Ext Groups and Ext Functors

In this note we discuss the Hom and Ext functors and their connection with extensions
of Abelian groups. The theory we develop has an analogue in the category of R-modules for
any ring R; however, we will restrict to R = Z by considering only Abelian groups. A nice
reference for this material is Chapter 11 of Introduction to the Theory of Groups, (3rd ed.)
by Rotman.

1 The Hom Functors

If A and B are Abelian groups, then Hom(A, B) is also an Abelian group under pointwise
addition of functions. In this section we will see how Hom gives rise to classes of functors.
Let A denote the category of Abelian groups. A covariant functor T : A A associates to
every Abelian group A an Abelian group T (A), and for every homomorphism f : A B a
homomorphism T (f ) : T (A) T (B), such that the following properties hold:

1. T (idA ) = idT (A) for all A A;

2. If f : A B and g : B C, then T (g f ) = T (g) T (f );

3. If f, g : A B, then T (f + g) = T (f ) + T (g).

If, instead, (2) is replaced by the equation T (g f ) = T (f ) T (g), we say that T is a

contravariant functor. Moreover, the definition of a functor usually requires only the first
two conditions. Adding (3) usually goes under the name additive functor.

Example 1.1. Let M be an Abelian group, and let T = Hom(M, ). In other words,
T is the functor for which T (A) = Hom(M, A). To define the action of T on maps, let
f : A B be a homomorphism. We want a map T (f ) : Hom(M, A) Hom(M, B). By
drawing a diagram, we see that 7 f is such a homomorphism. We define T (f ) by
T (f )() = f . It is then easy to see T is a covariant functor. On the other hand, if we
define S = Hom(, M ), then we have a contravariant functor, which acts on maps as follows:
If f : A B, then S(f ) : Hom(B, M ) Hom(A, M ) is given by S(f )() = f . Note the
position of f in this formula. Since S reverses the order of maps, it is contravariant.

1
 To understand better the Hom functors, we need the following property. A covariant
f g
functor T is said to be left exact if for every exact sequence 0 A B C 0, the
T (f ) T (g)
induced sequence 0 T (A) T (B) T (C) is exact. A contravariant functor S is left
S(g) S(f )
exact if for every such sequence, the induced sequence S(C) S(B) S(A) 0 is exact.
Right exactness is defined in a similar way. If T = Hom(M, ) (resp. S = Hom(, M )), we
will write f (resp. f ) for T (f ) (resp. S(f )) if f : A B is a homomorphism. While we do
not need it here, the tensor product functor is right exact.

Proposition 1.2. Let M be an Abelian group. Then Hom(M, ) is a left exact covariant
functor and Hom(, M ) is a left exact contravariant functor.
f g
Proof. We prove the first statement; the second is analogous. Suppose that 0 A B
f g
C 0 is exact. We need to show that 0 Hom(M, A) Hom(M, B) Hom(M, C)
is exact. First, we show that f is injective. Let Hom(M, A) with f () = 0. Then
f = 0. Thus, (M ) ker(f ) = 0. This yields = 0. Next, if Hom(M, A), we have
g (f ()) = g (f ) = (g f ) = 0. This proves im(f ) ker(g ). Finally, we must
show that ker(g ) im(f ). To do this, suppose that : M B is an element of ker(g ).
Then g = 0. Thus, (M ) ker(g) = im(f ). Therefore, for each m M there is an
a A with (m) = f (a). The element a is unique since f is injective. Thus, we may define
a function : M A by (m) = a. In other words, (m) = f 1 ((m)); we interpret f 1 as
a homomorphism from im(f ) to A. The map is the composition of two homomorphisms,
so it is an element of Hom(M, A). Moreover, f ()(m) = f ((m)) = (m). Therefore,
f () = ; thus, im(f ). This finishes the proof.

Example 1.3. It is not true that, in general, Hom(M, ) is right exact. To give an example,
let M = Zp , and consider the exact sequence 0 pZ Z Zp 0. Applying Hom(M, )
yields the sequence 0 Hom(Zp , pZ) Hom(Zp , Z) Hom(Zp , Zp ). However, the first two
terms are 0; there are no nontrivial homomorphisms from the finite group Zp into the group
Z = pZ. Moreover, Hom(Zp , Zp ) is a nontrivial group. The map Hom(Zp , Z) Hom(Zp , Zp )
is thus not surjective.

In some sense, homological algebra arose from the failure of Hom (and other functors)
to fail to be both right and left exact. We will define new functors, the Ext functors, which
explain this failure. Before we do this, we show for which groups the corresponding Hom
functors are exact. To aid in the understanding of the following proposition, we recall some
terms from a graduate algebra course. An Abelian group P is said to be projective if for
every diagram
P

g 
B /C /0

2
there is a homomorphism : P B with g = . In other words, the diagram can be
completed to the following commutative diagram

P


 g 
B /C /0

Similarly, an Abelian group I is injective if for every diagram

f
0 /A /B


I
there is a map : B I with f = ,
f
0 /A /B

 
I
yielding the commutative diagram above. We will show that, in the category of Abelian
groups, projectives are the same thing as free Abelian groups. To help us do this, we need
the following lemma, which is actually more general than the statement that projectives are
free. In the proof of this lemma, we use the fact that Z is projective, and so that a short
exact sequence of the form 0 K F Z 0 is split exact.

Lemma 1.4. A subgroup of a free Abelian group is free Abelian.

Proof. Let P be a subgroup of a free group F . We give the proof in the case that the basis
of F is countable; let {ei : i N} be a basis for F . Then F =
L Ln
i=1 Zei . Set Fn = i=1 Zei .

