MA3H6: Algebraic Topology

Martin Gallauer

January 22, 2024

Abstract
These are lecture notes for the course MA3H6 (Algebraic Topology) taught at the Uni-
versity of Warwick. The topics for this course are selected from Chapter 2 of Hatcher’s
“Algebraic Topology”. Much of this selection and indeed the content of this course are due
to the lecturers in previous years, including most recently Chris Lazda and John Greenlees.
Many pictures used are also due to Chris Lazda.
Please send any corrections to [email protected] or post them on the
forum for this course.

Contents
1 Introduction 2
1.1 Goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 The idea of homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 Simplicial homology 8
2.1 Δ-complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Chain complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

3 Singular homology . . . 19
3.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Low-degree interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

4 . . . and its fundamental theorems 26

4.1 Statement and immediate applications . . . . . . . . . . . . . . . . . . . . . . 26
4.2 Homotopy Invariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.3 Mayer-Vietoris: strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.4 Barycentric subdivision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

5 Applications 46
5.1 Fundamental classes for spheres . . . . . . . . . . . . . . . . . . . . . . . . . . 46
5.2 The Jordan curve theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.3 Relative homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

1
6 Degrees 55
6.1 Basic properties and examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
6.2 Antipodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
6.3 Local degrees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

7 Manifolds 61
7.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
7.2 Orientations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
7.3 Surfaces: topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
7.4 Surfaces: Homology and orientation . . . . . . . . . . . . . . . . . . . . . . . 69

8 Comparison 72
8.1 Simplicial = singular . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
8.2 CW complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
8.3 Cellular homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
8.4 Computing cellular homology . . . . . . . . . . . . . . . . . . . . . . . . . . 77

9 Thr Euler characteristic 80

9.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
9.2 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

10 Homology theories 89
10.1 Categories and functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
10.2 Axioms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
10.3 Generalized (co)homology theories . . . . . . . . . . . . . . . . . . . . . . . . 95

1 Introduction
1.1 Goals
Commentary 1.1. In very rough terms, Algebraic Topology is a set of tools for translating
questions in Topology to questions in Algebra:

Topology Algebra

In slightly more precise terms, it constructs algebraic invariants of topological spaces,

𝑋 ↦−→ 𝐴(𝑋 ).

Here, “algebraic” refers to the fact that 𝐴(𝑋 ) is a group, a vector space, etc. And “invari-
ant” indicates that if two spaces are homeomorphic (or just homotopy equivalent), then the
associated algebraic objects are isomorphic:

(1.2) 𝑋 '𝑌 ⇒ 𝐴(𝑋 )  𝐴(𝑌 )

Example 1.3. Given a topological space 𝑋 , let Z · 𝜋 0 (𝑋 ) denote the free abelian group on the
≈
set of path-connected components of 𝑋 . Of course, if 𝑓 : 𝑋 − → 𝑌 is a homeomorphism then

2
 ∼ 
it induces a bijection 𝜋 0 (𝑋 ) −
→ 𝜋0 (𝑌 ) and hence an isomorphism Z · 𝜋 0 (𝑋 ) −
→ Z · 𝜋0 (𝑌 ). The
same is true if 𝑓 is assumed to be a homotopy equivalence only.
We will encounter this algebraic invariant later on (Corollary 3.15) and identify it with the
zeroth homology group H0 (−).

Example 1.4. You will have encountered the fundamental group 𝜋1 (−) of a topological space
before. This is an important algebraic invariant although you will remember that it takes as
input not a topological space but rather a pointed topological space.
Recall that 𝜋1 (𝑋, 𝑥) is the set of homotopy classes of pointed maps (𝑆 1, ∗) → (𝑋, 𝑥), that is,
of loops in 𝑋 based at 𝑥. The group structure arises from concatenation of loops.

Example 1.5. The previous example admits a straightforward generalization. Given a pointed
space (𝑋, 𝑥) and 𝑛 ≥ 1, we denote by 𝜋𝑛 (𝑋, 𝑥) the set of homotopy classes of pointed maps
(𝑆 𝑛 , ∗) → (𝑋, 𝑥). There is a group structure on 𝜋𝑛 (𝑋, 𝑥) generalizing the one of the funda-
mental group. The resulting groups are called the homotopy groups of (𝑋, 𝑥).
For example,
(
Z 𝑙 =𝑛
(1.6) 𝜋𝑛 (𝑆 𝑙 ) 
0 𝑙 >𝑛

Commentary 1.7. Algebraic invariants may be used to tell topological spaces apart. By simple
contraposition of (1.2), if 𝐴(𝑋 )  𝐴(𝑌 ) then 𝑋 ; 𝑌 . For this to be a useful strategy, we would
like that:

1. it is ‘easy’ to compute 𝐴(−) and to tell such algebraic objects apart;

2. the algebraic invariant is ‘fine’ enough in that 𝐴(𝑋 )  𝐴(𝑌 ) actually happens ‘often’.

Let us see how the algebraic invariants considered in Examples 1.3 to 1.5 stack up against
these desiderata.

Example 1.8. The invariant Z·𝜋0 (−) performs well on the first but not so much on the second.
For example, it does not distinguish between any connected manifolds.

Example 1.9. We already mentioned (“pointed out”) the issue that the fundamental group—
and in fact, all homotopy groups—depends on a base point. Moreover, it is in general a non-
abelian group, which can make it quite tricky to tell whether or not two fundamental groups
are isomorphic. Similarly, the van Kampen theorem, while it promises a powerful tool to
compute 𝜋1 (−), involves amalgated sums and is algebraically a bit awkward to put to use.
On the positive side, the fundamental group is a complete invariant for compact surfaces.
On the negative side, it is not a fine invariant for higher dimensional spaces. In fact, for a CW-
complex 𝑋 , 𝜋1 (𝑋, 𝑥)  𝜋1 (𝑋 2, 𝑥) depends on the 2-skeleton 𝑋 2 ⊆ 𝑋 only. For illustration:

• 𝜋 1 ( [0, 1]) = ∗, 𝜋1 (𝑆 1 )  Z, 𝜋 1 (𝑆 2 ) = 0;
• 𝜋 1 can show that R2 and R3 are not homeomorphic: indeed, 𝜋 1 (R2 \{0})  𝜋1 (𝑆 1 )  Z
while 𝜋1 (R3 \{0})  𝜋1 (𝑆 2 ) = 0;

3
 • however, 𝜋 1 cannot tell R3 and R4 apart in this manner.

Example 1.10. As seen in Example 1.5, the higher homotopy groups combined remove this
shortcoming.1 They are powerful invariants, able to tell apart many topological spaces of
interest. Unfortunately, however, they are very difficult to compute. For example, it is not an
accident that (1.6) misses many cases (namely, 𝑙 < 𝑛): Most of these are unknown!

Remark 1.11. It seems from these examples that we cannot hope to simultaneously satisfy both
desiderata of Commentary 1.7. Either an invariant is easy to employ but not very powerful, or
it is able to tell many spaces apart but difficult to actually apply. We therefore see something at
play that is sometimes called the law of conservation of difficulty. This isn’t a law of nature but
rather a rule of thumb: in order to establish something difficult, some difficult mathematics
has to be done somewhere in the course of the proof.

Commentary 1.12. In this course, we will introduce and study another infinite sequence of
algebraic invariants. These are called homology groups and are denoted by H𝑛 (−), for 𝑛 ≥ 0.
Along the spectrum ‘easy to employ’—‘powerful’ they lie somewhere in the middle.2 For
example, as we will see, they are abelian groups and easier to compute than the homotopy
groups. (The homology groups of spheres are all known, and in fact we will compute them,
see Corollary 4.13.) On the other hand, they contain typically less information than those.

Commentary 1.13. The computability of homology groups comes at a price, however. Even
defining them is involved and will take us a week or so. And establishing some of their fun-
damental properties will require substantial effort.

1.2 The idea of homology

Example 1.14. Consider the following CW complex 𝑋 :

We may view the characteristic maps Φ𝛼 : [−1, 1] = 𝐷 1 → 𝑋 as defining paths ℓ𝛼 in 𝑋

from Φ𝛼 (−1) to Φ𝛼 (1).
1 The homotopy groups 𝜋𝑛 for 𝑛 ≥ 2 are also abelian.
2 Arguably, in the grand scheme of things, they are situated more towards the left end of the spectrum.

4
 Then every path in 𝑋 , say starting and ending at 𝑝 4 , is homotopic to a concatenation of
these basic paths ℓ𝛼 or their inverses. Examples:

• (ℓ5 ) −2 , the path starting at 𝑝 4 , running twice around the circle in clockwise direction,
and ending at 𝑝 4
• (ℓ4 ) −1 ℓ2 (ℓ3 ) −1 ℓ4 , the path running from 𝑝 4 to 𝑝 2 , around the ellipse counterclockwise
and returning to 𝑝 4

Commentary 1.15. In defining the fundamental group 𝜋1 (𝑋, 𝑝 4 ) we take into account the
order in which these basic paths are concatenated. As mentioned above (Commentary 1.12),
homology groups will be abelian so the following definition is a first step in the direction of
homology.

Definition 1.16. A 1-chain in 𝑋 is a formal linear combination of 1-cells

𝑛 1 ℓ1 + 𝑛 2 ℓ2 + 𝑛 3 ℓ3 + 𝑛 4 ℓ4 + 𝑛 5 ℓ5
5 Z·ℓ 
with 𝑛𝛼 ∈ Z. In other words, the group of 1-chains in 𝑋 is the free abelian group ⊕𝛼=1 𝛼
5
Z on the basis {ℓ𝛼 }.

Commentary 1.17. Some 1-chains in 𝑋 may be thought of as paths. For example, −2 · ℓ5

may be thought of as the first path considered in Example 1.14. Others, like ℓ2 + ℓ5 , don’t admit
such an interpretation. In any case we should remember that the order of ‘concatenation’ is
immaterial as far as 1-chains are concerned.
For a more accurate description of the relation between paths and 1-chains in 𝑋 see Com-
mentary 1.25.

Example 1.18. The two loops ℓ2 ℓ3−1 (based at 𝑝 2 ) and ℓ3−1 ℓ2 (based at 𝑝 3 ) are not distinguished
anymore when viewed as 1-chains. Indeed, the two linear combinations

(1.19) ℓ2 − ℓ3 and − ℓ3 + ℓ2

are equal. So, abelianization has the consequence of ‘forgetting about base-points’.

Commentary 1.20. We may now ask when a 1-chain represents a loop in 𝑋 . (Such 1-chains
will be called 1-cycles below.) In analogy with paths, the answer should be: if it enters and exits
every 0-cell the same number of times. We can formalize this as follows.

5
Definition 1.21. A 0-chain in 𝑋 is a formal linear combination of 0-cells:

𝑚 1𝑝 1 + 𝑚 2𝑝 2 + 𝑚 3𝑝 3 + 𝑚 4𝑝 4
4
with 𝑚 𝛽 ∈ Z. In other words, the group of 0-chains is the free abelian group ⊕𝛽=1 Z · 𝑝 𝛽  Z4
on the basis {𝑝 𝛽 }.

Definition 1.22. We define the boundary operator 𝜕1 that takes a basic 1-chain ℓ𝛼 to the formal
difference between start and end point of ℓ𝛼 , viewed as a 0-chain in 𝑋 . In other words, it is
the group homomorphism given on generators by:

𝜕1 : ⊕𝛼 Z · ℓ𝛼 −→ ⊕𝛽 Z · 𝑝 𝛽
ℓ1 ↦−→ 𝑝 2 − 𝑝 1
ℓ2 ↦−→ 𝑝 3 − 𝑝 2
ℓ3 ↦−→ 𝑝 3 − 𝑝 2
ℓ4 ↦−→ 𝑝 4 − 𝑝 2
ℓ5 ↦−→ 𝑝 4 − 𝑝 4 = 0

A 1-cycle in 𝑋 is a 1-chain in the kernel of 𝜕1 .

Example 1.23. • For any 𝑛 ∈ Z, the 1-chain 𝑛 · ℓ5 is a 1-cycle since 𝜕1 (𝑛 · ℓ5 ) = 𝑛 · 𝜕1 (ℓ5 ) =

𝑛 · 0 = 0.
• The 1-chain ℓ2 − ℓ3 = −ℓ3 + ℓ2 of (1.19) is a 1-cycle.

Remark 1.24. With the given basis elements for 1- and 0-chains, the boundary operator
𝜕1 : Z5 → Z4 is represented by the matrix

−1 0 0 0 0
­ 1 −1 −1 −1 0®
© ª
­0 1 1 0 0®
­ ®

«0 0 0 1 0¬

hence ker(𝜕1 ) = Z · (ℓ2 − ℓ3 ) ⊕ Z · ℓ5  Z2 . This is what we will later call the first homology
group of 𝑋 and denote by H1 (𝑋 ).

Commentary 1.25. The fundamental group of 𝑋 is (for example, by the van Kampen The-
orem) the free group on the two generators ℓ2 (ℓ3 ) −1 and ℓ5 . We therefore see in this example
that the morphism

Z ∗ Z  𝜋 1 (𝑋 ) −→ H1 (𝑋 )  Z2
ℓ2 (ℓ3 ) −1 ↦−→ ℓ2 − ℓ3
ℓ5 ↦−→ ℓ5

identifies the first homology group with the abelianization of the fundamental group. We will
see later (Theorem 3.28) that this is a general phenomenon.

6
Example 1.26. Let 𝑌 be the CW complex obtained from 𝑋 by filling in the ellipse with a new
2-cell:

More precisely, we choose the attaching map

(1.27) 𝜙1 : 𝑆 1 → 𝑋

described by the loop ℓ2 (ℓ3 ) −1 based at 𝑝 2 . At the level of 𝜋1 , attaching this new 2-cell 𝑑 1 has
the effect of killing the loop ℓ2 (ℓ3 ) −1 .

The following definition will not come as a surprise.

Definition 1.28. The group of 2-chains in 𝑌 is the free abelian group on 𝑑 1 . The boundary
operator 𝜕2 on 2-chains is given by:

𝜕2 : Z · 𝑑 1 −→ ⊕𝛼 Z · ℓ𝛼
𝑑 1 ↦−→ ℓ2 − ℓ3

Commentary 1.29. Here the image of 𝑑 1 should be thought of as the 1-cycle described by
the loop used to define the attaching map 𝜙.

Remark 1.30. If we define the first homology group of 𝑌 as the quotient

ker 𝜕1 : {1-chains} → {0-chains}

H1 (𝑌 ) =
img 𝜕2 : {2-chains} → {1-chains}

then we find that H1 (𝑌 )  Z is generated by the 1-cycle ℓ5 .

This is another instance of the phenomenon mentioned in Commentary 1.25. Indeed,
𝜋 1 (𝑌 )  𝜋1 (𝑆 1 )  Z is already abelian hence equal to its abelianization.

Example 1.31. Suppose we change the attaching map to 𝜙 0 : 𝑆 1 → 𝑋 described by the loop
ℓ2 (ℓ3 ) −1 ℓ2 (ℓ3 ) −1 to get another CW complex 𝑌 0. Then the boundary map sends the unique
2-cell 𝑑 10 to 2(ℓ2 − ℓ3 ) so that
H1 (𝑌 0)  Z ⊕ Z/2Z.

Example 1.32. We consider one last modification. Let 𝑍 be the CW complex obtained from 𝑌
by adding a second 2-cell 𝑑 2 via the same attaching map (1.27).

7
 Note that attaching this new 2-cell does not have any effect on the fundamental group so
we expect that H1 (𝑍 )  H1 (𝑌 )  Z.
The group of 2-chains in 𝑍 is the free abelian group Z · 𝑑 1 ⊕ Z · 𝑑 2 on the basis {𝑑 1, 𝑑 2 }
and the boundary operator takes both of these generators to the 1-cycle ℓ2 − ℓ3 . This means
ker(𝜕1 )
that H1 (𝑍 ) = img(𝜕 2)
is indeed the same as H1 (𝑌 )  Z. On the other hand, the kernel of the
boundary operator 𝜕2 : {2-chains} → {1-chains} is non-trivial: it is infinite cyclic generated
by 𝑑 1 −𝑑 2 . This is what we will eventually call the second homology group H2 (𝑍 ) = ker(𝜕2 ) 
Z.

Commentary 1.33. What is the upshot of these examples?

• You will now probably be in a position to guess how the definition of homology groups
of any CW complex 𝑋 will look like. Namely, one starts by defining the group of 𝑛-
chains in 𝑋 as the free abelian group on the 𝑛-cells 𝑒𝛼𝑛 in 𝑋 . The boundary operator
𝜕𝑛 : {𝑛-chains} → {(𝑛 − 1)-chains} takes a generator 𝑒𝛼𝑛 to an (𝑛 − 1)-cycle described
in terms of the attaching map for 𝑒𝛼𝑛 . (At this point of the discussion it is admittedly
not clear at all yet how this description looks like in general.) The 𝑛th homology group
H𝑛 (𝑋 ) is then the quotient ker(𝜕𝑛 )/img(𝜕𝑛+1 ).
More precisely, this is what we will call the cellular homology of 𝑋 in Section 8.3.
• The examples also suggest one way to think about the homology of a space 𝑋 . Namely,
roughly, the 𝑛th homology group H𝑛 (𝑋 ) measures ‘how many’ 𝑛-dimensional holes
there are in 𝑋 .

Commentary 1.34. The idea of using algebraic structures to distinguish non-homeomorphic

topological spaces was first systematically developed by Henri Poincaré in his series of papers
on “Analysis Situs” (1899–1904). For this reason these papers are commonly taken as the birth
of Algebraic Topology. And quite an impressive birth they were! Poincaré introduced not
only the fundamental group and a form of simplicial homology (see section 2 below), but also
formulated what is now known as Poincaré duality (see the fourth-year module MA4J7) and
the Poincaré Conjecture (now a celebrated theorem), among many other things.

2 Simplicial homology
As mentioned in the introduction, the definition of the homology groups of a topological
space is rather involved. For this reason we start in this section with a simplified version, called
simplicial homology, that takes as input not a topological space but rather a Δ-complex. In the
next section 3 we go on to study the generalized version applicable to all topological spaces.

8
2.1 Δ-complexes
Δ-complexes are topological spaces built out of simplices: points, lines, triangles, tetrahedrons,
etc. To make this precise, let us introduce:

Definition 2.1. The (standard) 𝑛-simplex Δ𝑛 ⊆ R𝑛+1 is the subspace

( 𝑛
)
∑︁
Δ𝑛 := (𝑥 0, . . . , 𝑥𝑛 ) | 𝑥𝑖 = 1, 𝑥𝑖 ≥ 0 .
𝑖=0

More generally, any homeomorphic space will also be called an 𝑛-simplex. Its vertices will
often be denoted 𝑣 0 , 𝑣 1 , . . . , 𝑣𝑛 . These correspond to the points (1, 0, . . . , 0), (0, 1, 0, . . . , 0), . . . ,
(0, . . . , 0, 1) in the standard 𝑛-simplex.

Example 2.2.
𝑥1 𝑥2
𝑥1
𝑥0 𝑥0 𝑥0

Δ0 Δ1 Δ2

Definition 2.3. The 𝑗th face of the 𝑛-simplex is the subspace 𝜕 𝑗 Δ𝑛 ⊆ Δ𝑛 of points whose 𝑗th
coordinate vanishes. That is,
( 𝑛
)
∑︁
𝜕 𝑗 Δ𝑛 = (𝑥 0, . . . , 𝑥𝑛 ) | 𝑥𝑖 = 1, 𝑥𝑖 ≥ 0 ∀𝑖, 𝑥 𝑗 = 0 .
𝑖=1

Note that 𝜕 𝑗 Δ𝑛 is canonically an (𝑛 − 1)-simplex with vertices 𝑣 0, . . . , 𝑣b𝑗 , . . . , 𝑣𝑛 .

Example 2.4. Here, the three faces of the standard 2-simplex are shown in color:

𝑥2
𝑥1

𝑥0

Note that the 𝑗th face is the one opposite the 𝑗th vertex.

Definition 2.5. The boundary of Δ𝑛 is the union of its (𝑛+1) faces, and will be denoted by 𝜕Δ𝑛 .

Example 2.6. Since the unique face 𝜕0 Δ0 is the empty set we have 𝜕Δ0 = ∅.

9
Exercise 2.7. Show that the 𝑛-simplex Δ𝑛 is homeomorphic to the 𝑛-ball
( 𝑛
)
2
∑︁
𝑛
𝐷 = (𝑦1, . . . , 𝑦𝑛 ) | 𝑦𝑖 ≤ 1 ⊂ R𝑛
𝑖=1

and its boundary 𝜕Δ𝑛 homeomorphic to the sphere 𝑆 𝑛−1 .

Definition 2.8. A Δ-complex is a topological space obtained inductively as follows:

1. Start with a (discrete) collection of 0-simplices, that is, points. This is the 0-skeleton 𝑋 0 .
2. Inductively, the 𝑛-skeleton 𝑋 𝑛 is obtained from 𝑋 𝑛−1 by attaching 𝑛-simplices Δ𝑛𝛼 whereby
each face 𝜕𝑖 Δ𝑛𝛼 gets identified with an (𝑛 − 1)-simplex Δ𝑛−1
𝛽
in 𝑋 𝑛−1 .

3. If 𝑋 = 𝑋 𝑘 and 𝑘 is minimal with this property then 𝑋 is of dimension 𝑘. More generally

we can have 𝑋 = ∪𝑛 𝑋 𝑛 in which case a subspace 𝑈 ⊆ 𝑋 is open iff 𝑈 ∩ 𝑋 𝑛 ⊆ 𝑋 𝑛 is open
for all 𝑛.

Example 2.9. Start with a single point 𝑋 0 = Δ0 . Attach now a single 1-simplex Δ1 , in the
only possible way. Namely, both boundary points are identified with the 0-skeleton:

In other words, we get a circle.

Example 2.10. Now we start with two points 𝑋 0 = {𝑝, 𝑞}. We then attach two 1-simplices
𝑎, 𝑏 as follows:
𝑎

𝑝 𝑞

The arrows indicate that the ‘line 𝑎 runs from 𝑝 to 𝑞’ (and similarly for 𝑏). More precisely, this
means that the 0th face of 𝑎 (the end point) gets identified with 𝑞, and the 1st face (the start
point) with 𝑝.
Note that we get again the circle. This is something to keep in mind: Δ-complexes can
typically be built out of simplices in different ways.

Example 2.11. Moving to 2-dimensional spaces, the filled-in square is a Δ-complex, as de-
picted:

𝐹
𝐺

10
It has four 0-simplices, five 1-simplices and two 2-simplices 𝐹 and 𝐺. Note how the arrows
specify an ordering of the vertices of each 2-simplex, and therefore a homeomorphism with Δ2 .
In particular, there is no ambiguity about which face is which. (For example, check that the
first face of 𝐹 is the diagonal. Or that the 0th face of 𝐺 is the right vertical line.)

Example 2.12. The torus T = 𝑆 1 × 𝑆 1 is obtained by identifying opposite edges of a square:

It is a Δ-complex, as follows:

𝑝 𝑎 𝑝

𝐹
𝑏 𝑏
𝑐 𝐺

𝑝 𝑎 𝑝

Here there is a single 0-simplex 𝑝, three 1-simplices 𝑎, 𝑏 and 𝑐, and two 2-simplices 𝐹 and 𝐺.

Exercise 2.13. See the first exercise sheet for further examples.

Remark 2.14. A Δ-complex is essentially just a combinatorial datum. Namely, it is deter-

mined (up to homeomorphism) by the sets of 𝑛-simplices 𝑆𝑛 , 𝑛 ≥ 0, together with the at-
taching rules. These are ‘face’ maps 𝑑𝑖𝑛 : 𝑆𝑛 → 𝑆𝑛−1 (for 0 ≤ 𝑖 ≤ 𝑛) specifying that 𝜕𝑖 Δ𝑛𝛼 gets
identified with Δ𝑑𝑛−1
𝑛 (𝛼) .
𝑖
These maps are not arbitrary but satisfy the relation (whenever 𝑖 < 𝑗)

(2.15) 𝑑𝑖𝑛−1 ◦ 𝑑 𝑛𝑗 = 𝑑 𝑛−1 𝑛

𝑗−1 ◦ 𝑑𝑖 ,

as you can easily check. Such a combinatorial datum 𝑆 = (𝑆 •, 𝑑 •• ) is called a Δ-set or semi-
simplicial set.
Given a Δ-set 𝑆 we denote the associated Δ-complex by |𝑆 |. (It is sometimes called its
geometric realization.) To be pedantic (but one rarely is), a Δ-complex is really a topological
space 𝑋 together with a homeomorphism 𝑋 ≈ |𝑆 | for some Δ-set 𝑆. The latter could then be
called a Δ-complex structure on 𝑋 . As we saw in Examples 2.9 and 2.10, a topological space can
admit distinct Δ-complex structures.

Example 2.16. In the case of the torus (Example 2.12) we found

𝑆 0 = {𝑝}, 𝑆 1 = {𝑎, 𝑏, 𝑐}, 𝑆 2 = {𝐹, 𝐺 }

with face maps that can be read off the picture. For example,

𝑑 02 (𝐹 ) = 𝑎, 𝑑 02 (𝐺) = 𝑏; 𝑑𝑖1 (𝑎) = 𝑑𝑖1 (𝑏) = 𝑑𝑖1 (𝑐) = 𝑝 for 𝑖 = 0, 1.

11
Commentary 2.17 (unimportant). You might have encountered simplicial complexes or CW-
complexes before. Every simplicial complex is automatically a Δ-complex, and every Δ-complex
is automatically a CW-complex. So, the relation between these is as follows:

{simplicial complexes} ⊆ {Δ-complexes} ⊆ {CW-complexes}

It turns out (we will not prove nor use that) that a topological space admits the structure of a
Δ-complex iff it admits the structure of a simplicial complex (albeit with different simplices in
general). So, in some sense, the first inclusion above is an equality.3 However, this is not true
for the second inclusion.

2.2 Homology
Commentary 2.18. In Section 1.2 we defined homology in the context of some simple CW
complexes. A key ingredient was the definition of certain boundary operators. It turns out
that moving from cells to simplices and from CW complexes to Δ-complexes makes things
even more transparent. Let us now see how.

• Start with the standard 1-simplex:

𝑣0 𝑣1

Taking a cue from Definition 1.22 we view its ‘oriented’ boundary as the formal differ-
ence between head and tail:
𝑣1 − 𝑣0. 4

• Similarly when passing to 2-simplices,

𝑣2

𝑏 𝑎

𝑣0 𝑣1
𝑐

we define its ‘oriented’ boundary as

𝑎 − 𝑏 + 𝑐. 5
3 You might now wonder why we don’t use simplicial complexes instead. The reason is that typically many
more simplices are required than for Δ-complexes, making the homology computations below more involved.
4 The difference 𝑣 − 𝑣 would work equally well.
0 1
5 Again, we have made a choice here. Orienting the triangle in a clockwise fashion works equally and would
have given the negative of this expression: −𝑎 + 𝑏 − 𝑐

12
 • The general formula for an 𝑛-simplex 𝑠 that emerges is then
𝑛
∑︁
(2.19) 𝜕𝑛 (𝑠) = (−1)𝑖 𝑑𝑖𝑛 (𝑠),
𝑖=0

the alternating sum of its faces.

In fact, this whole section is nothing but an elaboration of section 1.2 in the context of Δ-
complexes. With the little input given now, you should be able to define the (simplicial)
homology groups of a Δ-complex. Try it before reading on!

Definition 2.20. Let 𝑆 be a Δ-set.

1. The group of 𝑛-chains in 𝑆 is the free abelian group on 𝑆𝑛 . It is denoted by Δ𝑛 (𝑆).

2. The boundary operator 𝜕𝑛 : Δ𝑛 (𝑆) → Δ𝑛−1 (𝑆) is the homomorphism given on the gen-
erators 𝑠 ∈ 𝑆𝑛 by the formula in (2.19). (By convention, Δ−1 (𝑆) = 0 and 𝜕0 is the zero
map.)

Example 2.21. Recall from Example 2.9 the ‘minimal’ Δ-complex structure on 𝑆 1 . The only
possibly interesting boundary operator is 𝜕1 : Zℓ = Δ1 (𝑆) → Δ0 (𝑆) = Z𝑝. It takes the genera-
tor ℓ to 𝜕1 (ℓ) = 𝑑 0 (ℓ) − 𝑑 1 (ℓ) = 𝑝 − 𝑝 = 0 and hence is the zero map.

Example 2.22. Consider again the Δ-complex structure on the torus from Example 2.12. For
instance, we find that

𝜕2 (𝐹 − 𝐺) = 𝜕2 (𝐹 ) − 𝜕2 (𝐺) = (𝑏 − 𝑐 + 𝑎) − (𝑎 − 𝑐 + 𝑏) = 0

so we have found a non-trivial 2-chain in the kernel of 𝜕2 .

Definition 2.23. Let 𝑆 be a Δ-set. We define the following subgroups of Δ𝑛 (𝑆), for 𝑛 ≥ 0:

• the group of 𝑛-cycles 𝑍𝑛 (𝑆) := ker(𝜕𝑛 )

• the group of 𝑛-boundaries 𝐵𝑛 (𝑆) := img(𝜕𝑛+1 )
• the 𝑛th simplicial homology group is the quotient:

𝑍𝑛 (𝑆)
H𝑛 (𝑆) :=
𝐵𝑛 (𝑆)

Elements of this group are called homology classes.

Commentary 2.24. For the last expression to make sense we need to know that 𝐵𝑛 (𝑆) ⊆
𝑍𝑛 (𝑆). We will indeed prove this below in Lemma 2.34, but we will also see it the next couple
of examples directly. (In section 1.2 we also defined the homology as the quotient of cycles
by boundaries. In each case, it could be easily verified that this made sense, to wit, that the
𝑛-boundaries are all 𝑛-cycles.)

13
Example 2.25. Let 𝑆 be the Δ-set from Example 2.9. We saw in Example 2.21 that the bound-
ary operator 𝜕1 = 0 vanishes. It follows that:
𝑍 1 (𝑆) = 𝑍 0 (𝑆) = Z, 𝐵𝑛 (𝑆) = 0 ∀𝑛,
and therefore H0 (𝑆) = H1 (𝑆) = Z and all other homology groups 0.

Example 2.26. Let 𝑇 be the Δ-set from Example 2.10. We may represent the situation as
follows:
−1 1
 
1 −1
· · · → 0 → Z𝑎 ⊕ Z𝑏 −−−−−−−→ Z𝑝 ⊕ Z𝑞 → 0
𝜕1

with all other 𝑛-chains trivial. We deduce that

𝐵 0 (𝑇 ) = Z(𝑝 − 𝑞) ⊂ Z𝑝 ⊕ Z𝑞 = 𝑍 0 (𝑇 )
𝐵 1 (𝑇 ) = 0 ⊂ Z(𝑎 + 𝑏) = 𝑍 1 (𝑇 )
and therefore 𝐻 0 (𝑇 ) = 𝐻 1 (𝑇 ) = Z and all other homology groups 0.

Remark 2.27. We just saw that the two Δ-sets 𝑆 and 𝑇 of Examples 2.25 and 2.26 have the
same homology groups. This is not a coincidence (cf. also exercise 1.5 on sheet 1). We showed
in Examples 2.9 and 2.10 that their geometric realization is the same, namely the 1-sphere.
And it turns out, although it isn’t clear at all at this point, that the homology is an invariant
of the geometric realization. We will establish this in Corollary 8.5 and thereby justify the
following definition.

Definition 2.28. If 𝑋 is a topological space with a Δ-complex structure |𝑆 | ≈ 𝑋 we define its

𝑛th simplicial homology group to be
H𝑛Δ (𝑋 ) := H𝑛 (𝑆).
Commentary 2.29. Our informal discussion of homology, Commentary 1.33, lets us think
about these groups as follows. An 𝑛-cycle is an 𝑛-chain that could be the boundary of an (𝑛+1)-
chain. Therefore, a nonzero homology class detects an 𝑛-dimensional hole: something that
could be but isn’t a boundary.
(
Δ 1 Z : 𝑛 = 0, 1,
Example 2.30. So, we have seen that H𝑛 (𝑆 ) =
0 : else.

Example 2.31. Let us compute the simplicial homology of the torus, see Example 2.12. We
may represent the situation as follows:
1 1
 
1 1
−1 −1 (0 0 0)
· · · → 0 → Z𝐹 ⊕ Z𝐺 −−−−−−−→ Z𝑎 ⊕ Z𝑏 ⊕ Z𝑐 −−−−−−→ Z𝑝 → 0
𝜕2 𝜕1

We see that the 2-cycle from Example 2.22 generates the 2-cycles and H2Δ (T ) = Z. Of course,
H0Δ (T ) = Z. And in degree 1 we have:
𝐵 1 = Z(𝑎 + 𝑏 − 𝑐) ⊂ Z𝑎 ⊕ Z𝑏 ⊕ Z𝑐 = 𝑍 1

14
so that H1Δ (T ) = Z ⊕ Z.

Exercise 2.32. Compute the simplicial homology of the square, see Example 2.11. In fact,
before you start, ask yourself what the simplicial homology should be and then verify that.

Commentary 2.33. The first problem sheet asks you to compute more examples of simplicial
homology groups. At this point, it’s probably good to observe that once you’ve written down
the groups of chains and the boundary operators between them, computing the simplicial
homology is an entirely mechanical process. If you don’t have a preferred method already I
recommend the following one:
1. First, determine the Smith Normal Form for each boundary operator in matrix form.
(See the document with Preliminaries on the moodle page.)
2. Apply the following result: Given two matrices
𝐴 𝐵
Z𝑙 Z𝑚 Z𝑛

with 𝐵𝐴 = 0, we have

ker(𝐵)/img(𝐴)  ⊕𝑖=1 Z/𝑎𝑖 ⊕ Z𝑚−𝑟 −𝑠

𝑟

where
• 𝑎𝑖 are the invariant factors of 𝐴,
• 𝑟 = rk(𝐴), 𝑠 = rk(𝐵).
(This is Exercise 1.4 on sheet 1.)

We still need to justify Definition 2.23.

Lemma 2.34. Let 𝑆 be a Δ-set. Then 𝜕𝑛 ◦ 𝜕𝑛+1 = 0. Equivalently, 𝐵𝑛 (𝑆) ⊆ 𝑍𝑛 (𝑆).

Of course, we have set up things precisely so that this holds. For example, here is what hap-
pens when you apply twice the boundary operator to the standard 2-simplex Δ2 = [𝑣 0, 𝑣 1, 𝑣 2 ].
𝑣2
𝜕2 −[𝑣 0, 𝑣 2 ] [𝑣 1, 𝑣 2 ] 𝜕1 − +

+ −
[𝑣 0, 𝑣 1, 𝑣 2 ] [𝑣 0, 𝑣 1 ] 𝑣0 − + 𝑣1

In the resulting formal sum every 0-simplex appears once with a plus and once with a minus
sign, the two canceling each other out. This is what happens in general:

Proof. It is clear from the definitions of cycles and boundaries that the two statements are
equivalent. So we only prove the first. For this, let 𝑠 ∈ 𝑆𝑛+1 . Then
𝑛+1
! 𝑛 ∑︁
𝑛+1
∑︁ ∑︁
𝑗 𝑛+1
𝜕𝑛 𝜕𝑛+1 (𝑠) = 𝜕𝑛 (−1) 𝑑 𝑗 (𝑠) = (−1)𝑖+𝑗 𝑑𝑖𝑛𝑑 𝑛+1
𝑗 (𝑠).
𝑗=0 𝑖=0 𝑗=0

15
Fix a pair 0 ≤ 𝑖 < 𝑗 ≤ 𝑛 + 1. By (2.15), the two summands 𝐴 := (−1)𝑖+𝑗 𝑑𝑖𝑛𝑑 𝑛+1 𝑗 (𝑠) and
𝐵 := (−1)𝑖+𝑗−1𝑑 𝑛𝑗−1𝑑𝑖𝑛+1 cancel out. And every summand is of the form 𝐴 or 𝐵 (and not both)
so that the sum vanishes.
If you like it more formally, we can do this by breaking up the sum in two:
𝑛 ∑︁
∑︁ 𝑛+1 ∑︁ ∑︁
(−1)𝑖+𝑗 𝑑𝑖 𝑑 𝑗 (𝑠) = (−1)𝑖+𝑗 𝑑𝑖 𝑑 𝑗 (𝑠) + (−1)𝑖+𝑗 𝑑𝑖 𝑑 𝑗 (𝑠)
𝑖=0 𝑗=0 0≤𝑖< 𝑗 ≤𝑛+1 0≤ 𝑗 ≤𝑖 ≤𝑛
∑︁ ∑︁
𝑖+𝑗
= (−1) 𝑑 𝑗−1𝑑𝑖 (𝑠) + (−1)𝑖+𝑗 𝑑𝑖 𝑑 𝑗 (𝑠)
0≤𝑖< 𝑗 ≤𝑛+1 0≤ 𝑗 ≤𝑖 ≤𝑛
∑︁ ∑︁
𝑖+𝑗−1
= (−1) 𝑑 𝑗 𝑑𝑖 (𝑠) + (−1)𝑖+𝑗 𝑑𝑖 𝑑 𝑗 (𝑠)
0≤𝑖 ≤ 𝑗 ≤𝑛 0≤ 𝑗 ≤𝑖 ≤𝑛

=0


Commentary 2.35. While mathematicians had spoken of “cycles modulo boundaries” be-
fore, it seems like it was Poincaré who introduced formal sums of simplices and the boundary
operator, leading to a precise definition of homology classes. However, he did not see the
importance of studying the homology groups themselves as invariants of topological spaces.
This was only noticed years later by Emmy Noether and, independently, Leopold Vietoris6
and Walther Mayer. Mayer (1929) is also credited with introducing the term “chain complex”
that we are about to discuss.

