1

A.B.Sossinsky

TOPOLOGY-2

Math in Moscow Lecture Notes

Fall 2009

MCCME Publishers, Moscow 2009

2

TABLE OF CONTENTS

Foreword
Lecture 1. Homology functors
Lecture 2. CW-complexes
Lecture 3. Homotopy groups
Lecture 4. Cellular homology
Lecture 5. Simplicial homology
Lecture 6. Properties of simplicial homology
Lecture 7. Singular homology
Lecture 8. Applications of homology
Lecture 9. Cohomology
Lecture 10. Poincaré duality
Lecture 11. Obstruction theory
Lecture 12. Vector bundles and G-bundles
Lecture 13. Miscellany
 3

FOREWORD

The present booklet is based on lectures given in the fall semester of 2009
to second year IUM students in Russian and to Math in Moscow students
in English. I prepared and distributed handouts, typeset in LaTeX, at the
end of each lecture (the handouts, written in English, were given to the
American students and to the Russian ones as well). These handouts, with
slight revisions, are gathered in this small book.
The Topology-2 course at the IUM (and in the framework of the Math in
Moscow program) is traditionally an introductory course in algebraic topol-
ogy, mainly about homology theory. The students taking it have already had
Topology-1, which at the IUM is an elementary introduction to topology with
emphasis on its geometric and algebraic aspects. Topology-1 includes a min-
imal amount of general topology (topological spaces and continuous maps,
topological equivalence, compactness, connectedness, separability) and such
geometric topics as plane curves, surfaces, vector fields, covering spaces, ex-
amples of 3-manifolds, homotopy, treated by means of the corresponding
basic algebraic invariants (degree of circle maps, Whitney winding number,
Euler characteristic, Poincaré index, fundamental group, Morse index). For
more details about that course, the reader is referred to the book Topology-1
in the same series written jointly by V.V.Prasolov and myself.
As to the Topology-2 course, the titles of the 13 lectures will quickly give
an idea of its contents. As usual, I regard the ability of solving the problems
(appearing at the end of each lecture and used in the exercise class as well as
for homework assignments) as at least as important as mastering the theory.
In preparing the lectures, my main sources of information (besides my own
memories of the subject) were the wonderful books Course in Homotopic
Topology by A.T.Fomenko and D.B.Fuchs, V.V.Prasolov’s Homology Theory,
and S.V.Matveev’s Lectures on Algebraic Topology.
Teaching the Topology-2 course in the fall and winter of 2009 was a very
satisfying experience: the IUM class of some twenty students was very re-
ceptive and attentive, they corrected errors in the lectures and the handouts,
asked very pointed questions, and politely complained whenever the proofs
were only sketched. I am especially grateful to Vladimir Eisenshtadt and
Daniel Le for numerous useful remarks and to Alexey Deinega (who helped
prepare the first version of the illustrations). I would also like to thank Victor
Shuvalov for the final version of the illustrations and ??? , for rapidly and
efficiently editing this booklet.
4

Lecture 1
HOMOLOGY FUNCTORS

In this introductory lecture, we learn what homology theories (the main

protagonists of this course) are and how they work in topology. Here we
take their properties for granted (without actually constructing any of these
theories) and use them to give simple proofs of some deep topological facts,
thus motivating the study of homology.

1.1. Categories and functors

By definition, a category C is a pair consisting of a class Obj(C) of objects
and, for any ordered pair of objects (X, Y ), a set hom(X, Y ) of morphisms
with domain X and range Y ; if f ∈ hom(X, Y ), we write f : X → Y or
f
X −→ Y ; for every ordered triple of objects joined by two morphisms f, g,
f g
X −→ Y −→ Z, a morphism, called the composite of f and g and denoted
by gf (or g ◦ f ) is given; the objects and morphisms satisfy the two following
axioms:
f g h
Associativity. If X −→ Y −→ Z −→ W , then

h(gf ) = (hg)f : X → W.

Identity. For every object Y there is a morphism idY : Y → Y such that

if f : X → Y , then idY f = f , and if h : Y → Z, then h idY = h.
If the morphisms f : X → Y and g : Y → X satisfy gf = idX and
f g = idY , we say that f and g are inverse to each other, f (and g) are
isomorphisms, and X and Y are isomorphic.
In the most common examples of categories, the objects are sets with
an additional structure on them while the morphisms are maps preserving
this structure (in some sense). In this course, we will be using the following
examples of categories:
• T op: topological spaces and their continuous maps;
• Sim: simplicial spaces and their simplicial maps;
• CW: cell spaces (also called CW-complexes) and cellular maps;
• AGr: Abelian groups and their homomorphisms;
• GAG: graded Abelian groups and grading-preserving homomorphisms;
 5

• Mod: modules over commutative rings and their homomorphisms;

• Vect: vector spaces and linear operators;
• Fld: fields and their homomorphisms;
• Set: sets and their maps.
• Dir: sets supplied with a partial order ≤ such that hom(X, Y ) consists
of the pair (X, Y ) if X ≤ Y and is the empty set otherwise.
A covariant functor Φ from the category C to the category D assigns
to every object X from C an object Φ(X) from D and to each morphism
f : X → Y from hom(C), a morphism Φ(f ) : Φ(X) → Φ(Y ) from hom(D)
so that identical morphisms and composition are preserved, i.e.,
(i) Φ(idX ) =idΦ(X) , (ii) Φ(gf ) = Φ(g)Φ(f ).
The two properties (i) and (ii) are often referred to as functoriality. Note
that quite often the symbol Φ used here to denote both the assignment of
objects and the assignment of morphisms is replaced by two different symbols
for the different types of assignments.
A contravariant functor Ψ from the category C to the category D is defined
similarly, except that “the arrows are reversed”, i.e., Ψ(f ) : Ψ(Y ) → Ψ(X)
and Ψ(gf ) = Ψ(f )Ψ(g).
Here are some examples of functors.
• The forgetful functor from the category T op to the category Set which
assigns to every topological space its set of points (thus “forgetting” its topo-
logical structure) and to each continuous map the map itself (“forgetting”
about its continuity).
• The contravariant functor from the category of compact Hausdorff
spaces and their continuous maps to the category of Banach spaces and their
continuous homomorphisms that assigns to each space X the (Banach) space
of continuous real-valued functions on X.
• The covariant functor H0 from T op to AGr such that H0 (X) is the
free Abelian group generated by the set of connected components of X, and
if f : X → Y is a continuous map, then H0 (f ) : H0 (X) → H0 (Y ) is the
homomorphism such that if C is a connected component of X and C 0 is the
component of Y containing f (C), then H0 (f )(C) = C 0 .
• A direct system on a category C is a covariant functor from the category
of directed sets Dir to the category C. (This functor allows to define direct
limits in an arbitrary category.)
6

The main protagonists of this course are homology theories, which are
covariant functors from certain topological categories to certain algebraic
ones, and we discuss one of them in a section of its own.

