MA3403 Algebraic Topology

Lecturer: Gereon Quick

Lecture 19

19. Ext and the Universal Coefficient Theorem for cohomology

In the previous lecture, we introuced the singular cochain complex and defined
singular chomology. Along the way we ran into some exact sequences to which
applied the Hom-functor. In particular, we constructed the Kronecker map
κ : H n (X; M ) → Hom(Hn (X),M ).

Our goal for this lecture is to study the Hom-functor in more detail and to
prove the Universal Coefficient Theorem for singular cohomology which will
tell us that κ is surjective. However, κ is not injective in general, but the UCT
will tell us what the kernel is.
Again, for some this will be a review of known results in homological algebra.
Nevertheless, those who have not seen this before, should get a chance to catch
up.
We will again focus on the main ideas.

Let M be an abelian group. We would like to understand the effect of the

functor Hom(−,M ) on exact sequences.
Before we start, note that Hom is not symmetric in general, i.e., Hom(A,M )
and Hom(M,A) might be very different indeed. For example,
Hom(Z,Z/n) ∼
= Z/n, but Hom(Z/n,Z) = 0.

Our next observation tells us that Hom is left-exact:

Lemma: Hom is left-exact

(a) Let M be an abelian group. Suppose we have an exact sequence
i j
A→
− B→
− C → 0.
Then applying Hom(−,M ) yields an exact sequence
j∗ i∗
0 → Hom(C,M ) −
→ Hom(B,M ) −
→ Hom(A,M ).

1
2

In other words, the functor Hom(−,M ) is left-exact and sends cokernels

to kernels.
(b) Similarly, applying Hom(M,−) to an exact sequence of the form
i j
0→A→
− B→
− C
yields an exact sequence
j∗ ∗ i
→ Hom(M,B) −
0 → Hom(M,A) − → Hom(M,A).
In other words, the functor Hom(M,−) is left-exact and sends kernels to
kernels.

Proof: (a) To show that j ∗ is injective, assume that γ ∈ Hom(C,M ) satisfies

∗
j (γ) = 0. That means
j ∗ (γ)(b) = (γ ◦ j)(b) = γ(j(b)) = 0 for all b ∈ B.
But j is surjective, and hence every element in C is of the form j(b) for some
b ∈ B. Hence γ = 0 is the trivial homomorphism.
The composition i∗ ◦ j ∗ is clearly 0, since j ◦ i = 0 by assumption. Thus
Im (j ∗ ) ⊆ Ker (i∗ ).
Now if β ∈ Hom(B,M ) is in Ker (i∗ ), then
0 = i∗ (β)(a) = β(i(a)) for all a ∈ A.
In other words, β is trivial on the image of i and hence factors as
β : B → B/Im (i) → M.
But B/Im (i) ∼= C, since the intial sequence was exact. Hence β is the composition
j γ
of a map B →
− C→ − M for some γ ∈ Hom(C,M ). Thus, β ∈ Im (j ∗ ).
(b) The proof is of course similar. To show that i∗ is injective, let α ∈
Hom(M,A) be a map such that i∗ (α) = 0. That means
i∗ (α(m)) = i(α(m)) = 0 for all m ∈ M.
Since i is injective, this implies α(m) = 0 for all m ∈ M , and hence α = 0.
The composition j∗ ◦ i∗ is clearly 0, since j ◦ i = 0 by assumption. Thus
Im (i∗ ) ⊆ Ker (j∗ ).
If β ∈ Hom(M,B) is in Ker (j∗ ), then
0 = j∗ (β)(m) = j(β(m)) for all m ∈ M.
 3

