Homological Algebra

Notes by Sebastian Monnet

Based on lectures by Kobi Kremnitzer

October 10, 2021

1
Contents

1 Some Random Comments 5

2 Exact Sequences and Homology 6

2.1 Exactness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2.2 Short Exact Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.3 Chain Complexes and Homology . . . . . . . . . . . . . . . . . . . . . . 12

2.4 Homotopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3 Abelian Categories 18

3.1 Roadmap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3.2 Some Terminology from Category Theory . . . . . . . . . . . . . . . . . 18

3.3 Ab-enriched Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

3.4 Additive Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

3.5 Pre-abelian Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3.6 Abelian Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3.7 Connection with R−mod . . . . . . . . . . . . . . . . . . . . . . . . . . 24

4 Exact Functors 29

4.1 Left- and Right- Exact Functors . . . . . . . . . . . . . . . . . . . . . . 29

4.2 Exact Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

4.3 Specific Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

2
5 Projectives and Injectives 35

5.1 Projective Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

5.2 Injective Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

5.3 Resolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

5.4 R-mod Has Enough Injectives . . . . . . . . . . . . . . . . . . . . . . . 44

6 Derived Functors 49

6.1 δ-functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

6.2 Building Projective Resolutions . . . . . . . . . . . . . . . . . . . . . . 50

6.3 Left Derived Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

6.4 F -acyclic Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

6.5 Flat Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

6.6 Right Derived Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

7 Defining Ext and Tor 65

7.1 Tor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

7.2 Ext . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

8 Balancing Ext and Tor 68

8.1 Mapping Cones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

8.2 Double Complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

8.3 Total Complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

8.4 Non-Canonical Orderings on Double Complexes . . . . . . . . . . . . . 73

3
 8.5 Balancing Tor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

8.6 Balancing Ext . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

9 Ring Structure on Ext 82

9.1 Reinterpreting Ext . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

9.2 Yoneda Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

10 Universal Coefficient Theorem 88

11 Koszul Complexes 91

12 Ext and Extensions 95

12.1 Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

12.2 Yoneda Ext Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

13 Group (Co)homology 101

13.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

13.2 First Homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

13.3 The Norm Element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

13.4 Finite Cyclic Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

13.5 Free Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

13.6 Crossed Homomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . 108

13.7 Bar Complex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

13.8 Group Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

4
1 Some Random Comments

Here are some comments that would have been helpful for me when I was learning
Homological Algebra. They’re not important at all, so feel free to ignore them.

I will sometimes use the word “obvious” in contexts like “this is defined in the obvious
way”. This doesn’t mean that the thing is easy, but rather that it’s defined in the only
way that makes sense. It might still take some work to see what that way is.

Of the algebra-related courses in Michaelmas of fourth year, I think Homological Alge-

bra and Category Theory are the two hardest. However, a lot of the difficulty is shared.
Homological algebra is made much harder if you aren’t familiar with categories and
functors, and working with the concepts in homological algebra trivialises a fair bit of
the category theory course. Therefore, I recommend taking both. Also, if you want to
do a PhD in anything related to algebra, category theory is invaluable.

Homological algebra generally takes place on two different levels of abstraction, namely
the less abstract R-modules vs the more abstract “abelian category”. The distinction
between these settings is often blurred (indeed the Freyd-Mitchell Embedding Thereom
tells us that they are equivalent), and we will often work implicitly in R-modules, since
it is much simpler to do so.

5
2 Exact Sequences and Homology

The concept of an exact sequence seems at first a little arbitrary, but it turns out to be
incredibly useful. There is another type of sequence, called a chain complex, that is
in some sense close to being exact. From an algebraic standpoint, homology measures
the failure of a chain complex to be exact. The usual motivation for homology comes
from Algebraic Topology, and I recommend learning at least the definitions of simplicial
and singular homology, since they provide very useful context.

For now, we will work exclusively with R-modules. In the following section, we will
generalise the notions to the setting of abelian categories.

2.1 Exactness

Definition 2.1. A sequence of R-modules is a collection (An , fn )n∈Z , where the An

are R-modules and the fn : An → An−1 are module homomorphisms.

Definition 2.2. A cosequence1 of R-modules is a collection (An , f n )n∈Z , where the

An are R-modules, and the f n : An → An+1 are module homomorphisms.

Definition 2.3. The sequence

fn+1 fn
. . . → An+1 → An → An−1 → . . .

is exact at An if im fn−1 = ker fn .

Definition 2.4. A sequence is exact if it is exact at every term.

Exactness is defined in the same way for cosequences.

Example 2.5. Consider the sequence2

2
0 → Z → Z → 0.
1
This terminology is nonstandard, and a little silly, but I think it helps the flow of these notes.
2
Being pedantic, we defined sequences to be indexed by Z. This sequence, as written, has only
four terms, but we can imagine there being infinitely many zeros on either side, so it does fit our
definition of sequences.

6
This sequence is exact at the first copy of Z, but not the second.

Example 2.6. More generally, a sequence

f
0→A→B→0

is exact at A if and only if f is injective, and it is exact at B if and only if f is surjective.

Therefore, the statement that
0→A→B→0

is exact is equivalent to the statement that A → B is an isomorphism.

The most recent example hints at the power of exact sequences. They give us a new
language for rephrasing familiar statements about algebra. The fact that we specify the
maps involved allows us to discuss these concepts with more precision. For instance,
the exact sequences
2 3
0 → Z → Z, 0→Z→Z

both tell us that Z has a subgroup isomorphic to Z. However, the sequences are talking
about different subgroups (namely 2Z and 3Z respectively), and they tell us exactly
where those subgroups are.

Definition 2.7. Let {An } and {Bn } be sequences of R-modules. A morphism from
{An } to {Bn } is a collection of maps An → Bn such the diagram

... An+1 An An−1 ...

... Bn+1 Bn Bn−1 ...

commutes.

Sequences form a category with these morphisms, and the identity morphism is given
by the identity map An → An for each n. It is easy to check that a morphism of
sequence is an isomorphism1 if and only if each map An → Bn is an isomorphism.
1
In the sense of category theory.

7
2.2 Short Exact Sequences

In algebra, we often take quotients of modules by submodules. The “exact sequence

version” of this concept is the short exact sequence.

Definition 2.8. A short exact sequence is an exact sequence of the form

i π
0 → A → B → C → 0.

In rough terms, up to isomorphism, a short exact sequence just tells us that A is a

submodule of B and C is the quotient. The following lemma makes this precise.

Lemma 2.9. Let

i π
0→A→B→C→0

be a short exact sequence. There is a commutative diagram

i π
0 A B C 0
∼
= ∼
=

0 A0 B B/A0 0,

where A0 is a submodule of B and the maps on the bottom are the obvious maps.

Proof. By exactness at A, the map i is injective, so we may set A0 = i(A). Let c ∈ C.

Since π is surjective (by exactness at C, there is some b ∈ B with π(b) = c. Then
define ϕ(c) = b + A0 ∈ B/A0 . The map ϕ is well-defined, since if π(b) = π(b0 ) = c, then
b − b0 ∈ ker π = im i = A0 , so b + A0 = b0 + A0 . It is easy to check that ϕ is an R-module
homomorphism, and that the diagram commutes. To see that ϕ is an isomorphism, we
can define the inverse map
b + A0 7→ π(b),

and use similar ideas to check that it is a well-defined inverse to ϕ.

The kernel and cokernel of a map have the following universal properties from category
theory, which we take as definitions.

Definition 2.10 (Categorical kernel). The kernel of a map f : X → Y is a morphism

i : K → X such that

8
 1. f ◦ i = 0.

2. If ĩ : K̃ → X is a map with f ◦ ĩ = 0, then there is a unique morphism ϕ : K̃ → K

with i ◦ ϕ = ĩ.

In this situation, we say that i exhibits K as the kernel of f .

Definition 2.11 (Categorical cokernel). The cokernel of a map f : X → Y is a

morphism q : Y → C such that

1. q ◦ f = 0.

2. If q̃ : Y → C̃ is a map with q̃ ◦ f = 0, then there is a unique map ϕ : C → C̃

such that ϕ ◦ q = q̃.

In this situation, we say that q exhibits C as the cokernel of f .

Lemma 2.12. Let

i π
0→A→B→C→0

be a sequence. The following are equivalent:

1. The sequence is exact.

2. The map i exhibits A as the kernel of π, and the map π exhibits C as the cokernel
of i.

Proof. This basically follows from Lemma 2.9, together with the explicit constructions
of kernels and cokernels in R-mod (namely ker π = {x ∈ B : π(x) = 0} and coker i =
B/ im i).

Definition 2.13. A short exact sequence is split if it is isomorphic to one of the form

0 → A → A ⊕ B → B → 0,

where the maps are inclusion and projection.

i π
Lemma 2.14 (Splitting Lemma). Let 0 → A → B → C → 0 be a short exact sequence.
The following are equivalent:

9
 1. The short exact sequence is split.

2. There exists a retraction r : B → A.

3. There exists a section s : C → B.

Proof. The proof, as I have presented it, is much longer than any other version I have
seen elsewhere. This is because I have been very formalistic, being explicit about lots
of details that most authors find “obvious”. I’d probably recommend looking elsewhere
for an easier proof to follow, and consulting this one if you get stuck on any details.

We will show that (1) implies (2), then that (2) is equivalent to (3), and finally that
(2) and (3) implies (1).


Step 1: (1) =⇒ (2) and (3) .

Suppose that the short exact sequence is split (i.e. that (1) holds). Then we have an
isomorphism of short exact sequences:

i π
0 A B C 0
id ϕ id
i0 π0
0 A A⊕C C 0

where i0 and π 0 are the natural maps.

Let b ∈ B. Define r(b) to be the projection of ϕ(b) onto A, and for c ∈ C let s(c) =
ϕ−1 (0, c).

For a ∈ A, commutativity of the diagram tells us that i0 (a) = ϕ(i(a)). Also, by

definition of r we have i0 r(b) = ϕ(b) for all b ∈ B. It follows that

i0 (a) = ϕ(i(a)) = i0 ri(a),

so a = ri(a) by injectivity of i0 , which means that r is a retraction. So (2) holds.

For c ∈ C, we have
c = π 0 (0, c) = π 0 ◦ ϕ(s(c)) = π(s(c))

by commutativity of the diagram, so πs = idC , hence s : C → B is a section. So (3)

holds.

10
Step 2: (2) ⇐⇒ (3).

Suppose that (2) holds. Let r : B → A be the retraction.

Let c ∈ C. We would like to define s(c) = b − ir(b), where b ∈ π −1 (c). Such a b exists
because π is surjective.

To see that s is well-defined, suppose that b, b0 ∈ π −1 (c). Then b − b0 ∈ ker π = im i so

b = i(a) + b0 for some a ∈ A. It follows that

b − ir(b) = i(a) + b0 − ir(i(a) + b0 )

= i(a) + b0 − i(a) − ir(b0 )
= b0 − ir(b0 ).

So s is well-defined.

We have π ◦ s(c) = π(b − ir(b)) = π(b), since ir(b) ∈ im i = ker π. So πs = idC , hence
s is a section.

Suppose conversely the (3) holds, so a section s : C → B exists. Let b ∈ B. Then we

have
π(b − sπ(b)) = 0,

so b − sπ(b) ∈ ker π = im i, so b − sπ(b) = i(a) for some a ∈ A. By injectivity of i,

we have a well-defined function r : B → A, b 7→ a. It is easy to check that this is a
homomorphism (using the fact that i is a homomorphism).

Let x ∈ A, and write b = i(x), a = r(b). Then we have b − sπ(b) = i(a), which means
that
i(a) = i(x) − sπi(x) = i(x),

since πi = 0. So i(a) = i(x), hence a = x by injectivity of i, and we have x = a = ri(x).

So r is a retraction.

Step 3: (2) and (3) =⇒ (1).

Suppose that we have a section s : C → B and a retraction r : B → A. We define

maps

11
 ϕ
B A⊕C
ψ

by ϕ(b) = (r(b), π(b)) and ψ(a, c) = i(a) + s(c). It is easy to check that these maps are
mutually inverse isomorphisms.

2.3 Chain Complexes and Homology

Definition 2.15. A chain complex of R-modules is a sequence C∗ = {Cn , dn } such

that dn ◦dn+1 = 0 for all n. The map dn is called the differential of the chain complex.

Definition 2.16. A cochain complex1 is a cosequenece with the condition that the
composition of any two successive maps is zero.

Definition 2.17. A chain complex (resp. cochain complex) is called acyclic if it is

an exact sequence (resp. cosequence).

In other words, a chain complex is a sequence such that each “double map” is zero.
It is easy to see that chain complexes form a full subcategory of sequences, which we
denote Ch∗ (R).

Definition 2.18. A morphism of chain complexes (resp. cochain complexes) is called

a chain map (resp. cochain map).

Remark 2.19. It is common to abuse notation and denote all differentials by d. This
is justified more formally if we imagine that d is a function from the graded module
⊕n Cn to itself.

Remark 2.20. Annoyingly, there seems to be no agreed-upon name for the modules
Cn that constitute the chain complex. I have seen these referred to in many ways, such
as “the nth space”, “the degree n part”, “the nth term”, etc.

The condition that the double differential is zero is equivalent to im dn+1 ⊆ ker dn for
all n. Clearly then, we have a well-defined quotient module

ker dn
,
im dn+1
1
Unlike the cosequence, this terminology is completely standard.

12
and the chain complex C∗ is acyclic if and only if this quotient vanishes for all n.
Therefore, we may view the quotient as some sort of “obstruction to exactness”. It
turns out that this obstruction is very useful. We call it the nth homology of the
chain complex C∗ , and denote it by Hn (C∗ ). Since this is so important, we will restate
the definition more formally, while introducing some terminology.

Definition 2.21. Let C∗ be a chain complex. An element c ∈ Cn is called a cycle if

dn c = 0, and a boundary if c = dn+1 c0 for some c0 ∈ Cn+1 . We denote the cycles and
boundaries in degree n by Zn C and Bn C respectively. The nth homology of C∗ is the
quotient module
Zn C
Hn (C∗ ) = .
Bn C
Definition 2.22. The corresponding notions for a cochain complex are called cocy-
cles and coboundaries, and they are denoted Z n C and B n C respectively. The nth
cohomology is then defined to be

n Z nC
H (C) = n .
B C
Remark 2.23. The terms “cycle” and “boundary” are motivated by topology. In
topology, we work with a chain complex whose differential takes a “thing” (the thing
is kind of, but not really, a subspace of the topological space) to the boundary of the
thing. For instance, if the thing is a path, then the differential is the endpoint minus
the startpoint. If the thing is a disc, then the differential is (kind of) its boundary
circle. By definition, then, an element of the image of the differential is a boundary.
We call elements of the kernel “cycles”, since in the case of a path, the endpoint minus
the startpoint will be zero if and only if the path is a loop (i.e. it “cycles round”).

Definition 2.24. The nth homology of a chain complex C∗ is the module

Zn C ker dn
Hn (C∗ ) = = .
Bn C im dn+1

The slogan is that homology is “cycles modulo boundaries”. Homology has the useful
property that each Hn is a functor from chain complexes to R-modules. We now make
this precise.

Lemma 2.25. Let f : C∗ → D∗ be a morphism of chain complexes. There is a well-

13
defined module homomorphism

Hn (f ) : Hn (C∗ ) → Hn (D∗ ),

given by
c + Bn C 7→ f (c) + Bn D.

Proof. First of all, we claim that f restricts to a map Zn C → Zn D. This follows from
commutativity of
d
Cn Cn−1
fn fn−1

Dn d
Dn−1 ,

since any c ∈ Zn C has d(fn c) = fn−1 (dc) = fn−1 (0) = 0, so fn (c) ∈ Zn D. Similarly,
f takes boundaries to boundaries (use the same commutative diagram, but replace n
with n + 1), so the composition

Zn C → Zn D → Zn D/Bn D

kills Bn C, hence descends to a well-defined homomorphism

Hn (C∗ ) → Hn (D∗ ).

Definition 2.26. We call the homomorphism Hn (f ) the induced homomorphism

of f .

When n is clear, we often denote the induced map Hn (f ) by f∗ .

Lemma 2.27. Let induced homomorphism has the following useful properties.

1. If f, g : C∗ → D∗ are chain maps, and r ∈ R, then

(f + rg)∗ = f∗ + rg∗ .

2. If f : C∗ → D∗ and g : D∗ → E∗ are chain maps, then

(g ◦ f )∗ = g∗ ◦ f∗ .

14
 3. For any chain complex C∗ , we have (idC∗ )∗ = idHn (C) .

Proof. Follows straight from the definitions.

The above lemma tells us that Hn is a functor Ch∗ (R) → R-mod, and also that the
maps of Hom-sets are R-linear.

There are several notions of equivalence between chain complexes. The most obvious
is isomorphism. Also quite straightforward is the notion of quasi-isomorphism.

Definition 2.28. A quasi-isomorphism is a chain map f : C∗ → D∗ such that

f∗ : Hn (C∗ ) → Hn (D∗ ) is an isomorphism for all n.

Remark 2.29. The relation “there exists a quasi-isomorphism from C∗ to D∗ ” is not

an equivalence relation, because it is not transitive1 . The transitive closure of the
relation is useful, since it allows us to define something called the derived category
of chain complexes.

2.4 Homotopy

In topology, homotopy is an equivalence relation of maps, and we can use it to define

the notion of homotopy equivalence of spaces. We mirror the development, using chain
complexes instead of topological spaces.

Definition 2.30. Let f, g : C∗ → D∗ be chain maps. A chain homotopy from f to

g is a collection of maps hn : Cn → Dn+1 such that for each n, we have

(D)
fn − gn = hn−1 ◦ d(C)
n + dn+1 ◦ hn .

If there is a chain homotopy from f to g, then we write f ' g.

1
See [Link]
alence-relation for a counterexample.

15
It is common to represent the situation by the following diagram.

Cn+1 Cn Cn−1
f g
hn hn−1
Dn+1 Dn Dn−1

Note that this diagram is not commutative. It is just helpful for seeing where the
various maps point.

Lemma 2.31. The relation ' is an equivalence relation on chain maps C∗ → D∗ .

Proof. For reflexivity, just take hn = 0 for all n. For symmetry, if h is a chain homotopy
from f to g, then −h is a homotopy from g to f . For transitivity, if h and h0 are chain
homotopies from f to g and g to k respectively, then h + h0 is a chain homotopy from
f to k.

Lemma 2.32. Suppose that chain maps f, g : C∗ → D∗ are chain homotopic. Then
the induced maps f∗ , g∗ : Hn (C) → Hn (D) are equal.

