k(x)T − h(x)k ; x = a, a∗ a, aa∗ } .

The Inequality (5.1) holds even if the unit element of E is not necessarily
properly infinite, i.e., no orthogonality of SS ∗ and T T ∗ is required, i.e., it is not
necessary that C ∗ (S, T ) is an image of E2 .

Proof. Given h, k, S, T , consider the C *-morphisms

h1 := h(·) ⊗ 1 and k1 := k(·) ⊗ 1 ,

and define the “linking” completely positive contractions

h2 := (S ∗ h(·)S) ⊗ 1 and k2 := (T ∗ k(·)T ) ⊗ 1 .

We can take s1 := 1⊗r1 ∈ E ⊗O2 and s2 := 1⊗r2 in Remark 4.5.1 for the canonical
generators r1 , r2 of O2 = C ∗ (r1 , r2 ). Then s1 and s2 commute with the elements
of hi (D) ∪ kj (D), i, j ∈ {1, 2}, and the estimates in Remark 4.5.1 apply to the
C *-morphisms h1 , h3 := k1 and the c.p. contraction h2 = t∗ h1 (·)t for t := S ⊗ 1,
respectively to k1 , k2 = t∗ k1 (·)t for t := T ⊗ 1 and k3 := h1 (in place of h1 , t, h2 , h3
in Remark 4.5.1).




Notice that C s1 , s2 ; hi (a) , C s1 , s2 ; kj (a) and C s1 , s2 ; hi (a) − kj (a) are
zero for i, j = 1, 2, and that the estimates in Remark 4.5.1 changes therefore to
simpler ones, as e.g. to

kh1 (a) − (h1 ⊕t1 ,t2 k1 )(a)k ≤ 4 · var(h2 , k1 ; a)1/2 ,

and – with interchanged roles –

kk1 (a) − (k1 ⊕t3 ,t4 h1 )(a)k ≤ 4 · var(k2 , h1 ; a)1/2 ,

where we let t1 := (1 − SS ∗ ) ⊗ 1 + SS ∗ ⊗ r1 , and t2 := S ⊗ r2 (build by t := S ⊗ 1

from sk := 1 ⊗ rk ), and t3 := (1 − T T ∗ ) ⊗ 1 + T T ∗ ⊗ r1 and t4 := T ⊗ r2 (build by
t := T ⊗ 1 from sk := 1 ⊗ rk ).
 5. GENERALIZATIONS AND LIMITATIONS 599

If U := t1 t∗4 + t2 t∗3 then U (k1 ⊕t3 ,t4 h1 ) = (h1 ⊕t1 ,t2 k1 )U , and we obtain the
estimate
kU ∗ h1 (a)U − k1 (a)k ≤
kh1 (a) − (h1 ⊕t1 ,t2 k1 )(a)k + kk1 (a) − (k1 ⊕t3 ,t4 h1 )(a)k ≤
4 · var(h2 , k1 ; a)1/2 + 4 · var(k2 , h1 ; a)1/2 ≤
8 · µ(S, T ; h, k; a)1/2 .
Here we have used that
µ(S, T ; h, k; a) = max( kk2 (x) − h1 (x)k, kh2 (x) − k1 (x)k ; x = a, a∗ a, aa∗ ) ,
i.e., that


µ(S, T ; h, k; a) = max var(h2 , k1 ; a), var(k2 , h1 ; a) .
Notice finally that kU ∗ (h(a) ⊗ 1)U − k(a) ⊗ 1k = kU ∗ h1 (a)U − k1 (a)k . 

Remark 4.5.3. The methods of estimates in the proofs of Remark 4.5.1 and
Corollary 4.5.2 can be generalized to prove the following more flexible and general
result (with different U ∈ E):
If s1 , s2 , s01 , s02 ∈ E are isometries, s1 s∗1 +s2 s∗2 = 1 = s01 (s01 )∗ +s02 (s02 )∗ , h, k : D →
E are C *-morphisms, and S, T ∈ E are isometries, then there exists a unitary
U ∈ E that satisfies for all contractions a ∈ D the inequality
kU ∗ h(a)U − k(a)k ≤ C s1 , s2 ; h(a) + C s01 , s02 ; k(a) + 20 · µ(S, T ; h, k; a)1/2 ,

where µ(S, T ; h, k; a) is the maximum of

{kS ∗ h(x)S − k(x)k , kT ∗ k(x)T − h(x)k ; x = a, a∗ a, aa∗ } ,
as defined in Corollary 4.5.2.

Proof. Let h1 := h, h2 := S ∗ h(·)S, and h3 := k and use the Cuntz addition

⊕0 := ⊕t1 ,t2 with isometries t1 , t2 build by (s01 , s02 , S) in place of (s1 , s2 , t) in Remark
4.5.1, then we get that
kh(a) − (h(a) ⊕0 k(a))k ≤ C s01 , s02 ; k(a) + 10 · var(S ∗ hS, k; a)1/2 .

If we take for Remark 4.5.1 h1 := k, h2 := T ∗ k(·)T , h3 := h and use ⊕ := ⊕t1 ,t2

with t1 , t2 build by (s1 , s2 , T ) in place of (s1 , s2 , t) in Remark 4.5.1, then we obtain
the estimate:
kk(a) − (k(a) ⊕ h(a))k ≤ C s1 , s2 ; h(a) + 10 · var(T ∗ kT, h; a)1/2 .

Now we use that

µ(S, T ; h, k; a) = max var(S ∗ hS, k; a), var(T ∗ kT, h; a)

and take a unitary U ∈ E with U ∗ (x ⊕0 y)U = y ⊕ x for x, y ∈ E, then we get the

upper estimate
C s01 , s02 ; k(a) + C s1 , s2 ; h(a) + 20 · µ(S, T ; h, k; a)1/2

of kU ∗ h(a)U − k(a)k. 
600 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Remark 4.5.4. Let A := H0 (D) and C := A0 ∩ E. The algebras N (A, J),

β(N (A, J)) = N (A, βJ) and N (A, J + βJ) are hereditary C *-subalgebras of E,
N (A, J)∩N (A, βJ) = Ann(A, E) (because J ∩βJ = 0), and N (A, J)∪N (A, βJ) ⊆
N (A, J + βJ) .
The elements s2 , s3 ∈ E defined check ref!! Part (x) of Lemma 4.4.7 have
the property that the partial isometry Z := s2 s∗3 satisfies Z ∗ Z ∈ N (A, βJ), ZZ ∗ ∈
N (A, J), and Z ∗ Z, ZZ ∗ 6∈ Ann(A, J). Thus, Z ∈ N (A, J + βJ) \ Ann(A, J).
Conjecture: The partial isometry Z is not contained in the smallest β-invariant
C *-subalgebra of N (A, J + βJ) generated by N (A, J), i.e.,

Z 6∈ C ∗ (N (A, J) ∪ N (A, βJ)) .

A positive answer seems likely, because otherwise there exist n ∈ N,

a1 , . . . , an , b1 , . . . , bn ∈ N (A, J) such that Z = b∗1 β(a1 ) + . . . + b∗n β(an ).

Remark 4.5.5. The observations in Part (i) of Lemma 4.4.7 and the assump-
tions of Theorem 4.4.6 together are not enough to to give a general proof that
[k1 ] = [k2 ] and k1 (B) ⊆ J imply [k1 + βk3 ] = [k2 + βk3 ] for all C *-morphisms
kj : B → J, j ∈ {1, 2, 3}, of an arbitrary separable C *-algebra B.
But, since [k1 ] = [k2 ] and k1 (B) ⊆ J together imply that there exists u ∈
U0 (E) with u∗ k1 (·)u = k2 by Lemma 4.4.7(i), one can construct, with help of
suitable paths of positive contraction in J+ , a norm-continuous path of unitaries
U (ξ) ∈ U0 (J + C1) ⊂ U0 (E) for ξ ∈ [0, ∞) such that U (0) = 1 and

lim U (ξ)∗ k1 (b)U (ξ) = k2 (b) for all b ∈ B .

ξ→∞

Then U (ξ)c = c = cU (ξ) for all c ∈ βJ ⊂ Ann(J, E) and

lim U (ξ)∗ (k1 (b) + βk3 (b))U (ξ) = k2 (b) + βk3 (b) for all b ∈ B ,
ξ→∞

i.e., k1 + βk3 and k2 + βk3 are unitarily homotopic in E in the sense of Defini-
tion 5.0.1, which is sometimes weaker than k1 and k2 being homotopic or unitary
equivalent.
By symmetry – given by β – this shows also that for C *-morphisms kj : B → J,
j ∈ {1, 2, 3, 4}, the equations [k1 ] = [k2 ] and [k3 ] = [k4 ] imply that k1 + βk3 and
k2 + βk4 are unitarily homotopic in E.
We apply Theorem 4.4.6 mainly to special cases where the ideal J is moreover
sub-Stonean, i.e., has the property that, for each a, b ∈ J+ with ab = 0, there exist
contractions c, d ∈ J+ with ac = a, bd = b and cd = 0.
One can show that [k1 ] = [k2 ] and kj (B) ⊂ J for j ∈ {1, 2} imply [k1 + βk3 ] =
[k2 + βk3 ] for all C *-morphisms k3 : B → J if J is sub-Stonean in addition to the
other assumptions of Theorem 4.4.6.
It is not clear if the assumptions of Theorem 4.4.6 allow to deduce from [kj ] +
[h0 ] = [h0 ] (j ∈ {1, 2}) that [k1 + βk2 ] + [h0 + βh0 ] = [h0 + βh0 ] in the case where
J is not sub-Stonean.
 5. GENERALIZATIONS AND LIMITATIONS 601

Proof. If [k1 ] = [k2 ] and k1 (B) ⊆ J then there exists u ∈ U0 (E) with
∗
u k1 (·)u = k2 by Lemma 4.4.7(i).
∗
There are T1 , . . . , Tn ∈ E with Tm = −Tm for m = 1, . . . , n, i.e., with (iTm )∗ =
iTm in case of complex E, such that u = exp(T1 ) · . . . · exp(Tn ) .
We can define C *-morphisms hm : B → J (m = 1, . . . , n − 1) by h1 := k1 ,
hm+1 (·) := exp(−Tm )hm (·) exp(Tm ). Notice that k2 = hn+1 .
Something unclear here!
Unitary homotopy, as defined in Definition 5.0.1 is an equivalence relation (in
particular is transitive) and it suffices to show that of
????? of Theorem 4.4.6 ????
i.e., k1 +βk3 and k2 +βk3 are unitarily homotopic in E in the sense of Definition
5.0.1, which is sometimes weaker than being homotopic or unitary equivalent.
hm +βk3 and hm+1 +βk3 are unitarily homotopic for m = 1, . . . , n (respectively
are unitary equivalent by a unitary in U0 (J + C · 1) if J is sub-Stonean).
Thus, we may suppose that n = 1 and that k2 = exp(−T )k1 (·) exp(T ) with
∗
T = −T .
Let T ∈ E with T ∗ = −T , and suppose first that J is a sub-Stonean C *-
algebra. In fact, we need only the weaker property of J that for every g ∈ J+ there
exists f ∈ J+ with kf k ≤ 1 and gf = g. Then we find a positive contraction f ∈ J+
with xf = x for all x ∈ k1 (B) and a positive contraction e ∈ J+ with f e = e and
f T ` = f (eT e)` for all ` ∈ N.
Indeed:
k1 (B) contains a strictly positive contraction g1 , and there exists a contraction
f ∈ J+ with f g1 = g1 . The C *-subalgebra of J, generated by the countable subset
Y := { f, f T n ; n ∈ N } of J, contains a strictly positive contraction g2 . We find a
positive contraction e ∈ J+ with g2 e = g2 . It satisfies ye = y for all y ∈ Y , which
implies f T = f T e = (f e)(T e) = f (eT e) , and that

f T n+1 = f T n+1 e = (f T n )T e = (f T n e)T e = f T n (eT e) .

It follows by induction that f T n = f (eT e)n . Thus, f exp(T ) = f exp(eT e) and

k1 (b) exp(T ) = k1 (b)f exp(T ) = k1 (b)f exp(eT e) = k1 (b) exp(eT e) for all b ∈ B .

If we use that (eT e)∗ = eT ∗ e = − eT e and B ∗ · B = B, then we obtain that

exp(eT e) ∈ U0 (J + C · 1) ⊂ U0 (E) and, for all b ∈ B,

exp(−eT e)(k1 (b) + βk3 (b)) exp(eT e) = (exp(−T )k1 (b) exp(T )) + βk3 (b) .

We can reduce the proof of the general unitary homotopy to the sub-Stonean case if
we consider k1 and exp(−T )k1 (·) exp(T ) as maps into suitable sub-Stonean ideals,
e.g. if we use that
602 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

(1) I := Cb ([0, ∞), J)/ C0 ([0, ∞), J) is naturally isomorphic to a closed ideal
of
Cb ([0, ∞), E)/ C0 ([0, ∞), E) ⊃ E ,
(2) J = E ∩ I is a C *-subalgebra of I, and
(3) I is a sub-Stonean C *-algebra.

Property (3) is easy to see. The above considerations show that, for given T ∈ E
with T ∗ = −T , there is a positive contraction e ∈ I+ such that

exp(−T )k1 (·) exp(T ) = exp(−eT e)k1 (·) exp(eT e) .

It means equivalently that exp(−T )k1 (b) exp(T ) = limξ→∞ U (ξ)∗ k1 (b)U (ξ) for all
b ∈ B, where we define

U (ξ) := exp(g(ξ)T g(ξ)) ∈ U0 (J + C · 1) ⊂ U0 (E)

with g ∈ Cb ([0, ∞), J) a positive contraction that satisfies g + C0 ([0, ∞), J) = e. In

particular, ξ 7→ U (ξ) is norm-continuous and U (ξ) ∈ U0 (βJ 0 ∩ E). It follows that
k1 + βk3 and exp(−T )k1 (·) exp(T ) + βk3 are unitarily homotopic via ξ 7→ U (ξ) for
T ∈ E with T ∗ = −T .
This proves finally that [k1 ] = [k2 ] and kj (B) ⊆ J for j = 1, 2, 3 imply the
unitary homotopy of k1 + βk3 and k2 + βk3 . 

Remark 4.5.6. There are also isometries s1 and t1 with s1 s∗1 + t1 t∗1 = 1 and
t1 H0 (·) = H0 = H0 (·)t1 as proposed in Part (vi) of Lemma 4.4.7, but with inter-
changed role of p0 and 1 − p0 , i.e., we find those isometries s1 , t1 ∈ E with the
additional properties s∗1 p0 s1 = 1, (1 − p0 )t1 = (1 − p0 ) = t1 (1 − p0 ) instead of
s∗1 (1 − p0 )s1 = 1, p0 t1 = p0 = t1 p0 .
We do not know if in some applications (!) for given g ∈ (J + βJ)+ there exists
isometries s1 , t1 with above properties and the additional property that t1 g = g =
gt1 , i.e., that in addition to p0 s1 = s1 holds gs1 = 0. It means that there exists a
full properly infinite projection q ∈ p0 Ep0 with qg = 0.

Proof. By Proposition 4.3.6(v), it suffices to find an isometry R ∈ E with

(1 − p0 )R = 0 and R∗ H0 (·)R = 0, i.e., with R∗ p0 H0 (·)R = 0.
There exists u ∈ U(E) with s0 (h0 +βh0 )s∗0 = u∗ (h0 +βh0 )u by Lemma 4.4.7(i).
Let R := s0 ut0 . Then p0 = s0 s∗0 , s∗0 H0 s0 = h0 + βh0 and s∗0 t0 = 0 imply p0 R = R,
and R∗ H0 (·)R = 0. 

Here are some additional remarks concerning the C *-subalgebras C ∗ (p, q) and
C ∗ (p, q, 1) generated by projections p, q ∈ A.

Remark 4.5.7. Let A a unital C *-algebra that is Z2 -graded by βA ∈ Aut(A)

and let F1 , F2 ∈ A symmetries (i.e., Fk∗ = Fk and Fk2 = 1) of degree one (i.e.,
βA (Fk ) = −Fk ).
Then the unital C *-subalgebra B := C ∗ (F1 , F2 ) of A satisfies βA (B) = B.
 5. GENERALIZATIONS AND LIMITATIONS 603

It follows from Parts (iii) and (v) of Lemma 4.1.3 that there exists a grading pre-
serving unital C *-epimorphism ϕ from the universal (real or complex) C *-algebra
C ∗ (P, Q, 1) ⊂ C([0, π/2], M2 ) onto B with ϕ(1 − 2P ) = F1 and ϕ(1 − 2Q) = F2 .
Here we take the odd Z2 -grading on C ∗ (P, Q, 1) given as the restriction to
C ∗ (P, Q, 1) of the Z2 -grading β on C([0, π/2], M2 ) defined by β(f )(ϕ) := Z ∗ f (ϕ)Z
for f ∈ C([0, π/2], M2 ) with Z := [ζjk ] ∈ M2 ⊂ C([0, 1], M2 ) with ζjk := j − k,
j, k ∈ {1, 2}.
In particular, if kF1 − F2 k < 2 then there exists h ∈ C ∗ (F1 , F2 ) of degree zero
(i.e., βA (h) = h), with h∗ = −h, and khk < π/2 such that F2 = exp(−h)F1 exp(h).
The Z2 -grading β of C([0, π/2], M2 ) is odd in case M2 := M2 (R) and is even if
M2 := M2 (C), because there is no symmetry S in M2 (R) and µ ∈ R with Z = µS,
but (iZ)∗ = iZ =: S is a symmetry in M2 (C).

Remark 4.5.8. A continuous path ξ ∈ [0, 1] 7→ pξ ∈ A of projections pξ with

p0 = p and p1 = q for projections p, q ∈ A with kp − qk < 1, can be defined by
pξ := (U (ξ) + 1)/2 with symmetries U (ξ) := |T (ξ)|−1 T (ξ) obtained from the path
T (ξ) := sin(ξπ/2)(2q − 1) + cos(ξπ/2)(2p − 1)
of self-adjoint invertible operators T (ξ) with T (ξ)2 ≥ 2(1 − kp − qk2 ) in A + C · 1,
cf. [73, prop. 4.6.6] and proof of [816, prop. 5.2.6].
Parts (v) and (vi) of Lemma 4.1.3 show that the minimal length of a continuous
path ξ 7→ u(ξ) ∈ U(A) of unitary operators with u(0) = 1 and u(1)∗ pu(1) = q is
exactly arcsin kp − qk, because the lower bound for this length can be seen in M2
itself.

Remark 4.5.9. There is an alternative way in the complex case to get the pa-
rameters for the C *-morphisms of the universal unital C *-algebra C ∗ (P, Q, 1) gen-
erated by projections P, Q into M2 by considering the isomorphism of C ∗ (P, Q, 1)
with universal C *-algebras
C ∗ (P, Q, 1) = C ∗ (s, t) ∼
= C ∗ (Z2 ∗ Z2 ) ∼
= C ∗ (Z o Z2 ) ∼
= C(S 1 ) o Z2
with self-adjoint isometries s := 1 − 2P , t := 1 − 2Q, Z ⊂ Z2 ∗ Z2 given by
n ∈ Z 7→ (st)n ∈ Z2 ∗ Z2 and the Z2 -action n 7→ −n on Z realized by (st)n 7→
s(st)n s = (st)−n . Finally, C ∗ (Z) will be naturally identified with C(Z)
b = C(S 1 ).
We get a parametrization by ϕ ∈ (0, π/2) of the simple quotients of C(S 1 ) o Z2
with the 4 characters attached at 0 and π/2 in the same way as the parametrization
given by the embedding of C ∗ (P, Q, 1) into C([0, π/2], M2 ) with the 4 characters
attached at 0 and π/2.
Let F := [fjk ] = diag(−1, 1)Z ∈ M2 the flip-symmetry with fjk := |1 − δjk | .
Then the inner automorphism U ∗ (·)U of M2 (C) given e.g. by the unitary U :=
√ √
[ujk ] ∈ M2 (C) with entries u11 := −i/ 2 =: −u21 and u12 := u22 := 1/ 2
transforms all pairs of unitary elements diag(e2iϕ , e−2iϕ ), F ∈ M2 (C) into the pairs
t(ϕ), s defined in our case as
(1 − 2P )(1 − 2Q(ϕ)) , 1 − 2p ∈ O(2) ⊂ M2 (R) .
604 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Notice here that

st(ϕ) = (1 − 2P )(1 − 2Q(ϕ)) = (1 − 2P ) exp(−ϕZ)(1 − 2P ) exp(ϕZ) = exp(2ϕZ)

with P, Z, Q as in Lemma 4.1.3(iii). Then it suffices to check that U ∗ U = 12 , F U =

U diag(−1, 1) and diag(i, −i)U = U Z, to obtain that the parameter ϕ ∈ [0, π/2]
defines the same C *-morphism from C ∗ (P, Q, 1) into M2 .

6. On K-theory of corona C *-algebras

The homotopy of id and δ2 and that O2 is a pi-sun algebra

can be used to derive that they are unitary equivalent in

`∞ (O2 )/c0 (O2 ) .

It turns out that this property implies that O2 has a central sequence
of unital copies of O2 ...
Some almost E-theory results:
Suppose that B is a σ-unital C *-algebra and that M(B) is properly infinite, i.e.,
that O∞ is unitally contained in M(B), let A a separable C *-subalgebra of B, and
suppose that there is a sequence of point-norm continuous paths of c.p. contractions
Vn,ξ : A → B (n ∈ N, ξ ∈ [0, 1]) given, where M(B) is properly infinite, i.e., there
are isometries S, T ∈ M(B) with S ∗ T = 0.
Is this the Def. of ”properly infinite” C*-algebras?
We can manage that SS ∗ + T T ∗ = 1 if [1M(B) ] = 0 in K0 (M(B)) (e.g. this
is the case if B is stable) and other-wise we find the isometries S, T ∈ M(B) such
that 1 − SS ∗ + T T ∗ is a properly infinite full projection, e.g. we can replace S, T
by T S, ST and have T 2 (T 2 )∗ ≤ 1 − (T SS ∗ T ∗ + ST T ∗ S ∗ ).
Notice that [1 − SS ∗ + T T ∗ ] = −[1] in K0 (M(B)).
We define

(i) a unital C *-morphism λ : O∞ → M(B) by λ(Tn ) := Sn := T n−1 S (with

T 0 := 1) and
(ii) V∞,α : A → B∞ := `∞ (B)/c0 (B) for α ∈ {0, 1} by

V∞,α (a) := (V1,α (a), V2,α (a), . . .) + c0 (B) .

Now fix a linear filtration X1 ⊆ X2 ⊆ of A by finite-dimensional subsets with

S
n Xn dense in A.

Then there exists “Cuntz-averages”

kn
X
Wn := Sk Vn,ξk Sk∗
k=1

of the Vn,ξ (ξ ∈ [0, 1]) with suitable 0 = ξ1 < ξ2 < . . . < ξkn = 1 and unitaries
Un ∈ M(B) with

(iii) kVn,τk (a) − Vn,τk+1 (a)k < 2−n for a ∈ Xn , and

 6. ON K-THEORY OF CORONA C *-ALGEBRAS 605

(iv) S ∗ Un = S1∗ S ∗ , Sk∗ T ∗ Un = Sk+1

∗
S ∗ for k = 1, . . . , kn − 1 and Sk∗n T ∗ Un =
T ∗ , i.e., with suitable partial isometry Zn ∈ M(B) the unitary Un is given
by

Un := SS1∗ S ∗ + T S1 S2∗ S ∗ + . . . + T Sk Sk+1

∗
S ∗ + T Skn −1 Sk∗n S ∗ + T Skn T ∗ + Zn .

The partial isometry Zn ∈ M(B) used in the definition of the unitary Un exists
by our assumptions on the isometries S, T , because Sk = T k−1 S.

Y := SS1∗ S ∗ + T XS ∗ + T Skn T ∗ ,
with
n −1
kX
∗
X := Sk Sk+1 .
k=1
Pkn −1
Then X ∗ Skn = 0 = Sk∗n X , X ∗ X = ∗
k=1 Sk+1 Sk+1 and
X
XX ∗ =???? = k = 1kn −1 Sk Sk∗ .

∗
Sk Sk+1 S`+1 S`∗ = δk,` Sk Sk∗

n −1
kX n −1
kX
∗
Sk Sk+1 S`+1 S`∗ = Sk Sk∗
k,`=1 k=1

It follows
Xkn
Y ∗Y = ( Sj Sj∗ ) ⊕S,T 1
j=1

and
Xkn
Y Y ∗ = 1 ⊕S,T ( Sj Sj∗ )
j=1
Pkn
The projection Pkn := 1 − j=1 Sj Sj∗ is full and properly infinite by construction
of the Sk , because Pkn S` = S` for ` > kn .
Then W∞ (a) := (W1 (a), W2 (a), . . .) + c0 (B) is a c.p.c. map from A into B∞
and U∞ = (U1 , U2 , . . .) + c0 (B) is a unitary in M(B∞ ) that satisfies
∗
U∞ (V∞,0 ⊕S,T W∞ )U∞ = W∞ ⊕S,T V∞,1

in B∞ := `∞ (B)/c0 (B).
The same happens for Bω in place of B∞ , simply by passing to the quotient Bω
of B∞ and by observing that this epimorphism maps M(B∞ ) unitally into M(Bω ).
A bit more technical work is needed for the construction of ?????
The key trick of N.Ch. Phillips in [627] is the construction of suitable interpola-
tions between the above defined “Cuntz averages” for point norm continuous maps
Vt,ξ ∈ CP(A, B) (t ∈ [0, ∞), ξ ∈ [0, 1]), i.e., construct fitting strictly continuous
paths A 3 a 7→ Wt (a) ∈ B in the approximately inner c.p.c. maps Wt ∈ CP(A, B)
606 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

that satisfy that Vt,0 ⊕ Wt and Wt ⊕ Vt,0 are asymptotic unitarily equivalent by a
norm-continuous path t → U (t) of unitaries in M(B).
We do this in case of special classes of Vt,ξ ∈ C ⊆ CP(A, B) that cover the cases
considered by N.Ch. Phillips.
Further plan for topics:
Applications to unital endomorphisms of O∞
(all are unitarily homotopic),
permutation C *-morphisms of O∞ ⊗ O∞ , O∞ ⊗ O∞ ⊗ · · ·
(are all unitarily homotopic)
and to the n-repeat δn : On → On
(δn unitarily homotopic to idOn ,
unital endomorphisms are all unitarily homotopic to id
existence of asymptotic central path of copy of O2 in O2 ).
Following should be a remark in Chp.4?
But with target O2 replaced by Q(R+ , O2 ) ? Partly some appears
even in ?? Chp.1.!. But at beginning of Chp.5
there are some calculations in this direction
All unital C *-morphisms h : O2 → A are unitarily homotopic to each other,
if and only if, they are homotopic, – and in “good cases” of A (e.g. if A is K1 -
injective)– if and only if, the related unitaries are in U0 (A) ?
h1 , h2 : O2 → Q(R+ , A) both are in “general position” and represent the zero
element of [Homu (O2 , Q(R+ , A))], because they dominate each other. In general
the unitary equivalence can not be given by a unitary in U0 (Q(R+ , A)). Example:
The unital free product O2 ∗ C(S 1 ) ∼
= O2 ∗ O 2 .
The latter could be deduced from a very special case of Theorem A: There is a
unital g : O2 ⊗ O2 → O2 , and a 7→ h(a) := g(a ⊗ 1) is unitarily homotopic inside O2
to idO2 . Then h1 (g((·) ⊗ 1)) and h2 (g((·) ⊗ 1)) dominate each other asymptotically
by
Corollary ?? (on the asymptotic domination of residually nuclear C *-
morphisms hk : A → B of stable separable C *-algebras A and B, where A is exact
and separable and B is s.p.i.).
Suppose that E is strongly purely infinite, and that A is separable and exact,
(or if E is not s.p.i., that A is separable, s.p.i. and exact). Then every nuclear
C *-morphism ϕ : A → E extends to a C *-morphism ψ : A ⊗ O∞ → Q(R+ , E), by
Proposition 3.0.4.
Let gk : A ⊗ O∞ → E (k = 1, 2) C *-morphisms such that the C *-morphisms
hk (a) := gk (a ⊗ 1) generate the same m.o.c. cone in CP(A, E), then h1 and h2
1-step-dominate each other in Hom(A, E ∞ ) or Hom(A, Q(R+ , E)).
If the hk are injective and nuclear, then they generate the same m.o.c. cone, if
and only if, for each a ∈ A+ , the elements h1 (a) and h2 (a) generate the same ideal
of E.
 6. ON K-THEORY OF CORONA C *-ALGEBRAS 607

In particular, for any two injective unital nuclear C *-morphisms g1 , g2 : A ⊗

O2 → E that induce the same action of Prim(E) on A the C *-morphisms h1 and
h2 are unitarily homotopic.
Because one finds isometries S, T ∈ Q(R+ , E) with S ∗ h1 (·)S = h2 and
T ∗ h2 (·)T = h1 by simplicity and nuclearity of the hk and the pure algebraic fact
that there commutant algebras contain (possibly different) copies of O2 unitally.
Then we get from Chapter 4 that h1 and h2 both are unitarily equivalent in
Q(R+ , E) to h1 ⊕ h2 .
The situation of unital C *-morphisms form O2 into the stable corona Qs (B) is
different:
The (usual) homotopy classes of the unital h : O2 → Qs (B) are in bijective
correspondence to K0 (B) ∼= K1 (Qs (B)) by the K1 -bijectivity of stable coronas by
Proposition ??, but there is only one (sic !) class of unitary homotopy.
More generally, for every unital C *-algebra E with properly infinite unit that
satisfies 0 = [1E ] ∈ K0 (E) there is up to unitary homotopy only one unital C *-
morphism h : O2 → E, cf. Section ??, but the homotopy classes of unital h ∈
Hom(O2 , E) is identical with U(E)/U0 (E), which is isomorphic to K1 (E) if and
only if E is K1 -injective in addition (As is e.g. the case where E stable corona
E = Qs (A) of a σ-unital C *-algebra A or where E is strongly purely infinite.)
An example is the Calkin algebra C := C(`2 ) := L(`2 )/K(`2 ) . Then each
unital C *-morphism ψ : O2 → C is up to unitary equivalence given by multiplying
the πK ◦ ρ(sj ) with a unitary in C from the left. The multiplication with images of
Toeplitz operators T n or (T ∗ )m exhausts all cases (up to unitary equivalence in C
?).

Corollary 4.6.1. There is a copy of On unitally contained in On 0 ∩Q(R+ , On ),

i.e., there exists a point-norm continuous path of u.c.p. maps Vt : On ⊗ On → On
such that limt→∞ kVt (a ⊗ 1) − ak = 0 for all a ∈ On , limt→∞ k[a, Vt (1 ⊗ b)]k = 0
for all a, b ∈ On and

lim kVt (1 ⊗ b∗ b) − Vt (1 ⊗ b)∗ Vt (1 ⊗ b)k = 0 for all b ∈ On .

t→∞

The proof should follow from a “unsuspended” E-theory argument concerning

asymptotic homotopy of id and δn considered in Q(R+ × [0, 1], On ), respectively
the then unitary equivalence δn (a) = U ∗ aU in Q(R+ , On ) for a ∈ On ⊆ Q(R+ , On ).
The unital embedding On → On 0 ∩ Q(R+ , On ) is defined by the elements U sj
(j = 1, . . . , n).

Corollary 4.6.2. For every strongly purely infinite, unital, separable and
nuclear C*-algebra A there is a copy of O∞ ⊗ O∞ ⊗ · · · unitally contained in
A0 ∩ Q(R+ , A).
608 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

It uses that O∞ ⊆ M(A) (unitally) and the nuclear map a → s1 as∗1 + s2 as∗2
is asymptotical 1-step dominated by idA , because every nuclear ideal system pre-
serving c.p. map is approximately 1-step inner. (See study of nuclear c.p. maps in
s.p.i. algebras).
Find contraction T ∈ Q(R+ , M(A)) with s1 as∗1 + s2 as∗2 = T ∗ aT it implies that
δjk · a = s∗j T ∗ aT sk = t∗j atk for the contractions tk := T sk .
More ???? Tensor by K first or not? ...
We shall see in Chapter 10 that this implies that A ⊗ O∞ ∼ = A for every
purely infinite separable nuclear A, using that there exists a unital C *-morphism
η : O∞ → C([0, 1], O∞ ⊗ O∞ ) with π0 (η(b)) = b ⊗ 1 and π1 (η(b)) = 1 ⊗ b by
η(sn )(t) :=????? ?????????????? exact formulations needed!!! And where is
the proof? of .

Corollary 4.6.3. If B is separable, nuclear and strongly purely infinite then

there exists a norm-continuous path of *-monomorphisms ht : B ⊗ O∞ → B such
that limt→∞ kht (b ⊗ 1) − bk = 0.

Next corollary will be used in Chapter 5.

Corollary 4.6.4. Suppose that A is a unital C*-algebra and h1 , h2 ∈

Hom(O2 , A) are unital C*-morphisms, then there exists a norm-continuous path
t ∈ [0, ∞) 7→ u(t) ∈ U(A) such that h2 (x) = limt→∞ u(t)∗ h1 (x)u(t) for all x ∈ O2 .

Proof. There exist a unital C *-morphism k0 : O2 → O2 0 ∩ Q(R+ , O2 ), by

Corollary 4.6.1.
Give exact and precise reference!
We write O2 := C ∗ (s1 , s2 ), s∗1 s1 = 1 = s∗2 s2 , s1 s∗1 + s2 s∗2 = 1.
The unital C *-morphisms h1 , h2 : O2 → A extend to unital C *-morphisms H1
and H2 from Q(R+ , O2 ) to Q(R+ , A) with H` |O2 = h` (` ∈ {1, 2}).
Then k` := H` ◦ k0 are unital C *-morphisms from O2 into h` (O2 )0 ∩ Q(R+ , A)
(` ∈ {1, 2}).
It suffices to show that there exists a unitary U ∈ Q(R+ , A) with U ∗ h1 (sj )U =
h2 (sj ), j ∈ {1, 2}.
To be filled in ?? 

?? Is next Lemma 4.6.5 in a good position? Move it to Chapter 5?

Lemma 4.6.5. Suppose that B is a stable C*-algebra and that ψ : D → M(B)

is a C*-morphism from separable D to M(B).
The relative commutant δ∞ (ψ(D))0 ∩ M(B) of the image of the infinite repeat
δ∞ ◦ ψ has trivial K∗ -groups and is closed in the strict topology.
 6. ON K-THEORY OF CORONA C *-ALGEBRAS 609

sn bs∗n for
P
The infinite repeat δ∞ : M(B) → M(B) is given by δ∞ (b) :=
Pn
b ∈ M(B) where s1 , s2 , . . . is a sequence of isometries in M(B) with n sn s∗n = 1
strictly convergent, cf. Remark 5.1.1(8) and Lemma 5.1.2(i,ii).

Proof. Let F := δ∞ (ψ(D))0 ∩ M(B). It is closed under strict convergence.

Certainly, δ∞ (M(B))0 ∩ M(B) ⊆ F . By Lemma 5.1.2(i),
early citations ? ??
there are isometries t1 , t2 , . . . in δ∞ (M(B))0 ∩ M(B) such that
P ∗
tn tn strictly
converges to 1. By Lemma 5.1.2(ii), there are isometries r1 and r2 in δ∞ (M(B))0 ∩
M(B) with r1 r1∗ + r2 r2∗ = 1.
Let h : M(B) → M(B) denote the infinite repeat of id which is defined by
P
t1 , t2 , . . ., cf. Remark 5.1.1(8).The element h(b) = is in F if b ∈ F , because F
is strictly closed. Thus, h|F is a unital *-endomorphism of F . Let u := r1 t∗1 +
r2 tn t∗n+1 . The right side converges strictly to a unitary u in M(B) by proof
P

of Lemma 5.1.2(i) (see e.g. Remark 5.1.1(2) for the strict convergence). Since its
summands are in F , we get u ∈ F . It holds u∗ (b ⊕r1 ,r2 h(b))u = h(b) for b ∈ F by
the convergence argument in Remark 5.1.1(2) and in the proof of Lemma 5.1.2(ii).
Thus K∗ (F ) = 0 ( 9 ). 

The following Lemma 4.6.6 is a K1 -counterpart to Proposition 4.4.3(iii).

4.4.3(iii) correct? ??

Lemma 4.6.6. Suppose that B is stable and σ-unital, D is stable and separable,
and that H : D → M(B) is a non-degenerate C*-morphism such that δ∞ ◦ H is
unitarily equivalent to H (by an unitary in M(B)).
Let E := Qs (B) = M(B)/B , H0 := πB ◦ H , C := H0 (D)0 ∩ E , and let
U(H(D), B) denote the set of unitaries in M(B) that commute modulo B element-
wise with H(D).

(i) U(H(D), B) is closed under Cuntz addition and U 7→ [U +B] is an additive

epimorphism from U(H(D), B) onto the kernel of i1 : K1 (C) → K1 (E),
where i1 = K1 (i) for the inclusion map i : C ,→ E.
(ii) Let U1 , U2 ∈ (H(D), B) and let U := U1 ⊕ 1, U 0 := U2 ⊕ 1. Then
[U1 + B] = [U2 + B] in K1 (H0 (D)0 ∩ E),
if and only if,
there exist U 00 ∈ U(H(D), B) such that U 00 − U ∈ B and U 0 is homotopic
to U 00 in U(H(D), B),
if and only if,
there exist U 00 , V ∈ U(H(D), B) such that V ∗ U 0 V is homotopic to U 00 in
U(H(D), B) and (U 00 − U )H(D) ⊆ B.

9More generally, the existence of a unital endomorphism h : F → F with [h ⊕ id ] = [h] and

F
h ◦ C ⊆ C implies KK(C; A, F ) = 0 for all our generalized KK-theories in Chapter 8.
610 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Proof. There exists a unital C *-morphism O2 → δ∞ (M(B))0 ∩ M(B). The

latter is contained in H(D)0 ∩ M(B). It defines the needed isometries s1 , s2 ∈
H(D)0 ∩ M(B), cf. e.g. Lemma 5.1.2(ii).
?? Is it good to ref Chp 5 here?
(i): It is easy to see that U(H(D), B) is closed under Cuntz addition, because
O2 commutes with H(D). The image of U(H(D), B) under U → U + B ∈ E
goes into the unitary group of C . K1 (M(B)) = 0 because B is stable. Thus
[U + B]E = 0 in K1 (E) because [U ]M(B) = 0 in K1 (M(B)). The Cuntz addition in
M(B) induces the Cuntz addition in E: (U1 ⊕ U2 ) + B = (U1 + B) ⊕ (U2 + B). By
Lemma 4.2.6(v),the natural map from the unitaries of C into K1 (C) is an additive
epimorphism with respect to Cuntz addition on C.
It remains to show that U → [U + B] is an epimorphism onto the kernel of
K1 (C) → K1 (E). Let x ∈ K1 (C) be in the kernel of i1 : K1 (C) → K1 (E). Then
there is a unitary in u ∈ C with [u]C = x, [u]E = 0. Since 1E ⊆ O2 ⊆ C ⊆ E, we
can use Lemma 4.2.6(v,2) and get [u ⊕ 1]C = [u]C = x and that u ⊕ 1 is homotopic
to 1 in the unitaries of E. Thus, there is a unitary U ∈ M(B) with U + B = u ⊕ 1.
U commutes element-wise with H(D) modulo B, because U + B = u ⊕ 1 ∈ C =
((H(D) + B)/B)0 ∩ E.
(ii): Let U 0 := U1 ⊕ 1, U 00 := U2 ⊕ 1 and v1 := U1 + B and v2 := U2 + B
the corresponding unitaries in C. By Lemma 4.2.6(v,2), [v1 ]C = [v2 ]C if and only
if v1 ⊕ 1 = U 0 + B is homotopic to v2 ⊕ 1 = U 00 + B in the unitaries of C. This
homotopy can be lifted to a continuous arc t ∈ [0, 1] 7→ U (t) in the unitaries of
M(B) starting from U (0) = U 0 and ending at U 00 := U (1). Then U 00 − U ∈ B. But
if V is a unitary in M(B) such that V + B ∈ C then V commutes element-wise
with H(D) modulo B. Thus U (t) ∈ U(H(D), B) for t ∈ [0, 1].
Conversely, let U := U1 ⊕ 1, U 0 := U2 ⊕ 1 and suppose that there exist unitaries
V, U 00 ∈ U(H(D), B) and a continuous arc t ∈ [0, 1] 7→ U (t) into U(H(D), B) with
U (0) = V ∗ U 0 V , U (1) = U 00 and (U 00 − U )H(D) ⊆ B.
Let v1 := U1 + B, v2 := U2 + B, v3 := V + B and v4 := U 00 + B. Then
[v2 ]C = [v 3 (v2 ⊕ 1)v3 ]C = [v4 ]C in K1 (C), because [1]C = 0 and t 7→ U (t) + B
connects v 3 (v2 ⊕ 1)v3 and v4 in the unitaries of C.
On the other hand, v1 ⊕1, v4 ∈ C, (v1 ⊕1)−v4 is in the annihilator Ann(H0 (D))
of H0 (D) in E, and [v1 ⊕ 1]E = [v4 ]E = 0 in K1 (E). By Proposition 5.5.12(ii),
H0 : D → E dominates zero, because (πB )−1 (H0 (D)) = B +H(D) contains the sta-
ble and non-degenerate C *-subalgebra H(D). Therefore, the two-sided annihilator
Ann(H0 (D)) of H0 (D) ⊆ E is a full hereditary C *-subalgebra of E. The annihila-
tor Ann(H0 (D)) is an ideal of C = H0 (D)0 ∩ E. Thus, Lemma 4.2.20(ii) applies to
N := Ann(H0 (D)) ⊆ C ⊆ E and v1 ⊕1, v4 ∈ C . We get [v1 ] = [v1 ⊕1] = [v4 ] = [v1 ]
in K1 (C). 

Corollary 4.6.7. Suppose that J ⊆ E ⊆ F are C*-subalgebras of F and

satisfy the following conditions (i)–(v):
 6. ON K-THEORY OF CORONA C *-ALGEBRAS 611

(i) 1F ∈ E.
(ii) J and is an ideal of F .
(iii) For every b ∈ J+ there is c ∈ J+ with kck = 1 and cb = b.
(iv) For every c ∈ J+ there is an isometry t ∈ E with ct = 0.
(v) E contains a copy of O2 unitally.

If A is a σ-unital C*–algebra, and h1 , h2 : A → J are C*-morphisms, then the

following (a)–(c) are equivalent:

(a) h2 = U ∗ h1 (·)U for a unitary U in the connected component U0 (E) of 1

in the group U(E) of unitaries of E.
(b) h1 and h2 are unitarily equivalent in F .
(c) h1 and h2 are unitarily equivalent in M(J).

Proof. (a)⇒(b)⇒(c): There are natural unital C *-morphisms

E → F → M(J)

that fix the elements of J by assumptions (i) and (ii).

(c)⇒(a): Suppose that U ∈ M(J) is a unitary with h2 = U ∗ h1 (·)U . Let
e ∈ A+ a strictly positive element of A, and let b := h1 (e) + h2 (e). By (iii) and
(iv), there are c ∈ J+ and t ∈ E with kck ≤ 1, cb = b, ct = 0 and t∗ t = 1. It follows
chk (·) = hk = hk (·)c (k = 1, 2), and t∗ (h1 (a) + h2 (a))t = 0 for a ∈ A. Let x := cU c.
Then x is a contraction in J ⊆ E that satisfies x∗ h1 (·)x = h2 and xh2 (·)x∗ = h1 .
Since t ∈ E and E contains a copy of O2 unitally by (v), the Parts (iii) and (iv,c)
of Proposition 4.3.6 apply: There is V ∈ U0 (E) with V ∗ h1 (·)V = h2 . 

Remark 4.6.8. We consider in Chapters 7 and 9, the case, where F :=

Q(C0 (X, B)) := M(C0 (X, B))/ C0 (X, B), E := Cb (X, M(B))/ C0 (X, B) and J :=
Q(X, B) := Cb (X, B)/ C0 (X, B) for stable σ-unital B and a σ-compact locally
compact Hausdorff space X.
Note that in this particular case, J is the (two-sided) annihilator Ann(C, F ) of the
C *-subalgebra C := πC0 (X,B) (C0 (X) · 1) of the center of F .

Finally, we say some words about the groups G≈ (h0 ; D, E) in case where D and
h0 are unital and E is K1 -injective, because this is important for the comparison of
stable extensions with unital extension in case where E = Qs (B) = M(B)/B for
some σ-unital stable B.

Remark 4.6.9. Suppose that D is unital, that h0 : D → E is unital and that

isometries t1 , t2 ∈ E are canonical generators of a copy of O2 in E. Let h ∈
Hom(D, E).
Recall that [h]≈ ∈ [Hom(D, E)]≈ denotes the class of k ∈ Hom(D, E) with
k = u∗ h(·)u for a unitary in the connected component U0 (E) of 1 in in the unitary
group U(E), and that [Hom(D, E)]≈ becomes a semigroup with Cuntz addition
[h1 ]≈ + [h2 ]≈ := [h1 ⊕s1 ,s2 h2 ]≈ (see Proposition 4.3.2).
612 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Suppose now – in addition – that [h0 ]≈ + [h0 ]≈ = [h0 ]≈ in ([Hom(D, E)]≈ , +).
The proof of Proposition 4.3.5(iii) shows that this equivalently means that there is
u ∈ U0 (E) such that s1 := ut1 , s2 := ut2 ∈ h0 (D)0 ∩ E.
Notice that s∗1 s1 = s∗2 s2 = 1 and s1 s∗1 + s2 s∗2 = 1, and that [h1 ⊕t1 ,t2 h2 ]≈ =
[h1 ⊕s1 ,s2 h2 ]≈ for all h1 , h2 ∈ Hom(D, E). Therefore we can write ⊕ for ⊕s1 ,s2 .
We can define a subset S≈ (h0 ; D, E) ⊆ [Hom(D, E)]≈ :

S≈ (h0 ; D, E) := {[h]≈ ; ∃(h0 ∈ Hom(D, E)) : [h]≈ + [h0 ]≈ = [h0 ]≈ }.

Since [h0 ]≈ + [h0 ]≈ = [h0 ]≈ , the set S≈ (h0 ; D, E) is a sub-semigroup of

[Hom(D, E)]≈ . The set G≈ (h0 ; D, E) := [h0 ]≈ + S≈ (h0 ; D, E) is a sub-
group of S≈ (h0 ; D, E) that is naturally isomorphic to the Grothendieck group
Gr(S≈ (h0 ; D, E)) of S≈ (h0 ; D, E) by Lemma 4.2.3.
The definition of S≈ (h0 ; D, E) shows that [h]≈ ∈ S≈ (h0 ; D, E) implies that
there is u ∈ U0 (E) with h = s∗1 u∗ h0 (·)us1 . In particular, h : D → E is unital.
Conversely, if u ∈ U0 (E) and if h := s∗1 u∗ h0 (·)us1 is multiplicative, then
us2 (s2 u)∗ = 1−us1 (s1 u)∗ commutes with h0 (D) and s1 s∗1 = 1−s2 s∗2 commutes with
u∗ h0 (D)u (by Lemma 4.3.4(i)), and h0 := s∗2 u∗ h0 (·)us2 is a unital C *-morphism
with [h]≈ + [h0 ]≈ = [h0 ]≈ . [h]≈ is in G≈ (h0 ; D, E), if and only if, there are
u, v ∈ U0 (E) with h = s∗1 u∗ h0 (·)us1 and h0 = s∗1 v ∗ h(·)vs1 . Thus, there is a
natural semigroup morphism [h]≈ ∈ S≈ (h0 ; D, E) → [h] ∈ S(h0 ; D, E). It defines
a natural group morphism of Grothendieck groups: G≈ (h0 ; D, E) → G(h0 ; D, E).

Proposition 4.6.10. Suppose that D and E are unital, that h0 : D → E is

unital, and that isometries t1 , t2 ∈ E are canonical generators of a copy of O2 in
E with [h0 ]≈ = [h0 ⊕t1 ,t2 h0 ]≈ .
If E is K1 -injective, then the group-morphisms of Remark 4.6.9 define natural
exact sequences:

(i) K1 (h0 (D)0 ∩ E) → K1 (E) → G≈ (h0 ; D, E) → G(h0 ; D, E) → 0 . Here

K1 (h0 (D)0 ∩ E) → K1 (E) is induced by h0 (D)0 ∩ E ,→ E.
(ii) 0 → G(h0 ; D, E) → G(h0 ⊕ 0; D, E) → K0 (E), where [h] 7→ [h] + [0] for
[h] ∈ G(h0 ; D, E) and [k] 7→ [k(1)] ∈ K0 (E) for [k] ∈ G(h0 ⊕ 0; D, E).

Proof. We use the s1 , s2 ∈ h0 (D)0 ∩ E defined in Remark 4.6.9.

(i): An element h ∈ Hom(D, E) defines an element [h] of S(h0 ; D, E), if and
only if, there is h0 ∈ Hom(D, E) with [h] + [h0 ] = [h0 ]. It follows that the last obser-
vation of Remark 4.6.9 applies almost verbatim to S(h0 ; D, E) and G(h0 ; D, E),
in particular:
[h] ∈ G(h0 ; D, E), if and only if, there are u, v ∈ U(E) with h = s∗1 u∗ h0 (·)us1 and
h0 = s∗1 v ∗ h(·)vs1 .
Since u(1 ⊕ u∗ )s1 = us1 and since [u(1 ⊕ u∗ )] = 0 in K1 (E) we may suppose that,
in addition, [u] = 0 = [v] in K1 (E).
 6. ON K-THEORY OF CORONA C *-ALGEBRAS 613

Now we use the assumption that E is K1 -injective. Then for h ∈ Hom(D, E)

holds: [h] ∈ G(h0 ; D, E) if and only if [h]≈ ∈ G≈ (h0 ; D, E). Thus:
If E is K1 -injective, then the natural morphism G≈ (h0 ; D, E) → G(h0 ; D, E) is
surjective.
If [h]≈ is in G≈ (h0 ; D, E) then [h] ∈ G(h0 ; D, E). Then [h] = [h0 ] if and
only if h = u∗ h0 (·)u for some u ∈ U(E). Thus, the kernel of G≈ (h0 ; D, E) →
G(h0 ; D, E) is the set of equivalence classes [u∗ h0 (·)u]≈ with u ∈ U(E). If
u∗1 h0 (·)u1 = v ∗ u∗2 h0 (·)u2 v for v ∈ U0 (E), then (u2 vu∗2 )u2 u∗1 ∈ h0 (D) ∩ E, i.e.,
u2 u∗1 ∈ U0 (E) · U(h0 (D)0 ∩ E).
Conversely, [u∗1 h0 (·)u2 ]≈ = [u∗2 h0 (·)u2 ]≈ if u2 u∗1 ∈ U0 (E) · U(h0 (D)0 ∩ E). In par-
ticular, ς[u] ∈ K1 (E) 7→ [u∗ h0 (·)u]≈ is well-defined, because K1 (E) ∼ = U(E)/U0 (E)
by Lemma 4.2.6(iv) and Lemma 4.2.10.
Since (u∗ h0 (·)u) ⊕ v ∗ h0 (·)v = (u ⊕ v)∗ h0 (·)(u ⊕ v) the map ς : K1 (E) →
G≈ (h0 ; D, E) is additive by Lemma 4.2.6(iv). It follows: ker(ς) = {[u] ; u ∈
U0 (E) · U(h0 (D)0 ∩ E) } . In particular, ker(ς) is contained in the image of the
natural map η1 : K1 (h0 (D)0 ∩ E) → K1 (E), for the inclusion η : h0 (D)0 ∩ E ,→ E.
The natural map U(h0 (D)0 ∩ E)/U0 (h0 (D)0 ∩ E) → K1 (h0 (D)0 ∩ E) is surjective
by Lemma 4.2.6(v). Therefore, if [u] ∈ K1 (E) is in η1 (K1 (h0 (D)0 ∩ E)), there is
v ∈ U(h0 (D)0 ∩E) with [v] = [u]. We get u ∈ U0 (E)·U(h0 (D)0 ∩E) by K1 -injectivity
of E.
(ii): The map [h] → [h] + [h0 ] + [0] defines (obviously) a semigroup morphism
from S(h0 ; D, E) into G(h0 ⊕ 0 ; D, E).
The map is injective on G(h0 ; D, E):
Let [h] ∈ G(h0 ; D, E) with [h0 ⊕ 0] = [h ⊕ h0 ⊕ 0], then [h0 ] = [h ⊕ h0 ] : By assump-
tion, there is a unitary u ∈ E with u∗ s1 h0 (·)s∗1 u = s21 h(·)(s∗1 )2 + s1 s2 h0 (·)s∗2 s∗1 . In
particular, u∗ s1 s∗1 u = s1 s∗1 . Thus v := s∗1 us1 is unitary, and v ∗ h0 (·)v = h ⊕ h0 .
The map τ : [h] ∈ [Hom(D, E)] 7→ [h(1)] ∈ K0 (E) is a semigroup morphism
by Lemma 4.2.6(i). Since [1] = 0 we get [h(1) ⊕ h0 (1) ⊕ 0] = 0 in K0 (E) if
h(1) = 1. Hence, the image of G(h0 ; D, E) → G(h0 ⊕ 0 ; D, E) is in the kernel of
G(h0 ⊕ 0 ; D, E) → K0 (E).
Now let [k] ∈ S(h0 ⊕ 0 ; D, E) and [k(1)] = 0 in K0 (E) . We are going to
show that there is [h] ∈ G(h0 ; D, E) with [k] + [h0 ] + [0] = [h] + [h0 ] + [0] in
G(h0 ⊕ 0 ; D, E):
There is an isometry t ∈ E with p = k(1) ⊕ 1 ⊕ 0 ≥ tt∗ and [p] = 0 in K1 (E),
thus, there is an isometry s ∈ E with ss∗ = p, cf. Lemma 4.2.6(ii).
By assumption, there is an isometry r ∈ E with r∗ s1 h0 (·)s∗1 r = k. Let h1 :=
s∗ (k ⊕ h0 ⊕ 0)s, then h1 (1) = 1 and h1 := z1∗ h0 (·)z1 + z2∗ h0 (·)z2 = w∗ h0 (·)w for
z1 := s∗1 r(s21 )∗ , z2 := s∗2 s∗1 and w := s1 z1 + s2 z2 . Then w∗ w = h(1) = 1 and
h1 ∈ Hom(D, E), hence h1 ∈ S(h0 ; D, E).
There is a unitary u ∈ E with u∗ (h1 ⊕ h0 ⊕ 0)u = k ⊕ h0 ⊕ 0:
We have sh1 (·)s∗ = k ⊕ h0 ⊕ 0.
614 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Since (ss2 )∗ ss1 = 0, there is an isometry q ∈ E with qq ∗ = 1 − (ss1 )(ss1 )∗ . Let

v := s1 s∗1 s∗ + s2 q ∗ ∈ U(E) with v ∗ s1 = s1 s and v ∗ s2 = q. We get:

v ∗ (h1 ⊕ 0)v = s1 (k ⊕ h0 ⊕ 0)s∗1 = (k ⊕ h0 ⊕ 0) ⊕ 0 .

Since [0]+[0] = [0] and [h0 ]+[h0 ] = [h0 ] in ([Hom(D, E)], +), we get [h]+[h0 ]+[0] =
[k] + [h0 ] + [0] for h := h1 ⊕ h0 . Note that [h] ∈ G(h0 ; D, E). 

The proof of the isomorphism KK(C; A, B(1) ) ∼ = Ext(C; A, B) for trivially

graded A and B in Chapter 8 uses a characterization of the defining relations
of elements given in Chapter 5, and this characterization
Precise reference? To Chp.5 ?? ??
uses the following observation:

Lemma 4.6.11. Let h : D → E a C*-morphism such that C := h(D)0 ∩ E

contains a copy of O2 = C ∗ (s1 , s2 ) unitally, and that h dominates zero.
If p, q ∈ C are projections and V ∈ U(E) a unitary that satisfy V ∗ h(·)pV =
h(·)q, [p] = [q] in K0 (E) and [V ] = 0 in K1 (E), then there exist projections r1 , r2 ∈
Ann(h(D), E) ⊆ C such that [p ⊕ r1 ] = [q ⊕ r2 ] in K0 (C) .
Moreover, there exists a unitary U ∈ U0 (h(D)0 ∩ E) with U ∗ (p ⊕ r1 ⊕ 1 ⊕ 0)U =
q ⊕ r2 ⊕ 1 ⊕ 0 , where ⊕ = ⊕s1 ,s2 .

Proof. The proof can be reduced to the splitting of the 6-term exact into the
split-exact sequence

0 → K∗ (Ann(h(D), E)) → K∗ (C) → K∗ (C/ Ann(h(D), E)) → 0

with splitting maps K∗ (C) → K∗ (Ann(h(D), E)) ∼

= K∗ (E) given by the inclusion
map C ,→ E, see Lemma 4.2.20.
Here we give an alternative elementary approach, that gives explicit elements
that are useful for the study of generalized extension groups Ext(C ; A, B) consid-
ered in Chapter 5. We write ⊕ for the specified Cuntz sum ⊕s1 ,s2 .
Since K0 (C) and K0 (E) are groups, it suffices to consider the case where p is
the range of an isometry in C and q is properly infinite in C. We can do this simply
by passing from (p, q, V ) to the projections p1 := p ⊕ (1 − p) and q1 := q ⊕ (1 − p) ,
and the unitary V1 ∈ U(C) given by V1 := V ⊕s1 ,s2 1 . Then p1 = T T ∗ for the
isometry T := s1 p + s2 (1 − p) ∈ C. Thus, p1 is range of an isometry. and [p1 ]C =
[p]C +[1−p]C = 0 in K0 (C), [q1 ]C = [q]C −[p]C in K0 (C) and [q1 ]E = [q]E −[p]E = 0
in K0 (E). Since h(·) ⊕ h(·) = h, the unitary V1 satisfies V1∗ h(·)p1 V1 = h(·)q1 .
To keep notation simple we suppose in the following that p itself is a range of
an isometry. Let t1 ∈ C, t2 ∈ E the isometries defined in Proposition 4.3.6(i) with
t1 t∗1 + t2 t∗2 = 1 and h(D)t2 = {0}.
The projections p0 := t1 pt∗1 + t2 t∗2 and q 0 := t1 qt∗1 + t2 t∗2 are in C, satisfy
ph(·) = p0 h(·) and qh(·) = q 0 h(·) and have same classes 0 = [p]C = [p0 ] and
[q]C = [q 0 ]C in K0 (C) and [p0 ]E = [p]E = [q]E = [q 0 ]E in K0 (E). This is because
 7. ADDITIONAL REMARKS AND FURTHER PLANS FOR CHP.4 615

t2 t∗2 = 1 − t1 t∗1 ∈ C and t∗1 t1 = 1 imply [t2 t∗2 ] = 0 in K0 (C) and K0 (E). Moreover,
t2 O2 t∗2 is unitally contained in F := t2 Et∗2 = (1 − t1 t∗1 )E(1 − t1 t∗1 ) and f h(D) = {0}
for all f ∈ F . Notice that F is contained in the closed ideal Ann(h(D), E) of C.
The projections 1, t2 t∗2 , 1−t∗1 pt1 and 1−t∗1 qt1 are in C and majorize the properly
infinite projection t2 t∗2 of C. The have all the same class [1] = 0 in K0 (E) and t2 t∗2
is full in E. Thus, by Lemma 4.2.6(ii), there are partial isometries v, w ∈ E with
v ∗ v = 1 − t1 pt∗1 , vv ∗ = t2 t∗2 , w∗ w = 1 − t1 qt∗1 and ww∗ = t2 t∗2 . Let S := t1 pt∗1 + v
and T := t1 qt∗1 + w, then S and T are isometries in E with SS ∗ = p0 , T T ∗ = q 0 ,
St1 pt∗1 = t1 pt∗1 and T t1 qt∗1 = t1 qt∗1 . Thus, S ∗ h(·)S = t1 pt∗1 h(·)t1 pt∗1 = t1 h(·)pt∗1
and T ∗ h(·)T = t1 h(·)qt∗1 .
It follows, that W ∗ S ∗ h(·)SW = T ∗ h(·)T for the unitary W = t1 V t∗1 + t2 t∗2 .
Hence, SW T ∗ ∈ C by Lemma 4.3.4(ii), and defines a Murray–von-Neumann equiv-
alence between p0 and q 0 . Thus [p] = [p0 ] = [q 0 ] = [q] in K0 (C). By Lemma
4.2.6(iv,b), there is U ∈ U0 (C) with U ∗ (p ⊕ 1 ⊕ 0)U = q ⊕ 1 ⊕ 0 . 

7. Additional remarks and further plans for Chp.4

plan: ??
Some of following is contained in Lem.4.2.6 and around.
(0.1) V (A) common name?
Recall that V (A) denotes the set of all Murray–von-Neumann equivalence
classes [p] of projections p ∈ A. It has partially defined commutative “addition”
[p] + [q] := [r + s], if there exist projections r, s ∈ A with sr = 0 and [r] = [p] and
[s] = [q].
If the addition of MvN-classes – [p] with [q], [q] with [r], [p] + [q] with [r] and
[p] with [q] + [r] – are all defined, then the associative law holds: [p] + ([q] + [r]) =
([p] + [q]) + [r].
The addition is always defined if the unit element 1M(A) of the multiplier
algebra M(A) of A is properly infinite, and then V (A) is a commutative semigroup.
If, in addition, A contains a full projection p, then [q] ∈ V (A) → [e11 ⊗ q] ∈
V (K ⊗ A) is an isomorphism of semigroups, and K1 (A) is naturally isomorphic to
the Grothendieck group of the semigroup V (A).

Lemma 4.7.1. Suppose that p is a full and properly infinite projection in a

C*-algebra A.
For any projection q ∈ A there exists a projection r ∈ pAp such that p − r is
full and properly infinite in A and q ∼ r.
For any two projections r, s ∈ A the sum [r] + [s] exists in V (A). Its repre-
sentative is full and properly infinite if one of the projections is full and properly
infinite.
[q] ∈ [p] + V (A) if and only if q is full and properly infinite in A.
616 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

If p ∈ A is a full and properly infinite projection, then for any projection p ∈ A

there exists a full and properly infinite projection r ∈ A with [q] + [r] = [p].
In particular, the Murray–von-Neumann equivalence classes of full and properly
infinite projections in A build a subgroup [p] + V (A) of V (A) with (unique) neutral
element [r] that satisfies [p] + [r] = [p].
If p ∈ A is full and properly infinite, then [p] + [q] = [p] + [r] in V (A), if and
only if, [q] = [r] in K0 (A).
K0 (A) is naturally isomorphic to the sub-semigroup [p] + V (A) of V (A).
If p ∈ A is a properly infinite projection, i.e., if there exist partial isometries
u, v ∈ A with u∗ u = v ∗ v = p, uu∗ ≤ p, vv ∗ ≤ p and u∗ v = 0, then p0 := p − vv ∗
is a properly infinite projection in A that generates the same ideal as p, and p0 Ap0
contains a copy of O2 unitally, in particular [p0 ] = 2[p0 ] in V (A).

Lemma 4.7.2. If p ∈ A is full and properly infinite, then u ∈ U(pAp) 7→

[(1 − p) + u] ∈ K1 (A) is a surjective map.
If A is unital, u ∈ U(A), p ∈ A projection with 0 = [u] ∈ K1 (A), kpu − upk < 1
and p and 1 − p both full and properly infinite in A. Then u is in the connected
component U0 (A) of 1A in U(A).

Above lem contained in K1 -injectivity/surjectivity stuff?

(1.0) Suppose that A contains a full and properly infinite projection p ∈ A.
Then:
The map u ∈ U(pAp) 7→ [u + (1 − p)] ∈ K1 (A) defines a group epimorphism
from U(pAp) onto K1 (A).
The algebra A contains also a full projection p0 with [p0 ] = 2[p0 ] in V (A), i.e.,
with the property that p0 Ap0 contains a copy of O2 unitally.
The natural semigroup morphism [q] ∈ V (A) 7→ [q] ∈ K0 (A) is surjective.
The set V (A) + [p0 ] is a subgroup of V (A) that is identical with the set of the
MvN–equivalence classes of all full properly infinite projections in A.
The restriction of the semigroup epimorphism V (A) → K0 (A) to V (A) + [p0 ]
is a group isomorphism from V (A) + [p0 ] onto K0 (A).
Suppose that A is unital and 1A is properly infinite.
It allows to define “additions” in U(A) and in Proj(A) by U ⊕ V := s1 U s∗1 +
s2 V s∗2 + (1 − s1 s∗1 − s2 s∗2 ) and p ⊕ q := s1 ps∗1 + s2 qs∗2 .
It defines an addition in U(A)/U0 (A) and in V (A) that is compatible with the
multiplication in U(A)/U0 (A), – respectively with the addition in V (A ⊗ K).
It turns out that q → [q ⊗ e1,1 ] ∈ V (A ⊗ K) induces an isomorphism from
(V (A), +) onto V (A ⊗ K) with the usual bloc-wise addition. (In fact: (p + q) ⊗ e1,1
is MvN-equivalent to p ⊗ e1,1 + q ⊗ e2,2 if pq = 0 and for each projection p ∈ A ⊗ K
there exists q ∈ A with [q ⊗ e1,1 ] = [p] in V (A ⊗ K).)
 7. ADDITIONAL REMARKS AND FURTHER PLANS FOR CHP.4 617

Furthermore, U ⊕ V is homotopic to (U V ) ⊕ 1 in U(A). [U ⊕ 1] = [U ] in K1 (A)

for all U ∈ U(A). It holds U ⊕ 1 ∈ U0 (A) iff [U ] = 0 in K1 (A). Moreover, if
T ∈ A is an isometry, such that 1 − T T ∗ is a full properly infinite projection then
[T U T ∗ + (1 − T T ∗ )] = [U ] in K1 (A) for all U ∈ U(A), and [U ] = 0 implies that
T U T ∗ + (1 − T T ∗ ) ∈ U0 .
(1.1) In particular, every C *-algebra A with a properly infinite unit 1A is K∗ -
surjective, i.e., the natural semigroup morphism V (A) → K0 (A) and the natural
group morphism U(A) → K1 (A) are surjective.
(1.2) Every unital s.p.i. algebras is K∗ -injective in the sense that: This is now
shown in Chp. 4 in case of unital p.i. C *-algebras, with help of the squeezing
property. Refere to it!

(0) V (A)+[1A ] is the subgroup of V (A) of all full properly infinite projections.
The natural morphism from V (A) to K0 (A) defines an isomorphism from
V (A) + [1A ] onto K0 (A).
The element [1A ] is the neutral element of V (A) + [1A ], if and only if,
A contains a copy of O2 unitally.
(1) The natural group morphism from U(A) to K1 (A) is surjective and has
kernel U0 (A).

(1.3) If A is a unital s.p.i. algebra, then 1A is properly infinite and the group
V (A)+[1A ] consists of all MvN-equivalence classes of full projections in A (because
every element of A is properly infinite).
There is a copy of O∞ = C ∗ (s1 , s2 , . . .) unitally contained in A, because A is,
in particular,
purely infinite, i.e., each non-zero element of A is properly infinite.
Exact reference! ??
We have seen in Chapter 3 that every s.p.i. algebra B satisfies the following
property:
For every finitely generated Abelian C*-subalgebra A ⊆ Bω with at most 1-
dimensional maximal ideal space Prim(A) there exists a C*-morphism

h : A ⊗ O ∞ → Bω with h(a ⊗ 1) = a ∀ a ∈ B .

(Notice that U = h(U ⊗ (tt∗ )) + h(U ⊗ (1 − tt∗ )) is homotopic to the unitary

h(U ⊗ (tt∗ )) + h(1 ⊗ (1 − tt∗ )) = T U T ∗ + (1 − T T ∗ ) with T := h(1 ⊗ t) if 1B , U ∈ A
and U is unitary, because the unitary group of C(S 1 , O2 ) is connected.)
Now use that K∗ (Bω ) = (K∗ (B))ω (as Z2 -graded groups), and get: A unitary
U ∈ B (respectively a projection p ∈ B) is homotopic to 1 in U(B) (respectively
to q ∈ B in Proj(B)), if and only if, it is homotopic to 1 (respectively to q) in the
unitaries (respectively projections) of Bω .
Thus, it suffices to proof the K1 -injectivity of Bω . This follows from the homo-
topy of U ∼h T U T ∗ + (1 − T T ∗ ) in U(h(C ∗ (U ) ⊗ O∞ )), where T := h(1 ⊗ t), for
a non-unitary isometry t ∈ O∞ .
618 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

More generally, if for every commuting pair (b1 , b2 ) of self-adjoint elements in

a given unital C *-algebra B and ε > 0 there exists isometries s, t ∈ B with s∗ t = 0
and ksbj − bj sk < ε (j = 1, 2) then B is K1 -injective.
This is, because then 1B is properly infinite, and – with ε := 1/4 – the following
criteria holds for u = b1 + ib2 with [u] = 0 in K1 (B) and p := ss∗ .
which criterium ? from Kt1 -injectivity proposition?????, squeezing
property?? ks∗ utk < 2ε.

(2) Transfer to here, and generalize, the unitary homotopy for (id ⊕h0 )k con-
sidered in Chapter 11. ??.

(3) Let J ⊆ E a closed ideal, and let H0 : D → E a *-monomorphism such that

H0 (D)0 ∩ E contains a copy of O2 unitally, and that there is an isometry T ∈ E
with T T ∗ H0 (D) = {0}.
Recall that S(H0 ; D, E) is the set of unitary equivalence classes [h] where
h ∈ Hom(D, E) is dominated by H0 . Then, obviously, [h] ∈ S(H0 ; D, E) with
h(D) ⊆ J is a sub-semigroup S(H0 ; D, E; J) of S(H0 ; D, E).
Let B a C *-subalgebra of E. We denote by V (B) the Murray–von-Neumann
equivalence classes of projections in B with (partial) addition given by sums of
orthogonal representatives. If M(B) has a properly infinite unit, then V (B) is
a commutative semigroup such that [p] ∈ V (B) → [e11 ⊗ p] ∈ V (K ⊗ B) is an
isomorphism from V (B) onto V (K ⊗ B).
We define a map θ0 : S(H0 ; D, E) → V (H0 (D)0 ∩ E) by θ0 (h) := [tt∗ ] for some
isometry t ∈ E with h := t∗ H0 (·)t .
Let B := H0 (D)0 ∩ N (H0 (D), J), where N (H0 (D), J) ⊆ E denotes the here-
ditary C *-subalgebra of E of the elements x ∈ E with xH0 (D) ∪ H0 (D)x ⊆ J,
i.e.,
N (H0 (D), J) = (πJ )−1 (Ann(πJ (H0 (D)), E/J)) .
There is a natural semigroup morphism [p]B ∈ V (B) 7→ [p]C ∈ V (C) where C :=
H0 (D)0 ∩ E.
The following lemma is used to compare the semigroups SR(C ; A, B) of Chapter
7 with WvN-equivalence classes of projections in H0 (D)0 ∩ N (H0 (D), J).

Lemma 4.7.3. Let (H0 ; D, E; J), B and θ0 as above (with [H0 ] + [0] = [H0 ]
and [H0 ] + [H0 ] = [H0 ]). Let C := H0 (D)0 ∩ E. Then:

(o) B is a closed ideal of C and V (B) is a hereditary sub-semigroup of V (C).

(i) The map θ0 is well-defined and is a semigroup isomorphism from
S(H0 ; D, E) onto the sub-semigroup of classes [p] ∈ V (C) with the
property that there is an isometry t ∈ E with [tt∗ ] = [p] in V (C).
(ii,a) Let V0 (C) the kernel of the natural morphism V (C) → K0 (C).
If T ∈ E is an isometry with T T ∗ H0 (D) = {0}, then T T ∗ ∈ C,

θ0 (S(H0 ; D, E)) = [T T ∗ ] + V0 (C) .

 7. ADDITIONAL REMARKS AND FURTHER PLANS FOR CHP.4 619

(ii,b) The Grothendieck group of [T T ∗ ] + V0 (C) is naturally isomorphic to the

kernel of K0 (C) → K0 (E).
If F := Ann(H0 (D), E) denotes the two-sided annihilator of H0 (D)
in E, then F is an ideal of C, the Grothendieck group G(H0 ; D, E) :=
H0 + S(H0 ; D, E) of S(H0 ; D, E), and it holds:

G(H0 ; D, E) ∼
= K0 (C/F ) ∼
= ker(K0 (C) → K0 (E)) .

(iii) Let V0 (B) denote the kernel of V (B) → K0 (E). [T T ∗ ] ∈ V0 (B) and
θ0 (S(H0 ; D, E; J)) = θ0 (S(H0 ; D, E)) ∩ V (B).

If T ∈ E is an isometry with T T ∗ H0 (D) = {0} then T T ∗ ∈ V0 (B),

V0 (B) = V (B) ∩ V0 (C) and

θ0 (S(H0 ; D, E; J)) = [T T ∗ ] + V0 (B) .

Proof. Let C := H0 (D)0 ∩ E.

(o): If b, c, e ∈ C and bH0 (D) ⊆ J, then H0 (D)cbe = cbeH0 (D) = cbH0 (D)e ⊆
J. Thus, B is an ideal of C. The same argument shows that [p]C = [q]C in V (C)
and p ∈ B implies q ∈ B and [p]B = [q]B in V (B). It follows that V (B) is a
hereditary sub-semigroup of V (C), in the sense that [p] + [q] ∈ V (B) ⊆ V (C)
implies [p], [q] ∈ V (C).
(i): See the arguments in the proof of Part (i) of Proposition 4.4.3.
(ii,a): Since T T ∗ H0 (a) = 0 = H0 (a)T T ∗ for all a ∈ D, the projection T T ∗ is
in B ⊆ C. Let r, s ∈ C canonical generators of a unital copy of O2 , and let p a
projection in C with [p] = 0 in K0 (E). Then q := rT T ∗ r∗ + sps∗ is a projection in
C with [q] = [T T ∗ ] + [p] = [1C ] + [p] in V (C), and there exists an isometry t ∈ E
with tt∗ ∈ C and [q] = [tt∗ ] in V (C).
To see this, we can use the isomorphism V (E) + [1] ∼ = K0 (E) and the fact that
V (E) + [1] is identical with the MvN-classes of all properly infinite elements: Since
[q] = [p] + [T T ∗ ] = [p] + [1] in V (E), the projection q is properly infinite and full in
E. Since [1] = 0 in K0 (E) it follows that [q] = [p] = 0 in K0 (E). Thus [q] = [1] in
V (E).
(ii,b): F → C → E defines an isomorphism K0 (G) ∼
= K0 (E) that makes
that the sequence 0 → K∗ (F ) → K∗ (C) → K∗ (C/F ) → 0 is split exact with
K∗ (C) → K∗ (E) ∼
= K∗ (F ) as splitting map.
(iii): Let [h] ∈ S(H0 ; D, E; J) and h = t∗ H0 (·)t. Then tt∗ ∈ C by (i) and
tt H0 (a∗ a)tt∗ ∈ J for all a ∈ D. Thus, H0 (D)tt∗ ⊆ J and tt∗ ∈ B. It follows that
∗

θ0 (S(H0 ; D, E; J)) ⊆ V (B) ⊆ V (C).

If θ0 (h) ∈ V (B) and h = t∗ H0 (·)t for some isometry t ∈ E, then tt∗ ∈ C and
[tt∗ ]C = [q]C in V (C) for some q ∈ B. It follows tt∗ ∈ B and h = t∗ (tt∗ )H0 (·)t
maps D into J, i.e., h ∈ S(H0 ; D, E; J).
Hence, θ0 (S(H0 ; D, E; J)) = θ0 (S(H0 ; D, E)) ∩ V (B).
620 4. COMPARISON AND ADDITION OF SOME C *-MORPHISMS

Now notice that ([T T ∗ ]+V (C))∩V (B) = [T T ∗ ]+V (B), because [T T ∗ ] ∈ V (B)
and V (B) is hereditary in V (C). 

Can the Grothendieck group of S(H0 ; D, E; J) be calculated with help of K∗ -

groups of B, C and J ?

Remark 4.7.4. The following observation yields alternative proofs of some

parts of Lemma 4.2.6, e.g. of Part (v,5).
A unitary u ∈ U(E) satisfies u tt∗ = tt∗ for some isometry t ∈ E, if and
only if, u is conjugate to (u ⊕ 1) + p[−1] , i.e., if there is a unitary V ∈ U(E) with
V ∗ uV = (u ⊕ 1) + p[−1] . The unitary V can be chosen such that [V ] = 0 in K1 (E).

Proof. If V ∈ U(E) satisfies V ∗ uV = (u ⊕ 1) + p[−1] = s1 us∗1 + (1 − s1 s∗1 ),

then uV s2 = V s2 , thus utt∗ = tt∗ for the isometry t := V s2 .
If t ∈ E is an isometry with utt∗ = tt∗ , then let P := t(1 − s1 s∗1 )t∗ . The
projection P is splitting, because P ≥ (ts2 )(ts2 )∗ and P (ts1 )(ts1 )∗ = 0. Since P is
MvN-equivalent to 1 − s1 s∗1 , there exists a unitary V1 with V1∗ P V1 = 1 − s1 s∗1 . It
follows V1∗ uV1 (1 − s1 s∗1 ) = (1 − s1 s∗1 ) and u1 := s∗1 V1∗ uV1 s1 is a unitary in E such
that u = s1 u1 s∗1 + (1 − s1 s∗1 ), i.e., V1∗ uV1 = (u1 ⊕ 1) + p[−1] .
We use now the “corrected” direct sum formula U ⊕0 V := (U ⊕ V ) + p[−1]
on the unitary elements U, V ∈ U(E), and get V1∗ uV1 = u1 ⊕0 1. It follows that
V2∗ (u ⊕0 1)V2 = (u1 ⊕0 1) ⊕0 1 for V2 := V1 ⊕0 1. The corrected direct sum is still
commutative and associative up to unitary equivalence by the unitaries Uc and Ud
defined in Equations (2.3) and (2.4), i.e., (U ⊕0 V ) ⊕0 W and U ⊕0 (V ⊕0 W ) are
unitarily equivalent by unitaries in U0 (E). We get that there is a unitary V in the
classes U0 (E)V1 such that V ∗ uV = u ⊕0 1.
We can replace V by V 0 := V (1 ⊕0 V ∗ ) that satisfies 0 = [V 0 ] ∈ K1 (E). 
 CHAPTER 5

Generalized Weyl–von Neumann Theorems

?? Revise section structure !!

Remove two-fold text !! very urgent!!!
We generalize in this chapter the Weyl–von-Neumann theorem along lines indi-
cated by Voiculescu and Kasparov ([798], [404], see [42] for an elementary account
and use Lemma 2.1.22 to see that they are special cases of Theorem A). The defi-
nition, viewpoints and results are different from the generalization of some aspects
of the classical Weyl–von Neumann theorem by S. Zhang in [845].
The generalized Weyl–von Neumann Theorem 5.6.2 will be an ingredient of
the proofs of Theorem A and of the important special case of Theorem M, – the
Theorem 6.3.1 –, considered in Chapter 6 under the additional assumption that the
C *-algebra B in Theorem M has residual nuclear separation (cf. Definition 1.2.3),
i.e., that the m.o.c. cone Crn ⊆ CPnuc (B, B) is separating for B. This additional
assumption will be removed finally in Chapter 12, by showing that in ultrapowers
and asymptotic algebras those additional conditions are in the considered cases
for sufficiently big separable C *-subalgebras are satisfied, all this only by using
combinations of results obtained in all Chapters 7-11 together.
We assume in Chapter 6, in addition to the assumptions of Theorem M, that
the related universal weakly residually nuclear C *-morphism Hrn : B → M(B)
from B into its multiplier algebra M(B) is non-degenerate, faithful and that the
induced ideal-system action of 0(B) on B is simply the identity map ????
(1)
???? HERE and a bit below ???????
Explain Hrn by use of action and by use of suitable C.
In case of trivially graded stable separable A, B and non-degenerate m.o.c. cone
C we should have exact sequences

0 → Ext(C; A, B) → K0 (πB (HC (A))0 ∩ Q(B)) → K0 (Q(B)) ∼

= K1 (B) → 0 ,

and

0 → KK(C; A, B) → K1 (πB (HC (A))0 ∩ Q(B)) → K1 (Q(B)) ∼

= K0 (B) → 0 .

1 The author got in between, by a self-contained study of relations between coherent and non-

coherent Dini-spaces, refining and applying basic results from lattice theory, that each separable
stable C *-algebra has this property anyway. But here we use an other way of proof that considers
properties of ultra-powers of strongly purely infinite C *-algebras.

621
622 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

Since HC (A)0 ∩ M(B) contains a copy of O2 unitally we get K∗ -surjectivity.

What about K∗ -injectivity here??
For K0 (HC (A)0 ∩ M(B)) only the full properly infinite projections in HC (A)0 ∩
M(B) are uniquely defined by its value in K1 (HC (A)0 ∩ M(B)).
We must check if HC (A)0 ∩ M(B) is K1 -bijective! ?
... for the considered stable separable B and that there exists a C *-morphism
h : A → M(B) that defines the action of Prim(B) on A.
Relation to more general C and induced actions have to be explained / refer to
Chapter 3.
This ??? additional assumptions will be removed in Chapter 12 completely
using results of Chapters 3, and 7 -11 .
Here in Chapter 5 we allow weaker/stronger ??? assumptions.
Perhaps Chapter 3 should present the relations between actions and cones ...
In particular, C 0 ∩ Qs (B) should be σ-sub-Stonean ...???
Should here also obtain finally that O2 ⊗ O2 ⊗ · · · is (unitally) contained in
O2 , and that this allows to prove that O2 can be written as its own infinite tensor
product
??? if each unital ind-lim of O2 is O2 .
Need the approximate unitary equivalence of all unital endomorphisms of O2
and more generally of On , O∞ .
¡ ¡ ¡ Given at end of Chapter 4 ??
Its generalization to weakly residually nuclear maps – and, more generally and
systematic, to maps in a given m.o.c. cone C ⊆ CP(A, B) – will be used in Chapters
6 and 12 for the final proof of Theorem K.
Compare above and below text
First we prove an almost “tautological” universal version of a generalized Weyl–
von-Neumann theorem (Theorem 5.6.2) and show some of its consequences:
We apply it to give an alternative proof for the stability criteria of M. Rørdam
and J. Hjelmborg for σ-unital C *-algebras and its extensions (cf. [373], [688]).
We specialize it to the cases of extensions with splitting c.p. maps that are weakly
nuclear, weakly residually nuclear, or are in suitable general m.o.c. cones C ⊆
CP(A, B). It gives the base to define general cone-related Ext-groups Ext(C; A, B)
and we display explicit defining relations that will be used in Chapters 8 and 9 to
establish the functorial isomorphism of Ext(C; A, B) with KK(C; A, B(1) ) – alone
the lines of early work of G. Kasparov – and its isomorphism with the a generalized
version of “Rørdam groups” R(C; A, B) defined in Chapter 7, that generalize groups
of asymptotic morphisms appearing (implicitly) in early work of G. Elliott and
M. Rørdam.
 5. GENERALIZED WEYL–VON NEUMANN THEOREMS 623

Later applications of our general WvN-theorems (in the spirit of Voiculescu

and Kasparov) for the proofs of isomorphisms between the in Chapters 1, 4 and 7
– 9 considered types of operator class groups use the key results from Chapters 3
and 4.
We need some basic definitions, e.g. our version of unitarily homotopic mor-
phisms in the following Definition 5.0.1 will be used frequently.
Recall that the C *-algebra M(A) ⊆ L(A, A) denotes the multiplier algebra of
a C *-algebra A.

Definition 5.0.1. Let hj : A → M(B), j = 1, 2, C *-morphisms, and let V a

completely positive contraction from A into the multiplier algebra M(B) of B.
We call h1 and h2 unitarily homotopic if there is a norm-continuous map
t 7→ U (t) from the non-negative real numbers R+ into the unitaries in M(B), such
that,

(i) h2 (a) − U (t)∗ h1 (a)U (t) ∈ B for all t ∈ R+ and a ∈ A , and

(ii) limt→∞ k h2 (a) − U (t)∗ h1 (a)U (t) k = 0 for all a ∈ A .

Let B σ-unital and stable. Question:

What happens if we require only that
t 7→ U (t) satisfies (i, ii) and is only strictly continuous?
Comments on this question to be checked:
It says then only that h1 : A → M(B) and h2 : A → M(B) are unitarily


equivalent in C := Cb,st [0, ∞), M(B) / Cb ([0, ∞), B), where we take the natural


injective “constant” C *-morphism M(B) ⊂ Cb,st [0, ∞), M(B) .
In case of our Definition 5.0.1 we get that U ∈ Cb ([0, ∞), M(B)).
Is the strict topology – restricted to Cb ([0, ∞), B) – is the same as the norm
topology on Cb ([0, ∞), B)?
We have often to do with the case where B is stable and σ-unital. Then
U(M(B)) = U0 (M(B)), and there exist y1 , . . . , yn ∈ M(B) with yk∗ = −yk such
that U (1) = exp(y1 )·. . .·exp(y1 ). Then h2 (a) − U (1)∗ h1 (a)U (1) ∈ B for all a ∈ A .
Let V := πB (U (1)) Thus, πB ◦ h2 = V ∗ πB ◦ h1 (·)V ∈ Q(B) . and V ∈ U0 (Q(B)).
If, in addition, h2 is unitarily equivalent to δ∞ ◦ h2 in M(B), then we can replace
here πB ◦ h2 by πB ◦ δ∞ ◦ h2 .
The path t 7→ U (t) ∈ U(M(B)) defines an element U in U(M(C0 (R+ , B))) =
U0 (M(C0 (R+ , B))) with the property h2 (a)U − U h1 (a) ∈ C0 (R+ , B), where we use
the natural unital inclusions M(B) ⊂ M(C(R+ , B)) ⊆ M(C0 (R+ , B))
Notice here that C0 (R+ , B) is again stable and σ-unital if B has this properties.
If M(B) contains a copy of O2 unitally, then we can define Cuntz addition
h1 ⊕ h2 on Hom(A, M(B)) (cf. Chapter 4). We say that “h1 asymptotically
absorbs h2 ” if h1 and h1 ⊕ h2 are unitarily homotopic.
624 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

The C *-morphism h1 asymptotically dominates a completely positive con-

traction V : A → M(B) if there is a norm-continuous map t 7→ S(t) from the
non-negative real numbers R+ into the isometries in M(B), such that,

(i) V (a) − S(t)∗ h1 (a)S(t) ∈ B for t ∈ R+ and a ∈ A, and

(ii) limt→∞ k V (a) − S(t)∗ h1 (a)S(t) k = 0 for all a ∈ A.

Notice that h1 and h2 are unitarily homotopic, (respectively h1 asymptotically

dominates V ), if and only if, h1 and h2 are unitarily equivalent in (respectively h1
dominates V in) Cb (R+ , M(B))/ C0 (R+ , B) ⊃ M(B) . If B is unital, then this
algebra is the same as Q(R+ , B) := Cb (R+ , B)/ C0 (R+ , B).
This “algebraic” reformulation allows to apply the observations on Cuntz ad-
ditions from Sections 3 and 4 of Chapter 4, e.g. it implies immediately that h1
asymptotically absorbs h2 , i.e., h1 is unitarily homotopic to h1 ⊕ h2 , if and only
if, h1 asymptotically dominates h1 ⊕ h2 and h2 ⊕ h2 is unitarily homotopic to h2 ,
cf. Proposition 4.3.5(i,iii).
The asymptotic domination of all unital h1 , h2 : O2 → E
(of each other) should be clear ??
Compare all arguments with observations in Chapter 4!!!
∞
From nuclearity of O2 (and D2 := O2 ⊗ ∼
= O2 ?) one gets:
O∞ ∈ O2 0 ∩Q(R+ , O2 ) unitally, and that there exist paths of approximate inner
u.c.p. maps Vt : O2 → O2 with V∞ : O2 → Q(R+ , O2 ) is a unital C *-morphism with
V∞ (O2 ) ⊆ O2 0 ∩ Q(R+ , O2 ) implemented by some isometry in T ∈ Q(R+ , O2 ) with
T ∗ aT = V∞ (a) for a ∈ O2 . Thus, the images of h1 and h2 commute with unital
copies of O2 and 1-step dominate each other in Q(R+ , E) by nuclearity of O2 . It
implies that h1 and h2 are unitary equivalent by a unitary in Q(R+ , E). It means
that unital C *-morphisms h1 , h2 : O2 → E are unitary homotopic (with B := E
and A := O2 ).
The unitary homotopy of id with δ2 on O2 , implies that id is unitary homotopic
to an endomorphism k : O2 → O2 with k(O2 )0 ∩O2 ∼ = O2 , and moreover there exists
unital C *-morphisms k, k 0 : O2 → O2 with k 0 (O2 ) ⊆ k(O2 )0 ∩ O2 and linear span
of k(O2 ) · k 0 (O2 ) dense in O2 .
Thus, all unital h1 , h2 : O2 → E are unitarily homotopic inside hk (O2 ) ⊆ E to
unital homomorphisms gk : O2 → hk (O2 ) ⊆ E such that gk (O2 )0 ∩ E contains a
copy of O2 unitally. This implies that both hk (O2 )0 ∩ Q(R+ , E) contain (possibly
different) unital copies of O2 .
Since they 1-step dominate each other and commute with (possibly different)
copies of O2 in Q(R+ , E), they are unitarily equivalent in Q(R+ , E).
It means that unital h1 , h2 : O2 → E are unitarily homotopic with respect to
A := O2 and B := E. In particular h1 is homotopic to some unitary equivalent of
h2 , because near generators of copies of O2 in E are homotopic.
 5. GENERALIZED WEYL–VON NEUMANN THEOREMS 625

This applies to all unital C *-morphisms h1 , h2 : O2 → E of O2 into a unital C *-

algebra E and shows that they are unitarily homotopic (with respect to A := O2
and B := E):
For every unital C *-algebra E with properly infinite unit element that satisfies
0 = [1E ] ∈ K0 (E) there is up to unitary homotopy exactly one unital C *-morphism
h : O2 → E .
Does the latter not require that E is K1 -injective?
Answer: No, as above/below considerations show!
It shows that unitary equivalence implies unitary homotopy, but in general even
not point-norm homotopy of h1 and h2 in Hom(A, B).
A unitary homotopy with norm-continuous path of unitaries t 7→ U (t) ∈
U(M(B)) implies homotopy – in Hom(A, M(B)) with point-norm topology – if
A is σ-unital and U(M(B)) = U0 (M(B)) . In particular, h1 ⊗ idK and h2 ⊗ idK
are homotopic in Hom(A ⊗ K, J ⊗ K) , if h1 and h2 are unitarily homotopic in
M(B ⊗ K) and if J is an ideal of M(B) that contains h1 (A).
Unitary equivalent C *-morphisms h1 , h2 : A → M(B) are both non-degenerate
if one of its is non-degenerate (i.e., hk (A)B = B).
But this is not necessarily the case for unitarily homotopic C *-morphisms:
For example let B := A := K, h1 (a) := a, h2 (a) := T aT ∗ with T ∈ L(`2 ) an
isometry with (1−T T ∗ )`2 of infinite dimension (i.e., related to a unital copy of O2 ).
One can find here a norm-continuous path t ∈ [0, ∞) : U (t) ∈ U(M(K)) of unitaries
in L(`2 ) with U (0) = 1 and kU (t)∗ aU (t) − T aT ∗ k → 0 for t → ∞. This particular
sort of unitary homotopy can be given explicit by using that U(M(K)) = U0 (M(K))
and the fact that any trace-preserving *-endomorphism h2 of the algebra K of
compact operators of a separable (complex) Hilbert space is unitary homotopic to
h1 := idK .
(One can use here that the set of all isometries T ∈ L(`2 (N)) ∼ = M(K) with
∗
(1−T T )`2 of infinite dimension is path-connected in operator norm to the isometry
T (δn ) := δ2n , n = 1, 2, . . ..)
Notice that it is in general not possible to start with the additional condition
U (0) = 1 in the definition of unitary homotopy as the following counterexample
shows:
Let H := `2 (N), B := K(H), a0 ∈ `∞ (N) ⊆ L(H) the projection given by
(0, 1, 0, 1, . . .), and A := C ∗ (1, a0 ) ⊆ `∞ (N).
There exists T ∈ M(B) = L(`2 ) with T a0 − a0 T 6∈ K, T ∗ = −T , kT k < 1,
e.g. T := 3−1 (S ∗ − S) for the 1-step forward shift S of `2 (N) (Toeplitz operator
with index = −1), S(α1 , α2 , . . .) = (0, α1 , α2 , . . .) . Then exp(−T )a0 exp(T ) − a0 6∈
K =: B .
626 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

Let h1 := idA and h2 (a) := exp(−T )h1 (a) exp(T ), i.e., exp(T )h2 (a) −
h1 (a) exp(T ) = 0 , U (t) := exp(T ) for all t ∈ [0, ∞) (constant). In particular
h2 (a) − U (t)∗ h1 (a)U (t) = 0 ∈ K =: B for all t ∈ [0, ∞).
It does not exists t 7→ V (t) ∈ U(M(K)) norm-continuous, unitary, with V (0) =
1, V (t)∗ h1 (a)V (t) − h2 (a) ∈ K for all t ∈ [0, ∞) and a ∈ A, because it would imply
[exp(T ), h1 (a0 )] ∈ K.
A similar example is on some other place. ??
????????
because there are natural isomorphisms KK(A, B) ∼
= Ext(A, SB) for separable
stable σ-unital A and B, and because

KK(A, B) ∼
= ker(K1 (H0 (A)0 ∩ Q(B)) → K1 (Q(B)) ∼
= K0 (B)) ,

cf. Chapter 8.
We shall see below and in Chapter 7, that it does not matter if we replace in
the above definitions of asymptotic domination and of asymptotic absorption the
norm-continuity of t 7→ S(t), respectively of t 7→ U (t) by the – much weaker – strict
continuity if A is stable, h1 is non-degenerate and V (A) ⊆ B (respectively h2 (A) ⊆
B, respectively h2 is also non-degenerate and h1 (A) ∪ h2 (A) ⊆ B). It means, that
we can replace a weakly continuous path t 7→ S(t) (respectively t 7→ U (t)) in this
cases by norm-continuous paths that do the same job, but are rather different from
the given weakly continuous paths and have a different isometry S(t0 ) (respectively
unitary U (t0 )) at the starting point t0 ≥ 0.
Give Ref’s to places of proofs
of ‘‘strictly ..." to ‘‘norm- ...’’ !!!
Give example of different start points of paths !!!
The situation of unital C *-morphisms form O2 into the stable corona Qs (B)
of B (see Definition 5.4.7) is different:

The homotopy classes of the unital C *-morphisms h : O2 → Qs (B) are equiva-

lent to U(Qs (B))/U0 (Qs (B)), thus are in bijective correspondence to K1 (Qs (B)) by
the K1 -bijectivity of stable coronas, cf. Proposition 4.2.15. Thus, they are classified
by K0 (B) ∼= K1 (Qs (B)).
But there is only one full non-unital C *-morphism h : O2 → Qs (B) up to
unitary homotopy. <-- Example? Why?
Sort above/below text (red and blue)!!!
?????? Give Ref’s!!!
The unital C *-morphisms h : O2 → M(B) for σ-unital stable B are all homo-
topic, because of U(M(B)) = U0 (M(B)) by the Cuntz-Higson Theorem, cf. Propo-
sition 4.2.15, and unital C *-morphisms h0 , h1 : O2 → A are unitarily homotopic to
each other, if and only if, the unitary u := h0 (s1 )h1 (s∗1 ) + h0 (s2 )h1 (s∗2 ) is in U0 (A),
 1. STRICT CONVERGENCE IN MULTIPLIER ALGEBRAS 627

by Corollary ?? cf. Section ?? of Chapter 4. Clearly they are all homotopic if and
only if U(A) = U0 (A).
But the homotopy classes of unital h ∈ Hom(O2 , E) is identical with
U(E)/U0 (E), which is isomorphic to K1 (E) if and only if E is K1 -injective
(in addition).
All non-unital C *-morphisms h ∈ Hom(O2 , E) with the property that the
projections h(1) and 1 − h(1) are full in E and that 1 − h(1) is properly infinite are
homotopic in Hom(O2 , E) and are unitary homotopic. Clearly, the latter happens
for all non-unital h ∈ Hom(O2 , E) if E is simple and purely infinite.
Compare for the non-purely infinite case the elements of Hom(O2 , E) for the
stable corona E := Qs (M2∞ ) of M2∞ in Section 8.

1. Strict convergence in Multiplier algebras

Here we give some later used unconditional strict convergence results for com-
pletely bounded maps defined with help of certain strictly “unconditional square-
summable” sequences. We introduce some natural C *-subalgebras in M(B) of
stable (not necessarily σ-unital) B, and describe a certain “expanding” path of
projections in O2 ⊗ K ⊂ M(B) for σ-unital stable C *-algebras B.
A very important point for application of extension theory is the “stability”
of the σ-unital C *-algebras under consideration. We select for reference some ele-
mentary equivalent descriptions in Part (8). There are others and other proofs for
the here given equivalences and descriptions. We use (or misuse) our proof of the
following Remarks 5.1.1 – in particular of Parts (1) and (8) – to give also a survey
on multiplier algebras and their strict topology from our viewpoints and with our
terminology ( 2 ).

Remarks 5.1.1. The strict topology on M(B) ⊆ L(B) is given by the

semi-norms d ∈ M(B) 7→ kLd (a)k + kRd (b)k, where a, b ∈ B ( 3 ), this topology is
different to the strong topology on M(B) ⊆ L(B), induced by the strong operator
topology on L(B). The latter is defined by the semi-norms T 7→ kT (a)k with a ∈ B
and T ∈ L(B) only.
(Since M(B) has a natural C *-algebra imbedding as a C *-subalgebra of B ∗∗ ,
we can write Rd (a) as ad and Ld (a) as da for a ∈ B and d ∈ M(B) ⊆ L(B).)
The strong and the strict topology coincide on the set of normal elements
d ∈ M(B), because kadk = kda∗ k for d ∈ M(B) with d∗ d = dd∗ and a ∈ B. In
particular this topologies coincide on the unitary group of M(B) and on bounded
sets of self-adjoint elements in M(B).
A bounded net in {bτ } ⊂ M(B) converges strongly (resp. strictly = *strongly)
to an element of b ∈ M(B) if lim kbτ c − bck = 0 (resp. if lim kbτ c − bck = 0 and

2 Keep always in mind the simple observation in Part (2) of the the following Remarks.
3 In some cases, e.g. B = K, it is on bounded parts the same as the ∗ -strong topology.
628 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

lim kdbτ − dbk = 0) for all c in a subset X ⊆ M(B) (resp. for all d in a subset
Y ⊆ M(B) such that span(XB) and span(BY ) are dense in B.
If B is σ-unital and e ∈ B+ is a strictly positive element, the strict topology is
metrizable with metric ρ(b, c) := k(b − c)ek + ke(b − c)k on bounded parts of M(B).
P Pk
A series n dn means the sequence of its partial sums ek := n=1 dn with dn ∈
P
M(B). It converges unconditional strictly to T ∈ M(B) if n dγ(n) converges
P
also strictly in M(B) for each permutation γ of N. The series n dn denotes also
the sum (if it exists).
P
If n dn converges unconditional in strict topology, then for its sum holds
P P
n dγ(n) = n dn for each permutation γ of N, because this is true for the se-
∗ ∗
P
ries n f (d n and each linear functional f ∈ B , where B is considered as a
)
(complemented) subspace of M(B)∗ by B / M(B).
(1) A C *-morphism h : A → M(B) extends uniquely to a unital and strictly
continuous C *-morphism M(h) from M(A) into M(B), if and only if, the linear
span of the set h(A) · B is dense in B. (In fact the closed linear span of h(A) · B is
identical with the set of elements h(a) · b, a ∈ A and b ∈ B, because the proof of
G.K. Pedersen shows that the y ∈ X in the factorization e · y = x can be taken in
the closed convex cone B+ · x of X, cf. G.K. Pedersen’s proof a C *-version of the
Cohen Factorization Theorem outlined in Section 11 of Appendix A.
Def. of ‘‘non-degenerate’’ is used first in Part (7), but also
in Chapters 1, 2 and 3
(and perhaps Chapter 4 ?)
We call the C *-morphism h non-degenerate if h(A)B is dense in B. A C *-
subalgebra C ⊆ M(B) is a non-degenerate subalgebra if CB is dense in B.
(2) Suppose that x = (x1 , x2 , . . .) and y = (y1 , y2 , . . . ) are sequences in M(B)
xn x∗n and yn yn∗ converge strongly to elements X and
P P
such that the sums
Y of M(B)+ . We identify `∞ (M(B)) naturally with the multiplier algebra of
c0 (B) ∼
= B ⊗ c0 .
Original dn have been changed to bn
for later notational reasons.
Check changes in proofs (!), other references, applications!!
Then, for every bounded sequence b = (b1 , b2 , . . .) ∈ `∞ (M(B)), the series
yn bn x∗n converges strictly and unconditionally to an element Γ(b1 , b2 , . . .) in
P

M(B) . It defines a strictly continuous and completely bounded linear map

X
Γ : (b1 , b2 , . . .) ∈ `∞ (M(B)) = M(c0 (B)) 7→ yn bn x∗n ∈ M(B) .

The mapping Γ maps c0 (B) into B and has cb-norm

1/2
k Γ kcb ≤ kXk · kY k .

This convergence observation turns the proofs of Lemma 5.1.2, Proposition

5.4.1 and some others into straight-forward calculations – sometimes left to the
 1. STRICT CONVERGENCE IN MULTIPLIER ALGEBRAS 629

reader, as e.g. in the proof of Corollary 2.2.11(i) where this observation is used
implicitly.
(3) The applications of the unconditional strict convergence of the expressions
Γ – observed in Remark (2) – use sometimes quasi-central commutative approximate
units of σ-unital C *-algebras B to construct suitable sequences (xn ) and (yn ) that
satisfy the assumptions in Part (2):
Let e ∈ B+ any strictly positive element in B of norm kek = 1, A ⊆ M(B)
a separable C *-subalgebra, and take a linear filtration of A by vector sub-spaces
Xn ⊆ A of finite dimension with the properties that Xn = Xn∗ := {x∗ ; x ∈ Xn } ⊆
S
Xn+1 , dim(Xn ) < ∞ and n Xn is dense in A.
Separation arguments for convex combinations of f (e) with suitable non-
decreasing functions f show that there exists a strictly decreasing zero-sequence
α1 > α2 > · · · in (0, 1) and non-decreasing functions fn ∈ C0 ((0, 1])+ ,
n := 0, 1, 2, . . ., with properties f0 (t) := t, fn (t) = 0 for t ∈ [0, αn ],

fn fn+1 = fn , kfn k = 1, and kf0 − fn f0 k < 4−n ,

and the property that, for each b ∈ Xn and m > n, the commutators have estimates

k [fn (e), b] k + k[(fm (e) − fn (e))1/2 , b] k < 4−n kbk .

The properties imply fn |[ε, 1] = 1 and k e − fn (e)e k < ε for n ≥ 1 − log(ε).

(4) Especially let xn := yn := gn := (fn (e) − fn−1 (e))1/2 for n > 1 and
1/2
x1 := y1 := g1 := f1 (e) in Remark (2) with fn selected for Xn ⊆ A ⊆ M(B) as
in Remark (3).
The sum n gn∗ gn converges strictly to 1 and the in Remark (2) defined map
P

Γ : `∞ (M(B)) → M(B) with xn := yn := gn∗ is a unital completely positive map

with a − Γ(a, a, . . .) ∈ B for all a ∈ A.
(5) If g1 , g2 , . . . are as in Remarks (3) and (4), constructed for a given filtration
Xn of A ⊆ M(B) and a given strictly positive contraction e ∈ B+ , and if d1 , d2 , . . .
P
is a bounded sequence in M(B) with n kdn gn − gn dn k < ∞, then the series
P
n dn gn converges strictly to an element d ∈ M(B).

In particular, let P ⊆ N and (dn ) defined by dn := 1 if n ∈ P and dn := 0 if

P
n ∈ N\P . The corresponding positive contraction S := SP := n∈P gn ∈ M(B)+
is not necessarily in B.
But we get from Remark (4) that

Sa − aS , SaS + (1 − S 2 )1/2 a(1 − S 2 )1/2 − a ∈ B ∀ a ∈ A .

The operator S satisfies S 2 = n∈P gn2 if |p − q| > 1 for all p, q ∈ P with p 6= q.

P

(6) Suppose that a sequence of contractions d1 , d2 , . . . ∈ M(B), a self-adjoint

contraction a∗ = a ∈ M(B) and the sequence g1 , g2 , . . . defined in Remark (4)
satisfy
P

(i) ` ν` < ∞ for ν` := supn≥` max1≤k≤n kdk gn − gn dk k ,
630 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

P
(ii) ` µ` < ∞ for µ` := supn≥` kgn a − agn k, and
(iii) limn→∞ kgn d∗n adn+1 gn+1 k = 0 .
P
Then d1 , d2 , . . ., a and d := n dn gn satisfy

d∗ ad − Γ(d∗1 ad1 , d∗2 ad2 , . . .) ∈ B .

(7) D is a corner of B, if D is a hereditary C *-subalgebra of B and D+Ann(D)

is a non-degenerate C *-subalgebra of B.
It says equivalently that there is a projection p ∈ M(B) with D = pBp .
A subset X ⊆ B generates a corner of B, if the closure of the linear span of
C (X) · B · C ∗ (X) is a corner of B.
∗

(In fact the set of products c · b · c with b ∈ B and contractions c ∈ C ∗ (X)+

in an approximate unit of C ∗ (X) is identical with the hereditary C *-subalgebra of
B generated by C ∗ (X) as the Cohen factorization theorem [621, Thm. 4.1] shows,
cf. also Theorem A.11.1.)

(8) Recall that B is a stable C *-algebra if there exists an isomorphism ψ

from B ⊗ K onto B, where K := K(`2 ) denotes the compact operators K := K(`2 )
on `2 (N).
This is equivalent to the existence of a C *-algebra C and a C *-algebra isomor-
phism from C ⊗ K onto B.
The following criterium is necessary and sufficient for the stability of a (not
necessarily σ-unital) C *-algebra B:
There exists a sequence s1 , s2 , . . . of isometries in M(B) such that n sn s∗n con-
P

verges strictly to 1 in M(B).

Given a sequence s1 , s2 , . . . ∈ M(B) of isometries with n sn s∗n strictly con-
P
∗
P
vergent to 1, then δ∞ (b) := n sn bsn for b ∈ M(B) defines a faithful strictly
continuous unital *-endomorphism of M(B).
It holds u∗ δ∞ (b)u = n tn bt∗n by a unitary u ∈ M(B) with tn = usn for n =
P

1, 2, . . . if t1 , t2 , . . . is any other sequence of isometries in M(B) with n tn t∗n = 1

P

(strictly convergent).
The unitary equivalence class [δ∞ ] has representatives δ∞ : M(B) → M(B)
determined by s1 , s2 , . . .. We call this class (and sometimes the endomorphisms
P
n sn (·)sn in this class) the infinite repeat (on M(B)) of the identity mapping
idM(B) of M(B). Then δ∞ (b) (or its unitary equivalence class [δ∞ (b)]) is the
infinite repeat of the element b ∈ M(B).
There exists a C *-morphism µ : K → M(B) with following properties (a,b,c) :

(a) µ(K) ⊂ δ∞ (B)0 ∩ M(B),

(b) δ∞ (b)µ(k) ∈ B for b ∈ B and k ∈ K, and
 1. STRICT CONVERGENCE IN MULTIPLIER ALGEBRAS 631

(c) the (unique) C *-morphism ϕ : B ⊗ K → B with ϕ(b ⊗ k) = δ∞ (b)µ(k) is

an isomorphism from B ⊗ K onto B that maps J ⊗ K onto J for each ideal
J of B.

If a C *-morphism µ : K → M(B) satisfies conditions (a,b,c) then µ(K)·B = B.

The C *-morphisms µ : K → M(B) with properties (a,b,c) are uniquely defined
sn s∗n = 1 – up to
P
by the given sequence s1 , s2 , . . . ∈ M(B) of isometries with
0
unitary equivalence by unitaries in δ∞ (B) ∩ M(B).
One of the C *-morphisms µ : K → δ∞ (B)0 ∩ M(B) with properties (a,b,c) can
be defined by µ(pj,k ) := sj s∗k for j, k ∈ N, if the unitary equivalence class [δ∞ ] is
realized by given isometries s1 , s2 , . . ., i.e., if δ∞ (·) := n sn (·)s∗n .
P

The isomorphism ϕ from B ⊗ K onto B defined by µ : K → M(B) with proper-

ties (a,b,c) has the property that the corresponding C *-morphism b 7→ ϕ(b ⊗ p1,1 )
is a *-endomorphism of B, that is approximately 1-step inner as a c.p. contraction.
The C *-morphism ϕ((·) ⊗ p1,1 ) is unitarily homotopic to idB if B is σ-unital.
If b ∈ B+ , then ϕ(b ⊗ e1,1 ) is Murray–von-Neumann equivalent to b. In partic-
ular, ϕ(J ⊗ K) = J for each J ∈ I(B).
This property of the “natural” ϕ is different from the properties of very random
and possibly not well-behaved *-isomorphism ψ from B ⊗ K onto B – as allowed in
the definition of stability –, because in general the isomorphism ϕ ◦ ψ −1 of B does
not fix the ideal-system I(B) of B and is not approximately unitarily equivalent to
idB by unitaries in M(B).
The non-degenerate *-monomorphism µ : K → M(B) with properties (a,b,c)
has the“dual” property that

µ(K)0 ∩ M(B) = δ∞ (M(B)) .

It implies that δ∞ (M(B)) is closed in M(B) with respect to the strict topology on
M(B).
The strictly continuous extension M(µ) : M(K) ∼ = L(`2 ) → M(B) of µ is a
0
unital *-monomorphism, has image in δ∞ (M(B)) ∩ M(B) and the norm-closed
unit-ball of M(µ)(L(`2 )) is also closed in the strict topology of M(B).
The strictly closed C *-subalgebra δ∞ (M(B))0 ∩ M(B) of M(B) can be larger
than M(µ)(L(`2 )), e.g. if the center of M(B) is not trivial.
For each non-degenerate C *-morphism ψ : K(`2 ) → M(B), the topology in-
duced via M(ψ)−1 on M(K) = L(`2 ) by the strict topology on M(B) coincides on
bounded parts with the *-strong operator topology on L(`2 ).
(9) The (s1 , s2 , . . .)-existence criteria for stability in Remark (8) shows that
B and B +h(A) are stable if A is stable and h : A → M(B) is a non-degenerate
C*-morphism.
In particular, B is stable if B contains a non-degenerate stable C *-subalgebra.
632 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

To be worked out in detail:

(10) Let C1 , C2 , . . . a contable sequence of separable unital C *-algebras, K⊗D
a σ-unital stable C *-algebra, and let B := K ⊗ D ⊗ C1 ⊗ C2 ⊗ · · · the infinite tensor
product. We define unital C *-subalgebras G1 ⊂ G2 ⊂ · · · of M(B) by the following
conditions:
To be defined. Sort of infinite E(Cn , Cn+1 ) ⊆ M(B)
build with help an approximately central approximate unit of B that
is approximately central for h(A) of a given C *-morphism
h : A → M(B).

?????
Then for every C *-morphism h : A → M(B) of a separable C *-algebra A there
exists n0 := n0 (h(A)) ∈ N such that [h(A), Gn ] ⊆ B for all n ≥ n0 and, for each
a ∈ A, k[h(a), gn ]k → 0 for any sequence of contractions gn ∈ Gn .
(With n0 depending from an approximate unit in K ⊗ D that is quasi-central
for h(A).)
TO BE FILLED IN – for almost central sequences...
Lemma, ???? Idea ???, corona property ...
Case B := K ⊗ D ⊗ C1 ⊗ C2 ⊗ · · · A ⊆ M(B) separable C *-subalgebra.
Then one can find tensorial residuum if 1 ⊗ 11 ⊗ · · · ⊗ 1nk ⊗ Cnk+1 ⊗ · · · commuting
modulo B with A... provided that the Cn are unital simple and separable ... MORE
conditions ???

Proofs of Remarks (1)–(10):

We say here something about the different definitions of multipliers, multiplier
algebras, its relation to factorization properties of C *-algebra modules and the
interrelation between the (by them defined) strict, strong and weak* topologies on
bounded subsets of C *-subalgebrasof multiplier algebras. And we explain the often
required “non-degeneracy” of modular morphisms for the often needed verification
of unconditional strict convergence of series.
(0.1) General multiplier algebras:
If B is a Banach algebra and ρ : B → L(X) is an algebra homomorphism,
then the “action” ρ of B on X, respectively the (left) B-module X defined by
b · x := ρ(b)(x) is called non-degenerate if span(ρ(B)X) is dense in X.
To simplify some arguments we use later the special case for C *-algebras B of
the Cohen factorization theorem ( 4 ) given by G.K. Pedersen [621, Thm. 4.1] (with
13 lines of proof):
4 P. Cohen [?] proved 1959 that Banach algebras B with bounded approximate left-unit

have the factorization property. In 1964, E. Hewitt [363] and P. Koosis [491] showed that this
factorization property also holds for non-degenerate Banach B-modules.
 1. STRICT CONVERGENCE IN MULTIPLIER ALGEBRAS 633

If a C*-algebra B acts on a Banach space X from left with kb · xk ≤ kbkkxk and

if span(B · X) is dense in X, then for each ε > 0 and x ∈ X there exists e ∈ B+
and y ∈ X with e · y = x, kek ≤ 1, ky − xk < ε.
In fact, the proof of G.K. Pedersen, shows that the y ∈ X in the factorization
e · y = x can be chosen in the closed convex cone B+ · x of X, cf. Section 11 of
Appendix A.
Since one can here replace X likewise by B · x or by c0 (X) with left action
b · (x1 , x2 , . . .) = (b · x1 , b · x2 , . . .), we can find a positive contraction a ∈ B and
y ∈ B · x (respectively (y1 , y2 , . . .) ∈ c0 (X)) with b · y = x and kx − yk < ε
(respectively with b · yn = xn and kxn − yn k < ε for n = 1, 2, . . .). The latter
implies in particular for X := B that for a, b ∈ B there are e ∈ B+ , c ∈ B+ a and
d ∈ B+ b with ec = a, ed = b and kek ≤ 1.
If B is a σ-unital C *-algebra and X a non-degenerate B-module with kb · xk ≤
kbk kxk, then one can find, for each given strictly positive contraction b0 ∈ B+ and
ε > 0 and x ∈ X, a factorization e · y = x with ky − xk < ε where y ∈ X and
e := ϕ(b0 ), ϕ an increasing continuous function with f (0) = 0 and f (1) ≤ 1.
Except the latter, all this is also true for Banach algebras B with an approxi-
mate unit consisting of contractions. (But one has to use for such algebras B the
estimates from the generalization of the Cohen factorization given by E. Hewitt or
P. Koosis.)
In particular, the following definition and arguments work more generally for
Banach algebras B that contains an approximate unit {eτ } with k eτ k ≤ 1 .
Notice that the non-degeneracy of the action of B on X implies that x ∈ B · x
and and thus, that B · x = {0} is equivalent to x = 0.
This allows to extend the action of B on X in a unique way to a left module
action of the left-multiplier algebra of B , i.e., of the operator-norm closed
subalgebra M` (B) of L(B) of bounded left multipliers L ∈ L(B) of B, that are
defined by its property

L(ab) = L(a)b for all a, b ∈ B .

(The elements L ∈ M` are also called “left centralizers” of B, where we have to

consider B as subalgebra of L(B).)
The extension of the B-modul X to an M` (B)-modul X with (Lb ) ·0 x = b · x :=
ρ(b)(x) is uniquely defined for L ∈ M` (B) and y ∈ X via Cohen factorization
y := b · x by
L ·0 (b · x) := L(b) · x for b .

The definition of L ·0 y is justified, because the right side is equal to limτ L(eτ b) · x,
because L is bounded and eτ b → b. Since L(ea) · x = L(e) · (a · x) = L(eb) · y for
a·x = b·y, it follows that the action of M` (B) is well-defined. It is then not difficult
to check that (L, x) 7→ L ·0 x is bi-linear in L and x ∈ X and kL ·0 xk ≤ kLkkxk. See
[621, thm. 5.9] and proof of [621, thm. 5.10] for the case where B is a C *-algebra.
634 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

Obviously an operator T ∈ L(B) is in M` (B) if and only T Rb = Rb T for all

b ∈ B, where Rb (a) := ab for a ∈ B. This shows that M` (B) is closed in L(B)
with respect to the strong operator topology.
If a Banach algebra B contains (two-sided) bounded approximate unit, then the
Cohen factorization theorem, applied to B as a left module over B, shows that the
subalgebra M` (B) ⊂ L(B) is the closure of the (then to B isomorphic) subalgebra
{Lb ; b ∈ B} ⊂ L(B) with respect to the strong operator topology on L(B).
We use now that for every Banach space X, the space L(X) is complete as
locally convex vector space with respect to the strong operator topology (an imme-
diate consequence of the Banach-Steinhaus uniform boundedness theorem).
It implies that M` (B) equipped with the strong operator topology is the com-
pletion of B with respect to the uniform structure defined on B × B by the system
of semi-metrics ρc (a, b) := k(b − a)ck (c ∈ B), if B has a bounded two-sided approx-
imate unit.
Let ηB : B ,→ B ∗∗ the natural isometry from B into its second conjugate Banach
space B ∗∗ . We identify the elements b ∈ B later with its image ηB (b) ∈ B ∗∗ .
If B has a (left) approximate unit consisting of contractions in B then there
exists an element e ∈ B ∗∗ with the properties kek ≤ 1 and (Rb )∗∗ (e) = ηB (b) for
all b ∈ B.
It follows that the map T ∈ L(B) 7→ T ∗∗ (e) ∈ B ∗∗ defined on L(B) has an
isometric restriction λ to M` (B) that satisfies

(Rb )∗∗ λ(T ) = ηB (T b) and kλ(T )k = kT k for b ∈ B, T ∈ M` (B) . (1.1)

−1
If X ∈ B ∗∗ satisfies (Rb )∗∗ T ∈ ηB (B) for all b ∈ B then T (b) := ηB ((Rb )∗∗ X)
defines an element T ∈ M` (B) with λ(T ) = X.
If, moreover, the linear span of the elements (Rb )∗ ρ for b ∈ B and ρ ∈ B ∗ is
norm-dense in B ∗ then the element e ∈ B ∗∗ with the property (Rb )∗∗ (e) = b for all
b ∈ B and the map λ : M` (B) → B ∗∗ with the quoted property is unique.
The latter is the case for C *-algebras B, by the Kaplansky density theorem for
B ⊆ B ∗∗ .
Notice that that (Rb )∗∗ (X) = X · ηB (b) in the W*-algebra B ∗∗ for b ∈ B and
X ∈ B ∗∗ if B is a C *-algebra B, i.e., the equation (1.1) becomes

γ(T )ηB (a) = ηB (T a) .

It implies that γ is an isometric algebra isomorphism from M` into B ∗∗ with

γ(Lb ) = ηB (b).
If X ∈ B ∗∗ satisfies X· ∈ B ⊂ B ∗∗ then there exists T ∈ M` (B) with X = γ(T )
−1
for T (b) := ηB (XηB (b)).
It defines a multiplicative isometric algebra representation of M` (B) into the
W*-algebra B ∗∗ . The image of λ is the algebra of left-multipliers of ηB (B) ⊆ B ∗∗
inside B ∗∗ .
 1. STRICT CONVERGENCE IN MULTIPLIER ALGEBRAS 635

Recall that each C *-algebras B has an approximate unit consisting of contrac-

tions and the linear span of the elements (Rb )∗ (ρ) with b ∈ B and ρ ∈ B ∗ (where
Rb (a) := ba) is norm-dense in B ∗ , cf. [616, thm. 4.1.2] and use that each ρ is in the
norm closure of (Rb )∗ · ρ (b ∈ B, kbk ≤ 1), because B is *-ultra-strong dense in sec-
ond conjugate W*-algebra B ∗∗ by Kaplansky density theorem cf. [616, thm. 2.3.3].
In particular, in this case e = 1 := 1B ∗∗ and γ(T ) := T ∗∗ (1) for T ∈ M` (B).
Thus a natural isometric linear map λ : M` (ηB ) : M` (B) ,→ B ∗∗ is given by
the restriction of the natural linear map

T ∈ L(B) 7→ T ∗∗ (e) ∈ B ∗∗

to M` (B), where T ∗∗ here is the bi-adjoint of the operator T ∈ L(B) and e is the
unique cluster point of an approximate unit consisting of contractions in B.
The restriction of the map T 7→ T ∗∗ (e) to the elements in the algebra of left
multipliers M` (B) ⊆ L(B) of the map T → T ∗∗ (1) from L(B) to B ∗∗ is isometric
and multiplicative if the Banach algebra B contains a two-sided approximate unit
consisting of contractions and the elements (Rb )∗ (ρ)(·) := ρ((·)b) for ρ ∈ B ∗ and
b ∈ B are norm-dense in the dual Banach space B ∗ of B. Here e denotes the (then)
unique element in B ∗∗ with (Rb )∗∗ (e) = b for all b ∈ B. In particular the map
γ : M` (B) → B ∗∗ is uniquely defined for all C *-algebras B.
The general (two-sided) multiplier algebra of a Banach algebras B (not nec-
essarily with involution operation on B) is defined as the set of pairs (L, R) of
operators L, R ∈ L(B) that satisfy (R(b)) · c = b · (L(c)), L(ab) = L(a)b and
R(ab) = aR(b) for a, b, c ∈ B. Notice that left and right multiplication (La , Ra ) by
a ∈ B defines a two-sided multiplier for each a ∈ B.
The property (R(b)) · c = b · (L(c)) for all b, c ∈ B of the pair (L, R) ∈
L(B) × L(B) alone implies that a(L(bc) − L(b)c) = 0 = (R(ab) − aR(b))c for
all a, b, c ∈ B. Thus implies automatically that L and R are left and right multi-
pliers if e.g. B contains a (two-sided) approximate unit. Moreover, then the pair
(L, R) is determined uniquely by L alone and the property that to this (special sort
of) left multipliers L there exists a right multiplier such that (L, R) is a two-sided
multiplier.
In a Banach *-algebra B one can transform each right multiplier R into a left
b ∈ M` (B) by the anti-linear map on L(B) (of order 2) given by R(a)
multiplier R b :=
∗ ∗
R(a ) . If a Banach *-algebra B contains an approximate unit then the two-sided
∗
multipliers are the pairs of left multipliers (L, R)
b that satisfy (R(a))
b b = a∗ L(b),
where R is uniquely determined by L (if R exists for L). We rename them and
b
describe them as pairs (S, T ) ∈ M` (B) × M` (B) and the defining property is the
relation

a∗ S(b) = T (a)∗ b for all a, b ∈ B , (1.2)

that is equivalent to S(b)a

b = bT (a) for a, b ∈ B.
636 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

If B contains an approximate unit (eτ ) with keτ k ≤ 1 then Equation (1.2)

implies that kSk = kT k. In this way we can define on the two-sided multiplies
(S, T ) a norm k(S, T )k := max{kSk, kT k} = kSk.
(0.2) Two-sided multipliers of C *-algebras: We use the equivalent def-
inition for the two-sided multiplier algebra M(B) in case of C *-algebras B that
better suites for our applications:
We define M(B) as the norm-closed sub-algebra of M` (B) ⊆ L(B) consisting
of all operators T ∈ M` (B) with the property that there exists S ∈ L(B) that
satisfies Equation (1.2).
This equations yield automatically that the operator-norm inherited from L(B)
is the same as the C *-norm of the C *-algebra M(B).
The reader should notice the we have in fact a stronger observation (going back
to an old remark of J. von Neumann):
Alone the property that S and T are abstract maps from B into B (without sup-
posing linearity as precondition) that satisfy a∗ S(b) = T (a)∗ b for all b, c ∈ B allows
to see, using the property kd∗ dk = kdk2 = kdd∗ k of C *-norms and the closed graph
theorem, that S and T must be linear operators that are bounded and have the
properties kT k = kSk, T (bc) = (T b)c, S(bc) = (Sb)c and that S is uniquely deter-
mined by T . Then S := T ∗ ∈ M` (B) is well-defined for T in this closed subalgebra
M(B) of M` (B) by T ∗ := S.
The natural *-monomorphism of B onto an essential ideal of M(B) ⊆ L(B) is
given by b ∈ B 7→ Lb ∈ L(B), where Lb (c) := bc for b, c ∈ B. We identify B with
its image in M(B) via b ↔ Lb .
The so-called strong topology and strict topology on M(B) ⊆ L(B) is given
by the restriction to M(B) of the usual strong operator topology on L(B), that is
given by the family of semi-norms T 7→ kT bk with b ∈ B, and the strict topology
is given by the family of semi-norms T 7→ kT bk and T 7→ kT ∗ bk, where T ∗ ∈ L(B)
is defined as above by (T ∗ a)∗ b = a∗ (T b) for a, b ∈ B.
Obviously this implies that in case of σ-unital B we can take a strictly positive
contraction b0 ∈ B+ and get with T 7→ kT b0 k + kT ∗ b0 k a defining norm on M(B)
that defines the strict topology on bounded parts of M(B).
Let ηB : B ,→ B ∗∗ the natural *-monomorphism from B into its second con-
jugate W*-algebra B ∗∗ . Sometimes we identify the elements b ∈ B with its image
ηB (b) ∈ B ∗∗ .
A natural and unital C *-monomorphism M(ηB ) : M(B) ,→ B ∗∗ is given by
the restriction of the natural linear map

γ : T ∈ L(B) 7→ T ∗∗ (1) ∈ B ∗∗

to M(B), where T ∗∗ here is defined as the bi-adjoint of the operator T ∈ L(B) and
1 := 1B ∗∗ is the unit element of the W*-algebra B ∗∗ .
Look at next. Is it mentioned above?
 1. STRICT CONVERGENCE IN MULTIPLIER ALGEBRAS 637

The restriction of the map T 7→ T ∗∗ (1) (with 1 = 1B ∗∗ ) to the elements in the

algebra of left multipliers M` (B) ⊆ L(B) of the map T → T ∗∗ (1) from L(B) to
B ∗∗ is isometric and multiplicative if the Banach algebra B contains a two-sided
approximate unit consisting of contractions and the elements (Rb )∗ (ρ)(·) := ρ((·)b)
for ρ ∈ B ∗ and b ∈ B are norm-dense in the dual Banach space B ∗ of B. Here
1 := 1B ∗∗ denotes the (then) unique element in B ∗∗ with (Rb )∗∗ (1) = b for all
b ∈ B.
Thus, if B is a C *-algebra, then it is a C *-algebra monomorphism if restricted
to M(B) ⊆ L(B) because, B+ 3 eτ → 1 in B ∗∗ with σ(B ∗∗ , B ∗ ) topology, e.g.

T (b) = T ∗∗ (1 · b) = lim T (eτ · b) = T ∗∗ (eτ )b = T ∗∗ (1)b

τ

for all b ∈ B ∗∗ . To see this, use that T ∗∗ is σ(B ∗∗ , B ∗ )-continuous and that the
unit ball of B is σ(B ∗∗ , B ∗ )-dense in the unit ball of B ∗∗ .
The image of T ∈ M(B) 7→ T ∗∗ (1B ∗∗ ) is a C *-algebra isomorphism from
M(B) onto the C *-subalgebra of the W*-algebra B ∗∗ of elements d ∈ B ∗∗ with
the property dB ⊆ B and Bd ⊆ B, where we have the elements b ∈ B naturally
identified with its images ηB (b) ∈ B ∗∗ in the second conjugate W*-algebra B ∗∗ .
We say that a C *-morphism h : B → M from a C *-algebra B into a W*-algebra
M that is weakly non-degenerate if 1M is contained in the σ(M, M∗ )-closure of
h(B).
If h is weakly non-degenerate then there exists a unique C *-morphism
M(h) : M(B) → M with the property M(h)|B = h. The C *-morphism M(h) is
unital and maps M(B) into the C *-subalgebra of M given by the two-sided mul-
tipliers m ∈ M of h(B) in M , i.e., mh(B) ∪ h(B)m ∈ h(B) for m ∈ M(h)(M(B)).
If h : B → M is faithful and weakly non-degenerate then M(h) is faithful and
each two-sided multiplier m of h(B) is in the image M(h)(M(B)), because then
L : b 7→ h−1 (mh(b)) and R : b 7→ h−1 (h(b)m) define a pair (L, R) ∈ L(B) × L(B)
that defines an element T ∈ M(B) with M(h)(T )h(b) = mh(b). Now use that 1M
is in the σ(M, M∗ )-closure of h(B).
The latter applies to B := K(H), M := L(H) and to h : b ∈ K(H) → b ∈ K(H)
and yields that M(K(H)) is naturally isomorphic to L(H) by an isomorphism that
fixes the compact operators.
The strict closure of any convex subset Z ⊆ M(B) coincides with the inter-
section of M(B) ⊆ B ∗∗ with its closure Z in the σ(B ∗∗ , B ∗ )-topology, because all
strictly continuous linear functionals f : M(B) → C are of the form f = g ◦ M(ηB )
with g ∈ B ∗ ⊆ B ∗∗ .
This can be seen from the existence of b1 , . . . , bn ∈ B for a given strictly contin-
Pn
uous linear functionals f ∈ M(B)∗ that satisfy | f (T ) | ≤ ∗
k=1 kT bk k+kT bk k for
all T ∈ M(B), and from the fact that the unit-ball of B is strictly dense in the unit
ball of M(B) because each C *-algebra has an approximate unit {eτ } consisting of
positive contractions:
638 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

The net eτ 0 (T eτ ) ∈ B – with possibly different τ 0 , τ - converges strictly to T .

The strict density of the unit ball of B in the unit ball of M(B) implies also
directly that a strictly continuos linear functional f on M(B) satisfies kf k = kf |Bk.
Hence, f is determined by its restriction to B, i.e., is the same as g| M(B) of the
natural extension of g := f |B ∈ B ∗ to an σ(B ∗ , B ∗∗ )-continuous extension to an
element of M(B) ⊆ B ∗∗ .
If ρ : B → L(H) is a non-degenerate *-representation of B on a Hilbert space
H, i.e., if span(ρ(B)H) is dense in H, then ρ is also a non-degenerate C *-morphism
from B into M(K(H)).
This is because the closure of the linear span span(ρ(B)K(H)) is equal to
ρ(B) · K(H) by Cohen factorization and is a closed right ideal R of K(H).
Indeed: It is easy to see that K(H)H = H and that RH = P H and P K(H) = R
with a unique projection P ∈ L(H) for all closed right ideals R of K(H).
Since M(K(H)) is naturally isomorphic to L(H), a *-representation ρ : B →
L(H) is non-degenerate if and only if ρ is non-degenerate as a C *-morphism from
B into M(K(H)).
(This says equivalently that ρ(B)H = H if and only if ρ(B)K(H) = K(H).)
If the latter is the case we get the unique unital strictly continuous extension
M(ρ) : M(B) → M(K(H)) = L(H) with the property M(ρ)(b) = ρ(b) for b ∈ B.
Next has also overlaps with (0.3):
MOVE TO to M(K) study !?! CROSS REF to????
(uniqueness extensions M(h) of non-degenerate h are explained in ????????
where??? “uniqueness of M(h) for non-degenerate h”
Should be in Chp. 3 – or in Chp.2, or even in Chp.1)
The *-representation ρ : B → L(H) ∼ = M(K(H)) is non-degenerate as C *-
morphism from B to M(K(H)) if and only if ρ is non-degenerate as a representation
of B over H, i.e., the linear span of ρ(B)K(H) dense in K(H) if and only if the
linear span of ρ(B)H is dense in H.
See also further above/below for next:
The latter holds because K(H)H = H. (Alternative argument: A closed right
ideal R of K is equal to K if and only if the linear span of RH is dense in H.)
The proof of the uniqueness is similar to the below given proof in proof-part
(1a) of the uniqueness of M(h) : A → M(B) for non-degenerate h : A → M(B).
The existence of M(ρ) comes by restriction to M(B) = M(ηB )(M(B)) ⊆ B ∗∗
of the unital and normal extension ρ : B ∗∗ → L(H) of ρ to B ∗∗ , of ρ – using that
L(H) ∼ b ∗ , with ⊗
= (H⊗H) b the maximal uniform Grothendieck tensor product, i.e.,
H⊗Hb is the ideal of nuclear operators X with the norm kXknuc , and that H⊗H b
∗ ∗
is a complemented subspace of L(H) – as adjoint ρ of ρ |H⊗H. b The proofs are
straight-forward by calculations (cf. functional analysis textbooks).
 1. STRICT CONVERGENCE IN MULTIPLIER ALGEBRAS 639

Moreover, M(ρ)(M(B)) is contained in the bi-commutant ρ(B)00 , which is the

smallest von-Neumann subalgebra of L(H) that contains ρ(B).
It is the *-strong closure of ρ(B) if ρ is non-degenerate, by using the bi-
commutation theorem of J. von Neumann.
Indeed, if U ∈ L(H) is unitary and U ρ(b) = ρ(b)U for all b ∈ B, then
(U M(ρ)(T ) − M(ρ)(T )U )ρ(b) = 0 for all b ∈ B. Since ρ(B)H is dense in H it
follows U M(ρ)(T ) = M(ρ)(T )U , i.e., M(ρ)(T ) ∈ ρ(B)00 .
If, in addition, ρ is faithful on B then M(ρ) is faithful on M(B), because
M(ρ)(T ) = 0 implies ρ(T B) = {0}, i.e., T B = {0} and T = 0.
If the non-degenerate ρ is moreover faithful, then the image M(ρ)(M(B)) of
M(ρ) is identical with the set of operators Y ∈ L(H) with Y ρ(b), Y ∗ ρ(b) ∈ ρ(B)
for all b ∈ B. Indeed: Such Y defines bounded linear maps T, S ∈ L(B) by T b :=
ρ−1 (Y ρ(b)) and Sb := ρ−1 (Y ∗ ρ(b)) that satisfy T (bc) = (T b)c and (Sb)∗ c = b∗ (T c)
for b, c ∈ B. Thus, T ∈ M(B) and M(ρ)(T )ρ(b) = ρ(T b) = Y ρ(b) for b ∈ B. Since
ρ(B)H has dense span in H it follows M(ρ)(T ) = Y .
In particular, M(B) can be identified with the C *-subalgebra of the bi-dual
W*-algebra B ∗∗ ⊇ B given by the elements T ∈ B ∗∗ with T b, bT ∈ B for all b ∈ B.
This is because B ∗∗ is isomorphic to a von-Neumann algebra on a suitable Hilbert
space H and the corresponding representation of B is *-strongly dense in B ∗∗ .
We notice that any non-degenerate *-representation ρ : B → L(H) ∼
= M(K(H))
is also a non-degenerate C *-morphism into M(K(H)) = L(H), despite of different
definitions.
The norm-closed linear span R ⊆ K(H) of ρ(B) · K(H) is a non-degenerate left
B-module via ρ. Application of Cohen factorization shows that ρ(B) · K(H) = R
is a closed right-ideal of K(H). The closed right ideals R of K(H) are all of the
form R = P · K(H) for some orthogonal projection P ∈ L(H) given by the closed
subspace R · H of H.
Since ρ(B)H = H by non-degeneracy of ρ and – obviously – K(H) · H = H we
get P = 1. This shows that ρ : B → M(K(H)) is non-degenerate, i.e., ρ(B)·K(H) =
K(H), if we apply Cohen factorization for the non-degenerate left B-module K(H) .
We obtain that M(ρ) : M(B) → M(K(H)) ∼
= L(H) is strictly continuous.
NEW:
(0.3) The strict topology on M(B):
Recall that the strict topology on the two-sided multiplier algebra M(B) ⊆
L(B)⊕L(B) is given on the pairs (L, R) ∈ M(B) by semi-norms (L, R) 7→ kL(b)k+
kR(b)k with b ∈ B. Clearly the family defines the same locally convex topology as
the family of semi-norms given by (L, R) 7→ k L(b) k + k R(c ) k with b, c ∈ B.
In case of C *-algebras B (or Banach *-algebras with bounded approximate
unit) this strict topology is given by the semi-norms T ∈ M(B) 7→ kT bk + kT ∗ bk
for b ∈ B+ .
640 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

The norm topology on B is the same as the strict topology on B ⊆ M(B), if

and only if, B is unital, because B is strictly dense in M(B), an approximate unit
{ eτ } in the positive contractions converges strictly to 1 in M(B).
If B is non-unital and σ-unital, then B contains a sequence b1 , b2 , . . . B+ with
kbn k = 1 such that bn → 0 strictly in M(B).
Older stuff below in blue:
The finite rank operators a ∈ K(H) are norm-dense in K(H). Thus, is suffices
to find, for each finite rank projection p ∈ K(H) and ε > 0, a positive contraction
√
e = e(p, ε) ∈ B with k ρ(e)xk − xk k < ε/ n for k = 1, . . . n where n := rank of p
and x1 , . . . , xn ∈ H is an orthonormal basis of pH, because then kp − ρ(e)pk < ε.
Since the positive elements e ∈ B+ with kek < 1 build an approximate unit for
B and span(ρ(B)H) is dense in H we can find such e ∈ B+ . In fact, the Cohen
factorization theorem for non-degenerate left B-modules gives an e ∈ B+ with
ρ(e)xk = xk and kek = 1.
TO DO still here
(0.4) The strict topology on M(K) ∼
= L(`2 ):
HERE WE NEED FIRST that M(K) = L(`2 ).
Has been shown with several methods above(?).
Recall that we have seen above that M(K(H)) is naturally isomorphic to L(H)
by the isomorphism from L(H) to M(K(H)) that fixes K(H).
Let H a Hilbert space (of arbitrary dimension) and K := K(H) the essential
closed ideal of L(H) of compact operators on H.
There is natural isomorphism from M` (K) onto L(H) that fixes the elements of
K, because K is an ideal of L(H) and L(H) is in a natural way the second conjugate
of K.
In particular, M` (K) = M(K) ∼= L(H) and the natural isomorphism from
M(K) onto L(H) is continuous with respect to the strict topology on M(K) and
*-ultra-strong topology on L(H).
In the same way M` (K) → L(H) is continuous with respect to the strong
operator topology on M` (K) ⊂ L(K) induced from K and the strong operator
topology on L(H) induced from H.
It turns out that the strict topology on M(K) and the strong* topology on
L(H) coincide on bounded parts. (Similar: Both strong topologies on bounded
parts of M` (K) = L(H) coincide.)
It follows that for any (non-zero) non-degenerate representation ψ : K → M(B)
(or any weakly non-degenerate ψ : K → M into a W*-algebra M ) induce via
M(ψ)−1 on bounded parts of M(K) exactly the strict topology.
In particular, on bounded parts of M(K) the strict, *-strong and *-ultrastrong
topology coincide.
 1. STRICT CONVERGENCE IN MULTIPLIER ALGEBRAS 641

Alternatively:
Since K is essential in L(H) there is a natural unital *-monomorphism (say η)
from L(H) into M(K).
Conversely the natural *-representation T ∈ K 7→ T ∈ L(H) of K on H extends
to a faithful unital C *-morphism (say λ) from M(K) into the W*-algebra K∗∗ ∼=
L(H).
The unital C *-morphisms λ ◦ η form M(K) to M(K) fixes the elements of
K ⊆ M(K). The uniqueness of an extension of the non-degenerate C *-morphism
idK to a C *-morphism from M(K) into M(K) follows from above discussed Cohen
factorization factorization theorem and implies that λ ◦ η = idM(K) . Since η and λ
are faithful it follows that η must be surjective, i.e., η is an isomorphism from the
abstract C *-algebra M(K) ⊆ L(K) onto L(H) that fixes K(H).
We compare the strict topology on M(K(H)) ∼ = L(H) with the *-ultra-strong
topology on L(H) and consider the topologies induced on L(`2 (N)) by M(ρ)−1
from any non-degenerate *-representation ρ : K(`2 (N)) → L(H):
We get
(α)
Bounded norm-closed convex parts of M(ρ)(K(`2 (N))) are strictly closed ???
in M(K(H)) ???,
and that the by M(ρ)−1 induced topology on M(K(`2 (N))) is the same as the
strict topology with respect to K(`2 (N)).
(β) Strict topology in the multiplier algebra M(K)
(respectively M(B) ⊆ B ∗∗ is finer or equal to *-ultra-strong topology, and they
coincide on bounded parts.
(But it is not unlikely that they are globally different, because its restrictions
to `∞ (N) ⊆ L(`2 (N)) ∼
= M(K(`2 (N))) coincides with the strong operator topologies
given by the multiplier actions of `∞ (N) on c0 (N) respectively on `2 (N).)
Strict topology in the multiplier algebra M(K)
(respectively M(B) ⊆ B ∗∗ is finer or equal to *-ultra-strong topology, and they
coincide on bounded parts.
Older stuff below in blue:
The finite rank operators a ∈ K(H) are norm dense in K(H). Thus, is suffices
to find, for each finite rank projection p ∈ K(H) and ε > 0, a positive contraction
√
e = e(p, ε) ∈ B with k ρ(e)xk − xk k < ε/ n for k = 1, . . . n where n := rank of p
and x1 , . . . , xn ∈ H is an orthonormal basis of pH, because then kp − ρ(e)pk < ε.
Since the positive elements e ∈ B+ with kek < 1 build an approximate unit for
B and span(ρ(B)H) is dense in H we can find such e ∈ B+ . In fact, the Cohen
factorization theorem for non-degenerate left B-modules gives an e ∈ B+ with
ρ(e)xk = xk and kek = 1.
642 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

The same argument shows that the *-ultra-strong topology on M(K(H)) =

L(H) is coarser than the strict topology on M(K(H)), because the *-ultra-strong
topology is the *-strong topology for the left multiplication L(H) on the Hilbert-
Schmidt class operators (HS(H) ∼ = H ⊗2 H):
There the same argument as for given above for B := K(H) works for the left
multiplication on (new) B := HS(H), i.e., holds for the action of L(H) = M(K)
on the Hilbert space tensor product H ⊗2 H ∼
= HS(H).
Alternatively (done above):
The Cohen factorization theorem applies to left action of K(H) on the Hilbert-
Schmidt class operators and gives that the strict topology on M(K(H)) is finer
than the *-ultra-strong topology on all subsets of M(K(H)) ∼ = L(H).
It follows that they coincide on bounded parts of M(K).
But it says nothing about the behavior on unbounded parts.

(1.) Existence and uniqueness of M(h):

(1.0) Uniqueness and strict continuity of M(h):
Let A and B C *-algebras, h : A → M(B) a C *-morphism and ψ : M(A) →
L(B) any algebra homomorphism with the property ψ|A = h. Then ψ and h have
the property ψ(X)(h(a)b) = ψ(X)(ψ(a)b) = ψ(Xa)b = h(Xa)b for X ∈ M(A),
a ∈ A and b ∈ B.
Thus, ψ is uniquely determined by its restriction h := ψ|A on the closed linear
span L ⊆ B of h(A)B and ψ(1)b = b for b ∈ L. In particular ψ is unital and is
uniquely determined if the linear span of h(A)B is dense in B.
If span(h(A)B) is dense in B, then ψ is also a C *-morphism from M(A) into
M(B) ⊆ L(B) := L(B, B), because ψ(X)(h(a)b) = h(Xa)b implies

(ψ(X)(h(a1 )b1 ))∗ (h(a2 )b2 ) = b∗1 h(a∗1 X ∗ a2 )b2 = (h(a1 )b1 )∗ (ψ(X ∗ )(h(a2 )b2 )) ,

i.e., (ψ(X)b1 )∗ b2 = b∗1 (ψ(X ∗ )b2 ) if we pass to the closure of span(h(A)B). We get
that ψ(X) is in M(B) ⊆ L(B) and ψ(X ∗ ) = ψ(X)∗ if the linear span of h(A)B is
dense in B.
Such extensions ψ of h are automatically strictly continuous if h(A)B is dense
in B zz:
Then the B is also a non-degenerate left Banach module over A with left-action
a · b := h(a)b for a ∈ A and b ∈ B, and the Cohen factorization theorem applies to
it:
For each b1 , b2 ∈ B there exist contractions a1 , a2 ∈ A and c1 , c2 ∈ B with
kcj − bj k < 1/2 such that bj = h(aj )cj . It follows that the semi-norm

X 7→ kψ(X)b1 k + kψ(X ∗ )b2 k

on M(A) is less or equal to the semi-norm by

X 7→ (1 + kb1 k + kb2 k)(kXa1 k + kX ∗ a2 k) .

 1. STRICT CONVERGENCE IN MULTIPLIER ALGEBRAS 643

This latter semi-norms are continuous with respect to the strict topology on M(A).
It implies that ψ is strictly continuous.
If conversely a unital and strictly continuous ψ : M(A) → M(B) with ψ(a) =
h(a) exist, then an approximate unit (eτ ) ⊂ A+ of A converges strictly to 1 ∈
M(A), and h(eτ )b = ψ(eτ )b converges to b = ψ(1)b for each b ∈ B. Thus h(A)B is
dense in B i.e., h is non-degenerate.
(1.1) Existence of M(h):
Suppose that h : A → M(B) is a non-degenerate C *-morphism, i.e., the linear
span of h(A)B is dense in B.
The Cohen factorization theorem applies to the Banach left A-module B with
product a · b := h(a)b.
Then B is also in a unique way a Banach module over M` (A) with a product
L · b ∈ B defined by L ·0 b := L(a) · c = h(L(a))c for b = h(a)c and L ∈ M` (A).
0

It satisfies L(a) · b = L ·0 (h(a)b), kL ·0 bk ≤ kLk · kbk. This defines a contractive

Banach algebra homomorphism from M` (A) into L(B).
But in this special case holds the additional property L ·0 (bc) = (L ·0 b)c for
a ∈ A, b, c ∈ B and L ∈ M` (A). Indeed, let b = h(a)d, then bc = h(a)dc and
L ·0 bc = L(a) · dc = h(L(a))dc = (h(L(a))d)c == (L ·0 b)c.
Hence M` (L)(b) := L ·0 b defines a contractive Banach algebra homomorphism
M` (h) from M` (A) into M` (B) ⊆ L(B) with M` (La )(b) = h(a)b for a ∈ A and
b ∈ B.
Since – in our picture – M(A) ⊆ M` (A) ⊆ L(A) and M(B) ⊆ M` (B), we
can define M(h) : M(A) → M` (B) as the restriction of M` (h) to M(A), i.e., by
M(h)(X)b := X ·0 b for X ∈ M(A) and b ∈ B. This becomes automatically a
C *-morphism from M(A) into M(B) as we have seen in proof-part (1.0).
Older version for the existence proof:
The following is a special case of the above considered weakly non-degenerate
C *-morphism from B into a W*-algebra.
Let ρ : B → L(H) a faithful and non-degenerate *-representation on a Hilbert
space H. The representation ρ extends to a faithful unital *-representation
M(ρ) : M(B) → L(H) with M(ρ)(b) = ρ(b) for b ∈ B.
Let d : A → L(H) the *-representation defined by d(a) = M(ρ)(h(a)). In
particular, d(a)ρ(b) = ρ(h(a)b).
Since ρ and h are non-degenerate, the linear spans of h(A)B and ρ(B)H are
dense in B respectively in H. Thus, the linear span of d(A)ρ(B)H = ρ(h(A)B)H is
dense in H, and d : A → L(H) is non-degenerate. We know from Proof-part (1.1)
that a non-degenerate *-representation d : A → L(H) extend uniquely to a non-
degenerate *-representation M(d) : M(A) → L(H) with M(d)(X)d(a) = d(Xa).
644 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

We get

M(d)(X)ρ(h(a)b) = M(d)(X)d(a)ρ(b) = d(Xa)ρ(b) = ρ(h(Xa)b)

for all X ∈ M(A), a ∈ A and b ∈ B. Since the linear span of h(A)B is dense in B,
it follows that M(d)(X)ρ(b) ∈ ρ(B) and M(d)(X)∗ ρ(b) = M(d)(X ∗ )ρ(b) ∈ ρ(B)
for all b ∈ B. Thus, M(d)(M(A)) ⊆ M(ρ)(M(B)) by the characterization of the
image M(ρ)(M(B)) of M(B) in L(H).
We let M(h) := M(ρ)−1 (M(d)(X)). This is a unital C *-morphism from M(A)
into M(B) that satisfies M(h)(X)h(a)b = h(Xa)b and is the desired extension to
M(A), because

M(ρ)(M(h)(X)h(a)b) = M(d)(X)ρ(h(a)b) = d(Xa)ρ(b) = M(ρ)(h(Xa)b) .

(2) Proofs for Part(2): Let pj,k denote the matrix units of K. Then the series
P P
c := n xn ⊗p1,n and d := n yn ⊗p1,n are unconditional strictly convergent series
in M(B ⊗ K) ( 5 ).
P
Indeed, partial sums, i.e., c(S) := n∈S xn ⊗ p1,n over a finite set S ⊂ N of
indices, are norm-bounded by kXk1/2 respectively by kY k1/2 .
More precisely: Let pm := p1,1 +· · ·+pm,m , where pj,k ∈ K are partial isometries
that define canonical matrix units K for the standard basis of `2 . Then
X
gXh∗ ⊗ p1,1 − (g ⊗ 1)c(S)((h ⊗ 1)c(S))∗ = g(X − xn x∗n )h∗ ⊗ p1,1 ,
n∈S

∗
P P
for all g, h ∈ M(B), ≤ X, c(S)(e ⊗ pm ) = n∈S,n≤m (xn e) ⊗ p1,n and
n∈S xn xn
P
(e ⊗ pm )c(S) = n∈S,n≤m (exn ) ⊗ p1,n for each e ∈ M(B).
Since the partial sums c(S) are bounded in M(B ⊗ K) by kXk1/2 , and since
the semi-norms

ρm (s, t) := max(k(s − t)(e ⊗ pm )k, k(e ⊗ pm )(s − t)k)

for m ∈ N and positive contractions e ∈ B+ define the strict topology on bounded

P
parts of M(B ⊗ K), it follows that n xn ⊗ p1,n converges strictly in M(B ⊗ K).
That the convergence is also unconditional can be seen as follows:
For each contraction e ∈ B+ , m ∈ N and ε > 0 there exists n := n(e, m, ε) ∈ N
such that n ≥ m, c(S)(e ⊗ pm ) = c(S 0 )(e ⊗ pm ) and k(e ⊗ pm )(c(S) − c(S 0 ))k < ε for
all finite subsets S, S 0 ⊂ N with {1, . . . , n} ⊆ S 0 ∩S. The latter property implies the
P
unconditional strict convergence of c := n xn ⊗ p1,n in M(B ⊗ K), i.e., that for
P
each permutation γ of the positive integers N the series n xγ(n) ⊗p1,γ(n) converges
P
strictly to the same element of M(B ⊗ K) as c := n xn ⊗ p1,n does.
Thus, the series c and d are unconditional strictly convergent and its sums are
elements of M(B ⊗ K). We use again c and d as notation for the sums of the series,

5The convergence is usually not absolute !

 1. STRICT CONVERGENCE IN MULTIPLIER ALGEBRAS 645

because in this particular case the entries of the series can be restored from the
sum, e.g.
xn ⊗ p1,n = c(1 ⊗ pn,n ) .

The c and d are row matrices if one considers M(B ⊗ K) naturally as a subset
of the set of all infinite matrices [Ti,j ] ∈ M∞ (M(B)) with entries Ti,j from M(B):
If we use a fixed non-degenerate inclusion of c0 (B) = B ⊗ c0 into B ⊗ K then
we get “natural” unital C *-monomorphisms

M(B) ⊗ 1 ⊂ `∞ (M(B)) ∼
= M(c0 (B)) ⊂ M(B ⊗ K) ⊂ M∞ (M(B)) ,

where the first two strict inclusion are strictly continuous. This inclusions are
naturally defined: If we choose a system of matrix units pj,k for K, we get nat-
ural inclusions c0 ⊂ K, c0 (B) = B ⊗ c0 ⊂ B ⊗ K and can use the isomorphism
`∞ (M(B)) ∼ = M(c0 (B)). The map b ∈ M(B ⊗ K) 7→ dbc∗ is obviously a strictly
continuous completely bounded map with cb-norm equal to kck·kdk ≤ (kSkkT k)1/2 .
The in Remark (2) defined map Γ satisfies Γ(b) ⊗ p1,1 = dbc∗ for b ∈ `∞ (M(B)) ⊂
M(B ⊗ K) .
The proposed unconditional strict convergence of the series Γ(b1 , b2 , . . .) follows
P
now from the unconditional strict convergence of the sum n (δn,k bk ) to (b1 , b2 , . . .)
in M(c0 (B)) ∼
= `∞ (M(B)).
Since d∗ bc∗ ∈ B ⊗ p1,1 for d ∈ B ⊗ K, we get that Γ(b1 , b2 , . . .) ∈ B for
(b1 , b2 , . . .) ∈ c0 (B).
(3): Let ϕ0 (t) := t for t ∈ [0, 1] and

ϕk (t) := min(1, max(2k t − 1, 0)) = (2k t − 1)+ − (2k t − 2)+

for k = 1, 2, . . .. Then there exists 1 =: k0 < k1 < k2 < · · · and fn (t) in the convex
span of ϕkn , . . . , ϕkn+1 −1 such that the functions fn have the properties quoted in
Part (3). Here we use the method of Akemann and Pedersen as described in proof of
[43, thm. 2] or in [616, thm. 3.12.14, cor. 3.12.15]. See more details and discussion
in Section 3. The proposed estimates for norms of [(fn+1 (e) − fn (e))1/2 , x] follow
also from the estimates in Lemma 5.3.2.
(4): Let T (a) := a − Γ(a, a, . . .) the linear map from A into M(B) with norm
≤ 2, where Γ is defined as in Part (2) with xn := yn := gn := (fn (e) − fn−1 (e))1/2
1/2
for n > 1 and x1 := y1 := g1 := f1 (e) in Remark (2) with fn selected for
Xn ⊂ A ⊆ M(B) as in Remark (3).
xn := yn := gn := (fn (e) − fn−1 (e))1/2 .
Check Notations and estimates !?? ??
If a∗ = a ∈ Xn with kak ≤ 1 then T (a) = n (agn − gn a)gn and kagn − gn ak <
P

4−n by part (3). Since agn ∈ B, it follows T (a) ∈ B for a in a dense linear subspace
of A, and that T (A) ⊆ B by continuity of T .
646 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

(5): Let g1 , g2 , . . . as in (4) and d1 , d2 , . . . ∈ M(B) as in (5), then gn hn = gn =

hn gn for h1 := f2 (e)1/2 and hn := (fn+1 (e) − fn−2 (e))1/2 for n > 1 where f0 := 0,
X X X
dn gn = hn (dn gn )hn + (1 − hn )dn gn ,
P 2
k(1 − hn )dn gn k < kdn gn − gn dn k , and hn converges strictly to 2 in M(B).
Since gp gq = 0 if |p − q| > 1, it follows SP2 = n∈P gn2 if |p − q| > 1 for all
P

p 6= q ∈ P .
It holds Sa − aS ∈ B for the contractions a∗ = a ∈ Xn , because kgn a − agn k <
−n
4 by part (3). Thus, the continuous derivation a 7→ Sa − aS from A into M(B)
maps A into B, i.e., πB (S) ∈ πB (A)0 ∩ (M(B)/B).
It follow that πB (SaS + (1 − S 2 )1/2 a(1 − S 2 )1/2 ) = πB (a) for all a ∈ A.
P P P
(6): Since kdn gn − gn dn k ≤ n νn < ∞, the series n dn gn converges
P
(unconditional) strictly in M(B) by part (5). Let d := n dn gn denote its sum.
∗
P
Denote by Zk,j the sum of the series n g2n−j (d2n−j ad2n−j+k )g2n−j+k for
j = 0, 1 and k ∈ N, that is strictly convergent in M(B) by part (2), because
kd∗2n−j ad2n−j+k k ≤ kak and the sums n g2n−j+k
P 2 P 2
and n g2n−j are strictly con-
vergent in M(B) by part (5).
2
+ gn2 + gn+1
2 2
)1/2 (g0 := 0). The series n h2n converges
P
Let hn := (gn−1 + gn+2
strictly to some positive element in R+ − B+ . By condition (iii) and part (3),
X
Z1,0 + Z1,1 = hn (gn d∗n adn+1 gn+1 )hn ∈ B ,
n

because gn ∈ B and the sequence ( gn d∗n adn+1 gn+1 )n is in c0 (M(B)) by condition

(iii). It follows that S1 := (Z1,0 + Z1,1 )∗ + (Z1,0 + Z1,1 ) ∈ B .
If k > 1 then the elements zn := g2n−j (d∗2n−j ad2n−j+k )g2n−j+k satisfy

kzn k ≤ kag2n+k−j − g2n+k−j ak + 2 max kdk g2n+k−j − g2n+k−j dk k

1≤k≤2n−j+k

for k > 1 and j ∈ {0, 1}, by conditions (i) and (ii). Moreover, zn∗ zm = 0 = zm zn∗ for
P
n 6= m. Thus n zn converges absolute to Zk,j for k > 1, j = 0, 1, and its norm
can be estimated by
kZk,j k ≤ µ2+k−j + 2ν2+k−j .
P P
Since ( ` µ` ) + 2( ν` ) < ∞, it follows that
`
X X X
kZk,j k ≤ µ2+k−j + 2 ν2+k−j < ∞
k k k

+ Zk,1 )∗ + (Zk,0 + Zk,1 )

P
for j = 0, 1. It implies that the series S2 := k>1 (Zk,0
with entries in B converges to an element of B.
It is not difficult to see from the partial sums that the series corresponding to
d ad − Γ(d∗1 ad1 , d∗2 ad2 , . . .) converges strictly to the sum S1 + S2 ∈ B.)
∗

(7): The isomorphism D + Ann(D) ∼

= D ⊕ Ann(D) implies

M(D + Ann(D)) ∼
= M(D) ⊕ M(Ann(D))
 1. STRICT CONVERGENCE IN MULTIPLIER ALGEBRAS 647

and that the natural *-monomorphism ι : D + Ann(D) → B has a unital strictly

continuous extension M(ι) : M(D + Ann(D)) → M(B) if ι is non-degenerate.
Define p := M(ι)(1M(D) ). Then pBp = D. Conversely, if D = pBp then
(1 − p)B(1 − p) = Ann(D) and pep + (1 − p)e(1 − p) becomes an approximate unit
of B if e runs through an approximately central (for p) approximate unit of B.
(8): final part of TEXT in (8):
Recall that B is a stable C *-algebra if there exists an isomorphism ψ from
B ⊗ K onto B for the compact operators K = K(`2 ) on `2 (N).
This is equivalent to the existence of a C *-algebra C and a C *-algebra isomor-
phism from C ⊗ K onto B.
The following criterium is necessary and sufficient for the stability of a (not
necessarily σ-unital) C *-algebra B:
There exists a sequence s1 , s2 , . . . of isometries in M(B) such that n sn s∗n con-
P

verges strictly to 1 in M(B).

It holds u∗ δ∞ (b)u = ∗
P
n tn btn by a unitary u ∈ M(B) with tn = usn for
n = 1, 2, . . . if t1 , t2 , . . . is any other (countable) sequence of isometries in M(B)
with n tn t∗n = 1.
P

In this sense the unitary equivalence class [δ∞ ] has “representatives”

δ∞ : M(B) → M(B) determined by s1 , s2 , . . .. We call this class (and some-
P
times its members n sn (·)sn ) the infinite repeat (on M(B)) of the identity
mapping idM(B) of M(B).
Then δ∞ (b) (or its unitary equivalence class [δ∞ (b)]) is the infinite repeat of
the element b ∈ M(B).
Moreover, there exists C *-morphism µ : K → M(B) with the properties

(a) µ(K) ⊂ δ∞ (B)0 ∩ M(B),

(b) δ∞ (b)µ(k) ∈ B for b ∈ B and k ∈ K, and
(c) the (unique) C *-morphism ϕ : B ⊗ K → B with ϕ(b ⊗ k) = δ∞ (b)µ(k) is
an isomorphism from B ⊗ K onto B.

The C *-morphism µ : K → M(B) with properties (a,b,c) is uniquely defined

sn s∗n = 1 – up to
P
by the given sequence s1 , s2 , . . . ∈ M(B) of isometries with
unitary equivalence by unitaries in δ∞ (B)0 ∩ M(B).
One of the corresponding C *-morphisms µ : K → δ∞ (B)0 ∩M(B) can be defined
by µ(pj,k ) := sj s∗k for j, k ∈ N, if the unitary equivalence class [δ∞ ] is realized by
∗
P
given isometries s1 , s2 , . . ., i.e., δ∞ (·) := n sn (·)sn .

The isomorphism ϕ from B ⊗ K onto B defined by µ : K → M(B) – with

properties (a,b,c) – has the property that the corresponding C *-morphism b 7→
ϕ(b⊗e11 ) is a *-endomorphism of B, that is approximately 1-step inner if considered
as a c.p. contraction.
648 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

The C *-morphism ϕ((·) ⊗ p1,1 ) = s1 (·)s∗1 is unitarily homotopic to idB if B is

σ-unital:
???????????????
Moreover, if b ∈ B+ , then ϕ(b ⊗ p1,1 ) is Murray–von-Neumann equivalent to b.
In particular,
ϕ(J ⊗ K) = J ∀ J / B.

This property of the “natural” ϕ is different from the properties of very random
and possibly not well-behaved *-isomorphism ψ from B ⊗ K onto B – as allowed in
the definition of stability –, because in general the isomorphism ϕ ◦ ψ −1 of B does
not fix the ideal-system I(B) of B and is not approximately unitarily equivalent to
idB by unitaries in M(B).
The non-degenerate *-monomorphism µ : K → M(B) has the important “dual”
property that
µ(K)0 ∩ M(B) = δ∞ (M(B)) .
It shows that δ∞ (M(B)) is closed in M(B) with respect to the strict topology of
M(B).
The strictly continuous extension M(µ) : M(K) ∼ = L(`2 ) → M(B) of µ is a
unital *-monomorphism, has image in δ∞ (M(B))0 ∩ M(B) and the norm-closed
unit-ball of M(µ)(L(`2 )) is also closed in the strict topology of M(B).
The strictly closed C *-subalgebra δ∞ (M(B))0 ∩ M(B) of M(B) can be larger
than M(µ)(L(`2 )), e.g. if the the center of M(B) is not trivial, because the center of
M(B) is identical with the commutant of δ∞ (B) · µ(K) in M(B) , and δ∞ (B) · µ(K)
generates B if the center of M(B) is not trivial.
For each non-zero C *-morphism ρ : K(`2 ) → M(B), the topology induced by
M(ρ)−1 on M(K) = L(`2 ) from the strict topology on M(B) coincides on bounded
parts with the *-strong operator topology on L(`2 ). Moreover, M(ρ)(K) is strictly
closed if ρ is non-degenerate.
HERE: BEGIN NEW PROOF of (8) 27.10.2015 19:30
If B is stable then there exists a sequence of isometries s1 , s2 , . . . in M(B) such
sn s∗n converges strictly to 1 in M(B):
P
that
Is ϕ good notation?
If B is stable then there exist (by definition of stability) a C *-algebra C and a
C *-algebra isomorphism ϕ : C ⊗ K → B from C ⊗ K onto B, where K := K(`2 (N)).
If C is unital, then ρ(k) := ϕ(1 ⊗ k) defines a non-degenerate C *-morphism
from K into B.
If C is not unital, then the strictly continuous extension M(ϕ) : M(C ⊗ K) →
M(B) is a strictly continuous isomorphism from M(C ⊗ K) onto M(B).
The algebra C ⊗ K is an essential ideal of M(C) ⊗ K. Thus, there is a unique
*-monomorphism γ from M(C) ⊗ K into M(C ⊗ K) with γ(T ⊗ k1 )(c ⊗ k2 ) =
 1. STRICT CONVERGENCE IN MULTIPLIER ALGEBRAS 649

(T c) ⊗ (k1 k2 ) for T ∈ M(C), c ∈ C and k1 , k2 ∈ K. Then γ|(1 ⊗ K) is non-

degenerate. It implies that

ρ(k) := M(ϕ)(γ((1 ⊗ k)))

is a non-degenerate C *-morphism from K into M(B). Recall that here “non-

degenerate” means that ρ(K)B = B (by using e.g. Cohen factorization).
The *-monomorphism ρ extends uniquely to a faithful strictly continuous unital
C *-morphism M(ρ) from M(K) into M(B).
Any bijective map η from N × N onto N defines isometries tn ∈ L(`2 ) ∼
= M(K)
∗
P
with n tn tn converging *-strongly to 1 if we let tn (ek ) := eη(n,k) .
Since, as discussed above, on bounded parts of M(K) ∼ = L(`2 ) the *-strong,
*-ultra-strong and strict topology coincide, and since µ is strictly continuous, we
get that the isometries sn := M(γ)(tn ) define a sequence of isometries in M(B)
with the property that n sn s∗n converges strictly to 1 in M(B).
P

Suppose now that B is a C *-algebra such that its multiplier algebra M(B) con-
tains a sequence s1 , s2 , . . . ∈ M(B). We define a rather canonical C *-isomorphism
ϕ from B ⊗ K onto B :
The isometries s1 , s2 , . . . define (the unitary equivalence class of) the infinite
repeat
X
δ∞ (b) := sn bs∗n
n

for b ∈ M(B), because it follows from Remark (2) – with xn := yn := s∗n and bn := b
– that δ∞ (b) := n sn bs∗n is strictly unconditional convergent. The unconditional
P

convergence allows to check that δ∞ is a unital *-endomorphism of M(B).

It holds u∗ δ∞ (b)u = n tn bt∗n by a unitary u ∈ M(B) with tn = usn for n =
P

1, 2, . . . if t1 , t2 , . . . is any other sequence of isometries in M(B) with n tn t∗n = 1

P

(strictly convergent), because Part (2) shows that the sums n tn s∗n and n sn t∗n
P P

are unconditionally strictly convergent to contractions u, v ∈ M(B). Then the

unconditionality of the strict convergence of all four series allows to verify that
vu = uv = 1 by calculation.
The Remark (2) also implies that δ∞ is strictly continuous and is determined up
to unitary equivalence if it is build from different choices of sequence of isometries
s1 , s2 , . . . ∈ M(B) with n sn s∗n = 1 (strictly), compare Lemma 5.1.2(i,ii) and its
P

proof.
There is a natural C *-morphism µ : K → M(B) that is uniquely defined on
the elementary matrices pj,k by µ(pj,k ) := sj s∗k , as calculation shows.
Notice that µ can be different from the formerly defined ρ : K → M(B), e.g. if
B = C ⊗ K with unital C.
We show that above defined µ : K → M(B) has the in Part (8) listed properties
(a), (b) and (c):
650 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

Ad(a): µ(K) ⊂ δ∞ (B)0 ∩ M(B) because for b ∈ B and j, k ∈ N

X X
sj s∗k ( sn bs∗n ) = sj bs∗k = ( sn bs∗n )sj s∗k .
n n

Ad(b): δ∞ (b)µ(k) ∈ B for b ∈ B and k ∈ K, because the calculation in Part(a)

implies that (b ⊗ k) 7→ δ∞ (b)µ(k) extends to a C *-morphism ϕ : B ⊗ K → M(B)
with ϕ(b ⊗ pj,k ) ∈ B. Since the linear span of matrix units pj,k is dense in K it
follows that ϕ(B ⊗ K) ⊆ B.
Ad(c): The C *-morphism ϕ : B ⊗ K → B is with ϕ(b ⊗ k) = δ∞ (b)µ(k) is an
isomorphism from B ⊗ K onto B.
Indeed: By Lemma 2.2.3 the kernel of ϕ is = {0}, because from ϕ(b ⊗ k) = 0
it would follow that sj bs∗k = δ∞ (b)sj s∗k = 0 for a matrix unit pj,k ∈ K with
pj,j kpk,k = ξpj,k 6= 0.
The sequence s1 , s2 , . . . defines a partial isometry T ∈ M(B ⊗ K) by T :=
∗ ∗ ∗
P
n n ⊗ p1,n with T T = 1 and T T = 1 ⊗ p1,1 by sn sm ⊗ p1,n pm,1 = δm,n 1 ⊗ p1,1 .
s
T (b ⊗ pj,k )T ∗ = T n bs∗n ⊗ pj,k pn,1 = T ((bs∗k ) ⊗ pj,1 ) =?????
P

∗ ∗ ∗
P
n sn bsk ⊗ p1,n pj,1 = sj bsk ⊗ p11 = (δ∞ (b) · sj sk ) ⊗ p1,1 . This means T (b ⊗
∗ ∗
pj,k )T = ϕ(b ⊗ pj,k ) ⊗ p1,1 . Thus, T cT = ϕ(c) ⊗ p1,1 for all c ∈ B ⊗ K. Moreover,
T ∗ T = 1 and T (B ⊗ K)T ∗ = B ⊗ p1,1 .
Until HERE new 2015 Oct 30
??? by ??? the is a ???? natural C *-morphism µ : K → M(B) with the in Part
(8) quoted properties (a,b,c) uniquely defined by the given sequence s1 , s2 , . . . ∈
sn s∗n = 1 – up to unitary equivalence by unitaries in
P
M(B) of isometries with
δ∞ (B)0 ∩ M(B).
The below given arguments do not systematically
prove the statements in (8). Write new proofs!
??
(only for parts of 8?);
sn s∗n = 1 (strictly conver-
P
Suppose s1 , s2 , . . . ∈ M(B) are isometries with
gent) are given.
FROM “STABLE” TO s1 , s2 , . . .
∗
P
If, conversely, isometries s1 , s2 , . . . ∈ M(B) are given with n sn sn is (un-
conditional) strictly convergent to 1M(B) then we can define a strictly continuous
unital positive map δ∞ : M(B) → M(B). The strict continuity of δ∞ and the (easy
to check) multiplicative property of δ∞ |B imply that δ∞ is a strictly continuous
*-endomorphism of M(B).
??????
It turns out that there exists C *-morphism µ : K → M(B) with the properties
µ(K) ⊂ δ∞ (B)0 ∩ M(B) such that δ∞ (b)µ(k) ∈ B for b ∈ B and k ∈ K and such
 1. STRICT CONVERGENCE IN MULTIPLIER ALGEBRAS 651

that the (unique) C *-morphism ϕ : B ⊗ K → B with ϕ(b ⊗ k) = δ∞ (b)µ(k) is a

*-isomorphism from B ⊗ K onto B.
Indicate proof of non/uniqueness:
The C *-morphism µ with the described properties is determined up to unitary
equivalence by unitaries in δ∞ (B)0 ∩ M(B).
The isomorphism ϕ from B ⊗ K onto B defined by µ : K → M(B) with prop-
erties (a,b,c) has the property that the corresponding C *-morphism b 7→ ϕ(b ⊗ e11 )
is a *-endomorphism of B, that is approximately 1-step inner if considered as a
c.p. contraction. Moreover, if b ∈ B+ , then ϕ(b ⊗ e11 ) is Murray–von-Neumann
equivalent to b. In particular,

ϕ(J ⊗ K) = J ∀ J / B.

This property of the “natural” ϕ is different from the properties of very random
and possibly not well-behaved *-isomorphisms ψ from B ⊗K onto B – as allowed in
the definition of stability –, because in general the isomorphism ϕ ◦ ψ −1 of B does
not fix the ideal-system I(B) of Band is not approximately unitarily equivalent to
idB by unitaries in M(B).
The non-degenerate *-monomorphism µ : K → M(B) has the important prop-
erty that
µ(K)0 ∩ M(B) = δ∞ (M(B)) .
In particular:
δ∞ (M(B)) is closed in M(B) with respect to the strict topology of M(B).
The strictly continuous extension M(µ) : M(K) ∼ = L(`2 ) → M(B) of µ is a
unital *-monomorphism, has image in δ∞ (M(B))0 ∩ M(B) and the norm-closed
unit-ball of M(µ)(L(`2 )) is also closed in the strict topology of M(B).
The strictly closed C *-subalgebra δ∞ (M(B))0 ∩ M(B) of M(B) can be larger
than M(µ)(L(`2 )), e.g. if the center of M(B) is not trivial, because the center of
M(B) is identical with the commutant of δ∞ (B) · µ(K) in M(B) , and δ∞ (B) · µ(K)
generates B.
Next should be partly in the proof
Since µ is non-degenerate and B 6= {0}, µ is faithful and extends to a faithful
and strictly continuous C *-morphism M(µ) : M(K) → M(B) by the considerations
(0) given in the proof.
Let d : B → L(H) any non-degenerate *-representation on a Hilbert space H
(i.e., d(B)H = H, using factorization).
compare ?????????
∼ L(H) is non-degenerate in the sense of Part (1),
Then d : B → M(K(H)) =
and M(d) : M(B) → M(K(H)) is strictly continuous.
Thus M(d) ◦ M(µ) is strictly continuous on M(K(`2 )) and the topology in-
duced on M(K) by M(µ)−1 from the strict topology of M(B) is between the strict
652 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

topology of M(K) and the topology on M(K) that is induced by (M(d) ◦ M(µ))−1
on M(K) from the strict topology of M(K(H)).
The strict topology on M(K(H)) is on bounded parts the same as the *-strong
topology.
Every non-degenerate representation of K(H) on some other Hilbert space H2
induces on M(K(H)) = L(H) the same strict topology as its “native” coming from
the identical representation over H.
The topology induced by M(µ)−1 on M(K(`2 )) = L(`2 (N)) from the strict
topology on M(B) is the same for each non-degenerate C *-morphism µ : K(`2 ) →
M(B).
This “universal” strict topology on M(K) = L(`2 ) coincides on bounded parts
of L(`2 ) with the *-strong operator topology on L(`2 ). In particular, M(µ)(M(K))
is strictly closed in M(B) if µ(K) is non-degenerate.
str
The strict closure h(A) ⊆ M(B) of the image h(A) ⊆ M(B) of any non-
degenerate C *-morphism h : A → M(B) of a stable C *-algebra A contains the
image M(h)(M(A)) of the strictly continuous unital extension M(h) : M(A) →
M(B), just by strict continuity of M(h), but in general M(h)(M(A)) is not strictly
closed for non-degenerate h.
The closure of M(h)(M(A)) in M(B) with respect to the strict topology of
M(B) coincides with the intersection of M(B) ⊆ B ∗∗ with the σ(B ∗∗ , B ∗ ) closure
of M(h)(M(A)) in B ∗∗ .
(a) There exists a sequence s1 , s2 , . . . of isometries in M(B) such that n sn s∗n
P

converges strictly to 1 in M(B), if and only if, B is stable:

P ∗
There exists a sequence of isometries t1 , t2 , . . . ∈ M(K) = L(`2 ) with tn tn
strictly convergent to 1. The isometries tn can be given explicitly in L(`2 ) as
tn := I ◦ Jn by a fixed isometry I from `2 ⊗ `2 onto `2 composed with the isometries
Jn (x) := en ⊗ x for x ∈ `2 (N) and for the canonical basis {e1 , e2 , . . .} of `2 (N).
P ∗
The strict convergence tn tn = 1 in M(K) follows from the fact that the
*-strong topology on L(`2 ) ∼= M(K) coincides with the strict topology on bounded
parts of M(K) ∼ = L(`2 ).
The here used coincidence of the *-strong topology and the strict topology on
bounded parts of M(K) ∼ = L(`2 ) follows from the norm-density of the finite rank
operators in K.
Since the *-ultra-strong topology coincides with the *-strong topology on
bounded parts of L(`2 ), we get that strict (given by L(`2 ) ∼
= M(K(`2 )) and
*-ultra-strong topology coincide on bounded parts.
Do the strict topology and *-ultra-strong topology
coincide also on unbounded parts of L(H)?
It is – perhaps – the case if and only if, strict and *-ultra-strong topology
coincide on M(K(`2 (N))). The latter seems not to be the case (?!).
 1. STRICT CONVERGENCE IN MULTIPLIER ALGEBRAS 653

Are the two the strong operator topologies on `∞ (C) = M(c0 C) for its natural
actions on c0 or `2 different on unbounded parts?
Does there exist an unbounded net in `∞ that that converges in the strong
operator topology on L(`2 ) to zero but does not converge in the strong operator
topology in M(c0 ) = `∞ ⊂ L(c0 )?
It is the same problem with the “global” difference of the strong operator
topology for the left multiplication of L(`2 ) on K, Hilbert-Schmidt class, ore trace
class.
Semi-norms for *-ultra-strong topology on L(H) :
T 7→ ( n kT ∗ xn k2 + kT xn k2 )1/2 where xn ∈ `2 , n kxn k2 < ∞. It is the same
P P

as tr((T ∗ T + T T ∗ )S)1/2 = (kT S 1/2 k22 + kT ∗ S 1/2 k22 )1/2 for some positive trace class
operator S. Then R := S 1/2 is Hilbert-Schmidt, and

max(kT Rk2 , kT ∗ Rk2 ) ≤ (kT Rk22 + kT ∗ Rk22 )1/2 ≤ kT Rk2 + kT ∗ Rk2

Notice here that kXk = kXk∞ ≤ kXkp for all p ≥ 1 and X ∈ L(`2 ) with (kXkp )p =
tr((X ∗ X)p/2 ) < ∞ .
Alternatively:
The Hilbert-Schmidt class operator R can be Cohen factorized into aR1 = R
with a ∈ K and R1 in Hilbert-Schmidt class.
Since kT Rk2 = kT aR1 k2 ≤ kT ak · kR1 k2 for T ∈ L(`2 ) = M(K) it follows that
*-ultra-strong topology is coarser than strict topology or is equal to it:
Result:
If an (possibly unbounded) net converges to an operator T in strict topology, then
it converges also in *-ultra-strong topology.
Recall that strict topology on M(K) = L(`2 ) is defined by the semi-norms
T 7→ kT ak + kT ∗ bk where a, b ∈ K.
If we use Cohen factorization, then the strict topology on L(`2 ) is also defined
by the semi-norms T 7→ kT ck + kT ∗ ck with c ∈ c0 (N)+ ⊂ K(`2 (N)).
Restricted to T ∈ `∞ (N) it is the strong operator norm on `∞ (N) defined by
the multiplication c 7→ T c of elements c ∈ c0 (N) with T .
If one wants to compare this with the right multiplication action of `∞ on the
Hilbert-Schmidt class HS ∼= `2 (N × N), then one has to use the restriction of the
*-ultra-strong topology on L(`2 ) to `∞ ⊂ L(`2 ).
The (left-)actions of M(K) on K and HS are isometric with respect to the
operator norm on M(K) = L(`2 ).
The induced strong topologies have same (countable) zero sequences:
If X is a Banach space and T1 , T2 , . . . ∈ L(X) converges to zero in the strong
(or the “ultra-strong”) topology to zero, then kT1 k, kT2 k, . . . is a bounded sequence
by the uniform boundedness theorem.
654 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

Now one can use that the norms kT k of T ∈ M(K) ⊂ L(K) and T ∈ L(`2 ) =
M(K) coincide and that strict and *-ultra-strong topologies on bounded parts of
M(K) coincide.
It follows that differences in the convergence structures for *-ultra-strong or
strict topology on M(K) can be only found if one considers unbounded nets of high
cardinality that converge in *-ultra-strong topology but not in strict topology.
Suppose that there exists an isomorphism ψ from a C *-algebra C ⊗ K onto
B. Let M(ψ) denote its strictly continuous extension to an isomorphism from
M(C ⊗ K) onto M(B). Its inverse is given by M(ψ)−1 = M(ψ −1 ) and, therefore,
is strictly continuous. Notice that there are natural strictly continuous unital *-
monomorphisms M(ϕ1 ) : M(C) → M(C ⊗ K) and M(ϕ2 ) : M(K) → M(C ⊗ K)
that are the strictly continuous extensions of the non-degenerate C *-morphisms
ϕ1 : C → M(C ⊗K) and ϕ2 : K → M(C ⊗K) that are determined by the properties
ϕ1 (c)(c1 ⊗ k) = (c · c1 ) ⊗ k and ϕ2 (k)(c ⊗ k1 ) = c ⊗ (k · k1 ). We write 1 ⊗ t for
M(ϕ2 )(t) if t ∈ M(K) ∼ = L(`2 ).
There is a non-degenerate C *-morphism λ : K → M(B) that is given by λ(k) :=
M(ψ)(1⊗k) for k ∈ K, i.e., λ = M(ψ)◦ϕ2 . It extends to a unital *-monomorphism
M(λ) = M(ψ) ◦ M(ϕ2 ) : M(K) → M(B), that is strictly continuous on bounded
parts. The isometries sn := M(λ)(tn ) = M(ψ)(1 ⊗ tn ) have the property that
sn s∗n converges strictly to 1 in M(B), because
P P ∗
tn tn converges strictly to 1
and M(λ) is strictly continuous on bounded parts.
Conversely suppose that M(B) contains a sequence s1 , s2 , . . . of isometries in
M(B) such that n sn s∗n converges strictly to 1.
P

Then δ∞ (b) := n sn bs∗n converges strictly to an element δ∞ (b) of M(B) and

P

the map δ∞ : M(B) → M(B) is strictly continuous and unital on M(B).

Indeed: If b ∈ B and ε > 0, then there is m ∈ N with kb−(s1 s∗1 +. . .+sm s∗m )bk <
ε/2 and there exists a contraction e ∈ B+ with k s∗k b − es∗k b k < ε/2m . It gives
kb − δ∞ (e)bk < ε.
Thus δ∞ (B)B is dense in B, i.e., δ∞ |B is non-degenerate. It follows that
δ∞ = M(δ∞ |B) is the unique strictly continuous unital extension of δ∞ |B to M(B).
Then there is a unique C *-morphism µ : K → δ∞ (B)0 ∩ M(B) that satisfies
µ(ek,` ) := sk s∗` .
The range projections pn := sn s∗n satisfy pm ≤ 1 − pn because pn + pm ≤ 1
for m 6= n. It implies pm pn = 0. Thus n sn s∗n ≤ 1 implies s∗i sj = s∗i pi pj sj = 0
P
S
for i 6= j. In particular, B is the closure of its *-subalgebra B0 := n Pn BPn for
Pn := p1 + · · · + pn ∈ M(B).
Let ei,j denote the canonical matrix units of K ⊂ M∞ (C). We define an
isometry T ∈ M(B ⊗ K) with T T ∗ = 1M(B) ⊗ e1,1 by the “row matrix”

T := s1 ⊗ e1,1 + s2 ⊗ e1,2 + . . . . (1.3)

 1. STRICT CONVERGENCE IN MULTIPLIER ALGEBRAS 655

We use the natural embedding

M(B ⊗ K) ⊂ M∞ (M(B)) ⊆ B ∗∗ ⊗L(`2 (N))

that describes the elements of M(B ⊗ K) as matrices over M(B) then above T
becomes the row matrix

T ∼
= [s1 , s2 , . . .] ∈ M1,∞ (M(B)) ⊂ M∞ (M(B)) .
P
The sum n sn ⊗ e1,n is unconditional strictly convergent in M(B ⊗ K) by Remark
(2) if we take there yn := sn ⊗ 1, dn := 1 ⊗ 1 and xn := 1 ⊗ en,1 . Since

(s∗n ⊗ em,1 )T = (s∗n ⊗ em,1 )(sn ⊗ e1,n ) = 1 ⊗ em,n ,

we get T ∗ (sn ⊗e1,n ) = 1⊗en,n and T ∗ T = 1⊗1. The identities (sk ⊗e1,k )(s∗n ⊗en,1 ) =
δk,n sk s∗n ⊗ e11 , imply that T T ∗ = 1 ⊗ e11 . It follows that the map

H : b ∈ M(B) 7→ H(b) := T ∗ (b ⊗ e11 )T

defines a unital *-monomorphism H from M(B) into M(B ⊗ K) that is strictly

continuous (on bounded parts) and maps B into B ⊗ K, because T ∈ M(B ⊗ K).
The *-monomorphism H maps M(B) onto M(B ⊗ K), i.e., is an isomorphism,
H maps B onto B ⊗ K, and the inverse map H −1 : M(B ⊗ K) → M(B) is given
on M(B) K by H −1 (b ⊗ ejk ) = δ∞ (b)sj s∗k :
Indeed, if a ∈ M(B ⊗K), then a = T ∗ (T aT ∗ )T = H(T aT ∗ ) and T aT ∗ = b⊗e11
for some b ∈ M(B).
Since H is verified to be an isomorphism from M(B) onto M(B ⊗ K), H(B)
must be a closed ideal of M(B ⊗ K). The definition H := T ∗ ((·) ⊗ e11 )T of H with
T ∈ M(B ⊗ K) implies that H(B) ⊆ B ⊗ K.
Since B is an essential ideal of M(B) that is dense in M(B) with respect to
the strict topology, H(B) is a closed ideal of B ⊗ K that is strictly dense in B ⊗ K.
Thus H(B) = B ⊗ K.
(Alternatively one can prove that b ⊗ en,m ∈ T ∗ (B ⊗ e1,1 )T by showing that
B ⊗ 1 = T ∗ (δ∞ (B) ⊗ e1,1 )T , T ∗ (sn s∗m ⊗ e1,1 )T = 1 ⊗ en,m and δ∞ (B)sn s∗m ∈ B.
See more below.)
Let ψ := H|B and ϕ : B ⊗ K → B the inverse of ψ = H|B. Then M(ϕ) and
M(ψ) are strictly continuous and H = M(ϕ)−1 .
We can use the above defined isomorphism ϕ : B ⊗ K → B to define the other
in part (8) proposed C *-morphisms:
We have that (s∗` ⊗ e11 )T = 1 ⊗ e1,` . It implies T ∗ (sk bs∗` ) ⊗ e11 T = b ⊗ ek,` and
T ∗ (δ∞ (b) ⊗ e11 )T = b ⊗ 1 for all b ∈ M(B).
Next blue NEW / BETTER APPROACH?:
Let δ∞ (b) := n sn bs∗n where b ∈ M(B), s1 , s2 , . . . ∈ M(B) with n sn s∗n =
P P

1M(B) and s∗m sn = δm,n 1M(B) .

Let T ∈ M(B ⊗ K) the isometry defined by Equation (1.3).
656 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

There is a non-degenerate C *-morphism µ : K → δ∞ (B)0 ∩ M(B) with

T ∗ (µ(ei,j ) ⊗ e11 )T = 1 ⊗ ei,j , e.g. µ(c) ⊗ e11 := T (1 ⊗ c)T ∗ .
We can µ define µ : K → M(B) explicit by µ(ek,` ) := sk s∗` .
The equations T (1 ⊗ em,1 ) = sm ⊗ e1,1 imply (1 ⊗ e1,n )T ∗ = s∗n ⊗ e1,1 and

T (1 ⊗ em,n )T ∗ = T (1 ⊗ em,1 )(1 ⊗ e1,n )T ∗ = sm s∗n ⊗ e11 .

Since
δ∞ (b)sm s∗n = sm bs∗n = sm s∗n δ∞ (b) ,
∗
P
the C *-morphism [αjk ] → jk αjk sj sk extends to a C *-morphism µ : K →
δ∞ (M(B))0 ∩ M(B) and is uniquely defined by µ(k) ⊗ e11 = T (1 ⊗ k)T ∗ .
Better: To be shown directly with above H:
∗
P
By Remark (2), δ∞ (b) := n sn bsn is unconditional strictly convergent for
each b ∈ M(B), and δ∞ defines a strictly continuous unital *-monomorphism from
M(B) into M(B), cf. Proof of Lemma 5.1.2(i,ii), or compare the proof of the
equation δ∞ (b) ⊗ e1,1 = T (b ⊗ 1)T ∗ where T ∈ M(B ⊗ K) is the isometry defined
by Equation (1.3).
(b) There exists a non-degenerate *-monomorphism µ : K → M(B), where
K := K(`2 (N)), with the properties µ(K) ⊆ δ∞ (B)0 ∩M(B) such that δ∞ (b)µ(k) ∈ B
for b ∈ B and k ∈ K and such that the unique C*-morphism ψ : B ⊗ K → B with
ϕ(b ⊗ k) = δ∞ (b)µ(k) is a *-isomorphism from B ⊗ K onto B.
The above defined isomorphism ϕ : B ⊗ K → B, with H = M(ϕ)−1 , ψ :=
ϕ = H|B, T (b ⊗ k)T ∗ = (δ∞ (b)µ(k)) ⊗ e11 and H(b) = T ∗ (b ⊗ e1,1 )T , does the
−1

job if µ is defined by µ(k) ⊗ e1,1 := T (1 ⊗ k)T ∗ .

(c) The non-degenerate *-monomorphism µ : K → M(B) has the property
µ(K)0 ∩ M(B) = δ∞ (M(B)) . In particular, δ∞ (M(B)) is closed in M(B) with
respect to the strict topology of M(B).
This can be seen by using the isomorphism M(ϕ) : M(B ⊗ K) → M(B) from
M(B ⊗ K) onto M(B).
(d) The strictly continuous extension M(µ) : M(K) ∼ = L(`2 ) → M(B) of µ is
a strictly continuous unital *-monomorphism, has image in δ∞ (M(B))0 ∩ M(B)
and M(µ)(L(`2 )) is closed in the strict topology of M(B).
Next new version of proof ?
The natural extension θ : M(B) → B ∗∗ of the unital *-monomorphism
ηB : B → B ∗∗ from M(B) into the W*-algebra B ∗∗ is continuous with respect to
the strict topology on M(B) and the *-ultra-strong topology on B ∗∗ .
If h : A → M(B) is any non-degenerate *-morphism, then µ extends uniquely to
a unital and strictly continuous map M(h) : M(A) → M(B). Then M(h)(1) = 1
and M(h)(M(A)) is in M(B) the strict closure of h(A). Since θ : M(B) → B ∗∗
is continuous with respect to the strict topology on M(B) and the *-ultra-strong
topology on B ∗∗ , it follows that the *-ultra-strong closure of θ(h(A)) in B ∗∗ contains
 1. STRICT CONVERGENCE IN MULTIPLIER ALGEBRAS 657

θ(M(h)(A)). In particular, the *-ultra-weak closure of h(A) contains the unit

1 := 1B ∗∗ of B ∗∗ .
If M is a von-Neumann algebra and Φ : L(`2 ) ∼ = M(K) → M is a C *-morphism
such that the unit 1M of M is contained in the *-ultra-weak closure of Φ(K) then
Φ is normal on L(`2 ), i.e., is a unital W*-algebra isomorphism from L(`2 ) onto the
*-ultra-weak closure of Φ(K).
In particular, Φ(M(K)) is closed in the *-ultra-strong topology of B ∗∗ , and
the topology induced from this topology on M(K) ∼ = L(`2 ) is the same as the
*-ultra-strong topology on L(`2 ).
Now take K and µ : K → M(B) in place of A and h above. Let Ψ := θ ◦ M(µ),
then we get that the restriction of the strict topology on M(B) to M(µ)(M(K)) is
finer than the topology induced from the *-ultra-strong topology on M(K). Since
M(µ) : M(K) → M(B) is strictly continuous with respect to the strict topology
on M(K) and on M(B), it follows that the topology induced on M(K) by M(µ)−1
is equal to or coarser than the strict topology on M(K). So it is between the *-
ultra-strong topology and the strict topology on L(`2 ) ∼= M(K). It is easy to see -
by the definitions and by the remark in Part (i) – that the latter both coincides on
bounded parts of M(K).
Thus, on bounded parts of M(K) the topology induced from the strict topology
on M(B) via M(µ)−1 coincide.
The image M(µ)(M(K)) is strictly closed, because θ is continuous with respect
to the strict and *-ultra-strong topology, and θ(M(µ)(M(K))) is *-ultra-strongly
closed in B ∗∗ .
If one uses the isomorphism M(ϕ) from M(B ⊗ K) onto M(B), then one can
see that M(µ)(L(`2 )) is the isomorphic image of 1 ⊗ L(`2 ) ⊆ M(B ⊗ K).
It is not difficult to see that 1 ⊗ L(`2 ) is strictly closed in M(B ⊗ K) and that
the induced topology on M(K) ∼ = L(`2 ) coincides on bounded parts of L(`2 ) with
the *-strong topology.
Indeed, if e ∈ B+ is a strictly positive positive contraction, then C ∗ (e) ⊗ K is
a non-degenerate C *-subalgebra of B ⊗ K and 1 ⊗ M(K) is naturally contained
in M(C ∗ (e) ⊗ K) ⊆ M(B ⊗ K) and the representation T ∈ M(K) 7→ 1 ⊗ T ∈ is
strictly continuous, because it extends a non-degenerate representation of K. Thus,
the topology on M(K) induced from T 7→ 1 ⊗ T is coarser than the strict topology
on M(K). Let χ : C ∗ (e) → C a character. It defines a *-epimorphism λ from
C ∗ (e) ⊗ K onto K. Then M(λ) : M(C ∗ (e) ⊗ K) → M(K) is strictly continuous,
and its restriction to 1 ⊗ M(K) is continuous with respect to the strict topology on
M(C ∗ (e) ⊗ K) defined by C ∗ (e) ⊗ K.
The strict topology on M(C ∗ (e) ⊗ K) coming from C ∗ (e) ⊗ K coincides on
bounded parts with the restriction to M(C ∗ (e) ⊗ K) of the strict topology on
M(B ⊗ K), because on bounded parts the strict topology can be defined by the
658 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

semi-norms d ∈ M(B ⊗ K) 7→ kd(e ⊗ k1 )k + k(e ⊗ k2 )dk with k1 , k2 ∈ K+ of finite

rank. In particular this topologies coincide on bounded parts of 1 ⊗ M(K).
Summing up, we get strictly continuous C *-morphisms M(µ) : M(K) → 1 ⊗
M(K) ⊆ M(C ∗ (e)⊗K) ⊆ M(B⊗K) and M(λ) : M(C ∗ (e)⊗K) → M(K) such that
the strict topology on bounded parts of M(C ∗ (e) ⊗ K) coincides with the topology
induced from the strict topology of M(B ⊗ K). Moreover M(λ) ◦ M(µ) : M(K) →
M(K) is a strictly continuous isomorphism of M(K) Thus it is given by an inner
automorphism of M(K). It follows that all the induced strict topologies are the
same on bounded parts of M(K).
The topology on L(`2 ) ∼
= M(K) induced from the strict topology on M(K) is
on bounded parts the same as the *-strong topology on L(`2 ).
Thus, the topology induced by M(µ)−1 on L(`2 ) is on bounded parts the same
as the *-strong topology on L(`2 ).
Since the unit-ball of L(`2 ) is complete in the *-strong topology, it follows that
M(µ)(M(K)) is closed in M(B) with respect to the strict topology on M(B).
By Lemma 5.1.2(i,ii), all infinite repeats δ∞ are unitarily equivalent in M(B).
Therefore we need only to verify the observations by choosing case by case a suitable
“fitting” infinite repeat δ∞ .
The isomorphism ψ from B⊗K onto B extends uniquely to a strictly continuous
*-isomorphism M(ψ) from M(B ⊗ K) onto M(B). The natural *-monomorphism
λ : K → M(B ⊗ K) with λ(k1 )(b ⊗ k2 ) = b ⊗ k1 k2 is non-degenerate and extends to
a strictly continuous unital *-monomorphism M(λ) from M(K) into M(B ⊗ K).
Since there is an isomorphism ι from K ⊗ K onto K, given by an isomorphism
`2 ∼
= `2 ⊗2 `2 , there exists isometries t1 , t2 , . . . ∈ M(K) such that n tn t∗n strictly
P

convergent to 1, and that the commutant δ∞ (K)0 ∩ M(K) of δ∞ (K) contains a

non-degenerate copy H(K) of K, that satisfies δ∞ (k1 )H(k2 ) = ι(k1 ⊗ k2 ) where
δ∞ (k) := n tn kt∗n .
P

Let sn := M(ψ) ◦ M(λ)(tn ). Then s1 , s2 , . . . is a sequence of isometries in

M(B) such that n sn s∗n converges strictly to 1.
P

And δ∞ (b) := n sn bs∗n is a non-degenerate C *-morphism, such that µ : K →

P

δ∞ (B)0 ∩ M(B) (given by µ(k) := M(ψ)(H(k))) satisfies that δ∞ (b)µ(k) ∈ B.

It follows that there is a (new) C *-morphism ϕ from B ⊗ K into B that satisfies
ϕ(J ⊗ K) = J for all closed ideals of B and defines an isomorphism from B ⊗ K onto
B that is considerably different from the above defined by an arbitrary isomorphism
of B and some C ⊗ K.
In particular, the unique strictly continuous extension M(µ) = M(ψ) ◦
M(H) : M(K) → M(B) of µ : K → M(B) is a unital *-monomorphism from L(`2 )
into M(B), and has the property that δ∞ (M(B)) and M(µ)(L(`2 )) commute and
the C *-algebra generated by δ∞ (B) · µ(K) is strictly dense in M(B).
NEXT old or new ???
 1. STRICT CONVERGENCE IN MULTIPLIER ALGEBRAS 659

There exists isometries t1 , t2 , . . . in δ∞ (M(B))0 ∩ M(B) such that ∗

P
n tn tn
converges strictly to 1. This is because δ∞ (M(B))0 ∩ M(B) is strictly closed
and there is a non-degenerate *-monomorphism µ : K(`2 ) → M(B) with image in
δ∞ (M(B))0 ∩ M(B). Then M(µ) : M(K) ∼ = L(`2 ) → M(B) is strictly continuous,
0
unital and has image in δ∞ (M(B)) ∩ M(B).
(9): Recall that a C *-morphism h : A → M(B) is non-degenerate if and only
if the linear span of h(A)B is dense in B. The Cohen factorization theorem shows
that this is equivalent to B = h(A) · B (as set of products h(a)b).
It follows that non-degenerate h satisfies h(A)(h(A) + B) ⊆ h(A) + B and that
h(A)(h(A) + B) is norm-dense in the C *-subalgebra h(A) + B of M(B).
Indeed, if a ∈ A, b ∈ B and ε > 0 are given, then let γ := ε/(3 + kbk) We
find f ∈ A with kb − h(f )bk < γ. There exists a positive contraction e ∈ A+ with
kf − ef k < γ and ka − eak < γ. Straight calculation shows that k(h(a) + b) −
h(e)(h(a) + b)k < ε.
Let H : A → M(h(A)+B) the C *-morphism with H(e)(h(a)+b) := h(e)(h(a)+
b) for a, e ∈ A and b ∈ B. The elements tn := M(h)(sn ) ∈ M(B) (respectively
tn := M(H)(sn ) ∈ M(B) are isometries in M(h)(M(A)) ⊆ M(B) (respectively
P ∗
in M(H)(M(A)) ⊆ M(h(A) + B)), and tn tn converges strictly to 1 in M(B)
(respectively in M(h(A) + B)), because M(h) (respectively M(H)) is strictly con-
tinuous and unital by Part (1).
It follows that B and h(A) + B are stable by Part (8). 

Lemma 5.1.2. Suppose that B is a stable and σ-unital C*-algebra, and let
s1 , s2 , . . . ∈ M(B) a sequence of isometries such that n sn s∗n converges strictly to
P

1 in M(B), – see Remark 5.1.1(8) for the existence of {s1 , s2 , . . .}.

(i) The map δ∞ : M(B) → M(B) is a strictly continuous unital *-monomor-

0
phism. If δ∞ : M(B) → M(B) comes from another sequence t1 , t2 , . . . of
P ∗ 0
isometries in M(B) such that tn tn converges strongly to 1, then δ∞
2
is unitarily equivalent to δ∞ . In particular, δ∞ is unitarily equivalent to
δ∞ .
The algebra δ∞ (M(B))0 ∩ M(B) is strictly closed and contains a se-
qn qn∗ converges strictly to 1.
P
quence of isometries q1 , q2 , . . . such that
0
(ii) δ∞ (M(B)) ∩ M(B) contains a copy of O2 (with canonical generators
r1 , r2 ) unitally, and δ∞ ⊕r1 ,r2 idM(B) is unitarily equivalent to δ∞ .
(iii) Let D := O2 ⊗ K denote the stabilized Cuntz algebra O2 , q1 ≤ q2 ≤ . . . an
increasing sequence of projections in D and p0 is a nonzero projection in
D.
Then there exists norm-continuous maps t 7→ U (t) from R+ into
the set of unitaries in 1 + D ⊆ M(D) such that U (0) = 1 and qn ≤
U (t)∗ p0 U (t) and qn 6= U (t)∗ p0 U (t) for t ≥ n.
660 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

If, in addition, qn 6= qn+1 for n = 1, 2, . . ., then one can find the path
t 7→ U (t) ∈ M(D) such that qn ≤ U (t)∗ p0 U (t) ≤ qn+3 and U (t)∗ p0 U (t) 6∈
{qn , qn+3 } for t ∈ [n + 1, n + 2].
(iv) There are isometries S, T ∈ δ∞ (M(B))0 ∩ M(B) with SS ∗ + T T ∗ = 1 and
a norm-continuous map t ∈ [0, ∞) 7→ U (t) ∈ δ∞ (M(B))0 ∩ M(B) into
the unitaries of M(B), such that the norm-continuous maps

S0 , T0 : t ∈ R+ 7→ S0 (t), T0 (t) ∈ δ∞ (M(B))0 ∩ M(B)

given by S0 (t) := U (t)∗ SU (t) and T0 (t) := U (t)∗ T U (t) for t ∈ R+ satisfy
limt→∞ S0 (t)∗ b = 0 and limt→∞ T0 (t)∗ b = b for every b ∈ B.
(v) Let C a separable C*-subalgebra of M(B), T : C → M(B) a completely
positive contraction, and suppose that there exists a contraction g ∈ M(B)
with g ∗ cg − δ∞ (T (c)) ∈ B for every c ∈ C.
Then there exists a norm-continuous map t ∈ R+ 7→ S(t) ∈ M(B)
into the contractions in M(B) such that S(t)∗ cS(t) − δ∞ (T (c)) ∈ B for
every t ∈ R+ , and limn→∞ kδ∞ (T (c)) − S(t)∗ cS(t)k = 0 for all c ∈ C.
If, moreover, 1M(B) ∈ C and T is unital then t 7→ S(t) can be found
such that, moreover, S(t)∗ S(t) = 1 for t ∈ R+ .
(vi) Suppose that 1M(B) ∈ C ⊆ M(B), C is separable, and that T is a unital
C*-morphism such that there exists a contraction g ∈ M(B) with g ∗ cg −
δ∞ (T (c)) ∈ B for every c ∈ C. Then idC asymptotically absorbs T , i.e.,
idC ⊕T and idC are unitarily homotopic in the sense of Definition 5.0.1.
(vii) If f ∈ M(B) and if e ∈ B+ is a strictly positive element, then the norm
closures B0 of f ∗ Bf and B1 of δ∞ (f ∗ )Bδ∞ (f ) are σ-unital with strictly
positive elements f ∗ ef , respectively with δ∞ (f ∗ )eδ∞ (f ).
There is a natural isomorphism ψ from B1 onto B0 ⊗ K, that is given by

ψ(b) := [s∗j bsk ]jk ∈ B0 ⊗ K ⊆ M∞ (B0 ) .

sn ds∗n for
P
Proof. (i): The strict convergence of the n-th partial sums of
d ∈ M(B) and then of δ∞ (dτ ) to δ∞ (d) for a strictly convergent bounded net
in {dτ } ⊆ M(B) with strict limit d ∈ M(B) can be seen from the argument in
Remark 5.1.1(2). Since n sn s∗n = 1 (strictly), the map δ∞ is unital. It is clear
P

that δ∞ is an isometric and completely positive map. It is multiplicative because

δ∞ (d)δ∞ (e)sn s∗n = sn ds∗n sn es∗n = δ∞ (de)sn s∗n .

sn t∗n =: U is strictly convergent to a unitary because the strong

P
The sum
convergence of U and that of U ∗ can be seen by the argument in Remark 5.1.1(2).
This also shows that U ∈ M(B) is unitary and U ∗ δ∞ (d)U = δ∞ 0
(d) for d ∈ M(B).
The sufficient criterium in Remark 5.1.1(2) shows that n,m sn sm (sn sm )∗ con-
P
2
verges unconditional strictly to 1M(B) . Thus, δ∞ is an infinite repeat of idM(B) .
Compare Remark 5.1.1(8)
and use ρ(ejk ) := sj s∗k ????
 1. STRICT CONVERGENCE IN MULTIPLIER ALGEBRAS 661

‘‘It is clear that’’ ???????

The map T ∈ M(B) 7→ ST − T S ∈ M(B) is strictly continuous for each
S ∈ M(B), thus with S := δ∞ (d) we get that {δ∞ (d)}0 ∩ M(B) is strictly closed
for every d ∈ M(B)+ . It implies that δ∞ (M(B))0 ∩ M(B) is strictly closed as
intersection of strictly closed C *-subalgebras. It contains a non-degenerate copy
of K(`2 ) by Remark 5.1.1(8). Therefore, there is a strictly continuous unital *-
monomorphism ρ : M(K) ∼ = L(`2 ) → M(B) such that the image of ρ commutes
with δ∞ (M(B)).
P ∗
If t1 , t2 , . . . is a sequence of isometries in M(B) (with tn tn = 1 strictly)
then ρ(t1 ), ρ(t2 ), . . . are isometries in δ∞ (M(B))0 ∩ M(B) (with ρ(tn )ρ(tn )∗ = 1
P

strictly).
(ii): Let the isometries s1 , s2 , . . . ∈ M(B) the generators of a copy of O∞ as
established in Remark 5.1.1(8). By part (i), there exists a unital *-monomorphism
Λ from L(`2 ) ∼ = M(K) into δ∞ (M(B))0 ∩ M(B). Take isometries R1 , R2 ∈ L(`2 )
with R1 R1 + R2 R2∗ = 1 and let rk := Λ(Rk ) ∈ M(B) for k = 1, 2. Then C ∗ (r1 , r2 )
∗

is a copy of O2 in δ∞ (M(B))0 ∩ M(B) that contains 1M(B) .

0
(b) := n tn bt∗n of the identity
P
The map δ∞ ⊕r1 ,r2 idM(B) is an infinite repeat δ∞
map idM(B) of M(B) that is given by the sequence of isometries t1 := r2 , tn+1 :=
P ∗
tn tn = r2 r2∗ + r1 ( sn s∗n )r1∗ converges
P
r1 sn for n = 1, 2, . . ., because the series
strictly to 1. It follows that δ∞ ⊕r1 ,r2 idM(B) is unitarily equivalent to δ∞ by (i).
(iii): We use here (and only here !) the notation p < q for projections if p ≤ q
and p 6= q.
Where next is really used? For what?
Let p0 := 1 ⊗ e11 ∈ D and a p ∈ D := O2 ⊗ K non-zero projection. Since D
has an approximate unit consisting of projections we find projections r, s ∈ D and
a unitary v1 ∈ U0 (D + C · 1) in the connected component of 1 inside the unitaries
in D + 1 such that p0 < r < s, p < r < s, v1 (1 − s) = (1 − s) and v1∗ p0 v1 = p.
Is r really necessary for something above?
Let p, q, r, s ∈ D with p < q < s and p < r < s. Then there exists a con-
tinuous map t ∈ [0, 1] 7→ v(t) ∈ D with v(t)∗ v(t) = v(t)v(t)∗ = v(0) = s − p and
v(1)∗ qv(1) = r.
Use below that O2 has property (sq), because O2 = E2 /K.
It implies: O2 is K1 -bijective.

Every projection in O2 is properly infinite,

and
δ2 is homotopic to id.
(Implies K∗ (O2 ) = 0.
662 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

This happens because, by [172], any two projections P, Q ∈ O2 \ {0, 1} are

unitarily equivalent in O2 ∼
= (s − p)D(s − p), and U(O2 ) = U0 (O2 ).
The unitary equivalence of non-trivial projections in O2 implies that all non-
zero projections in D = O2 ⊗ K are MvN-equivalent, thus (s − p)D(s − p) is
isomorphic to O2 , and we can pass from O2 to (s − p)D(s − p).
If there is n0 ∈ N such that qn = qn0 for all n ≥ n0 , then we find projections
r, s ∈ D and a continuous map t 7→ w(t) ∈ D + 1 into the unitaries of D + 1 such
that w(0) = 1, qn0 < r < s and q := w(1)∗ p0 w(1) < s.
Thus 0 < r < s and 0 < q < s, and there is a continuous map t 7→ v(t) ∈ D with
v(t) v(t) = v(t)v(t)∗ = v(0) = s and v(1)∗ qv(1) = r. Let U (t) := w(t)(v(t)+(1−s))
∗

for t ∈ [0, 1] and U (t) = U (1) for t > 1. The map t 7→ U (t) is a continuous path in
the U0 (D + C1) ∩ (1 + D) of the desired type.
If the sequence q1 ≤ q2 ≤ . . . becomes not stationary, than we may assume that
0 < qn < qn+1 for all n ∈ N, – simply by re-indexing the non-equal qn and changing
the parameter t ∈ R. Therefore we find continuous maps t ∈ [0, 1] 7→ vn (t) ∈ D into
the partial unitaries of D such that vn (t)∗ vn (t) = vn (t)vn (t)∗ = vn (0) = qn+3 − qn
and vn (1)∗ qn+1 vn (1) = qn+2 .
We find a projection s ∈ D with 0 < q3 < s and a continuous map t 7→ w(t) from
[0, 1] into the unitaries of D + 1, such that q := w(1)∗ p0 w(1) < s, w(0) = 1. Then
there is a continuous map t 7→ v0 (t) ∈ D with v0 (t)∗ v0 (t) = v0 (t)v0 (t)∗ = v0 (0) = s
and v0 (1)∗ qv0 (1) = q3 .
Let U (t) := w(t)(v0 (t) + (1 − s)) for t ∈ [0, 1] and, by induction, U (n + t) :=
U (n)(qn+2 + vn+1 (t) + (1 − qn+4 )) for t ∈ [0, 1] and n = 1, 2, . . .. Then t 7→ U (t) is a
continuous map from R+ to the unitaries in D + 1, with U (0) = 1, U (n)∗ p0 U (n) =
qn+2 and qn ≤ U (t)∗ p0 U (t) and U (t)∗ p0 U (t) 6= qn for t ∈ [n, n + 1].
(iv): As noticed in Remark 5.1.1(8) (combined with part (i)), there is a non-
degenerate *-monomorphism ψ : K → M(B) such that M(ψ) : M(K) → M(B) is a
unital strictly continuous *-monomorphism with image in δ∞ (M(B))0 ∩M(B). We
identify M(K) with such a copy M(ψ)(M(K)) of M(K) in δ∞ (M(B))0 ∩ M(B).
Since `2 ⊗2 `2 ∼= `2 there is a non-degenerate copy D of O2 ⊗ K contained
in M(K) = L(`2 ), i.e., DK = K and the inclusion M(D) ⊆ M(K) is unital.
p0 := 1O2 ⊗ p11 is in D, but is not in K. There exists an increasing sequence of
projections q0 < q1 < q2 < · · · in 1O2 ⊗K ⊆ D with q0 ≤ q1 , q1 −q0 6= 0, q1 −q2 6= 0,
and limn→∞ kqn k − kk = 0 for every k ∈ K. It follows that (qn ) converges strictly
to 1 ∈ M(B).
By (iii) there is a norm-continuous map t 7→ U (t) into the unitaries of D + 1 ⊆
M(K) such that qn ≤ U (t)∗ p0 U (t) for t ≥ n.
In M(K) = L(`2 ) there are isometries T and S with p0 T = p0 , T T ∗ = q1 ,
∗
SS = 1 − q1 , because 1 − p0 , q1 − p0 and 1 − q1 are not in K (recall:
D ∩ K = {0}). Let S0 (t) := U (t)∗ SU (t) and T0 (t) := U (t)∗ T U (t). Then
qn S(t) = qn (U (t)∗ p0 U (t))U (t)∗ SU (t) = qn U (t)∗ p0 SU (t) = 0 and qn T (t) =
 1. STRICT CONVERGENCE IN MULTIPLIER ALGEBRAS 663

qn (U (t)∗ p0 U (t))U (t)∗ T U (t) = qn (U (t)∗ p0 U (t)) = qn for t > n, and the maps
t → S0 (t), T0 (t) ∈ M(K) ⊆ M(B) are norm-continuous.
Thus, limt→∞ S0 (t)∗ qn b = 0 and limt→∞ T0 (t)∗ qn b = qn b for every b ∈ B and
n ∈ N.
S
The desired result follows, because n (qn B) is dense in B.
(v): Let S(t) := gS0 (t) where S0 (t) ∈ δ∞ (M(B))0 ∩ M(B) is as in (iv). Then
t 7→ S(t) is a norm-continuous map from [0, ∞) into the contractions in M(B),

(δ∞ ◦ T )(c) − S(t)∗ cS(t) = S0 (t)∗ (δ∞ (T (c)) − g ∗ cg)S0 (t) ∈ B

for t ∈ R+ , and limt→∞ kS(t)∗ cS(t) − δ∞ (T (c))k = 0.

If 1 ∈ C and T (1) = 1, then 1−S(t)∗ S(t) = S0 (t)∗ (δ∞ (T (1))−g ∗ 1g)S0 (t) is in B
and tends to zero in norm. Since the norm-limit of S(t)∗ S(t) is 1, there is t0 ≥ 0 such
that S(t)∗ S(t) is invertible for every t ≥ t0 . Then t → (S(t0 + t)∗ S(t0 + t))−1/2 ∈ B
is norm-continuous, S 0 (t) := S(t0 + t)(S(t0 + t)∗ S(t0 + t))−1/2 is norm-continuous
and satisfies S 0 (t)∗ S 0 (t) = 1, S 0 (t)∗ cS 0 (t) − δ∞ (T (c)) ∈ B and lim S 0 (t)∗ cS 0 (t) =
δ∞ (T (c)) for c ∈ C.
(vi): The statement is independent of the chosen copy and generators {s1 , s2 },
{t1 , t2 } of O2 in M(B), because the unital completely positive map idC ⊕s1 ,s2 T is
unitarily equivalent to idC ⊕t1 ,t2 T by the unitary w = t1 s∗1 +t2 s∗2 . In the following ⊕
is formed with help of the same generators of O2 ⊆ M(B). Let V := δ∞ ◦T . By (ii),
there is a unitary U0 ∈ M(B) with U0∗ V (·) U0 = T ⊕ V . If we have found a norm-
continuous map t 7→ U1 (t) into the unitaries of M(B) with V (c)⊕c−U1 (t)∗ c U1 (t) ∈
B and limn→∞ kV (c) ⊕ c − U1 (t)∗ c U1 (t)k = 0 for all c ∈ C, then

U (t) := U1 (t)(U0 ⊕ 1)Ud (1 ⊕ U1 (t)∗ )

is as desired, where Ud is a unitary in U0 (O2 ) with Ud∗ (a ⊕ b) ⊕ c)Ud = a ⊕ (b ⊕ c)

as in Proposition 4.3.2(iv).
Thus, it is enough to prove the existence of U (t) for V = δ∞ ◦ T (in place of
T ). By (iv) there exists a norm continuous map t ∈ R+ 7→ S(t) ∈ M(B) into
the isometries of M(B) such that S(t)∗ c S(t) − V (c) ∈ B for every t ∈ R+ , and
limn→∞ kV (c) − S(t)∗ c S(t)k = 0 for all c ∈ C. The relative commutant of V (C)
in M(B) contains a copy of O2 unitally, by Part (ii).
Now we prove a more general result than stated in Part (vi):
Suppose that C is a C*-subalgebra C ⊆ M(B) with 1M(B) ∈ C and that V is
a unital C*-morphism V : C → M(B) such that there are a unital copy of O2 in
V (C)0 ∩M(B) and a norm-continuous map t 7→ S(t) from R+ into the contractions
of M(B) with S(t)∗ c S(t) − V (c) ∈ B and V (c) = lim S(t)∗ c S(t). Then there is
a norm-continuous map t 7→ U (t) from R+ into the unitaries of M(B) such that
c ⊕ V (c) − U (t)∗ c U (t) ∈ B for every t ∈ R+ and

lim k c ⊕ V (c) − U (t)∗ c U (t) k = 0 for c ∈ C .

t→∞
664 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

By Cb (R+ , M(B)) we denote the bounded continuous maps from R+ into

M(B). We let E := Cb (R+ , M(B))/ C0 (R+ , B) and consider M(B) as a subalge-
bra of E . Then the assumptions imply that C ,→ E dominates the C *-morphism
V : C → E and that V (C)0 ∩E contains a copy of O2 unitally. Thus, by Proposition
4.3.5(i) there is a unitary u ∈ E such that u∗ cu = c ⊕ V (c) for c ∈ C.
Let t 7→ W (t) a contraction in Cb (R+ , M(B)) which represents u. Then
1 − W (t)∗ W (t) and 1 − W (t)W (t)∗ are in B, lim k1 − W (t)W (t)∗ k = 0, lim k1 −
W (t)∗ W (t)k = 0. Thus there exists t0 ≥ 0 with k1 − W (t)∗ W (t)k ≤ 1/2 and
k1 − W (t)W (t)∗ k ≤ 1/2 for t ≥ t0 . Let U (t) := W (t + t0 )(W (t + t0 )∗ W (t + t0 ))−1/2
for t ≥ 0. Then t 7→ U (t) is as desired.
(vii): Let g := δ∞ (f ), pk := sk s∗k and qn := ∗
P
k≤n pk . Recall B0 := f Bf
∗ ∗ ∗ ∗
and B1 := g Bg . We have pk g = gpk = sk f sk , pk B1 ⊆ B1 B = sj (sj Bsk )sk ⊆
s∗j Bsk ⊆ B, and pj gBgpk = sj f ∗ s∗j Bsk f s∗k = sj f ∗ Bf s∗k . Thus, pj B1 pk = sj B0 s∗k
and s∗j B1 sk = B0 . Since qn g = gqn , kqn g ∗ bg − g ∗ bgk ≤ kgk2 k(qn b − b)gk → 0 for
n → ∞ and b ∈ B, we get qn B1 + B1 qn ⊆ B1 and limn→∞ kqn bqn − bk = 0 for all
b ∈ B1 .
Let ψ(b) := [s∗j bsk ]j,k ∈ M∞ (B0 ) . The restriction of ψ to qn B1 qn defines a
*-isomorphism from qn B1 qn onto Mn (B0 ). It follows that ψ is a *-isomorphism
from B1 onto B0 ⊗ K ⊆ M∞ (B0 ) because limn→∞ kqn bqn − bk = 0 for all b ∈ B1 .
Let ρ : B → C a positive state with ρ(f ∗ ef ) = 0 . Since b 7→ ρ(f ∗ bf ) is a
positive functional on B and e is strictly positive, we get ρ(f ∗ Bf ) = {0} and
ρ(B0 ) = {0}, i.e., f ∗ ef is a strictly positive element of B0 . In the same way one
gets that g ∗ eg is strictly positive in B1 . 

2. Some properties of corona C *-algebras

HERE STARTS DISCUSSION ON σ-sub-Stonean C *-algebras:

Definition 5.2.1. A C *-algebra E is sub-Stonean (also called SAW*-

algebra) if, for each positive contractions a, b ∈ E+ with ab = 0, there exists
a positive contraction c ∈ E+ with ca = a and bc = 0 ( 6 ).

???? Wrong Definition ?:

The C *-algebra E is a σ-sub-Stonean C *-algebra if A0 ∩ E is sub-Stonean

for every separable C *-subalgebra A of E.
????
My comment:
This (!) is an other property than given in my original definition in [448, def. 1.4].
This original definition seems to be at least formally “stronger”? It is unclear
if they are equivalent.

6 In particular, for each e ∈ E there exist a positive contraction f ∈ E with f e = e = ef .

+ +
 2. SOME PROPERTIES OF CORONA C *-ALGEBRAS 665

“Recall” that a C *-algebra C is an SAW*-algebra if any two σ-unital C *-

subalgebras A and B of C are “orthogonal” (i.e., A · B = {0}) if and only if
they are “separated” (i.e., there exists positive contractions e, f ∈ C with ef = 0,
A = eAe and B = f Bf . ⇐ is perhaps not the original definition of SAW*-algebras?
This applies to E := C ∗ (e) and F := C ∗ (f ). It causes that there exists contrac-
tions g, h ∈ C+ with gh = 0, F = gF g and G = hGh. It implies that ρϕ · ϕ(f ) ≤ g
for each ϕ ∈ C0 (0, 1]+ with suitable ρϕ ∈ (0, 1]. Does it imply that there exists a
positive contraction T ∈ gCg with T f = f = f T ?
What is it the correct Definition of sub-Stonean C *-algebras? (Look to Peder-
sen and others?)
???? Definition ???? A C *-algebra C is an SAW*-algebra if any two σ-unital
subalgebras A and B of C are “orthogonal”, if and only if, they are “separated”.
Original ?: positive f, g?: [292, def. 2.9]
My Definition [448, def. 1.4] in “green”:
We call a C *- algebra C σ-sub-Stonean if for every separable C *-subalgebra
A ⊆ C and every b, c ∈ C+ with bc = 0 and bAc = {0} there are positive contrac-
tions f, g ∈ A0 ∩ C with f g = 0, f b = b and gc = c .
Definition [292, def. 2.9]:
A C *-algebra C is σ-sub-Stonean if for every separable subalgebra A of C and
all positive b and c in C such that bAc = {0} there are contractions f and g in
A0 ∩ C such that f g = 0, f b = b and gc = c .
Gives f b = b and f c = 0, f ∗ f g = 0, f ∗ f c = 0 and f Pb = Pb .
Missing:
bc = 0 (becomes stronger, because of weaker assumptions)
and
Positivity of f, g (becomes weaker conclusion).
Therefore difficult to compare!
Is it really equivalent to my Def?
(Implies: f c = 0, f = v(f ∗ f )1/2 , (f ∗ f )1/2 g = 0, v(f ∗ f )1/2 Pb = Pb ???? . f ∗ f
has properties: f ∗ f g = 0, f Pb = Pb . Is g ∗ c = c ? gPc = Pc with Pc support of c.
Can here replace A by C ∗ (A, b) then it covers partly our Definition ... but I
can not see that one can deduce from [292, def. 2.9] that the commutant A0 ∩ C is
again σ-sub-Stonean as it is in our Definition ...
The definition [292, def. 2.9] is not identical to my Definition:
A C *-algebra C is σ-sub-Stonean if, for every separable subalgebra A of C and
all positive b and c in C such that bAc = {0}, there are contractions f and g in
A0 ∩ C such that f g = 0, f b = b and gc = c.
Definitions [292, def. 2.5] of Farah and Hart:
666 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

Two subalgebras A, B of an algebra C are orthogonal if ab = 0 for all a ∈ A

and b ∈ B.
They are separated if there is a positive element c ∈ C such that cac = a for
all a ∈ A and cb = 0 for all b ∈ B.
My definition [448, def. 1.4] is stated here again ( 7 ):

Definition 5.2.2. We call a C *-algebra C σ-sub-Stonean if for every sepa-

rable C *-subalgebra A ⊂ C and every b, c ∈ C+ with bc = 0 and bAc = {0} there
are positive contractions f, g ∈ A0 ∩ C with f g = 0, f b = b and gc = c.

My comments:
Obviously, if C is σ-sub-Stonean, then C is sub-Stonean (case A := {0}),
and B 0 ∩ C is σ-sub-Stonean for every separable C *-subalgebra B of C (consider
C ∗ (B, A) in place of A in the definition).
It is easy to see, that if D is a hereditary C *-subalgebra of C, then D is σ-sub-
Stonean, if and only if, for every d ∈ D+ there is a positive contraction e ∈ D with
ed = d.
Proof of last:
There is a positive contraction e in D with eb = b, ec = c and ea = a for all a ∈ A.
Then bC ∗ (A, e)c = {0}. If f 0 , g 0 are positive contractions in C that commute with
C ∗ (A, e) and satisfy f 0 g − 0, f b = b and g 0 c = c, then f := f 0 e and g = g 0 e in I+
are as desired.
In particular, Ann(d, C) is σ-sub-Stonean for every d ∈ C+ if C is σ-sub-
Stonean.
Further, if C is σ-sub-Stonean and I / C is a σ-sub-Stonean closed ideal of C,
then C/I is σ-sub-Stonean. (An exercise.)
Proof of last:
Let A ⊂ C/I separable, b, c ∈ (C/I)+ with bc = 0 and bAc = {0}, A1 ⊂ C separable
with πI (A1 ) = A, f ∈ C a self-adjoint contraction with f = b1/2 − c1/2 , d a strictly
positive element of C ∗ (f+ A1 f− ) ⊂ I. Let e ∈ I a positive contraction with ed = d,
and let b0 := f+ (1 − e)f+ , c0 := f− (1 − e)f− ∈ C+ . Then b0 A1 c0 = {0}, b0 c0 = 0
and πI (b0 ) = b, πI (c0 ) = c. Since C is σ-sub-Stonean there are f 0 , g 0 ∈ A01 ∩ C with
f 0 b0 − b0 and g 0 c0 = c0 . f := πI (f 0 ) and g := πI (g 0 ) are as desired.
If C is sub-Stonean, then C satisfies almost a sort σ-variant for this for given:
There are always contractions g, h ∈ C+ with gh = 0 that are “quasi-invariant”
under the unitaries in A + C · 1 ⊆ M(C), – in the sense that e.g. u∗ gu ∈ gCg for
all u ∈ U(A + C · 1) and satisfy b ∈ gCg and c ∈ hCh.
Definition of Farah and Hart ([292, def. 2.9]):

7 It was modified or wrongly displayed in [292, def. 2.9] to something that ?????? seems to

be even not equivalent to [448, def. 1.4].

 2. SOME PROPERTIES OF CORONA C *-ALGEBRAS 667

A C *-algebra C is σ-sub-Stonean if for every separable subalgebra A of C and

all positive b and c in C such that bAc = {0} there are contractions f and g in
A0 ∩ C such that f g = 0, f b = b and gc = c .
My question: Each SAW*-algebra (Def.???) has this property?
Let B the separable C *-algebra generated by AcA and e ∈ B+ a strictly positive
contraction in B. Then be = 0 and there are positive contractions f, g ∈ C+ with
f g = 0, bf = b and ge = e . It follows gbc = bc for all b ∈ B

Definition 5.2.3. The algebra E is σ2 -sub-Stonean if for every separable

C *-subalgebra A of E, every positive contraction e ∈ A0 ∩ E and every sequence
g1 , g2 , . . . of positive contractions with egn = 0, gn+1 gn = gn and limn kgn a−agn k =
0 for all a ∈ A+ there exists a positive contraction f ∈ A0 ∩ E with f gn = gn for
all n ∈ N and f e = 0.
Next blue argument not clear !
P −n
( If one can take a subsequence hk = gnk and then takes g := 2 gn or G :=
P −nk
k2 (hk+1 − hk ), then ge = 0 and Ge = 0, the proposed positive contraction
0
f ∈ A ∩ E satisfies f g = g – which is equivalent to f gn = gn for each n = 1, 2, . . .
–, and each a ∈ A derives D = gEg into D :

ageg − gega = (ag − ga)eg + (gaeg − geag) + (ge(ag − ga)) ∈ D

P −n
for each a ∈ A and e ∈ E, because for all a ∈ A ga = n2 gn a ∈ D by
limk k(gn a − agn )gk k .)
A C *-algebra C has AA-CRISP (asymptotically abelian, countable
Riesz separation property) if the following holds:
Assume an , bn , for n ∈ N, are positive elements of C such that

an ≤ an+1 ≤ bn+1 ≤ bn

for all n. Furthermore assume that D is a separable subset of C such that for every
d ∈ D we have
lim k[an , d]k = 0 .
n

Then there exists a positive c ∈ C such that an ≤ c ≤ bn for all n and [c, d] = 0 for
all d ∈ D.
The algebra E satisfies Kasparov’s technical property (KTP) if E has
the following property:
If B1 and B2 are σ-unital C*-subalgebras of E with B1 · B2 = {0} and ∆ ⊂ E
is a separable subspace with xb − bx ∈ B1 for all x ∈ ∆ and b ∈ B1 , then there
exists a positive contraction f ∈ E+ such that f · B1 = {0}, (1 − f ) · B2 = {0} and
f x = xf for all x ∈ ∆.
We say that a C *-algebra M has KTP if the following holds:
Assume A, B, and C are subalgebras of M such that A⊥B and C derives B.
Furthermore assume A and B are σ-unital and C is separable. Then there is a
668 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

positive element d ∈ M such that d ∈ C 0 ∩ M , the map x 7→ xd is the identity on

B, and the map x 7→ dx annihilates A.
Every derivation of a separable subalgebra of M is of the form δb for some
b ∈ M.
(Latter is Part of Definition? Is it ∂b ?)

Kasparov’s Technical Theorem: [73, thm.12.4.2]

Let J be a σ-unital C *-algebra. Let A1 and A2 be σ-unital C *-subalgebras of
M(J), and ∆ a separable subspace of M(J). Suppose A1 · A2 ⊆ J, and that ∆
derives A1 . Then there are M, N ∈ M(J) such that

0 ≤ M ≤ 1, N = 1 − M, M · A1 ⊆ J, N · A2 ⊆ J, and [M, ∆] ⊆ J .

The result can fail if A1 and A2 are not σ-unital, even if B1 ? and B2 ? are
commutative and ∆ = 0 [Choi and Christensen 1983]:
Kasparov’s Technical Theorem [73, thm.12.4.2] says – in an equivalent formu-
lation – that E := Q(A) := M(A)/A has “Kasparov’s Technical Property” if A is
σ-unital.
If B1 and B2 are orthogonal σ-unital C *-subalgebras of E with B1 · B2 = {0}
and ∆ a separable subspace with [x, b] ∈ B2 for all b ∈ B2 and x ∈ ∆. Then we can
find strictly positive contractions e1 ∈ B1 and e2 ∈ B2 and we get [x, b] ∈ e1 Ee1
for all x ∈ ∆ and b ∈ B1 .
The algebras B1 and B2 can not be completely re-discovered from (e1 , e2 , ∆).
But if we find a positive contraction T ∈ E+ with T e2 = e2 and T e1 = 0, and
[T, d] = 0 for all d ∈ ∆ then f := T satisfies Kasparov’s technical property for
(B1 , B2 , ∆).
We can conversely consider two positive e1 , e2 ∈ E+ with e1 e2 = 0 and a
countable subset X of E with the property that [x, e1 ] ∈ e1 Ee1 for all x ∈ X. (E.g.
by taking strictly positive contractions e1 ∈ A1 and e2 ∈ A2 ).
If we find a contraction T ∈ E+ with T x = xT for all x ∈ X, T e2 = e2 and
T e1 = 0. Then T has the required properties.
In fact it suffices to find T ∈ C ∗ (X)0 ∩ E, kT k ≤ 1, T ≥ 0 and T e1 = 0,
T e2 = e2 .
If e1 , e2 ∈ E+ with e1 e2 = 0 and a countable subset X of E with the property
that [x, e1 ] ∈ e1 Ee1 for all x ∈ X are given, then one can consider the C *-algebras
B2 := C ∗ (e2 ) and define ∆ as the linear span of X. A C *-algebra B1 can be defined
as closure of n B1,n , where the B1,n are inductively defined by B1,1 := C ∗ (e1 ),
S

B1,n+1 := C ∗ (B1,n , [y, x], x ∈ X, y ∈???).

Then e1 is a strictly positive element of B1 and [B1 , y] ⊆ B1 for all y in the
linear span of X.
 2. SOME PROPERTIES OF CORONA C *-ALGEBRAS 669

Assuming that E satisfies the Kasparov technical property, we get a contraction

f ∈ E+ with f · B1 = {0}, f ∈ C ∗ (X)0 ∩ E and f e1 = e1 .
A C *-algebra C has AA-CRISP (asymptotically abelian, countable Riesz sep-
aration property) if the following holds:
Assume an , bn , for n ∈ N, are positive elements of C such that an ≤ an+1 ≤
bn+1 ≤ bn for all n ∈ N. Furthermore assume D is a separable subset of C such
that for every d ∈ D we have
lim k[an , d]k = 0 .
n

Then there exists a positive c ∈ C such that an ≤ c ≤ bn for all n and [c, d] = 0 for
all d ∈ D.

So far we have not seen examples that separate the stronger definitions from the
sub-Stonean C *-algebras. For abelian C *-algebras the definitions are equivalent.
(But compare the study [292] of I. Farah and B. Hart in the case of corona
algebras.:
“ We present unified proofs of several properties of the corona of σ-unital C *-
algebras such as AA − CRISP , SAW ∗ , being sub-σ-Stonean in the sense of Kirch-
berg, and the conclusion of Kasparov’s Technical Theorem.
Although our results were obtained by considering C *-algebras as models of
the logic for metric structures, the reader is not required to have any knowledge of
model theory of metric structures (or model theory, or logic in general). The proofs
involve analysis of the extent of model-theoretic saturation of corona algebras.
Countable saturation of corona algebras Ilijas Farah, Bradd Hart,
arXiv:1112.3898v2 (revised 17 Oct 2012)
Related properties of coronas of σ-unital stable C *-algebras A is given in the
following proposition.

Proposition 5.2.4. Let A a stable σ-unital C*-algebra, D ⊂ A a σ-unital

hereditary C*-subalgebra.

(i) D is unitary homotopic to a corner of A.

(ii) The corona Q(A) := M(A)/A ∼ = Qs (A) is is a σ2 -sub-Stonean C*-algebra
(in sense of Definition 5.2.3).

Proof. (i): To be filled in ?? (ii): It is essentially Kasparov’s lemma. 

Question 5.2.5. Let B a sub-Stonean C *-algebra (in the sense that for each
self-adjoint contraction c ∈ B there exists a self-adjoint contraction d ∈ B such that
dc = |c| (equivalently expressed: for c1 , c2 ∈ B+ with c1 c2 = 0 exist contractions
d1 , d2 ∈ B+ with d1 d2 = 0 and dk ck = ck for k ∈ {1, 2}.)
Let e ∈ B+ a positive contraction, D := eBe and define the (two-sided) nor-
malizer of D in B by
N (D) := {b ∈ B ; bD ∪ Db ⊆ D} .
670 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

It is easy to see that N (D) is a C *-subalgebra of B and that ψD (c)d := cd (for

d ∈ D and c ∈ N (D)) defines a natural C *-morphism ψD : N (D) → M(D) from
N (D) into the (abstract two-sided) multiplier algebra M(D) of D. It is unital,
because there exists a contraction f ∈ B+ with f e = e.
Is ψD (N (D)) sub-Stonean?
Clearly, the kernel of ψD is equal to the two-sided annihilator

Ann(D) := Ann(B, D) := {b ∈ B ; bD = Db = {0} }

of D in B, which is a sub-Stonean hereditary C *-subalgebra of B.

The interesting question says:
Is ψD : N (D) → M(D) surjective?
I.e., M(D)/D ∼
= N (D)/ Ann(D) in a natural way.
It would give a nice new characterization of sub-Stonean C *-algebras.

Question 5.2.6. Let B a sub-Stonean C *-algebra. A special case of Question

5.2.5 is the following question:
Let b1 , b2 , . . . ∈ B+ a sequence of mutually orthogonal contractions.
Does B+ contain an contraction c ∈ B+ such that c is contained in the in N (D)
P −n
for D := dBd with d := n2 bn and c(bn − (bn − 1/2)+ ) = (bn − 1/2)+ for all
n ∈ N?

It seems (!) that Question 5.2.5 is equivalent to the following question:

Question 5.2.7. Let B a sub-Stonean C *-algebra and a ∈ B+ . Let D := aBa.

Suppose that d1 , d2 , . . . ∈ D+ are mutually orthogonal and such that d1 + d2 + · · ·
converges strictly to an element in T ∈ M(D)+ .
Something more elaborate is needed ?
Does there exist an element b ∈ B+ that satisfies bD ∪ Db ⊆ D and
X
lim kb(a − (1 − 1/n))+ a − dn (a − (1 − 1/n))+ ak = 0 ?
n
k≤n

3. Functions that respect quasi-central approximate units

If we want to verify in applications the estimates that appear in the assumptions

of the Parts (2)-(6) of Remark 5.1.1, then the below given Proposition 5.3.1 and its
consequence, the Lemma 5.3.2, are useful, despite the fact that also the arguments
in the proof of [43, thm. 2] can be used without controlling bound. This argument
of Arveson in [43] is in essence the following:
Let K a convex subset of the positive contractions in A that contains an ap-
proximate unit of A, and ϕ1 , ϕ2 , . . . ∈ C0 (0, 1]+ a given sequence of continuous
function.
 3. FUNCTIONS THAT RESPECT QUASI-CENTRAL APPROXIMATE UNITS 671

We need an element e ∈ K with ky − eyk ≤ εkyk for y in a given finite-

dimensional subspace Y ⊆ A that satisfies at the same time the commutation
properties k[e, x]k ≤ εkxk and k[ϕk (e), x]k ≤ εkxk for x ∈ X and k = 1, . . . , n,
where X ⊆ M(A) is a finite-dimensional linear subspace. This can be done by
solving following inequalities by an inductive selection procedure with nth step as
follows:
First choose for each given ε > 0, n0 and the considered continuous function
ϕk (λ) with ϕk (0) = 0, k ≤ n, a polynomial Pk (λ) with Pk (0) = 0 and |ϕk (λ) −
Pk (λ)| < ε/2 for λ ∈ [0, 1] and then select from before constructed approximate
units en with en em = emin m,n a finite convex combination e of elements en1 , . . . , enk
(n0 < n1 < n2 < . . . < nk ) such that k[e, x]k ≤ ηkxk for x in a given finite-
dimensional subspace Xk ⊆ M(A) in a (before chosen) filtration of A or of some
separable subspace, where η can be taken e.g. as η := ε/(2 max1≤k≤n {1 + Q0k (1)})
with polynomials Qk (λ) build from Pk (λ) by replacing all coefficients of Pk by its
absolute values and Q0k := d/dλ Qk is the derivative of Qk . Those estimates are
not perfect.
See [11, lem. 10.3] for the case of the polynomials P (λ) := λn .
Notice also that by [11, lem. 10.4], T ∗ = T , kT k < 1 and bounded X:

k(1 − T 2 )1/2 X − X(1 − T 2 )1/2 k ≤ (kT k · kXT − T Xk)/(1 − kT k2 )1/2 .

Unfortunately we have to consider the case where kT k is near to 1.

In our application we need a construction that produces a sufficiently fast
commutation property such that sub-selections satisfy automatically better com-
mutation and approximate-unit properties, i.e., we need to check estimates of
commutator-norms for commutators of elements in given different convex subsets
of an algebra. This must be used later to verify the quadratic unconditional strict
convergence of some series build by selections from given sequences of elements in
the multiplier algebra.
A control of such general estimate is possible by the above described inductive
selection. But we need a general control of the commutators for all needed functions
f itself, e.g. by suitable bounds for the – with respect to t ∈ [0, 1] non-decreasing –
function
G(f, ·) : [0, 1] 3 t 7→ G(f, t) ∈ [0, ∞] ,
defined by
G(f, t) := sup{ k[x, f (a)]k ; (a, x) ∈ P (t) }
for the set of pairs

P (t) := { (x, a) ; x, a ∈ L(`2 ), kxk ≤ 1, 0 ≤ a ≤ 1, kxa − axk ≤ t } .

Here f is a given non-decreasing continuous real function on [0, 1]. The defini-
tion shows that t 7→ G(f, t) is the minimal function that satisfies for given f the
inequalities
k[x, f (a)]k ≤ G(f, k[x, a]k)
672 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

for all contractions x, a ∈ L(H) with a ≥ 0. Below we observe that it suffices to

consider only self-adjoint contractions x = x∗ to define G(f, t).
Where Gu (f, t) is used?
Where Gu (f, t) is used? To get applicable estimates of G(f, t) it could be
useful to consider also the functions Gu (f, t) defined by

Gu (f, t) := sup{ ku∗ f (a)u − f (a)k ; (a, u) ∈ P U (t) }

where P U (t) is defined as the set of pairs

P U (t) := { (u, a) ; u, a ∈ L(H) with u∗ u = 1 = uu∗ , 0 ≤ a ≤ 1, k[a, u]k ≤ t } ,

i.e., it considers only differences kf (b)−f (a)k of functions of unitary transformations

b := u∗ au. Those have often easier estimates if the function f is operator monotone,
e.g. for f (t) = t1/2 it is easy to see that kf (b) − f (a)k ≤ f (kb − ak).
Always k[b, x]k ≤ k[b− , x]k + k[b+ , x]k ≤ 2kxk(kb− k + kb+ k) for all x ∈ L(H)
Is always k[b, x]k ≤ kxkkbk and each positive b ∈ L(H)?
It implies that for b ∈ L(H)+ and x ∈ L(H) holds that

k[b, x]k ≤ inf k[(b − t)− , x]k + k[(b − t)+ , x]k .

t∈R+

Is k[b, x]k ≤ kxkkbk for b ∈ L(H)+ and any x ∈ L(H) by the following estimates
??:
For each δ > 0 there are vectors θ, ξ ∈ H with norm = 1 such that h[b, x]θ, ξi ≥ 0
and k[b, x]k ≤ δ + h[b, x]θ, ξi . Notice that h[b, x]θ, ξi = hbθ, x∗ ξi − hxθ, bξi ≥ 0
It is a non-negative difference of two complex numbers of norm ≤ kxkkbk.
It follows that it is also a non-negative difference of two real numbers of norm
≤ kxkkbk. (Still it could be = 2kxkkbk .)
If P is the orthogonal projection onto the linear span of {θ, ξ, xθ, x∗ ξ}
then h[b, x]θ, ξi = hbP θ, P 2 x∗ P ξi − hP 2 xP θ, bP ξi ≥ 0. The latter is equal to
h[P bP, P xP ]θ, ξi ≤ k[P bP, P xP ]k.
Thus k[b, x]k ≤ δ + k[P bP, P xP ]k and we can reduce all to the case of H = C4 .
Can suppose that 0 ∈ Spec(P bP ).
Let P ∈ L(H) an orthogonal projection that commutes with b and satisfies
P b = b+ and (1 − P )b = b− . Then [b, x] = bP x − xP b + b(1 − P )x − x(1 − P )b ...
??? says b+ x − xb+ + b− x − xb− = [b, x]
and then ????? used where ???
check above and next again:
This ????????? can be seen by restriction to the case of self-adjoint contractions
X = X and B ≥ 0, if we replace b by B := diag(γ + b, γ + b) ∈ M2 (L(H)) ∼
∗
= L(H)
with γ := kb− k and replace x by X := diag(x , x) · Z ∈ M2 (L(H)), where Z ∗ =
∗

Z := [ζjk ] ∈ M2 is defined by ζjk := 1 − δjk and satisfies Z 2 = 1, Z diag(y, z) =

 3. FUNCTIONS THAT RESPECT QUASI-CENTRAL APPROXIMATE UNITS 673

diag(z, y)Z. Thus, ZBZ = B, ZX = diag(x, x∗ ), XZ = diag(x∗ , x). and X ∗ =

Z · diag(x, x∗ ) = X.
It follows that G(f, ·) restrict to the case of self-adjoint contractions X = X ∗ ,
because Then k[x, b]k = k[X, B]k and k[x, f (b)]k = k[X, f (B)]k.
B diag(x∗ , x) = diag((γ + b)x∗ , (γ + b)x),
ZBX = diag((γ + b)x, (γ + b)x∗ ),
ZB = Z diag((γ + b), (γ + b)) = diag((γ + b), (γ + b))Z = BZ,
Z(XB − BX) = diag(x, x∗ ) · diag(γ + b, γ + b) − diag((γ + b)x, (γ + b)x∗ ) is
the diagonal matrix with entries x(γ + b) − (γ + b)x and x∗ (γ + b) − (γ + b)x∗ .
Thus kXB − BXk = kx(γ + b) − (γ + b)xk = k[b, x]k if b∗ = b, γ := kb− k and
X := diag(x∗ , x) · Z = X ∗ .
Since one can X replace here again by X + kX− k ≥ 0 without changing the
norms, it shows that all estimates can be done using positive operators in L(`2 ).
In particular it suffices to consider self-adjoint contractions x and positive a if
we want to find an estimate for G(f, t).
Check next arguments again
Always G(f, 0) = 0, because f (a) commutes with x if a = a∗ commutes
with x and f (t) is a function on the spectrum of a. Moreover, if t1 > t2 > · · ·
we find (an , xn ) with k[xn , an ]k ≤ tn and G(f, tn ) − 2−n < k[xn , f (an )]k ≤
G(f, tn ). The elements a := (a1 , a2 , . . .) + c0 (L(H)) and x := (x1 , x2 , . . .) +
c0 (L(H)) of `∞ (L(H))/c0 (L(H)) generate a C *-subalgebra C ∗ (a, x) that can be
faithfully represented as a C *-subalgebra of L(H), then k[x, a]k ≤ t∞ := limn tn
and k[x, f (a)]k = lim supn→∞ G(f, tn ) = limn G(f, tn ). Since t 7→ G(f, t) is increas-
ing, and k[x, f (a)]k ≤ G(f, t∞ ) ≤ G(f, tn ), it follows that the monotone increasing
function t 7→ G(f, t) is continuous from the right side.
(i.e., is upper semi-continuous ?) What about the left side?
If f is continuous, then t 7→ G(f, t) is continuous, i.e., is also continuous from
the left side: Let t1 < t2 < . . . in [0, 1] and t∞ := limn tn . Suppose G(f, t∞ ) >
supn G(f, tn ) and let 0 < ε < G(f, t∞ ) − supn G(f, tn ).
By definition of G(f, t∞ ), there exists (a, x) ∈ P (t∞ ) with G(f, t∞ ) − ε <
k[x, f (a)]k ≤ G(f, t∞ ). If f is continuous on [0, 1], then limδ→0 kf ((1 − 2δ)(a +
δ1)) − f (a)k = 0.
Since [x, (1 − 2δ)(a + δ1)] = (1 − 2δ)[x, a], and G(f, (1 − 2δ)t∞ ) ≤ supn G(f, tn )
and

k[x, f ((1 − 2δ)(a + δ1))]k ≤ sup G(f, tn ) ≤ G(f, t∞ ) − ε ≤ G(f, t∞ ) .

n

If we suppose that f is continuous, then kf ((1−2δ)(a+δ1))−f (a)k → 0 for 0 < δ →

0. Thus, supn G(f, tn ) = G(f, t∞ ). This implies that t ∈ [0, 1] 7→ G(f, t) ∈ [0, 1] is
continuous on [0, 1]. The continuity implies – by the Lemma of Dini on automatic
674 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

uniform convergence from below applied to G(f, ·) on [0, 1] – that the value G(f, t)
is determined by the pairs (a, x) in P (t) with a and b operators of finite rank.
Some words about lower bounds for G(f, ·):
It is easy to check that f (t) ≤ G(f, t), e.g. a(t) := ts2 s∗2 + s3 s∗3 and x :=
s1 s∗2 of O∞ = C ∗ (s1 , s2 , . . .) ⊆ L(`2 ) satisfy kxk = 1, ka(t)k = 1, a(t) ≥ 0 and
k[x, f (a(t))]k = f (t), k[x, a(t)]k = t. This shows that f (t) ≤ G(f, t) for all t ∈ [0, 1].
Alternatively one can take elements a := diag(0, t, 1) ∈ M3 and x := [ξjk ] ∈ M3
with ξ13 := 1 and ξjk = 0 for (j, k) 6= (1, 3).
If we allow in case of M2 also the case with kak < 1, then a(t) := diag(t, 0)
and the above defined self-adjoint orthogonal matrix Z ∈ M2 again satisfy
k[Z, f (a(t))]k = f (t) and k[Z, a(t)]k = t for t ∈ [0, 1].
The above reduction to the case of self-adjoint x = x∗ and b ≥ 0 shows also
that kxb − bxk ≤ kxk(kb+ k + kb− k) for each b∗ = b = b+ − b− , and any x ∈ L(`2 ),
because for x∗ = x and b∗ = b the element T := i(bx − xb) is self-adjoint and its
norm is given by the supremum of |hT v, vi| = |hxv, bvi − hbv, xvi| for v ∈ H := `2
with kvk = 1. To see that |hT v, vi| ≤ kxkkbk, consider the orthogonal projection
P ∈ K of rank ≤ 2 from H onto the linear span of v and xv. Then

|hT v, vi| = |hP xP v, bP vi − hbP v, P xP vi| = |h(P bP xP − P xP bP )v, vi| .

This reduces the claim to the case where b and x are in M2 : If rank(P ) = 1 we get
hT v, vi = 0, and otherwise P bP is positive and P xP is self-adjoint in p(K)p ∼
= M2 .
It is easy to see that kbx − xbk ≤ 1 for contractions b, x ∈ M2 with b ≥ 0, because
one can reduce the general case to the case where b is diagonal and x has zero
diagonal.
If we allow both of b and x to be non-positive but self-adjoint contractions,
then x := Z with Z ∈ M2 as above defined and b := diag(1, −1) are self-adjoint
contractions that satisfy k[x, b] k = kxk(kb− k + kb+ k) = 2.
Since one can for positive a ∈ M2 and f ≥ 0 reduce the considerations to the
case of diagonal a and x = diag(α, β) · Z, one gets that k[x, f (a)]k ≤ f (k[x, a]k) for
all contractions x, a ∈ M2 (C) with a ≥ 0 and any continuous non-decreasing f on
[0, 1] with f (0) ≥ 0.
Def. of P and ?
k[x, f (a)]k = ihxv, f (a)vi = ihP xP v, f (a)P vi ≥ k[P xP, P f (a)P ]k
for P = P ∗ P , P v = v, P xv = xv ???
and k[P xP, P f (a)P ]k ≤??? if P f (a)P ≥ f (P aP )?
Some determination of G(f, t) should be possible in other cases because one
can restrict the definition to a, x ∈ Mn (C) and get Gn (f, t) ≤ G(f, t) and G(f, t) =
supn Gn (f, t) with uniform convergence by the Lemma of Dini. This reduction to
Mn can be seen as follows:
The positive contractions in c0 ⊆ K(`2 ) contain a approximate unit (en ) of K
that is (in norm) quasi-central for L(`2 ), i.e., for each separable subset of X ⊂
 3. FUNCTIONS THAT RESPECT QUASI-CENTRAL APPROXIMATE UNITS 675

L(`2 ) there exists a suitable sub-sequence that approximately commutes with the
elements of X and has the following properties: If 0 ≤ en ≤ 1 satisfy en em = em
for 1 ≤ m < n (that implies that each en is automatically of finite rank) and
en → 1 strictly in M(K) ∼ = L(`2 ) and k[en , a]k + k[en , f (a)]k → 0 for n → ∞, then
k[a, en xen ]k → k[a, x]k and k[f (a), en xen ]k → k[f (a), x]k for n → ∞.
Thus, it suffices to consider self-adjoint contractions x = x∗ ∈ K of fi-
nite rank to determine the value of F (f, t). After this reduction we can
modify the approximate unit (en ) such that in addition en x = x = xen
for all n, and get [en aen , x] = en [a, x]en and limn kf (en aen ) − f (ae2n )k = 0,
limn kf (en aen ) − f (e2n a)k = 0, limn kf (en aen ) − f (ae2n )k = 0, limn kf (ae2n )x −
f (a)xk = 0 and limn kxf (ae2n ) − xf (a)k = 0, but [f (en aen ), x] = [e2n f (a)e2n , x]
and ke2n a2 e2n − (en aen )2 k → 0 for n → ∞. We obtain finally the convergences
k[f (en aen ), x]k → k[f (a), x]k and k[en aen , x]k → k[a, x]k. Together with the
continuity of t 7→ F (f, t) this shows that F (f, t) = limn Fn (f, t).
T.A. Loring and F. Vides study in [541] among others the special case of
f (t) = t1/2 and get good reason for the conjecture that G(f, t) = t1/2 for f (t) = t1/2 ,
that was also suggested by G.K. Pedersen in [619]. Related results up to a (there
not explicitly given) constant can be obtained from results of A.B. Aleksandrov and
V.V. Peller, e.g. in [11, 12]. An estimate by Pedersen was given in [619, cor. 6.3]:
If 0 ≤ a ≤ 1 and kbk = 1, then k[a, b]k ≤ ε ≤ 1/4 implies k[a1/2 , b]k ≤ (1.25)ε1/2 .
We give explicit estimates in the following Proposition 5.3.1, that are likely not
the best possible, but its proof can be obtained easily by elementary observations.

Proposition 5.3.1. Let H a Hilbert space, and suppose that ϕ(t) a continuous
function on [0, ∞) that is operator-monotone in (0, ∞) and satisfies ϕ(0) ≥ 0.
For all positive a, b ∈ L(H),

kϕ(b) − ϕ(a)k ≤ ϕ(kb − ak) .

If u ∈ L(H) is unitary and a ≥ 0, then

k [u, ϕ(a)] k ≤ ϕ( k[u, a]k ) . (3.1)

If a, x ∈ L(H) are contractions with a ≥ 0, then

k [x, ϕ(a)] k ≤ 3ϕ( k [x, a] k ) . (3.2)

In particular, for all a ≥ 0, every contraction x and each β ∈ [1, ∞),

k[x, a1/β ]k ≤ 3k[x, a]k1/β .

Proof. It suffices to consider aε := (1 − 2ε)(ε + a) in place of a, because

xaε − aε x =: [x, aε ] = (1 − 2ε)[x, a] ,

and ϕ(k[x, a]k) = limε→0 ϕ(k[x, aε ]k) by continuity of ϕ on [0, ∞), and it implies
[x, ϕ(a)] = limε→0 [x, ϕ(aε )] by uniform continuity of ϕ on bounded subsets of
[0, ∞).
676 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

Notice that Spec(aε ) ⊆ [(1 − 2ε)ε, ∞) ⊂ (0, ∞) . Therefore we can use the
variant of the Löwner theorem given by F. Hansen [356, thm. 4.9, rem. 5.3]:
For every positive operator monotone function ϕ, defined in the positive half-
line (0, ∞), there is a unique bounded positive measure µ on the closed interval
[0, 1] such that for t ∈ (0, ∞), λ ∈ [0, 1] and the functions

−1

fλ (t) := (λ + (1 − λ)t)−1 t = 1 − 1 + ((1 − λ)/λ)t


/ 1−λ
holds Z 1
ϕ(t) = fλ (t) dµ(λ) for t > 0 .
0
Any function given by such a measure is operator monotone, and the measure µ
is a probability measure if and only if ϕ(1) = 1. The functions fλ : t 7→ fλ (t) are
easily seen operator monotone, in particular the fλ are monotone.
Let η > 0 and a, b ∈ L(`2 )+ such that η ≤ a and η ≤ b and let γ := kb − ak .
Since fλ is operator monotone and b ≤ a + γ, we get fλ (b) ≤ fλ (a + γ) . Moreover
fλ has the property fλ (t + γ) ≤ fλ (t) + fλ (γ) for γ > 0, because the function
x → x/(1 + x) is sub-additive for x ∈ [0, ∞). Thus we get
fλ (b) ≤ fλ (a + γ) ≤ fλ (a) + fλ (γ) .
If we interchange here a and b it implies together that kfλ (b) − fλ (a)k ≤ fλ (γ) . If
we use the above integral representations of ϕ(a) and ϕ(b), we obtain the estimate
Z 1
kϕ(b) − ϕ(a)k ≤ fλ (γ) dµ(λ) = ϕ(γ) = ϕ(kb − ak) .
0
To get that kϕ(b) − ϕ(a)k ≤ ϕ(kb − ak) for arbitrary positive a, b ∈ L(`2 ), we can
consider δ + a and δ + b for δ > 0 and then use that ϕ is uniformly continuous on
bounded parts of [0, ∞), e.g. that
lim kϕ(a) − ϕ(δ + a)k = 0 .
0<δ→0

If u is unitary and a ≥ 0, then uϕ(a)u∗ = ϕ(uau∗ ) can be seen easily with help of
uniform approximation of ϕ by polynomials on the interval [0, kak]. For b := u∗ au
holds kua − auk = kb − ak and kuϕ(a) − ϕ(a)uk = kϕ(b) − ϕ(a)k . This implies
inequality (3.1).
To find an estimate of k[x, ϕ(a)]k for a contraction x and a ≥ 0 by k[x, a]k,
we can replace a by the positive element A := diag(a, a) ∈ M2 (L(`2 )) and x
by the self-adjoint contraction X := diag(x, x∗ ) · Z ∈ M2 (L(`2 )), because then
k[X, A]k = k[x, a]k and k[X, ϕ(A)]k = k[x, ϕ(a)]k ( 8 ).
Let x∗ = x, b ∈ L(`2 ) with kxk < 1 and b ≥ 0. The Cayley transformation
of x ∈ L(`2 ) with kxk < 1 and x∗ x = xx∗ is defined as the continuous operator-
valued function
u(λ) := (λ + x)(1 + λx∗ )−1
for λ ∈ C with |λ| ≤ 1, cf. [616, p. 4]. Then u(λ) is analytic for |λ| < 1, u(0) = x,
u(eis ) is unitary for s ∈ [0, 2π].
8 Here Z ∈ M has entries z
2 j,k := 1 − δj,k .
 3. FUNCTIONS THAT RESPECT QUASI-CENTRAL APPROXIMATE UNITS 677

Is this next blue general study useful here?

The general Cayley transformation is given for all x ∈ L(`2 ) with kxk < 1 and
λ ∈ C with |λ| ≤ 1 by

u(λ) := (1 − xx∗ )−1/2 (λ1 + x)(1 + λx∗ )−1 (1 − x∗ x)1/2 .

Then u(λ) is unitary for |λ| = 1 and x = u(0). Moreover ku(λ)k < 1 for all |λ| < 1
(the latter by maximum modulus principle). See [222] or [402, ex. 10.5.5].
As in the considerations below, we get for v1 , v2 ∈ H := `2 and ρ(y) := hyv1 , v2 i
that Z 2π
ρ([x, b]) = (2π)−1 ρ([u(eis ), b]) ds .
0
??????
R 2π
Since 2πu(0) = 0
u(eis ) ds, we get
Z 2π
[x, b] = (2π)−1 [u(eis ), b] ds .
0
R 2π
It implies (2π) · k[x, b]k ≤ 0
k[u(e ), b]k ds . Hence, k[x, b]k ≤ sups k[u(eis ), b]k.
is

If we let here b := ϕ(a) then inequality (3.1) implies that

k[x, ϕ(a)]k ≤ sup{ ϕ(k[u(eis ), a]k) ; s ∈ [0, 2π] } .

P∞
We let v(λ) := (λ + x) and w(λ) := (1 + λx∗ )−1 = n=0 (λx∗ )n . Then u(λ) =
v(λ)w(λ), and, for all b ∈ L(`2 ), [b, v(λ)] = [b, x] , [b, w(λ)] = −λw(λ)[b, x∗ ]w(λ)
and
[b, u(λ)] = [b, x]w(λ) + λu(λ)[b, x∗ ]w(λ) .
Since u(λ) is unitary if |λ| = 1 we get for x = x∗ the estimate

k[u(eis ), b]k ≤ 2k[x, b]kkw(eis )k .

The inequality keis x∗ k = kxk < 1 implies kw(eis )k ≤ (1 − kxk)−1 . Thus,

ϕ(k[u(eis ), a]k) ≤ ϕ((1 − kxk)−1 2k[x, a]k) ,

and for all x∗ = x with kxk < 1,

k[x, ϕ(a)]k ≤ ϕ((1 − kxk)−1 2k[x, a]k) .

Since the latter formula holds for all x = x∗ with norm kxk < 1, we can replace x
by 3−1 x and get by monotony of ϕ the estimate

k[x, ϕ(a)]k ≤ 3ϕ(k[x, a]k) .

Compare next blue with above given similar remarks!!

A particular case is the operator monotone function ϕ(t) := t1/2 , where the
estimate shows that k[x, a]k2 ≤ 9k[x, a2 ]k. It gives that the above defined and
discussed continuous function G(t) := G(ϕ, t) ≤ 3t1/2 for ϕ(t) := t1/2 – defined as
678 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

the “minimal” function with G(k[a, x]k) ≥ k[a1/2 , x]k for contractions x, a ∈ L(`2 )
with a ≥ 0 – satisfies
t1/2 ≤ G(t) ≤ 3t1/2 .
Numerical methods indicate (but do not prove) that in this special case it could
be possible to improve the constant 3 down to 1 by other methods, compare [541].
This is in accordance with a suggestion of G.K. Pedersen in [619].
The following lemma gives for explicit estimates of unconditional strict conver-
gence at least some useful upper bounds (that are perhaps not minimal).

Lemma 5.3.2. Let x, y ∈ A and a, b ∈ A+ contractions and γ > 0. Suppose

that

max( k[x, a2 ]k , k[x, b2 ]k ) ≤ γ and ky − b2 yk + ky ∗ − b2 y ∗ k ≤ γ .

Then we get the estimates

k[x, (1 − b)a]k ≤ 6γ 1/2 and ky − byk + ky ∗ − by ∗ k ≤ γ .

Proof. Let a, b ∈ A+ and x ∈ A contractions. We get from Proposition 5.3.1

that
max( k[x, a]k , k[x, b]k )2 ≤ 9 max(k[x, a2 ]k, k[x, b2 ]k) ≤ 9γ .
This implies by [(1 − b)a, x] = (1 − b)[a, x] − [b, x]a that

k[(1 − b)a, x]k ≤ k[a, x]k + k[b, x]k ≤ 3k[a2 , x]k1/2 + 3k[b2 , x]k1/2 ≤ 6γ 1/2 .

The estimates for y follows from e := (1 − b)2 ≤ (1 − b2 )2 =: f if 0 ≤ b ≤ 1,

because e.g. ky − byk2 = ky ∗ eyk ≤ ky ∗ f yk = ky − b2 yk2 . 

4. A “tautologic” Weyl–von Neumann type result

We prove a very general “tautologic” version of the classical Weyl–von Neu-

mann theorem: the below stated Proposition 5.4.1.
Applications will be obtained later by verifying in some special cases the va-
lidity of assumptions (α) and (β) given in Proposition 5.4.1. The assumptions are
sufficient but are not necessary for some derived properties e.g. as in Part (iii) of
Proposition 5.4.1. Applications consider cases where the C *-algebra C does not
contain an element h with the property (α) required in Proposition 5.4.1 or is not
unital. But often one can build a larger algebra C1 and a suitable extension T1
of T that contains an element h that satisfies the required Conditions (α) and (β)
with C1 and T1 in place of C and T . But this extension, e.g. to an outer uniti-
zation is not always possible and needs some care. Here is a warning concerning
application in the theory of extension groups Ext(C ; C, B) : The nontrivial part
in those applications – by trying to establish the assumptions of Proposition 5.4.1
– consists in showing that condition (β) still holds also for the extended T1 on the
bigger C *-algebra C1 .
 4. A “TAUTOLOGIC” WEYL–VON NEUMANN TYPE RESULT 679

Adjoining a unit to C is only possible if the considered c.p. map T on C satisfies

that πB ◦ T is “zero-absorbing”, i.e., if πB · T “dominates zero” in the more precise
sense, as discussed in Chapter 4.
Does the Elliott-Kucerovsky result follow?? – from the special case
where C = C ∗ (E, d), with d ∈ B strictly positive, and πB (E) separable, T (d) = 0
and [T ] : πB (C) → M(B) has the property that, for each a ∈ B+ , the map
e ∈ E 7→ a[T ](πB (e))a or the map f ∈ πB (E) 7→ a[T ](f )a is nuclear?
I.e.: When does it satisfy moreover the Condition (β)?? (That we can then
apply Prop. 5.4.1 to it.)

Proposition 5.4.1. Let B a σ-unital C*-algebra, C a separable C*-subalgebra

of M(B), and T : C → M(B) a linear map with kT k ≤ 1 that satisfies the following
Conditions (α) and (β):

(α) There exists h ∈ C+ with T (h) = 0 and h1/n d → d if n → ∞ for every

d∈B.
(β) For every a ∈ B+ , every finite subset X ⊂ C+ of contractions, and every
ε > 0 there exists d ∈ M(B) with kd∗ cd − aT (c)ak < ε for c ∈ X.

Then the contraction T : C → M(B) with this properties is completely positive and
has following properties:

(i) There exists a sequence Sn of contractions in M(B) such that, for all
c ∈ C and k ∈ N, and, for all m, n ∈ N,

Sn∗ cSm − δm,n T (c) ∈ B and lim k δk,0 T (c) − Sn∗ cSn+k k = 0 .
n→∞

(ii) If B is stable then the completely positive contraction V := δ∞ ◦T satisfies

again Conditions (α) and (β) (with V there in place of T ) and there is
a norm-continuous map t ∈ R+ 7→ S(t) ∈ M(B) into the contractions of
M(B) such that, for s, t ∈ R+ , k ∈ Z+ = N ∪ {0} and c ∈ C holds

S(s)∗ c S(t) − δs,t T (c) ∈ B and lim S(t + k)∗ c S(t) = δk,0 · T (c) . (4.1)
t→∞

(iii) If in addition to the properties in (ii ) T is a C*-morphism and T (C)B

is dense in B, then the monomorphism ηC : c ∈ C 3 c 7→ c ∈ M(B)
asymptotically absorbs T , i.e., the C*-morphisms (idC ⊕T ) : C → M(B)
and ηC are unitarily homotopic in the sense of Definition 5.0.1 modulo
B.
(iv) If B is stable, 1M(B) ∈ C and T (1) = 1, or if πB ◦ T “dominates zero”
in the sense of Definition 4.3.3 ( 9 ), then the path t → S(t) in Part (ii )
can be modified such that Properties (4.1 ) still hold, but that S(t) has the
additional property that πB (S(t)) is an isometry for all t ∈ R+ .
If the operators πB (S(t)) are isometries, πB ◦ T is a C*-morphism
and (πB ◦ T )(C)0 ∩ (M(B)/B) contains a copy C ∗ (s1 , s2 ) of O2 unitally,

9 Compare also Lemma 4.3.4(i,iv).

680 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

then there exists a unitary U ∈ M(B)/B with

U ∗ πB (c)U = πB (c) ⊕s1 ,s2 πB (T (c)) for all c ∈ C .

(v) The contractions Sn in Part (i ), respectively the S(t) in Part (ii ), can be
chosen as isometries if 1M(B) ∈ C and T is unital.

Collection of remarks:
General remarks:
Check and reorganize the proof !!!
Items have been changed !!!
Condition (α) is satisfied by some h ∈ C+ with T (h) = 0 if B = R for the
closed right ideal R := {d ∈ B ; limn h1/n d = d}. In particular it suffices to check
this for d in a subset of B that generates B as a closed right ideal.
It suffices for Condition (β) to check the existence of da ∈ M(B) with

kd∗a cda − aT (c)ak < ε for eachc ∈ Xγ

and for each a ∈ B+ in an approximate unit for B and a family {Xγ } of finite
subsets Xγ ⊂ B+ that have the property that the linear spans L(Xγ ) build an
S
upward directed net of linear subspaces of B with γ L(Xγ ) dense in B. This
implies then that T is a point-wise approximately inner completely positive map.
Are there weaker formulation of Condition (β) possible? e.g. :
(New β): For every a ∈ B+ , every pair of positive contractions c1 , c2 ∈ C+ and
every ε > 0 there exists d ∈ M(B), – depending on (a, c1 , c2 , ε) – with

kd∗ cj d − aT (cj )ak < ε for j ∈ {1, 2}

We can replace the element d ∈ M(B) in Condition (New β) or in Condition

(β) by some d ∈ B, because aT (c)a ∈ B for a ∈ B+ , deλ ∈ B and

lim kaT (c)a − eλ aT (c)aeλ k = 0

λ

for any quasi-central approximate unit {eλ } ⊆ B+ of B ( 10 ).

Now we use Condition (α) and the positivity of T . They imply the following:
For any ϕ, ψ ∈ C0 (0, 1] and each c ∈ C we have that

T ((1 − ϕ(h))c(1 − ψ(h))) = T (c),

i.e., T (ϕ(h)c) = 0 for all c ∈ C and ϕ ∈ C0 (0, 1].

For any finite subset X ⊆ C+ , ε > 0 and contraction ϕ ∈ C0 (0, 1]+ there exists
for each x ∈ X an element dx ∈ B with kd∗x (1 − ϕ(h))x(1 − ϕ(h))dx − aT (x)ak < ε .
Since hBh = B by Condition (α), it implies that we can find a sequence of
mutually orthogonal positive contractions en ∈ C ∗ (h)+ ⊆ C ⊂ M(B) such that,

10 Quasi-central approximate units of B exist in B for every C *-algebra B.

+
 4. A “TAUTOLOGIC” WEYL–VON NEUMANN TYPE RESULT 681

for each a ∈ B+ and x ∈ C+ , there exists a sequence of elements dk ∈ B and

numbers nk ∈ N such that

kd∗k enk xenk dk − aT (x)ak < 2−k .

Unfortunately this seems only to work for each single element of finite subsets
X ⊂ C+ .
Condition (α) implies that there exists a contraction h ∈ C+ ⊆ M(B) with the
property that for any given ??????
Question:
Is it enough to require in place of (β) for every a ∈ B+ , every pair contraction
c1 , c2 ∈ C+ and every ε > 0 there exists d ∈ M(B) with kd∗ ck d − aT (ck )ak < ε ?
Could it be that Condition (α) allows to use a kind of an approximate induction
procedure:
Suppose we have verified (β) for all finite subsets X ⊂ C+ with cardinality ≤ n
(with fixed n ≥ 2) and all ε > 0. Is it possible to derive then that (β) also holds
for X with cardinality ≤ n + 1 .
CONCERNING PASSAGE TO THE STABLE CASE:
We show first that the conditions (α) and (β) on a linear map T : C → M(B)
of norm kT k ≤ 1 imply that T is completely positive (and T = 0 is possible).
Then each of the maps

T1 : C ⊗ 1 3 c ⊗ 1 7→ T (c) ⊗ 1 ∈ M(B ⊗ K)

– where here 1 means the unit element 1M(K) of M(K) –, and

T2 : C ⊗ K 3 c 7→ (T ⊗ idK )(c) ∈ M(B) ⊗ K ⊆ M(B ⊗ K)

satisfy again the Condition (α) with h ⊗ 1 ∈ C ⊗ 1 for T1 and for T2 e.g. with h ⊗ q
in place of h, where q := n n−2 pnn .
P

Notice that T1 is unitary equivalent to its infinite repeat, but T2 is usually not,
but the quotient C *-morphism πB⊗K ◦T2 : C ⊗K → Qs (B) dominates zero (in sense
of Definition 4.3.3).
The elements a ∈ (B ⊗ K)+ can be approximated in norm by elements in
(b0 ⊗ q)(B ⊗ K)+ (b0 ⊗ q) with a strictly positive element b0 ∈ B+ .
In case of T2 : C ⊗K → M(B ⊗K) it is enough to consider finite sets X ⊂ C ⊗K
containing only tensors c ⊗ p with c ∈ C+ and p ∈ K a projection.
It follows that it suffices to check the Condition (β) for completely positive
maps T1 or T2 (in place of T ) only on those finite subsets X ⊆ (C ⊗ 1)+ that
contain finitely many elements c ⊗ 1 with c ∈ C+ in case of T1 , and those finite
subsets X ⊆ (C ⊗ K)+ that contain only elements c ⊗ p with c ∈ C+ and p ∈ K in
case of T2 .
????? the elements (b0 ⊗ q)1/n in place of the required arbitrary elements
a ∈ (B ⊗ K)+ in the test criterium for Condition (β).
682 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

The linear span of the elements c ⊗ p with c ∈ C+ and p ∈ K a rank-one

projection in K is dense in X ⊆ (B ⊗ K)+ .
For every a ∈ B+ , every finite subset X ⊂ C+ of contractions and every ε > 0
there exists d ∈ M(B) with kd∗ cd − aT (c)ak < ε for c ∈ X.
h ⊗ ( n n−2 pnn ) ∈ B ⊗ K
P

in place of B,
??? map from T ⊗ idK also the properties (α) and (β) for B ⊗ K in place of B
with h ⊗ ( n n−2 pnn ) in place of h and d ⊗ 1 in place of d?
P

If we can shown that kT ⊗ idK k ≤ 1, then it suffices to consider stable B.

We have anyway kT ⊗ 1M(K) k ≤ 1, because kc ⊗ 1k = kck.
Desire / hope more:
Ad(i): Sk∗ S` = 0 ?? Perhaps at least sufficient that only Sk∗ S` ∈ B for k 6= ` ?
The strong hope is that (moreover) S1 , S2 , . . . can be found such that:
Sj∗ Sk = 0 for j 6= k
Or at least: Sj∗ Sk ∈ B and limn Sn+k
∗
Sn = 0 for j 6= k, k ≥ 1. ??
Could be related to extendability to unital case T˜ ??
Reality check to (i):
Contractions S1 , S2 , . . ..
Sj∗ cSk ∈ B for all c ∈ C (Suffices: Sj∗ cSk − δjk T (c) ∈ B For each c ∈ Lm and
m ∈ N.
For each m ∈ N and ε > 0 there exist n := n(m, ε) such that kSjc Sk k ≤ εkck
for all j, k ≥ n and c ∈ Lm .
and, for all c ∈ C and n, m ∈ N,

Sn∗ cSm − δn,m T (c) ∈ B and ∗

lim kδk,0 · T (c) − Sn+k cSn k = 0 .
n→∞

In the special case where Sn∗ Sn+1 = 0, let α(t) := t − n + 1 for t ∈ [n − 1, n]

and

S(t) := α(t)Sn+1 + (1 − α(t)2 )1/2 Sn .

Calc: S(t)∗ S(t) = α2 Sn∗ Sn + (1 − α2 )Sn+1

∗
Sn+1 + 0 if Sn∗ Sn+1 = 0.
∗ ∗
If Sn+1 cSn ∈ B and kSn+1 cSn k ≤ 2−(n+1) kck and kSn∗ cSn − T (c)k ≤ 2−n kck
∗
and kSn+1 cSn+1 − T (c)k ≤ 2−(n+1) kck for c ∈ Ln , then we get:
S(t)∗ cS(t) = α2 Sn∗ cSn + (1 − α2 )Sn+1
∗
cSn+1 + α(1 − α2 )1/2 Sn∗ cSn+1 + α(1 −
∗
α2 )1/2 Sn+1 cSn
It has estimates kS(t)∗ cS(t) − T (c)k ≤ kck · α2 2−n + (1 − α2 )2−(n+1) + α(1 −
2 1/2 −n
≤ 2−n


α ) 2
 4. A “TAUTOLOGIC” WEYL–VON NEUMANN TYPE RESULT 683

The map t ∈ R+ 7→ S(t) is norm-continuous path in the contractions in M(B)

and satisfies for s, t ∈ R+ and c ∈ C,

S(s)∗ cS(t) − δs,t T (c) ∈ B and lim kδs,0 T (c) − S(t + s)∗ cS(t)k = 0.
t→∞

Indeed: Let s ∈ [m − 1, m] and t ∈ [n − 1, n], β := s − m + 1, α := t − n + 1,

S(s) = βSm + (1 − β 1/2 )1/2 Sm+1 , S(t) := αSn + (1 − α2 )1/2 Sn+1 . If c ∈ M(B),
then S(s)∗ cS(t) = αβSm
∗ ∗
cSn + α(1 − β 1/2 )1/2 Sm+1 cSn +
∗ ∗
(1 − α2 )1/2 βSm cSn+1 + (1 − α2 )1/2 (1 − β 1/2 )1/2 Sm+1 cSn .
To (ii): requires to distinguish the unital, non-unital,
or stable case ??
It seems that second part of (ii)
is not so different to the case of NON-STABLE B.
Compare all with Part (iv) !!!
Ad (iii): ‘‘πB ◦ T is a C∗ -morphism’’ enough??
The unital, non-unital, or stable can give
different results??
Addition in (iii) requires that πB (Sn ) resp. πB (S(t)) is an
isometry and that O2 is unitally contained in πB (T (C))0 ∩ Q(B).
General remark to proof: Notice that for a verification of assumption (β) it
suffices to consider only positive contractions a := am in a given approximate unit
(am )m∈N of B.

Proof. The following are elementary consequences of assumptions (α) and

(β): If Y = {y1 , . . . , yn } ⊂ C is any finite subset of contractions in C, then we
can pass to the finite set X := {x1,1 , . . . , x1,4 , . . . , xn,1 , . . . , xn,4 } of the at most
4n positive contractions xk,1 , xk,2 , xk,3 , xk,4 , given by the polar decompositions
xk,1 − xk,2 = 2−1 (yk∗ + yk ) and xk,3 − xk,4 = i2−1 (yk∗ − yk ) of the real and imaginary
parts of the yk ∈ Y . An application of Condition (β) to X gives that there is
d ∈ M(B) with kd∗ yk d − aT (yk )ak < 4ε for k = 1, . . . , n.
If we apply this observation to the set Y := {c1,1 , c1,2 , . . . , cnn } of entries cj,k
of a positive contraction [cj,k ] ∈ Mn (C)+ and take a ∈ {em } for an approximate
unit em ∈ B+ of B, then we can see – with help of Condition (β) with ε/(4n2 ) in
place of ε –, that there exists dm ∈ M(B) for each a := em and ε := 1/m such
that in Mn (B) holds

k (dm ⊗ 1n )∗ [cj,k ]n,n (dm ⊗ 1n ) − (em ⊗ 1n )[T (cj,k )]n,n (em ⊗ 1) k < m−1 .

In particular, the matrix [T (cj,k )] is positive in Mn (M(B)) for each positive matrix
[cj,k ] ∈ Mn (C)+ , i.e., we get that the map T : C → M(B) is a completely positive
contraction.
Therefore, we can tensor it with idK and get a c.p. contraction Ts := T ⊗ idK
P −n
from C ⊗ K into M(B) ⊗ K ⊆ M(B ⊗ K). Let e := 2 pnn ∈ K .
684 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

The c.p. contraction Ts : C ⊗ K → M(B ⊗ K) satisfies again the conditions (α)

and (β) with C ⊗ K, B ⊗ K and h ⊗ e in places of C, B and h, but now with the
estimate

kTs (c) − (d ⊗ e1/n )∗ c(d ⊗ e1/n )k < ε ,

where d comes from condition (β) for suitably chosen n ∈ N.

It is clear that Ts (h ⊗ e) = 0 and lim(h ⊗ e)1/n c = c for each c ∈ B ⊗ K.
We can replace the d in Condition (β) by daγ ∈ B for sufficiently small γ > 0.
Thus, the element d ∈ M(B) in the inequality of Condition (β) for a given finite
set X ⊂ Cs.a. of selfadjoint contractions x = x+ − x− in C and a ∈ B+ can be
always chosen such that d ∈ B.
It follows for fixed a ∈ B+ that the c.p. map c ∈ C 7→ aT (c)a ∈ B is “approx-
imately 1-step inner” in the strict topology of M(B) as a map from C ⊆ M(B)
into B by assumption (β), and we can apply Lemma 3.1.8 by Remark 3.1.7 to this
approximately 1-step inner map. It gives an element d ∈ B that satisfies condition
(β) with norm kdk ≤ kak , because the c.p. map c ∈ C 7→ aT (c)a ∈ B has norm
kak2 kT k ≤ kak2 .
If b ∈ B is not positive, then a := (bb∗ )1/4 , b0 := a−1 b := limn (bb∗ + 1/n)−1/4 b
with norms kak = kbk1/2 = kb0 k and there exists d0 ∈ B, depending on a, X and
ε > 0, with kd0 k ≤ kak and kd∗0 xd0 − aT (x)ak < γ := ε/(1 + kbk1/2 ) for x ∈ X.
Thus d := d0 b0 satisfies kd∗ xd − b∗ T (x)bk ≤ kbk1/2 γ ≤ ε and kdk ≤ kbk.
Hence, we can use in the following also the c.p. maps b∗ T (·)b with non-
selfadjoint b ∈ B in place of aT (·)a in Condition (β), and find the element d ∈ B
with k d∗ c d − b∗ T (c)b k < ε for c ∈ X with kdk ≤ kbk and d ∈ B · b.
Now we bring Condition (α) into play:
If ϕ ∈ C0 (0, 1]+ is any non-negative continuous function with ϕ(0) = 0 and kϕk ≤ 1,
then we can find moreover an element d ∈ B depending on X, b ∈ B, ϕ and ε, such
that kdk ≤ kbk and

k d∗ (1 − ϕ(h))x(1 − ϕ(h))d − b∗ T (x)b k < ε for all x ∈ X . (4.2)

This holds because the assumption T (h) = 0 from Condition (α) and the positivity
of T imply T (cϕ(h)) = 0 for c ∈ C and ϕ ∈ C0 (0, 1], which gives that

T ((1 − ϕ(h))c (1 − ϕ(h))) = T (c) for all c ∈ C . (4.3)

To get the requested element d from Condition (β) – with a replaced by our non-
selfadjoint b (as discussed above) –, we replace X = {x1 , . . . , xn } by

gXg := { gx1 g, . . . , gxn g } ,

where we let g := 1 − ϕ(h), and use Equation (4.3). We find some d ∈ B depending
on X, b and ε, with kdk ≤ kbk such that d solves the Inequality (β) for gXg and b
in place of X and a, i.e., d solves the Inequality (4.2).
 4. A “TAUTOLOGIC” WEYL–VON NEUMANN TYPE RESULT 685

We can go one step further and consider the specific elements hn := ψn (h) for
increasing pice-wise linear functions ψn ∈ Cc (0, 1], as e.g. defined by

hn := ψn (h) for ψn (t) := min(1, max(0, 2n t − 1)) . (4.4)

Notice that the positive contractions hn+1 hn = hn for all n ∈ N. Above consider-
ations show that we can use in place of the before considered general ϕ(h) any of
the hn .
This allows to choose another solutions d ∈ B of the inequality in assumption
(β) in the following manner:
Let n0 ∈ N, and X a finite set of selfadjoint contractions x∗ = x ∈ C, an element
b ∈ B and ε > 0. Moreover we suppose that there is given an element f ∈ B and
let δ ∈ (0, ε), e.g. let δ := ε/(2 + 2kbk). Condition (α) implies that

lim hn b = lim h1/n b = b for all b ∈ B . (4.5)

n n

It implies that the decreasing sequence {(1 − hm )2 } converges strictly to zero in

M(B). Thus, there is a number m1 := m(f, δ) > n0 with kf − hm f k < δ and
kxf − hm xf k < δ for each m ≥ m1 and x ∈ X.
We can replace for m > m1 > n0 the finite set X of selfadjoint contractions by
the set (1 − hm )X(1 − hm ) and apply Equation T ((1 − hm )c(1 − hm )) = T (c) for
c ∈ C – coming from Equation (4.3) – and find d ∈ B with kdk ≤ kbk such that for
x ∈ X and ε > 0 holds

k d∗ (1 − hm )x(1 − hm )d − b∗ T (x)b k < ε/2 . (4.6)

The new element d ∈ B depends from X, b, m ∈ N and ε. Moreover our choice of

m ∈ N ensures that kd∗ (1 − hm )xf k < δkbk for x ∈ X and kd∗ (1 − hm )f k < δkbk.
Let δ ∈ (0, ε/(4 + 4kdk)) and let f ∈ B. The elements (1 − hm )d and xf are
in B. Therefore we find by Equation (4.5) a number n ∈ N such that n > m + 2,
k hn (1 − hm )d − (1 − hm )d k < δ and k hn xf − xf k < δ for x ∈ X.
Then element d0 := hn (1 − hm )d satisfies kd0 k ≤ kdk ≤ kbk, and for x ∈ X,

k d∗0 xd0 − d∗ (1 − hm )x(1 − hm )d k ≤ 2δkbk < ε/2

and

kd∗0 xf − d∗ (1 − hm )xf k ≤ kbk · khn (1 − hm )xf − (1 − hm )xf k ≤ δkbk < ε/2 .

It implies that hn , hm , d, b ∈ B and all x ∈ X satisfy following inequalities

k (hn (1 − hm )d)∗ x(hn (1 − hm )d) − b∗ T (x)b k < ε , (4.7)

k(hn (1 − hm )d)∗ xf k < ε/2 . (4.8)

and k(hn (1 − hm )d)∗ xf k ≤ kdk · kxf − hm xf k < ε/2 for x ∈ X.

This describes the idea for an induction procedure if b1 , b2 , . . . ∈ B, a zero-
sequence ε1 > ε2 > · · · > 0 (or finite sequence) and an increasing sequence of finite
subsets X1 ⊆ X2 ⊆ · · · of the selfadjoint contractions in C are given.
686 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

To explain in more detail the proposed induction procedure that is indicated

by the above observations, we rename above considered b, f ∈ B, X ⊆ C, ε > 0,
and the from them defined d ∈ B, m < n ∈ N by b1 , f1 , X1 , ε1 , d1 , m1 + 2 < n1 .
The f1 , i.e., in the above side condition involving f , was an “additional” infor-
mation, e.g. we can put f1 := f := 0 for n := 1. In the induction step (from k − 1
to k) it will be defined below more generally by
k−1
X
1/2
fk := (hnj (1 − hmj )dj )(hnj (1 − hmj )dj )∗ . (4.9)
j=1

Let X1 ⊆ X2 ⊂ C finite sets of contractions, b2 ∈ B an arbitrary element and

ε2 ∈ (0, ε1 ) given. Take from the former step d1 and n1 > m1 + 2 in N. Define

1/2
f2 := hn1 (1 − hm1 )d1 d∗1 (1 − hm1 )hn1 and find m2 > n1 + 2 such that

khm2 xf2 − xf2 k < ε2 /2

for all x ∈ X2 . Notice that automatically hm2 f2 = f2 because hm2 hn1 = hn1 .
We find d2 ∈ B with kd2 k ≤ kb2 k such that, for x ∈ X2 ,

k d∗2 (1 − hm2 )x(1 − hm2 )d2 − b∗2 T (x)b2 k < ε2 .

Let δ2 := ε2 /(2 + 2kb2 k) . After we have selected d2 we apply the above considera-
tion and find a suitable number n2 > m2 + 1 such that, for x ∈ X2 ,

k hn2 (1 − hm2 )d2 − (1 − hm2 )d2 k < δ2 and k hn2 xf2 − xf2 k < δ2 .

From now on we use the new notation h[m, n] := hn (1 − hm ) = hn − hm for

m < n that avoids iterated indices and seems to be more transparent ( 11 ). The
h[n, m] are positive contractions that satisfy the identities h[`, p]α h[m, n] = h[m, n] ,
h[m, n]h[p, q] = 0 , and h[m, n] + h[n, p] = h[m, p] for α ∈ (0, 1] and `, m, n, p ∈ N
with ` < m < n < p < q. Then we get for x ∈ X2 and h[m2 , n2 ] = hn2 (1 − hm2 )
that
k (h[m2 , n2 ]d2 )∗ x(h[m2 , n2 ]d2 ) − b∗2 T (x)b2 k < ε2 ,
(h[m2 , n2 ]d2 )∗ (h[m1 , n1 ]d1 ) = 0 by m2 > n1 , and

k (h[m2 , n2 ]d2 )∗ xf2 k < ε2 .

If we repeat this arguments for given contractions b1 , b2 , . . . , bk , . . . ∈ B, selfadjoint

contractions c1 := 1, c2 = h, c3 , . . . , ck , . . . ∈ C+ and a given decreasing zero se-
quence ε1 > ε2 > · · · > 0, then we obtain by the above described method for
each k ∈ N and Xk := {c1 , . . . , ck }, suitable positive integers with `1 := 1,
`k < mk < nk < `k+1 (with `1 := 1 and element dk ∈ B with kdk k ≤ kbk k
that satisfy the general Inequalities:

k (h[mk , nk ]dk )∗ x(h[mk , nk ]dk ) − b∗k T (x)bk k < εk for all x ∈ Xk . (4.10)

and k(h[mk , nk ]dk )∗ x(h[mj−1 , nj−1 ]dj−1 )k < εk for x ∈ Xk and 1 ≤ j < k.

11 Observe here and in some later proofs that the expressions [m, n] denote pairs of indices

and not any sort of commutators!

 4. A “TAUTOLOGIC” WEYL–VON NEUMANN TYPE RESULT 687

What happens with j = k−1 ??? Perhaps we have to replace in Equation (4.10)
(h[mk , nk ]dk )∗ x(h[mk , nk ]dk ) − b∗k T (x)bk simply by (h[mk , nk ]dk )∗ x(h[mj , nj ]dj ) −
δjk · b∗k T (x)bk for 1 ≤ j ≤ k
The latter inequalities follow from k (h[mk , nk ]dk )∗ x fn k < εk for fn as de-
fined in Equation (4.9). We have automatically (h[mk , nk ]dk )∗ (h[mj , nj ]dj ) = 0 for
j = 1, . . . , k − 1 by h[mk , nk ]h[mj , nj ] = 0 for j < k. This can be sees also from
the equations h[mk , nk ]h[`k , `k+1 ] = h[mk , nk ], for the suitable chosen sequence
1 := `0 < `1 < `2 < . . . in N with `k < mk < nk < `k+1 . Such a sequence exists
because we have chosen mk+1 > 1 + nk .
In particular, for a given finite set X ⊆ C of selfadjoint contractions in C, and
given elements b1 , b2 , . . . , bk ∈ B and ε > 0, we find elements d1 , . . . , dk ∈ B with
kdj k ≤ kbj k that satisfy

kd∗i xdj − δi,j b∗i T (x)bj k < ε

for 1 ≤ i, j ≤ k and x ∈ X, i.e., we get the following observation:

Observation (1):
Given contractions b1 , . . . , bn and X ⊂ C a finite subset of the selfadjoint contrac-
tions, then, for every ε > 0, there exists contractions d1 , . . . , dn ∈ B such that
d∗j dk = 0 for j < k and

k d∗j xdk − δj,k b∗k T (x)bk k < ε/n2 for k, j = 1, . . . , n, and x ∈ X . (4.11)

The Inequality (4.11) implies the approximate 1-step innerness of the c.p. map
Pn ∗
k=1 bk T (·)bk := NAME? S ???????? with respect to the strict topology on M(B),
i.e., the element d := d1 + d2 + · · · + dn ∈ B and all x ∈ X satisfy the Inequality
n
X
k d∗ xd − b∗k T (x) bk k < ε . (4.12)
k=1

Proof of Part(i):
We are going to choose the following objects µ, Xn , fk and Sn to make the
rest of the proof transparent and almost “constructive” – in a certain sense:
(A) A bijection µ form N × N onto N such that

µ((k, 1)) = min µ({k} × N) < µ((n, 1))

S
for k < n in N . It can be used to define a (disjoint) decomposition k Mk of N
given by bijective and order preserving maps νk : N → N with k ≤ νk (1) < νn (1)
for k < n in N. Here Mk = νk (N) = µ({k} × N). We have ν1 (n) = n for all
n < ν2 (1) = min M2 , ν1 (n) > n for n ≥ ν2 (1), and νk (1) > k for all k > 1. It
follows that νk (n) ≥ k + n for k > 1, n ∈ N and ν1 (n) ≥ n + 1 for n ∈ N.
(B) Filtration Xn of C and quasi-central approximate unit (en ) for V (Xn )
defined from strictly positive contraction e ∈ B+ .
688 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

(C) By induction:
Find solutions mk , nk , dk , and the such defined elements h[mk , nk ]dk =: gk of
1/2
(4.10) for bk := ek , with εk := 4−n and from them fk defined by Equation (4.9).
(D) The Sn are then (suitable) Γ-sums for the in (A) defined maps νn : N → N,

Sn := Γn ( gνk (1) , gνk (2) , . . .) ,

with Γn (·) defined by the x(n) and y (n) as where xn and yn have the components
components ?????????????
We use the above considerations concerning the inductive solutions of the In-
equalities (4.10) and Part (3) of Remarks 5.1.1 to define a filtration of C given by
linear spans Ln := span(Xn ) with Xn := {c1 , . . . , ckn }, defined from a sequence of
selfadjoint contractions c1 , c2 , . . . ∈ C that is dense in the unit ball of the self-adjoint
part Cs.a. of C.
Let e ∈ B+ a strictly positive contraction. We find a sequence of functions
fn ∈ C0 ((0, 1])+ such that en := fn (e) build an approximate unit of B that is
quasi-central with respect to finite subsets Xn ∪ T (Xn ) of M(B). See Part (3)
of Remarks 5.1.1 for more details. We take the there defined adjustment of the
sequence en , – in particular in relation to T (Xn ) – such that, for each (contraction)
x ∈ Xn and m > n,

k [en , T (x)] k + k[(em − en )1/2 , T (x)] k < 4−n .

It follows for infinite subsets M1 , M2 ⊆ N with min M1 < min M2 and order pre-
serving bijective maps λj from N onto Mj , λj : N → Mj , that
NEXT give good estimate:

kΓj,k (T (x), T (x), . . .) − δj,k T (x)k ≤?????

and Γj,k (T (x), T (x), . . .) − δj,k T (x) ∈ B for x ∈ Xn and j, k ∈ {1, 2}. Here
1/2
Γj,k (a1 , a2 , . . .) for j, k ∈ {1, 2} is defined as in Remark 5.1.1(2) with y1 := eλj (1) ,
1/2
yn+1 := (eλj (n + 1) − eλj (n))1/2 x1 := eλk (1) , and xn+1 := (eλk (n + 1) − eλk (n))1/2 .
We define by induction successive elements

g1 := h[m1 , n1 ]d1 , g2 := h[m2 , n2 ]d2 , . . . ∈ B

and a sequence p1 < p2 < . . . in N with pk + 1 < mk , mk + 1 < nk , nk + 1 < pk

such that
1/2 1/2
kgk∗ xg` − δk,` ek T (x)ek k ≤???
for x ∈ Xk , ` ≤ k, and that pk ????gk = gk .
Let M ⊆ N an infinite subset, and let λ : N → M the unique bijective order-
preserving map from N onto M , as described in Part (i,A).
allow to make the following observations (?0.1) - (?0.4):
????
such that corresponding ϕkn (h)(1 − ϕ`n (h))dn
 4. A “TAUTOLOGIC” WEYL–VON NEUMANN TYPE RESULT 689

with nk > 2 + mk , next mk+1 > 2 + n + k ????

can be used in place of dn for

kd∗m xdn − δn,m gn (e)T (x)gn (e)k < 2− max m,n kxk .

NEXT:
The sequences (ϕn ) and h1 , h2 , . . . can be defined by Equation (??).
(1.) h1 , h2 , . . . ∈ C ∗ (h)+ ⊆ C.
“ suitably ” (TO BE MADE PRECISE!!)
positive contractions with hn+1 hn = hn , hn → 1M(B) strictly, using that hB
is dense in B by condition (α).
Define a nice sequence of ϕn .
Such that h[mk , nk ]dk with nk > 2 + mk , next mk+1 > 2 + n + k
can be used in place of dn for
kd∗n xdn − gn (e)T (x)gn (e)k < 2−n kxk
(2.) fn ∈ C ∗ (e)+ ⊆ B for a strictly positive contraction e ∈ B+ with fn+1 fn =
fn and limn ke − fn ek = 0.
Since A := C ∗ (e, C, T (C)) is a separable C *-subalgebra of M(B) we find for
every linear filtration of A by finite-dimensional linear subspaces K1 ⊆ K2 ⊆ . . . of
S
A with n Kn dense in the rational convex hull of {f1 , f2 , . . .} an approximate unit
e1 , e2 , . . . with en+1 en = en and k[en , y]k ≤ 4−n kyk for y ∈ Kn . Compare Remark
5.1.1(3,4).
We can manage that Kn contains Ln ∪ T (Ln ) from a before defined linear
filtration L1 ⊆ L2 ⊆ . . . of C, given by Ln := span(Xn ), Xn := { c1 , c2 , . . . , cn } .
Thus, can find an approximate unit e1 , e2 , . . ., with k [(en+1 − en )1/2 , c] k ≤
√
3(2−n ) 2 kck
Compare next with above given definition of h[m, n]
without taking roots!
We define
h[j, k] := (hk − hj ) = (hk (1 − hj ))
for k > j. Notice that h[j + 1, k − 1](hk − hj ) = hk − hj and h[j1 , k1 ]h[j2 , k2 ] = 0
if k1 < j2 .
Find d1 , d2 , . . . ∈ B and m1 < n1 < m2 < n2 < · · · in N with the properties:
(1.) mk+1 > nk + 3
(2.) kdn k ≤ 1,
kd∗n cdn − en+1 T (c)en+1 k ≤ 8−n kck
for all c ∈ Ln , en , en+1 well-commute with c ∈ Ln and with T ( c ) if c ∈ Ln .
(3.) dk = (hnk − hmk )dk , in particular d∗k d` = 0 for ` 6= k.
(4.) k(hnk+1 − hmk+1 )c dk k ≤ 8−n k c k for all c ∈ Ln .
690 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

(5.) In particular ???, kc dk − hmk+1 c dk k <?????.