In other words, β(m) ∈ Ker (j) for all m ∈ M . Since Ker (j) = Im (i), we get
β(m) ∈ Im (i) for all m ∈ M . Hence β factors as
α i
β: M −
→A→
− B
for some α ∈ Hom(M,A). Thus, β ∈ Im (i∗ ). QED
However, suppose we have an injective homomorphism
A ,→ B.
Then it is in general not the case that the induced map
Hom(B,M ) → Hom(A,M )
is surjective.
2
For example, take the map Z →− Z given by multiplication by 2. It is clearly
injective. But if we apply Hom(−,Z/2), we get the map
2=0
Hom(Z,Z/2) ∼
= Z/2 −−→ Z/2 ∼
= Hom(Z,Z/2)
which is not surjective.
We would like to remedy this defect. And we can already guess how this can
be achieved. As we have seen in the previous lecture, Hom(−,M ) is not so far
from being exact. For, if we apply Hom(−,M ) to a short exact sequence of
free abelian groups, then the induced sequence is still short exact.
So let A be an abelian group and let us choose a free resolution of A as in a
previous lecture
0 → F1 ,→ F0  A.

Applying Hom(−,M ) to this equence yields an exact sequence

0 → Hom(A,M ) → Hom(F0 ,M ) → Hom(F1 ,M ).

The right-hand map is not necessarily surjective, or in other words, the cok-
ernel of the right-hand map is not necessarily zero.
This leads to the following important definition:

Definition: Ext
The cokernel of the map Hom(F0 ,M ) → Hom(F1 ,M ) is called Ext(A,M ).
Hence by definition we have an exact sequence
0 → Hom(A,M ) → Hom(F0 ,M ) → Hom(F1 ,M ) → Ext(A,M ) → 0.
4

Roughly speaking, the group Ext(A,−) measures how far Hom(A,−) is from
being exact.

Let us calculate some examples:

• Let A = Z/p. Then we can take F0 = F1 = Z and
p
Z→
− Z → Z/p → 0
as a free resolution of Z/p. For an abelian group M , the sequence defining
Ext looks like
p
0 → Hom(Z/p,M ) → Hom(Z,M ) →
− Hom(Z,M ) → Ext(Z/p,M ) → 0.
Since Hom(Z,M ) = M , this sequence equals
p
0 → p-torsion in M → M →
− M → Ext(Z/p,M ) → 0.
Thus
p
Ext(Z/p,M ) = Coker(M →
− M ) = M/pM.

• For a concrete case, let us calculate Ext(Z/2, Z/2). We use the free reso-
lution
2
Z→
− Z → Z/2 → 0.
Applying Hom(−,Z/2) yields
2
0 → Hom(Z/2,Z/2) → Hom(Z,Z/2) →
− Hom(Z,Z/2).
This sequence is isomorphic to
2=0
0 → Z/2 → Z/2 −−→ Z/2.
Since 2 = 0 in Z/2, the second map is trivial. Hence the cokernel of
this map is just Z/2. Thus
Ext(Z/2, Z/2) = Z/2.
• More generally, one can show
Ext(Z/n,Z/m) = Z/ gcd(n,m)
where gcd(n,m) denotes the greatest common divisor of n and m.
Now we should study Ext in more detail. As a first step we show that it can
be viewed as a cohomology group:
 5

Lemma: Ext and Hom as cohomology groups

j
Let A and M be abelian groups and 0 → F1 → − F0 → A → 0 be a free
resolution of A. Consider the cochain complex Hom(F∗ ,M ) given by
j∗
0 → Hom(F1 ,M ) −
→ Hom(F0 ,M ) → 0
with Hom(F1 ,M ) in dimension zero and Hom(F0 ,M ) in dimension one.
Then we have
H 0 (Hom(F∗ ,M )) = Hom(A,M ) and H 1 (Hom(F∗ ,M )) = Ext(A,M ).

Proof: By definition, Ext(A,M ) is the cokernel of j ∗ . Since the differential

out of Hom(F0 ,M ) is trivial, the first cohomology is just
H 1 (Hom(F∗ ,M )) = Hom(F0 ,M )/Im (j ∗ ) = Coker(j ∗ ) = Ext(A,M ).