2.3 Chain complexes

Commentary 2.36. Just now we have been computing homology groups of Δ-sets in a two-
step process, if you like:
𝑆 ↦→ (Δ• (𝑆), 𝜕• ) ↦→ H• (𝑆)
It will be useful to discuss the second step of this process independently of the situation here
since it will recur repeatedly throughout the course. We will see that the input should be
viewed as a chain complex and the step can be described as taking the homology of a chain
complex.

Definition 2.37. A chain complex 𝐶 • = (𝐶 •, 𝜕• ) is a family of abelian groups 𝐶𝑛 for 𝑛 ∈ Z and

maps7 (called differentials) 𝜕𝑛 : 𝐶𝑛 → 𝐶𝑛−1 such that 𝜕𝑛 ◦ 𝜕𝑛+1 = 0 for all 𝑛.

Remark 2.38. We often depict a chain complex as a sequence of abelian groups and maps
between them, like so:
𝜕𝑛+1 𝜕𝑛
· · · → 𝐶𝑛+1 −−−→ 𝐶𝑛 −−→ 𝐶𝑛−1 → · · ·
This should remind you of the pictures in, say, Examples 2.26 and 2.31. Of course, that’s not a
coincidence:
6 Did you know that Vietoris was the oldest verified Austrian man ever? He lived in three different centuries
and died at the age of 110.
7 Unless mentioned otherwise, all maps between (abelian) groups are group homomorphisms.

16
Example 2.39. Let 𝑆 be a Δ-set. We associated in Definition 2.20 abelian groups Δ𝑛 (𝑆) = Z𝑆𝑛
and maps 𝜕𝑛 : Δ𝑛 (𝑆) → Δ𝑛−1 (𝑆) that satisfy 𝜕𝑛 ◦ 𝜕𝑛+1 = 0, by Lemma 2.34. At least we did this
for 𝑛 ≥ 0. If we set Δ𝑛 (𝑆) = 0 for 𝑛 < 0 with zero maps between them we get a chain complex
Δ• (𝑆) that is called the simplicial chain complex associated with 𝑆.

Convention 2.40. In general if we only define 𝐶𝑛 for 𝑛 in some interval [𝑎, 𝑏] then it’s un-
derstood that 𝐶𝑛 = 0 for all 𝑛 ∉ [𝑎, 𝑏].

Example 2.41. Let 𝜄 : Z → R be the inclusion of the integers inside the real numbers. We
may view this as a chain complex placed in degrees, say, 1 and 0:
𝜕1 =𝜄
0 → Z −−−→ R → 0
Definition 2.42. Let 𝐶 • be a chain complex.
• The 𝑛-cycles are 𝑍𝑛 (𝐶 • ) = ker(𝜕𝑛 : 𝐶𝑛 → 𝐶𝑛−1 ).
• The 𝑛-boundaries are 𝐵𝑛 (𝐶 • ) = img(𝜕𝑛+1 : 𝐶𝑛+1 → 𝐶𝑛 ).
Note that since 𝜕𝑛 ◦ 𝜕𝑛+1 = 0 we have 𝐵𝑛 ⊆ 𝑍𝑛 .
• The 𝑛th homology group is H𝑛 (𝐶 • ) := 𝐵𝑛 .
𝑍𝑛

• If H𝑛 (𝐶 • ) = 0 one also says that the complex 𝐶 • is exact in degree 𝑛.

Example 2.43. For the simplicial chain complex Δ• (𝑆) of a Δ-set 𝑆 we recover the 𝑛-cycles,
𝑛-boundaries, and the 𝑛th homology group of 𝑆:
𝑍𝑛 (Δ• (𝑆)) = 𝑍𝑛 (𝑆), 𝐵𝑛 (Δ• (𝑆)) = 𝐵𝑛 (𝑆), H𝑛 (Δ• (𝑆)) = H𝑛 (𝑆)
Example 2.44. For the chain complex of Example 2.41 we get
(
R/Z : 𝑛 = 0
H𝑛 =
0 :𝑛 ≠0

The group R/Z can be identified with 𝑆 1 (viewed as a subgroup of the complex numbers with
multiplication), via the map exp(2𝜋𝑖 · −) : R → 𝑆 1 .

Example 2.45. Let 𝐶 ∞ (R3 ) be the group of smooth functions R3 → R, or smooth scalar fields
on R3 . Then 𝑉 ∞ (R3 ) = 𝐶 ∞ (R3 ) 3 is the group of smooth vector fields on R3 . The operators
div, grad and curl can be viewed as homomorphisms
div : 𝑉 ∞ (R3 ) → 𝐶 ∞ (R3 ),
grad : 𝐶 ∞ (R3 ) → 𝑉 ∞ (R3 ),
curl : 𝑉 ∞ (R3 ) → 𝑉 ∞ (R3 ).
The facts that div ◦ curl = 0 and curl ◦ grad = 0 mean that we get a chain complex
grad curl div
0 → 𝐶 ∞ (R3 ) −→ 𝑉 ∞ (R3 ) −→ 𝑉 ∞ (R3 ) −→ 𝐶 ∞ (R3 ) → 0
which we think of as being concentrated in [−3, 0]. We have H0 = R, the constant functions,
and you might have learned in Vector Calculus that H−3 = H−2 = H−1 = 0 (Poincaré Lemma).

17
Example 2.46. Let 𝐶 • be a chain complex with all differentials the zero maps. Then 𝑍𝑛 = 𝐶𝑛
and 𝐵𝑛 = 0 so that H𝑛 = 𝐶𝑛 .

Commentary 2.47. Whenever you define new objects it is a good practice (especially in
algebraic topology) to ask: what are the morphisms between these objects? For example, soon
after defining groups one also introduces group homomorphisms. After vector spaces linear
transformations. After topological spaces continuous maps. And so on.
We do the same for chain complexes.

Definition 2.48. Let (𝐶 •, 𝜕• ) and (𝐶 •0 , 𝜕•0 ) be two chain complexes. A chain map 𝑓• : 𝐶 • → 𝐶 •0
is a family of maps

𝑓𝑛 : 𝐶𝑛 → 𝐶𝑛0 , such that for all 𝑛 : 𝜕𝑛0 ◦ 𝑓𝑛 = 𝑓𝑛−1 ◦ 𝜕𝑛 .

In pictures we have ‘commuting squares’:

𝜕𝑛+1 𝜕𝑛
··· 𝐶𝑛+1 𝐶𝑛 𝐶𝑛−1 ···
𝑓𝑛+1 𝑓𝑛 𝑓𝑛−1
0
0
𝜕𝑛+1 𝜕𝑛0
··· 𝐶𝑛+1 𝐶𝑛0 0
𝐶𝑛−1 ···

Lemma 2.49. A chain map 𝑓• : 𝐶 • → 𝐶 •0 restricts to maps

• 𝑓𝑛 : 𝑍𝑛 (𝐶 • ) → 𝑍𝑛 (𝐶 •0 ), as well as
• 𝑓𝑛 : 𝐵𝑛 (𝐶 • ) → 𝐵𝑛 (𝐶 •0 ),

and hence descends to maps 𝑓𝑛 : H𝑛 (𝐶 • ) → H𝑛 (𝐶 •0 ).

Proof. See exercise sheet 1, problem 1.4. 

Definition 2.50. Let 𝑆 = (𝑆 •, 𝑑 • ) and 𝑆 0 = (𝑆 •0 , 𝑑 •0 ) be two Δ-sets. A map of Δ-sets 𝑓 : 𝑆 → 𝑆 0 is

a family of maps 𝑓𝑛 : 𝑆𝑛 → 𝑆𝑛0 such that for all 0 ≤ 𝑖 ≤ 𝑛,

𝑑𝑖0 ◦ 𝑓𝑛 = 𝑓𝑛−1 ◦ 𝑑𝑖 : 𝑆𝑛 → 𝑆𝑛−1

0
.

Lemma 2.51. A map of Δ-sets 𝑓• : 𝑆 → 𝑆 0 induces a chain map 𝑓• : Δ• (𝑆) → Δ• (𝑆 0).

Proof. We define 𝑓𝑛 : Δ𝑛 (𝑆) → Δ𝑛 (𝑆 0) on generators 𝑠 ∈ 𝑆𝑛 by

Z𝑆𝑛 −→ Z𝑆𝑛0
𝑠 ↦−→ 𝑓𝑛 (𝑠)

We now need to verify that these maps commute with the boundary operators. For this we

18
show:

𝜕𝑛0 ◦ 𝑓𝑛 (𝑠) = 𝜕𝑛0 (𝑓𝑛 (𝑠))

𝑛
∑︁
= (−1)𝑖 𝑑𝑖0 (𝑓𝑛 (𝑠)) definition of 𝜕𝑛0
𝑖=0
𝑛
∑︁
= (−1)𝑖 𝑓𝑛−1 (𝑑𝑖 (𝑠)) definition of map of Δ-sets
𝑖=0
𝑛
!
∑︁
= 𝑓𝑛−1 𝑖
(−1) 𝑑𝑖 (𝑠) linearity of 𝑓𝑛−1
𝑖=0
= 𝑓𝑛−1 ◦ 𝜕𝑛 (𝑠) definition of 𝜕𝑛

Remark 2.52. Combining the last two lemmas we see that every map of Δ-sets induces a map
in simplicial homology.

Exercise 2.53. What do you think is a/the appropriate notion of a morphism between Δ-
complexes?

3 Singular homology . . .
Commentary 3.1. In the previous section we defined the simplicial homology for Δ-complexes
and computed it in some examples. Several problems arise with this tool: First, topological
spaces do not typically come with an obvious Δ-complex structure. In fact, some topological
spaces admit no such structure at all. Secondly, even if a given space admits such a structure,
it might not be unique. And unfortunately we didn’t prove that the simplicial homology is
independent of the choice one then has to make. Thirdly, a (continuous) map 𝑓 : 𝑋 → 𝑌
between Δ-complexes doesn’t induce a map in simplicial homology, at least not in an obvious
way (see also Exercise 2.53).
In this section we are going to define a variant which avoids these difficulties (and even-
tually will allow us to prove the independence in the paragraph above). The idea of singular
homology is to allow all (as always, continuous) maps 𝜎 : Δ𝑛 → 𝑋 as simplices. Note the de-
parture: 𝑋 is not neatly built out of these simplices by identifying certain faces. For example,
𝜎 could be the constant map. To mark this departure we call such simplices singular, thus the
name of this homology theory.

3.1 Definition
Definition 3.2. Let 𝑋 be a topological space and 𝑛 ≥ 0. A singular 𝑛-simplex in 𝑋 is a contin-
uous map 𝜎 : Δ𝑛 → 𝑋 .

Example 3.3. 1. A singular 0-simplex Δ0 = {∗} → 𝑋 is nothing but a point 𝑥 ∈ 𝑋 . We

will sometimes identify the two.

19
 2. A singular 1-simplex 𝛾 : Δ1 = [0, 1] → 𝑋 is nothing but a path in 𝑋 from 𝛾 (0) to 𝛾 (1).
3. If 𝑋 is a Δ-complex then any 𝑛-simplex in 𝑋 gives rise to a singular 𝑛-simplex in 𝑋 .

Commentary 3.4. One good thing about the notion of singular simplices is that we already
know how to define its ‘oriented’ boundary since we know it for the standard 𝑛-simplex.
Namely, the ‘oriented’ boundary of 𝜎 : Δ𝑛 → 𝑋 should be:
𝑛
∑︁
(3.5) (−1)𝑖 𝜎 | 𝜕𝑖 Δ𝑛 .
𝑖=0

If we identify the 𝑖th face with the standard (𝑛 − 1)-simplex, as in Definition 2.3, then this
expression becomes a formal linear combination of singular (𝑛 − 1)-simplices in 𝑋 .

Definition 3.6. Let 𝑋 be a topological space and 𝑛 ≥ 0.

1. The group of singular 𝑛-chains in 𝑋 is the free abelian group on the singular 𝑛-simplices,
denoted 𝐶𝑛 (𝑋 ).
2. The boundary operator 𝜕𝑛 : 𝐶𝑛 (𝑋 ) → 𝐶𝑛−1 (𝑋 ) is the homomorphism defined on a basis
element 𝜎 : Δ𝑛 → 𝑋 by the formula in (3.5).

The same proof as in Lemma 2.34 then gives:

Lemma 3.7. Let 𝑋 be a topological space. Then 𝜕𝑛 ◦ 𝜕𝑛+1 : 𝐶𝑛+1 (𝑋 ) → 𝐶𝑛−1 (𝑋 ) is the zero map.

In other words, (𝐶 • (𝑋 ), 𝜕• ) is a chain complex and we can make the following definition.

Definition 3.8. Let 𝑋 be a topological space. The singular chain complex of 𝑋 is the com-
plex (𝐶 • (𝑋 ), 𝜕• ). The singular homology groups of 𝑋 are

H𝑛 (𝑋 ) := H𝑛 (𝐶 • (𝑋 )).

Example 3.9. Let 𝑋 = ∗ be a point. For each 𝑛 ≥ 0 there is a unique singular 𝑛-simplex, the
constant map 𝑐𝑛 : Δ𝑛 → ∗. Therefore the singular chain complex looks as follows:
𝜕𝑛 𝜕2 𝜕1
· · · → Z −−→ Z → · · · → Z −→ Z −→ Z → 0

Note that (
: 𝑛 > 0 even
𝑛 𝑛
∑︁ ∑︁ 𝑐𝑛−1
𝜕𝑛 (𝑐𝑛 ) = (−1)𝑛𝑐𝑛 | 𝜕𝑖 Δ𝑛 = (−1)𝑛𝑐𝑛−1 =
𝑖=0 𝑖=0
0 : else
and hence
 0  0
··· → Z −
→ Z→
− Z− − Z → 0.
→ Z→
We conclude that (
Z :𝑛 =0
H𝑛 (∗) =
0 :𝑛 ≠0

20
Commentary 3.10. Here we can already glimpse one of the disadvantages of singular ho-
mology. Even for such a simple space the singular chain complex is non-trivial. For larger
spaces, the number of singular 𝑛-simplices is so enormous that hands-on computations with
the singular chain complex are rarely feasible. We will now give interpretations of H0 and H1
of a topological space but eventually we will need some high-powered machine that allows us
to compute the singular homology groups without actually working with the singular chain
complex directly. This we will do in Section 4.

Commentary 3.11. The singular (in contrast to the simplicial) version of homology as defined
here is due to Samuel Eilenberg (1944) building on earlier work of Salomon Lefschetz (1933).
The improvement of the former over the latter consists in the realization that the simplices
should not be viewed as being oriented but rather as having an ordering on the vertices only.

3.2 Low-degree interpretation

Commentary 3.12. Let 𝛾 : Δ1 = [0, 1] → 𝑋 be a singular 1-simplex, that is a path, from 𝑥 =
𝛾 (0) to 𝑦 = 𝛾 (1), as in Example 3.3. Then we have

𝜕1 (𝛾) = 𝑦 − 𝑥 ∈ 𝐶 0 (𝑋 )

and we conclude that 𝑥 and 𝑦 are homologous, that is, their difference differs by a boundary.
(We also write 𝑥 ∼ 𝑦.) What is not clear, but we will prove in a moment, is that the converse
holds as well: Two points (viewed as singular 0-chains) are homologous if and only if there is
a path connecting them.

Lemma 3.13. Let 𝑋 be non-empty and path-connected.8 Then H0 (𝑋 ) = Z.

Proof. Define the degree homomorphism deg : 𝐶 0 (𝑋 ) → Z to send every basis element 𝑥 ∈ 𝑋
(that is, a singular 0-simplex) to 1 ∈ Z. We now show the following points:
• The map deg is surjective. Indeed, since 𝑋 is non-empty there exists 𝑥 ∈ 𝑋 with image a
generator 1 ∈ Z.
• We have an inclusion 𝐵 0 (𝑋 ) ⊆ ker(deg). Indeed, if 𝛾 : Δ1 → 𝑋 is a basis element in 𝐶 1 (𝑋 )
then
deg(𝜕1 (𝛾)) = deg(𝛾 (1) − 𝛾 (0)) = 1 − 1 = 0.

• Conversely, ker(deg) ⊆ 𝐵 0 (𝑋 ). Indeed, let 𝐿 = 𝑥 ∈𝑋 𝜆𝑥 · 𝑥 ∈ 𝐶 0 (𝑋 ) be an 𝑛-chain whose

Í
degree vanishes (𝜆𝑥 ∈ Z). In other words,
∑︁ ∑︁ ∑︁
Z30= 𝜆𝑥 = 𝜆𝑦 − (−𝜆𝑧 ).
𝑥 𝑦,𝜆 𝑦 >0 𝑧,𝜆𝑧 <0

Since both sums are equal we may rewrite 𝐿 as

∑︁
𝐿= (𝑦𝑖 − 𝑧𝑖 )
𝑖
8 Often one takes ‘path-connected’ to imply ’non-empty’. We’re just being extra careful.

21
 for some choice of points 𝑦𝑖 , 𝑧𝑖 ∈ 𝑋 , possibly repeating. Since 𝑋 is path-connected there
is a path 𝛾𝑖 from 𝑧𝑖 to 𝑦𝑖 so that 𝑦𝑖 − 𝑧𝑖 = 𝜕1 (𝛾𝑖 ) ∈ 𝐵 0 (𝑋 ). We conclude that 𝐿 ∈ 𝐵 0 (𝑋 ) as
we needed to show.
Altogether the first isomorphism theorem allows us to conclude:
𝑍 0 (𝑋 ) 𝐶 0 (𝑋 ) deg
H0 (𝑋 ) = = −−→ Z
𝐵 0 (𝑋 ) ker(deg) 


To deduce an interpretation of the 0th singular homology group for any space we need
the following intuitive fact.

Proposition 3.14. Let 𝑋 be a topological space and (𝑋𝛼 )𝛼 its path-connected components. Then we
have
H𝑛 (𝑋 ) = ⊕𝛼 H𝑛 (𝑋𝛼 ).

Proof. If 𝜎 : Δ𝑛 → 𝑋 is a singular 𝑛-simplex then its image is path-connected and therefore

lies entirely in one of the 𝑋𝛼 . In other words we have 𝐶𝑛 (𝑋 ) = ⊕𝛼 𝐶𝑛 (𝑋𝛼 ). Moreover, the
‘oriented’ boundary (3.5) of 𝜎 is a linear combination of (𝑛 − 1)-simplices all of which also lie
in 𝑋𝛼 . That is, 𝜕𝑛 is the sum of the boundary operators for each 𝑋𝛼 . (Another way of saying
this is that 𝐶 • (𝑋 ) = ⊕𝛼 𝐶 • (𝑋𝛼 ) as chain complexes.) This decomposition therefore passes to cycles
and boundaries, and eventually to homology. 

Corollary 3.15. Let 𝑋 be a topological space. Then H0 (𝑋 ) = Z𝜋0 (𝑋 ).

Proof. This is a combination of the two previous results, Lemma 3.13 and Proposition 3.14. 

Commentary 3.16. Corollary 3.15 was a warm-up for what will occupy us in the rest of this
section, namely an interpretation of H1 (𝑋 ). Our first goal is to construct a morphism of groups
(3.17) ℎ 1 : 𝜋1 (𝑋, 𝑥) → H1 (𝑋 ).
Subsequently we will show that this identifies H1 (𝑋 ) with the abelianization of the funda-
mental group (at least when 𝑋 is path-connected).
To define (3.17) on elements start with a loop 𝛾 : Δ1 → 𝑋 based at 𝑥, that is, 𝛾 (0) = 𝛾 (1) = 𝑥.
It follows that 𝜕1 (𝛾) = 𝛾 (1) − 𝛾 (0) = 𝑥 − 𝑥 = 0 ∈ 𝐶 0 (𝑋 ) so that 𝛾 ∈ 𝑍 1 (𝑋 ) is a 1-cycle. We
would like to define ℎ 1 ( [𝛾]) = [𝛾] ∈ H1 (𝑋 ) as the homology class associated with 𝛾. For this
we need to know that if we change 𝛾 to a homotopic loop 𝛾 0 then this homology class remains
constant.

Lemma 3.18. Let 𝛾 1, 𝛾 2 be two paths in 𝑋 , homotopic relative to their endpoints. Then 𝛾 1 ∼ 𝛾 2 .

Proof. We are given a homotopy 𝐻 between 𝛾 1 and 𝛾 2 relative to their endpoints:

𝛾2
𝑥 𝑦
𝛾1

22
Thus 𝐻 : [0, 1] 2 → 𝑋 looks as follows:
𝛾2

𝑐𝑥 𝑐𝑦

𝛾1

We can turn this into a 2-chain in 𝑋 like so,

𝛾2

𝜎2
𝑐𝑥 𝛾 𝑐𝑦
𝜎1

𝛾1

with 𝛾 (𝑡) = 𝐻 (𝑡, 𝑡) the ‘diagonal’. Hence

(3.19) 𝜕2 (𝜎2 − 𝜎1 ) = 𝛾 2 − 𝛾 + 𝑐 𝑥 − 𝑐 𝑦 + 𝛾 − 𝛾 1 = (𝛾 2 − 𝛾 1 ) + (𝑐 𝑥 − 𝑐 𝑦 )

and it suffices to show that 𝑐 𝑥 , 𝑐 𝑦 ∈ 𝐵 1 (𝑋 ). For this let 𝜎 : Δ2 → 𝑋 be the constant map with
value 𝑥. Then 𝜕2 (𝜎) = 𝑐 𝑥 − 𝑐 𝑥 + 𝑐 𝑥 = 𝑐 𝑥 which shows what we want for 𝑥, and for 𝑦 it’s the
same argument. 

At this point we have a well-defined map (3.17), ℎ 1 ( [𝛾]) = [𝛾]. (Note that the brackets on
the left mean ‘homotopy class’ while on the right they mean ‘homology class’.)

Lemma 3.20. The map (3.17) so defined is a group homomorphism.

Proof. We have already seen at the end of the last proof that the constant loop based at 𝑥 ∈ 𝑋
is a boundary and therefore vanishes in homology. In other words, ℎ 1 preserves the unit for
the group structure.
Let 𝛾 1, 𝛾 2 be two loops based at 𝑥 and let 𝛾 1𝛾 2 be their composite, given for 𝑡 ∈ [0, 1] by:
(
𝛾 1 (2𝑡) : 𝑡 ≤ 1/2
𝛾 1𝛾 2 (𝑡) =
𝛾 2 (2𝑡 − 1) : 𝑡 ≥ 1/2

In order to show that 𝛾 1𝛾 2 ∼ 𝛾 1 + 𝛾 2 we can produce a 2-simplex

𝑣2
𝛾 1𝛾 2 𝛾2

𝑣0 𝛾1 𝑣1

23
whose boundary is the difference 𝛾 1 + 𝛾 2 − 𝛾 1𝛾 2 . For this, project down orthogonally onto the
face [𝑣 0, 𝑣 2 ] and then apply 𝛾 1𝛾 2 . 

Remark 3.21. Let 𝛾 1, 𝛾 2 be two paths in 𝑋 with 𝛾 1 (1) = 𝛾 2 (0) so that they can be concatenated.
The preceding proof goes through and shows that in fact 𝛾 1𝛾 2 ∼ 𝛾 1 + 𝛾 2 . In particular, for any
path 𝛾 we have 𝛾 −1 ∼ −𝛾. (Indeed, 𝛾 + 𝛾 −1 ∼ 𝛾𝛾 −1 ∼ 0, by Lemma 3.18.)

Commentary 3.22. We cannot expect (3.17) to be an isomorphism in general for H1 (𝑋 ) is

abelian while 𝜋1 (𝑋, 𝑥) is not necessarily. But this means that the map factors through the
‘abelianization’:

Definition 3.23. Let 𝐺 be a group. Its commutator [𝐺, 𝐺] is the subgroup generated by the
elements 𝑔ℎ𝑔−1ℎ −1 for 𝑔, ℎ ∈ 𝐺. (It is normal.) The abelianization of 𝐺, denoted 𝐺 ab , is the
quotient 𝐺/[𝐺, 𝐺].

Example 3.24. If 𝐺 is abelian then 𝐺 ab = 𝐺.

Example 3.25. If 𝐺 = Z ∗ Z then 𝐺 ab = Z ⊕ 𝑍 . (See Exercise 2.2.)

For more examples, see the second exercise sheet.

Remark 3.26. By construction, 𝐺 ab is abelian, and it is in fact universal with respect to this
property. This means that whenever 𝜙 : 𝐺 → 𝐴 is a morphism to an abelian group 𝐴 there
exists a unique group morphism 𝜙¯ : 𝐺 ab → 𝐴 such that the composite 𝐺  𝐺 ab → 𝐴 is 𝜙:
∀
𝐺 𝐴

∃!
𝐺 ab
By this universal property of the abelianization and the fact that H1 (𝑋 ) is abelian, the
homomorphism (3.17) factors uniquely through a morphism between abelian groups,

(3.27) ℎ¯1 : 𝜋1 (𝑋, 𝑥) ab → H1 (𝑋 ),

and we will now prove:

Theorem 3.28. If 𝑋 is non-empty and path-connected then the map (3.27) is an isomorphism.

Proof. We will, in effect, define an inverse to this map.

Since 𝑋 is path-connected we choose, for every given 𝑦 ∈ 𝑋 , once and for all, a path 𝜂 𝑦
from 𝑥 to 𝑦. Given any path 𝛾 : Δ1 → 𝑋 we associate a loop based at 𝑥, as the following
concatenation:
(3.29) 𝑔(𝛾) := 𝜂𝛾 (0) 𝛾𝜂𝛾−1(1)

Linearly extending, we obtain a homomorphism

𝑔 : 𝑍 1 (𝑋 ) ⊆ 𝐶 1 (𝑋 ) → 𝜋 1 (𝑋, 𝑥) ab .

24
What happens to the boundaries? If 𝜎 : Δ2 → 𝑋 is a singular 2-simplex
𝑦1
𝛾1 𝛾2

𝑦0 𝛾3 𝑦2

then 𝛾 1𝛾 2 is homotopic to 𝛾 3 relative to their endpoints. To see this, embed the 2-simplex in
the square
𝛾 1𝛾 2

𝑐 𝑦0 𝑐 𝑦2

𝛾3

and project vertically onto the 2-simplex.

It follows that in 𝜋1 (𝑋, 𝑥) ab we have

𝑔(𝜕2𝜎) = 𝑔 (𝛾 1 + 𝛾 2 − 𝛾 3 )
= 𝑔(𝛾 1 ) + 𝑔(𝛾 2 ) − 𝑔(𝛾 3 )
= [𝜂 𝑣0𝛾 1𝜂 𝑣−1
1
] + [𝜂 𝑣1𝛾 2𝜂 𝑣−1
2
] − [𝜂 𝑣0𝛾 3𝜂 𝑣−1
2
]
= [(𝜂 𝑣0𝛾 1𝜂 𝑣−1
1
) (𝜂 𝑣1𝛾 2𝜂 𝑣−1
2
) (𝜂 𝑣2𝛾 3−1𝜂 𝑣−1
0
)]
= [𝜂 𝑣0𝛾 1𝛾 2𝛾 3−1𝜂 𝑣−1
0
]
= [𝜂 𝑣0 𝜂 𝑣−1
0
]
=0

so that (3.29) passes to 𝑔¯ : H1 (𝑋 ) → 𝜋 1 (𝑋, 𝑥) ab .

Now we check that the two composites are the identity:

• Let 𝛾 be a loop based at 𝑥. Then 𝑔( ¯ ℎ¯1 ( [𝛾])) = 𝑔(

¯ [𝛾]) = [𝜂𝑥 𝛾𝜂𝑥−1 ] = [𝜂𝑥 ] + [𝛾] − [𝜂𝑥 ] =
[𝛾] ∈ 𝜋1 (𝑋, 𝑥) so 𝑔¯ ◦ ℎ¯1 = id.
ab

• Let 𝐿 = 𝜆𝛾 𝛾 ∈ 𝑍 1 (𝑋 ) be a 1-cycle, 𝜆𝛾 ∈ Z. Replacing −𝛾 by 𝛾 −1 if necessary we

Í
may assume that all 𝜆𝛾 > 0 are positive, see Remark 3.21. Relabeling we can then write
𝐿 = 𝑘𝑖=1 𝛾𝑖 for paths 𝛾𝑖 in 𝑋 , possibly with repetitions. If 𝛾 1 is not a loop there must
Í
exist 𝑖 > 1 such that 𝛾 1 (1) = 𝛾𝑖 (0). Replacing 𝛾 1 + 𝛾𝑖 by the concatenation 𝛾 1𝛾𝑖 (again
using Remark 3.21) and doing induction we finally reduce to showing the claim for
𝐿 = 𝛾 a single loop in 𝑋 , based at 𝑦, say. In that case we have ℎ¯1 (𝑔( ¯ [𝛾])) = [𝜂 𝑦𝛾𝜂 𝑦−1 ] =
[𝜂 𝑦 ] + [𝛾] − [𝜂 𝑦 ] = [𝛾] and this completes the proof.

25
Corollary 3.30. If 𝑋 simply connected (hence non-empty and path-connected) then H1 (𝑋 ) = 0.9

Corollary 3.31. We have H1 (𝑆 1 ) = Z. A generator is given by the (homology class of the) obvious
surjective map 𝛾 1 : Δ1 → 𝑆 1 identifying the end points.

Remark 3.32. If we view 𝜋1 (𝑋, 𝑥) as the set of homotopy classes of pointed maps (𝑆 1, ∗) →
(𝑋, 𝑥) this Corollary provides another description of the map (3.17). As we will see in Sec-
tion 4.1 below, any continuous map 𝑓 : 𝑆 1 → 𝑋 induces a morphism in homology 𝑓∗ : H1 (𝑆 1 ) →
H1 (𝑋 ) and (3.17) can also be described as:

ℎ 1 ( [𝑓 ]) = 𝑓∗ ( [𝛾 1 ]) = [𝑓 ◦ 𝛾 1 ].

This suggests an analogous map in arbitrary degrees which was first studied by Hurewicz.
Once we prove that H𝑛 (𝑆 𝑛 ) = Z,10 generated by the homology class of some 𝛾𝑛 ∈ 𝑍𝑛 (𝑆 𝑛 ), we
can define

ℎ𝑛 : 𝜋𝑛 (𝑋, 𝑥) = [(𝑆 𝑛 , ∗), (𝑋, 𝑥)] −→ H𝑛 (𝑋 )

[𝑓 : 𝑆 𝑛 → 𝑋 ] ↦−→ 𝑓∗ ( [𝛾𝑛 ])

4 . . . and its fundamental theorems

So far our understanding of the singular homology is restricted to degrees 0 and 1. For ex-
ample, at this point we don’t even know the higher homology groups of such basic spaces
as the circle or higher-dimensional spheres. Since the singular chain complex is typically so
large, hands-on computations become all but infeasible. The goal of this section, therefore, is
to state and prove two fundamental theorems that allow us to compute the singular homology
of topological spaces without ever having to deal with the singular chain complex. To show
the power of these theorems we will also do some of these easy computations. More involved
applications will be forthcoming in later sections.

4.1 Statement and immediate applications

To state the first theorem we observe the following.

Lemma 4.1. Let 𝑓 : 𝑋 → 𝑌 be a continuous map. It induces a map in homology,

𝑓∗ : H𝑛 (𝑋 ) → H𝑛 (𝑌 ),

satisfying:11

1. id∗ = id,
2. (𝑓 ◦ 𝑔)∗ = 𝑓∗ ◦ 𝑔∗ .
9 Of course, if 𝑋 is empty then the conclusion holds as well.
10 See Corollary 4.13.
11 This says that H (−) is a functor from topological spaces to abelian groups.
𝑛

26
Proof. If 𝜎 : Δ𝑛 → 𝑋 is a singular 𝑛-simplex in 𝑋 then 𝑓 ◦ 𝜎 : Δ𝑛 → 𝑌 is a singular 𝑛-simplex
in 𝑌 , by continuity of 𝑓 . Linearly extending we obtain a homomorphism 𝑓∗ : 𝐶𝑛 (𝑋 ) → 𝐶𝑛 (𝑌 ).
Assembling these homomorphisms for all 𝑛 yields a chain map 𝑓∗ : 𝐶 • (𝑋 ) → 𝐶 • (𝑌 ) and there-
fore a map in homology (Lemma 2.49). The two properties stated are immediate from the
construction: These are true for the composition of maps. 

Theorem 4.2 (Homotopy invariance). Suppose 𝑓 , 𝑔 : 𝑋 → 𝑌 are homotopic maps. Then:

𝑓∗ = 𝑔∗ : H𝑛 (𝑋 ) → H𝑛 (𝑌 )

Corollary 4.3. If 𝑋 ' 𝑌 are homotopy equivalent then H𝑛 (𝑋 )  H𝑛 (𝑌 ) are isomorphic.

Proof. Let 𝑓 : 𝑋 → 𝑌 be a homotopy equivalence, with 𝑔 : 𝑌 → 𝑋 a homotopy inverse. That

is, 𝑓 ◦ 𝑔 ' id𝑌 , 𝑔 ◦ 𝑓 ' id𝑋 . Then:

𝑓∗ ◦ 𝑔∗ = (𝑓 ◦ 𝑔)∗ = (id𝑌 )∗ = id, 𝑔∗ ◦ 𝑓∗ = (𝑔 ◦ 𝑓 )∗ = (id𝑋 )∗ = id,

so that 𝑔∗ : H𝑛 (𝑌 ) → H𝑛 (𝑋 ) is an inverse to 𝑓∗ . 

Corollary 4.4. Let 𝑋 be a (non-empty) contractible space. Then

(
Z :𝑛 =0
H𝑛 (𝑋 ) =
0 :𝑛 ≠0

Proof. This follows from Corollary 4.3 together with Example 3.9. 

Example 4.5. We therefore know the homology of real Euclidean space R𝑘 , of the unit ball 𝐷 𝑘
and unit cube [0, 1] 𝑘 , among others.

Commentary 4.6. By itself, Homotopy Invariance might not be as powerful. Keeping one
of our motivating examples in mind, the sphere 𝑆 𝑘 is not contractible nor is it homotopy equiv-
alent to some other space whose homology we can compute (at least when 𝑘 ≥ 1). However,
we know how to build the sphere out of (contractible) 𝑘-balls, here for 𝑘 = 1:

𝑈1

(4.7)

𝑈2

More generally, if 𝑋 is a manifold we can cover it by open subspaces 𝑈𝑖 , each of which is

homeomorphic to R𝑘 , for some 𝑘, and therefore contractible. If we knew how to express the
homology of 𝑋 in terms of the homology of the 𝑈𝑖 then we could apply Homotopy Invariance
to conclude. Our second fundamental theorem, named after Mayer and Vietoris, allows us to do
that, with the single provision you are familiar with from the Seifert-van Kampen Theorem:
We also need to keep track of the intersection of 𝑈 1 and 𝑈 2 .

27
Commentary 4.8. The Seifert-van Kampen Theorem is very pertinent for the discussion
here so let us recall it in the following form. Let 𝑋 = 𝑈 1 ∪ 𝑈 2 be the union of two path-
connected open subspaces 𝑗1 : 𝑈 1 ↩→ 𝑋 , 𝑗2 : 𝑈 2 ↩→ 𝑋 such that 𝑈 1 ∩ 𝑈 2 is also path-connected.
For simplicity I will omit base-points. We then have a map
( 𝑗1 )∗ ∗( 𝑗2 ) ∗
𝜋1 (𝑈 1 ) ∗ 𝜋1 (𝑈 2 ) −−−−−−−−→ 𝜋1 (𝑋 ),
where:
• the map ( 𝑗1 )∗ ∗ ( 𝑗2 )∗ is surjective;
• its kernel is the normal subgroup generated by elements of the form 𝑖 (𝛾) = (𝑖 1 )∗ (𝛾) (𝑖 2 )∗ (𝛾) −1
and 𝑖 ℓ : 𝑈 1 ∩ 𝑈 2 ↩→ 𝑈 ℓ ;
Passing to homology we should abelianize everything in sight (Theorem 3.28) and this sim-
plifies the situation:
(𝑖 1 )∗ −(𝑖 2 ) ∗ ( 𝑗1 )∗ +( 𝑗2 )∗
(4.9) H1 (𝑈 1 ∩ 𝑈 2 ) −−−−−−−−→ H1 (𝑈 1 ) ⊕ H1 (𝑈 2 ) −−−−−−−−→ H1 (𝑋 )
where now
• ( 𝑗1 )∗ + ( 𝑗2 )∗ is surjective;
• its kernel is precisely the image of 𝑖.
We may express the last point as saying that the chain complex (4.9) is exact in the middle, see
Definition 2.42. (If we extend the chain complex by zero further to the right, then the first
point says precisely that the chain complex is also exact at H1 (𝑋 ). However, as we will discuss
now, this does not typically hold when 𝑈 1 ∩ 𝑈 2 is not path-connected.)

