EXT, TOR AND THE UCT

CHRIS KOTTKE

Contents
1. Left/right exact functors 1
2. Projective resolutions 2
3. Two useful lemmas 3
4. Ext 6
5. Ext as a covariant derived functor 8
6. Universal Coefficient Theorem for Cohomology 10
7. Tensor product 12
8. Tor 13
9. Universal Coefficient Theorem for Homology 15
References 16

1. Left/right exact functors

We start with the following observation.
Proposition 1.1. For fixed G ∈ AbGp, if 0 → A → A0 → A00 → 0 is a short exact
sequence of abelian groups, then
0 −→ Hom(A00 , G) −→ Hom(A0 , G) −→ Hom(A, G)
is exact.
We say the functor Hom(−, G) is only left exact. In general we have the following
definition.
Definition 1.2. Let F : AbGp −→ AbGp be a contravariant functor, and let
0 −→ A −→ A0 −→ A00 −→ 0 (1)
be a short exact sequence. If the corresponding sequence
0 −→ F (A00 ) −→ F (A0 ) −→ F (A) (2)
is exact, we say F is left exact. If instead
F (A00 ) −→ F (A0 ) −→ F (A) −→ 0
we say F is right exact. If F is both right and left exact, we say it is exact;
equivalently F is exact if
0 −→ F (A00 ) −→ F (A0 ) −→ F (A) −→ 0
is exact.
1
2 CHRIS KOTTKE

Similarly, if F is a covariant functor, we say it is left exact if

0 −→ F (A) −→ F (A0 ) −→ F (A00 )
is exact, and so on.
The theory of derived functors (see [Wei95]) gives a mechanism for computing
(in principle) a continuation of (2) to an exact sequence
0 −→ F (A) −→ F (A0 ) −→ F (A00 ) −→ R1 F (A)
−→ R1 F (A0 ) −→ R1 F (A00 ) −→ R2 F (A) −→ · · ·
where the objects Ri F (A) are known as the ith right derived functors of F.
We will not pursue the completely general theory1 but rather stick to the case
of Hom(−, G) and − ⊗ G in the category of abelian groups, whose derived functors
go by the names of Ext∗ (−, G) and Tor∗ (−, G), respectively. However, we will nev-
ertheless go through the “proper” construction of these objects, to get the essence
of the theory.

2. Projective resolutions
The main tool in this theory is the notion of a “projective” object, which in
essence allows us to infer the existence of lifting maps in particular situations.
Definition 2.1. We say an abelian group P is projective if given a homomorphism
f : P −→ G and a surjective homomorphism h : G0 −→ G, there exists a lift
f 0 : P −→ G0 such that
P
∃ f0
f (3)
h
G0 G 0
commutes.
Remark. The defining property of a projective object is a categorical one, and makes
sense in any so-called “abelian category” where things like kernels, cokernels and
exactness make sense. For instance, we can define projective objects in categories
of chain complexes, modules over a ring R, sheaves of such objects and so on.
Definition 2.2. Fix A ∈ AbGp. A projective resolution of A is an exact sequence
d d d
· · · −→ P1 −→ P0 −→ A −→ 0
where the objects Pi are projective.
It turns out that there are particularly simple resolutions for any abelian group
using free groups.
Lemma 2.3. Any free abelian group is projective. Furthermore, for any A ∈ AbGp,
there is a two step projective resolution
0 −→ R −→ F −→ A −→ 0
where R and F are free groups.
1which can be very general indeed, leading eventually to the theory of so-called “derived
categories” which are of importance in algebraic geometry.
 EXT, TOR AND THE UCT 3

Proof. For the first claim, suppose P is free. It suffices to define f 0 on generators
{pi }i∈I of P , and we can set
f (pi ) = gi0 , for any gi0 ∈ h−1 f (pi ) .

For the second claim, let F be the free abelian group on any set of generators for
A. Then F clearly surjects onto A and we let R be its kernel. It is then a standard
fact that any subgroup of a free group is free, so R is also free. 

Because of this result, we could restrict ourselves to free resolutions, and two
step ones at that. Nevertheless, we will continue to use projective resolutions to
illustrate general ideas.