For H 0 we remember that the augmented sequence

j∗
0 → Hom(A,M ) → Hom(F1 ,M ) −
→ Hom(F0 ,M )
is exact.
Hence Hom(A,M ) is isomorphic to its image in Hom(F1 ,M ) which is, by ex-
actness of the sequence, the kernel of j ∗ . But this kernel is the cohomology group
of Hom(F∗ ,M ) in dimension 0:
H 0 (Hom(F∗ ,M )) = Ker (j ∗ ) = Hom(A,M )
QED
We should check that Ext does not depend on the choice of free resolution. To
do this, we are going to apply the lemma we proved for the Tor-case which states
that maps can be lifted to resolutions and any two lifts are chain homotopic in a
suitable sense.

Proposition: Ext is independent of resolutions

i
Ext is independent of the choice of free resolutuon: If 0 → E1 →
− E0 → A
j
and 0 → F1 → − F0 → A are two free resolutions of A, there is a unique
isomorphism
∼
=
Coker(Hom(i,M )) −
→ Coker(Hom(j,M )).
6

Proof: We know from the result on lifting resolutions that we can lift the
identity map on A to a map of resolutions

0 / E1
i / E0 / A / 0
f1 f0
 j 
0 / F1 / F0 / A / 0.

The other way around we get a lift

0 / E1
i / E0 / A / 0
g1 g0
 j 
0 / F1 / F0 / A / 0.

We write E∗ for the complex 0 → E1 → E0 → 0 and F∗ for the complex

0 → F1 → F0 → 0.
Composition yields maps f∗ ◦ g∗ : E∗ → E∗ and g∗ ◦ f∗ : F∗ → F∗ which lift
the identity map on A. But since the identity maps on E∗ and F∗ , respectively,
also lift the identity on A, the lemma of a previous lecture implies that there is
a chain homotopy hE between f∗ ◦ g∗ and 1E∗ and a chain homotopy hF between
g∗ ◦ f∗ and 1F∗ .
Now we apply Hom(−,M ). Then hE induces a cochain homotopy Hom(hE ,M )
0 / Hom(E0 ,M ) / Hom(E1 ,M ) / 0
h∗
1Hom(E0 ,M ) g0∗ ◦f0∗ g1∗ ◦f1∗ 1Hom(E1 ,M )

v 

0 / Hom(E0 ,M ) / Hom(E1 ,M ) / 0.
between
Hom(f∗ ◦ g∗ ,M ) = Hom(g∗ ,M ) ◦ Hom(f∗ ,M ) and Hom(1E∗ ,M ) = 1Hom(E∗ ,M ) .

Whereas hF induces a cochain homotopy Hom(hF ,M )

0 / Hom(F0 ,M ) / Hom(F1 ,M ) / 0
h∗
1Hom(F0 ,M ) f0∗ ◦g0∗ f1∗ ◦g1∗ 1Hom(F1 ,M )

v 

0 / Hom(F0 ,M ) / Hom(F1 ,M ) / 0.
between
Hom(g∗ ◦ f∗ ,M ) = Hom(f∗ ,M ) ◦ Hom(g∗ ,M ) and Hom(1F∗ ,M ) = 1Hom(F∗ ,M ) .
 7

Thus, the maps induced by the compositions on cohomology are equal to the
respective identity maps. In other words, the induced maps f ∗ and g ∗ on coho-
mology are mutual inverses to each other.
Moreover, since the chain homotopy type of f∗ and g∗ is unique by the lemma
of the lecture on Tor, they induce in fact a unique isomorphism
∼
=
Coker(Hom(i,M )) = H 1 (Hom(E∗ ,M ) −
→ H 1 (Hom(F∗ ,M ) = Coker(Hom(j,M )).

QED

Lemma: Induced exact sequence

Let M be an abelian group and assume we have a short exact sequence of
abelian groups
i p
0→A→
− B→
− C → 0.
Then there is an associated long exact sequence
0 / Hom(C,M ) / Hom(B,M ) / Hom(A,M )

s
Ext(C,M ) / Ext(B,M ) / Ext(A,M ) / 0.

