The Functor Category

Categorical Methods in Representation Theory, Bristol, Sept. 2012

Mike Prest
School of Mathematics, Alan Turing Building, University of Manchester
Manchester M13 9PL, UK
[email protected]
September 26, 2012

Contents
1 Lecture 1: Projective Functors 1
1.1 Module categories and functors . . . . . . . . . . . . . . . . . . . 1
1.2 The Yoneda embedding and projective functors . . . . . . . . . . 4
1.3 Finitely presented functors in (mod-R, Ab) . . . . . . . . . . . . 8

2 Injective Functors 11
2.1 The tensor embedding . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Injective functors and pure-injective modules . . . . . . . . . . . 13
2.3 Duality of finitely presented functors . . . . . . . . . . . . . . . . 16
2.4 Injectives and projectives in the category of finitely presented
functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

3 Simple Functors 18
3.1 Simple functors and indecomposable modules . . . . . . . . . . . 18
3.2 An example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3 Serre subcategories and localisation . . . . . . . . . . . . . . . . . 20
3.4 Krull-Gabriel dimension . . . . . . . . . . . . . . . . . . . . . . . 22

1 Lecture 1: Projective Functors

1.1 Module categories and functors
Let A be a skeletally small preadditive category.

preadditive means that, for each pair, A, B, of objects of A, the set, A(A, B),
we write just (A, B), of morphisms from A to B is equipped with an abelian
group structure such that composition is bilinear: f (g + h) = f g + f h and
(g + h)f = gf + hf when these are defined.
These notes have been extracted and modified from [28].

1
A (skeletally) small category is one where the collection of (isomorphism
classes of) objects is a set.
For example, if R is a ring then consider the category which has just one
object, , and which has (, ) = R with the addition on R giving the abelian
group structure on (, ) and the multiplication on R giving the composition
(well use the convention for composition of morphisms that rs means do s then
r). This allows us to regard the ring R as a small preadditive category with just
one object.
L More generally if A has only finitely many objects, A1 , . . . , An , set R =
i,j (Ai , Aj ). Define addition and multiplication (i.e. composition) on R point-
wise, with the convention
S that if the domain of f is not equal to the codomain
of g, where f, g i,j (Ai , Aj ), then the product f g is zero. It is easy to check
that R is a ring and that 1 = e1 + + en , where ei is the identity map of Ai ,
is a decomposition of 1 R into a sum of orthogonal idempotents. This ring
R codes up almost all the information contained in the category A (the objects
Ai might not be recoverable from R, but thats not a big deal).
Example 1.1. Let A be the quiver
1 /2 /3 / ...
Let k be any field and let A be the k-path category of this quiver. That is,
the objects of A are the vertices of the quiver and the arrows of A are freely
generated as a k-category, that is under composition and forming well-defined
k-linear combinations, by the arrows of the quiver. Then dimk (i, j) = 1 if i j
and (i, j) = 0 if i > j. This is a preadditive category.
We consider only additive functors between preadditive categories; that is
functors F which satisfy F (f +g) = F (f )+F (g) whenever f +g is defined. Our
convention throughout is that functor means additive functor. A
left A-module is a functor from A to the category, Ab, of abelian groups. For
instance in the one-object () case, obtained from a ring R = End(), a functor
from A to Ab is determined by the image of , an abelian group - let us denote it
by M - together with, for each r R = (, ), an endomorphism of this abelian
group M. If we denote the action of this endomorphism as m 7 rm then, by
functoriality, (sr)m = s(rm). Thus a functor from this category, which we may
as well write as R, to Ab is a left R-module. It is easy to see that, conversely,
every left R-module gives rise to a functor from this one-point category to Ab
and that, furthermore, the natural transformations between functors are exactly
the R-linear maps between modules. Therefore, (R, Ab) ' R-Mod where the
latter is our notation for the category of left R-modules (we will use R-mod for
the category of finitely presented modules).
A natural transformation from the functor F to the functor G (where
F G : C D) is given by, for each object C C, a morphism C : GC
f
C 0 is a morphism
F C such that these all cohere in the sense that if C
of C then the diagram below commutes.

FC
C
/ GC

Ff Gf
 
F C0 / GC 0
C 0

2
 In the general case, where A is a small preadditive category, we may, there-
fore, write A-Mod for (A, Ab), especially since this reminds us that func-
tors are just generalised modules. For example, if R is a ring we could write
(mod-R)-Mod instead of (mod-R, Ab) (but we wont).
Theorem 1.2. If A is a skeletally small preadditive category then the functor
category A-Mod = (A, Ab) is abelian, indeed Grothendieck. There is, moreover,
a generating set of finitely generated projective objects, which we describe at 1.9.

An abelian category is a preadditive category which has finite direct sums

and a zero object, such that every morphism has a kernel and every
monomorphism is a kernel and, dually every morphism has a cokernel
and every epimorphism is a cokernel.
A Grothendieck category is an abelian category which has coproducts, in
which direct limits are exact and which has a generator.
A directed system is a collection of objects, Ci with i I where I is an
ordered set, and morphisms fij , where i, j I with i < j, such that
fjk fij = fik for all i < j < k and such that for all i, j I there is k I
with i, j < k.
The direct limit
(also called directed colimit) of a directed system (Ci )i , (fij :
Ci Cj )i<j is an object C and morphisms fi : Ci C with i I, such
that fi = fj fij for all i < j and which is universal in the sense that
if we also have an object D and system (gi : Ci D)iI of morphisms
such that gi = gj fij for all i < j, then there is a unique morphism
h : C D such that gi = hfi for all i.
Direct limits being exact means that if we have a directed system of exact
f g
sequences 0 A B C 0 then the direct limit sequence
lim f lim g
0 lim A
lim B

lim C 0 is exact.


A generating set for a preadditive category C is a set G of objects such for
every nonzero morphism f : A B in C there is G G and a morphism
g : G A such that f g 6= 0. If G = {G} then G is a generator.

Every Grothendieck category is a localisation of a module category (this is the

Gabriel-Popescu Theorem).
One may check that the category, Mod-R, of right R-modules is equivalent
to the category of contravariant functors to Ab from the one-point category
(corresponding to) R. More generally, and bearing in mind that contravariant
functors from a category are covariant functors from its opposite, we may write
(in the general case, replacing R by A) Mod-A for (Aop , Ab).
In a case such as Example 1.1, where the morphism sets are k-vectorspaces
and composition is k-bilinear, it is natural to consider, instead of the category of
additive functors to Ab, the naturally equivalent category of k-linear functors
to the category k-Mod ' Mod-k of k-vectorspaces. It is easily checked that the
forgetful functor from Mod-k to Ab induces a natural equivalence of functor
categories (A, Ab) ' (A, Mod-k).
Suppose that C is a skeletally small preadditive category and that D is an
abelian category (we have in mind D = Ab or D = Mod-k).

3
 The functor category (C, D) has, for its objects, the additive functors from
C to D and, for its morphisms, from a functor G to a functor F the natural
transformations from G to F . When one is trying to make sense of functor
categories, there is no harm in imagining that functor means module, that
subfunctor means submodule and that natural transformation between
functors means linear map between modules. For example the notion of
submodule generalises that of a subfunctor of F , which is simply a subobject
of F in the functor category: that is, a functor G and a natural transformation
: G F which is monic in (C, D), meaning that if , : H G are natural

transformations such that = then = , equivalently if H G is such
that = 0 then = 0.
The next lemma says that if D is abelian then subfunctors and quotient
functors are described pointwise.
Lemma 1.3. Suppose that C is a preadditive category, that D is an abelian
category and let : G F be a morphism in (C, D).
(i) is monic iff for every object C C the component C : GC F C at C is
a monomorphism in D;
(ii) is epi iff for every object C C the component C : GC F C at C is an
epimorphism in D.
The image and kernel of a morphism between functors in (C, D), where D
is abelian, are also given locally: if : G F is such a morphism then ker( )
is the subfunctor of G given at C C by ker( ) C = ker(C ) GC; also
im( ) F is given on objects by im( ) C = im(C ) F C. In each case the
action of the functor on morphisms is the obvious one.
Lemma 1.4. Suppose that C is preadditive and D is abelian. The sequence

0 F0 F F 00 0 of functors in (C, D) is exact iff for every C C the
C C
sequence 0 F 0 C F C F 00 C 0 is an exact sequence in D.
In particular, in the situation of 1.4, if F 0 F is an inclusion of functors
then the quotient F/F 0 is given on objects by (F/F 0 )C = F C/F 0 C. Also the
direct sum, F G, of two functors is given pointwise.

Direct limits also are computed pointwise. Let (F ) , ( : F F )
be a directed system of functors in (C, D) where now we assume that D is
Grothendieck abelian (for example, a module category or a functor category),
so has direct limits which are exact. Then, generalising 1.2,

the functor category
also is Grothendieck abelian. Let F, ( : F F ) be the direct limit of
this directed system in (C, D). For any
object C of C there is a directed system

, (( )C : F C F C) ) in D. Then F C, (( )C : F C
(F C)
F C) is, one may check, the direct limit of this system. As for the action of F
on morphisms, note that a morphism f : C C 0 gives rise to a morphism (the
obvious definition) between the corresponding directed systems in D and hence,
using the definition of direct limit, to a morphism which is F f : F C F C 0 .

