232A Final
232A Final
232A Final
MATH 232A
Name: Jiajie (Jerry) Luo
As its name may suggest, the fundamental group of a space is one of the most important
ideas from algebraic topology. It’s an algebraic invariant for topological spaces describing the
(closed) loops up to homotopy in that space. Much can be learned about topological spaces
through the fundamental group, as well as its generalizations - the higher homotopy groups.
While homotopy groups tell us quite a lot about a space, they’re limited in accessibility, as
computing them have been a challenge. This motivates the construction of another algebraic
invariant for topological spaces - homology groups.
Unlike, homotopy groups, homology groups are much more accessible, as they are easier to
compute. The drawback in this is that they are less intuitive and more of a challenge to define.
However, despite this, we will show the power that homology theory possesses in extracting
information about topological spaces.
We begin by discussing simplicial homology, which is much nicer and more intuitive than its
more advanced counterpart, singular homology. To do this, we first introduce the structure on
which simplicial homology is built upon - ∆-complex structures.
The construction for ∆-complex structures, in a more or less way, is representing a topological
space with (generalized) triangles. We define an n-simplex as the convex hull of n + 1 points,
{v0 , · · · , vn }, that sit in Rm , where {v0 , · · · , vn } do not lie in a hyperplane of dimension less than
n (note that this requires m ≥ n). A simple example is taking the convex hull of the standard
(orthonormal) basis elements on Rn+1 . We denote the n-simplex [v0 , · · · , vn ], where vi are the
points selected for the construction of the n-simplex. Moreover, we give an orientation on the
edges of our n-simplex by the ordering of the vertices, namely that the direction goes from
the vertex with the lower subscript to the vertex with the higher one. We refer to a n-simplex
n
as ∆n . Given an n-simplex [v0 , · · · , vn ], we give the coordinates of
P
tn vn as (t0 , · · · , tn ). We
i=0
call the set of points with none of the coordinates being 0 the interior (denoted as ∆ ˚ n ), and
n
the set of points whose coodinates have 0 the boundary (denoted as ∂∆ ). We notice that the
coordinates of [v0 , · · · , vn ] are in one to one correspondence with the coordinates (in under the
standard basis) of the n-simplex formed by the standard basis on Rn+1 (ie. taking its convex
hull).
With all this in place, we can now define a ∆-complex. A ∆-complex on a space X is a
collection of maps σα : ∆n → X, where n is indexed by α, such that:
1. σα |∆
˚ n is injective and each point of X is contained in exactly one image of the σα |∆
˚n .
1
on a space X is, in essence, just covering X up in (generalized) triangles. More specifically,
what we are doing is that we are mapping n-simplices into our space X in such a way where
our n-simplices cover X, but in a ‘non-overlapping’ way (ie. the interiors of the triangles do
not overlap). Moreover, while we can prescribe ∆-complexes to topological spaces, we can
also build topological spaces from disjoint simplices by way of identifying faces and taking
quotients.
Our next step towards simplicial homology is to define the chain complex of X. We denote
∆n (X) asPthe free abelian group generated by the n-simplices of X. These can be written of
the form α nα σα , for nα ∈ Z. We refer to the elements of ∆n (X) as n-chains. We recall from
earlier that the faces of an n-simplex, form the boundary. We use this definition to motivate
n
a homomorphism ∂n : ∆n (X) → ∆n−1 (X), defined by ∂n (σα ) = (−1)n σα |[v0 ,··· ,vˆi ,··· ,vn ] , where
P
i=0
[v0 , · · · , v̂i , · · · , vn ] is the n − 1 simplex [v0 , · · · , vi−1 , vi+1 , · · · , vn ] (that is, we remove the ith
vertex). Since the restriction of σa to a face is indeed another member of the ∆-complex, we see
that ∂n does indeed map into ∆n−1 (X). We call this map the boundary map, and rightfully so,
as it does map n-simplices to their boundaries. Indeed, as we iterate through i, we run through
all the n − 1-simplices that are faces of a given n-simplex. Moreover, the alternating sign takes
care of the orientation of our simplicies, so that heuristically, what’s happening is that we are
‘running around the boundary’. This is best seen in a 2-simplex, where given [v0 , v1 , v2 ], we
see that ∂2 ([v0 , v1 , v2 ]) gives [v1 , v2 ] − [v0 , v2 ] + [v0 , v1 ], which literally runs around the 2-simplex
(this is technically an abuse of notation, as ∂2 is defined on ∆2 (X), which consists of maps
from of the 2-simplices to X, and not the simplices themselves; however, seeing the n-chains
as simplicies is helpful in this context as to why the map is defined the way it is). This helps
motivate two definitions - we call elements of ker(∂n ) cycles, while the elements of Im(∂n+1 )
boundaries. We also observe that ∂n ◦ ∂n+1 is the 0 map. Since the context of the boundary
maps on chain complexes are quite clear, we will abbreviate them as ∂ when it is convenient to
do so. To put things together, we have the groups ∆n (X) and the boundary map (technically
plural, but they all do the same thing anyways) behaving in the following way
∂ ∂ ∂ ∂ ∂ ∂
· · · → ∆n+1 (X) → ∆n (X) → ∆n−1 (X) → · · · → ∆1 (X) → ∆0 (X) → 0.