That is not clear !!!
(6.) The contractions Sn should be defined as some Sn = Sµ,ν , where
X
Sµ,ν := dν(`) (eµ(`+1) − eµ(`) )1/2
`

for suitable injective maps ν : N → N and µ : N → N.

(6.0) We take a bijective map λ : N → N×N such that λ−1 (n, 1) < λ−1 (n+1, 1)
and λ−1 (n, 1) = min λ−1 ({n} × N) for all n ∈ N. E.g. take λ defined by the order
on N × N given by (1, 1) < (1, 2) < (2, 1) < (1, 3) < (2, 2) < (3, 1) < (1, 4) < . . .,
which comes from γ : (m, n) → k := m + n − 1 ∈ N, γ −1 (k) lexicographic ordered
as e.g. γ −1 (1) = {(1, 1)}, γ −1 (2) = {(1, 2) < (2, 1)}, and so on. Then λ defines a
decomposition of N into countably many pairwise disjoint infinite subsets Mn :=
λ−1 ({n} × N), such that kn := min Mn satisfies kn < kn+1 .
The natural maps order preserving bijections µn : N → Mn satisfy µ1 (k) = k
for k = 1, . . . , µ2 (1) − 1 and µn (k) > k for all other pairs (n, k) ∈ N × N.
find more flexible formulation by working with finite subsets!
??
For each `, m, n ∈ N, each finite-dimensional linear subspace ??? L ⊆ C ??? of
C, contractions bk ∈ B, γk ∈ (0, 1), k = 1, . . . , ` there exist ???? L ⊆ C ????

(i) numbers m(j, k), and n(j, k), j = 1, . . . , m, with n+2 < m(j, k), m(j, k)+
2 < n(j, k), n(j, k) + 2 < m(j + 1, k), n(m, k) + 2 < m(1, k + 1).
(ii) Contractions dj,k ∈ B , j = 1, . . . , m; k = 1, . . . , `, with

kd∗i,k xdj,k − b∗k xbk k < γk , and kd∗j,k xdi,k0 k < γmin(k0 ,k) .

(iii) a number ν ∈ N depending on dj,k ∈ B, j = 1, . . . , m, k = 1, . . . , `,

numbers n < n1 < n2 <???
Let b1 , b2 , . . . ∈ B, γ1 , γ2 , . . . ∈ (0, 1) finite or infinite sequence,
?????

Consider the functions

ϕn (ξ) := min 1, max(0, 2n ξ − 1) = (2n ξ − 1)+ − (2n ξ − 2)+ ,

and let hn := ϕn (h) for n = 1, 2, . . ..

The elements hn ∈ M(B)+ are positive contractions with hm hn = hm for
n > m and khn h − hk ≤ 21−n . Since h1/k converges strictly to 1 in M(B) by
(β), the sequence h1 , h2 , . . . tends to 1 unconditional strictly in M(B), because
k(1 − hn )h1/k k ≤ 2(1−n)/k .
We define rational numbers γk > 0, k = 1, 2, . . . , n and select by induction ele-
ments f1 , f2 , . . . , fn−1 and g1 , g2 , . . . , gn from the above defined increasing sequence
S := {h1 , h2 , . . .} ⊂ C ∗ (h)+ ⊂ M(B) , and choose simultaneously contractions
d1 , . . . , dn ∈ B such that they fulfill together following equations and inequalities,
where we let g1 := h1 at the first step:
 4. A “TAUTOLOGIC” WEYL–VON NEUMANN TYPE RESULT 691

(1.1) Find contractions dk ∈ B that fulfill inequality (4.13) for the given bk ,
x ∈ X and given gk ∈ S := {h1 , h2 , . . .}:

k d∗k (1 − gk )x(1 − gk )dk − b∗k T (x)bk k < γk . (4.13)

(1.2) Find fk ∈ S with fk gk = gk , k(1 − fk )dk k < γk and k(1 − fk )xdk k < γk
for x ∈ X, where gk , dk , X are given.
(1.3) Take gk+1 ∈ S with gk+1 fk = fk for given fk ∈ S (if k < n).

We have seen above the possibility to find the contraction dk ∈ B (for given X, γk ,
bk and gk ) that satisfies the inequality (4.13). The existence of fk and gk+1 follow
immediately from the properties of the sequence S.
We apply the conditions on the inductively selected bk , gk and fk to find an
estimate:
Conditions (1.3) and (1.2) on gk and fk imply that gk+1 ≥ fk ≥ gk , and by
induction that g` ≥ fk and (1 − g` )(1 − fk ) = (1 − g` ) for all ` > k. The elements
d` and (1 − gk ) are contractions by condition (1.2), and we get from condition (1.2)
that
kd∗` (1 − g` )x(1 − gk )dk k ≤ k(1 − fk )x(1 − gk )dk k < γ .
If we combine this with Inequalities (4.13) and (4.18) we get that the element
D := (1 − g1 )d1 + . . . + (1 − gn )dn ∈ B satisfies

kD∗ xD − V (x)k < ε for all x ∈ X ,

because
X
kD∗ xD − b∗k V (x)bk k < n2 δ = ε/2
k

by kd∗` (1 − g` )x(1 − gk )dk k < δ for k 6= `, and by the Inequalities (4.13).

Hence, V := δ∞ ◦ T : C → M(B) satisfies also condition (β).
2
Notice that δ∞ ◦ V = δ∞ ◦ T is unitarily equivalent to V by a unitary in
u ∈ M(B) by Lemma 5.1.2(i), and that T = s∗k V (·)sk for each k ∈ N if δ∞ (b) =
∗
P
k sk bsk .

Next assumption necessary?

It becomes relevant in Part (ii), where B is stable,
And not here!!!
We suppose from now on that T is unitary equivalent to δ∞ ◦ T . ????
Begin of a bit revised old text:
It seems better to define first the needed universal sequences of ek ∈ B and
dk ∈ B with the “good” properties with respect to a given filtration Ln of C. And
then select the corresponding hn(k) − hm(k) .
Then define S and more general Sn only with help of this sequences!
We define below a contraction S ∈ M(B) with the property S ∗ cS − T (c) ∈ B
for all c ∈ C.
692 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

The element S is defined as the sum of an unconditional strictly convergent

series by
X
S := (fn+1 − fn )1/2 dn (en+1 − en )1/2
n

where the fn = hkn ∈ S := {h1 , h2 , . . .} have to be suitably chosen, and

{e1 , e2 , . . .} ⊂ B+ is an approximate unit of B that is quasi-central with respect
of the C *-subalgebra A := C ∗ (e, C, T (C)) ⊆ M(B) build from a strictly positive
contraction e ∈ B+ of B and contractions dn ∈ B. They are “suitably chosen” in
the sense that the series S converges strictly and unconditional to a contraction in
B, kSk ≤ 1, and
??????
X
kd∗n (fn+1 − fn )1/2 c(fm+1 − fm )1/2 dm − δn,m en+2 fn+2 cfn+2 en+2 k < ∞ .
m,n

By assumption, B is σ-unital, i.e., there exists a strictly positive element e ∈ B+

with kek = 1.
Steps:
Choose:
(0) strictly positive contraction e ∈ B+ , kek = 1.
(1) linear filtration of C ∗ (e, C, T (C)) by taking a sequence in the unit ball of
∗
C (e, C, T (C))+ that contains a sub-sequence c1 , c2 , . . . that is dense in the positive
≤1
part of the unit ball of C (notation: C+ ???)
the positive rational convex combinations of the ϕn (e),
S := {h1 , h2 , . . .}, ϕn (hk ),
and with ????
defining a filtration Y1 ⊂ Y2 ⊂ . . . of A := C ∗ (e, C, T (C)) such that
Yn contains e, Xn := {h, hn , c1 , . . . , cn }, T (Xn ) and y1 , . . . , yn of a dense se-
≤1
quence in the positive parts C+ and A≤1 + of the unit balls of C+ and of A+ .

The strict convergence of S := {h1 , h2 , . . .} to 1M(B) implies that we can

find for each linear subspace Y ⊆ B of finite dimension and ε > 0 a number
m := m(Y, ε) ∈ N such that k(1 − hm )yk + ky(1 − hm )k < εkyk for all y ∈ Y .
A reformulation of conditions (α) and (β) together says:
Let L ⊂ C a finite-dimensional subspace of C, m ∈ N, a ∈ B a positive
contraction and ε > 0 then there exists d := d(L, ε) ∈ B with kdk ≤ 1 and

kd∗ (1 − hm )c(1 − hm )d − aT (c)ak ≤ εkck ∀ c ∈ L.

And, given any d ∈ B and m ∈ N, γ > 0 we find n ∈ N with n > m + 1 such that
k(1 − hn )dk < γ, k(1 − hn )cdk < γ and k(1 − hn )c(1 − hm )dk < γ for all c ∈ L.
We take a dense sequence in the positive contractions c1 , c2 , . . . ∈ C+ and
denote by Ln the linear span of {h, hn+2 , c1 , . . . , cn }.
 4. A “TAUTOLOGIC” WEYL–VON NEUMANN TYPE RESULT 693

Find positive e1 , e2 , . . . ∈ C ∗ (e)+ ⊂ B+ with en em = emin(m,n) and

k[en , T (x)]k < 4−(n+1) for x ∈ Ln , kxk ≤ 1.
1/2
Then k[en , T (x)]k < 2−n for x ∈ Ln , kxk ≤ 1, k[(en −em ), T (x)]k < 4−(n+1) +
−(m+1)
4 and k[(en − em )−1/2 , T (x)]k < 2− min(n,m) for x ∈ Ln , kxk ≤ 1.
The positive contractions en := ϕn (e) ∈ B with above ???
???? the above defined functions ϕn satisfy ???e0n e0m = e0m for m < n build an
approximate unit of B, i.e., the sequence (??e01 , ??e02 , . . .) converges unconditional
strictly to 1 in M(B).
If we build elements en (n = 1, 2, . . .) by convex combinations of e0kn , . . . , e0ln
with kn < ln < kn+1 then again en em = em for m < n and lim en b = b for all
b ∈ B.
By Remark 5.1.1(3)
– applied
to the ??? of the separable C *-subalgebra A := C ∗ (T (C) ∪ C ∪ {1}) ⊃ C of
M(B), and to a dense sub-sequence y1 , y2 , . . . of the positive contractions in A –
Let x1 , x2 , . . . a dense sequence in the positive contractions of C.
Let Ln and Kn the linear span of Xn := {h, hn , x1 , . . . , xn } ⊂ C respectively
of Xn ∪ T (Xn ) ∪ {y1 , . . . , yn }. Thus Ln ∪ T (Ln ) ⊆ Kn .
By Remarks ??(??,???) we can find positive contractions e1 , . . . , en , . . . ∈
C0 (e)+ that satisfy en em = emin(m,n) , limn en = 1M(B) and that the commuta-
tors [en , x] = en x − xen for x ∈ Kn satisfy k[en , x]k ≤ 8−n kxk ?????? [en , x] and
[en , (1 − x2 )1/2 ] for each x ∈ Kn
have norms

max{ k[en , x]k , k[en , (1 − x2 )1/2 ]k } < 8−2n for all x ∈ Kn , kxk ≤ 1 .

Convergence proof needs explicit dependence !!

1/2
By Lemma 5.3.2 we get the desired estimates of [en , x], [(1 − en )1/2 , x] and
[(en+k − en )1/2 , x].
If 1 ≤ k1 < k2 < . . . is any sequence Σ ⊆ N, and c ∈ C, then, – with ek0 := 0
–, the series
X∞
TΣ (c) := (ekn+1 − ekn )1/2 T (c)(ekn+1 − ekn )1/2
k=0

is unconditional strictly convergent in M(B) for each c ∈ C, and its “sum” c 7→

TΣ (c) defines a c.p. contraction TΣ : C → M(B) with the property TΣ (c)−T (c) ∈ B
for all c ∈ C
...?
Moreover kTΣ (c) − T (c)k ≤??????? for c ∈ Ln .
Indeed, let yn := xn := (ekn+1 − ekn )1/2 in Remark 5.1.1(2), then x2n =
P
n
1 = yn2 , the there defined map Γ corresponding to
P
694 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

??? Y := X := [x1 , x2 , . . .] ??? look this up in 5.1.1(2) ???

is a strictly continuous unital c.p. map from `∞ (M(B)) = M(c0 (B)) into
M(B) and TΣ (c) = Γ(T (c), T (c), . . .) for c ∈ C defines a c.p. contraction from C
into M(B) .
If ?????? ???????
Again by functional calculus – this time applied to the element h in condition
(α), we find a sequence of contractions h01 , h02 , . . . ∈ C ∗ (h)+ ⊆ C+ such that h0n h0m =
h0m for m < n and k(1 − h0n )en k < 8−n . Thus kh0n b − bk → 0 if n → ∞ for each
b ∈ B.
Why not hn (or gn , fn ) in place of h0n ?
If we use that T is positive, h ≥ 0 and T (h) = 0 by assumption (α), we get
T (h0n ) = 0, T (c h0n ) = 0, T (h0n c) = 0 and

T (c) = T ((1 − h0n )c(1 − h0n )), ∀c ∈ C , ∀n ∈ N.

Combining this with condition (β) we get the existence of contractions gn,k ∈ B
(k, n ∈ N) that satisfy, for c ∈ Ln ,

k (gn,k )∗ (1 − (h0k )2 )1/2 c(1 − (h0k )2 )1/2 gn,k − en+3 T (c)en+3 k ≤ 8−n−k kck . (4.14)

We define the desired contractions Sn,p as the sum of strictly converging series
of the kind
X
(h0`(n) − h0m(n) )g2n,m(n) e1/2
n + (h0`(k) − h0m(k) )g2k,m(k) (ek − ek−1 )1/2 ,
k>n

where m(k) < `(k) < m(k + 1) < `(k + 1) with choice depending from p ∈ N. Huge
gaps between `(k) and m(k + 1) can be used to make the Sn,p orthogonal modulo
∗ ∗
B, i.e., Sn,p Sn,q ∈ B and even such that Sn,p cSn,q ⊂ B for c ∈ C and

∗
lim kSn,p cSn,q − δp,q T (c)k = 0 .
n→∞

More precisely, we select inductively a subsequence hn := h0k(n) and finite

subsequences hn,m := h0`(n,m) , m = 1, . . . , n, from the above selected sequence
h01 , h02 , . . . with following properties :

(1) k(0) = 1,
(2) k(n) < `(n, m) < `(n, m + 1) ≤ k(n + 1) for m = 1, 2, . . . , n − 1.
(3) k(1 − h0`(n,m+1) )gn,`(n,m) k < 8−n ,
(4) k(1 − h0k(n+1) )gn,k(n) k < 8−n ,
(5) k(1 − h0k(n+1) )c(1 − h0k(n) )gn,k(n) k < 8−n and
k(1 − hn+1 )c(hj+1 − hj )gj,k(j) k < 8−n for c ∈ Xn+1 , j = 1, . . . , n − 1,
n = 1, 2, . . ., and
(6) (h`(n,m+1) − h`(n,m) )(hn+1 − hn ) = h`(n,m+1) − h`(n,m) , and then
(h`(n1 ,m+1) − h`(n1 ,m) )(h`(n2 ,m+3) − h`(n2 ,m+2) ) = 0
 4. A “TAUTOLOGIC” WEYL–VON NEUMANN TYPE RESULT 695

Notice that (1−hk )(1−hn ) = 1−hn for k < n, and (hk+1 −hk )(hn+1 −hn ) = 0
for k + 1 < n. Let
tn,m := (hn+2,m − hn )gn,k(n) en+2
for all n ∈ N and m ≤ n.
Then the tn,m are contractions with t∗n,m tn,m ≤ e2n+2 , t∗n,m1 tk,m2 = 0 and
∗
ktn,m(2) ctk,m(1) k < 8−n+1 2 for k + 1 < n, c ∈ Xk . We get (hn+2 − hn−1 )1/2 tn,m =
tn,m for n = 1, 2, . . ., m ≤ n.
It follows

kt∗k,m(1) ctn,m(2) − δk,n en+2 T (c)en+2 k < 8−n 3kck

for c ∈ Ln , because ktn,m − (1 − hn )gn,k(n) en+2 k < 8−n .

This follows from the definitions and the inequality k(1−hn+2,m )gn,k(n) k < 8−n ,
because hn+2,m = h0`(n+2,m) and k(1 − h0`(n+2,m) )gn,k(n) k < 8−n .
The latter inequality is a consequence of the following properties of h0n and
gn,k(n) :
The inequality k(1 − h0k(n+1) )gn,k(n) k < 8−n is property (3) of h0k(n) and gn,k(n) ,
and that k(n + 1) < k(n + 2) < `(n + 2, m) follows from property (2) of k(n) and
`(n, m). Then 0 ≤ h0n h0m = h0m ≤ 1 for m < n by construction of h0n . It follows
that
0 ≤ h0k(n+1) = h0k(n+1) h0`(n+2,m) ≤ h0`(n+2,m) ≤ 1 ,
It implies (1 − h0`(n+2,m) )2 ≤ (1 − h0k(n+1) )2 and, therefore

k(1 − h0`(n+2,m) )gn,k(n) k ≤ k(1 − h0k(n+1) )gn,k(n) k < 8−n .

The definition of Sn,p is justified by the unconditional strict convergence of the

sum at the right hand side. The strict convergence can be seen with the argument
mentioned in Remark 5.1.1(2):
We define Sn,p ∈ M(B) for n ∈ N and p ≤ n by
X
Sn,p := t2n,min(p,2n) e1/2
n + t2k,min(p,2k) (ek − ek−1 )1/2 .
k>n

We must show now:

(I) The series
X
t2k,min(p,2k) (ek − ek−1 )1/2
k>n

is unconditional strictly convergent, ....????

∗
(II) limn→∞ kSn,p k = 1, and Sn,p cSn,q − δp,q T (c) ∈ B for all c ∈ C and
n, p, q ∈ N, and
∗
lim kSn,p cSn,q − δp,q T (c)k = 0 .
n→∞

Old version:
696 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

because one can express Sn,p as product RDC of the bounded row matrix
1/2
R := [h2n+2 , (h2(n+1) − h2(n−1) )1/2 , (h2(n+2) − h(n+1)−2 )1/2 , . . .]

the contractive diagonal matrix

D := diag(α1 , α2 , . . .)

with diagonal entries α1 := t2n,m and α` := t2(`+n),m , and the bounded column
matrix C with transposed row

C > := [e1/2
n , (en+1 − en )
1/2
, (en+2 − en+1 )1/2 , . . .] .

Moreover, Remark 5.1.1(2) shows that the Sn,p are contractions.

It is not difficult to see that ?Sk,m −?Sn,m ∈ B and, therefore
∗ ∗
?Sk,m cSk,m −?Sn,m cSn,m ∈ B

for every c ∈ C.
Now let c ∈ Xm and, for n > m let
∗
am := e1/2 1/2
m t2m ct2m em − T (c)em , (4.15)
bn := (en − en−1 )1/2 t∗2n ct2n (en − en−1 )1/2 − T (c)(en − en−1 ), (4.16)
dm,n := (en − en−1 )1/2 t∗2n ct2m e1/2 1/2 ∗
m + em t2m ct2n (en − en−1 )
1/2
, (4.17)

and, for i < j,

fi,j := (ej − ej−1 )1/2 t∗2j ct2i (ei − ei−1 )1/2 + (ei − ei−1 )1/2 t∗2i ct2j (ej − ej−1 )1/2 .

Then from d1/2 T (c)d1/2 − T (c)d = (d1/2 T (c) − T (c)d1/2 )d1/2 we get kak k <
8−2m 3 + 8−k , kbn k < 8−2n 3 + 8−n , kdm,n k < 8−2n+1 2 and kfi,j k < 8−2j+1 2 for
m < n, m < i < j.
Since em + (em+1 − em ) + . . . = 1 strictly, for c ∈ Xm ,
∞
X ∞
X X
∗
Sm cSm − T (c) = am + bn + dm,n + fi,j .
n=m+1 n=m+1 m<i<j

The right hand side are absolutely convergent series in B. The sums of the norms
can be estimated by by 8−m 4.
∗
It follows that kSk,m cSk,m − T (c)k < 2−k and that Sk,m
∗
cSk,m − T (c) ∈ B for
∗ ∗ ∗
c ∈ Xk . Since Sk cSk − Sn cSn ∈ B, we get that Sn,m cSn,m − T (c) ∈ B for each
n = 1, 2, . . . and every element c of our dense sequence in the unit ball of C.
∗ ∗
Thus, Sn,m cSn,m − T (c) ∈ B and limn kSn,m cSn,m − T (c)k = 0 for every c ∈ C.
Our desire was to deepen the results by proving
orthogonality modulo B for the Sn ...
still to check:
∗ ∗
We have Sn,m 1
cSn,m2 ∈ B for m1 6= m2 and kSn,m 1
cSn,m2 k → 0 for n → ∞ if
c ∈ C or c = 1.
 4. A “TAUTOLOGIC” WEYL–VON NEUMANN TYPE RESULT 697

(ii): Recall that B is stable if and only if M(B) contains a sequence of

isometries s1 , s2 , . . . such that k sk s∗k = 1, cf. Remark 5.1.1(8). Then δ∞ (b) :=
P
∗
P
k sk bsk converges unconditional strictly for b ∈ M(B), is a strictly continuous
unital *-endomorphism and δ∞ (M(B))0 ∩M(B) = δ∞ (B)0 ∩M(B) contains a copy
of O2 unitally by Remark 5.1.1(8).
The contractive c.p. map V := δ∞ ◦ T , i.e., V (·) := k sk T (·)s∗k again satisfies
P

Conditions (α) and (β):

Clearly V satisfies condition (α), because V (h) = δ∞ (T (h)) = 0.
Let X ⊆ C+ a finite subset of the contractions in C+ , b ∈ B a contraction and
ε > 0.
There exists q ∈ N with

k (1 − Pq )b k < ε/4
Pq ∗
P ∗
for the projection Pq := k=1 sk sk , because k sk sk converges strictly to 1 ∈
M(B).
It implies for x ∈ X and the contractions bk := s∗k b ∈ B that
q
X
k b∗ V (x)b − b∗k T (x)bk k < ε/2 . (4.18)
k=1

We have seen at the end of the general observation, cf. Inequality (4.12), that,
for finite X ⊂ C+ , b1 , . . . , bq ∈ B and ε > 0, there exists an element d ∈ B with
q
X
kd∗ xd − b∗k T (x)bk k < ε/2
k=1

Thus d ∈ B solves the inequalities kd∗ xd − b∗ V (x)bk < ε (x ∈ X) for given finite
X ⊆ C+ , b ∈ B, ε > 0 and V := δ∞ ◦ T .
This confirms that δ∞ ◦ T satisfies also the Condition (β).
NEXT TO BE SORTED WITH NEW INSIGHT.
By Part (i), there exists a contraction g ∈ M(B) with V (c) − g ∗ cg ∈ B for all
c ∈ C.
Apply ???? Part (i) to V =: δ∞ ◦ T and C. Let g := S1 with S1 as in Part (i).
We use Lemma 5.1.2(v). Let t 7→ S 0 (t) the map for V = δ∞ ◦ T in Part (v) of
Lemma 5.1.2, and let s1 , s2 , . . . be the generators of a unital copy of O∞ in M(B)
with n sn s∗n = 1 (strictly convergent) such that δ∞ = n sn ( · )s∗n , cf. Remark
P P

5.1.1(8). Then t 7→ S(t) := S 0 (t)s1 is as desired for T .

Now notice that ?????????? ...
When S(t) is an isometry ? E.g. if S 0 (t) is a contraction with S 0 (t)∗ S(t) ≥ s1 s∗1
and s∗1 s1 = 1 ??
When absorption happens?
(iii): If 1M(B) ∈ C, then T is unital, because T (C)B is dense in B.
698 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

By Part (ii), V := δ∞ ◦ T satisfies (α) and (β). Moreover V (C)0 ∩ M(B) ⊂

δ∞ (M(B))0 ∩ M(B) contains a copy of O2 unitally.
By Part (i), there exists a contraction g ∈ M(B) with V (c) − g ∗ cg ∈ B for all
c ∈ C. The existence of U (t) now follows from Lemma 5.1.2(vi).
If 1M(B) is not in C, then let C
e := C + C1M(B) and Te(c + z1) := T (c) + z1
for c ∈ C and z ∈ C.
e and V := δ∞ ◦ Te again satisfy (α)
Then Te is a unital C *-morphism, and C
and (β) (in place of C and T there).
Indeed: V (h) = δ∞ (T (h)) = 0 gives (α).
We show (β) for the unital C *-morphism V : C e → M(B):
It suffices to check the condition (β) for V on sets of the form X ∪ {1} and a ∈ B+ ,
where kak ≤ 1 and X is a finite set of contractions in C.
By assumption, T (C)B is dense in B. The set δ∞ (B)B is dense in B, because
δ∞ is a unital and strictly continuous endomorphism of M(B). It follows that the
set V (C)B is dense in B, because V (C)δ∞ (B)B = δ∞ (T (C)B)B.
Let X be a finite subset of the contractions in C, a ∈ B+ a contraction and
ε > 0.
There exists a contraction
e was used as strictly positive in B
e ∈ C+ with kV (e)a − ak < ε/4 and kece − ck < ε/4 for c ∈ X, because V is
multiplicative and V (C)B is dense in B.
(β) of V was shown above from (β) of T
By assumption, V |C = δ∞ ◦ T satisfies (β).
Thus, there exists a contraction d1 ∈ B such that kd∗1 xd1 − aV (x)ak < ε/2 for
x ∈ eXe ∪ {e2 }, cf. proof of part (i).
Let d := ed1 ∈ B. Then kd∗ cd − aV (c)ak < ε for c ∈ X ∪ {1}.
(iv): See the argument at the end of the proof of Part (v) of Lemma 5.1.2. 

What are (α) and (β) in case T := 0 or

if T (c) := ρ(c) · 1 for a (pure) state ρ on C with ρ(C ∩ B) = {0} ?

Remark 5.4.2. Desirable !!! is the following version for countably generated
Hilbert D-modules and with a finite group action:
Let G a finite group, D a σ-unital C *-algebra with a G-action τD : G →
Aut(D), E a countably generated right Hilbert D-module with an isometric G-
action τE : G → Iso(E) into the linear isometries on E that is compatible with τD
in the sense that, for e ∈ E, g ∈ G, ??? and

τE (g)(e) · τD (g)(d) = τE (g)(e · d)

 4. A “TAUTOLOGIC” WEYL–VON NEUMANN TYPE RESULT 699

Now let C a separable C *-algebra with a G-action τC : G → Aut(C), and

suppose that Hk : C → L(E), k ∈ {1, 2}, are G-equivariant C *-morphisms.
If they are both in general position, and generate the same m.o.c cone, then
they should be unitarily homotopic ??

Corollary 5.4.3. Suppose that B is σ-unital and stable, A is separable, and

that H1 , H2 : A → M(B) are C*-morphisms, such that δ∞ ◦ Hk are non-degenerate
for k ∈ {1, 2}.
Then δ∞ ◦ H1 and δ∞ ◦ H2 are unitarily homotopic, if and only if, for every
contraction b ∈ B, Hj dominates weakly approximately inner the c.p. map b∗ Hk (·)b
for j, k ∈ {1, 2} in sense of Definition 3.10.1.
I.e., δ∞ ◦ H1 and δ∞ ◦ H2 are unitarily homotopic if, and only if, the (point-
norm closed) m.o.c. cones C(H1 ) and C(H2 ) in CP(A, B) corresponding to H1 and
H2 are identical.

Recall here that, by definition, the m.o.c. cone C(Hk ) ⊆ CP(A, B) is generated
by {b∗ Hk (·)b ; b ∈ B} and is point-norm closed.
We can weaken the requirement of non-degeneracy of δ∞ ◦ Hk by supposing
only that δ∞ ◦ Hk is unitarily homotopic to a non-degenerate C *-morphism from
A into M(B). Notice that this is a property of δ∞ ◦ Hk that is not the same as as
the property that δ∞ ◦ Hk non-degenerate.

Proof. The condition implies that H1 and H2 have the same kernel. Thus,
w.l.o.g. we can assume that H1 and H2 are faithful.
Let e ∈ B+ a strictly positive element of B, Fj := δ∞ (Hj (A)) and Cj :=
C ∗ (Fj , e) for j = 1, 2. Since Fj ∩ B = {0}, there is a unique C *-morphism Tj from
Cj onto Fi (for i 6= j ∈ {1, 2}) with Tj (e) = 0 and Tj (δ∞ (Hj (a))) = δ∞ (Hi (a)) for
a ∈ A.
2
The C *-morphism δ∞ ◦ Tj is unitarily equivalent to Tj , because δ∞ is unitarily
equivalent to δ∞ by Lemma 5.1.2(i) and Hj is uniquely defined up to unitary
equivalence.

Cj and Tj satisfy assumption (α) of Proposition 5.4.1 with h := e.

We show that (Cj , Tj ) satisfies also condition 5.4.1(β):
Let s1 , s2 , . . . a sequence of isometries in M(B) such that n sn (sn )∗ strictly con-
P

verges to 1M(B) , and let a ∈ B+ , X a finite subset of Cj and ε > 0 given. Then there
are finite subsets Y ⊆ A+ and Z ⊆ B such that X is a subset of {δ∞ (Hj (y))+z : y ∈
Y, z ∈ Z}. Our assumptions imply that there are b1 , . . . , bn ∈ B such that
X
ka(δ∞ )2 (Hi (y))a − b∗k Hj (y)bk k < ε/2 .
1≤k≤n

Since n sn (sn ) strictly converges to 1M(B) , there is m ∈ N such that kd∗ zdk <
∗
P

ε/2 for z ∈ Z and d := sm+1 b1 + . . . + sm+n bn . Thus kaVj (x)a − d∗ xdk < ε for
x = δ∞ (Hj (y)) + z in X.
700 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

By assumption, Tj (Cj )B = δ∞ (Hi (A))B is dense in B. Thus Proposition

5.4.1(iii) applies to (Cj , Tj ) and yields that H1 , H1 ⊕ H2 and H2 are mutually
unitarily homotopic. 

Corollary 5.4.4. Suppose that B is σ-unital and stable, A is separable, and

that S is a countably generated matrix operator-convex cone of completely positive
maps V from A into B (cf. Definition 3.2.2), such that S is non-degenerate in the
following sense: For every b ∈ B+ and ε > 0, there exists a ∈ A+ and V ∈ S such
that (b − ε)+ is in the closed ideal of B that is generated by V (a).
Then there is a non-degenerate C*-morphism H0 : A → M(B) with following
properties (i)–(iii):

(i) H0 is unitary equivalent to δ∞ ◦ H0 by a unitary in M(B).

(ii) For every V ∈ S there exists a sequence b1 , b2 , . . . ∈ B such that, for every
a ∈ A,
lim b∗n H0 (a)bn = V (a) .
n→∞

(iii) For every b ∈ B there exists a sequence V1 , V2 , . . . ∈ S such that, for every
a ∈ A,
lim Vn (a) = b∗ H0 (a)b .
n→∞

H0 is determined by (i)–(iii) up to unitary homotopy (cf. Definition 5.0.1).

H0 is injective, if and only if, S is separating for A, i.e., for every non-zero
positive a ∈ A+ \ {0}, there is V ∈ S with V (a) 6= 0 .

The example with B := A := O2 ⊗ K, S = CP(A, B), H0 (a) := a shows that

H0 is not determined by (ii) and (iii) of Corollary 5.4.4 alone.

Proof. The uniqueness up to unitary homotopy follows from Corollary 5.4.3.

Let a ∈ A+ . By (ii) and (iii), H0 (a) = 0 if and only if V (a) = 0 for every
V ∈ S. In particular, H0 is a monomorphism if and only if a ∈ A+ and V (a) = 0
for all V ∈ S imply a = 0.
By assumption there exists a countable sequence V1 , V2 , . . . ∈ S such that S is
contained in the point-norm closure of the matrix operator-convex cone generated
by {V1 , V2 , . . .}. We can suppose that every Vn appears infinitely times in this
sequence.
Kasparov–Stinespring dilations lead to Hilbert B-modules Hn and and C *-
morphisms Dn : A → L(Hn ) such that for b ∈ B and c ∈ A there is x ∈ Hn with
hx, Dn (a)xi = b∗ Vn (c∗ ac)b for a ∈ A, and that a ∈ A 7→ hy, Dn (a)yi ∈ B is in the
point-norm closure of the matrix operator-convex cone generated by Vn . (Recall
here that Hn is a completion of a quotient of the algebraic tensor product A B,
and that the B-valued Hermitian form is given there by
X
hx, yi := b∗j V (a∗j ck )dk
j,k
 4. A “TAUTOLOGIC” WEYL–VON NEUMANN TYPE RESULT 701

P P
for representatives j aj ⊗ bj of x and k ck ⊗ dk of y. We take it anti-linear in
the first variable.)
Let H := H1 ⊕ H2 ⊕ . . . and D : A → L(H) denote the Hilbert B-module sum,
respectively of the corresponding `∞ -sum of the C *-morphisms Dn .
Since every completely positive map Vn appears infinitely often in our sequence
(V1 , V2 , . . .), the maps Wy : a ∈ A 7→ hy, D(a)yi ∈ B for y ∈ H build a convex cone
S0 of completely positive maps from A into B, which is then automatically matrix
operator-convex. The map Wy is in the point-norm closure of S for every y ∈ H,
and every Vn is in the point-norm closure of S0 , i.e., S0 and S have the same point-
norm closure in CP(A, B). (Here we use the existence of approximate units in A
and B.)
Since Hn and (therefore) H are countably generated over B and since B is
σ-unital and stable, the Kasparov trivialization theorem leads to the Hilbert B-
module isomorphisms H ⊕ HB ∼ = HB ∼ = B.
Thus there is a (degenerate) C *-morphism h : A → M(B) that satisfies prop-
erties (ii) and (iii) (with H0 replaced by h).
Let B0 denote the closed span of δ∞ (h(A))Bδ∞ (h(A)), and let e ∈ B+ , f ∈
A+ strictly positive elements. B0 is a stable hereditary C *-subalgebra of B and
δ∞ (h(f ))eδ∞ (h(f )) is a strictly positive element of B0 , cf. Lemma 5.1.2(vii). B is
the closure of the linear span of Bh(A)B, because h satisfies property (ii) (with H0
replaced by h). Bh(A)B is contained in the closed span of BB0 B.
Thus, by a variant of the stable isomorphism theorem of L.G. Brown [107]
(cf. Corollary 5.5.6) there is an isomorphism ψ from B onto B0 ⊆ B, which is
approximately unitary equivalent to the idB by unitaries in M(B). H(a)b :=
ψ −1 (δ∞ (h(a))ψ(b)) defines a non-degenerate C *-morphism from A into M(B).
Since ψ is approximately unitary equivalent to idB , we get that that H satisfies
(ii) and (iii). Thus H0 := δ∞ ◦ H satisfies (i)–(iii) by Lemma 5.1.2(i). 

Remark 5.4.5. A “converse” of Corollary 5.4.4 holds obviously if A is sepa-

rable and B is σ-unital:
For every non-degenerate C *-morphism h : A → M(B), the set Sh of maps

Vh,b : a ∈ A 7→ b∗ δ∞ (h(a))b ∈ B

is a singly generated matrix operator-convex cone of completely positive maps, and

Vh,e is a generator of Sh if e is a strictly positive element of B. The corresponding
H0 is given by H0 := δ∞ ◦ h.

Corollary 5.4.6. Suppose that B is σ-unital and stable, A is separable

and unital, C ⊂ CP(A, B) a countably generated non-degenerate and faithful
point-norm closed operator-convex cone. Let H0 : A → M(B) the non-degenerate
*-monomorphism from A → M(B) as defined in Corollary 5.4.4. Further let
W : A → M(B) be a unital completely positive map with b∗ W (·)b ∈ C for all b ∈ B.
702 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

Then H0 (1) = 1, and there are isometries S, T ∈ M(B) with SS ∗ + T T ∗ = 1 and

W (a) − S ∗ H0 (a)S ∈ B for all a ∈ A.

Proof. Since A is unital and H0 is non-degenerate, we get H0 (1) = 1. Prop-

erty (i) in Corollary 5.4.4 implies H0 (A) ∩ B = {0}.
Let e ∈ B+ a strictly positive contraction, and let C := C ∗ (e, H0 (A)) ⊂ M(B).
Then C/(C ∩ B) ∼ = H0 (A). Thus, there is a unique *-morphism ψ : C → A with
ψ(e) = 0 and ψ ◦ H0 = idA . Let V (c) := W (ψ(c)) for c ∈ C. Then V is unital and
(C, h := e, V ) satisfy the condition (α) of Proposition 5.4.1.
We have C = (C ∩ B) + H0 (A). Let cj = bj + H0 (aj ) (j = 1, . . . , n) and
ε > 0, δ := ε/3. Since V ◦ H0 = W and since b∗ W (·)b ∈ C, there is d1 ∈ B with
kd∗1 H0 (aj )d1 − W (aj )k < δ for j = 1, . . . , n. By Lemma 5.1.2(iv) there exists

t 7→ S0 (t) ∈ δ∞ (M(B))0 ∩ M(B) ⊂ H0 (A)0 ∩ M(B)

with limt→∞ bS0 (t) = 0 for b ∈ B. There is t ∈ R+ with kd∗1 S0 (t)∗ bj S0 (t)(t)d1 k < δ
for j = 1, . . . , n. Thus kd∗ cj d − W (cj )k < ε for j = 1, . . . , n. Hence, for every
b ∈ B, the map c 7→ b∗ V (c)b ∈ B is approximately 1-step inner, i.e., condition (β)
of Proposition 5.4.1 is satisfied.
Thus C ⊂ M(B), h := e and V : C → M(B) satisfy the assumptions of
Proposition 5.4.1(i,iv). It gives an isometry S1 ∈ M(B) with S1∗ H0 (a)S1 − W (a) ∈
B for all a ∈ A.
By Lemma 5.1.2(iv), there exist isometries s, t ∈ H0 (A)0 ∩M(B) with ss∗ +tt∗ =
1. Let S := sS1 , then S ∗ H1 (·)S = S1∗ H1 (·)S1 , tt∗ ≤ 1 − SS ∗ and [1 − SS ∗ ] = [1] in
K0 (M(B)). By Lemma 4.2.6(ii), there is an isometry T with T T ∗ = 1 − SS ∗ . 

Definition 5.4.7. The quotient Qs (B) := M(B ⊗ K)/(B ⊗ K) is called stable

corona of the C *-algebra B.
If B itself is stable, i.e., B ∼
= B ⊗ K, we write instead Q(B) for M(B)/B ∼
=
s
Q (B). But the notation Q(B) for M(B)/B will also be used if B is non-unital
and not stable, i.e., it does not indicate that B is stable.

In Section 9 we define extension groups (depending on operator-convex cones

C ⊂ CP(A, B)) by

Ext(C ; A, B) := G(πB ◦ H0 ; A, Q(B))

where C is separating and satisfies the requirements of Corollary 5.4.4 and H0 : A →

M(B) is as in Corollary 5.4.4. The following Corollary 5.4.9 describes its elements.

Lemma 5.4.8. Let S ⊂ CP(A, J) a point-norm closed operator-convex cone,

T : A → C/J a c.p. contraction, and V : A → C a c.p. lift of T , i.e., πJ ◦ V = T ,
such that c∗ V (·)c ∈ S for every c ∈ J.
If A is separable, then there is a c.p. contraction W : A → C in the point-
norm closure of the operator convex hull C(V ) ⊂ CP(A, C) of V in CP(A, C) with
πJ ◦ W = T . In particular, c∗ V (·)c ∈ S for every c ∈ J.
 4. A “TAUTOLOGIC” WEYL–VON NEUMANN TYPE RESULT 703

If, in addition, A, C and T are unital, and if S is non-degenerate ( 12 ), then

there are a positive contraction e ∈ C+ and χ ∈ S such that such that χ(1) =
1 − eV (1)e ∈ J+ . Then W = χ(·) + eV (·)e is unital, πJ ◦ W = T and c∗ W (·)c ∈ S
for all c ∈ C.

Proof. Suppose that A is separable and that e ∈ A+ is a strictly positive

contraction. Let gδ (t) := min(tδ , t−1 ) for t ∈ (0, ∞) and gδ (0) := 0, where δ ∈ (0, 1].
For δ ∈ (0, 1] and a ∈ A + C · 1 let

Vδ (a) := gδ (V (e2δ )V (eδ aeδ )gδ (V (e2δ )) ,

and
Tδ (a) := (T (e2δ )δ T (eδ aeδ )(T (e2δ ))δ ,
Then πJ ◦ Vδ = Tδ , Vδ ∈ C(V ), and kVδ k = kVδ (1)k ≤ 1, because Vδ (1) = gδ (c)2 c
has norm ≤ 1 for c = V (e2δ ) ∈ C+ . Let D the hereditary C *-subalgebra of C/J
generated by T (e2 ), i.e., D is the closure of T (e2 )DT (e2 ). The span of eA+ e is
dense in A and 0 ≥ T (eae) ≤ T (e2 ). Thus T (A) ⊂ D, and T (e2 )δ ≤ T (e2δ )δ ∈ D+ ,
and fδ := T (e2δ )δ is an increasing approximate unit for D+ . Since eδ aeδ → a if
δ & 0 , it follows, that Tδ (a) → T (a) for a ∈ A if δ & 0 . Hence, T is in the
point-norm closure of the c.p. maps πJ ◦ Vδ where the Vδ ∈ C(V ) are contractions.
The set C(V )1 of contractions in C(V ) is an operator-convex set, in the sense
that a∗ V1 (·)a + b∗ V2 (·)b is a contraction in C(V ) for a, b ∈ C with ka∗ a + b∗ bk ≤ 1
and contractions V1 , V2 ∈ C(V )). Therefore one can “perturb” ( 13 ) the elements
of the sequence {V1/n } inside C(V )1 to a sequence Wn ∈ C(V )1 that converges in
point-norm to an element W ∈ CP(A, C) such that c∗ W (·)c ∈ S for every c ∈ C.

Suppose now that A, C and T are unital. Then 1 − g1 (V (1)) ∈ J+ and d1 :=

1 − V (1)g1 (V (1)) ∈ J+ . Let e1 := g1 (V (1))1/2 . There exists χ0 ∈ S with χ0 (1) ≥
(d1 − 1/4)+ . There is a contraction f ∈ J with f ∗ χ0 (1)f = (d1 − 1/2)+ , cf. Lemma
2.1.9. Let χ1 := f ∗ χ0 (·)f . Then χ1 ∈ S, χ1 (1) + e1 V (1)e1 = (d1 − 1/2)+ + 1 − d1 =:
h ≥ 1/2. Now let e := e1 h−1/2 ∈ C ∗ (d1 , 1) ⊂ C and χ := h−1/2 χ1 (·)h−1/2 ∈ S.
Then πJ (e) = 1 and W := χ + eV (·)e is unital with c∗ W (·)c ∈ S for all c ∈ J. 

Corollary 5.4.9. Suppose that A is separable, that B is stable and σ-unital,

and that ϕ : A → Q(B) = M(B)/B is a C*-morphism.
Let e ∈ B+ denote a strictly positive contraction in B and let S ⊂ CP(A, B)
countably generated non-degenerate operator-convex cone. Consider the non-
degenerate H0 : A → M(B) as defined in Corollary 5.4.4 for S, and let
h0 := πB ◦ (H0 ⊕ 0). Then:

(i) [h0 ] + [h0 ] = [h0 ],

(ii) The unitary equivalence class [ϕ] of ϕ is in S(h0 ; A, Q(B)) (= the semi-
group introduced in Section 4 of Chapter 4), if and only if, there is a

12I.e., for every e ∈ J and ε > 0 there are χ ∈ S and a ∈ A with (e − ε) ≤ χ(a)
+ + +
13 By “controlled” perturbations on a compact subset Ω ⊂ A with span(Ω) dense in A,

cf. e.g. the proof of the existence of nuclear lifts in [43].

704 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

completely positive map W : A → M(B) with πB ◦ W = ϕ such that

b∗ W (·)b in the point-norm closure of S for every b ∈ B ( 14 ).
(iii) [ϕ ⊕ 0] = [ϕ] + [0] ( 15 ) is in G(h0 ; A, Q(B)) = [h0 ] + S(h0 ; A, Q(B)) ,
( cf. Chapter 4, Section 4 ), if and only if, there is completely positive
contraction W : A → M(B) with πB ◦ W = ϕ such that
(a) b∗ W (·)b in the point-norm closure of S for every b ∈ B, and,
(b) for each V ∈ S, there is a sequence of elements d1 , d2 , . . . ∈ B such
that limn d∗n edn = 0 and limn d∗n W (a)dn = V (a) for all a ∈ A.

Proof. (i): The commutant h0 (A)0 ∩ Q(B) contains a copy of O2 unitally,

because H0 is unitarily equivalent to δ∞ ◦ H0 , δ∞ (M(B))0 ∩ M(B) contains a copy
of L(`2 ) unitally, and because L(`2 ) contains a copy of O2 ⊗ O2 unitally. Thus
the semigroup S(h0 ; A, Q(B)) is well-defined and is the set of unitary equivalence
classes (by unitaries in Q(B)) of C *-morphisms ψ : A → Q(B) dominated by h0
(cf. Definition 4.3.3).
(ii): Let D denote the point-norm closure of S.
Suppose that W : A → M(B) is a completely positive map with πB ◦ W = ϕ
such that b∗ W (·)b ∈ D for every b ∈ B. The c.p. lift W of ϕ can be modified that it
becomes contractive, cf. Lemma 5.4.8. It can be taken unital if A and ϕ : A → Q(B)
unital, because W 0 (a) := d1/2 H0 (a)d1/2 + cW (a)c is as desired, if c ∈ M(B)+ with
0 ≤ c ≤ 1, 1 − c ∈ B and d := 1 − cV (1)c ∈ B.
Let J ⊂ A denote the kernel of H0 : A → M(B) and let H1 = δ∞ ◦ H0 . H1
is unitarily equivalent to H1 (by condition (i) of Corollary 5.4.4) and, obviously,
πB ◦ H1 has the same kernel as H0 . W (J) = {0}, because b∗ W (·)b ∈ D for every
b ∈ B and H0 satisfies (ii). Thus, there is a completely positive contrcaction
[W ]J : A/J → M(B) with W = [W ]J ◦ πJ . Let C := C ∗ (e, H1 (A)), and define
V : C → M(B) by V (c) := [W ]J (Γ−1 (πB (c))), where Γ : A/J → πB (H1 (A)) is the
isomorphism from A/J onto πB (H1 (A)) with Γ(a + J) := πB (H1 (a)). Note that
C∩B = eCe , C = (C∩B)+δ∞ (H0 (A)) and that V (H1 (y)+z) = V (H1 (y)) = W (y)
for y ∈ A and z ∈ C ∩ B.
Then V (e) = 0 and e1/n d → d if n → ∞ for every d ∈ B. In particular,
V (C ∩ B) = {0}. If b ∈ B+ , ε > 0 and if X is a finite subset of C+ , then
there are f ∈ B and finite subsets Y ⊂ A, and Z ⊂ C ∩ B such that X ⊂
{H1 (y) + z : y ∈ Y, z ∈ Z} and kf ∗ H0 (y)f − bW (y)bk < ε/2 for y ∈ Y . The
latter holds by property (ii) of H0 in Corollary 5.4.4, because bW (·)b ∈ D. Since
δ∞ (g) = n sn gs∗n for all g ∈ M(B) with some sequence s1 , s2 , . . . of isometries
P

in M(B) with n sn s∗n strictly convergent to 1 in M(B), there is m ∈ N such

P

that kf ∗ s∗m zsm f k < ε/2 for z ∈ Z. It follows kd∗ xd − bV (x)bk < ε for x ∈ X and
d := sm f . Thus Proposition 5.4.1(i) applies: There is a contraction g ∈ M(B)
such that g ∗ cg − V (c) ∈ B for every c ∈ C. In particular, g ∗ H1 (a)g − W (a) ∈ B
14This particular c.p. lifts W of ϕ can be taken contractive. W can be chosen unital if A

and ϕ are unital.

15Note that, in general, [ϕ] + [0] 6= [ϕ] and [ϕ] + [0] 6= [ϕ] + [h ].
0
 4. A “TAUTOLOGIC” WEYL–VON NEUMANN TYPE RESULT 705

and πB (g)∗ πB (H1 (a))πB (g)∗ = ϕ(a) for a ∈ A. Since H0 is unitarily equivalent
to H1 by property (i) of Corollary 5.4.4, and since h0 := πB ◦ H0 ⊕ 0 (trivially)
dominates zero, it follows from Proposition 4.3.6(ii) that h0 dominates ϕ. This
means [ϕ] ∈ S(h0 , A, Q(B)).
Conversely, if ϕ is dominated by h0 = πB ◦ H0 ⊕ 0 in Q(B) = M(B)/B, then
there is a contraction g ∈ M(B) with πB (g ∗ H0 (·)g) = ϕ. For every b ∈ B, the
map W : a ∈ A 7→ b∗ g ∗ H0 (a)gb is in D by property (iii) of Corollary 5.4.4, i.e.,
W := g ∗ H0 (·)g is as desired.
We have seen that (i) is equivalent to [ϕ] ∈ S(h0 , A, Q(B)).
But [ϕ] ∈ S(h0 , A, Q(B)) implies [ϕ ⊕ 0] = [ϕ] + [0] ∈ S(h0 , A, Q(B)), because
(trivially) [0] ∈ S(h0 , A, Q(B)) and S(h0 , A, Q(B)) is a semigroup by Proposition
4.4.2(i). Since ϕ ⊕ 0 trivially dominates ϕ, [ϕ] + [0] ∈ S(h0 , A, Q(B)) if and only if
[ϕ] ∈ S(h0 , A, Q(B)).
By Proposition 4.3.5(i), [ϕ]+[0] ∈ G(h0 , A, Q(B), i.e., [ϕ]+[0] = [ϕ]+[0]+[h0 ],
if and only if ϕ ⊕ 0 dominates h0 .
Suppose that [ϕ]+[0] is in G(h0 , A, Q(B)). Then there are contractions g1 , g2 ∈
M(B) such that πB (g1∗ H0 (a)g2 ) = ϕ(a) and g2∗ g1∗ H0 (a)g1 g2 − H1 (a) ∈ B for all
a ∈ A and H1 := δ∞ ◦ H0 , because H1 is unitarily equivalent to H0 . Let W (a) :=
g1∗ H0 (a)g1 , then W satisfies (i), as we have seen above.
Let V ∈ D, there is a sequence of elements b1 , b2 , . . . ∈ B such that
limn b∗n H0 (a)bn = V (a) for all a ∈ A (by property (ii) of H0 in Corollary
5.4.4). One can see, with help of an approximate unit of A, that one manage that
kbn k2 ≤ kV k. Then dn := g2 sn bn satisfies limn d∗n W (a)dn = V (a) for all a ∈ A and
limn kd∗n edn k ≤ kV k lim ks∗n (g2∗ eg2 )sn k = 0, i.e., W satisfies also (ii).
Conversely, let W : A → M(B) a contractive and completely positive lift of ϕ
with (i) and (ii).
Then [ϕ ⊕ 0] ∈ S(h0 , A, Q(A)) by (i), as we have seen above. Since h0 and ϕ ⊕ 0
both dominate zero and since H0 is unitarily equivalent to H1 = δ∞ ◦ H0 , it suffices
to find a contraction g ∈ M(B) with g ∗ W (a)g − H0 (a) ∈ B for all a ∈ A, to get
that ϕ ⊕ 0 dominates h0 (cf. Proposition 4.3.6(ii)).
There is an isomorphism ψ from ϕ(A) onto H0 (A) such that ψ(ϕ(a)) = H0 (a)
for a ∈ A, because W (a) ∈ B if and only if H0 (a) = 0: Indeed, above we have
seen that H0 (a) = 0 implies H0 (a∗ a) = 0 and 0 ≤ W (a)∗ W (a) ≤ W (a∗ a) = 0.
W (a) ∈ B implies ϕ(a∗ a) = 0, i.e., W (a∗ a) ∈ B, and, by (ii), that V (a∗ a) = 0 for
all V ∈ D, i.e., H0 (a) = 0 by property (iii) of H0 in Corollary 5.4.4.
Let C := C ∗ (e, W (A)) and V (c) := ψ(πB (c)). Then C is separable, C ∩ B =
eCe, C = (C ∩ B) + W (A), V (e) = 0, limn e1/n b = b for b ∈ B, V (C ∩ B) = {0},
V (W (a)) = H0 (a) for a ∈ A, and δ∞ ◦ V is unitarily equivalent to V , because
H0 is unitarily equivalent to H1 := δ∞ ◦ H0 . Moreover, V : C → M(B) is a non-
degenerate C *-morphism.
706 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

If b ∈ B+ , X ⊂ A and Y ⊂ C ∩ B are finite and ε > 0, then there is d ∈ B

with kd∗ ydk < ε/2 and kd∗ V (x)d − bH0 (x)bk < ε/2 by property (ii) of W , because
bH0 (·)b ∈ D. Thus, Proposition 5.4.1 applies to C and V : There is a contraction
g ∈ M(B) with g ∗ cg − V (c) ∈ B for all c ∈ C, in particular g ∗ W (a)g − H0 (a) ∈ B
for all a ∈ A. 

Corollary 5.4.10. Suppose that A is separable, B is σ-unital, D ⊂ M(B) is

a
Is ‘‘simple’’ necessary?
simple, stable and σ-unital C*-subalgebra of M(B) such that DB is dense in
B.
Let H : A → M(D) ⊂ M(B) a non-degenerate *-monomorphism with H(A) ∩
D = {0} such that Hd : A 3 a 7→ d∗ H(a)d ∈ D is nuclear for all d ∈ D.
Then

(i) H0 := δ∞ ◦ H satisfies properties (i)–(iii) of Corollary 5.4.4 for the

operator-convex cone S := CPnuc (A, B).
(ii) H and H0 are unitarily homotopic if D ∼ = K or if D is purely infinite.

Proof. (i): Let s1 , s2 , . . . a sequence of isometries in M(D) such that

sn (sn )∗ converges strictly to 1. Then H0 is unitarily equivalent to H1 given
P

sn H(a)s∗n , and H1 is unitarily equivalent to δ∞ ◦ H1 , cf. Lemma

P
by H1 (a) :=
5.1.2(i). Let qm := n≤m sn s∗n , e ∈ D+ strictly positive, b ∈ B. Then
P
X
a ∈ A → b∗ e1/n qm H1 (a)qm e1/m b = b∗ e1/m sk H(a)s∗k e1/m b
1≤k≤m

is nuclear, because H : A → M(D) is weakly nuclear. It implies that Vb := b∗ H1 (·)b

is a nuclear map from A into B, because limm kqm e1/m b − bk = 0 for b ∈ B. Hence
H0 = U ∗ H1 (·)U satisfies condition (i) and (iii) of Corollary 5.4.4 for the m.o.c. cone
S := CPnuc (A, B).
The proof of Corollary 5.4.4 shows: Since H1 is unitarily equivalent to δ∞ ◦ H1
and is non-degenerate, we get that the maps Vb : a 7→ b∗ H1 (a)b build an operator-
convex cone C ⊂ CPnuc (A, B) (that is generated by Vf for a strictly positive element
f ∈ B+ ).
Let a ∈ A+ with kak = 1, b ∈ B+ and ε > 0. Then there is m ∈ N such
that ke1/m H(a)e1/m k > 1/2, and kb(1 − e1/m )bk < ε/2, because kH(a)k = 1
and because DB is dense in B. Let c := e1/m H(a)e1/m ∈ D. Since D is
P ∗
simple, there are f1 , . . . , fk ∈ D such that ke1/m − fj cfj k < ε/2 . Now let
d := 1≤j≤k s∗j e1/m fj b ∈ B. Then kd∗ H1 (a)d − b2 k < ε. Thus, Lemma 3.2.7 ap-
P

plies, and every nuclear map V : A → B is in the point-norm closure of C. Hence,

H0 satisfies condition (ii) of Corollary 5.4.4 for S := CPnuc (A, B).
sn as∗n is (up to unitary equivalence) independent
P
(ii): Since δ∞ (H(a)) =
sn s∗n = 1 (strictly), and since
P
from the sequence of isometries s1 , s2 , . . . with
M(D) → M(B) is unital and strictly continuous, it suffices to consider the case,
 4. A “TAUTOLOGIC” WEYL–VON NEUMANN TYPE RESULT 707

where B = D. Let e ∈ D+ a strictly positive contraction, and let C := C ∗ (H(A), e),

e ∈ C ∩ B =: J / C and there is a (unique) *-morphism ψ : C → A with ker(ψ) =
C ∩ B and ψ ◦ H = idA . Let V := H0 ◦ ψ = δ∞ ◦ T (for T := H ◦ ψ). Then
V : C → M(D) is a non-degenerate *-morphism with V (e) = 0. The map V is
weakly nuclear, because H0 : A → M(B) is weakly nuclear.
For b ∈ D = B the map Vb : c ∈ C 7→ b∗ V (c)b ∈ D is nuclear and Vb (C ∩ D) =
{0}. It follows from Proposition 3.2.13(i) that Vb is approximately 1-step inner if
D is purely infinite.
If D ∼
= K, then [Vb ] : C/(C ∩ D) → D is necessarily nuclear. Then the map Vb
is approximately 1-step inner by Remark 3.2.14. It follows that (C, V, e, B = D)
satisfy also condition (β) of Proposition 5.4.1 in both cases. Then H0 = δ∞ ◦ H =
H ⊕ δ∞ ◦ H is unitarily homotopic to H by Proposition 5.4.1(iii). 

QUESTIONS:
Suppose A separable (unital or stable), B σ-unital and stable.
Let H0 : A → Q(B) a C *-morphism, e.g. is defined by the universal bi-module
of all nuclear maps form A to B .
Is it next?
If we use that B ⊗ K ∼ = B, then this is given by H0 := πB ◦ δ∞ ◦ (1 ⊗ h) the
infinite repeat of any faithful representation h : A → M(K).
Now describe all H : A → Q(B) that absorb H0 . ...???
Define S 0 (H) as the set of “u.c.p. maps V : A → B” (??) with the property
that there exists a sequence of contractions in Tn ∈ Q(B) such that

πB (δ∞ ◦ V (a)) − Tn∗ Ho (a)Tn → 0 .

Let W : A → M(B) a c.p. contraction.

Does H dominate π ◦ W if b∗ W (·)b ∈ C(S 0 (H)) for all b ∈ B ?
The point is:
What happens with infinite strictly convergent sums
X
T (·) = b∗n Vn (·)bn
n

with Vn ∈ C(S 0 (H)) ?

Cases A = C, A = C0 (0, 1]?
What happens in unital/non-unital cases ?
Check next blue text again!!

Remark 5.4.11. G. Elliott and D. Kucerovsky, [264], have used – together

with their own proof – the following older version of Proposition 5.4.1:
708 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

Let C be a unital separable C *-algebra and let B be an essential closed two-

sided ideal of C, so that we may view C as a unital subalgebra of M(B):

B ⊆ C ⊆ M(B); 1∈C.

Let T : C → M(B) be a completely positive map which is zero on B, and suppose

that, for every b ∈ B, the map

b∗ T (·)b : C → B,

given by c 7→ b∗ T (c)b, can be approximated (on finite subsets of C) by the maps

c 7→ d∗ cd, d ∈ B .

It follows that there exists a sequence Sn ∈ M(B) such that

T (c) − Sn∗ cSn ∈ B, for all, c ∈ C ,

and limn Sn∗ cSn = T (c) for all c ∈ C.

This can be seen from Proposition 5.4.1, if one takes for the element h ∈ C+
in Condition (α) of Proposition 5.4.1 a strictly positive contraction of B. The
approximation condition implies Condition (β).

Comments to it: Find out the ingredients!!

Let B a σ-unital C *-algebra, C a separable C *-subalgebra of M(B), and
T : C → M(B) a linear map with kT k ≤ 1 that satisfies the following Conditions
( α) and ( β):

(α) There exists h ∈ C+ with T (h) = 0 and h1/n d → d if n → ∞ for every

d∈B.
(β) For every a ∈ B+ , every finite subset X ⊂ C+ of contractions and every
ε > 0 there exists d ∈ M(B) with kd∗ cd − aT (c)ak < ε for c ∈ X.

Question 5.4.12. Let A separable, B stable and σ-unital, C ⊆ CP(A, B) a

countably generated operator convex cone, e.g. C := CPnuc (A, B) (which is singly
generated).
Is there a characterization of those H : A → Q(B) that absorbs each πB ◦ δ∞ ◦
h : A → Q(B) with h : A → M(B) that satisfies b∗ h(·)b ∈ C for all b ∈ B?
In the case of [264] it is
In case of C := CPnuc (A, B) a necessary and sufficient criteria for absorbing
was given by G. Elliott and D. Kucerovsky [264], cf. [310] for the non-unital case.
Has to consider as C ⊂ M(B) some separable C *-subalgebra of M(B) with
the properties that πB (C) = H(A) and that C ∩ B contains e ∈ C ∩ B that is
strictly positiv in B. Then use B ∼
= B ⊗ K and take a faithful representation
ρ : A → M(K) ⊆ M(B ⊗ K) in “general position”, i.e., ρ unitary equivalent to its
infinite repeat.
To get absorbed it is necessary that H is faithful.
 5. Ψ-EQUIVARIANT STABILITY OF EXTENSIONS 709

Then let T (c) := ρ(H −1 πB (c)) . Let h := e for (α) a strictly positive element
of B ⊗ K .
Check (β) ... for C, T ...

5. Ψ-equivariant Stability of extensions

We deduce from Proposition 5.4.1 some results concerning stability and a Ψ-

equivariant version (respectively m.o.c. cone equivariant version) of L.G. Brown’s
stable isomorphism theorem [107, thm.2.8].
Recall that a hereditary C *-subalgebra D of a C *-algebra B is full if the linear
span of BDB is dense in B, and that D is called a corner if there exists a (unique)
projection p ∈ M(B) with D = pBp.
The following Corollary of the tautological Weil–von-Neumann type result
Proposition 5.4.1 gives an alternative proof of the stability criteria for σ-unital
C *-algebras of Hjelmborg and Rørdam [373, thm. 2.1, prop. 2.2(b)].

Corollary 5.5.1. Suppose that B is a non-zero σ-unital and stable C*-

algebra, and that D is a full corner of B, i.e., there is a projection p ∈ M(B)
with pBp = D and span BDB is dense in B.

(i) If, for every g ∈ D+ and δ > 0, there exist b ∈ D with kb∗ b − gk < δ, and
kb∗ gbk < δ, then there exists an isometry S ∈ M(B) with SS ∗ = p.
(ii) The isometry S of Part (i) defines an isomorphism ϕ from B onto D by
ϕ(b) := SbS ∗ for b ∈ B, and ϕ satisfies
ϕ(J) = D ∩ J for all J ∈ I(B) .
(iii) If S1 , S2 ∈ M(B) are isometries in M(B) such that the maps ϕk :=
Sk ( · )Sk∗ , k ∈ {1, 2}, from B onto D ⊆ B coincide (i.e., ϕ1 = ϕ2 ), then
there exists a unitary U in the center of M(B) with S1 · U = S2 .
(iv) The isometries S ∈ M(B) with SBS ∗ = D can be obtained from the polar
decompositions d := S(d∗ d)1/2 = (dd∗ )1/2 S, of any element d ∈ B with
the property that d∗ d is strictly positive in B and dd∗ is a strictly positive
element of D.

Proof. The idea of our proof is different from that in [373], because our
proof is based on the fact that full properly infinite projections are MvN-equivalent
if they have the same class in K0 (an observation of J. Cuntz, cf. proof of Lemma
4.2.6(ii)). We combine this with the “tautological” Weyl–von-Neumann type re-
sult in Proposition 5.4.1 to get Part (i). The other parts follow then by straight
calculation.
(In fact all this conclusions of Proposition 5.4.1 are almost trivial elementary
applications, consisting only in notational problems.)
(i):
TO BE SHOWN:
710 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

Let p ∈ M(B) a projection, and let D := pBp. Suppose that span BDB
is dense in B, and that for every g ∈ D+ and δ > 0, there exist b ∈ D with
kb∗ b − gk < δ, and kb∗ gbk < δ.
Then there exists an isometry S ∈ M(B) with SS ∗ = p.
Recall that K∗ ( M(B) ) = 0 because δ∞ is unitarily equivalent to δ∞ ⊕ id,
cf. Lemma 5.1.2(ii).
By Lemma 4.2.6(ii), it suffices to show that there is an isometry I in M(B) with
∗
II ≤ p, because then p and q := 1 are both projections that majorize orthogonal
ranges of two isometries in M(B) and have same class [p] = [q] in K0 (M(B)) = 0.
By Lemma 4.2.6(ii), the projections p and q = 1 are MvN-equivalent in M(B), i.e.,
there is an isometry S ∈ M(B) with SS ∗ = p.
(In fact, the existence of an operator S0 ∈ M(B) with x := 1 − S0∗ pS0 ∈ B
would be enough, because the stability of B causes the existence of an isometry
S1 ∈ M(B) with kS1∗ xS1 k ≤ 1/2. Then I := pS0 S1 (S1∗ S0∗ pS0 S1 )−1/2 is an isometry
with II ∗ ≤ p.)
We apply Proposition 5.4.1(i) to prove the existence of needed isometry I:
Take a strictly positive contraction e ∈ B+ and let C denote the unital com-
mutative C *-subalgebra of M(B), which is generated by the commuting elements
1, p, pep and (1 − p)e(1 − p). The intersection C ∩ B contains the strictly positive
contraction h := pep + (1 − p)e(1 − p) = 2−1 ((1 − 2p)e(1 − 2p) + e) of B.
The projection p ∈ M(B) is not contained in B, because otherwise p ∈ pBp =
D and the assumptions in part (i) propose the existence of b ∈ D (i.e., b ∈ pBp)
with pb = b = bp, kb∗ pbk < 1/4 and kb∗ b − pk < 1/4, which is impossible for a
non-zero projection p.
Let χ denote the (unique) character of C with χ(C ∩ B) = 0 and χ(p) = 1. It
satisfies χ(y) = χ(pyp) for all y ∈ C, in particular, χ((1 − p) + pep) = 0.
The map T (a) := χ(a)1 is a C *-morphism from C into M(B). Obviously,
T (p) = T (1) = 1 and T has the same kernel T −1 (0) = C ∩ B + C · (1 − p) as
χ : C → C has. In particular T (h) = 0 for the strictly positive element h of B. The
strict positivity implies that b = lim h1/n b for all b ∈ B.
Thus, C, T and h satisfy condition (α) of Proposition 5.4.1.
(β): For every a ∈ B+ , every finite subset X ⊂ C+ and every ε > 0 there
exists d ∈ M(B) with kd∗ cd − aT (c)ak < ε for c ∈ X.
Part to be checked:
We show that T satisfies also condition (β) of Proposition 5.4.1 :
Let a ∈ B+ , X a finite subset of C+ for C := C ∗ (1, p, pep, (1 − p)e(1 − p)) and
ε > 0. We define γ := max{kck : c ∈ X} and δ := ε/(4 + 4γ).
Notice that pCp = C ∗ (pep) + C · p, because c = pcp + (1 − p)c(1 − p) for all
c ∈ C.
 5. Ψ-EQUIVARIANT STABILITY OF EXTENSIONS 711

The kernel of T is identical with the kernel (1 − p)C + p(C ∩ D) of χ.

It follows that χ(C ∩ D) ⊆ χ(C ∩ B) = {0} and χ(1 − p) = 0
In particular pcp − χ(c)p is in C ∩ D for c ∈ C+ .
Since c = pc+(1−p)c = pcp+(1−p)c(1−p) for all c ∈ C and since χ(1−p) = 0
and χ(C ∩ B) = {0}, we get that
pcp − χ(c)p is in C ∩ D for c ∈ C+ . Thus, if b ∈ pB, then, for c ∈ C+ ,

b∗ cb = χ(c)b∗ b + b∗ (pcp − χ(c)p)b .

Since pcp − χ(c)p is selfadjoint,

kb∗ (pcp − χ(c)p)bk ≤ kb∗ |(pcp − χ(c)p)|bk .

We consider the sum f of the absolute values |pcp − χ(c)p|:

X
f := |pcp − χ(c)p| ∈ C ∩ D+ . (5.1)
c∈X

Then, for c ∈ X and b ∈ pB,

kb∗ cb − aV (c)ak ≤ |χ(c)|kb∗ b − a2 k + kb∗ f bk .

Thus, to verify (β), it suffices to find b ∈ pB with kb∗ b − a2 k < 2δ and kb∗ f bk < 2δ
for the f defined in Equation (5.1).
Since D is full, there exists b1 , . . . , bm ∈ pB with k b∗j bj − a2 k < δ. Let
P

ρ := δ/m(1+kak2 +δ). There is a contraction g ∈ D+ , such that kb∗j gbj −b∗j bj k < ρ
for j = 1, . . . , m. By induction, the assumption on D implies that, for every
f, g ∈ D+ , m ∈ N, and ρ > 0, there exist d1 , . . . , dm ∈ D, such that kd∗i dj −δij gk < ρ
and kd∗i f dj k < ρ for i, j = 1, . . . , m. Let b := dj bj . Then kb∗ f bk < m·(kak2 +δ)·ρ
P

and kb∗ b − b∗j bj k < m · (1 + kak2 + δ) · ρ. Thus kb∗ f bk < δ, kb∗ b − a2 k < 2δ, and
P

V = δ∞ ◦ T satisfies also condition (β) of Proposition 5.4.1.

We are now in position to apply Parts (i) and (iv) of Proposition 5.4.1, and
find an element S1 in M(B), such that kS1∗ pS1 − 1k < 1.
The element I := pS1 (S1∗ pS1 )−1/2 is an isometry in M(B) with II ∗ ≤ p.
Thus p is full and properly infinite in M(B) .
(ii): The isometry S ∈ M(B) with SS ∗ = p defines obviously an injective
C *-morphism ϕ(b) := SbS ∗ (b ∈ B) from B into D = pBp. It is surjective because
ϕ(SdS ∗ ) = d for d ∈ D.
The hereditary C *-subalgebra D of B is full in B. Therefore, the map J ∈
I(B) 7→ D ∩ J ∈ I(B) is a bijective map. Since S ∈ M(B) we get that SJS ∗ ⊆
D ∩ J. If d ∈ D ∩ J, then S ∗ dS ∈ J and ϕ(S ∗ dS) = d. Thus, ϕ(J) = D ∩ J.
(iii):
If ϕ(b) = S1 bS1∗ = S2 bS2∗ for isometries S1 , S2 ∈ M(B) and b ∈ B, then we
obtain that S1 S1∗ = S2 S2∗ =: p from S1 eS1∗ = S2 eS2∗ for for e ∈ B in an approximate
unit of B. Thus, U := T ∗ S ∈ M(B) is a unitary with U b = bU for all b ∈ B. Now
712 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

use that B is strictly dense in M(B) the unitary U is in the center of M(B).
Clearly T U = S = U T .
(iv): If d ∈ B has the property that d∗ d is strictly positive in B and dd∗ is
a strictly positive element of D, then the partial isometry v ∈ B ∗∗ in the polar
decomposition with d := v(d∗ d)1/2 = (dd∗ )1/2 v satisfies, – by definition of the
unique polar-decomposition in the W*-algebra B ∗∗ –, that v ∗ v is the open support
projection = 1 of d∗ d in B ∗∗ and vv ∗ is the support projection p of dd∗ in B ∗∗ .
Thus, vv ∗ = p, v ∗ v = 1.
Above formulas imply B = (d∗ d)1/2 B , dB = pB and B(dd∗ )1/2 = Bp. Using
this equations, then we get vB = dB = pB ⊆ B from d = v(d∗ d)1/2 , and Bv =
Bpv = Bd ⊆ B from (dd∗ )1/2 v = d.
Thus, v is an isometry in M(B) and can deserve as one of the above considered
S ∈ M(B). 

(StC) looks strange? Better notation?

Corollary 5.5.2 ([373]). A non-zero σ-unital C*-algebra A is stable, if and

only if, satisfies the following “stability criterium” (StC):

(StC) For every a ∈ A+ and ε > 0 there exists d ∈ A such that ka − d∗ dk < ε
and kd∗ adk < ε.

Proof. Let a ∈ A+ and ε > 0. If A is stable and t1 , t2 , . . . is a sequence

P ∗
of isometries in M(A) with tk tk strictly convergent to 1M(A) , then there exists
n ∈ N with ktn atn k < ε/(1 + kak) . If we let d := tn a1/2 , then d∗ d = a and
∗

kd∗ adk < ε.

Conversely suppose that each for a ∈ A+ and ε > 0, there exists d ∈ A with
kd adk < ε and kd∗ d − ak < ε. Let B := A ⊗ K and D := A ⊗ e11 .
∗

Then the corner D of the σ-unital stable C *-algebra B satisfies the assumptions
of Corollary 5.5.1. Thus, A is isomorphic to the stable algebra B. 

If A is σ-unital, then A/J is σ-unital. If A1 ⊆ A2 ⊆ · · · ⊆ A are σ-unital

S P −n
C *-subalgebras with n An dense in A, then A is σ-unital: e = n2 en is
strictly positive in A if en is strictly positive in An . This together implies that
indlim(hn : An → An+1 ) is σ-unital if each An is σ-unital.
It is easy to see that A satisfies the criteria of Corollary 5.5.2 if A is the inductive
limit of an (arbitrary upward directed) family of C *-algebras Aτ that satisfy the
criteria (STC) if the inductive limit is still σ-unital, e.g. if the directed diagram of
morphisms hn is countable. Thus we get:

Corollary 5.5.3 ([373]). The inductive limit of a sequence of σ-unital stable

C*-algebras is σ-unital and stable.

A generalization of the Zhang’s dichotomy [846] for simple purely infinite σ-

unital C *-algebras, is the following, cf. [462, thm. 4.24].
 5. Ψ-EQUIVARIANT STABILITY OF EXTENSIONS 713

Corollary 5.5.4 (Generalized Zhang dichotomy). A σ-unital purely infinite

C*-algebra B is stable if and only if B has no unital quotient.

Proof. Quotients of stable C *-algebras are zero or stable. Thus B can not
have a unital quotient if B is stable.
For the non-trivial direction suppose that B is p.i. and has no unital quotient.
Let 0 6= a ∈ B+ , 0 < ε < 1 and δ = ε/(1 + kak).
The closed ideal J of B which is generated by the non-zero (!) annihilator
D := Ann((a − δ)+ , B) of (a − δ)+ in B is equal to B, because – otherwise – B/J
is unital, Spec(πJ (a)) ⊂ {0} ∪ [δ/2, kak] and πJ (a) is a strictly positive element of
B/J, cf. Lemma 2.5.14(i,ii) for more details. Since, by assumption, B has no unital
quotient, it follows that D = Ann((a − δ)+ , B) is a full hereditary C *-subalgebra of
B. Thus, there are c1 , . . . , cn ∈ D+ and e1 , . . . , en ∈ B with ka − e∗k ck ek k < δ/2.
P

By the definition of p.i. algebras there is f ∈ B with ka − f ∗ cf k < δ, where c :=

c1 +. . .+cn ∈ D+ . Let d := c1/2 f . Then kd∗ d−ak < δ < ε, d∗ ad = d∗ (a−(a−δ)+ )d
and kd∗ adk ≤ δ(kak + δ) < ε. Thus, B satisfies the criteria (STC) of Corollary 5.5.2
and is stable. 

Remark 5.5.5.
(i) Corollary 5.5.4 implies Zhang’s dichotomy [846] for simple purely infinite alge-
bras:
A σ-unital simple purely infinite C*-algebra is unital or is stable.
(ii) Proposition 2.2.1, Brown’s stable isomorphism theorem and the dichotomy to-
gether immediately imply that every hereditary C *-subalgebra D of a simple purely
infinite C *-algebra A has an approximate unit consisting of projections. By a theo-
rem of Brown and Pedersen, [113], this property is equivalent to the property that
the real rank of A is zero, i.e., every selfadjoint element can be approximated by
selfadjoint elements of A with finite spectra.
Thus every simple purely infinite C*-algebra has real rank zero. See also the
proof of Proposition 2.2.1(x).
(iii) We don’t use for our results on classification that simple purely infinite algebras
A have zero real rank. (rr(A) = 0 can be seen also by the asymptotic methods in
Chapter 7 or immediately from Theorem B.)

The following Corollary 5.5.6 is implicitly contained in the proofs of [107]. more
precise ref We derive it from Corollary 5.5.1.

Corollary 5.5.6 (Ψ-equivariant Brown stable isomorphism). Suppose that

D is a stable σ-unital hereditary C*-subalgebra of a stable σ-unital C*-algebra B,
such that BDB is dense in B.
Then there exists d ∈ B, such that

(i) d∗ d is a strictly positive element of B,

(ii) dd∗ is a strictly positive element of D,
714 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

(iii) the polar decomposition v(d∗ d)1/2 of d in B ∗∗ defines an isomorphism

ϕ := v(·)v ∗ from B onto D ⊂ B, such that
(iv) ϕ : B → B is unitarily homotopic to idB , in particular ϕ(J) = D ∩ J for
J ∈ I(B).

It follows: if D1 is a σ-unital hereditary C*-subalgebra of a σ-unital C*-

algebra B1 , which is not contained in a closed ideal J 6= B1 , then there exists a
*-isomorphism ψ from D1 ⊗ K onto B1 ⊗ K such that, for J ∈ I(B1 ),

ψ((D1 ∩ J) ⊗ K) = J ⊗ K.

Proof. Let t and s isometries in M(B) with t∗ s = 0, and let E denote the
hereditary C *-subalgebra of B which is generated by F := sDs∗ +tBt∗ . Then id |F
is a non-degenerate *-monomorphism from F into E. It extends to a unital strictly
continuous *-monomorphism H := M(id |E) from M(F ) ∼ = M(sDs∗ ) ⊕ M(tBt∗ )
into E. It follows that sDs∗ and tBt∗ generate a corner of E. But both are
hereditary C *-subalgebras of B, which generate B as closed ideal.
Thus, sDs∗ and tBt∗ are corners of E that generate E as closed ideal.
F is stable, because sDs∗ and tBt∗ are stable, indeed: If s1 , s2 , . . . is a se-
∗
P
quence of isometries in M(F ), such that n sn sn converges strictly to 1, then
H(s1 ), H(s2 ), . . . has the same property for E. Thus, E is stable by Remark
5.1.1(8).
Since sDs∗ ∼ = D, tBt∗ and and E are stable, Corollary 5.5.2 implies that the
pairs sDs∗ ⊂ E and tBt∗ ⊂ E satisfy the assumptions of Corollary 5.5.1 (in place
of the pair D ⊂ B). It follows, that there are elements d1 , d2 ∈ E, such that d∗1 d1
and d∗2 d2 are strictly positive elements of E, d1 d∗1 is a strictly positive element of
sDs∗ and d2 d∗2 is a strictly positive element of tBt∗ . Then d := s∗ d1 d∗2 t satisfies
that d∗ d is a strictly positive element of B and dd∗ is a strictly positive element of
D. Let v(d∗ d)1/2 be the polar decomposition. By Remark 2.3.1, ϕ(a) = vav ∗ is an
isomorphism from B onto D, and ϕ−1 : D → B is given by a 7→ v ∗ av.
By Lemma 5.1.2(iv) there exists a norm-continuous map t ∈ [1, ∞) 7→ S0 (t)
into the isometries in M(B) with limt→∞ S0 (t)b = 0 for b ∈ B. Let G :=
Cb ([1, ∞), M(B))/ C0 ([1, ∞), B) ⊃ B. Then M(B) ⊂ G naturally and G con-
tains a copy of O2 unitally. Let h1 , h2 : B → E given by h1 (b) := b ∈ F ??? or G
and h2 (b) := h1 (ϕ(b)) for b ∈ B. Now consider the isometry s := {S0 (t)}1 ≤ t <∞ +
C0 (R+ , B) in G which corresponds to S0 ∈ Cb ([0, ∞), M(B)), and the contraction
y ∈ G with representative Y (t) := d∗ (dd∗ + 1/t)−1/2 for t ∈ [0, ∞).
Then s∗ hj (·)s = 0 for j = 1, 2, y ∗ h1 (·)y = h2 and yy ∗ h1 (·)yy ∗ = h1 Thus
Proposition 4.3.6(iii) applies: there is a unitary u in F with u∗ h1 (·)u = h2 .
The unitary u comes from a continuous map U (t) from [0, ∞) into the unitaries
of M(B) such that u = {U (t)}t∈[0,∞) + C0 (R+ , B), cf. proof of Lemma 5.1.2(vi).
Since the unitary group of M(B) is (norm-) connected ([180]), one can choose
such that t 7→ U (t) such that U (0) = 1. Clearly, u∗ h1 (·)u = h2 means that
 5. Ψ-EQUIVARIANT STABILITY OF EXTENSIONS 715

limt→∞ kU (t)∗ bU (t) − ϕ(b)k = 0 for b ∈ B, i.e., that idB and ϕ are unitarily
homotopic.
It follows ϕ(J) ⊂ D ∩ J and ϕ−1 (D ∩ J) ⊂ J, which means ϕ(J) = D ∩ J for
J ∈ I(B). 

next Lemma is the old Lemma 7.15,

which is now n o t cited in Chapter 7

Lemma 5.5.7. Suppose that A is a full hereditary C*-subalgebra of B (i.e.,

ABA = A and span(BAB) = B), and that A and B are both σ-unital and stable.
Then there exist x in B and an isomorphism ϕ from B onto A such that x∗ x is
strictly positive in B, xx∗ is a strictly positive element in A and ϕ(b) = ubu∗
where u can be chosen as any contraction in B ∗∗ that satisfies u∗ x = (x∗ x)1/2 and
xu∗ = (xx∗ )1/2 .
Moreover, ϕ : B → A ⊂ B is unitarily homotopic to idB .

Proof. The proof is implicitly contained in the proofs of Corollaries 5.5.1 and
5.5.6:
It is enough to find an x ∈ B with x∗ x strictly positive in B and xx∗ strictly
positive in A. We proceed as in [107]. Let D ⊂ M2 (B) be the hereditary C *-
subalgebra which is generated by diag(e, f ) where e is a strictly positive element
of B and f is a strictly positive element of A. Then F = diag(A, B) ∼ = B⊕A
is naturally a C *-subalgebra of D such that D is the closure of F DF . It follows
that M(F ) is unitally and strictly continuous contained in M(D). F is stable
because A and B are stable. Thus there exists a copy C of K in M(F ) with
CF = F = F C. Then C ⊂ M(D) and CD = CF DF = F DF = D and similarly
DC = D. Then M(C) ⊂ M(D) by a strictly continuous inclusion map such that
1M(C) 7→ 1M(D) . It follows that D is isomorphic to p11 Dp11 ⊗K where p11 is a rank
one projection in C. To see this, we take bi ∈ M(D) with b∗i bj = δij p11 and with
Pn
C the strict closure of the span of the bj bi (i, j = 1, 2, 3, . . .). Since { i=1 bi b∗i }
converges strictly to 1M(D) , d 7→ (b∗j dbi )ji defines a *-isomorphism from D onto
(p11 Dp11 ) ⊗ K. Since F is isomorphic to B ⊕ A, 1M(B) ⊕ 0 and 0 ⊕ 1M(A) define
projections p, q ∈ M(D) with p + q = 1 and p, q ∈ C 0 . By definition of F we have
that p, q ∈ M2 (B)∗∗ = M2 (B ∗∗ ) are given by p = diag(1, 0) and q = diag(0, r)
where r is the open support projection of A. Since p, q are in C 0 it follows that
under the isomorphism D ∼ = (p11 Dp11 ) ⊗ K (defined by C) p, q define p̄ = pp11 ,
q̄ = qp11 in M(p11 Dp11 ) with p̄ + q̄ = 1. Furthermore one gets p ∼ = p̄ ⊗ 1M(K) and
q∼ = q̄ ⊗ 1M(K) . ( M(D) ∼ = M(p11 Dp11 ) ⊗ K is described in Mat∞ (p11 Dp11 ) again
by T ∈ M(D) 7→ (b∗j T bi )ij and we get p 7→ (b∗j pbi ) = (b∗j bi p) = (δij p11 p) ⊗ 1M(K) .
Similarly q maps to p11 p ⊗ 1M(K) . p̄(p11 Dp11 p̄ generates p11 Dp11 as a closed ideal
because (p̄ ⊗ 1)(p11 Dp11 ⊗ K)(p̄ ⊗ 1) ∼= pDp and pDp generates D as a closed ideal.
A similar argument shows that q̄p11 Dp11 q̄ generates p11 Dp11 as a closed ideal, i.e.,
p̄ and q̄ are full projections and p̄ + q̄ = 1).
716 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

Since D contains a strictly positive element, p11 Dp11 also contains one.
Thus the conditions of [107, lem. 2.5] are fulfilled and we get isometries v̄, w̄ in
M(p11 Dp11 ⊗ K) with v̄ ∗ v̄ = w̄∗ w̄ = 1 and v̄v̄ ∗ = p̄ ⊗ 1M(K) , w̄w̄∗ = q̄ ⊗ 1M(K) .
In particular p̄ ⊗ 1M(K) and q̄ ⊗ 1M(K) are Murray- von Neumann equivalent
in M(p11 Dp11 ⊗ K). Now we use the isomorphism to M(D) and get a partial
isometry v ∈ M(D) with v ∗ v = p and vv ∗ = q. If we consider v as an element in
M2 (B ∗∗ ) ⊃ M(D) we get v ∗ v = diag(1, 0), vv ∗ = diag(0, r). Let y = v diag(e, 0),
then y satisfies y ∈ D, y ∗ y = diag(e2 , 0) is strictly positive in pDp ∼ = B ⊕ 0,
∗ 2 ∗ ∼
yy = v diag(e , 0)v is strictly positive in qDq = 0 ⊕ A. Thus we get that
y = (aij )ij ∈ M2 (B) with only a12 nonzero. Let x be a12 and ϕ(b) = ubu∗ for
1 1
any u in B ∗∗ with u∗ x = (x∗ x) 2 , xu∗ = (xx∗ ) 2 and kuk ≤ 1 (e.g. coming from
the polar decomposition of x). Since ϕ is a completely positive contraction, it is
sufficient to check that ϕ is multiplicative on the dense subset x∗ Bx of B and here
we only need that xx∗ = xu∗ ux∗ holds.
Similarly one checks that ϕ maps all onto the closure A of (xx∗ )1/2 B(xx∗ )1/2 ,
and that ψ(a) := u∗ au for a ∈ A is the inverse of ϕ, cf. Remark ??. 

Remark 5.5.8. Extensions E (given by an exact sequence 0 → B → E → A →

0) of stable separable algebras A by stable separable algebras B are in general not
stable. E.g. there is an extension E of A := K by B := C(Z, K) for the Tychonoff
product Z of countably many copies of the 2-dimensional sphere S 2 , cf. [685].
We show that every extension E of a separable stable C *-algebra A by a purely
infinite σ-unital stable B is stable.
This follows from some more general stability criteria that we discuss now.
Notice that it it suffices to consider the case A := C0 (R+ , K) by Remark
5.1.1(9).

Lemma 5.5.9. Suppose that B is stable and σ-unital, and that D is σ-unital.
Let Q(B) := M(B)/B and suppose that h1 , h2 : D → Q(B) are unitarily equivalent
C*-morphisms. Then:

(o) M(B) and Q(B) have property (sq) of Definition 4.2.14 (and are K1 -
bijective by Lemma 4.2.6(v) and Proposition 4.2.15).
(i) K∗ (M(B)) = 0 and U0 (M(B)) = U(M(B)).
(ii) E := Q(B) contains a copy of O2 unitally, and is K1 -bijective, i.e.,
the natural map U(E) → K1 (E) defines an isomorphism U(E)/U0 (E) ∼ =
∼
K1 (E) = K0 (B) .
(iii) If h1 dominates zero in the sense of Definition 4.3.3 then there is a unitary
U ∈ U(M(B)) such that πB (U )∗ h1 (·)πB (U ) = h2 (·) .
In particular, h1 and h2 are unitary equivalent by a unitary in U0 (E).

Proof. (i): The triviality K∗ (M(B)) = 0 easily follows from the unitary
equivalence of δ∞ and δ∞ ⊕ id.
The equation U0 (M(B)) = U(M(B)) was found by J. Cuntz and N. Higson [180],
 5. Ψ-EQUIVARIANT STABILITY OF EXTENSIONS 717

see also [557] in case that B contains a full projection. Modification of an observa-
tion in [180] leads to a slightly stronger observation in Lemma A.23.1, which implies
that M(B) satisfies the “squeezing” property (sq) of Definition 4.2.14, cf. Proposi-
tion 4.2.15. Thus, M(B) and, for every C *-algebra A, every C *-algebra quotient
of M(B) ⊗max A is K1 -bijective by Proposition 4.2.15.
(ii): O2 ⊂ L(H) = M(K) ⊆ M(B) unitally. Thus, E := M(B)/B contains a
copy of O2 unitally. E is K1 -surjective by Lemma 4.2.6(v) and is K1 -injective by
Part (i). This follows also from Proposition 4.2.15.
(iii): The maps h1 , h2 : D → E satisfy the requirements of Proposition
4.3.6(iv,d) by Part (ii): There exists W ∈ U(E) with W ∗ h1 (·)W = h2 . The
C *-morphism h1 dominates zero in the sense of Definition 4.3.3 if and only if there
exists an isometry s ∈ E with s∗ h1 (D)s = {0}. Then V := sW ∗ s∗ + (1 − ss∗ ) W

is a unitary in U(E) with V ∗ h1 (·)V = h2 and 0 = [V ] ∈ K1 (E). Thus, V ∈ U0 (E)

by the K1 -injectivity of E = Q(B) = M(B)/B. Since V is in U0 (M(B)/B) it is
a finite product of exponentials exp(tk ), of −t∗k = tk ∈ M(B)/B, k = 1, . . . , n.
There are Tk = −Tk∗ in M(B) with πB (Tk ) = tk . Thus, V = πB (U ) for the unitary
U := exp(T1 ) · . . . · exp(Tn ) in M(B). 

The following Lemma 5.5.10 summarizes consequences of Proposition 4.3.6 and

of the generalized Tietze extension theorem of Akemann, Pedersen and Tomiyama
in the σ-unital case (see [616, prop. 3.12.10] for the separable case).

Lemma 5.5.10. Suppose that A and B are σ-unital, and that h : A → M(B)/B
is a C*-morphism. Let J denote the kernel of h, and define
M(A, J) := {T ∈ M(A) ; T A ∪ AT ⊂ J} .

Further, let C := h(A), and let N (C) ⊂ M(B)/B the normalizer algebra of
C in M(B)/B, i.e., N (C) := {b ∈ M(B)/B ; bC ∪ Cb ⊂ C}, and let Ann(C) :=
{b ∈ M(B)/B bC = 0 = Cb} denote the (two-sided) annihilator of C in M(B)/B.
Define L : N (C) → M(C) by L(b)(c) := bc for c ∈ C and b ∈ N (C).
Then L is a C*-morphism, and the following properties are valid:

(i) h extends naturally to an epimorphism M(h) from M(A) onto M(C)

with kernel M(A, J) (generalized Tietze extension).
(ii) If, in addition, h has a non-degenerate lift to a C*-morphism h1 : A →
M(B), then πB (M(h1 )(M(A))) ⊂ N (C), and M(h) = L ◦ πB ◦ M(h1 ).
(iii) The natural C*-morphism L from the normalizer N (C) ⊂ M(B)/B of C
into the multiplier algebra M(C) of C is an epimorphism. The kernel of
L is Ann(C).
(iv) If h dominates zero, cf. Def. 4.3.3, and if S is any isometry in M(C), then
there exists a unitary U in the connected component of 1 in the unitaries
of M(B), such that πB (U )∗ h(·)πB (U ) = Sh(·)S ∗ .

Proof. (i): The morphism h : A → C is non-degenerate, because h is a *-

epimorphism from A onto C. Thus, there is a unique unital strictly continuous
718 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

C *-morphism M(h) from M(A) into M(C). By definition of M(h), M(h)(T ) = 0,

if and only if, h(T A ∪ AT ) = {0}.
The morphism M(h) is an epimorphism, because A is σ-unital. Indeed: This
is, — in the separable case —, the generalized Tietze extension theorem [616,
prop. 3.12.10]. If A is not separable — but σ-unital —, then one can reduce the
proof to the separable case ( 16 ), as follows:
Let c ∈ M(C)+ , let e ∈ A+ a strictly positive element of A and take b ∈ A with
h(b) = c · h(e), by using that h : A → C is surjective. Then the separable C *-
subalgebras F := C ∗ (b, e) ⊆ A, and G := h(F ) ⊆ C satisfy M(F ) ⊆ M(A) and
c ∈ M(G) ⊆ M(C), M(h)|M(F ) = M(h|F ) and M(h|F )(M(F )) = M(G).
(ii): Let b ∈ M(A), c ∈ C and a ∈ A with πB (h1 (a)) = h(a) = c. Then
M(h)(b)c = h(ba) = πB (h1 (ba)), and πB (M(h1 )(b))c = πB (M(h1 )(b)h1 (a)) .
(iii): It is obvious that Ann(C) ⊂ N (C), and that the natural C *-morphism L
from N (C) into M(C) has kernel Ann(C).
We show that N (C) → M(C) is surjective:
Consider D := (πB )−1 (C) and the *-epimorphism k := πB |D. Since C has
a strictly positive element d = h(e), and B has a strictly positive element p, the
element p + q is a strictly positive element of D, if q ∈ D+ is such that k(q) = d.
By part (i), M(k) is an epimorphism from M(D) onto M(C).
The lifted C *-morphism k1 := id |D of k is non-degenerate, because B ⊂ D.
Thus, M(k1 ) is a unital map from M(D) into M(B), and πB (M(k1 )(M(D))) ⊂
N (C).
By part (ii), the image of the natural C *-morphism N (C) → M(C) contains
the image of M(k), i.e., N (C) → M(C) is an epimorphism.
(iv): Let h2 := Sh(·)S ∗ . Then h2 is a C *-morphism from A into h(A) =
C ⊂ M(B)/B. By assumption, h dominates zero, i.e., there exists an isometry
V ∈ M(B)/B such that V ∗ h(A)V = {0}. Thus, V ∗ h2 (·)V = 0.
There exists a contraction y in N (C) such that L(y) = S by part (ii). It follows
y ∗ h2 (·)y = h and yh(·)y ∗ = h2 .
By Proposition 4.3.6(iii), h2 and h are unitarily equivalent by a unitary u1 ∈
M(B)/B. By Proposition 4.3.6(iv,c) the unitary equivalence can be realized by

16 There is a direct argument: Since M(h)(M(A)) contains C, it suffices to show that for

a given contraction c ∈ M(C)+ there exist d ∈ M(A)+ with M(h)(d) − c ∈ C. Let e ∈ A+

be a strictly positive element of A with kek = 1, and let d := h(e). By Remark 5.1.1(3), there
exist continuous functions 0 ≤ fn ≤ 1, with fn (0) = 0, fn+1 fn = fn , such that lim fn (t) = 1 for
1/2
t ∈ (0, 1] and that the commutator norms k[gn (d), c]k are less than 2−n , where g1 = f1 , and
gn := (fn+1 − fn )1/2 for n > 1. Let bn ∈ A+ contractions with h(bn ) = fn+2 (d)cfn+2 (d). By
P P
Remark 5.1.1(4), s := gn (a)bn gn (a) and t := gn (d)cgn (d) are strictly convergent series in
M(A) and M(C), respectively. Then M(h)(s) = t and c − t ∈ C, because h(gn (e)bn gn (e)) =
P
gn (d)cgn (d), M(h) is strictly continuous, and c − t is the sum of the norm-convergent series xn ,
where with xn := gn (d)[gn (d), c] ∈ C.
 5. Ψ-EQUIVARIANT STABILITY OF EXTENSIONS 719

a unitary u = πB (U ) in πB (U0 (M(B))) = U0 (M (B)/B), because V ∗ (h(a) +

h2 (a))V = 0 for a ∈ A. 

Lemma 5.5.11. Suppose that A and B are σ-unital and let h : A → Qs (B) a
C*-morphism.
Choose an isomorphism ψ : K ⊗ K → K from K ⊗ K onto K.
Let γ denote the natural isomorphism from (B ⊗ K) ⊗ K onto B ⊗ (K ⊗ K),
i.e., with γ((b ⊗ k1 ) ⊗ k2 ) = b ⊗ (k1 ⊗ k2 ) for b ∈ B, k1 , k2 ∈ K, and let

Φ1 : M(B ⊗ K) ⊗ K → M(B ⊗ K)

the non-degenerate monomorphism, is defined by

Φ1 := M((idB ⊗ψ) ◦ γ)|M(B ⊗ K) ⊗ K.

(i) There are isometries s0 , t0 ∈ M(K) and a norm-continuous map τ ∈

(0, 1] 7→ Iτ ∈ M(K) into the isometries in M(K) ∼
= L(H), such that, for
b ∈ K,

s0 s∗0 + t∗0 t0 = 1,
M(ψ)(1M(K) ⊗ p11 ) = s0 s∗0 ,
s∗0 ψ(k ⊗ p11 )s0 = k,
(Iτ )∗ M(ψ)(1M(K) ⊗ b)Iτ ∈ K, and
∗
lim k(Iτ ) M(ψ)(1M(K) ⊗ b)Iτ k = 0.
τ →0

(ii) Let r1 ∈ M(C0 ((0, 1], K)) be the isometry, which is defined by (r1 f )(τ ) :=
Iτ f (τ ) and (r1 f )(0) = 0 for f ∈ C0 ((0, 1], K), τ ∈ (0, 1]. Let e a strictly
positive element of B with kek = 1, and let µ the non-degenerate C*-
morphism from C0 ((0, 1]) into B with µ(g0 ) = e, where g0 (τ ) := τ . Now
define r = M(µ ⊗ idK )(r1 ).
Then Φ1 and the isometries s := 1M(B) ⊗ s0 , t := 1M(B) ⊗ t0 and r
in M(B ⊗ K) satisfy, for b ∈ M(B ⊗ K),

ss∗ + tt∗ = 1,
∗
ss = M(idB ⊗ψ)(1M(B) ⊗ (1M(K) ⊗ p11 )),
Φ1 (b ⊗ p11 ) = sbs∗ ,
r∗ Φ1 (M(B ⊗ K) ⊗ K)r ⊂ B ⊗ K, and
Φ1 ((B ⊗ K) ⊗ K) = B ⊗ K.

(iii) Φ := [Φ1 ] : Qs (B) ⊗ K → Qs (B) is a monomorphism, which dominates

zero and satisfies Φ(b ⊗ p11 ) = S ∗ bS with S := πB⊗K (s). The range
projection of S is SS ∗ = Φ(1 ⊗ p11 ).
(iv) h is unitarily equivalent to Φ(h(·) ⊗ p11 ) if and only if h dominates zero in
the sense of Definition 4.3.3. If this is the case, then the unitary equiva-
lence can be given by a unitary in the image of the unitaries in M(B ⊗K).
720 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

(v) Φ ◦ (h ⊗ idK ) is the Busby invariant of the stable extension

0 → (B ⊗ K) ⊗ K → E ⊗ K → A ⊗ K → 0,

where we identify B ⊗ K and (B ⊗ K) ⊗ K naturally by (idB ⊗ψ) ◦ γ, and

define
0 → B ⊗ K → E → A → 0,
as the extension corresponding to h : A → Qs (B).
(vi) If A is stable, then Φ(h(·) ⊗ p11 ) is the Busby invariant of a stable exten-
sion.

Proof. (i): p := M(ψ)(1M(K) ⊗ p11 ) and 1 − p are infinite projections in

M(K) ∼ = L(H). Thus, there are isometries s1 , t0 ∈ M(K) with s1 s∗1 = p and
t0 t∗0 = 1−p. Then b ∈ M(K) 7→ s∗1 M(ψ)(b⊗p11 )s1 is a unital weakly continuous *-
endomorphism on M(K). Its image contains the corner ψ(K⊗p11 ) of K. Therefore,
it is an isomorphism of M(K), and there exist a unitary u ∈ M(K) with ubu∗ =
s∗1 M(ψ)(b ⊗ p11 )s1 for b ∈ M(K). Let s0 := s1 u, then s0 and t0 have the desired
properties.
The construction of Iτ goes as follows. We let
n
X
pn := pjj
j=1

qn := M(ψ)(pn ⊗ (1 − pn ))
en := M(ψ)(pn+2 ⊗ (1 − pn )).

Then, qn ≤ en , qn+1 ≤ en , and qn , en − qn , en − qn+1 are infinite projections.

It follows that there exists a unitary v commuting with en , such that vqn v ∗ =
qn+1 . Since M(K) is isomorphic to the von-Neumann algebra L(H), there is a
T ∈ M(K)+ such that T commutes with en and v = eiT .
Let I an isometry in M(K) with II ∗ = qn . Then Iτ := eiT (τ ) v, with T (τ ) :=
((n + 1) − (n + 1)nτ )T , defines a norm-continuous map from [1/(n + 1), 1/n]. into
∗
the isometries of M(K), such that I1/n = I, In+1 In+1 = qn+1 and Iτ Iτ∗ ≤ en for
τ ∈ (1/(n + 1), 1/n).
By induction we get the norm-continuous map

τ ∈ (0, 1] 7→ Iτ ∈ M(K)

into the isometries of M(K) with the desired properties.

(ii): The properties of Φ1 , s and t can be seen by simple calculations on the
level of elementary tensors. Then use that all the maps in question are strictly
continuous, to get the general result.
On the property of r: It suffices to show that r∗ (1⊗M(ψ)(1⊗K))r is contained
in C ∗ (e) ⊗ K.
The properties of τ 7→ Iτ in part (i) mean that (r1 )∗ (1 ⊗ M(ψ)(1 ⊗ K))r1 is
contained in C0 ((0, 1], K). Now apply M(µ ⊗ idK ).
 5. Ψ-EQUIVARIANT STABILITY OF EXTENSIONS 721

(iii): With R := πB⊗K (r) we have R∗ R = 1 and R∗ Φ(·)R = 0 by (ii), i.e., Φ

dominates zero. The rest are straight calculations.
(iv): Since Φ dominates zero by (ii), also Φ(h(·) ⊗ p11 ) = Sh(·)S ∗ dominates
zero. Thus, if h is unitarily equivalent to h2 := Φ(h(·) ⊗ p11 ), then h dominates
zero.
Conversely, if V ∗ h(·)V = 0 for some isometry V ∈ Qs (B), and h2 = Sh(·)S ∗ ,
then W ∗ h2 (·)W = 0 for the isometry W := SV S ∗ +(1−SS ∗ ). Thus, by Proposition
4.3.6(iii), there is a unitary u with u∗ h(·)u = h2 , because h2 = Sh(·)S ∗ , h =
S ∗ h2 (·)S. By Lemma 5.5.9(iii) the unitary equivalence can be realized by a unitary
u = πB (U ) in πB (U0 (M(B)/B)).
(v): Straightforward calculations.
(vi): Let D be a σ-unital C *-algebra, and let ϕ an isomorphism from D ⊗ K
onto A. Let h1 := h ◦ ϕ.
Let ψ : K ⊗ K → K be the isomorphism from K ⊗ K onto K, and let s0 ∈ M(K)
an isometry with s0 bs∗0 = ψ(b ⊗ p11 ) for b ∈ K, see part (i).
Let γ denote the natural isomorphism from (D ⊗ K) ⊗ K onto D ⊗ (K ⊗ K).
We define an isomorphism λ from D ⊗ K onto (D ⊗ K) ⊗ K by λ := γ −1 ◦
(idD ⊗ψ −1 ).
Then ν(b) := λ−1 (b ⊗ p11 ), for b ∈ D ⊗ K, defines a *-endomorphism of D ⊗ K,
such that ν(d⊗b) := d⊗ψ(b⊗p11 ) for d ∈ D, b ∈ K. It follows, that s2 := 1M(D) ⊗s0
is an isometry in M(D ⊗ K), such that s2 bs∗2 = ν(b) for b ∈ D ⊗ K.
Let h2 := Φ ◦ (h1 ⊗ idK ) ◦ λ and h3 := Φ(h1 (·) ⊗ p11 ). Then h3 = h1 ◦ ν. The
h2 and h3 both dominate zero, because Φ dominates zero.
It suffices to show, that h2 and h3 are unitarily equivalent, by a unitary v =
πB (U ) for some unitary U of M(B⊗K), because the extension with Busby invariant
Φ ◦ (h ⊗ idK ) is stable, by part (v).
Let C := h2 (D ⊗ K). S := M(h2 )(s2 ) is an isometry in M(C), such that
h3 = Sh2 (·)S ∗ . Since h3 dominates zero, by Lemma 5.5.10(iv), there exists a
unitary U ∈ M(B) with v ∗ h2 (·)v = h3 for v := πB (U ). 
Proposition 5.5.12 (Stability of extensions). Suppose that
0→B→E→A→0
is an extension of A by B, where A and B are σ-unital and stable C*-algebras.
Let ι : B → E, η : E → A and ϕ : A → M(B)/B ∼ = Qs (B) denote the the defin-
ing morphisms of the above short-exact sequence, i.e., the corresponding monomor-
phism, respectively epimorphism, and the Busby invariant of E.
Further let θ denote an (arbitrary) isomorphism θ from B onto B ⊗ K, and
Θ : Q(B) → Qs (B) := Q(B ⊗ K) is the isomorphism that is induced by the isomor-
phism θ.
Then following properties (i)–(vi) are equivalent.
722 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

(i) The extension E is a stable C*-algebra.

(ii) The Busby invariant ϕ dominates zero.
(iii) For every a ∈ M(B)+ with πB (a) ∈ ϕ(A), b ∈ B+ and ε > 0, there exists
d ∈ B with kd∗ adk < ε and kd∗ d − bk < ε .
(iv) Θ ◦ ϕ and Φ(Θ ◦ ϕ(·) ⊗ p11 ) are unitarily equivalent, where Φ denotes the
natural monomorphism from Qs (B) ⊗ K into Qs (B) defined in Lemma
5.5.11(iii).
(v) There is an isomorphism λ from E ⊗ K onto E, such that B = λ(B ⊗ K),
and that b ∈ B 7→ λ(b ⊗ p11 ) and a ∈ A 7→ [λ](a ⊗ p11 ) are approximately
inner in B respectively in A, where [λ](a ⊗ p) := η(λ(f ⊗ p)) for a ∈ A,
f ∈ η −1 (a) and p ∈ K.
(vi) 1M(E) is properly infinite in M(E), i.e., the multiplier algebra M(E) of
E contains two isometries with orthogonal ranges.

The criteria (ii) and (vi) are not obvious, and (iii) is useful for applications.
Criterium (iii) was established by J. Hjelmborg and M. Rørdam in [373].
Compare also proof of [499, thm. 2.4].
Compare ?????

Proof. The implications (v)⇒(i) and (i)⇒(vi) are trivial. (i)⇒(iv) follows
from Lemma 5.1.2(ii).
(vi)⇒(iii): The natural C *-morphism h1 : E → M(B) with h1 (f )b =
−1
ι (f ι(b)) (for f ∈ E, b ∈ B) satisfies ϕ ◦ η = πB ◦ h1 for the Busby invari-
ant ϕ : A → Q(B) of the extension. This can be seen by the natural Busby
isomorphism:
E∼
= Eϕ := A ⊕ϕ,πB M(B) ⊂ A ⊕ M(B) ,

that is given by f 7→ (η(f ), h1 (f )).

Let C := ϕ(A) = πB (h1 (E)), and let a ∈ M(B)+ with πB (a) ∈ C, b ∈ B+ ,
ε > 0 and let δ := ε/(2(kak + kbk + 1)).
By Lemma 5.5.10(i) and (ii) (with E in place of A), the epimorphism h := πB ◦
h1 : E → C extends uniquely to a strictly continuous epimorphism M(h) : M(E) →
M(C).
By assumption there are isometries T1 , T2 ∈ M(E) with (T2 )∗ T1 = 0. Thus
rj := M(h)(Tj ) for j = 1, 2 are isometries in M(C) with orthogonal ranges.
Since A is stable, its image C = ϕ(A) by ϕ is again stable: A ∼ = D ⊗ K implies
that D has a closed ideal J such that C ∼ = (D ⊗ K)/(J ⊗ K) ∼
= (D/J) ⊗ K.
Let s1 , s2 , . . . a sequence of isometries in M(C) such that n sn (sn )∗ strictly
P

converges to 1. Then there is m ∈ N with ks∗m πB (a)sm k < δ. Since M(C) contains
a copy of O2 unitally, there is a unitary v in M(C), such that vr1 = sm . The
unitary group of M(C) is (norm-)contractible by [180]. Hence, there is a unitary
U ∈ M(E) with M(h)(U ) = v.
 5. Ψ-EQUIVARIANT STABILITY OF EXTENSIONS 723

The C *-morphism h1 : E → M(B) is non-degenerate, because h1 (ι(B)) =

B. Thus h1 uniquely extends to a strictly continuous unital C *-morphism
M(h1 ) : M(E) → M(B).
By Lemma 5.5.10(ii), πB (M(h1 )(M(E))) ⊂ N (C) and M(h) = L◦πB ◦M(h1 )
for the natural epimorphism L from N (C) onto M(C) with kernel Ann(C).
Thus the isometry I := M(h1 )(U T1 ) ∈ M(B) satisfies πB (I) ∈ N (C) and
dist(I ∗ aI, B) = kπB (I ∗ aI)k.
It follows πB (I ∗ aI) = (L ◦ πB )(I ∗ aI) == M(h)(U T1 )∗ πB (a)M(h)(U T1 ) =
s∗m asm, and, therefore, there is g = g ∗ ∈ B with kI ∗ aI − gk < δ.
Now let t1 , t2 , . . . a sequence of isometries in M(B) such that tn (tn )∗ strictly
P

converges to 1 (exists, because B is stable). There is k ∈ N such that kt∗k gtk k < δ.
Let d := Itk b1/2 ∈ B, then d∗ d = b and kd∗ adk < 2δkbk < ε.
(iii)⇒(ii): Let e ∈ A+ and b ∈ B+ strictly positive elements of A and B
respectively. We find a ∈ M(B)+ with πB (a) = ϕ(e). It is enough to find an
isometry S ∈ M(B) with S ∗ aS ∈ B.
Let C := C ∗ (a, b, 1) denote the unital C *-subalgebra of M(B) generated by
a, b and 1. Then C is the unitization D e of the C *-subalgebra D of C which is
generated by b and a. Let χ : C → C ∼ = C/D the natural character.
The C *-morphism V : C → M(B), with V (c) := χ(c)1 for c ∈ C, satisfies
condition (α) of Proposition 5.4.1 with h := b.
f := b + a is a strictly positive element of D, and, by assumption, for every
δ > 0, there is d ∈ B with kd∗ f δ dk < δ and kd∗ d − bδ k < δ.
Thus V and C satisfy also condition (β) of Proposition 5.4.1.
Since δ∞ ◦ V = V and V (a) = 0, by Proposition 5.4.1(ii) and (iv), there exists
an isometry S ∈ M(B) with S ∗ aS ∈ B.
(ii)⇒(iv): Θ ◦ ϕ dominates zero if ϕ dominates zero. But then Θ ◦ ϕ and
Φ((Θ ◦ ϕ)(·) ⊗ p11 ) are unitarily equivalent by Lemma 5.5.11(iv).
(iv)⇒(i): Since A is stable and σ-unital, Φ((Θ ◦ ϕ)(·) ⊗ p11 ) is the Busby in-
variant of a stable extension 0 → B ⊗ K → E1 → A → 0 by Lemma 5.5.11(vi).
Since Φ((Θ ◦ ϕ)(·) ⊗ p11 ) is unitarily equivalent to Θ ◦ ϕ, it follows that this equiv-
alence is given by a unitary in the image in M(B ⊗ K)/(B ⊗ K) of the unitaries
of M(B ⊗ K), cf. Lemma 5.5.11(iv). Thus, the extension E1 is equivalent to an
extension 0 → B ⊗ K → E2 → A → 0 with Busby invariant Θ ◦ ϕ –
in particular, E2 ∼
= E1 as C *-algebras, and
?? ????.
Since E2 and 0 → B → E → A → 0 are determined up to equivalence by the
Busby invariants Θ ◦ ϕ and ϕ, it follows that E2 and E are naturally isomorphic
(in particular, as C *-algebras). Thus E is a stable C *-algebra.
724 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

(i)⇒(v): Let µ : E ⊗ K → E any isomorphism from E ⊗ K onto E, and let

ι : B → E, η : E → A the defining morphisms for the exact sequence 0 → B →
E → A → 0. Then there is a closed ideal J of E such that µ(J ⊗ K) = ι(B) and
J ⊗ K is the kernel of η ◦ µ : E ⊗ K → A.
We define a new isomorphism λ from E ⊗ K onto E by

λ := µ ◦ (idB ⊗ψ) ◦ γ ◦ (µ−1 ⊗ idK ) ,

where ψ is an isomorphism from K ⊗ K onto K, and γ is the natural isomorphism

from (E ⊗ K) ⊗ K onto E ⊗ (K ⊗ K).
Then λ(ι(B) ⊗ K) = ι(B) and the *-monomorphisms

b ∈ B 7→ ι−1 (λ(ι(b) ⊗ p11 ))

and a ∈ A 7→ [η ◦ λ](a ⊗ p11 ) are approximately inner, because even the *-

endomorphisms g ∈ E ⊗ K 7→ (idE ⊗ψ)(γ(g ⊗ p11 )) and f ∈ E 7→ λ(f ⊗ p11 )
are unitarily homotopic to idE⊗K respectively idE . 

Remark 5.5.13. One can the criteria 5.5.12(iii) replace by the (formally
weaker) criteria:

(iii*) There exists k ∈ N, such that, for every b, g ∈ B+ , ε > 0, e1 , . . . , ek ∈

M(B), with πB (ej ) ∈ ϕ(A) and e∗i ej = δij e∗1 e1 for i, j = 1, . . . , n, there is
d ∈ B with kd∗ (e∗1 e1 )dk < ε, kd∗ gdk < ε and kd∗ d − bk < ε.

Indeed:
(iii)⇒(iii*): Take k := 1, and let a := g + e∗1 e1 in (iii).
(iii*)⇒(iii): Let C := ϕ(A), a ∈ M(B)+ with πB (a) ∈ C, b ∈ B+ , ε > 0, and
let c := πB (a), δ := ε/(2 + kbk + ε).
Since A is stable, C is stable, and there are isometries s1 , s2 , . . . in M(C), such
that pj strictly converges to 1, where pj := sj s∗j . Thus there is a projection q1 :=
P

p1 +. . .+pn in O∞ ⊂ M(C) such that kq1 cq1 −ck < δ. Let qj := p(j−1)n+1 +. . .+pjn .
Then there are partial isometries vj in M(C) with vj∗ vj = q1 and vj vj∗ = qj . Let
fj := vj (q1 cq1 )1/2 , j = 1, . . . , k. Then fj ∈ C and fi∗ fj = δij q1 cq1 . By Proposition
A.8.4, there exist e1 , . . . , ek ∈ M(B), with πB (ej ) = fj and e∗i ej = δij e∗1 e1 for
i, j = 1, . . . , n.
It follows, that g := (|a − e∗1 e1 | − δ)+ is in B+ . By assumption, there is d ∈ B
with kd∗ e∗1 e1 dk < δ, kd∗ gdk < δ and kd∗ d − bk < δ.
Since δ < ε and a ≤ e∗1 e1 + |a − e∗1 e1 |, we get kd∗ d − bk < ε and kd∗ adk < ε.

Corollary 5.5.14. Suppose that D and B are σ-unital C*-algebras and that
−1
ψ : D → M(B)/B is a *-monomorphism. Let E := πB (ψ(D)).

(I) If ψ has a non-degenerate “split” C*-morphism H : D → M(B) with

ψ = πB ◦ H, and if D is stable, then B and E are stable.
(II) If E is stable, then
 5. Ψ-EQUIVARIANT STABILITY OF EXTENSIONS 725

(i) B and D are stable and ψ ⊕0 is unitarily equivalent to ψ by a unitary

u ∈ U0 (Q(B)), and
(ii) there exists isometries S, T ∈ M(B) with SS ∗ +T T ∗ = 1, S ∗ ES ⊆ B
and (1 − T )E ∪ E(1 − T ) ⊂ B.

Proof. Ad(I): Suppose that D is stable and that H : D → M(B) is a non-

degenerate C *-morphism with ψ = πB ◦ H.
Then the σ-unital C *-algebra B is stable by Remark 5.1.1(9), because H(D)
is a stable non-degenerate C *-subalgebra of M(B), i.e., H(D)B = B.
If e ∈ E+ , then there exist b ∈ B and d ∈ D with e = b + H(d). Thus, for ε > 0
there exists a contraction f ∈ D+ with

ke − H(f )ek ≤ kb − H(f )bk + kd − f dk < ε ,

i.e., H(D)E = E and H(D) is a non-degenerate C *-subalgebra of M(E). It follows

that E is stable by Remark 5.1.1(9).
(II): Suppose that E is stable.
(i): The stability of E implies the stability of its ideal B and of its quotient D.
By Proposition 5.5.12(ii), the Busby invariant ψ : D → M(B)/B ∼ = Qs (B)
dominates zero, i.e., there is an isometry t ∈ Q(B) with t∗ ψ(D)t = {0}.
Define ⊕ := ⊕s1 ,s2 . Then ψ and ψ ⊕ 0 both dominate zero: t∗ ψ(·)t = 0 and
(t ⊕ 1)∗ (ψ(·) ⊕ 0)(t ⊕ 1) = 0.
Chose one of next:
Since B is σ-unital and stable, the stable corona Qs (B) of B is K1 -injective by
Lemma 4.2.10.
Stable coronas Qs (B) of σ-unital C *-algebras B have Property (sq) and are
therefore K1 -bijective by Proposition 4.2.15.
If B is stable, then Q(B) is K1 -surjective by Lemma 5.5.9.
It follows that ψ and ψ(·) ⊕ 0 are unitary equivalent with a unitary u ∈
U0 (Qs (B)) by last conclusion in Parts (iii) and (iv,b) of Proposition 4.3.6.
Prop./Lem./Rem. in Chp. 4
??
????????????????? ??
Since B is stable, there exist isometries S1 , S2 ∈ M(B) with S1 S1∗ + S2 S2∗ = 1.
Let sj = πB (Sj ). By Lemma 4.2.6(iii) there is a unitary u1 in Q(B) with u1 s2 = ss2 ,
because t1 := ss1 s2 and t2 := ss2 and r1 := s1 s2 and r2 := s2 are pairs of isometries
(t1 , t2 ) and (r1 , r2 ) with orthogonal ranges and the property that 1 − (t1 t∗1 + t2 t∗2 ) ≥
ss21 (ss21 ) and 1 − (r1 r1∗ + r2 r2∗ ) = s21 (s21 )∗ are full and properly infinite projections
in Q(B).
What is the application?
726 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

Then ψ = u∗1 ψ(·)u1 and ψ2 := ψ(·) ⊕s1 ,s2 0 have the property ψj (D)s2 = 0 for
j = 1, 2.
Since B is stable, Q(B) is K1 -injective by Lemma 5.5.9.
Better refer to Lemma for property sq in 4 ?
By Proposition 4.3.6(iv,a ???), there exists a a unitary V ∈ U0 (Q(B)) with
??? ?? Proceed proof ! ????????????????
(ii): By the proof of part (ii) of Lemma 5.5.9, there exist isometries S1 , S2 , T1 ∈
M(B) with πB (S1 ) = sπB (S2 ) and S1 S1∗ + T1 T1∗ = 1. Thus, πB (S1∗ ES1 ) =
πB (S2 )∗ s∗ ψ(D)sπB (S2 ) = {0}. It follows πB (S1 (1 − T1 )S1∗ E) = {0} = πB (ES1 (1 −
T1 )S1∗ ).
Now let P := T1 T1∗ = 1 − S1 S1∗ , T := P + S1 T1 S1∗ and S := S12 . Then
T T = 1 = S ∗ S, T T ∗ = P + S1 P S1∗ = 1 − SS ∗ and 1 − T = S1 (1 − T1 )S1∗ .
∗

Hence, S ∗ ES ⊂ B and (1 − T )E ∪ E(1 − T ) ⊂ B. 

Corollary 5.5.15. Suppose that Y is a σ-compact locally compact space, that

B is a σ-unital stable C*-algebra and that C is a σ-unital C*-subalgebra of the ideal
Q(Y, B) := Cb (Y, B)/ C0 (Y, B) of Q(C0 (Y, B)) = M(C0 (Y, B))/ C0 (Y, B).

(i) There are isometries S and T in Cb (Y, M(B)) ⊂ M(C0 (Y, B)) =
M(Cb (Y, B)) such that SS ∗ + T T ∗ = 1 , and that s := S + C0 (Y, B) ∈
Q(C0 (Y, B)) and t := T + C0 (Y, B) ∈ Q(C0 (Y, B)) satisfy s∗ Cs = {0}
and (1 − t)C = {0} = C(1 − t).
(ii) If D is a σ-unital stable C*-algebra, and 0 → C0 (Y, B) → Eϕ → D → 0 is
an extension which has Busby invariant ϕ : D → Q(C0 (Y, B)) with image
ϕ(D) in Cb (Y, B)/ C0 (Y, B) ⊂ Q(C0 (Y, B)), then Eϕ is stable.
(iii) If f is a contraction in Cb (Y, B), then there is an isometry T in
Cb (Y, M(B)) such that g ∗ ag = t∗ at for all a ∈ C, where g := f +C0 (Y, B)
and t := T + C0 (Y, B).
(iv) There is an *-monomorphism ρ : C ⊗ K ,→ Cb (Y, B)/ C0 (Y, B) with ρ(c ⊗
p11 ) = c for c ∈ C.

Proof. (i): We use that B = ∼ B ⊗ K, and replace B by B ⊗ K. Since Y

is σ-compact, there is a strictly positive function γ ∈ C0 (Y ) with 0 ≤ γ(y) ≤ 1.
Since γ ∈ C0 (Y )+ is strictly positive, the subsets Yn := γ −1 [1/n, 1] ⊂ Y are
S
compact and Y = n Yn . Let g ∈ C+ a strictly positive contraction for C. There
is f ∈ Cb (Y, B ⊗ K)+ with g = f + C0 (Y, B ⊗ K) and kf (y)k ≤ 1. There are
projections pn ∈ K with kf (y)(1 ⊗ pn ) − f (y)k < 1/(n + 1) for all y ∈ Yn+1 for
n = 0, 1, 2, . . .. This can be seen, because f |Yn is in C(Yn , B ⊗ K) ∼
= C(Yn , B) ⊗ K.
If we take linear interpolations p(t) := (n − t)pn−1 + (t + 1 − n)pn for t ∈ [n − 1, n],
then we get a norm-continuous path t ∈ [0, ∞) 7→ p(t) ∈ K from R+ into the
positive contractions of K such that kf (y)(1 ⊗ p(t)) − f (y)k < 1/n for all y ∈ Yn
and t ≥ n − 1. Thus, kf (y)(1 ⊗ p(γ(y)−1 )) − f (y)k ≤ γ(y) for all y ∈ Y . Note that
 5. Ψ-EQUIVARIANT STABILITY OF EXTENSIONS 727

W
p(t) ≤ qn := 1≤k≤n pk ∈ K, p(t) = p(t)qn and p(t)rn − p(t) = p(t)(qn rn − qn ) for
t ≤ n.
There is a non-degenerate *-representation λ : D := O2 ⊗ K → L(`2 ) ∼
= M(K).
We identify d ∈ D with λ(d) and M(K) with L(H) (for simplicity of notations).
Then we find projections rn ∈ D with 0 6= r1 ≤ r2 ≤ · · · , rn 6= rn+1 and
kqn rn − qn k < 1/n for t ≤ n because qn ∈ K. By Lemma 5.1.2(iii), there exists a
norm-continuous path t 7→ U (t) from R+ into the unitaries in 1 + D ⊂ M(K) such
that U (t)r1 U (t)∗ ≥ rn+1 for t ≥ n.
Since r1 and 1 − r1 ≥ r2 − r1 ∈ D \ {0} are properly infinite projections in
M(K), there are isometries S0 , T0 ∈ M(K) with S0 S0∗ = 1 − r1 and T0 T0∗ = r1 . It
follows kp(t)U (t)S0 k ≤ kp(t)−p(t)rn+1 k+krn+1 U (t)S0 k < 1/n + 1 for t ∈ [n, n+1],
because rn+1 U (t)(1 − r1 ) = 0 for t ≥ n. Thus, kp(γ(y)−1 )U (γ(y)−1 )S0 k ≤ γ(y) .
We define a norm-continuous map y 7→ V (y) from Y into the unitary elements
of 1 ⊗ M(K) ⊂ M(B ⊗ K) by V (y) := 1 ⊗ U (γ(y)−1 ) for y ∈ Y . Then S1 (y) :=
V (y)(1⊗S0 ) and T1 (y) := V (y)(1⊗T0 ) define isometries S1 , T1 ∈ Cb (Y, M(B ⊗K))
with S1 S1∗ + T1 T1∗ = 1 and, for y ∈ Y ,

kf (y)S1 (y)k ≤ kf (y)(1 ⊗ p(γ(y)−1 )) − f (y)k + kp(γ(y)−1 )U (γ(y)−1 )S0 k ≤ 2γ(y) .

Thus, f S1 ∈ C0 (Y, B⊗K), i.e., gs1 = 0 and s∗1 Cs1 = {0} for s1 := S1 +C0 (Y, B⊗K).
Now we proceed as in the proof of Proposition 5.5.14: The isometries S := S12 ,
T := T1 T1∗ + S1 T1 S1∗ and s := S + C0 (Y, B ⊗ K) are as stipulated.
(ii): By (i) with C := ϕ(D), Proposition 5.5.12 applies to ϕ, because ϕ domi-
nates zero.
(iii) follows from (i) by Proposition 4.3.6(ii), with Cb (Y, M(B))/ C0 (Y, B) in
place of E.
(iv): Let S, T ∈ Cb (Y, M(B)) the isometries from part (i), and let s := S +
C0 (Y, B),
t := T + C0 (Y, B), t1 := t and tn := stn for n = 2, 3, . . .. Then s, t, tn are
isometries in Cb (Y, M(B))/ C0 (Y, B) with t∗ s = 0, t∗m tn = 0 for m < n, and t1 ct∗1 =
tct∗ = c for all c ∈ C. There is a *-morphism ρ : C ⊗ K → Cb (Y, B)/ C0 (Y, B) with
ρ(c ⊗ pjk ) = tj ct∗k for canonical matrix units pjk in K. We get ρ(c ⊗ p11 ) = tct∗ = c
for c ∈ C . 

Compare here also [499, thm. 2.4.]: Let A and B be separable C *-

algebras such that B ⊗K has the CFP (= Corona Factorization Property). Suppose
that there is an extension of C *-algebras of the form 0 → B → E → A → 0. Then
E is stable if and only if A and B are stable.
(E.K.: It is clear that A and B must be stable if E ∼ = E ⊗ K. Does
Prop. 5.5.12(iii) apply in the opposite direction?
Implies “B is s.p.i. and σ-unital” that B ⊗ K has the CFP?)
728 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

Corollary 5.5.16. Extensions 0 → B → E → A → 0 of stable σ-unital

C*-algebras A by σ-unital purely infinite stable C*-algebras B are stable.

Proof. Suppose that B is a stable, σ-unital and purely infinite C *-algebra,

that A is a stable σ-unital C *-algebra, and that ϕ : A → Q(B) is a C *-morphism,
i.e., is the Busby invariant of an extension of A by B.
We show that the criterion (iii) of Proposition 5.5.12 is satisfied:
−1
Let C := ϕ(A), E := πB (C), a ∈ E+ , b ∈ B+ a strictly positive contraction in
B, let ε ∈ (0, 1), and put δ := ε/5 . The annihilator D := Ann((a − δ)+ ) of (a − δ)+
in E generates a closed ideal J of E, such that J = E or E/J is unital and πJ (a)
is invertible. Then E/(J + B) ∼ = C/πB (J) is a quotient of C. It is isomorphic to
−1
A/ϕ (πB (J)) and is unital (if non-zero). Since A is stable it has no unit. Thus
ϕ(A) = πB (J), i.e., J + B = E and B/(B ∩ J) ∼ = E/J is unital or zero. Since B is
stable and E/J is unital we get B = B ∩ J ⊂ J.
The hereditary C *-subalgebra D ∩ B generates B ∩ J = B as a closed ideal
of B, because B is an ideal of E. Therefore, there exist d1 , . . . , dn ∈ B with
kb − d∗j dj k < δ and dj d∗j ∈ (D ∩ B). Let c := dj d∗j , g := d∗j dj . Since B is
P P
∗
purely infinite, there exist f ∈ B with kf cf − gk < δ. It follows, that d := c1/2 f
satisfies kd∗ d − bk < 2δ < ε and kd∗ adk < kdk2 δ < ε, i.e., the stability criterion
(iii) of Proposition 5.5.12 applies. 

Remark 5.5.17. Suppose that A and B are stable and σ-unital, and that
ϕ : A ⊗ O2 → Q(B) is a *-monomorphism. The above results contain the following
−1
observations: (i) The extension E := πB (ϕ(A ⊗ O2 )) is stable if and only if E0 :=
−1
πB (ϕ(A ⊗ 1)) is stable.
(This is because ϕ(A ⊗ 1) is orthogonal to the range of an isometry in M(B)/B
if and only if ϕ(A ⊗ O2 ) is orthogonal to it. Recall that E is always stable if B is
purely infinite by Corollary 5.5.16.)
(ii) If E is stable, then there is a copy of O2 unitally contained in M(B), such that,
for a ∈ A and c ∈ O2 ,

ϕ(a ⊗ 1)πB (c) = ϕ(a ⊗ c) = πB (c)ϕ(a ⊗ 1) .

(Indeed, then M(E) contains a unital copy of O2 = C ∗ (s1 , s2 ), and the epimor-
phism λ := ϕ−1 πB |E : E → A ⊗ O2 extends to an epimorphism M(λ) from M(E)
onto M(A ⊗ O2 ), because λ is surjective and E and A ⊗ O2 are both σ-unital, i.e.,
one can apply Proposition ??.
(find precise cite! But it is also easy to check...) Pedersen Thm.:
M(A) → M(A/J) surjective if A is σ-unital.
There is a unitary V ∈ M(A ⊗ O2 ) with V (1 ⊗ sk ) = M(λ)(sk ). Since the
unitary group of M(A ⊗ O2 ) is connected by Remark ??,
Cuntz-Higson Theorem:
U(M(A)) = U0 (M (A)) if A stable and σ-unital
 6. W-VN TYPE RESULTS FOR WEAKLY NUCLEAR MAPS 729

there is a unitary U ∈ M(E) ⊂ M(B) with M(λ)(U ) = V . One can see, that
C (U ∗ s1 , U ∗ s2 ) ∼
∗
= O2 is a copy of O2 in M(B) with the desired property.)
(iii) If E0 is stable and ϕ0 (a) := ϕ(a ⊗ 1), then (ii) is equivalent to [ϕ0 ] = [ϕ0 ] + [ϕ0 ]
in [Hom(A, Q(B)], i.e., ϕ0 is unitarily equivalent to ϕ0 ⊕ ϕ0 by a unitary in Q(B).

Question 5.5.18. Is every extension

0 → B → E → C0 ((0, 1], O2 ⊗ K) → 0

stable if B is stable and separable? (Compare Remark 5.5.17(i).)

6. W-vN type results for weakly nuclear maps

We deduce from Propositions 5.4.1 and 4.3.5(i) a proof of Kasparov’s more

general version, cf. Corollary 5.6.1, of Voiculescu’s generalized Weyl–von Neumann
theorem. It is implicitly contained in [404], but is not explicitly stated there.
Voiculescu’s generalization of the Weyl-von Neumann theorem reads now as a spe-
cial case of Kasparov’s generalization. The proof of Corollary 5.6.1 shows that we
could deduce Corollary 5.6.1 as a logical sum of Voiculescu’s generalized Weyl–von
Neumann theorem and of Theorem 5.6.2.
Let A a (non-zero) separable C *-algebra and B a stable σ-unital C *-algebra.
We have seen in Chapter 3 that the universal *-monomorphism HC : A → M(B) for
the matrix operator-convex cone C := CPnuc (A, B) of completely positive nuclear
maps from A into B is given up to unitary homotopy by the following universal
construction:
Let d : A → M(K) ∼= L(`2 ) any faithful non-degenerate *-representation of A,
in sense of Remark 5.1.1(i).
We apply the infinite repeat δ∞ of Remark 5.1.1(8) to d, i.e., replace d by
δ∞ ◦ d to make sure that d is in general position, i.e., that d(A) ∩ K = {0} and that
d(A)K = K. A crucial property is that this “general position” implies that d and
δ∞ ◦ d are unitarily homotopic.
The *-representation d is unital if A is unital. If A is not unital, then the
extension given by the Busby invariant

β0 := πK ◦ d : A → M(K)/K

dominates zero in sense of Definition 4.3.3. Since B is stable there is up to

unitary equivalence (and up to unitary homotopy) a unique non-degenerate *-
monomorphism H0 : K → M(B)
Hilbert B-module that the generating

Corollary 5.6.1 (Kasparov,Voiculescu,Weyl–von-Neumann). Suppose that

B is a σ-unital C*-algebra, I : M(K) → M(B) a strictly continuous unital *-mono-
morphism and C ⊂ M(B) a separable C*-subalgebra such that C ⊂ I(M(K)).
Let T : C → M(B) be a completely positive contraction, such that
730 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

(α0 ) T (C ∩ I(K)) = 0,
(β 0 ) the completely positive contraction T 0 : C/(C ∩ I(K)) → M(B) is weakly
nuclear, and
(γ) T (c) = 1 for c ∈ C ∩ (1 + I(K)) (if 1M(B) ∈ C + I(K)).

Then:

(i) idC asymptotically dominates T in the sense of Definition 5.0.1.

(ii) If, moreover, T : C → M(B) is a C*-morphism, then idC asymptotically
absorbs T , i.e., idC ⊕T : C → M(B) and idC are unitarily homotopic in
the sense of Definition 5.0.1.

Proof. To keep notations simple, we identify the elements of M(K) and

I(M(K)) with help of I. (In fact, there is a σ-unital C *-algebra A and an iso-
morphism λ from A ⊗ K onto B such that I(k) = M(λ)(1M(A) ⊗ k) for k ∈ K, and
C lives in 1 ⊗ M(K).)
The strict continuity of the unital *-monomorphism I implies that KB is dense
in B. M(K) contains a copy of O∞ with generators given by isometries t1 , t2 , . . .
with t∗j tk = δjk 1 such that tk t∗k converges strictly to 1M(K) = 1M(B) . By Remark
P

5.1.1(8), it follows that B is stable.

The conditions (β 0 ) and (γ) allow us to reduce the proof to the case where C
is unital and contains K and that T is unital with T (K) = 0 : By assumption (β 0 ),
T 0 : C/(C ∩ K) → M(B) is weakly nuclear. Replace C by C + K if 1M(B) ∈ C + K.
Then T 0 is unital by assumption (γ). Thus we can replace T by T 0 πK , because
(C + K)/K ∼ = C/(C ∩ K).
If 1M(B) is not in C + K, then (C + K + C1)/K is naturally isomorphic to the
(outer) unitization of C/(C ∩K) and the unital extension T1 of T 0 to (C +K+C1)/K
is again completely positive and weakly nuclear by Lemma B.7.7(i). Thus we can
replace C by C + K + C1 and T by T1 πK .
Thus C and V := δ∞ ◦ T satisfy the condition (α) of Proposition 5.4.1, where
we take a strictly positive element h of K ⊂ C. The new C and V are now
unital. Moreover, V is a unital C *-morphism if the original T : C → M(B) is a
C *-morphism.
Next, we shall show that C and V satisfy condition (β) of Proposition 5.4.1.
Then (i) and (ii) will follow from parts (ii), (iv) and (iii) of Proposition 5.4.1.
Let ρ : C/K → M(K) ∼ = L(H) be given by a faithful unital *-representation
of the separable unital C *-algebra C/K over a separable Hilbert space H. We
define C1 := δ∞ (ρ(C/K)) and a unital completely positive map V1 : C1 → M(B)
by V1 (e) := δ∞ (T 0 (c)) for e ∈ C1 , c ∈ C/K, e = δ∞ (ρ(c)). Then, by Remark
3.1.2(iii), V1 is weakly nuclear and V = δ∞ ◦ ◦T = V1 γ, where γ := δ∞ ◦ ρ ◦ πK is a
unital C *-morphism from C ⊂ M(K) onto C1 ⊂ M(δ∞ (K)) with kernel K. Note
that M(δ∞ (K)) = δ∞ (M(K)) ⊂ M(K) by our choice of O∞ .
 6. W-VN TYPE RESULTS FOR WEAKLY NUCLEAR MAPS 731

By Lemma 5.1.2(ii), the commutant of δ∞ (M(K)) in M(B) contains a copy

of O2 unitally. D := C ∗ (O2 · δ∞ (K)) ∼
= O2 ⊗ K is a simple purely infinite C *-
subalgebra of M(B) such that δ∞ (M(K)) ⊂ M(D) and δ∞ (K)D is dense in D.
Thus C1 D ⊂ D. DB contains δ∞ (K)B and is therefore dense in B.
Let c1 , . . . , cn be contractions in C and ε > 0. Since C1 is a separable subalgebra
of M(D) ⊂ M(B) and since, for a ∈ B+ , the map e ∈ C1 7→ aV1 (e)a is nuclear, by
Proposition 3.2.15(i), there exists b ∈ B with kb∗ γ(cj )b − aV1 (γ(cj ))ak < ε/3 for
j = 1, . . . , n.
We find a projection p ∈ K with 6kb − pbk < ε/(1 + kbk), because KB is dense
in B.
By Lemma 2.1.22, we find a partial isometry v ∈ K such that v ∗ v = p and
3kS(cj ) − v ∗ cj vk < ε/(kbk2 + 1) for j = 1, . . . , n , where the unital completely
positive map S : C → pKp ∼ = Mk is given by S(c) := pγ(c)p , k := Dim(p`2 ) and
satisfies S(K) = 0 . Let d := vb . Since V (cj ) = V1 (γ(cj )) , the triangle inequality
gives that kd∗ cj d − aV (cj )ak < ε for j = 1, . . . , n . Thus C and V := δ∞ ◦ T satisfy
also condition (β) of Proposition 5.4.1. 

We study now weakly nuclear liftable semi-split essential extensions of sep-

arable C *-algebras A by simple purely infinite σ-unital stable algebras B. The
corresponding generalized Weyl–von Neumann Theorem 5.6.2 implies that the ex-
tension is split if its class in Extnuc (A, B) is zero. The pure infiniteness and the
weak nuclearity of the semi-split map is necessary in this statement, cf. Corollaries
5.7.1 – 5.7.3.
The following generalized Weyl–von Neumann Theorem 5.6.2 is in view of
Proposition 5.4.1 almost equivalent to the “local” result in Corollary 3.10.14.
If B nuclear then every unital completely positive contraction T with T (C ∩
I(K)) = {0} respectively T (C ∩ D) = {0} can be considered in Corollary 5.6.1 and
Theorem 5.6.2.
Our Theorem 5.6.2 is simply a corollary of Proposition 3.2.13, Proposition 5.4.1,
and Lemma B.7.7 together. (We call it a theorem, because of its importance.)

Theorem 5.6.2 (Generalized Weyl–von Neumann Theorem). Suppose that B

is σ-unital and stable, and that D ⊆ M(B) a
simple purely infinite
C*-subalgebra of M(B) such that DB is dense in B.
Let C be a separable C*-subalgebra of M(B) with CD ⊆ D and T : C → M(B)
a completely positive contraction such that

(α0 ) T (C ∩ D) = 0,
(α00 ) C ∩ D generates a corner of D,
(β 0 ) T : C → M(B) is weakly nuclear, and
(γ) T (C ∩ (1 + D)) ⊆ {1M(B) }.
732 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

Then

(i) idC asymptotically dominates T (in the sense of Definition 5.0.1).

(ii) If, moreover, T : C → M(B) is a C*-morphism then idC asymptotically
absorbs T , i.e., idC ⊕T : C → M(B) and idC are unitarily homotopic in
the sense of Definition 5.0.1.

Proof. The natural extension T1 : d + c 7→ T (c) of T to D + C is still weakly

nuclear by Lemma B.7.7(ii) and by assumptions (α0 ), (α00 ) and (β 0 ).
It is unital by assumption (γ), if 1M(B) = 1M(D) is in C + D. If 1M(B) is not in
C + D, then the unital extension T2 : d + z1 + c 7→ T (c) + z1 is again weakly nuclear
by Lemma B.7.7(i). The extension T1 (respectively T2 ) is a unital C *-morphism if
T is a C *-morphism from C into M(B).
Thus, we can assume that C and T are unital and that the natural extension
of T to C + D is weakly nuclear.
Let b ∈ B+ a strictly positive element of B. We find a sequence d1 , d2 , . . .
in D+ such that dn b tends to b, because DB is dense in B. Let C1 ⊂ C + D
be the separable unital C *-subalgebra of M(D) which is generated by C, 1 and
{d1 , d2 , . . .}, and let T1 be the natural extension of T to C1 . Then C1 , V := δ∞ ◦ T1
and a strictly positive element h of C1 ∩D fulfills condition (α) of Proposition 5.4.1.
We show that condition (β) of Proposition 5.4.1 is also satisfied:
Since C1 is a separable subalgebra of M(D) ⊂ M(B) and since, for a ∈ B+ , the
map c ∈ C1 7→ aV (c)a is nuclear, by Proposition 3.2.15(i), for c1 , . . . , cn ∈ C1 and
ε > 0, there exists d ∈ B with kd∗ cj d − aV (cj )ak < ε .
Thus C1 and V := δ∞ ◦ T1 are unital and satisfy the assumptions (α) and
(β) of Proposition 5.4.1 and the assumptions of Propositions 5.4.1(ii) and 5.4.1(iv).
Hence (i) follows from Proposition 5.4.1(ii) and (iv).
Moreover, V is a C *-morphism if the original T : C → M(B) is a C *-morphism,
and (ii) follows then from Proposition 5.4.1(iii). 

We explain now the important role of exactness in our consideration, and

why we need exactness at certain places as assumptions for generalized Weyl–von-
Neumann theorems.

Corollary 5.6.3. A C*-algebra A is exact if and only if, for every σ-unital
C*-algebra B, every weakly nuclear map V : A → M(B) is nuclear.

Proof. Suppose that each weakly nuclear V : A → M(B) is nuclear if B

is σ-unital. Let A ⊂ L(H) for some Hilbert space H and C ⊂ A a separa-
ble C *-subalgebra of A. Then there exists a faithful *-representation d : C →
M(K) ∼ = L(H1 ) of C on a separable Hilbert space H1 . Since M(K) is injective,
by Arveson extension theorem, [43], d extends to a completely positive contrac-
tion V : A → M(K). B := K is σ-unital and nuclear. Thus V is weakly nuclear
and therefore nuclear by assumption. d−1 : d(C) → L(H) extends to a completely
 6. W-VN TYPE RESULTS FOR WEAKLY NUCLEAR MAPS 733

positive contraction T : M(K) → L(H) by injectivity of L(H). T V is nuclear and

T V (a) = a for a ∈ C. Thus the inclusion map A ,→ L(H) is nuclear. But this is
equivalent to the exactness of A, cf. Remark 3.1.2(ii).
Now suppose conversely that A is exact and that B is σ-unital and V : A →
M(B) is a weakly nuclear contraction.
We can assume that A and V are unital: If A is not unital, then V1 (a + z1) =
V (a) + z1 is a unital extension to the unitization of A. The u.c.p. mapV1 is still
weakly nuclear by Lemma B.7.7(i), and the unitization of A is again exact.
If A is unital, then we take a state ψ on A and consider W (a) = V (a)+ψ(a)(1−
V (1)) for a ∈ A. W is unital and weakly nuclear. The tensor criteria in Remark
3.1.2(i) shows that V is nuclear if W is nuclear.
It suffices to approximate V on unital separable subalgebras C of A by unital
nuclear maps. The subalgebras of A are again exact by Remark 3.1.2(ii) and by
the Arveson extension theorem. Therefore, by Remark 3.1.2(ii), H0 : C → M(K) ⊂
M(B ⊗ K) is nuclear for a faithful unital *-monomorphism H0 : C → M(K) with
H0 (C) ∩ K = {0}, e.g. take H0 := δ∞ ◦ ρ for some unital faithful *-representation
ρ : C → L(H).
c ∈ C 7→ V (c) ⊗ 1M(K) ⊂ M(B ⊗ K)
is still weakly nuclear. Therefore, by Corollary 5.6.1, there exists a sequence of
isometries Sn ∈ M(B ⊗ K) such that limn kV (c) ⊗ 1 − Sn∗ H0 (c)Sn k = 0 for c ∈ C.
The nuclearity of H0 implies that V |C is nuclear. Thus V : A → M(B) is nuclear,
because the nuclearity is a “local” property of V by Definition 3.1.1. 

Let A a separable C *-algebra and ρ : A → L(H) ∼ = M(K) a faithful non-

degenerate *-representation in general position, i.e., ρ(A) ∩ K = {0}. If B is stable,
the composition H : A → M(B) of ρ with M(K) ,→ M(B) is a non-degenerate
*-monomorphism with H(A) ∩ B = {0}, and H is weakly nuclear. Let H0 :=
πB ◦ H the corresponding *-monomorphism from A into Qs (B) = Q(B). We use
the notation [H0 ] for the unitary equivalence class of H0 : A → Qs (B) by unitaries
in Qs (B).

Corollary 5.6.4 (Equivalence and Dominance in M(B) and Qs (B)). Sup-

pose that B is σ-unital and stable, Di ⊂ M(B) are σ-unital stable C*-subalgebra
with Di B dense in B, for i = 1, 2, and that A is a separable C*-algebra which is
unital or is stable. Furthermore, assume that (for i = 1, 2) Di is simple and purely
infinite or Di ∼= K.
Further let H and H0 be as above.

(i) Let h1 , h2 : A → Qs (B) be *-monomorphisms such that h1 (A) ⊂

πB (M(D1 )), h1 (A) ∩ πB (D1 ) = 0, and that h2 has a completely positive
weakly nuclear lift T2 : A → M(B). (Suppose in addition that that
hi (1) = 1 for i = 1, 2, if A is unital.)
Then h1 dominates h2 in Qs (B).
734 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

(ii) Let hi : A → M(Di ) ⊂ M(B) be weakly nuclear *-monomorphisms such

that hi (A)∩Di = 0 for i = 1, 2. (Suppose, in addition, that hi (1) = 1M(B)
for i = 1, 2 if A is unital.)
Then h1 and h2 are unitarily homotopic in the sense of Definition
5.0.1.
(iii) S(H0 , A, Qs (B)) is the ⊕-semigroup of the unitary equivalence classes of
those C*-morphisms from A into Qs (B) that have a weakly nuclear com-
pletely positive lift, and which are unital if A is unital.
(iv) If B itself is simple and purely infinite, then, for every *-monomorphism
h from A into Qs (B), the Cuntz sum h ⊕ H0 is unitarily equivalent to h
in Qs (B). (Here h is assumed to be unital, if A is unital.) In particular,

S(H0 , A, Qs (B)) ∩ [Mon(A, Qs (B))] = G(H0 , A, Qs (B)) .

Proof. In the stable case the unitized monomorphisms are still monomor-
phisms. Therefore it suffices to consider the unital case.
(i): Note that πB T2 = h2 . Consider a unital separable C ⊂ M(D1 ) such
that C contains a strictly positive element of D1 and πB (C) = h1 (A). Let T :=
T2 h−1
1 πB |C. T is weakly nuclear. Thus Theorem 5.6.2(i) or Corollary 5.6.1(i)
applies. Let s := πB (S(1)). Then s is an isometry in Qs (B) and s∗ h1 (·)s = h2 .
(ii): Let Ck the C *-algebra which is generated by hk (A) and a strictly positive
element of Dk , (k = 1, 2). Then there are unique epimorphisms ψk from Ck onto
A with ψk hk = idA .
Let T1 := h2 ψ1 and T2 := h1 ψ2 . Then Ck and Tk : Ck → M(B) satisfy
the assumptions of Theorem 5.6.2(ii) or of Corollary 5.6.1(ii). Thus we find norm
continuous maps t 7→ Uk (t) into the unitaries of M(B) such that, for a ∈ A, t ∈ R+ ,

(hk (a) ⊕ Tk (hk (a))) − Uk (t)∗ hk (a)Uk (t) ∈ B

and
lim k(hk (a) ⊕ Tk (hk (a))) − Uk (t)∗ hk (a)Uk (t)k = 0.
t→∞
Since T1 h1 = h2 and T2 h2 = h1 , we get the desired result with the unitaries
U (t) = U1 (t)V0 U2 (t)∗ , where V0 is a unitary with V0∗ (a ⊕ b)V0 = b ⊕ a.

By Chapter 4, (iii) and (iv) are special cases of (i). 

7. W-vN type results for simple p.i. algebras

The Part (ii) of below given Corollary 5.7.1 says among others that the ab-
sorption criteria of G. Elliott and D. Kucerovsky [264] (cf. [310] for the non-unital
case) is satisfied for all pi-sun algebras.

Corollary 5.7.1. Let B a nonzero σ-unital C*-algebra and t1 , t2 isometries

in M(B ⊗ K) with t1 t∗1 + t2 t∗2 = 1.
The following conditions on B are equivalent:
 7. W-VN TYPE RESULTS FOR SIMPLE P.I. ALGEBRAS 735

(i) B is stably isomorphic to a unital purely infinite simple C*-algebra or to

C;
(ii) For every unital separable C*-subalgebra C of M(B ⊗ K) such that C ∩
(B ⊗ K) contains a strictly positive element of B ⊗ K and for every unital
weakly nuclear map V : C → M(B ⊗ K) with V (C ∩ (B ⊗ K)) = {0} there
exists a sequence (sn ) of isometries in M(B ⊗ K) with
(1) V (b) − s∗n bsn ∈ B ⊗ K for all b ∈ C,
(2) kV (b) − s∗n bsn k → 0 for all b ∈ C;
(iii) For every unital separable C*-subalgebra C of M(B⊗K) such that C∩(B⊗
K) contains a strictly positive element of B⊗K, and for every unital weakly
nuclear C*-morphism h : C → M(B ⊗ K) with h(C ∩ (B ⊗ K)) = {0},
there exists a sequence (un ) of unitaries in M(B ⊗ K) with
(1) b ⊕t1 ,t2 ϕ(b) − u∗n bun ∈ B ⊗ K for all b ∈ C,
(2) kb ⊕t1 ,t2 ϕ(b) − u∗n bun k → 0 for all b ∈ C,
for some pair s1 , s2 of generators of O2 in M(B ⊗ K).

Proof. The implication (i)⇒(ii) follows immediately from Theorem 5.6.2(i)

and Corollary 5.6.1(i).
(ii)⇒(iii): If V is unital and weakly nuclear then δ∞ ◦ V is again unital and
weakly nuclear and has the same kernel as V . Therefore the proof of Lemma
5.1.2(vi) can be modified to obtain a proof of the implication (ii)⇒(iii) here. One
has only to replace E there by `∞ (M(B ⊗ K))/c0 (B ⊗ K) here. Then we apply
Proposition 4.3.5(i) as there.
(iii)⇒(i): By Corollary 2.2.11(i), it suffices to show that the stable corona
Qs (B) := M(B ⊗ K)/(B ⊗ K) of B is simple. This follows already from Part (iii,1):
Let a be a positive element of Qs (B) with kak = 1. We want to find an element
d such that d∗ ad = 1. This shows than that Qs (B) is not only simple but is also
purely infinite.
Let χ be a character on C ∗ (a) with χ(a) = 1, b ∈ M(B)+ a contraction with
πB (a) and q ∈ B ⊗ K a strictly positive element of B ⊗ K. Now let C := C ∗ (b, q)
be the C *-algebra generated by b and q. Then C ∩ (B ⊗ K) contains q and we can
apply Part(iii,1). The quotient map π from M(B ⊗ K) onto Qs (B) maps C onto
C ∗ (a) Let us define h : C → M(B ⊗ K) as h(c) = χ(π(c)) · 1. Clearly h is a nuclear
C *-morphism and annihilates C ∩ (B ⊗ K). Hence by Part (iii, α) there exists a
unitary u ∈ M(B ⊗ K) such that

b ⊕s1 ,s2 1 − u∗ bu ∈ B ⊗ K

(recall that h(b) = 1). Recalling the definition of ⊕s1 ,s2 , we get s1 bs∗1 +s2 s∗2 −u∗ bu ∈
B ⊗ K. Hence 1 − s∗2 (u∗ bu)s2 belongs to B ⊗ K. So, by taking d = π(us2 ), d∗ ad = 1
in Qs (B) as desired. 

Now we show a result that is stronger than the result pointed out in the proof
of the implication (iii)⇒(i) of Corollary 5.7.1:
736 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

Corollary 5.7.2. If B is σ-unital, simple and purely infinite and if A is

a simple separable unital C*-subalgebra of the stable corona Qs (B) such that the
inclusion map A ,→ Qs (B) has a completely positive lift T : A → M(B ⊗ K) which
is weakly nuclear, then A0 ∩ Qs (B) is purely infinite.
In particular, A0 ∩ Qs (B) is purely infinite for every simple separable nuclear
C*-subalgebra A ⊆ Qs (B) that contains the unit element of Qs (B).

Proof. By Corollary 5.7.1(iii), idA dominates the restriction of h0 to A ∼

= A⊗1
s
where h0 : A ⊗ O2 → Q (B) is a weakly nuclear liftable unital *-monomorphism,
e.g. coming from a faithful unital *-representation k : A ⊗ O2 → L(H) ∼ = M(K).
Thus A0 ∩ Qs (B) is not isomorphic to C.
Let V : A → M(B ⊗ K) be a weakly nuclear unital completely positive map
with π ◦ V = idA . If C is a unital separable C *-subalgebra of Qs (B) and h : C → A
is a unital C *-morphism, then V ◦ h is a weakly nuclear lift of h. Thus by Corollary
5.7.1(ii) there exists an isometry s ∈ Qs (B) with h(c) = s∗ cs for c ∈ C.
This shows that A ⊂ Qs (B) satisfies the criteria in Proposition 2.2.5(v), and,
therefore, A0 ∩ Qs (B) must be simple and purely infinite. 

Next stable absorption theorem characterizes the unital nuclear purely infinite
simple algebras B and B = C as the only C *-algebras by the “naive” one-to-one
generalization of the classical Weyl–von-Neumann theorem. It shows that a useful
more general generalization of the WvN-theorem needs more elaborate assumptions
and formulations, e.g. as ours.

Corollary 5.7.3. For a σ-unital C*-algebra B the following properties (i)

and (ii) are equivalent:

(i) B is stably isomorphic to C or is stably isomorphic to a simple purely

infinite unital nuclear C*-algebra A.
(ii) For every unital separable C*-algebra C ⊂ M(B ⊗ K) and every unital


C*-morphism h : C → M(B ⊗ K) with h C ∩ (B ⊗ K) = 0 there exists a
unitary U ∈ M(B ⊗ K) such that c ⊕ h(c) − U ∗ c U ∈ B ⊗ K for all c ∈ C.

Proof. (i)⇒(ii): Let E := B ⊗ K. Every C *-morphism k : F → M(E) is

weakly nuclear, because E is nuclear. Therefore, in this special case, the compo-
sition C + E → (C + E)/E ∼ = C/(C ∩ B) → M(B) is again weakly nuclear if
h(C ∩ (B ⊗ K)) = 0, and we can assume w.l.o.g. that C contains a strictly positive
element of E. Then (ii) follows from Corollary 5.7.1(iii,1).
(ii)⇒(i): We modify the proof of the implication (iii)⇒(i) of Corollary 5.7.1
as follows. Let C := C ∗ (b, 1), with b as there. Then we don’t use the assumption
that B is σ-unital, that we have used in the proof of Corollary 5.7.1 to check the
conditions in Corollary 2.2.11(i), because we can do here the same with every unital
separable C ⊂ M(B ⊗ K) by our assumptions in (ii).
 7. W-VN TYPE RESULTS FOR SIMPLE P.I. ALGEBRAS 737

By Corollary 2.2.11(i), this gives that B is σ-unital simple and purely infinite
and thus is stably isomorphic to a simple purely infinite unital C *-algebra A by
[172] and by L.G. Brown’s stable isomorphism theorem (cf. Corollary 5.5.6).
Let A unital with A⊗K ∼
= B ⊗K and let D be a separable unital C *-subalgebra
∼
of A and R : D → L(H) = 1A ⊗ M(K) a faithful unital *-representation with
R(D) ∩ K(H) = 0. Then (ii), for C := R(D), implies that there is an isometry
S ∈ M(A ⊗ K) with

S ∗ (1A ⊗ R(d))S − d ⊗ 1M(K) ∈ A ⊗ K .

If we use a sequence of states ψn on K which converges weakly to zero, then

we get that D ,→ A is a point-wise limit of the unital completely positive
maps d 7→ (idA ⊗ψn )(S ∗ 1A ⊗ R(d)S), which are nuclear. Here we have to
extend (idA ⊗ψn ) : A ⊗ K → A naturally to a unital completely positive map
M(A ⊗ K) → A. Thus A is nuclear. 

Remark 5.7.4. Above considerations show that the range of generalized Weyl–
von Neumann theorems seems to be very limited almost to extensions by K or stable
simple purely infinite algebras.
But if we consider special situations, i.e., additional requirements on C and V
or h, then Proposition 5.4.1 allows to deduce other types of generalizations of the
Weyl–von Neumann theorem in the spirit of Voiculescu’s generalization of BDF-
theory, e.g. some equivariant cases. Here we state a general result inspired by
E. Blanchard [89]. We do not use directly the fundamental conditions (α) and
(β) of the generalised Weyl–von-Neumann theorem but apply more comfortable
assumptions that imply conditions (α) and (β):
Assume that B is σ-unital and stable and that C ⊂ M(B) is unital and sepa-
rable. Let T : C → M(B) a unital completely positive map such that:

(α0 ) T (C ∩ B) = 0 and C ∩ B contains a strictly positive element of B;

(β 0 ) For every (non-degenerate) factorial representation ρ : B → L(H) there
exists a net Wτ of inner unital completely positive maps Wτ (a) = d∗k adk
P

from N := ρ(B)00 into N , such that ρ(V (c)) is the point-wise weak limit of
Wτ (ρ(c)) for every c ∈ C, where we have extended ρ naturally to a weakly
continuous homomorphism from B ∗∗ ⊃ M(B) into N .
(β 00 ) For every b ∈ B+ , c1 , . . . , cm ∈ C, ε > 0, n ∈ N there exist s1 , . . . , sn ∈
M(B) such that s∗i sj = δij and kck si b − si ck bk < ε for i = 1, . . . , n,
k = 1, . . . , m.

Then the conclusions (ii), (iii) and (iv) Proposition 5.4.1 hold.
Idea of the proof: Clearly (α0 ) implies (α) of Proposition 5.4.1.
A combination of a Hahn-Banach separation argument (which yields a ten-
sor product characterization of approximately inner completely positive maps,
cf. Chapter 3 or [443, appendix]) with a modification of the argument in the proof
738 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

of Proposition 2.2.5(iii) shows that (β 0 ) and (β 00 ) together imply (β) of Proposition

5.4.1 for V := δ∞ ◦ T .

8. Extensions and Ext(C; A, B)

Parts of the proofs of our theorems in Chapter 1 consists in showing that a

certain strong equivalence of extensions (related to the theorem in question) can be
deduced from – in a “formal sense” – considerably weaker equivalence relations on
extensions. In particular, we have to deduce that certain extensions are actually
split ( 17 ). It requires that the extensions that absorb certain split extensions
need to be characterized with help of generalized Weyl–von-Neumann–Voiculescu
theorems. This will be discussed here in this section.
We need urgently the following cases:
(1) B = O2 ⊗ K, A unital and separable (and exact), ϕ ∈ Hom(A, Q(B)) unital
(!) and with (automatically weakly nuclear) unital c.c. split T : A → M(B) with
πB ◦ T = ϕ.
If A is exact, then T is automatically nuclear. Thus, this is then the case if
ϕ : A → Q(B) is nuclear.
Is there an “elementary way” to find a splitting by C *-morphism?
E.g. it could be that ϕ extends to A ⊗ O2 , i.e., that one can find a unital copy
of O2 in ϕ(A)0 ∩ Q(B) : Then one can take a faithful unital nuclear *-representation
ρ : A → L(`2 ) = M(K) and define ψ : A → Q(B) by ψ(a) = πB⊗K (1O2 ⊗ ρ(a)). ψ
is again nuclear and commutes with a copy of O2 in Q(B).
Then ψ, ϕ 1-step dominate each other (needs a proof). It follows then from
results in Chapter 4, that they are unitary equivalent in Q(B). This defines a
unital splitting C *-morphism for ϕ.
To get such things there are two ways:
Replace A by A ⊗ O2 (covers the exact case), or try to find a way to prove that
ϕ ⊕ ϕ is unitary equivalent to ϕ in Q(B) if A is nuclear. (this should be necessary
to get the desired additional observations in case of nuclear A).
We can attempt to first use that A = O2 ⊗ O2 ⊗ · · · has this property (needs
homotopy invariance of Ext(A, O2 ) ?).
We find every exact C *-algebra as sub quotient of O2 in Q(B), such that each
c.p. lift sits in a copy of B ⊆ V (O2 ) ∩ M(B) for a u.c.p. map V : O2 → M(B).
Get then unital embedding into O2 , but first only of extensions of them by B.
Used later:
(1a) If ϕ and ψ : A → Q(B) have unital splits, then ψ and ϕ are unitarily
equivalent by a unitary in Q(B).
Moreover this should be the case if B = O2 ⊗ K
17 I.e., is not only stably split after adding a certain split extension.
 8. EXTENSIONS AND Ext(C; A, B) 739

One should check if ψ extends to ?????

Because, by Kasparov, Voiculescu, ????
(1b) U(Q(B)) = U0 (Q(B)) deduced from K1 -injectivity and K1 (Q(B)) = 0.
(By property (sq) of Q(B), K∗ (M(B)) = 0 and K∗ (O2 ) = 0. Ref. from Chp. 4)
(2) B arbitrary σ-unital, stable strongly p.i. C *-algebra. C given by

H0 : A → M(B)

non-degenerate, A separable and stable, with the property that H0 is unitarily

homotopic to δ∞ ◦ H0 .
2
Are δ∞ and δ∞ really unitarily homotopic?
Yes, they are, because they are unitary equivalent and each unitary equivalence
causes “homotopy” in M(B) by a norm-continuous path of unitaries in M(B) if B
stable and σ-unital, because then U(M(B)) = U0 (M(B)).
It is not clear if one can move them with a path U (t) in the unitaries such that

U (t)∗ δ∞
2
(H0 (a))U (t) − δ∞ (H0 (a)) ∈ B

for all a ∈ A.
But this should be possible if H0 = δ∞ ◦ h0 for some h0 : A → B ???
This could be almost possible with a “long” strictly continuous path W (t) with
lim W (t) = U for W (t) = exp(h1 (t)) · . . . · exp(hn (t)), U = exp(k1 ) · . . . · exp(kn ),
h` (t) → k` strictly, h` (t)∗ = −h` (t), k`∗ = −k` .
In the corona it does nothing.
Busby invariants in question are:
ϕ : A → Q(B)
with c.p. splits V : A → M(B): i.e., b∗ V (·)b ∈ C for each b ∈ B.
Pre-assumption in trivially graded case:
A and B stable, A separable, B σ-unital, C ⊆ CP(A, B) point-norm closed,
countably generated, non-degenerate m.o.c. cone.
HC : A → B non-degenerate in general position, in 1-1-relation to C, h0 :=
πB ◦ HC .

Ext(C; A, B) ∼
= kernel of K0 (h0 (A)0 ∩ Q(B)) → K0 (Q(B)) .

This is, because we consider only the zero-dominating maps in S(h0 ; A, Q(B)), i.e.,
where the extension is stable.
Since the σ-unital B is stable, and A is stable and σ-unital, it means equiv-
alently that we consider only stable extensions of A with B, by Proposition ??.
740 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

Alternatively: Classes of projections p ∈ M(B) with [p, h0 (a)] ∈ B for all

a ∈ A, equivalence modulo B in h0 (A)0 ∩ Q(B) (alternatively: isometries T with
T T ∗ = p, T ∗ T = p0 , T commutes mod B with h0 (A))
Unitary equivalence classes [ϕ] by unitaries in πB (U(M(B))) = U0 (Q(B)) de-
serve as elements as elements of SExt(C; A, B).
SExt(C; A, B) := S(πB ◦ H0 ; A, Q(B)).
implies Ext(C; A, B) ∼
= Gr(SExt(C; A, B))
Needs “absorbing elements”.
Or / Respectively, other (not equivalent) definition ????:
Ext(C; A, B) := [πB ◦ H0 ] + SExt(C; A, B)
The class [πB ◦ H0 ] deserves as a zero element.
The H0 for C := CPnuc (X; A, B), respectively C := CPnuc (A, B), and (if A and
B are both separable) C := CPnuc (A, B) should be given.
(3) Specify the “nuclear cases” for A unital, exact, and separable, and B simple
s.p.i., σ-unital, stable and separable.
Then Qs (B) = M(B)/B is simple and p.i. If ϕ, ψ : A → Qs (B) are two
unital nuclear *-monomorphisms. Then there exist isometries S, T ∈ Qs (B) with
S ∗ ϕ(·)S = ψ and T ∗ ψ(·)T = ϕ.
Need that in case B = O2 ⊗ K and separable unital C ⊆ Qs (B) that the unit
of C 0 ∩ Qs (B) is properly infinite in C 0 ∩ Qs (B), and K∗ (C 0 ∩ Qs (B)) = 0.
Can follow form ????
In a first step one could consider O2 ⊗ O2 ⊗ · · · in place of A.
Implies that O2 ⊗ O2 ⊗ · · · is unitally contained in O2 ?
Need: All unital *-endomorphisms of O2 and of O2 ⊗O2 ⊗· · · are approximately
inner.
Must be integrated in Chapter 3.
(4) Discuss:
Absorption like Elliott/Kucerovsky (+ correction by J. Gabe)
(5) Existence and uniqueness – up to unitary homotopy – of universal H0 : A →
M(B ⊗ K). In which cases of C?
In case of B a pi-sun algebra, there exists only CP(A, B) = CPnuc (A, B).
Then each nuclear c.p. map into Qs (B) is 1-compression of H0 and one has to
determine the K1 (πB⊗K ◦ H0 (A)0 ∩ Qs (B)).
The quickest way should be to show that O2 ∼= O2 ⊗ O2 ⊗ · · · and that it causes
that for each separable subset X of Q := Qs (O2 ⊗ O2 ⊗ · · · ) there exists a copy of
O2 unitally in X 0 ∩ Q.
(6??)
 8. EXTENSIONS AND Ext(C; A, B) 741

The definition of Elliott/Kucerovsky/Gabe of an “absorbing extension” is:

“An extension that absorbs any weakly nuclear extension”
Ell/Kuc in Pacific MJ: (Unital) “purely large” extensions absorb all (unitally)
weakly nuclear extensions.
The extension 0 → B → E → A → 0 is “purely large” (by Definition of “purely
large”) if for each x ∈ E \ B the hereditary C *-subalgebra x∗ Bx of B contains a
stable hereditary C *-subalgebra D ⊆ x∗ Bx that is full in B.
My observation:
Write B ∼= B ⊗ K for K := K(`2 (N)) The extension E must necessarily absorb the
extension given by Busby invariant πB ◦ h where h : A 3 a 7→ 1M(B) ⊗ δ∞ (d(a)) ∈
M(B) ⊗ L(`2 ) ⊆ M(B ⊗ K) ∼ = M(B) .
Here, d : A → L(`2 ) is any faithful C *-morphism that has the property that
d(A)`2 is not dense in `2 .
Full hereditary stable C *-subalgebras D of B ⊗ K contain an in D strictly
positive element e ∈ D+ that is M-vN equivalent in B ⊗ K to a strictly positive
element b ∈ (B ⊗ K)+ .
How to use it to show our assumptions for the absorption Theorem ??? for
A ⊆ Qs (B) and (πB ◦ h) : A → Qs (B)
An (unital) extension 0 → B → E → A → 0 is “(unitally) weakly nuclear” if
there exists a (unital) weakly nuclear c.p. map V : A → E with the property that
πB ◦ V is the Busby invariant of the extension.
Here V is “weakly nuclear” if A 3 a 7→ b∗ V (a)b ∈ B is nuclear for all b ∈ B.
(Kucerovsky/Ng, Houston J.Math.32, 2006): Any “full” extension by a σ-unital
stable C *-algebra B with (CFP) is “purely large”.
Here “full” means that each non-zero element of A ⊆ E/B is full in E/B.
Corona factorization property (CFP) (of stable C *-algebras B):
Each full projection P ∈ M(B) is properly infinite in M(B). (P is range of a
isometry in M(B)?)
J. Gabe: Lemma 1.2 :
There exists non-unital purely large extension such that the unitization is not
purely large.
Final Th.??
We recall some basic definitions from the theory of extensions: For every short
exact sequence 0 → B → E → A → 0 of C *-algebras we consider the Busby
diagram
0 /B /E /A /0

 
0 /B / M(B) / Q(B) /0
π
742 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

where Q(B) denotes the corona algebra M(B)/B. The right vertical arrow ϕ ∈
Hom(A, Q(B)) is the Busby invariant of the extension. There is a natural *-
isomorphism f ∈ E 7→ (η(f ), ρ(f )) ∈ Eϕ from E onto the pull-back

Eϕ := A ⊕ϕ,πB M(B) ⊂ A ⊕ M(B) ,

where η : E → A is the defining epimorphism of the extension E and ρ(f )b =

ι−1 (f ι(b)) for f ∈ E, b ∈ B and the defining monomorphism ι : B → E of the ex-
tension. Thus, the extension is (up to natural *-isomorphisms) uniquely determined
by the Busby invariant ϕ ∈ Hom(A, Q(B)).
For ϕ, ψ ∈ Hom(A, Q(B)), we write ϕ ≈ ψ if ψ(a) = π(u)ϕ(a)π(u∗ ), a ∈ A, for
some unitary u ∈ M(B), and call the extensions strongly equivalent . Then Eψ
and Eϕ are isomorphic as C *-algebras by an isomorphism that induces the identity
map on A and the automorphism b 7→ u∗ bu of B. That will be applied in the proofs
of Theorems A and B, and we need to establish strong equivalence in some special
cases from essentially weaker definitions of equivalence.
A Busby invariant ϕ ∈ Hom(A, Q(B)) is called trivial if there exists a C *-
morphism ϕ̂ ∈ Hom(A, M(B)) such that the diagram

A
ϕ
b
ϕ
v 
M(B) / Q(B)
π

commutes, i.e., if ϕ is liftable. Then Eϕ is a split extension, because the

natural epimorphism η from Eϕ onto A has a split morphism ηb : A → Eϕ with
η ◦ ηb = idA . (We do not suppose that the lift is unital if A and ϕ are unital.)
Now assume that M(B) contains a copy C ∗ (s1 , s2 ) of O2 unitally. (This is the
case e.g. for stable B, where M(B) ∼ = M(B ⊗ K) and Q(B) ∼ = Qs (B) := Q(B ⊗ K).)
Then Q(B) also contains the copy C ∗ (t1 , t2 ) of O2 unitally, where tk = πB (sk )
(k = 1, 2), and
Cuntz addition ⊕ := ⊕t1 ,t2 is defined on the classes [ϕ]≈ :
see Proposition 4.3.2 and notice that only canonical generators r1 , r2 ∈ Q(B) with
property t1 r1∗ + t2 r2∗ ∈ πB (U(M(B))) are considered.
We write ϕ ∼ ψ if there exist trivial elements τ1 , τ2 ∈ Hom(A, Q(B)) such that
ϕ ⊕ τ1 ≈ ψ ⊕ τ2 . In particular ϕ ∼ (ϕ ⊕ 0), because ϕ ⊕ 0 = ϕ ⊕ (0 ⊕ 0) ≈ (ϕ ⊕ 0) ⊕ 0
by Proposition 4.3.2(iv). Thus, we find for the ∼-classes [ϕ]∼ always a representing
element ϕ ∈ Hom(A, Q(B)) such that the inclusion map of the C *-subalgebra
−1
πB (ϕ(A)) in M(B) dominates zero (in the sense of Definition 4.3.3).
The *-morphism ϕ ⊕t1 ,t2 0 is unitarily equivalent to ψ ⊕t1 ,t2 0 in Q(B) if and
only if ϕ ⊕ 0 and ψ ⊕ 0 are conjugate by a unitary in the connected component
U0 (Q(B)) of 1 in the unitaries U(Q(B)) of Q(B) (cf. Proposition 4.3.6(iv,c)).


Since πB U0 (M(B)) = U0 (Q(B)), it follows that ϕ ∼ ψ if and only if, there
are “trivial” Busby invariants τ1 , τ2 ∈ Hom(A, Q(B)) such that ϕ ⊕ τ1 ⊕ 0 and
 8. EXTENSIONS AND Ext(C; A, B) 743

ϕ ⊕ τ2 ⊕ 0 are unitarily equivalent in Q(B). Hence, the relation ∼ is compatible

with Cuntz addition.
Recall that Ext(A, B) (or better notated as SExt(A, B)) is the semigroup
Hom(A, Q(B))/ ∼, where the zero element is the class of “trivial elements”. Let
Ext−1 (A, B) denote the group of invertible elements in Ext(A, B), i.e., of ∼-classes
of elements ϕ ∈ Hom(A, Q(B)) such that there are ψ, τ1 , τ2 ∈ Hom(A, Q(B)) such
that (ϕ ⊕ ψ) ⊕ τ1 = τ2 and τ1 , τ2 are trivial. Notice that V := ((s1 )2 )∗ τb2 (·)(s1 )2 is
a completely positive lift of of ϕ if ϕ is in Ext−1 (A, B).

In the remaining part of this section we suppose in addition, that B is stable

and σ-unital and that A is separable and is unital or stable ( 18 ).
Then, conversely, every completely positive liftable ϕ ∈ Hom(A, Q(B)) is “in-
vertible” in Ext(A, B), because for a suitable Kasparov-Stinespring dilation τb : A →
M2 (M(B)) ∼ = M(B) of a contractive c.p. lift V of ϕ there is ψ ∈ Hom(A, Q(B))
with ϕ ⊕ ψ ⊕ 0 ≈ τ ⊕ 0. Thus, for stable σ-unital B, Ext−1 (A, B) consists of the
the ∼-classes of completely positive liftable elements of Hom(A, Q(B)) (with Cuntz
addition).
Moreover, for trivial Busby invariants τ1 , τ2 ∈ Hom(A, Q(B)), there exists triv-
ial τ3 ∈ Hom(A, Q(B)) with

τ1 ⊕ τ3 ≈ τ3 ≈ τ2 ⊕ τ3 ,

(Indeed, τ3 := πB ◦ δ∞ ◦ (τb1 ⊕ τb2 ) does the job, because, with τb3 := δ∞ ◦ (τb1 ⊕ τb2 ),
τb1 ⊕ τb3 and τb2 ⊕ τb3 are both unitarily equivalent to τb3 in M(B) by Lemma 5.1.2(i,ii).)
It follows that h1 ∼ h2 for h1 , h2 ∈ Hom(A, Q(B)) if and only if there is trivial
τ ∈ Hom(A, Q(B)) with [h1 ] + [τ ] = [h2 ] + [τ ] (and [τ ] + [τ ] = [τ ]) in the semigroup
of unitary equivalence classes ([Hom(A, Q(B))], +).
(Indeed: h1 ∼ h2 ⇔ there exists liftable τ with h1 ⊕τ ≈ h2 ⊕τ ⇒ [h1 ]+[τ ] = [h2 ]+[τ ]
⇒ [h1 ⊕ τ ⊕ 0] = [h2 ⊕ τ ⊕ 0] ⇒ h1 ⊕ (τ ⊕ 0) ≈ h2 ⊕ (τ ⊕ 0). The last implication
comes from the K1 -injectivity of Q(B).)
Recall that Gr(S) denotes the Grothendieck group of an Abelian semigroup S.
The above observations imply – for σ-unital stable B – that Ext−1 (A, B) has the
following equivalent description as Grothendieck group of a certain semigroup:
Let Homl-cp (A, Q(B)) denote the set of Busby invariants ϕ : A → Q(B) that
have a completely positive lift V : A → M(B) ( 19 ). The set of unitary equivalence
classes [Homl-cp (A, Q(B))] of elements in Homl-cp (A, Q(B)) by unitaries in Q(B)
are invariant under Cuntz addition [ϕ] + [ψ] := [ϕ ⊕ ψ] (where we now have to allow
all generators t1 , t2 of a unital copy of O2 in Q(B) for the definition of ⊕).

18 Then ϕ ≈ ψ if and only if there is v ∈ U (Q(B)) with v ∗ ϕ(·)v = ψ , because U (Q(B)) =

0 0
πB (U0 (M(B))) and the unitary group of M(B ⊗ K) ∼ = M(B) is connected by [180]. Moreover,
one needs only that [v] = 0 in K1 (Q(B)), because Q(B) is K1 -injective if B is stable and σ-unital,
cf. Lemma 5.5.9(ii), or Proposition 4.2.15.
19 i.e., π ◦ V = ϕ, one can always replace V by a completely positive contraction with this
B
property
744 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

Ext−1 (A, B) is naturally isomorphic to the Grothendieck group

Gr([Homl-cp (A, Q(B))]) of the commutative semigroup ([Homl-cp (A, Q(B))], +).
(Indeed: the isomorphism is induced by the semigroup morphisms

[ϕ]≈ 7→ [ϕ] 7→ [ϕ]Gr

for ϕ ∈ Homl-cp (A, Q(B)), because, if [ϕ] + [λ] = [ψ] + [λ] for λ ∈ Homl-cp (A, Q(B))
then there exist κ ∈ Homl-cp (A, Q(B) and a liftable τ ∈ Homl-cp (A, Q(B)) such
that [τ ] = [λ ⊕ κ], and, therefore ϕ ⊕ τ ⊕ 0 ≈ ψ ⊕ τ ⊕ 0 . I.e., ϕ ∼ ψ if [ϕ]Gr = [ψ]Gr
for ϕ, ψ ∈ Homl-cp (A, Q(B)).
Conversely, ϕ ∼ ψ implies the existence of a liftable τ ∈ Homl-cp (A, Q(B)) with
ϕ ⊕ τ ≈ ψ ⊕ τ , which implies that [ϕ] + [τ ] = [ψ] + [τ ], i.e., that [ϕ]Gr = [ψ]Gr .)

We introduce the semigroup SExtnuc (A, B) and the group Extnuc (A, B) in a
similar way:
Let Homl-nuc (A, Q(B)) denote the elements ϕ ∈ Hom(A, Q(B)) which have weakly
nuclear lifts V : A → M(B). Note that h1 ≈ h2 if and only if u∗ h1 (·)u = h2 by
a unitary u ∈ U0 (Q(B)), that h1 ≈ h2 implies (by definition) that u∗ h1 (·)u =
h2 for a unitary u ∈ Q(B), and that u∗ h1 (·)u ∈ Homl-nuc (A, Q(B)) if h1 ∈
Homl-nuc (A, Q(B)) and u ∈ Q(B) is unitary: Indeed, V 0 (a) := d∗ V (a)d is a weakly
nuclear completely positive lift of u∗ h1 (·)u if πB (d) = u and V : A → M(B) is a
weakly nuclear lift of h1 . Thus, the subset Homl-nuc (A, Q(B)) ⊂ Hom(A, Q(B))
is closed under ≈, unitary equivalence and under ⊕: if ϕ, ψ ∈ Hom(A, Q(B)
have weakly nuclear lifts and χ ≈ ϕ, then χ and ϕ ⊕ ψ have weakly nuclear
lifts. It follows that Cuntz addition ⊕t1 ,t2 is defined on the set of classes [ϕ]≈
with ϕ ∈ Homl-nuc (A, Q(B)) for t1 = πB (T1 ), t2 = πB (T2 ), where the isometries
T1 , T2 ∈ M(B) are canonical generators of a copy of O2 in M(B). We write
ϕ ∼nuc ψ if there exist weakly nuclear C *-morphisms H1 , H2 : A → M(B) such
that ϕ ⊕ (πB ◦ H1 ) ≈ ψ ⊕ (πB ◦ H2 ). Note that h1 ∼nuc h2 implies h1 ∼ h2 , and it
happens that h1 ∼ h2 for h1 ∈ Homl-nuc (A, Q(B)) but h2 6∈ Homl-nuc (A, Q(B)) if
B is simple and is not nuclear (We do not know if h1 ∼ h2 implies h1 ∼nuc h2 for
h1 , h2 ∈ Homl-nuc (A, Q(B)).) The relation ∼nuc is compatible with the Cuntz ad-
dition on [Homl-nuc (A, Q(B))]≈ and h1 ∼ h2 if and only if there is a weakly nuclear
C *-morphism H : A → M(B) such that h1 ⊕ π ◦ H ⊕ 0 is unitarily equivalent to
h1 ⊕ π ◦ H ⊕ 0 by a unitary in Q(B). Since for every ϕ ∈ Homl-nuc (A, Q(B))
there exists a contractive weakly nuclear c.p. map V : A → M(B) and since
there is a weakly nuclear C *-morphism H : A → M(B) and an isometry S ∈
M(B) with V = S ∗ H(·)S, we can see that there is ψ ∈ Homl-nuc (A, Q(B)) such
that ψ ⊕ ϕ ⊕ 0 is unitarily equivalent to πB ◦ H ⊕ 0. We get that the quotient
([Homl-nuc (A, Q(B)]∼nuc , +) of the semigroup [Homl-nuc (A, Q(B))]≈ with respect
to the relation ∼nuc is a group. One can see that this group is isomorphic to
the below defined group Extnuc (A, B) (the nuclear Ext-group in the below given
Definition 5.8.2). The proof of the isomorphism

Homl-nuc (A, Q(B)) / ∼nuc ∼

= Extnuc (A, B)
 8. EXTENSIONS AND Ext(C; A, B) 745

is almost verbatim the above outlined proof of the natural isomorphism

Ext−1 (A, B) ∼
= Gr([Homl-cp (A, Q(B))], +).

If A is separable and unital, then we can consider the set Homul-nuc (A, Q(B))
of unital morphisms ϕ : A → Q(B) in Homl-nuc (A, Q(B)).
The subset Homul-nuc (A, Q(B)) ⊂ Homl-nuc (A, Q(B)) is closed under uni-
tary equivalence (hence under ≈-equivalence) and under Cuntz addition ⊕. The
Grothendieck group

Extu,strong
nuc (A, B) := Gr([Homul-nuc (A, Q(B))]≈ , +)

can be described equivalently by defining relations ∼u−nuc :

We write ϕ ∼u−nuc ψ if there are unital weakly nuclear C *-morphisms
H1 , H2 : A → M(B) such that ϕ ⊕ πB ◦ H1 ≈ ψ ⊕ πB ◦ H2 .
Then H3 := δ∞ ◦ (H1 ⊕ H2 ) is unital, weakly nuclear and is unitarily equivalent
to H3 ⊕ H1 and to H3 ⊕ H2 , and we get: ϕ ∼u−nuc ψ, if and only if, there is a
unital weakly nuclear *-morphism H : A → M(B) with ϕ ⊕ πB ◦ H ≈ ψ ⊕ πB ◦
u,strong
H. It follows that ϕ ∼u−nuc ψ implies that [ϕ]Gr = [ψ]Gr in Extnuc (A, B).
Conversely, if [ϕ]Gr = [ψ]Gr then there exists χ ∈ Homul-nuc (A, Q(B)) with ϕ ⊕ χ ≈
ψ ⊕ χ. By definition of Homul-nuc (A, Q(B)), there is a weakly nuclear c.p. lift
W : A → M(B) with πB ◦ W = χ. By a standard argument there are b ∈ B+
with kbk ≤ 1 and a unital weakly nuclear c.p. map U : A → M(B) such that
W 0 (a) := (1 − b2 )1/2 W (a)(1 − b2 )1/2 + bU (a)b satisfies kW 0 (1) − 1k < 1/2. Thus,
V (a) := W 0 (1)−1/2 W 0 (a)W 0 (1)−1/2 defines a unital weakly nuclear c.p. map with
πB ◦ V = χ. Let C := CPnuc (A, B)) in the (more general) Corollary 5.4.6. It
shows that there is a unital weakly nuclear C *-morphism H : A → M(B) and
isometries S, T ∈ M(B) with SS ∗ + T T ∗ = 1 and V (a) − S ∗ H(a)S ∈ B for all
a ∈ A. By Lemma 4.3.4(i) (applied to g := πB (SS ∗ ) and h1 := πB ◦H) we get that
SS ∗ H(a) − H(a)SS ∗ ∈ B and SV (a)S ∗ + T T ∗ H(a)T T ∗ − H(a) ∈ B for all a ∈ A,
i.e., χ ⊕s,t ϑ = πB ◦ H for ϑ(a) := πB (T ∗ H(a)T ) and s := πB (S) and t := πB (T ).
Hence, ϕ ⊕ πB ◦ H ≈ ψ ⊕ πB ◦ H, i.e., ϕ ∼u−nuc ψ.

Remark 5.8.1. Let E denote a unital C *-algebra and let O2 ∼ = C ∗ (s1 , s2 ) be

unitally contained in E. (Here we use only the unital embedding ι : O2 ,→ E up
to unitary equivalence by unitaries in the connected component U0 (E) of 1E in the
unitaries U(E) of E.)
We say that a C *-morphism ϕ : A → E strongly absorbs ψ : A → E if there
is a unitary u ∈ U0 (E) with u∗ ϕ(·)u = ϕ ⊕s1 ,s2 ψ, I.e., ϕ ≈ ϕ ⊕ ψ, if E = Q(B) for
stable and σ-unital B.
It allows to formulate a useful lifting criterium (that follows straight from the
definitions):
Suppose that a unital C*-morphism ϕ : A → Q(B) strongly absorbs every unital
C*-morphism πB ◦ H, where H : A → M(B) is unital and weakly nuclear:
There is a unital weakly nuclear C*-morphism ϕ b : A → M(B) with πB ◦ ϕ b = ϕ, if
u,strong
and only if, ϕ is in Homl-nuc (A, Q(B)) and [ϕ]≈ = 0 in Extnuc (A, B).
746 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

By the formulas in the proof of part (i) of Proposition 4.3.5, one can see that
ϕ strongly absorbs πB ◦ δ∞ ◦ H if and only if, there is an isometry S in M(B) such
that πB ◦ δ∞ ◦ H = πB (S)∗ ϕ(·)πB (S).
This links unital lifting problems to Weyl–von-Neumann type results and to
KK-theory.

Definition 5.8.2. Suppose that A is separable and B is σ-unital and stable.

Let
SExtnuc (A, B) := [Homl-nuc (A, Q(B))] ⊂ [Hom(A, Q(B))]
denote the unitary equivalence classes [ϕ] (by all unitaries in Q(B) !) of the weakly
nuclear liftable C *-morphisms ϕ : A → Q(B) equipped with Cuntz addition [ϕ] +
[ψ] := [ϕ ⊕ ψ].
We define the nuclear Ext-group Extnuc (A, B) as the Grothendieck group of
SExtnuc (A, B) :
Extnuc (A, B) := Gr(SExtnuc (A, B)) .
If C is σ-unital (but not necessarily stable) then we let
Extnuc (A, C) := Extnuc (A, C ⊗ K)
It is naturally isomorphic to the formerly defined group Extnuc (A, C) if C is stable
(cf. Lemma 5.8.5 below).
If A is unital, we let SExtunuc (A, B) denote the sub-semigroup of SExtnuc (A, B)
of classes [ϕ] ∈ SExtnuc (A, B) with ϕ(1A ) = 1Q(B) , i.e., , thus
SExtunuc (A, B) := [Homul-nuc (A, Q(B))] .
Then define the “unital” extension group Extunuc (A, B) of unital extensions with
weakly nuclear c.p. unital lifts V : A → M(B) by:
Extunuc (A, B) := Gr(SExtunuc (A, B)) .

Remark 5.8.3. The natural semigroup morphisms

[Homul-nuc (A, Q(B))]≈ → Extunuc (A, B)
induces an isomorphism
[h1 ]≈ + [Homul-nuc (A, Q(B))]≈ ∼
= Extunuc (A, B)
for h1 := πB ◦ ρ for any unital faithful *-representation ρ : A → M(K) ∼
= L(H) ⊂
M(B) with ρ(A) ∩ K = {0}.
If A is unital, then there are natural exact sequences:
0 → Extunuc (A, B) → Extnuc (A, B) → K0 (Q(B)) ∼
= K1 (B)
and
K1 (h0 (A)0 ∩ Q(B)) → K1 (Q(B)) → Extnuc
u,strong
(A, B) → Extunuc (A, B) → 0 .

The problem is now if every element of CPnuc (A, B) can be approximated via
compressions of H0 : A → M(B). It respects no ideals!
 8. EXTENSIONS AND Ext(C; A, B) 747

Remark 5.8.4. Suppose that A is separable and that B is σ-unital and stable.
Let H0 : A → L(`2 ) ∼
= M(K) ⊂ M(B) a non-degenerate *-monomorphism with
H0 (A) ∩ K = {0}.
Then H0 satisfies the assumptions of Corollary 5.4.10 with D := K, thus,
H0 satisfies the assumptions of Corollary 5.4.9 for the cone S = CPnuc (A, B).
Thus, h0 := πB ◦ H0 : A → Q(B) satisfies [h0 ⊕ h0 ] = [h0 ], S(h0 ; A, Q(A)) ⊂
S(h0 ⊕ 0; A, Q(B)) ⊂ [Homl-nuc (A, Q(B))], and [h] ∈ S(h0 ⊕ 0; A, Q(B)) for every
h ∈ Homl-nuc (A, Q(B)) , by Corollary 5.4.9. Thus:

SExtnuc (A, B) = [Homl-nuc (A, Q(B))] = S(h0 ⊕ 0; A, Q(B)) .

If A is unital, then H0 is unital the above considerations show that (??)

SExtunuc (A, B) = S(h0 ; A, Q(B))

and
Extunuc (A, B) = G(h0 ; A, Q(B)),
and finally,
Extu,strong
nuc (A, B) = G≈ (h0 ; A, Q(B)) .
If A is stable then

Gr([Homl-nuc (A, Q(B))], +) = [h0 ] + S(h0 ; A, Q(B)) = G(h0 ; A, Q(B))

because then [h0 ⊕ 0] = [h0 ] in Q(B) by Corollary 5.5.14.

But in general [Homl-nuc (A, Q(B))] ∼
6 S(h0 ; A, Q(B)), cf. Remark 5.5.8.
=
Since Extnuc (A, B) := Gr([Homl-nuc (A, Q(B))], +), it follows (for separable sta-
ble σ-unital A):

Extnuc (A, B) ∼
= [h0 ] + S(h0 ; A, Q(B)) = G(h0 ⊕ 0 ; A, Q(B))

and
Extnuc (A, B) ∼
= kernel(K0 (h0 (A)0 ∩ Q(B)) → K0 (B))
by Propositions 4.4.2 and 4.4.3.
In particular, the Corollary 5.7.1 implies the following:
Suppose that B is simple, purely infinite, σ-unital and stable, and that A is sepa-
rable, stable and exact. Then

[Monnuc (A, Q(B))] = Extnuc (A, B) .

And this formula is equivalent to the following observations (i)-(iv):

(i) Every nuclear *-monomorphism h : A ,→ Q(B) dominates h0 := πB ◦ ρ for

a non-degenerate *-monomorphism ρ : A ,→ L(H) ⊂ M(B) with ρ(A) ∩
B = {0} (i.e., [Monnuc (A, Q(B))] = [h0 ] + [Homnuc (A, Q(B))] ).
−1
(Indeed: Theorem 5.6.2 with D = B, C := πB (h(A)) and T := ρ ◦ h−1
yields the existence of an isometry S1 ∈ M(B) with S1∗ cS1 − T (c) ∈ B for
c ∈ C. Thus t∗1 h(·)t1 = h0 for t1 := S1 + B ∈ Q(B).)
748 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

(ii) If a C *-morphism h : A → Q(B) has a weakly nuclear lift V : A →

M(B) then h0 dominates h, and h is nuclear (i.e., [Homl-nuc (A, Q(B))] ⊂
S([h0 ], A, Q(B)) ⊂ Monnuc (A, Q(B))).
(Indeed: it can be managed that the weakly nuclear lift V of h is a
contraction, cf. Lemma 5.4.8, then Theorem 5.6.2, with D = B, C :=
−1
πB (h0 (A)) = B + ρ(A) and T := V ◦ h−1 0 , yields the existence of an
∗
isometry S2 ∈ M(B) with S2 ρ(a)S2 − V (a) ∈ B for a ∈ A, and t2 :=
S2 + B ∈ Q(B) satisfies t∗2 h0 (·)t2 = h.)
(iii) Homnuc (A, Q(B)) ⊆ Homl-nuc (A, Q(B)), because every nuclear c.p. con-
traction V : A → Q(B) has a nuclear lift T : A → M(B) (by a Theorem
of Effros and Choi, cf. [43]).
(iv) If h1 , h2 ∈ Hom(A, Q(B)) are unitarily equivalent, then they are unitarily
equivalent by a unitary u = πB (U ) with U ∈ M(B).
(Indeed: h1 dominates zero by Corollary 5.5.16 and Proposition 5.5.12(ii).
The unital C *-algebra Q(B) is K1 -injective by Lemmas 5.5.10(iv) and
4.2.6(viii).
The K1 -bijectivity of Qs (B) for σ-unital // has been
shown until here 3-times!! // Clear the places !!!
Thus, Proposition 4.3.6(iv,d). implies that h1 and h2 are unitarily
equivalent by a unitary u ∈ U0 (Q(B)) .)

Any two nuclear *-monomorphisms h, k : A ,→ Q(B) dominate each other (by (i)
and (ii)). Therefore, the unitary equivalence classes [h] ∈ [Monnuc (A, Q(B))] =
[h0 ] + S(h0 ; A, Q(B)) of nuclear *-monomorphisms h : A ,→ Q(B) is the
Grothendieck group of [Homnuc (A, Q(B))] by Proposition 4.4.2(ii).

Recall for applications of the following lemma that Extnuc (A, B ⊗ K)

onto Extnuc (A, B) := Ext(C; A, B) for the m.o.c. cone C := CPnuc (A, B)
and Extnuc (A, B ⊗ K) := Ext(C 0 ; A, B ⊗ K) for C 0 := CPnuc (A, B)K :=
CPnuc (A, B ⊗ K) = CPnuc (A, B) ⊗ CP(C, K).
Here CK := C ⊗ CP(C, K) denotes the (non-algebraic) tensor product of matrix
operator-convex cones in sense of Definition ??, i.e., is the smallest point-norm
closed m.o.c. cone in CP(A, B ⊗ K) that contains all tensor products T ⊗ S with
T ∈ C and S ∈ CP(C, K) ∼= K+ .

Lemma 5.8.5. Suppose that B is stable and σ-unital, that A is separable and C ⊆
CP(A, B) a countably generated non-degenerate full m.o.c. cone. Let λ : B ⊗K → B
be an isomorphism from B ⊗ K onto B such that b → λ(b ⊗ p11 ) is unitarily
homotopic to idB (cf. Corollary 5.5.6 and Lemma 5.5.11 for the existence of λ).
Denote by Q(λ) the induced isomorphism from Q(B ⊗ K) onto Q(B).
Then
ϕ ∈ Hom(A, Q(B ⊗ K)) 7→ Q(λ) ◦ ϕ ∈ Hom(A, Q(B))
defines a isomorphism from Ext(CK ; A, B ⊗ K) onto the 0-dominating Ext(C; A, B)
or unital version Ext(C; A, B) that is a functor with respect to A.
 8. EXTENSIONS AND Ext(C; A, B) 749

Check here:
It needs that we consider only zero-dominating
Busby invariants.
Discussion e.g. with B = M2∞ , B = O∞ , A = O2 .
The inverse is given by
Ext(C; A, B) → Ext( K CK ; A ⊗ K, B ⊗ K) → Ext(CK ; A, B ⊗ K) ,
i.e., is given by forming first the tensor product Eϕ ⊗ K of the extension Eϕ with
K and then by application of the map a ∈ A 7→ a ⊗ p11 ∈ A ⊗ K (cf. Lemma 5.5.11,
Proposition 5.5.12(v)).
In particular, there is a natural isomorphism
Extnuc (A ⊗ K, B) → Extnuc (A, B)
from Extnuc (A ⊗ K, B) onto Extnuc (A, B) that is induced by ϕ 7→ ϕ((·) ⊗ p11 ).

Proof. to be filled in ??
Do we need additional properties on the extensions,
e.g. zero absorption or more relaxed equivalence classes? 

Remark 5.8.6. The natural semigroup homomorphism

[Homul-nuc (A, Q(B))]≈ → SExtunuc (A, B)
and the induced group homomorphism
Extu,strong
nuc (A, B) = Gr([Homul-nuc (A, B)]≈ ) → Extunuc (A, B)
are in general not isomorphisms.
The definitions show that there are natural group and semigroup homomor-
phisms, e.g.
SExtnuc (A, B) → [Homl-cp (A, Q(B))] → Ext−1 (A, B) ,

Extnuc (A, B) → Ext−1 (A, B) .

If A is unital, then there are group homomorphisms
Extu,strong
nuc (A, B) → Extunuc (A, B) → Extnuc (A, B) → K0 (Q(B)) ∼
= K1 (B) ,
induced by ϕ ∈ Hom(A, Q(B)) → [ϕ(1)] ∈ K0 (Q(B)).
Further, there is the group morphism Extunuc (A, B) → Extnuc (A ⊗ K, B ⊗ K) ,
given by forming tensor products [Eϕ ⊗ K] of the corresponding extensions [Eϕ ] ∈
Extunuc (A, B) .
more ?? ?? ?? ??

Remark 5.8.7. Let ρ : A → L(H) ∼ = M(K) a faithful non-degenerate *-

representation of A on a Hilbert space H ∼= `2 (N) such that ρ(A) ∩ K = {0}.
Fix a unital strictly continuous *-monomorphism I : M(K) → M(B), and define
H : A → M(B) by H := I ◦ ρ. We let h0 := πB ◦ H : A → Q(B), i.e., consider
H(a) modulo B.
750 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

Then SExtnuc (A, B) = S(h0 ; A, Q(B)) if A is stable, and SExtunuc (A, B) =

S(h0 ; A, Q(B)) if A is unital, cf. Corollary 5.6.4(iii).
This implies the natural isomorphisms Extnuc (A, B) ∼ = G(h0 , A, Q(B)) , re-
spectively, in case of unital A, Extnuc (A, B) = G(h0 , A, Q(B)) and Extnuc (A, B) ∼
u ∼ =
G(πB ◦ (h0 ⊕ 0), A, Q(B)).
There are natural homomorphisms

K1 (πB (h0 (A))0 ∩ Q(B)) → K1 (Q(B)) ∼

= K0 (B)
and
K0 (B) ∼ u,strong
= K1 (Q(B)) → Extnuc (A, B) .
induced by u ∈ U(Q(B)) → [u∗ h0 (·)u]≈ ∈ [Homul-nuc (A, Q(B))]≈ .

The following Corollary 5.8.8 is a consequence of two of Kasparov’s theorems:

Corollary 5.6.1 and [73, prop. 17.6.5].

Corollary 5.8.8. Suppose that B is σ-unital and stable and that A is sepa-
rable.
If A or B is nuclear then

G(h0 , A ⊗ K, Q(B)) ∼
= Extnuc (A, B) = Ext−1 (A, B) .

Note that Ext−1 (A, B) ∼ = KK(A, C0 (R, B)) ∼ = KK(C0 (R, A), B) by theorems of
Kasparov (cf. [73, prop. 17.6.5, thm. 17.10.7, cor. 19.2.2.]).

Proof. By Lemma 5.8.5 and Remark 5.8.7, there is a natural isomorphism

from Extnuc (A, B) onto

Extnuc (A ⊗ K, B) = [h0 ] + [Homl-nuc (A ⊗ K, B)] = G(h0 , A ⊗ K, Q(B)) .

We have Homl-cp (A ⊗ K, Q(B)) = Homl-nuc (A ⊗ K, Q(B)) and ∼nuc =∼ for nuclear

B. If A is nuclear, then (trivially) Hom(A ⊗ K, Q(B)) = Homnuc (A ⊗ K, Q(B))
and Homnuc (A ⊗ K, Q(B)) ⊂ Homl-nuc (A ⊗ K, Q(B)) (by the Choi-Effros lift-
ing theorem, cf. [43]). In both cases Ext−1 (A, B) = Gr([Homl-cp (A, Q(B))]) =
Gr([Homl-nuc (A, Q(B))]) = Extnuc (A, B) . 

Corollary 5.8.9. Suppose that A is separable and unital, B purely infinite,

stable and σ-unital, h0 : A → L(H) ∼ = M(K) ⊂ M(B) a faithful and unital *-
representation with h0 (A) ∩ K = 0. Then

(i) Extunuc (A, B) = G(h0 , A, Q(B)); ??

(ii) 0 → Extunuc (A, B) → Extnuc (A, B) → K1 (B) is an exact sequence.
(iii) There is a natural exact sequence

K1 (πB (h0 (A))0 ∩ Qs (B)) → K0 (B) → Extnuc

u,strong
(A, B) → Extunuc (A, B) → 0.

In particular, Extu,strong
nuc (A, B) = Extunuc (A, B) if K0 (B) = 0.

Proof. to be filled in ?? 
 8. EXTENSIONS AND Ext(C; A, B) 751

Corollary 5.8.10. If A is a separable unital C*-algebra, then

Extu,strong
nuc (A, O2 ⊗ K) = Extunuc (A, O2 ⊗ K) = Ext−1 (A, O2 ) ∼
= 0.

In particular, if h : A → Qs (O2 ) is a unital *-monomorphism which has a completely

positive lift V : A → M(O2 ⊗ K), then there is a unital C*-morphism ϕ : A →
M(O2 ⊗ K) with πO2 ⊗K ϕ = h.

Proof.

Extu,strong
nuc (A, O2 ) = Extunuc (A, O2 ) ⊂ Extnuc (A, O2 ) = Ext−1 (A, O2 )

by Corollary 5.8.9(iii), because K0 (O2 ) = 0 and O2 is simple, purely infinite and

nuclear, cf. [172], [169].

Ext−1 (A, O2 ) ∼
= KK1 (A, (O2 )(1) ) ∼
= KK(SA, O2 )

by Kasparov’s isomorphisms and Bott periodicity, cf. [73, prop. 17.6.5, cor. 19.2.2]
(see also Chapter 8).
By [172], idO2 is homotopic to idO2 ⊕s,t idO2 in the unital endomorphisms of
O2 . Thus KK(SA, O2 ) = 0, because KK(SA, · ) is a homotopy invariant functor
such that the Cuntz addition ⊕ induces the addition in KK(SA, O2 ), cf. Chapter
8.
Now let h : A ,→ Qs (O2 ) a unital *-monomorphism that has a c.p. lift
V : A → M(O2 ⊗ K). Then Extu,strong nuc (A, O2 ) = 0 implies the existence of unital
*-monomorphisms h1 , h2 : A → M(O2 ⊗ K) and of a unitary U1 ∈ M(O2 ⊗ K) such
that (V (a)⊕h1 (a))−U1∗ h2 (a)U1 ∈ O2 ⊗K for a ∈ A, where V : A → M(O2 ⊗K) is a
unital completely positive lift of h : A → Qs (O2 ). We get a unitary U2 ∈ M(O2 ⊗K)
with U2∗ (V (a) ⊕ h1 (a))U2 − V (a) ∈ O2 ⊗ K for every a ∈ A from Theorem 5.6.2,
which has to be applied to C := V (A) + O2 ⊗ K and the completely positive unital
map
T := h1 ◦ h−1 πO2 ⊗K : C → M(O2 ⊗ K) .

Thus, ϕ : A 3 a 7→ U3∗ h2 (a)U3 with unitary U3 := U1 U2 is a unital C *-morphism

that is a lift of h. 

Corollary 5.8.11. Suppose that A is separable and unital.

(i) SExtunuc (A, O2 ⊗ K) ∩ [Mon(A, Qs (O2 ))] = Extunuc (A, O2 ⊗ K) = 0.

(ii) If a C*-morphism h : A → Qs (O2 ) has a unital completely positive lift
T : A → M(O2 ⊗ K) then h has a lift as a unital C*-morphism ϕ : A →
M(O2 ⊗ K), i.e., πO2 ⊗K ϕ = h.

Proof. (i) follows from Corollary 5.6.4(iv), because O2 ⊗ K is purely infinite

and simple.
???? ??
752 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

(ii): Let T : A → M(O2 ⊗ K) a unital completely positive map with πT = h.

Since O2 is nuclear, T is weakly nuclear, and thus h ∈ Homul-nuc (A, Q(O2 ⊗ K)),
i.e., [h] ∈ SExtunuc (A, O2 ⊗ K) by Definition 5.8.2.
Let d : A → L(H) ⊂ M(O2 ⊗ K) a faithful unital *-representation of A such
that d(A) ∩ K = 0. Let k := πd. By Kasparov’s generalized Weyl–von Neumann
theorem ([?], see Corollary 5.6.4), we have Extu (A, O2 ) := Gr([Ext−1
u (A, O2 )]) =
s
G(k, A, Q (O2 )), where we have used that O2 is nuclear (cf. [169]).
By Corollary 5.8.10, Extu (A, O2 ) = 0.
Thus there exists a unitary v in Qs (O2 ) such that h ⊕ k = v ∗ k(·)v.
From now on we use that O2 is purely infinite and simple, [169], [172].
h absorbs k by Corollary 5.6.4(iv), i.e., there exists a unitary w ∈ U0 (Qs (O2 ))
such that w∗ h(·)w = h ⊕ k.
It is easy to see, e.g. with help of Lemma 5.1.2(ii), that multiplier algebras of
stable algebras have trivial K-theory. Therefore we get K1 (Qs (O2 )) ∼
= K0 (O2 ) = 0
from the 6-term exact sequence of K-theory, [73].
By Corollary 5.7.2, Qs (O2 ) is simple and purely infinite. By a theorem of Cuntz
[172], the quotient of the unitary group of Qs (O2 ) by its connected component is
naturally isomorphic to K1 (Qs (O2 )) = 0. (More generally stable coronas of σ-unital
C *-algebras are K1 -bijectivity by Proposition 4.2.15.)
Thus vw∗ is in the connected component of 1 in the unitary group of Qs (O2 )
and therefore there exists a unitary U ∈ M(O2 ⊗ K) with π(U ) = vw∗ .
U ∗ d(·)U is a unital lift of h, i.e., π(U ∗ d(a)U ) = k(a). 

Corollary 5.8.12. Suppose that E is a unital separable C*-algebra, λ : E → A

is an epimorphism such that the kernel D of λ is essential in E, D ∼
= O2 ⊗ K, and
that there exists a unital completely positive map V : A → E such that λV = idA .
Then there exists a unital *-monomorphism ψ : A → E with λψ = idA .

Proof. Expressed in algebraic terminology, we show that a semi-split essential

exact sequence 0 → D → E → A → 0 splits unitally if D ∼= O2 ⊗ K.
Let π : M(O2 ⊗ K) → Qs (O2 ) be the natural epimorphism with kernel O2 ⊗ K
and let θ be the *-monomorphism from E into M(O2 ⊗ K) which is defined by the
isomorphism from D onto O2 ⊗ K. It is a monomorphism, because D is an essential
ideal of E (by our assumptions). In particular, θ(D) = O2 ⊗ K.
Then there is a unique unital *-monomorphism τ : A → Qs (O2 ) with πθ(E) =
τ (A) and πθ = τ λ.
W := θV is a unital completely positive lift of τ , i.e., πW = τ .
By Corollary 5.8.10 there exists a unital lift h of τ , i.e., π(h(a)) = τ (a). But
then h(A) ⊂ π −1 (τ (A)) = θ(E), and ψ(·) := θ−1 (h(·)) is a unital *-monomorphism
from A into E such that λψ = idA , because τ λψ = πθψ = πh = τ and τ is faithful.
Thus ψ is as desired. 
 9. THE Ψ-RESIDUALLY NUCLEAR CASE AND Ext(C; A, B) 753

9. The Ψ-residually nuclear case and Ext(C; A, B)

Start of collection for Chapter 5:

Needed for Theorem 6.3.1.
But is the here used notation ok?

Definition 5.9.1. Let X be a Dini-space, i.e., a sober second countable locally

quasi-compact T0 space.
Suppose that A and B are stable and σ-unital C *-algebras, and that
ΨA : OX → I(A) is a lower semi-continuous action of X on A and ΨB : OX → I(B)
an upper semi-continuous action of X on B (cf. Definition 1.2.6). A completely
positive map V : A → M(B) is called weakly ΨA -ΨB -residually nuclear,
(or shortly Ψ-residual nuclear) if, for every b ∈ B the completely positive map
a ∈ A 7→ b∗ V (a)b ∈ B is ΨA -ΨB -residually nuclear in the sense of Definition
1.2.8. We denote by Homl-nuc (X; A, Q(B)) ⊂ Hom(A, Q(B)) the set of C *-
morphisms h from A into the (stable) corona Q(B) = M(B)/B of B which
have a weakly ΨA -ΨB -residually nuclear completely positive lift V : A → M(B).
Then Homl-nuc (X; A, Qs (B)) is invariant under passage to unitary equivalent C *-
morphisms and under Cuntz addition. Let SExtnuc (X; A, B) denote the semigroup
of unitary equivalence classes of morphisms in Homl-nuc (X; A, Qs (B)) with Cuntz
addition.
We define:
Extnuc (X; A, B) := Gr(SExtnuc (X; A, B)).

We generalize the above definitions in the following manner, because all con-
structions of ideal-system equivariant versions of Kasparov’s Ext-theory factorize
anyway though the variants of Ext-theory for the corresponding m.o.c. cones:
Let A separable, B σ-unital and stable, and let C ⊂ CP(A, B) a matricial operator-
convex cone. We denote by Homl (C; A, Q(B)) ⊂ Hom(A, Q(B)) the set of the
C *-morphisms h : A → Q(B) ∼ = Qs (B) that have a c.p. lift V : A → M(B) with
∗
the property that b V (·)b ∈ C for all b ∈ B. Again, Homl (C; A, Q(B)) is invariant
under unitary equivalence and Cuntz addition, and we can define the semigroups
and its Grothendieck groups by

SExt(C; A, B) := [Homl (C; A, Q(B))] ⊂ [Hom(A, Q(B))]

Here [·] means the unitary equivalence classes of morphisms in Homl (C; A, Q(B))
with Cuntz addition.)

Ext(C; A, B) := Gr(SExt(C; A, B)) .

Note that Extnuc (X; A, B) = Extnuc (C; A, B) for the m.o.c. cone C ⊂ CP(A, B)
of ΨA –ΨB -residually nuclear c.p. maps from A into B. If A is exact and B is
separable, then there is a unique action ΨA of X := Prim(B) such that a given
point-norm closed m.o.c. cone C is the same as CPrn (X; A, B) with ΨB defined as
the identity action of X on X.
754 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

Remark 5.9.2. Special cases of the groups Ext(C; A, B) are Ext(A, B) =

Ext(CP(A, B); A, B), Extnuc (A, B) = Ext(CPnuc (A, B); A, B), Extnuc (X; A, B) =
Ext(CPrn (X; A, B); A, B), where CPrn (X; A, B) ⊂ CP(A, B) denotes the ΨA -ΨB -
residually nuclear c.p. maps, depending on fixed actions ΨA : O(X) → I(A) and
ΨB : O(X) → I(B) of X on A and B.
We can also define

Ext(X; A, B) := Ext(CP(X; A, B); A, B) ,

the group of ΨA -ΨB -equivariant extensions, where CP(X; A, B) ⊂ CP(A, B) is the

Ψ-equivariant c.p. maps.
Next Def’s could be in Chp.3
The cone C ⊂ CP(A, B) is faithful (respectively non-degenerate ) if there is
no non-zero ideal J of A with T (J) = {0} for all T ∈ C (respectively there is no
ideal I 6= B of B with T (A) ⊂ I for all T ∈ C ).
There is also an explanation of Ext(C; A, B) with help of extensions with Busby
invariant that have lifts that are limits of maps in C with respect to the strict
topology on M(B). It can be done in a similar manner as in the explanation given
in Section 8 for the case X =point (Note that Extnuc (X; A, B) = Extnuc (A, B) if
X = {p} is a point).
Corollaries 5.4.4 and 5.4.9 tell us:
If C is countably generated, then there is a “universal” element H0 ∈ Hom(A, M (B))
that defines the point-norm closure of C, with δ∞ ◦ H0 unitarily equivalent to
H0 , and H0 (A)BH0 (A) a split corner of B. It can be chosen non-degenerate
if C is non-degenerate , i.e., if C is not contained in CB(A, J) for some closed
ideal J of B. Let h0 := πB ◦ H0 . Then Ext(C; A, B) = [h0 ] + SExt(C; A, B) =
G(h0 ; A, B). The split extension B + H0 (A) = πB is stable if C is non-degenerate.
By Proposition 5.5.12(i+ii), this implies that every extension with Busby invariant
in Ext(C; A, B) is a stable C *-algebra. (Here we can see that the questions about
the non-degeneracy of CPrn (X; A, B) is of importance for actions of X on A and
B.)

Recall that a corner of a C *-algebra E is a hereditary C *-subalgebra F of E

such that there is a projection p ∈ M(E) with F = pEp, and that a subalgebra
G ⊂ E generates a corner of E, if C ∗ (G)EC ∗ (G) is a corner of E ( 20 ). The
following Ψ-equivariant generalization of Theorem 5.6.2 seems to be the maximal
possible, only special cases are used in our applications.

Theorem 5.9.3. Suppose that B is a σ-unital C*-algebra, D ⊂ M(B) is

strongly purely infinite, separable and stable C*-subalgebra such that DB is dense
in B (and thus M(D) ⊂ M(B) naturally), and that C ⊂ M(D) is a separable
C*-subalgebra. Define ΨC := Ψup D,B
D,C and ΨB := Ψdown as in Chapter 1.

20The set (!) of products C ∗ (G) · E · C ∗ (G) is always a hereditary C *-subalgebra of E by

the Cohen factorization theorem for Banach modules.

 9. THE Ψ-RESIDUALLY NUCLEAR CASE AND Ext(C; A, B) 755

Suppose further that

(α3) for every closed ideal J of D, the subalgebra M(πJ )(C) ∩ πJ (D) of
πJ (D) = D/J generates a corner of D/J, and
(β3) Let V : C → M(B) be a weakly ΨC -ΨB –residually nuclear completely
positive map such that, for every J ∈ I(D), ξ ∈ C and c ∈ C ∩ (ξ1 + D +
M(D, J)) the element V (c) is in the strict closure of ξ1 + span(BJB)
(21).

Then idC asymptotically dominates V in the sense of Definition 5.0.1.

If, moreover, V is a C*-morphism then idC asymptotically absorbs V , i.e.,
idC ⊕V : C → M(B) and idC are unitarily homotopic in the sense of Definition
5.0.1.

Proof. Since B and D are σ-unital, there are natural isomorphisms

M(B)/M(B, I) ∼ = M(B/I) and M(D)/M(D, J) ∼ = M(D/J) for closed ideals I
of B and J of D by the non-commutative version of the Tietze extension theorem
( 22 ).
They allow us to proceed as in the proof of Theorem 5.6.2:
First, (β3) implies V (C ∩ D) = {0} if we take ξ = 0 and J = {0}. Furthermore
C ∩D is a corner of D by (α3). The same properties hold for M(πJ )(C) = C/ΨC (J)
and πJ (D) = D/J and the unit element 1 of M(D/J) ∼ = M(D)/M(D, J) in place
of C, D and 1 ∈ M(D) by (β3) and (α3).
Notice that M(D, J) ⊂ M(D) ∩ M(B, ΨB (J)) and ΨC (J) = C ∩ M(D, J).
The latter identity is the definition of ΨC = ΨupD,C , and the first inclusion follows
from JB ⊂ ΨB (J), i.e., from J ⊂ M(B, ΨB (J)), because M(B, ΨB (J)) is strictly
closed in M(B), the inclusion M(D) → M(B) is strictly continuous and M(D, J)
is the strict closure of J in M(D) (Recall that ΨB (J) is the closure of span(BJB)).
It follows that M(D) ∩ M(B, ΨB (J)) is a strictly closed ideal of M(D). Thus
M(D) ∩ M(B, ΨB (J)) = M(D, J1 ) for the ideal J1 := D ∩ M(B, ΨB (J)) of D.
Then J ⊂ J1 and ΨB (J1 ) = ΨB (J). The arguments show that our J1 is the biggest
ideal of D with ΨB (J1 ) = ΨB (J), and that, for every closed ideal I of B, the
ideal ΨD (I) := D ∩ M(B, I) of D satisfies M(D, ΨD (I)) = M(D) ∩ M(B, I),
ΨC (ΨD (I)) = C ∩ M(B, I) and ΨB (ΨD (I)) ⊂ I. In particular, V (C ∩ M(B, I)) =
V (ΨC (ΨD (I))) ⊂ M(B, I) for all ideals I of B.
We let C1 := D + C + C1 and define Ve : C1 → M(B) by V e (d + c + ξ1) :=
V (c)+ξ1 for d ∈ D, c ∈ C, ξ ∈ C. It is well-defined, because d0 +c0 +ξ 0 1 = d+c+ξ1
implies that c0 − c ∈ (ξ − ξ 0 )1 + D, thus, V (c0 − c) = (ξ − ξ 0 )1 by (β3) with
J = {0}, which yields V (c0 ) + ξ 0 1 = V (c) + ξ1. By definition V e (D) = 0 and
[V e ]D : C1 /D → M(B) is the extension of [V ]D∩C : C/(D ∩ C) → M(B) to a
21I.e., V (C ∩ (ξ1 + D + M(D, J))) ⊂ ξ1 + M(B, Ψ (J)) for J ∈ I(D) and ξ ∈ {0, 1}.
B
22Cf. [616, Prop.3.12.10.] for the case of separable B. It extends to σ-unital C *-algebras B,

because then M(B/J) is the algebraic inductive limit of its C *-subalgebras M(A/(A ∩ J)) where
the C *-algebras A ⊂ B are separable and e ∈ A for some e ∈ B+ that is strictly positive in B.
756 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

∼ C/(D ∩ C) + C1 into M(B). Since V is a c.p. contraction

unital map of C1 /D =
with V (D∩C) = 0, it follows that V e : C1 → M(B) is a u.c.p. map with V e (D) = 0.
Thus, condition (α) of Proposition 5.4.1 is satisfied for C1 , V e and a strictly
positive contraction h ∈ D ⊂ C1 .
Let Φ(J) := C1 ∩ M(D, J) for ideals J of D. We have that V e (Φ(J)) ⊂
M(B, ΨB (J)) by (β3): Indeed, if d + c + ξ1 ∈ M(D, J), then c ∈ C ∩ ((−ξ)1 + D +
M(D, J)), and, by condition (β3), V (c) ∈ (−ξ)1 + M(B, ΨB (J)) , i.e., V e (d + c +
ξ1) = V (c) + ξ1 ∈ M(B, ΨB (J)) .
Lemma B.7.7(ii) shows that assumption (β3) implies that V e is even Φ-ΨB –
residually nuclear: The (induced class-)map
[V e ]J : C1 /Φ(J) → M(B/ΨB (J))
is weakly nuclear by Lemma B.7.7(ii) for every closed ideal J of D, because
C1 /Φ(J) = M(πJ )(C)+πJ (D)+C1 ⊂ M(D/J), [V e ]J (d+c+ξ1) = [V ]J (c)+ξ1 ∈
M(B/ΨB (J)), M(πJ ) ∩ πJ (D) generates a corner FJ of D/J = πJ (D) and
[V ]J : M(πJ )(C) → M(B/ΨB (J)) is weakly nuclear.
It follows that T := δ∞ ◦ V e is also a unital Φ-ΨB –residual weakly nuclear
u.c.p. map from C1 ⊂ M(D) into M(B). The strictly positive contraction h ∈ D+
again satisfies condition (α) of Proposition 5.4.1 for T : C1 → M(B).
Proposition 3.6.1 applies to T , and shows that the Φ-ΨB –residual nuclear maps
b∗ T (·)b (b ∈ B) are approximately 1-step inner in M(B). Thus, also condition (β)
of 5.4.1 is satisfied, and the conclusions follow from parts (iii) and (iv) of Proposition
5.4.1. 

Remark 5.9.4. The tricky points of the reduction to Proposition 5.4.1 in the
proof of Theorem 5.9.3 are contained in the proof of Lemma B.7.7 and in the proof
of the local approximation result in Proposition 3.6.1. The latter proof essentially
reduces to the case, where there is a weakly residually nuclear *-monomorphism
H1 : C → M(D) such that H1 (C) ∩ M(D, J) = H1 (C ∩ M(D, J)) for every closed
ideal J of D, as it is the case for the ultrapower Dω in place of D.

The following Corollary 5.9.5 is an immediate consequence of our above given

Weyl–von-Neumann–Voiculescu type Theorem 5.9.3 for weakly residually nuclear
maps. We define for D ⊆ B and X := Prim(D), an action
ΨE : O(X) ∼
= I(D) → I(Q(B))
of X on E := Q(B) by
ΨE (J) := πB (M(B, span(BJB))) for J ∈ I(D) .
If ΨA : O(X) → I(A) is an action of X on A, we denote by Mon(X; A, E) the set
of *-monomorphisms ϕ : A → E with ϕ(ΨA (U )) = ϕ(A) ∩ ΨE (U ) for U ∈ O(X).
(It might happen that Mon(X; A, E) is empty.)

Corollary 5.9.5. Suppose that A, and D ⊆ B are stable C*-algebras, such

that A and D are separable, D is strongly purely infinite and DB is dense in B.
 9. THE Ψ-RESIDUALLY NUCLEAR CASE AND Ext(C; A, B) 757

Let X := Prim(D). Further let H1 : A → M(D) be a weakly residually nuclear

*-monomorphism such that H1 is unitarily equivalent to δ∞ ◦ H1 and that H1 (A)D
is dense in D. Denote by H0 : A → Q(B) = M(B)/B ∼ = Qs (B) the corresponding
s
monomorphism from A into the stable corona Q (B) of B, i.e., H0 := πB H1 .
Consider the actions ΨB := ΨD,B
down of X on B and the action ΨA :=
H1−1 Ψup
D,H1 (A) of X on A.
Then there are natural isomorphisms

SExtnuc (X; A, B) ∼
= S(H0 , A, Q(B))
and
Extnuc (X; A, B) ∼
= G(H0 , A, Q(B)) .
If, in addition, B = D then

SExtnuc (X; A, B) ∩ [Mon(X; A, Q(B))] = G(H0 , A, Q(B)) .

Recall that G(H0 , A, Q(B)) is naturally isomorphic to the kernel of K0 (H0 (A)0 ∩
Q(B)) → K0 (Q(B)) by Proposition 4.4.3(ii).

Proof. to be filled in ?? 

Lemma 5.9.6. Suppose that B is σ-unital, A is a separable C*-subalgebra of

M(B)/B, and that V : A → M(B)/B is a completely positive contraction.
If there exist a sequence d1 , d2 , . . . of contractions in M(B)/B, such that, for
a ∈ A,

(i) limn→∞ kd∗n adn+1 k = 0 , limn→∞ kd∗n dn+1 k = 0 , and

(ii) limn→∞ kd∗n adn − V (a)k = 0,

then there exists a contraction d ∈ M(B), such that V (a) = πB (d)∗ aπB (d) for
a ∈ A.

Proof. Since A is separable, there is a strictly positive contraction c ∈ A+

with kck = 1.
The proof is similar to that of Proposition 5.4.1. Let Ω be a compact sub-
set of the contractions in A+ that linearly generates a dense linear subspace of
A, and let γ : Ω → M(B)+ (respectively τ : V (Ω) → M(B)+ ) a topological lift
of Ω ⊆ Q(B) (respectively of V (Ω) ⊆ Q(B)) into the positive contractions in
M(B). Further let e ∈ B+ a strictly positive contraction with kek = 1, and
let f1 , f2 , . . . ∈ M(B) contractive lifts of d1 , d2 , . . .. The unital C *-subalgebra
D := C ∗ (1, e, γ(Ω), τ (V (Ω)), {fn }n ) ⊆ M(B) generated by {e, 1, f1 , f2 , . . .}∪γ(Ω)∪
τ (V (Ω)) is separable.
By Remark 5.1.1(3), we find elements gn ∈ C ∗ (e)+ , namely, square roots of
suitable differences of elements in a suitable approximate unit of B that is approx-
imately central for D, with the following properties
−n
P 2
(1) n gn converges strictly to 1M(B) , gn gm = 0 for |m−n| > 1, kegn k < 2
758 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

(2) kfk gn − gn fk k < 2−n for k ≤ n ,

(3) kgn γ(a) − γ(a)gn k < 2−n and kgn τ (V (a)) − τ (V (a))gn k < 2−n for a ∈ Ω,
(4) kgn fn∗ bfn+1 gn+1 k ≤ 2−n + dist(fn∗ bfn+1 , B) for b ∈ γ(Ω) ∪ {1}, and
(5) kgn (fn∗ γ(a)fn − τ (V (a)))gn k ≤ 2−n + dist(fn∗ γ(a)fn − τ (V (a)), B) for
a ∈ Ω.
P
Let Γ(b1 , b2 , . . .) := gn bn gn for (b1 , b2 , . . .) ∈ `∞ (M(B)). By Remarks 5.1.1(2,4)
check numbers 2 and 4!, Γ is a unital c.p. map from `∞ (M(B)) into M(B).
It holds Γ(b1 , b2 , . . .) ∈ B if lim kgn bn gn k = 0, because then Γ(b1 , b2 , . . .) is the
P P
sum of the two series g2n−1 b2n−1 g2n−1 and g2n b2n g2n which have mutually
orthogonal summands by (1). Thus, by (5) and (ii),

Γ(f1∗ γ(a)f1 − τ (V (a)), f2∗ γ(a)f2 − τ (V (a)), . . .) ∈ B .

Moreover, by Remarks 5.1.1(4,5,6), τ (V (a))−Γ(τ (V (a)), τ (V (a)), . . .) ∈ B (by (3)),

P
fn gn converges strictly to an element f ∈ M(B), and, by (2)–(4)

f ∗ bf − Γ(f1∗ bf1 , f2∗ bf2 , . . .) ∈ B

for b ∈ γ(Ω) ∪ {1}, because limn kgn fn∗ bfn+1 gn+1 k = 0 by (4) and (i).
Thus, πB (f ) is a contraction and πB (f )∗ aπB (f ) = V (a) for a ∈ Ω. Let d :=
hλ(h∗ h), where λ(t) := min(1, t−1/2 ) for t ∈ [0, ∞). Then d is a contraction in
M(B) with πB (d)∗ aπB (d) = V (a) for a ∈ Ω, and thus for all a ∈ A. 

Remark 5.9.7. If (in addition to the assumptions of Lemma 5.9.6) B is stable

and A is contained in a σ-unital stable C *-subalgebra D of M(B)/B such that
−1
πB (D) is stable, then d can be replaced by an isometry s = πB (S) for some
isometry S ∈ M(B), i.e., d∗ ad = s∗ as for all a ∈ A.
(Indeed: there are isometries T1 , T2 ∈ M(B) such that T1 T1∗ + T2 T2∗ = 1, ct2 = 0
and t1 c = c = ct1 for all c ∈ D ⊇ A and tk = πB (Tk ),k = 1, 2, by Corollary 5.5.14.
Let S := T1 d + t2 (1 − d∗ d)1/2 ∈ M(B).)

Proposition 5.9.8. Suppose that A is separable, B is stable and σ-unital and

(1-)purely infinite, h : A → Q(B) is a C*-morphism, and that (Vk : Q(B) → Q(B))
is a sequence of approximately inner completely positive maps.
If h dominates h ⊕ h and Vn ◦ h converges point-wise on A to a c.p. contraction
T : A → Q(B), then h dominates T .

Proof. The multiplier algebra M(B) and the corona Q(B) contain copies of
O2 unitally, because B is stable. Since h dominates h ⊕ h, there is a copy of O∞
unitally contained in h(A)0 ∩ Q(B) (cf. Proposition 4.3.5(iv)). The sequence of
c.p. maps (Vk ◦ h) (with point-norm limit T ) allows to construct a sequence dn
of contractions in Q(B) such that d∗n+k bdn = 0 for b ∈ h(A) + C1, k > 0, and
limn→∞ kd∗n h(a)dn − T (a)k = 0 (cf. Lemma 3.10.5). Thus, Lemma 5.9.6 applies
and there is a contraction d ∈ Q(B) with d∗ h(·)d = T . Since A is stable and B is
−1
purely infinite, πB (h(A)) is stable by Corollary 5.5.16. Thus d can be replaced by
an isometry (by Remark 5.9.7). 
 9. THE Ψ-RESIDUALLY NUCLEAR CASE AND Ext(C; A, B) 759

Remark 5.9.9.
Where 5.9.9 is used? ??
If B ∼ = B ⊗ D∞ (with D∞ = O∞ ⊗ O∞ ⊗ . . .), then the latter approxi-
mate domination should be the case if and only if (for suitable topological lifts)
h : Ω → M(B ⊗ D∞ ) approximately dominates Te modulo a suitable commutative
e
approximate unit e1 ≤ e2 ≤ . . . of B. More precisely: Let e h and Te denote topo-
logical lifts of h and T on a linearly generating compact subset Ω of A. Then, for
every ε > 0 there should exist kε ∈ N, such that, for every m > n > kε , there is a
contraction dn,m ∈ B ⊗ D∞ with

kd∗n,m e
h(a)dn,m − em (1 − en )Te(a)(1 − en )em k < ε.

for a ∈ Ω. (Also some orthogonality relations of the dn,m are required, coming here
from a central sequence of copies of O∞ in M(B ⊗ D∞ ) for B ⊗ D∞ .)

Lemma 5.9.10. Let B a stable and σ-unital C*-algebra and b ∈ M(B)+ . De-
note by J(b) := span BbB the closed ideal of B, and by I(b) := span M(B)bM(B)
the norm-closed ideal of M(B) generated by b. Then, for every b ∈ M(B)+ ,




M B, J((b − ε)+ ) ⊆ I δ∞ (b) ⊆ M B, J(b) ,


and b ∈ I δ∞ (b) ∈ I(M(B)) .

Proof. Let e denote a strictly positive element of B+ , s1 , s2 , . . . ∈ M(B) a

sn s∗n = 1 strictly, and b ∈ M(B)+ . Then
P
sequence of isometries with

bB ⊆ J(b) := span BbB ,

because limn→∞ e1/n bc = bc for c ∈ B. Thus, Bb ∪ bB ⊆ J(b), i.e., b ∈ M(B, J(b))

for every b ∈ M(B)+ .
It follows that δ∞ (b) ∈ M(B, J(b)) for each b ∈ M(B)+ , because span BbB =
∗
P
span(Bδ∞ (b)B) by strict convergence of n sn bsn =: δ∞ (b), i.e., by norm-
Pm ∗
convergence in B of n=1 sn bsn a to δ∞ (b)a for all a ∈ B.
Clearly, b = s∗1 δ∞ (b)s1 ∈ I δ∞ (b) .

Let J := J((b − ε)+ ) = span(B(b − ε)+ B), and let t ∈ M(B, J)+ . If we
use functional calculus and convex combination, then we find en ∈ C ∗ (e) with
0 ≤ en ≤ 1, en en+1 = en , ken e−ek < 2−n , ken t−ten k < 2−n , and kgn t−tgn k < 2−n
for gn := (en − en−1 )1/2 with e0 := 0, cf. Remark 5.1.1(3).
Next still to be checked:
P
The sum V (t) := 0≤n≤∞ gn tgn converges strictly and the elements tgn and
V (t) − t are in J, cf. Remark 5.1.1(4). Since gn tgn ∈ J+ for t ∈ M(B, J) and since
J+ is the closed linear span of {b∗ (a − ε)+ b ; b ∈ B }, we get (a − ε)δ+ · B ⊆ J for
each δ ∈ (0, 1).
760 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

By definition of J and M(B, J), there exist kn ∈ N and bn,k ∈ B where

k ∈ {1, . . . , kn }, with
kn
X
k gn tgn − (bn,k )∗ (a − ε)+ bn,k k < 2−n .
k=1

sn s∗n strictly
P
Let s1 , s2 , . . . a sequence of isometries in M(B), such that
converges to 1. We define a bounded continuous function ϕ : [0, ∞) by ϕ(0) := 0
and ϕ(τ ) := ((τ − ε)+ /τ )1/2 for τ ∈ (0, ∞). Then ϕ(τ )2 ≤ ε−1 (τ − ε)+ because
τ ϕ(τ )2 = (τ − ε)+ . It follows that ϕ(a)bn,k ∈ J for each n ∈ N and k = 1, . . . , kn .
The elements fn := (en+2 − en−1 )1/2 are positive contractions with gn fn = gn
if we let e0 := 0.
We define inductively mn ∈ N, dn ∈ B and qn ∈ M(B), by m1 = 0, mn+1 :=
mn + kn ,
kn
X kn
X
dn := sk+mn ϕ(a)bn,k fn and qn := sk+mn s∗k+mn .
k=1 k=1
P
The sum qn of the projections qn ∈ M(B) converges strictly to 1M(B) . The dn
satisfy
kn
X
kgn tgn − d∗n δ∞ (a)dn k = kfn gn tgn − (bn,k )∗ (a − ε)+ bn,k fn k < 2−n

k=1
Pkn
and dn = qn xn fn , where xn ∈ B is defined by xn := k=1 sk+mn ϕ(a)bn,k . Notice
that xn and dn are in J, because bn,k ϕ(a) bn,k ≤ ε (bn,k )∗ (a − ε)+ bn,k ∈ J.
∗ 2 −1

The sequence x1 , x2 , . . . is bounded, because b∗n,k ϕ(a)2 bn,k ≤ ε−1 (bn,k )∗ (a − ε)+ bn,k
implies that
kn
X
x∗n xn = b∗n,k ϕ(a)2 bn,k ≤ ε−1 (2−n 1 + gn tgn ) ≤ ε−1 (1 + ktk) .
k=1

≤ 3 gn2 = 3 · 1M (B) converges strictly to some element F ∈

fn2
P P
Since also
P∞
M(B)+ with kF k ≤ 3, we get that n=1 dn is strictly convergent in M(B),
P∞ ∗
cf. Remark 5.1.1(2). Let d := n=1 dn ∈ M(B). Using that dm δ∞ (a)dn =
δm,n d∗n δ∞ (a)dn we get
X
d∗ δ∞ (a)d = d∗n δ∞ (a)dn .
n
The estimates kd∗n δ∞ (a)dn − gn tgn k < 2−n and that d∗n δ∞ (a)dn , gn tgn ∈ J show
that d∗ δ∞ (a)d − V (t) is in J, because d∗n δ∞ (a)dn , gn tgn ∈ J. It implies V (t) ∈
J + I(δ∞ (a)). Since J = J((a − ε)+ ) ⊆ J(a) ⊆ I(δ∞ (a)), the element t = V (t) +
(t − V (t)) ∈ M(B, J)+ is contained in I(δ∞ (a)) + J(a). Clearly I(δ∞ (a)) + J(a) =
I(δ∞ (a)) . 

The following Proposition 5.9.11 characterizes the invertible elements of

SExtnuc (Prim(B); A, B) ⊆ [Hom(A, Qs (B))]
for exact A ⊆ M(B) with the property that the imbedding map a ∈ A → a ∈ M(B)
is (weakly) nuclear, where B is σ-unital, stable and has the WvN-property. It
 9. THE Ψ-RESIDUALLY NUCLEAR CASE AND Ext(C; A, B) 761

generalizes the characterization of invertible elements in SExtnuc (A, B) given in

Remark 5.8.4 (using Theorem 5.6.2). This characterization will be used for the
proof of Theorem K.

Proposition 5.9.11 (Absorbing elements). Suppose that B is σ-unital and

stable, satisfies the WvN-property, and that A ⊆ M(B) is a separable stable ex-
act C*-subalgebra, such that idM(B) |A is weakly nuclear and non-degenerate. Let
ϕ : A → Qs (B) = M(B)/B a *-monomorphism, and let h0 the restriction of πB ◦δ∞
to A.
Define the “natural” lower semi-continuous action of Prim(B) on A by
ΨA (J) := A ∩ M(B, J) for J ∈ I(B) .
Then ϕ defines an element in G(h0 ; A, Q(B)) ∼= Extnuc (Prim(B); A, B) (i.e.,
ϕ and h0 dominate in Q(B) each other), if and only if,

(i) ϕ is nuclear, and

(ii) ϕ(ΨA (J)) = πB (M(B, J)) ∩ ϕ(A) for J ∈ I(B).

Proof. The weak nuclearity of id |A and the exactness of A imply, that

H0 := δ∞ |A is nuclear, cf. Chp.3: if A is exact then ‘‘weakly nuclear’’
H0 (norm-)‘‘nuclear’’ ??. Since δ∞ (M(B)) ∩ M(B, J) = δ∞ (M(B, J)), we
have H0 (ΨA (J)) = H0 (A) ∩ M(B, J) for J ∈ I(B). Since A is exact, this, together
with the nuclearity of H0 , implies that H0 is ΨA -Ψup -residually nuclear.
Let e denote a strictly positive element of B. Since Eϕ := {b ∈ M(B) : πB (b) ∈
ϕ(A)} is stable by Corollary 5.5.16 and Remark 3.10.9(iii), we find a stable separable
C *-subalgebra C of M(B), such that e ∈ C and πB (C) = ϕ(A).
Let h := ϕ−1 πB : C → A, then h is residually equivariant, because h(C ∩
M(B, J)) ⊆ ΨA (J), by (ii), because πB (C ∩ M(B, J)) ⊂ ϕ(A) ∩ πB (M(B, J)) =
ϕ(ΨA (J)).
Thus, T := H0 ◦ h is a non-degenerate weakly residually nuclear C *-morphism
from C into M(B).
By Theorem 5.9.3, there exists an isometry S in M(B), such that S ∗ cS −T (c) ∈
B for c ∈ C. This means that ϕ dominates h0 := πB ◦ H0 .

Let a ∈ A+ and ε > 0. By Lemma 5.9.10, (a − ε)+ is in ΨA (J) = A ∩ M(B, J)

for J := span(B(a − ε)+ B). By condition (ii), ϕ((a − ε)+ ) ⊆ πB (M(B, J)).
On the other hand, again by Lemma 5.9.10, M(B, J) is contained in the norm-
closed ideal of M(B) that is generated by H0 (a) = δ∞ (a). Thus, ϕ((a − ε)+ ) is
contained in the closed ideal of Qs (B), that is generated by h0 (a).
If follows that, for every closed ideal I of A, ϕ(I) is contained in the ideal of
Qs (B) that is generated by h0 (I). Since h0 ⊕ h0 is unitarily equivalent to h0 ,
and since ϕ is nuclear, it follows that h0 approximately dominates ϕ by Corollary
−1
3.10.8(ii). The C *-algebra πB (h0 (A)) = δ∞ (A) + B is stable, because δ∞ (A)
is stable and is a non-degenerate C *-subalgebra. The commutant h0 (A)0 ∩ Qs (B)
762 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

contains a copy of O∞ = C ∗ (s1 , s2 , . . . ) unitally, because δ∞ (A)0 ∩M(B) contains a

copy of O∞ = C ∗ (s1 , s2 , . . .) unitally by Lemma 5.1.2(i). If c1 , c2 , . . . is a sequence
of contractions in Qs (B) such that k c∗n h0 (a)cn − ϕ(a) k → 0, then dn := sn cn
satisfy the assumptions (i) and (ii) of Lemma 5.9.6 (for h0 (A) and ϕ in place of
A and V there). By Lemma 5.9.6 there exists an isometry T ∈ M(B) such that
πB (T )∗ h0 (·)πB (T ) = ϕ. Thus a 7→ T ∗ δ∞ (a)T is a ΨA residually nuclear lift of


ϕ. Thus [ϕ] ∈ G h0 ; A, Qs (B) . I.e. [ϕ] is the Busby invariant of an element in
Extnuc (Prim(B); A, B). 

Remark 5.9.12. Let D∞ := O∞ ⊗ O∞ ⊗ . . .. (It is isomorphic to O∞ by

Corollary H, or by a result of Lin and Phillips [531].)
We consider ψ1 : a ∈ D∞ → 1 ⊗ a ∈ D∞ ⊗ D∞ and an isomorphism ψ2 from
D∞ onto D∞ ⊗ D∞ . (ψ1 and ψ2 are unitarily homotopic by Corollary H.)
It holds C ∼
= C ⊗ D∞ for separable hereditary C *-subalgebras C of B ⊗ D∞ ,
cf. [443].
Suppose that A and B are separable, and that

0 → B ⊗ D ∞ → E → A ⊗ D∞ → 0

is an extension. Then there exists an isomorphism λ from E onto E ⊗ D∞ , cf. [443,

chp. 8].
If A and B are stable then E is stable by Corollary 5.5.16, because B ⊗ D∞ is
purely infinite.

Question 5.9.13. Suppose that A and B are separable stable C *-algebras and
that
ϕ1 , ϕ2 ∈ Hom(A ⊗ O∞ , Q(B ⊗ O∞ ))

are (not necessarily c.p. liftable) Busby invariants, and suppose that ϕ1 and ϕ2 are
unitarily homotopic (by unitaries in Q(B ⊗ O∞ )). Does it imply ϕ1 ≈ ϕ2 ?
(Recall that ϕ1 ≈ ϕ2 means the rather strong property that there is a unitary
u ∈ M(B ⊗ O∞ ) with ϕ1 (a) = π(u)ϕ2 (a)π(u∗ ) for all a ∈ A ⊗ O∞ .
NEXT TRUE?:
The question has a positive answer, if - in addition - ϕ1 has a completely
positive lift. The latter is e.g. the case, if A is nuclear, cf. [443].)

Collection/ to be sorted/ to be integrated: ??

next used in Chapter 8:
The next lemma shows that the relation introduced by J. Cuntz for an alterna-
tive definition of Kasparov’s Ext(A, B) also hold for our Ext(C; A, B) with obvious
modifications.

Lemma 5.9.14. Suppose that A is separable and that B is σ-unital and stable,
and that C ⊆ CP(A, B) is a point-norm closed matrix operator-convex cone.
 9. THE Ψ-RESIDUALLY NUCLEAR CASE AND Ext(C; A, B) 763

Let S denote the set of pairs (ϕ, p), where ϕ : A → M(B) is a C*-morphism
with b∗ ϕ(·)b ∈ C for all b ∈ B and p ∈ M(B) is a projection that satisfies ϕ(a)p −
pϕ(a) ∈ B for all a ∈ A.
We denote by [S] the set of unitary equivalence classes [(ϕ, p)] using unitaries
in M(B). An element (ϕ, p) ∈ S is degenerate if pϕ(a) = ϕ(a)p for all a ∈ A.

(i) [S] is a commutative semigroup (with Cuntz-Addition on maps and pro-

jections).
(ii) The map [ϕ, p] 7→ πB (ϕ(·)p) ∈ Hom(A, Q(B)) defines an additive semi-
group morphism ϑ from [S] onto SExt(C; A, B). The image of ϑ contains
[0] + SExt(C ; A, B).
(iii) The map S → Ext(C; A, B) induces an equivalence relation ∼ on S that
is compatible with Cuntz addition and is generated by the following oper-
ations and relations:
(a) unitary equivalence by unitaries in M(B), i.e., (ϕ, p) ∼ (ψ, q) if there
is a unitary u ∈ M(B) with ψ = u∗ ϕ(·)u and q = u∗ pu ,
(b) addition (ϕ ⊕ ψ, p ⊕ q) of degenerate elements (ψ, q) ∈ S to elements
(ϕ, p) ∈ S , i.e., (ϕ ⊕ ψ, p ⊕ q) ∼ (ϕ, p) , and
(c) unitary perturbation: (ϕ, p) ∼ (ϕ, q) if there is b∗ = b ∈ M(B) such
that ϕ(a)b − bϕ(a) ∈ B and (p − e−ib qeib )ϕ(a) ∈ B for all a ∈ A.
(iv) The natural semigroup morphism from SExt(C; A, B) into Ext(C; A, B)
is an epimorphism.
(v) If C is countably generated and non-degenerate, and if H : A⊗K → M(B)
is as in Corollary 5.4.4, then the unitary equivalence classes [(H, p)] ∈ [S]
build a sub-semigroup [P] of [S] such that ϑ maps [P] onto Ext(C; A, B).

Proof. Let S, T ∈ M(B) isometries with SS ∗ + T T ∗ = 1, and let s :=

πB (S) = S + B and t := πB (T ) = T + B in Q(B) := M(B)/B.
(i): Unitary equivalence (ϕ, p) ≈ (ψ, q) for (ϕ, p), (ψ, q) ∈ S means that there
is a unitary U ∈ M(B) with ψ = U ∗ ϕ(·)U and q = U ∗ pU . Now straight calculation
shows that the Cuntz addition

(ϕ, p) ⊕ (ψ, q) := (ϕ ⊕S,T ψ, p ⊕S,T q)

defines an operation on S. It is associative and commutative on the set [S] of

unitary equivalence classes [ϕ, p]≈ of elements (ϕ, p) ∈ S, by Proposition 4.3.2.
(ii): The map
γ(ϕ, p) : a ∈ A 7→ πB (ϕ(a)p) ∈ Q(B)

is a *-morphism from A into Q(B) if (ϕ, p) ∈ S, and the c.p. map V = pϕ(·)p ∈
CP(A, M(B)) satisfies b∗ V (·)b = (pb)∗ ϕ(·)pb ∈ C for each b ∈ B, because ψ(·) is
in the closure of C with respect to the strict topology (by definition of S). Thus,
γ(ϕ, p) is in Homl (C; A, Q(B)) and defines an Element [γ(ϕ, p)] of

SExt(C; A, B) := [ Homl (C; A, Q(B)) ] .

764 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

Clearly, γ(U ∗ ϕ(·)U, U ∗ pU ∗ ) = πB (U )∗ γ(ϕ, p)πB (U ) for unitary U ∈ M(B). One

gets [γ(ϕ ⊕ ψ, p ⊕ q)] = [γ(ϕ, p)] + [γ(ψ, q)] in SExt(C; A, B) by straight calculation.
We define ϑ : [S] → SExt(C; A, B) by

ϑ([(ϕ, p)]≈ ) := [γ(ϕ, p)] .

If ρ ∈ Homl (C; A, Q(B)), then there is a c.p. contraction V : A → M(B) with

b∗ V (·)b ∈ C for all b ∈ B (by definition of Homl (C; A, Q(B))). By Lemma 3.6.24,
there is a *-morphism ψ : A → M(B) with b∗ ψ(·)b ∈ C and isometries s1 , t1 ∈
M(B) such that t∗1 ψ(·)t1 = V and s1 s∗1 + t1 t∗1 = 1. Let U := s1 S ∗ + t1 T ∗ , ϕ(·) :=
U ∗ ψ(·)U and p := SS ∗ . Then (ϕ, p) ∈ S and 0 ⊕s,t ρ = γ(ϕ, p). Thus [0] +
SExt(C; A, B) ⊆ ϑ([S]).
(iii) +(iv): Recall that Ext(C; A, B) := Gr(SExt(C; A, B)) is the Grothendieck
group of the semigroup of unitary equivalence classes of morphisms ρ ∈
Homl (C; A, Q(B)) by unitaries u = πB (U ) for unitaries U in M (B).
Let (ϕ, p) ∈ S, q ∈ M(B) a projection and U ∈ M(B) a unitary with (p −
q)ϕ(A) ⊆ B, respectively with U ϕ(a) − ϕ(a)U ∈ B for all a ∈ A. Then (ϕ, q) and
(ϕ, U ∗ pU ) are in S, i.e., qϕ(a) − ϕ(a)q ∈ B and U ∗ pU ϕ(a) − ϕ(a)U ∗ pU ∈ B for
all a ∈ A. Moreover, γ(ϕ, p) = γ(ϕ, q) and πB (U )∗ γ(ϕ, p)πB (U ) = γ(ϕ, U ∗ pU ) in
Homl (C; A, Q(B)). Thus, the relations (a) and (c) on S are stronger (or equal) to
the relation on S induced by ϑ : [S] → SExt(C; A, B).
If (ϕ, p) ∈ S is degenerate, then ψ := δ∞ ◦ ϕ and q := δ∞ (p) ( 23 ) satisfy
b∗ ψ(·)b ∈ C for all b ∈ B and qψ(a) = ψ(a)q for all a ∈ A. Thus (ψ, q) ∈ S.
Since K∗ (M(B)) = 0 and s∗1 s2 = 0, it follows from Lemma 4.2.6(ii) that there is
an isometry R ∈ M(B) with RR∗ = 1 − s1 s∗1 . More explicitly the series sn+1 s∗n
P
∗
P
converges strictly in M(B) to an element R := sn+1 sn of M(B) by Remark
5.1.1(2). The element R is an isometry in M(B) with RR∗ = 1 − s1 s∗1 and Rsn =
sn+1 for n = 1, 2, . . ..
Then U := Ss∗n + T R∗ is unitary, U ∗ (ϕ ⊕S,T ψ)U = ψ and U ∗ (p ⊕S,T q)U =
q, i.e., ϑ([ϕ, p]) + ϑ([ψ, q]) = ϑ([ψ, q]) in SExt(C; A, B). Thus, [ϑ([ϕ, p])] = 0 in
Ext(C; A, B) if (ϕ, p) is a degenerate element in S. It implies that (ϕ1 , p1 )⊕S,T (ϕ, p)
and (ϕ1 , p1 ) define the same class in Ext(C; A, B) for all (ϕ1 , p1 ) ∈ S if (ϕ, p) is
degenerate.
It follows that the equivalence relation (ϕ, p) ∼ (ψ, q) on S defined by the oper-
ations (a), (b) and (c) is stronger or equal to the equivalence relation (ϕ, p) ∼ (ψ, q)
defined by [ϑ([ϕ, p])] = [ϑ([ψ, q])] in Ext(C; A, B). If we consider the semigroup [S]
then (a) becomes equality and (b) describes there addition by elements in the sub-
semigroup D of [S] of unitary equivalence classes of degenerate elements. Thus,
relation (b) is compatible with Cuntz addition on [S]. Straight calculations show
that also relation (c) is compatible with Cuntz addition on [S]. It follows that the

23Here δ : M(B) → M(B) is defined by a sequence of isometries s , s , . . . ∈ M(B) with

∞ 1 2
sn s∗n = 1, as considered in Remark 5.1.1(8) and Lemma 5.1.2.
P
strictly converging sum
 9. THE Ψ-RESIDUALLY NUCLEAR CASE AND Ext(C; A, B) 765

relation ∼ – coming from (a), (b), (c) – defines on [S] an equivalence relation that
is compatible with addition and that [(ϕ, p)] ∼ [(ψ, q)] implies [(ϕ, p)] ∼c [(ψ, q)].
To see that ∼c is equal to ∼, it suffices now to show that [S]/ ∼ is a group and
that ϑ([ϕ, p]) = ϑ([ψ, q]) implies (ϕ, p) ∼ (ψ, q).
We show first that ϑ([ϕ, p]) = ϑ([ψ, q]) implies (ϕ, p) ∼ (ψ, q), which is the
crucial and most involved part of the whole proof:

By definition of ϑ and of the equivalence classes [Homl (C; A, Q(B))] ,

there is a unitary U ∈ M(B) such that γ(ϕ, p) = γ(U ∗ ψ(·)U, U ∗ qU ) . Since
(U ∗ ψ(·)U, U ∗ qU ) ∼ (ψ, q) by relation (a). We may rename U ∗ ψ(·)U and U ∗ qU by
ψ respectively by q.
Thus we have to study the implication of the property

ϕ(a)p − ψ(a)q ∈ B ∀ a ∈ A.

Let ρ := (ϕ ⊕ ψ) ⊕ 0, p0 := (p ⊕ 0) ⊕ 0, q 0 := (0 ⊕ q) ⊕ 0, V := S(ST ∗ + T S ∗ )S ∗ ⊕
T T ∗ . Then ρ(·)T = 0, i.e., ρ dominates zero, (ρ, p0 ) ∼ (ϕ, p) and (ρ, q 0 ) ∼ (ψ, q)
by relation (b), and ρ(a)p0 − V ∗ ρ(a)q 0 V ∈ B for all a ∈ A. Let s1 , s2 , . . . ∈
sn s∗n = 1, and let λ := δ∞ ◦ ρ, p00 :=
P
M(B) a sequence of isometries with
s1 p s1 , q = s1 q s1 and W := δ∞ (V ). It holds (λ, p00 ) ∼ (ρ, p0 ), (λ, q 00 ) ∼ (ρ, q 0 ),
0 ∗ 00 0 ∗

and λ(a)p00 − W ∗ λ(a)q 00 W ∈ B for a ∈ A. Moreover, λ(·) dominates zero, the

algebra C ∗ (λ(A), W ) is contained in δ∞ (M(B)) and there is a copy C ∗ (S1 , T1 ) ∼= O2
0
unitally contained in δ∞ (M(B)) ∩ M(B), cf. Remark 5.1.1(8).
It follows that Lemma 4.6.11 applies to D := A, E := Q(B), h := πB ◦ λ,
the unitary πB (W ), and the projections p1 := πB (p00 ), q1 := πB (q 00 ), because
K∗ (M(B)) = 0 by Lemma 5.5.9(i):
We get the existence of u ∈ U0 (h(A)0 ∩ Q(B)) with u∗ (p1 ⊕ 1 ⊕ 0)u = (q1 ⊕ 1 ⊕ 0)
with Cuntz addition ⊕ taken with respect to the generators s1 := πB (S1 ) and
t1 := πB (T1 ) of a unital copy of O2 in πB (λ(A)0 ∩M(B)) ⊆ h(A)0 ∩Q(B). It follows
that there are self-adjoint h1 , h2 , . . . , hm ∈ M(B) with hk λ(a) − λ(a)hk ∈ B for
k = 1, . . . , m such that u = πB (U ) for U := eih1 · . . . · eihm .
Note that λ ⊕S1 ,T1 λ = λ. Thus (λ, p00 ) ∼ (λ, p00 ⊕ 1 ⊕ 0) = (λ, p00 ) ⊕ (λ, 1 ⊕ 0)
and (λ, q 00 ) ∼ (λ, q 00 ⊕ 1 ⊕ 0) . Let p2 := p00 ⊕ 1 ⊕ 0 and q2 := q 00 ⊕ 1 ⊕ 0, then
U ∗ p2 U − q2 ∈ B.
The equivalence relation ≈(c) generated by (c) is stronger than ∼ and yields
that (λ, p2 ) ≈(c) (λ, q2 ) if there are selfadjoint h1 , . . . , hm ∈ M(B) with hk λ(a) −
λ(a)hk ∈ B for k = 1, . . . , m and U ∗ p2 U − q2 ∈ B for U := eh1 · . . . · ehm .
All together shows that (ϕ, p) ∼ (ψ, q).
We check that [S]/ ∼ is a group:
For (ϕ, p) ∈ S, the element (ϕ, 1 − p) is in S and

(ϕ, 1 − p) ⊕ (ϕ, p) ∼ (ϕ, 1) ⊕ (ϕ, 0) ∼ (0, 0) .

766 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

This is because [u, ϕ(a) ⊕s,t ϕ(a)] ∈ B for all a ∈ A and u∗ (1 ⊕ 0)u = (1 − p) ⊕ p,
where u := exp(iH) = s(1−p)s∗ +t(1−p)t∗ +spt∗ −tps∗ for H := i(π/2)(tps∗ −spt∗ )
(respectively with z = s(1 − p)s∗ + spt∗ ) in (iii,b) ( 24 ). It follows that the natural
image of the semi-group SExt(C; A, B) in Ext(C; A, B) is a group, because the
image of ϑ : [S] → SExt(C; A, B) contains [0] + SExt(C; A, B) by part (ii) and
because [0] + [0] = [0].
If we combine this with the observation that the image of SExt(C; A, B)
generates Ext(C; A, B) , then we get that the natural semigroup morphism from
SExt(C; A, B) into Ext(C; A, B) is a semigroup epimorphism.
(v): Suppose that C is countably generated and that h0 : A → Q(B) is defined
as in Corollary 5.4.9. If (ϕ, p) ∈ S, then there is a unitary u ∈ M(B) and a projec-
tion q ∈ M(B) such that u∗ (H(a) ⊕ 0)qu − ϕ(a)p ∈ B for all a ∈ A, cf. Corollary
5.4.9(ii). Since there is a suitable copy of O2 = C ∗ (S1 , T1 ) unitally contained in
(H ⊕ 0)(A)0 ∩ ∩M(B), it follows that the elements (H ⊕ 0, q) build a sub-semigroup
of S under Cuntz-addition. 

next.ref: cor:5.X1.chp12 ??

Corollary 5.9.15. Let A and B stable and separable C*-algebras and let
Ψ : I(B) → I(A) a lower semi-continuous action of Prim(B) on A, such that
Ψ(0) = 0 and Ψ−1 (A) = {B}.
Suppose that there exists a non-degenerate weakly Ψ-residually nuclear *-
monomorphism H0 : A → M(B), such that δ∞ ◦ H0 is unitarily equivalent to H0 ,
and, for each J ∈ I(B),

Ψ(J) = H0−1 (H0 (A) ∩ M(B, J)) .

Then the H0 with this properties is uniquely determined up to unitary homotopy.

Proof. Let C(H0 ) ⊆ CP(A, B) the point-norm closed m.o.c. cone generated
by the maps a ∈ A 7→ b∗ H0 (a)b.
Then C(H0 ) is contained in the m.o.c. cone Cnuc (Ψ) of all Ψ-residually nuclear
c.p. maps V : A → B with V (Ψ(J)) ⊆ J for all J ∈ I(B) and [V ] : A/Ψ(J) → B/J
is nuclear.
But, by assumptions, for each J ∈ I(B),

Ψ(J) = H0−1 (H0 (A) ∩ M(B, J)) ,

i.e., that C(H0 ) is separating for the action Ψ.

By Corollary 3.9.1 this implies that the m.o.c. cone C(H0 ) is identical with the
m.o.c. cone of the – with respect to the l.s.c. action Ψ – residually nuclear c.p.
maps.

24 Use that C ∗ (s(1 − p)s∗ , sps∗ , tps∗ ) is naturally isomorphic to C ⊕ M .

2
 9. THE Ψ-RESIDUALLY NUCLEAR CASE AND Ext(C; A, B) 767

If H1 : A → M(B) has the same properties as H0 , – i.e., δ∞ ◦ H1 is unitary

equivalent to H1 and, for each J ∈ I(B),

Ψ(J) = H1−1 (H1 (A) ∩ M(B, J)) ,

and [H1 ] : A/Ψ(J) → M(B/J) is nuclear for every closed ideal J of B –, then this
implies by Corollary 3.9.1 that the m.o.c. cone C(H1 ) is again the m.o.c. cone of all
Ψ-residually nuclear c.p. maps. Thus, C(H0 ) = C(H1 ).
It implies that H0 and H1 are unitarily homotopic by Corollary 3.9.2. 

Ref. to 5.9.16 still exists in Chp. 12!!

Corollary 5.9.16. Suppose that A is separable and B is σ-unital and stable.

If h1 , h2 : A → M(B) are non-degenerate nuclear monomorphisms and define
the same action of I(B) on A, then δ∞ ◦ h1 and δ∞ ◦ h2 are unitarily homotopic.

Chapter 12 asks also for a Corollary

cor:5.?.same.lsc.action.gives.unitary.homotop
that says that for stable σ-unital B the same l.s.c. action of Prim(B) on
C (b) ⊂ M(B) and (– by pull-back also on C ∗ (b) –) on C ∗ (λ(b)) ⊂ M(B) – for
∗

some C *-morphism λ : C ∗ (b) → M(B) implies that δ∞ (b) and δ∞ (λ(b)) are unitary
homotopic in M(B).
Have the hk same kernel J?
Seems to need/check still that the [hk ] : A/J → M(B)
is again nuclear!! Then it reduces to the residually nuclear case
with exact A/J.
The following elementary corollary allows to (in proofs of Chapter 6) the
use of results on the generalized KK–theory KKnuc (X; A, B) (cf. Chapters 1 and
8). next.ref: cor:5.lifting-criterium ?? Is needed that the action
is monotonous upper s.c.?

Corollary 5.9.17. Suppose that A and B are separable and stable, that A is
exact and that B has the WvN–property of Definition 1.2.3. Further suppose that
ψ : A → Q(B) is a nuclear C*-morphism and H : A → M(B) is a non-degenerate
nuclear *-monomorphism. Furthermore suppose that, for J ∈ I(B),

ψ H −1 (H(A) ∩ M(B, J)) ∈ πB (M(B, J)) .

Then H0 := πB ◦ δ∞ ◦ H dominates ψ in Hom(A, Q(B)), i.e.,

[ψ] ∈ SExtnuc (X ; A, B) = S(H0 , A, Q(B))

for the action of X := Prim(B) on A given by

Ψ : J ∈ I(B) ∼
= O(X) 7→ Ψ(J) := H −1 (H(A) ∩ M(B, J)) ,

and the semi-group S(H0 ; A, Q(B) as defined in Chapter 4, Section 4.

768 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

If moreover

ψ −1 (ψ(A) ∩ πB (M(B, J))) = H −1 (H(A) ∩ M(B, J))

for J ∈ I(B), and if there exists a C*-morphism ϕ : A ⊗ O2 → Q(B) with ψ(a) =

ϕ(a ⊗ 1) for a ∈ A, then there exists a unitary U ∈ M(B) such that πB (U ∗ δ∞ ◦
H(a)U ) = ψ(a) for a ∈ A.

next.ref: prop:5.char-Extnuc. needed? ??

Proposition 5.9.18. Suppose that A and B are separable and stable, that
A is exact and that B has the WvN–property (cf. Definition 1.2.3). Further let
ψ : A → Q(B) and H : A → M(B) a weakly nuclear *-monomorphisms such that
BH(A)B generates B.
Let Ψ(J) := H −1 (H(A)∩M(B, J)) for J ∈ I(B), and h0 := πB ◦δ∞ ◦H : A →
Q(B).
Then:
Ψ is a lower semicontinuous action of Prim(B) on A (corresponding to H).

Extnuc (Prim(B), A, B) = [h0 ] + S(h0 ; A, Q(B))

[ψ] ∈ [h0 ] + S(h0 ; A, Q(B)), if and only if, ψ(A) ∩ πB (M(B, I)) = ψ(Ψ(I)) for
all I ∈ I(B).

The following Lemma allows to apply above to the asymptotic case.

next.ref:lem:asympt-absorbtion ??
Chp.7ref:lem:asympt-absorbtion.to.chp.7?
next: check if (1) and (2) are used in chp.7 !!

Lemma 5.9.19. (1) If B is a C*-algebra, then

πSB (M(SB, SI)) ∩ Q(R, B) = Q(R, I)

for all I ∈ I(B), where SB := C0 (R, B) and SI := C0 (R, I).

(2) Suppose that A is separable, B is σ-unital and stable, and h0 : A → B
satisfies [h0 ] + [h0 ] = [h0 ⊕ h0 ] = [h0 ]. Let s1 , s2 , . . . ∈ M(B) a sequence of
isometries such that sn s∗n converges strictly to 1 in M(B), and let H := δ∞ ◦h0 –
P

e.g. realized by s1 , s2 , . . .. Denote by HR (a) := η(H(a)), where η : M(B) → M(SB)

denotes the natural embedding that extends the inclusion map B ⊂ Cb (R, B) ⊂
M(SB).
If HR dominates k : A → Q(R, I) then h0 dominates k.

Proof. (1): Let b ∈ M(SB, SI)) and c ∈ Cb (R, B) ⊃ SB := C0 (R, B)

with b + SB = c + SB. There is a ∈ SB with b = c + a ∈ Cb (R, B). Thus
b(t) ∈ B ∩ M(B, I) = I for t ∈ R, i.e., b ∈ Cb (R, I). Since Cb (R, I) ∩ SB = SI it
follows b + SB ∈ Q(R, I) ∼
= πSB (Cb (R, I)).
 9. THE Ψ-RESIDUALLY NUCLEAR CASE AND Ext(C; A, B) 769

(2): Let d ∈ M(SB) a contraction with πSB (d∗ H R d) = k. By assumption,

there is a unitary u ∈ M(B) with u∗ s1 h0 (·)s∗1 u + u∗ s2 h0 (·)s∗2 u = h0 , and H =
∗ ∗ n ∗ ∗
P
n sn h0 (·)sn . Let tn = (u s1 ) u s2 . Then tn tm = δn,m 1 and tn h0 (·) = h0 (·)tn .

There exists a positive contraction e ∈ Cb (R, B) and a strictly increasing con-

Pk
tinuous function µ : R+ → (0, ∞) with e(t) ≤ n=1 sn s∗n for k ≥ µ(t), ed−de ∈ SB,
HR (a)e − eHR (a) ∈ SB for all a ∈ A and πSB (e)k(a) = k(a) = k(a)πSB (e) for
all a ∈ A. This can be seen with help of a convex commutative approximate
unit of B, that approximately commutes with the elements of H(A), s1 , s2 , . . .,
k(A)t (t ∈ [−n, n]) d|[−n, n]. Here e
e k : A → Cb (R, B) is some continuous topolog-
ical (not necessarily linear) lift of the C *-morphism k : A → Cb (R, B)/SB. Then
t 7→ g(t) := 1≤n≤[µ(t)]+1 tn s∗n e(t)d(t)e(t) is a norm-continuous map from R into
P

the contractions of B. It satisfies g ∗ h0 (a)g − d∗ HR (a)d ∈ SB for all a ∈ A, because

e2 d∗ HR (a)de2 − d∗ H R d ∈ SB.
to be filled in: check last 2 formulae ?? 

where is ref.cor:5.to-be-named used? ??

Corollary 5.9.20. Suppose that A and B are stable C*-algebras, where A is

separable B is σ-unital, and that H0 : A ,→ M(B) and ϕ : A ,→ Q(B) = M(B)/B
are nuclear *-monomorphisms with the properties that H0 (A)B is dense in B and,
for every a ∈ A, πB (H0 (a)) and ϕ(a) generate the same closed ideal of Q(B).
Then, if at least one of the following conditions (i) or (ii) is satisfied, there is a
unitary U ∈ M(B) such that ϕ = πB (U ∗ H0 (·)U ).

(i) B ∼
= C ⊗ O2 ⊗ O2 ⊗ . . ., or
−1
(ii) πB (ϕ(A)) is stable, and there are C*-morphisms H : A ⊗ O2 ,→ M(B)
and Φ : A ⊗ O2 ,→ Q(B) = M(B)/B with H0 (a) = H(a ⊗ 1) and ϕ(a) =
Φ(a ⊗ 1) for a ∈ A.

Proof. to be filled in ?? 

cite ? cor:5.X.1.chp8 ??

Corollary 5.9.21. Suppose that A and B are stable, A separable and B σ-

unital with strictly positive element e. Let H1 denote the infinite repeat of a faithful
*-representation ρ : A → L(H) ∼ = M(K) ⊂ M(B).
If h : A → M(B) is any C*-morphism such that a ∈ A 7→ eh(a)e generates
CPnuc (A, B) as a m.o.c. cone, then δ∞ ◦ h is unitarily homotopic to H1 : A →
M(B).
It holds Extnuc (A, B) = G(H0 , A, Q(B)) for H0 := πB ◦ H1 .

Proof. cite ? cor:5.X.1.chp8 ?? 

ref {cor:5.Y?1.chp9} (??) needed in this generality??

770 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

Corollary 5.9.22. Suppose that A and B are stable, A is separable and B is

σ-unital. Let C ⊆ CP(A, B) a countably generated non-degenerate operator convex
cone, and H0 : A → M(B) the non-degenerate C*-morphism described in Corollary
5.4.4. If Y is a locally compact space, let HY := πC0 (Y,B) ◦ X ◦ H0 and EY :=
M(C0 (Y, B))/ C0 (Y, B), where X : M(B) → M(C0 (Y, B)) is the natural strictly
continuous embedding of M(B) into M(C0 (Y, B)) ∼ = Cb,st (Y, M(B)).
Then:

(i)

Ext(C ⊗ CP(C, C0 (Y )); A, C0 (Y, B)) = G(HY , A, EY ) =: F (C, Y ) .

In particular,

Ext(C ⊗ CP(C, C0 (R)); A, SB) = G(HR , A, ER ) =: F (C, R)

for the above defined HR : A → ER .

(ii) In particular, if X is a T0 space that acts on A and B by ΨA respectively
ΨB , and if H0 (and then HR ) is formed with respect to the cone of ΨA -
ΨB –residually nuclear maps from A into B, then

Extnuc (X; A, SB) = G(HR , A, ER ) .

(iii) If H0 : A → M(B) is the infinite repeat of a non-degenerate h0 : A → B

with h0 ⊕ h0 ∼ = h0 , then HR dominates h0 : A → Q(R+ , B) ⊂ ER :=
Q(SB) . Every element h ∈ S(h0 ; A, ER ) is also in S(HR ; A, ER ) and
[h ⊕ H0 ] ∈ G(HR , A, ER ).
(iv) In particular, if A is exact and h0 is nuclear, then for h : A → Q(R+ , B)
holds [h ⊕ H0 ] ∈ Extnuc (X; A, SB) if and only if [h] ∈ S(h0 , A, ER ).
(Here the action of X := Prim(B) on A is defined by h0 .)

We have e.g. ER := M(SB)/SB if B is stable SB := C0 (R, B).

next: ref: cor:5.Y5.chp9

Corollary 5.9.23. Suppose that D is separable and stable, B is σ-unital and

stable, and that k0 : D → B is a non-degenerate *-monomorphism such that k0 ⊕ k0
is unitarily homotopic to k0 .
Let C ⊆ CP(D, B) denote the point-norm closed m.o.c. cone of c.p. maps gen-
erated by k0 . Let C(R) := C ⊗ CP(C, C0 (R)).
We define a *-monomorphism

H0 := δ∞ ◦ k0 : D → M(B).

Let HR := IH0 : D → ER := Q(SB), where I : M(B) → Q(SB) is the natural

embedding.
Then
Ext(C(R) ; D, SB) ∼
= G(HR ; D, ER ).
 9. THE Ψ-RESIDUALLY NUCLEAR CASE AND Ext(C; A, B) 771

If D is exact, we have moreover,

Extnuc (X; D, SB) = G(HR , D, ER ) = [HR ] + [Homnuc (X; D, ER )]

for X := Prim(B) and action ΨD (J) := k0−1 (k0 (D) ∩ J) for J ∈ I(B) ∼
=
O(Prim(B)).

Let A and B stable C *-algebras, where A is separable and B is σ-unital. Sup-

pose that the operator convex cone Crn (ΨA ) of all Ψ-residually nuclear maps from
A to B for a given lower semi-continuous action ΨA : O(Prim(B)) ∼ = I(B) → I(A)
of Prim(B) on A is countably generated (as e.g. in the case of separable B).
The universal C *-morphism Hrn : A → M(B) , that is determined (and de-
fined) by m.o.c. cone Crn (ΨA ) := Crn (ΨA , id; A, B) in the sense of Corollary 5.4.4,
will be called the universal weakly residually nuclear map from A into M(B).
Recall that Hrn is uniquely determined up to unitary homotopy by Corollary
5.4.4. It is now known that Crn (ΨA ) – and therefore also Hrn – defines the action
Ψ, because B has “weak Abelian separation” which is (formally ?) stronger than
residually separation for B. But the proof of “weak Abelian separation” for B
(which is a weaker property than the existence of an ideal-separating regular abelian
C *-subalgebra of B) deserves several steps, e.g. that every coherent Dini space X
is the primitive ideal space of a separable nuclear C *-algebra that has Abelian
separation, and study some “almost minimal” embedding of of Dini spaces into
coherent Dini spaces. It uses some results of this book, but can not summarized
here.

Proposition 5.9.24. If B is stable and separable, then the universal residually

nuclear map Hrn : B → M(B) exists for ΨB = idI(B) .
Separable B has residually nuclear separation (cf. Definition 1.2.3), if and only
if, Hrn is a non-degenerate *-monomorphism and Hrn (J) = Hrn (B) ∩ M(B, J) for
all J ∈ I(B).
If B 6= {0} is simple separable and stable, then Hrn : B → M(B) is unitarily
homotopic to M(ι) ◦ ρ for any non-degenerate *-morphism ι : K 7→ M(B) and any
non-degenerate *-representation ρ : B → L(`2 ) = M(K).
In particular, every simple separable C*-algebra has residually nuclear separa-
tion.

Proof. to be filled in ?? 

Used in Chp.8, ref: prop:5.stabilty.of.Ext.Co :

Stability, (partial) Functoriality of Ext?

Proposition 5.9.25. Suppose that A and B are σ-unital and that C ⊆

CP(A, B) is a point-norm closed m.o.c. cone.
772 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

Then the inclusions ιA : A ∼

= A ⊗ p11 ⊆ A ⊗ K and ιB : B ∼
= B ⊗ p11 ⊆ B ⊗ K
induce natural isomorphisms

i(A, B) := (ιA )∗ : Ext(C; A ⊗ K, B) → Ext(C; A, B) (9.1)

j(A, B) := (ιB )∗ : Ext(C; A, B) → Ext(C; A, B ⊗ K) , (9.2)

such that (ιB )∗ ◦ ((ιA )∗ )−1 = ((ιA )∗ )−1 ◦ (ιB )∗ in the sense

j(A ⊗ K, B) ◦ i(A, B) = i(A, B ⊗ K) ◦ j(A, B) .

If, moreover, C ⊆ CP(A, B) is countably generated, A, A0 are separable, B, C

are σ-unital, then ψ ∈ Hom(B, C), then there is group morphism

ψ∗ : Ext(C; A, B) → Ext(C(ψ) ◦ C; A, C) .

A *-morphism ϕ ∈ Hom(A0 , A) defines a group morphism

ϕ∗ : Ext(C; A, B) → Ext(C ◦ ϕ; A0 , B) .

Proof. ??
This was outlined in a proof of the corresponding (non-nuclear) Proposition [73,
17.6.5]. Note here that the stabilization of any relation that contain at least the
(unitary) isomorphisms, automatically allows in the class of each representative
also the Cuntz addition of so-called “degenerate” pairs, which can be defined as
those with the property that their infinite Hilbert B-module sum defines again a
C-compatible Kasparov module. Therefore they have no significance for us, because
we pass anyway to the stabilized relations, i.e., to the Grothendieck group. 

Compare following Remark with Remark 5.9.27

Split both Remarks into one ‘‘common Remark’’ and a Proposition
Change then all references to them!!!

Remark 5.9.26. The reader should notice that in general the closed ideal
I(δ∞ (B)) of M(B) generated by δ∞ (B) for a separable stable C *-algebra
B is not identical with M(B) e.g. even for B = C0 (0, 1] ⊗ K, because
M(B) ∼ = Cb,st (R+ , M(K)) is unital and 1 is not in the ideal J of Cb,st (R+ , M(K))
that is generated by δ∞ (C0 ((0, 1], K)), because it is contained in the ideal generated
by C0 ([0, ∞)) ⊗ 1.
Following properties of σ-unital stable B are equivalent:

(i) I(δ∞ (B)) = M(B).

(ii) Prim(B) is quasi-compact.

Equivalent to 0(B) quasi-compact?

(iii) B ⊗ O2 contains a full projection.

We can replace in (iii) B ⊗O2 by B ⊗O∞ , because O∞ ⊆ O2 (unital) and O2 ⊆ O∞

(non-unital).
 9. THE Ψ-RESIDUALLY NUCLEAR CASE AND Ext(C; A, B) 773

Proof. Since L(`2 ) = ∼ M(K) ⊆ δ∞ (B))0 ∩ M(B) (by a strictly continuous

unital monomorphism), δ∞ (B) commutes with a unital copy of O2 ∼ = C ∗ (t1 , t2 ) ⊆
M(K) and a sequence of isometries s1 , s2 , . . . ∈ M(K) with n sn s∗n strictly con-
P

vergent to 1 and ti sj = sj ti , t∗i sj = sj t∗i . Let e ∈ B+ a strictly positive contraction.

(i)⇒(ii): The element δ∞ (e) is a strictly positive contraction for B + δ∞ (B). And
δ∞ (e) is properly infinite because δ∞ ◦ δ∞ is unitarily equivalent to δ∞ . Thus,
there are ε > 0, X ∈ M(B) with X ∗ δ∞ ((e − ε)+ )X = 1. It follows that e is in the
closed ideal I((e − ε)+ ) of B generated by (e − ε)+ , i.e., B = I((e − ε)+ ) . Thus
kπJ ((e − ε)+ )k = 0 implies J = B for closed ideals J of B, and kJ + ak ≥ ε
for all J 6= B. Since the set Yε ⊆ Prim(B) of primitive ideals J of B with
kπJ (e)k = kJ + ek ≥ ε is a quasi-compact Gδ -subset, of Prim(B), it follows that
Prim(B) = Yε is quasi-compact.
(ii)⇒(iii): The algebra B ⊗ O2 is (strongly) purely infinite by Corollary ??. If
Prim(B) is quasi-compact, then the purely infinite algebra B ⊗ O2 has quasi-
compact primitive ideal space Prim(B ⊗ O2 ) ∼ = Prim(B).
By Proposition 2.10.4 B ⊗ O2 contains a full projection.
(iii)⇒(i): If B ⊗ O2 contains a full projection p, then there is a non-degenerate
C *-morphism h : K → B ⊗ O2 .
Above we have seen that we can define by µ(b ⊗ tj ) = δ∞ (b)tj a *-
monomorphism µ : B ⊗ O2 → M(B) with µ(b ⊗ 1) = δ∞ (b) and with δ∞ ◦ µ = µ
(if we take new isometries s1 , s2 , . . . for the definition of the new infinite repeat
δ∞ ). It follows, I(δ∞ (B)) = I(µ(B ⊗ O2 )) = I(µ ◦ h(K)) in M(B). The morphism
µ ◦ h(K) ⊆ M(B) is non-degenerate, i.e., M(µ ◦ h) : M(K) → M(B) exists and is
unital.
If we use the sequence s1 , s2 , . . ., then δ∞ (µ◦h(x)) = µ◦h(x) for each x ∈ K. We
find a sequence of isometries r1 , r2 , . . . in M(K) with n rn rn∗ = 1 for the definition
P

of δ∞ : K → M(K), then M(µ◦h)◦δ∞ is unitarily equivalent to δ∞ ◦µ◦h = µ◦h by

Lemma 5.1.2(i), because there is a unitary U ∈ M(B) with U ∗ M(µ ◦ h)(rn ) = sn
and U ∗ M(µ◦h)(·)U is strictly continuous. Thus, I(µ◦h(K)) ⊃ M(µ◦h)(I(δ∞ (K)).
But there is an isometry S ∈ M(K) ∼ = L(`2 ) with SS ∗ = δ∞ (e1,1 ). It follows
1 ∈ I(µ ◦ h(K)) = I(δ∞ (B)). 

Compare next with Remark 5.9.26

Remark 5.9.27. We consider sometimes elements that are in the ideal gener-
ated by δ∞ (B) in M(B), e.g. in some proofs of Chapters 2, 5, 6, 8 and 12. Therefore
the following rather strong equivalent properties of the ideal of M(B) generated
by δ∞ (B) should be noticed. They show that some care is needed if one establish
results with help of elements in the ideal generated by δ∞ (B).
Suppose that B is stable, then the following are equivalent:

(a) The ideal J of M(B) generated by δ∞ (B) is equal to M(B).

(b) B is σ-unital and prime(B) is quasi-compact.
(c) prime(B) quasi-compact.
774 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

(d) There exists a positive contraction e ∈ B+ such that e is strictly positive

and f := (2e − 1)+ is “full” in B – in the sense that f generates B as
closed ideal –, and that there exists a1 , . . . , an ∈ B such that
n
X
g := 2e − (2e − 1)+ = a∗k f ak .
k=1

If B contains a projection p ∈ B that generates B as a closed ideal, then δ∞ (p)

is the range of an isometry in M(B). In particular, then B satisfies the property
(a) and B ∼ = pBp ⊗ K.
If B is weakly purely infinite – in addition –, then property (a) implies that B
contains a properly infinite projection that generates B as a closed ideal.

Compare proof that stable stable B

with simple M(B)/B is σ-unital.
Is this proof here applicable?

Proof. For the proof of the existence of a countable approximate unit con-
sisting of contractions 0 ≤ b1 ≤ b2 ≤ · · · in B+ from the existence of e ∈ B+ and
Pn
d1 , . . . , dn ∈ M(B) with 1 − k=1 d∗k δ∞ (e)dk ∈ B , see Chapter 2[Cor.2.1.7(i)?] or
proof of Corollary 2.2.11 in case of simple B?
We remind the definition of Dini functions f on topological T0 spaces X (by
requiring that the generalized classical Lemma of Dini is valid for f ):
The function f is lower semi-continuous, and satisfies the Lemma of Dini: If
gτ is an upward directed net of non-negative lower semi-continuous functions on X
that converges point-wise to f on X, then the net converges uniformly on X to f .
An equivalent definition of bounded Dini functions is: The function f is non-
negative, lower semi-continuous and, for every decreasing sequence F1 ⊇ F2 ⊇ · · ·
T
of closed subsets Fn ∈ X, holds sup f ( n Fn ) = inf n sup f (Fn ).
On sober topological T0 spaces this is equivalent to saying that for each t > 0
the set f −1 [t, ∞) is a quasi-compact subset of X. (Notice that it is always a Gδ
subset of X).
The latter characterization implies that each Dini functions f on prime(B) is a
generalized Gelfand transformations N (b) : prime(B) → [0, ∞) where N (b)(J) :=
kb + Jk = kπJ (b)k of some b ∈ B, cf. [447] ( 25 ). This is fairly easy to see directly
for stable weakly purely infinite C *-algebras B.
Notice that prime(B) ⊇ Prim(B) is the “point-wise completion” of Prim(B)
because both have the same lattice of open subsets. The restriction of a Dini
function on prime(B) to Prim(B) is again a Dini function, and each bounded Dini
function f on Prim(B) is a generalized Gelfand transformation N (b) by [447].

25 Typo: Replace b by c in the last line of the proof of [447, lem.3.3]

 9. THE Ψ-RESIDUALLY NUCLEAR CASE AND Ext(C; A, B) 775

(a)⇒(b): Since the (non-zero) elements of δ∞ (B) are properly infinite inside
the hereditary C *-subalgebra D of M(B) generated by δ∞ (B), – because O2 ⊂
M(K) ⊆ δ∞ (B)0 ∩ M(B) – there exists a contraction b ∈ B+ and an operator
T ∈ M(B) with T ∗ δ∞ (b)T = 1.
It follows that there exists γ > 0 and S ∈ M(B) such that S ∗ T ∗ δ∞ ((b −
γ)+ )T S = 1 . That implies that c := (b − γ)+ generates B as a closed ideal. Thus
kπJ (c)k > 0 for all closed ideals of B.
Let e := γ −1 (b − (b − γ)+ ). Then kek ≤ 1 and ec = c. It follows kπJ (e)k ·
kπJ (c)k = kπJ (e)k > 0 for every closed ideal J 6= B. Thus the function 1 = N (e)
is a Dini function on prime(B). It implies that prime(B) is quasi-compact (by
definition of the Dini functions).
(c)⇒(b): prime(B) is quasi-compact, because kπJ (g)k = 1 for all J / B with
J 6= B follows from gf = f and kπJ (f )k > 0 for all J / B, J 6= B.
(b)⇒(c): The definition of Dini functions implies: prime(B) is quasi-compact,
if and only if 1 is a Dini function on prime(B).
Since 1 is the point-wise supremum of the norm functions N (b)(J) := kb+Jk =
kπJ (b)k (for J ∈ prime(B)) on prime(B), where b ∈ B+ with kbk < 1 and since the
positive elements in the open unit ball of B build an upward directed family, the net
of this functions N (b) converges point-wise to 1. Since 1 is Dini, it follows that the
net converges uniformly to 1. In particular, there exists a contraction b ∈ B+ with
kπJ (b)k > 2/3 for all closed ideals J 6= B of B. Let a := min(2b, 1) = 2b−(2b−1)+ .
Then kπJ (a)k = 1 for all closed ideals J 6= B of B.
Indeed: Clearly 0 ≤ a and kak ≤ 1. Take a character χ on C ∗ (πJ (b)) with
χ(πJ (b)) = kπJ (b)k > 2/3. Then χ((2πJ (b) − 1)+ ) = 2χ(πJ (b)) − 1 > 0 and
χ(πJ (a)) = 2χ(πJ (b)) − (2χ(πJ (b)) − 1)+ = 1. Thus, kπJ (a)k = 1.
It follows that ???? and kπJ (4(a − 1/4)+ )k = 1 for all J / B with J 6= B.
This implies that (a − 1/4)+ is again full in B and that there exists n ∈ N and
d1 , . . . , dn ∈ B such that k2b − d∗k (a − 1/4)+ dk k < δ.
P

It follows from Lemma 2.1.9

??????
e∗ d∗k (a − 1/4)+ dk e.
P
that there is a contraction e ∈ B with (2b − δ)+ =
more ???????
(c)⇒(a): We construct a projection Q ∈ δ∞ (B)M(B)δ∞ (B) that majorizes
δ∞ (f ). It follows that P := δ∞ (Q) is in the closed ideal generated by δ∞ (B)
2
(because δ∞ is unitary equivalent to δ∞ in M(B)) and P BP is stable full split-
corner of B. It follows from the stable isomorphism theorem of L.G. Brown [107],
respectively from Kasparov absorption theorem [404] that P is equivalent to 1 in
M(B).
δ∞ (B) commutes with a copy of C ∗ (t1 , t2 , . . .) = O∞ ⊆ M(B).
776 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

If B contains a projection p ∈ B that generates B as a closed ideal, then δ∞ (p)

is the range of an isometry in M(B). In particular, then B satisfies the property
(a) and B ∼ = pBp ⊗ K.
If B is weakly purely infinite –in addition–, then property (a) implies that B
contains a properly infinite projection that generates B as a closed ideal.
If B is weakly purely infinite then there exists n ∈ N such that F := s1 f s∗1 +
· · · + sn f s∗n is properly infinite in B. Let E := s1 es∗1 + · · · + sn es∗n and G :=
s1 gs∗1 + · · · + sn gs∗n . Then G = min(2E, 1), F = (2E − 1)+ and GF = F . Since
Pn
G = k=1 A∗k F Ak for Ak := s1 ak s∗1 + · · · + sn ak s∗n , and F is properly infinite,
there exist T1 , T2 ∈ B such that kTk∗ F Tk − Gk < 1/4 (k = 1, 2). We can manage
that T1∗ F T2 = 0 in addition, because for every ε > 0 there are D1 , D2 ∈ B with
Dj∗ F Dk = δjk (F − ε)+
(by a suitable 2 × 2-matrix arguments using Lemma 2.1.9 ??). It follows that
there exists S1 , S2 with (Tj Sj )∗ F (Tk Sk ) = δjk ·2(G−1/2)+ . Since GF = F it follows
that 2(G − 1/2)+ F = F . We get contractions Vk := F 1/2 Tk Sk with V1∗ V2 = 0 and
V1∗ V1 = V2∗ V2 and Vj∗ Vj Vk = Vk The isometry Z := ????? satisfies ????. 

Refer to inventer of this example!

Remark 5.9.28. An algebra B with properties (a)–(c) of Remark 5.9.27 can be

stably projection-less, e.g. there exist simple stable separable nuclear C *-algebras
B that are stably projection-less.
An easy example of a stably projection-less simple nuclear C *-algebra B is the
inductive limit of the C *-algebras An := C0 ((0, 1], M3n ), with *-monomorphisms
φn : An → An+1 given for f ∈ An and t ∈ (0, 1] by
φn (f )(t) := f (t/2) ⊕ f (t2 ) ⊕ f (t1/2 ) ∈ M3n+1 .
The closed ideals J of An are given by J = C0 (U ) ⊗ M3n , where U is a countable
union of pair-wise disjoint intervals (αn , βn ) 0 ≤ α < β ≤ 1 possibly further united
with (γ, 1] where supn βn ≤ γ < 1 .
If the ideal J(f ) generated by f ∈ A+ n contains C0 (α, β) ⊗ M3n for some 0 <
α < β < 1, then for each given δ > 0 there exists (sufficiently big) k ∈ N such that
the element
(φn+k ◦ · · · ◦ φn+1 φn )(f )
generates an ideal of An+k+1 that contains C0 ((δ, 1], M3n+k+1 ).
Thus, indlimn→∞ (An → An+1 ) is a simple C *-algebra. It is stably projection-
less, because An ⊗ K ∼
= C0 ((0, 1], K) does not contain a non-zero projection.

Remark 5.9.29. Suppose that B is stable and σ-unital, and that A is a pi-sun
algebra. Let ψ : A ⊗ K → Q(B) a C *-morphism with 0 = [ψ] ∈ Ext(A, B).
This implies that ψ has a non-degenerate lift h : A ⊗ K → M(B) where h is a
homomorphism with ψ = πB ◦ h and h(A ⊗ K)B = B.
It follows that M(h) : M(A ⊗ K) → M(B) is unital and injective ?????
 10. THE CASE OF PURELY LARGE EXTENSIONS 777

It follows that ψ has a non-degenerate lift h : A ⊗ K → M(B) , if an only if ,

the closed ideal I(ψ(A ⊗ K)) generated by ψ(A ⊗ K) is equal to Q(B)
????????? why this ???????
and ψ dominates zero
(i.e., there exists an isometry S ∈ M(B) with S ∗ h(A ⊗ K)S ⊆ B ).
The latter condition holds by Proposition 5.5.12, because B + h(A ⊗ K) is
(obviously) stable if h is non-degenerate.
Or is s.p.i. needed? If B is purely infinite, then ψ always dominates zero.
Thus, the question about existence of non-degenerate split morphisms for ψ
reduces to the verification of the necessary and sufficient condition I(Im(ψ)) =
Q(B).
It is always the case if B is, in addition, simple and ψ 6= 0.
There is no useful condition for the existence of possible degenerate lifts. But
the existence of a properly infinite projection p ∈ M(B) with πB (p)ψ(·)πB (p) = ψ
and πB (p) ∈ I(Im(ψ)) seems to be sufficient in the case of purely infinite B.

10. The case of purely large extensions

Definition, from [264] :

Let B be a C *-algebra, and let C be a C *-algebra containing B as a closed
two-sided ideal.
Let us say that C is purely large with respect to B if for every element c ∈ C \B,
the C *-algebra cBc∗ (the intersection with B of the hereditary C *-subalgebra of
C generated by cc∗ ) contains a C *-subalgebra which is stable (i.e., isomorphic
to its tensor product with the C *-algebra K of compact operators on an infinite-
dimensional separable Hilbert space) and is full in B (i.e., not contained in any
proper closed two-sided ideal of B).
EK comment: It implies that Ann(B, C) = {0}
(annihilator of B in C),i.e., B is essential in C.
Criterium:
Let A and B be C *-algebras, and let

0→B→C→A→0

be an extension of B by A (i.e., a short exact sequence of C *-algebras). Let us say

that the extension is purely large if the C *-algebra of the extension, C, is purely
large with respect to the image of B in it, in the sense described above.
Note that, if B is non-zero, a purely large extension of B by A is essential
(that is, the image of B in the C *-algebra of the extension C is an essential closed
two-sided ideal every non-zero closed two-sided ideal has non-zero intersection with
it).
778 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

My comment: B is ‘‘essential’’ in C because B ⊆ C is ‘‘purely

large’’ in C.
Lemma (in Section 7 of [264] ?):
Let C be a C *-algebra that is purely large with respect to a closed two-sided
ideal B (of C ?), in the sense of “Section 1”.
Then, for any positive element c of C which is not in B, any ε > 0, and any
positive element b of B, there exists b0 ∈ B with

kb − b0 cb∗0 k < ε .

If b is of norm one, and if the image of c in C/B is of norm one, then b0 may be
chosen to have norm one.
Theorem (in Section 6 of [264]).
Let A and B be separable C*algebras, with B stable and A unital.
A unital extension B → C → A ?? Notation ?? is absorbing, “in the nuclear
sense”, if, and only if, it is “purely large”.
Let us say, correspondingly, that an extension is absorbing in the nuclear sense
if it absorbs every extension which is trivial in the nuclear sense.
Again, “let us say” that a unital extension is absorbing in the nuclear sense
to mean that “ this holds within the semigroup of (equivalence classes of) unital
extensions”. (With triviality in the nuclear sense the existence of a unital weakly
nuclear splitting.)
EK remark:
Precise math. definition is not given!?
In the case of non-unital A, one has to require in addition
that φ : A → M(B)/B ‘‘dominates zero’’ in sense of Chapter 4,
because only then the map a + α · 1 → φ(a) + α · 1
defines again ???? a purely large extension
???? for the unitization (in general).
EK comments (to Lemma in Section 7 of [264]):
One can take (if B is σ-unital) cn = (1 − e2n )1/2 c(1 − e2n )1/2 , for a suitable
quasi-central unit of B and gets elements bn ∈ B with c − n b∗n cbn ∈ B + 1 .
P

When the algebra C/B is simple?

When elements of C/B are properly infinite in Q(B)?
My (or of others ???????????) conjecture:
This criterium is satisfied for all nuclear separable unital C *-subalgebras of
the stable corona Qs (B) of a σ-unital C *-algebra B, if and only if, B = C or B is
simple and purely infinite.
One could consider A := C[0, 1] and A := C.
 10. THE CASE OF PURELY LARGE EXTENSIONS 779

If B = C ⊗ K with pi-sun C, the the result holds for all unital separable
A ⊂ Qs (B)...?
Questions:
Is it always satisfied for all non-zero stable σ-unital B for given separable simply
purely infinite A, A ⊂ Qs (B), that is not contained in an ideal of Qs (B) ?
Has A ⊆ Qs (B) the property that for each non-zero a ∈ A+ there exist d ∈
Qs (B) with d∗ ad = 1?
What really implies the property that A ,→ Qs (B) absorbs all “hyper-nuclear”
and (at the same time) “hyper-inner” c.p. contraction maps V : A → Qs (B), in the
sense that there exists T ∈ Qs (B) with T ∗ (·)T = V .
Let B σ-unital and stable. It seems that the it would be enough to require that
every a ∈ A+ ⊂ Q(B) is properly infinite in Q(B). (And full??)
Take the stronger assumption that there exists x ∈ M(B) with πB (x)∗ aπB (x) =
kak · · · 1 (for a ∈ A+ ?). It is then equivalent to the property that each a ∈ A+ is
infinite properly infinite in B and ???????????????
Let B σ-unital and stable c ∈ M(B)+ with πB (c2 ) = a ∈ A+ , x ∈ M(B) with
∗
x cx ∈ 1 + B. e ∈ B+ strictly positive contraction, e1 , e2 , . . . suitable quasi-central
approximate unit in C ∗ (e) for C ∗ (c).
Find dn,1 , dn,2 with dn,k d∗n,k ∈ (c − δ)+ B(c − δ)+ , d∗n,1 dn,2 = 0, d∗n,k dn,k = en .
Good? For what?
...

