Homological Algebra

Steffen Oppermann

December 16, 2016

Contents

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

I General categories 5
2 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3 Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4 Natural transformations . . . . . . . . . . . . . . . . . . . . . . . 10
5 Equivalences of categories . . . . . . . . . . . . . . . . . . . . . . 13
6 Adjoint functors . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
7 Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
8 Limits and adjoint functors . . . . . . . . . . . . . . . . . . . . . 21
9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

II Additive and abelian categories 29

10 Additive categories . . . . . . . . . . . . . . . . . . . . . . . . . . 29
11 Kernels and cokernels . . . . . . . . . . . . . . . . . . . . . . . . 32
12 Abelian categories . . . . . . . . . . . . . . . . . . . . . . . . . . 33
13 Exact sequences, pullbacks and pushouts . . . . . . . . . . . . . . 34
14 Some diagram lemmas . . . . . . . . . . . . . . . . . . . . . . . . 40
15 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
N
III Hom and 53
16 Hom, projectives and injectives . . . . . . . . . . . . . . . . . . . 53
17 Tensor products . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
18 Hom-tensor adjunction . . . . . . . . . . . . . . . . . . . . . . . . 58
19 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

1
2 CONTENTS

IV Complexes and homology 63

20 The long exact sequence of homology . . . . . . . . . . . . . . . . 63
21 Cones and quasi-isomorphisms . . . . . . . . . . . . . . . . . . . 66
22 Homotopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
23 Projective and injective resolutions . . . . . . . . . . . . . . . . . 70
24 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

V Derived functors 77
25 Definition and first properties . . . . . . . . . . . . . . . . . . . . 77
26 Syzygies and dimension shift . . . . . . . . . . . . . . . . . . . . 80
27 Ext1 and extensions . . . . . . . . . . . . . . . . . . . . . . . . . 82
28 Total complexes - balancing Tor and Ext . . . . . . . . . . . . . . 86
29 Small global dimension . . . . . . . . . . . . . . . . . . . . . . . . 90
30 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

VI Triangulated categories 99
31 Motivation – triangles in the homotopy category . . . . . . . . . 99
32 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
33 Homotopy categories are triangulated . . . . . . . . . . . . . . . 106
34 Derived categories . . . . . . . . . . . . . . . . . . . . . . . . . . 109
35 Derived functors . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
36 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
1. INTRODUCTION 3

1 Introduction
Connections
Homological algebra

Algebraic topology Algebraic geometry Representation theory

simplicial homology; sheaf cohomology; extensions of modules
singular homology extensions of sheaves

Example 1.1. Let f : A B be a surjective map, and g : X B any map.

One may ask if there is a map h : X A such that g = f ◦ h.

• If we are just talking about sets, and maps, the answer is “yes”: for any
x ∈ X pick a preimage of g(x).
• If we are talking about vector spaces and linear maps the answer is also
“yes”: find a basis of X, then pick a preimage of g(x) for any basis element
x.
• If we are talking about abelian groups and linear maps the answer is
“sometimes”:
– Let A = Z/(4), B = X = Z/(2), and let f be the natural projection
and g the identity. Then there is no linear map h such that g = f ◦ h.
– Let A = Z/(6), B = X = Z/(2), and let f be the natural projection
and g the identity. Then there is a linear map h such that g = f ◦ h,
given by sending the residue class of 1 to the residue class of 3.
We will see: The obstruction to finding h is measure by the group Ext1 . (In
the examples above we have Ext1vector spaces = 0, and Ext1Z (Z/(2)Z/3) = 0, but
Ext1Z (Z/(2), Z/(2)) 6= 0.)
4 CONTENTS
Chapter I

General categories

2 Definition
In many situations in algebra (but also other parts of mathematics) we consider
some type of structures, say vector spaces, groups, rings, or similar. Typically
these are sets with some additional properties or structure. When studying
these kind of situations, there are two basic ingredients: We study the objects
having the desired structure themselves, and we study maps between objects
which preserve the structure (i.e. linear maps, group homomorphisms, ring ho-
momorphisms, . . . ). The concept of a category axiomatizes this.
Definition 2.1. A category C consists of
• a class of objects Ob C ;
• for any two objects X and Y a set of morphisms HomC (X, Y );
• for any three objects X, Y , and Z, a multiplication map

HomC (Y, Z) × HomC (X, Y ) HomC (X, Z) : (f, g) f ◦ g.

such that
• for any object X there is a morphism idX ∈ HomC (X, X) such that

∀Y ∈ Ob C ∀f ∈ HomC (X, Y ) : f ◦ idX = f,

∀Y ∈ Ob C ∀f ∈ HomC (Y, X) : idX ◦f = f.

5
6 CHAPTER I. GENERAL CATEGORIES

• multiplication is associative, that is for any objects X, Y, Z, and W and

f ∈ HomC (X, Y ), g ∈ HomC (Y, Z), and h ∈ HomC (Z, W ) we have

(h ◦ g) ◦ f = h ◦ (g ◦ f ).

Remark 2.2. Often HomC is just all maps with some additional nice property.
Example 2.3. • C = Set:

Ob Set = {sets}, and

HomSet (X, Y ) = {maps form X to Y }.

• C = Gp:

Ob Gp = {groups}, and
HomGp (G, H) = {group homomorphisms G to H}.

• C = Ab:

Ob Ab = {Abelian groups}, and

HomAb (G, H) = {group homomorphisms G to H}.

• C = Top:

Ob Top = {topological spaces}, and

HomTop (X, Y ) = {continuous maps X to Y }.

• For a ring R, C = Mod R:

Ob Mod R = {right R-modules}, and

HomMod R (M, N ) = {R-module homomorphisms M to N }.

• For a ring R, C = mod R:

Ob mod R = {finitely generated right R-modules}, and

Hommod R (M, N ) = HomMod R (M, N ).

Observation 2.4. For a category C , one can define the opposite category C op
by Ob C op = Ob C , and HomC op (X, Y ) = HomC (Y, X), together with the mul-
tiplication rule f ◦C op g = g ◦C f .
2. DEFINITION 7

One simple toy example of categories is the following

Construction 2.5. Let (X, 6) be a poset. The poset category C(X,6) is defined
by

Ob C(X,6) = X, and
(
{ιyx } if x 6 y
HomC(X,6) (x, y) =
∅ otherwise,

where ιzy ◦ ιyx = ιzx whenever x 6 y 6 z.

More generally, this construction works for a preordered set, that is a set
with an order that is not necessarily anti-symmetric.

Definition 2.6. A subcategory S of a category C consist of

• A subclass Ob S of Ob C ;

• for every S, T ∈ Ob S , a subset HomS (S, T ) ⊆ HomC (S, T );

such that the identity on any object in S is a morphism in S , and compositions

of morphisms in S are morphisms in S again.
The subcategory S ⊆ C is called full if for all S, T ∈ Ob S , HomS (S, T ) =
HomC (S, T ).

Example 2.7. • Ab is a full subcategory of Gp.

• For a poset (X, 6), and Y ⊆ X with induced poset structure, the poset
category C(Y,6) is a full subcategory of C(X,6) .

Definition 2.8. A morphism f ∈ HomC (X, Y ) is called

• monomorphism, if for any W and any g, h ∈ HomC (W, X) such that

f ◦ g = f ◦ h we have g = h;

• epimorphism if for any Z and any g, h ∈ HomC (Y, Z) such that g◦f = h◦f
we have g = h;

• split monomorphism (also called section) if there is g ∈ HomC (Y, X) such

that g ◦ f = idX ;

• split epimorphism (also called retraction) if there is g ∈ HomC (Y, X) such

that f ◦ g = idY ;
8 CHAPTER I. GENERAL CATEGORIES

• isomorphism if there is g ∈ HomC (Y, X) such that g ◦ f = idX and

f ◦ g = idY .

As the names suggest, split monomorphisms (split epimorphisms) are in fact

special monomorphsims (epimorphisms). See Exercise I.3
We often denote monomorphisms by arrows , and epimorphisms by ar-
rows .

Example 2.9. • In Set: monomorphism are split monomorphisms are in-

jective maps; epimorphisms are split epimorphisms are surjective maps;
isomorphisms are bijective maps.

• For a poset (X, 6), all morphisms in the poset category C(X,6) are both
mono- and epimorphisms. However, only identities are split monomor-
phisms or split epimorphisms.
In particular being a mono- and an epimorphism does not imply being an
isomorphism.

3 Functors
Definition 3.1. Let C and D be categories. A covariant functor F from C to
D consists of

• a map Ob C Ob D : X FX, and

• for any X, Y ∈ Ob C , a map HomC (X, Y ) HomD (FX, FY ), also de-

noted by F,

such that

• for any X ∈ Ob C we have F idX = idFX , and

• for any composable morphisms f and g in C we have F(g ◦ f ) = Fg ◦ Ff .

A contravariant functor from C to D is a covariant functor from C op to D.

In other words, it consists of maps HomC (X, Y ) HomD (FY, FX), and the
composition rule is F(g ◦ f ) = Ff ◦ Fg.

Example 3.2. • Let S be a subcategory of C . Then inclusion S C is

a (covariant) functor.
3. FUNCTORS 9

• Let C be a category, and X be an object. Then HomC (X, −) defines

a functor from C to Set: For any object Y ∈ Ob C , we obtain a set
HomC (X, Y ) by definition of category. For a morphism f ∈ HomC (Y1 , Y2 )
we define HomC (X, f ) by

HomC (X, f ) : HomC (X, Y1 ) HomC (X, Y2 ) : g f ◦ g.

This functor is called the covariant Hom-functor .

• Similarly one defines the contravariant Hom-functor HomC (−, X).
• For two posets (X, 6) and (Y, 6), a functor between the poset categories
is given by an order-preserving map X Y.

• Forming fundamental groups gives a functor Top∗ Gp from pointed

topological spaces to groups.

Definition 3.3. A functor F : C D is called

• faithful if for any X, Y ∈ Ob C the map HomC (X, Y ) HomD (FX, FY )

is injective;

• full if for any X, Y ∈ Ob C the map HomC (X, Y ) HomD (FX, FY ) is

surjective;
• dense if for any D ∈ Ob D there is C ∈ Ob C such that D ∼
= FC.
Example 3.4. • For an order preserving map f between two posets X and
Y , the associated functor between the poset categories is always faithful.
It is full if the images of two points are only comparable in Y if the two
points already were comparable in X. It is dense if and only if the map is
surjective.

• The forgetful functor Gp Set is faithful, but neither full nor dense.
Definition 3.5. Let X and C be categories. A C -valued presheaf on X is a
functor
X op C.
We denote by preshC X the collection of all C -valued presheaves on X .
By abuse of notation, for a poset (X, 6), we say a presheaf on (X, 6) is a
presheaf on the poset category C(X,6) .
10 CHAPTER I. GENERAL CATEGORIES

More explicitly, a C -valued presheaf F on a poset (X, 6) consist of

• for every x ∈ X, an object Fx ∈ ObC ;
• for every x, y ∈ X, such that x 6 y, a morphism resyx ∈ HomC (Fy , Fx );
such that resxx = idFx , and resyx ◦ reszy = reszx , whenever x 6 y 6 z.
Remark 3.6. Depending on the setup, and preferences of different authors,
various different notations are being used in the literature. These include
Fun(X op , C ), and, in partticular in the case of posets, “representations of X
in C ”.
This name “presheaves” which we will use here originates in the following
example.
Example 3.7. Let T be a topological space, and X the set of open subsets of
T . Then X is a poset with inclusion as partial order. Let S be a set (possibly
with some extra structure, for instance S = R or S = C).
Then we obtain a Set-valued presheaf F on X by setting F (U ) to be all
(nice) functions from U to S. Here the restriction morphisms are restriction of
functions to a smaller open subset of T .

4 Natural transformations
Definition 4.1. Let C and D be categories, and F and G be functors from C
to D. A natural transformation η from F to G consists of
• for every C ∈ Ob C a morphism ηC ∈ HomD (FC, GC),
such that for any morphism f ∈ HomC (C1 , C2 ) we have
ηC2 ◦ Ff = Gf ◦ ηC1 ,
that is the following diagram commutes in D:
ηC1
FC1 GC1

Ff Gf
ηC2
FC2 GC2

A natural transformation η is called natural isomorphism if all the ηC are

isomorphisms in D.
4. NATURAL TRANSFORMATIONS 11

Example 4.2. Let R be a ring. We denote by −∗ = HomR (−, R) the dual-

ity Mod R Mod Rop with respect to the ring. Then −∗∗ defines a functor
Mod R Mod R, and we have a natural transformation given by evaluation:

ev : idMod R −∗∗
evM (m) = [M ∗ 3 φ φ(m) ∈ R] ∈ HomR (M ∗ , R).

For R = F a field, we note that −∗∗ also defines a functor mod F mod F
between the categories of finite dimensional modules, and the induced natural
transformation
ev : idmod F −∗∗
is a natural isomorphism.
Observation 4.3. Let X be a category, such that the objects form a set. (Such
a category is called small .)
Then, for an arbitrary category C and functors F, G : X C the collection
of natural transformations from F to G forms a set. (In fact, it is a subset of
×X∈Ob X HomC (FX, GX).)
Thus, for a small category X , the C -valued presheaves on X form a cate-
gory, with

HompreshC X (F1 , F2 ) = {natural transformations F1 F2 }.

Obviously natural isomorphisms are precisely the isomorphisms in preshC X .

Example 4.4. Let (X, 6) be a poset, C a category, and F1 and F2 C -valued
presheaves on X.
A morphism f : F1 F2 consists of morphisms fx : (F1 )x (F2 )x for any
x ∈ X, such that resyx ◦fy = fx ◦ resyx whenever x 6 y. (Note that here the left
restriction refers to the structure of F2 , while the right restriction comes from
the structure of F1 .)
Example 4.5. • Let X = {1} be the poset with just one element. Then
preshC X = C .
• Let X = {1 6 2} be the poset with two comparable elements. Then the
objects of preshC X are morphisms in C , and morphisms of presheaves
are pairs of morphisms between domains and codomains, such that the
resulting square commutes.
12 CHAPTER I. GENERAL CATEGORIES

• Let X be the poset given by the Hasse diagram

ω

a b

The objects of preshC X are commutative squares in C . (Note that we

don’t need to specify resω ω a ω b ωb
0 , since res0 = res0 ◦ resa = res0 ◦ res .)

Theorem 4.6 (Yoneda Lemma). Let C be a category, C ∈ Ob C , and F a

functor C Set. Then the map

Y : {natural transformations HomC (C, −) F} FC

η ηC (idC )

is a bijection. In particular the natural transformations from HomC (C, −) to F

form a set.

Proof. We construct a map in the opposite direction. That is, given an element
x ∈ FC, we construct a natural transformation ζ x : HomC (C, −) F. For
D ∈ Ob C we set
x
ζD : HomC (C, D) FD : f (Ff )(x).

(Note that Ff ∈ HomSet (FC, FD), so this makes sense.)

Let us first check that ζ x is a natural transformation. Let g ∈ HomC (D1 , D2 ).
We have
x
ζD 2
◦ HomC (C, g) = [f (Ff )(x)] ◦ [f g ◦ f]
= [f (F(g ◦ f ))(x)]
= F(g) ◦ [f (Ff )(x)]
x
= F(g) ◦ ζD 1

We immediately see that

Y (ζ x ) = ζCx (idC ) = (F idC )(x) = idFC x = x.

It remains to see that for any natural transformation η : HomC (C, −) F

5. EQUIVALENCES OF CATEGORIES 13

we have η = ζ Y (η) . So Let D ∈ Ob C . Then

Y (η)
ζD = [f (Ff )(Y η)]
= [f (Ff ◦ ηC )(idC )]
= [f (ηD ◦ HomC (C, f ))(idC )] (η is a natural transformation)
= [f ηD (f )]
= ηD .

Corollary 4.7 (Yoneda embedding). Let X be a small category. Then the

functor
Y: X preshSet X : X HomX (−, X)
is fully faithful.

5 Equivalences of categories
Definition 5.1. A functor F : C D is called an equivalence if there is a
functor G : D C such that F ◦ G = idD and G ◦ F ∼
∼ = idC .
nat nat

Theorem 5.2. A functor F : C D is an equivalence if and only if it is full,

faithful, and dense.
Proof. Assume first that F is an equivalence, and let G as in the definition.
Let η : G ◦ F idC be a natural isomorphism. Then for any morphism
f ∈ HomC (C1 , C2 ) we have the commutative square
ηC1
GFC1 C1

GFf f
ηC2
GFC2 C2

−1
Thus f = ηC2 ◦ GFf ◦ ηC1
is uniquely determined by Ff . That is F is faithful.
Let ζ ba a natural isomorphism F◦G idD . In particular for any D ∈ Ob D
we have an isomorphism ζD : FGD D, showing that F is dense.
14 CHAPTER I. GENERAL CATEGORIES

To see that F is full, let f ∈ HomD (FC1 , FC2 ). Using the natural isomor-
phisms η and ζ as above, we construct the commutative diagram

FηC1 ζFC1
FC1 FGFC1 FC1

f g h
FηC2 ζFC2
FC2 FGFC2 FC2

where g and h are the unique maps making the squares commutative. By
naturality of ζ we know that g = FGh, and thus the commutativity of the left
hand square gives us that
f = FηC2 ◦ FGh ◦ (FηC1 )−1
−1
= F(ηC2 ◦ Gh ◦ ηC 1
)
showing that f is in the image of F.
Now assume conversely that F is full, faithful, and dense. By (a strong
version of) the axiom of choice, and since F is dense, we may fix, for any D ∈
ObD , an object GD in C and an isomorphism ζD : FGD D. For a morphism
f ∈ HomD (D1 , D2 ) we use the bijection

HomC (GD1 , GD2 ) HomD (FGD1 , FGD2 )

induced by F (since it is full and faithful), and define Gf to be the preimage of
−1
ζD2
◦ f ◦ ζD1 .
We claim that the above makes G a functor from D to C . Firstly we have
−1
G idD = F−1 (ζD ◦ idD ◦ζD ) = F−1 (idFGD ) = idGD .

f g
Secondly, for morphisms D1 D2 D3 ,
−1 −1 −1
G(g ◦ f ) = F−1 (ζD3
◦ g ◦ f ◦ ζD1 ) = F−1 (ζD3
◦ g ◦ ζD2 ◦ ζD2
◦ f ◦ ζD1 )
−1 −1
= F−1 (ζD3
◦ g ◦ ζD2 ) ◦ F−1 (ζD2
◦ f ◦ ζD1 ) = Gg ◦ Gf.

Next we claim that ζ defines a natural isomorphism F ◦ G idD . Let f ∈

HomD (D1 , D2 ). Then
−1
ζD2 ◦ FGf = ζD2 ◦ ζD2
◦ f ◦ ζD1 = f ◦ ζD1 .
6. ADJOINT FUNCTORS 15

Finally, we construct a natural isomorphism η : G ◦ F idC . First note that

ζ induces mutually inverse natural isomorphisms
−1
ζF− : F ◦ G ◦ F F and ζF− :F F ◦ G ◦ F.

Since F is fully faithful, we can find unique morphisms ηC : GFC C and

−
ηC :C GFC such that
−1 −
ζFC = F ηC and ζFC = F ηC .

If follows that η is a natural transformation, with inverse η − .

Example 5.3. Let F be a field.

Let MatF be the category given by

Ob MatF = N0 , and
HomMatF (m, n) = {n × m-matrices over F}

with matrix multiplication.

Let mod F be the category of finite dimensional F-vector spaces, with F-
vector space homomorphisms as morphisms.
Then the natural functor MatF mod F sending n to Fn is an equivalence.
We observe that constructing an equivalence in the other direction amounts
to choosing a basis for every finite dimensional F-vector space.

6 Adjoint functors
Definition 6.1. Let C and D be categories, and F : C D and G : D C
functors. We say that (F, G) is an adjoint pair if the functors

HomD (F−, −) and HomC (−, G−) : C op × D Set

are naturally isomorphic.

Example 6.2. Let (X, 6) be a poset, and C a category. For x ∈ X we have a

natural projection functor

πx : preshC X C:F Fx .
16 CHAPTER I. GENERAL CATEGORIES

We may also consider the diagonal functor ∆ : C preshC X given by

∆(C)x = C for any x ∈ X, and resxy = idC for any y 6 x.
If (X, 6) has a smallest element 0, then (π0 , ∆) is an adjoint pair. Similarly,
if there is a largest element ω, then (∆, πω ) is an adjoint pair.
Example 6.3 (Free modules). Let R be a ring. Then we have the forgetful
functor f : Mod R Set.
We construct a left adjoint R(−) : Set Mod R: For a set X,

R(X) = {functions f : X R | f (x) 6= 0 for only finitely many x ∈ X}.