1.2 The Yoneda embedding and projective functors

If C is an object of the preadditive category C then there is the corresponding
representable functor (C, ) : C Ab defined on objects by D 7 (C, D)
and on morphisms by f : D E maps to (C, f ), where (C, f ) : (C, D) (C, E)

4
is defined by (C, f )g = f g for g (C, D). There is also the corresponding
contravariant representable functor (, C) defined on objects by D 7 (D, C)
and on morphisms in the obvious way.
Lemma 1.5. (Yoneda Lemma) Let C be any preadditive category. Take C C
and F (C, Ab). Then there is a natural identification

((C, ), F ) ' (C, )

between the group of natural transformations from the representable fuctor (C, )
to F and the value group of F at C.
Naturality includes that if : F G is a natural transformation then
((C, ), ) : ((C, ), F ) ((C, ), G) is identified with the map C : F C
GC.
Similarly, if G (C op , Ab) then ((, C), G) ' GC.
Proof. The isomorphism = C,F : ((C, ), F ) ' F C is defined as follows. Let
((C, ), F ) and consider the component of at C, that is, C : (C, C) F C.
Define ( ) = C (1C ). Note that this element C 1C determines as follows:
the component of at D, D : (C, D) F D, is defined by D f = F f (C 1C ).

C (C, C)
C
/ FC

f (C,f ) Ff
  
D (C, D) / FD
D

It is straightforward to check that this works.

The first use of the term natural in the statement refers to the dependence
of on C and F . 

The Yoneda Lemma (theorem and proof) is completely general (the basic
version is for arbitrary categories and functors to Set). In particular it also
applies to the category, (C, Mod-k), of k-linear functors if C is a k-linear category.
It follows that, if (C, ) and (D, ) are representable functors then

((C, ), (D, )) ' (D, C).

In fact, it is immediate from 1.5 that we have a full contravariant embedding of

C to (C, Ab), called the Yoneda embedding.

A faithful functor F : A B between preadditive categories is one for which

F f = 0 implies f = 0; in particular F A = 0 implies A = 0.
A full functor F is one such that every morphism g : F A F B between
objects in the image of F has the form F f for some f : A B. So a full
and faithful functor allows allows its domain to be identified with a full
subcategory of its codomain.

Corollary 1.6. (see e.g. [8, 5.32]) Given any preadditive category C the functor
from C op to (C, Ab) given on objects by C 7 (C, ) and on morphisms by send-
ing f : C D to (f, ) : (D, ) (C, ) where (f, ) is given by (f, )g = gf
whenever g (D, E), is a full and faithful embedding.

5
 Dually, the (covariant) Yoneda embedding of C into (C op , Ab) given on ob-
jects by C 7 (, C), and on morphisms in the obvious way, is a full and faithful
embedding.

Example 1.7. Let R be a ring, regarded as a preadditive category with one

object. The Yoneda embedding from Rop to (R, Ab) = R-Mod is given by
taking the object of R to the functor (, ). This functor takes to (, ) = R
and takes r R = (, ) to the map s 7 rs (s R): (, ) (, ) - this is just
the left module R R.

s rs

 
/
r

t /

st s
 

The action of this Yoneda embedding on morphisms is to take t R = (, ) to
the endomorphism of R R which takes s R R to st. The fact that the Yoneda
embedding is full and faithful is just the assertion that End(R R) = Rop (i.e. the
right action of R).
Of course the dual Yoneda embedding from R (as a category) to (Rop , Ab) =
Mod-R takes to the right module RR and R = (, ) isomorphically to R =
End(RR ).
Lemma 1.8. If A is an object of the small preadditive category C then the
representable functor (A, ) is a finitely generated projective object of (C, Ab).
In particular, it is finitely presented.
P
Proof. If (A, ) = F is a directed sum of subfunctors then 1A (A, A) =
P
F A so 1A F A for some . By the Yoneda lemma (1.5) this gives a

morphism (A, ) F which, composed with the inclusion F (A, ), is

1(A,) ; for that is the endomorphism of (A, ) which Yoneda-corresponds to
1A . Therefore the inclusion F (A, ) is an epimorphism, as required.
For projectivity, take an epimorphism : F G of functors and a morphism
: (A, ) G which, by Yoneda, corresponds to an element a GA. Since
A : F A GA is epi there is b F A mapping to a.

(A, )


{ 
F /G

FA / GA
A

b /a

6
Suppose that ((A, ), F ) Yoneda-corresponds to b. Then = by the
Yoneda lemma. 

finitely generated An object C of a preadditive category C is finitely gener-

ated if it is not the direct
P limit of any directed system of proper subobjects.
Equivalently, if F = iI Fi then F = Fi1 +. . . Fin for some i1 , . . . , in I.
finitely presented An object C of a preadditive category C is finitely pre-
sented if the representable functor (C, ) : C Ab commutes with direct
limits (equivalently with filtered colimits).
Filtered colimits are more general than directed colimits = direct limits in
that they allow that in the underlying diagram there can be more than
one morphism f, g : Ci Cj but these also are directed in the sense that,
for any such pair, there is k > j and h : Cj Ck such that hf = hg (this
replaces the condition fik = fjk fij in the definition of direct limit.
If C is a module or, more generally functor, category then these definitions are
equivalent to the usual ones in terms of generators and relations, and to
those in terms of projective resolutions.

Proposition 1.9. (e.g. [8, p. 119]) If C is a skeletally small preadditive category

then the representable functors generate the functor category
L (C, Ab), indeed, for
every functor F : C Ab there is an epimorphism i (Ai , ) F for some
Ai C. A functor F is finitely generated iff this direct sum may be taken to be
finite.
Proof. For each isomorphism class of objects of C take a representative A.
(F A)
L
Define the morphism A (A, ) F to have component at a F A the
morphism fa : (A, ) F which Yoneda-corresponds to a. This map is surjec-
tive, essentially by definition. P
For the second
L statement, we have F = i im(fi ) where fi is the i-th com-
ponent map of i (Ai , ) F, so F finitely generated implies that F is a sum
of finitely many of these, therefore finitely many of the direct summands will
do. The converse follows since, by 1.8, each (Ai , ) is finitely generated (and
every image of a finitely generated object is finitely generated). 

Say that idempotents split in the preadditive category C if for every A C

each idempotent e = e2 End(A) has a kernel and the canonical map ker(e)
ker(1 e) A is an isomorphism.
Corollary 1.10. If C is a small preadditive category then the finitely generated
projective objects of (C, Ab) are the direct summands of finite direct sums of
representable functors. If C has split idempotents and has finite direct sums
then these are precisely the representable functors.
Proof. The first statement follows from 1.8 and 1.9.
For the second, if C has finite direct sums then (Ai , ) (Aj , ) ' (Ai
Aj , ) so, if F is finitely generated and projective, F is, without loss of gen-
erality, a direct summand of a functor of the form (A, ). Let : (A, ) F
split the inclusion and let f End(A, ) be followed by the inclusion: so
f 2 = f. The Yoneda embedding is full and faithful so there is e End(A) with

7
(e, ) = f hence with e2 = e. By assumption A = im(e) ker(e) and then it
follows quickly that F ' (im(e), ) so F is representable. 

In particular the finitely generated projectives of (R-mod, Ab) are the func-
tors (A, ) with A R-mod and, together, these generate the functor category.
Of course all the above applies if we replace C by C op . In particular the
representable functors (, A), A C, yield (closing under direct summands of
finite direct sums) the finitely generated projective objects of (C op , Ab). The
injective objects of the functor category (R-mod, Ab) are described in 2.15.
Example 1.11. We continue Example 1.1. We write i,i+1 for the arrow of
the quiver going from vertex i to vertex i + 1, more generally, ij (i j) for
compositions of these in the path category A, with ii the identity at i. Clearly
an additive functor from A to Mod-k is equivalent to a k-representation of the
quiver.
The simple functors/representations/modules are the Si , i 1, where Si has
dimension 1 at vertex i and is zero elsewhere. For each vertex i there is the in-
decomposable projective Pi (the projective cover of the simple object Si ) which
is 1-dimensional at each j i and 0 elsewhere. This is the representable functor
(i, ) and one can see that the embeddings P1 P2 . . . , of each indecom-
posable projective as the radical of the next, shows Aop Yoneda-embedded into
the functor category (A, Mod-k). Dually, the indecomposable injective repre-
sentations are the Ei = E(Si ), which have dimension 1 at each vertex j i and
zero elsewhere, plus one more, E = P1 , which has dimension 1 at each vertex.
The right modules - the contravariant functors - are the covariant functors for
the opposite quiver - the quiver with the same vertices but all arrows reversed.
f g
Lemma 1.12. (e.g. [8, 3.1, 3.21]) If A
B
C 0 is an exact sequence
(g,) (f,)
in mod-R then the induced sequence 0 (C, ) (B, ) (A, ) of
representable functors is exact in (mod-R, Ab).
f g (g,) (f,)
If A
B C is a sequence in mod-R such that (C, ) (B, )
(A, ) is exact in (mod-R, Ab) then the original sequence is exact. In particular
a morphism h of C is an epimorphism iff (h, ) is a monomorphism and, if
(h, ) is an epimorphism then h is a monomorphism.