Lemma 5.10.1. (“Kirchberg” cited in [264]). Let C be a unital separable C*-

algebra and let B be an essential closed two-sided ideal of C, so that we may view
C as a unital subalgebra of M(B):

B ⊆ C ⊆ M(B); 1∈C.

Let φ : C → M(B) be a completely positive map which is zero on B, and suppose

that, for every b0 ∈ B, the map

b∗0 φ(·)b0 : C → B,

given by c 7→ b∗0 φ(c)b0 , can be approximated (on finite sets) by the maps

c 7→ b∗ cb, b ∈ B .

It follows that there exists v ∈ M(B) such that

φ(c) − v ∗ cv ∈ B, c ∈ C .

The element v may be chosen so that the map c 7→ v ∗ cv also approximates φ on a

given finite subset of C (up to ε > 0).

Question 5.10.2. Let C ⊂ Qs (B) a separable unital C *-subalgebra of Qs (B)

with σ-unital B.
780 5. GENERALIZED WEYL–VON NEUMANN THEOREMS

Suppose that C satisfies the following criteria that is equivalent to the absorp-
tion criteria of G. Elliott and D. Kucerovsky in [264] (compare J. Gabe [310] for
the non-unital case):

The hereditary C*-subalgebra d∗ (B ⊗ K)d contains a stable and full hereditary

C*-subalgebra of B ⊗ K for each d ∈ M(B ⊗ K) with πB⊗K (d) ∈ C \ {0}.
−1
I.e., the ideal B ⊗ K is “purely large” in πB⊗K (C).
Does there exist a non-degenerate stable separable purely infinite simple C *-
subalgebra D of M(B ⊗ K) such that C is contained in πB⊗K (M(D)) ?
Is this even an equivalent criteria for absorption?

An answer is not obvious.

In case that C is non-unital then
one has to require here that the inclusion map
map C ,→ Qs (A) ‘‘dominates zero’’
in sense of Definition ??
see [310].
The closing question in [264]:
As pointed out above, any extension which is trivial in the nuclear sense has a
weakly nuclear splitting – i.e., is weakly nuclear. Is every weakly nuclear trivial
extension trivial in the nuclear sense?
Fit to better place and make it shorter.
It is for the proof of ‘‘Z-absorption implies (CFP)’’. Here CFP means
the corona factorisation property.

Remark 5.10.3. Suppose that P ∈ M(B ⊗K) is a projection with the property
that πB⊗K (P ) is properly infinite and full in Qs (B). Then there exists an isometry
T ∈ M(B ⊗ K) with T T ∗ = P .

Indeed: It says equivalently that there exists a contraction d ∈ M(B ⊗ K) with

X := 1 − d∗ P d ∈ (B ⊗ K)+ . On the other hand, exists an isometry R ∈ M(B ⊗ K)
with kR∗ XRk < 1/2 for given X ∈ B ⊗ K. Then R∗ D∗ P DR is a positive operator
with 1 ≤ 2R∗ D∗ P DR ≤ 2. Thus, there exists Q := (R∗ D∗ P DR)−1/2 ∈ M(B ⊗
√
K)+ that satisfies kQk ≤ 2 and 1 = S ∗ S = S ∗ P S for S := P DRQ. This
shows that P majorizes a range of an isometry and is – therefore – itself properly
infinite in M(B ⊗ K). The existence of an infinite repeat endomorphism δ∞ on
M(B ⊗ K) with tδ∞ (q)t∗ + s∗ qs unitarily equivalent to δ∞ (q) for all projections
q ∈ M(B ⊗ K) implies that 0 = [P ] ∈ K0 (M(B ⊗ K)) ∼ = {0}. Thus there exists an
∗
isometry T ∈ M(B ⊗ K) with T T = P .
 CHAPTER 6

Exact subalgebras of Purely infinite algebras

We give a proof of Theorem A in this chapter and study a special case of

Theorem K. The proof of Theorem A is simple and short, but it looks lengthy,
because we carry out every detail.
Let us give a short guide to a proof of Theorem A for the impatient readers:
We take Glimm’s theorem [324] (see [616, thm. 6.7.3]) to obtain that O2 contains
M2∞ as a quotient of a C *-subalgebra of O2 ( 1 ). Then we combine it with the re-
sult of [438] that every separable exact C *-algebra is a quotient of a C *-subalgebra
of the CAR-algebra M2∞ , in order to realise every separable exact C *-algebra as
a quotient of a C *-subalgebra of O2 . The embedding into O2 is then obtained
from the KK-triviality of O2 and from Corollary 5.8.12 of our generalized Weyl–
von-Neumann Theorem 5.6.2.
Certainly, for the first part of Theorem A, one could use also the natural em-
bedding of M2∞ into O2 instead of Glimm’s theorem. But then one gets not the
additional result in Remark 6.2.2, and one has to manage with additional tools that
the below considered extension 0 → O2 ⊗ K → E → A → 0 (with E ⊂ O2 ) becomes
essential. But one could here also first menage to find a sub-quotient C *-algebra of
s1 (s1 ) ∗ O2 s1 (s1 )∗ ∼
= O2 that is isomorphic to the given separable exact C *-algebra
A ( ... that can be considered as unital (!) ...) ... If the generated hereditary kernel
C *-subalgebra of the constructed extension is not stable, then the construction is
ready, because all non-zero hereditary C *-subalgebras of O2 are isomorphic to O2
or are isomorphic to O2 ⊗ K by Lemma 6.1.1(ii). Since the extension algebra is
contained in the nuclear C *-algebra O2 , it is an exact C *-algebra and as an exten-
sion it is locally semi-split. This allows to derive that we have only two cases: A
is automatically C *-subalgebra of O2 , or we get an extension B of A by O2 ⊗ K
that is contained up to isomorphisms in O2 . The KK-triviality and Ext-triviality
of O2 ⊗ K implies that the extension is split, as shown below ???ref???
Our proof of Theorem K follows a similar route, but the technical ingredients
of the proof of Theorem A have to be improved considerably. Therefore we prove
in Section 3 the intermediate result Corollary 6.3.2 (of Theorem 6.3.1), where we
assume, in addition to the assumptions on B in Theorem K, that

1Alternatively to the general theorem of Glimm we could alternatively use for the first part

of the proof the very different natural copy of M2∞ inside O2 that is just the closed linear span
of all elements T S∗ where T and S are all products of the generating isometries s1 , s2 of the
C *-algebra O2 := C ∗ (s1 , s2 ) with relations (s1 ) ∗ s1 = 1, (s2 ) ∗ s2 = 1 , (s2 ) ∗ s1 = 0 and
s1 (s1 ) ∗ +s2 (s2 )∗ = 1.

781
782 6. EXACT SUBALGEBRAS OF PURELY INFINITE ALGEBRAS

????
B is separable and that B ⊗ O2 contains a regular Abelian C *-subalgebra
(cf. Definition 1.2.9).
Decide:
Below transfer from old Chp. 5 more useful?
The generalized Weyl–von Neumann Theorem 5.6.2 will be an ingredient of
the proofs of Theorems A and of an important special case of Theorem M ????
in Chapter 6, under the additional assumption that the algebra B in Theorem
M has residual nuclear separation (cf. Definition 1.2.3), i.e., that the m.o.c. cone
Crn ⊆ CPnuc (B, B) is separating for B and its ideals. This extra assumption will
be removed finally in Chapter 12 that uses results from Chapters 7-11.
We assume in Chapter 6, – in addition to the assumptions of Theorem M
–, that the universal weakly residually nuclear C *-morphism Hrn : B → M(B)
from B into its multiplier algebra M(B) is non-degenerate for the in Theorem ??
considered stable separable B and that there exists a C *-morphism h : A → M(B)
that defines the action of Prim(B) on A. This additional assumptions will be
completely removed in Chapter 12 by using results of Chapters 3 and 7 -11 .
Its generalization to weakly residually nuclear maps – and, more generally, to
maps in a given m.o.c. cone C ⊆ CP(A, B) – will be used in Chapters 6 and 12 for
the proof of Theorem K.
This additional assumptions will be removed both in Chapter 12.
The here given proof of Theorem 6.3.1 gives also an alternative proof of Theo-
rem A that does not use [438], but requires to generalise some results of [437], see
Section 19 of Appendix A.

1. Normalizers of hereditary subalgebras (Part 1)

Recall that a *-subalgebra B of a C *-algebra A is an essential subalgebra

of A if it has no non-zero left annihilators in A, i.e., if aB = {0} implies a = 0 for
all a in A. E.g., always a dense *-subalgebra B of a C *-algebra C is an essential
subalgebra of the multiplier A := M(C) of C.
Next red: Is here a good place?
Our Proposition 6.2.1 implies that the classification of (the positions in O2 of)
the essential hereditary C *-subalgebras D of O2 – together with a certain semi-
splitting property – would in particular create invariants for the classification of all
nuclear separable unital C *-algebras. Unfortunately, up to now there is no method
to link the points of this “classifying space” with known invariants of K-theoretic
nature.

Lemma 6.1.1. Suppose that D ⊆ J ⊆ E ⊆ B are C*-subalgebras of a C*-algebra

B, and that J is an ideal of E.
 1. NORMALIZERS OF HEREDITARY SUBALGEBRAS (PART 1) 783

(i) If D is essential in B and J 6= E, then J is an essential ideal of E and is

an essential hereditary C*-subalgebra of B.
In particular, the C*-algebra J can not have a unit element.
(ii) If D ⊂ O2 is an essential hereditary C*-subalgebra of O2 and D 6= O2 ,
then D is isomorphic to O2 ⊗ K.

Proof. (i): Every annihilator of J in B ⊃ E is in an annihilator of D in B.

But the annihilators of D in B are zero.
Suppose J has a unit element e ∈ J, i.e., e = e∗ e and ex = xe = x for all
x ∈ J. The elements of (1 − e)B(1 − e) are two-sided annihilators of J. Thus
k(1 − e)b∗ b(1 − e)k = kb − bek2 = 0 for every b ∈ B, and e is the unit of B. Since
J is an ideal of E, this contradicts our assumption J 6= E.
(ii): All this is given by the equivalence of its Parts (ix) and (i) of Proposition
2.2.1, if we use that O2 is simple and purely infinite in the sense of J. Cuntz.
An alternative proof is the following: Let E := D + C · 1 ⊂ O2 . If D is an
essential hereditary C *-subalgebra of O2 and D 6= O2 , then D 6= E and D has not
a unit element.
The non-zero hereditary C *-subalgebra D of O2 , is simple, purely infinite and
separable. It follows that D is either unital or stable by Corollary 5.5.4 (Zhang
dichotomy). Thus, D ∼ = D⊗K ∼ = O2 ⊗ K by Corollary 5.5.6 (Brown stable isomor-
phism theorem, [107]). 

Some of the, in Remarks 6.1.2 mentioned, properties are also at

other places discussed. Check this!!!

Remarks 6.1.2. The study of quotients of C *-subalgebras of a given C *-

algebra C leads to the study of hereditary C *-subalgebras D of C and of its nor-
malizer algebra N (D) ⊂ C, which is defined as N (D) := {b ∈ C : bD + Db ⊂ D}
(and is usually very different from the multiplier algebra M(D) of D).
(1) The set of products CD is a closed left ideal of C and the sum CD + DC is
a closed linear subspace of C ( 2 ). We define C//D := C/(CD + DC) (considered
as an operator subspace of the W*-algebra C ∗∗ , see below).
We consider N (D)/D, its natural maps into C//D and into M(D)/D, and
the behavior of this maps on the system of closed ideals of C. They lead to the
guiding principles for all proofs in this chapter (except some additional arguments
on approximation and inductive limits in the Ψ-residually nuclear case).
(2) A hereditary C *-subalgebra D of C is a closed ideal of N (D) by definition
of N (D), and, therefore, there is a natural C *-morphism ρ : N (D) → M(D) from

2 The closed left ideal L := span(CD) is identical with the set of products CD by Cohen

factorization theorem, – or simply observe that every element a of a closed left ideal L can be
expressed as a = cd for c ∈ C suitable and d := (a∗ a)1/4 in D = L∗ ∩ L –. The set CD + DC is a
closed linear subspace of C, because it is the sum of a closed left ideal L and a closed right ideal
R, and because C ∗∗ p + pC ∗∗ is σ(C ∗∗ , C ∗ )-closed in C ∗∗ .
784 6. EXACT SUBALGEBRAS OF PURELY INFINITE ALGEBRAS

N (D) into the multiplier algebra M(D) of D. Clearly, the kernel of ρ : N (D) →
M(D) is the (two-sided) annihilator Ann(D) := {c ∈ C : cD + Dc = 0} ⊂ N (D)
of D in C. From the definition, one sees that Ann(D) is a closed ideal of N (D),
and that the *-epimorphism N (D) → N (D)/D is faithful on Ann(D).
(3) A hereditary C *-subalgebra D ⊂ C is essential in C, if and only if,
Ann(D) = {0}. Thus, the natural *-morphism N (D) → M(D) is faithful, if
and only if, D is essential in C. The natural morphism N (D) → M(D) is not
necessarily an epimorphism. Different essential hereditary C *-subalgebras D and
D0 have in general non-isomorphic N (D)/D and N (D0 )/D0 , even if D and D0 are
isomorphic.
The corona construction Q(D) := M(D)/D defines an isomorphism invariant
for all algebras D.

(4) Let F ⊂ C a C *-subalgebra and J a closed ideal of F , then the set D :=

JCJ is a hereditary C *-subalgebra of C such that F ⊂ N (D) and F ∩ D = J,
because an approximate unit of J is also an approximate unit of D. Therefore,
there is a natural *-monomorphism from F/J into N (D)/D and the unit of the
weak closure of J in the second conjugate C ∗∗ of C is also the unit of the weak
closure of D. Moreover, F + D is closed, because F D + DF ⊂ D, cf. [767, Vol. I,
chap. I,sec. 8, exercise 2].
(5) Let D be a hereditary C *-subalgebra of C. Then the unit element pD of the
weak closure of D in C ∗∗ is called the support projection of D. It is a so-called
open projection, and qD := 1 − pD is a closed projection.
(6) There are well-known one-to-one relationships between hereditary C *-
subalgebras, open projections, closed left ideals, weakly closed faces of the state
space of C, weakly closed right invariant subspaces of C ∗ , and the (with respect to
C) closed projections in C ∗∗ .
Indeed, let S(C) denotes the state space of C, and let X 7→ X o 7→ X oo the
passage to the polar and bipolar of a set, then the one-to-one relations are given
by the maps D 7→ L := CD, L 7→ D = L∗ ∩ L, D 7→ KD = Do ∩ S(C),
L 7→ Lo 7→ X := Lo ∩ S(C) = Do , D 7→ pD , p 7→ Lp := C ∩ (C ∗∗ p),
K 7→ LK := {a ∈ C : f (a∗ a) = 0 ∀f ∈ K}, p 7→ q = 1 − p, q 7→ C ∗ q,
X 7→ KX = S(C) ∩ X, q 7→ Dq := {a ∈ C : q(a∗ a + aa∗ )q = 0}.
We refer the reader to [767, Vol. I, chap. III, def. 6.19, cor. 6.20], and to
[616, prop. 3.11.9, thms. 3.10.7, 3.11.10]. (For our particular needs, the reader can
simplify the there given proofs if he uses instead 2 × 2-matrix arguments.)

(7) It is useful to consider also the quotient XD := C//D := C/(CD+DC) of C

by CD +DC (which is a closed linear subspace of C ), and let πCD+DC : C → C//D
denote the natural epimorphism. The space XD with its natural matrix order
structure and matrix norms ( 3 ) becomes a (special kind of) C *-system (defined

3It is given on the vector space M (C/(CD + DC)) by the natural isomorphism with the
n
quotient Banach space norms Mn (C) → Mn (C)/Mn (CD + DC).
 1. NORMALIZERS OF HEREDITARY SUBALGEBRAS (PART 1) 785

more generally in [437]). If C is unital, then the C *-system XD is unital with

matrix order unit πCD+DC (1). The bi-dual operator system of XD is nothing
else qD C ∗∗ qD (under natural identification given by the restriction of πCD+DC
∗∗
to
∗∗
qD C qD ), cf. [437]. The unital C *-systems are non-commutative generalizations
of the spaces of continuous affine functions on a Choquet simplex, cf. [437], [472].
The natural isomorphism qD C ∗∗ qD → XD ∗∗
defines a completely isometric iso-
morphism λ : qD cqD → c + (CD + DC) from the subspace qD CqD of C ∗∗ onto
XD , Therefore, the kernel of the completely positive map a ∈ C 7→ qD aqD is just
CD + DC.
It is easy to see that N (D) = {qD }0 ∩ C, and therefore, that there is a natural
unital C *-morphism N (D) → N (D)qD ⊂ qD CqD . By definition of pD = 1 − qD ,
the kernel of this *-epimorphism onto qD N (D) is just D. We get a natural *-
isomorphism τ (b + D) := bqD from N (D)/D onto qD N (D) ⊂ qD CqD , and we have
for a ∈ N (D):

dist(a, C ∗∗ pD + pD C ∗∗ ) = dist(a, CD + DC) = ka + Dk = kqD aqD k .

(8) We use for the proof of Proposition 6.2.1 results from the paper [438] which
rely on the following (non-trivial) fact [437, thm. 1.4(iii)] on normalizer algebras:
If C is unital, then the natural unital C*-morphism

0
πD : N (D) 3 a 7→ aqD = qD aqD ∈ qD C ∗∗ qD

defines a completely isometric and completely positive isomorphism from N (D)/D

∗∗ ∼
onto the (two-sided) multiplier algebra M(XD ) ⊂ XD of XD in XD = qD C ∗∗ qD
(the latter naturally identified by the map λ from (7)).

(9) An obvious consequence of (8) is the following:

Suppose that F is a C *-subalgebra of N (D) ⊂ {pD }0 ∩ C = {qD }0 ∩ C and that
the image F qD of the map c → qD cqD is σ(C ∗∗ , C ∗ )-dense in the image qD CqD of
C (which happens e.g. if qD CqD = qD F ). Then, πCD+DC (F ) = C/(CD + DC),
because it is a C *-subalgebra of M(C/(CD + DC))). In particular F + D = N (D)
and C = N (D) + DC + CD.
Notice that, conversely, qD CqD = qD N (D) if C = N (D) + CD + DC.
From the above collected results, we get that the following properties (i)-(v) of
hereditary C *-subalgebras D of a unital C *-algebra C are equivalent:

(i) The C *-system C//D is unitally and isometrically isomorphic to a C *-

algebra.
(ii) M(C//D) = C//D (i.e., C//D is a C *-subalgebra of qD C ∗∗ qD ).
(iii) C = N (D) + CD + DC.
(iv) There is a C *-algebra A and a completely positive contraction T : C → A
such that T (D) = {0}, T | N (D) is a *-epimorphism from N (D) onto
A with kernel D, and that there is a projection q ∈ C ∗∗ and an W*-
isomorphism S from qC ∗∗ q onto A∗∗ with T ∗∗ (c) = S(qcq) for all c ∈ C.
786 6. EXACT SUBALGEBRAS OF PURELY INFINITE ALGEBRAS

(v) There is a completely positive map T : C → N (D)/D with T | N (D) = πD

and ker(T ) = CD + DC.

(10) Let T : C → G be a unital completely positive map. We define the mul-

tiplicative domain MT ⊂ C of T as the set of d ∈ C with T (db) = T (d)T (b)
and T (bd) = T (b)T (d) for b ∈ C. M.D. Choi’s generalized Kadison inequality
T (a)∗ T (a) ≤ T (a∗ a) shows that d ∈ MT , if and only if, T (d∗ d) = T (d)∗ T (d) and
T (dd∗ ) = T (d)T (d)∗ . Unfortunately, the multiplicative domain (resp. the kernel)
of the unital completely positive map T ∗∗ : C ∗∗ → G∗∗ is not necessarily contained
in the weak closure of the multiplicative domain (resp. the kernel) of T , cf. Part
(11).
The multiplicative domain of the unital completely positive map
TD : a ∈ C 7→ qD aqD ∈ qD C ∗∗ qD
is just N (D) = {qD }0 ∩ C and N (D)∗∗ 6= {qD }0 ∩ C ∗∗ .
(11) The above remarks on the quotient spaces XD = C//D show that the
second conjugate (πCD+DC )∗∗ of πCD+DC : x ∈ C 7→ x + (CD + DC) ∈ C//D is
just the normalization a ∈ C ∗∗ 7→ qD aqD of TD in Part (10). The multiplicative
domain pD C ∗∗ pD + qD C ∗∗ qD of (πCD+DC )∗∗ is in general much bigger than the
weak closure of N (D) in C ∗∗ . But the kernel of (πCD+DC )∗∗ is just pD C ∗∗ +C ∗∗ pD ,
which is the weak closure of the kernel DC + CD of TD and of πD .
(12) Let X := XD := C//D ⊂ qD C ∗∗ qD , and let η1 : X → X ∗∗ ∼ = qD C ∗∗ qD ,
η2 : X ∗ → X ∗∗∗ and η3 : X ∗∗ → X ∗∗∗∗ the natural inclusions. Then (η1 )∗∗ : X ∗∗ →
X ∗∗∗∗ is a normal unital completely isometric map, (η2 )∗ : X ∗∗∗∗ → X ∗∗ is a normal
*-epimorphism, and η3 is a (non-normal) *-monomorphism. The map E := (η1 )∗∗ ◦
(η2 )∗ is a normal u.c.p. map on the fourth conjugate W*-algebra of X with E 2 = E.
We say that y ∈ X ∗∗ ∼
= qD C ∗∗ qD is a (two-sided) multiplier of X if yx, xy ∈ X
for all x ∈ X, where the multiplication is given by the W*-algebra structure on
X ∗∗ . Let M(X) ⊂ X ∗∗ denote the set of all multipliers of X. Then, clearly, M(X)
is a unital C *-subalgebra of X ∗∗ , and is contained in X, because qD = 1X ∗∗ ∈ X.
One can see by a separation argument, that y ∈ M(X), if and only if, η3 (y) is
in the multiplicative domain of E (see [437] for details). It follows from Remark
(10) that y ∈ M(X), if and only if, y, y ∗ y and yy ∗ are all in X.
(13) By [437, cor. 1.5], a ∈ N (D) 7→ aqD ∈ qD CqD ∼
= C//D is a *-epimorphism
from N (D) onto M(C//D) with kernel equal to D. It implies together with Remark
(12) the following observation:
Let x ∈ qD C ∗∗ qD . There is a ∈ N (D) with aqD = x if and only if, x, x∗ x, xx∗ ∈
C//D, if and only if, x ∈ M(C//D).
(14) If C is not unital, then we can pass to the unitization C
e of C, and get
immediately the following reformulation of (13):
Let x ∈ qD C ∗∗ qD . There is a ∈ N (D) with aqD = x if and only if, x, x∗ x, xx∗ ∈
C//D, if and only if, x ∈ C//D ∩ M(C//D).
 1. NORMALIZERS OF HEREDITARY SUBALGEBRAS (PART 1) 787

In particular, N (D)/D ∼
= (C//D) ∩ M(C//D) via identifications of a + D, apD
and a + (DC + CD) for a ∈ N (D).

Lemma 6.1.3. Suppose that E ⊂ B is a C*-subalgebra of a nuclear C*-algebra

B, D is a σ-unital hereditary C*-subalgebra of B, D ⊂ E, and that D is an ideal
of E.
Then every completely positive contraction ϕ from a separable unital C*-algebra
A into E/D has a completely positive and contractive lifting V : A → E, i.e.,
πD ◦ V = ϕ.
If, in addition, B and A are unital, 1B ∈ E, and ϕ : A → E/D is unital, then
a unital nuclear c.p. map V : A → E with πD ◦ V = ϕ exists.

Notice, that a lift V : A → E of an isomorphism ϕ : A → E/D from A onto

E/D can’t be nuclear if A is not a nuclear C *-algebra. But V : A → B ⊃ E is a
nuclear c.p. map from A into B, because B is nuclear.

Proof. If A is separable and unital, B is unital with 1B ∈ E and ϕ : A → E/D

is unital, then we can replace A by the separable unital C *-subalgebra F of E/D
generated by ϕ(A), ϕ by the identity map of this new algebra F , and finally E by
−1
G := πD (F ). The extension given by the exact sequence

0→D→G→F →0

has a splitting unital c.p. map T : F → E ⊆ B for πD by [238, thm. B, prop. 4.3].
It uses that D is nuclear as a hereditary C *-subalgebra of the nuclear algebra B,
i.e., D fulfills assumption (1) of [238, thm.B]. The C *-subalgebra G ⊆ B inherits
property (C) from the nuclear B by [238, prop. 5.3] (citing [35, thms. 3.2, 3.3]).
Property (C00 ) – which is equivalent to local reflexivity in the operator space
sense –, follows from property (C) and allows to verify the assumption (2) of [238,
thm.B], with help of [238, prop. 5.3(3)] and [238, prop. 5.5].
(Notice that the properties (C) and (C0 ) both are equivalent to exactness by
[438], because every separable exact C *-algebras is a quotient of some C *-sub-
algebras of the CAR algebra M2∞ .)
Check next again: Unit OK?
If one of B, E, A or ϕ are not unital, or if 1B 6∈ E, then we can adjoin to all
algebras (except D) an outer unit 1, and extend ϕ to a unital c.p. map

ϕ e → E/D
e: A e ∼ ],
= E/D
by ϕ(a+t1)
e e→E
:= ϕ(a)+t1 . There is a unital u.c.p. map W : A e with πD ◦W = ϕ. e
Then W (A) ⊂ E and the map V := W |A is a completely positive contraction from
A into E with πD ◦ V = ϕ. 

Remark 6.1.4. Other methods in Section 3 that are different from those of
[238] and are more near to the study of the convex cone of c.p. maps V : A → E
788 6. EXACT SUBALGEBRAS OF PURELY INFINITE ALGEBRAS

like methods of Arveson in the proof of [43, thm. 6]. The results are more general
lifting results than that stated in Lemma 6.1.3, e.g.
Suppose that B is separable and that B//D is a nuclear C*-space (i.e., that
qD B ∗∗ qD is an injective von-Neumann algebra), that D is an essential hereditary
C *-subalgebra of B and that D ⊆ E ⊆ N (D) ⊆ B.
Then for every separable C *-algebra A and every c.p. contraction ϕ : A → E/D
there exist a c.p. contraction V : A → E with πD ◦ V = ϕ such that V : A → B is
nuclear (but is not necessarily nuclear as a map from A to E).
If A and B are unital, 1B ∈ E and ϕ is unital, then one can manage that V is
unital.
(In fact, the reduction to [43, thm. 6] is done be showing that there exists a
(not necessarily unique) c.p. contraction T : B//D := B/(BD + DB) → M(D)/D
such that T | N (D)/D is the natural C *-morphism from N (D)/D into B//D.)

Lemma 6.1.5. Suppose that F is a C*-subalgebra of a C*-algebra C, and that

q ∈ F 0 ∩ C ∗∗ is a projection that is closed in C ∗∗ and satisfies F q = qCq.
Let D denote the hereditary C*-subalgebra of C with open support projection
1 − q. Then:

(i) N (D) = D + F .
(ii) Tq : c ∈ C 7→ qcq ∈ C ∗∗ is a c.p. contraction and has kernel equal to
DC + CD .
(iii) The multiplicative domain of Tq is given by {c ∈ C ; cq = qc }, which is
the same as N (D).
(iv) There are natural isomorphisms

Fq ∼
= F/(F ∩ D) ∼
= N (D)/D ∼
= C/(DC + CD)
that are given by

[Tq ] : πDC+CD (c) 7→ qcq ∈ F q = N (D)q ∼

= N (D)/D .
(v) C = N (D) + DC + CD.

Proof. to be filled in ?? 

Lemma 6.1.6. Suppose that D1 ⊂ C is a hereditary C*-subalgebra of a C*-

algebra C such that C = N (D1 )+D1 C+CD1 . Let ρ : N (D1 ) → B a *-epimorphism
from N (D1 ) onto a C*-algebra B with kernel equal to D1 , and let D2 ⊂ B a
hereditary C*-subalgebra of B.
Denote by J the hereditary C*-subalgebra J := ρ−1 (D2 ) of N (D1 ), and define
the hereditary C*-subalgebra D ⊂ C by D := JCJ.

(i) D1 ⊂ D, and N (D) = ρ−1 (N (D2 )) + D.

(ii) There is a unique positive contraction T : C → B that extends ρ. T is
completely positive, has kernel ker(T ) = D1 C + CD1 and multiplicative
domain N (D1 ).
 2. THE PROOF OF THEOREM A 789

(iii) T (D) = D2 , T (DC + CD) = D2 B + BD2 and T (N (D)) = N (D2 ).

(iv) The c.p. map T | N (D) defines an isomorphism of C*-algebras

η : N (D)/D 7→ N (D2 )/D2 ,

with η(πD (c)) = πD1 (T1 (c)) for c ∈ N (D).

In particular, η(c + D) = ρ(c) + D1 for c ∈ ρ−1 (N (D2 )).
(v) The class-map [T ](DC+CD) is a completely positive and completely isomet-
ric isomorphism from the C*-system C//D onto B//D2 .

Proof. to be filled in ??
Let F ⊂ N (D1 ) ⊂ C denote the inverse image ρ−1 (N (D2 )) =: F of N (D2 )
under the epimorphism ρ from N (D1 ) onto B.
The inverse image J := ρ−1 (D2 ) of D2 is the kernel of the epimorphism πD2 ◦
ρ : F → N (D2 )/D2 . In particular, J is an ideal of F , J contains the kernel D1
of ρ, ρ(J) = D2 . Thus, there is an isomorphism µ : F/J → N (D2 )/D2 with
µ ◦ πJ = πD2 ◦ ρ.
Now we consider the hereditary C *-subalgebra D := JCJ = JCJ of C . Then
D is a hereditary C *-subalgebra of C, and D1 ⊂ D. Since the ideal J of F contains
an approximate unit of D, we get that the hereditary C *-subalgebra F ∩D of F must
be equal to the ideal J of F , and that F ⊂ N (D). In particular, F/J is naturally
isomorphic to a C *-subalgebra of N (D)/D. The above defined natural unital
∼
monomorphism F/J → N (D)/D and the isomorphism µ−1 : N (D2 )/D2 → F/J
define together a unital *-monomorphism

ξ : N (D2 )/D2 ,→ N (D)/D ,

such that ξ(πD2 (ρ(a)) = πD (a) for a ∈ F . 

2. The Proof of Theorem A

Proposition 6.2.1. Suppose that A is a unital separable exact C*-algebra.

Then there exist a unital C*-subalgebra E of O2 , a closed ideal D of E, and a
*-isomorphism ϕ from A onto E/D, such that

(i) D is an essential hereditary C*-subalgebra of O2 ;

(ii) ϕ has a unital completely positive lifting V : A → E, i.e., ϕ = πD ◦ V ,
(iii) D ∼
= O2 ⊗ K .

If A is nuclear, then the subalgebras D ⊂ E ⊂ O2 , and the isomorphism

∼
ϕ : A → E/D can be chosen such that the epimorphism ϕ−1 ◦ πD : E → A extends
to a unital completely positive map T : O2 → A with kernel D · O2 + O2 · D.

Proof. Let B denote the CAR-algebra M2∞ and let C := O2 . We define

∼
D ⊂ E ⊂ C and ϕ : A → E/D and show later that they satisfy (i)–(iii).
First we consider the case of an exact separable A that is not necessarily nuclear:
790 6. EXACT SUBALGEBRAS OF PURELY INFINITE ALGEBRAS

Since C is not a C *-algebra of type I, we can apply the improved variant [616,
thm. 6.7.3] of Glimm’s theorem: There exist a unital subalgebra F1 of C and a
closed projection q1 in C ∗∗ such that q1 commutes with F1 , q1 Cq1 = q1 F1 and that
the C *-subalgebra q1 Cq1 of q1 C ∗∗ q1 is isomorphic to the CAR-algebra B := M2∞ .
Let p1 := 1 − q1 the open complement of q1 , and let D1 denote the hereditary
C *-subalgebra of B with open support projection p1 = pD1 , cf. Remark 6.1.2(5),
∼
i.e., (D1 )+ = {c ∈ C+ ; q1 cq1 = 0} . Furthermore, let h1 : q1 Cq1 → B a *-
isomorphism from q1 Cq1 ∼ = q1 F1 onto B, and define T1 : C → B by T1 (a) :=
h1 (q1 aq1 ).
By Lemma 6.1.5, T1 is a unital completely positive map, and the kernel of T1
is CD1 + D1 C. Moreover, N (D1 ) = F1 + D1 , and N (D1 ) is just the multiplicative
domain of T1 . It implies that C = N (D1 ) + CD1 + D1 C.
Let ρ := T1 | N (D1 ). Then ρ : N (D1 ) → B is a *-epimorphism from N (D1 )
onto B with kernel D1 = N (D1 ) ∩ (D1 C + CD1 ).

By [438, cor. 1.3, 1.4, thm. 4.1], for every separable unital exact C *-algebra
A, there exists a hereditary C *-subalgebra D2 ⊂ B and a unital *-monomorphism
ψ : A → N (D2 )/D2 .

Since D2 is an ideal of N (D2 ), the inverse image J := ρ−1 (D2 ) is an ideal of

the C *-subalgebra ρ−1 (N (D2 )) of N (D1 ) . We let D := JCJ.
The equation C = N (D1 ) + CD1 + D1 C and the Lemma 6.1.6 on iterated
normalizers show that that T1 : C → B is the unique completely positive extension
of the epimorphism ρ : N (D1 ) → B, that T1 (D) = D2 , that T1 (N (D)) = N (D2 )
and that T1 | N (D) defines a C *-isomorphism η from N (D)/D onto N (D2 )/D2
with η(πD (c)) = πD1 (T1 (c)) for c ∈ N (D).
We define a unital *-monomorphism ϕ : A → N (D)/D by ϕ := η −1 ◦ ψ and let

−1
E := πD (ϕ(A)) ⊂ N (D) ⊂ O2

be the inverse image of ϕ(A) under the quotient map πD : N (D) → N (D)/D.
Thus, we get a hereditary C*-subalgebra D of C := O2 , a C*-subalgebra E ⊂
∼
N (D) ⊂ O2 with D1 ⊂ D ⊂ E ⊂ N (D), and a unital *-isomorphism ϕ : A → E/D.

(i,iii): We use Lemma 6.1.1: Since D1 ⊂ D and D 6= O2 , it suffices to show

that D1 is essential in O2 , i.e., that Ann(D1 ) = {0}. Recall, that Ann(D1 ) is
a closed ideal of N (D1 ) and is a hereditary C *-subalgebra of C := O2 . The
epimorphism ρ : N (D1 ) → B := M2∞ has kernel D1 . Thus, ρ| Ann(D1 ) is faith-
ful, and ρ(Ann(D1 )) is a closed ideal of M2∞ . Since every non-zero hereditary
C *-subalgebra of O2 is stably infinite and since M2∞ is stably finite, we get
ρ(Ann(D1 )) = {0} and Ann(D1 ) = {0}.
(ii): By Lemma 6.1.3, there is a u.c.p. map V : A → E with πD ◦ V = ϕ.
This completes the proof for general exact A.
 2. THE PROOF OF THEOREM A 791

We consider now the special case of nuclear A:

If A is nuclear, then [438, cor. 1.5] says that the hereditary C *-subalgebra D2 ⊂
B := M2∞ and ψ : A → N (D2 )/D2 can be chosen such that ψ is an isomorphism
from A onto N (D2 )/D2 and, moreover, that B = N (D2 ) + D2 B + BD2 , i.e.,
ψ(A) = N (D2 )/D2 ∼ = B//D2 .
As in the case of (not-necessarily nuclear) exact A we define D ⊂ E ⊂ C, and
ϕ : A → E/D :
Let D := JCJ for J := ρ−1 (D2 ) ⊃ D1 with ρ := T1 | N (D1 ) (defined as in the
case of exact A), and use the isomorphism η : N (D)/D → N (D2 )/D2 to define
ϕ := η −1 ◦ ψ. Then ϕ : A → N (D)/D is an isomorphism of A onto N (D)/D. Thus
−1
E := πD (ϕ(A)) = N (D).
We apply Lemma 6.1.6 to (B, D2 , ψ −1 ◦ πD2 , A, 0), in place of the there
considered general system (C, D1 , ρ, B, D2 ), and get the unique u.c.p. extension
T2 : B → A of the *-epimorphism
ρ1 := ψ −1 ◦ πD2 : N (D2 ) → A ,
and T2 has kernel D2 B + BD2 and multiplicative domain N (D2 ).
Since C = N (D1 ) + CD1 + D1 C, we get from Lemma 6.1.6, that
T1 (DC + CD) = D2 B + BD2 .
It implies that the u.c.p. map T := T2 ◦ T1 : C → A has kernel DC + CD.
Let c ∈ N (D) ⊂ C. Then V1 (c) ∈ N (D2 ) and πD2 (V1 (c)) = η(πD (c)) by
Lemma 6.1.6. Thus,
V (c) = ρ1 (V1 (c)) = ϕ−1 (πD (c)) for c ∈ N (D) .
Thus, the u.c.p. map V : C → A extends the *-epimorphism from E = N (D) onto
A, and is given by ϕ−1 ◦ πD . 

We have seen so far, that there exists, for every separable unital exact C *-
algebra A, an essential semi-split exact sequence
0→D→E→A→0
where E is a unital C *-subalgebra of O2 , and where D ∼
= O2 ⊗ K is essential in O2 .
This extension splits unital by Corollary 5.8.12. The unital splitting C *-morphism
is a monomorphism from A onto a unital subalgebra of E ⊂ O2 .
The desired conditional expectation onto the image of A in O2 is then just the
composition of T : O2 → A with the splitting C *-morphism. More details are given
below in Remark 6.2.2.

Proof of Theorem A.. Ad(i): Let A be a separable exact C *-algebra. If

A is not unital then the unitization of A is again exact, as our A ⊗ (·) -exactness
definition of exactness of A in Chapter 3 immediately implies (e.g. use the 3×3-
lemma of category theory). Thus we can restrict our considerations to the case of
unital exact C *-algebras A.
792 6. EXACT SUBALGEBRAS OF PURELY INFINITE ALGEBRAS

The unital semi-split essential extension

λ
0 −→ D −→ E −→ A −→ 0

with λ := ϕ−1 ◦ πD from Proposition 6.2.1 has a unital split morphism by Corollary
5.8.12, because D ∼
= O2 ⊗ K and E is exact.
The unital split C *-morphism ψ : A → E ⊆ O2 with λ ◦ ψ = idA is the desired
unital *-monomorphism A into O2 .
Conversely, O2 is nuclear ([169]), nuclear C *-algebras are exact and C *-
subalgebras of exact C *-algebras are again exact ([432, prop. 7.1(i)] or [810,
2.5.1, 2.5.2], or see Remark 3.1.2(ii) and (iii)).
Ad(ii): If A is nuclear and T : O2 → A is the completely positive map from
Proposition 6.2.1, then P := ψ ◦ T is a conditional expectation from O2 onto
ψ(A) ⊂ E.
Indeed: P is a completely positive contraction that maps O2 into ψ(A). The
maps T , ψ and λ satisfy T (O2 ) = A, ψ(A) ⊂ E, T |E = λ, and λ ◦ ψ = idA .
Therefore, T ◦ ψ = idA , P 2 = P and

P (ψ(a)) = ψ(T (ψ(a))) = ψ(λ(ψ(a)) = ψ(a) ∀ a ∈ A.

Conversely, let P be a conditional expectation from a nuclear C *-algebra B onto a

C *-subalgebra A ⊂ B and ηA : A ,→ B the inclusion map. Then idA = P ◦ idB ◦ηA
is nuclear because idB is nuclear and nuclearity is preserved under composition with
other completely positive maps. 

Remark 6.2.2. The conditional expectation P := ψ ◦ T : O2 → ψ(A) that has

been defined in the proof of Part (ii) of Theorem A(ii) is an extreme point of the
convex set of all linear contractions from O2 into O2 .

Proof. Let M be a von Neumann algebra, q ∈ M a projection, and µ : M →

M a unital contraction such that µ(1 − q) = 0 (i.e., µ(q) = 1), qµ(qaq)q = qaq for
a ∈ M and µ|qM q is a C *-morphism.
Then µ is an extreme point of the convex set of linear contractions from M
into M . Indeed:
We use that each unitary U ∈ M is an extreme point of the closed unit-ball of
M , (cf. [704, thm. 1.6.4]). If T, S : M → M are contractions and θ ∈ (0, 1) with
θT + (1 − θ)S = µ, then θT (1) + (1 − θ)S(1) = 1. It implies that T (1) = 1 = S(1).
Hence T and S are positive unital maps. It follows T (1 − q) ≥ 0, S(1 − q) ≥ 0
and 0 = µ(1 − q) = θT (1 − q) + (1 − θ)S(1 − q), thus T (1 − q) = 0 = S(1 − q)
and T (a) = T (qaq), S(a) = S(qaq) for a ∈ M . For unitaries u of qM q we get that
U := µ(u) is unitary and U = θT (u) + (1 − θ)S(u). Thus µ(u) = T (u) = S(u) for
all unitaries of qM q. Together it implies µ = T = S.
The P : O2 → ψ(A) annihilates D and, therefore, M := O2 ∗∗ , µ := P ∗∗ and
q = 1 − pD satisfy the above assumptions on M , µ and q, where pD denotes the
 2. THE PROOF OF THEOREM A 793

support projection of D. Since P ∗∗ is an extreme point of the contractions on M ,

P must be an extreme point of the contractions on O2 . 

next Rem’s have overlapping statements

Remarks 6.2.3. Let ϕj : A → O2 (j = 1, 2) unital *-monomorphisms, then

they are unitarily homotopic by Theorem B. In particular, they are approximately
unitarily equivalent.
By Remark 6.2.2, one finds always for separable unital nuclear A at least one
unital embedding ι : A ,→ O2 such that there is an extremal conditional expectation
P from O2 onto ι(A).
For any other embedding κ : A ,→ O2 there is a norm-continuous path
t ∈ [0, ∞) 7→ U (t) into the unitary group of O2 , such that U (0) = 1 and
limt→∞ kU (t)∗ ι(a)U (t) − κ(a)k for all a ∈ A. One can use this to show that
for any (more general type of) unital *-monomorphism κ : A ,→ O2 there exists
a norm-continuous path t ∈ [0, ∞) 7→ V (t) into the isometries in O2 such that
limt→∞ kV (t)∗ κ(a)V (t) − κ(a)k for all a ∈ A and limt→∞ dist(V (t)∗ bV (t), κ(A))
for all b ∈ O2 . (This could replace the existence of conditional expectations in an
approximate sense.)
There can not be stronger results, even under very reasonable additional con-
ditions, because there exist a unital nuclear C*-algebra A unital *-monomorphisms
ϕ1 and ϕ2 from a unital nuclear C*-algebra A into O2 , such that there does not
exist an automorphism γ of O2 ⊗ O2 with γ(ϕ(a) ⊗ 1) = ψ(a) ⊗ 1 for a ∈ A.
Let us consider two nuclear unital C *-algebras A ⊂ B such that there is no
conditional expectation from B onto A:
If h : B → O2 is any unital *-monomorphism, and if ψ : A → O2 and P : O2 →
ψ(A) are as in Remark 6.2.2, then there can not exist an automorphism γ of O2 ⊗O2
with the property γ ◦ (h ⊗ id)|A ⊗ 1 = (ψ ⊗ id)A ⊗ 1.
This is, because, on one side, there is a conditional expectation Q := P ⊗(λ(·)1)
from O2 ⊗ O2 onto ψ(A) ⊗ 1, (where λ is a pure state on O2 ), but there can not
be a conditional expectation Q0 from O2 ⊗ O2 onto h(A) ⊗ 1, because otherwise
((h|A)−1 ⊗ 1)((Q0 |(h(B) ⊗ 1)) defines an conditional expectation from B onto A.
Suppose that A is a nuclear C *-subalgebra of a separable unital exact C *-
algebra B such that 1B ∈ A and such that there does not exist a conditional
expectation from B onto A. If k : B ,→ O2 is a unital embedding and if we define
i := k|A as the restriction of k to A, then there does not exist any conditional
expectation E from O2 onto i(A) = k(A), because otherwise k −1 ◦ E ◦ k would be
a conditional expectation from B onto A.
Examples:
(o) Suppose that D is a hereditary C *-subalgebra of a unital C *-algebra B
such that the normalizer algebra N (D) := {b ∈ B ; bD + Db ⊂ D} is different from
A := D + C1 and that bD 6= {0} for all b ∈ B+ \ {0}. Then there does not exist a
794 6. EXACT SUBALGEBRAS OF PURELY INFINITE ALGEBRAS

conditional expectation E from B onto A. (Here C denotes the complex numbers.)

Indeed: (E(b) − b)∗ (E(b) − b)d = 0 for all b ∈ N (D), d ∈ D, i.e. E|N (D) = id.
It yields e.g. the following two examples (i) and (ii):
(i) Consider O2 as a unital C*-subalgebra of L(`2 ) (by some *-representation).
Let K denote the compact operators, and let A := K + C1, B := K + O2 .
(ii) Let B := T , A := K+C1, where T denotes the Toeplitz algebra, generated
as C*-algebra by the Toeplitz operator T (i.e. the unilateral shift of `2 ).
(iii) Consider the natural continuous epimorphism σ from the Cantor Space
Ω := {0, 1}∞ onto [0, 1] given by
∞
X
σ : (a1 , a2 , ...) 7→ an 2−n
n=1

Then γ : f ∈ C[0, 1] 7→ γ(f ) := f ◦ σ ∈ C(Ω) gives a unital embedding of C[0, 1]

N∞
into C(Ω) = n=1 (C ⊕ C) ⊂ M2∞ .
Let A := C[0, 1] ⊂ C(Ω) =: B, where Ω is the Cantor set Ω := {0, 1}∞ , and
the monomorphism A ,→ B is given by f 7→ f ◦ ρ for the continuous map
X
ρ : (α1 , α2 , . . .) → αn 2−n
n=1

from Ω onto [0, 1].

The pairs A := γ(C[0, 1]) and B := C(Ω) – or B := M2∞ – have the property
that there does not exist a conditional expectation from B onto A, because the
continuous map σ is not open.
Still Question ?: Suppose that i, j : A ,→ O2 are unital embeddings such that
there are extremal conditional expectations from O2 onto i(A), respectively onto
j(A).
Is there an automorphism γ of O2 ⊗ O2 with γ(i(a) ⊗ 1) = j(a) ⊗ 1 for all a ∈ A
?
There are nuclear A and unital embeddings i, j : A ,→ O2 such that there is
a conditional expectation P from O2 onto i(A), but that there does not exist a
conditional expectation from O2 onto j(A).
It follows that there is no automorphism γ of O2 ⊗ O2 such that γ(i(A) ⊗ 1) =
j(A) ⊗ 1, because — otherwise — there is an automorphism β ∈ Aut(A) with
γ(i(a) ⊗ 1) = j(β(a)) ⊗ 1) for a ∈ A and the the conditional expectation P : O2 →
i(A) (with P (O2 ) = i(A)) defines a conditional expectation Q : O2 → j(A) (with
Q(O2 ) = j(A) = (j ◦ β)(A)) by Q(b) := id ⊗λ γ(P ⊗ λ(γ −1 (b ⊗ 1))) , where

λ(b) := ρ(b)1 ∈ O2 for b ∈ O2 .

What the hell is that ?:
One can show that there exist unital endomorphisms h1 , h2 : O2 → O2 , such
that there is no automorphism γ of O2 ⊗ O2 such that γ ◦ (h1 ⊗ id) = h2 ⊗ id. An
example is given by h1 , h2 as hk (a) = 1 ⊗ ιk (a) defined by isomorphisms ι1 from
 3. THE RESIDUALLY NUCLEAR CASE 795

M2∞ ⊗O2 onto O2 and ι2 from O2 ⊗O2 ∼ = O2 onto O2 : Here such an automorphism
γ would also define an isomorphism from M2∞ onto O2 .
???????????
Notice that, moreover, the latter endomorphisms hk have the property that
there is an extremal conditional expectation P from O2 ⊗ O2 onto hk (O2 ) ⊗ O2 .
The above examples suggest, that a separable unital exact C *-algebra A is
finite-dimensional if all unital *-monomorphisms h : A ,→ O2 are conjugate by an
automorphism γ of O2 . (The K0 -triviality of O2 implies immediately that all unital
*-monomorphisms h : A → O2 are unitarily equivalent by a unitary in O2 .)
Does there exist simple separable nuclear unital C *-subalgebras A ⊂ B such
that 1B ∈ A and such that there does not exist conditional expectation from B
onto A?
Does there exist a unital endomorphism ι : M2∞ ,→ M2∞ such that there is no
conditional expectation from M2∞ onto ι(M2∞ ) ?
Then it would follow that there are unital endomorphisms i : O2 ,→ O2 such
that there does not exist a conditional expectation from O2 onto i(O2 ), because
O2 ∼
= A ⊗ O2 ⊂ B ⊗ O2 ∼ = O2 .

3. The residually nuclear case

The proof of a Ψ-residually equivariant version of Proposition 6.2.1 requires

some generalizations of the above used ideas. In particular, we must generalize
some definitions and results of [437] and [438]. The following Theorem 6.3.1 is
half on the way to the proof of Theorem K. The proof of Theorem K will be
completed in Chapter 12. Compare Definitions 1.2.6, 1.2.8 and 1.2.3 and Chapters
3 and 5 for the used notation.

Theorem 6.3.1. Suppose that A and B are separable stable C*-algebras, and
Ψ : I(B) → I(A) is an action of Prim(B) on A, that has the following properties
(i)–(v):

(i) B has the WvN-property ( 4 ), and

(ii) B has residually nuclear separation ( 5 ),
(iii) A is exact ( 6 ).
(iv) Ψ is non-degenerate, lower semi-continuous, and monotone upper semi-
continuous ( 7 ).

4 Def. WvN-property ????

5 Residually nuclear separation is defined in Definition 1.2.3.
Equivalently, the universal
weakly residually nuclear *-morphism Hrn : B → M(B) of Definition ?? is a non-degenerate
*-monomorphism and separates the ideals of B by Proposition 5.9.24.
6 give Def. exact ??
7 Defs. non-degenerate, lsc, monotone usc.
796 6. EXACT SUBALGEBRAS OF PURELY INFINITE ALGEBRAS

(v) The m.o.c. cone C ⊂ CP(A, B) of Ψ-equivariant completely positive maps

( 8 ) from A into B is separating for Ψ, i.e., ΨC = Ψ by Proposition ??
I.e. there exists a *-monomorphism H : A → M(B), with the property
Ψ(J) = H −1 (H(A) ∩ M(B, J)) for all J ∈ I(B) ( 9 ).

The above properties lead to a non-degenerate nuclear *-monomorphism k : A ⊗

O2 ,→ B such that h(Ψ(J)) = h(A) ∩ J for all J ∈ I(B), where h(a) := k(a ⊗ 1).
The infinite repeats δ∞ h and δ∞ ◦ H are unitarily homotopic.
If, in addition, A is nuclear, A contains a regular Abelian C*-subalgebra, and
the action Ψ is continuous (i.e., Ψ(I) ∪ Ψ(J) = Ψ(I ∪ J) – in addition), then h
is unitarily homotopic to a non-degenerate *-morphism h1 : A → B such that there
exists an approximately inner conditional expectation P from B onto h1 (A).

( 10 ).
It follows from [464] that A ⊗ O2 contains a regular abelian C *-subalgebra for
every nuclear separable C *-algebra A.
If B is any separable C *-algebra, and B ⊗ O2 contains a regular Abelian C *-
subalgebra C ⊂ B ⊗ O2 , then every lower semi-continuous action Ψ : I(B) → I(A)
is realized by CPrn (Ψ; A, B) = CPrn (Prim(B); Ψ, id; A, B), cf. Proposition ??. In
particular, then B has residually nuclear separation (in the sense of Definition 1.2.3,
i.e., CPrn (B, B) defines the identity action of Prim(B) on B), cf. Corollary ??. A
separable C *-algebra B with residually nuclear separation has the WvN-property,
if and only if, B is strongly purely infinite (cf. Proposition ??). Hence, Theorem
6.3.1 implies:
Corollary 6.3.2. Theorem K is valid under the additional assumptions that B
is separable and that B ⊗O2 contains a regular Abelian C*-subalgebra C ⊂ B ⊗O2 .
Remark 6.3.3. We derive from Corollary 6.3.2 a complete proof of Theorem K
in Chapter 12, using the study of scale-invariant morphisms in Chapter 9. The
proof relies on the construction of a suitable strongly purely infinite separable
C *-subalgebra B1 of Q(R+ , B) := Cb (R+ , B)/ C0 (R+ , B) that contains a regu-
lar Abelian C *-subalgebra and a sufficiently large separable C *-subalgebra of B as
a non-degenerate subalgebra (of B1 ).
8 Give Ref to Def.Psi.equivariant.maps
9 I.e., Ψ = Ψ for the action Ψ of Prim(B) on A and the m.o.c. cone C ⊂ CP(A, B) of the
C
Ψ-equivariant nuclear maps. Then there is a C *-morphism h1 : A → M(B) such that h2 :=
M(Hrn ) ◦ h1 induces the action ΨA of Prim(B) on A by ΨA (J) = h−1 (h(A) ∩ M(B, J)) for
J ∈ I(B), cf. Proposition 5.9.24.
10 We assume, in addition to the assumptions of Theorem K, that B has residually nuclear

separation in sense of Definition 1.2.3, i.e., that the universal weakly residually nuclear *-morphism
Hrn : B → M(B) of Proposition 5.9.24
is non-degenerate and is separating for the ideals of B, and that ΨC = Ψ for the action Ψ
of Prim(B) on A and the m.o.c. cone C ⊂ CP(A, B) of the Ψ-residually nuclear maps, i.e., that
there is a C *-morphism h1 : A → M(B) such that h2 := M(Hrn ) ◦ h1 induces the action ΨA
of Prim(B) on A by ΨA (J) = h−1 (h(A) ∩ M(B, J)) for J ∈ I(B), cf. Proposition 5.9.24. This
additional assumptions will be removed both in Chapter 12.
 3. THE RESIDUALLY NUCLEAR CASE 797

Corollary 6.3.2 gives a nuclear *-monomorphism k : A ⊗ O2 → B1 ⊂ Q(R+ , B)

with image in the ideal generated by B ⊂ Q(R+ , B) and with the property that
k(A ⊗ 1) ∩ J = k(ΨA (B ∩ J) ⊗ 1) for any closed ideal J of Q(R+ , B).
The latter property induces the existence of a unitary U ∈ Cb (R+ , M(B)) and
of a non-degenerate nuclear *-monomorphism h : A ⊗ O2 → B with k(a ⊗ 1) =
U ∗ h(a ⊗ 1)U + C0 (R+ , B) for a ∈ A (see Chapter 9). The *-morphism h is as
stipulated in Theorem K.

Remark 6.3.4. Since every simple algebra has residually nuclear separation,
Theorem 6.3.1 contains Theorem A(i), if we consider O2 ⊗ A ⊗ K ⊂ L(`2 ) ∼ =
M(K) ⊂ M(B) for B = O2 ⊗ K. Our proof of Theorem 6.3.1 is independent
from almost all results on p.i.s.u.n. algebras, because for the existence of the lift
k : A ⊗ O2 → B we have only used that A ⊗ O2 ⊗ O2 is exact, and our elementary
characterization of absorbing liftable elements in SExtnuc (Prim(B); A, B).

We describe now the steps of the proof of Theorem 6.3.1.

Recall that H : A → M(B) is weakly nuclear, iff, or every b ∈ B, the map
A 3 a → b∗ H(a)b ∈ B is nuclear.
Since Ψ is non-degenerate, we get that H : A → M(B) is faithful and BH(A)B
is dense in B. We have see in Chapter 3 where exactly? that there exists a non-
degenerate weakly nuclear *-monomorphism H0 : A → M(B) with δ∞ ◦H0 unitarily
equivalent to H0 , such that H0 is unitarily homotopic to δ∞ ◦ H.
It follows that H0 (A) ⊂ δ∞ (M(B)). The commutant of δ∞ (M(B)) in M(B)
contains a copy of O2 unitally.
We rename δ∞ span(H0 (A) · O2 ) ∼


= A ⊗ O2 by A. Then, from now on,
∼
A = A ⊗ O2 . Then (i)–(v) imply:
(1) A is non-degenerate,i.e., AB is dense in B, and
(2) for every b ∈ B, the map A 3 a → b∗ H(a)b ∈ B is nuclear.
(3) The action Ψ : I(B) → I(A) is given by Ψ(J) := A ∩ M(B, J) and is
S S
monotone upper semicontinuous,i.e., n Ψ(Jn ) is dense in Ψ(J) for J := n Jn ,
J1 ⊂ J2 ⊂ · · · ⊂ B.
The proof of Theorem 6.3.1 will be given at the end of this Section. But here
we describe the steps of the proof:
First step:
With A satisfies also δ∞ (A) and C ∗ (δ∞ (A), O2 ) ∼
= A ⊗ O2 the properties (1)-(3).
Second step:

Next had disappeared and is restored from beamers

Steps of the embedding realization. Step 1:
We start with λ : B → M(B), a weakly continuous non-degenerate C *-morphism
λ : B → M(B) that defines the identity action J = λ−1 λ(B) ∩ M(B, J) on I(B)


798 6. EXACT SUBALGEBRAS OF PURELY INFINITE ALGEBRAS

and is in “general position” (i.e., δ∞ ◦ λ = u∗ λ(·)u). It exists because our B has in

particular “Abelian” factorization (and is separable, stable and s.p.i.).
Then obtain for A ∼ = A ⊗ D2 our non-degenerate Ψ-realization H1 : A → M(B)
– coming also from the “Abelian” factorization property of B that reduces it to a
classical selection problem. Get non-degenerate

H0 := M(λ) ◦ H1 : A ⊂ M(λ)(M(B)) ⊂ M(B) .

Here we list the former results that are needed for the “remaining” parts of the
proof:
(1) H0 (A) is non-degenerate, i.e., H0 (A)B is dense in B, and is in “general
position” (i.e., there exists a unitary U ∈ M(B) U ∗ H0 (·)U = δ∞ ◦ H0 ).
(2) The given lower s.c. action Ψ : I(B) → I(A) is realized by Ψ(J) :=
H0−1 (H0 (A) ∩ M(B, J)) and Ψ is monotone upper semi-continuous, i.e., n Ψ(Jn )
S
S
is dense in Ψ(J) for J := n Jn , if the sequence Jn ∈ I(B) is increasing:
J1 ⊂ J2 ⊂ · · · .
(3) For every b ∈ B, the ΨA -compatible map A 3 a 7→ b∗ H0 (a)b ∈ B is nuclear,
and can be approximated by compositions V2 ◦ V1 of the residually nuclear maps
V1 : a ∈ A 7→ b∗1 H1 (a)b1 ∈ B and V2 : b ∈ B 7→ b∗2 λ(b)b2 . I.e. V2 satisfies V2 (J) ⊂ J
for all J ∈ I(B) and that [V2 ]J : B/J → B/J is nuclear for all J ∈ I(B).
Step 2:
If A satisfies (1)–(3), then, – with a ∈ A identified with H0 (a) ∈ H0 (A) ⊂ M(B)
– for every a1 , . . . , an ∈ A and ε > 0, there exist completely positive contractions
V : M(B) → B and W : B → M(B) that satisfy the following conditions (a), (b)
and (c):

(a) kW ◦ V (aj ) − aj k < ε, for j = 1, . . . , n.

(b) V is strictly continuous and is residually equivariant, i.e., limn kV (bn ) −
V (b)k = 0 if bn → b in M(B) strictly and V (J) ⊂ J ∩B for J ∈ I(M(B)).
(c) W : B → M(B) is weakly residually nuclear, i.e., W (J)B ⊂ J for J ∈
I(B), and the maps [W ]J : B/J → M(B/J) ∼ = M(B)/M(B, J) satisfy
that ([W ]J )d : b ∈ B/J 7→ d∗ [W ]J (b)d ∈ B/J is a nuclear map for all
d ∈ B/J. Here [W ]J (a + J) := W (a) + M(B, J), i.e., d∗ [W ]J (b)d =
πJ (f ∗ W (a)f ) for b = a + J and d = f + J.

More on Step 2:
Consider the set of maps V := Vc : M(B) → B given by Vc : b ∈ M(B) 7→
∗
c M(λ)(b)c. The point-norm closure is an m.o.c. cone C1 .
Do the same with the maps W := WT : B → M(B) given by WT (b) :=
∗
T λ(b)T , for T ∈ M(B), and we denote this m.o.c. cone by C2 .
Both cones are singly generated (as m.o.c.c.), e.g. C1 by eM(λ)(·)e where e ∈
B+ is strictly positive, and C2 by λ (as u.c.p. map).
 3. THE RESIDUALLY NUCLEAR CASE 799

The properties (b) and (c) follows from the properties of λ and the fact that
M(λ) is the unique strictly continuous extensions of λ.
Continuation 1: More on Step 2:
Notice that λ(b) and δ∞ (b) for each self-adjoint b ∈ B must be unitarily
homotopic in M(B) by the generalized W-vN theorem, because both define *-
mono-morphisms from C ∗ (b) into M(B) that are in “general position” and de-
fine the same action of Prim(B) on C ∗ (b) by definition of λ, namely the action
J ∈ I(B) 7→ M(B, J) ∩ C ∗ (b).
It follows that each W ∈ C2 maps B into the closure I0 of M(B)δ∞ (B)M(B),
the closed ideal of M(B) generated by δ∞ (B).
Moreover, b ∈ M(B)+ will be mapped by all P ∈ C2 ◦C1 into the ideal of M(B)
that is the norm-closure of the union of the ideals M(B, J((ebe − 1/n)+ )), with
J(b) ∈ I(B) as defined above, e ∈ B+ strictly positive contractions.
A clear statement is now in Chapter 12:
The point-norm closed m.o.c. cone C1 ⊂ CP(A, M(B)) generated by a ∈ A →
W (V (a))) ∈ M(B) is contained in CPnuc (A, M(B)) and is element-wise approxi-
mating. Thus, is point-norm approximating for H0 : A → M(B).
Continuation 3: More on Step 2:
The condition (a) is equivalent to H0 ∈ C3 := C2 ◦ C1 ◦ CH0 . The m.o.c. cone
C3 is contained in the cone of (norm-) nuclear c.p. maps from A into M(B), and
the elements of C3 map A into the norm-closed ideal I(δ∞ (B)) of M(B) that is
generated by δ∞ (B).
More precisely, the elements of C3 map a ∈ A+ into the closed ideal of
2 3 2
I(δ∞ (B)) ⊂ M(B) generated by the element δ∞ (e)δ∞ (e)δ∞ (a)δ∞ (e)δ∞ (e).
If there is b ∈ B+ such that δ∞ (b) and a generate the same closed ideal of
M(B), then there are Tn ∈ C2 ◦ C1 such that kTn (a) − ak → 0.
Step 3:
We use that A ∼ = A ⊗ O2 ⊗ O2 ⊗ · · · , ( 11). and select suitable Vn : M(B) → B,
Wn : B → M(B) from Step 2, such that Wn ◦ Vn |A converges sufficiently fast
in point-norm to the inclusion A ,→ M(B) (where H0 (A) and A are naturally
identified). Then A is in the multiplier sub-algebra of the separable (non-unital)
C *-system

Y := indlimn (Wn ◦ Vn : M(B) → M(B)) ⊂ M(B)∞ .

By a standard Banach space theory argument, (cf. Section 14 of Appendix B), Y is

completely positive and completely isometric isomorphic to the (nuclear) C *-system

X := indlimn (Vn+1 ◦ Wn : B → B) ⊂ B∞ = `∞ (B)/c0 (B) .

11Notice that we do not use that O ∼ O ⊗ O ⊗ ·, because we want to give a proof that is
2 = 2 2
independent from the former results on p.i.s.u.n./ algebras, and gives also a new proof of Theorem
A.
800 6. EXACT SUBALGEBRAS OF PURELY INFINITE ALGEBRAS

Step 4:
An inspection of the natural completely positive and completely isometric isomor-
phism I from Y onto X shows that
(α) The isomorphism I maps A ⊂ Y into the intersection X ∩ M(X) of the
(two-sided) multiplier algebra M(X) ⊂ X ∗∗ of X with X ⊂ X ∗∗ .
(β) This happens in an ideal-system equivariant way, i.e.,

I(M(B, J)∞ ∩ Y ) = J∞ ∩ X for all J ∈ I(B) .

(γ) The operator system X is nuclear because the maps Tn : Vn+1 ◦ Wn are
nuclear, i.e., Tn can be approximated point-wise by “decomposable” contractions
that factorize approximately through suitable matrix-algebras Mkn .
Use here and above that inductive limits of – not necessarily unital – C*-systems
by using c.p. contractions always are C*-systems, and that the second conjugate
with the natural matrix-order and matrix-norm is a W*-algebra.
Step 5:
We construct a hereditary C *-subalgebra D ⊂ B and a completely positive and
completely isometric isomorphism ϕ from X onto B//D, that have the properties
(P1) D is residually essential – in particular D is stable,
(P2) ϕ(J∞ ∩ X) = πDB+BD (J),
(P3) ϕ(X ∩ M(X)) = πD (N (D)),
(P4) the natural map N (D)/D → M(D)/D ∼
= Q(B) extends to completely
positive map γ from B//D into M(D)/D with the property that γ(J//D) =
γ(B//D) ∩ M(D, D ∩ J) for all J . B.
Idea for finding D in Step 5 is the following
Lemma on replacements of inductive limits:
(See somewhere below)
End of Lemma.
Since B is stable, separable and s.p.i., we get that the (I(B)-) residually nuclear
contractions P : B → B can be approximated point-wise by maps T : B → B given
by TS (b) := S ∗ bS for isometries S ∈ M(B) with 1 − SS ∗ properly infinite. Thus
we can take a fixed copy of C ∗ (s, t) ∼
= O2 in M(B) and find the approximating TS
as S = U ∗ sU for suitable unitary U ∈ M(B).
Thus X becomes by the Lemma simply the same as indlim Tn : B → B with
Tn (b) = Un∗ s∗ Un bUn∗ sUn for a suitable sequence of unitaries Un ∈ M(B). It turns
then out that X is c.i. and c.p. isomorphic to B//D where D is found in a certain
inductive limit of copies of B by inner automorphisms of B.
Step 6:
We apply the Proposition on Ψ-compatible Busby invariants to the Busby invariant
γ ◦ ϕ ◦ I : A → M(D)/D. It has image in N (D)/D and admits a Ψ-equivariant lift
h : A → N (D) ⊂ B, that is nuclear as map from A into B.
 3. THE RESIDUALLY NUCLEAR CASE 801

(But h is not necessarily nuclear as a map from A into N (D).)

It turns out that h is unitarily homotopic to a non-degenerate nuclear mono-
morphism h0 : A → B that is Ψ-equivariant.
Step 7:

????????????????????????
The case of (bi-) continuous action Ψ of Prim(B) on A.
Target: If A is nuclear and h0 (A) regular in B, then there exists an Ψ-
equivariant conditional expectation from B to h0 (A) or at least a sequence of Ψ-
equivariant c.p. contractions Vn : B → h0 (A) with Vn (h0 (a)) → h0 (a) for n → ∞.
If Ψ : I(B) → I(A) is “continuous”, then h0 (A) must be a regular C *-
subalgebra of B. It follows, that there is a lower semi-continuous action
Φ : I(A) → I(B) with Φ(h(J)) := biggest K . B with K ∩ h(A) ⊂ h(J).
REF ??????????
Since A contains a regular abelian C *-subalgebra, the cone CPrn (Φ; B, A) of
Φ-residually nuclear maps W : B → A is separating for Φ, i.e., for each b ∈ B+
and each pure state λ on A with λ(J) 6= {0}, there is W ∈ CPrn (Φ; B, A) with
λ(W (Φ(J))) 6= {0}.
Since Φ(K ∩ h(A)) ⊃ K for all ideals K . B, it follows that Φ ◦ Ψ majorizes Id
of I(B).
??????? ??????????????????? If A is nuclear, ??????????????
??
Overview on proof Thm. 6.3.1 finished?
We need some simple general results on residually essential hereditary C *-
subalgebras D of a C *-algebra B, non-unital C *-systems and on inductive limits.
First we explain some properties of certain subspaces of C *–algebras.

Remark 6.3.5. Suppose that B is separable and that A ⊂ M(B) is a separable

stable exact C *-subalgebra, such that AB is dense in B and that the maps a ∈
A 7→ b∗ ab ∈ B are nuclear for all b ∈ B.
Then B is stable, and maps V : M(B) → B and W : B → M(B) with prop-
erties (a), (b) and (c) for (given) a1 , . . . , an ∈ A and ε > 0 exist, if and only
if,

(i) The m.o.c. cone CPrn (Prim(B), B, B) of residually nuclear maps from
B to B is non-degenerate, i.e., the universal weakly residually nu-
clear Hrn : B → M(B) C *-morphism (given by Corollary ?? for
C := CPrn (Prim(B); B, B)) is non-degenerate.
And separating for Prim(B)??
(ii) The representations a ∈ A 7→ M(Hrn )(a) ∈ M(B) and a ∈ A 7→ a ∈
M(B) define the same action of of Prim(B) on A.
802 6. EXACT SUBALGEBRAS OF PURELY INFINITE ALGEBRAS

(iii) The (lower semi-continuous) action of Prim(B) on A is monoton upper

semi-continuous.

(See the arguments in Chapter 12 for a proof.)

????????????
Move the needed arguments of Chapter 12 to here. ??
The conditions (i)–(iii) are satisfied if the action of Prim(B) on B defined by
Hrn is the identity map of I(B) ∼= O(Prim(B)).

Definition 6.3.6. A hereditary C *-subalgebra D ⊂ B is residually essential

in B, if for every closed ideal J ∈ I(B) holds that πJ (D) is essential in B/J, i.e.,
b ∈ B+ and bD ⊂ J imply b ∈ J (equivalently: Ann(πJ (D)) = 0 in B/J).

Lemma 6.3.7. Let T : X → C a positive and isometric linear map from a (not
necessarily unital) C*-system X into a C*-algebra C.
Then T −1 (C+ ∩ T (X)) = X+ .

Proof. We can pass to the second adjoint T ∗∗ : X ∗∗ → C ∗∗ . The map T ∗∗ is

still positive and isometric, T ∗∗ |X = T , X+ = (X ∗∗ )+ ∩X, and X ∗∗ is unitally order
isomorphic to a unital C *-algebra. Thus, we may suppose that X is a C *-algebra.
Let a = h + ik ∈ X with self-adjoint h, k ∈ X with polar decompositions
h = h+ − h− , k = k+ − k− , and suppose T (a) ∈ C+ . Then T (h+ ), T (h− ), T (k+ )
and T (k− ) are all in C+ . Thus T (k) = 0, a = h+ − h− and 0 ≤ T (h− ) ≤ T (h+ ).
Suppose that h− 6= 0. Then T (h+ ) 6= 0 and h+ 6= 0. Let x := kh− k−1 h− and
y := kh+ k−1 h+ . It follows x, y ∈ X+ , xy = 0, T (x), T (y) ∈ C+ , kT (x)k = kxk = 1,
kT (y)k = kyk = 1 and kT (x + y)k = kx + yk = max(kxk, kyk) = 1. Thus,
there exists a state ρ on C with ρ(T (x)) = 1 (by extending a suitable character
of C ∗ (T (x)) to C). Then 1 ≤ ρ(T (x)) + ρ(T (y)) = ρ(T (x + y)) ≤ 1. It implies
ρ(T (y)) = 0, which contradicts kh− kT (x) ≤ kh+ kT (y) if kh− k > 0. Thus h− = 0
and a = h+ . 

Lemma 6.3.8. Suppose that B and C are C*-algebras, that A is a C*-subalgebra

of C, that ΨC : I(B) → I(C) is a monotone map, that D ⊂ B is a hereditary C*-
subalgebra of B.
If T : B//D → C is a completely positive and completely isometric linear map,
such that A ⊂ T (B//D) and, for J ∈ I(B),

T (πD (J)) = ΨC (J) ∩ T (B//D) ,

then the *-monomorphism λ := T −1 |A : A → N (D)/D satisfies, for J ∈ I(B),

λ(A ∩ ΨC (J)) = λ(A) ∩ πD (J) .

Proof. The map λ : A → (B//D)∗∗ ∼= qD B ∗∗ qD is an isometric completely

positive map by Lemma 6.3.7, because

T ∗∗ ⊗ idn = (T ⊗ idn )∗∗ : qD B ∗∗ qD ⊗ Mn → C ∗∗ ⊗ Mn

 3. THE RESIDUALLY NUCLEAR CASE 803

is isometric and positive for each n ∈ N. Since kλk ≤ 1, we get from the generalized
Kadison inequality of Choi that λ(a∗ a) ≥ λ(a)∗ λ(a) for a ∈ A. If we apply T ∗∗ to
this inequality and use the Choi-Kadison inequality for T ∗∗ , we get

a∗ a = T ∗∗ (λ(a∗ a)) ≥ T ∗∗ (λ(a)∗ λ(a)) ≥ T ∗∗ (λ(a))∗ T ∗∗ (λ(a)) = a∗ a .

Thus, T ∗∗ (λ(a∗ a) − λ(a)∗ λ(a)) = 0. Using that T ∗∗ is faithful on C ∗ (λ(A)) we get

λ(a∗ a) = λ(a)∗ λ(a) for all a ∈ A.
Thus λ : A → qD B ∗∗ qD is a *-monomorphism with λ(A) ⊂ qD BqD = B//D.
By Remark 6.1.2(14), λ(A) ⊆ M(B//D) ∩ (B//D) = N (D)/D.
The equation follows from A ∩ ΨC (J) = A ∩ T (πD (J)), because T is isometric.


Lemma 6.3.9. Suppose that D is a σ-unital hereditary C*-subalgebra of a

C*-algebra B, and that A is a separable C*-subalgebra of the normalizer algebra
N (D) ⊂ B.
Let π0 : B → B//D and π2 : M(D) → M(D)/D denote the quotient maps, and
let ρ denote the natural C*-morphism from N (D) into M(D) with kernel Ann(D)
(the two-sided annihilator of D in B).
For a closed ideal J of B let

Ann(D, J) := {a ∈ B ; aD + Da ⊂ J} ⊃ J .

Then there exists a completely positive contraction T from the C*-system B//D
into the C*-algebra M(D)/D, such that T has the following properties (i)-(ii):

(i) T (π0 (Ann(D, J))) ⊂ π2 (M(D, D ∩ J)) for every J ∈ I(B).

(ii) (T ◦ π0 )|A is the C*-morphism (π2 ◦ ρ)|A from A into M(D)/D, and has
kernel A ∩ (D + Ann(D)).
(iii) T ◦ π0 (A ∩ Ann(D, J)) = T (π0 (A) ∩ π0 (Ann(D, J))) = T (π0 (A)) ∩
π2 (M(D, D ∩ J)) for J ∈ I(B).

In particular, – if B//D is nuclear –, then π2 ◦ρ defines a nuclear *-monomorphism

from A/(A ∩ (D + Ann(D))) into M(D)/D.

Proof. It holds N (D) ⊇ Ann(D) = Ann(D, {0}) ∼

= (Ann(D) + D)/D because
D + Ann(D) ∼ = D ⊕ Ann(D) naturally, thus D + Ann(D) is the kernel of π2 ◦
ρ : N (D) → Q(D) = M(D)/D.
Since D is σ-unital, it contains a strictly positive contraction e ∈ D+ . We can
suppose that e ∈ A (consider otherwise A1 := C ∗ (A, e)). Let a1 , a2 , . . . ∈ A a
dense sequence in the unit ball of A. There is an approximate unit e1 ≤ e2 ≤ . . . ∈
1/2
C ∗ (e) ⊂ D with en en+1 = en ≤ 1, ken e − en k < 2−n and k[ak , en ]k < 2−n−2 for
1/2
k ≤ n. Let g1 := e1 , gn+1 := (en+1 − en )1/2
P
b 7→ W (b) := n gn bgn is a completely positive contraction from B into M(D)
P
with W (e) = e, W (Ann(D)) = 0 and W (D) ⊂ D, because n gn bgn strictly
804 6. EXACT SUBALGEBRAS OF PURELY INFINITE ALGEBRAS

converges in M(D), cf. Remark 5.1.1(4). Moreover,

X
k(ρ(a) − W (a))(1 − em )k ≤ sup ka(en − em ) − gk agk (en − em )k ≤ 2−m .
n≥m
m−1≤k≤n+1

Thus ρ(a) − W (a) ∈ D for a ∈ A.

π2 ◦ W : B → M(D)/D is completely positive and contains D in its kernel.
It follows W (BD + DB) ⊂ D and that there is a unique completely positive map
T : B//D = B/(BD + DB) → Q(D) = M(D)/D with T ◦ π0 = π2 ◦ W .
(i): Let J ∈ I(B). Then W (Ann(D, J)) ⊂ M(D, D ∩ J), because eW (b)e =
P
egn bgn e ∈ D∩J for b ∈ Ann(D, J). Thus T (π0 (Ann(D, J))) ⊂ π2 (M(D, D∩J)).
(ii): T ◦ π0 |A = π2 ◦ ρ|A because π2 (W (a)) = π2 (ρ(a)) for a ∈ A. The kernel
of π2 ◦ ρ : N (D) → M(D)/D is ρ−1 (D) = D + Ann(D).
(iii): π2 ◦ ρ(A ∩ Ann(D, J)) = T ◦ π0 (A ∩ Ann(D, J)) ⊂ T (π0 (A) ∩
π0 (Ann(D, J))) ⊂ T (π0 (A)) ∩ π2 (M(D, D ∩ J)) = π2 (ρ(A)) ∩ π2 (M(D, D ∩ J),
because W (Ann(D, J)) ⊂ M(D, D ∩ J), T ◦ π0 = π2 ◦ W and T ◦ π0 |A = π2 ◦ ρ.
Conversely, if b ∈ N (D) and ρ(b) = c + d with d ∈ D and c ∈ M(D, D ∩ J)
then c = ρ(b) − d = ρ(b − d) ∈ ρ(N (D)) and ρ(b − d) ∈ M(D, D ∩ J), i.e.,
b − d ∈ Ann(D, J). Thus b ∈ N (D) ∩ Ann(D, J) + D and π0 (b) ∈ π0 (Ann(D, J)).
Since D ⊂ N (D), (N (D) ∩ Ann(D, J)) + D = N (D) ∩ (Ann(D, J) + D) =
(π2 ◦ ρ)−1 (N (D) ∩ M(D, D ∩ J)). 

Related/re-formulated text (new ϕ good?) for Lemma 6.3.9: Next

version better? ??
Recall that for C *-subalgebra D of a C *-algebra B, N (D) (= N (B, D)) is
defined as the set of two-sided normalizers b ∈ B of D: bD ∪ Db ⊆ D.
This definition of N (B, D) gives also a natural C *-morphism ϕD : N (D) →
M(D) given by ϕD (b)d := bd for b ∈ N (D) and d ∈ D. The kernel of ϕD is the
two-sided annihilator Ann(D) (= Ann(D, 0)) of D in B defined as the set of b ∈ B
with bD = {0} = Db.
Some of the following are discussed in Chapter 2 ? Ref’s??

Lemma 6.3.10. Suppose that D ⊂ B is a σ-unital residually essential hereditary

C*-subalgebra of B.

(i) If B has no unital quotient, then D has no unital quotient.

(ii) The map J ∈ I(B) 7→ D ∩ J ∈ I(D) is a bijection onto I(D).
(iii) The natural map ϕD : N (D) → M(D) given is injective, and ϕD (a) ∈
M(D, D ∩ J), if and only if, a ∈ N (D) ∩ J.
(iv) For a ∈ N (D) and J ∈ I(B) holds:

a ∈ D + (N (D) ∩ J) ⇔ πB//D (a) ∈ πB//D (J) ⇔ ϕD (a) ∈ D + M(D, D ∩ J) .

 3. THE RESIDUALLY NUCLEAR CASE 805

(v) For each separable C*-subalgebra A ⊂ B there is a completely positive

contraction TA : A → M(D)/D such that TA |(A ∩ N (D)) = πD ◦ ϕD and
TA (A ∩ J) ⊂ πD (M(D, D ∩ J)) for each closed ideal J of B.

Sort out: Where ‘‘ hereditary in B’’ is needed for D??

Proof. Recall that the definition of “residual essential” (in B) for a (not
necessarily hereditary) C *-subalgebra D ⊆ B is:
For each closed ideal K ⊆ B, there is no non-zero element of (B/K)+ that is
orthogonal to πK (D).
Give ref. to Definition?
This implies obviously that D ⊆ K := span(BDB) = B. Moreover, the
hereditary C *-subalgebra E := span(DBD) of B is again residually essential C *-
subalgebra of B.
(ii): If D is a hereditary C *-subalgebra of a C *-algebra B that is not contained
in any closed ideal K of B with K 6= B, then the map
K ∈ I(B) 7→ JK := K ∩ D ∈ I(D)
is a bijective lattice isomorphism.
cite TEXT books for this
(i): Suppose that B has no unital quotient, and suppose that there exists a
closed ideal J of D such that D/J is unital.
The closed ideal K := BJB of B is the smallest closed ideal of B that contains
J.
Since D is hereditary, each ideal J of D is the intersection of D with an ideal
L of B, i.e., J = D ∩ L. Thus J ⊆ D ∩ K ⊆ D ∩ L = J, i.e., J = D ∩ K.
It implies that D/J ∼ = πK (D) = p(B/K)p for some projection in p ∈ B/K,
because πK (D) is a unital hereditary C *subalgebra of B by πK (D)πK (B)πK (D) =
πK (DBD) = πK (D).
Since πK (D) is “essential” in B/K it follows that (1−p)(B/K)(1−p) = {0}, i.e.,
that B/K is unital, a contradiction to the assumption that all non-zero quotients
B/K of B are not unital.
(iii): The map ϕD : N (D) → M(D) is injective because the kernel is Ann(D)
and Ann(D) = {0} for essential C *-subalgebras D of B (by definition of “essen-
tial”).
For a ∈ N (D) ⊆ B holds that ϕD (a) ∈ M(D, D ∩ J), if and only if, a ∈
N (D) ∩ J:
If a ∈ N (D) ∩ J, then aD ∪ Da ⊆ D ∩ J. Thus, ϕD (a) ∈ M(D, D ∩ J). If
a ∈ N (D) and ϕD (a) ∈ M(D, D ∩ J), then aD ∪ Da ⊆ D ∩ J. It follows that
πJ (a)πJ (D) = {0} and πJ (D)πJ (a) = {0}. It says that πJ (a∗ a + aa∗ ) is orthogonal
to πJ (D).
806 6. EXACT SUBALGEBRAS OF PURELY INFINITE ALGEBRAS

The algebra D is “residually essential”, which means that πJ (D) is essential in

πJ (B) for each J ∈ I(B).
It follows that πJ (a∗ a + aa∗ ) = 0 or that πJ (D) = {0} must happen, i.e., a ∈ J
or D ⊆ J. Thus a ∈ N (D) ∩ J or D ⊆ J.
But essential C *-subalgebras in B generate B as ideal, i.e., if D ⊆ J and D is
essential in B imply that J = B and a ∈ N (D) ⊆ J = B.
(iv): For a ∈ N (D) are equivalent

a ∈ D + (N (D) ∩ J) ⇔ πB//D (a) ∈ πB//D (J) ⇔ ϕD (a) ∈ D + M(D, D ∩ J) .

Let a ∈ D + (N (D) ∩ J) then a = d + c with d ∈ D and c ∈ N (D) ∩ J. Then

trivially πB//D (a) = πB//D (c) because the kernel of πB//D is BD + DB ⊇ D. Thus
πB//D (a) ∈ πB//D (J).
Notice that N (D)/D =∼ M(B//D) ∩ (B//D) by a natural isomorphism. πB//D
coincides on N (D) with the C *-morphism a 7→ a + D. If πB//D (a) ∈ πB//D (J) and
a ∈ N (D), then a ∈ (J + BD + DB) ∩ N (D) = D + (N (D) ∩ J).
(To see more look to the second conjugate of B: N (D) = {PD }0 ∩ B ⊂ {PD }0 ∩
B ∗∗ .)
(v): To be filled in ??
For each separable C *-subalgebra A ⊂ B there is a completely positive con-
traction TA : A → M(D)/D such that TA |(A ∩ N (D)) = πD ◦ ϕD and TA (A ∩ J) ⊂
πD (M(D, D ∩ J)) for each closed ideal J of B:


Proposition 6.3.11. Suppose that D is a residually essential hereditary C*-

subalgebra of a separable stable C*-algebra B with WvN-property, and that the op-
erator space B//D := B/(DB + BD) is nuclear.
Let π1 : N (D) → N (D)/D and π2 : M(D) → M(D)/D denote the natural
epimorphisms, and let A is a stable separable C*-subalgebra of M(B), such that
A ,→ M(B) is weakly residually nuclear and AB is dense in B. Then:

(i) D is stable and there is an isomorphism ϕ from B onto D, which is

unitarily homotopic to idB , and, therefore, satisfies ϕ(J) = D ∩ J for
J ∈ I(B).
(ii) The natural C*-morphism ρ from N (D)/D into M(D)/D is a nuclear
*-monomorphism, and satisfies, for closed ideals J of B,

ρ(π1 (N (D) ∩ J)) = ρ(N (D)/D) ∩ π2 (M(D, D ∩ J)).

(iii) If λ : A → N (D)/D ⊂ B//D is a *-monomorphism with

λ(A ∩ M(B, J)) = λ(A) ∩ π1 (N (D) ∩ J) for J ∈ I(B) ,

then an element

H := Q(ϕ)−1 ◦ ρ ◦ λ : A → Q(B)
 3. THE RESIDUALLY NUCLEAR CASE 807

is nuclear and defines an element [H] of the semi-group

SExtnuc (Prim(B); A, B) := S(H0 ; A, Q(B))
where H0 := πB ◦ δ∞ |A with 2[H0 ] = [H0 ].
The element [H] is contained in the subgroup
Extnuc (Prim(B); A, B) = [H0 ] + SExtnuc (Prim(B), A, B) .
(iv) If, in addition to (iii), [H] = 0 in Extnuc (Prim(B), A, B), then there is a
non-degenerate nuclear *-monomorphism h from A to B, such that
h(A ∩ M(B, J)) = h(A) ∩ J for J ∈ I(B) .
The infinite repeats δ∞ ◦ h and δ∞ |A are unitarily homotopic.

Proof. Since D is residually essential, we have in particular that B is the

closed span of BDB.
(i): Since B has the WvN-property, B and its hereditary C *-subalgebra D
must be purely infinite in particular. D is separable and can not have a unital
quotient, because for every closed ideal I of D there is a closed ideal J of B with
I = D ∩ J, and D/I ∼ = πJ (D) is an essential subalgebra of the stable C *-algebra
πJ (B) (cf. assumptions). By Corollary 5.5.4, D is stable. Now use Corollary 5.5.6
to get ϕ with property(i).
(ii): The C *-algebras N (D) and N (D)/D are separable. The natural C *-
morphism ρ : N (D)/D → M(D)/D is faithful because Ann(D) = {0}. Let
T : B//D → M(D)/D be as in Lemma 6.3.9 with
T πBD+DB | N (D) = π2 ◦ η| N (D) = ρ ◦ π1 ,
where π2 : M(D) → Q(D) = M(D)/D is the quotient map and η : N (D) → M(D)
denotes the natural *-morphism with η(d) = d for d ∈ D. Then T = T ◦ idB//D is
nuclear, because B//D is nuclear. We obtain ρ = T ◦γ, where γ denotes the natural
isomorphism from N (D)/D onto M(B//D)∩(B//D) with γ ◦π1 = πBD+DB | N (D)
that is given by Lemma A.19.8(i). Thus ρ is a nuclear *-monomorphism.
By Lemma 6.3.9, Ann(D, J) ⊂ πJ−1 (Ann(πJ (D)) = J for residually essential D
and for J ∈ I(B). It follows:
ρπ1 (N (D) ∩ J) = ρπ1 (N (D)) ∩ π2 (M(D, D ∩ J)).

(iii): The C *-algebra A is a stable C *-subalgebra of M(B), such that id |A is

weakly residually nuclear, and AB is dense in B. If there is a *-monomorphism λ
from A into N (D)/D with
λ(A ∩ M(B, J)) = λ(A) ∩ π1 (N (D) ∩ J)
for J ∈ I(B), then
H := (Q(ϕ))−1 ◦ ρ ◦ λ : A ,→ Q(B) := M(B)/B
is a monomorphism from A into Q(B). Here Q(ϕ) is the isomorphism from Q(D) :=
M(D)/D onto Q(B) induced by ϕ, i.e., Q(ϕ)(d + D) = M(ϕ)(d) + B ∈ Q(B) for
808 6. EXACT SUBALGEBRAS OF PURELY INFINITE ALGEBRAS

d ∈ M(D). Note Q(ϕ)−1 = Q(ϕ−1 ). Then H : A → Q(B) is nuclear, because ρ is

a nuclear monomorphism. In particular, A is an exact C *-algebra. By Corollary
??, the inclusion map id : A ,→ M(B) is nuclear, because M(B) is weakly nuclear
and A is exact.
The map H satisfies, by asumptions in (ii),
H(A ∩ M(B, J)) = Q(ϕ)−1 ρ(λ(A) ∩ π1 (N (D) ∩ J)) (3.1)
−1
= H(A) ∩ Q(ϕ) (π2 (M(D, D ∩ J))). (3.2)
But Q(ϕ)−1 (π2 (M(D, D ∩ J))) = πB (M(B, J)), because ϕ(B) = D and
M(ϕ)(M(B, J)) = M(D, ϕ(J)) = M(D, D ∩ J).
The exactness of A gives that H is moreover residually nuclear with respect to
the natural actions of Prim(B) on A ⊂ M(B) and Q(B) (cf. ??).
Thus H : A ,→ Q(B), B and A ⊂ M(B) satisfy the assumptions of Proposition
??. ?? Therefore, [H] is in G(H0 ; A, Q(B)) = Extnuc (Prim(B), A, B), where H0 :=
πB ◦ δ∞ |A : A → Q(B).
to be filled in: show [H] ∈ G(H0 ; A, Q(B)) ??
(iv): Since [H] = 0 = [H0 ] in Extnuc (Prim(B); A, B), it follows by Remark ??
??). that H = w∗ H0 (·)w where w ∈ U0 (Q(B)) and H0 = πB ◦ δ∞ |A, because the
extensions given by elements of Extnuc (Prim(B), A, B) are stable C *-algebras ( 12 ).
We find a unitary u ∈ M(B), such that
H1 : a ∈ A 7→ u∗ δ∞ (a)u ∈ M(B)
is a residually nuclear lift of H : A → Q(B) with H1 (A) ∩ M(B, J) = H1 (A ∩
M(B, J)) for J ∈ I(B). Then
H2 := M(ϕ) ◦ H1 : A → M(D)
is a residually nuclear lift of ρ◦λ : A ,→ M(D)/D. Necessarily, H2 (A) ⊂ η(N (D)) =
π2−1 ρ(D). Thus, the *-monomorphism H3 := η −1 ◦ H2 maps A into N (D) ⊂ B. We
have H3 (A) ∩ J = η −1 (H2 (A) ∩ η(N (D) ∩ J)), Ann(D, J) := πJ−1 (Ann(πJ (D))) =
πJ−1 (0) = J , Ann(D, J) = {b ∈ B ; bD + Db ⊂ J} ??
and that b ∈ N (D) ∩ Ann(D, J), if and only if bD + Db ⊂ D ∩ J. It implies
η(N (D) ∩ J) = η(N (D) ∩ Ann(D, J))
= η(N (D)) ∩ M(D, D ∩ J).
Thus
H3 (A) ∩ J = η −1 (H2 (A) ∩ M(D, D ∩ J))
= η −1 M(ϕ)(H1 (A) ∩ M(B, J))
= η −1 M(ϕ)H1 (A ∩ M(B, J))
= H3 (A ∩ M(B, J))
12 Use here that A is stable, and that B is σ-unital, stable and purely infinite, cf. Corollary

5.5.16 .
 3. THE RESIDUALLY NUCLEAR CASE 809

for J ∈ I(B).
Further, H3−1 (0) = {0}, and H3 (A) must generate B as a two-sided closed
ideal, because J1 := span BH3 (A)B satisfies H3 (A) ∩ J1 = H3 (A) and,thus, A ⊂
M(B, J1 ), AB ⊂ J1 ⊂ B and J1 = B AB = B.
Since A and B are stable and separable, it follows, by Proposition ?? ??, that
H3 is unitarily homotopic to a nuclear non-degenerate *-monomorphism h : A ,→ B.
We get h(A) ∩ J = h(A ∩ M(B, J)) for J ∈ B.
Thus k1 := δ∞ ◦ h and k2 := δ∞ |A are both nuclear, non-degenerate *-
monomorphisms and satisfy ki (A) ∩ M(B, J) = ki (A ∩ M(B, J)) for J ∈ I(B)
and i = 1, 2. By Corollary ??, k1 and k2 are unitarily homotopic. 

Lemma 6.3.12. Suppose that E ⊂ F ⊂ B are hereditary C*–subalgebras of B.

(i) If for every a ∈ M(B//E)+ ∩ (B//E) with a 6= 0 there is b ∈ πBE+EB (F )

with ab 6= 0, then F is essential in B.
(ii) If D ⊂ C is hereditary, B = C ⊗M2∞ , E = D ⊗M2∞ and ρ is a pure state
on M2∞ , then B//E ∼ = (C//D) ⊗ M2∞ by a natural completely positive
and completely isometric map V from B//E onto (C//D) ⊗ M2∞ .
The hereditary C*-subalgebra F ⊂ B that ist generated by {b ∈
B+ ; (idC//D ⊗ρ)(V (πBE+EB (b))) = 0 } satisfies the assumption of (i).
In particular, F is essential in B.

Proof. Recall that the product ab in (i) has to be calculated in the von Neu-
mann algebra (B//E)∗∗ , that is naturally isomorphic to (1 − pE )B ∗∗ (1 − pE ).
(i): Let c ∈ B+ with cF = {0}. Then cE = {0} and, therefore, c ∈
Ann(E)+ ⊂ N (E)+ . It follows that a := πBE+EB (c) ∈ M(B//E)+ ∩ (B//E) and
that aπBE+EB (F ) = {0}. Thus a = 0 by assumption (i), i.e., (1−pE )c(1−pE ) = 0,
c1/2 (1 − pE ) = 0, c = cpE and c = pE c, where pE ∈ B ∗∗ denotes the open support
projection of E. Thus c ∈ E. Since cE = {0} and c ∈ E+ , we get c2 = 0 and c = 0.
(ii): Let L := CD and R := DC the closed left respectively right ideals
of C generated by D. Then EB + BE = (L + R) ⊗ M2∞ . Therefore, there
is a natural completely positive and completely isometric isomorphism V from
B//E := (C ⊗ M∞ )//(D ⊗ M∞ ) onto (C//D) ⊗ M∞ that satisfies V ((c ⊗ d) +
EB + BE) = (c + (CD + DC)) ⊗ d for c ∈ C and d ∈ M2∞ , i.e., V ◦ πBE+EB =
πCD+DC ⊗ id. Let A := M(C//D) ∩ (C//D). Using slice maps, one can see that
V |M(B//E) ∩ (B//E) is a *-isomorphism from M(B//E) ∩ (B//E) onto A ⊗ M2∞ ,
and N (E) = N (D) ⊗ M2∞ .
Let G := {g ∈ M2∞ ; ρ(g ∗ g+gg ∗ ) = 0}. The G is the hereditary C *-subalgebra
of M2∞ generated by {0 ≤ g ∈ M2∞ ; ρ(g) = 0}, and G is essential in M2∞ because
0 ≤ a ∈ M2∞ and aG = 0 implies that a = 0 or that the pure state ρ is faithful on
the hereditary C *-subalgebra of M2∞ generated by a is faithful, but the latter can
not happen because M2∞ is anti-liminal.
810 6. EXACT SUBALGEBRAS OF PURELY INFINITE ALGEBRAS

Thus, N (D) ⊗ G ⊂ F . If a ∈ (A ⊗ M2∞ )+ and aV (F ) = {0}, then a(A ⊗ G) =

{0}. But this implies that (f ⊗ id)(a)G = {0} for every positive linear functional f
on A, thus f ⊗ id(a) = 0 for all f ∈ A∗ , hence a = 0.
It follows that c ∈ M(B//E)+ ∩ (B//E) and cπBE+EB (F ) = {0} together
implies c = 0. 

Lemma 6.3.13. Suppose that D1 , D2 are hereditary C*–subalgebras of B, that

T : B//D2 → (B//D1 )⊗M2∞ is a completely positive and completely isometric map
from B//D2 onto (B//D1 ) ⊗ M2∞ with T (π2 (J)) = π1 (J) ⊗ M2∞ for every closed
ideal J of B, where πk : B → B//Dk := B/(BDk + Dk B) are the quotient maps for
k = 1, 2. Let ρ : M2∞ → C a pure state, and let D3 denote the hereditary C*-algebra
of B that is generated by the elements f ∈ B+ with (id ⊗ ρ)(T (π2 (f ))) = 0.
Then

(i) D3 ⊃ D2 , D3 is residually essential in B, and

(ii) there is a completely positive and completely isometric isomorphism I from
B//D3 onto B//D1 with
I(π3 (J)) = π1 (J)
for every J ∈ I(B), where π3 : B → B//D3 is the quotient map π3 (b) :=
b + BD3 + D3 B.

Proof. Clearly, V : b ∈ B 7→ (id ⊗ρ)(T (π2 (b))) ∈ B//D1 . is a completely

positive contraction from B onto B//D1 with V (D3 ) = 0 and V (J) = π1 (J).
Let A := M2∞ and G ⊂ A the hereditary C *-subalgebra G := {a ∈ A ; ρ(a∗ a+
∗
aa ) = 0}. It turns out that I := [V ] : B//D3 → B//D1 is a complete isometry: One
has T (π2 (D3 )) = (B//D1 ) ⊗ G and T (π2 (BD3 + D3 B)) = (B//D1 ) ⊗ (AG + GA).
The same happens with T ⊗ idn .
to be filled in: new proof coming from Lemma 6.3.12 or 3.1.8
??
Let G := {a ∈ M2∞ ; ρ(a∗ a) = 0 = ρ(aa∗ )}. G is essential in M2∞ . There
is no non-zero element x ∈ ((B//D1 ) ⊗ M2∞ )+ that is orthogonal to the support
projections of (B//D1 ) ⊗ G in ((B//D1 ) ⊗ M2∞ )∗∗ , because there is no non-zero
x ∈ (B//D1 )∗∗ ⊗ M2∞ which is orthogonal to (B//D1 )∗∗+ ⊗ e for a strictly positive
element e of G.
γ has the property that π1 (J) ⊗ M2∞ corresponds to the natural image of
J ⊗ M2∞ in (B ⊗ M2∞ )//(D1 ⊗ M2∞ ). Thus, one has natural isomorphisms
((B/J)//πJ (D1 )) ⊗ M2∞ ∼
= ((B//D1 ) ⊗ M2∞ )/(π1 (J) ⊗ M2∞ ) .
?? Clearly, D2 ⊂ D3 and π2 (D3 ) contains T ((B//D1 ) ⊗ G). Recall that T ∗∗ is
a W*-algebra isomorphism from ((B//D1 ) ⊗ M2∞ )∗∗ onto (B//D2 )∗∗ . If b ∈ B+
with bD3 + D3 b = {0} then π2 (b) is orthogonal to T ((B//D1 ) ⊗ G) in (B//D2 )∗∗ .
The element x := T −1 (π2 (b)) is orthogonal to (B//D1 ) ⊗ G in ((B//D1 ) ⊗ M2∞ )∗∗ ,
which implies (step by step) x = 0, π2 (b) = 0, b ∈ D1 ⊂ D3 and b2 = 0, b = 0.
 3. THE RESIDUALLY NUCLEAR CASE 811

Hence, D3 is essential in B.
If J is a closed ideal of B, then πk (J) is an M-ideal of B//Dk (i.e., there
is a projection zJ in the center of (B//Dk )∗∗ with πk (J)∗∗ = zJ (B//Dk )∗∗ ) and
there are natural completely positive and completely isometric isomorphisms
(B/J)//πJ (Dk ) ∼ = (B//Dk )/πk (J). Thus T defines a completely isometric and
completely positive isomorphism [T ]J from (B/J)//πJ (D2 ) ∼
= (B//D2 )/π2 (J) onto
((B/J)//πJ (D1 )) ⊗ M2∞ ∼
= ((B//D1 ) ⊗ M2∞ )/(π1 (J) ⊗ M2∞ ) .
Let [π2 ]J denote the quotient map B/J → (B/J)/πJ (D2 ). πJ (D3 ) is essential in
B/J, because πJ (D3 ) is the hereditary C *–subalgebra of B/J which is generated by
the elements g ∈ (B/J)+ = πJ (B+ ) with id ⊗ρ)([T ]−1 J ([π2 ]J (g))) = 0, and because
we can replace in the above arguments B, B//D2 , (B//D1 ), D2 , T and π2 by B/J,
(B/J)//πJ (D2 ), ((B/J)//πJ (D1 )), πJ (D3 ), [T ]J and [π2 ]J . 

Lemma 6.3.14. Suppose that B is stable and that Tn : B → B is a sequence of

completely positive contractions. Then there is a sequence of isomorphisms γn : B →
B⊗M2n , such that γn is unitarily homotopic to the map b ∈ B 7→ b⊗p11 ∈ B⊗M2n .
The sequence (γn ) defines a *-isomorphism γ∞ from B∞ := `∞ (B)/c0 (B) onto
!
Y M
E := B ⊗ M2n / (B ⊗ M2n ) ⊂ (B ⊗ M2∞ )∞
n n

by
γ∞ ((b1 , b2 , . . .) + c0 (B)) := (γ1 (b1 ), γ2 (b2 ), . . .) + c0 (B ⊗ M2∞ ) .
−1
Let Sn := γn+1 (Tn ⊗ ιn ) ◦ γn , where ιn : a ∈ M2n 7→ a ⊗ 12 ∈ M2n+1 . Consider
X := indlim(Tn : B → B) and Y := indlim(Sn : B → B) as subspaces of B∞ :=
`∞ (B)/c0 (B) , and let
Z := indlim(Tn ⊗ ιn : B ⊗ M2n → B ⊗ M2n+1 ) ⊂ E .
Then

(i) Z coincides with the image of X ⊗ M2∞ by the natural *-monomorphism

η from B∞ ⊗ M2∞ into E.
(ii) The isomorphism γ∞ from B∞ onto E maps Y onto Z = η(X ⊗ M2∞ )
and satisfies
γ∞ (Y ∩ J∞ ) = η((X ∩ J∞ ) ⊗ M2∞ )
for all closed ideals J of B.
(iii) If Tn is residually nuclear, then Sn is residually nuclear and X is a nuclear
(non-unital) operator system.

Proof. Let κn an isomorphism from the algebra of compact operators K onto

K ⊗ M2n . Since every endomorphism of K onto a hereditary C *–subalgebra D of
K is unitarily homotopic to the identity map on K (cf. e.g. Corollary 5.5.6), we
get that a ∈ K 7→ a ⊗ p11 ∈ K ⊗ M2∞ is unitarily homotopic to κn . There is an
isomorphism λ from B onto B ⊗ K, because B is stable and K ⊗ K ∼ = K. Then
γn := (λ ⊗ idn )−1 ◦ (idB ⊗κn ) ⊗ λ is an isomorphism from B onto B ⊗ M2n that is
812 6. EXACT SUBALGEBRAS OF PURELY INFINITE ALGEBRAS

unitarily homotopic to b ∈ B 7→ b ⊗ p11 ∈ B ⊗ M2n . In particular, γn (J) = J ⊗ M2n

for J ∈ I(B), which implies
Y M
γ∞ (J∞ ) = (J ⊗ M2n )/ (J ⊗ M2n ) .
n n
L
n B ⊗ M2 ⊂ c0 (B ⊗ M2 ) and let Vn : M2
(i): Let I := n ∞ ∞ → M2n the

∼
natural conditional expectation from M2∞ onto M2n = M2n ⊗ 1 ⊗ 1 ⊗ . . . ⊂ M2∞
with Vn (a ⊗ b) = tr(b)a for a ∈ M2n , b ∈ M2∞ . Then η : B∞ ⊗ M2∞ → E is the
C *-morphism that is defined by

η (((b1 , b2 , . . .) + c0 (B)) ⊗ c) := (b1 ⊗ V1 (c), b2 ⊗ V2 (c), . . .) + I

for (b1 , b2 , . . .) ∈ `∞ (B) and c ∈ M2∞ . This defines η, because B∞ ⊗ M2∞ =

B∞ ⊗max M2∞ by nuclearity of M2∞ .
Let Uk,∞ : B ⊗ M2k → Z ⊂ E the canonical maps for the inductive limit Z
defined by the maps Tn ⊗ ιn , i.e., starting with k − 1 zeros one has

Uk,∞ (d) = (0, . . . , 0, d, Tk ⊗ ιk (d), (Tk+1 Tk ) ⊗ (ιk+1 ιk )(d), . . .) + I ,

where d ∈ B ⊗ M2k .
Then Z is the closed linear span of Uk,∞ (b ⊗ c) for b ∈ B and c ∈ M2k , k ∈ N.
A straight calculation shows that this set coincides with η(X ⊗ M2∞ ).
(ii): We have (Tn ⊗ ιn ) ◦ γn = γn+1 ◦ Sn . Thus,

γ∞ ◦ Sm,∞ (b) = Um,∞ (γm (b)) ,

for b ∈ B, which implies that the *-isomorphism γ∞ from B∞ onto E maps the
closed subspace Y onto Z.
Clearly, η(J∞ ⊗ M2∞ ) ⊂ γ∞ (J∞ ). Since γ∞ (J∞ ) is an ideal, and since M2∞ is
simple and nuclear, there is a unique closed ideal K of B∞ such that η(K ⊗M2∞ ) =
η(B∞ ⊗ M∞ ) ∩ γ∞ (J∞ ). Let (b1 , b2 , . . .) ∈ `∞ (B) with (b1 , b2 , . . .) + c0 (B) ∈ K.
Then

η (((b1 , b2 , . . .) + c0 (B)) ⊗ 1) := (b1 ⊗ 12 , b2 ⊗ 14 , . . .) + I ∈ γ∞ (J∞ ) ,

i.e., there exists a sequence cn ∈ J ⊗ M2n with lim k(bn ⊗ 12n ) − cn k = 0. If we

apply the slice maps idB ⊗tr2n : B ⊗ M2n → B, we get that there is a sequence
d1 , d2 , . . . ∈ J with lim kbn − dn k = 0, i.e., (b1 , b2 , . . .) + c0 (B) ∈ J∞ and K = J∞ .
The equation (X ⊗ M2∞ ) ∩ (J∞ ⊗ M2∞ ) = (X ∩ J∞ ) ⊗ M2∞ holds by nuclearity
of M2∞ .
(iii): Let J a closed ideal of B. Then γn (J) = J ⊗ M2n and (Tn ⊗ ιn )(J ⊗
−1


M2n ) ⊂ J ⊗ M2n+1 . Thus, Sn (J) = γn+1 (Tn ⊗ ιn )(γn (J)) ⊂ J , i.e., Sn is
residually equivariant if Tn is residually equivariant. Since residually nuclear maps
are residually equivariant, Sn is residually equivariant if Tn is residually nuclear.
Then the completely positive quotient map [Sn ]J : B/J → B/J is given by [Sn ]J =
[βn+1 ] ◦ ([Tn ]J ⊗ ιn ) ◦ [βn ], where [Tn ]J : B/J → B/J is nuclear and βn is an
isomorphism from B/J onto (B/J) ⊗ M2n . Thus, Sn : B → B is residually nuclear.
 3. THE RESIDUALLY NUCLEAR CASE 813

The inductive limit X of the residually nuclear contractions Tn : B → B is a

nuclear C *-system by Lemma A.19.6, because residually nuclear maps in particular
are nuclear. 

Lemma 6.3.15. Suppose that B is a stable and σ-unital C*-algebra. Let d ∈

M(B) with kdk ≤ 1. Then there is a norm-continuous path t ∈ [0, ∞) 7→ s(t) ∈
M(B) into the isometries of M(B), such that limt→∞ s(t)∗ bs(t) = d∗ bd for all
b ∈ B.

Proof. By Lemma 5.1.2(iv), there exist norm-continuous maps t ∈ [0, ∞) 7→

T0 (t), T1 (t) ∈ M(B) with T0 (t)∗ T0 (t) = T1 (t)∗ T1 (t) = T0 (t)T0 (t)∗ +T1 (t)T1 (t)∗ = 1,
and limt→∞ Tk (t)∗ b = kb for b ∈ B, k = 0, 1. Let s(t) := T0 (t)(1−d∗ d)1/2 +T1 (t)d.
Then t 7→ s(t) is norm-continuous and limt→∞ s(t)∗ b = d∗ b . Since B = BB, it
follows limt→∞ s(t)∗ bs(t) = d∗ bd . 

Next has to be checked again ??

Lemma 6.3.16. Suppose that B a C*-algebra and that s1 , s2 , . . . ∈ M(B) is

a sequence of isometries in M(B). Let Sn (b) := s∗n bsn for b ∈ B and n ∈ N,
tn := s1 s2 · · · sn , qn := 1 − tn t∗n , and denote by E the closure of n qn Bqn .
S

Then E is a hereditary C*-subalgebra of B and there is a natural completely

positive and completely isometric map V from B//E onto indlim(Sn : B → B) ⊂
B∞ that satisfies

V (πBE+EB (J)) = V (B//E) ∩ J∞ for J ∈ I(B) .

Proof. Since qn ≤ qn+1 , the set E is the hereditary C *-subalgebra of B

S
generated by n qn Bqn .
Let pn := 1 − qn = tn t∗n , Rn (b) := pn bpn , πBE+EB : B → B//E the quotient
map, X := indlim(Sn : B → B), Tn (b) := t∗n btn and ρn (b) := tn bt∗n for b ∈ B.
Notice that Tn ◦ ρn = idB and Tk+n ◦ ρk−1 = Sn ◦ Sn−1 ◦ · · · ◦ Sk .
Let π∞ (b1 , b2 , . . .) := (b1 , b2 , . . .) + c0 (B). We define completely positive con-
tractions T∞ : B∞ → B∞ and ρ∞ : B∞ → B∞ by

T∞ (π∞ (b1 , b2 , . . .)) := π∞ (T1 (b1 ), T2 (b2 ), . . .)

ρ∞ ((b1 , b2 , . . .)) := π∞ (ρ1 (b1 ), ρ2 (b2 ), . . .) .

Notice T∞ ◦ ρ∞ = id on B∞ , and that T∞ (π∞ (b1 , b2 , . . .)) = t∗∞ π∞ (b1 , b2 , . . .)t∞

for the isometry

t∞ := π∞ (t1 , t2 , . . .) ∈ M(B)∞ ⊂ M(B∞ )

with range projection

p∞ := t∞ t∗∞ = π∞ (p1 , p2 , . . .) ∈ M(B)∞ .

We introduce a “diagonal” map ∆ : B → B∞ by ∆(b) := π∞ (b, b, . . .).

814 6. EXACT SUBALGEBRAS OF PURELY INFINITE ALGEBRAS

The C *-space B//E is isomorphic to Y := indlim(Rn : B → B), because

Rm Rn = Rn Rm = Rmax(m,n) (coming from pn+1 pn = pn+1 ) implies that a com-
pletely positive and isometric isomorphism from B//E onto

Y := indlimn (Rn : B → B) ⊂ B∞

is given by

b ∈ B 7→ P∞ (b) := π∞ ((p1 bp1 , p2 bp2 , . . .)) = p∞ ∆(b)p∞ ∈ B∞ ,

i.e., for y = [yij ] ∈ Mk (B//E) ∼ = Mk (B)//Mk (E) and any b := [bij ] ∈ Mk (B) with
π(bij ) = yij holds that k[P∞ (bij )]k = k[yij ]k for the norms in Mk (B∞ ) ∼
= (Mk (B))∞
respectively in Mk (B//E), and [P∞ (bij )] is positive if and only if y is positive in
Mk (B//E). This is because


k[yij ]k = dist [bij ], Mn (BE + EB) = lim sup k[pn bij pn ]k .
n

The argument for the complete positivity of the class-map η := [P∞ ] : B//E → Y
is similar. The completely isometric and completely positive isomorphism η from
B//E onto Y is determined by η ◦ πBE+EB = P∞ .
It follows that η(πBE+EB (J)) = P∞ (J) ⊆ J∞ ∩ Y for J ∈ I(B). The inclusion
“⊇” needs a less simple argument: If π∞ (p1 ap1 , p2 ap2 , . . .) ∈ J∞ , then there is a
bounded sequence b1 , b2 , . . . ∈ J with limn→∞ kpn apn −pn bn pn k = 0. It follows that
πBE+EB (a) = limn πBE+EB (pn bn pn ) in B//E, i.e., πBE+EB (a) is in the closure of
πBE+EB (J). But it turns out that J ∩ (BE + EB) = J(E ∩ J) + (E ∩ J)J and
that the natural map from J//(E ∩ J) to B//E is (completely) isometric. This
can be seen from the second conjugate spaces which are naturally isomorphic to
(1 − qE )B ∗∗ (1 − qE )qJ and to (1 − qE )B ∗∗ (1 − qE ). Here we denote by qE and qJ
the open support projection of E and J in B ∗∗ . This implies together that

η(πBE+EB (J)) = P∞ (J) = J∞ ∩ Y = (p∞ J∞ p∞ ) ∩ Y .

Since t∗n tn = pn , we get – for the isometry t∞ := π∞ (t1 , t2 , . . .) ∈ M(B)∞ ⊆

M(B∞ ) – that the restriction to Y of the c.p. contraction

T∞ (π∞ (b1 , b2 , . . .)) := t∗∞ π∞ (b1 , b2 , . . .)t∞

is a completely positive complete isometry from Y onto the image of the map
S1,∞ : B → X ⊂ B∞ .
In fact, S1,∞ (b) = T∞ (∆(b)) = T∞ (P∞ (b)). The defining maps Sk,∞ : B →
X ⊂ B∞ for the inductive limit X = indlimn (Sn : B → B) are given by

Sk,∞ (b) := π∞ (0, . . . , 0, Sk (b), Sk+1 (Sk (b)), . . .) .

(It is not important if one takes zero at places 1 to k − 1 or takes other elements
S
of B.) The inductive limit X = indlimn (Sn : B → B) is the closure of k Sk,∞ (B)
in B∞ . But, since S1,∞ = Sk,∞ ◦ Sk−1 ◦ · · · ◦ S1 , Sk,∞ = S1,∞ ◦ ρk−1 , and
ρn (B) = qn Bqn , it follows that

Sk,∞ (B) = S0,∞ (B) = {T∞ (∆(b)) ; b ∈ B} = T∞ (Y )

 3. THE RESIDUALLY NUCLEAR CASE 815

are the same linear subspace of B∞ . The image T∞ (Y ) = S1,∞ (B) of the (not nec-
essarily unital) C *-system Y ⊆ p∞ B∞ p∞ for the projection p∞ := π∞ (p1 , p2 , . . .) ∈
M(B)∞ ⊆ M(B∞ ) is closed, because the restriction of ρ∞ ◦ T∞ to p∞ B∞ p∞ is
the identity map, i.e., T∞ |p∞ B∞ p∞ must be isometric on Y .
The map T∞ ◦ η : B//E → X is induced from T∞ ◦ η ◦ πBE+EB = T∞ ◦ P∞ .
Since T∞ (P∞ (B)) = T∞ (Y ) = X, it follows that T∞ ◦ η = [T∞ ◦ P∞ ] : B//E → X
defines a complete isometry from B//E onto X.
Thus, ρ∞ |X is a completely positive and completely isometric map from
∼
X = B//E onto Y , and the map (ρ∞ |X)−1 is given by the class-map V :=
[S1,∞ ] : B//E → X with V ◦ πBE+EB = S1,∞ = T∞ ◦ ∆.
Since Tn (pn Jpn ) = Tn (J) = t∗n Jtn = J and ρn (J) = pn Jpn = J ∩ pn Bpn , we
get for all J ∈ I(B) that

T∞ (p∞ J∞ p∞ ) = T∞ (J∞ ) = J∞

and
ρ∞ (J∞ ) = p∞ J∞ p∞ = J∞ ∩ p∞ B∞ p∞ .

η(B//E) ∩ π(J)∞ = ρ∞ (X ∩ J∞ ) = P∞ (J) == η(πBE+EB (J)) .

It follows that V := [T∞ ◦ ∆] = T∞ ◦ η : B//E → X = indlimn (Sn : B → B) is a
complete isometry from B//E onto X and satisfies

V (πBE+EB (J)) = J∞ ∩ X = indlimn (Sn : J → J) .

Proposition 6.3.17. Suppose that B is separable and stable and has the WvN-
property, and that Tn : B → B is a sequence of residually nuclear completely positive
contractions. Consider X := indlim(Tn : B → B) naturally as operator subspace of
B∞ := `∞ (B)/c0 (B).
Then X is a nuclear subspace of B∞ and there exist a residually essential
hereditary C*-subalgebra D of B and a completely isometric and completely positive
isomorphism I from B//D onto X = indlim(Tn : B → B), such that

I(πBD+DB (J)) = J∞ ∩ X

for every closed ideal J of B.

Proof. The inductive limit X is nuclear by Lemma 6.3.14(iii), because the

maps Tn are residually nuclear contractions. Since B has the WvN-property, the
maps Tn are in the point-norm closure of the set of one-step inner maps b 7→ c∗ bc.
By Lemma 3.1.8 the Tn are in the point-norm closure of the contractive inner
c.p. maps b 7→ d∗ bd, i.e., with kdk ≤ 1. By Lemma 6.3.15, the maps b 7→ d∗ bd with
kdk ≤ 1 can be approximated in point-norm topology by maps b 7→ t∗ bt with an
isometry t ∈ M(B), because B is stable and σ-unital.
816 6. EXACT SUBALGEBRAS OF PURELY INFINITE ALGEBRAS

Now we can apply Lemma A.14.2 to the sequence of maps T1 , T2 , . . . and the
set of maps b 7→ t∗ bt with isometries t ∈ M(B), because B is separable. We get
that are isomeries tn ∈ M(B) such that X = indlim(Ln : B → B) for the maps
Ln (b) := t∗n btn .
Then, by Lemma 6.3.16, there is a hereditary C *-subalgebra F of B and com-
pletely isometric and completely positive isomorphism ϕ from B//F onto X with
ϕ(πBF +F B (J)) = J∞ ∩ X ⊂ B∞ for J ∈ I(B). By Lemma 6.3.14 there are residu-
ally nuclear contractions Sn : B → B such that there is a completely isometric and
completely positive isomorphism ψ from Y := indlim(Sn : B → B) onto X ⊗ M2∞
with ψ(J∞ ∩ Y ) = (J∞ ∩ X) ⊗ M2∞ for J ∈ I(B). In particular,

Y ∼
= (B//F ) ⊗ M2∞ ∼
= (B ⊗ M2∞ )//(F ⊗ M2∞ ) ,

If we repeat the above arguments with Sn (in place of Tn ), then we get that there
is a hereditary C *-subalgebra E ⊂ B and a completely isometric and completely
positive isomorphism χ from B//E onto Y such that χ(πBE+EB (J)) = J∞ ∩ Y . If
ρ : M2∞ → C is a pure state and V : X ⊗M2∞ → X ⊂ B∞ is the completely positive
contraction with V (c ⊗ d) = ρ(c)d, then there is a hereditary C *-subalgebra D of B
with E ⊂ D ⊂ B such that BD + DB is the kernel of the completely positive map
W := V ◦ψ◦χ◦πBE+EB : B → X from B into X. Let I := [W ]BD+DB : B//D → X.
By Lemma 6.3.13, D is residually essential in B and I is a completely isometric
and completely positive isomorphism from B//D onto X. If J is a closed ideal of
B then, for J ∈ I(B) ,




IπBD+DB (J) = V ψ χ (πBE+EB (J)) = V ψ(Y ∩ J∞ )


= V (X ∩ J∞ ) ⊗ M2∞ = X ∩ J∞ .

Proof of Theorem 6.3.1: Ad(I): Let ak be a dense sequence in the unit-ball

of A. There are completely positive maps Vn : M(B) → B and Wn : B → M(B)
that satisfy conditions (b) and (c) of Theorem 6.3.1 and

kWn Vn (aj ) − aj k < 2−n for 1 ≤ j ≤ n .

Then A ⊆ M(B) ⊂ M(B)∞ := `∞ (M(B))/c0 (M(B)) is naturally contained in

Y := indlim(Sn : M(B) → M(B)) ⊂ J∞

with Sn := Wn Vn . Indeed, the compositions of the upper rows Sn of the commu-

tative diagram

M(B) / M(B) / M(B) / ...

< <
V1 W1 V2 W2
" "
B /B / ...

converge asymptotically to idA on A ⊂ M(B) ⊂ M(B)∞ , in the sense, that

Sn,∞ : M(B) → M(B)∞ converges in point-norm on A to idA : A → M(B)∞ ,
 3. THE RESIDUALLY NUCLEAR CASE 817

where
Sn,∞ (b) := (0, . . . , 0, b, Sn (b), Sn+1 Sn (b), . . .) + c0 (B) .
Q
The product maps n≥1 Vn and (b1 , b2 , . . . ) 7→ (0, W1 (b1 ), W2 (b2 ), . . . ) de-
fine complete contractions V∞ from M(B)∞ into B∞ and σ ◦ W∞ from B∞ into
M(B)∞ . Let Tn := Vn+1 Wn : B → B. Then Tn is a residually nuclear contraction
by construction of Vn and Wn , indeed: Ue : B 3 b 7→ Vn+1 (eWn (b)e) ∈ B is a
residually nuclear map by conditions (a),(b) and (c) for every e ∈ B+ , and, since
Vn+1 strictly continuous as a map from M(B) into B (with norm topology), Tn is
the point-norm limit of {Ueτ } if eτ ⊂ B+ is an approximate unit of B. V∞ maps
Y into
X = indlim(Tn : B → B) ⊂ B∞
and σW∞ maps X into Y .
We have W∞ ◦ V∞ ◦ σ|Y = (S∞ ◦ σ)|Y = idY because S∞ ◦ σ ◦ Sm,∞ = Sm,∞ .
Clearly V∞ ◦ σ ◦ W∞ ) = T∞ . Since σ ◦ T∞ |X = idX , we get σV∞ ◦ (σW∞ )|X = idX .
If follows that σW∞ |X is a completely isometric and completely positive map from
X onto Y .
Let us consider the actions Ψ1 (J) = J∞ ⊂ B∞ and Ψ2 (J) = M(B, J)∞ ⊂
M(B)∞ for J ∈ I(B) of Prim(B) on B∞ and M(B, J)∞ . Since Wn and Vn are
weakly residually equivariant, we get

V∞ (Ψ2 (J)) ⊂ Ψ1 (J) and

σW∞ (Ψ1 (J)) ⊂ Ψ2 (J).

Thus V∞ (Ψ2 (J) ∩ Y ) ⊂ Ψ1 (J) ∩ X and

σW∞ (Ψ1 (J) ∩ X) ⊂ Ψ2 (J) ∩ Y.

Since (V∞ |Y )−1 = (σW∞ |X), we get σW∞ (Ψ1 (J) ∩ X) = Ψ2 (J) ∩ Y . By Propo-
sition 6.3.17 there is a residually essential hereditary C *-subalgebra D of B and a
∼
completely isometric and completely positive map T : B//D → X ⊂ B∞ , such that
BDB = B and
T (πD (J)) = Ψ1 (J) ∩ X
for J ∈ I(B). Let V := (σ ◦ W∞ ) ◦ T , C := J∞ and ΨC (J) := Ψ2 (J) for J in
I(B).
to be filled in! ??
Then Lemma 6.3.8 and Lemma A.19.8 define a *-monomorphism λ from A into
M(B//D) ∩ (B//D) = π1 (N (D)) such that λ = V −1 |A and that λ(A ∩ M(B, J)) =
λ(A) ∩ π1 (N (D) ∩ J), because V −1 (A ∩ M(B, J)) = V −1 (A ∩ ΨC (J)) = V −1 (A) ∩
π1 (J).
The operator space X is nuclear, because Tn : B → B is nuclear for every n ∈ N.
Thus B//D ∼= X is nuclear, and the assumptions of Proposition 6.3.11 are satisfied
for B and D ⊂ B, and A ⊂ M(B), λ : A ,→ π1 (N (D)) = M(B//D) ∩ (B//D)
818 6. EXACT SUBALGEBRAS OF PURELY INFINITE ALGEBRAS

are as in part (iii) of Proposition 6.3.11. By 6.3.11(iii) we get the class [H] ∈
Extnuc (Prim B, A, B).
Now suppose that A ∼= E ⊗O2 or B ∼= E ⊗O2 for a stable separable C *-algebras
E. Then we can use the natural isomorphisms
Extnuc (Prim(B); E ⊗ O2 , B) ∼
= KKnuc (Prim(B); SE ⊗ O2 , B)
13
(for separable stable B with WvN-property ) respectively
Extnuc (Prim(B); A, E ⊗ O2 ) ∼
= KKnuc (Prim(E) : SA, E ⊗ O2 )
cf. Chapter 8. The left sides of the above isomorphisms are zero because
KKnuc (Prim(B), ·, ·) is homotopy invariant in each variable and id is unitarily
homotopic to id ⊕ id for all C *-algebras ∼
= C ⊗ O2 .
Hence [H] = 0 in this cases, and Proposition 6.3.11(iv) gives the desired non-
degenerate nuclear *-monomorphism h : A ,→ B with h(A ∩ M(B, J)) = h(A) ∩ J.
δ∞ ◦ h is unitarily homotopic to δ∞ |A by Corollary ??.
Ad(II): Suppose, that we can find the Wn such that Wn (B) ⊂ A (in addition).
Then Sn (A) = Wn Vn (A) ⊂ A , A = indlim(Sn : A → A) ⊂ A∞ ⊂ B∞ , Sm,∞ (B) =
Sm+1,∞ (Sn (B)) ⊂ Sm+1,∞ (A) ⊂ A . It follows that Y = A in B∞ , because Y is
T
the closure of m Sm,∞ (B).
Thus, λ = V −1 : A ,→ πD (N (D)) is an isomorphism from A onto B//D and
πD (N (D)) = B//D, i.e., B = N (D) + BD + DB. In particular, A ∼ = B//D is
nuclear.
If [H] = 0 in Extnuc (Prim(D), A, D) ∼
= Extnuc (Prim(B), A, B), then again we
find a non-degenerate *-monomorphism H3 : A → N (D) ⊂ B with πD (H3 (a)) =
λ(a) such that
H3 (A ∩ M(B, J)) = H3 (A) ∩ J
(cf. proof of Proposition 6.3.11(iv)). As in the proof of Proposition 6.3.11(iv), it
follows that H3 (A) generates B as a two-sided closed ideal. The hereditary C *-
subalgebra D1 := H3 (A)BH3 (A) of B is stable.
There is a *-isomorphism ϕ from B onto D1 that is unitarily homotopic to idB
by 5.5.6. In particular, ϕ(J) = D1 ∩ J for J ∈ I(B). Let πD : B → B//D = λ(A) =
πD (H3 (A)) the natural quotient map. Then P1 := H3 λ−1 πD satisfies P1 H3 = H3
and P1 (B) ⊂ H3 (A).
It follows P1 (D1 ) ⊂ D1 . Thus, P := ϕ−1 ◦ P1 ◦ ϕ and h := ϕ−1 ◦ H3 are as
desired.
The extremality of the conditional expectation P : B → h(A) follows as in
Remark 6.2.2. 

13 Recall that our nuclear Ext-groups have rather strong equivalence relations that require

e.g. that the extensions are stable as C *-algebras.

 CHAPTER 7

Groups of asymptotic morphisms R(C; A, B)

We consider here “nuclear” (respectively “residually nuclear” , respectively

“C-compatible”) asymptotic morphisms. Their unitary equivalence classes modulo
C0 (R+ , B) form a semigroup SR(A, B) (respectively SR(X; A, B), SR(C; A, B))
if the multiplier algebra of B contains a copy of O2 unitally. The corresponding
Grothendieck group R(A, B) (respectively R(X; A, B), R(C; A, B)) is a continu-
ous analog of Rørdam’s group of asymptotic morphisms over N, cf. [677], [679],
[680] [681]. This groups are the mediators between the semi-groups [Hom(C; A, B)]
of unitary equivalence classes of the under Cuntz-addition invariant morphisms in
Hom(A, B) ∩ C and the C-equivariant KK-groups KK(C; A, B). If this is an isomor-
phism then KK(C; ·, ·)-equivalence implies Morita equivalence of A and B. It means
that we “reduce” some classification problems to questions on the “realization” of
KK-theory equivalence by isomorphisms.
The main tools are continuous modifications and generalizations of the Rørdam
groups R(A, B). They play a role in several generalizations, e.g. to the non-simple
case as described above, or others, mentioned in the remarks of this Chapter. Some
of the known stable invariants for simple separable nuclear C *-algebras are the same
for R(., .)-equivalent C *-algebras A and B (e.g. they are KK-equivalent, F (A) and
F (B) contain, up to isomorphisms, the same simple separable C *-subalgebras, the
stable rank, the real rank and decomposition rank are the same). A part of further
study on classification could be the study of the consequences of R(., .)-equivalence.
??

1. The semigroups SR(C; A, B)

By CP(A, B) ⊂ L(A, B) we denote the matrix operator-convex cone of the

completely positive contractions from a C *-algebra A into a C *-algebra B.
(We shall later require in addition that A and B are stable, that A is separable
and that B is σ-unital.)

Definition 7.1.1. Let C ⊂ CP (A, B) denote a point-norm closed matricial

operator-convex sub-cone of the cone of completely positive maps from A to B.
Suppose that V : [0, ∞) 3 t 7→ V (t) ∈ C ⊂ CP(A, B) is a point-norm continuous
map (i.e., is continuous with respect to the strong operator topology on L(A, B)).
The map V is called an asymptotic C-morphism from A to B if V satisfies the
following conditions (a), (b) and (c):
819
820 7. GROUPS OF ASYMPTOTIC MORPHISMS R(C; A, B)

(a) For every t ∈ R+ , V (t) : A → B has norm kV (t)k ≤ 1.

(b) The V (t) are asymptotically multiplicative , i.e.,

lim kV (t)(a∗ a) − V (t)(a)∗ V (t)(a)k = 0

t→∞

for every a ∈ A.
(c) for every a ∈ A+ and ε > 0 there exist b1 , . . . , bn ∈ B+ and c1 , . . . , cn ∈
Pn
Cb (R+ , B) such that lim supn→∞ kV (t)(a) − i=1 ci (t)∗ bi ci (t)k ≤ ε ( 1 ).

Compare next blue with def. of unitary homotopy in Definition

5.0.1 !!
Let V1 , V2 asymptotic C-morphisms. We call V1 unitarily homotopic to V2
if there is a map u : t ∈ R+ 7→ u(t) ∈ M(B) such that u(t) is unitary, is continu-
ous with respect to the strict topology on M(B) ( Is the footnote really true
?? Likely not. No proof exists so far. 2 ), and lim ku(t)∗ V1 (t)(a)u(t) −
V2 (t)(a)k = 0 for every a in A.
Suppose that M(B) contains a copy of O2 unitally (i.e., there are isometries
s1 , s2 ∈ M(B) with s1 s∗1 + s2 s∗2 = 1). It is easy to check that the Cuntz sum

(V1 ⊕ V2 )(t) := V1 (t) ⊕s1 ,s2 V2 (t) := s1 V1 (t)(·)s∗1 + s1 V1 (t)(·)s∗1

defines again an asymptotic C-morphism t 7→ (V1 ⊕ V2 )(t) from A to B.

The properties of Cuntz addition (cf. Chapter 4) show that the equivalence
classes [V ] of unitarily homotopic morphisms V form A to B build a semigroup
under Cuntz addition. This semigroup and its Grothendieck group will be denoted
by SR(C; A, B), respectively by R(C; A, B).
If C = CPnuc (A, B) is the cone of all nuclear maps from A to B the we shall
write SR(A, B) for SR(CPnuc (A, B); A, B) and call SR(A, B) the Rørdam semi-
group. Its Grothendieck group Gr(SR(A, B)) will be denoted by R(A, B) and call
it the Rørdam group.
Suppose that X is a T0 space and that ΨA : O(X) → I(A) and ΨB : O(X) →
I(B) are actions of X on A respectively B. Let CX ⊆ CP (A, B) denote the cone
of ΨA -ΨB -residually nuclear maps from A into B. We call the asymptotic CX -
morphisms residually nuclear asymptotic morphisms.
Then we write SR(X; A, B) and R(X; A, B) in place of SR(CX ; A, B) respec-
tively of R(CX ; A, B) . We call R(X; A, B) a Ψ-equivariant Rørdam group.

Clearly, all the definition depend from the actions ΨA and ΨB and not from
X alone. The point is that they all come from before chosen non-degenerate point-
norm closed matrix operator-convex cone C ⊆ CP(A, B).

1 Property (c) means, equivalently, that the morphism h : A → Q(R , B) corresponding

V +
to V : R+ → C with properties (a) and (b) maps A into the ideal of Q(R+ , B) generated by
B ⊂ Q(R+ , B).
2 The map t 7→ u(t) can be chosen norm-continuous if B is σ-unital and stable and A is

separable.
 1. THE SEMIGROUPS SR(C; A, B) 821

Remark 7.1.2. If t → V (t) ∈ C ⊆ CP(A, B) is an asymptotic C-morphism,

then V (a)(t) := V (t)(a) defines an element V (a) ∈ Cb (R+ , B) and

hV : a ∈ A 7→ V (a) + C0 (R+ , B)

is a C *-morphism from A into Q(R+ , B) := Cb (R+ , B)/ C0 (R+ , B).

If hV is unitarily equivalent to a C *-morphism k : A → B ⊂ Q(R+ , B) by a
unitary in M(Q(R+ , B)) then t 7→ V (t) and hV satisfies in particular the following
additional continuity conditions:

(d) For all lower semi-continuous 2-quasi-traces τ1 , τ2 : Q(R+ , B)+ → R+ ∪

{+∞} = [0, ∞] holds τ1 ◦ hV = τ2 ◦ hV if τ1 |B+ = τ2 |B+ .
(e) Let τγ , ρ : Q(R+ , B)+ → [0, ∞] denote lower semi-continuous 2-quasi-
traces on Q(R+ , B)+ , where the family of quasi-traces {τγ } are indexed
by elements γ is an element of a directed net Γ.
If {τγ ; γ ∈ Γ} and ρ satisfy that

sup lim τγ ((b − δ)+ ) = ρ(b)

δ>0 γ∈Γ

for all b ∈ B+ then we require here the condition of the map V that hV
satisfies in addition that

sup lim τγ (hV ((a − δ)+ )) = ρ(hV (a))

δ>0 γ∈Γ

for all b ∈ A+ .

We call the asymptotic C-morphism V continuous if it satisfies (in addition)

the above conditions (d) and (e). One can see that the continuous asymptotic
C-morphism V define a sub-semigroup SRc (C; A, B) of of SR(C; A, B).
The condition (d) implies in particular that

hV (A) ∩ I1 = hV (A) ∩ I2 if I1 ∩ B = I2 ∩ B

for all closed ideals I1 , I2 of Q(R+ , B), if we consider the lower semi-continuous
traces τk : Q(R+ , B)+ → {0, ∞} , k = 1, 2, given by τk (f ) := 0 if f ∈ (Ik )+ and
by τk (f ) := ∞ if f ∈ Q(R+ , B) \ Ik . Thus, conditions (a), (b) and (d) imply the
following properties (f) and (g). It shows that the conditions (a,b,d,e) together
imply moreover the following property (h):

(f) For every a ∈ A and every closed ideal J of B there exists

lim kV (t)(a) + Jk ,
t→∞

and Ψ(J) := (hV )−1 (hV (A) ∩ Q(R+ , J)) defines a monotone-continuous
action of Prim(B) on A in the sense of Definition 1.2.6.
(g) τV (a) := supδ>0 limt→∞ τ ((V (t)(a)−δ)+ ) ∈ [0, ∞] exists for every a ∈ A+
and for every lower semi-continuous 2-quasi-trace τ : B+ → [0, ∞].
The map τV : A+ → [0, ∞] is necessarily a lower semi-continuous 2-quasi-
trace.
822 7. GROUPS OF ASYMPTOTIC MORPHISMS R(C; A, B)

(h) The map τ 7→ τV is continuous from the set of l.s.c. 2-quasi-traces on B+

to those on A+ with respect to the topology of point-wise convergence
( 3 ).

Remark 7.1.3. Below we will show that the map t 7→ u(t) ∈ M(B) can be cho-
sen norm-continuous if asymptotic C-morphism V1 and V2 are unitarily homotopic
and if B is σ-unital and stable, cf. ??.
On the other hand, it is easy to check that V 0 (t)(a) := u(t)∗ V (t)(a)u(t) sat-
isfies conditions (a)-(c) for C = CPnuc (A, B), if t 7→ V (t) ∈ L(A, B) defines a
weakly nuclear contraction V : A → Cb (R+ , B) ⊆ M(C0 (R+ , B)) from A into
M(C0 (R+ , B)) such that hV := πC0 (R+ ,B) ◦ V is a C *-morphism from A into
M(C0 (R+ , B))/ C0 (R+ , B), and if t 7→ u(t) is a strictly continuous map from R+
to M(B).
It follows, that, equivalently, we can consider SR(A, B) as a the sub-semigroup
of the semigroup of unitary equivalence classes of those C *-morphisms h from A
into

Q(R+ , B) := Cb (R+ , B)/ C0 (R+ , B) ⊆ M(C0 (R+ , B))/ C0 (R+ , B)

that have a completely positive contractive lift V : A → Cb (R+ , B) which satisfies

the conditions (a)-(c).
By Lemma ??,
the unitaries for the unitary equivalences can be chosen in any of the following
unital algebras

Cb (R+ , M(B))/ C0 (R+ , B) ⊆ M(C0 (R+ , B))/ C0 (R+ , B) → M(Q(R+ , B)) .

Remark 7.1.4. The conditions (c), (d) and (e) are redundant in some cases,
e.g. if A has real rank zero or B is a pi-sun algebra. But (d) and (e) do not imply
(c) in general:
K is the only simple C *-algebra that has an asymptotic morphism into K with
(c).
For example let us consider only the requirements (a)–(d) for B = K :
A non-zero, simple, and separable C *–algebra A must be stably isomorphic to K
or is stably projection-less, if A admits a non-zero C *-morphism h : A : Q(R+ , K).
(In particular, that K is the only simple separable stable algebra which is SR-
equivalent to K.)
The closed ideal J of Q(R+ ; K) that is generated by K is Morita-equivalent to
the commutative algebra Q(R+ , C) = C(βR+ \ R+ ). In particular, every separable
C *-subalgebra of J is of type I, in particular it is very different from the huge
algebra Q(R+ ; K), that has every separable C *-algebra as a sub-quotient.

3 Notice that the l.s.c. quasi-traces build a compact set with topology of point-wise conver-

gence, because [0, ∞] is compact.

 1. THE SEMIGROUPS SR(C; A, B) 823

On the other hand, the cone of every separable C *-algebra is a C *-subalgebra

of Q(R+ , K). In particular, C0 (R+ , O2 ) ⊂ Q(R+ , K), and thus, (a), (b), (d), (e) (f),
(g), (h) are satisfied for a c.c.p. lift V of the inclusion map, but (c) is not satisfied.
If B is stably isomorphic to a unital C *-algebra B st (its “Cuntz standard
form”) which contains unitally a copy of O2 then B has no semi-finite lower semi-
continuous traces and then (c), (f), (g) and (e) follow from (a)-(b), because then
T + (B) = 0 and every b ∈ Cb (R+ , B) is in the ideal which is generated by B:
Indeed, by Lemma ??, there exists a projection p ∈ Cb (R+ , O2 ⊗ K) with b − pbp ∈
C0 (R+ , B). Thus pb − b ∈ C0 (R+ , B), and p(t) = u(t)p0 u(t) for p0 ∈ O2 ⊗ K ⊆ B.
.............................................................................
Is above OK? Compare Chp. 9 ??
.............................................................................
But here we add a few comments on the general case (and possible generaliza-
tions):
Since (b) implies that hV : a ∈ A 7→ V (a) + C0 (R+ , B) ∈ Q(R+ , B) is a C *-
morphism and that for every ω ∈ (βR+ \ R+ ) πω hV : A → Cb (R+ , B)/Jω is a
C *-morphism (f) means just that πω hV is a C *-morphism with the same kernel as
hV , i.e. khV (a)kω = kV (a)kω = dist(V (a), C0 (R+ , B)) for every ω in the corona
of R+ . E.g. if A is simple then h is zero or πω hV is a monomorphism for every
ω ∈ βR+ \ R+ .
If A has real rank zero then again (c), (d), (e), (g) and (h) follow from (a),
(b) and (f) since every projection of Q(R+ , B ⊗ K) is unitarily equivalent to a
projection in B ⊗ K by a unitary in the unitization of Q(R+ , B ⊗ K) and ideals
of A are determined by its projections. For RR(A) = 0 the topology on T + (A) is
induced by the pairing with K0 (A) thus (h) is satisfied. The same happens with
the space of 2-quasi-traces, which gives (d) and (e) from (a),(b) and (f).
The further challenge of the theory of (our type of) asymptotic morphisms are
variants of property (f). Here a lot of stronger or weaker versions are possible which
will be discussed somewhere else.

Next Cor. cited in Chapter 10.

Corollary 7.1.5. Suppose that A and B are C*-algebras, where A is separable

and B is a σ-unital and stable.
Let ϕ : A → Q(R+ , B) and ψ : A → Q(R+ , B) C*-morphisms that are unitarily
equivalent by a unitary W ∈ M(Q(R+ , B)).
Then ϕ and ψ are unitarily equivalent by a unitary U ∈ Q(R+ , M(B)) that is
a finite product of exponentials exp(Tj ), j = 1, . . . , n, where Tj ∈ Q(R+ , M(B))
with kTj k < 1, Tj∗ = −Tj .

Desired: Tj ∈ Q(R+ , B)
824 7. GROUPS OF ASYMPTOTIC MORPHISMS R(C; A, B)

Proof. Let W ∈ M(Q(R+ , B)) with W ∗ ϕ(·)W = ψ(·).

Let a0 ∈ A+ a strictly positive contraction for A. Then ϕ(a0 ) + ψ(a0 ) is a
strictly positive element of C ∗ (ϕ(A) ∪ ψ(A)). There exists contractions e0 , e1 , e2 ∈
Cb (R+ , B)+ such that, ei ej = ej ei and e1 (1 − e0 ), e2 (1 − e1 ) ∈ C0 (R+ , B), e3 :=
e2 + C0 (R+ , B) ∈ Q(R+ , B) satisfies e3 (ϕ(a0 ) + ψ(a0 )) = ϕ(a0 ) + ψ(a0 ).
It implies that e3 ϕ(·) = ϕ(·) = ϕ(·)e3 and e3 ψ(·) = ψ(·) = ψ(·)e3 . Let f :=
e3 W e3 . Then f ∈ Q(R+ , B) is a contraction with f ∗ ϕ(·)f = ψ(·) and f ψ(·)f ∗ =
ϕ(·). It follows that (1 − f f ∗ )ϕ(·)f f ∗ = 0 and ????
We can lift f to a contraction g ∈ Cb (R+ , B) with f = g + C0 (R+ , B) and
e1 g = g = ge1 .
Since B is stable and σ-unital, we get from
Lemma ??
isometries S, T ∈ Cb (R+ , M(B)) with SS ∗ + T T ∗ = 1 and (1 − S)e0 , e0 (1 −
S) ∈ C0 (R+ , B). It gives T ∗ e0 = 0 = e0 T . Let U0 := U (g) the Halmos unitary
of g with respect to (S, T ), cf. cf. Remark ??. Then U0 ∈ Cb (R+ , M(B)) and
πB (U0 (t)) = T S ∗ − ST ∗ , i.e., U0 ∈ T S ∗ − ST ∗ + Cb (R+ , B). We get U0∗ e1 =
U0∗ e0 e1 = Sge0 + T (1 − g ∗ g)S ∗ e0 = Sg = Se0 g = g,
Need U0∗ e1 = g ???? 

2. The asymptotic corona Q(X, A)

Let us fix some notations: Throughout this section X denotes a locally compact
Hausdorff space. In our later applications we use only closed subsets X of the plane
R×R ∼ = C that are not necessarily bounded.
Notice that a locally compact space X is a σ-compact space if and only if the
C *-algebra C0 (X) contains a strictly positive element e ∈ C0 (X) : Let X1 ⊆ X2 ⊆
. . . be a sequence of open subsets of X with compact closures such that the closure
S
of Xn is contained in Xn+1 and X = Xn .
By Urysohn Lemma, we find continuous functions µn ∈ C0 (X)+ with kµn k = 1,
P∞
µn |Xn = 1, µn |(X \ Xn+1 ) = 0. Then µn µn+1 = µn and e(x) := n=1 2−n µn ∈
C0 (X)+ , and 1 ≥ e(x) > 0 for every x ∈ X. Thus, C0 (X) is σ-unital with strictly
positive element e if X is σ-compact.
Conversely, suppose that the C *-algebra C0 (X) is σ-unital. There exists con-
tinuous function e ∈ C0 (X)+ that is a strictly positive element of norm one in
the C *-algebra C0 (X). The sets Un := e−1 ((n−1 , 1]) build a sequence of open
S
subsets of X with X = Xn , the closure of Xn is contained in compact set
Cn := e−1 ([(n + 1)−1 , 1]). Thus X is σ-compact.
We consider from now on only σ-compact l.c. spaces X.
Let Cb (X) denote the C ∗ -algebra of bounded continuous functions on X. By
βX we denote the Stone-Čech compactification of X, the corona γ(X) :=
 2. THE ASYMPTOTIC CORONA Q(X, A) 825

β(X) \ X consists of abstract limit points of the free ultra-filters on X. β(X) and
γ(X) are compact and X is a dense open subset of βX.
In terms of C ∗ -algebras we have natural isomorphisms Cb (X) ∼ = C(βX) ∼ =
∼ ∼
M(C0 (X)) and, therefore, C((βX \ X)) = Cb (X)/ C0 (X) = M(C0 (X))/ C0 (X).
The free ultra-filters on X (i.e. the points of the corona of X) correspond one-
to-one to the characters of Cb (X) which annihilate C0 (X), i.e. to the characters
of Cb (X)/ C0 (X). In the sequel we consider frequently every continuous bounded
function on X also as a continuous function on βX.
It follows that γ(R) = βR\R is the disjoint union of the open connected subsets
γ(R− ) and γ(R+ ), where R− = (−∞, 0] and R+ = [0, +∞).
More simply expressed:
If f is a bounded continuous function on R− and g a bounded continuous function on
R+ , then h(t) := f (t) for t ≤ −1, j(t) := g(t) for t ≥ 0, h(t) := |t|f (−1)+(1−|t|)g(0)
for t ∈ [−1, 0] is a bounded continuous function with h|γ(R− ) = f |γ(R− ) and
h|γ(R+ ) = g|γ(R+ ).
Now let A be a Banach space. By Cb (X, A) (resp. C0 (X, A)) we denote the Ba-
nach space of bounded continuous functions f : X → A with kf k = sup{kf (x)k : x ∈
X} (resp. continuous functions f : X → A vanishing at infinity, i.e., on γ(X)).
If A is a C *-algebra, then C0 (X, A) is an ideal of the C *-algebra Cb (X, A),
Cb (X) acts on Cb (X, A) by (gf )(x) = g(x)f (x) for f ∈ Cb (X, A) and g ∈ Cb (X).

Definition 7.2.1. Let f ∈ Cb (X, A). We denote by N (f ) ∈ Cb (X) the

continuous function N (f )(x) := kf (x)k.
We call the Banach space Q(X, A) := Cb (X, A)/ C0 (X, A) the asymptotic
corona of A with respect to X, and Q(R+ , A) the asymptotic corona of A.
If Y is a closed subset of β(X) then

JY := {f ∈ Cb (X, A) : N (f )(y) = 0 ∀y ∈ Y },

where we consider N (f ) as a function in C(βX). In particular, JX = C0 (X, A).

More generally, we define restrictions Cb (X, A)|Y of Cb (X, A) to closed sub-
sets Y of βX by
Cb (X, A)|Y := Cb (X, A)/JY .

In particular, Q(X, A) = Cb (X, A)|γ(X) . We write also Q(X, A)|Y := Cb (X, A)|Y
if Y ⊆ γ(X) .

The reader can easily check that:

(i) N (f ) : x 7→ kf (x)k is in Cb (X) ∼

= C(βX) if f ∈ Cb (X, A) and is Cb (X)+ -
homogeneous, i.e. gN (f ) = N (gf ) if f ∈ Cb (X, A), g ∈ Cb (X)+ .
(ii) C0 (X, A) = C0 (X) Cb (X, A), and, more generally,

JY = C0 ((βX) \ Y ) · Cb (X, A)
826 7. GROUPS OF ASYMPTOTIC MORPHISMS R(C; A, B)

if Y ⊆ γ(X) ( 4 ).

Part (ii) can be seen (e.g. in the case of σ-compact X) as follows: Let f ∈ JY
and let e ∈ C0 (X)+ ⊆ C(β(X)) a fixed continuous function with e(x) > 0 for
x ∈ X \ Y , e(y) = 0 for y ∈ γ(X) (i.e., take the above constructed strictly positive
element e ∈ C0 (X) of the σ-unital C *-algebra C0 (X \ Y )+ — with X replaced by
X \ Y ). C0 (X)+ . Then g = (N (f ) + e)1/2 ∈ C0 (βX \ Y ). The function f1 (x) :=
(N (f ) + δ)−1/2 (x)f (x) for x ∈ X is continuous and satisfies N (f1 ) ≤ N (f )1/2 . In
particular f1 ∈ JY and gf1 = f .

Lemma 7.2.2. Let A a C ∗ -algebra, and X a locally compact, σ-compact and

non-compact Hausdorff space.
There are following natural isomorphisms and properties:

(i) C0 (X, A) ∼
= C0 (X) ⊗ A by the a natural isomorphism that identifies f (x)a
and f ⊗ a for a ∈ A, f ∈ C0 (X). It defines a natural C*-morphism from
Cb (X) into the center of M(C0 (X, A)).
(ii) M(C0 (X, A)) is naturally isomorphic to the C*-algebra Cb,st (X, M(A))
of bounded strictly (= ∗ -strongly) continuous maps f from X into M(A).
(iii) Cb (X, A) is identical with the ideal of M(C0 (X, A)) given by

{ f ∈ M(C0 (X, A)) ; gf ∈ C0 (X, A) ∀ g ∈ C0 (X)} .

In particular, the natural *-monomorphism from Cb (X, A) extends to a

natural *-isomorphism

M(Cb (X, A)) ∼

= M(C0 (X, A)) .

(iv) Q(X, A) := Cb (X, A)/ C0 (X, A) is naturally isomorphic to an ideal of

M(C0 (X, A))/ C0 (X, A).
And M(Q(X, A)) = M(C0 (X, A))/ C0 (X, A).
The algebraic two-sided annihilator Ann(Q(X, A)) of Q(X, A) in
M(C0 (X, A))/ C0 (X, A) = M(Cb (X, A))/ C0 (X, A) is identical with

{T + C0 (X, A) ; T ∈ M(C0 (X, A)) , T f, f T ∈ C0 (X, A) ∀ f ∈ Cb (X, A)} .

It is identical with the C*-algebra C0 (X) · M(C0 (X, A)) / C0 (X, A).
(v) For f, g ∈ Q(X, A)+ , f g = 0 there exists contractions h1 , h2 ∈ Q(X, A)+
with f h1 = f , gh2 = g and h1 h2 = 0.
(vi) For f ∈ Cb (X, A ⊗ K)+ there exists an isometry s ∈ Cb (X, M(K)) ⊆
M(C0 (X, A ⊗ K)) such that f s ∈ C0 (X, A ⊗ K).
(vii) If M1 ⊂ M(C0 (X, A)) and M2 ⊆ Cb (X, A) are countable subsets then
there exist a positive contraction e ∈ Cb (X, A) such that eb − be, c − ec, c −
ce ∈ C0 (X, A) for all b ∈ M1 and c ∈ M2 .
(viii) Suppose that the C*-algebra B contains a continuous path t ∈ R+ 7→
p(t) ∈ B in the projections of B such that limt→∞ p(t)b = b for all b ∈ B.

4 The factorization holds for arbitrary l.c. X by Cohen factorization applied to the non-

degenerate Banach C0 (β(X) \ Y )-module JY .

 2. THE ASYMPTOTIC CORONA Q(X, A) 827

Let h : B → M(A) is non-degenerate C*-morphism, and

hX : Cb (X, B) → Cb,st (X, M(A)) ∼

= M(C0 (X, A))

its natural extension to Cb (X, B).

Then for every countable subset M ⊂ Cb (X, A) there exists a projec-
tion p ∈ Cb (X, B) such that (1 − hX (p))f, f (1 − hX (p)) ∈ C0 (X, A) for
all f ∈ M .

It is likely that M(Cb (X, B))/ C0 (X, B) ⊆ M(Q(X, B))

is not equal to M(Q(X, B)), e.g. if X = R+ and B is not σ-unital?

When M(A)/J ⊆ M(A/J) ??

Perhaps M(A/J) ∼
= M(A)/ N (A, J) if A and J are σ-unital ??
Notice that J := C0 (X, B) is not σ-unital if B is not σ-unital !

Proof. (i): Easy to see, e.g. [767, vol.I,chp.IV.,thm.4.14], [704, sec. 1.22].
(ii): For y ∈ X the ∗-epimorphism f ∈ C0 (X, A) 7→ f (y) ∈ A defines a
C *-morphism b ∈ M(C0 (X, A)) 7→ b(y) ∈ M(A) . If g ∈ C0 (X), a ∈ A then
(b(g ⊗ a))(y) = g(y)b(y)a ∈ C0 (X, A) . Thus y 7→ b(y) is strongly continuous. The
same happens for b∗ , y 7→ (b(y))∗ = b∗ (y) since C0 (X) A is dense in C0 (X, A) =
C0 (X) ⊗ A, for every b ∈ M(C0 (X, A)) there exists a strictly continuous function
x 7→ b(x) with sup kb(x)k ≤ kbk such that (bf )(x) = b(x)f (x). Conversely a
bounded strictly continuous map x 7→ b(x) ∈ M(A) defines a multiplier of C0 (X, A)
(respectively of Cb (X, A)), because the A-valued functions x → b(x)a(x) and x →
a(x)b(x) are in C0 (X, A) (respectively in Cb (X, A)) if x → b(x) ∈ M(A) is bounded
and strictly continuous.
Parts (iii) and (iv) can be seen from (ii) and (i).
(v): It suffices to find a contraction h1 ∈ Q(X, A)+ with h1 f = f and h1 g = 0,
because then one can use (g, h1 ) in place of (f, g) to produce in the same way h2
with h2 h1 = 0 and and h2 g = g.
Let b = b∗ ∈ Cb (X, A) with b + C0 (X, A) = f − g. The the positive and
the negative parts of b satisfy b+ (x) = max(0, b(x)), b− (x) = max(0, −b(x)), b+ +
C0 (X, A) = f and b− + C0 (X, A) = g. Take a strictly positive contraction e ∈
C0 (X)+ , i.e., 1 ≥ e(x) > 0 for all x ∈ X and limx→∞ e(x) = 0.
Define h(x) := e(x)−1 min(b+ (x), e(x) · 1), i.e., h(x) = e(x)−1 b+ (x) −
e(x)−1 (b+ (x) − e(x))+ . This h is a positive contraction in Cb (X, A), and with
kb+ (x)h(x) − b+ (x)k ≤ e(x) by spectral calculus, because (b+ (x) − e(x))+ ∈
C ∗ (b+ (x)) ⊆ A, x 7→ (b+ (x) − e(x))+ is continuous and t(1 − ε−1 min(t, ε)) ≤ ε
for all t ≥ 0 and ε > 0. Right from the definition we get b− (x)h(x) = 0. Thus
h1 := h + C0 (X, A) ∈ Q(X, A) has the desired properties.
(vi): Combine proof with (viii). New text in Lemma:
828 7. GROUPS OF ASYMPTOTIC MORPHISMS R(C; A, B)

For f ∈ Cb (X, A ⊗ K)+ there exists an isometry s ∈ Cb (X, M(K)) ⊆

M(C0 (X, A ⊗ K)) such that f s ∈ C0 (X, A ⊗ K).
Old proof: Let f ∈ Cb (X, A ⊗ K) and X1 ⊆ X2 ⊆ · · · a sequence of compact
S
subsets of X such that Xn is contained in the interior of Xn+1 and n Xn = X. We
find projections p1 ≤ p2 ≤ . . . ∈ K such that kf (x)1 ⊗ pn − f (x)k < n1 for x ∈ Xn .
Thus it suffices to find a strictly continuous map x 7→ s(x) ∈ L(H) = M(K) such
that s(x) is an isometry and kpn s(x)k ≤ n1 for x in the closure of Xn+1 \ Xn . Then
x 7→ 1 ⊗ s(x) is as desired.
For X = R+ and A := C this can be solved by Lemma 5.1.2(iv) with Xn :=
[0, n]: We find a norm-continuous map t ∈ R+ 7→ s0 (t) into the isometries of M(A)
such that kpn s0 (t)k ≤ n1 for t ∈ [n, n + 1].
For general X and A let s(x) := s0 (t(x)), where t : X → R+ is a continuous
P
function with t(X \ Xn ) ⊆ (n, ∞), e.g. t(x) := n (1 − gn (x)) with a continuous
functions gn : X → [0, 1] with gn |Xn−1 = 1 and gn |(X \ Xn◦ ) = 0.
(vii): stated text:
If M1 ⊆ M(C0 (X, A)) and M2 ⊆ Cb (X, A) countable subsets. Then there exist
a positive contraction e ∈ Cb (X, A) such that eb − be, c − ec, c − ce ∈ C0 (X, A) for
all b ∈ M1 and c ∈ M2 .
(viii): stated text:
Suppose that the C *-algebra B contains a continuous path t ∈ R+ 7→ p(t) ∈ B
in the projections of B such that limt→∞ p(t)b = b for all b ∈ B.
If h : B → M(A) is a non-degenerate C *-morphism, and hX : Cb (X, B) →
Cb,st (X, M(A)) ∼
= M(C0 (X, A)).
Then for every countable subset M ⊆ Cb (X, A) there exists a projection p ∈
Cb (X, B) such that (1 − hX (p))f, f (1 − hX (p)) ∈ C0 (X, A) for all f ∈ M .
To be filled in ?? 

Remark 7.2.3. The following example shows that in general Cb (X, A) is

strictly contained in the C *-algebra of the bounded strictly continuous functions
f ∈ Cb,st (X, M(A)) with f (X) ⊆ A, i.e., a bounded strictly continuous function
f on X with values in A is not necessarily continuous. Let X := {0} ∪ {1/n : n ∈
N} ⊆ [0, 1]. A := K and f (0) := 0, f (1/n) = pnn .
Then X is compact and limn→∞ pnn converges strictly to zero in M(A). Thus,
f ∈ Cb,st (X, M(A)). Clearly, f 6∈ C(X, M(A)), because certainly f 6∈ C(X, A).
Thus, f is in Cb,st (X, M(A)) ∼
= M(C(X, A)), has image in A, but f is not in
Cb (X, A) = C(X, A).

Lemma 7.2.4. Let (Ω, ρ) be a compact metric space, X a locally compact σ-

compact space and F a set of bounded nonnegative continuous functions on Ω × X,
such that

(i) |f (ω1 , x) − f (ω2 , x)| ≤ Cρ(ω1 , ω2 ) for all x ∈ X, f ∈ F and ω1 , ω2 ∈ Ω,

 2. THE ASYMPTOTIC CORONA Q(X, A) 829

(ii) For every bounded sequence f1 , f2 , . . . ∈ F and every countable cover-

ing {Un } of X by open subsets with compact closures (in X), there ex-
ist f ∈ F and a locally finite decomposition λ1 , λ2 , . . . ∈ C0 (X)+ of
1, such that for every k there exists nk with support λk ⊆ Unk and
P
f (ω, x) ≤ λk (x)fnk (ω, x) for x ∈ X.

Let f ∗ (x) := sup{f (ω, x) : ω ∈ Ω}, g(x) := inf{f ∗ (x) : f ∈ F }, h a bounded

continuous function on X such that g(x) < h(x) for every x ∈ X and Y ⊂ βX \ X
a compact subset of the corona.
Then there exist k1 , k2 ∈ F with k1∗ ≤ h, and k2∗ (y) ≤ sup{g(y) : y ∈ Y } for
every y ∈ Y .
It holds sup{g(x) : x ∈ βX \ X} ≤ limn→∞ (sup{g(x) : x ∈ X \ Xn }) where
Xn ⊆ Xn+1 ⊆ . . . X are open subsets of X with compact closures such that X =
S
Xn .

Remark 7.2.5. Here we consider only locally finite decompositions of 1 which

are finite in a sufficiently small neighborhood every point of X, i.e. for every x ∈ X
there is neighborhood U of x such that only a finite number of the λk is nonzero
on U .

Proof. By the Lipschitz-condition (i), the f ∗ are bounded continuous func-

tions on X. F ∗ = {f ∗ : f ∈ F } satisfies (i) and (ii) for Ω = point and F replaced by
F ∗ . The conclusions do not require Ω. Therefore it is enough to prove the lemma
in case of Ω = point, f = f ∗ ∀f ∈ F . Since g(x) < h(x) for x ∈ X we find for
every x ∈ X an f ∈ F and a neighborhood U (x) of x such that f (y) < h(y) for
every y ∈ U (x) and U (x) has compact closure. Since X is locally compact we find
S
a countable subsystem (fn , U (Xn )) with U (Xn ) = X. If (λk ) is the locally finite
decomposition of unit subordinated to {U (xn )} such that support of λk is contained
P
in some U (xnk ) and such that there is k1 ∈ F with k ≤ λk fnk , then g ≤ k1 ≤ h.
If Xn ⊆ Xn+1 ⊆ . . . X is a covering of X by open subsets with compact closure, the
a suitable construction of h ≥ g shows that sup{g(x) : x ∈ βX \ X} ≤ lim sup Cn
where Cn = sup{g(x) : x ∈ Xn+2 \ Xn }. Now consider Y ⊆ βX \ X, compact,
C = sup{g(y) : y ∈ Y }.
Then for every y ∈ Y , n ∈ N there exists f ∈ F ⊆ Cb (Y ) ∼
= C(βX), such that
f (y) < C + n1 . Thus for every point y ∈ Y there exists a neighborhood U (y) of
y ∈ βX and f ∈ F ⊆ C(βX) with f (x) ≤ C + n1 for x ∈ U (y).
Since Y is compact we find a y1 , . . . , yn ∈ Y , f1 , . . . , fn ∈ F with fk (x) ≤ C + n1
for x ∈ U (yk ) and Y ⊆ U (y1 ) ∪ . . . ∪ U (yn ). In this way we find a sequence
g1 , g2 , . . . ∈ F such that inf{gk (y) : k = 1, 2, . . .} ≤ C for y ∈ Y . Let δ ∈ C0 (X)+
with δ(x) > 0 for every x ∈ X (exists because X is σ-compact). By the first part
of the proof it is enough to find a bounded continuous function k with g ≤ k ≤
inf{gk }+δ on X. Because then there is k1 ∈ F with k1 ≤ k+δ which implies k1 (y) ≤
P
gn (y) for y ∈ βX \X and n = 1, 2, . . . . Let k = µ1 g1 + (µn+1 −µn )·inf{g1 , . . . , yn }
830 7. GROUPS OF ASYMPTOTIC MORPHISMS R(C; A, B)

where µn (x) ∈ [0, 1] for x ∈ Xn , µn (x) = 1 on Xn and µn (x) = 0 on X \ Xn+1 (use

Urysohn Lemma). 

Remark 7.2.6. If X = N and ω ∈ γ(N) := βN \ N is a point of the corona

γ(N) of N the ω is called a ”free” ultrafilter. Bω = Cb (N, B)/J{ω} = l∞ (B)/cω (B)
is then the well known (norm-)ultrapower of a Banach space B.
If X is locally compact, non-compact but σ-compact (e.g. X = R+ ) and ω ∈
βX \ X then in general there does not (!) exist a sequence x1 , x2 , . . . ∈ X and a
(free) ultrafilter ω1 ∈ βN \ N such that ω = λ(ω) (where λ : βN → βN is induced by
the map n 7→ xn ). Therefore in the following proof one needs really our reduction
to the bounded continuous functions on X (and not only ultrapower arguments or
pure logical partition arguments).
Another technical problem is the following: If Ω is a compact metric space
that contain infinitely many points, then the natural continuous epimorphism from
β(R+ × Ω) onto β(R+ ) × Ω is not an isomorphism. Thus Cb (R+ , C(Ω, A)) 
C(Ω, Cb (R+ , A)) in general. But always Cb (R+ , C(Ω, A)) = Cb (R+ × Ω, A).
.............................................................................
Suppose that B is a Banach space, Z ⊆ βX, f ∈ Cb (X, B). then
N (f )(y) := kf (y)k defines an element N (f ) of C(βX) ∼ = Cb (X). There-
fore kf kY := sup{N (f )(y) : y ∈ Y } is well defined for Y ⊆ βX closed and
kf kY = kπY (f )k = kf + JY k = dist(f, JY ).
Suppose, A, B are Banach spaces, T is a set of linear contractions from A into
B. T X denotes the set of strongly continuous maps from X into T . The elements of
T X operate on Cb (X, B) as linear contractions, indeed: if L := {S(y)}y∈X ∈ T X ,
f ∈ Cb (X, B) then g(y) := S(y)(f (y)) defines an element L(f ) = g ∈ Cb (X, B).
Let Ω be a compact metric space, γ1 : Ω → Cb (X, A) and Z ⊆ βX a subset
of βX. For fixed Ω, γ1 and γ2 , we define for Z ⊆ βX closed, L ∈ T X µ(L, Z) :=
sup{kL(γ1 (t)) − γ2 (t)kZ : t ∈ Ω} ν(Z) := inf{µ(L, Z) : L ∈ T X }.

With the above assumptions and notations we have:

Proposition 7.2.7. Assume that X is σ-compact and T is convex or that X

is totally disconnected. Then

(i) ν(Z) ≤ sup{ν({ω}) : ω ∈ Z}

(ii) ν(βX \ X) ≤ limn→∞ (sup{ν({y}) : y ∈ X \ Xn })
(iii) If Z ⊆ βX \ X is closed then there exists L ∈ T X with ν(Z) = µ(L, Z),
i.e. the infimum is attained.
(iv) ν(X) ≤ sup{ν({y}) : y ∈ X}.

Proof. The functions

g(y, t) := kL(γ1 (t)) − γ2 (t)ky = kL(y)(γ1 (t)(y)) − γ2 (t)(y)k,

 2. THE ASYMPTOTIC CORONA Q(X, A) 831

y ∈ X, t ∈ Ω satisfy the assumptions of Lemma ?? where we have to introduce

on Ω the metric ρ(t1 , t2 ) = ρΩ (t1 , t2 ) + kγ1 (t1 ) − γ1 (t2 )k + kγ2 (t1 ) − γ2 (t2 )k. In
particular the family {g(y, t)} is σ − C0 (X)-subconvex in the variable y ∈ X.
But then (i)-(iii) follows from Lemma ??, where for the proof of (iii) one has
to use the family max{0, g(y, t) − ν(Z)}, which again satisfies the assumptions of
Lemma ??. 

Corollary 7.2.8. Suppose that A and B are C*-algebras, that X is σ-compact

and that T is a set of c.p. contractions from A into B.
Then, for each ω ∈ β(X), T X defines a set of completely positive contractions
X
T |ω from Aω := Cb (X, A)/Jω into Bω := Cb (X, B)/Jω , given by the natural map
T X 7→ T X |Y from Cb,st (X, L(A, B)) in L(AY , BY ) for closed Y ⊆ β(X) \ X.
The restriction of T X |Y to each separable closed subspace C of AY is point-
norm closed in L(AY , BY ).
If T ⊆ CP(A, B) is convex, (respectively inner-invariant, operator-convex or is
a matrix operator-convex cone), then T X |Y is so as subset of CP(AY , BY ).
T X |Y satisfies the assumptions (i)-(iii) of Lemma ??.
???????
??
Transitivity needed???:

If T is convex, and for every a ∈ A+ and b ∈ B+ with kak = kbk = 1 and ε > 0
there exists V ∈ T with kV (a) − bk < ε, then T X |Y has this property for a ∈ (AY )+
and b ∈ (BY )+ with kaω k = kbω k for all ω ∈ Y .
In particular, if C ⊆ Aω is a separable C*-subalgebra and V : C → Bω is a
nuclear contraction, then there exists an L ∈ T X such that for V πω (a) = πω (L(a))
if a ∈ Cb (X, A) and πω (a) ∈ C.

Corollary 7.2.9. Let Y ⊆ βX\X be a compact subset, C a separable subspace

of Cb (X, A)/JY and S : C → Cb (X, B)/JY a linear map.
Suppose that for each ω ∈ Y , ε > 0 and every finite subset F ⊆ C there exists
a nuclear completely positive contraction Vω : πω (C) → Bω such that
kVω (πω (c)) − πω (S(c))k < ε
for all c ∈ F .
Then there exists a strongly continuous map L : x ∈ X →7 L(x) ∈ L(A, B) into
the nuclear completely positive contractions from A into B such that
S(πY (a)) = πY (L(a)) ∀ a ∈ πY−1 (C) .

Proof. Let Ω be a compact subset of C that generates C as a closed linear

subspace. and let γ1 : Ω → Cb (X, A) a topological lift, i.e., πY γ1 = idΩ . Let
γ2 : Ω → Cb (X, B) be a topological lift of (S|Ω) : Ω → Cb (X, B)/JY . Moreover,
832 7. GROUPS OF ASYMPTOTIC MORPHISMS R(C; A, B)

let T be the convex set of nuclear completely positive contractions from A into
B. Define ν(L, Y ) and µ(Y ) as explained above. We have to find L ∈ T X with
ν(L, Y ) = 0 because then L is as desired (since πY L and S are contractions).
By Proposition ?? (iii) and (i) it suffices to show that ν({y}) = 0 for each
y ∈ Y . But that we have seen above in Corollary ??. 

Now we get our asymptotic Weyl–von Neumann version of the Weyl–von

Neumann–Voiculescu–Kasparov theorem:

Theorem 7.2.10. Let X be a locally compact σ-compact space, Y a closed

subset of the corona βX \ X of X, B a C*-algebra and A a purely infinite simple
C*-subalgebra of B such that AB is dense in B. Moreover let C be a separable
C*-subalgebra of πY (Cb (X, A)) = Cb (X, A)/JY .
If V is a completely positive contraction from C into πY (Cb (X, B)) such that
for each ω ∈ Y there exists a nuclear completely positive contraction Vω : πω (a) =
πω (V (a)) for every a ∈ C, then there exists a contraction d ∈ Cb (X, B) such that
V (a) = πY (d)∗ aπY (d) for every a ∈ C.

Proof. The Theorem is the logical sum of Corollary ?? and Proposition ??.


Lemma 7.2.11. If a ∈ πY (Cb (X, A))+ satisfies kπω (a)k = 1 for every ω ∈ Y ,
then there exists b ∈ Cb (X, A)+ such that kπx (b)k = 1 for every x ∈ X and
a = πY (b).

Proof. Let d ∈ A+ with kdk = 1 and let c ∈ Cb (X, A)+ be a contractive

lift of a, i.e. πy (c) = a, c ≥ 0, kc(x)k = kπx (c)k ≤ 1 for every x ∈ X. Now let
g(x) := kc(x)k then g ∈ C(βX) ∼ = Cb (X), f (x) := c(x) + (1 − g(x))d is in Cb (X, A)
and kf k ≤ 1 and 1/2 ≤ max(g(x), 1 − g(x)) ≤ kf (x)k for x ∈ X. Then πy (f ) = a
because g(ω) = 1 for ω ∈ Y . Thus kπω (f )k = 1 for ω ∈ Y , i.e. h(ω) = 1 for ω ∈ Y ,
where h(x) := kf (x)k−1 . 

Next:
Suppose Cb (X, A) is not s.p.i.

Conjecture (1):
Then there exists a separable C *-subalgebra B of A
such that B has the same invariants:

ρ(a, b; B) = ρ(a, b; A)
for a, b ∈ B+ , where

ρ(a, b; C) := inf{kd∗ a2 d − a2 k + ke∗ b2 e − b2 k + kd∗ abek ; e, d ∈ C}

for a, b ∈ C.
 2. THE ASYMPTOTIC CORONA Q(X, A) 833

Conjecture (2):
Let A a C *-algebra. Then
for every separable C *-subalgebra B of M(A)
there exists a contraction d ∈ A+ with
BD ⊆ D for D := dAd
and kbk = supn kbd1/n k for all b ∈ B
such that
ρ(a, b; M(A)) = ρ(λ(a), λ(b); M(D))

for the faithful C *-morphism λ : B → M(dAd)

given by λ(b)x := bx for x ∈ D and b ∈ B

The σ-unitality of A and the σ-compactness

of X (used below) can be reduced if Conj. (1) and (2) hold.

If A is strongly p.i. and σ-unital and X is σ-compact,

then M(C0 (X, A)) is s.p.i. and above defined Aω strongly p.i.

This special case follows also from:

‘‘B s.p.i.’’ and ‘‘C exact’’ imply ‘‘B ⊗ C s.p.i.’’ and

‘‘M(b∗ Bb) is s.p.i.’’ for each b ∈ B
‘‘A s.p.i.’’ implies ‘‘A ⊗ C0 (X) = C0 (X, A) s.p.i.’’
implies ‘‘M(C0 (X, A)) s.p.i.’’ if X is σ-compact and A σ-unital

Cb (X, A) is ideal of Cb,st (X, M(A)) =∼ M(C0 (X, A)),

because f ∈ Cb (X, A), if and only if,
f ∈ Cb,st (X, M(A)) and gf ∈ C0 (X, A) for all
g ∈ C0 (X) · 1 ⊆ center of M(C0 (X, A)).

Property ‘‘s.p.i.’’ passes to ideals and quotients.

Corollary 7.2.12. If A is simple and purely infinite then every fiber Aω =

Cb (X, A)/Jω of Cb (X, A) (ω ∈ βX) is simple and purely infinite.
In particular, Cb (X, A) is a continuous field over β(X) with simple purely
infinite fibres Ax = A for x ∈ X and Aω for ω ∈ β(X) \ X.

Proof. If ω ∈ X then Jω is the kernel of the epimorphism πω : Cb (X, A) → A.

If ω ∈ β(X) \ X, then Jω = C0 (β(X) \ {ω}) · Cb (X, A) , because f ∈ Jω if
and only if N (f ) ∈ Cb (X) ∼
= Cb (βX) for the N (f )(x) := kf (x)k has the property
N (f )(ω) = 0.
Consider for a, b in (Aω )+ , kak = kbk = 1 the nuclear map V : c 7→ ρ(c)b where
ρ is a pure state on Aω with ρ(a) = 1.
834 7. GROUPS OF ASYMPTOTIC MORPHISMS R(C; A, B)

By Lemma 7.2.11 and Corollary 7.2.9 there exist elements c, d ∈ Cb (X, A) with
πω (c) = a, πω (d) = b, kc(x)k = kd(x)k = 1 for all x ∈ X.
Need reduction to σ-compact X !
if not supposed before.
For each compact subset K ⊆ X and ε > 0 there is a contraction f = f (K, ε) ∈
C(K, A) with kf (x)∗ c(x)f (x) − d(x)k < ε for all x ∈ K
???????????? (cf. Corollary ??).
Since X is σ-compact, we find continuous functions g0 = 0 and gn : X → [0, 1]
with supports in compact sets Kn of X such that gn+1 (x) = 1 for x ∈ Kn and
X = Kn . Let G1 (x) := n (g2n (x) − g2n−1 (x))1/2 f (K2n , 1/n)(x) and G2 (x) :=
S P
1/2
f (K2n+2 , 1/n)(x). Then Gj ∈ Cb (X, A) and d − G∗1 cG1 +
P
n (g2n+1 (x) − g2n (x))
G∗2 cG2 ∈ C0 (X, A). It follows that y1∗ ay1 + y2∗ ay2 = b for the contractions yj :=
πω (Gj ). In particular, Aω is simple.
Since A is naturally contained in Aω and since A is purely infinite, Aω is
not isomorphic to the compact operators on a Hilbert space. Thus Aω satisfies
the assumptions of part (iv) of Proposition ??, which implies that Aω is purely
infinite. 

Start of collections for Chapter 7

The *-epimorphism b 7→ b(t), for b ∈ C([0, 1], B), extends naturally to a *-
epimorphism πt : Cb (R+ × [0, 1], B) → Cb (R+ , B) by

πt ((bs )s∈R+ ) := (bs (t))s∈R+ .

The πt define evaluation semi-group morphisms

(πt )∗ : SR(C[0, 1]; A, B[0, 1]) → SR(C; A, B)

that naturally define group morphisms

(πt )∗ : R(C[0, 1]; A, B[0, 1]) → R(C; A, B) .

The following proposition confirms the homotopy invariance of the Rørdam groups.

Proposition 7.2.13. If B is stable and σ-unital h : A → Q(R+ , B[0, 1]), sat-

isfies [h] ∈ SR(C[0, 1]; A, B[0, 1]), then [h0 ] = [h1 ] in R(C; A, B), where [ht ] :=
(πt )∗ [h] ∈ R(C; A, B) is defined by the evaluation semi-group morphism

(πt )∗ : R(C[0, 1]; A, B[0, 1]) → R(C; A, B) .

Lemma 7.2.14. Suppose that X is a σ-compact locally compact space, B is

a stable and σ-unital algebra, and that D ⊆ M(B) is non-degenerate (i.e. DB is
dense in B).

(i) For every σ-unital C*-subalgebra A of Q(X, B) there exists an contraction

f ∈ Q(X, B)+ such that f a = af = a for every a ∈ A.
(ii) For each g ∈ Cb (X, B) there exists a contraction e ∈ Cb (X, D)+ with
g − eg, g − ge ∈ C0 (X, D) .
 2. THE ASYMPTOTIC CORONA Q(X, A) 835

(iii) For f ∈ Q(X, B) there exists an isometry T in Cb (X, M(B))/ C0 (X, B)

such that T ∗ f T = 0.


(iv) Each C*-morphism h : D → Q(X, B) dominates zero in Q X, M(B)
(and thus in Cb (X, M(B))/ C0 (X, B) ⊆ EX ).

Recall that Q(X, B) ⊆ Cb (X, M(B))/ C0 (X, B) ⊆ EX . Since the annihilator

of Q(X, B) in Cb (X, M(B))/ C0 (X, B) is just C0 (X, M(B))/ C0 (X, B), we have
also that Q(X, B) is naturally isomorphic to an ideal of Q(X, M(B)).

Proof. to be filled in ?? 

Important !
Observation:
For all C *-algebras B and l.c. Hausdorff spaces holds
M(C0 (X, B)) = M(Cb (X, B)) .
(1) C0 (X, B) ideal of Cb (X, B), f g = 0 for all g ∈ C0 (X, B) implies that f = 0.
And Cb (X, B) is non-degenerate C *-subalgebra of M(C0 (X, B)).
Induces faithful unital embedding of M(Cb (X, B)) into M(C0 (X, B)).
(2) Need that Cb (X, B) is ideal of M(C0 (X, B)) (then we are ready).
If f ∈ Cb (X, B), T ∈ M(C0 (X, B)) and ψ ∈ C0 (X) then T.(ψ.f ) ∈ C0 (X, B)
and ψ.(T.f ) = T.(ψ.f ). Thus, ψT f ∈ C0 (X, B) for all ψ ∈ C0 (X, B).
g ∈ Cb,st (X, M(B)) ∼ = M(C0 (X, B)) is in Cb (X, B) if and only if ψ · g ∈
C0 (X, B) for all ψ ∈ C0 (X).

Proposition 7.2.15. Suppose that A and B are stable and σ-unital, and that
X is a σ-compact, locally compact, non-compact Hausdorff space.
Then all “natural” definitions of inner equivalence of C*-morphisms hi : A →
Q(X, B), i = 1, 2, define the same classes, i.e., precisely that the following proper-
ties (i)–(v) of (h1 , h2 ) are all equivalent:

(i) There exists a unitary u1 in the multiplier algebra M(Q(X, B)) of Q(X, B)
such that u∗1 h1 (·)u1 = h2 , i.e., u∗1 h1 (a)u1 = h2 (a) for every a ∈ A.
(ii) There exists a unitary u2 in EX = M(C0 (X, B))/ C0 (X, B) =
M(Cb (X, B))/ C0 (X, B) such that u∗2 h1 (·)u2 = h2 for every a ∈ A.
(iii) There exists a unitary V3 in the C*-algebra Cb (X, M(B)) of bounded
norm-continuous functions on X with values in M(B) such that
u∗3 h1 (·)u3 = h2 for the unitary
u3 := πC0 (X,B) (V3 ) = V3 + C0 (X, B)
in Cb (X, M(B))/ C0 (X, B) ⊆ EX .
(iv) If Y is a σ-compact locally compact Hausdorff space such that the corona
β(X) \ X of X is an open subset of β(Y ) \ Y , then there exists an unitary
u4 ∈ EY such that u∗4 h1 (·)u4 = h2 , where we embed Q(X, B) naturally as
an ideal of EY .
836 7. GROUPS OF ASYMPTOTIC MORPHISMS R(C; A, B)

(v) There exists a contraction e in Q(X, B) such that e∗ h1 (·)e = h2 and

eh2 (·)e∗ = h1 .

In particular, we get for every C *-morphism h0 : A → Q(X, B) that the semi-

groups S h0 ; A, Q(X, M(B)) S h0 ; A, M(Q(X, B)) , and S(h0 ; A, EX ) become
naturally isomorphic semigroups if consider Q(X, B) as ideals in Q(X, M(B)) and
in EX .
If especially X := R+ and Y := R, then β(R+ ) \ R+ is naturally homeomorphic
to an open subset of β(R)\R ∼
U
= (β(R− )\R− ) β(R+ )\R+ . Let h0 : A → Q(R+ , B)


a C *-morphism with [h0 ] = [h0 ] + [h0 ] in Hom(A, ER+ ). Then S h0 ; A, ER ,



S h0 ; A, ER+ and S h0 ; A, Q(R+ , M(B)) are naturally isomorphic semigroups.

Proof. (i)⇒(v): Let f ∈ A+ a strictly positive contraction. The element

h0 (f ) + h1 (f ) is a strictly positive element of C ∗ (h0 (A) ∪ h1 (A)) ⊆ Q(X, B). Thus,
by Lemma 7.2.14(i) there exists a positive contraction C ∈ Cb (X, B) with cg = g =
gc for c := C + C0 (X, B) and g ∈ h0 (A) ∪ h1 (A). Let e := cu1 c.
(iii)⇒(ii): Use that Cb (X, B) ⊆ Cb (X, M(B)) ⊆ M(Cb (X, B)). The restric-
tion of the *-monomorphism

M(Cb (X, B)) → M(Q(X, B))

to the embedding of Cb (X, B) in Cb (X, M(B)) induces the quotient map

Cb (X, B) → Q(X, B).
(ii)⇒(i): Consider the C *-algebras

Cb (X, B) ⊆ Cb (X, M(B)) ⊆ Cb,st (X, M(B)) ∼

= M(C0 (X, B)) .

The C *-algebra Cb (X, B) is an ideal of M(C0 (X, B)), because f ∈ Cb (X, B) if

and only if ψ ·f ∈ C0 (X) for all functions ψ ∈ C0 (X) ∼
= C0 (X)1 ⊆ Cb,st (X, M(B)).
The induced C *-morphism from M(C0 (X, B)) into M(Cb (X, B)) is faith-
ful and unital and fixes Cb (X, B). It allows to identify the multiplier algebras
M(C0 (X, B)) and M(Cb (X, B)) naturally, as

Cb (X, B) ⊆ Cb,st (X, B) = M(C0 (X, B)) = M(Cb (X, B)) .

Since Q(X, B) = Cb (X, B)/ C0 (X, B), we get also a natural unital C *-
morphism from M(Cb (X, B)) into M(Q(X, B)). The restriction of the corre-
sponding *-morphisms

M(C0 (X, B)) = M(Cb (X, B)) → M(Q(X, B))

to the embedding of Cb (X, B) in M(C0 (X, B)) induces the quotient map
Cb (X, B) → Q(X, B). Thus, define unital C *-morphisms into M(Q(X, B))
that is compatible with Cb (X, B) → Q(X, B). This shows that (i) follows from (ii)
and (iii).
(v)⇒(iii):
to be filled in ??
 2. THE ASYMPTOTIC CORONA Q(X, A) 837

(iv)⇔(i): In case Y := X it is the same as the implication (ii)⇔(i), by the

above proven equivalences of (i),(ii),(iii) and (v).
If the compact set β(X) \ X is homeomorphic to an open subset of β(Y ) \ Y ,
let Z := (β(Y ) \ Y ) \ (β(X) \ X). It follows that Z is open and compact and
Q(Y, B) = Q(Y, B)|Z ⊕ Q(X, B). Thus M(Q(X, B)) is a quotient of M(Q(Y, B)).
It shows that (iv) implies (i) in general, because Q(X, B) ⊆ Q(Y, B) ⊆ EY is an
ideal of EY , that defines the natural group morphism U(EY ) → U(Q(X, B)).
To be filled in
???? The way from (iii) to (iv) goes over the natural C *-morphisms
Cb (Y, M(B))/ C0 (Y, B) → Cb (X, M(B))/ C0 (X, B) .


Corollary 7.2.16. Suppose that A and B are stable, where B is σ-unital and
A is separable.
Let h : A → Q(R+ , B) be a C*-morphism that dominates h ⊕ h, and let V : A →
Q(R+ , B) a completely positive contraction.
Further let Ω ⊆ A a (norm-)compact subset that generates a (norm-)dense
linear subspace of A, γ0 : Ω → Cb (R+ , B) a topological lift of h|Ω and γ1 : Ω →
Cb (R+ , B) a topological lift of V |Ω.
The following are equivalent:


(i) h dominates V by an isometry S ∈ Cb R+ , M(B) .
(ii) h approximately 1-dominates V .
(iii) h dominates weakly approximately inner V (in the second conjugate of
Q(R+ , B).
(iv) h n-dominates V for some n ∈ N.


(v) There exists a contraction T ∈ M C0 (R+ , B) such that
T ∗ δ∞ γ0 (a)T − γ1 (a) ∈ C0 (R+ , B) ∀ a ∈ A.

The criteria (v) is later needed only the case of constant h : A → B.

Proof. to be filled in ?? 

Lemma 7.2.17. Let h1 and h2 be C*-morphisms from A⊗O2 into a W*-algebra

N such that h1 (a ⊗ 1) = h2 (a ⊗ 1) for every a ∈ A.
Then h1 and h2 dominate weakly approximately inner each other.

In fact, h1 and h2 are unitarily homotopic in N (in norm-topology).

Proof. Let H(a) := h1 (a ⊗ 1) and let M denote the W*-algebra generated

by H(A) with unit element P = 1M ∈ N . Then there are unital C *-morphisms
gj : O2 → P N P such that H(a)gj (b) = hj (a ⊗ b). It follows that hj (O2 ) ⊆ E :=
M 0 ∩ P N P . Notice that K1 (M 0 ∩ P N P ) = 0 as it is for all W*-algebras, because
every unitary is an exponential of a some T with T ∗ = −T and kT k ≤ π.
838 7. GROUPS OF ASYMPTOTIC MORPHISMS R(C; A, B)

Any two unital *-morphisms from O2 into a unital C *-algebra E with U(E) =
U0 (E) are point-norm homotopic. The latter implies stable unitary homotopy for
copies of O2 . Since id and δ2 are unitary homotopic in O2 , it follows that stable
homotopy of unital copies of O2 implies that the copies unitarily homotopic in E.
The path of unitaries t 7→ U (t) + (1 − p) defines a unitary homotopy from h1
to h2 .
To be filled in: Check again ?? 
Proposition 7.2.18. Suppose that A and B are stable, where B is σ-unital
and A is separable, X a non-compact locally compact σ-compact Hausdorff space,
and that h : A → Q(X, B) is a C*-morphism which is unitarily equivalent to h ⊕ h
by a unitary in M(Q(X, B)).
Then there exists a C*-morphism k : A ⊗ O2 → Q(X, B) such that

(i) k(a ⊗ 1) = h(a) for all a ∈ A, and

(ii) k is unitarily equivalent to k ⊕ k.

A C*-morphism k with properties (i) and (ii) is unique up to unitary equivalence

by a unitary in Q(X, M(B)).

Proof. to be filled in ?? 
Corollary 7.2.19. Suppose that A and B are stable, where B is σ-unital and
A is separable and exact, and that hi : A → Q(R+ , B) are nuclear C*-morphisms
which are unitarily equivalent to hi ⊕ hi , i = 1, 2.
Furthermore, suppose that, for each closed ideal J of A, h1 (J) and h2 (J) gen-
erate the same ideal of Q(R+ , B).


Then h1 and h2 are unitarily equivalent, by a unitary in Q R+ , M(B) .

Proof. To be filled in ?? 

What happens if k : A⊗O2 ,→ E exists with k((·)⊗1) = h0 (·)? Is h0 ∼

h0 ⊕ h0 in Q(R+ , M(B))?
Corollary 7.2.20. Suppose that A is stable, separable and exact, that B
is stable and σ-unital, and that C0 (R, B) satisfies the WvN-property or that B is
strongly purely infinite.
Let h0 : A → B be a nuclear non-degenerate C*-morphism from A into B such
that h0 ⊕ h0 is unitarily homotopic to h0 . And let h : A → B is a C*-morphism
that satisfies for each J ∈ I(B) that
h−1 (h(A) ∩ J) = h−1
0 (h0 (A) ∩ J) ,

i.e., that h(a) and h0 (a) generate the same closed ideal of B for each a ∈ A+ .
Then h is unitarily homotopic to h ⊕ h0 .
In particular, every *-monomorphism h : A⊗K → B ⊗K is unitarily homotopic
to h ⊕ h0 if B is simple, purely infinite and σ-unital and A is unital and separable.
 2. THE ASYMPTOTIC CORONA Q(X, A) 839

Proof. The equation shows that h(A) generates B as a closed ideal, because
h0 (A)B is dense in B. Since A and B are stable and B is σ-unital, we get from
Proposition ??
IN CHAPTERS 7 or 5 ??
that h is unitarily homotopic to a non-degenerate C *-morphism from A into B,
i.e., there is a norm-continuous path t ∈ R+ → U (t) ∈ M(B) and a non-degenerate
C *-morphism H : A → B such that limt→∞ U (t)∗ h(a)U (t) = H(a) for each a ∈ A.
to be filled in, check from here ??
Therefore, we can assume that h(A) contains a strictly positive element of B.
Since h0 ⊕ h0 is unitarily homotopic to h0 , it follows by
Proposition ??,
that it suffices to show that h asymptotically absorbs h0 , i.e. that h dominates
h0 in the multiplier algebra M(Q(R+ , B)) ∼= Cb,st (R+ , B) of Q(R+ , B).
If we want to adapt the proof of
Proposition ??,
then we have to find a sequence of contractions dn ∈ B, such that, for a ∈ A
and j ∈ {0, 1},
lim kd∗n h(a)dn+j − (1 − j)h0 (a)k = 0 .
n→∞

Since (h0 ⊕ h0 ) ◦ h−1 : h(A) → B is residually nuclear by the assumptions, and since
B is strongly p.i. (or has WvN-property), we find such a sequence,
cf. Remark ??.
Thus h absorbs h0 asymptotically and, therefore h ⊕ h0 is unitarily homotopic
to h. 

Proofs of Theorems B(iii) and M(iii): First, Theorem B(iii) is a special

case of Theorem M(iii):
we have defined h0 : A⊗K → B⊗K in Chapter 1 before Theorem B. If we replace
A and B by its stabilizations, then with N := O2 ⊗K, Prim(N ) = Prim(B) =point,
and the assumptions of Theorem M(iii) are satisfied.
We explain the assumptions of Theorem M in the special situation of part
(iii): We suppose that A is a separable, stable and exact C *-algebra, that B is
strongly purely infinite, and that N is a strongly purely infinite separable stable
C *-subalgebra of B, such that N B is dense in B and that ΨB := ΨN,B 0 is an
isomorphism from I(N ) ∼ = O(Prim(N )) onto I(B). The latter conditions imply
that B is σ-unital and stable, and that, for every I ∈ I(N ), there is a unique
J ∈ I(B) with I = J ∩ N .
Furthermore, there is given a lower semi-continuous action ΨA : O(X) → I(A)
of X := Prim(N ) on A with ΨA (∅) = 0, Ψ−1 A (A) = X, such that ΨA satisfies
condition (ii) of Definition 1.2.6.
840 7. GROUPS OF ASYMPTOTIC MORPHISMS R(C; A, B)

Here we assume the existence of a non-degenerate nuclear *-monomorphism

k : A ⊗ O2 → N such that h0 := k((.) ⊗ 1) induces ΨA by ΨA (ZI ) = h−10 (h0 (A) ∩ I)
for I ∈ I(N ) and ZI ∈ O(X) corresponding to I. The existence of h0 follows from
Theorem K, which will be shown in Chapter 12. The proof of Theorem K uses
results of Chapter 3-6, of the first part of Chapter 9 and of Chapter 7 up to
Corollary ??.
For this h0 we have h0 (ΨA (Z)) = h0 (A) ∩ ΨB (Z), because N ∩ ΨB (Z) is
the closed ideal of N which corresponds to Z ∈ O(X). If h : A → B is a *-
monomorphism, such that h(ΨA (Z)) = h(A) ∩ ΨB (Z) for Z ∈ O(X), then, for
J ∈ I(B),
h−1 (h(A) ∩ J) = h−1
0 (h0 (A) ∩ J),

because ΨB maps O(X) onto I(B).

Thus, Corollary 7.2.20 applies, and h asymptotically dominates h0 . 

3. Remarks on the case of group actions

Remark 7.3.1. Our Definition 7.1.1 of asymptotic CPnuc (A, B)-morphisms

does not work well in the case where a T0 space X acts on A and B, as considered
in Section 2 of Chapter 1, or in the case where, moreover, a locally compact group
G acts ΨA -equivariant on A (via γA : G → Aut(A)) and ΨB -equivariant on B (via
γB ). Then it is useful to replace (a) by the following (a’):

(a’) For every t ∈ R+ , every open subset Z ⊆ X and every g ∈ G, V (t) is

Ψ-residually nuclear and V (t) is asymptotically γ-equivariant.
That means:
V (t)(ΨA (Z)) ⊆ ΨB (Z) and the naturally induced completely positive
contraction [V (t)]Z : A/ΨA (Z) → B/ΨB (Z) is nuclear, and,
for every g ∈ G and a ∈ A, V (t)(γA (g)(a)) − (γB (g)V (t)(a)) converges to
zero if t → ∞.

This is condition (a) in case of the cone C of Ψ-residually nuclear maps from A to
B if the action of G is trivial on A and B.
If the actions of G are not trivial, i.e., if there exists a g ∈ G such that γA (g) 6=
id or γB (g) 6= id, then also the definition of unitary homotopy should be stronger:
We could require that g 7→ M(γB (g))(u(t)) is norm-continuous for every t and that
limt→∞ ku(t) − M(γB (g))(u(t))k = 0 in the for every g ∈ G.
One has to make some extra assumption on the actions of G on B such that
that Cuntz addition makes sense:
There should exist norm-continuous maps t 7→ s1 (t), t 7→ s2 (t) from R+ into the
isometries in M(B) such that s1 (t)s1 (t)∗ + s2 (t)s2 (t)∗ = 1, M(γB (g))(si (t)) − si (t)
is in B for every t ∈ R+ and every g ∈ G, and tends to zero if t → ∞ for every
(fixed) g ∈ G.
 4. COLLECTION OF NEEDED RESULTS. 841

The semigroup of asymptotically γ-equivariant Ψ-equivariant nuclear asymp-

totic morphisms will be denoted by SRG (X, A, B). We write RG (X, A, B) for its
Grothendieck group, and R(X, A, B) if G is trivial.

Most of the theory ??????????????

more ??? ??

4. Collection of needed results.

PLAN:
General Plan for Chp. 7:
1. Asymptotic morphisms (def. and classes)
2. lifting of nuclear maps, cases where residually nuclear maps can be lifted,
lifting by continuous families of approximately inner maps. When does a continuous
family of approximately inner maps define a “locally” approximately inner map of
Q(X, B)?. Compare also the invertibility for SExtnuc (X; A, B) in Chapter 5.
3. ”car:” Asymptotic Weyl - von Neumann theorem (for asymptotically resid-
ually nuclear maps V : C → Q(X, B))
4. Absorption theorems for h : A → Q(X, B), k : A → C(Y, B), . . .
5. Asymptotic morphism as elements of

G(h0 ; A, ER ) ∼
= G(h0 ; A, Q(R+ , M(B))) ∼
= G(h0 ; A, Cb (R+ , M(B))/ C0 (R+ , B))
for ER := M(C0 (R, B))/ C0 (R, B), A separable, B σ-unital and stable.
R(Prim(B); A, B) ∼
= G(h0 ; A, ER ) if nuclear h0 : A ,→ B with [h0 ⊕ h0 ] = [h0 ]
defines the action.
R(C(h0 ); A, B) ∼
= G(h0 ; A, ER ) if h0 = h((·) ⊗ 1) for h : A ⊗ O2 → B.
6. Existence of zero domination:
For every h : A → Q(X, B) with stable B and σ-unital A there exist a positive
contraction e ∈ Q(X, B) and isometries S, T ∈ Cb (X, M(B)) with SS ∗ + T T ∗ = 1,
η(T )∗ eη(T ) = 0, h(a) = h(a)e = eh(a) for all a ∈ A, where η : Cb (X, M(B)) →
M(Q(R+ , B))
<== here X = R+ ?
denotes the natural *-morphism induced by the natural embedding
Cb (X, M(B)) ,→ Cb,st (X, M(B)) ∼
= M(C0 (X, B)) .
In particular, h : A → Q(X, B) dominates zero in Q(X, M(B)).
7. Asymptotic unitary equivalence of ”stable” monomorphisms to ”non-
degenerate” monomorphisms (see also Chapter 5)

to be added:
842 7. GROUPS OF ASYMPTOTIC MORPHISMS R(C; A, B)

refprop:7.Y1.chp9
For a C *-algebra F , let

Q(R+ , F ) := Cb (R+ , F )/ C0 (R+ , F ) .

Proposition 7.4.1. Q(R+ , M(F )) is in a natural way a C*-subalgebra of the

multiplier algebra of Q(R+ , F ).
If F is stable or unital, then, for every separable subset Y of Q(R+ , F ) and every
unitary v in the multiplier algebra of Q(R+ , F ) there is a unitary u ∈ Q(R+ , M(F ))
with u∗ bu = v ∗ bv for every b ∈ Y .
In general (for every σ-unital F ), if A ⊆ Q(R+ , F ) is a separable C*–subalgebra
of Q(R+ , F ) and v is a unitary in the connected component of 1 in M(D) for
D := A Q(R+ , F )A, then there is a unitary u ∈ Q(R+ , M(F )) with uD + Du ⊆ D
such that v = u + Ann(A).

ref{lem:7.Y2.chp9} or ref{prop:7.Y2.chp9}

Proposition 7.4.2. Suppose that M(B) contains a copy of O2 unitally, that

A is a C*–algebra, and that h0 : A ,→ Q(R+ , B) is a *-monomorphism. If h0 ⊕ h0
is unitarily equivalent to h0 in M(Q(R+ , B)) then there is a *-monomorphism
h : A ⊗ O2 ,→ Q(R+ , B) with h(a ⊗ 1) = h0 (a) for a ∈ A such that h ⊕ h is unitarily
equivalent to h by a unitary in M(Q(R+ , B)).
If A is σ-unital and stable and k : A⊗O2 → Q(R+ , B) satisfies k(a⊗1) = h0 (a)
for a ∈ A, then k is unitarily equivalent to h by a unitary in Q(R+ , M(B)).
In particular, then h ⊕ h is unitarily equivalent to h in Q(R+ , M(B)).

For the two latter statements it is only needed that A is σ-unital and the
unitary group of D := h0 (A) Q(R+ , B)h0 (A) is connected. ¡== for unital A or ???
or:
... U(M(h0 (A) Q(R+ , B)h0 (A))) is connected ?? ??
reflem:7.Y3.chp10? also: chp9?

Corollary 7.4.3. Suppose A is stable and separable, that B is stable, and that
h0 and k0 are nuclear *-monomorphisms from A into Q(R+ , B), such that h0 and
k0 “extend” to C*-morphisms h and k from A ⊗ O2 into Q(R+ , B).
Then h and h ⊕ h (respectively k and k ⊕ k) are unitarily equivalent in
Q(R+ , M(B)).
If for every I ∈ I(Q(R+ , B)),

h−1 −1
0 (h0 (A) ∩ I) = k0 (k0 (A) ∩ I) .

Then h and k are unitarily equivalent by a unitary in Q(R+ , M(B)) .

In particular, if h, k : A ⊗ O2 → B are nuclear *-monomorphisms that induce
the same map from I(B) to I(A), then h and k are unitarily homotopic.
 4. COLLECTION OF NEEDED RESULTS. 843

ref{lem:7.??}(??)
By Lemma lem:7.??(??), ( see ref in chp 9) k0 comes from k : D ⊗O2 → B, D, B
stable, D separable , B σ-unital there exists a contraction g1 ∈ Cb (R, 1B ⊗ K)+
such that g1 (x) = 0 for x ≤ 0 and limx→+∞ g1 (x)k0 (a) = k0 (a) for every a ∈ D,
because k0 (D) is separable.
refcor:7.Y7.chp9

Corollary 7.4.4. Suppose that A is separable and stable, B is σ-unital and

stable and that h, k : A ⊗ O2 → B are *-monomorphisms. Let h0 := h((.) ⊗ 1) and
k0 := k((.) ⊗ 1).
If there are approximately inner completely positive maps Vn , Wn : B → B such
that Vn ◦ h0 converges in point-norm to k0 and Wn ◦ k0 converges in point-norm to
h0 , then h and k are unitarily homotopic.

next: reflem:7.Y3.chp9

Lemma 7.4.5. There are natural inclusions and isomorphism

M(B) ⊆ Cb (R, M(B)) ⊆ Cb,st (R, M(B)) ∼
= M(C0 (R, B))

ref, cor:7.Y1.chp11

Corollary 7.4.6. Suppose that hu0 : A → O2 is a unital *-monomorphism,

and that the evaluations πy h at y ∈ R+ × X of h : A → Cb (R+ × X, D) are unital
monomorphisms. Let π : Cb (R+ × X, D) → Q(R+ × X, D).
Then πh : A → Q(R+ × X, D) dominates hu0 : A → O2 ⊆ Q(R+ × X, D).

ref; prop:7.Y2.chp10

Proposition 7.4.7. , Suppose that A and B are stable and σ-unital. and
that a C*-morphism h1 : A → B approximately dominates a completely positive
contraction h2 : A → B, (cf. Definition 3.10.1). (i.e., there exists a sequence of
contractions dn ∈ B such that lim kh2 (a) − d∗n h1 (a)dn k = 0 for a ∈ A.)
Then there are isometries tn ∈ M(B) such that lim kh2 (a) − t∗n h1 (a)tn k = 0
for a ∈ A.

ref; cor:7.Y4.chp12

Corollary 7.4.8. , h, k : A ⊗ O2 → B nuclear, h is unitarily homotopic to k

if h−1 (J) = k −1 (J) for all closed ideals of B

ref; cor:7.Y3.chp12

Corollary 7.4.9. Suppose that B ∼

= B ⊗ O∞ ⊗ O∞ · · · and that B is stable.
Let G = Q(R+ , B) or G := Bω and A ⊆ G separable. There is a *-monomorphism
h from A ⊗ O∞ ⊗ K into G with h(a ⊗ 1 ⊗ e1,1 ) = a.

ref; cor:7.Y2.chp12
844 7. GROUPS OF ASYMPTOTIC MORPHISMS R(C; A, B)

Corollary 7.4.10. G as in Corollary 7.4.9 A ⊆ G separable. The approxi-

mately inner completely positive contractions T : D → G are one-step inner, i.e.,
there is a contraction d ∈ G such that T (b) = d∗ bd.

ref; prop:7.Y1.chp12:

Proposition 7.4.11. Suppose that A1 , A2 , . . . are separable C*-algebras with

Ak ∼
Q
= Ak ⊗ O∞ ⊗ O∞ ⊗ . . ., and let D := ω (Ak ).
Then if B ⊆ D is separable, there exists a *-monomorphism ϕ from B ⊗ O∞
into D, such that ϕ(b ⊗ 1) = b for b ∈ B.

ref; lem:7.Y6.chp9

Lemma 7.4.12. Let a = (a1 , a2 ) ∈ Q(R, B) ∼

= Q(R− , B) ⊕ Q(R+ , B) a positive
contraction, then there exist a positive contraction b = (b1 , b2 ) ∈ Q(R, B) with
b1 = 0 and b2 a2 = a2 .

Lemma 7.4.13. Every non-zero projection in Cb (R, O2 ⊗ K) is equivalent to

1 ⊗ e1,1 ∈ O2 ⊗ K by unitaries in the unitization of Cb (R, O2 ⊗ K).

Proposition 7.4.14. (Compare also 7.4.18)

Suppose that O2 ⊗ K ⊆ M(B) non-degenerate with (O2 ⊗ K) ∩ B = 0. Define F
by Cb (R, O2 ⊗ K) ∼= F ⊆ Cb (R, M(B)). For every projection q ∈ F1 ∼
= Cb (R, O2 ⊗
K), and every element f1 ∈ Cb (R, B) there exists a projection p in F1 such that
pf1 p − f1 ∈ C0 (R, B), q ≤ p and q 6= p.

ref; {prop:7.Y5.chp9}(??), {prop:7.Y5new.chp9}(??),

Proposition 7.4.15. Suppose B is stable and σ-unital, and that D separable,

stable and exact.
Let ER := M(B ⊗ C0 (R))/B ⊗ C0 (R) and h0 : D → ER defined by
...????????????? is nuclear.
??
Let k ∈ Hom(X; D, Q(R+ , B)).

(i) The map k has unitary equivalence class [k] in S(h0 , D, ER ), if an only if
k is nuclear, where Q(R+ , B) ⊆ Q(R, B) ⊆ ER naturally.
(ii) R(X; D, B) = [h0 ] + S(h0 , D, ER ) = [h0 ] + S(h0 , D, Q(R+ , M(B))).
(iii) If h is nuclear as a map from D to Q(R+ , B), and if h is also Ψ-
equivariant, then h is dominated by h0 : D → Q(R+ , B).

The point is, that nuclear asymptotically equivariant h is residually nuclear

??????

really needed in chp. 9:

Let h : D → Q(R+ , B). Suppose that HR dominates h + βh0 , i.e., there exists an
 4. COLLECTION OF NEEDED RESULTS. 845

isometry T ∈ ER with T ∗ H0 (.)T = h + βh0 . Then h ∈ S(h0 ; D, Q(R+ , M(B))),

i.e., h0 dominates h if and only if HR dominates h + βh0 .
(True for any h0 : D → B non-degenerate with h0 ⊕ h0 ∼ h0 , i.e., H0 pseudo-
dominates h, if and only if, H0 dominates h, if and only if, h0 dominates h.)
ref; prop:7.Y5.chp9

By Proposition 7.4.15, S(h0 , D, ER ) is nothing else the semigroup of unitary

equivalence classes of Ψ-residually nuclear C *-morphisms h from D into Q(R+ , B),
where ΨA (J) and ΨQ(R+ ,B) are defined as above.
next: Proof of Thm. B(iii),
also Proof of Corollary 7.4.16

Corollary 7.4.16. The absorption result (iii) of Theorem B:

Suppose that B is simple, purely infinite, unital and contains a copy of O2 ⊆ B
unitally.
Let A a unital separable exact C*–algebra and let k1 : A → O2 , k2 : O2 ⊗ O2 →
O2 unital *-monomorphisms, and k0 := k2 (k1 (·) ⊗ 1).
Then, for every unital *-monomorphism h : A → Q(R+ , B), [h ⊕ k0 ] = [h].
More generally, R(D, (O2 ⊗ O2 ⊗ · · · ) ⊗ K) = 0 for every stable separable C*-
algebra D.

Lemma 7.4.17. Let h : D → B ⊗ K a C*-morphism where D is σ-unital. Then

there exists a contraction g1 ∈ Cb (R, 1M(B) ⊗ K)+ such that g1 (x) = 0 for x ≤ 0
and limx→+∞ g1 (x)h(a) = h(a) for every a ∈ D.

(because h(D) is σ-unital)

Next proposition, see chp9, but only the prop is cited:

Let G2 := (O2 ⊗ K) ⊆ M(B). Then G2 ∼ = O2 ⊗ K, G2 ∩ B = 0 and G2 · B is
dense in B. Let F1 := Cb (R, G2 ) ⊆ M(C0 (R, B)). Certainly β̃(F1 ) ⊆ F1 .
F1 ∩ C0 (R, B) = 0 and therefore F := πSB (F1 ) ⊆ ER is a C *-subalgebra of ER
which is isomorphic to Cb (R, O2 ⊗ K) and satisfies β(F ) = F .
H00 (D)G2 ⊆ B implies that F1 H00 (D) ⊆ Cb (R, B) and thus F H0 (D) ⊆ J + βJ.

Proposition 7.4.18. For every projection q ∈ F1 , and every element f1 ∈

Cb (R, B) there exists a projection p in F1 such that pf1 p − f1 ∈ C0 (R, B), q ≤ p
and q 6= p.

Proof. of Proposition 7.4.18 ?? to be filled in ... 

Next Corollary 7.4.19 is used in Chp. 9

Corollary 7.4.19. (iii) Suppose that D is a stable and separable, that B

is stable and σ-unital. Let k : D ⊗ O2 → B a non-degenerate *-monomorphism,
846 7. GROUPS OF ASYMPTOTIC MORPHISMS R(C; A, B)

k0 := k((·) ⊗ 1) and let C ⊆ CP(D, B) denote the point-norm closed matricial

operator-convex cone generated by k0 : D → B. Then:

(i) SR(C; D, B) = S(k0 ; D, Q(R+ , M(B))) (as semi-groups). By Corollary

??,
(ii) If k1 ∈ Hom(D, B) ∩ C is non-degenerate and satisfies that k1 ⊕ k1 is
unitarily homotopic to k1 and that k1 generates C, then k1 is unitarily
homotopic to k0 .
(iii) R(C; D, B) ∼
= SR(C; D, B) + [k0 ] ⊆ SR(C; D, B) .

Lemma 7.4.20. Suppose that D is separable, B stable, and let I1 : Q(R+ , B) →

ER := M(SB)/SB the canonical embedding.
The unitary equivalence classes of k ∈ Hom(D, Q(R+ , B)) and I1 ◦ k ∈
Hom(D, ER ) naturally coincide, i.e., [k] = [k 0 ] in [Hom(D, Q(R+ , B))] if and only
if [I1 ◦ k] = [I1 ◦ k 0 ] in [Hom(D, ER )].

Proposition 7.4.21. For separable stable D and σ-unital stable B, let

I1 : Q(R+ , B) → ER := M(SB)/SB the natural embedding. Suppose that
k : D ⊗ O2 ,→ B is non-degenerate. Let k0 := k((·) ⊗ 1) and h0 := I1 ◦ k0 .
Let C ⊆ CP(A, B) denote the point-norm closed matricial operator-convex cone
generated by k0 .

(i) A morphism k ∈ Hom(D, Q(R+ , B)) has unitary equivalence class [k] ∈
SR(C; D, B), if an only if, [I1 ◦ k] is in S(h0 ; D, ER ).
(ii) [k] ∈ [k0 ] + S(k0 ; D, Q(R+ , B)) ∼ = R(C; D, B), if and only if, [I1 ◦ k] ∈
[h0 ] + S(h0 ; D, ER ) = G(h0 ; D, ER ).
In particular, [(I1 ◦ k) ⊕ h0 ] ∈ R(C; D, B) for [k] ∈ SR(C; D, B).

We have
G(k0 ; D, Q(R+ , B)) ∼
= R(C; D, B) ∼
= G(I1 ◦ k0 ; D, ER )
if C = C(k0 ) for some k0 : D → B with k0 ⊕ k0 unitarily homotopic to k0 , if D is
separable and B is stable.

Proof. Use Proposition 4.4.3 ?????. 

Next Lemma {cor:7.YZnew.chp9} is used in Chp. 9

Lemma 7.4.22. Suppose that D is stable and separable, that B is σ-unital

and stable, and that k : D ⊗ O2 → B is a non-degenerate *-monomorphism. Let
k0 := k((·) ⊗ 1) and h0 := I1 ◦ k : D → J ⊆ Q(R, B) ∼ = Q(R− , B) ⊕ Q(R+ , B)
for the natural isomorphism I1 from Q(R+ , B) onto J := π(K) for K := {f ∈
Cb (R, B) ; f (t) = 0 for t < 0} (where

π := πC0 (R,B) : M(C0 (R, B)) → ER := M(C0 (R, B))/ C0 (R, B)

and where we consider B naturally as a subset Q(R+ , B)). Further, let

β : Q(R, B) → Q(R, B) the isomorphism with β(f + C0 (R, B)) = f˜+ C0 (R, B) for
f˜(t) := f (−t), and H0 := π ◦ (δ∞ ◦ k0 ⊗ 1) : D → ER .
 4. COLLECTION OF NEEDED RESULTS. 847

(i) If h : D → Q(R+ , B) ⊆ Q(R, B) is a C*-morphism then there exists a

contraction Y ∈ Cb (R, B)+ with Y (t) = 0 for t ≤ 0 such that yhy =
y(h + βh0 )(·)y = h for y := Y + C0 (R, B) ∈ J.
(ii)
H0 (D) ⊆ π(M(B)) ⊆ π(Q(R, M(B))) ⊆ ER ,

by the natural embeddings

M(B) ⊆ Cb (R, M(B)) ⊆ Cb,st (R, M(B)) = M(C0 (R, B)) ,

(iii) If z is a contraction in the ideal Q(R, B) = J + β(J) of ER :=

M(C0 (R, B))/ C0 (R, B)), then there is a contraction w ∈ Cb (R, B) with
w + C0 (R, B)) = z, and w∗ (δ∞ ◦ k0 )(·)w + C0 (R, B) = z ∗ H0 (·)z =: T .
(iv) There is a contraction v ∈ Cb (R, B) with v(t) = 0 for t < 0 and v ∗ k0 (·)v +
C0 (R, B) = T . Thus, d∗ h0 (·)d = d∗ (h0 + βh0 )(·)d = T for the contraction
d := v + C0 (R, B) ∈ J ∼
= Q(R+ , B).
(v) There is an isometry S ∈ Cb (R, M (B)) ⊆ M(C0 (R, B)) such that s :=
S + C0 (R, B) ∈ ER satisfies s∗ h0 (·)s = d∗ h0 (·)d = T .

Corollary 7.4.23. Suppose that N ⊆ B is a (non-simple) strongly purely

infinite C*–subalgebra of B with N B dense in B, and that k : D ⊗ O2 → N is a
non-degenerate nuclear *-monomorphism. Let k0 := k((·) ⊗ 1) ∈ Homn uc(D, B),
X := Prim(N ) and define actions ΨA (respectively ΨB ) of X on A (respectively on
B) by ΨA (J) := k0−1 (k0 (A) ∩ J) and ΨB (J) := the closed ideal of B generated by
J ∈ I(N ) ∼
= O(Prim(N )).
Then CPrn (Ψ) = C(k0 ) ⊆ CP(A, B) and

SR(X; D, B) = SR(CPnuc (Ψ); D, B) = S(k0 ; D, Q(R+ , M(B))) .

??
Next taken from chap.9. Need this here ??
The reader can see, that X := Prim(N ) acts on B upper semi-continuously
by ΨB where ΨB (J) is the closed ideal of B generated by the closed ideal J of
N . The T0 space X acts lower semi-continuously on D by ΨD , and, moreover, the
action ΨD is monotone upper semi-continuous, i.e., satisfies also condition (ii) of
Definition 1.2.6. Here ΨD (J) for J ∈ I(N ) is defined by

ΨD (J) := k0−1 (k0 (D) ∩ J) .

The later in Chapter 12 given proof of Theorem K shows that k0 : D → B (and

k : D ⊗ O2 → B) can be constructed up to unitary homotopy from the actions ΨD
and ΨB of X on D, respectively on B.
Next used in chp.9:
There is a natural isomorphism Cb,st (Y, M(B)) ∼
= M(C0 (Y, B)).
next: lem:7.ZZnew.chp.9
848 7. GROUPS OF ASYMPTOTIC MORPHISMS R(C; A, B)

Lemma 7.4.24. Suppose that D is separable and stable, B is σ-unital and stable
and that h0 ∈ Hom(D, B) is non-degenerate and is unitarily equivalent to h0 ⊕ h0 .
Let C := C(h0 ) ⊆ CP(D, B), the point-norm closed matrix operator-convex
cone generated by h0 .
Then a c.p. contraction T : D → Q(R+ , M(B)) is dominated by h0 in
Q(R+ , M(B))), if and only if, there is a point-norm continuous map t ∈ R+ 7→
Vt ∈ C such that V : D → C(R+ , B) satisfies kV k ≤ 1, V (a)(t) := Vt (a) and
V (a) + C0 (R+ , B) = T (a) for all a ∈ D.
If T ∈ CP(D, B) is a c.p. contraction, then h0 dominates T in Q(R+ , M(B)),
if and only if, T ∈ C.
In particular, h ∈ Hom(D, B) satisfies h ∈ C, if and only if, [h] ∈ SR(C; D, B),


if and only if, [h] ∈ Gr h0 ; D, Q(R+ , M(B)) .

The natural group homomorphism

G(h0 ; D, M(B)) → G(h0 ; D, Q(R+ , M(B)))

is neither surjective nor injective in general, e.g. for D = O2 ⊗ K, B = O2 ⊗ O2 ⊗ K

and h0 (·) := 1 ⊗ (·).

Proof. T ∈ CP(D, Q(R+ , M(B))) is dominated by h0 in Q(R+ , M(B)) ⊃ B,

if and only if, there is a norm continuous map t ∈ R+ 7→ S(t) ∈ M(B) into
the isometries in M(B) with s∗ h0 (·)s = T for s := S + C0 (R+ , M(B)). Let
Vt (a) := S(t)∗ h0 (a)S(t), then Vt ∈ C = C(h0 ) and V (a) + C0 (R+ , B) = T (a) for all
a ∈ A (because Cb (R+ , B))) ∩ C0 (R+ , M(B)) = C0 (R+ , B) and Cb (R+ , B) is an
ideal of Cb (R+ , M(B))).
If T ∈ CP(D, B) (in addition), this means that T is the limit of Vt :=
S(t)∗ h0 (·)S(t) in point-norm. Since C is closed in point-norm topology, this implies
T ∈ C.
More arguments:
The assumption implies that there are norm continuous paths

t ∈ [0, ∞) 7→ Sk (t) ∈ M(B)

(for k ∈ {1, 2}) of isometries, defining a path of copies of O2 in M(B) that commutes
with h0 (D) in Cb (R+ , M(B)) modulo C0 (R+ , B).
If T ∈ C then there exist point-norm approximation b∗n h(·)bn of T in
CP(D, B) ⊆ L(D, B) with suitable bn . With approximately central unit of
D one can manage that kbn k ≤ 1. This allows to replace the sequence by a
norm-continuous path of isometries in Cb (R+ , M(B)).
Conversely, if there is contraction V : D → Cb (R+ , B) with Vt (·) := V (·)(t) ∈ C,
then there is a norm-continuous map t ∈ R+ 7→ S(t) ∈ M(B) into the isometries
in M(B), such that limt→∞ kV (a)(t) − S(t)∗ h0 (a)S(t)k = 0 for all a ∈ D.
??
 5. RELATED QUESTIONS 849

Give reference for construction of S(t)

Perhaps as in above blue part?
Let h ∈ Hom(D, B). By Definition ??of [h] ∈ SR(C; D, B), if and only if, there
is a map t ∈ [0, ∞) 7→ Vt ∈ C, such that limt→∞ kVt (a) − h(a)k = 0 for every a ∈ D.
Thus h ∈ C.
If h ∈ C, then T := h is dominated by h0 in Q(R+ ; M(B))
(but in general not directly in M(B) itself, h0 dominates h only asymptotical
inside M(B)). 

Next cited from chp.8:

If an element b ∈ M(SB)/SB has a representative c ∈ Cb,st (R, M(B)) such that
c(t) ∈ B for every t. that it even does not say that c is in Cb (R, B) (as we have
seen in Chapter 7).
Next definition cited from chp.8:
SR(X; A, B) := Homnuc (X; A, Q(R+ , B)) for stable separable A and σ-unital stable
B.
for the action Ψ of X on Q(R+ , B)
one has to take Ψ(U ) := Q(R+ , B)ΨB (U ) Q(R+ , B)
(or less controlled, but NOT Ψ(U ) = Q(R+ , ΨB (U )) ???)
Somewhere in Chp.7: following should appear for use in Chp.9:
Q(R+ , M(F )) is a unital C *-subalgebra of the multiplier algebra M(Q(R+ , F )) of
Q(R+ , F ).

5. Related questions

Question:
Let hj : A → M weakly nuclear C *-morphisms (j = 1, 2), where A is separable and
exact and M is a semi-finite W*-algebra and kh1 (a)pk = kh2 (a)pk for all central
projections p of M and all a ∈ A, that the neutral ???? elements ej of the weak
closures of hj (A) are unitarily equivalent in M , and that τ (h1 (a)) = τ (h2 (a)) for
all a ∈ A+ and all l.s.c. additive traces τ : M+ → [0, ∞] (possibly degenerate).
Is h1 unitarily homotopic to h2 ?
Are they (at least) weakly approximately unitarily equivalent?
(There should be a reduction to the case of countably decomposable M and
then to M with separable pre-dual.)
If the image of h1 has only trivial intersection with the Brauer-ideal of M
(i.e., the norm-closed ideal generated by the finite projections in M ), then one has
only to check that the infinite repeats δ∞ ◦ hj are unitarily homotopic to hj (if
e1 = e2 = 1M ) and then apply Theorem M (here δ∞ is in End(M )).
 CHAPTER 8

The isomorphism of KK(C; A, B) and Ext(SC; A, SB)

We have reassuring good news for those readers who are interested only in
Theorem A and Corollaries C, D, F, G, H and J, and only in the nuclear case of
Theorem I:
Their proofs only use the triviality of Ext−1 (A, O2 ) and the special case of Theorem
B in the introductory Chapter 1, where A is nuclear. In this cases we have obviously
the identities Extnuc (A, B) = Ext−1 (A, B) and KK(A, B) = KKnuc (A, B). The
reader can see that our proof of Theorem B in this special cases involves only the
usual KK- and Ext-theory of Kasparov. So, the reader can find the needed facts for
this very special case and the here indicated approach in text books, e.g. in [73],
except some elementary observations in Sections 3 of this chapter
Is Section 3 all?
needed for the proofs of Theorems B and its generalization M in Chapter 9.
In this Chapter 8 we describe our viewpoint on a (weakly) nuclear version of
Kasparov’s KK-theory, and its isomorphism to Extnuc (A, SB), but in a more general
context that covers also our C-equivariant group Ext(C ; A, SB) – just by the any-
way needed more refined and general terminology. The needed definitions and facts
about Extnuc (A, B) and generalizations in different directions like Ext(X; A, B) and
Ext(C; A, B) can be found in Chapters 3 and 5. We generalize the ordinary KK-
functor in the following direction:
We introduce KK-groups KK(C; A, B) depending on (non-degenerate) countably
generated point-norm closed matrix operator-convex cones (m.o.c.c.) C ⊆ CP(A, B)
(cf. Chapter 3). It has the property, that the Kasparov product allows to define a
bi-additive map

KK(C1 ; A, B) × KK(C2 ; B, C) → KK(C3 ; A, C)

if C2 ◦ C1 ⊆ C3 and A and B are both separable, and C is σ-unital. A special case

is our residually nuclear KK-group KKnuc (X; A, B) that corresponds to a well-
defined m.o.c. cone C := CPrn (X; A, B) ⊆ CP(A, B) , the cone CPrn (X; A, B) :=
CPrn (X; ΨA , ΨB ) of Ψ-residually nuclear c.p. maps for given actions ΨA : O(X) →
I(A) of a T0 space X on A and ΨB of X on B (under the additional assumption
that CPrn (X; A, B) for the actions ΨA and ΨB is non-degenerate), see Chapters 1
and 3 for the
give Refs. for actions and its cones
851
852 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

definitions of this actions and the related m.o.c. cones. The group KKnuc (A, B)
of Skandalis [726] – even if they are defined directly as the homotopy classes given
by elements in E(A, B[0, 1]) ...
is the same as our KK(C ; A, B) for the special case of the m.o.c. cone C :=
CPnuc (A, B) ⊆ CP(A, B) (?? check def in [726] again ! ??),
if A is separable and B is σ-unital, because then CPnuc (A, B) is countably
generated as matrix operator convex cone.
If A and B are both separable, then C := CP(A, B) is itself countably gener-
ated and the usual Kasparov group KK(A, B) is the same as KK(CP(A, B) ; A, B) .
Check next! But next is true if B is not σ-unital!:
(Notice that CP(A, B) is not countably generated – which is equal to singly gen-
erated – as matrix operator-convex cone if A := K(`2 (N)) or A := C and B is not
separable.)
But CP(C, B) ∼ = B+ is generated as m. o. c. cone by the element C 3 z 7→ ze ∈
B if e ∈ B+ is a strictly positive element of B. But what happens if A := C0 (0, 1]?
All results of this chapter can be obtained by modifying the proofs of the cor-
responding results for the usual (non-nuclear) KK-theory, but use our observations
on operator-convex cones in Chapter 3. We apply the Ext(C; ·, ·)-groups defined in
Chapter 5 to give other proofs, which are more “elementary” (or “constructive”) and
allow to control all identifications and constructions which are needed during our
proofs of Theorems B and M, cf. Lemma 8.3.2. In fact, we need to work only with
unitary homotopy, because this is automatically C-equivariant, and we prove later
that C-equivariant homotopy can be realized by unitary homotopy (modulo stabi-
lization). This is also important for the X-equivariant theory KKnuc (X ; A, B) :=
KK(C ; A, B) given by the m.o.c. cone for C := CPrn (X ; ΨA , ΨB ; A, B).

The major results of this chapter are the homotopy invariance of KK(C ; A, B),
the observations in Section 5 (that are needed in Chapter 9) and the related “con-
trolled” isomorphism

Ext C ⊗ CP(C, C0 (R)); A, B ⊗ C0 (R) ∼

= KK(C; A, B) ,

that gives in particular Extnuc (X; A, SB) ∼= KKnuc (X; A, B) for actions ΨA
and ΨB of a T0 -space X on A respectively B if the cone CPrn (X; A, B) is
non-degenerate and countably generated, cf. Chapters 1 and 5.
In Chapters 9 and 12 we use results on KKnuc (X; A, B) and Extnuc (X; A, B) in
the case where X is often the primitive ideal space of a separable C *-algebra. The
definitions can be found in Chapters 1 and 5. The necessary changes from the theory
of KKnuc (A, B) and Extnuc (A, B) to this general KKnuc (X; ., .)- and Extnuc (X; ., .)-
theories or even to KK(C ; A, B) and Ext(C ; A, B) are often straight-forward, up
to more complicate notations:
Then one has only to replace the emphasized words “nuclear” by the words “Ψ-
residually nuclear”, respectively by “C-compatible”, if we consider the more flexible
 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB) 853

theory of the groups KK(C ; A, B) . Therefore we discuss sometimes only proper-

ties of the KKnuc -theory because of simpler notation. All differences to the basic
theory appears by a pre-selection of suitably chosen sub-classes of Kasparov mod-
ules. If this choice has been made, only two general conditions define the appro-
priate equivalence relation that makes the direct module-sum to the addition on
the corresponding classes that build the elements of the KK-groups. The formed
Grothendieck group defines in the case of C = CP(A, B) the relation ∼c on the
Kasparov modules considered by G. Kasparov and J. Cuntz.
The prove of the (operator-) homotopy invariance of J. Cuntz and G. Skandalis
carries over almost verbatim to B 7→ KK(C; A, B). The full homotopy invariance
can then derived by the ideas of A. Connes ...
Check next statement!!
A modification of parts of our proof of Theorem B can be used to show that
O∞ -stabilized unsuspended weakly nuclear ⊗ O∞ necessary?
E-theory is equivalent to KKnuc -theory in separable case with trivial grading
(an observation of N.Ch. Phillips, cf. Remark 8.3.6).
Check ??? At some points we could use that A0 ∩Q(B) with Q(B) :=
M(B)/B
is s.p.i. if A ,→ Q(B) is nuclear
and B is s.p.i. and σ-unital.

Then A0 ∩ Q(B) should be K1 -bijective.

VERY IMPORTANT!!!
We need !!! and use later that:
KK(C; A, B) is naturally isomorphic to the kernel of the natural map

K1 (πB ◦ HC )(A)0 ∩ Q(B) → K1 (Q(B)) ∼

= K0 (B)

if A is separable and stable and B is σ-unital and stable (and all is trivially graded).
And that the “natural” isomorphisms KK(C[0, 1] ; A, B[0, 1]) ∼
= KK(C ; A, B)
and KK([0, 1]C ; A[0, 1], B) ∼
= KK(C ; A, B). given by evaluations at points t ∈
[0, 1].
( – The latter are the homotopy invariances of KK(C; ·, ·) and works as in
the case C = CP(·, ·) using the G. Kasparov product construction modified by
A. Connes ... – ).
Claimed in Chapter 5:
There are natural isomorphisms KK(A, B) ∼ = Ext(A, SB) and KK(A, B) ∼ =
ker K1 (H0 (A) ∩ Q(B)) → Q(B) ∼
0


= K0 (B) for separable stable σ-unital A and B.
This should come from Cuntz picture for trivially graded A and B:
1
KK (C; A, B) is defined for separable A and σ-unital B and countably gener-
ated m.o.c. cone C ⊆ CP(A, B) as equivalence classes [(ψ, P )] in the set of all pairs
854 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

{(ψ, P ) ; ψ : A → M(B ⊗ K), P = P ∗ = P 2 ∈ M(B ⊗ K)} with the properties: ψ

is a C *-morphism, d∗ ψ(·)d ∈ C for all d ∈ B ⊗ K and P ψ(a) − ψ(a)P ∈ B ⊗ K.
Can all reduce to case E := H b B := HB ⊕ Hop . It has the “standard even”
B
b B) ∼
grading if B is trivially graded. And this induces on L(H = M(B ⊗ K) the
standard even grading.
If E is any trivially graded Hilbert B-module then E ⊕E op induces the standard
even grading on M2 (L(E)).
We should start with “Fredholm picture” in case of trivially graded separable
A and stable σ-unital trivially graded B:
Triples (E (0) ⊕ E (1) , φ0 ⊕ φ1 , T ) with T ∈ L(E (0) , E (1) ) and relations:
T φ0 (a)−φ1 (a)T ∈ K(E (0) , E (1) ) for all a ∈ A, 1−T ∗ T ∈ K(E (0) ) and 1−T T ∗ ∈
K(E (1) ) .
The φ∗ should satisfy hφ∗ (·)e, ei ∈ C.
Then try to “absorb” the φ∗ and E (∗) by HC : A → M(B) and E ∼
= B in case
of stable σ-unital B.
Should get new T ∈ M(B) with [T, HC (a)] ∈ B, (1 − T ∗ T )HC (a) ∈ B (1 −
∗
T T )HC (a) ∈ B for all a ∈ A. ... ???
Is this all at the beginning?
If one can make ????????????
Also very important:
There are natural isomorphisms

= ker(K0 (H0 (A)0 ∩ Q(B)) → K0 (Q(B)) ∼

Ext(C; A, B) ∼ = K1 (B))

for separable stable σ-unital A and σ-unital stable B.

⇐ This has been shown in Chp. 3 or 5 ??? Where? Ref.?

1. Some basics on KK and Ext

Think about interchanging the role of L(E) and B(E),

or use B(E) and L(E) !!??
The nuclear KK-theory KKnuc (A, B) was defined by Skandalis [726]. But we
use at the beginning a different definition that is more algebraic and refers not ex-
plicit to any sort of homotopy invariance. It allows us to give later a “constructive”
proof of the existence of a functor isomorphism:

Ext(C; A ⊗ K, B) ∼
= KK(C; A, B(1) ) (1.1)

in case where A and B are trivially graded, i.e., the corresponding Z2 -actions on
A and B are given by the identity map, A is separable, B is σ-unital, stable and
B(1) denotes the graded algebra B(1) := B ⊕ B with the standard odd grading
β((a, b)) := (b, a). In particular Extnuc (A ⊗ K, B) ∼
= KKnuc (A, B(1) ).
 1. SOME BASICS ON KK AND Ext 855

It implies then that Ext(SC; A, SB) = KK(C; A, B) for trivially graded sepa-
rable stable C *-algebras A and B.
The natural isomorphism B(1) ∼ = B ⊗ (C(1) ) of graded C *-algebras allows to
extend an operator-convex cone C ⊆ CP(A, B) naturally to a cone C ⊆ CP(A, B(1) ) :
Simply take the m.o.c. cone-tensor product C⊗CP(C, C⊕C), i.e., the smallest point-
norm closed hereditary sub-cone of CP(A, B(1) ) containing the c.p. maps V ⊗ W
with V ∈ C and W ∈ CP(C, C ⊕ C) ∼ = R+ ⊕ R+ .
Recall that the “tensor”-cones are in general not completions of algebraic tensor
products – see Definition 3.6.16: Here this “m.o.c.cone-tensor-product” is – as a
set – the same as the set of maps a ∈ A 7→ (V1 (a), V2 (a)) ∈ B1 with V1 , V2 ∈ C ⊆
CP(A, B).
We have to consider Z2 -graded C *-algebras, because we need formula (1.1)
and its consequences concerning the very important homotopy invariance. But the
reader should notice that the groups Extnuc (A, SB) and KKnuc (A, B) are in general
not isomorphic for non-trivially graded separable C *-algebras A and B.
Check again ??:
G. Skandalis [726] requires homotopy invariance as a part of his definition
of KKnuc (A, B) right from the beginning, so he was in the comfortable situation
that he didn’t need to prove the formula (1.1). But for us this is a non-obvious
key observation, that must be proven, because it is then often used in our proofs,
e.g. in Section 5 that establish the applicability the abstract results in Chapter 4
and allow with them together to show in Chapter 9 that
?? in the general case that we consider ??
Need: For stable separable C *-algebras A and B and countable generated non-
degenerate m.o.c. cone C for the canonical HC : A → M(B) holds (in the trivially
graded case):
The group KK(C ; A, B) is naturally isomorphic to the kernel of the natural
morphism

K1 (πB ◦ HC )(A)0 ∩ Qs (B) → K1 Qs (B) ∼

= K0 (B) .

A hint is: [73, subsec. 17.5.1] but then add the infinite repeat of φ0 ⊕ φ1 to it
and use suitable unitaries in M(B) to correct the formula ...
To be carried out in detail.
In case of Ext(C ; A, B) it is not difficult to see that Ext(C ; A, B) is isomorphic
to the kernel of

K0 (πB ◦ HC )(A)0 ∩ Qs (B) → K0 Qs (B) ∼

= K0 (B) .

We define KKnuc (A, B) as one of our more general groups KK(C; A, B) – spe-
cialized to the case where C := CPnuc (A, B). They have a completely algebraic
definition and that does not impose any sort of homotopy invariance in itself. We
856 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

shall later prove the homotopy invariance of our KK(C; A, B) by modifications of

ideas of G. Kasparov (discussed and modified by A. Connes, G. Skandalis, J. Cuntz,
N. Higson and others).
We work with an “algebraic” definition that does not refer to “homotopy invari-
ance” because the homotopy invariance of the purely algebraic defining relations
proves also the homotopy invariance of our Extnuc (A, B) groups (and of our groups
Ext(C; A, B), Extnuc (X; A, B) etc.) with help of our definition of KK(C; A, B). It
is not a triviality.
Even the much weaker (since as a relation stronger) operator-homotopy invari-
ance of the equivalence classes that build the elements of in KK(C; A, B) is not
evident. The proof of the homotopy invariance from stronger more algebraic def-
initions is the core of the application of generalized KK(C; ·, ·) to classification of
non-simple C *-algebras.
The so-called “degenerate” Kasparov modules represent zero automatically in
any version of KK-theory that is a quotient of the Grothendieck group of a sub-
semigroup of the semigroup of unitary equivalence classes of Kasparov modules,
because the “degenerate” Kasparov modules can be defined equivalently by its prop-
erty that their infinite direct sum is again a Kasparov module ( 1 ). The proof of the
operator homotopy can be done by modifications of the techniques of G. Kasparov
(respectively of J. Cuntz and G. Skandalis) in the proof of the homotopy invari-
ance of the below defined KKc (A, B)-functor for graded separable A and σ-unital
graded B, where KKc can be defined equivalently by the cobordism-picture, cf. [73,
def. 17.10.2, thm. 17.10.7] ( 2 ). After introduction of the Kasparov products up to
operator homotopy (!), the proof of [73, thm. 18.5.3] carries over to our situation
and establishes the general homotopy invariance, where one can use that Kasparov
products define the functor KKnuc to as an KK-“ideal” in the KK-category. See
below lemmata for a more details.
The way from operator homotopy invariance to the existence of Kasparov pro-
ducts (and then to homotopy invariance) can be found also in [389, chp. 2].
We use the approach of Kasparov to show that our KKnuc -bi-functor (respec-
tively our KK(C; ·, ·) bi-functor – generalizing our KKnuc (X; ·, ·) bi-functor for
actions Ψ of O(X) on ideal lattices) is homotopy invariant. But one has to be
careful because in general the identity maps idA does not represent an element of
KKnuc (A, A), and the relations are calculated up to stabilization with nuclear Kas-
parov modules and not up to stabilization with general Kasparov modules. But
with this in mind, one can go through the textbooks and over-carry step by step
every argument to the case of KKnuc (or even to KK(C; ·, ·), and thus especially to
KKnuc (X; ·, ·), if one uses the natural extension of the actions Ψ of X on A to an
action Ψ ⊗ C : U → Ψ(U ) ⊗ C on A ⊗ C, for C := C0 ((0, 1]), C0 (R), K, C([0, 1]) or
O∞ –).

1 Note that the infinite sums are again weakly nuclear, or more generally C-compatible.
2Note that [73, def. 17.10.1] must require [p, F ]φ(A) ⊆ K(E) in addition, cf. [187, def. 3.1].
 1. SOME BASICS ON KK AND Ext 857

Let us recall some general definitions concerning graded Hilbert modules used
in the “algebraic” definition of KK(C; ·, ·)-theory:
We prefer a picture (– the original picture of Kasparov himself –) that requires
to use graded σ-unital algebras for the most basic results, in particular, all C*-
algebras A, B, . . . in question are supposed to be σ-unital, of course except the related
algebras M(A), ... , etc .
Suppose that A and B are Z2 -graded C *-algebras, with grading automorphisms
βA ∈ Aut(A) and βB ∈ Aut(B). We drop the indices A and B, if it is obvious
where β acts. Thus, always β 2 = id. Further we suppose that C ⊆ CP(A, B) is a
(non-degenerate) point-norm closed matrix operator-convex (m.o.c.) cone with the
(minimally necessary) property that βB ◦ C ◦ βA ⊆ C .
In general a Z2 -grading of a vector space V is given by a linear operator β : V →
V with β 2 = idV .
compare def.s with book:
An element v ∈ V has degree deg(v) = 0 if v = β(v) and deg(v) = 1 if
v = −β(v). We use also the notation ∂v for deg(v).
Is notation ∂v really better? Not misleading?
The vector space V is the direct sum V = V (0) + V (1) ∼= V (0) ⊕ V (1) of its
(0) (1)
linear subspaces V = {v ∈ V ; deg(v) = 0} and V = {v ∈ V ; deg(v) = 1}.
Thus β(v) = β(v0 + v1 ) = v0 − v1 for the unique decomposition v = v0 + v1 with
deg(v0 ) = 0 and deg(v1 ) = 1.
The graded tensor product V ⊗ W of graded vector spaces V , W is usually
given by βV ⊗ βW ? ??
A Hilbert B-module E is a complete normed right B-module together with
positive definite hermitian B-module form (x, y) ∈ E × E 7→ hx, yi ∈ B that is
linear in the second variable and anti-linear (conjugate-linear) in the first variable
and satisfies kxk2 = khx, xik for all x ∈ E.
We denote by K(E) the closed linear span of the “B-rank-one” operators:

x ∈ E 7→ yhz, xi ∈ E .

It is a C *-algebra and an essential ideal in the C *-algebra L(E) of bounded and

h., .i-adjoint-able B-module endomorphisms of E. Moreover L(E) is naturally iso-
morphic to the multiplier C *-algebra (real or complex) M(K(E)) of K(E). During
Chapter 8 we denote by B(E) the Banach algebra of all bounded linear operators
on E with the usual operator norm. Then our L(E) is the subalgebra of T ∈ B(E)
with T (x · b) = (T x) · b for x ∈ E and b ∈ B. The norms on L(E) are the usual
operator norms. L(E) does not coincide with B(E) if B 6= C.

Assume that B is Z2 -graded by a C *-algebra automorphism βB of order 2.

2
A Hilbert B-module E is graded by βE , if βE is an isometry with βE = id and
858 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

βE (xb) = βE (x)βB (b) for x ∈ E, b ∈ B. One can easily see that the isome-
try property kβE (x)k = kxk (x ∈ E) of βE is equivalent to invariance property
βB (hx, yi) = hβE (x), βE (y)i for x, y ∈ E, e.g. simply by composing both sides with
positive functionals on B, and using the trapezoid rule and that khx, xik = kxk2 .
The grading βE induces on the C *-algebra L(E) the grading F → βE ◦ F ◦ βE .
By E op we denote the opposite Hilbert B-module of E, that is the same B-module
as E with the new “opposite” grading −βE .
An isomorphism of graded Hilbert B-modules λ : E1 → E2 is an isometric
B-module map from E1 onto E2 that satisfies λ ◦ βE1 = βE2 ◦ λ ( 3 ). The isomor-
phism classes of (over B) countably generated graded Hilbert B-modules build a
set, if B is σ-unital, because they are isomorphic to orthogonally complemented
grading invariant closed subspaces of HeB := HB ⊕B (HB )op , cf. [73, thm.14.6.1],
by Kasparov’s stabilization theorem [?], (or, by the almost equivalent Brown stable
isomorphism theorem, [107]).
The direct sum E ⊕∞ E 0 of Hilbert B-modules E and E 0 with norm
k(x, y)k∞ := max(kxk, kyk) is a Hilbert B ⊕ B-module, and therefore has also
the diagonal right B-module structure. If we consider E ⊕∞ E 0 as a right Ba-
nach B-module, then E ⊕∞ E 0 is naturally and bounded (but not isometrically!)
B-module isomorphic to the Hilbert B-module sum E ⊕B E 0 with B-valued
form h(x1 , y1 ), (x2 , y2 )i := hx1 , x2 i + hy1 , y2 i. Up to unitary equivalence, ⊕B is
associative and commutative, L(E) ⊕∞ L(E 0 ) ⊂ B(E ⊕∞ E 0 ) and L(E) ⊕∞ L(E 0 )
is a unital C *-subalgebra of L(E ⊕B E 0 ).

Some minimal basics on graded tensor products of operator-convex cones and

on natural extension are needed during this Chapter 8:
We denote by A ⊗b B the (outer) graded tensor product of Z2 -graded C *-algebras
A and B. The outer graded tensor product of graded Hilbert A-modules E1
and graded Hilbert B-modules E2 will be denoted by E1 ⊗E b 2 and is a Hilbert
A⊗B-module.
b

Since we have to define the groups KK(C ; A⊗C,b B ⊗D)

b for Z2 -graded separable
nuclear C *-algebras C and D, we must “extend” a given matrix operator-convex
cone C ⊆ CP(A, B) to an m.o.c. cone CC,D ⊆ CP(A⊗C, b B ⊗D)
b :
Simply, let CC,D := C ⊗ CP(C, D), where C ⊗ CP(C, D) ⊆ CP(A⊗C,
b b b B ⊗D)
b
denotes the point-norm closed m.o.c. cone that is generated by the graded
Hilbert (A⊗C,
b B ⊗D)-module
b (E1 ⊗E
b 2 , φ1 ⊗φ
b 2 : A⊗C
b → L(E1 ⊗E b 2 )) for a graded
Hilbert (A, B)-module (E1 , φ1 : A → L(E1 )) and a graded Hilbert (C, D)-module
(E1 , φ2 : C → L(E2 )) that generates C (respectively generates CP(C, D)) in the
sense of Definition 3.6.25. (Please observe that the notation C ⊗ b CP(C, D) does
not say that it is a sort of “completion” of the linear span of some kind of tensor

3 The grading λ is C-linear by the Cohen factorization theorem, because every Hilbert B-

module is non-degenerate, and it holds automatically hλ(x), λ(y)i = hx, yi for x, y ∈ E1 , because,
if a, b ∈ B+ and kc∗ ack = kc∗ bck for all c ∈ B then a = b, – as application of pure states to b − a
shows by using the “excision” Lemma ??.
 1. SOME BASICS ON KK AND Ext 859

products: It is only “generated” as point-norm closed m.o.c. cone by tensor

products V ⊗ W with V ∈ C and W ∈ CP(C, D). See Chapter 3 for more details.
Let us consider an easy example:

Example 8.1.1. We consider the groups KK(C; C[0, 1], C) that are used in the
proof of the later considered “homotopy invariance”. Where C ⊆ CP(C[0, 1], C) ∼
=
C[0, 1]∗+ is any point-norm closed m.o.c. cone ...
The minimal point-norm closed non-degenerate
??????
CP(C, C) = R+ · idC ∼
= R+
matrix operator-convex cone Cmin ⊆ CP(C[0, 1]) := CP[0,1] (C[0, 1], C[0, 1]) on
C[0, 1] are the “inner” c.p. maps, i.e., is here – with other words – simply the set
of all multiplication maps Vg (f ) := gf with positive elements g ∈ C[0, 1]+ . Thus
Cmin (C[0, 1]) ∼
= C[0, 1]+ by an isometric identification kVg k = kgk.
Indeed ????: CP(C[0, 1], C[0, 1]) is natural isomorphic to the set of weakly
continuous maps
[0, 1] 3 t 7→ µt ∈ C[0, 1]∗+
from [0, 1] into the finite positive measures on [0, 1].
Let C ⊆ CP(C[0, 1], C[0, 1]) is a point-norm closed m.o.c. cone, then for each
t ∈ [0, 1] there exists V ∈ C and f ∈ C[0, 1]+ such that V (f )(t) > 1. By continuity
of V (f )(·), order monotony of V and we find an open intervall I = (α, β) ⊆ [−1, 2]
and γ > 0 such that t ∈ I and V (1)(s) > γ for s ∈ I. The compactness of [0, 1]
shows then that there is µ > 0 finite sum of such kind of V ∈ C is a W0 ∈ C with
W0 (1)(t) ≥ µ. It shows that there exists W ∈ C with with W (1) = 1.
The only point-norm closed non-degenerate m.o.c. cone C ⊆ CP(C[0, 1], C) =
C[0, 1]∗+ is C[0, 1]∗+ itself ??
Seems to be ... wrong ...????
It implies that KK(C; C[0, 1], C) = KK(C[0, 1], C) ∼
= Z and then for all non-
degenerate point-norm m.o.c. cones C ⊆ CP(A, B) and, – as we see later – that

KK(C[0, 1] ; A, B[0, 1]) ∼

= KK(C ; A, B)

via any of the evaluation maps {Vt } ∈ C[0, 1] → V0 ∈ C and {b(t)} ∈ B[0, 1] :=
C([0, 1], B) 7→ b(0) ∈ B.
(Compare Definition 3.6.16 for the notations.)

For our later applications on the study of KK(C; A, B) and Ext(C; A, B) in

the trivially graded cases, we need only the below listed cases (I) and (II), where
the graded tensor products of algebras and the outer graded tensor-products of its
Hilbert modules become naturally isomorphic to the usual spatial C *-algebra- and
Hilbert module- tensor products, then with grading automorphisms on the tensor
product given by the tensor product of the grading automorphisms:
860 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

(I) A and B are graded, C ◦ βA ⊆ C and βB ◦ C ⊆ C ,

and C, D are trivially graded separable nuclear algebras.
(II) A and B are trivially graded and C, D are graded separable nuclear alge-
bras.

In case (I) we need only C[0, 1], C0 (R+ ), C0 (R), C0 (R2 ), C and K in place of C or D
for our applications, and in case of (II) the later application for the proofs of Theo-
rems ?? require only the cases (C, D) = (C(1) , C0 (R)) and (C, D) = (C0 (R), C(1) ).
The below given construction of the group morphisms

τB : KK(C, D) → KK(CPin (B, B) ⊗ CP(C, D) ; B ⊗C,

b B ⊗D)
b

becomes natural for all graded B, C and D.

b (1) ∼
In general, A⊗C = A oβ Z2 ⊆ A ⊗ M2 for the Z2 -action defined by the β and
with grading β ◦ βb on A oβ Z2 on the right side, where βb is just the dual Z2 -action
on A oβ Z2 of the action β : Z2 → Aut(A). It is a C *-morphism that maps a ∈ A
to a ∈ A and ε := (1, −1) to the Z2 -generator in Z2 ⊆ M(A oβ Z2 ). In the special
case C = CPrn (X; A, B) := CPrn (X; ΨA , ΨB ; A, B) we define a matrix o.c. cone
by

CC,D := CPrn (X ; A ⊗
b C , B ⊗D)
b := CPrn (X; ΨA⊗C
b , ΨB ⊗D
b )

b (U ) := ΨA (U )⊗C
for the actions ΨA⊗C b with U ∈ O(X) and (similar) ΨB ⊗F b For
simplicity, we assume here that actions ΨA and ΨB are invariant under the gradings
on A and B, e.g. βA (ΨA (U )) = ΨA (U ) for U ∈ O(X).
We know from Chapter 3 that for nuclear C and D, a c.p. map V ∈
CP(A ⊗
bC, B⊗b D) is in CC,D := C ⊗
b CP(C, D) if and only if

a ∈ A 7→ (idB ⊗ρ)(V
b (a⊗e))
b ∈B

is in C for each e ∈ C+ with βC (e) = e and for each positive functional ρ ∈ D∗

with ρ ◦ βD = ρ:
See Proposition ?? for the case of trivially graded C and D. The graded case
can be reduced to the case of an even grading by the embedding of C and D into
C o Z2 respectively D o Z2 (with the grading given by the self-adjoint unitary that
generates the copy of Z2 in M(C o Z2 )). Then the arguments are similar to those
in the proof of Proposition ?? (using that the point-norm closed m.o.c. cones are
just the point-norm closed convex cones that are hereditary in CP(A⊗C, b B ⊗D),
b
then properties of hereditary cones).
In case C := C, D := C0 (Y ) (with Y a locally compact Hausdorff space), a map
V ∈ CP(A, C0 (Y, B)) is in CC,D , if and only if, there is a point-strongly continuous
map W : y ∈ Y 7→ CP(A, B) ⊆ L(A, B) with W (y) ∈ C, the map y 7→ kW (y)(a)k is
in C0 (Y ) for every a ∈ A and W (y)(a) = V (a)(y). In this case we write also C(Y )
(C[Y ] or simply C) if CC,D = C ⊗ CP(C, C0 (Y )) for C = C and D = C0 (Y ). Thus,
 1. SOME BASICS ON KK AND Ext 861

the extensions C y CC,D are rather natural ( 4 ). KKnuc -theory is the special case
of KK(C; ·, ·)-theory where C = CPnuc (A, B),
Check if this is equivalent to the Def. of Skandalis [726]
and it is the special case of KK(X; ·, ·)-theory where X is a singleton, i.e.,
X = {p} with action on A given by ΨA (∅) = 0, ΨA ({p}) = A (and similar on B).
In the same way Extnuc (·, ·) is a special case of Extnuc (X; ·, ·) or of Ext(C; ·, ·). If A
is separable and B is nuclear and separable, then KK(C; A, B) = KKnuc (X; A, B)
for C := CPrn (X; A, B)
read again !!!!!! more precise? ??
(where X and non-degenerate actions ΨA and ΨB of X on A and B are given)
and X := Prim(B), with natural action on B and lower semi-continuous action of
Prim(B) defined by a “universal” Hilbert A-B-module H0 : A → L(HB ) (as defined
in Chapters 3 and 5, where we have to use that the corresponding cone is separating
and non-degenerate by the existence of a regular Abelian subalgebra in B (in the
sense of Definition B.4.1) using [464] and [359]).
A triple E = (E, φ, F ) is called a Kasparov module (more precisely: Kasparov
(A, B)–module ) if E is a graded right Hilbert B-module that is countably generated
over B, φ : A → M(K(E)) ∼ = L(E) is a grading preserving C *-morphism, and
F ∈ L(E) is an operator of degree 1 such that F φ(a)−φ(βA (a))F ( 5 ), (1−F 2 )φ(a)
and (F ∗ − F )φ(a) are in K(E) for every a ∈ A. The Kasparov module (E, φ, F ) is
degenerate if F φ(a) − φ(βA (a))F , (1 − F 2 )φ(a) and (F ∗ − F )φ(a) are all zero.
If the B-module E is trivially graded, then this definition implies that F = 0
and φ(A) ⊆ K(E) and that (E, 0, 0) is the only degenerate Kasparov module.
The Kasparov module E = (E, φ, F ) is isomorphic to E 0 = (E 0 , φ0 , F 0 ) (de-
noted by E 0 ≈u E) if there is an isometric B-module map U : E → E 0 from E
onto E 0 such that U βE = βE 0 U (i.e., U is of even degree), φ0 (·)U = U φ(·) and
F 0 U = U F . It is easy to see that the isometric B-module morphism U preserve
the values of the B-valued hermitian form: hU x, U yi = hx, yi for x, y ∈ E. By
definition, the Kasparov modules E and E 0 are unitary equivalent if they are
isomorphic in this way.
The sum E 0 ⊕E of Kasparov A-B-modules E 0 and E is defined by the B-module
sum

E 0 ⊕ E := (E 0 ⊕B E, φ0 ⊕ φ, F 0 ⊕ F ).

A Kasparov module (E1 , φ1 , F1 ) is a compact perturbation of the Kasparov

module (E, φ, F ) if E1 = E, φ1 = φ and (F − F1 )φ(a) is in K(E) for every a ∈ A.
It suffices here to consider only a ∈ A of degree ∂(a) = 0, because it is obvious that

4 See Chapter 3 for a Definition of tensor products of matrix operator-convex cones. We write

also C instead of CC,D to relax notation.

5Because ∂(F ) = 1, F φ(a) − φ(β (a))F = [F, φ(a)]
A gr for the graded commutator [·, ·]gr .
862 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

A(0) contains an approximate unit of A. Occasionally we use then the terminol-

ogy “φ-compact perturbation” to underline the difference to perturbations by
elements of K(E), or by generalized “compact” operators on the Banach space E.
The equivalence relation ∼scp which is generated by (unitary) isomorphisms
and compact perturbations is compatible with the addition of Kasparov modules.
Therefore the addition ?????? defines on the ∼scp -classes of Kasparov modules the
structure of a semigroup E(A, B)/ ∼scp . The relation ∼cp of [73, def.17.2.4] is
weaker, because the ∼cp -classes allow addition of degenerate Kasparov modules.
But ∼cp and ∼scp generate the same “stable” relation on E(A, B).

2. Isomorphism of Ext(C; A, B) and KK(C; A, B(1) )

Notice that the below considered natural map from KK(C; A, B(1) ) to
Ext(C; A, B)
Or: from Ext(C; A, B) to KK(C; A, B(1) )?
becomes in general only in the special case of separable trivially graded A and
trivially graded σ-unital B an isomorphism.
The here first considered algebras A, A1 , . . . , B, C, D, . . . are σ-unital C *-
algebras. We start with some general considerations that require to consider
also graded algebras. Thus, the algebras may have grading automorphisms
βA , βA1 , . . . , βB , . . ., and then we require that C ⊆ CP(A, B) is a point-norm closed
matrix operator-convex cone with βB ◦ C ◦ βA ⊆ C (cf. Chapter 3, Section on
m.o.c.c. operations and equivariant m.o.c. cones ?? for notations).

Definition 8.2.1. A Kasparov module (E, φ, F ) ∈ E(A, B) is called nuclear

(respectively Ψ-residually nuclear, C-compatible) if the C *-morphism
φ : A → M(K(E)) = L(E)
is weakly nuclear in the sense Definition B.7.6 (respectively is weakly Ψ-residually
nuclear, respectively is weakly C-compatible). Equivalently, this means that the c.p.
maps a ∈ A 7→ hφ(a)x, xi ∈ B are nuclear (respectively are Ψ-residually nuclear,
respectively are in C) for every x ∈ E.

In fact all KK-groups come from first selecting an m.o.c. cone C and then consid-
ering only the C-compatible Kasparov modules, because KK(A, B) = KK(C ; A, B)
with C = CP(A, B), KKnuc (A, B) = KK(CPnuc ; A, B), KK(Ψ ; A, B) =
KK(CP(Ψ ; A, B) ; A, B), where CP(Ψ ; A, B) ⊆ CP(A, B) denotes the m.o.c. cone
of Ψ-equivariant c.p. maps, and, finally, the Ψ-residually nuclear KK-groups are
KK(C ; A, B) with m.o.c. cone C defined by the “Ψ-residually nuclear” maps
C := CPnuc (Ψ ; A, B) ⊆ CP(Ψ ; A, B) ⊂ CP(A, B) .

This theory can be considered as a refinement of the original Kasparov theory.

Examples of nuclear Kasparov (A, B)-modules (respectively of C-compatible
Kasparov modules) are the modules (E, 0, F ) for arbitrary Hilbert B-modules E
 2. ISOMORPHISM OF Ext(C; A, B) AND KK(C; A, B(1) ) 863

and arbitrary F ∈ L(E) of degree one, and the difference construction (B, φ, 0)
for a nuclear C *-morphism φ : A → B (respectively for ψ ∈ Hom(A, B) ∩ C ), where
B is considered as right B-module.
The category of nuclear (respectively of C-compatible) Kasparov (A, B)-
modules is invariant under Hilbert B-module addition, because
φ ⊕ φ0 : A → M K(E ⊕B E 0 )

is again weakly nuclear (resp. is again weakly C-compatible) if φ and φ0 are weakly
nuclear (resp. are weakly C-compatible). If (E, φ, F ) is nuclear (respectively C-
compatible) and (E, φ, F ) ∼scp (E 0 , φ0 , F 0 ), then (E 0 , φ0 , F 0 ) is also nuclear (re-
spectively C-compatible). It follows that the ∼scp -classes of nuclear (respectively
C-compatible) Kasparov A–B-modules build a sub-semigroup Enuc (A, B)/ ∼scp (re-
spectively a sub-semigroup E(C; A, B)/ ∼scp ) of the semi-group E(A, B)/ ∼scp (with
Hilbert B-module sum as addition).

Definition 8.2.2. We define the nuclear Kasparov group KKnuc (A, B) as

the Grothendieck group of Enuc (A, B)/ ∼scp :
KKnuc (A, B) := KK(C; A, B) = Gr(Enuc (A, B)/ ∼scp ) = Gr(E(C; A, B)/ ∼scp ) ,
for C := CPnuc (A, B). The more general C-equivariant Kasparov group
KK(C; A, B) is the Grothendieck group of of E(C; A, B)/ ∼scp :
KKnuc (A, B) := Gr(E(C; A, B)/ ∼scp ) .

The X-equivariant KK-groups and residually nuclear KK-groups are defined

by
KK(X; A, B) := KK(C; A, B) for C := CP(X; A, B) ,
and
KKnuc (X; A, B) := KK(C; A, B) for C := CPrn (X; A, B) .

More precisely, KK(X; A, B) and KKnuc (X; A, B) depend on the actions

ΨA : O(X) → I(A) and ΨB : O(X) → I(B), that define the related point-
norm closed m.o.c. cones CP(X, ΨA , ΨB ; A, B) and CPrn (X, ΨA , ΨB ; A, B),
cf. Chapter 3. But there are other natural and functorial constructions of
m.o.c. cones C from “actions” ΨA and ΨB of X on A and B. The m.o.c. cones
CPrn (X, ΨA , ΨB ; A, B) ⊆ CPnuc (A, B) of ΨA -ΨB –residually nuclear maps play an
important role in our applications.

If A and B are trivially graded and if h : A → B ⊗ K = B ⊗ b K is a nuclear

C *-morphism (respectively if h ∈ Hom(A, B ⊗ K) ∩ C ), we can define [h − 0] ∈
KKnuc (A, B) (respectively [h − 0] ∈ KK(C; A, B)) as the class which is represented
by the difference construction
(HB , h : A → B ⊗ K, 0) ,
where HB denotes the Hilbert B-module
X
HB := {(b1 , b2 , . . .) : bn ∈ B, b∗n bn ∈ B}.
n
864 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

with trivial grading, and where we use the natural isomorphism B ⊗ K ∼

= K(HB ) .
It can be shown (as in the non-nuclear theory) that the bi-functor KKnuc (A, B)
is stable, and with help of Kasparov products, that our KKnuc groups are homotopy
invariant, hence, are the same as the KKnuc -groups of Skandalis [726] (at least if
A is exact). Therefore, they satisfy variants of Bott Periodicity, 6-term exact se-
quences for semi-split exact sequences, ... But in case of the KK(C; ·, ·)-theory
the situation is more complicate, because this are in essence functors from a “cat-
egory” of point-norm closed m.o.c. cones C ⊆ CP(A, B) into Abelian groups, and
the C *-algebras play only a formal role as a sort of indices.

The way to homotopy invariance for our functors KKnuc (A, B) := KK(C; A, B)
in case that C := CPnuc (A, B) and the much more general KK(C; A, B) is the
following:
Our definitions of KKnuc (A, B) is the weakly nuclear variant (respective C-
compatible variant) of the cobordism-class ∼c definition for KK, cf. [73,
defs. 17.2.4, 17.3.1, 17.10.2]. Indeed, weakly nuclear (respectively C-compatible)
Kasparov modules E1 and E2 define the same element of KKnuc (A, B), if and only
if, there exists a nuclear (respectively C-compatible) Kasparov module E3 such that
E1 ⊕ E3 is isomorphic to a compact perturbation of E2 ⊕ E3 . This is the equivalence
relation ∼c of [73, def. 17.2.4] (in case where C := CP(A, B)). More precisely, the
relation ∼c in [73] is the stabilization of the relation ∼cp that allows also addition
of degenerate elements. “Degenerate” Kasparov modules become zero even for the
stabilization of the relation ≈u of unitary equivalence in [73, def.17.2.1], but here
automatic inside Enuc (A, B) (respectively inside the semigroup E(C; A, B)): The
difference between the nuclear (respectively our more general C-compatible) case
and the usual KK-theory relation ∼c is, that the added module E3 has to be a
nuclear (respectively a C-compatible) Kasparov module as well.
We say that Kasparov modules (E, φ, F0 ) and (E, φ, F1 ) are operator homo-
topic if there exists a norm-continuous map t ∈ [0, 1] 7→ F (t) ∈ L(E) such that
F (0) = F0 , F (1) = F1 and that (E, φ, F (t)) is a Kasparov module for every t ∈ [0, 1].
The same arguments as in [73, sec. 17.10] show that operator homotopic nuclear
(respectively C-compatible) Kasparov modules define the same element of our group
KKnuc (A, B) (respectively of KK(C; A, B)).
The result that operator homotopic C-compatible Kasparov modules are stably
equivalent, implies that the Kasparov product is well-defined and is invariant under
operator homotopy. In fact, the proofs in [73] or in [389, sec. 2.2] show that Kas-
parov products E1 ⊗ E2 of representatives E1 and E2 of elements [E1 ] ∈ KKnuc (B, C)
and [E2 ] ∈ KK(C, D) define up to operator homotopy an element of KKnuc (B, C)
if B and C are separable. One has also that the Kasparov product defines a bi-
additive map
KK(A, B) × KKnuc (B, C) → KKnuc (A, C)
for σ-unital A, B and C and separable B. Its values are in KKnuc (B, D) respectively
KKnuc (A, C), and if A = B or D = C the multiplication with [idB −0] or [idC −0]
 2. ISOMORPHISM OF Ext(C; A, B) AND KK(C; A, B(1) ) 865

gives the identity homomorphism of KKnuc (B, C). Therefore, our KKnuc -functor
is homotopy invariant and coincides with the KKnuc -functor of Skandalis [726].
There are natural homomorphisms from KKnuc (A, B) into KK(A, B), as the defi-
nition tells us ( 6 ).
Certainly, KKnuc (A, B) = KK(A, B) if A or B is nuclear, because then every Kas-
parov (A, B)-module is nuclear.
We outline now the changes of the Kasparov approach, that are necessary in
the case of KK(C; A, B). Recall the other equivalence relations ∼c , ∼cp , ∼s , ∼soh ,
∼oh and ∼h and on E(C; A, B) that are defined as in [73, chp. 17] on all elements
of E(A, B) by

(∼cp :) E0 ∼cp E1 if there are degenerate E2 , E3 ∈ E(C; A, B), such that

E0 ⊕ E2 ∼scp E1 ⊕ E3 .

(∼c :) (E0 , φ0 , F0 ) ∼c (E1 , φ1 , F1 ) if [(E0 , φ0 , F0 )] = [(E1 , φ1 , F1 )] in

KK(C; A, B) = Gr(E(C; A, B)/ ∼scp ) .

In [73, chp. 17]: ‘‘stabilized version of ∼cp ’’

(∼soh :) (E0 , φ0 , F0 ) ∼soh (E1 , φ1 , F1 ) if there is a C *-morphism ψ : A → L(E)
and an operator-norm continuous map t ∈ [0, 1] 7→ F (t) ∈ L(E) such
that (E, ψ, F (t)) ∈ E(C; A, B) for every t ∈ [0, 1] and that (Ei , φi , Fi ) is
(unitarily) isomorphic to (E, φ, F (i)) for i = 0, 1.
(∼oh :) (E0 , φ0 , F0 ) ∼oh (E1 , φ1 , F1 ) if there are degenerate E2 , E3 ∈ E(C; A, B),
a Hilbert B-module E, a C *-morphism ψ : A → L(E) and a norm-
continuous map t ∈ [0, 1] 7→ F (t) ∈ L(E) such that (E, ψ, F (t)) ∈
E(C ; A, B) for every t ∈ [0, 1] and that (Ei , φi , Fi ) ⊕ Ei+2 is (unitarily)
isomorphic to (E, φ, F (i)) for i = 0, 1.
(∼h :) It is the equivalence relation that is generated by the reflexive, but not
necessarily transitive ???, relation ∼h0 :
(E0 , φ0 , F0 ) ∼h0 (E1 , φ1 , F1 ) if there exists

(E, ψ, F ) ∈ E(C ⊗ CP(C, C[0, 1]) ; A, B ⊗ C([0, 1]))

such that (Ei , φi , Fi ) is isomorphic to (E(i), ψi , F (i)) for i = 0, 1.

Straight calculations show that all this relations are compatible with addition in
E(C; A, B)/ ≈u .
The relation ∼cp is an equivalence relation on E(C; A, B), because the
degenerate Kasparov modules in E(C; A, B)/ ∼scp build a sub-semigroup of
E(C; A, B)/ ∼scp .
The relation ∼oh is an equivalence relation, because the degenerate Kasparov
modules in E(C; A, B)/ ∼soh build a sub-semigroup of E(C; A, B)/ ∼soh .

6 We don’t know if this homomorphisms are injective in general and if their images are

naturally complemented in KK(A, B).

866 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

Notice that the Kasparov module (E, ψ, F ) in the definition of ∼h can be

considered as a “continuous” family {(E(t), ψt , F (t))}0≤t≤1 in E(C; A, B), if we
consider the B[0, 1]-module E as a complemented B[0, 1]-submodule of H bB[0,1] ∼
=
C([0, 1], H)
b

but the unitary equivalences at the endpoints can not be specified ??????????
See proofs of part (vii) of Lemma 8.2.3.

Temporary, we define semi-groups

KK∗ (C; A, B) := E(C; A, B)/ ∼∗

for ∗ = h, oh, cp, c.

Lemma 8.2.3. Suppose that A and B are σ-unital and graded, C ⊆ CP(A, B)
is countably generated point-norm closed operator convex cone, with C ◦ βA ⊆ C
and βB ◦ C ⊆ C.

Then:

(i) The unitary equivalence classes E(C; A, B)/ ≈u build an Abelian semi-
group under addition of Kasparov modules.
(ii) The relations ∼h , ∼oh , ∼c , ∼cp , ∼scp and ≈u (respectively the rela-
tions ∼oh , ∼soh and ∼scp ) are successively stronger, and are equiva-
lence relations on E(C; A, B) that are all compatible with addition in
E(C; A, B)/ ≈u .
Thus, the set of equivalence classes E(C; A, B)/ ∼∗ are semi-groups,
and there are natural semi-group epimorphisms from E(C; A, B)/ ≈u onto
E(C; A, B)/ ∼∗ for each of the above listed equivalence relation ∼∗ .
If C 0 ⊆ C, then there are natural morphisms from E(C 0 ; A, B)/ ∼∗ into
E(C; A, B)/ ∼∗ .
(iii) KK∗ (C; A, B) := E(C; A, B)/ ∼∗ are groups for ∗ = oh, h and there are
natural semigroup epimorphisms

KKc (C; A, B) → KKoh (C; A, B) → KKh (C; A, B) .

The natural epimorphism E(C; A, B)/ ∼scp → E(C; A, B)/ ∼cp defines a
group isomorphism Gr(E(C; A, B)/ ∼scp ) ∼ = Gr(E(C; A, B)/ ∼cp ).
KKc (C; A, B) is a sub-semigroup of Gr(E(C; A, B)/ ∼cp )), and
the kernel of KKc (C; A, B) → KKoh (C; A, B) consists of the classes
[(E, φ, F )]c such that there is a degenerate element (E0 , φ0 , F0 ) ∈
E(C; A, B) with (E, φ, F ) ⊕ (E0 , φ0 , F0 ) ∼soh (E0 , φ0 , F0 ).
(iv) If f ∈ Hom(B, C) is a grading preserving epimorphism, let f∗ C := C(f )◦C
the point-norm closed matrix operator-convex cone generated by f ◦ C :=
{f ◦ V ; V ∈ C}. Then

f∗ [(E, φ, F )] := [(E ⊗
b f C, φ(·)⊗
b f 1, F ⊗
b f 1)]
 2. ISOMORPHISM OF Ext(C; A, B) AND KK(C; A, B(1) ) 867

defines semigroup morphism f∗ : E(C ; A, B)/ ≈u → E(f∗ C ; A, C)/ ≈u that

is compatible with all relations ∼∗ .
(v) If g ∈ Hom(D, A) is grading preserving let g ∗ C := {V ∈ CP(D, B) ; V ◦
g ∈ C } , then
g ∗ [(E, φ, F )] := [(E, φ ◦ g, F )]
defines semigroup morphism g ∗ : E(g ∗ C ; D, B)/ ≈u → E(C ; A, B)/ ≈u that
is compatible with all relations ∼∗ .
(vi) Let CPin (B, B) ⊆ CP(B, B) denote the cone of all approximately inner
c.p. maps of B.
If C and D are separable nuclear graded C*-algebras, then the map
 
[(E, φ, F )] 7→ (B ⊗E,
b idB ⊗φ,b 1⊗Fb )

maps E(C, D)/ ≈u into E(CPin (B, B) ⊗

b CP(C, D); B ⊗
b C, B ⊗
b D) and de-
fines a morphism
τB : KK(C, D) → KKoh (CPin (B, B)⊗
b CP(C, D); B ⊗C,
b B ⊗D)
b

with τB ([k − 0]) = [(idB ⊗k)

b − 0] for grading preserving k ∈ Hom(C, D).
In particular, [χt ] = [χt −0] ∈ KK(C([0, 1]), C) for χt (f ) := f (t) maps
to
τB ([χt ]) = [χB
t ] ∈ KKoh (CPin (B, B)⊗ CP(C[0, 1], C); B ⊗ C([0, 1]), B) ,
b b

where χB B
t denotes the epimorphism χt : f ∈ C([0, 1], B) 7→ f (t) ∈ B
(t ∈ [0, 1]).
Since [χt ] = [χ0 ] by [73, lem.18.5.2],it follows that [χB B
0 ] = [χt ] for
t ∈ [0, 1].
(vii) Let HbB := HB ⊕B HB with grading β∞ ⊕ (−β∞ ), where β∞ (x1 , x2 , . . .) :=
(βB (x1 ), βB (x2 ), . . .) for (x1 , x2 , . . .) ∈ HB ∼
= (B ⊗ K)(1 ⊗ p11 ).
There is a construction λ =: λB that assigns to each class [E] =
[(E, φ, F )] ∈ E(C; A, B)/ ≈u a special class λ([E]) ∈ E(C; A, B)/ ≈u that
has representative (H bB , ψ, U ) with U = U ∗ = U −1 , such that λ is additive,
E ∼cp λ(E) in E(C; A, B), λ(E) is degenerate if E is degenerate, λ(E1 ) ∼scp
λ(E2 ) if E1 ∼scp E2 , and E = (E, φ, F ) ∈ E(C[0, 1]; A, B[0, 1]) implies
(pt )∗ (λB[0,1] [E]) = λB ((pt )∗ [E]) .
Moreover, if E ∼=H bB then ψ is unitarily equivalent to φ ⊕ 0 by an unitary
W ∈ L(HB ) of degree zero, and, if F = F ∗ = F −1 then λ(H
b bB , φ, F ) is
isomorphic to (HbB , φ⊕s,t 0, F ⊕s,t (−F )) for (any) isometries s, t ∈ L(HbB )
of even degree.
In particular, all equivalence-operations for the relations ∼∗ can be
elaborated inside the class of special Kasparov modules (H bB , ψ, F ) with
∗ −1
F =F =F .
(viii) If M(B) contains isometries s, t of even degree with ss∗ + tt∗ = 1, then
addition in E(C; A, B)/ ≈u is compatible with
 : B ⊕∞ B → B , (a, b) 7→ sas ∗ +tbt∗ ,
868 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

i.e., the composition ∗ ◦ Σ of the map

Σ : [E(A, B)] × [E(A, B)] → [E(A, B ⊕ B)]

given by

Σ : ((E1 , φ1 , F1 ), (E2 , φ2 , F2 )) 7→ (E1 ⊕∞ E1 , φ1 (·) ⊕ φ2 (·), F1 ⊕ F2 )

with ∗ defines the addition in E(A, B)/ ≈u . In particular, 2[E] = δ ∗ [E] in

E(C; A, B)/ ≈u for E ∈ E(C; A, B) and δ(b) := sbs∗ + tbt∗ .

Proof. (i): The unitary equivalence classes E(C; A, B)/ ≈u build an Abelian
semigroup under addition of Kasparov modules, because E(A, B)/ ≈u is an Abelian
semigroup (is in particular a set), the sum of C-compatible Kasparov (A, B)-
modules is C-compatible, and the class of C-compatible Kasparov modules is in-
variant under unitary equivalence.
(ii): It is easy to check that ≈u , ∼scp and ∼soh are equivalence relations on
E(C ; A, B) that are compatible with sums of Kasparov modules.
If (E, φ, F 0 ) ∈ E(A, B) is a φ-compact perturbation of (E, φ, F ) ∈ E(C; A, B),
then (E, φ, F (t)) is in E(C; A, B) for F (t) := tF + (1 − t)F 0 , thus (E, φ, F 0 ) ∼soh
(E, φ, F ) in E(C; A, B), i.e., ∼scp is stronger than ∼soh on E(C; A, B).
Thus ∼soh , ∼scp and ≈u are successively stronger equivalence relations ∼∗ on
E(A, B) that are compatible with the property that (E, φ, F ) ∼∗ (E 0 , φ0 , F 0 ) and
(E, φ, F ) ∈ E(C; A, B) implies (E 0 , φ0 , F 0 ) ∈ E(C; A, B).
It is easy to see, that the infinite repeat E∞ := (E ⊗ `2 ; φ(·) ⊗ 1, F ⊗ 1) is a
Kasparov (A, B)-module, if and only if, E = (E, φ, F ) is a degenerate Kasparov
module. Then E∞ ⊕ E ≈u E∞ , and E is a degenerate element in E(C; A, B) if and
only if E∞ is in E(C; A, B).
compare above and below! ??
∼ (H ⊗ E, 1 ⊗ φ, 1 ⊗ F )
Infinite (B-module) sums (E, φ, F ) ⊕ (E, φ, F ) ⊕ · · · =
of degenerate Kasparov modules (E, φ, F ) ∈ E(C; A, B) are degenerate Kasparov
modules in E(C; A, B), and (E, φ, F ) ⊕ (H ⊗ E, 1 ⊗ φ, 1 ⊗ F ) is unitary isomorphic
to (H ⊗ E, 1 ⊗ φ, 1 ⊗ F ), i.e., [(E, φ, F )] = 0 in Gr(E(C; A, B)/ ∼scp ) Thus E ⊕
(E, φ, F ) ∼c E for every E ∈ E(C; A, B). If follows that ∼c is stronger than ∼oh .
Hence, there is a natural semigroup epimorphisms KKc (C; A, B) → KKoh (C; A, B).
Since the classes of degenerate modules in E(C; A, B) build the sub-semigroups
of E(C; A, B)/ ∼scp and E(C; A, B)/ ∼sop of elements that are additively absorbed
by some other element, we get that ∼op and ∼cp are successively stronger additive
equivalence relations on E(C; A, B) that are weaker than ∼sop respectively ∼scp .
Furthermore, KK(C ; A, B) = Gr(E(C ; A, B)/ ∼cp ) ,

Gr(E(C ; A, B)/ ∼soh ) = Gr(E(C ; A, B)/ ∼oh ) ,

 2. ISOMORPHISM OF Ext(C; A, B) AND KK(C; A, B(1) ) 869

and there is semigroup monomorphism KKc (C; A, B) ,→ KK(C; A, B), and semi-
group epimorphisms

E(C; A, B)/ ∼cp → KKc (C; A, B) → KKoh (C; A, B)

and a semigroup morphism KKoh (C; A, B) → Gr(E(C; A, B)/ ∼soh ) such that its
image generates Gr(E(C; A, B)/ ∼soh ). All its images are sub-semigroups that
generate the groups.
???????????????????????
The relation ∼soh is stronger than ∼oh by definition, and KKoh (C; A, B) =
E(C; A, B)/ ∼oh is a group ???? (if we include unitary equivalence and calculate
modulo addition of a degenerate module in E(C; A, B)), because ∼oh is compat-
ible with unitary equivalence, compact perturbation and addition of degenerate
elements.
???????????????????.
??
The relation ∼h is an equivalence relation, because homotopy ∼h0 is reflexive
and ≈u is an equivalence relation. Since homotopy “inside” E(C; A, B) are additive
as elements of E(C ⊗ CP(C, C[0, 1]); A, B ⊗ C[0, 1]), and since unitary equivalence
is compatible with addition of Kasparov modules, we get that ∼h is compatible
with addition in E(C; A, B)/ ≈u . Thus KKh (C; A, B) := E(C; A, B)/ ∼h is a
commutative semi-group.
The relation ∼soh is stronger than ∼h on E(C; A, B), because an operator
homotopy t 7→ (E, φ, F (t)) in E(C; A, B) defines an element (E ⊗C[0, 1], φ(·)⊗1, F )
in E(C([0, 1]); A, C([0, 1], B)) that defines a homotopy between (E, φ, F (0)) and
(E, φ, F (1)). Here:

F := {F (t)} ∈ C([0, 1], M(K(E))) ⊆ Cb,st ([0, 1], M(K(E))) ∼

= L(E ⊗ C[0, 1])
By definition of the relation ∼oh , E ∼oh ({0}, 0, 0) if E ∈ E(C ; A, B) is degenerate.
The same happens also for ∼h , because the homotopy connecting degenerate
E := (E, φ, F ) ∈ E(A, B) and the 0-module 0 := ({0}, 0, 0) , that is given the proof
of [73, prop.17.2.3], is in E(C([0, 1]); A, C([0, 1, B)) if E ∈ E(C; A, B). Thus ∼oh is
stronger than ∼h .
Finally, there are natural semigroup epimorphisms

E(C; A, B)/ ≈u → KKc (C; A, B) → KKoh (C ; A, B) → KKh (C; A, B) .

If (E, φ, F ) ∈ E(C ; A, B) then (E, φ ◦ βA , −F ) ∈ E(C; A, B), because C ◦ βA ⊆ C by

our assumptions.
The sum (E, φ, F ) ⊕ (E, φ ◦ βA , −F ) is operator-homotopic to a dege-
nerate element (cf. [73, prop.17.3.3]). Thus, the semigroups KKoh (C ; A, B)
and KKh (C ; A, B) are groups, and the above natural semigroup morphism
KKoh (C ; A, B) → Gr(E(C ; A, B)/ ∼sop ) is an isomorphism of groups.
to be filled in: proof of 8.2.3
870 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

Let ∼soh denote the equivalence relation generated by ≈u and operator ho-
motopies t ∈ [0, 1] 7→ (E, φ, F (t)) ∈ E(C2 ; B, C). Then ∼soh is compatible with
addition and is stronger than ∼oh .
Since KKoh (C ; B, C) = E(C2 ; B, C)/ ∼oh is a group, we get that there is a
natural epimorphism from Gr(E(C2 ; B, C)/ ∼soh ) onto KKoh (C; B, C). This must
be an isomorphism, because 0 = [E] ∈ Gr(E(C2 ; B, C)/ ≈u ) for every “degenerate”
E ∈ E(C2 ; B, C).
If C1 ⊆ C2 , then (obviously) E(C1 ; B, C) ⊆ E(C2 ; B, C).
For the relations ∼∗ =≈u , ∼scp or ∼soh It is easy to see, that elements Ej ∈
E(C1 ; B, C) (j = 1, 2) satisfy E1 ∼∗ E2 in E(C1 ; B, C), if and only if, E1 ∼∗ E2 in
E(C2 ; B, C).
Thus E(C1 ; B, C)/ ∼∗ is a subgroup of E(C2 ; B, C)/ ∼∗ . If we apply the
Grothendieck functor, we get natural semigroup morphisms KKc (C1 ; B, C) →
KKc (C2 ; B, C) and KKoh (C1 ; B, C) → KKoh (C2 ; B, C) .
In the case of the relation ∼∗ =∼h , one has at least that E1 ∼h E2 in E(C1 ; B, C)
implies that E1 ∼h E2 in E(C2 ; B, C), because E(C1 ⊗CP(C, C[0, 1]) ; B, C([0, 1], C))
is contained in E(C2 ⊗ CP(C, C[0, 1]) ; B, C([0, 1], C)).
??
Thus, if C1 ⊆ C2 , there are natural semigroup morphisms
KK∗ (C1 ; B, C) → KK∗ (C2 ; B, C) .
This semi-group morphisms are in general neither injective nor surjective.
??
semigroup epimorphisms
KKc (C2 ; B, C) → KKoh (C2 ; B, C) → KKh (C2 ; B, C) .

The natural injections B ∼

= C ⊗ p11 ⊆ B and C ∼ = C ⊗ p11 ⊆ C ⊗ K define
isomorphisms from KKoh (C3 ; B ⊗ K, C) onto KKoh (C2 ; B, C) respectively from
KKoh (C2 ; B, C) onto KKoh (C4 ; B, C ⊗ K), where the m.o.c. cones C3 ⊆ CP(B ⊗
K, C) and C4 ⊆ CP(B, C ⊗ K) are the natural extensions of C2 ⊆ CP(B, C) to the
stabilizations.
more???
(vi): Let χB B
t denote the epimorphisms χt : f ∈ C([0, 1], B) 7→ f (t) ∈ B (t ∈
[0, 1]), and let CPin (B, B) ⊆ CP(B, B) denote the cone of approximately inner c.p.
maps of B, and S := CPin (B, B) ⊗ CP(C[0, 1], C). Then the difference construction
7
(B , χBt , 0) is an Element in ( )



E S; C([0, 1], B), B = E CPin (B, B) ⊗ CP(C[0, 1], C) ; B ⊗ C[0, 1], B .
We denote by [χBt − 0] := [(B, χB
t , 0)] its class in KK(CPin (B, B); B ⊗
C([0, 1]), B) .
7The equation expresses only our conventions on notations, i.e., it is a trivial identity!
 2. ISOMORPHISM OF Ext(C; A, B) AND KK(C; A, B(1) ) 871

Since by ??? ??
C = (C ⊗ CP(C, C[0, 1])) ◦ (CPin (B, B) ⊗ CP(C[0, 1], C))
holds for all point-norm closed matrix operator-convex cones C ⊆ CP(A, B), one
has only to check that, for all σ-unital B and all t ∈ [0, 1] holds [χB B
0 − 0] = [χt − 0]
in KK(CPin (B, B); B ⊗ C[0, 1], B).
The cone CPin (B, B) is singly generated by {idB }. Thus, our theory applies
to KK(CPin (B, B); B ⊗ C[0, 1], B).

(E, φ, F ) 7→ (B ⊗E,
b idB ⊗φ,
b 1M(B) ⊗F
b )
defines an additive map from E(C, D) into E(CPin (B, B)⊗CP(C, D); B ⊗C, B ⊗D)
for trivially graded nuclear C and D (where e.g. the graded tensor product (B ⊗C,
b β)
coincides with (B ⊗ C, βB ⊗ idC )). It induces a group morphism τB : KK(C, D) →
KK(CPin (B, B); B ⊗ C, B ⊗ D). Obviously, τB ([ψ − 0]) = τB ([(D, ψ, 0)]) = [(B ⊗
D; id ⊗ψ, 0)] = [(id ⊗ψ) − 0] if ψ ∈ Hom(C, D).
We have the identities [(C, χt , 0)] = [χt − 0] = [χ0 − 0] in KKoh (C[0, 1], C) for
all t ∈ [0, 1], where χt : f ∈ C[0, 1] 7→ f (t) ∈ C, cf. [73, lem.18.5.2].
It is not difficult to see that the original proof of Kasparov of [χB 0 − 0] =
B
[χt −0] in KK(B⊗C[0, 1], B) works also in KK(CPin (B, B); B⊗C[0, 1], B), because
(B, χB
t , 0) = (B ⊗ C, idB ⊗ψt , 0) for the morphism ψt : g ∈ C[0, 1] 7→ g(t) ∈ C.

(vii): Let E op denote the Hilbert B-module E with grading −βE . Then
HbB ∼
= HB ⊕B HB . We use that there are unitary grading-preserving B-module
isomorphisms E ⊕B H bB ∼= HbB (Kasparov stabilization, [73, thm.14.6.1], [389,
op ∼ b
thm.1.2.12], HB ⊕B (HB ) = HB .
b b

We define λ[(E, φ, F )] for [E] = [(E, φ, F )] ∈ E(C; A, B)/ ≈u by the operations

λ = λ2 ◦ λ1 :
λ1 ([E, φ, F ]) := (E ⊕B H bB , φ ⊕ 0 , G1 (F ) ⊕ 0) ,
where G1 (F ) := g((F ∗ + F )/2) for the function g(t) = t on |t| ≤ 1 and g(t) := t/|t|
for |t| > 1. Then G1 (F ) = G1 (F )∗ and kG1 (F )k ≤ 1.
Note that (E, φ, G1 (F )) is degenerate if (E, φ, F ) is degenerate, G1 (F1 ⊕ F2 ) =
G1 (F1 ) ⊕ G1 (F2 ), G1 (U ∗ F U ) = U ∗ G1 (F )U for unitaries U , G1 (F2 ) is a φ-compact
perturbation of G1 (F1 ) if F2 is a φ-compact perturbation of F1 . and G1 (F )(t) =
G1 (F (t)) for F = {F (t)} ∈ L(E) if E = {E(t)} is a Hilbert B[0, 1]-module. Since
E ⊕(HbB , 0, 0) ∈ E(C; A, B), it follows from the arguments in [73, 17.4] that λ =: λB
has the quoted properties.
One can now go a step further and build from (E, φ, G) ∈ E(C; A, B) with
bB , G = G∗ and kGk ≤ 1 the Kasparov module
E := H
λ2 E, φ, (E ⊕B E op , φ ⊕ 0, U (G) ∈ E(C; A, B)

with U (G) ∈ M2 (L(E)) ∼

= L(E ⊕B E op ) where U (G) is the (selfadjoint) Halmos
unitary U (G) of G with U (G)11 := G, U (G)22 := −G and U (G)12 = U (G)21 :=
(1 − G2 )1/2 . Straight calculations show U (G)(βE (x), −βE (y)) = ((−βE ) ⊕
872 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

βE )(U (G)(x, y)) (i.e., U (G) has odd degree), and U (G) is a (φ ⊕ 0)-compact per-
turbation of G⊕−G. The graded commutator U (G)(φ(a)⊕0)−(φ(βA (a))⊕0)U (G)
is given by the 2 × 2-matrix with diagonal entries Gφ(a) − φ(βA (a))G and 0 and
with off-diagonal entries (1 − G2 )1/2 φ(a) and −(φ(βA (a))(1 − G2 )1/2 ), thus is in
K(E ⊕ E op ) ∼
= M2 (K(E)) (and is zero if (E, ψ, G) is degenerate). Now one can use
again that E ⊕ E op ∼=H bB . The operation λ2 also plays nicely together with unitary
equivalence, (Cuntz-)addition, φ-compact perturbation, operator homotopy and
homotopy (all inside the class of Kasparov C-modules with self-adjoint contractive
ψ-derivation G).
??
(viii):
(ix): ?????????????????? 

Lemma 8.2.4. Suppose that A and B are trivially graded and stable, A is sepa-
rable, B is σ-unital, and that C ⊆ CP(A, B) is a non-degenerate point-norm closed
m.o.c. cone that is countably generated.
Let B(1) denote B ⊕ B with odd grading βB⊕B (a, b) = (b, a).
Let H : A → M(B) defined by C such that δ∞ ◦ H is unitarily equivalent to H
and is non-degenerate.
Let s, t isometries in H(A)0 ∩ M(B) with ss∗ + tt∗ = 1.
H1 := H(·) ⊕ H(·) : A → M(B ⊕ B) = M(B) ⊕ M(B).

(o) CPnuc (A, B) is countably generated if A is separable and B is σ-unital.

(i) There is a natural grading preserving isomorphism B(1) ∼ = B ⊗C
b (1) , and
the graded standard Hilbert B(1) -module HB(1) is isomorphic to the B ⊕B-
b
module B ⊕ B with standard odd grading S(x, y) = (y, x).
Furthermore, K(B(1) ) ∼
= B(1) , and L(B(1) ) = M(B(1) ) = M(B)(1) .
(i.e., M(B) ⊕ M(B) with standard odd grading).
There is a grading-preserving unitary B(1) -module monomorphism γ
from B(1) ⊕B(1) B(1) onto B(1) that induces the Cuntz addition ⊕ on
M(B) ⊕ M(B), i.e.,

M(γ)((F1 , G1 ) ⊕B⊕B (F2 , G2 )) = (sF1 s∗ + tF2 t∗ , sG1 s∗ + tG2 t∗ )

for (F1 , G1 ), (F2 , G2 ) ∈ M(B) ⊕ M(B).

The morphism H1 : a ∈ A 7→ H(a) ⊕ H(a) ∈ M(B) ⊕ M(B) defines
a degenerate element (B(1) , H1 , 1 ⊕ (−1)) ∈ E(C; A, B(1) ).
(iii) The isomorphism classes of the Kasparov B(1) -modules (B(1) , H1 , U ), with
self-adjoint unitary U define a sub-semigroup Sp of E(C; A, B(1) )/ ≈u ,
such that the in E(C; A, B(1) )/ ≈u defined addition

[(B(1) , H1 , U1 )] + [(B(1) , H1 , U2 )] := [(B(1) , H1 , U3 )]

is realized by Cuntz addition (U1 , U2 ) 7→ U3 := SU1 S ∗ + T U2 T ∗ , where

S := (s, s) and T = (t, t).
 2. ISOMORPHISM OF Ext(C; A, B) AND KK(C; A, B(1) ) 873

It holds U = (1 − 2p, 2p − 1) with (1 − U )/2 =: p = p∗ = p2 if and

only if [p, H(A)] ⊆ B.
The semigroup morphism from Sp into

KKh (C; A, B(1) ) := E(C; A, B(1) )/ ∼h

is an epimorphism.
(iv) If we consider B as graded Hilbert B-module with even grading given by
the symmetry I = ss∗ − tt∗ ∈ M(B), then H bB ∼
= B. If V ∈ M(B)
is a selfadjoint unitary, then (B, H, V ) is in E(C; A, B) if and only if
U := s∗ V t is a unitary with [U, H(A)] ⊆ B. Then V = sU t∗ + tU ∗ s∗ .
The isomorphism classes of the Kasparov B-modules

(B, H, sU t∗ + tU ∗ s∗ ) ,

where U is a unitary in M(B) with [U, H(A)] ⊆ B build a sub-semigroup

Su of E(C; A, B)/ ≈u , such that the addition

[(B, H, sU1 t∗ + tU1∗ s∗ )] + [(B, H, sU2 t∗ + tU2∗ s∗ )] = [(B, H, sU3 t∗ + tU3∗ s∗ )]

in E(C; A, B(1) )/ ≈u is realized by Cuntz addition

(U1 , U2 ) 7→ U3 := sU1 s∗ + tU2 t∗ .

The semigroup morphism from Su into

KKh (C; A, B) := E(C; A, B)/ ∼h

is an epimorphism.

Proof. (o): ?????????????

The m.o.c. cone CPnuc (A, B) is countably generated if A is separable and B is
σ-unital, because it is generated by the c.p. map Vϕ,e (a) := ϕ(a)e, where e ∈ B+
is a strictly positive contraction and ϕ ∈ A∗+ is a faithful positive functional.
sn s∗n
P
(i): Since B is stable, there are isometries s1 , s2 , . . . ∈ M(B) with
strictly convergent to 1, cf. Remark 5.1.1(8). Then µ(b) := (s1 b, s2 b, . . .) defines an
isometric B-module isomorphism from B onto HB with inverse µ−1 (b1 , b2 , . . .) =
P ∗
n sn bn .

