Introduction to noncommutative

differential geometry
held by Dr. Tobias Mai in Summer ’19
Contents

1. Introduction 5

2. Spectral triples associated to manifolds 9

3. The geodesic distance in noncommutative geometry 23

4. The Riemann-Lebesgue measure in noncommutative geometry 31

Appendix 43

A. Basics of unbounded operators 45

B. Basics on Fourier transform 51

C. Solutions to the exercises 53

3
Chapter 1.

Introduction

Several “classical theories” in mathematics can be extended to the noncommu-

tative world. The appropriate framework is often obtained by the following
recipe:

(i) Take a classical space, i.e., a set X endowed with some additional structure
(e.g. a topological- or measure space, groups, manifolds, Lie groups, . . . );
(ii) Consider a suitable algebra of functions on X (e.g., C0 (X), C(X), L∞ (X),
C ∞ (X), . . . );
(iii) Transfer the additional structure of the space X to its associated (com-
mutative) algebra of functions and provide an intrinsic characterisation of that
structure;
(iv) Drop the assumption of commutativitiy.

Finding a good axiomatic description in (iii) that allows one to perform

step (iv) is clearly the core problem and by no means straight foreward. The
right choice confirms itself by a “reconstruction theorem”, by which, in the
commutative case, the underlying space can be “recovered” from that set of
axioms. We list some prominent examples in Table 1.1.
The actual “noncommutative space” is mostly just a “virtual” object behind
those algebras. The classical theories are thus rebuilt in an algebraic way,
immitating the dual picture on their associated algebras of functions.
This philosophy underlies also the theory of noncommutative differential
geometry that Alain Connes began to develop around the 80’s. His motivation
was to extend classical tools to

• spaces, that are “badly behaved” as point sets, but correspond naturally
to (noncommutative) algebras (e.g., Penrose tilings, the space of leaves of a
foliation, the phase space in quantum mechanics, . . . )

5
Chapter 1. Introduction

• general noncommutative situations without an underlying space.

But even for classical situations that are purely commutative, this point
of view gives new insights. Within noncommutative differential geometry,
manifolds are studied by some spectral data. The following definition is at the
heart of that approach.

Definition 1.1 (Alain Connes, 1994): A spectral triple is a triple (A, H, D)

where

• A is a unital complex ∗ -algebra,

• H is a separable complex Hilbert space with a faithful ∗ -representation
π : A → B(H),
• D is a (possibly unbounded) selfadjoint linear operator on H, say

D : H ⊇ dom D −→ H,

with compact resolvents, i.e., (D − λ1)−1 ∈ K(H) for all λ ∈ C − σ(D),

such that for all elements a ∈ A the following holds: π(a) dom D ⊆ dom D and
the commutator [D, π(a)] := Dπ(a) − π(a)D extends to an operator in B(H).

Example 1.2: Consider on T := {z ∈ C | |z| = 1} the arc length measure m,

i.e., the push-foreward of the Lebesgue measure on R via the map

γ : R −→ T, t 7−→ exp(it).

A function f : T → C is differentiable if and only if f ◦ γ : R → C is so;

its derivative f 0 : T → C is determined by f 0 (γ(t))γ 0 (t) = (f ◦ γ)0 (t) for all
t ∈ R. Take now H = L2 (T, m) and A = C ∞ (T) with the ∗ -representation
π : A → B(H) given by

π(f ) := Mf : L2 (T, m) −→ L2 (T, m), g 7−→ f g.

Further, we consider the densely defined operator

1
D0 : H ⊇ dom D0 −→ H, g 7−→ g 0
i
on dom D0 := C 1 (T), which is a symmetric operator. One can show that D0
is essentially self-adjoint; let D be its closure, which is thus selfadjoint. Then

6
 Chapter 1. Introduction

(A, H, D) is a spectral triple. Indeed, if F : H → `2 (Z) is the Fourier transform,

i.e.,
ˆ
1
2 2
F : L (T, m) −→ ` (Z), f 7−→ (fˆn )n∈Z , fˆn := f (ζ)ζ −n dm(ζ),
2π T

then FDF −1 is the multiplication by (n)n∈Z , hence we see that for all λ ∈ C−Z
it holds (D − λ)−1 ∈ K(H); moreover, for f ∈ A and g ∈ C 1 (T), it holds
1
[D, π(f )]g = D0 (f g) − f D0 g = π(f 0 )g.
i
We will see, that more general manifolds M induce spectral triples in a similar
way. Much of the structure of M can be recovered:

• d(p, q) := sup{|f (p) − f (q)| | f ∈ A : k[D, π(f )]k ≤ 1} is the geodesic

distance between p, q ∈ M,
´
• M f dvol = c(n)Tr(f |D|−n ) for all f ∈ A.

Exercises
Exercise 1.1: Let (A, H, D) be a spectral triple and let V ∈ B(H) be any
selfadjoint operator. Prove that (A, H, DV ) for the unbounded operator DV
given by DV := D + V with domain dom(DV ) := dom(D) is again a spectral
triple.

7
8
classical space noncommutative counterpart reconstruction theory
locally compact (compact) Haus- (unital) C ∗ -algebras, noncommuta- Gelfand-Naimark theorem, FA I,
dorff topological space tive topology Corollary 10.17
compact Hausdorff topological space von Neumann algebras, noncommu- FA II, Theorem 8.15
with finite radon measure tative measure theory
compact topological group compact quantum group, “noncom- Tannaka-Krein theorem
mutative” group theory
compact oriented smooth manifold spectral triple (unbounded) K- Connes’ reconstruction theorem2
Chapter 1. Introduction

cycle1 , noncommutative differential

geometry

Table 1.1.: Examples for outputs of the noncommutiser

There are several important “noncommutative spaces” that generalise classical ones. Although finding the right
axiomatic framework was in none of the cases straightforward, all these noncommutative counterparts have
in common that they mimic algebras of suitable functions over the respective classical spaces. For instance,
the definition of a C ∗ -algebra is modeled on C(X), i.e., the space of all continuous functions f : X → C on a
compact Hausdorff topological space X: It forms a unital complex ∗ -algebra (with operations defined pointwise
and the involution given by f 7→ f ), it becomes a Banach algebra when endowed with the maximum norm
kf k := maxx∈X |f (x)|, and all its elements f satisfy the crucial identity kf f k = kf k2 .

2
Connes, 1994
2
2013
Chapter 2.

Spectral triples associated to

manifolds

Spectral triples are (supposed to be) the right framework to extend classical
differential geometry to the noncommutative world. It is however not clear
offhand, how usual manifolds fit into that frame. In this chapter, we will see
that indeed each compact oriented smooth manifold induces a commutative
spectral triple in a natural fashion.

Definition 2.1 (Manifolds):

(i) An n-dimensional topological manifold is a Hausdorff topological space

M which is locally euclidean, i.e., each point x ∈ M has an open neigh-
bourhood that is homeomorphic to an open subset of Rn .
(ii) A (local) chart (U, ϕ) of M consists of an open subset U ⊆ M and a
homeomorphism ϕ : U → Ω = ϕ(U ) ⊆ Rn .
(iii) A family A = {(Ui , ϕi ) | i ∈ I} of charts satisfying M = i∈I Ui is called
S

an atlas of M. The homeomorphisms ψi,j : ϕi (Ui ∩ Uj ) → ϕj (Ui ∩ Uj )

given by ψi,j := ϕj ◦ ϕ−1
i |ϕi (Ui ∩Uj ) are called transition maps.

(iv) An atlas A of M is called smooth if all its transition maps are smooth,
i.e., C ∞ . A chart (U, ϕ) is said to be smooth with respect to a smooth
atlas A, if A ∪ {(U, ϕ)} is again a smooth atlas. A smooth atlas A is
called maximal, if every chart (U, ϕ) that is smooth with respect to A
already belongs to A. Every smooth atlas A induces a maximal one by

Amax := {(U, ϕ) | (U, ϕ) is a chart smooth with respect to A}.

(v) An n-dimensional smooth manifold is an n-dimensional topological mani-

fold M with a maximal smooth atlas A.

9
Chapter 2. Spectral triples associated to manifolds

Definition 2.2 (Tangent space): Let M be an n-dimensional smooth manifold.

We fix x0 ∈ M.

(i) A function f : V → R on an open subset V ⊆ M is said to be smooth, if

f ◦ ϕ−1 |ϕ(U ∩V ) : ϕ(U ∩ V ) → R is smooth for every smooth chart (U, ϕ).
(ii) On the set of all pairs (V, f ) consisting of an open neighbourhood V of x0
and a smooth function f : V → R, we introduce an equivalence relation
∼ by

(V1 , f1 ) ∼ (V2 , f2 ) :⇐⇒ ∃ V ⊆ V1 ∩ V2 open, x0 ∈ V : f1 |V = f2 |V .

The equivalence class of (V, f ), denoted by [f ]x0 , is called the germ of f

at x0 . We denote by Cx∞0 (M) the R-algebra of germs at x0 .
(iii) The R-vector space Tx0 M of all linear maps δ : Cx∞0 (M) → R satisfying
the product rule

δ([f ]x0 · [g]x0 ) = δ([f ]x0 ) · g(x0 ) + f (x0 ) · δ([g]x0 ) ∀ [f ]x0 , [g]x0 ∈ Cx∞0 (M)

is called the tangent space to M at x0 .

Remark 2.3: In the situation of Definition 2.2, let γ : (−ε, ε) → M be a smooth

path, i.e., γ is continuous and ϕ ◦ γ|γ −1 (U ) : γ −1 (U ) → Rn is smooth for every
smooth chart (U, ϕ), such that γ(0) = x0 . We call γ 0 (0) ∈ Tx0 M given by

γ 0 (0)([f ]x0 ) := (f ◦ γ)0 (0)

for all germs [f ]x0 ∈ Cx∞0 (M), the velocity vector of γ at x0 .

For every δ ∈ Tx0 M, there exists a smooth path γ : (−ε, ε) → M such that
γ(0) = x0 and γ 0 (0) = δ.

Next, we “glue” the tangent spaces, yielding the so-called tangent bundle.

Definition 2.4 (Topological vector bundle): Let X be a Hausdorff topological

space.

(i) An n-dimensional (real / complex) vector bundle over X is given by a

topological space E and a continuous map π : E → X such that the following
conditions are satisfied:
• The fibre Ex := π −1 ({x}) is a real / complex vector space of dimension n
for each x ∈ X.

10
 Chapter 2. Spectral triples associated to manifolds

• For each x0 ∈ X, there is an open neighbourhood U of x0 and a home-

morphism τ : π −1 (U ) → U × Kn1 such that π|π−1 (U ) = prU ◦ τ , where
prU : U × Kn → U is the projection onto the first component, i.e.,
prU (x, v) = x, and τ |Ex : Ex → {x} × Kn ∼ = Kn is a vector space iso-
morphism for all points x ∈ U . We call (U, τ ) a bundle chart (or a local
trivialisation).
(ii) A family A = {(Ui , τi ) | i ∈ I} of bundle charts (or local trivialisations)
S
satisfying X = i∈I Ui is called a bundle atlas. The transition maps
σi,j : (Ui ∩ Uj ) × Kn −→ (Ui ∩ Uj ) × Kn
given by σi,j := τj ◦ τi−1 |(Ui ∩Uj )×Kn satisfying

π −1 (Ui ∩ Uj )
τi τj

σi,j
(Ui ∩ Uj ) × Kn (x,v)7→σi,j (x,v)=(x,Si,j (x)v)
(Ui ∩ Uj ) × Kn

for a continuous map Si,j : Ui ∩ Uj → Gln (K) called the transition maps.
Definition 2.5 (Smooth vector bundle): Let M be a smooth manifold. An
n-dimensional smooth vector bundle over M is an n-dimensional topological
vector bundle over M, for which all transition maps are smooth.
Definition 2.6 (Tangent bundle): Let M be an n-dimensional smooth manifold
with maximal smooth atlas A = {(Ui , ϕi ) | i ∈ I}. We put T M := x∈M Tx M
`

and define π : T M → M by π(δ) = x if δ ∈ Tx M. We define the local

trivialisation by
τi : π −1 (Ui ) −→ Ui × Rn , (x, δ) 7−→ (x, Θ−1
i,x (δ))

with the isomorphism Θi,x : Rn → Tx M given by

n
v j ∂j (f ◦ ϕ−1
X
Θi,x (v)([f ]x ) := i )(ϕi (x))
j=1

for v = (v 1 , . . . , v n ) ∈ Rn , [f ]x ∈ Cx∞ (M). We endow T M with the topology

defined by the following requirement:
W ⊆ T M open :⇐⇒ ∀ i ∈ I : τi (Ui ∩ W ) ⊆ Ui × Rn is open.
Then T M is an n-dimensional smooth vector bundle over M, called the tangent
bundle of M.
1
Here, K denotes either the field of real numbers R or the field of complex numbers C.