For a map ϕ : X Y we set

 
X
R(ϕ) : R(X) R(Y ) : f y f (x) .
x∈ϕ−1 (y)

We claim that R(−) is left adjoint to f.

For x ∈ X, we let
(
1 if y = x
χx : X R: y .
0 6 x
if y =

Then χx ∈ R(X) .
Now we can define the mutually inverse natural transformations by

HomSet (X, fM ) HomR (R(X) , M )

X
ϕ [f ϕ(x) · f (x)]
x∈X

[x ψ(χx )] ψ

(Note that the sum in the second line is finite, since f (x) = 0 for almost all
x ∈ X.)

Example 6.4. Consider the forgetful functor f : Ab Gp. This functor has
a left adjoint, given by forming commutator factor groups.

Proposition 6.5 (Unit-counit adjunction). Let F : C D and G : D C be

two functors. Then the following are equivalent:
6. ADJOINT FUNCTORS 17

1. (F, G) is an adjoint pair.

2. there are natural transformations η : idC G ◦ F and ε : F ◦ G idD

(called unit and counit, respectively), such that

idF = εF− ◦ Fη and idG = Gε ◦ ηG− ,

i.e. for any C ∈ Ob C and D ∈ Ob D

idFC = εFC ◦ FηC and idGD = GεD ◦ ηGD .

Proof. Let α : HomD (F−, −) HomC (−, G−) be a natural transformation.

Then we may define a natural transformation η : idC G ◦ F by

ηC = αC,FC (idFC ).

To check that this defines a natural transformation, note that for a morphism
f ∈ HomC (C1 , C2 ) we have

GFf ◦ ηC1 = GFf ◦ αC1 ,FC1 (idFC1 )

= αC1 ,FC2 (Ff )
= αC2 ,FC2 (idFC2 ) ◦ f
= ηC2 ◦ f,

where the middle two equalities follow from the naturality of α in the second
and first argument, respectively.
Conversely, given a natural transformation η : idC G ◦ F we can define a
natural transformation α : HomD (F−, −) HomC (−, G−) by

αC,D (f ) = G(f ) ◦ ηC .

It is immediately checked that these two constructions are mutually inverse.

Similarly, we obtain a bijection between natural transformations

β : HomC (−, G−) HomD (F−, −) and ε : F ◦ G idD ,

sending β to the natural transformation given by εD = βGD,D (idGD ).

Now let α and η and β and ε correspond to each other as above. Then

β ◦ α = idHomD (F−,−)
⇐⇒ ∀C ∈ Ob C ∀D ∈ Ob D : βC,D ◦ αC,D = idHomD (FC,D)
18 CHAPTER I. GENERAL CATEGORIES

Moreover, since any morphism from FC to D is a multiple of idFC

⇐⇒ ∀C ∈ Ob C : βC,FC ◦ αC,FC (idFC ) = idFC

and, inserting αC,FC (idFC ) = G(idFC ) ◦ ηC = ηC , and βC,FC (ηC ) = εFC ◦ F(ηC ),
we obtain

⇐⇒ ∀C ∈ Ob C : εFC ◦ F(ηC ) = idFC

⇐⇒ εF− ◦ Fη = idF .

Similarly one can see that α ◦ β = idHomC (−,G−) if and only if Gε ◦ ηG− =
idG .

7 Limits
Definition 7.1. Let X be a small category (which we think of as indices, in
some sense), and C an arbitrary category. We denote by ∆ the functor

∆: C preshC X : C ∆C,

where ∆C is the functor sending any object of X to C, and any morphism to

idC .
Let F ∈ Ob preshC X .
• a limit (or inverse limit, projective limit) of F is an object lim F ∈ Ob C ,
←−
together with a natural isomorphism

HomC (−, lim F ) ∼

= HompreshC X (∆−, F )
←−

of functors C op Set;
• a colimit (or direct limit, injective limit) of F is an object lim F ∈ Ob C ,
−→
together with a natural isomorphism

HomC (lim F, −) ∼
= HompreshC X (F, ∆−)
−→

of functors C Set.
Observation 7.2. Note that a limit of F can equivalently be characterized as
an object lim F ∈ Ob C , together with morphisms ϕx : lim F Fx such that
←− ←−
7. LIMITS 19

f ◦ϕx = ϕy for any f : x y ∈ X , which is universal in the following sense: for

any other object C, together with maps ψx : C Fx such that f ◦ ψx = ψy for
any f : x y ∈ X there is a unique map Ψ : C lim F such that ψx = ϕx ◦ Ψ
←−
for all x ∈ Ob X .
The dual description holds for colimits.

Proposition 7.3. Let F ∈ Ob preshC X as above. If a limit lim F exists, then

←−
it is unique up to (unique) isomorphism. If a colimit lim F exists, then it is
−→
unique up to (unique) isomorphism.
Therefore it makes sense to speak about the limit or colimit.

Proof. Let (L, ϕx ) and (L0 , ϕ0x ) be two limits. Then, by the universal property
for L, there is a morphism Ψ : L0 L such that ϕ0x = ϕx ◦ Ψ. By the universal
property for L0 there is a morphism Ψ0 : L L0 such that ϕx = ϕ0x ◦ Ψ0 .
Now, again by the universal property of L, there exists a unique morphism
Φ: L L such that ϕx = ϕx ◦ Φ. But we know two candidates for Φ: idL and
Ψ ◦ Ψ0 . It follows that Ψ ◦ Ψ0 = idL . Similarly one sees that Ψ0 ◦ Ψ = idL0 . It
follows that Ψ and Ψ0 are mutually inverse isomorphisms.

Observation 7.4. If any F ∈ Ob preshC X has a limit, then lim defines a

←−
functor preshC X C , and this functor is right adjoint to ∆.
If any F ∈ Ob preshC X has a colimit, then lim gives a functor preshC X C,
−→
and this functor is left adjoint to ∆.

Example 7.5. Let (X, 6) be a poset with a smallest element 0. Then we have
seen in Example 6.2 that

lim = π0 : preshC X C:F F0 .

−→
Similarly, if X has a largest element ω, then lim = πω .
←−
Definition 7.6 (Product and coproduct). Let X be a set. We may regard X
as a poset with trivial poset structure. Let C be a category, and F ∈ preshC X.
(That is F is a collection of objects Fx , one for each x ∈ X.)

• If the limit Q
lim F exists, then it is called product of the objects Fx , and
←−
denoted by x∈X Fx .

• If the colimit lim

`F exists, then it is called coproduct of the objects Fx ,
−→
and denoted by x∈X Fx .
20 CHAPTER I. GENERAL CATEGORIES

Example 7.7. In the category Set, products are cartesian products, and co-
products are disjoint unions.
Example 7.8. In Mod R, both finite products and finite coproducts are given
by direct sums.
Definition 7.9 (Pullback and pushout). Let X be the poset given by the Hasse
diagram
a b

Let F ∈ preshC X. If the limit lim F exists, then it is called the pullback (or
←− Q
fibre product) of F , and denoted by Fa Fb .
F0
By abuse of notation we also call the commutative square

Fω Fa

Fb F0

a pullback, provided Fω is the pullback of the rest of the diagram.

More explicitly, a pullback is a commutative square as above, such that for
any other X with morphisms X Fa and X Fb making a similar square
commutative, there is a unique morphism X Fω making the two triangles in
the following diagram commutative.

Fω Fa

Fb F0

Let Y be the poset given by the Hasse diagram

ω

a b
8. LIMITS AND ADJOINT FUNCTORS 21

Let F ∈ preshC X. If
`the colimit −
lim F exists, then we call it the pushout of F ,
→
and denote it by Fa Fb .
Fω

Example 7.10. In the category Set, the pullback of

Fa Fb
resa
0 resb0
F0

is given by
Y
Fa Fb = {(a, b) ∈ Fa × Fb | resa0 (a) = resb0 (b)}.
F0

The pushout of
Fω
resω
a resω
b

Fa Fb

is given by
a a
Fa Fb = Fa Fb /(resω ω
a (x) ∼ resb (x) | x ∈ Fω ).
Fω

8 Limits and adjoint functors

Construction 8.1. Let F : C D be a functor, and X be a small category.
Then F induces a functor Fpresh
: preshC X preshD X .

Lemma 8.2. Let (F, G) be an adjoint pair of functors between categories C and
D. Let X be a small category.
Then the corresponding functors between presheaf categories

Fpresh : preshC X preshD X and Gpresh : preshD X preshC X

also form an adjoint pair.

Proof. We have a natural isomorphism η : HomD (F−, −) HomC (−, G−), i.e.
a collection of bijections ηX,Y : HomD (FX, Y ) HomC (X, GY ) such that
22 CHAPTER I. GENERAL CATEGORIES

• for any f : X X 0 ∈ C we have a commutative square

ηX,Y
HomD (FX, Y ) HomD (X, GY )

HomD (Ff, Y ) HomC (f, GY )

ηX 0 ,Y
HomD (FX 0 , Y ) HomD (X 0 , GY )

that is for any ϕ : FX 0 Y ∈ D we have

ηX,Y (ϕ ◦ Ff ) = ηX 0 ,Y (ϕ) ◦ f.

• similarly, for any g : Y Y 0 ∈ D and any ϕ ∈ HomD (X, Y ) we have

Gg ◦ ηX,Y (ϕ) = ηX,Y 0 (g ◦ ϕ).

Now we observe that

HompreshD X (Fpresh S, T )
={(fx )x∈Ob X ∈ ×Hom
x∈Ob X
D (FSx , Tx ) | ∀α ∈ HomX (x, y) : fx ◦ FSα = Tα ◦ fy }

Now fx ◦Sα and Tα ◦fy are morphisms from FSy to Tx . Since ηSy ,TX is a bijection
we may replace the conditions above by ηSy ,TX (fx ◦ FSα ) = ηSy ,Tx (Tα ◦ fy ).
Now note that by the two bullet points above the left hand side is equal to
ηSx ,Tx (fx ) ◦ Sα , while the right hand side is equal to GTα ηSy ,Ty (fy ). Thus,
writing gi for ηSx ,Tx (fx ), the above set is in bijection to

{(gx )x∈Ob X ∈ ×
x∈Ob X
HomC (Sx , GTx ) | ∀α ∈ HomX (x, y) : gx ◦ Sα = GTα ◦ gy }

presh
= HompreshC X (S, G T)

Theorem 8.3. Let (F, G) be an adjoint pair of functors between categories C

and D. Let X be a small category.
• Let X ∈ preshD X such that lim X exists. Then
←−
lim Gpresh X = G lim X.
←− ←−
(In particular this limit also exists.)
9. EXERCISES 23

• Let X ∈ preshC X such that lim X exists. Then

−→
lim Fpresh X = F lim X.
−→ −→

Motto: Right adjoints commute with limits, left adjoints commute with colim-
its.

Proof. We only prove the first claim, the second one is dual.
We have

HomC (−, G lim X) ∼

= HomD (F−, lim X)
←− ←−
∼
= Hompresh X (∆F−, X)
D

= HompreshD X (Fpresh ∆−, X)

= HompreshC X (∆−, Gpresh X).

Example 8.4. Consider the adjoint pair (R(−) , f) between Mod R and Set
from Example 6.3. We note that
`
R(X Y)
= R(X) ⊕ R(Y ) and f(M ⊕ N ) = fM × fN

by Theorem 8.3 above. (Of course in this example we could also have checked
that directly.)
However in general neither
a
R(X×Y ) = R(X) × R(Y ) nor f(M ⊕ N ) = fM fN.

9 Exercises
Exercise I.1. • Describe which morphisms in Set are monomorphisms,
epimorphisms, split monomorphisms, and split epimorphisms.

• Describe which morphisms in Top are monomorphisms and which mor-

phism are epimorphisms. Find an example of a morphism that is both a
monomorphism and an epimorphism, but not an isomorphism.

• Show that in the category Ring, the inclusion Z Q is both a monomor-

phism and an epimorphism.
24 CHAPTER I. GENERAL CATEGORIES

Exercise I.2. Let C be a category, f and g two composable morphisms.

• Show that if f and g are monomorphisms, then so is f ◦ g.

• Show that if f ◦ g is a monomorphism, then so is g.

• Find an example of morphisms such that f ◦ g is a monomorphism, but f

is not a monomorphism.

Exercise I.3. Show that

• Any split monomorphism is a monomorphism.

• Any split epimorphism is an epimorphism.

• The following are equivalent, for a morphism f :

– f is an isomorphism;
– f is a split monomorphism and an epimorphism;
– f is a monomorphism ans a split epimorphism.

• If f is an isomorphism, the the g in the definition is uniquely determined.

(And hence we denote it by f −1 .)

Exercise I.4. Which of the following functors are full? faithful? dense?

• the natural inclusion Ab Gp;

• forgetting the topology Top Set;

• the Hom-functor HomAb (Z/(2), −) : Ab Set.

a b
Exercise I.5. Let X be the poset given by the Hasse diagram that is
0
a > 0 and b > 0 with a and b incomparable.

• Determine all objects F ∈ preshSet X such that F (i) ∈ {∅, {?}} ∀i ∈ X.

• Which of the presheaves determined above are isomorphic to a presheaf

of the form HomCX (−, i) for some i ∈ X?

• Determine all objects F ∈ preshmod F X (where F is a field), such that

F (i) = F for all i ∈ X.
9. EXERCISES 25

• Which of the presheaves determined in the third part are isomorphic?

Exercise I.6. Recall that for a group G, we denote by Gop the opposite group,
that is the group with the same elements as G, but multiplication given by
g ·op h = h · g.

• Show that this construction defines a (covariant!) functor Gp Gp.

• Any group G is isomorphic to its opposite group: An isomorphism is given

by g g −1 . Investigate if this collection of isomorphisms defines a natural
isomorphism idGp −op .

Exercise I.7. Let G be a non-trivial group. We can consider the category CG

having only one object ?, with HomCG (?, ?) = G, and composition of morphisms
given by group multiplication. (Convince yourself that this is a category.)
Consider the following two functors F = HomCG (?, −) and H : CG Set
given by H(?) = G and H(g) = 1G .
Show that the functors F and H agree on all (i.e. the one) objects, but are
not naturally isomorphic.

Exercise I.8. Let X = {a, b, c} with the preorder given a 6 a, a 6 b, a 6 c,

b 6 a, b 6 b, b 6 c, c 6 c. (So a and b violate anti-symmetry). Let Y = {1, 2}
with the natural poset structure (i.e. 1 6 2). Show that the poset categories
C(X,6) and C(Y,6) are equivalent.
More generally, given an arbitrary preordered set X, find a poset Y such
that the cateories C(X,6) and C(Y,6) are equivalent.

Exercise I.9. Let X = {1 6 2} and Y = {1}. Let C be any category.

• Convince yourself that inclusion of Y into X induces a functor

preshC X preshC Y : F F ◦ incl .

• Find a right adjoint to the functor above.

• For C = Set or C = Ab, find a left adjoint to the functor above.

Exercise I.10. Find left adjoints to the functors

• forget : Ring Rng, the forgetful functor from rings with multiplica-
tive unit to rings without multiplicative unit.
26 CHAPTER I. GENERAL CATEGORIES

• forget : Ring∗ Ring, where Ring∗ is the category of “pointed rings”,

that is pairs (R, r) of a Ring R and an element r, and morphisms being
ring homomorphisms which send the distinguished element of the first ring
to the distinguished element of the second ring.
Find the unit and counit maps for both the above adjunctions.
Exercise I.11. Let A be an abelian group, S and T two subgroups.
• Show that the pullback of

incl.
incl.
T A

is given by S ∩ T .
• Show that the pushout of
proj.
A A/S

proj.

A/T

is given by A/(S + T ).
Exercise I.12. In the category Ab
• Show that the pullback of

L
α
β
M N

is given by Y
L M = {(l, m) ∈ L ⊕ M | α(l) = β(m)}
N
(with the obvious maps to L and M ).
9. EXERCISES 27

• Show that the pushout of

β
L M
α

is given by a
M N = M ⊕ N/{(β(`), −α(`)) | ` ∈ L}.
L

Exercise I.13. Let X be any poset, and F a Set-valued presheaf on X. Show

• that the limit lim F exists;
←−
• that the colimit lim F exists.
−→
Hint: Construct them explicitly, starting with product and coproduct, respec-
tively.
Exercise I.14. In any category, consider the following diagram

X Y Z

` ` `
W W X Y (W X Y) Y Z

`
W X Z

` `
Show that the “iterated pushout” (W X Y ) Y Z`is isomorphic to the pushout
along the composition of the horizontal arrows W X Z, provided all the push-
outs exist.
28 CHAPTER I. GENERAL CATEGORIES
Chapter II

Additive and abelian

categories

10 Additive categories
Definition 10.1. A pre-additive category is a category A such that all Hom-
sets are abelian groups, and composition of morphisms is bilinear.
An additive category is a pre-additive category A such that

• there is a zero-object, i.e. an object 0 such that for any X ∈ Ob A both

HomA (X, 0) and HomA (0, X) contain precisely one morphism.

• for any X, Y ∈ Ob A there is a biproduct, i.e. an object X ⊕ Y with

morphisms
ιX ιY

X X ⊕Y Y

πX πY

such that

idX = πX ◦ ιX , idY = πY ◦ ιY , and idX⊕Y = ιX ◦ πX + ιY ◦ πY .

Example 10.2. • Ab is an additive category.

29
30 CHAPTER II. ADDITIVE AND ABELIAN CATEGORIES

• For a ring R, the category Mod R is additive.

• Set and Top are not additive categories.

• For any small category X , and any additive category A , the category
preshA X is additive.

Observation 10.3. In the situation of the biproduct diagram, we have

πY ◦ ιX = πY ◦ ιX ◦ πX ◦ιX = πy ◦ ιX − πY ◦ ιY ◦πY ◦ ιX = 0,
| {z } | {z }
=idX⊕Y −ιY ◦πY =idY

and similarly
πX ◦ ιY = 0.

Lemma 10.4. Let A be an additive category. Then, for any two objects X and
Y , the biproduct X ⊕ Y is a product and a coproduct of X and Y .

Proof. We show that X ⊕ Y is a product, the proof that it is a coproduct is

dual.
We have to show that for any maps fX : H X and fY : H Y there is
precisely one map f : H X ⊕ Y such that πX ◦ f = fX and πY ◦ f = fY .
We see that

f = idX⊕Y ◦f = ιX ◦ πX ◦ f + ιY ◦ πY ◦ f = ιX ◦ fX + ιY ◦ fY .

Thus f is unique. On the other hand we can see that ιX ◦ fX + ιY ◦ fY fullfils

the requirements:

πX ◦ (ιX ◦ fX + ιY ◦ fY ) = πX ◦ ιX ◦fX + πX ◦ ιY ◦fY = fX .

| {z } | {z }
=idX =0

and similarly
πY ◦ (ιX ◦ fX + ιY ◦ fY ) = fY .

Remark 10.5. • In particular in an additive category any two objects have

isomorphic product and coproduct. This shows that neither Set nor Top
can be additive categories.
10. ADDITIVE CATEGORIES 31

• It is possible to show that the addition of morphism is completely deter-

mined by the biproducts, and not an additional part of the structure.
That is, an additive category is a category with a zero-object, such that
any two objects have a product and a coproduct which are isomorphic,
satisfying certain properties.

Remark 10.6. For n > 1, and objects X1 , . . . , Xn , we can iteratedly construct

X = (· · · (X1 ⊕ X2 ) ⊕ X3 ) · · · ) ⊕ Xn .

We note that for this object we have, similarly to the biproduct diagram and
with maps given by compositions of the maps there

πi : X Xi , and ιi : Xi X

such that
n
X
πi ◦ ιi = idXi ∀i, and ιi ◦ πi = idX .
i=1

Remark 10.7 (Matrix notation). We often use the following intuitive matrix
notation for morphisms from X = X1 ⊕ · · · ⊕ Xn to Y = Y1 ⊕ · · · ⊕ Ym :
A morphism f : X Y is represented by the matrix

(πYi ◦ f ◦ ιXj )i=1,...,m .

j=1,...,n

Conversely, given a matrix

(fij )i=1,...,m with fij : Xj Yi

j=1,...,n

we can interpret it as the map

m X
X n
ιYi ◦ fij ◦ πXj : X Y.
i=1 j=1

One easily sees that these constructions are mutually inverse to each other, and
thus we may identify matrices and maps between biproducts.
The main advantage of this notation is, that composition of maps is just
given by matrix multiplication:
32 CHAPTER II. ADDITIVE AND ABELIAN CATEGORIES

Given
o
M n
(fjk ) M m
(gij ) M
Xk Yj Zi
k=1 j=1 i=1

we have
Xm X
n Xn X
o
(gij )i,j ◦ (fjk )j,k = ( ιZi ◦ gij ◦ πYj ) ◦ ( ιYj ◦ fjk ◦ πXk )
i=1 j=1 j=1 k=1
Xm X n Xo
=( ιZi ◦ gij ◦ fjk ◦ πXk )
i=1 j=1 k=1
Xn
=( gij ◦ fjk )i,k .
j=1

11 Kernels and cokernels

Definition 11.1. Let f : X Y be a morphism in an additive category. The
kernel of f is (if it exists) the pullback of

0
f
X Y

In other words, the kernel is given by an object Ker f , together with a morphism
κ : Ker f X (the other morphism Ker f 0 necessarily being 0), such that
f ◦ κ = 0, and that for any object H and morphism h : H X such that
f ◦ h = 0 there is a unique morphism b h: H Ker f such that h = κ ◦ bh.
Dually, the cokernel of f is, if it exists, the pushout of

X Y
f

and consists of an object Cok f and a map π : Y Cok f .