Theorem 4.10 (Mayer-Vietoris long exact sequence). Let 𝑋 = 𝑈 1 ∪ 𝑈 2 be the union of two
open subspaces 𝑗1 : 𝑈 1 ↩→ 𝑋 , 𝑗2 : 𝑈 2 ↩→ 𝑋 . There are ‘connecting homomorphisms’ 𝜕 : H𝑛 (𝑋 ) →
H𝑛−1 (𝑈 1 ∩ 𝑈 2 ) such that
𝜕 (𝑖 1 )∗ −(𝑖 2 )∗ ( 𝑗1 )∗ +( 𝑗2 )∗ 𝜕
· · · → H𝑛+1 (𝑋 ) →
− H𝑛 (𝑈 1 ∩𝑈 2 ) −−−−−−−−→ H𝑛 (𝑈 1 )⊕H𝑛 (𝑈 2 ) −−−−−−−−→ H𝑛 (𝑋 ) →
− H𝑛−1 (𝑈 1 ∩𝑈 2 ) → · · ·
is an exact chain complex.

Remark 4.11. Recall that being a chain complex, the composite of any two maps in this se-
quence is zero. In other words, the image of each morphism is contained in the kernel of the
next. And being exact means that the converse holds, that is, the image of each morphism is
precisely the kernel of the next one. Equivalently, its homology vanishes in each degree. An
exact chain complex also goes by the name of long exact sequence thus the name of the theorem.

Example 4.12. Let us go back to the circle (4.7). Since the intersection 𝑈 1 ∩𝑈 2 ' 𝑆 0 consists of
two points (up to homotopy equivalence) we have H𝑛 (𝑈 1 ∩ 𝑈 2 ) = 0 for 𝑛 > 0, see Example 3.9
and Proposition 3.14. Since both 𝑈 ℓ are contractible, we also have H𝑛 (𝑈 ℓ ) = 0 for 𝑛 > 0. This
means that most terms in the Mayer-Vietoris l.e.s. vanish:
( 𝑗1 )∗ +( 𝑗2 ) ∗
0 ⊕ 0 −−−−−−−−→ H𝑛 (𝑆 1 ) →
𝜕
− 0

28
is exact for each 𝑛 > 1 so that H𝑛 (𝑆 1 ) = 0. With our known low-degree computations
(Corollary 3.31 and Lemma 3.13) or analyzing the Mayer-Vietoris l.e.s. further we obtain
(
Z : 𝑛 = 0, 1
H𝑛 (𝑆 1 ) =
0 : 𝑛 ≠ 0, 1

This generalizes to spheres of higher dimensions:

(
Z : 𝑛 = 𝑘, 0
Corollary 4.13. Let 𝑘 ≥ 1. Then H𝑛 (𝑆 𝑘 ) =
0 : 𝑛 ≠ 𝑘, 0

Proof. We may write 𝑆 𝑘 as the union of open neighborhoods 𝑈 1 (resp. 𝑈 2 ) of the lower (resp.
upper) hemisphere:

𝑈1 𝑈2

Each of these pieces is contractible and 𝑈 1 ∩𝑈 2 is homotopy equivalent to 𝑆 𝑘−1 . Mayer-Vietoris

therefore yields a sequence

... 𝐻𝑛 (𝑆 𝑘−1 ) 0 𝐻𝑛 (𝑆 𝑘 )

𝐻𝑛−1 (𝑆 𝑘−1 ) 0 𝐻𝑛−1 (𝑆 𝑘 )

𝐻 1 (𝑆 𝑘−1 ) 0 𝐻 1 (𝑆 𝑘 )

𝐻 0 (𝑆 𝑘−1 ) 𝐻 0 (𝑈 1 ) ⊕ 𝐻 0 (𝑈 2 ) 𝐻 0 (𝑆 𝑘 ) 0

in which the kernel of each map is equal to the image of the preceding one. We do induction
on 𝑘, with 𝑘 = 1 having been established in Example 4.12. For 𝑘 > 1, the last terms in the
sequence are (by induction hypothesis)
𝜕 (𝑖 1 )∗ −(𝑖 2 )∗ ( 𝑗1 ) ∗ +( 𝑗2 )∗
0 → H1 (𝑆 𝑘 ) →
− Z −−−−−−−−→ Z ⊕ Z −−−−−−−−→ H0 (𝑆 𝑘 ) → 0,
showing that H1 (𝑆 𝑘 ) = 0 and H0 (𝑆 𝑘 ) = Z. (Of course, we already knew that from Lemma 3.13
and Theorem 3.28.) And the upper parts of the sequence give isomorphisms for 𝑛 > 1:
(
Z :𝑛 =𝑘
H𝑛 (𝑆 𝑘 )  H𝑛−1 (𝑆 𝑘−1 ) =
0 :𝑛 ≠𝑘

29
by induction hypothesis. 

Convention 4.14. There is a more convenient formulation of the previous result. For this let
𝜋 : 𝑋 → ∗ be the unique morphism to the point. We then define reduced homology
 𝜋∗

H̃𝑛 (𝑋 ) := ker H𝑛 (𝑋 ) −−→ H𝑛 (∗) .

Note that if 𝑛 > 0 then H̃𝑛 (𝑋 ) = H𝑛 (𝑋 ) so the only difference is in degree 0. Also, by
convention, we only want to consider reduced homology of non-empty spaces. (Otherwise
all sorts of pathologies can arise.)

Remark 4.15. The following reformulation of Corollary 4.13 also holds when 𝑘 = 0:
(
Z :𝑛 =𝑘
H̃𝑛 (𝑆 𝑘 ) =
0 :𝑛 ≠𝑘

Remark 4.16. Mayer-Vietoris and Homotopy Invariance hold equally for reduced homology.
This follows from their original formulation by keeping track of the induced map 𝜋∗ . See
Exercise 3.2.

Commentary 4.17. We end this section with two celebrated applications of our fundamental
theorems. Both are due to Brouwer in the 1910’s (although with different proofs).
To motivate the first one, ‘the’ fixed-point theorem (there are in fact many of these), con-
sider a continuous map 𝑓 : [0, 1] → [0, 1]. It is quite intuitive and you probably already knew
that 𝑓 necessarily has a fixed point. (If you need a proof: use the Intermediate Value Theorem.)
What happens in higher dimensions? For example, imagine you go hiking in the Peak
District and get lost along the way. Imagine also you’re super old-fashioned and actually carry
a(n infinitely precise) map with you that you lay flat on the earth in front of you. The Brouwer
Fixed-point Theorem in dimension 2 asserts that there is a point on the map that represents
exactly the point where it touches the earth.
Or, take a glass of red wine and slosh it around. (This makes you look like a connaisseur.)
In dimension 3, the theorem asserts that there is a molecule in exactly the same spot as before
the sloshing!12

Corollary 4.18 (Brouwer Fixed-point Theorem). Every continuous map 𝑓 : 𝐷 𝑘 → 𝐷 𝑘 has a

fixed point.

Proof. Suppose not. Then the ray starting at 𝑓 (𝑥) in the direction of 𝑥 meets 𝑆 𝑘−1 in exactly one
point 𝑔(𝑥) ≠ 𝑓 (𝑥). This 𝑔 : 𝐷 𝑘 → 𝑆 𝑘−1 would define a retraction, contradicting Corollary 4.19
below. 

Corollary 4.19. 𝑆 𝑘−1 = 𝜕𝐷 𝑘 is not a retract of 𝐷 𝑘 .

12 Foranother fun application of the theorem see: David Gale, The Game of Hex and the Brouwer Fixed-Point
Theorem, The American Mathematical Monthly, vol. 86, 1979, pp. 818-827, weblink

30
Proof. Assume to the contrary that 𝑖 : 𝑆 𝑘−1 ↩→ 𝐷 𝑘 admits a retraction 𝑟 : 𝐷 𝑘 → 𝑆 𝑘−1 . Then
id𝑆 𝑘 = 𝑟 ◦ 𝑖 so that
𝑖∗ 𝑟∗
Z = H̃𝑘−1 (𝑆 𝑘−1 ) −
→ H̃𝑘−1 (𝐷 𝑘 ) −→ H̃𝑘−1 (𝑆 𝑘−1 ) = Z
is the identity map, by Lemma 4.1. However, H̃𝑘−1 (𝐷 𝑘 ) = 0 since 𝐷 𝑘 is contractible. We have
arrived at a contradiction. 

Corollary 4.20 (Invariance of domain). If 𝑘 ≠ ℓ then R𝑘 0 Rℓ .

≈
Proof. The argument is similar to Example 1.9. Namely, assume 𝑓 : R𝑘 − → Rℓ . (The statement
≈
is clear if one of 𝑘 or ℓ is zero so we may assume 𝑘, ℓ > 0.) Then also 𝑓 : R𝑘 \{0} −
→ Rℓ \{𝑓 (0)}.
Since R𝑘 \{0} ' 𝑆 𝑘−1 and Rℓ \{𝑓 (0)} ' 𝑆 ℓ−1 we obtain an isomorphism
𝑓∗
Z = H̃𝑘−1 (𝑆 𝑘−1 ) = H̃𝑘−1 (R𝑘 \{0}) −
→ H̃𝑘−1 (Rℓ \{𝑓 (0)}) = H̃𝑘−1 (𝑆 ℓ−1 )


which implies 𝑘 = ℓ. 

4.2 Homotopy Invariance

Commentary 4.21. The goal of this section is to prove Homotopy Invariance. Recall (The-
orem 4.2) that we start with a homotopy 𝐻 : 𝑋 × [0, 1] → 𝑌 between 𝑓 and 𝑔 and want to
prove 𝑓∗ = 𝑔∗ as maps in homology. The strategy is as follows:
1. Together with the prism operator, 𝐻 produces a chain homotopy between these two chain
maps 𝑓∗ and 𝑔∗ .
2. It is a very easy algebraic observation that chain homotopic chain maps induce the same
map in homology.
In order to see what exactly we ought to produce in the first step we start by introducing
chain homotopies.

Definition 4.22. Let 𝑎 •, 𝑏 • : 𝐶 • → 𝐶 •0 be two chain maps. A chain homotopy from 𝑎 to 𝑏 is a

collection of morphisms
𝜂𝑛 : 𝐶𝑛 → 𝐶𝑛+1
0

such that
(4.23) 0
𝑏𝑛 − 𝑎𝑛 = 𝜕𝑛+1 𝜂𝑛 + 𝜂𝑛−1 𝜕𝑛
for all 𝑛 ∈ Z. As usual, we say that 𝑎 and 𝑏 are chain homotopic if there exists a chain homotopy
between them.

Remark 4.24. In pictures, a chain homotopy looks like (for better readability I’m omitting
the subscripts that go with the maps)
𝜕 𝜕 𝜕
··· 𝐶𝑛+1 𝐶𝑛 𝐶𝑛 ···
𝑏−𝑎 𝜂 𝑏−𝑎 𝜂 𝑏−𝑎
𝜂
0 𝜕0 𝜕0 𝜕0
··· 𝐶𝑛+1 𝐶𝑛0 0
𝐶𝑛−1 ···

31
and where the defining identity (4.23) expresses the green path as the sum of the two brown
paths.

Here is the easy observation alluded to above:

Lemma 4.25. Let 𝑎 • and 𝑏 • be chain homotopic. Then their induced maps in homology are equal:

𝑎𝑛 = 𝑏𝑛 : H𝑛 (𝐶 • ) → H𝑛 (𝐶 •0 )

Proof. Let 𝑐 ∈ 𝑍𝑛 (𝐶 • ) be an 𝑛-cycle. Applying the defining equation (4.23) we see that
0 0
𝑏𝑛 (𝑐) − 𝑎𝑛 (𝑐) = 𝜕𝑛+1 𝜂𝑛 (𝑐) + 𝜂𝑛−1 𝜕𝑛 (𝑐) = 𝜕𝑛+1 (𝜂𝑛 (𝑐))

is a boundary. In other words, 𝑏𝑛 (𝑐) and 𝑎𝑛 (𝑐) are homologous as was to be proven. 

More material on chain homotopies is on Exercise Sheets 3, 4.

Commentary 4.26. We now turn to the prism operator 𝑃 and producing a chain homotopy
between 𝑎 = 𝑓∗ and 𝑏 = 𝑔∗ . The idea is as follows. Starting with a singular 𝑛-simplex 𝜎 : Δ𝑛 →
𝑋 and the homotopy 𝐻 : 𝑋 × [0, 1] → 𝑌 we compose them to a continuous map

𝐻 ◦ (𝜎 × id) : Δ𝑛 × [0, 1] → 𝑌 .

This provides a homotopy from 𝐻 ◦ (𝜎 × {0}) = 𝑓 ◦ 𝜎 = 𝑓∗ (𝜎) to 𝐻 ◦ (𝜎 × {1}) = 𝑔 ◦ 𝜎 = 𝑔∗ (𝜎).

𝑔∗ 𝜎

𝑓∗𝜎

How could we produce an (𝑛 + 1)-chain in 𝑌 out of this data? Well, we chop up the ‘prism’
Δ𝑛× [0, 1] into (𝑛 + 1)-simplices. We first consider the situation for small 𝑛 to get an idea
what’s going on. (To keep our notation straight let us label the vertices of Δ𝑛 = [𝑣 0, . . . , 𝑣𝑛 ].
Then the vertices of Δ𝑛 × [0, 1] are denoted by 𝑣𝑖 𝑗 = (𝑣𝑖 , 𝑗) where 0 ≤ 𝑖 ≤ 𝑛 and 0 ≤ 𝑗 ≤ 1.)
• For 𝑛 = 0 we have that Δ0 × [0, 1] = [𝑣 00, 𝑣 01 ] is already a 1-simplex so we set

𝑃 (Δ0 ) := [𝑣 00, 𝑣 01 ].

• For 𝑛 = 1 we get a square Δ1 × [0, 1]. We can turn this into a 2-chain, using the exact
same idea as in the proof of Lemma 3.18.

𝑣 01 𝑣 11

𝑣 00 𝑣 10

32
 As in loc. cit. (or, trial and error shows that) a good 2-chain to consider is

𝑃 (Δ1 ) := [𝑣 00, 𝑣 01, 𝑣 11 ] − [𝑣 00, 𝑣 10, 𝑣 11 ]

because its boundary is

𝜕2 (𝑃 (Δ1 )) = [𝑣 01, 𝑣 11 ] − [𝑣 00, 𝑣 11 ] + [𝑣 00, 𝑣 01 ] − [𝑣 10, 𝑣 11 ] + [𝑣 00, 𝑣 11 ] − [𝑣 00, 𝑣 10 ]

= ( [𝑣 01, 𝑣 11 ] − [𝑣 00, 𝑣 10 ]) − ( [𝑣 10, 𝑣 11 ] − [𝑣 00, 𝑣 01 ])
= ( [𝑣 01, 𝑣 11 ] − [𝑣 00, 𝑣 10 ]) − 𝑃 (𝜕Δ1 ),

as required by (4.23). (Here, we think of 𝑃 as linearly extended, so that 𝑃 (𝜕Δ1 ) = 𝑃 ( [𝑣 1 ] −

[𝑣 0 ]) = 𝑃 ( [𝑣 1 ]) − 𝑃 ( [𝑣 0 ]) = [𝑣 10, 𝑣 11 ] − [𝑣 00, 𝑣 01 ].)
• For 𝑛 = 2 we are dealing with Δ2 × [0, 1] which we may subdivide into three 3-simplices
as follows:

𝑣 21

𝑣 01 𝑣 11

𝑣 20
𝑣 00 𝑣 10

In order to satisfy the formula (4.23) we need to again take an alternating sum:

𝑃 (Δ2 ) = [𝑣 00, 𝑣 01, 𝑣 11, 𝑣 21 ] − [𝑣 00, 𝑣 10, 𝑣 11, 𝑣 21 ] + [𝑣 00, 𝑣 10, 𝑣 20, 𝑣 21 ]

We are now ready to guess the general formula.

Construction 4.27. We define the prism operator on Δ𝑛 = [𝑣 0, . . . , 𝑣𝑛 ] as follows:

𝑛
∑︁
𝑛
𝑃 (Δ ) = (−1)𝑖 [𝑣 00, . . . , 𝑣𝑖0, 𝑣𝑖1, . . . , 𝑣𝑛1 ] ∈ 𝐶𝑛+1 (Δ𝑛 × [0, 1])
𝑖=0

One way to think about this is that the boundary of the prism Δ𝑛 × [0, 1] is made of three
pieces:

Δ𝑛 × {1} Δ𝑛 × {0} 𝜕Δ𝑛 × [0, 1]

33
The signs in the definition of 𝑃 (Δ𝑛 ) ensure we can turn this into an equality of 𝑛-chains:

Lemma 4.28. For every 𝑛 ≥ 0 we have in 𝐶𝑛 (Δ𝑛 × [0, 1]):

𝜕𝑃 (Δ𝑛 ) = [𝑣 01, . . . , 𝑣𝑛1 ] − [𝑣 00, . . . , 𝑣𝑛0 ] − 𝑃 (𝜕Δ𝑛 )

Remark 4.29. Just to avoid any confusion, in this formula, 𝜕Δ𝑛 = 𝑛𝑗=0 (−1) 𝑗 [𝑣 0, . . . , 𝑣ˆ 𝑗 , . . . , 𝑣𝑛 ]
Í

is the 𝑛-th boundary operator applied to the standard 𝑛-simplex. And we have linearly ex-
tended 𝑃 so that the last term is
𝑛
∑︁
𝑛
𝑃 (𝜕Δ ) = (−1) 𝑗 𝑃 ( [𝑣 0, . . . , 𝑣ˆ 𝑗 , . . . , 𝑣𝑛 ])
𝑗=0

Before giving the proof of this lemma let us see how the prism operator can be used to
finish the proof of Homotopy Invariance.

Proof of Theorem 4.2. Choose a homotopy 𝐻 : 𝑋 × [0, 1] → 𝑌 from 𝑓 to 𝑔. Let 𝜎 : Δ𝑛 → 𝑋 be

a singular 𝑛-simplex in 𝑋 and consider 𝐻 ◦ (𝜎 × id), a map Δ𝑛 × [0, 1] → 𝑌 . It then induces a
map on (𝑛 + 1)-chains

𝐻 ◦ (𝜎 × id) ∗ : 𝐶𝑛+1 (Δ𝑛 × [0, 1]) → 𝐶𝑛+1 (𝑌 ).

We define

𝜂𝑛 : 𝐶𝑛 (𝑋 ) −→ 𝐶𝑛+1 (𝑌 )
𝜎 ↦−→ 𝐻 ◦ (𝜎 × id)


∗ (𝑃 (Δ𝑛 ))

so that

𝜕𝜂𝑛 (𝜎) = 𝜕 𝐻 ◦ (𝜎 × id) ∗ (𝑃 (Δ𝑛 )) definition of 𝜂𝑛

= 𝐻 ◦ (𝜎 × id) ∗ (𝜕𝑃 (Δ )) chain map identity

𝑛

= 𝐻 ◦ (𝜎 × id) ∗ ( [𝑣 01, . . . , 𝑣𝑛1 ] − [𝑣 00, . . . , 𝑣𝑛0 ] − 𝑃 (𝜕Δ )) Lemma 4.28

𝑛

= 𝑔∗𝜎 − 𝑓∗𝜎 − 𝜂𝑛−1 𝜕(𝜎) unwinding definitions

This shows that 𝜂 defines a chain homotopy from 𝑓∗ to 𝑔∗ so Lemma 4.25 kicks in. 

We are left to give the:

Proof of Lemma 4.28. This is a straightforward albeit tedious calculation.

We have
∑︁
(4.30) 𝜕𝑛+1 𝑃 (Δ𝑛 ) = (−1)𝑖+𝑗 [𝑣 00, . . . , 𝑣ˆ 𝑗0, . . . , 𝑣𝑖0, 𝑣𝑖1, . . . , 𝑣𝑛1 ] +
𝑗 ≤𝑖
∑︁
(−1)𝑖+𝑗+1 [𝑣 00, . . . , 𝑣𝑖0, 𝑣𝑖1, . . . , 𝑣ˆ 𝑗1, . . . , 𝑣𝑛1 ]
𝑗 ≥𝑖

34
 • Consider first the terms with 𝑖 = 𝑗. We get for these:
𝑛
∑︁ 𝑛
∑︁
[𝑣 00, . . . , 𝑣 (𝑖−1)0, 𝑣𝑖1, . . . , 𝑣𝑛1 ] − [𝑣 00, . . . , 𝑣𝑖0, 𝑣 (𝑖+1)1, . . . , 𝑣𝑛1 ]
𝑖=0 𝑖=0

all but two of which cancel each other out, leaving us with

[𝑣 01, . . . , 𝑣𝑛1 ] − [𝑣 00, . . . , 𝑣𝑛0 ].

• To account for the remaining terms (𝑖 ≠ 𝑗) we apply the prism operator 𝑃 to each face
[𝑣 0, . . . , 𝑣ˆ 𝑗 , . . . , 𝑣𝑛 ] of Δ𝑛 ,
∑︁
𝑃 ( [𝑣 0, . . . , 𝑣ˆ 𝑗 , . . . , 𝑣𝑛 ]) = (−1)𝑖 [𝑣 00, . . . , 𝑣𝑖0, 𝑣𝑖1, . . . , 𝑣ˆ 𝑗1, . . . , 𝑣𝑛1 ] +
𝑖< 𝑗
∑︁
(−1)𝑖+1 [𝑣 00, . . . , 𝑣ˆ 𝑗0, . . . , 𝑣𝑖0, 𝑣𝑖1, . . . , 𝑣𝑛1 ],
𝑗 <𝑖

hence, taking the alternating sum over all 𝑗, we get

∑︁
𝑃 (𝜕𝑛 Δ𝑛 ) = (−1)𝑖+𝑗 [𝑣 00, . . . , 𝑣𝑖0, 𝑣𝑖1, . . . , 𝑣ˆ 𝑗1, . . . , 𝑣𝑛1 ] +
𝑖< 𝑗
∑︁
(−1)𝑖+𝑗+1 [𝑣 00, . . . , 𝑣ˆ 𝑗0, . . . , 𝑣𝑖0, 𝑣𝑖1, . . . , 𝑣𝑛1 ].
𝑗 <𝑖

We note that this is precisely the negative of the terms in (4.30) yet to be accounted for
(that is, those with 𝑖 ≠ 𝑗). This concludes the proof.

4.3 Mayer-Vietoris: strategy

Commentary 4.31. The goal of this section is to prove (the existence of ) the Mayer-Vietoris
long exact sequence, Theorem 4.10. To re-familiarize ourselves with that result let us do one
more application.

Example 4.32. Let us compute the homology H• (𝑋 ) of the torus 𝑋 = T and the Klein bot-
tle 𝑋 = K. Recall that these can be obtained from the square by identifying opposite edges:

T K

In each case we may cover them by two open subsets 𝑉1 (in green) and 𝑉2 (in gray):

35
Note that in each case, 𝑉𝑖 ' 𝑆 1 and 𝑉1 ∩ 𝑉2 ' 𝑆 1 q 𝑆 1 which we label 𝐿 (for left) and 𝑅 (for
right). Here are the pictures for the torus on the left and the Klein bottle on the right:

T K

The black part depicts the common intersection. And the red arrows amount to choosing
generators of the first homology group (for 𝑉1, 𝑉2, 𝑉1 ∩ 𝑉2 ), see Corollary 3.31. The MV long
exact sequence vanishes in degrees > 2 and the only interesting bit is
𝜕 𝑖 𝑗 𝜕
0 → H2 (𝑋 ) →
− H1 (𝐿) ⊕ H1 (𝑅) →
− H1 (𝑉1 ) ⊕ H1 (𝑉2 ) →
− H1 (𝑋 ) →
−
H0 (𝐿) ⊕ H0 (𝑅) → H0 (𝑉1 ) ⊕ H0 (𝑉2 ) → H0 (𝑋 ) → 0
The bottom row is easy to analyze:
1 1
 
−1 −1 (1 1)
Z ⊕ Z −−−−−−−→ Z ⊕ Z −−−−→ Z → 0
so that the we get an exact sequence

− Z2 →
− Z2 →
𝜕 𝑖 𝑗 𝜕
0 → H2 (𝑋 ) → − H1 (𝑋 ) →
− Z → 0.
It follows from Exercise 2.6 that
H2 (𝑋 ) = ker(𝑖), H1 (𝑋 ) = Z ⊕ coker(𝑖).
In both cases, the inclusions 𝐿 ↩→ 𝑉1, 𝑉2 send generator to generator in H1 . Also in both cases,
𝑅 ↩→ 𝑉1 sends generator to the negative of the generator. For 𝑅 ↩→ 𝑉2 , the situation for the
torus and the Klein bottle are different. Putting everything together we find for 𝑖 the matrix:
1 −1 1 −1
   
T : K:
−1 1 −1 −1

36
so that


 Z : 𝑛 = 0, 2 

 Z :𝑛 =0
H𝑛 (T ) = Z2 H𝑛 (K) = Z ⊕ Z/2 : 𝑛 = 1

 

:𝑛 =1
 
0

: else 0

: else
 

(Note that the result for the torus coincides with the simplicial homology groups computed
in Example 2.31. This is an instance of the general comparison theorem between the two
homology theories to be discussed in Section 8.1.)

Commentary 4.33. As for the proof of Homotopy Invariance (section 4.2), the proof of
Mayer-Vietoris combines a topological step with a purely algebraic step:

1. The purely algebraic fact is that a ‘short exact sequence of chain complexes’ gives rise
to a long exact sequence in homology. (We will define short exact sequences of chain
complexes in a moment but you could try to guess already what it should be.)
2. The relevant short exact sequence of chain complexes associated with the cover 𝑋 =
𝑈 1 ∪ 𝑈 2 is
0 → 𝐶 • (𝑈 1 ∩ 𝑈 2 ) → 𝐶 • (𝑈 1 ) ⊕ 𝐶 • (𝑈 2 ) → 𝐶 • (𝑈 1 + 𝑈 2 ) → 0,
where 𝐶 • (𝑈 1 + 𝑈 2 ) ⊆ 𝐶 • (𝑋 ) is a subcomplex with the same homology groups. This last
bit (showing that they have the same homology) is the hardest part. This involves the
topological ingredient, a process called barycentric subdivision.

Commentary 4.34. Recall that a short exact sequence of abelian groups is an exact sequence
of the form
𝑓 𝑔
0→𝐴→ − 𝐵→ − 𝐶 → 0.
In other words,

• 𝑓 is injective,
• 𝑔 is surjective, and
• ker(𝑔) = img(𝑓 ).

We now transfer this concept to the level of chain complexes.

Definition 4.35. A short exact sequence of chain complexes

𝑓 𝑔
0 → 𝐴• → − 𝐶• → 0
− 𝐵• →

is the data of two chain maps, 𝑓 and 𝑔, such that for each 𝑛 ∈ Z,

𝑓𝑛 𝑔𝑛
0 → 𝐴𝑛 −→ 𝐵𝑛 −−→ 𝐶𝑛 → 0

is a short exact sequence of abelian groups.

37
Example 4.36. Let (𝐵 •, 𝜕• ) be any chain complex. Define a new chain complex 𝐴• by
 𝐵 :𝑛 >0
 𝑛



𝐴𝑛 = ker(𝜕0 ) :𝑛 =0

0

:𝑛 <0

The differentials are the ones from 𝐵 • restricted to 𝐴• . We may then define 𝐶 • levelwise as
𝐶𝑛 = 𝐵𝑛 /𝐴𝑛 with the induced differentials. The resulting short exact sequence 0 → 𝐴• →
𝐵 • → 𝐶 • → 0 of chain complexes looks as follows:

.. .. ..
. . .

0 𝐵2 𝐵2 0 0
𝜕2 𝜕2

0 𝐵1 𝐵1 0 0
𝜕1 𝜕1

0 ker(𝜕0 ) 𝐵0 𝐵 0 /ker(𝜕0 ) 0
𝜕0 𝜕0

0 0 𝐵 −1 𝐵 −1 0
𝜕−1 𝜕−1

0 0 𝐵 −2 𝐵 −2 0

.. .. ..
. . .
Proposition 4.37. Let
𝑓 𝑔
0 → 𝐴• → − 𝐶• → 0
− 𝐵• →
be a short exact sequence of chain complexes. There are ‘connecting homomorphisms’ 𝜕 : H𝑛 (𝐶 • ) →
H𝑛−1 (𝐴• ) and a long exact sequence in homology:
𝑓𝑛 𝑔𝑛
··· H𝑛 (𝐴• ) H𝑛 (𝐵 • ) H𝑛 (𝐶 • )
𝜕

𝑓𝑛−1 𝑔𝑛−1
H𝑛−1 (𝐴• ) H𝑛−1 (𝐵 • ) H𝑛−1 (𝐶 • )
𝜕

H𝑛−2 (𝐴• ) ···

Proof. This is a proof best done ‘live’ on the board or on paper/tablet etc. Give it a go yourself !
(Or watch the lecture recording. Or, it’s Hatcher, p. 116f.) In any case, here’s what one has to
do:

38
 • To construct 𝜕 let 𝑐 ∈ 𝑍𝑛 (𝐶 • ). Choose a lift 𝑏 ∈ 𝐵𝑛 under 𝑔𝑛 . Since 𝑔𝑛−1 𝜕𝑛 (𝑏) = 𝜕𝑛𝑔𝑛 (𝑏) =
𝜕𝑛 (𝑐) = 0 there exists 𝑎 ∈ 𝐴𝑛−1 such that 𝑓𝑛−1 (𝑎) = 𝜕𝑛𝑏. One checks that [𝑐] ↦→ [𝑎] is a
well-defined homomorphism 𝜕 : H𝑛 (𝐶 • ) → H𝑛−1 (𝐴• ).
• One checks exactness at each of the spots in the sequence.

Commentary 4.38. We now return to the topological situation. Let 𝑈 1, 𝑈 2 ⊆ 𝑋 be two

subspaces (not necessarily open). Denote by 𝐶𝑛 (𝑈 1 + 𝑈 2 ) ⊆ 𝐶𝑛 (𝑋 ) the subgroup of 𝑛-chains
that can be written as the sum of 𝑛-chains in 𝑈 1 and 𝑛-chains in 𝑈 2 .

Remark 4.39. The boundary of an 𝑛-chain in 𝑈 ℓ is an (𝑛 − 1)-chain in 𝑈 ℓ which implies

that the differentials in 𝐶 • (𝑋 ) restrict to 𝐶 • (𝑈 1 + 𝑈 2 ). With these induced differentials we
may view 𝐶 • (𝑈 1 + 𝑈 2 ) ⊆ 𝐶 • (𝑋 ) as a sub-chain complex. In other words, the inclusion maps
𝐶𝑛 (𝑈 1 + 𝑈 2 ) ↩→ 𝐶𝑛 (𝑋 ) define a chain map.

Proposition 4.40. Let 𝑗ℓ : 𝑈 ℓ ↩→ 𝑋 and 𝑖 ℓ : 𝑈 1 ∩ 𝑈 2 ↩→ 𝑈 ℓ be the inclusions, for ℓ = 1, 2. There is

a short exact sequence of chain complexes,

(𝑖 1 )∗ −(𝑖 2 )∗ ( 𝑗1 )∗ +( 𝑗2 )∗
0 → 𝐶 • (𝑈 1 ∩ 𝑈 2 ) −−−−−−−−→ 𝐶 • (𝑈 1 ) ⊕ 𝐶 • (𝑈 2 ) −−−−−−−−→ 𝐶 • (𝑈 1 + 𝑈 2 ) → 0.

Proof. We know these are chain maps, so we need to show exactness at the three spots (in each
degree 𝑛):

• The subgroup 𝐶𝑛 (𝑈 1 +𝑈 2 ) was defined precisely as the image of ( 𝑗1 )∗ + ( 𝑗2 )∗ so the latter

is surjective.
• The morphism (𝑖 1 )∗ : 𝐶𝑛 (𝑈 1 ∩𝑈 2 ) ↩→ 𝐶𝑛 (𝑈 1 ) is already injective hence so is (𝑖 1 )∗ − (𝑖 2 )∗ .
• The composite

(( 𝑗1 )∗ + ( 𝑗2 )∗ ) ◦ ((𝑖 1 )∗ − (𝑖 2 )∗ ) = ( 𝑗1 )∗ (𝑖 1 )∗ − ( 𝑗2 )∗ (𝑖 2 )∗ = ( 𝑗1𝑖 1 )∗ − ( 𝑗2𝑖 2 )∗ = 0

since 𝑗1𝑖 1 = 𝑗2𝑖 2 is simply the inclusion 𝑘 : 𝑈 1 ∩ 𝑈 2 ↩→ 𝑋 .

Conversely, if (𝑐 1, 𝑐 2 ) ∈ 𝐶𝑛 (𝑈 1 ) ⊕ 𝐶𝑛 (𝑈 2 ) such that ( 𝑗1 )∗ (𝑐 1 ) + ( 𝑗2 )∗ (𝑐 2 ) = 0 then also
( 𝑗1 )∗ (𝑐 1 ) = ( 𝑗2 )∗ (−𝑐 2 ) and we see that both sides must be supported on 𝑈 1 ∩ 𝑈 2 . That is,
there exists 𝑐 ∈ 𝐶𝑛 (𝑈 1 ∩ 𝑈 2 ) such that 𝑘 ∗ (𝑐) = ( 𝑗1 )∗ (𝑐 1 ) = ( 𝑗2 )∗ (−𝑐 2 ). By injectivity of
( 𝑗ℓ )∗ we deduce that already (𝑖 1 )∗ (𝑐) = 𝑐 1 and −(𝑖 2 )∗ (𝑐) = 𝑐 2 .

This completes the proof. 

Commentary 4.41. We now state the other property about 𝐶 • (𝑈 1 + 𝑈 2 ) that we need for the
proof of MV. (The proof will be given shortly.)

Proposition 4.42. Assume that the interiors of 𝑈 1 and 𝑈 2 jointly cover 𝑋 . Then the inclusion
𝐶 • (𝑈 1 + 𝑈 2 ) ↩→ 𝐶 • (𝑋 ) induces isomorphisms in homology.

39
Proof of Mayer-Vietoris (Theorem 4.10). By Proposition 4.37, the short exact sequence of Propo-
sition 4.40 induces a long exact sequence in homology:

𝜕 (𝑖 1 )∗ −(𝑖 2 )∗ ( 𝑗1 )∗ +( 𝑗2 )∗ 𝜕
· · · → H𝑛+1 (𝐶 • (𝑈 1 +𝑈 2 )) →
− H𝑛 (𝑈 1 ∩𝑈 2 ) −−−−−−−−→ H𝑛 (𝑈 1 )⊕H𝑛 (𝑈 2 ) −−−−−−−−→ H𝑛 (𝐶 • (𝑈 1 +𝑈 2 )) →
− ···

By Proposition 4.42, we may replace the homology groups H𝑛 (𝐶 • (𝑈 1 +𝑈 2 )) by H𝑛 (𝑋 ) every-

where thus the claim. 

Remark 4.43. As we have just seen, the Mayer-Vietoris long exact sequence also exists when
the subspaces 𝑈 1, 𝑈 2 ⊆ 𝑋 are not open. The important requirement is rather that their interiors
jointly cover 𝑋 .
(Of course, some condition is necessary since every space 𝑋 = 𝑉 ∪ 𝑉 𝑐 is the union of any
subspace and its complement. These intersect trivially so that a Mayer-Vietoris long exact
sequence would imply H𝑛 (𝑋 ) = H𝑛 (𝑉 ) ⊕ H𝑛 (𝑉 𝑐 ) for all 𝑛.)

Commentary 4.44. The remainder of this section is devoted to proving Proposition 4.42.
Basically, the idea is that every singular 𝑛-simplex Δ𝑛 → 𝑋 can be chopped up into 𝑛-simplices
that lie completely either in 𝑈 1 or in 𝑈 2 .
More precisely, we will construct ‘subdivision’ maps 𝑆 = 𝑆𝑛 : 𝐶𝑛 (𝑋 ) → 𝐶𝑛 (𝑋 ) such that

1. 𝑆 : 𝐶 • (𝑋 ) → 𝐶 • (𝑋 ) is a chain map;
2. 𝑆 ' id are chain homotopic.
3. For any 𝜎 : Δ𝑛 → 𝑋 there exists 𝑘 ≥ 1 such that 𝑆 𝑘 (𝜎) ∈ 𝐶𝑛 (𝑈 1 + 𝑈 2 ).

Since subdivision makes simplices only smaller, we have 𝑆 (𝐶𝑛 (𝑈 1 + 𝑈 2 )) ⊆ 𝐶𝑛 (𝑈 1 + 𝑈 2 ). And

the chain homotopy 𝑇 from id to 𝑆 will also satisfy 𝑇 (𝐶𝑛 (𝑈 1 + 𝑈 2 )) ⊆ 𝐶𝑛+1 (𝑈 1 + 𝑈 2 ).

Assuming these we may complete the proof as follows.