Then Fn+1 /Fn = Z for each n. Write Pj = P Fj . Then Pn+1 /Pn is isomorphic to a
subgroup of Fn+1 /Fn = Zen+1 . Thus, either Pn+1 /Pn = 0 or Pn+1 /Pn = Z. However, in
the latter case, from the exact sequence 0 Pn Pn+1 Z 0, we see, since Z is
free, that Pn+1 = Pj Z. Let An be a summand of Pn+1 such that Pn+1 = Pn An . By
induction, it follows that Pn
Ln
= i=1 Ai . Since P is the union of all the Pn , we conclude that
P
L
= i=1 Ai . From this we see that P is free, since it is the direct sum of the nonzero Ai ,
all of which are isomorphic to Z. The argument for an arbitrary sized basis {ei : i I} goes
through much the same as that above once we well order the index set I and use transfinite
induction.

The analogue of this result in the category of R-modules is false: submodules of free
R-modules are usually not free. For an easy example, if R = Z6 , then 2R is a submodule of
order 3. A direct sum of copies of R has order at least 6, so 2R is not free. We will see how
this fact about Abelian groups has applications to the Ext functors we will define below.

3
Lemma 1.5. An Abelian group P is projective if and only if P is free Abelian. Also, an
Abelian group I is injective if and only if I is divisible.
Proof. Recall that a group A is divisible if for every a A and n N, there is b A
with nb = a. To prove the second statement first, suppose that I is injective. Let a I
and n N. From the sequence 0 nZ Z and the homomorphism : nZ I given
by (nm) = ma, there is a homomorphism : Z I with |nZ = . If b = (1), then
f
nb = (n) = a. Thus, I is divisible. Conversely, suppose that I is divisible. Let 0 A B
be exact and let : A M be a homomorphism. We use a Zorns lemma argument to
extend to B. We view A B by identifying A with its isomorphic image f (A). Let S
be the set of all pairs (C, ) with C a subgroup of B satisfying A C and : C I
with |A = . Ordering S via (C, ) (C 0 , 0 ) if C C 0 and 0 |C = . A short argument
shows that we can apply Zorns lemma to obtain a maximal element (B0 , ). We claim that
B0 = B. If not, there is a b B B0 . Let B1 = B0 + Zb, the group generated by B0 and b.
If B0 Zb = 0, we may extend by defining 0 (x + nb) = (x) for every n. If B0 Zb 6= 0,
then, as a subgroup of the cyclic group Zb, there is an m with B0 Zb = Zmb. Since I is
divisible and mb B0 , there is an r I with mr = (mb). We then define 0 on B0 + Zb
by 0 (x + nb) = (x) + nr. A short argument shows that this extension is well-defined and
is a homomorphism. Maximality gives a contradiction; thus, B0 = B. Therefore, we have a
map : B I with f = , as desired.

For the first statement, first suppose that P is free Abelian, and let B C 0 be exact,
and f : P C be a homomorphism. Let X be a basis of P as a free Abelian group. For each
x X, choose b B with (b) = f (x). Define g : P B to be the unique homomorphism
arising from the function X B given by x 7 b. Then g = f . Conversely, suppose that
P is projective. It is then a summand of a free group. By the previous lemma, P is free
Abelian.
Proposition 1.6. The functor Hom(M, ) is exact if and only if M is a free Abelian group.
The functor Hom(, M ) is exact if and only if M is a divisible group.
Proof. By the lemma, it is enough to prove that Hom(M, ) is exact if and only if M
is projective, and Hom(, M ) is exact if and only if M is injective. The arguments for
these two statements are quite similar, so we prove only the first. Suppose that M is
f g
projective, and let 0 A B C 0 be an exact sequence. We know that 0
f g
Hom(M, A) Hom(M, B) Hom(M, C) is exact. It suffices to prove that g is surjective.
Take Hom(M, C). We then have the diagram

M

g 
B /C / 0.

Thus, since M is projective, there is Hom(M, B) with g = . In other words,

g () = . Thus, g is surjective. Conversely, suppose that Hom(M, ) is exact. To

4
show M is projective, suppose that there is a surjection g : B C and a homomorphism
g
: M C. We have an exact sequence 0 ker(g) B C 0. Applying Hom(M, )
to this sequence and using exactness yields the exact sequence 0 Hom(M, ker(g))
g
Hom(M, B) Hom(M, C) 0. Since g is surjective, there is Hom(M, B) with
g () = . This says g = . Thus, M is projective.

We will produce Ext functors which will have the property that given an exact sequence
0 A B C 0, we obtain an exact sequence, for every group M , of the form

0 Hom(M, A) Hom(M, B) Hom(M, C)

Ext(M, A) Ext(M, B) Ext(M, C) 0,

along with a related sequence for the functor Hom(, M ). Thus, the Ext functors in some
sense repair the failure of exactness.

2 Ext(C,A)
When we discussed group extensions of an Abelian group A by a group G, we saw that if G
acts trivially on A and if we have a symmetric cocycle, the corresponding group extension
is an Abelian group. In other words, the extension we obtained is an object in the category
A. To keep ourselves in this category, if A and C are Abelian groups with C acting trivially
on A, we say that z : C C A is a symmetric cocycle if z(0, c) = z(c, 0) = 0 for all c C,
if z(c, d) = z(d, c) for all c, d C, and if z satisfies the cocycle condition

z(c, d) + z(c + d, e) = z(d, e) + z(c, d + e).

We will denote by Z(C, A) the group of all symmetric cocycles. A symmetric coboundary is
a symmetric cocycle of the form b(c, d) = l(c) + l(d) l(c + d) for some function l : C A.
We denote by B(C, A) the group of all symmetric coboundaries. We then define

Ext(C, A) = Z(C, A)/B(C, A).