3. Two useful lemmas

The general yoga of derived functors is to execute the following steps
(1) Take a projective resolution
· · · −→ P1 −→ P0 −→ A −→ 0
of our given object.
(2) Chop off the A from the end to obtain a (no longer exact) sequence
· · · −→ P2 −→ P1 −→ P0 −→ 0
This will serve in a certain sense as a substitute for A. Observe that it is
still a complex since the composition of two maps is 0.
(3) Apply the left exact functor F (−) = Hom(−, G) (say, for definiteness) to
get
0 −→ F (P0 ) −→ F (P1 ) −→ F (P2 ) −→ · · ·
which is a complex since F takes the 0 morphism to the 0 morphism.
(4) Compute the (co)homology groups of this complex to get
Ker {F (Pi ) −→ F (Pi+1 )}
Ri F (A) :=
Im {F (Pi−1 ) −→ F (Pi )}
these are the right derived functors of F applied to A. We will observe that
R0 F (A) ≡ F (A), i.e. the 0th derived functor of F just gives us F back.
(5) Observe that for any exact sequence
0 −→ A −→ A0 −→ A00 −→ 0 (4)
there is a long exact sequence
0 −→ F (A00 ) −→ F (A0 ) −→ F (A) −→ R1 F (A00 )
(5)
−→ R1 F (A0 ) −→ R1 F (A) −→ R2 F (A00 ) −→ · · ·
In order to make this go through, we need to verify that the result is independent
of the projective resolution that we choose, and that we can make the projective
resolutions for (4) fit together into a (split) short exact sequence of complexes,
so that we get the long exact sequence (5) when we take homology. We’ll prove
these things in the following two lemmas. Observe that the proofs are mostly
applications of the projectivity property (3), and as such are valid in a much more
general context.
4 CHRIS KOTTKE

Lemma 3.1. Let f : A −→ B be a homomorphism of abelian groups and let

· · · −→ P1 −→ P0 −→ A −→ 0
and
· · · −→ Q1 −→ Q0 −→ B −→ 0
be projective resolutions. Then there is a chain map f∗ : (P∗ −→ A) −→ (Q∗ −→ B)
extending f , in other words a sequence of homomorphisms fi : Pi −→ Qi such that

··· P1 P0 A 0

f1 f0 f

··· Q1 Q0 B 0

commutes. Furthermore, any two such extensions of f are chain homotopic.

Proof. To prove existence of the fi , we proceed by induction on i, considering the
given map f−1 := f : A −→ B as the base case. Thus assume that fi : Pi −→ Qi
has been constructed. Denoting by Zi (Pi ) and Zi (Qi ) the cycles in Pi and Qi
respectively, observe that fi : Zi (Pi ) −→ Zi (Qi ) since fi−1 d = d fi . Thus consider
the diagram
d
Pi+1 Zi (Pi ) 0

fi+1 fi
d
Qi+1 Zi (Qi ) 0

The composition fi d gives a map from Pi+1 to Zi (Qi ), onto which Qi+1 surjects.
Thus the requisite map fi+1 is furnished by the defining property (3) of the pro-
jective group Pi+1 , completing the induction.
To show that two extensions {fi } and {fi0 } are chain homotopic, we consider
the difference gi := fi − fi0 and construct a chain homotopy {si : Pi −→ Qi+1 } such
that g = ds + sd. Again we proceed by induction. Observe that g−1 = f − f ≡ 0,
so that g0 maps P0 into cycles Z0 (Q0 ), and therefore lifts to a map s0 : P0 −→ Q1
as in the following diagram:

P0 A
g0 g−1 ≡ 0
s0
Q1 Z0 (Q0 ) 0

This gives the base case for the induction.

Suppose then that si : Pi −→ Qi+1 has been constructed such that gi = si−1 d +
d si , or equivalently
d si = gi − si−1 d.
It follows that the map gi+1 − si d maps Pi+1 into cycles Zi+1 (Qi+1 ) since



d gi+1 − si d = gi d − gi − si−1 d d = gi d − gi d = 0.
 EXT, TOR AND THE UCT 5

Thus we have the diagram

Pi+1

gi+1 − si d
si+1
Qi+2 Zi+1 (Qi+1 ) 0

and projectivity furnishes the map si+1 such that d si+1 = gi+1 − si d. 