12. ABELIAN CATEGORIES 33

Observation 11.2. In the definition of kernel above the map κ is a monomor-

phism: Let h, g : H Ker f such that κ ◦ h = κ ◦ g. Then clearly f ◦ κ ◦ h = 0,
and therefore κ ◦ h factors uniquely through κ, i.e. h = g.
Dually the map π in the definition of cokernel is an epimorphism.

Lemma 11.3. Let f : X Y be a morphism in an additive category. Then

f is a monomorphism if and only if 0 X is a kernel of f . Dually f is an
epimorphism if and only if Y 0 is a cokernel of f .

Proof. Assume first that f is a monomorphism. Then any morphism h : H X

such that f ◦ h = 0 is necessarily 0, and therefore factors (uniquely) through
0 X.
Conversely, assume 0 X is a kernel of f . Then any map h such that
f ◦ h = 0 factors through 0, that is is zero.

12 Abelian categories
Definition 12.1. A pre-abelian category is an additive category A , in which
every morphism has a kernel and a cokernel.

Definition 12.2. Let A be pre-abelian, and f : X Y a morphism. Let

ι π
Ker f X and Y Cok f be kernel and cokernel of f . Then

• the image of f , denoted by Im f , is the kernel of π;

• the coimage of f , denoted by Coim f , is the cokernel of ι.

Proposition 12.3. In the setup of Definition 12.2, there is a unique map f

making the diagram

ι f π
Ker f X Y Cok f

ρ κ
f
Coim f Im f

commutative.
34 CHAPTER II. ADDITIVE AND ABELIAN CATEGORIES

Proof. Uniqueness of f follows immediately, since morphisms ρ and κ are epi

and mono, respectively.
Since f ◦ ι = 0 there is a morphism f 0 : Coim f Y such that f 0 ◦ ρ = f .
Moreover, since ρ is epi, 0 = π ◦ f = π ◦ f ◦ ρ implies π ◦ f 0 = 0, hence f 0 factors
0

through κ. This proves the existence of f .

Definition 12.4. An abelian category is a pre-abelian category, in which, for
any morphism f : X Y the induced morphism f : Coim f Im f is an iso-
morphism.
Remark 12.5. In other words, an abelian category is an additive category
with kernels and cokernels, in which the first isomorphism theorem holds. (Re-
call that the first isomorphism theorem is precisely that X modulo kernel is
isomorphic with the image.)
Observation 12.6. In an abelian category
• every monomorphism is a kernel of its cokernel;
• every epimorphism is a cokernel of its kernel;
• every morphism that is both a monomorphism and an epimorphism is an
isomorphism.
Remark 12.7. One can show that the first two points above give an equivalent
definition of abelian category.

13 Exact sequences, pullbacks and pushouts

f g
Observation 13.1. Let A B C be morphisms in an abelian category,
such that g ◦ f = 0. Then we have the following commutative diagram

Im f B Cok f

Ker g B Im g

where the right part consists of the cokernels of the left horizontal maps, and
the left part consists of the kernels of the right horizontal maps.
13. EXACT SEQUENCES, PULLBACKS AND PUSHOUTS 35

It follows that the morphism Im f Ker g is an isomorphism if and only

if the morphism Cok f Im g is. (We may note that the former always is a
monomorphism, and the latter always is an epimorphism.)

f g
Definition 13.2. Let A B C be morphisms in an abelian category, such
that g ◦ f = 0. We say that this sequence of morphisms is exact if the natural
morphism Im f Ker g is an isomorphism.
We say that a longer sequence of morphisms is exact if it is exact in every
(inner) position.

f
Example 13.3. • The sequence 0 A B is exact if and only if Ker f =
0, that is if and only if f is a monomorphism.
g
• Dually the sequence A B 0 is exact if and only if Cok g = 0.

f g
• The sequence 0 A B C is exact if and only if

– f is a monomorphism (as before), and

f
– Ker g = Im f = A, that is A B is a kernel of g.

f g g
• Dually the sequence A B C 0 is exact if and only if B C is a
cokernel of f .

f g
• The sequence 0 A B C 0 is exact if both f is a kernel of g and
g is a cokernel of f . Such an exact sequence is called short exact sequence.

Proposition 13.4. Let A be an abelian category. Consider the morphisms in

the following commutative square.

f
A B
g h
i
C D
36 CHAPTER II. ADDITIVE AND ABELIAN CATEGORIES

• The square is a pullback if and only if the sequence

−f


g (h i)
0 A B⊕C D

is exact.
• The square is a pushout if and only if the sequence

−f


g (h i)
A B⊕C D 0

is exact.
Proof.

We first observe that the commutativity of the square means that (h i)◦
−f
g = −h ◦ f + i ◦ g = 0.
Now observe that the square is a pullback if and only if

e ∈ Ob A ∀fe: A
∀A e B ∀e
g: A
e C:
if h ◦ fe = i ◦ ge then ∃!ϕ : A
e A : fe = f ◦ ϕ and ge = g ◦ ϕ

assembling maps in matrices we obtain that this is equivalent to

 
∀Ae ∈ Ob A ∀ −fe : A e B ⊕C:
g
e
   
if (h i) ◦ −gef = 0 then ∃!ϕ : A A : −gef = −f


◦ϕ
e e e
g

Now note that this last statement is precisely the definition of a kernel.
The proof of the second point is dual.
Remark 13.5. Proposition 13.4 shows, in particular, that in abelian categories
pullbacks and pushouts always exist.
Corollary 13.6. Let A be an abelian category. If the square

f
A B
g h
i
C D
13. EXACT SEQUENCES, PULLBACKS AND PUSHOUTS 37

• is a pullback, and i is an epi, then it is also a pushout;

• is a pushout, and f is a mono, then it is also a pullback.
Proposition 13.7. Let A be an abelian category.
• If the square
f
A B
g h
i
C D

is a pullback, then the kernel morphism Ker f Ker i is an isomorphism.

• If the square is a pushout then the cokernel morphism Cok f Cok i is

an isomorphism.
Proof. We only prove the first part, the second one is dual.
Denote the inclusions of the kernels by ι : Ker f A and κ : Ker i C,
respectively, and the kernel morphism by ϕ : Ker f Ker i. Consider the mor-
phism 0 : Ker i B, as indicated in the following diagram.

ι f
Ker f A B
κ κ 0
ϕ
b
h
b
b
g
i
Ker i κ C D

Clearly i ◦ κ = 0 = h ◦ 0, so by the pullback property there is a morphism

κ
b : Ker i A such that κ = g ◦ κ b and 0 = f ◦ κ b. By the second equality κ b
factors through the kernel of f , that is there is a morphism κb : Ker i
b Ker f
such that κb=ι◦κ b.
b
Now it only remains to verify that κb is an inverse of ϕ. Firstly we have
b

κ◦ϕ◦κ
b=g◦ι◦κ
b b=g◦κ
b = κ,
and hence, since κ is a monomorphism,

ϕ◦κ
b = idKer i .
b
38 CHAPTER II. ADDITIVE AND ABELIAN CATEGORIES

Secondly we have
−f −f 0 −f ◦ι −f






g ◦ι◦κb◦ϕ=
b
g b ◦ ϕ = ( κ0 ) ◦ ϕ = κ◦ϕ
◦κ = g◦ι = g ◦ ι,
and hence, since both ι and −f


g are monomorphisms,

b ◦ ϕ = idKer f .
κ
b

Corollary 13.8. In an abelian category

• the pullback of a mono is a mono;
• the pullback of an epi is an epi;
• the pushout of a mono is a mono;
• the pushout of an epi is an epi.
Moreover, in the case of the second and third point, the square in question is
actually both a pullback and a pushout.
Proof. The first point follows immediately from Proposition 13.7 above.
For the second point, note first that the pullback now also is a pushout, by
Corollary 13.6. Now apply (the dual-part of) Proposition 13.7.
The third and fourth points are dual to the second and first, respectively.
f g
Proposition 13.9. In an abelian category, let A B C be morphisms such
that g ◦ f = 0. Then the following are equivalent.
f g
• The sequence A B C is exact.

• For any morphism x : X B, such that g ◦ x = 0, there is an object X,

b
and morphisms x
b and f as in the following diagram
b

fb
X
b X

x
b x
f
A B g C

such that the square commutes, and fb is an epimorphism.

13. EXACT SEQUENCES, PULLBACKS AND PUSHOUTS 39

• For any morphism y : B Y , such that y ◦ f = 0, there are an object Y̌ ,

and morphisms y̌ and fˇ as in the following diagram
f g
A B C
y y̌
ǧ
Y Y̌

such that the square commutes, and ǧ is a monomorphism.

Proof. We only prove the equivalence of the first two points. The equivalence
of the first and last point is dual to this.
f g
Assume first that A B C is exact, that is Im f B is a kernel of
g. Thus any morphism x such that g ◦ x = 0 factors through Im f B, as
indicated in the following diagram.

X
fb0
x0
Q
A X Im f
Im f
0
f
xb0
A B g C
f

We form the pullback as indicated above. By Corollary 13.8 the morphism fb0
is epi.
Now assume conversely that the second point holds. In particular we can
find a commutative diagram

fb
Ker
[g Ker g

ι
b ι
f
A B g C
40 CHAPTER II. ADDITIVE AND ABELIAN CATEGORIES

where ι : Ker g B is a kernel of g, and fb is an epimorphism.

Then Ker g is the image of ι ◦ fb = f ◦ bι, and the inclusion of Ker g into B
factors through the inclusion of Im f . It follows that the inclusion of Im f into
Ker g, which exists since g ◦ f = 0, is an isomorphism, i.e. that the sequence is
exact.

Remark 13.10. In the category Mod R of modules over a ring we can determine
exactness using elements: A sequence of morphisms

f g
A B C

with g ◦ f = 0 is exact if for every element x ∈ B such that g(x) = 0 there is a

preimage, that is xb ∈ A such that f (bx) = x.
Proposition 13.9 now tells us that the same holds for arbitrary abelian cate-
x
gories, if we replace “element” by “morphism X ”, and “preimage” by “com-
mutative square with epimorphism”.
We will see this kind of substitution in practice in the next section.

14 Some diagram lemmas

Theorem 14.1 (Five lemma). Let A be an abelian category. Consider the
following commutative diagram with exact rows.

a1 a2 a3 a4
A1 A2 A3 A4 A5

f1 f2 f3 f4 f5
b1 b2 b3 b4
B1 B2 B3 B4 B5

• Assume f2 and f4 are monomorphisms, and f1 is an epimorphism. Then

f3 is a monomorphism.

• Assume f2 and f4 are epimorphisms, and f5 is a monomorphism. Then

f3 is an epimorphism.

In particular, if all of f1 , f2 , f4 , and f5 are isomorphisms, then so is f3 .

14. SOME DIAGRAM LEMMAS 41

Proof for the case Mod R.

First point: Let x ∈ A3 , such that f3 (x) = 0. Then f4 (a3 (x)) = b3 (f3 (x)) = 0,
and, since f4 is a monomorphism, a3 (x) = 0.
Thus there is a preimage x b of x in A2 . We see that b2 (f2 (bx)) = f3 (a2 (b
x)) =
[
f3 (x) = 0, and thus there is a preimage f2 (bx) of f2 (b
x) in B1 .
Since f1 is assumed to be an epimorphism we can find a preimage f] 2 (b
x) of
f[
2 (b
x ) in A 1 .
Now note that

f2 (a1 (f] x))) = b1 (f1 (f]

2 (b 2 (b
x))) = f2 (b
x),

and, since f2 is a monomorphism this implies a1 (f]

2 (b
x)) = x
b.
Thus x = a2 (b ]
x) = a2 (a1 (f2 (b
x))) = 0.
Second point: Let x ∈ B3 . Since f4 is epi there is x0 ∈ A4 such that f4 (x0 ) =
b3 (x).
We note that f5 (a4 (x0 )) = b4 (f4 (x0 )) = b4 (b3 (x)) = 0. Thus, since f5 is
mono, we have a4 (x0 ) = 0. It follows that there is x x) = x0 .
b ∈ A3 such that a3 (b
Next observe that

b3 (x − f3 (b
x)) = b3 (x) − b3 (f3 (b
x)) = b3 (x) − f4 (a3 (b
x)) = 0.
| {z }
=x0

Hence there is y ∈ B2 such that b2 (y) = x − f3 (b

x). Moreover, since f2 is epi,
there is yb ∈ A2 such that f2 (b
y ) = y.
Now we have that

f3 (b
x + a2 (b
y )) = f3 (b
x) + b2 (f2 (b x) + x − f3 (b
y )) = f3 (b x) = x,
| {z }
=y

showing that an arbitrary x lies in the image of f3 .

Proof for arbitrary abelian categories. We only prove the first point, the second
one is dual.
Let x : X A3 be a morphism such that f3 ◦ x = 0. Since

f4 ◦ a3 ◦ x = b3 ◦ f3 ◦ x = 0,
42 CHAPTER II. ADDITIVE AND ABELIAN CATEGORIES

and f4 is mono by assumption, we have a3 ◦ x = 0. Thus, by Proposition 13.9,

we obtain X,
b x
b, and an epimorphism ab2 as indicated in the following diagram.

1 ◦ f1
b\ ab2
X
bb X
b X

x x
2◦x
f\ b b
a1 a2
A1 A2 A3

f1 f2 f3
b1 b2
B1 B2 B3

Now note that b2 ◦ (f2 ◦ x

b) = 0, and that, since f1 is epi,
b1 ◦ f 1 b2
A1 B2 B3

f2 ◦ x 1 ◦ f1 as indicated
bb \
is exact. Hence we can find X, b, and an epimorphism b\
above, such that b1 ◦ f1 ◦ f2 ◦ x
\ b = f2 ◦ x
b ◦ b\
1 ◦ f1 . Since b1 ◦ f1 = f2 ◦ a1 , and f2
is a monomorphism by assumption, this implies
a1 ◦ f\
2◦x b ◦ b\
b=x 1 ◦ f1 ,

and thus
x ◦ ab2 ◦ b\
1 ◦ f1 = a2 ◦ a1 ◦f2 ◦ x
\b = 0.
| {z }
=0

Since ab2 ◦ b\
1 ◦ f1 is an epimorphism this means that x = 0.
Thus we have seen that f3 ◦ x = 0 impies x = 0, which means that f3 is a
monomorphism.
Theorem 14.2 (Characterization of pullback and pushout). In an abelian cate-
gory, consider a commutative square, together with its kernel and cokernel mor-
phisms as in the following diagram.

ι f π
Ker f A B Cok f

k g h c
λ i ρ
Ker i C D Cok i
14. SOME DIAGRAM LEMMAS 43

Then

• the square is a pullback if and only if k is an isomorphism and c is a

monomorphism;

• the square is a pushout if and only if k is an epimorphism and c is an

isomorphism.

Proof. We only prove the first claim, the second one is dual.
Assume first that the square is a pullback. We have already seen – in Propo-
sition 13.7 – that the kernel morphism k is an isomorphism. Let x : X Cok f
be a morphism such that c ◦ x = 0.

bi π
b
X
bb X
b X
b
ϕ
x
b x
f π
◦x
h[b A B Cok f

g h c
i ρ
C D Cok i

π
Since B Cok f 0 is exact, by Proposition 13.9, there are X,
b xb, and an
epimorphism π
b as indicated in the diagram.
i ρ
Since C D Cok i is exact, and ρ◦(h◦bx) = c◦x◦b
π = 0, Proposition 13.9
also implies the existence of X, h ◦ x
bb [
b, and an epimorphism bi as above.
We get ϕ as indicated above by the pullback property of the original square.
Thus x ◦ π
b ◦ bi = π ◦ f ◦ ϕ = 0 implies x = 0, and thus c is a monomorphism.

Now assume conversely that k is an isomorphism, and c a monomorphism.

We have to show that the square is a pullback.
We consider the pullback of i and h, and, by the pullback property, we get
44 CHAPTER II. ADDITIVE AND ABELIAN CATEGORIES

a map ϕ to it from A as indicated in the following diagram.

ι f π
Ker f A B Cok f

ϕ
k 00 idB c00
k z
Q
B C bi x
Ker bi B Cok bi c
∼
= D

g
k0 h
b h c0
λ i ρ
Ker i C D Cok i

By the other implication of this theorem, we know that k 0 is an isomorphism.

It follows that also k 00 is an isomorphism. Moreover, since c is mono, so is c00 .
It now follows from the five lemma (Theorem 14.1) that ϕ is an isomorphism.

Theorem 14.3 (Snake lemma). In an abelian category, consider (solid part of )

the following diagram with exact rows and columns

Ker f1 Ker f2 Ker f3

ι1 ι2 ι3
∂
a1 a2
A1 A2 A3 0
f1 f2 f3

0 B1 B2 B3
b1 b2
π1 π2 π3

Cok f1 Cok f2 Cok f3

Then there is a map ∂ : Ker f3 Cok f1 , such that the dashed sequence

∂
Ker f1 Ker f2 Ker f3 Cok f1 Cok f2 Cok f3

is exact.
14. SOME DIAGRAM LEMMAS 45

Q
Proof. Construction of ∂: Consider the pullback A2 Ker f3 , and the pushout
` A3
B2 Cok f1 . By Theorem 14.2 we have induced exact sequences
B1

ab1 Y ab2
A1 A2 Ker f3 Ker f3 0
A3

and
be1 a be2
0 Cok f1 B2 Cok f1 B3 ,
B1

as indicated in the following diagram.

k1 k2
Ker f1 Ker f2 Ker f3

Ker f3 ab2
Q
ι 2 A2
A3
ι1 ι3
ab1 ∂
ιb3

A1 A2 A3
a1 a2
f1 f2 f3

b1 b2
B1 B2 B3

π
f1
be2
π1 B2
`
Cok f1 π3
B1 π2
be1
Cok f1 Cok f2 Cok f3
c1 c2

We consider the composition

f1 ◦ f2 ◦ ιb3
π
in the middle of the diagram.
46 CHAPTER II. ADDITIVE AND ABELIAN CATEGORIES

Note that both

π1 ◦ f2 ◦ ιb3 ) ◦ ab1 = be1 ◦ π1 ◦ f1 = 0

(f

and
be2 ◦ (f
π1 ◦ f2 ◦ ιb3 ) = f3 ◦ ι3 ◦ ab2 = 0.
f1 ◦ f2 ◦ ιb3 factors through both ab2 and be1 , that is we can (uniquely) find
Thus π
∂ such that
be1 ◦ ∂ ◦ ab2 = π
f1 ◦ f2 ◦ ιb3 .

Exactness in Ker f2 : We first note that k2 ◦k1 = 0 by functoriality of kernels.