In other words, Ext(C, A) = H 2 (C, A)sym . From what we did for group extensions, the
elements of Ext(C, A) are in 1-1 correspondence with equivalence classes of Abelian group
f g
extensions of C by A. We recall this correspondence. Let 0 A B C 0 be an
extension of A by C. Choose a function l : C B with g l = idC . Then the cocycle z
associated to l is the function given by

z(c1 , c2 ) = f 1 ((l(c1 ) + l(c2 ) l(c1 + c2 )) .

To understand this formula, we see that l(c1 ) + l(c2 ) l(c1 + c2 ) ker(g) = im(f ). Since
f is injective, there is a unique element a such that f (a) = l(c1 ) + l(c2 ) l(c1 + c2 ). Then
z(c1 , c2 ) = a. The group Ext(C, A) = 0 if and only if the only extension of A by C is the

5
direct product A C. Alternatively, Ext(C, A) = 0 if and only if every exact sequence of
the form 0 A B C 0 is split exact. Therefore, we see that Ext(A, D) = 0 if D
is divisible, and Ext(F, A) = 0 if F is free Abelian. Since we know that Hom(, D) is exact
if D is divisible and Hom(F, ) is exact if F is free, we do not need to repair exactness of
these functors via Ext functors; this rationale argues that, however we were to define Ext,
we should have these facts about Ext.

3 Pullbacks and Pushouts

We know, by the definition of addition of cocycles, that Ext(C, A) is an Abelian group in 1-1
correspondence with the set of extensions of A by C. Thus, we can translate this bijection
to give a definition of addition of extensions. We will describe this addition; to do so, we
first discuss pullbacks and pushouts.
Suppose that we have the following diagram
f
A /B

g

C

of Abelian groups. The pushout of this diagram is an Abelian group P , together with maps
: C P and : B P , such that f = g, and such that the following mapping
property holds: If Q is an Abelian group together with homomorphisms 0 : C Q and
0 : B Q such that 0 f = 0 g, then there is a unique homomorphism h : P Q such
that h = 0 and h = 0 .
f
A /B

g
  0
C
/P
??
??h
??
? 
0 +Q

The dual of this notion is called a pullback. If we have the following diagram

B
f
g 
C /A

then the pullback is an Abelian group P , together with maps : P B and : P C

satisfying g = f , such that for every Abelian group Q with maps 0 : Q B and
0 : Q C satisfying g 0 = f 0 , there is a unique homomorphism h : Q P with

6
 h = 0 and h = 0 .
Q?
?? 0
??h
??
 "
0 P
/B

f
  g 
C /A

We show that pushouts and pullbacks exist. To construct pushouts, suppose that we
have maps f : A B and g : A C. We define P = (B C) /S, where S =
{(f (a), g(a)) : a A}. It is clear that S is a subgroup of B C. We have maps : B P
and : C P given by (b) = (b, 0) + S and (c) = (0, c) + S. Furthermore, the definition
of S is exactly what we need to see that f = g, since

(f (a)) = (f (a), 0) + S = (0, g(a)) + S = (g(a))

for all a A. To show the mapping property, suppose that we have maps 0 : B Q and
0 : C Q with 0 f = 0 g. By the mapping property for direct sum, we have a map
B C Q, given by (b, c) 7 0 (b) + 0 (c). The kernel of this map contains S since 0 f =
0 g; therefore, there is an induced map h : P Q, given by h ((b, c) + S) = 0 (b) + 0 (c).
From this formula it follows that h = 0 and h = 0 . Finally, if k : P Q is
another map with k = 0 and k = 0 , then 0 (b) = k((b)) = k((b, 0) + S) and
0 (c) = k((c)) = k((0, c) + S). Combining these two equations, we get

k((b, c) + S) = k ((b, 0) + S) + k ((0, c) + S) = 0 (b) + 0 (c).

Thus, k = h. Therefore, P is the pushout of the diagram.

To construct pullbacks, we use similar ideas. Instead of constructing a quotient of a
direct sum, we use a subgroup of a direct product. Given f : B A and g : C A, we
set P = {(b, c) B C : f (b) = g(c)}. The canonical projection maps B C B and
B C C restrict to P to yield maps : P B and : P C. From the definition
of P it is clear that f = g . Suppose that there is a group Q and maps 0 : Q B
and 0 : Q C with f 0 = g 0 . The mapping property for direct product gives a map
Q B C, defined by q 7 (0 (q), 0 (q)). The equation f 0 = g 0 shows that the
image of Q is contained in P . Thus, we view this as a map h : Q P , and we have, by the
definition of h, that h = 0 and h = 0 . Moreover, an argument similar to that for
pushouts, we see that h is unique. Therefore, P is the pullback of the diagram.
f g f0 g0
A morphism (, , ) of extensions 0 A B C 0 and 0 A0 B 0 C 0 0
is a triple of homomorphisms such that the following diagram
f g
0 /A /B /C /0

 f0  g0 
0 / A0 / B0 / C0 /0

7
 f g f0
is commutative. An equivalence of two extensions 0 A B C 0 and 0 A
g0
B 0 C 0 of A by C is then a map (idA , , idC ). It is a consequence of the snake lemma
below that is automatically an isomorphism.

Lemma 3.1 (Snake Lemma). Suppose that

f g
A /B /C /0

 f0  g0 
0 / A0 / B0 / C0

is a commutative diagram. Then there is an exact sequence

ker() ker() ker() coker() coker() coker().

In particular, if and are injective (resp. surjective), then is injective (resp. surjective).