1.3 Finitely presented functors in (mod-R, Ab)

We will use the following facts about finitely presented objects (they hold in
any abelian category with a generating set of finitely presented objects).
Lemma 1.13. Let C be a module or functor category and let

0AB
C0

be an exact sequence.
(1) Suppose that B is finitely presented. Then C is finitely presented iff A is
finitely generated.
(2) If C is finitely presented and B is finitely generated then A is finitely gen-
erated.
Let F be a finitely generated functor in (mod-R, Ab). There is, by 1.9, some
A mod-R and an epimorphism (A, ) F. If F is finitely presented then the

8
kernel of this epimorphism will be finitely generated so there will be B mod-R
and an exact sequence (B, ) (A, ) F 0. By the Yoneda lemma every
morphism from (B, ) to (A, ) is induced by a morphism f : A B. This
gives the following description.
Lemma 1.14. Let F (mod-R, Ab) be finitely presented. Then there is
a
morphism f : A B in mod-R such that F ' coker (f, ) : (B, ) (A, ) .
Conversely, any functor of this form, coker(f, ) for some f mod-R, is finitely
presented.
Of course, given F , there are many choices for f above.
Given f : A B in mod-R write Ff = coker(f, ). Then Ff M (M
Mod-R) has the following description: Ff M = (A, M )/im(f, M ) is the abelian
group, (A, M ), of morphisms from A to M factored by the subgroup consisting of
those which factor initially through f. The next result shows that each finitely
generated subfunctor of (A, ) is determined by the property of morphisms
factoring through some specified morphism with domain A.
Lemma 1.15. Suppose that A mod-R and that G is a subfunctor of (A, ).
Then G is finitely generated iff G is finitely presented iff G has the form im(f, )
for some f : A B in mod-R.
Proof. Any functor (B, ) with B mod-R is finitely generated by 1.8, so any
functor of the form im(f, ) is finitely generated. Conversely, if G is a finitely
generated subfunctor of (A, ) then it is the image of a representable functor
(B, ) G (by 1.9). Compose this with the inclusion G (A, ) to obtain a
morphism (B, ) (A, ) which, by Yoneda, 1.5, has the form (f, ) for some
f
A B. It follows that G = im(f, ).
It remains to show that any such functor is finitely presented. Retaining
f g
the notation, let g : B C be the cokernel of f, so A B C 0 is
exact (and C is finitely presented). By 1.12 we obtain an exact sequence of
(f,)

functors 0 (C, ) (B, ) (A, ). Therefore ker (B, ) G is
(C, ) (B, ) and hence G is finitely presented, by 1.13. 

The following strengthening of 1.15, the fact that every finitely presented
functor in (mod-R, Ab) is coherent, is a key property, local coherence, of the
functor category. An object is coherent if it is finitely presented and every
finitely generated subobject is finitely presented. A ring is right coherent if
every finitely presented right module is coherent (equivalently if every inter-
section of two finitely generated right ideals is finitely generated and the right
annihilator of every element is finitely generated).
Corollary 1.16. Every finitely generated subfunctor of a finitely presented func-
tor in (mod-R, Ab) is finitely presented. That is, (mod-R, Ab) is locally coher-
ent.
Proof. Suppose H G with G finitely presented and H finitely generated.
By 1.14 there is B mod-R and an epimorphism : (B, ) G. Since G
is finitely presented ker() is finitely generated (1.13) so, if H 0 = 1 H, then
H 0 is finitely generated, being an extension of a finitely generated by a finitely
generated object, hence is finitely presented by 1.15. By 1.13, H ' H 0 /ker()
is finitely presented. 

9
 Denote by (mod-R, Ab)fp the full subcategory of finitely presented functors
from mod-R to Ab. The above result implies that this subcategory is abelian.
Proposition 1.17. (mod-R, Ab)fp is an abelian subcategory of (mod-R, Ab):
it is an abelian category and a sequence 0 F G H 0 which is exact
in the smaller category is also exact in the larger category.
Proof. By [8, 3.41] (where the term exact subcategory is used) it is enough
to check that the kernel and cokernel of every morphism : G F in
(mod-R, Ab)fp is also in this subcategory.
The cokernel of is a finitely presented object factored by a finitely generated
one, hence is finitely presented (1.13).
The image of is a finitely generated hence, by 1.16, finitely presented,
subfunctor of F. Therefore (1.13) the kernel of is a finitely generated, therefore
finitely presented, subfunctor of G, as required. 

Corollary 1.18. [2, 2] Every finitely generated subfunctor of a representable

functor in (mod-R, Ab) has projective dimension less than or equal to 1. Every
finitely presented functor has projective dimension less than or equal to 2.
Proof. The first statement follows from the end of the proof of 1.15, along with
1.8. Also, since the representable functors form a generating set of finitely gener-
ated projective functors (1.9) the second statement follows; for every finitely pre-
sented functor F therefore has a projective presentation 0 (C, ) (B, )
(A, ) F 0. 

Corollary 1.19. A finitely presented functor in (mod-R, Ab) has projective

dimension 1 iff it embeds in a representable functor.
Proof. For the direction not covered by 1.18 suppose that the sequence 0
(f,)
(B, ) (A, ) F 0 induced by f : A B is exact. So (1.12)
f is an epimorphism. Then the sequence 0 ker(f ) A B 0 is
exact. Now, ker(f ) is a finitely generated (but not necessarily finitely presented)
R-module, so there is an epimorphism C ker(f ) with C mod-R. From
the exact sequence C A B 0 one obtains (1.12) the exact sequence
0 (B, ) (A, ) (C, ) and so F, being coker((B, ) (A, )), embeds
in (C, ), as required. 

The next result comes from [2, p. 205]; we write gldim(C) = n if n is the
maximum of projective dimensions of objects in the category C.
Proposition 1.20. A ring R is von Neumann regular iff gldim(mod-R, Ab)fp =
0. Otherwise gldim(mod-R, Ab)fp = 2.
Proof. It has to be shown that if gldim(mod-R, Ab)fp 1 then gldim(mod-R, Ab)fp =
0.
If f : A B is a morphism in mod-R then, by assumption, Ff has projective
dimension 1 so, by 1.19, Ff embeds in a representable functor, hence there
(f,) (g,)
is an exact sequence (B, ) (A, ) (D, ) for some g : D A
g f
in mod-R. It follows from 1.12 that D A B is exact. Therefore mod-R
has kernels, which implies that R is right coherent; for this shows that ker(f )

10
is finitely generated and, since im(f ) is a typical finitely generated subobject
of B, it follows that every finitely generated submodule of a finitely presented
module is finitely presented. Furthermore, if f : A B is monic, then g = 0,
hence (f, ) is epi and so f : A B is split. Thus every embedding between
finitely presented modules is split and this is one of the standard equivalents to
a ring being von Neumann regular.
If R is von Neumann regular let F = Ff , where f : A B is in mod-R, be
a typical finitely presented functor (1.14). Let A00 = im(f ). Since R, being von
Neumann regular, is coherent A00 is finitely presented so A0 = ker(f ) is finitely
generated. The embedding of A0 in A is, by von Neumann regularity, split. So
Ff ' (A0 , ) has projective dimension 0 (1.8), as required. 

Since regularity is a right/left symmetric condition one has the immediate

corollary.
Corollary 1.21. gldim(mod-R, Ab)fp = gldim(R-mod, Ab)fp for every ring
R.

2 Injective Functors
2.1 The tensor embedding
We can move the whole module category Mod-R into the functor category -
though now we mean functors on finitely presented left modules. Here is how.
To any right R-module, M , we associate the corresponding tensor functor,
(M ) = (M R ) : R-mod Ab, given by (M )(L) = M R L on
objects L R-mod and with the obvious effect on morphisms: if g : L K
R-mod then (M )g = M R g : M L M K.
Thus from M Mod-R we obtain (M ) (R-mod, Ab).
Define the functor  : Mod-R (R-mod, Ab) by M = (M ) on objects
f
and, if M
N is a morphism of right R-modules, then f : (M ) (N )
is the natural transformation whose component at L R-mod is defined to be
f 1L : M L N L, that is, ( R L)f .

Lemma 2.1. Suppose that F, F 0 (R-mod, Ab) with F right exact and let
, 0 : F F 0 be natural transformations. If R = R0 then = 0 .

Proof. Let L be a finitely presented left module, say (Rm )Rn
L 0 is
exact. There is a commutative diagram with the top row exact

F Rn
F / FL /0
(0 ) L 0
R n L
 
F 0 Rn / F 0L /0
F 0

and with vertical maps being Rn = (R )n = (R0 )n = R0 n and either L or L0 for

the other. It follows that L = L0 since L .F = F 0 .Rn = F 0 .R0 n = L0 .F
and F is an epimorphism. 