HB(1) is HB⊕B with grading given by

((x1 , y1 ), (x1 , y2 ), . . .) 7→ ((y1 , x1 ), (y1 , x2 ), . . .) ,

and HB⊕B is B ⊕ B-module isomorphic to HB ⊕∞ HB by

((x1 , y1 ), (x1 , y2 ), . . .) 7→ ((x1 , x2 , . . .), (y1 , y2 , . . .))

This isomorphism becomes grading-preserving if we take on HB ⊕∞ HB the standard

odd grading (x, y) 7→ (y, x). It follows that HB(1) is isomorphic to B ⊕ B with odd
grading β(x, y) = (y, x).
Thus H bB ∼ (B ⊕ B) ⊕B⊕B (B ⊕ B) with grading β((x1 , y1 ), (x2 , y2 )) :=
(1) =

((y1 , x1 ), (−y2 , −x2 )).

874 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

Since B is stable, M(B) contains a copy

C ∗ (s, t ; ss∗ + tt∗ = s∗ s = t∗ t = 1)

of O2 unitally. The map ν((x1 , y1 ), (x1 , y2 )) := (sx1 + tx2 , sy1 − ty2 ) defines a
grading preserving B(1) -module map from (B ⊕ B) ⊕B⊕B (B ⊕ B) with grading β
onto B(1) .
Clearly L(B(1) ) = M(B(1) ) = M(B)(1) and K(B(1) ) = B(1) .
Since B is trivially graded, the graded tensor product B ⊗C b (1) is isomorphic to
B ⊗ (C ⊕ C) with grading idB ⊗α, where α(z1 , z2 ) = (z2 , z1 ) is the grading of C(1) .
b (1) ∼
Thus, B ⊗C = B(1) in this case.
(ii):
(iii):
(iv): 

The next proposition implies that relations ∼c , ∼oh and ∼h on E(C; A, B)

are the same if A is separable and B is σ-unital. We use the above introduced
conventions on the extensions of the m.o.c. cone C.

Proposition 8.2.5. Suppose that A, B, C are σ-unital graded C*-algebras,

where A is separable, and let C ⊆ CP(A, B) and C1 ⊆ C2 ⊆ CP(B, C) denote non-
degenerate countably generated point-norm closed matricial operator-convex cones
with C ◦ βA ⊆ C, βB ◦ C ⊆ C, Cj ◦ βB ⊆ Cj and βC ◦ Cj ⊆ Cj for j = 1, 2.

(i) The Kasparov product

[(E 0 , φ0 , F 0 )] ◦ [(E, φ, F )] = [(E 0 ⊗ b B 1 , , F 0 ]F )]

b B E , φ(·)⊗

defines a bi-additive map

KKoh (C; A, B) × KKoh (C2 , B, C) 7→ KKoh (C2 ◦ C; A, C) .

The natural injections B ∼ = C ⊗ p11 ⊆ B and C ∼ = C ⊗ p11 ⊆ C ⊗

K define isomorphisms from KKoh (C2 ; B ⊗ K, C) onto KKoh (C2 ; B, C)
respectively from KKoh (C2 ; B, C) onto KKoh (C2 ; B, C ⊗ K).
(ii) The Kasparov C-group KK(C; A, B) := Gr(E(C; A, B)/ ∼scp ) is homotopy
invariant in the – stronger – sense that the natural epimorphisms

KKc (C; A, B) → KKoh (C; A, B) → KKh (C; A, B)

and the natural monomorphism KKc (C; A, B) ,→ KK(C; A, B) are iso-

morphisms.
The groups KK(C ⊗ CP(C[0, 1], C); C([0, 1], A), B), KK(C; A, B) and
KK(C([0, 1]); A, C([0, 1], B)) are naturally isomorphic.
In particular, our above defined bi-functor

KKnuc (A, B) = KK(CPnuc (A, B); A, B)

 2. ISOMORPHISM OF Ext(C; A, B) AND KK(C; A, B(1) ) 875

is homotopy invariant, and coincides with the KKnuc -functor of Skandalis

[726]. Therefore KKnuc (A, B) is stable and is half exact (on semi-split
short exact sequences) in each variable.
(iii) There are natural isomorphisms
KK(C; A, B(1) ) ∼
= KK(C; A, SB) ∼
= KK(C; SA, B),
KK(C; A ⊗ K, B) ∼
= KK(C; A, B) ∼
= KK(C; A, B ⊗ K) .
In particular,
KKnuc (A, B(1) ) ∼
= KKnuc (A, SB) ∼
= KKnuc (SA, B) .

Proof. To deduce the ???????

??
By the Kasparov product construction of an connection F1 ]F2 , the Kasparov
module (E1 ⊗ b B E2 , φ1 ⊗
b B idE2 , F1 ]F2 ) exists (in E(A, C)) for (E1 , φ1 , F1 ) ∈
E(A, B) and (E2 , φ2 , F2 ) ∈ E(B, C), because A is separable. It is an element
in E(C2 ◦ C; A, C), if (E1 , φ1 , F1 ) ∈ E(C; A, B) and (E2 , φ2 , F2 ) ∈ E(C2 ; B, C),
cf. Proposition ?? where C2 ◦ C denotes the m.o.c. cone that is generated by the
compositions W ◦ V for V ∈ C and W ∈ C2 .
Since the connection F1 ]F2 is uniquely defined up to operator homotopy, it
defines the desired bi-additive map from KKoh (C; A, B) × KKoh (C2 ; B, C) into
KKoh (C2 ◦ C; A, C).
(ii): By definition two elements Ej = (Ej , φj , Fj ) ∈ E(C; A, B) (j = 0, 1)
are cobordant in E(C; A, B), if there are (E, φ, G) in E(C; A, B) and a partial
isometry v ∈ L(E) = M(K(E)) of degree 0, commuting with φ(A), such that
[v, G]φ(A) ⊆ K(E) and that the “restricted” Kasparov modules (E, φ, G)1−vv∗ :=
(pE, φ(·)|pE, pG|pE) and (E, φ, G)1−v∗ v are (unitary) equivalent to E0 respectively
E1 .
Then the system (E, φ, G, v) is a C–cobordism between E0 respectively E1 (inside
E(C; A, B)).
Inspections of the arguments in the proofs of [73, sec.17.10] shows that the
relation ∼c on E(C; A, B) is equivalent to the relation “cobordant” on E(C; A, B),
and that the relations ∼c is the same as ∼oh .
Indeed, the cobordism between 0 and (E, φ, F ) for a degenerate (E, φ, F 0 ) ∈
E(C; A, B) that is operator homotopic to (E, φ, F ) is given by (E ⊗`
b 2 , φ⊗1,
b G, 1⊗T
b )
(where T ∈ L(`2 ) is the Toeplitz operator and G is suitable). It is also C–cobordism
in E(C; A, B) between 0 and (E, φ, F ).
(Note that it suffices to prove that elements E ∈ E(C; A, B) which are operator
homotopic to a degenerate element in E ∈ E(C; A, B) are also cobordant to (0, 0, 0),
because this implies that KKc (C; A, B) is a group with the same representatives
(−E) for the inverse elements of [E]c as for [E]oh , and then, that E 0 ∼oh E implies
that E 0 ⊕ (−E) ⊕ E1 ∼soh E2 for degenerate E1 and E2 . Thus E 0 ⊕ (−E) ∼c 0 and
E ⊕ (−E) ∼c 0. Hence E 0 ∼oh E implies E 0 ∼c E by the cancelation property of ∼c .
876 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

The relation ∼op is different from ∼c because Enuc (C; A, B)/ ∼op has in general
not cancelation.)
Since KKc (C; A, B) is a sub-semigroup of


KK(C; A, B) := Gr E(C; A, B)/ ∼scp
that generates KK(C; A, B), it follows that the natural group morphism
KK(C; A, B) → KKoh (C; A, B)


is an isomorphism, i.e., our maximal group KK(C; A, B) = Gr E(C; A, B)/ ∼scp
is naturally isomorphic to
KKoh (C; A, B) := Gr(E(C; A, B)/ ∼oh ))
if A is separable and B is σ-unital.
If the invariance of the classes in KK(C; A, B) under unitary homotopy of its
representatives in E(C; A, B) is established for all separable A and all σ-unital B
and all (point-norm closed non-degenerate) operator-convex cones C ⊆ CP(A, B),
then Kasparov’s proof of the homotopy-invariance works also for KK(C; A, B) and
its special cases KKnuc (A, B) and KKnuc (X; A, B) :

Then one can use the (formally larger) classes of Kasparov modules (E, φ, F ) ∈
E(C; A, B) given by by the equivalence relation ∼oh (i.e., given by unitary isomor-
phisms and operator homotopy) on E(C; A, B).
This allows to use Kasparov products to show finally the homotopy invariance.
The Kasparov product reduces its proof to Lemma 8.2.3(vi) as follows:
shorten/complete above and below given arguments ??
The homotopy invariance of KK(C; A, B) = KKoh (C; A, B) finally follows from
Lemma 8.2.3(vi). Indeed, let
E := (E, φ, F ) = {(Et , φt , Ft )} ∈ E(CP(C, C[0, 1]) ⊗ C; A, C([0, 1], B)) .
The Kasparov modules (Et , φt , Ft ) = χB t (E, φ, F ) define the same element
B
[(Et , φt , Ft )] = (χt )∗ [(E, φ, F )] of KK(C; A, B) for t ∈ [0, 1].
This is because, for t ∈ [0, 1], [χt − 0] = [χ1 − 0] ∈ KK(C[0, 1], C) and
(χB
t )∗ [(E, φ, F )]
is the the same as the Kasparov product
[E] ⊗B⊗C[0,1] [(B, χB
t , 0)] = [E] ⊗ τB ([χt − 0]) .

Notice here that the smallest point-norm closed m.o.c. cone Cmin (C[0, 1], C)
that contains all characters
χt : f ∈ C[0, 1] 7→ f (t) ∈ C
is nothing else all of CP(C[0, 1], C) ∼
= C[0, 1]∗+
The m.o.c. cone CPin (B, B) of all approximately inner c.p. maps V from B to
B is generated by the identity map idB of B, and has the property that for each
point-norm closed m.o.c. cone C ⊆ CP(A, B) holds CPin (B, B) ◦ C = C.
 2. ISOMORPHISM OF Ext(C; A, B) AND KK(C; A, B(1) ) 877

check above and below formula!

Let C (1) := CPin (B, B)⊗b CP(C[0, 1], C) ⊆ B⊗?????? the m.o.c. cone tensor
product generated by { idB ⊗χt } in CP(B ⊗ C[0, 1], B), i.e., the point-norm closed
m.o.c. cone that is generated by idB ⊗ CP(C[0, 1], C) and CP(C[0, 1], C).
This construction reduces the proof of homotopy invariance of general
KK(C; A, B) to the check of the homotopy invariance of KK(C (1) ; B[0, 1], B),
which in turn reduces to KK(C[0, 1], C) ∼
= Z.
(iii): ????????
Part (i????) of ??????????
implies in particular, that our KKnuc -groups are the same as those of Skandalis,
[726]. Then periodicity (ii) comes now in the same way as it comes for Kasparov’s
KK-functor.
But there is ???????????? 

The following Proposition 8.2.6 is crucial for the later needed corollaries of the
homotopy invariance of our groups Ext(C; A, B). Therefore, we give a detailed
proof, that uses the natural isomorphism KKc (C; A, B(1) ) ∼ = KKh (C; A, B(1) ) and
the isomorphism from Ext(C; A, B) onto the kernel ( 8 ) of the group homomorphism

K0 (H0 (A ⊗ K)0 ∩ Qs (B)) → K0 (Qs (B)) .

Proposition 8.2.6. Suppose that A is separable and stable, that B is σ-

unital and stable, that both are trivially graded, and let C ⊆ CP(A, B) a point-norm
closed matrix operator-convex cone that is countably generated, non-degenerate and
faithful.

(i) Ext(C; A, B) is naturally isomorphic to KK(C; A, B(1) ), such that the iso-
morphism Φ(C; A, B) from Ext(C; A, B) onto KK(C; A, B(1) ) satisfies

Φ(δ ◦ C; A, C) ◦ δ∗ = (δ(1) )∗ ◦ Φ(C ; A, B)

and
Φ(C ◦ γ; A1 , B) ◦ γ ∗ = γ ∗ ◦ Φ(C ◦ γ; A1 , B)
for isomorphisms δ : B → C and γ : A1 → A.
In particular, Extnuc (A, B) is naturally isomorphic to KKnuc (A, B(1) ).
(ii) Ext(C(R+ ); A, C0 (R+ , B)) ∼= 0.
In particular, Extnuc (A, C0 (R+ , B)) ∼
= 0.
∼
(iii) KK(C(R+ ); A, C0 (R+ , B)) = 0.
(iv) Let H0 : A → Qs (B) the canonical C*-morphism associated to C and
S, T ∈ H(D)0 ∩ M(B) isometries with SS ∗ + T T ∗ = 1. Consider the
isometries s := πB (S) and t := πB (T ) in H0 (A)0 ∩ Qs (B).

8 Notice that Ext(A, B) is not equal to K (H (A ⊗ K)0 ∩ Qs (B)) itself. But one can calculate
0 0
Ext(A, B) as quotient by its subgroup K0 (Ann(H0 (A ⊗ K), Qs (B))). Use Lemma 4.2.20(o) to
compare it with our definition.
878 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

Then there exists a constant γ(C) < ∞ such that for each unitary
u ∈ U0 (H0 (D)0 ∩Qs (B)) the geodesic distance cel(sus∗ +tt∗ ) of the unitary
sus∗ + tt∗ to 1 inside U0 (H0 (D)0 ∩ Qs (B)), satisfies

cel(sus∗ + tt∗ ) ≤ γ(C) .

Proof. Recall that C := CPnuc (A, B) is singly generated, non-degenerate and

faithful if A is separable and B is σ-unital, because V := ρ(·)b0 is a generates
CPnuc (A, B) as m.o.c. cone if ρ ∈ A∗+ is a faithful positive functional on A and b0 ∈
B+ is a strictly positive element. (The more general m.o.c. cones CPrn (X; A, B) are
often not countably generated despite CPrn (X; A, B) itself is a matrix-hereditary
sub-cone of CPnuc (A, B).)
Notice that CPnuc (A, B) ⊗ CP(C, D) = CPnuc (A ⊗ C, B ⊗ D) for all nuclear
C *-algebras C and D by the local characterization in Proposition ??, and that


KKnuc (A, B ⊗C)
b = KK CPnuc (A, B ⊗ C); A, B ⊗C
b for all (graded) separable nu-
clear C. Further recall that

Extnuc (A, B) := Gr(SExtnuc (A, B)) ∼

= Ext(CPnuc (A, B); A, B) ,

with SExtnuc (A, B) := [Homl-nuc (A, Q(B ⊗ K))] ∼

= SExt(CPnuc (A, B); A, B), and
KKnuc (A, B) = KK(CPnuc (A, B); A, B) right from the definitions.
We proceed with the KK(C; ·, ·) and Ext(C; ·, ·) for general countably generates
point norm closed m.o.c. cones C ⊆ CP(A, B).
It follows that there is a natural isomorphism from KK(C; A, B(1) ) onto
Ext(C; A, B) for separable stable A, σ-unital stable B, (both trivially graded),
with countably generated non-degenerate and faithful m.o.c. cone C ⊆ CP(A, B).
It covers also the cases C := CPnuc (A, B) and C := CPrn (X; A, B), i.e., it suffices to
consider the (more general) case of a non-degenerate faithful countably generated
m.o.c. cone C in place of CPnuc (A, B).
(i): Life could be easier if a simply and elementary argument would show that
the stabilization ∼ of the relation that are given by unitary equivalence and by ψ-
“compact” perturbations on the below considered pairs (ψ, q) with ψ : A → M(B)
weakly C-compatible and q ∈ M(B) a projection with qψ(a)−ψ(a)q ∈ B (equipped
with with Cuntz addition ⊕ := ⊕S,T with a suitable copy C ∗ (S, T ) of O2 ) are the
same as the stabilization of relations induced on them by considering them modulo
B ⊗ K (up to unitary equivalence), i.e., as the corresponding classes of elements
that build our Ext(C; A, B) defined in Definition 5.8.2. The critical point is, that
one has to make sure that the relation ∼ is weak enough such that πB (p1 φ1 (·)p1 ) =
πB (p2 φ2 (·)p2 ) implies (φ1 , p1 ) ∼ (φ2 , p2 ).
In our approach we use that KKc (C; A, B(1) ) = KKh (C; A, B(1) ) to get the
relation (ψ1 , p1 ) ∼ (ψ2 , p2 ) from the homotopy invariance of KKc (C; A, B(1) ).
check next remark, give cross-ref
 2. ISOMORPHISM OF Ext(C; A, B) AND KK(C; A, B(1) ) 879

By Lemma 5.9.14, the general defining relations for classes of representing

elements (φ, p) that build the elements of Ext(C; A, B) are:

(φ0 , p0 ) ∼ (φ1 , p1 ) ⇔ ∃ (ψ0 , 1), (ψ1 , 1), ∃ U ∈ M(B) : (2.1)

∗
∀a ∈ A : p0 φ0 (a)p0 ⊕ ψ0 (a) − U (p0 φ1 (a)p0 ⊕ ψ1 (a))U ∈ B . (2.2)

Here ⊕ means Cuntz-addition in M(B).

From Lemma 5.9.14: Recall:
Let C ⊆ CP(A, B) is a countably generated point-norm closed matrix operator-
convex cone. We denote by S the set of pairs (ϕ, p), where ϕ : A → M(B) is a
C *-morphism with b∗ ϕ(·)b ∈ C for all b ∈ B and p ∈ M(B) is a projection that
satisfies ϕ(a)p − pϕ(a) ∈ B for all a ∈ A.
We denote by [S] the set of unitary equivalence classes [(ϕ, p)] by unitaries in
M(B).
An element (ϕ, p) ∈ S is degenerate if pϕ(a) = ϕ(a)p for all a ∈ A.

(i) [S] is a commutative semigroup (with Cuntz-Addition on maps and pro-

jections).
(ii) The map [ϕ, p] 7→ πB (ϕ(·)p) ∈ Hom(A, Q(B)) defines an additive semi-
group morphism ϑ from [S] onto SExt(C; A, B). The image of ϑ contains
[0] + SExt(C; A, B).
(iii) The map S → Ext(C; A, B) induces an equivalence relation ∼ on S that
is compatible with Cuntz addition and is generated by the following op-
erations and relations:
(a) unitary equivalence by unitaries in M(B), i.e., (ϕ, p) ∼ (ψ, q) if there
is a unitary u ∈ M(B) with ψ = u∗ ϕ(·)u and q = u∗ pu ,
(b) addition (ϕ ⊕ ψ, p ⊕ q) of degenerate elements (ψ, q) ∈ S to elements
(ϕ, p) ∈ S , i.e., (ϕ ⊕ ψ, p ⊕ q) ∼ (ϕ, p) , and
(c) unitary perturbation: (ϕ, p) ∼ (ϕ, q) if there is b∗ = b ∈ M(B) such
that ϕ(a)b − bϕ(a) ∈ B and (p − e−ib qeib )ϕ(a) ∈ B for all a ∈ A.
(iv) The natural semigroup morphism from SExt(C; A, B) into Ext(C; A, B)
is an epimorphism.
(v) If C is countably generated and non-degenerate, and if H : A⊗K → M(B)
is as in Corollary 5.4.4, then the unitary equivalence classes [(H(·), p)] ∈
[S] with p = T T ∗ for some isometry T ∈ M(B) build a sub-semigroup
[P] of [S] such that ϑ maps [P] onto Ext(C; A, B).

Recall that almost directly from the definitions one can see that there are
natural isomorphisms Ext(C; A, B) ∼ = Ext(C; A ⊗ K, B ⊗ K) (in particular this
happens for Extnuc ), see cf. Proposition 5.9.25.
One direction of the proof (such as in [73, 17.6]) is almost trivial, as the reader
can see from the last part of our proof. But for the other direction we make two
times very essential use of the homotopy invariance of KK(C; A, B(1) ), and use
the trivial isomorphism C([0, 1], B(1) ) ∼
= C([0, 1], B)(1) . Our proof is a compromise
880 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

between the suggestions in [73] before [73, prop. 17.6.5] and the detailed exposition
in [389, chp. 3] (both for ordinary KK and Ext).
We make use of the stability of the bi-functors KK(C; ·, ·) and Ext(C; ·, ·) to get
a constructive picture, that allows to see the C-compatibility of all constructions.
In particular, we suppose from now on that B is stable and σ-unital, and let D :=
A ⊗ K, E := Q(B) := M(B)/B that is naturally isomorphic to Qs (B) (because B
is stable).

Remark 5.8.7 and ?????????????????? ??

show that there are natural isomorphisms

Extnuc (A, B) ∼
= Extnuc (D, B) ∼
= G(H0 ; D, E) ,

respectively,

Ext(C; A, B) ∼
= Ext(C ⊗ CP(K, C); D, B) ∼
= G(H0 ; D, E) ,

where C is naturally extended to D = A ⊗ K as C ⊗ CP(K, C) ⊆ CP(D, B),

H0 = πB ◦ H1 , and H1 : D → M(B) is an infinite repeat of a faithful non-
degenerate representation of D in M(K) ⊆ M(B) (compare Corollary 5.4.10 in case
C = CPnuc (D, B)), or where H1 : D → M(B) is constructed from the countably
generated cone C by Corollary 5.4.4. Thus, there is a copy of O2 unitally contained
in the commutant of H1 (D) in M(B) by Lemma 5.1.2(ii), i.e., H0 (D)0 ∩ E contains
a unitally liftable copy of O2 . By Proposition 4.4.3(i), G(H0 ; D, E) is naturally
isomorphic to the kernel of K0 (H0 (D)0 ∩ E) → K0 (E) .
Furthermore, since C := H0 (D)0 ∩ E contains a unitally liftable copy of O2 ,
every element of this kernel has a representative q = p + B ∈ C which comes from
a projection p ∈ M(B), because q = q 0 ⊕ 1 ⊕ 0 is unitarily equivalent to 1 ⊕ 0 by
a unitary u ∈ U0 (E) = πB (U(M(B))) if [q 0 ] = 0 in E, but [q] = [q 0 ] in K0 (C),
cf. Lemma 4.2.6(iv,b).
Thus the representatives are projections p ∈ M(B) with pH1 (a) − H1 (a)p ∈ B
for every a ∈ D. By Lemma 4.2.6(i), the addition in K0 (H0 (D)0 ∩ E) coincides with
the Cuntz addition (if we use a unital copy of O2 in E that commutes element-wise
with H0 (D)). Let

P := {p ∈ M(B) ; p = p∗ = p2 , [p, H1 (a)] ∈ B ∀ a ∈ A } .

The set P is closed under Cuntz addition defined by a unital copy of O2 in

H1 (D)0 ∩ M(B) , and it is invariant under unitary equivalence by unitaries in the
−1
commutant of H1 (D) in M(B) and, more generally, by unitaries in πB (H0 (D)0 ∩
s
Q (B)). Therefore, the unitary equivalence classes of projections in P by unita-
ries in H1 (D)0 ∩ M(B) build a semigroup by Proposition 4.3.2. This semigroup
maps additively onto Ext(C; D, B) (respectively onto Extnuc (D, B)) by the above
consideration.
We are now going to describe the equivalence relation which is induced by this
additive map into K0 (C):
 2. ISOMORPHISM OF Ext(C; A, B) AND KK(C; A, B(1) ) 881

By Lemma 4.2.6(iv,b), two representatives p and p0 define the same element of

K0 (H0 (D)0 ∩ E) if and only if the Cuntz sums 0 ⊕ p ⊕ 1 and 0 ⊕ p0 ⊕ 1 are unitarily
equivalent modulo B by a unitary u ∈ M(B) that can be connected to 1 by a norm-
continuous path ξ ∈ [0, 1] 7→ u(ξ) ∈ M(B) of unitaries u(ξ) which all commute
element-wise with the image of H1 modulo B. Another equivalent condition is:
0 ⊕ p ⊕ 1 and 0 ⊕ p0 ⊕ 1 are (norm-)homotopic in the set P (the projections q of
M(B) with qH1 (a) − H1 (a)q ∈ B for every a ∈ D). (Because the latter implies
that their images in C := H0 (D)0 ∩ E are homotopic inside the projections of C.)
But a more delicate description of the equivalence relation on P induced by
P → K0 (H0 (D)0 ∩ E) comes from Propositions 5.5.12 and 4.4.3:
The *-morphism H0 : D → E dominates zero, because H1 (D) + B is sta-
ble, cf. Proposition 5.5.12(ii). Projections p, p0 ∈ P define the same class in
K0 (H0 (D)0 ∩ E), if and only if, there is a unitary u ∈ M(B) such that [u, H1 (a)]
and (u∗ (p0 ⊕ 1 ⊕ 0)u − (p ⊕ 1 ⊕ 0))H1 (a) are in B for every a ∈ D , cf. Proposition
4.4.3(iii).
By obvious reasons (e.g. consider the given map from P into K0 (H0 (D)0 ∩ E)
itself), the corresponding equivalence relations on P are compatible with Cuntz
addition.
Now we define a semigroup epimorphism from ([P], +) onto KKnuc (A, B(1) ) =
KKnuc (D, B(1) ) . (We show later that this map induces the same equivalence rela-
tions as the map from [P] to K1 (C).)
As in [73, 17.6.4], one can see that the elements of KK(C; D, B(1) ) (respectively
of KKnuc (D, B(1) )) can be represented by (HB ⊕∞ HB , φ, G) where
X
HB := {(b1 , b2 , . . .) ; bn ∈ B, b∗n bn ∈ B}

and the grading is given by β(x, y) = (y, x) for x, y ∈ HB , φ is weakly C-compatible

(respectively is nuclear ), and G is a selfadjoint unitary in L(HB ⊕∞ HB ) = M(B)⊕
M(B) = M(B)(1) of degree one. Thus, G = (U, −U ), U = U ∗ = U −1 ∈ M(B),
and φ(a)(x, y) = (φ1 (a)x, φ2 (a)y) satisfies ψ := φ1 = φ2 .
The passage to a selfadjoint unitary G ∈ M(B)(1) respects unitary
isomorphism classes and is additive on E(C; D, B(1) )/ ≈u (respectively in
Enuc (D, B(1) )/ ≈u ), the operation is compatible with “compact” perturbations and
transforms each homotopy into a homotopy. The same happens with the passage
from a graded B(1) -module E and φ : D → L(E) to ψ : D → HB ⊕ HB with help of
Kasparov stabilization: (E, φ, F ) ⊕ ((HB )(1) , 0, 1) ≈ ((HB )(1) , φ0 , F 0 ) for suitable
weakly C-compatible φ0 : D → L((HB )(1) )
???????.
??
But then Hilbert B-module addition becomes Cuntz addition with a unital
copy of O2 with degree zero generators. Thus the equivalence relations from our
definition becomes unitary equivalence with an unitary of degree zero and compact
perturbation, i.e., passage to an selfadjoint unitary F 0 with degree = 1 and (F 0 −
882 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

F )φ(D) ⊆ B. KK(C; D, B) is then again the Grothendieck group of the semigroup

which is defined by this equivalence classes.
It follows L(HB ⊕ HB ) = M(B) ⊕ M(B), φ = ψ ⊕ ψ and F = (1 − 2q, 2q − 1),
where ψ : D → M(B) is a weakly nuclear (possibly degenerate) C *-morphism and
q is a projection in M(B) with qψ(a) − ψ(a)q for all a ∈ D. Conversely (E, φ, F )
is in Enuc (D, B) if E := HB ⊕ HB , φ := ψ ⊕ ψ and F = (1 − 2q, 2q − 1) for every
pair (ψ, q) with weakly nuclear (respectively ψ ∈ C) ψ : D → M(B) and projection
p ∈ M(B) such that qψ(a) − ψ(a)q for all a ∈ D.
The addition in KK(C; D, B(1) ) agrees under this identification with the Cuntz
addition of the projections q and of the homomorphisms ψ. Isomorphisms become
unitary equivalences and “compact” perturbation means passage to a projection
p0 ∈ M(B) with (p0 − p)ψ(D) ∈ B. Therefore (H1 , 1) and (H1 , 0) represent the
zero element, and we can represent the elements of KK(C; D, B(1) ) by the elements
(ψ ⊕ H1 , q ⊕ 1). Let ψτ : D → M(B) be a point-norm continuous family of weakly
nuclear C *-morphisms and let p ∈ M(B) be a projection such that ψτ (a) − ψ0 (a)
and [ψ0 (a), p] are in B for all a ∈ D and τ ∈ [0, 1]. Then the homotopy invariance
of the KKnuc -functor implies that (ψ0 , p) and (ψ1 , p) represent the same element of
KK(C; D, B(1) ).
By Theorem 5.6.2(ii) – applied to C := H1 (A) and T = ψ ◦ (H1−1 ) –, we
find a norm-continuous map t 7→ U (t) into the unitaries of M(B) such that
U (t)∗ H1 (a)U (t) − (ψ(a) ⊕ H1 (a)) ∈ B for t ∈ R+ and

lim kU (t)∗ H1 (a)U (t) − (ψ(a) ⊕ H1 (a))k = 0

t→∞

for a ∈ A. Thus (U (0)∗ H1 U (0), q ⊕ 1) and (ψ ⊕ H1 , q ⊕ 1) represent the same

element of KK(C; D, B(1) ) ∼= KK(C; A, B(1) ) . But (U (0)∗ H1 U (0), q ⊕1) is unitarily
equivalent to (H1 , p) where p := U (0)(q⊕1)U (0)∗ , and p is in P. Thus every element
of KK(C; D, B(1) ) can be represented by (H1 , p), where p ∈ P.
Conversely, the maps

p ∈ P 7→ (H1 , p) 7→ (HB ⊕ HB , H0 ⊕ H0 , (1 − 2p, 2p − 1))

define together an additive map from P onto E(C; D, B(1) ).

If we use the copy of O2 in the commutant of H1 (D), then (H1 , p ⊕ 1 ⊕ 0) is
nothing else (H1 ⊕ H1 ⊕ H1 , p ⊕ 1 ⊕ 0), and represents therefore the same element
of KK(C; D, B(1) ) as it (H1 , p) does.
Let p, p0 ∈ P. If t ∈ [0, 1] 7→ u(t) is a norm-continuous map into the uni-
taries which element-wise commutes with H1 (D) modulo B, and if u(0) = 1 p0 −
u(1)∗ pu(1) ∈ B then (H1 , p) and (H1 , p0 ) define the same element of KK(C; D, B(1) )
because it defines a homotopy in the nuclear Kasparov modules followed by a “com-
pact” perturbation.
So far we have seen that p ∈ P 7→ (HB ⊕ HB , H1 ⊕ H1 , (1 − 2p, 2p − 1)) defines
a group epimorphism from Extnuc (D, B) onto KK(C; D, B(1) ).
 2. ISOMORPHISM OF Ext(C; A, B) AND KK(C; A, B(1) ) 883

It remains to show that the kernel is trivial, i.e., that [p] = 0 in K0 (H0 (D)0 ∩
Qs (B)) if (H1 , p) represents the zero of KK(C; D, B(1) ).
If (H1 , p) represents zero, then there is a pair (ψ, q) and a unitary u0 ∈ M(B)
such that H1 ⊕ ψ = u0 ψu∗0 and u0 qu∗0 is a H1 ⊕ ψ-“compact” perturbation of
p ⊕ q. We add to this relations the pair (H1 , 1). By Proposition 4.3.2 and Theorem
5.6.2(ii) (applied to C := H1 (A) and T := ψ ◦ H1−1 ), we find then unitaries u1 and
u2 such that H1 ⊕ (ψ ⊕ H1 ) = u1 (ψ ⊕ H1 )u∗1 , u∗2 H1 (a)u2 − (ψ ⊕ H1 )(a) ∈ B for
a ∈ D, and such that u1 (q ⊕ 1)u∗1 is a H1 ⊕ (ψ ⊕ H1 )-“compact” perturbation of
p ⊕ (q ⊕ 1).
Now we consider the relations modulo B in E := Qs (B) := cM (B)/B, and
let vj := πB (uj ), ϕ := πB ◦ ψ, p0 := πB (p) and q 0 := πB (q). Note that H0 ⊕
H0 = H0 , if we use a copy of O2 in H0 (D)0 ∩ Qs (B). We get v2∗ H0 v2 = ϕ ⊕ H0 ,
v1∗ (1 ⊕ v2 )∗ H0 (1 ⊕ v2 )v1 = v2∗ H0 v2 and

(q 0 ⊕ 1) − v1∗ (p0 ⊕ (q 0 ⊕ 1))v1 · v2∗ H0 v2 = {0} .

It follows that the unitary w := (1⊕v2 )v1 v2∗ and the projection r := v2 (q 0 ⊕1)v2∗
are in H0 (D)0 ∩ E, and that (r − w∗ (p0 ⊕ r)w)H0 (D) = {0}. Let p1 := w∗ (p0 ⊕ r)w.
If π denotes the quotient map, then It gives [p1 ] = [r] = 0 in K0 (Qs (B)) and
(p1 − r))H0 (D) = 0, [π(p)] = [p0 ] = [p1 ] − [r] in K0 (H0 (D)0 ∩ Qs (B)).
By Proposition 4.4.3(iii), [p1 ] = [r] in K0 (H0 (D)0 ∩ Qs (B)), because H0 domi-
nates zero. Thus [π(p)] = 0 in K0 (H0 (D)0 ∩ Qs (B)).
(ii): The functor

Y 7→ Ext(C(Y ); A, C0 (Y, B)) ∼

= KK(C(Y ); A, C0 (Y, B)(1) )

is homotopy invariant, because C0 (Y, B)(1) = C0 (Y, B(1) ) and the map Y 7→
KK(C(Y ); A, C0 (Y, C)) is homotopy invariant for each non-degenerate point-norm
closed m.o.c. cone C ⊆ CP(A, C) for any graded σ-unital graded C *-algebra C, by
Proposition 8.2.5.
(iii): Ext(C(R+ ); A, C0 (R+ , B)) ∼
= 0 follows from (ii), because C0 (R+ , B(1) ) is
a contractible.
(iv):
TEXT of (iv):
Let H0 : A → Qs (B) denote the canonical C *-morphism associated to C and
S, T ∈ H(D)0 ∩ M(B) isometries with SS ∗ + T T ∗ = 1. Consider the isometries
s := πB (S) and t := πB (T ) in H0 (A)0 ∩ Qs (B).
Then there exists a constant γ(C) < ∞ such that for each unitary u ∈
U0 (H0 (D)0 ∩ Qs (B)) the geodesic distance cel(sus∗ + tt∗ ) of the unitary sus∗ + tt∗
to 1 inside U0 (H0 (D)0 ∩ Qs (B)), satisfies

cel(sus∗ + tt∗ ) ≤ γ(C) .

884 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

Proof, To be filled in ?? 

Next Corollary 8.2.7 is important and has to be checked: ?? Move

parts of the Cor. (8.b2) to proof !!!
Recall Definitions (4.2) and (4.3) for a C *-subalgebra C ⊆ D of a C *-algebra
D and an ideal I / D of D: The derivation C*-subalgebra Der(C, I) ⊆ E is defined
by
Der(C, I) := {e ∈ D ; ea − ae ∈ I ∀ a ∈ C} ,
and the normalizer C*-subalgebra N (C, I) ⊆ E is defined as the hereditary C *-
subalgebra
N (C, I) := {e ∈ D ; ea, ae ∈ I ∀ a ∈ C} .
The definitions show that I ⊆ N (C, I) and that N (C, I) is a closed ideal of
Der(C, I).

Corollary 8.2.7. Suppose that A and B are stable C*-algebras (with trivial
gradings βA = idA and βB = idB ), A is separable, B is σ-unital and that C ⊆
CP(A, B) is a non-degenerate m.o.c. cone that is countably generated.
Let H : A → M(B) a non-degenerate C*-morphism with δ∞ ◦H unitarily equiv-
alent to H such that Vb : a ∈ A 7→ b∗ H(a)b is in C for each b ∈ B and the c.p. maps
{Vb ; b ∈ B} generate C, existing by Corollary 5.4.4.
Further let S, T ∈ H(A)0 ∩ M(B) two isometries with SS ∗ + T T ∗ = 1 that exist
by Remark 5.1.1(8).
Define H0 : A → M(B)/B = Qs (B) by H0 := πB ◦ H.
The group KK(C; A, B) is natural isomorphic to the kernel of the group mor-
phism
K1 H0 (A)0 ∩ Qs (B) → K1 Qs (B) ∼



= K0 (B) .

Alternatively, with D := M(B) and C := H(A) and I := B, we get

KK(C; A, B) ∼


= K1 Der(H(A), B) .

The elements of KK(C; A, B) can be represented by unitaries in u ∈ M(B) that

satisfy u ∈ Der(H(A), B).
Next has to be checked again!
It would be a dream. But not reality?
Two unitaries u, v ∈ Der(H(A), B) define the same element of KK(C ; A, B),
if and only if, there exist n ∈ N and f1 , . . . , fn ∈ Der(H(A), B) with fj∗ = −fj
(j = 1, . . . , n) and

(u∗ v) ⊕S,T 1 = exp(f1 ) · . . . · exp(fn ) .

In particular, u ∈ U Der(H(A), B) represents the zero element, if and only if,
there exist n ∈ N and f1 , . . . , fn ∈ Der(H(A), B) with fj∗ = −fj (j = 1, . . . , n) and

u ⊕S,T 1 = exp(f1 ) · . . . · exp(fn ) .

 2. ISOMORPHISM OF Ext(C; A, B) AND KK(C; A, B(1) ) 885

Proof. A description that is a straight reformulation of the picture coming

from the use of Kasparov modules and generalized Fredholm operators is given
by classes of elements F ∈ Der(H(A), B) with 1 − F ∗ F, 1 − F F ∗ ∈ N (H(A), B),
i.e., F + N (H(A), B) is a unitary in Der(H(A), B)/ N (H(A), B) and [F ] = [G] for
G ∈ Der(H(A), B) with 1 − G∗ G, 1 − GG∗ ∈ N (H(A), B) if and only if (G∗ F ⊕S,T
1) + N (H(A), B) is in U0 (Der(H(A), B)/ N (H(A), B)). The addition is defined by
[F1 ] + [F2 ] := [F1 F2 ].
Equivalently expressed:
Vj := Fj + N (H(A), B) are unitaries in Der(H(A), B)/ N (H(A), B).
And [F1 ] = [F2 ] if and only if (1 ⊕ V2 ) − (1 ⊕ V1 ) ∈ N (H(A), B)
???
Other approach ???
F1 , F2 ∈ Der(H(A), B) with 1−Fj∗ Fj , 1−Fj Fj∗ ∈ N (H(A), B) for some unitary
Fj + B = Vj ∈ M(B)/B, and [F1 ] = [F2 ] if and only if

1 − (V1∗ V2 ⊕ 1) ∈ Ann(H0 (A), Qs (B))

with ⊕ = ⊕πB (S),πB (T ) , i.e., (V1 ⊕ 1) − (V2 ⊕ 1) ∈ Ann(H0 (A), Qs (B)).

The equivalence follows from the fact that the hereditary C *-subalgebra
N (H(A), B) contains B and πB (N (H(A), B)) = Ann(H0 (A), Qs (B)).
to be filled in ??
By assumptions, A and B are σ-unital, A is stable and H0 is non-degenerate.
It follows that the σ-unital C *-subalgebras B, H(A) and H(A) + B of M(B) are
stable by Remark 5.1.1(9).
It follows from ???? Prop. ?? ?? that there exist an isometry S1 ∈ M(B)
with S1∗ H(A)S1 ⊆ B. Thus, V := πB (S1 ) ∈ Qs (B) is an isometry, V ∗ V = 1, that
satisfies V ∗ H0 (A)V = {0}. In particular, Ann(H0 (A)) := Ann(H0 (A), Qs (B)) is a
full hereditary C *-subalgebra of Qs (B). It follows

K∗ (H0 (A)0 ∩Qs (B))/ Ann(H0 (A)) ∼= kernel{K∗ H0 (A)0 ∩Qs (B) → K∗ (Qs (B))} .

We use that N (H(A), B) is a full hereditary C *-subalgebra of M(B) and con-

tains B. If we apply Lemma 4.2.20 to them we obtain that there is a natural isomor-


phism from K∗ Der(H(A), B)/ N (H(A), B) onto the kernel of the homomorphism
K∗ (Der(H(A), B) → K∗ (M(B))) induced by the inclusion Der(H(A), B) ⊆ M(B).
Using that B ⊆ N (H(A), B) ⊆ Der(H(A), B) we obtain that

Der(H(A), B)/B = H0 (A)0 ∩ Qs (B) (2.3)

s
N (H(A), B)/B = Ann(H0 (A), Q (B)) . (2.4)
886 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

Above we have seen that Ann(H0 (A), Qs (B)) is full in Qs (B). All this together
yields

Der(H(A), B)/ N (H(A), B) ∼

= (H0 (A)0 ∩ Qs (B))/ Ann(H0 (A), Qs (B))

and that Ann(H0 (A), Qs (B)) is a full hereditary C *-subalgebra of Qs (B) ... MORE
!!! ??? 

3. Ext(SC; A, SB) and KK(C; A, B) considered as K1 -groups

We fix now a σ-unital C *-algebra B – later we suppose in addition that B is

stable (which plays in the formal notation no role). The separability is then needed
for emphour interpretation of the Kasparov product.
C-compatible we need ????
Let A an arbitrary separable C *-algebra and C ⊆ CP(A, B) a non-degenerate
countably generated point-norm closed matricial operator-convex cone (“m.o.c.
cone”). We define point-norm closed m.o.c. cones C(Y ) := C ⊗ CP(C, C0 (Y ))
and C(Z, Y ) := C ⊗ CP(C0 (Z), C0 (Y )) ⊆ CP(C0 (Z, A), C0 (Y, B)) for the natural
extensions of C to the tensor products C0 (Z, A) = A ⊗ C0 (Z) with help of the
algebras C0 (Z). This is compatible with our in Section 1 introduced canonical
extensions CD,E := C ⊗ CP(D, E) of C to a m.o.c. cone in CP(A ⊗ D, B ⊗ E).
In case Y = R we write sometimes SC in place of C(R), but mostly we simply
drop the specification to the spaces Y and Z and write simply C in place of C(Y )
or C(Z, Y ), – if it is visible to which considered spaces the m.o.c. cone has to be
canonically extended.
Recall from Chapter 3, Section ??, that the possible extension of C to a point-
norm closed m.o.c. cone in CP(A ⊗ M, B ⊗ N ) is unique if M and N are simple
and nuclear (cf. Chapter 3 exact refs ??). This applies in particular to the case
where M = K and N = K. We write also C for this extension C ⊗ CP(K, K) ⊆
CP(A ⊗ K, B ⊗ K).
It is important to understand clearly that the definitions in Chapter 3 give not
“tensor products” in the usual algebraic or topological sense, because it are only
point norm closed m.o.c. cones that are “generated” by a set S of tensor products,
and the algebraic hull of this “tensors” is usually not dense in C(S).
Our below given results show that we only need to know that our func-
tors Y 7→ KKnuc (A, C0 (Y, B)), or Y 7→ KKnuc (X; A, C0 (Y, B)) respectively
Y 7→ KK(C(Y ); A, C0 (Y, B)) exists (and are homotopy invariant with respect to
the locally compact spaces Y ) to get the needed facts for our later applications
of Extnuc , Extnuc (X; ·, ·), Ext(C; A, ·) and to the Rørdam groups R(C; A, ·) in the
proofs of Theorems B and M.
 3. Ext(SC; A, SB) AND KK(C; A, B) CONSIDERED AS K1 -GROUPS 887

For the proofs of the general results in Chapter 9 we even need only that
Ext(C(R+ ); A, C0 (R+ , B)) = 0 (respectively Extnuc (A, C0 (R+ , B)) = 0, respec-
tively Extnuc (X; A, C0 (R+ , B)) = 0). In fact, it would be enough to know that

Ext C(S 1 , R+ ) ; D ⊗ Cb (S 1 ) , B ⊗ C0 (R+ ) = 0

for D = A ⊗ K, because e.g. Extnuc (A, B) = Ext(C; A, B) take C = CPnuc (A, B),
and for Extnuc (X; A, B) take the cone CPrn (X; A, B) of ΨA -ΨB -residually nu-
clear maps from A to B, where A is separable, B is σ-unital, and the action
ΨB is (downward !) induced by an action of X on a separable non-degenerate
C *-subalgebra N ⊆ M(B) (which makes sure that CPrn (X; A, B) is countably
generated, cf. Chapter 3).
We know from Prop./Cor. in Chapter 3
Give exact references !!! ??
and Corollary 5.4.4 how to find a non-degenerate C *-morphism H : A ⊗ K →
M(B ⊗ K) which determines the point-norm closure of C – in the case of, closed
with respect to the point-norm topology, countably generated and non-degenerate
matrix operator-convex cones C ⊆ CP(A, B) and of separable A and σ-unital B.
find ref’s for above ??? ??
For example, if A and B are both separable, then C = CP(A, B) satisfies our
requirements on C, but CP(A, B) is in general not singly generated if A is separable
and B is only σ-unital.
We know from Corollary 5.6.1, cf. also Remark 5.1.1(8) and Lemma 5.1.2, that
we can find a strictly continuous unital *-monomorphism M(K) ,→ M(B ⊗ K) and
Next H1 is only related to C := CPnuc (D, B).
we use also the universal H1 : D := A ⊗ K →→ M(B) in ‘‘general
position’’ in the sense of Definition 3.3.1,
related to some specified C ⊆ CP(A, B)
that is ‘‘non-degenerate’’ and countably generated.
(... and ) for D := A ⊗ K a faithful *-monomorphism H1 : D → M(K) ⊆
M(B ⊗ K) such that H1 is universal for the cone CPnuc (D, B) and that there is a
copy of O2 unitally contained in H1 (D)0 ∩ M(K) ⊆ M(B ⊗ K) . Here D := A ⊗ K.
The general bi-module to be considered for non-degenerate m.o.c. cones C ⊆
CP(A, B) is given by the universal H0 : D → M(B ⊗ K) in general position – in
the sense of Definition 3.3.1, with the property the c.p. maps d ∈ D 7→ e∗ H0 (d)e ∈
B ⊗ K generates the natural extension of C to C ⊗ CP(K, K) ⊆ CP(A ⊗ K, B ⊗ K).

To simplify notation, let us assume again that B is stable in this section,

i.e., that B ∼
= B ⊗ K. Further we suppose from now on that B is σ-unital, that
∼
A = D := A ⊗ K is stable and separable.
We know that there is a given non-degenerate C *-morphism H : D → M(B)
with δ ◦ H unitarily equivalent to H, such that H corresponds to a non-degenerate
888 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

countably generated m.o.c. cone C in the sense of Corollary 5.4.4. We fix H and C
from now on.
Suppose from now on that Y is a closed subspace of R2 . There is a natural
unital strictly continuous embedding

M(B) ∼
= 1 ⊗ M(B) ⊆ Cb,st (Y ∪ {∞}, M(B)) ⊆ M(C0 (Y, B)) .

Let EY := M(C0 (Y, B))/ C0 (Y, B) and HY (a) := πC0 (Y,B) (H(a)) ∈ EY for a ∈ D.
Then B ∼= C0 ({0}, B), H0 := H{0} = πB ◦ H, HY (D)0 ∩ EY contains a unital copy
of O2 that comes from a copy of O2 which is unitally contained in (1 ⊗ H(D))0 ∩
M(C0 (Y, B)). In particular, [HY ⊕ HY ] = [HY ] for the unitary equivalence classes.
Further let C ⊆ CP(D, B) denote the point-norm closed m.o. convex cone, that is
generated by the maps Vb : a ∈ D 7→ b∗ H(a)b (b ∈ B), and let

C(Y ) := C ⊗ CP(C, C0 (Y )) ⊆ CP(D, B ⊗ C0 (Y ))

denote the natural extension of C to a cone in CP(D, C0 (Y, B)) (cf. Definition 3.6.16
and Corollary 3.6.20. Don’t mix it up with a sort of completion of algebraic sums
of tensors!).
Then Ext(C(Y ); A, C0 (Y, B)) ∼
= G(HY ; D, EY ) by Proposition ?? and Corol-
lary 5.4.4 by putting C(Y ) and HY : D → EY in place of C and H : D → M(B).
Now we introduce a construction that is needed in the proof of Lemma 8.3.2
and Corollary 8.3.3. The Lemma 8.3.2 will be used in the proof of Theorems B(i,ii)
and M(i,ii) in Chapter 9.
Note that Qs (SB) = Qs (C0 (R, B)) = ER is naturally isomorphic to the pull-
back construction of the split epimorphisms π0+ : E (+) → E (0) and π0− : E (−) → E (0)
at t = 0, where E (+) := ER+ = Qs (C0 (R+ , B)), E (−) := ER− = Qs (C0 (R− , B))
and E (0) := E{0} = Qs (B):
This results from the natural isomorphism Cb,st (Y, M(B)) ∼
= M(C0 (Y, B)) in
cases Y = R, Y = R+ , Y = R− and Y = {0}, by dividing by C0 (Y, B). The
restriction and evaluation maps are induced by the corresponding restriction and
evaluation of strictly continuous functions:
H : D → M(B) ∼ = 1⊗M(B) goes into the constant functions of Cb,st (Y, M(B))
and then down to EY = Qs (C0 (Y, B)). Therefore we must not always distin-
guish the various restrictions of “constant” elements in M(B) ⊆ Cb (Y, M(B)) ⊆
Cb,st (Y, M(B)) .
But the reader should not misunderstand the terminology “constant”, “re-
striction” and “evaluation”: The point-“evaluations” πt (b) can all be zero but
b ∈ Qs (SB) is non-zero, because that says only that b has a representative c ∈
Cb,st (R, M(B)) such that c(t) ∈ B for every t. In particular this happens for all
c ∈ Cb (R, B).
Notice that the C *-subalgebra of elements

= Cb,st (X, M(B)) ∼

c ∈ M(Cb (X, B)) ∼ = M(C0 (X, B))
 3. Ext(SC; A, SB) AND KK(C; A, B) CONSIDERED AS K1 -GROUPS 889

with πx (c) ∈ B for all x ∈ X for a non-discrete second countable locally compact
Hausdorff spaces X is always bigger than Cb (X, B) if B is σ-unital and stable.
The criterium for c ∈ Cb (X, B) is πx (c) ∈ B for all x ∈ X and x 7→ kπx (c)k is
an upper semi-continuous
As we have seen in Chapter 7, this does not imply that c is in Cb (R, B), and,
in particular, it does not mean that c is in C0 (R, B). For example, the natural
embedding  : M(B) → Qs (C0 (Y, B)) of the constant functions into the stable
corona of C0 (Y, B) will be defined as b ∈ M(B) 7→ b + C0 (Y, B), i.e. (M(B)) =
M(B) + C0 (Y, B). It is faithful for Y = R and Y = R+ , but the kernel of πt ◦  is
B for every t ∈ Y . But this is just the nice property that we can explore now:

Remark 8.3.1. The algebra ER := Qs (SB) is a weak variant of a sort of

“pull-back” of epimorphisms π0+ : E (+) → E (0) and π0− : E (−) → E (0) if B is σ-
unital, where we use the natural identities and epimorphisms from M(C0 (R− , B)) =
M(Cb (R− , B)) → M(B) and M(C0 (R+ , B)) = M(Cb (R− , B)) → M(B)
We can define for unitaries U in M(B) that commute element-wise modulo B
with H(D), i.e., with U H(d) − H(d)U ∈ B for all d ∈ D, in an almost trivial way
a weakly nuclear (respectively weakly C-compatible) extensions hU : D → ER if H
is weakly nuclear (respectively H is weakly C-compatible).
More generally, we find in Ext(CH ; D, SB) = G(HR ; D, ER ) extensions by the
following method:
ER is in a natural way isomorphic to the subalgebra of E (−) ⊕ E (+) which
consists of the pairs (e, f ) with π0− (e) = π0+ (f ), i.e., e = π− (g− ) := πB (g− (0)),
f = π+ (g+ ) := πB (g+ (0)) for

gα ∈ Cb,st (Rα , M(B)) ∼

= M(C0 (Rα , B))

(α ∈ {+, −}), and the restriction maps gα 7→ gα |0 := gα (0). Notice that

πB (g− (0)) = πB (g+ (0)) is equivalent to g+ (0) − g− (0) ∈ B.
Is ER really the pull-back of E (−) and E (+) ?
Is C (0) := HR (D)0 ∩ M(SB) the pull-back of C+ and C−
with restriction epimorphisms Cα → C|0 (α ∈ {−, +}), for

Cα(0) := HRα (D)0 ∩ M(C0 (Rα , B))

and
C (0) |0 = H(D)0 ∩ M(B) .
The map c ∈ C (0) → c(0) ∈ C (0) |0 is a splitting epimorphism
via λ : C (0) |0 → C (0) defined by
λ(d)(t) := d ∈ C (0) |0 for all t ∈ R .

Is it the same for DX := Der(HX (D); C0 (X, B)),

with X = R, X = R+ X = R− or X = {0}?
890 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

What about Der(HR (D); K)/ C0 (R, B) with

K := {f ∈ Cb (R, B) ; lim kf k = 0} .
t→−∞

Let U ∈ Der(H(D), B) ⊆ M(B) a unitary, and let hα U: D → E

(α)
, for α ∈
(−) (+) ∗
{+, −}, be defined by hU (a) := H(a)|R− and hU (a) := U H(a)U |R+ for a ∈ D,
where we use the natural inclusions of M(B) into ER , E (−) and E (+) .
(+) (−) (+) (−)
Then π0+ hU (a) = H(a) = π0− hU (a) for a ∈ D. Thus, hU and hU define a
homomorphism hU : D → ER .
If we use the – a bit inaccurate – restriction notation, we can express this by
hU (a)|R− = H(a)|R− and hU (a)|R+ = U H(a)U ∗ |R+ for a ∈ D.
It turns out that hU has a completely positive contractive lift T : D →
M(SB) ∼ = Cb,st (R, M(B)) that is 2-step-dominated by H : D → M(B) ⊆
Cb,st (R, M(B)):
Let f ∈ Cb (R) be defined by f (t) := 0 for t ≤ 0, f (t) := t on [0, 1] and f (t) := 1
for t ≥ 1, further let T (d)(t) := (1 − f (t))H(d) + f (t)U H(d)U ∗ .
√ √
Thus T (·) = e∗1 H(·)e1 + e∗2 H(·)e2 for e1 := 1 − f and e2 := U ∗ f , and T is
weakly nuclear if H is weakly nuclear (respectively T is weakly C-compatible if H
is weakly C-compatible).
If M(B) contains a copy of O2 unitally and if [H] = [H] + [H] (which is the
case by our assumptions), then there are isometries s1 , s2 ∈ H(D)0 ∩ M(B) which
are canonical generators of O2 , cf. Proposition 4.3.5(iii). Then T (·) = IU∗ H(·)IU
for the isometry IU = (1 − f )1/2 s1 + f 1/2 s2 U ∗ .
Since U H(d)U ∗ − H(d) ∈ B, T (d) − H(d) is an element of the ideal I1 := {b ∈
Cb (R, B) : b(t) = 0 ∀t ≤ 0}, and T (d)−U H(d)U ∗ is a element of the ideal I2 := {b ∈
Cb (R, B) : b(t) = 0 ∀t ≥ 1}. But (I1 + SB)/SB = Q(R+ , B) and (I2 + SB)/SB =
Q(R− , B), where we identify Q(R, B) naturally with Q(R− , B) ⊕`∞ Q(R+ , B) and
consider it as an ideal of ER . The natural epimorphism ER → E (−) maps Q(R+ , B)
to zero and ER → E (+) maps Q(R− , B) to zero.
One can see that T (d) + SB = hU (d), because this equality can be calculated
separately in E (−) and E (+) .
So, if H(D)0 ∩ M(B) contains a unital copy of O2 and if the unitary group of
M(B) is connected, we get a map U 7→ hU from the intersection U0 (Q(B))∩H0 (D)0
into S(HR ; D, ER ), where H0 := πB ◦ H.
Thus we have seen that, in the case where B is stable and σ-unital and D is
stable and separable, for every unitary U ∈ M(B) with U H(d) − H(d)U ∈ B for
all d ∈ D,
the homomorphism hU defines an element [hU ] of S(HR ; D, ER ) and satisfies
hU (d) − H(d) ∈ J for every d ∈ D, where J is the ideal of Qs (SB) which is defined
by Q(R+ , B).
 3. Ext(SC; A, SB) AND KK(C; A, B) CONSIDERED AS K1 -GROUPS 891

It follows that [hU ⊕ HR ] is an element of Ext(C(H); D, SB) = G(HR ; D, ER ).

With the above introduced notations J, EY , HY , F (Y ) and H0 we get the

following lemma.

Lemma 8.3.2. Suppose that D is separable and stable, that B is σ-unital and
stable, and that H : D → M(B) is a non-degenerate faithful C*-morphism with
δ∞ ◦ H unitarily equivalent to H.
Furthermore, suppose that the following groups for R+ := [0, ∞) are trivial:

Ext(C; D, C0 (R+ , B)) = G(HR+ ; D, ER+ ) = {[HR+ ]} ∼

= 0.
Let [g], [T ] ∈ S(HR ; D, ER ) with [T ] = [HR ], and let s1 , s2 ∈ ER canonical genera-
tors of a copy of O2 that is unitally contained in ER . Then:

(i) There exists a unitary u ∈ ER such that

u∗ T (a)u − (g(a) ⊕s1 ,s2 T (a)) ∈ J for every a ∈ D ,

where J is the image of the natural embedding of Q(R+ , B) into

Q(R, B) ⊆ ER .
(ii) If σ : R → R is an order preserving homeomorphism of R, then
σ ◦ (g ⊕ T )] = [g ⊕ T ] in S(HR ; D, ER ).
[b

This applies to the case where H : D → M(B) corresponds to a countably gen-

erated non-degenerate m.o.c. cone C – as e.g. to C := CPnuc (A, B) if A is separable
and B is σ-unital – because then Ext(C; D, C0 (R+ , B)) ∼= 0, by Corollary 5.9.21
and Proposition 8.2.6.

Proof. We keep the notations of Remark 8.3.1. The Cuntz addition will be
taken at first with respect to fixed isometries s1 and s2 in HR (D)0 ∩ ER which
generate a copy of O2 in HR (D)0 ∩ ER . Then HR ⊕ HR = HR .
We show at the end of the proof of Part (i) the independence of the result from
the chosen representatives H for for [HR ] in G(HR ; D, ER ) and from the chosen
isometries s1 , s2 in HR (D)0 ∩ ER with s1 s∗1 + s2 s∗2 = 1.
(i): Recall that C0 (Rα , B) ∼
= C0 ((0, 1], B),

E (α) := ERα = Qs (C0 (Rα , B))

for α ∈ {−, +} and E (0) = Qs (B).

If g : D → ER := Qs (SB) defines an element in S(HR , D, ER ), then this means
that there is an isometry s ∈ ER with s∗ HR (d)s = g(d) for all d ∈ D, i.e., such
ss∗ ∈ HR (D)0 ∩ ER .
In fact we know at the beginning only that, by definition of SExt(C[R]; D, SB),
unify notations for C[X] !!!
892 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

there is a c.p. contraction V : D → M(SB) with πSB ◦ V = g and a∗ V (·)a ∈

C[R] for each a ∈ C0 (R, B) =: SB. From that we have obtained in Chapters 3 and
4 that there exists an isometry T ∈ M(SB) with T ∗ H(d)T − V (d) ∈ SB for all
d ∈ D. Thus s = πSB ◦ T has the desired property.
If we restrict now g(d) and s to R− and R+ , we get isometries s− ∈ E (−) ,
s ∈ E (+) and homomorphisms g − : D → E (−) and g + : D → E (+) such that
+

(sα )∗ Hα sα = g α for α ∈ {+, −} where H+ := H[0,∞) and H− := H(−∞,0] are

the “restrictions” of HR . Thus [g α ] ⊕ [Hα ] ∈ G(Hα , D, E (α) ) for α ∈ {+, −}. By
the definition of E (+) and E (−) and by assumption, we get [g α ⊕ Hα ] = [Hα ].
But this means that there are unitaries U + ∈ E (+) and U − ∈ E (−) such that
g α ⊕ Hα = (U α )∗ Hα U α for α ∈ {+, −}.
If we let H0 := πB ◦ H we get in particular

(π0+ (U + ))∗ H0 π0+ (U + ) = (π0− (U − ))∗ H0 π0− (U − )

at zero. From now on we use the shorter (but by Remark 8.3.1 somehow inaccurate)
notation
U + (0) := π0+ (U + ) and U − (0) := π0− (U − ).

Thus U + (0)U − (0)∗ ∈ Qs (B) is a unitary in the commutant H0 (D)0 ∩ Qs (B) of

H0 (D) in Qs (B). U − (0) is in the image of the unitaries of E (−) by π0− and U + (0)
is in the image of the unitaries of E (+) by π0+ .
We have K∗ (M(C0 (R+ , B))) = 0, because B is stable. This follows from
Lemma 5.1.2(ii) as in the proof of Lemma 4.6.5(ii), or from the contractibility of
the unitary groups of multiplier algebras of stable algebras, [557], [180]. Since
C0 (R+ , B) ∼= C0 (R− , B) is contractible, we get K1 (E (+) ) = 0 and K1 (E (−) ) =
0 from the 6-term exact sequence of K-theory. Therefore the epimorphisms π0+ ,
respectively π0− , maps unitaries of E (+) , respectively of E (−) , into the unitaries u
of E (0) := Qs (B) with [u] = 0 in K1 (Qs (B)). Thus [U + (0)U − (0)∗ ] = [U + (0)] −
[U − (0)] = 0 in K1 (Qs (B)), and, therefore, by Lemma 4.2.6(v,2), (U − (0)U − (0)∗ )⊕1
is in the connected component of 1 in the unitaries of Qs (B). Let V ∈ M(B) a
unitary lift of (U + (0)U − (0)∗ ) ⊕ 1. Then H(a)V − V H(a) ∈ B for a ∈ D, and
we can form the *-monomorphism hV : D → ER with [hV ] ∈ S(HR , D, ER ) and
hV (a) − HR (a) ∈ J.
Let W := V + C0 (R+ , B) in E (+) , then

π0+ (W ) = πB (V ) = (U + (0)U − (0)∗ ) ⊕ 1 .

The unitaries (U − ⊕ 1) ∈ E (−) and W ∗ (U + ⊕ 1) ∈ E (+) , satisfy

U − (0) ⊕ 1 = π0− (U − ⊕ 1) = π0+ (W )∗ (U + (0) ⊕ 1) = π0+ (W ∗ (U + ⊕ 1)).

Therefore they define a unitary u0 ∈ ER ⊆ E (−) ⊕ E (+) .

Let w ∈ O2 with w∗ ((a⊕b)⊕c)w = a⊕(b⊕c), and let u := u0 w. For the Cuntz
addition ⊕ and its transformation rules we refer to Chapter 4 for every details.
 3. Ext(SC; A, SB) AND KK(C; A, B) CONSIDERED AS K1 -GROUPS 893

Since H0 ⊕ H0 = H0 , we get from our definition of hV in Remark 8.3.1 that

g − (a) ⊕ H− (a) = w∗ (U − ⊕ 1)∗ (hV (a)|R− )(U − ⊕ 1)w

and
g + (a) ⊕ H+ (a) = w∗ (U + ⊕ 1)∗ V ∗ (hV (a)|R+ )V (U + ⊕ 1)w .

This means that g ⊕ H0 = u∗ hV u with u ∈ U(ER ) as above defined. Therefore,

(g(a) ⊕ H0 (a)) − u∗ H0 (a)u = u∗ (hV (a) − H0 (a))u is in J for a ∈ D.
We show the independence from the particular choice of T ∈ [HR ] and from
the generators s1 , s2 of O2 : Let T : D → Qs (SB) another representative of [HR ]
and t1 , t2 isometries in ER = Qs (SB) which generate a copy of O2 . Then there are
unitaries u1 , u2 ∈ ER such that T = u∗1 HR u1 and u2 tj = sj , j = 1, 2. Thus, with
⊕ = ⊕s1 ,s2 , g ⊕t1 ,t2 T = (1 ⊕ u1 )∗ u2 (g ⊕ HR )u∗2 (1 ⊕ u1 ). If u∗ hV u = g ⊕ HR , then
u∗3 (u∗1 hV u1 )u3 = g ⊕t1 ,t2 T for u3 := u∗1 uu∗2 (1 ⊕ u1 ), and T (a) − u∗1 hV (a)u1 ∈ J for
a ∈ D, because J is an ideal.
(ii): Let σs (t) := st+(1−s)σ(t) for s ∈ [0, 1] and t ∈ R. Since σ : R → R is order
preserving, the σs define a homotopy s → ρs := σbs | C0 (R, B) ∈ Aut(C0 (R, B)) be-
tween σb| C0 (R, B) = ρ0 and id = ρ1 , where σbs (f ) := f ◦σs for f ∈ Cb,st (R, M(B)) ∼=
M(C0 (R, B)). Then [σbs ◦h] = (ρt )∗ ([h]) in Ext(C(R); D, SB) ∼ = KK(C; D, (SB)(1) ),
by definition of (ρt )∗ ∈ Aut(Ext(C(R); A, B)).
Let τs := (ρs )(1) , i.e., τs (f, g) := (f ◦σs , g◦σs ). Then t → τs ∈ Aut(C0 (R, B)(1) )
is a (grading-preserving) homotopy between id and τ1 .
If Φ = Φ(C; D, SB) denotes the natural isomorphism from Ext(C(R); D, SB)
onto KK(C; D, (SB)(1) ), then (τs )∗ ◦ Φ = Φ ◦ (ρs )∗ .
Since (τ1 )∗ = (τ0 )∗ by homotopy invariance of KK(C; D, (SB)(1) ), we get [b σ◦
h] = (ρ0 )∗ ([h]) = (ρ1 )∗ ([h]) = [h] for [h] ∈ G(HR ; D, ER ). Since h = g ⊕ T satisfies
[h] ∈ G(HR ; D, ER ), we can see that [b σ ◦ (g ⊕ T )] = [g ⊕ T ] in S(HR ; D, ER ). 

In principle, Proposition 8.2.6, and Remark 5.8.7

Check ref. to 5.8.7
imply our below given Corollary 8.3.3 via Bott periodicity.
But we give a proof which uses only the homotopy invariance of KK(C; ·, ·) alone
to derive that Ext(C(R+ ); A, C0 (R+ , B)) = 0 and use the simple construction of
Remark 8.3.1, because our explicit constructions on the level of Busby invariants
and unitaries are needed in Chapter 9 to show that Theorem 4.4.6 yields the proofs
of Theorems B(i,ii) and M(i,ii).

Corollary 8.3.3. Suppose that D is separable, B is σ-unital and that both

are stable and trivially graded. Let C ⊆ CP(D, B) a non-degenerate point-norm
closed countably generated m.o.c. cone.
Let E := Q(B) := M(B)/B ∼ = Qs (B), and let H0 := πB ◦ H : D → E be a *-
monomorphism that comes from a faithful non-degenerate *-representation H : D →
894 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

M(B) with the property that δ∞ ◦ H is unitarily equivalent to H and that the maps
a ∈ D 7→ b∗ H(a)b ∈ B generate C ( 9 ).

(i) There is a natural isomorphism θ from the kernel of K1 (H0 (D)0 ∩ E) →

K1 (E) onto KK(C; A, B).
(ii) In particular, K∗ (HR+ (D)0 ∩ Qs (C0 (R+ , B))) = 0 where HR+ (a) means
the “restriction” of H0 (a) ∈ M(B) ⊆ ER = M(C0 (R, B))/ C0 (R, B) to
R+ .
(iii) The map

[g] ∈ G(H0 , D, Qs (SB)) → [U + (0)U − (0)∗ ] ∈ K1 (H0 (D)0 ∩ E),

where U − (0), U + (0) ∈ E are as in the proof of Lemma 8.3.2, defines a

group isomorphism ι from Ext(C; D, SB) = G(H0 ; D, Qs (SB)) onto the
kernel of K1 (H0 (D)0 ∩ E) → K1 (E).
(iv) Let h : D → B be a C*-morphism in C, Ch the extension of D by SB
given by the mapping cone construction for h, and let [h − 0] denote the
element of KK(C; A, B) that is represented by the difference construction
(B, h, 0).
Then the isomorphism θ ◦ ι from Ext(C; A, SB) onto KK(C; A, B)
maps the class [Ch ] ∈ Ext(C; A, SB) of Ch to [h − 0] ∈ KK(C; A, B), i.e.,
θ ◦ ι([Ch ]) = [h − 0].

Proof. (i): Let H0 (a) := H1 (a) + B, Q(B) := M(B)/B ∼ = Qs (B) and let
0 s 0
C := H0 (D) ∩ Q (B). Recall that H1 (D) ∩ M(B) contains a copy of O2 unitally,
and that K∗ (H1 (D)0 ∩ M(B)) = 0.
op
By E0 we denote the Hilbert B-module HB ⊕ HB , i.e., HB ⊕ HB with grading
βE0 : (x, y) 7→ (x, −y). Then naturally L(E0 ) ∼ = M2 (M(B)) with grading Z 7→
βZβ, β := diag(1, −1) and K(E0 ) ∼ = M2 (B) under this isomorphism. We define
φ0 : D → L(E0 ) by φ0 (a) := diag(H1 (a), H1 (a)) for a ∈ D.
For a unitary U ∈ M(B) let F (U ) ∈ L(E0 ) denote the selfadjoint unitary
which is given by the off-diagonal 2 × 2-matrix with entries F (U )11 = F (U )22 = 0,
F (U )12 = U and F (U )21 = U ∗ . Then

F (U )φ0 (a) − φ0 (a)F (U ) = F (1) diag U ∗ H1 (a) − H1 (a)U ∗ , U H1 (a) − H1 (a)U .

Thus, (E0 , φ0 , F (U )) is a C-compatible Kasparov module, if and only if, U ∈

U(H1 (D), B), with notation of Lemma 4.6.6. In particular, the product U0 U2 U1∗ is
in U(H1 (D), B) if U0 , U1 , U2 ∈ M(B) are unitaries and if

(E0 , diag(U0∗ H1 U0 , U1∗ H1 U1 ), F (U2 ))

is a Kasparov module. The latter is isomorphic to (E0 , φ0 , F (U0 U2 U1∗ )) via

diag(U0 , U1 ).

9 See Corollary 5.4.4. In the special C = CP

nuc (A, B), H = M(γ) ◦ ρ for a non-degenerate
*-morphism γ : K → M(B) and any faithful ρ : D ,→ M(K) with ρ(D) ∩ K = 0.
 3. Ext(SC; A, SB) AND KK(C; A, B) CONSIDERED AS K1 -GROUPS 895

We define a Hilbert B-module isomorphism T0 : E0 ⊕B E0 → E0 from the

Hilbert B-module sum E0 ⊕B E0 onto E0 , by

T0 : ((x1 , y1 ), (x2 , y2 )) 7→ (s1 x1 + s2 x2 , s1 y1 + s2 y2 ) .

Then T0∗ (φ0 ⊕ φ0 )T0 = φ0 and T0∗ (F (U1 ) ⊕ F (U2 ))T0 = F (U1 ⊕s1 ,s2 U2 ) for unitaries
U1 and U2 in M(B). By Lemma 4.6.6, U(H1 (D), B) is closed under Cuntz addition.
It follows that θ0 : U 7→ (E0 , φ0 , F (U )) is an additive map from U(H1 (D), B)
with Cuntz addition into the semigroup E(C; D, B)/ ≈ of isomorphism classes of
elements of E(C; D, B).
Since 1⊕1 = 1, we get that θ0 (1) represents the zero of KK(C; D, B). Therefore,
θ0 (U ) and θ0 (U ⊕ 1) represent the same element of KK(C; D, B).
A simple calculation shows that θ0 (U ) is a compact perturbation of θ0 (U 0 ) in
the sense of Definition 8.2.1 if and only if (U 0 − U )H1 (D) ⊆ B.
If t 7→ U (t) is a continuous map from [0, 1] into U(H1 (D), B), then t 7→ θ0 (U (t))
defines an operator homotopy in E(C; D, B), and, therefore, θ0 (U (0)) and θ0 (U (1))
represent the same element of KK(C; D, B) by the (operator) homotopy invariance
of KK(C; ·, ·).
By Lemma 4.6.6, it follows that θ0 (U1 ) and θ0 (U2 ) represent the same element
of KK(C; D, B) if [U1 + B] = [U2 + B] in K1 (H0 (D)0 ∩ Qs (B)).
This means that θ([U + B]) := [θ0 (U )] defines a group homomorphism θ from
the kernel of K1 (H0 (D)0 ∩ E) → K1 (E) into KK(C; D, B) ∼ = KK(C; A, B). It
remains to show:

(1) θ is an epimorphism, i.e., every element of KK(C; D, B) can be represented

by θ0 (U ) for a suitable U ∈ U(H1 (D), B).
(2) [θ0 (U )] = 0 in KK(C; D, B) implies [U + B] = 0 in K1 (C).

(1): For trivially graded D and σ-unital trivially graded B, one can use Kas-
parov’s stabilization theorem to obtain a representative (E, φ, F ) of an element of
KK(C; D, B), such that the Hilbert B-module E is isomorphic to the above con-
sidered “universal” graded Hilbert B-module E0 , cf. . [73, 17.4.1].
An element of KK(C; D, B) ∼ = KK(C; A, B) with representing element (E, φ, F )
can be also represented by C-compatible Kasparov modules (E ⊕ E op , φ ⊕ 0, G(F ))
with G(F ) = G(F )∗ = G(F )−1 . Here E op is the Hilbert B-module E with the new
grading operator −βE , and G(F ) ∈ L(E ⊕ E op ) ∼ = M2 (L(E)) is given by the matrix
with entries G(F )11 = F1 , G(F )22 = −F1 and G(F )12 = G(F )21 = (1 − F12 )1/2 ,
where F1 := f ((F ∗ + F )/2) is a contractive selfadjoint φ-“compact” perturbation
of F , and f ∈ Cb (R) is given by f (t) = t for t ∈ [0, 1], f (t) = −1 for t ≤ −1 and
f (t) = 1 for t ≥ 1.
(E ⊕B E op , φ ⊕ 0, G(F )) represents the same element of KK(C; D, B) because
(E ⊕B E op , φ ⊕ 0, G(F )) is a compact perturbation of the sum of (E, φ, F ) and
(E op , 0, −F ) (see [73, 17.4.2, 17.4.3, 17.6] for more details).
896 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

The passage to this special representatives is functorial in E(C; D, B) in the

sense that it is operator homotopy preserving and is additive modulo addition of
trivial elements of form (E 0 , 0, F 0 ) and compact perturbation. If (E, φ, F 0 ) is a
φ-“compact” perturbation of of (E, φ, F ), then
Is this correct needed?:
(E ⊕B E op , φ ⊕ 0, G(F 0 )) ∼cp (E ⊕B E op , φ ⊕ 0, G(F )).
Since we did start with a C-compatible Kasparov module (E, φ, F ) such that E
is isomorphic to our above defined E0 by a grading preserving isometric B-module
isomorphism, we have that E ⊕B E op is isometrically and grading preserving B-
module isomorphic to E0 .
Thus we find for each element of KK(C; D, B) a representing C-compatible
Kasparov module (E0 , φ, G) with G = G∗ = G−1 .
Since the degree of φ(a) is zero and the degree of G is one, the representa-
tive has the form φ(a) = diag(ψ0 (a), ψ1 (a)) with weakly C-compatible *-morphisms
ψ0 , ψ1 : D → M(B) and G = F (U ) for a unitary U ∈ M(B).
T0 defines an isomorphism from the sum of (E0 , diag(ψ0 , ψ1 ), F (U )) and of
θ0 (1) onto
(E0 , diag(ψ0 ⊕ H1 , ψ1 ⊕ H1 ), F (U ⊕ 1)). (∗ )

Since θ0 (1) represents the zero, every element of KK(C; D, B) is represented

by a Kasparov module of form (∗ ).
The homotopy invariance of KK(C; D, B) yields the following:
(τ ) (τ )
Let τ 7→ ψ0 : D → M(B) and τ 7→ ψ1 : D → M(B) be a point-norm continuous
families of weakly C-compatible C *-morphisms, and let U ∈ M(B) be a unitary
(0) (0) (τ ) (0)
such that U ψ1 (a)−ψ0 (a)U ∈ B for all a ∈ D. Suppose that ψ0 (a)−ψ0 (a) and
(τ ) (0) (0) (0)
ψ1 (a)−ψ1 (a) are in B for all a ∈ D and τ ∈ [0, 1]. Then (E0 , diag(ψ0 , ψ1 ), U )
(1) (1)
and (E0 , diag(ψ0 , ψ1 ), U ) represent the same element of KK(C; D, B).
By Theorem 5.6.2(ii) (applied to C := H1 (D) and T := ψ ◦ H1−1 ), we find
norm-continuous maps t 7→ U0 (t) and t 7→ U1 (t) into the unitaries of M(B) such
that, for k ∈ {0, 1}, t ∈ R+ and a ∈ A,

Uk (t)∗ H1 (a)Uk (t) − (ψk (a) ⊕ H1 (a)) ∈ B

and
lim k Uk (t)∗ H1 (a)Uk (t) − ψk (a) ⊕ H1 (a) k = 0 .


t→∞
Thus
E0 , diag(U0 (0)∗ H1 U0 (0), U1 (0)∗ H1 U1 (0)), F (U ⊕ 1) (∗∗ )

and (∗ ) represent the same element of KK(C; D, B) = KK(C; A, B).

(∗∗ ) is isomorphic to θ0 (U1 ), where U1 := U0 (0)(U ⊕ 1)U1 (0)∗ , and U1 is a
unitary in U(H1 (D), B), because (∗∗ ) is a Kasparov module. Thus, every element
of KK(C; D, B) ∼ = KK(C; A, B) can be represented by a C-compatible Kasparov
module θ0 (U ), where U ∈ U(H1 (D), B), and θ is an epimorphism.
 3. Ext(SC; A, SB) AND KK(C; A, B) CONSIDERED AS K1 -GROUPS 897

(2): Suppose that [θ0 (U )] = 0 in KK(C; D, B).

By Definition 8.2.1, [θ0 (U )] = 0 means that there exist a C-compatible Kasparov
module (E, φ, F ) and a compact perturbation (E, φ, F 0 ) of it, such that θ0 (U ) ⊕
(E, φ, F ) is isomorphic to (E, φ, F 0 ) via a grading preserving B-module unitary I.
In particular F 0 = I −1 (F (U ) ⊕ F )I. If we add trivial Kasparov modules, take
the above listed φ ⊕ 0-“compact” perturbations and pass to isomorphic Kasparov
modules (isomorphic in the strong sense of Definition 8.2.1), then we can assume
moreover that E = E0 , F = F (V1 ), F 0 = F (V2 ) for unitaries V1 and V2 in M(B),
as we have shown above. Then φ = diag(ψ0 , ψ1 ) for weakly C-compatible C *-
morphisms ψ0 and ψ1 from D into M(B).
By stability of B, there exists a sequence of isometries t1 , t2 , . . . in M(B) such
that tn t∗n converges strictly to 1, cf. Remark 5.1.1(8). Let λ be the infinite repeat
P

of the Cuntz sum (ψ0 ⊕ ψ1 ) ⊕ H1 i.e., λ := δ∞ ((ψ0 ⊕ ψ1 ) ⊕ H1 ), where ⊕ = ⊕s1 ,s2 .

Then λ is a strictly continuous C *-morphism from D into M(B).
The series ((tn ⊕ tn ) ⊕ tn )t∗n and converges strictly to a unitary in M(B).
P

Therefore λ is unitarily equivalent to the Cuntz sum of the infinite repeats of ψ0 ,

of ψ1 and of H1 . It follows that ψ0 ⊕ λ, ψ1 ⊕ λ and H1 ⊕ λ are unitarily equivalent
to λ.
Now Corollary 5.6.1 applies to the C *-subalgebra C := H1 (D) of M(B) and
to T = λ ◦ H1−1 . Thus there is a unitary v in M(B) such that v ∗ λ(a)v − H1 (a) ∈ B
for a ∈ D.
Let χ := v ∗ λv. Then again χ ⊕ ψ0 and χ ⊕ ψ1 are unitarily equivalent to χ.
By Lemma 4.6.5(ii), the relative commutant of χ(D) in M(B) has trivial K-theory,
in particular, [V + B] = 0 in K1 (C) if V is a unitary in M(B) which commutes
element-wise with χ(D). By Lemma 5.1.2, χ is unitarily equivalent to χ⊕χ, because
χ is an infinite repeat.
We add the Kasparov module (E0 , diag(χ, χ), F (1)) to (E0 , φ, F (Vj )), j = 1, 2.
Then we use our above defined isomorphism T0 from E0 ⊕ E0 onto E0 , and pass
to equivalent Kasparov modules, where we use the above described transformation
rules for U 7→ F (U ).
Then we get unitaries U1 , U2 , U3 and U4 in M(B) such that
(E0 , diag(χ, χ), F (Uj )) are Kasparov modules for j = 1, 2,
χ(D)(U1 − U2 ) ⊆ B,
U3∗ (H1 ⊕ χ)U3 = χ, U4∗ (H1 ⊕ χ)U4 = χ and U3∗ (U ⊕ U1 )U4 = U2 .
In particular V := U3∗ U4 commutes element-wise with the image of χ, and,
therefore, [V + B] = 0 in K1 (C).
Since H1 ⊕ H1 = H1 and χ(d) − H1 (d) ∈ B for d ∈ D, we get that the Uk are
in U(H1 (D), B) for k = 1, 2, 3, 4.
Let u0 , u1 , u2 , u3 , u4 denote the images of U, U1 , U2 , U3 , U4 in C = H0 (D)0 ∩ E.
898 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

Then [u1 ] = [u2 ] in K1 (C) by Lemma 4.6.6.

[u0 ] + [u1 ] = [u∗3 (u0 ⊕ u1 )u3 ] + [u∗3 u4 ] = [u2 ] .

Thus [U + B] = [u0 ] = 0 in K1 (H0 (D)0 ∩ E).

(ii): Let E (+) := Qs (C0 (R+ , B)). In the proof of Lemma 8.3.2 we have seen
that K∗ (E (+) ) = 0.
Is this CONCLUSION ‘‘thus’’ correct? ??
Thus,
by Propositions 4.4.3(i) and 8.2.6,

K0 (HR+ (D)0 ∩ E (+) ) = KK(C; D, C0 (R+ , B)(1) ) = 0 .

By Part (i) and Proposition 8.2.5(ii),

K1 (HR+ (D)0 ∩ E (+) ) = KK(C; D, C0 (R+ , B)) = 0 .

(iii): Let E := Qs (B), C := H0 (D)0 ∩ E, ER := Q(SB) E (α) := Qs (C0 (Rα , B))

for α ∈ {+, −}.
Since G(HR , D, ER ) = [HR ] + S(HR , D, ER ) , from the proof of Lemma 8.3.2 we
get the existence of unitaries U + ∈ E (+) and U − ∈ E (−) such that

g α = (U α )∗ Hα U α ,

where g α (a) := g(a)|Rα , Hα := HRα for α ∈ {+, −}. Here we use the “restriction”
notation as explained in Remark 8.3.1.
We have seen in the proof of Lemma 8.3.2 that U + (0)(U − (0))∗ is a unitary in
C such that [U + (0)(U − (0))∗ ] is in the kernel of the natural group homomorphism
from K1 (C) into K1 (E). Here U + (0) := π0+ (U + ) and U − (0) := π0− (U − ) for the
natural epimorphisms π0α : E (α) → E, where α ∈ {+, −}.
If U1+ ∈ E (+) and U1− ∈ E (−) are other unitaries with g α = (U1α )∗ Hα U1α for
α ∈ {+, −}. Then U1+ (U + )∗ is in ????? H+ (D)0 ∩E (+) . By part (ii), the K1 -images
of the natural epimorphism from H+ (D)0 ∩ E (+) into C = H0 (D)0 ∩ E is zero. This
implies that [U1+ (0)(U + (0))∗ ] = 0 in K1 (C). Similarly, [U − (0)(U1− (0))∗ ] = 0 in
K1 (C). Thus [U + (0)(U − (0))∗ ] = [U1+ (0)(U1− (0))∗ ] in K1 (C).
Therefore ι([g]) := [U + (0)(U − (0))∗ ] is a well-defined map from G(HR , D, ER ) =
Ext(C; D, SB) into the kernel of K1 (C) → K1 (E).
If g and h are representatives of elements in G(HR , D, ER ) and if U1+ , U2+ ∈ E (+)
and U1− , U2− ∈ E (−) are unitaries with hα = (U2α )∗ Hα U2α and g α = (U1α )∗ Hα U1α for
α ∈ {+, −}. Then U α := U1α ⊕U2α ∈ E (α) are unitaries with (g⊕h)α = (U1α )∗ Hα U1α .
Thus ι([g] + [h]) = [(U1+ (0)(U1− (0))∗ ) ⊕ (U2+ (0)(U2− (0))∗ )], which is ι([g]) + ι([h]) in
K1 (C). Therefore, ι is a group homomorphism.
By Lemma 4.6.6, ι is an epimorphism, because for U ∈ U(H1 (D), B) and hU
in Remark 8.3.1 we have, for g = hU ⊕ HR , U + = U ⊕ 1 and U − = 1, and therefore
ι[hU ⊕ HR ] = [U + B].
 3. Ext(SC; A, SB) AND KK(C; A, B) CONSIDERED AS K1 -GROUPS 899

Suppose that ι([g]) = [U + (0)U − (0)∗ ] = 0 in K1 (H0 (D)0 ∩ E). By Lemma

4.2.6(v,2), the unitary V := (U + (0)U − (0)∗ ) ⊕ 1 is in the connected component of
1 in the unitaries of C. Thus there is a continuous map t ∈ [0, 1] → U (t) into the
unitaries of M(B) such that U (0) = 1, U (1)+B = V and U (t)+B ∈ C for t ∈ [0, 1].
Thus U (t) ∈ U(H1 (D), B) for t ∈ [0, 1]. Let U := U (1) and W ∈ Cb (R, M(B)) ⊆
M(SB) the unitary which is defined by W (t) := 1 for t ≤ 0, W (t) := U (t) for
t ∈ [0, 1], and W (t) = U for t ≥ 1. Then, for the lift T of hU in Remark 8.3.1, we get
that T (a)(t) − W (t)∗ H1 (a)W (t) ∈ B for t ∈ R and T (a)(t) − W (t)∗ H1 (a)W (t) = 0
for t not in [0, 1]. Thus hU = w∗ (HR )w for w := W + C0 (R, B). In the proof of
Lemma 8.3.2 we have seen that [g ⊕ HR ] = [hU ]. Since [g] ∈ G(HR , D, ER ), we get
[g] = [g ⊕ HR ] = [hU ] = [w∗ HR w] = [HR ] = 0.
Thus ι is an isomorphism from Ext(C; A ⊗ K, SB) ∼
= G(HR , D, Qs (SB)) onto
0
the kernel of natural homomorphism from K1 (H0 (D) ∩ Qs (B)) into K1 (Qs (B)).

(iv): To simplify notation, we assume that B is stable. We keep the notations

E (−) , E (+) of Remark 8.3.1, but don’t use the notation ER used there for Qs (SB) =
M(SB)/SB. Let C := H0 (D)0 ∩ E.
If h : D → B is a *-morphism in C, then the mapping cone Ch is represented
as an element of S(HR ; D, Qs (SB)) by

gh : a ∈ D 7→ (0, h(a)) ∈ 0 ⊕ B ⊂ Q(R− , B) ⊕ Q(R+ , B) = Q(R, B) ⊆ E (−) ⊕ E (+) .

The corresponding element of G(HR , D, Qs (SB)) = Ext(C; A ⊗ K, SB) is [gh ⊕

HR ]. If we apply Theorem 5.6.2(ii) to C := H1 (D) and separately to T = 0 and
T = hH1−1 , we get norm-continuous maps t 7→ U0 (t) and t 7→ U1 (t) from R+ into
the unitaries of M(B), such that limt→∞ k(h ⊕ H1 )(a) − U0 (t)∗ H1 (a)U0 (t)k = 0,
(h ⊕ H1 )(a) − U0 (t)∗ H1 (a)U0 (t) ∈ B, limt→∞ k(0 ⊕ H1 )(a) − U1 (t)∗ H1 (a)U1 (t)k = 0
and (0 ⊕ H1 )(a) − U1 (t)∗ H1 (a)U1 (t) ∈ B for a ∈ D, t ∈ R+ .
Let us consider U0 as a unitary in Cb (R+ , M(B)) ⊆ M(C0 (R+ , B)) and V (t) :=
U1 (−t) for t ≤ 0 as a unitary V in M(C0 (R− , B)).
Let U + := π + (U0 ), U − := π − (V ). Then (gh ⊕ HR )|Rα = (U α )∗ Hα U α for
α ∈ {+, −}, U + (0) = π0+ (U + ) = U0 (0) + B and U − (0) = π0− (U − ) = U1 (0) + B.
Thus ι([gh ⊕ H0 ]) = [U0 (0)U1 (0)∗ + B]C ∈ K1 (C).
By construction of θ, we have that θ([U0 (0)U1 (0)∗ + B]C ) is represented by
(E0 , φ0 , F (U0 (0)U1 (0)∗ )).
Let (HB , h, 0) the C-compatible Kasparov module obtained by the difference
construction. Since the derivative F of the module is zero, the construction in the
proof of part (i) produces for the class of (HB , h, 0) in KK(C; D, B) the represen-
tative (E0 , diag(h, 0), F (1)), because F (1) = G(0), where the notation is as in the
proof of part (i).
In the proof of part (i) we have seen, that the Kasparov module (E0 , φ0 , F (1))
represents zero. If we add it to (E0 , diag(h, 0), F (1)), we get (E0 , diag(h ⊕ H1 , 0 ⊕
H1 ), F (1)).
900 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

By homotopy invariance of KK(C; ·, ·), (E0 , diag(U0∗ H1 U0 , U1∗ H1 U1 ), F (1)) re-

presents the same element, where U0 := U0 (0) and U1 := U1 (0) are the unita-
ries in M(B) which come from the above considered paths U0 (t) and U1 (t) in
U(H1 (D), B).
In the proof of part (i) we have seen that the latter Kasparov module is iso-
morphic to (E0 , φ0 , F (U0 U1∗ )), which represents ι([U0 U1∗ ]). Hence, ι ◦ θ([Ch ]) =
[h − 0]. 

Change following to C-picture !?

Corollary 8.3.4. Suppose that A is separable, that B is σ-unital and sta-

ble, and that both are trivially graded. Let D := A ⊗ K, ER := Qs (SB), and let
HR : D → ER be a *-monomorphism which comes from a faithful non-degenerate
*-representation of H : D ,→ M(B) ⊆ ER in “general position”, i.e., with δ∞ ◦ H
is unitarily homotopic to H.

(i) For every t ∈ [0, 1], the natural group homomorphism

πt : G(HR×[0,1] , D, Qs (S C([0, 1], B))) → G(HR , D, ER ),

which is induced by the natural epimorphism from Qs (S C([0, 1], B)) onto
ER is an isomorphism and is therefore independent of t ∈ [0, 1].
(ii) If σ is an orientation preserving homeomorphism of R and σb the automor-
phism of ER := Qs (SB) which is naturally induced by σ, then [b
σ ◦ h] = [h]
in Extnuc (A ⊗ K, SB) for every representing C*-morphism h : D → ER of
an element of Extnuc (A ⊗ K, SB).

Proof. (i): It follows from the natural isomorphisms G(H0 , D, ER ) =

Extnuc (D, SB) ∼ = KKnuc (D, B) (including G HR×[0,1] , D, Qs (S C([0, 1], B)) ∼


=
KKnuc (D, C([0, 1], B))) and the homotopy invariance of KKnuc .
(ii): Since σ is a strictly increasing continuous function from R into R, for
every f ∈ SB ∼ = C0 (R, B), the definition g(t, s) := f (st + (1 − s)σ(t)) defines an
element g in
S C([0, 1], B) ∼
= C0 (R × [0, 1], B) .

Let V : D → M(SB) be a completely positive

weakly nuclear, respectively C-compatible,
contractive lift of h : D → Qs (SB).
The curvature d 7→ V (d∗ d) − V (d)∗ V (d) of V maps D into SB = C0 (R, B).
Under the natural identifications of Cb,st (R, M(B)) with M(SB), respectively of
Cb,st (R × [0, 1], M(B)) with M(S C([0, 1], B)), we let

Ve (t, s)(d) := V (st + (1 − s)σ(t))(d)

for t ∈ R, s ∈ [0, 1], d ∈ D. Then Ve is

check again?:
 3. Ext(SC; A, SB) AND KK(C; A, B) CONSIDERED AS K1 -GROUPS 901

weakly nuclear, respectively C-compatible,

and the curvature d 7→ Ve (d∗ d)−Ve (d)∗ Ve (d) of Ve maps D into C0 (R×[0, 1], B) =
S C([0, 1], B). Thus it defines a
weakly nuclear, respectively C-compatible,
liftable C *-morphism k : D → Qs (S C([0, 1], B)).
The natural epimorphisms from Qs (S C([0, 1], B)) onto Qs (SB) at zero and
at one map k into σ bh and h, respectively. Thus k defines an element [k] of
Extnuc (D, S C0 ([0, 1], B)). with π1 ([k]) = [h] and π0 ([k]) = [bσ h]. Therefore, by
σ ◦ h] in Extnuc (D, SB), respectively is in Ext(C; D, SB).
(i), [h] = [b 

Remark 8.3.5. The natural map γ from KKnuc (A, B) to Hom(K∗ (A), K∗ (B))
is defined as (γ(x))(y) := y ⊗A x ∈ K∗ (B) for y ∈ K∗ (A), x ∈ KKnuc (A, B), where
we denote by y ⊗A x the Kasparov product of x and y, and K0 (.) and K1 (.) are
naturally identified with KK(C, .) and KK(C0 (R), .) respectively.
In terms of Extnuc (A, SB) this means equivalently, that γ maps the class with
representative g : D → Qs (SB) into the connecting maps

δ∗ : K∗ (A) → K(∗+1)mod2 (SB)

of the six term exact sequence for the extension of A ⊗ K by SB which is defined
by g. Then use K∗ (B) ∼= K(∗+1)mod2 (SB).
Both descriptions of γ allow to see that γ([h − 0]) = K∗ (h) if h : A ⊗ K → B is
a nuclear C *-morphism.

Remark 8.3.6. By Bott periodicity of KKnuc and by the isomorphism (1.1),

the similar notion of a strictly nuclear E-theory is functorial equivalent to KKnuc
(See further below). But E-theory is not useful for the proof of (1.1) and its
consequences. Note that Enuc (A, B) = E(A, B) if A is nuclear, but in general this
equation does hold for nuclear B and non-nuclear A.
Even in the trivially graded case there is no direct proof of the isomorphism

Extnuc (A, SB) ∼

= Enuc (A, B)

by using directly the weakly nuclear version SR(CPnuc ; SA, S 2 B) of [[SA⊗K, S 2 B]]
as defined in the beginning of Chapter 7. Therefore we outline here another proof :
Suppose that B is a stable σ-unital C *-algebra and denote by πs (s ∈ [0, 1])
the natural epimorphism

πs : Q(R+ , S C([0, 1], B)) → Q(R+ , SB) .

But notice that a weakly nuclear asymptotic morphism h = {V (t) : A ⊗ K → B}

in the sense of of Definition 7.1.1(a,b) represents in general not an element of the
Rørdam semigroup (with the additional requirement (c) of Definition 7.1.1 and with
C := CP (A ⊗ K, B ⊗ K)). This can be seen easily in the case A = C0 (R, O∞ ) and
B = C0 (R) ⊗ K. Note, that the nuclearity of V (t) is required for each t ∈ R+ , but
the uniform nuclearity on R+ is not required.
902 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

We say that weakly nuclear asymptotic morphisms h0 = {V0 (t)} and h1 =

{V1 (t)} from SA ⊗ K into SB are asymptotical homotopic if there is a weakly
nuclear asymptotic morphism h = {V (t)} from SA ⊗ K into C([0, 1], SB) such that
h0 = π0 ◦ h and h1 = π1 ◦ h.
As in [165] or [163, chp.II.B], one can see that the classes of asymptotically
homotopic weakly nuclear asymptotic homomorphisms from SA ⊗ K into SB ⊗ K
form a group Enuc (A, B) under Cuntz addition.
The generalized mapping cone construction, cf. Chapters 1 and 9, yields a group
homomorphism from Enuc (A, B) into Extnuc (SA ⊗ K, S 2 B).
Conversely the natural asymptotic morphism from SA ⊗ K into SB which
corresponds to an extension (extension-class) 0 → SB → E → A ⊗ K → 0 in
Extnuc (A ⊗ K, SB) is a weakly nuclear asymptotic morphism. It defines a group
homomorphism from Extnuc (A ⊗ K, SB) to Enuc (A, B).
The composition maps are equivalent (i.e. are element-wise homotopic) to the
natural map Enuc (A, B) → Enuc (SA, SB), respectively to

Extnuc (A ⊗ K, SB) → Extnuc (SA ⊗ K, S 2 B) .

Thus Proposition 8.2.5(iii) and Corollary 8.3.3(iii) imply that there is a natural
isomorphism KKnuc (A, B) ∼= Enuc (A, B) for separable A and σ-unital B.

4. The residually nuclear case

Remark 8.4.1. Suppose that A, B are stable, where A is separable and B is

σ-unital. They are assumed to be trivially graded.
Furthermore, we suppose that X is the primitive ideal space of a separable C *-
algebra C. We can then replace C by F ⊗ C with same primitive ideal space, where
we let F := K ⊗ O2 ⊗ O2 ⊗ . . .. Then, in addition, C is stable and C ∼
= C ⊗ O2 .
Suppose now that there are given monotonous actions ΨA and ΨB of X on A
and B, respectively, cf. Definition 1.2.6.
The group KKnuc (X; A, B) is defined in Definition 1.2.10 as KK(C; A, B) for
the cone C of ΨA -ΨB –residually nuclear maps.
Extnuc (X; A, B) ∼
= G(H0 ; A, B) and S(H0 , A, B) are defined in Chapter 5.
Note that here H0 : A → M(B) denotes the “universal” non-degenerate weakly
Ψ-residually nuclear C *-morphism from A to M(B) such that δ∞ H0 is unitarily
equivalent to H0 .
or is it only approximately unitary equivalent
with respect to the strict topology on M(B) ...?
See def. in chp. 5 !!
The existence and uniqueness of H0 is assured if the m.o.c. cone of Ψ-residually
maps from A → M(B) has a countable generating set (as it is e.g. in the case where
A and B are separable).
 4. THE RESIDUALLY NUCLEAR CASE 903

We list now the non-trivial changes of the above considered case X =point
(with trivial action ΨA (X) = A, ΨA (∅) = {0}).
In the beginning of the proof of Proposition 8.2.5(iii):
By Section 9, for D separable and stable and B σ-unital and stable,
Extnuc (X; D, B) = G(H0 ; D, E) where E := Q(B), H0 = πB ◦ H1 and
H1 : D → M(B) is an infinite repeat of a sufficiently general non-degenerate
weakly ΨA -ΨB –residually nuclear C *-morphism from D in M(B) (The existence
of such H1 for exact A follows later from results in Chapter 12).
The unitary homotopy between H0 ⊕ ψ and H0 comes in the residually nuclear
case from Theorem 5.9.3 in the same way as Theorem 5.6.2(ii) has been used in the
proof.

To-do-list: corollaries for chapter 9:

next:lem:8.Ynew.chp9

Lemma 8.4.2. There are natural isomorphisms

Ext C ⊗ CP(C, C0 (R)) ; D, B ⊗ C0 (R) ∼

= KK(C; D, B) .

The isomorphisms are compatible with respect to scaling of R.

The classes in KK(C ; A, B) are invariant under homotopy.
There are natural isomorphisms

Extnuc (X; D, SB) ∼

= KKnuc (X; D, B) .

Chp:9: By generalized Cor.8.3.3(iii) or Lem. 8.4.2:

Suppose that ???????
For a C *-morphism h from D to B with h ∈ C the mapping cone construction
defines an element α[h] = [Ch ] of Ext(C ⊗ CP(C, C0 (R)); D, SB) which is mapped
to the element [h − 0] ∈ KKnuc (X; D, B) under the natural isomorphism from
Ext(C ⊗ CP(C, C0 (R)); D, SB) onto KK(C; D, B).
next lem:8.4new used in chp:9

Lemma 8.4.3. , For every homomorphism k : D → ER that is dominated by

H0 , there exists a unitary u ∈ ER such that u∗ (k ⊕ H0 )(a)u − H0 (a) ∈ J for each
a ∈ D.

next mentioned in chp:1

There is a natural group homomorphism for tensor products:

KK(X; A, B) × KK(C, D) → KK(X; A ⊗ C, B ⊗ D) ,

where the action is given, e.g. on A ⊗ C, by

Z ∈ O(X) 7→ ΨA (Z) ⊗ C ∈ I(A ⊗ C) .

904 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

Is it the same as
KK(C; A, B) × KK(C, D) → KK(C ⊗ CP(C, D); A ⊗ C, B ⊗ D) ?

my view on split extensions:

Remark 8.4.4. Suppose that

0 −→ B −→ C −→ A −→ 0
is a short exact sequence of graded C *-algebras, and that C1 ⊆ CP(A, C),
?????
??

We define a map from the locally compact subspaces Y of R2 into the groups
by
F (Y ) := G(HY ; D, EY ) ∼
= Ext(C(Y ); A, B ⊗C0 (Y )) ∼
= KK(C(Y ), A, SB ⊗C0 (Y )) .
Then F (R+ ) = 0 because KK(C; ·, ·) is homotopy invariant and the identity map
of B ⊗ C0 (R+ ) is homotopic to 0.

Proposition 8.4.5. With the above notation holds that

K∗ (H[0,∞) (D)0 ∩ E[0,∞) ) = 0 ,
that there is a natural isomorphism from F (R) onto the kernel of K1 (H0 (D)0 ∩
Q(B)) → K1 (Q(B)) ∼ = K0 (B),
α ◦ h] = [h] ∈ Ext(C(R); D, B) for every h : D → ER with [h] ∈ F (R) and
and that [b
every order-preserving homeomorphism α from R to R with α(0) = 0.

The proof is contained in the further below given more explicit constructions,
The reduction to Mayer--Vietoris does not work??
but we give an independent proof, that reduce the proof to an application of
the K∗ -theory Mayer–Vietoris sequence to the Y -equivariant isomorphism of F (Y )
with the kernel of K0 (HY (D)0 ∩ EY ) → K0 (EY ) ∼
= K1 (B ⊗ C0 (Y )), (that we have
seen in Chapters 5 and 4).

Lemma 8.4.6. Let L(L2 (S 1 )) ∼= M(K) ⊆ H(D)0 ∩ M(B) given by a non-

0
degenerate map from K into H(D) ∩ M(B), and let
C0 (R) + C · 1 ∼
= C(S 1 ) ⊆ L(L2 (S 1 ))
given by multiplication of functions. Then the C*-morphism
H1 : D ⊗ C0 (R) → H(D)0 ∩ M(B)
with H1 (a ⊗ f ) := D(a)f satisfies the assumptions of Corollary 5.4.4 ??. for the
cone C ⊗ CP(C0 (R), C). Thus
Ext(C(R, Y ); D ⊗ C0 (R), B ⊗ C0 (Y )) ∼
= G(πB ◦ H1 ; D ⊗ C0 (R), EY )
 4. THE RESIDUALLY NUCLEAR CASE 905

and is isomorphic to the kernel of

K0 (T ) → K0 (EY ) ∼
= K1 (B ⊗ C0 (Y ))
where T := πB (H1 (D ⊗ C0 (R)))0 ∩ EY .
There is an epimorphism from Ext(C(R, R+ ) ; D ⊗ C0 (R), B ⊗ C0 (R+ )) onto
the kernel of the group morphism
K1 (H[0,∞) (D)0 ∩ E[0,∞) ) → K1 (E[0,∞) ) ∼
= K0 (B ⊗ C0 ([0, ∞))) .

Proof. ?? 

Alternatively: use explicit KK-equivalences in

E(C, C0 (R)(1) ) and E(C0 (R)(1) , C)

Proof of Proposition 8.4.5. : Since there are exact sequence

0 → F (R+ ) → K0 (H[0,∞) (D)0 ∩ E[0,∞) ) → K1 (B ⊗ C0 (R+ )),
0 → F (R− ) → K0 (H[0,−∞) (D)0 ∩ E[0,−∞) ) → K1 (B ⊗ C0 (R− )),
the monomorphism F (Y ) := G(HY ; D, EY ) → K0 (HY (D)0 ∩ EY ) and the mor-
phisms K∗ (HY (D)0 ∩ EY ) → K∗ (EY ) and the isomorphisms K0 (EY ) ∼= K1 (B ⊗
C0 (Y )) behave in a natural way equivariant with respect to homeomorphisms σ
b ◦ HY = HY . If X is a closed subset of Y such that for every
of Y , because σ
f ∈ Cb (Y )+ with f |X ∈ C0 (X) there exists g ∈ Cb (Y )+ with gf ∈ C0 (Y ) and
gf |X = f |X, then the natural map HY (D)0 ∩ EY → HX (D)0 ∩ EX is an epimor-
phism (that is induced by the natural epimorphism EY → EX ).
The algebra HR (D)0 ∩ ER is the pullback of H(−∞,0]) (D)0 ∩ E(−∞,0]) and
H[0,∞) (D)0 ∩ E[0,∞) by its natural morphism onto H[0,∞) ((D)0 ∩ E{0} . Since
K∗ (C0 (R+ , B)) = 0 and K∗ (H[0,∞) (D)0 ∩ E[0,∞) ) = 0 (by the first part), we get
that K1 (πB (H0 (D))0 ∩ Qs (B)) ∼
= K0 (HR (D)0 ∩ ER ).
The diagram (and thus the isomorphism) is invariant under σ b if σ is an order
preserving homeomorphism of R and σ(0) = 0. Then the morphisms of the Mayer–
Vietoris sequence are also σ σ |{0}) = K1 (id) on K1 (H0 (D)0 ∩
b-equivariant. Since K1 (b
b = id on K0 (HR (D)0 ∩ ER )
E0 ) it follows σ
More??? ??


One can check that the kernel of K1 (H0 (D)0 ∩ Q(B)) → K1 (Q(B)) ∼= K0 (B) is
nothing else KKnuc (A, B) if H0 corresponds to CPnuc (A, B) (and similarly in the
residually nuclear case KKnuc (X; A, B) or in the case of KK(C; A, B) with general
non-degenerate m.o.c. cone C ⊆ CP(A, B)). Notice here that A and B are trivially
graded.
Protocol (until 16.9.2009):
Some TO DO list for chp.8, applied in Chapter 11:
(0.1) In all points (*.*) and (*) below:
906 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

We should work with a suitable G-equivariant bi-functor (A, B) 7→ CA,B ⊆

CP(A, B) on suitable categories of G-algebras with G-equivariant morphisms, where
G is a Polish group.
(0.2) The only additional requirement for (G, X)-equivariant Kasparov modules
is that A and B are graded G algebras. That φ, and the gradings are G-equivariant,
and that the bi-module is G-modular, and that (g(F ) − F )φ(A) ∈ K(E) for all
g ∈ G, and g → g(F ) is norm-continuous.
(1) Each element of KKG (X; A, B) can be represented by a non-degenerate
(G, X)− Kasparov A–B-module (E, φ, F ) (i.e., φ(A)E = E).
(It gives the σ-additivity of KKG (X ; · , B) ???
But it is probably only for a special class of groups G possible, that contains
??????? compact ????? groups (?).
For them, one can reduce all to the stable case and then use the Brown stability
or Kasparov trivialization in a G-equivariant manner ????.)
(2) KKG (X; · , B) is half exact on (G, X)-semi-split sequences (and additive on
(G, X)-split-exact sequences). Needed (!!!), because of the KKG –proof in chapter
11.
(3) Desire: Give an explicit module for the KK-equivalence of split sequences.
(4) Morphisms:
If ψ : A → B satisfies ψ ◦ C1 ⊆ C2 , (that is symbolic ψ ∈ C2 C1−1 ⊆ CP(A, B)), then
ψ∗ : KK(C1 ; D, A) → KK(C2 ; D, B) is well-defined in the usual way.
If there are G-actions δ, α and β on D,A,B with α(g) ◦ C1 ◦ δ(g −1 ) ⊆ C1
and β(g) ◦ C2 ◦ δ(g −1 ) ⊆ C2 for all g ∈ G, and if ψ ◦ α(g) = β(g) ◦ ψ, then
ψ∗ : KKG (C1 ; D, A) → KKG (C2 ; D, B) is well-defined.
Suppose that G acts on a T0 -space X and that D, A, B are (G, X)-
algebras. If ψ : A → B is compatible with the (G, X)-actions on A and
B, then ψ∗ : KKG (X; D, A) → KKG (X; D, B) and ψ∗ : KKnuc G (X; D, A) →
KKnuc G (X; D, B) are well-defined. (This can be deduced from the case of cone-
depending KK, by taking C1 := C(X; D, A) and C2 := C(X; D, B), respectively
C1 := Crn (X; D, A) and C2 := Crn (X; D, B).)
Similar (obvious “opposite”) sufficient conditions can be given for the existence
of ψ ∗ : KKG (C4 ; B, E) → KKG (C3 ; A, E) : ψ should be G-equivariant and C4 ◦ψ ⊆
C3
We get that ψ ∗ : KKG (X; B, E) → KKG (X; A, E) is well-defined if ψ is
(G, X)-equivariant.
(5) The cones C needed for Mayer-Vietoris sequence:
(5a) Pull-back of epic morphisms ϕk : Ak → C :
In the case of the pull-back P := A1 ⊕ϕ1 ,ϕ2 A2 of ϕk : Ak → C (with ϕk ◦ CD,Ak ⊆
CD,C ) one has to take the “pull-back” of the cones CD,Ak as follows:
 4. THE RESIDUALLY NUCLEAR CASE 907

Let CD,P ⊆ CP(D, P ) denote the set of all maps d 7→ V1 (d)⊕V2 (d) with V1 ∈ CD,A1 ,
V2 ∈ CD,A2 , and ϕ1 (V1 (d)) = ϕ2 (V2 (d)) for all d ∈ D.
Notice that π ◦ CD,P ⊆ CD, C for π : a1 ⊕ a2 ∈ P → (1/2)(ϕ1 (a1 ) + ϕ2 (a2 )) ∈ C
and is G-equivariant, because the c.p. map W : a1 ⊕ a2 ∈ A1 ⊕ A2 7→ (1/2)(ϕ1 (a1 ) +
ϕ2 (a2 )) is G-equivariant and W ◦ (CD,A1 ⊕ CD,A2 ) ⊆ CD,C .
The co-homological case has to consider given CAk ,E ⊆ CP(Ak , E) and CC,E ⊆
CP(C, E). The assumptions need the G–invariance of the cones and maps, e.g. η(g)◦
CC,E ◦ γ(g −1 ) ⊆ CC,E , etc.
(5b) How to construct the (G, C)-lift from lifts of ϕk ?
In case that the ϕk are epimorphisms, the required lifts Tk : C → Ak (of ϕk for
homology) should satisfy Tk ◦ CD,C ⊆ CD,Ak (and should be G-equivariant). In the
co-homological situation, one has to require that CAk ,E ◦ Tk ⊆ CC,E etc.
Then T : c 7→ T1 (c) ⊕ T2 (c) is a G-equivariant c.p. split of the epimorphism
π : P → C with π ◦ T = idC , and T ◦ CD,C ⊆ CD,P .
(6) Mapping cones, Puppe sequence, Bott periodicity. (The modifications for
the C-version!):

ψ : A → B should be (G, C) equivariant (in the relevant sense, depending from

the homological or co-homological situation). The mapping cone Cψ of ψ is the
pull-back of f ∈ CB = B ⊗ C0 ((0, 1]) 7→ f (1) ∈ B and of ψ : A → B. C =: CD,B ⊆
CP(D, B) has to be extended to CD,CB := C ⊗ CP(C, C0 ((0, 1])) =: C(0, 1] ⊆
CP(D, CB).
This implies that CD,Cψ ⊆ CP(D, Cψ ) has to be the the set of all those c.p.
maps V1 ⊕ V2 : D → A ⊕ CB with V1 ∈ CD,A , V2 ∈ CD,CB and ψ(V1 (d)) = V2 (d)(1)
for all d ∈ D.
In the co-homological situation one has given CA,E and CB, E and has to
define CCB, E ⊆ CP(CB, E) and CCψ , E ⊆ CP(Cψ , E) : CCB, E := CB, E ⊗
CP(C0 ((0, 1]), C) CCψ , E as in the above given pull-back construction.
(7) The cones (and conditions in G-equivariant case) needed for the half-
exactness on (G, C)-semi-split exact sequences.
(On (G, X)-semi-split exact sequences it should be nothing in addition to re-
quire.)
Let G a second countable l.c. group, that acts on A and D via α(g) and δ(g),
and let J / A a G-invariant ideal with quotient map π : A → A/J and inclusion
ι : J → A. The action on J and A/J should be the induced one.
Further, CD,A ⊆ CP(D, A), CD,J ⊆ CP(D, J), and CD,A/J ⊆ CP(D, A/J)
are given. We require that π ◦ CD,A ⊆ CD,A/J and ι ◦ CD,J ⊆ CD,A , moreover,
that α(g)CD,A δ(g −1 ) ⊆ CD,A , α(g)CD,J δ(g −1 ) ⊆ CD,J , and [α](g)CD,A/J δ(g −1 ) ⊆
CD,A/J .
908 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

Definition 8.4.7. We say that V : A/J → A is a (G, C)-equivariant lift, if

V is completely positive, G-equivariant, and V ◦ CD,A/J ⊆ CD,A . The sequence
0 → J → A → A/J → 0 is called homological (G, C)-semi-split, if such V exists.
Similar, one has to require in the co-homological case that e.g. CA,D ◦ V ⊆
CA/J, D, etc.

The proof of the 6-term sequences of usual KK-theory applies now. because,
the (G, C)-analogs of [73, lem. 19.5.3, thm. 19.5.5] work.

Proposition 8.4.8. If the sequence 0 → J → A → A/J → 0 is homolog-

ical (G, C)-semi-split, then the six-term exact sequence of the “homology” X 7→
KKG (CD,X ; D, X) holds (cf. [73, thm. 19.5.7]).
If the sequence 0 → J → A → A/J → 0 is co-homological (G, C)-semi-split,
then the six-term exact sequence of the “cohomology” X 7→ KKG (CX,D ; X, D)
holds.

(8) Comments on the applications of KK G (C; ·, ·) in Chapters 9-12.

(8a) The σ-additivity of KKG (X; · , D) and its consequences:
The σ-additivity immediately follows from the fact that each element can be rep-
resented by a non-degenerate Kasparov module. One of the consequences is the
following lemma:

Lemma 8.4.9. Let B∞ denote the inductive limit of a sequence of (G, X)-
invariant injective morphisms ψn : Bn → Bn+1 , and suppose that the morphisms
ψn satisfy the following conditions (i) and (ii) for every n ∈ N.

(i) The element [(Bn+1 , ψn , 0)] of KKG (X; Bn , Bn+1 ) has an inverses ele-
ment in KKG (X; Bn+1 , Bn ).
(ii) For every finite subset F ⊆ Bn and each ε > 0, there are m2 > m1 ≥
n and a (G, X)-equivariant c.p. contraction V : Bm2 → Bm1 such that
kψnm1 (x) − V (ψnm2 (x))k < ε for all x ∈ F .

Then ψ1∞ : B1 → B∞ := indlim(ψn : Bn → Bn+1 ) defines a KKG (X; ·, ·)–

equivalence of B1 and B∞ , i.e., KK(ψ1∞ ) ∈ KKG (X; B1 , B∞ ) admits an inverse
in KKG (X; B∞ , B1 ).

Notice that (ii) is trivially satisfied for compact G and nuclear Bn (if the action
of X on Bn is continuous).

Proof. The condition (ii) ensures that there is a sequence n1 < n2 < · · · and
(G, X)-equivariant completely positive maps Tk : B∞ → Bnk such that

Sk := psi∞
k ◦ Tk

converges point-wise to idB∞ . It follows that the epimorphism from the ((G, X)-
contractible) mapping telescope T of the maps (ψ1 , ψ2 , . . .) onto the inductive limit
B∞ has an (G, X)-equivariant c.p. lift.
 4. THE RESIDUALLY NUCLEAR CASE 909

Then B∞ is KKG (X; ·, ·)-equivalent to the kernel of T → B∞ . After

realizing that, one is able to apply the standard arguments of cohomology
theory to h(·) := KKG (X; ·, D) (for fixed D), because of homotopy in-
variance and half-exactness on (G, X)-split-exact sequences. Then one uses
that proj − lim(1) (ψ1∗ , ψ2∗ , . . .) = 0, and gets that ψ1∞ : B1 → B∞ satisfies
that (ψ1∞ )∗ : KKG (X; B∞ , D) → KKG (X; B1 , D) defines an isomorphism for
KKG (X; B∞ , D) onto KKG (X; B1 , D) for each separable (G, X)-algebra D.
In particular, there is y ∈ KKG (X; B∞ , B1 ) with (ψ1∞ )∗ (y) = [idB1 ]. Since
ψ1∞ is (G, X)-equivariant, it defines an element z := [ψ1∞ ] := [(B∞ , ψ1∞ , 0)] of
KKG (X; B1 , B∞ ) with z ⊗B∞ y = (ψ1∞ )∗ (y) = [idB1 ]. Then

(ψ1∞ )∗ ([idB∞ ] − (y ⊗B1 z)) = z − (z ⊗B∞ y) ⊗B1 z = 0 ,

hence, y ⊗B1 z = [idB∞ ] by injectivity of (ψ1∞ )∗ . 

(8b) If A and B are trivially graded, then [(B, ψ, 0)] ∈ KKG (C; A, B) corre-
sponds to the element of ExtG (C ⊗ CP(C, C0 (R)); A, C0 (R, B)) that is defined by
the extension of A by C0 ((0, 1), B) given by the mapping cone.
(8c) Notice that the mapping cone Cψ of an (G, X)-equivariant morphism
ψ : A → B is in a natural way a (G, X)-algebra (because defined by passage to
the cone of B and a pull-back construction), and the lift a 7→ a ⊕ (a ⊗ f0 ) ∈ Cψ
is a (G, X)-equivariant. The canonical 6-term exact sequence for the (G, X)-semi-
split sequence 0 → SB → Cψ → A → 0 has as its connecting maps δ are just
the maps (Sψ)∗ KKG (X; SB, D) → KKG (X, SA, D) and ψ ∗ : KKG (X; B, D) →
KKG ; A, D) (for each separable (G, X)-algebra D). Thus, we get:

Lemma 8.4.10. The (G, X)-algebra Cψ is KKG (X; ·, ·)-trivial, if and only if,
ψ : A → B is a KKG (X; ·, ·)-equivalence. 

This has a nice application in chapter 11.

Remark 8.4.11. Suppose that G is a Polish l.c. group, and A is a separable

C *-algebra with two actions α and β of G on A. Let C = Cin ⊆ CP(A, A) the cone
of approximately inner c.p. maps of A.
If α and β are exterior equivalent then there is an invertible element in
KKG (C; (A, α), (A, β)).
Here α and β are exterior equivalent if there is a strictly continuous map
g 7→ U (g) from G into the the unitary group U(M(A)), such that U (gh) =
U (g)α(g)(U (h)), and β(g)(·) = U (g)α(g)(·)U (g −1 ) for g, h ∈ G. (Here we ex-
tend α(g) naturally to M(A). It is the same as our notation M(α(g)) in other
chapters.)

Proof. The arguments in the proof of [698, prop. 3.1] apply to our situation,
because the cone of approximately inner c.p. maps V is invariant under conjuga-
tions, i.e., β(g) ◦ V ◦ α(g −1 ) if α and β are exterior equivalent. 
910 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

5. Verifying Condition (DC) of the basic Theorem 4.4.6

The verification of Condition (DC) of Theorem 4.4.6 for our axiomatic picture
of KK(C; A, B) ∼ = Ext(C(R); A, SB) is completely contained in the following Propo-
sition 8.5.1. Let us first collect some informations that allow to give a proof at the
end of this Section.
In the following we make the overall assumptions that A and B are stable
C *-algebras (real or complex), A is separable, B is σ-unital, and E is any unital
C *-algebra.
We consider in the following Lemmata and in Proposition 8.5.4 also the (triv-
ially graded) real case. Clearly in the complex case the skew-adjoint operators
a∗ = −a are of the form a = ib with b∗ = b.
Say/Recall what is here a "non-degenerate" morphism.
Need to check first following facts:
Let A stable and separable, B stable and σ-unital, CB := C0 ([0, ∞), B),
H : A → M(B) a non-degenerate C *-morphism that is unitarily equivalent to
δ∞ ◦ H.
Consider the natural unital injection M(B) ⊂ M(CB) of M(B) in M(CB).
Then (by Kasparov ...?), U(πCB (H(A))0 ∩ (M(CB)/CB)) is equal to
U0 (πCB (H(A))0 ∩ (M(CB)/CB))
. ????

Proposition 8.5.1. Let A and B stable C*-algebras, where A is separable and

B is σ-unital, and H : A → M(B) a non-degenerate C*-morphism in “general
position” – in the sense of Definition 3.3.1, i.e., H(A)B = B and H is unitarily
homotopic to its infinite repeat δ∞ ◦ H, cf. Definition 5.0.1 and Remark 5.1.1(8).
Decide ! :
Is it equal to H(A)B = B and that H is approximately unitary
equivalent to δ∞ ◦ H in Norm or strict topology? ( 10 ).
Let S, T ∈ H(A)0 ∩ M(B) isometries with SS ∗ + T T ∗ = 1.
Suppose that there exists a strictly continuous map t ∈ R 7→ U (t) ∈ M(B) from
(−∞, +∞) into the unitaries of M(B) such that

(α) [U (t), H(a)] := U (t)H(a) − H(a)U (t) ∈ B for each a ∈ A,

(β) the map R 3 t 7→ [U (t), H(a)] from R to B is continuous for each a ∈ A,
and
(γ) limt→−∞ kU (t)H(a) − H(a)U (t)k = 0 for each a ∈ A.

Then there exist strictly continuous maps t ∈ R 7→ V (t) ∈ U(M(B)) and

t ∈ R 7→ W (t) ∈ U(M(B)) that have the following properties for each a ∈ A:

(i) U (t) ⊕S,T 1 = V (t)W (t) for all t ∈ R,

10 Then δ ◦ H is approximately unitary equivalent to H in M(B) with respect to the strict
∞
topology on M(B) in sense of Remarks 5.1.1.
 5. VERIFYING CONDITION (DC) OF THE BASIC THEOREM ?? 911

(ii) t 7→ [V (t), H(a)] is a continuous map from R into B,

(iii) limt→−∞ k[V (t), H(a)]k = 0 and limt→+∞ k[V (t), H(a)]k = 0,
(iv) W (t) = 1 for all t ≤ −1 and W (t) ∈ U(B + C · 1) ⊂ U(M(B)) for all
t ∈ R,
(v) t 7→ [W (t), H(a)] ∈ B is a continuous map.

Proposition 8.5.1 says with other words: If U ∈ M(C0 (R, B)) is a unitary with
U H(a)−H(a)U ∈ Cb (R, B) and and limt→−∞ kU (t)H(a)−H(a)U (t)k = 0 for each
a ∈ A, then U ⊕S,T 1 decomposes as U ⊕S,T 1 = V W , V, W ∈ U(M(C0 (R, B)))
with W ∈ 1 + Cb (R, B) – which implies obviously W H(a) − H(a)W ∈ Cb (R, B) –,
with W (t) = 1 for t ≤ −1 and V H(a) − H(a)V ∈ C0 (R, B) for all a ∈ A.
Notice that the assumptions of Proposition the map t ∈ R+ → U (t) ∈ M(B)
corresponding to U is only strictly continuous.

We remind first some simple facts and (easy to see) rough estimates concerning
skew-adjoint contactions −c∗ = c, c1 , . . . , cn ∈ E, kck ≤ 1 in some real or complex
unital C *-algebra E :

(1) exp(c) ∈ U0 (E) if c∗ = −c.

(If E is complex then c = ih for the self-adjoint element h = −ic ∈ E.)
(2) k [exp(c), T ] k ≤ 3 k [c, T ] k for all T ∈ E.
(Use that k[cn , T ]k ≤ (nkckn−1 ) · k[c, T ]k, kck ≤ 1 and e ≤ 3.)
(3) If c1 , . . . , cn ∈ E are skew-adjoint contraction and T ∈ E, then

k [ exp(c1 ) · exp(c2 ) · . . . · exp(cn ) , T ] k ≤ 3n max { k[ck , T ]k ; 1 ≤ k ≤ n } .

(Use part (2) and that k exp(c)k = 1 if c = −c∗ .)

In the following Lemma let B denote a σ-unital (real or complex) C *-algebra

and let D ⊆ M(B) a (real or complex) separable C *-subalgebra.
Recall that M(B) means the multiplier algebra of B and that M(B) ⊆ M(CB)
unitally by the natural non-degenerate embedding B ⊆ M(CB), where we define
here CB := C0 ((0, 1], B) ∼
= C0 ((α, β], B) for each α < β ∈ R. In this way also D ⊆
M(B) ⊆ M(CB) becomes a C *-subalgebra of M(CB) by natural identification
cf. prove of following Lemma 8.5.2 for details.

Lemma 8.5.2. Let B a σ-unital (real or complex) C*-algebra, D ⊆ M(B) a

separable C*-subalgebra of the multiplier algebra of B, b1 , . . . , bn ∈ M(CB) skew-
adjoint contractions with the property that [bk , d] ∈ CB for all d ∈ D and for each
k = 1, . . . , n.
Then, for each given fixed contractions d1 , . . . , dm ∈ D and ε > 0, there exist a
positive contraction e ∈ CB and a normal element g ∈ CB such that – with

ck := (1 − e2 )1/2 bk (1 − e2 )1/2

for k ∈ {1, . . . , n} – and for ` ∈ {1, . . . , m} holds:

(i) [ck , d] ∈ CB for all d ∈ D,

912 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

(ii) k[ck , d` ]k < ε/(6n + 7),

(iii) 1 + g ∈ U0 (C · 1 + CB), – respectively 1 + g ∈ U0 (R · 1 + CB) –, and
(iv) V · (1 + g) = exp(b1 ) · exp(b2 ) · . . . · exp(bn )
for the unitary V := exp(c1 ) · exp(c2 ) · . . . · exp(cn ) ∈ U0 (M(CB)).

It implies k [V, d` ] k < ε/2 for ` = 1, . . . , m, [V, d] ∈ CB and [g, d], gd, dg ∈ CB
for all d ∈ D.
There exist δ > 0 and −f ∗ = f ∈ CB such that, for ` = 1, . . . , m,

kf k ≤ 1 , exp(−f )(1 + g)|(0, δ] = 1|(0, δ] and k[f, d` ]k < ε/(3(n + 1)) .

It implies that h =: exp(−f )(1 + g) − 1 ∈ CB satisfies h(t) = 0 for t ∈ (0, δ],

V · (1 + g) = (V · exp(f )) · (1 + h) and

k [V · exp(f ), d` ] k < ε for ` = 1, . . . , m .

Moreover 1 + h = exp(−f )(1 + g) ∈ U0 (C1 + CB) (respectively 1 + h ∈ U0 (R1 + CB)

in case of a real C*-algebra B).

The later used point is to manage that the

product of exponentials and
the unitary in U(C0 ((−1, 0], B + C1))
becomes equal to 1 -- say at (−1, −1/2] if defined on (−1, +∞).

Proof. The natural C *-morphism from B into M(CB) is non-degenerate,

i.e., the span of B · CB is dense in CB. Thus, there is a natural strictly continuous
unital *-monomorphism from M(B) into M(CB). We identify the elements of
M(B) with there images in M(CB), i.e., M(B) ⊂ M(CB) in a natural manner.
Let f0 ∈ C0 (0, 1] the function with f0 (t) := t for t ∈ [0, 1] and let e0 ∈ B+
a strictly positive contraction in B. Then f0 ⊗ e0 ∈ CB is a strictly positive
contraction in CB and the intersection E of CB with the separable C *-subalgebra
A := C ∗ (D ∪ {b1 , . . . , bn , f0 ⊗ e0 }) of M(CB) is a separable C *-subalgebra of CB
and contains the strictly positive element f0 ⊗ e0 of CB.
Thus, the natural inclusion of E in CB defines a unital monomorphism
φ : M(E) → M(CB) with the property that A ⊆ φ(M(E)) ⊆ M(CB).
It follows that C ∗ (f0 ⊗ e0 )+ ⊆ CB contains a quasi-central approxi-
mate unit 0 ≤ e1 ≤ e2 ≤ . . . for the elements in C ∗ (f0 ⊗ e0 )+ , cf. [616,
thm. 3.12.14, cor. 3.12.15]. It is important that we can find the en with the
additional property ek e` = ek for 1 < k < `.
This allows to apply the “mirror method”. (The new application to the mir-
rored problem can be done without harming the before done steps.) The result is a
decomposition of U into unitaries U = V W with W ∈ 1 + J and [V, A] ⊆ C0 (R, B).

In particular, for each finite subsets X ⊂ A, Z ⊂ (A∩CB) and each δ > 0 there
exists a positive contraction e ∈ CB+ with kex−xek < δ and kz−zek+kz−ezk < δ
 5. VERIFYING CONDITION (DC) OF THE BASIC THEOREM ?? 913

for x ∈ X and z ∈ Z. (It suffices to take here Z := {f0 ⊗ e0 } and then the δ > 0
sufficiently small.)
?? See for this also Remark 5.1.1(1) and its proof in Section 1 and the Section
3 of Chapter 5 for more details.??
By Proposition 5.3.1, for all a ≥ 0, every contraction x and each β ∈ [1, ∞),

k[x, a1/β ]k ≤ 3k[x, a]k1/β .

Apply this to β := 2, a := 1 − e2 and x with kxk ≤ 1 and get

k[x, (1 − e2 )1/2 ]k ≤ 3k[x, e2 ]k1/2 ≤ 5k[x, e]k1/2 .

(i): Let e ∈ (CB)+ a positive contraction. Then b−(1−e2 )1/2 b(1−e2 )1/2 ∈ CB
for all b ∈ M(CB) because πCB (1) = πCB ((1 − e2 )1/2 ) for the quotient map
πCB : M(CB) → M(CB)/CB.
If we apply πCB to c := (1 − e2 )1/2 b(1 − e2 )1/2 ∈ M(CB) for b ∈ M(CB) with
the property [b, d] ∈ CB for all d ∈ D ⊆ M(CB), we get that πCB (c) = πCB (b)
and πCB ([c, d]) = πCB ([b, d]). Thus [c, d] ∈ CB for all d ∈ D if [b, d] ∈ CB for all
d ∈ D.
(ii): k[ck , d` ]k ≤ k[bk , d` ]k + 2k[(1 − e2 )1/2 , d` ]k ...
But we have only [bk , d` ] ∈ CB, need that k[ck , d` ]k is small.
Since bk ∈ M(CB), d` ∈ M(B) ⊆ M(CB) and [bk , d` ] ∈ CB for k = 1, . . . , n
` = 1, . . . , m – by assumptions –, we find for given δ > 0 a positive contraction
e ∈ C ∗ (f0 ⊗ e0 )+ ⊆ CB with k[e, bk ]k < δ, k[e, d` ]k < δ and k(1 − e)[bk , d` ]k +
k[bk , d` ](1 − e)k < δ f := (1 − e2 )1/2 , 1 ≤ k ≤ n and 1 ≤ ` ≤ m.
This ????????????
Let b := bk and d := d` . All are positive contractions.
(f bf d − df bf ) = f b(f d − df ) − (f d − df )bf + f (bd − db)f can be estimated by
2kbkkf d − df k + kf (bd − db)f k. Since bd − db ∈ CB
Let γ := ε/(6n + 7). Since

G := C ∗ (D, b1 , . . . , bn ) ⊆ φ(M(E)) ⊆ M(CB)

is separable we find a quasi-central approximate unit in the positive contractions

of CB for the elements of G by [616, thm. 3.12.14, cor. 3.12.15], i.e., we find a
positive contraction e ∈ CB with k[e, x]k < γ/2 for x ∈ {d1 , . . . , dm , b1 , . . . , bn }.
and k(1 − e)[?, ?]k + k[?, ?](1 − e)k <????? (small) for ????
Find contraction e ∈ CB+ with k[(1 − e2 )1/2 bk (1 − e2 )1/2 , d` ]k < γ ...
(iii,iv): Define g by

V −1 · exp(b1 ) · exp(b2 ) · . . . · exp(bn ) − 1

g :=

for V := exp(c1 ) · exp(c2 ) · . . . · exp(cn ) .

914 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

Since πCB (ck ) = πCB (bk ) it follows that πCB (V ) = πCB exp(b1 ) · exp(b2 ) · . . . ·
exp(bn ) and therefore πCB V −1 · exp(b1 ) · . . . · exp(bn ) = πCB (1), i.e., g ∈ CB.

Clearly 1 + g ∈ U(C · 1 + CB), respectively 1 + g ∈ U(R · 1 + CB) ∩ (1 + CB)

in real case. In complex case we have always U(C · 1 + CB) = U0 (C · 1 + CB) and
in case of real C *-algebras B holds U(R · 1 + CB) ∩ (1 + CB) ⊆ U0 (R · 1 + CB).
To be filled in??
It follows that k [V, d` ] k < ε/2 for ` = 1, . . . , m, [V, d] ∈ CB and [g, d], gd, dg ∈
CB for all d ∈ D.
There exist δ > 0 and −f ∗ = f ∈ CB such that kf k ≤ 1, exp(−f )(1 +
g)|(0, δ] = 1|(0, δ] and k[f, d` ]k < ε/(3(n + 1)) for ` = 1, . . . , m.
It implies that h =: exp(−f )(1+g)−1 ∈ CB, h(t) = 0 for t ∈ (0, δ], V ·(1+g) =
(V · exp(f )) · (1 + h) and

k [V · exp(f ), d` ] k < ε for ` = 1, . . . , m .

The following Lemma remains to be checked:

In the following let A and B denote stable (real or complex) C *-algebras. We
require that A is separable and that B is σ-unital. We fix a a non-degenerate C *-
morphism H : A → M(B) in general position, i.e., H(A)B = B and H is in M(B)
approximately unitary equivalent to its infinite repeat δ∞ ◦ H.

Lemma 8.5.3. Let A, B stable C*-algebras, A separable B σ-unital, H : A →

M(B) non-degenerate faithful and in “general position”, i.e., H and δ∞ ◦H unitary
homotopic in M(B).
Given U ∈ U(M(B[0, 2])) with U (t) = 1 for t ∈ [0, 1) and b1 , . . . , bn ∈
M(B[0, 2]), b∗k = −bk with [bk , H(a)] ∈ C0 ((1, 2], B) for k = 1, . . . , n, for all a ∈ A,
and such that
U (t) = exp(b1 (t)) · . . . · exp(bn (t))W (t)
with W |[0, 1] ∈ 1 + C([0, 1], B).
We find modifications b0k of bk and W 0 (t) of W (t) (the latter inside
1 + C([0, 1], B) ) such that such that W 0 (t) = 1 for t ∈ [0, 1/2], W 0 (1) = W (1),
b0k (1) = bk (1) and

U (t) = exp(b01 (t)) · . . . · exp(b0n (t)) · W 0 (t)

on [0, 1], and limt→0 b0k (t) = 0 strictly.

Perhaps use an approximately central unit of B? With b0k (t) := (1 −

e(t) ) bk (t)(1 − e(t)2 ). Converges strictly to 0.
2 1/2

πB (exp(b01 (t)) · . . . · exp(b0n (t))) = πB (exp(b1 (t)) · . . . · exp(bn (t)))

 5. VERIFYING CONDITION (DC) OF THE BASIC THEOREM ?? 915

Proposition 8.5.4. Let A and B stable C*-algebras (real or complex), where

A is separable and B is σ-unital, and let H : A → M(B) a non-degenerate C*-
morphism “in general position” (i.e., H is approximately unitary equivalent to δ∞ ◦
H).
in strict or norm topology ???
Compare Chp. 5
Moreover, let S, T ∈ H(A)0 ∩ M(B) isometries with SS ∗ + T T ∗ = 1.
The then following holds:
If α < β < γ ∈ R and t ∈ [α, γ] → u(t) ∈ U(M(B)) is a strictly continuous map
into the unitaries of M(B) with u(t) = 1 for t ∈ [α, β] and u(t)H(a)−H(a)u(t) ∈ B
for each a ∈ A and t ∈ [α, γ].
Assume moreover that the maps t 7→ u(t)H(a) − H(a)u(t) ∈ B is continuous
on [α, γ] for each a ∈ A (i.e., U := {u(t)} derives H(A) into into C([α, γ], B).
Then for each finite subset X ⊂ A and every ε > 0 there exist n ∈ N and a
decomposition u(t) ⊕S,T 1 = v(t) · w(t) into a product of strictly continuous maps
t ∈ [α, γ] → v(t) ∈ U(M(B)) and t ∈ [α, γ] → w(t) ∈ U(M(B)) such that w(t) =
1 = v(t) for t ∈ [α, (α + β)/2], w(t) ∈ B + C · 1 for t ∈ [α, γ]
one can only deduce that it w(t) is strictly continuous!!!!
with t ∈ [α, γ] → w(t) norm-continuous, even with w(t) ∈ 1 + B[α, γ] ??
and
v(t) := exp(b1 (t)) · exp(b2 (t)) · . . . · exp(bn (t))
7 bk (t) = −bk (t)∗ ∈ M(B) from ((α +
for suitable strictly continuous maps t →
P
β)/2, γ] with bk (t) ∈ Der(H(A), B) and k k[H(a), bk (t)]k < ε for all a ∈ X and
t ∈ ((α + β)/2, γ], and the maps
t ∈ ((α + β)/2, γ] 7→ [H(a), bk (t)] ∈ B
are (norm-) continuous and satisfy
lim k[H(a), bk (t)]k = 0 .
t→(α+β)/2

Proof. The problem is:

Given b1 (t), . . . , bn (t), with b∗k = −bk ∈ M(C0 ((−1/2, 0], B))
strictly continuous with respect to t ∈??? and bounded on (−1/2, 0]
with
t 7→ bk (t)H(a) − H(a)bk (t) ∈ C0 ((−1/2, 0], B)
and 1 − exp(b1 ) · . . . · exp(bn ) ∈ C0 ((−1/2, 0], B) .
Can we connect it in a controlled way inside
[−1, 0] and U(B + C · 1) with 1 ?? 

Notice that we do not require that limt→(α+β)/2 bk (t) = 0. Even not strictly.
But ?????
It can happen that distance of w(γ) from C · 1M(B) is equal to 2.
916 8. THE ISOMORPHISM OF KK(C; A, B) AND Ext(SC; A, SB)

Remark 8.5.5. It seems that one can show that the number n ∈ N in Propo-
sition 8.5.4 has an universal upper bound (clearly not depending on ε).
But it could be that ????????
Perhaps one get it by an indirect argument ?
For example one could consider for B the cone over c0 (B1 , B2 , . . .) from a
sequence with A = K ⊗ (C(0, 1] ∗ C(0, 1]) of hn : A → M(Bn ), α := −1, β := 0,
γ := 1, un : [−1, 1] → U(M(B)) –

We are now in position to give the Proof of Proposition 8.5.1.

Proof of Proposition 8.5.1. We can use a ‘‘mirror principle’’ and

apply induction ... 
 CHAPTER 9

Scale-invariant maps (i+ii of Thm’s B,M)

NEEDED???/CHECK???/PROBLEM:
In case A, B, stable and separable, that ???

Ext(C; A, B) ∼
= ker(U(πB (HC (A))0 ∩ Q(B)) → U(Q(B))) .

THIS should apply to Ext(C(0, 1] ; A, C0 ((0, 1], B)) = 0 and gives then that
each strictly continuous path of unitaries t ∈→ U (t) ∈ M(B) with

lim kU (t)a − aU (t)k = 0

t→0

and U (t)a − aU (t) ∈ B for all t ∈ (0, 1]

This has NOT been proven somewhere: is a finite product of exponentials
exp(h(t)) with h(t)∗ = −h(t) and limt→0 kh(t)a − ah(t)k = 0 and h(t)a − ah(t) ∈ B.
The number of this exponentials would then have moreover an universal upper
bound.
The injectivity of

ϑ : R(C; A, B) → KK(C; A, B)

requires to use that C = C(h0 ) and we must show that the elements in the kernel
of ϑ are homotopic in R(C; A, B) (via boundary maps from R(C[0, 1]; A, B[0, 1]) at
{0, 1} to R(C; A, B)) to the class [h0 ].
Moreover, one has to show with the method of N.Ch. Phillips that “boundary-
homotopic” elements of SR(C; A, B) define the same element of Gr(SR(C; A, B)).
??
In the meantime we rediscovered also our direct proof of the injectivity
of the natural group homomorphism from SR(C; A, B) onto Ext(C; A, SB) ∼ =
KK(C; A, B) is injective. The proof of the injectivity uses a mirror-principle and
triviality of KK(C; A, C(R− , B)), which follows from the homotopy invariance of
KK via KKc = KKoh = KKh (an observations of G. Kasparov).

We prove the parts (i) and (ii) of Theorem B and of Theorem M in this Chapter.
We assume the existence of the non-degenerate nuclear C *-morphism h0 : A → B as
described in Chapter 1 before Theorem M. The existence of h0 comes from Theorem
A (in the case of Theorem B), and, — if B is separable, stable, strongly purely
infinite and contains an Abelian regular C *-subalgebra —, comes from Corollary
6.3.2.
917
918 9. SCALE-INVARIANT MAPS (I+II OF THM’S B,M)

The general existence of h0 comes later from Theorem K (in the case of Theorem
M). Theorem K will be proven in Chapter 12 with help of Corollary 6.3.2 and
Corollary 9.1.7.
In Section 1 we derive also some useful applications of asymptotic maps that
are asymptotically scale invariant up to approximate unitary equivalence.
The proof, as given in the second part of this Chapter, is (in conjunction
with Theorem 4.4.6) a more detailed reformulation of the proof in the appendix
of our preprint from December 1994 (3rd draft), more precisely: We show that
the natural morphism from R(A, B) into KKnuc (A, B) (resp. from R(X, A, B) into
KKnuc (X; A, B) for X := Prim(B)) is an isomorphism, cf. Corollary 9.4.2. Here
we start with the *-monomorphism h0 described below Theorem A (resp. before
Theorem M) which defines the zero element of the R- and KKnuc -groups and reduce
the proofs of the parts (i) and (ii) of Theorems B and M to Theorem 4.4.6. This
reduction requires results of Chapters 5 and 7, and Lemma 8.3.2.
Note that the material of Chapter 8 are modifications of well-known results
of Kasparov, Skandalis and others, and can also be deduced from textbooks as
e.g. [73], [389], at least in the situation of nuclear C *-algebras, and only so far
as it is needed for Corollary C, if the reader is only interested in the classification
result Corollary C.
Then we use the isomorphism R(A, B) ∼ = KKnuc (A, B) (resp. R(X; A, B) ∼ =
KKnuc (X; A, B), R(C; A, B) ∼ = KKnuc (X; A, B)) to conclude with help of Corol-
lary 8.3.4,(ii) (respectively with help of the generalization of Corollary 8.3.4 for
Extnuc (X; ., .) and Ext(C; ., .)) that the elements of R(A, B) (respectively of
R(X; A, B)) are invariant under scaling in the sense of the below given Definition
9.1.1. We conclude that every element of R(A, B) (respectively of R(X; A, B),
R(C; A, B)) is given by a nuclear (respectively by a morphism ϕ : A ,→ B with
ϕ ∈ C, e.g. for C = CPrn (X; A, B)). But this proves parts (i) and (ii) of Theorems
B and M.
More generally, we show that
if A is separable, B is σ-unital, A and B are stable, h0 = k(· ⊗ 1) for some non-
degenerate *-monomorphism k : A ⊗ O2 ,→ B, and if C ⊆ CP(A, B) denotes the
point-norm closed matrix operator-convex cone generated by h0 , then the natural
morphism
R(C; A, B) → Ext(C ⊗ CP(C, C0 (R)); A, B ⊗ C0 (R))
is an isomorphism.
The homotopy invariance of KK(C; A, B) and the natural isomorphism

Ext(C ⊗ CP(C, C0 (R)); A, B ⊗ C0 (R)) ∼

= KK(C; A, B)

(shown in Chapter 8) then imply that every element of KK(C; A, B) can be repre-
sented by a morphism h from A into B that is asymptotically dominated by h0 (i.e.,
by h ∈ Hom(A, B) with [h] ∈ SR(C; A, B)). Then the definition of R(C; A, B) =
[h0 ] + SR(C; A, B) ∼
= Gr(SR(C; A, B)) and the relations in SR(C; A, B) yield that
 1. SCALE-INVARIANT ELEMENTS OF SR(X; A, B) 919

[h] = [k] in KK(C; A, B) if and only if h ⊕ h0 and k ⊕ h0 are unitarily equivalent.

(The latter can be deduced from the identity and isomorphism
G(h0 ; A, Q(R+ , M(B))) = R(C; A, B) ∼
= KK(C; A, B)
and the considerations of Chapter 4 on this type of groups.)

1. Scale-invariant elements of SR(X; A, B)

We introduce a notion of generalized invariance under scaling for asymptotic

maps:
Consider a topological isomorphism σ from R+ onto R+ . That is, σ is a bijective
continuous function of R+ onto R+ . Note that σ(0) = 0 and that σ has to be strictly
increasing.
Now let F be a C *-algebra. The homeomorphism σ induces an automorphism σ b
of Q(R+ , F ) in a natural way by mapping π(f ) to π(f ◦σ), for f ∈ Cb (R+ , F ), where
π : Cb (R+ , F ) → Q(R+ , F ) denotes the quotient map. Recall from Chapter 7 that
Q(R+ , M(F )) is a unital C *-subalgebra of the multiplier algebra M(Q(R+ , F )) of
Q(R+ , F ).

Definition 9.1.1. Let X be a set and h : X → Q(R+ , F ) a map. The map

h will be called invariant under scaling (up to unitary equivalence) , if, for
b ◦ h is unitarily equivalent to
every topological isomorphism σ from R+ onto R+ , σ
h by a unitary in Q(R+ , M(F )) ⊆ M(Q(R+ , F )).
We say that h is approximately scale-invariant , if, for every topological
b ◦ h is approximately unitarily equivalent to h
isomorphism σ from R+ onto R+ , σ
by unitaries in Q(R+ , M(F )).
An class [h] ∈ [Hom(A, Q(R+ , F ))] will be called invariant under scaling if
σ ◦ h] for every topological isomorphism σ from R+ onto R+ ( 1 ).
[h] = [b

Recall form Chapters 5 and 7 that, for a C *-algebra F ,

Q(R+ , F ) := Cb (R+ , F )/ C0 (R+ , F ) ,
that Q(R+ , M(F )) is in a natural way a C *-subalgebra of the multiplier algebra
M(Q(R+ .F )) of Q(R+ , F ), and that, for every separable subset Y of Q(R+ , F ) and
unitary v in the multiplier algebra of Q(R+ , F ) there is a unitary u ∈ Q(R+ , M(F ))
with u∗ bu = v ∗ bv for every b ∈ Y , so far as F is stable or is unital, cf. Proposition
7.4.1. Thus, one can replace Q(R+ , M(F )) by M(Q(R+ , F )) in Definition 9.1.1 if
F is unital or F is stable.
It is obvious from Definition 9.1.1 that, for every set X and every map
k : X → F into the “constant” elements of Q(R+ , F ) and for every unitary
u ∈ Q(R+ , M(F )), the map h(x) := u∗ k(x)u is, up to unitary equivalence, invari-
ant under scaling. The following Proposition 9.1.2 and Corollary 9.1.3 show that
1 That means, by our definition of the classes [ · ] in Chapters 4 and 7, that σ
b ◦ h is unitarily
equivalent to h in Q(R+ , M(F )), i.e., h is invariant under scaling up to unitary equivalence.
920 9. SCALE-INVARIANT MAPS (I+II OF THM’S B,M)

for separable h(X) this is the only example of a map which is scaling invariant up
unitary equivalence.

Proposition 9.1.2. Suppose that X is a compact metric space, F is a σ-unital

C*-algebra, and that h : X → Q(R+ , F ) is a continuous map.
If h is approximately scale-invariant, then there exists a continuous map h1
from X into F , such that h1 is unitarily equivalent to h by a unitary U in the
C*-subalgebra Q(R+ , M(F )), of the multiplier algebra of Q(R+ , F ).

The following proof shows that we can find the map h1 : X → F and the unitary
U such that U = W +C0 (R+ , M(F )) with a unitary W ∈ Cb (R+ , M(F )) such that
W (1) = 1M(F ) . It implies for the above example h := u∗ k(·)u, that there exists a
unitary W0 ∈ M(F ) ⊂ Cb (R+ , M(F )) such that h1 := W0∗ k(·)W0 and U = W0∗ u
are as desired.

Proof. Since h ∈ C(X, Q(R+ , F )) and since Q(R+ , F ) is a quotient C *-

algebra of Cb (R+ , F ), there exists T ∈ C(X, Cb (R+ , F )) such that h(b) = T (b) +
C0 (R+ , F ) for every b ∈ X, i.e., T : X → Cb (R+ , F ) is a topological lift of h.
If σ is a topological isomorphism of R+ , then the bounded continuous function
t 7→ T (b)(σ(t)) ∈ F is a lift of σ
b(h(b)) to Cb (R+ , F ).
Let u be a unitary in Q(R+ , M(F )) such that, for all b ∈ X,

ku∗ h(b)u − σ
b(h(b))k < ε/8 .

There exists a unitary w in Cb (R+ , M(F )) such that u = w + C0 (R+ , M(F )). w
is given by a norm continuous map w : t ∈ R+ → w(t) ∈ U(M(F )) ( 2 ). It follows
that
γ(b, t) := kT (b)(σ(t)) − w(t)∗ T (b)(t)w(t)k
satisfies, for every b ∈ X,

lim γ(b, t) = ku∗ h(b)u − σ

b(h(b))k < ε/8 .
t→∞

On the other hand, by the triangle inequality, for t ∈ R and a, b ∈ X,

|γ(a, t) − γ(b, t)| ≤ 2kT (a) − T (b)k

(where the last norm comes from C(X, Cb (R+ , F ))), i.e., γ(b, t) is uniformly con-
tinuous on the compact space X and, therefore,

lim (sup γ(b, t)) < ε/8 .

t→∞ b∈X

Let y ≥ 0. For x > y and ε > 0, consider the set M (x, ε) of the s > x with the
property that there exists a norm continuous map

v : t ∈ [x, s] → v(t) ∈ M(F )

2Note that the choice of w depends from u and thus from σ.

 1. SCALE-INVARIANT ELEMENTS OF SR(X; A, B) 921

from [x, s] into the unitary group U(M(F )) of the multiplier algebra of F with
v(x) = 1 and
kv(t)∗ T (b)(t)v(t) − T (b)(x)k < ε (1.1)
for b ∈ X and t ∈ [x, s] ( 3 ). From the definition of M (x, ε) we see that M (x, ε/2) ⊆
M (x, ε) and t ∈ M (x, ε) if x < t < s and s ∈ M (x, ε).
We define m(x, ε) := sup M (x, ε) ∈ [x, +∞]. The real number 2m(x, ε) is not
in M (x, ε) if m(x, ε) < ∞, because x > 0.
The properties of M (x, ε) and
???? what else? ??
lead to
m(x, ε/2) ≤ m(x, ε)
for ε > 0 . Moreover, x < m(x, ε/2) for every ε > 0, because T is (uniformly)
continuous.
Intuition says that for every y ∈ R+ and every ε > 0 there exists at least one
x > y with m(x, ε) = ∞. But we give a very detailed proof of this almost obvious
fact:
Let ε > 0 and y ∈ R+ be fixed. We show below that the assumption that

m(x, ε) < ∞ for every x > y (1.2)

contradicts that h is of approximate scaling invariant.

If m(x, ε) < ∞ for every x > y, then we can find a sequence (xn ) in (y, ∞)
such that, for every n,

n < xn < m(xn , ε/2) < 2m(xn , ε) < xn+1 < ∞ . (1.3)

This allows us to define a topological isomorphism of R+ by taking a strictly in-

creasing continuous piecewise linear map σ : R+ → R+ , satisfying

σ(xn ) = xn and σ(2m(xn , ε)) = xn + (m(xn , ε/2) − xn )/2 for every n.

As we have seen above, by the assumption of approximate invariance of h under

scaling (applied to σ), there exists a norm continuous map t 7→ w(t) ∈ U(M(F ))
with
lim sup kT (b)(σ(t)) − w(t)∗ T (b)(t)w(t)k ≤ ε/8.
t→∞ b∈X

Thus there exists xn in our sequence such that

kw(t)∗ T (b)(t)w(t) − T (b)(σ(t))k < ε/4 (1.4)

for t ≥ xn and b ∈ X.
On the other hand, by the definition of m(x, ε/2), we can find a norm continuous
map t 7→ z(t) ∈ U(M(F )) such that z(xn ) = 1 and

kz(t)∗ T (b)(t)z(t) − T (b)(xn )k < ε/2 (1.5)

3 Note here that v can depend on s and ε.

922 9. SCALE-INVARIANT MAPS (I+II OF THM’S B,M)

for t ∈ [xn , xn + (m(xn , ε/2) − xn )/2] and b ∈ X.

Since σ maps [xn , 2m(xn , ε)] onto [xn , xn + (m(xn , ε/2) − xn )/2] and since z(t)
is unitary, we get from equations (1.4) and (1.5) that, for t ∈ [xn , 2m(xn , ε)] and
b∈X,

kw(xn )T (b)(xn )w(xn )∗ − T (b)(xn )k < ε/4,

kz(σ(t))∗ w(t)∗ T (b)(t)w(t)z(σ(t)) − z(σ(t))∗ T (b)(σ(t))z(σ(t))k < ε/4,
kz(σ(t))∗ T (b)(σ(t))z(σ(t)) − T (b)(xn )k < ε/2 .

Let v(t) := w(t)z(σ(t))w(xn )∗ for t ∈ [xn , 2m(xn , ε)]. Then the equation v(xn ) = 1
and applications of the triangle inequality and of the (isometric) inner automor-
phisms w(xn )(·)w(xn )∗ show (together) that

kv(t)∗ T (b)(t)v(t) − T (b)(xn )k < ε

for every t ∈ [xn , 2m(xn , ε)] and b ∈ X. It gives 2m(xn , ε) ∈ M (xn , ε). This
leads to the (desired) contradiction by inequalities (1.3) because m(xn , ε) < ∞ (by
assumption).

We have seen above that, for every y ∈ R+ and ε > 0, there exist x > y
with m(x, ε) = ∞ . Hence, we find a sequence 0 < x1 < x2 < . . . ∈ R such that
xn+1 > max(xn , n) and m(xn , 2−n ) = ∞.
By the definition of M (xn , 2−n ) and of m(xn , 2−n ) there are norm continuous
maps vn from [xn , xn+1 ] into U(M(F )) such that vn (xn ) = 1 and

kvn (t)∗ T (b)(t)vn (t) − T (b)(xn )k < 2−n

for every t ∈ [xn , xn+1 ] and every b ∈ X.

By induction we define a norm continuous map W from R+ into the unitaries
of M(F ) by W (t) := 1 for t ∈ [0, x1 ] and W (t) := vn (t)V (xn ) for t ∈ [xn , xn+1 ].
Since vn (xn ) = 1 and vn is norm continuous, the map W : t 7→ W (t) is well-
defined and is continuous in norm. Then we have

kW (t)∗ T (b)(t)W (t) − W (s)∗ T (b)(s)W (s)k < 2(1−n)

for t, s ∈ [xn , xn+1 ] and b ∈ X. Iterated use of the triangle equation gives

kW (t)∗ T (b)(t)W (t) − W (s)∗ T (b)(s)W (s)k < 2(2−n) for all s, t ≥ xn .

Thus,
h1 (b) := lim W (t)∗ T (b)(t)W (t)
t→∞

exists in F for b ∈ X. If we consider h1 as a map from X to F ⊆ Q(R+ , F )

then h1 is unitarily equivalent to h by the unitary U = W + C0 (R+ , M(F )) in
Q(R+ , M(F )). Hence h1 is again continuous. 

Corollary 9.1.3. Let D be a separable C*-algebra, F a stable or a unital

C*-algebra and h : D → Q(R+ , F ) a C*-morphism.
 1. SCALE-INVARIANT ELEMENTS OF SR(X; A, B) 923

If h is approximately scale-invariant, then there exists a (“constant”) C*-

morphism k from D into F ⊆ Q(R+ , F ) such that h is unitarily equivalent to k
by a unitary in Q(R+ , M(F )).
In particular, for separable stable exact A and σ-unital stable B, an element [h]
of SR(X, A, B) is invariant under scaling if, and only if, [k] = [h] for a “constant”
C*-morphism k from A into B ⊆ Q(R+ , B).

Proof. Let d1 , d2 , . . . be a dense sequence in the unit ball of D and let X :=

{0} ∪ {2−n dn : n = 1, 2, . . .}. Then D is the closed span of the compact set X, and
the restriction of h to X satisfies the requirements of Proposition 9.1.2.
Thus, there is continuous map h1 : X → F and a unitary u in Q(R+ , M(F )),
which is considered as a C *-subalgebra of the multiplier algebra of Q(R+ , F ), such
that u∗ h(d)u = h1 (d) for d ∈ X.
The map k : d ∈ D 7→ u∗ h(d)u is a C *-morphism from D into Q(R+ , F ) which
extends h1 to a C *-morphism k from D into Q(R+ , F ), and k is unitarily equivalent
to h by a the unitary u in Q(R+ , M(F )).
k(D) ⊆ F , because the linear span of X is dense in D and k|X = h1 , i.e., k is
“constant”.

If D := A and F := B, then, by Proposition 7.4.1 the various versions of unitary

equivalence coincide and we may consider the unitary equivalence by unitaries in
Q(R+ , M(F )). Then the scale-invariance of [h] means that h is invariant under
scaling up to unitary equivalence. Hence h is approximately scale-invariant. 

We get the following useful Corollaries 9.1.4 and 9.1.6, which we apply in Chap-
ters 10 and 12.

Corollary 9.1.4. Suppose that A and B are C*-algebras, where A is stable

and separable. Let h0 : A → B be a non-degenerate *-monomorphism.
Then h0 is unitarily homotopic to h0 ⊕ h0 , if and only if, there exists a *-
monomorphism k : A ⊗ O2 → B, such that k0 (a) := k(a ⊗ 1) is unitarily homotopic
to h0 .
The monomorphism k can be chosen to be non-degenerate, and then k0 (A)
commutes with a copy of O2 which is unitally contained in M(B).

Proof. B is stable and σ-unital, because A is stable and separable and h0 is

non-degenerate, i.e., h0 (A)B is dense in B.
Suppose that k : A ⊗ O2 → B is such that k0 (a) := k(a ⊗ 1) is unitarily
homotopic to h0 . Then k(A ⊗ O2 ) is a stable C *-subalgebra that generates B as
a closed ideal, because h0 is non-degenerate. By Corollary 5.5.6(iv), k is unitarily
homotopic to a non-degenerate *-monomorphism from A ⊗ O2 into B, because B
is stable and σ-unital. Therefore, we can assume that k is non-degenerate.
924 9. SCALE-INVARIANT MAPS (I+II OF THM’S B,M)

The extension M(k) of k to a unital *-monomorphism from M(A ⊗ O2 ) to

M(B) maps 1 ⊗ O2 unitally into the commutant of k0 (A). Thus, k0 is unitarily
equivalent to k0 ⊕ k0 . It follows that h0 is unitarily homotopic to h0 ⊕ h0 .
Now suppose, that h0 is unitarily homotopic to h0 ⊕ h0 .
By Proposition 7.4.2, there exists, up to unitary equivalence by unitaries in
Q(R+ , M(B)), a unique *-monomorphism k1 : A⊗O2 → Q(R+ , B) with k1 (a⊗1) =
h0 (a) such that k1 ⊕ k1 is unitarily equivalent to k1 in Q(R+ , M(B)).
If we compose k1 with σb for a scaling σ of R+ , then σ
bk1 has the same properties
as k1 with respect to h0 . Thus, σ bk1 is unitarily equivalent to k1 by Proposition
7.4.2.
By Corollary 9.1.3, k1 is unitarily equivalent to a *-monomorphism k from
A ⊗ O2 into B by a unitary in Q(R+ , M(B)). But then k0 = k((·) ⊗ 1) is unitarily
equivalent to h0 in Q(R+ , M(B)), and this means that h0 and k0 are unitarily
homotopic. 

Remark 9.1.5. In view of the following assumptions on asymptotic embeddings

of a separable C *-algebra A it seems to be useful to remember the following basic
observations at the beginning of Chapter 3:
Let A a C *-algebra. Then the following are equivalent:

(i) A is an exact C *-algebra.

(ii) There exists a faithful nuclear *-representation ρ : A → L(H) over some
Hilbert space H.
(iii) Each C *-morphism h : A → B into a weakly injective C *-algebra B is
nuclear.
(iv) There exists a nuclear faithful C *-morphism h : A → C into some C *-
algebra C.
(v) There exists a nuclear completely isometric embedding of A into some
C *-algebra D.

Proof. The equivalence of (i) and (ii) is discussed in Remarks ?? ????? Sec-
tion 1 of Chapter 3. The equivalence of (ii)-(v) follow essentially from the Arveson
extension theorem [42], i.e., that each of the conditions (iii)-(v) implies that the
faithful *-representations of A on a Hilbert space H are nuclear, cf. Chapter 3 for
more details. 

Corollary 9.1.6. Suppose that A and B are stable C*-algebras, where A is

separable and exact, and that k is a nuclear *-monomorphism from A ⊗ O2 into
Q(R+ , B).
Let k0 (a) := k(a ⊗ 1) for a ∈ A.
If k0 (A) ∩ J1 = k0 (A) ∩ J2 for all pairs of closed ideals J1 and J2 of Q(R+ , B)
with J1 ∩B = J2 ∩B, then there exists a nuclear *-monomorphism h from A⊗O2 into
B, such that the *-monomorphisms h0 := h((·) ⊗ 1) and k0 from A into Q(R+ , B)
are unitarily equivalent by a unitary in Q(R+ , M(B)).
 1. SCALE-INVARIANT ELEMENTS OF SR(X; A, B) 925

If, moreover, B is σ-unital, and B is contained in the closed ideal of Q(R+ , B)

that is generated by the image of k, then h can be found such that, moreover, h0 is
non-degenerate and unitarily equivalent to h0 ⊕ h0 by a unitary in M(B).

Proof. If σ is a homeomorphism of R+ , a ∈ A, I ∈ I(Q(R+ , B)), then

b−1 (I), if and only if, k0 (a) ∈ I, because
bk0 (a) ∈ I, if and only if, k0 (a) ∈ σ
σ
B∩I =B∩σ b−1 (I).
Thus, k0 and k1 := σ b ◦ k0 are nuclear *-monomorphisms, such that k0 and
k1 extend to C *-morphisms k and σ bk from A ⊗ O2 into Q(R+ , B), and, for every
I ∈ I(Q(R+ , B)),
k0−1 (k0 (A) ∩ I) = k1−1 (k1 (A) ∩ I) .
By Corollary 7.4.3, this implies that k0 and k1 are approximately unitarily equiva-
lent.
Therefore k0 is approximately scale-invariant. By Corollary 9.1.3, k0 is unitarily
equivalent to a C *-morphism h1 from A into B by a unitary in Q(R+ , M(B)). By
Corollary 7.4.3, k0 and k0 ⊕ k0 are unitarily equivalent in Q(R+ , M(B)). Thus
h1 is unitarily homotopic to h1 ⊕ h1 . Now we can proceed as in the proof of
Corollary 9.1.4, and get a *-monomorphism h : A ⊗ O2 → B, such that h0 :=
h((·) ⊗ 1) is unitarily homotopic to h1 . Thus, h0 and k0 are unitarily equivalent in
Q(R+ , M(B)).
It follows that h0 is a nuclear map, if we consider it as a map from A into
Q(R+ , B). But B is relatively weakly injective in Q(R+ , B), i.e., there is a normal
conditional expectation from the second conjugate of Q(R+ , B) onto the second
conjugate of B ( 4 ). Thus h0 is also a nuclear map from A into B.
We use the nuclearity criteria (i) in Remark 3.1.2, to show that h is nuclear if
h0 is nuclear: Let C be a C *-algebra. Consider the natural C *-morphism
g := idC ⊗max h : C ⊗max (A ⊗ O2 ) → C ⊗max B .
The restriction of g to C ⊗max (A ⊗ 1) annihilates the kernel of the natural epi-
morphism from C ⊗max (A ⊗ 1) onto C ⊗ (A ⊗ 1), because h0 is nuclear. Since, by
nuclearity of O2 , on the algebraic tensor product (C ⊗ A) O2 there is a unique
C *-norm, we get that g annihilates the kernel of the natural epimorphism from
C ⊗max (A ⊗ O2 ) onto C ⊗ (A ⊗ O2 ). Thus h is nuclear.

Now we suppose, in addition, that the closed ideal I0 of Q(R+ , B), which is
generated by k0 (A), contains B, and that B is σ-unital.
Let J denote the closed ideal of B, which is generated by h0 (A). The ideal
Q(R+ , J) of Q(R+ , B) contains k0 (A) and I0 , because h0 (A) and k0 (A) generate
the same ideal of Q(R+ , B). Since B ∩Q(R+ , J) = J and B ∩I0 = B, we get B = J.
Thus, h(A ⊗ O2 ) is a stable C *-subalgebra of B and the closed ideal generated
by h(A ⊗ O2 ) is the σ-unital stable C *-algebra B. By Corollary 5.5.6(iv), h is
unitarily homotopic to a non-degenerate *-monomorphism h0 from A to B. The
4Use normalizations of suitable c.p. maps Q(R , B) → ` (B ∗∗ )/c (B ∗∗ ) → B ∗∗ .
+ ∞ 0
926 9. SCALE-INVARIANT MAPS (I+II OF THM’S B,M)

non-degenerate *-monomorphism h00 is again unitary equivalent to k0 by a unitary

in Q(R+ , M(B)). The image of h00 commutes element-wise with the unital copy
M(h0 )(1M(A) ⊗ O2 ) of O2 , if M(h0 ) is the unital extension of h0 to the multiplier
algebras. 

We can rephrase Corollary 9.1.6 in the language of actions of Prim(B) on A as

follows:

Corollary 9.1.7. Suppose that A and B are stable separable C*-algebras,

and that Ψ : I(B) → I(A) is a lower semi-continuous action of Prim(B) on A,
such that Ψ(0) = 0 and Ψ−1 (A) = {B}.
If k : A ⊗ O2 → Q(R+ , B) is a nuclear *-monomorphism, such that, for every
closed ideal I of Q(R+ , B) and for k0 := k((·) ⊗ 1),

k0 (Ψ(I ∩ B)) = k0 (A) ∩ I,

then there exists a non-degenerate nuclear C*-morphism h : A ⊗ O2 → B, such that

h0 (Ψ(J)) = h0 (A) ∩ J for every closed ideal J of B, where h0 (a) := h(a ⊗ 1).
The morphism h0 is unitarily equivalent to h0 ⊕ h0 . Every nuclear *-
monomorphism h1 : A → B which satisfies h1 (Ψ(J)) = h1 (A) ∩ J for J ∈ I(B),
and is unitarily homotopic to h1 ⊕ h1 , is unitarily homotopic to h0 .

Proof. Since the *-monomorphism k : A ⊗ O2 → Q(R+ , B) is nuclear, the

algebra A is exact.
By assumption, the intersection of the image of k0 with a closed ideal of
Q(R+ , B) depends only from the intersection of the ideal with B.
The closed ideal I0 of Q(R+ , B) which is generated by the image of k0 := k((·)⊗
1) satisfies k0 (Ψ(B ∩ I)) = k0 (A). Since Ψ−1 (A) = {B}, we have B ∩ I = B. Thus
Corollary 9.1.6 applies to k. It gives the existence of the desired non-degenerated
nuclear h with h0 unitarily equivalent to h0 ⊕ h0 by a unitary in M(B).
Since h0 is unitarily homotopic to k0 , for a ∈ A and closed ideals I of Q(R+ , B),
h0 (a) ∈ I if and only if k0 (a) ∈ I. This applies to the ideals I := Q(R+ , J), where
J ∈ I(B). We have J = I ∩ B. Thus, h0 (Ψ(J)) = h0 (A) ∩ J for J ∈ I(B).
If h1 : A → B is a nuclear *-monomorphism, such that h1 is unitarily homotopic
to h1 ⊕ h1 , and such that, for J ∈ I(B),

h−1 −1
1 (h1 (A) ∩ J) = h0 (h0 (A) ∩ J),

then h0 and h1 are unitarily homotopic by Corollary 7.4.3. 

2. Mapping-cone construction defines isomorphism

Let D a stable separable C *-algebra. We start with any non-degenerate *-

monomorphism k : D⊗O2 → B , i.e., we suppose that D, k and C with the following
properties (a)–(c) are given:
 2. MAPPING-CONE CONSTRUCTION DEFINES ISOMORPHISM 927

(a) D is a separable and stable C *-algebra,

(b) k : D ⊗ O2 → B is a non-degenerate *-monomorphism, and
(c) C ⊆ CP(D, B) is the point-norm closed matrix operator-convex cone that
is generated by the *-monomorphism k0 := k((·) ⊗ 1) ∈ Hom(A, B).

The p.i. algebra k(A ⊗ O2 ) is a non-degenerate C *-subalgebra of B by (b), i.e.,

k(A ⊗ O2 )B is dense in B and k extends uniquely to a strictly continuous unital
monomorphism M(k) from M(A ⊗ O2 ) into M(B). It follows that

(d) B is σ-unital and stable, cf. Remark 5.1.1(9),

(e) the morphism k0 := k((·) ⊗ 1) is is unitarily equivalent to k0 ⊕ k0 by a
unitary in M(B) by Proposition 4.3.5(iii), because k0 (A) commutes with
the unital copy M(k)(1M(A) ⊗ O2 ) of O2 in M(B)), and
(f) k0 : D → B is non-degenerate.

We know from Corollary 7.4.19, that

SR(C; D, B) = S(k0 ; D, Q(R+ , M(B))) ,

and that a non-degenerate *-morphism k1 ∈ Hom(A, B) ∩ C is necessarily unitarily

homotopic to k0 if k1 generates C and if k1 is unitarily homotopic to k1 ⊕ k1 . In so
far k0 and C determine each other uniquely.
We recall some definitions and results of Chapters 5, 7 and 8:

Remark 9.2.1. For locally compact Y , e.g. if Y equals one of R, R+ = [0, ∞)

or R− := (−∞, 0], we have defined in Chapter 7 the asymptotic coronas Q(Y, B) :

Q(Y, B) := Cb (Y, B)/ C0 (Y, B).

Recall that there is a natural isomorphism

Q(R, B) ∼
= Q(R− , B) ⊕ Q(R+ , B),
and that Q(R, B) is an ideal in ER = Qs (SB), where we write

SB := C0 (R, B) and ER := M(SB)/SB ∼

= Qs (SB).
We denote the canonical epimorphism from M(SB) onto ER by πSB .
Thus Q(R+ , B) contains B naturally and Q(R+ , B) is naturally isomorphic to
an ideal of ER . We let J denote the closed ideal of ER that is naturally isomorphic
to Q(R+ , B) and let I1 : Q(R+ , B) → J denote the natural isomorphism from
Q(R+ , B) onto J ⊆ ER . We define a monomorphism h0 : D → J, by

h0 := I1 ◦ k0 . (2.1)

To make I1 and all related calculations of later identities explicit, we can take
the continuous function λ(t) := min(1, max(0, t)) (for t ∈ R) and build an element
fe(t) := λ(t)f (t) in fe ∈ Cb (R, B) by fe(t) := λ(t) · f (t) (t ∈ [0, ∞)) and fe(t) := 0 for
t < 0, where f ∈ Cb (R+ , B) is a given representative of an element f + C0 (R+ , B)
in Q(R+ , B). Then


I1 f + C0 (R+ , B) = fe + C0 (R, B) .
928 9. SCALE-INVARIANT MAPS (I+II OF THM’S B,M)

All notions of unitary equivalence in Hom(D, Q(R+ , B)) coincide with the
equivalence induced by unitaries of ER := M(SB)/SB ∼ = Qs (SB), or by unitaries
coming from Cb (R, M(B)), because B is stable, cf. Proposition 7.4.1. Therefore,
the inclusion map I1 : Q(R+ , B) ,→ ER = Qs (SB) defines a natural semigroup
monomorphism ψ from the semigroup [Hom(D, Q(R+ , B))] of classes of unitar-
ily equivalent morphisms in Hom(D, Q(R+ , B)) (by unitaries in Q(R+ , M(B)) ⊆
M(Q(R+ , B))) to [Hom(D, ER )] those of Hom(D, ER ). Note that the additive map
ψ generalizes the usual mapping cone construction in a natural way.
The natural inclusions and isomorphism

M(B) ⊆ Cb (R, M(B)) ⊆ Cb,st (R, M(B)) ∼

= M(C0 (R, B))

of Lemma 7.4.5 induce natural inclusions

M(B) ⊆ Cb (R, M(B))/ C0 (R, B) ⊆ ER .

We denote by I2 is the unital inclusion map of M(B) into ER .

We define a *-monomorphism

H00 := δ∞ k0 : D → M(B) .

Let H0 := HR := I2 ◦ H00 : D → ER . Then H0 for H00 is related to C := C(k0 ) in the

sense of Corollary 5.4.4.

Let βe be the *-automorphism of M(C0 (R, B)) which is defined by β(f

e )(t) =
f (−t) for f ∈ Cb,st (R, M(B)) ∼
= M(C0 (R, B)). βe maps C0 (R, B) onto C0 (R, B)
and defines therefore an automorphism β of ER with β 2 = id.
Certainly, β(b) = b for b ∈ I2 (M(B)) ⊆ ER , and β(Q(R+ , B)) = Q(R− , B).
Thus β(J) ∩ J = 0, βH0 = H0 and J + β(J) = Q(R, B) ⊆ ER . Furthermore
h0 + βh0 = I2 ◦ k0 , because I2 (b) = I1 (b) + βI1 (b) for b ∈ B.

Since H0 dominates h0 , the map ψ induces a semigroup homomorphism

ϑ : [h] ∈ S(h0 ; D, ER ) → [h] + [H0 ] ∈ G(H0 ; D, ER ) .

Then ϑ|G(h0 , D, ER ) is a group homomorphism and ϑ([h] + [h0 ]) = [h] + [H0 ] by

Proposition 4.4.2(ii).
We have seen in Chapters 5 and 8 that

Ext(C ⊗ CP(C, C0 (R)); D, SB) = G(H0 , D, ER )

(cf. Corollary 5.9.23) and

Ext(C ⊗ CP(C, C0 (R)); D, SB) ∼

= KK(C; D, B)

(cf. Lemma 8.4.2).

The semigroup S(h0 ; D, ER ) is a sub-semi-group of S(H0 ; D, ER ) because
H0 dominates h0 (almost obviously). Thus, [h] 7→ [h ⊕ H0 ] = [h] + [H0 ]
defines a natural group morphism ϑ from R(C; D, B) = G(h0 ; D, ER ) into
Ext(C ⊗ CP(C, C0 (R)); D, SB) = G(H0 ; D, ER ).
 2. MAPPING-CONE CONSTRUCTION DEFINES ISOMORPHISM 929

By Chapters 4 and 7, we have

G(k0 ; D, Q(R+ , B)) = R(C; D, B) ∼
= G(h0 ; D, ER ) = [h0 ] + S(h0 ; D, ER ) .

Proof of Prop. 9.2.4 and Cor. 9.2.5 requires to prove first

homotopy invariance of R(C ; D, B),
i.e., that
π0 , π1 : R(C[0, 1] ; D, B[0, 1]) → R(C ; D, B)
satisfy π0 = π1 .
Is the homotopy invariance of R(C ; D, B) equivalent to the to the
injectivity of
R(C ; D, B) → KK(C ; D, B) .
???
But there is also a a direct proof of the criteria in Theorem
4.4.6 for the injectivity of G(h0 ; D, ER ) → G(H0 ; D, ER ), given by
an inductive decomposition procedure (see below).

Lemma 9.2.2. Suppose that D and B are stable, D is separable, B is σ-unital,

H : D → M(B) is a non-degenerate C*-morphism, and a1 , a2 are self-adjoint gen-
erators of D, i.e., D = C ∗ (a1 , a2 ).
Then there exists exists a constant Ω(H) ∈ N with the following property:
For every unitary U ∈ M C0 ((−∞, 1], B)), – given by a strictly continu-
ous map U : (−∞, 1] 3 t 7→ U (t) ∈ U(B) – with γU,k ∈ C0 ((−∞, 1], B) for
γU,k (t) := H(ak )U (t) − U (t)H(ak ) and every δ > 0 there exist n ≤ Ω(D, B, H)
and contractions T1 , . . . , Tn ∈ M C0 ((−∞, 1], B) with Tj∗ = −Tj and a unitary

V ∈ 1 + C0 ((−∞, 1], B), such that [Tj , H(ak )] ∈ C0 ((−∞, 1], B), k[Tj , H(ak )]k < δ
for j = 1, . . . , n, k = 1, 2, and
U ⊕s1 ,s2 1 = V · exp(T1 ) · . . . · exp(Tn ) .
In particular,
k[V ∗ (U ⊕ 1), H(ak )]k ≤ n · δ ≤ Ω(D, B, H) · δ .

Proof. Is not very likely that a proof exist!!!

To be filled in ?? 

Next lemma describes a correction method for paths of unitaries by mirroring

the interval [α, β) onto (−∞, α] via the bijective map
[α, β) 3 t 7→ α − (t − α)/(β − t) ∈ (−∞, α] .

Lemma 9.2.3. Let D and B stable C*-algebras, D separable and B σ-unital,

H : D → M(B) a non-degenerate C*-morphism in general position (i.e., δ∞ ◦
H unitarily homotopic to H), 0 ≤ α < β real numbers, a1 , a2 ∈ D self-adjoint
contractions,
(−∞, β] 3 t 7→ U (t) ∈ U(M(B))
930 9. SCALE-INVARIANT MAPS (I+II OF THM’S B,M)

a strictly continuous map.

Let λ(t) := U (t)H(a) − H(a)U (t) in M(B) for t ∈ (−∞, β] and suppose that
λ ∈ C((−∞, β], B) and lim∞ kλ(t)k = 0.
Let n, m ∈ N with max(m, n) ≤ Ω(D, B, H), and 0 < δ ≤ ε. Suppose that the
following are given (e.g. coming from Lemma 9.2.2):

(i) S1 , . . . , Sm ∈ Cb,st ((−∞, α], M(B)) with Sj∗ = −Sj , kSj k ≤ 1, [Sj , ak ] ∈
C0 ((−∞, α], B), k[Sj , H(ak )]k < ε and

V1∗ (U |(−∞, α]) ∈ 1 + C0 (−∞, α], B) ,

for V1 := exp(S1 ) · . . . · exp(Sm ).

(ii) T1 , . . . , Tn ∈ Cb,st ((−∞, α], M(B)) with Tj∗ = −Tj , kTj k ≤ 1,
[Tj , H(ak )] ∈ C0 ((−∞, β], B), k[Tj , H(ak )]k < ε and

V2∗ U ∈ 1 + C0 (−∞, β], B) ,

for V2 := exp(T1 ) · . . . · exp(Tn ).

Then there exists a unitary in W ∈ 1 + C([α, β], B) with W (β) = 1, W (α) =

V2 (α)∗ V1 (α) and, for t ∈ [α, β], ?????

k[W (t), H(ak )]k ≤ k[V1 , H(ak )]k+k[(V2 |(−∞, α]), H(ak )]k ≤ nδ+mε ≤ 2εΩ(?????) .

Then W (α)∗ V2 (α)∗ U (α) = V1 (α)∗ U (α), V2 (α)W (α) = V1 (α), V2 (β)W (β) =
V2 (β).
Moreover V3 (t) := V1 (t) for t ∈ (−∞, α] and V3 (t) := V2 (t)W (t) for t ∈ (α, β]
is strictly continuous on (−∞, β], with V3∗ U ∈ 1 + C0 ((−∞, β], B),

k[V3 (t), H(ak )]k ≤ mε + 2nδ ≤ 2εΩ(D, B, H)

and k[V3 (β), H(ak )]k = k[V2 (β), H(ak )]k ≤ nδ ≤ δΩ(D, B, H).

Proof. To be filled in ?? 

Proposition 9.2.4. The semigroup homomorphism ψ defines a surjec-

tive group homomorphism ϑ : [h] + [h0 ] 7→ [h] + [H0 ] from G(h0 ; D, ER ) onto
G(H0 ; D, ER ), if H0 : D → ER := M(B[R])/B[R] is as above defined.
The epimorphism ϑ is an isomorphism from G(h0 ; D, ER ) onto G(H0 ; D, ER ),
if Y 7→ R(C[Y ] ; D, B[Y ]) (Y compact metric space) is homotopy invariant, in the
sense that the evaluation maps



πt : R C[Y × I] ; D, B[Y × I] → R C[Y ] ; D, B[Y ]

have the same images for t ∈ {0, 1} ⊂ I := [0, 1].

Recall that we denote by C[R] (or likewise CR , SC) respectively by the ma-
trix operator convex cone C ⊗ CP(C, C0 (R)), that is the tensor product of C and
CP(C, C0 (R)) inside the category of m.o.c. cones, and we urge the reader to re-
mind that the tensor product in the category of m.o.c. cones has not not much to
 2. MAPPING-CONE CONSTRUCTION DEFINES ISOMORPHISM 931

do with a completion of some tensorproduct in the ordinary sense, because of the

many different “canonical” matrix-cones on tensor products, compare Chapter 3.
Remember that B[R] (or likewise BR , SB) are other notations for the C *-
algebra C0 (R, B) ∼
= B ⊗ C0 (R). The above listed results of Chapters 5, 7 and 8
immediately yield the following.
It suffices to consider the case Y := point in Proposition 9.2.4 by replacing B
by B[Y ] = Cb (Y, B).

Corollary 9.2.5. The semigroup homomorphism ψ induces a natural group

epimorphism ϑ of R(C; D, B) onto Ext(C[R]; D, B[R]) ∼
= KK(C; D, B). The epi-
morphism ϑ is an isomorphism if the evaluation maps

π0 , π1 : R(C[0, 1]; D, B[0, 1]) → R(C; D, B)

have the same image. 

We need several elementary observations for the proofs of Proposition 9.2.4 and
Corollary 9.2.5, e.g.

Lemma 9.2.6. Suppose that D is a stable and σ-unital C*-algebra and that k
is a non-degenerate *-monomorphism from D ⊗ O2 into B.
rn (rn )∗ converges
P
Let r1 , r2 , . . . a sequence of isometries in M(B) such that
strictly to 1, and let δ∞ := rn (·)(rn )∗ the corresponding infinite repeat endomor-
P

phism of M(B). Then:

(i) There exists a unital *-morphism  : M(O2 ⊗ K) ,→ M(B) such that

(0)  is a strictly continuous monomorphism,
(1) (M(O2 ⊗ K)) commutes element-wise with δ∞ (k(D ⊗ 1)),
(2) (O2 ⊗ 1M(K) ) commutes element-wise with k(D ⊗ 1), and
(3) r1 r1∗ = (1O2 ⊗ p11 ) , and
(4) δ∞ (k(D ⊗ 1)) · (O2 ⊗ K) ⊆ B .
(ii) If  : M(O2 ⊗K) → M(B) is any unital *-morphism that has the properties
(0)–(4) of Part (i), then

d ⊗ f 7→ δ∞ (k(d ⊗ 1))(f ⊗ p11 )

extends to a *-monomorphism k1 : D ⊗ O2 → B, such that d ∈ D 7→

k1 (d ⊗ 1) is unitarily homotopic to d ∈ D 7→ k(d ⊗ 1).
(iii) (O2 ⊗K)B is dense in B, and there exists an isometry t00 ∈ (M(O2 ⊗K))
such that t00 t0∗ ∗
0 = 1 − r1 r1 .

Proof. (i): First we consider the case B = D ⊗ O2 and k := id.

Let λ : K ⊗ K → K be an isomorphism. Then there exist sequences (In ),
In In∗ and sn s∗n strictly converge to 1,
P P
(sn ) of isometries in M(K), such that
λ(a ⊗ pnn ) = In aIn∗ and λ(pnn ⊗ a) = sn as∗n for a ∈ K.
In particular M(λ)(1 ⊗ p11 ) = In In∗ .
932 9. SCALE-INVARIANT MAPS (I+II OF THM’S B,M)

Then the infinite repeats δ (1) := In (·)In∗ and δ (2) := sn (·)s∗n have element-
P P

wise commuting images and δ (1) (a)δ (2) (b) = λ(a ⊗ b) for a, b ∈ K. Indeed, δ (1) =
M(λ)((·) ⊗ 1) and δ (2) = M(λ)(1 ⊗ (·)).
Let rn0 := 1 ⊗ In ⊗ 1, s0n := 1 ⊗ sn ⊗ 1, and 0 (f ) :=
P 0
sn (1 ⊗ M(κ)(f ))s0n
for f ∈ M(O2 ⊗ K), where κ denotes here the flip-isomorphism from O2 ⊗ K onto
K ⊗ O2 . Then 0 has the quoted properties with respect to r10 , r20 , . . . and id (in
place of , r1 , r2 , . . . , k):
For a ∈ A and b ∈ K, where D := A ⊗ K
Is A sufficiently well-defined? ??
let ??????
δ∞ (a ⊗ b ⊗ 1) = a ⊗ δ (1) (b) ⊗ 1,
and, for e ∈ K and f ∈ O2 ,

0 (e ⊗ f ) = 1 ⊗ δ (2) (e) ⊗ f.

In particular,

0 (1 ⊗ p11 ) = 1 ⊗ M(λ)(1 ⊗ p11 ) ⊗ 1 = r10 (r10 )∗ .

D ⊗ 1 and 0 (O2 ⊗ 1) commute, because, for a ∈ A, b, c ∈ K and f ∈ O2 ,

(a ⊗ λ(b ⊗ c)) · 0 (f ⊗ 1) = a ⊗ λ(b ⊗ c) ⊗ f .

In the general case, let M(k) : M(D ⊗ O2 ) ,→ M(B) the strictly continuous unital
extension of k. By Lemma 5.1.2(i) there is a unitary u ∈ M(B) such that rn =
uM(k)(rn0 ). Let  := uM(k)(0 (·))u∗ , then  satisfies (0)-(4).
(ii): By the properties listed in (i), there is a C *-morphism g from the algebraic
tensor product D O2 into B such that, for d ∈ D, f ∈ O2 ,

g(d ⊗ f ) = δ∞ (k(d ⊗ 1))(f ⊗ p11 ) ,

moreover g extends to a C *-morphism k1 from D⊗O2 into B, because O2 is nuclear

and simple.
Let k2 := k1 ((·) ⊗ 1) and k0 := k((·) ⊗ 1). Since r1 (r1 )∗ = (1 ⊗ p11 ),
k2 = r1 k0 (·)r1∗ . Thus k0 = r1∗ k2 (·)r1 . By Corollary 7.4.4, k2 and k0 are unitar-
ily homotopic, because B is stable, D is separable and r1 ∈ M(B).
(iii): (O2 ⊗ K)B is dense in B, because  is strictly continuous and unital.
By (i), (1 ⊗ (1 − p11 )) = 1 − r1 r1∗ .
Let T0 an isometry in M(K) ∼ = L(H) with T0 (T0 )∗ = 1 − p11 , and let t00 :=
(1 ⊗ T0 ). Then t0 is an isometry in (M(O2 ⊗ K)) such that t00 t0∗
0 ∗
0 = 1 − r1 r1 . 

Proof of Proposition 9.2.4. We use the notations and observations of Re-

mark 9.2.1 and reduce the prove to Theorem 4.4.6. It means, that we are going to
find F ⊆ H0 (D)0 ∩ER and elements s, t, p0 , s0 , t0 , u1 in ER , such that the above de-
fined D, ER , J, h0 , H0 , β, together with with the below defined F , s, t, p0 , s0 , t0 and
u1 satisfy the assumptions (i)–(vi) of Theorem 4.4.6. Then Theorem 4.4.6 implies
 2. MAPPING-CONE CONSTRUCTION DEFINES ISOMORPHISM 933

that the generalized mapping cone construction ψ : S(h0 ; E, ER ) → G(H0 ; D, ER )

defines a group epimorphism ϑ from G(h0 ; D, ER ) onto G(H0 ; D, ER ). It becomes
an isomorphism if we later can show in addition that that Γ(H0 (D), J, E) = 0.
We are now going to define the missing ingredients for this list. Parallel, we
check that the conditions (i)-(vi) of Theorem 4.4.6 are fulfilled for them:
rn (rn )∗ = 1,
P
Let r1 , r2 , . . . be a sequence of isometries in M(B), such that
and let  : M(O2 ⊗K) → M(B) the strictly continuous unital *-monomorphism that
satisfies (0)-(4) of Lemma 9.2.6(i) with respect to k : D ⊗ O2 → B and (r1 , r2 , . . .).
By Lemma 9.2.6(iii), there is an isometry t00 in (M(O2 ⊗K)), such that t00 (t00 ) =
1 − r1 (r1 )∗ . Thus t00 commutes element-wise with H00 (D).
Therefore δ∞ (k0 (D)) and k0 (D) both commute element-wise with the unital
copy η(O2 ) := (O2 ⊗ 1) of O2 in M(B), where

η(·) := ((·) ⊗ 1M(K) ) .

Let s1 , s2 generators of O2 and let s := I2 (η(s1 )), t := I2 (η(s2 )). Then both are
isometries in I2 (M(B)) ⊆ ER and generate a unital copy of O2 . Since β fixes the
elements of I2 (M(B)), we have β(s) = s and β(t) = t.
By Lemma 9.2.6(i), η(O2 ) ⊆ H00 (D)0 ∩ k0 (D)0 . If we apply I2 , we get that s
and t are in H0 (D)0 ∩ (h0 + βh0 )(D)0 ∩ ER .
Recall that k0 : D → B is non-degenerate.
By Lemma 7.4.17, there exists a contraction g1 ∈ Cb (R, (1 ⊗ K))+ such that
g1 (x) = 0 for x ≤ 0 and limx→+∞ g1 (x)k0 (a) = k0 (a) for every a ∈ D, because
k0 (D) is separable. Let g := g1 + SB ∈ ER . Then (h0 + βh0 )(a)g = h0 (a). The
positive contraction g commutes with s and t, because (1 ⊗ K) commutes with
(O2 ⊗ 1M(K) ). Thus s and t also commute element-wise with h0 (D).
Commutation of s, t with h0 (D)
follows also from commutation of s, t with
(h0 + βh0 )(D) and J ∩ βJ = 0. ??
Hence, s, t ∈ ER satisfy condition (i) of Theorem 4.4.6.

Our non-degenerate *-monomorphism H0 : D → ER satisfies Ext(C ⊗

CP(C, C0 (R)); D, SB) = G(H0 , D, ER ) by Corollary 5.9.22(i), as it was explained
above.
Therefore, by Lemma 8.4.3, for every homomorphism k : D → ER that is dom-
inated by H0 , there exists a unitary u ∈ ER such that u∗ (k ⊕ H0 )(a)u − H0 (a) ∈ J
for each a ∈ D.
Thus condition (ii) of Theorem 4.4.6 is valid.

Now we want to show that conditions (iii), (iv) and (v) of Theorem 4.4.6 are
satisfied:
For P := (O2 ⊗ K) ⊆ M(B) with  : O2 ⊗ K → M(B) we have that P ∼ = O2 ⊗ K,
and that P · B is dense in B (by Lemma 9.2.6). Moreover P ∩ B = {0} because
934 9. SCALE-INVARIANT MAPS (I+II OF THM’S B,M)

P is simple and P is not contained in B by condition (3) of Lemma 9.2.6(i). Let

F1 := Cb (R, P ) ⊆ Cb,st (R, M(B)) ∼= M(SB). Certainly β(F e 1 ) ⊆ F1 , and F1 ∩
C0 (R, B) = {0}, because (g(t)) = f (t) for g(t) ∈ O2 ⊗ K and because f (t) ∈ B
implies f (t) = 0. Therefore F := πSB (F1 ) ⊆ ER is a C *-subalgebra of ER that is
isomorphic to Cb (R, O2 ⊗ K) and satisfies β(F ) = F . Since H00 (D)P ⊆ B, we get
F1 H00 (D) ⊆ Cb (R, B) and F H0 (D) ⊆ J + βJ.
By Proposition 7.4.18, for every projection q ∈ F1 , and every element f1 ∈
Cb (R, B) there exists a projection p in F1 such that pf1 p − f1 ∈ C0 (R, B), q ≤ p
and q 6= p.
If we apply πSB , we get that condition (iii) of Theorem 4.4.6 is satisfied.
By Lemma 9.2.6(iii), there are isometries r1 , t00 in M(B) such that t00 (t00 )∗ =
1 − p00 , r1 (r1 )∗ = p00 ∈ P , r1∗ H00 (·)r1 = k0 , and t00 ∈ (M(O2 ⊗ K)) commutes with
the elements of H00 (D). If considered as elements of M(SB) they are fixed by β. e
0 0
Let s0 := I2 (r1 ), t0 := I2 (t0 ) and p0 := I2 (p0 ) the corresponding elements in ER .
Then s0 and t0 are fixed by β, p0 = s0 s∗0 ∈ F , and t0 commutes with the elements
of H0 (D) = I2 (H00 (D)). Since r1∗ H00 (·)r1 = k0 , we get s∗0 H0 (·)s0 = I2 k0 = h0 +βh0 .
Thus, condition (iv) of Theorem 4.4.6 is satisfied.
By [172] any two nonzero projections in O2 ⊗ K are unitarily equivalent by a
unitary in the *-semigroup 1 + O2 ⊗ K in M(O2 ⊗ K). Therefore, condition (v) of
Theorem 4.4.6 is satisfied by Lemma 7.4.13, because p0 is a nonzero projection of
F and F is isomorphic to Cb (R, O2 ⊗ K).

BEGIN (Old vi)

The old assumption (OLD vi) has been
proven in new Lemma 4.4.7
from the other assumptions (i)-(v) and (OLD vii).
Thus (OLD vi) is superfluous and Part (OLD vii)
is now the new Part (vi).
Check of condition (OLD vi) of Theorem 4.4.6:
Let h : D → J. Suppose that H0 dominates h + βh0 , i.e., there exists an isometry
T ∈ ER with T ∗ H0 (·)T = h + βh0 . By Lemma 7.4.22(i), there exists a contraction
y ∈ (J + β(J))+ = Q(R, B)+ such that y(h + βh0 )(·)y = h. It follows that
h = yT ∗ H0 (·)T y is dominated by H0 . Since z := T y is a contraction in the
ideal Q(R, B) of ER = M(C0 (R, B))/ C0 (R, B)) and, since

H0 (D) ⊆ πSB (M(B)) ⊆ πSB (Q(R, M(B))) ⊆ ER ,

by the natural embeddings

M(B) ⊆ Cb (R, M(B)) ⊆ Cb,st (R, M(B)) = M(C0 (R, B)) ,

there is a contraction w ∈ Cb (R, B) with w + C0 (R, B)) = z

h = w∗ (δ∞ ◦ k0 )(·)w + C0 (R, B) .

 2. MAPPING-CONE CONSTRUCTION DEFINES ISOMORPHISM 935

By 7.4.22(iv), there is a contraction v ∈ Cb (R, B) with v(t) = 0 for t < 0 and

v ∗ k0 (·)v + C0 (R, B) = h. Thus, d∗ h0 (·)d = d∗ (h0 + βh0 )(·)d = h for the contraction
d := v + C0 (R, B) ∈ J ∼ = Q(R+ , B). Since B is stable and h0 (D) is separable, there
is an isometry S ∈ Cb (R, M (B)) ⊆ M(C0 (R, B)) such that s := S + C0 (R, B) ∈ ER
satisfies s∗ h0 (·)s = d∗ h0 (·)d = h (cf. Lemma 7.4.22(v)). Thus assumption (OLD
vi) of Theorem 4.4.6 is satisfied.
END (OLD vi)

Check again assumption (vi) ?? ??

Now we verify assumption (vi) of Theorem 4.4.6:
First we define a continuous map ξ ∈ R → v(ξ) into the unitaries of O2 =
C ∗ (s1 , s2 ) for ξ ∈ [0, 1] by

v(ξ) := ξ(s1 s∗1 + s2 s∗2 ) + (1 − ξ 2 )1/2 (s2 s∗1 − s1 s∗2 ) ,

and by v(ξ) := 1 for ξ ≥ 1, v(ξ) := v(0) = s2 s∗1 − s1 s∗2 for ξ ≤ 0.

Let w(ξ) := η(v(ξ)), then w ∈ Cb (R, η(O2 )) is a unitary in Cb,st (R, M(B)) ∼
=
M(SB). Finally, let u1 := w + SB = πSB (w) . Then u1 ∈ J 0 ∩ ER , because every
a∈J ∼ = Q(R+ , B ⊗ K) has a representative b ∈ Cb (R, B) with b(ξ) = 0 for ξ ≤ 1,
and, for those b, wb = b = bw.
Since β fixes C ∗ (s, t) = I2 (η(O2 )) ⊆ I2 (M(B)) and s, t commutes element-wise
with h0 (D), we get that s and t commute element-wise with βh0 (D). The elements
of βJ can be represented by elements b in Cb (R, B) such that b(ξ) = 0 for ξ ≥ 0.
We get wb = η(s2 s∗1 − s1 s∗2 )b and bw = bη(s2 s∗1 − s1 s∗2 ). It implies that the equation

u1 βh0 (·)ss∗ = tt∗ βh0 (·)u1

is equivalent to

(ts∗ − st∗ )ss∗ βh0 (·) = βh0 (·)tt∗ (ts∗ − st∗ ) ,

i.e., is equivalent to the above shown element-wise commutation of ts∗ with βh0 (D).
Therefore condition (vi) of Theorem 4.4.6 is satisfied.
Now we have seen that all conditions of Theorem 4.4.6 are satisfied by D, ER ,
H0 , h0 , J and β. Therefore

ϑ : [k ⊕ h0 ] → [k ⊕ H0 ]

is a group epimorphism from G(h0 ; D, Qs (SB)) ∼

= R(C; D, B) onto

G(H0 ; D, Qs (SB)) = Ext(C ⊗ CP(C, C0 (R)); D, SB) ∼

= KK(C; D, B) .

The following Corollary 9.2.7 and its proof uses the notations and conventions
of Remark 9.2.1.
936 9. SCALE-INVARIANT MAPS (I+II OF THM’S B,M)

Corollary 9.2.7. Suppose that B and D are stable (and trivially graded), D is
separable, B σ-unital, and that h0 : D → B is non-degenerate with [h0 ⊕ h0 ] = [h0 ].
Let C := C(h0 ) and define Hom(C; D, B) := Hom(D, B) ∩ C . Then:

(i) S(h0 ; D, Q(R+ , M(B))) = SR(C; D, B).

If [k] ∈ S(h0 ; D, Q(R+ , M(B))) then h := k ⊕ h0 is in [h0 ] +
SR(C; D, B) ∼ = R(C; D, B), and [b σ ◦ h] = [h] in SR(C; D, B), for
every homeomorphism σ of R+ ,
i.e., h is invariant under scaling up to unitary equivalence.
(ii) The difference construction [h − 0] := [(B, h, 0)] ∈ KK(C; D, B) for h ∈
Hom(C; D, B) defines a semigroup epimorphism

α : [Hom(C; D, B)] → KK(C; D, B)

from the “constant” sub-semigroup [Hom(C ; D, B)] of R(C ; D, B) onto

KK(C; D, B).
Elements [h1 ], [h2 ] ∈ [Hom(C; D, B)] define the same element [h1 − 0] =
[h2 − 0] of KK(C; D, B), if and only if, h1 ⊕ h0 and h2 ⊕ h0 are unitarily
homotopic (cf. Definition 5.0.1).

Next example OK ???

More information ? ??
Note that, in “contrast” to Corollary 9.2.7, the natural group homomorphism
G(h0 ; D, B) → G(h0 ; D, Q(R+ , B)) is neither surjective nor injective in general,
e.g. for D = O2 ⊗ K, B = O2 ⊗ O2 ⊗ K and h0 (·) := 1 ⊗ (·).
Not clear, e.g.:
Then G(h0 ; D, Q(R+ , B)) contains only one element ?
What about this G(h0 ; D, B)?

Proof. The homotopy invariance of R(C; D, B) follows from different more

involved considerations ??
(i): Let ER := M(SB)/SB ∼ = Qs (SB) , let I1 : Q(R+ , B) ∼
= J ,→ ER be the
natural inclusion, and let I2 : M(B) → ER denotes the natural unital monomor-
phism induced by M(B) ⊆ Cb,st (R, M(B)) = M(SB).
By Proposition 7.4.21(i), a morphism k ∈ Hom(D, Q(R+ , B)) has unitary
equivalence class [I1 ◦ k] in S(h0 ; D, ER ), if an only if [k] ∈ SR(C; D, B). (The
unitary equivalence classes coincide, i.e., [k] = [k 0 ] in [Hom(D, Q(R+ , B))] if and
only if [I1 ◦ k] = [I1 ◦ k 0 ] in [Hom(D, ER )], cf. Lemma 7.4.20.)
The class [h] = [(I1 ◦ k) ⊕ h0 ] = [(I1 ◦ k)] + [h0 ] is in G(h0 ; D, ER ) = [h0 ] +
S(h0 ; D, ER ) ∼
= R(C; D, B), by Proposition 7.4.21(ii) (and by Proposition 4.4.3).
If σ is a topological isomorphism of R+ then σ(0) = 0 and therefore σ ex-
tends naturally to an orientation preserving topological isomorphism of R by let-
ting σ(t) = t for t ∈ R− . We denote this extension again by σ. It induces in a
natural way an automorphism σ b of ER = M(SB)/(SB) such that its restriction
 2. MAPPING-CONE CONSTRUCTION DEFINES ISOMORPHISM 937

to J ∼
= Q(R+ , B) is just the above considered induced automorphism of Q(R+ , B),
which we also denote by σb, because all is naturally related by the natural inclusion
and restriction maps.
The generalized mapping cone construction k 7→ I1 ◦ k satisfies σ
b ◦ (I1 ◦ k) =
I1 ◦ (b
σ ◦ k).
On the other hand the isomorphism ϑ of Proposition 9.2.4 satisfies, by definition
of ϑ,
σ ◦ (k ⊕ h0 )]) = [I1 ◦ (b
ϑ([b σ ◦ (k ⊕ h0 ))] + [H0 ] = [b
σ ◦ (I1 ◦ (k ⊕ h0 ))] + [H0 ] .
On the other hand σb ◦ H0 = H0 and σ b ◦ (H ⊕ H0 ) = (b
σ ◦ H) ⊕ H0 , because σ
b fixes
I2 (M(B)), H0 (D) is in I2 (M(B)) and ⊕ = ⊕s,t (up to unitary equivalence) with
canonical generators s, t of O2 in I2 (M(B)). Thus
σ ◦ (I1 ◦ (k ⊕ h0 ))] + [H0 ] = [b
[b σ ◦ ((I1 ◦ (k ⊕ h0 )) ⊕ H0 )] .
Since [I1 ◦ (k ⊕ h0 )] + [H0 ] ∈ G(H0 ; D, ER ) , we get from Corollary 8.3.4(ii) that
σ ◦ (I1 ◦ (k ⊕ h0 ) ⊕ H0 )] = [I1 ◦ (k ⊕ h0 ) ⊕ H0 ] = ϑ([k ⊕ h0 ]).
[b
Since [k ⊕ h0 ] is in R(C; D, B) and ϑ is faithful on R(C; D, B) by Proposition 9.2.4,
this shows the invariance of [k ⊕ h0 ] under scaling:
[k ⊕ h0 ] = [b
σ ◦ (k ⊕ h0 )].
By Corollary 7.4.19(i),
S(h0 ; D, Q(R+ , M(B))) = SR(C; D, B) .
The Definition of S(h0 ; D, Q(R+ , M(B))) yields that h = k ⊕h0 is scaling invariant
up to unitary equivalence (by unitaries in Q(R+ , M(B)). By Corollary 7.4.19(iii),
R(C; D, B) ∼
= SR(C; D, B) + [h0 ] ⊆ SR(C; D, B) .

(ii): Since h0 is unitarily equivalent to h0 ⊕ h0 we get from Lemma 7.4.24 that

Hom(D, B) ∩ C = {h ∈ Hom(A, B) ; [h] ∈ SR(C; D, B) } .

By part (i) and by Corollary 9.1.3, for every [h] ∈ G(h0 ; D, Q(R+ , M(B))) ∼ =
R(C; D, B), there exists a unitary u ∈ Q(R+ , M(B)) and a *-morphism k : D → B
such that u∗ k(·)u = h. It follows that k ∈ R(C; D, B) and [k ⊕ h0 ] = [k] + [h0 ] =
[h]+[h0 ] = [h]. Thus, k⊕h0 is unitarily homotopic to k, and k ∈ Hom(C; D, B). We
can replace k by k ⊕ h0 , and may assume that k is a monomorphism that dominates
h0 (in M(B) itself). It follows that k ∈ Hom(C; D, B) → [k] ∈ R(C; D, B) is
surjective.
Morphisms h1 , h2 ∈ Hom(C; D, B) define the same element [h1 ] = [h2 ] in
R(C; D, B) if and only if h1 ⊕ h0 and h2 ⊕ h0 are unitarily homotopic by defi-
nition of R(C; D, B) = G(h0 ; D, Q(R+ , M(B))).
It remains to check that the isomorphism R(C; D, B) ∼
= KK(C; D, B) given by
Proposition 9.2.4 and of Corollary 9.2.5 maps [h] ∈ R(C; D, B) to the difference
construction [h − 0] ∈ KK(C; D, B) if h ∈ Hom(C; D, B).
938 9. SCALE-INVARIANT MAPS (I+II OF THM’S B,M)

Let C(R) := C ⊗ CP(C, C0 (R)). By Corollary 5.9.23 and Corollary 8.3.3 there
are natural isomorphisms

Ext(C(R); D, SB) ∼
= G(H0 ; D, ER ) ,
and
Ext(C(R); D, SB) ∼
= KK(C; D, B)
that sends [H0 ] to zero and the mapping cones Ch ∈ SExt(C(R); A, SB) (cor-
responding to the Busby invariant [I1 ◦ h] ∈ S(H0 ; D, ER )) of a morphism h ∈
Hom(C; D, B) = Hom(D, B)∩C to the difference construction [h−0] = [(B, h, 0)] ∈
KK(C; D, B).
If we combine this isomorphism with the isomorphism [h] 7→ [h] ⊕ [H0 ] from
R(C; D, B) ∼= G(h0 ; D, Q(R+ , M(B))) onto G(H0 ; D, ER ) ∼
= Ext(C(R); D, SB),
then we get an epimorphism α from the Abelian semigroup [Hom(C; D, B)] ⊆
[Hom(D, B)] of unitary equivalence classes [h] of morphisms h ∈ Hom(C; D, B)
onto KK(C; D, B).
The construction of α shows (by inspection of the values of the composed maps)
that α([h]) = [(B, h, 0)] =: [h − 0] ∈ KK(C; D, B). Since the last two morphisms of
the compositions

[Hom(C; D, B)] → R(C; D, B) → Ext(C(R); D, B) → KK(C; D, B)

are isomorphisms, we get that [h1 − 0] = [h2 − 0] in KK(C; D, B) if and only if

h1 and h1 have the same class in G(h0 ; D, Q(R+ , M(B))). The latter means that
h1 ⊕ h0 and h2 ⊕ h0 are unitarily homotopic. 

3. Unsuspended stable E-theory versus Ext-groups

We consider the case where A is separable and stable, i.e., that D := A⊗K ∼
= A,
that B is σ-unital and stable and that C ⊆ CP(A, B) is a countably generated
point-norm closed m.o.c. cone. Moreover, we assume that C is “faithful” and “non-
degenerate” in the sense that if a ∈ A+ V (a) = 0 for all V ∈ C implies a = 0, and
that {V (a) ; V ∈ C, a ∈ A} generates B as a closed two sided ideal.
[With other words: There does not exist closed ideals I 6= {0} of A or J 6= B
of B such that V (I) = {0} or V (A) ⊆ J for all V ∈ C. We exclude this, because
otherwise we can in the below considerations A replace by A/I and B by J.]
Recall that in case where A is amenable and B is separable, the m.o.c. cone C
is the same as the m.o.c. cone of all ΨC -equivariant nuclear c.p.-maps from A to B,
where ΨC is the lower semi-continuous action of Prim(B) on A defined by C.
Respectively, C ⊆ CP(A, B) is the m.o.c. cone of all Ψ-equivariant c.p.-maps
for a given non-degenerate action Ψ : O(Prim(B)) → I(A) of Prim(B) on A that is
lower semi-continuous.
More specially we can take – in the situation where A is exact and the action
Ψ is lower s.c. and monotone upper s.c. – the m.o.c. cone C ⊆ CPnuc (A, B) of all
 3. UNSUSPENDED STABLE E-THEORY VERSUS Ext-GROUPS 939

Ψ-equivariant nuclear c.p.-maps V : A → B. The exactness of A yields that those

V are automatically Ψ-residual nuclear.
This equivalences show that working with an m.o.c. cone C covers all possible
cases of “ideal-equivariant” KK- and Ext-groups.
What is optimal notation?
h, H, H0 or H0 ? Mostly used?
Unify all notation?
What about h0 : A → B if C := C(h0 )?
Recall that there is a universal non-degenerate *-monomorphism H : A →
M(B) in “general position”, i.e., H is unitarily equivalent to its infinite repeat,
that is in 1-1-correspondence to C, cf. Chapters 3 and 5 Precise Refs ?? in
Chapters 3 and 5.
Let H1 := πB ◦ H : A → Q(B) := M(B)/B.
Basic observations:
Let A, B, C, H : A → M(B), H1 := πB ◦ H, as above.
Partly obvious generalizations of work of G. Kasparov, Connes/Higson and
J. Cuntz proves homotopy invariance of KK(C; A, ·), that

KK(C(1) ; A, B(1) ) ∼
= Ext(C ; A, B)

and that
KK(C; A, B) = Ext(C; A, SB) .
Moreover, if the m.o.c. cone C ⊂ CP(A, B) is countably generated and non-
degenerate, this results allow to give the following “reduction” to a description by
“ordinary” K∗ -theory (using also a result of Cuntz and Higson):

(1)


KK C(−∞, 1] ; A, C0 ((−∞, 1], B) = 0 .
(2)

KK(C; A, B) ∼
= kernel of K1 (H1 (A)0 ∩ Qs (B)) → K1 (Qs (B)) = K0 (B) .

(3) Qs (B) is K1 -bijective.

(4) Homotopy invariance of KK: Each evaluation map

B[0, 1] = C([0, 1], B) 3 f 7→ f (t) ∈ B

defines the same natural isomorphism

KK(C[0, 1] A, B[0, 1]) ∼

= KK(C ; A, B) .

(5) In particular, with CB := C0 ((0, 1], B),

KK(C(0, 1]; A, CB) = 0 .

We need the following natural isomorphisms and descriptions of KK(C ; A, B)

and and of representatives of its zero elements:
940 9. SCALE-INVARIANT MAPS (I+II OF THM’S B,M)

Proposition 9.3.1. Suppose that A is separable and stable, B is σ-unital and

stable, and if the m.o.c. cone C ⊆ CP(A, B) is non-degenerate and countably gen-
erated, and let then

KK(C ; A, B) ∼
= kernel of K1 (H(A)0 ∩ Q(B)) → K1 (Q(B)) ,

and
Ext(C ; A, B) ∼
= kernel of K0 (H(A)0 ∩ Q(B)) → K0 (Q(B)) .

(i) The kernel of K1 (H1 (A)0 ∩ Q(B)) → K1 (Q(B)) is the image of the map

u 7→ [πB (u)] ∈ K1 (H1 (A)0 ∩ Q(B))

where u ∈ M(B) is in the group of those unitary elements of M(B) that

satisfy uh(a) − h(a)u ∈ B for all a ∈ A.
(ii) The defining equivalence relation u1 ∼ u2 – given by [πB (u1 )] = [πB (u2 )]
in K1 (H(A)0 ∩ Q(B)) – can be equivalently expressed by the property that
there exist Tk ∈ M(B) (k = 1, . . . , n) with Tk∗ = −Tk and Tk h(a) −
h(a)Tk ∈ B, for a ∈ A, k ∈ {1, . . . , n}, such that

(u∗1 u2 ) ⊕s,t 1 = exp(iT1 ) · . . . · exp(iTn ) .

Here s, t ∈ h(A)0 ∩ M(B) are isometries with ss∗ + tt∗ = 1.

THIS formulation of Parts (i) and (ii) use the K1 -bijectivity

of Qs (B) for stable σ-unital B.
See Parts (c) and (iii) of Proposition 4.2.15.
It implies that all unitaries v ∈ Qs (B) with [v] = 0 in K1 (Qs (B)) are
products of exponentials. Thus, there is a unitary u ∈ M(B) with
π(u) = v if and only if [v] = 0 in K1 (Qs (B)).

Proof. The formulas for KK(C ; A, B) and Ext(C ; A, B)

have been mentioned/proven somewhere above!
(i): The unit 1 of the C *-algebra H1 (A)0 ∩ Q(B) is properly infinite.
Thus, each element z of K1 (H1 (A)0 ∩ Q(B)) has a representative [v] = z by a
unitary in v ∈ H(A)0 ∩ Q(B). and v ⊕ 1 defines the same class in K1 (H(A)0 ∩ Q(B))
as v.
If v = πB (u) ∈ H(A)0 ∩ Q(B) for some unitary in u ∈ M(B) then [v] = 0 in
K1 (Q(B)) because K1 (M(B)) = 0 by stability of B. Clearly, [u, h(a)] ∈ B for all
a ∈ A if and only if πB (u) ∈ H(A)0 ∩ Q(B).
Conversely, if v is unitary in Q(B) and [v] = 0 in K1 (Q(B)), then v ⊕ 1 ∈
U0 (Q(B)).
If [u ⊕ 1] = 0 in K1 (Q(B)), then u ⊕ 1 is a product of exponentials exp(X1 ) · . . . ·
exp(Xm ) with X`∗ = −X` and X` ∈ Q(B), cf. Lemma 4.2.6(v,2). And this implies
that there exists a unitary u ∈ M(B) with πB (u) = v ⊕ 1.
 3. UNSUSPENDED STABLE E-THEORY VERSUS Ext-GROUPS 941

(ii): If [u] = 0 in K1 (H(A)0 ∩Q(B)), then – moreover – there exists Y1 , . . . , Yn ∈

H(A)0 ∩ Q(B) with Yk∗ = −Yk such that u ⊕ 1 = exp(Y1 ) · . . . · exp(Yn ). See Lemma
4.2.6(v,2).
Given any finite subset M of A and ε > 0 then there exist Tk ∈ M(B) (k =
1, . . . , n) with πB (Tk ) = Yk and Tk∗ = −Tk and [Tk , h(a)] ∈ B for all a ∈ A and
k[Tk , h(a)]k < ε for all a ∈ M . The latter property can be obtained by lifting
the Yk to Sk ∈ M(B) with Sk∗ = −Sk and πB (Sk ) = Yk and then one defines
Tk := (1 − e)Sk (1 − e) for a suitable positive contraction e ∈ B+ in a approximate
unit of B that is quasi-central for h(A) ⊆ M(B). 

The following conjecture

Conjecture 9.3.2. The “logarithmic length” of the unitary πCB ((u∗1 u2 )⊕s,t 1)
with unitaries of u1 , u2 ∈ M(CB) such that [u1 , h(a)] − [u2 , h(a)] ∈ CB is bounded
by a universal constant, where above considered B is replaced here by CB :=


C0 (−∞, 1], B .

−1
One can correct this by passing to uk ⊕ 1 (k = 1, 2) in U(πB (H(A)0 ∩ Q(B))).
This is possible, because A is separable and stable, B is σ-unital and stable,
and h ≈u δ∞ h for h : A → M(B) with H = πB ◦ h, the map

u ⊕ 1 7→ πB (u ⊕ 1) ∈ U(H(A)0 ∩ Q(B)) → K1

is surjective map onto the kernel of

K1 (H(A)0 ∩ Q(B)) → K1 (Q(B)) .

Lemma 9.3.3. An unitary u ∈ M(B) represents the zero element of

KK(C ; A, B), if and only if, πB (u) represents the zero of K1 (H(A)0 ∩ Q(B)),
if and only if, there exist elements T1 , . . . , Tn ∈ M(B) such that Tj∗ = −Tj ,

exp(T1 ) · . . . · exp(Tn ) − (u ⊕ 1) ∈ B

and Tk h(a) − h(a)Tk ∈ B for all a ∈ A and k = 1, . . . , n.

Given ε > 0 and a finite subset M ⊂ A, then the Tk in this relations can be
chosen such that moreover kTk h(a) − h(a)Tk k < ε for a ∈ M .

Here is the same problem with the possibly missing K1 -injectivity of H(A)0 ∩
Q(B) if one hopes to replace u ⊕ 1 by u itself in this relations.
Is H(A)0 ∩ Q(B) K1 -injective if A or B is O∞ -absorbing?
Is it equivalent to the question about K1 -injectivity of all properly infinite unital
E?
The answer is positive if B is s.p.i. and H is nuclear.
Let B[0, 1] := C([0, 1], B) ∼
= B ⊗ C([0, 1]). Define C[0, 1] ⊆ CP(A, B[0, 1]) as
the set of c.p. maps V : A → B[0, 1] with V (·)(t) ∈ C for each t ∈ [0, 1]. The
942 9. SCALE-INVARIANT MAPS (I+II OF THM’S B,M)

corresponding realization of C[0, 1] is induced by the natural embedding η : b 7→

b ⊗ 1 ∈ B[0, 1] of B into B[0, 1] in the sense that M(η) ◦ h : A → M(B[0, 1]) is
the corresponding universal *-monomorphism from A into M(B[0, 1]) that defines
C[0, 1].
Thus, the kernel of

K1 ((πB[0,1] ◦ M(η) ◦ h)(A)0 ∩ Q(B[0, 1])) → K1 (Q(B[0, 1]))

is naturally isomorphic to KK(C[0, 1] ; A, B[0, 1]).

A natural generalization of Kasparov’s result of homotopy invariance of KK-
and Ext-groups is the following:
For (given fixed) t ∈ [0, 1], the evaluation map f ∈ B[0, 1] 7→ f (t) ∈ B defines a
natural isomorphism from KK(C[0, 1] ; A, B[0, 1]) onto KK(C ; A, B). (“Homotopy
invariance” of KK(C ; A, B).)
On the set of “representing” unitaries u ∈ M(B), respectively of {u(t)} =
U ∈ M(B[0, 1]) (with t 7→ u(t) ∈ M(B) strictly continuous), this isomorphism
corresponds to the evaluation map U = {u(t)} 7→ u(t) ∈ M(B).
Consequently:

Corollary 9.3.4. Suppose that A and B are stable, A separable and B σ-

unital, C ⊆ CP(A, B) a countably generated non-degenerate m.o.c. cone.
Let a0 ∈ A such that A = C ∗ (a0 ) and let h : A → M(B) a non-degenerate
C*-monomorphism that is unitary homotopic to is infinite repeat δ∞ ◦ h, satisfies
Vb := hh(·)b, bi ∈ C for all b ∈ B and that C is generated by {Vb ; b ∈ B}.
Let B[0, 1] := C([0, 1], B) and η : M(B) → Cb,st ([0, 1], M(B)) ∼
= M(B[0, 1])
the natural embedding.
If U ∈ M(B[0, 1]) is a representing unitary for an element of

KK(C[0, 1] ; A, B[0, 1]) ,

i.e., and if U (0) ⊕ 1 can be written modulo B as finite product of exp(Tk ), k =

1, . . . , n, with −Tk∗ = Tk ∈ M(B[0, 1]) and Tk h(a) − h(a)Tk ∈ B[0, 1] for a ∈ A,
k = 1, . . . , n, then U ⊕1 can be written modulo B[0, 1] as a finite product of exp(S` ),
` = 1, . . . , m, −S`∗ = S` ∈ M(B[0, 1]) and S` (M(η) ◦ h)(a) − (M(η) ◦ h)(a)S` ∈
B[0, 1] for a ∈ A, i.e.,

−S`∗ = S` ∈ Der(M(η)(h(A)), B[0, 1]) ⊆ M(B[0, 1])

and Commutator norms ????

?????????????????????????

Here M(η) : M(B) → M(B[0, 1]) is the natural inclusion map.

Proof. To be filled in ?? ?? 
 3. UNSUSPENDED STABLE E-THEORY VERSUS Ext-GROUPS 943

Corollary 9.3.5. Let A, B, C ⊆ CP(A, B) and h : A → M(B) as above,

M(η) : M(B) → Cb,st ((−∞, 1], M(B)) ∼
= M(C0 ((−∞, 1], B)) the natural
monomorphism.
Let u ∈ M(C0 ((−∞, 1], B)) a unitary given by the strictly continuous map
(−∞, 1] 3 t 7→ u(t) ∈ U(M(B)) from (−∞, 1] into the unitaries in M(B).
Suppose that [u(t), h(a)] ∈ B for all a ∈ A and t ∈ (−∞, 1] and that t 7→
[u(t), h(a)] is continuous, and

lim k[u(t), h(a)]k = 0 ∀ a ∈ A,

t→−∞

i.e., [u, h(A)] ⊆ C0 ((−∞, 1], B).

Then there exist S1 , . . . , Sn ∈ M(C0 ((−∞, 1], B)) with Sk∗ = −Sk , [sk , h(a)] ∈
B for all a ∈ A and
U = exp(S1 ) · . . . · exp(Sn ) · W
for some W ∈ U(C0 ((−∞, 1], B) + C · 1) with 1 − W ∈ C0 ((−∞, 1], B).
Such W is a product of exponentials W = exp(T1 ) · . . . · exp(T` ) with Tj =
−Tj∗ ∈ C0 ((−∞, 1], B).
In particular, W (t) ∈ U0 (B + C · 1) ∩ (1 + B) for each t ∈ (−∞, 1].

Question 9.3.6. Suppose that T1 , . . . , Tm ∈ Der(h(A), B) ⊆ M(B) and

S1 , . . . , Sn ∈ Der(h(A), B[0, 1]) ⊆ M(B[0, 1])

with Tj∗ = −Tj , Sk∗ = −Sk , V1 , V2 ∈ U0 (B + C · 1) ∩ (1 + B) and a ∈ A are given

with
exp(Tm ) · . . . · exp(T1 )V1 = W (0)V2 ,
for W (t) := exp(S1 (t)) · . . . · exp(Sn (t)).
Does there exist for each γ > 0 and ε ∈ (0, γ) a unitary

U ∈ U0 (B[0, 1] + C · 1) ∩ (1 + B[0, 1])

with U (0) = 1,

k[W (t)V2 U (t), h(a)]k ≤ γ + k[W (0)V2 , h(a)]k

for t ∈ [0, 1] and

k[W (1)V2 U (1), h(a)]k ≤ ε ?

Lemma 9.3.7. Suppose that M is a C*-algebra, B ⊆ M a closed ideal, X =

{a1 , . . . , an } ⊆ M a finite subset, and let

ν(a; S1 , . . . , Sn ) := exp(S1 ) · . . . · exp(Sn )a − a exp(S1 ) · . . . · exp(Sn )

for a, S1 , . . . , Sn ∈ M , with Sk = −Sk∗ .

Suppose that T1 , . . . , Tn ∈ M satisfy Tk = −Tk∗ and Tk a − aTk ∈ B for all
a ∈ X and k = 1, . . . , n.
944 9. SCALE-INVARIANT MAPS (I+II OF THM’S B,M)

Then for each ε > 0 there exists a positive contraction b = b(ε) ∈ B+ such that


ν a; (1 − b)T1 (1 − b), . . . , (1 − b)Tn (1 − b) ≤ ε ∀ a ∈ X .

But the QUESTION is ???:

Let B := C([0, 1], C), with C σ-unital and stable,
T1 , . . . , Tn ∈ M(B) self-adjoint contractions
with Tk h(a) − h(a)Tk ∈ B for all a ∈ A and k = 1, . . . , n, and, in
addition,

exp(T1 (0)) · . . . · exp(Tn (0)) = 1 .

Can we find the b ∈ B+ in this case such that b = {c(t)} satisfies
in addition the following inequality ?



k1 − exp i(1 − c(0))T1 (0)(1 − c(0)) · . . . · exp i(1 − c(0))T1 (0)(1 − c(0)) k < ε .

Proof. Without loss of generality, one can suppose that M is separable and
a 6= 0, because we can replace M and B in the proof by by C ∗ (X ∪ {T1 , . . . , Tn })
and B by B ∩C ∗ (X ∪{T1 , . . . , Tn }). Then B contains a strictly positive contraction
e ∈ B+ . If we define by functional calculus elements f (e) ∈ B+ with f ∈ C[0, 1]+ ,
kf k ≤ 1 and f ([0, δ]) = 0 for some δ ∈ (0, 1), then we can select from this family
of positive contractions a commutative approximate unit e1 , e2 , . . . of B that is
an approximately central sequence for M and satisfies em+1 em = em . It implies
(1 − em )(1 − em+k )` = (1 − em+k )` for k, ` ≥ 1.
Given ε > 0 we let γ := max(kT1 k, . . . , kTn k) and

δ := ε/n(1 + 2γ) · exp(γ) .

Since [a, Tk ] := aTk − Tk a ∈ B for a ∈ X, and since (em ) is an approximate unit of

B that is approximately central for the elements of M , there exists m0 ∈ N with
kem a − aem k < δ and k[a, Tk ](1 − em )k < δ for all m ≥ m0 and a ∈ X.
Recall that [a, x] := ax − xa satisfies the Leibnitz rule [a, xy] = [a, x]y + x[a, y].
We obtain with this rule that

k[a, (1 − em )Tk (1 − em )]k ≤ k[a, Tk ](1 − em )k + 2k[a, em ]kkTk k < δ · (1 + 2kTk k) .

It follows that

k[a, x1 · . . . · xn ]k ≤ γ n−1 (k[a, x1 ]k + . . . + ka, xn k)

for γ = max{kx1 k, . . . , kxn k}, and, for self-adjoint S, S1 , . . . , Sn ∈ Ms.a. that

n
X
ν(a; S1 , . . . , Sn ) ≤ k[a, exp(iSk )]k .
k=1

Moreover, k[a, exp(iS)]k ≤ k[a, S]k · (exp kSk).

 3. UNSUSPENDED STABLE E-THEORY VERSUS Ext-GROUPS 945

With γ := max{kS1 k, . . . , kSn k} and Sk ∈ Ms.a. , we get

n
X
ν(a; S1 , . . . , Sn ) ≤ exp(γ) · k[a, Sk ]k .
k=1

If Tk ∈ Ms.a. with Tk a − aTk ∈ B for all a ∈ X then we can find m0 ∈ N such

that k[a, (1 − em )Tk (1 − em )]k < δ · (1 + 2γ) for all m ≥ m0 . Thus, for m ≥ m0 ,

ν(a; (1 − em )T1 (1 − em ), . . . , (1 − em )Tn (1 − em )) < nδ · (1 + 2γ) · exp(γ) = ε .

We denote by ξ : M(B) → M(C0 (R, B)) the natural unital embedding given
by the strictly continuous extension of

B 3 b → 1 ⊗ b ∈ Cb (R, B) ⊆ M(C0 (R, B)) .

The next proposition is the base for the final proof the injectivity of

R(C; A, B) = G(h0 , A, E) → G(H0 , A, E) = KK(C; A, B) .

Proposition 9.3.8. Suppose that A and B are stable C*-algebra, A is separable

and B is σ-unital, and let h : A → M(B) a non-degenerate C*-morphism such that
δ∞ ◦ h is unitary equivalent to h. Denote by ξ : M(B) ,→ Cb,st (R, M(B)) the
natural inclusion.
If U ∈ M(C0 (R, B)) ∼
= Cb,st (R, M(B)) a unitary such that, for all t ∈ R,

U (t)h(a) − h(a)U (t) ∈ B

and
lim kU (t)h(a) − h(a)U (t)k = 0
t→−∞
for each a ∈ A, i.e.,

U · (ξ ◦ h)(a) − (ξ ◦ h)(a) · U ∈ J−∞

for J−∞ := {c ∈ Cb (R, B) ; lim t→−∞ c(t) = 0 }.

Then there exist unitaries V, W ∈ M(C0 (R, B)) such that U ⊕ 1 = V W ,

W (ξ ◦ h)(a) − (ξ ◦ h)(a)W ∈ C0 (R, B) ∀a ∈ A,

V ∈ 1 + Cb (R, B) and V (t) = 1 for t ≤ −1.

Proof. Let a0 ∈ A a contraction that generates A as C *-algebra.

First we define W0 ∈ M(SB) by W0 (t) := W (t) for t ≤ 0 and W0 (t) := W (−t)
for t > 0. Then W0 (t)h(a) − h(a)W0 (t) ∈ B for each t ∈ R and a ∈ A and
U0 (t) := W0 (t)∗ U (t) satisfies U0 (t) = 1 for t ≤ 0 and [U0 (t), h(a)] ∈ B for all t ≥ 0.
It follows that, for each (positive) n ∈ N, there exists a decomposition of Un :=
U0 |(−∞, 2n] of the form Un = Vn · Wn where Wn := exp(Tn,1 ) · . . . · exp(Tn,kn ) with
∗
−Tn,` = Tn,` ∈ M(C0 (−∞, 2n]) [Tn,` (t), h(a)] ∈ B, Vn (t) ∈ B + 1, limt→−1 Vn (t) =
1, k[Tn,` (t), h(a0 )]k < 4−n
946 9. SCALE-INVARIANT MAPS (I+II OF THM’S B,M)

Connect then by going back to −1 by both decompositions for Un and Un+1

and by shifting and rescaling the “mirrors” at −1.
It gives a path in [2n, 2n + 2] of elements in U(B + C · 1) that corrects Wn on
[2n, 2n + 1] in controlled way.


We could use the above mentioned results on intervals [n, n+1] (instead of [0, 1])
for a proof if we can find at the end-points a “fitting-together” procedure for the
above described approximate decomposition. Or if we can find similar approximate
decompositions with help of approximately central units in the ideal Cb ([n, ∞), B)
of M(C0 ([n, ∞), B)). (But suitable combined).
It has to do with the question if unsuspended m.o.c. cone-related E-
theory for strongly purely infinite separable C *-algebras is homotopy-invariant,
i.e., if for the Rørdam groups R(C; A, B) two asymptotic homomorphisms
h(0), h(1) : A → Q(R+ , B) are asymptotically stably unitary equivalent in
Cb (R+ , M(B))/ C0 (R+ , B) if they are homotopic, i.e., are boundaries of h : A →
Q(R+ , C([0, 1], B) with [h] ∈ R(C; A, B[0, 1]). One can directly prove this kind of
homotopy invariance for the Rørdam groups R(C; A, B).
We discuss now the natural relation of this homotopy invariance and the de-
composition condition (DC) of Theorem 4.4.6.
Suppose that A and B are stable, A is separable and B is σ-unital. Let h0 : A →
B a non-degenerate *-monomorphism such that h0 extends to a *-monomorphism
k0 : A ⊗ O2 → B with h0 = k0 ((·) ⊗ 1).
Let C ⊆ CP(A, B) the corresponding m.o.c. cone. H := δ∞ ◦h0 : A → M(B) de-
note the infinite repeat of h0 . Consider M(B) naturally as a unital C *-subalgebra of
constant elements of M(C0 (R, B)) ∼ = Cb,st (R, M(B)). Let consider also Cb (R, B)
as an ideal of M(C0 (R, B)).
There exists a *-monomorphism k1 : O2 ⊗K → H(A)0 ∩M(B) such that k1 (O2 ⊗
K)B = B, such that H(a)k1 (c) ∈ B for all a ∈ A and c ∈ k1 (O2 ⊗ K), and that
p0 := k1 (1 ⊗ p11 ) is a full in M(B) and 1 − p0 is a full and properly infinite
projection in H(A)0 ∩ M(B). Since p0 is properly infinite in k1 (O2 ⊗ K) with
[p0 ] = 0 ∈ K0 (k1 (O2 ⊗ K)), there exists an isometry in s0 ∈ M (B) with s0 s∗0 = p0
and an isometry t0 ∈ H(A)0 ∩ M(B) with t0 t∗0 = 1 − p0 .
Let F denote the image of Cb (R, O2 ⊗ K) ⊃ O2 ⊗ K. We consider F naturally
as a non-degenerate C *-subalgebra of Cb,st (R, M(B)) ∼ = M(C0 (R, B)) using the
*-monomorphism f 7→ {k1 (f (t))} ∈ Cb,st (R, M(B)). Then p0 ∈ F , 1 − p0 ∈ 1 + F .
It is easy to see that each non-zero projection p ∈ F is equivalent to p0 by a unitary
u ∈ 1 + F , because this is true for C([0, 1], O2 ⊗ K) (in place of F ).

Remarks 9.3.9. For each positive element g ∈ Cb (R, B) there is a projection

p ∈ F with g − pg ∈ C0 (R, B), p ≥ p0 .
F ⊆ H(A)0 ∩ M(C0 (R, B)).
 3. UNSUSPENDED STABLE E-THEORY VERSUS Ext-GROUPS 947

If V = {V (t)} ∈ M(C0 (R, B)) = ∼ Cb,st (R, M(B)) is a unitary with

limt→−∞ kV (t)H(a) − H(a)V (t)k = 0 and V H(a) − H(a)V ∈ Cb (R, B) for
all a ∈ A (i.e., V (t)H(a) − H(a)V (t) ∈ B for all t ∈ R and a ∈ A). Then
W ∈ M(C0 (R, B)) with W (t) := V (t)∗ for t ≤ 0 and W (t) := V (−t)∗ for t ≥ 0 is
a unitary that satisfies W H(a) − H(a)W ∈ C0 (R, B) for all a ∈ A. If U := V W
then U (t) = 1 for t ≤ 0 and U H(a) − H(a)U ∈ Cb (R, B) for all a ∈ A.
Let U ∈ M(C0 (R, B)) a unitary with U (t) = 1 for t ≤ 1 and U H(a) − H(a)U ∈
Cb (R, B) for all a ∈ A. Define Ue (t, s) := U (t − s−1 + 1) for t ∈ R and s ∈ (0, 1]
and Ue (t, 0) := 1 for all t ∈ R. Then Ue is a unitary in

M(C0 (R × [0, 1], B)) ∼

= Cb,st (R × [0, 1], M(B))

with Ue (t, s) = 1 if t ≤ 1 and Ue H(a) − H(a)Ue ∈ Cb (R × [0, 1], B) for all a ∈ A.

Notice Cb (R, C([0, 1], B)) ∼
= Cb (R × [0, 1], B) . Same for C0 in place of Cb , and
for (Cb,st , M(C([0, 1], B)), M(B)) in place of (Cb , C([0, 1], B), B).
For each s ∈ [0, 1] define the unitary Us (t) := Ue (t, s). The Us are unitaries in
M(C0 (R, B)) that satisfy U0 = 1 and Us (t) = 1 for t ≤ s−1 − 1 if s ∈ (0, 1]. In
particular, Us (t) = 1 for all t ≤ 0 and s ∈ [0, 1].
There exists a projection p ∈ F such that

(1 − p)(Ue H(a) − H(a)Ue ) ∈ C0 (R × [0, 1], B) for all a ∈ A .

Let v ∈ 1 + F a unitary with vpv ∗ = p0 . Then

(1 − p0 )(vUs H(a) − H(a)vUs ) ∈ C0 (R, B)

for all a ∈ A.
The projections (1 − p0 ) and p0 commute with H(a) and commute modulo
C0 (R, B) with (vUs )H(a)(vUs )∗ for each a ∈ A. We define for s ∈ [0, 1] the
completely positive map Ts : A → Cb (R, B) by

Ts (a) := p0 (vUs )H(a)(vUs )∗ p0 for a ∈ A

Then
Ts (a) − p0 (vUs )H(a)(vUs )∗ ∈ C0 (R, B)
and
Ts (a∗ a) − Ts (a)∗ Ts (a) ∈ C0 (R, B) .

Recall that Ts (a)(t) = p0 v(t)U (s, t)H(a)(v(t)U (s, t))∗ p0 for (t, s) ∈ (−∞, ∞) ×
[0, 1]. In particular, T0 (a) = p0 H(a), because U (0, t) = 1 and v(t) commutes with
H(a). For t ∈ R holds T1 (a)(t) = p0 v(t)U (t)H(a)U (t)∗ v(t)∗ p0 where U := {U (t)}
is the above considered unitary U ∈ M(C0 (R, B)) with U (t) = 1 for t ≤ 0.
The restriction to (t, s) ∈ [0, ∞)×[0, 1] of Ts defines a homotopy in SR(C; A, B)
from T0 = p0 H(·) to T1 (·)|[0, ∞).
It holds Ts (a)(t) = p0 H(a) for all t ≤ 0 and a ∈ A, and a ∈ A 7→ Ts (a) ∈
Cb (R, B) is a c.p. contraction, with Ts (a∗ a) − Ts (a)∗ Ts (a) ∈ C0 (R, B) for a ∈ A.
948 9. SCALE-INVARIANT MAPS (I+II OF THM’S B,M)

Proof. to be filled in ?? 

Proposition 9.3.10. Let U = {U (t)} ∈ Cb,st (R, B) = M(C0 (R, B)) and v =
{v(t)} ∈ 1 + F ∈ H(A)0 ∩ M(C0 (R, B)) the above considered unitary operators.
Then (1−p0 )(vU )H(a)(vU )∗ −(1−p0 )H(a) ∈ C0 (R, A), p0 (vU )H(a)(vU )∗ p0 ∈
Cb (R, A) and p0 (vU )H(a)(vU )∗ (1 − p0 ) ∈ C0 (R, A) for a ∈ A.
Let k(a) := s∗0 p0 (vU )H(a)(vU )∗ p0 s0 and h0 (a) := s∗0 p0 H(a)s0 for a ∈ A.
Suppose that there exists a unitary V ∈ Cb,st (R, M(B)) such that

V ∗ h0 (a)V − k(a) ⊕s0 ,t0 h0 (a) ∈ C0 (R, B) for all a ∈ A .

Then U = U1 U2 for suitable unitary operators U1 , U2 ∈ M(C0 (R, B)) with

H(a)(U1 − 1), U2 H(a) − H(a)U2 ∈ C0 (R, B) for all a ∈ A .

Proof. We consider first the special case where the c.p. map k : A → Cb (R, B)
given by
k(a)(t) := s∗0 (v(t)U (t))H(a)(v(t)U (t))∗ s0
is itself unitary equivalent modulo C0 (R, B) to h0 , i.e., we suppose in this special
case that there exists a unitary W = {W (t)} ∈ Cb,st (R, M(B)) with

W ∗ h0 (a)W − k(a) ∈ C0 (R, B) for all a ∈ A .

By definition of k(·) and Cuntz addition ⊕s0 ,t0 ,

s0 k(a)s∗0 + (1 − p0 )H(a) = k(a) ⊕s0 ,t0 H(a) ,

where t0 is an isometry in H(A)0 ∩ M(C0 (R, B)) with t0 t∗0 = 1 − p0 , and

s0 k(a)s∗0 = p0 (vU )H(a)(vU )∗ p0 .

Since (1 − p0 )H(a) − (1 − p0 )(vU )H(a)(vU )∗ ∈ C0 (R, B) and

p0 vU H(a)(vU )∗ − p0 (vU )H(a)(vU )∗ p0 ∈ C0 (R, B) ,

we get that

k(a) ⊕s0 ,t0 H(a) − (vU )H(a)(vU )∗ ∈ C0 (R, B)

for all a ∈ A .

Let V1 := s0 W s∗0 + (1 − p0 ). Then

V1∗ (h0 (a) ⊕s0 ,t0 H(a))V1 − (k(a) ⊕s0 ,t0 H(a)) ∈ C0 (R, B) for all a ∈ A .

Since h0 (a) ⊕s0 ,t0 H(a) = H(a) for all a ∈ A, it follows that V1∗ H(a)V1 −
(vU )H(a)(vU )∗ ∈ C0 (R, B) for all a ∈ A. Thus, V2 := V1 vU commutes with
H(a) modulo C0 (R, B) for a ∈ A. Notice U = U1 U2 with U1 := v ∗ V1∗ v and
U2 := v ∗ V2 . Since v ∈ 1 + F ⊆ H(A)0 ∩ M(C0 (R, B)) we get that U2 satisfies
U2 H(a) − H(a)U2 ∈ C0 (R, B) . The operator V1 − 1 = s0 W s0 − p0 is contained
in p0 M(C0 (R, B))p0 . Since p0 H(a) = H(a)p0 ∈ B for a ∈ A it follows that
H(A)(V1 − 1) ⊆ Cb (R, B). Since v commutes with H(a) and Cb (R, B) is an ideal
of M(C0 (R, B)), we get that H(A)(U1 − 1) ⊆ Cb (R, B).
Now we reduce the general case to the above considered special case:
 3. UNSUSPENDED STABLE E-THEORY VERSUS Ext-GROUPS 949

Recall:
p0 := k1 (1 ⊗ p11 ) ∈ k1 (O2 ⊗ K) ⊃ F
k(a) := s∗0 p0 (vU )H(a)(vU )∗ p0 s0
h0 (a) := s∗0 p0 H(a)s0 for a ∈ A.
It exits unitary V ∈ Cb,st (R, M(B)) with

V ∗ h0 (a)V − k(a) ⊕s0 ,t0 h0 (a) ∈ C0 (R, B) for all a ∈ A .

Want to show:
There exists p1 ≥ p0 in and a unitary W ∈ p1 M(B)p1 + (1 − p1 ) such that
W H(a)W ∗ − (vU )H(a)(vU )∗ ∈ C0 (A, B).
By properties of F , – or directly by simple properties of O2 ⊗ K –, there
exist unitary w1 , w2 ∈ 1 + F such that w1∗ p0 w1 ≤ 1 − p0 and w2∗ p1 w2 = p0 for
p1 := p0 + w1∗ p0 w1 . Notice

w1 (p1 − p0 )w1∗ = p0 ≤ w1 (1 − p0 )w1∗ .

For a ∈ A it follows that

p1 (vU )H(a)(vU )∗ − (s0 k(a)s∗0 + (p1 − p0 )H(a)) = p0 (vU )H(a)(vU )∗ (1 − p0 )

is in C0 (R, A).
Let h1 (a) := t∗0 (p1 − p0 )H(a)t0 = t∗0 (p1 − p0 )t0 H(a) and z := t∗0 w1∗ s0 .
The mapping h1 is a C *-morphism because p1 −p0 ≤ (1−p0 ) = t0 t∗0 and p1 −p0
commutes with H(A). Then

s0 k(a)s∗0 + t0 h1 (a)t∗0 = s0 k(a)s∗0 + (p1 − p0 )H(a) .

We show below that h1 is unitary equivalent to h0 modulo C0 (R, B).

Let V1 ∈ M(C0 (R, B)) a unitary with V1∗ h0 (a)V1 − h1 (a) ∈ C0 (R, B) and let
V2 := s0 s∗0 + t0 V1 t∗0 . Then

(s0 k(a)s∗0 + t0 h1 (a)t∗0 ) − V2∗ (s0 k(a)s∗0 + t0 h0 (a)t∗0 )V2 ∈ C0 (R, B) .

On the other hand, there ?????

It follows that ?????
We have

p1 (vU )H(a)(vU )∗ − (s0 k(a)s∗0 + t0 h1 (a)t∗0 ) ∈ C0 (R, B)

Next ?????????????
In the same way we get that p1 H(a) = s0 h0 (a)s∗0 + t0 h1 (a)t∗0

(s0 h0 (a)s∗0 + t0 h0 (a)t∗0 ) − V3∗ (s0 h0 (a)s∗0 + t0 h1 (a)t∗0 )V3 ∈ C0 (R, B) ,

and ?????
950 9. SCALE-INVARIANT MAPS (I+II OF THM’S B,M)

We want to show that ?????

Recall h0 (a) = s∗0 p0 H(a)s0 .
We get from t0 t∗0 = 1 − p0 that

z ∗ z = s∗0 w1 (1 − p0 )w1∗ s0 = 1 ,

and
zz ∗ = t∗0 w1∗ s0 s∗0 w1 t0 = t∗0 w1∗ p0 w1 t0 == t∗0 (p1 − p0 )t0 .
Notice that w1 , p1 , p0 , t0 commute with H(a). Thus (w1 (p1 − p0 )t0 )(t∗0 (p1 − p0 )) =
w1 (p1 − p0 ) commutes with H(a). Since

w1 t0 h1 (a)t∗0 w1∗ = s∗0 w1 (p1 − p0 )H(a)w1∗ = w1 (p1 − p0 )w1∗ H(a)

it follows

z ∗ h1 (a)z = s∗0 w1 (p1 − p0 )w1∗ H(a)s0 = s∗0 p0 H(a)s0 = h0 (a) ,

and

zh0 (a)z ∗ = zs∗0 H(a)s0 z ∗ = t∗0 w1∗ p0 H(a)w1 t0 =

t∗0 w1∗ p0 w1 H(a)t0 = t∗0 (p1 − p0 )H(a)t0 = h1 (a) .

Since h1 dominates h0 and h0 dominates 0 modulo C0 (R, B). It follows that

h1 dominates zero modulo C0 (R, B), i.e.,

[πC0 (R,B) ◦ h1 ] + [0] = [πC0 (R,B) ◦ h1 ]

in M(C0 (R, B))/ C0 (R, B). It follows that there exists isometries S, T ∈
M(C0 (R, B)) with Sh1 (a) − h1 (a)S ∈ C0 (R, B), SS ∗ + T T ∗ = 1 and T ∗ h1 (a)T ∈
C0 (R, B).
???????? 

We extend an m.o.c. cone C ⊆ CP(A, B) naturally to C[Y ] ⊆ CP(A, B[Y ]) by

letting B[Y ] := C0 (Y, B) and V ∈ CP(A, B[Y ]) is an element of C[Y ] if and only
if there is a point-norm continuous map y ∈ Y 7→ Vy ∈ C with y 7→ kVy (a)k is in
C0 (Y ), such that V (a)(y) = Vy (a) for a ∈ A.

Corollary 9.3.11. Suppose that A and B are stable C*-algebras, A is sepa-

rable, B is σ-unital, the m.o.c. cone C ⊆ CP(A, B) is generated by h0 ∈ C, where
h0 (a) := k(a ⊗ 1) for some a non-degenerate *-monomorphism k : A ⊗ O2 → B.
Then R(C; A, B[Y ]) → KK(C; A, B[Y ]) is injective, if and only if, Y 7→
R(C; A, B[Y ]) is homotopy invariant on locally compact metric spaces Y , – in the
sense that the canonical evaluation group morphisms

R(C; A, B[Y × [0, 1]]) → R(C; A, B[Y ])

are isomorphisms for t ∈ [0, 1].

Proof. to be filled in ?? 
 4. PROOFS OF PARTS (I) AND (II) OF THEOREMS B AND M 951

4. Proofs of Parts (i) and (ii) of Theorems B and M

Now we carry out the program for the proof of Parts (i) and (ii) of Theorems
B and M that we have described in the beginning of this Chapter 9. First we
explain why Parts (i) and (ii) of Theorem B are a special cases of Parts (i) and
(ii) of Theorem M (from the point of view of our program), and why they are both
special cases of Corollary 9.2.7.

Remark 9.4.1. Let A a separable exact C *-algebra. we use in the remaining

part of this section the definition D := A ⊗ K. Note that, by Corollary 5.5.6, the
*-monomorphism a ∈ A 7→ a ⊗ p11 is is unitarily homotopic to an isomorphism
from A onto D, if A is stable.
For notational simplicity we assume that in the sequel B denotes a stable σ-
unital C *-algebra. Then the assumptions of Theorem B can be reformulated as
follows: The C*-algebra A is unital, separable and exact, and B is a stable C*-
algebra that contains a copy N of O2 ⊗ K such that N B is dense in B .
We modify the definition of the *-monomorphism h0 : D → N ⊆ B ⊗ K (given
in Chapter 1 after Theorem A):
By Theorem A, we find a unital *-monomorphism h1 : A ⊗ O2 → O2 . Then
we define a unital *-monomorphism k1 : D → O2 ⊗ K by k1 := k1u ⊗ idK , where
k1u (a) := h1 (a ⊗ 1).
Since B contains the non-degenerate copy N of O2 ⊗ K, the monomorphism k1
defines a unital nuclear *-monomorphism k0 from D into B by k0 := ϑk1 , where ϑ
is an isomorphism from O2 ⊗ K onto N .
k0 is nuclear and non-degenerate, and there is a non-degenerate *-monomor-
phism k from D ⊗ O2 into N ⊆ B such that k0 = k((·) ⊗ 1). Thus k0 and k0 ⊕ k0
are unitarily equivalent.
The *-monomorphism k can be defined by ϑ(h1 ⊗ idK )γ, where γ means here
the natural isomorphism from (A ⊗ K) ⊗ O2 onto (A ⊗ O2 ) ⊗ K.
Later it follows, as a corollary of Theorem B, that any copy of O2 ⊗K in B, that
generates a full hereditary subalgebra of B, could be used in place of our particular
fixed copy N of O2 ⊗ K in B. Also later the reader can deduce from Corollary 9.4.2
that the above defined k0 could be replaced in the sequel by any *-monomorphism
from D = A ⊗ K into O2 ⊗ K, because of R(D, O2 ⊗ K) ∼ = KK(A, O2 ) = 0 and of
the absorption result (iii) of Theorem B, proved in Chapter 7, cf. Corollary 7.4.16.

Suppose, now that N ⊆ B is a (non-simple) strongly purely infinite C *-sub-

algebra of B with N B dense in B, and that k : D ⊗ O2 → N is a non-degenerate
nuclear *-monomorphism. Let k0 := k((·) ⊗ 1) ∈ Homnuc (D, B). With the def-
initions and elementary observations of Chapter 1 in hand, the reader can see,
that X := Prim(N ) acts on B upper semi-continuously by ΨB where ΨB (J) is the
closed ideal of B generated by the closed ideal J of N . The T0 space X acts lower
semi-continuously on D by ΨD , and, moreover, the action ΨD is monotone upper
952 9. SCALE-INVARIANT MAPS (I+II OF THM’S B,M)

semi-continuous, i.e., satisfies also condition (ii) of Definition 1.2.6. Here ΨD (J)
for J ∈ I(N ) is defined by

ΨD (J) := k0−1 (k0 (D) ∩ J).

The later in Chapter 12 given proof of Theorem K shows that k0 : D → B (and

k : D ⊗ O2 → B) can be constructed up to unitary homotopy from the actions ΨD
and ΨB of X on D respectively B. In fact, by Corollary 7.4.23 holds CPrn (Ψ) =
C(k0 ) and

SR(X; D, B) = SR(CPnuc (Ψ); D, B) = S(k0 ; D, Q(R+ , M(B)) .

Let X := Prim(N ), and use the natural lattice isomorphism between the
open subsets of of X and the closed ideals of N , cf. Section 2. Then ΨD (J) :=
(k0 )−1 (k0 (D) ∩ J) and ΨB (J) := ΨN,B
0 (J) = span(BJB) are actions of X on D
and B, respectively. Here ΨD is lower semi-continuous and ΨB is upper semi-
continuous.
We define actions of I(N ) on ER = Q(SB) and on Q(R+ , B) = ∼ J by ΨE (I) :=
R

πSB (M(SB, SΨB (I))), ΨQ(R+ ,B) (I) := Q(R+ , ΨB (I)) = ΨER (I) ∩ J for I ∈ I(N ).
Def.of action OK? ??
From the exactness of D it follows that h is Ψ-residually nuclear, if and only
if, h is nuclear and is Ψ-residually equivariant (cf. Chapter 3).
Thus H0 is nuclear and has a lift H00 := δ∞ ◦k0 which maps D into the multiplier
algebra
M(N ) ⊆ M(B) ⊆ Cb,st (R, M(B)) = M(C0 (R, B))
of the purely infinite C *-algebra N ⊆ B, but H00 (D) intersects N exactly in zero.
Therefore, by Corollary 5.9.23,

Extnuc (X; D, SB) = G(H0 ; D, ER ).

If D is exact, we have moreover G(H0 , D, ER ) = [H0 ] + [Homnuc (X; D, ER )], but

we don’t need this (for the proofs in section 9.2).
In Chapters 5 and 8 we have seen that there are natural isomorphisms

Extnuc (X; D, SB) ∼

= KKnuc (X; D, B) .

Since h0 : D → ER is dominated by H0 , every element in S(h0 ; D, ER )

is the unitary equivalence class [h] with nuclear h and defines an element
of Extnuc (X; D, SB) by the unitary equivalence class [h ⊕ H0 ], cf. Corollary
5.9.22(??).
By Proposition 7.4.15, S(h0 ; D, ER ) is nothing else the semigroup of unitary
equivalence classes Ψ-residually nuclear C *-morphisms h from D into Q(R+ , B),
where ΨA (J) and ΨQ(R+ ,B) are defined as above.
In Chapter 7 we have seen that

R(X; D, B) = G(h0 ; D, ER ) = [h0 ] + [Homnuc (X; D, J)] .

 4. PROOFS OF PARTS (I) AND (II) OF THEOREMS B AND M 953

(If h0 resp. k exists.)

Corollary 9.4.2. If D is exact and k : D ⊗ O2 → B is a non-degenerate

nuclear *-morphism that induces the above considered action of X := Prim(B) on
D, then ψ induces an isomorphism α of R(X; D, B) onto Extnuc (X; D, SB) ∼ =
KKnuc (X; D, B).

Proof. to be filled in ?? 

Following Lemma uses above conventions and explanations in Remark 9.4.1.

Lemma 9.4.3. If A is exact and h0 is nuclear, let ΨA denote the action of

Prim(B) induced by h0 .
For every nuclear Ψ-residually equivariant C*-morphism k : D → Q(R+ , B),
i.e., for [k] ∈ SR(X; D, B), the morphism h := k ⊕ h0 defines an element [h] in
R(X; D, B) that is approximately scale-invariant.

Proof. Use Corollary 9.2.7 with C := C(h0 ) = Crn (X; D, B).

Give new proof based on Corollary 9.2.7 ??
Let ER := Qs (SB). By Proposition 7.4.15(??), ?? for separable stable and exact
D, k ∈ Hom(X; D, Q(R+ , B)) has unitary equivalence class [k] in S(h0 ; D, ER ), if
an only if k is nuclear.
[h] = [k ⊕ h0 ] = [k] + [h0 ] is in R(X; D, B), because R(X; D, B) = [h0 ] +
S(h0 ; D, ER ) by Proposition 7.4.15(??) and Proposition 4.4.3.
If σ is a topological isomorphism of R+ then σ(0) = 0 and therefore σ ex-
tends naturally to an orientation preserving topological isomorphism of R by letting
σ(t) = t for t ∈ R− . We denote this extension again by σ. It induces in a natural
way an automorphism σ b of Qs (SB) such that its restriction to J ∼
= Q(R+ , B) is just
the above considered induced automorphism of Q(R+ , B), which we also denote by
σ
b, because all is trivially related by restriction maps.
Let I1 : Q(R+ , B) ∼
= J ,→ Qs (SB) the natural inclusion. Then for the general-
ized mapping cone construction k 7→ I1 ◦k we have, obviously, σ
b ◦(I1 ◦k) = I1 ◦(b
σ ◦k).
On the other hand the isomorphism α of Proposition 9.2.4 satisfies, by definition
of α,
σ ◦ (k ⊕ h0 )]) = [I1 ◦ (b
α([b σ ◦ (k ⊕ h0 ))] = [b
σ ◦ (I1 ◦ (k ⊕ h0 ))].
By Corollary 8.3.4(ii), we have

σ ◦ (I1 ◦ (k ⊕ h0 ))] = [I1 ◦ (k ⊕ h0 )] = α([k ⊕ h0 ]).

[b

Since [k ⊕ h0 ] is in R(X; D, B) and α is faithful on R(X; D, B) by Proposition

9.2.4, this shows the invariance of [k ⊕ h0 ] under scaling:

[k ⊕ h0 ] = [b
σ ◦ (k ⊕ h0 )] .


954 9. SCALE-INVARIANT MAPS (I+II OF THM’S B,M)

Proof of (i) and (ii) of Theorems B and M.. Give new proof based
on Corollary 9.2.7!!! ??
In case of Theorem B we pass to the stabilizations D := A ⊗ K and replace B
by B ⊗ K.
In the case of Theorem M we assume the existence of h0 (i.e., we prove (i) and
(ii) modulo the result of Theorem K, see our above made conventions).
By Corollary 9.4.2, the natural map h 7→ I1 ◦ h from Homnuc (X; D, Q(R+ , B))
into Homnuc (X; D, Qs (SB)) defines an isomorphism
∼
α : R(X; D, B) → Extnuc (X; D, SB) ∼
= KKnuc (X; D, B).

By Corollary 8.3.3(iii), for a nuclear C *-morphism h from D to B the mapping

cone construction defines an element α[h] = [Ch ] of Extnuc (X; D, SB) which is
mapped to the element [h − 0] ∈ KKnuc (X; D, B) under the natural isomorphism
from Extnuc (X; D, SB) onto KKnuc (X; D, B).
Thus the relations considered in Theorems B(ii) and M(ii) are just the relations
which are induced by the semigroup homomorphism from the unitary equivalence
classes of elements of Homnuc (X; D, B) into R(X; D, B). I.e., if h and k are in
Homnuc (X; D, B) then [h − 0] = [k − 0] ∈ KKnuc (X; D, B) if and only if they
define the same element of R(X; D, B). But this means that there is a strongly
continuous map t 7→ U (t) from R+ into the unitary group of M(B) such that
h ⊕ h0 = limt→∞ U (t)∗ (k ⊕ h0 )U (t). This proves the second part of Theorems B
and M.
To complete the proof of the first part of Theorem B, we have to show that
every element of R(X; D, B) can be represented by a Ψ-residually nuclear C *-
morphism h : D → B. But this follows from Lemma 9.4.3 and Corollary 9.1.3,
because R(X; D, B) = SR(X; D, B) + [h0 ] ⊆ SR(X; D, B) and [h0 ] is the neutral
element of R(X; D, B), cf. Chapter 4. 

Remark 9.4.4. Let us shortly mention alternative approaches as e.g. by

N. C. Phillips [627] in the case where X is a point:
If one finds another way to show the homotopy invariance of R(A, B) with respect
to the second variable B, then one can directly apply Corollary 9.1.3. And one gets
that the representatives of R(A, B) can be chosen as nuclear C *-morphisms from
A ⊗ K into B ⊗ K. Then it follows that there is a composition that replaces the
Kasparov products. But then it still remains to show that the natural morphism
from R(A, B) to KKnuc (A, B) ∼ = Enuc (A, B) is both faithful and surjective ( 5 ). The
zero functor satisfies all requirements of category theory descriptions of Kasparov
or E-theory type functors, except of the universality. A less trivial example, which
is also a two-sided ideal of KK-theory (and even of E-theory) is the subgroup of the
elements of KKnuc (A, B) that factorize through a commutative C *-algebra (it is a
5In principle, it could happen that h is – up to asymptotic unitary homotopy – the only
0
element of R(A, B).
 4. PROOFS OF PARTS (I) AND (II) OF THEOREMS B AND M 955

group by Bott periodicity). Therefore, formal category theoretic (algebraic) argu-

ments that prove the coincidence of R(A, B) and KKnuc (A, B) without assuming
UCT for A or B require some care.

The *-epimorphism b 7→ b(t), for b ∈ C([0, 1], B), extends naturally to a *-

epimorphism πt : Cb (R+ × [0, 1], B) → Cb (R+ , B) by
πt ((bs )s∈R+ ) := (bs (t))s∈R+ .
The πt define evaluation semi-group morphisms
(πt )∗ : SR(C[0, 1]; A, B[0, 1]) → SR(C; A, B)
that naturally define group morphisms
(πt )∗ : R(C[0, 1]; A, B[0, 1]) → R(C; A, B) .
 CHAPTER 10

Unitarily homotopic algebras and Theorem B(iv)

We give in this chapter the proof of part (iv) of Theorem B and obtain related
results. This finishes the proof of Theorem B, because parts (iii), (i)+(ii) have been
shown in Chapters 7 and 9 respectively.
We consider here also non-simple σ-unital C *-algebras.

1. Unitary homotopies and Proof of Theorem B(iv)

We recall that two morphisms ϕ, ψ : A → B are unitarily homotopic if there

is a norm-continuous map t ∈ R+ 7→ U (t) ∈ M(B) into the unitary operators in
the multiplier algebra M(B) of B such that ϕ(a) = limt→∞ U (t)∗ ψ(a)U (t) for all
a ∈ A, cf. Definition 5.0.1 ( 1 ).
We say that *-homomorphism ϕ : A → B defines a unitary homotopy be-
tween A and B, if there exists a *-homomorphism ψ : B → A such that ψ ◦ ϕ is
unitarily homotopic to idA and ϕ ◦ ψ is unitarily homotopic to idB .
A simple calculation shows that unitarily homotopic morphisms have the same
kernel. Therefore, a unitary homotopy ϕ : A → B always is a monomorphism.
If ϕ : A → B is a unitary homotopy and ψ : B → A is a *-homomorphism as in
the above definition, then ψ defines again a unitary homotopy between A and B.
In the following we denote by BXB the closed ideal of B which is generated
by a subset X of the multiplier algebra M(B) of B, i.e., BXB means the closure
of the linear span of BXB.
A unitary homotopy ϕ : A → B always satisfies that ϕ(A) is not contained in
a non-trivial closed ideal of B, i.e. Bϕ(A)B = B:

Proof. J := Bϕ(A)B is a closed ideal of B and ϕψ(B) ⊂ ϕ(A) ⊂ J. Since ϕψ

is unitarily homotopic to idB , this implies that every element of B is approximately
unitarily equivalent to an element of the ideal J. Thus B = J. 

1It suffices to suppose that t 7→ U (t) ∈ is strongly continuous. Then we say that ϕ and

ψ are “weakly unitarily homotopic”. Since the unitary operators are a topological group with
strict topology (and strong and strict topology coincides on the unitaries), the “weakly unitary
homotopy” of maps ϕ and ψ is an equivalence relation. If A and B are σ-unital and B is stable
then weakly unitarily homotopic ϕ and ψ are unitarily homotopic (by an other norm-contiuous
path t 7→ U
e (t)), cf. Corollary 7.1.5.

957
958 10. UNITARILY HOMOTOPIC ALGEBRAS AND THEOREM B(IV)

A similar argument shows that B and ϕ must be unital (resp. B is σ-unital) if

A is unital (resp. is σ-unital) and ϕ : A → B is a unitary homotopy.

In order to prove Theorem B(iv), we use the following observations (I)-(III):

(I) Suppose that A and B are σ-unital and stable, and h : A → B is a

monomorphism, such that Bh(A)B = B.
Then h is unitarily homotopic to a monomorphism k : A → B with
k(A)B = B ( 2 ).
(II) If h1 : A → B and h2 : B → C are unitarily homotopic to k1 : A → B
respectively k2 : B → C, and if k2 (B)C is dense in C, then h2 ◦ h1 is
unitarily homotopic to k2 ◦ k1 .
(III) Suppose that A and B are separable, that ϕ : A → B defines an unitary
homotopy of A and B, and that ϕ(A) contains a strictly positive element
of B. Then ϕ is unitarily homotopic to an isomorphism χ from A onto
B.

Observation (I) is a special case of Corollary 7.4.6. The observations (I), (II)
and (III) together imply that unitarily homotopic separable C *-algebras are iso-
morphic if they are σ-unital and stable or if they are unital, i.e., (I), (II) and (III)
imply Theorem B(iv).

Proof of the implication { (I),(II),(III) } ⇒ (iv) of Theorem B:.

Let ϕ : A → B a unitary homotopy and ψ : B → A a *-homomorphism such that
ϕ ◦ ψ is unitarily homotopic to idB and ψ ◦ ϕ is unitarily homotopic to idA . Above
we have shown that ϕ and ψ must be monomorphisms such that Bϕ(A)B is dense
in B and Aψ(B)A is dense in A. By (I) there exists monomorphisms ϕ1 : A → B
and ψ1 : B → A which are unitarily homotopic to ϕ and ψ respectively, such that
ϕ1 (A)B and ψ1 (B)A are dense in B and A respectively.
By (II), the ϕ1 and ψ1 again define a unitary homotopy between A and B.
Now (III) applies, and we get an isomorphism χ from A onto B such that χ is
unitarily homotopic to ϕ1 . But then χ is unitarily homotopic to ϕ by (II).
We have that B, ϕ and ψ are unital if A unital. Thus (III) applies to ϕ in the
unital case. 

Proof of (II): . The definition of unitary homotopy shows immediately that

k2 h1 is unitarily homotopic to h2 h1 . Since k2 (B)C is dense in C there is a unital
strictly continuous *-homomorphism ψ from the multiplier algebra M(B) of B into
the multiplier algebra M(C) of C such that ψ|B = k2 . If t ∈ R+ 7→ u(t) ∈ M(B) is
a strictly continuous map into the unitaries of M(B) such that lim ku(t)∗ h1 (a)u(t)−
k1 (a)k = 0 for a ∈ A, then t ∈ R+ → v(t) := ψ(u(t)) ∈ M(C) is strictly continuous
and lim kv(t)∗ k2 (h1 (a))v(t) − k2 (k1 (a))k = 0 for a ∈ A. Thus k2 h1 and k2 k1 are

2The norm-continuous path t ∈ R 7→ U (t) ∈ M(B) with lim ∗

+ t→∞ kk(a) − U (t) h(a)U (t)k =
0 (for all a ∈ A) can be taken in the unitization B
e = B + C · 1 ⊂ M(B).
 2. PROOF OF PROPERTY (III) 959

unitarily homotopic. Since unitary homotopy is an equivalence relation, k2 k1 is

unitarily homotopic to h2 h1 . 

2. Proof of property (III)

This Section is concerned with the proof of (III), that finishes the proof of
Theorem B. The proof of (III) is given above Corollary 10.3.1. Then we get some
other Corollaries of our technical Proposition 10.2.1.
We start with a very general Proposition 10.2.1 that is not exactly stated in
[438] as here, but is proved in [438], cf. proofs of [438, cor. 2.4(II), prop. 2.3].
Let Xn ⊂ Y be a sequence of Banach spaces and closed subspaces, and
let Vn : Xn → Xn+1 be a sequence of contractions. Then the inductive limit
indlim(Vn : Xn → Xn+1 ) is well-defined and naturally contained in `∞ (Y )/c0 (Y )
(or in a norm ultrapower Yω ) as a closed subspace. By

Vn∞ : Xn → indlim(Vn : Xn → Xn+1 )

we denote the natural contractions which are given in the definition of inductive
∞
limits by universal diagrams (and satisfy Vn+1 ◦ Vn = Vn∞ ). See [438, sec. 2] for
definitions and explicit formulas (in the case of operator systems).
In the following Proposition 10.2.1 let X be a separable Banach space and Y a
closed subspace of X, S ⊂ L(X) a semigroup of linear contractions on X, G be a
topological semigroup π : G → S a strongly continuous semigroup epimorphism.
To simplify notation we write V , Vn , V (t) and Vn (t) instead of π(g), π(gn ),
π(g(t)), π(gn (t)) etc.

Proposition 10.2.1. Let X, Y , S, G and π be as above.

(i) Assume that for every finite dimensional subspace Z1 of X and for every
finite dimensional subspace Z2 of Y and every ε > 0 there exist V ∈ S
with

kz − V zk ≤ εkzk ∀z ∈ Z2 and dist(V z, Y ) ≤ εkzk ∀z ∈ Z1 .

Then there exist a sequence V1 , V2 , . . . in S and an isometric isomorphism

∼
ϕ : Y → indlim(Vn : X → X)

from Y onto indlim(Vn : X → X), such that, for y ∈ Y ,

ϕ(y) = lim Vn∞ (y) .

n→∞

If, moreover, G is a group and π(1G ) = idX , then (V1∞ )−1 is an

isometric isomorphism from indlim(Vn : X → X) onto X and there is a
sequence (of isometries) Tn ∈ S such that, for y ∈ Y ,

(V1∞ )−1 (ϕ(y)) = lim Tn (y).

n→∞

(V1∞ )−1 ◦ ϕ is an isomorphism from Y onto X.

960 10. UNITARILY HOMOTOPIC ALGEBRAS AND THEOREM B(IV)

(ii) Suppose that, moreover, idX ∈ S, G is unital, π(1G ) = idX , and for every
pair of subspaces Z1 ⊂ X, Z2 ⊂ Y of finite dimension and for every ε > 0,
there exists a continuous map s ∈ [0, 1] 7→ g(s) ∈ G with g(0) = 1G , such
that, with V (s) := π(g(s)), for all s ∈ [0, 1],

kz − V (s)zk ≤ εkzk ∀z ∈ Z2 and dist(V (s)z, Y ) ≤ εkzk ∀z ∈ Z1 .

Then, for every dense sequence y1 , y2 , . . . ∈ Y the sequence

V1 , V2 , . . . ∈ S in (i) and the sequence g1 , g2 , · · · ∈ G for which π(gn ) = Vn
in (i) can be selected such that moreover for every n ∈ N there is a
continuous map t ∈ [0, 1] 7→ gn (t) ∈ G with kVn (t)yk − yk k < 2−n for
every k ≤ n, gn (0) = 1G , Vn (1) = Vn , V0 (t) ≡ 1.
∞
Furthermore, W (t) := Vn+1 ◦ Vn (n + 1 − t) (for n ≤ t ≤ n + 1) has
the property that

t ∈ R+ 7→ W (t)y ∈ indlim(Vn : X → X)

is continuous with W (0) = V1∞ , and

lim W (t)y = ϕ(y) .

t→∞

If, moreover, G is a group and π(1G ) = idX , then (V1∞ )−1 is an

isometric isomorphism from indlim(Vn : X → X) onto X.
The map t ∈ R+ 7→ g(t) ∈ G, which is defined by

g(t) := g1−1 · . . . · gn−1 gn+1 (t − n)−1 for n ≤ t ≤ n + 1 ,

is continuous and satisfies π(g(t)) = (V1∞ )−1 W (t).

The inclusion map Y ,→ X is homotopic to the isometric isomorphism
(V1∞ )−1 ◦ ϕ from Y onto X by the strongly continuous map t ∈ R+ 7→
V (t) := π(g(t)).

Proof. (i): The proof [438, cor. 2.4(II)] can be used here (word by word in
the same way), but the inductive existence proof of the Vn ∈ S with the desired
properties does not need in the induction step a Hahn-Banach separation argument,
as in [438], the existence is just our assumption in (i) that S is a semigroup and
has (!) the the needed approximation property.
If G is a group and π(1G ) = idX , then the Vn are isometries from X onto X
and thus V1∞ is an isometric isomorphism from X onto indlim(Vn : X → X). The
desired Tn can be found as iterates (products) of the Vn .
(ii): It is in principle the same proof as of [438, cor. 2.4(II)], but one has to
construct the finite-dimensional filtrations Xn ⊂ X, Yn ⊂ Y and the Vn = Vn (1)
in [438, prop. 2.3] such that the ball of V (t)Vmn (Xn ) is 8−n -contained in Xn+1 for
t ∈ [0, 1] and Vn (t) := π(gn (t)) with gn (0) = 1, V (1) = Vn . That can be done by
our assumptions in (ii).
Now assume that moreover G is a group and π(1) = idX . Than V1∞ is an
isometric isomorphism as one can easily see from the natural model for indlim in
 2. PROOF OF PROPERTY (III) 961

l∞ (X)/c0 (X), cf. e.g. beginning of [438, sec. 2]. The rest can be verified by straight
forward calculations. 

Note that on the unitary operators in M(B) the strong and the strict topology
coincide. We get the applications of Proposition 10.2.1 if we considered M(B) as
a closed subalgebra {T ∈ L(B) : ∃(S ∈ L(B)), (Sb)∗ a = b∗ T a, ∀a, b ∈ B} of the
bounded operators L(B) on B. The strict topology is the *-strong topology on
L(B) given by the seminorms T 7→ kT (b)k + kT (b∗ )k, b ∈ B.
The natural group epimorphism u ∈ M(B) 7→ ιu ∈ Int(B) ⊂ L(B), with
ιu (a) := uau∗ for a ∈ B, is continuous with respect to the strong topologies (on
both sides).

Corollary 10.2.2. Let ϕ : A → B be a *-homomorphism and suppose that A

and B are separable. Then ϕ is unitarily homotopic to an isomorphism ψ from A
onto B, if and only if, ϕ satisfies the following property (δ):
ϕ is a monomorphism and, for every ε > 0, a1 , . . . , an ∈ A, b1 , . . . , bm ∈ B
there exist a strongly continuous map t ∈ [0, 1] 7→ v(t) ∈ M(B) from [0, 1] into the
unitary operators in M(B) such that

(i) v(0) = 1,
(ii) k[v(t), ϕ(ai )]k < ε for i = 1, . . . , n, t ∈ [0, 1],
(iii) dist(v(1)∗ bj v(1), ϕ(A)) < ε for j = 1, . . . , m.

Proof. Assume ϕ : A → B satisfies the property (δ), then with X := B,

Y := ϕ(A), G = U (M(B)) := unitary group of M(B) with strict topology,
π(u) := ιu and S := {ιu : u ∈ U (M(B))} the assumptions of Proposition 10.2.1(ii)
are satisfied. Thus the inclusion map ϕ(A) ,→ B is unitarily homotopic to an
isomorphism from ϕ(A) onto B.
Since ϕ is a monomorphism, it follows that ϕ is unitarily homotopic to an
isomorphism from A onto B.
Conversely, suppose that ϕ is unitarily homotopic to an isomorphism ψ from
A onto B. Along a unitary homotopy the norm of an element remains constant.
Thus ϕ is a monomorphism.
Let t ∈ R+ 7→ u(t) ∈ M(B) be the strongly continuous map into the unitaries
corresponding to the unitary homotopy. Then there exists a t0 such that, for every
t ≥ t0 ,
kϕ(ai ) − u(t)ψ(ai )u(t)∗ k < ε/2.

Let fj := ψ −1 (u(t0 )∗ bj u(t0 )) ∈ A .

We find t1 > t0 with kϕ(fj ) − u(t)ψ(fj )u(t)∗ k < ε for every t ≥ t1 .
Let s(t) := t0 + t(t1 − t0 ) and v(t) := u(t0 )u(s(t))∗ for t ∈ [0, 1].
Then v(0) = 1, v(t) is unitary, t 7→ v(t) is strongly continuous and, by the
triangle inequality, for t ∈ [0, 1], kv(t)∗ ϕ(ai )v(t) − ϕ(ai )k < ε.
962 10. UNITARILY HOMOTOPIC ALGEBRAS AND THEOREM B(IV)

Furthermore, dist(v(1)∗ bj v(1), ϕ(A)) ≤ kv(1)∗ bj v(1) − ϕ(fj )k ≤ ε, because

v(1)∗ bj v(1) = u(t1 )ψ(fj )u(t1 )∗ . 

Corollary 10.2.3. Suppose that ϕn : An → An+1 , n = 1, 2, . . ., are *-homo-

morphisms, where An is separable, ϕn (An ) contains a strictly positive element of
An+1 , and there exists an isomorphism ψn from An onto An+1 such that ψn is
unitarily homotopic to ϕn .
Then ϕ∞
1 : A1 → indlim(ϕn : An → An+1 ) is unitarily homotopic to an isomor-
phism from A1 onto indlim(ϕn : An → An+1 ).

Proof. We have to check the criteria (δ) of Corollary 10.2.2. By Corollary

10.2.2 every ϕn is a monomorphism.
S ∞
ϕn (An ) is dense in B := indlim(ϕn : An → An+1 ).
The assumptions imply that ϕ∞ n (An ) contains, for every n, a strictly positive
element of B, because ϕn (An )An+1 is dense in An+1 , and then by induction:

ϕn+k ◦ . . . ◦ ϕn (An )An+k+1 = An+k+1 , ϕ∞

n (An )B = B .

Thus M(ϕ∞ ∼
n (An )) = M(An ) is unitally contained in M(B) and the strict
∞
topology on M(ϕn (An )) is the same as the topology which is induced by the strict
topology of M(B). Therefore it suffices to check that ϕ∞ ∞
1 : A1 → ϕn (An ) satisfies
the criteria (δ) of Corollary 10.2.2. But this follows from Corollary 10.2.2 and the
following observation:
By induction and (II), (ϕ∞n )
−1
◦ ϕ∞
1 = ϕn−1 ◦ . . . ◦ ϕ1 is unitarily homotopic
to the isomorphism ψn−1 ◦ · · · ◦ ψ1 from A1 onto An , where ψk : Ak → Ak+1 are
isomorphisms which are unitarily homotopic to ϕk : Ak → Ak+1 (k = 1, . . . , n −
1). 

Now we are ready for the proof of the above statement (III), and finish here
the proof of Theorem B:
check next proof again!!

Proof of (III). Suppose that A and B are separable.

Let h : A → B and k : B → A monomorphisms, such that h(A)B = B and
k(B)A = A, i.e., h(A) contains a strictly positive element of B, and h(B) contains
a strictly positive element of A. It follows

hk(B)B = hk(B)h(A)B = h(k(B)A)B = h(A)B = B.

In the same way kh(A)A = A. If, moreover, kh is unitarily homotopic to idA ,

then, by Corollary 10.2.3, there exists an isomorphism α from A onto the inductive
limit A∞ of the sequence of morphisms χn : A → A with χn := kh, such that α is
unitarily homotopic to χ∞1 : A → A∞ .
 3. SOME COROLLARIES 963

If hk is unitarily homotopic to idB , then there exists an isomorphism β from B

onto the inductive limit B∞ of ψn : B → B with ψn := hk, such that β is unitarily
homotopic to ψ1∞ : B → B∞ .
If we consider the natural inclusions

A∞ ⊂ l∞ (A)/c0 (A), B∞ ⊂ l∞ (B)/c0 (B),

we see, that

(a1 , a2 , . . .) 7→ (h(a1 ), h(a2 ), . . .) and (b1 , b2 , . . .) 7→ (k(b1 ), k(b2 ), . . .)

define (modulo zero-sequences) monomorphisms µ from A∞ into B∞ and ν from

B∞ into A∞ such that with νµ = id and µν = id. Therefore, µ is an isomorphism.
Thus ϕ := β −1 ◦ µ ◦ α : A → B is an isomorphism from A onto B. A simple
calculation shows that µ ◦ χ∞ ∞
1 = ψ1 ◦ h.

β −1 ◦ µ ◦ α is unitarily homotopic to β −1 ◦ µ ◦ χ∞
1 =β
−1
◦ ψ1∞ ◦ h.
Since ψ1∞ is unitarily homotopic to β, we get that ψ1∞ ◦h is unitarily homotopic
to β ◦ h.
Thus ϕ = β −1 ◦ µ ◦ α is unitarily homotopic to h. 

3. Some corollaries

We add some related corollaries of the first part of Proposition 10.2.1, which
can be proved by almost the same arguments as above.

Corollary 10.3.1. Let ϕ : A → B be a *-homomorphism, where A and B

are separable. Then the following are equivalent:

(i) ϕ is approximately unitarily equivalent to an isomorphism ψ from A onto

B.
(ii) ϕ is a monomorphism and for every ε > 0, a1 , . . . , an ∈ A, b1 , . . . , bm ∈
B there exists a unitary v ∈ M(B) with kv ∗ ϕ(ai )v − ϕ(ai )k < ε for
i = 1, . . . , n and dist(v ∗ bj v, ϕ(A)) < ε for j = 1, . . . , m.
(iii) ϕ is a monomorphism, and in the relative commutant ϕ(A)0 ∩ M(B)ω
of ϕ(A) in the ultrapower of M(B) there exists, for every finite sequence
b1 , . . . , bm ∈ B and every ε > 0, a unitary v ∈ ϕ(A)0 ∩ M(B)ω such that
v ∗ bj v has distance < ε from ϕω (Aω ) for j = 1, . . . , m.

Remark 10.3.2. Part (iii) of Corollary 10.3.1 is just an equivalent formulation

of (ii) in the more rigorous language of ultrapowers. There is a more elegant and
far reaching version in the ultrapower context.
One concludes Corollary 10.3.1 from Proposition 10.2.1(i) essentially in the
same way as we have deduced Corollary 10.2.2 from Proposition 10.2.1(ii). And
one can use the same arguments for the implications “Corollary 10.3.1 ⇒ Corol-
lary 10.3.3 ⇒ Corollary 10.3.4” below as for the implications “Corollary 10.2.2 ⇒
Corollary 10.2.3 ⇒ (III)” above.
964 10. UNITARILY HOMOTOPIC ALGEBRAS AND THEOREM B(IV)

Corollaries 10.3.4 and 10.3.5 overlap with (partly not published) results of El-
liott.

Corollary 10.3.3. Suppose that, for n = 1, 2, . . ., An is separable, ϕn : An →

An+1 are *-homomorphisms, ϕn (An ) contains a strictly positive element of An+1
and ϕn is approximately unitarily equivalent to an isomorphism from An onto An+1 .
Then ϕ∞1 : A1 → indlim(ϕn : An → An+1 ) is approximately unitarily equivalent
to an isomorphism from A1 onto indlim(ϕn : An → An+1 ).

Corollary 10.3.4. If A and B are separable C*-algebras, h : A → B and

k : B → A are C*-algebra morphisms such that h(A)B = B, k(B)A = A, kh and
hk are approximately unitarily equivalent to idA and idB respectively, then h is
approximately unitarily equivalent to an isomorphism from A onto B.

Corollary 10.3.5. Suppose that A is a separable C*-algebra, which is stable

or unital and that B is a separable unital C*-subalgebra of the multiplier algebra
M(Aω ) of Aω , such that B commutes element-wise with A, and such that b ∈ B 7→
b ⊗ 1 ∈ B ⊗ B and b ∈ B 7→ 1 ⊗ b ∈ B ⊗ B are approximately unitarily equivalent
(to each other in B ⊗ B), then

(i) B is simple and nuclear.

(ii) A ∼= A ⊗ B ⊗ B ⊗ . . . with an isomorphism ψ : A → A ⊗ B ⊗ B ⊗ . . . which
is approximately unitarily equivalent to a ∈ A 7→ a ⊗ 1 ⊗ 1 ⊗ . . .

Proof. (i): Let J be a closed ideal of B and J 6= B, ν a pure state of B

with ν(J) = 0. Since B is separable, we find a sequence of unitaries un ∈ B ⊗ B
and 2−n -approximations an of un in the algebraic tensor product B B, such that
b⊗1 = limn→∞ u∗n (1⊗b)un . Thus b = lim(id ⊗ν)(a∗n (1⊗b)an ) for every b ∈ B. The
right hand side is a sequence of completely positive finite rank maps with kernels
which contain J. Thus J = 0 and B is nuclear. Every closed ideal of B is trivial,
i.e. B is simple.
(ii): Since B is nuclear and simple, there is a copy of A⊗(B ⊗B) ∼= (A⊗B)⊗B
in the ultrapower (A ⊗ B)ω of A ⊗ B, such that A ⊗ 1 ⊗ B corresponds naturally to
the canonical embedding of A⊗B in (A⊗B)ω by constant sequences, and A⊗B ⊗1
is in C ∗ (A · (A0 ∩ M(Aω ))) ⊂ Aω ⊂ (A ⊗ B)ω , where Aω is naturally identified with
(A ⊗ 1B )ω .
Let v ∈ M(A ⊗ B)ω denote the unitary, which is represented by 1M(A) ⊗ un ,
where un is as in part (i). We see that the criteria (iii) of Corollary 10.3.1 is satisfied
for ϕ : a ∈ A 7→ a ⊗ 1B ∈ A ⊗ B. Thus ϕ is approximately unitarily equivalent to
to an isomorphism ψ1 from A onto A ⊗ B.
It follows by induction that ϕn : An → An+1 is approximately unitarily equiv-
alent to an isomorphism from An onto An+1 , where inductively A1 = A, An+1 =
An ⊗ B and ϕn (a) = a ⊗ 1B for a ∈ An . But then Corollary 10.3.3 gives that
the monomorphism a 7→ a ⊗ 1B ⊗ 1B ⊗ . . . from A into A ⊗ B ⊗ B ⊗ . . . =
 3. SOME COROLLARIES 965

indlim(ϕn : An → An+1 ) is approximately unitarily equivalent to an isomorphism

from A onto A ⊗ B ⊗ B ⊗ . . .. 

Remark 10.3.6. Since we have finished the proof of Theorem B with the above
given proof of (III), now we can use the results A-F and H as stated in Chapter 1.
One easily sees from Theorem B(ii)+(iii) and Corollaries C(ii) and F that B =
O2 and B = O∞ satisfy the requirements of Corollary 10.3.5. Therefore, also
Corollary G now follows from Corollary F and Corollary 10.3.5, if we do not use
the original paper [678].
We recall from Definition 3.10.1 that a *-homomorphism h1 : A → B approxi-
mately dominates a completely positive contraction h2 : A → B, if and only if, there
exists a sequence of contractions dn ∈ B such that lim kh2 (a) − d∗n h1 (a)dn k = 0 for
a ∈ A. Clearly, we can choose the dn ’s as isometries if A, B, h1 and h2 are unital.
If A and B are stable and σ-unital, then, by Proposition 7.4.7, we find isometries
tn ∈ M(B) such that lim kh2 (a) − t∗n h1 (a)tn k = 0 for a ∈ A.

Corollary 10.3.7. Suppose that A is a separable C*-algebra which is stable

or is unital and contains a copy of O2 unitally.
If idA approximately dominates idA ⊕ idA , then A ∼
= O∞ ⊗ A.
If idA is approximately unitarily equivalent to idA ⊕ idA by a sequence of uni-
taries in the multiplier algebra of A, then A ∼
= O∞ ⊗ A ∼ = O2 ⊗ A.

Proof. By the above Remark 10.3.6, the assumptions mean that there is an
isometry T ∈ M(A)ω such that T ∗ aT = sas∗ +tat∗ for a ∈ A and suitable canonical
generators of O2 ⊂ M(A). It follows that t1 := T t and s1 := T s are isometries in
M(A)ω such that t∗1 s1 = 0 and the elements t1 and s1 commute elementwise with
A ⊆ Aω . C ∗ (s1 , t1 ) contains a copy of B := O∞ unitally.
The natural unital *-homomorphism from M(A)ω into M(Aω ) fixes A ⊂ Aω
and maps, therefore, B unitally and, by simplicity of B, faithfully into A0 ∩M(Aω ).
By Corollary 10.3.5 and Remark 10.3.6, A ∼ = O∞ ⊗ A.
If idA is approximately unitarily equivalent to idA ⊕ idA , then the above con-
sidered T is unitary and t1 t∗1 + s1 s∗1 = 1, i.e., C ∗ (s1 , t1 ) is naturally isomorphic to
O2 . By Corollary 10.3.5 and Remark 10.3.6, A ∼ = O2 ⊗ A. 

Corollary 10.3.8. Suppose that A is a separable nuclear C*-algebra. Then A

satisfies the WvN-property of Definition 1.2.3, if and only if, A ⊗ K ∼
= A ⊗ K ⊗ O∞ .

Proof. We may suppose that A is stable, because A satisfies the WvN-

property if and only if A⊗K satisfies the WvN-property, as one can see immediately
from Definition 1.2.3.
Since O∞ ∼= O∞ ⊗O∞ ⊗. . ., we get from Proposition 3.10.15 that A is strongly
p.i. and has the WvN-property if A ∼
= A ⊗ O∞ .
966 10. UNITARILY HOMOTOPIC ALGEBRAS AND THEOREM B(IV)

If A is nuclear, then id ⊕ id is residually nuclear. Thus, if A satisfies the WvN-

property, then id ⊕ id is approximately dominated by id. Hence, A ∼ = A ⊗ O∞ by
Corollary 10.3.7. 

Revise next. Give ref.

Remark 10.3.9. In the paper with M. Rørdam [463] we show that all strongly
purely infinite C *-algebras have the WvN-property.
Mention the corrections that have to be done
in the paper with Rørdam [463]
(as in Info for Ando)!!!
It is an open question if every purely infinite separable nuclear C *-algebra has
the WvN-property.
Mention here:
“regular Abelian C *-subalgebra”
implies
“Abelian factorization”,
implies existence of
“ideal-separating residually nuclear map”
(= “residually nuclear separation”)
The progress is partly mentioned in Chp. 12.

Corollary 10.3.10. Suppose that A and B are stable separable nuclear C*-
algebras, and that h : A → B, k : B → A and ψ : A ⊗ O2 → A ⊗ O2 are *-
monomorphisms.

(i) If, for every primitive ideal J ⊂ A,

ψ(J ⊗ O2 ) = ψ(A ⊗ O2 ) ∩ (J ⊗ O2 ),

then ψ is unitarily homotopic to the identity map of A ⊗ O2 .

(ii) h ⊗ idO2 is unitarily homotopic to an isomorphism from A ⊗ O2 onto
B ⊗ O2 if kh(J) = kh(A) ∩ J and hk(K) = hk(B) ∩ K for every primitive
ideal J ⊂ A and every primitive ideal K ⊂ B.

Proof. (i): By assumption on ψ, ψ and id are both residually nuclear *-

monomorphisms and induce on A ⊗ O2 the same action of the primitive ideal space
of A. Thus, by Corollary 7.4.3, ψ and id are unitarily homotopic.
(ii): Since A and B are stable, we find by (I) and (II) *-monomorphisms
h1 : A → B and k1 : B → A, such that h1 is unitarily homotopic to h, k1 is unitarily
homotopic to k, h1 (A)B is dense in B, k1 (B)A is dense in A, and hk is unitarily
homotopic to h1 k1 , kh is unitarily homotopic to k1 h1 . Thus h1 k1 and k1 h1 again
take the system of closed ideals invariant. This carries over to h1 k1 ⊗ idO2 and
 3. SOME COROLLARIES 967

k1 h1 ⊗ idO2 . Thus, part(i) and the above proven (III) imply, that h1 ⊗ idO2 , and,
therefore, h ⊗ idO2 are unitarily homotopic to an isomorphism. 

We do not know if it is necessary to assume in the following Corollary 10.3.11

that the algebras Bn are stable.

Corollary 10.3.11. Suppose that (Bn ) is a sequence of separable stable C*-

algebras, and that hn : Bn → Bn+1 are *-homomorphisms.
Let B := indlim(hn : Bn → Bn+1 ).
If Bn ∼
= Bn ⊗ O∞ (resp. Bn ∼
= Bn ⊗ O2 ) for n = 1, 2, . . ., then B ∼
= B ⊗ O∞
∼
(resp. B = B ⊗ O2 ).

Proof. Let h∞ ∞
n : Bn → B the natural morphism and let Cn := hn (Bn ). Since
K, O∞ and O2 are simple and nuclear, Cn is stable and Cn ∼
= Cn ⊗ O∞ (resp. Cn ∼
=
Cn ⊗ O2 ).
The union of the stable subalgebras Cn is dense in B. By [373], cf. also
Corollary 5.5.3, B is stable, i.e. B ∼
= B ⊗ K.
Therefore, we can assume, that M(B) contains isometries s, t with ss∗ +tt∗ = 1,
such that dCn + Cn d ⊂ Cn for every d ∈ C ∗ (s, t) ∼
= O2 and n ∈ N. Otherwise, we
replace B by B ⊗ K, Cn by Cn ⊗ K, and hn by hn ⊗ idK .
By Corollary 10.3.7, idB |Cn approximately dominates (idB ⊕ idB )|Cn . There-
fore, by Corollary 10.3.7, B ∼
= B ⊗ O∞ .
If Cn ∼
= Cn ⊗ O2 , then, by Corollary 10.3.7, idB |Cn is approximately unitarily
equivalent to (idB ⊕ idB )|Cn by unitary elements of the multiplier algebra of Cn .
By [180], cf. [816, chp. 16], the unitary group of M(C) is contractible if C
is σ-unital and stable, cf. our version ?? of a proof. It follows, that the unitary
operators of M(C) can be approximated by unitary operators in the unitization of
C in the strict (= strong*) topology.
Therefore, idB |Cn is also approximately unitarily equivalent to (idB ⊕ idB )|Cn
by unitaries in the unitization of B. Thus, the morphisms idB and idB ⊕ idB are
approximately unitarily equivalent. Thus, by Corollary 10.3.7, B ∼ = B ⊗ O2 . 

Corollary 10.3.12. The class of separable nuclear C*-algebras with WvN-

property is closed under inductive limits.

Proof. There is a more general result for inductive limits of strongly p.i. al-
gebras (cf. [463]), but here we use Corollary 10.3.7.
Let hn : An → An+1 , n = 1, 2, . . ., a sequence of *-homomorphisms, where An
is separable, nuclear and has the WvN-property. Let A denote its inductive limit,
and let B := A ⊗ K.
By Definition 1.2.3, the class of C *-algebras with WvN-property is closed under
stabilization and passage to hereditary C *-subalgebras, i.e., it suffices to show that
B satisfies the WvN-property.
968 10. UNITARILY HOMOTOPIC ALGEBRAS AND THEOREM B(IV)

By Corollary 10.3.8, Bn := An ⊗ K, satisfies Bn ∼ = Bn ⊗ O∞ and B is the

inductive limit of the sequence hn ⊗ idK : Bn → Bn+1 .
Therefore, by Corollary 10.3.11, B ∼
= B ⊗ O∞ .
B is separable, stable and nuclear. Thus B has the WvN-property by Corollary
10.3.8. 
 CHAPTER 11

PI-SUN algebras represent the nuclear KK-classes

In this Chapter we prove Theorem I of Chapter 1 and some additional results.