Now we aim to apply Proposition 13.9 to show exactness.
Let x : X Ker f2 such that k2 ◦ x = 0. Then clearly also a2 ◦ ι2 ◦ x = 0.
Hence, since the sequence A1 A2 A3 is exact, by Proposition 13.9, there
is an object X and morphisms ab1 and ι[ 2 ◦ x such that ab1 is epi, and ι[
2 ◦ x ◦ a1 =
b
ab1 ◦ ι2 ◦ x, as indicated in the following diagram.

ab1
X
b X
ϕ x
k1 k2
2◦x
ι[ Ker f1 Ker f2 Ker f3

ι1 ι2 ι3
a1 a2
A1 A2 A3

f1 f2

0 B1 B2
b1

We note that b1 ◦ f1 ◦ ι[
2 ◦ x = f2 ◦ ι2 ◦ x ◦ ab1 = 0. Since b1 is mono this implies
that f1 ◦ ι[
2 ◦ x = 0, so 2 ◦ x = ι1 ◦ ϕ for some ϕ as indicated by the dashed
ι[
arrow above. Since

ι2 ◦ k1 ◦ ϕ = a1 ◦ ι1 ◦ ϕ = ai ◦ ι[
2 ◦ x = ι2 ◦ x ◦ ab1 ,

and since ι2 is a monomorphism, it follows that also the upper square of the
diagram commutes. Now exactness in Ker f2 follows from Proposition 13.9.
14. SOME DIAGRAM LEMMAS 47

Exactness in Ker f3 : We begin by noting that ∂ ◦ k2 = 0: This follows form

be1 ◦ ∂ ◦ k2 = π
f1 ◦ f2 ◦ ι2 = 0, since be1 is a monomorphism.
Now we proceed showing exactness by using Proposition 13.9. Thus, let
y : Ker f3 Y such that y ◦ k2 = 0. We construct the following diagram form
top to bottom:

k2 y
Ker f2 Ker f3 Y

ι2 ι3 ιe3
a1 a2 ye
A1 A2 A3 Ye

f1 f2 fe2
b1 ye
e
B1 B2 Ye
e

π1 π
f1
ye
e
e
Cok f1 e
Ye
e

Here we used Proposition 13.9 thrice:

• Ye , ye, and a monomorphism ιe3 exist since ι3 is a mono;
ι2 f2
• Ye , e
ye, and a monomorphism fe2 exist because Ker f2 A2 B2 is exact,
e
and (e y ◦ a2 ) ◦ ι2 = ιe3 ◦ y ◦ k2 = 0;
e f1 π1
• Ye , e
ye, and a monomorphism π f1 exist because A1 B1 Cok f1 is exact,
e e

and (ye ◦ a1 ) ◦ f1 = f2 ◦ ye ◦ a2 ◦ a1 = 0.
e e

We now claim that

π1 ◦ fe2 ◦ ιe3 ) ◦ y = e
(f ye ◦ ∂.
e
Q
To check this, consider the epimorphism ab2 : A2 Ker f3 Ker f3 as above.
A3
Note that we have
∂ ◦ ab2 = π1 ◦ h,
Q
where h is the unique map A2 Ker f3 B1 such that b1 ◦ h = f2 ◦ ιb3 .
A3
48 CHAPTER II. ADDITIVE AND ABELIAN CATEGORIES

Now we can calculate

ye ◦ π1 ◦ h
ye ◦ ∂ ◦ ab2 = e
e
e e

=πf1 ◦ eye ◦ b1 ◦ h
f1 ◦ e
=π ye ◦ f2 ◦ ιb3
f1 ◦ fe2 ◦ ye ◦ a2 ◦ ιb3
=π
f1 ◦ fe2 ◦ ye ◦ ι3 ◦ ab2
=π
f1 ◦ fe2 ◦ ιe3 ◦ y ◦ ab2 .
=π

And the claim follows since ab2 is an epimorphism.

Now exactness of the snake sequence in Ker f3 follows from Proposition 13.9.
Exactness in Cok f1 and Cok f2 : Are dual to the two positions we have
already treated.

Remark 14.4 (Construction of ∂ for Mod R). For an element x of Ker f3 , let
x
b be a preimage of ι3 (x) in A2 . Then b2 (f2 (b
x)) = f3 (a2 (b
x)) = f3 (ι3 (x)) = 0.
Therefore f2 (b [ x) in B1 . We define ∂(x) = π1 (f[
x) has a preimage f2 (b 2 (b
x)).
One may check that this is well-defined.

15 Exercises
Exercise II.1. Let A be a pre-abelian category, and X be a finite poset. Show
that any F ∈ preshA X has a limit and a colimit.

Exercise II.2. Consider an additive category.

• In the situation of the biproduct diagram

ιX ιY

X X ⊕Y Y

πX πY

with idX = πX ◦ ιX , idY = πY ◦ ιY , and idX⊕Y = ιX ◦ πX + ιY ◦ πY : Show

that πY is a cokernel of ιX .
15. EXERCISES 49

• Let f : X Y be a split monomorphism, which has a cokernel. Show

that there is an isomorphism ϕ : Y X ⊕ Cok f making the triangle

f
X Y

∼
= ϕ
idX
0
X ⊕ Cok f

commutative.

Exercise II.3. Let A be a pre-abelian category. Show that the following are
equivalent:

• A is abelian;

• every monomorphism is a kernel of some morphism, and every epimor-

phism is a cokernel of some morphism.

Hint: First show that if a monomorphism is a kernel of some morphism, then

it is in fact a kernel of its cokernel.

Exercise II.4. In an abelian category, consider a morphism f : X Y with

its image Im f .

• Let e : W X be an epimorphism. Show that Im(f ◦ e) = Im f .

• Let m : Y Z be a monomorphism. Show that Im(m ◦ f ) = Im f .

• Assume f = m ◦ e, with e : X I epi and m : I Y mono. Show that

I = Im f .

Exercise II.5. Consider the poset

X={a b },
0
50 CHAPTER II. ADDITIVE AND ABELIAN CATEGORIES

and the morphism of Ab-valued presheaves on it

0
0 2
0 Z
0
0 Z Z Z.
Z 0
(Here “2” is short for “the map given by multiplication by 2”.) Check that this
is a morphism. Calculate the kernel, image, and cokernel of this morphism.
Exercise II.6. Show that, for an abelian category A and a poset X, the
presheaf category preshA X is abelian again.
Exercise II.7. In an abelian category, let f : A B and g : B C be mono-
morphisms. Show that there is a short exact sequence
0 Cok f Cok g ◦ f Cok g 0.
Remark: If we think “Cok f = B/A”, “Cok g ◦f = C/A”, and “Cok g = C/B”,
C/A ∼
then this exercise gives us the isomorphism theorem B/A = C/B.
Exercise II.8 (3 × 3 Lemma). Consider the following diagram with exact rows
and columns.
0 0 0

A B C

0 A0 B0 C0 0

0 A00 B 00 C 00 0

0 0 0
15. EXERCISES 51

Show that A, B, and C also form a short exact sequence fitting into the above
diagram.

Exercise II.9 (Salamander lemma). Consider the following diagram in ModR,

where γβ = 0 and δγ = 0.

A
α
β γ δ
B C D E


Show that the sequence

Ker γ Ker γ Ker δ ∩ Ker  Ker δ F
Ker γα
Im β Im α + Im β Im γα Im γ Im γ
is exact.
52 CHAPTER II. ADDITIVE AND ABELIAN CATEGORIES
Chapter III
N
Hom and

16 Hom, projectives and injectives

Let A be a preadditive category, and A ∈ Ob A . Then HomA (A, −) and
HomA (−, A) define (additive) functors form A to Ab (covariant and contravari-
ant, respectively).
Now let A be abelian. We want to investigate what the Hom-functors do
to short exact sequences.

Example 16.1. In Ab, consider the short exact sequence

·2
0 Z Z Z/(2) 0.

Applying HomAb (Z/(2), −) we obtain

0 0 0 Z/(2) 0.

Applying HomAb (−, Z/2) we obtain

id 0
0 Z/(2) Z/(2) Z/(2) 0.

In both cases we observe that the resulting sequence is not exact any more.
However, in both cases we may note that the left map is still the kernel of
the right map. We will now see that this is a general feature of Hom-functors.

53
 N
54 CHAPTER III. Hom AND

Theorem 16.2 (Hom is left exact). Let A be an abelian category, and let
A ∈ Ob A .
• Let 0 X Y Z be exact. Then also

0 HomA (A, X) HomA (A, Y ) HomA (A, Z)

is exact, that is HomA (A, −) preserves kernels.

• Let X Y Z 0 be exact. Then also

0 HomA (Z, A) HomA (Y, A) HomA (X, A)

is exact, that is HomA (−, A) turns cokernels into kernels.

Such functors are called left exact.
Proof. We only prove the first claim, the second one is the same for the category
A op .
We denote by f : X Y and g : Y Z the maps of the original sequence.
We first check that f ∗ = HomA (A, f ) is injective. Let ϕ ∈ HomA (A, X) such
that f ∗ (ϕ) = 0. By definition f ∗ (ϕ) = f ◦ ϕ, and since f is a monomorphism
this can only be zero if ϕ already is zero.
Next let ϕ ∈ Ker g ∗ , that is g ◦ ϕ = 0. Then, since f is the kernel of g, there
is a map ψ : A X such that ϕ = f ◦ ψ, i.e. ϕ = f ∗ (ψ) ∈ Im f ∗ .
Definition 16.3. A functor between two abelian categories is called exact if it
preserves short exact sequences. It is called right exact if it preserves cokernels.
Observation 16.4. For a functor F between two abelian categories the following
are equivalent:
• F is exact;
• F is left exact and maps epimorphisms to epimorphisms;
• F is right exact and maps monomorphisms to monomorphisms.
Definition 16.5. Let A be an abelian category.
• An object P is called projective if the functor HomA (P, −) is exact.
• An object I is called injective if the functor HomA (−, I) is exact.
16. Hom, PROJECTIVES AND INJECTIVES 55

Clearly injective objects in A are just projective objects in A op .

Example 16.6. In the category Mod R, the object R is projective: Indeed

the functor HomR (R, −) : Mod R Ab is just the forgetful functor, and hence
clearly is exact.

Observation 16.7. Direct sums and direct summands of projective objects are
projective (and similar for injective). The zero object is projective and injective.

Observation 16.8. Let A be an abelian category.

• An object P is projective if and only if any given diagram as the solid

part of the following, with exact row

X Y 0

can be completed to a commutative diagram by a morphism as indicated

by the dashed arrow.
(This is just a diagrammatic restatement of the fact that the functor
HomA (P, −) maps epimorphisms to epimorphisms.

• An object I is injective if and only if any given diagram as the solid part
of the following, with exact row

0 X Y

can be completed to a commutative diagram by a morphism as indicated

by the dashed arrow.

Recall that the free R-module on a set I is

R(I) = {f : I R | f (i) 6= 0 for only finitely many i ∈ I}.

 N
56 CHAPTER III. Hom AND

Theorem 16.9. Let R be a ring, P an R-module. Then the following are

equivalent:

• P is projective in Mod R;

• There is a module Q such that P ⊕ Q ∼

= R(I) for some set I.

Proof. =⇒: Consider the natural map π : R(P ) P (the counit of the adjunc-
tion). It clearly is an epimorphism, and, since P is projective, it splits. Therefore
P ⊕ Ker π ∼ = R(P ) .
⇐=: Since HomMod R (R(I) , −) = HomSet (I, −), this functor maps epimor-
phisms to epimorphisms. Hence R(I) is projective. It follows that also all direct
summands of R(I) are projective.

Remark 16.10. It follows that for any R-module M , there is an epimorphism

P M from a projective module. (Take for instance P = R(M ) .)
It is also possible to show (but a lot more technical) that for every R-module
M there is a monomorphism M I into an injective R-module.

17 Tensor products
Definition 17.1. Ler R be a ring, M a right R-module and N a left R-module.
A map ϕ : M × N A to an abelian group A is called R-balanced if

∀m ∈ M, n1 , n2 ∈ N : ϕ(m, n1 + n2 ) = ϕ(m, n1 ) + ϕ(m, n2 ),

∀m1 , m2 ∈ M, n ∈ N : ϕ(m1 + m2 , n) = ϕ(m1 , n) + ϕ(m2 , n),
∀m ∈ M, n ∈ N, r ∈ R : ϕ(mr, n) = ϕ(m, rn)

A tensor product is an abelian group M ⊗R N , together with an R-balanced

map t : M × N M ⊗R N , such that for any R-balanced ϕ : M × N A there
is a unique morphism of abelian groups h : M ⊗R N A such that ϕ = h ◦ t.
In this situation we write m ⊗ n = t(m, n), and call it an elementary tensor .
Note that there is no reason for t to be surjective in general, that is not all
elements of the tensor product need to be elementary tensors.

Theorem 17.2. Tensor products exist and are unique up to isomorphism.

Proof. Uniqueness can be shown similarly to the proof of uniqueness of limits

and colimits (see Proposition 7.3).
17. TENSOR PRODUCTS 57

To prove existence we explicitly construct a tensor product. We start by

considering the free abelian group F = Z(M ×N ) . We have seen in Example 6.3
that
HomAb (F, A) = HomSet (M × N, A).
Now the idea of the proof is that we alter F in such a way that in the right hand
side only the R-balanced maps remain. We denote by U the abelian subgroup
of F generated by all expressions of the form
χ(m,n1 +n2 ) − χ(m,n1 ) − χ(m,n2 ) ,
χ(m1 +m2 ,n) − χ(m1 ,n) − χ(m2 ,n) , and
χ(mr,n) − χ(m,rn) .
Then it is immediately verified that F/U is a tensor product.
Observation 17.3. The above construction shows that, while not all elements
of the tensor product are elementary tensors themselves, they are finite sums of
elementary tensors.
Example 17.4. Note that both individual elementary tensors and entire tensor
products can be zero, even if they don’t “look like it”:
Consider Z/(2) ⊗Z Z/(3). Take an elementary tensor (a + (2)) ⊗ (b + (3)).
Then
(a + (2)) ⊗ (b + (3)) = (a + (2)) ⊗ 2(2b + (3))
= (a + (2))2 ⊗ (2b + (3))
= 0 ⊗ (2b + (3))
= 0 ⊗ 0(2b + (3))
=0⊗0
Thus all elementary tensors vanish, and hence the entire tensor product is zero.
Construction 17.5. Let f : M1 M2 be a morphism of right R-modules, and
N be a left R-module. Then the composition along the top and right of the
following diagram is R-balanced.
f × idN
M1 × N M2 × N

M1 ⊗R N M 2 ⊗R N
 N
58 CHAPTER III. Hom AND

Thus there is a unique map as indicated by the dashed arrow above, making
the diagram commutative. We denote this map by f ⊗R N . One immediately
verifies that
− ⊗R N : Mod R Ab
defines a functor.
Similarly, for a right R-module M one obtains a functor

M ⊗R − : Mod Rop Ab.

Example 17.6. Let R be any ring, and M ∈ Mod R. Then

M ⊗R R ∼
= M.

Indeed the map M ×R M : (m, r) mr is clearly R-balanced, thus induces a

homomorphism M ⊗R R M . An inverse is given by M M ⊗R R : m m⊗
1.

18 Hom-tensor adjunction
Let M be an R-S-bimodule. Then for any R-module L, the tensor product
L ⊗R M becomes an S-module via (l ⊗ m)s = l ⊗ ms. In fact we obtain a
functor
− ⊗R M : Mod R Mod S.
Similarly we have the functor

HomS (M, −) : Mod S Mod R,

where, for an S-module N , the R-module structure on HomS (M, N ) is given by

ϕ · r = ϕ(r · −).
The following result shows that these two functors are in fact adjoint.

Theorem 18.1. Let L be an R-module, M be an R-S-bimodule, and N be an

S-module. Then

HomS (L ⊗R M, N ) ∼
= HomR (L, HomS (M, N )),

and this isomorphism is natural in all arguments.

18. HOM-TENSOR ADJUNCTION 59

Proof. We have mutually inverse maps given by

HomS (L ⊗R M, N ) ∼
= HomR (L, HomS (M, N ))
ϕ [` ϕ(` ⊗ m)]
[` ⊗ m ψ(`)(m)] ψ

Corollary 18.2. Let M be an R-S-bimodule. Then the functor

− ⊗R M : Mod R Mod S

is right exact.

Proof. Since the functor is left adjoint to HomS (M, −) it commutes with all
colimits. But cokernels are certain colimits.

Remark 18.3. The above argument also shows that tensor products commute
with (infinite) coproducts.

Definition 18.4. A left R-module M is called flat if the tensor functor − ⊗R

M : Mod R Ab is exact.

Observation 18.5. • The R-module R is flat, since, by Example 17.6 ten-

soring with R is essentially identity.

• Any free R-module is flat, since, by Remark 18.3 tensoring commutes

with coproducts (which are special colimits), and since the coproduct of
a collection of exact sequences is exact again.

• Any projective R-module is flat, since it is a direct summand of a free

R-module by Theorem 16.9.

Remark 18.6. The converse of the last point above does not hold. For instance
Q is a flat Z module which is not projective.
However, for certain nice rings (for instance finite dimensional algebras over
a field), all flat modules are projective.
 N
60 CHAPTER III. Hom AND

19 Exercises
Exercise III.1. Let X be a poset, and F a field. By Exercise II.6 the category
preshmod F X is abelian.
For i ∈ X, we consider the special presheaves Pi and Ii given by
( (
F if j 6 i k if j > i
Pi (j) = and Ii (j) = .
0 otherwise 0 otherwise

Path I:
Verify directly: Pi is projective and Ii is injective in preshmod F X.
Path II:
Consider inclusion ι : {i} X. Show that the induced functor

ι∗ : preshmod F X preshmod F {i} = mod F

has a left adjoint L and a right adjoint R.

Check that
• ι∗ is exact;
• Pi = LF and Ii = RF;
• F is both projective and injective in mod F.
Conclude that Pi is projective and Ii is injective.
Independent of path: For X = {a > 0 < b}: Find a projective object P and
an epimorphism P I0 in preshmod F X.
Exercise III.2. Show that

Z/(m) ⊗Z Z/(n) = Z/(gcd(m, n)).

Hint: Recall that (gcd(m, n)) = (m, n).

Exercise III.3. Let L ∈ Mod R, M an R-S-bimodule, and N ∈ Mod S op .
Show
N∼
O O O O
(L M) =L (M N ).
R S R S

Hint: On elementary tensors, the map from left to right should send (l ⊗m)⊗n
to l ⊗ (m ⊗ n). The main issue is to show that this gives a well-defined map.
19. EXERCISES 61

Exercise III.4. Let A be an abelian category, and A an object in A . Convince

yourself that HomA (A, −) defines a functor A Mod R, where R = EndA (A).
Now assume that for any object X ∈ A , the R-module HomA (A, X) is
finitely generated. Show (without using the Freyd-Mitchell embedding theo-
rem), that HomA (A, −) : A mod R has a left adjoint.
Hints:
• Show that HomA (A, −) induces an equivalence between the subcategories
{An | n ∈ N} ⊆ A and {Rn | n ∈ N} ⊆ mod R.

• Show that HomA (A, X) is finitely presented as R-module.

 N
62 CHAPTER III. Hom AND
Chapter IV

Complexes and homology

20 The long exact sequence of homology

Definition 20.1. Let A be an abelian category. A (cochain) complex in A is
a sequence of objects and morphisms

d−2 d−1 d0 d1
A• = · · · A−1 A0 A1 ···
such that di ◦ di−1 = 0 for all i ∈ Z.
We denote by C(A ) the category of all complexes in A , where morphisms
are given by
HomC(A ) (A• , B • ) = {(f i )i∈Z |f i ∈ HomA (Ai , B i ) such that
f i ◦ di−1 i−1
A = dB ◦ f
i−1
∀i ∈ Z},
that is morphisms are commutative diagrams

d−2
A d−1
A d0A d1A
··· A−1 A0 A1 ···

f −1 f0 f1
d−2
B d−1
B d0B d1B
··· B −1 B0 B1 ···

Note that the category C(A ) is also abelian, with kernels and cokernels
being calculated position by position.

63
64 CHAPTER IV. COMPLEXES AND HOMOLOGY

Remark 20.2. One defines chain complexes in a very similar way, just using
lower indices and counting down. This distinction comes from the origins of
homological algebra in algebraic topology, where the index often is the dimension
of the objects involved. Thus it is natural that the boundary of an n-dimensional
object is n − 1-dimensional (chain complex) and not the other way around (thus
called cochain complex).
However in our course a complex is just an abstract sequence of objects and
maps, and thus the difference between counting up and counting down is of no
concern to us.

Definition 20.3. For a complex A• , and n ∈ Z, we set

Bn (A• ) = Im dn−1 and Zn (A• ) = Ker dn ,

called the n-boundaries and n-cycles, respectively.

Note that since, by definition, dn ◦ dn−1 = 0, the inclusion Bn (A• ) An
factors through the inclusion Zn (A• ) An . We denote by Hn (A• ) the cokernel
of this map Bn (A• ) Zn (A• ), and call it n-th homology of A• .
Note that all three of these constructions are functorial.

Remark 20.4. In case that our abelian category A is in fact a category of

modules (or any other category where it makes sense to talk about ‘elements’
of the objects) the above just means that Bn (A• ) and Zn (A• ) are submodules
of An , such that Bn (A• ) ⊆ Zn (A• ). Now homology is the quotient

Hn (A• ) = Zn (A• )/Bn (A• ).

There are obvious duals to the definition of boundaries, cycles, and homol-
ogy. (These are not what is called coboundaries, cocycles, and cohomology –
coboundaries are just the same as boundaries, but distinguishing between count-
ing up and counting down, see Remark 20.2.) However the next lemma tells us
that for homology it does not matter if we take this definition or its dual.

Lemma 20.5. Let A• be a complex in an abelian category. Then the epimor-

phism An Bn+1 (A• ) factors through the epimorphism An Cok dn−1 , and

Hn (A• ) = Ker[Cok dn−1 Bn+1 (A• )].

Proof. The factorization follows form dn ◦ dn−1 = 0.