Proof. Let h be a chain homotopy from f to g. We have

(D)
fn − gn = hn−1 ◦ d(C)
n + dn+1 ◦ hn ,

for each n. Let x ∈ Hn (C). Then x = [c] for some cycle c ∈ Zn C. We have

f∗ (x) − g∗ (x) = [fn (c) − gn (c)]

(D)
= [hn−1 ◦ d(C)
n (c) + dn+1 ◦ hn (c)]
(D)
= [dn+1 ◦ hn (c)]
= 0,

where the third equality comes from the fact that c is a cycle.

Now that we have defined homotopy of maps, we can define homotopy equivalence of
chain complexes just as we did in topology.

16
Definition 2.33. Let C∗ and D∗ be chain complexes. A chain homotopy equiva-
lence from C∗ to D∗ is a tuple (f, g, h1 , h2 ), where

f
C∗ D∗
g

are chain maps, h1 is a homotopy from gf to idC∗ , and h2 is a homotopy from f g to

idD∗ . If a chain homotopy equivalence enxists, then we say that C∗ and D∗ are chain
homotopy equivalent.

When it is clear what we mean (which is almost always), we drop the word “chain”
from these terms, just referring to homotopies and homotopy equivalences.

Lemma 2.34. If C∗ and D∗ are homotopy equivalent chain complexes via (f, g, h1 , h2 ),
then f∗ and g∗ are mutually inverse isomorphisms.

Proof. We have
g∗ ◦ f∗ = (g ◦ f )∗ = (idC∗ )∗ = idHn (C) ,

and similarly f∗ ◦ g∗ = idHn (D) .

Corollary 2.35. Homotopy equivalent chain complexes have isomorphic homology.

Proof. Immediate.

17
3 Abelian Categories

As was the case with Galois’s Galois Theory, freshly invented concepts are often messy.
It can take decades, or even centuries, for mathematicians to hammer the theory into a
more elegant shape. This was not the case for Grothendieck, who spewed out modern
homological algebra, more or less fully-formed, in his legendary “Tohoku Paper” of
1957.

Central to Grothendieck’s treatment is a type of category called an “abelian category”.

This is basically a generalisation of R-mod. It retains the nice properties we took
advantage of in the previous section, but allows us to extend the methods to many
exotic areas of mathematics.

3.1 Roadmap

To reach abelian categories, we need a few definitions. These can be difficult to absorb,
so we have broken the section up as much as possible. It may help to keep the following
diagram in mind. Its meaning will become clear as the definitions are introduced.

Category
Abelian group structure on Hom-sets

Ab-enriched category
0 object
Finite coproducts

Additive category
Kernels and cokernels

Pre-abelian category
Every mono is kernel of its cokernel
Every epi is cokernel of its kernel

Abelian category

3.2 Some Terminology from Category Theory

We rattle off some definitions from category theory. Hopefully, these are familiar.

18
Definition 3.1. Let C be a category, and let x ∈ C. We say that x is terminal if for
every c ∈ C, there is exactly one morphism c → x. Dually, we say that x is initial if
for every c ∈ C, there is exactly one morphism x → c.

Definition 3.2. A zero object in a category is an object that is both initial and
terminal.

Definition 3.3. A monomorphism is a morphism f such that f g1 = f g2 =⇒ g1 =

g2 for any morphisms g1 , g2 . Dually, an epimorphism is a morphism f such that
g1 f = g2 f =⇒ g1 = g2 for any g1 , g2 .

3.3 Ab-enriched Categories

Definition 3.4. A pre-additive or Ab-enriched category is a category in which

every hom-set is equipped with the structure of an abelian group, such that the com-
position
Hom(X, Y ) × Hom(Y, Z) → Hom(X, Z)

is Z-bilinear.

Proposition 3.5. In an Ab-enriched category, any initial object is also terminal.

Proof. Let ∗ be initial. Then id∗ is the unique element of Hom(∗, ∗), so id∗ is zero in
this group. Then since composition respects the group structures, we have for any map
f : ∗ → A,
f = f ◦ id∗ = f ◦ 0 = 0

so ∗ is terminal.

Proposition 3.6. If C is an Ab-enriched category, then so is its opposite category

C op .

Proof. For X, Y ∈ C op , the sets

HomC op (X, Y ) = HomC (Y, X)

19
are already endowed with the structure of an abelian group. Thus, we have only to
prove that composition is bilinear. Let X, Y, Z ∈ C and let

f, f 0 ∈ HomC op (X, Y ), g ∈ HomC op (Y, Z).

Then
g ◦op (f + f 0 ) = (f + f 0 ) ◦ g = f ◦ g + f 0 ◦ g = g ◦op f + g ◦op f 0 .

Similarly composition is bilinear in the other argument as well.

Proposition 3.7. In an Ab-enriched category, a binary product is also a binary co-

product.

Proof. Let X1 , X2 be elements of an Ab-enriched category C. Suppose that X1 and

X2 have a product X1 × X2 in C, with projections pk : X1 × X2 → Xk . By definition
of products, there are unique morphisms ik : Xk → X1 × X2 such that the following
diagrams commute.

X1 0 0 X2
i1
i2
id X1 × X 2 X1 × X2 id
p1 p2 p1 p2

X1 X2 X1 X2

Then we have

p1 ◦ (i1 p1 + i2 p2 ) = p1 , p2 ◦ (i1 p1 + i2 p2 ) = p2 .

By defintion of products, id : X1 × X2 → X1 × X2 is the unique morphisms with

pk ◦ id = pk for each k, so i1 p1 + i2 p2 = idX1 ×X2 . We claim that

X1 X2
i1 i2

X1 × X2

20
 `
is a universal cocone, so that X1 × X2 = X1 X2 . Suppose that

X1 X2

f1 f2
A

is another cocone. Then we have a map

ϕ = f1 ◦ p1 + f2 ◦ p2 : X1 × X2 → A,

which is easily seen to give a commutative diagram

X1 X2
i1 i2

f1
X1 × X2 f2
ϕ

A.

It remains to show that ϕ is unique. To see this, note that for any such ϕ we have

ϕ = ϕ ◦ idX1 ×X2
= ϕ ◦ (i1 p1 + i2 p2 )
= ϕi1 ◦ p1 + ϕi2 ◦ p2
= f 1 ◦ p1 + f 2 ◦ p2 .

Proposition 3.8. In an Ab-enriched category, all binary coproducts are also binary
products.

Proof. This is dual to Proposition 3.7. We will explain the duality explicitly.
`
Let C be Ab-enriched, and let X1 , X2 ∈ C have a coproduct X1 X2 . Then the
object X1 X2 is also an object of C op (since C and C op have the same objects), and
`

it is a product of X1 and X2 in the category C op . Since C is Ab-enriched, so is C op ,

which means that all binary products are binary coproducts in C op . Since X1 X2 is
`

a product in C op , it is therefore is also a coproduct in C op , which makes it a product of

21
X1 and X2 in C.

By Propositions 3.7 and 3.8, binary products and binary coproducts are the same
object in an Ab-enriched category. This motivates the following definition.

Definition 3.9. Let C be an Ab-enriched category, and let x, y ∈ C. If x and y have

a product in C, then it is called the biproduct of x and y, which we denote by x ⊕ y.

Definition 3.10. Let F : A → B be a functor between Ab-enriched categories. Then

F is said to be additive if it preserves finite biproducts.

Lemma 3.11. For any ring R, the category R-mod is Ab-enriched.

Proof. For any two left R-modules A and B, the group HomR (A, B) is naturally an
abelian group under pointwise addition. It is easy to check that composition is bilinear.

3.4 Additive Categories

Definition 3.12. A category is additive if it is Ab-enriched and admits finite coprod-

ucts.

Lemma 3.13. Let A be an additive category. Suppose that i : a → b is a monomor-

phism in A and i ∈ Hom(a, b) is the zero morphism. Then a = 0.

Proof. Let x ∈ A. Since Hom(a, x) is an abelian group, it contains at least one

morphism (zero). Let f : a → x be any morphism. Then

α ◦ 0 = 0 = α ◦ f.

Since α is a monomorphism, we have f = 0. Therefore a is initial, hence it is the zero

object.

Lemma 3.14. Let A be an additive category. Suppose that q : a → b is an epimorphism

in A. If q = 0, then b = 0.

22
Proof. Since A is additive, the opposite category Aop is too. The map q is a monomor-
phism q : b  a in Aop , and it is still the zero morphism. Therefore by Lemma 3.13, b
is the zero object in Aop , hence in A.

Lemma 3.15. For any ring R, the category R-mod is additive.

Proof. We know that the direct sum exists and is a coproduct in R-mod.

3.5 Pre-abelian Categories

Definition 3.16. An additive category is pre-abelian if every morphism has a kernel

and cokernel.

Lemma 3.17. Let A be a pre-abelian category. Every monomorphism has kernel 0,

and every epimorphism has cokernel 0.

Proof. Let i : a  b be a monomorphism in A. Let

ker i
Ker i a

be the kernel of i. Then i ◦ ker i = 0 = i ◦ 0, so ker i is the zero morphism (since i is a

monomorphism). Since ker i is a monomorphism, we have Ker i = 0.

Lemma 3.18. For any ring R, the category R-mod is pre-abelian.

Proof. Let f : A → B be a morphism in R-mod. It is easy to check that Ker f = {a ∈

A : f (a) = 0} is a kernel in the categorical sense. Similarly, Coker f = B/ Im f is a
categorical cokernel.

3.6 Abelian Categories

Definition 3.19. An pre-abelian category is abelian if every monomorphism is the

kernel of its cokernel and every epimorphism is the cokernel of its kernel.

23
We explain a bit more what this definition means. Let i : x → y be a monomorphism
in an abelian category A. Then there is a natural map q : y → Coker i. The content
of (the first part of) the definition is that

q
i
x y Coker i
0

is a universal cone, so that we may say x = Ker q. The statement that every epimor-
phism is the cokernel of its kernel is similar.
Lemma 3.20. The category of left R-modules is an abelian category.

Proof. Let i : A → B be a monomorphism of R-modules. Then Coker i = B/i(A) and

the cokernel map is the quotient q : B → i(A) with q(b) = b + i(A). It is clear that
i(A) = Ker q in the set-theoretic sense, so i exhibits A as the kernel of q.

Let q : A → B be an epimorphism of R-modules. Let i : Ker q → A be the inclusion.

Then Coker i = A/ Ker q ∼
= B, so q exhibits B as the cokernel of i.
Lemma 3.21. If A is abelian, then so is Aop .

Proof. The notions of kernel and cokernel are dual, as are monomorphisms and epi-
morphisms. In particular, if i is a monomorphism in Aop , then it is an epimorphism in
A, so it is the kernel of its cokernel in A, which means that it is cokernel of its kernel
in Aop .
Remark 3.22. In my opinion, the preceding proof is not a rigorous argument, since
there are a lot of things to check that I have taken to be obvious. This sort of argument
“by duality” is pretty common, and checking all the details is often hard.
Lemma 3.23. If A is an abelian category and C is any category, then Fun(C, A) is
abelian.

Proof. See [Link]

-category-aj-is-abelian-category-if-a-is-abelian.

3.7 Connection with R−mod

Lemma 3.24. The category R-mod is an abelian category.

24
Proof. It is easy to check that R-mod is pre-abelian. Let i : M → N be a monomor-
phism (i.e. an injective homomorphism of R-modules).

The cokernel of i is the quotient map π : N → N/ im i. Then we have a short exact

sequence
i π
0 → M → N → N/ im i → 0,

which means that i exhibits M as the kernel of π, so i is the kernel of its cokernel.
Showing that every epimorphism is the cokernel of its kernel is similar.

There are certainly abelian categories other than R-mod but, miraculously, there is a
partial converse to Lemma 3.24. In the statement of the following theorem, we refer to
an exact functor. We haven’t defined this properly yet, but it basically just means
that if a sequence is exact, then so is its image under the functor.
Theorem 3.25 (Freyd-Mitchell Embedding Theorem). Let A be a small abelian cat-
egory. Then there is a ring R and an exact, fully faithful functor F : A → R-mod.
This functor embeds A as a full subcategory in R−mod, by which we mean that for all
M, N ∈ A, we have

HomA (M, N ) ∼
= HomR (F (M ), F (N )).

Proof. See Weibel, Page 25, Theorem 1.6.1.

Lemma 3.26. The Freyd-Mitchell embedding preserves kernels and cokernels.

Proof. Let f : x → y be a morphism in an abelian category A, and let F : A → R-mod

be the Freyd-Mitchell embedding. Consider the sequence

i f q
0 → Ker f → x → y → Coker f → 0.

Lemma 3.27. Let A be an abelian category and let F : A → R-mod be the embedding
from Theorem 3.25. Then F (0) = 0.

Proof. We have HomR-mod (F (0), F (0)) ∼

= HomA (0, 0), so there is only one R-module
homomorphism F (0) → F (0). For any R-module M , we have homomorphisms 0 :
M → M and idM : M → M , which means that idF (0) = 0, so F (0) = 0.

25
The following proof is a powerful application of the Freyd-Mitchell Embedding Theo-
rem.

Theorem 3.28 (Snake Lemma). Let A be an abelian category, and suppose that we
have a diagram in A of the form

f g
A B C 0
a b c

0 A0 f0
B0 g0
C0

where the rows are exact. Then there is a morphism δ : Ker c → Coker a such that the
following sequence is exact

δ
Ker a → Ker b → Ker c → Coker a → Coker b → Coker c.

Furthermore, if f is a monomorphism, then so is the natural map Ker a → Ker b, and

if g 0 is an epimorphism then so is the natural map Coker b → Coker c.

Proof. Suppose first that A = R-mod. Let z ∈ Ker c. Since g is onto, there is some
y ∈ B such that z = g(y). We have g 0 b(y) = cg(y) = c(z) = 0, so b(y) ∈ Ker g 0 = Im f 0 ,
so there is some x0 ∈ A0 such that f 0 (x0 ) = b(y).

Suppose that y1 , y2 and x01 , x02 are choices for y, x0 above. Then we have g(y1 ) =
g(y2 ) = z, so y1 − y2 ∈ Ker g = Im f , so y1 − y2 = f (x) for some x ∈ A. Now
f 0 (x01 − x02 ) = b(y1 − y2 ) = bf (x) = f 0 a(x). The map f 0 is injective, so a(x) = x01 − x02 ,
hence x01 + Im a = x02 + Im a as elements of A0 / Im a = Coker a. Therefore we have a
well-defined map δ : Ker c → Coker a given by

δ(z) = x + Im a, where f 0 (x) = b(y) for some y ∈ B with g(y) = z.

We claim that δ is a an R-module homomorphism. Let z1 , z2 ∈ Ker c and for each

i let (xi , yi ) be the pair in the defition of δ(xi ). Then we have g(y1 − y2 ) = z1 − z2
and f 0 (x1 − x2 ) = b(y1 − y2 ), so δ(z1 − z2 ) = x1 − x2 . Similarly, for r ∈ R we have
δ(rx1 ) = rδ(x1 ). Therefore δ is indeed a module homomorphism.

The final part of the theorem follows immediately from the definitions of the natural
maps. Therefore the theorem is proved in R-mod.

26
Now let A be any abelian category. Let F : A → R-mod be the embedding from
Theorem 3.25. Then
F (f ) g
F (A) F (B) F (C) 0
F (a) F (b) F (c)

0 F (A0 ) F (f )0
F (B 0 ) F (g)0
F (C 0 )

is a diagram in R-mod with exact rows. Then by the case in R-mod, there is a
morphism δ∗ : Ker F (c) → Coker F (a) such that

∗ δ
Ker(F (a)) → Ker(F (b)) → Ker(F (c)) → Coker(F (a)) → Coker(F (b)) → Coker(F (c))

is exact.

Definition 3.29. A short exact sequence of chain complexes is a sequence

0 → A∗ → B∗ → C∗ → 0

of chain complexes, such that each

0 → An → Bn → Cn → 0

is a short exact sequence.

Corollary 3.30 (Sometimes also called the Snake Lemma). Suppose that 0 → A∗ →
B∗ → C∗ → 0 is a short exact sequence of chain complexes in an abelian category A.
Then for each n, there is a connecting map δn : Hn (C) → Hn−1 (A) such that we
have a long exact sequence

nδ
. . . → Hn (C) → Hn−1 (A) → Hn−1 (B) → Hn−1 (C) → . . . .

Proof. This follows from Lemma 3.28, applied to the diagram

Cn
An Bn ∂n+1 Cn+1
0

Zn−1 B
0 Zn−1 A Zn−1 B Zn−1 A
.

27
It is somewhat nontrivial what the maps in this diagram are, but if you think about
it, you’ll see that there is only one thing they could be, and that they are actually
well-defined with exact rows.

Remark 3.31. We can also prove this directly (and I think it makes more sense to do
so) using a diagram chase in R-mod, and then applying the Freyd-Mitchell Embedding
Theorem to transfer the result to general abelian categories.

Remark 3.32. It is actually possible to avoid the FM Embedding Theorem and do

diagram chases rigorously in any abelian category. For more information, see https:
//[Link]/2007/09/28/diagram-chases-done-right/.

28
4 Exact Functors

Since we care about exact sequences, it seems reasonable to study functors that preserve
exactness. In fact, it will be fruitful to study slightly more general functors as well,
that only preserve exactness on the left or on the right.

4.1 Left- and Right- Exact Functors

Definition 4.1. A functor F is left-exact if for every short exact sequence 0 → A →

B → C → 0, the sequence

0 → F (A) → F (B) → F (C)

is exact. Similarly, F is right-exact if instead

F (A) → F (B) → F (C) → 0

is always exact.

Lemma 4.2. If F : A → B is left-exact, and i is a monomorphism in A, then F (i) is

a monomorphism in B.

Proof. If i : A → B is a monomorphism, then we have a short exact sequence

0 → A → B → coker i → 0.

Therefore, 0 → F (A) → F (B) is exact, so F (i) is a monomorphism.

Lemma 4.3. Let F : A → B be a functor between abelian categories. The following

are equivalent:

1. F is left-exact.

2. For any exact sequence 0 → A → B → C, the corresponding sequence 0 →

F (A) → F (B) → F (C) is also exact.

29
 i π
Proof. It is trivial that (2) =⇒ (1). Suppose that (1) holds. Let 0 → A → B → C be
exact. Then we have a short exact sequence 0 → A → B → im π → 0, and therefore
the sequence
0 → F (A) → F (B) → F (im π)

is exact. Now, im π → C is a monomorphism, so F (im π) → F (C) is too by Lemma

4.2. Therefore



ker F (B) → F (C) = ker F (B) → F (im π) → F (C)


= ker F (B) → F (im π)


= im F (A) → F (B) .