Lemma 3.2. Let 0 −→ A −→ A0 −→ A00 −→ 0 be a short exact sequence. Then

there is a short exact sequence of projective resolutions P∗ −→ A, P∗0 −→ A0 and
P∗00 −→ A00 such that

0 0 0

··· P1 P0 A 0

··· P10 P00 A0 0

··· P100 P000 A00 0

0 0 0

commutes and the exact sequences 0 −→ Pi −→ Pi0 −→ Pi00 −→ 0 are split.

Proof. Choose any projective resolutions P∗ −→ A and P∗00 −→ A00 , giving a dia-
gram

··· P1 P0 A 0

A0

··· P100 P000 A00 0

We will fill in the middle with a projective resolution satisfying the claimed prop-
erties. (For this reason, this lemma is sometimes referred to as the “horseshoe
lemma”.)
Composition gives a map P0 −→ A0 , and a map P000 −→ A0 is furnished by
projectivity. These combine to give a map P0 ⊕P000 −→ A0 , and we set P00 := P0 ⊕P000 ,
6 CHRIS KOTTKE

obtaining the diagram

0 0

P0 A 0

P0 ⊕ P000 A0

P000 A00 0

0 0

The sequence P0 −→ P0 ⊕ P000 −→ P000 is obviously split exact; we will show that
the map P0 ⊕ P000 −→ A0 is surjective.
Fix 0 6= α0 ∈ A0 . There are two cases to consider. If α0 ∈ Ker {A0 −→ A00 }, then
it is in the image of the map from A, and therefore also in the image of the map
from P0 . If α0 ∈ / Ker {A0 −→ A00 } then it has nonzero image α00 ∈ A00 , and there is
some p ∈ P0 mapping onto α00 . By commutativity of (3), this p00 maps onto α0 ,
00 00

so every element in A0 is either in the image of P0 or P000 .

The proof continues by induction, replacing 0 −→ A −→ A0 −→ A00 −→ 0 by
the short exact sequence of cycles 0 −→ Zi (Pi ) −→ Zi (Pi0 ) −→ Zi (Pi00 ) −→ 0. 

4. Ext
We are now ready to construct the groups Exti (A, G) as derived functors of the
functor A 7−→ Hom(A, G).
Choose a projective resolution
· · · −→ P1 −→ P0 −→ A −→ 0
and apply Hom(−, G) to the truncated sequence P∗ −→ 0, obtaining
0 −→ Hom(P0 , G) −→ Hom(P1 , G) −→ · · · (6)
which is a chain complex.
Definition 4.1. The group Exti (A, G) is the ith (co)homology group of (6):
Ker {Hom(Pi , G) −→ Hom(Pi+1 , G)}
Exti (A, G) := (7)
Im {Hom(Pi−1 , G) −→ Hom(Pi , G)}

First we make some observations that are valid for right derived functors in
general.
Proposition 4.2. The groups Exti (A, G) are independent of the projective resolu-
tion of A.
Proof. Let P∗ −→ A and Q∗ −→ A be two projective resolutions. Lemma 3.1 gives
extensions of the identity map Id : A −→ A to chain maps f : P∗ −→ Q∗ and
g : Q∗ −→ P∗ . Observe that both g f : P∗ −→ P∗ and Id : P∗ −→ P∗ extend the
identity on A, and are therefore chain homotopic, via some s : P∗ −→ P∗+1 .
 EXT, TOR AND THE UCT 7

Applying the functor Hom(−, G), we obtain maps f ∗ : Hom(Q∗ , G) −→ Hom(P∗ , G),
g : Hom(P∗ , G) −→ Hom(Q∗ , G) and s∗ : Hom(P∗ , G) −→ Hom(P∗−1 , G), and the
∗

identity
f ∗ g ∗ − Id = s∗ δ + δs∗
holds where δ := d∗ , since Id∗ = Id. Thus the maps f ∗ and g ∗ induce isomorphisms
on homology groups. 

Proposition 4.3. The group Ext0 (A, G) is canonically isomorphic to Hom(A, G) :

Ext0 (A, G) ∼
= Hom(A, G), ∀ A, G

Proof. By left exactness, the sequence

0 −→ Hom(A, G) −→ Hom(P0 , G) −→ Hom(P1 , G)
is exact. Thus the zeroth homology group Ext0 (A, G), which is just the kernel of
the first map Hom(P0 , G) −→ Hom(P1 , G) in (6), is identified with Hom(A, G). 