20. THE LONG EXACT SEQUENCE OF HOMOLOGY 65

Now consider the following diagram, where K denotes the kernel of the
lemma.
K

Bn (A• ) An Cok dn−1

Zn (A• ) An Bn+1 (A• )

Hn (A• )

By the snake lemma we have a snake morphism as indicated by the dashed

arrow, and it is an isomorphism since both its kernel and cokernel are zero.

Theorem 20.6 (Long exact sequence of homology). Let A• B• C • be a

short exact sequence in C(A ), for some abelian category A . Then there is a
long exact sequence

··· Hn (A• ) Hn (B • ) Hn (C • ) Hn+1 (A• ) Hn+1 (B • ) ··· .

Proof. Note that Zn is left exact, and Cok dn is right exact. Thus we get exact
sequences in the rows of the following commutative diagram.

Cok dn−1
A Cok dn−1
B Cok dn−1
C 0

0 Zn+1 (A• ) Zn+1 (B • ) Zn+1 (C • )

where the vertical maps are induced by the maps dnA , dnB , and dnC , respectively.
In particular they are the composition Cok dn−1 A Bn+1 (A• ) Zn+1 (A• ), and
• •
similar for B and C . Thus their cokernels are H n+1
(A ), Hn+1 (B • ), and
•

Hn+1
(C ), respectively. Moreover, by Lemma 20.5, the kernels are Hn (A• ),
•

H (B • ), and Hn (C • ), respectively. Now the claim follows from the snake lemma.
n
66 CHAPTER IV. COMPLEXES AND HOMOLOGY

21 Cones and quasi-isomorphisms

In the setup of Theorem 20.6, one easily sees that the maps Hn (A• ) Hn (B • )
and Hn (B • ) Hn (C • ) are just homologies of the original maps in C(A ). How-
ever the maps Hn (C • ) Hn+1 (A• ) are induced by the snake morphism of the
snake lemma, and their description is less explicite. We make it explicite here
in the special case that all the short exact sequences An Bn C n are split.

Observation 21.1. Let A• B• C • be a short exact sequence in C(A ),

such that
∀n : B n = An ⊕ C n
and the maps are given by ( 10 ) and (0 1), respectively.
n n

Then dnB is given by a 2 × 2-matrix, say abn fcn .
In order for the first map to be a morphism of complexes we need
dnB ◦ ( 10 ) = ( 10 ) ◦ dnA ,
that is an = dnA and bn = 0. Similarly, in order for the second map to be a
morphism of complexes we need bn = 0 and cn = dnC .
Finally we require dnB ◦ dn−1
B = 0, with the above that gives
 n n   n−1 n−1   n n−1 n n−1 
d f d f
0 = 0A dn ◦ A0 dn−1 = 00 f ◦dC +d 0
A ◦f
C C

n
that is f ◦ (−dn−1
C )= ◦f dnA
. n−1

Conversely we see that any family (f n )n∈Z with this property gives rise to
a short exact sequence as above.
Definition 21.2 (Shift). Let A• be a complex. We denote by A• [n] the complex
obtained from A• by shifting every term n places to the left, that is with
(A• [n])i = Ai+n , and diA• [n] = (−1)n dn+i
A• .

Clearly [n] defines an autoequivalence of C(A ), with inverse [−n].

Also note that Hi (A• [n]) = Hi+n (A• ).

Definition 21.3 (Cone). Let f • : A• B • be a morphism in C(A ). Then the

cone Cone(f • ) is the complex

d−2 f −1 d−1
 
f0
 
d0B f 1
 
B B
0 −d−1A 0 −d0A 0 −d1A
··· B −1 ⊕ A0 B 0 ⊕ A1 ··· .
21. CONES AND QUASI-ISOMORPHISMS 67

By Observation 21.1 above we note that there is a degree-wise split short

exact sequence

( 10 ) (0 1)
•
B Cone(f • ) A• [1].

Moreover any degree-wise split short exact sequence is of this form.

Theorem 21.4. Let f • : A• B • be a morphism in C(A ). The long exact

sequence of homology associated to the short exact sequence

( 10 ) (0 1)
B• Cone(f • ) A• [1]

is

Hn (f • ) Hn ( 10 ) Hn (0 1)
• • •
··· n
H (A ) n
H (B ) n
H (Cone(f )) Hn+1 (A• ) ···

Proof. The fact that the second and third map are just the homologies of the
maps of complexes we started with follows immediately from the construction
of the long exact sequence. We need to check that the first map is indeed the
n-th homology of the map f • .

We follow the construction. To do so, we consider the following diagram

with exact rows, but not columns, where the middle part is just the diagram
68 CHAPTER IV. COMPLEXES AND HOMOLOGY

from the proof of Theorem 20.6.

( 01 ) Zn+1 (A• )
B n ⊕ An+1

0 Bn Cone(f • )n An+1 0

Hn+1 (A• )

Cok dn−1
B
Cok dn−1
Cone(f • ) Cok dnA 0

0 Zn+1 (B • ) Zn+1 (Cone(f • )) Zn+2 (A• )

Hn+1 (B • )
0 B n+1 Cone(f • )n+1 An+2 0

Cok dnB (1 0) B n+1 ⊕ An+2

Note that the composition along the columns are just the differentials dnB ,
dn n+1

B f
0 −dn+1
, and −dn+1
A , respectively.
A
Now recall the construction of the snake map from the snake lemma: Us-
ing the splitting indicated by the dashed arrows above, we first consider the
composition
 n n+1 
dB f
( 01 ) 0 −dn+1
A
Zn+1 (A• ) An+1 B n ⊕ An+1 B n+1 ⊕ An+2
(1 0)
B n+1 Cok dnB .
Multiplying the matrices we see that this is the composition

f n+1
Zn+1 (A• ) An+1 B n+1 Cok dnB .
Thus the induced map on homology is

Hn+1 (f • ) : Hn+1 (A• ) Hn+1 (B • ).

22. HOMOTOPY 69

Definition 21.5. A morphism f • : A• B • in C(A ) is called quasi-isomor-

n •
phism if H (f ) is an isomorphism for all n.

Corollary 21.6. Let f • : A• B • be a morphism in C(A ). Then f • is a

quasi-isomorphism if and only if the complex Cone(f • ) is exact.

22 Homotopy
Definition 22.1. A morphism f • : A• B • in C(A ) is called null-homotopic
if there are morphisms

hn ∈ HomA (An , B n−1 ) n ∈ Z

such that
∀n ∈ Z : f n = dn−1
B ◦ hn + hn+1 ◦ dnA .
Two morphisms f • and g • in HomC(A ) (A• , B • ) are called homotopic if f • −
g • is null-homotopic.

e• f• g•
Lemma 22.2. Let A• B• C• D• be morphisms in C(A ). If f • is
null-homotopic, then so is the composition g • ◦ f • ◦ e• .

Proof. By definition we have maps hi such that f n = dn−1

C ◦ hn + hn+1 ◦ dnB .
n n−1 n n
We choose h = g
e ◦ h ◦ e . Then

dn−1
D ◦e hn+1 ◦ dnA = dn−1
hn + e ◦ g n−1 ◦hn ◦ en + g n ◦ hn+1 ◦ en+1 ◦ dnA
| D {z } | {z }
=g n ◦dn−1 =dn
B ◦e
n
C

= g n ◦ (dn−1
C ◦ hn + hn+1 ◦ dnB ) ◦ en
= g n ◦ f n ◦ en

Definition 22.3. Let A be an abelian category. The homotopy category K(A )

is given by

Ob K(A ) = Ob C(A ) and

HomC(A ) (A• , B • )
HomK(A ) (A• , B • ) =
homotopy
70 CHAPTER IV. COMPLEXES AND HOMOLOGY

that is morphisms are considered the same if their difference is null-homotopic.

Lemma 22.2 shows that this indeed is a category, by making sure that mul-
tiplication of morphisms is well-defined.
It follows from the definition that K(A ) inherits the structure of an additive
category from C(A ) - the Hom-sets are by definition quotient abelian groups.
However K(A ) will typically not be abelian.

Proposition 22.4. Let f • : A• B • be null-homotopic. Then Hn (f • ) = 0 for

all n ∈ Z.
In particular the Hn define functors K(A ) A.

Proof. By assumption there are hn such that f n = dn−1B ◦ hn + hn+1 ◦ dnA .

First let ιA : Zn (A• ) An , and similar for ιB . Then Zn (f • ) is defined by

ιB ◦ Zn (f • ) = f n ◦ ιA

Inserting the above formula for f n we obtain that this is equal to

dn−1
B ◦ hn ◦ ιA + hn+1 ◦ dnA ◦ ιA = dn−1
B ◦ hn ◦ ιA .
| {z }
=0

Now note that this map clearly factors through Bn (B • ) B n , and thus the
induced map on homology vanishes.

23 Projective and injective resolutions

Definition 23.1. An abelian category A has enough projectives if for any
A ∈ Ob A there is an epimorphism P A from a projective object P to A.
Dually A has enough injectives if for any A ∈ Ob A there is a monomor-
phism A I from A to some injective object I.

Example 23.2. Let R be a ring. The category Mod R has enough projectives
and enough injectives.

Proposition 23.3. Let (X, 6) be a finite partially ordered set, and A an abelian
category.
If A has enough projectives, then so does preshA X. Dually, if A has
enough injectives, then so does preshA X.
23. PROJECTIVE AND INJECTIVE RESOLUTIONS 71

Proof. We show that preshA X has enough projectives, the claim about injec-
tives is dual. For a A ∈ Ob A and i ∈ X we define a presheaf PiA by
(
A A if j 6 i
Pi (j) = .
0 otherwise.
One easily sees that
HompreshA X (PiA , M ) = HomA (A, M (i)).
Therefore PiA is projective provided A is projective in A .
Now let M be an arbitrary A -valued presheaf on X. For i ∈ X, let
Ai M (i) be an epimorphism from a projective object in A . We set P =
L Ai
i∈X Pi . Then P is projective, and there is an epimorphism P M in
preshA X.
Definition 23.4. Let A be an abelian category with enough projectives, and
let A ∈ A . A projective resolution of A is a complex

d−2 d−1
··· P −2 P −1 P0 0 0 ···
with projective terms, which is exact, except in postion 0, where Cok d−1 = A.
Dually, if A has enough injectives then an injective resolution of A is a
complex
d0 1 d1 2
··· 0 0 I0 I I ···
with injective terms, which is exact, except in position 0, where Ker d0 = A.
Observation 23.5. Let A be an abelian category with enough projectives.
Then any object A ∈ A has a projective resolution. This can be constructed
iteratedly: Start with an epimorphism P 0 A, and call A−1 its kernel. Given
Ai , take an epimorphism P i Ai , and call Ai−1 its kernel. Concatenating
these short exact sequences we obtain a projective resolution.
Dually, if A has enough injectives, then any object has an injective resolu-
tion.
Construction 23.6. Let A and B be objects in an abelian category having
enough projectives. Let PA• and PB• be projective resolutions of A and B, respec-
f
tively. Given a morphism A B, we construct (non-canonically) a morphism
• • • 0 •
Pf : PA PB such that H (Pf ) = f :
72 CHAPTER IV. COMPLEXES AND HOMOLOGY

In the diagram below we construct the vertical morphisms from right to left,
starting with the given morphism f , such that everything commutes.

··· PA−2 PA−1 PA0

A−2 A−1 A

f
··· PB−2 PB−1 PB0

B −2 B −1 B

Here we obtain the morphisms PA−n PB−n using that PA−n is projective, and
thus the composition PA−n A−n B −n may be factored through the epi-
−n
morphism PB B −n . The morphisms A−n B −n are kernel morphisms.
Theorem 23.7. Let A have enough projectives. Then taking projective reso-
lutions defines a functor
p: A K(A ),
such that H0 ◦ p = idA and Hn ◦ p = 0 for n 6= 0.
Dually, if A has enough injectives we can define an injective resolution
functor
i: A K(A ),
such that H0 ◦ i = idA and Hn ◦ i = 0 for n 6= 0.
Proof. We have to show that there is a unique map Pf• : PA• PB• as in the
construction above, for any f : A B. Taking differences it is enough to show
this for f = 0.
So consider the solid part of the following commutative diagram

··· PA−2 PA−1 PA0

f −2 h−1 f −1 h0 A
f0

0
··· PB−2 PB−1 PB0

B −2 B −1 B
23. PROJECTIVE AND INJECTIVE RESOLUTIONS 73

Since the composition of f 0 with the epimorphism PB0 B vanishes we see that
f 0 factors as indicated by the rightmost dotted map above. Moreover, since PA0
is projective, we can lift this dotted map along the epimorphism PB−1 B −1
0 −1 0 0
to obtain a map h as above such that dPB ◦ h = f .
Now observe that d−1PB ◦ (f
−1
− h0 ◦ d−1 −1
PA ) = dPB ◦ f
−1
− f 0 ◦ d−1
PA = 0. Thus
f −1 −h0 ◦d−1
PA factors through the kernel B
−2
PB−1 as indicated by the second
−1
dotted arrow. Since PA is projective we may lift this along the epimorphism
PB−2 B −2 , and obtain a morphism h−1 such that d−2PB ◦ h
−1
= f −1 − h0 ◦ d−1
PA ,
or, in other words,
f −1 = h0 ◦ d−1 −2
PA + dPB ◦ h
−1
.
We iterate this construction to obtain a homotopy, thus showing that the map
of complexes we started with is in fact null-homotopic.

Proposition 23.8 (Horseshoe lemma). Let A B C be a short exact se-

quence in an abelian category. Assume PA• and PC• are projective resolutions of A
and C, respectively. Then there is a projective resolution PB• with PBi = PAi ⊕PCi ,
such that the following diagram commutes:

··· PA−2 PA−1 PA0 A

( 10 ) ( 10 ) ( 10 )

··· PA−2 ⊕ PC−2 PA−1 ⊕ PC−1 PA0 ⊕ PC0 B

(0 1) (0 1) (0 1)

··· PC−2 PC−1 PC0 C

Remark 23.9. In other words, the horseshoe lemma says that PB• may be
chosen as the cone of a certain map from PC• [−1] to PA• .
Proof. It suffices to consider the first step, and then iterate. Let us denote the
a b
given maps by A B C, and πA : PA0 A and πC : PC0 C. Since PC0 is
projective there is a map π˜C : PC0 B such that b ◦ π˜C = πC . It follows that
0 0
(a ◦ πA π˜C ) is a map PA ⊕ PC B making the right part of the diagram above
commutative. It follows from the five lemma (Theorem 14.1) that this also is
an epimorphism.
74 CHAPTER IV. COMPLEXES AND HOMOLOGY

Finally note that, by the snake lemma, the kernels also form a short exact
sequence, so we may iterate the argument.

24 Exercises
Exercise IV.1. Consider the poset
ω

X={a b },
0
and the complex of Ab-valued presheaves on it

0 3
0 2 0 0
0 Z Z
2 0
0 0 Z Z Z 0 0 0.

Z Z 0

(Here “2” is short for “the map given by multiplication by 2”.)

Calculate all homologies of this complex.
Exercise IV.2. Let A be an abelian category. Consider a morphism of com-
plexes over A , of the form

··· 0 0 A0 0 ···

··· 0 B −1 B0 0 ···

Assume this is a quasi-isomorphism.

Show that B 0 ∼= B −1 ⊕ A0 , such that the two non-zero maps above are the
canonical inclusions.
Exercise IV.3. Consider the X as below, and a field F. Calculate a projective
resolution of Iω in preshmod F X (see Exercise III.1).
24. EXERCISES 75

1. For X = {0 < ω};

ω

2. For X = { a b }.
0

Exercise IV.4. Let A• ∈ C(A ) be a complex such that Hi (A• ) = 0 for all
negative i. Show that there is a complex B • such that B i = 0 for all negative
i, and a quasi-isomorphism A• B•.
i i
Hint: You can take B = A for all positive i. What is a good choice for
B0?

Exercise IV.5. Let f • : A• B • be a morphism of complexes. Show that

B• Cone(f • ) is a weak cokernel of f • in the homotopy category K(A ).
f•
That is, the composition A• B• Cone(f • ) is zero, and for any mor-
phism g • : B • C • such that g • ◦ f • = 0 in K(A ) there is a (not necessarily
unique) factorization

f•
A• B• Cone(f • )

g•
C•

Exercise IV.6. Let A be an abelian category. Let A• ∈ C(A ). Show that

A• ∼
= 0 in K(A ) if and only if A• is isomorphic to a complex of the form

( 01 00 ) ( 01 00 ) ( 01 00 ) ( 01 00 )
−1 0 0 1 1 2
··· B ⊕B B ⊕B B ⊕B ··· .

Hint: For the “only if” direction, consider the short exact sequences

0 Bn (A• ) An Bn+1 (A• ) 0.

76 CHAPTER IV. COMPLEXES AND HOMOLOGY
Chapter V

Derived functors

25 Definition and first properties

Let A and B be abelian categories, and F : A B an (additive) functor. Then
F also induces a functor FK : K(A ) K(B). We use this construction to define
derived functors.

Definition 25.1. Let F : A B be a right exact functor. Assume that A has

enough projectives. Then we define the n-th left derived functor of F by

Ln F = H−n ◦ FK ◦ p.

That is, we take a projective resolution of the object, apply our functor to this
projective resolution, and then consider the homology groups of the result.
Dually, if F : A B is left exact and A has enough injectives, we can
construct right derived functors as

Rn F = Hn ◦ FK ◦ i.

Lemma 25.2. Let A have enough projectives, and let F : A B be right

exact. Then L0 F is naturally isomorphic to F.
Dually, if A has enough injectives and the functor is left exact, then R0 F ∼
=
nat
F.

Proof. We follow the definition of L0 F:

77
78 CHAPTER V. DERIVED FUNCTORS

d
Let A ∈ Ob A , and let · · · P −1 P 0 be a projective resolution of A.
Then A = Cok d. We apply F, and see that

Fd
L0 FA = H0 (· · · FP −1 FP0 ) = Cok Fd.

But since F is right exact we have

Cok Fd ∼
= F(Cok d) = FA.

Lemma 25.3. Let A have enough projectives, and let F : A B be exact.

Then Ln F = 0 for all non-zero n.
Dually, if A has enough injectives and the functor is exact, then Rn F = 0
except for n = 0.

Proof. Since the functor is exact it commutes with taking homology. That is
(
−n −n F ◦ idA if n = 0
Ln F = H ◦ FK ◦ p = F ◦ H ◦ p =
F◦0 if n 6= 0

Example 25.4. For the functors Hom and ⊗ the derived functors have special
names:

ExtnA (A, −) = Rn HomA (A, −)

ExtnA (−, A) = Rn HomA (−, A)
TorR
n (M, −) = Ln (M ⊗R −)

TorR
n (−, N ) = Ln (− ⊗R N )

In the second line, note that we consider HomA (−, A) as a left exact func-
tor A op Ab. In particular we calculate the derived functors by taking an
injective resolution in A op , that is a projective resolution in A .
We will see later that ExtnA (A, −)(B) = ExtnA (−, B)(A), and will simply
denote this by ExtnA (A, B). (And similar for Tor.)

Example 25.5. We calculate TorZi (Z/(n), −)(Z/(m)):

We start with a projective resolution of Z/(m). The simplest one is given
·m
by 0 Z Z 0.
25. DEFINITION AND FIRST PROPERTIES 79

Now we apply the (non-derived) functor Z/(n)⊗Z −, and obtain the complex
·m
0 Z/(n) Z/(n) 0. The kernel of the non-zero map here is

(n/gcd(m, n))/(n) ∼
= Z/(gcd(m, n)),

and the cokernel is also Z/(gcd(m, n)). Thus

(
Z/(gcd(m, n)) if i ∈ {0, 1}
TorZi (Z/(n), −)(Z/(m)) =
0 otherwise.

Theorem 25.6. Let A have enough projectives, and let F : A B be right

exact.
For any short exact sequence A B C in A there is a long exact se-
quence

··· L2 FC L1 FA L1 FB L1 FC FA FB FC 0

in B.

Proof. By the horseshoe lemma (Proposition 23.8) we may find projective res-
olutions of A, B, and C fitting into a diagram as follows:

··· PA−2 PA−1 PA0 A

( 10 ) ( 10 ) ( 10 )

··· PA−2 ⊕ PC−2 PA−1 ⊕ PC−1 PA0 ⊕ PC0 B

(0 1) (0 1) (0 1)

··· PC−2 PC−1 PC0 C

80 CHAPTER V. DERIVED FUNCTORS

Now, applying F to these projective resolutions, we obtain the diagram

··· FPA−2 FPA−1 FPA0

( 10 ) ( 10 ) ( 10 )

··· FPA−2 ⊕ FPC−2 FPA−1 ⊕ FPC−1 FPA0 ⊕ FPC0

(0 1) (0 1) (0 1)

··· FPC−2 FPC−1 FPC0

Note that while F is not exact, it does preserve direct sums and split short exact
sequences.
Now the long exact sequence of the theorem is just the long exact sequence
of homology (Theorem 20.6).

