Functoriality, Singular Homology, Relative

Homology

18 Nov, 2010

1 Overview
Last week, we defined the simplicial homology groups of a triangulated space and gave
a matrix reduction algorithm for their computation. This week, we will:

• Discuss functoriality: how continuous maps between topological spaces induce

algebraic maps between their respective homology groups.

• Give a slightly different definition of homology, called singular homology, which

is not so intimately tied to a triangulation.

• Define and compute the relative homology groups for pairs of spaces.

2 Functoriality
Now we come to probably the most important property of homology: the fact that maps
between spaces induce maps between homology groups. We’ll demonstrate this first
for simplicial maps between simplicial complexes. The whole discussion becomes a
lot cleaner after we define singular homology groups.

Induced maps. Suppose we have a simplicial map f : K → L between two sim-

plicial complexes; recall that this means f maps each simplex σ in K linearly onto
some simplex, of the same or lower dimension, f (σ) of L. For each integer p, we
thus see that f maps p-chains in K to p-chains in L. Formally, given a p-chain
c = Σi ai σi ∈ Cp (K), we define f# (σ) = Σi ai τi ∈ Cp (L), where ti = f (σi ) if
f (σi ) also has dimension p, and ti = 0 if f (σi ) has dimension less than p.
Note that this defines homomorphisms f# : Cp (K) → Cp (L) in each dimension
p. We also have, for each p, the two boundary maps ∂pK : Cp (K) → Cp−1 (K) and
∂pL : Cp (L) → Cp−1 (L). And it is easy to see that f# commutes with these two
boundary maps: ∂pL ◦ f# = f# ◦ ∂pK . In other words, f# has the defining characteristic
of a chain map between two chain complexes.

1
 D

A B A
B
S
E

C C

Figure 1: K is on the left, L is on the right, and the map is indicated by the labelling.

Hence f# maps cycles to cycles and boundaries to boundaries: f# (Zp (K)) ⊆

Zp (L) and f# (Bp (K)) ⊆ Bp (L). In other words, f# induces a map on homology
groups, which we denote by f∗ : Hp (K) → Hp (L); note that although we have
dropped all mention of p for notational clarity, there is really one map f∗ for each
dimension p.

Example. Probably the easiest situation to understood is a homology map induced

by an injective map, as illustrated in Figure 1. On the left we have a simplicial complex
K and on the right we have L. We define f : K → L in such a way the labelled edges
in Figure 1 correspond. Note that in this case, our map f is merely a way to mentally
include K as a subcomplex of L, and the chain map f# performs a similar service for
chains. The induced homology map f∗ is more interesting, and we’ll discuss it in each
dimension.

• f∗ : H0 (K) → H0 (L) is an isomorphism: each complex has only one compo-

nent, and f (K) contains a representative of this component.

• Notice that H1 (K) and H1 (L) are both rank one, but that f∗ : H1 (K) → H1 (L)
is not an isomorphism, because it has a non-zero kernel: namely the homology
class in H1 (K) represented by the 1-chain γ = A + B + C. This class does not
bound in K, but it does bound in L, since γ = ∂2L (S), and hence f∗ (γ) = 0.
Note also that the 1-dimensional homology class in L, represented by C +D+E,
is not in the image of f∗ .

• There are no 2-dimensional classes in either complex, so f∗ : H2 (K) → H2 (L)

is an isomorphism for trivial reasons.

As we’ll see in a few weeks, homology maps induced by inclusions will be the main
theoretical tool needed for the definition of persistent homology.

Functorial properties. The fact that a continuous map between topological spaces
(or a simplicial map between simplicial complexes, but it’s really the same thing as
we’ve seen!) induces an homomorphism between homology groups is incredibly im-
portant: basically, it means that we can associate algebraic data to each space, and
then attempt to transfer this information from space to space in order to make compar-
isons. The following two properties show that this information transfer takes place in
a consistent manner:

2
 • Let K be a simplicial complex and i : K → K denote the identity map given
by i(v) = v for all vertices in K. Then the induced map i∗ is also the identity
homomorphism.
• Let f : K → L and g : L → M be two maps between simplicial complexes, and
consider the composition (g ◦ f ) : K → M . Then (g ◦ f )∗ = g∗ ◦ f∗ ; in other
words, the induced map of the composition is the composition of the induced
maps.
As an easy consequence of these two properties, we have the following essential fact:
if f : K → L is a homeomorphism, then f∗ : Hp (K) → Hp (L) is an isomorphism for
every p.

proof:

This is most useful in the contrapositive. For example, we can now show that the
sphere S 2 is not homeomorphic to the torus T 2 , since there can never be an isomor-
phism between H1 (S 2 ) = 0 and H1 (T 2 ) = Z/2Z ⊕ Z/2Z.
In fact, something much more powerful is true: if f, h : K → L are two homotopic
maps, then the induced maps are equal: f∗ = h∗ . From this, we see that a homotopy
equivalence will also induce a homology isomorphism. Hence S 2 and T 2 are not even
homotopically equivalent!