11
A right exact functor F : A B between abelian categories is one which
takes right exact sequences in A to right exact sequences in B. If F is,
instead, contravariant then we mean that it takes left exact sequences in
A to right exact sequences in B. An exact functor is one which takes
exact sequences to exact sequences.

The basic results about M 7 M are stated in [13] and some more detail
may be found in [14, 1] (also see the exposition in [17, B16]).
Theorem 2.2. Let R be a small preadditive category. The functor  : Mod-R
(R-mod, Ab) given on objects by M 7 M is a full embedding and is
left adjoint to the functor evaluation at R from (R-mod, Ab) to Mod-R:
(M , F ) ' (M, F (R)).

adjoint functors Suppose that F : C D and G : D C are functors.

Then F is left adjoint to G and G is right adjoint to F (and that
(F, G) is an adjoint pair) if there is a natural bijection D(F (), ) '
C(, G()), meaning that for every C C and D D there is a natural
bijection C,D : D(F C, D) C(C, GD). By natural we mean that for
all f : C C 0 in C and g : D D0 in D the obvious diagrams commute.

Proof. If (M ) ' (N ) then, evaluating at R R, we obtain M ' M R R '

N R R ' N. If : (M ) (N ) is a morphism then its component,
R , at R R is a morphism from M ' M R R to N ' N R R. The natural
transformations and R agree at R so, since M and N are right
exact, 2.1 yields = R . Therefore  is a full embedding.
For the adjointness, first note that if F (R-mod, Ab) then, since End(R R) =
R (acting on the right), F (R R) has the structure of a right R-module: if
a F (R R) and s R then set as = F ( s) a and note that a(st) =
F ( st) a = F (( t)( s)) a = (as)t. The natural isomorphism
(M , F ) ' (M, F R) takes (M , F ) to R . By 2.1 this map is
monic. To define the inverse map: given g : M F R, let g be the natural
transformation from M to F the component of which at L R-mod is
defined by taking m l M L to F l g(m) where F l denotes the value of F
at the morphism from R R to L which takes 1 to l. One may check that g is
a natural transformation, that (g )R = g and that these processes do define an
adjunction. 

Example 2.3. The functor  is not left exact: we show this for the embedding,
j, of the Z-module Z into the Z-module Q.
A morphism : F G in (Ab, Ab) is monic iff for every L Ab the
component L : F L GL is monic (1.3). Set L = Z2 (i.e. Z/2Z). Then
Z Z2 ' Z2 but Q Z2 = 0 because every element of Q Z2 is (a linear
combination of elements) of the form a12 , and a12 = 12 a.212 = 12 a2.12 =
0. Thus we see that j L : Z Z2 Q Z2 is not monic.
j p
A short exact sequence 0 A B C 0 is pure-exact if it satisfies
the equivalent conditions of the next result, in which case we say that j is a
pure monomorphism (and that p is a pure epimorphism).
j
Theorem 2.4. The following conditions on a short exact sequence 0 A

p
B C 0 of right R-modules are equivalent.

12
(i) For every (finitely presented) left R-module L the morphism j 1L : AL
B L is monic.
(ii) The sequence is a direct limit of split-exact sequences.

Corollary 2.5. [20, 2.4] If 0 A B C 0 is pure-exact and C is
finitely presented then this sequence is split.
In particular any pure embedding in mod-R splits.

Theorem 2.6. A sequence 0 M N N 0 0 of right R-modules is

pure-exact iff the image 0 M N N 0 0 is an exact sequence of
functors. Furthermore  commutes with direct limits and products.
Proof. The first assertion is from the definition/2.4.
For every L R-mod the functor L, being a left adjoint, commutes with
direct limits and, from this, the first part of the second statement follows easily.
The assertion about products follows directly from 2.7 below. 

Proposition 2.7.Q [21, Satze

Q 1,2] A module M is finitely presented iff the canon-
ical map M ( i Ni ) i (M Ni ) is an isomorphism for all collections of
modules (Ni )i (and M is finitely generated iff each such map is a surjection).
The next result is immediate from the fact that  is full and faithful, hence
preserves and reflects idempotents (that is, projections to direct summands).
Corollary 2.8. A module M is indecomposable iff the corresponding functor
(M ) (R-mod, Ab) is indecomposable.

2.2 Injective functors and pure-injective modules

A module M is absolutely pure if every embedding M N in Mod-R with
domain M is pure. A module M is fp-injective if for every embedding i : A
B with finitely presented cokernel and every morphism f : A M there is
g : B M such that gi = f .
Proposition 2.9. For any module M the following are equivalent:
(i) M is absolutely pure;
(ii) M is fp-injective;
(iii) Ext1 (C, M ) = 0 for every finitely presented module C.
Clearly every injective module is absolutely pure. Over a right noetherian
ring absolutely pure = injective.

Proposition 2.10. A module is absolutely pure iff it is a pure submodule of an

injective module.
Proof. One direction is by the definition (and existence of injective hulls). For
the other suppose that i : M E is a pure embedding and that E is injective.
Let j : M N be any embedding. Since E is injective there is g : N E such
that gj = i. Since i is pure it follows easily that so is j, as required. 

An R-module N is pure-injective if, given any pure embedding f : A B

in Mod-R, every morphism g : A N lifts through f : there exists h : B N

13
such that hf = g.
 f
/B
A pure

g
 ~ h
N
Equivalently, N is pure-injective iff every pure embedding N M with domain
N is split.
Any direct product of pure-injective modules is pure-injective, as is any di-
rect summand of a pure-injective module. Any injective module is pure-injective
because such a module has the lifting property over all embeddings, pure or not.
A module is of finite endolength if, when regarded as a module over its
endomorphism ring, it has finite length.
Corollary 2.11. Every module of finite endolength is pure-injective.
In fact every module M of finite endolength has the stronger property of
being -pure-injective, meaning that every infinite direct sum of copies of M
is pure-injective.
Duals of modules are pure-injective.
Proposition 2.12. If S MR is an (S, R)-bimodule and S E is an injective left
S-module then M = HomS (S MR , S E) is a pure-injective left R-module.
In particular if M is a right R-module then HomZ (M, Q/Z) is a pure-
injective left R-module.
Example 2.13. [39, Thm. 1] The group ring RG is pure-injective as a right
module over itself iff the same is true for R and G is finite.
The following criterion for pure-injectivity is due to Jensen and Lenzing.
Theorem 2.14. [17, 7.1] A module M is pure-injective P iff for every index set,
I, the summation map : M (I) M , given by (xi )i 7 i xi , factors through
the natural (pure) embedding of M (I) into the corresponding direct product M I .
Theorem 2.15. ([13], [14, 1]) An exact sequence 0 M N N 0 0 in
Mod-R is pure exact iff the sequence 0 M N N 0 0 is a pure exact
sequence in (R-mod, Ab).
If M is a right R-module then M = (M ) is an absolutely pure object of
(R-mod, Ab), indeed every absolutely pure functor is isomorphic to one of this
form.
Furthermore, (M ) is injective iff M is pure-injective.
Proof. The given pure exact sequence is, 2.4, a direct limit of split exact
sequences. Since  commutes with direct limits (2.2) the image sequence is a
direct limit of split exact sequences, hence is pure exact. By 2.2 we also have
the converse.
Next we show that Q (R-mod, Ab) is absolutely pure [the definition given
for modules applies equally to functors] iff Q is a right exact functor.
Let F (R-mod, Ab)fp have projective presentation 0 (C, ) (B, )
(A, ) F 0 where A B C 0 is an exact sequence in R-mod
(see the proof of 1.18). Then the homology groups of the chain complex 0
((A, ), Q) ((B, ), Q) ((C, ), Q) 0, that is (by Yoneda) 0 QA

14
QB QC 0, are, by definition, precisely (F, Q), Ext1 (F, Q) and Ext2 (F, Q).
Therefore Q is right exact iff for all finitely presented functors F we have
Ext1 (F, Q) = 0 = Ext2 (F, Q). We have Ext2 (F, ) ' Ext1 (F 0 , ) where, with
the notation above, F 0 is the kernel of (A, ) F . Since (1.16) (R-mod, Ab)
is locally coherent F 0 also is finitely presented. So the condition on Q reduces
to Ext1 (F, Q) = 0 for all finitely presented F , that is (2.9), Q is an absolutely
pure functor.
It follows that every functor of the form M is absolutely pure.
For the converse, suppose that Q is absolutely pure hence, as shown above,
right exact. Note that Q(R R) is a right (End(R R) ')R-module. Define the
natural transformation (Q(R) ) Q to have component at L R-mod
the map Q(R) L QL defined by taking m l (m Q(R), l L) to
Q(l : R R L) m (a special case of that in the proof of adjointness in 2.2).
It is straightforward to check that this is, indeed, a natural transformation and
that at L = R it is an isomorphism. So, by 2.1, this is an isomorphism from Q
to (Q(R) ), as required.
Given M Mod-R there is the pure-exact sequence 0 M H(M )
H(M )/M 0, where H(M ) is the pure-injective hull (see below) of M , hence,
by 2.6, the sequence of functors 0 (M ) (H(M ) ) (H(M )/M
) 0 is pure-exact. If (M ) is injective then this sequence is split and
so, therefore (by 2.2,  is full), is the first, whence M is a direct summand of
H(M ), hence is pure-injective (and equal to H(M )). For the converse, take
an injective hull, E(M ), of M . Since any injective is absolutely pure,
E(M ) ' (N ) for some N Mod-R. So there is an exact sequence
0 (M ) (N ) (N )/(M ) ' (N/M ) 0 (the
isomorphism by right exactness of ) and, therefore, by the first part, there is
the pure-exact sequence 0 M N N/M 0 in Mod-R. If M is pure-
injective this sequence is split, so the sequence of functors is split and, therefore,
M is already injective. 

