HOMOLOGY

1. Singular homology
Let us start with heuristic definition of one-dimensional homology of topological space X.
We say that a one-dimensional cycle in X is homologically trivial (or that it is homologous
to zero) if it can be considered as a boundary of a surface. (Notice that a curve and a surface
can be regarded as linear combinations of several pieces; hence in more precise terminology
we should talk about one-dimensional and two-dimensional chains.) Two cycles belong
to the same homology class if their difference is homologically trivial. To calculate one-
dimensional homology we should describe all one-dimensional homology classes. To define
k-dimensional homology group Hk (X) we should factorize the group of k-dimensional cycles
Zk (X) with respect to its subgroup Bk (X) consisting of boundaries. For example, for the
sphere S n we have homologically non-trivial cycles only in dimensions 0 (points) and in
dimension n (the sphere itself can be considered as n-dimensional cycle).
To give a rigorous definition of homology we should define the group Ck (X) of k-
dimensional
P chains together with boundary operator ∂ acting in the direct sum C• (X) =
Ck (X) and transforming Ck (X) into Ck−1 (X); the boundary operator restricted to
Ck (X) will be denoted by ∂k . We require that ∂ 2 = 0 hence Im∂ ⊂ Ker∂. The elements
of Ker∂ = Z(X) are called cycles, the elements of Im∂ = B(X) are called boundaries,
the factor-groupPH(X) = Z(X)/B(X)
P is called homology group of X. It is easy to check
that H(X) = Hk (X) = Zk (X)/Bk (X). Here Zk (X) = Ker∂k is the group of k-
dimensional cycles, Bk (X) = Im∂k+1 is the group of k-dimensional boundaries, Hk (X)
stands for k-dimensional homology.
There exist various constructions of the group of chains leading to the same homology
groups. Our construction is based on consideration of cubic singular chains; the homology
we define is called singular homology. We define a k-dimensional singular cube in X as
a continuous map φ : I k → X where I = [0; 1]. In other words, a singular cube is a
k-dimensional surface in X parametrized by k numbers t1 , ..., tk where 0 ≤ ti ≤ 1. We say
that a singular cube lies in the set V ⊂ X if φ(I k ) ⊂ V.
A k-dimensional singular cubic chain (k-chain) is defined as linear combination of k-
dimensional singular cubes with integer coefficients; the set of linear combinations of this
kind will be denoted by C̃k (X). (Instead of integer coefficients we can use coefficients from
abelian group G ; then we obtain a definition of homology group with coefficients in G.
If the group of coefficients is not specified explicitly talking about linear combinations we
always have in mind combinations with integer coefficients.)1 We say that a singular cube
φ : I k → X is degenerated as if φ(t1 , ..., tk ) does not depend on one of the arguments.
1We say that a chain consists of cubes entering the linear combination specifying the chain with non-zero
coefficients. A chain lies in the set V if it consists of cubes lying in V .
1
2 HOMOLOGY

To define the group Ck (X) we factorize C̃k (X) with respect to the subgroup generated by
degenerate singular cubes. The (geometric) boundary of the cube I k is a union of 2k cubes
Ij± of dimension k − 1. The cube Ij− is singled out by the condition tj = 0, the cube Ij+
by the condition tj = 1. The coordinates on these cubes are t1 , ..., tk−1 , tk+1 , ..., tn (in this
order). We define the boundary of singular cube φ as a chain
X
−
∂φ = (−1)j−1 (φ+j − φj )

where the singular cube φ± ±

j is obtained as a restriction of φ to Ij . Extending the definition
of boundary by linearity we obtain the operator ∂ on C̃k (X) ; it is easy to check that it
descends to 2
PCk (X) and obeys ∂ = 0. This means that it can be used to define the homology
H(X) = Hk (X) (singular
P cubic homology). The operator ∂ can be considered as an
operator on C(X) = Ck (X); if we want to emphasize that we consider ∂ restricted to
Ck (X) we use the notation ∂k .
Let us consider some simple examples. Notice first of all that zero-dimensional chain
(a linear combination of points) is a cycle. Two points that can be connected by a path
are homologous cycles. We obtain that the group H0 (X) of a connected space is Z. For
a set with N connected component a zero-dimensional homology class corresponds to a
row of N integers i1 , ..., iN , i.e. the group H0 (X) is a direct sum of N copies of Z.2 (In
general, the group Hk (X) is a direct sum of k-dimensional homology groups of connected
components of X.)
Every closed path in X specifies a one-dimensional cycle, homotopic closed paths spec-
ify homologous cycles. (The homotopy is a map of I 2 ; we consider this map as a two-
dimensional chain. The boundary of this chain is a difference of cycles corresponding to
paths.) We obtain a homomorphism of fundamental group π1 (X, ∗) into one-dimensional
homology H1 (X); if the space X is connected this homomorphism is surjective. The funda-
mental group is in general non-commutative, but the homology is commutative. This means
that for connected space we obtain a surjective homomorphism π1 (X, ∗)/[π1 (X, ∗), π1 (X, ∗)] →
H1 (X); we will prove later that this homomorphism is an isomorphism.
If X is a one-point set , then H0 (X) = Z,Hk (X) = 0 for k > 0 (all singular cubes are
degenerate).
It is obvious that a continuous map f : X → Y induces a homomorphism f∗ : Ck (X) →
Ck (Y ) commuting with boundary operator ∂. This operator descends to homology and
specifies a homomorphism f∗ : Hk (X) → Hk (Y ) (we use the same symbol for the homo-
morphism on homology). It is easy to check that (f g)∗ = f∗ g∗ and id∗ = id (in other
words k-dimensional homology specifies a functor from the category of topological spaces
into the category of abelian groups).

Theorem 1. Two homotopic maps specify the same homomorphism of homology groups.