Now, we finally have all the tools we need in order to define the simplicial homology groups.
Since we have that ∂n+1 ◦∂n ≡ 0, we see that Im(∂n+1 ) ⊂ ker(∂n ). We define the nth homology
group of X, as Hn∆ (X) = ker(∂n )/Im(∂n+1 ).
Let us now take a moment to appreciate the complexity that is simplicial homology. This gives
us a brief taste of what topologists mean when they say ‘homology is hard to define.’ Here
already, we see the difficulty of its construction, and how it’s not completely clear (at least not
at first glance) what the elements of Hn∆ represent geometrically. Moreover, the construction of
the simplicial homology groups give rise a few questions. Do homotopy equivalent spaces have
isomorphic homology groups? More specifically, do homeomorphic spaces have isomorphic
homology groups? Better yet, does the homology group of a space depend on the choice of ∆-
complex structure? Unfortunately, the answers to these questions are not clear in the context
of simplicial homology. So, we shift our gaze to singular homology.
The setting for singular homology is similar to that of simplicial homology. Only now, our
class of maps is greatly expanded. Singular homology relies not on a ∆-complex structures to
2
‘restrict’ what maps we have, but rather, our maps, which we refer to as singular n-simplicies
σ : ∆n → X, can now be any (continuous) map. The analog of ∆n (X) is the nth chain group,
Cn (X), which is defined as the free abelian group generated by all singular n-simplices. We
similarly have the boundary map ∂n : Cn (X) → Cn−1 (X), which is defined exactly the same
way as it was in the context of simplicial homology. As before, we since we have ∂n ◦ ∂n+1 ≡ 0,
we can define the nth homology group in the same way as before, HL n (X) = ker(∂n )/Im(∂n+1 ).
A remark we can make from here is that if we split up Cn (X) = α Cn (Xα ), where Xα are
the path components of X, we note that ∂n isL a restricted to Cn (Xα ) maps into Cn−1 (Xα ).
From this, we can actually get that Hn (X) ∼ = α Hn (Xα ). The construction of the singular
homology groups are much more complicated than that of the simplicial homology groups, and
as such, they are much harder to think about. For example, the groups that make up the chain
complex of a space can potentially be massive (as it’s generated by all continuous mappings
from a n-simplex into X).
Let us begin by computing singular homology for the simplest space we can think of - that of
a point (we will refer to this space as X). Since there is only one map from any space to a
point, we see that each of the Cn (X) ∼= Z. We refer to the generator of Cn (X) as σn (which
n
is the unique map σn : ∆ → X). Let us now compute the boundary maps. We note that
n
since ∂n (σn ) = (−1)n σn−1 , the boundary map ∂n is an isomorphism if n is even and the zero
P
i=0
map when n is odd. From this, we calculate that Hn (X) = 0 when n > 0 and H0 (X) ∼ = Z.
The homology of X is a bit awkward since we have that the homology is trivial except for the
0th spot, which is isomorphic to Z. In fact, this is true for every space that’s retractable to a
point, so for convenience, we introduce reduced homology groups.
The construction for reduced homology groups is similar to that of singular homology groups,
except for one of the boundary maps being replaced with a different map. For a space X,
instead
having ∂0 : C0 (X) → 0, we replace it with : C0 (X) → Z, where we have
of
P P
nα σα = nα . That is, we have the following sequence:
α α
∂ ∂ ∂
· · · → C2 (X) → C1 (X) → C0 (X) → Z → 0.
As before, we have H̃n (X) = ker(∂n )/Im(∂n+1 ) for n > 0, and H̃0 (X) = ker()/Im(∂1 ).
By this construction, we note that Hn (X) = H̃n (X) for n > 0, and H0 (X) ∼ = H̃0 (X) ⊕ Z.
Moreover, if X was a singleton set, or any contractible space, we have that H̃n (X) = 0 for
every n. What’s more is that we can use the map to show that given a path connected space,
H0 (X) ∼
= Z, and by extension, if X has k path components, we would have that H0 (X) ∼ = Zk .
Now, suppose f : X → Y is a continuous map. We notice that f induces a map on the chain
groups, f] : Cn (X) → Cn (Y ) where f] (σ) = f σ (we can linearly extend f] , and see that it
forms a group homomorphism). A simple observation tells us that f] ∂ = ∂f] , and so, we have
the following commutative diagram:
∂ ∂ ∂ ∂
··· Cn+1 (X) Cn (X) Cn−1 (X) ···
f] f] f]
∂ ∂ ∂ ∂
··· Cn+1 (Y ) Cn (Y ) Cn−1 (Y ) ···
3
In general, if f] ∂ = ∂f] , we say that f] is a chain map. From this, we see that f] takes cycles
to cycles and boundaries to boundaries, and so, f] induces a homomorphism f∗ : Hn (X) →
Hn (Y ). Much like the induced map for the fundamental group, the induced map on homology
groups behaves nicely in many of the same ways. For example, given f : X → Y and
g : Y → Z, we have that (gf )∗ = g∗ f∗ . Moreover, we have that 1∗ = 1. One of the more
nontrivial facts that we also discover is that if f : X → Y and g : X → Y are homotopic, then
f∗ = g∗ on the level of homology groups. A direct consequence of this is that if f : X → Y
is a homotopy equivalence, then f∗ : Hn (X) → Hn (Y ) is an isomorphism. From this, we see
that homotopic equivalent spaces have isomorphic singular homology groups, something that
wasn’t so obvious under the setting of simplicial homology groups.