Proof. The proof of this result is just a long exercise in playing with commutative diagrams.
For a full proof, see a book on homological algebra. We will only prove the consequence stated
in the final sentence. First, suppose that and are injective, and suppose that (b) = 0.
Then 0 = g 0 (b) = g(b). Since is injective, g(b) = 0, so b = f (a) for some a A. Then
0 = (b) = f (a) = f 0 (a). Since both and f 0 are injective, a = 0, so b = f (a) = 0. Next,
suppose that and are surjective, and take b0 B 0 . Then g 0 (b0 ) = (c) for some c C.
Write c = g(b) for some b B. Then g 0 (b0 ) = (c) = g(b) = g 0 (b). Thus, b0 (b) = f 0 (a0 )
for some a0 A0 . Writing a0 = (a) for some a A, we have

b0 = (b) + f 0 (a0 ) = (b) + f 0 (a) = (b) + f (a)

= (b + f (a)).

Thus, is surjective.

The snake lemma is at the heart of several facts in homological algebra, most notably the
construction of the connecting homomorphisms of homology and cohomology groups. How-
ever, when we discuss the Ext groups below, we give ad-hoc constructions of the connecting
homomorphisms since we would have to discuss more homological algebra in order to make
use of the snake lemma.
The following two lemmas are the keys to constructing the Ext functors. To motivate their
statements, if A is an Abelian group, then we will prove that Ext(, A) is a contravariant
functor. This requires us to construct, for each homomorphism : C 0 C, a homomorphism
: Ext(C, A) Ext(C 0 , A). Similarly, if : A A0 is a map and C a group, to prove that
Ext(C, ) is a covariant functor, we need to produce a map : Ext(C, A) Ext(C, A0 ).
In terms of extensions, the first statement requires us to produce, for an exact sequence

8
 f g f0 g0
E : 0 A B C 0, an exact sequence E 0 : 0 A B 0 C 0 0 such that the
diagram
f0 g0
0 /A / B0 / C0 /0

id
 f  g 
0 /A /B /C /0

is commutative. In other words, we must produce a map (idA , , ) : E 0 E. The second

situation requires us to produce a map (, , idC ) : E E 0 . We will produce the required
extensions by means of pullbacks and pushouts.
f g
Lemma 3.2. Let E : 0 A B C 0 be an extension 0
and0
let : A A0 be a
f g
homomorphism. Then there is an extension E 0 : 0 A0 B 0 C 0, unique up to
equivalence, and a map (, , idC ) : E 0 E of extensions. Furthermore, if z is a cocycle
representing E, then z is a cocycle representing E 0 .

Proof. Let B 0 be the pushout of the diagram

f
A /B


A0

We then have maps : B B 0 and f 0 : A0 B 0 . Furthermore, from the map g : B C

and the zero map A0 C, the UMP for pushouts yields a unique map g 0 : B 0 C, yielding
the following commutative diagram.
f g
0 /A /B /C /0

id
 f0  g0 
A0 / B0 /C

Because the right square is commutative, the map g 0 is surjective. The composition g 0 f 0 = 0
by the definition of pushout. Finally, we must show that ker(g 0 ) = im(f 0 ) and that f 0
is injective. We do both by using the explicit description of B 0 = (A0 B) /S, where
S = {((a), f (a)) : a A}. The map f 0 is defined by f 0 (a0 ) = (a0 , 0) + S and is given by
(b) = (0, b) + S. Moreover, g 0 is defined by g 0 ((a0 , b) + S) = g(b). Thus,

ker(g 0 ) = {(a0 , b) + S : g(b) = 0} = {(a0 , b) + S : b ker(g) = im(f )}

= {(a0 , f (a)) : a0 A, a A} = {(a0 + (a), 0) + S : a0 A0 , a A}
= f 0 (A0 ).

Thus, ker(g 0 ) = im(f 0 ). Finally, we have ker(f 0 ) = {a0 A0 : (a0 , 0) S}. If (a0 , 0) S, then
there is an a A with (a0 , 0) = ((a), f (a)). From this we conclude that f (a) = 0, so
a = 0. Then a0 = (0) = 0.

9
 To determine the cocycle z 0 associated to E 0 , let l : C B with g l = idC . Then we
obtain a cocycle z representing E by the formula z(c1 , c2 ) = f 1 (l(c1 ) + l(c2 ) l(c1 + c2 )).
To obtain a cocycle for E 0 , we first need a map l0 : C B 0 with g 0 l0 = idC . Set
l0 = l. Commutativity of the diagram above shows that this is a valid choice. If z 0 is the
corresponding cocycle, then

z 0 (c1 , c2 ) = (f 0 )1 (l0 (c1 ) + l0 (c2 ) l0 (c1 + c2 ))

= (f 0 )1 (l(c1 ) + l(c2 ) l(c1 + c2 ))
= (f 0 )1 (f (z(c1 , c2 ))) = (f 0 )1 (f 0 (z(c1 , c2 )))
= (z(c1 , c2 )).

since is a homomorphism. Thus, z 0 = z. To finish the argument, we note that the

calculation to show z 0 = z did not need anything other than the fact that we have a map
(, , idC ) : E E 0 . Thus, if we have another extension Ee and a map (, 0 , idC ) : E E,
e
then z also represents E.e Since Ee and E 0 are represented by the same cocycle class, they
are equivalent extensions.
f g
Lemma 3.3. Let E : 0 A B C 0 be an extension and let : C 0 C be a map.
f0 g0
Then there is an extension E 0 : 0 A B 0 C 0 0, unique up to equivalence, and a
map (idA , , ) : E E 0 . Furthermore, if z is a cocycle representing E, then z ( ) is a
cocycle representing E 0 .

Proof. The ideas will be similar to the proof of the previous lemma, so we give fewer details.
Let B 0 be the pullback of the diagram

C0

g 
B /C

with associated maps g 0 : B 0 C 0 and : B 0 B. From the zero map A C and

f : A B the UMP yields a unique map f 0 : A B 0 with g 0 f 0 = 0 and f 0 = f . We
then have a diagram
f0 g0
A / B0 / C0

id
 f  g 
0 /A /B /C / 0.