2Sometimes it is convenient to consider so called reduced homology group H̃ (X) of non-empty space
k
X. For k = 0 it is a subgroup of H0 (X) singled out by relation i1 + ....iN = 0. For k > 0 the reduced
homology group is the same as the conventional one: H̃k (X) = Hk (X).
 HOMOLOGY 3

To prove this theorem we consider two maps f0 : X → Y and f1 : X → Y connected by

the homotopy fτ : X → Y . The statement follows from the remark that for any cycle ω
the cycles (f0 )∗ ω and (f1 )∗ ω are homologous; the homology is given by the surface covered
by the image of ω by the deformation fτ . To give more formal proof we construct for every
k-dimensional singular cube φ : I k → X a (k + 1)-dimensional singular cube ψ : I k+1 → Y
defined by the formula ψ(τ, t) = fτ (φ(t)). (Here a point of I k+1 is represented as a pair
(τ, t) where τ ∈ I, t ∈ I k .) This construction induces a map H : Ck (X) → Ck+1 (Y ). It is
easy to check that for every chain ω we have
∂Hω = (f1 )∗ ω − (f0 )∗ ω − H∂ω.
It follows immediately from this formula that for any cycle ω the cycles (f0 )∗ ω and (f1 )∗ ω
belong to the same homology class.
Corollary 2. If the map f : X → Y is a homotopy equivalence the map f∗ : Hk (X) →
Hk (Y ) is an isomorphism.
It follows from this statement that every contractible set ( a set homotopy equivalent to
a point) has the same homology as a point.

2. Cellular homology
It is not easy to calculate singular homology using the definition given in preceding
section. In this section we will give a definition of cellular homology of cell complex. It
is not clear from the definition that cellular homology is a topological invariant, but later
we will prove that it coincides with singular homology and therefore it is invariant with
respect to homeomorphisms. We will give some examples when using cellular homology
gives us an easy way to calculate the homology of topological space.
Recall that a k-dimensional cell σ k of cell complex X is pasted to (k − 1)-dimensional
skeleton X k−1 of complex X by means of a map f : S k−1 → X k−1 . Taking a composition
of this map with the identification map p : X k−1 → X k−1 /X k−2 we obtain a map pf :
S k−1 → X k−1 /X k−2 . The space X k−1 /X k−2 is a wedge sum of αk−1 spheres of dimension
k − 1 corresponding to (k − 1)-dimensional cells of X. (Here αk stands for the number
of k-dimensional cells in cell complex X. ) This means that the homotopy class of pf
is characterized by integer numbers m1 , ..., mi , ... where 1 ≤ i ≤ αk−1 . More precisely,
these numbers are well defined only if all spheres we consider are oriented (= all cells
are oriented). We define the boundary of the oriented k-cell σ k as a linear combination
P i k−1
m σi of (k − 1)-dimensional cells.
We define k -dimensional cellular chain (k-chain) as a linear combination of oriented k-
dimensional cells; the group of cellular k-chains will be denoted by Ckcell (X). The boundary
operator defined on cells extends to chains by linearity, it acts from Ckcell into Ck−1cell . It

can be extended also to an operator ∂ acting in C cell = Ckcell and obeying ∂ 2 = 0. (We
P
denote the the boundary operator on cellular chains by the same symbol as for singular
chains; hope this will not lead to confusion.) As usual we define the cellular P homology
as Ker∂/Im∂. We will use for cellular homology the same notation H(X) = Hk (X)
4 HOMOLOGY

as for singular homology anticipating the theorem about isomorphism between singular
homology and cellular homology.
As we have mentioned already one can modify the definition of homology considering
chains as linear combinations with coefficients in abelian group G (instead of integer coeffi-
cients). This modification leads to the definition of homology groups Hk (X; G) (homology
with coefficients in G).
Examples.
1. The sphere S n has a cell decomposition with one n-cell σ n and one 0-cell σ 0 . Both
cells are cycles, hence Hn (S n ) = H0 (S n ) = Z. All other homology groups vanish.
Considering homology with coefficients in G we obtain Hn (S n ; G) = H0 (S n ; G) =
G, Hk (S n ; G) = 0 for k 6= 0, n.
We can calculate the homology of the sphere taking any other cell decomposition. For
example, for S 2 we can take northern and southern hemispheres as two-dimensional cells
σ12 and σ22 and represent the equator as a union of two-semicircles σ11 ,σ21 and two points
σ10 ,σ20 . (More generally, for S n we can construct a cell decomposition having two cells in
any dimension ≤ n.) For appropriate choice of orientations we have ∂σ12 =∂σ22 =σ11 − σ21 ,
∂σ11 = ∂σ21 = σ10 − σ20 . We see that all two-dimensional cycles are of the form m(σ12 − σ22 )
where m ∈ Z, all one-dimensional cycles are boundaries.
2. Real projective space RP n has a cell decomposition with one cell σ k in every dimension
≤ n. To construct this decomposition we notice that RP n−1 is naturally embedded into
RP n and σ n = RP n \ RP n−1 is a cell. Hence by induction we can construct a cell complex
where the k-dimensional skeleton is RP k . We can obtain RP n identifying the points x
and −x on the sphere S n , then the same cell decomposition of RP n can be obtained from
a cell decomposition of S n having two cells in every dimension ≤ n. One can check that
∂σ 2m+1 = 0,∂σ 2m = 2σ 2m−1 for m > 0. (The simplest way to prove this is to start with the
decomposition of the sphere and to notice that the map x → −x preserves orientation of
odd-dimensional sphere and changes the orientation of even-dimensional sphere.) It follows
that odd-dimensional cells are cycles, the odd-dimensional homology groups of dimension
< n are isomorphic to Z2 , but Hn (RP n ) = Z for odd n. Even-dimensional homology vanish
(except H0 (RP n ) = Z).
To calculate the homology with coefficients in the group G we notice that Ckcell (RP n ) = G
for all k ≤ n. The boundary operator is does not vanish only for even k > 0 and transforms
g ∈ G into 2g. We obtain H2k+1 (RP n ; G) = G/2G for 2k + 1 < n , Hn (RP n ; G) = G if
n is odd, H2k (RP n ; G) = {g ∈ G|2g = 0} if 0 < 2k ≤ n. In particular, if G = Z2 we have
Hm (RP n , Z2 ) = Z2 for 0 ≤ m ≤ n.
3.Complex projective space CP n has a decomposition with one cell σ 2k in every even
dimension ≤ 2n. As in the real case CP n−1 is naturally embedded into CP n and σ 2n =
CP n \ CP n−1 is a 2n-cell. By dimensional reasons all cells are cycles. Even- dimensional
homology H2k (CP n ) = Z (or, more generally, H2k (CP n ; G) = G for k ≤ n), all other
homology groups vanish.
4. Two-dimensional surfaces.
 HOMOLOGY 5