11
Chapter 2. Spectral triples associated to manifolds

Definition 2.7 (Smooth sections): Let π : E → M be a smooth vector bundle

over a smooth manifold M and let V ⊆ M be open. A map s : V → E is called
a smooth section if
(i) π ◦ s = idV ,
(ii) For each local trivialisation τ : π −1 (U ) → U × Kn , we have a smooth
composition τ ◦s|U ∩V : U ∩V → (U ∩V )×Kn . In fact, (τ ◦s)(x) = (x, f (x))
for all x ∈ U ∩ V with f : U ∩ V → K being a smooth function.
We write Γ∞ (V, E) for the set of all smooth sections s : V → E, which is a
K-vector space.

Definition 2.8 (Vector fields): Let M be a smooth manifold. A vector field on

M is a smooth section of the tangent bundle T M. We write
X(V ) := Γ∞ (V, T M)
for the vector fields on the open subset V ⊆ M.

Theorem 2.9: Let M be a smooth manifold of dimension n and let V ⊆ M be

open. Then the map
Φ : X(V ) −→ der(C ∞ (V )), (Φ(X)f )(x) := X(x)([f ]x )
is an isomorphism of real vector spaces.

Note that der(C ∞ (V )) denotes the space of derivations on C ∞ (V ), i.e., linear

maps D : C ∞ (V ) → C ∞ (V ) that satisfy the product rule for all f, g ∈ C ∞ (V ):
D(f · g) = D(f ) · g + f · D(g).
Proof: (1) Take X ∈ X(V ) and f ∈ C ∞ (V ). Then the map
h : x −→ R, x 7−→ X(x)([f ]x )
is smooth.
To see this, consider the bundle atlas A = {(Ui , τi ) | i ∈ I} that we introduced
in Definition 2.6. It suffices to show that h|Ui ∩V is smooth for all i ∈ I. By
Definition 2.7 (ii), we find a smooth function g = (g1 , . . . , gn ) : Ui ∩ V → Rn
such that τi (X(x)) = (x, g(x)) for all x ∈ Ui ∩ V . Thus
X(x) = τi−1 (x, g(x)) = (x, Θi,x (g(x)))
and h(x) = X(x)([f ]x ) = Θi,x (g(x))([f ]x ) = nj=1 gj (x)(∂j (f ◦ ϕ−1
P
i )(ϕi (x)) for
all x ∈ Ui ∩ V , which shows that h is smooth on Ui ∩ V .

12
 Chapter 2. Spectral triples associated to manifolds

(2) Take X ∈ X(V ). Because of the preceeding consideration, we have a

well-defined map

Φ(X) : C ∞ (V ) −→ C ∞ (V ), (Φ(X)f )(x) = X(x)([f ]x ).

This Φ(X) is now in fact a deriavation.

To see this, let f, g ∈ C ∞ (V ) be given. Then, for all x ∈ V , we have

(Φ(X)(f · g))(x) = X(x)([f ]x · [g]x ) = X(x)([f ]x )g(x) + f (x)X(x)([g]x )

= (Φ(X)(f )) · g + f · (Φ(X)g))(x),

as desired.
(3) By the above steps, we now have a well-defined map

Φ : X(V ) −→ der(C ∞ (V )), X 7−→ Φ(X),

which is clearly linear.

(4) For the next steps, we need a technical result. Let U be an open
neighbourhood of a point x0 ∈ M. There exists a smooth function ρ : M →
[0, 1] with compact support

supp(ρ) = cl({x ∈ M | ρ(x) 6= 0}) ⊆ U

which is identically 1 in an open neighbourhood of x0 . We call ρ a bump function

for (U, x0 ) (there are such bump functions, take one in Rn and transport it to
M via a chart).
(5) Take D ∈ der(C ∞ (V )) and x0 ∈ V . We define the map

D|x0 : C ∞ (V ) −→ R, f 7−→ (D(f ))(x0 ).

If f1 , f2 ∈ C ∞ (V ) satisfy f1 |U = f2 |U for an open set U ⊆ V with x0 ∈ U , then

D|x0 (f1 ) = D|x0 (f2 ).
To see this, take a bump function ρ for (U, x0 ). Since ρ · (f1 − f2 ) ≡ 0 we get
that

0 = D|x0 (ρ · (f1 − f2 )) = D|x0 (ρ)(f1 − f2 )(x0 ) + ρ(x0 ) · D|x0 (f1 − f2 )

= D|x0 (f1 ) − D|x0 (f2 )

(as f1 and f2 agree on U and ρ(x0 ) = 1) and thus D|x0 (f1 ) = D|x0 (f2 ), as
desired.

13
Chapter 2. Spectral triples associated to manifolds

(6) We thus get a well-defined map

(Ψ(D))(x0 ) : Cx∞0 (V ) −→ R, [f ]x0 7−→ D|x0 (ρ · f |V )

where, for a (U, f ) representing [f ]x0 , ρ is any bump function for (U, x0 ) and
ρf ∈ C ∞ (M) is defined (in fact well-defined) by

0, if x ∈
/ supp(ρ),
(ρf )(x) =
ρ(x)f (x), if x ∈ U.

Clearly, (Ψ(D))(x0 ) ∈ Tx0 M.

(7) If we now take a derivation D ∈ der(C ∞ (V )), then the induced map

Ψ(D) : V −→ T M, x 7−→ (Ψ(D))(x)

belongs to X(V ).
(8) Finally, we get a linear map

Ψ : der(C ∞ (V )) −→ X(V ),

which satisfies Φ ◦ Ψ = idder(C ∞ (V )) and Ψ ◦ Φ = idX(V ) . 

Remark 2.10: Like vector spaces underly linear algebra, vector bundles underly
what can be seen as “parametrised” linear algebra. Indeed, various constructions
for vector spaces can be generalised to that setting.
Let X be a Hausdorff topological space and let E and F be vector bundles
over X (both real or complex) of dimension n and m, respectively.

(i) The Whitney sum (or direct sum) E ⊕ F is the vector bundle of dimension
n + m with (E ⊕ F )x = Ex ⊕ Fx for all x ∈ X.
(ii) The tensor product bundle E ⊗ F is the vector bundle of dimension n · m
with (E ⊗ F )x = Ex ⊗ Fx for all x ∈ X.
(iii) The homomorphisms bundle hom(E, F ) is the vector bundle of dimension
n · m with hom(E, F )x = hom(Ex , Fx ) for all x ∈ X.

Analogously, the dual bundle E ∗ (see assignment 2A, exercise 1 (i)), the exterior
product p E, the vector bundle multp (E) of all p-multilinear maps and the
V

vector bundles symp (E) and altp (E) of all symmetric– respectively alternating
p-multilinear maps can be defined.

14
 Chapter 2. Spectral triples associated to manifolds

Definition 2.11 (Riemannian metric): Let M be a smooth manifold of dimen-

sion n. A Riemannian metric on M is a smooth section g of the vector bundle
sym2 (T M) such that
gx : Tx M × Tx M −→ R
is an inner product on Tx M for all x ∈ X.

Remark 2.12: (i) A topological space X is said to be paracompact if every

open cover (Ui )i∈I of X has an open refinement (Vj )j∈J (i.e., (Vj )j∈J is an open
cover of X and for all j ∈ J there is i ∈ I such that Vj ⊆ Ui ) that is locally
finite (i.e., every x ∈ X has a neighbourhood V such that only finitely many
Vj (j ∈ J) have non-trivial intersection with V ).
(ii) Let X be a topological space. A parition of unity on X is a family (ρi )i∈I
of continuous functions ρi : X → [0, 1] such that
• Each x ∈ X has an open neighbourhood V such that only finitely many
ρi satisfy ρi |V 6≡ 0,
• For all x ∈ X we have i∈I ρi (x) = 1.
P

We say that (ρi )i∈I is subordinate to an open cover (Ui )i∈I of X, if supp(ρi ) ⊆ Ui
for all i ∈ I.
(iii) On a paracompact space X, each open cover (Ui )i∈I of X has a subor-
dinate partition of unity (ρi )i∈I . If X = M is a paracompact smooth manifold,
then each ρi can be chosen to be smooth.

Theorem 2.13: Let M be a smooth manifold of dimension n. Suppose that M

is paracompact. Then there is a Riemannian metric on M.

Proof: Take a bundle atlas A = {(Ui , τi ) | i ∈ I} of T M and let (ρi )i∈I be a

smooth partition of unity subordinate to (Ui )i∈I . We obtain a Riemannian
metric by
ρi (x)hΘ−1 −1
X
gx (δ1 , δ2 ) := i,x (δ1 ), Θi,x (δ2 )i
i∈I

for all x ∈ M and δ1 , δ2 ∈ Tx M, where h·, ·i is the standard inner product on

Rn . 

Definition 2.14 (Differential forms): Let M be a smooth manifold of dimen-

sion n.

(i) The dual bundle T ∗ M to the tangent bundle T M is called the cotangent
bundle.

15
Chapter 2. Spectral triples associated to manifolds

(ii) A smooth differential form of degree p (briefly p-form in the following) is

a smooth section of p T ∗ M. We put Ωp (M) := Γ∞ (M, p M ). We call
V V

Ω• (M) := p≥0 Ωp (M) the exterior algebra.

L

(iii) The exterior derivative is the unique family (dp )p≥0 of R-linear maps
dp : Ωp (M) → Ωp+1 (M) satisfiying
• For all f ∈ Ω0 (M) = C ∞ (M ) and all x ∈ M

(d0 f )(x) : Tx M −→ R, δ 7−→ δ([f ]x );

• For all p ≥ 0 it holds dp+1 ◦ dp = 0,

• For all ω ∈ Ωp (M), η ∈ Ωq (M) it holds

dp+q (ω ∧ η) = dp (ω) ∧ η + (−1)p ω ∧ dq (η)

where ∧ on Ω• (M) is defined pointwise, i.e., (ω ∧ η)(x) = ω(x) ∧ η(x)

in p+q Tx∗ M.
V

(iv) Let T ∗ MC be the complexification of T ∗ M, i.e., T ∗ MC = T ∗ M⊗(M×C),

where M×C is the real trivial bundle (i.e., π : M×C → M, (x, λ) 7→ x) of
dimension 2 (note that C ∼ p Vp
= R2 ). We put ΩC (M) := Γ∞ (M, C T ∗ MC ).

Definition 2.15 (Orientation perserving): Let U, V be open subsets of Rn . A

diffeomorphism f = (f 1 , . . . , f n ) : U → V is said to be orientation preserving
if for all x ∈ U it holds

det([∂j f i (x)]1≤i,j≤n ) > 0.

Definition 2.16 (Orientation of smooth manifolds): Let M be a smooth man-

ifold.

(i) A smooth atlas is called oriented if all its transition maps are orientation
preserving.
(ii) We say M is orientable if it admits an oriented smooth atlas.
(iii) An orientation on M is a maximal oriented smooth atlas.

Theorem 2.17 (Integration of smooth functions): Let M be an oriented n-

dimensional smooth manfold which is paracompact. Let g be a Riemannian
metric on M. Then there exists a unique linear map
ˆ ˆ
∞
: Cc (M) −→ R, f 7−→ f
M M

16
 Chapter 2. Spectral triples associated to manifolds

on the subspace Cc∞ (M) ⊆ C ∞ (M) of compactly supported functions such that
the following condition is satisfied: For every local chart (U, ϕ) in the maximal
oriented smooth atlas A of M and for each f ∈ Cc∞ (M) with supp(f ) ⊆ U we
have that ˆ ˆ
f= (f ◦ ϕ−1 )(det[gk,l ]1≤k,l≤n )1/2 dλn . (2.1)
M ϕ(U )

Here, λ is the Lebesgue measure on Rn and the functions gk,l ∈ C ∞ (ϕ(U )) are
n

for each x ∈ U determined by

gk,l (ϕ(x)) := gx ((dϕ)(x)−1 (∂k |ϕ(x) ), (dϕ)(x)−1 (∂l |ϕ(x) )

Note that {∂k |ϕ(x) | 1 ≤ k ≤ n} is the basis of Tϕ(x) Rn introduced in Exercise 2.1
(ii) and (dϕ)(x) : Tx M → Tϕ(x) Rn is the differential of ϕ at x, which is defined
∞
for δ ∈ Tx M and [f ]ϕ(x) ∈ Cϕ(x) (Rn ) by

((dϕ)(x)δ)([f ]ϕ(x) ) = δ([f ◦ ϕ]x ).