Now that we have seen some very nice properties of singular homology, a natural question
to ask is exactly how far we can go. In particular, does taking the quotient of a space X
by a subspace A play nicely on the level of homology groups? Or better yet, do we have
Hn (X/A) ∼ = Hn (X)/Hn (A)? A quick reality check tells us that while this would be nice, it’s
actually not true. For example, if we take any topological space X and embed it into its cone
CX, we have that CX is contractible, which means Hn (CX) = 0 for all n > 0. Fortunately,
hope is not lost, as there is a way we can relate X, A, and X/A in terms of long exact
sequences of reduced homology groups, if A is “good”. To define what “good” means, we say
that for a nonempty closed A ⊂ X, (X, A) is a good pair if A is a deformation retract of some
neighborhood of X. In particular, if (X, A) is a good pair, we have the long exact sequence
i ∗ j∗ ∂ ∗ i
· · · → H̃n (A) → H̃n (X) → H̃n (X/A) → H̃n−1 (A) → H̃n−1 (X) → · · · → H̃n (X/A) → 0
where i : A ,→ X is the inclusion map and j : X → X/A is the quotient map. The map
∂ takes inspiration from the boundary map on the level of chain complexes that we know
and love, and will be constructed as we go. Some of the consequences of this are that by
leveraging the fact that this sequence is exact, we can come up with some nice results as well
as easy ways to compute homologies, without even thinking much about what i∗ , j∗ , or ∂ are
doing on the level of homology groups. For example, we can relatively easily compute the
homology of the k-sphere, which are H̃k (S k ) = Z and H̃n (S k ) = 0 for all n 6= k (we note that
the homology group and reduced homology groups coincide for n > 0, and we can separately
compute H0 (S k ) = Z, as it is connected).
In order to show the exactness of the sequence we have above, we first introduce relative
homology groups. Relative homology groups, in a sense, measure how different Hn (X) is with
Hn (A). Given A ⊂ X, we say that Cn (X, A) = Cn (X)/Cn (A). Since the boundary map
takes elements from Cn (A) to Cn−1 (A), it induces the map ∂n : Cn (X, A) → Cn−1 (X, A), with
∂ 2 = 0 still holding. We call the elements of ker ∂ the relative cycles, and the elements of Im∂
the relative boundaries. Similar to how the other homology groups were defined, we define
Hn (X, A) = ker ∂n /Im(∂n+1 ). One quick remark we make is that for A 6= ∅, we have that
Hn (X, A) = H̃n (X, A). Although it may not be expected at first, we also have induced maps
for relative homology. If f : (X, A) → (Y, B) is a map of pairs (that is, f (A) ⊂ B), we can
make the construction as we did before and induce a map f] from the relative chain complex
Cn (X, A) to Cn (Y, B), which now induces a map f∗ : Hn (X, A) → Hn (Y, B). Indeed, the fact
that map of pairs is needed is makes quite a bit of sense, because if f (A) mapped outside
of B, none of the induced maps would make any sense. Moreover, if f : (X, A) → (Y, B)
4
and g : (X, A) → (Y, B) are homotopic as maps of pairs (that is, their restriction on A are
homotopic too), we have that f∗ = g∗ . So, we see many of the results from absolute homology
carrying over to relative homology, under “obvious” conditions, which is pretty nice.
We note that since
i j
0 → Cn (A) → Cn (X) → Cn (X, A) → 0
is exact, we also have the following short exact sequence of chain complexes:
.. .. ..
. . .
∂ ∂ ∂
i j
0 Cn+1 (A) Cn+1 (X) Cn+1 (X, A) 0
∂ ∂ ∂
i j
0 Cn (A) Cn (X) Cn (X, A) 0
∂ ∂ ∂
i j
0 Cn−1 (A) Cn−1 (X) Cn−1 (X, A) 0
∂ ∂ ∂
.. .. ..
. . .
By using the exactness of the rows, the fact that the diagram commutes, and that ∂ 2 = 0, we are
able to construct, after some diagram chasing, a well defined map ∂ : Hn (X, A) → Hn−1 (A).
As we can see, ∂ : Hn (X, A) → Hn−1 (A) is constructed with boundary map in mind, and
furthermore, can be thought of as the boundary map on relative cycles represented in Hn (X, A).
Moreover, with a big of diagram chasing, we have the following exact sequence which looks
quite similar to the long exact sequence we originally wanted1 :
∗ i j∗ ∂
Hn (A) → Hn (X) → Hn (X, A) → Hn−1 (A) → · · · → H0 (X, A) → 0.
We remark that by following the same process, we can get the same exact sequence for reduced
homology. In fact, we can do something similar with a triple (X, A, B) (with X ⊃ A ⊃ B).