The standard representation of sphere with g handles as 4g-gon gives a cell complex
with one 2-cell, 2g one-dimensional cells and one 0-cell. All cells are cycles, hence H2 =
Z, H1 = Z2g , H0 = Z.
A sphere with g handles and h Möbius bands can be represented as a polygon with
4g sides corresponding to handles and 2h sides corresponding to Möbius bands. We ob-
tain a cell complex with one 2-cell σ 2 , one 0-cell σ 0 and 2g + h one-dimensional cells
σ11 , ..., σ2g
1 , τ 1 , ..., τ 1 . The only non-trivial boundary is ∂σ 2 = 2(τ 1 + ... + τ 1 ). We obtain
1 h 1 h
that H2 = 0, H0 = Z and H1 is a direct sum of Z2 and 2g + h − 1 copies of of Z. This
agrees with the fact that the topological type of non-orientable surface depends only on
2g + h (if h > 2 one can replace two Möbius bands by one handle).
If we are interested in homology with coefficients in the group Z2 we obtain that H2 =
H0 = Z2 , and H1 is a direct sum of 2g + h copies of Z2 .

3. Relative singular homology

Let us consider a topological space X and its non-empty subset A. Then every chain
in A can be considered as a chain in X; we have an inclusion Ck (A) ⊂ Ck (X). We define
relative chains as elements of factor-group Ck (X)/Ck (A). The boundary operator ∂ de-
scends to Ck (X)/Ck (A); we define relative homology Hk (X, A) in terms of this operator.
In other words, we define the group of relative cycles Zk (X, A) as the kernel of boundary
homomorphism ∂k : Ck (X)/Ck (A) → Ck−1 (X)/Ck−1 (A) and the group of relative bound-
aries Bk (X, A) as the image of boundary homomorphism ∂k+1 : Ck+1 (X)/Ck+1 (A) →
Ck (X)/Ck (A); the homology Hk (X, A) is defined as a factor-group Zk (X, A)/Bk (X, A)
One can say that calculating relative homology we neglect everything in A. For example,
a relative cycle is a chain with boundary in A, relative cycle is homologous to zero if there
exists a chain with a boundary equal to a sum of this cycle and a chain lying in A.
It is clear that a continuous map f : X → Y obeying f (A) ⊂ B (map of pairs
f : (X, A) → (Y, B)) generates a homomorphism f∗ : Hk (X, A) → Hk (Y, B) (relative
homology is a functor from the category of pairs into the category of groups).
Theorem 3. If two maps of pairs f0 : (X, A) → (Y, B) and f1 : (X, A) → (Y, B) are
homotopic as maps of pairs (i.e. they can be connected by continuous family fτ : (X, A) →
(Y, B) of maps of pairs) the corresponding homomorphisms of homology groups coincide.
The proof repeats the proof of corresponding statement for absolute homology.
It is obvious that there exists a natural map j∗ : Hk (X) → Hk (X, A) (every absolute
cycle is a relative cycle). This map is induced by the map of pairs j : (X, ∅) → (X, A).
A boundary of a relative cycle is an absolute cycle in A; using this remark we construct
a homomorphism ∂ : Hk (X, A) → Hk−1 (A) (boundary homomorphism). The inclusion
i : A → X induces a homomorphism i∗ : Hk (A) → Hk (X). We see that one can consider a
sequence of groups and homomorphisms
(1) ... → Hk (A) → Hk (X) → Hk (X, A) → Hk−1 (A) → ...
Theorem 4. The sequence (1) is exact.
6 HOMOLOGY

We will skip a straightforward proof of this theorem, because later we will prove more
general statement.
The sequence (1) (exact homology sequence) can be used to calculate relative homology
if we know H(X) and H(A). For example, if X is connected and A is contractible we
obtain that Hk (X, A) = H̃k (X) where H̃k (X) = Hk (X) for k > 0 and H̃0 (X) = 0. The
groups H̃k (X) are reduced homology groups of connected space X.
It is important to emphasize that in good situations (for example, if A is a subcomplex
of connected cell complex X) the exact homology sequence can be rewritten in terms of
absolute homology in the following way:
(2) ... → Hk (A) → Hk (X) → H̃k (X/A) → Hk−1 (A) → ...
Here X/A stands for the space X where all points of A are identified.3
To prove (2) we need the following statement (excision theorem).
Theorem 5. Let us suppose that Z ⊂ A ⊂ X and there exists an open set U ⊂ A
containing the closure of Z. Then the homomorphism i∗ : Hk (X \ Z, A \ Z) → Hk (X, A)
induced by the inclusion i : (X \ Z, A \ Z) → (X, A) is an isomorphism.
The proof of this theorem is based on a lemma stating that calculating homology we can
work with chains containing only small cubes. To formalize this statement we introduce
a homomorphism τ : Ck (X) → Ck (X) called subdivision map. To define it we represent
the interval I as a union of its left half and right half. Taking direct product we obtain
the decomposition of I k into a union of 2k smaller cubes. Using this decomposition we
assign to every singular cube a sum of 2k singular cubes. Extending this construction to
cubic chains by additivity we obtain a homomorphism τkX : Ck (X) → Ck (X) (subdivision
map). We will denote this homomorphism by τ omitting k and X if this does not lead to
a confusion.
Lemma 6. There exists a homomorphism HkX : Ck (X) → Ck+1 (X) obeying
X
(4) Hk−1 ∂k + ∂k+1 HkX = τkX − 1.
To prove this lemma we notice first of all that τ commutes with ∂ and f∗ and τ0X =
1 (only these properties will be used in the proof). We take H0X = 0 and use (4) to
define HkX for k > 0 by induction. Rewriting (4) in the form ∂k+1 HkX c = σkX c where
X ∂ c we notice that ∂ σ X c = 0 hence σ X c is a cycle. We impose an
σkX c = τkX c − c − Hk−1 k k k k
additional condition that H commutes with f∗ ; then for arbitrary singular cube φ : I k → X
k
considered as singular chain in X we have HkX (φ) = φ∗ HkI c where c is the identity map
I k → I k considered as an element of Ck (I k ). We see that it is sufficient to construct HkX
for X = I k .The cube I k is contractible, its homology is same as homology of a point, hence
k k
σkI c is a boundary. This means that there exists b ∈ Ck+1 I k obeying ∂k+1 b = σkI c. We
3Sometimes it is convenient to use an equivalent exact homology sequence for reduced homology groups:

(3) ... → H̃k (A) → H̃k (X) → H̃k (X/A) → H̃k−1 (A) → ...
 HOMOLOGY 7

k
can take HkI c = b. It is easy to check that the formula HkX (φ) = φ∗ b (extended to all
chains by linearity) obeys (4).
It follows from (4) that the subdivision map τ does not change the homology class of a
cycle. Applying this map many times we see that any homology class can be represented
by a chain consisting of small cycles.
Let us start the proof of the theorem from the statement that the homomorphism i∗ is
surjective. We take an element of Hk (X, A) . This element can be represented by a chain ω
in X having boundary in A. Instead of ω we can represent this element by the chain τ N ω
consisting of smaller singular cubes . For N >> 0 the singular cubes in this chain cannot
have common points simultaneously Z and X \ A. Then deleting from τ N ω singular cubes
that lie in A we obtain a relative cycle that specifies the same element of Hk (X, A) and
lies in X \ Z. It determines an element of Hk (X \ Z, A \ Z) that maps into homology class
of ω in Hk (X, A).
To prove injectivity of i∗ we use very similar considerations. If a homology class belongs
to the kernel of this map a representative σ of this class is bounded by a chain in Ck+1 (X, A)
(more precisely, there exists a chain ρ ∈ C̃k+1 (X) obeying ∂ρ = σ+chain in A). Applying
the operator τ N where N >> 0 to the chain ρ and deleting singular cubes lying in A from
the chain τ N ρ we obtain a chain lying in X \ Z. The boundary of this chain is equal to
τ N σ+ a chain lying in A \ Z. Hence the homology class we started with vanishes.
Using the excision theorem one can express relative homology S as absolute homology of
some other space. Namely, we should consider the space X CA. Here CA stands for the
cone over A, i.e. for the product A×[0, 1] where all points of the form (a, 1) are considered as
one point denoted by ∗ (the tip of the cone). In the union of X and CA we identify a point
a ∈ A ⊂ X with the point (a, 0) ∈ CA. The cone CA is contractible, therefore for connected
S S
space X we have Hk (X CA, CA) = S H̃k (X CA). From S the other side we can use the
excision theorem to say that Hk (X CA, CA) = Hk (X CA \ ∗, CA \ ∗) = Hk (X, A).
(We have used that CA \ ∗ = A × [0, 1) is homotopically equivalent to A.)
We have proved that [
Hk (X, A) = H̃k (X CA).
S
The cone CA is contractible, therefore in good situations X CA is homotopy equivalent
S 4
to X CA/CA = X/A. We obtain that the exact sequence (1) can be rewritten in the
form (2).

4. Singular homology of cell complexes

Let us calculate first of all the singular homology n
n n 0
L of the sphere S . We will prove that
Hn (S ) = Z, H0 (S ) = Z for n > 0, H0 (S ) = Z Z, all other homology groups vanish.
Notice first of all that the answer for H0 is obvious (the sphere S n is connected for n > 0
and has two components for n = 0). Moreover, it is clear that Hk (S 0 ) = 0 for k > 0.
(The sphere S 0 consists of two points.) To calculate other singular homology groups we
4It is sufficient to assume that A is a subcomplex of cell complex X or, more generally, that A is an
S
NDR of X, hence CA is an NDR of X CA; see Sec....
8 HOMOLOGY

use the exact homology sequence in the form (2) for the pair (Dn , S n−1 ) where Dn is the
closed n-dimensional ball and S n−1 stands for its boundary. Taking into account that
Dn /S n−1 = S n we obtain exact sequence
... → Hk (S n−1 ) → Hk (Dn ) → H̃k (S n ) → Hk−1 (S n−1 ) → Hk−1 (Dn ) → ....
The ball Dn is contractible, therefore for k > 1 we obtain that Hk (S n ) = Hk−1 (S n−1 ) and
for k = 1 we see that H1 (S n ) is the kernel of the homomorphism of H0 (S n−1 ) → H0 (Dn ) =
Z. This permits us to calculate H1 (S n ) and by induction all other homology groups.5
A representative of the homology class generating Hn (S n ) = Z for n > 0 can be chosen
as a singular cube φ : I n → S n that maps the boundary of the cube into one point and
gives a homeomorphism of the interior of the cube and the complement of this point .
Notice that the generator of Z is defined up to a sign; the choice of this sign corresponds
to the choice of the orientation of the sphere.
Now we can calculate the singular homology of a wedge sum S n ∨ .... ∨ S n (bouquet) of
N spheres S n . In general, the wedge sum of N spaces can be obtained from their disjoint
union by means of identifying marked points in these spaces. The homology of disjoint
union can be represented as a direct sum of homology of summands; it follows from exact
homology sequence that the same is true for k-dimensional homology of wedge sum if k > 0
. We obtain that Hn (S n ∨ .... ∨ S n ) is a direct sum of N copies of Z, all other homology
groups vanish except H0 (S n ∨ .... ∨ S n ) = Z. Notice that our conclusions remain correct if
N = ∞.
Now we can calculate the singular homology of cell complex X by induction. (Talking
about singular homology of cell complex we have in mind the singular homology of corre-
sponding topological space -polyhedron. It would be more precise to use different notations
for cell complex and polyhedron. ) We will use the fact the X k /X k−1 is a wedge sum of
αk spheres of dimension k where αk is the number of k-dimensional cells. (Here X k stands
for k-dimensional skeleton of X.) The exact homology sequence (2) of the pair (X k , X k−1 )
takes the form
(5)
... → Hn (X k−1 ) → Hn (X k ) → H̃n (X k /X k−1 ) → Hn−1 (X k−1 ) → Hn−1 (X k ) → H̃n−1 (X k /X k−1 ) → ...
We obtain that Hn (X k ) = Hn (X k−1 ) for n < k − 1 or n > k. It follows by induction that
Hn (X k ) = 0 if n > k. Taking n = k in (5) we see that
Hk−1 (X k ) = Hk−1 (X k−1 )/ImH̃k (X k /X k−1 ).
(All (k − 1)-dimensional cohomology classes in X k come from X k−1 , but some classes
coming from X k−1 are bounded by k-dimensional cells.) For n = k + 1 we obtain from (5)
that
Hk (X k ) = Ker(H̃k (X k /X k−1 ) → Hk−1 (X k−1 )).
(All k-dimensional singular homology classes in X k come from linear combinations of k-
dimensional cells.)
5We could use the exact homology sequence in the form (3) to obtain the relation H̃ (S n ) = H̃ n−1
k k−1 (S )
valid for all k ≥ 0.
 HOMOLOGY 9