Proof: Let 0 → E1 → E0 → A → 0 be a free resolution of A, and 0 → F1 →

F0 → C → 0 be a free resolution of C. This data gives us a free resolution of B
by forming direct sums:

0 → E1 ⊕ F1 → E0 ⊕ F0 → B → 0.

By the result of the previous lecture, we can lift the maps in the short exact
sequence to maps of resolutions

0 / E1 / E1 ⊕ F1 / F1 / 0

  
0 / E0 / E0 ⊕ F0 / F0 / 0

  
0 / A / B / C / 0

  
0 0 0
8

The horizontal sequences are short exact, since the middle term is a direct sum
of the other terms. Hence we get a short exact sequence of chain complexes
0 → E∗ → E∗ ⊕ F∗ → F∗ → 0.

Since all three complexes consist of free abelian groups, applying Hom(−,M )
yields a short exact sequence of cochain complexes
0 → Hom(F∗ ,M ) → Hom(E∗ ⊕ F∗ ,M ) → Hom(E∗ ,M ) → 0.

By taking cohomology of these cochain compelxes, we get an induced long

exact sequence of the associated cohomology groups. This is the desired exact
sequence together with the identifiction of H 1 with Ext and H 0 with Hom of the
previous lemma. QED
This lemma also gives a hint to where the name Ext comes from:

Ext and extensions

• We can think of a short exact sequence of abelian groups
0→A→B→M →0
as an extension of M by A. We can then say that two extensions are
euqivalent if they fit into an isomorphism of short exact sequences
0 / A / B / M / 0
∼
=

0 /A / B0 /M / 0.

• Note that we can always construct a trivial extension by taking the direct
sum of A and M :
(1,0)
0 → A −−→ A ⊕ M → M → 0.
Recall that we say that such a sequence splits.
• The group Ext(A,M ) measures how far extensions of M by A can be from
being from the trivial extension. For, we have
Ext(A,M ) = 0 ⇐⇒ every extension of M by A splits.
Proof: Given an extension, applying Hom(−,M ) yields an exact sequence
Hom(B,M ) → Hom(M,M ) → Ext(A,M ).
1
Thus the identity map M → − M lifts to a map B → M if Ext(A,M ) = 0.
But that is equivalent to that the initial short exact sequence splits. QED
 9

• Now one can show in general that Ext(A,M ) is in bijection with the set
of all equivalence classes of extensions of M by A.
• For example, we computed Ext(Z/2,Z/2) = Z/2. The trivial element in
Ext corresponds to the trivial extension
0 → Z/2 → Z/2 ⊕ Z/2 → Z/2 → 0
whereas the non-trivial element corresponds to the extension
2
0 → Z/2 →
− Z/4 → Z/2 → 0.

We summarize some further properties of Ext:

• Ext is functorial: For any homomorphisms of abelian groups A → A0
and M → M 0 , there are homomorphisms
Ext(A0 ,M ) → Ext(A,M ) and Ext(A,M ) → Ext(A,M 0 ).
This follows from the lemma on liftings of resolutions.

• If A is free, then Ext(A,M ) = 0 for any abelian group A. This follows

1
from the fact that 0 → A →
− A → 0 is a free resolution of A.

• Ext commutes with finite direct sums, i.e.,

Ext(A1 ⊕ A2 ,M ) ∼
= Ext(A1 ,M ) ⊕ Ext(A2 ,M ).
This follows from the fact that the direct sum of free resolutions of each
A1 and A2 is a free resolution of A1 ⊕ A2 .

• Let A be a finitely generated abelian group and let T (A) denote its torsion
subgroup. Since Ext(Z/m,Z) = Z/m, the structure theorem for finitely
generated abelian groups and the previous two points imply that
Ext(A,Z) ∼= T (A).
Now we prove the main result which connects homology and cohomology and
answers the question we raised last time about the Kronecker map κ:

Theorem: Universal Coefficient Theorem

Let C∗ be a chain complex of free abelian groups and let M be an abelian
group. We write C ∗ = Hom(C∗ ,M ) for the induced cochain complex.
Then there are natural short exact sequences
κ
0 → Ext(Hn−1 (C∗ ),M ) → H n (C ∗ ) →
− Hom(Hn (C∗ ),M ) → 0
10

for all n. These sequences split, but the splitting is not natural.