1.2. Homology theory as a functor

Homology (more precisely, singular homology) is a covariant functor from
the category T op of topological spaces and their continuous maps to the cate-
gory GAG of graded Abelian groups and grading-preserving homomorphisms.
This means that to each topological space X and each positive integer n ∈ N
an Abelian group (denoted Hn (X)) is assigned and for each continuous map
f : X → Y , a homomorphism f∗ : Hn (X) → Hn (Y ) is given.
The singular homology functor will be constructed (and its main prop-
erties established) in Lecture 7. At this stage, the only properties that we
need, besides functoriality, i.e.,

(idX ) = idHn (X) and (f ◦ g)∗ = f∗ ◦ g∗

are the following isomorphisms

Hn (Sn ) ∼
= Z, Hn (Dn ) ∼
= 0 for any n ≥ 1,

where Sn and Dn are the n-dimensional sphere and disk, respectively.

1.3. Diagram chasing and some basic problems of topology

One of the main notions of category theory (quite useful in other branches
of mathematics as well) is the notion of commutative diagram. For example,
a commutative square is a diagram of the form
A −−−→ B
f
 

αy

βy

C −−−→ D
g

such that β ◦ f = g ◦ α. More generally, a commutative diagram is a con-

figuration of objects (e.g. groups) and arrows (morphisms) such that any
two paths along arrows starting at the same place and ending at the same
place have the same composition. Establishing the commutativity of a dia-
gram and using it to derive certain properties of the morphisms involved is a
 7

mathematical sport called diagram chasing that we shall be practicing a bit

in the exercise classes of this course.
The simplest commutative diagram is the commutative triangle, and it
helps visualize some of the basic problems of topology, namely the exten-
sion problem (and its particular case, the retraction problem) and the lifting
problem (and its particular case, the section problem). The extension prob-
lem is the following: given the (continuous) map f : A → B of a subset
A ⊂ X of a space X, to extend f to the whole space X, i.e., to find a map
F : X → B such that F (x) = f (x) for any x ∈ A if such a map exists. The
corresponding triangular diagram can be written in the form

XOO @
@F
@
i
@
A // B
f

where i denotes the inclusion i(x) = x for all x ∈ A, while the dashed
diagonal arrow is the map that we are searching for. The retraction problem
is the particular case of the extension problem in which A = B and f =id.
The lifting problem is as follows: given maps f : A → B and p : X → B,
to find the lift of f , i.e., a map F : A → X such that p ◦ F = f . The
corresponding triangular diagram has the form

X
~>>
F ~ p
~
~ 
A // B
f

where again the dashed diagonal arrow indicates the desired map. The
section problem is the particular case of the lifting problem in which we have
f = idB : A = B → B and p : X → B is a fiber bundle (a notion that will
be discussed in the next lecture).
Homology theory often provides easy negative solutions to the basic prob-
lems by reducing them to very simple algebraic questions, in particular, for
cases in which the direct topological (geometric) solutions are hopelessly dif-
ficult. We conclude this lecture by such an example. Many more will be
discussed in the exercise class.
8

1.4. The retraction problem and Brouwer’s fixed point theorem

Let us consider the retraction problem for the case in which X = Dn+1 ,
A = Sn = ∂Dn+1 , f is the identity map, i is the inclusion i : ∂Dn+1 ,→ Dn+1 .
Thus we are to find a retraction r of the (n+1)-disk to its boundary n-sphere.
Lemma. There does not exist any retraction of the (n + 1)-disk to its
boundary n-sphere for any n ∈ N.
Proof. Suppose that such a retraction r : Dn+1 → ∂Dn+1 exists. Con-
i r
sider the maps Sn ,→ Dn+1 −→ Sn . The corresponding n-homology homo-
morphisms are
i r
Hn (Sn ) ,→ Hn (Dn+1 ) −→ Hn (Sn ).
By the definition of a retraction, r ◦ i =idSn ; by functoriality

(r ◦ i)∗ = r∗ ◦ i∗ and (idSn )∗ = idHn (Sn ) ;

but since Hn (Sn ) = Z and Hn (Dn+1 ) = 0, we obtain the group homomor-

phisms
i∗ r∗
Z −→ 0 −→ Z
whose composition is the identity, which is impossible. 
An important application of this lemma is the famous Brouwer fixed point
theorem.
Theorem (Brouwer). Any continuous map f of the disk Dn+1 has a fixed
point, i.e., a point p such that f (p) = p.
Proof. Suppose it doesn’t. Then by setting r(x) equal to the intersection
of the ray [f (x), xi with the boundary sphere of the disk, we obviously obtain
a retraction of the disk on its boundary sphere, which is impossible by the
lemma. 
r(x)

f (x)
f (y)
y = r(Y )

Figure 1. Proof of Brouwer’s Theorem

 9

1.5. Problems

1.1. Prove that any functor T : C → D assigns isomorphic objects to iso-

morphic objects.
1.2. In an arbitrary category define the notion of right inverse and left
inverse of a morphism f : X → Y and show that if they both exist, then
they coincide and f is an isomorphism. Prove that any isomorphism has a
unique inverse.