26 Syzygies and dimension shift

Observation 26.1. Let F : A B be right exact, and assume that A has
enough projectives.
Then Li FP = 0 for all non-zero i and any projective P . (To see this, note
that 0 P 0 is a projective resolution.)
The aim of this section is to combine this observation with the long exact
sequence of derived functors.
Definition 26.2. Let A be abelian with enough projectives. For an object A,
we construct a syzygy of A as the kernel of an epimorphism from a projective
object to A, and denote it by ΩA. That is, by definition we have a short exact
sequence
0 ΩA P A 0
with P projective.
Remark 26.3. Note that ΩA is not uniquely determined by A: different epi-
morphisms from projectives may give different syzygies.
In particular Ω is not a functor. (It may be applied to morphisms, but this
again involves making choices.)
26. SYZYGIES AND DIMENSION SHIFT 81

It can be seen that Ω defines an auto-functor of the quotient category

A
morphisms factoring through
.
projective objects

Definition 26.4. Dually, if A has enough injectives, we define the cosyzygy of

an object A to be the cokernel of a monomorphism of A into an injective object.
The cosyzygy will be denoted by fA.
Remark 26.5. It is more usual to denote cosyzygies by Ω−1 . However is should
be noted that syzygy and cosyzygy are in general not mutually inverse to each
other, which this notation seems to suggest.
Theorem 26.6 (Dimension shift). Let F : A B be right exact, and assume
that A has enough projectives. Let A ∈ Ob A . Then
Ln FA = Ln−1 F(ΩA) ∀n > 2.

Moreover, given a short exact sequence ΩA P A with P projective, we

have
L1 FA = Ker[F(ΩA) FP ].

Proof. We consider the short exact sequence ΩA P A, and the long exact
sequence of derived functors associated to it. For n > 2 we obtain

0 = Ln FP Ln FA Ln−1 F(ΩA) Ln−1 FP = 0,

and thus the first claim.
For n = 1 we have the exact sequence

0 = L1 FP L1 FA F(ΩA) FP,
and thus the second claim.
We also have the dual for left exact functors:
Theorem 26.7. Let F : A B be left exact, and assume that A has enough
injectives. Let A ∈ Ob A . Then
Rn FA = Rn−1 F(fA) ∀n > 2.

Moreover, given a short exact sequence A I fA with I injective, we have

R1 FA = Cok[FI F(fA)].
82 CHAPTER V. DERIVED FUNCTORS

27 Ext1 and extensions

Let A be an abelian category, and A and B objects. We denote by E the
collection of all short exact sequences

E: 0 B E A 0

for some E.
We consider two short exact sequences E1 and E2 equivalent if there is a
commutative diagram

0 B E1 A 0
ϕ

0 B E2 A 0

Note that by the five lemma the map ϕ necessarily is an isomorphism. Using
this fact one may see that the above definition indeed gives rise to an equivalence
relation.

Definition 27.1. The Yoneda-Extension group is the collection of equivalence

classes
YExt1A (A, B) = E / ∼ .

To explain why this is a group, we first discuss that it is functorial in both

A and B:

Construction 27.2. Let f : B1 B2 . Then taking pushouts gives a map

1 1
YExtA (A, B1 ) YExtA (A, B2 ), denoted by f · −:

E: 0 B1 E A 0

`
f · E: 0 B2 B2 B1 E A 0

Here we use that pushouts of monos are mono, and have the same cokernel.
27. Ext1 AND EXTENSIONS 83

Dually, if g : A1 A2 , taking pullbacks gives a map

− · g : YExt1A (A2 , B) YExt1A (A1 , B).

It is possible to see that these constructions commute: (f · E) · g = f · (E · g).

Hence we may omit brackets in this setup.

Definition 27.3 (Baer sum). Let E1 and E2 be in YExt1A (A, B). We first
define their coproduct to be

E1 ⊕ E2 : 0 B⊕B E1 ⊕ E2 A⊕A 0 ∈ YExt1A (A ⊕ A, B ⊕ B),

where all maps are diagonal.

Now the Baer sum of E1 and E2 is

E1 + E2 = (1 1) · (E1 ⊕ E2 ) · ( 11 ) ∈ YExt1A (A, B).

Theorem 27.4. The Yoneda-Ext of two objects, together with Baer sum, forms
an abelian group (provided it is a set). The zero-element of this abelian group
is given by the split short exact sequence.
This group structure turns YExt1A into an additive functor A op ×A Ab.

Proof. It is clear from the construction that the Baer sum is commutative.
For E1 , E2 , E3 ∈ YExt1A (A, B) one may see that
1
E1 + E2 + E3 = (1 1 1) · (E1 ⊕ E2 ⊕ E3 ) · 1
1

independent of brackets, that is Baer sum is associative.

Next we observe that for any short exact sequence E we have that both 0 · E
and E · 0 are split short exact. Indeed we have the following pushout diagram

π
E: 0 B1 E A 0

0 ( π0 )
( 10 ) (0 1)
0·E 0 B2 B2 ⊕ A A 0

(and a similar one for the case of pullbacks along zero-morphisms).

84 CHAPTER V. DERIVED FUNCTORS

Now we check that for two maps f and g from B1 to B2 , and an extension
E ∈ YExt1A (A, B1 ), we have (f + g) · E = f · E + g · E. As a first step, consider
the commutative diagram

E: 0 B1 E A 0

( 11 ) ( 11 )
( 11 ) · E : 0 B1 ⊕ B1 E
e A 0

( 11 )

E ⊕ E: 0 B1 ⊕ B1 E⊕E A⊕A 0

where the dashed arrow exists by the pushout property of the upper left square.
We see that
( 11 ) · E = (E ⊕ E) · ( 11 ) .
Now we calculate

f · E + g · E = (1 1) · (f · E ⊕ g · E) · ( 11 )
= (f g) · (E ⊕ E) · ( 11 )
= (f g) ( 11 ) · E
= (f + g) · E

Finally we use the above observations to verify that the split exact sequences
are a neutral element, and that there are inverses:
Let E ∈ YExt1A (A, B), and let Esplit denote the split exact sequence between
the same two objects. Then

E + Esplit = 1 · E + 0 · E = (1 + 0) · E = E.

Similarly we check that (−1) · E is an inverse of E:

E + (−1) · E = (1 − 1) · E = 0 · E

is the split exact sequence.

Theorem 27.5. Assume A has enough projectives. Then

YExt1A (A, B) = Ext1A (−, B)(A).

27. Ext1 AND EXTENSIONS 85

Dually, if A has enough injectives then

YExt1A (A, B) = Ext1A (A, −)(B).
In particular if A has both enough projectives and enough injectives then
Ext1A (−, B)(A) = Ext1A (A, −)(B).
Proof. We prove the first claim. The second one is dual, and the third one then
follows immediately.
Consider a short exact sequence
ι π
Ep : 0 ΩA P A 0

with P projective. We may consider this an element of YExt1A (A, ΩA).

Now multiplication with Ep gives a map

− · Ep : HomA (ΩA, B) YExt1A (A, B).

We claim that this map is surjective, and that its kernel consist precisely of the
morphisms factoring through ι.
To see surjectivity, consider the following diagram for any E ∈ YExt1A (A, B):

Ep : 0 ΩA P A 0

E: 0 B E A 0

Here the right dashed arrow exists by the lifting property of projectives, and
the left dashed arrow is a kernel morphism. It follows from the characterization
of pushouts (see Theorem 14.2) that the left square is a pushout, that is that
E = f · Ep for the morphism f found in the diagram.
To determine the kernel of the map − · Ep , note that a morphism f is in the
kernel if and only if we can find a commutative diagram
ι π
Ep : 0 ΩA P A 0

f ( rs )
( 10 ) (0 1)
E: 0 B B⊕A A 0
86 CHAPTER V. DERIVED FUNCTORS

By commutativity of the right square we need s = π, and then the left square
commutes if and only if r ◦ ι = f . It follows that the kernel consists precisely of
the maps factoring through ι.
But, by dimension-shift (see Theorem 26.7) we have

Ext1A (−, B)(A) = Cok[HomA (P, B) HomA (ΩA, B)],

that is Ext1A (−, B)(A) is also the quotient of HomA (ΩA, B) modulo morphisms
factoring through ι.

28 Total complexes - balancing Tor and Ext

Definition 28.1. A double complex is an infinite commutative square pattern

.. .. ..
. . .
m−1,n−2
dv dm,n−1
v dm+1,n−1
v
m−2,n−1 m−1,n−1
dh dh dm,n−1
h dm+1,n−1
h
··· X m−1,n−1 X m,n−1 X m+1,n−1 ···
m−1,n−1
dv dm,n−1
v dm+1,n−1
v
m−2,n m−1,n
dh dh dm,n
h dm+1,n
h
··· X m−1,n X m,n
X m+1,n ···
m−1,n
dv dm,n
v dm+1,n
v
m−2,n+1 m−1,n+1
dh dh dm,n+1
h dm+1,n+1
h
··· X m−1,n+1 X m,n+1 X m+1,n+1 ···
m−1,n+1
dv dm,n+1
v dm+1,n+1
v

.. .. ..
. . .

such that the composition of any two vertical or any two horizontal morphisms
vanishes.
In other words, a double complex is just an object in the category C(C(A )).
•,•
Definition 28.2. Let
` X m,s−m be a double complex, and assume that for any s, the
infinite coproduct m∈Z X exists. (For instance this is true if on every
diagonal there are only finitely many non-zero objects.)
28. TOTAL COMPLEXES - BALANCING Tor AND Ext 87

Then the total complex of X •,• is given by

a
Tot(X •,• )s = X m,s−m ,
m∈Z

with the differential given on components by


dv
 m,s−m
if m0 = m
m0 ,s+1−m0
X m,s−m
X : (−1)s−m dm,s−m
h if m0 = m + 1

0 otherwise.


Remark 28.3. Note that cones are a special case of total complexes, where the
only non-zero objects lie in rows −1 and 0.
Proposition 28.4. Let X •,• be a double complex concentrated in finitely many
rows. (That is there are a 6 b such that X m,n = 0 whenever n < a or n > b.)
Assume that all rows of X •,• are exact. Then the total complex Tot(X •,• ) is
exact.
Proof. Let (
m,n X m,n if n > a
Y =
0 if n 6 a
that is Y •,• is obtained from X •,• by removing the top non-zero row.
Then one may observe that there is a natural map

f • : X •,a [−a − 1] Tot(Y •,• )

and
Tot(X •,• ) = Cone(f • ).
Now we may assume inductively that Tot(Y •,• ) is exact, and it then follows
from the long exact sequence of homology that also all homologies of Tot(X •,• )
vanish.
Corollary 28.5. Let X •,• be a double complex such that all diagonals are finite.
(That is for any s there are only finitely many m such that X m,s−m 6= 0.)
Assume all rows of X •,• are exact. Then the total complex Tot(X •,• ) is exact.
Proof. For any given position, we may disregard the rows of X •,• such that
X s−n,n = 0. Hence exactness in position s follows from Proposition 28.4 above.
Since this applies to any given position the entire complex is exact.
88 CHAPTER V. DERIVED FUNCTORS

Theorem 28.6 (Balancing Ext). Let A be an abelian category with enough

projectives and enough injectives. Then for any A, B ∈ Ob A

ExtnA (A, −)(B) = ExtnA (−, B)(A).

Proof. We choose a projective resolution P • of A, and an injective resolution

I • of B.
Recall that the two Ext-groups of the theorem are by definition the homolo-
gies of HomA (A, I • ) and HomA (P • , B), respectively. We will connect these
two complexes via the third complex Tot(HomA (P • , I • )), showing that there
are two quasi-isomorphisms

HomA (P • , B) Tot(HomA (P • , I • )) HomA (A, I • ).

It then follows immediately that all three complexes have the same homologies.

We denote the exact complex

··· P −1 P0 A 0 ···

•
by P , and similarly the exact complex

···0 B I0 I1 ···

•
by I .
• •
We consider the double complex HomA (P , I ), and its versions with P •
• •
instead of P and I • instead of I . (In the following picture we write (X, Y )
28. TOTAL COMPLEXES - BALANCING Tor AND Ext 89

for HomA (X, Y ) to save space.)

(A, B) (P 0 , B) (P 1 , B) (P 2 , B)

(A, I 0 ) (P 0 , I 0 ) (P 1 , I 0 ) (P 2 , I 0 )

(A, I 1 ) (P 0 , I 1 ) (P 1 , I 1 ) (P 2 , I 1 )

(A, I 2 ) (P 0 , I 2 ) (P 1 , I 2 ) (P 2 , I 2 )

We note that

•
HomA (P • , I ) = exact columns
=⇒ exact total complex

and

•
HomA (P , I • ) = exact rows
=⇒ exact total complex

On the other hand there is the morphism HomA (P • , B) Tot(HomA (P • , I • ))

(essentially given by the morphisms crossing the vertical dashed line above),
•
whose cone is Tot(HomA (P • , I )). In particular the cone is exact, so the mor-
phism is a quasi-isomorphism.
90 CHAPTER V. DERIVED FUNCTORS

Similarly the natural morphism HomA (A, I • ) Tot(HomA (P • , I • )) is a

• •
quasi-isomorphism, since its cone Tot(HomA (P , I ) is exact.
Now the two quasi-isomorphisms

HomA (P • , B) Tot(HomA (P • , I • )) HomA (A, I • )

give rise to isomorphisms

H−n (HomA (P • , B)) ∼

= H−n (Tot(HomA (P • , I • ))) ∼
= H−n (HomA (A, I • )).

Now note that the left hand term is by definition ExtnA (−, B)(A), while the
right hand term is ExtnA (A, −)(B).

Theorem 28.7 (Balancing Tor). Let R be ring, M a right and N a left R-

module. Then
TorR R
n (M, −)(N ) = Torn (−, N )(M ).

Proof. The proof is very similar to the proof of Theorem 28.6 above. Here we
•
start with two projective resolutions PM and PN• of M and N respectively. We
then proceed as before to show that we have quasi-isomorphisms

M ⊗R PN• •
Tot(PM ⊗R PN• ) •
PM ⊗R N.

It then follows that the homologies of all three complexes coincide.

29 Small global dimension

Throughout this section, let A be an abelian category that has enough projec-
tives or enough injectives.

Definition 29.1. The global dimension of A is

[Link] A = sup{n ∈ N0 | ∃A, B ∈ Ob A : ExtnA (A, B) 6= 0} ∈ N0 ∪ {∞}.

An abelian category is called

• semisimple if [Link] A = 0;

• hereditary if [Link] A 6 1.

Proposition 29.2. The following are equivalent:

29. SMALL GLOBAL DIMENSION 91

(1) A is semisimple;

(2) all objects in A are projective;

(3) all objects in A are injective;

(4) all epimorphisms in A are split epimorphisms;

(5) all monomorphisms in A are split monomorphisms.

Proof. We show (1) =⇒ (5) =⇒ (3) =⇒ (1). The proof of (1) =⇒ (4) =⇒
(2) =⇒ (1) is similar.
Assume first that A is semisimple. Then YExt1A = 0, hence all short exact
sequences split. In particular any monomorphism splits.
If any monomorphism splits then the lifting property for injectives is auto-
matic, so all objects are injective.
Finally, if all objects are injective, then ExtnA (A, −)(B) = 0 for all n > 0.
(Note that B is its own injective resolution.)

Example 29.3. Let F be a field. Then both Mod F and mod F are semisimple
abelian categories.

Definition 29.4. Assume A has enough projectives. Then the projective di-
mension pd A of an object A is the smallest n, such that A has a projective
resolution of the form

··· 0 P −n ··· P0 0 ··· .

(We say pd A = ∞ if all projective resolutions of A are infinite.)

Dually, if A has enough injectives, then the injective dimension id A of an
object A is the smallest n such that A has an injective resolution

··· 0 I0 ··· In 0 ··· .

Remark 29.5. Clearly an object A is projective if and only if pd A = 0, and

injective if and only if id A = 0.

Theorem 29.6. Assume A has enough projectives, and let A ∈ Ob A . Then

pd A = sup{n ∈ N0 | ∃B ∈ Ob A : ExtnA (A, B) 6= 0}.

92 CHAPTER V. DERIVED FUNCTORS

Proof. If pd A = n then ExtiA (A, B) = 0 for any i > n, and thus we have the
inequality >.
Assume now that ExtiA (A, −) = 0 for some i. By dimension shift it follows
that Ext1A (Ωi−1 A, −) = 0. Interpreting this as Yoneda-Ext, we see that any
epimorphism to Ωi−1 A splits, that is that Ωi−1 A is projective.
Now we have a projective resolution of length i − 1, given by

0 Ωi−1 A P 2−i ··· P0 0

showing that pd A 6 i − 1.

We also have the dual of the above theorem:

Theorem 29.7. Assume A has enough injectives, and let A ∈ Ob A . Then

id A = sup{n ∈ N0 | ∃B ∈ Ob A : ExtnA (B, A) 6= 0}.

Corollary 29.8. Assume A has enough projectives. Then

[Link] A = sup{pd A | A ∈ Ob A }.

Dually, if A has enough injectives, then

[Link] A = sup{id A | A ∈ Ob A }.

Proposition 29.9. Assume A has enough projectives. Then A is hereditary

if and only if all subobjects of projective objects are projective.

Remark 29.10. This explains the name “hereditary”: subobjects inherit the
property of being projective.

Proof. Assume first that any subobject of a projective is projective. Then it

follows that any object has a projective resolution with at most two non-zero
terms. Thus A is hereditary by Corollary 29.8.
Assume conversely that A is hereditary, and let A P be a subobject of a
projective. We denote by P/A the cokernel of this inclusion, and observe that

Ext1A (A, −) = Ext2A (P/A, −) = 0

where the first equality is dimension shift, and the latter comes from the defi-
nition of hereditary. It follows that A is projective.
29. SMALL GLOBAL DIMENSION 93

Theorem 29.11. Let R be right noetherian. Then the category mod R of finitely
generated right R-modules is hereditary if and only if all right ideals of R are
projective.

Proof. “only if” is clear, since right ideals are submodules of the projective
module R.
“if”: It suffices to show that any submodule of Rn is projective (since all
projective objects are direct summands of free modules). We show this by
induction on n, the case n = 1 holding by assumption.
Let M be a submodule of Rn , and consider the split short exact sequence

0 R R ⊕ Rn−1 Rn−1 0.
| {z }
∼
=Rn

We denote by I and K the image and kernel of the composition M Rn−1 .

Thus we have the following commutative diagram

K M I

R Rn Rn−1

where the dashed map is the kernel morphism, and it is mono by commutativity
of the left square.
Now inductively both K and I are projective, hence the upper short exact
sequence splits, and M ∼
= K ⊕ I also is projective.

Remark 29.12. More generally, one can show that the category Mod R is
hereditary if and only if all right ideals of R are projective.

Example 29.13. Let R be a principal ideal domain. (That is a commuta-

tive ring without zero-divisors, such that every ideal is generated by a single
element.) Then Mod R is hereditary.
In particular Mod Z is hereditary, and for any field F the category of modules
over the polynomial ring Mod F[X] is hereditary.

Remark 29.14. One can show that for a field F, one has

[Link] F[X1 , . . . , Xd ] = d.
94 CHAPTER V. DERIVED FUNCTORS

30 Exercises
Exercise V.1. Calculate
• ExtnZ (−, Z/(b))(Z/(a)) for alle a, b, n ∈ N;
• Extnpreshmod F X (−, P0 )(Iω ) for all n ∈ N, where X is the poset from Exer-
cise IV.3.
Exercise V.2. Let R as below, and S be the R-module which is F as F-vector
space, with all variables acting as 0. Calculate all ExtnR (S, S) for n > 0.
• R = F[X];
• R = F[X]/(X 3 );
• R = F[X, Y ];
• R = F[X, Y ]/(XY ).
Exercise V.3. Let R = F[X, Y ]/(XY ) for some field F, and M = R/(X).
Calculate ExtnR (M, M ) for all n ∈ N.
Exercise V.4 (Balancing Ext). The aim of this exercise is to give a different
(arguably simpler) proof for the fact that Ext is independent of with respect to
which argument we derive the Hom-functor.
Let A be an abelian category with enough projectives.
• For a short exact sequence A B C, and an object X, show that
there is a long exact sequence

··· ExtnA (−, A)(X) ExtnA (−, B)(X) ExtnA (−, C)(X)
Extn+1
A (−, A)(X) Extn+1
A (−, B)(X) Extn+1
A (−, C)(X) ···

• Show that ExtnA (−, I)(X) = 0 for all positive n, if I is injective.