Corollary 4.4. Let F : A → B be a functor between abelian categories. The following

are equivalent:

1. F is right-exact.

2. For any exact sequence A → B → C → 0, the corresponding sequence F (A) →

F (B) → F (C) → 0 is exact.

Proof. This follows from Lemma 4.3 by duality.

Lemma 4.5. Let A be an abelian category, and consider maps

f g
A→B→C

in A. Suppose that for all Z ∈ A, the sequence

−◦f −◦g
Hom(A, Z) ← Hom(B, Z) ← Hom(C, Z) ← 0

is exact. Then A → B → C → 0 is exact.

Proof. We need to show that g exhibits C as the cokernel of f . Suppose that α : B → Z

is some map with α ◦ f = 0. Then

α ∈ ker(− ◦ f ) = im(g ◦ −),

30
so α = ϕ ◦ g for a unique map ϕ : C → Z. This is precisely the universal property of
the cokernel.

Lemma 4.6. Suppose we have an adjunction

F
A B
G

of additive functors between abelian categories, where F is the left adjoint. Then F is
right-exact.

Proof. Let
0→A→B→C→0

be a short exact sequence in A, and let Z ∈ B. Then G(Z) ∈ A, so

−◦i −◦π
Hom(A, G(Z)) ← Hom(B, G(Z)) ← Hom(C, G(Z)) ← 0

is exact by left-exactness of Hom. Therefore,

−◦i −◦π
Hom(F (A), Z) ← Hom(F (B), Z) ← Hom(F (C), Z) ← 0

is exact, so
F (A) → F (B) → F (C) → 0

is exact by Lemma 4.5.

Corollary 4.7. If F, G are as above, then G is left exact.

Proof. We apply some hand-wavy duality. It might be worth working through the
details here to convince yourself that the claims are true. Since G is a right adjoint,
the functor G : Dop → C op is a left adjoint, so it is right exact. Therefore, G : D → C
is left exact.

4.2 Exact Functors

Definition 4.8. A functor is exact if it is left-exact and right-exact.

31
Lemma 4.9. Suppose that we have a long exact sequence

fn−1 fn
. . . → An−1 → An → An+1 → . . .

and and exact functor F . Then

. . . → F (An−1 ) → F (An ) → F (An+1 ) → . . .

is also exact.

Proof. Since we only have to check exactness at each term, it suffices to show that for
an exact sequence
f g
A → B → C,

the sequence
F (f ) F (g)
F (A) → F (B) → F (C)

is also exact. We prove this with a diagram-chase. Note that

0 → ker f → A → im f → 0,

0 → ker g → B → im g → 0,

and
0 → im g → C → coker g → 0

are short exact sequences. We can fit these short exact sequences into the following
commutative diagram:

0 0 0

ker f im g
π2 i2

f g
A B C
π1 i1

im f coker g

0 0 0

32
Note that the diagonals are exact. Applying F to the diagram (and removing some
redundant terms) gives a commutative diagram:

0 0

F (ker f ) F (im g)
F (π2 ) F (i2 )

F (f ) F (g)
F (A) F (B) F (C)
F (π1 ) F (i1 )

F (im f ) F (coker g)

0 0

Again the diagonals are exact. Since F (π1 ) is surjective, we have im F (f ) = im F (i1 ) =
ker F (π2 ) by exactness of at F (B). But F (i2 ) is injective, so ker F (g) = ker F (π2 ), and
it follows that
im F (f ) = ker F (π2 ) = ker F (g).

4.3 Specific Functors

Example 4.10. The following functors are not exact:

1. − ⊗Z Z/2,

2. HomZ (Z/2, −),

3. HomZ (−, Z/2).

To see this, consider the short exact sequence

·2
0 → Z → Z  Z/2 → 0.

Each functor takes this to a sequence that is not exact.

33
Lemma 4.11. Let A be an abelian category, and let M ∈ A be an object. Then the
functor HomA (M, −) : A → Ab is left-exact.

i π
Proof. Let 0 → A → B → C → 0 be a short exact sequence in A. We have to show
that the sequence

i◦− π◦−
0 → Hom(M, A) → Hom(M, B) → Hom(M, C)

is exact. For exactness at Hom(M, A), suppose that i ◦ ϕ = 0, where ϕ : M → A is

a map. Since i is a monomorphism and i ◦ ϕ = i ◦ 0, we have ϕ = 0. Therefore, the
sequence is exact at Hom(M, A).

Since π ◦ i = 0, we have im(i ◦ −) ⊆ ker(π ◦ −). Let ϕ ∈ ker(π ◦ −). Then π ◦ ϕ = 0.

Since i exhibits A as the kernel of π, there is a unique map f : M → A such that
i ◦ f = ϕ. Therefore, the sequence is exact at Hom(M, B).

Corollary 4.12. Let A be an abelian category and let M ∈ A be an object. Then the
functor
HomA (−, M ) : Aop → Ab

is left-exact.

Proof. This follows from the definition of the opposite category.

Corollary 4.13. Let R be a ring and M be an R-module. Then the functors

HomR (M, −) : R-mod → Ab, HomR (−, M ) : R-modop → Ab

are left-exact.

Lemma 4.14. For any ring R, the functor − ⊗R N : R-mod → R-mod is right-exact.

Proof. This follows from the adjunction

(− ⊗R N ) a HomR (N, −)

and Lemma 4.6.

34
5 Projectives and Injectives

For an object P of an abelian category A, the functor HomA (P, −) always left exact.
In this section, we study objects P such that the functor is also right exact. Such
objects are called projective. We also study the dual notion of injective objects.

5.1 Projective Objects

Definition 5.1. Let A be an abelian category. An object P ∈ A is said to be pro-

jective if HomA (P, −) : A → Ab is an exact functor.

We opted for the above definition of projective objects because it is the easiest to
state. However, the following equalence is very important, and Weibel uses it to define
projective objects.
Lemma 5.2. Let P be an object of an abelian category A. The following are equivalent:

1. The object P is projective.

2. The functor HomA (P, −) sends epimorphisms to epimorphisms.

3. For every epimorphism π : A → B and every morphism f : P → B, there exists

a morphism α : P → A such that the following diagram commutes:

A
∃α
∀π

P ∀f
B.

Proof. Clearly (1) =⇒ (2).

For (2) =⇒ (1), recall from Lemma 4.11 that F = HomA (P, −) is left-exact. Let

0→A→B→C→0

be a short exact sequence. Then by left-exactness of F , the sequence

0 → F (A) → F (B) → F (C) → 0

35
is exact at F (A) and at F (B). But the map B → C is an epimorphism, so (2) implies
that F (B) → F (C) is also an epimorphism, which means that we have exactness at
F (C) as well.

Condition (3) basically just says that when π is an epimorphism, the natural map

Hom(P, B) → Hom(P, A)

is surjective (i.e., for every f ∈ Hom(P, B), there is some α ∈ Hom(P, A) that maps to
f ). This precisely (2), so (2) ⇐⇒ (3).
Definition 5.3. We say that an R-module is projective if it is a projective object in
R-mod.
Lemma 5.4. Free R-modules are projective.

Proof. Let F = ⊕i Rei be a free R-module with basis {ei : i ∈ I}. Suppose that we
have a diagram
A
π

F f
B.

Since π is surjective, for each i there is some ai ∈ A with π(ai ) = f (ei ). Define the
map α : F → A by α(ei ) = ai .
Lemma 5.5. An R-module is projective if and only if it is a direct summand of a free
R-module.

Proof. Suppose that P is a direct summand of a free module. Then there is some R-
module P 0 such that P ⊕P 0 is free. Let π : A  B be a surjection and let f : P → B be
some map. Let f 0 : P ⊕ P 0 → B be the map f 0 (p, p0 ) = f (p). Since P ⊕ P 0 is free, hence
projective, f 0 has a lift α0 : P ⊕ P 0 → A. Now define α : P → A by α(p) = α0 (p, 0).

Suppose conversely that P is projective. Then we have a natural surjection

M
π: Rep → P, ep 7→ p,
p∈P

and taking f : P → P to be the identity gives us a section α of this surjection (since

P is projective). Therefore, the result follows by the Splitting Lemma.

36
5.2 Injective Objects

Injective objects are dual to projective objects.

Definition 5.6. An object I of an abelian category A is called injective if the object

I ∈ Aop is projective.

Lemma 5.7. For an R-module I, the following are equivalent:

1. I is injective.

2. The functor HomA (−, I) : Aop → Ab is exact.

3. The contravariant functor HomA (−, I) : A → Ab sends monomorphisms to epi-

morphisms.

4. for every monomorphism i : A → B and every map f : A → I, there exists a

g : B → I such that the following diagram commutes:

B
∃α
i

I f
A

Proof. This is the dual of Lemma 5.2.

Theorem 5.8 (Baer’s Criterion). Let M be a right R-module. The following are
equivalent.

1. M is injective.

2. For every right ideal I of R, every module homomorphism I → M can be extended

to a module homomorphism R → M .

Proof. The (1) =⇒ (2) is immediate from Lemma 5.7. The other direction is signifi-
cantly harder.

Assume that (2) holds. Fix some injection i : A  B of R-modules, and some map
f : A → M . Without loss of generality, assume that A ⊆ B and i is the inclusion. Let

37
Σ be the set whose elements are R-module maps α0 : A0 → M , where A ⊆ A0 ⊆ B and
α0 extends f (that is, the diagram

A A0
α0
f
M

commutes). We may give this set a partial order by saying that α0 ≤ α00 when A0 ⊆ A00
and α00 extends α0 . Suppose that α1 ≤ α2 ≤ . . . is an ascending chain in Σ, with
corresponding modules A1 ⊆ A2 ⊆ . . .. Let A0 = ∪An , and define α0 : A0 → M by
α0 (a) = αi (a) for a ∈ Ai . It is easy to check that α0 is a well-defined element of Σ,
and it is an upper bound on the chain (n.b. we are really just taking the colimit of the
chain).

Since Σ is a partially ordered set in which every ascending chain has an upper bound,
Zorn’s Lemma tells us that it has a maximal element, which we will call α0 : A0 toM .
To show that M is injective, we need to show that A0 = B, since we then have an
extension α of f to B.

Suppose that A0 6= B. Let b ∈ B \ A0 , and define A00 = A0 + Rb ⊆ B. Let I = {r ∈ R :

br ∈ A0 }. Then I is a right ideal of R, and we have a map

I → M, r 7→ α0 (br).

Since (2) holds, this extends to a map ϕ : R → M . We claim that there is a well-defined
map
α00 : A00 → M, a + br 7→ α0 (a) + ϕ(r),

where a ∈ A0 and r ∈ R. To see that this is well-defined, suppose that

a + br = a0 + br0 .

Then
a − a0 = b(r0 − r) ∈ A0 ∩ bR.

We have
α0 (a − a0 ) = α0 (b(r0 − r)) = ϕ(r0 − r),

38
since r − r0 ∈ I. Therefore, it follows that

α0 (a) + ϕ(r) = α0 (a0 ) + ϕ(r0 ),

so α00 is well-defined. But then α00 strictly extends α0 , contradicting maximality of α0 .

Therefore, A0 = B, so we are done.

Corollary 5.9. If R is a PID, then an R-module I is injective if and only if it is

divisible. That is, for all x ∈ I and r ∈ R \ {0} there exists q ∈ I such that x = rq.

The details of the proof can get in the way of the intuitive idea, which is quite simple.
Maybe try proving it yourself before reading on (use Baer’s Criterion).

Proof. Let I be an injective R-module, and let x ∈ I and r ∈ r \ {0}. Set J = rR

and define f : J → I by f (r) = x. By Baer’s Criterion, we may extend f to a
homomorphism f˜ : R → I. Then x = f (r) = f˜(r · 1) = r · f˜(1). So taking q = f˜(1),
we see that I is divisible.

Suppose conversely that I is a divisible R-module. Let J be an ideal of R and let

f : J → I be a module homomorphism. If J is the zero ideal, then trivially we may
extend f to the zero homomorphism R → I. Assume that J is nonzero.

Since R is a PID, we have I = rR for some nonzero r. Let x = f (r). Then since I is
divisible, there is some q ∈ I such that x = rq. Define f˜ : R → I by f˜(1) = q. Clearly
f˜ is an extension of f , so I is injective by Baer’s Criterion.

Corollary 5.10. The Z-module Q is injective.

Proof. Clearly Q is a divisible Z-module.

We will give a vast generalisation of Corollary 5.10, but it requires a bit more machinery.

Lemma 5.11. Let I be an injective right R-module and let I 0 be a direct summand of
I. Then I 0 is injective.

Proof. Write I = I 0 ⊕ M for some right R-module M . We will use Baer’s Criterion.

Let J be a right ideal of R, and let ϕ : J → I 0 be a module homomorphism. Then let

ϕ
ϕ̃ be the composition J → I 0 ,→ I. Since I is injective, Baer’s Criterion tells us that ϕ̃

39
 α̃ π
extends to a homomorphism α̃ : R → I. Let α be the composition R → I  I 0 , where
π is the projection onto I 0 along M .

Then for x ∈ J we have α(x) = π(α̃(x)) = π(ϕ̃(x)) = ϕ(x) by definition of ϕ̃, so

α : R → I 0 is an extension of ϕ, as required by Baer’s Criterion.

Lemma 5.12. Let R be a PID, and let p ∈ R be prime. Then the R-module M =
R[ p1 ]/R is injective.1

Proof. Note that R is a PID, hence a UFD. Let x + R ∈ M and r ∈ R \ {0}. Since R
is a UFD, we have x = pαk for k ≥ 0 and α ∈ R. We also have r = pl r0 for l ≥ 0 and
r0 ∈ R \ pR. Since r0 is coprime to pk+l , there is a q 0 ∈ R such that r0 q 0 ≡ α(mod pk )
q0
Let q = pk+l ∈ R[ p1 ]. Then

pl r 0 q 0 α r0 q 0 − α
rq − x = − = ,
pk+l pk pk

is in R since pk | r0 q 0 − α by definition of q 0 . Therefore r(q + R) = x + R ∈ M , so M

is divisible, hence injective.

Lemma 5.13. A Z-module is injective if and only if it is a direct sum of copies of Q

and Z[ p1 ]/Z for prime p.

Sketch Proof. Let I be an injective Z-module, and let Σ be the set of Z-submodules of
I that can be expressed as direct sums of Q and Z[ p1 ]/Z for prime p. For M, N ∈ Σ,
we say that M ≤ N if for any basis {ea : a ∈ A} of M , there is a basis {ea : a ∈ A0 } of
N extending it (that is, A0 ⊇ A).

Clearly (Σ, ≤) is a partially ordered set. Suppose that M1 ≤ M2 ≤ . . . is a chain in Σ.

Let {ea : a ∈ A1 } be any basis for M1 , and repeatedly extend it to bases {ea : a ∈ Ai }
for Mi . Then M = ∪i Mi is free with basis {ea : a ∈ ∪i Ai }, so it is in Σ. Since the basis
{ea : a ∈ A1 } for M1 was arbitrary, we have M1 ≤ M . Similarly, Mi ≤ M for all i.
Therefore, any chain in Σ has an upper bound, so by Zorn’s Lemma, Σ has a maximal
element, which we will call M .
1
I have tried to write the proof as concisely as possible, but this makes it seem a lot less intuitive
than it is. I would recommend trying to prove it yourself first. Maybe try the special case R = Z
before generalising to arbitrary PIDs.

40
Suppose that M 6= I. Then we have a short exact sequence

0 → M → I → I/M → 0.

Let I be an injective Z-module, and let Σ be the set of Z-submodules M of I equipped

with decompositions into direct sums of Q and modules of the form Z[ p1 ]/Z for prime
p. To be clear, an element of Σ is not just the module M , but the module M together
with the data of a specific direct sum decomposition.

We define a partial ordering on Σ by saying that M ≤ N if M ⊆ N and the inclusion

is compatible with the direct sum decompositions. Let M1 ≤ M2 ≤ . . . be a chain in
the partially ordered set (Σ, ≤). Then M = ∪i Mi is also in Σ, taking the union of the
direct sum decompositions.

Therefore, by Zorn’s Lemma, the partially ordered set (Σ, ≤) has a maximal element,
M . Since M is a direct sum of divisible modules, it is divisible, hence injective.
Therefore, the short exact sequence

0 → M → I → I/M → 0

splits, so
I∼
= M ⊕ I/M.

By maximality of M , the module I/M cannot have any submodules isomorphic to Q

or Z[ p1 ]/Z for any prime p. Let N = I/M , and suppose that N 6= 0. Let x ∈ N be
nonzero, and let C0 be the subgroup generated by x. Then C0 is a cyclic group. If it
is infinite, it is isomorphic to Z. If it is finite, then it has a subgroup of the form Z/p
for some prime p. Either way, we can take C ⊆ C0 to be a subgroup isomorphic either
to Z or to Z/p for prime p.

If C ∼= Z/p, define D = Z[ p1 ]/pZ ∼= Z[ p1 ]/Z. If C ∼

= Z, define D = Q. We have a
natural injection C ,→ D. Since N is a direct summand of I, it is injective, so by the
lifting property of injective modules,

N C,

41
we have a map ϕ : D → N extending C ,→ N . We claim that ϕ is injective. Let d ∈ D
have ϕ(d) = 0. Then there is some n ∈ N such that nd ∈ C. Since ϕ|C is injective and
ϕ(nd) = 0, we have nd = 0. But then d = 0, so ϕ is injective. So we have constructed
a submodule of N isomorphic to Q or Z[ p1 ]/Z. Either way, we have a contradiction, so
N = 0, and therefore I = M .

Lemma 5.14. For an algebraically closed field k, a k[x]-module is injective if and only
1
it is a direct sum of copies of k(x) and k[x][ (x−a) ]/k[x] for a ∈ k.

Proof. Similar to the previous Lemma. In fact, I’m pretty sure we can extend the result
to UFDs in general without much work, although I haven’t checked in detail.

5.3 Resolutions

Definition 5.15. Let M be an object of an abelian category A. A projective (resp.

free) resolution of M is a chain complex

. . . → P2 → P1 → P0 → 0,

together with a map ε : P0 → M , called the augmentation, such that the sequence

ε
. . . → P2 → P 1 → P0 → M → 0

is exact, and each Pi is projective (resp. free).

Remark 5.16. Since free modules are projective, a free resolution is a special type of
projective resolution.

Example 5.17. The Z-module Z/5 has free resolution

5
Z → Z,

since we have an exact sequence

5
Z → Z → Z/5 → 0.

The augmentation in this case is the quotient map Z → Z/5.

42
Definition 5.18. An injective resolution N → I • of an R-module N is a cochain
complex I • of injective R-modules that fits into an exact sequence

η
0 → N → I0 → I1 → . . . .