In order to solve the subsequent problems, you can assume known the
homology groups of Dn , Sn , RP n , CP n , T2 , Mg2 .
1.3. Can the torus be retracted on its meridional circle?
1.4. Is there a retraction of S5 on S4 , where S4 is the “equator” of S5 , i.e.
S5 = ΣS4 (here Σ denotes suspension) ?
1.5. Is there a retraction of RP 5 on RP 4 , where RP 4 ,→ RP 5 is the natural
embedding given by (x1 ; . . . ; x5 ) 7→ (x1 ; . . . ; x5 ; 0) ?
1.6. Is there a retraction of the solid torus S = S1 × D2 on its boundary
dS = T2 = S1 × S1 ?
1.7. Given a continuous map p : RP 2 → S1 × D17 and a homeomorphism
h : S1 × D17 → S1 × D17 , can h be lifted to a map H : S1 × D17 → RP 2 ?
1.8. Prove that Euclidean spaces of different dimensions are not homeo-
morhic.
10

Lecture 2
CW-COMPLEXES

Roughly speaking, a CW-complex is a topological space consisting of

disks of various dimensions glued together so that the boundary of each disk
is attached to disks of lower dimensions. In a certain sense, the category of
of CW-complexes and their morphisms (called cellular maps) is the widest
topological category that excludes unwieldy pathological topological spaces
and has a reasonable geometric visualization. In this lecture, we prove the
so-called cellular approximation theorem (which allows to replace arbitrary
continuous maps by cellular ones) and begin the study of fiber bundles, one
of the most important notions of geometric topology.

2.1. CW-complexes and their morphisms

Let X be topological space, X 0 its subset, and let σ : Dk → X be a
continuous map which is a homeomorphism on the interior int Dk = Dk \ ∂Dk
of the disk; then we say that σ is the characteristic map of a k-cell of X
attached to the subset X 0 ; we denote ek := σ(int Dk ), ēk := σ(int Dk ) and
call these sets open and closed k-cells, respectively; the restriction σ|∂Dk is
often called the attaching map of the k-cell ēk . A CW-space (also called cell
space or CW-complex) is defined as a topological space X presented as the
union X = k X , where X (k) (called the k-skeleton of X) is the union of
S (k)
X (k−1) and a certain number of k-cells attached to X (k−1) provided that the
two following axioms are satisfied:
• C (closure finiteness): the boundary ēk \ ek of each k-cell intersects a
finite number of l-cells of lower dimensions, i.e., cells with l < k;
• W (weak topology): a set F ⊂ X is closed iff the intersection F ∩ ēk is
closed for any cell ek .
If the number of cells constituting the CW-space is finite, the two ax-
ioms hold automatically. In that case (which we will usually consider) the
CW-space X is obviously compact, finite-dimensional, and Hausdorff; it is
useful to visualize it as the result of the following process: first take a fi-
nite number of points (the zero cells), attach a finite (possibly void) set of
1-cells (segments) by their endpoints to the set X (0) of zero cells, thus form-
ing the 1-skeleton X (1) , attach a finite (possibly void) set of 2-cells (2-disks)
along their boundary circles to X (1) , and so on, and finally attach the (finite
nonempty) set of n-cells of the highest dimension n to X (n−1) .
 11

Here are some simple examples:

• The sphere Sn , n ≥ 1, has a CW-space structure with two cells.
• The disk Dn , n ≥ 1, has a CW-space structure with three cells.
• The torus T2 := S1 × S1 has a CW-space structure with four cells.
• The sphere with h ≥ 1 handles Mg , g = h−1, has a CW-space structure
with 4h + 2 cells.
More interesting examples will be studied in the exercise class.
A continuous map f : X → Y of CW-spaces is said to be a cellular map
if f (X (k) ) ⊂ Y (k) for all k ∈ N. The class of all CW-spaces supplied with
cellular maps as morphisms forms one of the most important categories of
modern topology, called the category of CW-spaces.

2.2. Cellular approximation

The cellular approximation theorem is a very useful tool in homology
theory. It claims that any continuous map f : X → Y of CW-spaces is
homotopic to a cellular map. We shall prove a more general statement, the
so-called relative version of this theorem.
Theorem. Let X and Y be CW-spaces, let A ⊂ X be a CW-subspace
(possibly A = ∅), and suppose that there exists a continuous map
f : X → Y whose restriction to A is cellular. Then there exists a cellu-
lar map g : X → Y homotopic to f , and the homotopy on A is the identity.
Proof. We proceed by induction on the dimension of the cells σαn , ex-
tending the definition of the map g from the boundary of each cell to its
interior. To do this, we look all the cells emβ , m > n, which contain part of
the image of σα , and “blow out” this image to the boundary of em
n
β .
Remark. The image of a cell ēk in X can contain a cell of higher dimen-
sion ēm , as the famous Peano curve example shows.
Thus, in order to prove the theorem, it suffices, given a continuous map
f : Dn → Y and the characteristic map of an m-cell χ : Dm → Y , where
m > n and f (∂Dn ) ⊂ Y \ intχ(Dm ), to construct a map g : Dn → Y such
that
/ intχ(Dm ), then g(x) = f (x);
(i) if f (x) ∈
(ii) the map g is homotopic to f , the homotopy being the identity outside
intχ(Dm );
(iii) g(Dn ) ⊂ Y \ intχ(Dm ).
12

Actually it suffices to prove that there exists a map g : Dn → Y that

satisfies conditions (i) and (ii) and whose image does not contain at least
one point y ∈ intχ(Dm ). Indeed, if this is the case, it is easy to use y to
blow out the image of g to the boundary (see Fig.1); to do this, we consider
the composition g1 of the map g = g0 constructed above with the “blow
out” from y of the interior of Dm onto its boundary. This composition is
homotopic to g0 , as can be seen from the formula

gt (x) = (1 − t)g0 (x) + tg1 (x)).

This will complete the proof, because g1 now satisfies condition (iii) (as well
as the conditions (i) and (ii)).

y
−1
χ−1 (y)
(y)
−1(y)
χ
χ
χ
χ
χ −1
−1
−1
(y)
(y)
(y)
D n
g χ

Figure 2. Blowing out the image of g

To construct g, we consider the system of concentric disks

Dm m
ε := {x ∈ R : kxk ≤ ε},

so that Dm = Dm m
1 . For any 0 < ε < 1, the disk Dε is homeomorphic to
m m m
χ(Dε ) ⊂ Y , and so we identify Dε and χ(Dε ) ⊂ Y . The map f is uniformly
continuous on the compact set f −1 (Dm
3/4 ), so we can choose a δ > 0 such that

x, y ∈ f −1 (Dm
3/4 ) ⊂ D
n
& kx − yk < δ =⇒ kf (x) − f (y)k < 1/4.