In fact, (dϕ)(x) is bijective since ϕ is bijective.

Remark 2.18: (i) The assumption that M is oriented guarantees that the
right-hand side of Eq. (2.1) is well-defined, i.e., independent of the particular
choice of the chart (U, ϕ).
Using a partition of unity subordinate to the family (Ui )i∈I for a maximal
oriented smooth atlas A = {(Ui , ϕi ) | i ∈ I}, say (ρi )i∈I , one can then define
for general f ∈ Cc∞ (M)
ˆ ˆ
X
f := (ρi f ).
M i∈I M

Since supp(ρi f ) ⊆ Ui for each i ∈ I.

(ii) Theorem 2.17 merges actually two different concepts, namely on the one
hand the integration of compactly supported n-forms, i.e.,
ˆ ˆ
n
: Ωc (M) −→ R, ω 7−→ ω
M M

which requires only that M is oriented and on the other hand the volume
form dvol ∈ Ωn (M) of an oriented Riemannian manifold (M, g); in general,
ω ∈ Ωn (M) is called a volume-form if ω vanishes nowhere, and a paracompact
smooth manifold M is orientable if and only if a volume form exists; in fact,
fixing an equivalence class of volume forms specifies an orientation and vice versa;

17
Chapter 2. Spectral triples associated to manifolds

dvol is chosen such that dvol(x), for each x ∈ M, is normalised with respect to
the inner product on n Tx∗ M induced by g, i.e., hdvol(x), dvol(x)iVn Tx∗ M = 1.
V

One can show that for all f ∈ Cc∞ (M)

ˆ ˆ
f= f dvol,
M M

where f dvol ∈ Ωnc (M).

(iii) Let V be a finite dimensional real vector space and let h·, ·i : V × V → R
be an inner product on V . We thus have an isomorphism Φ : V → V ∗ , x 7→ h·, xi
which allows us to define an inner product h·, ·iV ∗ : V ∗ × V ∗ → R for ϕ, ψ ∈ V ∗
by
hϕ, ψiV ∗ := hΦ−1 (ϕ), Φ−1 (ψ)i.
For every p ∈ N, we extend the latter to an inner product
p p
V∗× V ∗ −→ R
^ ^
h·, ·iVp V ∗ :
Vp
b for ϕ1 ∧ · · · ∧ ϕp , ψ1 ∧ . . . ψp ∈ V ∗ by

hϕ1 ∧ · · · ∧ ϕp , ψ1 ∧ · · · ∧ ψp iVp V ∗ := det([hϕk , ψl iV ∗ ]1≤k,l≤n ).

When applied to each fibre of T M for an oriented paracompact smooth manifold

M with respect to the inner product induced by a Riemannian metric g on
M, we get for each p ≥ 0 an inner product h·, ·iΩpc (M) : Ωpc (M) × Ωpc (M) → R
for ω, η ∈ Ωpc (M) by
ˆ
hω, ηiΩpc (M) := hω(x), η(x)iVp Tx∗ M dvol(x) .
M

In the case p = n, h·, ·iVp Tx∗ (M) was used in (iii). The latter extend naturally
to inner products

h·, ·iΩpC,c (M) : ΩpC,c (M) × ΩpC,c (M) −→ C.

Theorem 2.19 (Hodge-de Rham triple): Let M be an oriented compact smooth

manifold of dimension n with Riemannian metric g. Consider

(i) the unital complex ∗ -algebra A := C ∞ (M, C),

18
 Chapter 2. Spectral triples associated to manifolds

(ii) the separable complex Hilbert space H := L2 ( •C T ∗ M, g), which is ob-

V

tained as the completion of the complex exterior algebra Ω•C (M) :=

L •
p≥0 ΩC (M) with respect to the inner product given by

n
X
h(ω0 , . . . , ωn ), (η0 , . . . , ηn )i := hωp , ηp iΩpC (M)
p=0

and the ∗ -representation π : A → B(H) given by multiplication, i.e.,

π(f )(ω) := f ω for every f ∈ A and ω ∈ H,
(iii) the unbounded operator D0 := d+d∗ , where d∗ is the adjoint of the densely
defined operator

d : H ⊇ dom d −→ H, ω 7−→ dRe(ω) + idIm(ω)

with domain dom d := Ω•C (M).

Then D0 is essentially self-adjoint; let D be its closure, which we call the Hodge-
de Rham operator. The Hodge-de Rham triple (A, H, D) is a commutative
spectral triple in the sense of Definition 1.1. We call ∆ := D2 the Hodge
Laplacian.

Definition 2.20: Let V be a K vector space with an inner product h·, ·i. Put
V• Vp
K V :=
L
p≥0 K V . Then

• p
(−1)k+1 hvk , viv1 ∧· · ·∧vbk ∧· · ·∧vp .
^ X
⌞ : V ×V −→ V, v⌞(v1 ∧· · ·∧vp ) :=
K k=1

Remark 2.21: The proof of Theorem 2.19 relies mostly on techniques that are
(not yet) at our disposal. We can understand, however, how commutators
[D, φ(f )] for f ∈ A look on H. They are given by the Clifford multiplication
with df from the left, i.e., for all ω ∈ Ω•C we have

[D, π(f )]ω = df •w. (2.2)

The Clifford multiplication is defined on fibres as follows: On the exterior

algebra •C VC := p≥0 pC VC for the complexification VC := V ⊗R C of a finite
V L V

dimensional real Hilbert space (V, h·, ·i), we define

•
^ •
^
⌞ : VC × VC −→ VC
C C

19
Chapter 2. Spectral triples associated to manifolds

by
p
(−1)k+1 hvk , viC v1 ∧ · · · ∧ vbk ∧ · · · ∧ vp .
X
v ⌞ (v1 ∧ · · · ∧ vp ) :=
k=0

Note that, for v = u ⊗ λ ∈ VC , we put v := u ⊗ λ; the inner product

h·, ·i : VC × VC → C is defined by

hu1 ⊗ λ1 , u2 ⊗ λ2 i := hu1 , u2 iλ1 λ1 .

V•
Then v • ω := v ∧ ω − v ⌞ ω for all v ∈ VC and ω ∈ C.

(1) For all f ∈ A, ω ∈ Ω•C (M) it holds

d∗ (f ω) = f d∗ ω − df ⌞ ω.

To see this, we take η ∈ Ω•C (M) and compute with respect to the inner
product h·, ·i := h·, ·iΩ•C (M) that

hdη, f ωi = hf dη, ωi
= hd(f η), ωi − hdf ∧ η, ωi
= hf η, d∗ ωi − hη, df ⌞ ωi = hη, f d∗ ω − df ⌞ ωi

from which the assertion follows.

(2) For all f ∈ A, ω ∈ Ω•C (M) it holds

[D, π(f )] = df • ω

To see this, consider the following computation:

[D, π(f )] = [d, π(f )]ω + [d∗ , π(f )]ω

= d(f ω) − f dω + d∗ (f ω) − f d∗ ω = df ∧ ω − df ⌞ ω = df • ω.

That [D, π(f )] extends to bounded linear operator on H, will be discussed later.
Further, we note that Ω•C (M) ⊆ dom d∗ , which justifies that D0 = d + d∗ is
densely defined with dom D0 = Ω•C (M). This can be shown with the help of the
Hodge star operator ∗ : Ω•C (M) → Ω•C (M) which associates to each ω ∈ ΩkC (M)
the unique ∗ω ∈ Ωn−k n−k
C (M) such that for all x ∈ M and η ∈ ΩC (M)

(ω ∧ η)(x) = hη(x), (∗ω)(x)iVn−k T ∗ M dvol(x) .

C x

In fact, one can show that

d∗ |ΩkC (M) = (−1)nk+1 ∗ d ∗ .

20
 Chapter 2. Spectral triples associated to manifolds

That D0 is essentially selfadjoint follows from results about general symmetric

differential operators on manifolds (based on Friedrichs modifiers). In order to
verify that D has compact resolvents, one defines the Sobolev spaces

Hs := {ω ∈ H | (1 + ∆)s/2 ω ∈ H}

for s ≥ 0 and uses the Rellich Lemma to show that H1 ,→ H0 is compact, which
implies that
(1 + ∆)−1/2 : H = H0 −→ H1 ,−→ H0 = H
is compact and hence (D − i1)−1 is compact.

Remark 2.22: If the manifold M carries more structure (i.e., spinc -manifold),
there is another spectral triple (A, D, H) associated to M, with D being the
Dirac operator. We do not go into details here.

Exercises
Exercise 2.1: (i) Let x0 ∈ Rn be given. For j = 1, . . . , n, we define a linear
∂f
map ∂j |x0 : Cx∞0 (Rn ) → R by ∂j |x0 ([f ]x0 ) := (∂j f )(x0 ) = ∂x j
(x0 ) for every germ
∞ n
[f ]x0 ∈ Cx0 (R ). Prove that {∂j |x0 | j = 1, . . . , n} forms a basis of the tangent
space Tx0 Rn .
(ii) Let M be a n-dimensional smooth manifold with the maximal smooth
atlas A = {(Ui , φi ) | i ∈ I}. Show that for every i ∈ I and each x0 ∈ Ui , the
linear map
Θi,x0 : Rn −→ Tx0 M
that is defined by
  n  
v j ∂j (f ◦ φ−1
X
Θi,x0 (v) [f ]x0 := i ) (φi (x0 ))
j=1

for each v = (v 1 , . . . , v n ) ∈ Rn and every germ [f ]x0 ∈ Cx∞0 (M), is an isomor-

phism of real vector spaces.

Exercise 2.2: (i) Let E be an n-dimensional (real or complex) vector bundle

over a Hausdorff topological space X.
Construct an n-dimensional (real or complex) vector bundle E ∗ over X, such
that for each x ∈ X the fibre Ex∗ of E ∗ is the dual space of the fibre Ex of E,
i.e., Ex∗ = hom(Ex , K).
We call E ∗ the dual bundle of E.

21
Chapter 2. Spectral triples associated to manifolds

(ii) Let E be an n-dimensional smooth (real or complex) vector bundle over

a smooth manifold M. Show that the dual bundle E ∗ of E is also smooth.

Exercise 2.3: Complete the proof of Theorem 2.9 of the lecture by proving the
following assertions for a smooth manifold M of dimension n and an open
subset V ⊆ M:

(i) For every D ∈ der C ∞ (V ), the map Ψ(D) : V → T M, x 7→ (Ψ(D))(x)

belongs to X(V ). Recall that (Ψ(D))(x0 ) ∈ Tx0 M for any point x0 ∈ V is
defined by

(Ψ(D))(x0 ) : Cx∞0 (M) −→ R, [f ]x0 7−→ D|x0 (ρ · f |V ),

where ρ : M → [0, 1], for a chosen representative (U, f ) of the given germ [f ]x0 ,
is a bump function for (U, x0 ).
(ii) The induced linear map Ψ : der C ∞ (V ) → X(V ), D 7→ Ψ(D) satisfies

Φ ◦ Ψ = idder C ∞ (V ) and Ψ ◦ Φ = idX(V ) ,

where Φ : X(V ) → der C ∞ (V ) is the linear map defined in Theorem 2.9.

Exercise 2.4: Let M be an oriented paracompact smooth manifold of dimension

n and let g be a Riemannian metric on M. Show that for every f ∈ Cc∞ (M)
with the property that supp(f ) ⊂ U for some
´ local chart (U, ϕ) in the maximal
oriented smooth atlas A of M, the value M f that is assigned to f by formula
(2.1) of the lecture, does not depend on the particular choice of (U, ϕ).