• Show that there is dimension shift with respect to the “inner” argument,
that is provided we have a short exact sequence

0 A I fA 0
with I injective, then
(
Extn−1
A (−, fA)(X) if n > 2
ExtnA (−, A)(X) ∼ =
Cok[HomA (X, I) HomA (X, fA)] if n = 1
30. EXERCISES 95

• Assume now that A in addition has enough injectives. Show that

ExtnA (−, A)(X) ∼

= ExtnA (X, −)(A).

Exercise V.5. Let A be an abelian category with enough injectives. We

consider the category of morphisms in A ,

mor(A ) = preshA {1 < 2}.

• Convince yourself that Ker defines a left exact functor mor(A ) A.

• Find out what the right derived functors Rn Ker are.
Hint: First consider the case that the morphism in question is an epimorphism.
Then generalize to arbitrary morphisms using a short exact sequence in mor(A )
where the other two objects are epimorphisms.
Exercise V.6. Calculate explicitly (i.e. by classifying equivalence classes of
short exact sequences) the following Yoneda-extension groups.
• YExt1Ab (Z/(2), Z/(3)),
• YExt1Ab (Z/(2), Z/(2)),
• YExt1preshmod F X (Iω , P0 ), where X = {0 < ω}.
(By genereal theory, these should coincide with the Ext calculated via projective
resolutions in Exercise V.1.)
Exercise V.7 (A short spectral sequence). Consider a double complex X •,•
with X m,n = 0 unless m, n ∈ {−1, 0} – that is essentially a commutative square
−1,−1
dh
−1,−1
X X 0,−1
−1,−1
dv d0,−1
v
−1,0
dh
X −1,0 X 0,0

We consider the kernels and cokernels of the horizontal maps, and denote by
k : Ker d−1,−1
h Ker d−1,0
h and c : Cok d−1,−1
h Cok dh−1,0 the kernel and cok-
ernel morphism, respectively.
Show that
96 CHAPTER V. DERIVED FUNCTORS

• H−2 (Tot(X •,• )) = Ker k;

• There is a short exact sequence Cok k H−1 (Tot(X •,• )) Ker c;

• H0 (Tot(X •,• )) = Cok c.
Exercise V.8. Let R = F[X, Y ]/(XY ) for some field F. Consider the double
complex X •,• given by
(
m,n m,n m,n X if m + n even
X = R, dh = dv = .
Y if m + n odd

Show that all rows and all columns of X •,• are exact, but its total complex is
not exact.
Exercise V.9. Let R be a ring, and X •,• a double complex of R-modules.
assume that X m,n = 0 whenever n > 0. (That is X •,• is concentrated on the
upper half plane.) Show that if all rows of X •,• are exact then so is its total
complex.
Exercise V.10. Let A be an abelian category with enough projectives. Con-
sider two short exact sequences C F B and B E A.
Assume that Ext2A (A, C) = 0.
Show that there is an object X completing the following diagram as indicated
by the dashed arrows.

C F B

C X E

A A

(That is, in the resulting diagram all squares commute and all rows and columns
are short exact sequences.)
Exercise V.11. Let A be an abelian category with enough projectives. Show
that [Link] A 6 2 if and only if any morphism between projectives has a pro-
jective kernel.
30. EXERCISES 97

Exercise V.12. Let A be an abelian category with enough projectives. Show

that
[Link] preshA {0 < ω} = [Link] A + 1.
98 CHAPTER V. DERIVED FUNCTORS
Chapter VI

Triangulated categories

31 Motivation – triangles in the homotopy cat-

egory
Throughout this section, let A be an abelian category. We have seen that for
a morphism of complexes f •

f•
A• B• Cone(f • ) A• [1]

is a complex in K(A ), giving rise to a long exact sequence of homology.

Now we take a different point of view, and say we consider the infinite
complex

··· A• [n] B • [n] Cone(f • )[n]

A• [n + 1] B • [n + 1] Cone(f • )[n + 1] ···

in the homotopy category. Since this complex is (up to shift) 3-periodic, we

denote it by the triangle of objects and morphisms

A• B•
|

Cone(f • )

99
100 CHAPTER VI. TRIANGULATED CATEGORIES

(where the decorated arrow indicates that this represents a morphism to A• [1]).
The next result shows that, while the triangle is defined starting with f • , it
has no preferred side.
Proposition 31.1. Let
f• ι• π•
A• B• Cone(f • ) A• [1]
be a triangle as above, that is ιn = ( 10 ) and π n = (0 1) for all n.
Then there is an isomorphism ϕ• : Cone(ι• ) A• [1] in K(A ) such that the
following diagram commutes.

ι• ( 10 ) (0 1)
B• Cone(f • ) Cone(ι• ) B • [1]

ϕ•
ι• π• −f • [1]
B• Cone(f • ) A• [1] B • [1]

Proof. We start by calculating that

Cone(ι• )n = Cone(f • )n ⊕ B n+1 = B n ⊕ An+1 ⊕ B n+1
and !
dn
B f
n+1
1
ιn+1
 
dCone(f • )
dnCone(ι• ) = 0 n+1
−dB
= 0 −dn+1
A 0 .
0 0 −dn+1
B

We consider the morphisms ϕ• : Cone(ι• ) A• [1] given by ϕn = (0 1 0) and

 0 
ψ • : A• [1] Cone(ι• ) given by ψ n = 1
n+1
. A straightforward calculation
−f
shows that these are indeed morphisms of complexes.
Now we check the following four claims:
(1) The square
( 10 )
Cone(f • ) Cone(ι• )

ϕ•
π•
Cone(f • ) A• [1]

commutes in the category C(A ).

32. DEFINITION 101

(2) The square

(0 1)
Cone(ι• ) B • [1]

ψ•
−f • [1]
A• [1] B • [1]

commutes in the category C(A ).

(3) ϕ• ◦ ψ • = 1 in the category C(A ).
(4) 1 − ψ • ◦ ϕ• is null homotopic.
(1), (2), and (3) are straightforward matrix calculations, which are left to the
reader. We only check (4) here. First we calculate
1   0  1 0 0
1 − ψ n ◦ ϕn = 1 − 1
n+1
◦ (0 1 0) = 0 0 0 .
n+1
1 −f 0f 1
0 0 0
Now we set hn = 000 : Cone(ι• )n Cone(ι• )n−1 and see that
100

dn−1 n
Cone(ι• ) ◦ h + h
n+1
◦ dnCone(ι• )
! !
dn
B f
n+1
1 0 0 0 0 0 0 dn
B f
n+1
1
= 0 −dn+1 A 0 ◦ 000 + 000 ◦ 0 −dn+1
A 0
100 100
0 0 −dn+1
B 0 0 −dn+1
B
 
 1 00
 0 0 0
= 0 00 + 0 0 0
−dnB 0 0
n
dB f n+1
1
1 0 0
= 0 0 0
n+1
0f 1

Now note that (3) and (4) together imply that ϕ• and ψ • are mutually
inverse isomorphisms in K(A ), and thus (2) implies that also the rightmost
square in the proposition commutes up to homotopy.

32 Definition
Definition 32.1. A triangulated category is an additive category T , together
with an autoequivalence [1] : T T , and a class ∆ of diagrams of the form
f g h
X Y Z X[1] such that
102 CHAPTER VI. TRIANGULATED CATEGORIES

(T1) • For any morphism f : X Y in T , there is a diagram

f
X Y Z X[1]

in ∆.
idX
• For any object X, the diagram X X 0 X[1] is in ∆.

• ∆ is closed under isomorphisms.

f g h
(T2) For any diagram X Y Z X[1] in ∆ also the diagrams

g h −f [1]
Y Z X[1] Y [1] and
−h[−1] f g
Z[−1] X Y Z

are in ∆.

(T3) Given the solid part of a diagram

X Y Z X[1]

u v w u[1]

X0 Y0 Z0 X 0 [1]

where the square commutes, and the rows are in ∆, one can always find
a morphism w as indicated above such that the entire diagram becomes
commutative.

(T4) Octahedral axiom: Given the solid part of the following diagram, where
32. DEFINITION 103

the two rows and the left column are in ∆,

f g
X Y Z0 X[1]

X Z Y0 X[1]

f [1]
h
X0 X0 Y [1]

h
g[1]
Y [1] Z 0 [1]

there are morphisms as indicated by the dashed arrows, such that also the
second column is in ∆, and the entire diagram commutes.

Remark 32.2. Sometimes morphisms Z X[1] are denoted by arrows

Z | X.

Then the elements of ∆ can be depicted as actual triangles

X Y
|

Z
104 CHAPTER VI. TRIANGULATED CATEGORIES

In particular the octahedron in Axiom (T4) becomes visible in this notation:

| Y0

X | X0
|
Y

|
Z0

Here all the oriented triangles lie in ∆, and all the non-oriented triangles and
squares commute.
Remark 32.3. (T3), by use of (T2), can be seen as a kind of “2 out of 3”-
property: Given any two morphisms connecting two triangles, one may find a
third.
Theorem 32.4 (Long exact Hom-sequence). Let T be a triangulated category,
X Y Z X[1] in ∆, and T ∈ Ob T . Then the sequences

··· HomT (T, X[n]) HomT (T, Y [n]) HomT (T, Z[n])
HomT (T, X[n + 1]) HomT (T, Y [n + 1]) HomT (T, Z[n + 1]) ···

and

··· HomT (Z[n], T ) HomT (Y [n], T ) HomT (X[n], T )

HomT (Z[n − 1], T ) HomT (Y [n − 1], T ) HomT (X[n − 1], T ) ···

are exact.
Proof. We prove the first claim, the second one is dual. (Note that T op is also
triangulated.)
32. DEFINITION 105

By the rotation axiom (T2) it suffices to check that the sequence

HomT (T, X) HomT (T, Y ) HomT (T, Z)

is exact. We do so by comparing the given triangle to the trivial triangle

T T 0 T [1].

T T 0 T [1]

g f g[1]

X Y Z X[1]

By (T3) the existence of the dashed arrow g is equivalent to the existence of the
middle dashed arrow. That is, for f ∈ HomT (T, Y ) we have

[Y Z] ◦ f = 0 ⇐⇒ ∃g ∈ HomT (T, X) : [X Y ] ◦ g = f.

Remark 32.5. The above theorem says that any morphism in a triangle is a
weak kernel of the next morphism, and a weak cokernel of the previous mor-
phism.

Theorem 32.6 (2 out of 3 property for isomorphisms). Let T be a triangulated

category, and consider two triangles connected by morphisms as in the following
diagram.

X1 Y1 Z1 X1 [1]

f g h f [1]

X2 Y2 Z2 X2 [1]

If two of the morphisms f , g, and h are isomorphisms, then so is the third one.

Proof. By (T2) we may rotate the triangles and assume f and g are isomor-
phisms. Now we apply the functor HomT (−, Z1 ) to the entire diagram, obtain-
106 CHAPTER VI. TRIANGULATED CATEGORIES

ing

(X1 , Z1 ) (Y1 , Z1 ) (Z1 , Z1 ) (X1 [1], Z1 ) (Y1 [1], Z1 )

−◦f −◦g −◦h − ◦ f [1] − ◦ g[1]

(X2 , Z1 ) (Y2 , Z1 ) (Z2 , Z1 ) (X2 [1], Z1 ) (Y2 [1], Z1 )

where (−, Z1 ) is short for HomT (−, Z1 ).

Since f and g are isomorphisms it follows that also the left two and right
two vertical maps in this diagram are isomorphisms. Now, by the five lemma,
the morphism
− ◦ h : HomT (Z2 , Z1 ) HomT (Z1 , Z1 )
is an isomorphism. In particular there is h̃ ∈ HomT (Z2 , Z1 ) such that h̃ ◦ h =
idZ1 , that is h is split mono.
Similarly, using the functor HomT (Z2 , −), one sees that h is split epi. Thus
h is an isomorphism.

33 Homotopy categories are triangulated

Theorem 33.1. Assume A is an additive category. Then the homotopy cat-
egory K(A ) is triangulated, with ∆ being the class of all diagrams isomorphic
to standard triangles A• • B • Cone(f • ) A• [1].
f

Proof. We have seen (the first half of) (T2) in Proposition 31.1.
The first and last point of (T1) hold by construction, for the second one note
idX
that 0 X• X• 0 is a triangle, so, since we already checked (T2), so is
idX
X• X• 0 X • [1].
For (T3) we may, up to isomorphism, assume the following setup:

f• ( 10 ) (0 1)
X• Y• Cone(f • ) X • [1]

u• v• ∃? u• [1]
f 0• ( 10 ) (0 1)
X 0• Y 0• Cone(f 0• ) X 0• [1]
33. HOMOTOPY CATEGORIES ARE TRIANGULATED 107

where the left square commutes up to homotopy. Explicitly that means that
there are maps hi : X i (Y 0 )i−1 such that

v i ◦ f i − (f 0 )i ◦ ui = di−1 i
Y0 ◦h +h
i+1
◦ diX .
 n n+1 
One easily sees that the map given by wn = v0 uhn+1 is a morphism of com-
plexes and fits into this diagram.
To check the octahedral axiom (T4), we again may assume that all triangles
are standard triangles, that is consider the commutative diagram

f• ( 10 ) (0 1)
X •
Y • Cone(f • ) X • [1]
gn 0
g•


0 1 n
g• ◦ f • ( 10 ) (0 1)
X• Z• Cone(g • ◦ f • ) X • [1]
1 0
( 10 ) 01
1 0
00 00
01 00
00 gn 0
Cone(g • )


Cone( 0 1 n
)
?
(0 1) ( 00 00 10 01 )
( 10 )
Y • [1] Cone(f • )[1]

The map marked ? in the diagram above is an isomorphism in K(A ), with

inverse given by
  n
1 0 00
Cone(g • ).


0f n+1
10 : Cone( g0 01 n )
n

It only remains to check that the square

(0 1)
Cone(g • ◦ f • ) X • [1]
1 0
01
00   f • [1]
00 1 0 00
0 f n+1 1 0 (0 1)
gn 0 n
Cone(g • ) Y • [1]


Cone( 0 1 n
) ∼
=
108 CHAPTER VI. TRIANGULATED CATEGORIES

commutes up to homotopy. In fact one easily checks that it even commutes in

C(A ).
Observation 33.2. Any object A ∈ Ob A may be considered as a complex
···0 A 0 · · · , with A in degree 0. This construction gives a fully faith-
ful embedding of A into C(A ) and into K(A ). (Note that no non-zero map
between complexes of this form can be null-homotopic.)
By abuse of notation, we identify the object X with the complex as above.
Lemma 33.3. Let X ∈ A , and A• ∈ C(A ). Then we may consider the
complex
HomA (X, A• ).
We have

Zn HomA (X, A• ) = HomC(A ) (X, A• [n]) and

Hn HomA (X, A• ) = HomK(A ) (X, A• [n]).

Dually

Zn HomA (A• , X) = HomC(A ) (A• , X[n]) and

n • •
H HomA (A , X) = HomK(A ) (A , X[n]).

Proof. We see that

HomC (A ) (X, A• [n]) = {ϕ ∈ HomA (X, An ) | dnA ◦ ϕ = 0}

= Ker[HomA (X, An ) HomA (X, An+1 )]
= Zn HomA (X, A• ).

Moreover a morphism from X to A• [n] is null-homotopic if and only if if factors

through dn−1
A , that is lies in

Bn HomA (X, A• ) = Im[HomA (X, An−1 ) HomA (X, An )].

The claim on homology now follows by taking quotients.

Recall that, provided an abelian category A has enough projectives, we have
the functor p : A K(A ) taking an object to its projective resolution. Recall
also that, by the horseshoe lemma (Proposition 23.8), for a short exact sequence

0 A B C 0 in A
34. DERIVED CATEGORIES 109

we have a triangle

pA pB pC pA[1] in K(A ).

Theorem 33.4. Assume A has enough projectives. Then

ExtnA (A, B) = HomK(A ) (pA, B[n]).

Dually, if A has enough injectives, then

ExtnA (A, B) = HomK(A ) (A, iB[n]).

Proof. We have ExtnA (A, B) = Hn HomA (pA, B) by definition. Now the claim
follows from Lemma 33.3 above.

Remark 33.5. In view of this theorem, the long exact Hom-Ext-sequence can
be seen as a long exact sequence coming from a triangle in the homotopy cate-
gory.

We proceed by extending the above to arbitrary derived functors. To do so,

we need the following two observations:

Observation 33.6. Let A be an abelian category. Taking homology H0 takes

triangles in K(A ) to long exact sequences. (This is just a restatement of the
long exact sequence of homology – see Theorem 20.6.)

Observation 33.7. Let F : A B be any additive functor. Then FK preserves

triangles.

Construction 33.8. Let F : A B be right exact. Then the long exact se-
quence of derived functors (associated to a short exact sequence A B C
in A ) is the long exact sequence of homology coming from the triangle

FK pA FK pB FK pC FK pA[1] in K(B).

34 Derived categories
Derived categories address the following two (closely related) issues with homo-
topy categories:
110 CHAPTER VI. TRIANGULATED CATEGORIES

• Short exact sequences are not triangles in the homotopy category. (How-
ever one may get triangles replacing the objects by projective or injective
resolutions.)
• Quasi-isomorphisms preserve all information on homology, but are not iso-
morphisms in the category K(A ). In particular, in the discussion above,
we had to take a projective or injective resolution, instead of the object
itself (which is quasi-isomorphic).
The answer to these issues it to (brute force) make quasi-isomorphisms in-
vertible.
Construction 34.1. Let A be an abelian category. A roof from a complex
X • to Y • is a diagram of the form

X̃ • f
qis q
X• Y•

with some middle object X̃ • , and where q is a quasi-isomorphism. For compact

notation we write the above roof as f · q −1 .
Two roofs f · q −1 and g · r−1 are called equivalent if there is a commutative
diagram
X̃ •
qis q f
qis

X• H• Y•

qis
qis r g
˜•
X̃

Remark 34.2. In other words, if we denote the middle quasi-isomorphisms by

q 0 and r0 respectively, we find a common denominator q ◦ q 0 = r ◦ r0 , and then
compare the enumerators f ◦ q 0 and g ◦ r0 .
We need to check that the above notion of equivalence defines an equivalence
relation. To that end (and in fact throughout the discussion of roofs) we need
the following observation.
34. DERIVED CATEGORIES 111

Lemma 34.3 (Ore condition). Let A be an abelian category. Given the solid
part of the following square, where q is a quasi-isomorphism, it is possible to
find the dashed part (including Ỹ • ), where r is a quasi isomorphism.

f
X̃ • Y•

qis q r qis
g
X• Ỹ •

Dually, given the dashed part, it is possible to find the solid part.

Proof. We complete q to a triangle as in the upper row of the following diagram

q
X̃ • X• Cone(q) X̃ • [1]

f g idCone(q) f [1]
r
Y• Ỹ • Cone(q) Y • [1]

Now by (T1) and (T3) we can complete the diagram as indicated by the dashed
arrows, such that the lower row is also a triangle. Since q is quasi-iso we know
that Cone(q) is exact. Now, since the cone of r is (isomorphic to) Cone(q), it
follows that also r is a quasi-isomorphism.

Lemma 34.4. The above defines an equivalence relation on the collection of

roofs from X • to Y • .

Proof. The definition of equivalence is clearly reflexive and symmetric.

Assume f · q −1 is equivalent to g · r−1 , which in turn is equivalent to h · s−1 ,
112 CHAPTER VI. TRIANGULATED CATEGORIES

as in the solid part of the following diagram.

X̃ •

qis
f
qis q H•
qis
qis Ĥ •
r g
X• ˜•
X̃ Y•
qis
qis qis

qis s h
H̃ •
qis

˜
˜•
X̃

By the Ore condition (Lemma 34.3) is is possible to find Ĥ • and the two dashed
quasi-isomorphisms such that the square in the middle commutes. (One easily
sees that if three sides of a square are quasi-iso, then so is the forth.)
Now the claim follows by considering Ĥ • between the two outer roofs.