Take a triangulation of Dm all of whose simplices are of diameter less than

δ. If the image under f of a simplex from this triangulation intersects the
sphere Sm−1 m m m
1/2 = ∂D1/2 , then this image is entirely contained in D3/4 \ D1/4 .
The simplices of the triangulation split into three distinct classes:
(a) those whose images are disjoint from Sm−11/2 ;
(b) those whose images are entirely contained in Sm−1
1/2 ;
(c) those whose images intersect Sm−1
1/2 , but are not contained in it.
 13

We shall construct the map g and the homotopy separately for each sim-
plex. In case (a), we set g(v) = f (v) for all vertices of the simplex and extend
the map by linearity. In case (b), we don’t change anything.
For a simplex whose image intersects Sm−11/2 (case (c)), the situation is more
complicated, because g is already defined on some of the faces (namely, on
those satisfying (a) or (b)), and we must extend g to the entire simplex. For
the vertices, we set g(v) = f (v). On a 1-face, if the map is not yet defined,
we extend it from the endpoints by linearity. On a 2-face ∆2 , if the map
is not yet defined, we define it as follows: we cover ∆2 by segments [m, x]
joining its baricenter m to points x on its boundary (at which g is already
defined), then set g(m) = f (m) and extend g to [m, x] by linearity. Further,
we perform the same construction for 3-faces, and so on, thus defining g on
the entire disk Dn .
A look at Fig.2, which shows only a “sector” of the disk Dm , should help
visualize what is going on.

1/4
2/4
3/4
Figure 3. The image of ∆k in Dm

By construction, for any simplex ∆k of the triangulation of Dn , its image

g(∆k ) is contained in the convex hull of f (∆k ). In case (c), this convex hull
m−1
does not intersect Dm k k
1/4 . Indeed, if y0 ∈ (∆ ) ∩ S1/2 , then f (∆ ) lies in a
disk of radius 1/4 centered at y0 , and this disk does not intersect Dm 1/4 . So
k m
the image g(∆ ) has no points in D1/4 .
The homotopy ft between f and g is constructed as follows. Whenever
g(x) = f (x), we set ft (x) = f (x) for all t. If g(x) 6= f (x), then both points
f (x) and f (y) belong to Dm and we set
ft (x) = (1 − t)f (x) + tg(x)).
Now the intersection of Dm
1/4 with the image of g lies in a finite number of
affine planes of dimension n < m, so that the disk Dm1/4 must contain a point
y nor belonging to this image, as required. 
14

2.3. Fiber bundles

A locally trivial fiber bundle, or a fiber bundle for short, is a quadruple
(E, B, F, p), where E, B, and F are topological spaces, p : E → B is a
surjective (continuous) map such that
• each point x ∈ B has a neighborhood U homeomorphic to U × F ;
• the homeomorphism U × F → p−1 (U ) is compatible with p, i.e., the
triangular diagram

U × FJ / p−1 (U )
JJ
JJ
JJ p
pr1 JJ
JJ 
%
U

in which pr1 is the projection on the first factor, is commutative. The

map p is called the bundle projection, B is the base of the bundle, F is the
fiber, and E is the total space of the bundle.
Here are some examples of fiber bundles.
• The projection on the first factor of any Cartesian product is a fiber
bundle, called trivial, the second factor playing the role of the fiber.
• Any covering space is a fiber bundle (with discrete fiber).
• The tangent bundle of an n-dimensional smooth manifold is a fiber
bundle with fiber the n-dimensional linear space.
• The open (i.e., with boundary circle removed) Möbius band is the total
space of a fiber bundle with base S1 and fiber R.
• The natural map of the Stiefel manifold V (k, n) onto the Grassmann
manifold G(k, n) is a fiber bundle with fiber the Lie group GL(k).
Note that all nontrivial fiber bundles have bases whose topology is, in
some sense, nontrivial. This is no accident: any (locally trivial) fiber bundle
is (globally) trivial if the base is “topologically trivial”, e.g. is a disk, as the
following statement shows.
Theorem [Feldbau]. Any fiber bundle p : E → Ik over the cube Ik is
trivial, i.e., is the Cartesian product of the cube by the fiber F .
Proof. First let us split the cube Ik = Ik−1 × [0, 1] into two half-cubes
Ik1 = Ik−1 × [0, 1/2] and Ik2 = Ik−1 × [1/2, 1], and, assuming that the bundles
over the half-cubes p1 : E1 → Ik1 and p2 : E2 → Ik2 are trivial, prove that so
is the bundle over the whole cube Ik .
 15

A point in E1 has coordinates (x, f ), x ∈ Ik1 , f ∈ F . Let us denote similar

coordinates in A2 by [x, f ]. If x belongs to both half cubes, then each point
e ∈ p−1 (x) ⊂ E1 ∩ E2 , has coordinates (x, f1 ) in E1 and [x, f2 ] in E2 (f1 is
not necessarily equal to f2 ). This defines a map fx : F → F , f1 7→ f2 . Let
us map the half cubes onto their intersection via the natural projection
π : (x1 , . . . , xk ) 7→ (x1 , . . . , xk−1 , 1/2)
and define the map ϕ : E2 → Ik2 × F by the formula
ϕ[x, f ] := (x, fπ(x) (f )).
Then ϕ is clearly a homeomorphism, which, together with the identical home-
omorphism E1 → Ik1 × F , constitutes a homeomorphism E → Ik × F ; this
shows that the bundle over the whole cube is trivial, as claimed.
Now let us divide the cube into halves again; then we can assume that
the bundle over one of the halves is nontrivial (otherwise the theorem follows
by what was proved above). Choosing such a nontrivial half-bundle p(1) , we
divide its base into two equal cubes and iterate the argument, choosing a
nontrivial quarter-bundle p(2) , and so on. After a certain number of itera-
tions, the base of the chosen bundle p(N ) will be so small that it will lie inside
an open set over which the given bundle is trivial (here we are using the local
triviality condition in the definition of fiber bundles), which is impossible
since p(n) was assumed nontrivial. This contradiction proves the theorem. 
Of course fiber bundles form a category if one defines bundle morphisms
in the natural way. Namely, suppose ξ = (E, B, F, p) and ξ 0 = (E 0 , B 0 , F 0 , p0 )
are two fiber bundles; then µ = (f, F ) is a morphism, µ = (f, F ) ∈ hom(ξ, ξ 0 ),
if the following square diagram
E −−−→ E 0
F
 
py p0 y
 

B −−−→ B 0
f

is commutative. We omit the definitions of bundle isomorphism and that of

the notion of sub-bundle, leaving them to the reader.
Given a fiber bundle p : E → B and a map f : X → B, one can define a
bundle pf : Ef → X over X, called the pullback of p via f , by setting