The map η is called the augementation.

Definition 5.19. Let A be an abelian category. We say that A has enough projec-
tives if for every M ∈ A, there is a projective object P and an epimorphism P  M .

Lemma 5.20. Let A be an abelian category with enough projectives. Then every object
of A has a projective resolution.

Proof. Let M ∈ A. Since A has enough projectives, we may take a projective object
∂0
P0 together with an epimorphism P0  M . Now we may construct our resolution
inductively as follows.

Suppose that we already have an exact sequence

k∂ ∂k−1
Pk → Pk−1 → Pk−2 → . . . → P0 → M → 0,

where the Pi are projective. Then there is a monomorphism ker ∂k  Pk , and since
A has enough projectives there is a surjection Pk+1  ker ∂k where Pk+1 is projective.
Then let ∂k+1 be the composition Pk+1  ker ∂k  Pk . Then im ∂k+1 = ker ∂k , so the
sequence
∂k+1 ∂k
Pk+1 → Pk → Pk−1 → . . . → P0 → M → 0,

is exact at Pk , and hence it is exact everywhere. By induction the sequence

. . . → P2 → P 1 → P0 → M → 0

is exact at every term, hence it is a projective resolution.

Definition 5.21. Let A be an abelian category. We say that A has enough injec-
tives if for every M ∈ A, there is an injective object I and a monomorphism M  I.

Lemma 5.22. If an abelian category A has enough injectives, then every object admits
an injective resolution.

43
Proof. Probably the “right” way to think about this is to note that an object I ∈ A
is injective if and only if I ∈ Aop is projective, and use the fact that left- and right-
resolutions are dual to each other. However, we will give a direct proof, similar to that
of Lemma 5.20.

Let M ∈ A. Then there is an injective object I 0 and a monomorphism ∂ −1 : M  I −1 ,

∂ −1
so the sequence 0 → M → I 0 is exact. Suppose that we have already constructed an
exact sequence
∂ −1 ∂ k−1
0 → M → I 0 → . . . → I k−1 → I k ,

where the I i are injective objects. Then we have an epimorphism I k  coker ∂ k ,

and (since A has enough injectives) a monomorphism coker ∂ k  I k+1 . Let ∂ k+1 be
the composition I k  coker ∂ k  I k+1 . Then ker ∂ k+1 = ker coker ∂ k = im ∂ k . We
conclude in the same manner as we did in Lemma 5.20.

Lemma 5.23. The category of left R-modules has enough projectives.

Proof. Since free modules are projective, it suffices to show that every R-module is the
image of some homomorphism from a free module.

To do this, let M be an R-module and let F be the free module with basis B = {vm :
m ∈ M }. By the universal property of free modules, there is a homomorphism ϕ :
F → M with ϕ(vm ) = m for each m ∈ M . Then is clearly a surjective homomorphism,
so we are done.

5.4 R-mod Has Enough Injectives

Proving that R-mod has enough injectives is hard enough to warrant a whole subsec-
tion to itself.
F
Definition 5.24 (Adjoint Functors). Let C D be categories and functors. We
G
say that the functors F and G are adjoint if for all x ∈ C and y ∈ D, there are
mutually inverse bijections bijections

Φx,y
HomC (x, G(y)) HomD (F (x), y)
Ψx,y

satisfying the so-called “naturality conditions”:

44
 f
1. For any map x → x0 in C, the following diagram commutes:

Φx0 ,y
HomC (x0 , G(y)) HomD (F (x0 ), y)
−◦f −◦F (f )
Φx,y
HomC (x, G(y)) HomD (F (x), y)

g
2. For any map y → y 0 in D, the following diagram commutes:

Φx,y
HomC (x, G(y)) HomD (F (x), y)
G(g)◦− g◦−
Φx,y0
HomC (x, G(y 0 )) HomD (F (x), y 0 )

In particular, we say that F is left adjoint to G and G is right adjoint to F , and

we call the relationship an adjunction. This situation is represented by the notation
F a G.
F
Lemma 5.25. Suppose that we have abelian categories and functors A B ,
G
where F a G and F is exact. Then G preserves injective objects.

Proof. Let I ∈ B be injective. Then we want to show that G(I) ∈ A is injective.

By Lemma 5.7, it suffices to show that the functor HomA (−, G(I)) : A → Ab sends
monomorphisms to epimorphisms. Recall that for a map ϕ : A → A0 in A, this functor
sends ϕ to the morphism

− ◦ ϕ : HomA (A0 , G(I)) → HomA (A, G(I)).

Let i : A → A0 be a monomorphism in A. Since the adjuncion F a G is natural in x,

we have a commutative diagram

ΦA0 ,I
HomC (A0 , G(I)) HomD (F (A0 ), I)
−◦i −◦F (i)
ΦA,I
HomC (A, G(I)) HomD (F (A), I)

Since F is exact, F (i) is a monomorphism, which means that −◦F (i) = HomB (F (i), I)

45
is an epimorphism. So the arrow on the right is a surjective function. Since the top
arrow is a bijection, the top path of the commutative diagram is a surjection, hence
the bottom path is too. Since the bottom arrow is a bijection, the left arrow must be
onto. Therefore − ◦ i is an epimorphism, as required.

Recall that an abelian group A is divisible if and only if for all a ∈ A and nonzero
n ∈ Z, there is some q ∈ A with a = nq.

Lemma 5.26. We have

L
1. If {Ai : i ∈ I} is a family of divisible abelian groups, then i∈I Ai is a divisible
abelian group.

2. If A is a divisible abelian group and K ⊆ A is any subgroup, then A/K is divisible.

Proof. Easy.

Lemma 5.27. The category of abelian groups has enough injectives.

Proof. Let A be an abelian group. By the proof of Lemma 5.23, we may express A as
a quotient of a free abelian group. In particular, we have
M 
A= Z /K
s∈S

L
where K is a submodule of s∈S Z. There is an obvious embedding of Z-modules
L L
s∈S Z ,→ s∈S Q, which induces an embedding

M  M 
A= Z /K ,→ Q /K =: I.
s∈S s∈S

Since Q is divisible, Lemma 5.26 tells us that I is divisible, hence injective, so we are
done.

The following theorem is the technical workhorse of this subsection. We will develop
some adjoint functor theory from scratch, making the proof longer than it needs to be
if you already know category theory. For a cleaner version, see [Link]
g/nlab/show/injective+module.

46
 F
Theorem 5.28. Let F a G be a pair of additive adjoint functors A B between
G
abelian categories A, B such that F is an exact, faithful functor. If B has enough
injectives, then so does A.

Proof. Let A ∈ A. Then F (A) ∈ B so there is an injective object I ∈ B together with

a monomorphism i : F (A)  I. Since i ∈ HomB (F (A), I), we have ĩ := ΨA,I (i) ∈
HomA (A, G(I)). Since F is exact, Lemma 5.25 tells us that G(I) ∈ A is injective.
Therefore it suffices to show that ĩ : A → G(I) is a monomorphism.

By definition of adjunctions, we have a map

ΦG(I),I
HomA (G(I), G(I)) HomB (F G(I), I)

and we define ε = ΦG(I),I (idG(I) ) : F G(I) → I. By definition of adjunctions, the

following diagram commutes:

ΦG(I),I
HomA (G(I), G(I)) HomB (F G(I), I)
−◦ĩ −◦F (ĩ)
ΦA,I
HomA (A, G(I)) HomB (F (A), I)

Therefore we have

ε ◦ F (ĩ) = ΦG(I),I (idG(I) ) ◦ F (ĩ)

= ΦA,I (idG(I) ◦ĩ)
= ΦA,I (ĩ)
= i,

where the final two equalities come respectively from commuativity of the diagram and
from the definition of ĩ. Clearly then ε ◦ F (ĩ) is a monomorphism, which implies that
F (ĩ) is a monomorphism (check this).

Now consider the exact sequence

ĩ
ker ĩ A G(I).

47
Since F is an exact functor, we have an exact sequence

F (ĩ)
F (ker ĩ) F (A) F G(I),

Since F (ĩ) is a monomorphism, exactness implies that the map F (ker ĩ) → F (A) is the
zero morphism, so by faithfulness of F the map ker ĩ → A is zero, which means that
ĩ : A → G(I) is a monomorphism, so we are done.

Definition 5.29. Let A be an abelian group and B be a left R-module. Then we give
HomAb (B, A) a natural right R-module structure via (f · r)(b) = f (rb).
F
Lemma 5.30. Define functors R-mod Ab by
G

1. F is the forgetful functor.

2. G(A) = HomAb (R, A)).

Then F a G.

Proof. Let M ∈ R-mod and A ∈ Ab. Then define

ΦM,A : HomR-mod (M, A) → HomAb (M, G(A))

by ΦM,A (f )(m) = f (mr) and

ΨM,A : HomAb (M, G(A)) → HomR-mod (M, A)

by ΨM,A (g)(m) = g(m)(1). It is an unpleasant exercise in abstract nonsense to check

that these are mutual inverses, and that they are natural in M and A.

Theorem 5.31. For any ring R, the category R-mod has enough injectives.

Proof. Let F, G be as in Lemma 5.30. It is clear that the forgetful functor F is exact
and faithful. We know that Ab has enough injectives, so since F a G, Lemma 5.30
gives the result.

48
6 Derived Functors

Suppose that F : A → B is a left exact functor between abelian categories, and that
0 → X → Y → Z → 0 is a short exact sequence in A. Then the sequence

0 → F (X) → F (Y ) → F (Z)

is exact at F (X) and F (Y ). It would be nice if we could find a long exact sequence
that extends this sequence. Derived functors give us a way of doing that.

6.1 δ-functors

Definition 6.1. Let A, B be abelian categories. A homological delta functor be-

tween A and B is a collection T = {Tn : A → B : n ≥ 0} of additive functors Tn such
that for any short exact sequence 0 → A → B → C → 0 in A there are morphisms

δn : Tn (C) → Tn−1 (A), n ∈ Z,

(where we write Tn = 0 for n < 0) satisfying

1. There is a long exact sequence

δn+1 n δ
. . . → Tn+1 (C) → Tn (A) → Tn (B) → Tn (C) → Tn−1 (A) → . . .

2. For each morphism of short exact sequences from 0 → A → B → C → 0 to

0 → A0 → B 0 → C 0 → 0, the following diagram commutes:

δn
Tn (C) Tn−1 (A)

0
δn
Tn (C 0 ) Tn−1 (A0 )

Definition 6.2. Let A, B be abelian categories. A cohomological delta functor

between A and B is a collection T = {T n : A → B : n ≥ 0} of additive functors
T n such that for any short exact sequence 0 → A → B → C → 0 in A there are

49
morphisms
δ n : T n (C) → T n+1 (A), n ∈ Z,

(where we write T n = 0 for n < 0) satisfying

1. There is a long exact sequence

δ n−1 δn
. . . → T n−1 (C) → T n (A) → T n (B) → T n (C) → T n+1 (A) → . . .

2. For each morphism of short exact sequences from 0 → A → B → C → 0 to

0 → A0 → B 0 → C 0 → 0, the following diagram commutes:

δn
T n (C) T n−1 (A)

(δ n )0
T n (C 0 ) T n−1 (A0 )

By the Snake Lemma, homology is a homological δ-functor from the category of chain
complexes Ch≥0 (A) to A, for any abelian category A. Similarly, cohomology gives a
cohomological δ-functor from Ch≥0 (A) to A.

Definition 6.3. A morphism S → T of homological (resp. cohomological) δ-functors

is a collection of natural transformatinos Sn → Tn (resp. S n → T n ) that commute with
δ.

Definition 6.4. A homological δ-functor T is universal if for any homological δ-

functor S and natural transformation f0 : S0 → T0 , there is a unique morphism of
δ-functors S → T exending f .

Definition 6.5. A cohomological δ-functor T is universal if for any cohomological

δ-functor S and natural transformation f 0 : T 0 → S 0 , there is a unique morphism
T → S of δ-functors extending f 0 .

6.2 Building Projective Resolutions

Our theory of left derived functors will depend heavily on projective resolutions. In
this subsection, we develop the tools we will need to manipulate projective resolutions
for our purposes.

50
Theorem 6.6. Let f 0 : M → N be a map of R-modules. Suppose that we have chain
complexes of the form


... P2 P1 P0 M 0,
f0
η
... Q2 Q1 Q0 N 0,

where all Pi are projective and the bottom row is exact.

Then there is a chain map f : P• → Q• extending f 0 , by which we mean that the

following diagram commutes:


... P2 P1 P0 M 0,
f2 f1 f0 f0
η
... Q2 Q1 Q0 N 0.

Furthermore, this chain map is unique up to homotopy equivalence.

Proof. (Existence.) We construct the fi by induction.

For the base case, since η : Q0 → M is surjective and Q0 is projective we may lift
f 0 ◦ ε : P0 → N to a map f0 : P0 → Q0 .

Assume that we already have maps f0 , f1 , . . . , fn that commute with the differentials
as in the diagram. Let P• , Q• have differentials ∂• , ∂e• respectively. By assumption, we
have the following commutative diagram:

∂n+1 ∂n
Pn+1 Pn Pn−1
fn fn−1
∂en+1 ∂en
Qn+1 Qn Qn−1

We claim that fn (im ∂n+1 ) ⊆ im ∂en+1 . To see this, let p ∈ im ∂n+1 . Clearly then
∂n p = 0, so by commutativity we have

∂en ◦ fn (p) = fn−1 ◦ ∂n (p) = fn−1 (0) = 0,

which means that f (p) ∈ ker ∂en = im ∂en+1 by exactness of Q• , so fn (im ∂n+1 ) ⊆ im ∂en+1

51
as we claimed. It follows that the composition fn ◦ ∂n+1 gives a well-defined map
Pn+1 → im ∂en+1 , so we have a diagram

Qn+1
∂en+1

fn ◦∂n+1
Pn+1 im ∂en+1

where the vertical arrow is a surjection. Since Pn+1 is a projective module, we may lift
to a map fn+1 : Pn+1 → Qn+1 such that the following diagram commutes:

Qn+1
fn+1
∂en+1
fn ◦∂n+1
Pn+1 Qn

Clearly we may redraw this diagram as:

∂n+1
Pn+1 Pn
fn+1 fn

Qn+1 Qn
∂en+1

which is exact what we need.

(Uniqueness.) Suppose first that f 0 = 0. Then we will inductively construct a

nullhomotopy s = {sn : Pn → Pn+1 }, that is fn = sn−1 ◦ ∂n + ∂en+1 ◦ sn for each n.
Let P0 = M, Q0 = N and Pi = Qi = 0 for i < 0. Also let f0 = f 0 : P0 → Q0 and
fi = 0 : Pi → Qi for i < 0.

For the base case, take si = 0 for i < 0. Then we have

f−1 = 0 = η ◦ s−1 + s−2 ◦ ∂−1 ,

so the claim is true for n = −1. Now assume that we have constructed si for i ≤ n − 1,

52
so that we have a diagram

∂n+1 ∂n ∂n−1
Pn+1 Pn Pn−1 Pn−2 ...

fn+1 fn fn−1 fn−2

sn−1 sn−2
∂en+1 ∂en ∂en−1
Qn+1 Qn Qn−1 Qn−2 ...

where the solid arrows commute and

fn−1 = ∂en ◦ sn−1 + sn−2 ◦ ∂n−1 .

Composing this last equation with ∂n gives

fn−1 ◦ ∂n = ∂en ◦ sn−1 ◦ ∂n + sn−2 ◦ ∂n−1 ◦ ∂n

= ∂en ◦ sn−1 ◦ ∂n ,

and by commutativity of the solid arrows we have ∂en ◦ fn = fn−1 ◦ ∂n , so

∂en ◦ fn = ∂en ◦ sn−1 ◦ ∂n ,

which means that the map (fn − sn−1 ◦ ∂n ) takes Pn into ker ∂en = im ∂en+1 , so we have
a diagram:
Pn
(fn −sn−1 ◦∂n )

∂en+1
Qn+1 im ∂en+1

By projectivity of Pn , this gives us a map sn : Pn → Qn+1 such that

∂en+1 ◦ sn = fn − sn−1 ◦ ∂n ,

as required.

Finally now consider the general case, where f 0 is not necessarily zero. Then suppose
that f = {fn } and g = {gn } are chain maps extending f 0 . Then f − g = {fn − gn } is
chain map extending f 0 − f 0 = 0, hence f − g ' 0 so f ' g.

Corollary 6.7 (Comparison Theorem). Let P• → M be a projective resolution of M

and let f : M → N be a map in R-mod. Then for any resolution Q• → N (not

53
necessarily projective!), there is a chain map P• → Q• lifting f , and this chain map is
unique up to homotopy.

Proof. Take P• to be exact in the theorem.

ε0 ε00
Theorem 6.8 (Horseshoe Lemma). Let P•0 → A0 and P•00 → A00 be projective resolutions
i π
in an abelian category A, and suppose that 0 → A0 → A → A00 → 0 is a short exact
sequence such that the following diagram commutes:

ε0
... P20 P10 P00 A0 0
iA

A
πA

ε00
... P200 P100 P000 A00 0

Set Pi = Pi0 ⊕ Pi00 for each i. Then there are maps Pi → Pi−1 such that P• → A is a
i π
projective resolution for A, and the short exact sequence 0 → A0 → A → A00 → 0 lifts
to a short exact sequence of chain complexes 0 → P•0 → P• → P•00 → 0.

Sketch Proof. We will prove the result in R mod , leaving the general abelian category
as an exercise in the Freyd-Mitchell Embedding Theorem. Consider the diagram

0 ε0
 P0 A0
1 i
 
0
P00 ⊕ P000 A
 
0 1 π

P000 ε00
A00 .

Since π is surjective and P000 is projective, there is a map α : P000 → A such that the

54
triangle
A
α
π

P000 ε00
A00

commutes. Define ε : P00 ⊕ P000 → A in matrix form by

 
ε = i ◦ ε0 α.

In other words,
ε(p0 , p00 ) = i ◦ ε0 (p0 ) + α(p00 ).

We claim that ε is surjective. Let a ∈ A. Then π(a) ∈ A00 , so π(a) = ε00 (p00 ) for some
p00 ∈ P000 . Then a − α(p00 ) ∈ ker π = im i, so

a = i(a0 ) + α(p00 )

for some a0 ∈ A0 . Since ε0 is surjective, we have a0 = ε(p0 ) for some p0 ∈ P00 . Then
a = ε(p0 , p00 ), so indeed ε is surjective. Now we have a diagram

0 ∂10
 P1 kerε0 
1 1
   
0 0
P10 ⊕ P100 ker ε
   
0 1 0 1

P100 ∂100
ker ε00 .