There are different methods to construct purely infinite algebras that are
G
KK (X; ·, ·)-equivalent to a given separable (G, X)-algebra, and have simple fibers
with respect to the action of X.
(1) inductive limit construction in the case of exact algebras and exact discrete
groups, 1a) direct inverses in Ext. 1b) using σ-additivity of KK w.r.t. first variable.
(2) construction of a suitable Toeplitz-Pimsner algebra (such that it is equal to
the corresponding Cuntz-Pimsner algebra).
Citation from Chapter 1, changes to bundle/equvariant case: ??
In fact, we prove in Chapter 11 a more general result for continuous fields
(Ax )x∈X of exact C *-algebras Ax on a compact Hausdorff space X with the addi-
tional property that the algebra A of continuous sections is exact (which is e.g. the
case if all fibers Ax are nuclear or if X is finite):
One has to take again At := A ⊗ O∞ st if A is unital, and define At as the
natural unital split extension of

C(X, O2 ) ∼
= [C(X)1M(A) ] ⊗ O2

by A ⊗ O∞ st if A is not unital. Then one can use the sub-trivialization theorem of

E. Blanchard [89] (that can be shown using Theorem A and is a special case of The-
orem K), to get a unital C(X)-module *-monomorphism hu0 : At → C(X, O2 ) ⊂ At .
We define again h := idAt ⊕hu0 and let P(A) := indlim(h : At → At ). All the opera-
tions are consistent with the above defined constructions of (Ax )t and P (Ax ) for the
fibers Ax of (Ax )x∈X . It is not difficult to see that (P (Ax ))x∈X is a continuous field
of C *–algebras, and that P(A) is the algebra of continuous sections of this field.
By construction, P(A) is exact. Further, the *-monomorphisms ηx : Ax → P (Ax )
define a C(X)-module monomorphism η from A into P(A).
We show in Chapter 11 the more general result:
The monomorphism η defines a KK(X; ·, ·)-equivalence between A and P(A) . ( 1 ).
At the end of Chapter 11 we outline an other construction that leads to
KK(X; ·, ·)-equivalent exact (respectively nuclear) algebras with primitive ideal
space isomorphic to X (cf. Theorems 11.4.1 and O):

1The KK(X; A, B) is for compact metric spaces X just the Kasparov functor R KKG (X; ., .)

for the base space X, G is the trivial group).

969
970 11. PI-SUN ALGEBRAS REPRESENT THE NUCLEAR KK-CLASSES

One starts with a suitable Hilbert bimodule and builds the corresponding Cuntz–
Krieger–Pimsner algebra. The construction has the advantage that G-actions of
locally compact groups G on (Ax )x∈X lead in a natural way to G-actions on
(P (Ax ) ⊗ K)x∈X ( 2 ).
Furthermore, we modify in Chapter 11 the above described construction of
A ⇒ P (A) in a way that we get (cf. Theorem 11.0.7):

Suppose that A is a unital separable exact C*-algebra, G is a countable (discrete)

exact group, and that α : G → Aut(A) is a group-morphism.
Then there exist a purely infinite exact unital C*-algebra B, a group-morphism
β : G → Aut(B) and a G-equivariant unital monomorphism η : A → B that defines
a KKG -equivalence [η] ∈ KKG (A, B).
It can be managed that η(A) is the range E(B) of a β(G)-equivariant conditional
expectation E on B.
If — in addition — A is nuclear, then one can find B, β and E, such that B
is nuclear — in addition —.

some new approach:

Definition: ????????????
To get some more general and farer going results on KK-equivalent embedding
into purely infinite algebras, we have to make some definitions, concerning maps
that changes the identity maps not to much in the sense of KK-theory.

Definition 11.0.1. Suppose that D is unital, separable and exact. A unital

monomorphism h : D → E will be called elementary if there are

(i) a unital *-monomorphism h0 : D → O2 ,

(ii) a *-monomorphism γ : O2 → O∞ , and
(iii) an isometry T ∈ O∞
(iv) an isomorphism ι : D ⊗ O2 → E such that
h(d) = ι((d ⊗ T T ∗ ) + (1 ⊗ γ(h0 (d)))) for all d ∈ D .

We call h plain if E = D ⊗ O∞ and ι = id.

More generally:
Suppose O∞ ⊂ M(D), k0 : D⊗O2 → D⊗O∞ , h0 (d) := k0 (d⊗1), “sufficiently”
degenerate. consider similar sum-construction variants
h(d) := T dT ∗ + h0 (d) etc.
Consider the cases of bundles, to see what is the right definition.
h0 : D → C0 (X, O2 )
2At least in the case where A is is nuclear, because then, by Theorem M(iii), there are
x
C(X)-module isomorphisms of P(A) ⊗ K onto any other purely infinite stable separable nuclear
C(X)-algebra B (with simple fibers Bx ) that is KK(X; ·, ·)-equivalent to A.
 11. PI-SUN ALGEBRAS REPRESENT THE NUCLEAR KK-CLASSES 971

generally:
Some continuous action of a lattice O on D should be given.
k0 : D ⊗ O2 → D ⊗ O∞
universally for that action (if k0 goes into D ⊗ O∞ ???)
Suppose that D1 → D2 → · · · → Dk are given by elementary morphisms
hn : Dn → Dn+1 and suppose that D1 ∼
= D1 ⊗ O∞ .
By definition there are isomorphisms ιn : Dn ⊗ O∞ → Dn+1 related to hn .
Consider
λj,k := ιj ⊗ id ⊗ · · · ⊗ id : Dj ⊗ O∞ ⊗ O∞ ⊗(k−j) → Dj+1 ⊗ O∞ ⊗(k−j)
ηk := λk ⊗ · · · ⊗ λ1 : D1 ⊗ O∞ ⊗k → Dk+1 .
The system is called homotopy-commutative, if ηk ⊗ id is unitarily homotopic
to ηk+1 and if every unitary homotopy on the boundary of a triangle extends to a
homotopy in the interior.
more precise !!! ???

Technical Lemma:

Lemma 11.0.2. Iterations of elementary morphisms h are elementary.

Elementary morphisms h1 , h2 : D → E are unitarily homotopic if
ι1 , ι2 : D ⊗ O∞ → E
are unitarily homotopic.
Elementary h is a KK-equivalence.
There is a conditional expectation Ph : E → h(D).
Others???

Proposition 11.0.3. Suppose that D1 is separable and exact, hn : Dn → Dn+1 ,

n ∈ N, is a sequence of elementary morphisms and D1 ∼= D1 ⊗ O∞ . Let
P := indlim(hn : Dn → Dn+1 ) .

(i) P is simple, purely infinite, separable, unital and exact (respectively is

pi-sun if the Dn are nuclear).
(ii) There is an approximately inner conditional expectation E : P → P onto
h1,∞ (D1 ), h1,∞ is injective.
(iii) The unital monomorphism h1,∞ : D1 → P defines a KK-equivalence be-
tween D1 and P , if the ιn -diagrams (to/for D1 ⊗ O∞ ) are homotopy com-
mutative.

The idea is, that one considers the m.o.c. cones Ck,` ⊂ CP(Dk , D` ), Ck,∞ ⊂
CP(Dk , P ) and C∞,k ⊂ CP(P, Dk ) that are generated by the hk,` , hk,∞ and condi-
tional expectations from P onto h1,k (D1 ) ⊆ Dk etc.
The consider (perhaps C and “action” - equivariant) asymptotic morphisms:
972 11. PI-SUN ALGEBRAS REPRESENT THE NUCLEAR KK-CLASSES

I.e. consider/construct/define a “natural” asymptotic morphism φ : P →

Q(R+ , D1 ), such that Q(R+ , h1,∞ ) : Q(R+ , D1 ) → Q(R+ , P ) has the property
that [Q(R+ , h1,∞ ) ◦ φ] : P → Q(R+ , P ) is homotopic to idP : P → Q(R+ , P ) and
φ ◦ h1,∞ : D1 → Q(R+ , D1 ) is homotopic to idD1 : D1 → Q(R+ , D1 ).
But all this inside the given m.o.c. cones.
The proof will be postponed to Section ??. (Sec. 11.2 ??)
Instead we first deduce from Proposition 11.0.3 the proof of Theorem I, the
proof of its C(X)-modular generalization Corollary 11.0.5, and the proof of Theo-
rem 11.0.7, that gives a G-equivariant generalization of Theorem I.
Theorem
Theorem 11.0.4. Suppose that A is a stable separable C*-algebra.
Let X a T0 space and Ψ : O(X) → I(A) an action of X on A such that there
exists a non-degenerate nuclear *-monomorphism k0 : A ⊗ O2 → A ⊗ O∞ that is
maximal with respect to the Ψ-residually nuclear maps and reproduces Ψ, i.e.,
J ∈ I(A) is in the image of Ψ, if and only if, k0 (J ⊗ O2 ) = k0 (A ⊗ O2 ) ∩ (J ⊗ O∞ ).
????????????????
Let h0 := k0 ((·) ⊗ 1) and h := id ⊕(h0 ⊗ id). Then P := indlim(h : A ⊗ O∞ →
A⊗O∞ ) is exact (respectively nuclear if A is nuclear), strongly purely infinite (more
precisely: P ∼= P ⊗ O∞ ), Prim(P ) ∼ = X, and η : a ∈ A 7→ h1,∞ (a ⊗ 1) ∈ P defines
a is a KK(X; ·, ·)-equivalence [η] ∈ KK(X; A, P ).
Corollary 11.0.5. Let X a locally compact Polish space. Suppose that A is
a separable exact C0 (X)-C*-algebra, such that x ∈ X → kax k is continuous for
every a ∈ A (i.e., A is the algebra of continuous sections vanishing at infinity of a
continuous field of C*-algebras Ax ). And let P(A) be defined as above.
Then P(A) is a strongly purely infinite exact C0 (X)-C*-algebra with simple
purely infinite fibers P (Ax ), and P(A) ∼
= P(A) ⊗ O∞ .
The natural map η : A → P (A) defines a KK(X; ·, ·)-equivalence.
Lemma 11.0.6. If D ∼ = D ⊗ O∞ and if discrete exact G acts on D by α : G →
Aut(D), then there exists a unitary representation U : G → U(O∞ ) and an ele-
mentary morphism h : D → D ⊗ O∞ with ι unitarily homotopic to id, such that
h ◦ α(g) = (α(g) ⊗ Ad(U (g))) ◦ h for g ∈ G.

Proof. Use an unital embedding η of the exact unital algebra D or,α G into
O2 . One may suppose that O2 is unitally contained in (1 − s1 s∗1 )O∞ (1 − s1 s∗1 ) for
O∞ = C ∗ (s1 , s2 , . . .) . Then U (g) = η(g) + T T ∗ has the property for T = s1 . 
Theorem 11.0.7. Suppose that A is an separable unital nuclear C*-algebra,
and α : G → Aut(A) is an action of a countable discrete exact group G on A.
Then there exist a pi-sun algebra B with an action β : G → Aut(B) of G on B
and a G-equivariant unital monomorphism η : A → B that defines a KK-equivalence
[η] ∈ KK(A, B).
 11. PI-SUN ALGEBRAS REPRESENT THE NUCLEAR KK-CLASSES 973

In particular, K∗ (β(g))◦K∗ (η) = K∗ (η)◦K∗ (α(g)) and K∗ (η) : K∗ (A) → K∗ (B)

is invertible.

It seems that (B, β) and η can be chosen, such that η defines even a KKG -
equivalence [η] ∈ KKG (A, B). (only ”seems” ???)

Proof. Let G act on D1 := A ⊗ O2 by α1 (g) := α(g) ⊗ id for g ∈ G.

Clearly the map k : a → a ⊗ 1 is a G-equivariant monomorphism from A into
D1 , and [k] ∈ KKG (A, D1 ) is a KKG -equivalence, in particular [k] ∈ KK(A, D1 )
is a KK-equivalence. Let Dn := A ⊗ (O∞ ⊗n ). By Lemma 11.0.6 there ex-
ists actions αn : G → Aut(Dn ) and G-equivariant plain elementary morphisms
hn : Dn → Dn+1 .
Now let B := indlim Dn and η := h1,∞ ◦k. Since the hn are G-equivariant, there
is a group morphism β : G → Aut(B) with h1,∞ ◦ α1 (g) = β(g) ◦ h1,∞ for g ∈ G.
By Proposition 11.0.3, B is pi-sun and the morphism η defines a KK-equivalence
[η] ∈ KK(A, B). 

Remark 11.0.8. We call a group G C *-liftable if, for every pi-sun algebra B
in the UCT-class with finitely generated K∗ (B) and every grading-preserving action
γ : G → Aut(K∗ (B)) of G on K∗ (B) that fixes the class [1] ∈ K0 (B), there exists
an action β : G → Aut(B) such that for the natural group morphism µ : Aut(B) →
Aut(K∗ (B)) holds µ ◦ β(g) = γ ◦ µ.
The Theorem 11.0.7 implies that the problem of the determination of all C *-
liftable groups G is equivalent to the question under which circumstances for a
finitely generated Abelian group H and an action γ0 : G → Aut(H) with G-fixed
element x0 ∈ H there is a unital nuclear C *-algebra A in the UCT-class and
an action α : G → Aut(A) such that K1 (A) = 0 and there is an isomorphism
γ : H → K0 (A)
???????????
Then the case K0 (A) = 0 and K1 (A) = H can then be studied by replacing A by
A ⊗ P∞ , where P∞ is the unique pi-sun algebra in the UCT class with K0 (P∞ ) = 0
and K1 (P∞ ) = Z ( 3 ).
The general case follows from the consideration of A = A1 ⊕ (A2 ⊗ P∞ ) in
Theorem 11.0.7, where K1 (Aj ) = 0 for j = 1, 2.
It is evident that every free group is in this class of groups.
It is not clear if Aut(B) → KK(B, B)−1 is surjective for pi-sun
B in UCT-class !!! ?? By the natural epimorphisms (???) Aut(B) →
KK(B, B)−1 → Aut(K∗ (B)), (given by the UCT and Theorem B).

3One can easily show that P = O st o Z for the flip action µ of Z on the standard form
∞ ∞ 2 2
O∞ ∼st
= P∞ ⊗ P∞ with K0 (µ)(x) = −x — it is simply the flip on the tensor product, cf. [448,
Rem.4.3(2)] and observe that, for the coordinate flip on R2 , the crossed product C0 (R2 ) × Z2 is
isomorphic to C0 (R) ⊗ C, with C := {f ∈ C0 (R0 , M2 ) ; f (0) = diagonal}, i.e., K∗ (C) = (Z, 0) .
974 11. PI-SUN ALGEBRAS REPRESENT THE NUCLEAR KK-CLASSES

Contributions of J. Spielberg [732] and (farer going) results of T. Kasura ([416],

[417]) show that in particular all cyclic groups are C *-liftable.
 11. PI-SUN ALGEBRAS REPRESENT THE NUCLEAR KK-CLASSES 975

Begin of old text:

Note that we can construct in the case of exact algebras only KK-equivalences,
but not KKnuc -equivalences because thre is no useful non-elementary criteria for
which exact separable A the identity map of A defines a class [idA ] in KKnuc (A, A).
(But there is a natural map from KKnuc (A, A) onto an ideal of the ring KK(A, A).)
Recall that O∞ st := (1 − s1 s∗1 )O∞ (1 − s1 s∗1 ) ⊂ O∞ . It contains a copy of O2
unitally.
If A is exact, separable and unital let At := A ⊗ O∞ st . If A is exact, separable
and non-unital, then let At denote the unital C *-subalgebra of M(A) ⊗ O∞ st given
by
At := A ⊗ O∞ st + 1 ⊗ O2 ⊆ M(A) ⊗ O∞ st ,
a split extension of O2 by A ⊗ O∞ st .
One can variate the construction if A is stable to get a stable version of At by
considering first a copy of K ⊂ M(A) with K ∩ A = {0} and K · A = A, existing by
Remark 5.1.1(8), and then let At := A⊗O∞ st +K⊗O2 , contained in M(A)⊗O∞ st .
Let S, T the generators of O2 = C ∗ (S, T ; SS ∗ + T T ∗ = 1, S ∗ S = 1 = T ∗ T ).
The algebra At is exact because A is exact and O∞ st and O2 are nuclear
(cf. [432]).
To avoid later problems with notations, now we change the notation for At :
Let B := At in the sequel. Note that the exact C*-algebra B is unital and contains
O2 unitally.
Define h := idB ⊕hu0 where the unital monomorphism hu0 : B → O2 ⊂ B is as
described in Chapter 1 before Theorem B. (Respectively, if A is stable, let h :=
idB ⊕h0 for the more general non-degenerate “maximal” ideal-system preserving
nuclear map h0 from A ⊗ O∞ into C ⊂ A ⊗ O∞ , where C is a nuclear and stable,
O2 absorbing C *-subalgebra of A ⊗ O2 with the ideal system of A)
P (A) := indlim(h : B → B) is purely infinite and simple by the criteria in
check if (iv) or (v) is used Proposition 2.2.5(iv)
for hn = h, Bn = B, because for ε > 0, a, b in B+ with kak = kbk = 1 we can
find an isometry v in O2 such that k1 − v ∗ h0 (a)vk < ε. Thus

kh(b) − h(b)1/2 v ∗ t∗ h(a)tvh(b)1/2 k < ε .

Since exactness is closed under inductive limits ([432]), P (A) is exact. P (A)
is nuclear if moreover A (and hence B) is nuclear.
The ideal system is the same as that from A because h is ideal system preserv-
ing.
For a proof of Theorem I it remains to show:

(i) There is a conditional expectation from the hereditary C*-subalgebra D of

P (A) generated by A ⊗ p onto A ⊗ p ∼ = A, and
976 11. PI-SUN ALGEBRAS REPRESENT THE NUCLEAR KK-CLASSES

(ii) the inclusion map A ∼= A⊗p ⊂ B ∼ = h∞

1 (B) ⊂ P (A) defines a KK-
equivalence between A and P (A).

The projection p = s2 s∗2 is in O∞ st = (1 − s1 s∗1 )O∞ (1 − s1 s∗1 ) ⊂ O∞ . A ∼

=
A ⊗ p → A ⊗ O∞ st is a KK-equivalence, because Cp → O∞ st defines a KK-
equivalence between C and O∞ st and because one can tensor KK-equivalences (by
nuclear algebras). It can be seen from the the 6-term exact sequences and from
KK(., O2 ) = KK(O2 , .) = 0 that A ⊗ O∞ st → B defines a KK-equivalence. Thus
A ∼= A ⊗ p → B defines a KK-equivalence between A and B. The hereditary
subalgebra of B generated by A ⊗ p is A0 := A ⊗ pO∞ p ⊂ A ⊗ O∞ st ⊂ B. Thus
A ⊗ p is the range of a conditional expectation from the hereditary C*-subalgebra
A0 onto A ⊗ p.
Above we have seen that the proof of Theorem I, (i.e. of (i) and (ii) above)
reduces to the proof of the following:

(i*) h∞
1 (B) is the range of a conditional expectation on P (A), and
(ii*) h1,∞ : B → P (A) defines a KK-equivalence [h1,∞ ] ∈ KK(B, P (A)).

We show (i∗) by an induction procedure and then use homotopy invariance of

unsuspended E-theory R(B, P (A)) and R(P (A), B) to prove (ii∗ ). More precisely
we construct a unital asymptotic morphism g : P (A) → Cb (R+ , B) such that g◦h1,∞
is homotopic to idB and h1,∞ ◦g is homotopic to idP (A) . By the homotopy invariance
of Rørdam groups it follows that h1,∞ is a KK-equivalence.
Cuntz addition of morphisms is the same as the “direct” sum of morphisms
(after stabilization) and thus induces the sum for the corresponding KK-elements:
[idB ] + [h0 ] = [idB ⊕h0 ] = [h] and 2[h0 ] = [h0 ⊕ h0 ] = [h0 ]. It follows that
[h0 ] = 0 and [h] = [id] in KK(B, B).
Thus it suffices to show that the following statements (α) and (β) are true in
order to prove (i*) and (ii*):

(α) There is a family of conditional expectations Enm from h∞ ∞

m (B) onto hn (B)
for n < m such that Enm+k h∞ m ∞ k k ∞
m = En hm . and En Em hk = En hk for
k ∞

n < m < k.
They define conditional expectations En : P (A) → h∞ n (B) ⊂ P (A)
with En (P (A)) = h∞ n (B) and E E
n m = E min(n,m) .
(β) There exists a completely positive unital map V : P (A) → Cb (R+ , B) such
that V (b∗ b) − V (b)∗ V (b) ∈ C0 (R+ , B) for every b ∈ P (A), and [(π ◦ V ) ◦
h∞ ∞
1 ] = [h] ∈ KK(B, B), [h1 ◦ (π ◦ V )] = [idP (A) ] ∈ KK(P (A), P (A)).

The completely positive maps V , V ◦ h∞ ∞

1 and h1 ◦ V in statement (β) de-
fine elements of Ext−1 (P (A), SB), Ext−1 (B, SB) and Ext−1 (P (A), SP (A)) and [.]
denote the corresponding elements of KK(., .) ∼= Ext−1 (., S(.)).
 2. PRELIMINARIES FOR THE PROOF OF (β) 977

1. Proof of (α)

We start with the proof of (α). The proof of (β) is conceptional quite simple
but requires some preparation.

Proof of (α): By induction we find (sn )∗ hn (b)sn = b for b ∈ B and n =

1, 2, . . .:

b = s∗ (sbs∗ + thu0 (b)t∗ )s = s∗ h(b)s = s∗ (sn )∗ hn (h(b))(sn )s = (sn+1 )∗ hn+1 (b)sn+1 .

Let Tn be given by Tn (b) := hn ((sn )∗ bsn ) and T0 = idB . Then Tn is a condi-

tional expectation from B onto hn (B), because Tn (B) ⊂ hn (B) and Tn (b) = b for
b in hn (B), because (sn )∗ hn (b)sn = b.
The conditional expectations Tn satisfy Tn+k hk = hk Tn and Tn Tm = Tmin(n,m) :
The first equation follows from hn+k ((sn )∗ (sk )∗ hk (b)sk sn ) = hk (hn ((sn )∗ bsn )).
Certainly Tn Tm = Tm if n ≤ m if n ≤ m and for m < n we have

Tn Tm (b) = hn ((sn )∗ hm ((sm )∗ bsm )sn ) = hn ((sn−m )∗ (sm )∗ bsm sn−m ) = Tn .

Now we define conditional expectations Enm (n < m) from h∞

m (B) onto
∞ ∞ m−n
hn (B) = hm (h (B)) by

Enm (b) := h∞ ∞ −1
m Tm−n (hm ) (b)

for b in h∞ m ∞ ∞
m (B), i.e. by En hm = hm Tm−n .

Then Enm h∞ k ∞
k = En hk for n < m and n < k ≤ m because

Enm h∞ m ∞ m−k
k = En hm h = h∞
m Tm−n h
m−k
= h∞
mh
m−k
Tk−n = h∞ k ∞
k Tk−n = En hk .

k
If n < m < k we have Enk Em = Enk because
k ∞
Enk Em hk = Enk h∞ ∞ ∞ k ∞
k Tk−m = hk Tk−n Tk−m = hk Tk−n = En hk .

The consistency of the Enm implies the existence and uniqueness of conditional
expectations En : P (A) → h∞ ∞
n (B) from P (A) onto hn (B) such that En hk =
∞

m ∞
En hk for n ≤ k ≤ m. Then En Em = Em if m ≤ n.
Let n < m < k ∈ N then

En Em h∞ k ∞ k k ∞ k ∞ ∞
k = En Em hk = En Em hk = En hk = En hk .

Thus En (P (A)) = h∞
n (B), En Em = Emin(n,m) and (α) is shown. 

2. Preliminaries for the proof of (β)

We need some observations for the proof of (β). They are based on elementary
properties of O2 and of Proposition 11.2.2 (which is related to Theorem B).
978 11. PI-SUN ALGEBRAS REPRESENT THE NUCLEAR KK-CLASSES

Lemma 11.2.1. Let A be a unital C*-algebra, then the unitary group of A ⊗ O2

is simply connected.
In particular, any continuous map u : ∂([0, 1] × [0, 1]) → U (O2 ) from the bound-
ary of the square [0, 1] × [0, 1] into the unitaries of O2 extends to a continuous map
from the square into the unitaries of O2 .

Proof. Let F := A ⊗ O2 . We consider O2 = ∼ 1 ⊗ O2 as a unital subalgebra of

F with generators {s, t} (to simplify notation). The unital endomorphism δ(a) =
sas∗ + tat∗ = a ⊕ a, is point-norm homotopic inside the endomorphisms of O2 to
idO2 , by [172, prop.2.2], O2 is K1 -injective by [172, thm.1.9], and K1 (O2 ) = 0 by
[172, thm.3.8].
Thus, U(O2 ) = U0 (O2 ), and it follows that idA ⊗δ is homotopic to idF and
u ∼h u ⊕ u in U(F ).
Notice that v ⊕ 1 is unitarily equivalent to 1 ⊕ v by the selfadjoint unitary
Uc := ts∗ +st∗ in M2 ⊂ O2 and (u1 ⊕u2 )⊕u3 is unitarily equivalent to u1 ⊕(u2 ⊕u3 )
by the unitary Ud ∈ U(O2 ) = U0 (O2 ), defined by equation (2.4), see Lemma 4.2.6(o)
or Proposition 4.3.2.
Since the unitary group of O2 is connected by [172, thm.1.9, thm.3.8], we get
that the unitary Ud ∈ O2 defined by equation (2.4) is homotopic to 1 in E. Together
we get the following homotopies in the unitary group of F :
u ∼ u ⊕ u = (u ⊕ 1)(1 ⊕ u) ∼ u2 ⊕ 1, and with v = u2

1 = (v ⊕ 1)(v ∗ ⊕ 1) ∼ v ⊕ v ∗ ∼ (v ⊕ v) ⊕ v ∗ ∼ v ⊕ (v ⊕ v ∗ ) ∼ v ⊕ 1 = u2 ⊕ 1.

Thus u ∼ 1.
For T := ∂([0, 1]×[0, 1]) and D := [0, 1]×[0, 1] this implies that the epimorphism
C(D, O2 ) → C(T, O2 ) maps the unitary group of C(D, O2 ) onto the unitary group
of C(T, O2 ) ∼
= C(T ) ⊗ O2 . 

The following Proposition 11.2.2 generalizes some aspects of Theorem B.

The unitary equivalence modulo C0 (R+ , C(X, D)) implies that k and h define
the same element of Extnuc (A, S C(X, D)).
Thus [h] = [k] in KKnuc (A, C(X, D)).
Parts (i) and (ii) of Proposition 11.2.2 simply mean that h ⊕ hu0 , h and k
are unitarily homotopic in C(X, D) if h(A) and k(A) are contained in C(X, D) ⊂
Cb (R+ , C(X, D)).
By πy : Cb (Y, D) → D we denote the evaluation at the point y ∈ Y = R+ × X.

Proposition 11.2.2. Suppose that X is a compact space, D is a unital simple

purely infinite C*-algebra which contains a copy of O2 unitally, that A a unital
separable exact C*-algebra, and that h, k : A → Cb (R+ × X, D) ∼
= Cb (R+ , C(X, D))
are unital C*-morphisms with the property that, for every y ∈ R+ × X, πy h and
πy k are monomorphisms. Then:
 2. PRELIMINARIES FOR THE PROOF OF (β) 979

(i) There is a unitary U0 ∈ Cb (R+ , C(X, D)) such that (h⊕hu0 )(a)−U0∗ h(a)U0
is in C0 (R+ , C(X, D)) for every a ∈ A.
(ii) If, moreover, h and k are nuclear and [h] = [k] in KKnuc (A, C(X, D)),
then there exist a unitary U ∈ Cb (R+ , C(X, D)) such that h(a) − U ∗ k(a)U
is in C0 (R+ , C(X, D)) for every a ∈ A.
(iii) For ϕ : A ⊗ K → C(X, D) ⊗ K with [ϕ(1 ⊗ p11 )] = 0 in K0 (C(X, D))
there exits a unital C*-morphism h : A → C(X, D) such that πy h is a
monomorphism for every y ∈ X and h ⊗ idK and ϕ ⊕ h0 (where h0 = hu0 ⊗
idK ) are unitarily homotopic. h is nuclear if ϕ is nuclear. In particular,
every z ∈ KKnuc (A, C(X, D)) with [1A ] ⊗A z = [1C(X,D) ] = 0 is of the
form z = [h] for a nuclear h of the type above.

Proof. Ad(i): Let Q(R+ , C(X, D)) := Cb (R+ , C(X, D))/ C0 (R+ , C(X, D))
and let π denotes the quotient map π : Cb (R+ , C(X, D)) → Q(R+ , C(X, D)).
Cb (R+ × X, D) ∼
= Cb (R+ , C(X, D)) and C0 (R+ × X, D) ∼
= C0 (R+ , C(X, D)) by
a natural isomorphism.
Consider C(X, D) as a unital C*-subalgebra of Q(R+ , C(X, D)). Then the
relative commutant hu0 (A)0 ∩ Q(R+ , C(X, D)) contains a copy of O2 unitally (by
the definition of hu0 in Chapter 1 before Theorem B).
Thus by Proposition 4.3.5 it suffices to show that πh : A → Q(R+ × X, D)
dominates hu0 : A → O2 ⊂ Q(R+ × X, D). But this follows from Corollary 7.4.6,
because the lift hu0 : A → O2 ⊂ Cb (R+ × X, D) is nuclear and the evaluations πy h
at y ∈ R+ × X of the lift h : A → Cb (R+ × X, D) of πh are unital monomorphisms.
Ad(ii): By the proof of Theorem B(i) and (ii) in Chapter 9 we have for
h0 := hu0 ⊗ idK that the morphisms h ⊗ idK ⊕h0 and k ⊗ idK ⊕h0 from A ⊗ K into
Cb (R+ , C(X, D) ⊗ K) are unitarily equivalent modulo C0 (R+ , C(X, D)) ⊗ K Since
h ⊗ idK ⊕h0 = (h ⊕ hu0 ) ⊗ idK the argument before Corollary C in Chapter 1 shows
that this implies that h⊕hu0 and k ⊕hu0 are unitarily equivalent in Cb (R+ , C(X, D))
modulo C0 (R+ , C(X, D)). By the assumptions on the πy h and πy k we get from
part(i) that moreover h and k are unitarily equivalent in Cb (R+ , C(X, D)) modulo
C0 (R+ , C(X, D)).
Ad(iii): By Theorem B(i) there exists for z ∈ KKnuc (A, C(X, D)) a nuclear C *-
monomorphism ϕ : A ⊗ K → C(X, D) ⊗ K with [ϕ] = z, because C(X, D) contains
a unital copy of O2 .
Then [ϕ(1 ⊗ p11 )] = z ⊗A [1A ] = 0 in K0 (C(X, D)).
[(ϕ ⊕ h0 )] = [ϕ] + [h0 ] = [ϕ] in KK(A, C(X, D)) because 2[h0 ] = [h0 ].
Since [ϕ(1 ⊗ p11 )] = 0 in K0 (C(X, D)), there is a unitary V in M (C(X, D) ⊗ K)
such that

1 ⊗ p11 = V ∗ (ϕ(1 ⊗ p11 ) ⊕ (1 ⊗ p11 ))V = V ∗ (ϕ ⊕ h0 )(1 ⊗ p11 )V.

980 11. PI-SUN ALGEBRAS REPRESENT THE NUCLEAR KK-CLASSES

Thus there is a unital C *-morphism h : A → C(X, B) such that, for a ∈ A

h(a) ⊗ p11 = V ∗ (ϕ ⊕ h0 )(a ⊗ p11 )V.

A ⊗ p11 is a full corner of A ⊗ K. Therefore, h ⊗ idK is unitarily homotopic to

∗
V (ϕ ⊕ h0 )V and thus to ϕ ⊕ h0 . If we apply the evaluation maps πy ⊗ idK to this
unitary homotopy, then we see that πy h is a monomorphism for every y ∈ X.
From the definition of a unitary homotopy it follows that a morphism is nuclear,
if it is unitarily homotopic to a nuclear morphism. 

Let X be a compact space and D a unital C *-algebra. Then cone(X) means

the one-point compactification of of R+ × X.
By the isomorphism [0, 1) ∼ = R+ , the C *-algebra C(cone(X), D) is the unital
∼
subalgebra of C([0, 1] × X, D) = C([0, 1], C(X, D)) which is naturally isomorphic to
the unital split extension of C0 (R+ , C(X, D)) by D ⊂ C(X, D).
Now we consider the case D = O2 .

Corollary 11.2.3. Suppose that X is a compact space, B is a unital separable

exact C*-algebra, and h : B → C(X, O2 ) is a unital C*-morphism such that πy h is
a monomorphism for every y ∈ X.
Then there exists k : B → C(cone(X), O2 ) with πy k = πy h for every y ∈ X ∼
=
X × {0} and πy k is a monomorphism for every y ∈ cone(X).

Proof. Since KK(A, C(X) ⊗ O2 ) = 0, h is unitarily homotopic to hu0 by

Proposition 11.2.2(ii). This homotopy yields a strongly continuous family of mor-
phisms k1 (τ, y) : B → O2 (τ ∈ [0, 1/2]) such that k1 (0, y) = h(y) := πy h for
y ∈ X. And there is a continuous map u : (0, 1/2] × X → U (O2 ) such that
k1 (τ, y) = u(τ, y)∗ hu0 u(τ, y) for y ∈ X and τ ∈ (0, 1/2]. The unitary v ∈ C(X, O2 )
with v(y) := u(1/2, y) is homotopic to 1 in the unitaries of C(X, O2 ) ∼ = C(X) ⊗ O2
by Lemma 11.2.1.
Thus there is an extension w of u to (0, 1] × X with w(1, y) = 1 for y ∈ X and
w(τ, y) = u(τ, y) for y ∈ X, τ ∈ [0, 1/2], k(τ, y) := k1 (τ, y) for y ∈ X, τ ∈ [0, 1/2],
k(τ, y) := w(τ, y)∗ h0 (.)w(τ, y) for y ∈ X, τ ∈ (1/2, 1] is the desired extension of h
to cone(X). 

Lemma 11.2.4. (i) If ϕ : B → Cb (R+ , O2 ) is a unital C*-morphism such

that πτ ϕ is a monomorphism for every τ ∈ R+ then there exists a unitary
U ∈ Cb (R+ , O2 ) such that ϕ(a) − U ∗ hu0 (a)U ∈ C0 (R+ , O2 ) for every
a ∈ B, where hu0 : B → O2 is as defined before Theorem B in Chapter 1.
In particular, every unital *-monomorphism k : B → O2 is unitarily
homotopic to hu0 .
(ii) Suppose that h := idB ⊕s,t hu0 , and that uk are unitaries in O2 with uk s =
sk+1 , u0 = 1.
Then hk+1 = uk (idB ⊕s,t hk )u∗k , where hk : B → O2 is a unital C*-
morphism.
 2. PRELIMINARIES FOR THE PROOF OF (β) 981

In particular, h2 is unitarily homotopic to h by unitaries τ ∈ [0, 1) 7→

u(τ ) ∈ O2 such that u(1/2) = u1 , u(0) = 1, u(1/2 + τ ) = u1 (1 ⊕s,t
v(2τ )) for τ ∈ [0, 1/2), where h1 (b) = limτ →1 v(τ )h0 (b)v(τ )∗ is a unitary
homotopy between h1 and h0 with v(x) ∈ O2 and v(0) = 1.
(iii) For k ∈ N there exist unitaries vk ∈ O2 with vk (sk ) = s.
If k ≥ 2 and vk is a unitary in O2 with vk (sk ) = s, then there is exactly
one unital *-monomorphism ψk : B → O2 such that hk = vk∗ (idB ⊕ψk )vk ,
where Cuntz addition is taken with {s, t}.
(iv) Let u(t) as in (ii), and define a unital C*-morphism ϕ from B into
Cb (R+ , B) by ϕ(b)(t) := h(b) for t ∈ [0, 1] ϕ(b)(n) := hn (b) for n =
1, 2, . . . and ϕ(b)(n + t) := u(t)ϕ(b)(n)u(t)∗ for n = 1, 2, . . ., t ∈ [0, 1).
Suppose n + k ≥ 1 and that uk+n+2 are unitaries in O2 with
uk+n+1 · sk+n+1 = s and uk+n+2 · sk+n+2 = s.
Then there are a continuous map u : [0, 1] → O2 into the unitaries of O2
and a strongly continuous map ψ : [0, 1] → Hom(B, O2 ) ⊂ L(B) into the
unital C*-morphisms from B into O2 such that u(0) = uk+n+1 , u(1) =
uk+n+2 and
hk ϕ(1 + τ )hn = u(τ )∗ (idB ⊕(h0 ⊕ ψ(τ )))u(τ ).

Proof. Ad(i): Since KKnuc (B, O2 ) = KK(B, O2 ) = 0, (i) follows as a special

case from Proposition 11.2.2(ii) where X= point, k = hu0 and D = O2 .
Ad(ii): Above we have seen that (sk )∗ hk (b)sk = b, cf. construction of En .
Suppose that hk−1 : B → O2 and uk−1 ∈ O2 with hk = uk−1 (id ⊕s,t hk−1 )u∗k−1
are given.
Let h0k := h0 ⊕s,t (hk−1 ◦ h). Then h0k is a unital morphism from B into O2 .
We have
hk+1 (b) = uk−1 (s2 b(s∗ )2 + sth0 (b)t∗ s∗ + t(hk−1 ◦ h)(b)t∗ )u∗k−1 .

Let w be the unitary in O2 with ws2 = s, wst = ts, wt = t2 .

Then (id ⊕s,t hk−1 )h = h ⊕s,t (hk−1 ◦ h) = (idB ⊕s,t h0 ) ⊕s,t (hk−1 ◦ h) =
w∗ (idB ⊕s,t (h0 ⊕s,t hk−1 ◦ h))w = w∗ (idB ⊕s,t h0k )w.
It follows hk+1 = uk−1 w∗ (idB ⊕s,t h0k )w(uk−1 )∗ .
On the other hand, uk−1 w∗ s = sk+1 = uk s and uk−1 , uk , w ∈ O2 .
Thus uk = uk−1 w∗ (1 ⊕s,t v) for some unitary v ∈ O2 . Now let hk := v ∗ h0k v.
Ad(iii): There exist unitaries vk in O2 with vk (sk ) = s, because sk (sk )∗ and
1 − sk (sk )∗ are all Murray–von Neumann equivalent for k ∈ N, by [172].
The existence of ψk follows from (ii): Let ψk := hk−1 for uk−1 := (vk )∗ .
ψk is uniquely determined by ψk (b) = t∗ vk hk (b)(vk )∗ t.
Ad(iv): By definition of ϕ, we have ϕ(1 + τ )(b) = u(τ )∗ (b ⊕ h(τ )(b))u(τ ),
where u : [0, 1] → O2 is a continuous map into the unitaries of O2 with u(0) = 1,
982 11. PI-SUN ALGEBRAS REPRESENT THE NUCLEAR KK-CLASSES

u(1)s2 = s, and h : [0, 1] → Hom(B, O2 ) is a strongly continuous map into the

unital *-monomorphisms from B into O2 with h(0) = h0 .
If u1 , u2 in O2 are unitaries with u∗0 s = sk , u∗1 s = sk+1 such that

λ(τ )(b) = u(τ )∗ (b ⊕ ψ(τ )(b))u(τ ) (∗ )

where ψ : τ 7→ ψ(τ ) ∈ Hom(B, O2 ), and u : τ 7→ u(τ ) ∈ U (O2 ) are chosen with

u(0)∗ s = sk , u(1)∗ s = sk+1 then u(0) = (1 ⊕ v0 )u0 , u(1) = (1 ⊕ v1 )u1 , where v0
and v1 are unitaries in O2 .
Let v : [0, 1] → U (O2 ) be continuous with v(0) = v0 , v(1) = v1 , then χ(τ )(b) =
w(τ )∗ (⊕χ(τ )(b))w(τ ) where χ(τ )(b) := v(τ )∗ ψ(τ )(b)v(τ ), w(τ ) := (1 ⊕ v(τ ))∗ u(τ ).
Thus in all cases in question it suffices to find a decomposition (*) for at least
one pair (ψ(τ ), u(τ )) for every τ ∈ O2 , such that ψ(τ ) is a unital monomorphism and
u(0)sk = s, u(1)sk+1 = s. This is the case for λ(τ ) := ϕ(1 + τ ) by definition of ϕ. If
λ(τ ) has the decomposition (*) of the desired type, then (by a simple computation)
hλ(τ ) = u1 (τ )∗ (b ⊕ ψ1 (τ )(b))u1 (τ ) and λ(τ ) = u2 (τ )∗ (b ⊕ ψ2 (τ )(b))u2 (τ ) where
u1 (τ ) := w(u(τ ) ⊕ 1), u2 (τ ) := wu(τ ), ψ1 (τ ) := ψ(τ )(b) ⊕ h0 (λ(τ )(b)), ψ2 (τ ) :=
h0 (b)⊕ψ(τ )((b)) and w := s(t∗ )2 +tst∗ s∗ +t2 s∗ . Thus by induction we get (iv). 

Now we define a unital C *-morphism ϕ : B → Cb (R+ , B) by ϕ(b)(t) := h(b)

for t ∈ [0, 1] ϕ(b)(n) := hn (b) for n = 1, 2, . . . and ϕ(b)(n + t) := u(t)ϕ(b)(n)u(t)∗
for n = 1, 2, . . ., t ∈ [0, 1), where u(t) is as in Lemma 11.2.4(ii). The definition fits
because u(0) = 1.
Then by the above Lemma 11.2.4(iv) we have ϕ(b) = w(b ⊕s,t ψ(b))w∗ , where
w ∈ Cb (R+ , O2 ) is defined inductively by w(t) = 1 for t ∈ [0, 1], w(n+t) = u(t)w(n)
for n = 1, 2, . . ., t ∈ [0, 1) (note that u(0) = 1 ), w(n + t) = u(1/2)w(n) for
n = 1, 2, . . ., t ∈ (1/2, 1] and where ψ(b) := t∗ w∗ ϕ(b)wt is a unital C *-morphism
from B into O2 ⊂ B.
Thus by Lemma 11.2.4(i) there exists a unitary v ∈ Cb (R+ , O2 ) such that
ψ(b) − vhu0 (b)v ∈ C0 (R+ , O2 ) for every b ∈ B and thus ϕ(b) − (w(1 ⊕ v))h(b)(w(1 ⊕
v))∗ is in C0 (R+ , O2 ) for every b in B, and w(1 ⊕ v) is a unitary in Cb (R+ , O2 ).

Lemma 11.2.5. There exist unital C*-morphisms

θn : B → C([0, 1] × [0, 1], B)

that satisfy the following boundary conditions for σ ∈ [0, 1], τ ∈ [0, 1]:
θn (0, σ) = hϕ(1 + nσ), θn (1, σ) = hϕ(1 + (n + 1)σ),
θn (τ, 0) = h2 , and θn (τ, 1) = hn+1 ϕ(1 + τ ).

Proof. By Lemma 11.2.1 and Corollary 11.2.3 it is enough to show that there
is a continuous map u : ∂([0, 1]×[0, 1]) → O2 into the unitaries of O2 and a strongly
continuous map h : ∂([0, 1] × [0, 1]) → Hom(B, O2 ) ⊂ L(B, O2 ) into the unital *-
monomorphisms from B into O2 , such that the boundary conditions on θn are given
by ψ : ∂([0, 1] × [0, 1]) → Hom(B, B) with ψ(b)(τ, s) = u(τ, s)∗ (b ⊕ h(τ, s)(b))u(τ, s)
 3. PROOF OF (β) 983

for (t, s) ∈ ∂([0, 1] × [0, 1]) because then θn (b)(τ, s) := w(τ, s)∗ (b ⊕ k(τ, s)(b))w(τ, s),
(τ, s) ∈ [0, 1] × [0, 1] is as desired, where w : [0, 1] × [0, 1] → U (O2 ) is a continuous
extension of u and k : [0, 1] × [0, 1] → Hom(B, O2 ) is a strongly continuous map
from [0, 1] × [0, 1] ∼= cone(∂([0, 1] × [0, 1])) into the unital *-monomorphisms from
B into O2 such that h = k|∂([0, 1] × [0, 1]).
On the sides of the square [0, 1] × [0, 1], the boundary conditions are given
by strongly continuous maps into Hom(B, B), which coincide on the vertices: h2
in (0, 0) and (1, 0) , hn+2 in (0, 1) and hn+3 in (1, 1). By the construction of ϕ
we have ϕ(1 + k + ξ) = ϕ(1 + ξ)hk for ξ ∈ [0, 1]. Thus we find m = 2n + 3
points z1 , z2 , . . . , zm on ∂([0, 1] × [0, 1]) (containing the vertices z1 = (0, 0), z2 =
(1, 0)) such that λk (τ ) := θn (zk+1 + τ (zk+1 − zk )) is one of the following functions:
λ1 (τ ) ≡ h2 , λ2 (τ ) = ϕ(1 + τ )h, λ3 (τ ) = ϕ(1 + τ )h2 , . . . , λn+2 (τ ) = ϕ(1 + τ )hn+1 ,
λn+3 (τ ) = hn+1 ϕ(2 − τ ), λn+4 (τ ) = hϕ((n + 1) − τ ) = hϕ(2 − τ )hn+1 , . . . ,
λ2n+3 (τ ) = hϕ(2 − τ ).
Now apply Lemma 11.2.4(iv) to this function to get the above desired particular
form.


3. Proof of (β)

We construct in some sense a “completely positive liftable unsuspended E-

equivalence” as an element of R(P (A), B) , given by by a path of u.c.p. maps
Vt : P (A) → B with the property that Vt ◦ h1,∞ : B → B is equivalent in R(B, B)
to h : B → B. for the maps h1,∞ : B → P (A) and h : B → B:
Since h : B → B is a KK-equivalence, (and is equivalent to idB in R(B, B)), it
is sufficient to define a unital completely positive map V : P (A) → Cb (R+ , B) with
the following properties:

(i) V (b∗ b) − V (b)∗ V (b) ∈ C0 (R+ , B) for every b ∈ P (A), i.e., V defines a
unital and completely positively liftable homomorphism from P (A) into
Q(R+ , B).
(ii) There are a *- homomorphism ϕ1 : B → Cb (R+ , B) and a unitary U ∈
Cb (R+ , O2 ) such that ϕ1 (a) − U ∗ h(a)U ∈ C0 (R+ , B)
(Is not necessary! “Stable” homotopy of ϕ1 with h : B → B is
enough.)
and
{ϕ1 (a)(t) − V (h1,∞ (a))(t)} ∈ C0 (R+ , B) for a ∈ B, τ ∈ R+ .
(It gives – by R(C, D) = KK(C, D) for unitally O∞ -containing C,D
– [V ] ◦ [h1,∞ ] = [h] = [idB ] in KK(B, B))
This is equivalent to V ◦ h1,∞ homotopic to U ∗ h(·)U for suitable path
U (t). Thus [h] = [V ] ◦ [h1,∞ ] in KK(B, B)
(iii) Need a proof that h1,∞ (V (·)(t)) is homotopic to [idP (A) ] in R(P (A), P (A)).
(A kind of “reconstruction”.)
984 11. PI-SUN ALGEBRAS REPRESENT THE NUCLEAR KK-CLASSES

(iv) Another way to describe it would be the following:

There is unital completely positive W : P (A) → Cb (R+ ×[0, 1], P (A)),
such that limτ →∞ kW (b∗ b)(τ, t) − W (b)(τ, t)∗ W (b)(τ, t)k = 0
and
limτ →∞ kW (b)(τ, 1) − bk = 0
and
limτ →∞ kW (b)(τ, 0) − h1,∞ (V (b)(τ ))k = 0.

This means that [π ◦ V ] ∈ Ext−1 (P (A), SB) ∼ = KK(P (A), B) has the property
that
[h∞
1 ] ⊗B [π ◦ V ] = [idP (A) ] ∈ KK(P (A), P (A))

and
[π ◦ V ] ⊗P (A) [h∞ ∼
1 ] = [h] = [idB ] ∈ KK(B, B).

Thus Theorem I is proved if we have shown the existence of V , ϕ1 , U and W

with the above listed properties (i)-(iii).
We define the unital completely positive map V : P (A) → Cb (R+ , B) by the
following constructions.
For n = 1, 2, . . . let ϕn (b)(τ ) := ϕ(b)(τ − n) for τ ≥ n and ϕn (b)(τ ) := h(b) for
0 ≤ τ < n. Then ϕn (n + k) = hk (b), ϕn (b)(n + τ ) = h(b) for k = 1, 2, . . ., τ ∈ [0, 1]
and ϕn+k (hk (b))(τ ) = ϕn (b)(τ ) for τ ≥ n + k. We let Vn := ϕn (h∞ n )
−1
En .
Then Vn h∞ ∞ ∞
n = ϕn , and we have Vm (hn (b))(τ ) = Vn (hn (b))(τ ) for m > n,
τ ≥ m:
Vm ◦ h∞ ∞ m−n
n = Vm ◦ hm h = ϕm hm−n .
For the parameter τ ≥ m we get ϕm (hm−n (b))(τ ) = ϕn (b)(τ ) = Vn (h∞
n (b))(τ ). Let

V (d)(τ ) := (n + 2 − τ )Vn (d)(τ ) + (τ − n − 1)Vn+1 (d)(τ )

for n + 1 ≤ τ ≤ n + 2 and V (d)(τ ) := V1 (d)(τ ) for 0 ≤ τ ≤ 2.
The definition is fits, because on the boundary points of the intervals in question
we have V (d)(n+1) = Vn (d)(n+1) and V (d)(n+2) = Vn+1 (d)(n+2) for n = 1, 2, . . ..
Then V : P (A) → Cb (R+ , B) is a unital completely positive map and
V (h∞
n (b))(τ ) = ϕn (b)(τ )

for τ ≥ n+1, n ≥ 1, because if τ ≥ n+1 then there is an m ∈ N with n+1 ≤ m+1 ≤

τ ≤ m+2 and for this m we get V (h∞ ∞
n (b))(τ ) = (m+2−τ )Vm (hn (b))(τ )+(τ −m−
1)Vm+1 (h∞ ∞
n (b))(τ ) and the right hand side is equal to Vn (hn (n))(τ ) = ϕn (b)(τ ).
S ∞
In particular, since P (A) is the closure of hn (B), we get that V is asymp-
totically multiplicative, i.e., V (d∗ d) − V (d)∗ V (d) ∈ C0 (R+ , B) for d ∈ P (A).
Thus V is an asymptotic morphism (which is not nuclear if B is not nuclear)
from P (A) in B with V (h∞ 1 (b)) − ϕ1 (b) ∈ C0 (R+ , B). As we have seen above, ϕ1
(as a scaling of ϕ) is unitarily equivalent to h : B → B ⊂ Cb (R+ , B) by a unitary
U in Cb (R+ , O2 ). Thus
[V ] ⊗P (A) [h∞
1 ] = [h] = [idB ]
 3. PROOF OF (β) 985

in KK(B, B).
We construct an asymptotic morphism

W : P (A) → Cb (R+ × [0, 1], P (A))

with

W (d)(τ, 0) − h∞
1 (V (d)(τ )) ∈ C0 (R+ , P (A)) and W (d)(τ, 1) − d ∈ C0 (R+ , P (A)).

We use the unital C *-morphisms θn : B → C([0, 1] × [0, 1], B) of Lemma 11.2.5.

Let us list some equations for V , h∞
n , ϕ and θn that we have shown above or
are obtained by straight forward calculations from the definitions:
Just from the definition of ϕ, ϕn , V and from V (hn (b))(n+1+σ) = ϕn (n+1+σ)
we get ϕ(1 + σ)(b) = V (h∞ n (b))(n + 1 + σ)) for σ ∈ [0, 1].

Then we need moreover the equations h∞ k

m+k (h (b)) = hm (b) and the boundary
conditions listed in Lemma 11.2.5.
From the boundary conditions it follows that h(θn (b)(1, τ )) = θn+1 (h(b))(1, τ ).
At first we define unital, completely positive maps Wn .
Let Wn : P (A) → C([n + 1, n + 2] × [0, 1], P (A)) be defined for τ, σ ∈ [0, 1] by

Wn (d)(n + 1 + σ, τ ) := h∞ ∞ −1
n+2 (θn ((hn ) En (d))(σ, τ )).

It implies
Wn (h∞ ∞
n (b))(n + 1 + σ, τ ) = hn+2 (θ( b)(σ, τ )),

and from En (P (A)) = h∞ n (B) and the above listed equations (and the boundary
conditions of θn in Lemma 11.2.5) the reader easily get by straight forward calcu-
lations (by replacing En (d) by h∞n (b)) the following three equations (where n ∈ N,
d ∈ P (A), and σ, τ ∈ [0, 1]):

Wn (En (d))(n + 1 + σ, 0) = En (d),

Wn (En (d))(n + 1 + σ, 1) = h∞
1 (V (En (d))(n + 1 + σ))

and
Wn (En (d))(n + 2, τ ) = Wn+1 (En (d))(n + 2, τ ).

The latter equation shows that

W (d)(n + 1 + σ, τ ) := Wn ((1 − σ)En−1 (d) + σEn (d))(n + 1 + σ, τ )

for n = 2, 3, . . . and σ, τ ∈ [0, 1] and W (d)(η, σ) := W2 (E1 (d))(3, σ) for τ, σ ∈ [0, 1],
0 ≤ η ≤ 3 is a well-defined completely positive and unital map from P (A) into
Cb (R+ , P (A)).
Since the union of the En (P (A)) is dense in P (A) we see from first two the
above listed three equations for the Wn ’s that

lim kW (d)(σ, 0) − dk = 0
σ→∞
986 11. PI-SUN ALGEBRAS REPRESENT THE NUCLEAR KK-CLASSES

and
lim kW (d)(τ, 1) − h∞
1 (V (d)(τ ))k = 0
σ→∞
for every d ∈ P (A).
This means [h∞ 1 ◦ (π ◦ V )] = [id] in KK(P (A), P (A)). On the other hand
∞ ∞
[h1 ◦ (π ◦ V )] = [h1 ] ⊗B [V ].
This completes the proof of (β) and ends the proof of Theorem I.
 4. EXAMPLES VIA CUNTZ-PIMSNER ALGEBRAS 987

4. Examples via Cuntz-Pimsner algebras

The following (X, G)-equivariant construction of generalized Fock–Toeplitz al-

gebra (respectively of a Cuntz–Krieger–Pimsner algebra) using a suitable Hilbert
bi-module.

Theorem 11.4.1. Text from lectures on equivariant reconstruction.

Proof. 

Cite from Chapter 1: One starts with a suitable “universal” Hilbert bimodule
and builds the corresponding Cuntz–Krieger–Pimsner algebra.
The universality of the construction has the advantage that G-actions of locally
compact groups G on (Ax )x∈X lead in a natural way to G-actions on (P (Ax ) ⊗
K)x∈X
Next taken from Toronto 2014 lectures
Definition (in Chp.1??) of regular subalgebras:
Let C ⊂ A a C *-subalgebra. C is regular for A if

(i) C separates the ideals J of A: J1 ∩ C = J2 ∩ C implies J1 = J2 .

(ii) C ∩ (J1 + J2 ) = (C ∩ J1 ) + (C ∩ J2 ) for all J1 , J2 ∈ I(A).

Theorem 11.4.2 (Realization of Ψ, H.H.,E.K.). Suppose that B is separable

and stable. Let Ψ : I(B) → I(A) a non-degenerate lower s.c. action of Prim(B) on
A.
If B ⊗ O2 contains a regular abelian C*-subalgebra C then Ψ = ΨC for C := CΨ .
In particular, Ψ comes from a non-degenerate *-monomorphism h : A⊗K → M(B),
that is unique up to unitary homotopy of its infinite repeats.

Corollary 11.4.3 (Reconstruction Theorem, H.H.,E.K.). Suppose that A is a

nuclear and stable, that Ω is a sup–inf closed sub-lattice of I(A) ∼
= O(Prim(A)) with
Prim(A), ∅ ∈ Ω . Then there is a non-degenerate *-monomorphism H0 : A → M(A)
with following properties:

(i) The infinite repeat δ∞ ◦ H0 is unitarily equivalent to H0 .

(ii) For every U ∈ O(Prim(A)) holds H0 (J(V )) = H0 (A)∩M(A, J(U )) where
S
V ∈ Ω is given by V = {W ∈ Ω ; W ⊂ U }.

The H0 is uniquely determined up to unitary homotopy.

The uniques up to unitary homotopy means by Definition 5.0.1:

If H1 : A → M(A) also satisfies the conditions (i) and (ii) then there is a norm-
continuous path t ∈ R+ → U (t) ∈ U(M(A)) such that U (t)∗ H2 (a)U (t)−H0 (a) ∈ A
for all a ∈ A and t ∈ R+ and limt→∞ U (t)∗ H2 (a)U (t) = H0 (a).

Corollary 11.4.4 (Continuation of Reconstruction Theorem). The Cuntz-

Pimsner algebra OH of the Hilbert A-A-module H := (A, H0 ) is stable and strongly
purely infinite; and it is the same as the C*-Fock algebra F(H) of H.
988 11. PI-SUN ALGEBRAS REPRESENT THE NUCLEAR KK-CLASSES

The natural embedding of A into OH defines a lattice isomorphism from Ω onto

O(Prim(OH )) and is a KK(X; ·, ·)-equivalence for X := prime(Ω) ∼
= Prim(OH ).

Discussion of some general ways:

The algebra A will be changed to nice Cuntz “standard” form B := At up to
KK, such that O2 ⊆ M(B), α : A → B is KK(C ; A, B) a KK-equivalence (!!!!),
e.g. given by tensoring with O2 ⊗ K and “adding” O2 or O2 ⊗ K, depending on C.
Better would be: Can it be chosen stably homotopic to direct sum A ⊕s1 ,s2 O2
sitting in B = At ?
What is the inverse? Can stabilize ? Use C ⊗
b CP(O∞ )?

– after stabilization also an E-equivalence?? –

and the endomorphism γ := idB ⊕γ0 , γ0 , C-compatible “inner homotopic”: γ02
and γ0 ⊕γ0 “controlled” (inner?) homotopic to γ0 , γ should have has in KK(C; B, B)
the same class as idB .
C := indlimn (γn : B → B) with γn = γ, τ : C → Q([0, ∞), B) by intertwining
γn and γn+1 in suitable way and using conditional expectations onto ????,
given by moving the conditional expectations from C onto γn,∞ (B) to that of
with γ1,∞ : B → C, satisfies τ ◦ γ1,∞ h B in a controlled sense (“inner” sense ?),
with B (b) = b + B[0, ∞).
An idea would be convex combinations Ts := (n + 1 − s)Tn + (s − n)Tn+1 for
−1
s ∈ [n, n + 1] as maps from C into B[n, n + 1], where Tn := γ1,n ◦ Pn : C → B,
Pn : C → γ1,n (B) ⊂ C.
Thus, [B ] = [idB ] in E(B, B) ∼
= KK(C; B, B), [τ ] ◦ [γ1,∞ ] = [B ] = [idB ],
γ1,∞ : Q([0, ∞), B)
Does it explain the KK-equivalence ??
It gives classes X =???? ∈ KK(C ; C, B), Y ∈ KK(C ; B, C) inclusion, with
Y = [γ1,∞ ], X = [τ ] corresponds to asymptotic morphism from C to B.
that is KK(C; ·, ·) equivalent to A.
B has equivariant endomorphism that is KK(C; ·, ·) equivalent to idB , ...
Use the “fact” that certain kinds of ??? and a non-commutative variant of a
classical result of J. Milnor ???? given by L.G. Brown [?] . give simply ref to
Blackadar Rosenberg [1987, 1.12]. not found????
Countable additivity w.r.t. first variable: Blackadar’s K-theory book, [73,
thm. 19.7.1]
Continuity in first variable [73, thm. 21.5.2] with Milnor lim1 -sequence.
 CHAPTER 12

Non-commutative Selection and Proof of Thm. K

In this chapter we use the in Chapters 4 -11 confirmed results for a complete
proof of Theorems K (e.g. using Corollary 6.3.2) and Theorem O in Chapter 1
(e.g. using Theorem ??). As it was explained in the introductory Chapter 1,
this completes the proofs of all the results stated in Chapter 1, because we have
already proved Theorem M in Chapter 9 under the additional assumption, that there
exists h0 : A ,→ B with the properties that h0 is injective, non-degenerate, nuclear,
is unitarily homotopic to h0 ⊕ h0 , and realizes the given non-degenerate action
Ψ : I(B) → I(A) via Ψ(J) = h−1 0 (J ∩ h0 (A)). The existence of h0 is now the main
point of Theorem K, because we have seen the uniqueness of such h0 up to unitary
homotopy in Chapter 9. Moreover we have shown there, that, for σ-unital stable
B, there is — up to unitary equivalence (by unitaries in Q(R+ , M(B))) — at most
one nuclear *-monomorphism k0 : A → Q(R+ , B) with k0 approximately unitarily
equivalent (by unitaries in the bigger algebra M(Q(R+ , B))) to k0 ⊕ k0 , such that
k0 realizes the given action Ψ of Prim(B) on A via Ψ(J ∩ B) = k0−1 (J ∩ h0 (A)) for
any closed ideal J of Q(R+ , B). We have seen in Chapter 9, that this properties
of k0 imply that such k0 must be unitarily equivalent to some h0 : A → B of the
desired kind ( 1 ).
The Theorem 6.3.1 implies the existence of h0 : A → B (realizing Ψ) under
the additional assumptions that B is separable and contains a regular abelian C *-
subalgebra in the sense of Definition 1.2.9.
Thus, to complete the proof of Theorem K, we must show only the following
(but get more general) results:

(i) B can be replaced by a suitable separable subalgebra B0 of B and Ψ

by a suitable action Ψ0 of Prim(B0 ) on B0 , i.e., there exists a separable
C *-subalgebra B0 of B such that
(a) B0 is stable and contains a strictly positive element of B.
(b) Each ideal I of B0 is the intersection I = B0 ∩ J of an ideal J of B
with B0 .
(c) Ψ(J1 ) = Ψ(J2 ) if J1 ∩ B0 = J2 ∩ B0 , J1 , J2 ∈ I(B).
(d) B0 is strongly purely infinite.

1Combine Corollaries 9.1.6 and 9.1.4 – with the B in 9.1.4 replaced by our
k0 (A) Q(R+ , B)k0 (A) .

989
990 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

(2) If B is separable ( 2 ), then we show the existence of a separable C *-

subalgebra B1 ⊂ Q(R, B) such that
(e) B ⊂ B1 and B contains a strictly positive element of B1 (in partic-
ular, B1 is stable).
(f) Each ideal I of B1 is the intersection I = B1 ∩ J of an ideal J of
Q(R+ , B) with B1 .
(g) B1 contains a regular abelian C *-algebra C ⊂ B1 .
(h) B1 is strongly purely infinite.

If we consider B1 (in place of B) and the action Ψ1 (I) := Ψ(I ∩ B) of Prim(B1 ) on

A, then it turns out that the action Ψ1 is non-degenerate, lower semi-continuous and
monotone upper semi-continuous ( 3 ). Thus, Theorem 6.3.1 (respectively Corollary
6.3.2 — a special case of Theorem K) applies to (A, B1 , Ψ1 ) and gives that there is
nuclear
k0 : A ,→ B1 ⊂ Q(R+ , B0 ) ⊂ Q(R+ , B)
with k0 unitarily equivalent to k0 ⊕ k0 , and k0 (A) ∩ J = k0 (Ψ1 (B1 ∩ J)) = k0 (Ψ(J ∩
B)) for all closed ideals J of Q(R+ , B). Then Corollaries 9.1.6 and 9.1.4 imply
that there is non-degenerate nuclear h : A ⊗ O2 → B such that h0 := h((·) ⊕ 1) is
unitarily equivalent to k0 . Thus, h0 is as desired, and Theorem K follows from the
existence of (B0 , Ψ0 ) and (B1 , Ψ1 ) with the above listed (1a)–(2h).
The technical preparations for the existence of (B0 , Ψ0 ) and (B1 , Ψ1 ) lead also
to the asymptotic non-commutative ‘selection’ Theorem 12.1.8.
The below given characterization of the Dini functions on Prim(A), Proposition
12.2.6, the reduction result in Lemma 12.2.14, together with the results in the last
sections of Chapters 5 and 6 yield a proof of Proposition 12.2.15.
For notations we refer the reader to Chapter 1, Definitions 1.2.1–1.2.8.

1. Non-commutative asymptotic Selection

First we need some lemmata for our proof of the asymptotic non-commutative
Selection Theorem 12.1.8.

Lemma 12.1.1. Suppose that H : A → M(B) is a C*-morphism, where A and

B are σ-unital. For J ∈ I(B), let

Ψ(J) := H −1 (Ψup
B,H(A) (J)) := H
−1
(H(A) ∩ M(B, J)) .

(i) Ψ is a lower semi-continuous action of Prim(B) on A, i.e., it satisfies

parts (iii) and (iv) of Definition 1.2.6.
2The separability of B can be supposed, because we can replace B by B , and Ψ by Ψ (J) :=
0 0
Ψ(K) for arbitrary K .B with J = B0 ∩K. The properties (b) and (c) imply that Ψ0 is well-defined
and is lower s.-c. and monotone upper s.-c., because Φ(J) := span(BJB) is upper semi-continuous,
T T
satisfies B0 ∩ Φ(J) = J, and B0 ∩ τ Φ(Jτ ) = τ Jτ , and Ψ is lower s.-c. and monotone upper
s.-c.
3The l. s.-continuity and monotone u. s.-continuity of Ψ follows from the that continuity
1
properties of Ψ and of I 7→ I ∩ B.
 1. NON-COMMUTATIVE ASYMPTOTIC SELECTION 991

(ii) Ψ(0) = ker(H), and Ψ−1 (A) is the set of J ∈ I(B) with BH(A)B ⊂ J.
(iii) Ψ satisfies (ii) of Definition 1.2.6 if H(A) ⊂ B.
(iv) If C is a σ-unital C*-subalgebra of M(B), such that CB is dense in B,
H(A) ⊂ M(C) (i.e. H(A)C ⊂ C), and, for I ∈ I(C),

Φ(I) := H −1 (Ψup
C,H(A) (I)) = H
−1
(H(A) ∩ M(C, I)) ,

then, for J ∈ I(B),

Ψ(J) = Φ(Ψup
B,C (J)) .

(v) The morphism H : A → M(B) is weakly Ψ-residually nuclear if H satisfies

at least one of the following properties (a)–(c):
(a) H : A → M(C) is Φ-residually nuclear.
(b) A is exact and H : A → M(C) is weakly nuclear.
(c) C is nuclear.

Proof. (i)-(iv) follow straight from the definitions. The details are left to the
reader. E.g., (iv) follows from the easily poved identity

M(C) ∩ M(B, J) = M(C, C ∩ M(B, J))

for non-degenerate C *-subalgebras C of M(B).

Note that (iv) implies (v,a), because, for J ∈ I(B), I := Ψup
B,C (J), we have the
natural inclusions
[H] : A/Φ(I) ,→ M(C/I) ⊂ M(B/J),
the map a + Φ(I) 7→ b∗ (d∗ (H(a) + M(C, I))d)b is a nuclear map from A/Φ(I) into
B/J for b ∈ B/J, d ∈ C/I ⊂ M(B/J), and C/I is a non-degenerate C *-subalgebra
of M(B/J).
It is obvious that (v,c) implies (v,a). By Proposition ??, H : A → M(C)
is (norm-) nuclear if A is exact and H is weakly nuclear. Then all the quotient
maps [H] : A/Φ(I) → M(C)/M(C, I) ⊂ M(C/I) are nuclear by exactness of A
(cf. Remark ??). 

Lemma 12.1.2. Suppose that A and B are stable and separable C*-algebras,
and that Ψ : I(B) → I(A) is a lower semi-continuous action of Prim(B) on A,
such that Ψ(0) = 0 and Ψ−1 (A) = {B}.
Then the following are equivalent:

(i) There exists a non-degenerate weakly Ψ-residually nuclear *-monomor-

phism H0 : A → M(B), such that δ∞ H0 is unitarily equivalent to H0 ,
and, for J ∈ I(B),

Ψ(J) = H0−1 (H0 (A) ∩ M(B, J)).

(ii) For every J ∈ Prim(B), a ∈ A \ Ψ(J) there exist a Ψ-residually nuclear

completely positive map V : A → B such that V (a) 6∈ J.

If H0 with (i) exists, then H0 is unique up to unitary homotopy.

992 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

Proof. (i)⇒(ii): If a ∈ A and H0 (a) 6∈ M(B, J), then there exists b ∈ B+

with bH0 (a)b 6∈ J. V = bH0 (.)b is a Ψ-residually nuclear map from A to B with
V (a) 6∈ J.
(i)⇒(ii): If I is a closed ideal of B, and if a ∈ A is not in Ψ(I), then there
exists a primitive ideal J ∈ Prim(B), such that I ⊂ J and a 6∈ Ψ(J), because Ψ
is lower semi-continuous and I is the intersection of the primitive ideals J which
contain I.
Thus, there is a Ψ-residually nuclear map V : A → B with V (a) 6∈ I.
The set K of Ψ-residually nuclear completely positive contractions V from A to
B is a convex set. It is a separable metric space as a subset of CP(A, B) ⊂ L(A, B)
with strong topology on L(A, B) (i.e., topology of point-wise convergence in norm,
given by the system of semi-metrics ρa (V1 , V2 ) := kV1 (a) − V2 (a)k).
Let V1 , V2 , . . . be a dense sequence in K. We get weakly Ψ-residually nuclear
C *-morphisms dn from A to L(HB ), if we apply Stinespring-Kasparov dilation to
Vn , n = 1, 2, . . .. Here HB denotes the (right) Hilbert-B-module which is given by
the sequences (b1 , b2 , . . .) with norm convergent series b∗n bn in B. Since B is stable
P

and σ-unital, we get that there is a natural isomorphism HB ∼ = B as (right) Hilbert-

B-modules and L(HB ) ∼ = M(B) as C *-algebras under this isomorphism. If we use
this isomorphism, we get weakly Ψ-residually nuclear C *-morphisms hn : A → B,
such that Vn is the point-norm limit of Ψ-residually nuclear completely positive
contractions of form W = b∗ hn (.)b, with b ∈ B a contraction.
∗
P
Let h : A → M(B) be the direct sum of h1 , h2 , . . ., i.e., h(a) := n sn hn (a)sn
sn (sn )∗
P
for a ∈ A, where s1 , s2 , . . . is a sequence of isometries in M(B) with
strictly convergent to 1 in M(B).
Then h is weakly Ψ-residually nuclear and every Ψ-residually nuclear com-
pletely positive contraction V : A → B is the point-norm limit of maps b∗ h(.)b with
contractions b ∈ B.
It follows, that h(Ψ(I)) ⊂ M(B, I) and h(a) 6∈ M(B, I) for I ∈ I(B), a 6∈ Ψ(I).
Thus h : A → M(B) is a weakly Ψ-residually nuclear *-monomorphism, such
that, for I ∈ I(B),
h(Ψ(I)) = h(A) ∩ M(B, I).
The hereditary C *-subalgebra D of B, which is generated by h(A)Bh(A), is stable,
because A is stable.
The existence of an approximate unit in A shows that h(A)D ⊂ D and that
h(a)Bh(a∗ ) is contained in the closure of h(a)Dh(a∗ ) for a ∈ A.
Let k(a) := h(a)|D for a ∈ A. It follows, that k : A → M(D) is a non-
degenerate *-monomorphism from A into M(D), such that, for J ∈ I(B) and
a ∈ A+ , k(a) ∈ M(D, D ∩ J) if and only if h(a) ∈ M(B, J). Since every closed
ideal of D is the intersection of a closed ideal of B with D, we get that k is weakly
Φ-residually nuclear for Φ := ΨD,Bdown , and that Φ(D ∩ J) = Ψ(J) for J ∈ I(B). In
particular, Ψ(J1 ) = Ψ(J2 ) if J1 ∩ D = J2 ∩ D.
 1. NON-COMMUTATIVE ASYMPTOTIC SELECTION 993

The ideal I which is generated by D, satisfies h(A) ⊂ M(B, I). Thus Ψ(I) = A
and, by assumption, I = B. By Corollary 5.5.6, there is a *-isomorphism γ from
B onto D, which is approximately inner as a *-monomorphism from B to B and
satisfies γ(I) = D ∩ I for I ∈ I(B).
Let H0 denote the infinite repeat of M(γ)−1 k. Then H0 is a non-degenerated
weakly (Φ ◦ γ)-residually nuclear *-monomorphism from A to B with H0 (a) ∈
M(B, I), if and only if, k(a) ∈ M(D, γ(I)). Since δ 2 is unitarily equivalent to δ,
γ(I) = D ∩ I, and Ψ = Φγ, we get that H0 has the desired properties.
By Corollary 5.9.15, H0 with (i) is unique up to unitary homotopy. 

Lemma 12.1.3. Suppose that C is a separable commutative C*-algebra, that A

is a separable C*-algebra, and that

Ψ : I(C) ∼
= O(Prim(C)) → I(A)

is a lower semi-continuous action of X := Prim(C) on A with Ψ(0) = 0 and

Ψ−1 (A) = {C}.
Then, there is a *-monomorphism H from A into M(C ⊗ K) such that, for
every closed ideal J of C,

H(Ψ(J)) = H(A) ∩ M(C ⊗ K, J ⊗ K) .

If A is stable, then H can be taken as a non-degenerate *-monomorphism, i.e.,

such that H(A)(C ⊗ K) is dense in C ⊗ K.

Proof. Since J 7→ J ⊗ K defines a lattice isomorphism from I(A) onto I(A ⊗

K), it suffices to consider the case where A is stable. Moreover, it suffices to show
that condition (ii) of Lemma 12.1.2 is satisfied for B := C ⊗ K under the above
natural identification J ⊗ K ↔ J of I(B) and I(C).
Note that here X := Prim(C) is the maximal ideal space of C, and that X itself
or its one point compactification Y := X ∪ {C} is a metrizable compact space.
We define a map g from Y into the set of closed split faces of the quasi-state
space Sq (A) := {f ∈ A∗ : 0 ≤ f, kf k ≤ 1} of A as follows: For J ∈ Y let

g(J) := {f ∈ Sq (A) : f (Ψ(J)) = 0}.

Then the lower semicontinuity of Ψ implies that the map (J, f ) 7→ J is an open
map from {(J, f ) : J ∈ Y, f ∈ g(J)} ⊂ Y × Sq (A) onto Y .
Let a ∈ A and I a closed ideal of C. The lower semicontinuity of Ψ implies that,
if a 6∈ Ψ(I), then there is a maximal ideal J0 of C such that I ⊂ J0 and a 6∈ Ψ(J0 ).
Thus there is a quasi-state f0 ∈ g(J0 ) with f0 (a) = ka + Ψ(J0 )k. Since g satisfies
the requirements of the Michael selection principle [553], we find a continuous map
J ∈ Y 7→ f (J) ∈ Sq (A) such that f (J) ∈ g(J) and f (J0 ) = f0 . In particular,
f (C) = 0. The map

V : b ∈ A 7→ {f (J)(b)}J∈X ∈ C0 (X) ∼
= C ⊗ p11
994 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

is a Ψ-equivariant completely positive contraction from A into B with V (a) 6∈

I ⊗ K. 

Lemma 12.1.4. Suppose that B, C and D are separable C*-subalgebras of a

C*-algebra G, such that

(i) B and D are stable,

(ii) D contains B and DBD = D, and,
(iii) C is a commutative C*-subalgebra of D with the following properties (α)
and (β):
(α) For every closed ideals J1 and J2 of D, J1 ∩ C = J2 ∩ C implies
J1 ∩ B = J2 ∩ B.
(β) There exists a sequence of (in D) approximately inner completely
positive contractions Vn : D → C with limn→∞ Vn (a) = a for every
a ∈ C.

Suppose moreover that A is a separable stable C*-algebra and that Ψ : I(B) →

I(A) is a lower semi-continuous action of Prim(B) on A with Ψ(0) = 0 and
Ψ−1 (A) = {B}.
Let E := B · G · B ⊆ G and Φ(J) := Ψ(B ∩ J) for J ∈ I(E).
Then there exists a non-degenerate weakly Φ-residually nuclear *-monomor-
phism H1 : A ,→ M(E), such that δ∞ H1 is unitarily equivalent to H1 , and, for
every closed ideal J of E,

H1 (Φ(J)) = H1 (Ψ(B ∩ J)) = H1 (A) ∩ M(E, J).

H1 with this property is unique up to unitary homotopy.

In particular, for every automorphism ϕ of E with ϕ|B = idB , the composition
M(ϕ)H1 = ϕH1 ϕ−1 is unitarily homotopic to H1 .

Condition (iii,β) implies that (J1 ∩ C) + (J2 ∩ C) = (J1 + J2 ) ∩ C for all

closed ideals J1 and J2 of D. If a commutative C *-subalgebra C ⊆ D satisfies
this condition and has property (iii,α) then C is called regular Abelian C *-
subalgebra of D.
The map I / C 7→ J(I) / D considered in the proof of Lemma 12.1.4 is lower
semi-continuous. Thus, the family of c.p. maps W : D → C with W (J(I)) ⊂ I
is closed under point-norm convergence, satisfies W (J) ⊂ J ∩ C for all J / D by
J(J ∩ C) ⊇ J (hence W is approximately inner in D). Since C is nuclear, it follows
that W is residually nuclear.
Can we find, for c1 , . . . , cn ∈ C+ and ε > 0, a contraction V ∈ CPin (A, A) with
V (C) ⊂ C and kV (ck ) − ck k < ε ?

Proof. The hereditary C *-subalgebra E of G is σ-unital and stable because

B is σ-unital and stable. Therefore, the uniqueness of H1 follows from Lemma
12.1.2. The ϕ-invariance up to unitary homotopy follows from the uniqueness.
 1. NON-COMMUTATIVE ASYMPTOTIC SELECTION 995

Let J1 , J2 ⊂ D closed ideals, then C ∩ Ji ⊂ C ∩ (J1 + J2 ) for i = 1, 2. Now let

a ∈ (J1 + J2 ) ∩ C. There exist b ∈ J1 and c ∈ J2 with a = b + c. By assumption
(iii, β), for ε > 0 there is an approximately inner completely positive contraction
V : D → C with kV (a) − ak < ε. Then V (b) ∈ J1 ∩ C, V (c) ∈ J2 ∩ C and
V (a) ∈ (C ∩ J1 ) + (C ∩ J2 ). But (C ∩ J1 ) + (C ∩ J2 ) is a closed ideal of C.
Thus, for closed ideals J1 and J2 of D,

C ∩ (J1 + J2 ) = (C ∩ J1 ) + (C ∩ J2 ) .

Therefore, for every closed ideal I of C, the set of closed ideals J of D with
C ∩ J ⊂ I is upward directed. It is easy to see, that this set is also upward
monotonous closed. Hence, for every closed ideal I of C, there is a biggest ideal
J(I) in the set of closed ideals J of D with J ∩ C ⊂ I.
Let ΨD (I) := J(I) for I ∈ I(C).
The definition of ΨD shows that ΨD : I(C) → I(D) defines a lower semi-
continuous action of Prim(C) on D.
Prim(D) acts lower semi-continuous on B by J ∈ I(D) 7→ B ∩ J ∈ I(B).
It follows that the composition with the given action Ψ of Prim(B) on A,

ΨA : J ∈ I(C) 7→ Ψ(B ∩ ΨD (J))

is a lower semi-continuous action of Prim(C) on A.

Let J ∈ I(D), then, by definition of ΨD , J ⊂ ΨD (C ∩ J) and C ∩ ΨD (C ∩ J) ⊂
C ∩ J. By assumption (iii,α), this implies B ∩ J = B ∩ ΨD (C ∩ J). Thus, for every
J ∈ I(D),
Ψ(B ∩ J) = ΨA (C ∩ J).
In particular, ΨA (0) = Ψ(0) = 0 and ΨA (C) = Ψ(B) = A.
Let I ∈ I(C) and suppose that ΨA (I) = A, i.e., Ψ(B ∩ ΨD (I)) = A. Since, by
assumption on Ψ, Ψ−1 (A) = {B}, this implies B ⊂ ΨD (I). By (ii) and definition of
ΨD , it follows ΨD (I) = D, C = ΨD (I) ∩ C ⊂ I, and therefore, (ΨA )−1 (A) = {C}.
Thus, ΨA is a lower semi-continuous action of Prim(C) onto A with ΨA (0) = 0
and (ΨA )−1 (A) = {C}.
By Lemma 12.1.3, there exists a non-degenerate *-monomorphism H from A
into M(C ⊗ K) such that, for I ∈ I(C),

H(ΨA (I)) = H(A) ∩ M(C ⊗ K, I ⊗ K).

We have D ∩ E = BDB, because both are hereditary C *-subalgebras of D

which are generated by B. Therefore, D ∩ E is σ-unital and stable. By (ii), D is
the smallest closed ideal of D which contains B. Let J := DCD the closed ideal of
D which is generated by C. Then C ∩ J = C = C ∩ D, thus J ∩ B = D ∩ B = B
and J = D, by (iii,α) and (ii). It follows, that (D ∩ E) ⊗ p11 and C ⊗ K are stable
σ-unital C *-subalgebras of D ⊗ K which both generate the same ideal D ⊗ K.
996 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

By Corollary 5.5.6, there is a *-monomorphism

k : C ⊗ K ,→ D ∩ E
such that k(C ⊗ K) contains a strictly positive element of D ∩ E, and that, for every
closed ideal J of D,
k(C ⊗ K) ∩ J = k((C ∩ J) ⊗ K).

Since k is non-degenerate, M(k(C ⊗ K)) ⊂ M(E) and, for I ∈ I(G),

M(k(C ⊗ K)) ∩ M(E, I ∩ E) = M(k(C ⊗ K), k((C ∩ I) ⊗ K)).
Here we used that C ∩ I = C ∩ (I ∩ D).
Let H1 := M(k)δ∞ H = kδ∞ (H(.))k −1 . Then H1 is a non-degenerate *-
monomorphism, H1 (A) is contained in M(k(C ⊗ K)), and, for every closed J of E,
and
H1 (Φ(J)) = H1 (Ψ(B ∩ J)) = H1 (A) ∩ M(E, J) .
Note here that B ∩ J = B ∩ (I ∩ D), and C ∩ I = C ∩ (I ∩ D) for the unique ideal
I of G with J = I ∩ E.
Since H1 (A) is contained in M(k(C ⊗ K)), we get from Lemma 12.1.1(v) that
H1 is weakly Φ-residually nuclear. δ∞ H1 is unitarily equivalent to H1 , because
δ∞ M(k)δ∞ is unitarily equivalent to M(k)δ∞ by Lemma 5.1.2. 

We denote by [a, b] the commutator ab − ba in the following.

Lemma 12.1.5. For every n ∈ N, there is a continuous function fn with fn (0) =

0 which has the following universal property:
If A is a C*-algebra with A ∼ = A ⊗ O∞ ⊗ O∞ ⊗ . . ., a1 , . . . , an , b ∈ A+ are
contractions, and ε > 0, with k[aj , ak ]k < ε for j, k = 1, . . . , n, then, there exists a
contraction c ∈ A with
k[aj , c]k < fn (ε) , k(1 − c∗ c)aj k < fn (ε) and k[aj , c∗ bc]k < fn (ε) .

Proof. Let C be a commutative C *-subalgebra of a C *-algebra D. Then the

set K(C) of completely positive maps V (a) := (ci )∗ aci with c1 , . . . , cm ∈ C and
P

kV (1)k ≤ 1 is convex. Its point weak closure on D∗∗ is K(C ∗∗ ), where we identify
C ∗∗ naturally with the weak closure of C in D∗∗ . For b ∈ D, a1 , . . . , an ∈ C, and
δ > 0, there exist V ∈ K(C ∗∗ ) with
k(1 − V (1))aj k < δ and k[aj , V (b)]k < δ. ( ∗)
To see this, approximate a1 , . . . , an by elements in the span of mutually orthogonal
projections p1 , . . . , pk ∈ C ∗∗ and consider j pj bpj . A Hahn-Banach separation
P

argument shows that we can find V with (∗ ) even in K(C).

Suppose that there is a C *-algebra B and a *-monomorphism ϕ from B ⊗ O∞
P
into D, such that C ⊂ ϕ(B ⊗ 1) and b ∈ ϕ(B ⊗ 1). Let c := 1≤i≤m ϕ(di ⊗ si ),
where ϕ(di ⊗ 1) = ci and si denote the canonical generators of O∞ . Then c
satisfies c∗ bc = V (b), V (1) = c∗ c, and [aj , c] = 0. Therefore, k(1 − c∗ c)aj k < δ,
k[aj , c∗ bck < δ, and c is a contraction in D commuting with aj for j = 1, . . . , n.
 1. NON-COMMUTATIVE ASYMPTOTIC SELECTION 997

Now we define numbers g(a, b) := k[a, b]k, where [a, b] = ab − ba,

η(a1 , . . . , an ) := max{g(aj , ak ) : j, k = 1, . . . , n},

µ(a; b; c) := max(g(a, c), g(a, c∗ bc), k(1 − c∗ c)ak),

and, for C *-algebras A and contractions a1 , . . . , an , b, c ∈ A+ ,
v(a1 , . . . , an ; b; A) := inf {max(µ(a1 ; b; c), . . . , µ(an ; b; c))},
c

where the infimum runs over all contractions c ∈ A.

For every t ∈ R+ , and for every C *-subalgebra A, we denote by Y (t, A) the set
of sequences (a1 , . . . , an , b) of contractions in A+ such that η(a1 , . . . , an ) ≤ t. Now
we define increasing functions F (t, A) and F (t) as follows:
F (t, A) := sup{v(a1 , . . . , an ; b; A) : (a1 , . . . , an , b) ∈ Y (t, A)},
and F (t) := supA {F (t, A)}, where the supremum runs over all separable C *-
subalgebras A of L(`2 ) with A ∼= A ⊗ O∞ ⊗ O∞ ⊗ . . .. F (t, A) and F (t) are
increasing on R+ .
It is not hard to see, that fn with the desired properties exists if and only if
limt→0 F (t) = 0. Since F (t) is increasing, we have limt→0 F (t) = inf{F (1/k) : k ∈
N}.
Suppose that F (1/k) ≥ 3γ > 0 for k ∈ N. Then we can find, for k = 1, 2, . . .,

(i) separable C *-algebras Ak with Ak ∼ = Ak ⊗ O∞ ⊗ O∞ ⊗ . . ., and

(k) (k) (k) (k) (k)
(ii) (a1 , . . . , an , b ) ∈ Y (1/k, Ak ) such that v(a1 , . . . , an ; b(k) ; Ak ) > 2γ.
Q
Now we use ultrapowers: Let D := ω (Ak ), and a1 , . . . , an , b ∈ D with rep-
(1) (2)
resentatives (aj , aj , . . .) of aj , and (b(1) , b(2) , . . .) of b. Let C and B be the
C *-subalgebras of D that are generated by {a1 , . . . , an } and {b, a1 , . . . , an } respec-
tively. Then C is commutative and B is separable. By Proposition 7.4.11, there
exists a *-monomorphism ϕ from B ⊗ O∞ into D, such that ϕ(d ⊗ 1) = d for d ∈ B.
As we have seen above, there is a contraction c ∈ D such that
max(µ(a1 ; b; c), . . . , µ(an ; b; c)) < γ .
This contradicts the above property (ii). 

Lemma 12.1.6. Suppose that B ∼ = B ⊗ O∞ ⊗ O∞ ⊗ . . .. Let G := Q(R+ , B),

the corona of B, or let G := Bω , the ultrapower of B.
If D is a separable C*-subalgebra of G and C a separable commutative C*-
subalgebra of D, then there exist a contraction d ∈ C 0 ∩ G and a commutative
C*-subalgebra A of G such that C ⊂ A, d∗ da = a for a ∈ C, and d∗ Dd ⊂ A.

Proof. We consider the case G = Q(R+ , B). The proof for the case G = Bω
is similar.
Let e1 , e2 , . . . be a dense sequence in the positive contractions of D.
By Corollary 7.4.10, the approximately inner completely positive contractions
T : D → G are one-step inner, i.e., there is a contraction d ∈ G such that T (b) =
998 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

d∗ bd. If T |C = id |C, then [a, d] = (1 − dd∗ )ad and k(1 − dd∗ )1/2 adk2 = kT (a∗ a) −
T (a∗ )T (a)k = 0 for a ∈ C. It follows d ∈ C 0 ∩ G and and d∗ da = d∗ ad = a for
a ∈ C.
Thus it suffices to construct a sequence of approximately inner completely pos-
itive maps Tn : G → G, such that Tn |C = id |C, limn [Tn (ek ), Tn (ej )] = 0 and that
limn Tn (ek ) exists for k, j ∈ N .
The approximately inner completely positive contractions T : G → G with
T |C = id |C form a semigroup for every C *-subalgebra C ⊂ G. Therefore, we
attempt to construct a sequence of approximately inner completely positive con-
tractions Tn : G → G in the special way Tn := Sn+1 Sn . . . S1 . Here Sn : G →
G should be approximately inner completely positive contractions which satisfy
Sn |Cn = id |Cn and Sn (gn ) commutes with Cn , where C1 := C, g1 := e1 , and, by
induction, gn := Sn−1 . . . S1 (en ), Cn+1 := C ∗ (Cn , Sn (gn )).
Then the sequence of restrictions Tn |D has the desired properties, and A :=
indlim Cn , T (b) := lim Tn (b) for b ∈ D have the properties as stated in Lemma
12.1.6.
Thus it remains to show that for every separable commutative C *-subalgebra
C ⊂ G and every contraction g ∈ G+ there is an approximately inner completely
positive contraction S : G → G with S|C = id |C such that S(g) commutes elemen-
twise with C.
Let a1 , a2 , . . . a sequence of positive contractions in Cb (R+ , B), such that the
sequence of the cn := an + C0 (R+ , B) is dense in the positive contractions in C. Let
b ∈ Cb (R+ , B)+ a contraction which represents g. With the notation of Lemma
12.1.5, we find a sequence s1 > s2 > . . . in (0, 1] such that fm (ε) < 1/n for
ε ∈ (0, sn ], m ≤ n. This is possible by Lemma 12.1.5.
Since [cj , ck ] = 0, there are positive real numbers xn with xn+1 > max(xn , n),
such that k[ai (y), aj (y)]k < sn for y ≥ xn and 1 ≤ i, j ≤ n.
Let In := [xn , xn+2 ]. Since C(In , B) absorbs O∞ tensorial, we can apply
Lemma 12.1.5 to C(In , B), aj |In , j = 1, . . . , n, and b|In , and find contractions dn ∈
C(In , B) such that, for j = 1, . . . , n, k(1 − d∗n dn )(aj |In )k, k(aj |In ) − d∗n (aj |In )dn k
and k[(aj |In ), d∗n (b|In )dn ]k are smaller then 1/n.
Now we choose a continuous function µ1 (y) on R+ with 0 ≤ µ1 (y) ≤ 1,
µ1 (x2n ) = 0 and µ1 (x2n−1 ) = 0, for n ∈ N. Let µ2 (y) := 1 − µ1 (y).
We define elements h1 , h2 ∈ Cb (R+ , B) as follows:
Let hk (y) = 0 for y ∈ [0, xk ], k = 1, 2, and, for y ∈ [x2n−k , x2n+2−k ], n ∈ N,

hk (y) := (µk (y))1/2 d2n−k .

Then the corresponding elements d1 , d2 ∈ G satisfy kd∗1 d1 + d∗2 d2 k ≤ 1 and

define an inner completely positive contraction S(a) := d∗1 ad1 + d∗2 ad2 on G with
S|C = id |C, such that S(g) commutes with C. 
 1. NON-COMMUTATIVE ASYMPTOTIC SELECTION 999

Lemma 12.1.7. Suppose that B is separable and stable, and that B is isomor-
phic to B ⊗ O∞ ⊗ O∞ ⊗ . . ..
Let G := Q(R+ , B). Then there exist separable C*-subalgebras C and D such
that B, C, D and G satisfy the assumptions (i), (ii) and (iii) of Lemma 12.1.4.

Proof. Below, we show the existence of sequences of separable C *-subalgebras

Cn ⊂ Dn ⊂ Fn ⊂ Dn+1 of G, of contractions dn ∈ G, and of *-monomorphisms hn
from Fn ⊗ O∞ ⊗ K into G, such that

(i) D1 := B,
(ii) Fn is the C *-subalgebra of G generated by Dn and dn ,
(iii) hn (f ⊗ 1 ⊗ p11 ) = f for every f ∈ Fn
(iv) Dn+1 is the image hn (Fn ⊗ O∞ ⊗ K) of hn ,
(v) Cn is a commutative C *-subalgebra of Dn such that, for every closed ideal
J of Dn , J is the smallest closed ideal of Dn which contains Cn ∩ J, and
(vi) dn commutes elementwise with Cn , d∗n dn a = a for every a ∈ Cn , and
d∗n Dn dn generates a commutative C *-subalgebra An of Fn with Cn ⊂ An ,
(vii) An ⊂ Cn+1 .

In particular, we have B ⊂ Dn ⊂ Dn+1 and Cn ⊂ An ⊂ Cn+1 ⊂ Dn+1 .

Let D and C be the closures of the unions of the sequences (Dn ) and (Cn )
respectively. Then C ⊂ D and B ⊂ D. D is stable by (iv) and [373]. B is stable
by assumption. By (ii),(iii) and (iv), Dn+1 is the closed span of Dn+1 Dn Dn+1 .
Thus, by (i), the span of DBD is dense in D. C is commutative.
Let J1 and J2 be closed ideals of D with J1 ∩C = J2 ∩C. Then J1 ∩Cn = J2 ∩Cn
and, by (v), J1 ∩ Dn = J2 ∩ Dn . But Jk is the inductive limit of Jk ∩ Dn for
n = 1, 2, . . . and k = 1, 2.
We can define approximately inner completely positive contractions Tn : D →
C with Tn |Cn = id |Cn , as follows: Let enk := dn dn+1 . . . dn+k and Tn (a) :=
limk→∞ (enk )∗ aenk Note here that enm is in Dn+m+1 and that, therefore, by (vi) and
(vii), (enk )∗ a(enk ) is in Cn+m+2 and (enk )∗ a(enk ) = (enl )∗ a(enl ) for a ∈ Dn+m+1 ,
and k, l > m.
Thus, B, C, D and G satisfy the assumptions (i), (ii) and (iii) of Lemma
12.1.4.

Now we show the existence of Cn , Dn , dn , An , Fn and hn by induction:

We start at n = 0. Let C0 := 0, d0 := 0, A0 := 0 and let ϕ denote an
isomorphism from B ⊗ O∞ ⊗ K onto B ⊂ G. ϕ exists by our assumptions on B.
We define F0 ⊂ G as ϕ(B ⊗ 1 ⊗ p11 ). Then there is a unique *-monomorphism
h0 from F0 ⊗ O∞ ⊗ K onto B, such that, for b ∈ B, e ∈ O∞ and k ∈ K,

h0 (ϕ(b ⊗ 1 ⊗ p11 ) ⊗ e ⊗ k) = ϕ(b ⊗ e ⊗ k).

In particular, h0 (f ⊗ 1 ⊗ p11 ) = f for f ∈ F0 .

1000 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

In other words, h0 (f ⊗ e ⊗ k) = ϕ(ψ(f ) ⊗ e ⊗ k) , where ψ denotes the inverse

of the isomorphism b 7→ ϕ(b ⊗ 1 ⊗ p11 ) from B onto F0 .
By (ii), for n = 1, 2, . . ., Fn := C ∗ (Dn , dn ) is defined by Dn and dn .
If we have found Fn , we find a *-monomorphism hn from Fn ⊗ O∞ ⊗ K into G
with property (iii) by Corollary 7.4.9.
Since Dn+1 is the image of hn , we have B = D1 , and it is enough to to define
dn and Cn with properties (v), (vi) and (vii). If Cn ⊂ Dn are given, then the
existence of dn and An := C ∗ (d∗n Dn dn ) with (vi) follows from Lemma 12.1.6.
So we may assume that An ⊂ Fn and hn are given. It remains to construct
Cn+1 with properties (vii) and (v), where n has to be replaced by n + 1 in (v).
Let (fk ) ⊂ Fn a dense sequence in the set of positive contractions in Fn , andlet
Xn denote the commutative C *-subalgebra of Fn ⊗K which is generated by An ⊗p11
and the set {fk ⊗ pkk : k = 2, 3, . . .}.
We identify Fn ⊗ K with the subalgebra Fn ⊗ 1 ⊗ K of Kn := Fn ⊗ O∞ ⊗ K.
For every closed ideal J of Kn , we have that Xn ∩ J generates J as a closed ideal
of Kn . Since hn (Kn ) = Dn+1 , we can define Cn+1 as hn (Xn ). Now properties (v)
and (vii) follow from property (iii) of hn . 

Theorem 12.1.8 (Asymptotic non-commutative Selection). Suppose that A

and B are separable and stable C*-algebras, where B is isomorphic to B ⊗ O∞ ⊗
O∞ ⊗ . . ., and that
Ψ : I(B) → I(A)

defines a lower semi-continuous action of Prim(B) on A in the sense of Definition

1.2.6 with Ψ(0) = 0 and Ψ−1 (A) = {B}.
Let E denote the hereditary C*-subalgebra of Q(R+ , B) which is generated by
B, and let Φ(I) := Ψ(I ∩ B) for I ∈ I(E).
Then, there exists a non-degenerate weakly Φ-residually nuclear *-monomor-
phism H0 from A into the multiplier algebra M(E) of E such that δ∞ H0 is unitarily
equivalent to H0 , and, for I ∈ I(E),

Ψ(I ∩ B) = Φ(I) = H0−1 (H0 (A) ∩ M(E, I)).

H0 with this property is unique up to unitary homotopy.

Conversely H0 determines Ψ in the sense of Lemma 12.1.1, i.e., for J ∈ I(B),

Ψ(J) := H0−1 (H0 (A) ∩ M(E, E ∩ Q(R+ , J))) .

b ◦ H0 are unitarily homotopic, for every automorphism

In particular, H0 and σ
b of M(E), which is induced by a scaling homeomorphism σ of R+ .
σ

Proof. The existence of H0 is the logical sum of Lemma 12.1.4 and Lemma
12.1.7, because B is separable, stable and is isomorphic to B ⊗ O∞ ⊗ O∞ ⊗ . . ..
 2. SELECTIONS AND Ψ-EQUIVARIANT EMBEDDING 1001

The uniqueness up to unitary homotopy of a non-degenerate weakly Φ-resid-

ually nuclear *-monomorphism H0 : A → M(E) with
Φ(I) = H0−1 (H0 (A) ∩ M(E, I)).
and δ∞ H0 unitarily equivalent to H0 follows from Lemma 12.1.2.
For J ∈ I(B), Q(R+ , J) is an ideal of Q(R+ , B) and J = B ∩ Q(R+ , J).
Note that σ σ (I) = B∩I for every I ∈ I(E)
b fixes the elements of B and that B∩b
and every homeomorphism σ of R+ . Thus Φb σ = Φ, and, therefore the uniqueness
result applies to σ
bH0 .
Since Ψ(J) = Φ(E ∩ Q(R+ , J)) for J ∈ I(B), H0 determines Ψ in the sense of
Lemma 12.1.1. 

2. Selections and Ψ-equivariant embedding

We give now some definitions and lemmata that are needed for the generaliza-
tion of Theorem A to the case of non-simple strongly purely infinite algebras as
target algebras instead of O2 (cf. Theorem K).

Definition 12.2.1. Let I(A) denote the set of closed ideals of A with the
below described topology and let Prim(A) denote the space of primitive ideals of
A with the hull-kernel topology, i.e. J ∈ I(A) is in Prim(A) if it is the kernel of a
(non-zero) irreducible representation of A, and the closed subsets of Prim(A) are
given by the sets of “hulls” h(I) := {J ∈ Prim(A) : I ⊂ J}. Thus the closure
operation X ⊆ Prim(A) 7→ X ⊂ Prim(A) is given by X := h(k(X)), where the
T
“kernel” ideal k(X) of X is defined as k(X) := J∈X J.
We define a sort of generalized Gelfand transform b ∈ A 7→ bb from A into
the bounded functions on I(A), where bb : I(A) → R+ , for b ∈ A, is defined by
bb(J) := kb + Jk := inf{kb + ck : c ∈ J}.

Recall that the ideal space I(A) of a C *-algebra A is a compact Hausdorff space
if we take the coarsest topology under which every generalized Gelfand transform
is continuous. The latter is the Fell topology on I(A), cf. [559].
The compactness follows simply from the fact that every character on the
commutative C *-algebra C ∗ ({bb ; b ∈ A}) defines a C *-seminorm on A. Here
C ∗ ({bb ; b ∈ A}) means the C *-subalgebra of `∞ (I(A)discrete ) that is generated
by the Gelfand transforms bb (b ∈ A). All Gelfand transforms vanish on the point
J = A, i.e. bb(A) = 0. Thus, for every b ∈ A, bb is in C0 (I(A) \ {A}).
We shall use also the notation bb for the restriction bb | Prim(A) of bb to Prim(A) ⊂
I(A). This is justified, because bb is determined by its restriction to Prim(A), for
I ∈ I(A), i.e.,
bb(I) = sup{ bb(J) ; J ∈ h(I) }.

If A is commutative, then Prim(A) is the space of maximal ideals of A and bb

√
is the ordinary Gelfand transform of the absolute value |b| = b∗ b of b ∈ A.
1002 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

Remark 12.2.2. If A is separable, then, with the homeomorphism which we

have introduced in Chapter 1, I(A) becomes a closed subspace of the Hilbert cube
[0, 1]∞ , which contains (0, 0, . . .) and is closed under component-wise maximum α∨β
for α, β ∈ I(A). Since I(A) is compact, it follows that, for every subset K ⊂ I(A),
W V∞
K ∈ I(A), and, for every decreasing sequence α1 ≥ α2 ≥ . . ., n=1 αn is in I(A).
Note here, that the natural map from the ideals of A into the Hilbert cube
reverses the natural order, i.e., the elements α in the corresponding subset of the
Hilbert cube correspond better to the closed subsets of the primitive ideal space
of A. If A ∼= C(X) for a compact metric space X, the image is just the classical
compact space of closed subsets of X, the topology is then given by the Hausdorff
distance metric.
The reader should note that, as explained in Chapter 1, Prim(A) has not the
relative topology induced by I(A):
Prim(A) is a subset of I(A), but is not a topological subspace. The topology
of Prim(A) is given by the above cited hull-kernel topology (Jacobson topology). It
is easy to see, that the system O(Prim(A)) of open sets is the coarsest T0 -topology
on Prim(A) under which every generalized Gelfand transform b a of an element a in
A is lower semi-continuous. Thus, Prim(A) has the topology that is induced by
the T0 topology on the Hilbert cube [0, 1]∞ given by the family of open subsets in
Hausdorff topology, that are upward directed (with respect to the coordinate-wise
order on elements of [0, 1]∞ ).
If A is separable, one can re-discover Prim(A) as the ∨-prime elements of I(A)
with system of closed sets induced by the intersection of I(A) with the set {β : β ∨
α = α} of points below α ∈ I(A).
The reader, who is not familiar with this topics, could learn from considering
the following simple example:
Let X a point-complete T0 -space with a countable base of its topology (e.g. X ∼
=
Prim(A) with separable A), and let f : X → R+ be a bounded non-negative function
on X with kf k∞ := sup f (X) = 1. We can define an action of O(X) on C0 ((0, 1]) by
Ψf (Z) := C0 ((sup f (X \Z), 1]) for open subsets Z of X. Here we define sup(∅) := 0,
because we consider only subsets and elements of [0, 1].
The reader easily checks that the action Ψf is lower semi-continuous if and only
if f is lower semi-continuous.
It is interesting to note, that Ψf satisfies property (ii) of Definition 1.2.6, if and
only if, for every decreasing sequence K1 ⊃ K2 ⊃ . . . of closed subsets Kn of X,
\
sup f ( Kn ) = inf{sup f (Kn ) : n = 1, 2, . . .}.

Now let X := Prim(A) for a separable C *-algebra A. As we shall see below, Ψf

satisfies properties (ii), (iii) and (iv) of Definition 1.2.6, if and only if, J 7→ f (h(J))
is a continuous function on I(A), and this is the case, if and only if, f = bb for an
element b ∈ A ⊗ O2 .
 2. SELECTIONS AND Ψ-EQUIVARIANT EMBEDDING 1003

The property (i) of Definition 1.2.6 is satisfied for Ψf only in the rare and
exceptional case where f nsatisfies sup f (K1 ∩ K2 ) = min(sup f (K1 ), sup f (K2 ))
for every closed subsets K1 , K2 ⊂ Prim(A). E.g., this happens for commutative A,
if and only if, f is 0 except on one point of Prim(A).

Let s, t ∈ M(A) isometries, a, b, c, d ∈ A elements and denote by β(a, b; t) the

maximum of kt∗ bt − ak, kt∗ b∗ bt − a∗ ak and kt∗ bb∗ t − aa∗ k. Clearly t ∈ M(A) 7→
β(a, b; t) is strictly continuous and β(c, d; t) ≤ β(a, b; t) + ka − ck + kb − dk.
In the following Lemma 12.2.3, let

D(a, b; s, t) := max(β(b, a, s), β(a, b, t)) .

The definition causes D(a, b; s, t) ≤ D(c, d; s, t) + ka − ck + kb − dk. A similar

argument shows that D(a, b; s, t) is strictly continuous in s, t ∈ M(A). We define a
sort of “distance” ρ1 of contractions a, b ∈ A by

ρ1 (a, b) := ρ1 (a, b, A) := inf{D(a, b; s, t) : s, t ∈ M(A), s∗ s = t∗ t = 1 } .

Lemma 12.2.3. For every C*-algebra A with A = ∼ A ⊗ D2 , and for every

contractions a, b ∈ A with ρ1 (a, b, A) < ε, there exists a unitary U ∈ M(A) such
√
that kU ∗ aU − bk < 8 ε .

Proof. It suffices to consider separable C *-algebras A (up to isomorphisms).

The properties of D(a, b; s, t) imply ρ1 (a, b) ≤ ρ1 (c, d) + ka − ck + kb − dk. Obviously
kU ∗ aU − bk ≤ kU ∗ cU − dk + ka − ck + kb − dk for a, b, c, d ∈ A by triangle inequality
for unitary U ∈ M(A).
Recall that D2 := O2 ⊗ O2 ⊗ · · · . By the above inequalities we can suppose
for our estimates that a = c ⊗ 1, b = d ⊗ 1 ∈ A ⊗ 1 and that there are isometries
s, t ∈ M(A) with D(a, b; s ⊗ 1, t ⊗ 1) = D(c, d; s, t) < ε, by approximation of
contractions a, b ∈ A ∼
= A ⊗ D2 by those in A ⊗ 1 and by strict approximation of
the isometries s, t ∈ M(A) by those in M(A) ⊗ 1 ⊂ M(A). Then we have to find
U ∈ M(A) ⊗ O2 ⊂ M(A) with kU ∗ (c ⊗ 1)U − d ⊗ 1k ≤ 8 · D(c, d; s, t)1/2 .
Let B := C ∗ (x) denote the universal C *-algebra generated by a single con-
traction x ∈ B and define representations h, k : B → E := M(A) by h(x) := c,
k(x) := d, then we find a unitary U ∈ E ⊗ O2 that satisfies the inequality

kU ∗ (h(x) ⊗ 1)U − k(x) ⊗ 1k < 8 · µ(s, t; h, k; x)1/2 ,

by Corollary 4.5.2, where µ(s, t; h, k; x) denotes the maximum of the 6 values

 ∗
ks h(y)s − k(y)k , kt∗ k(y)t − h(y)k ; y ∈ { x, x∗ x, xx∗ } .

This definition of µ(s, t; h, k; x) shows that D(c, d; s, t) = µ(s, t; h, k; x), i.e., U is

as desired. 

Lemma 12.2.4. Suppose that A is a separable C*-algebra, and that F a simple

and exact C*-algebra (e.g. F = O2 or F = K). Let B := A⊗O2 and X := Prim(A).
Then:
1004 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

(i) The map J 7→ J ⊗ F defines a natural topological isomorphism X = ∼

Prim(A ⊗ F ).
(ii) {b
a : a ∈ A ⊗ K} is closed under point-wise maximum.
(iii) Every lower semi-continuous non-negative function f on X is the point-
wise l.u.b. of a sequence (can ) with an ∈ A+ .
(iv) {ba : a ∈ B} ⊂ `∞ (Xdisk ) is a maximum-closed and k.k∞ norm-closed
subset of `∞ (Xdisk ),
(v) The k . k∞ - closure of {ba : a ∈ A ⊗ K} is {b
a : a ∈ B}.

Proof. Ad(i): By Proposition B.4.2, J 7→ J ⊗ F defines topological isomor-

phism from Prim(A) onto Prim(A⊗F ), for every simple separable exact C *-algebra
F , e.g. F = O2 or F = K.
Ad(ii): Let s, t be canonical generators of a unital copy of O2 in M(A ⊗ K).
The Cuntz sum a ⊕ b := sas∗ + tbt∗ satisfies
k(a ⊕ b) + Jk = max(ksas∗ + Jk, ktbt∗ + Jk) = max(ka + Jk, kb + Jk) ,
a ⊕ b = max( b
i.e., \ a , bb ).
Ad(iii): Let f : X → R+ be a lower semi-continuous function on X. For
c ∈ R+ , let Y (c) := f −1 ((c, ∞)), and let χ(c) denote the characteristic function of
Y (c). Then cχ(c) ≤ f . If z ∈ X and 0 < c < f (z), then c = cχ(c)(z). Thus f is
the point-wise the l.u.b. of the at most countable set {cχ(c) : c = m/n, m, n ∈ N}.
Since f is lower semi-continuous, Y (c) is an open subset of X and, therefore χ(c)
is lower semi-continuous. Moreover, by the definition of the topology on X, there
is a closed ideal I(c) of A such that Y (c) = X \ h(I(c)).
Thus it suffices to show that, for every closed ideal I of A and c ∈ R+ , there
exist a ∈ A+ , such that the characteristic function χ of X \ h(I) is the l.u.b. of
gn := c a 1/n where bn = a1/n . Since I is separable, we find a strictly positive
bn = b
contraction a in I+ . But this means: 0 < ka + Jk ≤ 1 if J ∈ X is a primitive ideal
of A which is not in h(I), and ka + Jk = 0 for J ∈ h(I), i.e., ba ≤ χ and ba(J) > 0 if
1/n
χ(J) > 0. Thus, sup b a(J) is the characteristic function χ of X \ h(I).
Ad(iv): Let b ∈ B, d := (b∗ b)1/2 and c ∈ (O2 )+ with Spec(c) = [0, 1], then
kb + Jk = k(d ⊗ c) + J ⊗ O2 k.
There is an isomorphism ϕ from B ⊗ O2 onto B, such that g 7→ ϕ(g) ⊗ 1 is an
approximately inner endomorphism of B ⊗ O2 . In particular ϕ(J ⊗ O2 ) = J, and,
for a := ϕ(d ⊗ c),
kb + Jk = ka + Jk.

This, together with the logical sum of Corollary 3.10.11 and Lemma 12.2.3,
says that, for every uniformly convergent sequence fn = bbn with bn ∈ B, there
is a sequence (an ) ⊂ B+ , such that (an ) converges in B and bbn = ac
n . a 7→ b
a
is a norm-continuous map from B into `∞ (Prim(B)discrete ). Thus lim fn = b for
b
b = lim an .
Part (v) follows from (iv), because O2 is nuclear. 
 2. SELECTIONS AND Ψ-EQUIVARIANT EMBEDDING 1005

Definition 12.2.5. Suppose X is a T0 -space and that f : X → R+ := [0, ∞) is

a non-negative function on X. We say that f is a Dini function on X (or say f is
Dini) if f is lower semi-continuous and, for every increasing sequence g1 ≤ g2 ≤ . . .
of lower semi-continuous functions gn : X → R+ , (gn ) converges uniformly to f in
`∞ (X), i.e., limn→∞ kf −gn k∞ = 0, if f is point-wise the supremum of the sequence
(gn ), i.e., if f (x) = sup{gn (x) : n ∈ N }.

Every Dini function on X is automatically bounded, because we can consider

gn := min(n, f ).
If X is a metrizable, locally compact and σ-compact Hausdorff space, then a
function f is a non-negative Dini function in the sense of the above given Definition,
if and only if, f ∈ C0 (X). This follows from the classical Lemma of Dini.
The following is a weak version of a stronger result in [447], where B = A ⊗ O2
in (ii) is replaced by A itself.

Proposition 12.2.6. Suppose that A is a separable C*-algebra and let B :=

A ⊗ O2 , X := Prim(A) and h(I) := {J ∈ X : I ⊆ J} for I ∈ I(A).
Then for a non-negative function f on X the following properties are equivalent:

(i) f is a Dini function.

(ii) There exists b ∈ B with bb = f , i.e., f (J) = bb(J ⊗ O2 ) for J ∈ X.

(iii) f is lower semi-continuous, bounded, and, for every increasing sequence

I1 ⊂ I2 ⊂ . . . of closed ideals In of A,
sup f (h(I)) = inf{sup f (h(In )) : n = 1, 2, . . .},
S∞
where I is the closure of n=1 In .

Proof. (i)⇒(ii): Suppose that f : X → R+ is a Dini function. By Lemma

12.2.4(iii), there is a sequence (an ) in A such that f is the point-wise supremum
of the sequence (c n (J)) for J ∈ X. Then gn is
an ). Let gn (J) := max(ab1 (J), . . . , ac
an increasing sequence of non-negative lower semi-continuous functions on X with
f = sup{gn : n ∈ N} point-wise. Since f is Dini, (gn ) converges uniformly to f .
Let bn := (a1 ⊗ 1) ⊕ . . . ⊕ (an ⊕ 1) in B, then gn (J) = kbn + (J ⊗ O2 )k.
Thus gn ∈ {bb : b ∈ B}. By Lemma 12.2.4(iv), f = bb for some b ∈ B.
(ii)⇒(iii): Suppose that there is b ∈ B such that f (J) = bb(J ⊗ O2 ) for J ∈ X.
Let g(I) := kb + (I ⊗ O2 )k for every closed ideal I of B. Then, for every subset
T
Z ⊂ I(A), g(I) = sup{g(J) : J ∈ Z}, where I := {J : J ∈ Z}. If K is the
closure of an increasing sequence I1 ⊂ I2 ⊂ . . . of closed ideals In in A, g(K) =
inf{g(In ) : n ∈ N}.
Since g|X = f , it follows that f satisfies sup f (Z) = g(k(Z)) for every subset
T
Z of X, where k(Z) = {J : J ∈ Z} is the kernel-operation on I(A).
1006 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

But hk(Z) is the closure of Z in X and kh(I) = I for every closed ideal I of
A. We get that f is lower semi-continuous, and that sup f (h(I)) = g(I) for every
closed ideal I of A. Now the upward monotone continuity of g implies, for every
increasing sequence I1 , I2 , . . . of closed ideals In of A, that, for the closure I of
S
{I1 , I2 , . . . },

sup f (h(I)) = inf{sup f (h(In )) : n = 1, 2, . . .}.

(iii)⇒(i): We define a map T from the bounded functions f on X into the

functions F on I(A), by T (f )(I) := sup f (h(I)) for I ∈ I(A). Then T (g) ≤ T (f )
if g ≤ f . If I1 , I2 ∈ I(A) then T (f )(I1 ∩ I2 ) = max(T (f )(I1 ), T (f )(I2 )), because
h(I1 ∩ I2 ) = h(I1 ) ∪ h(I2 ). In particular, T (f ) is monotone, i.e., T (f )(I) ≤ T (f )(J)
if J ⊂ I.
If f is lower semi-continuous on X, then T (f )(J) = f (J) for every primitive
ideal J of A:
The set {y ∈ X : f (y) ≤ f (J)} is closed and contains J. The closure of the
one-point set {J} is h(J). Thus f (J) ≤ T (f )(J) = sup f (h(J)) ≤ f (J).
If f is lower semi-continuous on X, then T (f ) is also a lower semi-continuous
function on I(A):
Let M be a subset of I(A), and let J(M ) denote the intersection of the ideals
I ∈ M . Then h(J(M )) is the closure in X of of the union of the subsets h(I) ⊂ X
for I ∈ M . Since f is lower semi-continuous,
[
sup f (h(J(M ))) = sup f ( {h(I) : I ∈ M }) = sup{sup f (h(I)) : I ∈ M }.
S S
The right hand equality comes from f ( {h(I) : I ∈ M }) = {f (h(I)) : I ∈ M }.
Let c ∈ R+ and P ∈ I(A) a point in the closure of the set M(c) :=
{I ∈ I(A) : T (f )(I) ≤ c}. Then J(M(c)) ⊂ P and, therefore, T (f )(P ) ≤
T (f )(J(M(c))). By the above formula for f , T (f )(J(M(c))) ≤ sup{T (f )(I) : I ∈
M(c)} ≤ c. Thus T (f ) is lower semi-continuous on I(A).

Now suppose that f satisfies the assumptions of (iii).

We show that T (f ) is then also an upper semi-continuous function, i.e., T (f )
is a continuous function on the metrizable compact Hausdorff space I(A):
Let c ∈ R+ , and let (Pn ) be a convergent sequence in I(A) with limit P ∈ I(A)
T
such that c ≤ T (f )(Pn ) for n = 1, 2, . . .. Let In = k≥n Pk . Then In ⊂ In+1 ,
In+1 ⊂ Pn+1 and, therefore, c ≤ T (f )(Pn+1 ) ≤ T (f )(In+1 ) ≤ T (f )(In ).
S
It follows c ≤ T (f )(I) for the closure I of {I1 , I2 , . . . }.
Let a ∈ P , then b
a(P ) = 0, b a(Pk ) : k ≥ n}, and
a(In ) = sup{b

a(P ) = lim b
b a(Pn ) = lim b
a(In ) = b
a(I).
n

Thus, P ⊂ I and c ≤ T (f )(I) ≤ T (f )(P ). Which proves the upper semicontinuity

of T (f ).
 2. SELECTIONS AND Ψ-EQUIVARIANT EMBEDDING 1007

Let 0 ≤ g1 ≤ g2 ≤ . . . be an increasing sequence of lower semi-continuous

functions on X, such that f (y) = sup{gn (y) : n ∈ N} for every y ∈ X. Let I ∈ I(A)
and ε > 0. We find y ∈ h(I) ⊂ X such that f (y) + ε > sup f (h(I)). There is
n ∈ N with gn (y) + ε > sup f (h(I)). Thus T (g1 ) ≤ T (g2 ) ≤ . . . is an increasing
sequence of lower semi-continuous functions on I(A), which converges point-wise
to the continuous function T (f ).
By the lemma of Dini, the sequence T (gn ) converges uniformly to T (f ) on
I(A). This happens also with gn = T (gn )|X and f = T (f )|X.
Hence, f is a Dini function. 

Corollary 12.2.7. Suppose that A and B are separable C*-algebras, and that
γ is a topological isomorphism from Prim(A) onto Prim(B). Let E := A ⊗ O2 and
F := B ⊗ O2 . Then f 7→ f ◦ γ defines an isomorphism from {bb : b ∈ F } onto
{b
a : a ∈ E}, where we naturally identify Prim(A) with Prim(E) and Prim(B) with
Prim(F ).

Proof. f 7→ f ◦ γ defines an isomorphism from the Dini functions on Prim(B)

onto the Dini functions on Prim(A). Now apply Proposition 12.2.6. 

Corollary 12.2.8. Suppose that A and B are separable C*-algebras, and that
Ψ is a map from I(B) into I(A) with Ψ(J1 ) ⊂ Ψ(J2 ) for J1 ⊂ J2 , J1 , J2 ∈ I(B).
Then:

(i) Ψ is lower semi-continuous, if and only if, for every a ∈ A, J 7→ b a(Ψ(J))

is a lower semi-continuous function on Prim(B), and for every I ∈ I(B)
and a ∈ A with b a(Ψ(I)) > 0 there exists J ∈ Prim(B) with b a(Ψ(J)) > 0
and I ⊂ J.
(ii) Ψ is, moreover, countably monotone upper semi-continuous, i.e., satisfies
(ii) of Definition 1.2.6, if and only if, for every a ∈ A, J 7→ b
a(Ψ(J)) is a
Dini function on Prim(B).

aΨ|X for a ∈ A.
Proof. Let X := Prim(B) and fa := b
Suppose that Ψ is lower semi-continuous. Then, for every subset Z of X,
\ \
a(Ψ( J ∈ Z) = b
b a( Ψ(J)) = sup{ba(J) : J ∈ Z}.
J∈Z

In particular, sup fa (Z) = sup fa (hk(Z)) and b

a(Ψ(I)) = sup fa (h(I)), because
T T T
{J ∈ Z} = {J ∈ hk(Z)} and I = kh(I) = {J ∈ h(I)}. But this means that
fa is lower semi-continuous on X and that b a(Ψ(I)) > 0 implies the existence of
J ∈ h(I) with fa (J) > 0.
Conversely, suppose that fa is lower semi-continuous for every a ∈ A and
that for every I ∈ I(B) and a ∈ A with b a(Ψ(I)) > 0 there exists J ∈ h(I)
with fa (J) > 0. This says that fa (Z) = {0} implies fa (hk(Z)) = {0} and that
a(Ψ(I)) = 0 if and only if fa (h(I)) = {0}.
b
1008 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

T T S
Let I := α Iα , K := α Ψ(Iα ) and Z := α h(Iα ). By monotony of Ψ, we
have Ψ(I) ⊂ K. If a ∈ K, then b a(Ψ(Iα )) = 0 for every α in the index set. Thus
{0} = fa (Z) = fa (hk(Z)). But k(Z) = I. Therefore, ba(Ψ(I)) = 0, i.e., a ∈ Ψ(I).
Hence K = Ψ(I) and Ψ is lower semi-continuous.
Ad(ii): We have seen in the proof of the first part, that b
a(Ψ(I)) = sup fa (h(I))
for I ∈ I(B) and that fa is lower semi-continuous, if Ψ is lower semi-continuous.
fa is bounded by kak.
Let I1 ⊂ I2 ⊂ . . . an increasing sequence of closed ideals of B, and let I and K
S S
denote the closures of In and Ψ(In ), respectively. Then
a(K) = inf{sup fa (h(In )) : n = 1, 2, . . .}.
b
By Proposition 12.2.6(iii), it follows that b a(Ψ(I)) = b a(K) if fa is a Dini function.
Thus, Ψ satisfies (ii) of Definition 1.2.6 if fa is a Dini function on X for every a ∈ A.
If Ψ satisfies (ii) of Definition 1.2.6 then Ψ(I) = K and, therefore,
sup fa (h(I)) = inf{sup fa (h(In )) : n = 1, 2, . . .}.
Thus, by Proposition 12.2.6(iii), fa is a Dini function on X for every a ∈ A, if Ψ
satisfies property (ii) of Definition 1.2.6. 

Remark 12.2.9. Recall from Remark 3.11.3 that for a C *-algebra B the
following Properties (i)–(iv) are equivalent to each other:

(i) B strongly purely infinite in the sense of Definition 1.2.2.

(ii) For every a, b ∈ B+ and ε > 0 there exist contractions s, t ∈ B such that
ka2 − s∗ a2 sk, kb2 − t∗ b2 tk and ks∗ abtk are all less then ε.
(iii) The ultrapower Bω of B is strongly purely infinite.
(iv) The asymptotic corona Q(R+ , B) of B is strongly purely infinite.
(v) Q(R+ , B) has the WvN-property of Definition 1.2.3.
(vi) B admits no non-zero character, and for every separable C*-subalgebra C
of Q(R+ , B) the following holds:
If t ∈ R+ 7→ V (t) is a strongly continuous map from R+ into the
approximately inner completely positive contractions from B into B, T is
the completely positive contraction from Q(R+ , B) into Q(R+ , B) given by
T (a + C0 (R+ , B)) := V (a) + C0 (R+ , B),
where a ∈ Cb (R+ , B) and V (a)(t) := V (t)(a(t)), and if T |C is residually
nuclear, then, there exists a contraction d ∈ Q(R+ , B) such that T (b) =
d∗ bd for b ∈ C.

From (ii) it follows that the class of strongly purely infinite algebras is closed
under inductive limits.
It is obvious that quotient algebras and hereditary C *-subalgebras of strongly
purely infinite C *-algebras are strongly purely infinite.

Remark 12.2.10. In [431, sec. 3] we have discussed several equivalent proper-

ties of C *-subalgebras A of a C *-algebra B, among them the property, that there
 2. SELECTIONS AND Ψ-EQUIVARIANT EMBEDDING 1009

is a normal conditional expectation from the second conjugate B ∗∗ of B onto the

second conjugate A∗∗ of A, where we have naturally identified A∗∗ with the weak
closure of A in B ∗∗ .
Among others, in [431] it was shown that this property is equivalent to the
relative weak injectivity (in short: rwi) of A in B, which means that the natural
C *-morphism from the maximal C *-algebra tensor product A ⊗max C to B ⊗max C
is injective, and thus isometric, for every C *-algebra C, cf. [431, prop. 3.1]. Here
it suffices even to take C = C ∗ (F∞ ), the full group C *-algebra of the free group on
countably many generators.
It follows immediately that the closure A of the union of an increasing sequence
A1 ⊂ A2 ⊂ . . . of relatively weakly injective subalgebras of B is again relatively
weakly injective in B.
Unfortunately, relative weak injectivity does not pass to quotients in general:
There exist B, A ⊂ B, closed ideals J of A, and I1 ⊂ I2 of B such that

(i) J := I1 ∩ A = I2 ∩ A and A is relatively injective in B,

(ii) A/J is relatively injective in B/I1 , but
(iii) A/J is not relatively weakly injective in B/I2 .

We define a stronger property as the relative weak injectivity such that it plays
together with some particular quotient maps in a natural manner:
We say that A ⊂ B is residually relatively weakly injective in B, if,
for every J ∈ I(A), the natural monomorphism maps A/(A ∩ ΨA,B down (J)) onto a
A,B 4
relatively weakly injective C *-subalgebra of B/Ψdown (J) ( ).
The above mentioned results from [431] and the short exactness of the maximal
C *-algebra tensor product imply immediately the following equivalent formulation:
A ⊂ B is residually relatively weakly injective in B, if and only if,

A ⊗max C ∗ (F∞ ) ⊂ B ⊗max C ∗ (F∞ )

and, for every J ∈ I(A),

(A ∩ ΨA,B
down (J)) ⊗
max
C ∗ (F∞ ) = (A ⊗max C ∗ (F∞ )) ∩ (ΨA,B
down (J) ⊗
max
C ∗ (F∞ )).

If A is relatively weakly injective in B, and if I is a closed ideal of B ⊗max C

such that
I ∩ (A ⊗max C) = (A ∩ ΨA,B
down (J)) ⊗
max
C,
then I contains ΨA,B
down (J) ⊗
max
C.
It follows, that A ⊂ B is residually relatively weakly injective in B if, e.g., A
is relatively weakly injective in B and every closed ideal of A ⊗max C ∗ (F∞ ) is the
intersection of A ⊗max C ∗ (F∞ ) with a closed ideal of B ⊗max C ∗ (F∞ ).

4Recall that ΨA,B (J) is the smallest closed ideal of B which contains J ∈ I(A).
down
1010 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

Lemma 12.2.11. Suppose that A1 ⊂ A2 ⊂ . . . is a sequence of C*-subalgebras

of B which are residually relatively weakly injective in B. Then the closure A of its
union is residually relatively weakly injective in B.

Proof. First, A is again relatively weakly injective in B, i.e., A ⊗max C and

An ⊗max C are naturally C *-subalgebras of B ⊗max C for every C *-algebra C,
because the functor (.) ⊗max C is continuous with respect to inductive limits.
Let J ∈ I(A). Then (A ∩ ΨA,B down (J)) ⊗
max
C is contained in ΨA,B
down (J) ⊗
max
C,
A,B
and Ψdown (J) is the closure of the union of the increasing sequence of closed ideals
ΨA n ,B
down (An ∩ J).

For n < m, the intersection of ΨAn ,B

down (An ∩ J) ⊗
max
C with Am ⊗max C is
Am ,B
max
contained in A ∩ Ψdown (Am ∩ J) ⊗ C.
Thus, the intersection of ΨAn ,B
down (An ∩ J) ⊗
max
C with A ⊗max C is contained
in (A ∩ ΨA,B
down (J)) ⊗
max
C.
By continuity of (.) ⊗max C , we get that (A ∩ ΨA,B
down (J)) ⊗
max
C is the inter-
max A,B max
section of A ⊗ C with Ψdown (J) ⊗ C. 

Lemma 12.2.12. Suppose that A ⊂ B is non-degenerate and residually rela-

tively weakly injective in B, and that A ∩ ΨA,B
down (J) = J for every J ∈ I(A).

Let C be a C*-subalgebra of M(A) and V : C → M(A) a completely positive

map.
Then V is weakly residually nuclear as a map into M(A), if and only if, it is
weakly residually nuclear as a map into M(B).

Proof. By Lemma 12.1.1(iv), since M(A) ⊂ M(B) unitally, we have M(A) ∩

M(B, I) = M(A, A ∩ I) for I ∈ I(B). It follows that M(A, J) = M(A) ∩
M(B, ΨA,B
down (J)) . Therefore, V (C ∩ M(B, I)) ⊂ M(B, I) for every I ∈ I(B),
if and only if, V (C ∩ M(A, J)) ⊂ M(A, J) for every J ∈ I(A).
The argument for the proof of Lemma 12.1.1(v) shows, that V is also weakly
residually nuclear as a map from C ⊂ M(B) into M(B), if V is weakly residually
nuclear as a map from C into M(A).
Now suppose that V is weakly residually nuclear as a map from C into M(B),
and let J ∈ I(A) and I := ΨA,B down (J). Then J = A ∩ I, and C ∩ M(B, I) =
C ∩ M(A, J), because C ⊂ M(A). C/(C ∩ M(A, J)) is in a natural way a C *-
subalgebra of M(A/J) ⊂ M(B/I). The natural map [V ]I : C/(C ∩ M(B, I)) →
M(B/I) is weakly nuclear as a map from C/(C∩M(B, I)) ⊂ M(B/I) into (B/I)∗∗ ,
and [V ]I (C/(C ∩ M(B, I))) is contained in M(A/J). By assumption, A/J ⊂ B/I
is a non-degenerate and relatively weakly injective C *-subalgebra of B/I, i.e., there
is a normal conditional expectation P from (B/E)∗∗ onto the weak closure (A/J)∗∗
of A/J in (B/E)∗∗ . The natural map [V ]J from C/(C ∩ M(A, J)) into (A/J)∗∗
equals P [V ]I . Thus [V ]J is weakly nuclear.
Thus V : C → M(A) is weakly residually nuclear. 
 2. SELECTIONS AND Ψ-EQUIVARIANT EMBEDDING 1011

Remark 12.2.13. Recall that, by Definition 1.2.3, a C *-algebra B has resid-

ually nuclear separation if, for every separable C *-subalgebra C ⊂ B, every
a ∈ C+ and every ε > 0, there exists a residually nuclear completely positive
contraction V : C → B such that kV (a) − ak < ε. Obviously, residually nuclear
separation passes to hereditary C *-subalgebras and to quotient algebras. An ap-
plication of Lemma 12.1.2 to A := B and Ψ := ΨB shows that
a separable stable C*-algebra B has residually nuclear separation, if and only
if, there is a non-degenerate weakly residually nuclear *-monomorphism H0 : B →
M(B), such that δ∞ H0 is unitarily equivalent to H0 , and, for J ∈ I(B),

H0 (J) = H0 (B) ∩ M(B, J).

On can conlude from [463] and [443] that

a C*-algebra B with residually nuclear separation has the WvN-property if and

only if it is strongly purely infinite.

Lemma 12.2.14. Suppose that B is strongly purely infinite, and that A is a

separable C*-subalgebra of M(B).
Then, for every separable C*-subalgebra C of B, there exists a separable C*-
subalgebra D of B, such that

(i) C ⊂ D,
(ii) D is strongly purely infinite,
(iii) for every closed ideal J of D, there exists a closed ideal I of B such that
J = D ∩ I, i.e., J = D ∩ ΨD,Bdown (J),
(iv) there is a weakly continuous conditional expectation from (B/ΨD,B down (J))
∗∗

onto (D/J)∗∗ , i.e., D is residually relatively weakly injective in B, and

(v) AD ⊂ D and aD = {0} implies a = 0 for a ∈ A, i.e., the natural
C*-morphism L : a ∈ A → La |D ∈ M(D) from A to M(D) is a mono-
morphism, where La (b) := ab.

The subalgebras D of B with the properties (i)-(v) are closed under inductive
limits, i.e., closures D of increasing sequences D1 ⊂ D2 ⊂ . . . of C*-subalgebras of
B with properties (i)-(v) satisfy again (i)-(v).
If B is stable, then D can be chosen as stable subalgebra of B.
If B is stable and has residually nuclear separation, then D can be chosen
such that D is stable and there exists a non-degenerate residually nuclear *-
monomorphism H0 : D → M(D), such that δ∞ H0 and H0 are unitarily equivalent,
and H0 (J) = H0 (D) ∩ M(D, J) for every closed ideal J of D.
If B is σ-unital (with strictly positive element e ∈ B+ ) then D with (i)-(iv) for
C replaced by C ∗ (C, e) satisfies IL(A ∩ M(B, I)) = L(A) ∩ M(D, D ∩ I) for every
I ∈ I(B).
If id |A : A ,→ M(B) is weakly residually nuclear, then L : A → M(D) is weakly
residually nuclear.
1012 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

Proof. If a1 , a2 , . . . is a dense sequence in A, then we find contractions bn ∈

B+ with kan bn k+1/n > kan k. The smallest C *-subalgebra C1 of B with AC1 ⊂ C1 ,
C ⊂ C1 and {b1 , b2 , . . .} ⊂ C1 is separable, and L : a 7→ La |C1 is a *-monomorphism
from A to M(C1 ).

Let C1 be a separable C *-subalgebra of B, and let M be a countable dense

subset of the positive cone of C1 .
By Remark 12.2.9 we can choose, for a, b ∈ M and k ∈ N, contractions c, d ∈ B,
such that ka2 − c∗ a2 ck, kb2 − d∗ b2 dk and kc∗ abdk are less then 1/k. The chosen
elements are in a countable subset S of B. Let C2 be the separable C *-subalgebra
of B which is generated by C1 and S.
Then for positive elements a, b ∈ C1 and δ > 0 there exist contractions c, d ∈ C2
such that ka2 − c∗ a2 ck, kb2 − d∗ b2 dk and kc∗ abdk are less then δ.
In the same way we construct an increasing sequence C1 ⊂ C2 ⊂ . . . of separable
C *-subalgebras Cn of B such that, for positive elements a, b ∈ Cn and δ > 0 there
exist contractions c, d ∈ Cn+1 such that ka2 − c∗ a2 ck, kb2 − d∗ b2 dk and kc∗ abdk are
less then δ.
Let D1 be the closure of the union of the Cn . Then, for positive elements
a, b ∈ D1 and δ > 0 there exist contractions c, d ∈ D1 such that ka2 − c∗ a2 ck,
kb2 − d∗ b2 dk and kc∗ abdk are less then δ. Thus D1 is a separable strongly purely
infinite C *-subalgebra of B and contains C1 .

Now let G be a C *-algebra. For every C *-subalgebra C of G and a, b ∈ G, k ∈

N, we denote by ρ(a, b, k, C) the infimum of the numbers ka − c∗j bcj k, where cj ∈
P

C, 1 ≤ j ≤ k, with k c∗j cj k ≤ 1. If C1 ⊂ C2 then ρ(a, b, k, C2 ) ≤ ρ(a, b, k, C1 ).

P

For fixed C and k, ρ(a, b, k, C) is uniformly continuous in a and b.

Let C ⊂ G and a, b ∈ C positive contractions. Then a is in the closed ideal
generated by b if and only if the infimum of {ρ(a, b1/n , k, C) : n, k ∈ N } is zero.
Thus, the equations ρ(a, b, k, C) = ρ(a, b, k, G) for every a, b ∈ C+ , k ∈ N,
imply that, for every J ∈ I(C),

ΨC,G
down (J) ∩ C = J.

Now let E1 be a separable C *-subalgebra of G and let M denote a dense

countable subset of (E1 )+ .
By definition of ρ, we can choose, for a, b ∈ M and k, m ∈ N, contractions
Pk Pk
c1 , . . . , ck ∈ G with k j=1 c∗j cj k ≤ 1 such that ka − j=1 c∗j bcj k is less then
ρ(a, b, k, G) + 1/m. The chosen elements are in a countable subset S of G. Let E2
be the separable C *-subalgebra of G which is generated by E1 and S.
Then ρ(a, b, k, E2 ) = ρ(a, b, k, G) for positive elements a, b ∈ E1 and k ∈ N.
In the same way we construct an increasing sequence E1 ⊂ E2 ⊂ . . . of sepa-
rable C *-subalgebras En of G such that ρ(a, b, k, En+1 ) = ρ(a, b, k, G) for positive
elements a, b ∈ En and k ∈ N.
 2. SELECTIONS AND Ψ-EQUIVARIANT EMBEDDING 1013

Let E be the closure of the union of the En . Then E is separable, E1 ⊂ E, and

ρ(a, b, k, E) = ρ(a, b, k, G) for positive elements a, b ∈ E and k ∈ N. Here we used
the uniform continuity of ρ in a and b.
Thus, for every separable C *-subalgebra E1 of a C *-algebra G, there exist a
separable C *-subalgebra E of G, such that E1 ⊂ E and every closed ideal J of E
is the intersection of E with a closed ideal of G, i.e., E ∩ ΨE,G
down (J) = J.

By [431, lem. 3.4], for every separable C *-subalgebra D1 of B, there is a

separable relatively weakly injective C *-subalgebra D2 of B with D1 ⊂ D2 .
Let F := C ∗ (F∞ ). Thus E1 := D2 ⊗max F is in natural way a separable
C *-subalgebra of G := B ⊗max F .
We find a separable C *-subalgebra E2 of G which contains E1 and has the
property that every closed ideal of E2 is the intersection of E2 with a closed ideal
of G. Since the algebraic tensor product B F is dense in G, there is a separable
C *-subalgebra D3 of B such that D2 ⊂ D3 , and the closure in G of the algebraic
tensor product D3 F contains E2 .
By [431, lem. 3.4], we find a separable relatively weakly injective C *-subalgebra
D4 of B with D3 ⊂ D4 . Then E2 ⊂ D4 ⊗max F .
In this way we get sequences D1 ⊂ D2 ⊂ . . . and E1 ⊂ E2 ⊂ . . . of separable
C *-subalgebras of B and G, respectively, such that D2n is relatively weakly injective
in B, D2n ⊗max F = E2n−1 , and every closed ideal of E2n is the intersection of E2n
with a closed ideal of G.
Let D ⊂ B denote the closure of the union of algebras Dn . Then D is relatively
weakly injective in B, and D⊗max F is just the closure E of the union of the algebras
En .
Let J ∈ I(E). The closure I of the union of In := ΨE 2n ,G
down (E2n ∩ J) contains
J, because the union of E2n ∩ J is dense in J.
In ∩ E is contained in J, because, for n ≤ m, In ⊂ Im , and

E2m ∩ In ⊂ E2m ∩ Im = E2m ∩ J,

and the union of E2m ∩ In is dense in In ∩ E. Thus J = I ∩ E. By the last

observation in Remark 12.2.10, D is residually relatively weakly injective in B.
Now let J ∈ I(D). Then ΨD,B down (J) ⊗
max
F is the closed ideal of B ⊗max F
which is generated by the closed ideal I := J ⊗max F of E = D ⊗max F , i.e., is
ΨE,G D,B
down (I). It contains (D ∩ Ψdown (J)) ⊗
max
F , and its intersection with E = D is
max D,B
I. Thus J ⊗ F contains (D ∩ Ψdown (J)) ⊗max F . If we apply a character χ of
F to this, we get that J = D ∩ ΨD,B
down (J)).

Now we change the notation and iterate the constructions:

We find inductively sequences (Cn ), (Dn ) and (En ) of separable C *-subalge-
bras of B, such that C ⊂ C1 , Cn ⊂ Dn ⊂ En , and En ⊂ Cn+1 , ACn ⊂ Cn ,
1014 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

kLa |C1 k = kak for a ∈ A, Dn is strongly p.i., En is residually relatively weakly

En ,B
injective in B, and Ψdown (J) ∩ En = J for every J ∈ I(En ).
Let D be the closure of the union of this sequence. D contains C. By Remark
12.2.9, D is strongly p.i. and, by Remark 12.2.10, D is residually relatively weakly
injective in B. It is easy to see that D satisfies condition (v).
Let J ∈ I(D) and Jn := En ∩ J. Then ΨD,B down (J) is the closure of the increasing
En ,B
sequence of closed ideals In := Ψdown (Jn ), and In ∩Em ⊂ Im ∩Em = Jm for n < m.
Thus In ∩ D ⊂ J and, therefore ΨD,Bdown (J) ∩ D = J. Thus D satisfies (i)-(v).

Now let D1 ⊂ D2 ⊂ . . . be an increasing sequence of separable C *-algebras Dn

of B, and suppose that Dn satisfies (i)-(v).
S
Let D the closure of Dn . By Remarks 12.2.9 and 12.2.10, D has also proper-
ties (ii) and (iv). Above we have seen that (iii) is preserved under inductive limits,
i.e. D satisfies (iii). It is easy to check that D satisfies (v).

Now suppose that B is stable. By Remark 5.1.1(8), this means that there exists
tn (tn )∗ strictly converges
P
a sequence t1 , t2 , . . . of isometries in M(B), such that
to 1. We replace A by the separable C *-subalgebra A1 of M(B) which is generated
by A and {t1 , t2 , . . .}. Then a separable C *-subalgebra D of B is stable, if D
satisfies (i)-(v) for A1 in place of A:
sn (sn )∗ strictly converges to
P
sn := Ltn |D are isometries in M(D) such that
1 in M(D).

Suppose that B is stable and has residually nuclear separation.

Let D1 := D be above found separable stable C *-subalgebra D of B with
(i)-(v) for A1 , let B1 denote the closure of DBD, and let M1 be a countable dense
subset of the positive cone of D1 .
Residually nuclear separation passes to hereditary C *-subalgebras. Therefore
we can find a countable set S1 of residually nuclear completely positive contractions
from D1 into B1 such that, for a ∈ M1 and n ∈ N, there is a V ∈ S1 with
ka − V (a)k < 1/n.
The union the images of the V ∈ S1 is a separable subset of B1 . Thus it is
contained in a separable C *-subalgebra C2 of B1 . Necessarily, C2 contains M1 and,
therefore, also D1 .
We find a separable C *-subalgebra D2 of B1 with (i)-(v) for A1 , C2 and B1 in
place of A, C and B, because B1 is again strongly purely infinite, C2 ⊂ B1 , and
A1 ∼= L(A1 )|B1 is contained in M(B1 ). If M2 is a countable dense subset of the
positive cone of D2 , then we can again find a countable set S2 of residually nuclear
completely positive contractions from D2 into B1 such that, for a ∈ M2 and n ∈ N,
there is a V ∈ S2 with ka − V (a)k < 1/n.
If we go so on, we get sequences D1 ⊂ D2 ⊂ . . ., S1 , S2 , . . . of separable C *-
subalgebras Dn of B1 and of countable sets Sn of residually nuclear completely
positive contractions from Dn to B1 such that, for n ∈ N :
 2. SELECTIONS AND Ψ-EQUIVARIANT EMBEDDING 1015

(i) C ⊂ D1 , D1 is stable and contains a strictly positive element of Dn

(ii) Dn is strongly purely infinite,
(iii) for every closed ideal J of Dn , J = Dn ∩ ΨD n ,B
down (J), and,
(iv) Dn is residually relatively weakly injective in B,
(v) A1 Dn ⊂ Dn and kak = kLa |D1 k for a ∈ A.
(vi) V (Dn ) ⊂ Dn+1 for V ∈ Sn , and,
(vii) for every a ∈ (Dn )+ and every δ > 0 there is V ∈ Sn with kV (a) − ak < δ.

Let D denote the closure of the union of the C *-algebras Dn . Then D satis-
fies conditions (i)-(v) of Lemma 12.2.14 and D is stable. By Lemma 12.2.12, the
completely positive maps V ∈ Sn are also residually nuclear if we consider them as
maps from Dn to D.
Now we argue as in the proof of the implication (ii)⇒(i) in the proof of Lemma
12.1.2:
Since, by Kasparov stabilization theorem, HD ∼
= D as (left) Hilbert D-module,
we can take Kasparov-Stinespring dilations of the V in Sn , infinite repeats and
direct sums. We get a sequence of non-degenerate weakly residually nuclear *-
monomorphisms hn : Dn → M(D), such that δ∞ hn is unitarily equivalent to hn ,
and hn (J ∩ Dn ) = hn (Dn ) ∩ M(D, J) for J ∈ I(D). By Lemma 12.1.2, it follows
that hn |Dm is unitarily homotopic to hm for m < n.
This allows to replace the hn by unitarily equivalent *-monomorphisms kn such
that the sequence kn is convergent in point-norm on every Dm .
The limit is a *-monomorphism h from D into M(D) with h(J) = h(D) ∩
M(D, J) for J ∈ D, and h|Dn is weakly residually nuclear for every n ∈ N. It
follows that h is residually nuclear, and that D is the closure of Dh(D)D.
By Lemma 12.1.2, we can construct from h with the above properties a
non-degenerate residually nuclear *-monomorphism H0 from D into M(D) with
H0 (J) = H0 (D) ∩ M(D, J) for J ∈ I(D) such that H0 is unitarily equivalent to
its infinite repeat δ∞ H0 .
Now suppose that B is σ-unital and let e ∈ B+ strictly positive.
Then we can find D with the properties (i)-(v) and e ∈ D (simply by replacing
C by C ∗ (C, e)). It holds L(A∩M(B, I)) = L(A)∩M(D, D ∩I) for every I ∈ I(B),
because, for a ∈ A

a ∈ M(B, I) ⇔ ae ∈ I ⇔ aD ⊂ D ∩ I ⇔ a ∈ M(D, D ∩ I) .

If id |A : A ,→ M(B) is weakly residually nuclear, then L : A → M(D) is

weakly residually nuclear, by (iv) and Lemma 12.2.12, because we replace B by its
hereditary C *–subalgebra DBD (with strictly positive e ∈ D+ ). 

The following Proposition 12.2.15 generalizes the first part of Theorem A. Here
the target algebra B replaces O2 in Theorem A, and B is in general not simple.
1016 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

Theorem A(i) is implied by Proposition 12.2.15, because a separable exact algebra

A always has a nuclear *-monomorphism into L(`2 ) ⊂ M(O2 ⊗ K).

Proposition 12.2.15 (Trivialization and Embedding).

Suppose that B is strongly purely infinite, stable, σ-unital and has residually
nuclear separation, and that A is stable, separable and exact.
Let H be a non-degenerate *-monomorphism from A into M(B), such that
a ∈ A 7→ b∗ H(a)b ∈ B is nuclear for every b ∈ B.
The following properties (i) and (ii) are equivalent:

(i) There exists a non-degenerate nuclear *-monomorphism h : A ⊗ O2 ,→ B

such that δ∞ h0 and δ∞ H are unitarily homotopic, where h0 (a) := h(a⊗1)
for a ∈ A, and δ∞ : M(B) → M(B) is the infinite repeat, as defined in
Remark 5.1.1(8).
(ii) The lower semi-continuous action
Ψ : J ∈ I(B) 7→ H −1 (H(A) ∩ M(B, J)) ∈ I(A),
of Prim(B) on A is also monotone upper semi-continuous (i.e., satisfies
condition (ii) of Definition 1.2.6).

Note that h0 = h0 ⊕s,t h0 if we use the isometries s, t ∈ M(B) which are given
by s := M(h)(1 ⊗ r1 ) and t := M(h)(1 ⊗ r2 ) if r1 and r2 denote the canonical
generators of O2 .

Proof. (i)⇒(ii): For a ∈ A, δ∞ (H(a)) ∈ M(B, J) if and only if δ∞ (h0 (a)) ∈

M(B, J), because they are unitarily homotopic in M(B). For stable σ-unital B,
we have that
δ∞ (M(B)) ∩ M(B, J) = δ∞ (M(B, J)),
i.e., b ∈ M(B, J) if and only if δ∞ (b) ∈ M(B, J) for b ∈ M(B). Since B ∩
M(B, J) = J, it follows, that h0 (A) ∩ J = h0 (Ψ(J)). By Lemma 12.1.1(iii),
J 7→ h−10 (h0 (A) ∩ J) satisfies condition (ii) of Definition 1.2.6.

(ii)⇒(i): We consider at first the case that B is separable.

By Remark 5.1.1(8), there is a strictly continuous unital C *-morphism from
M(K) into the relative commutant R of δ∞ (H(A)) in M(B). It follows that R
contains a unital copy of O2 and a sequence of isometries r1 , r2 , . . ., such that
rn (rn )∗ strictly
P
the isometries rn commute with the elements of O2 , and that
converges to 1 in M(B).
There is a non-degenerate *-monomorphism K : A ⊗ O2 → M(B) with K(a ⊗
b) = δ∞ (H(a))b for a ∈ A and b ∈ O2 . The isometries rn commute with the image
of K. Thus, by Lemma 5.1.2(i), δ∞ K is unitarily equivalent to K.
If we can find a non-degenerate nuclear *-monomorphism h from A ⊗ O2 into
B, such that δ∞ h is unitarily homotopic to δ∞ K, then the infinite repeat of h0 :=
h((.) ⊗ 1) is unitarily homotopic to (δ∞ )2 H. But δ∞
2
is unitarily equivalent to δ∞
by Lemma 5.1.2(i).
 2. SELECTIONS AND Ψ-EQUIVARIANT EMBEDDING 1017

Thus, by Theorem 6.3.1, it suffices to show that, for a1 , . . . , an ∈ K(A ⊗ O2 )

and for ε > 0, there exist completely positive contractions V : M(B) → B and
W : B → M(B) such that

(a) kW ◦ V (aj ) − aj k < ε, for j = 1, . . . , n.

(b) V is strictly continuous and is residually equivariant, i.e., limn kV (bn ) −
V (b)k = 0 if bn → b in M(B) strictly and V (J) ⊂ J ∩B for J ∈ I(M(B)).
Old version was only: V is weakly residually nuclear.
(c) W : B → M(B) is weakly residually nuclear, i.e., W (J)B ⊂ J for J ∈
I(B), and the c.p. maps Wa : b ∈ B 7→ a∗ W (b)a ∈ B are residually nuclear
for all b ∈ B. I.e. the maps [W ]J : B/J → M(B/J) ∼ = M(B)/M(B, J)
have the property that ([W ]J )d : b ∈ B/J 7→ d∗ [W ]J (b)d ∈ B/J is a
nuclear map for all d ∈ B/J. Here [W ]J (a + J) := W (a) + M(B, J), i.e.,
d∗ [W ]J (b)d = πJ (f ∗ W (a)f ) for b = a + J and d = f + J.
Old version was only: W is weakly residually equivariant,
i.e. W (J) ⊂ M(B, J) for J ∈ I(B).

change the old version for Thm. 6.3.1, as executed below: ??

NEXT comes from some beamer presentation:
(1) H0 (A) is non-degenerate, i.e., H0 (A)B is dense in B, and is in “general
position” (i.e., there exists a unitary U ∈ M(B) U ∗ H0 (·)U = δ∞ ◦ H0 ).
(2) The given lower s.c. action Ψ : I(B) → I(A) is realized by Ψ(J) :=
H0−1 (H0 (A) ∩ M(B, J)) and Ψ is monotone upper semi-continuous, i.e., n Ψ(Jn )
S
S
is dense in Ψ(J) for J := n Jn , if the sequence Jn ∈ I(B) is increasing:
J1 ⊂ J2 ⊂ · · · .
(3) For every b ∈ B, the ΨA -compatible map A 3 a 7→ b∗ H0 (a)b ∈ B is nuclear,
and can be approximated by compositions V2 ◦ V1 of the residually nuclear maps
V1 : a ∈ A 7→ b∗1 H1 (a)b1 ∈ B and V2 : b ∈ B 7→ b∗2 λ(b)b2 . I.e. V2 satisfies V2 (J) ⊂ J
for all J ∈ I(B) and that [V2 ]J : B/J → B/J is nuclear for all J ∈ I(B).
Step 2:
If A satisfies (1)–(3), then, – with a ∈ A identified with H0 (a) ∈ H0 (A) ⊂ M(B)
– for every a1 , . . . , an ∈ A and ε > 0, there exist completely positive contractions
V : M(B) → B and W : B → M(B) that satisfy the above listed conditions (a),
(b) and (c).
More on Step 2:
Consider the set of maps V := Vc : M(B) → B given by Vc : b ∈ M(B) 7→
∗
c M(λ)(b)c . The point-norm closure is an m.o.c. cone C1 .
What is λ : B → M(B)??
Perhaps λ : B → M(B) a residually nuclear separation?
Do the same with the maps W := WT : B → M(B) given by WT (b) :=
T ∗ λ(b)T , for T ∈ M(B), and we denote this m.o.c. cone by C2 .
1018 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

Both cones are singly generated (as m.o.c.c.), e.g. C1 by eM(λ)(·)e where e ∈
B+ is strictly positive, and C2 by λ (as u.c.p. map).

The above listed properties (b) and (c) for suitable V = Vc and W = WT follow
from the properties of λ and the fact that M(λ) is the unique strictly continuous
extensions of λ.
Continuation 1: More on Step 2:
Notice that λ(b) and δ∞ (b) for each self-adjoint b ∈ B must be unitarily
homotopic in M(B) by the generalized W-vN theorem, because both define *-
mono-morphisms from C ∗ (b) into M(B) that are in “general position” and de-
fine the same action of Prim(B) on C ∗ (b) by definition of λ, namely the action
J ∈ I(B) 7→ M(B, J) ∩ C ∗ (b).
It follows that each W ∈ C2 maps B into the closure I0 of M(B)δ∞ (B)M(B),
the closed ideal of M(B) generated by δ∞ (B).
Moreover, b ∈ M(B)+ will be mapped by all P ∈ C2 ◦C1 into the ideal of M(B)
that is the norm-closure of the union of the ideals M(B, J((ebe − 1/n)+ )), with
J(b) ∈ I(B) as defined above, e ∈ B+ strictly positive contractions.
The following Lemma supports the conjectures in ‘‘Continuation
1’’ and ‘‘Continuation 2’’.

Lemma 12.2.16. Let B a σ-unital stable C*-algebra, e ∈ B+ a strictly positive

contraction, and let a ∈ M(B)+ such that the natural l.s.c. action of Prim(B) on
C ∗ (a) ∼
= C0 (Spec(a) \ {0}) is also monotone upper semi-continuous.
Then a is in the closed ideal of M(B) that is generated by δ∞ (eae).
Suppose that – more generally – λ : B → M(B) is a non-degenerate *-
monomorphism such that δ∞ ◦ λ is approximately unitarily equivalent to λ and that
span(BbB)) = span(Bλ(b)B)) for each b ∈ B+ .
Then a is in the closed ideal of M(B) that is generated by λ(eae).

Proof. Let I(b) denote the closed ideal of M(B) generated by b ∈ M(B)+
and let J(b) the closed ideal of B generated by BbB. Note J(δ∞ (b)) = J(b) =
J(ebe) = J(b1/2 e2 b1/2 ) = J(beb).
For b ∈ M(B)+ and γ > 0 holds, by Lemma 5.9.10, that
δ∞ ((b − γ)+ ) ∈ M(B, J((b − γ)+ )) ⊆ I(δ∞ (b)) .
Let ε > 0. Since, the natural action of Prim(B) on C ∗ (a) ∼ = C0 (Spec(a) \ {0}) is
S
monotone upper semi-continuous, and since J(eae) is the closure of n J((eae −
1/n)+ ), we get that a is in the closed union of the ideals C ∗ (a)∩M(B, J((eae−γ)+ ))
for γ ∈ (0, 1).
It follows that (a − ε)+ ∈ M(B, J((eae − γ)+ )) for some suitable γ > 0. Then
(a − ε)+ ∈ I(δ∞ (eae)). Since I(δ∞ (eae)) is closed it follows a ∈ I(δ∞ (eae)).
Add the missing cross references:
 2. SELECTIONS AND Ψ-EQUIVARIANT EMBEDDING 1019

The conditions on λ imply that, for b ∈ B+ , the upper s.c. action of Prim(C ∗ (b))
on B defined by C ∗ (b) ⊂ M(B) and λ : C ∗ (b) → M(B) are the same. This implies
that the l.s.c. actions of Prim(B) on C ∗ (b) defined by id |C ∗ (b) or λ|C ∗ (b) are
the same, because they are both the Galois adjoint of the same upper s.c. action,
cf. Proposition ??.
It follows by Corollary ?? that δ∞ (λ(b)) and δ∞ (b) are approximately uni-
tarily equivalent in M(B). This implies I(δ∞ (λ(b))) = I(δ∞ (b)). Since λ(b) is
approximately unitary equivalent to δ∞ (λ(b)) it follows I(λ(b)) = I(δ∞ (b)).
This applies in particular to b := eae.
What about the a and e here ??
It implies that a ∈ I(δ∞ (eae)) = I(λ(eae)). 

Continuation 2: More on Step 2:

This ideal is contained in the closed ideal of M(B) generated by δ∞ (ebe) for a
strictly positive element e ∈ B+ of B.
??????????????????????
b ∈ I(δ∞ (ebe)) if and only if b ∈ I(δ∞ (e)) = the closed ideal of M(B) generated
by δ∞ (B).
??????????????????????
Conjecture: There exists a sequence Pn ∈ C2 ◦ C1 with kPn (b) − bk → 0 if and
only if b ∈ I(δ∞ (e)).
¡– Seems to require further assumptions.
Continuation 3: More on Step 2:
The condition (a) is equivalent to H0 ∈ C3 := C2 ◦ C1 ◦ CH0 . The m.o.c. cone
C3 is contained in the cone of (norm-) nuclear c.p. maps from A into M(B), and
the elements of C3 map A into the norm-closed ideal I(δ∞ (B)) of M(B) that is
generated by δ∞ (B).
More precisely, the elements of C3 map a ∈ A+ into the closed ideal of
2 3 2
I(δ∞ (B)) ⊂ M(B) generated by the element δ∞ (e)δ∞ (e)δ∞ (a)δ∞ (e)δ∞ (e).
If there is b ∈ B+ such that δ∞ (b) and a generate the same closed ideal of
M(B), then there are Tn ∈ C2 ◦ C1 such that kTn (a) − ak → 0.
END of TEXT from old beamer presentation
Idea: take unital weakly residual nuclear monomorphism H
of M(B) in general position and use that H|C and idC are
approximately unitarily equivalent by unitaries u1 , u2 , . . ..
Then replace W by suitable u∗n H(W (·))un .
Let C := K(A ⊗ O2 ), let a1 , a2 , . . . be a dense sequence in the unit ball of C,
and let Xn denote the span of {a∗1 , a1 , . . . , a∗n , an }.
1020 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

idM(B) |C is weakly residually nuclear, because C is exact and δ∞ |A = K(( · )⊗

1) is nuclear by the exactness of A. In fact, id |C : C ,→ M(B) is even nuclear, see
Chapters 3 and 5, where we have discussed those topics in every detail.
Since K is unitarily equivalent to δ∞ , id |C is unitarily equivalent to δ∞ |C.
By Remark 12.2.13, there exists a non-degenerate weakly residually nuclear
*-monomorphism d : B → M(B) such that δ∞ d is unitarily homotopic to d, and,
for J ∈ B,
d(J) = d(B) ∩ M(B, J).

Thus M(d) (M(B, J)) = M(d)(M(B)) ∩ M(B, J) for J ∈ I(B).

It follows, that the *-monomorphisms M(d)|C and id |C from C to M(B) are
both nuclear *-monomorphisms and induce the same action of Prim(B) on C, i.e.,
for J ∈ I(B),
M(d)(C ∩ M(B, J)) = M(d)(C) ∩ M(B, J).

Therefore, by Corollary 5.9.16, δ∞ M(d)|C and δ∞ |C are unitarily homotopic. This

implies, together with the unitary equivalence of δ∞ |C and id |C, the existence of
a sequence U1 , U2 , . . . of unitaries in M(B), such that, for n = 1, 2, . . . and b ∈ Xn ,

kUn∗ δ∞ M(d)(b)Un − bk < 4−n kbk . (1)

Let e ∈ B+ a strictly positive contraction in B of norm one. By Remark

5.1.1(3), we find a sequence of functions gn ∈ C0 ((0, 1])+ , such that gn gn+1 = gn ,
kgn k = 1, gn → 1 point-wise, and, for b ∈ Xn and m > n,

k[gn (e), b]k + k[(gm (e) − gn (e))1/2 , b]k < 4−n kbk.

We define real numbers γ(n, k, m, T ), γ(n, k, m) and γ(n, k) for k, m, n ∈ N and

for inner completely positive contractions T : B → B as follows:

γ(n, k, m, T ) := sup{kT (gk (e)agk (e)) − gm (e)agm (e)k ; a ∈ Xn , kak ≤ 1}

γ(n, k, m) := inf γ(n, k, m, T )
T
γ(n, k) := sup γ(n, k, m).
m

Then γ(n, k, m) = 0 for m ≤ k, γ(n, k, m) ≤ γ(n + 1, k, m), γ(n, k, m) ≤

γ(n, k + 1, m) and γ(n, k, m) ≤ γ(n, k, m + 1).
Thus, γ(n, k + 1) ≤ dzγ(n, k) ≤ γ(n + 1, k).
Now we show that limk→∞ γ(n, k) = 0.
Since k 7→ γ(n, k) is decreasing and m 7→ γ(n, k, m) is increasing, it is enough
to show that limj→∞ γ(n, kj , kj+1 ) = 0 for every sequence k1 < k2 < . . ..
Let pj := gkj (e) for j ∈ N, and P : C → `∞ (B), Q : C → `∞ (B) the com-
pletely positive maps which are defined by P (c) := (p1 cp1 , p2 cp2 , . . . ) and Q(c) :=
(p2 cp2 , p3 cp3 , . . . ).
 2. SELECTIONS AND Ψ-EQUIVARIANT EMBEDDING 1021

To see limj→∞ γ(n, kj , kj+1 ) = 0, it suffices to show that, for every ε > 0, there
exists an inner completely positive contraction T on `∞ (B), such that, for a ∈ Xn ,
dist(T (P (a)) − Q(a), co (B)) < kakε.

Let h1 denote the C *-morphism from C ⊗ C0 ((0, 1]) ∼ = C0 ((0, 1], C) to

`∞ (B)/c0 (B), with h1 (c ⊗ f ) = π((cf (p1 ), cf (p2 ), . . .)) for c ∈ C and f ∈ C0 ((0, 1]),
where π means the natural quotient map from `∞ (B) onto `∞ (B)/c0 (B).
Further let h2 the *-epimorphism from C ⊗ C0 ((0, 1]) onto C, with h2 (c ⊗ f ) =
f (1)c.
Note that h1 (c ⊗ f02 ) = π(P (c)), where f0 (t) = t. We show that h1 approxi-
mately inner dominates V := πQh2 . Then the proof of limk γ(n, k) = 0 is ready,
because it follows that, for ε > 0, there is an inner completely positive contraction
T 0 on `∞ (B)/c0 (B), such that, for a ∈ Xn ,
kT 0 (π(P (a))) − π(Q(a))k < kakε/2.
(because C is exact).
Since every completely positive contraction c 7→ pj cpj , for j > 1, is a nuclear
map, its direct sum Q is a nuclear map from C into `∞ (B) ⊂ M(c0 (B)). By
Corollary 5.6.3, Q is nuclear, because C is exact. Thus V is nuclear.
Now we show that V (a ⊗ f ) = f (1)π(Q(a)) is in the closed ideal generated by
h1 (a ⊗ f ) for a ∈ C+ , f ∈ C0 ((0, 1])+ :
V (a ⊗ f ) = 0, if f (1) = 0. But if f (1) > 0, then there is δ ∈ (0, 1), such
that f (t) > f (1)/2 for t ∈ (δ, 1]. Define λ ∈ C0 ((0, 1]) by λ(t) := 0 on [0, δ],
λ(t) :=], min(1, 2(t − δ)/(1 − δ)) for t ∈ (δ, 1]. Then f (1)(a ⊗ λ) ≤ a ⊗ f , and it
suffices to show that πQ(a) is in the closed ideal generated by π(b), where
b := (a1/2 λ(p1 )a1/2 , a1/2 λ(p2 )a1/2 , . . .).
By Corollary 3.10.12 and Proposition 2.12.8(iii),
what is needed: prop:2.24 ??? ??
this follows if we can show that the sequence of real numbers
λn = max(0, sup{kpn+1 apn+1 + Jk − ka1/2 λ(pn )a1/2 + Jk : J ∈ Prim(B)})
converges to zero. We have
kpn+1 apn+1 + Jk = ka1/2 p2n+1 a1/2 + M(B, J)k ≤ ka + M(B, J)k.

By definition of λ, λ(pn )λ(pn+1 ) = λ(pn ), kλ(pn )k = 1 and lim kλ(pn )e−ek = 0.

Since a1/2 λ(pn )a1/2 ≤ a and λ(pn ) is an increasing approximate unit of B,
we get that ξn (J) := ka1/2 λ(pn )a1/2 + Jk converges to ka + M(B, J)k for every
J ∈ Prim(B).
The natural lower semi-continuous action Ψup
B,C from Prim(B) on C satisfies,
for J ∈ I(B),
δ∞ (H(Ψ(J))) = δ∞ (H(A)) ∩ ΨupB,C (J) .
1022 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

Here we used that δ∞ (M(B, J)) is the same as δ∞ (M(B)) ∩ M(B, J) . Since O2
is simple and nuclear, and since K(A ⊗ 1) = δ∞ (H(A)), we get that, for J ∈ I(B),

Ψup
B,C (J) = K(Ψ(J) ⊗ O2 ) .

Thus Ψup
B,C satisfies the condition (ii) of Definition 1.2.6, because Ψ satisfies
this condition by assumption.
By Corollary 12.2.8(ii), the property (ii) of Definition 1.2.6 for Ψup
B,C implies
that, for every a ∈ C+ , the real function

a(Ψup
J 7→ b B,C (J) = ka + M(B, J)k

is a Dini function on Prim(B).

Therefore, the functions ξn converge on Prim(B) uniformly to the function
J 7→ ka + M(B, J)k, because ξn is an increasing sequence of lower semi-continuous
functions which converges point-wise to J 7→ ka + M(B, J)k on Prim(B).
Let θn := sup{ka + M(B, J)k − ξn (J) ; J ∈ Prim(B)} .
Then 0 ≤ lim λn ≤ lim θn = 0. Thus V (a ⊗ f ) is in the closed ideal generated
by h1 (a ⊗ f ) for a ∈ C+ , f ∈ C0 ((0, 1])+ .
By Proposition B.4.2(ii), it follows now that, for every closed ideal J of C ⊗
C0 ((0, 1]), V (J) is in the closed ideal of `∞ (B)/c0 (B) that is generated by h1 (J).
Thus, the kernel of h1 is contained in the kernel of V , and the natural completely
positive map S from the image of h1 into `∞ (B)/c0 (B) is residually equivariant.
Since C ⊗ C0 ((0, 1]) is exact, by Remark 3.1.2(iv), h1 (C ⊗ C0 ((0, 1])) is exact
and S is nuclear.
By Corollary 3.10.7, S is approximately inner.
This completes the proof of limk→∞ γ(n, k) = 0.

We can take a sequence 1 < k1 < k2 < . . . of positive integers such that
γ(n, kn ) < 4−n , and get an approximate unit en := gkn (e) ∈ B+ , of B with
en+1 en = en , ken k = 1, and, for b ∈ Xn ,

k[en , b]k + k[(en+1 − en )1/2 , b]k < 4−n kbk. (2)

Since γ(n, kn ) < 4−n , we find inner completely positive contractions Tn : B →

B, such that, for b ∈ Xn

kTn (en ben ) − en+1 ben+1 k < 4−n kbk. (3)

check next construction with new (a),(b),(c) of 6.3.1 ??

We define the desired weakly residually equivariant completely positive con-
tractions Wn : B → M(B) and Vn : M(B) → B, such that Vn is weakly residually
nuclear, and (Wn ◦ Vn )|C tends to id |C in point-norm topology:
Let n ∈ N fixed and k > n. The inner completely positive contractions

Lk := Tk+1 ◦ Tk ◦ . . . ◦ Tn+1 Tn
 2. SELECTIONS AND Ψ-EQUIVARIANT EMBEDDING 1023

Pmk ∗
of B are given by Lk (b) := j=1 (ck,j ) bck,j for b ∈ B, where ck,j ∈ B, j =
Pmk
1, . . . , mk , j=1 (ck,j )∗ ck,j is a contraction. Note that, by induction, (3) implies

kLk (en ben ) − ek+2 bek+2 k < 41−n /3, (4)

for b ∈ Xn and k > n.

Let t1 , t2 , . . . denote the sequence of isometries in M(B) with tk (tk )∗ strictly
P

convergent to 1, such that our infinite repeat δ∞ is defined by tk (·)t∗k , cf. Remark
P

5.1.1(8).
1/2
Let fn := en , and fk := (ek+1 − ek )1/2 for k > n. We define, by induction,
integers p(n + 1) := 1 and, p(k + 1) = p(k) + mk for k > n, and elements of
Pmk
M(B) by dn := 1, dk := j=1 (tp(k)+j )ck,j . Note dk is a contraction, because
(dk )∗ dl = δk,l j (ck,j )∗ ck,j .
P
P
Then k>n dk fk is strictly convergent in M(B) and
X
Sn := dk fk
k≥n

is a contraction in M(B), cf. Remark 5.1.1(???????). ??

Let, for b ∈ M(B),
Wn (b) := Sn∗ δ∞ (b)Sn ,
and
Vn (b) := en Un∗ δ∞ (M(d)(b))Un en .

Then, Wn |B is a weakly residually equivariant contraction from B into M(B),

Vn is a weakly residually nuclear contraction from M(B) to B, and, for b ∈ Xn ,

kWn Vn (b) − bk < 41−n 3kbk.

This estimate summarizes the estimate (1) and the estimates for kΓ(x − y)k and
kΓ(y) − bk otained by summing the inequalities (4) and (2). Here Γ denotes the
completely positive contraction from `∞ (B) into M(B), which is given by
X
Γ(b1 , b2 , . . .) := (en )1/2 bn (en )1/2 + fk bk fk ,
k>n

and x, y denote the elements of `∞ (B), which are defined by xk := Lk (en ben ) and
yk := ek+2 bek+2 for k = 1, 2, . . ..
Note that Γ(x) = Wn (en xen ) and Γ(y) = (en )1/2 b(en )1/2 + k>n fk bfk . (See
P

Remark 5.1.1(?????) ?? for the calculation of the needed estimates from those of
(2) and (4).)
Thus Theorem 6.3.1 applies in the separable case.
Here end of use of 6.3.1

If B is not separable, then we can consider instead of B a separable C *-

subalgebra D of B, which contains a strictly positive element of B, is invariant
by right and left multiplication by elements of H(A), and satisfies all the prop-
erties listed in Lemma 12.2.14 with respect to H(A), B and D. In particular,
1024 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

D is residually weakly injective in B, H(A)D ⊂ D, and D ∩ ΨD,Bdown (I) = I for

I ∈ I(D). It follows H(A) ⊂ M(D) ⊂ M(B) and, therefore, by Lemma 12.1.1(iv),
H(A) ∩ M(B, J) = H(A) ∩ M(D, D ∩ J) for J ∈ I(B). Thus, Ψ(J) = Φ(D ∩ J)
for J ∈ I(B), where Φ(I) := H −1 (H(A) ∩ M(D, I)) for I ∈ I(D).
By Lemma 12.2.12, it follows that H : A → M(D) is is a weakly Φ-residually
nuclear map from A to M(D).
Since ΨD,B D,B
down is upper semi-continuous, and Φ(I) = Ψ(Ψdown (I)) for I ∈ I(D),
we get from (ii), that the lower semi-continuous action Φ also satisfies condition
(ii) of Definition 1.2.6.
By construction, D is again strongly purely infinite and has residually nuclear
separation.
Thus, by the above considered separable case, there exists a non-degenerated
nuclear *-monomorphism h : A ⊗ O2 → D ⊂ B, such that the infinite repeat of
h0 := h((·) ⊗ 1) is unitarily homotopic to the infinite repeat of H in M(D). Since
D is a stable and non-degenerate C *-subalgebra of B, this implies that δ∞ h0 is
unitarily homotopic to δ∞ H in M(B). 

Remark 12.2.17. For every C *-algebra B, every separable C *-subalgebra

D of Q(R+ , B) and every commutative separable C *-subalgebra C of D, there
exists a strongly continuous map t 7→ V (t) from R+ into the approximately inner
completely positive contractions on B, such that the completely positive contraction
T on Q(R+ , B) defined by V fixes the elements of C and T (D) is contained in a
commutative C *-algebra of Q(R+ , B).
This can be seen as follows: If g is a contraction in Cb (R+ , B ⊗ O∞ ) and f a
pure state on O∞ then

V (t)(b) := (id ⊗f )(g(t)∗ (b ⊗ 1)g(t))

defines an approximately inner completely positive contraction V (t) of B such that

t 7→ V (t) is strongly continuous. For the corresponding completely positive contrac-
tion T on Q(R+ , B) and the contraction d := g+C0 (R+ , B) in Q(R+ , B⊗O∞ ) holds
T (a) = P (d∗ (a⊗1)d) for a ∈ Q(R+ , B). Here the completely positive contraction P
from Q(R+ , B ⊗O∞ ) onto Q(R+ , B) is given on representatives e ∈ Cb (R+ , B ⊗O2 )
by e(t) 7→ (id ⊗f )(e(t)) for t ∈ R+ , and Q(R+ , B) ⊗ O∞ is naturally embedded in
Q(R+ , B ⊗ O∞ ). Now choose d for D ⊗ 1 and C ⊗ 1 as in Lemma 12.1.6 to get the
desired result.
If B is purely infinite then, for every positive element a in Q(R+ , B), T (a) is
in the closed ideal generated by a: For δ > 0, (T (a) − δ)+ = e∗ ae + f ∗ af for some
e, f ∈ Q(R+ , B). It follows that then T |D is residually nuclear. This shows:
For every purely infinite C*-algebra B the asymptotic corona Q(R+ , B) has
residually nuclear separation in the sense of Definition 1.2.3 (cf. Remark 12.2.10).
 2. SELECTIONS AND Ψ-EQUIVARIANT EMBEDDING 1025

If, moreover, B is strongly purely infinite then there exists a contraction d in

Q(R+ , B) with T (b) = d∗ bd for b ∈ D. d commutes elementwise with C, as we have
shown on the beginning of the proof of Lemma 12.1.6.

Lemma 12.2.18. Suppose that B is stable and σ-unital. Let B1 := B ⊗ F ,

where F := O∞ ⊗ O∞ ⊗ . . .. Then there is a non-degenerate *-monomorphism
d : B1 ,→ M(B), such that

(i) δ∞ d is unitarily equivalent to d, and

(ii) d(B1 ) ∩ M(B, J) = d(J ⊗ F ) for J ∈ I(B).

Proof. Since B is stable, we can apply the infinite repeat δ∞ to M(B). There
is a unital copy of L(`2 ) in the relative commutant of δ∞ (M(B)) in M(B). There-
fore, a unital copy of F ⊂ M(B) commutes with δ∞ (B).
Since F is nuclear and simple, it follows that there is a unique C *-morphism
h : B1 → M(B) with h(b ⊗ f ) = δ∞ (b)f for b ∈ B, f ∈ F . h is a non-degenerate
*-monomorphism, because δ∞ |B := h((.)⊗1) is a non-degenerate *-monomorphism
and F is simple.
Let d := δ∞ h. Then d is non-degenerate *-monomorphism, because d(B ⊗ 1) =
2
δ∞ (B) is a non-degenerate C *-subalgebra of M(B).
2
δ∞ d is unitarily equivalent to d, because δ∞ is unitarily equivalent to δ∞ .
Let J ∈ I(B). Then I := d−1 (d(B1 ) ∩ M(B, J)) is a closed ideal of B1 . By
Proposition B.4.2(iii), there is a closed ideal K of B such that K ⊗ F = Ψ1 (J). We
2
get that b ∈ K if and only if δ∞ (b)B ⊂ J. Since b ∈ B, this is the case if and only
if b ∈ J. Thus K = J. 

Proof of Theorem K:. Let Ψ(J) := ΨA (ZJ ) for J ∈ I(B). Here ZJ ⊂

Prim(B) denotes the hull of J.
The uniqueness up to unitary homotopy of the nuclear *-monomorphism h0 :=
h((.) ⊗ 1) follows from h0 (Ψ(J)) = h0 (A) ∩ J for J ∈ I(B) and from the unitary
homotopy of h0 with h0 ⊕ h0 , cf. Corollary 9.1.4.

We reduce the general case to the case where B is separable and Ψ is non-
degenerate.
If B is separable, it suffices to consider, instead of of B, the smallest ideal I of
B with Ψ(I) = A, to prove the existence of h. The existence of I follows from the
lower semicontinuity of Ψ.
Thus, we may assume, in addition, that Ψ−1 (A) = {B}.
Let F := O∞ ⊗ O∞ ⊗ . . ., B1 := B ⊗ F , and let E1 and E denote the hereditary
C *-subalgebras of Q(R+ , B1 ) and Q(R+ , B) which are generated by B1 and B,
respectively. By Remarks 12.2.9 and 12.2.17, E1 and E are strongly purely infinite
and have residually nuclear separation.
1026 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

Since F is simple and nuclear, we get from Proposition B.4.2(iii), that

J ↔J ⊗F

defines a natural topological isomorphism from Prim(B) onto Prim(B1 ) and,

therefore, a lattice isomorphism from I(B) onto I(B1 ). We denote the ac-
tion of Prim(B1 ) on A by Ψ1 , i.e., Ψ1 (J ⊗ F ) = Ψ(J). Then Ψ1 is again
lower semi-continuous, satisfies property (ii) of Definition 1.2.6, Ψ1 (0) = 0 and
Ψ−1
1 (A) = {B1 }.

The action Φ(J) := Ψ1 (B1 ∩ J) for J ∈ I(E1 ) is a lower semi-continuous action

of Prim(E1 ) on A, and satisfies property (ii) of Definition 1.2.6, Φ(0) = 0 and
Φ−1 (A) = {E1 }. This is the case, because Ψ1 and Ψup E1 ,B1 : J 7→ B1 ∩ J satisfy
conditions (ii)-(iv) of Definition 1.2.6, and B1 ⊂ J implies J = E1 .
By Theorem 12.1.8, there is a non-degenerate weakly residually nuclear mono-
morphism G0 from A to M(E1 ) such that G0 is unitarily equivalent to δ∞ G0 , and,
for J ∈ I(E1 ),
G0 (Φ(J)) = G0 (A) ∩ M(E1 , J).

By Proposition 12.2.15, there is a non-degenerate nuclear *-monomorphism

k : A ⊗ O2 → E1 such that, for k0 := k((.) ⊗ 1), δ∞ k0 is unitarily homotopic to G0
in M(E1 ), and k0 is unitarily equivalent to k0 ⊕ k0 .
In particular, for J ∈ I(E1 ),

δ∞ k0 (Φ(J)) = δ∞ (k0 (A)) ∩ M(E1 , J).

This implies k0 (Ψ1 (B1 ∩ I)) = k0 (A) ∩ I for every closed ideal I of Q(R+ , B1 ).
E1 is a hereditary C *-subalgebra of Q(R+ , B1 ). Therefore, k is also nuclear as
a *-monomorphism from A ⊗ O2 into Q(R+ , B1 ).
Since Ψ−1
1 (A) = {B1 }, it follows from Corollary 9.1.7, that there exists a nuclear
*-monomorphism h from A⊗O2 into B1 , such that h0 := h((.)⊗1) is non-degenerate
and satisfies h0 (Ψ1 (I)) = h0 (A) ∩ I for I ∈ I(B). Thus, for J ∈ I(B),

h0 (Ψ(J)) = h0 (A) ∩ (J ⊗ F ).

Now, it follows from Lemma 12.2.18, that H0 = dh0 is a non-degenerate nuclear

*-monomorphism from from A into M(B), such that δ∞ H0 is unitarily equivalent
to H0 , and, for J ∈ I(B),

H0 (Ψ(J)) = H0 (A) ∩ M(B, J).

Since B contains a strictly positive element of E, we get from Lemma 12.1.1

that M(B) ∩ M(E, I) = M(B, B ∩ I) for I ∈ I(E), and we can consider H0 as a
non-degenerate nuclear *-monomorphism from A into M(E). Thus, for I ∈ I(E),

H0 (Ψ(B ∩ I)) = H0 (A) ∩ M(E, I).

Now we can repeat the above applications of Proposition 12.2.15 and Corollary
9.1.7, where we have to replace B1 by B, E1 by E, and G0 by H0 : A → M(E).
 2. SELECTIONS AND Ψ-EQUIVARIANT EMBEDDING 1027

We get the desired non-degenerate nuclear *-monomorphism h from A ⊗ O2

into B with h0 (Ψ(J)) = h0 (A) ∩ J for J ∈ I(B) and h0 = h((.) ⊗ 1). 

Corollary 12.2.19. Suppose that B is strongly purely infinite, separable and

stable.
Then {bb : b ∈ B} is the set of all Dini functions on Prim(B).

Proof. Let B1 := B ⊗ O2 , and let f be a Dini function on X.

By Proposition 12.2.6, there exists a ∈ (B1 )+ with f (J) = b
a(J ⊗ O2 ) for J ∈
Prim(B). Let C denote the commutative C *-subalgebra of B1 which is generated
by a, and let A := C ⊗ K. Then A is a nuclear C *-subalgebra of B1 ⊗ K.
Let Φ(J) := J ⊗ O2 ⊗ K.
By Lemma 12.2.4, Φ is a lattice isomorphism from I(B) onto I(B1 ⊗ K).
It follows that Ψ(J) := A ∩ Φ(J) for J ∈ I(B) defines a lower semi-continuous
action of Prim(B) on A. It satisfies Ψ(0) = 0, Ψ(B) = A, and the condition (ii) of
Definition 1.2.6.
By Theorem K, there is a *-monomorphism h0 : A ,→ B, such that, for J ∈
I(B),
h0 (Ψ(J)) = h0 (D) ∩ J.

Let b := h0 (a ⊗ p11 ). Then, for J ∈ Prim(B),

kb + Jk = k(a ⊗ p11 ) + Φ(J)k = b

a(J ⊗ O2 ).

This means bb = f . 

The following Corollary 12.2.20 is an equivalent reformulation of Corollary L.

Corollary 12.2.20. Suppose that A and B are separable, stable and nuclear
C*-algebras, and that there is a topological isomorphism γ from X := Prim(A) onto
Prim(B).
Then there exists an isomorphism ϕ from A ⊗ O2 onto B ⊗ O2 , such that ϕ
induces γ, i.e., for J ∈ Prim(A),

ϕ(J ⊗ O2 ) = γ(J) ⊗ O2 .

ϕ with this properties is unique up to unitary homotopy.

In particular, the natural group morphism from Aut(A ⊗ O2 ) to the homeomor-
phisms of Prim(A) is an epimorphism with kernel equal to the group of automor-
phisms of A ⊗ O2 that are unitarily homotopic to id.
In particular, every approximately inner automorphism of A ⊗ O2 is unitarily
homotopic to id.

Proof of Corollary L and of Corollary 12.2.20. By Theorem K,

there exist non-degenerate *-monomorphisms h : A ⊗ O2 ,→ B ⊗ O2 and
k : B ⊗ O2 ,→ A ⊗ O2 such that h and k induce the corresponding actions of
1028 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

X on B, i.e., for J ∈ Prim(A), h(J ⊗ O2 ) is the intersection of h(A ⊗ O2 ) with

γ(J) ⊗ O2 , and similar for k. By Corollary 10.3.10(ii), h is unitarily homotopic to
an *-isomorphism ϕ from A ⊗ O2 onto B ⊗ O2 with the desired property. Here we
have used that every unital *-endomorphism of O2 is unitarily homotopic to id,
and that O2 ∼= O2 ⊗ O2 , cf. Corollary F(iii) and Corollary H(ii).
The uniqueness of ϕ follows from Corollary 10.3.10(i). 

3. Interjection: Non-commutative Selection Conjecture

Corollary 12.3.1. Suppose that A is a separable, stable and exact C*-algebra.

Then there are

(i) a separable stable nuclear C*-algebra B such that B ∼= B ⊗ O2 , and

(ii) a non-degenerate *-monomorphism ϕ : A ,→ B such that ϕ induces an
isomorphism λ from I(B) onto I(A) by λ(I) := ϕ−1 (I ∩ ϕ(A)), i.e., for
closed ideals I1 , I2 ∈ I(B), I1 ∩ ϕ(A) = I2 ∩ ϕ(A) implies I1 = I2 , and,
λ(I) = J for J ∈ I(A) and I := Bϕ(J)B ∈ I(B).

B is unique up to λ-equivariant isomorphisms, and ϕ is unique up unitary homo-

topy.
If, in addition, A strongly purely infinite, then there is also a non-degenerate
*-monomorphism ψ : B ,→ A that induces a ”kind of inverse” for λ. The map ϕ ◦ ψ
is unitarily homotopic to idB .
The nuclear map ψ ◦ ϕ is unitarily homotopic to idA : A → A, if and only if, A
is nuclear and A ∼
= A ⊗ O2 .

Proof. The uniqueness of B follows from Corollary L. The uniqueness, up to

unitary homotopy, of ϕ follows from Corollary 7.4.8.
We can replace A by A ⊗ O2 , to prove the existence of B and ϕ. Here we
use that O2 ⊗ O2 ∼= O2 and that every unital *-endomorphism of O2 is unitarily
homotopic to the identity map of O2 , cf. Corollary F(ii) and Corollary H(ii).
Suppose A ∼
= A ⊗ O2 . Then A is strongly purely infinite. By Theorem K, there
is a non-degenerate nuclear *-monomorphism h : A → A, such that h induces the
natural action of Prim(A) on A, i.e. h(J) = h(A) ∩ J for every closed ideal J of A.
Let B denote the inductive limit indlim(hn : A → A) where hn = h for n =
1, 2, . . .. Since h is nuclear, B is a nuclear C *-algebra. By Corollary 10.3.11, B is
separable, stable and B ∼ = B ⊗ O2 .
By Corollary 7.4.8, the nuclear maps hn and h are unitarily homotopic for
n = 2, 3, . . ., because hn (J) = hn (A) ∩ J (by induction). This implies that J is the
closure of Ahn (J)A for every J ∈ I(A).
If we apply the canonical maps h∞ n ∞
n+k h = hk , we obtain that, for J ∈ I(A),

h∞ ∞ ∞
k (J) = hk (A) ∩ hn+k (J),
 3. INTERJECTION: NON-COMMUTATIVE SELECTION CONJECTURE 1029

and h∞ ∞ ∞ ∞
n (J) is the closure of hn (A)h1 (J)hn (A).

Let ϕ denote the canonical C *-morphism h∞ 1 from A into B := indlim(hn : A →

A), and, for J ∈ I(A), let µ(J) denote the closure of Bϕ(J)B. Then µ(J) is the
inductive limit of h∞
n (J), and ϕ(A) ∩ µ(J) = ϕ(J).

If I ∈ I(B), then I is the inductive limit of h∞

T ∞
n (A) ∩ I, because n hn (A) is
dense in A. If we apply the above observations to

Jn := (h∞
n )
−1 ∞
(hn (A) ∩ I)

we get Jn = Jm for m, n ∈ N and

h∞
n (ϕ
−1
(ϕ(A) ∩ I)) = h∞
n (A) ∩ I.

Thus I = µ(ϕ−1 (ϕ(A) ∩ I)), and λ defines a lattice isomorphism from I(B) onto
I(A) with inverse µ. (In particular, it preserves prime elements and the hk-topology
on them.)
If A strongly purely infinite, then, by Theorem K there is also a non-degenerate
*-monomorphism ψ : B ,→ A that defines the “inverse” µ = λ−1 of λ : I(B) →
I(A), i.e., J ∩ ψ(B) = ψ(µ(J)).
Then ϕ ◦ ψ induces the identity on I(B), and ϕ ◦ ψ is unitarily homotopic to
idB by Theorem K.
If the nuclear map ψ ◦ϕ is unitarily homotopic to idA : A → A then A is nuclear
and idA is approximately unitarily equivalent to idA ⊕ idA , because ψ and ϕ can be
taken non-degenerate (which implies that ψ ◦ (idB ⊕ idB ) ◦ ϕ is unitarily equivalent
to (ψ ◦ ϕ) ⊕ (ψ ◦ ϕ)), and idB is unitarily homotopic to idB ⊕ idB .
If A is nuclear and A ∼
= A ⊗ O2 , then Theorem K applies also to ψ ◦ ϕ, and is
unitarily homotopic to idA : A → A. 

Conjecture 12.3.2. Suppose that B is a separable stable C*-algebra with

B ∼= B ⊗ O∞ . Let E denote the closure of B Q(R+ , B)B, let H : A → M(E) a
non-degenerate C*-morphism from a separable stable C*-algebra A into the multi-
plier algebra M(E) of E, and let Ψ1 denote the lower semi-continuous actions of
Prim(E) on A which is defined, for I ∈ I(E), by

Ψ1 (I) = H −1 (H(A) ∩ M(E, I)).

Furthermore, suppose that

(i) H is unitarily equivalent to its infinite repeat δ∞ H,

(ii) H and σ b of M(E),
bH are unitarily homotopic, for every automorphism σ
which is induced by a scaling homeomorphism σ of R+ , and
(iii) H is weakly Ψ1 -residually nuclear.

The conjecture is:

Then there exists a C*-morphism H0 from A into M(B), which is in M(E) uni-
tarily homotopic to H.
1030 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

Remark 12.3.3. If the Conjecture 12.3.2 is true, then (i)-(iii) imply the addi-
tional condition

(iv) Ψ1 (I1 ) = Ψ1 (I2 ) if I1 ∩ B = I2 ∩ B.

This is because M(B) ∩ M(E, I) = M(B, I ∩ B). The latter formula comes from
B ⊂ BEB = E.
One could believe that the proof of Conjecture 12.3.2 can be given with ideas
from Section 1. But there is a difficult point:
Let b ∈ B+ a strictly positive contraction of B. Let X be a compact subset of A+
with linear span dense in A. For T in M(E) let kT k# := kT bk + kbT k.
The difficulty consists in the proof the following additional condition:

(v) For every linearly generating compact subset X of A, there exists a se-
quence of functions fn ∈ C0 ((0, 1])+ with
kfn k = 1, fn fn+1 = fn and k[fn (b), H(a)]k# < 2−n
for n ∈ N and a ∈ X, such that, in addition, for every ε > 0 and every
homeomorphism σ of R+ , there is a unitary U ∈ M(E) with the following
properties:
For every a ∈ X and n ∈ N, k[fn (b), U ]k# < 2−n , and
kU ∗ H(a)U − σ bH(a)U ∗ k# < ε.
bH(a)k# + kH(a) − U σ

Note that the existence of approximately commutative approximate units im-

plies that, for every countable set S of homeomorphisms σ of R+ , such a sequence
fn exists (uniformly for all ε > 0 and σ ∈ S).
If, conversely, H is approximately unitarily equivalent to a “constant” H0 , then
the same argument shows the existence of the desired sequence (fn ) and the unitary
U ∈ M(E). I.e., the sequence (fn ) exists, if Conjecture 12.3.2 is true.
If we make the additional assumption of the above described “uniform” ex-
istence of the sequence fn , then the proof of the Conjecture 12.3.2 is similar to
the proofs of Proposition 9.1.2 and its Corollary 9.1.3 together. But the operator
norms, that we have considered there, must be replaced by the above strictly con-
tinuous norms k . k# . This norm is invariant under scaling automorphisms σ b, but
it is only quasi-invariant with respect to inner automorphisms.

Remark 12.3.4. Let A, B, E and H : A → M(E) as in Conjecture 12.3.2.

Suppose that there exists a non-degenerate C *-morphism H0 from A into M(B),
such that H0 is unitarily homotopic to H in M(E).
Let Ψ2 denote the lower semi-continuous actions of Prim(B) on A defined by
Ψ2 (J) := H0−1 (H0 (A) ∩ M(B, J)).

Then Ψ1 (J) = Ψ2 (J ∩ B) for J ∈ I(E) and H0 is weakly Ψ2 -residually nuclear.

Indeed:
Since H0 and H are unitarily homotopic in M(E) , H0 (a) is in M(B) ∩ M(E, J)
 3. INTERJECTION: NON-COMMUTATIVE SELECTION CONJECTURE 1031

if and only if H(a) is in M(E, J). But it is easy to see that M(B, J ∩ B) =
M(B) ∩ M(E, J). Thus Ψ1 (J) = Ψ2 (J ∩ B) for J ∈ I(E).
Let I ∈ I(B). Then J := Q(R+ , I) ∩ E is a closed ideal of E, such that
J ∩ B = I. B/I is relatively weakly injective in E/J, because E/J is naturally
isomorphic to the hereditary C *-subalgebra of Q(R+ , B/I), that is generated by
B/I.
It follows, that I = B ∩ ΨB,E
down (I), and that B/I is relatively weakly injective
in E/ΨB,E
down (I). Therefore, B is residually relatively weakly injective in E.
Thus H0 is Ψ2 -residually nuclear by Lemma 12.2.12, where we have to replace
(C, A, B) by (H0 (A), B, E).

Definition 12.3.5. A separable C *-algebra B has the (non-commutative)

selection property if, for every separable C *-algebra A and every lower semi-
continuous action

Ψ : I(B) → I(A)

of Prim(B) on A in the sense of Definition 1.2.6 with Ψ(0) = 0 and Ψ−1 (A) = {B},
there exists a non-degenerate weakly Ψ-residually nuclear *-monomorphism H0
from A into the multiplier algebra M(B) of B such that δ∞ H0 is unitarily equivalent
to H0 , and, for J ∈ I(B),

Ψ(J) = H0−1 (H0 (A) ∩ M(B, J)) .

(By Corollary 5.9.16, H0 is unique up to unitary homotopy.)

Lemma 12.1.2(ii) allows to give an equivalent definition which also works for
non-stable B.
In other words (in the terminology of Section ??):
For every separable A and for every non-degenerate lower semi-continuous action
Ψ of Prim(B) on A, CPrn (Ψ; A, B) realizes the action Ψ in the sense that Ψ = ΨC
for C := CPrn (Ψ; A, B).
The “selection” procedure is the following:
Let be given a0 ∈ A+ and a pure state λ0 on B with λ0 (Ψ(span(Aa0 A)) 6= {0}.
Then we have to construct a continuous map f from the Polish space P (B) of pure
states on B into the compact metric space S(A∗ ) of positive linear functionals ρ on
A with kρk ≤ 1 with properties:
(i) f (λ)(Ψ(Jλ )) = 0 for all λ ∈ S(A∗ ), where Jλ is the kernel of the irreducible
representation dλ corresponding to λ.
(ii) For each a ∈ A+ there is b ∈ B with λ(b) = f (λ)(a) for all λ ∈ S(A∗ ).
The condition (ii) is not covered by usual selection theorems.

Conjecture 12.3.6 (Non-commutative Michael selection principle). Every

separable and stable C*-algebra B has the selection property of Definition 12.3.5.
1032 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

Now there exists a proof of the NC Micheal selection principle.

??
It uses that coherent Dini spaces are primitive ideal spaces of nuclear separable
C *-algebras and that every lower s.c. action of an Abelian separable C *-algebra C
on a separable and stable C *-algebra B comes from a H0 : C → M(B).

Remark 12.3.7.
(1) It is known that every exact separable stable C *-algebra B has the selection
property, because B ⊗ O2 contains a “regular” commutative C *-subalgebra C ⊂
B ⊗ O2 if B is exact and separable (cf. Remark 12.3.8).
(2) Since B has the selection property, if and only if, B ⊗ O2 has the selection
property, we may suppose that B ∼ = B ⊗ O2 and let D ⊂ B a non-degenerate copy
of C ⊗ K that is also regular in B. Then the non-commutative Michael selection
principle, the observation that M(B, B ∩ J) is equal to the set of T ∈ M(B) with
T D ⊂ J, Lemma 12.1.2 and Corollary 5.9.15 imply together a proof of Conjecture
12.3.2 for all stable separable B with the property that B ⊗ O2 contains an Abelian
regular C *-subalgebra.
(3) If the Conjecture 12.3.2 is true then also the non-commutative Michael
selection principle of Conjecture 12.3.6 holds:
Let B1 := B ⊗O∞ . By Theorem 12.1.8 and Conjecture 12.3.2, there is a weakly
Ψ-residually nuclear C *-morphism H1 : A → M(B1 ) such that, for J ∈ I(B),

Ψ(J) = H1−1 (H1 (A) ∩ M(B1 , J ⊗ O∞ )).

Let J ∈ I(B) and a ∈ A \ Ψ(J). Then there is f ∈ B1 , such that f ∗ H1 (a)f is not
in J ⊗ O∞ .
Let d : B1 → M(B) denote the non-degenerate *-monomorphism of Lemma
12.2.18.
By Lemma 12.2.18(ii), d(f ∗ H1 (a)f ) is not in M(B, J), i.e., there is e ∈ B, such
that V (a) is not in J, where we let V (c) := e∗ d(f ∗ H1 (c)f )e for c ∈ A.
V is Ψ-residually nuclear for f ∈ B1 and e ∈ B, because f ∗ H1 (.)f is Ψ-
residually nuclear, and e∗ δ∞ (d(.))e is equivariant on I(B) ∼
= I(B1 ).
By Lemma 12.1.2, there exists the desired Ψ-residually nuclear non-degenerate
*-monomorphism H0 from A to M(B).
Thus a proof of Conjecture 12.3.2 would imply a proof of the non-commutative
Michael selection principle.

Remark 12.3.8. If one limits to those separable C *-algebras B such that

Prim(B) is homeomorphic to Prim(C) for a separable exact C *-algebra C, then,
by Corollary 12.3.1, Theorem K, Corollary L and Lemma 12.2.18, it suffices to
prove the selection property for the unique stable separable nuclear C *-algebra A
with Prim(A) homeomorphic to Prim(B), and with A ∼ = A ⊗ O2 . Thus, in this class
 3. INTERJECTION: NON-COMMUTATIVE SELECTION CONJECTURE 1033

of C *-algebras B, the selection property is essentially a property of the T0 -space

Prim(B).
Lemma 12.1.3 implies, that every separable stable C *-algebra B with Hausdorff
primitive ideal space has the selection property.
Lemma 12.1.4 and the below given Lemma 12.3.10 show that stable separable
B has the selection property if B contains a commutative C*-subalgebra C which is
regular in B in the sense of Definition B.4.1.
This implies, e.g., that B has the selection property if its primitive ideal space
is isomorphic to the primitive ideal space of an inductive limit of C *-algebras with
Hausdorff primitive ideal spaces. But the second example considered after Corollary
N in Chapter 1, is not such an inductive limit but contains a commutative C *-
subalgebra which is regular in the algebra.
Examples of those B where considered by Mortensen [559]. We generalize his
construction as follows:
Let S be a compact metric semi-group with jointly continuous multiplication,
and let C := C(S). The multiplication defines a unital C *-morphism d : C → C ⊗C,
which is given by d(f )(x, y) = f (xy). Let A := C ⊗ O2 . We choose a unital *-
monomorphism h from C into O2 , and a *-isomorphism k from O2 ⊗ O2 onto O2 .
Let g denote the unital *-endomorphism of A, which is given by

g := (idC ⊗k)((idC ⊗h)d) ⊗ idO2 ).

Now let B := indlim(hn : A → A), with hn = h for n = 1, 2, . . ..

Then I(B) is the projective limit of the iterates of the map from O(S) into
O(S), which is given, for open subsets Z of S, by

Z 7→ S \ ((S \ Z)S).

If S is a linearly ordered space, which is separable and compact in its interval

topology, then S is topologically and order isomorphic to a closed subset of the
interval [0, 1] (cf. [360]). S carries semigroup structures, e.g. the multiplication
(x, y) 7→ min(x, y). The corresponding map on O(S) is given by

Z 7→ S ∩ (max(S \ Z), 1],

and the T0 -space of the corresponding is S with S, ∅ and S ∩ (t, 1], t ∈ [0, 1) as
open subsets. (Together with Corollary L, this implies the main result of [559].)

Remark 12.3.9. The realization H0 : A → M(E) of a lower semi-continuous

actions ΨB : I(A) → I(B) of Prim(A) on B (as given Theorem 12.1.8 with E :=
B Q(R+ , B)B) can be used to obtain realizations of upper semi-continuous actions
ΨA : I(B) → I(A) of Prim(B) on A, if the action ΨA is ‘residually non-singular’
in the following sense:
We call an action ΨA : I(B) → I(A) of Prim(B) on A residually non-
singular, if ΨA (J1 ) = ΨA (J2 ) and J1 ⊂ J2 imply J1 = J2 .
1034 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

If ΨA satisfies the condition (i) or the condition (iii) of Definition 1.2.6, then
ΨA is residually non-singular if and only if ΨA is an injective map from I(B) into
I(A).
Let B and A stable and separable, and let X := Prim(B). Since in the following
considerations only the lattices of ideals are considered, we assume — for simplicity
— that B ⊗ F ∼ = B and A ⊗ F ∼ = A, where F ∼ = O∞ ⊗ O∞ ⊗ . . ..
Suppose that ΨA : I(B) ∼ = O(X) → I(A) satisfies the conditions (i) and (ii) of
Definition 1.2.6, and ΨA (0) = 0, Ψ−1
A (A) = {B}.

We define, for I ∈ I(A), a subset of I(B) by

γ(I) := {J ∈ I(B) : ΨA (J) ⊂ I} .

The set γ(I) is closed under sums J1 + J2 of closed ideals, and J ∈ γ(I) if J is
the closure of an increasing sequence in γ(I). Therefore γ(I) contains a unique
maximal element ΨB (I). ΨB (I) is the closure of the sum of all elements in γ(I).
It is easy to see that ΨB : I(A) → I(B) is a lower semi-continuous action of
Prim(A) on B, that Ψ−1 B (B) = {A}, and that, for J ∈ I(B), I ∈ I(A), J ⊂ ΨB (I)
if and only if ΨA (J) ⊂ I. In particular, J ⊂ ΨB (ΨA (J)). It follows that, for
J ∈ I(B),
\
ΨA (J) = {I ∈ I(A) : J ⊂ ΨB (I)}.
If ΨA is residually non-singular, it implies that ΨB (0) = 0, and that ΨB is a left
inverse of ΨA .
Let E denote the closure A Q(R+ , A)A, and let H0 : B → M(E) be the ΨB -
residually nuclear non-degenerated *-monomorphism of Theorem 12.1.8 for ΨB ,
then an easy calculation shows that, for J ∈ I(B),
H (B),E
0
ΨA (J) = A ∩ Ψdown (H0 (J)) .

If, moreover, A has the selection property, then even H0 can be found such
that H0 (B) ⊂ M(A), and we get, for J ∈ I(B),
H (B),A
0
ΨA (J) = A ∩ Ψdown (H0 (J)) .

Lemma 12.3.10. Suppose that A is a nuclear stable C*-subalgebra of a separable

stable C*-algebra B, and that A has the selection property of Definition 12.3.5.
Then the following are equivalent:

(i) A is a regular subalgebra of B in the sense of Definition B.4.1.

(ii) Ψup
B,A : J ∈ I(B) 7→ A ∩ J ∈ I(A) is a residually non-singular action,
which is continuous in the sense of Definition 1.2.5, (i)-(iv).
(iii) A separates the ideals of B, and, for every I ∈ I(A), the set γ(I) := {J ∈
I(B) : A ∩ J ⊂ I} contains a unique maximal element.
(iv) A separates the ideals of B, and there exists a sequence (Tn ) of residually
equivariant maps from B into B, such that Tn (B) ⊂ A for n ∈ N, and
Tn |A converges in point-norm topology to id |A.
 3. INTERJECTION: NON-COMMUTATIVE SELECTION CONJECTURE 1035

Proof. If J1 , J2 ∈ I(B), I := (A ∩ J1 ) + (A ∩ J2 ), and γ(I) of (iii) contains

a maximal element K, then J1 ⊂ K and J2 ⊂ K. Therefore J1 + J2 ⊂ K, which
means I = A ∩ (J1 + J2 ).
This observation shows that (iii) and (ii) are obvious equivalent formulations
of (i).
(iv)⇒(i): Since A separates I(B), it remains to show that A ∩ (J1 + J2 ) is
contained in (A ∩ J1 ) + (A ∩ J2 ), for J1 , J2 ∈ I(B). The argument in the second
part of the proof of Lemma 12.1.4 can be easily adapted, to get this.
(iii)⇒(iv): ΨB (I) := max γ(I) is the lower semi-continuous action of Prim(A)
on B, which corresponds to the upper semi-continuous action ΨA := Ψup B,A in the
sense of Remark 12.3.9.
ΨB (0) = 0 and Ψ−1
B (B) = {A}, because A separates the ideals of B.

Since A is separable, stable, and has the selection property, there is a non-
degenerate weakly ΨB -residually nuclear *-monomorphism H0 from B into M(A),
such that, for I ∈ I(A)
H0 (ΨB (I)) = H0 (B) ∩ M(A, I).

It follows A ∩ ΨB (I) ⊂ I, and

H0−1 (H0 (A) ∩ M(A, I)) = A ∩ ΨB (I) .
Thus, by Corollary 3.10.6(II), H0 |A approximately dominates V := idA , because A
is also nuclear.
The hereditary C *-subalgebra D ⊂ B generated by A is stable and generates
B as an ideal, because A is stable and separates I(B).
If we consider H0 |D as a *-monomorphism from D into M(D), then H0 |D is
weakly residually nuclear, because A is nuclear, H0 (D) ⊂ M(A), and, for J ∈ I(B),
J = ΨB (A ∩ J) and
H0 (D) ∩ M(D, D ∩ J) = H0 (ΨB (A ∩ J)) .

Thus T := a∗ H0 (b∗ (.)b)a is a residually nuclear map from B into B with image
in A, if a ∈ A and d ∈ D.
The convex combinations of restrictions T |A approximate id |A in the point-
norm topology, because, the map H0 |A approximately dominates V := idA . 

The following Proposition 12.3.11 generalizes part (ii) of Theorem A.

Proposition 12.3.11. Suppose that A, B are stable and separable, that A

is nuclear and has the selection property, B is strongly purely infinite, and that
Ψ : I(B) → I(A) is a residually non-singular lower semi-continuous action of
Prim(B) on A with Ψ(0) = 0 and Ψ(B) = A.
Then the following are equivalent.

(i) Ψ is a continuous action in the sense of Definition 1.2.6, (i)-(iv).

1036 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

(ii) There is a non-degenerate *-monomorphism h0 : A ,→ B with h0 (Ψ(J)) =

h0 (A) ∩ J and [h0 ] = [h0 ] + [h0 ] and a residually equivariant conditional
expectation P from B onto h0 (A).
(iii) Every non-degenerate *-monomorphism h from A into B, with h(Ψ(J)) =
h(A) ∩ J and [h] = [h] + [h], maps A onto a regular subalgebra of B.

Proof. Combine Theorem K, Lemma 12.3.10, and the last part of Theorem
6.3.1. 

The next observations allows to extend the main results of this section consid-
erably (see Corollary 12.3.18).
Alternative version from Dini3 :
Suppose that A is a stable separable C *-algebra, and that g : Prim(A) → [0, ∞)
is a bounded lower semi-continuous function,such that the set X := {sup g(F ) ; F ∈
F(Prim(A))} is closed in [0, sup g(X)].
Then there exists a ∈ M(A)+ with the following properties (i)-(iii).

(i) Spec(a) = X,
(ii) g(J) = ka + M(A, J)k for all J ∈ Prim(A), and
(iII) Spec(a + M(A, I)) = X ∩ [0, sup g(hull(I))] for every closed ideal I / A.

If a ∈ M(A)+ satifies (i)-(iii) then its infinite repeat δ∞ (a) also satisfies (i)-
(iii). If b ∈ M(A)+ satisfies (i)-(iii) (with b in place of a), then there is
norm-continuous map t ∈ [0, ∞) 7→ U (t) ∈ U(M(A)) such that U (0) = 1 and
limt→∞ U (t)δ∞ (a)U (t)∗ = δ∞ (b).

Proposition 12.3.12. Suppose that B is stable and separable, 1 ∈ K ⊂ [0, 1]

a closed set and that f : Prim(B) → K is a lower semi-continuous function on
Prim(B) with supremum sup f (Prim(B)) = 1. Then there is a ∈ M(B)+ with
following properties:

(i) Spec(a) = K
(ii) ka + M(B, J)k = f (J) for all J ∈ Prim(B),
(iii) a is unitarily equivalent to δ∞ (a).
(iv) The action of Prim(B) on K = Spec(a) ∼ = Prim(C ∗ (a, 1)) is given by
Ψ(I) = (ka + M(B, I)k, 1] ∩ K for all closed ideals I of B.

The a ∈ M(B)+ with (i)–(iv) is unique up to unitary homotopy.

Proof. We find isometries r, t, s1 , s2 , . . . ∈ M(B) with tt∗ + rr∗ = 1 and

with n sn (sn )∗ strictly converges to 1. We define ∆ : `∞ (M(B)) → M(B) by
P

sn bn (sn )∗ . If we identify `∞ (M(B)) with M(c0 ⊗ B) then we

P
∆(b1 , b2 , · · · ) :=
can see that ∆ is a unital non-degenerate strictly continuous *-monomorphism.
If J is closed ideal of B, then J is stable, and tBt∗ +sJs∗ is stable and generates
a full hereditary C *–subalgebra D of B. Then D is stable, and, by Corollary 5.5.6
(which is the Ψ-equivariant version of the Brown stable isomorphism theorem),
 3. INTERJECTION: NON-COMMUTATIVE SELECTION CONJECTURE 1037

there exists an approximately inner isomorphism ψ from B onto D. Thus, there

is an projection p ∈ M(B) such that pBp and (1 − p)B(1 − p) are stable, pBp
generates J and (1 − p)B(1 − p) is full in B . It follows that there is an isometry
v ∈ M(B) with vv ∗ = 1 − p, and that the strictly closed ideal of M(B) generated
by p is just M(B, J).
We define a lower semi-continuous increasing function γ : K → K by γ(0) := 0
and γ(s) := sup(K ∩ [0, s)) for s > 0. The function γ satisfies t ≤ γ(s) ≤ s for all
t < s ∈ K, and γ(s) < s holds if and only if (γ(s), s) ∩ K = ∅.
It holds inf{γ(s) ; s ∈ (t, 1] ∩ K} = t for all t ∈ K. Indeed, if t = inf K ∩ (t, 1],
then there is a sequence s1 > s2 > · · · in K with lim sn = t, then sn+1 ≤ γ(sn ) ≤ sn ,
thus lim γ(sn ) = t. For each s ∈ (t, 1] ∩ K there is n ∈ N with s > sn . Hence
γ(s) ≥ γ(sn ) ≥ t and inf{γ(s) ; s ∈ (t, 1] ∩ K} = t . If t < s0 := inf K ∩ (t, 1], then
s0 ∈ K (by compactness of K) and (t, s0 ) ∩ K = ∅. It follows (t, 1] ∩ K = [s0 , 1] ∩ K,
[0, s0 ) ∩ K = [0, t] ∩ K and inf{γ(s) ; s ∈ (t, 1] ∩ K} = g(s0 ) = t.
Now let (x1 , x2 , . . .) a dense sequence of (pairwise different) points in K \ {0}
that contains all boundary points of K (in (0, 1]).
Let Jn denote the ideal of B corresponding to the open subset f −1 (γ(xn ), ∞)
of Prim(B). For each n ∈ N, we find a projection pn ∈ M(B, Jn ) such that pn Bpn
is stable and generates Jn .
Define a := ∆(x1 p1 , x2 p2 , . . .). We show that a ∈ M(B) satisfies (i) and (ii),
and we modify a later to get also (iii) and (iv).
(i): Let p0 := 1 − ∆(p1 , p2 , . . .), and define a unital *-morphism λ : C(K) →
M(B), for g ∈ C(K), by

λ(g) := g(0)p0 + ∆(g(x1 )p1 , g(x2 )p2 , . . .) .

Then λ is faithful, because {0, x1 , x2 , . . .} is dense in K. Clearly a = λ(f0 ) for

f0 (t) := t, t ∈ K. Thus Spec(a) = Spec(f0 ) = K by spectral permanence.
(ii): Notice that (sn )∗ Bsn = B, (sn )∗ Jsn = J and pn B generates Jn . Thus
pn (sn )∗ Bsn ⊂ (sn )∗ Jsn is equivalent to Jn ⊂ J. If J is primitive, then the inclusion
Jn ⊂ J means that f (J) ≤ γ(xn ) (by definition of Jn ).
Let J ∈ Prim(B). Then ka + M(B, J)k ≤ t, if and only if, k(a − t)+ +
M(B, J)k = 0, if and only if, (a − t)+ B ⊂ J, if and only if, min(xn −
t, 0)pn (sn )∗ Bsn ⊂ (sn )∗ Jsn for all n ∈ N, if and only if, f (J) ≤ γ(xn ) for
all n with t < xn .
Thus, ka + M(B, J)k ≤ t, if and only if, f (J) ≤ t, because

inf{γ(xn ) ; t < xn } = inf{γ(s) ; s ∈ (t, ∞) ∩ K} = t .

It follows (ii).
(iii): Replace a by δ∞ (a) (if necessary) and use that δ∞ ◦ δ∞ is unitarily
equivalent to δ∞ by Lemma 5.1.2(i).
1038 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

(iv): Suppose that a0 ∈ M(B) satisfies (i)–(iii) (in place of a). By property (iii)
and by Lemma 5.1.2(ii), one can find a unital *-monomorphism ϕ : C(K) ⊗ O2 →
M(B) with ϕ(f0 ⊗1) unitarily equivalent to a0 with property (i)–(iii), where f0 (t) =
t for t ∈ K. Take a unital monomorphism ρ : C(K) → O2 (e.g. using a continuous
epimorphism of the Cantor set onto K), and let g ∈ C(K × K) ∼ = C(K) ⊗ C(K)
denote the function g(s, t) := min(s, t).
Then Spec(g) = g(K × K) = K, and, for every closed subset F of K, one
has g(F × K) = [0, max F ] ∩ K. It follows that b := (id ⊗ρ)(g) ∈ C(K) → O2
has spectrum Spec(b) = F and that a natural action of Prim(C(K) ⊗ O2 ) ∼ = K
∗ ∼
on Prim(C (b, 1)) = K is given by F 7→ K ∩ [0, max F ] (all expressed by the
maps for the closed subsets F ). In particular, if I is a closed ideal of C(K) ⊗ O2
(corresponding to the open subset U of K), then I ∩ (C(K) ⊗ 1) = C0 (U ) ⊗ 1
and I ∩ C ∗ (b, 1) = µ(C0 (K ∩ (max(K \ U ), 1])), for the unital *-monomorphism
µ : h ∈ C(K) → h(b) = (id ⊗ρ)(h◦g) ∈ C(K) → O2 . It implies kb+Ik = kf0 ⊗1+Ik
and I ∩ C ∗ (b, 1) = µ(C0 (K ∩ (kb + Ik, 1]), for all closed ideals of C(K) → O2 .
Let a := δ∞ (ϕ(b)). Then a satisfies (i)–(iv), because the l.s.c. action of Prim(B)
on C ∗ (a, 1) (respectively on C ∗ (a0 )) factorizes over the action on of Prim(C(K) →
O2 ) ∼
= K on C(K) ⊗ 1 (respectively on C(K) ⊗ 1). 

Lemma 12.3.13. Suppose that B is a σ-unital C*–algebra, and that J, K ∈

I(B).

(i) (J + K)+ = J+ + K+ , and

(ii) M(B, J + K) = M(B, J) + M(B, K) .

Proof. (i): See [401, exercise 4.6.64]. Alternatively: It is easy to see that the
set J + K is a closed ideal. Let c ∈ (J + K)+ , and d := c1/2 . There are selfadjoint
a ∈ J and b0 ∈ K with a + b0 = d. Then c ≤ a2 + b2 for b := (b20 + (ab0 + b0 a)− )1/2
and b ∈ K+ . By asymmetric Riesz decomposition [616, prop. 1.4.10], there are
e, f ∈ B with ee∗ ≤ a2 , f f ∗ ≤ b2 and c = e∗ e + f ∗ f .
(ii): Since J, K and J + K are closed ideals of B, the M(B, J), M(B, K),
M(B, J) + M(B, K) and M(B, J + K) are closed ideals of M(B). The inclusion

M(B, J) + M(B, K) ⊂ M(B, J + K)

comes immediately from the Definition of M(B, ·). Let T ∈ M(B, J + K)+ and
ε > 0. Since B is σ-unital, there is an approximate unit e0 = 0 ≤ e1 ≤ e2 ≤ · · · ≤ 1
P P
in B with en en+1 = en for n ∈ N, such that n (en+1 − en ) and n≥0 (en+1 −
en )1/2 T (en+1 − en )1/2 are strictly convergent with sums 1 and T1 and kT1 − T k < ε.
Let e−1 := 0 and fn := en+2 − en−1 . By part (i), there are an ∈ J+ and bn ∈ K+
1/2
an (en+1 − en )1/2 and
P
with an + bn = fn T fn . Then S1 := n≥0 (en+1 − en )
S2 := n≥0 (en+1 −en )1/2 bn (en+1 −en )1/2 are strictly convergent, S1 ∈ M(B, J)+ ,
P

S2 ∈ M(B, K)+ , T1 = S1 +S2 . Thus, T has distance < ε from M(B, J)+M(B, K)
(for each ε > 0). 
 3. INTERJECTION: NON-COMMUTATIVE SELECTION CONJECTURE 1039

Remark 12.3.14. The (algebraic) definition of M(B, J) immediately shows

T T
that γ M(B, Jγ ) = M(B, γ Jγ ) for for every family {Jγ } ⊂ I(B), i.e., that
M(B, ·) is lower semi-continuous.
The monotone map M(B, ·) is not monotone upper semi-continuous, e.g. if
B := c0 = C0 (N) and Jn := C({1, 2, . . . , n}) (respectively B := C0 (N, O2 ⊗ K) and
S S
Jn := C({1, 2, . . . , n}, O2 ⊗ K) ), then B = n Jn , but n M(B, Jn ) is not dense in
M(B) = M(B, B).

Next need Def.’s of Dini function and Dini space

Lemma 12.3.15. Suppose that B is separable and stable, that X is a Dini space,
and that Ψ : I(B) ∼
= O(Prim(B)) → O(X) is a lower semi-continuous action.
Let J ∈ I(M(B)) and let Ψ∗ (J) := the biggest open subset U of X such that,
for every Dini function g : X → [0, ∞) on X with support in U and sup g(X) = 1,
af ∈ J for the element af ∈ M(B)+ of Proposition 12.3.12 for the l.s.c. function
f (J) := sup g(X \ Ψ(J)) for every Dini function g : X → [0, ∞) on X with support
in U and sup g(X) = 1.
It holds

(o) Ψ∗ : I(M(B)) → O(X) is well-defined.

(i) U ⊂ Ψ∗ (J), if and only if, M(B, Φ(V )) ⊂ J for all V << U and the
upper s.c. adjoint Φ of Ψ.
(ii) The map J → Ψ∗ (J) is l.s.c. and monotone upper s.c. on the ideals J
with J ∩ δ∞ (M(B)) generates J.

Proof. Important, to be filled in ??, 

Proposition 12.3.16. Suppose that A is separable and contains a regular exact

C*-subalgebra C. Then A has the Abelian factorization property.

Theorem 12.3.17. Suppose that A and B are separable and stable, and that A
has the Abelian factorization property.
Then, for each non-degenerate lower semi-continuous action Ψ : I(B) → I(A),
there is a non-degenerate faithful weakly Ψ-residually nuclear C*-morphism
H : A → M(B) (with H unitarily equivalent to δ∞ ) such that

Ψ(J) = H −1 (H(A) ∩ M(B, J)) for all J ∈ I(B) .

Proof. to be filled in ??, 

Corollary 12.3.18. Suppose that A and B are separable C*-algebras, A is

exact and B is stable. If Ψ : I(B) → I(A) is a lower semi-continuous action
of Prim(B) on A with Ψ({0}) = {0} and Ψ−1 (A) = {B}, then there is a non-
degenerate (norm-) nuclear C*-morphism H : A → M(B) with δ∞ ◦ H unitarily
equivalent to H and Ψ(J) = H −1 (H(A) ∩ M(B, J)) for all J ∈ I(B).
The nuclear morphism H is unique up to unitary homotopy.
1040 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

Proof. to be filled in ?? 

4. Proof of (old !!!) Theorem O

We conclude this chapter with the proof of the old !!!! Theorem O.
beginning from here, old text has to be replaced ??
The proof is fairly elementary and allows the reader to check his insight in the
above developed methods for Ψ-residually nuclear maps.
We show that any C *-algebra with primitive ideal space homeomorphic to [0, 1]
and fibres =∼ O2 respectively ∼
= O2 ⊗ K are isomorphic to C([0, 1], O2 ) respectively
to C([0, 1], O2 ⊗ K).
??
Next there is only a proof of the old Theorem O

Suppose that A is a unital separable C *-algebra with Prim(A) ∼ = [0, 1], the
interval [0, 1] with ordinary topology, and that the fibers, i.e. the simple quotients,
are isomorphic to O2 . Thus, A is as a C *-algebra bundle with base space [0, 1] and
fibers isomorphic to O2 . This implies that A nuclear and is strongly p.i., cf. ends
of Chapters 2 and 3. The K-theory of A is trivial, see below.
For nuclear C *-algebra bundles A, B over a locally compact σ-compact
metrizable spaces X, we have that KKnuc (X; A, B) = R KKtrivial (X; A, B).
By Corollary N, the bundle is stably isomorphic to C([0, 1], O2 ) if and only if
R KKtrivial ([0, 1], A, A) = 0.
This allows to use the tautological epimorphism and the six-term exact sequence
to check
the ‘‘following proposition’’, that has been erased,
which reduces the triviality question to the ordinary KK-theory.
We show the non-trivial direction of the statement of the old version of Theorem
O, i.e., that A is isomorphic to C([0, 1], O2 ), if A is a separable unital C*-algebra
with Prim(A) ∼ = [0, 1], if every primitive quotient of A is isomorphic to O2 , and if
A satisfies the UCT for the ordinary KK-theory of C*-algebras.
All non-zero projections in C([0, 1], O2 ⊗K) are equivalent. Therefore, it suffices
to show that A ⊗ K and C([0, 1], O2 ⊗ K) are isomorphic.
Every primitive quotient of A is nuclear. Since A is separable, this implies that
every factor representation of A generates an injective factor. Thus, A is a nuclear
C *-algebra.
By Corollary L, A ⊗ O2 ∼= C([0, 1], O2 ) with an isomorphism that implements
the isomorphism χ from [0, 1] onto Prim(A).
The isomorphism and the natural embedding of A ⊗ 1 into A ⊗ O2 maps χ(t)
onto the intersection of A ⊗ 1 with the kernel of the epimorphism g 7→ g(t) from
 4. PROOF OF (OLD !!!) THEOREM O 1041

C([0, 1], O2 ) onto O2 , and it maps the center of A onto C([0, 1]). This gives a unital
isomorphism λ from C([0, 1]) onto the center of A such that

χ(t) = λ(C0 ([0, 1] \ {t}))A and πt (λ(f )a) = f (t)πt (a).

Thus A is a (continuous) C *-bundle with quotient maps πt : A → A/χ(t) in

the sense of [92], [471]. Since [0, 1] has dimension one, and since the fibers A/χ(t)
are simple and purely infinite, by Remarks 2.15.12 and 3.11.6(ii), the algebra A is
strongly purely infinite, cf. [93]. (The latter can be seen also directly and elementary
with a modification of the idea in step one, below.)
It follows that Theorem M applies to A:
A ⊗ K is isomorphic to C([0, 1], O2 ⊗ K), if and only if, KK([0, 1]; A, A) = 0.
We show the equation KK([0, 1]; A, A) = 0 in the following five steps.
Let A|X denote the restriction of the bundle A to X: Let Ψ denote the action
of [0, 1] on A, which is defined by a homeomorphism χ from [0, 1] onto Prim(A).
Then Ψ(Z) = λ(C0 (Z))A for open subsets Z of [0, 1] and the primitive ideal χ(t)
is the ideal Ψ([0, 1] \ {t}). If X is a relatively closed subset of an open subset Z
of [0, 1], we define A|X as Ψ(Z)/Ψ(Z \ X). Then πt : A → A|{t} is the natural
epimorphism on the fiber at t. Let 0 < x < y ≤ 1, then the natural epimor-
phisms A|[0, y] → A|[0, x] and A|[0, y] → A|[x, y] define just the pull back of the
natural epimorphisms A|[0, x] → A|{x} and A|[x, y] → A|{x}, i.e., we can compose
elements of A|[0, x] and A|[x, y] to an element of A|[0, y] if they coincide at x.
First, A contains a copy of O2 unitally:
Since A|{x0 } is isomorphic to O2 there exist contractions a and b in A such
that a∗ b = 0 and that the continuous functions x 7→ kπx (c)k are zero at x = x0 for
c in G := {1 − a∗ a, 1 − b∗ b, 1 − aa∗ − bb∗ }.
It follows that there exists δ > 0 such that kπY (c)k < 1/8 for c ∈ G, where
Y := [max(0, x0 − δ), min(x0 + δ, 1)] and πY denotes the natural epimorphism from
A onto A|Y . Then s := πY (a)πY (a∗ a)−1/2 and t := πY (b)πY (b∗ b)−1/2 are canonical
generators of a unital copy of O2 in A|Y .
We find x0 = 0 < x1 < . . . < xn = 1 such that A|[xj , xj+1 ] contains isometries
sj and tj with sj s∗j + tj t∗j = 1 for j = 0, . . . , n − 1, because [0, 1] is compact.
We modify them by induction, fit them together and get a unital copy of O2
in A:
Suppose that k < n and that s, t are canonical generators of O2 in A|[0, xk ]. Let
π+ and π− denote the natural epimorphisms A|[0, xk ] → A|{xk } and A|[xk , xk+1 ] →
A|{xk } respectively.
Then u := π+ (s)π− (s∗k ) + π+ (t)π− (t∗k ) is a unitary in A|{xk } ∼
= O2 . Since the
unitary group of O2 is connected, we find a unitary v in A|[xk , xk+1 ] such that
π− (v) = u. Let s̃ := vsk and t̃ := vtk . Then π− (s̃) = π+ (s) and π− (t̃) = π+ (t).
1042 12. NON-COMMUTATIVE SELECTION AND PROOF OF THM. K

Thus there are S, T in A|[0, xk+1 ] with restrictions to [0, xk ] and [xk , xk+1 ] equal
to s, t and s̃, t̃ respectively. Thus S ∗ S = 1 = T ∗ T and SS ∗ + T T ∗ = 1.
Second, K0 (A) = 0: Since A contains copy of O2 unitally, it suffices to show
that, for every non-zero projection p in A, pAp contains a copy of O2 unitally.
t ∈ [0, 1] 7→ kπt (p)k is continuous, non-zero and takes 1 or 0 as its values. Thus
this function is one on [0, 1], and, therefore, Prim(pAp) = Prim(A) = [0, 1], and
the primitive quotients are isomorphic to O2 . The first step shows that the unital
algebra pAp contains a copy of O2 unitally.
Third, K1 (A) = 0: Let B := A ⊗ P∞ , where P∞ is the unique unital pi-sun
algebra in the UCT-class with K0 (P∞ ) = 0 and K1 (P∞ ) = Z, cf. Corollary H(ii).
Then K1 (A) = K0 (B) by Künneth theorem. But Prim(B) = [0, 1] and every simple
quotient of B is isomorphic to O2 ∼
= O2 ⊗ P∞ . Thus from the second step we see
that K0 (B) = 0.
Fourth, separable A with K∗ (A) = 0 satisfies the UCT if and only if
KK(A, B) = 0 for all separable C *-algebras B.
The six term exact sequence implies that KK(A|I, A|I) = 0 for every half-open
interval I = [0, t) or I = (t, 0], because there is a natural semi-split short exact
sequence
0 → A|[0, t) ⊕ A|(t, 1] → A → O2 → 0,

and KK is additive in each variable with respect to direct sums. It follows that
KK(A|I, A|I) = 0 for every closed subinterval I = [s, t] ⊂ [0, 1] of [0, 1], because
A|I = A/(A|[0, s) + A|(t, 1]). For open or half-open subintervals I of [0, 1] we have
short exact sequences 0 → A|I → A|K → B → 0 where K is the closure of I and
B is O2 or O2 ⊕ O2 .
Fifth: To calculate KK([0, 1]; A, A), we use the six-term exact sequence with
respect to the second variable and the following C([0, 1])-equivariant semi-split short
exact sequence:
0 → J → C([0, 1]) ⊗ A → A → 0,

where the action on C([0, 1]) ⊗ A is induced by

d : f ∈ C([0, 1]) 7→ f ⊗ 1 ∈ C([0, 1]) ⊗ A

and the map C([0, 1]) ⊗ A → A is the tautological epimorphism given by

f ⊗ a ∈ C([0, 1]) ⊗ A 7→ f a ∈ A .

The Ψ-equivariant semi-splitness of the exact sequence can be seen as follows:

Tensor with O2 , use that A ⊗ O2 ∼= C([0, 1], O2 ), take the obvious lift and cut it
down to A and C([0, 1], A) with help of a state on O2 .
J is the closed sum of ideals C0 (K) ⊗ A|I with certain open subintervals K and
I of [0, 1], cf. Proposition B.4.2(ii).
We consider now the action of [0, 1] on C0 (K) ⊗ A|I which is induced by d.
 4. PROOF OF (OLD !!!) THEOREM O 1043

For open or closed sub-intervals I and K of [0, 1],

KK([0, 1]; C0 (K) ⊗ A|I, C0 (K) ⊗ A|I) = 0,
because A|I is KK-equivalent to O2 and we can tensor the identity map of C0 (K)
with this KK-equivalence to obtain a KK([0, 1]; ., .)-equivalence of C0 (K) ⊗ A|I and
C0 (K) ⊗ O2 .
The Mayer-Vietoris sequence gives that also KK([0, 1]; B, C) = 0 if B ⊂ J and
B is finite sum of ideals C0 (K) ⊗ A|I of C([0, 1]) ⊗ A.
KK([0, 1]; ., .) is in the first variable continuous with respect to Ψ-equivariant
monomorphic inductive limits of nuclear C *-algebras Jn which are KK([0, 1]; ., .)-
equivalent to C([0, 1], O2 ). It follows KK([0, 1]; J, ·) = 0, because the corresponding
Milnor lim(1) -sequence vanishes.
Since also KK([0, 1]; C([0, 1])⊗A, C([0, 1])⊗A) = 0, the six term exact sequence
shows that KK([0, 1]; A, A) = 0.
 APPENDIX A

Cuntz equivalence, multiplicative domains,

?related topics?

?? TO DO list of topics, to add:

0 ≤ b ≤ n diag(b) for b ∈ Mn (A)+ etc.

( <-- this is partly done below )
Applications of n[a] ≤ m[b] for a, b ∈ A+ .

Remarks on Glimm halving in ultra-powers. !!!!!!! That is:

Aω has Glimm halving property,
if and and only if,
A has uniform Glimm halving property.
Is Glimm halving and uniform Glimm halving really different?
See also Remarks in Chapter 2 on Condition (ii) of Definition ??.
See also Lemma A.6.8(iii).
Conjecture?:
Let A a separable stable C *-algebra.
For separable C *-subalgebras B of Qs (A) ∼
= M(A)/A
and unital separable C *-subalgebras C of
F (A) := (A0 ∩ Aω )/ Ann(A, Aω )
there exists a unital C *-morphism
of the ‘‘winding around’’ (or ‘‘joining’’) algebra E(C, C) into B 0 ∩
Qs (A).
OK: It is only a very weak hope ... !!!

Generalities:
Flip of self-absorbing approximately inner ?
U(A) = U0 (A) for all A := On , A := En ?
U(A) = U0 (A) for all tensorial self-absorbing A?
Jiang-Su?
Each tensorial self-absorbing A absorbs Jiang-Su? Result of Winter?
Are K1 -injective?
Has stable rank one, by result of Rørdam

What happens with the commutators of unitaries in C(S 1 ) ⊗ E(Mm , Mn )

1045
1046 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

if m 6= n ?
Is it K1 -injective?
In particular, is each commutator of two unitaries in U(E(Mm , Mn )),
or -- more generally -- those contained in the connected component
U0 (E(Mm , Mn )) of 1 in the unitaries of it?

1. (*) The algebras On , En and O∞

We outline an approach to get the used basic facts on En and On , n ∈

{2, 3, . . . , ∞}. All is based on elementary observations (or can be replaced by such
observation in our particular cases). We work with formal algebras generated by
relations, because it allows to see what carries over to general non-commutative
rings... and what does not so ... E.g. the existence of suitable very important
C *-subalgebras in the central sequence algebra requires the analytic theory of
some rather special types of operator algebras, defined by their defining relations.
(1) C *-algebras defined by relations .
Let A denote an algebraic *-algebra over C (or likewise over R) that has
finitely, countable or uncountably many generators a1 , a2 , . . . and defining re-
lations R1 , R2 , . . . that are given explicitly by non-commutative polynomials
Pn (a1 , . . . , an , a∗1 , . . . , a∗n ) with coefficients in C (or R in case of real C *-algebras)
and the relation are given by requiring the equations Pn = 0, A is the quotient
of the free *-algebra by the ideal generated by the polynomials Pn . Sometimes
the relations are written more efficient as matrix conditions and then the Pn are
matrix polynomials. But this can be described also by ordinary n.c. polynomials
by writing the conditions for the entries separately.
If we require bounds, e.g. that certain expressions P (a1 , . . . , a∗n ) should
become a contractions in any representation then we can do this by introduc-
ing new elements, say x, y, and then Q1 (a1 , . . . , a∗n , x) := P (a1 , . . . , a∗n ) − x
and Q2 (x, x∗ , y, y ∗ ) = 1 − x∗ x − y ∗ y, but consider C ∗ (a1 , . . . , an ; R) with sys-
tem of relations R containing P as a C *-subalgebra of the unital C *-algebra
C ∗ (a1 , . . . , an , x, y, 1; R ∪ {Q1 , Q2 }). We go here not into details, because we use
only very special types of relations that are easily to understand and are usually
“bounded” in the following sense:
The algebraic *-algebra A is called “ bounded” if, for each generator an , there
A := A + C1, 0 ≤ γn < ∞ such that a∗n an + c∗1 c1 + · · · + c∗m cm =
exist c1 , · · · , cm ∈ f
γn2 1 . If A is bounded then every *-representation d of A on a pre-Hilbert space H
is bounded in the sense that kd(an )k ≤ γn for all generating elements. (Compare
[426, lem. 5.3(iii)] that was inspired by ideas of T. Tannaka [768] and its abstract
version introduced for the study of block-algebras of M.G. Krein [493]). (Today:
dual of compact “quantum groups”?)
In particular, the supremum kak := sup{d} kd(a)k of all C *-semi-norms on A
exists and is finite for all elements on A. The set J0 := {a ∈ A ; kak = 0 } is a
 1. (*) THE ALGEBRAS On , En AND O∞ 1047

*-ideal of A, and the C *-completion of A/J0 with respect to the induced (maximal)
C *-norm is a C *-algebra, that we denote by C ∗ (a1 , a2 , . . . ; R1 , R2 , . . .).
One can check that the below considered (!) special examples A all have the
property that J0 = {0} (exercises for the reader). – But in more general cases and
other types of relations it can happen that J0 6= {0}. –
To make this definitions more flexible, one can here the “scalar coefficients”
replace by elements in some other ring/algebra.
(2) On the defining relations for En and On
One can easily show – on a fairly algebraic level – that the universal C *-algebras
On := C ∗ (s1 , . . . , sn ; s∗j sj = 1 = s1 s∗1 + · · · + sn s∗n )
with n > 1 is “naturally” isomorphic to the universal unital C *-algebra
C ∗ (Bn ∪ {S} ; S ∗ S = 1 ∈ Bn , SaS ∗ = e1,1 ⊗ a, a ∈ Bn ) (1.1)
that contains Bn := Mn∞ .
Notice that this formula is only a formal “reminder” if we hold up our restricted
view that we, if possible, are willing to describe the basic model algebras as quotients
of “free” algebras by ideals generated by “relations” (as above described in an
informal manner), then Bn or a suitable dense *-subalgebra itself has to be described
as quotient of free *-algebra. It means here that we first have to define Bn as the
competition of the inductive limit of Ak,n := Mnk with a ∈ Ak,n 7→ Ak+1,n given by
a 7→ a ⊗ 1n and S is an “additional variable” that satisfies the relation polynomials
P1 (S) := S ∗ S − 1 = 0 with 1 = indlimk→∞ 1nk , and S(an ⊗ 1n )S ∗ = e11 ⊗ an ∈
Ak+1,n for all an ∈ Ak,n . It looks childish, but careful consistence would require a
detailed list of generators and how the generators of one algebra can be expressed
by the generators of the other algebra and vice versa. All this should be done on
the algebraic level. With other words: to compare this relation-defined algebras we
should her replace – for calculation with (countably many) relations on the pure
algebraic level – the algebra Bn by its algebraic inductive limit of the sequence
Mnk → Mnk+1 given by the maps a 7→ a ⊗ (e11 + · + enn ) = a ⊗ 1n and the proceed
as e.g. in [686, proof thm. 4.2.2] or in proof of [169, thm. 1.13]. The isomorphism
of On with the corner cross-product in Equation (1.1) becomes true for the C *-
algebras after completion with the maximal C *-norm on them. In more detail it
goes as follows:
Let Ik := {1, . . . , n}k and define
W (i1 , . . . , ik ) := Wp := si1 · . . . · sik
for the “word” p := (i1 , . . . , ik ) ∈ Ik . Then the linear span of the elements 1,
Wp , (Wp )∗ and Wp (Wq )∗ (p ∈ Ik , q ∈ I` ) is a dense *-subalgebra of On . Take
S := s1 and identify Mnk with the linear span of the elements Wp Wq∗ , p, q ∈ Ik
(k = 1, 2, . . .).
It shows that On ∼
= Bn oψ N for the corner endomorphism ψ : Bn 3 a 7→
e1,1 ⊗ a ∈ Bn . This algebra is stably isomorphic to the crossed product of a
1048 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

trace-scaling automorphism α of Bn ⊗ K. Trace-scaling automorphisms α of simple

C *-algebras have always the property that αk is strictly outer for k ∈ N, and
therefore, (Bn ⊗ K) oα Z is simple (use here the more general results of Kishimoto
[478] or of Olesen and Pedersen [578]). Crossed products of nuclear C *-algebras by
Abelian groups are always nuclear ????? give citation ?????????? (more generally
crossed products of nuclear C *-algebras by amenable groups are nuclear).
Missing citation/reference to place?
A proof would be:
Let G locally compact group, that acts point-norm continuous ( 1 ) on a C *-
algebra A by ρ : G → Aut(A), and let B any C *-algebra. There are natural iso-
morphisms of full crossed products

C ∗ (G, A; ρ) ⊗max B ∼
= C ∗ (G, A ⊗max B; ρ(·) ⊗max idB )

and of reduced crossed products

∗
Cred (G, A; ρ) ⊗min B ∼ ∗
= Cred (G, A ⊗min B; ρ(·) ⊗min idB )

and an obvious canonical epimorphism from the first to the second that is compat-
ible with the canonical epimorphism (·) ⊗max B → (·) ⊗min B.
Thus, if A nuclear and G amenable, then C ∗ (G, A, ρ) = Cred
∗
(G, A, ρ) is again
nuclear.
Here is what we need:
But in our case we use the following general observation:
Suppose that G is a compact group that acts point-norm continuous on a C *-
algebra A and that the fix-point algebra Aρ(G) of the action ρ of G on A is nuclear.
Then A nuclear.
Indeed: The kernel J of A ⊗max B → A ⊗min B is ρ(g) ⊗max id-invariant. And
the integral over G defines a faithful conditional expectation from J to the kernel of
Aρ(G) ⊗max B → Aρ(G) ⊗min B that is trivial by nuclearity of Aρ(G) . Thus, J = {0}
and A must be nuclear.
We have used here the following variant of the Definition of Nuclearity of C*-
algebras (compare cite :
Choi–Effros , [141], [142], Nuclear C*-algebras and injectivity: the general case.
(Shows: A is nuclear , if and only if, A∗∗ is an injective W*-algebra.) [145]
A C *-algebra A is nuclear, if and only if, the natural *-epimorphisms A ⊗max
B → A ⊗min B are faithful ( = are isomorphisms) for every C *-algebra B.
In fact, it suffices to take here as B the algebras C ∗ (F∞ ), L(`2 ) and
L(`2 )/K(`2 ). ( ¡ – is this from Kirchberg, E. [426], or from one of the upper
citations? )

1 Means that g ∈ G 7→ ρ(g)(a) is a continuous map from G into A for each a ∈ A

 1. (*) THE ALGEBRAS On , En AND O∞ 1049

It follows that On is simple and nuclear. By definition, On (2 ≤ n < ∞)

is finitely generated and O∞ is countably generated, in particular they all are
separable.
We have this ‘‘exercise in footnote’’ carried out somewhere!
Find and Cite the place of it!!
( 2 ).
The simplicity of the corner-endomorphism crossed product Bn oψ N and the
property that
for each a, b ∈ (Bn )+ with kbk = 1 ≥ kak and for each ε > 0 there are k ∈ N and a
contraction d ∈ Bn with kψ k (a) − d∗ bdk < ε
imply that Bn oψ N is purely infinite (cf. Corollary 2.18.3). Hence, On is a pi-sun
algebra in the UCT-class.
(Urgent Question:
Where it is shown that the On are in the UCT-class, and what has to be shown
to recognize the ”pi-sun” algebras are in the UCT-class? For the On one has
”only” to show first that all AF-algebras are in the UCT-class, and what imply the
products of UCT-class algebras by circle actions, action by an endomorphism or
automorphisms.)
The above mentioned property can be seen directly by the following argument
of J. Cuntz:
Let b ∈ Bn ⊆ On with b ≥ 0 and kbk = 1. Then we find k ∈ N and c ∈ (Mnk )+
with kb − c ⊗ 1k < ε and kck = 1. Since the k-fold tensor product ek of e1,1 ∈ Mn is
a minimal idempotent in Mnk there is a unitary u ∈ Mnk with ek u∗ cuek = ek . Let
d := (u⊗1)ψ k (a1/2 ), then kdk ≤ 1 and d∗ (c⊗1)d = ψ k (a). Thus kd∗ bd−ψ k (a)k < ε.
The Pimsner–Voiculescu exact sequence for crossed products (more precisely:
its variant for crossed products by N, [636]) gives “immediately” the exact sequence
1 ×(n−1) 1
0 → K1 (On ) → Z[ ] −→ Z[ ] → K0 (On ) → 0 ,
n n
?? to be checked again !!!
where 1 ∈ Z[ n1 ] maps to [1] ∈ K0 (On ). It shows that K1 (On ) = 0 and K0 (On ) ∼=
Zn−1 (in a way that the class of [1] generates Zn−1 ). In particular, O2 is K∗ -trivial.

(3) The non-simple algebras En .

We consider the (non-simple) universal C *-algebras

En := C ∗ (s1 , . . . , sn ; s∗i sj = δi,j 1) ,

and its (simple) natural inductive limit

O∞ := C ∗ (s1 , s2 , . . . ; s∗i sj = δi,j 1) .

2 In fact, it is easy to see (as an exercise for the readers?) that every separable C *-algebra A

with properly infinite multiplier algebra M(A), i.e., with 1M(A) ∈ E2 ⊆ M(A), is single generated.
This applies to all separable pi-sun algebras, in particular to En , On and O∞ .
1050 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

It turns out ( 3 ), that e := en := 1 − (s1 s∗1 + · · · + sn s∗n )

The element en is called ‘‘p’’ on some other places.
is a non-zero projection in En with eEn e = Ce, that the closed ideal I(e)
(generated by e) is naturally isomorphic to K, and that the left multiplication of
En on I(e) is faithful (as one can see with help of the gauge action z ∈ S 1 7→ σz ∈
Aut(En ) with σz (sn ) = zsn that fixes e).
Indeed:
X · En · e = {0} implies X ∗ X · En · e = {0}. Since σz (En · e) = En · e for all z ∈ S 1 ,
we get also σz (X ∗ X) · En · e = {0} for all z ∈ S 1 . Thus P (X ∗ X) · En · e = {0}.
R
Here P (a) := S 1 σz (a)dz for a ∈ En .
It is not difficult to see: The fix-point algebra of any (continuous) circle action
on the compact operators K of a separable Hilbert space has a non-degenerate fix-
point C *-subalgebra of K. (Because p ∈ P (p) · K · P (p) for all continuous circle
actions σ(z) on K.)
It follows that P (En ) acts faithful on the ideal I(e) by left-multiplication if and
only if P (En ) acts faithful on I(e) ∩ P (En ).
It is not difficult to see that the fix-point algebra P (En ) of the circle action on
En is an AF-algebra that contains p. If P (En ) acts not faithful on I(e) then there
must exist a non-zero projection in the kernel of this action.
Such a projection must be equivalent to a projection in the canonical filtration
of P (En ) by finite AF-algebras and each “smaller” projection is also on the kernel.
It causes that for some W := sk1 · . . . · skm holds W W ∗ · I(e) = {0}, in particular
W W ∗ (W e) = 0. But this is impossible, because W is an isometry, i.e., W ∗ W = 1
and W ∗ (W e) = e.
Thus, P (En ) acts faithful on En · e and on I(e). By above considerations it
implies that P (X ∗ X) = 0 if XEn · e = {0}. Since P is a faithful conditional
expectation (for any continuous circle action) it follows that X = 0 if X ∈ En and
X · En · e = {0}. Thus, I(e) is a non-degenerate ideal of En , i.e., there exists no
ideal orthogonal to I(e).
(This arguments show that every σ(z)-invariant non-zero ideal of En has non-
zero intersection with P (En ).)
Therefore, I(e) is an essential ideal of En .
Clearly, the quotient En /I(e) is naturally isomorphic to the – simple – C *-
algebra On .
(This way of arguments requires to prove before !!! the simplicity of On , to
obtain above conclusion.)

3Use that E is the span of elements W W ∗ of words, where W and W are = 1 or are
n r s r s
products of some of the generators s1 , . . . , sn .
 1. (*) THE ALGEBRAS On , En AND O∞ 1051

Above we have seen, that there is a short-exact sequence

0 → K → En → On → 0 .

In particular, En must be nuclear (if On is proven to be nuclear). The corre-

sponding six-term exact sequence in K∗ -theory shows that (necessarily) K1 (En ) = 0
and that there is an exact sequence

0 → Z → K0 (En ) → Zn−1 → 0 ,

such that the class [1] ∈ K0 (En ) is mapped onto the 1 ∈ Zn−1 and 1 ∈ Z is mapped
into [e] ∈ K0 (En ). The definition of e shows that [e] = (n − 1)[1] in K0 (En ).
The only remaining possibility is now K0 (En ) ∼
= Z with generator [1] of Z.
One can see now (e.g. from [e] = (n − 1)[1]) that the extension En of On by K
is not a split extension.
One can show also directly that the natural C *-morphisms En−1 → En → On
for n ≥ 2 define a faithful embedding from En−1 into On , – and therefore also a
faithful embedding from En−1 into En that does not intersect the closed ideal of En
Pn
generated by en := 1 − k=1 sk s∗k .
Indeed: Let ρ : En−1 → L(H) a faithful unital *-representation of En−1 . We
can here take a separable Hilbert space H, because En−1 is separable. The H has
infinite dimension because L(H) contains a non-unitary isometry. Thus H ∼ = `2 (N).
If we let L := ρ(en−1 )H, then there exists an isometry T ∈ L(H ⊗2 H) with
T T ∗ = ρ(en−1 ) ⊗ idH . The isometries tk := ρ(sk ) ⊗ idH for k = 1, . . . , n − 1 and
Pn
tn := T satisfy the relation k=1 tk t∗k = 1. Thus C ∗ (t1 , . . . , tn ) is a natural image of
On and the C *-representation a 7→ ρ(a)⊗idH is a faithful unital C *-morphism from
En−1 into C ∗ (t1 , . . . , tn ) that factorizes through the canonical unital C *-morphisms
En−1 → En → On → C ∗ (t1 , . . . , tn ). It shows that the natural unital C *-morphism
En−1 → En is injective and does not intersect the ideal of En generated by en .
It follows that the natural unital *-morphisms from En into O∞ are injective
(because they factorize over the natural C *-morphism En → En+1 , and that is
injective) and that, therefore, K∗ (O∞ ) = indlimn K∗ (En ) by continuity of K∗ (·)
with respect to inductive limits ...
(Unfortunately, it is bit more complicate ...!!!)
We get that O∞ is separable, unital and nuclear, that K0 (O∞ ) ∼ = Z with genera-
tor [1] ∈ K0 (O∞ ) and that K1 (O∞ ) = 0. The unital C *-morphisms ηn : En → En+1
satisfy the sufficient criteria in part (iv) of Proposition 2.2.5, because I(e) ∼
= K is
an essential ideal.
Indeed: It implies that, for each a ∈ (En )+ with kak = 1, there is a contraction
c ∈ I(e) with kc∗ ac − ek < ε, thus kd∗ ηn (a)d − 1k < ε for d := ηn (c)sn+1 .
Thus O∞ is simple and purely infinite by Proposition 2.2.5(iv), i.e., if we
all above put together the we see that O∞ is a (simple) pi-sun algebra with
1052 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

K0 (O∞ ) =∼ Z with generator [1] and K1 (O∞ ) = {0}. The pure infiniteness im-
plies that K1 (O∞ ) = 0 is equivalent to U0 (O∞ ) = U(O∞ ).
(4) δn is homotopic to id in End(On )
δ2 is unitarily homotopic to id in the space of unital *-endomorphism of O2
(with point-norm topology).
Indeed: Notice that X := {s2 , st, ts, t2 } are four isometries with range sum = 1
(X defines a unital embedding of O4 into O2 ). Thus B := span{ab∗ ; a, b ∈ X }
is naturally isomorphic to M4 . Let u1 := δ2 (s)s∗ + δ2 (t)t∗ = s2 (s∗ )2 + ts(st)∗ +
st(ts)∗ +t2 (t∗ )2 . Then u1 is a selfadjoint unitary in B, δ2 (s) = u1 s and δ2 (t) = u1 t .
It follows that there is a (norm-)continuous path [0, 1] 3 τ 7→ u(τ ) ∈ U(B) with
u(0) = 1 and u(1) = u1 . The unital endomorphisms hτ of O2 with hτ (s) = u(τ )s
and hτ (t) = u(τ )t define a point-norm continuous path in the unital endomorphisms
of O2 with h0 = id and h1 = δ2 .
Similarly, one can see that δn is homotopic to id in the unital *-endomorphisms
of On for 3 ≤ n < ∞ , by showing that
n
X n
X
u1 := δn (sk )s∗k = sk s` (s` sk )∗
k=1 k,`=1

is a selfadjoint unitary in the natural image of Mn2 (Z) spanned by

{sk s` s∗j s∗i ; k, `, j, i ∈ {1, . . . n} .

(5) The unitary groups are contractible U(On ) = U0 (On )

For the case of the real C *-algebras generated by s1 , . . . , sn , one can consider
Pn
the unitary u := k=1 δn (sk )s∗k in the bigger C *-subalgebra Mn3 of On spanned
by Wp (Wq )∗ (p, q ∈ {1, . . . , n}3 ).
The determinant of u ⊗ 1n in O(n3 ) is always ±1, but if n = 2m then again
u ∈ 1n ∈ SO(n3 ) and SO(k) is connected for all k ∈ N.
Thus, for even n ∈ 2N, we have again that δn is homotopic to id inside the
unital endomorphisms of the real version (On )R of On .
check next again? was used! Where?? The sign of the determinant of
u ⊗ 1 ∈ M22 ⊗ 12 ⊆ M23 is positive. Thus u ⊗ 1 ∈ SO(23 ), .... ????? But elements
of SO(23 ) are connected to 1 in SO(23 ). This implies also the K∗ -triviality of O2 R .
(Moreover F (O2 R ) = 0 for every homotopy invariant “additive” functor from the
category of “real” C *-algebras into “additive” categories of groups.)
For the other “real” On R the situation is more complicate ... Compare e.g. [718]
in conjunction with [173].
(6) The groups U(B ⊗ O2 ) are contractible.
U(B ⊗ O2 ) is connected for all unital separable C *-algebras: The application
of the path idB ⊗hτ from id to idB ⊗δ2 to U defines a path from U to δ2 (U ) in
U(B ⊗ O2 ).
Now proceed as in Chapter 11. more general? ??.
 1. (*) THE ALGEBRAS On , En AND O∞ 1053

Notice that every continuous map λ : S n → U(B ⊗ O2 ) is an element of

U(C(S n , B) ⊗ O2 ). Thus, U(B ⊗ O2 ) has trivial homotopy groups. Since U(B ⊗ O2 )
is a Polish space, that is locally homeomorphic to the self-adjoint part of B ⊗ O2 ,
it follows that U(B ⊗ O2 ) is contractible (By a classical result of J. Milnor [556]).
In particular, the unitary group of C(S 1 , O2 ) is simply connected.
(7) A proof that On is pi-sun
There is an alternative direct proof of the simplicity, nuclearity and pure in-
finiteness of On , considered as the universal C *-algebras

C ∗ (s1 , . . . , sn ; s∗j sj = 1 = s1 s∗1 + · · · + sn s∗n ) ,

that does not imply (immediately) the above derived additional informations that
On is in the UCT class and K∗ (On ) ∼
= Zn−1 ⊕ 0 with generator [1].
If z ∈ S 1 ⊂ C is a complex number with |z| = 1 then, by universality of On ,
there is a unique automorphism σz of On with σz (sj ) = zsj . It is easy to check on
elementary products of the sj and s∗k that z → σz is a (point-norm continuous) circle
action on On , and that the fixed point algebra of σ is just Bn , i.e., P (On ) = Bn
for the faithful conditional expectation
Z 2π
1
P (x) := σeit (x) dt .
2π 0
Let F any C *-algebra and h ≥ 0 in the kernel J of On ⊗max F → On ⊗min F .
Since the kernel is invariant under the action σz ⊗max idF by the functoriality
of the maximal and minimal C *-tensor products, we can integrate and get that
1
R 2π
2π 0 (σe ⊗ id)(h) dt is in J. The conditional expectation defined by this integral
it

can be calculated on elementary tensors, and is just P ⊗max idF . It is necessarily

a faithful conditional expectation from On ⊗max F onto Bn ⊗max F = Bn ⊗min F
(because Bn is nuclear). It follows that J ∩ (Bn ⊗max F ) = P ⊗max id(J) = 0 and,
thus P ⊗max idF (h) = 0 and h = 0, i.e., J = {0} and On is nuclear.
The proof of the simplicity and pure infiniteness uses that the faithful condi-
tional expectation P is one-step approximately inner (by a sequence of isometries
D1 , D2 , . . . ∈ On ). Refer to the place where the Dn are defined.
What is here Dk ? How is it defined? See further below.
An argument of J. Cuntz in [172] implies that On is both simple and purely
infinite:
Let ε > 0 and a ∈ On with 0 ≤ a and kak = 1, then P (a) 6= 0 (because P
is faithful on (On )+ by continuity of the circle action σz ), and we find k ∈ N
and a minimal projection p ∈ Mnk with kDk∗ aDk − P (a)k < (εkP (a)k)/2 and
kpP (a)p − kP (a)kpk < (εkP (a)k)/2 .
There is an isometry T in On with T (T ∗ ) = p . We get k d∗ ad − 1 k < ε for
d := kP (b)k−1/2 Dk T . Then calculation shows that kT ∗ Dk∗ aDk T − 1kP (a)kk <
(2/3)εkP (a)k .
(8) Approximate Innerness of P : On → Mn∞
1054 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

An algebraically defined explicit sequence of isometries D1 , D2 , . . . ∈ On , first

described by De Schreye and Van Daele in [209, prop. 6], with property

lim k Dk∗ xDk − P (x) k = 0 for all x ∈ On

k

can be given by the following modification of definitions by De Schreye and Van

Daele in [209, prop. 5] (it uses an argument of J. Cuntz in [169, lem. 1.8]):
We let Ik := { 1, . . . , n }k and define isometries Wp := si1 si2 · . . . · sik for p =
(i1 , . . . , ik ) ∈ Ik . The linear span of the elements Wr Ws∗ is a dense *-subalgebra of
On , and each element Wr Ws∗ (r ∈ I` , s ∈ Im ) with ` ≤ m ≤ k is in the linear span
of elements Wp Wq∗ with q ∈ Ik and p ∈ Ik−(m−`) . Notice that Mnk is naturally
isomorphic to the linear span of Wr Ws∗ with r, s ∈ Ik . Define Yk := s1 (s2 )k and
X
Dk := Wp Yk Wp∗ .
p∈Ik

The Wp are isometries with mutually orthogonal ranges that sum up to 1. Therefore
Dk is an isometry, and satisfies Dk∗ Wr Ws∗ Dk = P (Wr Ws∗ ) if r ∈ I` , s ∈ Im and
max(`, m) ≤ k .
Indeed:
It suffices to consider the case m = k ≥ `, then Ws∗ Dk = Yk Ws∗ , which implies that
Dk∗ bDk = b for all b ∈ Mnk . In case ` < k, one has Dk∗ Wr = ∗ ∗
P
i∈Ij Wr Wi Yk Wi
for j := k − `. Thus, the remaining case m = k > ` leads (with j := k − ` > 0) to
X
Dk∗ Wr Ws∗ Dk = Dk∗ Wr Yk Ws∗ = Wr (Wi Yk∗ Wi∗ Yk )Ws∗
i∈Ij

which is zero: If Wi Yk∗ 6= 0 then necessarily Wi = sj2 , which gives Wi Yk∗ Wi∗ Yk =
sj2 (sk2 )∗ s∗1 (sj2 )∗ s1 (s2 )k = 0 , because (sj2 )∗ s1 = 0 .
Change to more consistent notation!!
(9) Question if flip has algebraic approximation
An interesting question for application to flip maps is the following:
It holds that limk (Dk ⊗ Dk )∗ (x ⊗ y)(Dk ⊗ Dk ) = P (x) ⊗ P (y) for all x, y ∈ On
using suitable contractions D1 , D2 , . . . ∈ On with the property

lim k Dk∗ xDk − P (x) k = 0 for x ∈ On .

k

(It would be desirable to have them also for En in place of On to get it also for
O∞ .)
If this is the case then the question about the simplicity of the fixed point
algebra of the flip map on On ⊗ On becomes a question about the simplicity of the
fix-point algebra Fn∞ of the flip on Mn∞ ⊗ Mn∞ . This is because Fn∞ is the range
of the restriction of P ⊗ P to the fix-point algebra of the flip map on On ⊗ On :
Each (possibly existing) non-trivial ideal J of the fix-point algebra of the flip map
on On ⊗ On has a nontrivial intersection with the Fn∞ with help of the in the
flip-algebra contained Dk ⊗ Dk .
 1. (*) THE ALGEBRAS On , En AND O∞ 1055

??? Is there an opposite application ???: To show simplicity (!) of On , could

we show directly that the flip of On ⊗ On is approximately inner as c.p. map?
???
The unitary u = n s∗n ⊗ sn that satisfies u∗ u = 1 = uu∗ , u(sk ⊗ 1) = 1 ⊗ sk ,
P

and u∗ (1 ⊗ sk ) = sk ⊗ 1.
∗ ∗ ∗ ∗ ∗
P P P
n,m sn sm ⊗ sn sm = n 1 ⊗ sn sn = 1 and n,m sn sm ⊗ sn sm = 1 and
P
n ... = 1

(applies also to the ultrapower ?).

P ∗ P ∗
u(sk ⊗ 1) = (1 ⊗ sk )u ??? n sn sk ⊗ sn ?? =?? n sn ⊗ (sk sn ) No !!!!

Can not work??

because extremal ideals need not be maximal.
Could only work with extremal S1 -invariant ideals,
and would show that K ⊆ En
is the only non-trivial S1 -invariant ideal.
Attempt: Go the following way:
The kernel ideal of En → On is generated by p := 1 − s1 s∗1 − . . . − sn s∗n , and
the projection p satisfies pWi Wj∗ p = 0 and pWi = 0 for Wi ∈ Ik and Wj ∈ I` . It
follows that p is a minimal projection, i.e., satisfies p(En )p = C · p and the ideal
generated by p is isomorphic to K(`2 (N)).
(Question: In the real case one has to take ±1 actions (by Z2 or Zn2 ) ??)
Moreover, the circle action fixes the projection p. Thus, the ideal generated by
p is invariant under circle action.
next is nonsense, because ‘‘extremal’’ is not ‘‘maximal’’.
Check if this used on other places!!!
Since On and En are unital, they
???? have a maximal (? only extremal !! ?) nontrivial closed ideal (say J) that
contains the ideal generated by p.
Without knowing first the existence or non-existence of an ideal we have to
proof with other means that the conditional expectation P is approximately inner.
Just this is established with D1 , D2 , . . ..
The fixed points of the circle action on “this maximal ideal” must be necessarily
an ideal of the fixed point algebra.
(10) On simplicity of algebraic On version
The reader can check, that the above arguments show also that the universal
algebraic *-algebra (over R, C or over the rational numbers) with the above suitably
generalized defining relations of On is a simple *-algebra.
1056 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

In fact, the above arguments also show that the (algebraic) algebra generated by
P
elements s1 , . . . , sn , t1 , . . . , tn with defining relations tj sk = δjk 1 and k sk tk = 1
is simple.
(With suitably modified D1 , D2 , . . ..)
One can easily see that there is a unique anti-isomorphism x 7→ x∗ of order 2
with s∗k = tk . (The involution ∗ has to be defined as conjugate linear if we consider
the algebraic algebra over C defined by this relations.)
(11) List K∗ (On )
(K0 (On ), [1], K1 (On )) = (Zn−1 , 1, 0) (n = 3, . . .), (K0 (O2 ), K1 (O2 )) = (0, 0)
and (K0 (O∞ ), [1], K1 (O∞ )) = (Z, 1, 0).
Has been discussed above in Old(0.1) ?= New (3)? .
(12) Unital endomorphisms of On are homotopic.
Every unital endomorphism of On is homotopic to id (in particular δn is homo-
topic to id).
[And thus, (n − 1) K∗ (On ) = 0]
(See Old (0.2) ?= ? New (4) or (A1.4, it follows also from UCT and classification
for n = 3, 4, . . .).
But we need before that O∞ and O2 are pi-sun with (K0 (O2 ), K1 (O2 )) = (0, 0)
(It follows from: δ2 unitarily homotopic to id) and (K0 (O∞ ), [1], K1 (O∞ )) =
(Z, 1, 0) (should come from K∗ (En ) = (Z, [1En ] = 1, 0) and O∞ = indlimn En with
unital En → En+1 ).
En → E2 → En homotopic to id, and E2 → En → E2 homotopic to id, where
Em → En for n ≥ m natural sk 7→ sk , and O∞ → E2 given by s1 7→ s1 and
sn+1 7→ sn2 s1 . Notice s∗1 (sm ∗ n
2 ) s2 s1 = 0 for m = 0, 1, 2, . . ., m < n.

The proof of homotopy follows from the K1 -injectivity implied by the squeezing
property (sq) of the En .
We must show before that O2 ∼
= D2 := O2 ⊗ O2 ⊗ · · · and O∞ ∼
= D∞ :=
O∞ ⊗ O∞ ⊗ · · · ???.
(How to get that shortly, elementary, sufficient ???)
(13) General F (O2 ) = 0 for homotopy functors.
For every
(enough: “with respect to unitary homotopy” ??)
homotopy-invariant additive functor F : C ∗ → G into an Abelian (or only ad-
ditive) category G holds F (O2 ) = 0 where C is a subcategory of the category of
C *-algebras that contains O2 (as object) and all unital endomorphisms of O2 (as
morphisms).
Also every functor F that takes on approximately unitary equivalent C *-
morphisms into the same element/morphism in G.
 1. (*) THE ALGEBRAS On , En AND O∞ 1057

(14) On as Cuntz-Pimsner algebras.

There is another way to study On (and En ) – including all the former results on
nuclearity, simplicity, pure infiniteness, UCT and the calculation of the K-theories.
This algebras are very special cases of Cuntz-Pimsner algebras (respectively
generalized Fock–Toeplitz algebras) as considered by M. Pimsner in [633], because
On ∼
= O(H) and En ∼ = T (H) for the Hilbert C –C- bimodule H := `2 (n), O∞ ∼ =
O(H) = T (H) for the C –C-bimodule H := `2 (N). The paper of M. Pimsner unifies
and generalizes some parts of related work of J. Cuntz.

1.1. Remarks on On and O∞ (2). Where is the homotopy a ⊗ 1 ∼ 1 ⊗ a in

O∞ , O2 ?
(and On ?? depends on the class in K1 (On ⊗ On ) of ??? a possible flipping
unitaries vn , discussed somewhere below ?? with limn vn∗ (sk ⊗ 1)vn = 1 ⊗ sk ,
limn vn∗ (1 ⊗ sk )vn = sk ⊗ 1 ??? or V := n−1/2 k,` s∗k ⊗ s` ??? with V ∗ = V )
P

constructed respectively described?

For O2 ⊗ A we have always K∗ (O2 ⊗ A) = 0. Indeed, δ2 ⊗ idA is (unitary)
homotopic to id on O2 ⊗ A, because δ2 on O2 has this property.
If A is unital, then this implies U(O2 ⊗ A) = U0 (O2 ⊗ A) by Proposition 4.2.15
because O2 satisfies as a quotient of E2 the “squeezing” Property (sq) of Definition
4.2.14 and O2 ⊗ A inherits it from O2 .
It follows moreover that U(O2 ⊗ A) is contractible, because O2 ⊗ (A ⊗ C(Sn ))
has also contractible C(Sn , U(O2 ⊗ A)) ∼
= U(O2 ⊗ (A ⊗ C(Sn ))) .
The case of O∞ requires to show that K1 (O∞ ) = 0 and K1 (O∞ ⊗ O∞ ) = 0. It
was shown in Section ?? of Chapter 2 that O∞ is p.i. and simple.
< -- NO! That On is simple has to be shown -- before we can
apply this arguments !!
The definitions of O∞ , On and En show that all have an properly infinite unit
element, thus are K1 -surjective by Lemma 4.2.6(v).
By Proposition 4.2.15, En , O∞ and On have the squeezing property (sq) of
Definition 4.2.14, because the En have it and O∞ respectively On inherits it as
inductive limit of all En respectively quotients of En .
Those C *-algebras A that absorb O2 or O∞ tensorial are also strongly purely
infinite (and are simple if A is simple), and then are K1 -injective by Proposition
4.2.15 or ?????????????
Since O∞ is in the “bootstrap” category (as inductive limit of the En ), the
Künneth theorem for tensor products gives that K1 (O∞ ⊗ O∞ ) = 0 and that
K0 (O∞ ⊗ O∞ ) = Z with [1 ⊗ 1] = 1. This argument can be repeated and yields
K1 (O∞ ⊗ O∞ ⊗ · · · ) = 0 and K0 (O∞ ⊗ O∞ ⊗ · · · ) = Z with [1] = 1 by induction.
Use vn get what??
1058 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

Pn
It needs to consider the partial isometries vn := k=1 sk ⊗ s∗k in O∞ ⊗ O∞
Pn
(respectively the unitaries in On ⊗ On ). They satisfy vn∗ vn = 1 ⊕ ( k=1 sk s∗k ) and
Pn
vn vn∗ = ( k=1 sk s∗k ) ⊕ 1.
Are the isometries sk ⊗ s` generators of O∞ ⊗ O∞ ? (Certainly the sk ⊗ 1 and
1 ⊗ s` are generators.)

If one uses K1 -injectivity, then one has to use also Künneth theorem to calculate
K1 (O∞ ⊗ O∞ ).
Needs K1 -injectivity of all s.p.i. algebras:
Some has to be shown (more practical?) in Section ?? of Chapter 2.
The notion of K1 -injectivity is defined and studied in Section 2 of Chapter 4.
Gives that all unital *-morphisms O∞ → E are unitarily homotopic if and
only if E is K1 -injective:
Let S1 , S2 , . . . ∈ E generators of O∞ . sn := h1 (Sn ), tn := h2 (Sn ).
s1 , s2 , . . . homotopic to s1 , s1 sn2 and t1 , t1 tn2 , then to t1 , t2 , . . ..
Gives embeddings E2 ,→ E3 ,→ E.
All unital *-monomorphisms En ,→ En+k (n, k ≥ 1) are homotopic in En+k and
are – moreover – homotopic in E – if E is K1 -injective.
If E unital and has properly infinite unit, then this means U(E)/U0 (E) ∼
= K1 (E)
if E is K1 -bijective.
Uses U0 (En ) = U(En ) and that En ∩ K = {0} in En+1 , because K ⊆ Em is an
essential ideal of Em for each m ≥ 2 and otherwise the unital map En → On+1
would define a unital C *-morphism from On into On+1 .
It follows that D∞ has a flip automorphism. that is homotopic to id.

SOME COLLECTION OF RESULTS USED IN CHAPTERS 2 and

4 about On :
Transfer or Repeat ??
Needs to refer to “Property (sq)” !!!
Give the precise reference to definition and Prop. with results in Chp. 4 !!!
?? From Chp. 4:
(a): En := C ∗ (s1 , . . . , sn ; s∗i sj = δij ), n = 2, . . . , ∞, satisfy Property (sq):
Indeed: Let w := k1 k2 . . . kp , v := `1 `2 . . . `q , “words” of length p, q ∈ N with
“letters” kj , `i ∈ {1, . . . , n} from the “alphabet” {1, . . . , n}.
Define T0 := 1 and an isometry by the product

Tp (w) := sk1 sk2 · . . . · skp .

It is easy to see ??? that the linear span of the element 1 and of the “elementary”
products Tq (w), Tq (w)∗ and Tq (w)(Tr (v)∗ ) is dense in the C *-algebra En (in both
of real or complex case), because
 1. (*) THE ALGEBRAS On , En AND O∞ 1059

(i) the product Tq (v)∗ Tp (w) is equal to 1 if v = w,

(ii) is equal to zero if there exists g ≤ r := min(p, q) with kg 6= `g ,
(iii) is equal to T(p−q) (u1 ) if q < p and w = vu1 and
(iv) is equal to T(q−p) (u2 )∗ if p < q and v = wu2 .

We denote by G one of this elementary products or let G = 1.

Perhaps change L(...) to λ(...)???
A sort of “length” L(G) is defined by L(1) := 0, L(Tq ) := q, L((Tq )∗ ) := q and
L(Tq · (Tr )∗ ) := q + r. Clearly, s∗k Tp (w) = δk,k1 Tp−1 (v) for all k = 1, . . . , n with
word v := k2 k3 . . . kn for w = k1 k2 . . . kn .
Let G := Tp (w)Tq (v)∗ with w = k1 . . . kp , v = `1 . . . `q (p, q ≥ 1), then,
L(s∗k Gs` )
= L(G) − 2 if and only if k1 = k and `1 = `, otherwise only s∗k Gs` = 0
can happen.
In case G = Tp (w), w = k1 . . . kp we get s∗k Gs` 6= 0 if and only if k = k1
and then L(s∗k Gs` ) = L(G), because then G = Tp (u) with u = k2 . . . kp `. Since
L(G∗ ) = L(G) the same holds for G = Tp (w)∗ with the role of k and ` interchanged.
Pm
Therefore, one can find, for each linear combination C = α0 1 + j=1 αj Gj
(with scalars αj ), an index 1 ≤ j0 ≤ m and suitable k, ` ∈ {1, . . . , n} such that
k 6= `, s∗k Gj0 s` = 0 for at least one j0 ∈ {1, . . . , m} and L(s∗k Gj s` ) ≤ L(Gj ) for
j 6= j0 . Thus s∗k Cs` = j6=0,j6=j0 αj s∗k Gj s` with elementary products s∗k Gj s` (it
P

can be equal to 1).

Iteration – say m-times – of this operation leads to words w and v in the alpha-
bet {1, . . . , n} such that for the isometries Tm (w) and Tm (v) holds Tm (w)∗ Tm (v) =
0 and Tm (w)∗ CTm (v) = 0. It says that En has Property (sq).
It follows that its quotients On also have Property (sq).
END OF (sq) PROOF for En !
We can also use also the proof of J. Cuntz in [172, thm. 1.9, thm. 3.7] to get
U0 (O∞ ) = U(O∞ ).

On the real version (O2 )R of O2 :

The homotopy id ∼h δ2 exists also in the semi-group of unital *-endomorphisms
of (O2 )R . It follows that U((O2 )R ) = U0 ((O2 )R ) and that all unital *-
endomorphisms ϕ of (O2 )R are homotopic to the identity isomorphism of (O2 )R .
Indeed, ϕ is uniquely defined by ϕ(sj ) = uϕ sj (j = 1, 2), where
2
X
uϕ := ϕ(sj )s∗j ,
j=1

and, for each unitary u ∈ (O2 )R holds:

u ⊕s1 ,s2 u ∼h u in U((O2 )R ). Since u ⊕ v ∼h v ⊕ u and uv ⊕ 1 by Part (v,4) of
Lemma 4.2.6 we get that

(u ⊕ u) ⊕ u∗ ∼h u ⊕ u∗ ∼h 1 .
1060 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

Since Ud or Ud∗ first?

Ud∗ ((u ⊕ u) ⊕ u∗ )Ud = u ⊕ (u ⊕ u∗ )

it follows
u ⊕ 1 ∼h u ⊕ (u ⊕ u∗ ) ∼h 1
in U((O2 )R ). We get that

u ∼h u ⊕ u ∼h (u ⊕ 1) · (1 ⊕ u) ∼h 1

for every u ∈ U((O2 )R ), i.e., U((O2 )R ) = U0 ((O2 )R ).

The self-adjoint unitary C(2) := 1≤i,j≤2 si sj s∗i s∗j in the real C *-algebra
P

A2 := C ∗ (si sk (sj s` )∗ ; 1 ≤ i, j, k, ` ≤ 2) ∼
= M4 (R)
has determinate = −1. But δ2 (C(2)) is unitary equivalent to C(2) in

M8 (R) ∼
= A3 := C ∗ (si sk sm (sj s` sn )∗ ; 1 ≤ i, j, k, ` ≤ 2)
∗
P
by the unitary C(3) := 1≤i,j,k≤2 si sj sk (sj sk si ) . This implies that C(2) ∈
U0 ((O2 )R ) and then that δ2 is homotopic to id inside the unital *-endomorphisms
of (O2 )R .
The equality U(E2 ) = U0 (E2 ), can be seen (in case of complex E) from the short-
exact sequence K → E2 → O2 , and from the equalities U0 (K + C · 1) = U(K + C · 1)
and U0 (O2 ) = U(O2 ) .
Transfer or Repeat? FROM Chp. 2:
In proof of property (sq) for En in Chp. 4
is already some said about the algebraic version of En .
The C *-algebra En is the universal C *-algebra generated by n isometries with
orthogonal ranges, i.e., En := C ∗ (s1 , . . . , sn ; s∗k s` = δk` 1) . The C *-algebra O∞ is
the inductive limit of the sequence of the natural unital C *-morphisms hn : En →
En+1 defined by hn (sk ) := sk for k = 1, . . . , n.
The *-monomorphisms hn satisfy the criteria 2.2.5(iv):
The closed ideal J(pn ) of En generated by pn := 1 − s1 s∗1 − · · · − sn s∗n is
isomorphic to K(`2 (N)) and pn is a minimal projection of J(pn ) ∼ = K, because
∗
pn Tq (w1 )Tr (w2 ) pn = 0 for w1 , w2 words in the “alphabet” {1, . . . , n} with at least
one of w1 and w2 is not the “empty” word (i.e., the q + r > 0 for the lengths q of
w1 and r of w2 ).
It implies that pn · En · pn = C · pn .
The ideal J(pn ) is an essential ideal of En , because On ∼
= En /J(pn ) is simple.
(Refer here to the place where simplicity of On is proven!)
The simplicity of On follows from the fact that Mn∞ is the fixed point algebra
of the “gauge” circle action on On and that (!!!!) the conditional expectation from
On onto Mn∞ defined by this action is an approximately inner u.c.p. map.
(Where is the appox innerness is shown here?)
 1. (*) THE ALGEBRAS On , En AND O∞ 1061

(The point is, that one proves first – e.g. Cuntz – that the conditional expecta-
tion Pn : On → Mn∞ is element-wise approximately inner by taking approximation
be elementary .)
Ref’s to simplicity (! where shown?) of On ??:
J(pn ) ∼
=K
pn En pn = Cpn
On simple, implies that
J(pn ) is essential ideal of En .
(But the argument for “J(pn ) essential” needs only that On is G-simple for
G := S 1 and the “gauge” circle action.)
Hence, a ∈ En 7→ H(a) ∈ M(J(pn )) with H(a)b = ab for b ∈ J(pn ) is a faithful
C *-morphism, and for a ∈ (En )+ with kak = 1 there exists a partial isometry
z ∈ J(pn ) ∼
= K with z ∗ z = pn and k pn − z ∗ az k < ε .
Define c := hn (z)sn+1 hn (b1/2 ). If we use that hn (pn ) = pn+1 + sn+1 s∗n+1 ∈
En+1 and pn+1 sn+1 = 0, then we get

kc∗ hn (a)c − hn (b)k < ε .

New ?? concepts for Appendix of book:

Try to find a direct simple proof that the gauge action of On is approximately
inner ...???
NOT found yet !!!
Definitions of On , En , O∞ .
Its K∗ -groups:
K∗ (En ) = (Z, 0), K∗ (On ) = (Z/(n − 1)Z, 0), n = 2, 3, . . ., i.e., cyclic with generators
[1] K∗ (O∞ ) = (Z, 0).
O∞ is the inductive limit of the inclusion maps En → En+1 , n = 2, 3, . . . . It is
stationary on [1]. Shows (Z, 1) ∼
= (K0 (O∞ ), [1]).
It was shown in Chp.2, sec.2.2:
indlim En → En+1 is simple and p.i.
But it did use that the ideal I(p) is essential in En ,
with p = p0 = en defined below/above several times.
Requires that I(p) is essential for En .
On := En /I(p), where I(p) is the closed ideal of En generated by the projection
p := 1 − s1 s∗1 − . . . − sn s∗n .
Here En := C ∗ (s1 , . . . , sn ; s∗i sj = δij 1) = C ∗ (An ) is the C *-completion of
An , defined as the universal algebraic *-algebra generated by s1 , . . . , sn subject to
the relations s∗i sj = δi,j 1.
The existence and well-defined-ness of C ∗ (An ) follows from an old observation
of M.G. Krein in 1949 [493] mentioned in his study of Krein bloc-algebras:
1062 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

Let A a unital *-ring with relations that allow to prove that for (some)
given generators a1 , a2 , . . . ; a∗1 , a∗2 , . . . and relations R1 , R2 , . . . of A there exists
b1,n , . . . , bk,n ∈ A and γn ≥ 0 (depending from an ) such that in the unitization
Ae := A + C · 1 of A holds

a∗n an + b∗1,n b1,n + · · · + b∗k,n bk,n = γn 1 .

Then there is a well-defined maximal C *-seminorm on A. We denote the corre-

sponding C *-algebra then by C ∗ (a1 , a2 , . . . ; R1 , R2 , . . .). The natural C *-morphism
from the algebraic A into its C *-algebraic version C ∗ (a1 , a2 , . . . ; R1 , R2 , . . .) is not
necessarily faithful.
(In non-unital case one can adjoin a unit element e and require that ??? for
desired “contractions” w ∈ A there exist v ∈ Ae in an extended version Ae ⊇ A
with w∗ w + v 2 = 1 or w∗ w + v ∗ v = 1.)
We show that our An is the linear span of the expressions Wα Wβ∗ for α and β
words in from the “alphabet” {1, 2, . . . , n}, where Wα := 1 for the empty word (of
length 0) or Wα = sk1 sk2 · . . . · sk` for the word α = k1 k2 . . . k` of length ` = `(α)
with k1 , k2 , . . . , k` ∈ {1, . . . , n}.
Notice that Wα and Wβ satisfy Wα∗ Wβ = δα,β 1 if α and β have the same
length. If `(α) < `(β), then we have to write β as β = β 0 γ with `(β 0 ) = `(α) and
get Wα∗ Wβ = δα,β 0 · Wγ . In this way only terms like 1 and Wα · Wβ∗ appear in the
sub-semigroup of An generated by 1, s1 , . . . , sn , s∗1 , . . . , s∗n . Thus An is the linear
span of this terms.
If we let p0 := 1 − s1 s∗1 − · · · − sn s∗n then p0 Wα = 0 if α has length ≥ 1, i.e.,
if Wα 6= 1, because p0 sk = 0 for all k ∈ {1, . . . , n}. It follows that p0 An p0 = C · p0
and p0 En p0 = C · p0 .
It follows that W (α)p0 W (α)∗ W (β)p0 W (β)∗ = 0, i.e., that the ideal I(p0 ) of An
generated by {p0 } contains a countable number of mutually orthogonal projections
pα := W (α)p0 W (α)∗ with 1-dimensional pα An pβ .
Clearly p0 = p∗0 , and, for k 6= ` and i = j, or i 6= j,

(si )k p0 ((si )k )∗ (sj )` p0 ((sj )` )∗ = 0 .

It implies that the closed ideal I(p0 ) of En must be isomorphic to K .

There is a circle action σz on An (by universality of An ) given on generators by
σz (sk ) := zsk for z ∈ S 1 := {z ∈ C ; |z| = 1}. The action fixes p0 and is compatible
with the relations of An , i.e., the tk := σz (sk ) define the same relations t∗j tk = δjk 1
and p0 := 1−t∗1 t1 −· · ·−t∗n tn as the s1 , . . . , sn . Moreover σz (a∗ ) = σz (a)∗ , because
this happens on generators.
By universality of An it follows that σ extends to a circle action on En that
fixes p0 . In particular, it takes I(p0 ) ∼
= K invariant.
(In fact one has moreover a natural U(Mn ) action on En that fixes p0 .)
 1. (*) THE ALGEBRAS On , En AND O∞ 1063

Thus, σ extends naturally to a circle action σ on En , and then to an action

on On respectively on O∞ , because they are the completions with the maximal
possible C *-norms on it.
Notation below differs from that of above:
The fix-point-algebra of σ on On is generated by the elements Wα Wβ∗ with Wα
and Wβ words in “letters” s1 , . . . , sn of the same length `(α) = `(β). We have
Pn
sn s∗n = 1 in On . It follows that Wα Wβ∗ = j=1 (Wα sj )(Wβ sj )∗ is contained in
P

the span of the Wγ Wδ with 1 + `(α) = `(γ) = `(δ) = 1 + `(β).

It is easy to see that this is naturally isomorphic to the inductive limit of unital
embeddings
Mnk → Mnk ⊗ 1n ⊆ Mnk+1 .
Thus On σ ∼
= Mn∞ and Z
Pn (a) := σz (a)dz
z∈S 1
is a conditional expectation from On onto Mn∞ ⊆ On .
It turns out that Pn is 1-step approximately inner in On (cf. [209]), but this
has been shown also by J. Cuntz itself.
∗
The corresponding contractions Tm with Pn (a) := limm Tm aTm are given by
X X X
Tm := Vα Wβ (Vα )∗ (Wβ )∗ = δnm (Wβ )(Wβ )∗ = Vα δnm (Vβ ) ,
α,β β α
m
where α and β run through {1, . . . , n} .
If leads to the observation that On is simple, p.i. (hence s.p.i.) and nuclear:
The nuclearity follows from the general observation that compares maximal and
minimal tensor products and shows that a C *-algebra with a continuous action of
a compact group is nuclear if its fix-point-algebra is nuclear.
Where explained in detail? Refer to this.
Since Pn (On ) ∼= Mn∞ is simple and Pn is faithful on the positive cone of On , it
follows that “one-step approximate innerness” of Pn would show also the simplicity
of On , cf. [169], [172] or [209].
Nothing found in the cited papers?!
The argument of J. Cuntz goes a follows:
(1a.) The conditional expectation Pn : On → Mn∞ ⊆ On is faithful, because it
is the integral of a circle action.
(1a is My or his argument?)
(1b.) The nonzero projections in Mn∞ ⊆ On are all infinite in On .
In particular Why it should be a special case of what???, there exists for each
nonzero projection p ∈ On an isometry s ∈ On with s∗ ps = 1.
(1c.) It implies that each nonzero positive element a ∈ Mn∞ is infinite in On ,
and there exists d ∈ On with d∗ ad = (kak/2) · 1.
1064 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

(2a.) Let b ∈ On an element in W? ????? then there exists d ∈ W?? such that
d b bd = Pn (b∗ b) ...??
∗ ∗

(Range of such d must almost commute with given finite subset in the fixed
point algebra of the circle action but must have orthogonal ranges if composed with
a finite subset of non-fixed points ...?)
(2b.) Let a ∈ On positive with kak = 1 and ε > 0. Then there exists an
element e ∈ A with ke∗ ae − Pn (a)k ≤ ε.
WHERE THIS IS SHOWN?
(2a.)
If we can show that Pn (a) is contained in {d∗ ad ; d ∈ On } for each a ∈ On with
a ≥ 0.
Thus, if 0 6= a ≥ 0 and Pn is faithful on (On )+ then On is simple.
Let a ∈ On positive and ε > 0, then find n ∈ N and b∗ = b ∈ Wn with
kb − ak < ε. Leads to kPn (a) − Pn (b2 )k < ε.
2

Then find element (contraction?) D ∈ Wm with kD∗ b2 D − Pn (b2 )k < ε. ...

If one can show that each projection in Pn (On ) dominates the range of an
isometry in On then the one-step approximation of Pn shows moreover that On
must be purely infinite.
That is, each minimal projection of the span of {Wα Wβ∗ ; `(α) = `(β) = k} =
Mnk is the range of an isometry in On , because all non-zero minimal projections in
Mnk are equivalent.
Since I(p) ∼
= K (where shown? ref!) and On := En /I(p) is nuclear, it follows
that the algebras En are nuclear. This passes to its inductive limit O∞ , i.e., O∞ is
nuclear.
The defining element u ∈ O2 with usj = δ2 (sj ) of the endomorphism δ2 : O2 →
O2 given by a ∈ O2 7→ δ2 (a) := s1 as∗1 + s2 as∗2 can be written down explicitly as

u := s1 s1 s∗1 s∗1 + s2 s1 s∗2 s∗1 + s1 s2 s∗1 s∗2 + s2 s2 s∗2 s∗2 .

It is a symmetry in the real C *-algebra M4 (R) ∼

= A2 := CR∗ (si sk s∗` s∗j ; i, j, k, ` =
1, 2) with determinate = −1.
∼ A3 :=
Since CR∗ (si sk s∗` s∗j ; i, j, k, ` = 1, 2) is a unital *-subalgebra of M8 (R) =
∗ ∗ ∗ ∗
CR (si sk sm sn s` sj ; i, j = 1, 2), and O(4) ⊗ 12 ⊆ SO(8), there is a continuous path
inside SO(8) that connects u ⊗ 12 in SO(8) with 1. This happens then also in O2 .
An explicit path can be found with help of suitable elementary transformations:
Bit more general for n ∈ N, n > 1, we consider On and En . Let
W (k) := {1, . . . , n}k . If w = (j1 , . . . , jk ) ∈ W (k) is a “word” of length k,
let λw := (jk , j1 , . . . , jk−1 ) the cyclic change of the word w ∈ W (k) and
w := (jk , jk−1 , . . . , j2 , j1 ) the reversed word in W (k). They define isometries
∗
P
S(w) := sj1 · . . . · sjk . Notice that the elements Z(k) := w∈W (k) Z(λw)Z(w)
 1. (*) THE ALGEBRAS On , En AND O∞ 1065

∗
P
and U (k) := w∈W (k) Z(w)Z(w) of On are orthogonal operators in Ak :=
span({S(w1 )S(w2 )∗ ; w1 , w2 ∈ Wk }) ∼= Mnk respectively a symmetry (i.e.,
U (k)∗ = U (k), U (k)2 = 1).
Notice that Ak ⊆ Ak+1 because
n
X
∗
S(w1 )S(w2 ) = (S(w1 )s` )(S(w2 )s` )∗
`=1

and S(w)s` = S((w, `) where (w, `) := (j1 , . . . , jk , `) ∈ W (k + 1).

Z(k + 1)∗ S(w1 )S(w2 )∗ Z(k + 1) = δn (S(w1 )S(w2 )∗ ) for w1 , w2 ∈ W (k) .

This follows from Z(k + 1)∗ S(w1 )s` s∗` S(w2 )∗ Z(k + 1) = s` S(w1 )S(w2 )∗ s∗` or
s∗` S(w2 )∗ Z(k + 1) = S(w2 )∗ s∗` .
In case n = 2 and k = 2, one gets

U (2) = Z(2) = s21 (s21 )∗ + s22 (s22 )∗ + s1 s2 (s2 s1 )∗ + s2 s1 (s1 s2 )∗

Moreover U (2)s` = δ2 (s` ) for ` = 1, 2.

More generally, for n ≥ 2, U (2)s` = δn (s` ) for ` = 1, . . . , n . It follows that δn is
homotopic to id on On , if and only if, U (2) is homotopic to 1 in U(On ) (respectively
in O(On ) for the real version of On ).
In case n = 2 we get with V := (s1 s∗2 − s2 s∗1 ) that

Z(3)∗ U (2)Z(3) = δ2 (U (2)) = (s1 U (2)s∗1 + s2 s∗2 )V ∗ (s1 U (2)s∗1 + s2 s∗2 )V .

Notice that exp((π/2)V ) = V + (1 − V 4 ) if V ∗ = −V = V 3 . In particular,

V = exp((π/2)V ) if V ∗ = −V and V 4 = 1. Let C = [αjk ] ∈ SO2 the 90◦ rotation
matrix, i.e., α11 = α22 = 0 and α21 = −α12 = 1 and recall [s1 , s2 ]∗ = [s∗1 , s∗2 ]> ∈
M2,1 (O2 ) ⊆ M2 (O2 ). We define a unitary in (the real version of) O2 by

Vθ := exp(θ(π/2)V ) = [s1 , s2 ] exp(θπ/2C)[s1 , s2 ]∗ ∈ A3

Since V ∗ = −V , we get that Vθ is in SO(A3 ) ∼

= SO3 for θ ∈ [0, 1].
It follows that

Uθ := Z(3)(U (2) ⊕ 1)V (θ)∗ (U (2) ⊕ 1)V (θ)Z(3)∗

is a path in SO(A3 ) that connects the unitary U (2) = U1 ∈ O(A2 ) with 1 = U0 .

It follows that the unital *-endomorphism δ2 of O2 is homotopic inside the
unital endomorphisms End1 (O2 ) of O2 to the identity isomorphism.
Similar considerations show that δn ∈ End1 (On ) is homotopic inside the unital
endomorphisms End1 (On ) of On to the identity isomorphism id of On .

If one can find a unitary U in the ultra-power (O2 )ω of O2 , or in Q(R+ , O2 ) :=

Cb (R+ , O2 )/ C0 (R+ , O2 ), or in (O2 )∞ := `∞ (O2 )/c0 (O2 ) with U δ2 (x)U ∗ = x for
x ∈ O2 then one gets for tj := U sj that t1 , t2 commute element-wise with O2 , i.e.,
O2 ∼= C ∗ (t1 , t2 ) is a copy of O2 that is unital contained in O2 0 ∩ (O2 )ω (respectively
1066 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

in O2 0 ∩ Q(R+ , O2 ) or in O2 0 ∩ (O2 )∞ ). Then, by repeat of this construction, we

find a unital copy of O2 ⊗ O2 ⊗ · · · ⊆ O2ω that commutes with O2 element-wise in
this asymptotic algebras. It allows to conclude that O2 ∼ = D2 := O2 ⊗ O2 ⊗ · · · .

Next is contained in proof prop. 2.2.5?

The following argument for pure infiniteness of O∞ uses (and needs) that En →
O∞ is injective. (Somewhere above proven !)
and that e ∈ En satisfies sn+1 s∗n+1 ≤ e.
Claim: The algebra O∞ is purely infinite.
Idea of arguments:
“Have somewhere above shown” that the natural map η : En → En+1 is injective
Pn
and that en := 1 − k=1 sk s∗k satisfies en En en = C · en .
(E.g. via looking to En−1 → En → On and extending a faithful unital represen-
tation ρ : En−1 → L(H) with separable H to ρ ⊗ id : En−1 ⊗ L(H) → L(H ⊗ H) and
finding an isometry T ∈ L(H ⊗ H) with

T T ∗ = 1 ⊗ 1 − ρ(s1 s∗1 + . . . + sn−1 s∗n−1 ) ⊗ 1 .

– Here we denote idH by 1. )

It follows that I(e) embeds injective into En+1 and s∗n+1 η(e)sn+1 = 1. If c ∈ En
is positive and g ∈ I(e) with γ · e = g ∗ cg 6= 0, g ∗ g = e, then s∗n+1 g ∗ cgsn+1 = γ1.
In particular, c is full and properly infinite in En+1 .
Let a ∈ O∞ positive with kak = 1. Then there exists n ∈ N and a positive
contraction b ∈ En with ka − bk < 1/8. Thus, kbk ≥ 7/8 and k(b − 1/8)+ k ≥ 3/4.
There is a contraction d ∈ O∞ with d∗ ad = (b − 1/8)+ , cf. Lemma 2.1.9. It follows
that d∗ ad ∈ En and kd∗ adk ≥ 3/4.
Rest of the proof: e = 1 − k sk s∗k satisfies I(e) ∼
P
= K and e ∈ I(e) minimal and
e ∈ En+1 ⊆ O∞
Need for final conclusion that
(o) I(e) ∼
= K,
(i) e is a properly infinite full projection in En+1 ⊆ O∞ and
(i) I(e) essential for En .
To be shown:
The closed ideal I(e) ∼
= K of En is an essential of En .
Steps:
(1.) I(e) ∼
= K, because:
It is a closed ideal of En generated by the projection e := 1 − (s1 s∗1 + . . . + sn s∗n ).
It satisfies esj = 0 for j = 1, . . . , n. Thus, eEn e = C · e, i.e., the projection e is a
minimal idempotent in En .
(2.) I(e) is invariant under the canonical circle action, because e is fixed by the
circle action
(3.) The set of positive elements orthogonal to I(e) is (by (2.)) invariant under the
 1. (*) THE ALGEBRAS On , En AND O∞ 1067

circle action and is a closed ideal J of En orthogonal to I(e).

(4.) The image πI(e) (J) of J in On := En /I(e) is a closed ideal of On that is
isomorphic to J and is invariant under the circle action on On and ...
(5.) πI(e) (J) does not contain the class 1 + I(e) ∈ En /I(e):
Otherwise it means that (1 − Y ) · En · e = {0} for some Y ∈ I(e) with 0 ≤ Y ≤ 1.
It implies that Y a = a for all a ∈ I(e) and that Y − Y 2 = 0. Thus, Y is a unit
element of I(e), but I(e) is not unital.
(6.) The circle-action invariant ideal πI(e) (J) has – if it would be non-zero – a non-
zero intersection with P (On ) that is a closed ideal of P (On ), that can not contain
1 by (5.).
Since P (On ) ∼
= Mn∞ it follows that J ∼
= πI(e) (J) = 0.
This finishes the proof that I(e) is an essential ideal of En (without showing
before that On is simple).
Follows also immediately from the simplicity of On ( – if this has
been shown before together with the property of e := 1 − s1 s∗1 − · · · − sn s∗n that
e · En · e = Ce – but we give an other proof that does not use the simplicity of On ):
eW (α) = 0 for `(α) > 0, where W (α) :=????.
Recall: a ∈ O∞ with kak = 1 there exists n ∈ N and a positive contraction
b ∈ En with ka − bk < 1/8. Thus, kbk ≥ 7/8 and k(b − 1/8)+ k ≥ 3/4. There is a
contraction d ∈ O∞ with d∗ ad = (b − 1/8)+ , cf. Lemma 2.1.9.
We find a minimal projection q in I(p) with

kq(b − 1/8)+ qk ≥ k(b − 1/8)+ k − 1/4 ≥ 1/2 .