Ef := (e, x) ∈ E × X : p(e) = f (x) and pf (e, x) = p(x).
16

2.4. Covering homotopy theorem

Roughly speaking, the covering homotopy theorem says that the homo-
topy of any map to the base of a fiber bundle can be lifted to its total space,
provided that a lift of the given map is specified. We shall prove a more
general statement, the so-called relative version of the covering homotopy
theorem. In it (and throughout this lecture), I denotes the closed interval
[0, 1] and t is the parameter running over I. Given a homotopy H : X×I → Y ,
the notation Ht : X → Y will always be used for the map defined by the
formula Ht (x) := Y (x, t).
Theorem. Suppose p : E → B is a fiber bundle, X is a CW-space,
X 0 ⊂ X a CW-subspace, and the following maps are given:
• a homotopy Ht : X → B and a lift H e 0 : X → E of its initial map H0 ;
• a homotopy H e 0 : X 0 → E that covers the restriction H 0 of Ht to X 0 ;
t t
Then there exists a homotopy H e t : X → E which covers the homotopy
Ht and coincides with H e 0 on X 0 .
t
Proof. We will need the following lemma (which is also useful in other
contexts).
Lemma (Borsuk). Suppose X is a CW-space, X 0 is a CW-subspace of
X, and f : X → Y is a map. Then any homotopy F 0 : X 0 × I → Y of the
map f 0 := f |X 0 can be extended to a homotopy of f .
Proof. We extend the homotopy by induction on the dimension n of the
cells of the CW-space. Let n = 0; if x0 ∈ X 0(0) , then the map {x0 } × I → Y
is already defined, while if x0 ∈ / X 0(0) , we send {x0 } × I to f (x0 ).
Now suppose the homotopy has been extended to the skeleton X n , where
n ≥ 0. Then for each (n + 1)-cell, we have a map defined on Sn × I and
on Dn+1 × {0} that we must extend to Dn+1 × [0, 1]. To do this, embed the
cylinder Dn+1 × I in Rn+2 , choose a point P on the axis of the cylinder above
its top Dn+1 × {1}, and consider the projection x 7→ ϕ(x) from P of the
cylinder to the union of its lateral surface Sn × I and its bottom Dn+1 × {0}.
Now send the point x to the image of ϕ(x) under the map already defined
by the inductive assumption. This proves the lemma by induction. 
We now return to the proof of the theorem. Three cases will be considered.
Case 1. Suppose that the bundle is trivial, i.e., E = B×F and p(b, f ) = b.
Then any map to E is the product of two maps (to B and to F ), in particular
we can write
e t0 (x0 ) = Ht0 (x0 ), Φt (x0 ) ,


H
 17

where Φt (x0 ) ∈ F is the F -coordinate of the point H e 0 (x0 ). But Φt can be

t
regarded as a homotopy to F given on a subset X 0 ⊂ X, and so, by Borsuk’s
Lemma, it can be extended to a homotopy Φ b t of the entire set X. Now if we
put
b t (x0 )


He t (x) = Ht (x), Φ

we obtain the desired covering homotopy H

et
Case 2. We now consider the case in which p : E → B is an arbitrary
fiber bundle, but
X = Dn and X 0 = ∂Dn = Sn−1 .
By assumption, we are given a homotopy Ht : Dn → B, a lift H e 0 : Dn → E,
0 n−1 0
and a homotopy H et : S → E that covers the restriction Ht of Ht to Sn−1 .
Let us take the pullback of the bundle p via the map H, obtaining the
fiber bundle
p1 : E1 → Dn × I = Dn+1 ,
where
E1 = (d, e) ∈ Dn+1 × E : p(e) = H(d)


and p1 (d, e) = p(e). By the Feldbau Theorem, the bundle H ∗ (p) is trivial.
Now consider the identity Dn+1 → Dn+1 as a homotopy Jt : Dn → Dn+1 ,
define
Je0 : Dn → E1
by setting
Je0 (d) = ((d, 0), H0 (d)),
denote Jt0 = Jt |Sn−1 , and define Jet0 ; Sn−1 → E1 by setting

Jet0 (s) = ((s, 0), H

e 0 (s)).
t

Then we are clearly in the situation of Case 1 with

X Dn , X 0 Sn−1 , Ht Jt , Ht0 Jt0 , H

e0 e0
Je0 , H t Jet0 .

Using the result of Case 1, we obtain a homotopy Jet : Dn → E1 .

Now we conclude Case 2 by defining H e t : Dn → E via the formula
e t (d) := ϕ(Jet (d)), where ϕ : E1 → E is the map (d, e) 7→ e. The verification
H
of the required properties of Ht is straightforward from the construction.
18

The above construction is summarized by the following diagram:

Jet
-E /9 E
qq q8 @ 1 ϕ
sssB
 ss 
qqq 
Je0 qqq  He 0 sss 
q  ′ H ∗ (p) ss  p
qqqqq  Jet sss
s
 Het ′
q  J ′ ss 
n
qqq n−1 t
 ss 
D ⊃ S / Dn+1 ⊃ D n
⊃ S n−1
6; B
4
Jt Ht

Case 3. We now suppose that the bundle p : E → B and the pair (X, X 0 )
are arbitrary. The proof readily proceeds by induction on the dimension n
of the cells constituting X n ⊂ X, using the results of Case 2 to define the
required map for each n-cell. 
 19