This is of the same form as the diagram we started with, so we get a surjection ∂1 :
P10 ⊕ P100 → ker ε. Continuing inductively, we get a projective resolution

. . . → P20 ⊕ P200 → P10 ⊕ P100 → P00 ⊕ P000 → A → 0.

It may be worth going through a few steps of this inductive procedure to convince
yourself that it does actually work.

55
6.3 Left Derived Functors

Definition 6.9. Let F : A → B be a a right exact functor between abelian categories,

and suppose that A has enough projectives. For an object A ∈ A, take a projective
resolution P• → A, and define the nth left derived functor

Ln F (A) = Hn (F (P• )).

Lemma 6.10. For any choice of projective resolution in Definition 6.9, we have
L0 F (A) ∼
= F (A).

Proof. By definition, L0 F (A) is the zeroth homology of the chain complex

F (∂2 ) F (∂1 )
. . . → F (P2 ) → F (P1 ) → F (P0 ) → 0,

so we have

L0 F (A) = H0 (F (P∗ ))
= coker F (∂1 ).

Now, the sequence

∂
P1 →1 P0 → coker ∂1 → 0

is exact, so by Lemma 4.4, the sequence

F (∂1 )
F (P1 ) → F (P0 ) → F (coker ∂1 ) → 0

is also exact. In particular, we have

coker F (∂1 ) ∼
= F (coker ∂1 ) ∼
= F (A),

and the result follows our earlier observation that L0 F (A) = coker F (∂1 ).

Lemma 6.11. If P• → A and Q• → A are projective resolutions of A ∈ A, then

Hn (F (P• )) ∼
= Hn (F (Q• ))

for all n. In other words, the left derived functor is well-defined.

56
Proof. We addressed the case n = 0 in Lemma 6.10. By the Comparison Theorem,
we may lift id : A → A to a chain map f : P• → Q• so that the following diagram
commutes:
... P2 P1 P0 A 0
f2 f1 f0 id

... Q2 Q1 Q0 A 0

Let f∗ : H• F (P• ) → H• F (Q• ) be the map induced by f∗ . Similarly there is a chain map
g : Q• → P• lifting id : A → A, and we have g∗ f∗ = (gf )∗ . Since gf and id : P• → P•
are both chain maps lifting id : A → A, by the Comparison Theorem they are chain
homotopic. The functor F takes this chain homotopy to a chain homotopy between
F (gf ) and F (idA ) = idF (P• ) so (gf )∗ = (id)∗ : H• F (P• ) → H• F (Q• ) is the identity on
homology.

Similarly f∗ g∗ is is the identity on H• F (Q• ), so f∗ , g∗ are isomorphisms.

Corollary 6.12. If A is projective, then Ln F (A) = 0 for n 6= 0.

Proof. Trivial.

Lemma 6.13. Let f : A → A0 be a map in an abelian category A. Then there is a

canonical map Ln F (f ) : Ln F (A) → Ln F (A0 ) for all n.

Proof. Take projective resolutions P• → A and P•0 → A0 . By the Comparison Theorem,

we may lift f to a chain map P• → P•0 , which induces a map Ln F (A) = Hn (P• ) →
Hn (P•0 ) = Ln F (A0 ). Furthermore, any two such chain maps P• → P•0 are chain ho-
motopic (by the Comparison Theorem), so they induce the same map on homology.
Therefore the map Ln F (A) → Ln F (A0 ) is natural.

Corollary 6.14. The nth left derived functor is a functor A → B.

Proof. Let A ∈ A have projective resolution P• → A. Then idP lifts idA , so Ln F (idA ) =
(idP )∗ = idLn F (A) .
f g
Let A → B → C be a diagram in A. Then we need to show that Ln F (g ◦ f ) =
Ln F (g) ◦ Ln F (f ). Let
P• → A, Q• → B, R• → C

57
be projective resolutions in A. Then the chain maps from the proof of Lemma 6.13
give a commutative diagram

P• Q• R•

A f
B g C,

and the result follows by naturality of Ln F on maps.

Corollary 6.15. The nth left derived functor is an additive functor A → B.

f
ε η
Proof. Let A B be a diagram in A. Let P• → A and Q• → B be projective
g
fe
resolutions. By the Comparison Theorem, lift to chain maps P• Q• such that
g
e
the following diagrams commute.

fe g
P• Q• P• Q•
e

ε η ε η

A f
B A g B

We also have a commutative diagram

fe+e
g
P• Q•
ε η

A f +g
B,

so
Ln F (f + g) = Hn (fe + ge) = Hn (fe) + Hn (e
g ) = Ln F (f ) + Ln F (g).

Lemma 6.16. Let F : A → B be a right-exact functor, and let A ∈ A. Suppose that

∂ ε
M1 →1 M0 → A → 0

58
is an exact sequence. Then

F (M0 )/ im F (∂1 ) ∼
= F (A).

Proof. This follows immediately from the fact that

F (M1 ) → F (M0 ) → F (A) → 0

is exact, by Lemma 4.4.

Corollary 6.17. If F is a right-exact functor, then L0 F = F .

Proposition 6.18. Let F : A → B be a right exact functor between abelian categories,

and let U : B → C be an exact functor between abelian categories. Then

U (Ln F ) = Ln (U F ).

Theorem 6.19. The collection {Li F }i forms a homological δ-functor.

Proof. We shall only prove that the long exact sequence exists. For functoriality, see
Weibel, Page 45, Theorem 2.4.6.

Let 0 → A0 → A → A00 → 0 be a short exact sequence in an abelian category A, and

let F : A → B be a right exact functor. Let P•0 → A0 and P•00 → A00 be projective
resolutions. Then by the Horseshoe Lemma, there is a projective resolution P• → A
fitting into the commutative diagram

0 P•0 P• P•00 0

0 A0 A A00 0,

where both rows are exact. Since the Pn00 are projective (or by construction in the proof
of the Horseshoe Lemma), the short exact sequence 0 → Pn0 → Pn → Pn00 → 0 is split
for each n. Since F is an additive functor it preserves split exact sequences, which
means that
0 → F (Pn0 ) → F (Pn ) → F (Pn00 ) → 0

59
is split1 exact for each n. Therefore we have a short exact sequence

0 → F (P•0 ) → F (P• ) → F (P•00 ) → 0

of chain complexes in B, and by the Snake Lemma this gives us a long exact sequence

δ δn−1
. . . → Ln F (A00 ) →
n
Ln−1 F (A0 ) → Ln−1 F (A) → Ln−1 F (A00 ) → Ln−2 F (A0 ) → . . . .

Theorem 6.20. If A has enough projectives, then the collection {Li F }i forms a uni-
versal homological δ-functor.

Proof. See Weibel, Page 47, Theorem 2.4.7.

6.4 F -acyclic Objects

Definition 6.21. Let F : A → B be a left-exact functor between abelian categories.

An object P ∈ A is F -acyclic if Li F (P ) = 0 for all i ≥ 1.

Lemma 6.22. Projective modules are F -acyclic for any F .

Proof. Compute the derived functor using the constant projective resolution . . . →
P → P → P → 0 with differential idP .

Lemma 6.23. Let 0 → M → P → A → 0 be an exact sequence, where P is F -acyclic.

Then

1. Li F (A) ∼
= Li−1 F (M ) for i ≥ 2.

2. L1 F (A) ∼
= ker((F (M ) → F (P )).

Proof. These both follow from the long exact sequence of Li F .

Theorem 6.24. Let P∗ → A be a resolution of A be F -acyclic objects. Then Li F (A) =

Hi F (P∗ ) for all i.
1
The point here is that F is not exact, so we need the splitting to say that it preserves the exactness
of this sequence.

60
Proof. Let P∗ → A have differentials ∂i : Pi → Pi−1 and augmentation ε : P0 → A. We
proceed by induction on i, with two base cases i = 0 and i = 1.

For i = 0, we know that L0 F (A) = F (A), and we have H0 F (P∗ ) ∼

= F (A) by Lemma
6.16. Let M = ker ε. Then 0 → M → P0 → A → 0 is a short exact sequence. Since F
is right-exact, it preserves cokernels, so

F (P1 )
F (M ) = F (coker ∂2 ) = coker F (∂2 ) = .
im F (∂2 )

Therefore,
 F (P ) 
1
L1 F (A) = ker → F (P0 )
im F (P2 )
= H1 F (P∗ ).

For the inductive step, let i ≥ 2. Then 0 → M → P0 → A → 0 is exact, so by Lemma

6.23, we have

Li F (A) ∼
= Li−1 F (M ) ∼
= Hi−1 (. . . → F (P2 ) → F (P1 ) → 0) ∼
= Hi F (P∗ ),

where the second isomorphism is by induction.

6.5 Flat Modules

Definition 6.25. A module F is flat if − ⊗R F is an exact functor.

Lemma 6.26. Let B be a left R-module. The following are equivalent:

1. B is flat.

2. TorR
n (A, B) = 0 for all n ≥ 1 and all left R-modules A.

3. TorR
1 (A, B) = 0 for all left R-modules A.

Proof. Suppose that B is flat. Let F∗ → A be a free resolution of A. Since − ⊗R B is

exact, the sequence

. . . → F2 ⊗R B → F1 ⊗R B → F0 ⊗R B → A ⊗R B → 0

61
is exact, so the homology of

. . . → F2 ⊗R B → F1 ⊗R B → F0 ⊗R B → 0

vanishes in positive degree. Therefore, we have (1) =⇒ (2).

The implication (2) =⇒ (3) is trivial. Finally, (3) =⇒ (1) follows from the long
exact sequence of Tor, since for any short exact sequence 0 → X → Y → Z → 0, we
have that
0 = TorR1 (Z, B) → X ⊗ B → Y ⊗ B → Z ⊗ B → 0

is exact.

Corollary 6.27. Every projective module is flat.

Proof. If P is projective, then 0 → P is a projective resolution, so TorR

n (A, P ) = 0 for
all A and all n ≥ 1.

Let I be a partially ordered set. We say that I is filtered if for all i, j in I, there exists
k ∈ I such that i, j < k. Recall that we may view a partially ordered set as a category,
where Hom(i, j) has precisely one morphism if i ≤ j, and is empty otherwise.

For categories C and D, we write Fun(C, D) for the category of functors from C to D.

Lemma 6.28. Let I be a filtered partially ordered set. Then the functor

colim : Fun(I, R-mod) → R-mod

is exact.

Proof. See Weibel, Page 57, Theorem 2.6.15.

Corollary 6.29. A filtered colimit of flat R-modules is flat.

Proof. The tensor-hom adjunction tells us that the functor − ⊗R M : R-mod →

R-mod is a left adjoint, so it preserves colimits. Let F : I → R-mod be a filtered
diagram of flat R-modules, and let

0→X→Y →Z→0

62
be a short exact sequence in R-mod. The F (i) are flat, so each sequence

0 → F (i) ⊗ X → F (i) ⊗ Y → F (i) ⊗ Z → 0

is exact. Since colimI is exact, we have a short exact sequence

0 → colimI (F (i) ⊗ X) → colimI (F (i) ⊗ Y ) → colimI (F (i) ⊗ Z) → 0.

Finally, the result follows from the fact that − ⊗R M commutes with colimits for all
R-modules M .

We note the two following facts. I’m not sure if they’re examinable, and I won’t give
proofs, but I’m stating them to be on the safe side.

Fact 6.30. A Z-module is flat if and only if it is torsion-free.

Fact 6.31. Let R be a ring and S ⊂ Z(R) be a central multiplicative set. Then S −1 R
is a flat R-module.

Corollary 6.32. The Z-module Q is flat.

We already know that Tor can be computed using projective resolutions. We also
know that all projective resolutions are flat. In fact, it turns out that we can relax the
projective assumption altogether and computer Tor using any flat resolution. This is
the content of the following lemma.

Theorem 6.33 (Flat Resolution Lemma). Let A be an R-module and let F• → A be

a flat resolution of A. Then for any R-module B, and all n, we have

TorR ∼
n (A, B) = Hn (F∗ ⊗R B).

Proof. This is immediate from Lemma 6.24, since the flat modules Fi are − ⊗R B-
acyclic.

6.6 Right Derived Functors

Let F : A → B be a left exact functor between abelian categories, and assume that A
has enough injectives. For an element A ∈ A, let A → I • be a projective resolution.

63
Definition 6.34. The nth right derived functor of F is

Rn F (A) = H n (F (I • )).

Since an injective object of A is precisely a projective object of Aop , the injective

resolution A → I • is a projective resolution in Aop . Therefore1 , right derived functors
are just left derived functors in the opposite category, and we get lots of results about
them for free, by duality. Most significantly, we get that right derived functors are
universal cohomological δ-functors. In particular, we have the following result.

Lemma 6.35. Let F : A → B be left exact, and let

0→A→B→C→0

be a short exact sequence in A. Then we get a long exact sequence

0 → F (A) → F (B) → F (C) → R1 F (A) → R1 F (B) → R1 F (C) → R2 F (A) → . . . .

1
As is often the case with duality, we are sweeping a lot of technicalities under the rug. It is
generally fine to do this, as long as we know what needs to be checked.

64
7 Defining Ext and Tor

The two most famous derived functors are Ext and Tor. Like most homological con-
structions, these find uses throughout pure mathematics. One of the most obvious (to
me) applications is the Universal Coefficient Theorem from algebraic topology, which
allows us to compute the homology and cohomology of topological spaces with all sorts
of funky coefficients, using only the homology with Z-coefficients.

7.1 Tor

We know that − ⊗R B is a right exact functor R-mod → R-mod.

Definition 7.1. For any R-modules A and B, the modules TorR

∗ (A, B) are defined to
be the values of the left derived functor

TorR
∗ (A, B) = L∗ (− ⊗R B)(A).

Example 7.2. We will compute

TorZ∗ (Z/6, Z/9).

We want the derived functor of − ⊗Z Z/9, evaluated at Z/6. Therefore we take a

projective resolution of Z/6:

6
0 → Z → Z → Z/6 → 0.

The modules TorZ∗ (Z/6, Z/9) are then the homology of

6⊗Z/9
0 → Z ⊗Z Z/9 → Z ⊗Z Z/9 → 0.

By the natural isomorphism R ⊗R M ∼

= M , this chain complex is isomorphic to

6
0 → Z/9 → Z/9 → 0.

65
Let f : Z/9 → Z/9 be the multiplication-by-6 map. Then

coker f if n = 0,



TorZn (Z/6, Z/9) = ker f if n = 1,


0

else.

Finally, we have
ker f = {[n] ∈ Z/9 : 9 | 6n} = 3Z/9Z ∼
= Z/3,

and
Z
6Z + 9Z
∼
coker f = / = Z/3.
9Z 9Z
Therefore, we have

Z/3 if n = 0, 1,
TorZn (Z/6, Z/9) ∼
=
0 else.

Example 7.3. Similarly, if R is a PID and a, b ∈ R, then


R/d if n = 0, 1,
R
Torn (R/a, R/b) =∼
0 else,

where d = gcd(a, b). This is basically exactly the same proof as the previous example.

We also know that, for any R-module A, the functor HomR (A, −) is left exact. There-
fore, it has right derived functors.

7.2 Ext

Definition 7.4. For R-module A and B, we define the modules Ext∗R (A, B) to be the
values of the right derived functors

Ext∗R (A, B) = R∗ (HomR (A, −))(B).

66
Example 7.5. We will compute

Ext∗Z (Z/2, Z/4).

Since we are evaluating a right derived functor at Z/4, we need to take an injective
resolution. We use the injective resolution

17→[1/4] 4
Z/4 → Q/Z → Q/Z → 0.

Therefore, Ext∗Z (Z/2, Z/4) is the cohomology of the cochain complex

4◦−
HomZ (Z/2, Q/Z) → HomZ (Z/2, Q/Z),

which is isomorphic to
0
Z/2 → Z/2.

Therefore, 
Z/2 if n = 0, 1,
n ∼
ExtZ (Z/2, Z/4) =
0 else.

67
8 Balancing Ext and Tor

We defined TorR R
∗ (−, B) as the left derived functor of −⊗R B. In other words, Tor∗ (A, B)
is the homology of P∗ ⊗R B, where P∗ → A is a projective resolution. It turns out that
we can also compute TorR ∗ (A, B) by taking the homology of A ⊗R Q∗ , where Q∗ → B
is a projective resolution. The proof of this fact is colloquially known as “balancing
Tor”.

Similarly, we defined Ext∗R (A, B) as the cohomology of Hom(A, I ∗ ), where B → I ∗

is an injective resolution. It turns out that we can also compute Ext by taking the
cohomology of Hom(P∗ , B), where P∗ → A is a projective resolution.

To prove these facts, we need some more machinery, which we will develop in this
section.

8.1 Mapping Cones

Definition 8.1. Let f : B• → C• be a chain map. We define the mapping cone of f

to be the chain complex cone(f ) with degree n part
 
cone(f ) n = Bn−1 ⊕ Cn ,

and differential
∂(bn−1 , cn ) = (−∂B (bn−1 ), ∂C (cn ) − f (bn−1 )).

The definition of the differential on cone(f ) is very unpleasant. It is easier to think of

it as being represented by a matrix
!
−∂B 0
,
−f ∂C

acting on column vectors of the form (bn−1 , cn )T . Yet another way of thinking about

68
the differential is via the following diagram:

−∂
Bn−1 Bn−2
−f

∂
Cn Cn−1 .

For the sake of completeness, we also define the dual notion.

Definition 8.2. If f : B • → C • is a cochain map, then the mapping cone of f is the
cochain complex cone(f ) with degree n part

cone(f ) n = B n+1 ⊕ C n ,
 

and whose differential is

∂(bn+1 , cn ) = (−∂B (bn+1 ), ∂C (cn ) − f (bn+1 )).

Lemma 8.3. Let f : B → C be a chain map. There is a long exact sequence of

homology

π f∗ i
∗
. . . → Hn+1 (cone(f )) →∗ Hn (B) → Hn (C) → Hn (cone(f )) → . . . ,

where f∗ = H• (f ) is the map on homology induced by f .

Proof. We define a short exact sequence of chain complexes

i π
0 → C → cone(f ) → B[−1] → 0

by i(c) = (0, c) and π(b, c) = −b. It is easy to check that these are chain maps, and
that the sequence is exact.

Note that Hn+1 (B[−1]) = Hn (B). By the Snake Lemma, this induces a long exact
sequence of homology

π δ i
. . . → Hn+1 (cone(f )) →∗ Hn (B) → Hn (C) →
∗
Hn (cone(f )) → . . . ,

for a connecting map δ : Hn (B) → Hn (C).