Exercise 2.5: Let M be an oriented compact smooth manifold of dimension n

and let g be a Riemannian metric on M. Prove the identity

hdf ∧ η, ωiΩ•C (M) = hη, df ⌞ ωiΩ•C (M)

for all f ∈ C ∞ (M, C) and all ω, η ∈ Ω•C (M).

22
Chapter 3.

The geodesic distance in

noncommutative geometry

Associating a spectral triple such as the Hodge-de Rham triple to a compact

oriented smooth Riemannian manifolds follows the philosophy of “spectral
geometry”, where such classical geometric objects are studied by spectral
properties of canonically associated differential operators. At the same time,
this allows to carry classical concepts over to a noncommutative setting.
In this chapter, we discuss the geodesic distance within the framework of
spectral triples.
Definition 3.1 (Geodesic distance): Let (M, g) be a Riemannian manifold, i.e,
M is a paracompact smooth manifold with Riemannian metric g, and let
x0 , x1 ∈ M. Then
(i) We denote by Γ(x0 , x1 ) the set of all smooth paths γ : [0, 1] → M sat-
isfying γ(0) = x0 , γ(1) = x1 . Note that M is connected if and only if
Γ(x0 , x1 ) is non-empty for every choice of points x0 , x1 ∈ M.
(ii) If γ ∈ Γ(x0 , x1 ) is given, we define γ 0 (t) ∈ Tγ(t) M for each t ∈ [0, 1] by
γ 0 (t)([f ]γ(t) ) := (f ◦ γ)0 (t) for all [f ]γ(t) ∈ Cγ(t)
∞
(M).1 The length L(γ) of
γ is then defined as
ˆ 1
L(γ) := gγ(t) (γ 0 (t), γ 0 (t))1/2 dt
0

(iii) Suppose that M is connected. The geodesic distance dg (x0 , x1 ) between

x0 and x1 is defined as

dg (x0 , x1 ) := inf{L(γ) | γ ∈ Γ(x0 , x1 )}.

1
It happend in the same way in Remark 2.3.

23
Chapter 3. The geodesic distance in noncommutative geometry

Remark 3.2: (i) On a connected Riemannian manifold (M, g), the geodesic
distance induces a metric
dg : M × M −→ [0, ∞), (x0 , x1 ) 7−→ dg (x0 , x1 )
called the Riemannian distance function. Note that if x0 6= x1 , the geodesic
distance dg (x0 , x1 ) is a positive number. Indeed, for γ ∈ Γ(x0 , x1 ) and a local
chart (U, ϕ) with x0 ∈ U and x1 ∈ / U , we have for all t ∈ [0, T ] the equality
gγ(t) (γ 0 (t), γ 0 (t)) = hG(γ(t))v 0 (t), v 0 (t)i
where G = (gk,l )1≤k,l≤n : ϕ(U ) −→ Mn (R) defined for all x ∈ U by
 
gk,l (ϕ(x)) := gx (dϕ)(x)−1 (∂k |ϕ(x) ), (dϕ)(x)−1 (∂l |ϕ(x) )

(see Theorem 2.17) and the smooth map v : [0, T ] → Rn , t 7→ ϕ(γ(t)), where
T ∈ [0, 1] is chosen such that γ([0, T ]) ⊂ U .
Take r > 0 such that cl(B(ϕ(x0 ), r)) ⊂ ϕ(U ) and V := ϕ−1 (B(ϕ(x0 ), r))
which is an open subset of U . We find δ ∈ (0, 1] such that for all y ∈ B(ϕ(x0 ), r)
and ξ ∈ Rn it holds
δkξk ≤ hG(y)ξ, ξi1/2 ≤ δ −1 kξk.
Thus
ˆ T0
L(γ) ≥ gγ(t) (γ 0 (t), γ 0 (t))1/2 dt
0
ˆ T0
= hG(γ(t))v 0 (t), v 0 (t)i1/2 dt
0
ˆ T0 ˆ T0
0
≥δ kv (t)k dt ≥ δ v 0 (t) dt = δkv(T 0 ) − ϕ(x0 )k
0 0

for every T 0 ∈ (0, T ] with γ([0, T 0 ]) ⊂ V . By enlarging T and taking the limit
in T 0 , we infer that L(γ) ≥ δr > 0 and thus we have dg (x0 , x1 ) ≥ δr > 0.
(ii) The topology on M induced by the metric dg agrees with the given
topology on M. This can be shown by arguments similar to (i). In fact, one
shows that for each x0 ∈ M and a local chart (U, ϕ) with x0 ∈ U , an open
neighbourhood V ⊆ U of x0 and δ ∈ (0, 1], it exists r > 0 such that for all
x ∈ V it holds
δ|ϕ(x) − ϕ(x0 )| ≤ dg (x, x0 ) ≤ δ −1 |ϕ(x) − ϕ(x0 )|,
and dg (x, x0 ) ≥ δr for all x ∈ M − V .

24
 Chapter 3. The geodesic distance in noncommutative geometry

(iii) A connected paracompact smooth manifold is second countable (i.e.,

admits a countable base). Thus, it follows from the Urysohn metrisation
theorem that the topology on M must be metrisable; this is in keeping with
(ii).
(iv) Let (M1 , g1 ) and (M2 , g2 ) be two connected Riemannian manifolds.
Then every isometry ϕ : (M1 , g1 ) → (M2 , g2 ), i.e., a map ϕ : M1 → M2 that
satisfies for all x0 , x1 ∈ M1 that
dg2 (ϕ(x0 ), ϕ(x1 )) = dg1 (x0 , x1 )
is necessarily smooth and satisfies ϕ∗ g2 = g1 — this is the result of the Myers-
Steenrod theorem from 1939.
Note that if ϕ : M1 → M2 is a smooth immersion between (paracompact)
smooth manifolds M1 and M2 and if g is a Riemannian metric on M2 , then
ϕ∗ g is the Riemannian metric on M1 given by
 
(ϕ∗ g)x (α, β) := gϕ(x) (dϕ)(x)(α), (dϕ)(x)(β)

for all x ∈ M1 and α, β ∈ Tx M1 , where (dϕ)(x) : Tx M1 → Tx M2 is defined

as in (Theorem 2.17); ϕ is called an immersion, if (dϕ)(x) is injective for each
x ∈ M.

Our goal is to “dualise” the definition of the geodesic distance such that it
fits into the framework of spectral triples.
Theorem 3.3 (Musical isomorphisms): Let (M, g) be a Riemannian manifold.
Then g can be seen as a positive definite pairing on smooth vector fields, i.e., a
map
g : X(M) × X(M) −→ C ∞ (M)
which is C ∞ (M)-bilinear and satisfies g(X, X) ≥ 0 for all X ∈ X(M) and
g(X, X)(x) = 0 at x ∈ M if and only if X(x) = 0.
This induces an isomorphism (in fact, a C ∞ (M)-bimodule map)
[ : X(M) −→ Ω1 (M), X 7−→ X [ := g(X, ·).
Its inverse ] : Ω1 (M) → X(M), ω 7→ ω ] is determined by ω(X) = g(ω ] , X) for
all X ∈ X(M). The inner product on Ω1 (M) defined in (Remark 2.18) (iii)
satisfies ˆ
hω, ηiΩ1 (M) = g(ω ] , η ] )
M

for all ω, η ∈ Ω (M, since hω(x), η(x)iTx∗ (M) = gx (ω ] (x), η ] (x)) for each
1

x ∈ M.

25
Chapter 3. The geodesic distance in noncommutative geometry

Definition 3.4: Let (M, g) be a Riemannian manifold. The gradient grad f of

a function f ∈ C ∞ (M) is the vector field

grad f := (d0 f )] ∈ X(M)

with d0 f ∈ Ω1 (M) as defined in (Definition 2.14) (iii). We thus have that for
each x ∈ M and δ ∈ Tx M

gx (gradx f, δ) = (d0 f )(x)(δ). (3.1)

Definition 3.5: Let (M, g) be a compact Riemannian manifold. On X(M), we

define a norm k·k∞ by

kXk∞ := max gx (X(x), X(x)) ∀ X ∈ X(M).

x∈M

Theorem 3.6: Let (M, g) be a compact and connected Riemannian manifold.

Then, for all x0 , x1 ∈ M, we have that

dg (x0 , x1 ) = sup{|f (x1 ) − f (x0 ) | f ∈ C ∞ (M) : kgrad f k∞ ≤ 1}.

Proof: Take any γ ∈ Γ(x0 , x1 ). Then, for every f ∈ C ∞ (M), it holds

ˆ 1
f (x1 ) − f (x0 ) = f (γ(1)) − f (γ(0)) = (f ◦ γ)0 (t) dt
0

and for each t ∈ [0, 1], we have

(f ◦ γ)0 (t) = (d0 f )(γ(t))(γ 0 (t)) = gγ(t) (gradγ(t) f, γ 0 (t)),

so that the Cauchy-Schwarz inequality yields

|(f ◦ γ)0 (t)| ≤ gγ(t) (gradγ(t) f, gradγ(t) f )1/2 gγ(t) (γ 0 (t), γ 0 (t))1/2
≤ kgrad f k∞ gγ(t) (γ 0 (t), γ 0 (t))1/2 .

In summary, we get |f (x1 ) − f (x0 )| ≤ kgrad f k∞ L(γ). We infer from the latter
that

sup{|f (x1 ) − f (x0 ) | f ∈ C ∞ (M) : kgrad f k∞ ≤ 1} ≤ dg (x0 , x1 ).

In order to prove “≥”, we consider the function f0 : M → R, x 7→ d(x0 , x).

While f0 is not (necessarily) smooth, the triangle inequality for dg impiles that
f0 is at least Lipschitz continuous with Lipschitz constant 1. For every ε > 0,

26
 Chapter 3. The geodesic distance in noncommutative geometry

we find hε ∈ C ∞ (M) such that kf0 − hε k∞ < ε and kgrad hε k∞ ≤ 1 + ε; put

1
fε := 1+ε hε ∈ C ∞ (M). Then kgrad fε k∞ ≤ 1 and it holds

1
|fε (x1 ) − fε (x0 )| = |hε (x1 ) − hε (x0 )
1+ε
1
= |f0 (x1 ) − (f0 (x1 ) − hε (x1 )) + (hε (x0 ) − f0 (x0 ))|
1+ε
1  
≥ |f0 (x1 )| − (|f0 (x1 ) − hε (x1 )| + |f0 (x0 ) − hε (x0 )|
1+ε
1 2ε
≥ dg (x0 , x1 ) − ,
1+ε 1+ε
which shows that for every ε > 0 it holds
dg (x0 , x1 ) 2ε
sup{|f (x1 ) − f (x0 )| | f ∈ C ∞ (M) : kgrad f k∞ ≤ 1} ≥ − .
1+ε 1+ε
We conclude by taking the limit ε ↓ 0. 

The preceeding motivates the following definition:

Definition 3.7: Let (A, H, D) be a spectral triple. We define by

A := clk·k∞ (π(A)) ⊆ B(H)

a C ∗ -algebra and denote by S(A) the state space of A. We define for ϕ, ψ ∈ S(A)
by

dD (ϕ, ψ) := sup{|ψ(π(a)) − ϕ(π(a))| | a ∈ A : k[D, π(a)]k ≤ 1} ∈ [0, ∞]

the spetral distance between ϕ and ψ.

In view of Remark 3.2 (iv), the following theorem says that the Hodge-de
Rham triple remembers the metric.
Theorem 3.8: Let (M, g) be a compact oriented Riemannian manifold and
(A, H, D) be the Hodge-de Rham triple for (M, g). Then the faithful representa-
tion π : A → B(H) extends to a faithful ∗ -representation π̂ : C(M, C) → B(H)
which induces an isometric ∗ -isomorphism
∼
=
πb : C(M, C) A := clk·k (A) ⊆ B(H).

Define δx ∈ S(A) for x ∈ M by δx (π̂(f )) := f (x) for f ∈ C(M, C). If M is

commutative, then dg (x0 , x1 ) = dD (δx0 , δx1 ) for all x0 , x1 ∈ M.