The proof builds on the same ideas as for the UCT in homology. But let us do
it anyway to get more practice.
Proof: • We write Zn for the kernel and Bn−1 for the image of the differential
d : Cn → Cn−1 . Since Cn and Cn−1 are free, both Zn and Bn−1 are free as well.
By definition of Zn and Bn , the restriction of the differentials to these groups
vanish. This implies that (Z∗ ,d) and (B∗ ,d) are chain complexes (with trivial
differentials).
Hence we get a short exact sequence of chain complexes
d
(1) 0 → Z∗ → C∗ →
− B∗−1 → 0.

• Since all groups in these chain complexes are free, applying the functor
Hom(−,M ) yields again a short exact sequence of cochain complexes
0 → Hom(B∗−1 ,M ) → Hom(C∗ ,M ) → Hom(Z∗ ,M ) → 0.
This follows from the lemma we proved in the previous lecture.
• Since the differentials in Z∗ and B∗ are trivial, the nth cohomology of
Hom(B∗−1 ,M ) is just Hom(Bn−1 ,M ), and the nth cohomology of Hom(Z∗ ,M ) is
just Hom(Zn ,M ).
Hence the long exact sequence in cohomology associated to the short exact
sequence (1) looks like
∂ d∗ i∗ ∂
· · · → Hom(Zn−1 ,M ) → → H n (Hom(C∗ ,M )) −
− Hom(Bn−1 ,M ) − → Hom(Zn ,M ) →
− Hom(Bn ,M ) → · · ·

∂
• The connecting homomorphism Hom(Zn ,M ) → − Hom(Bn ,M ) in this se-
quence is i∗n = Hom(in ,M ), where in : Bn ,→ Zn denotes the inclusion. For, the
connecting homomorphism is defined as follows. Consider the maps

Hom(Cn ,M ) / Hom(Zn ,M )
δ

Hom(Bn ,M ) / Hom(Cn+1 ,M ).
δ

A preimage of ϕ ∈ Hom(Zn ,M ) is any map ψ : Cn → M which restricts to Zn .

Such a preimage exists since the upper horizontal map is surjective. Then ψ is
 11

mapped to ψ ◦ d ∈ Hom(Cn+1 ,M ) by δ. Since every boundary is a cycle, we have

ψ ◦ d = ϕ ◦ d.
Now it remains to find a map ϕ̃ : Bn → M such that
ψ ◦ d = ϕ ◦ d = ϕ̃ ◦ d.
There is a canonical candidate for ϕ̃, namely the restriciton of ϕ to Bn . This is
exactly i∗n (ϕ).

16 M
ϕ FO

ψ
Zn / CO n ϕ̃ ϕ

d
 in
/ Bn  /
d
Cn+1 Zn

• A long exact sequence can always be cut into short exact sequences of
the from
0 → Coker(Hom(in−1 ,M )) → Hn (C ∗ ) → Ker (Hom(in ,M )) → 0.

Since the functor Hom(−,M ) sends cokernels to kernels, the kernel on the
right-hand side is just
Ker (Hom(in ,M )) = Hom(Coker(in ),M ) = Hom(Zn /Bn ,M ) = Hom(Hn (C∗ ),M ).

For the cokernel on the left-hand side, we use that

in−1
Bn−1 −−→ Zn−1 → Hn−1 (C∗ ) → 0
is a free resolution of Hn−1 (C∗ ).
Hence, after applying Hom(−,M ), we get an exact sequence
i∗n−1
0 → Hom(Hn−1 (C∗ ),M ) → Hom(Bn−1 ,M ) −−→ Hom(Zn−1 ,M ) → Coker(Hom(in−1 ,M )) → 0.