(Note: The following paragraph is hard to understand. Probably the best way to under-

69
stand it is to work out for yourself what δ(b̄) is using the proof of the Snake Lemma.)

In particular, for a cycle b ∈ Bn , we have that δ(b̄) = c̄ for any cycle c ∈ C with
i(c) = ∂(α, β), where (α, β) is an element of cone(f ) such that π(α, β) = b.

By definition of π, it suffices to take (α, β) = (−b, 0). Then we have i(c) = ∂(−b, 0) =
(∂b, f (b)) = (0, f (b)), since b is a cycle. Therefore we have c = f (b) and hence δ(b̄) =
f (b) = f∗ (b̄). So δ = f∗ and the result follows.

Corollary 8.4. A chain map f : B → C is a quasi-isomorphism if and only if the

mapping cone cone(f ) is acyclic.

Proof. Suppose that f is a quasi-isomorphism. Then we have an exact sequence

∼
= ∗ i π ∼
=
Hn (B) → Hn (C) → Hn (cone(f )) →∗ Hn−1 (B) → Hn−1 (C).

By exactness at Hn (C) and Hn−1 (B) we have that ker(i∗ ) = Hn (C) and im(π∗ ) = 0.

So i∗ kills everything, which means that im(i∗ ) = 0, and π∗ kills everything, so ker(π∗ ) =
Hn (cone(f )). By exactness we have

0 = im(i∗ ) = ker(π∗ ) = Hn (cone(f )),

so cone(f ) is acyclic. The converse is very straightforward; if cone(f ) is acyclic then

we have an exact sequence

f∗
0 → Hn (B) → Hn (cone(f )) → 0.

8.2 Double Complexes

We now define something called a double complex. Annoyingly, there are two very
similar but different conventions. We will follow Weibel, but see André Henriques’s
notes for the other convention.

Definition 8.5. A double complex C•• is a set of abelian groups Cp,q indexed by
(p, q) ∈ Z2 together with maps dh : Cp,q → Cp−1,q and dv : Cp,q → Cp,q−1 satisfying

70
d2h = d2v = 0 and dh dv = −dv dh .

In words, a double complex is an infinite two-dimensional grid of abelian groups where

each row (respectively each column) is a chain complex, and the horizontal and vertical
differentials anticommute. The following diagram is a picture of a double complex.

.. .. ..
. . .

... C−1,1 C01 C11 ...

... C−1,0 C00 C10 ...

... C−1,−1 C0,−1 C1,−1 ...

.. .. ..
. . .

8.3 Total Complexes

The total complex is a way of turning a double complex into a chain complex by taking
“diagonal slices”.

Definition 8.6. The total chain complex Tot(C•• ) defined by C•• is the chain
complex with h i M
Tot(C•• ) = Cp,q ,
n
p+q=n

together with differential

h i h i
Tot
d : Tot(C•• ) → Tot(C•• ) , dTot = dh + dv .
n n−1

The total complex is illustrated by the colours in the following diagram; each “diagonal
slice” is given a different colour. For example, Tot(C•• )0 is the direct sum of all the

71
red terms.
.. .. ..
. . .

... C−1,1 C01 C11 ...

... C−1,0 C00 C10 ...

... C−1,−1 C0,−1 C1,−1 ...

.. .. ..
. . .

There is also a version of the total chain complex where everything is the same, but
Q
we take the direct product instead of the direct sum. This is denoted by Tot (C•• ).
To avoid confusion, we will often write Tot⊕ (C•• ) for Tot(C•• ).

Let C•• be a double complex. We say that C•• is an upper half plane complex if
there is some q0 such that Cpq = 0 whenever q < q0 . Similarly C•• is a right half
plane complex if there is some p0 such that Cpq = 0 whenever p < p0 .

The following lemma tell us that in certain situations, the total complex is not only a
chain complex, but actually an exact sequence in R-mod. We defer to Weibel for the
proof.

Lemma 8.7 (Acyclic Assembly Lemma). Let C•• be a double complex in R-mod.
Consider the four following situations.

1. C•• is an upper half plane complex with exact columns.

2. C•• is a right half plane complex with exact rows.

3. C•• is an upper half plane complex with exact rows.

4. C•• is a right half plane complex with exact columns.

Q
If (1) or (2) holds, then Tot (C•• ) is exact. If (3) or (4) holds, then Tot⊕ (C•• ) is
exact.

72
Proof. See Weibel, Page 59, Lemma 2.7.3.

We also note part of Henriques’s formulation. The proof is similar.

Lemma 8.8 (Henriques’s AAL). Let C•• be a double complex such that for every n,
there exist only finitely many pairs (p, q) such that p + q = n and Cpq 6= 0. If C•• has
exact rows, then Tot⊕ (C•• ) is exact.

8.4 Non-Canonical Orderings on Double Complexes

As if things weren’t confusing enough, we will sometimes write double complexes where
the arrows do not point down and to the left. You might see the term “double cochain
complex” bandied about, but we will not bother with this. Instead, we call everything
a double complex, regardless of where the arrows point.

When the arrows do point down and to the left, we will say that the double complex
is canonically ordered. Non-canonical orderings get a bit dangerous when we want
to apply the Acyclic Assembly Lemma. For instance, the double complex

.. ..
. .

C01 C11 ... (1)

C00 C10 ...

is not an upper half plane complex, even though we have drawn it in the upper half
plane. This is because, in order to make it into a canonically ordered double complex,
we would have to reverse the vertical arrows.

Another important point1 is that in order to take the total complex, strictly speaking
you have to put your double complex into canonical order. However, you can usually
skip this step and take diagonal slices of the non-canonically ordered double complexes
straight away. To do this, note that each arrow must point from one diagonal slice into
1
This paragraph is hard to understand. It might be best to leave it to one side until you see why
it’s relevant.

73
the next. For instance, the complex C•• illustrated in (1) must have diagonal slices in
the % direction. That is, we have
Q Y
Tot (C•• )n = Ci,i−n .
i

8.5 Balancing Tor

For the following definition, let M and N be right and left R-modules respectively,
and let P• and Q• be projective resolutions for M and N . Let ∂ (P ) and ∂ (Q) be the
differentials for P• and Q• .

Definition 8.9. The double complex P• ⊗R Q• has components

(P• ⊗R Q• )p,q = Pp ⊗R Qq .

The differentials on P• ⊗R Q• are the maps

dh : Pp ⊗R Qq → Pp−1 ⊗R Qq , dv : Pp ⊗R Qq → Pp ⊗R Qq−1 ,

given by
dh = ∂ (P ) ⊗ 1, dv = (−1)p ⊗ ∂ (Q) .

The following lemma tells us that P• ⊗R Q• is a legitimate double complex.

Lemma 8.10. The differentials in P• ⊗R Q• anticommute.

Proof. We need to check that the following diagram anticommutes:

∂ (P ) ⊗Qq
Pp ⊗R Qq Pp−1 ⊗R Qq
(Q)
Pp ⊗∂q Pp−1 ⊗∂ Q

Pp ⊗R Qq−1 (P ) Pp−1 ⊗R Qq−1

∂ ⊗Qq−1

By definition of the fuctors Pm ⊗R − and − ⊗R Qn , the diagram anticommutes for

simple tensors in Pp ⊗R Qq . Since these simple tensors generate the entire module, the
diagram commutes in general.

74
Lemma 8.11. Let M, N, P• , Q• be as above. There are natural quasi-isomorphisms

P• ⊗R N ← Tot(P• ⊗R Q• ) → M ⊗R Q• .

Proof. View M and N chain complexes with only one nonzero component (of degree
0). Then we can form double complexes P• ⊗ Q• , P• ⊗ N and M ⊗ Q• in the obvious
way.

Let  : P• → M and η : Q• → N be the augmentation maps (that is they send elements

all the way down the exact sequence in the obvious way). Then we have maps

 ⊗ Q• : P• ⊗ Q• → M ⊗ Q• ,

and
P• ⊗ η : P• ⊗ Q• → P• ⊗ N.

These maps in turn induce chain maps

f : Tot(P• ⊗ Q• ) → Tot(M ⊗ Q• ) ∼
= M ⊗ Q• ,

and
g : Tot(P• ⊗ Q• ) → Tot(P• ⊗ N ) ∼
= P• ⊗ N.

We claim that f and g are quasi-isomorphisms.

Let C•• be the double complex obtained from P• ⊗ Q• by adding M ⊗ Q• in the column
p = −1. To clarify, we represent C•• by the following diagram, where the newly added
terms are shown in red.

.. .. ..
. . .

M ⊗ Q2 P0 ⊗ Q 2 P1 ⊗ Q 2 ...

M ⊗ Q1 P0 ⊗ Q 1 P1 ⊗ Q 1 ...

M ⊗ Q0 P0 ⊗ Q 0 P1 ⊗ Q 0 ...

75
One can check that

= (M ⊗ Qn+1 ) ⊕ Tot(P• ⊗ Q• ) n ∼
     
Tot(C•• ) n
= Tot(P• ⊗ Q• ) n ⊕ (M ⊗ Qn+1 ).

It follows that, at least as sets, we have

Tot(C•• )[−1] ∼
= cone(f ).

We will not verify that the isomorphism is a map of double complexes, but this is not
too hard to do.

Since each − ⊗ Qq is an exact functor (because the Qq are projective, hence flat),
the rows of C are exact. Since C is an upper half plane complex, the chain complex
Tot(C•• ) is acyclic by the Acyclic Assembly Lemma. Thus, cone(f ) is an acyclic chain
complex, so by Corollary 8.4, we have that f is a quasi-isomorphism.

The proof that g is a quasi-isomorphism is similar, so we omit it.

Corollary 8.12. We have an isomorphism

Hn (P• ⊗ N ) ∼
= Hn (M ⊗ Q• ),

for each n, so Tor can be computed by

TorR ∼
n (M, N ) = Ln (M ⊗ −)(N ) = Ln (− ⊗ N )(M ).

8.6 Balancing Ext

We define two more types of double complex. For illustrations, see the proofs of Lemma
8.5 and Lemma 8.16. Let C• and D• be chain complexes. Then we define the double
complex Hom(C• , D• ) to have terms

Hom(C• , D• )pq = Hom(Cp , Dq ),

and differentials are

(D)
dhpq ϕ = (−1)q × ϕ ◦ ∂p+1 , dvpq = ∂q(D) ◦ ϕ.

76
This is a non-canonical ordering, with dh pointing to the right, and dv pointing down.

If instead D• is a cochain complex, then we define Hom(C• , D• ) to have the same terms
(i.e. the (p, q) term is Hom(Cp , Dq )) and the same horizontal differentials, but with
vertical differentials
q
dvpq ϕ = ∂(D) ◦ ϕ.

This is again a non-canonical ordering, with differentials pointing up and to the right.

Theorem 8.13. Let M and N be R-modules (either both left or both right). Then we
have quasi-isomorphisms

HomR (P• , N ) → Tot(HomR (P• , I • )) ← HomR (M, I • ),

and hence ExtiR (M, N ) can be computed by any of

H i (HomR (P• , N )) ∼
= H i (Tot(HomR (P• , I • )) ∼
= H i (HomR (M, I • ).

Proof. The double complex HomR (P• , I • ) looks like this.

.. ..
. .

Hom(P0 , I 1 ) Hom(P1 , I 1 ) ...

Hom(P0 , I 0 ) Hom(P1 , I 0 ) ...

Let C•• be the double complex obtained by adding Hom(M, I 0 ) on the left.

.. .. ..
. . .

Hom(M, I 1 ) Hom(P0 , I 1 ) Hom(P1 , I 1 ) ...

Hom(M, I 0 ) Hom(P0 , I 0 ) Hom(P1 , I 0 ) ...

By the injectivity of the I n , the double complex C•• has exact columns. If we reversed
the arrows, we could turn C•• into a canonically ordered double complex satisfying the

77
hypotheses of Henriques’s Acyclic Assembly Lemma. Therefore Tot⊕ (C•• ) is acyclic.
The nth term of this total complex is

Tot⊕ (C•• )n = Hom(M, I n+1 ) ⊕ Tot⊕ Hom(P• , I • ).

Now, the bottom vertical arrows in the most recent diagram give us a chain map
f : Hom(M, I • ) → Tot⊕ (P• , I • ), which has mapping cone

cone(f )n = Hom(M, I n+1 ) ⊕ Tot⊕ Hom(P• , I • ).

Therefore cone(f ) is isomorphic to the total complex, hence it is acyclic, so f is a

quasi-isomorphism.

Similarly there is a map g : Hom(P• , N ) → Tot⊕ (P• , I • ). The corresponding double

complex has exact rows because the Pi are projective, and it is not too hard to show
that cone(g) is acyclic, so g is a quasi-isomorphism.

Corollary 8.14. The functor Ext can be computed by either of

ExtnR (M, N ) = Rn (Hom(−, N ))(M ) ∼

= Rn (Hom(M, −))(N ).

Since Hom(−, ) is contravariant, its right derived functor is computed using injective
resolutions (since they are projective in the opposite category). Therefore, Corollary
8.14 tells us that Ext∗R (A, B) is the cohomology of the cochain complex HomR (P∗ , B),
where P∗ → A is a projective resolution. The following (quite hard) example uses this
fact to compute Ext.

Example 8.15. Let a ≥ b ≥ c be integers. We will compute Ext∗Z/2a (Z/2b , Z/2c ). Let
R = Z/2a Z. Then Z/2b /Z has projective resolution

2a−b 2b 2a−b 2b 1
. . . → R → R → R → R → R → Z/2b Z.

Therefore Ext∗R (Z/2b , Z/2c ) is the cohomology of

2a−b 2b 2a−b 2b
. . . ← Z/2c ← Z/2c ← Z/2c ← Z/2c ← Z/2c .

Since b > c, the map 2b : Z/2c → Z/2c is zero, so Ext∗R (Z/2b , Z/2c ) is the cohomology

78
of
2a−b 0 2a−b 0
. . . ← Z/2c ← Z/2c ← Z/2c ← Z/2c ← Z/2c .

Therefore, we have

c
Z/2


 if i = 0,
ExtiR (Z/2b , Z/2c ) = ker(f ) if i ≥ 1 is odd,


coker(f ) if n ≥ 2 is even,


2a−b
where f is the map Z/2c → Z/2c . Suppose that a − b ≥ c. Then f = 0, so

ExtiR (Z/2b , Z/2c ) = Z/2c , for all i.

Suppose that a − b > c. Then

ker f = {n̄ : 2c | 2a−b n}

= {n̄ : 2b+c−a | n}
= 2b+c−a Z/2c Z
= Z/2a−b .

We also have

im f = 2a−b Z/2c ,

so
coker f = (Z/2c Z)/(2a−b Z/2c Z) ∼
= Z/2a−b Z.

Therefore, we have

Z/2c if i = 0,
ExtiR (Z/2b , Z/2c ) =
Z/2a−b Z else.

The following result gives us yet another way of computing Ext∗R (M, N ).

Lemma 8.16. The modules Ext∗R (M, N ) can be obtained by computing the cohomology
of the cochain complex:
Q
Tot (HomR (P• , Q• )).

79
Proof. We start by drawing a picture of the double complex HomR (P• , Q• ).

.. ..
. .

Hom(P0 , Q1 ) Hom(P1 , Q1 ) ...

Hom(P0 , Q0 ) Hom(P1 , Q0 ) ...

Let C•• be the double complex obtained by adding the cochain complex Hom(P• , N )
as follows.
.. ..
. .

Hom(P0 , Q1 ) Hom(P1 , Q1 ) ...

Hom(P0 , Q0 ) Hom(P1 , Q0 ) ...

Hom(P0 , N ) Hom(P1 , N ) ...

Since the Pi are projective, the functor Hom(Pi , −) is exact, so C•• has exact columns.
Q
Since C•• is upper half plane, the Acyclic Assembly Lemma tells us that Tot (C•• ) is
acyclic. Looking at the picture of C•• we see that
 Q   Q 
Tot (C•• ) n = Tot (Hom(P• , Q• ) n ⊕ Hom(Pn−1 , N ).

The vertical maps Hom(Pn , Q0 ) → Hom(Pn , N ) give a morphism of chain complexes

Q
f : Tot (Hom(P• , Q• )) → Hom(P• , N ),

which has mapping cone

 Q   Q 
cone(f )n = Tot (Hom(P• , Q• ) n ⊕ Hom(Pn−1 , N ) = Tot (C•• ) n .

One can check that the differentials of the mapping cone agree with those of the total
Q
complex, so cone(f ) = Tot (C•• ) is acyclic. Therefore f is a quasi-isomorphism and
the result follows.

80
Example 8.17. We will use Lemma 8.16 to compute Extk[x] (k, k) for a field k. Let
R = k[x] and take the projective resolution

x
0 R R k 0.

For 0 ≤ i, j ≤ 1 we have HomR (Pi , Pj ) = HomR (R, R) ∼

= R, and out double complex
is:
−x
R R
x x

R x R
Q
Then the cochain complex Tot (Hom(P• , Q• )) is

∗ = −1 ∗=0 ∗=1
(1,0)7→x
17→(x,−x) (0,1)7→x
0 R R⊕R R 0,

which has cohomology 

k if n = 0 or 1,
Hn =
0 else.

It is easy to check that this is the same as Ext∗R (k, k).

It turns out that the R-module

∞
M
Ext∗R (A, A) = ExtnR (A, A)
n=0

can be given the structure of a graded ring. We start by reinterpreting ExtnR (A, A)
the set of degree −n chain maps from P∗ to itself (where P∗ → A is a projective
resolution), modulo chain homotopy. We then use this characterisation to define a
product on Ext∗R (A, A) as composition of chain maps.

81
9 Ring Structure on Ext

So far, we have viewed Ext∗R (A, B) as a collection of R-modules. Taking the direct
sum, we obtain a single R-module
∞
M
Ext∗R (A, B) = ExtiR (A, B).
i=0

We will define a product Ext∗R (A, B) × Ext∗R (A, B) → Ext∗R (A, B), called the Yoneda
product, which gives Ext∗R (A, B) the structure of a graded ring.

9.1 Reinterpreting Ext

Let A and B be R-modules with projective resolutions P• and Q• respectively. Then

by Lemma 8.16 we have
Q
Ext∗R (A, B) = H ∗ (Tot (Hom(P• , Q• ))).
Q
Write T for the total complex Tot (Hom(P• , Q• )). Let ϕ ∈ T n . Then ϕ =
P
i ϕi for
ϕi ∈ Hom(Pi , Qi−n ). For each i, let ϕ
ei = εi−n ϕi , where

1 if i ≡ 0, 3 (mod 4),
εi =
−1 if i ≡ 1, 2 (mod 4).

e is an injective module homomorphism from T n to the

It is easy to see that ϕ 7→ ϕ
module of graded module homomorphisms P• → Q• [−n] (i.e. collections of module
homomorphisms Pi → Qi−n , without any assumptions about the chain complex struc-
ture).