Proposition 4.4. For any projective group P , all the higher Ext groups vanish:
Exti (P, G) ≡ 0, i≥1

Proof. The sequence

Id
0 −→ P −→ P −→ 0
is a projective resolution of P ! 

Proposition 4.5. If 0 −→ A −→ A0 −→ A00 −→ 0 is an exact sequence, there is a

long exact sequence
0 −→ Hom(A00 , G) −→ Hom(A0 , G) −→ Hom(A, G)
(8)
−→ Ext1 (A00 , G) −→ Ext1 (A0 , G) −→ Ext1 (A, G) −→ · · ·

Proof. Choose projective resolutions according to Lemma 3.2. Since the sequences
0 −→ Pi −→ Pi0 −→ Pi00 −→ 0 are split exact, it follows that
0 −→ Hom(P∗00 , G) −→ Hom(P∗0 , G) −→ Hom(P∗ , G) −→ 0
is a short exact sequence of complexes. (A priori it is only left exact, but then it
follows from the splitting that it is also exact on the right.) The sequence (8) then
follows from the usual long exact sequence in homology and Proposition 4.3. 

Finally, we have the following results, which are particular to the Hom(−, G)
functor and/or the category of abelian groups. The proof of the next Proposition
follows directly from Lemma 2.3.
Proposition 4.6. For any abelian groups A and G, the only nonzero Ext groups
are in degrees 0 and 1:
Exti (A, G) = 0, i ≥ 2.
8 CHRIS KOTTKE

For this reason, the groups Ext1 (A, G) are often just denoted Ext(A, G). Note
however that this construction can be applied to the setting of modules over a ring
R (generalizing abelian groups which are equivalent to modules over Z), and then
the groups ExtiR (M, N ), i ≥ 2 are nonzero in general.
Proposition 4.7. Exti ( α Aα , G) = i i
L L Q
α Ext (Aα , G) and Ext (A, α Gα ) =
Q i
α Ext (A, Gα )
Proof. These follow directly from the corresponding identities
M M
Hom( Aα , G) = Hom(Aα , G), and
α α
Y Y
Hom(A, Gα ) = Hom(A, Gα ).
α α
L
by and by taking direct sums of projectives to form a resolution for α Aα . 

A particularly important computation is the following.

Proposition 4.8. Ext1 (Zm , Z) = Zm . More generally, for any group G, Ext1 (Zm , G) =
G/mG.
Proof. Use the resolution
m
0 −→ Z −→ Z −→ Zm −→ 0.
Then Ext1 (Zm , Z) is the second cohomology group of
m
0 −→ Z ∼
= Hom(Z, Z) −→ Z ∼
= Hom(Z, Z) −→ 0
which is Zm . Similarly, since Hom(Z, G) = G, it follows that Ext1 (Zm , G) = G/mG.


Remark. The name Ext comes from “Extension.” We say X is an extension of A

by B if
0 −→ B −→ X −→ A −→ 0
is exact. Given A and B there is always the trivial extension X = A ⊕ B, corre-
sponding to the isomorphism class of the split exact sequence. It can be shown (see
[Wei95]) that isomorphism classes of extensions of A by B are in 1-1 correspondence
with elements of Ext1 (A, B), with the trivial extension corresponding to 0.

5. Ext as a covariant derived functor

There is another way to define the groups Ext1 (A, G), namely, as the derived
functors of the covariant functor
Hom(A, −) : G 7−→ Hom(A, G).
Proposition 5.1. Hom(A, −) is left exact. Thus if
0 −→ G −→ G0 −→ G00 −→ 0
is exact, then
0 −→ Hom(A, G) −→ Hom(A, G0 ) −→ Hom(A, G00 )
is exact.
 EXT, TOR AND THE UCT 9