Because the left square is commutative, f 0 is injective. We can show the top sequence yields
f0 g0
an exact sequence 0 A B 0 C 0 0 by a similar argument to that of the previous
lemma; we leave out the details.
To determine the cocycle for E 0 , let l : C B with g l = idC , and let z be the
corresponding cocycle. Recalling that B 0 = {(b, c0 ) B C 0 : g(b) = (c0 )}, we produce a
map l0 : C 0 B 0 with g 0 l0 = idC 0 by defining l0 (c0 ) = (l(c0 ), c0 ). Recalling the definition of

10
g 0 , we have g 0 l0 (c0 ) = g 0 (l(c0 ), c0 )) = c0 . Thus, g 0 l0 = id. The cocycle z 0 corresponding to
this choice of l0 is

z 0 (c01 , c02 ) = (f 0 )1 (l0 (c01 ) + l0 (c02 ) l0 (c01 + c02 ))

= (f 0 )1 ((l(c01 ), c01 ) (l(c02 ), c02 ) (l(c01 + c02 ), c01 + c02 ))
= (f 0 )1 ((l(c01 ) + l(c02 ) l(c01 + c02 ), 0))
1
= (f 0 ) (f z((c01 ), (c02 ), 0) = z((c01 ), (c02 ))

Since f 0 (a) = (f (a), 0).Therefore, z 0 = z ( ), as desired.

Finally, we show that E 0 is unique, up to equivalence. Suppose there is an extension
fe g
Ee : 0 A B e C 0 0 and a map (idC , , e) : Ee E. This map corresponds to the
e
e

commutative diagram
fe g
0 /A / e e
/ C0 /0
B
id
e
e
 f  g 
0 /A /B /C / 0.

We have the commutative diagram

B
e
g
e

g0 #
B0 / C0

e

  g 
B / C.

Since ge = g , e B 0 such that g 0 h = ge and

e the UMP for pullbacks gives a map h : B
h = .
e We then have a diagram

fe g
0 /A / e e
/ C0 /0
B
id h id
 f0  g0 
0 /A / B0 / C0 /0

id
 f  g 
0 /A /B /C /0

where we know that every square commutes except for perhaps to the top left corner. We
need to show that f 0 = h fe to see that it does commute. However, the uniqueness part of
the UMP for pullbacks will give this to us. We have maps f : A B and 0 : A C 0 such

11
that g f = 0 = 0, yielding the following commutative diagram

AA 0
AAAA hfe
AAAA
AAAA
f0 A g0 
B0 / C0
f

"  g 
B / C.

Since f 0 , hfe : A B 0 , the UMP for pullbacks will show that they will be equal if g 0 f 0 =
0 = g 0 hfe and f 0 = f = hfe. However, all of these equations hold from commutativity of
the various diagrams above. So, we do have a map (idA , h, idC ) : Ee E 0 ; therefore, these
extensions are equivalent.
f g f0 g0
Let E : 0 A B C 0 and E 0 : 0 A B 0 C 0 be two extensions of A by
C. To construct the Baer sum of two extensions, we first consider the direct sum
f f 0 gg 0
0 A A B B0 C C 0

of the extensions. It is easy to see that this sequence is exact. Define homomorphisms :
AA A and : C C C by (a1 , a2 ) = a1 +a2 and (c) = (c, c). By the two lemmas,
we have two extensions E1 : 0 A E1 C C 0 and E2 : 0 A E2 C 0
such that the following diagram

f f 0 gg 0
0 /AA / B B0 /C C /0

1 id
 f1  g1 
0 /A / E1 /C C /0
O O O
id 2
f2 g2
0 /A /E
2
/C /0

commutes. We call the extension E2 the Baer Sum of E and E 0 . We denote it symbolically
by E + E 0 .
f g f0 g0
Proposition 3.4. Let : E : 0 A B C 0 and E 0 : 0 A B 0 C 0 be
extensions of A by C. If z and z 0 are cocycles for these extensions, respectively, then z + z 0
is a cocycle for E + E 0 .

Proof. Let z and z 0 be the cocycles for E and E 0 , respectively. We recall how to define them.
We choose functions l : C B and l0 : C B 0 such that g l = idC and g 0 l0 = idC . Then

z(c1 , c2 ) = f 1 (l(c1 ) + l(c2 ) l(c1 + c2 )) ,

z 0 (c1 , c2 ) = (f 0 )1 (l0 (c1 ) + l0 (c2 ) l0 (c1 + c2 )) .

12
To have similar data for the direct sum extension, we define l l0 : C C B B 0 by
(l l0 ) ((c1 , c2 ), (c01 , c02 )) = (l(c1 , c2 ), l0 (c01 , c02 )).
It is easy to check that (g g 0 ) (l l0 ) = idCC . Furthermore, the cocycle arising from this
choice, which we denote by z z 0 , is given by
(z z 0 ) ((c1 , c2 ), (c01 , c02 )) = (z(c1 , c01 ), z 0 (c2 , c02 )).
To prove the proposition we need to write out the cocycle for the extension E+E 0 . We combine
the results of the two lemmas to do this. Since z z 0 is the cocycle for the top extension,
(z z 0 ) is the cocycle for the middle extension, and then y = ( (z z 0 )) ( ) is
the cocycle for the bottom extension. To be more explicit, if c1 , c2 C, then
y(c1 , c2 ) = (z z 0 )((c1 , c1 ), (c2 , c2 )) = (z(c1 , c2 ), z 0 (c1 , c2 ))
= z(c1 , c2 ) + z 0 (c1 , c2 ).
Thus, z + z 0 is a cocycle representing our extension E + E 0 .
The proposition then tells us that addition in Ext(C, A) corresponds to Baer sum of
extensions, via the identification of an element of Ext(C, A) with an extension of A by C.