Then qd∗ adq = αq for some α ≥ 1/2. There is a contraction z ∈ En+1 with
z ∗ qz = 1, because there is a partial isometry v ∈ I(p) ∼= K with v ∗ qv = p and
∗ −1/2 ∗
sn+1 psn+1 = 1. Thus f := α dqz satisfies f af = 1.
It follows that O∞ is simple and purely infinite (if we have shown before the
simplicity of On or only that I(e) is an essential ideal of En ).

1.2. Remarks on Real On and O∞ (2). The defining unitary u ∈ O2 with

usj = δ2 (sj ) for the endomorphism

δ2 : O2 3 a 7→ δ2 (a) := s1 as∗1 + s2 as∗2 ∈ O2

can be written down explicitly as

u := s1 s1 s∗1 s∗1 + s2 s1 s∗2 s∗1 + s1 s2 s∗1 s∗2 + s2 s2 s∗2 s∗2 .

It is a symmetry with determinate = −1 in the real C *-algebra M4 (R) ∼

= A2 :=
∗ ∗ ∗
CR (si sk s` sj ; i, j, k, ` = 1, 2).
But since CR∗ (si sk s∗` s∗j ; i, j, k, ` = 1, 2) is a unital *-subalgebra of
Check formula for A3 :

M8 (R) ∼
= A3 := CR∗ (si sk sm s∗n s∗` s∗j ; i, j, k, `, m ∈ {1, 2})
1068 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

, and the natural embedding A2 ⊂ A3 , O(4) ⊗ 12 ⊆ SO(8), there is a continuous

path inside SO(8) that connects u ⊗ 12 in SO(8) with 1. This happens then also
in the real version of O2 .
An explicit path can be found using an isomorphism from M8 (R) onto A3 .
Bit more general (but only in complex case) for n ∈ N, n > 1, we con-
sider On and En . Let W (k) := {1, . . . , n}k . If w = (j1 , . . . , jk ) ∈ W (k)
is a “word” of length k, let λw := (jk , j1 . . . , jk−1 ) the cyclic change of
the word w ∈ W (k) and w := (jk , jk−1 , . . . , j2 , j1 ) the reversed word in
W (k). The define isometries S(w) := sj1 · . . . · sjk . Notice that the elements
∗ ∗
P P
Z(k) := w∈W (k) Z(λw)Z(w) and U (k) := w∈W (k) Z(w)Z(w) of On are
orthogonal operators in Ak := span({S(w1 )S(w2 ) ; w1 , w2 ∈ Wk }) ∼
∗
= M nk
∗ 2
respectively a symmetry (i.e., U (k) = U (k), U (k) = 1).
Notice that Ak ⊆ Ak+1 because
n
X
S(w1 )S(w2 )∗ = (S(w1 )s` )(S(w2 )s` )∗
`=1

and S(w)s` = S((w, `) where (w, `) := (j1 , . . . , jk , `) ∈ W (k + 1).

Z(k + 1)∗ S(w1 )S(w2 )∗ Z(k + 1) = δn (S(w1 )S(w2 )∗ ) for w1 , w2 ∈ W (k) .

This follows from Z(k + 1)∗ S(w1 )s` s∗` S(w2 )∗ Z(k + 1) = s` S(w1 )S(w2 )∗ s∗` or
s∗` S(w2 )∗ Z(k + 1) = S(w2 )∗ s∗` .
In case n = 2 and k = 2, one gets
U (2) = Z(2) = s21 (s21 )∗ + s22 (s22 )∗ + s1 s2 (s2 s1 )∗ + s2 s1 (s1 s2 )∗
Moreover U (2)s` = δ2 (s` ) for ` = 1, 2.
More generally, for n ≥ 2, U (2)s` = δn (s` ) for ` = 1, . . . , n. It follows that δn is
homotopic to id on On , if and only if, U (2) is homotopic to 1 in U(On ) (respectively
in O(On ) for the real version of On ).
In case n = 2 we get with V := (s1 s∗2 − s2 s∗1 )
Z(3)∗ U (2)Z(3) = δ2 (U (2)) = (s1 U (2)s∗1 + s2 s∗2 )V ∗ (s1 U (2)s∗1 + s2 s∗2 )V .

Notice that exp((π/2)V ) = V + (1 − V 4 ) if V ∗ = −V = V 3 . In particular,

V = exp((π/2)V ) if V ∗ = −V and V 4 = 1. Let C = [αjk ] ∈ SO2 the 90◦ rotation
matrix, i.e., α11 = α22 = 0 and α21 = −α12 = 1 and recall [s1 , s2 ]∗ = [s∗1 , s∗2 ]> ∈
M2,1 (O2 ) ⊆ M2 (O2 ) . We define a unitary in (the real version of) O2 by
Vθ := exp(θ(π/2)V ) = [s1 , s2 ] exp(θπ/2C)[s1 , s2 ]∗ ∈ A3
Since V ∗ = −V , we get that Vθ is in SO(A3 ) ∼
= SO3 for θ ∈ [0, 1].
It follows that
Uθ := Z(3)(U (2) ⊕ 1)V (θ)∗ (U (2) ⊕ 1)V (θ)Z(3)∗
is a path in SO(A3 ) that connects the unitary U (2) = U1 ∈ O(A2 ) with 1 = U0 .
 1. (*) THE ALGEBRAS On , En AND O∞ 1069

It follows that the unital *-endomorphism δ2 of O2 is homotopic inside the

endomorphisms of O2 to the identity isomorphism.

This is an important information, because it allows to show that with an ap-

proximation argument one can apply the following observation:
If one can find a unitary U in the ultra-power (O2 )ω of O2 , or in Q(R+ , O2 ) :=
Cb (R+ , O2 )/ C0 (R+ , O2 ), or in (O2 )∞ := `∞ (O2 )/c0 (O2 ) with U δ2 (x)U ∗ = x for
x ∈ O2 then one gets for tj := U sj that t1 , t2 commute element-wise with O2 in
the asymptotic algebra, i.e., O2 ∼ = C ∗ (t1 , t2 ) is unital contained in O2 0 ∩ (O2 )ω
(respectively in O2 0 ∩ Q(R+ , O2 ) or in O2 0 ∩ (O2 )∞ ). Then, by repeat of this
construction, we find a unital copy of O2 ⊗ O2 ⊗ · · · ⊆ O2ω that commutes with O2
element-wise in this asymptotic algebras. It allows to conclude that O2 ∼ = D2 :=
O2 ⊗ O2 ⊗ · · · . See Chapter 11.
Remarks on real O2 and O∞ (3):
The endomorphism δ2 (a) := s1 as∗1 + s2 as∗2 is homotopic to id in the endomor-
phisms of real Cuntz algebra (O2 )R .
It follows U(O2 ) = U0 (O2 ) by [172, thm. 1.9, thm. 3.7].
We have for the “real version” of O∞ also U0 (O∞ ) = U(O∞ ) by [172,
thm’s. 1.9, 3.7], (cf. arguments in Chapter 4, modify them for the real case).
U0 ((O2 )R ) = U((O2 )R ) and U0 ((O∞ )R ) = U((O∞ )R ) hold also for the “real”
version (O2 )R of O2 .
Here we use in the real cases E = ER , (O2 )R and (E2 )R also the “complex”
notations “unitary” and U(E), U0 (E) instead of O(E), respectively SO(E), for the
orthogonal elements u ∈ E, respectively u ∈ E in the connected component of 1 in
O(E).
The homotopy id ∼h δ2 exists also in the semi-group of unital *-endomorphisms
of (O2 )R . It follows that U((O2 )R ) = U0 ((O2 )R ) and that all unital *-
endomorphisms ϕ of (O2 )R are homotopic to the identity isomorphism of (O2 )R .
Indeed, ϕ is uniquely defined by ϕ(sj ) = uϕ sj (j = 1, 2) where
2
X
uϕ := ϕ(sj )s∗j ,
j=1

and, for each unitary u ∈ (O2 )R holds:

u ⊕s1 ,s2 u ∼h u in U((O2 )R ). Since u ⊕ v ∼h v ⊕ u ∼ uv ⊕ 1 by part (v,4) of
Lemma 4.2.6 we get that

(u ⊕ u) ⊕ u∗ ∼h u ⊕ u∗ ∼h 1 .

Since Ud or Ud∗ first?

Ud∗ ((u ⊕ u) ⊕ u∗ )Ud = u ⊕ (u ⊕ u∗ )

it follows
u ⊕ 1 ∼h u ⊕ (u ⊕ u∗ ) ∼h 1
1070 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

in U((O2 )R ). We get that

u ∼h u ⊕ u ∼h (u ⊕ 1) · (1 ⊕ u) ∼h 1

for every u ∈ U((O2 )R ), i.e., U((O2 )R ) = U0 ((O2 )R ).

The selfadjoint unitary C(2) := 1≤i,j≤2 si sj s∗i s∗j in the real C *-algebra A2 :=
P

C ∗ (si sk (sj s` )∗ ; 1 ≤ i, j, k, ` ≤ 2) ∼
= M4 (R) has determinate = −1.
But δ2 (C(2)) is unitary equivalent to C(2) in

M8 (R) = A3 := C ∗ (si sk sm (sj s` sn )∗ ; 1 ≤ i, j, k, ` ≤ 2)

∗
P
by the unitary C(3) := 1≤i,j,k≤2 si sj sk (sj sk si ) . This implies that C(2) ∈
U0 ((O2 )R ) and then that δ2 is homotopic to to id in the unital *-endomorphisms of
(O2 )R .
The equality U(E2 ) = U0 (E2 ) , can be seen (in case of complex E) from the
short-exact sequence

K → E2 → O2
and from the equalities
U0 (K + C · 1) = U(K + C · 1)
and U0 (O2 ) = U(O2 ).
But one can also use the general property (sq) to prove K1 -injectivity.
[209, prop. 5]:
Tk Tk∗ = 1.) Define Rk := T2 T1 · . . . · T1 ,
P
(Let T1 , . . . , Tn isometries with k
where we take k-times T1 .
??
Is it not simply the original argument of J. Cuntz ?
Let (Uk is not a unitary !)
X
Uk = Vp Rk Vp∗
p∈Γk

Then φ(x) = limk→∞ Uk∗ xUk for x ∈ On .

???????
(What happens in En in place of On ??)
R
Here φ(x) := G σ(g)(x) dµG (g). It is different defined in the paper. (Above
defined as P (x) for the circle action.)
How is the group G defined?? Likely by circle action – or by Z2 -action:
χ : {−1, 1} ?
Here

Γk := {1, 2, . . . , n}k := p = (p1 , . . . , pk ) ; pj ∈ {1, 2, . . . , n}



and Vp := Tp1 · . . . · Tpk .

 2. SEMI-PROJECTIVITY OF THE En AND On 1071

(Also written somewhere as Tk (w) for w = p1 . . . pk , pj ∈ {1, . . . , n}).

2. Semi-Projectivity of the En and On

Recall that E2 := C ∗ (s1 , s2 ; s∗j sk = δjk 1). The C *-algebra E2 satisfies a

version of weak semi-projectivity shown in the following lemma.
Lemma A.2.1. If A is the closure of an increasing sequence of C*-subalgebras
A1 ⊆ A2 ⊆ · · · and ψ : E2 → A is a (not necessarily unital) C*-morphism, then
for every ε ∈ (0, 1) there exist n = n(ε) ∈ N, a C*-morphism ψn : E2 → An and
elements h1 , h2 ∈ A of norm khk k ≤ ε (k = 1, 2) such that h∗k = −hk and ψn (1) :=
exp(−h1 )ψ(1) exp(h1 ), kψ(sj ) − ψn (sj )k < ε and ψn (sj ) = exp(h2 )ψ(sj ) exp(h1 )
for j = 1, 2.

Reorganize following proof:

Explain the relations between

‘‘k1 − uk < 1 small’’ and u := exp(h) for h ∈ A with h∗ = −h and ‘‘khk
small’’.

Start then (! 1st) the unital case with

range projections in An for all n.

Do (2nd) the reduction from unital case where

the range projections are not in An
to the case where they are in all An

then (3rd) the reduction from most general

to the unital case.

Proof. Let pk := ψ(sk s∗k ) for k = 1, 2. We consider first the C *-morphism

ψ0 : C ⊕ C → A defined by ψ0 (ξ1 , ξ2 ) := ξ1 p1 + ξ2 p2 .
Let y := p1 − p2 . It has spectrum {−1, 1} ⊆ SpecA (y) ⊆ {−1, 0, 1}. We find
a sequence of selfadjoint contractions yn ∈ An such that y = lim yn . It follows
that there exists n0 such that kyn − yk < 1/8 for all n ≥ n0 . Thus, Spec(yn ) is in
[−1, −3/4] ∪ [−1/4, 1/4] ∪ [3/4, 1] for all n ≥ n0 . Let γ(t) the continuos pice-wise
linear function with (only) break points {−3/4, −1/4, 1/4, 3/4}, values γ(t) := −1
on [−1, −3/4], γ(t) := 0 on [−1/4, 1/4], γ(t) := 1 on [3/4, 1] and γ(t) linear on
[−3/4, −1/4] ∪ [1/4, 3/4].
Then Spec(γ(yn )) ⊆ {−1, 0, 1} for all n ≥ n0 , and p1,n := γ(yn )+ , p2,n :=
γ(yn )− are projections for n ≥ n0 and ψ0,n (ξ1 , ξ2 ) := ξ1 p1,n + ξ2 p2,n are C *-
morphisms from C ⊕ C into An (n ≥ n0 ) that satisfy limn ψ0,n (ξ1 , ξ2 ) = ψ0 (ξ1 , ξ2 ).
More precisely, there exist h1,n ∈ A with h∗1,n = −h1,n and kh1,n k ≤ arcsin????,
such that
ψ0,n (1, 1) = exp(−h1,n )(p1 + p2 ) exp(h1,n )
1072 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

for all n ≥ n0 .
Let zj := ψ(sj ), P := z1∗ z1 = z2∗ z2 , pj := zj zj∗ ≤ P (j = 1, 2). Notice that
p1 + p2 ≤ P (implies p1 p2 = 0).
S
Since the self-adjoint contractions in the algebraic *-algebra n An are dense
in the self-adjoint contractions of A, we can start with a sequence x∗n = xn ∈ An
with kxn k ≤ 1 and lim kxn − P k = 0. Let γn := kxn − P k. Then the spectrum
Spec(xn ) of xn is contained in [−γn , γn ] ∪ [1 − γn , 1]. Thus, Pn := ϕ(xn ) ∈ An is a
projection with kxn − Pn k ≤ γn for all n with γn < 1/4 where

ϕ(t) := min(1, 2 max(0, 2t − 1)) = 4[(t − 1/2)+ − (t − 3/4)+ ] .

We get the existence of n0 such that there are projections Pn ∈ An with kPn −P k ≤
1/2 for n ≥ n0 and that limn kPn − P k = 0.
Let λ(t) a “suitable” continuous strictly increasing functions on [0, 2] with
λ(0) = 0.
By ????
Check Ref’s for changes!!
Part (v) of Lemma 4.1.3
there exist hn ∈ A+ with khn k ≤ arcsin kPn − P k such that e−ihn P eihn = Pn .
Notice that here eihn is build in A + C1 after adjoining a unit to A and that
|eit − 1| ≤ |t| (for t ∈ [−π, π]) implies

k1 − eihn k ≤ khn k ≤ arcsin kPn − P k .

An opposite estimate follows from 2| sin(t/2)| = | exp(−it/2) − exp(it/2)| =

|1 − exp(it)| for t ∈ [−π, π] and gives khk ≤ 2 arcsin(k1 − exp(ih)k/2) for h∗ = h
with khk ≤ π/2.
Notice also that k1−uk < 1 implies for unitary u that s := log(u) = log(1−(1−
u)) exists, satisfies s∗ = −s, ksk ≤ π/3 and exp(s) = u. (One can consider C ∗ (u) as
a quotient of C(S 1 ) and reformulate it as properties of characters of C(S 1 ) i.e., as
conditions on points of S 1 ⊆ C.) Using that | log(1−t)| ≤ − log(1−|t|) ≤ |t|/(1−|t|)
for |t| < 1, we get also the estimate ksk ≤ − log(1 − k1 − uk) ≤ k1 − uk/(1 − k1 − uk)
if u is a unitary with k1 − uk < 1. Let h := −is, then u = exp(ih) and

khk ≤ k1 − uk/(1 − k1 − uk) .

If we take some suitable ` > n0 such that kh` k < λ(ε), then we can replace A
by P` AP` , ψ by the then unital C *-morphism ψ 0 : E2 → P` AP` b ∈ E2 → A given
by ψ 0 (b) := e−ih` ψ(b)eih` and consider only the P` An P` with n ≥ ` > n0 . We store
n0 , n` and H0 := h` in our bookkeeping.
This reduces the considerations to unital A and a unital C *-morphism ψ : E2 →
A, i.e., now zj∗ zk = δj,k 1. We proceed in the same way as above with the projections
pj = zj zj∗ and find self-adjoint h1 ∈ A, n1 < n2 in N and self-adjoint h2 ∈ (1 −
q1 )A(1 − q1 ) both with kh1 k, kh2 k ≤ λ(ε), such that e−ih1 p1 eih1 =: q1 ∈ An1
 3. ON THE JIANG-SU ALGEBRA Z 1073

and e−ih2 e−ih1 p2 eih1 eih2 =: q2 ∈ An2 . Notice that also e−ih2 e−ih1 p1 eih1 eih2 = q1 ,
because q1 eih2 = eih2 q1 = q1 . Again store n1 < n2 and H1 := h1 and H2 := h2 to
our list.
In this way we have reduced the consideration to the case where we can start
with the more comfortable assumptions zj∗ zk = δjk 1 and pj = zj zj∗ ∈ An for
all n ∈ N. Let Sn denote the closed unit-ball of An and S the closed unit-ball
of A. Since (pj Sn )n is an increasing sequence of closed convex subsets of pj S
T
and pj S is the closure of n pj Sn there exists for given ε > 0 and m ∈ N an
n := n(ε, m) ∈ N with n > m and dist(zj , pj Sn ) < λ(ε) for j = 1, 2. It means that
we can find wj ∈ pj Sn with kwj − zj k < λ(ε) (j = 1, 2). It implies that wj wj∗ ≤ pj ,
1 − 2ε ≤ wj∗ wj ≤ 1 and kwj wj∗ − pj k < 2λ(ε). Thus, (wj∗ wj )−1/2 exists and satisfies

1 ≤ (wj∗ wj )−1/2 ≤ (1 − 2λ(ε))−1/2 .

Let uj := wj (wj∗ wj )−1/2 . Then uj is an isometry with pj uj = uj and

kuj u∗j − pj k ≤ 2λ(ε) + [(1 − 2λ(ε))−1/2 − 1] < 1 ,

i.e., with uj u∗j = pj and

kuj − zj k ≤ λ(ε) + [(1 − 2λ(ε))−1/2 − 1] .

The functions λ(t) should be chosen such that λ(ε) + [(1 − 2λ(ε))−1/2 − 1)] < ε/3,
and that |eit − 1| ≤ 2λ(ε) + 2[(1 − 2λ(ε))−1/2 − 1] implies |t| ≤ ε/2.
It follows that U := u1 z1∗ + u2 z2∗ + (1 − p1 − p2 ) is a unitary with U zj = uj and

kU − 1k ≤ ku1 − z1 k + ku2 − z2 k ≤ 2λ(ε) + 2[(1 − 2λ(ε))−1/2 − 1] .

Then uj = U zj with unitary U such that kU − 1k is sufficiently small.

Summing up we get that that uj = U V ∗ zj V with V := eiH0 eiH1 eiH2 and
k1 − V k????? and k1 − U V k ≤ k1 − U k + k1 − V k.
Since |1 − exp(it)|2 = 2(1 − cos(t)) the norm kHk ≤ π/2 of H ∗ = H approaches
zero if k1 − exp(iH)k goes to zero. This allows to show that products of unitaries
sufficiently near to 1 are exponentials exp(iH) with H ∗ = H of small norm. 

3. On the Jiang-Su algebra Z

Let m, n ∈ N = {1, 2, . . .} natural numbers with g.c.d.(n, m) = 1, the

dimension-drop C *-algebra E(Mm , Mn ) is defined as the C *-subalgebra of
C([0, 1], Mm ⊗Mn ) consisting of the maps f : [0, 1] → Mm ⊗Mn with f (0) ∈ Mm ⊗1n
and f (1) ∈ 1m ⊗ Mn .
(We call it also “winding around” algebra ... build by a construction on pairs
of algebras similar to the construction of a “Join” polyhedron in Rm × Rn from two
polyhedra in Rm and Rn ... )
Need: Suitable central sequences, ...
There exists a trace-collapsing unital endomorphism from E(M2∞ , M3∞ ) into
E(M2∞ , M3∞ ).
1074 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

Definition A.3.1. The Jiang-Su algebra is isomorphic to the inductive limit

of this endomorphism ...

There are several other definitions. See original Jiang-Su paper ... Needed
properties of the Jiang-Su algebra...
Containment of Jiang-Su algebra in other algebras, e.g. infinite tensor products
of suitable amenable unital C *-algebras.
Need:
Special Central sequences of the JS algebra that allow to apply p.i. and s.p.i.
criteria.
Each tensorial self-absorbing C *-algebra absorbs the Jiang-Su algebra. (Win-
ter)
Compare with my homotopy criterium ...
Something lost??

4. Fix-point algebras of compact group actions

Suppose that G is a compact group and α : G → Aut(A) a point-norm contin-

uous group homomorphism from G into the group Aut(A) of automorphisms of A.
It is not difficult to see, that the map
Z
G
a 7→ P (a) := α(g)(a) d(g)
G

defines a faithful conditional expectation P G from A onto the fix-point algebra AG

of G. Here “faithful” means that P G (a) 6= 0 for all non-zero a ∈ A+ .
Notice that one can not conclude similar properties for A from the correspond-
ing property of the fix-point C *-subalgebra of a Z-action of A (by an automorphism
of A).

Proposition A.4.1. Suppose that a compact group G acts continuous in point-

norm topology on a C*-algebra A.
The fix-point subalgebra AG of A is nuclear (respectively is exact, weakly injective,
QWEP, has the local lifting property), if and only if, A is nuclear (respectively is
exact, weakly injective, QWEP, has the local lifting property).
Moreover, if the conditional expectation PG can be approximated point-wise by
inner c.p. maps
n(τ )
X
Vτ (a) := d∗k,τ adk,τ ,
k=1

then each non-zero closed ideal of A has non-zero intersection with AG .

It is interesting that the question if PG respects also local reflexivity – as defined

for operator spaces, which is different from the well-known (and almost trivial) local
reflexivity of Banach spaces – remains open even in case of circle actions!
 4. FIX-POINT ALGEBRAS OF COMPACT GROUP ACTIONS 1075

It could be that the new operator exact sequences criteria for left-ideals of
(stable) locally reflexive C *-algebras implies now a positive answer for “local re-
flexivity” ... ???
But this requires to check what happens with the the fixed points under com-
pact actions?
Question: What about “relative” properties? In particular, “relative weak
injectivity” could be a candidate ...
Equivalent formulation of r.w.i. is given by: A ⊆ B r.w.i. in B if and only if
A ⊗max C ∗ (F∞ ) ⊆ B ⊗max C ∗ (F∞ ) .
If G acts on B by ρ : g ∈ G 7→ ρ(g) ∈ Aut(B) and ρ(g)(A) = A for all g ∈ G,
then B G ∩ A is r.w.i. in B G ?
Here B G denotes the fix-point algebra of the G-action on B. Seems to work
for compact G.

Proof. Wa adapt an idea of A. Grothendieck for the study of locally convex

vector spaces with help of tensor product functors. A simple variant of his idea
transferred to the study of analytic properties of C *-algebras is to express this
properties with help of tensor product functors and use the automatic G-invariance
of the functors. We use this proof to say something more general about the appli-
cations of this method and its limitation.
Let (at the moment) C denote a fixed C *-algebra. Let α : G → Aut(A) denote
the G-action on A, and let

(B, C) → B ⊗µ C and (B, C) → B ⊗ν C

denote C *-algebra tensor product “functors” that satisfy k · kµ ≥ k · kν on the

algebraic tensor product B C for all C *-algebras B, i.e., the closed ideals Jµ (B)
and Jν (B) of maximal C *-algebra tensor product B ⊗max C given as kernels of the
natural *-epimorphisms B ⊗max C → B ⊗µ C and B ⊗max C → B ⊗ν C satisfy
Jµ (B) ⊆ Jν (B). (In fact, a suitable functorial selection of those “kernels” contains
all necessary information on the tensor product functors!)
They are defined by a B-functorial choice of C *-norms k · kµ ≥ k · kν on the
algebraic tensor products B C and are the corresponding topological tensor-
products are the completions of B C with this norms. The functoriality means
that for each C *-morphism h : A → B and x ∈ A C in the algebraic tensor
product holds that kh ⊗ id(x)kµ ≤ kxkµ and similar for k · kν . Or, equivalently,
(h ⊗max id)(Jµ (A)) ⊆ Jµ (B) for each C *-morphism h : A → B and fixed C.
With other words, we allow only “natural” functor transformations

ϕ B : B ⊗µ C → B ⊗ν C

that satisfy
ϕB ◦ (h ⊗µ id) = (h ⊗ν id) ◦ ϕA
for h : A → B .
1076 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

In addition, we suppose, that V ⊗µ id and W ⊗ν id remain contractions, whenever

V and W are c.p. contractions. It is not difficult to see, – with help of Kasparov-
Stinespring dilations of V and using the outer-unital extension Ve : A e→B e –, that
this property is equivalent to the property of ⊗µ (and similar of ⊗ν ) that B ⊗µ C ⊆
Be ⊗µ C and pBp ⊗µ C ⊂ B ⊗µ C for all C, and all full projections p ∈ B . (We let
it to reader to check this.)
Let ???????
Thus, AG is nuclear (respectively, weakly injective, QWEP, locally liftable) if
A is so.
Exactness always passes to C *-subalgebras.
The conditional expectations P G are faithful for compact G. Indeed, if a ≥ 0
and χ is a positive functional on A with χ(a) > 0, then χ(P G (a)) > 0, because
it is the integral over the non-zero, non-negative, and continuous function g 7→
χ(α(g)(a)).
The maps η ⊗µ id : AG ⊗µ C → A ⊗µ C and V ⊗µ id : A ⊗µ C → AG ⊗µ C satisfy
(V ⊗µ id)(η ⊗µ id) = id. Thus, AG ⊗µ C is naturally isomorphic to the closure of
the subspace AG C in A ⊗µ C, i.e., AG ⊗µ C ⊂ A ⊗µ C.
Now let the inclusion A ⊂ B be G-equivariant, for some (point-norm) contin-
uous G-action β : G → Aut(B), i.e., αg (a) = βg (a) for a ∈ A and g ∈ G. Then it
holds:
If AG ⊗µ C ⊂ B G ⊗ν C, then A ⊗µ C ⊂ B ⊗ν C.
Indeed, let J denote the kernel of the natural C *-morphism ϕ : A ⊗µ C →
B ⊗ν C. The functoriality of ⊗µ and ⊗ν implies
ϕ ◦ (α(g) ⊗µ idC ) = (β(g) ⊗µ idC ) ◦ ϕ .
Thus, α(g) ⊗µ idC (J) = J for all g ∈ G.
One can see, with help of elementary tensors a ⊗ c, that g 7→ (α(g) ⊗µ idC ) and
g 7→ (β(g) ⊗ν idC ) are point-norm continuous, and that the corresponding faithful
G
: x 7→ (α(g) ⊗ id)(x) dg and P2 := PYG : y 7→
R
conditional expectations P1 := PX
(α(g) ⊗ id)(x) dg map X := A ⊗µ C onto the closure of AG C in A ⊗µ C,
R

respectively, maps Y := B ⊗ν C onto the closure of B G C in B ⊗ν C and satisfy

ϕ ◦ P1 = P2 ◦ ϕ. for ϕ : A ⊗µ C → B ⊗ν C. It follows, that P1 (J) ⊂ J ∩ AG ⊗ C
Since AG ⊗µ C ⊂ A ⊗µ C and B G ⊗ν C ⊂ B ⊗ν C are complemented subspaces,
we get that PG (J) is in the kernel of the monomorphism AG ⊗µ C → B G ⊗ν C
(after natural identifications of AG ⊗µ C and B G ⊗ν C with its images). Thus
PG (J) = {0}, and J = 0.

Now we apply the latter to our cases:

One obtains that A is nuclear if AG is nuclear if one uses above ⊗µ := ⊗max ,
⊗ν := ⊗min , B := A and C arbitrary.
It has been shown in [431, prop.1.1] that B is weakly injective, if an only if,
B ⊗max C ∗ (F ) = B ⊗min C ∗ (F ), where F is the free group on countably many
 4. FIX-POINT ALGEBRAS OF COMPACT GROUP ACTIONS 1077

generators. Thus, we obtain that A is weakly injective if AG is weakly injective, if

we let ⊗µ := ⊗max , ⊗ν := ⊗min , A = B and C := C ∗ (F ).
By [431, prop.1.1], B has the local lifting property, if and only if, B ⊗max
L(`2 ) = B ⊗min L(`2 ) . We get that A has the local lifting property if AG has the
local lifting property, if we take ⊗µ := ⊗max , ⊗ν := ⊗min , A = B and C := C ∗ (F )
To see that A has the QWEP if AG has the QWEP, one can take ⊗µ :=
⊗max , ⊗ν := ⊗` , B := A, C := C ∗ (F ), because B has QWEP, if and only if,
B ⊗max C ∗ (F ) = B ⊗` C ∗ (F ) .
Here (generally) B ⊗` C – “the minimal left-exact C *-tensor-product functor”
– is defined by the completion of B C with respect to the C *-norms
X X
k bk ⊗ ck k` := sup NF ( bk ⊗ ck ) ,
F,π
k k

given by (sufficiently many) C *-algebras F and epimorphisms π : F → B and,

X X
NF ( bk ⊗ ck ) := dist( fk ⊗ ck , I ⊗min C) ,
k k

where π(fk ) = bk and I := ker(π), and with distance taken in F ⊗min C.

That the exactness of A follows from the exactness of AG , can be seen by
putting ⊗ν := ⊗min , ⊗µ := ⊗r , B := A and consider arbitrary C.
Here we define the functor (B, D) 7→ B ⊗r D – “the minimal right exact C *-
tensor-product functor” – from D ⊗` B by
X X
k bk ⊗ dk kr := k dk ⊗ bk k` ????? .
k k

??? More ????

If PG can be approximated point-wise by inner c.p. maps
n(τ )
X
Vτ (a) := d∗k,τ adk,τ ,
k=1

then PG (J) ⊂ J ∩ AG .
Since PG is faithful on positive elements, each non-zero closed ideal of A has
non-zero intersection with AG . 

Question A.4.2. Is the class of locally reflexive C *-algebras is invariant under

crossed product by abelian compact groups G?
(Could this follows from Takai Duality: A ⊗ K ∼
= (A o G) o G.
b For discrete
groups H acting on B – e.g. for H := G
b and

????
holds B ⊆ B o H. Now take B := A o G and H := G.)
b

The Question is:

1078 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

Can we deduce from the local reflexivity of the crossed product A o G (re-
spectively of the fixed point algebra AG ) of A by a continuous action of a compact
group G on A that A itself is locally reflexive?
Even in the case of circle actions G := T (= S 1 ) the answer is not known.

5. Comparison of elements in C *-algebras

HERE is some collection of observations on ≈ and - and some easy

consequences:
Notice a ≈ a∗ a ∼M vN aa∗ in A.
It holds a ⊗ b ≈ (a + b) ⊗ 0 in M2 (A) if a∗ b = 0 = ba∗ . Indeed: It implies
b a = 0, a∗ ab∗ b = 0 and Thus (a + b) ≈ (a + b)∗ (a + b) = a∗ a + b∗ b,
∗

It follows that the ≈-classes [K(H)]approx for all (non-zero) Hilbert spaces H is
isomorphic to {0, 1, . . . , Dim(H)} if H has finite dimension and {0, 1, 2, . . .} ∪{+∞}
if H ∼
= `2 (N).
S, T ∈ K(`2 (N)) satisfy S ≈ T if an only if the ranks (= dimension of image)
of S ∗ S and T ∗ T are the same.
The positive operators g(α) := (α, α2 , α3 , . . .) ∈ K(`2 (N)) for α ∈ (0, 1) are
properly infinite in K(`2 (N)) with trace = α/(1 − α), because g(α) - g(α2 ) =
(α2 , α4 , α6 , . . .) , and g(α) - α · g(α2 ) = (α3 , α5 , α7 , . . .) imply g(α) ⊕ g(α) - g(α),
because αg(α) ∼M vN g(α2 ) ⊕ α · g(α2 ) in K(`2 (N)) and ∼M vN implies ≈. But
αg(α) ≈ g(α).
Notice also that inside the diagonal operators in K(`2 (N)) it holds g(α)2 =
g(α2 ). It gives g(α) ⊕ g(α) - g(α)2 ⊕ αg(α)2 ≈ α · g(α).
The g(α) ∈ K+ are all properly infinite and limn→∞ kg(1/n)k ≤ 1/n with
traces Tr(g(α)) = α/(1 − α) .
Asymptotic: 2α/(1 − α) versus (1 + α)α2 /(1 − α2 ) on α ∈ (0, 1) behave simi-
lar/different for α % 1.
A flexible almost “topological” version of comparison has been introduced in
special cases in the papers [170, 171] and was later generalized e.g. by M. Rørdam
in [690] and others to the following definition.

Definition A.5.1. The Cuntz comparison between elements a, b ∈ A is

defined as follows:
We say that b majorizes a (in A and in the sense of J. Cuntz [170, 171]),
and write it as a - b or b % a), if there are sequences c1 , c2 , . . . ; d1 , d2 , . . . ∈ A such
that a := limn cn bdn .
Below it is shown that a - b is a transitive order relation, i.e., a - b and b - c
imply a - c. Obviously always a - a.
Thus, we can define an equivalence relation a ≈ b by a - b and b - a. It has
the obvious property a ≈ a for all a ∈ A.
 5. COMPARISON OF ELEMENTS IN C *-ALGEBRAS 1079

move to/ or see one of the below given lemmas:

It is evident that a - b is a transitive relation: a - a because e.g. always
a = limn (aa∗ )1/n a(a∗ a)1/n , and a = limn gn c hn for suitable choices gn := ckn e`n
and hn := f`n dkn if e1 , e2 , . . . ; f1 , f2 , . . . ∈ A satisfies b = limn en c fn .
Elements a, b ∈ A are Cuntz equivalent (notation: a ≈ b) if a - b and b - a.
Since a - a and - is transitive, ≈ is an equivalence relation in the usual sense. The
a ≈ b–equivalence classes of a ∈ A ⊗ K in A ⊗ K will be denoted sometimes by [a]≈
or simply by [a] for elements a in the algebraic inductive limit

indlimn Mn (A) = indlimn A ⊗ Mn ⊆ A ⊗ K

(but sometimes even for a ∈ A ⊗ K). The corresponding semigroup of this classes
with Cuntz-addition has later been written as W (A) by M. Rørdam, cf. [690]. We
denote by Cu(A) the classes in A ⊗ K, and call it the large Cuntz semi-group.
Since M(A ⊗ K) contains a copy of E2 := C ∗ (S1 , S2 ; Sj∗ Sk = δj,k · 1) unitally
(It contains also a copy of O2 , but that is for the definitions here not of interest), we
can define a (Cuntz-)addition of elements by a⊕b := S1 aS1∗ +S2 bS2∗ for a, b ∈ A⊗K,
it induces an addition of the ≈ classes, and allow to define the “general” and the
“local” Cuntz semigroups, namely the “general” for all elements of A ⊗ K and the
“local” for the Pedersen ideal P (A ⊗ K) of A ⊗ K. Notice here that P (A ⊗ K) is
usually different from the – naturally in P (A ⊗ K) as inductive limit contained –
T
union (respective inductive limit) n∈N Mn (P (A)), but from the definition of the
Pedersen ideals P (A) of A and P (K ⊗ A) of K ⊗ A as “the minimal dense algebraic
ideals” it is easy to see that each element x ∈ P (A ⊗ K)+ is in the stronger sense of
Murray–von-Neumann equivalent (see below) to some element in y ∈ Mn (P (A))+
for some n := n(x) ∈ N.
S
The semigroup W (A) for elements in n Mn (A) as introduced by M. Rørdam,
[690], is in general strictly between the “large” and the “small” Cuntz semi-groups
Cu(A) and CS(A).
It is often considered as the algebraic inductive limit

Mn ⊗ A = Mn (A) → Mn+1 (A) = Mn+1 ⊗ A

via the embedding Mn ⊕ {0} ⊆ Mn+1 . But this is inessential because - and ≈ are
weaker than ≤ and ∼MvN combined, but this requires some care.
A stronger equivalence relation ∼ is the Murray–von-Neumann equivalence:
Elements a, b ∈ A+ are MvN-equivalent (denoted by a ∼ b or a ∼MvN b), if they
are equivalent in sense of Murray and von Neumann, i.e., if there exists d ∈ A with
d∗ d = a and dd∗ = b. Indeed, a ≈ b because one can take dn := d((d∗ d)1/2 +1/n)−1
and cn := d∗n to get a := lim d∗n bdn and b := lim dn ad∗n .
We define n-step majorization (denoted b n a) by

a ⊕ 0n−1 ∼
= a ⊗ p11 - b ⊗ 1n .
1080 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

?? Check all chapters for closed ideals: J(b), Jb , I(b)

generated by an element or subset of B.
Decide which of them is the best and change others!!!

Definition A.5.2. Let B a C *-algebra. A non-zero contraction c ∈ B will be

called a scaling element if c∗ cc = c. We call a scaling element c ∈ B “strictly
scaling” if in addition c∗ c 6= cc∗ .
A positive contraction b ∈ B+ is a scale generator if there exists d ∈ B and
0 < δ < ε < 1 such that d∗ (b − ε)+ d = (b − δ)+
In particular then, (b − δ)+ is in the closed ideal generated by (b − ε)+ .
A positive element b ∈ B+ is a compactly supported element in B if there
exist δ ∈ (0, kbk) such that b is contained in the closed ideal generated by (b − δ)+ .
(In particular each projection p ∈ B is compactly supported.)
The element b ∈ B is called compactly n-supported if there exists δ > 0
such that b ⊗ p11 - (b − δ)+ ⊗ 1n .

Remark A.5.3. An equivalent definition of a scaling element c ∈ B is that

e := c∗ c is a positive contraction in B with ecc∗ = cc∗ = cc∗ e.
A scaling element c ∈ B that is not strictly scaling in sense of our Definition
A.5.2 is simply a normal element in B with the property that Spec(c) ⊆ {0} ∪ S 1 .
Notice that B. Blackadar and J. Cuntz call elements c ∈ B in [78, def. 1.1]
“scaling” only if they are strictly scaling elements in sense of our Definition A.5.2.

Remark A.5.4. The usual definition of compactness implies for Hausdorff

spaces X that compact subsets of X are closed. But notice that for non-Hausdorff
spaces this is not the case if we take the usual definition of a “compact” subset
S
Y ⊆ X that we find in every family V ⊆ O(X) with Y ⊆ U ∈V U a finite collection
U1 , . . . , Un ∈ V with Y ⊆ U1 ∪ . . . ∪ Un . This kind of subsets Y of X has been
called in the Bourbaki monographs “quasi-compact” for non-Hausdorff spaces.
We have to consider sometimes the T0 spaces X := Prim(A) for separable
C *-algebras A – or in cases where A is non-separable its “point” completions
X := prime(A) ( 4 ). We write often “compact” instead of “quasi-compact” in
our considerations.
Moreover this kind of (“quasi -”) compact subsets K of X are even not “satu-
rated ”, i.e., are not the intersections of all open subsets of X that contain K. Only
the very special “coherent” spaces X defined by the property that each (quasi-)
compact subset of X is saturated in this sense.
The class of saturated spaces does not contain Prim(A0 ) for

A0 := { f ∈ C([0, 1], M2 ) ; f (0) ∈ C ⊕ C = diag(M2 ) } .

4 Such “adding” of all prime closed subsets of a T space X to the set of points in X also
0
called a “sobrification” of X, in a sense it is a kind of completion of X that does not change its
lattice of open subsets.
 5. COMPARISON OF ELEMENTS IN C *-ALGEBRAS 1081

The notions of Definition A.5.2 will be justified by the following Lemma:

Lemma A.5.5. Let b ∈ B+ , c ∈ I((b − γ)+ )+ for some γ > 0.

(i) For each ε > 0 there exists n := n(ε) ∈ N such that (b − ε)+ ⊗ p11 - c ⊗ 1n
in A ⊗ Mn , if and only if, b is in the closed ideal generated by c. In
particular, then b ∈ I((b − γ)+ ).
(ii) If b is contained in the closed ideal I((b − γ)+ ), then the Dini function
fb : J ∈ Prim(B) 7→ kπJ (b)k on Prim(B) has an open and compact subset
Ub of Prim(B) as its support, and fb (J) ≥ γ for all J ∈ Ub . In particular,
the closed ideal I(b) = span(BbB) has (quasi-) compact primitive ideal
space Prim(I(b)).
(iii) If the closed ideal I(b) of B generated by b ∈ B+ has compact primitive
ideal space, then there exists γ > 0 such that b is in the closed ideal
generated by (b − γ)+ .
In particular, for each ε > 0 there exists n := n(ε) ∈ N and
Pn
d1 , . . . , dn ∈ B such that (b − ε)+ = k=1 d∗k (b − γ)+ dk .
(iv) Let c ∈ B a scaling element, i.e., kck ≤ 1 and c∗ cc = c . The element
V (c) := c + i(1 − c∗ c)1/2 is an isometry in B + C · 1, and the projection
p := p(c) := 1 − V (c)V (c)∗ = (c∗ c − cc∗ ) + i(c(1 − c∗ c)1/2 − (1 − c∗ c)1/2 c∗ )
is a projection in the closed ideal I(c) of B.
The ideal I(p) generated by p contains a sequence of mutually orthog-
onal Murray–von-Neumann equivalent projections p = p(c) =: p1 , p2 , . . ..
(v) The closed ideal I(p) ⊆ I(c) given by the projection p ∈ I(c) of part (iv)
is generated as a closed ideal by c∗ c − cc∗ ∈ I(c).
The ideal I(p) contains c∗ c(1 − c∗ c) and each d ∈ B that satisfies
dc = 0 and dc∗ c = d.
(vi) If b and γ satisfy (iii) and that (b − γ)+ is properly infinite, then for each
ε ∈ (0, γ) there exists dk := dk (ε) ∈ B (k = 1, 2) such that
dj (b − γ)+ dk = δjk min(1, ε−1 (b − ε)+ ) = δjk ε−1 (b − ε)+ − (b − (1 + ε))+ .
∗

1/2
In particular, the elements ck := (b − γ)+ dk are “scaling” elements in
B with
c∗j ck = δjk c∗1 c1 , c∗k ck ck = ck and kck k = 1
that generates the closed ideal I(b) := span(BbB) of B.
(vii) The closed ideal I(p1 ) of B generated by p1 := 1−V (c1 )V (c1 )∗ with scaling
elements c1 , c2 ∈ B as in Part (vi) coincides with I(b). In particular,
I(b) is generated by a hereditary stable C*-subalgebra D that is generated
by a sequence of mutually orthogonal and mutually equivalent projections
p1 , p2 , . . ..

Proof. (i): Let b ∈ B+ , and c ∈ I((b − γ)+ )+ for fixed γ > 0 .

If b is contained in the closed ideal I(c) generated by c then for each ε > 0 there
exists n ∈ N and d1 , . . . , dn ∈ B with kb − k d∗k cdk k < ε/2 . By norm-continuity
P
1082 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

of δ 7→ (c − δ)+ and by Lemma 2.1.9, there exists a contraction e ∈ B such that

X
(b − ε)+ = (dk e)∗ (c − δ)+ (dk e) .
1≤k≤n

We can this equation express by (b − ε)+ ⊗ p11 = D∗ ((c − δ)+ ⊗ 1n )D in B ⊗ Mn

P
with column D := k (dk e) ⊗ pk1 ∈ Mn,1 (B).
In particular, for each ε > 0 there exists n := n(ε) ∈ N such that (b−ε)+ ⊗p11 -
c ⊗ 1n in A ⊗ Mn .
Suppose now that for each ε > 0 there exists n := n(ε) ∈ N such that (b−ε)+ ⊗
p11 - c ⊗ 1n in A ⊗ Mn . It says that there exists a sequence of column matrices
D1 , D2 , . . . ∈ Mn,1 (B) ⊆ Mn (B) with (b−ε)+ ⊗p11 = limn Dn∗ (c⊗1n )Dn in Mn (B).
It follows that (b − ε)+ ⊗ p11 ∈ Mn (I(c)) with I(c) / B the ideal generated by c. It
follows that (b − ε)+ ∈ I(c) ⊆ I((b − γ)+ ) for each ε > 0. Thus, b is in the closed
ideal I(c) generated by c, in particular, b ∈ I((b − γ)+ ).
(ii): Recall here that fb−1 [γ, ∞) ⊆ Prim(A) is (quasi-) compact for every γ > 0,
cf. [213, prop. 3.3.7], where we consider here the Dini function fb on Prim(A)
defined by fb (J) := kπJ (b)k for b ∈ B and J ∈ Prim(B).
If b is contained in the closed ideal
\
I((b − γ)+ ) = {J ∈ Prim(B) ; kπJ (b)k ≤ γ } ,
then the Dini function fb : J ∈ Prim(B) 7→ kπJ (b)k ∈ [0, ∞) on Prim(B) has
support
Ub := fb−1 (0, ∞) = {J ∈ Prim(B) ; fb (J) > 0} .
This set Ub is identical with the (quasi-) compact subset
fb−1 [γ, ∞) = {J ∈ Prim(B) ; fb (J) ≥ γ} .

Indeed: The inequality fb (J) = kπJ (b)k < γ for J ∈ Prim(A) implies that
there are ε > 0 and δ ∈ [fb (J), γ) such that πJ ((b − δ)+ ) = 0 and, – by assumption
b ∈ I((b − γ)+ ) –, that there exist d1 , . . . , dn ∈ B with (b − ε)+ = k d∗k (b − δ)+ dk .
P

If we apply to this πJ , we see that

X
(fb (J) − ε)+ = kπJ ((b − ε)+ )k ≤ kdk k2 kπJ ((b − δ)+ )k = 0 .
k

Thus, fb (J) ≤ ε for all ε > 0, i.e., fb (J) = 0 if fb (J) < γ. It means Ub :=
fb−1 (0, ∞) = fb−1 [γ, ∞).
Since the set fb (J) = kπJ (b)k ≥ γ is a compact subset of Prim(B), the closed
T
ideal I(b) = span(BbB) = {J ∈ Prim(B) ; fb (J) = 0 } has compact primitive
ideal space Prim(I(b)) = fb−1 [γ, ∞).
(iii): Suppose that I(b) has compact primitive ideal space Prim(I(b)) ⊆
Prim(B). The corresponding open subset of Prim(B) is given by fb−1 (0, ∞). The
family of open subsets fb−1 (δ, ∞) of fb−1 (0, ∞) with δ > 0 covers the compact space
fb−1 (0, ∞). Thus, there exists γ > 0 such that fb−1 (γ, ∞) = fb−1 (0, ∞). It is easy
to see that fb−1 (γ, ∞) = fc−1 (0, ∞) for c = (b − γ)+ . It follows that (b − γ)+ and b
generate the same closed ideal of B.
 5. COMPARISON OF ELEMENTS IN C *-ALGEBRAS 1083

This implies that for each ε > 0 there exists n := n(ε) ∈ N and d1 , . . . , dn ∈ B
Pn
such that (b − ε)+ = k=1 d∗k (b − γ)+ dk .
(iv,v): Let c ∈ B a contraction with cc∗ c∗ c = cc∗ . The element V := V (c) :=
c + i(1 − c∗ c)1/2 of B + C · 1 is an isometry, because cc∗ (1 − c∗ c) = 0 implies
c∗ (1 − c∗ c) = 0, c∗ (1 − c∗ c)1/2 = 0 and (1 − c∗ c)1/2 c = 0.
The projection p := p(c) := 1 − V (c)V (c)∗ , is in I(c) ⊆ B. The projections

pn := V n−1 (V ∗ )n−1 − V n (V ∗ )n

are mutually orthogonal, where we let p1 := p(c). If m < n then pm ∼M vN pn by

Z = V n−m pm , i.e., Zpm Z ∗ = pn . In particular, pn ∈ I(p(c)) for all n ∈ N.
The equations c∗ (1 − c∗ c)1/2 = 0 and (1 − c∗ c)1/2 c = 0 imply p := 1 −
V (c)V (c)∗ = (c∗ c − cc∗ ) + i(c(1 − c∗ c)1/2 − (1 − c∗ c)1/2 c∗ ) and that (1 − c∗ c)1/2 p(1 −
c∗ c)1/2 = c∗ c(1 − c∗ c). Thus, J := I(c∗ c(1 − c∗ c)) ⊆ I(p). Since c(1 − c∗ c) ∈ J ,
we get c(1 − c∗ c)1/2 ∈ J and that πJ (p) = πJ (c∗ c − cc∗ ) , i.e., p − (c∗ c − cc∗ ) ∈ J .
The epimorphism πJ maps I(p) ⊇ J onto I(p)/J and has kernel J ⊆ I(p). We get
that the positive element c∗ c − cc∗ must be in I(p).
Let d ∈ B with d∗ c = 0 and dc∗ c = d. Then dd∗ V (c) = 0, because d(1 −
∗
c c) = 0. It follows dd∗ ≤ kdk2 p, which implies d ∈ I(p).
1/2

(vi): Suppose that b ∈ B+ and γ satisfy (iii) and that (b − γ)+ is properly
infinite.
It implies that for each τ > γ and n ∈ N the existence of e1 , . . . , en ∈ B such
that e∗j (b − γ)+ ek = δij (b − τ )+ .
Indeed, for each µ > 0 and n ∈ N there exists a matrix R ∈ Mn (B) that satisfies
∗
R ((b − γ)+ ⊗ p11 )R = ((b − γ)+ − µ)+ ⊗ 1n . Here we use again Lemma 2.1.9 in
Mn (B) to obtain equality. Define µ := τ − γ and e1 , . . . , en by [e1 , . . . , en ] := R
with R for µ. The e1 , . . . , en have the desired property.
Let ε ∈ (0, γ). There exists τ > γ and m := m(ε, τ ) ∈ N g` := g` (ε, τ ) ∈ B
(` = 1, . . . , m) such that
X
g`∗ (b − τ )+ g` = min(1, ε−1 (b − ε)+ ) = ε−1 (b − ε)+ − (b − (1 + ε))+ ,

1≤`≤m

because we find f` ∈ B with

X
k f`∗ (b − γ)+ f` − (b − ε/3)+ k < ε/3
`

and can apply the continuity of τ 7→ (b − τ )+ and the Lemma 2.1.9 to get the
equation proper by multiplying the f` with a suitable contraction e from the right
to reach the equality
X
(f` e)∗ (b − τ )+ (f` e) = (b − 2ε/3)+ .
1≤`≤m

Finally we let g` := f` eh(b) where

h(t) := min(ε, (t − ε)+ )/(ε(t − 2ε/3)+ ) .

1084 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

We take n := 2m in the above given construction of the ek and let d1 :=

P P
1≤`≤m e` g` and d2 := 1≤`≤m em+` g` . Then

d∗j (b − γ)+ dk = δjk min(1, ε−1 (b − ε)+ ) .

1/2
In particular, the ck := (b − γ)+ dk are “scaling” elements in B with c∗j ck =
δjk c∗1 c1 , ck c∗k c∗k ck = ck c∗k , such that I(ck ) = I(b) for the the closed ideals I(ck ) :=
span(Bck B) and I(b) := span(BbB) of B. 

We have sometimes to deduce estimates for the infimum γ(n, a, b, ε) of the

norms k d∗k dk k for d1 , . . . , dn ∈ A with kb − k d∗k adk k < ε. Notice that then
P P

there are contractions e1 , . . . , em ∈ A such that kb − j e∗j aej k < ε and m ≤

P

n(γ(n, a, b, ε) + 1).

Lemma A.5.6. Let X = [bk` ] ∈ Mn (B)+ , d1 , . . . , dn ∈ B, and let diag(X) ∈

Mn (B) denote the diagonal matrix with entries b11 , . . . , bnn in the diagonal of X.

(i) The linear map X 7→ n diag(X) − X on Mn (B) is completely positive. In

particular, X ≤ n diag(X) for all X ∈ Mn (B)+ .
(ii) Let Trn ([αjk ]) := (α11 + · · · + αnn ) · 1n for [αjk ] ∈ Mn . The map X 7→
n(idB ⊗ Trn )(X) − X on Mn (B) is completely positive.
In particular, X ≤ n(idB ⊗ Trn )(X) = n(b11 + . . . + bnn ) ⊗ 1n for all
X = [bk,` ] ∈ Mn (B)+ .
Pn
(iii) 0 ≤ k,`=1 bk` ≤ n(b11 + . . . + bnn ) for all X = [bk,` ] ∈ Mn (B)+ .
(iv) For every b ∈ B+ and d1 , . . . , dn ∈ B,

(d1 + · · · + dn )∗ b(d1 + · · · + dn ) ≤ n (d∗1 bd1 + · · · + d∗n bdn ) .

Proof. (i): We identify Mn (M(B)) naturally with M(B)⊗Mn , and Mn with

1 ⊗ Mn . Let z := exp((2πi)/n) and U := diag(1, z, . . . , z n−1 ) ∈ Mn ⊆ Mn (M(B)) .
Then, for a ∈ M(B),
n−1
X X
(U k )∗ (a ⊗ eij )U k = a ⊗ ( z (j−i)k )eij = nδij (a ⊗ eij ) .
k=0 k

0 ∗ 0
Pn−1 k ∗ k
Thus, X = (U ) XU ≤ k=0 (U ) XU = n diag(X) , and the map
n−1
X
X 7→ n diag(X) − X = (U k )∗ XU k
k=1

is completely positive, because is a sum of c.p. maps.

(ii): The map n(idB ⊗ Trn )(X) − X = (idB ⊗(n Trn − id))(X), idB is c.p. on B
P
and we show tat n Trn − id is c.p. on Mn , where we define Tr([αjk ]) := ( k αkk )·1n .
The complete positivity of n Trn − id on Mn can be seen from the formula
X X
n−1 ( αkk ) · 1n = n−2 g −1 [αjk ]g ,
k g∈G

because this implies n( k αkk )·1n −[αjk ] is the sum of n2 −1 inner *-automorphisms
P

g −1 [αjk ]g of Mn .
 5. COMPARISON OF ELEMENTS IN C *-ALGEBRAS 1085

Here G := G(U, V ) ⊆ U(n) = U(Mn ) is the irreducible finite subgroup

G of U(Mn ) with n2 unitary elements that is generated by the unitary U :=
diag(z 0 , z 1 , . . . , z n−1 ) with z := exp (2πi)/n and by the cyclic permutation V ∈
U(Mn ) of the canonical basis {e1 , . . . , en } of Cn , i.e., V (ek ) = ek+1 (k = 1, . . . , n−1)
and V (en ) = e1 . To see |G| = n2 , notice that U j V k = z jk V k U j for j , k ∈
{0,..., n − 1}.
Alternative proof:
The map

S : X 7→ diag(X) = (b11 , . . . , bnn ) ∈ B n = B ⊕ · · · ⊕ B ∼

= diag(Mn (B)) ⊆ Mn (B)
is a c.p. map, because it is a conditional expectation and

diag(X) = n−1 (n diag(X) − X) + n−1 X .

P
It is clear that the maps Tj (b1 , . . . , bn ) := k6=j bk are completely positive, because
they are sums of c.p. maps.
The map (b1 , . . . , bn ) →
7 (b1 + . . . + bn ) ⊗ 1n − diag(b1 , . . . , bn ) ∈ Mn (B) is
completely positive in B n ∼ = diag(Mn (B)), because it is the orthogonal sum of the
c.p. maps Tj .
It follows the positivity of

X 7→ [n(b11 +. . .+bnn )]⊗1n −X = n((idB ⊗ Trn )(X)−diag(X))+(n diag(X)−X) .

(iii): Let C the column C := [1, 1, . . . , 1]> ∈ Mn,1 (M(B)). Apply the c.p. map
Y 7→ C ∗ Y C to Y := n diag(X) − X. Since Y ≥ 0 by part (i), we get
n
X
bk` = C ∗ XC ≤ nC ∗ diag(X)C = n(b11 + . . . + bnn ) .
k,`=1

(iv): Take X := [bjk ] in part (iii) with bj,k := d∗j bdk , i.e., X = D∗ D for the
row matrix D := [b1/2 d1 , · · · , b1/2 dn ] ∈ Mn,1 (B). 

Next remark considers the equivalence given by polar decomposition of a of the

hereditary C *-subalgebras generated by a∗ a and by aa∗ .
Next appears also before proof of THM.E. ??

Remark A.5.7. Let a ∈ A non-zero and a := v(a∗ a)1/2 its polar decomposition
in the W*-algebra A∗∗ , let D := a∗ aAa∗ a and E := aa∗ Aaa∗ the hereditary C *-
subalgebras of A generated by a∗ a, respectively by aa∗ .
Then D = a∗ Aa, E = aAa∗ , and the map ϕ : d ∈ D → vd v ∗ ∈ A∗∗ has image in
E and defines a C *-algebra isomorphism from D onto E. The inverse isomorphism
ϕ−1 is given by e ∈ E 7→ v ∗ ev ∈ D.
It holds ϕ(D ∩ I) = E ∩ I for every closed ideal I / A of A. In particular, if D
is a full hereditary C *-subalgebra of A, i.e., if span(ADA) = A, then also E is full
in A.
1086 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

In particular, D is simple if and only if E is simple.

Proof. Let a ∈ A and a = v(a∗ a)1/2 = (aa∗ )1/2 v its unique (right) polar
decomposition of a in A∗∗ . Then vx ∈ A for every x in the norm-closed right-ideal
R := a∗ aA = a∗ A of A that is generated by (a∗ a)1/2 .
The map x 7→ vx is an isometric right-module isomorphism from R onto the
closed right-ideal aA = aa∗ A . The inverse map from aA onto R is given by
y 7→ v ∗ y, because a∗ = v ∗ (aa∗ )1/2 is the polar decomposition of a∗ in A∗∗ .
This follows from vf (a∗ a) = f (aa∗ )v for each non-negative continuous function
f ∈ C0 (0, kak2 ], and can be easily seen in A∗∗ .
easily seen: more details?
In particular, y := vz α ∈ A and y ∗ y = z 2α for all z ∈ A+ with z ≤ a∗ a and all
α > 0. 
Remark A.5.8. Let D a closed hereditary C *-subalgebra of A.
Each closed ideal J of D is the intersection J = D ∩ I with D of the closed
ideal I := I(J) := span(AJA) of A.
For I1 , I2 ∈ I(A) holds I1 ∩ D = I2 ∩ D if and only if I1 ∩ I(D) = I2 ∩ I(D),
where I(D) := span(ADA).
If I(D) = A then the map I ∈ I(A) 7→ J := I ∩ D ∈ I(D) is a lattice
isomorphism from the lattice I(A) of closed ideals of A onto the lattice D. The
inverse of this map is given by J 7→ I := span(AJA).

6. Basic properties of Cuntz semigroups

Compare also with W-vN equivalence ??

Let A a C *-algebra. Recall that a - b (with more precise notation a -A b)
for a, b ∈ A is defined by the existence of sequences of elements dn , en ∈ A with
a = limn dn ben .
In particular, always a - a∗ a - aa∗ - a∗ .
The relation depends from A and passes in general not to C *-subalgebras: Let
a, b ∈ A+ non-zero with ab = 0 and a -A b define D ⊆ A as D := aAa + bAb then
a 6- b in D.
Notice that for A 6= 0, ?????
It follows that a, b ∈ (A ⊗ K)+ then a - b is equivalent to the existence of
c1 , c2 , . . . ∈ A ⊗ K with a = limn c∗n bcn . Moreover, if a 6= 0, the cn can be replaced
1/k 1/k
here by (b−δn )+ n cn (a−δn )+ n with suitable choices kn ∈ N, < δn < min(kak, kbk)
with kn → ∞ and δn → 0, depending on the norms kcn k, kak and kbk in an
“universal” way. (The explicit formula should be an exercise.) Thus, if a, b are in
the Pedersen ideal P (A) of A – which is by definition the minimal dense ideal of
A denoted by P (A) – then a, b ∈ P (A) and a ≈A b implies a ≈P (A) b. Recall also
that a ≈ b means a - b and b - a.
 6. BASIC PROPERTIES OF CUNTZ SEMIGROUPS 1087

Lemma A.6.1. The relation a - A b has following properties:

(i) The relation - is transitive:

Always a - a in C ∗ (a) ⊆ A itself. The relations a - A b and b - A c
imply a - A c.
The relation a ≈ b, i.e., a - b and b - a together, is an equivalence
relation on A, and it satisfies that a - b, a ≈ c and b ≈ d together imply
that c - d.
(ii) The relations ≈ and - are compatible with ⊕, e.g. a ⊕ c - b ⊕ d in
M2 (A) if a - b and c - d in A. If, moreover, b, d ∈ A are “orthogonal”
in the sense that bd∗ = 0 and b∗ d = 0, then a - b and c - d imply
a ⊕ c - (b + d) ⊕ 0 ≈ b ⊕ d in M2 (A).
(iii) The relation - is compatible with *-morphisms: If ψ : A → B is a *-
morphism, and a - A b in A, then ψ(a) - B ψ(b) in B.
(iv) The set ξ(b) of a ∈ A with a - b is closed in A and satisfies cξ(b)d ⊆ ξ(b)
for c, d ∈ M(A).
More precisely, ξ(b) is the closure in A of the set of all elements
cbd ∈ A with c, d ∈ M(A).
(v) It holds aa∗ ≈ a∗ a ≈ a∗ ≈ a, and g(a∗ a) - a for all continuous functions
g on Spec(a∗ a) with g(0) = 0.
If, in addition, g(t) > 0 for t > 0, then g(a∗ a) ≈ a.
(vi) If a, b ∈ A+ , then a - b, if and only if, for every ε > 0, there are
δ = δ(ε) > 0 and d = d(ε) ∈ A+ such that d∗ (b − δ)+ d = (a − ε)+ . In
particular, a - b if a ∼M vN c for some 0 ≤ c ≤ b.
If p and q are projections in A, then p - q, if and only if, there is a
projection r ∈ A with p ∼ r ≤ q.
(vii) If a ∈ A+ then a - b if and only if (a − ε)+ - b for all ε > 0.
(viii) If D is a hereditary C*-subalgebra of A, and a, b ∈ D ⊆ A, then a - D b
in D, if and only if a - A b in A.
Particular cases are:
Always a - D b for a ∈ D := b∗ Ab and b ∈ A.
x ⊕ 0n -B y ⊕ 0n in B := Mn+1 (A), if and only if, a - A y (i.e., in
A itself ).
If a, b ∈ A, then a - A b in A, if and only if, a - M(A) b in the
multiplier algebra M(A).
(ix) (x ⊕ y) ≈ (y ⊕ x) in M2 (A) and ((x ⊕ y) ⊕ z) ≈ (x ⊕ (y ⊕ z)) in M3 (A)
for all x, y, z ∈ A.
(x) (x + y) ⊕ 0 - x ⊕ y in M2 (A) for all x, y ∈ A.
(xi) If a, b ∈ A+ , πJ (a) - πJ (b) for J / A, then, there is c ∈ J+ with a - b ⊕ c.
In particular, there exists c ∈ J+ ∩ bAb such that (b ⊕ b) - (b ⊕ c) if
πJ (b) ⊕ πJ (b) - πJ (b).
(xii) Suppose that a ∈ A+ and that there is b ∈ Ann(a, A)+ such that a ∈
span(AbA).
1088 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

Then, for each closed ideal J / A, each δ > 0, each c ∈ J+ , and every
ε > 0 there is d = d(ε) ∈ J+ ∩ Ann((a − δ)+ , A) and n = n(δ, ε) ∈ N such
that
(a − δ)+ + (c − ε)+ - a ⊕ (d ⊗ 1n ) .

(xiii) Parts (xiii) and (xiv) should be separated

Suppose that A is an AW*-algebra, pn , q ∈ A projections with p1 ≤
p2 ≤ · · · and with pn - q for each n ∈ N. Then p - q for p := supn pn .
(xiv) If A is an AW*-algebra, and if a, b ∈ A+ with a - b, then pa - pb for the
support projections pa , pb ∈ A of a and b, and az - bz for each z ∈ A+ in
the center of A.
(xv) Suppose that an , bn ∈ A, a = lim an , b = lim bn . If there is a continuous
function ψ ∈ C[0, 1]+ with ψ(δ) > 0 for δ > 0, such that (a∗n an − δ)+ -
(b∗n bn − ψ(δ))+ for all n ∈ N and all rational δ > 0. Then a - b.
(xvi) Suppose that C ⊆ B ⊆ A are C*-subalgebras of A, and that S ⊆ C is a set
of positive contractions with the property that, for each c1 , c2 ∈ S and each
positive rational numbers r1 , r2 with property (c1 − r1 )+ - A (c2 − r2 )+
in A holds also (c1 − r1 )+ - B (c2 − r2 )+ in B.
If S ⊆ C ⊆ B satisfy the following condition (D), then for all a, b ∈ C,

a -A b if and only if a -B b.

(D) For each positive contraction a ∈ C+ and δ > 0, there exists c ∈ S

and rational r > 0 with (a − δ)+ - B (c − r)+ - B a.
Condition (D) is satisfied, if S is dense in the set of contractions in C+ .
(xvii) Let a, b ∈ A and B := Aω or B := Cb (X, A)/ C0 (X, A) with any locally
compact Hausdorff space X, and A ⊆ B the natural embedding. Then
a - A b, if and only if, a - B b .

Proof. The properties (i) and (iii) come straight form the definition.
(ii): a⊕c = limn (en ⊕gn )(b⊕d)(fn ⊕hn ) if a = limn en bfn and c = limn gn dhn .
If (moreover) b, d ∈ A satisfy bd∗ = 0 and b∗ d = 0, then (b + d) ⊕ 0 ≈ b ⊕ d
in M2 (A) by part (v), because (b + d)(b + d)∗ ⊕ 0 = [b, d][b, d]∗ and [b, d]∗ [b, d] =
(b ⊕ d)∗ (b ⊕ d) for the row matrix [b, d] ∈ M1,2 (A).
(iv): If a = limn fn ben then cad = limn (cfn )b(en d) . Thus, cad ∈ ξ(b) if
a ∈ ξ(b) for all c, d ∈ M(A).
Let a := lim an with an - b for each n ∈ N, and let ε > 0. Put δ := ε/2. There
are n = n(δ) ∈ N with ka − an k < δ and elements f, g ∈ A with kan − f bgk < δ .
Thus ka − f bgk < ε for suitable f, g ∈ A. It implies that a - b. Thus, a ∈ ξ(b) if
a ∈ ξ(b).
The closed set ξ(b) is the closure in A of the set of all elements cbd ∈ A with
c, d ∈ M(A), because cbd - b, i.e., cbd ∈ ξ(b), for all elements c, d ∈ M(A), and if
a ∈ ξ(b), i.e., if a - b, then the there exist sequences (cn ) (dn ) in A ⊆ M(A) such
that a = limn cn bdn .
 6. BASIC PROPERTIES OF CUNTZ SEMIGROUPS 1089

(v): The relations g(b) - A b for b ∈ A+ and continuous g : Spec(b) → R+

with g(0) = 0 can be seen in C ∗ (b) ∼
= C0 (Spec(b) \ {0}), using later part (iii). In
this way one can also see that g(b) ≈ b if g(t) > 0 for all 0 < t ∈ Spec(b).
In particular, b ≈ b2 , thus a(a∗ a)a∗ = (aa∗ )2 ≈ aa∗ . It implies aa∗ - a∗ a. If
we replace a by a∗ , then we get aa∗ ≈ aa∗ .
Hence b ≈ c if b, c ∈ A+ and b ∼MvN c.
The definition of - shows that a - b implies a∗ - b∗ . Thus, a ≈ b if and only
if a∗ ≈ b∗ .
We have a∗ a = limn cn adn with cn = a∗ and dn = (a∗ a)1/n . On the other
hand, a = limn en a∗ afn for en = a, fn := gn (a∗ a), where gn (t) = (t + 1/n)−1 t1/n .
It follows a ≈ a∗ a, and then a∗ ≈ (a∗ a)∗ = a∗ a.
(vi): If a, b ∈ A+ and a - b, then a ≈ a1/2 - b1/2 ≈ b by parts (i) and (v).
Thus, there are sequences cn , dn ∈ A with limn cn b1/2 dn = a1/2 . Let ε > 0. There
is n ∈ N with kcn b1/2 dn d∗n b1/2 c∗n − ak < ε/3.
Thus, ε/3 + kdn k2 cn b(cn )∗ ≥ a, and Lemma 2.1.9 gives the existence of a
contraction e ∈ A with f ∗ bf = e∗ b1 e = (a − ε/2)+ for b1 := kdn k2 cn b(cn )∗ and
f := kdn kc∗n e. It follows kf ∗ (b − δ)+ f − ak < ε for δ = ε/(4(kf k2 + 1)) > 0. Now
again Lemma 2.1.9 applies and gives the existence of g ∈ A with d∗ (b − δ)+ d =
g ∗ (f ∗ (b − δ)+ f )g = (a − ε)+ where d := f g.
Conversely, if there are δ = δ(ε) > 0 and d = dε ∈ A+ such that d∗ (b − δ)+ d =
(a − ε)+ for every ε > 0, then (a − ε)+ - (b − δ)+ . Since (b − δ)+ - b by part (v),
then (a − 1/n)+ - b for all n ∈ N. It implies a - b by part (iv).
By part (v), a ∼ c implies a ≈ c. If 0 ≤ c ≤ b, then there are contractions
dn ∈ A with d∗n bdn = (c − 1/n)+ by Lemma 2.1.9. Thus c ≈ a - b.
It follows, that the existence of r ≤ p = p∗ = p2 with r ∼ p implies p - q, if p
and q are projections in A.
If p - q then there are δ > 0 and d ∈ A with (1 − δ)d∗ qd = d∗ (q − δ)+ d =
(p − 1/2)+ = (1/2)p . It follows that z := (2(1 − δ))1/2 qd satisfies z ∗ z = p and
r := zz ∗ ≤ q.
(vii): It follows from parts (v) and (iv), because
(a − ε)+ - a and a = lim(a − (1/n))+ .

(viii): Use an approximate unit of D.

(ix); Part (viii) allows to use suitable (unitary) permutation matrices in
M2 (M(A)) respectively M3 (M(A)).
(x): d∗ (x ⊕ y)d = (x + y) ⊕ 0 for d ∈ M2 ⊆ M2 (M(A)) with matrix entries
d11 = d21 = 1, d12 = d22 = 0.
(xi): Let a, b ∈ A+ , πJ (a) - πJ (b) and ε > 0. Then Part (vi) shows that there
are δ = δ(ε) > 0 and d = d(ε) ∈ A, such that xε := d∗ (b − δ)+ d − (a − ε)+ ∈ J .
It follows (a − ε)+ ≤ b ⊕ |xε | . If we take here ε ∈ {1/n ; n ∈ N} and let c :=
1090 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

2−n (1+kx1/n k)−1 |x1/n | , then c ∈ J+ , kck ≤ 1, |x1/n | - c and (a−1/n)+ - b⊕c
P
n
for all n ∈ N. Thus a - (b ⊕ c) by (vii).
(xii): Let a ∈ A+ and b ∈ Ann(a, A)+ := {c ∈ A ; ca = 0 = ac} such that
a ∈ span(AbA). Let δ > 0 and let K denote the closed ideal, that is generated by
Ann((a − δ)+ , A). Then b ∈ Ann(a, A) ⊂ Ann((a − δ)+ , A). It follows that a ∈ K.
On the other hand, πK (a) is invertible in A/K, because K contains (1 − fδ (a))a,
i.e., Spec(πK (a)) ⊆ [δ, kak]. Thus K = A, which means that D := Ann((a−δ)+ , A)
is a full hereditary C *-subalgebra of A. Then J ∩ D is full in J for each J / A.
Thus, for every ε > 0 and c ∈ J+ , there are e1 , . . . en ∈ (J ∩D)+ and f1 , . . . , fn ∈
J such that (c − ε)+ = k fk∗ ek fk . Thus (c − ε)+ - d ⊗ 1n for d := k ek . Now
P P

(a − δ)+ + c - a ⊕ (d ⊗ 1n ) follows from Parts (ii), (vi) and (x).

Put (xiii) and (xiv) on extra place (xiii): There is a projection y ∈ qAq
such that y is central in qAq, y = yq is finite (or zero) and (q −y) is properly infinite
(or zero). There exist a central projection z of A such that zq = y. Notice that the
projection p := supn pn ∈ A+ exists, because A is an AW*-algebra. If a ∈ A is a
contraction, then the left and right support projections paa∗ := supn (aa∗ )1/n and
pa∗ a := supn (a∗ a)1/n of a are Murray–von-Neumann equivalent in A by the partial
isometry v ∈ A of the polar decomposition a = v(a∗ a)1/2 of a in A, cf. [64].
By Part (vi), there exist qn ≤ q such that pn ∼ qn for n ∈ N . It implies
zpn ∼ zqn ⊆ zq = y and (1 − z)pn ∼ (1 − z)qn ⊆ (1 − z)q = q − y.
If q 6= y then q − y is properly infinite, i.e., there are s, t ∈ A with s∗ s = t∗ t =
q − y and s∗ t = 0 and ss∗ + tt∗ ≤ q − y. Then there exist a sequence s1 , s2 , . . . ∈
(q − y)A(q − y) with s∗m sn = δm,n (q − y). Let wn ∈ A with (1 − z)pn = wn∗ wn and
wn (wn )∗ ≤ q − y. Now we define

X
a := s1 w1 + 2−n sn+1 wn+1 (1 − z)(pn+1 − pn ) .
n=1

Then kak ≤ 1, (1 − z)p = supn (a∗ a)1/n and aa∗ ≤ (q − y). Thus paa∗ =
supn (aa∗ )1/n ≤ (1 − z)q = q − y and (1 − z)p = pa∗ a ∼ paa∗ ≤ (1 − z)q .
The AW*-algebra yAy is finite. Let v0 a partial isometry with v0∗ v0 = zp1 and
v0 (v0 )∗ ≤ y = zq. By induction, we find partial isometries v2 , v3 , . . . ∈ A with
vn∗ vn = z(pn+1 − pn ), vn (vn )∗ ≤ zq = y and vn∗ vk = 0 for k < n. Indeed:
The partial isometry wn := v0 + · · · + vn−1 satisfies wn∗ wn = zpn and wn (wn )∗ ≤ y.
By Part (vi), there is a partial isometry x ∈ A with x∗ x = zpn+1 and xx∗ ≤ y,
because zpn+1 - y. Then xzpn x∗ ∼ wn (wn )∗ in the finite AW*-algebra yAy. We
find a unitary u ∈ yAy with uxzpn x∗ u∗ = (wn wn )∗ , because Murray–von-Neumann
equivalent projections in finite AW*-algebras are unitarily equivalent, cf. [64, chap.
6]. Let vn := uxz(pn+1 −pn ), then vn∗ wn = 0 and vn∗ vn = z(pn+1 −p) and vn vn∗ ≤ zq.
P −n
Now let b := 2 vn . Then kbk ≤ 1, bb∗ ≤ zq, zp = supn (b∗ b)1/n . Thus
zp = pb∗ b ∼ pbb∗ ≤ zq.
It follows p = pa∗ a + pb∗ b ∼ paa∗ + pbb∗ ≤ q .
 6. BASIC PROPERTIES OF CUNTZ SEMIGROUPS 1091

(xiv): If z ∈ A+ is in the center of A, and if (dn ), (en ) ⊆ A satisfy lim en bdn =

a, then lim en (zb)dn = za, i.e., a - b implies za - zb.
Let a, b ∈ A+ with a - b, then (1/n)pn ≤ a - b ≤ kbkq, p1 ≤ p2 ≤ · · ·
and p = supn pn for the support projections p := pa , pn := pan and q := pb of a,
an := (a − (1/n))+ and b respectively. In particular, pn - q for all n ∈ N. Thus,
Part (xiii) implies pa - pb .
(xv): Suppose that a = lim an , b = lim bn , and that ψ is a continuous function
ψ(δ) > 0 such that (a∗n an − δ)+ - (b∗n bn − ψ(δ))+ for all n ∈ N and all rational
δ > 0. Let ε > 0, and δ ∈ (0, ε/2) rational. There is n0 ∈ N with ka∗ a − a∗n an k < δ
and kb∗ b − b∗n bn k < ψ(δ)/2 for all n ≥ n0 . This implies a∗ a ≤ (a∗n an − δ)+ + 2δ
and b∗n bn ≤ b∗ b + ψ(δ)/2. By Lemma 2.1.9, we find d1 , d2 ∈ A with (a∗ a − ε)+ =
d∗1 (a∗n an − δ)+ d1 and (b∗n bn − ψ(δ))+ = d∗2 b∗ bd2 . It follows (a∗ a − ε)+ - b∗ b (for all
ε < 0), and (finally) a ≈ a∗ a - b∗ b ≈ b by Parts (i), (vii) and (v).
(xvi): Let a1 , b1 ∈ C \ {0} with a1 - A b1 , then a ≈ C a1 , b ≈ C b1 and a - A b
for the contractions a := ka1 k−2 a∗1 a1 ∈ C+ and b := kb1 k−2 b∗1 b1 ∈ C+ . By Part (vi)
there exist 0 < δn ≤ εn < 1/4 with limn εn = 0 such that (a − εn )+ - A (b − δn )+ .
For each n ∈ N there are cn , dn ∈ C+ and rational rn , sn > 0 with (a − 2εn )+ - B
(cn − rn )+ - B (a − εn )+ and (b − δn )+ - B (dn − sn )+ - B b.
It follows (cn − rn )+ - A (dn − sn )+ . Then (cn − rn )+ - B (dn − sn )+ by
assumptions on S. It yields (a − 2εn )+ - B b for all n ∈ N, by Part (i). Thus,
a - B b and a1 - B b1 by Parts (vii) and (i).
Condition (D) is satisfied, if S is dense in the set of contractions in C+ , because
kc − ak < δ/3 implies the existence of contractions d1 , d2 ∈ C and rational r ∈
(δ/3, δ/2] with (a − δ)+ = d∗1 (c − r)+ d1 and (c − r)+ = d∗2 ad2 , by Lemma 2.1.9.
proofs complete ??
(xvii): Let a, b ∈ A and B := Aω or B := Cb (X, A)/ C0 (X, A), and consider
the natural embedding a 7→ πω (a, a, . . .) ∈ Aω , respectively a 7→ fa + C0 (X, A)
for fa (x) := a. Then a - A b implies a -B b by Part (iii). We use a ≈ a∗ a for
the opposite direction, cf. Part (v). Let a, b ∈ A+ with a - B b and δ > 0, then
Part (v) says that there is d = (d1 , d2 , . . .) ∈ `∞ (A) with limn d∗n bdn = (a − δ)+
(respectively d ∈ Cb (X, A) with g ∈ C0 (X, A) for g(x) := d(x)∗ bd(x) − (a − δ)+ ).
Thus (a − δ)+ - A b in both cases (since X is not compact). It follows a - A b by
Part (vii). 

We call a C *-subalgebra B of A --preserving if a, b ∈ B and a - A b imply

that a - B b.

Lemma A.6.2. Let C ⊆ A a separable C*-subalgebra and B1 ⊆ B2 ⊆ · · · ⊆ A a

sequence of --preserving C*-subalgebras of a C*-algebra A.

(o) Each hereditary C*-subalgebra D ⊆ A is --preserving.

S
(i) The closure B of n Bn is --preserving.
1092 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

(ii) There exists a separable --preserving C*-subalgebra B of A with C ⊆ B ⊆

A.
(iii) Consider a C*-algebra E naturally as a C*-sub-algebra of A := Eω , (re-
spectively of A := Cb (X, E)/ C0 (X, E) for a locally compact Hausdorff
space X). Then E is a --preserving C*-subalgebra of A.

Proof. (o): See Lemma A.6.1(viii).

(i): We prove a stronger result: Suppose that a - A b and a, b ∈ Bn imply
S
a - Bn+1 b. Then, obviously, the relation a -A b for a, b ∈ n Bn implies a - B b.
S
Let C := B and let S ⊆ C+ denote the set of positive contractions in n Bn . Then
S, C, B and A satisfy the assumptions of Lemma A.6.1(xvi). Thus, for all a, b ∈ B,
a - A b if and only if a - B b .
(ii): Since C is separable, there exists a countable dense subset S in the set of
positive contractions in C+ .
If c, d ∈ S and r, t ∈ [0, 1] are rational with (c − r)+ - A (d − t)+ , then there
is a sequence d1 , d2 , . . . ∈ A with (c − r)+ = limn d∗n (d − t)+ dn .
The set S × S × ([0, 1] ∩ Q) × ([0, 1] ∩ Q) is at most countable. It follows, that
there is a separable C *-subalgebra C ⊆ B1 ⊆ A such that (c − r)+ - A (d − t)+
implies (c − r)+ - B1 (d − t)+ for c, d ∈ S and r, t ∈ [0, 1]. By Lemma A.6.1(xvi),
for each a, b ∈ C, a - A b is equivalent to a - B1 b.
We can repeat this argument for B1 in place of C, and get a separable C *-
subalgebra B1 ⊆ B2 ⊆ A such that, for each a, b ∈ B1 , a - A b is equivalent to
a - B2 b. Going on this way, we can find a sequence B1 , B2 , . . . of separable C *-
subalgebras Bn ⊆ Bn+1 ⊆ A such that B0 := C ⊆ B1 and, for each a, b ∈ Bn ,
S
a - A b is equivalent to a - Bn+1 b . Then the closure B of n Bn is a --preserving
separable C *-subalgebra of A that contains C.
(iii): The cases E ⊆ Eω and E ⊆ Cb (X, E)/ C0 (X, E) follow from Lemma
A.6.1(xvii). 

Remark A.6.3. Suppose that B ⊆ A is a C *-subalgebra. Let a ∈ A+ , b ∈ B+

and let JA (b) := span(AbA) (respectively JB (b) := span(BbB) ) the closed ideal of
A (respectively of B) generated by b.
Notation I(a) was in conflict with absorption ideal

IA (a) = {b ∈ A ; [b] + [a] ≤ [a]}

The following observation (o) is trivial, and the implications (o)⇒(i)⇒(ii) are
straight-forward:
(o) The smallest number t ≥ 0 with (a − t)+ ∈ JA (b) is given by

dist(a, JA (b)) = ka + JA (b)k = kπJA (b) (a)k .

 6. BASIC PROPERTIES OF CUNTZ SEMIGROUPS 1093

(i) For each δ > 0 there exists n = n(δ) ∈ N such that (a − (t + δ))+ - b ⊗ 1n ,
if and only if, t ≥ dist(a, JA (b)).
(ii) dist(a, JA (b)) = dist(a, JB (b)) for all b ∈ B+ if B ⊗ K ⊆ A ⊗ K is --
preserving.

Let B ⊆ A a C *-sub-algebra of A and a, b ∈ A positive contractions, and let

d1 , d2 ∈ A contractions. We define

ρ(a, b; d1 , d2 ) := ka2 − d∗1 a2 d1 k + kb2 − d∗2 b2 d2 k + kd∗1 abd2 k ,

and denote by ρ(a, b; B) the infimum of the non-negative numbers

{ρ(a, b; d1 , d2 ) ; d1 , d2 ∈ B, kd1 k ≤ 1, kd2 k ≤ 1, } .

The following Lemma establishes the continuity of the property s.p.i. in the category
of C *-algebras with respect to inductive limits.

Lemma A.6.4. Let C ⊆ A and B1 ⊆ B2 ⊆ . . . ⊆ A separable C*-subalgebras.

S
Let B denote the closure n Bn . Then:

(i) For contractions a, b, a0 , b0 ∈ A+ holds

| ρ(a0 , b0 ; A) − ρ(a, b; A) | ≤ 5ka0 − ak + 5kb0 − bk

and ρ(a, b; C) ≥ ρ(a, b; A) if a, b ∈ C.

(ii) ρ(a, b; A) = ρ(a, b; Aω ) for all contractions a, b ∈ A+ .
(iii) If ρ(a, b; Bn+1 ) = ρ(a, b; A) for all positive contractions a, b ∈ Bn for n =
1, 2, . . . , then ρ(a, b; B) = ρ(a, b; A) for all positive contractions a, b ∈ B.
(iv) There exists a separable C*-subalgebra B of A with C ⊆ B ⊆ A, such
that ρ(a, b; B) = ρ(a, b; A) for all contractions a, b ∈ B+ .

Proof. (i): The inequality ρ(a, b; C) ≥ ρ(a, b; A) is immediate from the defi-
nition. Straight calculation shows

| ρ(a0 , b0 ; d1 , d2 ) − ρ(a, b; d1 , d2 ) | ≤ 5ka0 − ak + 5kb0 − bk .

(ii): By Part (i), ρ(a, b; A) ≥ ρ(a, b; Aω ). Here we have identified a ∈ A with

πω (a, a, . . .) = ∆(a) + cω (A) in Aω .
Let a, b ∈ A+ contractions and δ > 0 . There are sequences of contraction
e1 , e2 , . . . ; f1 , f2 , . . . ∈ A such that, for e := πω (e1 , e2 , . . .) and f := πω (f1 , f2 , . . .),

ρ(a, b; Aω ) + δ > ρ(a, b; e, f ) = lim ρ(a, b; en , fn ) ≥ inf ρ(a, b; en , fn ) .

ω n

Clearly, inf n ρ(a, b; en , fn ) ≥ ρ(a, b; A) .

(iii): By assumptions and Part (i), we have ρ(a, b; B) = ρ(a, b; A) for all pos-
S
itive contractions a, b ∈ n Bn . It implies ρ(a, b; B) = ρ(a, b; A) for all positive
contractions a, b ∈ B, because (a, b) 7→ ρ(a, b; B) − ρ(a, b; A) is continuous.
(iv): Let X ⊆ C+ a countable subset of contractions, that is dense in the set of
all positive contractions in C. For each a, b ∈ X there are sequences of contractions
1094 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

e1 , e2 , . . . ; f1 , f2 , . . . ∈ A, such that limn ρ(a, b; en , fn ) = ρ(a, b; A). Thus, there is a

countable subset X ⊆ Y ⊆ A of the contractions in A, such that

inf{ ρ(a, b; d1 , d2 ) ; d1 , d2 ∈ Y } = ρ(a, b; A) ∀ a, b ∈ X .

Let C ⊆ B1 ⊆ A denote the separable C *-subalgebra that is generated by Y . Then

ρ(a, b; B1 ) = ρ(a, b; A) for all a, b ∈ X. This is true also for all positive contractions
a, b ∈ C, because (a, b) 7→ ρ(a, b; B) − ρ(a, b; A) is continuous.
Now we can repeat the argument with B1 in place of C, and get a separable
C *-subalgebra B1 ⊆ B2 ⊆ A with ρ(a, b; B2 ) = ρ(a, b; A) for all a, b ∈ B1 . Thus,
we find a sequence B1 ⊆ B2 ⊆ · · · ⊆ A that satisfies the assumptions of Part (iii).
S
Then the closure B of n Bn is a separable C *-subalgebra of A with C ⊆ B and
ρ(a, b; B) = ρ(a, b; A) for all a, b ∈ B. 

Lemma A.6.5. Let M a W*-algebra, a ∈ M+ with kak = 1, let Pb ∈ M denote

P ∗
the support projection of b ∈ M+ , and let d1 , . . . , dm ∈ M with dj dj ≤ 1.
Suppose that the support projections of z(a − t)+ and of z(1 − a − t)+ (a − s)+
are either zero or infinite for all t, s ∈ (0, 1) and for all central projections z ∈ M .
Then:

(i) The support projections of (a − t)+ and (1 − a − t)+ · (a − s)+ are properly
infinite or zero for every t, s ∈ [0, 1].
(ii) For each δ > 0, there exists mutually orthogonal projections p1 , . . . , pn ∈
{a}0 ∩ M and 0 ≤ α1 ≤ · · · ≤ αn ≤ 1 such that pk is properly infinite in
P P
M for k = 1, . . . , n, k pk = Pa and (a − δ)+ ≤ k αk pk ≤ a.
(iii) For each δ > 0 and m ∈ N there exists partial isometries v1 , . . . vm ∈
Pa M Pa with vj∗ vk = δjk Pa and k vj a − avj k < δ for j, k = 1, . . . , m.
(iv) For each ε > 0 there exist self-adjoint hk ∈ M with khk k ≤ π (1 ≤ k ≤ 4)
such that h2 = −h1 , h4 = −h3 and
next ok ?? ??

X 4
X
d∗j adj ) − ε


( +
≤ exp(ihk )a exp(−ihk )
j k=1

Proof. check proof of (i) again (i): Suppose that the support projec-
tions of z(a − t)+ and of z(1 − a − t)+ (a − s)+ are either zero or infinite for all
t ∈ (0, 1) and for all central projections z ∈ M .
Let b ∈ M+ such that the support projection Pb ∈ M of b is infinite. Then
there exists a central projection z ∈ M such that zPb is finite (or is zero) and
that (1 − z)Pb is properly infinite (or is zero). The support projection of zb is
given by zPb . Thus, if the support projection of zb is not finite, then zPb = 0 and
Pb = (1 − z)Pb is properly infinite.
Thus, all support projections P(a−t)+ and and P(1−a−t)+ (a−s)+ of the elements
(a − t)+ respectively (1 − a − t)+ (a − s)+ are properly infinite ore zero in M .
 6. BASIC PROPERTIES OF CUNTZ SEMIGROUPS 1095

If follows that (a − t)+ and (1 − a − t)+ (a − s)+ are zero or are properly infinite
elements of M for each s, t ∈ [0, 1], by Lemma 2.5.3(xv).
(ii): By Lemma A.6.2(ii), there exists a separable C *-subalgebra B ⊆ M with
a ∈ B, such that c - B d if and only if c - M d for all c, d ∈ B . Let M0 ⊆ M the
σ(M, M∗ )-closure (i.e., ultra-weak closure) of B in M .
Then M0 is a direct sum M0 = ∼ Q Nα of W*-algebras Nα with separable
α
predual, and the elements (a − t)+ and (1 − a − t)+ (a − s)+ are properly infinite or
Q
zero in M0 for all s, t ∈ [0, 1]. It follows, that the image aα of a = {aα }α ∈ α Nα
have the properties that the elements (aα − t)+ and (1 − aα − t)+ (aα − s)+ are
properly infinite or zero.
Fix some of the α. Let b := aα ∈ Nα =: N . Then 0 ≤ b ≤ 1 and the elements
(b−t)+ and (t−b)+ (b−s)+ are properly infinite for all s, t ≥ 0. Let E := Pb ∈ N the
support projection of b in N . By 2.5.3(xv), the support projections P (t) := P(b−t)+
of (b − t)+ and the support projections Q(t) := P(t−b)+ b of (t − b)+ b are properly
infinite projections in N (but, clearly, they can’t be properly infinite in {b}00 ⊆ N ,
and they are not necessarily properly infinite in {b}0 ∩N ). It holds Q(t), P (t) ∈ {b}00 ,
Q(t)P (t) = 0, Q(s) ≤ Q(t), P (s) ≥ P (t) for s < t Q(t) + P (t) ≤ E. The projection
R(t) := E − Q(t) − P (t) satisfies bR(t) = tR(t) for each t ∈ [0, 1]. In particular,
R(t)R(s) = 0 for s 6= t. Since N has separable pre-dual, there is a faithful normal
state ρ on N . It follows ρ(R(t)) > 0 if R(t) 6= 0, and that {t ∈ [0, 1] ; ρ(R(t)) ≥ 1/n}
contains at most n points. It implies, that the set S = {t ∈ [0, 1] ; R(t) 6= 0} is
countable. We have 0 6∈ S, because P (0) = E.
If s < t and R(t) = 0, then P (s) − P (t) = P (s)Q(t). But P (s)Q(t) is the
support projection of (t − b)+ (b − s)+ , and (t − b)+ (b − s)+ is zero or properly
infinite properly infinite. Now, Lemma 2.5.3(xv) implies that the support projection
P (s) − P (t) = P (s)Q(t) of (t − b)+ (b − s)+ is properly infinite if P (s) 6= P (t) and
t 6∈ S.
Since b = sup+ tP (t) (in the lattice ({b}00 )+ ), we can approximate b from below
P
in norm by tk (P (tk ) − P (tk+1 )) = supk tk P (tk ) with 0 = t1 < t2 < . . . < tn < 1,
tk+1 6∈ S and where tk ∈ ((k − 1)/n, k/n].
It follows that each Rk = P (tk ) − P (tk+1 ) (with P (tn+1 ) := 0 = P (1)) is
a properly infinite projection in Nα or is zero, Rk ∈ {b}00+ , b − 1/n ≤
P
(k −
P
1)/nRk tk Rk ≤ b.
If we do this for each summand Nα of M0 , then we get projections pk = (Rkα )α ∈
M0 , k = 1, . . . , n, such that the pk commute with a, are mutually orthogonal, are
P P
properly infinite or zero, have sum pk = Pa , and a − (2/n) ≤ (k − 1)/npk ≤ a.
(iii): Let δ > 0 and m ∈ N. Let n = [2/δ] + 1. We find pk ∈ {a} ∩ M ,
k = 1, . . . , n, with the properties in (ii). It follows, that, for each j ∈ {1, . . . , n},

X
kpj T pj b − bpj T pj k ≤ kb − (k − 1)/n)pk kkpj T pj k ≤ (2/n)kpj T pj k .
k
1096 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

Since the projections pk are properly infinite, we find z`,j ∈ pj M pj , ` = 1, . . . , m

∗
with z`,j zk,j = δ`,k pj .
Let v` = j z`,j . Then v1 , . . . vm ∈ Pa M Pa and v`∗ vk = δ`,k Pa . Since the pj
P
P
commute with a and v` , we have that v` a − a v` = j pj (z`,j a − az`,j )pj . Thus,
the above estimate gives k v` a − a v` k < δ for `, k = 1, . . . , m,
end above? more?
P ∗
(vi): Given d1 , . . . , dm ∈ M with dj dj ≤ 1, and ε > 0, we let δ := ε/(1 +
√
m). We find the v1 , . . . , vm with the properties of (iii), and define a contraction d ∈
Pm
M by `=1 v` d` . Then d∗ ad − ` d∗` ad` = CBC ∗ for the row C = [d1 , . . . , dm ] ∈
P

M1,m (M ) and the matrix B = [b`,k ] ∈ Mm (M ) with entries

b`,k = δ`,k a − v`∗ avk = v`∗ (vk a − avk ) .

Thus, B = V ∗ D for the rows V := [v1 , . . . , vm ] and D = [v1 a−av1 , . . . , vm a−avm ] in

√
M1,m (M ). Since kCk ≤ 1, V ∗ V ≤ Pa ⊗ 1n , and kDk ≤ δ m, it follows kCBC ∗ k ≤
kBk ≤ kDk < ε. Hence,
X
d∗` ad` − ε ≤ d∗ ad .
`

Let e := (d + d )/2, f := (d − d∗ )/2i, h1 := arcsin(e), h2 := −h1 , h3 := arcsin(f )

∗

and h4 := −h3 , then kek ≤ 1, kf k ≤ 1, and khk k ≤ π for k = 1, 2, 3, 4.

d∗ ad ≤ d∗ ad + dad∗ = 2(eae + f af ) ≤
2(eae + f af ) + 2(cos(h1 )a cos(h1 ) + cos(h3 )a cos(h3 )) =
X
4 exp(ihk )a exp(−ihk ) .
k

Lemma A.6.6. For every a ∈ A and every unitary u ∈ M(A), the element
u au is in the norm-closed convex hull of the set {exp(ih)a exp(−ih) ; h∗ = h ∈
∗

A, khk < π} .
In particular, if C ⊆ A+ is a hereditary closed convex sub-cone of A+ , then

K := {a ∈ A+ ; exp(ih)a exp(−ih) ∈ C for all h∗ = h ∈ A with khk < π}

is the positive part I+ := K of a closed ideal I of A.

The closed ideal I is the largest ideal of A with I+ := I ∩ A+ ⊆ C.

Proof. Recall that M(A) ⊆ A∗∗ . The set {exp(ih) ; h∗ = h ∈ A, khk < π} is
*-ultra-strongly dense in the unitaries of A∗∗ , because exp : A∗∗ → A∗∗ is *-ultra-
strongly continuous on bounded parts, and because every unitary V of a W*-algebra
is of the form V = exp(ih) for some self-adjoint h ∈ A∗∗ with khk ≤ π.
Use Hahn-Banach separation from the closed convex hull of

{exp(ih)a exp(−ih) ; h∗ = h ∈ A, khk < π} .

 6. BASIC PROPERTIES OF CUNTZ SEMIGROUPS 1097

If exp(ih)a exp(−ih) ∈ C for all h∗ = h ∈ A with khk < π and if C is closed and
convex, then u∗ au ∈ C for all unitaries u ∈ M(A), i.e.,
\
a ∈ K := u∗ Cu .
u∈U (M(A))

If C is a hereditary closed convex sub-cone of A+ , then K := u∈U (M(A)) u∗ Cu

T

is a hereditary closed convex sub-cone of A+ that is invariant under “inner” auto-

morphisms Ad(u) of A.
Thus, the hereditary C *-subalgebra I := KAK of A is a closed ideal of A with
I+ = K.
If J / A is a closed ideal with J+ ⊆ C, then u∗ au ∈ C for all a ∈ J+ and
u = exp(−ih) with h∗ = h ∈ A. Thus, J ⊆ I. 

Lemma A.6.7. Suppose that A is a C*-algebra with the property that every
additive lower semi-continuous trace τ : A+ → [0, ∞] takes only the values 0 and
∞.
Let Sin denote the set of approximately inner completely positive contractions,
and let Su denote the point-norm closed convex hull of the (exponential) inner
automorphisms
a 7→ exp(ih)a exp(−ih)
with h∗ = h of norm < π.
Then, for each contraction a ∈ A+ , V ∈ Sin and ε > 0, there exists T ∈ Su
with
V (a) − ε ≤ 4T (a) .

Proof. to be filled in ?? 

Lemma A.6.8. Let a ∈ A+ , and suppose that there is m ∈ N such that, for every
contraction b ∈ A+ in the (algebraic) ideal generated by {a} there are d1 , . . . , dm ∈
Pm
A (depending on b) with k=1 d∗k adk = (b − 1/2)+ .
We denote by m(a) the smallest m ∈ N with this property.
Let I(a) denote the closed ideal of A generated by a ∈ A+ . Then:

(i) For every c ∈ I(a)+ and ε > 0 there are δ > 0 and d1 , . . . , dm(a) ∈ aAc
with k d∗k (a − δ)+ dk = (c − ε)+ and k k d∗k dk k ≤ 2δ −1 kck.
P P

check above estimate

(ii) If A1 , A2 , . . . is a sequence of C*-algebras, such that there is µ ∈ N with
m(a) ≤ µ for each n ∈ N and a ∈ (An )+ then m(b) ≤ µ for every positive
Q
b ∈ n An .
(iii) If m(a) ≤ m0 for all a ∈ A+ , and if J is a closed ideal of A, then
m(b) ≤ m0 for every positive b ∈ (A/J)+ .
(iv) Suppose that, for each b ∈ (Aω )+ and ε > 0, there exists m = m(b, ε) ∈ N
such that (b − ε)+ ⊗ 12m - b ⊗ 1m .
Then there exists n ∈ N such that a ⊗ 12n - a ⊗ 1n for all a ∈ A+ .
1098 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

Proof. (i): Let c ∈ I(a)+ and ε > 0 . There are n ∈ N and f1 , . . . , fn ∈ A

with kc − k fk∗ afk k < ε/2 .
P

Question:
d∗k (a − δ)+ dk = (c − ε)+ and
P
Are there δ > 0 and d1 , . . . , dm ∈ aAc with k
k dk dk k ≤ 2δ −1 kck?
P ∗

(If it does not work in A∗∗ then not in A!)

check above estimates
to be filled in ?? 

Question A.6.9. If 0 ≤ a ≤ b and kbk ≤ 1, then, for every ε > 0 there is a

contraction d ∈ C ∗ (a, b) ⊆ A with d∗ bd = (a − ε)+ and kdb − bdk ≤ f (kab − bak, ε),
where

f (t, ε) := sup{µ(a, b; ε) ; a, b ∈ L(`2 )+ , 0 ≤ a ≤ b ≤ 1 , kab − bak ≤ t }

for t > 0 and

µ(a, b; ε) := inf{kdb − bdk ; d ∈ C ∗ (a, b), kdk ≤ 1, d∗ bd = (a − ε)+ } .

Is limt→0 f (t, ε) = 0 for each (fixed) ε > 0?

That is “in the limit” (i.e., in L(`2 )ω ) the case of commuting 0 ≤ a ≤ b and
1/2
d := lim (b + δ)−1/2 (a − ε)+ .
δ→0

The question has to do with the study of images in Q(X, B)|ω of the fibers
F (X; A, B)|ω of the unital C(γX)-algebra

F (X ; A, B) := (A0 ∩ Q(X, B))/ Ann(A, Q(X, B))

for separable C *-subalgebra A ⊆ Q(X, B).

Question A.6.10. Is there a universal continuous function ρ : R+ → R+ with

ρ(0) = 0, such that ρ is increasing and has the following property (*)?
(*) If a, b ∈ A+ , 0 ≤ γ < ε satisfy b ≤ a + γ, then there is a contraction
d := d(a, b; γ, ε) ∈ A with d∗ ad = (b − ε)+ and

kda − adk ≤ ρ( sup ka(b − δ)+ − (b − δ)+ ak) .

γ≤δ≤ε

What about using ultrapowers here?

7. Basics on Quasi-traces

Is only collection/list of topics now

Definition A.7.1. To be defined:

Dimension function d, dimension function
Rank function = subadditive local rank function, local rank function r, quasi-
trace τ , 2-quasi-trace bounded, unbounded, l.s.c. traces = additive q-traces, ...
 7. BASICS ON QUASI-TRACES 1099

For every compact Hausdorff space X of dimension ≥ 2 there exists a compact

Hausdorff space Y of dimension ≤ Dim(X) + 1, a continuous injective map ϕ : X →
Y (with adjoint epimorphism ϕ b : C(Y ) → C(X) given by ϕ(f
b )(x) := f (ϕ(x)) for
x ∈ X) and a unital extension

0 → K ⊗ c0 → A → C(Y ) → 0

(– with fixed canonical epimorphism π : A → C(Y ) –) such that for each quasi-
measure µ on X (in the sense of Aarnes [3]) with µ(X) = 1 the composition
τµ := νµ ◦ ϕb ◦ π is a unital quasi-trace on A+ , where νµ : C(X)+ → [0, ∞) is the
local quasi-trace defined by the quasi-measure µ. It is then easy to see that τµ can’t
be 2-sub-additive on the positive part A+ of the type-I C *-algebra A+ if νµ is not
additive on C(X)+ , because otherwise the (again) 2-sub-additive class map [τµ ] on
C(X)+ coincides with νµ , but the quasi-state νµ on C(X)+ can’t be 2-sub-additive
if the quasi-measure µ is not a (then finite) Borel measure on X.
Recall here that a quasi-trace τ is 2-sub-additive if it has the property τ (a+b) ≤
2(τ (a) + τ (b)) for a, b ∈ A+ . This property is equivalent to both of τ (a + b)1/2 ≤
τ (a)1/2 + τ (b)1/2 (U. Haagerup) and the existence of a quasi-trace τ2 on M2 (A)+
with τ2 (a ⊗ p11 ) = τ (a),
Aarnes has defined in [3] a quasi-state on C([0, 1]2 ) that is not additive, there-
fore it can not be 2-sub-additive. Thus, there are bounded quasi-traces that are not
2-quasi-traces on unital type-I C*-algebras.
Haagerup [342] (cf. also [348]) has shown that every bounded 2-quasi-trace on
an exact unital C *-algebra is additive (i.e., is a bounded trace). One can technically
improve the arguments of Haagerup [342] and Blackadar/Handelman [79] to get
this result also for all (not necessarily bounded !) lower semi-continuous 2-quasi-
traces τ : A+ → [0, ∞].

Remark A.7.2. The lower semi-continuous dimension functions d : M∞ (A) →

S
[0, ∞], where M∞ (A) := n Mn (A), are in one-to-one correspondence to lower
semi-continuous 2-quasi-traces τ : A+ → [0, ∞] by :
Z ∞
τd (a) := lim d((a − t)+ ) dt
ε&0 ε

and dτ (b) for b ∈ Mn (A) and given τ – extended to τn : Mn (A)+ → [0, +∞], by

dτ (b) := lim sup τn ((b∗ b)1/m ) .

m

Check last definition!

Is here ((b∗ b)1/m − 1/m)+ better??
The local quasi-trace τ (respectively rank function d ) is on A+ lower semi-
continuous, if and only if, for every a ∈ A+ : τ (a) = supδ>0 τ ((a − δ)+ )
( respectively d(a) = supδ>0 d((a − δ)+ ) ). All on A1+ := {a ∈ A+ ; kak ≤ 1}
bounded local quasi-traces are lower semi-continuous.
An l.s.c. quasi-trace τ : A+ → [0, ∞] is a 2-quasi-trace, if and only if, τ (a + b) ≤
2(τ (a) + τ (b)) for all a, b ∈ A+ , if and only if, τ (a + b)1/2 ≤ τ (a)1/2 + τ (b)1/2 for all
1100 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

a, b ∈ A+ . It is equivalent to the original definition of J. Cuntz that proposes the

existence of a quasi-trace τ2 on M2 (A)+ with τ (a) = τ2 (diag(a, 0)) for all a ∈ A+ .
The dimension functions d : M∞ (A) → [0, ∞] are in one-to-one correspondence
to the monotone additive maps λ : CS(A ⊗ K) → R+ ∪ {+∞}.
Need definitions of Cu(A), CS(A) and M∞ (A) given in Chp. 2 ! Or move them
to here? ??? Or only remind them here? ???
If d : M∞ (A) → [0, ∞] is a dimension function (respectively a rank func-
tion r : A → [0, ∞], respectively τ : A+ → [0, ∞] a quasi-trace) then d∗ (b) :=
supε>0 d((b∗ b − ε)+ ) for b ∈ M∞ (A) (respectively r∗ (b) := supε>0 r((b∗ b − ε)+ ) for
b ∈ A, τ∗ (a) := supε>0 τ ((a − ε)+ ) for a ∈ A+ ) is a lower semi-continuous dimen-
sion function on M∞ (A) (respectively, lower semi-continuous rank function, lower
semi-continuous quasi-trace). r∗ is sub-additive on A if r is sub-additive. (τ∗ is a
2-quasi-trace if τ is a 2-quasi-trace.
It holds τ∗ = τdτ and d∗ = dτd .
Let a ∈ A+ and denote by S the semigroup of x ∈ CS(A⊗K) with the property
that there is n ∈ N with x ≤ n[a] = [a ⊗ 1n ], i.e., b - a ⊗ 1n for [b] = x.
????
then (S, ≤, [a]) is a scaled abelian semigroup (with the induced preorder ≤ and
order unit [a]). Every monotone and additive map λ : S → [0, ∞] extends to a
monotone and additive map λe : CS(A ⊗ K) → [0, ∞] by λe (x) := λ(x) for x ∈ S
and by λe (x) := +∞ for x 6∈ S.

Proof. If x ≤ y and λe (y) < ∞, then x, y ∈ S, in case that λe (x) ≤ λe (y). If

λe (x + y) < ∞ then x + y ∈ S. Since 0 ≤ y and 0 ≤ x in CS(A ⊗ K) it follows that
x = 0 + x ≤ x + y ≤ n[a].
and ?????
If λe (x) + λe (y) < ∞, then λe (x) < ∞ and λe (y) < ∞, i.e., x, y ∈ S. 

Proof. to be filled in ?? 

Corollary A.7.3. Let A denote a C*-algebra, and a ∈ A+ with the property

that d(a) = 0 for every lower semi-continuous dimension function d on M∞ (A) :=
S
n Mn (A) with d(a) < ∞, then, for every ε > 0, there is n(a, ε) ∈ N (depending
on ε) such that (a − ε)+ ⊗ 12n - a ⊗ 1n in M2n (A) for all n ≥ n(a, ε).

Proof. Let I denote the algebraic ideal of A generated by a. It contains

the minimal dense ideal (Pedersen ideal) of the hereditary C *-subalgebra aAa, in
particular, (a − ε)+ ∈ I for all ε > 0. Consider the sub-semigroup S of classes
[c] ∈ CS(aAa) with the property that there is n ∈ N with [c] ≤ n[a], i.e., c - a ⊗ 1n .
Then (S, ≤, [a]) is a “scaled” semigroup with the preorder induced from CS(aAa).
Since every monotone additive map λ : S → R+ extends to monotone additive
map λe : CS(A) → [0, ∞], and D(x) := λe ([x]) is a dimension function on M∞ (a),
 7. BASICS ON QUASI-TRACES 1101

and since D∗ (x) := supε>0 D(((x∗ x)1/2 −ε)+ ) is a lower semi-continuous dimension
function on A, we have D∗ (a) = 0 . Thus, λ((a − ε)+ ) = 0 for all ε > 0.
By A.13.10, for every ε > 0, there is n0 = n0 (ε) ∈ N with 2n[(a − ε)+ ] ≤ n[a]
in CS(aAa) for all n ≥ n0 , i.e., (a − ε)+ ⊗ 12n - a ⊗ 1n for all n ≥ n0 . 

Lemma A.7.4. Suppose that τ : A+ → [0, ∞] is a lower semi-continuous quasi-

trace.

(i) If τ is a 2-quasi-trace, and if e ∈ A+ satisfies τ (e) < ∞ and kek = 1, then

e + → R+ with ρ(d ⊕ 0) = τ (d) for
there is a bounded quasi-trace ρ : M2 (D)
d ∈ D+ , where D := {d ∈ A ; ed = de = d}.
(ii) The l.s.c. quasi-trace τ is additive on A+ , if τ |D+ is sub-additive for
every hereditary C*-subalgebra D ⊆ A with the property that there is
e ∈ A+ with τ (e) < ∞, kek = 1 and de = ed = d for all d ∈ D.

Proof. (i): We have τ (d) ≤ kdkτ (e) for all d ∈ D. By [79, cor.II.2.5], the
bounded 2-quasi-trace τ 0 := τ |D+ on D extends to a bounded 2-quasi-trace τ 00 |D
e+
on the unitization D
e of D with

τ 00 (1) = sup{τ (d) ; d ∈ D+ , kdk ≤ 1} ≤ τ (e) .

The proof of [79, cor.II.2.5] is based on the technically engaged theorem [79,
thm.I.4.1]. A detailed and elementary proof of Part (i) goes as follows:
It follows that f := 2(e − 1/2)+ satisfies df = f d = d for all d ∈ D and
f 1/n ≤ 2e for all n ∈ N. In particular, f is a strictly positive contraction in the
center of the C *-subalgebra C ∗ (D, f ) of A, because h(f )d = h(1)d for h ∈ C0 (0, 1]
and d, which implies C ∗ (D, f ) = D + C ∗ (f ) and DC ∗ (f ) = D.
The restriction of τ to C ∗ (D, f )+ is bounded, because, for c ∈ C ∗ (D, f )+ ,

τ (c) ≤ kck sup τ (f 1/n ) ≤ 2kckτ (e) .

n

The element f (1 − f ) is orthogonal to D and is a strictly positive element of an

e ∼
ideal I of C ∗ (D, f ), with D = C ∗ (D, f )/I . The isomorphism is defined by the
epimorphism φ : C ∗ (D, f ) → De with kernel C ∗ (f (1 − f )), given by φ(f ) = 1 and
φ(d) = d.
Then we have M2 (D) = {b ∈ M2 (A) ; b(e ⊕ e) = (e ⊕ e)b = b } . Thus
M2 (C ∗ (D, f )) is generated by M2 (D) and M2 (C ∗ (f )), f ⊕f = f ⊗12 is in the center
of M2 (C ∗ (D, f )), and the ideal M2 (C ∗ (f (1 − f ))) is the hereditary C *-subalgebra
of M2 (C ∗ (D, f )) that is generated by g := (f (1 − f )) ⊗ 12 = (f − f 2 ) ⊕ (f − f 2 ) .
Use the identification M2 (A) = A ⊗ M2 , then e ⊕ e = e ⊗ 12 , f ⊕ f = f ⊗ 12 ,
M2 (C ∗ (D, f )) = C ∗ (M2 (D), M2 (C ∗ (f )), M2 (C ∗ (f (1 − f ))) = C ∗ (f (1 − f )) ⊗ M2 ,
f ⊕ f is in the center of M2 (C ∗ (D, f )).
check again!::
e ∼
We define on the unital C *-algebra M2 (D) = M2 (C ∗ (D, f ))/M2 (I) a function
ν(a + s) := inf n τ2 ((1 − g )(a + f ⊗ s)(1 − g ) for selfadjoint a∗ = a ∈ M2 (D)
1/n 1/n
1102 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

and s∗ = s ∈ M2 with ???????? ≤ a + s ≥ 0. It is obvious that ν(d ⊕ 0) = τ (d) for

all a ∈ D+ , and it is not hard to see, that ν|C+ is additive for every commutative
C *-subalgebra C ⊆ M2 (D) e , because C ∗ (C · (f ⊗ 12 ), g) ⊆ M2 (C ∗ (D, f )) is a
commutative C *-subalgebra of M2 (C ∗ (D, f )) .
Thus ν1 (d + t) = ν((d + t) ⊕ 0) (for d∗ = d ∈ D and t ∈ R+ with d + t ≥ 0) is
a bounded 2-quasi-trace on De + with ν1 (d) = τ (d) for d ∈ D+ .

(ii): Let a, b ∈ A+ . Since τ is in particular a lower s.c. 2-quasi-trace, one has

τ (a + b) ≤ 2(τ (a) + τ (b)) and max(τ (a), τ (b)) ≤ τ (a + b). Thus τ (a) + τ (b) = +∞ ,
if and only if, τ (a + b) = +∞ . Hence, we may suppose that τ (a + b) < ∞ .
Let gδ (t) := min((2/δ) max(t − δ/2, 0), 1) for t ∈ R+ and δ ∈ (0, 1] , and notice
that gδ gε = gε for δ ≤ ε/2 .
If τ (a + b) < ∞ and δ > 0, let e := gδ (a + b) and D := {x ∈ A ; xe = ex = x } .
The elements c := g4δ (a+d)1/2 and d := ka+bkg2δ (a+d)1/2 are in D+ , d commutes
with all elements in cAc and d ≥ c(a+b)c. In particular, τ (d−cac)+τ (cac) = τ (d) ,
τ (d − cbc) + τ (cbc) = τ (d) , and
2τ (d) − τ (c(a + b)c) = τ (2d − c(a + b)c) = τ ((d − cac) + (d − cbc))
The sub-additivity of τ implies τ (c(a + b)c) ≤ τ (cac) + τ (cbc) , and
2τ (d) − τ (c(a + b)c) ≤ τ (d − cac) + τ (d − cbc) = 2τ (d) − (τ (cac) + τ (cbc)) .
Hence, τ (c(a + b)c) = τ (cac) + τ (cbc) .
The elements a, b, a + b are in E := (a + b)A(a + b). Since
τ (y 1/2 g2δ (a + b)y 1/2 ) = τ (g2δ (a + b)1/2 yg2δ (a + b)1/2 )
and
τ (y) = lim τ (y 1/2 g2δ (a + b)y 1/2 )
δ&0
for all y ∈ E+ by monotony and lower semi-continuity of τ , it follows τ (a + b) =
τ (a) + τ (b). 

Remark A.7.5. Similar arguments show:

A lower s.c. local quasi-trace τ : A+ → [0, ∞] is 2-subadditive,
i.e., τ (a + b) ≤ 2(τ (a) + τ (b)) for all a, b ∈ A+ , if and only if
τ (a + b) = ∞ implies τ (a) + τ (b) = ∞, and, for every e ∈ A+ with τ (e) < ∞
and kek = 1, τ |D+ is 2-subadditive, where D is the the hereditary C*-subalgebra
D := {a ∈ A ; ae = ea = a } .

Corollary A.7.6. Let A an exact C*-algebra. Then every lower semi-

continuous 2-quasi-trace τ : A+ → [0, ∞] is an additive trace.

Proof. If A is exact, unital and τ (1) < ∞, then this was shown by U.
Haagerup in [342] (see also [348]).
It suffices to show in the non-unital case that τ |D+ is sub-additive for every
hereditary C *-subalgebra D ⊆ A with the property that there is e ∈ A+ with
τ (e) < ∞, kek = 1 and de = ed = d for all d ∈ D, cf. Lemma A.7.4(ii). But then
 8. PROJECTIVITY OF SOME C *-ALGEBRAS 1103

e + → [0, ∞) of τ |D+ by Lemma A.7.4(i). Where we use

there is an extension τe : D
that the unitization D of D ⊆ A is exact if A is exact.
e 

Remark A.7.7. There exist unital quasi-traces that are not 2-quasi-traces on
the unital C *-algebra A := C([0, 1]) ∗ C([0, 1]), i.e., on the unital free product of
two copies of C([0, 1]), and on a type-I C *-algebra A that is a unital extension

0 → c0 (K) → A → C([0, 1]2 ) → 0 .

Cite! place of exact definition.

8. Projectivity of some C *-algebras

Definition A.8.1. A C *-algebra B is projective if for every C *-algebra A,

closed ideal J / A of A and C *-morphism ψ : B → A/J there exists a C *-morphism
φ : B → A with ψ = πJ ◦ φ (φ is a lift of ψ).

Obvious examples of projective C *-algebras are the free algebras generated by

a finite, countable or uncountable number of contractions. (The latter by using the
Axiom of Choice.)
Here we display a proof of the projectivity of the cones CF := C0 ((0, 1], F )
over finite-dimensional C *-algebras F .
The following Lemma A.8.2 reformulates [540, thm. 3.3]. We use a variant of
[616, prop. 1.5.10] for a proof.

Lemma A.8.2. Let J / A, a, b ∈ A+ and d ∈ A/J with d∗ d ≤ πJ (a) and

dd∗ ≤ πJ (b). Then there exists g ∈ A with g ∗ g ≤ a, gg ∗ ≤ b and πJ (g) = d.

Remark A.8.3. Lemma A.8.2 is a two-sided version of an order lifting theorem

of Combes [155], cf. also [616, prop. 1.5.10]:
Let ρ : B → C a C *-algebra epimorphism. If x ∈ B+ and y ∈ C satisfy
y y ≤ ρ(x) then there exists some z ∈ B with ρ(z) = y and z ∗ z ≤ x.
∗

Inspection of the proof of [616, prop. 1.5.10] allows to see moreover the follow-
ing:
For each b ∈ B with ρ(b) = y there exists z ∈ B with ρ(z) = y, z ∗ z ≤ x and
(in addition) zz ∗ ≤ bb∗ .
Indeed: Let b ∈ B with ρ(b) = y and J := ρ−1 (0). Then (x − b∗ b)− ∈ J+ , and
c = (x − b∗ b)− + x satisfies ρ(c) = ρ(x), b∗ b ≤ c and x ≤ c.
Let z := limn→∞ b(n−1 1 + c)−1/2 x1/2 . This limit exists because the elements
tn := b(n−1 1 + c)−1/2 x1/2 satisfy via monotony and centrality of C *-algebra norms
the inequalities

ktm − tn k2 ≤ k (m−1 1 + c)−1/2 − (n−1 1 + c)−1/2 c k .

The equation πJ (x) = πJ (c) implies πJ (z) = πJ (b). The additional inequality
zz ∗ ≤ bb∗ comes from k(n−1 1 + c)−1/2 x1/2 k2 ≤ 1 via x ≤ c.
1104 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

Proof of Lemma A.8.2: Let J / A , a, b ∈ A+ and d ∈ A/J with

d∗ d ≤ πJ (a) and dd∗ ≤ πJ (b) .
We find some e ∈ A with πJ (e) = d by surjectivity of πJ . There exists f ∈ A with
f ∗ f ≤ a and πJ (f ) = d, by Remark A.8.3.
We can apply again Remark A.8.3, but this time with (d∗ , f ∗ , b) in place of
(y, b, x), because πJ (f ) = d and dd∗ ≤ πJ (b). Get z ∈ A with πJ (z) = d∗ , z ∗ z ≤ b
and zz ∗ ≤ f ∗ f .
The element g := z ∗ has the desired properties: g ∗ g ≤ f ∗ f ≤ a, gg ∗ = z ∗ z ≤ b
and πJ (g) = πJ (z)∗ = d . 

Proposition A.8.4. [540, cor. 3.8] For each C*-algebra F of finite dimension,
the cone CF := C0 ((0, 1], F ) of F is projective in sense of Definition A.8.1.

Proof. The following simple observation works only for σ-unital C *-algebras
B1 and B2 . This is necessary to observe because the (free) universal C *-algebra
generated by un-countably many contractions is projective (by using the Axiom of
Choice) but is not σ-unital.
If B1 and B2 are projective σ-unital C*-algebras, then B1 ⊕ B2 is projective.
Indeed: Let ψ : B1 ⊕ B2 → A/J a C *-morphism, e1 ∈ B1 and e2 ∈ B2 strictly
positive contractions, and let a ∈ A with πJ (a) = ψ(e1 ⊕ (−e2 )). The elements
a1 := (a∗ + a)+ , a2 := (a∗ + a)− ∈ A+ and hereditary C *-subalgebras Dk :=
ak Aak (k ∈ {1, 2}) satisfy D1 D2 = {0}, ψ(B1 ⊕ {0}) ⊆ πJ (D1 ) ∼ = D1 /(J ∩ D1 )
∼
and ψ({0} ⊕ B2 ) ⊆ πJ (D2 ) = D2 /(J ∩ D2 ). Then ψ1 (b1 ) := ψ(b1 ⊕ 0) (b1 ∈ B1 )
and ψ2 (b1 ) := ψ(0 ⊕ b2 ) (b2 ∈ B2 ) are C *-morphisms into D1 /(J ∩ D1 ) respectively
D2 /(J ∩ D2 ). By projectivity of B1 and B2 there are C *-morphisms φk : Bk → Dk
with πJ (φk (bk )) = ψk (bk ) for bk ∈ Bk (k ∈ {1, 2}). The map φ(b1 ⊕ b2 ) :=
φ1 (b1 ) + φ2 (b2 ) (bk ∈ Bk ) is a linear lift of ψ. It is a C *-morphism, because
D1 D2 = {0}.
This shows that finite orthogonal direct sums of σ-unital projective C *-algebras
are projective.
Let k1 , k2 , . . . , kn ∈ N and F := Mk1 ⊕ · · · ⊕ Mkn . It is easy to see that
CF := C0 ((0, 1], F ) ∼
= C0 ((0, 1], Mk1 ) ⊕ · · · ⊕ C0 ((0, 1], Mkn ) .
The additive invariance of the class of σ-unital projective C *-algebras shows that it
is enough to prove the projectivity of CF only in the cases F := C and of F := Mn
for n > 1.
If F := C then CF ∼ = C0 (0, 1] and has the generator f0 defined by f0 (t) = t
(t ∈ [0, 1]). ψ : CF → A/J is determined by the positive contraction b := ψ(f0 ).
Let d ∈ A with πJ (d) = b1/2 . Then πJ (f (a0 )) = ψ(f ) for f ∈ C0 (0, 1] and
a0 := d∗ d − (d∗ d − 1)+ . Thus φ : f 7→ f (a0 ) is a lift of ψ.
Since projectivity is invariant under direct sums we get that CF is projective
for F = Cn = C ⊕ · · · ⊕ C.
 8. PROJECTIVITY OF SOME C *-ALGEBRAS 1105

If F := Mn with n > 1 then the C *-algebra CF = CMn := C0 ((0, 1], Mn ) is

isomorphic to the universal C *-algebra An := C ∗ (x2 , . . . , xn ; Rn ) with conditions
and relations Rn given by kx2 k ≤ 1, x∗j xk = δjk x∗2 x2 and xj xk = 0 for j, k ∈
{2, . . . , n}.
Indeed (by using arguments in the proof [538, prop. 2.7]): It is easy to see that
the elements γn (xk ) := f0 ⊗ pk,1 ∈ C0 ((0, 1], Mn ) – with the canonical matrix units
pj,k ∈ Mn – generate C0 (0, 1] ⊗ Mn = C0 ((0, 1], Mn ) and satisfy the norm-condition
and relations Rn and therefore define a *-epimorphism onto C0 ((0, 1], Mn ). More-
over, it is not difficult to check that the positive element

Zn := (x1,2 x∗1,2 + x∗1,2 x1,2 + . . . + x∗1,n x1,n )1/2

is a contraction in the center of An and γn (Zn ) = f0 ⊗1n . Since it is in the center, it

must be strictly positive in An (by the defining relations). Thus, every irreducible
*-representation ρ of An maps Zn to a scalar ρ(zn ) = α1 for some α ∈ (0, 1] and
that ρ(An ) ∼= Mn with ρ(x1,k ) = αp1,k .
Use now that γn (Zn ) = f0 ⊗ 1n to get that the natural C *-morphism γn from
An onto C0 ((0, 1], Mn ) is an isomorphism, because any irreducible representation
of A is the compositions of γn with the irreducible representations of C0 ((0, 1], Mn )
and those exhaust all the corresponding α ∈ (0, 1] .
The relations Rn (or its realization in CMn = C0 (0, 1]⊗Mn ) show the existence
of canonical C *-morphisms from An into An+1 that agrees with the natural C *-
morphisms from CMn into CMn+1 given by the inclusion Mn ⊆ Mn+1 (as left
upper corner). Moreover the natural inclusion CF ⊂ C0 ((0, 1], Mn ) for F := Cn
defines an isomorphism from CF onto

C ∗ ((x2 x∗2 )1/2 , (x∗2 x2 )1/2 , . . . , (x∗n xn )1/2 ) ⊆ An ,

in a way such that f0 ⊗ 1n ∈ CF corresponds to Zn , f0 ⊗ p11 to (x∗2 x2 )1/2 and the

f0 ⊗ pkk map to (xk x∗k )1/2 for k = 2, . . . , n.
Let ψ : An → A/J a C *-morphism. The projectivity of CF ∼ = C0 ((0, 1], Cn )
for F := C shows that there is a C *-morphism φ1 : `n (C0 (0, 1]) = CCn → A
n

that is always given by mutually orthogonal positive contractions a1 , . . . , an ∈ A+

with φ1 ((x∗2 x2 )1/2 ) = a1 and φ1 ((xk x∗k )1/2 ) = ak for k = 2, . . . , n, and satisfies the
equations

πJ (a1 )2 = ψ(x∗2 x2 ) and πJ (ak )2 = ψ(xk x∗k ) for k ≥ 2 . (8.1)

Let a1 , . . . , an ∈ A mutually orthogonal positive contractions, and suppose that

there are given C *-morphism ψ : An → A/J with ψ(x∗2 x2 ) ≤ πJ (a21 ) and ψ(xk x∗k ) ≤
πJ (a2k ) for k ≥ 2. We show that there exists a C *-morphism φ : An → A with
πJ ◦ φ = ψ .
In addition we show that we can find the φ with φ(x∗2 x2 ) ≤ a21 and φ(xk x∗k ) ≤ a2k
for k ≥ 2.
We proceed by induction over k ≥ 2 and produce from above defined
φ1 : CCn → A, given by mutually orthogonal positive contractions a1 , . . . , an ∈ A+ ,
1106 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

inductively C *-morphisms φk : Ak → A with φk (x2 )∗ φk (x2 ) ≤ a21 , φk (xj )φk (xj )∗ ≤

a2j and πJ (φk (xj )) = ψ(xj ) for j = 2, . . . , k.
If k = 2 then this follows immediately from Lemma A.8.2 with a := a21 , b := a22 ,
d := ψ(x2 ).
Let 2 < k < n and suppose that we have a C *-morphism φk : Ak → A with
the above listed properties. We apply Lemma A.8.2 to a := φk (x2 )∗ φk (x2 ) ≤ a21 ,
b := a2k+1 and d := ψ(xk+1 ), and obtain g ∈ A with g ∗ g ≤ φk (x2 )∗ φk (x2 ), gg ∗ ≤
a2k+1 and πJ (g) = ψ(xk+1 ).
We use φk and g to define a C *-morphism φk+1 : Ak+1 → A with πJ ◦ φk+1 =
ψ|Ak+1 , φk+1 (x∗2 x2 ) ≤ a21 and φk+1 (xj x∗j ) ≤ a2j .
Recall that φk (xi )∗ φk (xj ) = δij φk (x2 )∗ φk (x2 ) for i, j ∈ {2, . . . , k} and let S :=
(φk (x2 )∗ φk (x2 ))1/2 . Then φk (xj ) := vj S is the polar decomposition of φk (xj )
with the partial isometries vj ∈ A∗∗ given by vj = limn φk (xj )(n−1 1 + S)−1 . It
follows that vj∗ vi = δij v2∗ v2 . Since g ∗ g ≤ S 2 , we can define yj := vj (g ∗ g)1/2 =
limn φk (xj )(n−1 1 + S)−1 (g ∗ g)1/2 (j = 2, . . . , k) and yk+1 := g.
The yj satisfy yj yj∗ ≤ φk (xj )φk (xj )∗ ≤ a2j , yi∗ yj = δij y2∗ y2 = δij g ∗ g. for
∗
2 ≤ i, j ≤ k, and yk+1 yj = g ∗ yj = 0 for j ≤ k because gg ∗ ≤ a2k+1 and yj yj∗ ≤ a2j .
Moreover, yi yj = 0 because yi∗ yi ≤ g ∗ g ≤ a21 and yj yj∗ ≤ a2j and a1 aj = 0 for
j = 2, . . . , n. ky2 k = kgk ≤ kφk (x2 )k ≤ 1
Thus, the elements {y2 , . . . , yk+1 } ⊂ A satisfy the condition and relations Rk+1
of Ak+1 . Moreover, they satisfy y2∗ y2 ≤ a21 and yj yj∗ ≤ a2j for j = 2, . . . , k + 1. Let
φk+1 : Ak+1 → A the corresponding C *-morphism.
We check that πJ ◦ φk+1 is the restriction of ψ : An → A/J to Ak+1 :
Recall that πJ (yk+1 ) = πJ (g) = ψ(xk+1 ) by construction of g, and
πJ (φk (xj )) = ψ(xj ) for 2 ≤ j ≤ k by assumption on φk . In particular
πJ ((g ∗ g)1/2 ) = (ψ(x2 )∗ ψ(x2 ))1/2 , and πJ (S) = (ψ(x2 )∗ ψ(x2 ))1/2 by defini-
tion of S := (φk (x2 )∗ φk (x2 ))1/2 . Let T := (ψ(x2 )∗ ψ(x2 ))1/2 . It satisfies
ψ(xj )∗ ψ(xj ) = T 2 . Then the definition of the yj for j = 2, . . . k show the norm
convergence
πJ (yj ) = lim ψ(xj )(n−1 1 + T )−1 T = ψ(xj ) .
n

It says that φk+1 , – defined by φk+1 (xj ) := yj –, satisfies πJ ◦ φk+1 = ψ|Ak+1 . 

Corollary A.8.5. For each n ∈ N the universal C*-algebra A generated by n

contractions t1 , . . . , tn with mutually orthogonal ranges is projective.

Proof. Notice that A is given by universal C *-algebra with relations

A := C ∗ (t1 , . . . , tn ; ktk k ≤ 1, t∗k t` = 0 for k 6= ` , k, ` ∈ {1, . . . , n} )

with contractions tk . The relations imply the additional relation t1 t∗1 + · · · + tn t∗n ≤
1 . Let B a C *-algebra, J ⊆ B a closed ideal of B and ρ : A → B/J a C *-morphism.
We let sk := ρ(tk ) . All we need is to find contractions c1 , . . . , cn ∈ B with c∗j ck = 0
 9. PROJECTIVITY AND C.P.C. ORDER ZERO MAPS 1107

for j 6= k and πJ (ck ) = sk := ρ(tk ), because then there is a C *-morphism ψ : A → B

with ψ(tk ) = ck , which implies πJ ◦ ψ = ρ.
The C *-subalgebra of A generated by (tk t∗k )1/4 (k = 1, . . . , n) is the image of
a unique C *-morphism γ : CF → A for F := Cn and

CF ∼
= C ∗ (y1 , . . . , yn ; 0 ≤ yk ≤ 1, yk y` = 0, for k 6= ` )

with γ(yk ) = (tk t∗k )1/4 . By the projectivity of CF there exists a C *-morphism λ
from CF into B that is a lift of ρ ◦ γ : CF → B/J. The C *-morphism λ satisfies
the equations
πJ (λ(yk )) = ρ((tk t∗k )1/4 ) = (sk s∗k )1/4 .
The polar decomposition sk = vk (s∗k sk )1/2 = (sk s∗k )1/2 vk of sk in (B/J)∗∗ has
the property that (sk s∗k )1/4 vk = vk (s∗k sk )1/4 ∈ B/J . Let xk ∈ B elements with
πJ (xk ) = (sk s∗k )1/4 vk . Then the elements bk := λ(yk )xk satisfy b∗` bk = 0 for k 6= `
and πJ (bk ) = sk for k = 1, . . . , n. We define elements ck ∈ B by ck := bk f (b∗k bk ) =
limp→∞ bk f (1/p+b∗k bk ) for the continuous function f (t) := max(1, t)−1/2 on [0, ∞).
Use here that the element f (b∗k bk ) is a multiplier of C ∗ (b∗k bk ). The elements
c1 , . . . , cn are contractions in B with c∗k cj = 0 for j 6= k (i.e., with orthogonal
ranges) and πJ (ck ) = ρ(tk ). 

Contents of next Remark should be exist on

other places! Find it! And delete one of them.
Delete following Lemma ref.lem:A.old.2.5
Replace citations as follows:
ref.lem:A.old.2.5(i,ii) by Remark A.8.6????,
ref.lem:A.old.2.5(iii) by Remark 2.1.16(ii),
Old ref.lem:A.1.25 =? ref.lem:A.old.2.5(iv) by ?????
ref.lem:A.old.2.5(v) by ?????
ref.lem:A.old.2.5(iv) = Old A.1.25?? by ??????
Begin: Old Lemma ref.lem:A.old.2.5

Remark A.8.6. The projectivity of CF can be equivalent expressed by the

following formulation:
If F is a C *-algebra of finite linear dimension then every extension

0 → J → E → C0 ((0, 1], F ) → 0

of C0 ((0, 1], F ) by a C *-algebra J is a split extension.

9. Projectivity and c.p.c. order zero maps

Elements a, b ∈ A of a C *-algebra A are orthogonal if (aa∗ +a∗ a)(bb∗ +b∗ b) = 0.

We denote this by a⊥b. It is equivalent to the 4 equations ab = 0, ba = 0, a∗ b = 0
and ab∗ = 0 . And it says equivalently that h1,j h2,k = 0 for j, k ∈ {1, 2}, where
h1,1 , h1,2 , h2,1 , h2,2 ∈ A are the selfadjoint elements with a = h1,1 + ih1,2 and
1108 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

b = h2,1 + ih2,2 . In particular, for self-adjoint a, b ∈ A holds a⊥b if and only if

ab = 0.

Definition A.9.1. Let A and B C *-algebras and ψ : A → B a bounded linear

map that is invariant under passage to adjoint elements, i.e., ψ(a∗ ) = ψ(a)∗ for all
a ∈ A ( 5 ). We call a map ψ with this property orthogonality preserving if ψ has
the additional property that ψ(h)ψ(k) = 0 for selfadjoint h, k ∈ A with hk = 0.
(6)
If ψ is moreover a completely positive contraction, then ψ is called an “order-
zero map” by W. Winter and J. Zacharias in [836, def. 1.3].
(We call it here sometimes “order-zero morphism”.)

If ϕ : C0 ((0, 1], A) → B is a C *-algebra homomorphism then ψϕ (a) := ϕ(f0 ⊗a)

is the contractive completely positive order-zero map defined by ϕ.
It turns out that there exists a natural decomposition of ψϕ into element-
wise commuting C *-morphisms λ : C0 (0, 1] → M(D) and ρ : A → M(D) with
ψϕ (f ⊗a) = λ(f )ρ(a) for all f ∈ C0 (0, 1] and a ∈ A, where D denotes the hereditary
C *-subalgebra B generated by ψϕ (A) ...
We use in this book only the C *-morphisms ϕ : C0 ((0, 1], A) → B instead of
order-zero morphisms, because by our
Lemmata ??, ??, ??, and Proposition ??
there is a natural bijective relation between contractive, 2-positive and orthogo-
nality preserving maps ψ : A → B and the usual C *-morphisms ϕ : C0 ((0, 1], A) →
B that is given by ψ(a) = ϕ(f0 ⊗ a) for all a ∈ A. It is a bijective relation be-
cause the 2-positive contraction ψ determines ϕ uniquely by the following obvious
equations for all n ∈ N and a ∈ A+ :
ϕ(f0n ⊗ a) = ϕ(f0 ⊗ a1/n )n = ψ(a1/n )n .
In particular, this implies that all 2-positive and orthogonality preserving maps
ψ : A → B are automatically completely positive!
The following Proposition A.9.2 should be moved below general observations!
One application of the projectivity of C0 ((0, 1], F ) for a finite-dimensional C *-
algebra F and the characterization of 2-positive order-zero maps ϕ : F → A/J is
the following proposition.

Proposition A.9.2. Let F a C*-algebra of finite dimension and ϕ : F → A/J

a 2-positive order-zero map, then there exists a completely positive order zero map
ψ : F → A with πJ ◦ ψ = ϕ .

Proof. For each 2-positive order-zero map ϕ from F into the C *-algebra
B := A/J there exists a C *-morphism λ0 : C0 ((0, 1], F ) → B and f ∈ C0 (0, 1]+
5 Such maps ψ are sometimes called “symmetric”.
6 Equivalently, a ⊥a implies ψ(a )⊥ψ(a ), because ψ(a∗ ) = ψ(a)∗ on A. M. Wolff calls
1 2 1 2
our “orthogonality preserving” maps “disjointness preserving” in [838].
 9. PROJECTIVITY AND C.P.C. ORDER ZERO MAPS 1109

with ϕ(b) = λ0 (f ⊗b) for all b ∈ B. By Proposition A.8.4, the algebra C0 ((0, 1], F ) is
projective. Thus, there exists a C *-morphism λ1 : C0 ((0, 1], F ) → A with πJ ◦ λ1 =
λ0 . Then ψ(b) := λ1 (f ⊗ b) defines a completely positive order-zero from F into A
with πJ ◦ ψ = ϕ. 

Here are some citations from the paper [836] of W. Winter and J. Zacharias.
Erase them or replace them by my own version (with citations).
Corollary 3.1 of WILHELM WINTER AND JOACHIM ZACHARIAS in
COMPLETELY POSITIVE MAPS OF ORDER ZERO:
[836, cor. 3.1]
Let A and B be C *-algebras, and ϕ : A → B a c.p.c. order zero map. Then,
the map given by ρϕ (f0 ⊗ a) := ϕ(a) (for a ∈ A) induces a *-homomorphism
ρϕ : C0 ((0, 1], A) → B .
Conversely, any *-homomorphism ρ : C0 (0, 1] ⊗ A → B induces a c.p.c. order
zero map ϕρ : A → B via ϕρ (a) := ρ(f0 ⊗ a). These mutual assignments yield
a canonical bijection between the point-norm closed set of c.p.c. order zero maps
from A to B and the *-homomorphisms from C0 (0, 1] ⊗ A to B.
My Remarks:

The positive case:

If A is abelian and ϕ : A → B positive, then ϕ is completely positive, because the
second conjugate ϕ∗∗ : A∗∗ → B ∗∗ is again positive. Then ϕ and ϕ∗∗ are moreover
a completely positive map because each finite subset of A∗∗ can be approximated
(in norm) by the linear span of finitely many projections in A∗∗ .
Thus, then there are unique (!) C *-morphisms ρC : C(0, 1] ⊗ C → B with
ρC (f0 ⊗ c) = ϕ(c) for each commutative C ⊆ A ...
Check if it is on intersections of different abelian C ⊆ A the same. (Seems to
be ...)
But the PROBLEM (!) is, that general elements of (C(0, 1] ⊗ A)+ are not
contained in some C(0, 1] ⊗ C ... Thus, this old idea of proof don’t work.

The following theorem reformulates the theorem [838, thm.2.3] of M. Wolff on

Jordan algebras for the more special case of C *-algebras, where he calls the above
defined “symmetric” and “orthogonality preserving” maps “disjointness preserv-
ing”.

Theorem A.9.3. Let A and B be C*-algebras, with unital A, and let ϕ : A → B

be a disjointness preserving map. Define D := {ϕ(1A )}0 ∩ B ⊆ B and

C := ϕ(1A ) · D .

Then, ϕ(A) ⊆ C and there is a Jordan *-homomorphism ψ : A → M(C) from A

into the multiplier algebra of C satisfying ϕ(a) = ϕ(1A )ψ(a) for all a ∈ A .
1110 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

The ideas of his proof are related to some study of Lamperti operators by
W. Arendt in [41].
Notice that ϕ(1A ) is selfadjoint, but is here not required to be positive. The
proof of Theorem A.9.3 reduces to the case where A = C[0, 1], but is not obvious.
We use the below given Lemma ?? and consider then a slightly more general
case where A is not necessarily unital, ϕ : A → B is a bounded linear map with the
properties that ϕ is “symmetric” – in the sense that ϕ(a∗ ) = ϕ(a)∗ for all a ∈ A
– and is “orthogonality preserving” – in the sense that ϕ(a1 )ϕ(a2 ) = 0 if a1 a2 = 0
for orthogonal selfadjoint elements a1 = a∗1 , a2 = a∗2 (i.e., with a1 a2 = 0).

Remark A.9.4. First (see below explained): Banach case for V : Y ∗ →

X ∗.
Then V is σ(Y ∗ , Y )-σ(X ∗ , X) continuous
if and only if
there exists bounded linear map U : X → Y such that V = U ∗ .
But look to the notations!! Chaos there???
The continuity properties for S : A∗∗ → B ∗∗ and T : A∗∗ → B ∗∗ with respect
to σ(A∗∗ , A∗ ) and σ(B ∗∗ , B ∗ ) (... play role of V := U ∗ : Y ∗ → X ∗ , e.g. with
suitable S∗ : B ∗ → A∗ in the role of U : X → Y with continuity of V with respect
to σ(Y ∗ , Y ) and σ(X ∗ , X).)
Because then, for example, the linear functional given by a ∈ A∗∗ 7→ ϕ(T (a))
where ϕ ∈ B ∗ is equal the linear functional V (ϕ) ∈ A∗ . This proves the continuity
properties for T = V ∗ .
V : Y ∗ → X ∗.
Then V is σ(Y ∗ , Y )-σ(X ∗ , X) continuous
if and only if
there exists bounded linear map U : X → Y such that V = U ∗ .

Conversely, the quoted continuity properties e.g. for T on A∗∗ implies that for
each ϕ ∈ B ∗ the map ϕ ◦ T is in A∗ and has norm ≤ kϕk · kT k. Thus, V (ϕ) := ϕ ◦ T
defines a bounded linear map V : B ∗ → A∗ .
Compare notational chaos with above:
From Banach space theory it follows that a continuous linear map V : X → Y
(here U := V ∗ = T∗ ) for Banach spaces X and Y (here in our case X := B ∗ and
Y := A∗ ) has an adjoint T := L∗ : Y ∗ = A∗∗ → X ∗ = B ∗∗ that is σ(A∗∗ , A∗ ) –
σ(B ∗∗ , B ∗ ) continuous, i.e., is continuous with respect to the topologies σ(A∗∗ , A∗ )
on A∗∗ and σ(B ∗∗ , B ∗ ) on B ∗∗ , because this means that L(ϕ) = ϕ ◦ T is in A∗ for
each ϕ ∈ B ∗ .
In particular, then ????
 9. PROJECTIVITY AND C.P.C. ORDER ZERO MAPS 1111

Indeed, we can consider ϕ : B → C as linear operator. Then ϕ∗∗ : B ∗∗ → C

is in B ∗ ⊆ (B ∗∗ )∗ and (ϕ ◦ V )∗∗ = ϕ∗∗ ◦ T . But this says that T : A∗∗ → B ∗∗ is
continuous with respect to σ(A∗∗ , A∗ ) and σ(B ∗∗ , B ∗ ).
In the special case of ???????? we have to require in addition also that U (a∗ ) =
U (a)∗ and V (a∗ ) = V (a)∗ for all a ∈ A, such that S := U ∗∗ and T := V ∗∗ map
selfadjoint elements to selfadjoint elements.

Remark A.9.5. We consider on the class of C *-algebras only “symmetric”

Jordan morphisms T : A → B for C *-algebras A and B, i.e., linear maps with the
properties that T (a∗ ) = T (a)∗ for all a ∈ A and T (h2 ) = T (h)2 for h∗ = h ∈ A.
It follows then that T (a∗ b + b∗ a) = T (a)∗ T (b) + T (b)∗ T (a) for all a, b ∈ A.
The second conjugate T ∗∗ : A∗∗ → B ∗∗ of a C *-Jordan morphism is again a
C *-Jordan morphism.
PROOF of this ??:
At first T and T ∗∗ are positive linear contractions on As.a. and (A∗∗ )s.a. , be-
0
cause T (a) = T (a1 /2 )T (a1/2 ) ≥ 0 for a ∈ A+ and χ(T (·)) ∈ (A∗ )+ for positive
χ ∈ B ∗ gives that its natural extensions is the positive functional χ(T ∗∗ (·)) ∈ A∗
and is again positive on A∗∗+ and T
∗∗
is σ(A∗∗ , A∗ ) – σ(B ∗∗ , B ∗ ) continuous.
Let 0 ≥ c ∈ A∗∗ with kck = 1. Then there exists a net of self-adjoint contrac-
tions {aλ } ⊆ A+ that converges to c *-ultra-strongly in A∗∗ . Then T (a2λ ) → T (c2 )
???????
If ρ ∈ B ∗ is a state then ρ(T (aλ )T (aλ )) →???

Lemma A.9.6. Let A and B C*-algebras and T : A → B a C*-Jordan mor-

phism. Then the following properties of T are equivalent:

(i) T is 2-positive, i.e., T ⊗ idM2 : A ⊗ M2 → B ⊗ M2 is positive.

(ii) The linear map T2 := T ⊗ idM2 : A ⊗ M2 → B ⊗ M2 is contractive, i.e.,
has norm ≤ 1.
(iii) T is C*-algebra morphism.

Proof. It is somewhere calculated (further down?).

It holds (T2 )∗∗ = T2∗∗ . Thus one can restrict the considerations to W*-algebras.


Remark A.9.7. Let A denote a C *-algebra.

The open support projection pt ∈ A∗∗ of (a − t)+ ∈ A+ is the unit element of
Dt∗∗⊆ A∗∗ , where Dt := (a − t)+ A(a − t)+ with t ∈ [0, kak]. The projection pt is
contained in C ∗ (a)∗∗ ⊆ A∗∗ and is the unit element of C ∗ ((a − t)+ )∗∗ ⊆ A∗∗ .
We can here replace A by C ∗ (a) and A∗∗ by the second conjugate (C ∗ (a))∗∗ ⊆
A∗∗ of C ∗ (a).
We denote the projection p0 ∈ A∗∗ – respectively more generally the p0 for a
positive contraction in any W*-algebra M –, for a ∈ A+ also by pa , respectively
1112 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

by pb for b ∈ A+ (or b ∈ M ), to distinguish the support projections of different

elements a, b ∈ A+ .
If a 6= 0 and t ∈ [0, kak] then the projection pt is a projection that is the
smallest positive contraction q ∈ A∗∗ with the property

(k(a − t)+ k−1 (a − t)+ )1/n ≤ q for all n ∈ N .

This remains true for every W ∗ -algebra M (in place of the special case M :=
A∗∗ ):
The “support projection” pa ∈ M for some positive contraction a ∈ M+ is the
smallest positive contraction in M with a1/n ≤ pa for all n ∈ N; i.e.:
Let 0 ≤ x ≤ 1 in M with a1/n ≤ x for all n ∈ N then pa ≤ x.
In particular, if N is another W*-algebra and ψ : M → N is an injective unital
C *-algebra mono-morphism from M into N and E : N → ψ(M ) ⊆ N is a condi-
tional expectation of N onto a C *-subalgebra ψ(M ) of N , then E(pψ (a)) = ψ(pa ),
where pψ (a) the support projection in N of ψ(a) ∈ N and pa the support projection
in N of a positive contraction a ∈ N .
This applies to the natural unital conditional expectation from N := M ∗∗ onto
M ⊆ M ∗∗ , where ψ denotes here the canonical embedding of the W*-algebra M in
its second conjugate N .
Let 0 ≤ s < t ≤ 1, a ∈ A+ with kak ≤ 1 and ps , pt ∈ A∗∗ the open support
projection of (a−s)+ and (a−t)+ then (a−t)+ ≤ (a−s)+ , ps pt = pt and (a−t)+ =
((a−s)+ −(t−s))+ ≤ (t−s)·ps +(a−t)+ . It follows that (a−s)+ −(a−t)+ ≤ (t−s)ps .
Moreover, pt ((a − s)+ − (a − t)+ ) = (t − s)pt ≤ (a − s)+ − (a − t)+ .
Thus, (t − s)pt ≤ (a − s)+ − (a − t)+ ≤ (t − s)ps for the support projections pt
for (a − t)+ and ps for (a − s)+ and 0 ≤ s ≤ t ≤ 0.

Lemma A.9.8. Let A a C*-algebra.

(i) Every a ∈ A+ is contained in the norm closure of the convex set generated
by the open support projections pt := p(a−t)+ ∈ A∗∗ of (a − t)+ ∈ A+ with
t ∈ (0, kak) .
(ii) All elements of A are contained in the (norm-)closed linear span of the
open support projections pa ∈ A∗∗ of elements a ∈ A+ .
(iii) Let T : A → M a bounded linear map into a W*-algebra M , and let
Te : A∗∗ → M denote its unique natural extension to a σ(A∗∗ , A∗ )–
σ(M, M∗ ) continuous linear map (i.e., the normal extension of T ).
The point is:
It uses the natural conditional expectation from M ∗∗ onto
M,
where M is considered as C*- subalgebra of M ∗∗ .
The linear map T is a Jordan morphism (cf. Remark A.9.5) if and
only if, Te(pa ) is an orthogonal projection in M for the open support pro-
jections pa ∈ A∗∗ of each elements a ∈ A+ .
 9. PROJECTIVITY AND C.P.C. ORDER ZERO MAPS 1113

Proof. ( i): Let a ∈ A+ with kak = 1 and pn,k ∈ C ∗ (a)∗∗ ⊆ A∗∗ the open
support projection of ak := (a − (k − 1)/n)+ ∈ C ∗ (a) ⊆ A for k ∈ {1, . . . , n + 1}.
Then a` ≤ ak for k ≤ `, and

ak+` (ak − ak+` ) = kak − ak+` kak+` .

Then a1 = a, a2 = (a − 1/n)+ , . . . , ak = (a − (k − 1)/n)+ , . . . , an = (a − (n −

Pn
1)/n)+ , an+1 = 0. Thus, a = k=1 (ak − ak+1 ) . Notice that ak ≥ a` for ` > k.
We denote hereby pn,k the open support projection of ak , k = 1, . . . , n + 1.
Then pn,n+1 = 0. Thus

1/npn,k+1 ≤ ak − ak+1 ≤ 1/npn,k .

It follows that
n
X n
X
( pn,k+1 ) ≤ n · a ≤ ( pn,k ) .
k=1 k=1

Since pn,n+1 = 0, the difference of the left and right estimates is = pn,1 . It
shows that a ∈ A+ can be approximated arbitrarily well by scalar multiples of sums
of open projections in A∗∗ .
(ii): Each element a ∈ A is the (complex) linear combination of at most 4
elements in A+ .
(iii): If T : A → M is a C *-Jordan morphism and a ∈ A+ is a positive
contraction then T (a) = T (a1/n )n for all n ∈ N. Thus, T (a)1/n = T (a1/n ) for
n ∈ N.
Is T ∗∗ : A∗∗ → M ∗∗ again C *-Jordan???
Is the “natural” map M ∗∗ → M multiplicative?
The increasing sequence T (a)1/n converges in M to the support projection of
T (a) with respect to the τ (M, M∗ )-topology, that is stronger than the σ(M, M∗ )-
topology.
The increasing sequence of elements a1/n converge τ (A∗∗ , A∗ )-strongly in A∗∗
to the support projection pa ∈ A∗∗ of a. Here pa A∗∗ pa is the τ (A∗∗ , A∗ ) closure of
the hereditary C *-subalgebra aAa
of ??????
In general the map Te : A∗∗ → M is continuous with respect to the topologies
σ(A , A∗ ) and σ(M, M∗ ), because Te = PM ◦ T ∗∗ where T ∗∗ : A∗∗ → M ∗∗ is the
∗∗

weakly continuous bi-dual map from T and PM : M ∗∗ → M

together τ -continuous ???
normal (!!! ???) W∗ -algebra epimorphism from M ∗∗ onto M .
Is this correct ????
T : A → M (positive because Jordan)
T ∗∗ : A∗∗ → M ∗∗ σ-continuous.
1114 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

(One could also use an orthogonality argument,

Or reduction to abelian case?
Take maximal commutative C *-subalgebra C of A that contains the given
element a ∈ A+ .
Then C ∗∗ ⊆ A∗∗ T |C is a C *-algebra morphism from C into M .
Second conjugate T ∗∗ |C = (T |C)∗∗ : C ∗∗ → M ∗∗ The algebra T (C) ⊆ M is
commutative.
in particular pa ∈ A∗∗ → Te(pa ) ∈ M


Lemma A.9.9. Let A and B denote C*-algebras and suppose that S, T : A∗∗ →
B ∗∗ are σ(A∗∗ , A∗ )-σ(B ∗∗ , B ∗ ) continuous “symmetric” linear maps in the sense
that S(a∗ ) = S(a)∗ and T (a∗ ) = T (a)∗ for all a ∈ A .

(i) Let a, b ∈ A+ positive contractions with a ∈ D := bAb and p, q ∈ A∗∗ the

open projection p = pq ≤ q corresponding to the hereditary C*-subalgebras
D and E := aAa ⊆ D of A.
Then S(p)T (1 − q) ∈ A∗∗ is contained in the σ(A∗∗ , A∗ )-closure of the
set of elements S(e)T (1 − d) with positive contractions e ∈ E+ , d ∈ D+
with de = e.
Moreover, each element S(e)T (1 − d) with positive contractions d ∈
D+ and e ∈ E+ and de = e is in the σ(A∗∗ , A∗ )-closure of the set of
elements S(e)T (g) with positive contractions g ∈ (1 − d)A(1 − d).
(ii) In particular, S(p)T (1 − q) = 0 if p, q ∈ A∗∗ are open projections with the
properties that qp = q and for each contractions a, b, c ∈ A+ with cb = b
and ca = a. holds S(a)T (c−b) = 0 for all positive contractions a, b, c ∈ A+
with the properties that a ≤ p and b ≤ q and cb = b a(c − b) = 0.
????????????????
S((a − 1/n)+ )T (1 − fm (b)) = 0 for all a ∈ A+ with 0 ≤ a ≤ p and
n, m ∈ N with (a − 1/n)+ (1 − fm (a)) = 0.

Remarks A.9.10. Notice that Part (iii????) of Lemma A.9.9 contains the case
a = b and D = E.
The continuity properties of S and T are equivalent to the existence bounded
linear maps U, V : A → B with S := U ∗∗ and T := V ∗∗ . Because in general Banach
space theory one has that a continuos map V : A → B from
In our case U (a∗ ) = U (a)∗ and V (a∗ ) = V (a)∗ for all a ∈ A, such that S := U ∗∗
and T := V ∗∗ .
Proofs not ready, because of the partial continuity of ??? !!!
Ad(iii):
 9. PROJECTIVITY AND C.P.C. ORDER ZERO MAPS 1115

TEXT: Let a, b ∈ A+ positive contractions with a ∈ D := bAb and p, q ∈ A∗∗

the open projection p = pq ≤ q corresponding to the hereditary C *-subalgebras D
and E := aAa ⊆ D of A.
Then S(p)T (1 − q) ∈ A∗∗ is contained in the σ(A∗∗ , A∗ )-closure of the set of
elements S(e)T (1 − d) with positive contractions e ∈ E+ , d ∈ D+ with de = e.
We use the piece-wise linear continuous functions ϕn (t) on [0, 1] given by
ϕn (t) := 2n+1 ((t − 2−(n+1) )+ − (t − 2−n )+ ) for n = 1, 2, . . .. This continuous
functions are non-negative, ϕn (0) = 0, kϕn k = 1, ϕn+1 ϕn = ϕn and ϕn converges
on each interval [s, 1] ⊆ (0, 1] uniformly to 1. For every n ∈ N and protection
P ∗ = P = P 2 holds ϕn (P ) = P .
Let en := ϕn (a) ∈ E+ , where a is a strictly positive contraction in E. Then
the en are positive contractions that satisfy em en = em for m < n and the sequence
e1 , e2 , . . . converges to the support projection p ∈ A∗∗ of a in the τ (A∗∗ , A∗ ) topol-
ogy, that is stronger than the σ(A∗∗ , A∗ ) topology on A∗∗ . By our assumption that
S : A∗∗ → A∗∗ is σ(A∗∗ , A∗ )-continuous shows that the sequence S(e1 ), S(e2 ), . . .
converges to the element S(p) ∈ A∗∗ in the σ(A∗∗ , A∗ ) topology.
Let p ≤ q ∈ A∗∗ denote the open support projection E = aAa ⊆ D = bAb.
Since en ≤ p ≤ q, we get that en + (1 − en )1/2 q(1 − en )1/2 = q for each n ∈ N, i.e.,
q − en = (1 − en )1/2 q(1 − en )1/2 .
Let fn := ϕn (a) ∈ D+ . This positive contractions satisfy fm fn = fm for
m < n, and the increasing sequence f1 , f2 , . . . converges to the support p of a
Need to find contractions dn ∈ D+ (for D = bAb) with dn en = en and dn → q
in A∗∗ .
dn−1 = en + (1 − en )1/2 fn (1 − en )1/2
1/2 1/2 1/2
for suitable contractions fn ∈ D+ with fn → q in A∗∗ and fn en − en fn
and en − fn en small ...
Since en ≤ p ≤ q, we get en + (1 − en )1/2 q(1 − en )1/2 = q ????

We call a projection p ∈ A∗∗ open in the following Lemma A.9.11 if pA∗∗ p is

the σ(A∗∗ , A∗ )-closure of a hereditary C *-subalgebra D of A. We say that p is a
σ-unital open projection if the corresponding D ⊆ A is σ-unital, i.e., if D = dAd for
some d ∈ A+ . The following Lemma A.9.11 considers special cases of orthogonally
preserving pairs of symmetric maps V, W : A → B.
(It seem that the general theory of such pairs, without assuming the additional
equivalent conditions (i)–(ix) in Lemma A.9.11 has to do with approximately inner
gradings on C *-algebras and questions of different nature.)

Lemma A.9.11. Let A and B C*-algebras and let V : A → B and W : A → B

linear maps that are “symmetric” in the sense that V (a∗ ) = V (a)∗ and W (b∗ ) =
V (b∗ ). and are relatively orthogonality preversing that satisfy

(1) T := V ∗∗ (1A∗∗ ) = W ∗∗ (1A∗∗ ) ∈ B ∗∗ and

1116 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

(2) V (a1 )W (a2 ) = 0 for all a1 , a2 ∈ A+ with a1 a2 = 0.

Then the following properties (i)–(ix) of V and W , respectively of its bi-adjoints

∗∗
V and W ∗∗ , are equivalent:

(i) T V (a) = V (a)T for all a ∈ A.

(ii) T W (a) = W (a)T for all a ∈ A.
(iii) T V ∗∗ (p) = V ∗∗ (p)T for all open projections p ∈ A∗∗ .
(iv) T W ∗∗ (p) = W ∗∗ (p)T for all open projections p ∈ A∗∗ .
(v) T V ∗∗ (p) = V ∗∗ (p)T for all σ-unital open projections p ∈ A∗∗ .
(vi) T W ∗∗ (p) = W ∗∗ (p)T for all σ-unital open projections p ∈ A∗∗ .
(vii) V = W.
(viii) V (a)V (b) = 0 for all a, b ∈ A+ with ab = 0.
(ix) W (a)W (b) = 0 for all a, b ∈ A+ with ab = 0.

Proof. 

Proposition A.9.12. If V : A → B is symmetric and orthogonality preserving

and has norm kV k ≤ 1, then there are two uniquely defined symmetric Jordan
morphisms ψk : C0 ((0, 1], A) → B (k ∈ {1, 2} with the following properties:

(i) ψ1 (C0 ((0, 1], A)) · ψ2 (C0 ((0, 1], A)) = {0} and
(ii) V (a) = ψ1 (f0 ⊗ a) − ψ2 (f0 ⊗ a) for all a ∈ A+ .

Proof. Next proof wrong here?

By Lemma A.9.9 we get with W = V that W ∗∗ (1 − p)V ∗∗ (p) = 0 = V ∗∗ (1 −
p)W ∗∗ (p)
Thus T W ∗∗ (p) = V ∗∗ (p)W ∗∗ (p), V ∗∗ (p)T = V ∗∗ (p)W ∗∗ (p) and T V ∗∗ (p) =
W ∗∗ (p)V ∗∗ (p). Thus T W ∗∗ (p) = V ∗∗ (p)T . It follow that T W (a) = V (a)T for all
a ∈ A.


Lemma A.9.13. Let ϕ : A → B an orthogonality preserving map in sense of

Definition A.9.1. Then the restriction ψ : C ∗ (A, 1A∗∗ ) → B ∗∗ of ϕ∗∗ : A∗∗ → B ∗∗
is again an orthogonality preserving map from C ∗ (A, 1A∗∗ ) = A + C · 1A∗∗ into
B ∗∗ .

Proof. We write here 1 for the unit element 1A∗∗ of A∗∗ . If A is unital then
nothing is to prove. If A is not unital then A + C · 1 = C ∗ (A, 1) ⊆ A∗∗ is natural
isomorphic to the standard unification of A.
It implies: If x∗ = x and y ∗ = y elements of A and α, β ∈ R such that
(x + α1)(y + β1) = 0, then necessarily αβ = 0, and xy = yx.
If α = β = 0 then xy = 0 and ψ(x)ψ(y) = ϕ∗∗ (x)ϕ∗∗ (y) = ϕ(x)ϕ(y) = 0,
because ϕ preserves orthogonality.
 9. PROJECTIVITY AND C.P.C. ORDER ZERO MAPS 1117

By symmetry of the cases α 6= 0 and β 6= 0 for αβ = 0, it suffices to consider

the case α = 0 :
Let z := −β −1 y. It gives zx = x = xz, i.e., (1 − z)x = 0.
If ϕ∗∗ (1 − z)ϕ∗∗ (x) = 0, then ϕ(x)(ϕ(y) + βϕ∗∗ (1)) = 0, because βϕ∗∗ (1 +
β −1 y) = ϕ(y) + βϕ∗∗ (1) is equivalent to ϕ∗∗ (x)ϕ∗∗ (1 + β −1 y) = 0.
Suppose that zx = x, x∗ = x and z ∗ = z. Then also xz = x, x+ = zx+ and
1/n
x− = zx− , because for example x(x+ ) → x+ , and, in the same way, x− = zx− .
A similar argument shows that z+ x+ − z− x+ = zx+ = x+ implies z− x+ = 0
and z+ x+ = x+ . If we take here x− in place of x+ , then we get z− x− = 0 and
z+ x− = x− .
It follows that (z+ − 1)+ x+ = 0 from (z+ − 1)x+ = 0, because (z+ − 1) and x+
commute. (In general, if a selfadjoint element is orthogonal to a positive element,
then its positive and negative parts are also orthogonal to the positive element.)
z+ x = x, (z+ − (z+ − 1)+ )x = x, 0 ≥ v := 2(z+ − (z+ − 1/2)+ ) ≤ 1 satisfies
v(z+ − (z+ − 1)+ ) = (z+ − (z+ − 1)+ ), vx = x and (1 − v)1/2 x = 0 in A∗∗
...
Positive contraction eτ ∈ A+ suitably chosen ...
Need: v + (1 − v)1/2 eτ (1 − v)1/2 ∈ A+ converges in A∗∗ to 1 in σ(A∗∗ , A∗ )
topology.
Since ψ is linear, ...


Corollary A.9.14. Let A and B C*-algebras, ϕ : A → B an orthogonality

preserving

My blue Remarks:
Let ϕ : A → B a 2-positive and orthogonality preserving map.
Question:
Does there exists a C *-algebra morphism h : C0 (0, 1] ⊗ A → B and an element
gϕ ∈ C0 (0, 1]+ such that ϕ(a) = h(g ⊗ a) for all a ∈ A.
(Is there a certain uniqueness for such h and g?)
If A is not unital or B is not unital, then take ϕ∗∗ : A∗∗ → B ∗∗ .
(It is here not clear if ϕ∗∗ is still orthogonality preserving, – except for “open”
projections in A∗∗ corresponding to hereditary C *-subalgebras of A. But ϕ∗∗ maps
selfadjoint elements to selfadjoint elements, respectively is positive, 2-positive or
c.p. if ϕ has this properties.)
One has now to check if the (again 2-positive) restriction ϕ∗∗ |C ∗ (A ∪ {1A∗∗ })
of ϕ∗∗ is again orthogonality preserving:
Let a1 , a2 ∈ A selfadjoint, α1 , α2 ∈ R and suppose that A is not unital. If
x := a1 + α1 · 1A∗∗ and y := a2 + α2 · 1A∗∗ are “orthogonal” in the sense xy = 0,
1118 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

then at least one of α1 and α2 is equal to 0. Therefore, up to change of indices

and multiplication by a non-zero real number, this reduces all to the cases where
a1 a2 = 0 and a1 a2 +a1 = a1 (a2 +1) = 0. Then a1 a2 = 0 leads to ϕ(a1 )ϕ(a2 ) = 0 by
using the assumption that ϕ preserves orthogonality of selfadjoint elements. The
case a1 a2 = −a1 induces that a1 and a2 are commuting selfadjoint elements and
that e∗ = e := −a2 is a local unit for a1 .
Reduces to the case of A0 := C ∗ (a, e) with ae = a, a∗ = a, e∗ = e, ϕ0 : A0 → B
linear, bounded, involution and orthogonality preserving. C := A0 + C · 1 ⊆ A∗∗
(where 1 denotes here the unit element 1A∗∗ of A∗∗ 0 ).

NEED that ϕ0 (a)(ϕ∗∗ (1C )) = (ϕ∗∗ (1C ))ϕ0 (a) and

???????????????????????
Let T := ϕ∗∗ (1A∗∗ ) ∈ B ∗∗ .
Does T commute element-wise with ϕ(A)?
Suppose (!!!) we have found a C *-Jordan morphism ψ : A → {T }0 ∩ B ∗∗ with
ϕ(a) = T · ψ(a), and suppose that ϕ is 2-positive on A.
Can we deduce that ψ : A → B ∗∗ is 2-positive if ϕ is 2-positive?
It would be enough to show that

ψδ (a) := Tδ ϕ(a)Tδ

for Tδ := (T + δ1B ∗∗ )−1/2 is 2-positive for all δ ∈ (0, kT k/2).

(THIS ARE THE KEY QUESTION!)
Since ψ(A) commutes ???????
It should reduce to the study of the images of the support projection P ∈ A∗∗
of positive elements in the Pedersen ideals of A:
They should have the important property ϕ∗∗ (1 − P )ϕ∗∗ (P ) = 0. It is then the
key for all other study!
(Seems to be OK at least in case of positive maps ϕ, because if a, b ∈ Ped(A)+
are contractions with ab = a and b ≤ P , then (1−b)a = 0 and a ≤ P , (1−P ) ≤ 1−b.
... )
To obtains that ϕ∗∗ (1 − b)ϕ∗∗ (a) = 0 we have to check if (1 − b) the σ(A∗∗ , A∗ )
- limit of elements (1 − b)1/2 c(1 − b)1/2 , where c ∈ Ped(A)+ with kck ≤ 1.
All this should work also in case of self-adjoint non-positive case of ϕ : A → B.
Precise estimates in self-adjoint case?
This could give that ϕ∗∗ (P ) = T · Φ(P ) with Φ(P )2 = Φ(P ) = Φ(P )∗ , where
T := ϕ∗∗ (1A∗∗ ) ∈ B ∗∗ is a *-preserving map (self-adjoint maps).
It seems that the Φ can be integrated to to a C *-Jordan morphism from A to
∗∗
B .
It is not clear if Φ(A)B ⊆ B ...(likely not!)
 9. PROJECTIVITY AND C.P.C. ORDER ZERO MAPS 1119

Is this new bounded linear map again orthogonality preserving on self-adjoint

elements?
(Seems to be by the above outlined decomposition for the images of the “com-
pact” projections!)
General observations:
A C *-Jordan morphism ψ : A → C is a linear map with ψ(a∗ ) = ψ(a)∗ for all
a ∈ A, and satisfies ψ(ib) = iψ(b) and ψ(b2 ) = ψ(b)2 for all b ∈ A+ . In particular,
ψ is a positive linear map: ψ(A+ ) ⊆ C+ .
Then ψ(b2 ) = ψ(b)2 for all b∗ = b ∈ As.a. . As.a. is a real Jordan algebra
with [a, b] := (1/2) · (ab + ba) for a, b ∈ As.a. , and ψ|As.a. is a real Jordan algebra
morphism from As.a. into Cs.a. , because ab + ba = (a + b)2 − (a2 + b2 ). Clearly ψ
is determined by its restriction to As.a. = A+ − A+ .
Lemma A.9.15. If ψ : A → C is a C*-Jordan morphism then properties (i),
(ii) and (iii) are equivalent:

(i) ψ is a C*-morphism.
(ii) ψ(a∗ a) = ψ(a)∗ ψ(a) for all a ∈ A.
(iii) ψ(a∗ a) ≥ ψ(a)∗ ψ(a) for all a ∈ A.

Proof. The implications (i)⇒(ii)⇒(iii) are obvious.

(iii)⇒(ii): Always ψ(a∗ ) = ψ(a)∗ for C *-Jordan morphisms ψ and
ψ(a)∗ ψ(a) + ψ(a)ψ(a)∗ = ψ(a∗ a + aa∗ ) = ψ(a∗ a) + ψ(aa∗ )
Suppose ψ(a∗ a) ≥ ψ(a)∗ ψ(a) and ψ(aa∗ ) ≥ ψ(a)ψ(a)∗ , then this all together im-
plies that the sum of the two positive elements ψ(a∗ a) − ψ(a)∗ ψ(a) and ψ(aa∗ ) −
ψ(a)ψ(a)∗ is zero. Thus, ψ(a∗ a) = ψ(a)∗ ψ(a) and ψ(aa∗ ) = ψ(a)ψ(a)∗ .
(ii) ⇒ (i): For C *-Jordan morphisms ψ holds in general
ψ(h)ψ(k) + ψ(k)ψ(h) = ψ(hk + kh) = ψ(hk) + ψ(kh)
by using the equations ψ((h + k)2 ) = (ψ(h) + ψ(k))2 , ψ(h2 ) = ψ(h)2 and ψ(k 2 ) =
ψ(k)2 . It delivers the general formula:
ψ(h)ψ(k) + ψ(k)ψ(h) = ψ(hk) + ψ(kh).
Now use ψ(a∗ a) = ψ(h)2 + ψ(k)2 + i(ψ(hk − kh)) and
ψ(a)∗ ψ(a) = ψ(h)2 + ψ(k)2 + i(ψ(h)ψ(k) − ψ(k)ψ(h))
The property ψ(a∗ a) = ψ(a)∗ ψ(a) and that ψ(hk − kh) = ψ(hk) − ψ(kh) imply
the equation
ψ(h)ψ(k) − ψ(k)ψ(h) = ψ(hk) − ψ(kh) .
Add this to the general formula for C *-Jordan morphisms (obtained by pentagon
rule): ψ(h)ψ(k) + ψ(k)ψ(h) = ψ(hk) + ψ(kh) and get 2ψ(h)ψ(k) = 2ψ(hk).
It says that ψ is multiplicative. Thus ψ : A → C is a C *-algebras morphism. 

Remark: Let T : A → C a C-linear map with kT k ≤ 1, T ∗∗ (1A∗∗ ) = 1C ∗∗ then

T is positive.
1120 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

Lemma A.9.16. Let A and C C*-algebras, T : A → C a positive linear map

(i.e., T is linear and T (A+ ) ⊆ C+ ).
Then kT k = kT ∗∗ (1A∗∗ )k and T (a)2 ≤ kT k · T (a2 ) for all a∗ = a ∈ A.
If, moreover, T is 2-positive, in the sense that

T2 := T ⊗ IdM2 : A ⊗ M2 → C ⊗ M2

is positive, then, for each a ∈ A,

T (a)∗ T (a) ≤ kT k · T (a∗ a) .

If T is positive and unital and T ⊗idM2 has norm ≤ 1 on A⊗M2 , then T ⊗idM2
is positive.
Is generally a unital hermitian contraction T automatically positive:
Restrict T to C ∗ (1, a) for a ∈ A+ , kak ≤ 1. Take pure state ρ on C with
ρ(T (a)) = min{t ∈ Spec(T (a))}. Then ρ · T is a unital linear functional on A with
norm ≤ 1.
(Is equivalent for positive T ∗∗ , to decompose T ∗∗ as T ∗∗ = GS(·)G with
G := T ∗∗ (1A∗∗ )1/2 for some suitable positive S : A∗∗ → C ∗∗ with S(1) = sup-
port projection of G. Then require (!) that S2 := S ⊗ idM2 has norm = 1. This
requirement is equivalent to the 2-positivity of T .)
(The 2-positivity for positive T could be also equivalent to

Z · diag(T (a1,2 ), T (a2,1 )) ≤ diag(T (a11 ), T (a22 )) .

Here Z ∈ M2 (M(A)) has entries [1 − δj,k ], and [aj,k ] ∈ M2 (A)+ .

This should be equivalent to k(1/n + T (a1,1 ))−1 T (a1,2 )(1/n + T (a2,2 ))−1 k ≤ 1
and k(1/n + T (a2,2 ))−1 T (a2,1 )(1/n + T (a1,1 ))−1 k ≤ 1 ...

Proof. Consider in place of A the commutative C *-algebra C ∗ (a). Then

T |C ∗ (a) is the restriction of the completely positive map S := (T |C ∗ (a))∗∗ from
C ∗ (a)∗∗ ⊆ A∗∗ . Let P ≤ 1A∗∗ denote the unit-element of C ∗ (a)∗∗ . Passage
to linear sums of projections show that S is c.p. with norm kSk = kS(P )k ≤
kT ∗∗ (1A∗∗ )k ≤ kT k and therefore T (a)2 = S(a)2 ≤ kS(P )kS(a2 ) ≤ kT kT (a2 ) .
The map T2 := T ⊗ idM2 has norm kT2 k = kT ⊗ 12 k = kT k if T2 is again
positive (by the additional assumption). We can build for a ∈ A the selfadjoint
element b∗ = b := [ajk ] in M2 (A) ∼
= A⊗M2 with zero diagonal entries b11 = b22 = 0
and off-diagonal entries b12 := a and b21 := a∗ .
Can now apply the general formula for positive maps and self-adjoint elements:
Get T2 (b)2 ≤ kT kT2 (b2 ) by positivity of T2 and kT2 k = kT k. The upper left entries
show that T (a∗ )T (a) ≤ kT k · T (a∗ a). 

The following Proposition is a straight conclusion of above Lemmata A.9.15

and A.9.16.
 10. THE “SOCLE” OF A C *-ALGEBRA. 1121

Proposition A.9.17. Let A and C C*-algebras and ψ : A → C a C*-Jordan

morphism.
Then ψ is a C*-algebra morphism, if and only if, ψ is a 2-positive contraction,
i.e., the algebraic tensor product

ψ2 := ψ ⊗ idM2 : A ⊗ M2 → C ⊗ M2

is a positive contraction.

Suppose we have a C *-Jordan morphism (or a C *-morphism) π : A → B ∗∗ and

T ≥ 0 in B ∗∗ such that π(A) ∈ {T }0 ∩ B ∗∗ and T · π(a) = ϕ(a) ∈ B.
The C *-Jordan morphism π is 2-positive (and therefore is a C *-algebra mor-
phism) if ϕ is 2-positive, because π(·) is point-wise limit of (T + 1/n)−1/2 ϕ(·)(T +
1/n)−1/2 for n → ∞, and the maps b ∈ B 7→ (T + 1/n)−1/2 b(T + 1/n)−1/2 are
completely positive.
(Is important to know, for my version of a complete proof!).
??? Let D ⊆ B the hereditary C *-subalgebra generated by ϕ(A), i.e., D :=
ϕ(A)Bϕ(A), then it is likely that T ∈ M(D) (modulo annihilator of D?) and
π(a) ∈ M(D) for all a ∈ A. ?????

10. The “socle” of a C *-algebra.

Recall that the “socle” of Banach algebra A is the (algebraic) ideal of A gen-
erated by all elements a ∈ A with aAa of finite linear dimension. In case of
C *-algebras A the socle(A) it is an algebraic *-ideal generated by all projections
p∗ = p2 = p ∈ A with pAp of finite dimension.
The closure of the socle of a C *-algebra A is the largest closed ideal of A
that is a c0 -sum of “elementary” C *-algebras ∼
= K(H) with H of finite or infinite
dimensions.

Lemma A.10.1. Let ρ a pure state on A with ρ(socle(A)) = {0} and a ∈ A+

with kak = 1 = ρ(a), then, for every δ > 0 and n ∈ N, in the commutant of
(a − (1 − δ))+ in (a − (1 − δ))+ A(a − (1 − δ))+ contains pairwise orthogonal positive
elements a1 , . . . , an with kai k = 1.

Notice that the element a ∈ A+ could be a projection a = p and that then

(a − (1 − δ))+ = δp for all δ ∈ [0, 1] .

Proof. If 1 is not isolated in the spectrum of a, then a1 , . . . , an exist by

functional calculus applied to (a − (1 − δ/2))+ . If 1 is isolated, then there are
γ > 0, b ∈ A+ and a projection q ∈ A with bq = 0, a = b + q, kbk = 1 − γ . Then
(a − (1 − δ))+ A(a − (1 − δ))+ = qAq for 0 < δ < γ . Since also ρ(a2 ) = 1, we get
ρ(b(1 − b)) = ρ(a − a2 ) = 0, ρ(b) = 0 and ρ(q) = 1 . It follows that q 6∈ socle(A).
Thus any maximal commutative C *-subalgebra C of qAq is infinite-dimensional,
i.e., for every n ∈ N we find non-zero orthogonal a1 , . . . , an ∈ C+ . 
1122 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

Remark A.10.2. Let J := socle(A) and D a hereditary C *-algebra of A, then

D ∩ socle(A) = socle(D) and D ∩ J = socle(D).
Indeed: The elements a ∈ socle(A) are characterized by the property that a∗ Aa
is finite-dimensional. It implies D socle(A)D ⊆ D ∩ socle(A) = socle(D). And it
follows socle(D) ⊆ DJD = J ∩ D and DJD ⊆ D socle(A)D ⊂ socle(D).

Lemma A.10.3. Let J := socle(A), and ψ : C0 ((0, 1], Mn ) → A a C*-morphism

with kπJ ◦ ψ(f0 ⊗ e1,1 )k = 1, D the hereditary C*-subalgebra of A generated by
ψ(f0 ⊗ 1), where f0 (t) = t for t ∈ [0, 1].
Then there exist an increasing continuous map λ : [0, 1] → [0, 1] with λ(0) = 0
and λ(1) = 1, and a C*-morphism ϕ : C0 ((0, 1], Mn ) → D with ϕ(f ◦ λ) = ψ(f )
for f ∈ C0 ((0, 1], Mn ), such that 1 is not isolated in the spectrum of ϕ(f0 ⊗ 1)
and ψ(f0 (1 − f0 ) ⊗ 1n ) is contained in the hereditary C*-subalgebra E generated by
ϕ((f0 (1 − f0 ) ⊗ 1n ). Moreover ϕ(g − (g ◦ λ)) ∈ E for g ∈ C0 ((0, 1], Mn ).

Proof. We take ϕ := ψ and λ := f0 if 1 is not isolated in Spec(ϕ(f0 ⊗ 1)).

Thus we may suppose that 1 is isolated in Spec(ϕ(f0 ⊗ 1n )).
First we consider the case where n = 1 and where e := ψ(f0 ) is a strictly
positive element of A:
Then there is µ ∈ (0, 1) such that the hereditary C *-algebra F of A generated by
(e − µ)+ is unital with unit p := 1F = (1 − µ)−1 (e − µ)+ . Then pe = p = pe and
k(1 − p)ek ≤ µ. Since kπJ (e)k = 1, it follows that p is not in J. Let C ⊆ F a
maximal commutative C *-subalgebra of F . The compact Hausdorff space X :=
Prim(C) can’t be finite, because p ∈ C \ J and the minimal idempotents of C are
in the socle of F = pAp and, therefore, are in J. Moreover, by the same argument,
if the open subset U of X := Prim(C) corresponding to the ideal C ∩ J of C is
closed in X, then the open and closed set Y := X \ U (that is homeomorphic to
Prim(C/J ∩C)) is not finite. Thus, in any of this cases, there is a point x0 ∈ Y that
is not isolated in X. It allows to find a function h : Prim(C) → [0, (1 − µ)/2] with
h(x0 ) = 0 such that there is a sequence xn ∈ X with 0 < h(xn ) and h(xn ) → 0. This
means that 0 ∈ Spec(πJ (g)), and that 0 is not isolated in Spec(g) ⊆ [0, (1 − µ)/2]
if g ∈ C+ has Gelfand transform gb = h.
Now let a := e(1 − p) + (p − g) = e − g ∈ A+ . Then µe ≤ a ≤ e , kπJ (a)k =
kπJ (p) − πJ (g)k = 1, and 1 is not isolated in Spec(a). Moreover, e(1 − e) =
(1 − p)(e − e2 ) ≤ e(1 − e) + (p − g) − (p − g)2 = a(1 − a) , i.e., e(1 − e) is contained
in the hereditary C *-subalgebra that is generated by a(1 − a). We denote by ϕ
the (unique) C *-morphism ϕ : C0 (0, 1] → A with ϕ(f0 ) = a and define a pice-wise
linear continuous function λ : [0, 1] → [0, 1] by λ(t) := t for t ∈ [0, µ], λ(t) := 1 for
t ∈ [(µ + 1)/2, 1] and λ|[µ, (µ + 1)/2] linear. Then ϕ(λ) = λ(a) = e = ψ(f0 ), and
ϕ(f ◦ λ) = f ◦ λ(a) = f (e) = ψ(f ) for all f ∈ C0 (0, 1].
In the general case we let D denote the hereditary C *-subalgebra D of A that
is generated by ψ(f0 ⊗ 1n ) ∈ A+ , and let A1 ⊆ D denote the hereditary C *-
subalgebra of A that is generated by ψ(f0 ⊗ e1,1 ). Then ψ1 (f ) := ψ(f ⊗ e1,1 )
 11. C *-VERSION OF COHEN FACTORIZATION 1123

defines a C *-morphism from C0 (0, 1] into A. There is a natural isomorphism θ

from A1 ⊗ Mn onto D such that θ(ψ1 (f ) ⊗ α) = ψ(f ⊗ α) for all f ∈ C0 (0, 1] and
θ(a ⊗ e1,1 ) = a for all a ∈ A1 . Since D ∩ J = socle(D) and A1 ∩ J = socle(A1 ),
the C *-morphism ψ1 : C0 (0, 1] → A1 satisfies the assumptions for the case n = 1 .
With ϕ1 : C0 (0, 1] → A1 and λ : [0, 1] → [0, 1] constructed as above, we have that
ϕ := θ ◦ (ϕ1 ⊗ idn ) and λ have the required properties. (Notice here that λ(1) = 1
and λ(0) = 0 implies g − (g ◦ λ) ∈ C0 (0, 1) ⊗ Mn for all g ∈ C0 ((0, 1], Mn ).) 

11. C *-version of Cohen factorization

The following theorem is a C *-version of the Cohen factorization theorem,

given by G.K. Pedersen in [621, Thm. 4.1] with a proof on 13 lines. We give a more
detailed proof. Only elementary knowledge on functional calculus and approximate
units of C *-algebras is used.

Theorem A.11.1. Suppose that a C*-algebra A acts on the Banach space X

from left with ka · xk ≤ kakkxk and span(A · X) dense in X.
Then for given ε > 0 and x ∈ X, there exist y ∈ {a · x ; a ∈ A+ } and e ∈ A+
with e · y = x, kek ≤ 1 and ky − xk < ε.

Proof. Since the span of A · X is dense in X, any approximate unit (uλ ) for
A will converge strongly to the operator 1 in L(X). We let A e := A + C1 be the
unitized C *-algebra acting on X, and by induction (setting a0 := 1 and x0 := x)
we define sequences (an ) and (xn ) in A
e+ and X, respectively, by

an := an−1 − 2−n (1 − un ) , xn := a−1

n · x.

It is easy to verify (by induction) that an ≥ 2−n 1 (so that k2−n a−1
n k ≤ 1), and that
dist(an , A) = 2−n .
Moreover
−1
xn − xn−1 = a−1
n (an−1 − an )an−1 · x = 2
−n −1
an (1 − un ) · xn−1 .

Having chosen the un ’s properly we can therefore assume that kxn − xn−1 k < 2−n ε
for all n.
Let e := lim an and y := lim xn . Since an · xn = x for all n, we have ey = x.
Moreover, kx − yk < ε by construction. Finally, e ∈ A+ and kek ≤ 1. In fact,
e = n≥1 2−n un .
P


The additional condition y ∈ {a · x ; a ∈ A+ } is not mentioned in the orig-

inal formulation of [621, Thm. 4.1], but can be seen from its proof, or can be
shown from the decomposition of x = e · y and combination with the corresponding
decomposition e0 f = e of e with f ∈ C ∗ (e)+ e and a contraction e0 ∈ C ∗ (e)+ .
Of course, Theorem A.11.1 works also for non-degenerate right-actions: Con-
sider the left multiplication a ·` x := a · x by a ∈ Aop , where Aop denotes the
opposite C *-algebra of A, i.e., A with opposite multiplication.
1124 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

An easy separation argument and the fact that {e ∈ A+ ; kek < 1} is an

approximate unit for A show that the non-degeneracy condition span(A · X) = X
implies that each x ∈ X is contained in the closure of {a · x ; a ∈ A+ , kak < 1}.
In particular, A · x = {0} implies x = 0 if X is a non-degenerate Banach
A-module.
It allows to extend the left-multiplication (a, x) ∈ A × X 7→ a · x ∈ X uniquely
to a left multiplication (b, x) ∈ M(A) × X → X that is strictly continuous with
respect to b ∈ M(A).
Since one can replace the Banach module X by c0 (X) in Theorem A.11.1, this
Theorem implies the following formally stronger result:
For every finite sequence x1 , . . . , xn in X and ε > 0 there exists a positive
contraction e ∈ A+ and y1 , . . . , yn ∈ X with yk ∈ {a · xk ; a ∈ A+ } such that
e · yk = xk , kyk − xk k < ε and kek ≤ 1.

12. On centralizers in asymptotic coronas

The discrete version of asymptotic coronas for a C *-algebra A is A∞ :=

`∞ (A)/c0 (A).

Lemma A.12.1. The algebra F∞ (C, A) := (C 0 ∩ A∞ )/ Ann(C, A∞ ) is a unital

C*-algebra if C is any σ-unital C*-subalgebra of A∞ .

For a better understanding what is going on, notice here that an arguments in
[448] can be modified to obtain the stronger result for separable C *-subalgebras
C ⊆ A∞ and D := C(A∞ )C that there exists a natural isomorphism
(C 0 ∩ A∞ )/ Ann(C, A∞ ) ∼
= C 0 ∩ M(D) .

Proof. If e ∈ C+ is a strictly positive contraction then obviously C ⊆ D :=

e · A∞ · e = C · A∞ · C ,
D0 ∩ A∞ ⊆ C 0 ∩ A∞ ⊆ {e}0 ∩ A∞
and Ann({e}, A∞ ) = Ann(D, A∞ ) = Ann(C, A∞ ) . (We can suppose that kek = 1,
but this is not important.) Then e = (e1 , e2 , . . .)+c0 (A) with contractions en ∈ A+ .
Let (g1 , g2 , . . .) a sequence of positive contractions with gn ∈ C ∗ (en ) and ken −
gn en k < 1/n.
Then g := (g1 , g2 , . . .) + c0 (A) satisfies 0 ≤ g ≤ 1, g − g 2 ∈ Ann({e}, A∞ )+
and g ∈ D0 ∩ A∞ . The latter because gd = d = dg for all d ∈ D, i.e., the
positive contraction d is a “local” unit for all elements in D ⊇ C. It follows that
g + Ann({e}, A∞ ) is a projection in (D0 ∩ A∞ )/ Ann(D, A∞ ).
Thus, g is a positive contraction in D0 ∩ A∞ ⊆ C 0 ∩ A∞ with gc = c = cg for
all c ∈ C and g + Ann(C, A∞ ) is a projection in (C 0 ∩ A∞ )+ / Ann(C, A∞ ) .
If h ∈ C 0 ∩ A∞ and c ∈ C then (h − gh)c = 0 = c(h − gh) because ghc = gch =
ch = hc and cgh = ch, i.e., (h − gh) ∈ Ann(C, A∞ ) .
 12. ON CENTRALIZERS IN ASYMPTOTIC CORONAS 1125

We obtain gh, hg ∈ C 0 ∩ A∞ for h ∈ C 0 ∩ A∞ and h − gh, h − hg, g − g 2 ∈

Ann(C, A∞ ) in a similar way. It implies that g + Ann(C, A∞ ) is the unit element
of (C 0 ∩ A∞ )/ Ann(C, A∞ ). 

Remark A.12.2. Let X a Polish l.c. space, that is not compact, and let ω ∈
γX := βX \ X a point of the corona γX of X.
Recall that Cb (X, B) and Q(X, B) := Cb (X, B)/ C0 (X, B) are C *-bundles over
βX respectively γX := βX \X, i.e., are algebras of continuous section of continuous
fields of C *-algebras Bω (if ω ∈ γX) and Bx = B (if x ∈ X).
∼ C(βX) and Cb (X)/ C0 (X) ∼
In particular, Cb (X) = = C(γX). Cb (X, B) is
a C(βX)-algebra, and Q(X, B) is a C(γX)-algebra. This explains the notations
Cb (X, B)|Y and Q(X, B)|Z for closed subsets Y ⊆ βX and Z ⊆ γY . We write
also Bω for Q(X, B)|{ω} if ω ∈ γX.
Let A ⊆ Q(X, B) a separable C *-subalgebra. We define

F (X; A, B) := (A0 ∩ Q(X, B))/ Ann(A, Q(X, B)) .

It is always a unital algebra that contains a copy of C(γX) in its center.

(Indeed : Clearly, A0 ∩ Q(X, B) is a C(γX)-algebra. Since Ann(A, Q(X, B)) is
a hereditary C *-subalgebraof Q(X, B), it is a C(γX)-algebra. An approximate unit
of a separable C *-subalgebra C ⊆ Cb (X, B) with C|γX = A allows to construct a
positive contraction f ∈ Cb (X, B)+ with kf (x)k = 1 for each x ∈ X and ea = a =
ae for all a ∈ A, where e := f |γX.)
Let DA := A Q(X, B)A denote the σ-unital C *-subalgebra of Q(X, B) that is
generated by A, and let

N (DA ) := { f ∈ Q(X, B) ; f DA ∪ DA f ⊆ DA } .

Then (obviously) A0 ∩ Q(X, B) ⊆ N (DA ), and the natural C *-morphism from

A0 ∩ Q(X, B) into M(DA ) has kernel Ann(A, Q(X, B)) = Ann(DA , Q(X, B)) . The
image is contained in A0 ∩ M(DA ). It is not hard to show that it is really all
of A0 ∩ M(DA ), because Ann(A, Q(X, B)) ⊆ A0 ∩ Q(X, B) and the natural *-
monomorphism
N (DA )/ Ann(A, Q(X, B)) → M(DA )
is surjective. The latter can be deduced from the facts that A contains a strictly
positive contraction e of DA , and each element of M(DA )+ is the sum T + S + a,
where a ∈ A, and T, S ∈ M(DA )+ are in M(DA ) strictly convergent sequences of
mutually orthogonal positive elements of A.
Thus, the natural *-isomorphism from N (DA )/ Ann(A, Q(X, B)) → M(DA )
defines a natural C(γX)-algebra isomorphism

F (X; A, B) ∼
= A0 ∩ M(DA ) .

This isomorphism allows to see, that F (X; ·, ·) is a “stable” invariant, i.e., F (X; A⊗
K, B ⊗K) is naturally C(γX)-module isomorphic to F (X; A, B). In fact this follows
1126 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

easily from the natural C(γX)-module isomorphisms

(DA ) ⊗ K ∼
= D(A⊗K)

and
(A ⊗ K)0 ∩ M(DA ⊗ K) ∼
= A0 ∩ M(DA ) .

Let πY : Q(X, B) → Q(X, B)|Y = Cb (X, B)|Y denote the natural C(γX)-
modular epimorphism, and denote by D|Y the image πY (D) ⊆ Q(X, B)|Y of a
subset D ⊆ Q(X, B), in particular, f |Y := πY (f ).
We generalize F (X; A, B) to closed subsets Y ⊆ γX :

F (X, Y ; A, B) := ((A|Y )0 ∩ (Q(X, B)|Y ))/ Ann((A|Y ), (Q(X, B)|Y )) .

Again one can show that F (X, Y ; A, B) is a stable invariant in the above consid-
ered sense. Our old definition of F (X; A, B) becomes F (X, γX; A, B) in the new
terminology.
The map πY maps Ann(A, Q(X, B)) into Ann((A|Y ), (Q(X, B)|Y )), and maps
A ∩ Q(X, B) into (A|Y )0 ∩ (Q(X, B)|Y ).
0

It is not clear, for which Y one has (A0 ∩ Q(X, B))|Y = (A|Y )0 ∩ (Q(X, B)|Y )
see Questions A.12.3.
We have Ann(A, Q(N, B))|Y 6= Ann((A|Y ), (Q(N, B)|Y )) for B := C and Y =
{ω} ⊆ γN a suitable free ultrafilter and A = C ∗ (f ) for non-zero non-negative
f ∈ Q( C) = `∞ (N)/c0 (N) with 0 ∈ Spec(f ) not isolated and ω ∈ γN a boundary
point of the support of f .
It follows, that πY does not necessarily define an isomorphism from
F (X; A, B)|Y onto F (X, Y ; A, B). In particular, the fibers F (X; A, B)|ω of
the C(γX)-algebra can not be identified with F (X, {ω}; A, B).
The situation is less complicate if A ⊆ B. In particular, the algebras
F (X, Y ; A, A) are stable invariants that have fibers F (X, {ω}; A, B) with ω ∈ Y .

Questions A.12.3. Let A0 ∩ Q(X, B) and F (X; A, B) as in Remark A.12.2.

Here are some questions related to a possible or impossible reduction of prop-
erties of asymptotic invariants F (X; A, B) to properties of its fibers F (ω; A, B):
When (A0 ∩ Q(X, B))|Y = (A|Y )0 ∩ (Q(X, B)|Y ) ?
What are the fibers F (X; A, B)ω of F (X; A, B) ?
Is it clear from the definition of F (X; A, B) that F ({ω}; A, B) is a quotient of
F (X; A, B)|{ω} ?
Let a, b ∈ Q(X, B) self-adjoint contractions with commuting πω (a) and πω (b).
Does there exist commuting self-adjoint a0 , b0 ∈ Q(X, B) with πω (a0 ) = πω (a) and
πω (b0 ) = πω (b) ?
Let f, g ∈ Cb (X, B) self-adjoint contractions with a = f + C0 (X, B) and b =
g + C0 (X, B). When we can find self-adjoint f0 ∈ C0 (X, B) and a continuous
 12. ON CENTRALIZERS IN ASYMPTOTIC CORONAS 1127

function µ : X → [1, ∞) such that, for f1 = f + f0 ,

Z µ(x)
h(x) := (2µ(x))−1 eitf1 (x) g(x)e−itf1 (x) dt
−µ(x)

has the property πω (b0 ) = πω (b) and ab0 = b0 a for b0 := h + C0 (X, B) ?

Is there a “universal” continuous function τ : R+ → R+ with τ (0) = 0, such
that τ is increasing and has the following property?
If a, b ∈ C+ are contractions with kabk < ε, then, for every δ > 0, there is a
contraction d ∈ C with d(a − δ)+ = 0 and kdb − bdk + kd∗ bd − bk ≤ τ (ε).
Continuous maps into canonical commutation relations could give counterex-
amples?

Remark A.12.4. Let ω ∈ β(N), f ∈ `∞ (N)+ with f (ω) = 0. It is not difficult

to see that there is a sequence S = {n1 , n2 , . . .} ⊆ N (where nk < nk+1 ), such that
limk f (nk ) = 0, i.e., f |S ∈ c0 (S).
(On finds a sequence n1 < n2 < . . . in N with nk ∈ Xk := f −1 [0, 1/k] ⊆ N,
because each Xk is an infinite subset of N.)
A subset S with f |S ∈ c0 (S) can be found in the ultrafilter defined by ω, i.e.,
with χ(S)(ω) = 1 , if and only if, ω ∈ γN is in the interior of βN ∩ f −1 (0).

Remark A.12.5. Suppose that X and Y are non-compact σ-compact locally

compact spaces, that X is finite-dimensional, and that ψ : X → Y is a proper
continuous map.
Then ψ extends naturally to a continuous map βψ : βX → βY with βψ(γX) =
βψ(βX) ∩ γY . Take ω ∈ γX := βX \ X and let ω 0 := βψ(ω).
Now let B a C *-algebra, and S ⊆ CPin (B) a convex set of approximately inner
completely positive contractions, such that S is closed in point-norm topology.
If, for each (point-norm) continuous map x ∈ X 7→ Vx ∈ S, the map Vω : Bω →
Bω is ideal preserving, then, for each (point-norm) continuous map y ∈ Y 7→ Wy ∈
S, the map Wω0 : Bω0 → Bω0 is ideal preserving.
Notice that every σ-compact non-compact l.c. space Y has a proper continuous
map ψ : Y → R+ . This leads to the following question.

Question A.12.6. Let ω ∈ β(R+ ) \ R+ , and B a C *-algebra.

Suppose that Vω : Bω → Bω is ideal preserving for every bounded (point-norm)
continuous map t ∈ R+ 7→ Vt ∈ CPin (B).
Does B satisfy the condition (i) of Definition 2.0.4 ?
The problem here is the following: Let n ∈ N 7→ Vn ∈ CPin (B) is a given
sequence of approximately inner c.p. contractions. Can it be extended to t ∈
R+ 7→ Vt ∈ CPin (B) point-norm continuous, with same properties of Vω on Bω for
ω ∈ γ(R+ ) as for Vω with ω ∈ γ(N) ?
1128 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

13. Order unit spaces and pre-ordered semigroups

Order-unit spaces are the spaces of continuous affine functions on compact

convex subsets of locally convex real vector spaces. We use them to derive some
results on the extension of states of pre-ordered abelian semigroups with order unit.
We denote by Q the rational numbers (and not the quaternions) during Section 13.
The results are basic, but they are also known and proved with other or similar
methods a long time ago.

Definition A.13.1. An order unit space is a triple (B, B+ , e), where B is a

real Banach space, B+ is a closed convex cone in B and e ∈ B+ is a distinguished
element of the cone B+ , such that −e 6∈ B+ and for every x ∈ B holds:
The vectors te + x and te − x are both in B+ , if and only if, t ≥ kxk.
B is partially ordered by

x ≤ y ⇔ y − x ∈ B+ .

A state of B is a monotone (i.e., order preserving) R-linear map λ : B → R from

B into R with λ(e) = 1.
We denote by E(B) ⊆ B ∗ the set of states of B equipped with the σ(B ∗ , B)-
topology. Then E(B) is a compact convex set with this topology (cf. Part (ii) of
Proposition A.13.2).

Proposition A.13.2. Let B = (B, B+ , e) and B1 = (B1 , (B1 )+ , e1 ) order

unit spaces (in particular real Banach spaces), and let Y ⊆ B denote an additive
subgroup of B with e ∈ Y .

(i) Closed linear subspaces L of B with e ∈ L are again order unit spaces, if
we let L+ := B+ ∩ L.
(ii) An additive map λ : B → R is positive (i.e., λ(B+ ) ⊆ R+ ), if and only if,
λ is R-linear and bounded with kλk = λ(e).
In particular, the positive states functionals λ ∈ B ∗ with λ(e) = 1 and
kλk ≤ 1, and E(B) is a compact convex subset of B ∗ with *weak topology
(i.e., σ(B ∗ , B)-topology).
(iii) For each x ∈ B there is a state λ with |λ(x)| = kxk. In particular, B is
in a natural way an order unit subspace of C(E(B)).
(iv) For every continuous linear functional f ∈ B ∗ there are (not necessarily
unique) states λ1 , λ2 ∈ E(B) and α, β ∈ R+ with f = αλ1 − βλ2 and
α + β = kf k.
(v) An element x ∈ B is in the interior (B+ )◦ of B+ , if and only if, there is
n ∈ N with e ≤ nx.
(vi) An additive subgroup Y of B with e ∈ Y separates the states of B, if and
S
only if, n (1/n)Y = {(1/n)x ; x ∈ Y } is dense in B.
(vii) Suppose that φ : B → B1 is a unital additive map, (i.e., φ(e) = e1 and
φ(x + y) = φ(x) + φ(y) for all x, y ∈ B), and let Y ⊆ B an additive
subgroup such that {(1/n)x ; x ∈ Y } is dense in B.
 13. ORDER UNIT SPACES AND PRE-ORDERED SEMIGROUPS 1129

The following properties (1)–(5) are equivalent and imply that φ is

R-linear.
(1) φ is monotone (i.e., is order preserving: φ(B+ ) ⊆ (B1 )+ ).
(2) ρ ◦ φ ∈ E(B) for every state ρ ∈ E(B1 ).
(3) φ is a contraction.
(4) φ is continuous, and, for x ∈ Y , the existence of n ∈ N with e ≤ nx
implies the existence of m ∈ N with e1 ≤ mφ(x).
(5) φ((B+ )◦ ) ⊆ (B1 )+ .
(viii) If Y is as in (vii), and the unital additive map φ : B → B1 is monotone,
then the following properties are equivalent:
(a) B+ = φ−1 (φ(B) ∩ (B1 )+ ).
(b) For every λ ∈ E(B) exists ρ of B1 with λ = ρ ◦ φ.
(c) φ is isometric.
(d) For x ∈ Y , the existence of m ∈ N with e1 ≤ mφ(x) implies the
existence of n ∈ N with e ≤ nx.
(e) b ∈ (B+ )◦ , if and only if, φ(b) ∈ ((B1 )+ )◦ .
(ix) Suppose that a subset F ⊆ E(B) is norming for B (or that F is norming
on a rationally dense additive subgroup Y of B in the sense that kyk =
supf ∈F |f (y)| for each y ∈ Y ).
Then the convex hull of F is σ(B ∗ , B)-dense in E(B).
In particular, this implies: If K is a compact convex subset of a locally
convex real vector space V , then the space Aff c (K) ⊆ C(K) of all contin-
uous affine functions f : K → R is an order unit space, and E(Aff c (K))
is naturally isomorphic to K.
(x) (B, B+ , e) is naturally isomorphic to the ordered space Aff c (E(B)) ⊆
C(E(B)) of all real-valued continuous affine functions on the (*weakly)
compact convex subset E(B) ⊆ B ∗ .

Proof. (i): (L, L+ := B+ ∩ L, e) satisfies the axioms of an order unit space,

because the definition is “local”, i.e., requires only the (norm-)completeness of the
space and the cone, and that any (at most) two-dimensional subspace Re + Rx is
an order unit space.
(ii): Since −e 6∈ B+ and e ∈ B+ , it follows that te − e, te + e ∈ B+ , if and only
if, t ≥ 1. Thus, kek = 1.
Let U := {x ∈ B ; kxk < 1} denote the open unit ball of B. The sets δe + B+
(δ > 0) are contained in the interior of B+ , because e + U ⊆ B+ and δe + B+ ⊆
δ(e + U ) + B+ .
Let λ : B → R an additive map with λ((B+ )◦ ) ⊆ [0, ∞). Then λ is Q-linear.
Since re ± x = (r − kxk)e + (kxke − x) ∈ (B+ )◦ for rational r > kxk, we get
0 ≤ λ(re + x) = rλ(e) ± λ(x) for all rational numbers r > kxk. Thus, kλk < ∞ and
λ(e) ≥ kλk . On the other hand, kλk = kλk · kek ≥ λ(e).
1130 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

Let b ∈ B, then b ∈ B+ , if and only if, k(kbke) − bk ≤ kbk. Indeed: If b 6= 0,

then a := kbk−1 b ∈ B+ , if and only if, e − (e − a) = a ∈ B+ and e + (e − a) ∈ B+ ,
if and only if, ke − ak ≤ 1.
Let λ ∈ B ∗ with kλk = λ(e). Then | λ(e)kbk − λ(b) | ≤ λ(e)kbk for all b ∈ B+ ,
which implies that λ(b) ≥ 0.
Since kek = 1, there exists λ ∈ B ∗ with λ(e) = 1, by Hahn-Banach extension,
i.e., E(B) 6= ∅.
The subset E(B) of λ ∈ B ∗ with kλk ≤ 1 and λ(e) = 1 is a *weakly closed
subset of the unit ball (B ∗ )≤1 of B ∗ . The latter is compact in the *weak topology
(:= σ(B ∗ , B)-topology).
(iii): It suffices to consider the case kbk = 1.
Let δ := 2 − max(ke − bk, ke + bk) ≥ 0. Then (1 − δ)e ± b = (2 − δ)e − (e ∓ b) ∈ B+ ,
which implies (1 − δ) ≥ kbk = 1 and δ = 0. It follows that ke − bk = 2 or
ke + bk = 2. We may suppose ke + bk = 2, because we can replace b by (−b). By
Hahn-Banach extension, there is λ ∈ B ∗ with kλk = 1 and λ(e + b) = 2. It follows
that kλk = λ(e) = 1 = λ(b). Thus, λ ∈ E(B) by Part (ii), and kbk = |λ(b)|.
(iv): It suffices to consider the case of f ∈ B ∗ with kf k = 1. Let T (λ1 , λ2 , α) :=
αλ1 − (1 − α)λ2 for λ1 , λ2 ∈ E(B) and α ∈ [0, 1]. Then T is a (*weakly) continuous
map from the compact set K := E(B) × E(B) × [0, 1] into the (*weakly) compact
∗
set B≤1 := {g ∈ B ∗ ; kgk ≤ 1 }. In particular, the image T (K) of T is closed.
Straight calculation shows that T (K) is also convex and T (K) = −T (K). Since
the subset E(B) = T (E(B) × E(B) × {1}) is norming for B, a standard Hahn-
Banach separation argument shows that T (K) = (B ∗ )≤1 . In particular, there are
α, β ∈ [0, ∞) and λ1 , λ2 ∈ E(B) with α + β = kf k = 1 and αλ1 − βλ2 = f .
(v): If x ∈ (B+ )◦ then x − (1/n)e ∈ (B+ )◦ for sufficiently big n ∈ N, in
particular nx − e ∈ B+ . Conversely, if nx − e ∈ B+ , then x = (1/n)e + y with
y := x − (1/n)e ∈ B+ . Thus, x ∈ y + (1/n)(e + U ) ⊆ B+ for the open unit ball U
of B, cf. proof of Part (ii). Hence, x ∈ (B+ )◦ .
(vi): Let λ1 , λ2 ∈ E(B). λ1 (x) = λ2 (x) for all x ∈ Y , if and and only if,
Z := {(1/n)x ; x ∈ Y } is in the kernel of λ1 − λ2 . If Z is dense in B, then λ1 = λ2 ,
and the elements of Y separate the states.
Now let Y an additive subgroup of B with e ∈ Y . Then Z := {(1/n)x ; x ∈
Y } = spanQ (Y ) is norm-dense in spanR (Y ). Thus, if Z is not dense in B, then
there exists f ∈ B ∗ with kf k = 1 and f (x) = 0 for all x ∈ Y . By Part (iv), there
are states λ1 , λ2 ∈ E(B) and α ∈ [0, 1] with f = αλ1 − (1 − α)λ2 . Since e ∈ Y , it
follows that 0 = f (e) = 1 − 2α. Hence, λ1 (x) = λ2 (x) for x ∈ Y and λ1 6= λ2 .
It follows, that Z must be dense in B if Y separates the states.
(vii): The additivity of φ implies that φ(rb) = rφ(b) for all rational numbers
r ∈ Q. Thus, if φ satisfies (3), then φ is R-linear. The implications (1)⇒(5) is
obvious.
 13. ORDER UNIT SPACES AND PRE-ORDERED SEMIGROUPS 1131

(5)⇒(2): The map ρ ◦ φ : B → R is additive and ρ ◦ φ(e) = ρ(e1 ) = 1 for

ρ ∈ E(B1 ). If φ((B+ )◦ ) ⊆ (B1 )+ , then ρ ◦ φ is an additive map from B to R that
is not negative on (B+ )◦ . It follows that ρ ◦ φ is continuous by an argument in the
proof of Part (ii). Then ρ ◦ φ(B+ ) ⊆ [0, ∞) , and Part (ii) shows ρ ◦ φ ∈ E(B).
(2)⇒(3): If ρ ◦ φ ∈ E(B) for every state ρ ∈ E(B1 ), then kφ(b)k =
supρ∈E(B1 ) |ρ ◦ φ(b)| ≤ kbk by Parts (ii) and (iii).
(3)⇒(1): Let x ∈ B+ and m, n ∈ N with m/n > kxk. Then me ± (me −
nx) ∈ B+ and, thus, kme − nxk ≤ m. It follows kme1 − nφ(x)k ≤ m. Thus
nφ(x) = me1 − (me1 − nφ(x)) ∈ (B1 )+ .
(1)⇒(4): Let x ∈ Y and n ∈ N with nx − e ∈ B+ , then nφ(x) − e1 ∈ (B1 )+ .
The map φ is continuous, because (1) implies (3).
(4)⇒(5): Let Z := n (1/n)Y = spanQ (Y ). If z = (1/k)y ∈ Z ∩(B+ )◦ with y ∈
T

Y and k ∈ N, then y ∈ Y ∩(B+ )◦ . By Part (v), there is n ∈ N with ny −e ∈ B+ . By

assumptions, it follows that there is m ∈ N with mφ(y) − e1 ∈ (B1 )+ . In particular,
φ(z) = (1/(mk))((mφ(y) − e1 ) + e1 ) ∈ (B1 )+ . Hence, φ( (B+ )◦ ∩ Z ) ⊆ (B1 )+ .
Since φ is continuous by assumptions, φ maps the closure M of (B+ )◦ ∩ Z into
(B1 )+ . By assumption, Z is dense in B. Thus Z is dense in each open subset of
B, in particular (B+ )◦ ⊆ M .
(viii): Since φ is unital and increasing (i.e., φ(B+ ) ⊆ (B1 )+ ), φ is a unital
contraction by Part (vii). By (v), x ∈ (B+ )◦ , if and only if, there is n ∈ N with
nx − e ∈ B+ . Thus nφ(x) − e1 ∈ (B1 )+ , and φ(x) ∈ ((B1 )+ )◦ , if x ∈ (B+ )◦ . It
shows that, under our assumptions, the properties (a), (c) and (e) are equivalent
to:
(a’) φ(B+ ) ⊃ (B1 )+ ∩ φ(B).
(c’) kψ(x)k ≥ kxk for all x ∈ B.
(e’) φ((B+ )◦ ) ⊃ ((B1 )+ )◦ .

(c)⇒(e): Let x ∈ B with φ(x) ∈ ((B1 )+ )◦ . There is m ∈ N with φ(b) :=

mφ(x) − e1 ∈ (B1 )+ for b := mx − e ∈ B, cf. Part (v). In the proof of Part (ii) we
have seen that φ(b) ∈ (B1 )+ implies kte1 − φ(b)k ≤ t for t := kφ(b)k = kbk. Thus
kkbke − bk = kte1 − φ(b)k ≤ kbk, which implies b = nx − e ∈ B+ , as shown in proof
of Part (ii). Thus, x ∈ (B+ )◦ by Part (v).
(e)⇒(a): Let φ(x) ∈ (B1 )+ , then φ(x) + δe1 ∈ ((B1 )+ )◦ for each δ > 0, by
Part (v). It follows x + δe ∈ B+ for each δ > 0, i.e., x ∈ B+ .
(a)⇒(c): If B+ = φ−1 (φ(B) ∩ (B1 )+ ), then we get from kφ(b)ke1 ± φ(x) ∈
(B1 )+ , that kφ(b)ke ± x ∈ B+ . Thus kxk ≤ kφ(x)k for all x ∈ B, and φ is
isometric.
(c)⇒(b): The map φ defines an isometric unital isomorphism from B onto the
closed subspace L := φ(B) with e1 ∈ L. If λ ∈ E(B), then ρ0 := λ ◦ φ−1 : L → R
is in E(L). There is an extension ρ ∈ (B1 )∗ of ρ0 with kρk = kρ0 k = 1. Thus,
ρ ∈ E(B1 ) by Part (ii), and λ ◦ φ = λ0 ◦ φ = ρ.
1132 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

(b)⇒(c): By Part (iii), for b ∈ B,

kbk = sup |λ(b)| = sup |ρ(φ(b))| = kφ(b)k .

λ∈E(B) ρ∈E(B1 )

(e)⇒(d): Suppose that x ∈ Y and m ∈ N satisfy e1 ≤ mφ(x). Then φ(x) ∈

((B1 )+ )◦ by Part (v), and (e) implies that x ∈ (B+ )◦ . Again, Part (v) implies that
there is n ∈ N with e ≤ nx.
(d)⇒(c): Let y ∈ Y and p, q ∈ N with (p/q) > kφ(y)k. Then pe1 ± qφ(y) ∈
(B1 )+ . Thus φ(xk ) − e1 ∈ (B1 )+ for xk := (p + 1)e + (−1)k qy ∈ Y , k ∈ {1, 2}.
By assumption (d), there are n1 , n2 ∈ N with nk xk − e ∈ B+ . Since (nk (p + 1) −
1)/(nk q) ≤ (p + 1)/q, we get Thus ((p + 1)/q)e ± y ∈ B+ , and (p + 1)/q ≤ kyk.
Since inf{(p + 1)/q ; p/q > t} = t for each t ∈ R+ , we obtain kφ(y)k ≥ kyk for all
S
y ∈ Y . This carries over to all z ∈ Z = n (1/n)Y , i.e., kzk ≤ kφ(z)k for all z ∈ Z.
It follows kbk = kφ(b)k for all b ∈ B, because Z is dense in B and kφk ≤ 1.
(ix): Suppose that Y ⊆ B is an additive subgroup such that spanR (Y ) is
dense in B, and that F ⊆ E(B) satisfies kyk = supf ∈F |f (y)| for each y ∈ Y . Let
M (b) := supf ∈F |f (b)| for b ∈ B. Then M (b) is a semi-norm on B with M (b) ≤ kbk,
the set Z := {(1/n)x ; x ∈ Y } = spanQ (Y ) is norm-dense in B, and M (z) = kzk
for all z ∈ Z. Thus, M (b) = kbk on all b ∈ B, i.e., F is norming for B.
Suppose now that F is norming on B. Let G ⊆ E(B) denote the *weak closure
of the the convex hull of F . Then G is convex, compact, and is norming on B.
Suppose that G 6= E(B). Then, by Hahn-Banach separation, there exist b ∈ B
and f ∈ E(B) such that f (b) 6∈ {g(b) ; g ∈ G}. Since G is compact and convex,
there are α, β ∈ R with {g(b) ; g ∈ G} = [α, β]. Let c := 2b − (α + β)e, then
f (c) 6∈ {g(c) ; g ∈ G} = [α − β, β − α]. Thus, kck = supg∈G |g(c)| = β − α and
|f (c)| > kck, which contradicts f ∈ E(B).
Let K a compact convex subset of a locally convex vector space V . Then the
continuous linear functionals f ∈ V ∗ separate the point of K and f |K ∈ Aff c (K).
The set Aff c (K) of continuous affine functions on K is a closed subspace of C(K)
that contains 1.
The natural map k ∈ K 7→ δk E(C(K)) ⊆ C(K)∗ is continuous. The re-
strictions b
k of δk to the elements of Aff c (K) are in the state space E(Aff c (K)) of
Aff c (K). The map k 7→ b k is continuous, affine (= convex) and injective. Thus, cK
is a norming closed convex subset of the state space of Aff c (K). Thus, k 7→ k isb
surjective. It follows that k 7→ b
k is an affine topological isomorphism from K onto
E(Aff c (K)).
(x): The natural map B 3 b 7→ bb ∈ Aff c (E(B)) ⊆ C(E(B)) with bb(λ) := λ(b).
Is unital and isometric by Part (iii). By Part (ix), the natural map from E(B) into
E(Aff c (E(B))) is bijective. It follows that B b ⊆ Aff c (E(B)) separates the states
of Aff c (E(B)). Now Part (vi) shows that B b = Aff c (E(B)). Thus, b 7→ bb defines
a natural unital isometric isomphism from (B, B+ , e) onto the order unit space
Aff c (E(B)) ⊆ C(E(B)). 
 13. ORDER UNIT SPACES AND PRE-ORDERED SEMIGROUPS 1133

Lemma A.13.3. Let V be a real vector space, K ⊆ V a convex cone in V and

u ∈ K such that −u 6∈ K and that, for every x ∈ V there exists n = n(x) ∈ N with
nu − x ∈ K.
Define P (x) := inf{m/n ; m, n ∈ N, mu ± nx ∈ K } .
Then P is a seminorm on V .
Let V0 the space of v ∈ V with P (v) = 0, let π(x) := x + V0 the quotient map
π : V → V /V0 and define a Norm k · k on V /V0 by kπ(x)k = P (x).

(i) The completion B of the normed space V /V0 becomes an order unit space
with order unit e := π(u) = u + V0 and B+ := π(K).
(ii) (B, B+ , e) is naturally order-unit isomorphic to the space of affine func-
tions on the σ(V 0 , V )-compact convex set of linear maps λ : V → R with
λ(K) ⊆ R+ and λ(u) = 1.
(iii) π(x) is in the interior of B+ , if and only if, there is n ∈ N with nx−u ∈ K.
(iv) Suppose that X is an additive subgroup of of V , such that u ∈ X and V
is the real linear span of X, and that ψ : X → B1 is an additive map from
X into an order unit space (B1 , (B1 )+ , e1 ) with ψ(u) = e1 .
If the existence of n ∈ N with nx − e ∈ K implies the existence of
m ∈ N with mψ(x) − e1 ∈ (B1 )+ , then there is a unique unital linear
contraction η : B → B1 with η(π(x)) = ψ(x) for all x ∈ X.
This is in particular the case, if the map ψ : X → B1 is order pre-
serving, i.e., if ψ(x) ≤ ψ(y) if y − x ∈ K.
(v) The map η : B → B1 in (iv) is isometric, if and only if, for each x ∈ X,
the existence of n ∈ N with nx − e ∈ K is equivalent to the existence of
m ∈ N with mψ(x) − e1 ∈ (B1 )+ .

Proof. Clearly P (x) = P (−x), P (u) = 1 (by −u 6∈ K) and P (0) = 0. If

t ∈ R+ satisfies tu ± x ∈ K, then su ± x = (tu ± x) + (s − t)u ∈ K for all s ≥ t.
Thus, P (x) = inf{t ≥ 0 ; tu ± x ∈ K}. Since, for t, s > 0, tu ± x ∈ K ⇔
(st)u ± sx ∈ K, we get P (sx) = sP (x) for s > 0 and x ∈ V .
If t1 , t2 ≥ 0 and t1 u ± x, t2 u ± y ∈ K implies (t1 + t2 )u ± (x + y) ∈ K. Thus
P (x + y) ≤ P (x) + P (y). Thus, P is a semi-norm on V .
(i): Since (P (x) + ε)u ± x ∈ K for all ε > 0, we have (kπ(x)k + ε)e ± π(x) ∈ B+
for all x ∈ V . The map y 7→ (kyk + ε)e − y is continuous on B and takes values
in B+ for all y ∈ π(V ) = V /V0 . Since B+ is closed, we get kyke − y ∈ B+ for
all y ∈ B. It follows te − y = (kyke − y)+ ∈ B+ for all y ∈ B and t ≥ kyk. In
particular, e + U ⊆ B+ for the open unit ball U of B, and e is contained in the
interior of B+ .
If y ∈ B and s ≥ 0 satisfies se ± y ∈ B+ then for every ε > 0, there are
x1 , x2 ∈ K and z ∈ V with kse + y − π(x1 )k < ε, kse − y − π(x2 )k < ε and
ky−π(z)k < ε. It implies P (su+z−x1 ) < 2ε, P (se−z−x2 ) < 2ε and kyk ≤ P (z)+ε.
1134 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

It follows (2ε)u + (su + z − x1 ) ∈ K and (2ε)u + (su − z − x2 ) ∈ K. Since

x1 , x2 ∈ K, we get (2ε) + s ≥ P (z) and, finally, s ≥ kyk − 3ε. Thus, se ± y ∈ B+
and s ≥ 0 imply s ≥ kyk, and (B, B+ , e) is an order unit space.
(ii): We have (P (x)+ε)u±x ∈ K for x ∈ V and all ε > 0. Thus, |λ(x)| ≤ P (x)
for all linear maps λ : V → R with λ(u) = 1 and λ(K) ⊆ R+ . Since V /V0 is
dense in B, there is a unique linear functional ρ : B → R with ρ ◦ π = λ and
kρk ≤ 1. The functional ρ is positive and unital, because π(K) is dense in B+ and
ρ(e) = λ(u) = 1.
Clearly, λ := ρ ◦ π is a linear map from V to R with λ(K) ⊆ R+ and λ(u) = 1, if
ρ : B → R is a positive linear functional on B with ρ(e) = 1.
The σ(Lin(V, R), V )-topology on the convex set of linear maps λ : V → R with
λ(K) ⊆ R+ and λ(u) = 1 gives it a structure of a compact convex set E(V, K, u).
The (algebraic and affine) isomorphism E(B) → E(V, K, u), given by ρ 7→ ρ◦π, is is
an affine homeomorphism, because π(V ) is dense in B. By Proposition A.13.2(ix),
the order unit space (B, B+ , e) is naturally order-unit isomorphic to the space of
real-valued continuous affine functions on the compact convex set E(B) of states
on B.
(iii): If nx − u ∈ K, then nπ(x) − e ∈ B+ , which implies that π(x) is in the
interior of B+ by Proposition A.13.2(v).
Suppose that π(x) is in the interior of B+ , then there is n ∈ N with π(x) −
(2/n)e ∈ B+ , i.e., π(nx − 2u) ∈ B+ . Since B+ is the closure of π(K), there exists
y ∈ V and k ∈ K with nx − 2u = y + k and P (y) = kπ(y)k < 1. It follows that
u + y ∈ K (by definition of P (y)). Hence nx − u = (y + u) + k ∈ K.
(iv): Suppose that ψ : X → B1 is an additive map with ψ(u) = e1 , and suppose
that, for each x ∈ X, the existence of n ∈ N with nx − u ∈ K implies the existence
of m ∈ N with mψ(x) − e1 ∈ (B1 )+ .
Let p, n ∈ N with p/n > P (x) = kπ(x)k, then pu±nx ∈ K, and ((p+1)u±nx)−
e ∈ K. Thus, there is m ∈ N with m((p+1)e1 ±nψ(x))−e1 ∈ (B1 )+ . It follows, that
(m(p + 1) − 1)e1 ± mnψ(x) ∈ (B1 )+ . Hence kψ(x)k ≤ (p + 1 − 1/m)/n ≤ p/n + 1/n.
Since we can approximate P (x) ≥ 0 from above by rational p/n with arbitrary
large denominator n, we get kψ(x)k ≤ P (x) for all x ∈ X.
S
The Q-linear hull of X is given by W := n (1/n)X. The map ψ : X →
B1 naturally extends to W by ψ0 ((1/n)x) := (1/n)ψ(x). The extension is well-
defined and Q-linear, because (nm)−1 (mpx ± nqy) = (1/n)(px) ± (1/m)(qy) maps
to (nm)−1 (mpψ(x) ± nqψ(y)) = (1/n)(pψ(x)) ± (1/m)(qψ(y)). We have P (w) ≥
kψ0 (w)k for all w ∈ W , because this holds for w ∈ X. In particular, ψ0 (w) = 0 if
w ∈ V0 .
Consider the set Y := π(X) ⊆ B. It satisfies the assumptions on Y of Parts
(vi, vii, viii) of Proposition A.13.2, because V is the real span of X. In particular,
S
the rational span Z := n (1/n)π(X). of Y is dense in the real span of X, hence is
dense in B. Let z = (1/n)π(x) ∈ Z.
 13. ORDER UNIT SPACES AND PRE-ORDERED SEMIGROUPS 1135

We define η0 (z) := ψ0 (w) for z ∈ Z and w ∈ W with π(w) = z. The definition

is correct, because π(w0 ) = π(w) is equivalent to P (w0 − w) = 0 by definition of V0
and π : V → B. Thus kψ0 (w0 − w)k ≤ P (w0 − w) = 0 and ψ0 (w0 ) = ψ0 (w). We
have kzk = P (w) ≥ kψ0 (w)k = kη0 (z)k.
The map η0 : Z → B1 satisfies η0 ◦ π = ψ0 . It follows that η0 is Q-linear and
contractive.
Since V is the real span of X, and since π(V ) is dense in B, the rational span
Z of Y is dense in B. Thus, the unital contractive Q-linear map η0 : Z → B1
uniquely extends to a unital contractive R-linear map η : B → B1 . This map
satisfies η ◦ π(x) = ψ(x).
Since η : B → B1 is a unital linear contraction, the map η is increasing.
(v): Let η : B → B1 an isometric linear map with η(π(x)) = ψ(x) for all x ∈ X.
Then η(e) = η(π(u)) = ψ(u) = u1 . Since η unital and contractive, η(B+ ) ⊆ (B1 )+ ,
by Proposition A.13.2(vii). If nx − u ∈ K, then nπ(x) − e ∈ B+ and η(nπ(x) − e) =
nψ(x) − e1 ∈ (B1 )+ . If there is m ∈ N with η(mπ(x) − e) = mψ(x) − e1 ∈ (B1 )+ ,
then mπ(x) − e ∈ B+ by Proposition A.13.2(viii,a).
Conversely, suppose that ψ : X → B1 is additive and unital, and suppose that,
for each x ∈ X, the existence of n ∈ N with nx − u ∈ K is equivalent to the
existence of m ∈ N with mψ(x) − e1 ∈ (B1 )+ . Then, by part (iv), there is a unique
unital linear contraction η : B → B1 with η(π(x)) = ψ(x) for all x ∈ X.
The set Y := π(X) satisfies the assumptions of Proposition A.13.2(viii,d):
S
Z = n (1/n)Y is dense in B, and for y ∈ Y ,the existence of m ∈ N with e1 ≤ mη(y)
implies the existence of n ∈ N with e ≤ ny. Indeed: If e1 ≤ mη(y) and x ∈ X with
y = π(x), then mψ(x) − e1 = mη(y) − e1 ∈ (B1 )+ , which implies, by assumption
on X, the existence of n ∈ N with nx − u ∈ K, i.e., ny − e = π(nx − u) ∈ B+ .
Thus, the unital linear contraction η is isometric by the equivalence of properties
(viii,c) and (viii,d) in Proposition A.13.2(viii). 

Now we use the above elementary material on order-unit spaces for a study of
scaled abelian semigroups.
Definition A.13.4. Let S denote an abelian semigroup. S is preordered by
the relation x <S y if the relation is transitive and invariant under addition, i.e.,
x <S z if x <S y and y <S z, x + w <S y + w for all w ∈ S. We do not require that
x <S x, therefore we write x ≤S y if x <S y or x = y. The preordered semigroup
(S, <S ) is partially ordered if x ≤S y and y ≤S x imply x = y. (If it is clear,
which S and preorder <S is ment, we simply write < and ≤ for <S respectively
≤S .)
An element u ∈ S is an order unit in the preordered abelian semigroup (S, <S )
if x ≤ x + u for all x ∈ S, and, for every y ∈ S, there is n = n(y) ∈ N such that
y ≤ nu and u ≤ y + nu ( 7 ). Usually there are many different order units in S.
7The properties x ≤ x + u and that u ≤ y + nu follow automatically, if S has a zero element

0 with 0 ≤ x for all x ∈ S.

1136 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

An abelian preordered semigroup (S , <S , u) with a distinguished order unit

u ∈ S will be called a scaled (abelian) semigroup.
A state on S is an order preserving additive map λ : S → R with λ(u) = 1.
Lemma A.13.5. Suppose that V and W are vector spaces over the rational
numbers Q.

(i) The natural Z-modul morphism V ⊗Z W → V ⊗Q W is an isomorphism.

(ii) η : v →
7 v ⊗Z 1 is a natural Q-linear monomorphism from V into the real
vector space V ⊗Z R = V ⊗Q R.
(We identify V with its image in V ⊗Q R.)
(iii) Let U ⊆ V ⊗Z R an open subset with respect to an locally convex topology
O on V ⊗Z R. Then V ∩ U is dense in U with respect to O.
(iv) For every subset X ⊆ V , and for every (not necessarily separated) locally
convex topology on V ⊗Z R, the rational convex hull convQ (X) of X in V
is dense in the real convex hull convR (X) of X in V ⊗Z R.
(v) convR (X) ∩ V = convQ (X) holds in V ⊗Z R.

The observations (i)-(v) imply the following observation:

Suppose that T : V → W is an additive map, ν : W → [0, ∞) is a semi-norm
on W ⊗Z R, X ⊆ W is a subset of W , and that there exists v1 ∈ V ⊗Z R, such
that (T ⊗Z idR )(v1 ) is in the interior of convR (X) ⊆ W ⊗Z R with respect to the
topology on W ⊗Z R defined by ν.
Then there exists v2 ∈ V such that T (v2 ) ∈ W is in the interior of convQ (X)
with respect to ν.

Proof. Since the tensor products are the algebraic tensor products, the state-
ments reduces to V, W of finite dimension (over Q, i.e., V ∼
= Qn and W ∼ = Qm as
vector spaces over Q), and to finite subsets X of V ∼
= Qn .
Then the proofs automatically reduce to the following easy observations:
(i): There are natural isomorphisms Q ⊗Z Q ∼
= Q, and, therefore, also natural
isomorphisms
Qn ⊗Z Qm ∼ = Qnm ∼= Qn ⊗Q Qm .
∼ V ⊗Q R, because R
(ii): By Part (i), there is a natural isomorphism V ⊗Z R =
m
is a Q-vector space. In case V = Q , the natural isomorphism induces the natural
embedding of Qm in Rm . The injectivity of η : v 7→ v ⊗Z 1 then follows from the
injectivity of natural embedding Qn → Rm = Qm ⊗Z R.
(iii): The finest locally convex topology on Rn is identical with the topology
induced by the norm k(x1 , . . . , xn )k∞ := maxj |xj |. Thus, every open subset U of
Rn with respect to any (not necessarily separated) locally convex topology O of Rn
is the union of products I1 × · · · × In of open intervals Ik ⊆ R. It follows that the
vectors with rational coordinates are dense U with respect to to the topology O,
because U1 ∩ U = ∅ if U1 ∈ O and U1 ∩ U does not contain a vector with rational
coordinates.
 13. ORDER UNIT SPACES AND PRE-ORDERED SEMIGROUPS 1137

Pm
(iv): Any R-convex combination j=1 αj vj = p ∈ Rn of v1 , . . . , vm ∈ Qn can
be approximated by a convex combination of the v1 , . . . , vm with rational αj ≥ 0
P
with j αj = 1, because the rational points are dense in the (m − 1)-dimensional
standard simplex S ∈ Rm (by part (iii) applied to the inverse T −1 (S) of a suitable
affine injective map T : Rm−1 → Rm with T (0) = (1/m, . . . , 1/m) and rational
entries in its coefficient matrix).
(v): convR (X) ∩ V = convQ (X) comes from the case where V ∼ = Qn . Let
X = {v1 , . . . , vm } ⊆ Qn is a finite subset, and j αj vj = v0 ∈ Qn , with αj ∈
P
P
(0, ∞), αj = 1. Then we can form the matrix A ∈ Mm,n+1 (Q) with columns
aj = (vj , 1)> , and let b ∈ Mn+1,1 (Q) the column b = (v0 , 1). Then A · x0 = b with
x0 = [α1 , . . . , αn ]> ∈ Mn,1 (R) . It follows that the matrices A and the extended
matrix (A, b) have same rank (calculated in R). Since both matrices have rational
entries, the ranks are also equal if calculated in Q. Thus there is a solution of
A · x1 = b with x1 = [β1 , . . . , βn ]> ∈ Mn,1 (Q) . If B ∈ Mn+1,m (Q) is a generalized
inverse of A then T : y ∈ Mm,1 (R) 7→ (1m − BA)y + Bb is an affine map from Rm ∼ =
(
Mm,1 (R) onto the set of all solutions x ∈ Mm,1 R) of Ax = b. The set T −1 (0, ∞)m
is an open convex subset of Rm that is not empty, because (0, ∞)m is open in Rm
and x0 ∈ (0, ∞)m by assumption. By part (ii), we get that Qm ∩ T −1 (0, ∞)m is
not empty, i.e., there is x2 ∈ (0, ∞)m ∩ Qm with Ax2 = b. This means that v0 is a
rationally-convex combination of v1 , . . . , vm . 

The following key lemma is an observation of B. Blackadar and M. Rørdam

[86]. It provided a useful “almost” cancellation property.

Lemma A.13.6. If (S, <S , u) is a scaled abelian semigroup, then, for x, y ∈ S,

there exists m, p ∈ N with mx + (p + 1)u ≤ my + pu , if and only if, there are n ∈ N
and z ∈ S with nx + u + z ≤ ny + z.

Proof. If mx + (p + 1)u ≤ my + pu, then we let n := m and z := pu. To

show the non-trivial direction, it suffices to show that x + u + z ≤ y + z implies the
existence of m, p ∈ N with mx + (p + 1)u ≤ my + pu, because we can rename nx
and ny by x and y in the statement.
If xj , yj , z ∈ S satisfy xj + z ≤ yj + z (j = 1, 2), then
x1 + x2 + z ≤ x1 + y2 + z ≤ y1 + y2 + z .
Induction gives that nx + z ≤ ny + z for all n ∈ N if x + z ≤ y + z.
If x1 , x2 ∈ S, then there is n ∈ N with x1 ≤ x2 + nu: Indeed, there are
n1 , n2 ∈ N with a ≤ n1 u, b ≤ n2 u and u ≤ b + n2 u. It follows n1 b ≤ b + (n1 − 1)n2 u
and a ≤ b + nu for n := (2n1 − 1)n2 .
Let x, y, z ∈ S with x + u + z ≤ y + z. There are k, ` ∈ N such that u ≤ z + ku
and z ≤ ku, u ≤ y+`u, y ≤ `u. It follows z ≤ k(y+`u). Let x0 = x+`u, y 0 = y+`u.
Then z ≤ ky 0 , x0 + u + z ≤ y 0 + z, thus nx0 + nu + z ≤ ny 0 + z for all n ∈ N. Take
above a := kx0 + u and b := z, then there is n ∈ N with kx0 + u ≤ nu + z. We get
(n + k)x0 + u = nx0 + kx0 + u ≤ nx0 + nu + z ≤ ny 0 + z ≤ (n + k)y 0 .
1138 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

It implies mx + (p + 1)u ≤ my + pu for m := n + k, p := (n + k)`. 

Proposition A.13.7. Suppose that (S, < , u) is a scaled abelian semigroup
such that m, n ∈ N and mu ≤ nu imply m ≤ n.
Then there is an order unit space B(S) = (B, B+ , e) and a unital additive map
γ : S → B with γ(u) = e, such that, for every state λ on (S, < , u), there is exactly
one state ρ on B with λ = ρ ◦ γ . Furthermore:

(i) Natural universality: If B1 is an oder unit space, and ψ : S → B1 is

monotone, unital and additive, then there is a unique monotone unital
linear map T : B(S) → B1 with T ◦ γ = ψ.
(ii) There are equivalent:
(ii.1) γ(y) − γ(x) is in the interior of B+ .
(ii.2) There exist m, p ∈ N such that mx + (1 + p)u ≤ my + pu.
(ii.3) λ(x) < λ(y) for all states λ of S.
(iii) An additive unital map ψ from S to an order unit space (B1 , (B1 )+ , e1 ) is
monotone, if and only if, the inequality x+u ≤ y in S implies ψ(x) ≤ ψ(y)
for x, y ∈ S ( 8 ).
(iv) For a monotone additive unital map ψ : S → B1 are equivalent:
(iv.1) There is an isometric unital map I : B → B1 with I(γ(x)) = ψ(x)
for x ∈ S.
(iv.2) If there is n ∈ N with nψ(x) + e1 ≤ nψ(y), then there exist m, p ∈ N
with mx + (p + 1)e ≤ my + pe (i.e., γ(y) − γ(x) is in the interior of
B+ by (ii)).
(iv.3) For every state λ on S there exists a state ρ on B1 with λ = ρ ◦ ψ.

Proof. Let Gr(S) denote the Grothendieck group of S, and let

Gr(S)+ := {[y] − [x] ; x, y ∈ S, x ≤S y} .
Then for each g ∈ Gr(S) there are s, t ∈ S with g = [s] − [t].
By Definition A.13.4, there are m, n ∈ N with s ≤ nu, u ≤ t + nu and t ≤ nu.
We get g = ([nu] − [t]) − ([nu] − [s]) ∈ Gr(S)+ − Gr(S)+ and g ≤ g + ([nu] − [s]) =
[nu] − [t] ≤ 2n[u] . Thus, (Gr(S), Gr(S)+ , [u]) is a scaled abelian group with order
g ≤ h ⇔ g − h ∈ Gr(S)+ and order unit [u].
It turns out for x, y ∈ S that [x] ≤ [y] in Gr(S), if and only if, there is z ∈ S with
x + z ≤ y + z. Indeed, x + z ≤ y + z implies [y] − [x] ∈ Gr(S)+ . If [y] − [x] ∈ Gr(S)+
then there are x1 , x2 , s ∈ S with x1 ≤ x2 and y + x1 + s = x + x2 + s =: t. Thus
x + z = t + x1 ≤ t + x2 = y + z for z := x1 + x2 + s.
By Lemma A.13.6 it follows that there is n ∈ N with n[x] + [u] ≤ n[y] in Gr(S)
(i.e., that there exists z ∈ S with nx + z + u ≤ ny + z), if and only if, there are
m, p ∈ N with mx + (p + 1)u ≤ my + pu in S.
We define a pre-ordered R-vector space (V, K, w) by V := Gr(S) ⊗Z R, w :=
[u] ⊗Z 1 and K is the convex cone of V := Gr(S) ⊗Z R generated by Gr(S)+ ⊗Z 1 .
8It implies then that φ(y) − φ(x) is in the interior of the positive cone (B ) .
1 +
 13. ORDER UNIT SPACES AND PRE-ORDERED SEMIGROUPS 1139

Since Q⊗Z R = Q⊗Q R = R, we get from Lemma A.13.5, that Gr(S)⊗Z Q → V

S
is injective and that (Gr(S)⊗Z Q)∩K = n (Gr(S)+ ⊗Z 1/n). Hence, (g ⊗1) ∈ K, if
and only if, there exists n ∈ N with ng ∈ Gr(S)+ . In particular, ([y] − [x]) ⊗ 1 ∈ K,
if and only if, there are m ∈ N and z ∈ S with mx + z ≤ my + z.
Suppose −w = ([u] − [2u]) ⊗ 1 ∈ K, then there is m ∈ N (w.l.o.g.) with
2mu + z ≤ mu + z in S. By LemmaA.13.6, there are n, q ∈ N with n((2m − 1)u) +
(q + 1)u ≤ nmu + qu, i.e., ku ≤S `u with mn + q =: ` < k := n(2m − 1) + q + 1.
This contradicts our assumption that ku ≤ `u implies k ≤ `.
Since −w 6∈ K, (V, K, w) is a scaled R-vector space, as considered in Lemma
A.13.3. We define, B(S) := (B, B+ , e) as the order unit space that is obtained from
the scaled R-vector space (V, K, w) in Lemma A.13.3, and let γ(x) := π([x] ⊗Z 1),
where π : V → B is the quotient map v ∈ V 7→ v + V0 ∈ V /V0 ⊆ B as considered
in Lemma A.13.3.
Since g ∈ G 7→ g ⊗Z 1 ∈ V and π are monotone and unital, we get that
γ : S → B is a monotone, unital additive map. Thus λ := ρ ◦ γ is a monotone
additive unital map, if ρ is a state of B(S).
It is obvious, that every state λ : S → R uniquely factorizes through Gr(S)
and then through (V, K, w), i.e., for every state λ of S there is a unique state ρ0
on (V, K, w) and then a unique state ρ on (B, B+ , e) (by Lemma A.13.3(iv) with
(B1 , (B1 )+ , u1 ) := (R, R+ , 1)) with ρ(γ(x)) = ρ0 ([x] ⊗ 1) = λ(x) for all λ ∈ E(S).
It follows from Lemma A.13.3(ii), that there is a unique isometric unital order
isomorphism ι from B onto Aff c (E(S)) with ι(γ(x)) = ι(π([x] ⊗ 1))(λ) = λ(x) for
all x ∈ S.
(i): Suppose that ψ : S → B1 is monotone, unital and additive map into an
order unit space B1 . Since B1 is an R-vector space, the map factorizes through a
linear map [ϕ] : V = Gr(S) ⊗Z R → B1 with ψ(x) = [ψ]([x]) for x ∈ S.
The map [ψ] is unital and order preserving, and X := Gr(S) ⊗Z Q satisfies the
assumptions of Lemma A.13.3(iv).
Have to show:
There is a unique monotone unital linear map T : B(S) → B1 with T ◦ γ = ψ.
That means: Existence by construction ... and proof of uniqueness ... e.g. by
use of separating states on B1 .
If x ≤ y then mx + u ≤ (mx + u) for any m ∈ N. It follows that there are
n, q ∈ N with n(mφ(x)) + (q + 1)v ≤ n(mφ(x) + v) + qv. Thus nmλ(φ(x)) ≤
(n − 1) + nmλ(φ(x)) and λ(φ(x)) ≤ λ(φ(x)) + (1/m) for all m ∈ N and all states λ
on (B1 , (B1 )+ , e1 ).
to be filled in ?? 

Corollary A.13.8. Let (S , < , u), (T , < , v) scaled abelian semigroups, and
φ : S → T an additive map with φ(u) = v.
1140 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

If mv ≤T nv implies m ≤ n (for n, m ∈ N), then φ has the following properties

(i) and (ii).

(i) λ ◦ φ is a state of S for each state λ of T ,

if and only if,
for x, y ∈ S, the inequality x + u ≤ y implies the existence of n, q ∈ N
with nφ(x) + (q + 1)v ≤ nφ(y) + qv.
(ii) For every state ρ of S there is a state λ of T with ρ = λ ◦ φ,
if and only if,
for x, y ∈ S, the existence of m, p ∈ N with mx + (p + 1)u ≤ my + pu is
equivalent to the existence of n, q ∈ N with nφ(x) + (q + 1)v ≤ nφ(y) + qv.

Proof. (i): Suppose that for each x, y ∈ S with x + u ≤ y there are n, q ∈ N

with nφ(x) + (q + 1)v ≤ nφ(y) + qv, then 1/n ≤ λ(φ(y)) − λ(φ(x)) for all states
λ : T → R on (T, <T , v). Thus x+u ≤ y implies λ(φ(x)) ≤ λ(φ(y)). Since (R, R+ , 1)
is an order unit space, we get from Part(iii) of Proposition A.13.7, that the additive
unital maps λ ◦ φ : S → R are monotone, i.e., that the λ are states on (S , <S , u).
Suppose, conversely, that λ ◦ φ is a state on (S, <S , u) for all states λ on
(T, <T , v). If x + u ≤ y then λ(φ(y) − φ(x)) ≥ 1 for all states λ on (T, <T , v).
This implies the existence of n, q ∈ N such that nφ(x) + (l + q)v ≤ nφ(y) + qv, by
part (ii) of Proposition A.13.7,
(ii): By part (i) and by Proposition A.13.7, there is a unique unital monotone
linear map ψ : B(S) → B1 := Aff c (E(T )) ∼= B(T ) with ψ(γS (x))(λ) = λ(φ(x)) for
all x ∈ S and λ ∈ E(T ).
Suppose that nφ(x) + (p + 1)v ≤ nφ(xy) + pv in T implies the existence of
m, p ∈ N with mx+(p+1)e ≤ my+pe in S. If there is k ∈ N with kψ(x)+e1 ≤ kψ(y)
in (B(T ), B(T )+ , eT ) then 1/k ≤ λ(φ(y))−λ(φ(x)) for every state λ on T . If follows
the existence of n, q ∈ N with nφ(x) + (p + 1)v ≤ nφ(xy) + pv in T , by part (ii) of
Proposition A.13.7. By assumption, there are m, p ∈ N with mx+(p+1)e ≤ my+pe.
Thus, ψ : B(S) → Aff c (E(T )) satisfies the condition in part (iv.2) of Propo-
sition A.13.7(iv). It implies that for every state ρ on S there is a state ρ0 on
Aff c (E(T )) with ρ(x) = ρ0 (ψ(γS (x))) for x ∈ S. Since E(T ) is compact and con-
vex, it follows that there is λ ∈ E(T ) with ρ0 (γT (y)) = λ(y) for all y ∈ T . Thus,

ρ(x) = ρ0 (ψ(γS (x)) = ρ0 (γT (φ(x)) = λ(φ(x)) .

By Proposition A.13.7, there is a state λ on T such that ρ0 (γT (y)) = λ(y) for
all y ∈ T .
γ(y) − γ(x) is in the interior of B(S)+ , iff, there are m, p ∈ N with mx + (p +
1)u ≤ my + pu.
For every state λ on S there exists a state ρ on B1 with λ = ρ ◦ ψ.
Suppose that for every state ρ on S there is a state λ on T with λ ◦ φ = ρ, then
ψ is isometric by
 13. ORDER UNIT SPACES AND PRE-ORDERED SEMIGROUPS 1141

?????????????? reference?
For every state ρ of S there is a state λ of T with ρ = λ ◦ φ, if and only if, for
x, y ∈ S, the existence of m, p ∈ N with mx + (p + 1)u ≤ my + pu is equivalent to
the existence of n, q ∈ N with nφ(x) + (q + 1)v ≤ nφ(y) + qv.
Let η : T → B(T ) ∼
= Aff c (E(T )) denote the natural unital additive map from T
into is universal order unit space (B(T ), B(T )+ , eT ) (cf. Proposition A.13.7), and
let ψ := η ◦ φ.
By part (iii) of Proposition A.13.7, the additive unital map ψ : S → B(T )
from S to the order unit space (B(T ), B(T )+ , eT ) is monotone, if and only if, the
inequality x + u ≤ y in S implies φ(x) ≤ φ(y) for x, y ∈ S.
to be filled in ??
A.13.7 

Corollary A.13.9. If (S1 , < , u) is a scaled subgroup of (S, < , u) (with in-
duced order), then B(S1 ) → B(S) is isometric. In particular, every state of S1
extends to a state of S.

Proof. Consider the inclusion map φ : S1 3 x 7→ x ∈ S and let x, y ∈ S1 .

There are n, p ∈ N with nx+(p+1)u ≤ ny+pu in S1 , if and only if, nφ(x)+(p+1)u ≤
nφ(y) + pu. Now apply part (ii) of Corollary A.13.8 and Proposition A.13.7. 

Corollary A.13.10. Suppose that the scaled abelian semigroup (S, < , u) has
a zero element 0 with 0 ≤ x for all x ∈ S.
If a ∈ S satisfies λ(a) = 0 for all states λ of S, then there exists n(a, u) ∈ N
such that
2na ≤ nu for all n ≥ n(a, u) .

Proof. The proof includes the case, that φ ≡ 0 is the only monotone additive
map φ : S → R.
If φ(x) ≡ 0 is the only monotone linear map φ : S → R, then there is no state
on the sub-semigroup T = N · u of S with induced preorder and order-unit u (by
Corollary A.13.9). It follows that there is m ∈ N with (m + 1)u ≤ mu, because
otherwise there is a state on (T, <S |T, u) that extends to (S, <S , u) by Corollary
A.13.9. We get (n + `)u ≤ nu for all n ≥ m and ` ∈ N (by induction over `). There
is k ∈ N with a ≤ ku. Thus, 2na ≤ 2nku ≤ nu for all n ≥ m.
If mu ≤ nu always implies m ≤ n, then, by Proposition A.13.7(ii), there are
m, p ∈ N with 3ma + (p + 1)u ≤ (m − 1)u + (p + 1)u, because 1 ≤ λ(u) − λ(4a) for
all states λ of S.
It implies q3ma + (p + 1)u ≤ q(m − 1)u + (p + 1)u for all q ∈ N (by induction
over q, cf. proof of Lemma A.13.6). There is k ∈ N with a ≤ ku and k ≥ p + 1. We
get (q3m + 1)a ≤ (q(m − 1) + k)u for all q ∈ N. If n ∈ N ∩ [q(m − 1) + k, (q3m + 1)/2]
then 2na ≤ nu. Let q ≥ (2m−3+2k)/(m+2) then q3m+1 ≥ 2((q +1)(m−1)+k).
1142 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

It follows that
[
[n0 , ∞) ⊆ [q(m − 1) + k, (q3m + 1)/2]
q∈N

for n0 := q0 (m − 1) + k with q0 := 1 + [(2m − 3 + 2k)/(m + 2)]. 

14. Approximate intertwining of inductive limits

We have to work sometimes with C *-systems that are inductive limits of com-
pletely positive contractions on non-unital C *-algebras.
Some text missing? ??
We write B∞ := `∞ (B)/c0 (B) for a Banach space B. Recall that B∞ is a
C *-algebra in case that B is a C *-algebra. Then B∞ is the same as B∞ in the
terminology Q(X, B) := Cb (X, B)/ C0 (X, B) used in Chapter 6 for locally compact
spaces X (here X := N). More generally, we define B∞ for a sequence B1 , B2 , . . .
of Banach algebras or C *-algebras by
Y M
B∞ := ( Bn )/( Bn ) = `∞ (B1 , B2 , . . .)/c0 (B1 , B2 , . . .).
n n

A sequence of contractions Vn : An → Bn defines a contraction V∞ : A∞ → B∞

by

V∞ ((a1 , a2 , . . .) + c0 (A1 , A2 , . . .) := (V1 (a1 ), V2 (a2 ), . . .) + c0 (B1 , B2 , . . .)

for (a1 , a2 , . . .) ∈ `∞ (A1 , A2 , . . .), a norm-bounded sequence with an ∈ An .

to be filled in: basics on indlim(Tn : Bn → Bn+1 ) ??
The following lemmata collect some basic facts on inductive limits of linear con-
tractions on Banach spaces and c.p. contractions on C *-algebras. We use notations
⊕n Bn = c0 (B1 , B2 , . . .) ⊆ c0 (L(H)) and Πn Bn = `∞ (B1 , B2 , . . .) ⊆ `∞ (L(H)) for
closed subspaces of L(H) with sufficiently big Hilbert space H (which plays only
an intermediate role).

Lemma A.14.1. Suppose that A1 , A2 , . . . and B1 , B2 , . . . are Banach spaces, and

that Sn : An → An+1 and Tn : Bn → Bn+1 are contractions.
There is a natural isometric embedding indlim(Tn : Bn → Bn+1 ) ⊆ B∞ given
by the maps ηn : Bn → B∞ defined by

ηn (b) := (0, · · · , 0, b, Tn (b), Tn+1 Tn (b), . . . ) + (⊕n Bn ) ,

where b ∈ Bn is on the n-th position. We write also Tn,∞ instead of ηn .

Let Vn : An → Bn , Wn : Bn → An+1 contractions, and let Sn := Wn Vn , Tn :=
Vn+1 Wn .
only up to small εn ?
Then there is a natural isometry

indlim(Sn : An → An+1 ) ∼
= indlim(Tn : Bn → Bn+1 )
 14. APPROXIMATE INTERTWINING OF INDUCTIVE LIMITS 1143

given by V∞ | indlim(Sn : An → An+1 ) and W∞ | indlim(Tn : Bn → Bn+1 ), where

V∞ : A∞ → B∞ and W∞ : B∞ → A∞ are given on representatives of elements in
A∞ and B∞ by
(a1 , a2 , . . .) 7→ (V1 (a1 ), V2 (a2 ), . . .) ,

respectively
(b1 , b2 , . . .) 7→ (0, W1 (b1 ), W2 (b2 ), . . .) .

Suppose now that the An and Bn are operator spaces (respectively are C*-spaces
or C*-algebras). The above given isomorphism is completely isometric (respectively
completely positive, respective is a C*-algebra isomorphism), if, for every k ∈ N
there is n(k) ∈ N such that Vn ⊗ idk and Wn ⊗ idk are contractive (respectively
are positive, i.e., have positive second conjugates; respective the Vn , Wn are C*-
morphisms).

Proof. We have Tk,∞ (Vk (a)) = V∞ (Sk,∞ (a)) for a ∈ Ak because, for n > k
and a ∈ Ak ,

Vn (Wn−1 Vn−1 (· · · (Wk Vk (a)) · · · )) = Tn−1 (· · · (Tk (Vk (a)) · · · )) .

Thus V∞ maps indlim(Sn : An → An+1 ) into indlim(Tn : Bn → Bn+1 ).

A similar argument shows that W∞ maps indlim(Tn : Bn → Bn+1 ) into
indlim(Sn : An → An+1 ).
Straight calculation shows that W∞ ◦ V∞ = S∞ and V∞ ◦ W∞ = T∞ , for the
maps S∞ : A∞ → A∞ and T∞ : B∞ → B∞ given on representatives by

(a1 , a2 , . . . , an , an+1 . . .) 7→ (0, S1 (a1 ), S2 (a2 ), . . . , Sn−1 (an−1 ), Sn (an ), . . .) ,

and

(b1 , b2 , . . . , bn , bn+1 . . .) 7→ (0, T1 (b1 ), T2 (b2 ), . . . , Tn−1 (bn−1 ), Tn (bn ), . . .) .

The operators S∞ and T∞ are contractions and fix the images of

indlim(Sn : An → An+1 ) ⊆ A∞ , respectively of indlim(Tn : Bn → Bn+1 ) ⊆ B∞ .
Indeed: On representatives one has S∞ (Sn,∞ (an )) = Sn,∞ (an ) for an ∈ An .
Thus, the restriction of V∞ defines an isometric linear isomorphism from
indlim(Sn : An → An+1 ) onto indlim(Tn : Bn → Bn+1 ) with the restriction of W∞
to indlim(Tn : Bn → Bn+1 ) as inverses.
Suppose now that the An and Bn are operator spaces, matrix operator sys-
tems, or C *-spaces (all not necessarily unital). It follows, that this isomorphism is
completely isometric (respectively are matrix-order isomorphisms) if V∞ and W∞
are completely contractive (respectively completely positive) on the images of the
respective images of the inductive limits.
It suffices to suppose that for each k ∈ N there is n(k) ∈ N such that Vn ⊗ idk
and Wn ⊗ idk are contractions (respectively are positive) for each n ≥ n(k). 
1144 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

Lemma A.14.2. Suppose that B1 , B2 , . . . are separable Banach spaces and that
Gn ⊆ L(Bn , Bn+1 ) are subsets of linear contractions from Bn into Bn+1 . If, for
each n ∈ N, the operator Tn : Bn → Bn+1 is in the point-norm closure of Gn ( =
closure of Gn in strong operator topology), then there exist g1 , g2 , . . . G, such that in
`∞ (B1 , B2 , . . .)/c0 (B1 , B2 , . . .) the related inductive limits coincide:
indlim(gn : Bn → Bn+1 ) = indlim(Tn : Bn → Bn+1 ) .

We identify here the inductive limits with their canonical corresponding closed
subspaces of B∞ := `∞ (B1 , B2 , . . .)/c0 (B1 , B2 , . . .).

Proof. There exist sequences b1,n , b2,n , . . . ∈ Bn that are dense in Bn , for each
n ∈ N.
We define Tm,n : Bm → Bn for n ≥ m, by Tm,m = id (identity map of Bm ),
Tn,n+1 := Tn Tm,n := Tn−1 ◦ · · · ◦ Tm+1 ◦ Tm for n > m. We denote by
Tm,∞ : Bm → X := indlim(Tn : Bn → Bn+1 ) ⊆ B∞
the defining morphisms from Bm into X. Recall that for our realization X ⊆ B∞
of indlim(Tn : Bn → Bn+1 ) holds
Tm,∞ (b) = (0, . . . , 0, b, Tm,m+1 (b), Tm,m+2 (b), . . .) + c0 (B1 , B2 , . . .)
for b ∈ Bm . Then Tm,∞ = Tn,∞ ◦ Tm,n .
The set {Tm,∞ (bk,m ) ; k, m ∈ N} is dense in X := indlim(Tn : Bn → Bn+1 ),
S
because X is the closure of m Tm,∞ (Bm ).
We find linear subspaces Ln ⊆ Bn and gn ∈ Gn of finite dimension, such that

(i) Ln contains the sets Tn−1 (Ln−1 )∪gn−1 (Ln−1 ) , { b1,n , b2,n , . . . , bn,n } , and
{Tk,n (bj,k ) ; 1 ≤ k ≤ n − 1, j ≤ n} , and {gk,n (bj,k ) ; 1 ≤ k < n − 1, j ≤
n} , where gk,n := gn−1 ◦ · · · ◦ gk+1 ◦ gk for 1 ≤ k < n − 1.
(ii) kTn (b) − gn (b)k ≤ 2−n kbk for b ∈ Ln .

Then Tn (Ln ) ⊆ Ln+1 , gn (Ln ) ⊆ Ln+1 , and indlim(Tn : Ln → Ln+1 ) and

Q L
indlim(gn : Ln → Ln+1 ) are the same closed subspaces of ( n Ln )/( n Ln ) ⊆ B∞ ,
because, Tm,∞ (Lm ) ⊆ Tn,∞ (Ln ) and gm,∞ (Lm ) ⊆ gn,∞ (Ln ) for n ≥ m, and be-
cause, for b ∈ Lm ,
X
kTm,∞ (b) − gm,∞ (b)k ≤ kbk · k(Tn − gn )|Ln k ≤ kbk · 21−m .
n≥m

The set {Tm,∞ (bk,m ) ; m, k ∈ N} is dense in indlim(Tn : Bn → Bn+1 ). The element

Tm,n (bk,m ) is in Ln for n > max(k, m). Thus, Tm,∞ (bk,m ) = Tn,∞ (Tm,n (bk,m )) ∈
Tn+1,∞ (Ln+1 ). It shows that indlim(Tn : Ln → Ln+1 ) is the same closed subspace of
B∞ as indlim(Tn : Bn → Bn+1 ). Since also gm,n (bk,m ) is in Ln for n > max(k, m),
a similar argument shows that
indlim(gn : Ln → Ln+1 ) = indlim(gn : Bn → Bn+1 ) .

It is obvious that the resulting isomorphisms are, e.g. , C *-algebra isomor-

phisms if the intertwining maps are e.g. C *-morphisms. 
 15. INTERSECTION WITH SUMS OF LEFT AND RIGHT IDEALS 1145

15. Intersection with sums of left and right ideals

The following Remark allows to circumvent in some cases the rather engaged
general (semi-constructive) reduction to relatively weakly injective separable sub-
spaces of operator spaces given in Section 14.

Remark A.15.1. The sum L + R ⊆ B of a closed left ideal L and a closed right
ideal R of a C *-algebra B is always a closed subspace of B, see Proposition A.15.2.
Thus, if D is a C *-subalgebra of B with the property that (D ∩ L) + (D ∩ R)
is dense in D ∩ (L + R) then automatically (D ∩ L) + (D ∩ R) = D ∩ (L + R).
Density of (D ∩ L) + (D ∩ R) in D ∩ (L + R) is a very special situation. The
following trivial example shows this:
Let B = C⊕C, L := C⊕0, R := 0⊕C and D := C·(1, 1). Then D∩(L+R) = D
but D ∩ L = {0} and D ∩ R = {0}.
But in the very particular case of Lemma B.14.1 where B is a C *-algebra,
G ⊆ B separable subset, X is a closed left ideal and Y is a closed right ideal of B,
we need only Part (2) of Sublemma B.14.2 to get a separable C *-algebra D with
G ⊆ D and (D ∩ X) + (D ∩ Y ) dense in D ∩ (X + Y ).
Then the equality follows from next Proposition A.15.2.

Proposition A.15.2. Let L a closed left ideal and R a closed right ideal of a
C*-algebra A. Consider the intersections D := L∗ ∩ L and E := R∗ ∩ R. Then

(i) D and E are hereditary C*-subalgebras A that are identical with the sets
of products L∗ · L = {a∗ b ; a, b ∈ L} = D, respectively R · R∗ = E .
(ii) L and R are the element-wise products L = A · D and R = E · A,
where, e.g. , A · D := { a · d ; d ∈ D , a ∈ A } .
(iii) Let η := πL∩R : a 7→ a + (L ∩ R) denotes the quotient map from A onto
A/(L ∩ R). For any x ∈ L and y ∈ R holds



max kη(x)k, kη(y)k ≤ kη(x + y)k ≤ 2 max kη(x)k, kη(y)k .

(iv) The algebraic sum L + R is a closed linear subspace of A.

(v) The second conjugate operator space of A/(L + R) is naturally isomorphic
to the ternary algebra (1 − q)A∗∗ (1 − p), where p and q are the “open”
support projections in A∗∗ for the hereditary C*-subalgebras D = L∗ ∩ L
and E = R∗ ∩ R of A. In particular,

kπL+R (a)k = k(1 − q)a(1 − p)k for all a ∈ A . (15.1)

Here L∗ (respectively R∗ ) mean the right ideal (respectively left ideal) given by
L∗ := {a∗ ; a ∈ L} and R∗ := {a∗ ; a ∈ R} , – not the dual spaces. Our approach
here ignores the general theory of operator spaces, because we try to derive all from
basic knowledge about Functional analysis, as e.g. the “Krein bi-polar theorem” for
convex subsets in locally convex vector spaces ... All that we show here prove in
the same way the corresponding results for the operator-matrix norms, because
1146 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

∼ Mn (A)/Mn (L) and Mn (A) ∼

e.g. Mn (A/L) = = Mn ⊗ A, ... etc. The following proof
is not minimal, because contains some additional observations.

Proof. Clearly, D := L∗ ∩ L = D2 because D is a C *-algebra and every

element of a C *-algebra is the product of two elements, e.g. by using polar decom-
position and roots of positive elements.
The closed left ideal L is a non-degenerate right module for the C *-algebra D.
Thus L = L · D by the Cohen factorization theorem A.11.1.
Then A · D ⊆ L, L = L · D and L ⊆ A give that A · D = L . Since D is a
hereditary C *-subalgebra, we get D = D3 ⊆ DAD ⊆ D. It follows D = D2 ⊆
L∗ · L = DA2 D ⊆ D.
Similar considerations, – or passage to the left ideal R∗ := {a∗ ; a ∈ R} –, show
also the corresponding results for closed right ideals of A and E := R∗ ∩ R.
Since (L ∩ R) · D ⊆ L ∩ R and E · (L ∩ R) ⊆ L ∩ R we get that

dist(f · x, L ∩ R) ≤ kf k · dist(x, L ∩ R)

for f ∈ E and x ∈ A, and

dist(x · g, L ∩ R) ≤ kgk · dist(x, L ∩ R)

for g ∈ D, and x ∈ A.
Let η := πL∩R . Then obviously kη(x + y)k ≤ kη(x)k + kη(y)k ≤


2 max kη(x)k, kη(y)k .


We show that max kη(x)k, kη(y)k ≤ kη(x + y)k :
Let x ∈ L and y ∈ R. Then η((x + y)(x∗ x)1/n ) = η(x(x∗ x)1/n ), because
y(x∗ x)1/n ∈ L ∩ R. We get an estimate

dist((x + y)(x∗ x)1/n , L ∩ R) ≤ kxk2/n dist((x + y), L ∩ R) ,

because (x∗ x)1/n ∈ D .

It follows that for all n ∈ N, kη(x(x∗ x)1/n )k = kη((x + y)(x∗ x)1/n )k ≤
kxk kη(x + y)k. On the other hand, kη(x)k ≤ kη(x(x∗ x)1/n )k + kx − x(x∗ x)1/n k.
2/n

Since limn→∞ kxk2/n = 1 and limn→∞ kx − x(x∗ x)1/n k = 0, we obtain that

kη(x)k ≤ kη(x + y)k. In the same way we get that kη(y)k ≤ kη(x + y)k.
It follows that the natural linear map from B := L/(L ∩ R) ⊕∞ R/(L ∩ R) onto
the image C := η(L + R) ⊆ A/(L ∩ R) of the sum L + R ⊆ A is a bijective linear
map that maps the Banach space B onto the normed vector space C. The above
given estimates show now that the norm on B induced by this map is equivalent to
the given complete norm on B. Thus, B is also a Banach space with this new norm,
and C is a Banach space with respect to the norm on C induced from A/(L ∩ R).
This implies that C is closed in A/(L ∩ R). It follows that L + R = η −1 (C) is closed
in A.
 15. INTERSECTION WITH SUMS OF LEFT AND RIGHT IDEALS 1147

Therefore A/(L + R) is a well-defined Banach space. It is in a natural way

an operator space, because we can in above observations A, L and R replace by
Mn (A) ∼
= Mn ⊗ A, Mn (L) and Mn (R).
The general theory of Banach spaces A shows that for closed linear subspaces
X of A the bi-dual space (A/X)∗∗ is naturally isomorphic to A∗∗ /wcl(X), where
wcl(X) is the σ(A∗∗ , A∗ )-closure of X. It is identical with the bi-polar X oo of X
in A∗∗ . It holds dist(a, X) = dist(a, X oo ) measured with the norm in A∗∗ . Thus
kπX (a)k = kπX oo (a)k for all a ∈ A.
In case of (matrix-normed) operator spaces this carries over to the matrix spaces
Mn (A), Mn (X) and Mn (A/X) ∼ = Mn (A)/Mn (X).
The σ(A∗∗ , A∗ ) closure of L + R in A∗∗ is A∗∗ p + qA∗∗ , where p, q ∈ A∗∗ are
the “open” projections corresponding to the hereditary C *-subalgebras D and E
respectively.
The second conjugate of A/(L + R) is the quotient of A∗∗ by the bi-polar of
L + R in A∗∗ . The bi-polar of L + R is just the σ(A∗∗ , A∗ )-closure of L + R in A∗∗ .
It contains the sum of A∗∗ p ∼ = L∗∗ and qA∗∗ ∼ = R∗∗ , and it is not difficult to see
that the unit-ball of A∗∗ p + q A∗∗ is σ(A∗∗ , A∗ )-compact. Here p and q are the open
projections in A∗∗ corresponding to the hereditary C *-subalgebras D = L∗ ∩ L and

∗∗
E := R∗ ∩R of A. Thus, A/(L+R) is natural isomorphic to A∗∗ /(A∗∗ p+q A∗∗ ).
It is easy to check that the norm of π(A∗∗ p+q A∗∗ ) (a) for a ∈ A∗∗ is equal to
k(1 − q)a(1 − p)k .
In particular, kπL+R (a)k = k(1 − q)a(1 − p)k for all a ∈ A .
In every W∗ -algebra M holds: If p, q ∈ M are projections, then

dist(a, M p + qM ) = k(1 − q)a(1 − p)k for all a ∈ M .

This is the case because (1 − q)a(1 − p) − a ∈ M p + qM and the map T : M 3 a 7→

(1 − q)a(1 − p) is a linear contraction with T 2 = T and kernel = M p + qM .
It follows that kπL+R (a)k = dist(a, A∗∗ p + qA∗∗ ) = k(1 − q)a(1 − p)k for all
a ∈ A. 

Corollary A.15.3. If L and R are closed left and right ideals of a C*-algebra
A and B ⊆ A a separable C*-subalgebra of A then there exists a separable C*-algebra
C ⊆ A such that B ⊆ C and

B ∩ (L + R) ⊆ (C ∩ L) + (C ∩ R) .

Proof. By Proposition A.15.2(iv,v), the linear subspace L + R is closed in A.

Thus B ∩ (L + R) is a closed separable subspace of L + R and of B. We find a dense
sequence {b1 , b2 , . . .} in the unit-ball of the separable Banach space B ∩ (L + R).
In L + R are elements cn ∈ L and dn ∈ R such that bn = cn + dn . The separable
C *-algebra C of A generated by B and {c1 , c2 , . . . ; d1 , d2 , . . .} has the properties
that B ⊆ C, and that C ∩ L and C ∩ R are closed left and right ideals of C, and, by
1148 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

Proposition A.15.2, (C ∩L)+(C +R) is a closed subspace of C that contains a dense

subset of B ∩ (L + R) ⊂ C. Thus, B ⊆ C and B ∩ (L + R) ⊆ (C ∩ L) + (C ∩ R) . 

16. Surjectivity and control of perturbations

Let E and F be Banach spaces and T a bounded linear map from E into
F . If X and Y are subsets of E, then cl(X) means the (norm-) closure of X,
wcl(X) ⊆ E ∗∗ is the σ(E ∗∗ , E ∗ )-closure of X in the second conjugate space E ∗∗ of
E, and distH (X, Y ) denotes the Hausdorff distance between subsets X, Y ⊆ E (or
X, Y ⊆ E ∗∗ or – more generally – subsets of any metric space E) that is given by

distH (X, Y ) := sup{ dist(a, Y ), dist(b, X) ; a ∈ X, b ∈ Y } .

We denote by S the open unit ball of E.

Does def. of f (x, T ) fit the case of T := π(c(.)d) ?
Suppose that there exists a function f (t) of t ∈ (0, 2] with the properties that
limt→0 f (t) = 0 and that if a, b ∈ S and kT (a) − T (b)k < x then dist(a, S ∩
T −1 (T (b)) ≤ f (x) .
This allows to show that T maps the closed unit ball S of E onto the closure
of T (S) :
Since T is bounded, T (S) is contained in the closure of T (S).
Let g1 , g2 , . . . ∈ T (S) ⊆ F a Cauchy sequence that converges to some h∞ ∈ F .
Find a subsequence h1 , h2 , . . . of (gn ) such that for each e ∈ T −1 (hn ) ∩ S there
exists e0 ∈ T −1 (hn+1 )∩S with ke0 −ek ≤ f (khn+1 −hn k) and n f (khn+1 −hn k) <
P

∞.
The latter requires only that limx→0 f (x) = 0 if we pass to a suitable sub-
sequence of (hn ) of (gn ).
Then one finds step-wise en ∈ T −1 (hn ) with n ken − en+1 k < ∞. It follows
P

that (en ) converges to some e∞ ∈ S and T (e∞ ) = h∞ .

We define a function f (x, T ) for x ∈ (0, ∞) by

Definition A.16.1.

f (x, T ) := sup{ distH S ∩ T −1 (T (a)), S ∩ T −1 (T (b)) ; a, b ∈ S, kT (a − b)k ≤ x } .

Clearly f (x, T ) is an increasing function of x ∈ (0, ∞). The perturbation con-

stant per(T ) will be defined by per(T ) := limx→0 f (x, T ) .

Proposition A.16.2. Let x 7→ f (x, T ) the above defined function.

(i) f (x, T ) is a continuous increasing function on ]0, ∞[.

(ii) per(T ) = 0 implies T (cl(S)) = cl(T (S)).
(iii) f (x, T ) = f (x, T ∗∗ ) for every x > 0.
 16. SURJECTIVITY AND CONTROL OF PERTURBATIONS 1149

Here T ∗∗ means the second adjoint operator of T . f (x, T ) is the infimum of all
numbers y > 0 such that given b ∈ S, d ∈ T (S) with kT (b) − dk ≤ x there exists
a perturbation b + h of b inside S such that khk ≤ y and T (b + h) = d. Thus,
per(T ) = 0 says that a small perturbation inside T (S) can be realized by a small
perturbation inside S. The proofs and further results will be given in
Section ?? (= in old A).
Now let A be a C *-algebra and L, R closed left and right ideals with open
support projections l and r in A∗∗ respectively, i.e., l, r are the open projec-
tions in A∗∗ satisfying A∗∗ l = wcl(L) and rA∗∗ = wcl(R), cf. [767, vol.I,
chp.III.,def.6.19,cor.6.20], [616, thm.3.10.7, prop.3,11.9, rem.3.11.10].
Put q = 1 − l and p = 1 − r. Let b, d be elements of A∗∗ in the multiplier
algebra M(A) of A such that pbb∗ p is invertible in pA∗∗ p with inverse g and qd∗ dq
is invertible in qA∗∗ q with inverse h.
The algebraic sum R + L is a closed linear subspace of A, wcl(L + R) = A∗∗ l +
rA∗∗ and kπL+R (a)k = k(1 − r)a(1 − l)k for all a ∈∈ A by Proposition A.15.2.
We denote by π = πL,R the quotient map A → A/(L+R) given by c → c+L+R
and denote by π(b(.)d) the map given by c → bcd + L + R for c ∈ A.

Proposition A.16.3. f (x, T ) ≤ x(kgk khk)1/2 + (2x(kgk khk)1/2 )1/2 , where

T := π(b(.)d).

The Proof follows from Corollaries ?? (old cor.A.2.9 ???) and ?? (old cor.A.3.2
???) (cf. Section ?? old A.4?).
We obtain from Propositions A.16.2(ii) and A.16.3 immediately the following
corollary:

Corollary A.16.4. Under the above assumptions concerning A, L, R, b and

d the quotient map A → A/(L + R) maps b(cl(S))d onto a closed set.
In particular the quotient map maps the closed unit ball of A onto the closed
unit ball of A/(L + R).

For every positive integer n, the map [bj,k ]n → [pbj,k q]n from Mn (A∗∗ ) ∼
=
Mn (A)∗∗ onto Mn (pA∗∗ q) is a contraction and has kernel Mn (A∗∗ l + rA∗∗ ) =
Mn (wcl(L+R)) ∼= wcl(Mn (L+R)); passage to the quotient space defines an isomet-
ric Mn –bimodule isomorphism from (Mn (A)/Mn (L+R))∗∗ ∼ = Mn (A∗∗ )/Mn (wcl(L+
R)) onto Mn (pA∗∗ q). On the other hand there is a natural Mn –bimodule isomor-
phism from Mn (A/(L + R)) onto Mn (A)/Mn (L + R). The matrix norms induced
by these isomorphisms give A/(L + R) the structure of a matrix normed space in
the sense of
Effros [EF2]
such that the second conjugate matrix normed space is completely isometrically
isomorphic to pA∗∗ q, an operator subspace of A∗∗ and C *-triple system,
cf. Section ??.
1150 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

If moreover A is unital and L = {b∗ : b ∈ R} then p = q and, under the above

identifications, A/(L + R) = pAp ⊆ pA∗∗ p becomes a matrix order unit space in
the sense of [235] with matrix order unit p ∈ A/(L + R) such that the second
conjugate matrix order unit space is just the unital C *-algebra pA∗∗ p, thus then
A/(L + R) is an operator space in the sense of [245]. More generally we call a
matrix order unit space X a C*-system if its second conjugate matrix order unit
space X ∗∗ is unitally matrix order isomorphic to a unital C *-algebra. Then X is
an operator system in the sense of
??? ??? [C/E2],
i.e., a closed unital and selfadjoint linear subspace of a C *-algebra together
with the matrix order inherited from the inclusion. The C *-algebra structure on
X ∗∗ is uniquely determined as the second conjugate matrix order unit structure of
the given one on X and we can define the left multiplier algebra M` (X), the right
multiplier algebra Mr (X) and the multiplier algebra M(X) of X as follows:
M` (X) := {b ∈ X ∗∗ : bX ⊆ X}, Mr (X) := {b ∈ X ∗∗ : Xb ⊆ X}, M(X) :=
M` (X) ∩ Mr (X). Here we identify X with its canonical and isometric image in
X ∗∗ by the evaluation map evX : X → X ∗∗ . The algebras M` (X) and Mr (X)
are closed subalgebras of X ∗∗ contained in X (more precisely: in the image of
evX : X → X ∗∗ ), M` (X) = {b∗ : b ∈ Mr (X)} and the multiplier algebra M(X)
of X is a unital C *-algebra which is unitally completely positively and completely
isometrically contained in the operator system X. In our special case A/(L + R) is
identified with pAp(⊆ pA∗∗ p ⊆ A∗∗ ) and Mr (A/(L + R)) = {pbp ; b ∈ A, pApbp ⊆
pAp}, M(A/(L + R)) = {pbp ; b ∈ A, pbpAp + pApbp ⊆ pAp}.
Let D be a hereditary C *-subalgebra of A (i.e. D closed, selfadjoint and DAD ⊆
D). An element b ∈ A that satisfies Db ⊆ D (respectively bD ⊆ D, bD + Db ⊆ D)
a right normalizer (respectively left normalizer, normalizer ) of D in A. The right
normalizers, left normalizers, normalizers obviously form closed operator algebras
N r (D), N ` (D) and N (D) respectively, N r (D) = {b∗ : b ∈ N ` (D)} and N (D) =
N ` (D) ∩ N r (D) is a C *-subalgebra of A. Notice that L = A · D and R = D · A are
closed left and right ideals of A, respectively, whose support projections in A∗∗ are
equal that of D (i.e., are equal to the unit element of D∗∗ = wcl(D) ⊆ A∗∗ ). From
definitions we see that L = cl(AD) = A · D ⊆ N r (D), R = cl(DA) ⊆ N ` (D) and
D ⊆ N (D) are closed ideals of N r (D), N ` (D) and N (D) respectively.
We define A//D := A/(cl(AD) + cl(DA)), the (unital) quotient–C*-system of
the unital C *-algebra A with respect to the hereditary C*-subalgebra D of A. We
denote again by πD : A → A//D the quotient map b 7→ b + cl(AD) + cl(DA). Now
we are in position to state the main result of this section.

Theorem A.16.5. Let A be a unital C*-algebra. D a hereditary C*-subalgebra

of A, A//D := A/(AD + DA) the quotient–C*-system of A with respect to D and
πD : A → A//D the quotient map.
 16. SURJECTIVITY AND CONTROL OF PERTURBATIONS 1151

(i) The restriction of πD to N r (D) (resp. to N ` (D)) is a Banach algebra

epimorphism onto Mr (A//D) (resp. onto M` (A//D)) with kernel cl(AD)
(resp. cl(DA)),
(ii) for every positive integer n, πD ⊗ idn maps the closed unit ball of

N r (Mn (D)) = Mn (N r (D)) ⊆ Mn (A)

onto the closed unit ball of

Mn (Mr (A//D)) ⊆ Mn (A//D) ∼

= Mn (A)//Mn (D) ,

(iii) πD | N (D) is a C*-algebra epimorphism from the (two-sided) normalizer

algebra N (D) ⊆ A of D onto the multiplier algebra M(A//D) of A//D
with kernel ideal ker(πD | N (D)) = D .

As a corollary we get the following which can be seen as alternative formulation

of
REFERENCE ?
Theorem ?? in view of Proposition ??.

Corollary A.16.6. Let A and C be unital C*-algebras, B a unital closed

subalgebra of C and V : A//D → C a unital completely isometric map such that
B ∪ {b∗ b : b ∈ B} ⊆ Im(V ). Then there exists a unital closed subalgebra E of A
such that

(i) E ∩ (cl(AD) + cl(DA)) = cl(AD),

(ii) V ◦ πD |E is a Banach algebra epimorphism form E onto B with kernel
cl(AD) and
(iii) the induced map [V ◦ πD ]0 : E/cl(AD) → B is completely isometric, where
B is equipped with the matrix norms induced by C and E/cl(AD) is
equipped with the matrix norms induced by the inclusion A/cl(AD) ⊆
A∗∗ p.

If moreover B is a C*-subalgebra of C then there exists a unital C*-subalgebra F

of A such that

(iv) F ∩ (cl(DA) + cl(AD)) = D and

(v) V ◦ πD |F is a C*-algebra epimorphism from F onto B with kernel D.

With other words and under the assumptions of Corollary A.16.6: B is a C *-

quotient algebra of a C *-subalgebra of A if B is a C *-subalgebra of C. Theorem
?? and Corollary ?? are proven in Section ??.
??? Give ref. to suitable place In a forthcoming paper we shall show:
The assumptions of Corollary ?? are satisfied with the CAR-algebra A := M2∞
and with D,C and V suitably chosen if and only if B is separable and exact in the
sense of [KI2], cf. [KI4].
1152 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

Notice that all proofs work both in real and complex case. There are shorter
proofs if one is interested only in the results of Theorem ??(iii) and Corollary
??(iv,v).
The results presented here are essentially reworking of material which has been
circulating as preprint since 1989. Results similar to those of Corollary ???1.3???
and Theorem ???1.4(i)??? are also contained in preprints of L. Brown [BR], where
he gives proofs using different methods and estimates.

17. On perturbation of unitaries

Change here all signs to Halmos unitary !

If b is a contraction on a Hilbert space we denote by U (b) ∈ M2 (A) the “Halmos
unitary” matrix defined in Remark 4.2.4, i.e., the matrix
" #
b , −(1 − bb∗ )1/2
.
(1 − b∗ b)1/2 , b∗

We are going to prove the following Proposition A.17.1 and its Corollary A.17.2.

Proposition A.17.1. Let A denote a complex (or real) unital C*-algebra, u

a unitary in A and p, q a pair of projections in A such that kpuqk < 1. Then for
every contraction b in pAq there exists a unitary ũ in A such that pũq = b and
ku − ũk = kU (puq) − U (b)k .

Corollary A.17.2. Let A be a unital C*-algebra (real or complex), p, q

nonzero projections in A and T the natural map from A onto pAq given by
T (a) = paq (a in A). Then the perturbation function f (x, T ) of T , cf. Definition
A.16.1, can be estimated by f (x, T ) ≤ x + (2x)1/2 .

We need some preliminary lemmata. To simplify notation, let diag(a, b, . . .)

denote the diagonal matrix with diagonal elements a, b, . . . , let M (c) be the matrix
" #
c , −(p − cc∗ )1/2
(q − c∗ c)1/2 , c∗

if c is a contraction in pAq and let us denote by Z the matrix

" #
0, 1
.
1, 0

Lemma A.17.3. Let A be a C*-algebra, p, q projections in A, a in pAq, d in

qAp such that a and
" #
a , −(p − aa∗ )1/2
N :=
(q − a∗ a)1/2 , d
are contractions in M2 (A).
Then M (a) satisfies M (a)∗ M (a) = diag(q, p) , M (a)M (a)∗ = diag(p, q) .
If moreover kak < 1 then d = a∗ .
 17. ON PERTURBATION OF UNITARIES 1153

Proof. (1 − aa∗ )1/2 p = p(1 − aa∗ )1/2 = (p − aa∗ )1/2 and q(1 − a∗ a)1/2 =
(1 − a∗ a)1/2 q = (q − a∗ a)1/2 because aa∗ ≤ p, aa∗ ≤ q and paq = a . It follows that
a∗ (p − aa∗ )1/2 = (q − a∗ a)1/2 a∗ , (p − aa∗ )1/2 a = a(q − a∗ a)1/2 , and (q − a∗ a)1/2 d =
(1−a∗ a)1/2 d. Put M := M (a). Using these identities straightforward computations
show M ∗ M = diag(q, p) and M M ∗ = diag(p, q) . Moreover
M ∗ N = diag(q, p − aa∗ − ad) + diag(c, 0)Z
where c := a∗ (p − aa∗ )1/2 + (q − a∗ a)1/2 d. But kM ∗ N k ≤ kM ∗ k · kN k ≤ 1 by
the assumptions on N and the above observations concerning M . Looking to the
(1, 1)-element of (M ∗ N )(M ∗ N )∗ we obtain kq + cc∗ k ≤ 1. On the other hand
c = qc by the above identities. Thus q + cc∗ = q(q + cc∗ )q ≤ q, i.e., c = 0. It
follows:
(1 − a∗ a)1/2 (−a∗ ) = −a∗ (p − aa∗ )1/2 = (q − a∗ a)1/2 d = (1 − a∗ a)1/2 d .
If kak < 1 then (1 − a∗ a)1/2 is invertible and d = −a∗ . 

Lemma A.17.4. Let be A, a, p and q as in Lemma A.17.3 and kak < 1. If the
matrix  
a, (p − aa∗ )1/2 , g
V :=  (q − a∗ a)1/2 , d, h 
 

f, k, e
is a partial isometry in M3 (A) such that V ∗ V = diag(q, p, s) and V V ∗ =
diag(p, q, r) then f = g = h = k = 0 , d = −a∗ , e∗ e = s and ee∗ = r.

Proof. The upper left 2×2-sub-matrix of V must be a contraction in M2 (A).

By Lemma A.17.3, d = −a∗ because kak < 1. Using now the equality V V ∗ =
diag(p, q, r) from d = −a∗ we get gg ∗ = 0, hh∗ = 0, and using V ∗ V = diag(q, p, s),
we obtain f ∗ f = 0, k ∗ k = 0. From f = g = h = k = 0 it follows that e∗ e = s and
ee∗ = r . 

Lemma A.17.5. Let p, q, r, s be projections in a unital C*-algebra A and v, w ∈

A such that v ∗ v = q, ww∗ = p, p + r + vv ∗ = 1 and q + s + w∗ w = 1. Put
     
p, v, r q, w∗ , s y, 0, 0
D :=  0, 0, 0  , E :=  0, 0, 0  , F (y) :=  0, 0, 0  .
     

0, 0, 0 0, 0, 0 0, 0, 0
And G(y) := D∗ F (y)D , H(y) := E ∗ F (y)E , T (y) := D∗ F (y)E if y is in A.
Then, for each y ∈ A,

(i)  
pyq, pyw∗ , pys
T (y) =  v ∗ yq, v ∗ yw∗ , v ∗ ys  .
 

ryq, ryw∗ , rys

(ii) G and H are injective C*-morphisms from A into M3 (A) such that G(1) =
diag(p, q, r) and H(1) = diag(q, p, s).
(iii) T is an isometry from A into M3 (A) such that T (y)T (y)∗ = G(yy ∗ ) and
T (y)∗ T (y) = H(y ∗ y) for y ∈ A.
1154 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

(iv)
 
a, c, 0
T (a + vb + cw + vdw + e) =  b, d, 0 
 

0, 0, e
if e ∈ rAs, a ∈ pAq, b ∈ qAq, c ∈ pAp and d ∈ qAp.
(v) If y ∈ A then the equality
 
a, c, 0
T (y) =  b, d, 0 
 

0, 0, e
implies y = a+vb+cw +vdw +e with e ∈ rAs, a ∈ pAq, b ∈ qAq, c ∈ pAp
and d ∈ qAp.
(vi) An element y = a+vb+cw+vdw+e satisfying the conditions of Part(iv) is
a unitary of A if and only if the upper left 2×2–submatrix is a partial isom-
etry (say W ) in M2 (A) with W ∗ W = diag(q, p) and W W ∗ = diag(p, q)
and e is a partial isometry with e∗ e = s and ee∗ = r .

Proof. (i) is obvious.

(ii,iii): We denote M3 (A) by B and the projection diag(1, 0, 0) = F (1) by
P . Then F defines a unital C *-algebra isomorphism from A onto P BP . By our
assumptions DD∗ = P = EE ∗ . Thus z → D∗ zE, z → D∗ zD and z → E ∗ zE
define linear isometries from P BP onto D∗ DBE ∗ E, D∗ DBD∗ D and E ∗ EBE ∗ E
respectively. Moreover the latter two are unital C *-algebra isomorphisms.

T (y)T (y)∗ = D∗ F (y)EE ∗ F (y ∗ )D = D∗ F (yy ∗ )D = G(yy ∗ )

and T (y)∗ T (y) = H(y ∗ y). Hence T is an isometry from A onto D∗ DBE ∗ E and
G, H are unital *-isomorphisms from A onto D∗ DBD∗ D and E ∗ EBE ∗ E respec-
tively. Computations show that G(1) = D∗ D = diag(p, q, r) and H(1) = E ∗ E =
diag(q, p, s).
(iv): By the assumptions on p, q, r, s, w, v we have ws = qs = rv = rp = pv =
wq = 0 . Put y := a + vb + cw + vdw + e . Then by the assumptions on a, b, c, d, e
we have ry = e = ys = rys. With g = y − e by (i) it follows that
 
pgq, pgw∗ , 0
T (y) =  v ∗ gq, v ∗ gw∗ , 0  .
 

0, 0, e
Using pv = wq = 0 we get pgq = paq = a, pgw∗ = cww∗ = cp = c, v ∗ gq = v ∗ vb =
qb = b and v ∗ gw∗ = v ∗ vdww∗ = qdp = d.
(v): As we have seen in the proofs of (ii) and (iii), T defines an isometry from
A onto diag(p, q, r)B diag(q, p, s).
From T (y) = diag(p, q, r)T (y) diag(q, p, s) we see that a = paq, b = qbq, etc.
Put z := a + vb + cw + vdw + e. Then by (iv), T (z) = T (y). But ker(T ) = 0, i.e.
y = z.
 17. ON PERTURBATION OF UNITARIES 1155

(vi): By (ii), (iii) and (iv),

W ∗ W = diag(q, p), W W ∗ = diag(p, q), e∗ e = s and ee∗ = r if and only if

T (y)∗ T (y) = H(1) and T (y)T (y)∗ = G(1) if and only if
H(1 − y ∗ y) = 0 and G(1 − yy ∗ ) = 0 if and only if
y ∗ y = 1 = yy ∗ .

Proof of Proposition A.17.1. :

We put a = puq. Then kak < 1, (1 − a∗ a)−1/2 and (1 − aa∗ )−1/2 exist. Let
v = (1 − p)uq(1 − a∗ a)−1/2 and w = (1 − aa∗ )−1/2 pu(1 − q). We have v ∗ v =
(1 − a∗ a)−1/2 qu∗ (1 − p)uq(1 − a∗ a)−1/2 = (1 − a∗ a)−1/2 (q − a∗ a)(1 − a∗ a)−1/2 = q
because qa∗ a = a∗ a = a∗ aq. Similarly we obtain ww∗ = p. In particular w and v
are partial isometries. By definitions of v and w we have (1−p)v = v, w(1−q) = w,

(1 − p)uq = v(1 − a∗ a)1/2 = vq(1 − a∗ a)1/2 = v(q − a∗ a)1/2 ,

and pu(1 − q) = (p − aa∗ )1/2 w. Thus

vv ∗ ≤ 1 − p, w∗ w ≤ 1 − q, v ∗ uq = (q − a∗ a)1/2

and puw∗ = (p − aa∗ )1/2 . Put r = 1 − p − vv ∗ and s = 1 − q − w∗ w. Then

p, v, r, q, w, s satisfy the assumptions of Lemma A.17.5 and T (u) defined there satis-
fies T (u)∗ T (u) = H(u∗ u) = H(1) = diag(q, p, s), T (u)T (u)∗ = G(1) = diag(p, q, r)
by Lemma A.17.5(ii,iii). Moreover, by Lemma A.17.5(i) and the above equations,
the Lemma A.17.4 applies to T (u):
 
a, (p − aa∗ )1/2 , 0
T (u) =  (q − a∗ a)1/2 , −a∗ , 0 
 

0, 0, e

with e∗ e = s, ee∗ = r. We put ũ = b + v(q − b∗ b)1/2 + (p − bb∗ )1/2 w − vb∗ w + e.

By Lemma A.17.5(iv),
 
b, (p − bb∗ )1/2 , 0
T (ũ) =  (q − b∗ b)1/2 , −b∗ , 0 .
 

0, 0, e

By Lemma A.17.3 and Lemma A.17.5(vi), the element ũ is a unitary in A. We have

ku − ũk = kT (u − ũ)k = kM (a) − M (b)k.
Now if c ∈ pAq then U (c) = M (c) + diag(1 − p, 1 − q)Z.
Thus ku − ũk = kU (a) − U (b)k = kU (puq) − U (b)k. 

Lemma A.17.6. Let be a, b contractions and h, k positive selfadjoint operators

on a Hilbert space. Then

(i) kh1/2 − k 1/2 k ≤ kh − kk1/2 ,

(ii) ka∗ a − b∗ bk ≤ 2ka − bk,
(iii) kU (a) − U (b)k ≤ ka − bk + (2ka − bk)1/2 .
1156 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

Proof. (i): Let t = kh − kk. Then h + t ≤ (h1/2 + t1/2 )2 and k ≤ h + t. The

function g(t) = t1/2 is operator monotone on [0, ∞), cf. [767, Prop.I.6.3].
Thus k 1/2 ≤ (h + t)1/2 ≤ h1/2 + t1/2 and we can interchange k and h in this
inequality.
(ii): a∗ a − b∗ b = a∗ (a − b) + (a − b)∗ b.
(iii): Let

c := diag (1 − aa∗ )1/2 − (1 − bb∗ )1/2 , (1 − a∗ a)1/2 − (1 − b∗ b)1/2 .

By (i) and (ii) we get kck ≤ (2ka − bk)1/2 .

We have U (a) − U (b) = diag(a − b, (a − b)∗ ) + cZ. 

Proof of Corollary A.17.2. :

Let be a, b ∈ A and x > 0 such that pbq = b, kak < 1, kbk < 1 and kpaq − bk < x.
Define P := diag(p, 0), Q := diag(q, 0), B = diag(b, 0) and u := U (a) (almost
Halmos unitary). Then u is a unitary in M2 (A) such that P uQ = diag(paq, 0),
kP uQk ≤ kak < 1 and kP uQ − Bk = kpaq − bk < x .
Moreover P BQ = B and kBk = kbk < 1. By Proposition A.17.1 there exists
a unitary ũ ∈ M2 (A) such that ku − ũk = kU (P uQ) − U (B)k and diag(pcq, 0) =
P ũQ = B = diag(b, 0) where c means the (1,1)–element of the unitary 2 × 2–matrix
ũ. Looking at the (1,1)–element of u − ũ by Lemma A.17.6(iii) we obtain ka − ck ≤
ku − ũk ≤ kU (P uQ) − U (B)k < x + (2x)1/2 . On the other hand, kck ≤ kũk = 1 and
pcq = b. Now let 0 < t < 2. Put s = 1 − (t/2) and e = (p + s(1 − p))c(q + s(1 − q)).
Then peq = pcq = b, ke − ck ≤ 2(1 − s) = t, ka − ek < t + x + (2x)1/2 ,

k(p + s(1 − p))cqk ≤ (1 − s)kbk + skcqk ≤ 1 − (1 − s)(1 − kbk)

and kek ≤ 1 − (1 − s)2 (1 − kbk) < 1.