We can perform almost the same procedure as before, though because of the
combination of covariance and left exactness, the procedure gets dualized. In par-
ticular, we must consider injective objects instead of projective ones.
Definition 5.2. An abelian group I is injective if for any homomorphism f :
G −→ I and injective homomorphism h : G −→ G0 , there exists a homomorphism
f 0 : G0 −→ I such that
I
∃ f0
f (9)
h
G0 G 0
commutes.
Given G ∈ AbGp, an injective resolution is an exact sequence
0 −→ G −→ I0 −→ I1 −→ · · ·
where the groups Ij are injective.
Taking an injective resolution G −→ I∗ for G and applying the functor Hom(A, −)
to the truncated complex 0 −→ I∗ results in the complex
0 −→ Hom(A, I0 ) −→ Hom(A, I1 ) −→ · · · .
Taking the cohomology of this sequence results in groups
Ker {Hom(A, Ii ) −→ Hom(A, Ii+1 )}
Exti (A, G) := (10)
Im {Hom(A, Ii−1 ) −→ Hom(A, Ii )}
which turn out to be the same as those defined by deriving the functor Hom(−, G).
Proposition 5.3. The groups defined by (10) coincide with those defined by (7).
Proof. See [Wei95], Theorem 2.7.6. 
Again, in the category of abelian groups it turns out that there is always a two
step injective resolution which gives another way to show that Exti (A, G) = 0 for
i ≥ 2.
Furthermore, dualizing the lemmas in Section 3 (whose proofs are the same
up to replacing “projective” by “injective” and reversing all arrows) shows that
Exti (A, G) is independent of the injective resolution of G, and we have the following
analogue of Proposition 4.5.
Proposition 5.4. From a short exact sequence
0 −→ G −→ G0 −→ G00 −→ 0
we obtain a long exact sequence
0 −→ Hom(A, G) −→ Hom(A, G0 ) −→ Hom(A, G00 )
−→ Ext1 (A, G) −→ Ext1 (A, G0 ) −→ Ext1 (A, G00 ) −→ 0

There is also the obvious analogue of Proposition 4.4.

Proposition 5.5. If I ∈ AbGp is injective, then
Exti (A, I) = 0, ∀A
10 CHRIS KOTTKE

6. Universal Coefficient Theorem for Cohomology

The universal coefficient theorem for cohomology quantifies

the difference be-
tween the groups H n (X, G) = H n
Hom(C∗ (X), G), δ and Hom(Hn (X), G) in
terms of the groups Ext Hn−1 (X), G . Observe that there is a natural map

h : H n (X, G) −→ Hom Hn (X), G ,

where h([f ])([c]) = f (c).

Here [f ] denotes the cohomology class of a cocycle f ∈ C ∗ (X, G) = Hom(C∗ (X), G),
and [c] denotes the homology class of a cycle c ∈ C∗ (X). The map does not depend
on the choices of representative: since δf = 0 and ∂c = 0 it follows that

h([δg])([c]) = δg(c) = g(∂c) = 0, and h([f ])([∂d]) = f (∂d) = δf (d) = 0.

We will prove a completely algebraic version of the universal coefficient theorem

first. Let (C∗ , ∂) be a chain complex of free groups, and denote by

Zn = {α ∈ Cn : ∂α = 0} , Bn = {∂β : β ∈ Cn+1 }

the cycles and boundaries of Cn , respectively. Additionally, we will use the short-
hand Hn to denote the homology group Hn (C∗ ), ∂).
We have short exact sequences

i ∂
0 −→ Zn −→ Cn −→ Bn−1 −→ 0 (11)

and
i q
0 −→ Bn −→ Zn −→ Hn −→ 0, (12)

and we note that the groups Bn and Zn are free, being subgroups of the free groups
Cn . From the theory of Ext groups, these lead to exact sequences

δ i∗
0 −→ Hom(Bn−1 , G) −→ Hom(Cn , G) −→ Hom(Zn , G) −→ 0 (13)

and

q∗ i∗
0 −→ Hom(Hn , G) −→ Hom(Zn , G) −→ Hom(Bn , G) −→ Ext(Hn , G) −→ 0 (14)

Note that since (11) is a sequence of free groups it splits, though not naturally with
respect to the boundary maps ∂ : Cn −→ Cn−1 .