Now, if only we can replace Hn (X, A) with H̃n (X/A)... Turns out, we can! But with a slight
caveat: (X, A) must be a good pair. In order to get there, we need the excision theorem, which
tells us that given Z ⊂ A ⊂ X where cl(Z) ⊂ Å, we have that the inclusion (X − Z, A − Z) ,→
(X, A) induces an isomorphism Hn (X − Z, A − Z) → Hn (X, A) for all n. Another way we can
say this is that if A, B ⊂ X such that Å ∪ B̊ = X, then the inclusion (B, A ∩ B) ,→ (X, A)
induces an isomorphism Hn (B, A ∩ B) → Hn (X, A) for all n. The proof of the excision
theorem is rather long and involved, but the result gets us what we want. Namely if we have
1
We note that this construction is purely algebraic. In fact, given a short exact sequence of chain complexes
(not necessarily derived from a topological space!), we can construct a long exact sequence of homology groups.
This result is known as the Snake Lemma.
5
a good pair (X, A), then the quotient map q : (X, A) → (X/A, A/A) induces an isomorphism
q∗ : Hn (X, A) → Hn (X/A, A/A) for all n. The way the excision theorem is used is that to
prove this, we find a neighborhood of A, which we call V , that deformation retracts back onto
A, and we excise A out. We then use a series of isomorphisms to give us that Hn (X, A) and
Hn (X/A, A/A) are indeed isomorphic. We note that Hn (X/A, A/A) ∼ = H̃n (X/A), and so, this
proposition gives us the final piece required to show Hn (X, A) ∼
= H̃n (X/A), which finally gives
us
i∗ j∗ ∂ ∗ i
· · · → H̃n (A) → H̃n (X) → H̃n (X/A) → H̃n−1 (A) → H̃n−1 (X) → · · · → H̃n (X/A) → 0.
Let us now take a moment to appreciate this piece of machinery that was just discussed. The
long exact sequence actually gives us quite a few results. From the last proposition, we’re able
to deduce that the identity map id : ∆n → ∆n (viewed as a singular n-simplex) generates
Hn (∆n , ∂∆n ), W
which will become useful later. Another result we obtain from our proposition
is that if X = α Xα where the wedge sum is formed at points xα ∈ Xα L L (Xα , {xα })
such that
is good, then the inclusion map i : Xα → X induces an isomorphism α : α H̃n (Xα ) →
H̃n (X). That is, the homology group of wedge sums break up into the homology groups of
their individual components provided that the point the chosen basepoint is good. Another
nice application that uses the excision theorem, as well as our long exact sequence, is that we
can use them to show if nonempty open sets U ⊂ Rn and V ⊂ Rm are homeomorphic, then
m = n. We notice that in many of these examples, a good pair was required in order to make
our machinery run. But what if we didn’t have a good pair? It turns out that in general, we
have the following isomorphism: H̃n (X ∪ CA) ∼ = Hn (X, A).
Now that we’ve discussed simplicial and singular homology in quite a bit of detail, the natural
question is if they agree. That is, given a ∆-complex structure (indeed, the way we defined
simplicial homology was based on such construction), are simplicial homology groups and sin-
gular homology isomorphic? We’d better hope so, because otherwise, we would be in some
trouble. Fortunately, they are. In fact, we can show this in greater generality, that relative
simplicial homology groups (which are defined the same way as relative singular homology
groups) are isomorphic to relative singular homology groups. Since we have the embedding of
∆n (X) ,→ Cn (X) and ∆n (A) ,→ Cn (A), we have a chain map ∆n (X, A) → Cn (X, A). This
induces a homomorphism Hn∆ (X, A) → Hn (X, A), which we can show is actually an isomor-
phism by considering the k-skeleton of X, recalling that the identity map on ∆k generates
Hk (∆k , ∂∆k ), using the long exact sequence we obtained from earlier, and applying the five
lemma - a key tool from homological algebra. We note that the case where A = ∅ reduces down
to an isomorphism of absolute homology. To wrap things up, we note that since we have the
equivalence between simplicial and singular homology, results from one carries into the other.
For example, by the way simplicial homology was constructed, the number of generators for
Hn∆ (X) is at most the number of n-cells in the ∆-complex. As such, this puts a bound on
the number of generators of Hn (X) as well. On the other hand, singular homology tells us
that homotopy equivalent spaces have the same homology group. This means that this holds
in the context of simplicial homology too, which wasn’t very clear at all when we were just
considering simplicial homology. Indeed, it’s nice to see that this equivalence doesn’t just give
us closure, but also results that would otherwise be difficult to verify.
Having discussed the two ways to compute homology, we still have a gaping issue that still
6
needs to be further resolved: computation. While computation using simplicial homology is
pretty straight forward and doable, it’s rather limited in application, as we require a ∆-complex
structure to be imposed onto our space. On the other hand, while singular homology is (in
theory) able to determine a lot about a much wider class of spaces, it can be an utter pain to
make direct computations with (which is why we so frequently had to use roundabout ways to
compute, such as using exact sequences). Now if only we had a homology theory somewhere in
the middle, that can both compute efficiently and across a wide variety of spaces... Fortunately,
we do - cellular homology! Cellular homology is, in a sense, a cross between simplicial homology
and singular homology, in that it has the computational flavor of simplicial homology, but over
a much wider scope of spaces (ie. cell complexes) using facts we know from singular homology.