Construction 34.5. Let A be an abelian category. Assume that for any

complexes X • and Y • , the collection of roofs from X • to Y • up to equivalence
is a set. Then we define the derived category by

Ob D(A ) = Ob K(A ) and

HomD(A ) (X , Y ) = {roofs from X • to Y • }/ ∼,
• •

with composition given as follows:

Given f · q −1 : X • Y • , and g · r−1 : Y • Z • as in the solid part of the
34. DERIVED CATEGORIES 113

following diagram,

˜•
X̃
qis r̃ f˜

X̃ • Ỹ •
qis q f qis r g

X• Y• Z•
˜ • and the two dashed maps by the Ore condition (Lemma 34.3)
we may find X̃
. We now define the product to be
(g · r−1 ) ◦ (f · q −1 ) = (g f˜) · (qr̃)−1 .
One may check that this is well-defined up to equivalence of roofs, and only
depends on the equivalence class of the factors. Then it is easy to see that this
multiplication is associative.
Observation 34.6.
• The derived category comes with a natural functor K(A ) D(A ) which
is sends every complex to itself, and a morphism f to the trivial roof
f · id−1 .
• A complex X • becomes isomorphic to 0 in D(A ) if and only if it is exact.
• A morphism f in K(A ) is mapped to the zero-morphism in D(A ) if there
is a quasi-isomorphism q such that f ◦ q = 0. One can prove that this is
equivalent to f factoring through an exact complex. (To see this, consider
the cone of q.)
• For a quasi-isomorphism q, also the shift q[n] is a quasi-isomorphism for
any n. It follows that [1] defines an autoequivalence of D(A ).
Theorem 34.7. Let A be an abelian category, such that D(A ) is defined. Then
D(A ) is a triangulated category, where ∆ consists of all triangles isomorphic
to standard triangles

f
X• Y• Cone(f ) X • [1],
where f is a morphism of complexes.
114 CHAPTER VI. TRIANGULATED CATEGORIES

Proof. We check the axioms. Here we make heavy use of the fact that we already
checked the axioms for K(A ).
(T1) For the first bullet point (there is a triangle starting with any morphism)
we proceed as follows: Given the morphism f · q −1 we first find a standard
triangle starting with f , and then alter it by the isomorphism q.
The second bullet point (triangle with identity as first morphism) follows
from the same statement for K(A ).
The third one (∆ closed under isos) holds by definition.
(T2) Up to isomorphism, the triangle is a standard triangle. For such trian-
gles we know that the rotations are isomorphic to standard triangles in K(A ),
and therefore also in D(A ).
(T3) Up to isomorphism we may assume that the two triangles we want to
connect are standard triangles. That is we have the solid part of the following
diagram.

a
A• B• Cone(a) A• [1]
qis q qis q[1]
à qis r s Ã[1]
qis q̃ qis q̃[1]
ã ˜
Ø B̃ Cone(ã) Ã[1]
f g f [1]
h
a
C• D• Cone(c) C • [1]

By the Ore condition (Lemma 34.3) we can find the two dashed maps as in the
diagram above, such that the upper pentagon commutes. We may even choose
them in such a way that also the lower pentagon commutes.
We complete ã to a triangle, and apply (T3) for the homotopy category to
find the morphisms s and h making the diagram commutative. Taking homology
and applying the five lemma (Theorem 14.1), we can see that r and q ◦ q̃ being
quasi-isomorphisms automatically also makes s a quasi-isomorphism. Thus the
morphism h · s−1 in D(A ) is the desired morphism between cones.
(T4) For the octahedral axiom, one may argue (similarly to the above) that
all the input data lies in the homotopy category, and thus (T4) follows from the
same axiom for K(A ).
Remark 34.8. It might seem like we gained little, since the triangles in the
34. DERIVED CATEGORIES 115

derived category are “the same” as the triangles in the homotopy category.
However, since there are now more isomorphisms (all quasi-isos have become
isomorphisms), there are in fact “more” triangles.
f g
Example 34.9. Let 0 A B C 0 be a short exact sequence in an
abelian category. Then there is a triangle

f g
A B C A[1] in D(A ).
To see this, consider the standard triangle

f ( 10 ) (0 1)
A B Cone(f ) A[1].
We note that Cone(f ) is the complex in the upper row of the following diagram,
and that the vertical map here is a quasi-isomorphism

f
··· 0 A B 0 ···
q g

··· 0 0 C 0 ···

Thus we also have the (isomorphic) triangle

f q ◦ ( 10 ) (0 1) · q −1
A B C A[1]
in D(A ). Finally note that q ◦ ( 10 ) = g.
Theorem 34.10. Let A be abelian, and A, B ∈ Ob A . Then

0
 if n < 0,
HomD(A ) (A, B[n]) = HomA (A, B) if n = 0,
1

YExtA (A, B) if n = 1.


Proof. We consider roofs

E• f
qis q
A B[n]
116 CHAPTER VI. TRIANGULATED CATEGORIES

where q is a quasi-isomorphism.
We first consider the truncation of E • to the right as in the following dia-
gram.

τ 60 E • : ··· E −1 Ker d0E 0 ···

E• : ··· E −1 E0 E1 ···

One may observe that the natural map r : τ 60 E • E • indicated above induces
an isomorphism on all non-positive homologies. Since in our setup E • is quasi-
isomorphic to A it has non-zero homology only in degree 0. Therefore r is a
quasi-isomorphism.
It follows that the roof f · q −1 is equivalent to the roof f r · (qr)−1 . In other
words, up to equivalence we may assume that E • is concentrated in non-positive
degrees.
This proves the first claim, since B[n] is concentrated in (the in that case
positive) degree −n.
Now assume n > 0. Then we may (similarly to the above) cut off the left
part of E • as indicated in the following diagram.

E• : ··· E −n−1 E −n E −n+1 ···

−n−1
τ >−n E • : ··· 0 Cok dE E −n+1 ···

As before we see that the map s : E • τ >−n E • is a quasi-isomorphism (since

n > 0).
Now observe that both q and f factor through s (since both A and B[n]
are concentrated in degrees > −n), say via q 0 and f 0 . Then we see that the
roof f · q −1 is equivalent to f 0 · (q 0 )−1 . Thus now we may assume that E • is
concentrated in degrees −n, . . . , 0.
Now we consider the case n = 0. Then, by the above discussion, we may
assume that E • is concentrated in degree 0. Thus q is an isomorphism, and hence
f · q −1 lies in the image of the natural map HomA (A, B) HomD(A ) (A, B).
Conversely this map is also injective, since no non-zero morphism from A to B
has vanishing homology.
34. DERIVED CATEGORIES 117

Proposition 34.11. Let A be an abelian category, and P • a right bounded

complex of projectives. (That is all P n are projective, and ∃N ∀n > N : P n = 0.)
Then

(1) Let E • be an exact complex. Then HomK(A ) (P • , E • ) = 0.

(2) Any quasi-isomorphism P̃ • P • from any complex P̃ • to P • is a split

epimorphism in the category K(A ).

(3) Let X • be any complex. Then the map

HomK(A ) (P • , X • ) HomD(A ) (P • , X • )

is an isomorphism.

Proof. (1) Let f • : P • E • be a morphism of complexes. We construct a null-

homotopy iteratedly from right to left. So let n be some index, and assume we
i−1
already have hi : P i E i−1 for i > n, such that f i = dE ◦ hi + hi+1 diP . (Note
that this is automatic for n > N - thus we have a starting point for our iterated
construction.)
The setup is depicted in the following diagram.

dnP
Pn P n+1
fn
hn hn+1 f n+1
dn−1
E
E n−1 En E n+1
dnE

Im dn−1
E

We observe that dnE ◦ (f n − hn+1 ◦ dnP ) = 0, and hence f n − hn+1 ◦ dnP factors
through Im dn−1
E = Ker dnE as indicated by the dashed arrow above. Since P n is
projective we may lift along the epimorphism E n−1 Im dn−1
E , obtaining h
n

as indicated by the dotted arrow.

(2) Let q : P̃ • P • be a quasi-isomorphism. Then, in the triangle

q
P̃ • P• Cone(q) P̃ • [1]
118 CHAPTER VI. TRIANGULATED CATEGORIES

in K(A ), the complex Cone(q) is exact (Corollary 21.6), so by (1) the middle
map vanishes.
It now follows that q is a split epimorphism.
(3) By (2) in any fraction f · q −1 : P • X • the quasi-isomorphism q is a
split epimorphism. Thus we may find q̃ such that q ◦ q̃ = idP • . One easily checks
that f · q −1 = (f q̃) · id−1 , so it lies in the image of the functor K(A ) D(A ).
On the other hand, we know that a morphism f in K(A ) vanishes in D(A ) if
and only if there is a quasi-isomorphism q such that f ◦ q = 0 (Observation 34.6.
However, by (2) such a quasi-isomorphism is a split epimorphism, hence f =
0.

Corollary 34.12. Assume A has enough projectives or enough injectives.

Then
HomD(A ) (A, B[n]) = ExtnA (A, B)
for any n and objects A and B of A .

Proof. Assume A has enough projectives, and let pA be a projective resolution

of A. Then the natural projection q : pA A is a quasi-isomorphism, and so

HomD(A ) (A, B[n]) = HomD(A ) (pA, B[n])

= HomK(A ) (pA, B[n]) (by Proposition 34.11)
= ExtnA (A, B) (by Theorem 33.4)

Remark 34.13. • By Example 34.9 a short exact sequence in A “is” a

triangle in D(A ). Thus, by Corollary 34.12 above, the long exact Hom-
Ext sequence can be interpreted as the long exact Hom-sequence coming
from this triangle.

• Corollary 34.12 and Theorem 34.10 also show that all definitions of Ext
coincide, when they are defined (Yoneda-Ext, deriving by first argument,
and deriving by second argument).

35 Derived functors
Let F : A B be an additive functor. We observed that applying this functor
position by position gives rise to a functor FK : K(A ) K(B). We would now
35. DERIVED FUNCTORS 119

like to do the same thing for derived categories, that is we would like to have a
functor FD making the following square commutative
FK
K(A ) K(B)

πA πB
FD
D(A ) D(B)

where πA and πB are the canonical functors from the homotopy categories to
the corresponding derived categories.
Unfortunately, however, it is not possible to find such a functor FD in general:
Lemma 35.1. Let F : A B be an additive functor between abelian categories.
Then a functor FD : D(A ) D(B) making the diagram above commutative
exists if and only if F is exact.
Proof. If the functor F is exact, then it preserves homology, and thus in particu-
lar FK preserves quasi-isomorphisms. It follows that we can define FD (f ·q −1 ) =
FK (f ) · FK (q)−1 . (Note that FK preserves equivalence of roofs, so this is in fact
well-defined.)
On the other hand, if F is not an exact functor, then there will be a short
exact sequence A1 A2 A3 in A such that the image

0 FA1 FA2 FA3 0

is not exact. Interpreting this sequence as an element of K(A ), we see that the
object is sent to 0 by πA , but not by πB ◦ FK . Clearly this makes it impossible
to obtain a commutative square as above.
Since it usually is not possible to find a functor FD as above, one is lead to
consider functors that make the square “as commutative as possible”.
Definition 35.2. Let F : A B be an additive functor between abelian cat-
egories. A (total) left derived functor of F is a functor LF : D(A ) D(B),
together with a natural transformation φ : LF ◦ πA πB ◦ FK , which is univer-
sal in the following sense:
For any other functor G : D(A ) D(B), together with a natural transfor-
mation ψ : G◦πA πB ◦FK , there is a unique natural transformation ζ : G LF
such that ψ = φ ◦ ζπA .
120 CHAPTER VI. TRIANGULATED CATEGORIES

Dually, a (total) right derived functor of F is a functor RF : D(A ) D(B),

together with a natural transformation φ : πB ◦ FK RF ◦ πA satisfying a dual
universal property.
Remark 35.3. In general, there is no reason for a total left or right derived
functor to exist.
However, if one does exist, then the universal property guarantees that it is
unique (up to unique natural isomorphism). Therefore we can talk about the
left derived functor or the right derived functor in this case.
It is a bit technical to construct total derived functors between the entire de-
rived categories in general (and requires additional assumptions on A ). To sim-
plify our situation here a bit we consider the full subcategory of right bounded
complexes
C− (A ) = {A• | ∃N ∀n > N An = 0} ⊂ C(A ),
and its counterparts K− (A ) ⊂ K(A ) and D− (A ) ⊂ D(A ). Similarly we
may consider the category of left bounded complexes C+ (A ), the homotopy
category of left bounded complexes K+ (A ), and the derived category of left
bounded complexes D+ (A ).
Proposition 35.4. Assume A has enough projectives. For any right bounded
complex A• there is a right bounded complex p(A• ) of projectives and a quasi-
isomorphism p(A• ) A• .
This construction gives a functor

p : D− (A ) K− (A )

which is left adjoint to projection π : K− (A ) D− (A ). Moreover, the unit of

the adjunction  : idD− (A ) πp is a natural isomorphism.
•
Proof. We construct p(A ) iteratedly from right to left. Assume all Ai with i >
n are already projective. Pick an epimorphism P n An with P n projective,
and consider the following diagram

An−1 Pn
Q
··· An−2 An Pn An+1 ···

id id

··· An−2 An−1 An An+1 ···

35. DERIVED FUNCTORS 121

where the map P n An+1 is composition, and the map An−2 An−1 An P n
Q
is obtained from the pullback property. Since the pullback is taken along an
epimorphism the middle square is in fact exact, and thus this morphism of
complexes is a quasi-isomorphism.
Iterating this construction one obtains the desired quasi-isomorphism

ηA• : p(A• ) A• .

Now we can first turn p into a functor D− (A ) D− (A ) by setting p(f ) =

−1 • •
ηB ◦ f ◦ ηA for any morphism f : A B . Since by Proposition 34.11(3) the
morphism sets in the derived and homotopy category coincide on right bounded
complexes of projectives, p defines a functor D− (A ) K− (A ).
The fact that p is left adjoint to π follows from

34.11(3)
HomK− (A ) (pA• , B • ) ∼
= HomD− (A ) (pA• , B • ) ∼
= HomD− (A ) (A• , B • ),

where the second isomorphism is due to the fact that the quasi-isomorphism
ηA• becomes an isomorphism in the derived category.
Finally we note that the unit is given by A• = (ηA• )−1 – which is defined
on the derived level.

Now we can prove that total right derived functors can be understood using
projective resolutions at least in the setup of right bounded complexes.

Theorem 35.5. Let F : A B be an additive functor between abelian cate-

gories.

• Assume that A has enough projectives. Then on the subcategories of right

bounded complexes there is a total left derived functor

LF : D− (A ) D− (B)

given by LF = πB ◦ FK ◦ p.

• Dually, if A has enough injectives, then there is a total right derived

functor
RF : D+ (A ) D+ (B)
with respect to left bounded complexes, given by RF = πB ◦ FK ◦ i.
122 CHAPTER VI. TRIANGULATED CATEGORIES

Proof. We only prove the first claim, the second one is dual.
First note that in the diagram
FK
K− (A ) K− (B)

πA πB
LF
D− (A ) D− (B)

we do have a natural transformation φ : LF ◦ πA πB ◦ FK . Recalling that

p is left adjoint to πA (see Proposition 35.4 above) we have the counit η : p ◦
πA idK− (A ) . We now choose

φ = (πB ◦ FK )(η) : πB ◦ FK ◦ p ◦πA πB ◦ F K .

| {z }
=LF

It only remains to verify that our choice of LF and φ satisfy the universal
property of Definition 35.2. Let G : D− (A ) D− (B) be a different functor,
together with a natural transformation ψ : G ◦ πA πB ◦ FK . If ζ : G LF is
a natural transformation such that ψ = φ ◦ ζπA then
ψp = φp ◦ ζπA ◦p

Since the unit  : idD− (A ) πA ◦ p is a natural isomorphism, and since

φp = (πB ◦ FK )(ηp ) = (πB ◦ FK ◦ p)(−1 )
we obtain
ζ = LF()−1 ◦ ζπA ◦p ◦ G() = ψp ◦ G().
In particular ζ as uniquely determined.
Conversely, with the choice ζ = ψp ◦ G(), we obtain
φ ◦ ζπA = (πB ◦ FK )(η) ◦ ψpπA ◦ G(πA )
= ψ ◦ (G ◦ πA )(η) ◦ G(πA )
= ψ ◦ G(πA (η) ◦ πA )
| {z }
=idD− (A )

= ψ.
Thus our choice of LF and φ does satisfy the universal property, so it is the total
left derived functor of F.
36. EXERCISES 123

Remark 35.6. • Theorem 35.5 shows, in particular, that the appearance

of projective and injective resolutions in the definition of left and right
derived functors is not an arbitrary choice / coincidence. On the con-
trary, the definition of derived functors via a universal property forces
this construction.

• In many cases an obvious analog of Theorem 35.5 also holds for unbounded
complexes. However, for such a result one typically needs that A has
certain colimits (essentially along the poset Z), and that these colimits
are exact.

36 Exercises

Exercise VI.1. Let T be a triangulated category, and X Y Z X[1]

a distinguished triangle. Assume the map Z X[1] is 0. Show that the (rest
of the) triangle then is a split short exact sequence.

Exercise VI.2. Let T be a triangulated category. Assume that T is in addi-

tion abelian. Show that T is semisimple.

Exercise VI.3. Let A be an abelian category. Show that any complex is

isomorphic to its homology in K(A ) if and only if A is semisimple.
(Here the homology of a complex X • is considered as the complex
0 0 0
··· H−1 (X • ) H0 (X • ) H1 (X • ) · · · .)

f
Exercise VI.4. Let T be a triangulated category, and C F B C[1]
g
and B E A B[1] two distinguished triangles. Assume the composition
f [1] ◦ g : A C[2] vanishes. Show that there is an object X and morphisms
as indicated by the dashed arrows below, such that the diagram commutes and
124 CHAPTER VI. TRIANGULATED CATEGORIES

the new row and new column are distinguished triangles too.

C F B C[1]

C X E C[1]

A A

F [1] B[1]

(Compare to Exercise V.10.)

Exercise VI.5. Let A be abelian. Let A• be a complex concentrated in nega-

tive degrees (An = 0 ∀n > 0), and B • be a complex concentrated in non-negative
degrees (B n = 0 ∀n < 0).
Show that HomD(A ) (A• , B • ) = 0.
Exercise VI.6. Let T be a triangulated category, and U be a triangulated
subcategory. (That is a full subcategory closed under [1] and [−1], and such
that the cone of any morphism in the subcategory is in U again.)
Let S be the collection of all morphisms in T , whose cone lies in U .
Show that (up to set theoretical issues) one can define a triangulated category
S −1 T making all morphisms in S invertible in the same way as we defined
the derived category in the lectures.
Index

3 × 3 Lemma, 50 covariant functor, see functor, covari-

ant
Ab, 6 cycles, 64
abelian category, 34
additive category, 29 dense functor, see functor, dense
adjoint pair of functors, 15 derived category, 112
derived functor, 77
Bn (−), 64 total, 119
Baer sum, 83 dimension shift, 81
balanced map, 56 double complex, 86
biproduct, 29
boundaries, 64 elementary tensor, 56
enough injectives, 70
C(−), 63 enough projectives, 70
category, 5 epimorphism, 7
small, 11 split, 7
coboundaries, see boundaries equivalence of categories, 13
cocycles, see cycles exact functor, 54
cohomology, see homology exact sequence, 35
coimage, 33 short, 35
cokernel, 32 Ext, 78
colimit, 18 is balanced, 88
complex, 63 Yoneda-Ext, 82
cone, 66
contravariant functor, see functor, co- faithful functor, see functor, faithful
variant Five lemma, 40
coproduct, 19 flat module, 59
cosyzygy, 81 free module, 16
counit is projective, 56
of an adjunction, 17 full functor, see functor, full

125
126 INDEX

full subcategory, see subcategory, full morphism, 5

functor, 8
contravariant, 8 natural isomorphism, 10
covariant, 8 natural transformation, 10
dense, 9 null-homotopic, 69
faithful, 9
full, 9 object, 5
octahedral axiom, 102
global dimension, 90 opposite category, 6
Gp, 6 Ore condition, 110

Hn (−), 64 poset category, 7

hereditary, 90 pre-abelian category, 33
Hom-functor pre-additive category, 29
contravariant, 9 presheaf
covariant, 9 on a category, 9
Hom-tensor-adjunction, 58 on a poset, 9
homology, 64 product, 19
long exact sequence, 65 projective, 54
homotopic morphisms, 69 projective dimension, 91
homotopy category, 69 projective resolution, 71
Horseshoe lemma, 73 as functor, 72
pullback, 20
image, 33 pushout, 20
injective, 54
injective dimension, 91 quasi-isomorphism, 69
injective resolution, 71
isomorphism, 8 retraction, 7
right derived functor
kernel, 32 total, 119
right exact functor, 54
left derived functor roof, 110
total, 119
left exact functor, 54 Salamander Lemma, 51
limit, 18 section, 7
semisimple, 90
mapping cone, see cone Set, 6
matrix notation, 31 shift, 66
monomorphism, 7 short exact sequence, 35
split, 7 small category, 11
INDEX 127

Snake lemma, 44
split epimorphism, 7
split monomorphism, 7
subcategory, 7
full, 7
suspension, see shift
syzygy, 80

tensor product, 56
is right exact, 59
Top, 6
Tor, 78
is balanced, 90
total complex, 87
total derived functor, 119
total left derived functor, 119
total right derived functor, 119
triangulated category, 101

unit
of an adjunction, 17

Yoneda embedding, 13
Yoneda lemma, 12
Yoneda-Ext, 82

Zn (−), 64
zero-object, 29