2.6. Problems

2.1. Let C be the union of all circles of center (1/n, 0) in the xy-plane and
radius 1/n. Prove that C is not homeomorphic to a CW-space.
2.2. a) Find a space satisfying (W) but not (C).
b) Find a space satisfying (C) but not (W).
c) Is the closure of cell necessarily a subspace ?
2.3. Find mimimal CW-complex structure on CP n , RP n , #qi=1 T2i , #qi=1 RPi2 .
2.4. Define S∞ , RP ∞ , CP ∞ and supply them with CW-structure.
2.5. Prove that S∞ is contractible.
2.6. Give an example of a nontriangulable two-dimensional CW-complex.
2.7. Prove that π1 X lives in X (2) (the 2-skeleton of X), i.e., show that
π1 (X) ≡ π1 (X (2) ).
2.8. Prove that a CW-complex is connected iff its 1-skeleton X (1) is con-
nected.
2.9. Prove that a CW-complex is connected iff it is path-connected.
2.10. Prove that πk Sn = 0 for all k < n.
2.11. Show that the Cartesian product, the cone, the suspension, and the
join of CW-complexes have natural CW-complexes structures.
2.12. Prove that any finite CW-complex X n can be embedded in RN , where
N = (n + 1)(n + 2)/2.
2.13. Show that πn (S1 ) are trivial for all n > 1. (Hint : Consider the bundle
p : R1 → S1 whose projection is given by formula p(x) = exp(xi) )
2.14. Prove that a (locally trivial) fiber bundle p : Sn → B whose base B
consists of more than one point is not homotopic to a constant map.
20

Lecture 3
HOMOTOPY GROUPS

Like homology theory, homotopy group theory is a functor from the cat-
egory of topological spaces and their (continuous) maps to the category of
graded groups and their homomorphisms. The construction of homotopy
groups is much simpler than that of homology groups and they have many of
the properties of homology groups, so that they could be used to solve topo-
logical problems in the spirit of Lecture 1. Unfortunately, they are much
more difficult to compute than the homology groups, and this restricts their
applications. Thus the computation of the homotopy groups of spheres is
still an open problem (despite over 50 years of efforts by the world’s best
topologists), but we shall compute one of them π3 (S2 ), which will give us the
occasion to discover (rediscover?) the beautiful Hopf bundle.

3.1. Homotopy group theory as a functor

We shall now define the homotopy groups πn (X, x0 ) of an arbitrary topo-
logical space X with fixed base point x0 ∈ X for all n ≥ 2. In the case n = 1,
the group π1 (X, x0 ) is the fundamental group, which we assume known.
Let (X, x0 ) be a topological space supplied with a base point, and let
(In , ∂In ) the pair consisting of the n-cube and its (topological) boundary
sphere. The group πn (X, x0 ), n ≥ 2, is defined as the set of spheroids, i.e.,
base point preserving homotopy classes of maps α : (In , ∂In ) → (X, x0 ),
i.e. continuous maps such that α(∂In ) = x0 and homotopies Ht such that
Ht (∂In ) = x0 for all t ∈ I. (The explanation for the term “spheroid” is that
(In /∂In ) = Sn , see Fig.3.1.)

In
ϕ

x0
X

Sn
Figure 3.1. Spheroids

Further we define the product of two spheroids as shown in Fig.2, in which

the shaded part of the cube In is mapped to the base point of x0 ∈ X.
 21

ϕ1
I1n

x0
X
I2n
ϕ2

Figure 3.2. Product of two spheroids

Proposition. The set πn (X, x0 ), n ≥ 2, with the product operation as

defined above, is an Abelian group.
Proof. It is easy to see that the neutral element in πn (X, x0 ) is the con-
stant spheroid Sn → x0 . For the inverse element to an arbitrary spheroid
α : (In , ∂In ) → (X, x0 ), we take the map α0 (x, s) := α(x, −s); then the com-
position is homotopic to the constant map, as will be shown in the exercise
class.
If n ≥ 2, the product operation is commutative. To see this, it suffices to
take a good look at Fig.3.

f g
f g
g f
g f

Figure 3.3. Commutativity of the product operation

Finally, associativity can readily be shown (it will probably be discussed

in the exercise class). 
Of course homotopy groups constitute a functor, and we now define, for
a given base point preserving map f : (X, x0 ) → (Y, y0 ) of topological spaces
and any n ≥ 2, the corresponding induced homomorphism

f∗ : πn (X, x0 ) → πn (Y, y0 ).

This is done in the natural way, i.e., by setting f∗ := {f ◦ α} for any spheroid
{α} ∈ πn (X, x0 ), where the curly brackets denote base point preserving ho-
motopy classes. The fact that f∗ is well defined (i.e., does not depend on the
choice α ∈ {α}) is a straightforward verification.
22

The proof of the following theorem is also a straightforward verification.

Theorem. The homotopy groups are homotopy invariant, i.e.,

X ' Y =⇒ πn (X, x0 ) ∼
= πn (Y, y0 ) for all n > 0.

As pointed out above, although the homotopy groups are easy to define,
they are hard to compute. Here are some examples and properties:
• πn (point) = 0 for all n ≥ 0;
• πk (Sn ) = 0 for all k < n (this easily follows from the Cellular Approxi-
mation Theorem);
• πn (Sn ) = Z for all n ≥ 1 (this is a consequence of the Hurewicz Theorem,
which will be proved in subsequent lectures, and the fact that Hn (Sn ) = Z);
• πn (X × Y ) = πn (X) ⊕ πn (Y ) (obvious);
An important property of the homotopy groups is the fact that there
exists a natural action of the fundamental group π1 (X, x0 ) on πn (X, x0 ),
n ≥ 2. Because of this action, π2 (S1 ∧ S2 ) is not isomorphic to Z, as one
might naively suppose. Indeed, we can modify the spheroid id :S2 → S2
by first mapping S2 to the wedge S2 ∧ I by identifying the parallels within
the Arctic Circle to points and then wrapping the “tail” I around S1 several
times. The element of π2 (S1 ∧ S2 ) thus obtained differs from the spheroid id.
In fact π2 (S1 ∧ S2 ) ∼
= Z ⊕ ··· ⊕ Z ⊕ ....