We record the following point which was established in the course of the
proof above.
Proposition 2.16. An object of (R-mod, Ab) is absolutely pure iff it is a right
exact functor.
Corollary 2.17. There is a bijection between isomorphism classes of indecom-
posable pure-injective right R-modules, N , and isomorphism classes of inde-
composable injective objects, Q, in the functor category (R-mod, Ab), given by
N 7 (N ) and Q 7 Q(R R).
A pure-injective hull for a module M is a pure embedding M N with N
pure-injective and N minimal such, in the sense that there is no factorisation of
this map through any direct summand of N. Existence and uniqueness of pure-
injective hulls is most easily obtained by using  to pull back the corresponding
results for injective functors to pure-injective modules. We denote the pure-
injective hull of M by H(M ).
Corollary 2.18. The embedding M N is a pure-injective hull in Mod-R
iff (M ) (N ) is an injective hull in (R-mod, Ab): E(M ) '
(H(M ) ).

15
Corollary 2.19. Every module M has a pure-injective hull which is unique to
isomorphism over M : if j : M N and j 0 : M N 0 are pure-injective hulls
of M then there is an isomorphism f : N N 0 such that f j = j 0 .

Corollary 2.20. Every indecomposable pure-injective has local endomorphism

ring.
Proof. This follows by using  to pull back the corresponding fact for indecom-
posable injective objects in Grothendieck categories. 

2.3 Duality of finitely presented functors

The two functor categories, (R-mod, Ab)fp and (mod-R, Ab)fp , are dual. The
explicit definition of the duality comes from [3, 7] and [14, 5.6] and is as follows.
Let F (R-mod, Ab)fp . Define the functor dF (mod-R, Ab) by dF A =
(F, A ) for A mod-R; it is the case, 2.21 below, that A is a finitely
presented functor if A is a finitely presented module. If g : A B is in mod-R,
then dF g : (F, A ) (F, B ) is defined to be (F, g ), so is given by
dF g = (g ). That is, dF , the dual of F , is given by the representable
functor (F, ) restricted to the image of the embedding, (2.2) A 7 A , of
mod-R into (R-mod, Ab)fp . The fact that dF is a finitely presented functor is
2.24.
Theorem 2.21. ([2, 6.1]) Let M Mod-R. The functor (M R ) : R-mod
Ab is finitely presented iff M is finitely presented.
The definition of d on morphisms is as follows: given f : F G in
(R-mod, Ab)fp define df : dG dF to be (f, ), that is, if M mod-R then
the component of df at M is given by taking (G, M ) to f (F, M ).
Dually, if F (mod-R, Ab)fp , then we use the same notation for the functor
dF (R-mod, Ab) given on objects by dF L = (F, L).
Example 2.22. If A mod-R and L R-mod then, d(L, ) '
( L) and
d(A ) ' (A, ). For the first, d(L, ) B = (L, ), B which (1.5) is
naturally isomorphic to B L = ( L)B and, for the second, d(A ) B =
(A , B ) which is naturally isomorphic to (A, B) = (A, )B.

Lemma 2.23. d is a contravariant exact functor from (R-mod, Ab)fp to (mod-R, Ab)fp .
Proof. Suppose that 0 H F G 0 is an exact sequence in
(R-mod, Ab)fp . For any A mod-R the sequence 0 (G, A ) (F, A
) (H, A ) Ext1 (G, A ) is exact. By 2.15, (A ) is an absolutely
pure functor hence (see 2.9), since G is finitely presented, Ext1 (G, A ) = 0,
so d is indeed exact. That dF is finitely presented is shown next. 

Proposition 2.24. If F (R-mod, Ab)fp then dF (mod-R, Ab)fp .

(f,)
Proof. Suppose that (L, ) (K, ) F 0 is a projective presentation
f
of F with K L R-mod (1.14). Apply d to obtain the, exact by 2.23,
sequence 0 dF d(K, ) d(L, ) that is, by 2.22, 0 dF ( K)
( L). Since K, L are finitely presented so, by 2.21, are the functors K

16
and L so the image of K, being finitely generated, is finitely presented
(1.16). Therefore (1.13) the kernel, dF, of this map is finitely generated, hence
(1.16 again) finitely presented, as required. 

Theorem 2.25. The functor

d : ((R-mod, Ab)fp )op (mod-R, Ab)fp

is an equivalence of categories. If we use d also to denote the corresponding func-

tor from ((mod-R, Ab)fp )op to (R-mod, Ab)fp then d2 is naturally equivalent to
the identity functor on (R-mod, Ab)fp .
Proof. Applying d to a presentation (L, ) (K, ) F 0 of F
(R-mod, Ab)fp gives the exact sequence 0 dF ( K) ( L) in
(mod-R, Ab)fp , by the proof of 2.24. Applying d again gives, by 2.23 and 2.22,
the exact sequence (L, ) (K, ) d2 F 0. From this and what has been
shown already the assertions follow. 

Proposition 2.26. If M is any right R-module and if F is any functor in

(mod-R, Ab)fp then there is a natural isomorphism (dF, M ) ' F M (more



accurately, F M where F denotes the unique extension of F to a functor on all
of Mod-R which commutes with direct limits - this uses that every module is a
direct limit of finitely presented modules).
Proof. For M finitely presented we have this, by 2.25 and definition of d,
since F M ' d.dF.M ' (dF, M ). For arbitrary modules we can argue as
follows. Suppose that (B, ) (A, ) F 0 is a projective presentation of
F in (mod-R, Ab)fp . By 2.23 and 2.22 this gives the exact sequence 0 dF
(A ) (B ) in (R-mod, Ab)fp . In turn this gives the exact sequence
(B , M ) (A , M ) (dF, M ) 0 (exactness at the last
place because, 2.15, M is absolutely pure), that is, (B, M ) (A, M )
(dF, M ) 0. Since (B, M ) (A, M ) F M 0 also is exact we deduce
(dF, M ) ' F M . 

2.4 Injectives and projectives in the category of finitely

presented functors
We can see from the first lecture that the category of finitely presented func-
tors has enough projectives (since the whole category has a generating set of
projectives which are themselves finitely presented). So, by the duality 2.25, we
deduce that there are enough injectives, meaning that every functor embeds
into a functor which, as an object of (R-mod, Ab)fp , is injective. We identify
these functors.
Proposition 2.27. [14, 5.5] The category (R-mod, Ab)fp of finitely presented
functors has enough injectives; the injective objects are the functors (isomorphic
to one) of the form A with A mod-R.
Proof. Any functor of the form A R with A mod-R is finitely presented
(2.21) and is, by 2.15, absolutely pure in (R-mod, Ab), hence is injective in

17
(R-mod, Ab)fp ; for every pure embedding in the latter category is split since it
has finitely presented cokernel - see 2.5.
Conversely, if G (R-mod, Ab)fp is injective in this category then Ext1 (F, G) =
0 for all F (R-mod, Ab)fp , that is, by 2.9, G is absolutely pure in (R-mod, Ab)
so, by 2.15, G is isomorphic to a functor of the form A with A Mod-R.
By 2.21, A mod-R.
That proves the second statement: to see the first, let F (R-mod, Ab)fp .
There is, by 1.9, an epimorphism (A, ) dF for some A mod-R. This
dualises (2.25) to an embedding F d(A, ) ' (AR ) (by 2.22), as required.


In general (R-mod, Ab)fp does not have injective hulls (i.e. minimal injec-
tive extensions) equivalently, by 2.25, (mod-R, Ab)fp does not have projective
covers. A ring R is Krull-Schmidt if every finitely presented right R-module
is a direct sum of indecomposable modules with local endomorphism rings. For
instance artin algebras are Krull-Schmidt. As defined, this is right Krull-
Schmidt but, in fact, the notion is right/left symmetric.
Proposition 2.28. Let R be a ring. Then
(i) the category (R-mod, Ab)fp has injective hulls iff
(ii) the category (R-mod, Ab)fp has projective covers iff
(iii) R is Krull-Schmidt.