But before we can talk further about cell complexes, we need to take a slight detour into
computation on the homology of spheres (ie. S n ).
One of the main ingredients to build up cellular homology is the notion of degree for a map
n n
on a sphere.( Consider f : S → S , where n is nonzero. We recall that when n > 0, we have
Z n=k
Hn (S k ) = . From this, we see that f∗ : Hn (S n ) → Hn (S n ) must be multiplication
0 n 6= k
by some integer (ie. f∗ (α) = dα for some d ∈ Z). We define the degree of f , deg(f ), to be this
integer obtained from the induced map on the nth homology2 . In a way, the degree measures
how f “wraps around” S n . The degree of f acts as an invariant on spheres to itself, in that it
has many nice properties, which include the following (among others, which can be referenced
in the very beginning of Section 2.2 of Hatcher)3 :
1. If f and g are homotopic, then deg(f ) = deg(g) as f∗ = g∗ .
2. The degree is multiplicative. That is, deg(f g) = deg(f ) deg(g), which can be easily seen
in that (f g)∗ = f∗ g∗ .
3. If f is a reflection, then deg(f ) = −1, which also means that the antipodal map has
deg(−1) = (−1)n+1 (as it is a composition of n + 1 reflections).
Our next order of business to develop is a rather handy lemma about CW complexes, which
will be the domain on which we build cellular homology. Let X be a CW complex. Then we
have the following:
1. Hk (X n , X n−1 ) = 0 when k 6= n and is free abelian for k = n. Moreover, there is a one to
one correspondence between the n-cells and generators for Hn (X n , X n−1 ).
2. Hk (X n ) = 0 when k > n.
3. The inclusion map induces an isomorphism i∗ : Hk (X n ) → Hk (X) when k < n. If k = n,
then the induced map is surjective.
Here, (a) can be realized by observing (X n , X n−1 ) is a good pair, and X n /X n−1 is a wedge
sum of n-spheres, with each n-sphere corresponding to an n-cell of X. (b) and (c) can be
2
In particular, we want to make the distinction that we are mapping from our sphere to itself, rather than
another sphere. Thus, the integer for the degree is well defined (as otherwise, there would be confusion in
regards to whether the degree is the positive or negative integer of a certain magnitude).
3
Although there is a lot more that we can say about the degree of a map on a sphere, we will omit most of
the details that are less relevant for the purposes of cellular homology.
7
realized by considering the long exact sequence for the pair (X n , X n−1 ), although the case
where dim(X) = ∞ for (c) requires a bit more work. One observation we can make is that
Hk (X) = 0 when k > dim(X).
We now have the tools that we need in order to define cellular homology. For our chain
complex, we let our nth chain group be Hn (X n , X n−1 ). Although this construction of our
chain complex is rather strange, there’s actually a rather “natural” construction our boundary
map. By considering the long exact sequence for (X n , X n−1 ) (for each n), and using the lemma
that we mentioned above, in the following way: we can construct boundary maps (which we
also refer to as degree maps) in the following way:
Here, we notice that we can construct dn by composing jn−1 ∂n from their respective long exact
sequences. Moreover, our chain groups alongside the maps dn do indeed form a chain complex,
as we notice that dn dn+1 = 0, since inside the composition contains ∂n jn , which is a part of
the long exact sequence for Hn (X n , X n−1 ) (by exactness, this composition must be the zero
map). As we’ve done so for simplicial and singular homology, the nth cellular homology group
is defined to be HnCW (X) = ker(dn )/Im(dn+1 ). Moreover, by observing the properties of the
maps involved in the construction of HnCW (ie. ∂n+1 , jn , ∂n , jn−1 ), we can show that (surprise
surprise) HnCW (X) ∼= Hn (X).
Let us now take a moment to understand the construction that we just concocted. Since for
each Hn (X n , X n−1 ), the generators are in one to one correspondence with the n-cells of X,
we see that the generators of Hn (X) is bounded above by the n-cells of X. In particular, it
is quite evident that if X has no n-cells, then Hn (X) ∼ = 0. Moreover, if there are no n ± 1
cells, then it’s easy to see that Hn (X) is a free abelian group with generators in one to one
correspondence to the n-cells of X. This is quite similar to some of the properties we previously
noted about simplicial homology, except now, it’s applied to a much bigger class of spaces (ie.
CW complexes).
In fact, it’s useful to think of our nth chain group as being free abelian groups generated by
the n-cells of X themselves. As we mentioned before, we sometimes call the cellular boundary
map dn the degree map. The reasons for this is that there is a “natural” way to relate the
cellular boundary map with the notion of degree. Let {enα } be the set of n-cells of X, and
{en−1 n n−1
β } be the set of n − 1-cells of X. For an n-cell eα and an n − 1-cell eβ , consider the
attaching map aα : ∂enα → X n−1 and the quotient map πβ : X n−1 → X n−1 /(X n−1 \ en−1β ). We
8
notice that ∂enα is homeomorphic to S n−1 , as is X n−1 /(X n−1 \ eβn−1 ) (since we are taking our
n − 1 cell and collapsing everything outside of it). So, πβ ◦ aα can be thought of as aPmap from
S n−1 to itself, and so, we can take the degree. We then have the formula dn (enα ) = dαβ en−1β ,
β
where dαβ = deg(πβ ◦ aα ). That is, the coefficient on en−1
for the image of
β enα is, roughly
speaking, the number of times the attaching map “wraps around” en−1
β .