27
Chapter 3. The geodesic distance in noncommutative geometry

Proof: (1) By definition of H, it is easily seen that each f ∈ C(M, C)

defines an operator πb (f ) ∈ B(H) by πb (f )ω := f ω for all ω ∈ H with
kπb (f )k ≤ kf k∞ ; in fact, we have that kπb (f )k = kf k∞ , since πb (f ) restricts to
the ordinary multiplication operator on L2 (M, g) := cl(Ω0C (M)), for which
we know kπb (f )|L2 (M,g) k = kf k∞ . Thus, πb : C(M, C) → B(H), f 7→ πb (f ) is a
faithful (isometric) ∗ -representation, which extends π : A → B(H) and induces
an isometric ∗ -isomorphism C(M, C) ∼ = A.
(2) By equation (Eq. 2.2) in (Remark 2.20), we have for f = f ∈ A, and
forms ω ∈ Ω•C (M), and each point x ∈ M that

([D, π(f )]ω)(x) = (df )(x) • ω(x) =: c((df )(x))ω(x),

where, for each real v ∈ Tx∗ MC , the operator

• •
Tx∗ MC −→ Tx∗ MC ,
^ ^
c(v) : ω 7−→ v · ω
C C

is an isometry (refer to (Exercise 3B 1(ii))). Hence

k([D, π(f )]ω)(x)kV• Tx∗ MC ≤ k(df )(x)kTx∗ MC · kω(x)kV• Tx∗ MC .

C C

We conclude that
 
k[D, π(f )]ωk Ω•C (M) ≤ maxk(df )(x)k Tx∗ MC · kωkΩ•C (M)
x∈M

and so, by (Definition 3.4) and (Definition 3.5),

k[D, π(f )]k ≤ maxk(df )(x)kTx∗ MC = kgrad f k∞ .

x∈M

Optimising ω, we get that k[D, π(f )]k = kgrad f k∞ and thus, by (Theorem
3.6) and (Exercise 4AB - 1(ii))

dg (x0 , x1 ) = sup{|f (x1 ) − f (x0 )| | f ∈ C ∞ (M) : kgrad f k∞ ≤ 1}

= sup{|δx1 (f ) − δx0 (f )| | f = f ∈ A : k[D, π(f )]k ≤ 1} = dD (δx0 , δx1 )

which concludes the proof. 

28
 Chapter 3. The geodesic distance in noncommutative geometry

Exercises
Exercise 3.1: Let (A, H, D) be a spectral triple with the faithful ∗ -representation
π : A → B(H). Consider the state space S(A) for the associated C ∗ -algebra
A := clk·k (π(A)) ⊆ B(H). Prove the following assertions:
(i) If the image of the set {a ∈ A | k[D, π(a)]k ≤ 1} under the canonical
projection in the quotient Banach space A/C1 is a norm bounded set,
then the spectral distance satisfies dD (φ, ψ) < ∞ for all φ, ψ ∈ S(A) and
induces a metric dD : S(A) × S(A) → [0, ∞).
(ii) For all φ, ψ ∈ S(A), we have that
dD (φ, ψ) = sup{|ψ(π(a)) − φ(π(a))| | a = a∗ ∈ A : k[D, π(a)]k ≤ 1}.
Hint: To prove “≤”, establish first that the set {a ∈ A | k[D, π(a)]k ≤ 1}
is closed under the following maps: a 7→ ζa for each ζ ∈ C with |ζ| = 1,
a 7→ a∗ , a 7→ Re(a) = 12 (a + a∗ ), and a 7→ Im(a) = 2i1 (a − a∗ ).
Exercise 3.2: Let (A1 , H1 , D1 ) and (A2 , H2 , D2 ) be spectral triples with the
faithful ∗ -representations π1 : A1 → B(H1 ) and π2 : A2 → B(H2 ), respectively.
We call these two spectral triples equivalent, if there exists a ∗ -isomorphism
Φ : A1 → A2 and a unitary operator U : H1 → H2 such that for all a ∈ A1 it
holds
U π1 (a)U ∗ = π2 (Φ(a))
and U D1 U ∗ = D2 .
Show that in this case adU : B(H1 ) → B(H2 ), x 7→ U xU ∗ is an isometry
which satisfies adU (A1 ) = A2 , where A1 and A2 are the C ∗ -algebras associated
to A1 and A2 , respectively, and prove that ad∗U : S(A2 ) → S(A1 ), φ 7→ φ ◦ adU
defines an isometry for the spectral distances, i.e.,
dD1 (ad∗U φ, ad∗U ψ) = dD2 (φ, ψ) for all φ, ψ ∈ S(A2 ).
Exercise 3.3: Consider the complex unital ∗ -algebra A = C ⊕ C with entry-
wise operations. Let H1 and H2 be finite dimensional complex Hilbert spaces
and put H := H1 ⊕ H2 . Define the ∗ -homomorphism π : A → B(H) for all
a = (a1 , a2 ) ∈ A by !
a1 idH1 0
π(a) := .
0 a2 idH2
Further, take any linear operator M : H1 → H2 and consider the operator
!
0 M∗
D := .
M 0

29
Chapter 3. The geodesic distance in noncommutative geometry

(i) Verify that (A, H, D) is a spectral triple. Compute for all a = (a1 , a2 ) ∈ A
the commutator [D, π(a)] and show that its norm is given by

k[D, π(a)]k = |a2 − a1 |kM k.

(ii) Consider the states δ1 , δ2 : A → C that are respectively given by δ1 (a) = a1

and δ2 (a) = a2 for each a = (a1 , a2 ) ∈ A. Compute the spectral distance
dD (δ1 , δ2 ).
(iii) Show that the spectral triple (A, H, D) is even, i.e., there is a selfadjoint
operator Γ ∈ B(H) with the properties that Γ2 = idH , DΓ + ΓD = 0,
and π(a)Γ = Γπ(a) for all a ∈ A. We call Γ a grading on (A, H, D).

30
Chapter 4.

The Riemann-Lebesgue measure in

noncommutative geometry

In (Chapter 3), we have seen that the geodesic distance on a connected, compact
and oriented Riemannian manifold can be recovered from its associated Hodge-
de Rham triple via Connes spectral distance.
In this chapter, we will discuss the noncommutative integral, by which
integration of (smooth) functions with respect to the Riemann-Lebesgue measure
on Riemannian manifolds as introduced in (Theorem 2.17) is generalised to the
framework of spectral triples.
Like in quantum mechanics, the underlying idea is that operators on a
separable complex Hilbert space H with dim H = ∞ take over the role of
complex variables, while selfadjoint operators on H correspond to real variables.
Remark 4.1: Recall (Theorem 9.8 from the Functional Analysis I lecture notes)
that T ∈ B(H) is compact if and only if T can be approximated in operator
norm on B(H) by finite rank operators; equivalently,
∀ ε > 0 ∃ V ⊆ H subspace, dim V < ∞ : kT |V ⊥ k < ε,
where T |V ⊥ : V ⊥ → H is the restriction of T to V ⊥ and k·k is the norm on
B(V ⊥ , H).
For a compact operator T ∈ B(H), we call the non-zero eigenvalues (µn (T ))n≥0
of |T | := (T ∗ T )1/2 , arragned in decreasing order and repeated according to
multiplicity, the characteristic values of T . Note that µn (T ) converges to 0 as
n → ∞. We have that for all n ∈ N0
µn (T ) = inf{kT − Sk | S ∈ B(H) : dim ran S ≤ n}
(4.1)
= inf{kT |V ⊥ k | V ⊆ H subspace, dim V = n}
and in particular µ0 (T ) = kT k.

31
Chapter 4. The Riemann-Lebesgue measure in noncommutative geometry

In view of Remark 4.1, compact operators are considered as “infinitesimals”

in our “quantised calculus”; their “size” is measured by the rate of decay of
their sequence of characteristic values.
Definition 4.2: Let T ∈ K(H) and α > 0 be given. We say that T is an
infinitesimal of order α, if µn (T ) = O(n−α ) as n → ∞, i.e., if there is a
constant C < ∞ such that for all n ∈ N it holds µn (T ) ≤ Cn−α . For α > 0, we
denote by Iα (H) the set of all T ∈ K(H) which are infinitesimals of order α.

Remark 4.3: Recall (Theorem 9.5 from the Functional Analysis I lecture notes)
that K(H) is a norm-closed two-sided ideal in B(H). It follows from Eq. (4.1)
that for all T ∈ K(H) and S ∈ B(H)

µn (T S) ≤ kSkµn (T ) and µn (ST ) ≤ kSkµn (T )

and that for T1 , T2 ∈ K(H)

µn+m (T1 + T2 ) ≤ µn (T1 ) + µm (T2 );

thus each Iα (H) forms a (non-closed) two-sided ideal in B(H). Furthermore,

we have for T1 , T2 ∈ K(H) that

µn+m (T1 T2 ) ≤ µn (T1 )µm (T2 )

which implies the following rule for infinitesimals: If T1 is of order α1 and T2 is

of order α2 , then T1 T2 is of order α1 + α2 .

We want to find an “integral” that is defined on I1 (H) and neglects all

infinitesimals of order α > 1.
Remark 4.4: An operator T ∈ B(H) is said to be in trace class, if ∞ k=0 h|T |ξk , ξk i
P

is finite for some (and in turn for each) orthonormal basis (ξk )k∈N0 of H. In
this case, the sum ∞ k=0 hT ξk , ξk i is absolutely convergent and its value Tr(T )
P

is indepentend of the choice of the orthonormal basis (ξk )k∈N0 of H; we call

Tr(T ) the trace of T .
The set L 1 (H) of all trace class operators on H forms a (non-closed) two-
sided ideal in B(H). However, L 1 (H) is a Banach space with respect to the
norm kT k1 := Tr(|T |). Note that L 1 (H) ⊆ K(H). If T ∈ L 1 (H) is positive,
then ∞
X
Tr(T ) = µn (T ). (4.2)
n=0

Thus, Tr : L 1 (H) → C is not appropriate for our purpose, because:

32
 Chapter 4. The Riemann-Lebesgue measure in noncommutative geometry

• I1 (H) is not a subset of L (1,∞) (H), so that Tr is not defined on all

infinitesimals of order 1.
• Tr does not vanish even on finite rank operators, but those belong to
Iα (H) for every α > 1.

Definition 4.5: For T ∈ K(H), we define for each N ∈ N

N −1
X 1
σN (T ) := µn (T ) and γN (T ) := σN (T ).
n=0 log N

Remark 4.6: Note that σN (T ) is a partial sum in Eq. (4.2). For T ∈ I1 (H),
we find constants C, C 0 < ∞ such that for all N ∈ N
N −1
C
≤ C 0 log(N ).
X
σN (T ) = kT k +
n=1 n

Consequently, (γN (T ))N ∈N is a bounded sequence. Thus I1 (H) ⊆ L (1,∞) (H),

where n o
L (1,∞) := T ∈ K(H) kT k(1,∞) := sup γN (T ) < ∞
N ∈N

is the Dixmier ideal. Note that L (H) ⊆ L

1 (1,∞)
(H).

Proposition 4.7: Consider operators T1 , T2 ∈ K(H). For all N ∈ N, we have

that

σN (T1 + T2 ) ≤ σN (T1 ) + σN (T2 )

and if T1 , T2 are positive, we have in addition that

σ2N (T1 + T2 ) ≥ σN (T1 ) + σN (T2 ).

It follows that for any positive T1 , T2 ∈ K(H)

!
log 2
γN (T1 + T2 ) ≤ γN (T1 ) + γN (T2 ) ≤ γ2N (T1 + T2 ) 1 + .
log N

Proof: Exercise! 