It is clear from our calculations that all homology classes in X come from linear com-
binations of cells. To give a precise meaning to this vague statement we use the notion of
cellular homology and prove that cellular homology coincides with singular homology.
We define a cellular k-chain as a linear combination of oriented k-dimensional cells.
(To orient a cell in X k \ X k−1 we should orient the corresponding sphere in X k /X k−1 .)
It follows from the above considerations that the group of cellular k-chains Ckcell (X)
can be identified with Hk (X k , X k−1 ) = Hk (X k /X k−1 ) for k > 0 and with H0 (X 0 ) for
k = 0. We define a boundary operator ∂k : Ckcell (X) → Ck−1 cell (X) as a composition
cell
of the boundary operator Ck (X) = Hk (X , X k k−1 ) → Hk−1 (X k−1 ) and the homo-
morphism Hk−1 (X k−1 ) → Hk−1 (X k−1 , X k−2 ) = Ck−1 cell (X). We obtain an operator ∂ on

C cell (X) = Ckcell (X) that obeys ∂ 2 = 0. (Notice that in Sec ...we gave different def-
P
inition of operator ∂; these definitions are equivalent.) We define the cellular homology
H cell (X) = Hkcell (X) by means of the operator ∂.
P

Theorem 7. The cellular homology of cell complex X is isomorphic to singular homology

of corresponding topological space : Hkcell (X) = Hk (X).
To prove this theorem we notice first of all that one can give also a definition of relative
cellular homology of a pair (X, A) where A is a subcomplex of cell complex X. (We should
consider the boundary operator on C cell (X)/C cell (A).) Moreover, one can construct the
exact homology sequence of a pair in the same way as for singular homology. In particular,
one can write the exact sequence (5) replacing all singular homology group by their cellular
counterparts. Assuming that we have proved already that (k − 1)-dimensional cellular
and singular homology are isomorphic and taking into account that the cellular homology
of X k /X k−1 coincide with singular homology of this space we obtain the isomorphism
between k-dimensional cellular and singular homology. Hence we can prove the theorem
by induction.

5. Cohomology
Let us start with a sequence of abelian groups Cn and homomorphisms ∂n : Cn → P Cn−1
obeying ∂n−1 ∂n = 0. Equivalently we can consider a graded abelian group C = Cn
and an operator ∂ transforming Cn into Cn−1 and obeying ∂ 2 = 0. We can say that
these data specify a chain complex or,
Pin different terminology, a differential graded group.
We define the homology H(C) = Hn (C) as Ker∂/Im∂. In other words, Hn (C) =
Zn (C)/Bn (C) where Zn (C) = Ker∂n is the group of cycles and Bn (C) = Im∂n+1 is the
group of boundaries.
P Let us fix now an abelian group G. Then we construct P homology groups H(C; G) =
Hn (C; G) as homology of the chain complex C ⊗ G = Cn ⊗ G. (We use the fact
that ∂ induces an operator C ⊗ G → C ⊗ G obeying ∂ 2 = 0 (we use the same notation
for the induced operator). These homology groups are called homology groups of C with
coefficients in G. If C is the group of singular chains of topological space X or cellular
chains of cell complex X this construction leads to homology of X with coefficients in G.
(If C is a free abelian group with free system of generators ei then the elements of C are
10 HOMOLOGY

linear combinations of generators ei with integer coefficients and the elements of C ⊗ G

can be considered as linear combinations of generators with coefficients in G.)
We can define also cohomology groups of C with coefficients in G in the following way.
We consider the groups C n (G) = Hom(Cn ; G) as groups of homomorphisms of Cn into
G (elements of these groups are called G-valued cochains) . Notice that the boundary
operator ∂n : Cn → Cn−1 induces a homomorphism ∇n : C n−1 (G) → C n (G) (coboundary
homomorphism) . (If ψ : Cn−1 → G is a homomorphism then ∇n (ψ) : Cn → G is a
composition of homomorphisms ∂ and ψ. If G is a field (more precisely, a field considered
as abelian group with respect to addition) then C n (G) is a vector space dual to vector
space Cn ⊗ G and ∇n is the operator adjoint to ∂n . If Cn = Cncell (X) then C n (G) is the
group of G-valued functions on the set of n-dimensional cells. In particular, every cell σ
specifies a Z-valued cochain taking value 1 on σ and 0 on all other cells. The coboundary
of this cochain does not vanish only on cells having σ in their boundary.
It follows immediately from the definition that ∇n ∇n−1 = 0. This relation allows us
to define cohomology groups of chain complex C with coefficients in G by the formula
H n (C) = Z n (G)/B n (G), where Z n (G) is the group of cocycles (the group of n-dimensional
cochains ω obeying ∇n+1 ω = 0) and B n (G) is the group of coboundaries (the group of
n-dimensional cochains of the form ω = ∇n σ).
Applying this definition to the chains in topological space or in cell complex we obtain
a definition of cohomology of topological space:
H n (X) = Z n (X; G)/B n (X; G),
where Z n (X; G) stands for the group Ker∇n+1 of n-dimensional cocycles in X and B n (X; G)
stands for the group Im∇n of n-dimensional coboundaries. (We are working here either
with singular cochains in topological space X or with cellular cochains in cell complex X.
We obtain either singular cohomology of topological space X or cellular cohomology; as in
the case of homology we will prove that singular and cellular cohomology coincide.) Apply-
ing the definition of cohomology to relative chains in topological space or in cell complex we
obtain a definition of relative cohomology. Notice that that Hom(Cn (X)/Cn (A); G) can be
identified with the subgroup of the group Hom(Cn (X); G) consisting of homomorphisms
that vanish on chains in A.
Examples.
The group of zero-dimensional chains in topological space X is generated by points
of X. Therefore zero-dimensional singular cochains with coefficients in G are in one-to-
one correspondence with G-valued functions on X. Similarly, a one-dimensional cochain
corresponds to a G-valued function on one-dimensional singular cubes (on the space of
continuous maps φ : [0, 1] → X). If a zero-dimensional cochain corresponds to a function
f : X → G then its coboundary corresponds to a function ∇(f )(φ) = f (φ(1)) − f (φ(0)).
This means that the function f specifies a zero-dimensional cocycle iff it is constant on
every connected component of X. We see that H 0 (X; G) is isomorphic to direct sum GN of
N copies of G where N is the number of connected components. Sometimes it is convenient
to work with reduced cohomology group H̃ n (X; G) of non-empty space X that is equal to
H n (X; G) for n > 0 and is defined as G+...+G/G for n = 0. (In the definition of H̃ 0 (X; G)
 HOMOLOGY 11