Theorem 6.1. There are short exact sequences

j
h
0 −→ Ext(Hn−1 , G) −→ H n Hom(C∗ , G) −→ Hom(Hn , G) −→ 0 (15)

for each n, which are natural in G and C∗ , and which split (though not naturally
with respect to C∗ ).
 EXT, TOR AND THE UCT 11

Proof. The proof (taken from [Bre97]) involves a bunch of chasing around the some-
what complicated diagram

δ
Ext(Hn−1 , G) Hom(Cn+1 , G) Hom(Bn , G) 0
j
δ i∗
δ i∗
0 Hom(Bn−1 , G) Hom(Cn , G) Hom(Zn , G) 0

i∗ δ q∗
h
i∗
0 Hom(Zn−1 , G) Hom(Cn−1 , G) Hom(Hn , G)

A few notes about the diagram: The columns on the right and left are pieces of
the sequences (14) in degrees n − 1 and n, respectively. The middle column is just
the chain complex for cohomology. Finally, the horizontal maps are pieces of the
sequences (13) in degrees n + 1, n and n − 1, respectively, from top to bottom. The
dashed map is induced by a choice of splitting Cn −→ Zn of the sequence (11). The
dotted maps are the ones we are defining; they are not well-defined at the level of
cochains, but we’ll show they are well-defined on cohomology classes.
The map h which we defined earlier is equivalent to the following. It takes a
cocycle f ∈ Hom(Cn , G) in the middle of the diagram and goes [right, down] (since
δf = 0, [right, up] results in 0 so the resulting element of Hom(Zn , G) is in the
image of Hom(Hn , G)). If f = δg to begin with, then it comes from Hom(Bn−1 , G)
by commutativity (from g, going [up] is equivalent to [left, up, right]), and then
going [right] is equivalent to a path involving two consecutive steps in the middle
row, resulting in 0.
The map j taking Ext(Hn−1 , G) to a cohomology class in Hom(Cn , G) is defined
similarly, by lifting [down] and going [right]. The resulting element f ∈ Hom(Cn , G)
satisfies δf = 0 since δ [up] is equivalent to [right, up, left] and that involves two
steps in the middle row, giving 0. Similarly, the ambiguity in the initial lift [down]
comes from Hom(Zn−1 , G), which maps into the image of δ in Hom(Cn , G) and so
the image of j is well-defined in cohomology.
It is straightforward to see that composing j and h results in 0, as is injectivity
of j and surjectivity of h.
To see exactness in the middle of (15), note that any f ∈ Hom(Cn , G) which
goes to 0 by h must already vanish at Hom(Zn , G), hence it lies in the image of
Hom(Bn−1 , G) and comes from some element in Ext(Hn , G) by j.
Naturality with respect to G and C∗ will be left to the contemplation of the
reader.
12 CHRIS KOTTKE

Finally, the splitting of (15) is induced by the dashed map, which is natural with
respect to G but not with respect to the indexing by n (i.e. it is not natural with
respect to chain maps such as ∂ : C∗ −→ C∗−1 ). 
Applying Theorem 6.1 to the chain complex C∗ (X), we obtain what is usually
called the Universal Coefficient Theorem for Cohomology.
Corollary 6.2 (Universal Coefficient Theorem). There are short exact sequences
0 −→ Ext Hn−1 (X), G −→ H n (X; G) −→ Hom Hn (X), G −→ 0



(16)
for each n, which are natural in G and X, and which split (though not naturally
with respect to X). There are similar short exact sequences
0 −→ Ext Hn−1 (X, A), G −→ H n (X, A; G) −→ Hom Hn (X, A), G −→ 0

for a pair (X, A).

If a space X has finitely generated homology in all degrees, then we can decom-
pose its homology groups as
Hn (X) ∼ = Fn ⊕ Tn
L
where Fn = Z is a maximal finitely generated free abelian subgroup, and the
torsion Tn is a finite direct sum of finite cyclic groups. Combining this with Propo-
sition 4.8, we obtain the following
Proposition 6.3. If X has finitely generated homology, then
H n (X) ∼
= Fn ⊕ Tn−1 (17)
n n
where H (X) = H (X, Z) denotes cohomology with Z coefficients.
Proof. Since Hom(Zm , Z) = 0, Hom(Z, Z) ∼
= Z, Ext(Zm , Z) = Zm and Ext(Z, Z) =
0 it follows that
Hom Hn (X), Z ∼= Fn , Ext Hn−1 (X), Z ∼



= Tn−1 .
The isomorphism (17) then follows from the splitting of (16). 

7. Tensor product
We recall without proof some results about the tensor product.
Definition 7.1. If A, B ∈ AbGp, the group A ⊗ B is the group generated by
{a ⊗ b : a ∈ A, b ∈ B} subject to the relations
(a + a0 ) ⊗ b = a ⊗ b + a0 ⊗ b, and
a ⊗ (b + b0 ) = a ⊗ b + a ⊗ b0 .