3 Simple Functors
3.1 Simple functors and indecomposable modules
Every simple functor in the category (R-mod, Ab) has, since it is finitely gener-
ated and by 1.9, the form S = (A, )/G where A mod-R and G is a maximal
subfunctor of (A, ). There are two cases: if G is a finitely generated functor
then the quotient S is a finitely presented functor and has a projective presenta-
(f,) f
tion of the form (B, ) (A, ) S 0 where A B is a morphism in
mod-R; otherwise S, though finitely generated, does not lie in (R-mod, Ab)fp .
In the case that S is finitely presented the dual, dS (mod-R, Ab)fp , is, by
2.25, a finitely presented simple functor.
In both cases, by 2.17, the injective hull of S in the functor category (R-mod, Ab)
has the form (N ) for a unique indecomposable pure-injective module NR .
Assume that R is Krull-Schmidt. Let S (R-mod, Ab)fp be a finitely
presented simple functor and let NR be the indecomposable pure-injective such
that (N ) is its injective hull, that is such that (S, N ) 6= 0. Consider the
dual dS (mod-R, Ab)fp of S and let (A, ), where A is an indecomposable
finitely presented right module, be its projective cover; so we have an exact
sequence 0 F (A, ) dS 0 of finitely presented functors (F is finitely
presented by coherence of the functor category and 1.13). Dualise this sequence
to obtain, using 2.22, the exact sequence 0 ddS ' S (A ) dF 0.
If we assume now that A is a pure-injective module (which, by 2.11, will be the
case if R is an artin algebra), then we have (N ) ' (A ), hence A ' N
(since there is just one indecomposable injective to which S has a nonzero map).
This gives the following.

18
Theorem 3.1. Suppose that R is an artin algebra. Then there are natural bi-
jections between indecomposable right R-modules A of finite length and simple
finitely presented functors S on mod-R (respectively, on R-mod). This is given
by S 7 A where (A, ) is the projective cover of S (resp. A is the injective
hull of S) and, in the other direction, A 7 (A, )/J(A, ) where J(A, ) de-
notes the unique maximal proper subfunctor of (A, ) (resp. by A 7 soc(A),
the unique minimal nonzero subfunctor of A ).
Regarding the detailed part of the statement, one can check that the functor
F which is defined by taking B mod-R to the set of all non-isomorphisms
from A to B is the unique maximal proper subfunctor of (A, ).
In fact, over an artin algebra every simple functor is finitely presented and
this is a starting point for Auslander-Reiten theory.

3.2 An example
Very occasionally we can get a complete picture of the functor category.
Let K be a field and let R = K[] = K[X]/hX 2 i. Then RR is indecompos-
able, projective and injective, with top and socle both isomorphic to the unique
simple module U . It follows that every module has the form R(I) U (J) for
some I, J. Thus R is of finite representation type, which means that every
R-module is a direct sum of indecomposable modules and there are, up to iso-
morphism, only finitely many indecomposable modules. The Auslander-Reiten
quiver of R is easily computed (for Auslander-Reiten theory see, e.g., [4]). The
vertices are R and U and it has arrows the embedding i : U R of rad(R) into
R and the epimorphism : R U of R to R/rad(R). Note that i = 0 and
i =  (i.e. multiplication by ).
The Auslander algebra of R is the endomorphism ring S = End(M ) where
M is a direct sum of one copy of each of the indecomposable (finitely presented)
R-modules, which  in this case is S= End(R
 U ). We may represent
 S as
(R, R) (U, R) R = K1R K Ki
the matrix ring ' and this
(R, U ) (U, U ) K K1U
decomposes as a left module as, say, Q1 Q2 where Qi is the i-th column. Set
T1   1 ). One can check that Q2 ' rad(Q1 ). Right multiplication
= Q1 /rad(Q
0 0
by gives an epimorphism Q1 rad(Q2 ) so, noting the lengths of
0
these modules, we conclude that rad(Q2 ) ' T1 . Let T2 denote the other simple
module (the top of Q2 ).
Thus far we have four indecomposable modules - the two indecomposable
projectives and the two simples. One can also see I2 - the injective hull of
T2 , with socle T2 and I2 /T2 ' T1 . It is not difficult to check that there are
no more indecomposable modules so, since S is an artin algebra, S is of finite
representation type (this is not always the case for the Auslander algebra of
a finite representation type algebra). We can compute the Auslander-Reiten
quiver for left S-modules to be as shown, where the 1st (respectively 2nd) and
5th (respectively 6th) columns should be identified (and dotted lines indicate
Auslander-Reiten translates).

19
 Q; 1 = I1 > Q1

#
Q2 ; I2 > Q2

# 
T2 T1 T2
Now we use the fact, see e.g. [43, 4.9.4], that the functor category (mod-R, Ab)
is equivalent to the category of left modules over the Auslander algebra1 : S-Mod '
(mod-R, Ab). We deduce that, in our example, there are just five indecompos-
able functors. To identify these we can use the explicit description, see [43,
p. 121], of this equivalence, which takes a functor F to F M regarded as a left
S-module and, in the other direction, a finitely presented left S-module S A is
sent to the functor HomS (S HomR (, S MR ), S A). In our example the functors
are easily identified directly:
Q1 is visibly the representable functor (RR , ), which is also ( R R), it is
also the injective hull of T1 and the projective cover of T1 ;
Q2 is visibly the representable functor (U, ) and is the projective cover of T2 ;
I2 is the injective hull of T2 and is also ( R U );
To illustrate the localisation process described in the next section, con-
sider the Serre subcategory, hT2 i, generated by T2 . In the quotient category
S-mod/hT2 i, since T2 becomes isomorphic to 0, all of Q2 , T1 and I2 become
isomorphic and we see that there are just two indecomposables, a simple and
a non-split self-extension of that simple - which looks like, and can checked to
be, (equivalent to) mod-R (in general, however, a localisation of a module or
functor category need not be equivalent to a module or functor category).

3.3 Serre subcategories and localisation

Suppose that B is an abelian category and that S is a Serre subcategory,
that is, a subcategory of B such that whenever 0 A B C 0 is an
exact sequence in B then B S iff A, C S. Then there is a localisation
functor from B to a category of fractions which is obtained by inverting those
morphisms of B whose kernels and cokernels are in S (see, e.g., [23, 4.3] for
details). This category is denoted B/S and referred to as the quotient of B by
S. As stated in the next result, this is again an abelian category, so this is not
the same as factoring by an ideal of S (the process used to obtain the stable
module category).
Theorem 3.2. [23, pp. 172-174] Suppose that B is an abelian category and
that S is a Serre subcategory of B. The quotient ategory B/S is abelian and the
localisation functor F : B B/S is exact. The kernel of F in the sense of
ker(F ) = {B B : F B = 0} is S. Every exact functor G : B B 0 from B to
an abelian category B 0 with ker(G) S factors uniquely through F : B B/S
via an exact and faithful functor.
1 All that is going on here is that mod-R is the additive closure of, hence has the same

category of representations as, the preadditive category of indecomposables which, having

only finitely many objects, is equivalent to a ring. We could make the same definition in the
general case but wed get a ring with many objects rather than an actual ring.

20
 If C is a locally coherent Grothendieck abelian category, in particular if C


is a functor category, then the closure under direct limits, S , of any Serre
subcategory S of C fp is a hereditary torsion class in C, that is, a class closed
under subobjects, factor objects, extensions and arbitrary direct sums. An
object is defined to be torsionfree if it contains no non-zero torsion subobject.
The associated (hereditary) torsion functor, : C C assigns to every object
its largest torsion subobject.


We may form the localisation (in the sense of [10], see [38]) of C at S . As


above, this is done by forcing every object of S to become 0 in the new, quotient,
abelian category, but we also require the localisation functor to commute with
direct limits.
There are a couple of ways of constructing this functor. The first method we
describe is most naturally presented as a functor from C to a full subcategory.
Take an object C C. First we make its torsion subobject zero, by forming
C/ (C). Then we embed this into its injective hull E(C/ (C)) and take the
maximal extension of C

1 = C/ (C) within E(C1 ) by a torsion object, namely
C = 1 E(C1 )/C1 where : E(C1 ) E(C1 )/C1 is the natural projec-
tion. Then C is -injective, meaning that it has no non-split extension by a
torsion object. This is the action of the localisation functor, which we denote
Q , on objects; there is a naturally induced action on morphisms. We denote
by C the image of this functor; it is abelian and a full, but not exact=abelian,
subcategory of C. This category is referred to as the localisation (or quotient
category) of C at .
Theorem 3.3. (see [38, IX.1, X.1], [23, 4.3.8, 4.3.11, 4.4, 4.6.2]) Let
be a hereditary torsion functor on a Grothendieck abelian category C. Then the
localised category C also is Grothendieck abelian, the localisation functor Q :
C C is exact and, if F : C C 0 is any exact functor to a Grothendieck
category C 0 such that F commutes with direct limits and ker(F ) T - the
torsion class corresponding to - then F factors uniquely through Q .
As said already, the localisation functor Q : C C has a right adjoint,
namely the inclusion, i, of C in C: C(C, iD) ' C (C , D) for every C C and
D C . The image of i is, up to natural equivalence, the full subcategory of
-torsionfree, -injective objects of C.
For any -torsionfree, -injective object C, one has Q C ' C and the injec-
tive objects of (i)C are exactly the -torsionfree injective objects of C.
If G is a generating set for C then Q G is a generating set for C .
The inclusion functor i : C C is not in general right exact: if 0
D0 D D00 0 is an exact sequence in C then, regarded as a sequence in
C, the quotient D/D0 , though certainly -torsionfree, need not be -injective,
so the corresponding exact sequence in C would replace D00 by the, perhaps
proper, submodule D/D0 (of which D00 would be the -injective hull). Under
the alternative construction, outlined below, the functor Q can be seen as
sheafification with respect to a certain Grothendieck-type topology and then
this corresponds to the fact that the inclusion of the category of sheaves in the
category of presheaves is not right exact. Also the inclusion functor does not,
for instance, commute with infinite direct sums: in general, the direct sum of
objects in C , regarded as embedded in C, is obtained by taking their direct sum
in C and then taking the -injective hull of that.