Proof of Proposition 4.42. We show that the map 𝜄 : H𝑛 (𝐶 • (𝑈 1 + 𝑈 2 )) → H𝑛 (𝑋 ) is injective and

surjective:

• Surjectivity: Let 𝑧 ∈ 𝑍𝑛 (𝑋 ) be an 𝑛-cycle in 𝑋 . This can be written as a linear combination

of finitely many 𝑛-simplices 𝜎𝑖 : Δ𝑛 → 𝑋 . Choose 𝑘  0 such that 𝑆 𝑘 (𝜎𝑖 ) ∈ 𝐶𝑛 (𝑈 1 + 𝑈 2 )
for all 𝑖. Hence also 𝑆 𝑘 (𝑧) ∈ 𝐶𝑛 (𝑈 1 + 𝑈 2 ) and this is a cycle. As shown in Exercise 3.1, we
have 𝑆 𝑘 ' id via some chain homotopy 𝜂 hence 𝑧 − 𝑆 𝑘 (𝑧) = 𝜕𝜂 (𝑧) + 𝜂𝜕(𝑧) = 𝜕𝜂 (𝑧). In
particular, 𝑧 ∼ 𝑆 𝑘 𝑧 and in H𝑛 (𝑋 ) we have [𝑧] = 𝜄 ( [𝑆 𝑘 𝑧]).
• Injectivity: Let 𝑤 ∈ 𝑍𝑛 (𝐶 • (𝑈 1 + 𝑈 2 )) such that 𝜄 ( [𝑤]) = 0, that is, 𝑤 = 𝜕(𝑧) for some
𝑧 ∈ 𝐶𝑛+1 (𝑋 ). As before there exists 𝑘 and 𝜂 such that 𝑧 − 𝑆 𝑘 (𝑧) = 𝜕𝜂 (𝑧) + 𝜂𝜕(𝑧). But then
𝜕𝑆 𝑘 (𝑧) = 𝜕(𝑧) − 𝜕 2𝜂 (𝑧) − 𝜕𝜂𝜕(𝑧) = 𝑤 − 𝜕𝜂𝑤 so that [𝑤] = 0 as required.13 
13 Note that in the last step we used that 𝜂 (𝑤) ∈ 𝐶𝑛+1 (𝑈 1 + 𝑈 2 ). Indeed, 𝜂 = − 𝑘−1
𝑖=0 𝑇 𝑆 sends 𝐶𝑛 (𝑈 1 + 𝑈 2 ) into
Í 𝑖
𝐶𝑛+1 (𝑈 1 + 𝑈 2 ).

40
4.4 Barycentric subdivision
Commentary 4.45. Start with a singular simplex 𝜎 : Δ𝑛 → 𝑋 . Since the interiors of 𝑈 1 and 𝑈 2
cover 𝑋 , we have an open cover
𝜎 −1 (𝑈˚ 1 ) ∪ 𝜎 −1 (𝑈˚ 2 )
of Δ𝑛 . Let 𝐴𝑖 = Δ𝑛 \𝜎 −1 (𝑈˚𝑖 ) be the closed complement, 𝑖 = 1, 2, and define a function 𝑓 : Δ𝑛 →
R,
𝑑 (𝑥, 𝐴1 ) + 𝑑 (𝑥, 𝐴2 )
𝑓 (𝑥) = ,
2
the average distance of 𝑥 to each of the two closed subsets. (If one of the 𝐴𝑖 is empty then
𝜎 (Δ𝑛 ) ⊆ 𝑈𝑖 and the discussion is moot.) By compactness of Δ𝑛 , this function attains a mini-
mum 𝛿, which is necessarily > 0 (otherwise we wouldn’t have a cover). One can then easily
check that every simplex [𝑤 0, . . . , 𝑤𝑛 ] ⊆ Δ𝑛 of diameter < 𝛿 is entirely contained in one of the
𝜎 −1 (𝑈˚𝑖 ). Such a number 𝛿 is called a Lebesgue number for the given open cover of Δ𝑛 .
The upshot of this is that as soon as we chop up Δ𝑛 into simplices which are of diameter < 𝛿
then the restriction of 𝜎 to each of these lies in 𝐶𝑛 (𝑈 1 + 𝑈 2 ). Barycentric subdivison is a process
which systematically chops up linear simplices into smaller ones such that, iterating the process,
the diameter tends to 0.

Convention 4.46. We have been using the following notion implicitly already, for example
when discussing the prism operator. Given some euclidean space 𝑉 and elements 𝑣 0, . . . , 𝑣𝑛 ∈ 𝑉 ,
recall that the linear simplex [𝑣 0, . . . , 𝑣𝑛 ] is the subspace
∑︁𝑛 ∑︁
{ 𝑥𝑖 𝑣𝑖 | 𝑥𝑖 ≥ 0, 𝑥𝑖 = 1} ⊆ 𝑉 .
𝑖=0

Remark 4.47. For example, the standard 𝑛-simplex in 𝑉 = R𝑛+1 is obtained by taking the
standard basis vectors as the vertices 𝑣𝑖 . For vectors 𝑣𝑖 in ‘general position’, the linear simplex
is at least homeomorphic to Δ𝑛 (although not isometric). However, weird things can happen
if we don’t require the difference vectors [𝑣 0, 𝑣 1 ], . . . , [𝑣 0, 𝑣𝑛 ] to be linearly independent. (For
example, think of the 𝑛-simplex [𝑣 0, 𝑣 0, . . . , 𝑣 0 ].) So, typically one also requires the vectors to
satisfy this condition.

Definition 4.48. The barycenter of a simplex [𝑣 0, . . . , 𝑣𝑛 ] is the point

1 ∑︁
𝑛
𝑏 := 𝑣𝑖 .
𝑛 + 1 𝑖=0
(Recall that the “barycenter” means center of mass.)

Example 4.49. Here are some examples in low dimensions:

𝑣2

𝑣0 = 𝑏 𝑣0 𝑏 𝑣1 𝑏1 𝑏 𝑏0

𝑣0 𝑣1
𝑏2

41
Commentary 4.50. In fact, one can find the barycenter recursively. Knowing the barycen-
ter 𝑏𝑖 on the 𝑖th face [𝑣 0, . . . , 𝑣ˆ𝑖 , . . . , 𝑣𝑛 ], let ℓ𝑖 be the line connecting 𝑏𝑖 and 𝑣𝑖 . Then 𝑏 is the
intersection of all these lines ℓ𝑖 .
We will not need this fact but it is helpful to have in mind since it will lead our thinking
in subdividing [𝑣 0, . . . , 𝑣𝑛 ]. The pieces will all be cones with apex 𝑏, and base given by one of
the pieces we obtained, inductively, by subdividing one of the faces. (So there will be (𝑛 + 1)!
many simplices.) The only thing to be careful about, as always, are the signs. Here is the
construction.

Construction 4.51. Given a linear (𝑛 − 1)-simplex [𝑤 1, . . . , 𝑤𝑛 ] ⊆ Δ𝑛 we define its barycentric

cone
𝑏 [𝑤 1, . . . , 𝑤𝑛 ] := [𝑏, 𝑤 1, . . . , 𝑤𝑛 ] ⊆ Δ𝑛 .
We extend this to linear combinations of linear (𝑛 − 1)-simplices.
The barycentric subdivision 𝑆 (Δ𝑛 ) ∈ 𝐶𝑛 (Δ𝑛 ) of Δ𝑛 is defined by induction on 𝑛:

• If 𝑛 = 0, we set 𝑆 (Δ0 ) = Δ0 .
• If 𝑛 > 0, we set
𝑛
∑︁
𝑆 (Δ𝑛 ) = 𝑏𝑆 𝜕Δ𝑛 := (−1)𝑖 𝑏𝑆 (𝜕𝑖 Δ𝑛 ).14
𝑖=0

Example 4.52. Compare with the pictures in Example 4.49:

1. 𝑆Δ1 = 𝑏𝑆 [𝑣 1 ] − 𝑏𝑆 [𝑣 0 ] = [𝑏, 𝑣 1 ] − [𝑏, 𝑣 0 ]

2. 𝑆Δ2 = [𝑏, 𝑏 0, 𝑣 2 ] − [𝑏, 𝑏 0, 𝑣 1 ] − [𝑏, 𝑏 1, 𝑣 2 ] + [𝑏, 𝑏 1, 𝑣 0 ] + [𝑏, 𝑏 2, 𝑣 1 ] − [𝑏, 𝑏 2, 𝑣 0 ].
3. In dimension 3 we offer the following picture instead of the lengthy formula (and ig-
noring the signs):

𝑣2

𝑏1
𝑏3
𝑏 𝑏0
𝑣0

𝑣3
𝑏2

𝑣1
14 Note that by induction, 𝑆 (𝜕 Δ𝑛 ) is a linear combination of linear simplices hence the barycentric cone is defined.
𝑖

42
Commentary 4.53. The signs are chosen so that the boundary of 𝑆Δ𝑛 is the (subdivision of
the) boundary of Δ𝑛 , that is, all the internal boundaries cancel out. For example,
𝜕𝑆Δ1 = 𝜕( [𝑏, 𝑣 1 ] − [𝑏, 𝑣 0 ]) = 𝑣 1 − 𝑏 − 𝑣 0 + 𝑏 = 𝑣 1 − 𝑣 0 = 𝜕Δ1 .
We now prove this in general.

Lemma 4.54. 1. 𝜕𝑏 (𝜎) + 𝑏𝜕(𝜎) = 𝜎 for every linear simplex 𝜎.

2. We have 𝜕𝑆Δ𝑛 = 𝑆 (𝜕Δ𝑛 ).

If we ignore the signs, then the first identity says something intuitively clear: the boundary
of the cone consists of the base and the cones on its faces.
Proof. Let 𝜎 = [𝑤 1, . . . , 𝑤𝑛 ]. Then
𝜕𝑏 [𝑤 1, . . . , 𝑤𝑛 ] = 𝜕[𝑏, 𝑤 1, . . . , 𝑤𝑛 ]
𝑛
∑︁
= (−1)𝑖 𝜕𝑖 [𝑏, 𝑤 1, . . . , 𝑤𝑛 ]
𝑖=0
= [𝑤 1, . . . , 𝑤𝑛 ] − 𝑏𝜕[𝑤 1, . . . , 𝑤𝑛 ]
and the first identity is established.
The second identity follows from the first, by induction. Namely, for 𝑛 = 0 we have zero
on both sides. And for 𝑛 > 0 we have
𝜕𝑆Δ𝑛 = 𝜕𝑏𝑆 𝜕Δ𝑛 = (id −𝑏𝜕)𝑆 𝜕Δ𝑛 = 𝑆 𝜕Δ𝑛 − 𝑏𝑆 𝜕 2 Δ𝑛 = 𝑆 𝜕Δ𝑛
where we used the first identity in the second equality, and induction in the third. 

Definition 4.55. Let 𝑋 be a topological space and 𝜎 : Δ𝑛 → 𝑋 a singular 𝑛-simplex. We define

the barycentric subdivision of 𝜎 to be the 𝑛-chain
𝑆 (𝜎) := 𝜎∗𝑆Δ𝑛 ∈ 𝐶𝑛 (𝑋 ).
By linear extension we obtain a homomorphism
𝑆 : 𝐶𝑛 (𝑋 ) → 𝐶𝑛 (𝑋 )
Corollary 4.56. 𝑆 : 𝐶 • (𝑋 ) → 𝐶 • (𝑋 ) is a chain map.

Proof. We have:
𝜕𝑆 (𝜎) = 𝜕𝜎∗𝑆Δ𝑛 definition of 𝑆
= 𝜎∗ 𝜕𝑆Δ 𝑛
𝜎∗ chain map
= 𝜎∗𝑆 (𝜕Δ ) 𝑛
Lemma 4.54
𝑛
∑︁
= (−1)𝑖 𝜎∗𝑆 (𝜕𝑖 Δ𝑛 )
𝑖=0
𝑛
∑︁
= (−1)𝑖 𝑆 (𝜕𝑖 𝜎) definition of 𝑆
𝑖=0
= 𝑆 (𝜕𝜎)

43
This establishes 𝜕𝑆 = 𝑆 𝜕, as required. 

Proposition 4.57. The barycentric subdivision is chain homotopic to the identity: 𝑆 ' id : 𝐶 • (𝑋 ) →
𝐶 • (𝑋 ).

Proof. That is, we need to define a chain homotopy 𝑇 : 𝐶𝑛 (𝑋 ) → 𝐶𝑛+1 (𝑋 ) from 𝑆 to id. The
idea is to chop up the prism Δ𝑛 × [0, 1] into (𝑛 + 1)-simplices in such a way that the ‘bottom
face’ stays intact (the ‘identity’) and the ‘top face’ is subdivided:

𝑣 01 𝑏 𝑣 11

Δ1 × [0, 1]:

𝑣 00 𝑣 10

Let us write Δ𝑛0 for the bottom face Δ𝑛 × {0}, Δ𝑛1 for the top face Δ𝑛 × {1}, and 𝑏 for the
barycentric cone of Δ𝑛1 . Then we set, recursively,

𝑇 (Δ𝑛 ) := 𝑏Δ𝑛0 − 𝑏𝑇 𝜕Δ𝑛0 ∈ 𝐶𝑛+1 (Δ𝑛 × [0, 1]).

• So, for example, for 𝑛 = 0 we get 𝑇 (Δ0 ) = [𝑏, 𝑣 00 ] = [𝑣 01, 𝑣 00 ].

• And if 𝑛 = 1 we get 𝑇 (Δ1 ) = [𝑏, 𝑣 00, 𝑣 10 ] − [𝑏, 𝑣 11, 𝑣 10 ] + [𝑏, 𝑣 01, 𝑣 00 ].
• Here is a higher-dimensional example (with the appropriate signs omitted):

𝑤2
𝑏1
𝑤0 𝑏
𝑏0
𝑏2

𝑤1
𝑣2

𝑣0

Δ2 × [0, 1] 𝑣1

We now verify the following identity reminiscent of the chain homotopy condition (4.23):

(4.58) 𝜕𝑇 (Δ𝑛 ) + 𝑇 (𝜕Δ𝑛0 ) = Δ𝑛0 − 𝑆Δ𝑛1

44
Indeed, for 𝑛 = 0 both sides equal [𝑣 00 ] − [𝑣 01 ], and for 𝑛 > 0 we have

𝜕𝑇 (Δ𝑛 ) = 𝜕𝑏Δ𝑛0 − 𝜕𝑏𝑇 𝜕Δ𝑛0 definition of 𝑇 (Δ𝑛 )

= Δ𝑛0 − 𝑏𝜕Δ𝑛0 − 𝑇 𝜕Δ𝑛0 + 𝑏𝜕𝑇 𝜕Δ𝑛0 Lemma 4.54
= Δ𝑛0 − 𝑇 𝜕Δ𝑛0 + 𝑏 (𝜕𝑇 𝜕Δ𝑛0 − 𝜕Δ𝑛0 )
= Δ𝑛0 − 𝑇 𝜕Δ𝑛0 − 𝑏 (𝑆 𝜕Δ𝑛1 + 𝑇 𝜕 2 Δ𝑛0 ) induction
= Δ𝑛0 − 𝑇 𝜕Δ𝑛0 − 𝑆Δ𝑛1 𝜕 2 = 0, definition of 𝑆
𝜎
Given a singular 𝑛-simplex 𝜎 : Δ𝑛 → 𝑋 let 𝜎 0 : Δ𝑛 × [0, 1] → Δ𝑛 −
→ 𝑋 be the composition of 𝜎
with the projection away from the second factor. If we define

𝑇 : 𝐶𝑛 (𝑋 ) → 𝐶𝑛+1 (𝑋 )
𝜎 ↦→ 𝜎∗0𝑇 (Δ𝑛 )

then (4.58) implies that T defines a chain homotopy 𝑆 ' id as claimed. 

Commentary 4.59. In view of Commentary 4.45, the last thing to establish in Commen-
tary 4.44 is that the diameters of the simplices in 𝑆 𝑘 Δ𝑛 tend to zero as 𝑘 → ∞, that is, as we
repeatedly subdivide. This follows from the next lemma.

Lemma 4.60. Let [𝑤 0, . . . , 𝑤𝑛 ] be a simplex in the barycentric subdivision of [𝑣 0, . . . , 𝑣𝑛 ]. Then

𝑛
diam( [𝑤 0, . . . , 𝑤𝑛 ]) ≤ diam( [𝑣 0, . . . , 𝑣𝑛 ]).
𝑛+1
Exercise 4.61. Verify this by hand in low dimensions, see Example 4.49!

Proof. If 𝑛 = 0 the claim is true since [𝑤 0 ] = [𝑣 0 ] and both have diameter 0. So from now on
we assume 𝑛 > 0.
We start with the following observation about any linear simplex [𝑥 0, . . . , 𝑥𝑛 ] whatsoever:

For every 𝑥, its maximum distance to points in the simplex is attained at a vertex 𝑥𝑖 .

Indeed, let 𝑦 ∈ [𝑥 0, . . . , 𝑥𝑛 ] with k𝑥 − 𝑦 k maximal. So we may write 𝑦 = 𝑡𝑖 𝑥𝑖 with 𝑡𝑖 = 1

Í Í
and 𝑡𝑖 ≥ 0. Then we have
∑︁ ∑︁ ∑︁
k𝑥 − 𝑦 k = k𝑥 − 𝑡𝑖 𝑥 𝑖 k = k 𝑡𝑖 (𝑥 − 𝑥𝑖 ) k ≤ 𝑡𝑖 k𝑥 − 𝑥𝑖 k ≤ maxk𝑥 − 𝑥𝑖 k

with equality if 𝑦 is one of those vertices 𝑥𝑖 with k𝑥 − 𝑥𝑖 k maximal. This establishes the obser-
vation.
In particular, applying this observation twice we see that the diameter of [𝑤 0, . . . , 𝑤𝑛 ] is
the length of the longest edge [𝑥, 𝑦]. Let us distinguish two cases:

• If none of 𝑥, 𝑦 is the barycenter 𝑏 then they must be vertices of a simplex in the barycen-
tric subdivision of one of the faces [𝑣 0, . . . , 𝑣ˆ𝑖 , . . . , 𝑣𝑛 ]. By induction we then have

𝑛−1 𝑛
diam[𝑤 0, . . . , 𝑤𝑛 ] = k𝑥 − 𝑦 k ≤ diam[𝑣 0, . . . , 𝑣ˆ𝑖 , . . . , 𝑣𝑛 ] ≤ diam[𝑣 0, . . . , 𝑣𝑛 ]
𝑛 𝑛+1

45
 • If, say, 𝑥 = 𝑏, then 𝑦 lies on some face of [𝑣 0, . . . , 𝑣𝑛 ] and the observation above allows
us to assume 𝑦 = 𝑣𝑖 is one of the vertices of that face. Let 𝑏𝑖 be the barycenter of
[𝑣 0, . . . , 𝑣ˆ𝑖 , . . . , 𝑣𝑛 ], that is:
1 ∑︁ 1 ∑︁ 1 𝑛
𝑏𝑖 = 𝑣𝑗, hence 𝑏= 𝑣𝑗 = 𝑣𝑖 + 𝑏𝑖 .
𝑛 𝑗≠𝑖 𝑛+1 𝑗 𝑛+1 𝑛+1

𝑥 =𝑏 𝑏𝑖

𝑦 = 𝑣𝑖

It follows that
𝑛 𝑛
diam[𝑤 0, . . . , 𝑤𝑛 ] = k𝑣𝑖 − 𝑏 k ≤ k𝑣𝑖 − 𝑏𝑖 k ≤ diam[𝑣 0, . . . , 𝑣𝑛 ]
𝑛+1 𝑛+1


5 Applications
5.1 Fundamental classes for spheres
Commentary 5.1. We saw in Corollary 4.13 that the homology of the sphere 𝑆 𝑘 (with 𝑘 ≥ 1)
is a copy of Z in degrees 0, 𝑘 (and vanishes in all other degrees). A generator of H𝑘 (𝑆 𝑘 ) is
called a fundamental class. Our goal in this section is to describe explicitly fundamental classes
(that is, cycles representing them) for all spheres.
More generally, orientable compact connected manifolds 𝑀 of dimension 𝑘 have H𝑘 (𝑀) 
Z, and a generator is called a fundamental class for 𝑀. It may be obtained by choosing a
triangulation of the manifold and then choosing a compatible orientation of the simplices in the
triangulation. (The fact that a triangulation exists is due to compactness. And that a compatible
orientation exists is precisely the orientability condition. For disconnected manifolds, one can
do this process on each connected component separately.)

Example 5.2. In Example 4.12 we already described a fundamental class for 𝑆 1 . Viewing the
circle as obtained from an interval by identifying the endpoints, the quotient map

Δ1 = [0, 1] → [0, 1]/(0 ∼ 1) ≈ 𝑆 1

is a 1-cycle which generates H1 (𝑆 1 ).

Commentary 5.3. When we try to reproduce this construction for 𝑆 2 we run into a problem.
The singular 2-simplex given by the quotient map

𝜎 : Δ2 → Δ2 /𝜕Δ2 ≈ 𝑆 2

46
has boundary
𝜕𝜎 = ∗ − ∗ + ∗ = ∗
where ∗ : Δ1 → 𝑆 2 denotes the constant path at the point to which 𝜕Δ2 was contracted. In
other words, 𝜎 is not a cycle.
In fact, this construction gives a fundamental class for 𝑆 𝑘 precisely when 𝑘 is odd. Below,
we will instead describe a construction that works in all dimensions.

Example 5.4. We gave two Δ-complex structures on 𝑆 1 in Examples 2.9 and 2.10. The first
one is related to the cycle of Example 5.2. The second one suggests looking at the cycle

𝜎 := 𝜎+ − 𝜎−

instead, where 𝜎+ : Δ1 → 𝑆 1 picks out the upper hemicircle, and 𝜎− : Δ1 → 𝑆 1 picks out the
lower hemicircle, running in the opposite direction.

By Theorem 3.28 (or a direct verification), this cycle represents a fundamental class. This
is the cycle that we will generalize to all dimensions.

Remark 5.5. Recall that 𝑆 𝑘 ≈ 𝜕Δ𝑘+1 are homeomorphic, see Exercise 2.7. The 𝑘-boundary
𝜕(Δ𝑘+1 ) ∈ 𝐶𝑘 (Δ𝑘+1 ) actually lies in the subgroup 𝐶𝑘 (𝜕Δ𝑘+1 ). Writing Δ𝑘+1 = [𝑣 0, . . . , 𝑣𝑘+1 ] we
therefore have explicitly

𝑘+1
∑︁
𝜕Δ 𝑘+1
= (−1)𝑖 [𝑣 0, . . . , 𝑣ˆ𝑖 , . . . , 𝑣𝑘+1 ].
𝑖=0

Proposition 5.6. The cycle 𝜕Δ𝑘+1 ∈ 𝐶𝑘 (𝜕Δ𝑘+1 ) represents a generator in homology.

Proof. It is clear that this is a cycle, since it is a boundary in the chain complex 𝐶 • (Δ𝑘+1 ). We
now prove the statement by induction on 𝑘, using Mayer-Vietoris.
Choose open subspaces 𝑈 1, 𝑈 2 ⊂ Δ𝑘+1 as follows (see also the picture).

47
 𝑈 1 is an open neighborhood of the last face 𝜕𝑘+1 Δ𝑘+1 which deformation retracts onto the
latter. 𝑈 2 is an open neighborhood of the remaining faces ∪0≤𝑖 ≤𝑘 𝜕𝑖 Δ𝑘+1 which deformation
retracts onto these. We choose them in such a way that 𝑈 1 ∩ 𝑈 2 deformation retracts onto
𝜕(𝜕𝑘+1 Δ𝑘+1 ) = 𝜕[𝑣 0, . . . , 𝑣𝑘 ] and 𝑈 1 ∪𝑈 2 deformation retracts onto 𝜕[𝑣 0, . . . , 𝑣𝑘+1 ]. By induction
hypothesis, we know that H𝑘−1 (𝑈 1 ∩ 𝑈 2 )  Z is generated by
𝑘
∑︁
(5.7) 𝜕[𝑣 0, . . . , 𝑣𝑘 ] = (−1)𝑖 [𝑣 0, . . . , 𝑣ˆ𝑖 , . . . , 𝑣𝑘 ].
𝑖=0

In the Mayer-Vietoris l.e.s. for the pair 𝑈 1, 𝑈 2 covering 𝑈 1 ∪ 𝑈 2 ' 𝜕Δ𝑘+1 , the connecting
homomorphism
H𝑘 (𝑈 1 ∪ 𝑈 2 ) → H𝑘−1 (𝑈 1 ∩ 𝑈 2 )
is an isomorphism and we only need to show that 𝜕Δ𝑘+1 is sent to the generator (5.7) or its
negative.
We shall now use the explicit construction of the connecting homomorphism, see Propo-
sition 4.37. The cycle 𝜕Δ𝑘+1 has an obvious lift to 𝐶𝑘+1 (𝑈 1 ) ⊕ 𝐶𝑘+1 (𝑈 2 ), namely
𝑘
!
∑︁
(−1)𝑘+1 [𝑣 0, . . . , 𝑣𝑘 ], (−1)𝑖 [𝑣 0, . . . , 𝑣ˆ𝑖 , . . . , 𝑣𝑘+1 ] .
𝑖=0

It follows that its image under the connecting homomorphism is the unique (𝑘 − 1)-cycle 𝜎
in 𝑈 1 ∩ 𝑈 2 which satisfies
   𝑘
∑︁ 
(𝑖 1 )∗ (𝜎) , −(𝑖 2 )∗ (𝜎) = (−1) 𝑘+1
𝜕[𝑣 0, . . . , 𝑣𝑘 ] , (−1) 𝜕[𝑣 0, . . . , 𝑣ˆ𝑖 , . . . , 𝑣𝑘+1 ] .
𝑖

𝑖=0

It is clear that 𝜎 = (−1)𝑘+1 𝜕[𝑣 0, . . . , 𝑣𝑘 ] does the trick. 

Remark 5.8. Let 𝑆 +𝑘 (resp. 𝑆 −𝑘 ) be the upper (resp. lower) hemisphere of 𝑆 𝑘 . Choose homeo-
morphisms
≈ ≈
𝜎+ : Δ𝑘 −
→ 𝑆 +𝑘 , 𝜎− : Δ𝑘 −
→ 𝑆 −𝑘
such that

• both 𝜎+ and 𝜎− map the boundary 𝜕Δ𝑘 homeomorphically onto the equator 𝑆 +𝑘 ∩ 𝑆 −𝑘 ; and
𝜎+ (𝜎− ) −1
• the composition 𝜕Δ𝑘 −−→ 𝑆 +𝑘 ∩ 𝑆 −𝑘 −−−−−→ 𝜕Δ𝑘 is the identity.

48
Corollary 5.9. The cycle 𝜎+ − 𝜎− ∈ 𝐶𝑘 (𝑆 𝑘 ) represents a fundamental class for 𝑆 𝑘 .

Proof. For 𝑘 = 1 we have already seen this in Example 5.4. We may therefore assume 𝑘 > 1.
Our two assumptions on 𝜎+ and 𝜎− imply that 𝜎+ −𝜎− is indeed a cycle. Choose open neighbor-
hoods 𝑈 +, 𝑈 − of the two hemispheres, respectively, which deformation retract on to them and
whose intersection deformation retracts onto the equator. The connecting homomorphism

H𝑘 (𝑆 𝑘 ) → H𝑘−1 (𝑈 + ∩ 𝑈 − )

in the Mayer-Vietoris sequence is an isomorphism and sends 𝜎+ − 𝜎− to 𝜕𝜎+ = 𝜕𝜎− . We saw in

Proposition 5.6 that 𝜕Δ𝑘 is a generator of H𝑘−1 (𝜕Δ𝑘 ) so we win. 

5.2 The Jordan curve theorem

Recall that a Jordan curve is a simple closed curve in R2 . The Jordan curve theorem, due to
Jordan (1887), states:

Every Jordan curve splits the plane into two regions.

One of the two regions is necessarily bounded and is thus interpreted as the interior, while the
other region is necessarily unbounded, thus interpreted to be the exterior. The Jordan curve is
then the boundary of each of these regions.
The theorem is intuitively clear if one imagines a ‘nice’ (say smooth) curve. However,
imagine for example fractal curves or an Osgood curve. In these cases, the intuition breaks
down. Luckily, homology comes to our rescue! (Exercise 3.6 treats another class of ‘nice’
curves.)

Proposition 5.10 (Jordan Curve Theorem). Let 𝛾 : 𝑆 1 ↩→ R2 be an injective continuous map with
image 𝐶. Then:
 Z2 : 𝑛 = 0


H𝑛 (R2 \𝐶) = Z : 𝑛 = 1



0

:𝑛 >1


Remark 5.11. According to Proposition 3.14 (and the fact that R2 \𝐶 is locally path-connected),
the part H0 (R2 \𝐶) = Z2 of the proposition says precisely that the complement of 𝐶 has two
connected components.
We will translate the problem as follows. Let R2 ↩→ R2 ∪ {∞} ≈ 𝑆 2 be the one-point
compactification. We are going to prove the following:
(
2 Z2 : 𝑛 = 0
(5.12) H𝑛 (𝑆 \𝐶) =
0 :𝑛 >0

49
Exercise 5.13. Let 𝑋 = 𝑆 2 \𝐶 and 𝑈 = R2 \𝐶, and let 𝑉 be an open disk around ∞ which
does not meet 𝐶. Apply Mayer-Vietoris to the cover 𝑋 = 𝑈 ∪ 𝑉 to deduce Proposition 5.10
from (5.12).

We start by proving the following auxiliary result (which we are going to apply to parts
of the curve 𝐶). This says in particular that a non-intersecting path in R2 cannot separate it.

Lemma 5.14. Let 𝜅 : [0, 1] → 𝑆 2 be an injective continuous map with image 𝐷 = 𝜅 ( [0, 1]). Then
(
Z :𝑛 =0
H𝑛 (𝑆 2 \𝐷) =
0 :𝑛 >0

Proof. For an interval 𝐼 ⊆ [0, 1] we set 𝐷 𝐼 := 𝜅 (𝐼 ) so that 𝐷 = 𝐷 [0,1] . Let 𝑈 = 𝑆 2 \𝐷 [0,1/2] and
𝑉 = 𝑆 2 \𝐷 [1/2,1] . Note that

𝑈 ∩ 𝑉 = 𝑆 2 \𝐷, 𝑈 ∪ 𝑉 = 𝑆 2 \𝐷 [1/2,1/2] ≈ R2 ' ∗.

The Mayer-Vietoris l.e.s. gives isomorphisms for 𝑛 ≥ 1,

H𝑛 (𝑆 2 \𝐷)  H𝑛 (𝑆 2 \𝐷 [0,1/2] ) ⊕ H𝑛 (𝑆 2 \𝐷 [1/2,1] ),

and a short exact sequence

0 → H0 (𝑆 2 \𝐷) → H0 (𝑆 2 \𝐷 [0,1/2] ) ⊕ H0 (𝑆 2 \𝐷 [1/2,1] ) → Z → 0.

Assume that for some 𝑛 ≥ 1 there exists 𝜎 ∈ H𝑛 (𝑆 2 \𝐷) non-zero. By the isomorphism just
exhibited it remains non-zero in at least one of the two groups H𝑛 (𝑆 2 \𝐷 𝐼1 ), 𝐼 1 being the first
or second half of the interval. Repeating this argument we find a nested sequence of intervals

[0, 1] = 𝐼 0 ⊃ 𝐼 1 ⊃ 𝐼 2 ⊃ 𝐼 3 ⊃ · · ·

50
with = {𝑝} and such that 𝜎 ≠ 0 in all H𝑛 (𝑆 2 \𝐷 𝐼ℓ ).
Ñ
ℓ ≥0 𝐼 ℓ

But note that 𝑆 2 \𝐷 [𝑝,𝑝 ] = 𝑆 2 \{𝜅 (𝑝)} ≈ R2 is contractible hence 𝜎 vanishes in H𝑛 (𝑆 2 \{𝜅 (𝑝)}).
Let 𝜏 ∈ 𝐶𝑛+1 (𝑆 2 \𝐷 [𝑝,𝑝 ] ) such that 𝜕𝜏 = 𝜎. Write 𝜏 as linear combination of singular simplices.
Each of these has compact image in 𝑆 2 \𝐷 [𝑝,𝑝 ] . The union of these images is covered by the
open subsets (𝑆 2 \𝐷 𝐼ℓ )ℓ so by compactness, there exists ℓ such that 𝜏 ∈ 𝐶𝑛+1 (𝑆 2 \𝐷 𝐼ℓ ). But then
𝜎 = 0 ∈ H𝑛 (𝑆 2 \𝐷 𝐼ℓ ), contradicting our assumption. We deduce that H𝑛 (𝑆 2 \𝐷) = 0 for all
𝑛 > 0.
We argue similarly in degree 𝑛 = 0. Assume 𝑥, 𝑦 ∈ 𝑆 2 \𝐷 are two points in different path-
connected components. Then we find a nested sequence of intervals 𝐼 ℓ as before such that 𝑥, 𝑦
are in different path-connected components of 𝑆 2 \𝐷 𝐼ℓ for all ℓ. Since 𝑆 2 \{𝜅 (𝑝)} is contractible,
it must contain a path connecting 𝑥 with 𝑦. By compactness again, this path misses 𝐷 𝐼ℓ for
ℓ  0 and hence 𝑥 and 𝑦 lie in the same path-connected component of 𝑆 2 \𝐷 𝐼ℓ for ℓ  0. A
contradiction. We deduce that H0 (𝑆 2 \𝐷) = Z. 

Proof of Proposition 5.10. By Remark 5.11, we should compute the homology of 𝑆 2 \𝐶. Choose
𝑆 +1 and 𝑆 −1 to be the upper and lower hemicircles so that 𝑆 +1 ∩ 𝑆 −1 = 𝑆 0 . We apply MV with

• 𝑋 = 𝑆 2 \𝛾 (𝑆 0 ),
• 𝑈 + = 𝑆 2 \𝛾 (𝑆 +1 ),
• 𝑈 − = 𝑆 2 \𝛾 (𝑆 −1 ),
• 𝑈 + ∩ 𝑈 − = 𝑆 2 \𝐶.

As 𝑆 ±1 ≈ [0, 1] we know the homology groups of 𝑈 ± from Lemma 5.14. And as 𝑆 2 \𝛾 (𝑆 0 ) ' 𝑆 1
we also know its homology groups. Thus Mayer-Vietoris gives the result.
In more detail, for 𝑛 > 0 we have an exact sequence

H𝑛+1 (𝑆 2 \𝛾 (𝑆 0 )) → H𝑛 (𝑆 2 \𝐶) → H𝑛 (𝑈 + ) ⊕ H𝑛 (𝑈 − ),

in which both outer terms vanish. Hence so does H𝑛 (𝑆 2 \𝐶).

And for 𝑛 = 0 we have an exact sequence

0 → H1 (𝑆 2 \𝛾 (𝑆 0 )) → H0 (𝑆 2 \𝐶) → H0 (𝑈 + ) ⊕ H0 (𝑈 − ) → H0 (𝑆 2 \𝛾 (𝑆 0 )) → 0

which is of the form

(1 1)
0 → Z → H0 (𝑆 2 \𝐶) → Z2 −−−−→ Z → 0.
It follows that H0 (𝑆 2 \𝐶) = Z2 , as claimed. 

51
5.3 Relative homology
Given a subspace 𝐴 ⊆ 𝑋 , we are interested in the relation between H𝑛 (𝐴) and H𝑛 (𝑋 ). As
you observed in Exercise 2.4, the induced map H𝑛 (𝐴) → H𝑛 (𝑋 ) is not injective in general
(nor surjective, of course). The goal of this section is to give a precise measure of the failure
of injectivity and surjectivity. This will be expressed in terms of the relative homology groups
H• (𝑋, 𝐴) and we will prove excision, a powerful tool in computing these groups.

Commentary 5.15. Before passing to homology, that is, at the level of singular chains,
𝐶𝑛 (𝐴) ⊆ 𝐶𝑛 (𝑋 )
is a subgroup. Namely, tautologically, the free abelian group on singular 𝑛-simplices in 𝐴 ⊆ 𝑋 .
Of course, this actually defines a sub-chain complex 𝐶 • (𝐴) ⊆ 𝐶 • (𝑋 ).