we factorize the direct sum of N copies of G with respect to G diagonally embedded in

this direct sum.)
As in the case of homology the calculation of singular cohomology for n > 0 directly from
definition is quite difficult. However, again for cell complexes the calculations are much
easier. If we are working with finite cell complex C the boundary operator Cn → Cn−1
can be regarded as a matrix mji having integer entries.( Here Cn = Zαn where αn denotes
the number of n-dimensional cells.) The same matrix specifies the boundary operator
Cn (G) = Gαn → Cn−1 (G) = Gαn−1 . The group of cochains C n (G) also can be identified
with Gαn ; the coboundary operator ∇n : C n−1 (G) = Gαn−1 → C n (G) = Gαn corresponds
to a matrix transposed to the matrix mji .
Applying this remark to the examples of the Sec ... we obtain that for the sphere
we have H 0 (S n ; G) = H n (S n ; G) = G, all other cohomology groups vanish. For real
projective space RP n all cochain groups C k (RP n = G for 0 ≤ k ≤ n; the coboundary
operator vanishes for even k and for k = n; it transforms g ∈ G into 2g ∈ G for other
k. The cohomology H k (RP n ) = G/2G for even k ≤ n. If k is odd and k < n we have
H k (RP n ) = {g ∈ G|2g = 0}, if n is odd H n (RP n ) = G.

6. Exact sequences
P P
A homomorphism of chain complex (A = An , ∂) into a chain complex (C = Cn , ∂)
is a homomorphism ψ : A → C preserving grading and commuting with the boundary
operator ∂. ( We use the same notation for boundary operator in both complexes.) In other
words, a homomorphism of chain complexes is a collection of homomorphisms ψn : An →
Cn obeying ∂n ψn = ψn−1 ∂n . It is obvious that a homomorphism of chain complexes induces
a homomorphism of corresponding homology denoted by ψ∗ ; it preserves the grading :
ψ∗ : Hn (A) → Hn (C).
If ψ is injective we can consider An as a subgroup of Cn . In this case we say that A is a
subcomplex of X ; we can talk about P a pair of complexes (C, A) and introduce the relative
homology H(C, A) = H(C/A) P= Hn (C/A) (homology of the pair) as the homology of
the chain complex (C/A = Cn /An , ∂). (The boundary operator ∂ descends to C/A in
obvious way.) It is easy to define a homomorphism Hn (C/A) → Hn−1 (A) also called a
boundary operator and denoted by ∂. Namely, for every element of ω ∈ Zn (C/A) we take
a representative in Cn . Its boundary is a cycle in An−1 ; we send the homology class of ω
into the homology class of this cycle.
Theorem 8. The sequence
... → Hn (A) → Hn (C) → Hn (C/A) → Hn−1 (A) → ...
is exact.
The proof is straightforward. Let us check, for example, the exactness in the term
Hn (C/A). Let us consider a relative cycle ω and corresponding homology class [ω] ∈
Hn (C/A). If ∂[ω] = 0 then ∂ω is homologous to zero in A, hence ∂ω = ∂σ where σ ∈ A.
We can say that ω − σ is an absolute cycle in C; its homology class is mapped into
relative homology class [ω] ∈ Hn (C/A). We proved that the kernel of boundary operator
12 HOMOLOGY

is contained in the image of absolute homology. To check that the image is contained in
the kernel we notice that an absolute cycle is a relative cycle with zero boundary.
An exact sequence for homology with coefficients in abelian group G can be derived as
a corollary of this theorem.
A homomorphism of chain complexes ψ : A → C induces homomorphisms of groups
of cochains C n (G) → An (G) commuting with coboundary operator. This means that ψ
generates a map of corresponding cohomology ψ ∗ : H n (C; G) → H n (A; G) (cohomology
is a contravariant functor on a category of chain complexes with values in the category of
abelian groups). If ψ is injective we consider A as a subgroup of C and define relative coho-
mology H ( C, A, G) in terms of relative cochains Hom(C n /An ; G). (The boundary operator
∂n : C n /An → C n−1 /An−1 induces the coboundary operator ∇n : Hom(C n−1 /An−1 ; G) →
Hom(C n /An ; G). We define relative cohomology using this coboundary operator.)
Notice, that a relative cochain ω ∈ Hom(C n /An ; G) can be interpreted as an absolute
cochain [ω] ∈ Hom(C n ; G) that vanishes on An . We will assume that every homomorphism
An → G can be extended to a homomorphism C n → G. This assumption permits us to
define a homomorphism H n (A; G) → H n+1 (C, A; G) (coboundary homomorphism) as a
map sending cohomology class of a cocycle in A to a cohomology class of a relative cocycle
obtained as a coboundary of an extension of a cocycle in A (obviously this coboundary
vanishes on A).
Theorem 9. The sequence
... → H n−1 (A; G) → H n (C, A; G) → H n (C; G) → H n (A; G) → ...
is exact.
The proof is simple and very similar to the proof of corresponding statement for homol-
ogy.
It is important to notice that homology and cohomology groups with values in arbitrary
group G can be expressed in terms of homology with coefficients in Z. This statement (the
universal coefficients theorem) will be proven in the next section for the case when G has
finite number of generators. In the case when G is a field of characteristic zero (or, more
generally, G is a torsion-free group ) it is easy to check that Hn (C; G) = Hn (C) ⊗ G. If G
is a field then H n (C; G) is dual to Hn (C; G) as a vector space: H n (C; G) = Hn (C; G)∗ .
The main theorems that we have proven for homology with integer coefficients can be
generalized for homology with arbitrary coefficients and for cohomology. Let us formulate
cohomological versions of these theorems.
Theorem 10. Every continuous map f : X → Y induces a homomorphism f ∗ : H n (Y ; G) →
H n (X; G). More generally, a map of pairs f : (X, A) → (Y, B) induces a homomorphism
of relative cohomology f ∗ : H n (Y, B) → H n (X, A). The induced homomorphisms obey
(f g)∗ = g ∗ f ∗ , (id)∗ = id (in other words, cohomology is a contravariant functor from the
category of topological spaces into the category of abelian groups and relative cohomology is
a contravariant functor from the category of pairs in the category of abelian groups).
If two maps are homotopic (as maps of topological spaces or as maps of pairs) corre-
sponding maps of cohomology groups (of relative cohomology groups) coincide.
 HOMOLOGY 13