21
 The alternative construction is essentially a sheafification process, and can
be found in [1], or see [38, IX.1]. In this construction no change is made in
the objects but the morphisms of the new category are defined by C (C, D) =
limC 0 limD (C 0 , D/D0 ) where C 0 ranges over subobjects of C such that C/C 0 is
0
torsion (these are directed by intersection) and D0 ranges over -torsion sub-
objects of D (directed by sum). In this category, an enlargement of C by mor-
phisms, each object C is isomorphic to the corresponding object C obtained
by the first construction.
Theorem 3.4. (see [23, 4.4.9]) Suppose that C and C 0 are abelian categories,
that C is Grothendieck and that Q : C C 0 is an exact functor. Suppose also
that Q has a full and faithful right adjoint. Then ker(Q) = {C C : QC = 0} is
a hereditary torsion class in C and the right adjoint of Q induces an equivalence
between C 0 and the corresponding localisation of C.
Example 3.5. Take C = Ab and take torsion to mean 2-torsion by declaring
an abelian group M to be torsion if each element of M is annihilated by a power
of 2.
Then the objects of the quotient category are the abelian groups D such that
D has no 2-torsion and such that D has no non-trivial extension by a 2-torsion
module.
To get from an arbitrary abelian group, C, to the corresponding object of
the quotient category, first factor out the 2-torsion subgroup, C, the subgroup
consisting of all elements of C which are annihilated by some power of 2; set C1 =
C/ C - a 2-torsionfree group. Let E(C1 ) denote the injective hull of C1 (this
will be a direct sum of (p 6= 2)-Prufer groups and copies of Q) and consider the
factor group E(C1 )/C1 ; consider the 2-torsion subgroup (E(C1 )/C1 ) (clearly
there will be no contribution from the Prufer components) and let C2 be its full
inverse image in E(C1 ) - so C1 C2 E(C1 ) and C2 /C1 = (E(C1 )/C1 ) (thus
C2 is obtained by making C1 fully divisible by powers of 2). The localisation
functor takes C to C2 . For instance, if C = Z then C2 = Z[ 21 ]. Indeed this
localisation functor is easily seen to be equivalent to tensoring with Z[ 12 ].
Say that (or the corresponding torsion class) is of finite type if the torsion
class is generated as such by the finitely presented torsion objects. For example,


if C is locally coherent and S is a Serre subcategory of C fp then S is a torsion
class of finite type.
Corollary 3.6. If C is a locally coherent abelian category and is of finite type
then C is locally coherent and has, for its class (up to isomorphism) of finitely
presented objects, the localisations of finitely presented objects of C: (C )fp =
(C fp ) = {C : C C fp }.

3.4 Krull-Gabriel dimension

Let C be a locally coherent abelian (hence, [7, 2.4], Grothendieck) category; so
the subcategory C fp of finitely presented=coherent objects is abelian.
Let S be the Serre subcategory of C fp generated by all the finitely presented
simple functors2 . This consists of all finitely presented objects of C of finite
2 If we were to use all the simple functors then we would obtain Gabriel dimension [12].

22


length. Then S is a finite type torsion class in C and we may form the locali-


sation C1 = C/ S . By 3.6, C1 is locally coherent, so we can repeat the process,
with C1 in place of C, and obtain C2 . Etc. Transfinitely. We obtain a sequence
of locally coherent categories C indexed by ordinals. By the last statement in
3.6 this sequence eventually must stabilise, either with the trivial category, or
with a nontrivial category with no finitely presented simple object. In the first
case, if is the smallest ordinal such that C = 0 then we say that 1 (or if
this is a limit ordinal) is the Krull-Gabriel dimension of C, KGdim(C) = .
In the second case the Krull-Gabriel dimension of C is undefined and we write
KGdim(C) = . If C is the functor category (mod-R, Ab) where R is a ring
or small preadditive category then we write KG(R) for KGdim(mod-R, Ab) (
above cannot be a limit ordinal in this case).
Since the duality of Section 2.3 takes finitely presented simple functors to
finitely presented simple functors, KG(R) is right/left symmetric.



Corollary 3.7. KGdim (mod-R, Ab) = KGdim (R-mod, Ab) .
That there is some connection between this dimension and representation
type of artin algebras can be seen in the following results.

Dimension 0 KG(R) = 0 iff R is of finite representation type.

Dimension 1 This value is not attained by any artin algebra ([16, 3.6], [19,
11.4]).

Dimension 2 Algebras with Krull-Gabriel dimension 2 include the tame hered-

itary algebras ([11], see also [25] and [33]).

Dimension defined and > 2 The domestic string algebras n (see [6, 2.3,
4], [35, Thm. 1]) have KG(n ) = n + 1. It is conjectured that any domestic
string algebra has finite Krull-Gabriel dimension.

Dimension Wild finite-dimensional algebras have Krull-Gabriel dimension

(see [24, pp. 281-2] or [18, 8.15]).

Hereditary artin algebras In particular for hereditary artin algebras the

possible representation types - finite, tame/domestic, wild - correspond to the
values 0, 2, for Krull-Gabriel dimension.

String algebras/canonical algebras There are examples of domestic string

algebras with any finite Krull-Gabriel dimension 2. Exactly what values of
Krull-Gabriel dimension can occur for domestic (string) algebras is a subject
of conjecture (see [29], also [27], [37]). On the other hand, non-domestic string
algebras have Krull-Gabriel dimension (see [26], [36], and also [30]). Tubular
algebras have Krull-Gabriel dimension (see [26], [15]).

23
Some group rings Puninski, Puninskaya and Toffalori [31] showed that the
integral group ring of a nontrivial finite group has Krull-Gabriel dimension .
For group rings KG over a field they determined [32, 4.11] the value of Krull-
Gabriel dimension (which turns out to be 0, 2 or ), except in the case that
K has characteristic 2, does not contain a primitive cube root of 1 and G has a
generalised quaternion group for a Sylow 2-subgroup.

References
[1] M. Artin, A. Grothendieck and J.L. Verdier, Theorie des Topos et Co-
homologie Etale des Schemas (SGA4), Lecture Notes in Mathematics,
Vol. 269, Springer-Verlag, 1972.
[2] M. Auslander, Coherent functors, pp. 189-231 in Proceedings of the
Conference on Categorical Algebra, La Jolla, 1965, Springer-Verlag,
1966.
[3] M. Auslander, Isolated singularities and existence of almost split se-
quences, (Notes by Louise Unger pp. 194-242 in Representation Theory
II, Groups and Orders, Ottawa 1984, Lecture Notes in Mathematics,
Vol. 1178, Springer-Verlag, 1986.
[4] M. Auslander, I. Reiten, S. Smal, Representation Theory of Artin Alge-
bras, Cambridge Studies in Advanced Mathematics, Vol. 36, Cambridge
University Press, 1995.
[5] D. J. Benson, Representations and Cohomology I: Basic representation
theory of finite groups and associative algebras, Cambridge Studies in
Advanced Math., Vol. 30, Cambridge University Press, 1991.
[6] K. Burke and M. Prest, The Ziegler and Zariski spectra of some domes-
tic string algebras, Algebras and Representation Theory, 5(3) (2002),
211-234.

[7] W. Crawley-Boevey, Locally finitely presented additive categories,

Comm. Algebra, 22(5) (1994), 1641-1674.
[8] P. Freyd, Abelian Categories, Harper and Row, 1964.
[9] P. Freyd, Representations in abelian categories, pp. 95-120 in Proceed-
ings of the Conference on Categorical Algebra, La Jolla, 1965, Springer-
Verlag, 1966.
[10] P. Gabriel, Des categories abeliennes, Bull. Soc. Math. France, 90
(1962), 323-448.

[11] W. Geigle, The Krull-Gabriel dimension of the representation theory

of a tame hereditary artin algebra and applications to the structure of
exact sequences, Manus. Math., 54(1-2) (1985), 83-106.
[12] R. Gordon and J. C. Robson, The Gabriel dimension of a module, J.
Algebra, 29(3) (1974), 459-473.