Definition 5.16. We define 𝐶𝑛 (𝑋, 𝐴) to be the quotient 𝐶𝑛 (𝑋 )/𝐶𝑛 (𝐴) with the differential
𝐶𝑛 (𝑋, 𝐴) → 𝐶𝑛−1 (𝑋, 𝐴) induced from 𝜕 : 𝐶𝑛 (𝑋 ) → 𝐶𝑛−1 (𝑋 ). We define the relative homology
groups as the homology groups of the relative singular chain complex 𝐶 • (𝑋, 𝐴):
H𝑛 (𝑋, 𝐴) := H𝑛 (𝐶 • (𝑋, 𝐴))
Commentary 5.17. Let us take a moment to break this construction down a bit.
• Let us call an 𝑛-chain 𝑧 ∈ 𝐶𝑛 (𝑋 ) a relative 𝑛-cycle if 𝜕(𝑧) ∈ 𝐶𝑛−1 (𝐴). For example, for
a singular 𝑛-simplex 𝜎 : Δ𝑛 → 𝑋 this just means that image of the boundary 𝜕Δ𝑛 is
contained in the subspace 𝐴.
• Let us also call an 𝑛-chain 𝑧 ∈ 𝐶𝑛 (𝑋 ) a relative 𝑛-boundary if it is homologous to an 𝑛-
chain in 𝐴, that is if there exist 𝑤 ∈ 𝐶𝑛+1 (𝑋 ) and 𝑎 ∈ 𝐶𝑛 (𝐴) such that 𝑧 = 𝑎 + 𝜕𝑤. Note
that every relative 𝑛-boundary is a relative 𝑛-cycle since 𝜕𝑧 = 𝜕𝑎.
By construction,
relative 𝑛-cycles
H𝑛 (𝑋, 𝐴)  .
relative 𝑛-boundaries
Therefore, the intuition is that H𝑛 (𝑋, 𝐴) measures the homology of 𝑋 ‘discarding 𝐴’. We will
see later that in good cases this intuition can be made precise (Theorem 5.23, Proposition 5.28).

A relative 1-cycle that is not a cycle in 𝑋

We now come back to the question raised at the start of this section. The long exact sequence
for the pair (𝑋, 𝐴) appearing in the following result provides a precise way of measuring the
failure of the maps H𝑛 (𝐴) → H𝑛 (𝑋 ) to be isomorphisms.

52
Corollary 5.18. There is an exact sequence

··· H𝑛 (𝐴) H𝑛 (𝑋 ) H𝑛 (𝑋, 𝐴)

H𝑛−1 (𝐴) H𝑛−1 (𝑋 ) H𝑛−1 (𝑋, 𝐴) ···

Proof. By construction, we have a short exact sequence of chain complexes

0 → 𝐶 • (𝐴) → 𝐶 • (𝑋 ) → 𝐶 • (𝑋, 𝐴) → 0.

Therefore the claim follows from Proposition 4.37. 

Exercise 5.19. Show that the connecting homomorphism takes a homology class [𝑧] ∈ H𝑛 (𝑋, 𝐴)
represented by a relative cycle 𝑧 ∈ 𝐶𝑛 (𝑋 ) (cf. Commentary 5.17) to [𝜕𝑧] ∈ H𝑛−1 (𝐴).

Exercise 5.20. Let 𝑥 ∈ 𝑋 be a point and set 𝐴 = {𝑥 }. We write (𝑋, 𝑥) for the pair (𝑋, {𝑥 }).
Show that there is an isomorphism

H𝑛 (𝑋, 𝑥)  H̃𝑛 (𝑋 )

for all 𝑛.

Remark 5.21. Let 𝑘 ≥ 1. Recall (Commentary 5.3) that the canonical singular 𝑘-simplex

𝜎 : Δ𝑘 → Δ𝑘 /𝜕Δ𝑘 ≈ 𝑆 𝑘 ,

given by the quotient map, described a fundamental class for 𝑆 𝑘 precisely when 𝑘 was odd. The
problem for even 𝑘 may seem silly: It’s that 𝜎 is not a cycle. We can rectify that using relative
homology (and reduced homology, given Exercise 5.20). Indeed, 𝜎 is a perfectly acceptable
relative 𝑘-cycle for the pair
(Δ𝑘 /𝜕Δ𝑘 , 𝜕Δ𝑘 /𝜕Δ𝑘 ) ≈ (𝑆 𝑘 , ∗).
Moreover, its class in homology generates the group

H𝑘 (𝑆 𝑘 , ∗).
∼
Note that H𝑘 (𝑆 𝑘 ) −
→ H𝑘 (𝑆 𝑘 , ∗) is an isomorphism so, in some sense, 𝜎 does define a fundamental
class in all dimensions. Can you explain the apparent contradiction with Commentary 5.3?

Commentary 5.22. We now come to excision which is essentially equivalent to Mayer-

Vietoris (Theorem 4.10). The way to think about it is as follows. If it is indeed true that
H𝑛 (𝑋, 𝐴) ‘ignores’ what’s going on inside of 𝐴 then, surely, this group won’t change if I re-
move (=excise) a piece of 𝐴?

Theorem 5.23 (Excision). Let 𝑍 ⊆ 𝐴 ⊆ 𝑋 be subspaces and assume that 𝑍¯ ⊆ 𝐴.

˚ Then

H𝑛 (𝑋, 𝐴)  H𝑛 (𝑋 \𝑍, 𝐴\𝑍 ).

53
Proof. We are going to employ the following trick. Set 𝐵 = 𝑋 \𝑍 . Note that:

• 𝐴 ∩ 𝐵 = 𝐴\𝑍 ,
• 𝐴˚ ∪ 𝐵˚ = 𝐴˚ ∪ (𝑋 \𝑍¯ ) = 𝑋 by our assumption, and
𝐶𝑛 (𝐵) 𝐶𝑛 (𝐴+𝐵)
• 𝐶𝑛 (𝑋 \𝑍, 𝐴\𝑍 ) = 𝐶𝑛 (𝐵, 𝐴 ∩ 𝐵) = 𝐶𝑛 (𝐴∩𝐵)  𝐶𝑛 (𝐴) by the second isomorphism theorem.

We now have a ‘morphism of short exact sequences of chain complexes’ (meaning simply, a
commutative diagram albeit in three dimensions):

0 𝐶 • (𝐴) 𝐶 • (𝐴 + 𝐵) 𝐶 • (𝑋 \𝑍, 𝐴\𝑍 ) 0

=

0 𝐶 • (𝐴) 𝐶 • (𝑋 ) 𝐶 • (𝑋, 𝐴) 0

Passing to homology we get a morphism of chain complexes

H𝑛 (𝐴) H𝑛 (𝐴 + 𝐵) H𝑛 (𝑋 \𝑍, 𝐴\𝑍 ) H𝑛−1 (𝐴) H𝑛−1 (𝐴 + 𝐵)

= =

H𝑛 (𝐴) H𝑛 (𝑋 ) H𝑛 (𝑋, 𝐴) H𝑛−1 (𝐴) H𝑛−1 (𝑋 )

in which both rows are exact. By barycentric subdivision, the second and the fifth vertical
arrow are isomorphisms, see Proposition 4.42. By the (aptly-named) five lemma (Exercise 4.1),
the middle vertical arrow is an isomorphism as well. This shows the claim. 

Commentary 5.24. Recall that a topological manifold of dimension 𝑘 is a Hausdorff space so

that every point has an open neighborhood homeomorphic to R𝑘 . Of course, every smooth
manifold is a topological manifold.

Corollary 5.25. Let 𝑀 be a 𝑘-dimensional topological manifold and 𝑥 ∈ 𝑀 a point. Then

(
Z :𝑛 =𝑘
H𝑛 (𝑀, 𝑀\𝑥)  H𝑛 (R𝑘 , R𝑘 \∗) 
0 :𝑛 ≠𝑘

Proof. Let 𝑥 ∈ 𝑈 be an open neighborhood such that 𝑈 ≈ R𝑘 . Then Excision gives the first
isomorphism (with 𝑋 = 𝑀, 𝐴 = 𝑀\𝑥, and 𝑍 = 𝑀\𝑈 ). For the second isomorphism we consider
the long exact sequence of the pair (R𝑘 , R𝑘 \∗). (Note that the claim is clear if 𝑘 = 0 so we
will assume 𝑘 ≥ 1 from now on.) Using that R𝑘 \∗ ' 𝑆 𝑘−1 and that R𝑘 is contractible we get
H𝑛 (R𝑘 , R𝑘 \∗) = 0 for 𝑛 > 𝑘, and for 𝑛 < 𝑘, while for 𝑛 = 𝑘 we find an exact sequence

0 → H𝑘 (R𝑘 , R𝑘 \∗) → H𝑘−1 (𝑆 𝑘−1 ) → H𝑘−1 (R𝑘 ).

If 𝑘 > 1 the last term vanishes, giving the required isomorphism. If 𝑘 = 1, the last map
identifies with ( 1 1 ) : Z2 → Z giving, again, the required isomorphism. 

The next result generalizes Corollary 4.20.

54
Corollary 5.26 (Invariance of domain, revisited). Let 𝑈 ⊆ R𝑘 , 𝑉 ⊆ Rℓ be non-empty open
subsets. If 𝑈 ≈ 𝑉 then 𝑘 = ℓ.

We conclude this section with another version of the intuition, according to which relative
homology H𝑛 (𝑋, 𝐴) ‘discards 𝐴’.

Definition 5.27. A pair (𝑋, 𝐴) is called good if

1. 𝐴 ⊆ 𝑋 is a closed subset, and
2. there exists an open neighborhood 𝐴 ⊆ 𝑉 which deformation retracts onto 𝐴.

Proposition 5.28. Let (𝑋, 𝐴) be a good pair. Then the quotient map 𝑋 → 𝑋 /𝐴 induces isomorphisms

H𝑛 (𝑋, 𝐴)  H𝑛 (𝑋 /𝐴, 𝐴/𝐴)  H̃𝑛 (𝑋 /𝐴).

Proof. The long exact sequence for the pairs (𝑋, 𝐴) and (𝑋, 𝑉 ), respectively, and the five lemma
give an isomorphism
∼
H𝑛 (𝑋, 𝐴) −
→ H𝑛 (𝑋, 𝑉 ).
Excision gives a further isomorphism H𝑛 (𝑋, 𝑉 )  H𝑛 (𝑋 \𝐴, 𝑉 \𝐴). Combining the two we
have

(5.29) H𝑛 (𝑋, 𝐴)  H𝑛 (𝑋 \𝐴, 𝑉 \𝐴).

Now, repeat this argument with (𝑋 /𝐴, 𝐴/𝐴) instead of (𝑋, 𝐴), yielding

(5.30) H𝑛 (𝑋 /𝐴, 𝐴/𝐴)  H𝑛 ((𝑋 /𝐴)\(𝐴/𝐴), (𝑉 /𝐴)\(𝐴/𝐴)) = H𝑛 (𝑋 \𝐴, 𝑉 \𝐴)

Finally, (5.29) and (5.30) together prove the Proposition. 

Example 5.31. Clearly, (Δ𝑘 , 𝜕Δ𝑘 ) is a good pair. We deduce that

(
Z :𝑛 =𝑘
H𝑛 (Δ𝑘 , 𝜕Δ𝑘 )  H̃𝑛 (𝑆 𝑘 ) 
0 :𝑛 ≠𝑘

6 Degrees
The degree of a map 𝑆 𝑘 → 𝑆 𝑘 , despite its simplicity, is a powerful concept. We will illustrate
this in this section, after studying some of its properties.

6.1 Basic properties and examples

Definition 6.1. Let 𝑓 : 𝑆 𝑘 → 𝑆 𝑘 be a (as always, continuous) map. Then the induced homo-
morphism in homology,
𝑓∗ : H̃𝑘 (𝑆 𝑘 ) → H̃𝑘 (𝑆 𝑘 ),
is an endomorphism of an infinite cyclic group, and hence given by multiplication by a unique
integer deg(𝑓 ) ∈ Z. This is called the degree of 𝑓 .

55
 Let us record some basic properties around this construction.

Lemma 6.2. Let 𝑓 , 𝑔 : 𝑆 𝑘 → 𝑆 𝑘 . Then:

1. deg(id) = 1
2. deg(𝑔 ◦ 𝑓 ) = deg(𝑔) · deg(𝑓 )
3. if 𝑓 ' 𝑔 then deg(𝑓 ) = deg(𝑔)
4. if 𝑓 is a homotopy equivalence then deg(𝑓 ) = ±1
5. if 𝑓 is not surjective ( for example, constant) then deg(𝑓 ) = 0

Proof. The first two properties are clear, and the third follows from Homotopy Invariance.
The next follows from the first three, cf. the proof of Corollary 4.3. For the last one, let 𝑥 ∈ 𝑆 𝑘
not in the image of 𝑓 . Then 𝑓 factors as
𝑓
𝑆𝑘 𝑆𝑘

𝑆 𝑘 \𝑥

Upon passing to reduced homology we see that 𝑓∗ factors through H̃𝑘 (𝑆 𝑘 \𝑥) = 0 since 𝑆 𝑘 \𝑥 is
contractible:
𝑓∗
Z Z

0
We deduce that deg(𝑓 ) = 0. 

Commentary 6.3. Note that these basic properties do not produce a map 𝑓 : 𝑆 𝑘 → 𝑆 𝑘 whose
degree is different from 0, 1. However, for 𝑘 = 1 we know how to produce maps of arbitrary
degree (see Example 6.5 below). For 𝑘 > 1 we will deduce the existence of such maps from
the case 𝑘 = 1, see Proposition 6.6.

Exercise 6.4. Let 𝑘 = 0. What are the possible degrees? Give an example of each.

Example 6.5. Let 𝑛 be an integer, and let 𝑓 : 𝑧 ↦→ 𝑧𝑛 be the 𝑛-power map of 𝑆 1 ⊆ C× viewed
as the complex numbers of norm 1. We saw in Example 5.2 that the loop

𝜎 : [0, 1] → 𝑆 1
𝑡 ↦→ 𝑒 2𝜋𝑖𝑡

represents a generator in H1 (𝑆 1 )  H̃1 (𝑆 1 ). Then 𝑓∗ ( [𝜎]) = [𝑓 ◦ 𝜎] is represented by the loop

𝑡 ↦→ 𝑒 2𝜋𝑖𝑛𝑡 which is homologous to 𝑛𝜎. (For example, you can use Theorem 3.28 to see this
most easily.) We deduce that 𝑓 has degree 𝑛.

Proposition 6.6. Let 𝑘 ≥ 1. For every integer 𝑛 ∈ Z there exists a map 𝑓 : 𝑆 𝑘 → 𝑆 𝑘 of degree 𝑛.

56
Proof. Recall (Exercise 2.9) the suspension 𝑆𝑋 of a space 𝑋 , which is obtained from 𝑋 × [−1, 1]
by collapsing each of 𝑋 × {1} and 𝑋 × {−1} to a point. The two opens

𝐶 +𝑋 = 𝑋 × (−𝜖, 1]/(𝑋 × {1}), 𝐶 −𝑋 = 𝑋 × [−1, 𝜖)/(𝑋 × {−1})

are contractible, and Mayer-Vietoris gives an isomorphism

∼
𝜕 : H𝑘+1 (𝑆𝑋 ) −
→ H𝑘 (𝑋 ).

A map 𝑓 : 𝑋 → 𝑋 induces a map 𝑆 𝑓 : 𝑆𝑋 → 𝑆𝑋 in an obvious way, and one checks (using the
definition of the connecting homomorphism) that the square

𝑆 𝑓∗
H𝑘+1 (𝑆𝑋 ) H𝑘+1 (𝑆𝑋 )
𝜕 𝜕
𝑓∗
H𝑘 (𝑋 ) H𝑘 (𝑋 )

commutes.
Applying this with 𝑋 = 𝑆 𝑘−1 and noticing that 𝑆 (𝑆 𝑘−1 ) ≈ 𝑆 𝑘 , we get deg(𝑆 𝑓 ) = deg(𝑓 ) and
thereby reduce, inductively, to 𝑘 = 1. This we saw in Example 6.5. 

Commentary 6.7. Given a continuous map 𝑓 : 𝑆 𝑘 → 𝑆 𝑘 , how would one go about com-
puting its degree? This is certainly not obvious just from the definitions. We have described
in Corollary 5.9 a rather explicit fundamental class for 𝑆 𝑘 , represented by a cycle 𝜎+ − 𝜎− .
Therefore we need to find the integer 𝑛 such that the 𝑘-chains

(6.8) 𝑛𝜎+ − 𝑛𝜎− and 𝑓 ◦ 𝜎+ − 𝑓 ◦ 𝜎−

are homologous. Unfortunately, we don’t seem to have a good way of solving this problem.
Since the isomorphism H𝑘 (𝑆 𝑘 )  Z was obtained by induction on 𝑘, through the connecting
𝜕
homomorphism in the Mayer-Vietoris sequence H𝑘 (𝑆 𝑘 ) −
→ H𝑘−1 (𝑆 𝑘−1 ), the natural thing to
∼
try would be to apply 𝜕 to both 𝑘-chains. Eventually, we would reduce to the case 𝑘 = 1 where
things might be sufficiently explicit. However, you can see how these reductions become
quite rapidly impractical in most cases. (Think about how the connecting homomorphsm
was defined!)
We will actually see examples (okay: one example) where the integer 𝑛 in (6.8) can be de-
termined explicitly. However, besides that we will mostly develop general tools for computing
degrees. And at the same time, and arguably more importantly, we will learn some interesting
things about spheres.

6.2 Antipodes
Lemma 6.9. Let 𝑆 𝑘 ⊂ R𝑘+1 be the unit circle. Let 𝑓 : 𝑆 𝑘 → 𝑆 𝑘 be the reflection in a hyperplane
through 0 in R𝑘+1 . Then deg(𝑓 ) = −1.

57
Proof. Let 𝐻 ⊆ R𝑘+1 be the fixed hyperplane. It splits the sphere 𝑆 𝑘 into two hemispheres 𝑆 +𝑘
≈
and 𝑆 −𝑘 as in Remark 5.8. If we pick some homeomorphism 𝜎+ : Δ𝑘 − → 𝑆 +𝑘 and set 𝜎− = 𝑓 ◦ 𝜎+
then these satisfy the assumptions of Remark 5.8. By Corollary 5.9, 𝑠 = [𝜎+ − 𝜎− ] generates
H̃𝑘 (𝑆 𝑘 ), and
𝑓∗ (𝑠) = [𝑓 ◦ 𝜎+ ] − [𝑓 ◦ 𝜎− ] = [𝜎− ] − [𝜎+ ] = −𝑓∗ (𝑠),
so that deg(𝑓 ) = −1. 

Example 6.10. Let 𝑇 : R𝑘+1 → R𝑘+1 be a linear orthogonal transformation. It restricts to a

homeomorphism
≈
𝑓 : 𝑆𝑘 −
→ 𝑆𝑘 .
We therefore already know deg(𝑓 ) = ±1. Let us show that in fact deg(𝑓 ) = det(𝑇 ).
Choosing a suitable orthonormal basis, T can be represented by a block sum matrix where
each block is either a 2×2-rotation matrix, or a 1×1-matrix with entry
 ±1, with at mostone −1.
cos(𝛼𝑡) − sin(𝛼𝑡)
One can homotope any rotation matrix to the identity, via 𝑡 ↦→ , so we
sin(𝛼𝑡) cos(𝛼𝑡)
further reduce to the case where 𝑇 is the identity, or a reflection at a hyperplane through 0.
In both cases we have computed the degree and it coincides with the determinant of 𝑇 .

Corollary 6.11. Let 𝑓 : 𝑆 𝑘 → 𝑆 𝑘 be the antipodal map 𝑥 ↦→ −𝑥. Then deg(𝑓 ) = (−1)𝑘+1 .

Proof. The map 𝑥 ↦→ −𝑥 : R𝑘+1 → R𝑘+1 is the composition of 𝑘 + 1 reflections through the
coordinate hyperplanes. We conclude with Lemma 6.9. 

Corollary 6.12. Suppose 𝑓 : 𝑆 𝑘 → 𝑆 𝑘 has no fixed points. Then deg(𝑓 ) = (−1)𝑘+1 .

Proof. It is sufficient to prove that 𝑓 is homotopic to the antipodal map, by Corollary 6.11. The
line through 𝑓 (𝑥) and −𝑥 goes through the origin if and only if 𝑓 (𝑥) = 𝑥.
−𝑥
𝑓 (𝑥)

Since we are excluding this possibility, the following expression defines a homotopy from
the antipodal map to 𝑓 :
𝑡 𝑓 (𝑥) + (1 − 𝑡) (−𝑥)
(𝑡, 𝑥) ↦→
k𝑡 𝑓 (𝑥) + (1 − 𝑡) (−𝑥) k


Commentary 6.13. Recall that a vector field on 𝑆 𝑘 is a continuous map v : 𝑆 𝑘 → R𝑘+1 . It is a

tangent vector field if v(𝑥)⊥𝑥 for all 𝑥 ∈ 𝑆 𝑘 .

58
 v(𝑥)
𝑥

Example 6.14. Obviously, the constant map v(𝑥) = 0 is a tangent vector field on every sphere.
We are interested in tangent vector fields that vanish nowhere.

Example 6.15. The map

𝑥 = (𝑥 1, . . . , 𝑥 2𝑚 ) ↦→ v(𝑥) = (−𝑥 2, 𝑥 1, . . . , −𝑥 2𝑚 , 𝑥 2𝑚−1 )
defines a non-vanishing tangent vector field on 𝑆 2𝑚−1 .
Obviously this particular map can only be defined on odd-dimensional spheres. Is there
another, maybe more clever, construction for even-dimensional ones? The answer is: No!

Corollary 6.16 (Hairy Ball Theorem). Every tangent vector field on an even-dimensional sphere
vanishes at some point.

Proof. Assume to the contrary that v : 𝑆 𝑘 → R𝑘+1 is a non-vanishing tangent vector field with 𝑘
even. Consider then the map
𝑆 𝑘 × [0, 1] 3 (𝑥, 𝑡) ↦→ cos(𝜋𝑡)𝑥 + sin(𝜋𝑡)v(𝑥) ∈ R𝑘+1 .
Since 𝑥⊥v(𝑥) and v(𝑥) ≠ 0 the two vectors 𝑥, v(𝑥) are linearly independent. It follows that the
map lands in R𝑘+1 \{0} and we can divide by the norm to get a homotopy 𝑆 𝑘 × [0, 1] → 𝑆 𝑘 .
At time 𝑡 = 0 (resp. 𝑡 = 1) it is the identity (resp. antipodal) map. But this is impossible since
deg(id) = 1 while deg(− id) = (−1)𝑘+1 = −1, by Corollary 6.11. 

Commentary 6.17. Take an (ordinary, 3-dimensional) ball and imagine it comes with hairs
on its surface, for some weird reason. (For example, like a coconut comes with hairs. Or you
may think of a hedgehog that is curling up into a ball instead.) The theorem says that you
cannot comb all these hairs flat at the same time. Here is a picture (from wikipedia) illustrating
a failed attempt:

You can see how there are tufts at each of the two poles.15
15 It is not a coincidence that there are two tufts here either (at least if the vector field is also supposed smooth). If
counted with the correct ‘multiplicities’ (=indices) the number is always equal to 2 which is the Euler characteristic
of the 2-sphere, as we will discuss later. This statement is known as the Poincaré-Hopf Theorem.

59
 The theorem is due to Brouwer (1912). For a (maybe slightly) more practical implication
see Exercise 5.8.

6.3 Local degrees

Commentary 6.18. We will now redo the definition of the degree, but this time as a local
invariant around a point. The main result is Proposition 6.21 that expresses the ‘global’ degree
in terms of local ones. This is a very useful tool for computing the degree. See Example 6.22
for a nice application.

Remark 6.19. Let 𝑘 ≥ 1, let 𝑓 : 𝑆 𝑘 → 𝑆 𝑘 be a continuous map and assume 𝑓 −1 (𝑦) = {𝑥 1, . . . , 𝑥𝑛 }

is finite. We may choose disjoint open balls 𝑈𝑖 ⊆ 𝑆 𝑘 around 𝑥𝑖 .
𝑥1 𝑦
𝑓
𝑥2

𝑥3

So 𝑓 induces a map of pairs (𝑈𝑖 , 𝑈𝑖 − 𝑥𝑖 ) → (𝑆 𝑘 , 𝑆 𝑘 − 𝑦) and hence by excision, a morphism in

homology,
𝑓∗
(6.20) H𝑘 (𝑆 𝑘 , 𝑆 𝑘 − 𝑥𝑖 )  H𝑘 (𝑈𝑖 , 𝑈𝑖 − 𝑥𝑖 ) −
→ H𝑘 (𝑆 𝑘 , 𝑆 𝑘 − 𝑦).

We saw in Corollary 5.25 that both of these groups are infinite cyclic. In fact, the map in the
∼
long exact sequence for the pair H𝑘 (𝑆 𝑘 ) − → H𝑘 (𝑆 𝑘 , 𝑆 𝑘 − 𝑥𝑖 ) is an isomorphism, and similarly
∼
H𝑘 (𝑆 𝑘 ) −
→ H𝑘 (𝑆 𝑘 , 𝑆 𝑘 − 𝑦). Therefore we may view (6.20) as an endomorphism

𝑓 |𝑥𝑖 : H𝑘 (𝑆 𝑘 ) → H𝑘 (𝑆 𝑘 )

of an infinite cyclic group, which is multiplication by some integer, called the local degree of 𝑓
at 𝑥. We denote it by deg(𝑓 |𝑥𝑖 ).

Proposition 6.21. In the situation above we have

𝑛
∑︁
deg(𝑓 ) = deg(𝑓 |𝑥𝑖 )
𝑖=1

Example 6.22. Let 𝑝 (𝑧) = 𝑝𝑛 𝑧𝑛 + · · · + 𝑝 0 ∈ C[𝑧] be a complex polynomial viewed as a map

C → C. It extends to a continuous map on the one-point compactification

𝑓 : 𝑆 2 ≈ C ∪ {∞} → C ∪ {∞} ≈ 𝑆 2 .

Let 𝑤 ∈ C be such that 𝑝 0 (𝑧) ≠ 0 for all 𝑧 ∈ 𝑓 −1 (𝑤). This means that 𝑓 is invertible around 𝑧
and therefore deg(𝑓 |𝑧 ) = ±1. In fact, since polynomials are orientation preserving, we must
have deg(𝑓 |𝑧 ) = 1. It follows from Proposition 6.21 that

deg(𝑓 ) = #𝑓 −1 (𝑤) = 𝑛 = deg(𝑝).

60
Proof of Proposition 6.21. Consider the relative homology group H𝑘 (𝑆 𝑘 , 𝑆 𝑘 − 𝑓 −1 (𝑦)) which by
excision may be identified with
H𝑘 (q𝑈𝑖 , q(𝑈𝑖 − 𝑥𝑖 ))  ⊕ H𝑘 (𝑈𝑖 , 𝑈𝑖 − 𝑥𝑖 )  H𝑘 (𝑆 𝑘 , 𝑆 𝑘 − 𝑥𝑖 ).
The composite with the obvious maps in the long exact sequences for the various pairs
Í
𝑓∗ ∼
H𝑘 (𝑆 𝑘 ) −
→ ⊕ H𝑘 (𝑆 𝑘 , 𝑆 𝑘 − 𝑥𝑖 )  H𝑘 (𝑆 𝑘 , 𝑆 𝑘 − 𝑓 −1 (𝑦)) −
→ H𝑘 (𝑆 𝑘 , 𝑆 𝑘 − 𝑦) ←
− H𝑘 (𝑆 𝑘 )
is, by definition, multiplication by deg(𝑓 |𝑥𝑖 ). But the composition of the first two arrows
Í
H𝑘 (𝑆 𝑘 ) → H𝑘 (𝑆 𝑘 , 𝑆 𝑘 − 𝑓 −1 (𝑦)) is simply the canonical map in the long exact sequence of the
pair, and hence fits into the following commutative square

H𝑘 (𝑆 𝑘 ) H𝑘 (𝑆 𝑘 , 𝑆 𝑘 − 𝑓 −1 (𝑦))
𝑓∗ 𝑓∗

∼
H𝑘 (𝑆 𝑘 ) H𝑘 (𝑆 𝑘 , 𝑆 𝑘 − 𝑦)

Unwinding the definitions, this amounts to precisely the claim. 

7 Manifolds
Many of the topological spaces one typically encounters “in nature” are manifolds. In this
section we collect some topics which are specific to these nice objects. This includes foremost
the notion of orientation. We will also see how homology is a fine enough invariant to classify
low-dimensional manifolds.

7.1 Examples
We already recalled the notion of a (topological) manifold in Commentary 5.24: A Hausdorff
space so that every point has an open neighborhood homeomorphic to R𝑘 , for some fixed 𝑘,
which is called the dimension of the manifold.

Example 7.1. 1. Of course, R𝑘 itself is a manifold of dimension 𝑘.

2. Let 𝑃, 𝑄 be the north and south poles of 𝑆 𝑘 . Stereographic projection gives homeomor-
phisms:

𝜋𝑃 : 𝑆 𝑘 \ {𝑃 } −→ R𝑘

𝜋𝑄 : 𝑆 𝑘 \ {𝑄 } −→ R𝑘
Which shows that 𝑆 𝑘 is a 𝑘-manifold.

61
 3. An open subspace of a 𝑘-manifold is itself a 𝑘-manifold.

Example 7.2. Besides R2 and 𝑆 2 , all of T , K and RP2 are 2-manifolds:

We will see many more examples of 2-manifolds in Section 7.3.

Remark 7.3. The interval [0, 1] is not a manifold. Indeed, no open neighborhood of 0 (or 1)
is homeomorphic to R𝑘 for any 𝑘. (Instead, it is what is called a manifold with boundary. We
will not discuss these in this course.)

Example 7.4. A 0-dimensional manifold is simply a discrete set.

Commentary 7.5. How does a 1-dimensional manifold 𝑀 look like? Start at a point 𝑥 ∈ 𝑀.
Since locally around 𝑥, 𝑀 looks like the real line, we may pick one of the two directions and
follow it. If we ever come back to 𝑥 we have found 𝑀 ≈ 𝑆 1 . Otherwise we could follow the
other direction and clearly could not come back to 𝑥 either. For example, this could occur if
𝑀 ≈ R.16 The following exercise makes this argument more precise.

Exercise 7.6. Show that the only compact, connected 1-manifold is (homeomorphic to) 𝑆 1 .

Sketch. Let 𝑀 be a compact, connected 1-manifold. By assumption there are open subsets
𝑈 1, . . . , 𝑈𝑛 ⊆ 𝑀 such that 𝑈𝑖 ≈ R1 . We may choose 𝑛 to be minimal. Note that 𝑛 > 1. (Why?)
Since 𝑀 is connected we may assume, after relabeling, that 𝑈 1 ∩ 𝑈 2 ≠ ∅. If 𝜋0 (𝑈 1 ∩ 𝑈 2 ) = ∗
then 𝑈 1 ∪ 𝑈 2 ≈ R1 , contradicting minimality. The hardest bit is to show that the only other
possibility is 𝜋0 (𝑈 1 ∩ 𝑈 2 ) = ∗ ∗. In that case 𝑀 = 𝑈 1 ∪ 𝑈 2 ≈ 𝑆 1 . (In particular, 𝑛 = 2.)
Ý


7.2 Orientations
Let 𝑉 be a non-zero finite-dimensional real vector space. Recall that two bases define the same
orientation if the change of basis transformation from one to the other has positive determi-
nant. This defines an equivalence relation whose two equivalence classes are the two possible
orientations of 𝑉 . Reversing the logic, we can think of invertible transformations with posi-
tive (resp. negative) determinant as orientation-preserving (resp. orientation-reversing). For
example, one of the basic observations is that a reflection is orientation-reversing.
Since manifolds locally look like finite-dimensional vector spaces we expect that orien-
tations can be generalized to manifolds. What could play the role of bases and change of
16 Theseare in fact the only two possibilities if one assumes 𝑀 to be connected and not too wild, for example
second-countable. Look up the long line for a wild 1-manifold.

62
basis transformations in this context? The result (Lemma 6.9) that reflections on spheres have
degree −1 suggests the following dictionary:
• basis ! generator of the infinite-cyclic group H𝑘 (𝑆 𝑘 )
• linear transformation ! endomorphism of H𝑘 (𝑆 𝑘 )
• determinant ! degree
The connection with manifolds is given by Corollary 5.25: Namely, the 𝑘th local homology
group of a 𝑘-dimensional manifold 𝑀 is always infinite-cyclic and can be identified with
H𝑘 (𝑀, 𝑀\𝑥)  H𝑘−1 (𝑆 𝑘−1 ).

Definition 7.7. A local orientation of 𝑀 at 𝑥 is a choice of generator of H𝑘 (𝑀, 𝑀\𝑥).

Commentary 7.8. Here is one way to think about this. To fix our ideas let 𝑀 be a 2-manifold
and choose an open neighborhood 𝑈 ≈ R2 around 𝑥. The long exact sequence for the pair
(𝑈 , 𝑈 \𝑥) induces an isomorphism
∼
→ H1 (𝑈 \𝑥)  H1 (𝑆 1 )
H2 (𝑈 , 𝑈 \𝑥) −
since 𝑈 \𝑥 deformation retracts onto a small circle around 𝑥. Choosing a local orientation 𝜔𝑥
at 𝑥 therefore amounts to choosing in which direction to loop around this circle.

𝜔𝑥 𝜔𝑥

Commentary 7.9. It is clear how choosing a local orientation at 0 ∈ R2 determines a local

orientation at every other point 𝑥 ∈ R2 . We can globally choose the clockwise or counter-
clockwise orientation. (Similarly, we can globally choose the right-handed or left-handed
orientation of R3 , etc.) The same is not true on the (open) Möbius strip, however. If we
choose a local orientation at 𝑥 and try to ‘transport’ it along a loop around the strip we end up
in 𝑥 with the opposite orientation.
We will now make a series of definitions which precisely distinguishes between these
two behaviours. We will then express this by saying that the plane is orientable, while the
open Möbius strip is not. Compared to the theory for vector spaces discussed above, non-
orientability is a genuinely new concept.

Convention 7.10. Let 𝐵 ⊆ 𝑀 be a subset of a 𝑘-manifold. We say that 𝐵 is a small open (resp.
closed) ball if it has an open neighborhood 𝐵 ⊂ 𝑈 ≈ R𝑘 in which it identifies with an open
(resp. closed) ball of finite radius.
The point of this convention is that by excision (and where 𝐵 identifies with the open ball
˚
𝐵(𝑥, 𝑟 ) of radius 𝑟 around 𝑥 ∈ R𝑘 )
˚ 𝑟 ))  H𝑘−1 (𝜕𝐵(𝑥, 𝑟 )),
H𝑘 (𝑀, 𝑀\𝐵)  H𝑘 (𝑈 , 𝑈 \𝐵)  H𝑘 (R𝑘 , R𝑘 \𝐵(𝑥,

63
which is infinite-cyclic. So we may think of a generator here as an orientation of the boundary
sphere of 𝐵.
Note that for every 𝑦 ∈ 𝐵 we thereby get an induced local orientation through the canon-
ical map
∼
H𝑘 (𝑀, 𝑀\𝐵) − → H𝑘 (𝑀, 𝑀\𝑦).
We say that a family of local orientations (𝜔 𝑦 , 𝑦 ∈ 𝐵) is consistent if there is a generator 𝜔 𝐵 ∈
H𝑘 (𝑀, 𝑀\𝐵) mapping to each 𝜔 𝑦 .

𝜔𝑦

𝜔𝐵
𝐵
𝑈

Definition 7.11. • An orientation of a 𝑘-manifold 𝑀 is a function 𝑥 ↦→ 𝜔𝑥 ∈ H𝑘 (𝑀, 𝑀\𝑥)

assigning to 𝑥 ∈ 𝑀 a local orientation, which is locally consistent in the sense above. That
is, every 𝑥 ∈ 𝑀 sits inside a small open ball 𝐵 such that the local orientations (𝜔 𝑦 , 𝑦 ∈ 𝐵)
are consistent.
• 𝑀 is called orientable if it admits an orientation. Otherwise it is called non-orientable.

Example 7.12. The 𝑘-sphere is orientable. Indeed, after choosing a fundamental class in H𝑘 (𝑆 𝑘 ),
the canonical map

(7.13) H𝑘 (𝑆 𝑘 ) → H𝑘 (𝑆 𝑘 , 𝑆 𝑘 \𝑥)

induces local orientations at each point 𝑥 ∈ 𝑆 𝑘 . And these are locally consistent since (7.13)
factors through H𝑘 (𝑆 𝑘 , 𝑆 𝑘 \𝐵) for any small open ball 𝐵 around 𝑥.

Construction 7.14. We now construct the orientation bundle 𝑀 e associated with 𝑀. Define 𝑀 e
as a set to be the pairs (𝑥, 𝜔𝑥 ) where 𝑥 ∈ 𝑀 and 𝜔𝑥 is a local orientation at 𝑥. It comes with an
obvious map 𝜋 : 𝑀 e → 𝑀 sending such a pair to 𝑥.
For the topology, if 𝐵 ⊆ 𝑀 is a small open ball then we saw that there are exactly two
collections of consistent local orientations (𝑦, 𝜔 𝑦 ) where 𝑦 ranges over the points in 𝐵. In other
words, 𝜋 −1 (𝐵) = 𝐵 q 𝐵, and this gives 𝑀 e the structure of a manifold itself so that 𝜋 : 𝑀
e → 𝑀 is
a 2-sheeted cover.

Exercise 7.15. Show that for any manifold 𝑀, its orientation bundle 𝑀 e is orientable. In fact,
it has a canonical orientation. Moreover, the deck transformation (𝑥, 𝜔𝑥 ) ↦→ (𝑥, −𝜔𝑥 ) reverses
this orientation.

64
Example 7.16. Let 𝑀 be the open Möbius strip

𝑀 := [0, 1] × (0, 1)/∼

where
(0, 𝑦) ∼ (1, 1 − 𝑦)
for all 𝑦 ∈ (0, 1).

e  𝑆 1 × (0, 1).
So 𝑀

Lemma 7.17. Giving an orientation of 𝑀 is equivalent to giving a continuous section to 𝜋.