4 The Ext Functors

Just as Hom yields a class of functors, so does Ext. We describe these functors now. For
each Abelian group M we will obtain two functors Ext(M, ) and Ext(, M ). The first
is covariant and the second contravariant. We know what these functors do to objects,
since they send a group A to Ext(M, A) and Ext(A, M ), respectively. We must define their
action on maps. Fix an Abelian group M . If : A A0 is a homomorphism, we define
: Ext(M, A) Ext(M, A0 ) by
(z + B(M, A)) = z + B(M, A0 ).
Similarly, we define : Ext(A0 , M ) Ext(A, M ) by
(z + B(A0 , M )) = z ( ) + B(A, M ).
To help to understand these definitions, we note that these formulas arise from Lemmas 3.2
and 3.3. In terms of extensions, if E : 0 A B M 0 is an extension of C by A,
then (E) : 0 A0 B 0 M 0 is the unique up to equivalence extension making the
following diagram commute.
f g
0 /A /B /M /0

id
 f0  g0 
0 / A0 / B0 /M /0

A similar diagram helps to understand .

13
Lemma 4.1. Ext(M, ) is a covariant functor and Ext(, M ) is a contravariant functor.
Proof. We prove only the first statement; the proof of the second is similar. Let z Z(M, A),
and let : A A0 be a homomorphism. We show that z is a symmetric cocycle. First,
z(m, 0) = (z(m, 0)) = (0) = 0 and z(0, m) = (z(0, m)) = 0. Second, z(m, n) =
(z(m, n)) = (z(m, n)) = z(m, n). Finally, to check the cocycle condition, let m, n, p
M . Then z(m, n) + z(m + n, p) = z(n, p) + z(m, n + p). Applying to this equation and
remembering that is a group homomorphism gives z(m, n) + z(m + n, p) = z(n, p) +
z(m, n + p). Thus, z is a cocycle. If z is a coboundary, then there is a function
l : M A with f (m, n) = l(m) + l(n) l(m + n) for all m, n M . Then z(m, n) =
l(m) + l(n) l(m + n). Thus, as l : M C, we see that z is a coboundary.
Therefore, is well-defined. Because is a homomorphism and addition in Z(M, A) is
pointwise, it is clear that is a homomorphism.
To show that Ext(M, ) is a functor, we have to show that it behaves correctly on maps.
First, consider the identity id : A A. Then id : Ext(M, A) Ext(M, A) is defined by
z + B(M, A) 7 id z + B(M, A) = z + B(M, A). Thus, id is the identity on Ext(M, A).
Next, let : A C and : C D be homomorphisms. Then

() (z + B(M, A)) = z + B(M, D) = (z + B(M, C))

= ( (z + B(M, A)).

Therefore, () = . Finally, if , : A C are homomorphisms, then

( + ) (z + B(M, A)) = ( + ) z + B(M, C) = (z + z) + B(M, C)

= (z + B(M, C)) + (z + B(M, C)).

Thus, ( + ) = + . This finishes the proof that Ext(M, ) is a covariant functor.

f g
Our goal is, for each exact sequence 0 A B C 0 and each group M , to
produce an exact sequence
f g
0 Hom(M, A) Hom(M, B) Hom(M, C)
f g
Ext(M, A) Ext(M, B) Ext(M, C) 0

and an exact sequence

g f
0 Hom(C, M ) Hom(B, M ) Hom(A, M )
g f
Ext(C, M ) Ext(B, M ) Ext(A, M ) 0.

We need to define the connecting homomorphisms . We define : Hom(M, C) Ext(M, A)

as follows. Let Hom(M, C). If z is the cocycle associated to the exact sequence, we
define () = z ( ) + B(M, A). In other words, () is represented by the cocycle
that sends (m, n) to z((m), (n)). Similarly, we define : Hom(A, M ) Ext(C, M ) by
() = z + B(C, M ).

14
 f g
Theorem 4.2. Let 0 A B C 0 be an exact sequence of Abelian groups. For
every Abelian group M , the sequences
f g
0 Hom(M, A) Hom(M, B) Hom(M, C)
f g
Ext(M, A) Ext(M, B) Ext(M, C) 0

and an exact sequence

g f
0 Hom(C, M ) Hom(B, M ) Hom(A, M )
g f
Ext(C, M ) Ext(B, M ) Ext(A, M ) 0

are exact.

Proof. We prove the exactness of the second sequence and leave the proof of the first sequence
to the reader. Let z be a cocycle representing the exact sequence. We first show that
the composition of any pair of consecutive maps is 0. For the pair of g and f , we have
f g = (g f ) = 0 = 0 since Hom(, M ) and Ext(, M ) is a contravariant functor.
Next, to see that f = 0, take Hom(B, M ). Then (f ()) = ( f ) = f z. We
recall that z(c1 , c2 ) = f 1 (l(c1 ) + l(c2 ) l(c1 + c2 )) for an appropriate function l : C B.
Then f z(c1 , c2 ) = l(c1 ) + l(c2 ) l(c1 + c2 ). Since l : C M , we see that f z satisfies
the definition of a coboundary. Thus, ( f ) (z) = 0 in Ext C, M ).
Next, we see that g = 0. Take Hom(A, M ). Then g (()) is the class of the
cocycle y, where
y(b1 , b2 ) = f 1 (lg(b1 ) + lg(b2 ) lg(b1 + b2 )) .
The function l satisfies g l = idC . Therefore, for each b B there is an a A with
lg(b) b = f (a). Consequently, there is a function t : B A such that lg(b) b = f (t(b)).
The cocycle y can then be represented as

y(b1 , b2 ) = f 1 (f t(b1 ) + f t(b2 ) f t(b1 + b2 ))

= t(b1 ) + t(b2 ) t(b1 + b2 ).