Theorem 4.8 (Dixmier traces): Let ω : `∞ (N, R) → R be a linear map such

that

(i) ω((αN )N ∈N ) ≥ 0 if αN ≥ 0 for all N ∈ N,

33
Chapter 4. The Riemann-Lebesgue measure in noncommutative geometry

(ii) ω((αN )N ∈N ) = limN →∞ αN , if (αN )N ∈N is convergent,

(iii) ω((α2N )N ∈N ) = ω((αN )N ∈N ) for each (αN )N ∈N ∈ `∞ (N, R).1
(1,∞)
Then, Trω : L (1,∞) (H+ → [0, ∞) for all λ ≥ 0, T , T1 , T2 ∈ L+ satisfies the
conditions

Trω (T1 + T2 ) = Trω (T1 ) + Trω (T2 ), Trω (λT ) = λ Trω (T )

and extends uniquely to a positive linear map Trω : L (1,∞) (H) → C, which for
all S ∈ B(H), T ∈ L (1,∞) (H) and U ∈ L 1 (H) ⊆ L (1,∞) (H) satisfies

Trω (ST ) = Trω (T S) (4.3)

and
Trω (U ) = 0. (4.4)
⊆ L 1 (H). We call Trω a Dixmier trace.
S
Note that α>1 Iα (H)

Proof: For every T ∈ L (1,∞) (H)+ , (γN (T ))N ∈N is bounded (see Remark 4.6);
hence, Trω (T ) is well-defined and clearly Trω (T ) ≥ 0 by (i). For T1 , T2 ∈
L (1,∞) (H)+ , it follows from Proposition 4.7 and property (i) of ω that

Trω (T1 + T2 ) ≤ Trω (T1 ) + Trω (T2 )

! !
  log(2)
≤ ω (γ2N (T1 + T2 )N ∈N + ω γ2N (T1 + T2 ) .
log(N ) N ∈N

By property (ii) of ω, we get that

! !
log(2) log(2)
ω γ2N (T1 + T2 ) = lim γ2N (T1 + T2 ) = 0,
log(N ) N ∈N
N →∞ log(N )

since γ2N (T1 + T2 ) is bounded and property (iii) yields that

   
ω (γ2N (T1 + T2 )N ∈N = ωω (γN (T1 + T2 )N ∈N = Trω (T1 + T2 ).

In summary, we get that Trω (T1 + T2 ) = Trω (T1 ) + Trω (T2 ). That for all λ ≥ 0
and T ∈ L (1,∞) (H)+ it holds Trω (λT ) = λ Trω (T ) is clear since ω is linear and
γN (λT ) = λγN (T ) for each N ∈ N.
The extension of Trω to L (1,∞) (H) uses that
1
This condition is called the scale invariance.

34
 Chapter 4. The Riemann-Lebesgue measure in noncommutative geometry

• each T ∈ L (1,∞) (H) can be written uniquely as T = <(T ) + i=(T ) with

selfadjoint <(T ), =(T ) ∈ L (1,∞) (H);
• each T = T ∗ ∈ L (1,∞) (H) can be written as T = P+ |T |P+ − P− |T |P− for
projections P+ , P− ∈ vN(T ) satisfying P+ − P− = 1 (using the measurable
functional calculus Φ̃ : Bb (Sp(T )) → vN(T ) from Remark 6.12 (iv), Functional
Analysis II, with P+ := Φ̃(χ[0,∞) ) and P− := Φ̃(χ(−∞,0) ), where P+ |T |P+ ,
P− |T |P− ∈ L (1,∞) (H)+ due to Remark 4.3 since |T | ∈ L (1,∞) (H)+ .

Let U ∈ B(H) be unitary. To prove Eq. (4.3), we note first that Eq. (4.1)
implies for n ∈ N0 and for all T ∈ K(H) that µn (U T U ∗ ) = µn (T ), which yields
by definition for N ∈ N that for all T ∈ K(H) it holds γN (U T U ∗ ) = γN (T )
and hence it holds for all T ∈ L (1,∞) (H) that Trω (U T U ∗ ) = Trω (T ). Because
L (1,∞) (H) is a two-sided ideal, this is equivalent to the statement that for all
T ∈ L (1,∞) (H) it holds Trω (U T ) = Trω (T U ).
Since every S ∈ B(H) is a linear combination of (in fact four) unitaries (see
the proof of Lemma 6.13 in the Functional Analysis II lecture notes), the latter
yields Eq. (4.3).
To verify Eq. (4.4), we take T ∈ L 1 (H) and without loss of generality, we
may assume that T ≥ 0. Note that (σN (T ))n∈N is bounded by kT k1 due to
Eq. (4.2), thus γN (T ) → 0 as N → ∞, so that by property (iii) we have
 
Trω (T ) = ω (γN (T ))N ∈N = lim γN (T ) = 0.
N →∞

Note that Iα (H) ⊆ L 1 (H) for all α > 1, since for each T ∈ Iα (H), we find
C < ∞ such that
∞ ∞
n−α < ∞.
X X
kT |1 = µn (T ) ≤ kT k + C
n=0 n=1 

Definition 4.9: Let T ∈ L (1,∞) (H). We say that T is measurable

ffl of the value
of Trω (T ) is independent of ω; we denote this common value by T and call it
the noncommutative integral of T . Moreover, we put

M(H) := {T ∈ L (1,∞) (H) | T measurable}.

Remark 4.10: (i) The existence of (in fact infinitely many) linear maps
ω : `∞ (N, R) → R satisfying the conditions (i), (ii) and (iii) in Theorem 4.8
was proved by Dixmier in 1966: With the construction of Trω , he proved the
existence of singular traces on B(H) (i.e., traces that vanish on L 1 (H)) and
settled to the negative the question of the uniqueness of the trace on B(H).

35
Chapter 4. The Riemann-Lebesgue measure in noncommutative geometry

(ii) An alternative approach was developed by Connes. It relies on the

(piecewise linear) interpolation of (σN (T ))N ∈N given by

σλ (T ) := inf{kRk1 + λkSk | R ∈ L 1 (H), S ∈ K(H) : T = R + S}

for each λ > 0. For T ∈ L (1,∞) (H) and any a > e,

σλ (T )
γ : [a, ∞) −→ R, λ 7−→
log(λ)
is a continuous and bounded function, its Cesàro mean with respect to the
Haar measure du
u
on the multiplicative group (0, ∞) is for each λ ∈ [0, ∞) given
by ˆ λ
1 du
τλ (T ) := γ(u)
log(λ) a u
and defines a function λ 7→ τλ (T ) in Cb ([a, ∞)). For T ∈ L (1,∞) (H)+ , let τ̇ (T )
be the class of λ 7→ τλ (T ) in the quotient C ∗ -algebra B := Cb ([a, ∞))/C0 ([0, ∞)).
One can show that τ̇ : L (1,∞) (H)+ → B extends to a positive linear map
τ̇ : L (1,∞) (H) → B with the property that for each S ∈ B(H) and for each
T ∈ L (1,∞) (H) it holds τ̇ (ST ) = τ̇ (T S).
For every state ω on B, one defines Trω : L (1,∞) (H) → C by

Trω (T ) := ω(τ̇ (T ))

for all T ∈ L (1,∞) (H), this map Trω then satisfies Eq. (4.3) and Eq. (4.4) in
Theorem 4.8. Moreover, we have that T ∈ L (1,∞) (H) is measurable if and only
if limλ→∞ τλ (T ) exists, in which case

T = lim τλ (T ).
λ→∞

Note that exhibiting a state ω on the (non-separable) C ∗ -algebra requires the

axiom of choice.
(iii) M(H) is a vector space and satisfies ST S −1 ∈ M(H) for all invertible
S ∈ B(H) and T ∈ M(H). Moreover, one can check that M(H) ( M(H).
(iv) With the help of real interpolation theory, one can construct out of
L 1 (H) ⊆ K(H) a family L (p,q) (H) with 1 < p < ∞ and 1 ≤ q ≤ ∞ of two-
sided ideals in B(H). In fact, L (p,q) (H) for q < ∞ consists of those T ∈ K(H)
which satisfy
∞
1
N ( p −1)q−1 σN (T )q < ∞,
X

N =1

36
 Chapter 4. The Riemann-Lebesgue measure in noncommutative geometry

while L (p,∞) (H) consists of those T ∈ K(H) for which

1
kT k(p,∞) := sup N q −1 σN (T ) < ∞;
N ∈N

it follows that L (p,∞) (H) = I1/p (H) for each p > 1. On the diagonal, one finds
the Schatten-ideals L p (H) := L (p,p) (H), where the interpolation norm k·k(p,p)
on L (p,p) (H ) is equivalent to the Schatten-p-norm
kT kp := Tr(|T |p )1/p
for T ∈ L p (H). A noncommutative integration theory for which L p (H) serves
as a analoge of the Lp -space in Lebesgue integration theory, was developed by
Segal in the fifties.
Definition 4.11: Let (A, H, D) be a spetrac triple. We say that (A, H, D) is
(i) p-summable if (1 + D2 )−1/2 ∈ L p (H),
(ii) (p, ∞)-summable if (1 + D2 )−1/2 ∈ L (p,∞) (H),
(iii) θ-summable if e−tD ∈ L 1 (H).
2

Example 4.12: Consider the spectral triple (A, H, D) from Example 1.2, where
A = C ∞ (T, C), H = L2 (T, m) and D was the closure of
1
D0 : H ⊃ dom D0 −→ H, g 7−→ g 0
i
with domain dom D0 := C 1 (T). Then the operator ∆ := D2 has spectrum
Sp(∆) = {|n|2 | n ∈ Z}. We conclude that (1 + ∆)−1/2 ∈ L (1,∞) (H), i.e.,
the spectral triple (A, H, D) is (1, ∞)-summable. Indeed, since the Fourier
transform F : L2 (T, m) → `2 (Z) is a unitary and FDF −1 = M(n)n∈Z , where
Mλ , for any sequence λ = (λn )n∈Z of complex numbers, is the closed operator
Mλ : `2 (Z) ⊇ dom Mλ −→ `2 (Z), (an )n∈Z 7−→ (λn an )n∈Z
with domain dom Mλ := {(an )n∈Z ∈ `2 (Z) | (λn an )n∈Z ∈ `2 (Z)} we conclude
that
F∆F −1 = M(|n|2 )n∈Z , F(1 + ∆)−1/2 F −1 = M(1+n2 )−2 )n∈Z ∈ B(`2 (Z))
and finally (1 + ∆)−1/2 ∈ L (1,∞) (H); in fact, we have
!

−1/2
 1 1 1 1
µn ((1 + ∆) ) = 1, √ , √ , . . . , √ , √ ,... ,
n∈N0 2 2 1 + n2 1 + n2
so that
ffl γN ((1 + ∆)−1/2 ) → 2 as N → ∞ and hence even (1 + ∆)−1/2 ∈ M(H)
with (1 + ∆)−1/2 = 2.

37
Chapter 4. The Riemann-Lebesgue measure in noncommutative geometry

Example 4.13: In the situation of Example 4.12, let P ∈ B(H) be the orthogo-
nal projection onto ker D = C1 ⊂ L2 (T, m). By Exercise 1B-1, (A, H, D̃) with
D̃ := DP = D + P gives another spectral triple. Note that ∆ ˜ := D̃2 = ∆ + P ,
˜ becomes invertible. We have ∆
so that ∆ ˜ −1/2 ∈ L (1,∞) (H) since

˜ −1/2 ))n∈N = 1, 1, 1, 1 , 1 , . . . , 1 , 1 , . . .
 
(µn (∆ 0
2 2 n n

and ˜ −1/2 ) → 2 as N → ∞; in fact, we have ∆

thus γ̃N (∆ ˜ −1/2 ∈ M(H) and
ffl −1/2
∆˜ = 2.
˜ −1/2 ∈ L (1,∞) (H) because
More generally, for f ∈ A = C ∞ (T, C), we have f ∆
f · : L (T, m) → L (T, m), g 7→ f g is bounded and L (1,∞) (H) is a two-sided
2 2

ideal in B(H). It is a consequence of Connes trace theorem that f · ∆ ˜ −1/2 is

measureable and ˆ
˜ −1/2 1
f ·∆ = f (ζ) dm(ζ) . (4.5)
π T
The general version of Connes trace theorem (1988) is about pseudodifferential
operators on compact Riemannian manifolds. This theory has its origins in the
work of Kohn, Nirenberg, Hörmander and others in the sixties.

Definition 4.14: Let M be a compact smooth manifold of dimension n and let

πE : E → M be a k-dimensional smooth vector bundle.

(i) A differential operator of order m is a linear operator

P : Γ(M, E) −→ Γ(M, E)

which, in local coordinates x = (x1 , . . . , xn ) of M, is of the form

∂ α1 ∂ αn
Aα (x)(−i)|α|
X
P = · · ·
|α|≤m
∂xα1 1 ∂xαnn

where α = (α1 , . . . , αn ) is a multi-index with entries 0 ≤ α1 , . . . , αn ≤ n

and cardinality |α| := nj=1 αj , the Aα ∈ Mk (C ∞ (M)) for each |α| ≤ m
P

with Aα 6≡ 0 for some multiindex α with |α| = n.