Proof. Giving a section 𝜔 : 𝑀 → 𝑀 e (not necessarily continuous) amounts to choosing, for each
𝑥 ∈ 𝑀, a local orientation 𝜔𝑥 at 𝑥. The map 𝜔 is continuous if and only if for each small open
ball 𝐵 ⊆ 𝑀, and 𝜋 −1 (𝐵) = 𝐵 q 𝐵 =: 𝐵 + q 𝐵 − , the preimages 𝜔 −1 (𝐵 + ) and 𝜔 −1 (𝐵 − ) are both open
in 𝐵. Since these two preimages are disjoint and jointly cover 𝐵 this condition is equivalent to
𝜔 (𝐵) = 𝐵 + or 𝜔 (𝐵) = 𝐵 − . Which means precisely that the local orientations (𝜔 𝑦 , 𝑦 ∈ 𝐵) are
consistent. 

Remark 7.18. Let 𝑥 ∈ 𝑀 and choose a local orientation 𝜔𝑥 . A path in 𝑀 from 𝑥 to 𝑦 has a
unique lift to 𝑀
e starting at (𝑥, 𝜔𝑥 ) and ending at (𝑦, 𝜔 𝑦 ) for some 𝜔 𝑦 . In other words, the path
determines a unique local orientation at 𝑦.

Corollary 7.19. Assume 𝑀 is a connected manifold. Then:

• either, 𝑀
e → 𝑀 is a non-trivial 2-sheeted cover and 𝑀 is non-orientable;
• or, 𝑀
e = 𝑀 q 𝑀 and 𝑀 admits precisely two orientations.

Proof. Assume 𝜋 : 𝑀 e → 𝑀 is a non-trivial 2-sheeted cover, and 𝜔 : 𝑀 → 𝑀 e a continuous

section to 𝜋. Let 𝑥 ∈ 𝑀 and 𝜋 −1 (𝑥) = {(𝑥, 𝜔𝑥 ), (𝑥, −𝜔𝑥 )}. By assumption there is a path 𝛾 in 𝑀
e
from (𝑥, 𝜔𝑥 ) to (𝑥, −𝜔𝑥 ). Then
𝛾, 𝜔𝜋𝛾
are two paths in 𝑀
e lifting 𝜋𝛾 = 𝜋𝜔𝜋𝛾 and starting at (𝑥, 𝜔𝑥 ). But the first path ends at (𝑥, −𝜔𝑥 )
while the second one ends at (𝑥, 𝜔𝑥 ). This is a contradiction to uniqueness of lifting paths, see
Remark 7.18. We deduce that 𝑀 is non-orientable.
The second case is clear. 

65
Example 7.20. We deduce from Example 7.16 that the open Möbius strip is not orientable.

Corollary 7.21. Any simply-connected manifold is orientable.

Example 7.22. In particular, euclidean space R𝑘 is orientable.

Example 7.23. Let 𝑘 ≥ 1. The quotient map 𝑆 𝑘 → RP𝑘 is the unique non-trivial two-sheeted
cover of real projective space. Moreover, the non-trivial desk transformation is given by the
antipodal map 𝑥 ↦→ −𝑥 which has degree (−1)𝑘+1 . Hence:

• If 𝑘 is even so that this degree is −1 we deduce that the desk transformation is orientation-
reversing and 𝑆 𝑘 = RP g𝑘 . As 𝑆 𝑘 is connected, RP𝑘 must be non-orientable in this case.

• If 𝑘 is odd so that this degree is 1 we deduce that the desk transformation is orientation-
preserving and 𝑆 𝑘 ≠ RP g𝑘 . The latter must then be the trivial 2-sheeted cover and RP𝑘
is orientable.

Remark 7.24. Suppose 𝑀 is a manifold and 𝑀 ≈ 𝑁 a homeomorphism with another topo-

logical space. Then 𝑁 is also a manifold (of the same dimension). Moreover, 𝑀 is orientable if
and only if 𝑁 is orientable.
On the other hand, if 𝑀 is a manifold, with 𝑀 ' 𝑁 for some topological space 𝑁 then 𝑁 is
not necessarily a manifold. And even if it is, (non-)orientability is not preserved. For example,
the open Möbius strip is homotopy equivalent to 𝑆 1 but only the latter is orientable.

7.3 Surfaces: topology

In the remainder of this section we discuss a particular class of manifolds, namely surfaces. A
surface (in this course) is a compact, connected 2-manifold. (In particular, a surface is non-
empty.) After recalling the classification of surfaces17 we will (in section 7.4) apply the ‘alge-
braic topology’ we have learnt so far to these objects. This is a particularly instructive illustra-
tion of some of the theory we’ve developed. Of course, the mathematics here is very old even
if in the, say, 19th century it wasn’t expressed in this exact language.

Example 7.25. We have seen several examples of surfaces already: 𝑆 2 , T , K, RP2 . We will
now describe a general procedure for constructing new examples from old ones.

Construction 7.26. Let 𝑆 1 and 𝑆 2 be two surfaces and let 𝐷𝑖 ⊆ 𝑆𝑖 be two small closed disks.
We can then glue 𝑆 1 \𝐷˚ 1 and 𝑆 2 \𝐷˚ 2 along 𝜕𝐷 1 ≈ 𝜕𝐷 2 . We call the resulting space the connected
sum 𝑆 1 #𝑆 2 .
17 You may have seen this classification in other courses before. In any case, this isn’t a central part of this course
which is why we will leave some results without proof.

66
 Connected sum of 𝑆 1 = T and 𝑆 2 = 𝑆 2 with 𝑈𝑖 := 𝐷˚ 𝑖

Remark 7.27. It is not too hard to show that the connected sum of two surfaces is a surface. It
is non-trivial but also true that the homeomorphism type of 𝑆 1 #𝑆 2 is independent of the choice
of the disks 𝐷𝑖 and the homeomorphism between their boundaries. In fact, the #-operation
becomes associative, commutative and unital on the set of homeomorphism types of surfaces.
As the picture in the construction above suggest, the unit for this operation is 𝑆 2 .

Example 7.28. One can obtain the 𝑔-holed torus as the connected sum of 𝑔 tori, Σ𝑔 = T # · · · #T :

Alternatively, Σ𝑔 is what one gets by identifying the boundary edges in a 4𝑔-gon according
to the word
𝑊𝑔 = 𝑎 1𝑏 1𝑎 −1 −1 −1 −1
1 𝑏 1 · · · 𝑎𝑔𝑏𝑔 𝑎𝑔 𝑏𝑔 = [𝑎 1, 𝑏 1 ] · · · [𝑎𝑔 , 𝑏𝑔 ].

This follows from the next lemma.

Remark 7.29. More generally, whenever 𝑊 is a word in 𝑛 letters, with each letter occurring
twice, one can consider the regular 2𝑛-gon 𝑃 (𝑊 ) with edges labeled according to the word 𝑊 .
And then 𝑀 (𝑊 ) = 𝑃 (𝑊 )/∼ is the quotient obtained by identifying the corresponding oriented
edges. For example, here is the 4𝑔-gon giving rise to Σ𝑔 :

67
Lemma 7.30. For words 𝑊 and 𝑊 0 as above (with disjoint alphabets) we have 𝑀 (𝑊 )#𝑀 (𝑊 0) ≈
𝑀 (𝑊𝑊 0) where 𝑊𝑊 0 is the concatenation of the two words.

Proof. We will not prove this but the idea is to ‘lift’ the operation of connected sum to the level
of 2𝑛-gons. 

Example 7.31. The surface 𝑁 1 := 𝑀 (𝑧 2 ) is homeomorphic to RP2 . By the lemma, 𝑁ℎ :=

𝑀 (𝑧 12 · · · 𝑧ℎ2 ) ≈ RP2 # · · · #RP2 . In Exercise 5.7 you show that 𝑁 2 ≈ RP2 #RP2 ≈ K is a familiar
surface, namely the Klein bottle.

Example 7.32. The surface associated with the empty word is the 2-sphere. We can think of
it as Σ0 = 𝑀 (∅) = 𝑀 (𝑥𝑥 −1 ).

Remark 7.33 (Classification of surfaces). We now recall the result that each surface is home-
omorphic to exactly one surface of the following list:

1. Σ𝑔 , 𝑔 ≥ 0;18
2. 𝑁ℎ , ℎ > 0.19

Remark 7.34. It follows that every surface is homeomorphic to a quotient of a regular poly-
gon.

Remark 7.35. Given all we said so far, you should be wondering: what happens if I take the
connected sum of an 𝑁ℎ and a Σ𝑔 ? After all I should end up with another surface of this list.
Which one is it? It turns out that
T #RP2 ≈ 𝑁 3,
which is due to Walther von Dyck (1888). So the commutative monoid of homeomorphism
types of surfaces is isomorphic to

h𝑡, 𝑟 | 𝑡 + 𝑟 = 3𝑟 i.20
18 The integer 𝑔 is of course the genus of the surface.
19 The integer ℎ is also called the non-orientable genus of the surface. We will see below that the 𝑁ℎ are indeed
precisely the non-orientable surfaces in this list.
20 Beware that the monoid is not cancellative. That is, from 𝑡 + 𝑟 = 3𝑟 we cannot deduce 𝑡 = 2𝑟 (which is false).

68
7.4 Surfaces: Homology and orientation
Commentary 7.36. Recall that a compact 0-manifold is just a finite discrete set. Hence the
0th homology group classifies them (is a ‘complete invariant’). Recall also that a compact 1-
manifold is just a finite disjoint union of circles.21 Hence, again, H0 classifies them. Moreover,
taking into account H1 as well we can distinguish 1-manifolds from 0-manifolds.
In the remainder of this section, we will prove that the story continues in dimension 2.
Namely, the classification of Remark 7.33 is reflected in the homology groups H0 , H1 , H2 of
the surfaces. Moreover, they are also distinguished from 1- and 0-manifolds. In summary:
H0 ⊕ H1 ⊕ H2 is a complete invariant for compact manifolds (up to homeomorphism) of di-
mension at most 2.

Commentary 7.37. One thing that is new in the 2-dimensional story is the occurrence of
non-orientable manifolds. This doesn’t exist in lower dimensions. On the other hand, in
dimension 3, classification questions like this become much harder. And homology is not
a complete invariant anymore. For example, there are many non-homeomorphic compact
3-manifolds with the homology of 𝑆 3 . (As an aside, if you also assume that the manifold is
simply-connected then 𝑆 3 is the only one. This is the celebrated Poincaré Conjecture (cf.
Commentary 1.34) proved by Grigori Perelman in the early 2000’s.)

In preparation of the homology computations below we recall from Exercise 5.10 the fol-
lowing notion.

Definition 7.38. Let (𝑋𝛼 , 𝑥𝛼 ) be a family of pointed spaces. Their wedge sum is the quotient
space Ý
Ü 𝑋𝛼
(𝑋𝛼 , 𝑥𝛼 ) := Ý 𝛼 .
𝛼 𝛼 {𝑥 𝛼 }

It is naturally a pointed space.

Example 7.39. (𝑆 1, ∗) ∨ (𝑆 1, ∗) is the figure eight: two circles touching at a point.

Lemma 7.40. Assume that each (𝑋𝛼 , 𝑥𝛼 ) is a good pair. Then H̃𝑛 ( (𝑋𝛼 , 𝑥𝛼 ))  ⊕H̃𝑛 (𝑋𝛼 ).
Ô

Proof. As in Exercise 5.10, the statement follows from Proposition 5.28 together with Proposi-
tion 3.14. 

Proposition 7.41. The homology of the 𝑔-holed torus is



 Z : 𝑛 = 0, 2
H𝑛 (Σ𝑔 ) = Z2𝑔


:𝑛 =1

0

:𝑛 >2


Proof. We compute its homology using the following open cover:

21 In this discussion, we always mean: up to homeomorphism.

69
 The subspace 𝑉 is contractible, 𝑈 ∩ 𝑉 is homotopy equivalent to a loop 𝛾,22 and 𝑈 de-
formation retracts onto the boundary, which is a wedge sum of 2𝑔 loops 𝑎 1, 𝑏 1, . . . , 𝑎𝑔 , 𝑏𝑔 . By
Lemma 7.40, H̃𝑛 (𝑈 ) is then the direct sum of the reduced homologies of the loops (which is
concentrated in degree 1). We deduce that the only non-vanishing bit in the reduced MV
exact sequence is

0 → H̃2 (Σ𝑔 ) → H̃1 (𝑈 ∩ 𝑉 ) → H̃1 (𝑈 ) → H̃1 (Σ𝑔 ) → 0

where the middle map Z𝛾 → Z2𝑔 = h𝑎 1, 𝑏 1, . . . , 𝑎𝑔 , 𝑏𝑔 i sends the generator 𝑔 to

𝑎 1 + 𝑏 1 − 𝑎 1 − 𝑏 1 + · · · + 𝑎𝑔 + 𝑏𝑔 − 𝑎𝑔 − 𝑏𝑔 = 0.

The claim follows. 

Corollary 7.42. The surfaces Σ𝑔 , 𝑔 ≥ 0, are orientable.

Proof. We use the computation of the homology in Proposition 7.41. According to that, we
have H2 (Σ𝑔 ) = Z. We now claim that, as in Example 7.12 (for 𝑔 = 0), the homomorphism
H2 (Σ𝑔 ) → H2 (Σ𝑔 , Σ𝑔 − 𝑥) is an isomorphism for every point 𝑥 ∈ Σ𝑔 . As there, this would
show that the surface Σ𝑔 is orientable. To establish the isomorphism we look at the long exact
sequence for the pair:23

(7.43) H2 (Σ𝑔 − 𝑥) → H2 (Σ𝑔 ) → H2 (Σ𝑔 , Σ𝑔 − 𝑥) → H1 (Σ𝑔 − 𝑥) → H1 (Σ𝑔 )

Letting 𝑈 be as in the proof of Proposition 7.41 the inclusion 𝑈 ↩→ Σ𝑔 − 𝑥 is a homotopy

equivalence. In particular, Σ𝑔 − 𝑥 is a 2𝑔-fold wedge of circles and H2 (Σ𝑔 − 𝑥) = 0. Moreover,
∼ ∼
we saw in that proof that H1 (𝑈 ) − → H1 (Σ𝑔 ) hence also H1 (Σ𝑔 − 𝑥) − → H1 (Σ𝑔 ), that is, the
last map in (7.43) is invertible. This shows that the third map in (7.43) is zero. Altogether the
second map in (7.43) is an isomorphism as required. 

Remark 7.44. Using this proposition one can describe quite explicitly a fundamental class
for Σ𝑔 ,24 and observe the close connection between fundamental classes and orientations. (See
22 What we mean by this is that 𝑈 ∩ 𝑉 ' 𝑆 1 and we choose a generator 𝛾 ∈ H1 (𝑆 1 ). Say in clockwise direction.
23 In Example 7.12, the terms just before and after the map in question vanished since 𝑆 𝑘 − 𝑥 was contractible.
This is not true if 𝑔 > 0 so we need to be more careful.
24 An alternative is to wait for Theorem 8.1 below that shows H (Σ ) = HΔ (Σ ) and describe a fundamental class
2 𝑔 2 𝑔
via a Δ-structure. We did this for the torus in Example 2.31.

70
 the lectures.) While the argument in the Corollary (and also in Example 7.12) does not apply
to other manifolds, it is still an indication of a general phenomenon. Namely, a compact,
connected 𝑘-manifold 𝑀 is orientable if and only if H𝑘 (𝑀)  Z. In that case, the canonical
map H𝑘 (𝑀) → H𝑘 (𝑀, 𝑀 − 𝑥) is an isomorphism for every 𝑥 ∈ 𝑀, and a fundamental class
for 𝑀 induces an orientation. We will not prove this here but see Theorem 3.26 of Hatcher’s
book.

Next we turn to the other class of surfaces 𝑁ℎ , ℎ > 0.

Proposition 7.45. The homology of 𝑁ℎ is



 Z :𝑛 =0
H𝑛 (𝑁ℎ ) = Zℎ−1 ⊕ Z/2Z


:𝑛 =1

0

:𝑛 ≥2

Proof. We take the same open cover as in the previous proof. Now, the boundary is a wedge
of ℎ circles and the interesting ‘middle’ map Z𝛾 → Zℎ = h𝑥 1, . . . , 𝑥ℎ i in the MV exact sequence
sends the generator 𝛾 to
2(𝑥 1 + · · · 𝑥ℎ ),
thus the claim. 

Proposition 7.46. The surfaces 𝑁ℎ , ℎ > 0, are non-orientable.

This follows from the computation in Proposition 7.45, in particular the fact that H2 (𝑁ℎ ) =
0, together with the characterization of orientable compact connected manifolds mentioned
in Remark 7.44. However, we can also give a direct proof as follows.
Proof. We use the observation that removing a small closed disk from RP2 yields a space home-
omorphic to the open Möbius strip. (Presumably you show this in the course of doing Exer-
cise 5.7. If you haven’t done this yet: Think of the projective plane as obtained from the square
in the usual way. Remove a disk which is cut in half by the top and bottom edges. Identify
the corresponding quotient with the open Möbius strip.) It follows that for ℎ > 0, the space
𝑁ℎ = RP2 # · · · #RP2 contains the open Möbius strip as an open subspace. Since the latter is
non-orientable (Example 7.20), we deduce that the former isn’t either. 

Remark 7.47. This also gives an arguably more geometric criterion for the orientability of
surfaces. Namely, it shows that a surface is non-orientable if and only if it contains an open
Möbius strip.

The rest of We now sketch an alternative proof orientability of surfaces, based on the classification of
this section surfaces and the following intuitive fact.
is non-
examinable Lemma 7.48. Let 𝑆 1, 𝑆 2 be two surfaces. Then 𝑆 1 #𝑆 2 is orientable if and only if both 𝑆 1 and 𝑆 2 are.
but you
might find it Commentary 7.49. We won’t give the proof of this statement here but if you would like to
instructive. familiarize yourself better with both, connected sums and orientations, this would provide a
good opportunity. See Exercise 6.1.

71
Alternative proof of Corollary 7.42 and Proposition 7.46. We know that RP2 is non-orientable, see
Example 7.23. It follows from the lemma that every 𝑁ℎ is non-orientable.
Assume T was non-orientable. Then, by the lemma, all Σ𝑔 would be non-orientable. And
with the observation in the first paragraph we would conclude that all surfaces apart from
the sphere are non-orientable. But consider the orientation bundle of the torus, e T . As a
two-sheeted cover of T it is itself a surface and orientable by Exercise 7.15. It follows that
T = 𝑆 2 . Since the latter is simply-connected it must be the universal cover, and we deduce that
e
𝜋 1 (T ) = Z/2. This is absurd since we know that 𝜋 1 (T ) = Z2 . 

Remark 7.50. How does one figure out the orientation bundle of the non-orientable surfaces,
fℎ ? You are asked to do this for the Klein bottle K ≈ 𝑁 2 in Exercise 5.7 and we will develop a
𝑁
very efficient tool, the Euler characteristic, to answer this question later in this course. But here
is a picture to guide you.
Place the surface Σ𝑔 inside R3 in such a way that every reflection at a coordinate hyperplane
sends the surface to itself. You’ll easily convince yoursef that this is indeed possible. It follows
that the antipodal map − id : R3 → R3 also takes Σ𝑔 to itself. The induced map in homology
− id : H2 (Σ𝑔 ) → H2 (Σ𝑔 ) is multiplication by −1. (You can see this from the identification of
H2 (Σ𝑔 ) with H1 (𝑈 ∩ 𝑉 ) in the proof of Proposition 7.41 and the fact that reflection of a circle
(' 𝑈 ∩ 𝑉 ) has degree −1.) If you chose the embedding Σ𝑔 ↩→ R3 in the way I’m thinking of,
the antipodal map also has no fixed-points so that the quotient Σ𝑔 /(𝑥 ∼ −𝑥) is again a surface 𝑆.
It follows that 𝑆˜ = Σ𝑔 and hence that 𝑆 is non-orientable. In fact, we will see later that 𝑆 = 𝑁𝑔+1 .
Of course, this recovers the known fact that 𝑆 2 /(𝑥 ∼ −𝑥) = RP2 .

8 Comparison
Recall that our definition of simplicial homology for Δ-complexes is still lacking justification:
We haven’t proven at this point that H𝑛Δ (𝑋 ) is independent of the Δ-complex structure on 𝑋 .
Our first goal in this section is to show that, in fact, whatever Δ-complex structure one puts on
the space 𝑋 , the resulting simplicial homology is isomorphic to the singular homology of 𝑋 .
Some of the ideas in the proof will recur when we introduce yet another homology theory:
cellular homology for CW complexes. And again, we will show that cellular homology and
singular homology agree. So, this section is all about comparing different homology theories.

8.1 Simplicial = singular

Let 𝑋 be a topological space endowed with a Δ-complex structure (𝑇 , |𝑇 | ≈ 𝑋 ). Every 𝑛-
simplex 𝑠 ∈ 𝑇 gives rise to a canonical continuous map (abusively still denoted) 𝑠 : Δ𝑛 → 𝑋 .
This extends to a homomorphism Δ𝑛 (𝑇 ) → 𝐶𝑛 (𝑋 ) and, in fact, to a chain map Δ• (𝑇 ) → 𝐶 • (𝑋 )
since the boundary operator is defined in the same way on both sides. Recall that we ‘defined’
H𝑛Δ (𝑋 ) = H𝑛 (𝑇 ) (= H𝑛 (Δ• (𝑇 ))).
∼
Theorem 8.1. The induced map H𝑛Δ (𝑋 ) −
→ H𝑛 (𝑋 ) is an isomorphism.
∼
Proof. In other words, our goal is to show that H𝑛 (𝑇 ) −
→ H𝑛 (|𝑇 |) is an isomorphism.

72
 For sub-Δ-sets𝑇 00 ⊆ 𝑇 0 we define Δ• (𝑇 0,𝑇 00) = Δ• (𝑇 0)/Δ• (𝑇 00) and accordingly, H𝑛 (𝑇 0,𝑇 00) =
H𝑛 (Δ• (𝑇 0,𝑇 00)). By Proposition 4.37, it sits inside a familiar long exact sequence involving the
simplicial homology of 𝑇 0 and 𝑇 00. We may apply this to 𝑇 0 = 𝑇 𝑘 and 𝑇 00 = 𝑇 𝑘−1 , the Δ-sets of
simplices in 𝑇 of dimension at most 𝑘 and 𝑘 − 1, respectively. This gives rise to a morphism of
exact sequences, like so (and continuing in both directions):

H𝑛+1 (𝑇 𝑘 ,𝑇 𝑘−1 ) H𝑛 (𝑇 𝑘−1 ) H𝑛 (𝑇 𝑘 ) H𝑛 (𝑇 𝑘 ,𝑇 𝑘−1 ) H𝑛−1 (𝑇 𝑘−1 )

H𝑛+1 (|𝑇 𝑘 |, |𝑇 𝑘−1 |) H𝑛 (|𝑇 𝑘−1 |) H𝑛 (|𝑇 𝑘 |) H𝑛 (|𝑇 𝑘 |, |𝑇 𝑘−1 |) H𝑛−1 (|𝑇 𝑘−1 |)

When 𝑘 = 0, |𝑇 0 | is a discrete topological space on the set 𝑇0 , and the map H𝑛 (𝑇 0 ) → H𝑛 (|𝑇 0 |)
is clearly an isomorphism. By induction and the five lemma, the middle vertical arrow in the
diagram above is an isomorphism if the first and the fourth are. We will show this in Lemma 8.2
below. Assuming this for now, let us complete the proof as follows. (When 𝑇 = 𝑇 𝑘 for some 𝑘
then we are already done. Only when 𝑇 has simplices of arbitrarily large dimension we need
to supply an additional argument.)
Let 𝑧 ∈ 𝑍𝑛 (𝑇 ) be an 𝑛-cycle whose image in H𝑛 (|𝑇 |) vanishes. In other words, there
exists 𝜏 ∈ 𝐶𝑛+1 (|𝑇 |) with 𝜕𝜏 = 𝑧. It is a general fact about the geometric realization of a
Δ-set that every compact subspace of |𝑇 | must lie in some |𝑇 𝑘 |. Hence 𝜏 ∈ 𝐶𝑛+1 (|𝑇 𝑘 |) for
some 𝑘 > 𝑛 and we deduce that 𝑧 maps to zero in H𝑛 (|𝑇 𝑘 |). By the previous argument then
𝑧 = 0 ∈ H𝑛 (𝑇 𝑘 ) = H𝑛 (𝑇 ). This shows injectivity of H𝑛 (𝑇 ) → H𝑛 (|𝑇 |).
Similarly, let 𝜎 ∈ 𝑍𝑛 (|𝑇 |) be an 𝑛-cycle. As we observed before 𝜎 ∈ 𝑍𝑛 (|𝑇 𝑘 |) for some 𝑘 > 𝑛
and by the argument above, [𝜎] comes from H𝑛 (𝑇 𝑘 ) = H𝑛 (𝑇 ). This shows surjectivity of
H𝑛 (𝑇 ) → H𝑛 (|𝑇 |) and we are done. 
∼
Lemma 8.2. The canonical homomorphism H𝑛 (𝑇 𝑘 ,𝑇 𝑘−1 ) −
→ H𝑛 (|𝑇 𝑘 |, |𝑇 𝑘−1 |) is an isomorphism.

Proof. We are going to identify both sides independently with the free abelian group on the
set 𝑇𝑘 of 𝑘-simplices, and then observe that generators are sent to generators in the obvious
way.
Note that Δ• (𝑇 𝑘−1 ) is a chain complex in degrees 𝑘 − 1, 𝑘 − 2, . . . , 0. And Δ• (𝑇 𝑘 ) is the
same chain complex with an additional term Z𝑇𝑘 in degree 𝑘. It follows that Δ• (𝑇 𝑘 ,𝑇 𝑘−1 ) is
the chain complex with Z𝑇𝑘 concentrated in degree 𝑘. In particular, we have:
(
Z𝑇𝑘 : 𝑛 = 𝑘
(8.3) H𝑛 (𝑇 𝑘 ,𝑇 𝑘−1 ) =
0 :𝑛 ≠𝑘

By construction of the geometric realization, we have homeomorphisms

|𝑇 𝑘 | Δ𝑘 × 𝑇𝑘 Ü Δ𝑘
≈ ≈ .
|𝑇 𝑘−1 | 𝜕Δ𝑘 × 𝑇𝑘 𝑇
𝜕Δ𝑘
𝑘

73
We deduce from Lemma 7.40 and Example 5.31 that25
(
Z𝑇𝑘 :𝑛 =𝑘
(8.4) H𝑛 (|𝑇 |, |𝑇
𝑘 𝑘−1
|)  H̃𝑛 (|𝑇 |/|𝑇
𝑘 𝑘−1
|) 
0 :𝑛 ≠𝑘

with generators corresponding to elements 𝑠 ∈ 𝑇𝑘 via the relative cycles 𝑠 : Δ𝑘 → |𝑇 𝑘 |. It

follows that the map in the statement identifies the groups in (8.3) and (8.4). 

Corollary 8.5. The simplicial homology H•Δ (𝑋 ) depends on 𝑋 only (and not on the Δ-complex
structure).

Corollary 8.6. Suppose 𝑋 has a Δ-complex structure with simplices in dimension ≤ 𝑘 only. Then
H𝑛 (𝑋 ) = 0 for all 𝑛 > 𝑘.

Example 8.7. Recall (Corollary 5.9) our construction of a fundamental class for the sphere 𝑆 𝑘 .
We now describe an alternative approach, using Theorem 8.1.
Namely, 𝑆 𝑘 has a Δ-complex structure obtained by glueing two copies of Δ𝑘 along the
boundary. Thus

Δ𝑘 (𝑆 𝑘 ) = Z2 = h𝜎+, 𝜎− i, Δ𝑘−1 (𝑆 𝑘 ) = Z𝑘+1 = h𝜏0, . . . , 𝜏𝑘 i

with 𝜕(𝜎± ) = (−1)𝑖 𝜏𝑖 . It follows that

Í

H𝑘 (𝑆 𝑘 )  H𝑘Δ (𝑆 𝑘 ) = Z = h𝜎+ − 𝜎− i.

Remark 8.8. Similarly, one could construct fundamental classes for the orientable surfaces Σ𝑔
using a Δ-complex structure, cf. Remark 7.44.

8.2 CW complexes
Recall that a CW complex is a topological space 𝑋 obtained as follows:

1. Start with a discrete space 𝑋 0 , the 0-cells.

2. Inductively, the 𝑛-skeleton 𝑋 𝑛 is obtained from 𝑋 𝑛−1 by attaching 𝑛-cells 𝐷𝛼𝑛 along maps
𝜙𝛼 : 𝜕𝐷𝛼𝑛 = 𝑆𝛼𝑛−1 → 𝑋 𝑛−1 .
3. If 𝑋 = 𝑋 𝑘 and 𝑘 is minimal with this property then 𝑋 is of dimension 𝑘. More generally
we can have 𝑋 = ∪𝑛 𝑋 𝑛 , in which case a subspace 𝑈 ⊆ 𝑋 is open iff 𝑈 ∩ 𝑋 𝑛 ⊆ 𝑋 𝑛 is open
for all 𝑛.

The 𝜙𝛼 are called attaching maps. Since 𝑋 𝑛 = 𝑋 𝑛−1 ∪ q𝜙𝛼 (q𝛼 𝐷𝛼𝑛 ), the attaching map extends to
a characteristic map Φ𝛼 : 𝐷𝛼𝑛 → 𝑋 𝑛 → 𝑋 .

Remark 8.9. As for Δ-complexes, this terminology is abusive. More accurately, we described
what a CW complex structure on a space 𝑋 consists of. There can be many such structures.
25 We use Exercise 6.3 to argue that these are good pairs.

74
Example 8.10. The sphere 𝑆 𝑘 (for 𝑘 > 0) admits a CW complex structure with a single 0-cell ∗
and a single 𝑘-cell. The attaching map 𝜕𝐷 𝑘 = 𝑆 𝑘−1 → ∗ is the unique map.

Commentary 8.11. Every Δ-complex is a CW complex. The main difference between the
two is that the attaching maps 𝜙𝛼 : 𝜕𝐷𝛼𝑛 → 𝑋 𝑛−1 for CW complexes are allowed to be any
continuous map while for Δ-complexes, they restrict to the inclusion of an (𝑛 − 1)-cell for
each face of the 𝑛-simplex.
Example 8.10 shows how the greater flexibility allows for more parsimonious cell struc-
tures.

Example 8.12. A 1-dimensional CW complex is the same thing as a 1-dimensional Δ-complex.

Both of these can be identified with graphs.

Example 8.13. The torus T admits a CW complex structure with a single 0-cell to which one
attaches two loops (1-cells) 𝑎, 𝑏, and finally one 2-cell via the attaching map described by the
loop [𝑎, 𝑏] = 𝑎𝑏𝑎 −1𝑏 −1 .

Example 8.14. We may think of real projective 𝑘-space RP𝑘 as the quotient 𝑆 𝑘 /(𝑥 ∼ −𝑥) of the
sphere under the antipodal map. Alternatively, it is the quotient of one of the two hemispheres
𝐷 𝑘 with antipodal points on the boundary 𝜕𝐷 𝑘 = 𝑆 𝑘−1 identified. In other words, RP𝑘 is
obtained by attaching a 𝑘-cell to RP𝑘−1 along the canonical quotient map 𝑆 𝑘−1 → RP𝑘−1 .
Inductively, it follows that RP𝑘 has a CW structure with one cell in each dimension 0, 1, . . . , 𝑘.

Example 8.15. We don’t need to stop the process in the previous example. Continuing we
get the infinite union RP∞ = ∪𝑘 RP𝑘 as a CW complex with a single cell in each dimension.

More examples of CW complexes can be found on the exercise sheet 6.

8.3 Cellular homology

Commentary 8.16. Our next goal is to define a homology theory for CW complexes, sim-
ilarly to how we defined simplicial homology in Section 2. Of course, we would like to take
as 𝑛-chains 𝐶𝑛CW (𝑋 ) the free abelian group on the 𝑛-cells. But how to define the boundary
operator? The greater flexibility in attaching maps also means that there isn’t an obvious way
to say what the ‘oriented boundary’ of an 𝑛-cell is.
We will proceed somewhat differently. In this section, we define the cellular chain complex
of a CW complex in terms of relative singular homology. The definition is quite abstract but:

• Lemma 8.17 shows that its degree-𝑛 term is the free abelian group on the 𝑛-cells;
• Corollary 8.25 shows that the cellular chain complex computes singular homology.
• In the next section will we identify the boundary operator and thereby make cellular
homology an eminently computable theory.

75
Lemma 8.17. Let 𝑋 be a CW complex with 𝑛-cells {𝐷𝛼𝑛 }, 𝑛 > 0. Then
(
⊕𝛼 Z · [𝐷𝛼𝑛 ] : 𝑚 = 𝑛
H𝑚 (𝑋 𝑛 , 𝑋 𝑛−1 ) 
0 :𝑚 ≠𝑛
Proof. This is exactly as in the proof of Lemma 8.2:
H𝑚 (𝑋 𝑛 , 𝑋 𝑛−1 )  H̃𝑚 (𝑋 𝑛 /𝑋 𝑛−1 )  H̃𝑚 (∨𝛼 𝐷𝛼𝑛 /𝜕𝐷𝛼𝑛 )  ⊕𝛼 H̃𝑚 (𝐷𝛼𝑛 /𝜕𝐷𝛼𝑛 ) 
Remark 8.18. We may describe the isomorphism in the lemma explicitly as follows. Choose
a homeomorphism Δ𝑛 ≈ 𝐷 𝑛 . We then get a continuous map
Φ𝛼
Δ𝑛 ≈ 𝐷 𝑛 −−→ 𝑋 𝑛
which is in fact a relative cycle for the pair (𝑋 𝑛 , 𝑋 𝑛−1 ). Its relative homology class generates
the copy of Z corresponding to 𝐷𝛼𝑛 .

The following is the main technical lemma we need to define the cellular chain complex
and identify its homology with singular homology. You could compare it with simplicial
homology for Δ-complexes where the corresponding statements are obviously true.

Lemma 8.19. Let 𝑋 be a CW complex.

1. If 𝑋 is of dimension 𝑘 then H𝑛 (𝑋 ) = 0 for all 𝑛 > 𝑘.
2. The map H𝑛 (𝑋 𝑚 ) → H𝑛 (𝑋 ) induced by the inclusion 𝑋 𝑚 → 𝑋 is an isomorphism if 𝑛 < 𝑚
and surjective if 𝑛 = 𝑚.

Proof. 1. We do induction on 𝑘, with 𝑘 = 0 being clear (Example 3.9). Assuming the

statement for 𝑘 − 1 let us consider the long exact sequence for the pair (𝑋 𝑘 , 𝑋 𝑘−1 ):
(8.20) · · · → H𝑛+1 (𝑋 𝑘 , 𝑋 𝑘−1 ) → H𝑛 (𝑋 𝑘−1 ) → H𝑛 (𝑋 𝑘 ) → H𝑛 (𝑋 𝑘 , 𝑋 𝑘−1 ) → · · ·
By induction, the second term vanishes for 𝑛 > 𝑘 > 𝑘 − 1. By Lemma 8.17, the last term
vanishes for 𝑛 > 𝑘. Therefore the third term vanishes for 𝑛 > 𝑘.
2. Consider again the long exact sequence (8.20) for the pair above. When
∼
• 𝑛 < 𝑘 − 1, both outer terms vanish and the middle arrow H𝑛 (𝑋 𝑘−1 ) −→ H𝑛 (𝑋 𝑘 ) is
an isomorphism;
• 𝑛 < 𝑘, the last term vanishes and the middle arrow H𝑛 (𝑋 𝑘−1 )  H𝑛 (𝑋 𝑘 ) is surjec-
tive.
In particular, for 𝑛 < 𝑚 we have isomorphisms
∼ ∼ ∼
(8.21) H𝑛 (𝑋 𝑚 ) −
→ H𝑛 (𝑋 𝑚+1 ) −
→ H𝑛 (𝑋 𝑚+2 ) −
→ ···
A compactness argument as in the proof of Theorem 8.1 shows that these groups are
necessarily H𝑛 (𝑋 ).
On the other hand, if 𝑛 = 𝑚, then the first arrow in (8.21) is (in general) only surjective
while the remaining ones are isomorphisms. This completes the proof.


76
 We will now contemplate the following diagram that is ‘spliced together’ from the long
exact sequences for the various pairs and which define the horizontal arrows in red.

50

0 H6 𝑛 (𝑋 𝑛+1 )
(
H (𝑋 𝑛 )
6 𝑛
𝛼 𝛽
(
H𝑛+1 (𝑋 𝑛+1, 𝑋 𝑛 )
𝑑𝑛+1
/ H𝑛 (𝑋 𝑛 , 𝑋 𝑛−1 ) 𝑑𝑛
/ H𝑛−1 (𝑋 𝑛−1, 𝑋 𝑛−2 )
5
𝛾 ( 𝛿
H5 𝑛−1 (𝑋 𝑛−1 )

Note that the various zeros come from Lemma 8.19.