It is characterized by the following universal property: if f : A × B −→ C is

a bilinear homomorphism (thus f (a + a0 , b) = f (a, b) + f (a0 , b) and f (a, b + b0 ) =
f (a, b) + f (a, b0 )) then f factors through a unique map fe : A ⊗ B −→ C.
Proposition 7.2. The tensor product satisfies the following properties.
(i) AL⊗B ∼= B ⊗ A. L
(ii) ( α Aα ) ⊗ B ∼
= α (Aα ⊗ B).
(iii) (A ⊗ B) ⊗ C ∼
= A ⊗ (B ⊗ C).
(iv) Z ⊗ B ∼= B.
 EXT, TOR AND THE UCT 13

(v) Zn ⊗ B ∼= B/nB.
(vi) Given homomorphisms f : A −→ C and g : B −→ C there is a natural
homomorphism f ⊗ g : A ⊗ B −→ C. (In particular A ⊗ − and − ⊗ B are
covariant functors.)

Remark. The natural setting for the tensor product is for modules M, N over a
ring R, generalizing abelian groups which are Z modules. The module M ⊗R N
is defined as in Definition 7.1 with the additional relation ra ⊗ b = a ⊗ rb for
r ∈ R. It is characterized by an analogous universal property. If M and N are
R-algebras (say for commutative R), then M ⊗R N is naturally an R algebra, and
the tensor product can be seen to be the direct product (coproduct) in the category
of R-algebras.

8. Tor
As with Hom and Ext, we begin with the failure of exactness for the functor
− ⊗ G.
Proposition 8.1. For fixed G ∈ AbGp and any short exact sequence
0 −→ A −→ A0 −→ A00 −→ 0
the resulting sequence
A ⊗ G −→ A0 ⊗ G −→ A00 ⊗ G −→ 0 (18)
is exact. In other words, the covariant functor − ⊗ G : AbGp −→ AbGp is only
right exact.
Proof. Exercise. 
The derived functors procedure again gives us a way to extend (18) leftward to
a long exact sequence. In this case the groups we obtain, called Tori (A, G) are
called the left derived functors of − ⊗ G, since they extend (18) to the left. The
combination of right exactness and covariance leads us to use projective groups
once again.
Thus, consider the functor − ⊗ G for fixed G, and let A ∈ AbGp be arbitrary.
Let
· · · −→ P1 −→ P0 −→ A −→ 0 (19)
be a projective resolution. We apply − ⊗ G to the truncated sequence · · · −→
P1 −→ P0 −→ 0 to obtain the complex
· · · −→ P1 ⊗ G −→ P0 ⊗ G −→ 0 (20)
i
Definition 8.2. The group Tor (A, G) is the ith homology group of (20):
Ker {Pi ⊗ G −→ Pi−1 ⊗ G}
Tori (A, G) :=
Im {Pi+1 ⊗ G −→ Pi ⊗ G}

Recalling the properties of Exti (A, G) proved in Section 4, we have a similar

package of results about Tori (A, G).
Proposition 8.3. The groups Tori (A, G) satisfy the following.
(i) Tori (A, G) is independent of the choice of projective resolution (19).
(ii) Tor0 (A, G) ∼
= A ⊗ G.
14 CHRIS KOTTKE

(iii) If P is a projective group (in particular if P is free), then Tori (P, G) = 0 for
i ≥ 1.
(iv) For any abelian group A, Tori (A, G) = 0 for i ≥ 2.
(v) If 0 → A → A0 → A00 → 0 is a SES, then there is a long exact sequence

0 −→ Tor1 (A, G) −→ Tor1 (A0 , G) −→ Tor1 (A00 , G)

(21)
−→ A ⊗ G −→ A0 ⊗ G −→ A00 ⊗ G −→ 0

(vi) Tori ( α Aα , G) = α Tori (Aα , G) and

L L

The proof is similar to those in Section (4) and will be left to the reader.
We mention the following analogue of Proposition 5.3.

Proposition 8.4. The groups Tori (A, G) are the same as the left derived functors
of A ⊗ − (obtained by taking the homology of A ⊗ P∗ −→ 0 where P∗ −→ G is a
projective resolution). In other words, Tori (A, G) is symmetric:

Tori (A, G) ∼
= Tori (G, A).