The function t : B M , and so this is the definition for y to be a coboundary. Thus,

y = 0 in Ext(B, M ). Therefore, g = 0.
At this point we have verified that the sequence is a zero sequence. Since Hom is right
g f
exact, 0 Hom(C, M ) Hom(B, M ) Hom(A, M ) is exact. To prove exactness, we then
have to prove exactness at the final Hom term and the three Ext terms.
ker() = im(f ) : Take Hom(A, M ) with () = 0. The extension corresponding to
() is then split exact. By Lemma 3.2, we have a commutative diagram
f g
0 /A /B /C /0

id
  
0 /M i /BC /C / 0.

15
Let j : M C M be the canonical projection. Set = j Hom(B, M ). We have
f = jf = ji = , so = f ( ). Thus, ker() = im(f ).
ker(g ) = im() : Let z 0 Ext(C, M ) such that g (z 0 ) = 0. By Lemma 3.3, there is a
commutative diagram
f g
E: 0 /A /B /C /0

j id
f0  g0 
E0 : 0 /M /E /C /0
O O O
id g

E 0 : 0 /M i /BM /B / 0.

Let j : B B M be the canonical map. Then g 0 jf = gjf = gf = 0. Therefore, for

each a A, there is an m M with g 0 jf (a) = f 0 (m). Since f 0 is injective, m is unique.
Thus, there is a map : A M with g 0 jf = f 0 . We thus have a commutative diagram
f g
E: 0 /A /B /C /0

j id
 f0  g0 
E0 : 0 /M /E /C /0

By the description of the cocycle z 0 for the extension E 0 , we have z 0 = z = (). Thus,
z 0 im().
f0 g0
ker(f ) = im(g ) : Suppose that 0 M E B is an extension of M by B whose
image extension in Ext(A, M ) is 0. By Lemma 3.3, we construct this image extension by
forming the pullback of the maps f : A B and g 0 : E B. We have the following
commutative diagram
f0 g0
0 /M /E /B /0
O O O
id f

0 /M i /P /A / 0.

Since the bottom extension represents 0 in Ext(A, M ), the sequence is split exact. Thus,
there are maps j : A P and k : P M with j = idA and ki = idM . Let Q be the
pushout of the maps id : M M and f 0 : M E. By the UMP for pushouts, there is a
map h : Q C with h = g g 0 and h t = 0. Thus, the following diagram commutes

0 /M t /Q h /C /0
O O O
id g
f0 g0
0 /M /E /B /0
O O O
id f

0 /M i /P /A / 0.

16
If we show that the top row is exact, then it will represent an element of Ext(C, M )
mapping to the given element of Ext(B, M ). We recall that Q = (M E)/S, where
S = {(m, f 0 (m) : m M }. By definition of h, the sequence is a zero sequence. The
map t is injective, since t(m) = (m, 0) + S, and (m, 0) S if and only if m = 0. Also, h is
surjective, since if c C, then c = g(b) for some b B, and so c = gg 0 (e) for some e. Then
h ((0, e) + S) = c. Finally, we show that ker(h) = im(t). Suppose that h ((m, e) + S) = 0.
Then gg 0 (e) = 0. So, g 0 (e) = f (a) for some a A. But, f (a) = g 0 j(a). Set p = j(a).
Then (p) = f 0 k(p). Consequently, g 0 (e) = g 0 f 0 k(p) = 0. So, e = f 0 (m0 ) for some m0 , and so
(m, e) + S = (m, f 0 (m0 ) + S = (m m0 , 0) + S = t(m m0 ), as desired.
The hardest step is to show that f is surjective. For this we need to use that we are
in the category of Abelian groups and not a category of modules; the fact we need is that
subgroups of free Abelian groups are free Abelian. Let y be a cocycle representing a class in

Ext(M, A), and let 0 M E A 0 be the corresponding extension. We then obtain
f g
an exact sequence 0 M E B C 0. There is a free Abelian group F and an

exact sequence 0 K F C 0. Thus, K is also free Abelian, and so both K and F
are projective. Therefore, there is a homomorphism : F B with g = . Furthermore,
g = = 0; thus, ((K)) ker(g) = im(f ). Thus, since K is projective, there is
a homomorphism : K E with f = . We then have the following commutative
diagram

0 /K /F  /C /0

id
 f  g 
0 /M /E /B /C / 0.

Let E 0 be the pushout of the diagram

K /F


E
We then have a map h : E 0 B such that the following diagram commutes

K /F

0
 
M
/E / E0
BB
BB h
BB
BB
! 
f +
B.
h
We claim that 0 M E 0 B 0 is an extension of M by B whose image in Ext(M, B)
maps to the given extension in Ext(M, A). We have several things to verify. First, we
see that h = f = 0 by the exactness of the 6-term sequence above. Next, we show
that h is surjective. To do this, let b B. Then g(b) = (x) for some x F . Then

17
g(x) = (x) = g(b). Thus, (x) = b + f (a) for some a A. If we write a = (e), then
h(e) = f (e) = f (a). Thus,
b = (x) f (a) = h0 (x) h(e) im(h),
h
as desired. To finish the proof that the sequence 0 M E 0 B 0 is exact, we
show that ker(h) = im(). Recall that E 0 = (E F ) /S with S = {((k), (k)) : k K}.
Suppose that h ((e, x) + S) = 0. Then by definition of h, we have f (e)+(z) = 0. Applying
g yields 0 = g(f ) = (z). Thus, since 0 K F C 0 is exact, z = (k) for some
k K. Then
f (e) = (z) = (k) = f (k).
This yields f (e + (k)) = 0, so e + (k) = (m) for some m M . Finally, we see that
(e, x) + S = (e + (k), x (k)) + S
= ((m), 0) + S = (m).
This finishes the proof that the sequence is exact. To show that this sequence maps onto
the given one, we have the following diagram