Lemma 8.22. 1. The composition 𝑑𝑛 ◦ 𝑑𝑛+1 is zero.

2. The quotient group ker(𝑑𝑛 )/img(𝑑𝑛+1 ) is canonically isomorphic to H𝑛 (𝑋 ).

Proof. 1. The composition 𝑑𝑛 ◦ 𝑑𝑛+1 factors through the two blue arrows in the diagram.
These are part of a long exact sequence hence their composite is zero.
2. Since 𝛿 is injective, we have

(8.23) ker(𝑑𝑛 ) = ker(𝛾) = img(𝛽).

We also know by Lemma 8.19 that H𝑛 (𝑋 𝑛+1 )  H𝑛 (𝑋 ) canonically. That is,

(8.23)
H𝑛 (𝑋 )  coker(𝛼)  img(𝛽)/img(𝛽 ◦ 𝛼) = ker(𝑑𝑛 )/img(𝑑𝑛+1 ),

where the second isomorphism comes from the fact that 𝛽 is injective.


Definition 8.24. Let 𝑋 be a CW complex. We define the cellular chain complex 𝐶 •CW (𝑋 ) so
that
CW
(𝑋 ) = H𝑛+1 (𝑋 𝑛+1, 𝑋 𝑛 ) −−−→ 𝐶𝑛CW (𝑋 ) = H𝑛 (𝑋 𝑛 , 𝑋 𝑛−1 ) −−→ · · ·
𝑑𝑛+1 𝑑𝑛
· · · → 𝐶𝑛+1
The cellular homology groups are the homology of this chain complex:

H𝑛CW (𝑋 ) = H𝑛 (𝐶 •CW (𝑋 ))

Corollary 8.25. For any CW complex 𝑋 there are canonical isomorphisms H𝑛CW (𝑋 )  H𝑛 (𝑋 ).

8.4 Computing cellular homology

By Lemma 8.22, we know that cellular homology theoretically computes singular homology
for CW complexes. Our goal now is to make this a practical tool.

77
Remark 8.26. Let 𝑋 be a CW complex so that for 𝑛 > 0

𝑋 𝑛 = 𝑋 𝑛−1 ∪𝜙𝛼 (q𝛼 𝐷𝛼𝑛 )

for the attaching maps 𝜙𝛼 : 𝜕𝐷𝛼𝑛 → 𝑋 𝑛−1 . Note that we have26

(8.27) 𝑋 𝑛 /𝑋 𝑛−1 ≈ ∨𝛼 𝐷𝛼𝑛 /𝜕𝐷𝛼𝑛 = ∨𝛼 𝑆𝛼𝑛 ,

and in particular there are canonical quotient maps, obtained by collapsing all but one sphere
to a point:
𝜋𝛼 : 𝑋 𝑛  𝑋 𝑛 /𝑋 𝑛−1  𝑆𝛼𝑛

Remark 8.28. By Lemma 8.17, the cellular chain complex in degree 𝑛 is the free abelian
group on the 𝑛-cells. So, it remains to identify the boundary operator in terms of a chosen
basis. Recall that a basis ( [𝐷𝛼𝑛 ])𝛼 for 𝐶𝑛CW (𝑋 ) may be obtained by choosing a generator [𝐷 𝑛 ] ∈
H𝑛 (𝐷 𝑛 , 𝜕𝐷 𝑛 ):

(8.29) [𝐷𝛼𝑛 ] = (Φ𝛼 )∗ ( [𝐷 𝑛 ]) ∈ H𝑛 (𝑋 𝑛 , 𝑋 𝑛−1 )

Proposition 8.30. There are generators [𝐷 𝑛 ] such that in terms of the basis (8.29) for the groups
𝐶 •CW (𝑋 ), the boundary operator is given by the following formula:
• in degree 𝑛 = 1:
𝑑 1 ( [𝐷𝛼1 ]) = [𝜙𝛼 (1)] − [𝜙𝛼 (−1)]

• in degrees 𝑛 > 1: ∑︁
𝑑𝑛 ( [𝐷𝛼𝑛 ]) = 𝑑𝛼 𝛽 [𝐷 𝑛−1
𝛽 ],
𝛽

where  
𝜙𝛼 𝜋𝛽
𝑑𝛼 𝛽 = deg Δ𝛼 𝛽 : 𝑆𝛼𝑛−1 −−→ 𝑋 𝑛−1
−−→ 𝑆 𝛽𝑛−1

Remark 8.31. If one changes the generator [𝐷 𝑛 ] (for all 𝛼 simultaneously!) then the formula
for the boundary operator changes by a sign. Of course, this does not affect the homology
of the complex so in practice this is not so important. In the sequel we will not dwell on this
point.

Before giving the proof let us do some examples.

Example 8.32. Whenever there is a single 0-cell, the first boundary operator 𝑑 1 : 𝐶 1CW (𝑋 ) →
𝐶 0CW (𝑋 ) is the zero map.

Example 8.33. Let us consider the ‘minimal’ CW complex structure on the sphere 𝑆 𝑘 , 𝑘 > 0,
as in Example 8.10. The cellular chain complex has a copy of the integers in degrees 0 and 𝑘.
If 𝑘 = 1, the boundary operator 𝑑 1 = 0 by the previous example. If 𝑘 > 1, then all boundary
operators are clearly zero as well. We deduce that HCW
• (𝑆 𝑘 ) = 𝐶 •CW (𝑆 𝑘 ) has two copies of the
integers, in degrees 0 and 𝑘. (Of course, we knew that already.)
26 To avoid any ambiguities in the sequel we fix once and for all homeomorphisms 𝐷 𝑛 /𝜕𝐷 𝑛 ≈ 𝑆 𝑛 for all 𝑛 > 0.

78
Example 8.34. Let 𝑋 = Σ𝑔 be the orientable surface of genus 𝑔 ≥ 0. It admits a CW complex
structure with a single 0-cell, 2𝑔 1-cells 𝑎 1, 𝑏 1, . . . , 𝑎𝑔 , 𝑏𝑔 , and a single 2-cell with attaching
𝜋𝛽
− 𝑋 1 −−→ 𝑆 𝛽1 is
map 𝛾 : 𝜕𝐷 2 = 𝑆 1 → 𝑋 1 given by the loop [𝑎 1, 𝑏 1 ] · · · [𝑎𝑔 , 𝑏𝑔 ]. The map 𝑆 1 →
𝛾

homotopic to the constant map since each 𝑎𝑖 and 𝑏𝑖 appears together with its inverse. It follows
that the cellular chain complex has the shape:
0 0
− Z2𝑔 →
0 → Z→ − Z→0

We deduce that the cellular homology groups HCW

• (Σ𝑔 ) coincide with those found in Propo-
sition 7.41.

Exercise 8.35. Repeat this computation for the non-orientable surface 𝑁ℎ , cf. Exercise 6.4.

Example 8.36. Let us compute the cellular homology of RP𝑘 , with the CW complex structure
of Example 8.14. Thus it has a single cell in each dimension 𝑛 = 0, . . . , 𝑘. To determine the
boundary operator 𝑑𝑛 we consider—as we should—the map
2:1
𝑓 : 𝑆 𝑛−1 −−→ RP𝑛−1  RP𝑛−1 /RP𝑛−2 ≈ 𝑆 𝑛−1,

and use the local degree formula (Proposition 6.21) to compute its degree. At the ‘North
pole’ 𝑁 (importantly, away from the equator), the map is locally a homeomorphism hence
has degree ±1.27 At the ‘South pole’ 𝑆, the map is also locally a homeomorphism, namely the
antipodal map followed by the one at 𝑁 . Since the antipodal map has degree (−1)𝑛 , it follows
that (
±2 : 𝑛 even
deg(𝑓 ) = deg(𝑓 | 𝑁 ) + (−1) deg(𝑓 | 𝑁 ) =
𝑛
0 : 𝑛 odd
We conclude that the cellular chain complex looks as follows:

𝑑𝑘 ±2 0 ±2 0
0 → Z −−→ Z → · · · → Z −−→ Z → − Z → 0,
− Z −−→ Z →

with 𝑑𝑘 = ±2 if 𝑘 is even, and 𝑑𝑘 = 0 if 𝑘 is odd. We conclude that



 Z : 𝑛 = 0, and 𝑛 = 𝑘 if odd
H𝑛 (RP ) = Z/2Z


𝑘
: 0 < 𝑛 < 𝑘 odd

0

: else


Example 8.37. It follows from the computations in the previous example that



 Z :𝑛 =0
H𝑛 (RP ) = Z/2 : 0 < 𝑛 odd
∞



0

: else

27 Pedant remark (which you should probably ignore): The sign of the degree actually depends on the chosen
homeomorphism 𝐷 𝑛−1 /𝜕𝐷 𝑛−1 ≈ 𝑆 𝑛−1 which we have fixed throughout the discussion.

79
 Proof is Proof of Proposition 8.30. We choose the ‘obvious’ generator [𝐷 0 ] of H0 (𝐷 0, ∅) = Z given by
non- the unique simplex Δ0 → 𝐷 0 = ∗. Next we choose a homeomorphism Δ1 ≈ 𝐷 1 which gives a
examinable path from −1 to 1. As in Remark 8.18, this gives a generator [𝐷 1 ] ∈ H1 (𝐷 1, 𝜕𝐷 1 ). Inductively,
we define [𝐷 𝑛 ] (for 𝑛 > 1) as the generator corresponding to [𝐷 𝑛−1 ] ∈ H𝑛−1 (𝐷 𝑛−1, 𝜕𝐷 𝑛−1 )
under the sequence of isomorphisms
𝜕
(8.38) H𝑛 (𝐷 𝑛 , 𝑆 𝑛−1 ) →
− H𝑛−1 (𝑆 𝑛−1 )  H𝑛−1 (𝐷 𝑛−1 /𝑆 𝑛−2 )  H𝑛−1 (𝐷 𝑛−1, 𝑆 𝑛−2 ).

Here, the first isomorphism is the connecting homomorphism in the long exact sequence
for the pair, the second isomorphism is induced by our chosen homeomorphism 𝑆 𝑛−1 ≈
𝐷 𝑛−1 /𝜕𝐷 𝑛−1 , and the last isomorphism is an instance of Proposition 5.28.
With this chosen basis in each dimension we now verify the formula for the boundary
operator in the statement:
In degree 𝑛 = 1, the boundary operator is given by the connecting homomorphism
H1 (𝑋 1, 𝑋 0 ) → H0 (𝑋 0 ). Start with a 1-cell 𝐷𝛼1 . The ‘characteristic map’
Φ𝛼
Φ̃𝛼 : Δ1 ≈ 𝐷𝛼1 −−→ 𝑋 1

is a relative cycle and, by Exercise 5.19, the connecting homomorphism takes this to the class
of 𝜕 Φ̃𝛼 = 𝜙𝛼 (1) − 𝜙𝛼 (−1), by our choice of the homeomorphism Δ1 ≈ 𝐷 1 . We deduce that

(8.29)
 
𝑑 1 ( [𝐷𝛼𝑛 ]) = 𝑑 1 ( Φ̃𝛼 )∗ ( [Δ1 ]) = 𝑑 1 ( [ Φ̃𝛼 ]) = [𝜙𝛼 (1)] − [𝜙𝛼 (−1)].

In degree 𝑛 > 1, we consider the following diagram:

(Δ𝛼 𝛽 )∗
H𝑛 (𝐷𝛼𝑛 , 𝜕𝐷𝛼𝑛 ) H𝑛−1 (𝑆𝛼𝑛−1 ) H𝑛−1 (𝑆 𝛽𝑛−1 ) H𝑛−1 (𝐷 𝑛−1 H𝑛−1 (𝐷 𝑛−1
𝜕
∼ ∼ 𝛽
/𝜕𝐷 𝑛−1
𝛽
) ∼ 𝛽
, 𝜕𝐷 𝑛−1
𝛽
)
(Φ𝛼 )∗ (𝜙𝛼 )∗
(𝜋 𝛽 )∗
∼
H𝑛 H𝑛−1 H𝑛−1 (𝑋 𝑛−1, 𝑋 𝑛−2 ) H𝑛−1 (𝑋 𝑛−1 /𝑋 𝑛−2 )
𝜕
(𝑋 𝑛 , 𝑋 𝑛−1 ) (𝑋 𝑛−1 )

It is clearly commutative. Note that the composite of the two red arrows is the boundary
operator 𝑑𝑛 . Now, start with [𝐷 𝑛 ] in the top left. Following the path down-right-up-right all
the way to the top right, we get a multiple of [𝐷 𝑛−1 𝛽
], the coefficient of 𝑑𝑛 ( [𝐷𝛼𝑛 ] for the basis
element [𝐷 𝑛−1
𝛽
]. On the other hand, following the path straight to the right we get 𝑑𝛼 𝛽 · [𝐷 𝑛−1𝛽
],
by (8.38). 

9 Thr Euler characteristic

We often measure mathematical objects by their ‘size’: sets by their cardinality, vector spaces
by their dimension, abelian groups by their rank etc. When you try to define an analogous
notion of size in algebraic topology, you end up with the Euler characteristic. Even more than
the homology groups it is a way of breaking down the complexity of a space, in this case to
a single number (okay, integer). Given this, it is quite surprising how powerful this notion
turns out to be.

80
9.1 Definition
Euler didn’t speak of ‘Euler characteristics’, obviously, but he was one of the first to systemat-
ically apply this notion. Let us recall the formula he found (Proposition 9.4).

Definition 9.1. A plane graph is a finite 1-dimensional CW complex embedded in the real
plane R2 . Equivalently, it is a finite graph in the plane in which the edges don’t cross.

a plane graph not a plane graph

Commentary 9.2. There are finite graphs that admit no such an embedding in R2 , and these
are called nonplanar. (Note that the example on the right above can be embedded in R2 , as a
square, hence it is planar.) An example of a nonplanar graph is the complete graph 𝐾5 on five
vertices. Another one is the complete bipartite graph 𝐾3,3 :

There are theoretical criteria as well as practical algorithms for deciding whether a graph is
planar.

Remark 9.3. A face of a plane graph G is a connected component of R2 \ G:

𝐹3

𝐹1 𝐹2

Recall now:

Proposition 9.4 (Euler’s formula). For a planar graph with 𝑣 vertices, 𝑒 edges and 𝑓 faces we have

𝑣 − 𝑒 + 𝑓 = 2.

81
Example 9.5. Recall the complete bipartite graph 𝐾3,3 . We have 𝑣 = 6 and 𝑒 = 9. Moreover,
every face has an even number of boundary edges, that is, at least 4. So 4𝑓 ≤ 2𝑒 = 18, or
𝑓 ≤ 4. Hence
𝑣 − 𝑒 + 𝑓 ≤ −3 + 4 = 1.
Euler’s formula shows that 𝐾3,3 cannot be embedded in R2 , that is, 𝐾3,3 is nonplanar.

Commentary 9.6. The argument for Proposition 9.4 we are going to give will link the
situation with homology and introduce several important ideas. That’s not to say that Euler
thought of it in this way.

Remark 9.7. By adding a point at ∞, a planar graph yields a CW complex structure on the
2-sphere, with 𝑣 many 0-cells, 𝑒 1-cells and 𝑓 2-cells. Right?

The discussion around the Jordan curve theorem should have made you wary of such a claim.
And rightly so: What if the edges aren’t as nice? Is it still true that they form the boundaries
of 2-cells?
The answer is yes, but this is (a special case of ) a non-trivial result called the Schoenflies’
Theorem. (Which we are going to assume without proof. Alternatively, we give a proof of
Euler’s formula for those plane graphs that underlie a CW structure on 𝑆 2 .) So, Euler’s formula
will follow from the following statement since the minimal CW complex structure on 𝑆 2 has
a single 0-cell and a single 2-cell (Example 8.10).

Proposition 9.8. Let 𝑋 be a topological space that admits the structure of a finite CW complex.
Then the alternating sum ∑︁
𝜒 (𝑋 ) = (−1)𝑛 #{𝑛-cells}
𝑛 ≥0

is independent of the choice of CW complex structure.

Remark 9.9. The alternating sum 𝜒 (𝑋 ) is called the Euler characteristic of 𝑋 . We will define
it in greater generality below.

Definition 9.10. Let 𝐴 be a finitely generated abelian group. It decomposes as 𝐴 = 𝑇 ⊕ 𝐹 , the

direct sum of the torsionfree part 𝑇  Z𝑟 (which is necessarily free abelian) and a finite abelian
group 𝐹 . The integer 𝑟 is well-defined and called the rank of 𝐴.
Alternatively, rk(𝐴) is the dimension of the Q-vector space 𝐴 ⊗Z Q. This follows from the
fact that Z𝑟 ⊗ Q = Q𝑟 while 𝐹 ⊗ Q = 0.

Exercise 9.11. Here is a warm-up exercise for you:

82
 1. Let 0 → 𝑉1 → 𝑉2 → 𝑉3 → 0 be a short exact sequence, with 𝑉𝑖 finite dimensional
Q-vector spaces. Then dim(𝑉0 ) − dim(𝑉2 ) + dim(𝑉3 ) = 0.
𝑑𝑛+1 𝑑𝑛
2. Let · · · → 𝐶𝑛+1 −−−→ 𝐶𝑛 −−→ 𝐶𝑛−1 → · · · be an exact chain complex. Then its Q-
𝑑𝑛+1 𝑑𝑛
linearization · · · → 𝐶𝑛+1 ⊗ Q −−−→ 𝐶𝑛 ⊗ Q −−→ 𝐶𝑛−1 ⊗ Q → · · · is still exact.
3. Conclude that for a short exact sequence 0 → 𝐴1 → 𝐴2 → 𝐴3 → 0 of finitely generated
abelian groups we have
rk(𝐴1 ) − rk(𝐴2 ) + rk(𝐴3 ) = 0.

Corollary 9.12. Let 𝐶 • be a chain complex with only finitely many non-zero terms, all of which are
finitely generated abelian groups. Then
∑︁ ∑︁
(−1)𝑛 rk(𝐶𝑛 ) = (−1)𝑛 rk(H𝑛 (𝐶 • ))
𝑛 𝑛

Proof. We have two short exact sequences, for all 𝑛:

0 → 𝑍𝑛 → 𝐶𝑛 → 𝐵𝑛−1 → 0
0 → 𝐵𝑛 → 𝑍𝑛 → H𝑛 → 0

We therefore get
∑︁ ∑︁
(−1)𝑛 rk(𝐶𝑛 ) = (−1)𝑛 [rk(𝑍𝑛 ) + rk(𝐵𝑛−1 )] Exercise 9.11
𝑛 𝑛
∑︁
= (−1)𝑛 [rk(𝑍𝑛 ) − rk(𝐵𝑛 )]
𝑛
∑︁
= (−1)𝑛 rk(H𝑛 ) Exercise 9.11
𝑛

Proof of Proposition 9.8. By the Corollary, 𝜒 (𝑋 ) = rk(H𝑛 (𝑋 )) and the right-hand side
Í
𝑛 (−1)
𝑛

is independent of the cell structure. 

The proof suggests the following definition.

Definition 9.13. Let 𝑋 be a space with only finitely many non-zero homology groups, all of
which are finitely generated abelian groups. Then its Euler characteristic is defined as
∑︁
𝜒 (𝑋 ) := (−1)𝑛 rk(H𝑛 (𝑋 )).
𝑛 ≥0

Remark 9.14. To emphasize, if 𝑋 is a finite CW complex then Corollary 9.12 shows that
∑︁ ∑︁
𝜒 (𝑋 ) = (−1)𝑛 rk(H𝑛 ) = (−1)𝑛 #{𝑛-cells},
𝑛 𝑛

so this is compatible with the notation earlier in Proposition 9.8. As we will see below, going
back and forth between these two expressions can be a powerful tool.

83
Example 9.15. As we observed above, 𝜒 (𝑆 2 ) = 2. More generally, we have
(
2 : 𝑘 even
𝜒 (𝑆 𝑘 ) =
0 : 𝑘 odd

This can be seen

• either from the Definition 9.13 and our homology computation in Corollary 4.13,
• or from the ‘minimal’ cell structure Example 8.10.
Note that although there are many CW complex structures on 𝑆 𝑘 none of it has an odd number
of cells!

Remark 9.16. Let 𝐶 • be a chain complex that has only finitely many non-zero terms, all of
which are finitely generated abelian groups. If 𝐶 • is exact then 𝑛 (−1)𝑛 𝜒 (𝐶𝑛 ) = 0. This
Í
follows from Corollary 9.12.

Example 9.17. Consider a CW structure on 𝑆 2 made of 𝑣 vertices, 𝑒 edges, and 2-cells made
of 𝑝 pentagons and ℎ hexagons. For example, consider a football:

© Pumbaa80 / Wikimedia Commons / CC-BY-SA-3.0

We have
• 2𝑒 = 6ℎ + 5𝑝,
• 3𝑣 = 2𝑒,
• 2 = 𝜒 (𝑆 2 ) = 𝑣 − 𝑒 + 𝑝 + ℎ.
Putting these together we get

6𝑝 = 12 + 6𝑒 − 6𝑣 − 6ℎ = 12 + 2𝑒 − 6ℎ = 12 + 6ℎ + 5𝑝 − 6ℎ = 12 + 5𝑝

that is, 𝑝 = 12.

Note that for the football pictured above one has ℎ = 20. However, the regular dodecahe-
dron has ℎ = 0, for example.

© DTR / Wikimedia Commons / CC-BY-SA-3.0

84
Example 9.18. Let us come back to the complete bipartite graph 𝐾3,3 . As we showed in
Example 9.5, any drawing would need 𝑓 ≤ 4, and the graph is therefore nonplanar. However,
it can be drawn on T , K with 𝑓 = 3, and on RP2 with 𝑓 = 4:

So (
0 on T or K
𝑣 −𝑒 + 𝑓 =
1 on RP2
These are the Euler characteristic of the three spaces as we know from either their cell structure
(for example, as Δ-complexes) or the homology we computed earlier.
See also Exercise 7.4.

9.2 Properties
In this section we prove many of the properties one would expect a measure of size to satisfy.
To get an intuition for what these are note that any set 𝑋 can be viewed as a topological space
with the discrete topology. If 𝑋 is finite then 𝜒 (𝑋 ) = rk(H0 (𝑋 )) = |𝑋 | is its cardinality. In
the case of finite discrete spaces, some of the properties we will prove reduce to the following
trivialities:
• |𝑋 ∪ 𝑌 | = |𝑋 | + |𝑌 | − |𝑋 ∩ 𝑌 |
• |𝑋 × 𝑌 | = |𝑋 | · |𝑌 |
• If 𝑓 : 𝑌 → 𝑋 is a 𝑑 : 1-map then |𝑌 | = 𝑑 · |𝑋 |.

Proposition 9.19. Let 𝑋 = 𝑈 ∪ 𝑉 and assume that

1. either, 𝑋 is a CW complex and 𝑈 , 𝑉 are subcomplexes,
2. or 𝑈 , 𝑉 ⊆ 𝑋 are open subsets.
If 𝜒 (𝑈 ), 𝜒 (𝑉 ) and 𝜒 (𝑈 ∩ 𝑉 ) are all defined then so is 𝜒 (𝑋 ) and we have

𝜒 (𝑋 ) = 𝜒 (𝑈 ) + 𝜒 (𝑉 ) − 𝜒 (𝑈 ∩ 𝑉 ).

Proof. In the first case, if 𝑈 and 𝑉 are finite 𝐶𝑊 complexes then so is 𝑋 . Looking at the
number 𝑐𝑛 of cells in each dimension 𝑛 we have

𝑐𝑛 (𝑋 ) = 𝑐𝑛 (𝑈 ) + 𝑐𝑛 (𝑉 ) − 𝑐𝑛 (𝑈 ∩ 𝑉 )

which implies the identity in the statement.

For the second case look at the Mayer-Vietoris long exact sequence and use that the alter-
nating sum of the ranks is 0, by Remark 9.16. 

85
Example 9.20. Recall the connected sum 𝑆 1 #𝑆 2 of surfaces (section 7.3). We have for a small
disk 𝐷 ⊆ 𝑆𝑖 ,
˚ − 𝜒 (𝑆 1 )
𝜒 (𝑆𝑖 ) = 𝜒 (𝐷) + 𝜒 (𝑆 2 \𝐷)
˚ = 𝜒 (𝑆𝑖 ) − 1. It follows that
so that 𝜒 (𝑆 2 \𝐷)

˚ + 𝜒 (𝑆 2 \𝐷)
𝜒 (𝑆 1 #𝑆 2 ) = 𝜒 (𝑆 1 \𝐷) ˚ − 𝜒 (𝑆 1 )
= 𝜒 (𝑆 1 ) + 𝜒 (𝑆 2 ) − 2.

˚ 𝐷, or 𝑆 1 but
(To be sure, we are applying Proposition 9.19 not directly to the (sub)spaces 𝑆𝑖 \𝐷,
rather to suitable open neighborhoods of these.)

Corollary 9.21. We have:

𝜒 (Σ𝑔 ) = 2 − 2𝑔
𝜒 (𝑁ℎ ) = 2 − ℎ

Remark 9.22. Alternatively, this follows directly from the homology computations in Propo-
sitions 7.41 and 7.45.

Remark 9.23. The Euler characteristic of a surface can therefore attain any integer ≤ 2. The
sphere is the only surface with 𝜒 = 2 and 𝑁ℎ the only one with 𝜒 = 2 − ℎ as long as ℎ is
odd. Every even nonpositive integer is the Euler characteristic of precisely two surfaces, one
orientable and one non-orientable.
In particular: Surfaces are completely classified by

• whether they are orientable or not, and

• their Euler characteristic.

Proposition 9.24. Let 𝑋 and 𝑌 be finite CW complexes. Then so is 𝑋 × 𝑌 and

𝜒 (𝑋 × 𝑌 ) = 𝜒 (𝑋 ) · 𝜒 (𝑌 ).

Proof. This follows from Exercise 7.6 in which you show that
∑︁
𝑐𝑛 (𝑋 × 𝑌 ) = 𝑐 𝑎 (𝑋 ) · 𝑐𝑏 (𝑌 ). 
𝑎+𝑏=𝑛

Remark 9.25. In fact, the statement remains true for arbitrary topological spaces whenever
𝜒 (𝑋 ) and 𝜒 (𝑌 ) are defined. Obviously this imposes restrictions on when a space can be written
as a product of certain other spaces. For example, if a space has non-zero Euler characteristic
it isn’t homeomorphic to 𝑆 3 × 𝑌 , for any 𝑌 .
We will not prove nor use that result, however.

Proposition 9.26. Let 𝑝 : 𝑌 → 𝑋 be a 𝑑-sheeted cover and assume that 𝑋 is a finite CW complex.
Then so is 𝑌 and we have
𝜒 (𝑌 ) = 𝑑 · 𝜒 (𝑋 ).

86
Proof. Let Φ𝛼 : 𝐷𝛼𝑛 → 𝑋 be one of the characteristic maps. By the lifting property (and since
𝐷 𝑛 is simply-connected) there are precisely 𝑑 lifts Φ𝛼,𝑖 : 𝐷𝛼,𝑖
𝑛 → 𝑌 , and these, I claim, are the

characteristic maps for a CW complex structure. So, the main point is that 𝑌 admits a CW
structure such that above each cell 𝐷𝛼𝑛 in 𝑋 there are precisely 𝑑 cells 𝐷𝛼,1
𝑛 , . . . , 𝐷 𝑛 . This is a
𝛼,𝑑
good thing to know. And of course it immediately yields the identity in the proposition.
In the remainder I will prove the claim above. It relies on basic properties of covers and
is not terribly important in the sequel. We use Exercise 6.2 where you show (or read in
Hatcher’s book) that the collection (Φ𝛼,𝑖 )𝑛,𝛼,𝑖 defines a CW complex structure on 𝑌 if and only
if the following two conditions hold:28
≈
• Each Φ𝛼,𝑖 restricts to a homeomorphism from the interior 𝐷˚ 𝛼,𝑖
𝑛 − 𝑛 to its image and
→ 𝑒𝛼,𝑖
Ý 𝑛 .
𝑌 = 𝑛,𝛼,𝑖 𝑒𝛼,𝑖
• The image Φ𝛼,𝑖 (𝜕𝐷𝛼,𝑖
𝑛 ) is contained in cells of dimension less than 𝑛.

For the first condition, since Φ𝛼 = 𝑝 ◦Φ𝛼,𝑖 is injective on the interior so is Φ𝛼,𝑖 . Let 𝑈 ⊆ 𝐷˚ 𝛼,𝑖
𝑛 be a

(non-empty) open subset. To show that Φ𝛼,𝑖 (𝑈 ) ⊆ 𝑒𝛼,𝑖 is open pick a point 𝑦 in the image, with
𝑛

𝑥 = 𝑝 (𝑦) ∈ 𝑋 . There exists a small open neighborhood 𝑥 ∈ 𝑉 ⊆ 𝑒𝛼𝑛 such that 𝑝 −1 (𝑉 ) = 𝑑𝑗=1 𝑉𝑗
Ý

with each 𝑉𝑗 = 𝑉 . It follows that 𝑉𝑖 ∩ 𝑝 −1 (Φ𝛼 (𝑈 ) ∩𝑉 ) is an open neighborhood of 𝑦 inside 𝑒𝛼,𝑖 𝑛 .

If 𝑦 ∈ 𝑌 is an arbitrary point with image 𝑥 = 𝑝 (𝑦) downstairs, there exist unique 𝑛, 𝛼 such
that 𝑥 ∈ 𝑒𝛼𝑛 . By the unique lifting property, 𝑦 ∈ 𝑒𝛼,𝑖
𝑛 for exactly one 𝑖. And 𝑦 cannot belong to

some 𝑒 𝛽,𝑗 with (𝑛, 𝛼) ≠ (𝑚, 𝛽) either as otherwise 𝑥 ∈ 𝑒𝑚

𝑚
𝛽
. At this point we have established the
first condition.
For the second condition, if 𝑦 = Φ𝛼,𝑖 (𝑧) for some point 𝑧 ∈ 𝜕𝐷𝛼,𝑖 𝑛 on the boundary, then

𝑝 (𝑦) = 𝑥 ∈ 𝑒 𝛽 for some 𝑚 < 𝑛 and some 𝛽. Therefore 𝑦 ∈ 𝑒 𝛽,𝑗 for some (unique) 𝑗.
𝑚 𝑚 

Remark 9.27. Again, under suitable finiteness assumptions this statement is true for spaces
without CW complex structures.

Example 9.28. We know that all surfaces 𝑁ℎ for ℎ > 0 have a non-trivial orientation bun-
dle 𝑁˜ℎ . Necessarily it is one of the orientable surfaces Σ𝑔 . In Remark 7.50, we described how to
visualize the covering map Σ𝑔 → 𝑁ℎ . As an application of Proposition 9.26 we can determine
the genus 𝑔 very easily.
Indeed, it gives us the identity

𝜒 (Σ𝑔 ) = 2 · 𝜒 (𝑁ℎ ) ⇔
2 − 2𝑔 = 2 · (2 − ℎ) ⇔
𝑔 =ℎ−1

Example 9.29. For any 𝑘 ≥ 0, 𝑆 𝑘 → RP𝑘 is a 2-sheeted cover. It follows that

(
𝑘 1 𝑘 1 : 𝑘 even
𝜒 (RP ) = 𝜒 (𝑆 ) =
2 0 : 𝑘 odd
28 The third condition there is superfluous for finitely many cells.

87
If we endow RP𝑘 with the cell structure of Example 8.14 which has one cell in each dimen-
sion ≤ 𝑘 then the cell structure on 𝑆 𝑘 produced in the proof of Proposition 9.26 has two cells
in each dimension ≤ 𝑘. In fact, it can be described very easily as follows.
Start with ±1 as 0-cells. Attach two 1-cells as the upper and lower hemicircles. Think of
the result as the equator in 𝑆 2 . Attach then two 2-cells as the upper and lower hemisphere.
Continue like that to get 𝑆 𝑘 . The cover map 𝑆 𝑘 → RP𝑘 precisely identifies the two cells in
each dimension.
One advantage of this cell structure on 𝑆 𝑘 (in contrast to the minimal one of Example 8.10)
is that the 𝑛-skeleton of 𝑆 𝑘 identifies with 𝑆 𝑛 as long as 𝑛 ≤ 𝑘. So, the union 𝑆 ∞ := ∪𝑘 ≥0𝑆 𝑘
makes sense and has an obvious structure of CW complex.

Example 9.30 (Unimportant quiz). What should the Euler characteristic 𝜒 (RP∞ ) of infinite
real projective space be? To be sure, this is not something we have defined since this space has
infinitely many nonvanishing homology groups, as we saw in Example 8.37. So, it is more of
an idle question.

• However, you might argue, while the homology groups are non-zero in all odd degrees,
these are cyclic of order 2 hence their rank is zero. They should not contribute to the
Euler characteristic. Which is then just

𝜒 (RP∞ ) = rk(H0 (RP∞ )) = 1.

More generally, this may suggest that we could define the Euler characteristic for spaces 𝑋
with finitely generated abelian homology groups and such that rk(H𝑛 (𝑋 )) = 0 for
𝑛  0. But: read on!
• Of course, when your CW complex has infinitely many cells you cannot expect to be
able to compute the Euler characteristic as an alternating sum of numbers of cells in each
dimension. Just to drive this point home, for RP∞ we have one cell in each dimension
hence we should get ∑︁
𝜒 (RP∞ ) = (−1)𝑛 ,
𝑛 ≥0

which is a divergent series. Another way of saying this is that since RP∞ = ∪𝑘 ≥0 RP𝑘
we would expect
1 + (−1)𝑘
𝜒 (RP∞ ) = lim 𝜒 (RP𝑘 ) = lim ,
−−→
𝑘→∞
−−→
𝑘→∞
2

which does not converge.

• The CW complex 𝑆 ∞ considered in Example 9.29 comes with a canonical 2:1-cover
map 𝑆 ∞ → RP∞ . Now the homology of 𝑆 ∞ is very easy, by Lemma 8.19. Since H𝑛 (𝑆 𝑘 )
vanishes whenever 0 < 𝑛 < 𝑘 we have
(
Z :𝑛 =0
H𝑛 (𝑆 ) =
∞
0 :𝑛 >0

88
 and therefore 𝜒 (𝑆 ∞ ) = 1. (Note that this is defined!) By the formula for covers in
Proposition 9.26 we would therefore expect

1
𝜒 (RP∞ ) = .
2

So, the point is that in extending the notion of Euler characteristic, one needs to be careful in
extending the basic properties we established.29

10 Homology theories
Let’s take stock. We defined no less than three homology theories,

• simplicial,
• singular, and
• cellular,

each with its own advantages and disadvantages. Importantly, we also proved that they are
all the ‘same’. The goal of this last section is to explain why this is not a coincidence: that
everything ‘behaving like a homology theory’ is in fact the same.
Of course, the important bit here is what is in scare quotes. The insight of Eilenberg and
Steenrod (1945) was that a few axioms are sufficient for the characterization of a homology
theory. It won’t surprise you that some of the most important theorems we proved in the
The whole course of the past weeks will show up again as axioms.
section 10 is
non- Commentary 10.1. The topics we have discussed up till this point were all developed in what
examinable. one might call the ‘early period of algebraic topology’ (with the already mentioned Analysis
Situs papers by Poincaré around 1900 seen as its starting point). One could say that it took a
couple of decades for the subject of homology to become so stable that an axiomatization was
made possible and universally accepted (as far as I know). This axiomatization, then, seems
like a good place to end our foray into algebraic topology. Needless to say, homology theory
is only a small part of algebraic topology, with cohomology and homotopy theory two very
natural directions to head to thereafter, see Section 10.3.
29 There is actually an invariant you can associate to a space 𝑋 that has infinitely many non-zero homology
groups all of which are finitely generated abelian. This is the Poincaré series
∑︁
𝑃 (𝑋, 𝑡) := rk(H𝑛 (𝑋 ))𝑡 𝑛 ∈ Z[[𝑡]]
𝑛 ≥0

which is a formal series. If the Euler characteristic is defined then 𝜒 (𝑋 ) = 𝑃 (𝑋, −1). And it satisfies similar properties
as the Euler characteristic with respect to unions and products.
Also, there is an invariant that spits out 1/2 for RP∞ . It is called the homotopy cardinality. However, it behaves
more multiplicatively than additively.

89
10.1 Categories and functors
It’s certainly no accident that Eilenberg is one of the authors (the other being Mac Lane) who
introduced around the same time the notions of categories, functors, and natural transforma-
tions. These are used in the axiomatization of homology theory.

Commentary 10.2. So, what is a category?

A category is simply a formalization of the good practice stipulated in Commentary 2.47.
It is a language for speaking about objects and morphisms without reference to the subject
matter at hand. It is therefore necessarily abstract. Here, we will only introduce the very
basics.
Slightly more precisely, a category C is a collection of objects Ob( C) and for any pair of
objects 𝑐, 𝑐 0 ∈ Ob( C) a set of morphisms denoted Hom C(𝑐, 𝑐 0), which can be composed.

Before unpacking the last part of the definition let us look at some examples.