3 Singular Homology
So far, we have defined homology in terms of various sets of simplices which arise from
a given triangulation of a topological space. Although this is necessary for algorithmic
purposes, it is also theoretically cumbersome; we do not actually want to mentally
triangulate a space every time we wish to think about its homology groups! Fortunately,
there’s a way around this problem, although we are forced to deal with yet one more
level of abstraction.
Let X be a topological space, and for each integer p, let ∆p denote the standard
p-simplex with the topology inherited from Euclidean space. A singular p-simplex in
X is a continuous mapping σ : ∆p → X.

We can formally add singular p-simplices to form singular p-chains, and the set of
such chains forms a vector space Sp (X), called the p-th singular chain group in X. As
before, the chain groups are connected by boundary maps: ∂p : Sp (X) → Sp−1 (X),
defined on a basis as follows. Given a singular p-simplex σ : ∆p → X, we define
∂p (σ) to be the formal sum of the singular (p − 1)-simplices obtained by restricting the
map σ to each of the (p − 1)-faces of ∆p in turn.

3
 Finally, we can define singular p-cycles and singular p-boundaries exactly as be-
fore, and then define the p-th singular homology group of X, Hp (X), to be p-cycles
modulo p-boundaries, again exactly as before. Here are several important facts about
singular homology:

• Amazingly, the singular homology groups of a triangulable space X are isomor-

phic to the simplicial homology groups of any simplicial complex which trian-
gulates X.
• As defined, the singular homology groups are clearly not algorithmically com-
putable (for one, the vector spaces Sp (X) are uncountably infinite in dimen-
sion!); on the other hand, they’re useful for mental arithmetic, and we will see
later that there are some systematic ways of understanding them.

Functoriality. Induced maps are much easier to understand in the context of singular
homology. Let f : X → Y be a continuous map between topological spaces and let
σ : ∆p → X be a singular p-simplex in X. Then f# (σ) is defined to be the singular
p-simplex in Y given by the composition f ◦ σ : ∆p → Y.

Application. As a simple demonstration of what can be shown with these ideas, we

prove the following famous theorem in all dimensions. Let Bd denote the closed d-
dimensional unit ball.
BROUWERS’S FIXED-POINT THEOREM: every continuous map f : Bd → Bd
has a fixed point.

proof: for purposes of contradiction, suppose that we have a continuous map f :

Bd → Bd which satisfies f (x) 6= x for all x ∈ Bd . Hence for every x, there is a unique
ray starting at f (x), passing through x, and ending at some point r(x) on the boundary
sphere Sd−1 . Mapping each x to its r(x) in this manner, we define another continuous
map r : Bd → Sd−1 .
Now let I : Sd−1 → Sd−1 be the identity map on the sphere and let i : Sd−1 ֒→ Bd
denote the inclusion of the sphere into the ball. Note that the composition r ◦ i = I,
and so, applying the second functoriality property above, we get r∗ ◦ i∗ = I∗ for all
dimensions p. Let’s focus on dimension p = d − 1. By the first property above, we
know that I∗ : Hd−1 (Sd−1 ) → Hd−1 (Sd−1 ) takes the non-zero (in fact, rank one)
homology group of the sphere onto itself. On the other hand, i∗ : Hd−1 (Sd−1 ) →
Hd−1 (Bd ) = 0 must be the zero map, since the target is the zero group! Hence we
have factored an isomorphism through a zero map, which must be a contradiction.

4
4 In-Class Exercises
E b

A Referring to the figure above, let K consist of the vertices a and b, and let L be
the edge E along with its vertices. Let i : K → L denote the inclusion map.

(a) Is i# (a) = i# (b)?

(b) Let α, β ∈ H0 (K) denote the homology classes represented by a and b,

respectively. Is i∗ (α) = i∗ (β)?. What is i∗ (α + β)?

B Look at Figure 1 again. Can you find a simplicial map g : K → L such that g∗
is an isomorphism for every dimension p?

5
 Figure 2: X is the annulus, and X0 is the union of the outer and inner circles

5 Relative Homology
The homology groups measure the higher-order connectivity of a space X, counting, in
some well-defined algebraic sense, the number of components, tunnels, voids, and so
forth in X. On the other hand, suppose we have a nested pair of spaces (X, X0 ) with
X0 ⊆ X, or pair of simplicial complexes (K, K0 ) with K0 a subcomplex of K. The
relative homology groups of this pair will measure the connectivity of the large space
relative to the small space, in a sense we now define. If a distinction is needed, we will
use absolute homology to refer to the homology of a single space.