Thus, f (x, T ) ≤ x + (2x)1/2 by Lemma 2.3(i) (of some other paper!) ?. 

18. Open projections, C *-spaces and Kadison transitivity

Others:
C *-space,
(unital) C *-system,
Non-unital C *-system,
adding a unit,
inductive limits of C *-systems,
nuclear C *-systems: nuclearity criteria,
inductive limits with residually nuclear maps,

Check if ‘‘quasi-state’’ can be mixed up with ‘‘quasi-trace state’’

???
Partly TRANSPORT from old Chp. 2:
 18. OPEN PROJECTIONS, C *-SPACES AND KADISON TRANSITIVITY 1157

List of equivalent properties

of open projections is given below.

Definition A.18.1. Let A a C *-algebra.

The projection p ∈ A∗∗ is open (with respect to A) if A∗∗ p is the σ(A∗∗ , A)-
closure of the left ideal Lp := {a ∈ A ; a(1 − p) = 0}.
A projection q ∈ A∗∗ is closed if p := 1 − q is an open projection.

The following proposition given equivalent characterizations of open projections

p ∈ A∗∗ , respectively of closed projections q = 1 − p ∈ A∗∗ . (We use for a linear
functional ϕ ∈ A and its normal extension to A∗∗ the same notation ϕ, i.e., we
identify A∗ and the predual (A∗∗ )∗ of the bi-dual W*-algebra A∗∗ in a natural
way.)

Proposition A.18.2. Let A a C*-algebra, p ∈ A∗∗ a projection and q := 1 − p.

The following are equivalent:

(i ) p is open (i.e., q is closed).

(ii ) p is the (unique) smallest projection in A∗∗ with ap = a for all a ∈ A+
with a(1 − p) = 0.
(iii ) The convex cone C(q) of all normal positive functionals on A∗∗ with
ϕ(q) = kϕk (i.e., with ϕ(p) = 0) is σ(A∗ , A)-closed in A∗ .
(iv ) The convex subset

Q(q) := {ϕ ∈ (A∗ )+ ; ϕ(q) = 0 , kϕk ≤ 1}

of the quasi-state space Q(A) ⊆ A∗ of A is σ(A∗ , A)-closed (i.e., is a

weakly closed face of Q(A)).
(v ) The left A∗∗ -module A∗ q is σ(A∗ , A)-closed in A∗ .
(vi ) A∗∗ p is the σ(A∗∗ , A∗ )-closure of A ∩ (A∗∗ p).
(vii) The projection p is the least upper bound in A∗∗ for the set {a ∈ A+ ; ap =
a, kak ≤ 1} of positive contractions.
(viii) The projection p ∈ A∗∗ is the support projection of the hereditary C*-
subalgebra D ⊆ A of A defined by D := {a ∈ A ; (aa∗ + a∗ a)q = 0 }.
(ix) This property is not? sufficient ?:
For every pure state ϕ of A with ϕ(p) = 1, (i.e., ϕ(q) = 0) there exists
a ∈ A+ with kak = 1, ap = p and ϕ(a) = 1.
(It delivers that p ≥ q for some open projection. Let Lp = {b ∈
A b(1 − p) = 0} it is a left ideal with open support projection
Hope? Perhaps next has to be added:
If ψ is a quasi-state on A with ψ(p) = 1 then there exists a pure state
ϕ in the weakly closed face of Q(A) generated by ψ with ϕ(p) 6= 0.
Or: p is determined by its discrete part.

The equivalences of Parts (i)-(viii) are – at least implicitly – contained in the

book of G.K. Pedersen, cf. [616, thm. 3.6.11,thm. 3.10.7] and its proofs, or are
1158 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

obvious reformulations of each other. We prove here only the equivalence of the
Parts (viii) and (ix).

Proof. (viii)⇒(ix): If p ∈ A∗∗ is an open projection and ϕ ∈ A∗+ is a pure

state on A with ϕ(1 − q) = 0, then ϕ|D is a state on D.
This restriction to D is pure on D: If ψ1 and ψ2 are states on D with ψ1 + ψ2 =
2ϕ then this happens also for the all (the unique) normal extensions ψk of ψk to
states on A∗∗ and implies that ψ1 = ψ2 = ϕ.
By ????? excision lemma??? reference ???? there exists for the pure state ϕ|D
an positive contraction a ∈ D+ with ϕ(a) = 1. Clearly, 1 = ϕ(a) ≤ kak ≤ 1, and
ap = a. because a ∈ D+ .
(ix)⇒(viii): Suppose that for every pure state ϕ of A with ϕ(p) = 1, (i.e.,
ϕ(q) = 0) there exists a ∈ A+ with kak = 1, ap = p and ϕ(a) = 1.
Let L := {a ∈ A ; a(1 − p) = 0} = {a ∈ A ; ap = a} . Then L is a closed
left ideal of A and the positive contractions in L generate L. Clearly the open
support projection q ∈ A of the hereditary C *-subalgebra L∗ ∩ L satisfies again
(1 − p)q(1 − p) = 0. Thus qp = q and r ≤ q for all open projections r ∈ A∗∗ with
r(1 − p) = 0. 

Notice that Part (ix) of Proposition A.18.2 is a non-commutative version of the

original definition of F. Hausdorff of open subsets U ⊆ X of a metric space X: U
is open, if and only if, for each point x ∈ U there exists a (bounded) continuous
function f ∈ Cb (X)+ with f (x) = 1, kf k ≤ 1 such that the support of f is contained
in U .
Notice that ???????
Before next:
Here we have first to show that
a ∈ A 7→ a + (R + L) ∈ A/(R + L) maps
the closed unit-ball onto the closed unit ball.!!!
One of the obvious consequences is the following Corollary is equivalent to the
“advanced Kadison transitivity” in Lemma 2.1.15(i,ii).

Corollary A.18.3. If p ∈ A∗∗ is a projection such that pA∗∗ p has finite

dimension (as vector space) then p is a closed projection and there is a closed left-
ideal L ⊆ A such that A∗∗ (1 − p) is the σ(A∗∗ , A∗ )-closure of L in A∗∗ .
In particular, the natural map a ∈ A → pap defines an isometric isomorphism
from A/(L∗ + L) onto pA∗∗ p, that maps the closed unit-ball of A onto the closed
unit-ball of pA∗∗ p.

It is well known for Banach spaces that πX maps the open unit-ball of a Banach
space A maps onto the open unit ball of A/X for each closed linear subspace of X.
Notice that R + L is a closed subspace of a C *-algebra A for any closed left-
ideals L and right ideals R, cf. Proposition A.15.2.
 18. OPEN PROJECTIONS, C *-SPACES AND KADISON TRANSITIVITY 1159

The bipolar of R + L (= σ(A∗∗ , A∗ )-closure of R + L) is identical with A∗∗ P +

QA∗∗ , where P and Q are the open support projections of the hereditary C *-
subalgebras R∗ ∩ R and L∗ ∩ L. The second adjoint of (A/(R + L))∗∗ of A/(R + L)
is naturally isomorphic to A∗∗ /(A∗∗ P + QA∗∗ ). All this spaces are matrix-normed
operator spaces and the canonical isomorphisms are complete isometries. It is easy
to see that A∗∗ /(A∗∗ P + QA∗∗ ) is naturally completely isometric isomorphic to the
ternary C *-algebra (1 − Q)A∗∗ (1 − P ).

Remark A.18.4. The Proposition A.18.6 and Theorem ??

main content:
(A//D) ∩ M(A//D) ∼
= N (D)/D
yields an alternative, more general, conceptual and algebraic proof of the exis-
tence of a C *-subalgebra B ⊆ A with the property that a ∈ B 7→ pap ∈ pA∗∗ p is
a *-epimorphism onto pA∗∗ p if pA∗∗ p is finite-dimensional, as derived in Remark
A.20.1 from the Kadison transitivity theorem.
In fact the proof is completely independent from the use of the non-commutative
Lusin theorem [616, thm. 2.7.3] and is independent from all textbook versions of the
Kadison transitivity theorem. It gives an alternative and “almost algebraic” proof
of the Kadison transitivity theorem without any approximation arguments. (But
uses the uniform Hölder continuity of that what we call “restricted perturbation”).
Let p∗ = p ∈ A∗∗ a non-zero projection in the socle of A∗∗ , i.e., pA∗∗ p is finite-
dimensional. The set Sp (A) of f ∈ A∗ with kf k ≤ 1 = f (p) is identical with the
set of all normal states on pA∗∗ p. The set of states on pA∗∗ p is compact in norm
of pA∗ p ⊆ A∗ if pA∗∗ p is finite dimensional.
Since the σ(A∗ , A)-topology is Hausdorff and is continuous with respect to the
norm on A∗ , it follows that Sp (A) is compact in the σ(A∗ , A)-topology.
By ??? Proposition ??, the σ(A∗ , A)-compactness of Sp (A) = Cp (A∗∗ )
It seems better to consider the
quasi-states
in Q(A) ?... No consequent terminology!!!
ρ on A with ρ(1 − p) = 0. And then require that this set is σ(A∗ , A) closed.
Then apply then bi-polar theorem ...
Give -- or refer to -- Definition of the convex set
Cp (M ) := {f ∈ M∗ ; kf k = 1, f (p) = 1}.

Open-ness of 1 − p is not proven now!!!

implies that p ∈ A∗∗ is a closed projection, i.e., there is a closed left ideal L ⊆ A
such that A∗∗ (1 − p) is the σ(A∗∗ , A∗ )-closure of L.
Then p is a closed projection in A∗∗ with respect to A, i.e., the hereditary C *-
subalgebra D := {a ∈ A ; ap = 0 = pa} has (1 − p) as its open support projection
1160 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

∼ D∗∗ is equal to the σ(A∗∗ , A∗ )-

qD of D in A∗∗ , respectively (1 − p)A∗∗ (1 − p) =
closure of D ⊆ A∗∗ . We let L := A · D = span(A · D). It is a closed left-ideal of A
with σ(A∗∗ , A∗ )-closure equal to A∗∗ (1 − p).
If L is a closed left-ideal of A, and q ∈ A∗∗ is the open support projection of
L, i.e.,
σ(A∗∗ ,A∗ )
L∗∗ ∼= A∗∗ q = L ,
then there is a natural isomorphism pAp ∼ = (A/(R + L))∗∗ for R := {a∗ ; a ∈ L}
and p := 1 − qL .
Let D := R ∩ L. The C *-space A//D := A/(R + L) ∼
= pAp ⊆ A∗∗ is c.p. and
completely isometrically isomorphic to pAp ⊆ pA p and (A//D)∗∗ ∼
∗∗
= pA∗∗ p as
matrix-normed and matrix-ordered spaces. Since pA∗∗ p has finite dimension, it
follows that (A//D) ∼
= pAp = pA∗∗ p.
By Theorem ?? the quotient N (D)/D by D of the (two-sided) normalizer C *-
algebra N (D) := {a ∈ A ; aD ∪ Da ⊆ A} of D is naturally isomorphic to the
intersection pAp ∩ M(A//D) of A//D ∼= pAp with the C *-subalgebra
M(A//D) := {T ∈ pA∗∗ p ; T pAp ∪ pApT ⊆ pAp}
of two-sided multipliers of the subset A//D ⊆ (A//D)∗∗ ∼
= pA∗∗ p .
Since pAp ∼= A//D is σ(A∗∗ , A∗ ) dense in pA∗∗ p ∼
= (A//D)∗∗ and, in our special
case, pA p is of finite dimension, we get that M(A//D) = pAp = pA∗∗ p. Thus,
∗∗

the restriction V |B to B := N (D) of the c.p. contraction V : a ∈ A 7→ pap ∈ pA∗∗ p

is a *-epimorphism from B onto pA∗∗ p ∼ = M(A//D) ∼ = N (D)/D.

New attempt Dec.2016:

Define a projection q = q ∗ q ∈ A∗∗ as open projection if q is the support
projection q := qD of a hereditary C *-subalgebra D of A.
A projection p ∈ A∗∗ is - by definition - a closed projection if 1 − p ∈ A∗∗ is
open.
Lemma A.18.5. Let p ∈ A∗∗ a projection.
The projection q := 1 − p ∈ A∗∗ is open, if an only if, the set Sp (A) of f ∈ A∗
with kf k = f (p) ≤ 1 is σ(A∗ , A)-closed and for each pure state g ∈ A∗ with g(p) = 0
there exists an element a ∈ A+ with kak = g(a) = 1 and f (a) = 0 for all f ∈ Sp (A).

Proof. If the projection 1 − p is open then the set of b ∈ A with bp = 0 is a

closed left ideal L of A and q := 1−p ∈ A∗∗ is the support projection of D := L∗ ∩L
(this by definition of open projections).
In particular f (a) = 0 for all f ∈ Sp (A) and a ∈ D, and conversely f (p) = 1 for
all normal states f on A∗∗ with f (D) = 0. Clearly, the set of normal states with
f (D) = {0} is closed in the σ(A∗ , A)-topology.
Thus, if g ∈ A∗ is a pure state with g(p) = 0 then g|D is a pure state of D. It
follows that there exists a contraction a ∈ D+ with g(a) = 1, cf. [616, thm. 2.7.5].
Thus f (a) = 0 for all f ∈ Sp (A).
 18. OPEN PROJECTIONS, C *-SPACES AND KADISON TRANSITIVITY 1161

Notice that generally a ∈ A+ and f (a) = 0 for all f ∈ Sp (A) is equivalent to

a(1 − p) = a.
The set L := {b ∈ A ; f (b∗ b) = 0 ∀ f ∈ Sp (A)} is the maximal closed left-ideal
of A with the property L · p = {0} .
The support projection Q ∈ A∗∗ of the hereditary C *-subalgebra D := L∗ ∩ L
of A is an open projection with Qp = 0.
Suppose that Q 6= q := 1 − p. Then there exists a pure state ρ ∈ A∗ with
ρ(Q) = 0 and ρ(p) < 1:
This is the case because S1−Q (A) is a σ(A∗ , A)-compact convex set of states
with the property that each extreme point f of S1−Q (A) is also an extreme point
of of the quasi-state space of A, hence is a pure state of A with f (Q) = 0.
Each f ∈ S1−Q (A) is the σ(A∗ , A) limit of convex combinations of extreme
points in S1−Q (A) (as it holds for every compact convex set). Thus, if there is no
pure state h ∈ A∗ with h(Q) = 0 and h(p) < 1, then Q = 1 − p, which contradicts
our assumption Q 6= q := 1 − p.
But if h ∈ A∗ is a pure state on A with h(q) = 0 and h(p) < 1, then we can
consider the irreducible representation R : A → L(H) corresponding to h. There is
a cyclic vector x ∈ H with kxk = 1 such that h(a) := hR(a)x, xi for all a ∈ A.
Let R : A∗∗ → L(H) also denote the normalization of R. Then R(Q)x = 0 and
kR(p)xk < 1. Thus y := γ(x − R(p)x) ∈ H with γ := kx − R(p)xk−1 satisfies
R(Q)y = 0, R(p)y = 0 and kyk = 1. The normal state g(a) := hR(a)y, yi on A∗∗
is pure and satisfies g(1 − (Q + p)) = 1.
In particular, the assumption 1 − Q − p 6= 0 gives a pure state on A with
g(Q) = 0 and g(p) = 0.
But by assumption, there exists a ∈ A+ with kak = 1 = g(a) and ap = 0. This
implies a ≤ Q and g(Q) 6= 0.
Thus, Q = 1 − p (i.e., 1 − p is open) if for each pure state g ∈ A∗ with
g(p) = 0 there exists an element a ∈ A+ with kak = g(a) = 1 and f (a) = 0 for all
f ∈ Sp (A). 

Is it also equivalent to each of the following ???

The norm-closed left-ideal

L := (A∗∗ (1 − p)) ∩ A = { a ∈ A ; ap = 0 }

of A is σ(A∗∗ , A∗ )-dense in A∗∗ (1 − p).

This is equivalent to:
A∗ p := {f ∈ A∗ ; f (ap) = f (a)} is σ(A∗ , A)-closed.
Clearly, L is a closed left-ideal of A and L ⊆ A∗∗ (1 − p).
Let qL ∈ A∗∗ denote the open support projection corresponding to L, i.e.,
qL := sup{d ∈ A+ ; d ≤ 1, d ∈ L} .
1162 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

Then the σ(A∗∗ , A∗ )-closure of L is A∗∗ qL and qL ≤ 1 − p.

Let L be σ(A∗∗ , A∗ )-dense in A∗∗ (1 − p). Then A∗∗ qL = A∗∗ (1 − p), hence
qL = 1 − p.
A∗ p = {f ∈ A∗ ; f (ap) = f (a), for all a ∈ A } = {f ∈ A∗ ; f (A(1 − p)) = 0} .
It says that A∗ p = {f ∈ A∗ ; f (L) = {0} } . The right side is L⊥ in A∗ and is
there σ(A∗ , A)-closed.
THE HARD stuff is the opposite direction:
Suppose p ∈ A∗∗ is a projection with the property that A∗ ·p is σ(A∗ , A)-closed.
(Then pA∗ p = (p · A∗ ) ∩ (A∗ · p) is σ(A∗ , A)-closed by σ(A∗ , A)-continuity of
f → f on A∗ given by the map f (a) := f (a∗ ) .)
Thus, there is closed subspace X of A such that ?????
(A/X)∗ ∼
= A∗ · p ???
?????
Let Lp := {a ∈ A ; ap = 0} i.e., Lp = A∩(A∗∗ (1−p)) = {a ∈ A ; a = a(1−p)}.
And Lp is a closed left-ideal of A. Let Q denote the open support projection
of A with A∗∗ Q = Lp with respect to σ(A∗∗ , A∗ )-topology. We get Q ≤ (1 − p).
Thus p ≤ 1 − Q, and pA∗ p is a subspace of (A/(L∗p + Lp ))∗ ∼
= (1 − Q)A∗∗ (1 − Q)
HERE starts the difficulty of this rather special approach:
Is Q = 1 − p ????
Are p and Q determined by the there restrictions to the “discrete” type-I part
of A∗∗ ?
There are natural bijections between the following fairly different types of ele-
ments in A∗∗ , quotient spaces of A, subspaces of A and certain subsets of A∗ :
(a) closed projections in A∗∗
(b) open projections in A∗∗
(c) quotient C *-spaces A//D := A/(L∗ + L) (with matrix order and matrix
norms induced from A (respectively A∗∗ ).
(d) closed left-ideals L ⊆ A
(e) closed right-ideals R ⊆ A
(f) hereditary C *-subalgebras D of A given by D := L∗ ∩ L respectively D :=
∗
R ∩ R.
(g ??) hereditary σ(A∗ , A)-closed convex sub-cones K of A∗+ . (The K should
be of form R+ · Cp (A∗∗ ) for some projection p ∈ A∗∗ .)
(h ??) the set Sp (A) of f ∈ A∗ with kf k ≤ 1 = f (p) is σ(A∗ , A)-closed.
(if 1 − p is open ???).
Uses Lemma:
 18. OPEN PROJECTIONS, C *-SPACES AND KADISON TRANSITIVITY 1163

Each hereditary norm-closed sub-cone K of the positive part (M∗ )+ of the

predual M∗ of a W*-algebra M is of the form Kq = (qM∗ q)+ ∼ = ((qM q)∗ )+ for
some projection q ∈ M .
0 6= p ∈ M := A∗∗ is a closed projection, if and only if, the unit ball of Kp is
σ(A∗ , A)-closed, if and only if, Sp (A) = Cp (M ) := {f ∈ M∗ ; kf k ≤ 1 = f (p)} is
σ(A∗ , A)-compact.
In particular, every projection p ∈ A∗∗ with pA∗∗ p of finite dimension (as a
vector-space over R) is a closed projection.
This gives, e.g. by [616, thm. 1.5.2, thm. 3.10.7, cor. 3.10.8, prop. 3.11.9], the
equivalences:
D ⊆ A hereditary C *-subalgebra,
D 7→ D+ , D+ is hereditary closed convex cone in A+ ,
M ⊆ A+ hereditary closed convex cone, build M 7→ L(M ) := {a ∈ A ; a∗ a ∈
M } then L(M ) is closed left-ideal of A.
L ⊆ A closed left-ideal, L 7→ DL := D(L) := L∗ ∩ L, DL := L∗ ∩ L is a
hereditary C *-subalgebra of A.
Directly to L from D without passing through D+ :
Let L(D) := D · A. To see that this a closed left-ideal simply apply Cohen factor-
ization to the left Banach D-module L := span(D · A) ⊆ A . It gives L = D · L ⊆
D · A ⊆ L.
The projection p ∈ A∗∗ is a closed projection for A, with p := 1 − q, if and
only if,
q ∈ A∗∗ open projection for A.
A projection q ∈ A∗∗ is an open projection,
if and only if,
q = l.u.b.{a ∈ A+ ; a ≤ q} in A∗∗ ,
if and only if,
the closed left ideal Lq := {a ∈ A ; aq = a} has σ(A∗∗ , A∗ )-closure = A∗∗ q.
A non-zero projection 0 6= p ∈ M := A∗∗ is closed,
if and only if,
Sp (A) = Cp (M ) := {f ∈ M∗ ; kf k ≤ 1 = f (p)} with M := A∗∗ is σ(A∗ , A)-
closed,
if and only if,
Cp (A∗∗ ) is σ(A∗ , A)-compact,
if and only if,
q := 1 − p is open in A∗∗ ,
if and only if,
1164 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

A∗ p := {f ∈ A∗ ; f ((·)p) = f } is closed in A∗ with respect to σ(A∗ , A)-

topology.
Missing:
D 7→ A//D := A/(L∗ + L),
X := A//D 7→ q := 1X ∗∗ ∈ A∗∗ closed projection, unit of the second conjugate
X ∼
∗∗
= qA∗∗ q of X. Given by q := 1 − pD .
Passage to the open support projection qD ∈ A∗∗ of D is given by qD := 1 − q.

Proposition A.18.6. Let A a C*-algebra and p ∈ A∗∗ a (self-adjoint) projec-

tion.
The following are equivalent:

(i) p is a closed projection,

(ii) 1 − p is an open projection,
(Definition of “open” ?
1 − p is the l.u.b. in A∗∗ of an upward directed family positive con-
tractions in A?)
(iii) 1 − p is the least upper bound of {a ∈ A+ ; a ≤ 1 − p}.
(iv) The convex subset Cp (A∗∗ ) = Sp (A) := {f ∈ A∗ ; kf k ≤ 1 = f (p)} of A∗
is σ(A∗ , A)-closed.
(v) Cp (A∗∗ ) (= Sp (A)) is σ(A∗ , A)-compact.
(vi) The closed left-ideal L := {a ∈ A ; ap = 0} of A has σ(A∗∗ , A∗ )-closure
equal to A∗∗ (1 − p).

If p ∈ A∗∗ is a non-zero projection, then the σ(A∗ , A)-compactness of Sp :=

{f ∈ A∗ ; kf k ≤ 1 = f (p)} implies that p ∈ A∗∗ is a closed projection, i.e., there is
a closed left ideal L ⊆ A such that A∗∗ (1 − p) is the σ(A∗∗ , A∗ )-closure of L.

Next is cited where !!!

Compare with following lemmata.
The proof of the below stated Proposition A.21.4 concerning excision of pure
states use a reduction method to the separable case that is established for arbitrary
operator spaces and Banach spaces in full generality by the following Lemma B.14.1
and then applied for our basic observation on operator space quotients of C *-
algebras in Remark A.18.9.
Since we use them here the first time, we place here some “easy” facts – i.e.,
those that does not discuss the residually equivariant behavior of quotients by sums
L + R of left and right ideals L and R. We need reduction to separable subspaces
with same perturbations conditions for quotients, because Ext-theory, KK-theory
and our version of a stable but un-suspended E-theory (continuous versions of the
approximate kind of E-theory used by Elliott and Rørdam) have to do with non-
separable corona spaces.
We start with a very general observation:
 18. OPEN PROJECTIONS, C *-SPACES AND KADISON TRANSITIVITY 1165

Lemma A.18.7. Let (R1 , ρ1 ) and (R2 , ρ2 ) metric spaces of bounded diameters
≤ 2, T : R1 → R2 a contractive map with T (R1 ) dense in R2 .
Suppose that (R1 , ρ1 ) is complete (i.e., each Cauchy sequence in R1 with respect
to the metric ρ1 has a limit in R1 ).
Need an increasing continuous function γ ∈ C([0, 1], R) with γ(0) = 0, t ≤ γ(t)
and with the property:
If a, b ∈ R1 then there exists for each ε > 0 an element c ∈ R1 such that
ρ2 (T (c), T (b)) < ε and ρ1 (a, c) < γ(ρ2 (T (a), T (b))) + 2ε ≤ 3γ(ρ2 (T (a), T (b))).
Suppose that there exists a function a continuous function λ on [0, 2] with
λ(0) = 0 and the property that for p1 , p2 ∈ R1 and ε > 0 there exists p3 ∈ R1
with ρ2 (T (p3 ), T (p2 )) < ε and ρ1 (p1 , p3 ) ≤ λ(ρ2 (p1 , p2 )) + ε ????
???? Then T is surjective ????.

Proof. Find an decreasing sequence r1 > r2 > · · · in (0, 1) such that

n
P
n λ(3 rn ) < ∞. In particular lim rn = 0.

Let q ∈ R2 . There exist p1 , p2 , . . . ∈ R1 with ρ2 (pn , q) < rn /2. Thus

ρ2 (pm , pn ) ≤ (rm + rn )/2.
Find Pm,n ∈ R1 with ???
??
Idea: ????
Take a point q ∈ R2 . Given any sequence qn ∈ T (R1 ) with limn ρ2 (qn , q) = 0
“quickly enough”, we can select a sub-sequence rk := qnk such that in the inverses
T −1 (B(rk , δk )) of small balls around the rk one finds inductively elements pk that
build a Cauchy sequence in R1 ... 

Let E a Banach space and S the open unit ball of E. Recall that the Hausdorff
distance between two bounded convex subsets K1 , K2 of the Banach space E is
given by

distH (K1 , K2 ) := inf { t ∈ [0, ∞) ; K1 ⊆ K2 + tS and K2 ⊆ K1 + tS } .

Definition A.18.8. Let E a real or complex Banach spaces X ⊆ E a (not

necessarily closed) linear subspace, S the open unit-ball of E and t ∈ (0, ∞).
Define for a, b ∈ S the Hausdorff distance between the intersection K(a, X) :=
S ∩ (a + X) and K(b, X) = S ∩ (b + X) of the affine spaces a + X and b + X with
the open unit-ball S of E by

ρ(a, b) := distH (K(a, X), K(b, X)) .

We define the function of restricted perturbation t 7→ f (t; X) of the quotient map

πX by
f (t; X) := sup{ρ(a, b) ; a, b ∈ S, kπX (a − b)k ≤ t }
1166 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

It turns out that t → f (t; L∗1 + L2 ) is Hölder continuous near 0 if L1 and L2

are closed left ideals of a C *-algebra E. The point is that

f (t; L∗1 + L2 ) ≤ f (t; p1 E ∗∗ + E ∗∗ p2 )

for the (open) support projections p` of the σ(E ∗∗ , E ∗ ) closures E ∗∗ p` of L` , an the

latter is easily to estimate by an at zero Hölder continuous function of t.
The function f (t; X) estimates in a symmetric way for given elements a, b ∈ E
with max(kak , kbk) < 1 the best value for the solution of the following “restricted
perturbation” problem:
Let a + X = πX (a) and b + X = πX (b). What is the infimum =: λ(a, b) ≤ ka − bk
of ka − ck over all c ∈ b + X with kck < 1?
Because then, for each ε > 0, we can find c = c(ε) ∈ E with kck < 1 and
πX (c) = πX (b) and

ka − bk + ε ≥ ε + λ(a, b) ≥ ka − ck ≥ kπX (a) − πX (b)k .

We have that λ(a, b) ≤ f (kπX (a) − πX (b)k; X) . ?????

Need place for A CLEAR STATEMENT on the estimate of

f (t; L∗1 + L2 ) ≤ f (t; p1 A∗∗ + A∗∗ p2 ) ≤ t + (2t)1/2

!! ?? !!

Remark A.18.9. Let A denote a C *-algebra, L ⊆ A a closed left ideal, R ⊆ A

a closed right ideal, D a hereditary C *-subalgebra of A. We use the notations
DL := L∗ ∩ L, DR := R∗ ∩ R, LD := span(A · D).
Let ψ(α) := α + (2α)1/2 for α ∈ [0, ∞) .
??
Then there are the following elementary observations, cf. [431]:

(i) The vector space sum R + L is a closed linear subspace of A.

The bi-duals and bi-polars in A∗∗ are given by L∗∗ ∼= (Lo )o = A∗∗ pL
where pL := pD ∈ A∗∗ is the open support projection of L respectively of
D := L∗ ∩ L. Similarly R∗∗ = (Ro )o for closed right ideals.
The bi-polar ∼
= (R + L)∗∗ in the W*-algebra A∗∗ is identical with the
(always σ(A∗∗ , A∗ )-closed) subspace

A∗∗ q + pA∗∗ = (1 − p)A∗∗ q + pA∗∗ q + pA∗∗ (1 − q)

where p, q ∈ A∗∗ are the “open” support projections of the hereditary

C *-algebras DL := L∗ ∩ L and DR := R∗ ∩ R of A.
The bi-dual operator space (A/(R + L))∗∗ is naturally isomorphic to
the ternary algebra (triple product algebra) (1 − q)A∗∗ (1 − p) ⊆ A∗∗ .
In particular, (A/(L∗ + L))∗∗ ∼
= (1 − q)A∗∗ (1 − q) is a W*-algebra.
The isomorphisms and natural embeddings are all completely isomet-
ric, and the quotient maps are completely contractive, because we can
 18. OPEN PROJECTIONS, C *-SPACES AND KADISON TRANSITIVITY 1167

here alway replace A, R, L, D and p, q by Mn (A), Mn (R), Mn (L), Mn (D)

and p ⊗ 1n , q ⊗ 1n for n ∈ N.
The below considered constructions remain compatible via the obvi-
ous canonical completely isometric isomorphisms if we tensor by Mn , as
e.g. N (D, A) ⊗ Mn ∼ = N (D ⊗ Mn , A ⊗ Mn ) for N (D, A) in Part (iv).
(ii) Let x, y ∈ A/(R + L) with kxk, kyk ≤ 1 and a ∈ A with kak ≤ 1 and
πR+L (a) = x, then there exist b ∈ A with kbk ≤ 1 that satisfies πR+L (b) =
y and
ka − bk ≤ ψ(k y − x k) .
In particular, we get the important fact that the quotient map πR+L
maps the closed unit-ball of A onto the closed unit-ball of A/(R + L).
(iii) Let D := L∗ ∩ L ⊆ A and A//D := A/(L∗ + L) the operator system
defined by D.
Then L := A · D = span(A · D) (because D ⊂ L and L is a right D-
module), and (A//D)∗∗ is an operator system that is naturally isomorphic
to the W*-algebra (1−qD )A∗∗ (1−qD ), where qD denotes the open support
projection of D in A∗∗ .
(iv) Let M(A//D) ⊂ (1 − pD )A∗∗ (1 − pD ) denote the C *-algebra of two-sided
multipliers
T · A//D , A//D · T ⊆ A//D
∗∗
of the space A//D ⊆ (1 − pD )A (1 − pD ).
The natural epimorphism πD from the two-sided normalizer algebra

N (D, A) := {a ∈ A ; aD ∪ Da ⊆ D}

of D in A into

A//D := A/(L∗ + L) ⊆ (1 − pD )A∗∗ (1 − pD )

maps N (D, A) onto (A//D) ∩ M(A//D), and has kernel

Ann(D, A) := {a ∈ A ; aD = {0} = Da } .

(v) An element a ∈ A is in N (D, A) if and only if πL∗ +L (a∗ a) =

πL∗ +L (a∗ )πL∗ +L (a) and πL∗ +L (aa∗ ) = πL∗ +L (a)πL∗ +L (a∗ ) , i.e., if
and only if a is in the multiplicative domain of πL∗ +L .
(vi) The first part of this conclusion is also a corollary of a
more elementary proposition: Corollary A.15.3!!
For every separable C *-subalgebra G ⊆ A there exists a separable
C *-subalgebra B ⊆ A such that G ⊆ B,

B ∩ (L∗ + L) = (B ∩ L)∗ + (B ∩ L) ,

and B/(B ∩ L) → A/L is completely isometric, and (in addition) that

b ∈ B and b(D ∩ B) ⊆ D ∩ B for D := DL implies bD ⊆ D and the
natural map B//(B ∩ D) → A//D is completely isometric and defines a
*-monomorphism

(B//(B ∩ D)) ∩ M(B//(B ∩ D)) → (A//D) ∩ M(A//D) .

1168 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

(vii) If A is unital, then naturally M(A//D) = ∼ N (A, D)/ Ann(A, D) and B

can be chosen such that in addition to the properties in (vi), B unital with
1B = 1A , N (B, B ∩ D) ⊆ N (A, D) and Ann(B, B ∩ D) = B ∩ Ann(A, D).

19. On non-unital C *-systems

needed:
operator systems, C *-systems (definitions), nuclear operator system,
c.p. and c.i. maps
needed statements:
indlim nuclear if maps are nuclear,
perhaps also:
indlim is C *-space, B//D is C *-space,
nuclear OS is C *-space
definitions and listing of some result appear also further below.
?? ????
Matrix-normed operator spaces (X, k · kn ) can be defined as closed linear sub-
spaces of L(H). An axiomatic characterization as Banach spaces X with a family of
matrix norms k·kn on Mn (X) = X ⊗Mn has been given by E. Effros and Z.-J. Ruan,
[245], that satisfy ka ⊕ bkm+n = max(kakm , kbkn ) and kαcβkn ≤ kαkkckm kβk for
a, c ∈ Mm (X), b ∈ Mn (X), α ∈ Mn,m , β ∈ Mm,n , cf. [245]. If X is completely
isometric embedded in L(H) then the Banach space second conjugate X ∗∗ of X is
naturally embedded in the C *-algebra L(H)∗∗ as a σ(L(H)∗∗ , L(H)∗ )-closed linear
subspace. One can see immediately that the norm-closed unit balls of the bi-dual
Banach spaces
(Mn (X))∗∗ ∼
= Mn (X ∗∗ ) ⊆ Mn (L(H)∗∗ ) = (Mn (L(H))∗∗
are just the σ((Mn (X))∗∗ , (Mn (X))∗ ) closures of the unit-balls of Mn (X). Notice
that this definition is completely independent from any chosen norm on Mn (X ∗ ) ∼=
Mn ⊗X ∗ that is equivalent on the subspaces X ∗ ∼= eij ⊗X ∗ to the usual dual Banach
space norm on X ∗ , because the σ((Mn (X))∗∗ , (Mn (X))∗ ) topology depends only
form the equivalence classes of norms on Mn (X)∗ = Mn (X ∗ ) = Mn ⊗ X ∗ and
nothing else. For example, one can here use the usual Banach dual norms k · k∗n
on Mn ⊗ X ∗ that satisfy kf ⊕ gk∗m+n = kf k∗m + kgk∗n instead of kf ⊕ gk∗m+n =
max(kf k∗m , kgk∗n ). This can be also abstractly described as the bi-dual matrix-
normed space that is given by applying construction of matrix normed operator
space X ∗ of X with matrix norms given the isomorphisms Mn (X ∗ ) ∼ = CB(X, Mn )
by E. Effros and Z.J. Ruan in [245, sec. 3.2]. The natural inclusion Mn (X) ,→
Mn (X ∗∗ ) is isometric by [245, prop. 3.2.1]. But the above mentioned canonical
isometry of Mn (X ∗∗ ) ∼= Mn (X)∗∗ is a – formally – stronger statement.

Lemma A.19.1. Let A a C*-algebra, ρ1 , ρ2 ∈ A∗ Hermitian with kρ1 k = kρ2 k =

1. Suppose that there exists b, c ∈ (A∗∗ )+ with
kb + ck = kb − ck = max(kbk, kck) = 1 = ρ1 (b − c)
 19. ON NON-UNITAL C *-SYSTEMS 1169

and ρ1 (βb + αc) = ρ2 (βb + αc) for all α, β ∈ R. Then for the polar decompositions
ρj = ρj+ − ρj− holds ρ1 (b) = ρ2 (b) = ρj+ (b) = kρj+ k, and ρ1 (c) = ρ2 (c) =
−ρj− (c) = −kρj− k.

Proof. Recall that the polar decompositions ρj = ρj+ − ρj− of ρ2 and ρ2

satisfy 1 = kρj+ + ρj− k = kρj+ k + kρj− k.
Let tj := ρj+ (b) + ρj− (c), sj := ρj+ (c) + ρj− (b)). Then 1 = ρj (b − c) = tj − sj
and 1 ≥ (ρj+ + ρj− )(b + c) = tj + sj . They imply sj = 0 and ρj (b) = ρj+ (b),
ρj (c) = −ρj− (c), for j = 1, 2.
On the other hand, 1 = kρj+ k + kρj− k ≥ tj = 1, kρj+ k ≥ ρj+ (b) and kρj− k ≥
ρj− (c), that gives ρj+ (b) = kρj+ k and ρj− (c) = kρj+ k. Now use that ρ1 = ρ2 on
Rb + Rc. 

Remark A.19.2. A general (= not necessarily order-unital) operator system

X is a closed linear subspace X of a C *-algebra A such that X is invariant under
passage to adjoints (i.e., a ∈ X ⇔ a∗ ∈ X), and such that for each n ∈ N, a = a∗ ∈
Mn (X) ⊆ Mn (A) and δ > 0 there exist b, c ∈ Mn (X)+ := Mn (A)+ ∩ Mn (X) with

kb + ck = max(kbk , kck) ≤ kak and ka − (b − c)k < δ .

If A is unital and 1 ∈ X, then this property is trivially satisfied with b := (kak1 +

a)/2, c := (kak1 − a)/2 and δ = 0. The condition implies that functionals
f : Mn (X) → C with f (T ) ≥ 0 for T ∈ Mn (X)+ satisfy kf k = sup{f (T ) ; T ∈
Mn (X)+ kT k = 1} , and (therefore) norm-preserving extensions of f to a her-
mitian functional g : Mn (A) → C are positive functionals on Mn (A). The con-
dition also implies that hermitian linear functionals f on Mn (X) have decompo-
sitions f = f+ − f− into positive functionals f+ and f− on Mn (X) that satisfy
kf k = kf+ + f− k = kf+ k + kf− k . The decompositions with this properties are
not necessarily unique, but the norms kf+ k and kf− k are independent from the
chosen decomposition with this property, cf. Lemma A.19.1 and use the *weak
compactness in A∗∗ ⊕ A∗∗ of the set of pairs (b, c) with b, c ∈ X+ ∗∗
⊆ A∗∗ and
kb + ck ≤ 1.
It gives immediately, that the *weak closure (∼ = Mn (X ∗∗ )) of Mn (X) in
∗∗ ∼ ∗∗
Mn (A) = Mn (A ) contains a unique element En with the property f (En ) = kf k
for each positive functional of f on Mn (X). In this way, each Mn (X ∗∗ )sa becomes
an order-unit space, and X ∗∗ a unital operator system (with norms defined by the
order-unit norms on Mn (X ∗∗ ).
It allows to show that every completely positive contraction V : X → L(H)
extends to a completely positive contraction V e : A → L(H). In this way we
get a “universal” c.p.c. map Vu : X → B for some suitable C *-algebra B. We
∗
choose B such that Vu (X) generates B and call Cmax (X) := B (together with
∗
Vu : X → Cmax (X)) the universal C*-algebra generated by the (not necessarily
∗
unital) operator system X. There is a natural C *-morphism h : Cmax (X) → A
such that h ◦ Vu (x) = x for all x ∈ X.
1170 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

The induced norms on Mn (X) and the closed cones of positive elements
Mn (X)+ provide X with a structure of matrix-norms and matrix orders ( 9 ).
Let, more generally, X ⊆ A and Y ⊆ B self-adjoint closed subspaces of C *-
algebras A and B, and let T : X → Y ⊆ B a linear map. We call T completely
positive (respectively completely contractive, completely isometric) if the
natural extensions T ⊗ idn of T to maps from Mn (X) ∼ = X ⊗ Mn to Mn (Y ) are
positive (respectively contractive, isometric) for every n = 1, 2, . . .. We write later
c.p. , c.c. and c.i. maps (respectively).
We say that two self-adjoint closed subspaces X ⊆ A ⊆ L(H) and Y ⊆ B ⊆
L(K) are c.i. and c.p. isomorphic , if and only if, there are c.p. and c.c. maps
V : L(H) → L(K) and W : L(K) → L(H) with W V |X = idX and V W |Y = idY .
The second conjugate space X ∗∗ of a selfadjoint closed subspace X ⊆ A is
naturally isomorphic to the σ(A∗∗ , A∗ )-closure of X ⊆ A∗∗ . It is easy to check
(because Mn (X)∗∗ ,→ Mn (A)∗∗ is isometric) that the matrix-norms induced by
Mn (A∗∗ ) ∼
= Mn (A)∗∗ on Mn (X ∗∗ ) are the same as the second conjugate norms (i.e.,
bi-polar norms) on Mn (X ∗∗ ) under the natural isomorphisms Mn (X)∗∗ ∼= Mn (X ∗∗ )
(of vector spaces).
We say that inclusion X ,→ A is admissible if the σ(Mn (X)∗∗ , Mn (X)∗ )-
closure of Mn (A)+ ∩ Mn (X) is the same as Mn (A)∗∗+ ∩ Mn (X)
∗∗
for each n ∈ N.
This is e.g. the case if every bounded positive functional on Mn (A)+ ∩ Mn (X)
extends to a positive functional on Mn (A)+ with same norm. In general, the
inclusion X ,→ A is not admissible. But an arguments with bi-polars shows that
X ,→ A is admissible if X is an operator system (not necessarily unital) and is
equipped with the matrix norms and orders induced on Mn (X) by Mn (A) . On the
other hand, there always exists a C *-algebra B and c.p. and c.c. maps V : X → B
and W : B → L(H) such that W V (x) = x for x ∈ X and V (X) ,→ B is admissible
( 10 )
If X ⊆ A is admissible in A then an argument using the bi-polar shows:
X ⊆ A∗∗ is an operator system, if and only if, X ⊆ A is an operator system.
∗∗

Suppose that X is a self-adjoint closed subspace of A and that X ∗∗ (with bi-

dual matrix norms and orders of the matrix norms coming from matrix orders of X)
is c.i. and c.p. isomorphic to a C *-algebra M . Then M is necessarily a W*-algebra.
Any (not necessarily unital) completely isometric and completely positive map
V from a unital C *-algebra C into an other C *-algebra B (in our case with B =
A∗∗ ) satisfies V ⊗ idn (c) ∈ Mn (B)+ if and only if c ∈ Mn (C)+ . (In fact, it is
enough to require that C is a matrix ordered space with an order unit such that
the matrix norms are given by the induced order norm: kck ≤ 1 for c∗ = c ∈ Mn (C)
if and only if −1 ⊗ 1n ≤ c ≤ 1 ⊗ 1n .)

9One can give axiomatic descriptions of, not necessarily order-unital, operator systems inde-

pendent from inclusions X ,→ A into C *-algebras – using ideas of Effros and Ruan [245].
10 Here we consider A as C *-subalgebra of L(H), and that V is a c.i. and c.p. isomorphism

from X onto V (X).

 19. ON NON-UNITAL C *-SYSTEMS 1171

It follows: If X ∗∗ is c.i. and c.p. isomorphic to a C*-algebra, then X is a – not

necessary order-unital – operator system, ( 11 ).
Then any c.i. and c.p. embedding V : X ,→ B into a C *-algebra B satisfies
that Mn (B)+ ∩ V ⊗ idn (Mn (X)) = V ⊗ idn (Mn (A)+ ∩ Mn (X)), and – therefore –
that V (X) ,→ B is admissible. In particular, the natural embedding X ,→ X ∗∗ is
admissible.
Since completely positive and completely isometric maps T from a C *-algebra
onto an other C *-algebra is always a C *-algebra isomorphism, we get that the C *-
algebra structure on X ∗∗ is uniquely determined if X ⊆ A is admissible and X ∗∗
is c.i. and c.p. isomorphic to a C *-algebra. We can define the multiplier algebra
M(X) of a C *-system X by
M(X) := {a ∈ X ∗∗ ; aX ∪ Xa ⊆ X} .
It is a C *-subalgebra of X ∗∗ . If I : X → Y is a c.i. and c.p. map from X onto
a self-adjoint closed subspace Y ⊆ B of a C *-algebra B then Y is an operator
system with Y ∗∗ c.i. and c.p. isomorphic to a C *-algebra, I ∗∗ : X ∗∗ → Y ∗∗ is an
isomorphism of C *-algebras and M(I) := I ∗∗ |M(X) is a C *-isomorphism from
M(X) onto M(Y ).

Examples where X ∗∗ (together with its matrix norms and orders) is isomorphic
to a C *-algebra are:
(1) Let X := B//D := B/(BD + DB) ⊆ A := (1 − pD )B ∗∗ (1 − pD ) where D is
a hereditary C *-subalgebra of a C *-algebra B and pD ∈ B ∗∗ is the “open” support
projection of D with D∗∗ = pD B ∗∗ pD (12). The matrix norms and matrix orders
on Mn (X) that are induced by the inclusion X ⊆ A are easily seen to be the same
as those induced by the quotient maps Mn (B) → Mn (B//D) ∼ = Mn (B)//Mn (D).
Therefore, X ∗∗ is naturally c.i. and c.p. isomorphic to the von Neumann algebra
(1 − pD )B ∗∗ (1 − pD ).
(2) More generally, Y ⊆ X and the quotients X//Y := X/Z are in a natural
manner operator systems (with second conjugates isomorphic to the C *-algebras
pX ∗∗ p respectively (1 − p)X ∗∗ (1 − p)) if

(i) X ∗∗ is c.i. and c.p. isomorphic to a C *-algebra,

(ii) the σ(X ∗∗ , X ∗ )-closure of Y in X ∗∗ is a hereditary C *-subalgebra pX ∗∗ p
of X ∗∗ for a projection p ∈ X ∗∗ , and
(iii) Z is the set of z ∈ X with ρ(z) = 0 for all positive functionals ρ on A ⊃ X
with ρ(Y ) = {0}, i.e., Z = X ∩ (pX ∗∗ + X ∗∗ p).

(Then there is a natural c.i. and c.p. embedding X//Y ⊆ A//D where D is the
hereditary C *-subalgebra of A generated by Y .)
11 We consider X ∗∗ together with the second conjugate matrix norms on M (X ∗∗ ) ∼
n =
Mn (X)∗∗ and the σ(Mn (X)∗∗ , Mn (X)∗ )-closure of Mn (A)+ ∩ Mn (X) as matrix-order structure.
12 The closed linear span R ⊆ B of DB := {db ; d ∈ D, b ∈ B} is a non-degenerate left

D-module. Thus DR = R ⊆ DB ⊆ R by the Cohen factorization theorem, because D has a

bounded approximate unit.
1172 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

(3) A special case of (2) with X//Y = X/Y , Y = Z is obtained if Y ⊆ X is an

M-ideal of X (i.e., the subspace Y ⊥ of X ∗ is `1 -complemented in X ∗ : X ∗ = Y ⊥ +L
with kf + lk = kf k + klk for f, l ∈ X ∗ with f (Y ) = {0} and l ∈ L), because then
there is a central projection p in the center of the W*-algebra X ∗∗ such that pX ∗∗
is the weak closure of Y in X ∗∗ .
(4) Inductive limits indlim(Tn : Bn → Bn+1 ) are C *-systems, if B1 , B2 , . . . are
C *-algebras and Tn : Bn → Bn+1 are c.p. contractions, cf. Lemma A.19.6(v).
(5) All nuclear operator systems (that are not necessarily unital), are C *-
systems (cf. Proposition A.19.4).

Definition A.19.3. Let A a C *-algebra and X ⊆ A a self-adjoint closed

subspace of A. Then X is a (not necessarily with order-unit equipped) C *-
system if X is a (not necessarily order-unital) operator system and there exists
a c.i. and c.p. map I from a C *-algebra M onto X ∗∗ , i.e., the second conjugate
operator system X ∗∗ is c.i. and c.p. isomorphic to a C *-algebra.
A similar definition for general C *-spaces would be completely wrong, because
there exist separable operator spaces that have injective C *-algebras as second
conjugate, but are not C *-spaces and are not nuclear (are even not exact).
The unique C *-algebra structure on X ∗∗ allows to define the multiplier al-
gebra M(X) ⊆ X ∗∗ by

M(X) := {a ∈ X ∗∗ ; aX ∪ Xa ⊆ X} .

A closed subspace X ⊆ A is nuclear if, for every x1 , . . . , xn ∈ X and δ > 0,

there are k ∈ N and completely positive contractions V : A → Mk , W : Mk → A
with W (Mk ) ⊆ X and kW V (xj ) − xj k < δ for j = 1, . . . , n.
It is easy to see that nuclear X must be an operator system (that is not neces-
sarily order-unital).
Below we see that X is moreover a C *-system, cf. Lemma ??. (???
lem:6.?nuc.cst.syst ???)
A C *-system X is nuclear if X is a nuclear subspace of the C *-algebra X ∗∗ .

Proposition A.19.4. Let X 6= 0 a separable (not necessarily unital) matrix-

ordered and matrix-normed space. TFAE:

(i) X is nuclear.
(ii) X is completely isometric and completely order isomorphic to the inductive
limit of completely positive contractions Vn : Mkn → Mkn+1 .
(iii) X ∗∗ is completely isometric and completely order isomorphic to an injec-
tive W*-algebra.
(iv) There exists hereditary C*-subalgebras D ⊆ E ⊆ M2∞ of the CAR-algebra
M2∞ = M2 ⊗ M2 ⊗ · · · and a completely isometric matrix-order isomor-
phism from X onto E//D := E/(E · D + D · E).
 19. ON NON-UNITAL C *-SYSTEMS 1173

Remarks A.19.5. Before we start with the proof of Proposition A.19.4 we

remind the reader that a (not necessarily unital) operator system X is a very special
kind of much more general operator spaces defined and considered by E. Effros and
Zh.-J. Ruan in [245].
The (as Banach space) separable operator systems are always completely iso-
metric isomorphic to the quotient space A/(R + L) of some separable C *-algebra
A by a sum R + L of a closed right ideal R of A and and closed left ideal L of A.
The algebraic sum is always closed, cf. Proposition A.15.2(iv)
An operator system has in addition to its matrix norms an involution map
x ∈ X → x∗ ∈ X and a matrix order structure with the property that the second
order operator space X ∗∗ contains an order unit 1 ∈ X ∗∗ with the property that
−1n ⊗ 12 ≤ x ⊗ p12 + x∗ ⊗ p21 ≤ 1n ⊗ 12 is equivalent to kxk ≤ 1 for x ∈ Mn (X ∗∗ ).
(Here the order unit 1 ∈ X ∗∗ is often not in X itself, and pjk ∈ M2 denote matrix
units and 12 := p11 + p22 .)
In this case holds: If X is an operator system and is in addition a C*-space (as
reminded below), then X ∗∗ is a W*-algebra.
A C *-ternary space (also called “C *-space”) X is by definition is an oper-
ator space that has the following two properties:

(c*-sp 1) The second conjugate operator-space X ∗∗ of X is completely isometric

isomorphic to a ternary algebra (triple product algebra) pM q for some
W*-algebra M and projections p, q ∈ M .
(c*-sp 2) The natural completely isometric embedding η : X ,→ X ∗∗ of X into X ∗∗
is weakly injective, i.e., there is a completely contractive projection P from
X 4∗ onto the σ(X 4∗ , X ∗∗∗ ) closure of the image of η ∗∗ (X) of the natural
C *-ternary morphism η from the W*-ternary algebra X ∗∗ into X 4∗ .

The general definition of relative weakly injective linear maps T : X ,→ A

into a C *-ternary algebra A is that T is completely isometric and that there is a
completely contractive σ(A∗∗ , A∗ )-continuous linear projection

P : A∗∗ → T ∗∗ (X ∗∗ ) ⊆ A∗∗

with the property that P ◦ T ∗∗ = T ∗∗ . The existence of this embedding shows that
the operator space X is a C *-space (C *-ternary system).
An operator space X is weakly injective , if and only if, a suitable completely
isometric linear embedding of X into L(H) for some Hilbert space H is relative
weakly injective.
All separable C *-ternary spaces are completely isometric isomorphic to quo-
tient operator spaces C ∗ (T )/(L + R) of the universal C *-algebra C ∗ (T ) generated
by a single contraction T , with sum L + R of a suitable closed left-ideal L and a
closed right-ideal R of C ∗ (T ) (this follows easily from the results of [472]).
1174 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

An operator system X is a (not necessarily unital) C *-system if X ∗∗ is com-

pletely isometric and order isomorphic to a W*-algebra and the natural inclusion
map η : X ,→ X ∗∗ is relative weakly injective in X ∗∗ .
It is known that all (non-unital) separable C *-systems X admit a canonical
unitization X b that is completely order isomorphic to C ∗ (F∞ )/(L∗ + L) for some
closed left ideal L of C ∗ (F∞ ), cf. [472].
It is not difficult to see that all inductive limits A1 → A2 → · · · defined by
completely positive unital maps Vn : An → An+1 are C *-systems, cf. [472].

Proof of Proposition A.19.4. (i)⇒(ii): The definition of nuclearity for

separable operator systems X is equivalent to the existence of
S
(α) linear subspaces X1 ⊆ X2 ⊆ · · · ⊆ X with dim(Xn ) < ∞ and n Xn
dense in X, and
(β) completely positive contractions Tn : Mkn → Xn+1 and Sn : Xn → Mkn
such that k(Tn ◦ Sn )|Xn k < 2−n .

It follows that X = indlimn (Tn ◦ Sn ) : Xn → Xn+1 . One can check this in X∞ :=

`∞ (X)/c0 (X) by comparing it in X∞ on the subsets Xn with the inductive limit
X = indlimn (ιn : Xn → Xn+1 ) for the natural inclusion maps ιn from Xn into
Xn+1 . The asymptotic of the approximately commuting diagrams allow to see that

X∼
= indlimn→∞ Wn := Tn ◦ Sn : Xn → Xn+1
by comparing the restrictions of the iterates

Wk,` := W` ◦ . . . ◦ Wk : Xk → X`

on Xn ⊆ Xk for fixed n ∈ N and large n < k < `. It shows that the natural
embedding in X modulo c0 (X) in X∞ coincides with this inductive limit, because
it has a natural embedding into X∞ given for x ∈ Xn by

Wn,∞ (x) = (0, . . . , 0, x, Wn (x), Wn+1 Wn (x), . . .) + c0 (X) .

The complete isometry from X onto the map

x ∈ X 7→ (S1 (x), S2 (x), . . .) + c0 (Mk1 , Mk2 , . . .)

defines an a completely isometric and completely positive map from X onto the
inductive limit defined by the sequence of c.p. contractions Vn := Sn+1 ◦Tn : Mkn →
Mkn+1 . The defining morphisms for the inductive limit of the Vn are given by

Vn,n+` := Vn+` ◦ . . . ◦ Vn+1 ◦ Vn : Mkn → Mkn+`

and for an ∈ Mkn by

Vn,∞ (an ) = (0, . . . , 0, an , Vn (an ), Vn+1 Vn (an ), . . .) + c0 (Mk1 , Mk2 , . . .) .

Clearly this happens also for the map ???????

To be filled in ??
(ii)⇒(iv):
 19. ON NON-UNITAL C *-SYSTEMS 1175

(iv)⇒(iii):
(iii)⇒(i): 

Lemma A.19.6. Suppose that A, B1 , B2 , . . . are C*-algebra, that Tn : Bn →

Bn+1 (n = 1, 2, . . .) are completely positive contractions and that X ⊆ A is a
closed subspace of A, that is invariant under passage to adjoints, i.e., X = X ∗ :=
{x∗ x ∈ X}.
Then

(i) Suppose that there is a c.p. contraction V : A → A∗∗ such that V (x) = x ∈
A ⊆ A∗∗ for x ∈ X, V (A) is contained in the σ(A∗∗ , A∗ )-closure of X
in A∗∗ and (V ⊗ idn )(Mn (A)+ ) is contained in the σ(Mn (A)∗∗ , Mn (A)∗ )-
closure of Mn (A)+ ∩ Mn (X)) in Mn (A∗∗ ) ∼ = Mn (A)∗∗ .
Then X is a C*-system.
(ii) If X is a C*-system and ψ : X → C is a c.i. and c.p. map into a C*-
algebra C such that ψ(X) generates C, then there exists a unique *-algebra
morphism ϕ from C into X ∗∗ with ϕ(ψ(x)) = x for x ∈ X.
(iii) If, for every x1 , . . . , xn ∈ X and δ > 0, there is a completely positive
contractions V : A → A, such that V (A) ⊆ X and kxj − V (xj )k < δ for
j = 1, . . . , n, then X is a (not necessarily unital) C*-system.
(iv) X ⊆ A is a C*-system if X is a nuclear subspace of A.
(v) indlim(Tn : B → B) ⊆ `∞ (B)/c0 (B) is a C*-system.
(vi) If the maps Tn : B → B are nuclear, then the C*-system
indlim(Tn : B → B) is nuclear.

Proof. (vi): We show: If the Bn are separable (not necessarily unital)

C *-algebras and the maps Tn : Bn → Bn+1 are nuclear c.p. contractions, then
indlimn (Tn : Bn → Bn+1 ) is a separable nuclear C *-space.
This reduces to the case, where Bn is of finite dimension by Lemma A.14.2 and
Lemma A.14.1.
In the non-separable case is indlimn (Tn : Bn → Bn+1 ) the union of nuclear
subspaces given by indlimn (Sn : An → An+1 ) for separable C *-subalgebras An ⊆
Bn with Tn (An ) ⊆ An+1 , such that Sn := Tn |An is still an nuclear map from An
into An+1 .
to be filled in ??
Let X := indlimn (Sn : An → An+1 ), where the An are separable and the Sn
are nuclear. Consider the natural embedding X ⊆ `∞ (A1 , A2 , . . .)/c0 (A1 , A2 , . . .) .
In the separable case we can find a increasing sequence Z1 ⊆ Z2 ⊆ · · · of finite-
dimensional subspaces of X, and subspaces Yk ⊆ Ak of finite dimension, such that
S
k Zk is dense in X, Sk (Yk ) ⊆ Yk+1 and Zk = Sk,∞ (Yk ) for all k ∈ N. Then we
find ?????????? and
c.p. contractions Vk : Mµ(k) → Ak+1 , Wk : Ak → Mµ(k) , such that Vk ◦
Wk : Ak → Ak+1 have the property that for Lk := Yk ∪ Vk−1 (Mµ(k−1) ) ⊆ Ak holds
1176 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

k(Vk ◦ Wk − Sk )|Lk k < 8−k . It follows that X = indlimk→∞ (Sk : Yk → Yk+1 )

and X = indlimk→∞ (Vk ◦ Wk : Lk → Lk+1 ) with matrix order and matrix norm
inherited from `∞ (A1 , A2 , . . .)/c0 (A1 , A2 , . . .).
This implies, by Lemma A.14.1, that the matrix-normed and matrix-
ordered space X is completely isometric and completely order isomorphic to
C := indlimk→∞ (Wk+1 ◦ Vk : Mµ(k) → Mµ(k+1) . The Tk := Wk+1 ◦ Vk are
c.p. contractions.
By Proposition A.19.4 it follows that C and therefore X is a nuclear (not
necessarily unital) C *-system.
????? 

Remark A.19.7. “Non-unital” modifications of the proofs in [438] show that:

Compare next Definitions with those in work of J.Pisier or Effros–Lance ???
(1) There is a C *-algebra C and a hereditary C *-subalgebra D ⊆ C such that
C//D ∼ = indlim(Tn : B → B) by a completely isometric and completely positive
isomorphism.
(Use Stinespring dilation and [438, lem.2.5], compare the arguments in [472].)
(2) Every separable C *-system X is isomorphic to C//D for some separable
C *-algebra C and hereditary C *-algebra D ⊆ C.
(Compare [472, prop.5] for the unital case.)
(3) Let X a C *-system, i.e., X (considered e.g. as X ⊆ L(H) for some H)
is a matrix-ordered operator space with second conjugate X ∗∗ isometrically order
isomorphic to a W*-algebra M and there is a normal conditional expectation PX
from M ∗∗ onto the σ(M ∗∗ , M ∗ )-closure of the image of the canonical map from X
into M ∗∗ given by the natural map X → X ∗∗∗∗ ∼ = M ∗∗ . This latter assumption is
equivalent to the requirement that X ⊆ X is relatively weakly injective in X ∗∗ ,
∗∗

which can be expressed equivalently that the natural map from X ⊗max C ∗ (F2 ) to
X ∗∗ ⊗max C ∗ (F2 ) is injective (in an isometric sense). Here ⊗max is taken on the
category of tannery algebras extended to operator spaces X with X ∗∗ isomorphic
to a W*-ternary algebra pM q.
(To understand the latter extra requirement check by examples that PX defines
a normal c.p. contraction from M into M ∗∗ that is usually very different from both
of the central normal embedding M ∼ = pM · M ∗∗ and of the non-normal natural
inclusion of M into M ∗∗ .)
Then: X is nuclear,
if and only if,
the second conjugate operator system X ∗∗ is an injective C *-algebra,
if and only if,
the inclusion map from X into the W*-algebra X ∗∗ is weakly nuclear.
 19. ON NON-UNITAL C *-SYSTEMS 1177

(Modify here [438, lem.2.8] for the implication

X ∗∗ injective ⇒ X nuclear .

The other directions follows from (X ∗∗ )op ⊗max X = (X ∗∗ )op ⊗min X (where the
tensor products are defined by the natural inclusion of X in the W*-algebra X ∗∗ )
???????? Find reference ????????? by an argument of E. Effros and Ch. Lance
[241, thm. ???] that uses here the standard representation of X ∗∗ and (X ∗∗ )op in
the sense of U. Haagerup).
If this is the case and X is separable then one can take C := M2∞ in (2).
(4) Similar arguments show the analogs for C *-operator spaces X ∼ = C/(L + R)
of the results (2) and (3) for separable C *-spaces X, (by definition operator spaces
X with with X ∗∗ completely isometric to pM q for some von-Neumann-algebra M
and projections p, q ∈ M .)

Lemma A.19.8. Suppose that D is a hereditary C*-subalgebra of a C*-algebra

B with corresponding “open” support projection p := pD ∈ B ∗∗ . Let

M(B//D) := {a ∈ (B//D)∗∗ : a(B//D) ∪ (B//D)a ⊆ B//D } .

(o) The natural isomorphisms

= B ∗∗ /(pB ∗∗ + B ∗∗ p) ∼
(B//D)∗∗ ∼ = (1 − p)B ∗∗ (1 − p)
are unital and completely isometric. They equip (B//D)∗∗ with the unique
von-Neumann algebra structure that is compatible with the bi-adjoint ma-
trix order unit structure on B//D. It holds B//D = (1 − p)B(1 − p) if we
naturally identify (1−p)B ∗∗ (1−p) ∼= (B//D)∗∗ and consider all operators
as elements of (1 − p)B ∗∗ (1 − p).
(i) M(B//D) is a C*-subalgebra of (B//D)∗∗ , and M(B//D) ∩ (B//D) is
a closed ideal of M(B//D) .
The ideal M(B//D) ∩ (B//D) is natural isomorphic to N (D)/D by
the canonical epimorphism B 3 b 7→ b + R + L ∈ B//D = B/(R + L),
where L and R denote the closures of BD respectively DB.
(ii) An element x ∈ (1 − p)B ∗∗ (1 − p) is in M(B//D) ∩ (B//D), if and only
if, x, x∗ x, xx∗ ∈ B//D ⊆ (1 − p)B ∗∗ (1 − p).
(iii) If we identify N (D)/D and M(B//D) ∩ (B//D) naturally via πR+L ,
then, for every closed ideal J of B,

πD (N (D) ∩ J) = πR+L (J) ∩ πD (N (D)) .

(iv) Suppose that that T : B//D → C is a completely positive and completely

isometric map from B//D into a C*-algebra C.
If a C*-subalgebra A of C is contained in T (B//D) then γ := T −1 |A
is a *-monomorphism from A into B//D ∩ M(B//D) ∼ = N (D)/D.
(v) Let Ann(D, J) := {b ∈ B ; bD + Db ⊆ J } .
If J is closed ideal of B, then J ⊆ Ann(D, J) and

πJ (Ann(D, J)) = Ann(πJ (D)) := Ann(πJ (D), {0}) .

1178 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

In particular, πJ (D) is essential in B/J, if and only if, Ann(D, J) = J.

In particular, N (D) ∩ Ann(D, J) = N (D) ∩ J for all J / B.
If πJ (D) is essential in B/J then

N (D) ∩ Ann(D, J) = N (D) ∩ J .

(vi) Suppose that Ann(D, J) = J for all J / B. Then there is a unique *-

monomorphism
λ : N (D) → M(D)
such that λ|D = idD and

λ(N (D) ∩ J) = λ(N (D)) ∩ M(D, D ∩ J)

for all J / B.
In particular,

πD ◦ λ(N (D)) ∩ πD (M(D, D ∩ J)) = πD ◦ λ(N (D) ∩ J)) .

(vii) If B is separable, then there exist a completely positive contraction

V : B//D := B/(L + R) → M(D)/D such that V | N (D) = πD ◦ λ and



V πL+R (J) ⊆ πD M(D, D ∩ J) for all J / B .

Proof. Ad (o) and (i): If B is unital, then M(B//D) ⊆ B//D and the
statements (o) and (i) are contained in [437, thm.1.4] (in the unital case).
If B is non-unital, we can pass to the unitization Be = B + C · 1 of B and
have Be ∼
∗∗ ∗∗ ∗∗
= B ⊕ C. Let e denote the unit of B in B e ∗∗ . Then p ≤ e, B//D =
(e − p)B(e − p) = (1 − p)B(1 − p) and

B//D ⊆ B//D
e = (1−p)B(1−p)
e ⊆ (e−p)B(e−p)⊕C·(1−e) = B//D ⊕C·(1−e) .

It follows B//D
e = (1 − p)B(1 − p) + C · (1 − p) = B//D + C(1 − p) and, therefore

B//D = (B//D)
e ∩ (e − p)B ∗∗ (e − p) ,

and M(B/D)
e ∩ (e − p)B ∗∗ (e − p) = M(B//D)
e ∩ B//D .
If x ∈ M(B//D) ∩ B//D, then x(B//D) + (B//D)x ⊆ B//D and x(1 − e) =
(1 − e)x = x ∈ B//D.
e Thus x ∈ M(B//D)
e ∩ B//D. If y ∈ M(B//D),
e then
y = x + z(1 − p) with x ∈ B//D and z ∈ C. Since (1 − p)c = c = c(1 − p) for
∗∗
c ∈ B//D,
e we get x(B//D)+(B//D)x ⊆ (B//D)∩(e−p)B
e (e−p). The right side
is equal to B//D, which shows that M(B//D)
e = (M(B//D) ∩ B//D) + C · (1 − p).
The normalizer algebra N (D) ⊆ B of D in B, is the intersection of B with the
normalizer algebra N (D) + C · 1 of D in B.
e

By [437, thm.1.4(iii), lem.4.9], the natural epimorphism from B

e onto the unital
C *-system B//D defines a unital *-epimorphism from N (D) + C · 1 ⊆ B
e e = B + C·
onto M(B/D)
e = M(B//D) + C · (1 − p) ⊆ B//D e = B//D + C · (1 − p) with
kernel D. It follows that the natural epimorphism from B onto B//D defines a
*-epimorphism from N (D) onto M(B//D) ∩ B//D with kernel = D.
e ∗∗ = B ∗∗ as in the proof of parts (o,i).
Ad(ii): Let e with eB
 20. OPEN PROJECTIONS AND KADISON TRANSITIVITY 1179

If x, x∗ x, xx∗ ∈ B//D ⊆ B//D,

e then x ∈ M(B//D)
e ∩ B//D = M(B//D) ∩
B//D by [437, lem.4.8(ii)] and proof of part(i).
Conversely, x, x∗ x, xx∗ ∈ B//D if x is in M(B//D)∩B//D ⊆ (1−p)B ∗∗ (1−p),
because M(B//D) ∩ B//D is a closed ideal of the C *-subalgebra M(B//D) of
(1 − p)B ∗∗ (1 − p) by part(i).
Ad(iii): Because πD and πL+R | N (D) are the same, πD (N (D)∩J) is contained
in πD (N (D)) ∩ πL+R (J).
If a ∈ J and πL+R (a) ∈ πD (N (D)), then there exists b ∈ N (D) with b − a ∈
R + L. If we apply the quotient map πJ from B onto B/J to b, then πJ (b) ∈
πJ (R) + πJ (L), and πJ (b) is in the normalizer algebra N (πJ (D)).
Since πJ (R) and πJ (L) are the closed right and left ideals generated by πJ (D),


we have that N πJ (D) ∩ (πJ (R) + πJ (L)) = πJ (D) . It follows πJ (b) ∈ πJ (D),
i.e., b ∈ D + J . If b = d + c with d ∈ D and c ∈ J, then c = b − d ∈ N (D) ∩ J and
πD (c) = πD (a).
Ad(iv): Since T is completely positive and completely isometric, it fol-
lows that γ : A → B//D is completely positive and completely isometric.
Thus, in (B//D)∗∗ ∼ = (1 − p)B ∗∗ (1 − p) holds γ(a)∗ γ(a) ≤ γ(a∗ a) and
a∗ a ≤ T ∗∗ (γ(a)∗ γ(a)) ≤ a∗ a. It gives γ(a∗ a) = γ(a)∗ γ(a) for a ∈ A, i.e.,
γ : A → γ(A) ⊆ (B//D)∗∗ is a C *-morphism. Hence γ(A) ⊆ M(B//D) by part
(ii).
Ad(v): Let J / B. We get J ⊆ Ann(D, J) from the definition Ann(D, J) :=
{b ∈ B ; bD + Db ⊆ J } .
An element b ∈ B satisfies πJ (bD) = πJ (b)πJ (D) = {0} if and only if bD ⊆ J.
Thus,
πJ (Ann(D, J)) = Ann(πJ (D)) := Ann(πJ (D), 0) .

It follows that N (D) ∩ Ann(D, J) = N (D) ∩ J if πJ (D) is essential in B/J.

If πD (N (D) ∩ Ann(D, J)) = πD (N (D) ∩ J), and πJ (bD) = πJ (b)πJ (D) = {0},
then b∗ b ∈ Ann(D, J)
More????? proof: Lemma A.19.8 ??
Ad(vi):
Ad(vii): 

20. Open projections and Kadison transitivity

Remark A.20.1. We use the following lemmata and the Lemma 2.2.3 in the
proof of Proposition 2.2.5 and in proofs in Chapter 3.
The proofs do not allow something “up to ε” instead they require “precise”
algebraic variants of the Kadison transitivity theorem (cf. [616, thm. 2.7.5]) for
1180 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

projections p∗ = p ∈ A∗∗ with pA∗∗ p of finite linear dimension (i.e., p is in the

C *-algebra-“socle” of A∗∗ considered as C *-algebra).
The variant [616, thm. 2.7.5] of the Kadison transitivity theorem implies that,
for each projection q ∈ pA∗∗ p, there exists a self-adjoint contraction a∗ = a ∈ A,
kak ≤ 1 with p − 2q = (1 − 2q)p = ap .
It shows that the natural C *-morphism A0 := p0 ∩ A 3 a 7→ ap ∈ pM p is
surjective.
The theorem [616, thm. 2.7.5] covers also that case of any projections p in
the “socle” socle(A∗∗ ) of A∗∗ , because there exists a normal non-degenerate *-
representation D : A∗∗ → L(H) such that D|pA∗∗ p is faithful and D(p) has finite
rank, i.e., Dim(D(p)H) < ∞, cf. Lemma 2.1.15(i).
Here the (french) notation “socle” (“pedestal”, “base”?) of a Banach *-algebra
B is defined (since ca. 1950) by

socle(B) := {b ∈ B ; Dim(b∗ Bb) < ∞ } .

See ?????

It gives that for each selfadjoint contraction b∗ = b ∈ pA∗∗ p = pAp there is a

self-adjoint contraction a∗ = a in A with pa = ap = b .
Notice that all this happens also in the cases of real C *-algebras, C *-ternary
algebras etc., e.g. in the case of real C *-algebras, the same result with skew-adjoint
elements b∗ = −b ∈ A :
Then c := e12 ⊗ b − e21 ⊗ b is self-adjoint in (p ⊕ p)M2 (A)(p ⊕ p) = M2 (pAp) .
A self-adjoint contraction a∗ = a ∈ M2 (A), a(p ⊗ p) = (p ⊗ p)a and a(p ⊗ p) = c
satisfies for its entries

[ajk ] = e11 ⊗ a11 + e12 ⊗ a12 + e21 ⊗ a21 + e22 ⊗ a22

that a∗12 = a21 , a∗11 = a11 , a∗22 = a22 , pajk = ajk p, pa11 = 0, pa22 = 0, pa12 = b,
pa∗12 = −b. Thus, a := (1/2)(a12 − a∗12 ) is skew-adjoint and pa = ap = b.
It follows that p commutes with a and pap2 = pap. Hence, a is in the multi-
plicative domain Mult(V ) ⊆ A of the completely positive map
e → pap ∈ pA∗∗ p .
V:a∈A

Since the finite-dimensional real C *-algebra pA∗∗ p is the linear span of its pro-
jections, the restriction of V to the C *-subalgebra Mult(V ) of A is a *-epimorphism
from Mult(V ) onto pA∗∗ p.
In particular, for each irreducible representation d : A → L(H), we get that for
x, y ∈ H with kxk = kyk = 1 there exists e ∈ A with d(e)x = y and kek = 1. Then
ψ(e∗ e) = 1 if ψ denotes the pure state ψ(a) := hd(a)x, xi .
Compare Remark A.18.4 for an alternative proof of the existence of a C *-sub-
algebra B ⊆ A such that B 3 b 7→ pbp ∈ pA∗∗ p is a *-epimorphism.
 21. ON EXCISION OF PURE STATES 1181

21. On excision of pure states

We need non-separable and “precise” versions of well-known excision results

e.g. in [6]. Here we consider only excision for pure states on C *-algebras A. The
idea is, to show that every separable C *-subalgebra is contained in an eventually
bigger separable C *-subalgebra with the property that the restriction of the pure
state to this bigger one is again a pure state. For use in proofs, recall here that all
hereditary C *-subalgebras D of separable C *-algebras are σ-unital.
HERE some comments (important ??, check it, erase unnecessary discussions):
Before the proofs of some Lemmata and the Proposition A.21.4 let us mention
some of the conclusions from them. One of it is the following important property:
For every C*-algebra A and any pure state ϕ on A, the separable C*-subalgebras
B of A with the property that ϕ|B is again an irreducible state on B build an upward
directed net of separable C*-subalgebras of A such that each separable C*-subalgebra
of A is contained in one member of this net.
In particular, A is the (algebraic, set-theoretic) union of this special separable
C *-subalgebras, and one can combine it with a countable number of other properties
of separable C *-subalgebras of A, as considered in other sections in the Appendices
A and B, e.g. B can be chosen that it is relative weakly injectivity in A, that all
ideals of B are intersections of ideals of A with B .... And we can see that A is the
set-theoretic union of the upward directed family of all separable C *-algebras that
have all this countably many properties together...
The sequence (en )n≥1 of elements in B ⊆ A with G ⊆ B in Proposition ??
“excises” in a very sharp sense the restriction ϕ|B of the pure state ϕ . If one
allows to do the same with some more general nets of positive contractions for
some state ϕ on A then this ϕ can be “excised”, if and only if, ϕ is in the σ(A∗ , A)-
closure of the pure states on A – compare [6, def. 2.1, prop. 2.2] for this more
general excision property, which is not equivalent to our demand of accuracy for
our definition of “strict excision”.
Here we equip the reader with necessary information on this topic to understand
our methods for a detailed study of sufficient conditions on simple C *-algebras
A that cause pure infiniteness of A. (In fact we provide a big list of equivalent
properties for simple C *-algebras in Sections 2 to 4 of Chapter 2 that each imply
or are equivalent to pure infiniteness.)
The here constructed special fixed element e is a contraction that “supports”
ϕ sharply if A is separable and G = A, i.e., then the extreme point ϕ of the quasi-
state space A, i.e., the set of all linear functional on A of norm one, is the unique
peak point for suitable e ∈ A+ and e ≤ 1 with A = eAe.
For given pure states ϕ on separable or non-separable A there are many positive
contractions e of norm kek = 1 with the property that ϕ is a peak point of e ∈ A+
if A has no sub-quotient of type I, e.g. et for t ∈ (0, ∞).
1182 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

We use another terminology that describes all with help of quotient C *-spaces
A/(L∗ + L) build from C *-algebras by using left-ideals L because those anyway
play a role in some of our proofs e.g. for the embedding theorem.
For non-separable A one gets usually only a convex subset of positive contrac-
tion k ∈ A+ with ϕ(k) = 1. Then ϕ(k 2 ) = 1. If ϕ 6= ψ for some other pure state ψ
then there exists a positive contraction k ∈ A+ with
min(ϕ(k), ψ(k)) < max(ϕ(k), ψ(k)) = 1
by application of Kadison transitivity to projections in p ∈ A∗∗ with pA∗∗ p isomor-
phic to M2 or C ⊕ C.
This follows e.g. from Remark A.18.9 by considering the C *-algebra
A/ (Lϕ ∩ Lψ )∗ + (Lϕ ∩ Lψ ) .

It is an, – at most 4-dimensional and at least 2-dimensional –, quotient C *-space of

A (which is automatically a C *-algebra because of its finite dimension as a vector
space).

Lemma A.21.1. Let E a C*-algebra and J / E a non-zero σ-unital closed ideal

of E such that E/J is one-dimensional.
Then E contains a strictly positive contraction e ∈ E+ such that πJ (e) = 1E/J
and that e − e2 is a strictly positive element of J.

Proof. Let ψ(t) := min(t, 1) for t ∈ [0, ∞). Since E/J is one-dimensional,
there is unique element d ∈ (E/J)+ with kdk = 1, i.e., d = 1E/J . Let g ∈ E a lift
of d, i.e., πJ (g) = d1/2 = d . Then c := ψ(g ∗ g) is a positive contraction c ∈ E+
with πJ (c) = d = 1E/J . By assumption, J+ contains a strictly positive element h.
We may suppose that khk = 1/2, – otherwise replace h by k2hk−1 h.
We have now c ∈ E+ with 0 ≤ c ≤ 1 in E + C · 1 ⊆ M(E) ⊆ E ∗∗ and
πJ (c) = 1E/J and a strictly positive element h ∈ J+ for J with norm khk = 1/2 .
Then 1/2 ≤ 1 − h ≤ 1 and 2 · 1 ≤ (1 − h)−1 in M(E) . (Here 1 denotes the unit of
M(E).)
This implies that c + h is strictly positive in E and c + h ≤ (1 − h)−1 ∈ M(E) .
We define a positive contraction by
e := (1 − h)1/2 (c + h)(1 − h)1/2 ∈ E+ .
The element e is strictly positive in E because (1 − h)1/2 is invertible in M(E) and
c + h is strictly positive in E.
Obviously (1 − h)1/2 h(1 − h)1/2 = h − h2 ∈ J+ and c1/2 hc1/2 ∈ J+ . Thus
πJ (e) = πJ (c) = 1E/J by the MvN-equivalence
(1 − h)1/2 c(1 − h)1/2 ∼M vN (c − c1/2 hc1/2 ) .
It implies e − e2 ∈ J+ because 0 ≤ e ≤ 1 .
If we use that 0 ≤ c ≤ 1 to get e ≤ (1 − h)1/2 (1 + h)(1 − h)1/2 = 1 − h2 , i.e.,
that h2 ≤ 1 − e. It implies that e1/2 h2 e1/2 ≤ e − e2 ∈ J+ .
 21. ON EXCISION OF PURE STATES 1183

The contraction h2 is strictly positive in J because h is strictly positive in J.

Now observe that if J is a closed ideal, h2 is strictly positive in J and e is strictly
positive in E, then e1/2 h2 e1/2 is strictly positive in J. 

Definition A.21.2. Let A C *-algebra and ρ a state on A. We say that ρ has

the excision property if there exists a net of positive contractions {eτ }τ ∈Σ ⊆ A+
that satisfy that ρ(eτ ) = 1 for all τ ∈ Σ and

lim k eτ aeτ − ρ(a)e2τ k = 0 .

τ ∈Σ

We say that a state ρ on A has the strict excision property if A contains a

positive contraction e ∈ A+ with ρ(e) = 1 such that

lim ken aen − ρ(a)e2n k = 0 for all a ∈ A .

n→∞

It is not difficult to see that a state ρ on A with the property that there exists
e ∈ A+ with kek ≤ 1, ρ(e) = 1 and limn ken aen − ρ(a)e2n k = 0 must be necessarily
a pure state on A:
The assumptions show that kek = 1, i.e., there is a pure state λ on A with λ(e) = 1.
Then λ(xek ) = λ(ek x) = λ(x) for all x ∈ A, and this implies limn (λ(a)−ρ(a)) = 0 ,
for all a ∈ A, i.e., ρ = λ is pure.

Lemma A.21.3. Let B separable C*-algebra, ρ a state on B.

The state ρ is a pure state on B, if and only if, there exists e ∈ B+ with
kek = 1, ρ(e) = 1 and

{b ∈ B; ρ(b) = 0} ⊆ (e − e2 )B + B(e − e2 ) . (21.1)

In particular, then ρ has the strict excision property in Definition A.21.2, i.e., there
there exists e ∈ B+ with ρ(en ) = 1 = kek for all n ∈ N, and

lim ken ben − ρ(b)e2n k = 0 for all b ∈ B .

n→∞

Proof. Suppose that ρ is a state on B and e ∈ B+ with ρ(e) = 1 = kek.

Then 1 ≤ ρ(e)2 ≤ ρ(e2 ) ≤ ρ(e) = 1, It follows that 0 ≤ e − e2 ≤ e and 0 ≤
ρ((e − e2 )2 ) ≤ ρ(e − e2 ) = 0 imply that ρ((e − e2 )b∗ b(e − e2 )) = 0 for all b ∈ B.
Thus, ρ(b∗ b) = 0 for all b ∈ L := B(e − e2 ). It says that the closed left ideal L
is contained in Lρ := {b ; b ∈ B, ρ(b∗ b) = 0}, i.e., L ⊆ Lρ . It is easy to see that
always L∗ρ + Lρ ⊆ ρ−1 (0). If we combine this containment with the containment
ρ−1 (0) ⊆ L∗ + L in (21.1), then we get that

ρ−1 (0) = L∗ρ + Lρ .

But this shows that the state ρ is a pure state, because [616, prop. 3.13.6.(i,ii)]
says that ρ is pure state on B, if and only if, L∗ρ + Lρ = ρ−1 (0), i.e., ρ is a pure
state, if and only if, L∗ρ + Lρ is equal to the null-space of ρ.
Now let ρ be a pure state on the separable C *-algebra B. We are going to
show that there exists a strictly positive contraction e ∈ B+ with ρ(e) = 1, that
1184 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

satisfies with L := B(e − e2 ) the equation

ρ−1 (0) = L∗ + L .

In particular, then e is a strictly positive contraction in B, L∗ + L + C · e = B, and

e − e2 is a strictly positive element of L∗ ∩ L.
A special case of the Kadison transitivity theorem shows then the existence of
c ∈ B+ with kck = ρ(c) = 1 , cf. [616, thm. 2.7.5, thm. 3.13.2(vi)]. – or use our
more engaged Lemma 2.1.15(ii,iii) for that.
The positive contraction c ∈ B+ satisfies then 1 = ρ(c)2 ≤ ρ(c2 ) ≤ ρ(c) = 1.
Thus, ρ(c2 ) = 1 and ρ(c − c2 ) = 0 .
Let h : B → L(H) the corresponding irreducible representation with cyclic
vector v ∈ H corresponding to ρ, i.e., kvk = 1 and hh(b)v, vi = ρ(b) for b ∈ B. Then
kck = 1 = ρ(c) = ρ(c2 ) and c ≥ 0 imply that h(c)v = v, because k(1 − h(c))vk2 =
hh(c)v, h(c)vi + hv, vi − hv, h(c)vi − hh(c)v, vi = ρ(c2 ) + 1 − 2ρ(c) = 0 , – here with
idH is denoted by 1.
It follows that ρ(b) = ρ(bc) = ρ(cbc) = ρ(cb) for all b ∈ B . In particular,
ρ(cb bc) = ρ((bc)∗ bc) = ρ(b∗ b) for all b ∈ B.
∗

Thus, the closed left ideal L := {b ∈ B ; ρ(b∗ b) = 0} of B has the property

L · c ⊆ L . Notice that B = L∗ + L + C · c, because L∗ + L is the kernel of the pure
state ρ and ρ(c) = 1 .
Since B is separable, the hereditary C *-subalgebra D := L∗ ∩ L of B contains
a strictly positive element g ∈ D+ for D. The positive part D+ of D consists of all
b ∈ B+ with ρ(b) = 0. In particular, c − c2 ∈ D+ . It follows that g + c is a strictly
positive element of B.
The relation L · c ⊆ L impies that D · c ⊆ D and c · D ⊆ D. It says that
the E := D + C · c is a C *-subalgebra of B, D is a closed ideal of E, ρ|E is a
character on E with kernel equal to D and ρ(ξ · c) = ξ for all ξ ∈ C, i.e., E/D is
one-dimensional. If we identify C naturally with πD (E) = E/D, then the quotient
map πD is just the character ρ|E.
By Lemma A.21.1, E contains a strictly positive contraction e ∈ E+ such that
πD (e) = 1E/D and that e − e2 is a strictly positive element of D = L∗ ∩ L that is
the kernel of ρ|E – if we adjust the identification of C with E/D such that 1 ∈ C
corresponds to 1E/D , i.e., ρ|E maps e and c in D + C · c onto the unit-element of
E/D ∼ = C · ρ(c) = C by the character ρ|E with ρ(e) = ρ(c) = 1.
Now we return to B = L∗ + L + C · c. We got also B = L∗ + L + C · e, because
ρ(e) = ρ(c) and L∗ + L is the kernel of ρ. Since e − e2 is a strictly positive element
of D = L∗ ∩ L (by above construction of e), it follows that L = B · D = B · (e − e2 )
and L∗ = (e − e2 ) · B .
Finally we estimate the norms ken ben − ρ(b)e2n k for large n ∈ N:
Each element b ∈ B = L∗ + L + C · e can be written as b = b1 + b2 + α · e with
b∗1 ∈ L, b2 ∈ L, α ∈ C. Then ρ(b) = α and the equation L = Be(1 − e) implies
 21. ON EXCISION OF PURE STATES 1185

that, for given δ ∈ (0, 1), there exist c1 , c2 ∈ B with kb1 − e(1 − e)c1 k < δ and
kb2 − c2 e(1 − e)k < δ. Notice that ken c1 (1 − e)en k ≤ kc1 kk(1 − e)en k and use
that limn→∞ ken (1 − e)k = 0 to get that limn ken ben − ρ(b)e2n k ≤ 3δ for arbitrary
δ ∈ (0, 1). 

The following Proposition is essentially [93, lem. 2.14]. But we give here to-
gether with Lemma A.21.3 a more detailed proof, because it plays a basic role for
proofs of some other results.

Proposition A.21.4. Let ϕ be a pure state on a C*-algebra A and G ⊆ A a

separable C*-subalgebra.
Then there exist a separable C*-subalgebra B ⊆ A and an element e ∈ B+ with
the properties G ⊆ B, kek = 1, ϕ(e) = 1 and

{b ∈ B; ϕ(b) = 0} ⊆ (e − e2 )B + B(e − e2 ) .

In particular, ϕ|B is a pure state on B and has the strict excision property in
Definition A.21.2 by Lemma A.21.3.
It shows that every pure state on a C*-algebra A has the excision property of
Definition A.21.2.

Proof of Proposition A.21.4. Let ϕ ∈ A∗+ a pure state, i.e., an extreme

point of the convex set of positive linear functionals on A of norm = 1. There
are many equivalent properties of a state ϕ that imply that ϕ is a pure state,
e.g. by [616, prop. 3.13.6(i,ii)]: The closed left ideal L := Lϕ of A defined by
L := Lϕ := {a ∈ A ; ϕ(a∗ a) = 0} satisfies ϕ−1 (0) = L∗ + L, if and only if, ϕ is a
pure state on A.
compare with text:
Then a special case of the Kadison transitivity theorem shows the existence of
c ∈ A+ with kck = ϕ(c) = 1 , e.g. use [616, thm. 2.7.5, thm. 3.13.2(vi)] or use our
farer going Lemma 2.1.15(ii,iii) to get such c ∈ A+ .
The purity of ϕ implies that

{a − ϕ(a)c ; a ∈ A} = L∗ + L = ker(ϕ) ,

cf. [616, thm. 3.13.2(iv), prop. 3.13.6] for the second equation. It yields that
P : A → A defined by P (a) := a − ϕ(a)c maps A into ker(ϕ) and satisfies P (a) = a
for all a ∈ ker(ϕ), kP k ≤ 2 and P 2 = P , i.e., is a linear projection of norm kP k ≤ 2
from A onto ker(ϕ).
The maps a ∈ A 7→ ϕ(a) · c ∈ A and idA are completely positive contractions.
Thus, the map P (a) := a − ϕ(a)c on A has a cb-norm kP kcb ≤ 2 , in particular
kP k ≤ 2 . Obviously, ϕ(P (a)) = 0 for all a ∈ A, – i.e., P (A) ⊆ ϕ−1 {0} –, and
ϕ(a) = 0 implies P (a) = a. Thus, ϕ−1 {0} ⊆ P (A). It shows that

P (A) = ϕ−1 {0} = L∗ + L and P 2 = P .

1186 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

In particular, P (a) = a if and only if a ∈ L∗ + L . The same arguments (– but with

b ∈ B in places of a –) show that P (B) = B ∩ (L∗ + L) for every linear subspace B
of A with c ∈ B ( 13 ).
The projection map P has following property: If B1 ⊆ B2 ⊆ · · · and B :=
S S
n Bn then P (B) = n P (Bn ) .

Now let G ⊆ A a separable C *-subalgebra of A, and let B1 the separable C *-

subalgebra of A generated by G ∪ {c} . Then (L∗ + L) ∩ B1 = P (B1 ), in particular
P (B1 ) ⊆ B1 .
By the – here very crucial – Corollary A.15.3, there exists a separable C *-
subalgebra B2 ⊆ A with the properties that B1 ⊆ B2 and

B1 ∩ (L∗ + L) ⊆ (L∗ ∩ B2 ) + (L ∩ B2 ) .

We can repeat this construction and argument by induction, and get an increasing
sequence of separable C *-subalgebras B1 ⊆ B2 ⊆ B3 ⊆ · · · of A with the property

Bn ∩ (L∗ + L) ⊆ (L∗ ∩ Bn+1 ) + (L ∩ Bn+1 ) ⊆ Bn+1 ∩ (L∗ + L) .

S
Let B denote the closure of n∈N Bn . Then, for all n ∈ N ,

Bn ∩ (L∗ + L) ⊆ (L∗ ∩ B) + (L ∩ B) ⊆ B ∩ (L∗ + L) .

(Use here that (L ∩ B) ⊆ B ∩ (L∗ + L) and L ∩ Bn+1 ≤ L ∩ B.)

The space n Bn ∩ (L∗ + L) is dense in B ∩ (L∗ + L):
S

P (Bn ) = Bn ∩ (L∗ + L) and P (B) = B ∩ (L∗ + L), because c ∈ Bn ⊆ B . Since P

S S
is a bounded projection and n Bn is dense in B we get that n P (Bn ) is dense in
P (B).
The sum (L∗ ∩ B) + (L ∩ B) is the sum of a closed left and closed right ideal
of B. Sums K ∗ + K := {x∗ + y ; x, y ∈ K ⊆ B} build by closed left ideals K ⊆ B
of B are always closed subspaces of B by Proposition A.15.2.
We get that P (B) ⊆ (L∗ ∩ B) + (L ∩ B) , because Bn ∩ (L∗ + L) = P (Bn ) ⊆
Bn+1 ∩ L∗ + Bn+1 ∩ L and kP k ≤ 2.
It follows that P (B) = B ∩ (L∗ + L) = B ∩ ϕ−1 {0} is equal to (L∗ ∩ B) +
(L ∩ B) and is the kernel of the restriction ψ := ϕ|B of ϕ to B. Notice that
b ∈ L ∩ B, if and only if, ψ(b∗ b) = 0 , i.e., L ∩ B = Lψ , where Lψ := {b ∈
B ; ψ(b∗ b) = 0}. The characterization of pure states on B, given by the above
cited [616, prop. 3.13.6(i,ii)], shows that ψ := ϕ|B is a pure state on B, because ψ
satisfies the purity criterium ψ −1 (0) = L∗ψ + Lψ .
Notice that G ⊆ B1 ⊆ B and that B is separable.

13 Notice here that it could be that ϕ|B is not a pure state on a C *-subalgebra B of A with

c ∈ B ... except in some special cases where ϕ|B is the only state ρ on B with ρ(c) = 1. For
example, take A := M2 , c := 12 and let B ⊆ A the diagonal elements in A. Then ϕ([αjk ]) =
2−1 jk αjk is a pure state on A but not on B.
P
 22. SOME TECHNICAL FUNCTIONS ON [0, 1] 1187

By Lemma A.21.3 there exists an element e ∈ B+ with the properties kek = 1,

ϕ(e) = 1 and
{b ∈ B; ϕ(b) = 0} ⊆ (e − e2 )B + B(e − e2 ) ,
and B has the strict excision property of Definition A.21.2.
Since we can do this selection procedure for every given separable subset G of
A with with respect to a given fixed pure state of A, we can see that each pure state
of A has the (in general not “strict”) excision property of Definition A.21.2. 

22. Some technical functions on [0, 1]

We use often for some very elementary approximation technics the increasing
piecewise linear functions hn ∈ C0 (0, 1]+ with break points at 2−n , n = 0, 1, 2, . . .,
given by
hn (t) := min(1, max(2n t − 1, 0)) for t ∈ [0, 1], n = 0, 1, . . . .
In the C *-algebra C0 (0, 1] and n ∈ N they are given by
h0 := 0 , h1 := (2f0 − 1)+ , . . . , hn := (2n f0 − 1)+ − (2n−1 f0 − 1)+ ,
where f0 ∈ C0 (0, 1] denotes the identity map f0 (t) := t on (0, 1].
The below listed properties of the hn can be easily seen on the intervals
[2 , 2−(n−1) ], because the functions hn are linear on two of this intervals and
−n

take outside them only the values 0 and 1:

(i) hn |[0, 2−n ] = 0, hn |[2−n , 2−(n−1) ] is the linear map

hn (t) := (2−(n−1) − 2−n )−1 (t − 2−n ) for t ∈ [2−n , 2−(n−1) ]
and hn |[2−(n−1) , 1] = 1. In particular, hn ∈ Cc (0, 1]+ ⊂ C0 (0, 1]+ for all
n ∈ N, and h0 := 0 on [0, 1] .
(ii) The set (2−n , 1] is the open support of hn in [0, 1], and is contained in
h−1
n+1 (1) = [2
−n
, 1] . In particular,
0 ≤ hn+1 hn = hn ≤ hn+1 ≤ 1 for all n ∈ N .
(iii) The non-negative function hn+1 − hn has support in the open interval
(2−(n+1) , 2−(n−1) ), i.e., is zero on [0, 2−(n+1) ] and [2−(n−1) , 1], is linear
increasing on [2−(n+1) , 2−n ] with value (hn+1 − hn )(2−n ) = 1 and is linear
decreasing on [2−n , 2−(n−1) ].
In particular, hn+2 (t) − hn+1 (t) = 1 − hn+1 (t) for t ∈ [2−(n+1) , 2−n ] .
(iv) The restriction to [2−(n+1) , 2n ] of the, – in C0 (0, 1] absolute convergent
and on all intervals [1/n, 1] uniformly convergent –, series
∞
X
2−n (hn+1 (t) − hn (t))
n=0
−(n+1) n
is on [2 , 2 ] just equal to the increasing linear function
2−(n+1) (1 + hn+1 )|[2−(n+1) , 2n ]
with values 2−(n+1) and 2−n at the end-points.
1188 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

P∞
Thus, the series n=0 2−n (hn+1 − hn ) converges uniformly on [0, 1],
and converges absolute in the Banach space C0 (0, 1] to f0 (t) := t, because
∞
X
2−n khn+1 − hn k ≤ 2 .
n=0

(v) The hn have the othogonality property hn (hn+2 − hn+1 ) = 0 .

(vi) For each γ ∈ (0, 1] only finitely many functions hn − hn−1 have support
P
in [γ, 1]. In particular, n≥1 (hn − hn−1 ) converges on each interval [γ, 1]
uniformly to 1.
(vii) The functions ψn := (hn − hn−1 )1/2 ∈ C0 (0, 1] (applied to t = f0 (t)) are
well-defined, have support in in [2−n , 2−n 4].
In particular, ψn ψm = 0 for |n − m| > 1.
ψ 2 converges in the multiplier algebra M C0 (0, 1] ∼
P

(viii) The sum =

n=1 n P
Cb (0, 1] strictly to 1, because n=1 (hn − hn−1 ) converges on each in-
terval [2−n , 1] uniformly to 1 = limn hn . (See Remark 5.1.1(2) concerning
strict convergence.)

An Application of the functions h1 , h2 , ... is the below given simple Example

A.22.1. It shows that the ≈-classes in W (A) and the corresponding ≈-classes in
Cu(A) are not the same for A := C0 (0, 1]. Notice that the Pedersen ideal of this
C *-algebra A is Cc (0, 1] and our “small Cuntz semigroup CS(A)” defined and used
in
where ????????
is also different from W (A) and Cu(A) . (But here Cu(A) has “good compari-
son” because of Dim((0, 1]) = 1 !)
Check calculations and notations in example:
Seems to be no real progress with comparison.
But it is likely to be true ... Perhaps pass to M2n (C0 (0, 1]) ... and combine with
the hn

Example A.22.1. The ≈-classes in W (A) and the corresponding ≈-classes in

Cu(A) are not the same for A := C0 (0, 1].

1/2
Consider A := C0 (0, 1] and define gn (t) := 2−n (hn+1 (t) − hn (t)) . Let
P∞
T := n=1 gn ⊗ p1,n .
Then T ∈ C0 (0, 1] ⊗ K and T T ∗ = ( gn2 ) ⊗ p11 = f0 ⊗ p11 for f0 (t) := t, but
P

the “almost diagonal” matrix

check formula for T ∗ T !!!

X X
T ∗T = gn gn+1 ⊗ pn,n+1 + gn gn−1 ⊗ pn+1,n + 2−n (hn+1 − hn ) ⊗ pnn
n n
S
is not contained in n Mn (A).
 23. K1 -INJECTIVITY AND GENERALIZED KUIPER THEOREM 1189

23. K1 -injectivity and generalized Kuiper theorem

Compare with related topics in Section 2 of Chapter 4

in particular with the squeezing property in Definition 4.2.14.
It is the basic observation used in the proof of Part(c) of
Proposition 4.2.15.
We reformulate some technical lemmata of J. Cuntz and N. Higson in [172]
and [180], that we use for example to derive generalizations of Kuiper’s Theorem
[505] in the spirit of J. Cuntz, N. Higson [180] and J. Mingo [557] by generalizing
their ideas to C *-algebras that have our “squeezing” Property (sq) of Definition
4.2.14.
We modify ideas in [180] to get a proof of the following easier applicable
Lemma.

Lemma A.23.1. If B is a σ-unital and stable, then, for every separable C*-sub-
algebra C of M(B), e ∈ C and ε > 0, there exist isometries S, T ∈ M(B) with
T ∗ CS ⊆ B, T ∗ S = 0 and kT ∗ eSk < ε.

Proof. The stability of B causes that for each given b ∈ B and ε > 0 there
exists an isometry r ∈ M(K) ⊆ M(B) such that kr∗ brk < ε (with r depending on
b and ε).
Thus, if we later are only interested in showing that M(B) has the “squeezing”
Property (sq) in Definition 4.2.14 for A := M(B), then it is enough to find, for a
given a ∈ M(B), isometries S, T ∈ M(B) with T ∗ aS ∈ B and S ∗ T = 0.
The claimed property S ∗ T = 0 is not important for this proof, because we
could replace a considered sequence b1 , b2 , . . . of contractions in M(B) anyway by
Q∗ b1 R, Q∗ b2 R, . . . with isometries Q, R ∈ M(B) with Q∗ R = 0 for this proof,
because the isometries QS and RT have the additional orthogonality property
(RT )∗ (QS) = 0.
We describe now a selection method that shows for a given separable C *-
subalgebra C of M(B) the existence of isometries S, T ∈ M(B) with orthogonal
ranges and the property that S ∗ CT ⊆ B .
Let c1 , c2 , . . . ∈ M(B) a sequence of contractions in M(B). Likewise this could
be a sequence that is dense in the unit ball of a given separable C *-subalgebra C
of M(B) and with c1 := c the particular element that should be squeezed below
ε > 0 with help of “squeezing isometries” S and T in M(B), i.e., with the additional
property kS ∗ cT k < ε .)
We describe an inductive selection method for the construction of isometries
S, T ∈ M(B) with S ∗ T = 0 and S ∗ cn T ∈ B for all n ∈ N , such that we can apply
Remark 5.1.1(2) concerning strict convergence on M(B) for σ-unital stable B :
The stability of B is equivalent to the existence of a sequence s1 , s2 , . . . ∈ M(B)
of isometries with the property that k sk s∗k converges strictly to 1 ∈ M(B), i.e.,
P
1190 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

with respect to the strict topology on M(B), cf. Remark 5.1.1(8) and Lemma 5.1.2.
In particular, the sequence k ∈ N → ks∗k bk + kbsk k converges to 0 for each b ∈ B.
Describe first the selection process.
Because now comes an other observation
about producing new isometries:
The sums n sλ(n) s∗λ(n) converge in M(B) strictly to a projection if λ : n ∈
P
P
N → λ(n) ∈ N is an injective map. Moreover, the sum n sλn fn converges strictly
to an isometry Sλ ∈ M(B) if f1 , f2 , . . . ∈ M(B) is any sequence with the property
that n fn∗ fn converges strictly to 1 in M(B) by Remark 5.1.1(2).
P

??
We use the following method to construct isometries in M(B):
Let e ∈ B+ a strictly positive contraction with kek = 1. Apply the functions
hn , ψn ∈ C0 (0, 1]+ as defined in Section 22 with ψn = (hn − hn−1 )1/2 to e ∈ A, i.e.,
fn := ψn (e) = (hn (e) − hn−1 (e))1/2 ∈ A+ . The n fn2 converges strictly in M(A)
P

to 1M(A) and n 2−n hn (e) is absolute convergent to e in A.

P

< Check above?

P∞
Thus S := n=1 sλ(n) fn converges strictly to an isometry in M(A) if
∗
P
n=1 sn sn converges strictly to 1M(A) by (two-fold application of) Remark
5.1.1(2).
If κ(n) ∈ N is another strictly increasing sequence, then for each b ∈ M(A) the
sequence

X
fn s∗κ(m) bsλ(n) fn
m,n

unconditional ??? Has to be checked !!!

is strictly convergent in M(A) with limit = T ∗ bS ∈ M(A) for the isometries
P∞
T := n=1 sκ(n) fn and above defined S.
Let b1 , b2 , . . . ∈ M(A)+ a sequence of positive contractions.
We find inductively a sequences of even numbers kn ∈ 2 · N with k1 := 2,
kn < kn+1 and odd numbers `n ∈ 1 + 2 · N with `n < `n+1 that satisfy
k(s`n )∗ bj skm fm k < 4−n for all j ≤ n and m < n and kfm (s`m )∗ bj skn k < 4−n
for all j ≤ n and m < n.
Beginning step:
Since bj s2 f1 ∈ A for j ∈ {1, 2} there exists an odd number `1 > 1 with
ks∗`1 bj s2 f1 k < 1/4 for j ∈ {1, 2}.
Then use that f1 s∗`1 bj are in A for j ≤ 3. It follows that there exists even
k2 ∈ N with k2 > 2 and kf1 (s`1 )∗ bj sk2 k < 4−n for all j ≤ 3.
Still to define/arrange the selection procedure!
 23. K1 -INJECTIVITY AND GENERALIZED KUIPER THEOREM 1191

Let m ∈ N an odd natural number, and given isometries skj , j = 1, . . . , m then

find an even index `m ∈ N such that `m > `m−1 and ks∗`m bn skj fj k < 4−n for all
n ≤ m + 2 and j < m .
Then go one step higher and select the next odd index km+1 in a similar way.
We select below suitable sequences fn ∈ B+ and injective maps κ : N → N with
odd values and λ : N → N with even values such that Sλ∗ bn Sκ ∈ B for all n ∈ N.
Let e ∈ B+ a strictly positive element with kek = 1 . The stability of B implies
that 0 is not isolated in the spectrum Spec(e) ⊆ [0, 1] . We take the piecewise linear
functions hn (ξ) := min(1, max(2n ξ − 1, 0)) (with break points at 0, 2−n , 21−n and
1), that is defined in Section 22 (and studied there in every detail), and consider

1/2
the non-negative functions ϕn (ξ) := hn (ξ) − hn−1 (ξ) for n = 1, 2, . . ., where
we let h0 := 0. Notice that ϕm · hn = ϕm for m ≤ n − 1.
Then the series n ϕn (ξ)2 ξ converges uniformly on [0, 1] to f0 , where f0 (ξ) :=
P

ξ, cf. Property (iv) in Section 22.

It follows that the contractions fn := ϕn (e) ∈ A+ and en := hn+1 (e) ∈ A+
have the property that fm en = fm for all m ≤ n and n fn2 e converges in norm
P

inside B to e ∈ B.
Refere to Property (iv???) in Section 22???
Hence, n fn2 converges strictly to 1M(B) in M(B).
P
P
The above observations imply that Sκ := n sκ(n) fn is an isometry in M(A)
for each strictly increasing map κ : N 3 n 7→ κ(n) ∈ N for the above defined
f n ∈ A+ .
If the ranges of two injective maps κ : N → N and λ : N → N are disjoint
then the isometrics Sλ and Sκ have orthogonal ranges. (And if the intersection
κ(N) ∩ λ(N) is a finite subset of N then Sκ∗ Sλ ∈ B .)
Start with k1 := 1, and define inductively strictly increasing maps κ : n 7→ kn ∈
N with odd kn and λ : n 7→ `n ∈ N with even `n such that

ken+2 s∗`m bj skn+1 k < ε · 2−(n+3) /(n + 1)

for all m ≤ n and j ≤ n + 2, and kn+1 > max(kn , `n ) and that in the next step

ks∗`n+1 bj skm en+2 k < ε · 2−(n+4) /(n + 2)

for all m ≤ n + 1 and j ≤ n + 2, and `n+1 > max(`n , kn ).

It is possible to find the desired odd number kn+1 and then the desired even
number `n+1 , because the sum n sn s∗n of the range projections converges strictly
P

to 1 and this implies, e.g. for the finitely many elements en+2 s∗`m bj ∈ B with
m ≤ n + 1 and j ≤ n + 2 that limν→∞ ken+2 s∗`m bj sν k = 0 .
Estimates to be checked again.
In particular the case j = 1. ??
For each j ∈ M the element Sλ∗ bj Sκ ∈ M(B) is contained in B :
1192 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

For fixed bj in the sequence (b1 , b2 , . . .) the summands X1,j := f1 s∗`1 bj sk1 f1
and partial sums
X `
X`,j := fn s∗`n bj skm fm
m,n=1

are in B. For each bj ` kX`+1,j −X`,j k < ∞, because kX`+1,j −X`,j k < ε·2−(n+1)
P

for n ≥ j. It implies convergence in B itself.

In the particular case b1 := c , we get kSλ∗ cSκ k < ε, because we have managed
that the norm of X1,1 and of all (X`+1,1 − X`,1 ) is small enough to get the desired
estimate.
To see that X := limn Sn ∈ B and S ∗ bj T ∈ M(B) are equal for it suffices to
see that cS ∗ bj T d = cXd for all c, d ∈ B . But this is evident even with c = d := e
and easy to see
It follows S ∗ CT ⊆ B and S, T are isometries with S ∗ T = 0 if the sequence
b1 , b2 , . . . is dense in the positive part of the unit ball of C ⊆ M(B) .


Remark A.23.2. Lemma A.23.1 implies that A := M(B) has the “squeezing”
Property (sq) of Definition 4.2.14. Therefore, all non-zero quotients of A ⊗max D
for unital C *-algebras D are K1 -bijective, by Proposition 4.2.15.
This covers the results of [180], e.g. that U(M(B)) = U0 (M(B)) and that all
pointed homotopy groups πn (U(M(B)), 1) are trivial ( 14 ).

M. Mingo [557] obtained the following generalization of Kuiper’s theorem in

the special case where B = C ⊗ K for unital C *-algebras C.

Theorem A.23.3 (J.Cuntz, N. Higson,[180]). Let B σ-unital stable C*-algebra,

and A a unital C*-algebra.
The K∗ -groups of M(B) ⊗ A and M(B) ⊗max A are trivial, and any quotient
algebra of M(B) ⊗max A is K1 -injective.
The unitary groups U of the algebras M(B) ⊗max A are connected (with respect
to the operator-norm topology), and have only trivial homotopy groups πn (U, 1) = 0.
Every continuous map from a locally finite CW-complex X into U is homotopic
in U to the constant map X 3 x 7→ 1 .

Proof. The C *-algebra M(B) has the “squeezing” property (sq) of Defi-
nition 4.2.14 by Lemma A.23.1. It causes the property (sq) for all C *-algebras
(M(B) ⊗max A) ⊗ C(X), all unital C *-algebras A and compact Polish spaces X
(e.g. let X := S n ). The property (sq) of M(B) implies K1 -injectivity of this alge-
bras.
14 The there deduced contractibility of the non-separable (!) Banach-space manifold
U (M(B)) requires some set-theoretic discussion that is not given in detail. We doubt that it
is possible without using a version of the Axiom of Choice (which is not provable), that should be
mentioned.
 24. TENSOR INTERSECTION LEMMA 1193

The infinite repeat D := δ∞ ⊗max idA satisfies D ⊕ id ∼

= D. It implies that
max
K∗ (M(B) ⊗ A) = 0 for all C *-algebras.
The K1 -triviality and K1 -injectivity for all X = S n imply together that every
continuous map from a locally finite CW-complex into U is homotopic in U to the
constant map 1. 

The following Corollary A.23.4 (see J.Cuntz, N. Higson,[180]) of Theorem

A.23.3 uses that a complete metrizable space that is locally uniformly homeomor-
phic to a (not necessarily separable !) Banach space is homotopic to a (locally
finite) CW-complex, cf. [556] for open subsets of separable Banach spaces and
[542, chp. IV, lem.5.2, cor.5.5] for the non-separable case ( 15 ). Then the vanish-
ing of the – by 1 – pointed homotopy groups yields that U(C) is contractible if
U(C(S n ) ⊗ C) = U0 (C(S n ) ⊗ C) for each n ∈ N. It implies:

Corollary A.23.4. If B is stable and σ-unital and A is unital, then the uni-
tary groups U(M(B) ⊗ A) and U(M(B) ⊗max A) are contractible.

Remark A.23.5. The following independent conclusion does not use arguments
from [542] (dependent from additional axioms on set theory?):
If A is unital and B is stable and σ-unital, then each C*-quotient C 6= {0} of
M(B) ⊗max A satisfies U(C)/U0 (C) ∼= K1 (C) .
Indeed: The quotients C have Property (sq) of Definition 4.2.14 and are K1 -
bijective by Proposition 4.2.15.

24. Tensor Intersection lemma

Next lemma to App.B? To Reduction to separable case?

In the following lemma X ⊗ν Z means the closure of X Z in the completion
A ⊗ν B of the algebraic tensor product A B with respect to a C *-norm k · kν on
A B. Notice that, e.g. in the case of the maximal C *-tensor product A ⊗max B
of C *-algebras A and B, one has for C *-subalgebras C ⊂ A not necessarily that
the natural C *-morphism C ⊗max B → A ⊗max B is injective. This C *-morphism
is injective for all C *-algebras B if and only if C is relatively weakly injective in
A (i.e., if there is a c.p. map V : A → C ∗∗ with V (c) = c for c ∈ C ⊆ A), because
– otherwise – the natural *-epimorphism C ⊗max B → C B ⊂ A ⊗max B is not
injective for B = C ∗ (F2 ).
The following Lemma is the Intersection Lemma [438, lem. 3.9].

Lemma A.24.1. Let A, B Banach spaces and N : A B → R+ a norm on

A ⊗ B with N (a ⊗ b) ≤ γ1 kak · kbk for some γ1 < ∞. Let A ⊗N B denote the
completion of A B w.r.t. N . Furthermore let X ⊂ A and Y ⊂ B closed linear

15 [542] does not say what kind of set theory axioms are precisely needed in case of open

subsets of non-separable Banach spaces.

1194 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

subspaces such that there is net of linear maps Tµ : B → Y and γ2 < ∞ with
N ((id ⊗Tµ )(z)) ≤ γ2 N (z) for z ∈ A B, and kTµ (y) − yk → 0 for all y ∈ Y . Then

X B∩A Y =X Y

in the completion of A B w.r.t. N .

In particular,
(C ⊗ν B) ∩ (A ⊗ν Y ) = C ⊗ν Y
for C*-algebras A, B, C*-norms k · kν on the algebraic tensor product A B,
any C*-subalgebra C ⊂ A and sums Y = L + R ⊆ B (respectively intersection
Y = L ∩ R) of closed right ideals R and closed left ideals R of B.

Proof. Let z ∈ X B ∩ A Y and ε > 0. Define δ := ε/(3 + 2γ2 ). We find

z1 ∈ X B and z2 ∈ A Y with N (zj − z) < δ for j = 1, 2. Let ak ∈ A and yk ∈ Y
P
with z2 = k ak ⊗ yk . By assumption, there is Tµ with N (z2 − (id ⊗Tµ )(z2 )) ≤
P
γ1 k kak kkTµ (yk ) − yk k < δ . Since (id ⊗Tµ )(z1 ) ∈ X ⊗ Y and

z − (id ⊗Tµ )(z1 ) = (z − z2 ) + (z2 − (id ⊗Tµ )(z2 )) + (id ⊗Tµ )(z2 − z1 ) ,

we get dist(z, X Y ) ≤ δ + δ + γ2 (2δ) < ε, – with respect to the semi-norm N on

A B. 

25. Characterization of closed ideals by inner automorphisms

Lemma A.25.1. Let A a (complex) C*-algebra and let X ⊆ A+ a hereditary,

closed convex cone in A+ , exp(−ih)X exp(ih) ⊆ X for all h = h∗ ∈ A with khk < π,
then X is the positive part J+ of a closed ideal J of A.

Proof. There should be also some reference to text books

e.g. Pedersen, Kadison, Dixmier, ??? ...
It is ”easy to see” that the hereditary C *-subalgebra J of A generated by the
linear span of X has the property that X is dense in J+ := A+ ∩ J. Thus, X = J+ .
The hereditary C *-subalgebra J is an ideal of A, because it is invariant under
conjugation with unitaries exp(ih) (and exp(−ih)) from any set S ⊆ A of elements
h = h∗ that has dense linear span in A, because then the open support projection
of the hereditary C *-subalgebra J is necessarily in the center of A∗∗ . (In case of a
”real” C *-algebra A one has here to take as S a suitable set of k ∈ A with k ∗ = −k
and to apply that exp(−k)X exp(k) ⊆ X and to rediscover X.)
Therefore it suffices for a proof to show that the linear subspace J := (X −
X) + i(X − X) is a hereditary C *-subalgebra and that J+ = X if X is a hereditary
sub-cone of A+ that satisfies exp(−ih)X exp(ih) ⊆ X for all h∗ = h ∈ A.
We show that a hereditary, closed convex cone X ⊆ A+ , has the property
A+ ∩ (X − X) = X and that this implies that X − X is closed in A and, therefore,
J := (X − X) + i(X − X) is a closed linear subspace of A with the property
J+ := A+ ∩ J = X.
 26. ON THE CORONA FACTORIZATION PROPERTY (CFP) 1195

Moreover we show that X = A+ ∩ LX for the subset LX ⊆ A defined by

LX := {a ∈ A ; a∗ a ∈ X } and that LX is a closed left-ideal of A. It implies that
J = L∗X ∩ LX , i.e., that J is a hereditary C *-subalgebra of A.
If x1 − x2 ≥ 0, x2 ∈ A+ and x1 ∈ X then x1 ≥ x1 − x2 ≥ 0. It follows
x1 − x2 ∈ X because X is hereditary.
Let LX := {a ∈ A ; a∗ a ∈ X}. Then LX is a closed left ideal of A and
X = A+ ∩ LX .
Indeed: Use that x2 ≤ kxkx and (a−b)∗ (a−b)+(a+b)∗ (a+b) = 2(a∗ a+b∗ b) and
that X + X ⊂ X, 2X = X, X is hereditary and closed. In particular x ∈ A+ ∩ LX
for all x ∈ X, because x2 ∈ X for all x ∈ X by x2 ≤ kxkx .
Conversely, if y ∈ A+ ∩ LX then y 2 ∈ X. But δ(y − δ)+ ≤ y 2 for all δ ∈ (0, 1).
It yields that (y − δ)+ ∈ X for each δ ∈ (0, 1). Since X is closed, it results that
X = A+ ∩ LX . In particular (X − X) + i(X − X) = L∗X ∩ LX is a hereditary
C *-subalgebra of A. 

26. On the Corona Factorization Property (CFP)

Proposition A.26.1. Let A denote a non-zero simple σ-unital C*-algebra of

real rank zero.
If A ⊗ K contains an infinite projection and A has the corona factorization
property (CFP), then A is purely infinite.

By definition, a C *-algebra A has the Corona Factorization Property

(CFP), if and only if, every full projection P in the multiplier algebra M(A ⊗ K)
of A ⊗ K is properly infinite, i.e., P ⊕S,T P - P in M(A ⊗ K) for isometries
S, T ∈ M(A ⊗ K) with S ∗ T = 0.
Here a projection P ∈ M(A⊗K) is full in M(A⊗K) if the ideal J of M(A⊗K)
generated by P is equal to M(A ⊗ K).
This is obviously equivalent to the property that a finite Cuntz sum Q :=
P ⊕ P ⊕ · · · ⊕ P ∈ M(A ⊗ K) and an isometry R ∈ M(A ⊗ K) exists with the
property RR∗ ≤ Q.
How to produce full projections? By sequences of properly infinite pro-
jections in A ⊗ K?
Others from old Part in Part B:
We discuss the corona factorization property (CFP) for some classes of σ-unital
C *-algebras of real rank zero.
Does the Elliott conjecture hold for simple, separable, unital, nuclear C *-
algebras A of real rank zero that have no non-trivial lower semi-continuous 2-quasi-
trace?
1196 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

The latter property (no non-trivial l.s.c. 2-quasi-traces) is equivalent to the

existence of a properly infinite projection in A ⊗ K, and we can pass to some
algebra p(A ⊗ Mn )p that contains a unital copy of O2 .
A more general open question is:
Has every separable nuclear simple C*-algebra A of real rank zero the following
“corona factorization property”?
(CFP): Every full projection in P ∈ M(A ⊗ K) is properly infinite.
(Then the projection P is the range P = T T ∗ of an isometry T ∈ M(A ⊗ K),
because M(A ⊗ K) has trivial K∗ -groups.)
(It is equivalent to:
If p ∈ M(A ⊗ K) is a projection with 1 - (p ⊕ p) then 1 - p.)
Every simple purely infinite σ-unital C *-algebra A has property (CFP) because,
– for σ-unital (!) A –, the stable corona Qs (A) := M(A⊗K)/(A⊗K) is simple (and
then automatically purely infinite), if and only if, A is simple and purely infinite,
cf. Corollary 2.2.11(i). It implies that M(A ⊗ K) is (strongly) purely infinite. Since
K∗ (M(A ⊗ K)) = {0} it follows that each full projection in M(A ⊗ K) is the range
of an isometry.
For simple σ-unital A ⊗ K and every non-zero projection p ∈ A ⊗ K, p(A ⊗
K)p ⊗ K ∼= A ⊗ K. Thus we may suppose that A is unital if A is simple and p.i.
An ideal system preserving isomorphism p(A ⊗ K)p ⊗ K ∼ = A ⊗ K exists if A
is σ-unital and A ⊗ K contains a projection p ∈ A ⊗ K that generates A ⊗ K as a
closed ideal.
It comes from P := δ∞ (p) ∈ M(A ⊗ K) as P (A ⊗ K)P ∼ = p(A ⊗ K)p ⊗ K and
the fact that there is an isometry s ∈ M(A ⊗ K) with ss∗ = P .
Let T ∈ M(A ⊗ K)+ a positive contraction with kπA⊗K (T )k = 1 . Define with
matrix units ejk ∈ K and identify c0 (C) with c0 := C ∗ (e11 , e22 , · · · ) ⊂ K and let
Pn
gn := k=1 1 ⊗ ekk ∈ 1 ⊗ c0 .
We find an increasing quasi-central approximate unit for C ∗ (T ) ⊆ M(A ⊗ K)
in the convex hull of the approximate unit consisting of the projections gn ∈ A ⊗ K.
This can be seen from our cf. Lemma B.23.1 or arguments in the proof
for [616, thm. 3.12.14] and combine them with the arguments for [616,
cor. 3.12.15,cor. 3.12.16].
To be shown:
Ptn
We find projections pn ∈ c0 ⊂ K with pn := k=r n
ekk where tn−1 + 1 < rn ≤ tn
are in N, k(1 ⊗ pn )T (1 ⊗ pn )k ≥ (n − 1)/n, pn pm = 0 for m 6= n, (1 ⊗ pn )gn = 0,
and k(1 ⊗ pk )T (1 ⊗ pn )k < 1/n2 for k < n .
Pn−1
Notice here that for T ∈ M(A ⊗ K)+ and q := k=1 pk ∈ K holds (1 ⊗ q)T =
lim` (1 ⊗ q)T 1 ∈ M(A ⊗ K) It has to be checked if there exists the pn !
P
Notice that n 1⊗pn converges strictly to a projection Q ∈ (1M(A) ⊗M(K)) ⊆
M(A ⊗ K) .
 26. ON THE CORONA FACTORIZATION PROPERTY (CFP) 1197

If such p1 , p2 , . . . exist, then there are contractions dn ∈ A ⊗ K

(likely such that d∗n dn ≤ 1 ⊗ enn and dn ∈ (1 ⊗ pn )(A ⊗ K)(1 ⊗ enn ))
with 1 ≥ d∗n (1 ⊗ pn )T (1 ⊗ pn )dn ≥ (1 − 2/n)(1 ⊗ enn ). Check Conjecture:
P
n (1 ⊗ pn )dn (1 ⊗ enn ) is strictly convergent to a contraction S ∈ M(A ⊗ K) with
1 − S ∗ T S ∈ A ⊗ K.
One can find/use a partial isometries U ∈ M(K) with enn U ∗ = en,rn , i.e.,
with enn U ∗ = en,n en,rn where ern ,rn ≤ pn for all n ∈ N. The sum Γ :=
P
n (1 ⊗
pn )dn (1 ⊗ en,`n ) converges strictly in M(A ⊗ K) and defines a contraction. Let
S := Γ ◦ U .
Then S is a contraction in M(A ⊗ K) such that 1 − S ∗ T S ∈ A ⊗ K and this
shows that Qs (A) is simple if A is σ-unital, simple and purely infinite. See above
attempt.
The strong pure infiniteness of M(A) is known if A is s.p.i. and σ-unital !
More general: (With K := K(`2 (N)).) If σ-unital A is s.p.i. then M(A ⊗ K) is
s.p.i. (It requires to show that M(A) is s.p.i. if A is σ-unital and A is s.p.i. )
It implies in particular that A has (CFP), because each (non-zero) projection
in M(A ⊗ K) \ (A ⊗ K) is properly infinite if the stable corona Qs (A) is purely
infinite, because K0 (Qs (A)) = 0.
The stable corona Qs (A) is simple if and only if A σ-unital, simple and p.i. if
and only if Qs (A) is simple and purely infinite, cf. Corollary 2.2.11.
Check:
Are some relations between [670, prop. 9.3] (on CFP) and our non-existence
of “infinitesimal” sequences??!!
(Is not very likely.)
Corona Factorization Property (CFP):
“Every full projection in the multiplier algebra M(A ⊗ K) of A ⊗ K is properly
infinite.”
(Who has really introduced this definition?)
Pn
If T1 , . . . , Tn ∈ M(A ⊗ K) exists with k=1 Tk∗ P Tk = 1, then P is properly
infinite, i.e., there exists S ∈ M(A ⊗ K) with 1 − S ∗ P S ∈ A ⊗ K.
???? Is Def. of what???
Is it equivalent to the following if A is σ-unital?:
(CFPI) := “Corona-full are properly infinite” defined by:
“Every full projection of Qs (A) is properly infinite.”
Implies (CFP), because if P is a full projection in M(A ⊗ K) then πA⊗K (P ) is
a full projection in Qs (A). By (CFPI) there exists a contraction R ∈ M(A ⊗ K)
with 1 − R∗ P R ∈ A ⊗ K. We can find an isometry X ∈ M(K) such that k1 −
(1 ⊗ X)∗ R∗ P R(1 ⊗ X)k < 1/4. Thus, there exists an element Y ∈ M(A ⊗ K) with
1198 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

Y ∗ P Y = 1. Then T := P Y is an isometry with T T ∗ ≤ P . It follows that P is

properly infinite. Now use that K0 (M(A ⊗ K)) = {0} and get that [P ] = [1], i.e.,
there is an isometry Z ∈ M(A ⊗ K) with ZZ ∗ = P .
(Similar arguments show that each full element of M(A ⊗ K)+ dominates the
range of an isometry if this is the case for every full element of Qs (A).)
Thus, we can show that (CFPI) implies (CFP).
Consider now the property (of ?????):
If 1 - P ⊕ P then 1 - P for projections P ∈ M(A ⊗ K).
This property implies (CFP), because a projection P is full in M(A ⊗ K) if
there exist n ∈ N such that [1] ≤ 2n [P ] in Cu(M(A ⊗ K)).
There is a projection Q ∈ Cu(M(A⊗K)) with [Q] = 2n−1 [P ]. Thus 1 - Q⊕Q.
reduces to [1] ≤ [Q] = 2n−1 [P ].
It follows also from (CFP), because 1 - P ⊕P implies that P is a full projection
in M(A ⊗ K).
More on the question if (CFP) implies (CFPI) (in case that A has real rank
zero):
Let T ∈ M(A ⊗ K)+ a positive contraction, with π(T ) = T + (A ⊗ K) a
projection in Qs (A) for π := πA⊗K .
When does there exist Yk := 1 ⊗ Xk that such that π(Yk ) is unitary and Y1∗ T Y2
is a partial isometry?
If π(T ) is a full projection in Qs (A) the there are elements S1 , . . . Sn ∈ M(A⊗K)
with 1 − k Sk∗ T Sk ∈ A ⊗ K
P

There exists an isometry I ∈ 1 ⊗ M(K) with k I ∗ Sk∗ T Sk I = 1M(A) ⊗ 1M(K) .

P

Seems that we need a stronger property, e.g. that every full positive contraction
T in M(A ⊗ K) with π(T ) a projection is properly infinite ...
Then (1 − T )T ∈ A ⊗ K.
Let G := f (T ) for f (t) := 1 for t ≥ 1/4 and f (t) := 0 for t ≤ 1/8 and
f (t) := 8t − 1 on [1/8, 1/4]. Then π(G) = π(T ), G(T − 1/4)+ = (T − 1/4)+ and
G ≤ 4T .
If π(T ) is full in Qs (A), then π(sT s∗ ) is full in Qs (A).
But it is not clear if sT s∗ can be “extended” to a projection in M(A ⊗ K) such
that ????
1/2
Let S := (T − 1/2)+ ∈ M(A ⊗ K), and a0 ∈ A+ , b0 ∈ K strictly positive
contractions. Notice GS = S = SG. And let D := S(A ⊗ K)S the hereditary
C *-subalgebra of A ⊗ K generated by the strictly positive element S(a0 ⊗ b0 )S of
D.
We find d1 , . . . , dn ∈ M(A ⊗ K) with j d∗j Sdj = 1. Thus, L := (A ⊗ K)S is
P

not contained in a non-trivial closed ideal of A.

 26. ON THE CORONA FACTORIZATION PROPERTY (CFP) 1199

Thus, D a “full” hereditary C *-subalgebra of A ⊗ K. Gd = d = dG for all

d ∈ D.
Next ‘‘green’’ to be shown:
We find a projection P ∈ M(A ⊗ K) and an element z ∈ A ⊗ K such that
zz = S(a0 ⊗ b0 )S and z ∗ z is a strictly positive element of P (A ⊗ K)P .
∗

Does there exists an element Y ∈ M(A⊗K) with Y Y ∗ ≥ P and Y ∗ Y G = Y ∗ Y ?

Is P full in M(A ⊗ K)?
The true question is:
Can we find for each full projection Q in Qs (A) a full projection P ∈ M(A ⊗ K)
such that π(P ) - Q.
OK. Above seems not to work. !!!
Other approach:
π(dk )∗ π(T )π(dk ) = 1 and
P
Let d1 , . . . , dn ∈ M(A ⊗ K) (w.l.o.g.) with k
∗
P
k k dk dk k ≤ 1.

Here “w.l.o.g.”, because we can replace the dk by ek := T 1/2 dk S, with some

suitable isometry S ∈ M(A ⊗ K).
Then k π(ek )∗ π( ek ) = 1, and we can lift the column
P

[π(e1 ), . . . , π(en )]> , which defines a partial isometry in Mn (Qs (A)), of norm = 1 to
a contraction in M(A ⊗ K).
Then there exists isometry S ∈ 1M(A) ⊗ M(K) with k1 − k S ∗ d∗k T dk Sk < 1/4
P

and 1 − k S ∗ d∗k T dk S ∈ A ⊗ K.
P

Let D := ( k S ∗ d∗k T dk S)−1/2 . Then 1 = k Rk∗ T Rk in M(A ⊗ K) for Rk :=

P P

dk SD. ...
Suppose (!) now that T ∈ M(A ⊗ K) itself is a full projection and that each
full projection in Qs (A) is properly infinite.
Then we can take n = 1 and get isometry R ∈ M(A ⊗ K) with R∗ T R = 1, i.e.,
T ≥ RR∗ , R∗ R = 1.
Thus, T is full and infinite, RO2 R∗ ∈ T M(A ⊗ K)T , RsT s∗ R∗ ≤ T and
RtT t∗ R∗ ≤ T . Thus [T ] + [T ] ≤ [T ] ≤ [1], [1] ≤ [T ].
Can we lift each projection in Qs (A) to a projection in M(A ⊗ K) ?
Is not possible in general, because the elements of the K0 -groups are represented
by projections and K0 (M(A ⊗ K)) = 0 .
Then we can consider the case ?????
If every closed (two-sided) ideal of A has the Corona Factorization Property,
then we say that A has the strong Corona Factorization Property.
Or is this the “strong” (CFP)? : See the green text further below!
It is known that A has property (CFP) if A0 ∩ Aω does not have a character,
cf. [467, thm. 4.2]. But the converse implication is wrong:
1200 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

The example [467, ???] is a simple purely infinite exact unital separable C *-
algebra A that has property (CFP) but where A0 ∩ A∞ has a character.
Where it is shown that A0 ∩ A∞ has a character for A in example ????
Thus the opposite direction is wrong.
Give citation for ‘‘no character’’ on A0 ∩ A∞ of example. Likely
in [467]
It follows from [467, thm. 4.2] that for every σ-unital C *-algebra A the algebra
A ⊗ Z has property (CFP), because of Z ⊗ Z ⊗ · · · ∼ = Z.
This is because one can show that each separable C *-subalgebra of M(K ⊗ A ⊗
Z) commutes modulo K ⊗ A ⊗ Z with a unital copy of Z.
Check now outlined proof
of property (CFP) for A ⊗ Z !
The preparations for [467, thm. 4.2] and the many needed technical facts from
other papers make it useful to give a more elementary proof of property (CFP) for
the algebras A ⊗ Z.
A less engaged proof can be based on the property Z ∼ = Z ⊗ Z ⊗ · · · and that
there is a unital C *-morphism η : Z → E(Z, Z) ⊆ C([0, 1], Z ⊗ Z) with π0 ◦ η(a) =
1 ⊗ a and π1 ◦ η(a) = a ⊗ 1.
Thus, in the stable corona each “full” projection P is also properly infinite,
because – by the above mentioned property – the elements t1 , . . . , tn ∈ M(K⊗A⊗Z)
with t∗1 P t1 + . . . + t∗n P tn = 1 (calculated modulo K ⊗ A ⊗ Z) and a unital copy of
O2 generate a separable C *-subalgebra of M(K ⊗ A ⊗ Z) that commutes modulo
K ⊗ A ⊗ Z with a unital copy of Z.
Let B := C ∗ (P, t1 , . . . , tn , s1 , s2 ; R) the free C *-algebra – with relations R
given by s∗1 s1 = s∗2 s2 = s1 s∗1 + s2 s∗2 = 1, P ∗ = P = P 2 and
P ∗
k tk P tk = 1,
ktk k ≤ 1 –. Then one can see that 1 ⊗ P is in the C *-algebra tensor product
Z ⊗ C ∗ (P, t1 , . . . , tn , s1 , s2 ; R) a properly infinite full projection:
(n + 1)[1] = [1] ≤ n[P ] ≤ n[1] = [1]
After Z-tensoring this becomes [1] = [P ] in the Cuntz semigroup W (Z ⊗
∗
C (P, t1 , . . . , tn , s1 , s2 ; R)) and carries then over to the stable corona.
Look to the explicit formula in the green below given calculation.
What about replacing Z by infinite tensor products of A(n, 2) with fixed n ≥ 3,
where A(n, 2) := C ∗ (a1 , . . . an , b1 , . . . bn , R) with relations (R) given by:
n
X
(R) := { a∗k ak = 1, b∗j aj = 0, b∗j bj = a∗j aj , j = 1, . . . , n} .
k=1

The A(n, p) are similar defined and are contained ???? in suitable tensor products
of A(n, 2):
Instead a1 , . . . an , b1 , . . . bn we consider ak` (k = 1, . . . , n, ` = 1, . . . , p) with
Pn ∗ ∗ ∗
k=1 ak,1 ak,1 = 1, ak,` ak,m = δ`,m ak,1 ak,1 .
 26. ON THE CORONA FACTORIZATION PROPERTY (CFP) 1201

Which elements of O∞ ⊗ A(n, p) are full in O∞ ⊗ A(n, p) ???

In particular, those of form 1 ⊗ X with X looking how ???
∗
P
`=1 t` P t` = 1 Gets only elements Yk with

Yk∗ (P ⊗ 1)Yk = 1 ⊗ a∗k,1 ak,1 . (Does not work so simple as it was hoped.)
Question:
Can we prove Z-absorption for separable simple unital nuclear A if A satisfies
(CFP)?
There exist a separable simple exact unital C *-algebra A that is p.i. (thus
Qs (A) := M(A ⊗ K)/(A ⊗ K) is simple and s.p.i., ... give citation for simplicity
and p.i. of Qs (A) !!!), but A0 ∩ Aω has a character:
The algebra A := C ∗ (F2 ) ⊗ R is an example with is property.
Give citation for the character on A0 ∩ Aω in case A := C ∗ (F2 ) ⊗ R !!!
Where is the algebra R defined ?
It follows (from which conditions ?) that each full projection in M(A ⊗ K) is
the range of an isometry in M(A ⊗ K). Thus, this algebra A has the (CFP) but
A0 ∩ A∞ has a character.
Has (at least) A0 ∩ A∞ a character if A is a simple unital separable and nuclear
C *-algebra that does not satisfy (CFP)?
It is known that separable unital A has (CFP) if A0 ∩ A∞ has no character,
see: [467, thm. 4.3].
But it is not known if stably infinite simple A with (CFP) is infinite ?????
Does (CFP) of simple separable A imply that A0 ∩ A∞ has no character?
Answer: No! Counterexample with exact A is in [467]. (But this counterex-
ample is not nuclear!)
What happens for exact simple separable unital p.i. A? Has A (CFP) ??
(Answer: Yes). Notice that simple p.i. A are s.p.i. !!!
Proposition ?? says that M(A ⊗ K) s.p.i. if A is a σ-unital s.p.i. C *-algebra.
Thus, s.p.i. σ-unital A have property (CFP).
Part (2) of Theorem [499, thm.3.1] says that property (CFP) for separable B
passes to non-zero quotients.
This shows that P is properly infinite modulo K ⊗ A ⊗ Z.
By Remark 5.10.3, this implies that P = T T ∗ for some isometry
T ∈ M(K ⊗ A ⊗ Z).
(Reference? What about converse in simple nuclear case?)
(Here “full” means equivalently that a finite Cuntz sum P ⊕ P ⊕ . . . ⊕ P of P
in M(A ⊗ K) majorize 1, which implies that this sum is properly infinite.)
Open question:
1202 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

Are all simple (separable, nuclear) stably infinite C *-algebras of real rank zero
automatically purely infinite?
Long standing question? Compare Chp. 2.!
By [581, cor. 5.16], a separable simple C *-algebra of real rank zero with the
Corona Factorization Property (CFP) is either stably finite or purely infinite.
This was also proved in an unpublished paper by S. Zhang, who didn’t explic-
itly introduce/mention the (CFP). (He simply considered something like it – or
something equivalent to it – for granted).
S. Zhang: (Reference ??? unpublished)
If A is a simple C *-algebra of real rank zero with Property (CFP), then A is
either stably finite or purely infinite.
(One can pass to the case of σ-unital simple C *-algebras, by passage to suitable
separable subalgebras to drop the separability assumption by passing to a suitable
separable C *-subalgebra.)
∗
The example A0 := Cred (F2 ) ⊗ R is simple, exact and p.i. (thus is s.p.i.).
Simple σ-unital purely infinite C *-algebras have the (CFP) (cite for proof?). But
F (A0 ) has a character. Thus (CFP) does not imply that F (A) has no character if
A is exact, simple and p.i.
A result for non-simple σ-unital C *-algebra A of real rank zero says:
[581, cor. 5.15]: Let A be a σ-unital C *-algebra of real rank zero, and let p be
a projection in A such that the m-fold direct sum p ⊕ p ⊕ . . . ⊕ p is properly infinite
(in Mm (A)) for some natural number m. Then p itself is properly infinite if one of
the following two conditions (i) or (ii) below hold:

(i) Every ideal in A has the (CFP)

(ii) A has the (CFP) and p is a full projection in A.

(The Part (i) should immediately follow from Part (ii).)

It comes from:
[581, thm. 5.13]: Let A be a σ-unital C *-algebra of real rank zero. Then V (A)
has the strong Corona Factorization Property (for monoids ), if and only if, every
ideal I in A has the Corona Factorization Property (for C *-algebras).
Is here A suppose to be stable? [581, lem. 5.6]: Let A be a σ-unital C *-algebra,
let P be a properly infinite, full projection in M(A) and let p ≤ P be a projection
in A. Then P − p is properly infinite and full in M(A).
[581, lem. 5.7]: Let A be a C *-algebra, let {pn } and {qn } be sequences of
P∞
pairwise orthogonal projections in A such that the sums P = n=1 pn and Q =
P∞
n=1 qn are strictly convergent in the multiplier algebra M(A), and hence define
projections P and Q in M(A).
 26. ON THE CORONA FACTORIZATION PROPERTY (CFP) 1203

(i) Suppose that there are sequences {kn } and {`n } of natural numbers such
that 1 ≤ k1 < `1 < k2 < `2 < k3 < · · · , and such that
[pn ] ≤ [qkn ] + [qkn +1 ] + . . . + [qln ]
for all n ∈ N. Then P - Q in M(A).
(ii) If P - Q in M(A), then for every natural number k there exists a natural
number ` such that
[p1 ] + [p2 ] + . . . + [pk ] ≤ [q1 ] + [q2 ] + . . . + [q` ]
in V (A).

Proof of (i): For each n, let sn ∈ A be a partial isometry with

s∗n sn = pn , sn s∗n ≤ qkn + qkn +1 + . . . + qln .
P P
As the sums pn and qn are strictly convergent, it follows that the sum
P∞
S := n=1 sn is strictly convergent in M(A). Hence P = S ∗ S ∼ SS ∗ ≤ Q.
From [467, thm. 4.3] (When central sequence algebras have a character.) we
get:
Theorem 4.3 :
Let A be a unital separable C *-algebra such that the central sequence algebra
F (A) has no characters. Then A has the strong Corona Factorization Property.
D. Kucerovsky and P.W. Ng show in [499, thm. 3.1] that the quotient of any
separable C *-algebra with the Corona Factorization Property again has the Corona
Factorization Property.
“It follows from this result that the quotient of any separable C *-algebra with
the strong Corona Factorization Property again has the strong Corona Factorization
Property.”
N
It was shown in [670, prop. 6.3] that A ⊗max ( max D) has the strong Corona
Factorization Property.
(For D with what kind of properties??? Likely D has to be unital and without
characters???)
The corollary [581, cor. 5.16]: says that a separable simple C *-algebra of real
rank zero with the Corona Factorization Property is either stably finite or purely
infinite.
(It shows that we could add the (CFP) to the assumptions of (Q1) then the
answer becomes positive. Moreover (Q1.implies.CFP for A), (Q1.implies.pi of A),
(Q1.implies.F (A) has no character) and (Q1.implies. ... are equivalent positive
answers.)
(It was also proved in an unpublished paper by S. Zhang, but without explicit
realizing / recognition that something like the (CFP) is needed.)
(One can drop the separability assumption by passing to suitable separable
C *-subalgebras.)
1204 A. CUNTZ EQUIVALENCE, MULTIPLICATIVE DOMAINS, ?RELATED TOPICS?

There exist exact simple separable unital C *-algebras, e.g. Cρ∗ (F2 )⊗R, that are
purely infinite but have not the (CFP). (Because F (Cρ∗ (F2 ) ⊗ R) has a character.)
If A is separable simple and σ-unital and F (A) has not a character then A has
(CFP).
( Recall: (CFP) = ”Corona factorization property” )
END of (CFP) Discussion !

27. On dense algebraic Ideals

The following definition is perhaps not identical with the definition given by
Pedersen in his book [616] on C*-algebras A ... He requires (!) that his dense ideal
is ”hereditary”.
We consider later the set A0,+ of all ε-cut downs of positive elements in A+
and show that this set is contained in any dense algebraic ideal J of A. We do not
know if the Pedersen ideal is identical with the algebraic ideal A0 of A generated
by all cut-down elements (a − ε)+ for a ∈ A+ .
The ideal A0 is dense in A and is contained in any dense algebraic ideal of A.
Since the Pedersen ideal of A contains all the cut-down elements (a − ε)+ (for
a ∈ A+ and ε > 0), it is likely that they are the same: Find a proof ...
Consider first the case A := C0 (0, ∞) ... ???
But ?????
Are semi-finite (2-)quasi-traces on the Pedersen ideal l.s.c.?
Thm. 1.3.3.: (in Pedersen Book) The set A+ is a closed real cone in Asa , and
x ∈ A+ , if and only if, x = y ∗ y for some y ∈ A.
Thm. 1.3.5.: (in Pedersen Book) If 0 ≤ x ≤ y then a∗ xa ≤ a∗ ya for all a ∈ A
and kxk ≤ kyk.
Prop. 1.4.5.: (in Pedersen Book) Let x ∈ A and x∗ x ≤ a ∈ A+ . If 0 < α < 1/2,
then there exists u ∈ A with kuk ≤ ka(1/2)−α k such that x = uaα .
(The element u ∈ A is not necessarily unitary.)
Sec. 5.6.: The minimal dense ideal. (But Pedersen considered only ”hereditary”
algebraic ideals. The true minimal dense ideal is the ideal ideal
Thm. 5.6.1.: (in Pedersen Book, p.175) !!!!!!! For each C*-algebra A there is a
dense hereditary ideal K(A), which is minimal among all dense ideals. (of A).
My question: Is it identical with the algebraic ideal of A that is generated by
the set of all ε-cut-downs (a − ε)+ ? (Here a ∈ A+ and ε ∈ (0, kak).)
Proof of Pedersen Thm. 5.6.1: Let K((0, ∞)) := set of all continuous function
on (0, ∞) with compact support (inside (0, ∞)). Define

K(A)0 := {f (x) : x ∈ A+ , f ∈ K((0, ∞))+ }.

 28. DIMENSION-FUNCTIONS IN SENSE OF J.CUNTZ 1205

Let
n
X
K(A)+ := {x ∈ A+ : x ≤ xk , xk ∈ K(A)0 }
k=1
so that K(A)+ is the smallest hereditary cone in A+ containing K0 (A).
If K(A) denotes the linear span of K(A)+ we conclude as in (Pedersen Book)
Proposition(5.1.3) that K(A) is a hereditary (!!!) *-algebra with (K(A))+ =
K(A)+ .
Since u∗ f (x)u = f (u∗ xu) for any unitary u ∈ A
e we have u∗ K(A)0 u = K(A)0 ,
and (as in subsection 5.2.1 of Pedersen book) this implies that K(A) is an ideal:
If u is a unitary in A
e = A + C1 then
3
X
4u∗ x = 4(x1/2 u)∗ x1/2 = ik (1 + ik u)∗ x(1 + ik u)
k=0

(the polarization identity), ... ?

Let (fn ) be a sequence in K(A) ...
Now use that each element of A
e is a combination of unitaries ...

28. Dimension-Functions in Sense of J.Cuntz

The following definition of Dimension functions (is quoted to J.Cuntz).

Definition A.28.1. Let D : XA →??? be a (non-negative) semi-finite Dimen-

S
sion function on XA := n Mn (Ped(A)) ”in sense of J.Cuntz”.
Claim: D(a) = D(a∗ a), D(a) ≤ D(b) if a - b.

D∗ (a) := supε>0 D((a − ε)+ )

Then

α 0 ≤ D(a) ≤ D∗ (a) for all a ∈ X,

β D∗ is again a Dimension function,
γ D − D∗ is a Dimension function,
δ If D is a Dimension function, then then the kernel Ker(D) satisfies
[ [
Ker(D) = ( Mn (Ped(A))) ∩ ( Mn (J))
n n

for some (closed ?) ideal J of A.

ε ? (D − D∗ )∗ = D∗ − D∗ = 0 ?

Has it following property?: D(a) 6= 0 implies D(b) 6= 0 for some b.

Are semi-finite (2-) quasi traces on the “Pedersen ideal” l.s.c.?
Are semi-finite Dimension functions (in the sense of J.Cuntz ) “lower semi-
continuous”?
 APPENDIX B

Exact C*-algebras and examples

We discuss in this Appendix B some facts related e.g. to the following questions:
(0) Passage to suitable separable subalgebras that preserve important proper-
ties of the algebra.
(1) Monotone maps between partially ordered sets.
(2) Some properties of exact C *-algebras.
(3) Example of a nuclear and approximately inner c.p. map that is not residually
nuclear, and its relation to the possible existence of group vN-algebras that are not
weakly exact. (Is there an example of Ozawa of vN-alg. that is not “weakly exact”?)
(4) Example of a (simple?) unital nuclear separable C *-algebra A and a unital
*-monomorphisms ι : A ,→ O2 of A into the Cuntz algebra O2 such that there is no
conditional expectation from O2 onto ι(A) .
It shows that there exist unital *-monomorphisms ι1 and ι2 of a (simple?)
unital nuclear separable C *-algebra A with the property that if we tensor the ιk
with idO2 , then the unital *-monomorphisms idO2 ⊗ιk : O2 ⊗ A → O2 ⊗ O2 ∼ = O2 of
the nuclear C *-algebras O2 ⊗ A are not unitarily equivalent, – even not equivalent
if we change it by an automorphism of O2 ∼ = O2 ⊗ O2 and an automorphism of
O2 ⊗ A, because the existence/non-existence of conditional expectations onto the
image remains unchanged if we tensor with O2 and apply any of this operations.)
(5) Some open questions concerning p.i. algebras.
(6) Examples of D where O(Prim(D)) is not the projective limit of lattices
O(Xn ) of Hausdorff l.c. spaces Xn with lower s.c. and monotone upper s.c. maps
from O(Xn+1 ) into O(Xn ).
(7) A family of pi-sun C *-algebras that exhausts the UCT-classes up to KK-
equivalence.
(8) This is more a conjecture than a given proof:
If the “exponential length” of all unitaries in U0 (A) in a unital A is bounded
by some constant γ ∈ (0, ∞) then U0 (A) is uniformly contractible, i.e., there exists
an (uniformly) continuous map

ψ : U0 (A) × [0, 1] → U0 (A)

such that ψ((u, 0)) = 1 and ψ((u, 1)) = u

1207
1208 B. EXACT C*-ALGEBRAS AND EXAMPLES

(The proof uses [553, thm. 3.2”], because

φ(u) := {(h1 , . . . , hn ) ∈ An ; h∗k = −hk , u = exp(h1 ) · . . . · exp(hn ) }

is an l.s.c. carrier φ : U0 (A) 7→ F(An ). if φ(u) is non-empty on each u ∈ U0 (A).

The l.s.c.-property of φ has to be checked again !!!
Open and at the same time closed subsets of complete metric spaces are com-
plete metric spaces, hence are para-compact.
We use this to get an alternative proof of the uniform contractibility of
U(M(A)) for all σ-unital stable C *-algebras A. (NOW STILL OPEN QUESTION
!!!).

1. Operations on Multiplier algebras (1)

List of properties of Multiplier algebras:

A σ-unital (and e ∈ A+ strictly positive contraction):
If e ∈ B ⊆ A then the natural C *-morphism h : B → A given by the inclusion
map defines defines a *-monomorphism from M(B) into M(A). The image consists
of all T ∈ M(A) with T e, eT ∈ B.
For each separable C *-subalgebra C of M(A) exists separable B ⊆ A with
e ∈ B, and eC ∪ Ce ∈ B.
In particular C ⊆ M(B) ⊆ A.
If J ⊆ A is closed ideal then A/J has the strictly positive element f := πJ (e).
If D ⊆ M(A/J) is a separable C *-subalgebra, then there exists a separable C *-
subalgebra B of A such that e ∈ B and Df ∪f D ⊆ πJ (B). Thus, D ⊆ M(πJ (B)) ⊆
M(A/J). Then the natural epimorphism πJ |B from B onto πJ (B) defines an epi-
morphism from M(B) ⊆ M(A) onto M(πJ (B)), and D ⊆ M(πJ (B)) ⊆ M(A/J).
In particular, D ⊆ M(πJ )(M(A)). Thus, M(πJ ) : M(A) → M(A/J) is surjective.
Compare with Remark B.1.1 !!!

Remark B.1.1 (Non-commutative Tietze extension). Let A a σ-unital C *-

algebra, i.e., there exists a an element e ∈ A+ with kek = 1 that is a strictly
positive element of A.
Then for every closed ideal J of A the natural (on bounded parts strictly con-
tinuous) C*-morphism M(πJ ) : M(A) → M(A/J) is surjective.
The kernel of M(πJ ) is N (A, J) := {T ∈ M(A) ; T e, eT ∈ J}.
In case where A is separable, this is [616, prop. 3.12.10].
If A is not separable and e ∈ A+ as strictly positive element with kek = 1 then,
for every unital separable C *-subalgebra B ⊆ M(A/J) there exists a separable C *-
subalgebra E ⊆ A with e ∈ E and BD ∪ DB ⊆ D for D := πJ (E) ∼ = E/(J ∩ E).
The strictly continuous extension M(πJ∩E ) : M(E) → M(E/(J ∩ E)) of the
epimorphism πJ∩E : E → E/(J ∩ E) ∼ = D is surjective by [616, prop. 3.12.10].
 2. ACTIONS AND MONOTONE MAPS 1209

The algebra B is contained in M(D) ⊆ M(A/J) – if we naturally identify

M(D) with the two-sided multipliers of D in M(A/J). In the same way M(E)
is naturally isomorphic to algebra the two-sided multipliers of E in M(A). The
strictly continuous extension M(πJ ) : M(A) → M(A/J) of πJ maps M(E) into
M(D). It becomes there under natural identification of E/(J ∩E) with D the same
as the natural strictly continuous unital C *-morphism M(πE∩J ) from M(E) into
M(E/(J ∩ E)).
Thus, B ⊆ M(A/J) is contained in the image of M(πJ ) : M(A) → M(A/J).

Above has to be discussed / reformulated

2. Actions and monotone maps

The Chapter 0 of [321] could be a reference for this section, but all results
mentioned here are elementary exercises for the reader. Recall that a set Y with
transitive, reflexive and antisymmetric relation x ≤ y is a partially ordered set
(poset). We need only very special posets, namely those that are order isomorphic
or order anti-isomorphic to the lattice of the ideals (or: ideal lattice) of a ring.
Obviously, the family O(X) of open subsets of a T0 space X and the family
F(X) of closed subsets of X, and the set X itself are all partially ordered sets with
the inclusions U ⊆ V , F ⊆ G, and x ≤ y if and only if x ∈ {y} .
The order reversing isomorphisms F 7→ X \ F and U 7→ X \ U show that F(X)
is just the opposite ordered space of O(X).
Subsets Z of O(X) or F(X) have unique inf Z a and sup Z, i.e., they are
complete lattices. Moreover O(X) is a Heyting algebra (also called frame), because
[ [
W∩ Uγ = W ∩ Uγ .
γ γ

The infimum {Uγ } is the interior ( γ Uγ )◦ of the intersection of a family {Uγ } ⊆

V T

O(X). The supremum in F(X) of a family of closed subsets is the closure of its
union.
A closed subset F of a T0 space X is prime if it is not the union of two closed
subsets F1 , F2 ⊆ F , that are both different from F . Equivalently this means: If
F ⊆ Z1 ∪ Z2 , then F ⊆ Z1 or F ⊆ Z2 . A T0 spaces X is sober (also called
point-wise complete, or point-complete ) if each prime closed subset F ⊆ X is the
closure of a point of X.
For any T0 space X there is a unique (up to natural isomorphisms) sober space
X c and an embedding η : X → X c such that U ∈ O(X c ) 7→ η −1 (U ) ∈ O(X) is a
lattice isomorphism. X c is called the sobrification of X.
In particular, each sober T0 space X is completely determined likewise by each
of the lattices O(X) or F(X).
The sobrification Prim(A)c of the space of primitive ideals Prim(A) is naturally
isomorphic to prime(A) (the space of prime ideals of A) if A is a C *-algebra.
1210 B. EXACT C*-ALGEBRAS AND EXAMPLES

If A is a C *-algebra, then there are natural order isomorphisms I(A) = ∼

O(Prim(X)) and order reversing isomorphism from the lattice of closed ideals I(A)
onto onto F(A).
I(A), O(X), F(X) as posets
lower s.c. and upper s.c. maps
case I(A) (lattice of closed sets of
notation ↑ X, ↑ t, ↓ X , ↓ t
Galois connections

Definition B.2.1. Let X and Y posets and φ : Y → X and ψ : X → Y maps

that satisfy:

(i) φ and ψ are monotone (order preserving), and

(ii) the relations φ(y) ≥ x and y ≥ ψ(x) are equivalent for all (x, y) ∈ X × Y .

Then (φ, ψ) is called Galois connection or adjunction between X and Y .

The map φ is the upper adjoint and ψ is the lower adjoint.

Let C ⊆ CP(A, B) a matricially operator-convex cone. Consider the lower

semi-continuous action ΨC : I(B) → I(A) and the upper semi-continuous action
ΦC : I(A) → I(B) as defined in Definition ??. Then (ΨC , ΦC ) is a Galois connection,
and ΨC is the upper adjoint and ΦC is the lower adjoint of the connection.
The following proposition equips us with the in Chapters 3 and 6 needed ob-
servation on Galois connections.

Proposition B.2.2. Let X and Y partially ordered sets and φ : Y → X and

ψ : X → Y monotone (increasing) maps.

(i) The formula ψ(x) := inf φ−1 (↑ x) for x ∈ X defines a lower adjoint of φ
(such that (φ, ψ) is a Galois connection), if φ(inf Z) = inf φ(Z) so far as
inf Z exists in X, and Y is a complete lattice, or Y is a complete semi-
lattice and φ : Y → X is co-final (i.e., for all x ∈ X there is y ∈ Y with
x ≤ φ(y) — or equivalently: φ−1 (↑ x) 6= ∅ ).
(ii) If (φ, ψ) is a Galois connection, then ψ(x) := min φ−1 (↑ x) for all x ∈ X
and φ(y) := max ψ −1 (↓ y) for all y ∈ Y .
In particular, they determine each other.
(iii) If (φ, ψ) is a Galois connection, then the upper adjoint φ is inf-preserving
and the lower adjoint ψ is sup-preserving.
(iv) compare products with id
(v) when ψ ◦ φ or ψ ◦ φ is identity?

definitions and uniqueness of adjoints

uniqueness of the upper and lower Galois adjoints
in particular, the upper and the lower map of a Galois connection determine
each other uniquely.
 3. TOPOLOGICAL ACTIONS 1211

A reference could be [321, chp. 0].

3. Topological actions

Let L denote a set (or a lattice), X and Y topological spaces, Ψ1 : L → O(X)

and Ψ2 : L → O(Y ) map (increasing if L is a lattice). We say that an increasing
map Ψ : O(Y ) → O(X) is fitting (or Ψ1 -Φ2 compatible) if Ψ(Ψ2 (`)) ⊃ Ψ1 (`) for
` ∈ L.
Recall that a map Ψ : O(Y ) → O(X) is is lower semi-continuous if the image
Ψ(V ) of the interior V := ( α Vα )◦ of the set α Vα is the interior of α Ψ(Vα ) ,
T T T

for each family {Vα } of open subsets Vα ⊆ Y . Lower s.c. actions Ψ are increasing.

Lemma B.3.1. Let L denote a set (or a lattice), X and Y topological spaces,
Ψ1 : L → O(X) and Ψ2 : L → O(Y ) maps (increasing if L is a lattice).

(i) Let S and A sets and (ψ, α) ∈ S × A 7→ Wψ,α ∈ O(X) a map. Then
  ◦ ◦  !◦ ◦
\ \ \ \
  Wψ,α   =  Wψ,α  .
α ψ ψ α

(ii) If S 6= ∅ is a set of lower semi-continuous maps Ψ : O(Y ) → O(X), then

the map
\
Ψ0 : O(Y ) 3 V 7→ ( Ψ(V ))◦ ∈ O(X)
Ψ∈S

is lower semi-continuous.
(iii) There is a minimal element Ψ0 : O(Y ) → O(X) in the set S of the fitting
lower s.c. maps, i.e., Ψ0 (V ) ⊆ Ψ(V ) for every open subset V of Y and
every fitting l.s.c. map Ψ : O(Y ) → O(X) .
Moreover, Ψ0 (V ) ⊃ `∈L(V ) Ψ1 (`) for every open subset V of Y , where
S

L(V ) := {` ∈ L ; Ψ2 (`) ⊆ V } .
(iv) If Ψ2 (`1 ) ⊆ Ψ2 (`2 ) implies Ψ1 (`1 ) ⊆ Ψ1 (`2 ) for `1 , `2 ∈ L, and if x ∈
T
α Ψ1 (`α ) 6= ∅ implies the existence of ` ∈ L with Ψ2 (`) ⊆ Ψ2 (`α ) for all
α ∈ A and x ∈ Ψ1 (`), then the minimal fitting map Ψ0 in part (iii) satisfies
Ψ0 (V ) = `∈L(V ) Ψ1 (`) for each V ∈ O(Y ), and Ψ0 (Ψ2 (`)) = Ψ1 (`) for
S

all ` ∈ L.
(v) An increasing map Ψ : O(Y ) → O(X) is lower semi-continuous,
if and only if,
the map

λ : Y 3 y → λ(y) := X \ Ψ(Y \ {y}) ∈ F(X)

S
from Y to F(X) satisfies λ(y) ⊆ z∈Z λ(z) for each subset Z ⊆ Y and
y ∈ Z,
if and only if,
the function g(y) := sup f (λ(y)) is lower semi-continuous for every
bounded lower semi-continuous function f : X → [0, ∞).
1212 B. EXACT C*-ALGEBRAS AND EXAMPLES

Proof. (i): Since ( ψ Wψ,α )◦ is contained in Wψ,α for each (ψ, α), we get
T

that α ( ψ Wψ,α )◦ is contained in α Wψ,α for each ψ ∈ S. Thus, the interior of

T T T
◦ ◦
T T T
α ( ψ Wψ,α ) is contained in ( α Wψ,α ) for each ψ ∈ S. This implies that

 ◦ !◦
 WΨ,α  ◦ ⊆
\ \
\ \
◦
Wψ,α .
α ψ ψ α

◦ ◦
T T T T
The interior of ψ ( α Wψ,α ) is contained in α ( ψ Wψ,α ) by a similar argument.

(ii): Let {Vα }α∈A a family of open subsets of Y , and let WΨ,α := Ψ(Vα )
for Ψ ∈ S and α ∈ A. Then Ψ0 (( α Vα )◦ ) is the interior of Ψ Ψ ( α Vα )◦ =
T T T

WΨ,α )◦ (by lower semi-continuity of Ψ ∈ S). On the other hand,

T T
(
TΨ 0 α T T ◦ 0
α Ψ (Vα ) = α ( Ψ WΨ,α ) by definition of Ψ and of WΨ,α . Now apply part (i)
and the definition of lower semi-continuity.
(iii): Let Ψ0 (V ) := X for all open subsets V ⊆ Y (including the case V = ∅).
It is lower s.c. and is fitting. Thus, the set S of fitting l.s.c. maps Ψ : O(Y ) → O(X)
is not empty.
Let Ψ0 (V ) ∈ O(X) denote the interior of the set Ψ∈S Ψ(V ) for each V ∈ O(Y ).
T

Then, Ψ1 (`) ⊆ Ψ0 (Ψ2 (`)) for all ` ∈ L and Ψ0 : O(Y ) → O(X) is lower semi-
continuous by part (ii). Since Ψ0 ∈ S, it is the minimal l.s.c. map Ψ ∈ S. Since Ψ0
is fitting, Ψ1 (`) ⊆ Ψ0 (V ) for each ` ∈ L with Ψ2 (`) ⊆ V .
T
(iv): We define L(V ) := {` ∈ L ; Ψ2 (`) ⊆ V } . Then, α L(Vα ) =
L( ( α Vα )◦ ) , ` ∈ L(Ψ2 (`)) for ` ∈ L, and L(V1 ) ⊆ L(V2 ) for V1 ⊆ V2 .
T
S
Now let Φ(V ) := `∈L(V ) Ψ1 (`) for V ∈ O(Y ). Then Ψ1 (`) ⊆ Φ(Ψ2 (`)) for
` ∈ L, Φ(V ) ⊆ Ψ0 (V ) for all V ∈ O(Y ) (by Part (iii)), and Φ(V1 ) ⊆ Φ(V2 ) for
V1 ⊆ V2 . In particular, Φ(( α Vα )◦ ) is contained in the interior of α Φ(Vα ).
T T

The reversed inclusion can be derived from the rather strong assumption, that
T
x ∈ α Ψ1 (`α ) 6= ∅ implies the existence of ` ∈ L with Ψ2 (`) ⊆ Ψ2 (`α ) for all
α ∈ A and x ∈ Ψ1 (`).
In particular Φ is lower s.c. and fitting. Thus, Φ(V ) ⊃ Ψ0 (V ).
Since `2 ∈ L(Ψ2 (`2 )) = {`1 ∈ L ; Ψ2 (`1 ) ⊆ Ψ2 (`2 )}, we have Ψ1 (`1 ) ⊆ Ψ1 (`2 )
for `1 ∈ L(Ψ2 (`2 )), if Ψ2 (`1 ) ⊆ Ψ2 (`2 ) implies Ψ1 (`1 ) ⊆ Ψ1 (`2 ). Thus Ψ0 (Ψ2 (`)) =
Ψ1 (`) in this case.
S
(v): We write λ(Z) for the union z∈Z λ(z) ⊆ X.
Suppose that Ψ : O(Y ) → O(X) is an increasing lower semi-continuous map,
i.e.,
\ \
( Ψ(Uτ ))◦ = Ψ(( Uτ )◦ ) .
τ τ

Let Z ⊆ Y , then y ∈ Z, if and only if,

\
Y \ {y} ⊃ Y \ Z = ( Y \ {z})◦ .
z∈Z
 4. SOME PROPERTIES OF EXACT C *-ALGEBRAS 1213

It follows that
\
Ψ(Y \ {y}) ⊃ (Ψ( Y \ {z}))◦ .
z∈Z

The right side is the same as X \ λ(Z), by the rule X \ ( τ Vτ )◦ = τ (X \ Uτ ).

T S

Thus, λ(y) ⊆ λ(Z) for each subset Z ⊆ Y and y ∈ Z.

Suppose that the map

λ : Y 3 y → λ(y) := X \ Ψ(Y \ {y}) ∈ F(X)

from Y to F(X) satisfies:

λ(y) ⊆ λ(Z) for each subset Z ⊆ Y and y ∈ Z.
Let f : X → [0, ∞) l.s.c., and let g(y) := sup f (λ(y)). Then W := f −1 [0, t] is closed,
and y ∈ Z := g −1 [0, t], if and only if, λ(y) ⊆ W . Since W is closed and λ(Z) ⊆ W
we have λ(y) ⊆ W for all y ∈ Z. Thus, g −1 [0, t] is closed for all t ∈ [0, ∞), i.e., g is
l.s.c.
Suppose that the function g(y) := sup f (λ(y)) is lower semi-continuous for
every bounded lower semi-continuous function f : X → [0, ∞).
Consider the map F : F(Y ) → F(X) given by F (Z) := X \Ψ(Y \Z). F is increasing.
F ({y}) = λ(y).
S S
We must show that F (Zτ ) has closure F ( Zτ ) for each family {Zτ } of closed
subsets of Y . Since F is increasing,
[ [
F (Zτ ) ⊆ F ( Zτ ) .
S S
Let Z := Zτ . Then Z = z∈Z {z}, and
[ [
F (Zτ ) ⊃ F ({z}) = λ(Z)
z∈Z

holds because F is increasing. Let U := X \ λ(Z) and let f : X → [0, 1] the

characteristic function of U . Then g(y) := sup f (λ(y)) is lower s.c. and the closed
subset G := g −1 (0) is given by y ∈ Y with λ(y) ∈ λ(Z). In particular, G contains
the closure of Z. We get F (Z) ⊆ X \ U . Thus,
[ [
F ( Zτ ) = F (Z) ⊆ λ(Z) ⊆ F (Zτ ) .

4. Some properties of exact C *-algebras

Definition B.4.1. We say that a C *-subalgebra E ⊆ A is a regular subal-

gebra of A, if, for all closed ideals J1 , J2 of A,

(i) E ∩ (J1 + J2 ) = (E ∩ J1 ) + (E ∩ J2 ), and

(ii) J1 = J2 if E ∩ J1 = E ∩ J2 .
1214 B. EXACT C*-ALGEBRAS AND EXAMPLES

In other words this says, that Ψup

A,E (J) := E ∩ J defines a continuous action of
X := Prim(A) on E in the sense of Definition 1.2.6, because of (i), and this action
is injective if considered as a map from OX ∼ = I(A) into I(E), by (ii).
It is shown in [464] that A ⊗ O2 contains a regular Abelian C *–subalgebra
E for every separable nuclear C *-algebra A. We obtain in Chapter 12, that this
implies the same for separable exact C *-algebras A.

The Lemma 2.2.3 yields another useful result:

Proposition B.4.2. Suppose that A and B are separable C*-algebras, and

that A is exact.

(i) Every primitive ideal J of A ⊗ B is a sum J = (J1 ⊗ B) + (A ⊗ J2 ), where

J1 and J2 are primitive ideals of A and B, respectively.
(ii) Every closed ideal J of A ⊗ B is the closure of the sum of the family of all
elementary ideals J1 ⊗ J2 ⊆ J containing in J, where J1 ⊆ A and J2 ⊆ B
are closed ideals.
Here ‘‘elementary’’ ideals are defined as the
tensor products J1 ⊗ J2 ??
(iii) The 1-1-map from Prim(A) × Prim(B) onto Prim(A ⊗ B) which is given
by (i) is a homeomorphism from the Tychonoff product of Prim(A) and
Prim(B) onto Prim(A ⊗ B).
(iv) If E ⊆ A and F ⊆ B are C*-subalgebras which separate the closed ideals
of A and B, in the sense that e.g. E ∩ J1 = E ∩ J2 implies J1 = J2 , then
again E ⊗ F separates the closed ideals of A ⊗ B.
If, moreover, E and F are regular subalgebras of A and B in the sense
of Definition B.4.1, then E ⊗ F is a regular subalgebra of A ⊗ B.

Proof. (i): The tensor product d = d1 ⊗d2 of irreducible representations d1 of

A and d2 of B is irreducible. d defines a faithful representation of (A/J1 ) ⊗ (B/J2 ),
where J1 and J2 are the kernels of d1 and d2 . Since A and, therefore, A/J1 and J1
are exact, the kernels of A ⊗ B → A ⊗ (B/J2 ), of (A/J1 ) ⊗ B → (A/J1 ) ⊗ (B/J2 )
and of J1 ⊗ B → J1 ⊗ (B/J2 ) are A ⊗ J2 , (A/J1 ) ⊗ J2 and J1 ⊗ J2 , respectively.
Since exact C *-algebras are locally reflexive, the kernels of A ⊗ B → (A/J1 ) ⊗ B
and of A ⊗ (B/J2 ) → (A/J1 ) ⊗ (B/J2 ) are J1 ⊗ B and J1 ⊗ (B/J2 ). Thus the kernel
of d is (J1 ⊗ B) + (A ⊗ J2 ), as the 3×3-lemma shows.
Thus λ : (J1 , J2 ) 7→ J1 ⊗B +A⊗J2 maps Prim(A)×Prim(B) into Prim(A⊗B).
A general irreducible representation d of A ⊗ B defines commuting factorial
representations d1 of A and d2 of B, such that d(a ⊗ b) = d1 (a)d2 (b). By a result of
J. Dixmier, the kernels J1 of d1 and J2 of d2 are primitive. (J1 ⊗ B) + (A ⊗ J2 ) is
contained in the kernel of d, and d defines a C *-norm on the algebraic tensor product
A/J1 B/J2 which majorize the spatial norm, cf. [704, prop.1.20.5,prop.1.22.7].
Thus the kernel of d is (J1 ⊗ B) + (A ⊗ J2 ).
(ii):
 4. SOME PROPERTIES OF EXACT C *-ALGEBRAS 1215

Check again
Let J0 be the closure of the sum of all elementary ideals which are contained
in J. Suppose that there exists d ∈ J+ such that d is not in J0 . Then, by (i), there
exists primitive ideals J1 ⊆ A and J2 ⊆ B such that J0 ⊆ J1 ⊗ B + A ⊗ J2 and A ⊗
B → (A/J1 ) ⊗ (B/J2 ) maps d into a non-zero positive element of (A/J1 ) ⊗ (B/J2 ).
Thus there exist pure states ϕ on A and ψ on B such that ϕ ⊗ ψ(J0 ) = 0 and
ϕ ⊗ ψ(d) > 0. By Lemma 2.2.3, there exists z ∈ A ⊗ B, e ∈ A+ and f ∈ B+ with
z ∗ z ∈ dAd, e ⊗ f = zz ∗ , ϕ(e) > 0 and ψ(f ) > 0. Hence e ⊗ f ∈ J, ϕ ⊗ ψ(e ⊗ f ) > 0
and the elementary ideal generated by e ⊗ f is contained in J0 . A contradiction to
the choice of ϕ and ψ.
(iii): Every open subset of the Tychonoff product Prim(A) × Prim(B) is the
union of cartesian products Y × Z of open subsets Y of Prim(A) and Z of Prim(B).
They correspond to closed ideals K1 ⊆ A and K2 ⊆ B, and λ maps Y × Z onto
the open subset of Prim(A ⊗ B) which corresponds to K1 ⊗ K2 . Thus, by (ii), λ
maps the open subsets of the Tychonoff product Prim(A) × Prim(B) onto the open
subsets of Prim(A ⊗ B).
(iv): The natural map J 7→ E ∩ J defines a monotone map Ψ1 := Ψup A,E from
the open subsets of Prim(A) into the open subsets of Prim(E) (∼ = closed ideals
of E), which satisfies the conditions (ii),(iii) and (iv) of Definition 1.2.6. E and
F separate the closed ideals of A and B, if and only if, Ψ1 and Ψ2 := Ψup B,F are
injective.
We consider Ψ3 := Ψup A⊗B,E⊗F as a map from the open subsets of Prim(A) ×
Prim(B) into the open subsets of Prim(E) × Prim(F ). By the monotony and
by condition (ii) of Definition 1.2.6 for Ψ3 , it is enough to check the injectivity
of the map Ψ3 on the base of the topology of Prim(A) × Prim(B). This base
is given by the cartesian products of open subsets of Prim(A) and of Prim(B).
Ψ3 (Y × Z) = Ψ1 (Y ) × Ψ2 (Z), for open subsets Y of Prim(A) and Z of Prim(B),
because, for closed ideals J ⊆ A and K ⊆ B, (E ⊗F )∩(J ⊗K) = (E ∩J)⊗(F ∩K).
The latter identity follows from the exactness of E ⊆ A. The cartesian product
map Ψ1 × Ψ2 is injective, because Ψ1 and Ψ2 are injective.
Now suppose that, moreover, that E is regular in A and that F is regular in
B. E and F are regular, if and only if, moreover, Ψ1 and Ψ2 satisfy condition (i)
of Definition 1.2.6, e.g., Ψ1 (Z1 ∪ Z2 ) = Ψ1 (Z1 ) ∪ Ψ1 (Z2 ) for open subsets Z1 and
Z2 of Prim(A). We want to deduce that Ψ3 (X1 ∪ X2 ) = Ψ3 (X1 ) ∪ Ψ3 (X2 ) for open
subsets X1 and X2 of Prim(A ⊗ B).
By the monotony and by condition (ii) of Definition 1.2.6 for Ψ3 , it suffices to
show that, for open subsets Yk of Prim(A) and Zk of Prim(B),

[ [
Ψ3 ( Yk × Zk ) ⊆ Ψ3 (Yk × Zk ) .
1≤k≤n 1≤k≤n

Note that Ψ1 (Prim(A)) = Prim(E) and Ψ2 (Prim(B)) = Prim(F ).

1216 B. EXACT C*-ALGEBRAS AND EXAMPLES

If J ⊆ A and K ⊆ B are closed ideals, then the natural epimorphism from

A ⊗ B onto (A/J) ⊗ (B/K) has kernel A ⊗ K + J ⊗ B, because A is exact and,
therefore, locally reflexive. The same applies to J ∩ E, K ∩ F , E and F , and
E/(E ∩ J) ⊗ F/(F ∩ K) is a subalgebra of A/J ⊗ B/K. Thus

((A ⊗ K) + (J ⊗ B)) ∩ (E ⊗ F ) = (E ⊗ F ) ∩ (A ⊗ K) + (E ⊗ F ) ∩ (J ⊗ B),

which means that

Ψ3 (Y × Prim(B)) ∪ Ψ3 (Prim(A) × Z) = Ψ3 ((Y × Prim(B)) ∪ (Prim(A) × Z)) .

Let (I1 , I2 ) be a point of Prim(E) × Prim(F ) which is not in in the union of

the sets Ψ1 (Yk ) × Ψ2 (Zk ) for k = 1, . . . , n. Then there are subsets S and T of
S
{1, . . . , n}, with S ∪ T = {1, . . . , n}, such that I1 is not in k∈S Ψ1 (Yk ) and I2 is
S
not in k∈T Ψ2 (Zk ).
S
Since E and F are regular subalgebras, we have Ψ1 (Y ) = k∈S Ψ1 (Yk ) and
S S S
Ψ2 (Z) = k∈T Ψ2 (Zk ), where Y := k∈S Yk and Z := k∈T Zk .
Thus (I1 , I2 ) is not in (Ψ1 (Y ) × Prim(F )) ∩ (Prim(E) × Ψ2 (Z)). The latter is
the same as Ψ3 ((Y × Prim(B)) ∩ (Prim(A) × Z)) .
S
But Ψ3 is monotonous, and Yk × Zk ⊆ (Y × Prim(B)) ∩ (Prim(A) × Z). 

Corollary B.4.3. Suppose that A1 , A2 , . . . is a sequence of separable exact

C*-algebras (with An unital for n ≥ n0 ).

(i) There is a natural homeomorphism from the Tychonoff product

Prim(A1 ) × Prim(A2 ) × . . . onto Prim(A1 ⊗ A2 ⊗ . . .).
(ii) If Bn are regular C*-subalgebras of An , n = 1, 2, . . ., then B1 ⊗ B2 ⊗ . . .
is a regular C*-subalgebra of A2 ⊗ A2 ⊗ . . . .

Proof. Let A := A1 ⊗ A2 ⊗ . . ., Cn := A1 ⊗ A2 ⊗ . . . ⊗ An , Dn := An+1 ⊗

An+2 ⊗ . . ., and let 1n denote the unit element of Dn . Then, for every closed
ideal J of A, A/J is the inductive limit of πJ (Cn ⊗ 1n ). Thus J is the inductive
limit both of Jn ⊗ 1n and, therefore, of Jn ⊗ Dn , where Jn ⊆ Cn is defined by
Jn ⊗ 1n = (Cn ⊗ 1n ) ∩ J.
(i): By Proposition B.4.2, we get that every primitive ideal of A is the closure
of the sum of ideals Cn−1 ⊗ Jn ⊗ Dn , where Jn is a primitive ideal of An for
n = 1, 2, . . . .
Conversely, let Jn ⊆ An a sequence of primitive ideals. We find pure states ψn
on An which define irreducible representations with kernel Jn . By [704, prop. 4.3.4],
the infinite tensor product of pure states is pure. Thus, the kernel of the natural
epimorphism from A onto the infinite tensor product of the quotients An /Jn is
primitive. This shows the set-theoretic isomorphism of Prim(A) with the Tychonoff
product of the sequence Prim(An ). The topological isomorphism means that one
can find for both a bases of the topology which is mapped by the set-theoretic
isomorphism onto each other. By definition of the Tychonoff product this means
 4. SOME PROPERTIES OF EXACT C *-ALGEBRAS 1217

that every ideal of A is the closure of a sum of ideals J1 ⊗ . . . ⊗ Jn ⊗ Dn . This

follows from Proposition B.4.2 and the remark at the beginning of the proof.
(ii): By, Proposition B.4.2(iv), for n = 1, 2, . . ., Gn := B1 ⊗ . . . ⊗ Bn ⊗ 1n
is a regular subalgebra of Cn ⊗ 1n . This and Gn ⊆ Gn+1 imply that the closure
S S
B1 ⊗ B2 ⊗ . . . of n Gn is regular in the closure A of n Cn . 

If we combine Proposition B.4.2 with the criteria in [432], we get:

Corollary B.4.4. A separable C*-algebra A is exact, if and only if, for every
separable C*-algebra B, the natural map from Prim(A) × Prim(B) to Prim(A ⊗ B),
which is induced by the tensor product of irreducible representations, is a topological
isomorphism. The isomorphism is the same as in Proposition B.4.2(i).

Remark B.4.5.
(1) A lower s.c. quasi-trace τ : A+ → [0, ∞] is 2-additive if and only if it is an
R∞
integral τ (a) = 0+ D((a − t)+ ) dt of a lower s.c. sub-additive dimension function
D : A → [0, ∞]. One gets D back from τ by D(a) = supδ>0 τ (fδ (a∗ a)), where
fδ (t) := min(δ −1 (t − δ)+ , 1) for δ > 0.
(2) Every lower semi-continuous 2-quasi-trace τ : A → [0, ∞] on an exact
C *-algebra is additive. (In case of a unital A this is a result of Haagerup and
Thorbjørnsen, [342] and [348, cor. 9.14]. It extends to the non-unital case by the
below given Lemma B.4.6.)
(3) Every lower semi-continuous dimension function D : A → [0, ∞] on A inte-
R∞
grates to a l.s.c. 2-quasi-trace τD on A+ by τD (a) = 0+ D((a − t)+ )dt.
(4) A stable simple C *-algebra A contains a (non-zero) properly infinite pro-
jection, if and only if, there is no non-zero dimension function D on the Pedersen
ideal of A (cf. Blackadar and Cuntz [78]).
(??? This should be also my remark in the 1994 book from Canada workshop
???)
(5) Summing up (1)–(4), we get:
If A is simple and exact, then T + (A) = {0} if and only if A is stably infinite.
In particular, simple stably projection-less exact C*-algebras have always non-zero
additive traces, cf. [441]. The T + (A) = {0} implies for simple nuclear A that
A ⊗ K contains a (non-zero) properly infinite projection.

Lemma B.4.6. Let τ : A+ → [0, ∞] a lower semi-continuous 2-quasi-trace.

(i) The map τ and its dimension function D : A → [0, ∞] are order monotone
on A+ , and D with respect to Cuntz majorization a - b on A+ .
(ii) If, for every contraction c ∈ A+ with τ (c) < ∞, the restriction τ |B+ is
additive for B := {a ∈ A ; ac = a = ca}, then τ is additive on A+ .
(iii) If τ (A+ ) ⊆ [0, ∞) then γ := sup{τ (a) ; kak ≤ 1, a ∈ A+ } < ∞ and
τe(b + z1) := zγ + τ (b+ ) − τ (b− ) (for b + z1 ∈ A
e+ ) defines a bounded
2-quasi-trace on the unitization Ae of A.
1218 B. EXACT C*-ALGEBRAS AND EXAMPLES

Proof. (i): Let D : A → [0, ∞] the l.s.c. dimension function corresponding to

τ , i.e.,

D(a) := lim sup τ ((((a∗ a) − δ)+ )γ )

δ>0,γ>0

and let 0 ≤ a ≤ b or a - b.
Then (a−t)3+ ≤ (a−t)+ (b−t)+ (a−t)+ ≤ kbk(a−t)2+ , because (b−t)+ ≤ b ≤ kbk
by (b − t)+ = (b − t) + (b − t)− ≤ b and (a − t)+ − (a − t)− = a − t ≤ b − t ≤
(b − t)+ and the order preserving map X 7→ (a − t)+ X(a − t)+ on X ∗ = X implies
(a − t)3+ ≤ (a − t)+ (b − t)+ (a − t)+ .
It induces: For ε > δ > 0 there is x ∈ A with x∗ x = fε (a) and xx∗ fδ (b) = xx∗ .
It follows τ (fδ (a)) ≤ τ (f (b)) ≤ D(b). Thus D(a) ≤ D(b). From D(aα ) = D(a)
and the monotony and centrality of D it follows D((a − t)+ ) ≤ D((b − t)+ ) for all
R∞
t > 0 if 0 ≤ a ≤ b. Since τ (a) = 0+ D((a − t)+ ) dt, we get from it that τ (a) ≤ τ (b),
(ii): Let a, b ∈ A+ with τ (a) + τ (b) < ∞.
Then ?????
show that τ ((a − ε)+ + (b − ε)+ ) < ∞ ???? Fill Proof in !!! ??
(iii): Suppose that γ = ∞ then there exists positive contractions a1 , a2 , . . . ∈
P −n
A+ with τ (an ) > 4n . Then a := 2 an is a positive contraction in A with
n n n
an ≤ 2 a, and 4 < 2 τ (a) for all n ∈ N by part (i). Thus τ (a) = ∞, contradicting
????? that ???
If b ∈ A and z ∈ C then b + z1 ≥ 0 implies z ≥ 0, b∗ = b and kb− k ≤ z.
Thus τe(b + z1) ≥ 0 for b + z1 ≥ 0. If 0 ≤ bj + zj 1 (j = 1, 2) commute, then b1
and b2 commute. The additivity of τ on C ∗ (b1 , b2 ) induces the additivity of τe on
C ∗ (b1 , b2 , 1)+ .
It follows that τe is a bounded local quasi-trace on A
e with bounded lower semi-
continuous rank-function D(x + z1) = γ if z 6= 0 and D(x) = D(x) for x ∈ A and
e e
z ∈ C.
???????????????????? check
the point is the existence of D
f2 !! 

5. Approximate divisible algebras

Remark B.5.1. Recall that a separable C *-algebra B is called approximately

divisible if there exist a sequence of unital *-morphisms hn : M2 ⊕ M3 → M(B)
with limn→∞ khn (a)b − bhn (a)k = 0 for all a ∈ M2 ⊕ M3 and b ∈ B.
B := A ⊗ O∞ is approximately divisible for every separable C*-algebra A.
(Proof: C ∗ (s1 , s2 , · · · ) ∼
= O∞ ∼
= O∞ ∼
= O∞ ⊗ O∞ ⊗ · · · by Corollary F(ii), –
or the argument following Corollary H –, and M2 ⊕ M3 is unitally contained in
O∞ = C ∗ (s1 , s2 , . . .).
 6. POSITIONS OF C *-SUBALGEBRAS IN O2 1219

This is because we can take M3 := C ∗ (si s∗j ; 1 ≤ i, j ≤ 3) and M2 :=

C ∗ (tk t∗` ; 1 ≤ k, ` ≤ 2), where t1 , t2 are suitable partial isometries with t1 t∗1 + t2 t∗2 =
p := 1 − (s1 s∗1 + s2 s∗2 + s3 s∗3 ) and t∗1 t1 = t∗2 t2 = q := 1 − (s1 s∗1 + s2 s∗2 ).
The existence of t1 and t2 follows from K0 (O∞ ) ∼ = Z, with [p] = −2 and
[q] = −1 and the pure infiniteness of pO∞ p, e.g. by cf. Lemma 4.2.6.
In fact, the projections r := s4 qs∗4 + s5 qs∗5 and p satisfy r ≤ p and −2 = [p] =
[r] ∈ K1 (O∞ ). Thus p and r are MvN-equivalent in O∞ . Let z ∈ O∞ a partial
isometry with zz ∗ = p and z ∗ z = r. The partial isometries t1 := zs4 q, t2 := zs5 q
have the desired property.
It is known that the infinite tensor product (M2 ⊕ M3 )⊗∞ := (M2 ⊕ M3 ) ⊗
(M2 ⊕ M3 ) ⊗ · · · contains the GICAR-algebra unitally, which is a simple unital
AF-algebra defined by the “Pascal triangle” with multiplicities given by the integer
coefficients of the polynomials Pn (t) := (t + 1)n , cf. [270].
Every infinite-dimensional simple unital AF-algebra contains the Jiang-Su al-
gebra Z, [391, cor. 6.3]. It implies that A ⊗ Z ∼ = A for all “approximately
divisible” A, i.e., those A that have a unital C *-morphism of M2 ⊗ M3 into
F (A) := (A0 ∩ Aω )/ Ann(A, Aω ). (The F (A) is a“stable” invariant of separable
C *-algebras A in the sense that F (A) ∼
= F (A ⊗ K).)
It seems that one can replace the assumption of approximate divisibility in
most applications by the weaker requirement that A tensorial absorbs Z, i.e., that
A⊗Z ∼ = A. The main reason could be something like the pull-back proposition
[690, prop. 6.5] of M. Rørdam.
E.g., separable exact A is strongly p.i., if A ∼
= A ⊗ Z and every l.s.c. trace
τ : A+ → [0, ∞] takes only values in {0, ∞}. (For separable nuclear A holds also
the converse, because then A ∼= A ⊗ O∞ and one can use that O∞ ⊗ Z = O∞ .)
Is above partly shown in Chapter 2?
That A ⊗ Z s.p.i. for nuclear separable
trace-less A is [690, thm. 5.2].
The proof in [690, thm. 5.2] shows
that this holds also for exact trace-less A,
and more generally for all C *-algebras
that have only trivial l.s.c. traces. ???????

Further topics. To be filled in. ??

6. Positions of C *-subalgebras in O2

There exist a separable unital nuclear C *-algebra A and a unital monomor-

phisms i : A → O2 of A into the Cuntz algebra O2 , such that, there does not exist
a conditional expectation from O2 onto i(A).
But all separable unital nuclear C *-algebras A have a unital *-monomorphism
1220 B. EXACT C*-ALGEBRAS AND EXAMPLES

ι : A → O2 such that there exists a conditional expectation from O2 onto ι(A), i.e.,
a completely positive contraction E : O2 → ι(A) with E ◦ ι = ι.
There is an interesting connection to some conjecture on equivalence of embed-
dings that are in “sufficiently general position” (see below).

Weaker results on embeddings ι : A ,→ O2 , as there are (for example):

One finds always for separable unital nuclear C *-algebras A at least one unital
embedding ι : A ,→ O2 such that there is a conditional expectation E from O2 onto
ι(A).
On the other hand, for any other embedding κ : A ,→ O2 , there is a norm-
continuous path t ∈ [0, ∞) 7→ U (t) into the unitary group of O2 such that U (0) = 1
and limt→∞ kU (t)∗ ι(a)U (t) − κ(a)k for all a ∈ A. One can use this to show that for
the (more general type of) embedding κ : A ,→ O2 there is a norm-continuous path
t ∈ [0, ∞) 7→ V (t) into the isometries in O2 such that limt→∞ kV (t)∗ κ(a)V (t) −
κ(a)k for all a ∈ A and limt→∞ dist(V (t)∗ bV (t), κ(A)) = 0 for all b ∈ O2 . (This
could replace the non-existing conditional expectations onto κ(A) by expectations
in an approximate sense.)

The examples below produce also counter-examples

to the above question ???,
but are also counter-examples to the following more interesting question (which
was suggested by the above recalled results and by the example of finite-
dimensional A):
We call a separable unital nuclear C *-algebra A transportable in O2 , if, for
any two unital *-monomorphisms i : A ,→ O2 , and j : A ,→ O2 in general position
there is an automorphism ψ of O2 with ψ ◦ i = j.
We say that a unital *-monomorphisms i : A ,→ O2 , is in general position if
the commutant i(A)0 ∩ O2 of the image i(A) in O2 contains a copy of O2 unitally.
Let i : O2 → O2 and j : O2 → O2 in general position. Is there an automorphism
of O2 ⊗ O2 that conjugates i ⊗ id and i ⊗ id?
Case: O2 ⊗ C[0, 1] → O2 unital and injective and O2 ⊗ O2 → O2 an isomor-
phism?
Is there an automorphism of O2 ⊗ O2 that maps 1 ⊗ i(A) onto 1 ⊗ j(A)?
Obviously all unital *-monomorphisms of finite-dimensional C *-algebras A are
unitary equivalent in O2 . It implies that each *-monomorphism of A is in general
position, and that A is “transportable” inside O2 .
We do not know if separable unital AF-algebras A are transportable in O2 .
Perhaps one could find a unital endomorphism ι : M2∞ ,→ M2∞ such that there
does not exist a conditional expectation from M2∞ onto ι(M2∞ ). (Compare the
below given reasoning.)
Then the ??????
 6. POSITIONS OF C *-SUBALGEBRAS IN O2 1221

The below given examples show that C[0, 1] is not transportable in O2 :

There exists nuclear A and unital embeddings i : A ,→ O2 and j : A ,→ O2
such that there is a conditional expectation from O2 onto i(A), but that there does
not exist a conditional expectation from O2 onto j(A) (see the examples discussed
below). It yields that for i0 (a) = i(a) ⊗ 1 and j 0 (a) = j(a) ⊗ 1 there does not exist
a *-isomorphism ψ of O2 ⊗ O2 (∼ = O2 ) with ψ ◦ i0 = j 0 (see the explanation below).
Let A a separable unital nuclear C *-algebra.
First:
If one uses the proof of the existence of embeddings j : A ,→ O2 with help of the
generalized Voiculescu-Weyl-vonNeumann theorem 5.4.1 (with O2 ⊗ K in place of
the compact operators K) and with help of Ext(A, O2 ) = 0, then one gets (as an
additional result) that the unital embedding j : A ,→ O2 can be found such that
there is a conditional expectation P from O2 onto j(A). (In fact the conditional
expectation can be chosen “extreme”, i.e., that P is an extreme point in the convex
set of contractive linear maps from O2 into O2 .)
Second:
If i : A ,→ O2 is an other unital embedding such that there does not exist a
conditional expectation from O2 onto i(A), then there does not exist an isomor-
phism ψ of O2 ⊗ O2 with ψ(j(a) ⊗ 1) = i(a) ⊗ 1 for a ∈ A, because otherwise
E(b) := id ⊗ f (ψ(P (b) ⊗ 1)) for b ∈ O2 and a (fixed pure) state f on O2 defines a
conditional E expectation from O2 onto i(A).
Third:
If A is a nuclear C *-subalgebra of a separable unital nuclear C *-algebra B such
that 1B ∈ A and such that there does not exist a conditional expectation from B
onto A (We list below some examples), then:
If k : B ,→ O2 is a unital embedding and if we define i := k|A as the restriction
of k to A, then there can not exist any conditional expectation E from O2 onto
i(A) = k(A), because otherwise k −1 ◦ E ◦ k would be a conditional expectation from
B onto A.

Some Examples:
(o) Suppose that D is a hereditary C *-subalgebra of a unital C *-algebra B such
that the normalizer algebra N (D) := {b ∈ B ; bD + Db ⊆ D} is different from
A := D + C1 and that bD 6= {0} for all b ∈ B+ \ {0}. Then there does not exist a
conditional expectation E from B onto A.
(Here C denotes the complex numbers.)
Indeed, (E(b) − b)∗ (E(b) − b)d = 0 for all b ∈ N (D), d ∈ D, i.e., E| N (D) = id.
It yields e.g. the following two examples (i) and (ii):
(i) Consider O2 as a C *-subalgebra of L(`2 ) (by some unital *-representation).
Let K denote the compact operators, and let A := K + C1, B := K + O2 .
(ii) Let B := T , A := K+C1 , where T denotes the Toeplitz algebra, generated
as C *-algebra by the Toeplitz operator T (i.e. the unilateral shift of `2 ).
1222 B. EXACT C*-ALGEBRAS AND EXAMPLES

(iii) Consider the natural continuous epimorphism σ from the Cantor space
Ω := {0, 1}∞ onto [0, 1] given by
∞
X
σ : (a1 , a2 , ...) 7→ an 2−n .
n=1

Then γ : f ∈ C[0, 1] 7→ γ(f ) := f ◦ σ ∈ C(Ω) gives a unital embedding of C[0, 1]

N∞
into C(Ω) = n=1 (C ⊕ C) ⊆ M2∞ .
The pairs A ⊆ B with A := γ(C[0, 1]) and B := C(Ω) or with B := M2∞ have
the property that there does not exist a conditional expectation from B onto A,
because the continuous map σ is not open.
(It is an exercise to check that σ is not open, and that this implies the non-
existence of the conditional expectation onto A .)
Some Remarks:
A related open question is:
Suppose that i, j : A ,→ O2 are unital embeddings such that there are extremal
conditional expectations from O2 onto i(A), respectively onto j(A). Is there an
automorphism ψ of O2 ⊗ O2 with ψ(i(a) ⊗ 1) = j(a) ⊗ 1 for all a ∈ A?
It is likely that there exists simple separable nuclear unital C *-algebras A ⊆ B
such that 1B ∈ A and such that there does not exist conditional expectation from
B onto A. (Perhaps, even with A ∼ =B∼ = M2∞ ?)
Then it would follow that there are unital endomorphisms i : O2 ,→ O2 such that
there does not exist a conditional expectation from O2 onto i(O2 ), because O2 ∼ =
A ⊗ O2 ⊆ B ⊗ O2 ∼ = O2 . (This would show that the position of i(O2 ) in O2 could
be very random.)
(It seems likely that O2 is not transportable in O2 ).

7. Quotients of nuclear maps

Example of a nuclear and approximately inner c.p. map that is not residually
nuclear.
Its relation to the existence of group vN-algebras that are not weakly exact.
Here we explain the reason for principal difficulties with nuclear maps (coming
from the existence of non-exact discrete groups), and we point out, that nuclear
and approximately inner (respectively nuclear and Ψ-equivariant) c.p. maps are in
general not residually nuclear if they are ideal-system preserving.
We introduce also some tools that allow to overcome this problems at least in
special cases.

Lemma B.7.1. Suppose that M is a W*-algebra with faithful normal state ρ

and let b1 , b2 , . . . ∈ M+ .
 7. QUOTIENTS OF NUCLEAR MAPS 1223

P
(i) If n ρ(bn ) ≤ 1, then for each ε > 0 there exists a projection p ∈ M such
that
k
X
ρ(1 − p) ≤ ε and k pbn pk ≤ ε−1 for all k ∈ N . (7.1)
n=1
P
(ii) Let d, c1 , c2 , . . . ∈ M contractions such that n ρ(bn ) ≤ 1 for the elements
bn := (d − cn )∗ (d − cn ) ∈ M+ .
If p ∈ M is a projection that satisfies the inequalities (7.1) of part (i)
for this b1 , b2 , . . . ∈ M+ then

lim f (pcn p) = f (pdp) for all f ∈ M ∗ .

n

Proof. (i): The functions gt (x) := x/(1 + t−1 x) = t(1 − 1/(1 + t−1 x)) on
[0, ∞) are operator monotone continuous functions, cf. [616, sec. 1.3.7]. They
satisfy gs (x) ≤ gt (x) ≤ x for s ≤ t, gt (x) ≤ t for all x ≥ 0, and |x − gt (x)| ≤ t−1 α2
Pk
for x ∈ [0, α]. Let ak := n=1 bn , then 0 ≤ gt (ak ) ≤ gt (ak+1 ), kgt (ak )k ≤ t,
more ?? check again?
It follows that the increasing sequence (gt (ak ))k of positive elements of M has
an ultra-weak limit (at first only an ultra-weak cluster point) A(t) ∈ M+ with
kA(t)k ≤ t and gt (ak ) ≤ A(t). Then dk := (A(t) − gt (ak ))1/2 ∈ M+ is a bounded
sequence with limk ψ(d∗k dk ) = 0 for all positive normal functionals ψ on M . It
follows that (dk ) converges ultra-strongly to zero, and implies that d2k also converges
ultra-strongly to zero, i.e., A(t) is uniquely determined and is the ultra-strong limit
of (gt (ak ))k in M .
Since gt (ak )gs (ak ) = gs (ak )gt (ak ), gs (ak ) ≤ gt (ak ) for s ≤ t and ρ(gt (ak )) ≤
ρ(ak ) ≤ 1, we get for the strong limits limk gt (ak ) = A(t) ∈ M+ that A(t)A(s) =
A(s)A(t), A(s) ≤ A(t) for s ≤ t and ρ(A(t)) ≤ 1 for all t ∈ [0, ∞). Let C denote
the commutative W*-subalgebra of M generated by 1 and A(m) (m ∈ N). Let
qm ∈ C the support projection of (A(m) − ε−1 )+ , i.e., qm = 1 − pm , where pm
is the support of the annihilator of (A(m) − ε−1 )+ in C. Then qm ≤ qm+1 and
ε−1 qm ≤ A(m). Thus, ρ(qm ) ≤ ε for all m ∈ N. It follows that p := 1 − m qm ∈ C
W

satisfies pA(m)p = A(m)p ≤ ε−1 p for all m ∈ N and ρ(1 − p) = supm ρ(qm ) ≤ ε.
We get pgm (ak )p ≤ pA(m)p ≤ ε−1 p for all k, m ∈ N. If we fix (arbitrary) k ∈ N
and let m tend to ∞, then this shows that pak p ≤ ε−1 p for all k ∈ N.
(ii): If f is a (not-necessarily normal) state on M , then |f (p(d − cn )p)|2 ≤
f (pbn p) . Thus, n |f (pdp) − f (pcn p)|2 ≤ ε−1 f (p) . In particular limn f (pcn p) =
P

f (pdp) . Finally use that M ∗ is the linear span of the states on M . 

Lemma B.7.2. Suppose that M is a W*-algebra, A is a separable C*-algebra

and C ⊆ CP(A, M ) a matrix operator-convex cone.
If V : A → M is in the point-ultraweak closure of C, then there is a net of
projections pµ such that pµ V (·)pµ is in the point-norm closure of C and the net
{pµ } converges strongly to 1M .
1224 B. EXACT C*-ALGEBRAS AND EXAMPLES

In particular, if e ∈ A+ satisfies V (e) = 0, then there is a net of contractions

Tµ ∈ C in the point-norm closure of C with Tµ (e) = 0 such that Tµ converges in
point-ultra-strong topology to V .

Proof. We can replace in the statement C ⊆ CP(A, M ) by its point-norm clo-

sure C, which is again an m.o.c. cone. Then C is a hereditary sub-cone of CP(A, M )
in the sense that W1 +W2 ∈ C implies W1 , W2 ∈ C, cf. Corollary 3.6.28. Let R denote
the set of b ∈ M such that b∗ V (·)b ∈ C. Since (b+c)∗ V (·)(b+c)+(b−c)∗ V (·)(b−c) =
2b∗ V (·)b+2c∗ V (·)c and C is hereditary, point-norm closed and invariant under com-
position with inner c.p. maps, it follows that R is a (norm-) closed right-ideal of M .
The hereditary C *-subalgebra D := R∗ ∩ R of M contains an approximate unit pσ
of D consisting of projections. Let R1 denote the weak closure and let E ∈ M the
projection with EM = R1 . Then pσ converges ultra-strongly to E, pσ V (·)pσ ∈ C,
and {pσ V (·)pσ }σ converges to EV (·)E in point-ultra-strong topology. In particular,
E = {p ; p ∈ R, p2 = p = p∗ } . It remains to show that E = 1M .
W

Since for every non-zero projection P ∈ M there is a non-zero countably decom-

posable projection Q ∈ M with Q ≤ P , it suffices to show that for every countably
decomposable projection Q there is a projection p ≤ Q with p ∈ R. If we replace
M , V and C by QM Q, QV (·)Q and QCQ := {QW ( · )Q ; W ∈ C}, then the proof
reduces to the case, where M has a faithful normal state ρ.
We introduce the norm kbkρ := ρ(b∗ b)1/2 on M . The topology defined by the
norm k · kρ coincides on bounded parts of M with the (ultra-)strong topology on M .
A separation argument shows that that V : A → M is moreover in the point-
*ultra-strong closure of of C if V is in the point-ultra-weak closure of C. We may
suppose that kV k ≤ 1, then Lemma 3.1.8 implies that V can be approximated in
point-strong topology by contractions W ∈ C .
Let C1 the convex set of W ∈ C with kW k ≤ 1, and let (an ) denote a sequence
that is dense in the unit ball of A. We find Wn ∈ C1 with kV (ak )−Wn (ak )kρ < 8−n
for k ≤ n. Let bn := k≤n (V (ak )−Wn (ak ))∗ (V (ak )−Wn (ak )) . Then ρ(bn ) ≤ 1.
P P

Thus, for each ε > 0, there is a projection p := p(ε) ∈ M with ρ(1 − p) ≤ ε and
P∞ −1
n=1 pbn p ≤ ε p , cf. Lemma B.7.1(i). It follows from Lemma B.7.1(ii) that
limn f (pWn (ak )p) = f (pV (ak )p) for every bounded linear functional f ∈ M ∗ on
M and every k. Since C1 ⊆ L(A, M ) consists of contractions and since {a1 , a2 , . . .}
is dense in the unit ball of A, the map pV (·)p is in the point-σ(M, M ∗ ) closure of
the convex subset p(C1 )p := {pW (·)p ; W ∈ C1 }. Thus pV (·)p is the point-norm
closure of p(C1 )p ⊆ C, i.e., pV (·)p ∈ C, p ∈ R. 

Remark B.7.3. A special case of Lemma B.7.2 is:

Suppose that M is a W*-algebra, A is separable, V : A → M is weakly nuclear and
that e ∈ A+ satisfies V (e) = 0. Then there is a net of (norm-)nuclear contractions
Tµ : A → M with Tµ (e) = 0 and Tµ → V in point-ultrastrong topology.
It is in general not possible to find factorable maps Tµ with this property. See the
following Remark B.7.4.
 7. QUOTIENTS OF NUCLEAR MAPS 1225

Remark B.7.4. Let C ⊆ M(D) and V : C → B with V (C ∩D) = {0} a nuclear

map. Then VD (c + d) := V (c) for c ∈ C and d ∈ D is a well-defined completely
positive map from the C *-algebra C + D into B, because (C + D)/D ∼
= C/(C ∩ D)
naturally.
When VD : C + D → B is again a nuclear map ? (It is an equivalent formula-
tion of the general quotient problem for nuclear maps in Remark 3.1.2(iv), as our
considerations below indicate.)
From [238] and [438] it follows that VD is nuclear if C is exact (see Remark
B.7.8). Clearly, VD is nuclear if [V ]C∩D : C/(C ∩ D) → B is nuclear.
For a long time it was expected that VD is always nuclear. But N. Ozawa [597]
has shown that the von Neumann algebra generated by the regular representation
of a non-exact discrete group is not weakly exact. M. Gromov [335] has shown the
existence of discrete non-exact groups. (And Guentner and Kaminker [338] and
Ozawa [592] discovered this connections between Gromov’s work to non-exactness.)
By Definition 3.1.1, a map T : C → N from a C *-algebra C to a von Neumann
algebra N is weakly nuclear if T is the point-wise weak limit of factorable maps
( 1 ), compare Definition 3.1.1.
By [597] there exists a von-Neumann algebra that is not weakly exact. That
means (equivalently) that there exists a von Neumann factor N of type II1 with
separable predual N∗ and a weakly nuclear maps T from a separable C *-algebra C
into N and a closed ideal J of C such that T (J) = {0} and that [T ]J : C/J → N
is not weakly nuclear.
It implies that the same happens with a (norm-)nuclear map V : C → N with
V (J) = {0} (in place of T ): An Egoroff type argument (see B.7.2) shows that
every weakly nuclear c.p. contraction T : C → N from a separable C *-algebra C
into a von Neumann algebra with separable predual N∗ is the point-strong limit
of a sequence of (norm-)nuclear c.p. contractions Vn : C → N with Vn (J) = {0}.
It follows that [Vn ]J : C/J → N is not weakly nuclear for almost all n ∈ N. In
particular, [V ]J : C/J → N is not nuclear for suitable n ∈ N and V := Vn .
Let I := Ann(J) denote the annihilator of J in C, let p ∈ C ∗∗ be the support
projection of J in the second conjugate C ∗∗ of C and consider E := C ∗ (C, p) ⊆ C ∗∗ ,
D := I + pE. Then E = C + pE = C + D is naturally a C *–subalgebra of M(D),
a ∈ C → (1p )a ∈ E defines a C *-morphism W : C/J → C + D = E such that
[V ]J = VD ◦ W . Thus VD : C + D → N can not be (weakly) nuclear.
In a similar way one finds a C *-subalgebra C1 ⊆ B := C ⊕ N ⊗L(`2 ) and
an inner c.p. contraction W : B → B such that (W |C1 ) : C1 → B is nuclear but
not residually nuclear (with respect to the action of Prim(B) on C1 ): Stinespring
dilation and application of an infinite repeat lead to a *-representation ρ of C into
the von Neumann algebra tensor product N ⊗L(`2 ) and a projection e ∈ L(`2 ) of

1It is also a point-τ (N, N )-limit of factorable maps, because the set of factorable maps is an
∗
operator convex cone.
1226 B. EXACT C*-ALGEBRAS AND EXAMPLES

rank one, such that V (a) ⊗ e = (1 ⊗ e)ρ(a)(1 ⊗ e) for all a ∈ C. Let ρ1 (a) :=
(a, ρ(a)) ∈ B, C1 := ρ1 (C) and W (b) := (0, 1 ⊗ e)b(0, 1 ⊗ e) for b ∈ B. Then W |C1
is nuclear and is inner in B, but is not residually nuclear on C1 ⊆ B.

We use the following Lemmata B.7.5 and B.7.7 and the Definitions B.7.6 to
bypass the above mentioned problems at least under certain good conditions.

Lemma B.7.5. Let V : A → M(B) ⊆ B ∗∗ a completely positive map from A

into the multiplier algebra M(B) of B. Then the following are equivalent:

(i) For every d ∈ B the completely positive map a ∈ A 7→ d∗ V (a)d ∈ B is

nuclear.
(ii) V is a weakly nuclear map from A to the W*-algebra B ∗∗ in sense of
Definition 3.1.1 if V is considered as a completely positive map from A to
the second conjugate B ∗∗ of B

Proof. One can see from the definition (of weakly nuclear maps), that the
weakly nuclear maps build a point-weakly closed m.o.c. cone ⊆ L(A, B ∗∗ ). Since
the unit-ball B is *-strongly dense in the unit-ball of B ∗∗ (by Kaplansky density
theorem), a map V : A → B ∗∗ is weakly nuclear, if and only if, the maps a 7→
d∗ V (a)d are weakly nuclear for every d ∈ B.
Old Prop. 3.1.9(i) has been changed. Compare again!!
No! Next is not Part of Proposition 3.1.9(i-iv) !!!
By Proposition 3.1.9(i), the map a ∈ A 7→ d∗ V (a)d ∈ B is nuclear, if and only
if, it is weakly nuclear as a map from A to B ∗∗ , cf. Proposition 3.1.9(i). 

Definition B.7.6. We call a completely positive map V : A → M(B) weakly

nuclear if V satisfies the equivalent conditions (i) and (ii) of Lemma B.7.5.
An example of a weakly nuclear completely positive map is given by B := K(H),
V : A → L(H) a faithful *-representation of A. It is nuclear if and only if A is exact,
cf. Remark 3.1.2(ii) and Corollary 5.6.3.
Let D ⊆ B be a hereditary C *-subalgebra of a C *-algebra B. D is a corner of
B, if there is a projection p ∈ M(B) such that D = pBp. Certainly the (two-sided)
annihilator {b ∈ B : bD = 0 = Db} of D is just the orthogonal corner (1−p)B(1−p)
of pBp. A corner is σ-unital if B is σ-unital, because the compression pbp of a
strictly positive element b of B is a strictly positive element of pBp. The unital, in
particular σ-unital, Calkin algebra Q(K) contains hereditary C *-subalgebras which
are not σ-unital.
We say that a *-subalgebra E ⊆ B generates a corner of B if the closure D
of EBE is a corner of B. The annihilator of E in B is then a corner of B, and is
therefore σ-unital if B is σ-unital.
A hereditary C *-subalgebra D ⊆ B is full , if it generates B as a closed ideal.
A theorem of L.G. Brown [107] says that B ⊗ K and D ⊗ K are isomorphic if B is
σ-unital and D is a σ-unital full hereditary C *-subalgebra of B. The isomorphism
 7. QUOTIENTS OF NUCLEAR MAPS 1227

is induced in the sense of Remark 2.3.1 by an element z ∈ B ⊗ K such that z ∗ z is

an strictly positive element of B ⊗ K and zz ∗ is a strictly positive element of D ⊗ K,
cf. Corollary 5.5.6.
A subalgebra D of a C *-algebra A will be called an essential subalgebra if
zero is the only left annihilators of D in A, i.e., if aD = {0} implies a = 0.
It is easy to see, that an ideal I of a C *-algebra A is essential if and only if it
has non-zero intersection with every non-zero ideal of A.
An extension E, i.e. an exact sequence 0 → J → E → A → 0 (defined by the
epimorphism η : E → A), is semisplit if there is a completely positive contraction
V : A → E with η ◦ V = idA .

Lemma B.7.7. Suppose C is C*–subalgebras of M(D), and that V : C → M(B)

is a weakly nuclear completely positive contraction. Then:

(i) Ve (c + z1) := V (c) + z1 is a unital weakly nuclear map from the (outer)
unitization Ce of C into M(B).
(ii) If C ∩ D generates a corner of D, then every weakly nuclear contraction
V : C → M(B) with V (C ∩ D) = {0} and with V (C ∩ (1 + D)) ⊆ {1}
extends naturally to a unital weakly nuclear map W : D +C +C1 → M(B)
with W (D) = 0 by W (d + c + ξ1) := V (c) + ξ1 for d ∈ D, c ∈ C 1 ∈ M(D)
and ξ ∈ C.

Proof. (i): For a contraction d ∈ C holds V (d∗ d) ≤ 1, b∗ V (d∗ cd)b + zb∗ b =

b V (d∗ (c + z)d)b + zb∗ (1 − V (d∗ d))b, and c + z1 7→ z 7→ zb∗ (1 − V (d∗ d))b is a
∗

nuclear c.p. map, because 1 is not in C if the unit element 1 is outer adjoined.
Therefore, for contractions d ∈ C, c + z1 7→ b∗ V (d∗ cd)b + zb∗ b is nuclear as a sum
of two nuclear maps. The map c + z1 7→ b∗ Ve (c + z1)b is the point-norm limit of
those maps.
(ii): Let C1 := D + C + C1 ⊆ M(D) .
If 1 ∈ C + D then C1 = D + C and C1 /D ∼ = C/(D ∩ C) . If 1 6∈ D + C
∼ ^
then C1 = D + C (respectively C1 /D = C/(D ∩ C)) is the outer unitization of
^
D + C (respectively of C/(D ∩ C) ). The condition V (C ∩ (1 + D)) ⊆ {1} says that
[V ]D∩C : C/(D ∩ C) ∼ = (D + C)/D → M(B) is unital if 1 ∈ C + D. We can define
in both cases a unital c.p. map W : C1 → M(B) by W := [V ] ◦ πD , respectively by
its extension to the outer unitization of C + D. Here [V ] := [V ]D∩C : (C + D)/D ∼
=
C/(D ∩ C) → M(B) is well-defined because V (D ∩ C) = {0}. Thus, W is a
well-defined, unital c.p. map with W (D) = {0}. In both cases holds
W (d + c + ξ1M(D) ) = W (c + ξ1) = V (c) + ξ1M(B) .
It suffices to prove that for every b ∈ B the c.p. map Wb : c+d 7→ b∗ V (c)b is nuclear,
because the outer unitization of W is then also weakly nuclear.
Let p ∈ M(D) the orthogonal projection such that pDp is the closure of (C ∩
D)D(C ∩ D). Then E := (1 − p)D(1 − p) is the (two-sided) annihilator {d ∈
D ; d(C ∩ D) = 0 = (C ∩ D)d} of C ∩ D. Therefore CE ⊆ E, EC ⊆ E and
1228 B. EXACT C*-ALGEBRAS AND EXAMPLES

pc(1 − p) = 0, i.e., pc = cp for c ∈ C. Thus F := E + C is a C *-algebra, E is an

ideal of F ⊆ M(D), F commutes with p and pF = pC. C ∩ E = C ∩ (D ∩ E) =
(C ∩ D) ∩ E = 0. Therefore T |F is the composition of F → F/E ∼ = C with
∗
c ∈ C → b V (c)b. Thus T |F is nuclear and T (F ∩ D) = 0.
Let X be any C *-algebra. By the criteria in Remark 3.1.2(i), the algebraic
tensor product idX (T |F ) : X F → X B extends to completely positive map
S1 from the minimal (=spatial) tensor product X ⊗ F into the maximal C *-algebra
tensor product X⊗max B. Since S1 (X (F ∩D)) = 0, we have that S1 (X⊗(F ∩D)) =
0. Since F commutes with p, F ∩ D = p(F ∩ D)p + (1 − p)(F ∩ D)(1 − p). But
E = (1 − p)(F ∩ D)(1 − p) and C ∩ D ⊆ p(F ∩ D)p ⊆ p(C ∩ D)p = C ∩ D. Thus
F ∩ D = (C ∩ D) + E . It follows that D is the closure of D(F ∩ D).
It is now the essential point of the proof that, therefore, the distance lemma
[438, lem. 3.9] applies and gives the identity (X ⊗ F ) ∩ (X ⊗ D) = X ⊗ (F ∩ D) in
the C *-algebra X ⊗ (C + D) for every C *-algebra X .
It follows (X ⊗ (C + D))/(X ⊗ D) ∼
= (X ⊗ F )/(X ⊗ (F ∩ D)), because X ⊗ D
is a closed ideal of F ⊗ (C + D).
Let S2 : (X ⊗ F )/(X ⊗ (F ∩ D)) → X ⊗max B be the completely positive map
that is induced by S1 . The composition of X ⊗ (C + D) → (X ⊗ F )/(X ⊗ (F ∩ D))
with the completely positive map S2 defines a completely positive map S3 from
X ⊗ (C + D) into X ⊗max B. The restriction of S3 to the algebraic tensor product
is just idX T .
Since S3 can be found for every C *-algebra X, T is nuclear by the criteria in
Remark 3.1.2(i). 

Remark B.7.8. To understand the essential point of the proof of Lemma

B.7.7(ii), the reader should note, that the intersection formula (F ⊗ C) ∩ (F ⊗ D) =
F ⊗(C ∩D) holds for every C *-algebra D and C ⊆ M(D), if and only if, F is exact.
This happens even if we consider here only separable C and require, in addition,
that D is a simple, purely infinite and separable C *-algebra.
The intersection formula it true for every pair of C *-algebras C, D ⊆ L(H), if
and only if, F satisfies the slice map property (S) of S. Wassermann. The Property
(S) implies exactness, but it is unknown if (S) is equivalent to exactness.
On the other hand, for every C *-algebra F , every locally reflexive C *-algebra
C ⊆ M(D), one can show that (F ⊗ C) ∩ (F ⊗ D) = F ⊗ (C ∩ D). (The ideal C ∩ D
of C is a locally reflexive C *-subalgebra of D.)
Thus our proof of Lemma B.7.7(ii) also shows that the natural extension VD
of V to C + D with VD (D) = 0 is weakly nuclear if C is locally reflexive in the
sense of [238] (without any assumptions about the behavior of C ∩ D in D). All
exact C *-algebras are locally reflexive, cf. [432, rem. on p. 71] (compare [802] for
the case of separable C *-algebras).
 8. QUESTIONS ABOUT PURE INFINITENESS 1229

8. Questions about pure infiniteness

compare and sort following questions !! Below, we list some open ques-
tions concerning the verification of pure infiniteness, and K1 -injectivity.

Question B.8.1. Denote by A ∗ B the unital universal free C *-algebra of

unital A and B, and define the free joint Efree of A and B by

Efree (A, B) := {f ∈ C([0, 1], A ∗ B) ; f (0) ∈ A ∗ 1 , f (1) ∈ 1 ∗ B } .

(1) Suppose that A is non-unital, separable and locally purely infinite.

Does there exist n ∈ N such that the unit of Mn (M(A)) is properly infinite?
(It is the case for weakly purely infinite A, or if M(A) itself is locally purely infinite.
in the sense of Definition 2.0.3)
(2) Suppose that A is unital and weakly p.i. (respectively locally p.i.) and that
1A/J is properly infinite for every non-zero closed ideal {0} =6 J / A.
Is 1A properly infinite if A is not prime ?
It leads to the –equivalent– question, whether or not n-purely infinite quotients of
the unital C *-subalgebra Efree (O∞ , O∞ ) of C([0, 1], O∞ ∗O∞ ) have properly infinite
unit elements. See the farer going question (8) below.
(3) Let D := {z ∈ C ; |z| ≤ 1} and S 1 = ∂D. Then C(D) is naturally
isomorphic to the (unital) algebra cone(C(S 1 )) and the corresponding epimorphism
from C(D) onto C(S 1 ) is given by ϕ(f ) := f |S 1 . Let f0 (z) := z and u := ϕ(f0 ).
The unitary u defines an inner automorphism σ(x) := u∗ xu of O∞ ∗ C(S 1 ). One
can show ( 2 ) that a positive answer to the following question implies a positive
answer to question (2):
Does the pull-back
(O∞ ∗ C (D)) ⊕π,σ◦π (O∞ ∗ C (D))
of the unital epimorphisms

π := id ∗ϕ : (O∞ ∗ C (D)) → O∞ ∗ C S 1

and of σ ◦ π have a properly infinite unit?

(Unfortunately, it is likely that (3) has a negative answer. And it seems that the
same happens for O2 – in place of O∞ .)
(4) Suppose that A is a purely infinite (non-simple) algebra, and a ∈ A+ .
Are the (unital!) fibers of the upper (!) semi-continuous C*–bundle

F (C ∗ (a), A) := ({a}0 ∩ Aω )/ Ann(a, Aω )

over Spec(a) purely infinite?

(The fibers are 2-p.i. by [443].)
More precise and actual citation?

2 by [448, prop.1.6,rem.1.15(2), prop.A.4], because two unital *-endomorphisms of O

∞ are
approximately unitarily equivalent by Theorem B
1230 B. EXACT C*-ALGEBRAS AND EXAMPLES

(5) Are the fibers Bt (of F (C ∗ (a), A) , t ∈ Spec(a) ) K1 -injective ?

(I.e., is U0 (Bt ) the kernel of the natural map U(Bt ) → K1 (Bt ) .)
(6) Is O∞ ∗ C(S 1 ) K1 -injective ?
With other words:
Does there exist v ∈ U0 (O∞ ∗ C(S 1 )) with ut1 = vt1 , where t1 , t2 , . . . (respec-
tively u) are the canonical generators of O∞ (respectively of C(S 1 ))?
If the answer is positive, then U(B)/U0 (B) ∼ = K1 (B). for every unital algebra B
that contains a a copy of O∞ unitally. (The answer could be negative. The question
is possibly related to question (3). See also Proposition 4.3.6(iv).)
(7) Is O∞ ∗ O∞ K1 -injective?
Note that (6) and (7) are related, because unital copies of O∞ in a unital C *–
algebra A are unitarily homotopic by a continuous path u(t) ∈ U(A), and one can
manage that [u(t)]1 = 0 in K1 (A).
(Since hi : O∞ → A, i = 1, 2, are unital, it follows from Theorem B – and from the
UCT for O∞ – that there is a continuous path u(t) in the unitaries of A such that
limt→∞ ku(t)∗ h1 (b)u(t) − h2 (b)k = 0.)

Perhaps, O∞ ∗ O∞ and O∞ ∗ C(S 1 ) are KK-equivalent.

At least, there are unital *-morphisms α : O∞ ∗ O∞ → O∞ ∗ C(S 1 ) (α(a ∗ 1) :=
a ∗ 1 and α(1 ∗ a) := (1 ∗ U )∗ (a ∗ 1)(1 ∗ U )) and β : O∞ ∗ C(S 1 ) → Cb (R+ , O∞ ∗ O∞ )
(β(a ∗ 1) = a∗ 1, β(U ) := u) such that π ◦ β◦ is the natural – constant – embedding
from O∞ ∗ O∞ into Q(R+ , O∞ ∗ O∞ ).)
(8) Does Efree (O∞ , O∞ ) contain a non-trivial projection?
Is the unit of Efree (O∞ , O∞ ) properly infinite?
(9) Let J a separable weakly purely infinite C *–algebra such that there
is a sequence of positive contractions e1 , e2 , . . . ∈ J+ and contractions vn,k ,
such that en vm,k = vm,k , vn,k em = vm,k , en em = em , for m < n, k, l ∈ N,
∗
vm,k vn,l = δk,l δm,n e2n for k, l, m, n ∈ N, moreover, we suppose that (en ) is an
approximate unit of J.
Is the unit 1M(J) of M(J) infinite?

(10) Suppose, in addition to the assumptions of (9), that J of (9) is an ideal of

a unital C *-algebra A and that there is a *-monomorphism ϕ : C0 (0, 1] ⊗ O∞ ,→ A
with ϕ(f0 ⊗ 1) = n 2−n en .
P

Is the unit of A properly infinite?

A positive answer would imply that weakly purely infinite algebras are purely
infinite, and that purely infinite algebras A satisfy the assumption of the question
(12).
(11) Suppose that a1 , a2 ∈ A+ are stable elements of A (i.e., Dk := ak Aak is
stable for k = 1, 2 ).
 8. QUESTIONS ABOUT PURE INFINITENESS 1231

Is a1 + a2 a properly infinite element of A?

A positive answer would imply that locally purely infinite algebras are purely
infinite, and that every purely infinite algebra A satisfies the assumption of the
question (12).
(But it is even unknown whether or not sums of σ-unital stable ideals have a
properly infinite strictly positive element.
What happens in the case where D1 and D2 are ideals of A and where A is
moreover locally purely infinite? )
(12) Suppose that F (C, A) := (C 0 ∩ Aω )/ Ann(C, Aω ) has a properly infinite
unit for every separable commutative C *-subalgebra C of A.
Is A strongly purely infinite?
(13) Suppose that A is p.i. Is C([0, 1], A) p.i.?. (It is 2-p.i.)
Special case:
Let A a purely infinite separable nuclear C *-algebra. Is C([0, 1], A) purely
infinite?.
For simple A the answer is positive, because simple p.i. algebras are strongly p.i.
(14) Are local p.i. C*-algebras A with the global Glimm halving property purely
infinite?
(A is p.i. if and only if A is l.p.i. and the sum D = (D1 + D2 )A(D1 + D2 ) of any
two stable hereditary C *-subalgebras D1 and D2 of A contains a family (Dτ )τ of
stable hereditary C *-subalgebras of A such that each element of the Pedersen ideal
S
of D can be approximated by elements from τ A(Dτ )A – only elements of the
union, not sums of them! –.)

The next question asks what happens with pure infiniteness if we have a con-
ditional expectation of finite index onto a subalgebra.

Question B.8.2. Suppose that A ⊆ B (both not necessarily simple) and that
P : B → A is a conditional expectation from B onto A of finite index, i.e., P |A =
idA , kP k = 1, and there exists 0 < λ < ∞ with λ−1 b ≤ P ⊗ idn (b) for all b ∈
(B ⊗ Mn )+ and n ∈ N. (Note that the existence of P implies that A is unital if and
only if B is unital, and that then P must be necessarily unital. More generally, A
is a non-degenerate C *-subalgebra of B in the non-unital case. This can be seen
by passage to the second conjugate P ∗∗ .)
What about of the following statements (1)–(6) concerning versions of pure
infiniteness?
(1) Every l.s.c. 2-quasi-trace of A is trivial, if and only if, every l.s.c. 2-quasi-
trace of B is trivial.
(2) A is locally p.i., if and only if, B is locally p.i.
(3) A is p.i. (respectively is weakly p.i., strongly p.i.), if and only if, B is p.i.
(respectively is weakly p.i., strongly p.i.).
1232 B. EXACT C*-ALGEBRAS AND EXAMPLES

(4) If B is unital, then 1A is stably properly infinite (respectively is prop-

erly infinite), if and only if, 1B is stably properly infinite (respectively is properly
infinite).
(5) There are a *-monomorphism h : B → A ⊗ K and a conditional expectation
0
P : D → h(B) of finite index from the hereditary C *-algebra D of A ⊗ K generated
by h(B) onto h(B).
(6) If ϕ : A → F is a *-epimorphism onto a finite AW*-factor F , then there
exists a finite AW*-algebra G with finite-dimensional centre and a *-epimorphism
ψ : B → G, a unital *-morphism  : F → G and a conditional expectation E : G →
(F ) of finite index, such that  ◦ ϕ = ψ|A and ψ ◦ P = E ◦ ψ.
(Alternatively, and almost equivalent: If the multiplier algebra M(X) ∩ X of a
not-necessarily unital C *-system X contains a finite AW*-factor F ⊆ M(X) ∩ X
and if there is a conditional expectation from X onto F of finite index, then X =
M(X) and X is a finite AW*-algebra with finite centre.)
(7) For every b ∈ B+ there are n ∈ N, a projection q ∈ Mn and a ∈ (A ⊗ Mn )+
such that b⊗q is Cuntz-equivalent to a in B ⊗Mn . (At least, if J = (A∩J)B(A∩J)
for each closed ideal J of B, and I = A ∩ span(BIB) for each closed ideal of I of
A.)

Questions B.8.3. Are inductive limits A of pi(n) algebras A1 , A2 , . . . again

pi(n)?
Q
Can the ultra-product ω Ak of pi(n) algebras Ak (k = 1, 2, . . . ) have a non-
zero finite-dimensional quotient? (Equivalently: Is the c0 -sum of pi(n) C*-algebras
L
k Ak a pi(n) algebra? )
Is there a general function f : N → N such that A ⊗ K is pi(f (n)) if A is pi(n) ?

The implication
{ A is pi(n) } ⇒ { A has property pi-m for some m ≥ n }
(m depending on A) was shown by an indirect argument (showing that Aω is l.p.i.
and, therefore, Aω is traceless). Our there given argument does not indicate a way
to find a uniform bound for m (if n ∈ N is given).
Fix n ∈ N. Suppose that A is σ-unital, has no non-zero finite-dimensional
quotient A/J of dimension ≤ n, and that for each a ∈ A+ , positive b ∈ span(AaA)
and ε > 0 there are d1 , . . . , dn ∈ A such that kb − j d∗j adj k < ε.
P

Further let e ∈ A a strictly positive element of A. Then each hereditary C*-

subalgebra Ak := (e − 1/k)+ A(e − 1/k)+ is pi(n), by Lemma 2.12.4(iii).
Hence, positive answers to the above (equivalent) questions, would imply, that
one can replace the condition (ii) in Definition 2.0.4 by the (possibly weaker) con-
dition that A has no quotient A/J of dimension ≤ n2 .

Question B.8.4. Is the unit 1F of a unital weakly purely infinite C*-algebra

F properly infinite?
 8. QUESTIONS ABOUT PURE INFINITENESS 1233

By definition, F is pi(n) for some n ∈ N. We know at least that 1F ⊗ 1m is

properly infinite for some m ∈ N, because pi(n) implies that F is l.p.i.
Proof of Corollary 2.7.18
Proposition ??, that there exists m ≥ n such that a ⊗ 1m is properly infinite
for every element 0 6= a ∈ A. In particular 1A ⊗ 1m is properly infinite.
A positive answer would imply that every weakly p.i. algebra is purely infinite.
Moreover, this would imply that each weakly p.i. algebra is commutant-p.i. in the
sense of Definition 2.16.6.
Question B.8.5. Is every commutant-p.i. algebra A (cf. Def. 2.16.6), strongly
purely infinite?
Question B.8.6. Suppose that Prim(A) is isomorphic to the Hilbert cube
[0, 1]∞ and that each fiber (= each simple quotient) is isomorphic to O2 .
Is A purely infinite?
(The question considers a special case of [462, ques. 4.8].)
Is A purely infinite if A is weakly purely infinite?
Question B.8.7. l.p.i. implies w.p.i.?
w.p.i. implies p.i.?
p.i implies l.c.p.i.?
l.c.p.i. implies c.p.i.?
c.p.i. implies s.p.i.?
(Is special case of [462, ques. 4.8].)
Merge below blue questions into above questions
??????
Question B.8.8. Is the unit of M(aAa) p.i. if a ∈ A+ is properly infinite in
A? ([462, ques.3.10])
A special open question is: Is the unit of M(A) properly infinite if A is a
separable purely infinite algebra?
Question B.8.9. a, b ∈ A+ commuting and (a − t)+ and (b − t)+ properly
infinite or zero for all t > 0. Is a + b properly infinite?
(It is a special case of [462, ques.3.10].)
Question B.8.10. Let A a separable unital C *-algebra and that J and K are
closed ideals of A with J ∩ K = {0}. Suppose that the units 1A/J and 1A/K are
properly infinite in A/J and A/K. It is easy to see that 1A ⊗ 12 is properly infinite
in M2 (A).
Is 1A properly infinite in A ?
If the unitary group of A/(J + K) is connected, then the answer is positive.
(Notice that A = A/(J ∩K) is just the pull-back A/J ⊕µ,ν A/K of the epimorphisms
µ : A/J → A/(J + K) and ν : A/K → A/(J + K).)
1234 B. EXACT C*-ALGEBRAS AND EXAMPLES

The question considers a test case of the question whether the proper infinite-
ness of a projection p ∈ A can be verified by looking only to p + J in all prime
quotients A/J.

Question B.8.11. Suppose that A is separable and is w.p.i. and a, b ∈ A+ are

stable contractions such that a + b is strictly positive in A.
Does there exist a stable element c = c(γ) ∈ A+ such that (a + b − γ)+ is in
the closed ideal of A generated by c ?
The question is equivalent to the question, if every w.p.i. algebra is p.i. (And
then, automatically, every p.i. algebra is c.p.i.)

Question B.8.12. Is M(A) weakly purely infinite if A is weakly purely infinite

and σ-unital?

Question B.8.13. Is the unit of M(A) properly infinite if A is purely infinite

and σ-unital?

9. On the properties “IR” and “stable rank one”

It’s all related to the almost ridiculous Part (xvii) of Proposition 2.2.1.
ALL HAS TO BE SORTED!!
DESIRE OF Remark 2.2.2:
See Section 9 in Appendix B for the definitions of “Property IR” and “stable
rank one”, and for an explanation, why Part (xvii) of Proposition 2.2.1 is one of the
needed observations for the proof of the Brown-Friis-Rørdam alternative for simple
C *-algebras with Property IR between pure infiniteness and “stable rank one”.
BEGIN OF LIST OF MATERIAL

Definition B.9.1. A C *-algebra B has stable rank one if each element

of B + C · 1M(B) is in the operator-norm closure of the invertible elements in
C ∗ (B, 1M(B) ) = B + C · 1M(B) ⊆ M(B).

Obviously, C *-algebras B that contain an infinite projection do not have this

is never the case for any simple purely infinite C *-algebras.)
A is simple, and has the property that each element T of A+C·1 ⊆ M(A) that
is not in the norm-closure of invertible elements of A + C · 1 must be right-invertible
or left-invertible in A+C·1, – which would be the correct interpretation of Property
IR in the special case of simple C *-algebras A – then this property holds also for
(A ⊕ +C · 1M(A) ) ⊕ C and for any non-zero hereditary C *-subalgebra D ⊆ A, i.e.,
also for D + C · 1 ⊆ A + C · 1 holds that any element that can not be approximated
by invertible elements in D + C · 1 is left- or right-invertible (but is not invertible).
This passage to hereditary C *-subalgebras is a key point and is equivalent to
the Morita equivalence of property IR on the class of simple C *-algebras.
 9. ON THE PROPERTIES “IR” AND “STABLE RANK ONE” 1235

It seems that it suffices to consider: passage to M2 (A), to inductive limits, and

to (non-zero) corners of (simple) C *-algebras, to get Morita equivalence of property
IR.
The same happens with property “real rank zero”.
as the property that there exists a hereditary C *-subalgebra
there exist elements in Ae := A + C · 1 ⊆ M(A) that are not in the closure of
of the invertible elements ... ,if and only if, A contains an infinite projection. Thus
????
Each element in A + C · 1 ⊆ M(A) that is not in the closure of the set of
invertible elements in A + C · 1 has a left or a right inverse (but not both of them)
and A + C has not stable rank one, i.e., the invertible elements are not dense in
A + C. (Or is Mn (A + C) to consider also? And its stabilizations ...? Is it a stable
problem?)
This gives immediately that A + C contains a non-unitary isometry ... Delivers
a projection in A. Then to every non-zero hereditary C *-subalgebra of A...?
It carries over to D + C for all non-zero hereditary C *-subalgebras D of A (by
stabilization ?)
ABOVE TEXT VERY RANDOM!
COMPARE WITH BELOW
GIVE ONLY SHORT SUMMARIES ...
The C *-algebras with stable rank one are – in a sense – opposites to (weak)
purely infinite C *-algebras, ... at least in case of simple C *-algebras.
Definition B.9.2. A C *-algebra B has “stable rank one” (denoted by “SR1 ”)
if each element of B +C·1M(B) ⊆ M(B) can be approximated in norm by invertible
elements in B + C · 1 .

It is easy to see from the definition that the class of C *-algebras with stable
rank one are invariant under following operations:

( i) forming direct sums B1 ⊕ B2 ,

( ii) passage to quotients B/J,
(iii) forming of inductive limits, e.g. B has SR1 if there are C *-subalgebras
S
B1 ⊂ B2 ⊂ · · · ⊂ B of C *-subalgebras Bn with SR1 such that n Bn is
dense in B,
(iv) pBp has SR1 if B has SR1 and p∗ = p ∈ M(B) is a projection.
(v) M2 (B) has SR1 if B has SR1. (use a modification of the classical Gauss
algorithm here in an approximate manner). It follows that Mn (B) has
SR1 (in conjunction with Part (iv)).
(vi) B ⊗ K has SR1 if B has SR1.
(vii) B has SR1 and separable and A is a separable C *-algebra that is Morita
equivalent to B then A has SR1.

WHAT about hereditary subalgebras/ideals?

1236 B. EXACT C*-ALGEBRAS AND EXAMPLES

The following Lemma B.9.3 is due to L.G. Brown [109].

Lemma B.9.3. Let D a hereditary C*-subalgebra of B and d ∈ D+C·1 ⊆ M(B).

Suppose that there exists a sequence (bn ) of elements in bn ∈ B + C · 1 that are
invertible in B + C · 1 and satisfy limn kbn − dk = 0. Then there exists a sequence
of cn ∈ D + C · 1 that are invertible in D + C · 1 and satisfy limn kcn − dk = 0 .

It allows to show that (non-zero) simple C *-algebras A are purely infinite if A

has not SR1 but satisfies property IR.
Next is my Definition!! Better name needed?

Definition B.9.4. We say (for a moment and only here) that an element
x ∈ B + C · 1M(B) ⊆ M(B) is “well-behaved ” if, for every closed ideal J ⊆ B of B,
the element πJ (x) is invertible in (B + C)/J if πJ (x) is left or right invertible in
(B + C)/J.

If B is simple, then this says that x ∈ B + C · 1M(B) is “well-behaved” if x is

invertible if it is left or right invertible.
If, for example, πJ (x) is neither left- nor right-invertible for all closed ideals J
of B, then x is well-behaved
Notice that P. Friis and M. Rørdam have 1996 in [305] the following property IR
that sounds a bit like a weakening of the property SR1 ???, but is rather different.

Definition B.9.5. A C *-algebra B has the property IR if each element b ∈

B + C · 1 ⊆ M(B), that is “well-behaved” in sense of Definition B.9.4, is in the
norm-closure of the invertible elements in B + C · 1 .

Thus property IR causes:

If b ∈ B + C · 1 is not in the norm-closure of the invertible elements in B + C · 1
and B has property IR, then there exists an closed ideal J of B such that πJ (b) is
left- or right-invertible but is not invertible in (B + C · 1)/J .
The point is now the following: Is it the Proposition 2.2 ???
If b = c + λ1M(B) (with c ∈ B) can be approximated by invertible elements
in B + C · 1M(B) and D is a hereditary C *-subalgebra of B with c ∈ D, then
c + λ1M(D) can be approximated by invertible elements of c + λ1M(D)
(Lemma of Brown, which one?).

Theorem B.9.6. Simple C*-algebras with property IR have stable rank one or
are purely infinite.

For the proof see [305] and [109].

Can proceed as follows:

If A is simple and has property IR and A has not stable rank one.
 9. ON THE PROPERTIES “IR” AND “STABLE RANK ONE” 1237

Then every non-zero hereditary C *-subalgebra D of A has not stable rank one,
because “stable rank one” is invariant under Morita equivalence.
So, if A has not stable rank one, then D can not have stable rank one.
If D has not stable rank one then there exists d ∈ D such that d or d + 1M(A)
can not be approximated by invertible elements in D + C · 1.
The point is now, that one must show that d (or d+1) can not be approximated
by invertible elements in A+C·1 if d (or d+1) can not be approximated by invertible
elements in D + C · 1 ⊆ M(A).
– This is because IR is not proven to be Morita equivalent so far –
Possible way:
1.) Passage to M2 (A), 2.) Passage to corner, 3.) to inductive limits, ...
If there is no passage to σ-unital hereditary C *-subalgebras, this does not help
...
How to check if IR passes to hereditary C *-subalgebras D: All closed ideals of
D are intersections J ∩ D with closed ideals J of A.
Let d ∈ D ⊆ A well-behaved in D + C · 1M(D) , is it then also well-behaved in
A + C · 1M(D) ?
And conversely?
Suppose that πJ (d) + z1 is approx-invertible in A + C · 1, the also approx-
invertible in D + C · 1 ? And conversely?
(Is it not so, that there exists ρ > 0 such that d − z1M(A) is not invertible for
every z ∈ C with |z| < ρ?
Otherwise, we find a zero-sequence zn ∈ C with d − zn invertible in for every
n ∈ N and |zn | → 0. Then d is in the closure if the invertible elements in D + C · 1.
Multiplying d or d + 1 with a constant we may suppose that d − z1 is not
invertible for all z ∈ C with |z| ≤ 1.
That implies that, e.g. , d can not be approximated by invertible elements in
A + C1 because ...
If T is invertible, then (T − z) is invertible for all z ∈ C with |z| < kT −1 k ??
and kT − dk < γ, implies Spec(T ) ⊆ (γ ◦ D2 ) + Spec(d),
k1 − T −1 dk ≤ kT −1 k · kT − dk
?????
Now in the class of C *-algebras ?????
(Other idea:
Try to show directly that D + C1 has also property IR, because then we get directly
that d (or d + 1) are right or left invertible. It is a sort of invariance of property IR
under Morita equivalence.)
1238 B. EXACT C*-ALGEBRAS AND EXAMPLES

(But a Lemma/Preposition of L.G. Brown gives this!).

L.G. Brown:

Is it the proposed Proposition 2.2 ???

The intersection of D + C · 1M(A) with the operator-norm closure in M(A) of
the invertible elements in A + C · 1M(A) is equal to the operator-norm closure of
the invertible elements in D + C · 1 in M(A).
All ???? inside ????? in M(A)?
If ????
If A has property IR, this implies that d (or d + 1) is right-invertible or left-
invertible in A or A + C · 1M(A) .
If d + 1 is not invertible but is right-invertible in A + C1, it implies that there
exists right-inverse R = e+z1 ∈ A+C·1cM for d+1M shows that (e+z1)(d+1) = 1
(respectively (e + z1)d = 1).
Thus, ke + z1k2 (1 + d)∗ (d + 1) ≥ 1 and d + 1 (respective d) is right invertible
in D + C1 by [(1 + d)∗ (d + 1)]−1 (1 + d)∗ . Thus, D + C · 1 contains a non-unitary
isometry ...
Perhaps there exists a proof that works without Lemma of Brown?
The “opposite” direction is “Corollary 1.11”:
Any purely infinite simple C *-algebra satisfies property IR.
Here is a collection of excerpts:
Material for Part (xvii) of Proposition 2.2.1!!!
ON BROWN CRITERIUM:
M. Rieffel: “Stable rank one”:
GL(A)
e is dense in A.e

For a unital C *-algebra A, P. Friis and M. Rørdam [305] defined R(A) as

follows:
The element a is in R(A) if and only if there does not exist a (closed two-sided) ideal
I such that πI (a) is one-sided invertible but not invertible, where πI : A → A/I is
the quotient map.
They then said that A satisfies IR if R(A) is in the (norm) closure of GL(A).
Of course it is obvious that GL(A) ⊆ R(A), since no element that is one-sided
invertible but not invertible (in some quotient algebra of A) can be approximated
by invertible elements of A.
For non-unital A, they said A satisfies IR if A
e does. Here Ae = A if A is unital and
∗
A = C (A, 1) ⊆ M(A) if A is not unital.
e

For expository purposes we introduce a formally weaker property, but in Propo-

sition 2.2 we will show it is equivalent.
An arbitrary C *-algebra A satisfies IR0 if every element of R(A+ ) ∩ (1 + A) is in
 9. ON THE PROPERTIES “IR” AND “STABLE RANK ONE” 1239

the closure of GL(A+ ) .

Obviously IR0 is equivalent to IR in the unital case, but for non-unital A we are
e ∩ A is not contained in the closure
leaving open (for now) the possibility that R(A)
+
of GL(A ) .
Lemma 1.2. (cf. [4, Theorem 3.5]).
Let B be a proper hereditary C *-subalgebra of a unital C *-algebra A, and identify
B + with B + C1. For t in 1 + B, if t ∈ GL(A), then t ∈ GL(B + ).
Proposition 1.9.
If A is the direct limit of an upward directed family {Bi } of hereditary C *-
subalgebras, and if each Bi satisfies IR, then A satisfies IR.
Corollary 1.10.
If A has an approximate identity of projections, in particular if A is of real rank
zero, then A satisfies IR if and only if pAp satisfies IR for each projection p in A.
Corollary 1.11.
Any purely infinite simple C *-algebra satisfies property IR.
Theorem 2.5.
Let I be a closed two-sided ideal of a C *-algebra A. Then A has IR if and only if
both I and A/I have IR and invertibles lift from A/I
e to A.e

Theorem 3.2.
Let A be a C *-algebra of real rank zero. Then A has IR, if and only if, whenever
p and q are projections in A generating the same ideal, then (1 − p) ∼ (1 − q) in
A+ implies p ∼ q.
Proposition 4.1.
If A is a simple C *-algebra, then A has IR if and only if either A has stable rank
one or A is purely infinite.
My comment: “stable rank one” is in case of simple C *-algebras what ? ...
It was pointed out in [305] that stable rank one implies IR
(“Stable rank in this sense was introduced by Rieffel [9] and stable rank one
means that GL(A)
e is dense in A.”)
e

(Thus, in case of simple A, excluding “stable rank one” means that there exist
elements in A
e that can not be approximate by invertible elements.)

Rieffel showed in [9] that C *-algebras of stable rank one satisfy a stronger
cancellation property: If p and q have the same image in K0 (A), then p ∼ q. Of
course p ∼ q implies that p and q generate the same ideal I and have the same
image in K0(I). We show next that stable IR implies a
?????
Proof.
1240 B. EXACT C*-ALGEBRAS AND EXAMPLES

For the direction not already proved, assume that A has IR and is not of stable
rank one. We need to show that every non-zero hereditary C *-subalgebra B of A
contains a non-zero projection and that every non-zero projection is infinite.
For the first, note that B cannot have stable rank one, since B is strongly
Morita equivalent to A and the stable rank one property is preserved by strong
Morita equivalence. We may assume B is not unital, and thus B e can be identified
with B + C · 1 ⊆ A. Then there must be t ∈ 1 + B such that t 6∈ R(B)
e e (since B
has IR by Proposition 1.4). Since the only relevant quotient of B e is B
e itself, t is
one-sided invertible but not invertible, and B contains a proper isometry u. Then
1 − uu∗ is the desired non-zero projection in B. For the second, let p be a non-zero
projection in A, and let B = pAp. The same sort of reasoning as above shows that
there is a proper isometry in B, whence p is infinite. 
Decide what is really needed from below!!
Some More Text from L.G. Brown, to support the relation of this this Di-
chotomy Property with the rather simple property in Part (xvii) of Proposition
2.2.1:
Text comes mainly from [109]:
For a non-unital C *-algebra A, A
e denotes the result of adjoining an identity,
+ e if A is non-unital and A+ := A ⊕ C if A is
and Ae if A is unital. Let A denote A
unital.
DEFINITION OF R(A) and IR:
For a unital C *-algebra A, Friis and Rørdam [305] defined R(A) as follows:

The element a is in R(A), if and only if, there does not exist a (closed two-
sided) ideal J such that πJ (a) is one-sided invertible but not invertible, where
πJ : A → A/J is the quotient map. They then said that A satisfies IR if R(A) is
in the (norm) closure of GL(A).
Of course it is obvious that GL(A) ⊆ R(A), since no element that is one-sided
invertible but is not invertible (in some quotient algebra of A) can be approximated
by invertible elements of A.
For non-unital A, they said A satisfies IR if A
e does.

We introduce a formally weaker property, but in Proposition 2.2 we will show

it is equivalent.
An arbitrary C*-algebra A satisfies IR0 if every element of R(A+ ) ∩ (1 + A) is
in the closure of GL(A+ ).
Obviously IR0 is equivalent to IR in the unital case, but for non-unital A we
e ∩ A is not contained in the
are leaving open (for now) the possibility that R(A)
closure of GL(A).
e

Lemma [305, lem. 4.3] shows that direct products (also known as `∞ -direct
sums) of C *-algebras with IR have IR, and the proof contains the assertion that IR
 9. ON THE PROPERTIES “IR” AND “STABLE RANK ONE” 1241

passes to ideals (see Remark 1.5 below for more on this). We proceed to generalize
the latter result. Our proof relies on Proposition 2.2, but the result is included in
this section for expository purposes.
Lemma 1.2.
([114, thm. 3.5]).
Let B be a proper hereditary C *-subalgebra of a unital C *-algebra A, and identify
B + with B + C · 1, where 1 means 1A . For all t in 1 + B holds: If t ∈ GL(A), then
t ∈ GL(B + ).
Proposition 1.4.
Any hereditary C *-subalgebra of a C *-algebra with IR also has IR.
(HERE A CRITICAL REMARK TO [305, lem. 4.3])

Remark 1.5.
The argument given in the proof of [305, lem. 4.3] for the fact that any ideal I in
a C *-algebra with IR also has IR actually shows only that I has IR0 . Of course
this is remedied by our Proposition 2.2, but it is not hard to see, without using any
results from the present paper, that [305, lem. 4.3] is correct as stated.
Proposition 1.6.
Let B be a hereditary C *-subalgebra of a unital C *-algebra A. Then (1 + B) ∩
R(A) ⊆ R(B + C · 1).
DEFINITION OF R(A) is where ???
Example 1.7.
The last result is not true for elements of B; i.e., B ∩ R(A) need not be contained
in R(B + C · 1).
To see this let A0 be a non-unital purely infinite simple C *-algebra, let A := A
f0 ,
and let B := pAp for a non-zero projection p in A0 . If u is a proper isometry in
B, then u 6∈ R(B + C · 1), since B + C · 1 ∼ = B ⊕ C, but u ∈ R(A), since the only
relevant quotient of A is A itself.
Lemma 1.8.
If A is the direct limit of a directed family {Bi } of hereditary C *-subalgebras, and
if each Bi satisfies IR0 , then A satisfies IR0 .
Proposition 1.9.
If A is the direct limit of
an upward directed

family {Bi } of hereditary C *-subalgebras, and if each Bi satisfies IR, then A

satisfies IR.
What is correct??
1242 B. EXACT C*-ALGEBRAS AND EXAMPLES

If A is the direct limit of

a directed family {Bi } of hereditary C *-subalgebras, and if each Bi satisfies IR0 ,
then A satisfies IR0 .

10. Uniform global Glimm halving

Remark B.10.1. Let Aω := `∞ (A)/Jω , and B ⊆ Aω a C *-subalgebra. We

define
F (B, A) := (B 0 ∩ Aω )/ Ann(B, Aω )
and the special case F (A) := F (A, A).
One can formulate “uniform variations” of the global Glimm halving property
for A, respectively of the non-existence of characters on A:

(i) There is (universal) n ∈ N such that, for every a ∈ A+ , c ∈ A+ with

kck ≤ 1, ca = a, and ε > 0, there exists b, d1 , . . . , dn ∈ A, b ∈ aAa, b2 = 0,
P ∗ ∗
kbk = 1, k[b, a]k < ε, k[dj , a]k < ε, dj bb dj = c.
(ii) For all a∗ = a ∈ Aω , F (C ∗ (a), A) has the global Glimm halving property
(Notice here that F (C ∗ (a+ ), A) ⊕ F (C ∗ (a− ), A) ∼ = F (C ∗ (a), A).)
(iii) Aω has the global Glimm halving property, i.e., for all a ∈ (Aω )+ , ε > 0,
there exist bε,a ∈ a(Aω )+ a with b2ε,a = 0, (a − ε)+ ∈ I(bε,a ) (i.e., for
every δ ∈ (0, ε), there exist n = n(δ) and d1 , . . . , dn ∈ Aω such that
P ∗ ∗
j dj (bε,a bε,a )dj = (a − (ε + δ))+ ).
(iv) There is (universal) n ∈ N such that for a, c ∈ A+ with kck = 1 and
ca = a there exists b ∈ A with b2 = 0, cb = bc = b and d1 , . . . , dn ∈ A
P ∗ ∗
with dj (bb + b∗ b)dj = a.
(v) No hereditary C *-subalgebra of Aω has a character 6= 0. (It suffices to
consider σ-unital hereditary subalgebras.)
(vi) There exists (universal) n ∈ N, such that for all a, c ∈ A+ , kck = 1, ca = a
there are b1 , . . . , bn , d1 , . . . , dn ∈ A with b2j = 0, kbj k = 1, cbj = bj c = bj ,
P ∗ ∗
cdj = dj c = dj , and dj (bj bj + bj b∗j )dj = a.

Then (i)⇔(ii), (iii)⇔(iv), and (v)⇔(vi). Moreover (iv)⇒(vi).

11. Examples that exhaust the UCT-class

The following Lemma B.11.1 shows that every Z2 -graded countable Abelian
group G = G0 ⊕ G1 is isomorphic to K∗ (C0 (X)) for some Polish l.c. space X that
is locally homeomorphic to an at most 3-dimensional CW-complex – optical not
visibly because some of them can be considered as topological subspaces of R7 but
not of R6 or the they can be in the “visible” class of subsets of R3 . In particular,
each separable C*-algebra in the UCT class is KK-equivalent to C0 (X) for at least
one of such X.

Lemma B.11.1. Let W denote the compact Alexandroff (one-point)-completion

of N × R. If G is a countable Abelian group, then there exists a continuous map
 11. EXAMPLES THAT EXHAUST THE UCT-CLASS 1243

ϕ : W → W such that ϕ(∞) = ∞ and that K1 (X) = 0 and K0 (X) ∼

= G for
X := Y \ ((0, 1] × {∞}) where Y denotes the mapping cone of

Y := ((0, 1] × W ) ∪ϕ W

of ϕ.

Proof. Let Z∞ denote the free Abelian group with free generators y1 , y2 , . . ..
If x1 , x2 , . . . ∈ G is a sequence that generates G, then there is an epimorphism
λ : Z∞ → G with λ(yp ) := xp . We can suppose that 0 appears in the sequence (xn )
infinitely often. Then the kernel F ⊆ Z∞ of λ is not finitely generated. By [308,
thm. 14.2], there is an isomorphism µ : Z∞ → F from Z∞ onto F .
It is not difficult to see that there is an automorphism σ of Z∞ (given by an
inductively selected new basis y10 , y20 , . . . of Z∞ ), such that ν := µ ◦ σ : Z∞ → Z∞
has the property that
lp
X
ν(yp ) = gp,j yj
j=kp

with kp ≤ lp , gp,j ∈ Z and limp→∞ kp = ∞.

It follows that it suffices to define suitable continuous maps ϕp from Cp :=
{p} × S 1 ∼= (R ∪ {∞}) (where we identify R with {eis ; s ∈ (0, 2π)}) onto the
rosette
Wp := ({kp , kp + 1, . . . , lp − 1, lp } × R) ∪ {∞} .
The latter is the identification of the points (j, 1) (representing our infinite point
of Wp ) in the mp := 1 + lp − kp disjoint copies Cj = {j} × S 1 of S 1 . There exists
maps ϕp : Cp → Wp that satisfy ϕp (∞) := ∞ and that the maps

K1 (ϕp ) : K1 (Wp ) → K1 (Cp )

satisfy K1 (ϕp )[Uj ] = gj [Vp ] in K1 (Wp ). Here Uj : Wp → S 1 and Vp : Cp → S 1 are

defined by Uj (k, z) := 1 if k 6= j and Uj (j, z) := z, and by Vj (j, z) := z. The map
ϕp : Cp → Wp can be defined as follows:
Select t0 = 0 ≤ t1 ≤ · · · ≤ tmp = 2π with tj−1 = tj if gj = 0 and tj−1 < tj if gj 6= 0.
If gj = 0, let ϕp ((p, etj 2πi )) := (j, 1) = ∞. If gj 6= 0, then define homeomorphisms
γj : [tj−1 , tj ] → [0, 2|gj |π] by the affine maps γj (s) := (s − tj−1 )(tj − tj−1 )−1 2gj π if
gj > 0 and γj (s) := (tj −s)(tj −tj−1 )−1 2|gj |π if gj < 0. Now define ϕp ((p, e2πs )) :=
(j, eiγj (s) ) ??? for s ∈ [tj−1 , tj ] (where we identify all (j, 1) with ∞).
Then we define ϕ : W → W by ϕ((p, z)) := ϕp ((p, z)) -
Let Y := ((0, 1] × W ) ∪ϕ W denote the mapping cone of ϕ : W → W . It
follows, that X := Y \ ((0, 1] × {∞}) is locally a 2-dimensional CW-complex, that
K1 (W ) = K1 (N × R) ∼= Z∞ and that K1 (ϕ) : K1 (N × R) → K1 (N × R) defines the
monomorphism ν : Z∞ → Z∞ . In particular,

K1 (C0 (ϕ)) : K1 (C0 (N × R)) → K1 (C0 (N × R))

is an injective map with co-kernel isomorphic to G.

1244 B. EXACT C*-ALGEBRAS AND EXAMPLES

The Mayer-Vietoris sequence shows that the 6-term K∗ -sequence of the short-
exact sequence
C0 ((0, 1) × (N × R) → C0 (X) → C0 (N × R)
(defined by a pull-back) has injective boundary map

∂ : K1 (C0 (N × R)) → K0 (C0 ((0, 1) × (N × R)) ∼

= K1 (C0 (N × R))

given by K1 (C0 (ϕ)). Since K1 (C0 ((0, 1) × (N × R)) = K0 (C0 (N × R)) = 0, we get
K1 (X) = K1 (C0 (X)) = 0 and K1 (X) = K0 (C0 (X)) ∼ = G. 

Remark B.11.2. If the group G is finitely generated then one can take for X
with K0 (X) ∼
= G and K1 (X) = 0 likewise:

(i) X := a finite disjoint union of points and of pointed Moore spaces

(D/ ∼n ) \ {0}

(For construction of Moore spaces see [692, ex.12.2]), or – alternatively –

(ii) X := the mapping cone of a continuous map from a not necessarily finite
cloverleaf (A × R) ∪ {∞} into a not necessarily finite cloverleaf (or a point)
(B ×R)∪{∞}, where A is a finite set of cardinality |A| = minimal number
of generators of G and |B| ≤ |A| (and B can be empty).

12. S.p.i. subalgebras B with Hausdorff Prim(B)

We say that a C *-algebra D can be locally approximated by C*-subalgebras B

with Hausdorff primitive ideal space Prim(B), if, for every finite subset X ⊆ D and
for each ε > 0, there is a C *-subalgebra B := B(X, ε) ⊆ D of D such that Prim(B)
is Hausdorff and dist(x, B) := inf b∈B kx − bk < ε for every x ∈ X.
It is likely that Theorem ??K?? can be used to show, that this property can
be reformulated as a property of the primitive ideal space Prim(D) alone, provided
that D is separable and nuclear and D ∼ = D ⊗ O2 ⊗ K :
If B ⊆ D has Hausdorff Y := Prim(B), then the action I(D) ∼
= O(Prim(D)) →
O(Y ) of Prim(D) on Y is l.s.c. and monotone upper s.c. Conversely those actions
define C *-morphisms from C0 (Y, O2 ⊗ K) into D.
Consider all local approximations of this type.
Is the above local “Hausdorff” approximation property equivalent to the fol-
lowing property:
Property (???): The family of adjoint maps from C0 (Y )+ to the Dini functions
on Prim(D) contain in its images every finite set of Dini functions on Prim(D) up
to ε > 0.
We give an almost minimal example D, that can not be locally approximated
by C *-subalgebras B with Hausdorff primitive ideal space Prim(B), and contains
regular abelian C *-subalgebras with rather different properties.
Proofs are not complete ??
 12. S.P.I. SUBALGEBRAS B WITH HAUSDORFF Prim(B) 1245

Remark B.12.1. Let A := {f ∈ C([0, 1], M2 ) ; f (1) ∈ ∆} where ∆ denotes the

diagonal matrices of M2 .
We fix a unital monomorphism h1 : M2 (O∞ ) → O2 and a non-degenerate mono-
morphism h2 : (O2 ⊕ O2 ) ⊗ K → O∞ ⊗ K.
Further let D := A ⊗ O2 ⊗ K, D1 := {f ∈ C([0, 1], O2 ) ; f (1) ∈ h1 (∆ ⊗ O∞ )},
and let

D2 := {f ∈ C([0, 1], O∞ ⊗ K) ; f (1) ∈ h2 ((O2 ⊕ O2 ) ⊗ K)} .

Then:

(1) The abelian C *-subalgebra B := C([0, 1], ∆) of A is regular in A and in

D, but B ,→ A does not define a KK(Prim(A), ·, ·) equivalence.
(2) D can not be locally approximated by C *-subalgebras of D that have
Hausdorff primitive ideal spaces.
(3) The algebras D0 := D, D1 and D2 have same primitive ideal spaces, and
contain regular abelian C *-algebras C0 , C1 respectively C2 , such that the
inclusion maps define KK(Prim(Dk ); ·, ·)-equivalences (k = 0, 1, 2).

Proof. We embed A into D by A 3 a 7→ a ⊗ r ∈ D := A ⊗ O2 ⊗ K, where r

is a non-zero projection in O2 ⊗ K. Notice that Prim(A) ∼
= Prim(D) by restriction
of primitive ideals I / D 7→ I ∩ A / A.
The two characters χ1 and χ2 on ∆ define characters on A and, therefore, define
closed points x1 and x2 of Prim(A) (and of Prim(D)). Since [0, 1) is naturally
homeomorphic to the primitive ideal space of the closed ideal C0 ([0, 1), M2 ) of A,
there is an open subset U of Prim(A) that is homeomorphic to [0, 1). The closed
complement is Prim(A) \ U = {x1 , x2 } , where x1 and x2 denote the kernels of χ1
respectively χ2 .
Since A ⊆ C([0, 1], M2 ), the set-valued map µ(t) := {t} for t ∈ [0, 1) and
µ(1) := {x1 , x2 } from [0, 1] = Prim(C([0, 1], M2 )) into Prim(A) is lower semi-
continuous and is montone upper semi-continuous. The corresponding map

Φ0 : O([0, 1]) → O(Prim(A)) ∼

= O([0, 1) ∪ {x1 , x2 })

is given by Φ0 (V ) := V if 1 6∈ V and Φ0 (V ) := (V ∩ (0, 1]) ∪ {x1 , x2 }.

Clearly, B ∼
= C([0, 1]) ⊕ C([0, 1]) is a commutative C *-subalgebra of A (and,
hence of D) that separates the ideals of A (and, hence, that of D).
The set-valued map Φ from Prim(A) into the closed subsets of Prim(B) ∼ =
[0, 1] × {1, 2} satisfies Φ(xj ) = {(1, j)} for j = 1, 2 and Φ(t) = {(t, 1), (t, 2)} for
t ∈ [0, 1).
The montone upper semi-continuity and the lower semi-continuity of the corre-
sponding map Ψ from the lattice of open subsets of Prim(A) to the lattice of open
subsets of Prim(B) imply that in good cases (e.g. if B separable and exact)
?????????????????
1246 B. EXACT C*-ALGEBRAS AND EXAMPLES

we may suppose that B ⊆ A ⊗ O2

We have for the injective map Ψ : O(Prim(A)) → O(Prim(B)), that an open
subset of W of Prim(B) ∼ = [0, 1] × {1, 2} is in the image of Ψ, if and only if,
(t, 1), (s, 2) ∈ W and s, t ∈ [0, 1) imply that (t, 2), (s, 1) ∈ W .
It follows ????
that the image of Ψ consists of the family of open subsets of V × {1, 2} \ G with
V an open subset of [0, 1] and G a subset of {(1, 1), (1, 2)}.
If I is a closed ideal of A, then (I ∩ B) contains a strictly positive element of
I : This is clear, if I ⊆ C0 ([0, 1), M2 ), because then I = C0 (Y, M2 ) for some open
subset of Y of [0, 1).
In the other cases, there is an open subset Y of [0, 1] with
????????????????????????
Suppose now that for every finite subset Z ⊆ D. and every epsilon there is
C *-subalgebra F of D such that dist(z, F ) < ε for all z ∈ Z.
The two projections p = 1 ⊗ e11 ⊗ r and q = 1 ⊗ e22 ⊗ r of B and its sum
1 ⊗ 12 ⊗ r are (up to MvN-equivalence) the only non-zero projections in D.
Indeed: If Q ∈ D is a projection, it defines a continuous map Q(t) from [0, 1]
to M2 ⊗ O2 ⊗ K with Q(1) ∈ ∆ ⊗ O2 ⊗ K. If we let P (t) := Q(1) for t ∈ [0, 1], then
Q and P are in D and
Qs := Q(s + ts )
?????????
is a homotopy that connects Q and P in the projections of D.
Now use that all non-zero projections in ∆ ⊗ O2 ⊗ K are MvN-equivalent to
the above mentioned three,
????
We have: If. P, Q, R1 , R2 are projections in D such that R1 is MvN-equivalent
to R2 in B, R1 + P = p and R2 + Q = q, then R1 = R2 = 0.
Indeed: The Ri must be in the intersection of the primitive ideals defined by
the characters χk on A, Thus, the Ri are in the ideal C0 ([0, 1), M2 ) ⊗ O2 ⊗ K of D,
which contains only the zero projection.
???????????
check last until here
Since ?????
The reason should be the contradictory observation, that such an inductive
limit decomposition eventually would lead to a continuous map f from [0, 1] into
the Hausdorff space of subsets of {1, 2, 3} which is not constant:
 12. S.P.I. SUBALGEBRAS B WITH HAUSDORFF Prim(B) 1247

Let C := C([0, 1], M2 ) ⊗ O2 . To see such a map, the reader should remark that
then also A ⊗ O2 has such an inductive limit decomposition by unital subalgebras
with Hausdorff primitive ideal spaces. Thus, eventually, one finds, at least after
application of an inner automorphism of A ⊗ O2 , a C *-subalgebra B of A ⊗ O2
which has Hausdorff primitive ideal space Y := Prim(B) and contains the constant
diagonal projections p = diag(1, 0) and q = diag(0, 1) of M2 ∩ A ⊆ C, and contains
also a contraction b of A with kbb∗ (0) − pk < 1/8 and kb∗ b(0) − qk < 1/8. In
particular, B is unital. The two primitive ideals I1 and I2 of A ⊗ O2 corresponding
to the point 1 of Prim(C) ∼ = [0, 1] separate p and q. Thus there are primitive ideals
J1 ⊃ I1 ∩ B and J2 ⊃ I2 ∩ B of B with p ∈ J1 , q 6∈ J1 , p 6∈ J2 and q ∈ J2 .
Every primitive ideal J3 of B which contains the intersection of B with the kernel
of the evaluation map g 7→ g(0) contains neither p nor q. Let Z1 and Z2 be the
open and closed subsets of Y which correspond to the norm functions pb and qb
respectively. Since p + q = 1, Z1 ∪ Z2 = Y . Therefore, the open and closed subsets
W1 := Z1 \ Z1 ∩ Z2 , W2 := Z2 \ Z1 ∩ Z2 and W3 := Z1 ∩ Z2 define a partition of Y
with Jn ∈ Wn , n ∈ {1, 2, 3}.
Since Prim(B) is Hausdorff and B is unital, we get that there are four orthog-
onal projections P, Q, R1 , R2 in B such that R1 is MvN-equivalent to R2 in B,
R1 + P = p and R2 + Q = q.
Above, we have seen, that this can not happen in A ⊗ O2 .
??????
To be explained ??
We define a continuous and open map λ from Y onto {1, 2, 3} by λ(x) := n for
x ∈ Wn and n ∈ {1, 2, 3}.
Then
f : t ∈ [0, 1] 7→ λ(Y \ Ψup
B,C ([0, 1] \ {t}))

is a map from [0, 1] into the Hausdorff space of subsets of {1, 2, 3}, and f (0) = {3},
f (1) = {1, 2}, which leads to a contradiction. Recall here that ΨupB,C is a set-valued
upper semi-continuous and monotone lower semi-continuous map. Thus f satisfies
f (S) = f (S) for every subset S of [0, 1]. 

Next proofs not complete ??

Remark B.12.2. Let A a unital C *-algebra with Hausdorff Prim(A) and p, q ∈

A projections with kp + q − 1k < 1 such that {(p − q)2 }0 ∩ A has again Hausdorff
primitive ideal space.
Then A decomposes into a direct sum A ∼ = B ⊕C such that C ∗ (π2 (p), π2 (q), 1C )
is a commutative C *-subalgebra of C, π2 (q) = 1C − π2 (p), and that 1 ⊗
C ∗ (π1 (p), π1 (q), 1B ) is contained in a 2-homogenous C *-subalgebra ∼= C(Y, M2 ) of
O2 ⊗ B with Y an open and closed subset of X × [0, π/2].
Next has to be checked !
1248 B. EXACT C*-ALGEBRAS AND EXAMPLES

Moreover, it can be managed that the induced unital C *-morphism

h : C ∗ (P, Q, 1) → 1 ⊗ C ∗ (π1 (p), π1 (q), 1B )

extends to a unital C *-morphism C([0, π/2], M2 ) → O2 ⊗ B.

Above seems not to be obvious !!
Check if extension exists for irreducible rep’s of B !!
Let D := {f ∈ C([0, 1], M2 ) ; f (1) diagonal}. The quotient

O2 ⊗ (C ∗ (P, Q, 1)|[π/4, π/2]) ∼

= O2 ⊗ D

of O2 ⊗ C ∗ (P, Q, 1) does not contain a unital C *-subalgebra A that satisfies 1 ⊗

(P |[π/4, π2 ]) ∈ A and dist(1⊗(Q|[π/4, π2 ]); A) < 1/32 and has Hausdorff primitive
ideal space.
This non-existence implies that the C *-algebras

K ⊗ O2 ⊗ [C0 ([0, π/2), M2 ) + CP + C(1 − P )] ∼

= K ⊗ O2 ⊗ D

and
K ⊗ O2 ⊗ C ∗ (P, Q, 1)
can not be an inductive limit of C *-algebras An with Hausdorff primitive ideal
spaces Prim(An ).
It follows that the lattice I(C ∗ (P, Q, 1)) ∼
= O(Prim(C ∗ (P, Q, 1))) of open sub-
sets of the (non-coherent) primitive ideal space Prim(C ∗ (P, Q, 1)) of C ∗ (P, Q, 1) can
not be an algebraic limit of lattices O(Xn ) for locally compact Hausdorff spaces Xn
with respect to lower s.c. and monotone upper s.c. actions of Xn+1 on Xn .

Proof. The C *-algebra A0 := C([0, π/2], M2 ) and its C *-subalgebra

C (P, Q, 1) can be described by a universal C *-algebra C ∗ (R, S) with follow-
∗

ing universal relations:

R is a partial isometry, i.e., RR∗ R = R, with relations R2 = 0 and (R − R∗ )4 =
−(R − R∗ )2 . And S satisfies 0 ≤ S ∗ = S ≤ −(R − R∗ )2 and SR = RS.
In particular, −(R − R∗ )2 is the unit element of C ∗ (R, S), C ∗ (R) ∼ = M2 , S
∗
is in the centre of C (R, S) and S has spectrum Spec(S) ⊆ [0, 1]. It follows that
arcsin(S) is a well-defined positive element in the centre of C ∗ (R, S), has spectrum
in [0, π/2] and C ∗ (arcsin(S)) = C ∗ (S). It follows that the natural C *-morphism
from M2 ⊗C(Spec(arcsin(S))) onto C ∗ (R, arcsin(S)) = C ∗ (R, S) is an isomorphism.
To define a unital C *-morphism λ from C ∗ (R, S) onto C([0, π/2], M2 ) ∼
= M2 ⊗
C[0, π/2] we can choose the matrix-unit R(ϕ) := e21 and S(ϕ) := sin(ϕ)12 . This
epimorphism shows in particular that Spec(S) = [0, 1] and Spec(arcsin(S)) =
[0, π/2]. Notice that λ(S 2 ) = (P − Q)2 is a positive contraction in the centre
of A0 and λ(S) = |P − Q|.
An inverse µ : C([0, π/2], M2 ) → C ∗ (R, S) of λ can be defined by the natural
epimorphism from C([0, π/2], M2 ) = M2 ⊗ C(Spec(arcsin(S))) onto C ∗ (R, S), i.e.,
is determined by a µ(e2,1 ) := R and µ(f0 · 12 ) := arcsin(S) for f0 (t) := t on [0, π/2].
 12. S.P.I. SUBALGEBRAS B WITH HAUSDORFF Prim(B) 1249

Then µ(M2 ) = C ∗ (R) ⊆ C ∗ (R, S), µ(P ) = R∗ R ∈ C ∗ (R), µ(Z) = R − R∗ ,

µ((P − Q)2 ) = S 2 , µ(H) = arcsin(S) · Z, and µ(Q) = exp(−H)R∗ R exp(µ(H).
Let A a unital C *-algebra and p, q ∈ A projections such that p and 1 − p are
full projections in {(p−q)2 }0 ∩A. Let B := O2 ⊗({(p−q)2 }0 ∩A). Then there exists
a partial isometry V ∈ B with V ∗ V = 1 ⊗ p and V V ∗ = 1 ⊗ (1 − p), because 1B is
properly infinite and 1 ⊗ p and 1 ⊗ (1 − p) are full and properly infinite projections
in O2 ⊗ B, cf. Lemma 4.2.6(ii).
If we let T := 1 ⊗ |p − q|, then U (1 ⊗ q) = (1 ⊗ p)U for U := exp(arcsin(T ) ·
(V − V ∗ )) and there is a unique unital C *-morphism ρ : C ∗ (R, S) → B ⊆ O2 ⊗ A
with ρ(R) = V and ρ(S) := T . Then ρ(Z) = ρ(R + R∗ ) = V + V ∗ , ρ((P − Q)2 ) =
ρ(S 2 ) = T 2 = 1 ⊗ (p − q)2 ,

ρ(H) = ρ(arcsin(S) · Z) = (1 ⊗ arcsin(|p − q|))(V − V ∗ ) .

Why ρ(Q) = 1 ⊗ q ???

It follows that ρ(P ) = ρ(R∗ R) = 1 ⊗ p, ρ(H) = ρ(arcsin(S) · (R − R∗ )) =
(1 ⊗ arcsin(|p − q|)) · (V − V ∗ ), and

ρ(Q) = exp(−ρ(H))(1 ⊗ p) exp(ρ(H)) = 1 ⊗ q .

Let A a unital C *-algebra with Hausdorff primitive ideal space X := Prim(A)

and p ∈ A a projection. Then X is compact and the generalized Gelfand transforms
p : x ∈ X 7→ kπx (p)k = kp + Jx k ∈ [0, 1] and
b

1 − p : x ∈ X 7→ kπx (1 − p)k = k(1 − p) + Jx k ∈ [0, 1]

\

are continuous functions on X with values in {0, 1} such that max( b 1 − p ) = 1,

p, \
because Dini functions on compact Hausdorff spaces X are continuous on X.
The support U1 of b p (respectively U2 of \1 − p ) is the open subset of X cor-
responding to the closed ideal J1 of A generated p (respectively J2 generated by
1 − p). Since 1 − pb (respectively 1 − 1[
− p) is continuous and non-negative, its sup-
port is open. Thus, U1 and U2 are open and compact subsets of X. In particular
U1 and U2 are also closed subsets of the Hausdorff space X. Then V := U1 ∩ U2 ,
W1 := U1 \ V and W2 := U2 \ V are disjoint open and closed subsets of the compact
Hausdorff space X with X = V ∪ W1 ∪ W2 .
Since the Jk correspond to Uk = Wk ∪ V , the ideal K := J1 ∩ J2 has open and
closed support V in X = Prim(A). Recall that 1 ∈ J1 + J2 = A, πJ1 (p) = 0 and
πJ2 (p) = 1.
Let I denote the closed ideal of A corresponding to the open and compact subset
V of X. Then I + K = A, J1 + J2 = A and K ∩ I = {0}. If we define B1 := A/I,
D1 := A/J1 and D2 := A/J2 . Then πI |K is an isomorphism from K onto B1 ,
and πK |I defines an isomorphism from I onto C1 := A/K ∼ = D1 ⊕ D2 , where the
isomorphism from C1 onto D1 ⊕ D2 is induced by a ∈ A 7→ πJ1 (a) ⊕ πJ2 (a). In
1250 B. EXACT C*-ALGEBRAS AND EXAMPLES

particular, πI (p) and πI (1 − p) are full projections of B1 . Since πJ1 (p) ⊕ πJ2 (p) =
0 ⊕ 1, it follows that πK (p) is in the centre of the unital C *-algebra C1 .
Is next the critical case?
Let q ∈ A another projection and and let Y := arcsin(Spec(|p − q|)). Recall
that there is a natural unital C *-morphism from C(X × Y ) into {(p − q)2 }0 ∩ A
which makes {(p − q)2 }0 ∩ A into a C(F )-algebra for some closed subset F of X × Y .
Our assumptions imply that the C(F )-algebra {(p − q)2 }0 ∩ A is in a natural way
a C *-bundle over X × arcsin Spec(|p − q|).
Then πK (q) and πK (p) commute in C1 , and we can repeat the arguments with
(πI ({(p − q)2 }0 ∩ A), πI (q)) in place of (A, p).
We get finally a direct sum decomposition A ∼ = B ⊕ C such that π1 : A → B
and π2 : A → C have the property that π1 (p), 1 − π1 (p), π1 (q) and 1 − π1 (q) are
projections in B which are full in {π1 (p − q)2 }0 ∩ B, and that π2 (qp) = π2 (pq).
Notice that B = π1 (A) and {π1 (p − q)2 }0 ∩ B have Hausdorff primitive ideal space
Y that is homeomorphic to an open and closed subset of X ∼ = Prim(A), respectively
of X × arcsin(Spec(|p − q|)).
To simplify notation we suppose from now on that A itself has the property that
p, 1 − p, q, 1 − q are full projections (and that X := Prim(A) is Hausdorff) and that
(p − q)2 is an essential element of the centre of A, i.e., that Ann((p − q)2 , A) = {0}.
In the considered case, we still have that our additional assumption k(p + q) −
1k = k(1 − p) − qk < 1 remains in charge in the new A, – our renamed B and
π1 (p), π1 (q).
If we apply part (v) of Lemma 4.1.3 to 1 − p, q in place of p, q then we get
−h∗ = h ∈ C ∗ (p, q, 1) ⊆ A with khk = arcsin kp + q − 1k < π/2 and php = 0 =
(1 − p)h(1 − p) such that exp(−h)(1 − p) exp(h) = q. Thus h = a − a∗ for some
a ∈ A with a = pa(1 − p) and kak = arcsin kp + q − 1k.
Need ???
the existence of c ∈ O2 ⊗ pA(1 − p) with cc∗ = 1 ⊗ p, c∗ c = 1 ⊗ (1 − p) and
b ∈ A+ with bp = pb, bq = qb and a = bc
???????
We can use that 1 ⊗ (p − q)2 is in the centre of O2 ⊗ A and that 1 ⊗ p and
1 ⊗ (1 − p) are both full and properly infinite projections of O2 ⊗ A.
Explicit formula for H|[0, ψ] for study of the inclusion
C ∗ (P, Q, 1)|[0, π/4] ⊆ C([0, π/4], M2 ).
An explicit formula for H|[0, ψ] ∈ C ∗ (P, Q)|[0, ψ] for ψ ∈ [0, π/2) is given by
H(ϕ) := ϕ(e2,1 − e1,2 ).
Notice that

P Q(ϕ)(1 − P ) = P exp(−H)(ϕ)P · P exp(H)(ϕ)(1 − P ) = − cos(ϕ) sin(ϕ)e1,2 .

 12. S.P.I. SUBALGEBRAS B WITH HAUSDORFF Prim(B) 1251

and
P Q(ϕ)(1 − P ) − (1 − P )Q(ϕ)P = cos(ϕ) sin(ϕ)Z
and (P − Q(ϕ))2 = sin(ϕ)2 12 , because

Q(ϕ) − P = [cos(ϕ), − sin(ϕ)]> [cos(ϕ), − sin(ϕ)] − P = sin(ϕ)I

with I = [αjk ] with α11 = − sin(ϕ) = −α22 , α12 = α21 = − cos(ϕ), i.e., orthogonal
with determinant −1, I 2 = 12 . Similarly: (P + Q(ϕ) − 1)2 = (Q(ϕ) − (1 − P ))2 =
cos(ϕ)2 12 .
I.e., always (p − q)2 + (p + q − 1)2 = 1, |p − q| and |p + q − 1| are in the centre
of C ∗ (p, q, 1).
Alternatively one can use symmetry of sin and cos at π/4 (= 45◦ ).
H := a − a∗ , a = −ϕe1,2 for a = CP Q(1 − P ) with C := D−1 arcsin |P − Q|
where D := (1 − (P − Q)2 )1/2 · |P − Q|
?????????.
Let A ⊆ O2 ⊗ (C ∗ (P, Q, 1)|[π/4, π/2]) a unital C *-subalgebra such that p0 :=
1O2 ⊗ P |[π/4, π/2] ∈ A. Suppose that X := Prim(A) is Hausdorff. Since A is
separable and unital, X is second countable and compact, i.e., is a compact Polish
space.
Each quotient C *-algebra of C[π/4, π/2] ⊗ A has again Hausdorff primitive
ideal space. Thus, the C *-subalgebra C ∗ (C[π/4, π/2] · 1, A) generated by A and
C[π/4, π2 ] · 1 is a again a unital C *-subalgebra of O2 ⊗ (C ∗ (P, Q, 1)|[π/4, π/2]) that
contains p0 and has again a Hausdorff primitive ideal space. Thus, we can suppose
in addition that 1 ⊗ C[π/4, π/2] is contained in A.
If we use that O2 ∼ = O2 ⊗ O2 ⊗ · · · , then we can tensor here all algebras
in question again by O2 and suppose moreover that O2 ⊗ A ∼ = A. Then A is
purely infinite and the MvN-equivalence classes of projections in A are in 1-1-
correspondence with the open and compact subsets of X.
Since A ⊆ O2 ⊗C ∗ (P, Q, 1)|[π/4, 1π/2] is contained in C([π/4, π/2], O2 ⊗M2 ) ∼ =
O2 ⊗ C[π/4, π/2] and 1 ⊗ C[π/4, π/2] ⊆ A is contained in the centre of O2 ⊗
(C ∗ (P, Q, 1)|[π/4, π/2]), the C *-algebra A is C *-bundle over [π/4, π/2], i.e., is the
C *-algebra of continuous sections of a continuous field

ϕ ∈ [π/4, π/2] 7→ Aϕ ⊆ O2 ⊗ M2 ⊆ O2 .

The same happens with the C *-subalgebras p0 Ap0 and (1 − p0 )A(1 − p0 ) of A,

because, they are invariant under multiplications with 1 ⊗ f where f ∈ C[π/4, π/2].
The generalized Gelfand transforms b a(J) := ka + Jk for J ∈ Prim(A) ∼
= X are
Dini functions on X. The class of Dini functions on Hausdorff spaces X coincides
with the bounded non-negative continuous functions f ∈ C0 (X) with σ-compact
locally compact open support f −1 (0, ∞) in X. The Gelfand transforms qb ∈ C0 (X)
of projections q ∈ A are the characteristic functions of open and compact subsets
of X corresponding to the closed ideal span(AqA) generated by q.
1252 B. EXACT C*-ALGEBRAS AND EXAMPLES

Let J1 ⊆ A the closed ideal of A generated by p := 1 ⊗ (P |[π/4, π/2]), J2 ⊆ A

the closed ideal generated by 1 − p = 1 ⊗ (1 − P )|[π/4, π/2] and U1 , U2 ⊆ X are the
corresponding open subsets.
− p = 1 − pb. It implies
Their characteristic functions are given by pb and by 1[
that U1 and U2 are compact (and open) and satisfy U1 ∪ U2 = X.
The set F := X \ (U1 ∩ U2 ) is an open and compact subset of X and is a metric
space with F1 := F ∩ U1 and F2 := F ∩ U2 , that are disjoint open and closed subsets
of F with F = F1 ∪ F2 .
To be filled in ??
Suppose that the lattice I(C ∗ (P, Q, 1)) is an algebraic limit of the lattices
O(Xn ), where the corresponding actions are l.s.c. and monotone upper s.c.
Then the corresponding actions

Ψn : O(Xn+1 ) → O(Xn )

are induced from suitable C *-morphisms

C0 (Xn , K ⊗ O2 ) → C0 (Xn+1 , K ⊗ O2 )

and K ⊗ O2 ⊗ C ∗ (P, Q, 1) is isomorphic to an inductive limit of C0 (Xn , K ⊗ O2 )

coming from suitable compatible C *-morphisms

φn : C0 (Xn , K ⊗ O2 ) → K ⊗ O2 ⊗ C ∗ (P, Q, 1) .

For some n0 there exists a unitary u ∈ M(K ⊗ O2 ⊗ C ∗ (P, Q, 1)) and projections
r, s ∈ C0 (Xn , O2 ) such that s ≤ r e11 ⊗ 1 ⊗ 1 = u∗ φn0 (e11 ⊗ r)u, e11 ⊗ 1 ⊗ p =
u∗ φn0 ((e11 ⊗ s)u and dist( q; φn0 (e11 ⊗ (r C0 (Xn0 , O2 )r)) < 1/32.
The proof depends from above un-verified local uniqueness !!!


13. Approximate locally liftable maps

Let J / A a closed ideal. We say that a completely contractive linear map

T : C → A/J (e.g. T = idA/J ) is ε-approximately locally liftable if for every finite-
dimensional subspace X ⊆ C there is a completely contractive linear map S =
S(T,X,ε) : X → A with kπJ ◦ S − ηX k ≤ ε. The map T is approximately locally
liftable if T is ε-approximately locally liftable for all ε > 0. The map T is locally
liftable if T is ε-approximately locally liftable for ε = 0.
Let A ⊆ L(H) and denote by F the hereditary C *-subalgebra of L(H) that is
generated by J. Then A∩F = J, F is approximately injective (in the sense of [238]),
and A+F is C *-algebra with the intersection properties (F ⊗D)∩(A⊗D) = J ⊗D
and ((A + F ) ⊗ D)/(F ⊗ D) = (A ⊗ D)/(A ⊗ J) (because J contains an approximate
unit of F ). The approximation arguments in [238] show that on each separable
C *-subalgebra C1 ⊆ C there is a complete contraction S : C1 → A + F with
πF ◦ S = T |C1 if T : C → A/J (and then T : C → (A + F )/F ) is approximately
locally liftable.
 14. REDUCTIONS TO SEPARABLE SUBSPACES (SEP 1) 1253

In fact, it follows (by using arguments in [238]) that T : C → A/J is approxi-

mately locally liftable (into A), if and only if, T : C → (A + F )/F is locally liftable,
if and only if, on every separable C *-subalgebra C1 ⊆ C there is a completely
contractive map S1 : C1 → A + F with πF ◦ S1 = T |C1 .
If we apply this to T = idA/J , then we get: idA/J is approximately locally
liftable, if and only if, J ⊗ D → A ⊗ D → (A/J) ⊗ D is exact for every C*-algebra
D.
For exact A (and more generally for “locally reflexive” A in the sense of [238])
it is true that W : A/J → B is nuclear if V = W ◦ πJ is nuclear.
Indeed, more generally, one can check with help of the criteria (i) in Remark 3.1.2.
????? that the identity map idA/J of A/J is locally liftable, in the sense that
for every subspace X ⊂ A/J of finite dimension and every ε > 0 there exists a
completely (1 + ε)-contractive map T : X → A with πJ ◦ T = idX , then this is
sufficient to prove the nuclearity of W if W ◦ πJ is nuclear on A:
This can be seen by an (elementary but unpleasant) approximation and extension
argument, or by the following more conceptual argument.
If idA/J is locally liftable, then (the reader can check that) Give argument for
proof!
J ⊗ D → A ⊗ D → (A/J) ⊗ D
is an exact sequence for each C *-algebra D.
If A is locally reflexive, then this sequence is exact, – almost directly by the
definition of (matricial) local reflexivity.
One can show that the local reflexivity of A is equivalent to the (metric) ex-
actness of all sequences

L ⊗ D → A ⊗ D → (A/L) ⊗ D

for all closed left-ideals L ⊆ A and all C *-algebras D.

Give refs or arguments for proof of this!

14. Reductions to separable subspaces (sep 1)

The following lemma on “reductions to the separable case” is used in some cases,
where we need to apply a kind of “excision” methods. Such type of reductions can
be managed by iterated application of the variant [553, prop. 7.2] by E. Michael of
the Bartle-Graves theorem [52, thm. 4]. Our elementary “separable” selection of
sequences does not involve the axiom of choice for sets of uncountable cardinalities
(as all selections principles do implicitly!), but we use the Open Mapping Theo-
rem that is an application of the (almost – but not really – “constructive”) Baire
category theorem.

Lemma B.14.1. Let B denote a Banach space, and let X, Y ⊂ B closed linear
subspaces of B, such that the (algebraic) sum X + Y is closed in B.
1254 B. EXACT C*-ALGEBRAS AND EXAMPLES

Then for every separable linear subspace G ⊆ B there exist a separable closed
linear subspace D of B such that D has following properties:

(i) G ⊆ D,
(ii) (D ∩ X) + (D ∩ Y ) = D ∩ (X + Y ) ,
(iii) dist(d, V ) = dist(d, V ∩ D) for all d ∈ D, where V means each of the
closed subspaces X + Y , X, Y and X ∩ Y of B.

If B is a C*-algebra, then a separable C*-subalgebra D of B with properties

(i)–(iii) can be found.
If B is an operator space, and Mn (X) and Mn (Y ) are equipped with the induced
matrix norms then a separable closed subspace D of B can be found such that D
satisfies in addition the matrix versions (iii*) of Part (iii):

(iii*) dist(d, Mn (V )) = dist(d, Mn (V ∩ D)) for all d ∈ Mn (D), n ∈ N, where V

stands for each of the spaces X ∩ Y , X, Y and X + Y .

That the linear space X + Y of B is closed in B is a necessary extra assumption

because there exist closed linear subspaces X and Y e.g. of C[0, 1] and of L1 [0, 1]
such that the algebraic sum X + Y is not closed.
Give Refs. to counter examples !
The proof of Part (ii) uses the open mapping theorem.
We prove first a sub-lemma that describes an inductive selection procedure.

Sublemma B.14.2. Let B a closed linear subspace of L(H) for some Hilbert
space H, and X, Y closed linear subspaces of B such that X + Y is closed in B.
Let G ⊆ B a separable subset of B.
Then there are separable closed linear subspaces G1 , G2 , . . . of B that have the
following properties:

(1) G ⊆ Gn ⊆ Gn+1 for all n ∈ N

(2) Gn ∩ (X + Y ) is contained in the closure of the linear subspace
(Gn+1 ∩ X) + (Gn+1 ∩ Y ) of Gn+1 ∩ (X + Y ).
(3) k(idn ⊗πV ∩Gn+1 )(b)k ≤ k(idn ⊗πV )(b)k for all b ∈ Mn (Gn ) , where V
stands for each of the four closed subspaces X, Y , Z := X + Y and X ∩ Y
of B, and idn ⊗πV denotes the quotient map Mn (B) 7→ Mn (B)/Mn (V ).

If B is a C*-algebra, then separable C*-subalgebras Gn of B with properties (1)–(3)

can be found for n = 1, 2, . . ..

Proof of Sublemma B.14.2. The Property (3) splits into the 4 cases where
V := X, V := Y , V := X + Y or V := X ∩ Y . Each of (1), (2) and of the 4
cases of (3) can be considered separately for any given separable subspace Gn ⊆
B (in place of any given separable G ⊆ B), i.e., we get six separable subspaces
Gn+1,1 , Gn+1,2 , Gn+1,3 , . . . , Gn+1,6 of B that are as desired in steps (1),(2) and
(3) independently. Then any separable subspace Gn+1 ⊆ B that contains the six
 14. REDUCTIONS TO SEPARABLE SUBSPACES (SEP 1) 1255

subspaces Gn+1,k (k = 1, . . . , 6) satisfies the conditions (1)–(3) with respect to the

given Gn .
If, moreover, B is a C *-algebra, then this shows that we can take as Gn+1 any
separable C *-subalgebra of B that by the Gn+1,k (k = 1, . . . , 6).
Therefore it suffices to find independently for given separable G ⊆ B (e.g. for
G := Gn with before constructed Gn ⊆ B) and for each property (k), k = 1, 2, 3,
and each part of (3) a “solution” Gn+1 ⊇ Gn with the in (1)–(3) listed properties.
Ad(1): It is an “input” condition: Given any separable subset G, e.g. G :=
Gn ⊇ B, then any separable subspace Gn+1,1 of B with G ⊆ Gn+1,1 satisfies (1).
Ad(2): Let G any closed separable linear subspace of B, (e.g. G := Gn if Gn
is constructed before). Then G ∩ (X + Y ) is separable and closed. Take a dense
sequence z1 , z2 , . . . ∈ G ∩ (X + Y ), and find xn ∈ X and yn ∈ Y with zn = xn + yn .
Let F any closed separable subspace of B that contains the closed linear span of
the elements of G and of {x1 , y1 , x2 , y2 , . . .}. Then G ∩ (X + Y ) is contained in the
closure of the linear subspace (F ∩ X) + (F ∩ Y ) of F ∩ (X + Y ). And we can take
Gn+1,2 := F .
Ad(3): Recall that k(idn ⊗πV )(b)k = dist(b, Mn (V )) for all b ∈ Mn (B).
Given any G (e.g. G := Gn , a before selected Gn ) and let V one of the 4 listed
spaces (resp. operator spaces) X, Y, X + Y, X ∩ Y in B. Take a sequence of matrices
g1 , g2 , . . . that is dense in the unit ball of Mn (G). For each gk ∈ Mn (G) we find
hk ∈ Mn (V ) such that
kgk − hk k ≤ (1 + 2−k ) · dist(gk , Mn (V )) .

We can take as Gn+1,k (k ∈ {1, 2, 3, 4} here depending on the considered V )

any separable closed linear subspace of B that contains all entries of the matri-
ces h1 , h2 , . . ., because it is not difficult to see that it has then the property that
dist(g, Mn (Gn+1,k ∩ V )) = dist(g, Mn (V )) for all g ∈ Mn (G) ( 3 ).
If this is done for all V ∈ {X, Y, X + Y, X ∩ Y } then any separable closed linear
subspace Gn+1 of B that contains Gn+1,1 ∪. . .∪Gn+1,4 has the property that, for all
g ∈ M (G) holds dist(g, Mn (Gn+1 ) ∩ Mn (V )) = dist(g, Mn (V )) for the considered
spaces V ⊆ B, 

Proof of Lemma B.14.1. If B is a Banach space B, and X ⊆ B and Y ⊆ B

are closed linear subspaces, then we could consider B as a closed subspace of the
C *-algebra `∞ (M ) for a set M of suitable cardinality. Thus it suffices to consider in
the following always operator spaces, i.e., closed linear subspaces of a given closed
linear subspace B ⊆ L(H).
S
Let G, G1 , G2 , . . . as in Sublemma B.14.2, define G∞ := n Gn and D := G∞ .
Notice that Part (3) of Sublemma B.14.2 is equivalent to dist(b, Mn (V ∩ Gn+1 )) =
dist(b, Mn (V )) for all b ∈ Mn (Gn ). It implies dist(b, Mn (V ∩ D)) = dist(b, Mn (V ))
3 Use here that for each contraction g ∈ M (G) and ε > 0 there is an infinite subset R of N
n
such that kg − gk k < ε for each k ∈ R.
1256 B. EXACT C*-ALGEBRAS AND EXAMPLES

for b ∈ Mn (G∞ ) and each n ∈ N. This carries over to all b ∈ Mn (D) and V ∩ D in
place of V ∩ G∞ , because Mn (D) is the norm-closure of Mn (G∞ ) . It implies that
for every n ∈ N and V ⊆ B (V ∈ {X + Y, X, Y, X ∩ Y }) holds that the natural
injective linear map idn ⊗πV from Mn (D)/Mn (D ∩ V ) into B/Mn (V ) of norm ≤ 1
is a complete isometry, as one can see from the formula

dist(b, Mn (D ∩ V )) ≤ dist(b, Mn (V )) for all b ∈ Mn (D) .