To fully appreciate the work of wonder that is cellular homology, let’s see it in action. One
of the wonderful thing about cellular homology is that it computes the nth homology directly
from the n-cells. What’s more is that we can also use it to work backwards and build spaces to
fit certain conditions homological descriptions. Given an abelian group G and integer n ≥ 1,
we say that if there’s a CW complex X such that Hn (X) ∼ = G and H̃i (X) ∼
= 0 for i 6= n, then
we say that X is a Moore space, or more specifically, X is a M (G, n). We can use cellular
homology to construct Moore spaces in the following manner. First, we know that S n is a
M (Z, n). Now, we take S n and attach an n + 1-cell, en+1 , in such a way where the attaching
map has degree m, we see that the resultant space is actually a M (Zm , n). By taking wedge
sums, we can construct M (G, n) for any finitely generated G. In fact, we can use this idea and
extend this construction for infinitely generated abelian groups to construct spaces X that
are M (G, n) for G being any abelian group. Furthermore, given any set of abelian groups
{Gn }n∈N , we can take wedge sums of M (Gn , n) (for different n) in order to construct a space
X such that Hn (X) ∼ = Gn .
Another example of cellular homology at play is calculating the homology groups of RP n .
First, we note that we can construct S n with two k-cells, for each 0 ≤ k ≤ n, where we attach
each k-cell along the boundary of each of the two k − 1-cells. That is, given any of the two
k-cells eki (i = 1 or i = 2), dk (eki ) = ek1 − ek2 . Taking on this view, RP n would have one k-cell for
each 0 ≤ k ≤ n (we denote the k-cell as ek ). Moreover, because of the antipodal identification,
we have that dk (ek ) = ek−1 − (−1)k+1 ek−1 . From this, for the chain complex of RP n , the kth
degree map alternates from the zero map and the ·2 map, depending on whether k is odd or
even. That is, we have the following chain complex:
2 0 2 0
0 → Z → Z → · · · → Z → Z → 0 if n is even and
0 2 2 0
0 → Z → Z → · · · → Z → Z → 0 if n is odd.
So, we see
Z
k = 0, k = n ≡ 1 mod 2
n
Hk (RP ) = Z2 k ≡ 1 mod 2, 0 < k < n
0 otherwise
In fact, RP 2n−1 is a special case of a class of spaces called lens spaces4 . The construction of
a CW structure for a lens space is similar to that of RP 2n−1 , albeit much more complicated5 .
Given a lens space Lm (`1 , · · · , `n ), where each `1 ∈ Z (which are coprime to m) corresponds
to a rotation, we can construct the following chain complex:
0 m m 0
0 → Z → Z → ··· → Z → Z → 0
4
The 2n − 1 comes from the fact that we are viewing our construction in Cn . In this case, our unit vectors
form a copy of S 2n−1 , which have an even number of coordinates when viewed as a real space.
5
The construction can be found on Example 2.43 in Hatcher.
9
So, we see
Z
k = 0, 2n − 1
Hk (Lm (`1 , · · · , `n )) = Zm k ≡ 1 mod 2, 0 < k < 2n − 1
0 otherwise
Before talking about more computational techniques, let us now consider some places where
homology (in particular, cellular homology) comes up. One of these
P places is the Euler char-
n
acteristic. For a finite CW complex X, we say that χ(X) = (−1) cn , where cn is the
n
number of n-cells of X. We note that for the case of platonic solids (which are represented
as 2 dimensional complexes), this agrees with the P
(less general but better known) definition
of V − E + F . We can also show that χ(X) = (−1)n rankHn (X), which actually shows
n
that the Euler characteristic is a topological invariant, as the homology groups are invariant
under homotopy. This (equivalent) definition of Euler characteristic shouldn’t be too big of a
surprise, because in order to make one of the n-cells not count towards the free part of the nth
homology (ie. make it torsion), we need to either attach an n + 1 cell in a way such that the
n + 1 cell also doesn’t contribute to the free part of the n + 1st homology or attach the n-cell
in such a way that it makes one of the n − 1-cells not count towards the rank of the n − 1st
homology.
Another use of is to study groups. In particular, as it was shown in §1.B, the homotopy type
of a K(G, 1)) (ie. a space X such that π1 (X) ∼ = G) is uniquely determined by the group G.
From this, we can find the homology of a group G by taking Hn (K(G, 1)). The construction
for such a space can be made with a CW complex, from which we can calculate its homology
with cellular homology. This is a rather interesting application, as it shows us yet another
example of the symbiotic relationship different fields of mathematics can have. In particular,
while topology benefits from algebra via the fundamental (and upper homotopy) groups and
homology groups, algebra benefits from topology through homological algebra (including the
motivation for the snake lemma, five lemma, etc.) and group (co)homology.