0 /M /E /A /0

id f
  
0 /M / E0 h /B /0

which clearly commutes. If y 0 is a cocycle representing the bottom extension, then (y 0 ) = y;

this follows from Lemma 3.3. This proves that f is surjective.
Example 4.3. Let p be a positive integer. We compute the group Ext(Zp , Zp ); in fact,
p
we determine Ext(Zp , B) for any group B. Consider the exact sequence 0 Z Z
Zp 0, where the map Z Z is multiplication by p. Recall that, since Z is free Abelian,
Ext(Z, B) = 0. Applying the theorem to B, we have the exact sequence
p
0 Hom(Zp , B) Hom(Z, B) Hom(Z, B)
p
Ext(Zp , B) Ext(Z, B) Ext(Z, M ) 0.
Recall that Hom(Z, B) = B via the map 7 (1). Also, Hom(Zp , B) = p B , the p-torsion
subgroup of B, which is defined by p B = {b B : pb = 0}. This isomorphism is given by
7 (1). A short computation, using the isomorphism Hom(Z, B) = B shows that p
translates to the multiplication by p map on B. The long exact sequence above then reduces
to
p
0 p B B B Ext(Zp , B) 0.
Therefore, Ext(Zp , B) is the cokernel p , so Ext(Zp , B)
= B/pB. In particular, Ext(Zp , Zp )
=
Zp
= Ext(Z p , Z). As we saw in an example earlier, applying Hom(, Zp ) to the sequence
p
0 Z Z Zp 0 ruins exactness; we obtain the sequence 0 0 0 Zp . By using

the Ext groups, we fill it out to the exact sequence 0 0 0 Zp Zp 0 0.

18
5 Homological Algebra view of Ext
We give a brief description of how Ext groups arise in the larger context of homological
algebra. This section assumes some knowledge of chain complexes. Let A and M be Abelian
groups. For n 0 let An = Map(An+1 , M ) be the set of all functions An+1 M . By using
pointwise addition, we view An as an Abelian group. If Fn is the free Abelian group on the
set An+1 , then the UMP for free Abelian groups implies that An = Hom(Fn , M ). We make
{An } into a cochain complex by defining dn1 : An1 An by
n1
X
dn1 (f )(a0 , . . . , an ) = f (a1 , . . . , an ) + (1)i f (a0 , . . . , ai + ai+1 , . . . , an )
i=1
+ (1)n f (a0 , . . . , an1 ).

For some notation, we write A1 = 0 and d1 = 0. By a boring calculation, one sees that
dn dn1 = 0 for all n. We then define Extn (A, M ) to be the homology at the n-th stage of
this complex. In other words,

Extn (A, M ) = ker(dn )/ im(dn1 ).

For n = 0, we have Ext0 (A, M ) = ker(d0 )/ im(d1 ) = ker(d0 ) since d1 = 0. As a special case
of dn , we see that d0 (f )(a0 , a1 ) = f (a1 ) f (a0 + a1 ) + f (a1 ). Thus, f ker(d0 ) if and only if
f is a homomorphism, and d0 (f ) im(d0 ) is a 2-coboundary. Moreover, if f : A A M ,
then d1 (f ) = 0 if and only if, for all a0 , a1 , a2 A, we have

f (a1 , a2 ) f (a0 + a1 , a2 ) + f (a0 , a1 + a2 ) f (a0 , a1 ) = 0.

Thus, d1 (f ) = 0 if and only if f is a 2-cocycle. Thus, we see that Ext1 (A, M ) = Ext(A, M )
and Ext0 (A, M ) = Hom(A, M ).
An alternative, more abstract view of this situation is via free resolutions. If F :
Fn F1 A 0 is a free resolution of A; that is, an exact sequence with
each Fi a free Abelian group, then Extn (A, M ) is the n-th homology group of the complex
{Hom(F, M )}. In other words, we consider the cochain complex

Hom(F1 , M ) Hom(F2 , M )

and take homology of the complex. Note that the maps changed direction since Hom(, M )
is contravariant. The Fi defined above provide a free resolution of A, since we have the
complex
Fn F1 F0 A 0
defined as follows: the map d0 : F0 A is the map arising from the UMP for free Abelian
groups and from the identity map A A; recall that F0 is the free Abelian group on A. If
n 1, then dn : Fn Fn1 arises from the UMP and the map An+1 An Fn1 given by
n
X
(a0 , . . . , an ) 7 (1)i (a0 , . . . , ai + ai+1 , . . . , an ),
i=0

19
One gets the maps An = Hom(Fn , M ) An+1 = Hom(Fn+1 , M ) by applying the functor
Hom(, M ) to the map Fn+1 Fn . It is a theorem of homological algebra that the homology
groups one gets do not depend on the choice of free resolution of A.
If A is an Abelian group and F is a free Abelian group such that there is an exact
sequence 0 K F A 0, then K is also free Abelian. Thus, we can take this to be a
free resolution. If we apply the machinery to this extension, then we see that all homology
groups for n 2 are 0. This implies that Extn (A, M ) = 0 if n 2. Therefore, it was
sufficient for us to define Ext = Ext1 .
As a final note, if R is a ring, we could define groups ExtR (A, M ) for any two R-modules
A and M , by considering the problem of extensions in the category of R-modules. In
fact, we can define a sequence {ExtnR (A, M ) : n N} of groups, for which Ext0R (A, M ) =
HomR (A, M ) and Ext1R (A, M ) = ExtR (A, M ).

20