Theorem 11. The sequence

... → H n−1 (A; G) → H n (X, A; G) → H n (X; G) → H n (A; G) → ...
is exact.
In this statement X is a topological space (or a cell complex) and A is its subspace (or
a a subcomplex). The statement follows from Theorem 9 applied to the complex of chains
in X and its subcomplex consisting of chain in A.
Theorem 12. If A is a subcomplex of cell complex X (or, more generally, A is an NDR
of topological space X) one can prove that
H n (X, A; G) = H̃ n (X/A; G).
Theorem 13. Cellular cohomology of cell complex X is isomorphic to singular cohomology
of polyhedron X. Similar statement is true for relative cohomology.

7. Product
In this section we will study the homology and cohomology of direct product of topo-
logical spaces. We will apply the results to define a product of cohomology classes.
We will work with cell complexes, hence our statements will be proven for polyhedra.
However, they are correct also in more general situation.
Notice first of all that a product of k-dimensional cube I k and l-dimensional cube I l is a
(k + l)-dimensional cube I k+l and that the orientation of factors induces an orientation of
the product. It is easy to prove the following formula for the chain ∂I k+l (for the boundary
of the product):
(6) ∂I k+l = ∂I k × I l + (−1)k I k × ∂I l .
(We define the product of chains extending the definition of the product of cubes by
linearity.)
If we have cell decompositions of polyhedra X and Y we can define a cell decomposition
of direct product X × Y taking as cells direct products of cells in X and cells in Y . Then
C(X × Y ) = C(X) ⊗ C(Y ) ( the group of cellular chains in X × Y is a tensor product
P groups of cellular chains in X and Y ). More precisely, we can say that Cn (X × Y ) =
of
k+l=n Ck (X) ⊗ Cl (Y ). (Cells of dimension n that constitute a basis of Cn (X × Y ) are
products of cells of dimension k and n−k.) If the group of coefficients R has also a structure
of a commutative ring we can consider groups of chains as R-modules; then the R-module
C(X × Y ; R) is tensor product of R-module C(X; R) and C(Y ; R). We see that one can
assign to a pair of chains ω ∈ Ck (X; R) and σ ∈ Cl (Y ; R) a chain ω ⊗ σ ∈ Ck+l (X × Y ; R).
It follows from (6) that
(7) ∂(ω ⊗ σ) = ∂ω ⊗ σ + (−1)k ω∂σ.
If ω and σ are cycles then ω ⊗ σ is a cycle; it is easy to check that homology class of this
cycle depends only on homology classes of ω and σ . This means that we constructed a
14 HOMOLOGY

homomorphism Hk (X; R) ⊗ Hl (Y ; R) → Hk+l (X × Y ; R). 6 One can prove that in the case
when R is a field the homomorphism
X
Hk (X; R) ⊗ Hl (Y ; R) → Hn (X × Y ; R)
k+l=n

is an isomorphism (see the next section).

Notice that for arbitrary topological spaces X and Y we can define a product of k-
dimensional singular cube in X and l-dimensional singular cube in Y as an (k+l)−dimensional
singular cube in X × Y :
(φ × ψ)(t1 , ...., tk+l ) = (φ(t1 , ..., tk ), ψ(tk+1 , ..., tk+l ).
By linearity we extend this definition to singular chains with coefficients in a commutative
ring R. We obtain a homomorphism Ck (X; R) ⊗ Cl (Y ; R) → Ck+l (X × Y ; R) for singular
chains; it obeys (7 and induces a homomorphism Hk (X; R) ⊗ Hl (Y ; R) → Hk+l (X × Y ; R)
for singular homology. It is obvious that this homomorphism coincides with homomorphism
we constructed using cellular homology. It follows that the latter homomorphism does not
depend on the choice of cell decomposition.
Very similar considerations can be applied to cohomology of direct product. For any
commutative ring R we can say that
X
C n (X × Y ; R) = C k (X; R) ⊗ C l (Y ; R).
k+l=n

The coboundary operator is given by the formula dual to the (7):

∇(f ⊗ g) = ∇f ⊗ g + (−1)dimf f ⊗ ∇g.
Again a product of cocycles is a cocycle; we have a homomorphism
H k (X; R) ⊗ H l (Y ; R) → H k+l (X × Y ; R).
If R is a field the R-module H n (X × Y ; R) is isomorphic to
X
H k (X; R) ⊗ H l (Y ; R).
k+l=n

However, in the case of cohomology we can define not only this homomorphism (called
cross-product), but also a product of cohomology classes of space X (cup product).
Notice, that the diagonal inclusion x → (x, x) specifies a map i : X → X × X and
corresponding map of cohomology i∗ : H n (X ×X; R) → H n (X; R). Taking the composition
of cross-product with this map we obtain cup-product f ∪ g ∈ H k+l (X; R) of cohomology
classes f ∈ H k (X; R), g ∈ H l (X; R):
f ∪ g = i∗ (f ⊗ g).
This operation is bilinear (because it is defined as a map of tensor product);
P k it specified
•
a structure of associative graded ring on cohomology H (X; R) = H (X; R). This
6More generally, for arbitrary abelian groups G and G one can construct a homomorphism H (X; G )⊗
1 2 k 1
Hl (Y ; G2 ) → Hk+l (X × Y ; G1 ⊗ G2 ).
 HOMOLOGY 15

ring (cohomology ring of X) is not commutative in general: even (=even-dimensional)

elements of the ring commute with all other elements, but odd (=odd-dimensional) elements
anticommute
(8) f ∪ g = (−1)dim f dim g g ∪ f
(This follows immediately from the remark that interchanging first k coordinates with the
last l coordinates we change the orientation if k and l are odd and preserve the orientation
if one of them is even.)
It is easy to check that a continuous map Φ : X → Y induces a homomorphism of
cohomology ring: Φ∗ (f ∪ g) = Φ∗ f ∪ Φ∗ g.
Notice, that similar construction of a product can be applied to homology in X if X is
a topological group (or, more generally, an H-space). The multiplication of elements of X
determines a map X × X → X and corresponding map of homology: Hn (X × X; R) →
Hn (X; R). Composing this map with cross-product Hk (X; R) ⊗ Hl (X; R) → Hk+l (X ×
X; R) we obtain a structure of associative graded ring on homology (Pontryagin product).