24
[13] L. Gruson and C. U. Jensen, Modules algebriquement compact et fonc-
teurs lim(i) , C. R. Acad. Sci. Paris, 276 (1973), 1651-1653.

[14] L. Gruson and C. U. Jensen, Dimensions cohomologiques reliees aux
foncteurs lim(i) , pp. 243-294 in Lecture Notes in Mathematics, Vol. 867,

Springer-Verlag, 1981.

[15] R. Harland, Pure-injective Modules over Tubular Algebras and String

Algebras, Doctoral Thesis, University of Manchester, 2011, available at
www.maths.manchester.ac.uk/mprest/publications.html
[16] I. Herzog, The endomorphism ring of a localised coherent functor, J.
Algebra, 191(1) (1997), 416-426.

[17] C. U. Jensen and H. Lenzing, Model Theoretic Algebra; with particular

emphasis on Fields, Rings and Modules, Gordon and Breach, 1989.
[18] H. Krause, The Spectrum of a Module Category, Habilitationsschrift,
Universitat Bielefeld, 1997, published as Mem. Amer. Math. Soc., No.
707, 2001.

[19] H. Krause, Generic modules over artin algebras, Proc. London Math.
Soc. (3), 76(2) (1998), 276-306.
[20] D. Lazard, Autour de la platitude, Bull. Soc. Math. France, 97 (1969),
81-128.

[21] H. Lenzing, Endlich prasentierbare Moduln, Arch. Math., 20(3) (1969),

262-266.
[22] J. C. McConnell and J. C. Robson, Noncommutative Noetherian Rings,
John Wiley and Sons, 1987.
[23] N. Popescu, Abelian Categories with Applications to Rings and Mod-
ules, Academic Press, 1973.
[24] M. Prest, Model Theory and Modules, London Math. Soc. Lect. Note
Ser., Vol. 130, Cambridge University Press, 1988.
[25] M. Prest, Ziegler spectra of tame hereditary algebras, J. Algebra, 207(1)
(1998), 146-164.
[26] M. Prest, Morphisms between finitely presented modules and infinite-
dimensional representations, pp. 447-455 in Canad. Math. Soc. Conf.
Proc., Vol. 24 (1998).
[27] M. Prest, Topological and geometric aspects of the Ziegler spectrum, pp.
369-392 in H. Krause and C. M. Ringel (Eds.), Infinite Length Modules,
Birkhauser, 2000.
[28] M. Prest, Purity, Spectra and Localisation, Encyclopedia of Mathemat-
ics and its Applications, Vol. 121, Cambridge University Press, 2009.

[29] M. Prest and J. Schroer, Serial functors, Jacobson radical and repre-
sentation type, J. Pure Appl. Algebra, 170(2-3) (2002), 295-307.

25
 [30] G. Puninski, Superdecomposable pure-injective modules exist over some
string algebras, Proc. Amer. Math. Soc., 132(7) (2004), 1891-1898.

[31] G. Puninski, V. Puninskaya and C. Toffalori, Superdecomposable pure-

injective modules and integral group rings, J. London Math. Soc. (2),
73(1) (2006), 48-64.
[32] G. Puninski, V. Puninskaya, and C. Toffalori, Krull-Gabriel dimension
and the model-theoretic complexity of the category of modules over
group rings of finite groups, J. London Math. Soc., 78 (2008), 125-142.
[33] C. M. Ringel, The Ziegler spectrum of a tame hereditary algebra, Col-
loq. Math., 76 (1998), 105-115.
[34] J.-E. Roos, Locally Noetherian categories and generalised strictly lin-
early compact rings: Applications, pp. 197-277 in Category Theory,
Homology Theory and their Applications, Lecture Notes in Mathemat-
ics, Vol. 92, Springer-Verlag, 1969.
[35] J. Schroer, On the Krull-Gabriel dimension of an algebra, Math. Z.
233(2) (2000), 287-303.

[36] J. Schroer, On the infinite radical of a module category, Proc. London

Math. Soc.(3), 81(3) (2000), 651-674.
[37] J. Schroer, The Krull-Gabriel dimension of an algebra - open problems
and conjectures, pp. 419-424 in H. Krause and C. M. Ringel (Eds.),
Infinite Length Modules, Birkhauser, 2000.

[38] B. Stenstrom, Rings of Quotients, Springer-Verlag, 1975.

[39] W. Zimmermann, (-)algebraic compactness of rings, J. Pure Appl.
Algebra, 23(3) (1982), 319-328.

Additional references

[40] M. Auslander, Functors and objects determined by objects, pp. 1-244

in R. Gordon (Ed.), Representation Theory of Algebras, Philadelphia
1976, Marcel Dekker, New York, 1978.
[41] M. Auslander, A functorial approach to representation theory, pp. 105-
179 in Representations of Algebras, Puebla 1980, Lecture Notes in
Mathematics, Vol. 944, Springer-Verlag, 1982.

[42] D. Baer, W. Geigle and H. Lenzing, The preprojective algebra of a tame

hereditary Artin algebra, Comm. Algebra, 15(1-2) (1987), 425-457.
[43] D. J. Benson, Representations and Cohomology I: Basic representation
theory of finite groups and associative algebras, Cambridge Studies in
Advanced Math., Vol. 30, Cambridge University Press, 1991.

[44] K. Burke, Some Model-Theoretic Properties of Functor Categories for

Modules, Doctoral Thesis, University of Manchester, 1994.

26
[45] P. M. Cohn, On the free product of associative rings, Math. Z. 71(1)
(1959), 380-398.

[46] W. Crawley-Boevey, Modules of finite length over their endomorphism

rings, pp. 127-184 in Representations of Algebras and Related Topics,
London Math. Soc. Lect. Note Ser. Vol. 168, Cambridge University
Press, 1992.
[47] W. Crawley-Boevey, Infinite-dimensional modules in the representation
theory of finite-dimensional algebras, pp. 29-54 in I. Reiten, S. Smaland
. Solberg (Eds.), Algebras and Modules I, Canadian Math. Soc. Conf.
Proc., Vol. 23, Amer. Math. Soc, 1998.
[48] A. Facchini, Module Theory: Endomorphism rings and direct sum de-
compositions in some classes of modules, Progress in Math., Vol. 167,
Birkhauser, Basel, 1998.
[49] C. Faith and E. A. Walker, Direct sum representations of injective mod-
ules, J. Algebra, 5(2) (1967), 203-221.
[50] P. Gabriel and F. Ulmer, Lokal prasentierbare Kategorien, Lecture
Notes in Mathematics, Vol. 221, Springer-Verlag, 1971.

[51] P. Gabriel and M. Zisman, Calculus of Fractions and Homotopy Theory,

Springer-Verlag, 1967.
[52] K. R. Goodearl, Von Neumann Regular Rings, Pitman, London, 1979,
2nd Edn., Krieger Publishing, Florida, 1991.

[53] V. E. Govorov, On flat modules (in Russian), Sibirsk. Mat. Z., 6 (1965),
300-304.
[54] L. Gruson, Simple coherent functors, pp. 156-159 in Representations
of Algebras, Lecture Notes in Mathematics, Vol. 488, Springer-Verlag,
1975.

[55] I. Herzog, The Ziegler spectrum of a locally coherent Grothendieck cat-

egory, Proc. London Math. Soc., 74(3) (1997), 503-558.
[56] R. Kielpinski, On -pure injective modules, Bull. Polon. Acad. Sci.
Math., 15 (1967), 127-131.

[57] H. Krause, The spectrum of a locally coherent category, J. Pure Applied

Algebra, 114(3) (1997), 259-271.
[58] H. Krause, Exactly definable categories, J. Algebra, 201(2) (1998), 456-
492.

[59] S. Mac Lane, Categories for the Working Mathematician, Springer-

Verlag, 1971.
[60] S. Mac Lane and I. Moerdijk, Sheaves in Geometry and Logic: A First
Introduction to Topos Theory, Springer-Verlag, 1992.
[61] B. Mitchell, Theory of Categories, Academic Press, 1965.

27
[62] B. Mitchell, Rings with several objects, Adv. in Math., 8 (1972), 1-161.
[63] U. Oberst and H. Rohrl, Flat and coherent functors, J. Algebra, 14(1)
(1970), 91-105.
[64] M. Raynaud and L. Gruson, Criteres de platitude et de projectvite,
Seconde partie, Invent. Math., 13(1-2) (1971), 52-89.
[65] B. Stenstrom, Purity in functor categories, J. Algebra, 8(3) (1968), 352-
361.

[66] R. B. Warfield, Purity and algebraic compactness for modules, Pacific

J. Math., 28(3) (1969), 699-719.
[67] C. A. Weibel, An Introduction to Homological Algebra, Cambridge Uni-
versity Press, 1994.

[68] W. Zimmermann, Rein injektive direkte Summen von Moduln, Comm.

Algebra, 5(10) (1977), 1083-1117.

28