Let us now recall from earlier all that we were able to compute using the long exact sequence
of homology groups, given a pair (X, A). In particular, the main tool we leveraged was the fact
that our sequence was exact - very rarely did we have to reference the chain complex of either
X or A, nor even what the maps of our exact sequence were. Even though we can compute
homology groups more directly now using cellular homology, let us not forget the power that
comes from exact sequences.
i j
Given a short exact sequence of groups 0 → A → B → C → 0, we say that our short exact
sequence splits if one of the following equivalent definitions hold:
1. There’s a homomorphism p : B → A such that pi = 1A .
2. There’s a homomorphism s : C → B such that js = 1C .
3. There’s an isomorphism φ : B → A ⊕ C such that the following diagram commutes,
where the lower maps are a 7→ (a, 0) and (a, c) 7→ c.
10
i j
0 A B C 0
φ
A⊕C
Showing that these statements are equivalent is rather straightforward, and can even general-
izes to nonabelian cases (with the exception of 2. implying 1. and 3.). It is not difficult to see
how this theorem can be rather useful, in that it can simplify the computation of certain ho-
mology groups. To give a concrete example, we can use this theorem to show the nonexistence
of certain retracts r : X → A for certain spaces, because if such existed, we would necessarily
have that Hn (X) ∼ = Hn (A) ⊕ Hn (X, A),6 which is not true in general.
We now introduce the Mayer-Vietoris sequence, which is, in a sense, the homology equivalent
for van Kampen’s theorem for the fundamental group. Just like how van Kampen’s theorem
gives us the fundamental groups of a space X given two fundamental group of two subspaces,
the union of whose interiors give us all of X, the Mayer-Vietoris sequence gives us a long exact
sequence relating the homology of X with the homology of two subspaces whose interiors7 ,
when unioned, give us all of X. Explicitly, the Mayer-Vietoris sequence is given below, where
the maps will later be specified:
Φ Ψ ∂
· · · → Hn (A ∩ B) → Hn (A) ⊕ Hn (B) → Hn (X) → Hn−1 (A ∩ B) → · · · .
The construction of the Mayer-Vietoris sequence is in the same spirit as the long exact sequence
of pairs. Let Cn (A + B) be the set of n-chains in X that are the sum of n-chains of A and B.
That is, Cn (A + B) = Cn (A) + Cn (B) (note that this is NOT the direct sum!). We see that
the boundary map takes elements of Cn (A) to Cn−1 (A) and Cn (H) to Cn−1 (B), so it follows
that elements of Cn (A + B) get mapped to Cn−1 (A + B). As such we see that we our set of
Cn (A + B) actually form a chain complex.
In fact, we see that Cn (A + B) is, in a sense, Cn (A) ⊕ Cn (B)/Cn (A ∩ B), since Cn (A + B)
count the chains in Cn (A ∩ B) once, while in Cn (A) ⊕ Cn (B), they are counted twice. From
this, we have the short exact sequence
φ ψ
0 → Cn (A ∩ B) → Cn (A) ⊕ Cn (B) → Cn (A + B) → 0,
where φ(x) = (x, −x) and ψ(x, y) = x + y. Moreover, by the main proposition that was used
in the proof of the excision theorem, since the interiors of A and B give X, we have that the
inclusion Cn (A + B) → Cn (X) induces an isomorphism on the homologies. Since we have a
short exact sequence of chain complexes, as well as the isomorphism Hn (A + B) ∼= Hn (X), we
have an induced long exact sequence given by what we have above, where Φ and Ψ are the
induced maps of φ and ψ respectively, and ∂ is the connecting map obtained using the Snake
Lemma. By making a slight modification on the maps from the 0th chain groups, we also have
the Mayer Vietoris sequence for reduced homology.
6
Since if r is a retract, then i : A ,→ X must induce an injection i∗ : Hn (A) ,→ Hn (X).
7
It should be noted that this condition is also that of the excision theorem. This is not a coincidence, as
will be later mentioned.
11
As we can see, the Mayer Vietoris gives us a way to relate the homology of a space X by
the homologies of two subspaces whose interiors give us X based on how they intersect. It
should also be mentioned that we can actually leverage the sequence structure to for inductive
arguments. For example, suppose A1 , · · · , An are open sets sets of X, such that A1 ∪· · ·∪An =
X. Moreover, suppose any intersection of the Ai ’s is either empty or has trivial reduced
homology. Using the (reduced) Mayer Vietoris sequence, we can induct on the number of open
sets that make up X to show Hi (X) ∼= 0 when i ≥ n − 1, by observing that
12
We now consider when G = F , where F is a field. If F has characteristic 2, then each
boundary map is the 0 map, from which we see Hk (RP n ; F ) ∼ = F , for 0 ≤ k ≤ n. However, if
charF > 2, then we see that each multiplication by 2 is an isomorphism, from which we have
that Hk (RP n ; F ) ∼
= F for k = 0 and k odd, and 0 otherwise (again, for 0 ≤ k ≤ n).
Now that we’ve gone over quite a bit of the theory and computation of homology, let us further
abstract what we’ve been doing. The reason for this is that when homology theory was first
developed, there were multiple theories of homology, including simplicial homology, singular
homology, cellular homology, and Cech homology, as well as various others. For a while, most
of these theories seemed to agree, except, well, when they didn’t10 . What could be shown is
that the various theories of homologies agreed when we had a CW complex. As a result, the
following definition for homology theory developed.