8. Universal coefficient formula. Künneth theorem

The topological considerations of preceding section are closely related to algebraic results
that will be proved in present section.
Let K denote a graded free abelian differential group (=a chain complex of free abelian
groups ... → Kn → Kn−1 → ...). Let us denote corresponding homology by Hn (K).
For any abelian group G we define the homology of K with coefficients in G as the
homology of the complex
... → Kn ⊗ G → Kn−1 ⊗ G → ...
It is denoted by Hn (K; G). We would like to obtain an expression for Hn (K; G) in terms
of H( K) (the universal coefficient formula for homology)
Theorem 14. There exists an exact sequence
0 → Hn ⊗ G → Hn (K; G) → T or(Hn−1 ; G) → 0
and this exact sequence splits (i.e. Hn (K; G) is isomorphic to direct sum of Hn ⊗ G and
T or(Hn−1 ; G)).
The operation T or(Γ; G) in this theorem obeys T or(Γ1 +Γ2 ; G) = T or(Γ1 ; G)+T or(Γ2 ; G),
m
T or(Z; G) = 0, T or(Zm ; G) = {g ∈ G|mg = 0} = Ker(G −→ G)
These properties allow us to calculate T or(Γ; G) for every finitely generated abelian
group Γ. This is sufficient for us, because we would like to prove Theorem 14 only in the
case when all groups Kn are finitely generated.
Notice, that T or(Γ; G) = 0 if G has no torsion. Hence Hn (K; G) = Hn ⊗ G if G is
torsion-free.
To prove the theorem we need the following
16 HOMOLOGY
P
Lemma 15. If K = Kn is a chain complex of finitely generated free abelian groups we
can chose free generators in groups Kn in such a way that the matrix of boundary operator
∂ is diagonal (i. e. ∂ transforms every generator into a multiple of another generator).
It follows from this lemma every chain complex of finitely generated free abelian group
can be represented as a direct sum of elementary complexes of the form 0 → Z → 0 and
0 → Z → Z → 0. This means that it is sufficient to check the statement of Theorem 14
only for these two complexes. For the first one the statement is obvious. Let consider
the second one, assuming that the boundary operator Z → Z is given by multiplication by
m > 0. Then the homology of the complex are 0, 0, Zm , 0 and the homology of this complex
m
with coefficients in G are 0, Ker(G −→ G), G/mG, 0. This agrees with the statement of
Theorem 14.
It remains to give a proof of the lemma. Let us start with the homomorphism ∂ : Kn →
Kn−1 . Changing free generators in these groups (= using elementary column and row
operations) we can transform the matrix of this homomorphism into a diagonal matrix.
Some of new generators of Kn−1 can be obtained up to a factor from new generators of Kn ;
it follows from the relation ∂ 2 = 0 and our assumption that the groups Ki are free that
these generators belong to the kernel of ∂; we will not change these generators anymore.
Now we will consider the homomorphism ∂ : Kn−1 → Kn−2 restricted to a subgroup
spanned by remaining new generators of Kn−1 . Again we change the generators in this
subgroup and in Kn−2 to diagonalize the matrix of this homomorphism. Repeating this
construction we obtain a proof of the lemma.
For any abelian group G we define the cohomology of K with coefficients in G as the
homology of the complex
... → Hom(Kn−1 , G) → Hom(Kn ; G) → ...
It will be denoted by H n (K; G).
If the the abelian groups Kn are free one can prove an expression for H n (K; G) in terms
of Hn (K) ( universal coefficient formula for cohomology).
Theorem 16. There exists an exact sequence
0 → Ext(Hn−1 ; G) → H n (K; G) → Hom(Hn ; G) → 0;
this exact sequence splits (i.e. H n (K; G) is isomorphic to direct sum of Hom(Hn ; G) and
Ext(Hn−1 ; G)).
The operation Ext in this theorem can be calculated for all finitely generated abelian
groups Γ by means of formulas Ext(Γ1 +Γ2 ; G) = Ext(Γ1 ; G)+Ext(Γ2 ; G), Ext(Z; G) = 0,
Ext(Zm ; G) = G/mG.
Again Lemma 15 allows us to reduce the proof in the case of finitely generated groups
to the straightforward consideration of elementary complexes.
Very similar arguments permit us to prove Künneth formula expressing homology of
tensor product of two chainPcomplexes in terms
P of homology of factors. Let us consider
two chain complexes K = Kn and L = Ln of free abelian groups. We define their
 HOMOLOGY 17
P P
tensor product K ⊗ L = (K ⊗ L)n where (K ⊗ L)n = k+l=n Kk ⊗ Ll as a chain complex
with boundary operator defined by the formula
∂(ω ⊗ σ) = ∂ω ⊗ σ + (−1)k ω∂σ
where ω ∈ Kk , σ ∈ Ll .
Theorem 17. For every n there exist exact sequence
X X
0→ Hi (K) ⊗ Hn−i (L) → Hn (K ⊗ L) → T or(Hi (K); Hn−i−1 (L)) → 0.
This sequence splits.
In the case of finitely generated groups the proof can be based on simple application of
Lemma 15.
Combining this theorem with Theorems 14 and 16 or using the fact that Lemma 15 can
be proved for free modules over principal rings one can analyze homology and cohomology
of tensor product in more general situations.
We will not discuss these generalizations, however, we will formulate some simple and
very useful statements about homology and cohomology with coefficients in a field F .
These statements follow from the above theorems. They also can be derived independently
from the fact that an analog of Lemma 15 is valid for vector spaces over F.
H k (K; F) = Hk (K, F)∗ ,
X
Hn (K ⊗ L; F) = Hk (K; F) ⊗ Hl (L; F),
k+l=n
X
n
H (K ⊗ L; F) = H k (K; F) ⊗ H l (L; F).
k+l=n