Definition We give the simplicial definition and leave the obvious singular changes
to you. Assume we have a nested pair of subcomplexes (K, K0 ). The key concept is
that of relative chain groups: for every p, we define Cp (K, K0 ) = Cp (K)/Cp (K0 ). In
other words, a relative p-chain is a coset c+Cp (K0 ). Thus, two p-chains c, c′ ∈ Cp (K)
define the same relative p-chain iff their difference consists entirely of p-simplices in
K0 . For each p, we have a boundary map ∂p : Cp (K, K0 ) → Cp−1 (K, K0 ) which
comes from the boundary map on K: on each coset, we define ∂p (c + Cp (K0 )) =
∂p (c) + Cp−1 (K0 ). Notice that this map is indeed well-defined, since the boundary of
a p-chain in K0 is a (p − 1)-chain in K0 .
Now we follow the same algebraic procedures as in the definition of absolute ho-
mology, setting the group of relative p-cycles Zp (K, K0 ) = ker ∂p , the group of rela-
tive p-boundaries Bp (K, K0 ) = im ∂p−1 , and finally the p-th relative homology group
Hp (K, K0 ) = Zp (K, K0 )/Bp (K, K0 ).

Example To help unpack the definitions above, we work an example in some detail.
Consider the pair of spaces (X, X0 ) where X is the annulus S1 ×[0, 1] as drawn in Figure
2, and X0 is the union of the outer and the inner circles. We will mentally compute the
relative homology group Hp (X, X0 ) in all dimensions. To make this easier, we imagine
that we have triangulated X in such a way that X0 forms a subcomplex; if you don’t
like doing this, then you can also just think of singular chains.

6
We can show that H0 (X, X0 ) = 0:

and H1 (X, X0 ) = Z/2Z:

and H2 (X, X0 ) = Z/2Z

7
Matrix reduction algorithm. The matrix reduction algorithm for computing the rel-
ative homology groups of a pair of subcomplexes (K, K0 ) is almost identical to the
absolute homology algorithm; the only difference lies in the ordering of the columns
and the interpretation of the reduced form. First we choose an ordered basis for each
chain group Cp (K) in such a way that the simplices in K0 come first. Then for each
p, we again create a boundary matrix Dp to represent the boundary map ∂p ; note then
the columns and the rows are first indexed by simplices from K0 and then finally by
simplices in K − K0 . We then reduce the matrix to Smith Normal Form exactly as
before, as shown below. But now we are only interested in the lower right submatrix
indexed by simplices in K − K0 :

The form of the algorithm makes an important fact apparent. Suppose have a further
pair of subcomplexes (L, L0 ) which satisfies the inclusions L ⊆ K, L0 ⊆ K0 , and the
equality L − L0 = K − K0 .

In this case, the EXCISION THEOREM says that the pair have isomorphic relative
homology groups; that is, Hp (K, K0 ) ∼
= Hp (L, L0 ) for all dimensions p. For a proof,
we need only look at the matrix reduction above. We can obtain the boundary matrix
for L by removing all rows and columns which correspond to simplices in K − L. But
it’s easy to see that K − L = K0 , and thus our row and column removal only affects
the irrelevant upper left submatrix.

8
 Figure 3: X = Y is the torus, and Y0 is everything below the thick line.

Maps of pairs. Recall that a continuous (or simplicial) map between topological
spaces (or simplicial complexes) induces a homomorphism between absolute homol-
ogy groups. The same thing happens for relative homology. More precisely, suppose
that we have a continuous map of pairs f : (X, X0 ) → (Y, Y0 ); this means a continu-
ous map f : X → Y with the property that f (X0 ) ⊆ Y0 . As before we can map every
chain in X to a chain in Y, but we also note that every chain in X0 maps to one in Y0 .
In other words, for each p we have a chain map f# : Cp (X, X0 ) → Cp (Y, Y0 ) which
leads to a map between relative homology groups f∗ : Hp (X, X0 ) → Hp (Y, Y0 ).
For a simple example, consider the map of pairs i : (X, ∅) → (X, X0 ), where X the
2-torus and X0 is everything sitting below the thick line in Figure 3. We describe the
induced relative homology homomorphism i∗ Hp (X) → Hp (X, X0 ) in every dimension
p.
For p = 0, we note that H0 (X) is rank one, while H0 (X, X0 ) is trivial, and so i∗
must be the zero map in this dimension.
For p = 1, the group H1 (X) is rank two; let’s take as a basis the homology class
α represented by the loop drawn in Figure 3 and the homology class β represented by
the meridian circle. It is easy to see that i∗ (α) = 0, since α has a representative in X0 .
On the other hand, i∗ (β) is in fact just the generator of the rank-one group H1 (X, X0 ).
Thus, in dimension one, i∗ is surjective and has a rank-one kernel.
For p = 2, the group H2 (X) is rank one, with generator represented by the entire
torus; this generator maps via i∗ to the generator of the rank-one group H2 (X, X0 ).
Hence i∗ is an isomorphism in dimension two.

9
6 Even More Exercises
A Compute (at least mentally) the relative homology groups Hp (B2 , S1 ), where we
think of the circle S1 as the boundary of the closed 2-dimensional unit ball B2 .
Generalize your answer to arbitrary dimension:

B Consider the annulus X as drawn in Figure 2 and recall that X0 is the union of
the inner and the outer circles. Recall that we have already computed Hp (X, X0 )
in every dimension.

• Compute the absolute homology groups Hp (X0 ) and Hp (X):

• Let i denote the inclusion of X0 into X and j denote the inclusion of pairs
(X, ∅) → (X, X0 ). Prove that im i∗ = ker j∗ :

10