(ii) For ξ ∈ Tx∗ M, written as ξ = nj=1 ξj dxj , we define the complete symbol
P

of P (as the polynomial in ξ1 , . . . , ξn given) by pP (x, ξ) := m P

P
d=0 pd (x, ξ),
where
pPd (x, ξ) := Aα (x)ξ α := Aα (x)ξ1α1 · · · ξnαn .
X X

|α|=d |α|=d

38
 Chapter 4. The Riemann-Lebesgue measure in noncommutative geometry

The principal symbol of P is defined as

σ P (x, ξ) := pPm (x, ξ) = Aα (x)ξ α .
X

|α|=d

It induces a linear map σ P (ξ) : Ex → Ex for each ξ ∈ Tx∗ M.

(iii) We say that the differential operator P is elliptic if its principal symbol
σ P (ξ) : Ex → Ex is invertible for each x ∈ M and any ξ ∈ Tx∗ M − {0}.
(iv) For a local section u of E, one can write
ˆ
1
(P u)(x) = eihξ,xi pP (x, ξ)ub(ξ) dξ1 · · · dξn (4.6)
(2π)n/2 Rn
with the Fourier transform ub of u which is given by
ˆ
1
ub(ξ) = n/2
e−ihξ,xi u(x) dx1 · · · dxn .
(2π) Rn

A linear operator P : Γ(M, E) → Γ(M, E) is called pseudodifferential

operator of order m, written P ∈ Ψm (M, E), if Eq 4.6 holds locally for a
matrix-valued function pP in the symbol class Symm (M, E), i.e., in local
coordinates, a matrix of smooth functions whose derivatives satisfy the
growth conditions
|∂xα ∂ξβ pi,j (x, ξ)| ≤ Cα,β (1 + |ξ|)m−|β| .
The principal symbol of P is then defined as
σ P := [pP ] ∈ Symm (M, E)/Symm−1 (M, E).
(v) Suppose that g is a Riemannian metric on M. The Wodzicki residue of
P ∈ Ψ−n (M, E) is defined by
ˆ
1
ResW (P ) := tr σ P (x, ξ)ωξ ∧ dx,
(2π)n S ∗ M
where S ∗ M := {(x, ξ) ∈ Tx∗ M | h·, ·iTx∗ M = 1} is the co-sphere bundle
over M, tr is the matrix trace, dx := dx1 ∧ · · · ∧ dxn and
n
ˆ j ∧ · · · ∧ dξn .
(−1)j−1 ξ dξ1 ∧ · · · ∧ dξ
X
ωξ :=
j=1

Theorem 4.15 (Connes’ trace theorem, 1988): Let (M, g) be a compact Rie-
mannian manifold of dimension n. For P ∈ Ψ−n (M, E) with a complex vector
bundle E over M, the following statements hold:

39
Chapter 4. The Riemann-Lebesgue measure in noncommutative geometry

(i) P extends to a bounded linear operator on the Hilbert space L2 (M, E),
which is obtained by completion of Γ(M, E) with respect to the inner
product given by
ˆ
hu1 , u2 i := u2 (x)∗ u1 (x) dvol(x) .
M

Moreover, P ∈ L (1,∞) (L2 (M, E).

Exercises
Exercise 4.1: Let H be an infinite dimensional separable complex Hilbert space.
(i) Let T ∈ K(H) and N ∈ N be given. Prove the formula
n o
σN (T ) = inf kRk1 + N kSk R ∈ L 1 (H), S ∈ K(H) : T = R + S

for the value σN (T ) that was defined in Definition 4.5 of the lecture.
(ii) Like in Remark 4.10 (ii), we define for every T ∈ K(H) and each λ > 0
n o
σλ (T ) := inf kRk1 + λkSk R ∈ L 1 (H), S ∈ K(H) : T = R + S .

Due to (i), this interpolates the values σN (T ). Show that this interpolation
is in fact piecewise linear, i.e., prove that σλ (T ) = λkT k holds for every
λ ∈ [0, 1) and that
σN +λ (T ) = (1 − λ)σN (T ) + λσN +1 (T )
holds for each N ∈ N and every λ ∈ [0, 1).

Exercise 4.2: Let (A1 , H1 , D1 ) and (A2 , H2 , D2 ) be spectral triples with infinite
dimensional separable complex Hilbert spaces H1 , H2 and suppose that the
operator Γ1 ∈ B(H1 ) is a grading on (A1 , H1 , D1 ). Put
A := A1 ⊗C A2 , H := H1 ⊗C H2 , and D := D1 ⊗ idH2 +Γ1 ⊗ D2 .
Prove the following assertions:
(i) (A, H, D) is a spectral triple.
(ii) (A, H, D) is θ-summable2 whenever at least one of the spectral triples
(A1 , H1 , D1 ) and (A2 , H2 , D2 ) is θ-summable.
2
Recall from Definition 4.11 that a spectral triple (A, H, D) is said to be θ-summable if
2
e−tD ∈ L 1 (H) for each t > 0.

40
 Chapter 4. The Riemann-Lebesgue measure in noncommutative geometry

Exercise 4.3: Consider the Fourier transform F : S(Rn ) → S(Rn ) on the

Schwartz space S(Rn ). Prove the following properties:

(i) For each u ∈ S(Rn ) and each multi-index α = (α1 , . . . , αn ) ∈ Nn0 , we

have that ∂ α (Fu) = (−i)|α| F(mα u), where mα denotes the function

mα : Rn −→ C, (x1 , . . . , xn ) 7−→ xα1 1 · · · xαnn .

(ii) For each u ∈ S(Rn ) and each multi-index α ∈ Nn0 , we have that

F(∂ α u) = i |α| mα Fu.

6 Ω ⊆ Rn be open. Consider a differential operator

Exercise 4.4: Let ∅ =
∞ ∞
P : C (Ω) → C (Ω) which is of the form

aα (−i)|α| ∂ α
X
P =
|α|≤m

for some integer m ≥ 0 and with coefficients aα ∈ C ∞ (Ω) for each |α| ≤ m.
Let
pP : Ω × Rn −→ R, aα ξ α
X
(x, ξ) 7−→
|α|≤m

be the complete symbol of P . Prove that for each u ∈ S(Rn ) and every point
x ∈ Rn ˆ
1
(P u|Ω )(x) = n/2
eihξ,xi pP (x, ξ)û(ξ) dλn (ξ) .
(2π) Rn

41
Appendix

43
Appendix A.

Basics of unbounded operators

Let (H1 , h·, ·i) and (H2 , h·, ·i2 ) be complex Hilbert spaces; the norms induced
by the inner product are denoted by k · k1 and k · k2 , respectively.

I. The notion of unbounded linear operators

By an unbounded (linear) operator from H1 to H2 , we mean a linear map

T : H1 ⊇ dom T −→ H2

that is defined on a linear subspace dom T of H1 , called the domain of T . We

say that T is densely defined if dom T is dense in H1 , i.e., if clk·k1 (dom T ) = H1 .
The graph of T , which we will denote by G(T ), is defined as

G(T ) := {(x, T x) ∈ H1 × H2 | x ∈ dom T }.

It is thus linear subspace of the Hilbert space H1 ⊕ H2 with the inner product
given by

h(x1 , y1 ), (x2 , y2 )i := hx1 , x2 i1 + hy1 , y2 i2 for (x1 , y1 ), (x2 , y2 ) ∈ H1 ⊕ H2 .

Lemma A.1: A linear subspace G ⊆ H1 ⊕H2 is the graph of an unbounded linear

operator (i.e., there is an unbounded linear operator T : H1 ⊇ dom T → H2
such that G = G(T )) if and only if G ∩ ({0} × H2 ) = {(0, 0)}.

II. Closed and closable operators

Let T : H1 ⊇ dom T → H2 be an unbounded operator from H1 to H2 .

45
III. The adjoint operator

An unbounded operator S : H1 ⊇ dom S → H2 is called an extension of T ,

written as S ⊆ T , if G(T ) ⊆ G(S) holds, i.e., if dom T ⊆ dom S and Sx = T x
for every x ∈ dom T .
We say that the unbounded operator T is

• closed, if G(T ) is closed in H1 ⊕ H2 ; explicitly, this means that for every

sequence (xn )n∈N in dom T which converges in H1 to a point x ∈ H1 and for
which (T xn )n∈N is convergent in H2 to a point y ∈ H2 , it holds true that
x ∈ dom T and y = T x.
• closable, if T admits an extension S that is closed.

Theorem A.2: For an unbounded operator T : H1 ⊇ dom T → H2 , the follow-

ing statements are equivalent:

(i) T is closable;
(ii) For every sequence (xn )n∈N in dom T which converges to 0 in H1 and for
which (T xn )n∈N converges in H2 to a point y ∈ H2 , we necessarily have
that y = 0;
(iii) cl(G(T )) ∩ ({0} × H2 ) = {(0, 0)}.

It is worthwhile to take a closer look on the proof that (iii) implies (i). It
follows from Lemma A.1 that if (iii) holds, then cl(G(T )) must be the graph
of an unbounded linear operator, say T : H1 ⊇ dom T → H2 . The operator
T is thus a closed extension of T ; in fact, it is the (unique) minimal closed
extension (i.e., for every other closed operator S that satisfies T ⊆ S, it follows
that T ⊆ S), called the closure of T . Furthermore, its domain dom T is the
closure of dom T with respect to the graph norm k · kT which is defined by
kxk2T := kxk21 + kT xk22 for each x ∈ dom T .

III. The adjoint operator

Let now T : H1 ⊇ dom T → H2 be a densely defined unbounded linear operator.
For every y ∈ H2 , we introduce a linear functional

φy : dom T −→ C, x 7−→ hT x, yi2 .

Using this notation, we may define

dom T ∗ := {y ∈ H2 | φy is continuous on dom T with respect to k · k1 },

46
which is clearly a subspace of H2 . Since dom T is dense in H1 , φy for every
y ∈ dom T ∗ extends uniquely to a bounded linear functional Φy on H1 ; by the
Riesz representation theorem, the latter must be of the form Φy (x) = hx, T ∗ yi1
for all x ∈ H1 with a unique vector T ∗ y ∈ H1 . The assignment y 7→ T ∗ y is in
fact linear on dom T ∗ , so that this construction results in an unbounded linear
operator
T ∗ : H2 ⊇ dom T ∗ −→ H1 ,
called the adjoint of T .

Theorem A.3: Let T : H1 ⊇ dom T → H2 be a densely defined unbounded

linear operator.

(i) The adjoint operator T ∗ is always closed.

(ii) If T is closed, then T ∗ is densely defined and the operator T ∗∗ := (T ∗ )∗
satisfies T ∗∗ = T .
(iii) The operator T is closable if and only if its adjoint T ∗ is densely defined;
in this case, we have that T ∗∗ = T .

IV. Symmetric and selfadjoint operators

Throughout the following, let (H, h·, ·i) be a complex Hilbert space. A densely
defined operator T : H ⊇ dom T → H is called

• symmetric, if T ⊆ T ∗ , or in other words, if hT x1 , x2 i = hx1 , T x2 i for all

x1 , x2 ∈ H.
• selfadjoint, if T = T ∗ .
• maximally symmetric, if T is symmetric and if the following holds: when-
ever S is a symmetric extension of T , it follows that S = T .
• essentially selfadjoint, if T is symmetric with selfadjoint closure T .

Lemma A.4:

(i) Every symmetric operator is closable.

(ii) Every selfadjont operator is maximally symmetric.
(iii) A densely defined operator T : H ⊇ dom T → H is essentially selfadjoint
if and only if T = T ∗ .

47
V. Resolvent set and spectrum

Suppose that T is a densely defined and symmetric operator. The defect

indices n± (T ) ∈ [0, ∞] of T are defined by

n+ (T ) := dim(T + i)⊥ = dim ker(T ∗ − i)

and n− (T ) := dim(T − i)⊥ = dim ker(T ∗ + i).

Theorem A.5: Let T : H ⊇ dom T → H be densely defined and symmetric.

Then the following statements are equivalent:

(i) T is essentially selfadjoint;

(ii) n+ (T ) = n− (T ) = 0;
(iii) ran(T + i) and ran(T − i) are dense.

Suppose in addition that T is closed. Then the following statements are equiva-
lent:

(i) T is selfadjoint;
(ii) n+ (T ) = n− (T ) = 0;
(iii) ran(T + i) = ran(T − i) = H.