We can formalize our notion of homology using the language of category theory. We define a
reduced homology theory h̃ to be a functor between the category of topological spaces to the
category of sequences of abelian groups such that the following hold:
1. If f, g : X → Y are homotopy equivalent, then f∗ = g∗ : h̃n (X) → h̃n (X).
2. For all CW pairs (X, A), there is a group homomorphism ∂ : h̃n (X/A) → h̃n−1 (A) such
that the following sequence is long exact:
i ∗ q∗ ∂ ∗ i
· · · → h̃n (A) → h̃n (X) → h̃n (X/A) → h̃n−1 (A) → h̃n−1 (X) → · · · .
Moreover, given a map f : (X, A) → (Y, B), there is an induced map f¯ : X/A → Y /B
such that the following diagram commutes
∂
h̃n (X/A) h̃n−1 (A)
f¯∗ f∗
∂
h̃n (Y /B) h̃n−1 (B)
W
3. If X = α Xα and iα : Xα ,→ X is the inclusion map for each α, then
M M
iα∗ : h̃n (Xα ) → h̃n (X) is an isomorphism.
α α
We notice that each of these statements have been shown to be true for (reduced) singular
homology as we developed it. As such, the reason for choosing these axioms is none other than
to “force” a homology theory to agree with singular homology for CW complexes. This, in a
sense, makes singular homology our “model homology theory”.
One thing to notice is that this construction is for reduced homologies. In the case of unre-
duced homology, we need to formalize the notion of relative homology groups hn (X, A) as a
10
See the Hawaiian Earring: Example 1.25 in Hatcher.
13
functor, where absolute homology groups are represented hn (X, ∅). Given the notion of rel-
ative homology, the axioms of unreduced homology is similar to that of reduced homology.
The second axiom would need to be modified where the long exact sequence is that of the
unreduced homology groups (with relative homology) as well as a word on excision, such as
hn (X, A) ∼
= hn (X/A, A/A), for CW pairs (X, A). We would also replace the third axiom with
that of disjoint unions instead of wedge sums.
Let us now discuss some of the applications of homology theory as a whole. One result that
we can show using homology theory is the famed Jordan Curve theorem, which tells us that a
circle in R2 has an inside and outside. In more precise terms, if h : S 1 → S 2 is an embedding,
then S 2 \h(S 1 ) has exactly two connected components, each homeomorphic to open disks11 . In
fact, we can prove a more general version of this theorem, which goes as follows: If h : Dk → S n
(k ≤ n) is an embedding, then H̃i (S n \ h(Dk )) = 0. Moreover, if h : S k → S n is an embedding
(again, k ≤ n), then H̃i (S n \ h(S k )) ∼
= Z when i = n − k − 112 .
The first part of this can be proved by induction, by using the Mayer-Vietoris sequence, with
A = S n \ h(I k−1 × [0, 1/2]) and B = S n \ h(I k−1 × [1/2, 1]). From this, we notice that A ∪ B =
S n \ h(I k−1 × {1/2}) and A ∩ B = S n \ h(I k ), from which sets up our induction and (with some
work) gives us the desired result. The second part of the theorem is also proved inductively
with the (reduced) Mayer-Vietoris sequence, where the inductive step requires splitting up
S k into its northern and southern hemisphere D+ k
and D− k
, and letting A = S n \ h(D+
k
) and
n k n k ∼ n
B = S \ h(D− ), which gets us H̃i (S − h(S )) = H̃i+1 (S \ h(S )). k−1
Another result that we can obtain using homology theory is the Borsuk Ulam theorem, which
tells us that an odd map f : S n → S n (ie. f (−x) = −f (x)) has odd degree. The proof of
this13 uses the induced map f¯ on RP n , and as well as homology with coefficients (specifically
we use coefficients from Z2 ), which shows that it’s sometimes useful to consider coefficients
outside of Z. A corollary that follows from this is that any map g : S n → Rn has a point x
such that g(x) = g(−x).
Altogether, we see that homology theory is a rich theory readily available for use in a topological
setting. It allows us to port over algebraic methods and ideas to prove and disprove statements
about topological spaces. One thing that homology is especially good for is to prove the
nonexistence of certain objects. For example, as it has been shown before, if we have A ⊂ X
such that there’s a retract r : X → A, then we necessarily have that Hn (X) = Hn (A) ⊕
Hn (X, A), which is not true in general.
One of the great advantages of homology theory, unlike homotopy theory, is that it’s relatively
easy to compute (especially with cellular homology). Even though homology is a bit tricky
to define and the objects we’re dealing with are rather nontrivial to wrap our heads around,
it doesn’t fail to give us a plethora of results. In other words, homology theory gives us a
practical set of algebraic tools to utilize on our quest to further unlock the world of topology.
11
This result can be obtained through differential methods if we allow our embedding to be smooth. However,
with algebraic techniques, we can show this for any embedding.
12
We note that the Jordan Curve theorem is a special case of this specifically.
13
See pg. 174 in Hatcher.
14