Proof. See [Wei95], Theorem 2.7.2. 

Finally some computations.

Proposition 8.5. Tor0 (Zm , A) = Zm ⊗ A = A/mA and Tor1 (Zm , A) = mA :=

{a ∈ A : ma = 0} .

Proof. Using the resolution

m
0 −→ Z −→ Z −→ Zm −→ 0

it follows that Tori (Zm , A) are the homology groups of

m
= A −→ Z ⊗ A ∼
0 −→ Z ⊗ A ∼ = A −→ 0

Remark. This hints at the reason for the name Tor, which stands for “Torsion.”
In fact it is possible to show by taking direct limits (see [Wei95]) that for any
A ∈ AbGp, Tor1 (Q/Z, A) is the torsion subgroup of A.

Definition 8.6. An abelian group G is called flat if − ⊗ G is exact.

Evidently projective groups are flat (though the converse is not true), and if G
is flat then Tor1 (A, G) = 0 for all A. (In fact G is flat if and only if Tor1 (A, G) = 0
for all A which is easily seen using symmetry of Tor.) We recall the following fact
from algebra.

Proposition 8.7. Q is flat. More generally, any localization ring S −1 Z is flat.

Corollary 8.8. Tor1 (Q, A) = 0 for all A.

 EXT, TOR AND THE UCT 15

9. Universal Coefficient Theorem for Homology

We briefly recall the construction of singular homology with coefficients in a
group G.

Definition 9.1. The group of singular n-chains on X with G coefficients is the

group

( )

nα σα  σα : ∆n −→ X, nα ∈ G ∼
X
Cn (X; G) = = Cn (X) ⊗ G

α

It is evidently a direct sum over {σ : ∆n −→ X} of copies of G. It follows that

Cn (X, A; G) := Cn (X; G)/Cn (A; G) ∼

= Cn (X, A) ⊗ G.

The singular homology groups with coefficients in G are the groups

Hn (X; G) := Hn C∗ (X) ⊗ G , Hn (X, A; G) = Hn (C∗ (X, A) ⊗ G)

In analogy to the UCT for cohomology, the Universal Coefficient Theorem for Ho-
mology relates the groups Hn (X; G) to the groups Hn (X)⊗G and Tor1 (Hn−1 (X), G).
As before, we prove a completely algebraic version first.

Theorem 9.2. Let C∗ be a chain complex of free groups, and let Hn denote its nth
homology group. There are short exact sequences

h j
0 −→ Hn ⊗ G −→ Hn (C∗ ⊗ G) −→ Tor1 (Hn−1 , G) −→ 0

which are natural in G and C∗ and which split (though not naturally with respect
to C∗ .)

Proof. From the short exact sequence (11) we obtain the sequence

i⊗1 ∂⊗1
0 −→ Zn ⊗ G −→ Cn ⊗ G −→ Bn−1 ⊗ G −→ 0

which splits, and the (12) gives rise to the sequence

i⊗1 q⊗1
0 −→ Tor1 (Hn , G) −→ Bn ⊗ G −→ Zn ⊗ G −→ Hn ⊗ G −→ 0

using the fact that Zn is free.

16 CHRIS KOTTKE

The proof follows from a diagram chase analogous to the one in the proof of
Theorem 6.1, using the diagram

∂
0 Bn ⊗ G Cn+1 ⊗ G Tor1 (Hn−1 , G)
j
i ∂
i ∂
0 Zn ⊗ G Cn ⊗ G Bn−1 ⊗ G 0
q ∂ i
h
i
Hn ⊗ G Cn−1 ⊗ G Zn−1 ⊗ G 0

0

Corollary 9.3 (Universal coefficient theorem for homology). There are short exact
sequences
0 −→ Hn (X, A) ⊗ G −→ Hn (X, A; G) −→ Tor1 Hn−1 (X, A), G −→ 0

which are natural in G and (X, A) and which split (though not naturally with respect
to (X, A).)

References
[Bre97] G.E. Bredon, Topology and geometry, Graduate Texts in Mathematics, vol. 139, Springer,
1997.
[Wei95] C.A. Weibel, An introduction to homological algebra, Studies in Advanced Mathematics,
vol. 38, Cambridge University Press, 1995.

Brown University, Department of Mathematics

E-mail address: ckottke@math.brown.edu