X
x0

Figure 3.4. The action of π1 (X) on πn (X)

3.2. Exact homotopy sequence for fiber bundles

A sequence of groups and homomorphisms
ϕi ϕi−1
. . . −→ Gi −→ Gi−1 −→ Gi−2 −→ . . .
 23

is called exact at the term Gi−1 if Im(ϕi ) = Ker(ϕi−1 ), and simply exact if it
is exact at all its terms. It is easy to prove (see the exercise class) that in
any exact sequence of the form
ϕ
0 −→ G2 −→ G1 −→ 0

the homomorphism ϕ is necessarily an isomorphism.

Now suppose p : E → B is a fiber bundle with fiber F , e0 and b0 are base
points such that p(e0 ) = b0 and p−1 (b0 ) = F 3 e0 . Denote by i the inclusion
(F, e0 ) ,→ (E, e0 ). We already have two of the homomorphisms that will
appear in our sequence (namely p∗ and i∗ ); we now construct the third one

∂∗ : πn (B, b0 ) → πn−1 (f, e0 )

as follows. Let α : (Sn , s0 ) → (B, b0 ) be a spheroid from πn (B, b0 ). Think of

the sphere Sn as lying in Rn+1 and cut it up into (n−1)-spheres with common
point the North Pole s0 by rotating (n − 1)-dimensional hyperplanes passing
through s0 . Then the map α can be regarded as a homotopy αt : Sn−1 → B
joining two copies α0 , α1 of the constant map Sn−1 → b0 ∈ B (see Fig.5).

α
s0
b0
ϕt
Stn−1
n
S
B
Figure 3.5. Definition of ∂∗

et : Sn−1 → B.
By the covering homotopy theorem, there exists a covering ϕ
We define ∂∗ by setting ∂∗ (α) := ϕ
f1 .
Theorem. The sequence of homomorphisms
i
∗ p∗ ∗ ∂
· · · → πn (F, e0 ) −→ πn (E, e0 ) −→ πn (B, b0 ) −→ πn−1 (F, e0 ) → . . .

defined above is exact.

Proof. To show the exactness of our sequence, we must prove six inclu-
sions of the form Im(.)⊂ Ker(.) and Ker(.)⊂ Im(.). They follow from the
definitions and the Covering Homotopy Theorem; we shall prove only one
24

inclusion (namely Ker(∂∗ ) ⊂ Im(p∗ )), leaving the other ones for the exercise
class.
We can represent any spheroid α0 : (Sn , s0 ) → (B, b0 ) as a homotopy
αt : Sn−1 → B and (by the Covering Homotopy Theorem) and consider
the covering homotopy α et : Sn−1 → E. If α0 was chosen in Ker(∂∗ ), then
n−1
α
e1 : S → F is homotopic to the constant map. Let βs be the homotopy
in F joining α
e1 to the constant map. Consider the homotopy
(
αe2t if t ∈ [0, 1/2],
ψet =
βe2t−1 if t ∈ [1/2, 1].

Corresponding to this spheroid is the spheroid ge : Sn−1 → F for which the

map g := p ◦ ge is homotopic to α0 . Hence α0 ∈ Im p∗ , as required. 

3.3. Exact homotopy sequence for pairs

We shall now define the homotopy groups πn (X, A, a0 ) of a pair of topo-
logical space with base point a0 ∈ A ⊂ X for all n ≥ 2. Define a relative
spheroid as a base point preserving homotopy class of maps

α : (Dn , ∂Dn , s0 ) → (X, A, x0 ),

i.e., continuous maps such that α(s0 ) = x0 and homotopies Ht such that
Ht (∂Dn ) ⊂ A and Ht (s0 ) = x0 for all t ∈ I. It is often more convenient
to interpret spheroids as (classes of) maps (In , ∂In , s0 ) → (X, A, a0 ) (this
interpretation gives the same result, because the pair In /∂I n is homeomorphic
to (Dn , ∂Dn ) The main definition, that of the product of two relative spheroids
is relegated to the exercise class, together with the proof of the fact that
πn (X, A, a0 ) is a group for any n ≥ 2 and is Abelian for n ≥ 3.
Of course the homotopy group πn (·, ·) of pairs can be regarded, for n ≥ 3,
as a functor from the category of pairs of topological spaces to the category
of Abelian groups; the main definition, that of the induced homomorphism

f∗ : πn (X, A, a0 ) → πn (Y, B, b0 )
corresponding to the map of pairs f : (X, A) → (Y, B), is left as an exercise.
Given a pair of spaces with base point, (X, A, a0 ), we have the inclusion
map i : (A, a0 ) ,→ (X, a0 ), which determines the induced homomorphism i∗ .
Since any absolute spheroid (Dn , s0 ) → (X, a0 ) can be regarded as a relative
 25

spheroid (Dn , Sn−1 , s0 ) → (X, A, a0 ) (which takes Sn−1 to a0 ), we also have

the homomorphism p∗ : πn (X, a0 ) → πn (X, A, a0 ). Finally, we define
∂∗ : πn (X, A, a0 ) → πn (A, a0 )
by assigning to the relative spheroid (Dn , Sn−1 , s0 ) → (X, A, a0 ) its restriction
to Sn−1 .
Theorem. The sequence of homomorphisms
i ∗ p∗ ∗∂
· · · → πn (A, a0 ) −→ πn (X, a0 ) −→ πn (X, A, a0 ) −→ πn−1 (A, a0 ) → . . .
defined above is exact.
Proof. Again we must verify six inclusions, but we will check only one,
namely Im p∗ ⊂ Ker ∂∗ . Let α : In → X be a spheroid whose restriction to
In−1 is homotopic to the constant map to a0 (in the class of maps In−1 → A).
Let gt : In−1 → A be a homotopy between the restriction of α to In−1 and
the constant map. Consider the homotopy ft : ∂In → X that coincides with
gt on In−1 and takes ∂In \ In−1 to a0 . Using the Borsuk Lemma, we extend
this homotopy to a homotopy of α. Thus we obtain a homotopy in the class
of relative spheroids that joins α to a relative spheroid taking ∂In to a0 . 