It turns out that D is a closed linear subspace of B that satisfies all conditions
(i), (ii) and (iii*) in Lemma B.14.1:
Clearly condition (i) follow from G ⊆ G1 .
If, in addition, B is a C *-algebra, then the Gn in Conditions (1)–(3) of Sub-
lemma B.14.2 can be chosen as C *-subalgebras of B. I.e. we can then suppose that
S
all Gn (n = 1, 2, . . .) are C *-algebras, that G∞ := n Gn is a *-subalgebra of B,
and that D := G∞ is a separable C *-subalgebra of B.
The condition (iii) follows from the Property (3) of Sublemma B.14.2 for the
operator system D. The isomorphisms Mk ∼ = pMn p for all projections p ∈ Mn with
S
rank = k < n, shows that the operator space D := n Gn satisfies

dist(d, Mn (V ∩ D)) = dist(d, Mn (V )) for all d ∈ Mn (D),

for each V ∈ {X + Y, X, Y, X ∩ Y } and every n ∈ N, because this is true for all

d ∈ Gk with k > n, by Part (3) of Sublemma B.14.2.
Clearly the sum (D ∩ X) + (D ∩ Y ) is contained in D ∩ (X + Y ), because D ∩ X
and D ∩ Y are closed subspaces of the closed subspace D ∩ (X + Y ) of B.
We show that the linear space (D ∩ X) + (D ∩ Y ) is dense in D ∩ (X + Y ) :
Property (2) of Sublemma B.14.2 with V := X + Y and the there constructed
Gn ⊆ Gn+1 ⊆ · · · satisfy

Gn ∩ (X + Y ) ⊆ (Gn+1 ∩ X) + (Gn+1 ∩ Y ) ⊆ (D ∩ X) + (D ∩ Y ) .

It implies that
[
( Gn ) ∩ (X + Y ) ⊆ (D ∩ X) + (D ∩ Y ) ⊆ D ∩ (X + Y ) .
n
S
Thus, (D ∩ X) + (D ∩ Y ) is dense in D ∩ (X + Y ) if ( n Gn ) ∩ (X + Y ) is dense
in D ∩ (X + Y ).
To show the latter, let V := X + Y and y ∈ D ∩ V , Then y ∈ D, πV (y) = 0
S
and, by density of n Gn in D, there exist a sequence nk ∈ N with nk < nk+1 and
elements dk ∈ Gnk such that y = limk dk and limk πV (dk ) = 0.
Recall that any closed subspace V of B holds (idn ⊗πV )(p11 ⊗ b) = p11 ⊗ πV (b)
for b ∈ B for the quotient map πV : B → B/V from B, i.e.,

dist(b, V ) = kπV (b)k = k(idn ⊗πV )(p11 ⊗ b)k .

Thus, if we take n := nk in Part (3) of Sublemma B.14.2, then we obtain that

kπG1+nk ∩V (dk )k ≤ kπV (dk )k for dk ∈ Gnk . It follows that there exists ek ∈
 15. PASSAGE TO SEPARABLE SUBALGEBRAS (SEP 2) 1257

G1+nk ∩ V with kek − dk k ≤ kπV (dk )k . Hence limk kek − dk k = 0 and limk kek −
S
yk = 0, i.e., y ∈ n (Gn ∩ V ).
S
It says that n (Gn ∩ V ) is dense in D ∩ V and implies that the subspace
(D ∩ X) + (D ∩ Y ) is dense in D ∩ (X + Y ) = D ∩ V .
Next we show that the dense linear subspace (D∩X)+(D∩Y ) of D∩(X +Y ) is
closed in D ∩ (X + Y ). This shows then the predicted equality (D ∩ X) + (D ∩ Y ) =
D ∩ (X + Y ) in Part(ii), and finishes the proof.
From now on let V := X ∩ Y ⊆ B.
By Part (3) of Sublemma B.14.2 we have that dist(g, Gn+1 ∩V ) ≤ dist(g, V ) for
g ∈ Gn . Thus, dist(g, D ∩ V ) = dist(g, V ) for all g ∈ G∞ . Since G∞ is dense in D
it follows that the restriction to D of the quotient map πV : b ∈ B → b + V ∈ B/V
defines an linear isometry η from the quotient space D/(D∩V ) into B/V with kernel
D ∩ V = D ∩ X ∩ Y . In particular, η|(D ∩ X)/(D ∩ V ) and η|(D ∩ Y )/(D ∩ V )
are isometries into the subspaces X/V respectively Y /V of (X + Y )/V . We denote
this restrictions of η by η1 and η2 .
Recall that the space D ∩ V is a closed subspace of each of the closed subspaces
D ∩ X, D ∩ Y and (D ∩ X) + (D ∩ Y ) = D ∩ (X + Y ). Thus, if the linear space
η(D ∩ X) + η(D ∩ Y ) is closed in (X + Y )/V then (D ∩ X) + (D ∩ Y ) is closed in
D ∩ (X + Y ), because

(D ∩ X) + (D ∩ Y ) = η −1 (η(D ∩ X) + η(D ∩ Y )) .

This implies finally the predicted equation (D ∩ X) + (D ∩ Y ) = D ∩ (X + Y ).

The natural linear map T from the Banach space X/V ⊕∞ Y /V with norms


k(x + V, y + V )k := max kπV (x)k, kπV (y)k onto (X + Y )/V that is given by

T (x + V, y + V ) := (x + y) + V

is surjective and has norm ≤ 2. The T is also injective because x + y = v ∈ V

implies that x = v − y ∈ X ∩ Y = V , thus x, y ∈ V . The Open Mapping Theorem
shows that the inverse T −1 : (X + Y )/V → X/V ⊕∞ Y /V of the linear isomorphism
T is bounded. In particular, T maps each closed subspace of X/V ⊕1 Y /V onto a
closed subset of (X + Y )/V .
The map η1 ⊕ η2 from (D ∩ X)/(D ∩ V ) ⊕∞ (D ∩ Y )/(D ∩ V ) into X/V ⊕∞ Y /V
is an isometry. Thus, its image is a closed subspace of X/V ⊕1 Y /V . It follows that
T maps this image onto a closed subspace of (X + Y )/V . Calculation shows that
T ◦ (η1 ⊕ η2 ) maps (D ∩ X)/(D ∩ V ) ⊕∞ (D ∩ Y )/(D ∩ V ) onto η(D ∩ X) + η(D ∩ Y ).
Thus, η(D ∩ X) + η(D ∩ Y ) is closed in (X + Y )/V . 

15. Passage to separable subalgebras (sep 2)

We define for a, b ∈ A and ε > 0 minimal numbers n := n(b, a, ε) with the

property that there exist c1 , . . . , cn , d1 , . . . , dn ∈ A with δ := kb − k d∗k ack k < ε .
P
1258 B. EXACT C*-ALGEBRAS AND EXAMPLES

The latter inequality implies by Lemma 2.1.9 that there exists a rational number
ρ and ` ∈ N with δ < ρ < ε such that there exist c1 , . . . , cn ∈ A with (b − ρ)+ =
P ∗
k ck ack if a, b ∈ A+ .

New more precise text:

We define numbers n(b, a, ε; B) and γ(b, a, ε; B) ≥ n(b, a, ε; B) (the latter only
in case a, b ∈ B+ and kak = kbk = 1) for C *-subalgebras B of A and non-zero a, b ∈
B with b in the closed ideal J(a) of B generated by a, by letting n := n(b, a, ε; B)
the minimal number n ∈ N such that there exist c1 , . . . , cn , d1 , . . . dn ∈ B with
Pn
kb − j=1 d∗j acj k < ε . We can here take dj = cj if a, b ∈ B+ , cf. Remark B.15.1.
Moreover n(b, a, ε) = n((b∗ b)1/2 , (a∗ a)1/2 , ε).
The number ν := ν(b, a; ε) for a, b ∈ B+ is the smallest number ν ∈ N with the
Pn
property that there exists contractions c1 , . . . , cν ∈ B such that kb− j=1 d∗j acj k <
ε.
What about definition of above defined γ?
Equivalently: There exists X, Y ∈ Mn (B) with kb ⊗ p11 − Y ∗ (a ⊗ 1n )Xk < ε .
This is because we can replace here X, Y ∈ Mn (B) by the columns [c1 , . . . , cn ]> :=
X · (1 ⊗ p11 ) and [d1 , . . . , dn ]> := Y · (1 ⊗ p11 ) and vice versa. Here 1 denotes the
unit element of the multiplier algebra of M(A).
Remark B.15.1. If a, b ∈ B+ and there exists by c1 , . . . , cn , d1 , . . . , dn ∈ A
with kb − j d∗j acj k < ε then there exists f ∈ A such that ej := (cj + dj )f satisfies
P

kb − j e∗j aej k < ε.

P

Indeed, if a, b ∈ B+ then we can use that for the positive elements g := b ⊗ p11
and h := a ⊗ 1n holds also k2g − (Y ∗ hX + X ∗ hY )k < 2ε. It implies that there
exists δ ∈ (0, ε) such that 0 ≤ 2g ≤ (Y ∗ hX + X ∗ hY ) + 2δ(1 ⊗ 1n ).
Clearly, by X ∗ hY = 0, (Y ∗ hX+X ∗ hY )+(X ∗ hX+Y ∗ hY ) = (X+Y )∗ h(X+Y ),
√
which gives that 0 ≤ g ≤ T ∗ hT +δ·(1⊗n) for T := 2(X +Y ) ∈ Mn (A). By Lemma
2.1.9, there exists a contraction d ∈ Mn (A)+ with d∗ (T ∗ hT )d = (g − (ε + δ)/2)+ .
In particular, kg − Z ∗ hZk < ε for Z := T d.
Pn
Thus, with [e1 , . . . , en ]> := Z(1 ⊗ p11 ) we get kb − j=1 e∗j aej k < ε.

Important are Parts (iii) and (iv)

Proposition B.15.2. For every C*-algebra A and every separable subset X ⊂
A of A there exists a separable C*-subalgebra B of A with X ⊂ B such that B has
the following additional properties:

(i) The natural C*-morphism C ⊗max B → C ⊗max A is injective for all

separable C*-algebras C (i.e., B is ”relative weakly injective” in A),
( Are properties (i) and (iii) equivalent ? )
(ii) The natural maps (C ⊗min B)/(Y ⊗min B) 7→ (C ⊗min A)/(Y ⊗min A) are
isometric for all separable C*-algebras C and closed subspaces Y of C.
??? Or better ?:
Require metric version of the “intersection property”:
 16. REDUCTIONS TO SEPARABLE CASES (SEP 3) 1259

(Y ⊗min A) ∩ (C ⊗min B) = Y ⊗min B .

Let Z1 ⊆ Z2 and Z3 ⊆ Z2 operator spaces (closed linear subspaces of
L(`2 )).
What has to be required on this spaces?
that the natural map from Z1 /(Z1 ∩ Z3 ) to Z2 /Z3 is completely iso-
metric ???
(iii) For each k ∈ N, a, b ∈ Mk (B)+ and δ > 0 holds that for every c1 , . . . , cn ∈
Mk (A) there exists d1 , . . . , dn ∈ Mk (B) with kdj k ≤ kcj k for j = 1, . . . , n
and
X X
kb − d∗k adk k ≤ δ + kb − c∗k ack k .
k k

(iv) For every a, b ∈ B and ε > 0, n(b, a, ε; B) = n(b, a, ε; A) and

γ(b, a, ε; B) = γ(b, a, ε; A) if a, b ∈ B+ and kak = kbk = 1.
(Where n(...) and γ(...) are defined ?)
(v) (B ∩ I1 ) + (B ∩ I2 ) = B ∩ (I1 + I2 ) for any closed ideals I1 and I2 of A.
??? ( ⊇ is in question!!! and by out-devision of I1 ∩I2 we can suppose
that l1 ·I2 = {0}. But the question is if one finds separable Bm that satisfies
at least one of the properties and this Bm are contained in the Bn with
n > m ... then the inductive limits gives this ”all properties” B ...)

In particular, every closed ideal J of B is the intersection J = B ∩ I of a closed

ideal I of A with B.

Proof. To be filled in ??
??? Reduce (v) to σ-unital A with approximate unit in B and A = I1 + I2
generated as ideals by B ∩ (I1 + I2 ). ???
B ⊆ A ⊆ A∗∗ leads to B ∗∗ ⊆ A∗∗ in the sense that B ∗∗ is naturally isomorphic
to the σ(A∗∗ , A∗ )-closure of B ⊆ A∗∗ in A∗∗ .
It seems that B is ”relative weakly injective” in A, if and only if there exists a
c.p. contraction P from A into B ∗∗
(A positive contraction would be enough, because its restriction can be nor-
malised to a completely positive contraction S from A∗∗ into B ∗∗ ⊆ A∗∗ with the
property that S(b) = b for all b ∈ B ⊆ A∗∗ .
Since one can do this with B ⊗ Mn ⊆ A ⊗ Mn , one can obtain by cluster-point
arguments that there is a completely positive contraction T from A into B ∗∗ ⊆ A∗∗
with the property that T (b) = b for all b ∈ B.


16. Reductions to separable cases (sep 3)

An important technical property in the study of residually nuclear maps be-

tween non-simple C *-algebras is the notion of local reflexivity for closed subspaces
B ⊆ L(H) introduced by Effros and Haagerup in [238]:
1260 B. EXACT C*-ALGEBRAS AND EXAMPLES

Definition B.16.1. A closed subspace B ⊆ L(H) (i.e., an operator space with

its system of matrix norms) is locally reflexive if, for every finite-dimensional
subspaces X ⊆ B ∗∗ , F ⊆ B ∗ , and ε > 0 there is a linear map T : X → B that
satisfies the following properties (i) and (ii):

(i) |f (T (x)) − x(f )k ≤ εkxk · kf k for all x ∈ X, f ∈ F , and

(ii) T is completely contractive, i.e., kT ⊗ idn k ≤ 1 for all n ∈ N for the
maps T ⊗ idn : X ⊗ Mn → B ⊗ Mn with norms on X ⊗ Mn induced from
B ∗∗ ⊗ Mn ⊆ L(H ⊗ Cn )∗∗ .

Remark B.16.2. One can find for each n ∈ N a linear map T : X → B that
satisfies property (i) and kT ⊗idn k ≤ 1 by the local reflexivity of the Banach spaces
B ⊗ Mn , but in general such T does not exist that it satisfies also kT ⊗ idn+k k ≤ 1
for k ∈ N.
Does it say that there could exist Xk ⊆ B ∗∗ and Yk ⊆ B ∗ of
finite dimension such that,
for all Tk ∈ L(Xk , B) with (i) holds that
limk→∞ kTk ⊗ idk k = ∞ ?
All exact C *-algebras (or, more generally, 1-exact operator spaces) are locally
reflexive. It is not known if locally reflexive C *-algebras are exact in general.
Local reflexivity passes to C *-quotients and C *-subalgebras (in particular,
∗
Q
n M n is not locally reflexive, because it contains C (F∞ ) as a C *-subalgebra).

The class of locally reflexive C *-algebras A is invariant under tensor products

with K and Mn .
But it is so far unknown if tensor products of locally reflexive A with M2∞
(and – therefore – also min-tensor products with arbitrary exact C *-algebras) again
become locally reflexive. It is also not known if the class of locally reflexive C *-
algebras are invariant under inductive limits (at least for suitably “controlled” nets
of C *-morphisms?).
Local reflexivity is not preserved under passage to stable coronas, asymptotic
coronas, multiplier algebras, and ultra-powers of A if A is not sub-homogenous (i.e.,
if A has only irreducible representations of dimension ≤ n for some n ∈ N).
Indeed, it is not difficult to see that stable coronas, asymptotic coronas, mul-
tiplier algebras, and ultra-powers of A are sub-Stonean C *-algebras. Recall that
sub-Stonean C *-algebras C have the property that for orthogonal positive contrac-
tions a, b ∈ C+ there are orthogonal positive contractions c, d ∈ C+ with ca = a
and db = b.
Sub-Stonean C *-algebras C have the property that B1 (`∞ ) = V (B1 (C)) for
every c.p. contraction V : C → `∞ with V (B1 (C)) ∩ c0 dense in B1 (c0 ) (where
B1 (X) means the closed unit ball of a Banach space X).
 16. REDUCTIONS TO SEPARABLE CASES (SEP 3) 1261

If a C *-algebra C with the latter property contains a C *-subalgebra D, that

L
has n Mn as a quotient, then C contains a C *-subalgebra E, such that E has a
Q
quotient that is isomorphic to n Mn = `∞ (C, M2 , M3 , . . .).
The following more general property of a C *-algebra C, stated as follows, is
equivalent to the property that C has no non-zero closed ideal of type-I:
For every non-zero positive e ∈ C+ in the C*-algebra C, the σ-unital heredi-
Q
tary C*-subalgebra eCe contains a C*-subalgebra D ⊆ eCe such that n Mn =
c0 (C, M2 , M3 , . . .) is a quotient of the C*-algebra D.
A non-zero C *-algebra C is sub-homogenous, i.e., there exists nC ∈ N such
that all irreducible representations of C have dimension ≤ nC , if and only if, there
Q
does not exist a C *-subalgebra D ⊆ C with the property that n Mn is a quotient
C *-algebra of D. (For a proof of this criteria it suffices to consider only the type-I
algebras by Glimm’s sub-quotient theorem, saying that each separable non-type-I
C *-algebra C contains a hereditary C *-subalgebra D such that C = D · C + C ·
D + N (D, C) and N (D, C)/D ∼ = M2∞ ).
The full group C *-algebras C ∗ (G) of a lattices G in non-compact semi-simple
Lie groups are not exact (and even not locally reflexive) because they all contain a
copy of C ∗ (F2 ) and every separable unital C *-algebra is a quotient of its subalgebra
C ∗ (F∞ ).
∗
The reduced group C *-algebras Cred (G) are all exact for G in this class of
discrete groups.

Remark B.16.3. Suppose that A ⊆ M is a C *-subalgebra of a W*-algebra M .

Then centre of the *-ultra-strong closure of A in M contains a family of mutually
P
orthogonal projections {pκ } such that Q := 1− κ pκ satisfies QA = AQ = {0}, and
that each pκ can be decomposed in M into a sum of a family mutually orthogonal
countably decomposable projections in M of cardinality ≤ density of A.
(It suffices to consider the case where M = A∗∗ .)
What was the original remark ??? Good for? ??

Questions B.16.4. Given a semi-finite properly infinite W*-algebra M and a

separable C *-subalgebra A ⊆ M .
Does there exist a family of mutually orthogonal countably decomposable pro-
jections pγ ∈ A0 ∩ M with
P
pγ = 1 such that following properties (i)–(iii) are
satisfied?

(i) pγ is properly infinite,

(ii) for every weakly l.s.c. trace τ : N+ → [0, ∞] (on the positive part of N :=
{pγ ; γ ∈ Γ }0 ∩ M ) there is a weakly l.s.c. trace τ 0 : M+ → [0, ∞] with
τ |A+ = τ 0 |A+ ,
(iii) for every projection p in the centre of N there is a projection q in the
centre of M with kapk = kaqk for all a ∈ A .
1262 B. EXACT C*-ALGEBRAS AND EXAMPLES

Does there exist a separable unital C *-subalgebra A ⊆ C ⊆ M (with M a

W*-algebra) such that following properties (i)–(iv,v????) are satisfied

(i) O2 is unitally contained in C, δ∞ (C) ⊆ C and δ∞ ⊕id unitarily equivalent

to δ∞ by unitaries in C (for the weakly continuous δ∞ : M → M ).
(ii) Every closed ideal I of C is the intersection of a closed ideal of M with
C.
(iii) For every projection p in the centre of M there is q in the centre of C
with kapk = kaqk for all a ∈ A.
(iv) C ∩ K is an essential ideal of C.
(v) C ∩ K is stable (????????).

local lifting, local lifting property,

locally liftable exact sequences
weakly injective algebras,
relatively weakly injective subalg.,
approximately injective algebras
Recall that A ⊗ B denotes the minimal C *-algebra tensor product of A and B.

Lemma B.16.5. Let A and B C*-algebras, where A is separable and B is σ-

unital.

(i) If X is a separable Banach space, C ⊆ L(X, B) a point-norm closed set

of contractions, J a σ-unital closed ideal of B. If for all S, T ∈ C and for
every contraction a ∈ J+ , the element

aS(·)a + (1 − a2 )1/2 T (·)(1 − a2 )1/2

is again in C, then πJ ◦ C is point norm closed in L(X, B/J) .

(ii) A c.p. map V : A → B/J is locally c.p. liftable (or c.b. liftable ???) if

J ⊗ L(`2 ) → B ⊗ L(`2 ) → (B/J) ⊗ L(`2 )

is exact. If, moreover ?????. See Effros –Haagerup.

(iii) B is locally reflexive, if and only if,

Jτ ⊗ L(`2 ) → Bτ ⊗ L(`2 ) → B ∗∗ ⊗ L(`2 )

is exact, if and only if, the natural C*-morphism form B ∗∗ ⊗max L(`2 )
into (B ⊗ L(`2 ))∗∗ factorizes over B ∗∗ ⊗min L(`2 ).
(iv) More ??????????????

Proof. (i): We use metrics on L(X, B) and L(X, B/J) that define the point-
norm topologies: Let x1 , x2 , . . . ∈ X a sequence that dense in the unit-ball of
P −n
X. Then let ρ0 (T, S) := n2 kT (xn ) − S(xn )k and similarly defined ρ1 on
L(X, B/J). Obviously ρ1 (πJ ◦ T, πJ ◦ S) ≤ ρ0 (T, S) and ρ0 (T, S) ≤ kT − Sk.
Moreover, if E := {eτ } ⊂ J+ is a with respect to the separable C *-subalgebra
C ∗ (T (X) ∪ S(X)) ⊆ B quasi-central approximate unit, then ρ1 (πJ ◦ T, πJ ◦ S) is
 17. SINGLY GENERATED SEPARABLE C *-ALGEBRAS (SEP. 4) 1263

equal to

inf{ ρ0 (1 − eτ )1/2 T (·)(1 − eτ )1/2 , (1 − eτ )1/2 S(·)(1 − eτ )1/2 ; eτ ∈ F } .

Let C 3 Tn : X → B such that πJ ◦ Tn converges in point-wise to some map

S : X → B/J.
Can select nk ∈ N that ρ1 (πJ ◦ Tnk , πJ ◦ Tnk ) < 4−k . Then we define a new
sequence Rk ∈ L(X, B) with the property that ρ0 (Rk , Rk+1 ) < 2−k with help step
by step modification by replacing the Tnk+1 by
1/2 1/2
Rk+1 := (1 − fk )1/2 Tnk+1 (·)(1 − fk )1/2 + fk Rk (·)fk

with suitably chosen fk ∈ F ⊆ J+ . It follows that the Rk ∈ C converge in point-

norm on X to a map R ∈ C with πJ (R) = S .
To be filled in ??


17. Singly generated separable C *-algebras (sep. 4)

Lemma B.17.1. Let E denote a unital C*-algebra with properly infinite unit
and s1 , s2 ∈ E isometries with orthogonal ranges: s∗j sk = δj,k 1, and let e ∈ E+
denote a positive contraction.
Then the singly generated C*-subalgebra C ∗ (X) = C ∗ (X ∗ , X) of E generated
by the element

X := X(s1 , s2 , e) := s1 · 1 + s1 (2 + e)−1 s∗2 + s2 (2 + e)−1 s∗1

is identical with the C*-subalgebra C ∗ (s1 , s2 , e) ⊆ E generated by the elements

{s1 , s2 , e} ⊆ E .

Proof. Obviously X(s1 , s2 , e) ∈ C ∗ (s1 , s2 , e) ⊆ E by definition of X and

C ∗ (s1 , s2 , e) = C ∗ (s1 , s2 , a) for a := (2 + e)−1 by e = a−1 − 2.
Recall that s1 , s2 ∈ A, s∗j sk = 1A and that a := (2 + e)−1 ∈ A+ is invertible
with estimates 1/3 ≤ a ≤ 1/2 .
We define Y := s1 (1 + s1 as∗2 + s2 as∗1 ). Consider f := s1 as∗2 + s2 as∗1 . Then
f = f , f s1 = s2 a and f 2 = s1 a2 s∗1 + s2 a2 s∗2 . Thus kf k ≤ 1/2, and it follows that
∗

1/2 ≤ h := 1 + f ≤ 3/2 .
This element h allows to rewrite the above defined Y as X = s1 h .
We show that the C *-subalgebra C ∗ (X) = C ∗ (X ∗ , X) of A generated by X
contains s1 , s2 and a, i.e., that C ∗ (X) = C ∗ (s1 , s2 , a) . Then 1 ∈ C ∗ (X) and
e = a−1 − 2 ∈ C ∗ (X), i.e., C ∗ (X) = C ∗ (s1 , s2 , e) .
The above discussion allows to see the following formulas: X ∗ X = h2 ≥ 1/4 ,
s1 = X(X ∗ X)−1/2 = Xh−1 ∈ C ∗ (X), 1 = s∗1 s1 , f = h − 1 = (X ∗ X)1/2 − 1
and f s1 = s2 a . Thus, s1 , f, 1 ∈ C ∗ (X) and a2 = as∗2 s2 a = s∗1 f 2 s1 . Since, by
1264 B. EXACT C*-ALGEBRAS AND EXAMPLES

assumptions, a ≥ 1/3 we get that a, a−1 , e = a−1 − 2 ∈ C ∗ (X), and finally that
s2 = f s1 a−1 ∈ C ∗ (X). 

Compare next Proposition with [770, thm.2.3] and the Remark B.17.3 con-
cerning other attempts to prove Proposition B.17.2.

Proposition B.17.2. If E is a separable C*-algebra with a properly infinite

unit element, then there exists a unital C*-morphism ψ : O∞ → E and E is singly
generated as C*-algebra.

Proof. Let s1 , s2 ∈ E isometries with s∗1 s2 = 0. Then let tk := sk1 s2 for

k = 1, 2, . . . . The elements tk ∈ E satisfy that t∗j tk = δj,k 1E for j, k ∈ N, i.e., the
t1 , t2 , t3 , . . . define a unital C *-morphism from O∞ into E .
By separability of E, there exists a norm-dense sequence a1 , a2 , . . . in the pos-
itive contractions in E. The positive contraction
∞
X
e := sk (2k + ak )−1 s∗k ∈ E+
k=1

has the property that the C *-subalgebra C ∗ (t1 , t2 , e) of E generated by {t1 , t2 , e} is

identical with E, because sk , 1E ∈ C ∗ (t1 , t2 ) for all k ∈ N and (2k + ak )−1 = s∗k esk
gives that ak = (s∗k esk )−1 − 2k 1E is in C ∗ (t1 , t2 , e) .
Thus, E is singly generated by the element X(t1 , t2 , e) ∈ E defined in Lemma
B.17.1. 

Remark B.17.3. The authors H. Thiel and W. Winter of [770, thm.2.3] quote
Nagisa [567] who had referenced the result [770, thm.2.3] (and thus Proposition
B.17.2) to the author (E.K.) of this book. Probably M. Nagisa was informed that
the author wrote in some evening discussion ( or in a break between talks ) some-
times in the 1990th in MF Oberwolfach a formula on the blackboard in the discus-
sion room that did show that every properly infinite unital (4) separable C *-algebra
is singly generated. It could be that this proof was different from the here given
proof of Proposition B.17.2, but it was definitely different from all of the (incom-
plete) proof of [770, thm.2.3]. Thus, unfortunatly we couldn’t use it as citation
here, had to give our own proof. H. Thiel and W. Winter quote for the proof of
[770, thm.2.3] an argument in the proof of [580, thm. 9]. But this quoted argument
∗ k
P
does not carry over to an element Y := k≥1 (sk ak sk + 1/2 sk ) with spectrum of
ak = a∗k in [4−n /2, 4−n ] as considered in their proof, because they claim – and
need/use it (!) in their proof –
Check next red text carefully again!!!
Perhaps, the constructions give two self-adjoint elements that
generates all?
S∞
... that the spectrum of Y is contained in {0}∪ k=1 [4−n /2, 4−n ]. But this is not
the case in general, e.g. consider the there given formula in case where A := O∞
4It is the case if and only if the algebra contains a copy of O
∞ unitally!
 17. SINGLY GENERATED SEPARABLE C *-ALGEBRAS (SEP. 4) 1265

with generating isometries s1 , s2 , . . . satisfying s∗j sk = δj,k the spectrum of the

element k 4−k sk s∗k +2−k sk is exactly equal to {0}∪ k≥1 (4−k +2−k D), where D is
P S

the unit-disc, in particular the spectrum is not contained in {0}∪ k≥1 [4−k /2, 4−k ] .
S

Unfortunately this gap in the arguments for the proof of [770, thm.2.3] given by
H. Thiel and W. Winter harms then any further argument of their proof.

Further details concerning sums of invertible positive elements and non-

unitary isometries s1 , s2 , . . . ∈ E with s∗j sk = δj,k :
Let us consider elements X ∈ E defined by
X
X := (sk ak s∗k + γk sk ) ,
k≥1

where the ak are positive with Spec(ak ) ⊆ [αk , βk ] for “suitable” 0 < βk+1 <
αk < βk . One can show that the spectrum of such element X is contained {0} ∪
S
k≥1 ([αk , βk ] + γk D) where D := {z ∈ C ; |z| ≤ 1} is the unit disc.

Below we give a more precise formula for Spec(X) that allows to deduce that
P∞ −n
our, in Remark B.17.3 mentioned, special example Y := n=1 4 sn sn + 2−n sn
S −n −n
in O∞ it is precisely equal to {0} ∪ n (4 + 2 D) .
Pn ∗
Let Pn := k=1 sk sk and consider stepwise the “triangular” decompositions
of X as
X = Pn XPn + (1 − Pn )X(1 − Pn ) + Pn X(1 − Pn ) .
It allows to identify X with an upper triangular matrix in M2 (E) with entries
X11 = Pn XPn , X12 = Pn X(1 − Pn ) and X22 = (1 − Pn )X(1 − Pn ). This is
possible, because all X of the above defined type satisfy (1 − Pn )XPn = 0 for this
projections Pn . It implies that

{0} ∪ Spec(X) = {0} ∪ Spec(Pn XPn ) ∪ Spec((1 − Pn )X(1 − Pn )) .

If we proceed with n ∈ N by induction then we can see that this leads to

∞
[
Spec(X) ∪ {0} = {0} ∪ Spec(ak + γk sk ) .
k=1

Here we have used that (sk ak s∗k + γk sk )sk s∗k = sk (ak + γk sk )s∗k . Using that
Spec(S) = D for every non-unitary isometry S ∈ E, this leads to the estimate
Spec(ak + γk sk ) ⊆ {0} ∪ (Spec(ak ) + γk D) .
In the special case where X is build with scalars ak = βk · 1E , and limk βk = 0
and limk γk = 0 it gives the precise equation
[
Spec(X) = {0} ∪ (βk + γk D) .
k

This applies in particular to the above defined special element Y ∈ O∞ .

P 2
An other special case is the case where λ := k γk < ∞ and limk βk = 0
−1/2
P P
we get that S := λ γk sk is an isometry in E and with a := k sk ak sk the
C *-algebra C ∗ (a, S) contains automatically all sk and ak , i.e., is identical with
C ∗ (a1 , a2 , . . . ; s1 , s2 , . . .) .
1266 B. EXACT C*-ALGEBRAS AND EXAMPLES

Our conclusion is that the basic question for the study of above discussed
phenomena should be:
Under which circumstances holds that a, S ∈ C ∗ (a + S) ⊂ L(`2 ) for a given positive
invertible operator a ∈ L(`2 ) and a non-unitary isometry S ∈ L(`2 )?
Here `2 := `2 (N) and the isometry S should have an infinite dimensional co-
kernel (1 − SS ∗ )`2 , e.g. up to unitary equivalence S ∈ L(`2 ) can be defined with an
ONB by Sen := e2n .
More calculations. To be checked again.
No final decision obtained if the weakly separated case works:

Even if one selects the positive numbers more carefully, namely such that

([αk , βk ] + γk D) ∩ ([α` , β` ] + γ` D) = ∅ ,

for k 6= `, then we still have to struggle to obtain at least two of the isometries to
get the rest of all. This “weak” disjointness holds, e.g. in the case where we take
αk := 2−2k , βk := 2αk and γk := (αk − βk+1 )/4 = αk /8 , because then

αk+1 = αk /4 < βk+1 = αk /2 < αk < 2αk = βk

and
βk+1 + γk+1 < αk − γk .

It follows from αk+1 = αk /4 βk+1 = αk /2 and γk = αk+1 /2 = αk /8 .

This particular choice causes at least that there exist unitaries u1 , u2 , . . . ∈
C (X) that have the very weak property that at least sk uk ∈ C ∗ (X), because
∗

the above defined very particular αk , βk and γk satisfy the additional conditions
γk < αk2 /(2βk ) . (obtained form βk = 2αk and γk = αk /8 ).
It is not unlikely that there is some possible repair of the arguments in the
proof of [770, thm.2.3] if one takes the γk very small in relation to to the lower
bounds αk > 0 of invertible elements ... but we doubt this.
But then one can only find (not necessarily selfadjoint) idempotents Pk ∈
C (X) with Pk X = sk ak s∗k + γk sk .
∗

An other calculation gives sk ak s∗k +γk sk = Y +Z with Y := sk (ak +γk sk )s∗k and
Z := γk sk (1 − sk s∗k ) satisfy ZY = 0, Z 2 = 0, (1 − sk s∗k )Z = 0 and (1 − sk s∗k )Y = 0.
In particular,

{0} ∪ Spec(Y + Z) = {0} ∪ Spec(Y ) = Spec(ak + γk sk )

for the invertible positive contraction ak .

18. On asymptotically estimating functions

Let A := L(`2 ), and let F : (a, b) ∈ A × A → A and G : (a, b) ∈ A × A → A

with the properties that G(a, b) = 0 implies that F (a, b) = 0 and that there exists
 19. AN ELEMENTARY MATRIX NORM BOUND 1267

a bund γ ∈ (0, ∞) such that max{ kF (a, b)k, kG(a, b)k } ≤ γ for all contractions
a, b ∈ A . We require moreover that for (a1 , a2 , . . .), (b1 , b2 , . . .) ∈ `∞ (A) holds

F ( πc0 (A) (a1 , a2 , . . .), πc0 (A) (b1 , b2 , . . .) ) = πc0 (A) (F (a1 , b1 ), F (a2 , b2 ), . . .)

– by suitable “local” re-imbedding of parts of `∞ (A)/c0 (A) into L(`2 ) –, and

G( πc0 (A) (a1 , a2 , . . .), πc0 (A) (b1 , b2 , . . .) ) = πc0 (A) G(an , bn )k .

Likewise, for F is would be enough to have

kF ( πc0 (A) (a1 , a2 , . . .), πc0 (A) (b1 , b2 , . . .) )k ≥ lim inf kF (an , bn )k ,
n

and for G that

kG( πc0 (A) (a1 , a2 , . . .), πc0 (A) (b1 , b2 , . . .) )k ≤ lim sup kG(an , bn )k .
n

Then there is a general method for such F and G that decides if there exists
an increasing continuous function ϕ ∈ C0 (0, 1]+ with kF (a, b)k ≤ ϕ(kG(a, b)k) for
all contractions a, b ∈ A :
Fix t ∈ (0, 1] and take the supremum S(t) over the norms kF (a, b)k for all
contractions a, b with kG(a, b)k ≤ t and call it ??? ...
It is certainly ≤ γ and t 7→ S(t) is increasing.
There exist increasing continuous functions ϕ(t) on (0, 1] with S(t) ≤ ϕ(t) and
limt→0 ϕ(t) = limt→0 S(t).
If 0 < µ := limt→0 S(t), then there exist sequences of contractions a1 , a2 , . . .
and b1 , b2 , . . . with kG(an , bn )k < 2−n and kF (an , bn )k ≥ µ − 2−n .

19. An elementary matrix norm bound

A simple explanation for the estimate in Remark 2.1.10 goes as follows:

We can write each matrix a = [ajk ] ∈ Mn (A) as a sum of n sub-matrices RL(`)

or as sum of n sub-matrices LR(`) that have the property that RL(`)∗ RL(`) and
LR(`)∗ LR(`) are diagonal matrices. Thus the sub-matrices have norms kRL(`)k
and kLR(`)k that is ≤ the maximum of the norms of there entries ajk . Then use
that
Xn X n
a= RL(`) = LR(`) .
`=1 `=1

Such sub-matrices can be selected as follows :

RL(`) for ` ∈ {1, . . . , n} the sub-matrix given with entries ajk at the places (j, k)
with j + k = ` mod (n) i.e., with the elements

{ a1,`−1 , a2,`−2 , . . . , a`−2,2 , a`−1,1 ; a`,n , a`+1,n−1 , . . . , an−1,`+1 , an,`

on its given places and zero’s otherwise. It is on 45o degree from right to left
downward directed lines.
1268 B. EXACT C*-ALGEBRAS AND EXAMPLES

Similarly let LR(`) for ` ∈ {1, . . . , n} denote the sub-matrix with entries on
the left to right downward directed lines, parallel to the main diagonal LR(n) =
diag(a11 , a22 , . . . , ann ) :
The sub-matrix LR(`) is non-zero only on places (j, k) with j − k = ` mod (n) :

{ a1,` , a2,`+1 , . . . , a`−1,n−1 , a`,n ; a`+1,1 , a`+2,2 , a`+3,3 , . . . , an−1,`−1 , an,`

All sub-matrices RL(`) and LR(`) have the property that the products
??????
Only the case of the right-to-left downward diagonal a1,n , a2,n−1 , . . . , an,1 (with
j+k = 1 mod(n)) splits not into two – in orthogonal boxes contained – sub-matrices.
The same happens with the lines that are parallel to the main diagonal
a1,1 , a2,2 , . . . , an,n .

20. On flip-invariant subalgebras of A ⊗ A

Let A a C *-algebra and let γ : A ⊗ A → A ⊗ A the unique *-automorphism

given by γ(a ⊗ b) = b ⊗ a for a, b ∈ A. We call it the flip automorphism on A ⊗ A.
It would be helpful to know more on the fix-point algebras of the flips on
O2 ⊗ O2 , O∞ ⊗ O∞ and P∞ ⊗ P∞ ∼ = (1 − s1 s∗1 )O∞ (1 − s1 s∗1 ) = O∞ st .
Also on Z ⊗ Z ?
More generally what about the fix-point algebra of the flip on On ⊗ On , Mn∞ ⊗
Mn∞ ? Are they simple?
The flip algebra (A ⊗ A)γ is nuclear if (and only if ??) A is nuclear.
Is (A ⊗ A)γ also simple ??? (seems to be if we can say something on the case of
tensorial Z absorption, but the flip algebras in Mn ⊗ Mn or K ⊗ K are not simple.)

if A is simple and unital without quotient of finite dimension? The fix-point

algebras of the flips on Mn∞ ⊗ Mn∞ are simple (except possible the case P∞ ⊗ P∞ ),
and they have all a central sequence of copies of E2 (the latter with exception of
Mn∞ ⊗ Mn∞ or of the Jiang-Su algebra).
It makes them pi-sun C *-algebras if the fix-point algebras of the flip on Mn∞ ⊗
Mn∞ is simple ????
and Why this?
The inductive limit for the latter is give by unital Mf (k) ⊕ Mg(k) → Mf (k+1) ⊕
Mg(k+1) . This happens with some “over-crossing” and Mf (k) ⊕ Mg(k) is unitally
contained in Mnk ⊗ Mnk , in particular n2k = f (k) + g(k)). f (k) := nk (nk + 1)/2,
g(k) := nk (nk − 1)/2.
Iteration rules are given by f (k + 1) := βn (f (k), g(k)) := f (k)n(n + 1)/2 +
g(k)n(n − 1)/2,
 20. ON FLIP-INVARIANT SUBALGEBRAS OF A ⊗ A 1269

g(k + 1) := γn (f (k), g(k)) := f (k)n(n − 1)/2 + g(k)n(n + 1)/2 .

It comes from the rule for eigenvalues of the flip-unitary for Mn ⊗ Mn on
`2 (n) ⊗ `2 (n):
Plus: n + (n2 − n)/2 = n(n + 1)/2 and Minus: (n2 − n)/2 = n(n − 1)/2.
Check of f (k) + g(k) = n2k is OK:

g(k + 1) = nk (nk + 1)n(n − 1)/4 + nk (nk − 1)n(n + 1)/4 = nk+1 (nk+1 − 1)/2 ,

and

f (k + 1) := f (k)n(n + 1)/2 + g(k)n(n − 1)/2 = nk+1 (nk+1 + 1)/2 .

Is there a variant of next prop.

for A = B ⊗ Z?

Proposition B.20.1. Let A := B ⊗ O∞ and let γ : A ⊗ A → A ⊗ A the flip

isomorphism γ(a1 ⊗ a2 ) := a2 ⊗ a1 .
Then the fix-point algebra (A ⊗ A)γ has only ideals I that are intersections of
ideals J of A ⊗ A with (A ⊗ A)γ :

I = (A ⊗ A)γ ∩ J .

The flip-algebra (A ⊗ A)γ contains in its multiplier algebra a central sequence

of unital copies of O∞ .
In particular, (A ⊗ A)γ is simple and purely infinite if B is simple and exact.

Proof. It is easy to see (why ⊆ ??) that for each closed ideal I of A ⊗ A the
set I + γ(I) is a γ-invariant closed ideal of A ⊗ A with

(A ⊗ A)γ ∩ (I + γ(I)) = (A ⊗ A)γ ∩ I .

We show that if 0 ≤ b ∈ (A ⊗ A)γ and c ∈ (A ⊗ A)+ is contained in the ideal I

of A⊗A generated by b then the element c+γ(c) is in the closed ideal J of (A⊗A)γ
generated by b :
Let 0 ≤ b ∈ (A⊗A)γ with kbk = 1, c ∈ (A⊗A)+ in the ideal of A⊗A generated
Pn
by b and ε ∈ (0, 1/4). Then there exists g1 , . . . , gn ∈ A⊗A with kc− k=1 gk∗ bgk k <
ε. Define η := ε/(n2 + 1)(1 + C 2 ), where δ := max{kg1 k, . . . , kgn k}.
Since b can be approximated by elements in the algebraic tensor product A A
there exist isometries s1 , . . . , sn , sn+1 , . . . , s2n ∈ M(A) with mutually orthogonal
ranges and the property that k(sk ⊗ sn+k )b − b(sk ⊗ sn+k )k < η for above defined
η and k ∈ {1, . . . , n}.
The norms of (sk ⊗ sn+k )∗ b(sk ⊗ sn+k ) − b and of (sj ⊗ sn+j )∗ b(sk ⊗ sn+k ) for
j 6= k are less than η.
1270 B. EXACT C*-ALGEBRAS AND EXAMPLES

−2
Pn
h∗ bh − k=1 gk∗ bgk can be
P

Now let h := k (sk ⊗ sn+k )gk . The norm of δ
estimated by
X X
k(sk ⊗ sn+k )∗ b(sk ⊗ sn+k ) − bk + k(sj ⊗ sn+j )∗ b(sk ⊗ sn+k )k ≤ n2 η .
k j6=k
P
Recall that γ(h) = k (sn+k ⊗ sk )γ(gk ) .
It follows that γ(h)∗ h = 0, kγ(h)∗ bγ(h) − γ(c)k < ε and
n
X
kγ(h)∗ bhk ≤ δ 2 k(sn+j ⊗ sj )∗ b(sk ⊗ sn+k )k ≤ δ 2 n2 η .

k,`=1

It gives that

k (c + γ(c)) − (h + γ(h))∗ b(h + γ(h)) k ≤ 2(ε + δ 2 n2 η) < 4ε .

Notice that c ≤ c + γ(c) and that c + γ(c) and h + γ(h) are in (A ⊗ A)γ .
Thus, each ideal of the fix-point algebra (A ⊗ A)γ of the flip automorphism on
A ⊗ A is the intersection J ∩ (A ⊗ A)γ of an ideal J of A ⊗ A with (A ⊗ A)γ .
Let (sn ), (tn ) ⊂ M(A) sequences of isometries with limn→∞ ksn a − asn k +
ktn a − atn k = 0 for each a ∈ A, and s∗n tn = 0, then sn ⊗ sn , tn ⊗ tn ∈ M(A) ⊗



γ

M(A) ⊆ M (A ⊗ A)γ ) , the isometries (sn ⊗ sn )n and (tn ⊗ tn )n are central
sequences for all elements of A ⊗ A. 

Question B.20.2. What happens if A ∼

= B ⊗ E(M2∞ , M3∞ )?
What happens in case that A is approximately divisible?

Corollary B.20.3. If A is a pi-sun C*-algebra then its flip-algebra (A ⊗ A)γ

is again a pi-sun algebra.

Proof. It is clear that (A ⊗ A)γ is separable, unital and nuclear because

V (a) := (1/2)(a + γ(a)) is a unital conditional expectation from the separable,
unital and nuclear algebra A ⊗ A onto (A ⊗ A)γ . By Corollary F(ii), A ∼ = A ⊗ O∞
for all pi-sun algebras A. Thus Proposition B.20.1 applies to A and gives that
(A ⊗ A)γ is simple and contains a central sequence of copies of O∞ . (Recall here
that O∞ ∼ = O∞ ⊗ O∞ ⊗ · · · and that C ∗ (skn tn , k = 1, 2, . . .) is isomorphic to O∞
for each pair of isometries sn , tn with s∗n tn = 0.) 

21. On ideal structure of crossed products

Remark B.21.1. Let X a non-compact locally compact Hausdorff space. Then

X is a completely regular space, i.e., for each closed subset F ⊆ X and x ∈ X \ F
there exists g ∈ C0 (X)+ with g(x) = 1, kgk = 1 and g(F ) = {0}. We denote
by β(X) the Stone-Cech compactification of X (maximal compact Hausdorff space
with X as a dense subspace). Recall that it can be defined equivalently by the
property that each continuous map φ : X → Z from X into a compact (Hausdorff)
space Z uniquely extends to a continuous map β(φ) : β(X) → Z. An other descrip-
tion of β(X) is the the set of characters of Cb (X) with σ(Cb (X)∗ , Cb (X))-topology,
 21. ON IDEAL STRUCTURE OF CROSSED PRODUCTS 1271

i.e., as Gelfand character space of Cb (X). Thus, there is a natural isomorphism

Cb (X) ∼ = C(β(X)) by Gelfand-Naimark. and each continuous map from X into a
compact metric space M – e.g. a closed disk – extends to a continuous map from
β(X) into M .
Let A a C *-algebra and let Y ⊂ γ(X) a non-empty compact subset of the
corona space γ(X) := β(X) \ X of X.
If we use the natural isomorphism C(β(X)) ∼= Cb (X) then Cb (X, A) becomes
the C(β(X))-algebra of continuous sections in a suitable continuous field of C *-
algebras {Ay ; y ∈ β(X)} over β(X). This is because the norm function map
f ∈ Cb (X, A) 7→ N (f ) ∈ Cb (X) defined by N (f )(x) := kf (x)k. It satisfies with
N (f )h = N (hf ) for h ∈ C0 (X). This defines Cb (X, A) as continuous field over
β(X) :
(1. ) Cb (X, A) is a Cb (X)-algebra:
If g : X → A in Cb (X, A) and f ∈ Cb (X), then (f ·g) ∈ Cb (X, A) for (f ·g)(x) =
f (x)g(x).
(2. ) There is a natural norm function N : g ∈ Cb (X, A) → N (g) ∈ Cb (X)+
that satisfies N (f ·g) = |f |·N (g), N (g1 +g2 ) ≤ N (g1 )+N (g2 ) and N (g ∗ g) = N (g)2 .
The property (1.) and (2.) together impliy that Cb (X, A) the algebra of con-
tinuous sections for the continuous field ( Aω ; ω ∈ β(X) ). Here Aω is the quotient
defined the C *-seminorm N (·)(ω) and is isomorphic to Cb (X, A)/(Jω · Cb (X, A))
with Jω equal to the kernel of the evaluation character f 7→ f (ω) on Cb (X) for
ω ∈ β(X). For our applications it is an important point that Cb (X, A) is also a
lower s.c. field of C *-algebras. and not only an upper s.c. field of C *-algebras over
the compact space β(X).
If Y is a closed subspace of β(X), then we can form the “restriction” C *-
algebra Cb (X, A)|Y := Cb (X, A)/JY , where JY is the ideal of bounded A-valued
continuous functions f ∈ Cb (X, A) with the property that the function g(x) :=
kf (x)k in Cb (X) = C(β(X)) satisfies g|Y = 0, i.e., JY is the closed ideal of Cb (X, A)
generated C0 (β(X) \ Y ) · Cb (X, A).
One can here also think of Cb (X, A) as an ideal of Cb (X, f
A ) where f
A denotes
A if A is unital and A + C1 ⊆ M(A) otherwise
We write always Aω for Cb (X, A)|Y if Y = {ω} for a point ω ∈ β(X) \ X and,
sometimes, also AY for Cb (X, A)|Y .
(There are good reasons for the requirement Y ∩ X = ∅ on Y .)
Well-known examples are A∞ := Cb (X, A)|Y for X = N, Y := β(N) \ N, Aω :=
Cb (X, A)|Y for X = N, Y := {ω} where ω ∈ β(N) \ N is a point (corresponding to
a free ultra-filter on N) or Q(X, A) := Cb (X, A)|Y for Y := γ(X) = β(X) \ X.
Let fa (x) := a for x ∈ X and a ∈ A. The map A 3 a 7→ fa (x) ∈ Cb (X, A)
defines a natural *-monomorphism from A into Cb (X, A) with JY ∩ A = {0} for all
non-empty Y ⊂ β(X) \ X.
1272 B. EXACT C*-ALGEBRAS AND EXAMPLES

Therefore, there are natural *-monomorphisms A ,→ Cb (X, A)|Y . We

identify A with the image of A under this embedding. The hereditary C *-
subalgebras A Cb (X, A)A (respectively A(Cb (X, A)|Y )A) of Cb (X, A) (respec-
tively of Cb (X, A)|Y ) will be denoted by EA or E(X, A) (respectively EA |Y or
E(X, A)|Y ).
It turns out that σ-unital A (or E(X, A)|Y ) is essential in Cb (X, A)|Y for at
least one Y 6= ∅, if and only if, A is unital.
In particular, A is a degenerate C *-subalgebra of Cb (X, A)|Y if A is non-unital
but σ-unital (and X is non-compact).
It follows that there are natural unital *-monomorphisms ϕ1 : M(A) ⊂
Cb (X, M(A))|Y →???? and ϕ2 : M(A) → M(EA |Y ). The morphism ϕ2 is
strictly continuous, because ϕ2 |A is essential in EA (by definition of EA |Y ).
The natural morphism ϕ : M(A) ⊂ Cb (X, M(A))|Y is unital and injective and
satisfies ϕ(a) ∈ A ⊂ Cb (X, A)|Y .
The algebra Cb (X, A)|Y is an ideal of the unital algebra Cb (X, M(A))|Y , there-
fore, there is a natural unital *-morphism ψ : Cb (X, M(A))|Y → M(Cb (X, A)|Y ) ,
with kernel π(I), where π denotes the natural epimorphism from Cb (X, M(A))
onto Cb (X, M(A))|Y and I is the ideal of g ∈ Cb (X, M(A)) with gf ∈ JY for all
f ∈ Cb (X, A).
It holds ϕ(M(A)) ∩ ker(ψ) = {0}, because (ϕ(b)f )(x) = af (x) for b ∈ M(A),
x ∈ X, and bf |Y = 0 for all b ∈ Cb (X, A) implies bA = {0}. Thus, ψ ◦ ϕ : M(A) →
M(Cb (X, A)|Y ) is a unital *-monomorphism with ψ ◦ϕ(a) = a ∈ A ⊂ Cb (X, A)|Y .
Thus, ψ ◦ ϕ(M(A)) ⊂ N (Cb (X, A)|Y, EA ) and the induced unital *-morphism from
M(A) into M(EA ) is the unique strictly continuous extension of the non-degenerate
*-monomorphism A → EA .
If T ∈ L(A, B) is a bounded linear map, then T X (f )(x) := T (f (x)) defines
a bounded linear map T X : Cb (X, A) → Cb (X, B) with T X (JY ) ⊂ T X (JY ) for
Y ⊂ β(X), i.e., T X (f )|Y = 0 if f |Y = 0. It allows to define the linear quotient
map T Y : Cb (X, A)|Y → Cb (X, B)|Y , i.e., T Y satisfies T Y (f |Y ) = T X (f )|Y and
T Y (a) = T (a) for all a ∈ A.
The map L(A, B) 3 T → T Y ∈ L(Cb (X, A)|Y, Cb (X, B)|Y ) is linear and isometric.
If α : A → B is a *-morphism (respectively *-monomorphism, *-isomorphism, com-
pletely positive) then αY : Cb (X, A)|Y → Cb (X, B)|Y is a *-morphism (respec-
tively *-monomorphism, *-isomorphism, completely positive) with αY (a) = α(a) ∈
B for all a ∈ A. If T (a) = cad for c, d ∈ M(A), then T Y : Cb (X, A)|Y →
Cb (X, A)|Y satisfies T Y (f |Y ) = ψ ◦ ϕ(c) · (f |Y ) · ψ ◦ ϕ(d) for f ∈ Cb (X, A), and
T Y (f |Y ) = c · (f |Y )d if f |Y ∈ EA = A(Cb (X, A)|Y )A.
In particular, if α ∈ Aut(A) and M := (Cb (X, A)|Y )∗∗ p the support projection
of EA in M , then αY ∈ Aut(Cb (X, A)|Y ), αY (A) = A, αY (EA ) = EA , (αY )∗∗ (A0 ∩
M ) = A0 ∩ M and (αY )∗∗ (p) = p.
 21. ON IDEAL STRUCTURE OF CROSSED PRODUCTS 1273

Moreover, if α = Ad(u) (i.e., α(a) = uau∗ for a ∈ A, where u ∈ M(A) is

unitary), then αY = Ad(U ) for U = V |Y , where V (x) := u for x ∈ X.
???????????????????????

Lemma B.21.2. Let u ∈ M(A) unitary, and B := Cb (X, A)|Y where Y 6= ∅ is

a closed subset of β(X) \ X (e.g. if X = N then B := A∞ , B := Aω are of this
kind).
The *-isomorphism µ := Ad(u)Y ∈ Aut(B) (cf. Remark B.21.1) is induced on
B by γu (f )(t) := Ad(u)(f (t)) := uf (t)u∗ for t ∈ R+ and f ∈ Cb (R+ , A), and is
given by the inner automorphism Ad(U ) : B 3 b 7→ U bU ∗ ∈ B, for the unitary
U := ψ ◦ ϕ(u) in ψ(ϕ(M(A))) ⊂ ψ(Cb (X, M(A)) ⊂ M(B).
In particular, Ad(U )(a) = Ad(u)(a) for all a ∈ A.
The second conjugate Ad(U )∗∗ : B ∗∗ → B ∗∗ is given by Ad(U )∗∗ (b) =
η(U )bη(U )∗ , where η : M(B) → B ∗∗ means the natural strictly–*ultra-strongly
continuous unital monomorphism.
The σ(B ∗∗ , B ∗ )-closure of A in B ∗∗ is naturally W*-isomorphic to A∗∗ via a
natural isomorphism η1 : A∗∗ → B ∗∗ . Let q = η1 (1) ∈ B ∗∗ denote the unit of
η1 (A∗∗ ) ⊂ B ∗∗ , and identify M(A) naturally with its image in A∗∗ .
Then

(i) η1 (a) = q · η ◦ ψ ◦ ϕ(a) for all a ∈ M(A).

(ii) A0 ∩ B ∗∗ = η1 (A∗∗ )0 ∩ B ∗∗ and Ad(U )∗∗ (A0 ∩ B ∗∗ ) = A0 ∩ B ∗∗ .
(iii) Ad(U )∗∗ (q) = q ∈ A0 ∩ B ∗∗ and qAd(U )∗∗ (b) = qb for all b ∈ A0 ∩ B ∗∗ ,
In particular ab = aAd(U )∗∗ (b) for all a ∈ A and b ∈ A0 ∩ B ∗∗ .

Proof. ??
pB ∗∗ p is the second conjugate of ABA,
????? 

Lemma B.21.3. Suppose that A ⊂ B := Cb (X, A)|Y is as in Remark B.21.1,

α ∈ Aut(A), and q ∈ B ∗∗ denotes the support projection of the hereditary C*-
subalgebra ABA = EA |Y of B. Let β := αY : B → B denote the natural extension
of α to B (cf. Remark B.21.1).
Then β(q) = q, q is in the centre of A0 ∩ B ∗∗ , and β(A0 ∩ B ∗∗ ) = A0 ∩ B ∗∗ .
If there are κ ∈ (0, 1) and a unitary u ∈ M(A) with kα − Ad(u)k ≤ 2κ, then
kβ(p) − pk ≤ κ for all projections p ∈ (A0 ∩ B ∗∗ )q.
In particular, if p is a projection in A0 ∩ B ∗∗ , then aβ ∗∗ (p)p = 0 for all a ∈ A
implies that qp = 0 = qβ ∗∗ (p).

Proof. The linear map β − Ad(U ) is equal to the natural extension (α −

Ad(u))Y of α − Ad(u) ∈ L(A, A) to a bounded linear operator of Cb (R+ , B)|Y ,
where U ∈ M(B) is defined as in Lemma B.21.2. It follows kβ − Ad(U )k ≤ 2κ .
1274 B. EXACT C*-ALGEBRAS AND EXAMPLES

Since β(a) = α(a) for all a ∈ A, we get β ∗∗ (A0 ∩B ∗∗ ) = A0 ∩B ∗∗ and β ∗∗ (q) = q.

The projection q is in the centre of B ∗∗ , because q is in the σ(B ∗∗ , B ∗ )-closure of
A and commutes with A.
By Lemma B.21.2, Ad(U )∗∗ (b) = b for all b ∈ (A0 ∩B ∗∗ )q. Thus, for projections
p ∈ (A0 ∩ B ∗∗ )q,

2kp − β ∗∗ (p)k = k(q − 2p) − β ∗∗ (q − 2p)k ≤ 2κ .

The equation Aβ ∗∗ (p)p = {0} implies β ∗∗ (qp)qp = 0. But then kβ ∗∗ (qp)−qpk ≤

κ < 1 implies that 0 = qp = qβ ∗∗ (p) = β ∗∗ (qp). 

22. On semi-projective relations

Next Def. makes no sense !? In this formulation... only?

Definition B.22.1. An operator-polynomial relation R : F n → B∞ (on ele-

ments of a subset F of a C *-algebra A) is semi-projective on F n if there exists for
each element X ∈ F n with R(X) = 0 a sequence Xk ∈ F n with R(Xk ) = 0 and
limk kXk − Xk = 0 .

23. Non-2-sub-additive quasi-traces exist.

We list some more details than given at the reference places:

Where is the ”reference place”?
The unital type-I C *-algebra A is a (very special) unital extension E :

0 → c0 (K) → E →ψ C([0, 1]2 ) → 0 .

It means that the closed ideal J generated by all commutators in the unital C *-
algebra E is isomorphic to c0 (K) ∼ = c0 ⊗ K. The commutative quotient E/J of
E is isomorphic to C([0, 1]2 ) and, if ψ : E → C([0, 1]2 ) is the defining map for the
extension, then the (central) unital quasi-trace τE on E is given by τA ◦ πc0 (K) ,
where τA is the Aarnes quasi-state on C([0, 1]2 ).
The following says: C := C([0, 1]n ) is a ”projective” C *-algebra ? In the sense:
If ψ : C → B/J is a unital *-monomorphism, then there exists a unital *-morphism
φ : C → B with φ = πJ ◦ ψ.
But what happens with the question concerning additivity, if µ is a quasi trace
then for all strictly commuting elements a, b in the image holds additivity ... ???
THE NEW OBSERVATION IS:
Because C([0, 1]2 ) is projective inside the class of unital Abelian C *-algebras
one gets that an extension with `∞ (K) can be modified to an extension by c0 (K)
such that πc0 (K) (C ∗ (S, T )) ∩ `∞ (K) = {0} .
It gives an extension 0 → c0 (K) → E → C([0, 1]2 ) that has the property that
(1) E is a type I C *-algebra with Abelian quotient of Dimension = 2.
 23. NON-2-SUB-ADDITIVE QUASI-TRACES EXIST. 1275

(2) This extension has the property that every local unital quasi-trace τ on
C([0, 1]2 ) is additive on τ |C ∗ (x, 1)+ for all normal elements x ∈ E.
(3) The composition τE := τA ◦ πc0 (K) is a is additive on each commutative
C *-subalgebra of E. But is not 2-sub-additive.
Why is it not 2-sub-additive ??
(4) There exist elements x, y ∈ E+ with τE (x) = 0, τE (y) = 0 and τE (x+y) = 1.
The restriction map

C0 (X) 3 f 7→ f |Z ∈ C(Z) ∼
= C([0, 1]2 )

defines an epimorphism from A/J ∼

= C0 (X) onto C([0, 1]2 ) and has the property
that ?????????
It is not unlikely that one can manage that that X = Z ∼ = [0, 1]2 using the
index-preserving compressions (T ∗ )n (·)T n on the Toeplitz algebra

0 → K(`2 ) → T = C ∗ (T ) → Cf (S1 ) .

Then one would get a honest extension 0 → c0 (K) → A → C([0, 1]2 ) .

But we need that k[(T ∗ )n aT n , (T ∗ )n bT n ]k becomes small if ab − ba ∈
K(`2 )
for positive contraction a, b ∈ L(`2 ) with ab − ba ∈ K.
We use the following Lemma:

Lemma B.23.1. Let a, b ∈ L(`2 ) positive contractions that satisfy ab − ba ∈ K,

p ∈ K a projection and ε(0, 1/2]. Then there exists a positive contraction c ∈ K
such that p ≤ c, kca − ack < ε , kcb − bck < ε and

k(1 − c)a(1 − c)2 b(1 − c) − (1 − c)a(1 − c)2 b(1 − c)k < ε .

I.e. (1−c)a(1−c) and (1−c)b(1−c) commute up to ε and are compact perturbations

of a and b.

Proof. It is essentially the existence of an approximate unit {en ∈ A+ ; n =

1, 2, . . .} in σ-unital C *-algebras A that satisfies en en+1 = en , ken k = 1 and is
quasi-central with respect to a separable C *-subalgebra B ⊆ M(A), – in the sense
that limn kben − en bk = 0 for all b ∈ B.
The proof uses obvious modifications of the proof of [616, thm. 3.12.14], and
has to observe that the arguments for [616, cor. 3.12.15,cor. 3.12.16] work also
for σ-unital A and separable B ⊆ M(A). It shows that we can find a sequence
(en ) with this properties inside the (algebraic) convex hull of any given countable
approximate unit {p1 , p2 , . . .} of σ-unital C *-algebra A that have the additional
property pn pn+1 = pn . (Such an approximate unit {pn ; n ∈ N } of A exists in
every σ-unital C *-algebra A.)
In our special case we apply this to A := K(`2 ) and take the projections pn ∈ A
of rank n ∈ N as approximate unit of A, that maps `2 (N) to the linear span of the
1276 B. EXACT C*-ALGEBRAS AND EXAMPLES

first n elements of the canonical basis of `2 (N). Now let B := C ∗ (K(`2 ), a, b) ⊆

L(`2 ) = M(A). 

Remark B.23.2. Let f0 ∈ C([0, 1]) = C ∗ (f0 , 1) denote the identity map
f0 (t) := t for all t ∈ [0, 1] and let 1 ∈ C([0, 1]) denote the function that is constant
equal to 1.
The functions g, h ∈ C([0, 1]2 ) = C([0, 1]) ⊗ C([0, 1]) are the coordinate maps
g(x, y) = x and h(x, y) = y, corresponding to f0 ⊗1 and 1⊗f0 in the tensor product
notation.
Notice that C([0, 1]2 ) = C ∗ (g, h, 1) and that this means in the tensor notation
that C([0, 1]) ⊗ C([0, 1]) = C ∗ (f0 ⊗ 1, 1 ⊗ f0 , 1 ⊗ 1).

Lemma B.23.3. The algebra C([0, 1]2 ) is the universal C*-algebra with respect
to the above given relations and has following properties:
If A is a unital C*-algebra that is generated by commuting positive contactions
S, T ∈ A and 1A , then there exists a unique C*-algebra morphism ϕ : C([0, 1]2 ) → A
with ϕ(g) = S, ϕ(h) = T and ϕ(1) = 1A .
In particular, if S, T ∈ A are commuting positive contractions in a unital C*-
algebra A and if ψ : C ∗ (S, T, 1A ) → C([0, 1]2 ) is a unital C*-morphism with ψ(S) =
f and ψ(T ) = g then ψ is an isomorphism form C ∗ (S, T, 1A ) onto C([0, 1]2 ).

Proof. The universality of C([0, 1]2 ) in the class of unital commutative C *-

algebras is easy to see, because C([0, 1]) is the universal unital C *-algebra generated
by a positive contraction, and C([0, 1]) is nuclear. This implies that any pair of
element-wise commuting unital C *-morphisms φ1 , φ2 : C([0, 1]) → A define a unital
C *-morphism ϕ : C([0, 1]2 ) → A with ϕ(g) = φ1 (f0 ) and ϕ(h) = φ2 (f0 ).
Let ψ : C ∗ (S, T, 1A ) → C([0, 1]2 ) a unital C *-morphism with ψ(S) = g and
ψ(T ) = h. Then the universal epimorphism ϕ from C([0, 1]2 ) onto C ∗ (S, T, 1A )
satisfies ψ◦ϕ = id of C([0, 1]2 ), because ψ◦ϕ is unital, ψ◦ϕ(g) = g and ψ◦ϕ(h) = h .
Therefore ψ must be surjective and ϕ injective. Thus ψ and ϕ are isomorphisms. 

Let γn : S 1 → [0, 1]2 a sequence of piecewise smooth continuous map that

contains in its uniform closure the set of all continuous maps γ : S 1 → [0, 1]2 that
are piecewise smooth Jordan curves ( 5 ).
Denote by T ⊂ L(`2 ) the Toeplitz algebra. Notice that K ⊂ T and that there
is a natural isomorphism from C(S 1 ) onto T /K ⊂ C, where C := L(`2 )/K denotes
the Calkin algebra.
The maps γn define a sequence of unital C *-morphisms ϕn : C([0, 1]2 ) →
C(S ) ⊂ C and, therefore, a unital C *-morphism Φ : C([0, 1]2 ) → `∞ (C(S1 )) ⊂
1

`∞ (C) .

5 I.e. the closure of the set of maps γ – but considered as positive contractions in C(S 1 ) ⊕
n
C(S 1 ) – contains all those maps γ : S 1 → [0, 1]2 that are piecewise smooth Jordan curves.
 23. NON-2-SUB-ADDITIVE QUASI-TRACES EXIST. 1277

The C *-morphism Φ is faithful on C([0, 1]2 ), and is a unital C *-monomorphism,

because for every point x ∈ [0, 1]2 there exists a piece-wise smooth Jordan curve
γ : S 1 → [0, 1]2 with x ∈ γ(S 1 ).
The generators {1, Φ(g), Φ(h)} ⊂ `∞ (C(S1 )) of can be lifted to 1 ∈ `∞ (T ) ⊂
`∞ (L(`2 )) and to, not necessarily commuting, positive contractions X, Y ∈ `∞ (T ) .
The positive contractions X = (x1 , x2 , . . .) and Y = (y1 , y2 , . . .) satisfy XY −Y X ∈
`∞ (K). It follows that the n-th entries with πK (xn ) = ϕn (g) and πK (yn ) = ϕn (h)
can be modified step by step by a compact perturbation with elements in 1 + K,
as considered and shown in Lemma B.23.1. We get new positive contractions sn
and tn with xn − sn , yn − tn ∈ K and ksn tn − tn sn k < 2−n . The new positive
contractions S := (s1 , s2 , . . .) and T := (t1 , t2 , . . .) in `∞ (T ) ⊂ `∞ (L(`2 )) have the
properties that S − X, T − Y ∈ `∞ (K),
ST − T S ∈ c0 (K) ⊂ `∞ (K) ⊂ `∞ (T ) ,
S + `∞ (K) = Φ(g) and T + `∞ (K) = Φ(h). The new positive contractions S +
c0 (K) and T + c0 (K) are commuting positive contractions in `∞ (T )/c0 (K) and
the commutative unital C *-algebra D ⊂ `∞ (T )/c0 (K) generated by them (and
1) maps onto the C *-algebra Φ(C([0, 1]2 )) ⊆ `∞ (C(S 1 )) in a way that S + c0 (K)
maps to Φ(g) and T + c0 (K) maps to Φ(h). By the injectivity of Φ, it follows from
Lemma B.23.3 that π`∞ (K) maps the commutative C *-algebra πc0 (K) (C ∗ (S, T, 1))
onto Φ(C([0, 1]2 )) in a way that the canonical that the described generators are
respected.
??????? Alternative description: ?????
Since ST − T S ∈ c0 (K) it follows that πc0 (K) (S) and πc0 (K) (T ) generate com-
mutative C *-subalgebra A of `∞ (T )/c0 (K) with the property that it is generated
by two commuting positive contraction, that gives
?????
It follows, that B := C ∗ (S, T, 1) + c0 (K) is a C *-algebra with the property
B/c0 (K) ∼
= C([0, 1]2 ), because C ∗ (S, T, 1) ∩ `∞ (K) = C ∗ (S, T, 1) ∩ c0 (K) .
To be checked if we have moreover ???
For all normal elements b ∈ B (respectively b ∈ C ∗ (S, T, 1) + `∞ (K)) holds that
the function f := πc0 (K) (b) (respectively the function f := π`∞ (K) (b)) in C([0, 1]2 )
has zero winding numbers, – with respect to alls closed curves γ : S 1 → [0, 1]2 in
[0, 1]2 and all points in C \ f (γ(S 1 )). (It is here not of interest if clockwise or
anti-clockwise parametrized.)
The extensions have the property that
f ∈ C([0, 1]2 ) ∼
= (C ∗ (S, T, 1) + `∞ (K))/`∞ (K) ⊆ `∞ (C(S 1 ))
has only zero winding numbers,
???? if and only if, ???? “if” is O.K.
???? But what about the converse ?:
???? Suppose that f has only zero winding
1278 B. EXACT C*-ALGEBRAS AND EXAMPLES

???? numbers. Then f can be approximately

???? factorized over 1-dimensional spaces:
???? Means if X is compact and metric space:
???? If f ∈ C(X) and ε > 0,
???? there exists g : X → Y
???? with Y 1-dimensional, and
???? h : Y → C with ???? kh ◦ g − f k < ε. ????
if there is a normal element in x ∈ `∞ (K) + C ∗ (S, T, 1), i.e., x satisfies x∗ x =
∗
xx , with f = π`∞ (K) (x) ( 6 ).
The latter is the case, if and only if, there exists, – for a given y ∈ C ∗ (S, T, 1) +
c0 (K) ⊂ `∞ (T ) with πc0 (K) (y) = f –, an element z ∈ `∞ (K) with the property
that z + y is normal: (z + y)∗ (z + y) = (z + y)(z + y)∗ . The element y satisfies
y ∗ y − yy ∗ ∈ c0 (K) by commutativity of (C ∗ (S, T, 1) + c0 (K))/c0 (K) .
(If one could here manage that y ∗ z, zy ∗ ∈ c0 (K), then one gets at least that
z ∗ z − zz ∗ ∈ c0 (K), but we don’t know that.)
It leads to the following question about contractions with index zero in the
Toeplitz algebra T :
Question B.23.4. Do we need for the determination of the winding numbers
only to look for invertible elements in C(S 1 )?

Let f ∈ C([0, 1]). We need for the conclusion that the restriction τA | C(f, 1)+
of the Aarnes quasi-state τA on C([0, 1])+ to the positive part of C(f, 1)+ is there
additive, the following property of the given f ∈ C([0, 1]):
For each continuous map γ : S 1 → [0, 1]2 and z 6∈ f (γ(S1 )) the winding number
w(f ◦ γ, z) is equal to zero.
It can happen that z ∈ f ◦ γ1 (S1 ) for some other Jordan curve γ1 (S1 ) ⊆ [0, 1]2
... So z can be in the spectrum of some other factor f ◦ γ1 (S1 ) ⊆ [0, 1].
But we have at least the positive direction: If there exists a normal element
g ∈ C ∗ (S, T, 1) + `∞ (K) with π`∞ (K) (g) = f then all winding numbers are zero.
Question: Suppose that one of it is not zero. Does there exist a non-additive
quasi-state on C(f, 1)+ ?????
Does there exist for an increasing function g ∈ C0 ((0, 1]) with the following
property?:
Let X ∈ T a contraction with index index(X) = 0 and kX ∗ X − XX ∗ k < ε
then there exists Y ∈ T with Y ∗ Y = Y Y ∗ , X − Y ∈ K and kX − Y k ≤ g(ε).
Can take here contraction X ∈ T with polar decomposition ZA = X for
positive self-adjoint A := (X ∗ X)1/2 ∈ T+ and Z ∈ L(H) an isometry with index
zero. It shows that one can replace Z by a unitary
6 We can here suppose that kxk = kf k = 1, because we can f multiply with a positive scalar

without changing the winding numbers of f and the normality of x, because all happens in the
commutative C *-algebra C ∗ (x, x∗ ) and its quotients.
 23. NON-2-SUB-ADDITIVE QUASI-TRACES EXIST. 1279

????. Get ???

Unfortunately, we do not know if this implies also the existence of a normal
element y ∈ C ∗ (S, T, 1) + c0 (K).
Is this is equivalent (!) to the property that all the indices of the f := πc0 (K) (y)
are zero?
It can be represented by a normal element in B [ or only that there is an element
x in `∞ (K) such that f + x is normal ].
[Need only that the indices vanish !!!! This should not change by compact
perturbations ...]
(This is true if we consider the bigger C ∗ (S, T ) + `∞ (K) instead.
Have then to show: If we have a normal representative in C ∗ (S, T ) + `∞ (K)
then we have a normal representative in B. )
??? Similar observations ?? show that the map µ := τA ◦ ϕ : B+ → [0, ∞) from
the free product B := C[0, 1] ∗ C[0, 1] into [0, ∞) is a unital quasi-trace that has
additive restrictions µ|C+ for each commutative C *-subalgebra C of B. It requires
only the additivity on C+ for all C := C ∗ (a + ib) with commuting a, b ∈ B+ .
Check next red and blue again! Look to the original notes and
compare.
The non-subadditive quasi-measure on µA on [0, 1]2 and the corresponding
non-2-sub-additive quasi-state τA on C([0, 1]2 )+ , given by J. Aarnes in [3], has
the property that there exists g, h ∈ C([0, 1]2 )+ with τA (g) = 0, τA (h) = 0 and
τ (g + h) > 2. And it has the – for us important – property that the restriction
τA |C ∗ (1, f )+ is additive, if and only if, the function f : [0, 1]2 → C has zero “winding
numbers” with respect to closed (Jordan) paths in P ⊂ [0, 1]2 and points in C\f (P ).
Now take the universal unital free product C *-algebra B := C([0, 1]) ∗ C([0, 1])
and the natural unital C *-algebra epimorphism ϕ : B → C([0, 1]2 ). One can show
that all images f = ϕ(b) ∈ C([0, 1]2 ) of normal elements, i.e., b ∈ B with b∗ b = bb∗ ,
have only zero winding numbers. (The proof uses the natural relation between
winding numbers of parametrized closed curves and the index of operators in the
Toeplitz algebra 0 → K(`2 ) → T → C(S 1 ) → 0 .) It implies that the restriction
µ|B+ of µ := τA ◦ ϕ is a unital quasi-trace of B = C([0, 1]) ∗ C([0, 1]). This µ can
not be 2-sub-additive because otherwise τA must be 2-sub-additive on C([0, 1]2 )+ ,
by using here that the set of elements x ∈ B with µ(x∗ x) = 0 must be a closed
ideal of B if µ is 2-sub-additive.
One can explore the relation between winding numbers of elements of C([0, 1]2 )
and indices of the Toeplitz algebra T even more and take a sequence ψn : S 1 →
[0, 1]2 of pice-wise smooth closed curves ψn : S 1 → [0, 1]2 ∼
= [0, 1] + i[0, 1] ⊆ C that
are uniformly dense in the set of all closed curves ψ : S → [0, 1]2 . It defines a
1

sequence of unital C *-epimorphisms ϕn : C([0, 1]2 ) → C(S 1 ) ∼ = T /K(`2 ).

1280 B. EXACT C*-ALGEBRAS AND EXAMPLES

The latter sequence ϕn can be explored to show that there exist l.s.c. quasi-
traces τ on certain type I C *-algebras A :
It is an extension 0 → J → A →φ C([0, 1]2 ) → 0 with an ideal J that is an
extension of C0 ((0, 1)) (or of C0 (0, 1] ?) by c0 (K), that has the property that µA ◦ φ
is not a 2-sub-additve, but is bounded and unital, is defined on all of A+ , and has
additive restrictions to each commutative C *-subalgebras of A. Moreover, there
exists no “bound” γ ∈ (0, ∞) with the property τ (a + b) ≤ γ · max(τ (a), τ (b)),
because it can happen that τ (a + b) > 0 – but µ(a) = 0 = µ(b) for some non-
commuting a, b ∈ A+ .
HERE is an older summary: J.F. Aarnes [3] described a ???????
There exists a unital extension 0 → c0 (K) → A → C([0, 1]2 ) → 0 such that τ
is given by τ = τA ◦ φ, where τA denotes the example of a globally non-additive
unital quasi-state on C([0, 1]2 ) that is additive on each C *-subalgebra C ∗ (f, 1)
where f : [0, 1]2 → C factorizes approximately over a tree (that is in this case of
J.F. Aarnes described in [3]).
It should be better to give a citation, not an explanation?!!
Much more easy to handle, despite not of of type I, is the unital free product
B := C[0, 1]∗C[0, 1] and the non-2-subadditive quasi-state τ0 := τA ◦π, where here π
denotes the natural C *-epimorphism π from B onto C([0, 1])⊗C[0, 1]) = C([0, 1]2 ) .
The key observation uses a suitable chosen unital *-monomorphism η from
C([0, 1]2 ) into

`∞ (C(S 1 )) ∼
= `∞ (T )/`∞ (K) ,

where here T := C ∗ (T ) ⊂ L(`2 ) means the Toeplitz algebra generated by the

forward shift T en = en+1 on `2 (N), and the defining countable family of maps
from C([0, 1]2 ) to C(S 1 ) are given by a uniformly dense sequence in all pice-wise
linear continuous maps (could also take smooth maps) from S 1 into [0, 1]2 = [0, 1] +
i[0, 1] ⊆ C. For each map γ : S 1 → [0, 1]2 define a unital C *-morphism γ
b : f 7→ f ◦ γ
All this C *-morphisms lift to unital C *-morphisms from B into T , and define a
unital C *-morphism from B into `∞ (T ).
Then the free product B = C[0, 1] ∗ C[0, 1] has an “universal lifting property”
that implies that all unital C *-morphisms h : B → C into the Calkin algebra C :=
L(`2 )/K(`2 ) lift to unital C *-morphisms H : B → L(`2 ).
By using the BDF theory, this implies that each image of each normal element
maps to an element with the property that all entries have zero indices ...
????
BDF = Brown Douglas Fillmore
supports an index argument for the characterization of normal operators (mod-
ulo compact perturbation) among the essentially normal operators:
 24. CENTRAL SEQUENCE ALGEBRAS, THE ε-TEST AND ITS APPLICATION. 1281

An operator S ∈ L(`2 ) is a compact perturbation of a normal operator, if and

only if, all indices of S − λ are equal to zero for all λ 6∈ Spec(πK (S)), cf. [207,
cor. IX.7.4].
This implies that all winding-numbers for π(S) ∈ C([0, 1]2 ) are zero.
The construction shows that a function f ∈ C([0, 1]2 ) is the image π(g) of
a normal element g ∈ B, if and only if, f can be uniformly approximated by
functions fn ∈ C([0, 1]2 ) that factorize over a finite tree Tn (which is then automatic
homotopic to the root of Tn ), i.e., fn = ϕn ◦ ψn with ψn : [0, 1]2 → Tn and
ϕn : Tn → [0, 1]2 , limn kfn − f k = 0 .
An important property of such tree-factorable functions fn is that they are
homotopic inside there images fn ([0, 1]2 ) ⊂ C to a constant map (or at least “inside
any neighbourhood of its image”, by a simpler proof).
It is not difficult to check that the Aarnes quasi-state µ0 on C([0, 1] × [0, 1])
becomes additive on the subalgebras C ∗ (f ) ⊂ C([0, 1]2 ) generated by f with this
special property (being a uniform limit of tree-factorable functions on [0, 1]2 ). Thus
the restriction of τ0 := µ0 ◦ π every abelian C *-subalgebra of B := C[0, 1] ∗ C[0, 1]
is additive. But µ0 ◦ π can not be additive, because π is surjective and the Aarnes
quasi-state µ0 is not additive on C([0, 1]2 ).
An application of quasi-diagonal elements shows that this implies the existence
of a unital type I C *-algebra A that has a non-additive quasi-state is an extension
c0 (K) → A → C0 ((0, 1], ψ(B)) by modification of a C *-morphism ψ from the uniti-
zation of C0 ((0, 1], B) into `∞ (T ) ⊆ `∞ (L(`2 )) constructed from the C *-morphism
that we have described above.
This equivalent properties of τ are long known for a sort of “normal” quasi-
states on finite AW*-algebras M , and the proof of the equivalence of the three
properties of τ goes over variants of suitable ultra-power constructions that give
on suitable hereditary C *-algebras D ⊆ A with τ |D+ bounded a C *-morphism
ϕ : D → M where M is a hereditary C *-subalgebra of an AW*-algebra such that
all maximal abelian C *-subalgebras of M are hereditary C *-subalgebras of Abelian
W*-algebras.

24. Central sequence algebras, the ε-test and its application.

The here given formulation and proof of the ε-test Lemma given in the Appen-
dix of [448] corrects also a really obvious(!) typo in the there given formulation
and proof of this Lemma.
To each free filter ω on the natural numbers N and to each C *-algebra A one
can associate the ultrapower Aω and the central sequence C *-algebra Aω ∩ A0 as
follows:
1282 B. EXACT C*-ALGEBRAS AND EXAMPLES

Let cω (A) denote the closed two-sided ideal of the C *-algebra `∞ (A) of bounded
sequences from A given by

cω (A) = (an )n≥1 ∈ `∞ (A) | lim kan k = 0 .


n→ω

We use the notation limn→ω αn (and sometimes just limω αn ) to denote the limit
of a sequence (αn )n≥1 along the filter ω. This limit exists for all free ultra-filters ω
on N.
The ultrapower Aω is defined to be the quotient C *-algebra `∞ (A)/cω (A); and
we denote by πω the quotient mapping `∞ (A) → Aω . Let ι : A → `∞ (A) denote
the ”diagonal” inclusion mapping ι(a) = (a, a, a, . . . ) ∈ `∞ (A), a ∈ A; and define
ιω := πω ◦ ι : A → Aω . Both mappings ι and ιω are injective. We shall often
suppress the mapping ιω and view A as a sub-C *-algebra of Aω . The relative
commutant, Aω ∩ A0 , then consists of elements of the form πω (a1 , a2 , a3 , . . . ), where
(an )n≥1 is a bounded asymptotically central sequence in A. The C *-algebra Aω ∩A0
is called a central sequence algebra.
We shall most often insist that the free filter ω is an ultrafilter, and we avoid
using A∞ := `∞ (A)/c0 (A) and A∞ := A0 ∩ A∞ , which have similar properties and
produce similar results (up to different selection procedures). One of the reasons is
that we need an epimorphism from Aω onto the W*-algebra N ω := `∞ (N )/cτ,ω (N )
(to be defined below), cf. Theorem B.24.3 below, where N is the weak closure of
A in the GNS representation determined by a tracial state τ on A. The sequence
algebra N ∞ := `∞ (N )/cτ,0 (N ) (with cτ,0 (N ) the bounded sequences in N with
limn kak2,τ = 0) is not a W*-algebra. Another reason for preferring free ultra-
filters to general free filters is that, for A 6= C, simplicity of Aω is equivalent to pure
infiniteness and simplicity of A. The algebras Aω are the fibers of the continuous
field A∞ with base space β(N) \ N. The structure of this bundle appears to be
complicated.
Recall from [448] that if B is a C *-algebra and if A is a separable sub-C *-
algebra of Bω , then we define the central sequence algebra

F (A, B) := (A0 ∩ Bω )/Ann(A, Bω ).

This C *-algebra has interesting properties. For example, F (A, B) := F (A ⊗ K, B ⊗

K). We let F (A) := F (A, A). The C *-algebra F (A) is unital if A is σ-unital. If
A is unital, then F (A) = A0 ∩ Aω . We refer to [448] for a detailed account on
the C *-algebras F (A, B) and F (A). We shall often, in the unital case, denote the
central sequence algebra Aω ∩ A0 by F (A).
We have the following useful selection principle for sequences in the filter ω
with a countable number conditions from [448, Lemma A.1]. For completeness we
add a proof (7).

7 We correct here a misleading typo in the original proof given in [448]!.

 24. CENTRAL SEQUENCE ALGEBRAS, THE ε-TEST AND ITS APPLICATION. 1283

Lemma B.24.1 (The ε-test). Let ω be a free ultrafilter. Let X1 , X2 , . . . be any

(k)
sequence of sets and suppose that, for each k ∈ N, we are given a sequence (fn )n≥1
(k)
of functions fn : Xn → [0, ∞].
(k) Q∞
For each k ∈ N define a new function fω : n=1 Xn → [0, ∞) by
∞
Y
fω(k) (s1 , s2 , . . .) := lim fn(k) (sn ), (sn )n≥1 ∈ Xn .
n→ω
n=1

Suppose that, for each m ∈ N and each ε > 0, there exists s = (s1 , s2 , . . .) ∈
Q∞ (k)
n=1 Xn such that fω (s) < ε for k = 1, 2, . . . , m. It follows that there is t =
Q∞ (k)
(t1 , t2 , . . .) ∈ n=1 Xn with fω (t) = 0 for all k ∈ N.

Proof. For each n ∈ N define a decreasing sequence (Xn,m )m≥0 of subsets of

Xn by Xn,0 = Xn and

Xn,m = s ∈ Xn ; max{fn(1) (s), . . . , fn(m) (s)} < 1/m ,



for m ≥ 1. We let m(n) := sup{m ≤ n ; Xn,m 6= ∅}; and for each integer k ≥ 1,
let Yk := {n ∈ N ; k ≤ m(n)}. Fix some k ≥ 1. By assumption there exists
Q (j)
s = (sn ) ∈ n Xn such that fω (s) < 1/k for 1 ≤ j ≤ k. This entails that
(j)
there exists a set Zk ∈ ω such that fn (sn ) < 1/k for 1 ≤ j ≤ k and for all
n ∈ Zk . This again implies that Xn,k 6= ∅ for all n ∈ Zk ; which finally shows that
m(n) ≥ min{k, n} for all n ∈ Zk . It follows that Zk \ {1, 2, . . . , k − 1} ⊆ Yk , from
which we conclude that Yk ∈ ω (because ω is assumed to be free). Now,
1 1 1 1
lim = lim inf ≤ inf sup ≤ inf = 0.
n→ω m(n) n→ω m(n) k n∈Yk m(n) k k

By definition of m(n) we can find tn ∈ Xn,m(n) ⊆ Xn for each n ∈ N. Put

(k)
t = (tn )n≥1 . Then fn (tn ) ≤ 1/m(n) for all k < m(n) by definition of Xn,m(n) , so
1
fω(k) (t) = lim fn(k) (tn ) ≤ lim = 0,
n→ω n→ω m(n)
for all k ≥ 1 as desired. 

Let N be a W*-algebra with separable predual and let τ be a faithful normal tracial
state on N . Consider the associated norm, kak2,τ = τ (a∗ a)1/2 , a ∈ N , on N . Let
N ω denote the W*-algebra `∞ (N )/cω,τ (N ), where cω,τ (N ) consists of the bounded
sequences (a1 , a2 , · · · ) with limω kan k2,τ = 0. (As mentioned above, if ω is a free
filter, which is not an ultrafilter, then N ω is not a W*-algebra.)

Remark B.24.2. Since the pre-dual N∗ of the W*-algebra N is separable,

the algebra N 0 ∩ N ω contains a copy of the hyper-finite II1 factor R with sepa-
rable predual, if and only if, for each k ∈ N there exists a sequence of unital *-
homomorphisms ψn : Mk → N such that limn→∞ k[ψn (t), a]k2,τ = 0 for all t ∈ Mk
and a ∈ N , if and only if, there is a ∗ -homomorphism ϕ : M2 → N 0 ∩ N ω such that
t ∈ N 7→ t · ϕ(1) ∈ N ω is faithful. (It suffices to consider elements t in the centre
of N .)
1284 B. EXACT C*-ALGEBRAS AND EXAMPLES

This equivalence was shown by Dusa McDuff, [550], in the case where N is a
factor. She gets moreover that N ∼= N ⊗R, if and only if, N has a central sequence
that is not hyper-central. Such a II1 factor is called a McDuff factor.

The result below was proved by Y. Sato in [707, Lemma 2.1] in the case where A
is nuclear. We give here an elementary proof of this useful result, that does not
assume nuclearity of A. The result implies that the central sequence C *-algebra
Aω ∩ A0 has a subquotient isomorphic to the hyperfinite II1 factor R whenever
A has a factorial trace so that the corresponding II1 factor, arising from GNS
representation with respect to that trace, is a McDuff factor. In Remark ?? in the
next section we show how one can give an easier proof of the theorem below using
that the kernel Jτ of the natural *-morphism Aω → N ω is a so-called σ-ideal. The
proof given below does not (explicitly) use σ-ideals.

Theorem B.24.3. Let A be a separable unital C*-algebra, τ be a faithful tracial

state on A, let N be the weak closure of A under the GNS representation of A with
respect to the state τ , and let ω be a free ultrafilter on N. It follows that the natural
morphisms
Aω → N ω , A 0 ∩ Aω → N 0 ∩ N ω
are surjective.

Proof. Let πA and πN denote the quotient mappings `∞ (A) → Aω and

`∞ (N ) → N ω , respectively. Denote the canonical map Aω → N ω by Φ, and
e : `∞ (A) → Nω denote the map Φ ◦ πA .
let Φ
We show that Φ : Aω → N ω is surjective: If x = πN (x1 , x2 , . . . ) is an element
in N ω , then, by Kaplansky density theorem, there exists ak ∈ A with kak k ≤
kxk k and with kak − xk k2,τ ≤ 1/k. It follows that (a1 , a2 , . . . ) ∈ `∞ (A) and that
Φ(a
e 1 , a2 , . . . ) = x.

To prove that the natural map A0 ∩ Aω → N 0 ∩ N ω is surjective, it suffices to

show that if b = (b1 , b2 , . . . ) ∈ `∞ (A) is such that Φ(b)
e ∈ N ω ∩ N 0 , then there is
an element c ∈ `∞ (A) such that πA (c) ∈ Aω ∩ A0 and Φ(c) e = Φ(b).
e Let such a
∞ ∗ ∞ −1
b ∈ ` (A) be given, put B = C (A, b) ⊆ ` (A), and put J = (Φ) (0) ∩ B. Notice
e
that an element x = (x1 , x2 , . . . ) ∈ `∞ (A) belongs to the kernel (Φ)e −1 (0) of Φe if
and only if limn→ω kxn k2,τ = 0. Notice also that ba − ab ∈ J for all a ∈ A.
Let (d(k) )k≥1 be an increasing approximate unit for J consisting of positive con-
tractions which is asymptotically central with respect to the separable C *-algebra
B. Then

0 = lim k(1 − d(k) ) ba − ab (1 − d(k) )k

k→∞

= lim k(1 − d(k) )b(1 − d(k) )a − a(1 − d(k) )b(1 − d(k) )k

k→∞

for all a ∈ A.
We use the ε-test (Lemma B.24.1) to complete the proof:
Let (ak )k≥1 be a dense sequence in A. Let each Xn be the set of positive contractions
 24. CENTRAL SEQUENCE ALGEBRAS, THE ε-TEST AND ITS APPLICATION. 1285

(k)
in A. Define fn : Xn → [0, ∞) by
fn(1) (x) = kxk2,τ , fn(k+1) (x) = k(1 − x)bn (1 − x)ak − ak (1 − x)bn (1 − x)k, k ≥ 1 .
(1) (`)
Notice that fω (d(`) ) = limn→ω kdn k2,τ = 0 for all ` because each d(`) belongs to
J. Note also that
fω(k) (d(`) ) = k(1 − d(`) )b(1 − d(`) )ak − ak (1 − d(`) )b(1 − d(`) )k.
It is now easy to see that the ε-test in Lemma B.24.1 is satisfied, so there exists a
(k)
sequence d = (dn )n≥1 of positive contractions in A such that fω (d) = 0 for all k.
(1)
As fω (d) = 0 we conclude that Φ(d)
e = 0.
(k)
Put c = (1 − d)b(1 − d). Then Φ(ce − b) = 0, so Φ(c)
e = Φ(b).
e Since fω (d) = 0,
we see that cak − ak c = 0 for all k ≥ 2. This shows that πA (c) ∈ Aω ∩ A0 . 

Correction to:
‘The UCT, the Milnor Sequence, and a Canonical Decomposition of the Kas-
parov Groups’ Author: Schochet C.L.
Source: K-theory, Volume 14, Number 2, June 1998 , pp. 197-199(3)
Publisher: Springer
Abstract: In this note we correct a mistake in K-Theory 10 (1996), 49–72. In
that paper we asserted that under bootstrap hypotheses the short exact sequence
(injective at the beginning and surjective at the end):

lim1 HomZ (K∗ (Ai ), K∗ (B)) → Ext1Z (K∗ (A), K∗ (B)) → lim Ext1Z (K∗ (Ai ), K∗ (B))
←− ←−

which arises in the computation of KK∗ (A, B) is a split sequence. This is not
always the case. Thus KK∗ (A, B) decomposes into the three components
HomZ (K∗ (A), K∗ (B)), lim Ext1Z (K∗ (Ai ), K∗ (B))
←−
and
lim1 HomZ (K∗ (Ai ), K∗ (B)).
←−

However, this is a decomposition in the sense of composition series, not as three

direct summands. The same correction applies to the Milnor sequence. If there is
no prime p for which both K∗ (A) and K∗ (B) have p-torsion then the decomposi-
tion is indeed as direct summands. The other results of the paper are unaffected.
Keywords: KK-theory; Kasparov groups; Universal Coefficient Theorem; Milnor
sequence; KK-filtration; fine structure
Language: English Document Type: Regular paper
Affiliations: 1: Mathematics Department, Wayne State UniversityDetroit, MI
48202, U.S.A. E-mail: [email protected]
 APPENDIX C

Temporary (!!) monitor and To-Do List

1. Circulation of Changes

We still have no example of a nuclear separable C *-algebra that is purely

infinite but is not strongly purely infinite !!! (2014, first discussed ?).
On simple C*-algebras A this two properties are the same! (For obvious reason!)
Def. equivalent to ”purely infinite” (p.i.):
The element a ∈ A+ is ”purely infinite” if a ⊕ a - a (inside M2 (A) and means
there that diag(a, a) - diag(a, 0) ). Here one can take also non-positive a ∈ A with
a ⊕ a - a, because a ≈ a∗ a ≈ (a∗ a)∗ = a(a∗ ).
This is equivalent to the property of a ∈ A that If b ∈ A+ and b 6= 0 is in the
closed ideal I(a) of A generated by a, then b ⊕ a - a.
( This is equivalent to a ⊕ a - a and is the original definition of a being ”p.i.”
given by J. Cuntz ! Give here precise Citation of Definition in paper of Cuntz!!! )
The Definition C.1.1 is equivalent to a ⊕ a - a for all a ∈ A+ .
In general a ∈ A+ is called ”infinite” if there exists non-zero x ∈ A with
x ⊕ a - a.
Lemma:
, Let A a C*-algebra and X ⊆ A a countable subset of A. The there exists a
( separable) C*-subalgebra B of A with the properties that X ⊆ B and for all
elements c, d ∈ K(`2 ) ⊗ B holds:
c - d in K(`2 ) ⊗ B, if and only if, c - d in K(`2 ) ⊗ A.
This separable C*-subalgebra B of A is mostly much bigger then the C*-
subalgebra generated by the separable X, but B is still a separable C*-algebra.
Moreover one can manage (by suitable enlarging of B by its construction) that
B is nuclear if A is nuclear.
(But notice here that exactness is anyway a hereditary property! And remind
that exactness is the same as Subnuclearity.) Give here citation!
This implies not only that elements of B are infinite (resp. purely infinite) in
B, if and only if, this elements are infinite (resp. purely infinite) in A (which is
usually a weaker property). This property of B implies also that each closed ideal
J of B is the intersection I ∩ B = J

1287
1288 C. TEMPORARY (!!) MONITOR AND TO-DO LIST

Definition C.1.1. A (non-zero) C*-algebra A is called purely infinite, if and

only if,

(i) The C*-algebra A has no characters, and

(ii) for every element a, b ∈ A+ \ 0 with b in the closed ideal I(a) of A, gener-
ated by a, there is a sequence of elements c1 , c2 , . . . ∈ A with limn c∗n acn =
b. (In the sense limn kb − c∗n acn k = 0.)

Observations concerning Definition C.1.1:

Part (ii) of the Definition C.1.1 of purely infinite (p.i.) C*-algebras A (also
called ”properly infinite” C*-algebras ?) is equivalent to:
For each d1 , d2 ∈ A, a ∈ A+ and ε > 0 there exists d3 ∈ A with
kd∗1 ad1 + d∗2 ad2 − d∗3 ad3 k < ε .
This equivalence can be shown by complete induction, because each element b ∈
I(a)+ for the closed ideal I(a) of A generated by a can be approximated arbitrary
near by elements of the kind
e∗1 ae1 + e∗2 ae2 + · · · + e∗n aen ∈ I(a)+ .

We discuss here the ”permanence properties” of this definition of purely infinite

(= properly infinite ?) C*-algebras:

Proposition C.1.2. The class of purely infinite C*-algebras A as defined in

Definition C.1.1 has following permanence properties:

(i) Quotients, ideals, and hereditary C*-subalgebras,

(ii) Inductive limits of purely infinite C*-algebras,
(This is still not proven !!! But every positive element a ∈ A+ of
an inductive limit is the limit of a sequence of properly infinite positive
elements. It seems that the limit of p.i. elements is p.i., which gives the
proof.)
(iii) Every separable subset X of A is contained in a separable purely infinite
C*-subalgebra B of A. Moreover one can find such B with the property
that each ideal J of B is the intersection IcapB = J of a closed ideal of
B.
(Then X contained in B, that can be much bigger than C ∗ (X) ⊆ A.
Can we manage to find such a C*-subalgebra B that is in addition
relatively weakly injective in A? Unfortunately: It is Not known!?)

( But what about extensions? )

Observations:
Question: Can we prove that (all sorts of) purely infinite C*-algebras have the
Cuntz property:
For all a ∈ A+ holds a ⊕ a - a.
This property should work also for inductive limits.
 1. CIRCULATION OF CHANGES 1289

If this is shown than one must show that Cuntz-pi of A implies that A is a p.i.
C*-algebra.
(But this depends from the definition of ”p.i.” etc.)
(It is then to show that hereditary C*-subalgebras of D of A have no characters.
But that is equivalent to the property that ρ(A) ∩ K(H) = {0} for each irreducible
representation ρ : A 7→ L(H) of A holds that ρ(A) ∩ K(H) = 0.)

Remark C.1.3. Notice that, moreover, one can manage to find separable C*-
subalgebras Aτ of A in part (iv) of Proposition C.1.2 that are in addition relatively
weakly injective in A.
This follows from the general observation that each separable C*-subalgebra
B of A is contained in a separable C*-subalgebra D of A that is relatively weakly
injective in A, i.e. there exist a completely positive contraction E : A → D∗∗ with
E(d) = d for all d ∈ D.
Where is this (= the latter) shown ?

Proof. Ad(i): Suppose that A is a p.i. C*-algebra, and J ⊂ A a (non-zero)

closed ideal. Then J and A/J can not have no characters.
(See considerations below.)
The part (ii) of Definition C.1.1 is moreover shown for all hereditary C*-
subalgebras.
Ad(ii): The case of inductive limits:
Since, by part (i), the properties of purely infinite C*-algebras pass to (non-
zero) quotients we can suppose that A is the inductive limit of C*-subalgebras
Aτ ⊆ A of A and the Aτ are purely infinite.
If A is a C*-algebra with a character ξ then every (e.g. separable) C*-
subalgebra B of A is in the kernel of ξ or the restriction ξ|B to B is a character.
Thus, if A is an inductive limit of C*-algebras Aτ without characters, then A has
no character.
Suppose that A is an inductive limit of an upward directed net of C*-
subalgebras Aτ ⊆ A and each Aτ is purely infinite. Then for each a, b ∈ A+ with
the property that b is contained in the closed ideal J of A generated by a, and
each chosen δ ∈ (0, 1) (small enough) there exists n ∈ N and d1 , . . . , dn ∈ A with
kb − k d∗k adk k < δ.
P

The problem is about the estimates ...

can be simplified if we use that we must show only that there exists for each
d1 , d2 ∈ A and ε > 0 an element d3 ∈ A with the property that

kd∗3 ad3 − d∗1 ad1 − d∗2 ad2 k < ε .

1290 C. TEMPORARY (!!) MONITOR AND TO-DO LIST

Let γ > 0 (sufficiently small) and find b, e1 , e2 , e3 ∈ Aτ , with norms of a − b,

dj − ej (j = 1, 2) with norms below γ, and

ke∗3 be3 − e∗1 be1 − e∗2 be2 k < γ .

then ?????????
??? This gives ???:
For each b ∈ A+ and each positive element c of the closed ideal J generated by
b there exists a sequence of elements xn ∈ A such that kc − x∗n bxn k converges to
zero.
Have the same problem: norm- estimate ke∗3 be3 − e∗3 ae3 k and a way to come
from a to b ...
dj − ej for j = 1, 2, a − b have to be estimated:
Now we take γ > 0 small enough such that
e − b, a − f , gk − dk
ke − bk , ka − f k, kgk − dk k < γ can be chosen arbitrary small?
Find Aτ ⊆ A, such that it contains ????
Find h ∈ Aτ with kh∗ f h − k gk∗ f gk k (arbitrary small !) say < µ for some
P

before given µ > 0.

The problem is: khk and kh∗ f h − h∗ ahk <???.
???? we can find Aτ ⊆ A and e, f, ? ∈ Aτ ????
Has to be studied more careful !!! It seems that the cut-down possibility (a−ε)+
plays a role ...
(1) If J ⊆ A is a nonzero closed ideal of A with J 6= A then J and A/J have
no characters:
Obviously all (closed) ideals J of A and quotients A/J have no characters if A
has no characters, because each character ξ of A/J defines the (non-zero) character
a 7→ χ(a) := ξ(πJ (a)) on A. But this can not exist by property (i) of A.
And if η is a character of J with kernel K. Then K is a closed ideal of A and
J/K ⊆ A/K is a closed ideal of A/K that is one-dimensional. Thus, J/K = C · p
for some projection p ∈ A/K. Since J/K is a closed ideal of A/K this implies that
x 7→ xp ∈ C · p for x ∈ A/K defines a character on A/K. Above we have seen that
the quotients A/K can not have a character. Thus, J can not have a character.
So far we have seen that closed ideals and quotients of A can not have a
character.
(The question about characters of hereditary C*-subalgebras of A will be con-
sidered further below in ????? .... )
Here is the general observation concerning this direction:
 1. CIRCULATION OF CHANGES 1291

Lemma:
Let D ⊆ A a hereditary C*-subalgebra of A (i.e. D = DAD = DAD) and I ⊆ D
a closed ideal of D. Then the closed ideal J ⊆ A of A generated by the hereditary
C*-subalgebra I of A has the property that J ∩ D = I.
Remark:
The same happens with von-Neumann algebras M : Let P ∗ P = P ∈ M a projection
and Q a projection in the centrum c(P M P ) of P M P . Then let R in the centre c(M )
of M the smallest projection with the property RQ = Q, which means equivalently
that RM = RM R is the weak closure (i.e. the σ(M, M∗ )-closure) of the linear span
of M QM .
Then R is the smallest projection in the centrum c(M ) of M with the property
that RQ = Q, and it turns out that R, P and Q satisfy that RP = Q.
(3) Now let D ⊆ A a (non-zero) hereditary C*-sub-algebra of A. Suppose
that D has a character, say η. Let I ⊆ D denote the kernel-ideal of η. Then the
closed ideal J of A generated by I has the property that J ∩ D = I (It is an exercise
Lemma! Cited above.) and that – therefore – C ∼ = D/I ⊆ A/J is a one-dimensional
hereditary C*-subalgebra of A/J that is naturally isomorphic to C
and is needed for what ???
Thus, A/F contains an isomorphic image of some algebra of compact compact
operators ... on a Hilbert space (of dimension ≥ 1). But this was excluded by the
requirements that no hereditary C*-sub-algebra of A has a character.
(0) The elements cn in Part (ii) can be taken in aAb because the Part (ii) of
this definition can be expressed equivalently:
If nonzero b ∈ A+ is contained in the closed ideal of A generated by a ∈ A+
then, for every 0 < ε < kbk, there exists δ > 0 and d ∈ A with d∗ (a−δ)+ d = (b−ε)+ .
In particular, the Part(ii) passes automatically to all (non-zero) hereditary C*-
subalgebras D ⊆ A of A, i.e., if A is p.i., then D satisfies Part (ii) of Definition
C.1.1.
Let J ⊆ A a closed ideal and x, y ∈ (A/J)+ such that y is in the closed ideal
of x ∈ (A/J)+ , and ε ∈ (0, kyk).
Then there exist elements a ∈ A+ and d1 , . . . dn ∈ A with πJ (a) = x and
kπJ (b) − yk < δ for b := d∗1 ad1 + · · · + d∗1 ad1 and δ := ε/3. By Part (ii) of Definition
C.1.1 there exists c ∈ A with kc ∗ ac − bk < δ. Thus, kπJ (c)∗ xπJ (c) − yk < ε .
The Parts (i) is also satisfied for D: Suppose that D has a character φ, then
there exists a contraction e ∈ D+ with φ(e) = 1. It extends to a pure state ρ on A.
(Because in general states of D extend to states on A. Then the extreme points of
this extensions turn out to be pure state on A.)
And then ???
x, y ∈ (A/J)+ 
1292 C. TEMPORARY (!!) MONITOR AND TO-DO LIST

(XXXX????)
It is not clear if strong pure infiniteness and the pure infiniteness are really
different !!!
Here is a definition that is equivalent to ”strongly purely infinite” (s.p.i.):
If S = [aj,k ] ∈ M2 (A)+ then there exists a sequence of diagonal matrices
Tn = sn ⊕tn ∈ M2 (A) such that Tn∗ STn converges to the diagonal matrix a1,1 ⊕a2,2 .
(Gives s∗n a1,1 sn ≈ a1,1 , t∗n a2,2 tn ≈ a2,2 , s∗n a1,2 tn ≈ 0, and t∗n a2,1 sn ≈ 0.)
Gives with aj,k := a for all cases j, k ∈ {1, 2} that a is properly infinite in A,
because s∗n asn ≈ a, t∗n atn ≈ a s∗n atn ≈ 0. It is equivalent a ⊕ a - a, by considering
the case a1,1 := a, ai,j = 0 for (i, j) 6= 0.
Need to show my and others notations and definitions!
Definition of: ”approximately divisible” ?? where ??? Def. of: ”???” etc.
Some old citation changes:
Lemma ?? (Old Ref. lem:2.pure.state.excision ) was A.1.24 or: A.old.2.4 ??
Lemma ?? (Old Ref. lem:2.pure.state.excision ) or: A.old.2.4 also in old: A.8.1
Only 1-cited in proof of Lemma 2.2.3.
Lemma A.6.1 old A.1.9.
Lemmas = Lemmata ?? – ??? now named ??? (Old Ref. lem:2.pure.state.excision
)
New 2.1.22 = old.. lem:A.old.3.4c, also ??? old A.1.24 (=lem:A.old.2.4) - old
A.1.28,

2. Proposed changes (1)

Change it now, because it can produce misunderstandings !

1.) General notations for matrices (to be decided and then changed where
different notation was used):
Adjust in all chapters the use of

Mm,n (A) ∼
= A ⊗ Mm,n (C) ∼
= Mm,n (C) ⊗ A

by this rules:
The columns with n-entries should be in (1, n) -matrices ... ∈ Mn,1 (A) ... not
in M1,n (A) !!!
Lin. Alg. book of Boseck S.76: Matrix of type (m, n) has m rows and n
columns. Set of this matrices denoted by Am,n . Am,r · Ar,n ⊂ Am,n .
M. Koecher: K (m,n) set of matrices with m rows and n columns. K (m,r) ·
K (r,n) ⊂ K (m,n) .
H. Anton: m × n-matrix has m rows and n columns. m × r-matrix multiplied
with r × n-matrix is m × n-matrix. Transposed matrix of A denoted by AT .
 2. PROPOSED CHANGES (1) 1293

H.-J. Kowalsky: (m, n) -matrix has m rows and n columns.

G. Strang: m × n-matrix has m rows and n columns:
All have first rows: a1,1 , a1,2 , a1,3 , ... next row: a2,1 , a2,2 , a2,3 , ...
TO DO LIST: from Nov. 2014!!! Not done July/Aug 2021!!!
Adjust in all chapters of the book the use of ∼ and ≈ !!!
And the same with [a]∼ , hai∼ , [a]≈ , hai≈ .
Also adjust all the matrix calculations to the above
cited terminology. ??
Chp 1. Careful check of Introduction.

Ad Chp.2:
Ad 2.1. Shorten: bundle version of C(X, A) with simple A.
Ad 2.2. Other parts only as overview article.

Ad Chp. 3. Check what is really needed later.

It would be good if we could show that:
Question:
Let A denote a separable exact unital C *-algebra and consider a unital nuclear
c.p. contraction V : A → M(O2 ⊗ K) with V (a∗ a) − V (a)∗ V (a) ∈ O2 ⊗ K.
Find a c.p. extension W : A⊗O2 → M(O2 ⊗K) with W (a⊗1)−V (a) ∈ O2 ⊗K
for a ∈ A and W (x∗ x) − W (x)∗ W (x) ∈ O2 ⊗ K for x ∈ A ⊗ O2 .
Perhaps, we can replace O2 by the infinite tensor product O2 ⊗ O2 ⊗ · · · and
try to adjust V with help of suitable elements of 1 + O2 ⊗ K ... ???
In particular it should be interesting for nuclear A. (It is true if A is in
the UCT-class. Perhaps here is a critical point if canceling the important UCT-
assumption.)
( < −− Why it is true in UCT-class ?)
For exact A the question is less interesting because we can replace A by A ⊗ O2
at beginning of the embedding-proof for exact A in O2 .
Since O2 ∼ = O2 ⊗ O2 ⊗ · · · , it is likely that O2 has the property that any
separable C *-subalgebra of Qs (O2 ) commutes with a unital copy of O2 in Qs (O2 ).
(Qs (O2 ) means here the ”stable” corona M(O2 ⊗ K)/O2 ⊗ K of O2 ?)
It could replace homotopy invariance considerations, because otherwise one gets
only unital embedding of A ⊗ O2 that are “extremal” in O2 for nuclear A but are
not extremal for A itself. (What means here “extremal”?)
It seems to be necessary to study following questions:
1294 C. TEMPORARY (!!) MONITOR AND TO-DO LIST

Question: If A is a separable unital nuclear C *-algebra and ϕ : A → Qs (O2 )

a unital *-monomorphism where

Qs (O2 ) := Q(O2 ⊗ K) := M(O2 ⊗ K)/(O2 ⊗ K)

It is likely that ϕ extends to ψ : A ⊗ O2 → Qs (O2 ) with ψ(a ⊗ 1) = ϕ(a).

”Likely” only ???
Look what happens with the unital c.p. maps V from Mn to Qs (O2 ). Try to
find an ”arbitrary” small perturbation W of V that commutes with a unital copy
of O2 in Qs (O2 ).
Alternatively we could proof that:
Nuclear *-monomorphisms φ of exact unital separable A into s.p.i. “corona”
spaces C extends to A ⊗ O∞ , i.e., that φ(A)0 ∩ C has a properly infinite unit.
(Where this is proved? Could prove?)
(What about c.p. maps from Mn to C? By doubling the rows that define this
maps ?)
Should be a consequence of a conclusion of a generalized Weyl–von-Neumann
theorem that could prove that φ dominates φ ⊕ φ. If C is s.p.i., e.g. in the case
where C is the corona C = Qs (B) of a σ-unital strongly p.i. C *-algebra B.
It gives that φ extends at least to ψ : A ⊗ O∞ → C with ψ(a ⊗ 1) = φ(a).
Remind here the definition of the Cuntz algebra:

O∞ := C ∗ (s1 , s2 , . . . ; s∗j sk = δjk 1 ) .

Then ψ(1⊗p) for p = 1−s1 s∗1 is the unit of a copy of O2 in pO∞ p, and q := ψ(1⊗p)
is a full properly infinite element of C with 0 = [q] ∈ K0 (C). (The embedding γ of
O2 in pO∞ p will be denoted by γ : b ∈ O2 7→ γ(b) ∈ pO∞ p and satisfies γ(1) = p.)
If, in addition, the algebra C has a properly infinite unit 1 with 0 = [1] ∈ K0 (C)
(– as it is the case e.g. for all stable coronas C := Qs (B) or for quotients C of
Qs (B)∞ –) then there is an isometry T ∈ C with T T ∗ = q. (In fact, the requirement
that ψ(1 ⊗ p) is full and properly infinite implies that 1 is properly infinite in C.)
We can define a new unital ψ0 : A ⊗ O2 → C by ψ0 (a ⊗ b) := T ∗ ψ(a ⊗ γ(b))T .
Then a 7→ φ0 (a) := ψ0 (a ⊗ 1) is a nuclear unital *-monomorphism that com-
mutes with a copy of O2 unitally contained in φ0 (A)0 ∩ C.
We can define φ back from this map by

φ(a) = ψ(a⊗1) = ψ(1⊗s2 )∗ ψ(a⊗p)ψ(1⊗s2 ) = (ψ(1⊗s2 )∗ T )ψ0 (a⊗1)(T ∗ ψ(1⊗s2 )) .

Notice that V := T ∗ ψ(1 ⊗ s2 ) is an isometry with 1 − V V ∗ full and properly infinite

in O2 .
The remaining problem is the following:
Suppose that the “liftable” copy (in sense of given split extension) λ : A⊗O2 →
???? ???
 2. PROPOSED CHANGES (1) 1295

Now suppose that there exists a (“liftable”) nuclear unital *-monomorphism

λ : A ⊗ O2 → C with the property that there exists sequences of isometries t1 , t2 , . . .
and r1 , r2 , . . . such that, for all a ∈ A, λ(a ⊗ 1) = limn t∗n φ0 (a)tn and φ0 (a) =
limn rn∗ λ(a ⊗ 1)rn .
Apply a “fitting together” procedure to get with properties of the “corona
algebra” that there are isometries t and r with r∗ λ(a ⊗ 1)r = φ0 (a) and s∗ φ0 (a)s =
λ(a ⊗ 1) for a ∈ A. Since both commute with (different ) unital copies of O2 in
its commutants it follows from Corollary 4.3.7 that φ0 and λ((·) ⊗ 1) are unitary
equivalent in C. (If C is a stable corona, then this equivalence is given by a product
of exponentials.)
In particular: check the material of concurring sections.
Use the beamer presentations (Muenster !!! details???), compare with its con-
tents.
Add needed case:

A ⊆ M(D), D ⊆ M(B) non-degenerate (necessary ??? only for M(D) ⊆

M(B), could be e.g. enough that DBD is a corner of B), D s.p.i.)
Then residually nuclear V : A → B
– with respect to the action

J ∈ I(B) 7→ A ∩ M(D, D ∩ M(B, J))

–, is 1-step approximately inner in M(B).

(Or is section 2.2 ?? of Chp.???? 2? the right place for the proof of this?).

We could use the minimality of the m.o.c.c. of residually nuclear c.p. maps
among the cones with same invariant ideals:
It is enough to consider separable A, because non-separable A is the inductive
limit of a net of separable C *-sub-algebras Aτ ⊆ A with the following additional
properties:
For each countable subset X ⊆ A there is an separable Aτ with X ⊆ Aτ .
Each closed ideal of Aτ is the intersection of a closed ideal of A with Aτ , and Aτ
is residually relatively weakly injective in A in the sense that (Aτ /(Aτ ∩ J)) ⊗max F
is naturally a C *-subalgebra of (A/J) ⊗max F for each C *-algebra F .
(It seems to turn out that it suffices to consider here only separable C *-algebras
F ?)
If A s.p.i. then Aτ can be chosen such that Aτ is in addition s.p.i.
It implies that the restriction V |Aτ of V to Aτ of each residually nuclear map
V : A → B with respect to some l.s.c. action Ψ of Prim B on A is also residually
nuclear with respect to the induced action of Prim B on Aτ , i.e., for I := Ψ(J) and
Iτ := I ∩ Aτ := Ψτ (J) holds: V (Iτ ) ⊆ J and [V |Aτ ] : Aτ /Iτ → B/J is nuclear,
1296 C. TEMPORARY (!!) MONITOR AND TO-DO LIST

because [V |Aτ ] is equivalent to the restriction of [V ] : A/I → B/J to πI (Aτ ) ⊆ A/I

by the isomorphism [πI ] : Aτ /Iτ → πI (Aτ ).
If Bτ is a separable C *-subalgebra of B such that V (Aτ ) ⊆ Bτ , “symbolic”
invariance HΨ (Aτ )Bτ ⊗ K ⊆ Bτ ⊗ K, each closed ideal of Bτ is the intersection
of a closed ideal of B with Bτ and Bτ is residually weakly injective in B, then
V |Aτ : Aτ → Bτ is still residually nuclear with respect to the “induced” action
Φ : I(Bτ ) → I(A) given by Φ(J) := Ψ(BJB) ∩ Aτ .
Here we simplify notation:
BJB denotes the closed linear span of the set of products b1 cb2 with c ∈ J and
b1 , b2 ∈ B. Indeed, J = Bτ ∩ BJB,

V (Φ(J)) ⊆ V (Aτ ) ∩ BJB ⊆ Bτ ∩ BJB = J

and [V ] : Aτ /Φ(J) → Bτ /J is nuclear, because V (Aτ ) ⊆ Bτ , [V ] : Aτ /Φ(J) =

Aτ /(Aτ ∩ BJB) → B/BJB is nuclear, and there is a natural isomorphism from
Bτ /J onto π(Bτ ) ⊆ B/BJB which is relatively weakly injective in B/BJB by
assumption.
Another question is to find suitable separable C *-sub-algebras Bτ of B such
that for a given action H(Aτ )Bτ ⊆ Bτ and, for J ∈ I(B),

(H|Aτ )−1 (M(Bτ , J ∩ Bτ )) = Aτ ∩ H −1 (M(B, J)) .

More generally: Let E ⊆ M(B) and F ⊆ B separable C *-subalgebras, S a

countable family of closed ideals of B.
Find a separable C *-subalgebra C ⊆ B that contains F and satisfies EC ⊆ C,
each closed ideal I of C is the intersection I = C ∩ J of a closed ideal J of B with
C, and E ∩ M(B, J) = H −1 (M(C, C ∩ J)) for all closed ideals J of B (for the
*-homomorphism H : E → M(C) with H(e)c = ec for e ∈ E and c ∈ C).
What about the desire that C is relatively injective in B? Or is approximately
injective in B?
Can we find such C with the additional property that there exists a normal
conditional expection from B ∗∗ (or from D∗∗ with D := CBC) onto C ∗∗ ? Perhaps,
this seems to be equivalent to the injectivity of the natural morphism

C ⊗ F 7→ B ⊗ F

for all (separable) C *-algebras F .

Reduce to separable case? Here necessary?
The idea would be to compare the composition Cnuc of the l.s.c. action Ψ1 of
Prim B on D composed with Cnuc of the l.s.c. action Ψ2 of Prim D on A with the
Cnuc of the l.s.c. action of Prim B on A:
If the composition Ψ2 ◦ Ψ1 defines the same action as Ψ, and if the m.o.c. cone
Cnuc (Ψ2 ; A, D) and Cnuc (Ψ1 ; D, B) separate Ψ2 respectively Ψ1 , then the minimality
 2. PROPOSED CHANGES (1) 1297

of Cnuc (Ψ; A, B) among the m.o.c.c. with same induced action delivers that

Cnuc (Ψ; A, B) ⊆ Cnuc (Ψ2 ; A, D) ◦ Cnuc (Ψ1 ; D, B) .

Since – at the present state of text – we do not know that (for separable A, B
and D) every l.s.c. action Ψ between there ideal lattices can be separated by the
corresponding m.o.c.c. of Ψ-residually nuclear maps, we get some more work:
We have to show that we can the question reduce to the cases of (other) separa-
ble A, B and D with the additional property that B and D contain regular abelian
C *-subalgebras C.
WHERE is ”regular C *-subalgebra” defined?
Perhaps by the property that each pure state can be extended to a pure state
in a unique manner?
We fix a residually nuclear c.p. map V : A → B with respect to an action Ψ of
Prim(B) that has the following property:
The – lower semi-continuous – action Ψ : I(B) → I(A) is given by Ψ(J) :=
−1
H (M(B, J)) where H : A → M(B) is a *-homomorphism with the property that
H – in the topology of point-wise strict convergence – 1-step innerly approximately
factorizes through H1 : A → M(D) and H2 : D → M(B) in a sense that the maps
a 7→ b∗ H(a)b can be approximated in point-norm topology of maps of the form
a 7→ c∗ H2 (d∗ H1 (a)d)c for suitable c ∈ B and d ∈ D depending on b ∈ B. Moreover
we require that D is strongly p.i.
Since it suffices to show that for given a1 , . . . , an ∈ A+ , and ε > 0 there is
b ∈ B with kV (aj ) − b∗ H(aj )bk < ε for j = 1, . . . , n, it suffices to consider suitable
separable C *-sub-algebras Aτ , Bτ and Dτ of A, B and D, such that a1 , . . . , an ∈ Aτ
and that the Aτ , Bτ , Dτ same property the restrictions of H, H1 and H2 have the
same
This can be done by replacing first A by its “good” separable C *-algebras Aτ
that contains a1 , . . . , an – as described above. Since V (aj ) is contained in the closed
ideal generated by b∗ H(aj )b (b ∈ B) and is – at the same time – also in the closed
ideal generated by c∗ H2 (d∗ H1 (aj )d)c (c ∈ B, d ∈ D) we find a separable C *-sub-
algebra Aτ 3 a1 , . . . an of A, a separable C *-subalgebra Bτ of B and a separable
strongly p.i. subalgebra Dτ of D such that H(Aτ )Bτ ⊆ Bτ , H2 (Dτ )Bτ ⊆ Bτ ,
H1 (Aτ )Dτ ⊆ Dτ and that a 7→ b∗ H(a)b can be approximated by the restrictions
to Aτ .
Then we replace D and B by suitable σ-unital hereditary C *-subalgebras Dτ
and Bτ such that H1 (Aτ )Dτ = Dτ and H2 (Dτ )Bτ = Bτ .
This is possible, because ?????
The following covers only the case where the action of prime(B) on A is also
monotone upper s.c.
Can’t see any reason for this!
1298 C. TEMPORARY (!!) MONITOR AND TO-DO LIST

The good approach to separable Dτ and Bτ is the study of the Dini-function:

X Dini-space, Y locally q-compact sober T0 space, e.g. X := Prim(A), Y :=
prime(B). Ψ : O(Y ) → O(X) l.s.c. and monotonous upper s.c. Then (equivalently)
each Dini-function f on X defines a a Dini function on Y .
Since the set of Dini-functions on X (e.g. X = Prim(A)) is uniformly separable,
one has to find in a C *-algebra corresponding to Y a sequence of elements that
correspond to a dense sequence in the image of the Dini functions on X.
Then one enlarge it to a suitable Bτ with all the needed properties.
Older approach:
(1) Reduction to A separable, D and B σ-unital. Then all can be reduced to
the case where D, B are also separable, by careful selection of the separable full
C *-sub-algebras of B and D that have the same “semi-metrics” ρ(a, b; n) ∈ [0, ∞]
than those coming from B respectively from D.
(Use here: If A separable, then V (A) ⊆ B generates a separable C *-
subalgebra.)
(2) Then consider the maps that factorize over abelian C *-subalgebras of Dω ⊆
M(B)ω → M(Bω ).
(2’) Or (perhaps better), consider equivariant c.p. maps A → Dω that factorize
over abelian C *-subalgebras of Dω . Approximate them by 1-step inner c.p. maps.
(3) Then go from separable abelian C *-subalgebras C ⊆ Dω → M(Bω ) equi-
variant to Bω and approximate the residually nuclear c.p. maps by 1-step inner
maps.
Use here the embedding C ⊆ C ⊗ O∞ ⊆ Dω
(4) Show that the maps from (1)-(3) define the l.s.c. action of I(Bω ) → I(A)
of prime(Bω ) on A.
(5) Show that the point norm closure of the set of maps A → Bω defined by
(1)-(3) is an m.o.c. cone. (The convexity via Cuntz criterion has to be shown).
(6) All? ???
Passage to the non-degenerate case by compressions:
B and D first by the hereditary C *-subalgebras C := ADAB · (Bω ) · BADA
and E := AD · (ADA)ω · DA (that is naturally contained in the multiplier algebra
M(C) of C) of the ultra-powers of B and E, to obtain separable “regular” abelian
C *-subalgebras, that are suitable for the problem.
One of the points is to show that residual nuclear V : A → B is also residual
nuclear as map from A into C – one with respect to the l.s.c. action Ψ1 of I(B)
on I(A), the other with respect to the l.s.c. action Ψ2 of I(C) on I(A) –, which
needs a check that I := Ψ2 (J) = Ψ1 (J ∩ B). Then [V ]I : A/I → B/(J ∩ B) ⊆ C/J
is again nuclear, and it follows that V : A → C is again residually nuclear.
 2. PROPOSED CHANGES (1) 1299

Now fix a residually nuclear map from V : A → B. We find separable C *-

algebras F of C and G of E such that AG = G, GF = F , G is strongly p.i., G and
F contain regular Abelian C *-subalgebras, each closed ideal of G is the intersection
of a closed ideal of E with G, each closed ideal C
????????????????

Concerning work on Chp. 4. :

4a) Perhaps: Study of K1 -injectivity on extra place?
Give short !!! proof of Cuntz lemma on ” full properly infinite projections ” ?
4b) Clean presentation of the “basic proposition”? Now a Theorem ?
4c) Again case: A ⊆ E (via H0 ) J ideal of E, V : A → J 1-step inner (approx-
imately ?).
Suppose E ⊃ A separable, for all separable X ⊆ E and Y ⊆ J, there exists a
positive contraction e ∈ E (?) with ex = xe for all x ∈ X and ey = y = ye for all
y ∈Y.
How the Grothendieck group looks like?
When there is a “majorizing” element? At least of larger semigroup?
Consider an abelian semigroup S with following property:
There exists x1 ∈ S with:
For each x ∈ S there exists y ∈ S with x + y = x1 . Then there exists x0 with
2x1 + x0 = x1 . Thus 2x0 + 2x1 = x0 + x1 . Then x + y + x0 + x1 = x0 + 2x1 = x1
shows that S + x1 is a subgroup of the semigroup S, and is isomorphic to the
Grothendieck group of S and its ”zero” element is equal x0 + x1 .
(h1 : A → J) ∈ S(A, E, H0 , J) (last good notation?) with property:
If [h] ∈ S(A, H0 , E) with h(A) ⊆ J we get/require that [h] ∈ S(A, h1 , E).
Special case: x = s∗ t∗ H0 ts, y =???, (1 − ss∗ )t∗ H0 (·)t(1 − ss∗ ) ∈ S(A, H0 , E).
implies h1 ⊕ h1 - h1 ?
Does it imply the existence and uniqueness of h0 ??

Ad Chp. 5
Ad 5a) Partial results for A ⊆ O2 perhaps only to list?
Anyway now shifted to Appendix B ?
At least of consequences of the more general results (on WvN, Ext)?

5b) More systematic “stable” section. Select only the later needed, e.g. for
asymptotic morphisms, Rørdam’s criteria ...

5c) Ext(C ; A, B) in case B s.p.i. “Kasparov for C”.

1300 C. TEMPORARY (!!) MONITOR AND TO-DO LIST

5d) Homotopy invariance (even of Ext(A, O2 ) = Extnuc (A, O2 )) follows first

from Kasparov’s homotopy invariance of KK -theory,
and from δ2 ∼ id on O2 :
Perhaps one shows (before) that the unital endomorphism δ2 is unitarily homo-
topic to identity map of O2 .

5e)
Ext(C ; A, B) in front of other Ext-notion?

5f) Characterization of “zero” element of Ext(C ; A, B):

B σ-unital and stable,
H : A ⊗ O2 ,→ M(B)/B,
A stable, A exact, inclusion H nuclear, in case of C = CPnuc (Ψ; A, B).
But H must be fitting with the related “action” Ψ.
It must be underlined that “stability” of the corresponding extension algebra
is only granted if the map H dominates zero.
(This is always reached if one tensors – again ? – with K.)
One problem is: When Qs (B) is strongly p.i? Same with M(B)?
∼ B ⊗ O∞ ( =
Positive if B = ∼ B ⊗ D∞ ). Then moreover every separable C *-
subalgebra of B “commutes approximately” with a unital copy of O∞ in B∞ .
This should carry over to Qs (B). (Still a conjecture!)

Equivariant H:
πB (M(B, J)) ∩ H(A ⊗ 1) = H((A ∩ M(B, J)) ⊗ 1) for all closed ideals J of B.
It should follow that a 7→ H(a⊗1) is unitary equivalent to πB ◦H0 : A → Qs (B).

Important:
Is it better to start with the classification of A ⊗ O2 → Qs (B) nuclear?
It is not so useful in case of A nuclear and B = K ⊗ O2 , because it would not
allow to show that the lift is the range of a c.p. projections that is an ?? what ??
in the set of unital positive maps.

5g) Is it possible to prove the triviality of

Ext(C; A ⊗ D2 , B) and Ext(C; A, B ⊗ D2 ) directly? Not using KK ? (Here D2 =
O2 ⊗ O2 ⊗ · · · and the point is that we do not borrow the homotopy - invariance
from KK.)
Partly from suitable material from Chapter 3?
A general Theorem about unitary equivalence in Q(B) (for B stable and σ-
unital) of morphisms h1 , h2 : A ⊗ O2 ,→ Q(B) would be useful.
 3. LIST OF CHANGES: LABELS AND NAMES 1301

Minimally necessary: h1 (a ⊗ 1) and h2 (a ⊗ 1) generate the same closed ideal of

Q(B) for each a ∈ A+ .
There is (by some Corollary in 4) a unitary in Q(B) that makes them unitary
equivalent. The Q(B) is K1 -bijective.
In case B = O2 ⊗ K we can use that B ∼ = B ⊗ O2 ⊗ O2 ⊗ · · · . It gives that
any separable C *-subalgebra of Q(O2 ⊗ K) commutes with a unital copy of O2 .
Since it commutes also with some countable approximations by c.p. maps of a C *-
morphism from a separable C *-algebra A into Q(O2 ⊗ K), and since Q(O2 ⊗ K)
has trivial K∗ -groups
Similar problems appear in the study of Prim-space extensions. (By getting
them manageable by tensor them ⊗O2 .)
E.g. if Hk (a) := hk (a ⊗ 1) have weakly nuclear c.p. lifts, A separable and exact,
H1 (a) and H2 (a) generate the same ideal of Q(B) for each a ∈ A+ .
If A is stable (respectively the corresponding general extension stable), then it
should be possible to give the equivalence by a unitary in U0 (Q(B)).
The needed basic informations have to be developed (in parts) in Chapters 3
and 4.

Chp. 6:
Embedding Theorem needs the characterization of the “big zero-element” of
Ext(C ; A, B).
A exact, stable and separable, B stable, σ-unital and strongly p.i.
C = CPnuc (A, B) ∩ CP(Ψ; A, B) with Ψ : O(Prim(B)) → O(Prim(A)) l.s.c. and
monotone u.s.c. (that is weaker than u.s.c. ???)
Requires to use and study C ⊗ idO2 What is needed really and minimally?

Chp 7.
New chapter 7:
equal to old Chapter 10 (or 11?)
plus first parts of old Chapter 9.
More ????

3. List of Changes: labels and names

(Wanted: Ref. to Definition of “properly infinite element” c ∈ A+ ?) for

Reference ‘sec2:basics.factorize.infinite’ ??? Possible: 1
LaTeX Warning: Reference ‘NOref:WvN.Aomega.implies.WvN.A.?’ on page
48 undefined on input line 5174. (?? seems not to exist in book ??? chapter 12 ??)
1302 C. TEMPORARY (!!) MONITOR AND TO-DO LIST

LaTeX Warning: Reference ‘thm:3.?black.out??PrfUses:WvN.implies.Spi’ on

page 48 undefined on input line 5187. [48]
eksec3..Part1.tex [236] Chapter 3

4. Ref ’s and Cite to be fixed — not up to date

5. Temporary references / compare with my way

What is the precise definition of ”Cuntz semigroup”?

From Black.Robert.Tik.Toms.Wint.2010 [85]:
sec:2.2. The Cuntz semigroup. Let A be a C*-algebra. Let us consider on
(A ⊗ K)+ the relation a - b if vn bvn∗ → a (i.e., the sequence (vn bvn∗ )n converges
to a for the considered a, b ∈ (A ⊗ K)+ ) for some sequence (vn ) in A ⊗ K. Let us
write a ∼ b if a - b and b - a. In this case we say that a is Cuntz equivalent to
b. Let Cu(A) denote the set (A ⊗ K)+ / ∼ of Cuntz equivalence classes. We use hai
to denote the class of a in Cu(A). It is clear that

hai 6 hbi ⇔ a - b

defines an order on Cu(A). We also endow Cu(A) with an addition operation by

setting
hai + hbi := ha0 + b0 i ,
where a0 and b0 are orthogonal and Cuntz equivalent to a and b respectively (the
choice of a0 and b0 does not affect the Cuntz class of their sum). The semigroup
W(A) is then the sub-semigroup of Cu(A) of Cuntz classes with a representative in
S
n Mn (A)+ .

Alternatively, Cu(A) can be defined to consist of equivalence classes of count-

ably generated Hilbert modules over A. The equivalence relation boils down to
isomorphism in the case that A has stable rank one, but is rather more compli-
cated in general. We do not require the precise definition of this relation in the
sequel, and so omit it; the interested reader may consult [5] or [1] for details.
[1] P. Ara, F. Perera, and A. S. Toms. K-theory for operator algebras. Classi-
fication of C*-algebras, Commun. Contemp. Math. To appear.(2008?)
[5] Coward, K.T., Elliott, G.A., Ivanescu, C., [2008], The Cuntz semigroup as
an invariant for C*-algebras, J. Reine Angew. Math. 623, 161–193.
” We note, however, that the identification of these two approaches to Cu(A) is
achieved by associating the element hai to the class of the Hilbert module a`2 (A).
If X is a countably generated Hilbert module over A, then we use [X] to denote its
Cuntz equivalence class; with this notation the sub-semigroup W(A) is identified
with those classes [X] for which X is finitely generated. ”
( E.K.: I can not see how this fits bijectively together with the - classes in
A ⊗ K...)
2.4. Functionals and Cu.
 5. TEMPORARY REFERENCES / COMPARE WITH MY WAY 1303

Let S be a semigroup in the category Cu. A functional on S is a map λ : S →

[0, ∞] that is additive, order preserving, preserves suprema of increasing sequences
and satisfies λ(0) = 0. We use F (S) to denote the functionals on S. We will make
use of a lemma, established in [268].
3. THE RADIUS OF COMPARISON
3.1.Original definition.
The radius of comparison was originally introduced in [18 ??] as an invariant
for unital C*-algebras. (18 is cited in [268])
Let A be a unital C*-algebra, and let QT12 (A) denote the set of normalized 2-
quasi-traces on A. The radius of comparison of A, denoted by rc(A), is the infimum
of the set of real numbers r > 0 with the property that a, b ∈ t∞
n=1 Mn (A) (or noted
F∞
a, b ∈ n=1 Mn (A) ) satisfy a - b whenever

dτ (hai) + r < dτ (hbi), τ ∈ QT12 (A) . (5.1)

By the results of Subsection 2.4 (??) (of which paper/book ??), this is equivalent
to the demand that x, y ∈ Cu(A) satisfy x 6 y whenever

λ(x) + r < λ(y) ,

for all λ ∈ F (Cu(A)) which are normalized in the sense that λ(h1A i) = 1. The
motivation for this definition comes from the stability properties of topological
vector bundles.
4.4.?? When is W(A) hereditary?
Recall that W (A) is the sub-semigroup of Cu(A) of elements hai with a ∈
Mn (A)+ for some n. In fact, W(A) is the original definition of the Cuntz semigroup.
Here we consider the question of when this sub-semigroup is hereditary, i.e., has
the property that if x 6 y in Cu(A) and y ∈ W(A), then x ∈ W(A).
We prove that finite ”radius of comparison” suffices. This result was previously
unknown, even in the case of ”strict comparison”.
Theorem 4.4.1.???
Let A be a C*-algebra for which the projections in every quotient of A ⊗ K
are finite. Let a ∈ A+ be strictly positive and suppose that rA,a < k ∈ N . If
hbi ∈ Cu(A) is such that

λ(hbi) 6 nλ(hai) for all λ ∈ F (Cu(A)) ,

for some n ∈ N, then b is Murray–von-Neumann equivalent to an element of

M2(n+k) (A)+ .
Lemma 4.4.2.???
If b1 and b2 are Murray–von-Neumann equivalent to elements in Mn (A) then
b1 + b2 is Murray–von-Neumann equivalent to an element in M2n (A).
(Case n = 1 checked ???)
Question 4.4.3.???
1304 C. TEMPORARY (!!) MONITOR AND TO-DO LIST

Is there a C*-algebra for which W(A) is not a hereditary subset of Cu(A)?

6. Collection of Definitions of pure infinite C*-algebras

In the separate paper is a shorter and more transparent (??) calculation of the
inequalities below?
Beginning here: Collections of Defs for p.i.
This is NOW (2022 !!!) only for my private use!!! (improving of estimates
started July 2021)
It could be useful for the study of extensions of p.i. algebras ???
The below considered estimates are used to prove and improve the following
estimate:
Let a, b ∈ A+ positive contractions, then, using polar decomposition of a − b in
positive and negative parts (a − b)+ and (a − b)− = (b − a)+ of

a − b = (a − b)+ − (a − b)− .

It gives that a − (a − b)+ = b − (b − a)+ , by using that (a − b)− = (b − a)+ .

This symmetry allows to estimates the norms of a − (a − b)+ and b − (b − a)+
by using the norms of (a + b) − |a − b| and b − (a − b)− = b − (b − a)+ . Since
|a − b| = (a − b)+ + (a − b)− it follows that

a + b − ((a − b)+ + (a − b)− ) = (a − (a − b)+ ) + (b − (a − b)− )

Since the two terms at the right side are equal and
p
|a − b| = (a − b)2 = (a − b)+ + (a − b)− ,

we get here immediately that

2(a − (a − b)+ ) = (a + b) − |a − b| = 2(b − (b − a)+ ) .

p
If we use that |a − b| = (a − b)2 and a + b are (positive) roots of the positive
operators X := (a − b)2 and Y := (a + b)2 . For all positive operators X, Y ∈ A+
holds
2ka − (a − b)+ k = kY 1/2 − X 1/2 k ≤ (kY − Xk)1/2 .

We get in our special case that Y − X = 2(ab + ba),

kY 1/2 − X 1/2 k ≤ kY − Xk1/2 ≤ 2kabk1/2 ,

and finally that

kb − (b − a)+ k = ka − (a − b)+ k = (1/2)k(a + b) − |a − b|k ≤ kabk1/2

PROOF of kY 1/2 −X 1/2 k ≤ kY −Xk1/2 ? (Refer to Pedersen book, concerning

citation on operator monotone functions. But there exits other papers on this –
cite them also!).
 6. COLLECTION OF DEFINITIONS OF PURE INFINITE C*-ALGEBRAS 1305

Y ≤ X + γ · 1 with γ := kY − Xk implies

Y 1/2 ≤ (X + γ)1/2 ≤ X 1/2 + γ 1/2

The estimate (β + γ)1/2 ≤ β 1/2 + γ 1/2 for β, γ ∈ (0, ∞) is easy to see by taking on
both sides quadrants. Thus kY 1/2 − X 1/2 k ≤ kY − Xk1/2 , and in our special case
where X and γ · 1 we get ?????? )
In this special case we get Y − X = 2(ab + ba), and kY − Xk ≤ 4kabk, hence

k(a − (a − b)+ )k = k(b − (b − a)+ )k ≤ kabk1/2 .

See below, but with other notations, because of desired generality.

We use here the ”operator monotony”: It is e.g. given in the book C*-algebras
and their automorphism groups of G.K. Pedersen:
Prop. (1.3.8): Let X, Y ∈ L(H) positive operators. If 0 < kY − Xk ≤ 1, then
The operator function t 7→ tβ is operator monotone on R+ for β ∈ (0, 1).
We use this in case β = 1/2 for the proof of following Result:
Lemma: For all positive operators X, Y ∈ L(H)+ holds the inequality of norms:

kY 1/2 − X 1/2 k ≤ kY − Xk1/2 .

And, therefore,
kY − Xk ≤ kY 2 − X 2 k1/2

Proof: Notice that this follows by the symmetries in X and Y , – as e.g. kY −

Xk = kX − Y k that we consider here as the kY − Xk·??? · idH –, from the general
observation that the operator monotony (which says that 0 ≤ a ≤ b implies 0 ≤
a1/2 ≤ b1/2 ) implies that

Y 1/2 ≤ X 1/2 + kY − Xk1/2 ,

because Y ≤ (X+kY −Xk·idH ) and, always, (X+kY −Xk)1/2 ≤ X 1/2 +kY −Xk1/2 ,
because C ∗ (X, kY − Xk · idH ) is commutative.
By symmetry between X and Y it follows that

X 1/2 ≤ Y 1/2 + kY − Xk1/2

using kY − Xk = kX − Y k. Thus,

kX 1/2 − Y 1/2 k ≤ kY − Xk1/2 .

So, it is enough to apply the operator monotony with β = 1/2 to the obvious
inequality Y ≤ X + kY − Xk. We get that Y 1/2 ≤ (X + kY − Xk)1/2 by operator
monotony. And,- similarly -, X 1/2 ≤ (Y + kY − Xk)1/2 .
The operator X and the scalar kY − Xk (– here considered as the operator
kY − Xk · idH ∈ L(H)+ –) are in a commutative C*-subalgebra of L(H). There, in
the commutative C*-algebra C ∗ (X, kY −Xk·idH ) ⊆ L(H), all monotone increasing
continuous functions apply to the positive elements in commutative C*-algebras as
1306 C. TEMPORARY (!!) MONITOR AND TO-DO LIST

order-monotone function ... like: 0 ≤ Z ∗ = Z 7→ Z := X + kY − Xk. Thus,

(X + kY − Xk)1/2 ≤ X 1/2 + kY − Xk1/2 ,
By taking squares one can see that this is in the commutative C*-subalgebra
∗
C (X, kY − Xk · 1) equivalent to the obvious inequality

X + kY − Xk ≤ X + kY − Xk + 2(kY − Xk · X)1/2 .

?????????
This works because for all nonnegative numbers(!) u and v holds that

(u + v)1/2 ≤ u1/2 + v 1/2 ,

because we have then that u + v ≤ u + v + 2(uv)1/2 .

This carries over to all positive and elements u, v ∈ A+ ... ???
Here is something missing:
It follows for positive contractions A, B that
p
k(A + B) − (A − B)2 k ≤ 2kABk1/2

for positive contractions A, B. ??

(Above it is better explained!)
Want to know:

kA − (A − B)+ k ≤ 2kABk1/2

It comes from C := A + B and D := |A − B| and

C 2 − D2 = (A2 + AB + BA + B 2 ) − (A2 − AB − BA + B 2 ) = 2(AB + BA) .

Thus, kC 2 − D2 k ≤ 4kABk and

kC − Dk ≤ (kC 2 − D2 k)1/2 ≤ 2kABk1/2 .

Above calculations have the following desired application:

Let 0 ≤ A, B ≤ 1 in a C*-algebra then
p
2(A − (A − B)+ ) = (A + B) − (A − B)2 = (A + B) − |A − B|

It follows that
kA + B − |A − B|k ≤ 2kABk1/2
and kA − (A − B)+ k ≤ kABk1/2 .
Notice here also that: AB = 0 if and only if A − (A − B)+ = 0.
The above shown later used application is:

kB − (A − B)− k = kA − (A − B)+ k ≤ kABk1/2 .

!!! Other topic now below !!!:

 7. ON PASSAGE TO SUITABLE SEPARABLE (!) C*-SUBALGEBRAS 1307

7. On passage to suitable separable (!) C*-subalgebras

Let X ⊆ A a countable (or norm-separable) subset of a C*-algebra A, then

there exists a separable C*-subalgebra B of A that contains X, and such that
X ⊆ B ⊆ A have following properties:
1.) b1 -Mn (B) b2 , if and only if, b1 -Mn (A) b2 for all b1 , b2 ∈ Mn (B) and n ∈ N.
2.) B ⊆ A is exact, if A is exact.
( (2.) Means ”Exactness is hereditary”. Where is it proved ? Give reference
and/or outline of proof.)
Let (more general) B a C*-subalgebra of an ”exact” C*-algebra A. A ⊗min C
is defined by the minimal C*-norm on the algebraic tensor product A ⊗ C.
The algebras B ⊗min J − − > B ⊗min C − − > B ⊗min C/J are natural
subalgebras of the algebras

A ⊗min J − − > A ⊗min C − − > A ⊗min C/J

Notice here that A ⊗min C − − > A ⊗min C/J and B ⊗min C − − > B ⊗min
C/J are always surjective, and are well-defined from the algebraic tensor products
A ⊗ C − − > A ⊗ C/J by completion with the minimal C*-norms.
We see elow that B ⊗min J is the kernel of B ⊗min C − − > B ⊗min C/J if
A ⊗min J is the kernel of A ⊗min C − − > A ⊗min C/J.
If the kernel of B ⊗min C − − > B ⊗min C/J would be different from B ⊗min J
then it must be bigger than B ⊗min J. But if A is exact, then this kernel is always
contained in (A ⊗min J) ∩ (B ⊗min C).
If we take approximate units eσ ∈ B+ and fτ ∈ J+ of B and of J, then

Yσ,τ := (eσ ⊗ fτ )X(eσ ⊗ fτ )

converges to X for each element of X ∈ (A ⊗min J) ∩ (B ⊗min C). But, each Yσ,τ
is contained in B ⊗min J. Thus,

(A ⊗min J) ∩ (B ⊗min C) = B ⊗min J

and it shows that B ⊆ A is exact, if A is exact.

Notice here that there is a natural isomorphism A ⊗min C/J ∼
= (A ⊗min
C)/(A ⊗min J) for exact C*-algebras A.
The question is equivalent to:
Is the kernel (A ⊗min J) ∩ B ⊗min C ⊂ A ⊗min C of B ⊗min C − − > (A ⊗min
C)/(A ⊗min J) bigger than B ⊗min J ?
In general this kernels are contained in

F (B, C; J) := {x ∈ B ⊗min C; ϕ ⊗ id(x) ∈ J}?

And we can ask the stronger question: Is F (B, C; J) = (A ⊗min J) ∩ B ⊗min C

under which circumstances?
1308 C. TEMPORARY (!!) MONITOR AND TO-DO LIST

If one can it reduce (!) to the separable case for A the one can use that
B ⊆ A ⊆ O2 proves exactness of B.
This would then require to show that exactness is preserved under inductive
limits ... in addition.
(give Ref’s !!!)
3.) If A is nuclear, then B can be chosen nuclear, by an iteration argument for
B becoming a suitable inductive limit X ⊆ B1 ⊆ B2 ⊆ · · · and let B the closure of
S
n Bn .

(One can moreover manage that B is approximately injective, weakly injective

etc. if A has such properties, also (??) slice-map properties (??) could be earned
from A ...??? Check this carefully !!!). (Below is an other reduction attempt ...)
Now let A be a separable C*-algebra and let J a closed ideal of A, and a0 ∈ J
a strictly positive contraction for J, i.e. with

J = Aa0 A = a0 Aa0 = a0 Ja0 .

Assume (from now on) that a0 is properly infinite, i.e., a0 ⊕ a0 -J a0 in J (and

then also a0 ⊕ a0 -A a0 and vice versa). Notice that a0 ⊕ a0 -B a0 also for every
C*-subalgebra B of A with J ⊆ B.
Suppose that A/J contains an infinite element x ∈ (A/J)+ , i.e., there exists
nonzero y ∈ (A/J)+ with x ⊕ y -A/J x. Then the closed ideal I of A/J generated
by all the elements y ∈ (A/J)+ with x ⊕ y - x contains (by separability of A/J)
a strictly positive contraction z ∈ I+ . (And z is non-zero because because I+ is
non-zero.)
This implies that z(A/J)z and (A/J)z(A/J) are both dense in I.
It turns out that x ⊕ z -A/J x and – since I is an ideal where z is a strictly
positive contraction – that also z ⊕ z -A/J z, i.e., that z is properly infinite in A/J.
We can lift z to a positive contraction a1 ∈ A+ with πJ (a1 ) = z.
Then can consider the positive element e := a0 + (1 − a0 )1/2 a1 (1 − a0 )1/2 with
πJ (e) = a1 and e ≥ a0 .
Thus, J ⊆ eAe and πJ (eAe) = I.
Is e infinite? Even this is not clear!
Is a0 ⊕ e - e and is a1 ⊕ e - e ?
1/2 1/2
Can we also take a0 + a1 (1 − a0 )a1 ???
Here some minimal conditions that let hope that A is p.i. ( =: all nonzero
elements are ”infinite”):
(0) This part (0) is true for all C*-algebras:
The J. Cuntz relation a -A b is compatible with with the corresponding relation
in hereditary C*-subalgebras:
 7. ON PASSAGE TO SUITABLE SEPARABLE (!) C*-SUBALGEBRAS 1309

If D ⊆ A is a hereditary C*-subalgebra of A and a, b ∈ D satisfy a -A b in A,

then also a -D b in D.
If cn bdn converges to a in A where cn , dn ∈ A, then one finds en cn fn , gn dn hn ∈
rAr ⊆ D , for r := a∗ a + aa∗ + b∗ b + bb∗ , such that (en cn fn )b(gn dn hn ) converges
to a ∈ rAr.
It is possible to find for every separable subspace X ⊆ A (or every countable
subset Y ⊆ A of A) a separable C*-subalgebra C ⊆ A that contains X (respectively
contains Y ) such that for a, b ∈ K ⊗ C holds a - b in K ⊗ C, if and only if, a - b
in K ⊗ C. Remind here a -D b for positive elements a, b ∈ D in a C*-algebra D
means that there exists a sequence (dn ) in D with a = lim1→∞ - d∗n bdn .
(In addition one can C select as a nuclear C*-subalgebra of A if A is nuclear,
this is automatic for exactness, but it is an open question if C ⊇ X can be always
chosen that there exists in addition a conditional expectation from A∗∗ onto C ∗∗ .)
(0) Original definition of pure infiniteness of a C*-algebra A is that a ⊕ a -A a
for all a ∈ A+ . (It says that there exists sequences (cn ) and (dn ) in A such that
c∗n acn and d∗n adn converge both to a, but c∗n adn converges to 0.)
(1) Minimal requirements (!) to the idea of ”p.i.” C*-algebras are:
(1a) For every hereditary C*-subalgebra D of A and every pure state ρ on D
there exists elements d, e ∈ D+ with ρ(e) > 0 and e ⊕ d - d.
(Then A contains a closed ideal J with e ∈ J, and f ⊕ d - d for all f ∈ J.)
Notice here that the question if every non-zero positive contraction b ∈ A+ is
properly infinite (p.i.) is equivalent to the following question:
If D := bAb is the hereditary C*-subalgebra of A that contains b as strictly
positive element of D, then for each element c of the closed ideal J of A generated
by D there exists sequences (en ) and (fn ) in A with c = lim en bfn .
Here one can pass to suitable separable C*-subalgebras B ⊆ A of A, or of D,
with the property that for b, c ∈ B holds b -B c, if and only if, b -A c. This implies
also that each (closed) ideal of B is the intersection of B with a closed ideal of A.
The same can be done with A ⊗ K(`2 ) and its C*-subalgebra B ⊗ K(`2 ).
To exclude also the cases C ∼
= D or K ∼
= D for hereditary C*-subalgebras of
A/J one has to suppose that ??????
But we must to say much more! For example:
If E is a hereditary C*-subalgebra ... ????
Then B satisfies again (1a) and the question if every positive contraction of b
of B ?????
(1c) Possibly (formally) ”stronger” as (1a) is the following:
For every closed ideal J 6= A (that can be J = {0}) and (non-zero) hereditary
C*-subalgebra D of A/J, the algebra D contains an ”infinite” element 0 6= d ∈ D+
(i.e. that d is non-zero).
1310 C. TEMPORARY (!!) MONITOR AND TO-DO LIST

It is defined by the property that there exists non-zero e ∈ D+ with e ⊕ d -D d,

i.e. it means that there exists a sequences of elements fn , gn ∈ D with limn fn ∗
dfn = d, limn gn ∗ dfn = 0 and limn gn ∗ dgn = e.
??? Thus, ???? e⊕??? is in the ideal of A/J one-step map by d ?????
Since ????
(Or is here also e ∈ A/J allowed?
Yes is allowed, because the elements e ∈ A/J with e ⊕ d -A/J d build a
closed ideal K of A/J. The intersection K ∩ D of K with D is identical with
......................... ?????)
In particular, then each (!) nonzero hereditary C*-sub-algebra D of A has no
character.
Equivalently expressed: A has no irreducible representation that contains in its
image non-zero compact operators on a Hilbert space. Or: A does not contain two
ideals I ⊆ J ⊆ A such that J/I is isomorphic to the algebra of compact operators
K(H) on an infinite dimensional or (non-zero) finite dimensional Hilbert space H.
We call a ∈ A+ infinite if there exists some non-zero element b ∈ A+ such that
a ⊕ b - a. Here we use the Cuntz-majorization x - y in A ⊗ K, and a ⊕ b means
the the 2 × 2 matrix in M2 (A) with diagonal entries a and b.
Let the following be a ”suitable”(?) definition of a ”locally purely infinite”
C*-algebra A:
For every closed ideal J of A and (nonzero) hereditary C*-sub-algebra D of
A/J there exists an infinite element in D (which is then infinite A/J), but for
hereditary C*-sub-algebras D of A/J this is the same as being infinite in D).
Questions:
(Q1) Is this class of C*-algebras invariant under passage to hereditary sub-
algebras and quotients?
(Q2) Is every separable subset X of A contained in a (suitable) separable C*-
sub-algebra B of A that is again a ”locally purely infinite C*-algebra”?
The reduction to the separable case is very important. One should take those
separable C*-sub-algebras B ⊆ A with the property that elements in B ⊗ K have
the same - −− relations in B ⊗ K as in A ⊗ K.
But the important point is: If b ∈ B is infinite in A then b should be infinite
in B. ... And this for all hereditary sub-algebras ...
It seems ”easy” to show that inductive limits of locally purely infinite C*-
algebras are again locally purely infinite.
The point is: How looks like a hereditary subalgebra of a quotient of this
inductive limit?
Is it itself an inductive limit? (Of useful things).
 9. ON IDEALS GENERATED BY n-HOMOGENOUS ELEMENTS 1311

On reduction to the separable case:

Let X ⊆ A a separable subset of A.
Then there exists a separable C*-sub-algebra B of A that contains X and has
the property that for all a, b ∈ B holds: a -B b, if and only if, a -A b. It implies
also that each closed ideal of B is is an intersection of a closed ideal of A with B.
(But in general B does not separate the ideals of A.)

8. MORE on infinite elements

Let A a C*-algebra ...

Let A a separable C*-algebra and D a hereditary C*-sub-algebra. Let h ∈ D,
denote by I(h) the closed ideal of elements a ∈ A with a ⊕ h -A h, then the
intersection I(h)∩D has the property that b ∈ (I(h)∩D), if and only if, b⊕h -D h.
(Check this again! Seems to work.)
Moreover I(h) is generated by I(h) ∩ D.
If I(h) or I(h)∩D contain a strictly positive contraction c ∈ A+ (i.e. cAc dense
in I(h), or cAc is dense in I(h) ∩ D) then c is properly infinite and I(h) = I(c)
respectively I(h) ∩ D = I(c) ∩ D.
Such element c ∈ A+ exists if A or D is separable.
It should be equivalent to require that for each ????
A question of Bruce Blackadar:
Is it possibly true that the classifiable simple C*-algebras are precisely the
(simple) inductive limits of sequences of semiprojective (separable) nuclear C*-
algebras?
E.K.: What is the Definition of semiprojective C*-algebras?
Kind of approximate lifting property? ????

9. On ideals generated by n-homogenous elements

Let A a C*-algebra and Hn the (smallest) hereditary C*-subalgebra of A that

is generated (or only ”defined”) by the n-homogenous elements in A+ .
Definition: An element a ∈ A+ is n-homogenous, if and only if, there exists a
C*-algebra morphism

ϕ : C0 ((0, 1], Mn ) = C0 ((0, 1]) ⊗ Mn 7→ A

with the property that a = γ · ϕ(f0 ⊗ 1n ) for some γ ∈ (0, ∞) and f0 (t) := t, for
t ∈ (0, 1].
1312 C. TEMPORARY (!!) MONITOR AND TO-DO LIST

Hn is the smallest hereditary C*-subalgebra of A that is generated by the

convex combinations of the n-homogenous elements in A+ . Since the set of n-
homogenous is invariant under automorphisms of A, – especially by inner auto-
mophisms –, it follows that Hn is a closed ideal of A.
This ideal Hn has the property that A/Hn has only irreducible representations
on Hilbert spaces of dimensions ≤ (n − 1).
(as ideal and not necessarily fixing ???)
GO ON NOW !!!!!!

10. On order in A+ for C*-algebras A:

It is very likely that the result in the following proposition is well-known, but
we could not find a reference for it in text-books.
Proposition:
If two elements a, b ∈ A+ satisfy a ≤ b, then for each γ > 1 there exists an operator
d ∈ C ∗ (a, b) ⊆ A with d∗ bd = aγ .
What happens ????
What we really later need is that:
If two elements a, b ∈ A+ satisfy a ≤ b then there exists for each ε ∈ (0, kak)
an element dinA with the property d∗ bd = (a − ε)+ .
Similar problem: If ka − bk < γ < min(kak, kbk) Does there exists d ∈ A with
(a − γ)+ = d∗ bd ?
Proof: ...
It suffices to consider the case a 6= 0 (by rescaling later a, b and d with non-
negative real numbers). Moreover, it suffices to consider only the case where A :=
C ∗ (a, b).
We consider the multiplier algebra M(A) and write γ (and δ) in place of γ ·
1M(A) (and in place of δ · 1M(A) ), and consider mainly δ = 2−n for n := 1, 2, ....
Let en := (b+2−n )−1/2 a1/2 and dn := en ·aλ with λ = (γ −1)/2, i.e., γ = 1+2λ.
We (want to ???) show that the dn converge to an element of d ∈ C ∗ (a, b)
with d∗ bd = aγ and kbk ≤ 1.
??? is really
Use that ken k2 = ken e∗n k = ke∗n en k
and then that kdn k ≤ kakλ · ken k and estimate by

en e∗n = (b + 2−n )−1/2 a(b + 2−n )−1/2 ≤ (b + 2−n )−1/2 b(b + 2−n )−1/2 = (b + 2−n )−1 b .

In particular ken k ≤ 1 for all n = 1, 2, ....

Recall that en = (b + 2−n )−1/2 a1/2 .
 11. ON DENSE (ALGEBRAIC) IDEALS OF C*-ALGEBRAS: 1313

The definition of en shows that e∗n (b + 2−n )en = a. It is equivalent to e∗n ben =
a − 2−n e∗n en , If the (en ) converges in A to an element e ∈ C ∗ (a, b) ⊆ A then
e∗ be = a and kek ≤ 1.
The problem is that this en do not converge in A (itself)?
dn := en · aλ with λ = (γ − 1)/2, i.e., γ = 1 + 2λ.
All is now proven up to the estimate of kdn − dm k:
kdn − dm k ≤ kakλ ken − em k If kak≤ 1 then we get same estimates ... ???
Recall that en := (b + 2−n )−1/2 a1/2 and dn := en · aλ ,
Xn := (b + 2−n )−1/2 b1/2 converges in the point norm ???
Let fn,m := (b + 2−n )−1/2 − (b + 2−m )−1/2 . Since 0 ≤ x ≤ y implies
∗
Then fn,m = fn,m and (dn − dm ) · (dn − dm )∗ = fn,m afn,m . Now a ≤ b implies
that

fn,m afn,m ≤ fn,m bfn,m = [b1/2 (b + 2−n )−1/2 − b1/2 (b + 2−m )−1/2 )]2 .

??? It follows that (dn ) is a Cauchy sequence in A if ??? (??)

Then kdn − dm k2 = k(dn − dm )(dn − dm )∗ k, and a ≤ b implies that ?????

(dn −dm )(dn −dm )∗ ≤ ((b+2−n )−1/2 −(b+2−m )−1/2 )b((b+2−n )−1/2 −(b+2−m )−1/2 )

11. On dense (algebraic) ideals of C*-algebras:

Let A a C*-algebra and let J ⊆ A an algebraic ideal of A that is dense in A,

i.e., for every a ∈ A there exists a sequence (cn ) in J with ka − cn k < 1/n. Here
we do not require that J is invariant under the involution c 7→ c∗ , i.e. we do not
here require that c ∈ J implies that c∗ ∈ J.
Precisely expressed: J is a norm-dense linear subspace of A with the properties
b · a, a · b ∈ J for all b ∈ J and a ∈ A.
One can see from this defining properties that J∗ := {c∗ ; c ∈ J} ⊆ A is again a
dense algebraic ideal in A, because the involution x → x∗ is isometric in A and the
inclusions a∗ · J ⊆ J and J · a∗ ⊆ J for a ∈ A imply that J∗ · a ⊆ J∗ and a · J∗ ⊆ J∗
for all a ∈ A.
It follows that the sets of products J∗ · J and J · J∗ are contained in J∗ ∩ J and
are both dense sets in A, that are invariant under multiplication by elements from
A.
Thus, the intersection J∗ ∩ J ⊆ J is again a dense ideal of A because (J∗ · J) ∪
(J · J∗ ) ⊆ J∗ ∩ J.
If J is a dense algebraic ideal of A, then J has following properties:
(1) for every a ∈ A+ and 0 < ε < 1 there exists b ∈ J+ := A+ ∩ J with
ka − bk < ε.
1314 C. TEMPORARY (!!) MONITOR AND TO-DO LIST

Proof of property (1): It is enough to consider the case where a ∈ A+ has norm
kak = 1. Then ka1/2 k = kak1/2 = 1. Let β := (1 + ε)1/2 − 1, i.e., (1 + β)2 = 1 + ε
and ε = β 2 + 2β.
Since J is dense in A, there exists c ∈ J with ka1/2 −ck < β. Then c∗ c ∈ A+ ∩J
and it follows that ka − c∗ ck < β(2 + β) = ε by using the following general estimate
in C*-algebras A:
If x, y ∈ A and kx − yk < β, then
kx∗ x − y ∗ yk < β · (kxk + kyk) ≤ β · (β + 2kxk)
because kyk ≤ kxk + ky − xk. Then take here x := a1/2 and y := c.
 Bibliography

[1] Aarnes, J.F. [1969], Physical states on a C*-algebra, Acta Math. 122, 161–172.
[2] Aarnes, J.F. [1970], Quasi-states on C*-algebras, Trans. Amer. Math. Soc. 149, 601–625.
[3] Aarnes, J.F. [1991], Quasi-states and quasi-measures, Advances in Math. 86, 41–67.
[4] Akemann, Ch.A., Anderson, J., Weaver, N., [1999], A geometric spectral theory for n-tuples
of self-adjoint operators in finite von Neumann algebras, J. Funct. Anal. 165, 258–292.
[5] Akemann, Ch.A., Ostrand, Ph.A., [1975], On a tensor product C*-algebra associated with
the free group on two generators, J. Math. Soc. Japan 27(4), 589–599.
[6] Akemann, Ch.A., Anderson, J.H., Pedersen, G.K., [1986], Excising States of C*–algebras,
Can. J. Math. 38, 1239–1260.
[7] Akemann, Ch.A., Pedersen, G.K., Tomiyama, J., [1973], Multipliers of C*–Algebras,
J. Funct. Analysis 13, 277–301.
[8] Akemann, Ch.A., Pedersen, G.K., [1977], Ideal perturbation of elements in C*-algebras,
Math. Scand. 141, 117–139. hfill
[9] Akemann, Ch.A., Pedersen, G.K., [1979], Central sequences and inner derivations of sepa-
rable C*–algebras, Amer. J. Math. 101, 1047–1061.
[10] Akemann, Ch.A., Weaver N., [2008], B(H) has a pure state that is not multiplicative on
any masa, Proc. Natl. Acad. Sci. USA. 2008 Apr 8; 105(14):5313-4.
[11] Aleksandrov, A.B., Peller, V.V., [2010], Operator Hölder-Zygmund functions, Adv. Math.
224, 910–966. (arXiv:09073049v2)
[12] , , [2011], Estimates of operator moduli of continuity, J. Funct. Analysis 261,
2741–2796. (arXiv:1104.3553v1)
[13] Alfsen, E.M., [1971], Compact convex sets and boundary integrals, 2nd edition, Springer-
Verlag, New York, 1971, Ergebnisse der Mathematik und ihrer Grenzgebiete, Band 57.
[14] Al-Rawashdeh, Ah., [2016], Non-isomorphic C*-algebras with isomorphic unitary groups,
Adv. Oper. Theory 1(2), 160–163.
[15] Al-Rawashdeh, Ah., Booth, A., Giordano, Th., [2012],
[16] Amini, M., Golestani, N., Jamali, S., Phillips, N.C., [2021], Simple tracially Z-absorbing
C*-algebras, (arXiv:2109.05192v1)
[17] An, Q., Liu, Zh., [2022], A total Cuntz semigroup for C*-algebras of stable rank one,
arXiv:2201.10763v1 (26 Jan 2022).
[18] Anantharaman-Delaroche, C., [1995], Classification des C*-algèbres purement infinies
[d’après E. Kirchberg], Séminiare Bourbaki 805.
[19] , [1997], Purely infinite C*-algebras arising from dynamical systems,
Bull. Soc. math. France 125, 199–225.
[20] , [1997], C*-algèbres de Cuntz-Krieger et groups Fuchsiens, Operator Theory, Op-
erator Algebras and Related Topics (Timisoara 1996), pp. 17–35, The Theta Foundation,
Bucharest, 1997.
, [2002], Amenability and exactness for dynamical systems and their C*-algebras,
Trans. Amer. Math. Soc. 354, 4153–4178.
[21] , Renault, J., [2000], Amenable groupoids, Monographies de L’ Enseignement
Mathématique, Genève, vol. 36. With a foreword by Georges Skandalis and an Appendix
(B) by E. Germain.

1315
1316 BIBLIOGRAPHY

[22] Anderson, J., [1978], A C ? -algebra A for which Ext(A) is not a group, Ann. of Math. 107,
455–458.
[23] Ando, H., Haagerup, U., [2013], Ultrapowers, QWEP von Neumann Algebras, and the
Effros-Maréchal Topology,
appeared where?
Received: 2013-11-13, Revised: 2013-12-17 Published Online: 2014-3-7 (but then all
with Winsløw, C). It is likely that it is published as part of [25] see/compare also ...
arXiv:1306.0460.
[24] Ando, H., Haagerup, U., [2014], Ultraproducts of von Neumann algebras, J. Funct. Analysis
266, 6842–6913.
[25] Ando, H., Haagerup, U., Winsløw, C., [2016], Ultraproducts, QWEP von Neumann algebras,
and the Effros-Maréchal topology, J. Reine Angew. Math. 715, 231–250.
[26] Ando, H., Kirchberg, E. [2016], Non-commutativity of the central sequence algebra for sep-
arable non-type I C*-algebras, J. London Math. Soc. 94(1), 280–294.
[27] Ara, P., Brown, N.P., Guido, D., Lledó, F., Toms, A., Perera, F., [2011], The Cuntz semi-
group and the classification of C*-algebras, Contemporary Mathematics 534, vii+168pp. In
Aspects of operator algebras and applications, Preprint: arXiv:0902.3381v1.
[28] Ara, P., Goodearl, K.R., O’Meara, K.C., Raphael, R., [2000], K1 of separative exchange
rings and C*-algebras with real rank zero, Pacific J. Math., 195(2), 261–275.
[29] Ara, P., Lledó, F., Perera, F., [2011], Basic definitions and results for operator algebras, in
Appendix of [27].
[30] Ara, P., Pedersen, G.K., Perera, F., [2001], A closure operation in rings, International
Journal of Mathematics, 12(7), 791–812.
[31] Araki, H., Woods, E.J., [1968], A classification of factors, Publ. RIMS. Kyoto Univ. 4,
51–130.
[32] Archbold, R.J., [1975], On the Centre of a Tensor Product of C*-Algebras, J. Lon-
don Math. Soc. 10(3), 257–262.
[33] Archbold, R.J., [1977], An averaging process for C*-algebras related to weighted shifts,
Proc. London Math. Soc., (3), 35, 541–554.
[34] Archbold, R.J., [1979], On the ”Flip-Flop” Automorphism of C ∗ (S1 , S2 ),
Quart. J. Math. Oxford Ser. (2) 30, 129–132.
[35] Archbold, R.J., Batty, C.J.K., [1980], C*-tensor norms and slice maps, J. Lond. Math. Soc.
22 (2), 127–138.
[36] Archbold, R.J., Gogić, I., Robert, L., [2021], Local variants of the Dixmier Property and
weak centrality for C*-Algebras, arXiv:2106.0009v1 (math.OA) 31 May 2021.
[37] Archbold, R.J., Robert, L., Tikuisis, A., [2017], The Dixmier property and tracial states for
C*-algebras, arXiv:1611.08263v2 (math.OA) 29 Jun 2017.
[38] Archbold, R.J., Spielberg, J.S., [1994], Topologically free actions and ideals in discrete C*-
dynamical systems, Proceedings of the Edinburgh Mathematical Society 37(1), 119–124.
[39] Archey, D., Buck, J., Phillips, N.C., [2017], Centrally large subalgebras and tracial Z-
absorption, International Mathematics Research Notices 292, 1–21.
[40] Archey, D., Phillips, N.C., [2020], Permanence of stable rank one for centrally large subal-
gebras and crossed products by minimal homeomorphisms, J. Operator Th. 83, 353–389.
[41] Arendt, W., [1983], Spectral properties of Lamperti operators, Indiana Univ. Math. J. 32,
199–215.
[42] Arveson, W. [1969], Subalgebras of C*-algebras, Acta Math. 123, 141–224.
[43] , [1977], Notes on extensions of C*-algebras, Duke Math. J. 44, 329–355.
[44] Baaj, S., Julg, P., [1983], Théorie bivariante de Kasparov et opérateurs non bornés dans les
C*-modules hilbertiens, CR Acad. Sci. Paris Sér. I Math., 296(21), 875–878.
[45] Barlak, S., Enders, D., Matui, H., Szabó, G., Winter, W., [2015], The Rokhlin property vs.
Rokhlin dimension 1 on unital Kirchberg algebras, J. Noncommut. Geom. 9(4), 1383–1393.
 BIBLIOGRAPHY 1317

[46] Barlak, S., Li, X., [2020], Cartan subalgebras and the UCT problem, II, Math. Ann. 378,
255–287.
[47] Barlak, S., Szabó, G., [2016], Sequentially split *-homomorphisms between C*-algebras,
Int. J. Math. 27(12), 48 pages.
[48] Barlak, S., Szabó G., [2020], Approaching the UCT problem via crossed products of the
Razak-Jacelon algebra, Groups Geom. Dyn., 14(1), 137–149.
[49] Barlak, S., Szabó G., [2017], Rokhlin actions of finite groups on UHF-absorbing C*-algebras,
Trans. Amer. Math. Soc. 369, 833–859.
[50] Barlak, S., Szabó G., [2019], On diagonal quasi-free automorphisms of simple Cuntz-Krieger
algebras, Math. Scand. 125(2), 210–226.
[51] Barlak, S., Szabó G., Voigt C., [2017], The spatial Rokhlin property for actions of compact
quantum groups, J. Funct. Anal. 272(6), 2308–2360.
[52] Bartle, R.G., Graves, L.M., [1952], Mappings between function spaces, Trans. AMS 72,
400–413.
[53] Batty, C. J. K., [1979], Fubini and martingale theorems in C*-algebras, Math. Proc. Cam-
bridge Philos. Soc., 86, 57–69.
[54] Bauval, A., [1998], RKK(X)-nucléarité (d’après G. Skandalis), K-theory 13, 23–40.
[55] Bedos, E., Conti, R., [2012], On discrete twisted C*-dynamical systems, Hilbert C*-modules
and regularity, Münster J. of Math. 5, 183–208.
[56] Bédos, E., Kaliszewski, S., Quigg, J., [2021], Skew products of finitely aligned left cancellative
small categories and Cuntz–Krieger algebras, Münster J. of Math. 14, 59–99.
[57] Beltita, I., Beltita, D., [2017], Topological aspects of group C ∗ -algebras, Banach Center
Publications.
[58] Benson, D.J., Kumijan, A., Phillips, N.Ch. [2003], Symmetries of Kirchberg algebras, E-
print: arXiv:math/0302273v1 [math.OA].
[59] Bentmann, R., [2010], Filtrated K-theory and classification of C*-algebras, Diploma thesis,
Göttingen 2010.
[60] Bentmann, R., [2014], Kirchberg X-algebras with real rank zero and intermediate cancella-
tion, J. Noncommut. Geom. 8(4), 1061–1081.
[61] Bentmann, R., Dadarlat, M. [2015], One-parameter continuous fields of Kirchberg algebras
with rational K-theory, Proceedings of the AMS 143(8), 3455–3463.
[62] Bentmann, R., Köhler, M. [2011], Universal Coefficient Theorems for C*-algebras over finite
topological spaces, arXiv:math/1101.5702v3.
[63] Bentmann, R., Meyer, R. [2014], A more general method to classify up to equivariant KK-
equivalence, arXiv:math/1405.6512.
[64] Berberian, S. [1972], Baer *-Rings, Grundlehren der Math. Wissenschaften 195, Springer,
Heidelberg/New York.
[65] Berg, I.D. [1971], An extension of the Weyl–von Neumann theorem to normal operators,
Trans. Amer. Math. Soc. 160, 365–371.
[66] , Davidson, K.R. [1991], A quantitative version of the Brown-Douglas-Fillmore the-
orem, Acta. Math. 166, 121–161.
[67] Bhattacharya, A., Wang, Sh., [2016], Kirchberg’s factorization property for discrete quan-
tum groups, Bulletin London Mathematical Society, VL 48, arXiv:1512.08737v2 [math.OA]
(2016).
[68] Blackadar, B. [1980], A simple C*-algebra with no nontrivial projection, Proc. Amer. Math.
Soc. 78, 504–508.
[69] Blackadar, B. [1980], Traces on simple AF C*-algebras, J. Funct. Anal., 38(2), 156–168.
[70] Blackadar, B. [1985], Shape theory for C ? -algebras, Math. Scand., 56(2), 249–275.
[71] , [1985], Nonnuclear subalgebras of C ? -algebras, J. Operator Theory 14, 347–350.
[72] , [1990], Symmetries of the CAR algebra, Ann. of Math. 131(2), 589–623.
[73] , [1998], K-theory for operator algebras, MSRI publications 5, Cambridge University
Press.
1318 BIBLIOGRAPHY

Compare all citations of !!! old book !!! of Bruce Blackadar with new cite or
place in [73]!!! VERY important!!! ??
[74] , [2004], Semiprojectivity in Simple C*-Algebras, Advanced Studies in Pure Mathe-
matics 38, Operator Algebras and Applications, 1–17.
[75] , [2004], The stable rank of full corners in C*-algebras, Proc. Amer. Math. Soc.
132(10), 2945–2950.
[76] , [2006], Operator algebras. Theory of C*-algebras and von Neumann algebras, En-
cyclopaedia Mat. Sci. 122, Springer-Verlag, Berlin 2006.
[77] , [2019], Classification of C*-Algebras, (Draft of Incomplete) Preliminary Edition,
24. 09. 2019.
[78] , Cuntz, J. [1982], The structure of stable algebraically simple C*-algebras,
Amer. J. Math. 104, 813–822.
[79] , Handelman, D. [1982], Dimension functions and traces on C*-algebras, J. Funct.
Analysis, 45, 297–340.
[80] , Kirchberg, E., [1997], Generalized inductive limits of finite-dimensional C*-
algebras, Math. Ann., 307, 343–380.
[81] , Kirchberg, E., [2001], Inner quasidiagonality and strong NF algebras, Pacific J.
Math., 198, 307–329.
(There is an obvious misprint in the statement of Part(ii) of Proposition 3.16! Compare the
proof and read it there so as it is used in the proof of Prop. 3.16.(ii) !)
[82] , Kirchberg, E., [2000], Generalized inductive limits and quasidiagonality, in Alge-
bren, Proceedings of the SFB-Workshop on C*–algebras, Münster, Germany, March 8-12,
1999/J. Cuntz, S. Echterhoff (ed.), Berlin etc., Springer (2000), p.23–41.
[83] , Kirchberg, E., [2011], Irreducible representations of inner quasidiagonal C*-
algebras, Canad. Math. Bull. 54(3), 385–395.
arXiv:0711.4949v2 (12.Dec.2007).
[84] , Kumjian, A., Rørdam, M., [1992], Approximately central matrix units and the
structure of non-commutative tori, K-theory 6, 267–284.
[85] , Robert, L., Tikuisis, A.P., Toms, A.S., Winter, W. [2012], An algebraic approach
to the radius of comparison, Trans. A.M.S. 364, 3657–3674.
arXiv:1008.4024v1 (24 Aug 2010)
[86] , Rørdam M. [1992], Extending states on preordered semigroups and the existence of
quasitraces of C*-algebras, Journal of Algebra 152, 240–247.
[87] Blanchard, E. [1995], Tensor products of C(X)-algebras over C(X), Astérisque 232, 81–92.
[88] , [1996], Deformations de C*-algèbres de Hopf, Bull. Soc. Math. France 124, 141–215.
[89] , [1997], Subtriviality of continuous fields of nuclear C*-algebras, (with an appendix
by E. Kirchberg), J. reine angew. Math. 489, 133–149.
[90] , [2005?], Deformations of infinite projections, Preprint 2005?.
[91] , [2010], K1 -injectivity for properly infinite C*-algebras, Clay Math. Proc. 11, 49–54.
[92] , Kirchberg, E. [2004], Global Glimm halving for C*-bundles, J. Operator Theory
52, 385–420.
[93] , Kirchberg, E. [2004], Non-simple purely infinite C*-algebras: the Hausdorff case,
J. Funct. Analysis 207, 461–513.
[94] , Rohde R., Rørdam M. [2008], Properly infinite C(X)-algebras and K1 -injectivity,
J. Noncommutative Geometry 2(3), 263–282.
[95] , Wassermann S., [2007], Exact C*-bundles, Houston Journal of Mathematics, 33(4),
1147–1159.
[96] Boca, F., [1991], Free products of completely positive maps and spectral sets, J. Funct. Anal-
ysis 97, 251–263.
[97] Bonkat, A., [2002], Bivariante K-Theorie für Kategorien projektiver Systeme von C*-
Algebren, Ph.D. thesis, Westfälische Wilhelms-Universität, 2002.
 BIBLIOGRAPHY 1319

[98] F. F. Bonsall, J. Duncan, Numerical ranges of operators on normed spaces and of elements
of normed algebras, London [1971].
[99] F. F. Bonsall, J. Duncan, Complete Normed Algebras, Springer, Berlin etc. [1973].
[100] J. Bosa, J. Gabe, A. Sims, S. White, The nuclear dimension of O∞ -stable C*-algebras,
arXiv:1906.02066v1, math.OA.
[101] Böttcher, A., Spitkovsky, I.M., [2010], A gentle guide to the basics of two projections theory,
Linear Algebra Appl. 432, 1412–1459.
[102] Bowen, R., Franks, J., [1977], Homology for zero-dimensional nonwandering sets, Ann. of
Math. 106, 73–92.
[103] Brake, L., and Winter, W., [2019], The Toeplitz algebra has nuclear dimension one,
Bull. Lond. Math. Soc. 51(3), 554–562.
[104] Bratteli, O., [1979], Crossed products of UHF-algebras by product type actions, Duke
Math. J. 46, 1–23.
[105] Bratteli, O., Elliott, G., [1978], Structure Spaces of Approximately Finite-Dimensional C*-
Algebras, II, J. Functional Analysis 30, 74–82.
[106] Bratteli, O., Elliott, G., Evans, D., Kishimoto, A., [1991], Noncommutative spheres, Inter-
nat. J. Math. 2, no.2, 139–166.
[107] Brown, L.G., [1977], Stable isomorphism of hereditary subalgebras of C*-algebras, Pacific
J. Math. 71, 335–348.
[108] , [1984], The Universal coefficient theorem for Ext and quasidiagonality, in: Oper-
ator Algebras and Group Representations, Proc. Int. Conf. Neptun (Romania) 1980, vol.1,
Pitman Press, Boston, Monogr. Stud. Math. 17, 60–64.
[109] , [2018], On the property IR of Friis and Rørdam, Preprint, arXiv:1809.07808v1
[110] , Douglas, R.G., and Fillmore, P.A. [1973], Unitary equivalence modulo the compact
operators and extensions of C*-algebras, Proceedings of a Conference in Operator Theory,
Lecture Notes in Mathematics 345, New York, Springer-Verlag.
[111] , Douglas, R.G., and Fillmore, P.A. [1977], Extensions of C*-algebras and K-
homology, Ann. of Math. 105(2), 265–324.
[112] , Green, P., Rieffel, M.A., [1971] Stable isomorphisms and strong Morita equivalence
of C*-algebras, Pacific J. Math. 71, 349–363.
[113] Brown, L.G., Pedersen, G.K., [1991], C*-algebras of real rank zero, J. Funct. Analysis 99,
131–149.
[114] Brown, L.G., , [1995], On the geometry of the unit ball of a C*-algebra, J. Reine
Angew. Math. 469, 113–147.
[115] Brown, L.G., , [2009], Limits and C*-algebras of low rank or dimension, J. Operator
Theory 61, 381–417.
[116] Brown, N.P., [2004], On quasidiagonal C*-algebras, Operator algebras and applications,
Adv. Stud. Pure Math., vol. 38, Math. Soc. Japan, Tokyo, pp. 19–64.
[117] Brown, N.P., [2004], Connes’ embedding problem and Lance’s WEP, Int. Math. Res. Not.,
2004, 501–510.
[118] Brown, N.P., [2006], Invariant means and finite representation theory of C*-algebras,
Mem. Amer. Math. Soc. 184, no. 865, viii+105.
[119] Brown, N.P., Browne, S.L., Willett,R., Wu,J., [2020], The UCT-Problem for nuclear C*-
algebras, Preprint (May 8, 2020.)
[120] Brown, N.P., Choda, M., [2001], Approximation entropies in crossed products
with an application to free shifts, Pacific Journal of Mathematics, 198(2), 331–346.
[121] Brown, N.P., Ciuperca, A., [2009], Isomorphism of Hilbert modules over stably finite C*-
algebras, J. Funct. Analysis, 257(1), 332–339.
[122] Brown, N.P., Ozawa, N.G., [2008], C*-algebras and finite-dimensional approximations,
Grad. Stud. Math. 88, American Mathematical Society, Providence 2008 (509 pp.)
[123] Bunce, J.W., Salinas, N., [1976], Completely positive maps and the left matricial spectrum
of an operator, Duke Math. J. 43, 747–774.
1320 BIBLIOGRAPHY

[124] Bunce, L.J., J.D. Maintland Wright, [1985], Quantum measures and states on Jordan alge-
bras, Comm. Math. Phys. 98, 187–202.
[125] Burgstaller, B., [2009], Notes on the nuclearity of random subalgebras of O2 , Monatshefte
für Math. 157(1), 1–11.
[126] Busby, R.C., [1968], Double centralizers and extensions of C*-algebras, Trans. Amer. Math.
Soc. 132, 79–99.
[127] Busby, R.C., Smith, H.A., [1970], Representations of twisted group algebras, Trans. Amer.
Math. Soc. 149, 503–537.
[128] Cantier, L., [2020], A unitary Cuntz semigroup for C*-algebras of stable rank one,
arXiv:2012.04338v1 (math.OA) 8 Dec 2020.
[129] Capraro, V., [2010], A survey on Connes’ embedding conjecture, arXiv:1003.2076.
[130] Capraro, V., Radulescu, F., [2011], Cyclic Hilbert spaces and Connes’ Embedding Problem,
Complex Analysis and Operator Theory, February 2011, arXiv:1102.5430v3 (math.OA) 17
Sep 2013.
[131] Castillejos, J., Evington, S., Tikuisis, A., White, S., Winter, W., [2021], Nuclear dimension
of simple C*-algebras, Invent. Math. 224(1), 245–290.
[132] Catterall, S., Wassermann, S., [2006], Continuous bundles of C*-algebras with discontinuous
tensor products, Bulletin of the London Mathematical Society, 38(4), 647–656.
[133] Cave, C. and Zacharias, J., [2018], Kirchberg-Wassermann exactness vs exactness: reduction
to the unimodular totally disconnected case, Münster Journal of Mathematics, 11, 217–224.
[134] Chavez-Dominguez, J.A., [2019], An Ando-Choi-Effros lifting theorem respecting subspaces,
J. Lond. Math. Soc. (2) 100, 914–936.
[135] Choi, M.-D., [1974], A Schwarz inequality for positive linear maps on C*-algebras, Illinois
J. Math., 18, 565–574.
[136] Choi, M.-D., [1975], Completely Positive Linear Maps on Complex Matrices, Linear Algebra
Appl., 10, 285–290.
[137] Choi, M.-D., [1979], A simple C*-algebra generated by two finite-order unitaries, Canad.
J. Math., 31(4), 867–880.
[138] Choi, M.-D., [1983], Lifting projections from quotient C*-algebras, J. Operator Theory
10(1), 21–30.
[139] , Christensen, E., [1983], Completely order isomorphic and close C*-algebras need
not be *-isomorphic, Bull. London Math. Soc. 15, 604–610.
[140] , Effros, E.G., [1976], The completely positive lifting problem for C*-algebras, Ann. of
Math. 104, 585–609.
[141] , Effros, E.G., [1976], Separable nuclear C*-algebras and injectivity, Duke Math. J.
43, 309–322.
[142] , , [1977], Nuclear C*-algebras and injectivity: the general case, Indiana Univ.
J. of Math.26, 443–446.
[143] , , [1977], Lifting Problems and the Cohomology of C*-algebras, Can. J. Math.
24(5), 1092–1111.
[144] , Effros, E.G., [1977], Injectivity and operator spaces, J. Funct. Analysis 24, 156–209.
[145] , , [1978], Nuclear C*-algebras and the approximation problem, Amer.
J. Math. 100, 61–79.
[146] , Elliott, G.A. [1990], Density of the selfadjoint elements with finite spectrum in an
irrational rotation C*-algebra, Math. Scand., 67, 73–86.
[147] Christensen, E., Effros, E.G., Sinclair, A., [1987], Completely bounded multilinear maps
and C*-algebraic cohomology, Invent. Math., 90, 279–296.
[148] Christensen, E., Sinclair, A., Smith, R., White, S., Winter, W., [2012], Perturbation of
nuclear C*-algebras, Acta Mathematica, 208, 93–150.
[149] Ciuperca, A., [2008], Some properties of the Cuntz semigroup and an isomorphism theorem
for a certain class of non-simple C*-algebras. Thesis for the Doctor degree on University of
Toronto, in the Library and Archives Canada, Ottawa.
 BIBLIOGRAPHY 1321

[150] Ciuperca, A., Elliott, G.A., [2008], A Remark on Invariants for C*-Algebras of Stable Rank
One, International Mathematics Research Notices, Volume 2008, rnm 158. Published: 01
January 2008. https://doi.org/10.1093/imrn/rnm158.
[151] Ciuperca, A., Giordano, T., Ng, P.W., and Niu, Z., [2013], Amenability and uniqueness,
Adv. Math. 240, 325–345.
[152] Ciuperca, A., Robert, L., Santiago, L., [2010], The Cuntz semigroup of ideals and quotients
and a generalized Kasparov stabilization theorem, J. Operator Theory 64, 155–169.
[153] Coburn, L.A., [1967], The C*-algebra generated by an isometry, Bulletin of the American
Mathematical Society 73, 722–726.
[154] Cohen, P.J. [1959], Factorization in group algebras, Duke Math. J. 26, 199–205.
[155] Combes, F., [1969], Sur les faces d’une C*-algèbre, Bull. Sci. Math. 93, 37–62.
[156] Comfort, W.W., Negrepontis, S. [1974], The theory of ultrafilters, Grundlehren der
math. Wiss. 211, Springer.
[157] Connes, A. [1975], A factor not anti-isomorphic to itself, Bull. Lond. Math. Soc. 7, 171–174.
[158] Connes, A. [1976], Classification of injective factors. Cases II1 , II∞ , IIIλ , λ 6= 1, Ann. of
Math. 104, 73–115.
[159] , [1978], On the cohomology of operator algebras, J. Funct. Analysis 28, 248–253.
[160] , [1979], Sur la théorie non-commutative de l’ integration, LMN 753, 19–143.
[161] , [1980], On the spatial theory of von Neumann algebras, J. Functional Analysis,
35,153–164.
[162] , [1981], An analogue of the Thom isomorphism for crossed products of a C*-algebra
by an action of R, Adv. in Math. 39, 31–55.
[163] , [1994], Noncommutative geometry, Academic Press, Inc., San Diego, New York etc.
[164] Connes Embedding Problem (and the equivalent QWEP conjecture) will be explained in
Chapter 11 ???...of what? (Perhaps in Pisier’s new book?).
[165] , Higson, N. [1990], Déformations, morphisms asymptotiques et K-théorie bivariante,
C.R. Acad. Sci. Paris, Sér. 1. Math. 311, 101–106.
[166] Coward, K.T., Elliott, G.A., Ivanescu, C., [2008], The Cuntz semigroup as an invariant for
C*-algebras, J. Reine Angew. Math. 623, 161–193.
[167] Cowling, M., Haagerup, U., [1989], Completely bounded multipliers of the Fourier algebra
of a simple Lie group of real rank one, Invent. Math. 96, 507–549.
[168] Crabb, M.J. [2006], A new prime C*-algebra that is not primitive, J. Funct. Analysis 236,
630–633.
[169] Cuntz, J. [1977], Simple C*-algebras generated by isometries, Commun. Math. Phys. 57(2),
173–185.
[170] , [1977], The structure of multiplication and addition in simple C*-algebras,
Math. Scand. 40, 215–233.
[171] , [1978], Dimension functions on simple C*-algebras, Math. Ann. 233, 145–153.
[172] , [1981], K-theory for certain C*-algebras, Ann. of Math. 113, 181–197.
[173] , [1981], K-theory for certain C*-algebras, II, J. Operator Theory, 5, 101–108.
[174] , [1981], A class of C*-algebras and topological Markov chains. II. Reducible chains
and the Ext-functor for C*-algebras, Invent. Math. 63, 181–197.
[175] , [1982], The internal structure of simple C*-algebras and applications, Proc. Symp.
Pure. Math. (R. V. Kadison, ed.) 38(1), American Mathematical Society, 85–115.
[176] , [1983], K-theoretic amenability for discrete groups, Journal für die reine und ange-
wandte Mathematik 344, 180–195.
[177] , [1984], On the homotopy groups of the space of endomorphisms of a C*-algebra
(with applications to topological Markov chains), in: Operator algebras and group represen-
tations, Vol. I (Neptun, 1980), volume 17 of Monogr. Stud. Math., pages 124–137. Pitman,
Boston, MA, 1984.
1322 BIBLIOGRAPHY

[178] , [2008], C*-algebras associated with the ax + b -semigroup over N, K-theory and
noncommutative geometry, EMS Ser. Congr. Rep., Eur. Math. Soc., Zürich, 2008, pp. 201–
215.
[179] , [2011], C*-algebras associated with integral domains and crossed products by actions
on adele spaces, J. Noncommutative Geometry 5, 1–37.
[180] , Higson, N. [1987], Kuiper’s theorem for Hilbert modules, pp. 429–434, in: Pro-
ceedings of the Summer Conference on Operator algebras and Physics, 1985: University
of Iowa, editors: Jorgensen, P. E.T., Muhly, P.S., Contemp. Math. 62, Amer. Math. Soc.,
Providence.
[181] , Deninger, Ch., Laca, M., [2013], C*-algebras of Toeplitz type associated with alge-
braic number fields, Math. Ann. 355(4), 1383–1423.
[182] , Echterhoff, S., Li, X., Yu, G., [2017], K-Theory for Group C*-Algebras and Semi-
group C*-Algebras, Oberwolfach seminars, vol. 47, Birkhäuser/Springer, Cham, 2017.
[183] , Krieger, W. [1980], A class of C*-algebras and topological Markov chains, In-
vent. Math. 56, 251–268.
[184] , Li, X. [2010], The regular C*-algebra of an Integral Domain, Quanta of maths, Clay
Math. Proc. 11, 149–170, Amer. Math. Soc., Providence, RI.
[185] , Li, X. [2009], K-theory of ring C*-algebras associated to function fields,
arXiv:0911.5023v1, 2009.
[186] , Pedersen, G.K. [1979] Equivalence and traces on C*-algebras, J. Funct. Analysis
33, 135–164.
[187] , Skandalis, G. [1986], Mapping cones and exact sequences in KK-theory, J. Operator
Theory 15:1, 163–180.
[188] Dadarlat, M. [1994], Shape theory and asymptotic morphisms for C*-algebras, Duke
Math. J. 73, 687–711.
[189] , [2000], Nonnuclear subalgebras of AF-algebras, American J. Math. 122:3, 581–597.
[190] , [2000], Approximate Unitary Equivalence and the Topology of Ext (A, B). In: Cuntz
J., Echterhoff S. (eds) C*-Algebras. Springer, Berlin, Heidelberg.
[191] , [2006], Continuous fields of C*-algebras over finite dimensional spaces, Preprint:
math.OA/0611405.
[192] , [2007], The homotopy groups of the automorphism group of Kirchberg algebras,
J. Noncommut. Geom. 1(1), 113–139.
[193] , [2009], Fiberwise KK-equivalence of continuous fields of C*-algebras, J. K-Theory 3,
205–219.
[194] , [2011], A stably contractible C*-algebra which is not contractible, Preprint (2011?),
all references are from before 2003, Paper published?
[195] On the classification of nuclear C*-algebras, Proc. London Math. Soc. (3) 85 (2002) 168–210
[196] , Elliott, G. [2007], One-parameter continuous fields of Kirchberg algebras, Com-
mun. Math. Phys. 274(3), 795–819.
(first preprint in 2005)
[197] , Hirshberg, I., Toms, A.S., Winter, W., [2009], The Jiang-Su algebra does not always
embed, Math. Res. Lett. 16(1), 23–26. International Press 2009.
(Preprint arXiv:0712.2020v1)
[198] , Loring, T. [1996], Classifying C*-algebras via ordered, mod-p K-theory,
Math. Ann. 305, 601–616.
[199] , Pasnicu, C. [2005], Continuous fields of Kirchberg C*-algebras, Journal of Func-
tional Analysis 226(2), 429–451.
(Preprint: November 17, 2004.)
[200] , Penning, U. [2013], Unit spectra of K-Theory from strongly self-absorbing C*-
algebras, Algebr. Geom. Topol. 15(2015),137–168. arXiv:
1306.2583v1
 BIBLIOGRAPHY 1323

[201] , Penning, U. [2014], A Dixmier-Douady Theory for strongly self-absorbing C*-

algebras II: the Brauer group, Journal of Noncommutative Geometry 9(4), 1661–6952.
[202] , Toms, A.S., [2009], A universal property for the Jiang–Su algebra, Adv. Math. 220,
341–366.
[203] , Toms, A.S., [2010], Ranks of operators in simple C*-algebras, J. Funct. Analy-
sis 259(5), 1209–1229.
[204] , Winter, W. [2007], On the KK-theory of strongly self-absorbing C*-algebras, pre-
print: math.OA/0704.0583
[205] , Winter, W. [2008], Trivialization of C(X)-algebras with strongly self-absorbing
fibres, Bull. Soc. Math. France, 136, 575–606.
[206] Davidson, K.R., [1991], Lifting positive elements in C*-algebras, Integral Eq. and Operator
Theory 14, 183–191.
[207] Davidson, K.R., [1996], C*-algebras by example, Fields Institute monographs, AMS, Provi-
dence, Rhode Island.
[208] Davie, A.M., [1976], Classification of essentially normal operators, in “Spaces of Ana-
lytic Functions”, Seminar Kristiansand, Norway, June 9-14, 1975, Eds.: O. B. Bekken,
B.K. Øksendal, A. Stray, Lecture Notes in Mathematics, Vol. 512, 31–55.
[209] De Schreye, D., Van Daele, A. [1981], A new technique to deal with Cuntz-algebras,
Math. Annalen 257, 257–501.
[210] Dieudonné, J., [1942], Sur le socle d’un anneau et les anneaux simples infinis, Bull. Soc.
Math. France 70, 46–75.
[211] Dixmier, J. [1950], Les moyennes invariantes dans les semi-groupes et leurs applications,
Acta Sci. Math. (Szeged) 12, 213–227.
[212] Dixmier, J. [1952], Algèbres quasi-unitaires, Comm. Math. Helv. 26, 275–322.
[213] Dixmier, J. [1960], Les C*-algèbres et leurs représentations, Bull. Soc. Math. France 88,
95–112.
[214] , [1964], Traces sur les C*-algèbres II, Bull. Sci. Math. 88, 39–57.
[215] , [1968], Sur les espaces localement quasi-compact, Canad. J. Math. 20, 1093–1100.
[216] , [1969], Les Algèbres d’opérateurs dans l’espace Hilbertien, (Algèbres de von Neu-
mann), Gauthier-Villars, Paris, 1969.
[217] Dixmier, J. [1969], Les C*-algèbres et leurs représentations, Gauthier-Villars, Paris.
(Engl. translation by Ch. Lance: C*-algebras, North Holland, Amsterdam (1982).)
[218] , Douady, A., [1963], Champs continus d’espaces hilbertiens et de C*-algèbres,
Bull. Soc. Math. France 91, 227–284.
[219] Dor-On, A., Eilers, S., Geffen, S., [2020], Classification of irreversible and reversible Pim-
sner operator algebras, Compositio Mathematica, 156(12), 2510–2535.
Published online by Cambridge University Press: 13 January 2021 (Print publication: De-
cember 2020).
[220] Douglas, R.G., [1966], On majorization, factorization, and range inclusion of operators on
Hilbert space, Proc. Amer. Math. Soc. 17, 413–415.
[221] Dye, H.A., [1955], On the geometry of projections in certain operator algebras, Ann. of
Math. (2) 61, 73–89.
[222] Dye, H. A., Russo, B., [1966], A note on unitary operators in C*-algebras, Duke Math. J.
33, 413–416.
[223] Dykema, K.J., [1998], Faithfulness of free product states, J. Funct. Anal. 154, 223–229.
[224] Dykema, K.J., [1999], Simplicity and the stable rank of some free product C*-algebras,
Trans. Amer. Math. Soc. 351, 1–40.
[225] Dykema, K.J., [2002], Purely infinite, simple C*-algebras arising from free product con-
structions, II, Math. Scand. 90, 73–86.
[226] Dykema, K.J., Rørdam, M., [1998], Purely infinite, simple C*-algebras arising from free
product construction, Can. J. Math. 50(2), 323–341.
1324 BIBLIOGRAPHY

[227] Dykema, K.J., Rørdam, M., [1998], Projections in free product C*-algebras, GAFA,
Geom. Funct. Anal. 8(1), 1–16. Erratum [2000]: Geom. Funct. Anal. 10(4), p. 975.
[228] Dykema K., Rørdam, M., [2000], Projections in free product C*-algebras, II, Math. Z. 234,
103–113.
[229] Dykema, K.J., Shlyakhtenko, D., [2001], Exactness of Cuntz-Pimsner C*-algebras, Proceed-
ings of the Edinburgh Math. Soc. 44, 425–444.
[230] Dykema, K.J, Smith, R.R., The completely bounded approximation property for extended
Cuntz-Pimsner algebras, (preprint 11 Nov 2003 or later?)
[231] Eagle, Ch.J., Farah, I., Kirchberg, E., Vignati, A., [2015], Quantifier elimination in C*-
algebras, arXiv:1502.00573v3, revised 13 Jun 2015.
[232] Easo, Ph., Garijo, E., Kaubrys, S., Nkansah D., Vrabec M., Watt D., Wilson C.,
Bönicke Ch., Evington S., Forough M., Pacheco S.G., Seaton N., White S., Whittaker M.F.,
Zacharias, J., [2020], The Cuntz-Toeplitz algebras have nuclear dimension one, Journal of
Functional Analysis 279(7), 108690.
[233] Effros, E.G., [1963], Order ideals in a C ∗ -algebra and its dual, Duke Math. J. 30, 391–412.
[234] Effros, E.G., [1965], Convergence of closed subsets in a topological space, Proc. AMS 16,
929–931.
[235] , [1978], Aspects of noncommutative order, Lecture Notes in Math. 650: C ∗ -algebras
and applications to physics, Springer, Berlin, Heidelberg, New York.
[236] , [1981], AF-algebras, Dimensions and C ∗ -algebras, CBMS Regional conference series
in Mathematics, 46, AMS, Providence R.I.
[237] , [1982], On the structure theory of C ∗ -algebras: Some old and new problems, Op-
erator algebras and applications, Part 1 (Kingston, Ont., 1980), pp.19–34, Proc. Sympos.
Pure Math. 38, Amer. Math. Soc., Providence, R.I.
[238] Effros, E.G., Haagerup, U., [1985], Lifting problems and local reflexivity for C*-algebras,
Duke Math. J. 52, 103–128.
[239] , Hahn, F., [1967], Locally compact transformation groups and C*-algebras, Memoirs
of the AMS, 75, American Mathematical Society, Providence, R.I., 1967.
[240] , Kaminker, J., [1986], Homotopy continuity and shape theory for C*-algebras, In:
Geometric methods in operator algebras (Kyoto, 1983), pages 152–180. Longman Sci. Tech.,
Harlow, 1986.
[241] , Lance, Ch. [1977], Tensor products of operator algebras, Advances in Math. 25,
1–34.
[242] , Ozawa, N., Ruan, Zh.-J. [2001], On injectivity and nuclearity for operator spaces,
Duke J. Math. 110, 489–521.
[243] , Rosenberg, J., [1978], C*-algebras with approximately inner flip, Pacific
J. Math. 77, 417–443.
[244] , Ruan Zh.J., [1990], On approximation properties for operator spaces, International
Journal of Math.(IJM) 1, 163–187.
[245] , Ruan Zh.J., [2000], Operator spaces, (LMS Monographs, New Series 23) Clarendon
Press, Oxford.
[246] , Shen Ch.L., [1980], Approximately finite C*-algebras and continued fractions, In-
diana Univ. Math. J. 29(2),191–204.
[247] , Størmer E., [1979], Positive projections and Jordan structure in operator algebras,
Math. Scand. 45, 127–138.
[248] Eilers S., Loring T.A., [1999], Computing contingencies for stable relations, Inter-
nat. J. Math. 10, 301–326.
[249] Eilers, S., Restorff, G., [2006], On Rørdam’s Classification of Certain C*-Algebras with
One Non-Trivial Ideal, In: Bratteli O., Neshveyev S., Skau C. (eds) Abel Symposia, vol 1.
Springer, Berlin, Heidelberg, Operator algebras: The Abel symposium 2004, Abel Symp.,
1, Springer, Berlin, 2006, pp. 87–96.
 BIBLIOGRAPHY 1325

[250] Eilers, S., Restorff, G., and Ruiz, E., [2011], Non-splitting in Kirchberg’s Ideal-related KK-
Theory, Canad. Math. Bull. 54(1), 68–81.
[251] Eilers, S., Restorff, G., and Ruiz, E., [2019], The complete classification of unital graph
C*-algebras: Geometric and strong, Preprint, arXiv:1611.07120(v2), last revised 25 Aug
2019.
[252] Eilers, S., Tomforde, M., [2009], On approximately finite-dimensional nonsimple graph C*-
algebras, Math. Ann. 346, Article: 393 (2010).
[253] Elliott, G.A., [1978], On approximately finite-dimensional von Neumann algebras. II,
Canad. Math. Bull. 21, 415–418.
[254] , [1980], Some simple C*-algebras constructed as crossed products with discrete outer
automorphism groups, Publ. Res. Inst. Math. Sci. Kyoto Univ. 16, 299–311.
[255] , [1990], Dimension groups with torsion, Int. J. Math. 1,1 (1990), 361–380.
[256] , [1993], A classification of certain simple C*-algebras, pages 373–385 in: Quantum
and Non-Commutative Analysis, H. Araki etc. (eds.), Kluwer, Dordrecht, 1993.
[257] , [1993], On the classification of C*-algebras of real rank zero, J. reine
angew. Math. 443, 179–219.
[258] , [1995], The classification problem for amenable C*-algebras, Proceedings of the
International Congress of Mathematicians, Zürich, 1994, Birkhäuser Verlag, Basel, 922–932.
[259] , [1996], An invariant for simple C*-algebras, pp. 61–90, in “Invited Papers, Cana-
dian Mathematical Society 1945-1995, Vol.3” (Eds. J.B. Carrell and R. Murphy). Canadian
Mathematical Society, Ottawa.
[260] , [2007], Hilbert modules over a C*-algebra of stable rank one, C.R. Math.
Acad. Sci. Soc. R.Can. 29,48–51.
[261] , Evans, D.E. [1993], The structure of irrational rotation C*-algebras, Ann. of
Math. 138, 477–501.
[262] , Gong, G. [1996], On the classification of C*-algebras of real rank zero, II, Ann. of
Math. 144(3), 497–610.
[263] , Gong, G., and Li, L. [2007], On the classification of simple inductive limit C*-
algebras, II: The isomorphism theorem, Invent. math. 168, 249–320. (preprint 1999)
[264] , Kucerovsky, D., [2001], An abstract Voiculescu–Brown–Douglas–Fillmore absorp-
tion theorem, Pacific J. Math. 198(2), 385–409.
(Compare a correction given in [310].)
[265] , Niu, Z. [2008], On tracial approximation, J. Funct. Analysis 254, 394–440.
[266] , Niu, Z. [2012], Extended rotation algebras: Adjoining spectral projections to rotation
algebras, J. Reine Angew. Math. 665, 1–71.
[267] , Niu, Z. [2017], The C*-algebra of a minimal homeomorphism of zero mean dimen-
sion, Duke Math. J. 166, 3569–3594.
[268] , Robert, L., Santiago, L., [2011], The cone of lower semicontinuous traces on a
C*-algebra, Amer. J. Math. 133, 969–1005. arXiv:0805.3122v2 (21 Jan 2009).
[269] , Rørdam, M. [1995], Classification of certain infinite simple C*-algebras II, Com-
ment. Math. Helv. 70, 615–638.
[270] , , [2006], Perturbation of Hausdorff moment sequences, and an application
to the theory of C*-algebras of real rank zero, pages 97–115 in Operator algebras, The Abel
Symposium 2004, (Eds.) O. Bratteli,S. Neshveyev,C. Skau, Springer, Berlin, etc., X+279 p.
[271] Elliott, G.A., Sierakowski, A., [2012(?)], Paper and title forgotten or not existing !!!
[272] , Thomsen, K., [1994], The state space of the K0 -group of a simple separable C*-
algebra, Geom. Funct. Analysis 4, 522–538.
[273] , Villadsen, J. [2000], Perforated ordered K0 -groups, Canad. J. Math. 52, 1164–1191.
[274] Enders, D., [2015], On the nuclear dimension of certain UCT-Kirchberg algebras, published
2014-05-26, J. Funct. Anal., 268(9): 2695–2706.
[275] Enders, D., [2015], Semiprojectivity for Kirchberg algebras, arXiv:1507.06091 [math.OA] 22
Jul 2015.
1326 BIBLIOGRAPHY

[276] Enders, D., Shulman, T., [2021], Commutativity of central sequence algebras, 2106.00271v1
[math.OA ] 1 Jun 2021.
[277] Enock, M., Schwartz, J.M., [1975], Une duality dans les algèbres de von Neumann, Mémoire
Soc. Math. France, suppl. Bull., 44, 1–144.
[278] Enock, M., Schwartz, J.M., [1986], Algèbres de Kac moyennables, Pacific J. Math., 125,
363–379.
[279] Evans, D.E., [1980], On On , Publ. Res. Inst. Math. Sci., 16, 915–927.
[280] Evans, D.E., [1982], Gauge actions on OA , J. Operator Theory, 7, 79–100.
[281] Exel, R., [1987], Rotation numbers for automorphisms of C ∗ algebras, Pacific Journal of
Mathematics, 127, 31–89.
[282] Exel, R., [1993], The soft torus and applications to almost commuting matrices, Pacific
Journal of Mathematics, 160, 207–217.
[283] Exel, R. [1994], Circle actions on C*-algebras, Partial automorphisms and a Generalized
Pimsner–Voiculescu exact sequence, J. Funct. Anal., 122, 361–401.
[284] Exel, R. [1997], Twisted partial actions: A classification of regular C*-algebraic bundles,
Proc. London. Math. Soc., 74, 417–443.
[285] Exel, R. [1997], Amenability for Fell bundles, J. Reine Angew. Math., 492, 41–73.
[286] Exel, R. [2000], Partial representations and amenable fell bundles over free groups, Pacific
Journal of Mathematics 192, 39–63.
[287] Exel, R., [2002], Exact groups and Fell bundles, Math. Annalen 323, 259–266.
[288] Eymard, P., [1964], L’algèbre de Fourier d’un groupe localement compact,
Bull. Soc. Math. France 92, 181–236.
[289] Fack, T., Skandalis, G., [1982], Sur les représentations et idéaux de la C*-algèbre d’un
feuilletage, J. Operator Theory 8, 95–129.
[290] Fang X., Hua, J., Xu, X.-M., [2019], Continuity of functors with respect to generalized
inductive limits, Frontiers of Mathematics in China 14(2), 551–566.
[291] Farah, I., [2011], All automorphisms of the Calkin algebra are inner, Ann. of Math.,
173(2),,619–661.
[292] Farah, I., Hart, B., [2013], Countable saturation of corona algebras, C.R. Math. Rep. Acad.
Sci. Canada 35, 35–56.
(arXiv:1112.3898v2, revised 17 Oct 2012)
[293] Farah, I., Hart, B., Rordam, M., Tikuisis, A., [2017], Relative commutants of strongly self-
absorbing C*-algebras, Selecta Math. (New Series) 23 (1), 363–387. (arXiv:1502.05228v3
revised 16.April 2016.)
[294] Farah, I., Phillips, N.Ch., [2013], The commutant of L(H) in its ultrapower may or may
not be trivial, arXiv:0808.3763v2 (1 Oct 2009).
[295] Farenick, D., Kavruk, A.S., Paulsen, V.I., Todorov, I.G., Characterisations of the weak
expectation property, arXiv:13071055v1 (3 Jul 2013).
[296] Fell, J.M.G., [1960], The dual space of C*-algebras, Trans. AMS 94, 365–403.
[297] , [1962], A Hausdorff topology for the closed subsets of a locally compact Hausdorff
space, Proc. AMS, 13, 472–476.
[298] , [1962], The structure of algebras of operator fields, Acta Math., 106, 237–268.
[299] Fell, J.M.G., Doran, R.S., [1988], Representations of *-algebras, locally compact groups,
and Banach *-algebraic bundles. Vol.1, Pure and Applied Mathematics, vol. 125, Aca-
demic Press Inc., Boston, MA, 1988. Basic representation theory of groups and algebras.
MR90c:46001.
[300] Fillmore, P.A., [1996], A user’s guide to operator algebras, Canadian Mathematical Society
series of monographs and advanced texts, Wiley-Intersience.
[301] Forough, M., Gardella, E., Thomsen, K., [2021], Asymptoting lifting for completely positive
maps, arXiv:2103.09176.
[302] Forough, M., Golestani, N., [2021], The weak tracial Rokhlin property for finite group ac-
tions on simple C*-algebras, Doc. Math. 25, 2507–2552.
 BIBLIOGRAPHY 1327

[303] Frank, M., [1999], Geometrical aspects of Hilbert C*-modules, Positivity 3, 215–243.
[304] Frank, M., Kirchberg, E., [1998], On conditional expectations of finite index, J. Operator
Theory, 40, 87–111.
[305] Friis, P., Rørdam, M., [1996], Almost commuting self-adjoint matrices - a short proof of
Huaxin Lin’s theorem, J. Reine Angew. Math. 479, 121–131.
[306] Fritz, T., [2012], Tsirelson’s problem and Kirchberg’s conjecture, Reviews in Mathematical
Physics, 24(05):1250012.
[307] Fu, X., Lin, H., [1970], Non-amenable simple C*-algebras with tracial approximation,
arXiv:2101.07900.
[308] Fuchs, L. [1970], Infinite Abelian Groups, Volume 1. Academic Press, New York and London.
[309] Gabe, J., [2016], A note on non-unital absorbing extensions, Pacific J. Math.284(2), 383–
393.
[310] , [2016], Lifting theorems for completely positive maps, Preprint, arXiv:1508.00389v3.
[311] , [2017], Quasidiagonal traces on exact C*-algebras, J. Funct. Anal., 272(3), 1104–
1120.
[312] , [2020], Traceless AF embeddings and unsuspended E-theory, Geom. Funct. Anal.,
30(2), 323–333.
[313] , [2020], A new proof of Kirchberg’s O2 -stable classification, J. Reine Angew. Math.
761, 247–289. (Online available since 16.05.2018)
[314] , [2019], Classification of O∞ -stable C*-algebras, arXiv:1910.06504v3.
[315] , Ruiz, E. [2015], Absorbing representations with respect to operator convex cones,
arXiv:1503.07799.
[316] , Ruiz, E. [2020], The unital Ext-groups and classification of C*-algebras, Glasgow
Math. J., 62(1), 201–231.
[317] Gardella, E., [2014], Existence and uniqueness theorems for circle actions on Kirchberg
algebras, Lecture at Fields Institute, Toronto 2014.
[318] Gardella, E., Hirshberg, I., Santiago, L., [2021], Rokhlin dimension: duality, tracial proper-
ties, and crossed products, Ergodic Th. Dynam. Syst. 41(2), 408–460.
[319] Ge, L., Hadwin, D., [2001], Ultraproducts of C*-algebras, Operator theory: advances and
applications, 127, 305–326.
[320] Ghez, P., Lima, R., Roberts, J.E., [1985], W*-categories, Pacific J. Math., 120(1), 79–109.
[321] Gierz, G., Hofmann, K.H., Keimel, K., Lawson, J.D., Mislove, M., and Scott, D.S., [2003],
Continuous Lattices and Domains, Encyclopedia of Mathematics and its Applications,
Vol. 93. Cambridge Univ. Press.
[322] Glimm, J. [1960], On a certain class of operator algebras, Trans. AMS 95, 318–340.
[323] , [1960], A Stone-Weierstrass theorem for C*-algebras, Ann. of Math. 72, 216–244.
[324] , [1961], Type I C*-algebras, Ann. of Math. 73, 572–612.
[325] Isaac Goldbring and Thomas Sinclair, [2015], On Kirchberg’s embedding problem, Journal
of Functional Analysis 269(1), 155–198.
[326] Gong, G. [1994], Approximation by dimension drop C*-algebras and classification,
C.R. Math. Rep. Acad. Sci. Canada, Vol. XVI, No.1, February 1994, 40–44.
[327] Gong, G., [2002], On the classification of simple Inductive limit C*-Algebras, I: The reduc-
tion theorem, Documenta Math. 7, 255–461.
[328] Gong G., Lin H., Niu Zh., [2018], A classification of finite simple amenable Z-stable C*-
algebras, I: C*-algebras with generalized tracial rank one, Preprint, arXiv:1812.11590v1.
[329] Gong G., Lin H., Niu Zh., [2019], Classification of finite simple amenable Z-
stable C*-algebras, II: C*-algebras with rational generalized tracial rank one, Preprint,
arXiv:1909.13382v1.
[330] , Jiang, X., Su, H., [2000], Obstructions to Z-stability for unital simple C*-algebras,
Canadian Math. Bull. 43, no. 4, 418–426.
[331] Goodearl, K., Handelman, D. [1976], Rank functions and K0 of regular rings, Appl. Alge-
bra 7, 195–216.
1328 BIBLIOGRAPHY

[332] Graetzer, G., (or Grätzer, G., ?), Lattice theory: Foundation, Birkhäuser, Basel, 2010.
or
Grätzer, G., General lattice theory, Springer, Basel 1998
WE need explicit references to this books or equivalent results in other Lattice books!
[333] Greenleaf, F.P., [1969], Invariant means on topological groups and their applications,
Van Nostrand Mathematical Studies, No.16, Van Nostrand Reinhold Co., New York -
Toronto, Ont. - London, 1969.
[334] Grigorchuk, R., Musat, M., and Rørdam, M., [2018], Just-infinite C*-algebras, Commentarii
Math. Helvetici, 93(1), 157–201. (arXiv:1604.08774v3).
[335] Gromov, M., [2003], Random walk in random groups, GAFA 13, 73 –146.
[336] Grothendieck, A. [1955], Produits tensoriels topologiques et espaces nucléaires, Memo.
Amer. Math. Soc., 16.
[337] Grove, K., Pedersen, G.K., [1984], Sub-Stonean spaces and corona sets J. Funct. Analysis
56, 124–143.
[338] Guentner, E., Kaminker, J., [2002], Exactness and the Novikov conjecture, Topology 41(2),
411–418.
[339] Haagerup, U. [1978], An example of a non nuclear C*-algebra, which has the metric ap-
proximation property, Invent. Math. 50(3), 279–293.
[340] Haagerup, U. [1983], All nuclear C*-algebras are amenable, Invent. Math. 74, 305–319.
[341] , [1985], The Grothendieck inequality for bilinear forms on C*-algebras,
Adv. Math. 56, 93–116.
[342] , [1991,2014], Quasitraces on exact C*-algebras are traces, C.R.M. Rep. Acad. Sci.
Canada 36, 67–92. Manuscript 1991, was revised 2014 as Preprint arXiv:1403.7653v1.
[343] , Kraus, J., [1994], Approximation properties for group C*-algebras and group von
Neumann algebras, Trans. Amer. Math. Soc. 344, 667–699.
[344] , Musat, M., [2008], The Effros-Ruan conjecture for bilinear forms on C*-algebras,
Invent. Math. 174, 139–163.
[345] , Musat, M., [2011], Factorization and dilation problems for completely positive maps
on von Neumann algebras, Comm. Math. Phys. 303(2), 555–594.
[346] , Musat, M., [2015], An asymptotic property of factorizable completely positive maps
and the Connes embedding problem, Comm. Math. Phys., 338(2), 721–752.
[347] , Pisier, G., [1993], Bounded linear operators between C*-algebras, Duke Math. J., 71,
889–925.
[348] , Thorbjørnsen, S. [1999], Random matrices and K-theory for exact C*-algebras,
Documenta Math. 4, 341–450.
[349] , Thorbjørnsen, S. [2005], A new application of random matrices: Ext(Cred ∗ (F )) is
2
not a group, Annals of Math. 162, 711–775.
[350] , Winsløw, C. [2000], The Effros-Maréchal topology in the space of von Neumann
algebras, Journal of Functional Analysis 171, 401–431.
[351] , Zsido, L., [1984], Sur la propriété de Dixmier pour les C*-algèbres, C. R. Acad. Sci.
Paris, Ser. I Math. 298, no. 8, 173–176.
[352] Halmos, P.R. [1950], Normal dilations and extension of operators, Summa Brasiliensis
Math. 2, 125–134.
[353] , [1969], Two subspaces, Trans. Amer. Math. Soc. 144, 381–389.
[354] Halperin, S., [1992], Universal Enveloping Algebras and Loop Space Homology, J. Pure
Appl. Alg. 83, 237–282.
[355] Handelman, D., [2018], Invariants for permutation-Hermite equivalence and critical dimen-
sion groups, Münster J. of Math. 11, 49–156.
[356] Hansen, F. [2013], The fast track to Löwner’s theorem, Linear Algebra and its Applications
438(11), 4557–4571. (arXiv:1112.0098v3)
[357] , Pedersen, G.K., [1982], Jensen’s inequality for operators and Löwner’s theorem,
Math. Ann. 258, 229–241.
 BIBLIOGRAPHY 1329

[358] , , [2003], Jensen’s operator inequality, Bulletin LMS 35:04, 553–564.

(arXiv:math/0204049v1)
[359] Harnisch, H., Kirchberg, E. [2005], The inverse problem for primitive ideal spaces, SFB
478-preprint, Heft 399, www.math.uni-muenster.de
[360] Hausdorff, F. [1914], Grundzüge der Mengenlehre, Teubner, Leipzig.
[361] Herman, R.H., Vaserstein, L.N. [1984], The stable range of C*-algebras, Invent. Math. 77,
553–555.
[362] Herman, R.H., Putnam, I.F., Skau, Ch.F. [1992], Ordered Bratteli diagrams, dimension
groups and topological dynamics, International Journal of Math. 3(6), 827–864.
[363] Hewitt, E. [1964], The ranges of certain convolution operators, Math. Scand. 15, 147–155.
[364] Higson, N. [1987], A characterization of KK-theory, Pacific J. Math. 126, 253–276.
[365] , Lafforgue, V., Skandalis, G., [2002], Counterexamples to the Baum-Connes conjec-
ture, Geometric and Functional Analysis (GAFA) 12, 330–354.
[366] Hirshberg, I., Kirchberg, E., White, S. [2012], Decomposable approximations of nuclear
C*-algebras, Advances in Math. 230(3),1029–1039.
[367] Hirshberg, I., Orovitz, J., [2013], Tracially Z-absorbing C*-algebras, J. Funct. Anal. 265,
765–785.
[368] Hirshberg, I., Phillips, N.C., [2015], Rokhlin dimension: obstructions and permanence prop-
erties, Doc. Math. 20, 199–236.
[369] Hirshberg, I., Rørdam, M., Winter, W. [2007], C0 (X)-algebras, stability and strongly self-
absorbing C*-algebras, Math. Annalen 339, 695–732. preprint: arXiv 0610344v1.
[370] Hirshberg, I., Szabo, G., Winter, W., Wu, J. [2017], Rokhlin dimension for flows,
Comm. Math. Phys. 353(1), 253–316.
[371] Hirshberg, I., Winter, W. [2007], Rokhlin actions and self-absorbing C*-algebras, Pacific
J. Math. 233, 125–143.
[372] Hirshberg, I., Winter, W., Zacharias J. [2015], Rokhlin dimension and C*-dynamics, Com-
munications in Mathematical Physics 335(2), 637–670.
[373] Hjelmborg, J.v.B., Rørdam, M. [1998], On stability of C*-algebras, J. Funct. Analysis 155,
153–170.
[374] Holub, J.R., [1973], Reflexivity of L(E, F ), Proc. Amer. Math. Soc. 39, 175–177.
[375] Huruya, T., [1979], An intersection result for tensor products of C*-algebras, Proc. AMS 75,
186–187.
[376] Huruya, T., [1980], Fubini products of C*-algebras, Tôhoku Math. Journ. 32, 63–70.
[377] Isono, Y., [2013], Weak exactness for C*-algebras and application to condition (AO),
J. Funct. Analysis 264, 964–998.
[378] Izumi, M., [1993], Subalgebras of C*-algebras with finite Watatani indices, I. Cuntz Algebras,
Commun. Math. Phys. 155, 157–182.
[379] , [1998], Subalgebras of infinite C*-algebras with finite Watatani indices II, Cuntz-
Krieger algebras, Duke Math. J. 91, 409–461.
[380] , [2002], , [2002], Inclusions of simple C*-algebras, J. Reine
Angew. Math. 547, 97–138.
[381] , [2004], Finite group actions on C*-algebras with the Rohlin property I, Duke
Math. J. 122(2), 233–280.
[382] , [2004], Finite group actions on C*-algebras with the Rohlin property II,
Adv. Math. 184, 119–160.
[383] , [2001], The Rohlin property for automorphisms of C*-algebras Mathematical
physics in mathematics and physics (Siena, 2000), 191–206, Fields Inst. Commun., 30,
Amer. Math. Soc., Providence, RI.
[384] , Matui, H., [2010], Z2 -actions on Kirchberg algebras, Advances in Math. 224, 355–
400.
[385] , Matui, H., [2010], A weak homotopy equivalence type result related to Kirchberg
algebras, Journal of Noncommutative Geometry, 14(4).
1330 BIBLIOGRAPHY

[386] Jacelon, B., [2013], A simple, monotracial, stably projectionless C*-algebra, J. London
Math. Soc. (2) 87, 365–383.
[387] Jacelon, B., Winter, W., [2015], Z is universal, J. Noncommut. Geom. 8, 1023–1042.
[388] Jacobson, N. [1956], Structure of Rings, AMS Colloq. Publ. Providence (1956).
[389] Jensen, K.K., Thomsen, K. [1991], Elements of KK-theory, Birkhäuser, Boston, Basel,
Berlin.
[390]
[391] Jiang, X., Su, H. [1999], On a simple unital projectionless C*-algebra, Amer. J. Math. 121,
359–413.
[392] Johanesová, M. , Winter, W., [2012], The similarity problem for Z-stable C*-algebras, Bul-
letin of The London Mathematical Society, 44(6), 1215–1220.
[393] Johnson, B.E., Williams, J.P., [1975], The range of a normal derivation, Pacific J. Math. 58,
105–122.
[394] Jolissaint, P. , Robertson, G., [2000], Simple purely infinite C*-algebras and n-filling actions,
J. Funct. Analysis 175, 197–213.
[395] Junge, M., [2004], A Fubini type theorem for non-commutative Lp spaces,
Canad. J. Math. 56, 983–1021.
[396] Junge, M., Musat, M., [2007], Trans. Amer. Math. Soc., 359(1), 115–142.
[397] Junge, M., Pisier, G., [1995], Bilinear forms on exact operator spaces and B(H) ⊗ B(H),
Geometric & Functional Analysis (GAFA Journal) 5, 329–363.
[398] Kaad, J., [2017], Differentiable absorption of Hilbert C*-modules, connections, and lifts of
unbounded operators, J. Noncommut. Geom. 11, 1037–1068.
[399] Kadison, R.V., [1952], A generalized Schwarz inequality and algebraic invariants for operator
algebras, Ann. of Math. 56, 494–503.
[400] , Ringrose, J.R. [1983], Fundamentals of the theory of operator algebras, Volume I,
Elementary theory, Academic Press, San Diego, etc.
[401] , , [1986], Fundamentals of the theory of operator algebras, Special Topics,
Volume III, Elementary Theory – An Exercise Approach, Academic Press, San Diego, etc.
[402] , , [1986] Fundamentals of the theory of operator algebras, Special topics,
Volume IV, Advanced Theory – An Exercise Approach, Academic Press, San Diego, etc.
[403] Kaplansky, I., [1953], Modules over operator algebras, Amer. J. Math. 75, 839–858.
[404] Kasparov, G.G. [1980], Hilbert C*-modules: Theorems of Stinespring and Voiculescu, J. Op-
erator Theory 4, 133–150.
[405] , [1981], The operator K-functor and extensions of C*-algebras, Math. USSR
Izvestiya 16, 513–572.
[406] , [1988], Equivariant KK-theory and the Novikov conjecture, Invent. Math. 91, 147–
201.
[407] , [1991], Relative K-homology and K-homology of an ideal, K-theory 5, 47–49.
[408] , Skandalis, G. [1991], Groups acting on buildings, Operator K-theory and Novikov
conjecture, K-theory 4, 303–337.
[409] Katsura, T. [2004], On C*-algebras associated to C*-correspondences, J. Funct. Analysis
217, 366–401.
[410] , [2004], A class of C*-algebras generalizing both graph algebras and homeomorphism
C*-algebras I, fundamental results, Trans. Amer. Math. Soc. 356(11), 4287–4322.
[411] , [2006], Non-separable AF algebras, pages 165–173 in Operator algebras, The Abel
Symposium 2004, (Eds.) O. Bratteli, S. Neshveyev, C. Skau, Springer, Berlin, etc., X+279
p. Preprint: arXiv:math/0504423v1.
[412] , [2006], A class of C*-algebras generalizing both graph algebras and homeomorphism
C*-algebras II, examples, Internat. J. Math. 17, 791–833.
[413] , [2006], A class of C*-algebras generalizing both graph algebras and homeomorphism
C*-algebras III, ideal structures, Ergodic Theory and Dynam. Systems 26, 1805–1854.
 BIBLIOGRAPHY 1331

[414] , [2008], A class of C*-algebras generalizing both graph algebras and homeomorphism
C*-algebras IV, pure infiniteness, J. Funct. Analysis 254, 1161–1187.
[415] , [2007], Ideal structure of C*-algebras associated with C*-correspondences, Pacific
J. Math. 230, 107–145.
[416] , [2008], Permutation presentations of modules over finite groups, Journal of Algebra
319, 3653–3665. Preprint 2006: arXiv:0608091v1.
[417] , [2008], A construction of actions on Kirchberg algebras which induce given actions
on their K-groups, J. Reine Angew. Math. 617, 27–66.
Preprint 2007: arXiv:0608093v2.
[418] Keller, O.H., [1931], Die Homoiomorphie der kompakten konvexen Mengen im Hilbertschen
Raum, Mathematische Annalen 105, 748–758.
[419] Kerr, D., [2020], Dimension, comparison, and almost finiteness., J. Eur. Math. Soc. 22,
3697–3745.
[420] Kerr, D., Szabo, G. [2018], Almost finiteness and the small boundary property, Preprint
arXiv: 1807.04326v2 (11.Juni 2019).
[421] Khoshkam, M., [1984], Perturbations of C*-algebras and K-theory, J. Operator Theory, 12,
89–99.
[422] Khoshkam, M., [1984], On the unitary equivalence of close C*-algebras, Michigan
Math. J., 31, 331–338.
[423] Khoshkam, M., [1988], Close C*-algebras are KK-equivalent, In Operator theory: operator
algebras and applications, Part 2 (Durham, NH, 1988), volume 51 of Proc. Sympos. Pure
Math., pages 159–162. Amer. Math. Soc., Providence, RI, 1990.
[424] Khoshkam, M., Skandalis, G. [1997], Toeplitz algebras associated with endomorphisms and
Pimsner-Voiculescu exact sequences, Pacific J. Math. 181, 315–331.
[425] Khoshkam, M., Skandalis, G. [2002], Regular representation of groupoid C*-algebras and
applications to inverse semigroups, J. reine angew. Math. 546, 47–72.
[426] Kirchberg, E. [1977], C*-Nuclearity implies CPAP, Math. Nachrichten 76, 203–212.
[427] , [1978], Positive maps and C*-nuclear algebras, Proceedings of the International
Conference on operator algebras, ideals and their applications in theoretical physics (Leipzig,
1977), pp. 255–257, Teubner Texte, Leipzig, 1978.
[428] , [1983], The Fubini theorem for exact C*-algebras, J. Operator Theory 10, 3–8.
[429] , [1989], On the matricial approximation property, preprint, Universität Heidelberg,
Math. Institut, Heft Nr. 16, April 1991.
[430] , [1991], Exactness properties of von Neumann algebras and the equivalence of Ex-
actness to Property C, preprint, Universität Heidelberg, Math. Institut, Heft Nr. 16, April
1991.
[431] , [1993], On non-semisplit extensions, tensor products and exactness of group C*-
algebras, Invent. Math. 112, 449–489.
[432] , [1994], Commutants of unitaries in UHF-algebras and functorial properties of ex-
actness, J. reine angew. Math. 452, 39–77.
[433] , [1994], Discrete groups with Kazhdan’s property T and factorization property are
residually finite, Math. Ann. 299, no. 3, 551–563.
[434] , [1994], The classification of purely infinite C*-algebras using Kasparov’s theory,
Preprint, 3rd draft, Dec 1994. (It is a survey of some
results and methods described in more detail in [436], [442] and this book all together. But
it can serve as a first reading on the status of results around 1994. In particular the reader
can exercise his understanding by the indicated arguments for the there only outlined ideas
for some proofs.) (We also recommend the more detailed german version of it ...)
[435] , [1994], C*-algebras using Kasparov’s theory, Fields Institut Workshop on: The
Classification of Amenable C*-algebras, Dec. 6–10, 1994.
1332 BIBLIOGRAPHY

[436] , [1995], Exact C*-algebras, tensor products and the classification of purely infi-
nite algebras, Proceedings of the International Congress of Mathematicians, Zürich, 1994,
Birkhäuser Verlag, Basel, Bd.2, 943–954.
[437] , [1995], On restricted perturbations in inverse images and a description of normal-
izer algebras in C*-algebras, J. Funct. Analysis 129, 1–34.
[438] , [1995], On subalgebras of the CAR-algebra, J. Funct. Analysis 129, 35–63.
[439] , [1995], On ” inner ” quasi-diagonal C*-algebras, and Weakly exact vonNeumann
algebras, two Lectures on Fields Institute Workshop on: Quantum Groups and their Con-
nection with Quantized Functional Analysis, June 24–28, 1995.
[440] , [1996], The derivation problem and the similarity problem are equivalent, J. Oper-
ator Theory 36, 59–62.
[441] , [1997], On the existence of traces on exact stably projectionless simple C*-algebras,
Operator algebras and their applications, ed. P. A. Fillmore and J. A. Mingo, Fields
Inst. Commun. 13, 171–172.
[442] , [2000], Das nicht-kommutative Michael-Auswahlprinzip und die Klassifikation
nicht-einfacher Algebren, in C*-Algebras, p. 92–141 in Proceedings of the SFB-Workshop
on C*-algebras, Münster, Germany, March 8 –12, 1999/J. Cuntz, S. Echterhoff (ed.), Berlin
etc., Springer, 272 pp.
[443] , [2003], On permanence properties of strongly purely infinite C*-algebras, SFB478-
preprint:
Heft 284 of Preprint series from SFB 478 Geometrische Structuren in der Mathematik des
Mathematischen Instituts der Westfälischen Wilhelms-Universität Münster, August 2003,
ISSN 1435-1188.
[444] , [2003], The inverse problem for primitive ideal spaces, Preprint (Nov. 7th, 2003).
[445] , [2004], Dini functions on spectral spaces, SFB478-preprint, 321, Universität
Münster, Münster (2004).
[446] , [2004], Dini spaces and Polish equivalence relations, Preprint: Febr 28, 2004, Some
typos corrected: 2016, 2020.
[447] , [2006], The range of generalized Gelfand transforms on C*-algebras, J. Operator
Theory 55, 239–251. (Proof of Lemma 3.3 contains obvious typos: The element ”b” has to
be replaced in the last two lines of the proof – and only there! – by the element ”c”.)
[448] , [2006], Central sequences in C*-algebras and strongly purely infinite algebras, pages
175–232 in Operator algebras, The Abel Symposium 2004, (Eds.) O. Bratteli, S. Neshveyev,
C. Skau, Springer, Berlin, etc., X+279 p.
[449] , [2006], An example of a quasi-trace on a unital separable Typ-1 C*-algebra that is
not a 2-quasi-trace, Preprint: April 9, 2006. // (A unital (not 2-) quasi-trace on a certain
unital extension of C([0, 1]2 ) by c0 (K(`2 )) is given... OK? )
[450] , [2008.July14], Maximal Abelian C*-subalgebras of the CAR-algebra with totally
disconnected spectrum, Preprint: July 14, 2008, Preliminary version.
[451] , [June 2009], An exotic R-action on the Cuntz Algebra O2 , Lecture: Glasgow, June
2009. 11 pages.
[452] , [Oct.2010], Property (C) and and local reflexivity of C*-algebras, Preprint Oct 2010
(36 pages, still some typos).
[453] , [June 2010], Spectra of C*-algebras and line actions on O2 , Lecture on 23rd Inter-
national Conference On Operator Theory (Timisoara, June 29 – July 04, 2010).
[454] , [July 2011], Extensions of C*-algebras and coherent Dini spaces, First Preprint
July 2011 (18 pages, some typos June 2014).
[455] , [2013], Actions of locally quasi-compact X on F (X), Notes (April 2013), not pub-
lished now.
[456] , [2015], Abelian factorization of separable C*-algebras and realizations of Dini
spaces, preprint in preparation 2015. See Lectures in 2013 or 2014?
 BIBLIOGRAPHY 1333

[457] , [2014], Central sequence algebras with and without characters, June 23-27, 2014,
(Talk in: 42nd Canadian Annual Symposium on Operator Algebras and Their Applications
(COSy), Fields Institute, 222 College St.)
[458] , [2015], On the matricial approximation property, Preprint, 10 pages, not published.
[459] , [2016], Remark on the universal coefficient theorem, Manuscript. (Parts of this
lecture notes are used in this book!)
[460] , Phillips, N.C. [2000], Embedding of exact C*-algebras in the Cuntz algebra O2 ,
J. reine angew. Math. 525, 17–53.
[461] , Phillips, N.C. [2000], Embedding of continuous fields of C*-algebras in the Cuntz
algebra O2 , J. reine angew. Math. 525, 55–94.
[462] , Rørdam, M. [2000], Non-simple purely infinite C*-algebras, Amer. J. Math. 122(3),
637–666.
[463] , Rørdam, M., [2002], Infinite non-simple C*-algebras: absorbing the Cuntz algebra
O∞ , Advances in Math. 167(2), 195–264.
[464] , Rørdam, M. [2005], Purely infinite C*-algebras: ideal-preserving zero homotopies,
Geom. Funct. Analysis (GAFA) 15, 377–415.
[465] , Rørdam, M. [2012], Some cases where A = ∼ A ⊗ Z, Preliminary (!) Version: Sept
4, 2012.
[466] , Rørdam, M. [2014], Central sequence C*-algebras and tensorial absorption of the
Jiang-Su algebra, J. Reine Angew. Math. 695, 175–214. MR 3276157.
Erratum to it in: J. Reine Angew. Math. 695, 215–216.
[467] , Rørdam, M., [2015], When central sequences C*-algebras have characters, Interna-
tional J. Math. 26(07), 32 pp. Preprint: arXiv 1409.1395v2.
[468] , A. Sierakowski, [1918], Strong pure infiniteness of crossed products, Ergodic Theory
and Dynamical Systems, 38(1), 220–243. Published online: 28 July 2016.
[469] , A. Sierakowski, [2015], Filling families and strong pure infiniteness, Preprint:
arXiv:1503.08519v2. (and Google Scholar).
[470] , Vaillant, G., [1992], On C*-algebras having linear, polynomial and subexponential
growth, Invent. Math. 103, 635–652.
[471] , Wassermann, S., [1995], Operations on continuous bundles of C*-algebras,
Math. Ann. 303, 677–697.
[472] , Wassermann, S. [1998], C*-algebras generated by operator systems,
J. Funct. Analysis 155(2), 324–351.
[473] , Wassermann, S. [1999], Exact groups and continuous bundles of C*-algebras,
Math. Ann. 315, no.2, 169–203.
[474] , Wassermann, S. [2000], Permanence properties of C*-exact groups, Documenta
Math. 5 (2000) 513–558.
[475] , Wassermann, S. [2010], Property (C) and local reflexivity of C*-algebras, Preprint
(36 pages, not complete and with some typos), Oct. 22, 2010. Some later corrections.
[476] , Winter, W. [2004], Covering Dimension and Quasidiagonality, Internat. J. Math.
15(1), 63–85.
[477] Kishimoto, A. [1980], Simple crossed products of C*-algebras by locally compact abelian
groups, Yokohama Math. J. 28, 69–85.
[478] Kishimoto, A. [1981], Outer automorphisms and reduced crossed products of simple C*-
algebras, Comm. Math. Phys. 81, 429–435.
[479] Kishimoto, A. [1982], Universally weakly inner one-parameter automorphism groups of sim-
ple C*-algebras, Yokohama Math. J. 30, 141–149.
[480] Kishimoto, A. [1996], A Rokhlin property for one-parameter automorphism groups, Comm.
Math. Phys. 179, 599–622.
[481] Kishimoto, A. [1996], The Rohlin property for shifts on UHF algebras and automorphisms
of Cuntz algebras, J. Funct. Anal., 140(1), 100–123.
[482] Kishimoto, A. [1999], Pairs of simple dimension groups, Internat. J. Math. 10, 739–761.
1334 BIBLIOGRAPHY

[483] Kishimoto, A. [2002], Rohlin flows on the Cuntz algebra O2 , Internat. J. Math., 13(10),1065–
1094.
[484] Kishimoto, A. [2004], Central sequence Algebras of a purely infinite simple C*-algebra,
Canad. J. Math. Analysis 56, 1237–1258.
[485] Kishimoto, A. [2005], The one-cocycle property for shifts, Ergodic Theory Dynam. Systems,
25(3), 823–859.
[486] Kishimoto, A., Kumjian, A., [1996], Simple stably projectionless C*-algebras arising as
crossed products, Canad. J. Math., 48(5), 980–996.
[487] Kishimoto, A., Kumjian A., [1997], Crossed products of Cuntz algebras by quasi-free
automorphisms, Operator Algebras and Their Applications (Waterloo, Ontario, Canada,
1994/1995), Fields Inst. Commun. 13, 173–192, Amer. Math. Soc., Providence, RI, 1997.
[488] Kishimoto, A., Ozawa N., Sakai S. [2003], Homogeneity of the pure state space of a separable
C*-algebra, Canad. Math. Bull. 46, 365–372.
[489] Kodaka, K., Strong Morita equivalence for inclusions of C*-algebras induced by twisted
actions of countable discrete group, preprint, arXiv:1910.03774, Math. Scand. to appear.
[490] Kodaka, K., Teruya, T. [2018], The strong Morita equivalence for inclusions of C*-algebras
and conditional expectations for equivalence bimodules, J. Aust. Math. Soc., 105, 103–144.
[491] Koosis, P. [1964], Sur un theorème de Paul Cohen, Comptes Rendus de l’Academie de
Sciences Paris, 259, 1380–1382.
[492] Kraus, J., [1991], The slice map problem and approximation properties, J. Funct. Anal. 102,
116–155.
[493] Krein, M.G., [1949], A principle of duality for a bicompact group and square block algebra,
Dokl. Akad. Nauk. SSSR, 69, 725–728.
[494] Krein, M.G., [1949/50], Hermitian positive kernels on homogenous spaces (in russian),
Ukrain. Math. Journ. 1 (1949), 64–98, 2 (1950), 10–59.
[495] Kucerovsky, D., [1994], Kasparov products in KK-theory and unbounded operators with
applications to index theory, Thesis, Oxford University, 1994.
[496] Kucerovsky, D., [2006], Large Fredholm Triples, J. Funct. Anal. 236(2), 395–408.
[497] , Ng, P.W., [2005], Decomposition rank and absorbing extensions of type I algebras,
J. Funct. Analysis 221, 25–36.
[498] , Ng, P.W., [2006], The corona factorization property and approximate unitary equiv-
alence, Houston Journal of Math. 32, 531–550.
[499] , Ng, P.W., [2006], S-regularity and Corona Factorization Property, Math. Scand.
99, 2213–2215.
[500] , Ng, P.W., [2007], AF-Skeletons and Real Rank Zero Algebras with the Corona
Factorization Property, Canad. Math. Bull. 50(2), 227–233.
[501] , Ng, P.W., [2009], A simple C*-algebra with perforation and the corona factorization
property, J. Operator Theory 61(2), 227–238.
[502] , Ng, P.W., Perera, F., [2010] Purely infinite corona algebras of simple C*-algebras,
Math. Ann. 346, 23–40.
[503] , Perera, F., [2011], Purely infinite corona algebras of simple C*-algebras with real
rank zero, J. Operator Theory 65, 131–144.
[504] , Mathieu, M., [2018], Relative double commutants in coronas of separable C*-
algebras, arXiv:1810.01413v1 [math.OA] 2 Oct 2018.
[505] Kuiper, N.H., [1965], The homotopy type of the unitary group of Hilbert space Topology 3,
19–30.
[506] Kwasniewski, B.K., Szymanski, W., [2017], Pure infiniteness and ideal structure of C*-
algebras associated to Fell bundles, J. Math. Anal. Appl. 445, 898–943.
[507] Laca, M., Spielberg, J.S., [1996], Purely infinite C*-algebras from boundary actions of dis-
crete group, J. Reine Angew. Math. 480, 125–139.
[508] Lamperti, J., [1958], On the isometries of certain function spaces, Pacific J. Math. 8, 459–
466.
 BIBLIOGRAPHY 1335

[509] Lance, Ch., [1973], On Nuclear C*-Algebras, J. Funct. Analysis 12, 157–176.
[510] , [1976], Tensor products of non-unital C*-algebras, London Math. Soc. 12, 160–168.
[511] , [1982], Tensor products and nuclear C*-algebras, Proc. Sym. Pure Math. 38 (I):
400–421.
[512] , [1995], Hilbert C*-modules, LMS Lecture notes series 210, Cambridge University
Press.
[513] Li, X., [2012], Semigroup C*-algebras and amenability of semigroups, J. Funct. Anal. 262,
4302–4340.
[514] Li, X., [2013], Nuclearity of semigroup C*-algebras and the connection to amenability,
Adv. Math. 244, 626–662.
[515] Liang, Jian, Rezvani, Sepiden, [2015], Operator-valued Kirchberg theory, arXiv:1410.5379v1
(20 Oct 2014).
[516] Lin, Huaxin, [1989], The simplicity of the quotient algebra M (A)/A of a simple C*-algebra,
Math. Scand. 65, 119–128.
[517] , [1991], Simple C*-algebras with continuous scales and simple corona algebras,
Proc. AMS, 112, 871–880.
[518] , [1991], Generalized Weyl – von Neumann theorems, Int. J. Math. 2, 725–739.
[519] , [1992], Skeleton C*-algebras, Canad. J. Math. 44, 324–341.
[520] , [1993], Extensions by C*-algebras with real rank zero, Int. J. Math. 4, 231–252.
[521] , [1993], Exponential rank of C*-algebras with real rank zero and the Brown–Pedersen
conjectures, J. Funct. Analysis 114, 1–11.
[522] , [1998], Almost commuting unitaries and classification of purely infinite simple C*-
algebras, J. Funct. Analysis 155, 1–24.
[523] , [2000], Tracially AF C*-algebras, Trans. AMS, 353(2), 693–722.
[524] , [2001], Tracial topological rank of C*-algebras, Proc. London Math. Soc. 83(3),
199–234.
[525] , [2001], An introduction to the classification of amenable C*-algebras, World Scien-
tific, New Jersey etc.
[526] , [2004], Simple corona C*-algebras, Proc. AMS 132, 3215-3224.
[527] , [2007], Simple nuclear C*-algebras of tracial topological rank one, J. Funct. Analysis
251(2), 601–679.
[528] , [2009], Embedding Crossed Products into a Unital Simple AF-algebra, Preprint
2009. arXiv:math/0604047v1.
[529] , Niu, Z., [2008], Lifting KK-elements, asymptotic unitary equivalence and classifi-
cation of simple C*-algebras, Adv. Math. 219, 1729–1769.
[530] , Niu, Z., [2011], The range of a class of classifiable separable simple amenable C*-
algebras, J. Funct. Anal. 260(1), 1–29.
[531] , Philips, N.Ch., [1995], Approximate unitary equivalence of homomorphisms from
O∞ , J. Reine Angew. Math. 464, 173–186. (Preprint: arXiv:funct-an/9506008)
[532] , Zhang, S., [1991], On infinite simple C*-algebras, J. Funct. Analysis 100, 221–231.
[533] Lin Q., Phillips N.Ch., [2002 ??], Direct limit decomposition for C*-algebras of minimal
diffeomorphisms, Advanced Studies in Pure Mathematics 38, ?? 2004 ??????
(Preprint: arXiv:math/0208086v1 exists check arXiv and source again !!! (Preprint:
arXiv:funct-an/9506008 does not exist ???)
[534] Longo, R., [1987], Simple injective subfactors, Adv. Math. 63, 152–171.
[535] Longo, R., [1994], A duality for Hopf algebras and for subfactors. I., Comm. Math. Phys.,
159, 133–150.
[536] Longo, R., Roberts, J.E., [1997], A theory of dimension, K-Theory 11(2), 103–159.
[537] Loring, T.A., [1993], Projective C*-algebras, Math. Scand. 73, 274–280.
[538] , [1993], C*-algebras generated by stable relations, J. Funct. Analysis 112, 159–201.
[539] , [1997], Lifting Solutions to Perturbing Problems in C*-algebras, Field Institute
monographs, vol.8, Amer. Math. Soc., Providence, Rhode Island.
1336 BIBLIOGRAPHY

[540] , Pedersen, G.K. [1998], Projectivity, transitivity and AF-telescopes, Trans.

Amer. Math. Soc. 350, 4313–4339.
[541] Loring, T.A., Vides, F., [2013] Estimating norms of commutators, arXiv:1301.4252v3, see
also https://repository.unm.edu/handle/1928/23462 for a more detailed version.
[542] Lundell, A.T., Weingram, S. [1969], The topology of CW complexes, Van Nostrand Reinhold
(New York, 1969).
[543] Ma, X., [2021], Purely infinite locally compact Hausdorff étale groupoids and their C*-
algebras, To appear in Int. Math. Res. Not.(IMRN), doi:10.1093/imrn/rnaa360.
[544] Ma, X., [2021], A categorical study on the generalized type semigroup, arXiv:2107.10816v1
(22 Jul 2021)
[545] Magaina, B., [2021], Cones of completely bounded maps, Positivity 25(1),1–29.
[546] Manuilov, V., Thomsen, K., [2004], E-theory is a special case of KK-theory, Proc. London
Math. Soc. (3) 88, no.2, 455–478.
[547] Matui, H., [2008], Classification of outer actions of ZN on O2 , Adv. Math. 217, 2872–2896.
(arXiv:0708.4074v2)
[548] Matui, H., Sato, Y., [2012], Z-stability of crossed products by strongly outer actions, Comm.
Math. Phys. 314,193–228. (arXiv: 0912.4804v2)
[549] , , [2012], Strict comparison and Z-absorption of nuclear C ∗ -algebras, Acta
Mathematica 209,173–196. (arXiv:1111.1637v1)
[550] McDuff, D. [1970]. Central sequences and the hyperfinite factor, Proc. London Math. Soc. 21,
443–461.
[551] Meyer, R., Nest, R. [2012], C*-algebras over topological spaces: filtrated K-theory,
Canad. J. Math. 64, 368–408.
[552] Michael, E. [1951], Topologies on spaces of subsets, Trans. Amer. Math. Soc., 71, 152–182.
[553] Michael, E. [1956], Continuous selection I, Ann. of Math., 63, 361–382.
[554] Michael, E. [1956], Continuous selection II, Ann. of Math. 64, 562–580.
[555] Michael, E. [1957], Continuous selection III, Ann. of Math. 65, 375–390.
[556] Milnor, J.W., [1952], On spaces having the homotopy type of a CW-complex, Trans.
Amer. Math. Soc. 90, 272–280.
[557] Mingo, J., [1987], K-theory and multipliers of stable C*-algebras, Trans. Amer. Math. Soc.
299, 397–411.
[558] Mori, M., Ozawa N., [2020], Mankiewicz’s theorem and the Mazur–Ulam property for C*-
algebras, Studia Math., 250, 250–281.
[559] Mortensen, J., [1999], Classification of certain non-simple C*-algebras, J. Operator Theory
41, 229–259.
[560] Murphy, G.J., [1990], C*-Algebras and Operator Theory, Academic Press, San Diego –
London, 1990.
[561] Murphy, G.J., [1991], Ordered groups and crossed products of C*-algebras, Pacific J. Math.
148, 319–349.
[562] Murphy, G.J., [1996], C*-algebras generated by commuting isometries, Rocky Mountain J.
Math. 26, 237–267.
[563] Murphy, G.J., [2001?], Primitivity Conditions for Full Group C*-algebras, (Preprint, later
to - or equal to - 2001)
[564] Musat, M., [2005], On the operator space UMD property for noncommutative Lp -spaces,
arXiv:math/0501033v1.
[565] Musat, M., Rørdam, M., [2019], Factorizable maps and traces on the universal free product
of matrix algebras, Published Dec 2019 in International Mathematics Research Notices,
Preprint: arXiv:1903.10182v4 (Oct 2019).
[566] Musat, M., Rørdam, M., [2020], Non-closure of quantum correlation matrices and factor-
izable channels that require infinite dimensional ancilla (with an appendix by Narutaka
Ozawa), Comm. Math. Phys., 375(3), 1761–1776. (arXiv:1806.1042)
 BIBLIOGRAPHY 1337

[567] Nagisa, M., [2004], Single generation and rank of C*-algebras, in: Operator algebras and
applications, Proceedings of the US-Japan seminar held at Kyushu University, Fukuoka,
Japan, June 7–11, 1999. Math. Soc. Japan, Tokyo, 2004, Adv. Stud. Pure Math. 38, 135–
143.
[568] Nawata, N., [2013], Picard groups of certain stably projectionless C*-algebras,
J. Lond. Math. Soc. 88, no. 2, 161–180.
[569] Nawata, N., [2019], Trace scaling automorphisms of the stabilized Razak-Jacelon algebra,
Proc. Lond. Math. Soc. 118, no. 3, 545–576.
[570] Nawata, N., [2019], Rohlin actions of finite groups on the Razak-Jacelon algebra, to appear
in Int. Math. Res. Not. IMRN, preprint in arXiv:1905.09469.
[571] Ng, Ch.-K., Wong, N.Ch., [2021], A classification of C*-algebras through open projections,
Preprint.
[572] Niu, Zh., [2019], Comparison radius and mean topological dimension: Zd-actions,
arXiv:1906.09171.
[573] Niu, Zh. [2021], Comparison radius and mean topological dimension: Rokhlin property,
comparison of open sets, and subhomogeneous C*-algebras, To appear in J. Anal. Math.
(2021?)
[574] Niu, Zh., Wang, Q., [2021], A tracially AF algebra which is not Z-absorbing, (With an
appendix by Caleb Eckhardt), Münster J. of Math. 14, 41–57.
[575] Nou, D., [2006], Asymptotic matricial models and QWEP property for q-Araki–Woods al-
gebras, Journal of Functional Analysis 232, 295–327.
[576] Oikhberg, T., [2006], The non-commutative Gurarii spaces, Arch. Math. 86, 356–364.
[577] Olesen, D., [1975], Inner automorphisms of simple C*-algebras, Commun. Math. Phys. 44,
175–190.
[578] Olesen, D., Pedersen, G.K. [1981], Applications of the Connes Spectrum to C*-dynamical
Systems, III, J. Funct. Analysis 45, 357–390.
[579] Olsen, C.L., Pedersen, G.K. [1989], Corona C*-algebras and their applications to lifting
problems, Math. Scand. 64, 63–86.
[580] Olsen, C.L., Zame W.R. [1976], Some C*-algebras with a single generator, Trans. AMS 215,
205–217.
[581] Ortega, E., Perera, F., Rørdam, M., [2011], The Corona Factorization Property and Refine-
ment Monoids, Trans. AMS 363(9), 4505–4525.
[582] , , , [2012], The Corona Factorization Property, stability, and the
Cuntz semigroup of a C*-algebra, Int. Math. Res. Notices (IMRN) 1, 34–66.
[583] Ortega, E., Rørdam, M., Thiel, H., [2011], The Cuntz semigroup and comparison of open
projections, J. Funct. Analysis 260(12), 3474–3493.
[584] Osaka, H., Phillips, N.C., [2012], Crossed products by finite group actions with the Rokhlin
property, Math. Z. 270, 19–42.
[585] Osaka, H., Teruya, T., [2006], Topological stable rank of inclusions of unital C*-algebras,
Internat. J. Math. 17, 19–34.
[586] , , [2007], Stable rank of depth two inclusions of C*-algebras,
C.R. Math. Acad. Sci. Soc. R. Can. 29(1), 28–32.
[587] , , [2010], Strongly self-absorbing property for inclusions of C*-algebras with
a finite Watatani index, arxiv:math.OA/1002.4233.
[588] , , [2011], Nuclear dimension for inclusions of unital C*-algebras,
arXiv:1111.1808v1 (8 Nov 2011).
[589] , , [2018], Permanence of nuclear dimension for inclusions of unital C*-
algebras with the Rokhlin property, Adv. Oper. Theory 3(1), 123–136.
[590] , , [2018], The Jiang-Su Absorption for Inclusions of Unital C*-algebras,
Canadian Journal of Mathematics (???).
[591] , , [2020], The Rokhlin property for inclusions of C*-algebras, Rocky Moun-
tain Journal of Mathematics 50(5), 1785–1792.
1338 BIBLIOGRAPHY

[592] Ozawa, N., [2000], Amenable actions and exactness for discrete groups, C.R. Acad. Sci. Paris
Sr. I Math. 330(8), 691–695.
[593] Ozawa, N., [2001], A non-extendable bounded linear map between C*-algebras, Proc. Ed-
inb. Math. Soc. (2) 44, 241–248.
[594] Ozawa, N., [2002], An application of expanders to B(`2 ) ⊗ B(`2 ), arXiv:math/0110151v3
(math.OA) 27 Oct 2002.
[595] Ozawa, N., [2004], About the QWEP conjecture, International Journal of Mathemat-
ics, 15(5), 501–530.
[596] Ozawa, N., [2004], There is no separable universal II1 factor, Proc. Amer. Math. Soc. 132,
487–490.
[597] , [2007], Weakly exact von Neumann algebras, Journal of the Mathematical Society
of Japan, 59(4), 985–991. (Preprint: arXiv:0411473v1).
[598] , [2008], Weak amenability of hyperbolic groups, Groups Geom. Dyn. 2, 271–280.
[599] , [2012], Examples of groups which are not weakly amenable, Kyoto J. Math., 52,
333–344.
[600] , [2013], About the Connes embedding conjecture: algebraic approaches, Japanese
Journal of Mathematics, 8(1), 147–183.
[601] , [2013], Tsirelson’s problem and asymptotically commuting unitary matrices,
J. Math. Phys., 54, 032202 (8 pages).
[602] , Pisier, G. [2014], A continuum of C*-norms on B(H) ⊗ B(H) and related tensor
products, Glasg. Math. J. 58 (2016), 433–443. (Preprint: arXiv:1404.7088v1)
[603] , Suzuki, Y., [2021], On characterizations of amenable C*-dynamical systems and
new examples, Selecta Math. (N.S.), 27, Article number: 92. 29pp.
[604] Pasnicu, C., [1986], ? Classification and structure for non-simple C*-algebras, ?????
[605] Pasnicu, C., [1986], On certain inductive limit C*-algebras, Indiana University Mathematical
Journal, 35(2), 269–288.
[606] Pasnicu, C., [1992], Homomorphisms of Bunce-Deddens Algebras, Pacific Journal of Math-
ematics, 155(1), 157–167.
[607] Pasnicu, C., [2021], Ideals in tensor products, topological dimension zero and the weak ideal
property, J. Math. Anal. Appl. 501, 125–210.
[608] Pasnicu, C., [2021], On the weak ideal property and some related properties,
Bull. Math. Soc. Sci. Math. Roumanie, Tome 64 (112), No. 2, 147–157.
[609] Pasnicu C., Phillips, N.Ch., [2014], Permanence properties for crossed products and fixed
point algebras of finite groups, Trans. AMS 366(9), 4625–4648.
[610] Pasnicu, C., Rørdam, M., [2007], Purely infinite C*-algebras of real rank zero, J. reine
angew. Math. (Crelle’s Journal) 613, 51–74.
[611] Paulsen, V. I. [1982], Completely bounded maps on C*-algebras and invariant operator
ranges, Proc. Am. Math. Soc. 86, 91–96.
[612] Paulsen, V. I. [1986], Completely bounded maps and dilations, Pitman Research Notes in
Math., Vol. 146, Sci. Tech., Harlow.
[613] Paulsen, V. I. [2011], Weak expectations and the injective envelope, Trans. Amer. Math. Soc.
363(9), 4735–4755.
[614] Pedersen, G. K. [1968], Measure theory for C*-algebras II, Math. Scand. 22, 63–74.
[615] Pedersen, G. K. [1972], Some operator monotone functions, Proc. Amer. Math. Soc. 36,
309–310.
[616] Pedersen, G. K. [1979], C*-algebras and their automorphism groups, Academic Press, Lon-
don.
[617] , [1980], The linear span of projections in simple C*-algebras, J. Operator Theory
4, 289–296.
[618] , [1990], The corona construction, in Operator Theory: Proceedings of the 1988
GPOTS-Wabash Conference (Indianapolis, IN, 1988), vol. 225 of Pitman Res. Notes
Math. Ser., Longman Sci. Tech., Harlow, 1990, 49–92.
 BIBLIOGRAPHY 1339

[619] , [1993], A commutator inequality, in Operator algebras, mathematical physics, and

low-dimensional topology (Istanbul, 1991), vol. 5 of Res. Notes Math., A K Peters, Wellesley,
MA, 1993, 233–235.
[620] Pedersen, G. K. [1997], A strict version of the non-commutative Urysohn lemma, Proc. AMS
125, 2657–2660.
[621] Pedersen, G. K. [1998], Factorization in C*-algebras, Exposition. Math. 16, 145–156, Spec-
trum Akadem. Verlag, Heidelberg 1998.
[622] Pedersen, G. K. [1999], Pullback and pushout constructions in C*-algebra theory, Journal
of Functional Analysis, 167(2), 243–344.
[623] , Takesaki, M., [1972], The Operator Equation T HT = K, Proc. AMS, 36, 311–312.
[624] Perera, F., Rørdam, M., [2004], AF-embeddings into C*-algebras of real rank zero,
J. Funct. Anal. 217, 142–170. Preprint(2003): arXiv:math/0310340v1.
[625] Phillips, N. Ch., [1988], Inverse limits of C*-algebras and applications, in Operator algebras
and applications, vol. 1 , London Mathematical Society Lecture Note Series 135, 127–185.
Cambridge University Press.
[626] Phillips, N. Ch., [1988], Inverse limits of C*-algebras, J. Operator Theory 19, 159–195.
[627] Phillips, N. Ch., [2000], A classification theorem for nuclear purely infinite simple C*-
algebras, Documenta Math. 5, 49–114.
[628] [2002], Real rank and exponential length of tensor products with O∞ , J. Operator
Theory 47, 117–130.
[629] [2006], The tracial Rokhlin property for actions of finite groups on C*-algebras,
Preprint, arXiv:math.OA/0609782.
[630] [2009], Freeness of actions of finite groups on C*-algebras, Operator structures and
dynamical systems, 503, 217–257.
[631] , Viola, M.G., [2013], A simple separable exact C*-algebra not anti-isomorphic to
itself, Math Ann. 355(2), 783–799.
[632] Phillips, N.C., Weaver, N., [2007], The Calkin algebra has outer automorphisms, Duke
Math. J. 139(1), 185–202.
[633] Pimsner, M.V., [1997], A class of C*-algebras generalizing both Cuntz-Krieger algebras and
crossed products by Z, Fields Inst. Commun. 12, 389–457.
[634] , Popa, S., [1986], Entropy and index for subfactors, Ann. Scient. Ec. Norm. Sup. 4,
57–106.
[635] , Popa, S. , Voiculescu, D. [1979], Homogeneous C*-extensions of C(X) ⊗ K(H), I,
J. Operator Theory 1, 55–108.
[636] , Voiculescu, D. [1980], Exact sequences for K-groups of certain cross-product C*-
algebras, J. Operator Theory 4, 93–118.
[637] Pisier, G. [1992], Factorization of operator valued analytic functions, Adv. Math. 93, 61–
125.
[638] Pisier, G. [1995], Exact operator spaces, Colloque sur les algèbres d’opérateurs, Astérisque
232, Soc. Math. France (1995), 159–187. (1993 preprint arXiv:9308204).
[639] Pisier, G. [1996], Espaces d’opérateurs : une nouvelle dualité
[640] Pisier, G. [1996], A simple proof of a theorem of Kirchberg and related results on C*-norms,
J. Operator Theory 35(2), 317–335.
[641] Pisier, G. [2000], Remarks on the similarity degree of an operator algebra,
arXiv:math/0009052v2 (math.FA) 5 Oct 2000.
[642] Pisier, G. [2001], Similarity problems and completely bounded maps,
Lecture Notes in Mathematics (1618), Springer-Verlag, Berlin.
[643] Pisier, G., Xu, Q., [2003], Non-commutative Lp - spaces, In: Handbook of the Geometry of
Banach Spaces, vol. 2, Amsterdam: Elsevier, 2003, pp.1459-1517.
[644] Pisier, G., [2019], On a Characterization of the Weak Expectation Property (WEP),
arXiv:1907.06224v2 (23 Jul 2019).
1340 BIBLIOGRAPHY

[645] Pisier, G. [2020], Tensor Products of C*-algebras and Operator spaces. The Connes–
Kirchberg problem, Kindle book 2020.
[646] Pop, C.,Smith, R.R., [2004], Crossed Products and Entropy of Automorphisms, Journal of
Functional Analysis 206(1), 210–232.
(or: arXiv:math/0209016v2)
[647] Popa, S., [1981], On the Russo-Dye theorem, Michigan Math. J. 28, 311–315.
[648] Popa, S., [1986], A short proof of ”injectivity implies hyperfiniteness” for finite von Neu-
mann algebras, J. Operator Theory 16, 261–272.
[649] Popa, S., [1997], On local finite-dimensional approximation of C ? -algebras, Pacific
J. Math. 181, 141–158.
[650] Popa, S., [2015?], Correspondences, Preprint. 2015?
[651] Powers, R.T., [1975], Simplicity of the C*-algebra associated with the free group on two
generators, Duke Math. J. 42, 151–156.
[652] Putnam, I.F., [1990], The invertible elements are dense in the irrational rotation C*-
algebras, J. Reine Angew. Math. 410, 160–166.
[653] Putnam, I.F., Spielberg, J.S., [1999], The structure of C*-algebras associated with hyperbolic
dynamical systems, J. Funct. Analysis 163, 279–299.
[654] Radulescu, F. [2002], A comparison between the max and min norms on C ∗ (Fn ) ⊗ C ∗ (Fn ),
arXiv:math/0202061v1. [math.OA] / 7 Feb 2002.
[655] Raeburn, I., [2005], Graph Algebras, CBSM Reg. Conf. Ser. Math., 103, vi +113 pp.
[656] Razak, Sh., [2002], On the classification of simple stably projectionless C*-algebras,
Canad. J. Math., 54(1),138–224.
[657] Renault, J., [1980], A groupoid approach to C*-algebras, Lecture Notes in Math., vol. 793,
Springer, Berlin, 1980.
[658] Renault, J. [2000], Cuntz-like algebras, p. 371–386 in: Operator theoretical methods, Theta
Found., Bucharest, 2000.
[659] Renault, J. [2008], Cartan subalgebras in C*-algebras, Irish Math. Soc. Bull. 61, 29–63.
[660] Reid, G.A., [1965], A generalisation of W*-algebras, Pacific J. Math. 15(3), 1019–1026.
[661] Rickart, C.E., [1960], General theory of Banach algebras, D. Van Nostrand, New York.
[662] Rieffel, M.A., [1974], Induced representations of C*-algebras, Advances in Math. 13, 176–
257.
[663] Rieffel, M.A., [1983], Dimension and stable rank in the K-theory of C*-algebras, Proc. Lon-
don Math. Soc. 46, 301–333.
[664] Rieffel, M.A., [1987], The homotopy groups of the unitary groups of noncommutative tori,
Journal of Operator Theory 17, 237–254.
[665] Ro, M.Y., [2014], Approximate Diagonalization of homomorphisms from C(X) to C*-
algebras of tracial rank one, (2014), ArXiv:1412.3682v1.
[666] Robert, L., [2007], The Cuntz semigroup of some spaces of dimension at most 2,
(Preprint 4. Sept 2008: arXiv:0711.4396v3)
[667] Robert, L., [2012], Classification of inductive limits of 1-dimensional NCCW complexes
Advances in Mathematics 231, 2802–2836.
[668] Robert, L., [2013], Remarks on Z-stable projectionless C*-algebras, 22 December, 2013;
arXiv:1112.6069v3.
[669] Robert, L., [2015], Nuclear dimension and sums of commutators, Indiana Univ. Math. J.
64 559–576.
[670] Robert, L., Rørdam, M., [2013], Divisibility properties for C*-algebras, Proc. London
Math. Soc. 106(6), 1330–1370.
[671] Robert, L., Tikusis, A., [2017], Nuclear dimension and Z-stability of non-simple C*-
algebras, Trans. AMS 369:7, 4631–4670.
[672] Roe, J., Willet, R., [2014], Ghostbusting and property A, J. Funct. Anal. 266(3), 1674–1684.
[673] Rørdam, M., [1988], Advances in the theory of unitary rank and regular approximation,
Ann. of Math. 128, 153–172.
 BIBLIOGRAPHY 1341

[674] , [1991], Ideals in the multiplier algebra of a stable C*-algebra,

J. Operator Theory 25(2), 283–298.
[675] , [1991], On the structure of simple C*-algebras tensored with a UHF-algebra,
J. Funct. Analysis 100, 1–17.
[676] , [1992], On the structure of simple C*-algebras tensored with a UHF-algebra, II,
J. Funct. Analysis 107, 255–269.
[677] , [1993], Classification of inductive limits of Cuntz algebras,
J. Reine Angew. Math. 440, 175–200.
[678] , [1994], A short proof of Elliott’s theorem: O2 =∼ O2 ⊗ O 2 ,
C.R. Math. Rep. Acad. Sci. Canada 16, 31–36.
[679] , [1995], Classification of certain infinite simple C*-algebras, J. Funct. Analysis 131,
415–458.
[680] , [1995], Classification of Cuntz-Krieger algebras, K-theory 9, 31–58.
[681] , [1997], Classification of certain infinite simple C*-algebras III, Operator Algebras
and Their Applications, pages 257–282, Fields Inst. Commun. 13, Amer. Math. Soc., Prov-
idence, RI.
[682] , [1997], Stability of C*-algebras is not a stable property, Doc. Math. J. DMV, 2,
375–386.
[683] , [1997], On sums of finite projections, Operator algebras and operator theory
(Shanghai, 1997), Amer. Math. Soc., Providence, RI, (1998), 327–340.
[684] , [1997], Classification of extensions of certain C*-algebras by their six term exact
sequences in K-theory, Math. Ann., 308, 93–117.
[685] , [2001], Extensions of stable C*-algebras, Documenta Math. 6, 241–246.
[686] , [2002], Classification of nuclear, simple C*-algebras, p. 1–145 in: Operator Alge-
bras and Non-Commutative Geometry VII, Encyclopedia of Mathematical Sciences, Vol.126,
J. Cuntz, V.F.R. Jones (Eds.) Springer, Berlin etc.
[687] , [2003], A simple C*-algebra with a finite and an infinite projection, Acta Math.
191, 109–142.
[688] , [2004], Stable C*-algebras, Advanced Studies in Pure Mathematics 38, “Operator
Algebras and Applications”. Edited by Hideki Kosaki, 177–199.
[689] , [2004], A purely infinite AH-algebra and an application to AF-embedability, Israel
Jour. of Math. 141, 61–82.
[690] , [2004], The stable and the real rank of Z-absorbing C*-algebras, International
J. Math. 15(10), 1065–1084. (Preprint: arXiv:math/0408020v2)
[691] , [2005], The real rank of certain simple C*-algebras, Advances in Operator Algebras
and Mathematical Physics, Theta, Bucharest (2005), 197–206.
[692] , Larson, F., Laustsen. N.J. [2000], An Introduction to K-theory for C*-algebras,
L.M.S. Student Texts 49, Cambridge University Press.
[693] , Winter, W., [2010], The Jiang-Su algebra revisited, J. Reine Angew. Math. 642
(2010), 129–155. (Preprint: arXiv:math/0801.2259v1)
[694] Rørdam, M., Sierakowski, A., [2012], Purely infinite C*-algebras arising from crossed pro-
ducts, Ergod. Th. & Dynam. Sys., 32(1), 273–293.
[695] Rosenberg, J., [1977], Amenability of crossed products of C*-algebras, Com-
mun. math. Phys., 57, 187–191.
[696] , [1997], The algebraic K-theory of operator algebras, K-Theory 12:1, 75 – 99.
[697] , Schochet, C. [1981], The classification of extensions of C*-algebras, Bull. Amer.
Math. Soc., 4, 105–110.
[698] , , [1986], The Künneth theorem and the universal coefficient theorem for
equivariant K-theory and KK-theory, Memoirs of the AMS, 348, AMS Providence, 1986.
[699] , , [1987], The Künneth theorem and the universal coefficient theorem for
Kasparov’s generalized K-functor, Duke Math. J. 55, 431–474.
1342 BIBLIOGRAPHY

[700] Efren Ruiz, [2007], A classification theorem for direct limits of extensions of circle algebras
by purely infinite C*-algebras, J. Operator Theory 58(2), 311–349.
[701] Ruiz, E., Sims A., Sørensen, A.P.W., [2015], UCT Kirchberg algebras have nuclear dimension
one, Advances in Mathematics, 279, 1–28.
[702] Ruiz, E., Sims, A., Tomforde, M., [2015], The nuclear dimension of graph C*-algebras, Adv.
Math. 272, 96–123.
[703] ”Russo-Dye theorem”, see “Dye” in citation [222].
[704] Sakai, S., [1971], C*-algebras and W*-algebras, Ergebnisse der Mathematik, Bd. 60,
Springer, Berlin Heidelberg New York.
[705] Sarkowicz, P., [2019], Exact C*-algebras and the Kirchberg-Phillips Nuclear Embedding The-
orem, A thesis presented to the University of Waterloo, Ontario, Canada.
[706] Sato, Y., [2008], The Rohlin property for automorphisms of the Jiang-Su algebra,
arXiv:0908.0135.
[707] , [2011], Discrete amenable group actions on von Neumann algebras and invariant
nuclear C*-algebras, arXiv:1104.4339.
[708] , [2012], Trace spaces of simple nuclear C*-algebras with finite-dimensional extreme
boundary, Preprint 2012, arXiv:1209.3000v1.
[709] , [2019], Actions of amenable groups and crossed products of Z-absorbing C*-
algebras, Advanced Studies in Pure Mathematics, 80, 189–210.
[710] Sato, Y., White, S., Winter, W., [2015], Nuclear dimension and Z-stability, Inventiones
mathematicae, 202, 893–921.
[711] Schafhauser, C.P., [2015], AF embeddings of graph algebras, J. Operator Theory 74, 177–
182.
[712] Schafhauser, C.P., [2015], Cuntz–Pimsner algebras, crossed products, and K-theory,
J. Funct. Anal., 269, 2927–2946.
[713] Schafhauser, Ch., [2018], A new proof of the Tikuisis-White-Winter theorem, J. Reine
Angew. Math., 759, 291–304 (2000) published 8. March 2018.
[714] Schochet, C., [1981], Topological methods for C*-algebras I: spectral sequences, Pacific J.
Math., 96, 193–211.
[715] , [1982], Topological methods for C*-algebras II: geometric resolutions and the
Künneth formula, Pacific J. Math., 98, 443–458.
[716] , [1984], Topological methods for C*-algebras, III: axiomatic homology, Pacific J.
Math., 114, 399–445.
[717] , [1996], The UCT, the Milnor sequence, and a canonical decomposition of the Kas-
parov groups, K-theory 10, 49–72. Correction: K-theory 14 (1998), 197–199.
[718] Schröder, H., [1993], K-theory for real C*-algebras and applications, Pitman research notes
in mathematics, Longman 1993.
[719] Schwartz, J., [1963], Two finite, non-hyperfinite, non-isomorphic factors, Comm. Pure
Appl. Math. 16, 19–26.
[720] Schweizer, J., [2001], Dilations of C*-correspondences and the simplicity of Cuntz-Pimsner
algebras, J. Funct. Analysis 180, 404–425.
[721] Scott, J., [2005], Hopf algebras up to homotopy and the Bockstein spectral sequence, Alge-
braic & Geometric Topology, 5, 119–128.
[722] Shulman, T., [2010], The C*-algebras qA ⊗ K and S 2 (A) ⊗ K are asymptotically equivalent,
J. Operator Theory 63, 85–100.
[723] Sims, A., [2020], Hausdorff étale groupoids and their C*-algebras, In: F. Perera, editor,
Operator algebras and dynamics: groupoids, crossed products, and Rokhlin dimension, Ad-
vanced Courses in Mathematics. CRM Barcelona. Birkhäuser, 2020.
[724] Skandalis, G., [1984], Some remarks on Kasparov theory, J. Funct. Analysis 56, 337–347.
[725] , [1985], Exact sequences for the Kasparov groups of graded algebras, Can. J. Math.
37, 193–216.
[726] , [1988], Une notion de nucléarité en K-théorie (d’après J. Cuntz), K-theory 5, 1–7.
 BIBLIOGRAPHY 1343

[727] Skoufranis, P., [2014], Completely positive maps, available online at ”pskoufra.info.yorku.ca”.
[728] Sparr, G., [1980], A new proof of Löwner’s theorem on monotone matrix functions,
Math. Scand. 47, 266–274.
[729] Spielberg, J.S., [1988], Embedding C*-algebra extensions into AF algebras, J. Funct.
Anal. 81, 325–344.
[730] Spielberg, J.S., [2002], A functorial approach to the C*-algebras of a graph, Internat. J.
Math. 13(3), 245–277.
[731] , [2009], Semiprojectivity for certain purely infinite C*-algebras,
Trans. Amer. Math. Soc. 361(6), 2805–2830. (arXiv:0102229v2, Jun 2002).
[732] , [2005], Weak semiprojectivity for purely infinite C*-algebras, Canad. Math. Bull.,
to appear. 613(2), 211–230. (arXiv: 0206185v2, 13 Apr 2005).
[733] , [2007], Non-cyclotomic presentations of modules and prime-order automorphisms
of Kirchberg algebras, J. Reine Angew. Math.,613(2), 211–230. (arXiv: 0504287v1).
[734] , [2007], Graph-based models for Kirchberg algebras, J. Operator Theory 57(2), 347–
374.
[735] , [2008], Cuntz–Krieger algebras associated with Fuchsian groups, Published online
by Cambridge University Press: 19 September 2008.
[736] , [2012], C*-algebras for categories of paths associated to the Baumslag-Solitar
groups, J. Lond. Math. Soc. (s2) 86(3), 728–754.
[737] , [2014], Groupoids and C*-algebras for categories of paths, Trans. Amer. Math.
Soc. 366(11), 5771–5819.
[738] , [2020], Groupoids and C*-algebras for left cancellative small categories, Indiana
Univ. Math. J., 69(5), 1579–1626.
[739] Stevens, K.H., [1998], The classification of certain non-simple approximate interval algebras,
pages 105–148 in: Operator Algebras and their Applications, II, (Waterloo, ON, 1994/1995),
Fields Inst. Commun. vol.20. Amer. Math. Soc., Providence, RI, 1998.
[740] Stinespring, W.F., [1955], Positive functions on C*-algebras, Proc. AMS 6, 211–216.
[741] Størmer, E., [1963], Positive linear maps of operator algebras, Acta Mathematica 120, 233–
278.
[742] Størmer, E., [1965], On the Jordan structure of C*-algebras, Trans.ÃMS 120(3), 438–447.
[743] Størmer, E., [1969], Symmetric states of infinite tensor products of C*-algebras, J. Func-
tional Analysis 3, 48–68
[744] Størmer, E., [1982], Decomposable positive maps on C*-algebras, Proc. AMS 86, 402–404.
[745] Summers, S.J., Werner, R.F., [1995], On Bell’s inequalities and algebraic invariants, Letters
in Mathematical Physics 33, 321–334.
[746] Sundar, S., [2011], Inverse Semigroups and the Cuntz-Li Algebras, arXiv: 1104.1085v2 (12
Apr 2011).
[747] Suzuki, Y., [2013], Haagerup property for C*-algebras and rigidity of C*-algebras with prop-
erty (T), Journal of Functional Analysis 265(8), 1778–1799.
[748] Suzuki, Y., [2015], Construction of minimal skew products of amenable minimal dynamical
systems, arXiv:1503.01317v1.
[749] Suzuki, Y., [2017], Elementary constructions of non-discrete C*-simple groups, Proc. AMS,
145, 1369–1371.
[750] Suzuki, Y., [2017], Amenable minimal Cantor systems of free groups arising from diagonal
actions, J. Reine Angew. Math. 722, 183–214.
[751] Suzuki, Y., [2017], Group C*-algebras as decreasing intersection of nuclear C*-algebras,
American Journal of Mathematics, 139(3), 681–705.
[752] Suzuki, Y., [2017], Minimal ambient nuclear C*-algebras, Advances in Mathematics 304,
421–433.
[753] Suzuki, Y., [2019], Simple equivariant C*-algebras whose full and reduced crossed products
coincide, J. Noncommut. Geom. 13, 1577–1585.
[754] Suzuki, Y., [2019], Non-amenable tight squeezes by Kirchberg algebras, arXiv:1908.02971v2.
1344 BIBLIOGRAPHY

[755] Suzuki, Y., [2020], Complete descriptions of intermediate operator algebras by intermediate
extensions of dynamical systems, Comm. Math. Phys. 375, 1273–1297.
[756] Suzuki, Y., [2020], On pathological properties of fixed point algebras in Kirchberg algebras,
Proceedings of the Royal Society of Edinburgh, Section A: Mathematics 150(6), 3087–3096.
[757] Suzuki, Y., [2021], The approximation property and exactness of locally compact groups,
Journal of the Mathematical Society of Japan, 73 (1), 263–275.
[758] Suzuki, Y., [2021], Equivariant O2 –absorption theorem for exact groups, Compositio Math-
ematica 157 (7), 1492–1506.
[759] Suzuki, Y., [2021], Rigid sides of approximately finite dimensional simple operator algebras
in non-separable category, International Mathematics Research Notices, Volume 2021, Issue
3, February 2021, Pages 2166–2190.
[760] Szabó, G., [2018], Strongly self-absorbing C*-dynamical systems, Trans. Amer. Math. Soc.
370, 99–130. Corrigendum: Trans. Amer. Math. Soc. 373 (2020), 7527–7531.
[761] , [2018], Strongly self-absorbing C*-dynamical systems, II, arXiv:1602.00266. J.
Noncomm. Geom. 12(1), 369–406.
[762] , [2017], Strongly self-absorbing C*-dynamical systems, III, Advances in Mathematics
316(3), 356–380.
[763] , [2017], On the nuclear dimension of strongly purely infinite C*-algebras, Advances
in Math. 306, 1262–1268.
[764] , [2018] Equivariant Kirchberg-Phillips-type absorption for amenable group actions,
Comm. Math. Phys. 361(3), 1115–1154.
[765] Takesaki, M., [1958], A note on the cross-norm of the direct product of operator algebras,
Kodai Math. Sem. Rep. 10, 137–140.
[766] , [1964], On the cross-norm of the direct product of C*-algebras, Tohoku Math. Jour-
nal 16, 111–122.
[767] , [1979], Theory of operator algebras I, Springer, New York Inc.
Reprinted 2002 as Volume 124 of the “Encyclopaedia of Mathematical Sciences”
[768] Tannaka, T., [1939], Über den Dualitätssatz der nichtkommutativen topologischen Gruppen,
Tôhoku Math. J. 45, 1–12.
[769] Thiel, H., [2017], The Cuntz Semigroup, Lecture notes University of Münster, Winter Se-
mester 2016-17.
[770] Thiel, H., Winter, W., [2014], The generator problem for Z-stable C*-algebras, Trans. AMS
366, 2327–2343. (arXiv:1201.3879v1)
[771] Thom, A., [2010], Examples of hyperlinear groups without factorization property, Groups
Geom. Dyn. 4, 195–208.
[772] Thomsen, K., [2000], On absorbing extensions, Proc. AMS 129:5, 1409–1417.
[773] Thomsen, K., [1991], Nonstable K-Theory for operator algebras, K-Theory 4, 245–267.
[774] Thomsen, K., [2006], A Lemma, Short notes: March 20, 2006.
[775] Tikusis, A., [2014], Nuclear dimension, Z-stability, and algebraic simplicity for stably pro-
jectionless C*-algebras, Math. Ann. 358, 729–778.
[776] Tikusis, A., White, S., Winter, W., [2017], Quasidiagonality of nuclear C*-algebras, Annals
of Mathematics 185, 229–284.
[777] Tomiyama, J., [1967], Applications of Fubini type theorem to the tensor product of C*-
algebras, Tohoku Math. J. 19, 213–226.
[778] Tomiyama, J., [1975], Tensor products and approximation problems of C*-algebras, Publ.
RIMS, Kyoto Univ. 11, 153-183.
[779] Toms, A.S., [2005], On the independence of K-theory and stable rank for simple C*-algebras,
J. Reine Angew. Math. 578, 185–199.
[780] , [2006], Flat dimension growth for C*-algebras, J. Funct. Analysis 238, 678–708.
[781] , [2008], An infinite family of non-isomorphic C*-algebras with identical K-theory,
Trans. Am. Math. Soc. 360(10), 5343–5354.
 BIBLIOGRAPHY 1345

[782] , [2008], Stability in the Cuntz semigroup of a commutative C*-algebra, Proc. Lond.
Math. Soc. 96, 1–25.
[783] , [2008], On the classification problem for nuclear C*-algebras, Ann. of Math. 167(3),
1029–1044.
[784] , [2009], Comparison theory and smooth minimal C*-dynamics, Comm. Math. Phys.
289, 401–433.
[785] Toms, A.S., Winter, W., [2005], Z-stable ASH algebras, Canad. J. Math. 60, 703–720.
(arXiv:math.OA/0508218)
[786] Toms, A.S., Winter, W., [2007], Strongly self-absorbing C*-algebras, Trans. Amer. Math.
Soc. 359(8), 3999–4029.
[787] Toms, A.S., Winter, W., [2013], Minimal dynamics and K-theoretic rigidity: Elliott’s con-
jecture, Geom. Funct. Analysis 23, 467–481.
[788] Toms, A.S., Winter, W., White, S., [2012], Z-stability and finite-dimensional tracial bound-
aries, arXiv:1209.3292
[789] Troitskij, E.V., [1986], Contractibility of the complete linear group of the Hilbert C*-module
`2 (A), Funct. Anal. and its Applications 20, 301–307.
[790] Tsirelson, B.S., [1980], Quantum generalizations of Bell’s inequality, Letters in Mathemat-
ical Physics, 4(2), 93–100.
[791] Tsirelson, B.S., [1993], Some results and problems on quantum Bell-type inequalities,
Hadronic J. Suppl., 4, 329–345.
[792] Tsirelson, B.S., [1994] ?, Bell inequalities and operator algebras, Available as pdf in internet:
https://www.tau.ac.il/ tsirel/download/bellopalg.pdf
[793] Tu, J.-L., [1999], La conjecture de Baum-Connes pour les feuilletages moyennables, K-
Theory 17(3), 215–264.
[794] Vietoris, L., [1921], Stetige Mengen, Monatshefte der Math. Phys. 31, 173–204.
[795] Villadsen, J., [1998], The range of the Elliott invariant of the simple AH-algebras with slow
dimension growth, K-theory 15, 1–12.
[796] , [1998], Simple C*-algebras with perforation, J. Funct. Analysis 154, 110–116.
[797] , [1999], On the stable rank of simple C*-algebras, J. Amer. Math. Soc. 12, 1091–
1102.
[798] Voiculescu, D., [1976], A non-commutative Weyl – von Neumann theorem, Rev. Roumaine
Math. Pures et Appl. 21, 97–113.
[799] Wassermann, S. [1975], Tensor products of *-automorphisms of C*-algebras, Bulletin of the
London Mathematical Society, 7(1), 65–70.
[800] Wassermann, S. [1975], Extension of normal functionals on W*-tensor products, Math.
Proc. Camb. Philos. Soc. 78, 301–307.
[801] Wassermann, S. [1976], Liftings in C*-algebras: a counterexample, Proc. London Math.
Soc. 9, 201–202.
[802] Wassermann, S. [1976], The slice map problem for C*-algebras, Proc. London Math.
Soc. 32(3), 537–559.
[803] Wassermann, S. [1976], On tensor products of certain group C*-algebras, J. Functional
Analysis 23, 239–254.
[804] Wassermann, S. [1977], Injective W*-algebras, Proc. Cambridge Phil. Soc. 82, 39–47.
[805] Wassermann, S. [1978], A pathology in the ideal space of L(H) ⊗ L(H), Indiana Univ. Math.
Journ. 27(6), 1011–1020.
[806] Wassermann, S. [1990], Tensor products of free-group C*-algebras, Bull. London Math. Soc.
22, 375–380.
[807] Wassermann, S., [1991], C*-algebras associated with groups with Kazhdan’s property T, Ann.
of Math. 134(2), 423–431.
[808] Wassermann, S., [1991], A separable quasidiagonal C ? -algebra with a nonquasidiagonal quo-
tient by the compact operators, Math. Proc. Cambridge Philos. Soc. 110, 143–145.
1346 BIBLIOGRAPHY

[809] Wassermann, S., [1994], Subquotients of UHF C*-algebras, Math. Proc. Cambridge
Phil. Soc. 115, 489–500.
[810] Wassermann, S., [1994], Exact C*-algebras and related topics, Lecture Notes Series, No. 19,
Global Analysis Research Center, Seoul National University.
[811] Wassermann, S., [2005], Exact C*-algebras and related topics, Revised and corrected Version,
June 2005.
[812] Watatani, Y., [1990], Index for C*-subalgebras, Mem. Amer. Math. Soc. v. 83, No. 424.
[813] Watson, N., [2014], On the Structure of Nuclear C*-algebras with Real Rank Zero, A the-
sis submitted in conformity with the requirements for the degree of Doctor of Philosophy
Graduate, Department of Mathematics, University of Toronto.
[814] Weaver, N., [1999], Lipschitz algebras, World Scientific Publishing Co., Inc., River Edge,
NJ, 1999, xiv+223 pp.
[815] Weaver, N., [2003], A prime C*-algebra that is not primitive, J. Funct. Analysis 203, 356–
361. (Preprint: arXiv:math.OA/0106252v2, 6 Jul 2001)
[816] Wegge-Olson, N.E. [1993], K-theory and C*-algebras, Oxford University Press (1993), Ox-
ford, New York, Tokyo.
[817] Werner, W., [2002], Subspaces of L(H) that are *-invariant, J. Funct. Anal. 193, 207–223.
[818] Werner, W., [2004], Multipliers on matrix ordered operator spaces and some K-groups,
J. Funct. Anal. 206, 356–378.
[819] Willett, R., [2009], Some notes on property A, In: Limits of graphs in group theory and
computer science, EPFL Press, Lausanne, 2009, pp. 191–281.
[820] Willett, R., [2015], A non-amenable groupoid whose maximal and reduced C*-algebras are
the same, Münster J. Math. 8, 241–252.
[821] Willett, R., Yu, G., [2021], Controlled KK-theory I: a Milnor exact sequence,
arXiv:2011.10906v2 (21 Apr 2021).
[822] Willett, R., Yu, G., [2021], Controlled KK-theory II: a Mayer-Vietoris principle and the
UCT, arXiv:2104.10766v1, 21 Apr 2021.
[823] Williams, D. P., [2007], Crossed products of C*-algebras,
Mathematical surveys and monographs, Vol. 134, Amer. Math. Society.
[824] Winter, W., [2003], Covering dimension for nuclear C*-algebras I,
J. Funct. Anal. 199, 535–556.
[825] , [2004], Decomposition rank of subhomogenous C*-algebras,
Proc. London Math. Soc. 89, 427–456.
[826] , [2005], On topologically finite-dimensional simple C*-algebras, Math. Ann. 332(4),
843–878.
[827] , [2006], On the classification of simple Z-stable C*-algebras with real rank zero and
finite decomposition rank, J. London Math. Soc., 74(1), 167–183.
ArXiv:0502181v2 (2005).
[828] , [2007], Simple C*-algebras with locally finite decomposition rank,
J. Funct. Anal. 243(2), 394–425.
[829] , [2007], Localizing the Elliott conjecture at strongly-self absorbing C*-algebras (with
appendix by H. Lin), arXiv: 0708.0283v3.
[830] , [2009], Covering dimension for nuclear C*-algebras. II.,
Trans. Amer. Math. Soc. 361(8), 4143–4167.
[831] , [2011], Strongly self-absorbing C*-algebras are Z-stable, J. Noncom-
mut. Geom. 5(2), 253–264. (preprint: arXiv:0905.0583v1)
[832] , [2010], Decomposition rank and Z-stability, Invent. Math. 179(2), 229–301.
[833] , [2012], Nuclear dimension and Z-stability of pure C*-algebras,
Invent. Math. 187(2), 259–342.
[834] , [2016], QDQ vs. UCT, arXiv:1603.01493v2 (For Abel Symposium 2015).
 BIBLIOGRAPHY 1347

[835] , [2018], Structure of nuclear C*-algebras: From quasidiagonality to classification

and back again, Proceedings of the International Congress of Mathematicians (ICM 2018),
pp. 1801–1823.
[836] , Zacharias, J., [2009], Completely positive maps of order zero, Münster J. Math., 2,
311–324.
[837] , Zacharias, J., [2010], The nuclear dimension of C*-algebras, Adv. Math. 224(2),
461–498.
[838] Wolff, M., [1994], Disjointness preserving operators on C*-algebras, Arch. Math. 62, 248–
253.
[839] Xia, Jingbo, [2000], Diagonalization and unitary equivalence modulo Schatten p-classes,
Journal of Functional Analysis 175, 279–307.
[840] Zhang, S. [1990], Algebras with real rank zero and internal structure of their corona and
multiplier algebras III, Canad. J. Math. 42, 159–190.
[841] , [1990], A property of purely infinite simple C*-algebras, Proc. Amer. Math. Soc.
109, 717–720.
[842] , [1990], A Riesz decomposition property and ideal structure of multiplier algebras,
J. Operator Theory 24, 209–225.
[843] , [1990], Diagonalizing projections in multiplier algebras and in matrices over a C*-
algebra, Pacific J. Math. 145, 181–200.
[844] , [1991], Matricial structure and homotopy type of simple C*-algebras with real rank
zero, J. Operator Theory 26, 283–312.
[845] , [1991], K1 groups, quasidiagonality, and interpolation by multiplier projections,
Trans. Amer. Math. Soc. 325, 739–818.
[846] , [1992], Certain C*-algebras with real rank zero and their corona and multiplier
algebra, Part I, Pacific J. Math. 155, 169–197.
[847] , [1993], Exponential Rank and Exponential Length of Operators on Hilbert C*-
Modules, Annals of Math. 137, 129 –144.