Example 10.3. 1. The category of topological spaces Top has as objects topological spaces
and morphisms are continuous maps.
2. The category of abelian groups Ab has as objects abelian groups and morphisms are
homomorphisms of abelian groups.
3. The category of chain complexes Cpx has as objects chain complexes and as morphisms
chain maps.

Commentary 10.4. Let 𝑓 : 𝑋 → 𝑌 and 𝑔 : 𝑌 → 𝑍 be continuous maps between topological

spaces. We can compose the two maps to obtain another continuous map

𝑔 ◦ 𝑓 : 𝑋 → 𝑍, 𝑥 ↦→ 𝑔(𝑓 (𝑥)).

This composition rule satisfies the following:

Associativity ℎ ◦ (𝑔 ◦ 𝑓 ) = (ℎ ◦ 𝑔) ◦ 𝑓 ,
Identity the identity map id𝑌 : 𝑌 → 𝑌 is a neutral element for composition: 𝑔 ◦ id𝑌 = 𝑔 and
id𝑌 ◦𝑓 = 𝑓 .

We require a category to come with a composition rule that satisfies the analogous properties:

Definition 10.5. A category C is a collection of objects and sets of morphisms as in Commen-

tary 10.2 together with composition rules (for every triple of objects),

◦ : Hom C(𝑐 0, 𝑐 00) × Hom C(𝑐, 𝑐 0) −→ Hom C(𝑐, 𝑐 00),

that is associative and admits an identity element.

We write 𝑓 : 𝑐 → 𝑐 0 if 𝑓 ∈ Hom C(𝑐, 𝑐 0).

Commentary 10.6. This notion is rather abstract and therefore general. In particular, it does
not stipulate that the composition rule must be actual ‘composition of maps’. The following
two examples are meant to illustrate that. However, they won’t play a role in the sequel.

90
Example 10.7. Let 𝐺 be a group. Define a category 𝐵𝐺 with a single object ∗ and morphism
set Hom𝐵𝐺 (∗, ∗) = 𝐺 where the composition is multiplication of group elements:

𝑔 ◦ 𝑔 0 := 𝑔 · 𝑔 0

Associativity follows from the associativity of group multiplication. The identity is given by
the identity element of the group. (We did not use inverses. 𝐵𝐺 remains a category if 𝐺 is a
monoid only.)

Example 10.8. Let (𝑃, ≤) be a partially ordered set. Define a category 𝑃˜ with objects the
elements of 𝑃 and morphism sets
(
∗ :𝑥 ≤𝑦
Hom𝑃˜ (𝑥, 𝑦) :=
∅ : else

I leave it as an exercise to check that there is a unique composition rule making 𝑃˜ into a
category.

Definition 10.9. A morphism 𝑓 : 𝑐 → 𝑑 in a category C is an isomorphism if there exists

𝑔 : 𝑑 → 𝑐 such that 𝑓 ◦ 𝑔 = id𝑑 and 𝑔 ◦ 𝑓 = id𝑐 . The morphism 𝑔 is unique if it exists and is
called the inverse of 𝑓 .

Example 10.10. This recovers the following notions in the examples considered above:
1. in Top: homeomorphisms
2. in Ab: isomorphisms (that is, bijective homomorphisms)
3. in Cpx: isomorphisms in each degree
4. in 𝐵𝐺: every homomorphism is an isomorphism if 𝐺 is a group
˜ (if you require the partially ordered set to be antisymmetric then) the only isomor-
5. in 𝑃:
phisms are the identities

Definition 10.11. Let C and D be categories. A functor 𝐹 : C → D associates to each object

𝑐 ∈ Ob( C) an object 𝐹 (𝑐) ∈ Ob( D), and comes with maps

𝐹 : Hom C(𝑐, 𝑐 0) → Hom D (𝐹 (𝑐), 𝐹 (𝑐 0))

that are compatible with composition and identities. In short: 𝐹 (id) = id and 𝐹 (𝑓 ◦ 𝑔) =
𝐹 (𝑓 ) ◦ 𝐹 (𝑔).

Here is an immediate consequence.

Lemma 10.12. If 𝑓 is an isomorphism in C then 𝐹 (𝑓 ) is an isomorphism in D.

Proof. Let 𝑔 be an inverse to 𝑓 . Then

𝐹 (𝑓 ) ◦ 𝐹 (𝑔) = 𝐹 (𝑓 ◦ 𝑔) = 𝐹 (id) = id

91
and
𝐹 (𝑔) ◦ 𝐹 (𝑓 ) = 𝐹 (𝑔 ◦ 𝑓 ) = 𝐹 (id) = id
so that 𝐹 (𝑔) is an inverse to 𝐹 (𝑓 ). 

Example 10.13. Here are some examples of functors we have seen during the course of the
past weeks:
1. H𝑛 : Top → Ab,
2. H𝑛 : Cpx → Ab,
3. 𝐶 • : Top → Cpx,30
4. (−) ab : Grp → Ab

Example 10.14. Let 𝑓 : 𝐺 → 𝐺 0 be a group homomorphism. There is then an induced functor

𝐵𝑓 : 𝐵𝐺 → 𝐵𝐺 0
which (necessarily) sends the unique object of 𝐵𝐺 to the unique object of 𝐵𝐺 0 and on morphism
sets is given by
𝑓 : 𝐺 = Hom𝐵𝐺 (∗, ∗) → Hom𝐵𝐺 0 (∗, ∗) = 𝐺 0 .
In fact it is clear that group homomorphisms 𝐺 → 𝐺 0 and functors 𝐵𝐺 → 𝐵𝐺 0 are in bijection.

10.2 Axioms
To state the axioms for homology theory we need to introduce a variant of the category of
topological spaces.

Definition 10.15. We denote by CW2 the category of CW pairs. Its objects are pairs of CW
complexes (𝑋, 𝑌 ) where 𝑌 ⊆ 𝑋 is a subcomplex. A morphism (𝑋, 𝑌 ) → (𝑋 0, 𝑌 0) is a continuous
map 𝑓 : 𝑋 → 𝑋 0 such that 𝑓 (𝑌 ) ⊆ 𝑌 0. These are composed in the obvious way.
We identify CW complexes 𝑋 with pairs (𝑋, ∅).

Definition 10.16. A homology theory is a collection of functors (for 𝑛 ≥ 0),

ℎ𝑛 : CW2 → Ab,
and natural maps
𝜕𝑛 : ℎ𝑛 (𝑋, 𝑌 ) → ℎ𝑛−1 (𝑌 ) := ℎ𝑛−1 (𝑌 , ∅)
satisfying the following axioms:
30 In fact, one can compose functors in a pretty obvious way.
(What does that suggest? Right, there is a category
of (small) categories with functors as morphisms.) And then the following triangle commutes by definition:
𝐶•
Top Cpx
H𝑛
H𝑛
Ab

92
Homotopy invariance If 𝑓 ' 𝑔 : 𝑋 → 𝑌 then ℎ𝑛 (𝑓 ) = ℎ𝑛 (𝑔) : ℎ𝑛 (𝑋 ) → ℎ𝑛 (𝑌 ).
Collapse 31 The map ℎ𝑛 (𝑋, 𝑌 ) → ℎ𝑛 (𝑋 /𝑌 , 𝑌 /𝑌 ) is an isomorphism.
Exactness The sequence
𝜕𝑛+1 𝜕𝑛
· · · → ℎ𝑛+1 (𝑋, 𝑌 ) −−−→ ℎ𝑛 (𝑌 ) → ℎ𝑛 (𝑋 ) → ℎ𝑛 (𝑋, 𝑌 ) −−→ ℎ𝑛−1 (𝑌 ) → · · ·

is exact.
Additivity For any collection (𝑋𝛼 , 𝑌𝛼 ), the map ⊕𝛼 ℎ𝑛 (𝑋𝛼 , 𝑌𝛼 ) → ℎ𝑛 (q𝛼 (𝑋𝛼 , 𝑌𝛼 )) is an iso-
morphism.
Dimension ℎ𝑛 (∗) = 𝛿𝑛0 Z

Proposition 10.17. Singular homology defines a homology theory.

Proof. Homotopy invariance is Theorem 4.2, Exactness is Corollary 5.18, Collapse is Propo-
sition 5.28, Additivity is Proposition 3.14, and Dimension is Example 3.9. 

Remark 10.18. Here, naturality of the connecting homomorphism 𝜕𝑛 means that for every
𝑓 : (𝑋, 𝑌 ) → (𝑋 0, 𝑌 0) of CW pairs the following square commutes:

𝜕𝑛
ℎ𝑛 (𝑋, 𝑌 ) ℎ𝑛−1 (𝑌 )
ℎ𝑛 (𝑓 ) ℎ𝑛−1 (𝑓 )
𝜕𝑛
ℎ𝑛 (𝑋 0, 𝑌 0) ℎ𝑛−1 (𝑌 0)

For singular homology this follows from the construction of the connecting homomorphism
in Proposition 4.37.
In categorical language, this says that 𝜕𝑛 is a natural transformation (of functors CW2 → Ab):

Definition 10.19. A natural transformation 𝜙 : 𝐹 → 𝐺 between two functors 𝐹, 𝐺 : C → D is a

collection of maps 𝜙𝑐 : 𝐹 (𝑐) → 𝐺 (𝑐) in D such that the following square commutes for every
𝑓 : 𝑐 → 𝑐 0:
𝜙𝑐
𝐹 (𝑐) 𝐺 (𝑐)
𝐹 (𝑓 ) 𝐺 (𝑓 )
𝜙𝑐 0
𝐹 (𝑐 0) 𝐺 (𝑐 0)

For the rest of this section we fix a homology theory ℎ𝑛 . Let us see what we can com-
pute just from the axioms. We define, as before, the reduced homology group as ℎ˜𝑛 (𝑋 ) :=
ker (ℎ𝑛 (𝑋 ) → ℎ𝑛 (∗)).

31 This
is non-standard terminology. Presumably, others would call this Excision instead. The two results
(Theorem 5.23 and Proposition 5.28) are certainly related but I would like to distinguish them.

93
Example 10.20. 1. For each triple 𝑋 ⊇ 𝑌 ⊇ 𝑍 , the sequence32
𝜕𝑛+1 𝜕𝑛
· · · → ℎ𝑛+1 (𝑋, 𝑌 ) −−−→ ℎ𝑛 (𝑌 , 𝑍 ) → ℎ𝑛 (𝑋, 𝑍 ) → ℎ𝑛 (𝑋, 𝑌 ) −−→ ℎ𝑛−1 (𝑌 , 𝑍 ) → · · ·

is exact. This is a diagram chase, starting with Exactness. Cf. Exercise 4.5.
2. If 𝑋 is a contractible space then ℎ˜𝑛 (𝑋 ) = 0 for all 𝑛. Indeed, the map ℎ𝑛 (𝑋 ) → ℎ𝑛 (∗) is
an isomorphism, by Homotopy Invariance, cf. Corollary 4.3.
3. Consider the exact sequence above for the triple (𝐷 𝑘 , 𝜕𝐷 𝑘 , ∗):

· · · → ℎ˜𝑛 (𝐷 𝑘 ) → ℎ𝑛 (𝐷 𝑘 , 𝑆 𝑘−1 ) → ℎ˜𝑛−1 (𝑆 𝑘−1 ) → ℎ˜𝑛−1 (𝐷 𝑘 ) → · · ·

As the two outer terms vanish we have an isomorphism in the middle. And by Collapse,
Additivity and Dimension we get

ℎ˜𝑛 (𝑆 𝑘 ) ← → ℎ˜𝑛−1 (𝑆 𝑘−1 ) ← → ℎ˜𝑛−𝑘 (𝑆 0 ) = 𝛿𝑛𝑘 Z.

∼ ∼ ∼ ∼
− ℎ𝑛 (𝐷 𝑘 , 𝑆 𝑘−1 ) − − ··· −

4. We have by Collapse

→ ℎ˜𝑛 (𝑋 𝑘 /𝑋 𝑘−1 ) −
→ ℎ˜𝑛 (∨𝛼 𝑆𝛼𝑘 ) ←
− ⊕𝛼 ℎ˜𝑛 (𝑆𝛼𝑘 ) = ⊕𝛼 𝛿𝑛𝑘 Z
∼ ∼ ∼
ℎ𝑛 (𝑋 𝑘 , 𝑋 𝑘−1 ) −

where the wrong-way isomorphism is deduced from Additivity and Collapse, as in

Lemma 7.40. Here, the 𝛼 range over the 𝑘-cells in 𝑋 .

In fact, everything we computed (in terms of homology) in this course can be computed
just from the axioms:

Theorem 10.21. If (ℎ𝑛 , 𝜕𝑛 ) is a homology theory then there are natural isomorphisms

ℎ𝑛 (𝑋, 𝑌 )  𝐻𝑛 (𝑋, 𝑌 ).

Sketch of proof. Example 10.20 allows us to define a ‘cellular chain complex’

𝐶𝑛CW,ℎ (𝑋 ) := ℎ𝑛 (𝑋 𝑛 , 𝑋 𝑛−1 )  ⊕𝛼 Z[𝐷𝛼𝑛 ]

with differentials exactly as for singular homology. To deduce that the homology of this
chain complex is our old cellular homology, we need to identify the differentials. If you
recall the cellular boundary formula Proposition 8.30, you see that it suffices to show that
deg (𝑓 ) = deg(𝑓 ) for maps 𝑓 : 𝑆 𝑘 → 𝑆 𝑘 , 𝑘 > 0. You can prove the local degree formula
ℎ

Proposition 6.21 for the theory ℎ and deduce that this is true at least for all 𝑓 we constructed
in Proposition 6.6. The result for general 𝑓 then follows from the following non-trivial fact (see
Hatcher, Corollary 4.25, and also section 10.3 below): The degree map induces an isomorphism
∼
deg : [𝑆 𝑘 , 𝑆 𝑘 ] −
→Z
32 The connecting homomorphisms are the composites ℎ𝑛+1 (𝑋, 𝑌 ) → ℎ𝑛 (𝑌, ∅) → ℎ𝑛 (𝑌, 𝑍 ).

94
between homotopy classes of maps 𝑆 𝑘 → 𝑆 𝑘 and the integers.33 That is, up to homotopy, we
have already constructed all such self-maps in Proposition 6.6.
At this point we have proved ℎ𝑛 (𝑋 )  ℎ𝑛CW (𝑋 )  H𝑛CW (𝑋 )  H𝑛 (𝑋 ), by Corollary 8.25.
We deduce that ℎ𝑛 (𝑋, 𝑌 )  ℎ˜𝑛 (𝑋 /𝑌 )  𝐻˜𝑛 (𝑋 /𝑌 )  H𝑛 (𝑋, 𝑌 ) which completes the sketch of
the proof. 

Remark 10.22. One can make the natural isomorphism in the theorem commute with the
‘connecting homomorphisms’ 𝜕𝑛 . It follows that in a suitable category of homology theories,
all objects are isomorphic.

Example 10.23. If ℎ𝑛 is a homology theory (for example, singular homology ℎ𝑛 = 𝐻𝑛 ) then so

is ℎ𝑛0 where ℎ𝑛0 (𝑋, 𝑌 ) = ℎ𝑛 (𝑋, 𝑌 ) and 𝜕𝑛0 = −𝜕𝑛 . By the previous remark, these two homology
theories are isomorphic nonetheless. An explicit isomorphism is given by (−1)𝑛 id : ℎ𝑛 (𝑋, 𝑌 ) →
ℎ𝑛0 (𝑋, 𝑌 ).

10.3 Generalized (co)homology theories

Perhaps you find Theorem 10.21 satisfying (I do), or perhaps you find it rather disappointing.
Indeed, wasn’t the goal of algebraic topology to distinguish spaces, as discussed in Section 1.1?
And if two spaces can’t be distinguished using singular homology then—Theorem 10.21 tells
us—they can’t be distinguished by any homology theory.
But not so fast! Looking at the history of ‘homology theory’ since Theorem 10.21 was
established, what seems to have happened is the following. Rather than giving up on the
subject, mathematicians started weakening and modifying the axioms and studying theories
that satisfied these new sets of axioms. I will touch upon 2 ½ modifications below, very briefly.
These can also be read as suggestions of where to go next if you would like to study more
algebraic topology.

Coefficients Actually even before the axiomatization, mathematicians had considered ho-
mology that spit out abelian groups other than the integers on a point. Here is a precise
definition.

Definition 10.24. Let 𝐴 be an abelian group. A homology theory with coefficients in 𝐴 is a

collection of functors ℎ𝑛 : CW2 → Ab together with natural maps 𝜕𝑛 satisfying the same axioms
as in Definition 10.16 except that the dimension axiom is replaced by ℎ𝑛 (∗) = 𝛿𝑛0𝐴.

Remark 10.25. To obtain such a theory replace the singular chain complex 𝐶 • (𝑋 ) by the
tensor product 𝐶 • (𝑋 ) ⊗Z 𝐴. That is, in degree 𝑛 you consider the group
Ê
𝐶𝑛 (𝑋 ; 𝐴) := 𝐴[𝜎]
𝜎 : Δ𝑛 →𝑋

33 The addition on the left-hand side takes two (pointed) maps 𝑓 , 𝑔 : 𝑆 𝑘 → 𝑆 𝑘 to the composite
𝑓 ∨𝑔
𝑆 𝑘  𝑆 𝑘 ∨ 𝑆 𝑘 −−−→ 𝑆 𝑘

where the first map collapses the equator (containing the base point). See Section 10.3 below for more details.

95
with the ‘same’ differential as before. Then we set H𝑛 (𝑋 ; 𝐴) := H𝑛 (𝐶 • (𝑋 ; 𝐴)) and similarly
for relative homology. This singular homology with coefficients in 𝐴 satisfies the axioms as in the
previous definition.

A particularly useful group to consider is 𝐴 = Z/2Z.

Example 10.26. The cellular chain complex for RP𝑘 with Z/2-coefficients looks as follows
(compare with Example 8.36):
0 0 0
0 → Z/2 →
− Z/2 → · · · → Z/2 → − Z/2 → 0,
− Z/2 →
so that (
Z/2 : 0 ≤ 𝑛 ≤ 𝑘
H𝑛 (RP𝑘 ; Z/2) =
0 :𝑛 >𝑘
and
H𝑛 (RP∞ ; Z/2) = Z/2
for all 𝑛 ≥ 0.

Here is an application related to Exercise 5.8.

∼
Proposition 10.27. Let 𝑓 : 𝑆 𝑘 → 𝑆 𝑘 be an odd map. Then 𝑓∗ : Z/2 = H𝑘 (𝑆 𝑘 ; Z/2) −
→ H𝑘 (𝑆 𝑘 ; Z/2) =
Z/2 is an isomorphism.

Proof. See Hatcher, Proposition 2B.6, which really uses Z/2-coefficients in an essential way.


Corollary 10.28 (Borsuk-Ulam). For any map 𝑓 : 𝑆 𝑘 → R𝑘 there is 𝑥 ∈ 𝑆 𝑘 such that 𝑓 (−𝑥) =
𝑓 (𝑥).

Proof. See Exercise 5.8. 

Commentary 10.29. This implies that at any point in time there are two antipodal points on
earth with the exact same temperature and wind speed.

On the other hand, in terms of distinguishing more spaces, this modification of homology
theory is of not much help:

Proposition 10.30. If H𝑛 (𝑋 )  H𝑛 (𝑌 ) for all 𝑛 then ℎ𝑛 (𝑋 )  ℎ𝑛 (𝑌 ) for all homology theories

with coefficients in 𝐴.

Proof. There is a similar uniqueness theorem for homology theories with coefficients in 𝐴 as
Theorem 10.21, see Hatcher, Theorem 4.59. That is, every such homology theory is actually
isomorphic to H• (−; 𝐴), the one constructed just above. And for this theory one has the uni-
versal coefficient theorem (Hatcher, Corollary 3A.4) which expresses it in terms of the ordinary
singular homology:
H𝑛 (𝑋 ; 𝐴)  (H𝑛 (𝑋 ) ⊗ 𝐴) ⊕ Tor(H𝑛−1 (𝑋 ), 𝐴)

96
where the Tor-groups are some purely algebraic invariant of abelian groups. 

Dimension axiom Here is a more drastic departure from Definition 10.16.

Definition 10.31. A generalized homology theory is a collection of functors ℎ𝑛 : CW2 → Ab

together with natural maps 𝜕𝑛 satisfying the same axioms as in Definition 10.16 except for
(possibly) the dimension axiom.

Commentary 10.32. Several theories that satisfied only these weaker axioms were found in
the following decades, arguably most prominently bordism and stable homotopy. A lot of work
in algebraic topology since then has gone into investigating these and other examples, but
many questions remain open!
However, one thing that is clear is that no uniqueness theorem in the form of Theo-
rem 10.21 holds for generalized homology theories. The two examples just mentioned (and
many others) are genuinely distinct.

My goal is to define stable homotopy for you, and give you an example of a question about
them that is still open.

Definition 10.33. The homotopy groups for 𝑛 > 0 of a pointed space are homotopy classes
of pointed maps,
𝜋𝑛 (𝑋, 𝑥) = [𝑆 𝑛 , 𝑋 ].
The group law is defined by the following diagram.

Here, the first map is the ‘pinching map’, sending the equator to the common base point.
(The base point of the sphere on the left is assumed to lie on the equator so this pinching map
becomes a map of pointed spaces.)

Remark 10.34. The group 𝜋𝑛 (𝑋, 𝑥) turns out to be abelian if 𝑛 ≥ 2. To see this, imagine
thickening the equator in the sphere on the left, thereby pinching a whole strip in the middle.
Then you can swap the remaining caps around the north and south pole continuously without
them meeting the base point (which, remember we placed somewhere on the equator). Finally
you shrink the strip back to the equator. This shows that 𝑓 · 𝑔 ' 𝑔 · 𝑓 .

Exercise 10.35. Why does the previous argument not show that 𝜋 1 (𝑋, 𝑥) is abelian? In other
words, what goes wrong on the 1-sphere?

Commentary 10.36. Of course, we know that homotopy equivalences or homeomorphisms

induce isomorphisms in homology, and we saw very few instances (surfaces, mainly) where

97
the converse holds. The following gives an idea of how powerful homotopy groups are. Note
that it applies to any CW complex whatsoever.

Theorem 10.37 (Whitehead’s Theorem). Let 𝑓 : 𝑋 → 𝑌 be a continuous map between CW

complexes. If it induces a bijection on connected components and
∼
𝑓∗ : 𝜋𝑛 (𝑋, 𝑥) −
→ 𝜋𝑛 (𝑌 , 𝑓 (𝑥))
for all 𝑥 ∈ 𝑋 and all 𝑛 then 𝑓 is a homotopy equivalence.

Proof. See Hatcher, Theorem 4.5. 

Given this power, it is not surprising that computing these homotopy groups is very dif-
ficult in general. As mentioned in Example 1.10, the homotopy groups of spheres are mostly
unknown.

Example 10.38 (Hopf fibration). Consider the map 𝑆 3 → C ∪ {∞},

𝑥 1 + 𝑖𝑥 2
(𝑥 1, 𝑥 2, 𝑥 3, 𝑥 4 ) ↦→
𝑥 3 + 𝑖𝑥 4
which maps to ∞ when the denominator vanishes. We may identify the one-point compact-
ification C ∪ {∞} with the 2-sphere, in which case this map 𝜂 : 𝑆 3 → 𝑆 2 is known as the Hopf
fibration (due to Heinz Hopf (1931)). Over each point in 𝑆 2 the fiber of 𝜂 consists of a circle 𝑆 1 ,
and for any pair of distinct points in 𝑆 2 these circles pass through each other precisely once!
(See the wikipedia page for some illustrations, or even better, watch this video.)
What is the relation with homotopy groups? One might think that just as 𝜋𝑛 (𝑆 𝑘 ) = 0 for
𝑛 < 𝑘, one also has 𝜋𝑛 (𝑆 𝑘 ) = 0 for 𝑛 > 𝑘. (On the ‘diagonal’ 𝜋𝑘 (𝑆 𝑘 ) = Z one always has a copy
of the integers, see below.) Hopf put that idea to rest. He used the fact that the linking number
between any two distinct fibers is 1 to prove that the Hopf fibration is not null-homotopic. In
fact, 𝜋 3 (𝑆 2 )  Z[𝜂] generated by the class of the Hopf fibration.

However, there is a sort of stability phenomenon happening that was discovered by Hans
Freudenthal (1936).

Theorem 10.39 (Freudenthal Suspension Theorem). Let 𝑋 be a pointed CW complex. The

suspension map
𝜋𝑛+𝑘 (Σ𝑘 𝑋 ) = [𝑆 𝑛+𝑘 , Σ𝑘 𝑋 ] → [Σ𝑆 𝑛+𝑘 , Σ𝑘+1𝑋 ] = [𝑆 𝑛+𝑘+1, Σ𝑘+1𝑋 ] = 𝜋𝑛+𝑘+1 (Σ𝑘+1𝑋 )
is an isomorphism for 𝑘  𝑛.

Proof. See Hatcher, Corollary 4.24. 

Example 10.40. In fact, there are precise bounds on 𝑘, 𝑛 (depending on 𝑋 ) for this map to
become an isomorphism. For example it implies rather easily that all maps in the sequence
∼ ∼
Z  𝜋 1 (𝑆 1 ) −
→ 𝜋 2 (𝑆 2 ) −
→ 𝜋 3 (𝑆 3 ) → · · · ,
are isomorphisms.

98
Example 10.41. Again by the Freudenthal suspension theorem, there is a sequence
∼ ∼
Z[𝜂] = 𝜋 3 (𝑆 2 )  𝜋4 (𝑆 3 ) −
→ 𝜋 5 (𝑆 4 ) −
→ ···

It turns out that the first map is not an isomorphism. Instead it identifies 𝜋4 (𝑆 3 ) with Z[𝜂]/2[𝜂].

The theorem says that the group 𝜋𝑛+𝑘 (Σ𝑘 𝑋 ) is independent of 𝑘 for large 𝑘. Therefore we
can define:

Definition 10.42. Let 𝑋 be a pointed CW complex. The stable homotopy groups are

𝜋𝑛𝑠 (𝑋 ) = lim 𝜋𝑛+𝑘 (Σ𝑘 𝑋 ).

−−→
𝑘→∞

Note that these make sense even for 𝑛 < 0, and by Remark 10.34, they are all abelian groups.

So, these groups record only stable phenomena at the level of homotopy groups. For this
reason, they are easier to compute than the original ‘unstable’ homotopy groups, although still
(too) difficult.

Example 10.43. 1. From Example 10.40 we deduce that 𝜋 0𝑠 (∗) = Z.

2. It is not too hard to see that 𝜋𝑛𝑠 (∗) = 0 if 𝑛 < 0.
3. It is also known (Serre (1953)) that 𝜋𝑛𝑠 (∗) is a finite group for all 𝑛 > 0. For example,
𝜋1𝑠 (∗) = Z/2 generated by the class of 𝜂, as seen in Example 10.41.
4. Except for small-ish values of 𝑛 > 0 (in the region of two to three digits), the groups
𝜋𝑛𝑠 (∗) are largely unknown. Nevertheless it is known (at least since Adams (1965)) that
infinitely many are non-zero. You can find some intriguing information about these in
Hatcher, p. 384ff.

Commentary 10.44. The stable homotopy groups of spheres are linked, among many other
things, with the geometry of higher-dimensional manifolds. Because the latter is hard, it is
not too surprising that the former is hard as well. Just as one example, computations of 𝜋𝑛𝑠 (∗)
allowed mathematicians to determine the number of smooth structures on spheres.

Proposition 10.45. Stable homotopy groups 𝜋𝑛𝑠 define a generalized homology theory.

Proof. Hatcher, Proposition 4F.1. 

Remark 10.46. Note that this theory fails the dimension axiom in an extreme way: The values
𝜋𝑛𝑠 (∗) are non-zero for infinitely many 𝑛 and they are not even known!

Exercise 10.47. You could say: Okay, the groups 𝜋𝑛𝑠 (∗) are hard to compute but assuming
these, would it then be possible to compute 𝜋𝑛𝑠 (𝑋 ) for any CW complex 𝑋 ? An obvious idea
would be to try to repeat the proof of Theorem 10.21, and to construct a cellular chain complex
with terms given by stable homotopy groups by spheres. What goes wrong?
Hint: Observe that 𝜋•𝑠 (𝑆 𝑛 ) = 𝜋 •−𝑛
𝑠 (∗).

99
Cohomology Another modification to the axioms of a homology theory one can make is
to ‘dualize’ everything. This leads to the notion of (ordinary and generalized) cohomology
theory.
Thus, a cohomology theory is a collection of contravariant functors ℎ𝑛 : CW2 → Ab and
natural transformations ℎ𝑛−1 (𝑌 ) → ℎ𝑛 (𝑋, 𝑌 ) satisfying axioms dual to the ones for a homology
theory.34 We will not make this precise here but see Hatcher, p. 202. Instead, it turns out that
the ‘ordinary’ cohomology theories are again uniquely determined by their value on a point,
so there is essentially one example which we will discuss now.

Definition 10.48. The singular cochain complex of a space 𝑋 has terms the abelian groups

𝐶𝑛∗ (𝑋 ) := Hom(𝐶 −𝑛 (𝑋 ), Z)

and differential given by precomposition with the differential in the singular chain complex.
Note that it has non-zero terms in non-positive degrees only.
The singular cohomology of 𝑋 is the collection of groups

H𝑛 (𝑋 ) := H−𝑛 (𝐶 •∗ (𝑋 ))

Remark 10.49. This dualization process actually has some positive side-effects, making co-
homology far from a trivial modification of homology. The starting point is the observation
that there are maps
Δ∗
∪ : H𝑚 (𝑋 ) × H𝑛 (𝑋 ) → H𝑚+𝑛 (𝑋 × 𝑋 ) −−→ H𝑚+𝑛 (𝑋 ),

where the first map can be thought of as taking an ‘𝑚-cell 𝑒𝑚 and an 𝑛-cell 𝑒𝑛 ’ in 𝑋 and
producing the ‘(𝑚 + 𝑛)-cell 𝑒𝑚 × 𝑒𝑛 in 𝑋 × 𝑋 . This can be made precise using the cellular
cochain complex (for CW complexes). The second map is induced by the diagonal embedding
Δ : 𝑋 → 𝑋 × 𝑋 (recall that maps in cohomology go in the opposite direction!).
This is called the cup product and it makes H∗ (𝑋 ) := ⊕𝑛 H𝑛 (𝑋 ) into a (graded) ring, a
structure that is not visible at the level of homology. The cup product admits a more geometric
interpretation in good cases. For example, on surfaces 𝑆, the product H1 (𝑆) × H1 (𝑆) → H2 (𝑆)
can be determined by computing intersections (after deformation) of their fundamental loops
(cf. Section 7.3).

Example 10.50. For 𝑘 > 0, the cohomology ring for the sphere is H∗ (𝑆 𝑘 ) = Z[𝑡]/𝑡 2 (with 𝑡
in degree 𝑘).

Example 10.51. One also has cohomology with coefficients (by taking Hom(𝐶 −𝑛 (𝑋 ), 𝐴) in-
stead of Hom(𝐶 −𝑛 , Z)). As rings,

H∗ (RP∞ ; Z/2) = Z/2[𝑡]

(with 𝑡 in degree 1).

34 A functor 𝐹 : C → D is contravariant if it induces morphisms in the opposite direction, that is, comes with
maps
Hom C(𝑐, 𝑐 0 ) → Hom D (𝐹 (𝑐 0 ), 𝐹 (𝑐)),
still assumed compatible with composition and identities.

100
Example 10.52. Recall from Exercise 5.10 the space 𝑋 = 𝑆 1 ∨ 𝑆 1 ∨ 𝑆 2 whose homology groups
are isomorphic to the ones of the torus T . It is ‘clear’ that these spaces are not homeomorphic
(nor homotopy equivalent), and you proved this in that exercise. Here is a way to see this
using cohomology. In both cases, one has isomorphic cohomology groups and they coincide
with the homology groups:

H0 = Z, H1 = Z2, H2 = Z,

with all other groups zero. The cup product of the two generators 𝛼, 𝛽 in degree 1, however,
is non-zero for the torus, and zero for 𝑋 . Hence X ; T .
That 𝛼 ∪ 𝛽 ≠ 0 in T can be explained by the fact that however you deform the two circles

they will continue to intersect.

That 𝛼 ∪ 𝛽 = 0 in 𝑋 can be explained by the fact that the cohomology ring of a wedge
sum is a subring of the product:

H∗ (𝑆 1 ∨ 𝑆 1 ∨ 𝑆 2 ) ⊆ H∗ (𝑆 1 ∨ 𝑆 1 ) × H∗ (𝑆 2 )

and since 𝛼 ∪ 𝛽 ∈ H2 (𝑆 1 ∨ 𝑆 1 ) = 0.

Remark 10.53. There is also a notion of generalized cohomology theory, obtained by dropping
the dimension axiom. A famous such theory is topological K-theory. It was used by Adams and
Atiyah (1966) to give an elegant proof of the Hopf invariant one problem to which we now
turn.

Definition 10.54. A real division algebra is a finite-dimensional real vector space A together
with an operation · : A2 → A such that for all 𝑎 ∈ A, the maps

A→A
𝑥 ↦→ 𝑎 · 𝑥
𝑥 ↦→ 𝑥 · 𝑎

are R-linear, and invertible if 𝑎 ≠ 0. (No assumptions are made on associativity, identity, or
commutativity. Nor need these be normed algebras.)

Example 10.55. The classical examples are R itself, the complex numbers C, the quaternions H,
and the octonions O. The first two are commutative, associative and have an identity. The
third is associative with identity but is not commutative. The last is made of pairs of quater-
nions with multiplication given by
¯ 𝑑𝑎 + 𝑏𝑐).
(𝑎, 𝑏) (𝑐, 𝑑) = (𝑎𝑐 − 𝑑𝑏, ¯

101
It is not even associative (but has an identity).
Another example is the complex numbers with the non-standard multiplication 𝑧𝑤 = 𝑧𝑤.
It has no identity.
Note that these division algebras have real dimension 1,2,4,8.

Remark 10.56 (Hopf invariant). Let 𝜙 : 𝑆 2𝑛−1 → 𝑆 𝑛 be a continuous map for 𝑛 > 1. We may
use this to construct a CW complex 𝐶𝜙 with cells 𝑒 0, 𝑒𝑛 , 𝑒 2𝑛 , with 𝜙 the attaching map for the
2𝑛-cell. By cellular cohomology, we have as only non-trivial cohomology groups

H0 = Z[𝑒 0 ], H𝑛 = Z[𝑒𝑛 ], H2𝑛 = Z[𝑒 2𝑛 ].

The cohomology ring is completely determined by the cup product of a generator [𝑒𝑛 ] ∈ H1
with itself. We have
[𝑒𝑛 ] ∪ [𝑒𝑛 ] = ℎ(𝜙) · [𝑒 2𝑛 ]
for some integer ℎ(𝜙) ∈ Z, defined up to sign. This is called the Hopf invariant of 𝜙. (With
more care, it can be defined without the sign indeterminacy.)

Example 10.57. For 𝑛 = 2, we have ℎ(𝜂) = 1 which is related to the fact that the knots in
distinct fibers have linking number 1. There are analogues of the Hopf fibration in higher
dimensions,

𝜂 4 : 𝑆 7 → 𝑆 4 (with fibers 𝑆 3 ), 𝜂 8 : 𝑆 15 → 𝑆 8 (with fibers 𝑆 7 ),

also with Hopf invariant one.

In fact, 𝑋𝜂 = CP2 , 𝑋𝜂4 = HP2 , 𝑋𝜂8 = OP2 and in each case the cohomology ring is Z[𝑡]/𝑡 3
with 𝑡 in degree 2, 4, 8. (For 𝑛 = 1 one could take Z/2-coefficients. The canonical 2:1-cover
𝜂 1 : 𝑆 1 → 𝑆 1 produces 𝑋𝜂1 = RP2 and H∗ (RP2 ; Z/2) = Z/2[𝑡]/𝑡 3 .)

Remark 10.58 (Hopf invariant one problem). The Hopf invariant problem asks: are there
maps other than 𝜂 = 𝜂 2 , 𝜂 4 , and 𝜂 8 with Hopf invariant one?

Theorem 10.59 (Adams (1960); Adams-Atiyah (1966)). Maps with Hopf invariant one only occur
for 𝑛 = 2, 4, 8 (and 𝑛 = 1 if you wish).

The first proof which is quite long uses operations in cohomology (an ordinary cohomol-
ogy theory), the second is shorter and uses K-theory (a generalized cohomology theory).

Corollary 10.60. There are no real division algebras except in dimensions 1, 2, 4, 8.

Remark 10.61. The connection between Hopf fibrations and division algebras is indicated by
the following observation. Let A be one of the four classical real division algebras of Exam-
ple 10.55 of dimension 𝑛 = 1, 2, 4, 8. Then there is a map

S(A2 ) → AP1

from the elements in A2 of unit norm to the projective line over A. It sends an element 𝑎 of
unit norm to the line through the origin and 𝑎.

102
 Now, the left-hand side identifies with the sphere 𝑆 2𝑛−1 , while the right-hand side identifies
with 𝑆 𝑛 so this is a continuous map 𝑆 2𝑛−1 → 𝑆 𝑛 . Guess which one? Right, it’s one of the Hopf
fibrations from Example 10.57.

103