Thus, since Ext(−,M ) is independent of the chosen free resolution,

Coker(Hom(in−1 ,M )) = Ext(Hn−1 (C∗ ),M ).

Finally, to obtain the asserted splitting we use that subgroups of free abelian
groups are free. That implies that sequence (1) splits and we have
Cn ∼
= Zn ⊕ Bn−1 .
12

Applying Hom(−,M ) yields

Hom(Cn ,M ) ∼
= Hom(Zn ,M ) ⊕ Hom(Bn−1 ,M ).

Now one has to work a little bit more to get that this induces a direct sum
decomposition in homology.
It remains to check that the right-hand map in the theorem is in fact the
previously defined map κ. We leave this as an exercise. QED
Now we can prove the result we claimed in the previous lecture:

Corollary: From isos in homology to isos in cohomology

Let C∗ and D∗ be two chain complexes of free abelian groups. Let M be
an abelian group.
ϕ
Assume that there is a map C∗ −
→ D∗ which induces an isomorphism in
homology
∼
=
ϕ∗ : H∗ (C∗ ) −
→ H∗ (D∗ ).
Then this map also induces an isomorphism in cohomology with coeffi-
cients in M
∼
=
ϕ∗ : H ∗ (C ∗ ) −
→ H ∗ (D∗ ).

Proof: Since the construction of the long exact sequence we used in the proof
of the theorem is functorial, we see that ϕ induces a commutative diagram

0 / Ext(Hn−1 (C∗ ),M ) / H n (C ∗ ) / Hom(Hn (C∗ ),M ) / 0

(ϕ∗ )∗ ϕ∗ (ϕ∗ )∗
  
0 / Ext(Hn−1 (D∗ ),M ) / H n (D∗ ) / Hom(Hn (D∗ ),M ) / 0

The assumption that ϕ∗ induces an isomorphism implies that the two outer
vertical maps are isomomorphisms. The Five-Lemma implies that the middle
vertical map ϕ∗ is an isomorphism as well. QED

Our previous oberservations about Ext and torsion subgroups together with
the theorem imply:
 13

Corollary: Computing cohomology from homology

Assume that the homology groups Hn (C∗ ) and Hn−1 (C∗ ) of the chain com-
plex are finitely generated. Let Tn ⊆ Hn (C∗ ) and Tn−1 ⊆ Hn−1 (C∗ ) denote
the torsion subgroups. Then we can calculate the integral cohomology of
C ∗ = Hom(C∗ ,Z) by
H n (C ∗ ; Z) ∼
= (Hn (C∗ )/Tn ) ⊕ Tn−1 .

Since the singular chain complex S∗ (X,A) is an example of a chain complex

of free abelian groups, the theorem implies:

Corollary: UCT for singular cohomology

For each pair of spaces (X,A) there are split short exact sequences
0 → Ext(Hn−1 (X,A),M ) → H n (X,A; M ) → Hom(Hn (X,A),M ) → 0
for all n, and these sequences are natural with respect to maps of pairs
(X,A) → (Y,B).

As a final remark, we mention that there are versions of Ext for the category of
R-modules for any ring. The corresponding Ext-groups ExtR (M,N ) will depend
on the ring R as well as on the modules M and N . Moreover, there might be
non-trivial higher Ext-groups ExtiR (M,N ) for i ≥ 2, in general.
But the theory is very similar to the case of abelian groups, i.e., Z-modules,
as long as R is a principal ideal domain (PID). For, then submodules of free R-
modules are still free over R (which is not true in general). Hence free resolutions
of length two exist, and higher Ext groups vanish also in this case.
For example, fields are examples of PIDs. However, note that, for example,
Ext(Z/2,Z/2) = Ext1Z (Z/2,Z/2) = Z/2 whereas Ext1Z/2 (Z/2,Z/2) = 0. Hence the
base rings matter.