Proposition 9.1. The map ϕ 7→ ϕ

e restricts to an isomorphism

Z nT ∼
= Ch∗ (R)(P• , Q• [−n]),

where the left hand side is the cocycles in degree n, and the right hand side denotes
chain maps P• → Q• [−n].

82
 e : P• → Q• [−n] is a chain
Proof. It suffices to show that ϕ is a cycle if and only if ϕ
map. We have
X
dϕ = (dh ϕi + dv ϕi )
i
X Q
= ((−1)i−n × ϕi ◦ ∂i+1
P
+ ∂i−n ◦ ϕi ).
i

e : P• → Q• [−n] is a chain map if and only if

The map ϕ

P Q
ei ◦ ∂i+1
ϕ = ∂i+1−n ◦ϕ
ei+1

for all i. By definition of ϕ,

e this is equivalent to

P Q
εi−n ϕi ◦ ∂i+1 = εi+1−n ∂i+1−n ◦ ϕi+1 ,

hence
P Q
ϕi ◦ ∂i+1 = (−1)i+1−n × ∂i+1−n ◦ ϕi+1 (2)

for all i.

The map ϕ ∈ T n is a cycle if and only if

dh ϕi + dv ϕi+1 = 0

for all i, which is equivalent to

Q
(−1)i−n ϕi ◦ ∂i+1
P
+ ∂i+1−n ◦ ϕi+1 = 0 (3)

e : P• → Q• [−n]
for all i. Conditions 2 and 3 are equivalent, so ϕ is a cycle if and only if ϕ
is a chain map.

Proposition 9.2. For ϕ ∈ Z n T , the cycle ϕ is a boundary if and only if ϕ

e : P• →
Q• [−n] is chain homotopic to zero.

Proof. Suppose that ϕ is a coboundary. Then ϕ = dψ for some ψ ∈ T n−1 . Then we

have
Q
ϕi = (−1)i−n × ψi−1 ◦ ∂iP + ×∂i+1−n ◦ ψi ,

83
so
Q
ei = (−1)i−n × εi−n ψei−1 ◦ ∂iP + εi−n+1 ∂i+1−n
i−n ϕ ◦ ψei .

Since ε2i = 1 for all i, this gives

Q
ei = (−1)i−n × ψei−1 ◦ ∂iP + εi−n εi−n+1 ◦ ∂i+1−n
ϕ ψei ,

and it is easily checked that εi−1 εi = (−1)i for all i, so

Q
ei = (−1)i−n × ψei−1 ◦ ∂iP + (−1)i−n+1 × ∂i+1−n
ϕ ◦ ψei

Set hi : Pi → Qi−n+1 as hi = (−1)i+1−n × ψei . Then we have

Q
ei = hi−1 ◦ ∂iP + ∂i+1−n
ϕ ◦ hi ,

hence h is a chain homotopy between ϕ

e and 0.

Suppose conversely that ϕ

e is chain homotopic to 0. Let h be a the chain homotopy, so
that
Q
ei = hi−1 ◦ ∂iP + ∂i+1−n
ϕ ◦ hi

for all i. Then define ψei = (−1)i+1−n hi : Pi → Qi−n+1 for each i and

ψi = εi−n+1 ψei ∈ T n−1 .

Since εi−1 εi = (−1)i for all i, we have

ψi = εi−n+1 εi−n εi−n+1 hi = εi−n hi .

Then

Q
(dψ)i = (−1)i−n ψi−1 ◦ ∂iP + ∂i−n+1 ◦ ψi
Q
= (−1)i−n εi−1−n hi−1 ◦ ∂iP + εi−n ∂i−n+1 ◦ hi
Q
= εi−n hi−1 ◦ ∂iP + εi−n ∂i−n+1 ◦ hi
= εi−n ϕ
ei
= ϕi .

Therefore ϕ = dψ is a coboundary.

84
Therefore ExtnR (A, B) is isomorphic to the quotient of the module of chain maps P• →
Q• [−n] by the submodule of nullhomotopic chain maps.

9.2 Yoneda Product

Let A, B be as above, with projective resolutions P• and Q• . Suppose that C is a third

R-module with projective resolution R• → C. Suppose that we have chain maps

ϕ : P• → Q• [−m], ψ : Q• → R• [−n].

Then ψ ◦ ϕ is a chain map P• → R• [−m − n]. Since homotopy commutes with compo-
sition of maps, if either of ϕ and ψ is nullhomotopic, then ψ ◦ ϕ is too. Therefore the
composition induces a well-defined map

^: Extm n m+n
R (A, B) × ExtR (B, C) → ExtR (A, C).

One can check that this map is R-bilinear, and in the case A = B = C it is an
associative binary operation

^: Ext∗R (A, A) × Ext∗R (A, A) → Ext∗R (A, A),

hence it gives Ext∗R (A, A) the structure of a graded ring.

Example 9.3. Let R = k[x]/(x2 ), and view the field k as the R-module R/xR. Then,
as a ring, we have
Ext•R (k, k) ∼
= k[y], |y| = 1.

Proof. We will use the projective resolution

x x x
... R R R 0

for k. Call this projective resolution P• .

A chain map P• → P• [−n] is a collection of module homomorphisms fi : R → R for

85
i ≥ n, such that the following diagram commutes.

x x x x
... R R R R ...
fn+2 fn+1 fn

... x R x R x R 0

The commutativity of the diagram is equivalent to fi (x) = xfi+1 (1) for all i ≥ n, which
is equivalent to (fi − fi+1 )(1) ∈ xR for all i ≥ n. Therefore the fi (1) are all congruent
modulo x.

The chain map f• is nullhomotopic if and only if there are module homomorphisms
hi : R → R for i ≥ n − 1 such that fi (1) = hi−1 (x) + xhi (1) for all i ≥ n, as illustrated
by the following diagram.

x x x x
... R R R R ...
fn+2 fn+1 fn
hn+2 hn+1 hn hn−1
... x R x R x R 0

In other words, f• is nullhomotopic if and only if fi (1) = x(hi−1 (1) + hi (1)) for all
i ≥ n. Clearly if f• is nullhomotopic, then fi (1) ∈ xR for all i ≥ n.

We claim that, conversely, if fi (1) ∈ xR for all i ≥ n, then f• is nullhomotopic. Suppose

that fi (1) ∈ xR for all i ≥ n. Then define hn−1 = 0, and let hn (1) ∈ R be an element
with fn (1) = xhn (1). Inductively, for i > n define hi (1) ∈ R to be an element such that
fi (1) − xhi−1 (1) = xhi (1). Then h is a nullhomotopy, so f• is nullhomotopic. Therefore
“chain maps modulo homotopy” is the same as “chain maps modulo x”.

We already showed that for any chain map f• : P• → P• [−n], all the fi (1) are congruent
modulo x. Modulo homotopy (i.e. modulo x), we may assume that for each i, we have
fi (1) = ai for some ai ∈ k. Since the fi (1) are all congruent modulo x (because f• is
(n) (n)
a chain map), the ai are all equal. Therefore ExtnR (k, k) = k · f• , where f• is the
degree −n chain map

x x x x
... R R R R ...
1 1 1

... x R x R x R 0.

86
 (m) (n) (m+n)
It is clear that f• ◦ f• = f• , so the Yoneda product gives

f•(m) ^ f•(n) = f•(m+n) .

(1)
Set y = f• . Then for each n, we have

ExtnR (k, k) = k · y n ,

so, as a ring, we have Ext•R (k, k) = k[y], where y is a degree −1 chain map, so |y| =
1.

87
10 Universal Coefficient Theorem

Theorem 10.1 (Künneth Formula). Let P• be a chain complex of flat R-modules such
that each ∂Pn is also flat. Then for all left R-modules M , and all integers n, there is
a natural short exact sequence

0 → Hn (P ) ⊗R M → Hn (P ⊗R M ) → TorR
1 (Hn−1 (P ), M ) → 0.

Proof. For each n, we have a short exact sequence

0 → Zn P → Pn → ∂Pn → 0.

For each M ∈ R-mod, the long exact sequence of TorR

∗ (−, M ) has terms

TorR R R
2 (∂Pn , M ) → Tor1 (Zn P, M ) → Tor1 (Pn , M ).

Since Pn and ∂Pn are flat, the left and right terms are zero, so TorR
1 (Zn P, M ) = 0.
Since M was arbitrary, Zn P is flat for all n.

Since ∂Pn is flat, we have TorR

1 (∂Pn , M ) = 0, so we have a short exact sequence

0 → Zn P ⊗R M → Pn ⊗R M → ∂Pn ⊗R M → 0,

hence we have a short exact sequence of chain complexes

0 → Z• P ⊗R M → P• ⊗R M → ∂P• ⊗R M → 0,

which gives us a long exact sequence

. . . → Hn+1 (∂P• ⊗R M ) →Hn (Z• P ⊗R M ) → Hn (P• ⊗R M ) →

Hn (∂P• ⊗R M ) → Hn−1 (Z• P ⊗R M ) → . . . .

The chain complexes ∂P• ⊗R M and Z• P ⊗R M both have zero differential. Therefore,
this long exact sequence gives an exact sequence

∂Pn+1 ⊗ M → Zn P ⊗ M → Hn (P• ⊗ M ) → ∂Pn ⊗R M → Zn−1 P ⊗R M (*)

88
The chain complex
∂Pn+1 → Zn P → Hn (P• ) → 0

is a projective resolution for Hn (P• ), so TorR

∗ (H∗ (P• ), M ) is the homology of the chain
complex
ϕn
0 → ∂Pn+1 ⊗R M → Zn P ⊗R M → 0,

where ϕn is the natural map. Now, (*) gives a short exact sequence

coker ϕn → Hn (P• ⊗R M ) → ker ϕn−1 .

This is exactly the result we are trying to prove.

Theorem 10.2 (Universal Coefficient Theorem). Let P• be a chain complex of free

Z-modules. Then for all M ∈ Z-mod, we have

Hn (P• ⊗Z M ) ∼
= (Hn (P• ) ⊗Z M ) ⊕ TorZ1 (Hn−1 (P• ), M ),

for all n. However, the decomposition is not natural.

Proof. Since each Pn is a free abelian group, so is the subgroup ∂Pn . Therefore ∂Pn is
projective, so the short exact sequence

0 → Zn P → Pn → ∂Pn → 0

splits, and hence Pn ∼

= Zn P ⊕ ∂Pn . Tensoring with M , we get

Pn ⊗ M ∼
= (Zn P ⊗ M ) ⊕ (∂Pn ⊗ M ),

so the inclusion
Zn P ⊗ M → Pn ⊗ M

has a retract. This retract restricts to a retract of

Zn P ⊗ M → ker(∂n ⊗ idM ),

89
We have a commutative diagram

Pn+1 ⊗ M Pn+1 ⊗ M

Zn P ⊗ M ker(∂n ⊗ idM ).

Modding out by the cokernels of the vertical arrows, we get that the natural map

Hn (P ) ⊗ M → Hn (P∗ ⊗ M ),

which has a retract. We know from Theorem 10.1 that there is a short exact sequence

0 → Hn (P ) ⊗R M → Hn (P ⊗R M ) → TorR
1 (Hn−1 (P ), M ) → 0,

which is therefore split.

90
11 Koszul Complexes

Let R be a ring and let x ∈ Z(R) be a central element. Then we define the Koszul
complex K(x) of x to be the chain complex

x
0 → R → R → 0,

concentrated in degrees 1 and 0. We denote the generator of K(x)1 by ex , so that

K(x)1 = R · ex , K(x)0 = R · 1, and dex = 1.

Suppose that x = (x1 , . . . , xn ) is a finite sequence of central elements in R. Then K(x)

is the chain complex
K(x1 ) ⊗R . . . ⊗R K(xn ).

For an R-module A, we define the Koszul homology and Koszul cohomology to

be

Hp (x, A) = Hp (K(x) ⊗R A),

H p (x, A) = H p (Hom(K(x), A)).

Lemma 11.1. For x ∈ Z(R), we have

H0 (x, A) = A/xA, H1 (x, A) = {a ∈ A : xa = 0}.

Proof. Easy.

Lemma 11.2. The module Kp (x) is free with generators

ei1 ∧ ei2 ∧ . . . ∧ eip , i1 < i2 < . . . < ip ,

and differentials
p
X
d(ei1 ∧ . . . ∧ eip ) = (−1)k+1 xik (ei1 ∧ . . . ∧ êik ∧ . . . ∧ eip ).
k=1

Corollary 11.3. We have

H0 (x, A) = A/(x1 , . . . , xn )A, H 0 (x, A) = Hom(R/(x1 , . . . , xn ), A).

91
We have a version of a the Künneth Theorem.

Theorem 11.4 (Künneth Formula). Let C• ∈ Ch∗ (R), and let x ∈ Z(R). Then there
is a short exact sequence

0 → H0 (x, Hq (C)) → Hq (K(x) ⊗R C) → H1 (x, Hq−1 (C)) → 0.

Proof. Consider the natural short exact sequence

0 → R → K(x) → R[−1] → 0.

Tensoring this short exact sequence with C yields a sequence

0 → C → K(x) ⊗R C → C[−1] → 0.

It is easy to check (just consider the chain maps degreewise) that this is a short exact
sequence. The Snake Lemma gives us an exact sequence

δ δ
Hn+1 (C[−1]) → Hn (C) → Hn (K(x) ⊗R C) → Hn (C[−1]) → Hn−1 (C).

Clearly this is isomorphic to

δ δ
Hn (C) → Hn (C) → Hn (K(x) ⊗R C) → Hn−1 (C) → Hn−1 (C).

We claim that δ : Hn (C) → Hn (C) is the multiplication by x map. To see this, note
that the diagram

0 Cn (K(x) ⊗R C)n C[−1]n 0

0 Cn−1 (K(x) ⊗R C)n−1 C[−1]n−1 0

92
is given explicitly by
 
1  
 
0 0 1
0 Cn Cn ⊕ C
n−1  Cn−1 0
∂ (−1)n x
∂   ∂
0 ∂
0 Cn−1   Cn−1 ⊕ Cn−2   Cn−2 0
1 0 1
 
0

Chasing this diagram, we see that that indeed δ = (−1)n x as we claimed. Now, it
follows that we have a short exact sequence

0 → xHn (C) → Hn (K(x) ⊗R C) → {a ∈ Hn−1 (C) : xa = 0} → 0.

By Lemma 11.1, this is the desired result.

Let A be an R-module. A nonzero element r ∈ R is called a zero divisor of A if there

is a nonzero element a ∈ A such that ra = 0. A regular sequence on A is a finite
sequence (x1 , . . . , xn ) in R such that the following two conditions hold.

1. The element x1 is not a zero-divisor of A.

2. For each i ≥ 2, the element xi is not a zero-divisor of A/(x1 , . . . , xi−1 ).

Proposition 11.5. Let x be a regular sequence on A ∈ R-mod. Then Hq (x, A) = 0

for q > 0.

Proof. We proceed by induction on n. For n = 1, the result is clear.

Let n ≥ 2. Let y = (x1 , . . . , xn−1 ) and x = (x1 , . . . , xn ). By the Künneth formula for
Koszul complexes, we have a short exact sequence

0 → H0 (xn , Hq (y, A)) → Hq (x, A) → H1 (xn , Hq−1 (y, A)) → 0.

For q ≥ 2, the first and last modules are zero, so Hq (x, A) = 0. The case q = 1 gives

H1 (x, A) ∼
= H1 (xn , H0 (y, A)),

93
and
H0 (y, A) = yA,

so H1 (xn , H0 (y, A)) is the first homology of

n x
A/yA → A/yA,

and this is zero since xn is not a zero-divisor of A/yA.

Proposition 11.6. If x is a regular sequence on R, then K(x) is a free resolution of

R/I, where I = (x1 , . . . xn ), so

TorR
p (R/I, A) = Hp (x, A), ExtpR (R/I, A) = H p (x, A).

94
12 Ext and Extensions

The module Ext1R (A, B) can be thought of as the set of equivalence classes of “exten-
sions of A by B”, which we will define in this section. In fact, even the abelian group
structure of Ext1R (A, B) can be expressed in terms of these extensions, using somethng
called the “Baer Sum”.

12.1 Extensions

Let A and B be R-modules. An extension of A by B is an exact sequence 0 → B →

X → A → 0. Extensions ξ and ξ 0 are equivalent if there is a commutative diagram

0 B X A 0
= ∼
= =

0 B X0 A 0.

(0,1)
An extension is split if it is equivalent to 0 → B → A ⊕ B → A → 0.

Lemma 12.1. In Z-mod, there are exactly p equivalence classes of extensions of Z/p
by Z/p.

Proof. Let
i π
0 → Z/p → X → Z/p → 0

be an extension of Z/p by Z/p. If X =∼ = Z/p ⊕ Z/p, then the extension is split, so

there is only one possible equivalence class.

Suppose that X ∼
= Z/p2 . Then we have

i π
0 → Z/p → Z/p2 → Z/p → 0.

The map π : Z/p2 → Z/p is surjective, so π([1]) is nonzero, which means that ker π =

95
pZ/p2 Z. Let π([1]) = [a] for a ∈ {1, . . . , p − 1}. We have a commutative diagram

π
Z/p2 Z/p
a−1

Z/p2 1
Z/p.

Therefore, up to equivalence of extensions, we may assume that a = 1. The equivalence

class is now determined by i([1]), which could be any of {[p], [p2 ], . . . , [(p − 1)p]}.

Therefore there are 1 + (p − 1) = p possibilities.

Lemma 12.2. If Ext1 (A, B) = 0, then every extension of A by B is split.

Proof. Applying Ext(A, −) to the extension ξ gives us a short exact sequence

0 Hom(A, B) Hom(A, X) Hom(A, A) 0,

s
so in partuclar there is some s ∈ Hom(A, X) such that A → X → A is the identity.
Therefore the extension is split.

In general, applying Ext(A, −) to an extension ξ gives an exact sequence

δ
0 → Hom(A, B) → Hom(A, X) → Hom(A, A) → Ext1 (A, B) → . . . .

Let Θ(ξ) = δ(idA ) ∈ Ext1 (A, B). Then we have a map

Θ : {extensions of A by B} → Ext1 (A, B).

Lemma 12.3. The map Θ gives a bijection between equivalence classes of extensions
of A by B and Ext1 (A, B).

Sketch Proof. We only sketch the proof, because the details are pretty unseemly.