For closed operators, we actually have the following.

Theorem A.6: Let T : H ⊇ dom T → H be densely defined, closed, and sym-

metric. Then we have the following:

(i) T is selfadjoint if and only if n+ (T ) = n− (T ) = 0.

(ii) T is maximally symmetric if and only if n+ (T ) = 0 or n− (T ) = 0.
(iii) T has a selfadjoint extension if and only if n+ (T ) = n− (T ).

V. Resolvent set and spectrum

For any densely defined unbounded linear operator T : H ⊇ dom T → H, we
define its resolvent set ρ(T ) by
n o
ρ(T ) := λ ∈ C (T − λ1) : dom T → H is bijective and (T − λ1)−1 ∈ B(H)

and its spectrum σ(T ) by σ(T ) := C − ρ(T ).

Lemma A.7: Suppose that T : H ⊇ dom T → H is densely defined and closed.

48
 (i) If (T − λ1) : dom T → H is bijective for a λ ∈ C, then its inverse
(T − λ1)−1 is bounded.
(ii) The spectrum σ(T ) ⊆ C is closed.
(iii) If T is selfadjoint, then σ(T ) ⊆ R.
(iv) If T is symmetric and satisfies σ(T ) ⊆ R, then T is selfadjoint.

VI. The spectral theorem and functional calculus

Theorem A.8: Let T : H ⊇ dom T → H be selfadjoint. Then there is a unique
spectral measure E such that
ˆ
hT x, yi = λ dhE(λ)x, yi for all x ∈ dom T , y ∈ H.
R

If h : R → R is measurable, then
ˆ
hh(T )x, yi = h(λ) dhE(λ)x, yi
R

defines a selfadjoint operator h(T ) : H ⊇ dom h(T ) → H with domain

 ˆ 
dom h(T ) := x ∈ H |h(λ)|2 dhE(λ)x, xi < ∞ .
R

49
Appendix B.

Basics on Fourier transform

We recall some basic facts about the Fourier transform on Rn which can be
used for the solution of the exercises without proof.

(i) Let λn be the Lebesgue measure on Rn . For every function u ∈ L1 (Rn , λn ),

we define its Fourier transform û = Fu by
ˆ
1
(Fu)(ξ) := e−ihξ,xi u(x) dλn (x) for each ξ ∈ Rn ,
(2π)n/2 Rn

where h·, ·i denotes the standard inner product on Rn , i.e., hξ, xi = nj=1 ξj xj for
P

each x = (x1 , . . . , xn ) and ξ = (ξ1 , . . . , ξn ) in Rn . It is known that Ff ∈ C0 (Rn ).

(ii) Let S(Rn ) be the Schwartz space, i.e., the space of all smooth functions
f : Rn → C satisfying

sup (1 + |x|m )|(∂ α f )(x)| < ∞

x∈Rn

for each m ∈ N0 and each multi-index α = (α1 , . . . , αn ) ∈ Nn0 . The Fourier

transform F, if restricted to S(Rn ), has the remarkable property that it induces
a bijection F : S(Rn ) → S(Rn ), with inverse given by
ˆ
−1 1
(F v)(x) = eihξ,xi v(ξ) dλn (ξ) for each x ∈ Rn .
(2π)n/2 Rn

51
Appendix C.

Solutions to the exercises

Solution (to Exercise 1.1): According to Definition 1.1, we have to check that
(i) DV is selfadjoint,
(ii) DV has compact resolvent,
(iii) For all a ∈ A, π(a) dom DV is contained in DV and [DV , π(a)] is bounded
on H.
Ad (i): This is a general fact: If T : H ⊇ dom T → H and S ∈ B(H),
(T + S)∗ = T ∗ + S ∗ . In particular, if T and S are selfadjoint, then T + S is
also selfadjoint.
To see this, note that

dom(T + S)∗ = {y ∈ H | x 7→ h(T + S)x, yi = hT x, yi + hSx, yi is bounded}

= {y ∈ H | x 7→ hT x, yi is bounded}
= dom T ∗

since x 7→ hSx, yi is bounded for every y ∈ H. Thus, for all x ∈ dom T and for
all y ∈ dom T ∗ it holds

hT x, yi + hSx, yi = h(T + S)x, yi = hx, (T + S)∗ yi = hx, (T ∗ + S ∗ )yi,

i.e., (T + S)∗ = T ∗ + S ∗ .
Ad (ii): Take any λ ∈ C − σ(DV ), choose λ1 ∈ C − σ(D) and λ2 := λ − λ1 .
Then

(DV − λ1)−1 − (D − λ1 1)−1

 
= (D − λ1 1)−1 (D − λ1 1) − (DV − λ1) (DV − λ1)−1
= −(D − λ1 1)−1 (V − λ2 1)(DV − λ1)−1 ,

53
Appendix C. Solutions to the exercises

i.e., we have

(DV − λ1)−1 = (D − λ1 1)−1 (1 − (V − λ2 1)(DV − λ1)−1 ),

where (D − λ1 1)−1 is compact and (1 − (V − λ2 1)(DV − λ1)−1 ) is bounded,

which yields that (DV − λ1)−1 is compact (recall that K(H) is a two-sided
ideal in B(H)).
Ad (iii): For all a ∈ A, π(a) dom DV ⊆ dom DV , as dom DV = dom D.
Furthermore,
[DV , π(a)] = [D, π(a)] + [V, π(a)],
where [V, π(a)] ∈ B(H), extends to a bounded operator on H.

Solution (to Exercise 2.1): (i) First we want to show, that

{∂j |x0 | 1 ≤ j ≤ n}
Pn
is R-linearly independent. Let therefore α1 , . . . , αn ∈ R with j=1 αj ∂j |x0 = 0
in Tx0 Rn . Then, for each [f ]x0 ∈ Cx∞0 (R), it holds
n n
∂f
αj ∂j |x0 ([f ]x0 ) = αj
X X
0= (x0 ).
j=1 j=1 ∂xj

For 1 ≤ i ≤ n, we apply this to the equivalence classes [fi ] of the functions

fi : Rn −→ R, (x1 , . . . , xn ) 7−→ xi − xi0 ,

∂fi
which gives 0 = nj=1 αj ∂x i
P
j (x0 ) = α , as desired.

For {∂j |x0 | 1 ≤ j ≤ n} to be a basis of Tx0 Rn , we need to show that indeed

Lin{∂j |x0 | 1 ≤ j ≤ n} = Tx0 Rn . To see this, we first need a technical result.
For every open set U ⊆ Rn and each smooth function f : U → R, we find by
Taylor’s theorem on every open ball B(x0 , r) ⊆ U with r > 0 a smooth function
ϕ : B(x0 , r) → R such that for all x ∈ B(x0 , r) we have

f (x) = f (x0 ) + h(grad f )(x0 ), x − x0 i + ϕ(x),

where ϕ(x)/(x − x0 ) → 0 for x → x0 . In fact, we can give such a function ϕ

explicitly: For all x ∈ B(x0 , r) we put
n
g j (x)fi (x)
X
ϕ(x) := hg(x), x − x0 i =
j=1

54
 Appendix C. Solutions to the exercises

where g = (g 1 , . . . , g n ) : B(x0 , r) → Rn is given by

ˆ 1" #
j ∂f ∂f
g (x) := (x0 + s(x − x0 )) − j (x0 ) ds .
0 ∂xj ∂x

Note that each g j , 1 ≤ j ≤ n, is smooth with g j (x0 ) = 0.

For any δ ∈ Tx0 Rn , any representative f of the class [f ]x0 in Cx∞0 (R) and the
corresponding smooth function ϕ defined as above we thus find
n
X  n  
[g j ]x0 [fj ]x0 = δ([g j ]x0 )fj (x0 ) + g j (x0 )δ([fj ]x0 ) = 0
X
δ([ϕ]x0 ) = δ
j=1 j=1

and hence
n
X ∂f
δ([f ]x0 ) = δ([fj ]x0 ) (x0 ) ∈ Lin{∂j |x0 | 1 ≤ j ≤ n},
j=1 ∂xj

since δ([fj ]x0 ) ∈ R for 1 ≤ j ≤ n, as we wanted to show.

(ii) Note that we have an isomorphism

φi,x0 : Cx∞0 (M) −→ Cϕ∞i (x0 ) (Rn ), [f ]x0 7−→ [f ◦ ϕ−1

i ]ϕi (x0 )

and thus an isomorphism

φbi,x0 : Tϕi (x0 ) Rn −→ Tx0 M, δ 7−→ δ ◦ Φi,x0 .

Take the isomorphism from (i), i.e., the map

n
ψϕi (x0 ) : Rn −→ Tϕi (x0 ) Rn , (v 1 , . . . , v n ) 7−→ v j ∂j |ϕi (x0 ) ,
X

j=1

this yields an isomorphism Θi,x0 := φbi,x0 ◦ ψϕi (x0 ) : Rn → Tx0 M, which at

v = (v 1 , . . . , v n ) ∈ Rn and [f ]x0 ∈ Cx∞0 (M) looks as
n
X 
Θi,x0 (v)([f ]x0 ) = φbi,x0 v j ∂j |ϕi (x0 ) ([f ]x0 )
j=1
n
X 
= v j ∂j |ϕi (x0 ) (φi,x0 ([f ]x0 ))
j=1
n
X  n
v j ∂j |ϕi (x0 ) ([f ◦ ϕ−1 v j (∂j (f ◦ ϕ−1
X
= i ]ϕi (x0 ) ) = i ))(ϕi (x0 ))
j=1 j=1

which is the given assertion.

55
Appendix C. Solutions to the exercises

Solution: (i) We define

E ∗ := Ex∗ = {(x, f ) | x ∈ X, f ∈ Ex∗ }, π ∗ : E ∗ −→ X, (x, f ) 7−→ x.

a

x∈X

Let A = {(Ui , τi ) | i ∈ I} be a maximal bundle atlas for the bundle π : E → X,

with local trivialisation τi : π −1 (Ui ) → Ui × Kn . By definition, for each x ∈ Ui
the map τi |Ex : Ex → {x} × Kn is an isomorphism, thus

(τi |Ex )0 : (K n )∗ −→ Ex∗ , g 7−→ g ◦ (τi |Ex )

is an isomorphism as well. Fixing an isomorphism Φ : Kn → (Kn )∗ allows us to

define τi∗ : (π ∗ )−1 (Ui ) → Ui × Kn fibre-wise by
 −1
τi∗ |Ex∗ := (τi |Ex )0 ◦ Φ : Ex∗ −→ Kn = {x} × Kn .

Endowing E ∗ with the topology, whose open sets are characterised by “W ⊆ E ∗

is open if and only if τi∗ (Ui ∩ W ) ⊆ Ui × Kn is open for every i ∈ I” makes E ∗
a vector bundle with bundle atlas A∗ = {(Ui , τi∗ ) | i ∈ I}.
∗
(ii) The transition maps σi,j of E ∗ are of the form
∗ ∗
σi,j (x, v) = (x, Si,j (x)v) ∀ (x, v) ∈ (Ui ∩ Uj ) × Kn
∗
with the transition matrices Si,j : Ui ∩ Uj → Gln (K) that are determined at any
point x ∈ Ui ∩ Uj by

LEE (Si,j
∗
(x)) = τj∗ |Ex∗ ◦ (τi∗ |Ex∗ )−1
= ((τj |Ex )0 ◦ Φ)−1 ◦ ((τi |Ex )0 ◦ Φ)
= Φ−1 ◦ ((τi |Ex ) ◦ (τj |Ex )−1 )0 ◦ Φ
= Φ−1 ◦ (LEE (Sj,i (x)))0 ◦ LEE ∗ (In )
∗
= Φ−1 ◦ LEE ∗ (Sj,i (x)> ) ◦ Φ = LEE (Sj,i (x)> )

where Si,j : Ui ∩ Uj → Gln (K) are the transition matrices for E, E is the
standard basis of Kn , E ∗ is the dual basis to E and we fix Φ that satisfies
Φ(ei ) = e∗i for i ∈ {1, . . . , n}.
∗
From the above calculation we conclude Si,j (x) = Sj,i (x)> = (Si,j (x)−1 )> . In
∗
particular, if E is smooth, then E is smooth as well.

56