First of all, we claim that Θ is surjective. Let x ∈ Ext1 (A, B). We will construct an
extension ξ of A by B such that Θ(ξ) = x. Pick some short exact sequence

j π
0 → B → I → N → 0,

96
where I is injective. This exists because R-mod has enough injectives, and we can
just take N to be the cokernel of some monomorphism to an injective object. Since I
is injective, we have Ext1 (A, I) = 0, so the long exact sequence of Ext gives an exact
sequence

δ
0 → Hom(A, B) → Hom(A, I) → Hom(A, M ) → Ext1 (A, B) → 0.

Since δ is surjective, there is some β ∈ Hom(A, B) such that δ(β) = x. Define the
R-module X to be the pullback

X A
β

I π N.

By the universal property of pullbacks, we get a unique map B → X such that the
diagram
B
0

j
X A
β

I π N
commutes. Therefore, we get a commutative diagram

0 B X A 0
β

0 B j
I π N 0.

It turns out that the top row of this diagram is exact. You can prove this using the
explicit construction of pullbacks and diagram chasing. By naturality of the long exact
sequence of Ext, we get a commutative diagram

Hom(A, A) Ext1 (A, B)

β◦−

Hom(A, N ) δ
Ext1 (A, B).

97
This diagram gives
idA Θ(ξ)

β δ(β) = x,

where ξ is the short exact sequence

0 → B → X → A → 0.

Therefore, Θ is onto. It turns out that the ξ we constructed is unique up to equivalence,

so we get a well-defined map

Ψ : Ext1 (A, B) → {equivalence classes of extensions}

with Φ ◦ Ψ = id. It turns out that this is a two-sided inverse for Φ, and the result
follows.

Let
1i π i π
ξ1 : 0 → B → X1 →1 A → 0, 2
ξ2 : 0 → B → X2 →2 A → 0

be extensions of A by B. Let

X 00 = X1 ×A X2 = {(x1 , x2 ) ∈ X1 × X2 : π1 (x1 ) = π2 (x2 )},

and let Y = X 00 /{(i1 (b), −i2 (b)) : b ∈ B}. We have maps

i : B → Y, b 7→ (i1 (b), 0)

and
π : Y → A, (x1 , x2 ) 7→ π1 (x1 ) + π2 (x2 ).

The sequence
i π
0→B→Y →A→0

is called the Baer sum of ξ and ξ 0 .

Lemma 12.4. The Baer sum is a well-defined extension of A by B.

Lemma 12.5. The set of equivalence classes of extensions of A by B is an abelian

group under the Baer sum, and the map Θ is an isomorphism of abelian groups.

98
12.2 Yoneda Ext Groups

Using extensions of A by B, we can define Ext1 (A, B) in any abelian category (i.e. no
need for projectives or injectives). We call this the Yoneda Ext group.

More generally, we define the Yoneda Extn (A, B) to be the equivalence classes of exact
sequences
ξ : 0 → B → Xn → . . . → X1 → A → 0,

under the equivalence relation generated by ξ ∼ ξ 0 if there is a diagram

0 B Xn ... X1 A 0
= =

0 B Xn0 ... X10 A 0.

Note that the arrows Xi → Xi0 do not have to be isomorphisms. At first glance, this
seems different to our definition of equivalence for extensions of A by B. However, by
the 5-lemma, this definition does actually generalise the previous one.

We again define a notion of a Baer sum. Let ξ and ξ 0 be representatives of elements of

Extn (A, B). Let X100 be the pullback of

X1

X10 A,

and let Xn00 be the pushout of

B Xn

Xn0 .

Let Yn be the quotient of Xn00 by the antidiagonal. Then the Baer sum is

0
0 → B → Yn → Xn−1 ⊕ Xn−1 → . . . → X2 ⊕ X20 → X100 → A → 0.

Suppose that A has enough projectives and P• → A is a projective resolution. Consider

99
the diagram

... Pn+1 Pn Pn−1 ... P0 A 0

=

0 B Xn ... X1 A 0.

By the comparison theorem, there is a chain map from the top row to the bottom row
(P )
lifting id : A → A. Setting M = ker ∂n gives a diagram

0 M Pn−1 ... P0 A 0
β =

0 B Xn ... X1 A 0

with exact rows.

Fact 12.6. There is a natural isomorphism between Yoneda Extn and the standard
Extn .

100
13 Group (Co)homology

We are often interested in studying mathematical objects by extracting algebraic in-

variants from them. One such invariant is the homology or cohomology of a group.
Group (co)homology is a rich application of the theory we have developed, drawing on
tools from throughout these notes.

13.1 Definitions

Definition 13.1. Let G be a group. A G-module is an abelian group A together

with a group homomorphism ρ : G → GLn (A).

As usual, we abbreviate ρ(g)(a) by g · a.

Definition 13.2. A morphism (A, ρ) → (B, σ) of G-modules is a Z-linear map ϕ :

A → B such that
ϕ(g · a) = g · ϕ(a)

for all g ∈ G and a ∈ A.

These morphisms make G-modules into a category, G-mod.

Lemma 13.3. There is an equivalence of categories

G-mod ' ZG-mod,

where ZG-mod is the group algebra over Z.

Definition 13.4. A G-module is trivial if g · a = a for all g ∈ G and a ∈ A.

Definition 13.5. Let A ∈ G-mod. Then the submodule of invariants if

AG = {a ∈ A : g · a = a∀g ∈ G},

and the module of coinvariants is

AG = A/hg · a − a : g ∈ G, a ∈ Ai.

101
Lemma 13.6. The assignments A 7→ AG and A 7→ AG are functorial. That is, we
have functors
−G , −G : G-mod → Ab.

There is a functor triv : Ab → G-mod taking A to the G-module on A with trivial

action.

Lemma 13.7. We have adjunctions −G a triv a −G .

Proof. This is basically because, for any G-module M and abelian group A, the Z-linear
homomorphisms M → AG are literally the same thing as G-linear homomorphisms
M → A, and G-linear homomorphisms triv(A) → M are precisely the abelian group
homomorphisms A → M with hga − a : g ∈ G, a ∈ Ai contained in their kernel.

Corollary 13.8. The functor −G is right-exact and the functor −G is left-exact.

Lemma 13.9. We have

AG = Z ⊗ZG A.

Proof. Let N = hga − a : g ∈ G, a ∈ Ai. Define ϕ

e : A → Z ⊗ZG A by

ϕ(a)
e = 1 ⊗ a.

Then ϕ
e kills N , so it descends to a Z-module homomorphism

ϕ : AG → Z ⊗ZG A.

There is a ZG-bilinear map B : Z × A → AG given by

(n, a) 7→ na + N.

This map induces a map ψ : Z ⊗ZG A → AG with

ψ(n ⊗ a) = na + N.

Clearly ϕ and ψ are mutual inverses.

Lemma 13.10. We have

AG = HomZG (Z, A).

102
Since −G and −G are right- and left-exact respectively, we can take their derived
functors.

Definition 13.11. The group homology of G with coefficients in A is

H∗ (G, A) = L∗ (−G )(A),

and the group cohomology of G with coefficients in A is

H ∗ (G, A) = R∗ (−G )(A).

Lemma 13.12. We have

H∗ (G, A) ∼
= TorZG
∗ (Z, A)

and
H ∗ (G, A) ∼
= Ext∗ZG (Z, A).

Example 13.13. Let G = hti be infinite cyclic and let A be a G-module. Then ZG is
the Laurent polynomial ring Z[t, t−1 ].
−1 ]
We start by computing the group homology H∗ (G, A). This is the same as Tork[t,t
∗ (Z, A).
−1
Write R = Z[t, t ]. Then the trivial R-module Z has a projective resolution

(t−1)
0 → R → R → Z → 0,

so the group homology is the homology of the chain complex

(t−1))⊗A
0 → R ⊗R A → R ⊗R A → 0,

which is the same as the homology of

(t−1)
0 → A → A → 0.

It follows that

G
{a ∈ A : ta = a} = A if n = 1,



Hn (G, A) = A/(t − 1)A = AG if n = 0,


0 else.


103
Similarly, we have
H ∗ (G, A) ∼
= Ext∗R (Z, R).

Using the same projective resolution, the group cohomology is the cohomology of the
cochain complex
(t−1)
0 ← HomR (R, A) ← HomR (R, A) ← 0,

which is equivalent to
(t−1)
0 ← A ← A ← 0.

Therefore H ∗ (G, A) = H1−∗ (G, A).

13.2 First Homology

Let ε : ZG → Z be the augmentation map g 7→ 1 for all g ∈ G. Let J = ker ε ⊆ ZG.

Then J is a free Z-module with basis {g − 1 : g ∈ G \ {1}}.

Lemma 13.14. We have

J/J 2 ∼
= G/[G, G].

Proof. Define the map θ : G → J/J 2 by θ(g) = g − 1. For a, b ∈ G, we have

θ(ab) = ab − 1 + J 2 = (a − 1) + (b − 1) + J 2 = θ(a) + θ(b),

so θ is a group homomorphism. Since J/J 2 is abelian, we have

θ(aba−1 b−1 ) = θ(a) + θ(b) − θ(a) − θ(b) = 0,

so [G, G] ⊆ Ker θ, so θ descends to a homomorphism

θ̄ : G/[G, G] → J/J 2 .

Define σ : J → G/[G, G] to be the unique homomorphism with g − 1 7→ g + [G, G].

For a, b ∈ G, we have

σ((a−1)(b−1)) = σ(ab−a−b+1) = σ(ab−1−(a−1)−(b−1)) = aba−1 b−1 +[G, G] = 0.

So σ descends to a homomorphism J/J 2 → G/[G, G]. Clearly θ̄ and σ̄ are mutual

104
inverses, so we are done.

Lemma 13.15. We have

JG ∼
= J/J 2 .

Proof. We have
JG ∼
= J ⊗ZG Z = J ⊗ZG ZG/J ∼
= J/J 2 .

Theorem 13.16. We have

H1 (G, Z) ∼
= G/[G, G].

Proof. We have a short exact sequence

0 → J → ZG → Z → 0

of G-modules. Since TorZG

∗ is a homological δ-functor, we obtain an exact sequence

H1 (G, ZG) → H1 (G, Z) → JG → (ZG)G → ZG → 0.

Since ZG is a projective ZG-module, we have H1 (G, ZG) = 0. The right-hand map is

the isomorphism
ZG ∼ = ZG/J ∼= Z,

so we have
H1 (G, Z) ∼
= JG = J/J 2 ∼
= G/[G, G].

13.3 The Norm Element

Let G be a finite group. The norm element of ZG is

X
N= g ∈ ZG.
g∈G

Clearly N ∈ (ZG)G .

105
Lemma 13.17. The subgroup

H 0 (G, ZG) = (ZG)G

is the two-sided ideal ZN of ZG generated by N .

Proof. Easy. It’s just Z · N .

13.4 Finite Cyclic Groups

Let Cm = hσ : σ m = 1i be the cyclic group of order m. Then the norm element is

1 + σ + . . . + σ m−1 .

We have
0 = σ m − 1 = (σ − 1)N.

Lemma 13.18. The chain complex

σ−1 N σ−1 ε
. . . → ZCm → ZCm → ZCm → ZCm → Z → 0

is a projective resolution for Z as a ZCm -module.

Proof. Exactness follows from a diagram chasing, using the fact that the sequences

N σ−1
0 → J → ZCm → ZN → 0, 0 → ZN → ZCm → J → 0

are exact.

106
Theorem 13.19. Let A be a Cm -module. Then

A/(σ − 1)A if n = 0,



Hn (Cm , A) = AG /N A if n = 1, 3, 5, . . .,


{a ∈ A : N a = 0}/(σ − 1)A if n = 2, 4, 6, . . ..


G
A if n = 0,



H n (Cm , A) = {a ∈ A : N a = 0}/(σ − 1)A if n = 1, 3, 5, . . .,


AG /N A if n = 2, 4, 6, . . ..


Corollary 13.20. We have


Z


 if n = 0,
Hn (Cm , Z) = Z/m if n ≥ 1 is odd,


0 else.


Z if n = 0,



H n (Cm , Z) = Z/m if n ≥ 2 is even,


0 else.


13.5 Free Groups

Theorem 13.21. Let G be a free group on a set X. Then J is a free ZG-module with
basis {x − 1 : x ∈ X}.

Proof. See, Page 169, Proposition 6.2.6.

Corollary 13.22. If G is a free group on X, then Z has free resolution

0 → J → ZG → Z → 0.

Therefore, Hn (G, A) = H n (G, A) = 0 for n 6= 0, 1, and H0 (G, Z) ∼

= H 0 (G, Z) ∼
= Z,
while
H1 (G, Z) ∼ Z, H 1 (G, Z) ∼
M Y
= = Z.
x∈X x∈X

107
13.6 Crossed Homomorphisms

Historically, the maps we are interested in are called “crossed homomorphisms”. In

these notes, however, we will adopt the more modern term “derivations”. Many of the
proofs from now on are omitted; we have opted just to define the objects and sketch
the theory. The proofs can all be found in Weibel.
Definition 13.23. Let G be a group and A be a left G-module. A derivation of G
in A is a map D : G → A with D(gh) = gD(h) + D(g) for all g, h ∈ G.
Remark 13.24. For those familiar with derivations more generally, Definition 13.23
is perhaps a little odd. Usually, you would expect

D(gh) = gD(h) + D(g)h.

The reason we drop the h on the far right is that we are viewing A as having trivial
G-action on the right. In other words, we have D(g)h = D(g).

Write Der(G, A) for the set of derivations. Then Der(G, A) is an abelian group under
pointwise addition. For a ∈ A, let Da : G → A be the map Da (g) = ga − a.
Definition 13.25. A derivation of the form Da is a principal derivation.

Write PDer(G, A) for the set of principal derivations. It is easy to see that Da + Db =
Da+b , so PDer(G, A) is a subgroup of Der(G, A). Recall that J is the augmentation
ideal of ZG. Let ϕ : J → A be a G-module homomorphism. Define Dϕ : G → A by

Dϕ (g) = ϕ(g − 1).

Then Dϕ : G → Ai is a G-module map.

Lemma 13.26. The map ϕ 7→ Dϕ is a natural isomorphism

HomG (J, A) → Der(G, A)

of abelian groups.
Theorem 13.27. We have

H 1 (G, A) = Der(G, A)/ PDer(G, A).

108
Proof. The short exact sequence

0 → J → ZG → Z → 0

of ZG-modules gives a long exact sequence beginning with

0 → Hom(Z, A) → Hom(ZG, A) → Hom(J, A) → H 1 (G; A) → 0,

and the natural isomorphism Hom(J, A) ∼ = Der(G, A) takes the image of Hom(Z, A) to
PDer(G, A), so
H 1 (G; A) ∼
= Der(G, A)/ PDer(G, A).

Corollary 13.28. Let A be a trivial G-module. Then

H 1 (G, A) = Der(G, A) ∼
= HomGrp (G, A).

Theorem 13.29 (Hilbert Theorem 90). Let L/K be a finite Galois extension with
Galois group G. Let L∗ be the unit group of L. Then L∗ is naturally a G-module, and

H 1 (G, L∗ ) = 0.

13.7 Bar Complex

Throughout this section, Z is a trivial G-module.

Definition 13.30. The unnormalised bar complex is the chain complex

ε
. . . → B2u → B1u → B0u → Z → 0

with B0u = ZG and Bnu = (ZG)⊗n . The differential d : Bnu → Bn−1

u
is given by
n
X
d= (−1)i di ,
i=0

109
where

d0 (g1 ⊗ . . . ⊗ gn ) = g1 · (g2 ⊗ . . . ⊗ gn ),
di (g1 ⊗ . . . ⊗ gn ) = g1 ⊗ . . . ⊗ gi gi+1 ⊗ . . . ⊗ gn for 1 ≤ i ≤ n − 1,
dn (g1 ⊗ . . . ⊗ gn ) = g1 ⊗ . . . ⊗ gn−1 .

Definition 13.31. The normalised bar complex is


. . . → B2 → B1 → B0 → Z → 0,

where B0 = ZG, and for n ≥ 1, the group Bn is free abelian on basis

{[g1 | . . . |gn ] : gi ∈ G \ {1}}.

Pn i
The differential d : Bn → Bn−1 is d = i=0 (−1) di , where

d0 (g1 | . . . |gn ) = g1 · (g2 | . . . |gn ),

di (g1 | . . . |gn ) = g1 | . . . |gi gi+1 | . . . |gn for 1 ≤ i ≤ n − 1,
dn (g1 | . . . |gn ) = g1 | . . . |gn−1 .

We write [] for 1 ∈ B0 = ZG. If any of the gi is 1, we write [. . . |gi | . . .] for 0 ∈ Bn .

Example 13.32. We have

d([g|h]) = g[h] − [gh] + [g]

and
d([f |g|h]) = f [g|h] − [f g|h] + [f |gh] − [f |g].

Theorem 13.33. The normalised and unnormalised bar complexes are free resolutions
of Z as a ZG-module.

Corollary 13.34. We have that H ∗ (G, A) is the cohomology of the chain complexes
HomG (B∗u , A) and HomG (B∗ , A).

This allows us to give an explicit description of group cohomology. Define an n-cochain

to be a function f : Gn → A. An n-cochain ϕ is normalised if ϕ(g1 , . . . , gn ) = 0

110
whenever one of the gi is 1. Then

HomG (Bnu , A) ∼
= {n-cochains}.

Define the differential d : {n-cochains} → {(n + 1)-cochains} by

n−1
X
dϕ(g0 , . . . , gn ) = g0 ϕ(g1 , . . . , gn ) + (−1)i ϕ(. . . , gi gi+1 , . . .) + (−1)n ϕ(g0 , . . . , gn−1 ).
i=1

If dϕ = 0, then ϕ is an n-cocycle, and for all ϕ, the cochain dϕ is an n-coboundary.

Write Z n (G, A) and B n (G, A) for the abelian groups of n-cocycles and n-coboundaries
respectively. Then
H n (G, A) = Z n (G, A)/B n (G, A).

Lemma 13.35. We have

H 1 (G, A) = Der(G, A)/ PDer(G, A).

13.8 Group Extensions

Let A be an abelian group and let G be a group. An extension of G by A is a short

exact sequence
π
0 → A → E → G → 1.

The extension splits if π has a section. That is, if there is a group homomorphism
s : G → E such that π ◦ s = id G. Extensions

π
0 → A → Ei → G → 1

for i = 1, 2 are equivalent if there is a group isomorphism E1 → E2 such that the

obvious diagram commutes.

Theorem 13.36. There is a natural bijection

H 2 (G, A) ⇐⇒ {Equivalence classes of extensions of G by A}.

111