3.4. The Hopf bundle and π3 (S2 )

The Hopf bundle, one of the most beautiful constructions in topology
and in all of mathematics, is the fiber bundle h : S3 → S2 constructed in the
following way: present the sphere S3 in the form
S3 = (z1 , z2 ) ∈ C2 : |z1 |2 + |z2 |2 = 1


and define an action of the circle S1 = {w ∈ C : |w| = 1} by the rule

(z1 , z2 ) 7→ (wz1 , wz2 ); it is easy to prove (see the exercise class) that the
orbit space of S3 under this action is the 2-sphere.
This definition is very simple but hard to visualize. A more visual defini-
tion, based on the fact that the 3-sphere can be obtained by gluing together
two solid tori, will be discussed in the exercise class, but the best way to
visualize the Hopf bundle is to look at the marvelous animation of the con-
struction due to Etienne Ghys (see “Dimensions” on his web page).
The Hopf bundle can be used to obtain the following remarkable formula
for π3 (S2 ), which in its day was a little mathematical sensation:

π3 (S2 ) = Z.
26

To prove this, let us write out part of the homotopy sequence for the Hopf
bundle:
p∗ ∂ i
· · · → π2 (S3 ) −→ π2 (S2 ) −→
∗
π1 (S1 ) −→
∗
π1 (S3 ) → . . . .
Since π2 (S3 ) = π1 (S3 ) = 0 (see the previous section), we obtain the isomor-
phism π2 (S2 ) ∼= π1 (S1 ), and since the fundamental group of the circle is Z,
2 ∼
we have π2 (S ) = Z. (This is particular case of a fact mentioned but not
proved in the previous section.)
Now let us write out another part of the same homotopy sequence for the
Hopf bundle:
i p∗ ∂
· · · → π3 (S1 ) −→
∗
π3 (S3 ) −→ π3 (S2 ) −→
∗
π2 (S1 ) → . . .
Since the two extreme terms are zero, we obtain π3 (S3 ) ∼ = π3 (S2 ), and our
claim follows from the isomorphism (mentioned in the previous section for
any n, not only 3) π3 (S3 ) ∼= Z. 
3.5. Homotopy groups of spheres: some information
The computation of πk (Sn ) for k > n, arguably a useless task, was one
of the central topics of mathematics in the 1950ies and 1960ies, mostly out
of sheer curiosity. The remarkable work of L.S.Pontryagin, J.-P.Serre, and
J.F.Adams eventually turned out to have other, much more useful applica-
tions. Although open questions still remain, the computation of the homo-
topy groups of spheres is no longer very fashionable today. Here we list only
a few results, chosen more or less at random, that may seem very strange at
first glance.
• πn (Sn ) = Z, n ≥ 1;
• πn+1 (Sn ) = Z2 , n ≥ 3;
• πn+2 (Sn ) = Z2 , n ≥ 4;
• πn+3 (Sn ) = Z24 , n ≥ 5;
• πn+4 (Sn ) = 0, n ≥ 6;
• πn+7 (Sn ) = Z240 , n ≥ 9;
• πn+9 (Sn ) = Z2 ⊕ Z2 ⊕ Z2 , n ≥ 11;
• πn+11 (Sn ) = Z504 , n ≥ 13.
The main tool for proving these results are the so-called spectral se-
quences, in particular those due to J.-P.Serre and J.F.Adams. Spectral se-
quences are a very sophisticated and effective means for computing homology
groups, but are beyond the framework of this course.
 27

3.6. Problems

3.1. Prove that πk X lives in X (k+1) , more precisely ik : X (k) ,→ X induces

an isomorphism (ik )∗ : πk X (k+1) → πk X for k < n and an epimorphism for
k = n (n is the dimension of X)
3.2. Prove that πn (S1 ∨ Sn ) = Z ⊕ Z ⊕ · · · ⊕ Z ⊕ . . . .
3.3. Prove that the product of spheroids is well defined.
3.4. Prove that the induced homomorphism f∗ : πn X → πn Y is well defined.
3.5. Prove the exactness of the homotopy sequence for fiber bundles at the
other terms.
3.6. Prove the exactness of the homotopy sequence for pairs at the other
terms.
3.7. Prove that CP 2 = D4 ∪ CP 1 , where p : S3 → CP 1 is the Hopf bundle
and S3 = ∂D4 .
3.8. Does there exist a retraction r : CP 2 → CP 1 , where CP 1 is embedded
in CP 2 in the natural way?
3.9. If A is a retract of X, then
a) the map i∗ : πn (A) → πn (X) is injective ;
b) the map p∗ : πn (X) → πn (X, A) is surjective;
c) the map ∂∗ : πn (X, A) → πn−1 (A) is a zero homomorphism;
d) πn (X) = πn (X, A) ⊕ πn (A).
3.10. If there exists a homotopy ft : X → X for which we have f0 = id and
f1 (X) ⊂ A, then
a) the map i∗ : πn (A) → πn (X) is surjective ;
b) the map p∗ : πn (X) → πn (X, A) is a zero homomorphism;
c) the map ∂∗ : πn (X, A) → πn−1 (A) is injective;
d) πn (A) = πn+1 (X, A) ⊕ πn (X).
28

Lecture 4
CELLULAR HOMOLOGY

Cellular homology theory is a functor from the category of CW-complexes

to graded Abelian groups that has the advantage (over other homology the-
ories) of greatly simplifying the actual computations of the homology groups
for the basic topological spaces, e.g. manifolds. In this lecture, we are rather
high-handed about providing rigorous proofs of the properties of this functor,
so as to be able to perform these computations as quickly as possible. The
theory is based on the notion of degree of a map f : Sn → Sn of the sphere
to itself, with which we begin.

4.1. Degree of self maps of the n-sphere

The simplest way to define the degree of a map f : Sn → Sn is to set
deg(f ); = f∗ (1) ∈ Z, where f∗ : πn (Sn ) → πn (Sn ) = Z is the homomorphism
induced by f . But we can’t do that, because we have not established that
πn (Sn ) = Z, a fact which is usually proved by using homology theory (specif-
ically, the Hurewicz theorem, to appear in Lecture 6), a theory that we are
only beginning to construct; thus such an approach would constitute a logical
vicious circle. So we need a geometric definition of deg(f ), and will give it
in the case of a rather restricted class of maps f .
Let us call a continuous map ϕ : Sn1 → Sn2 neat if, for some point p ∈ S
the set Sn1 \ ϕ−1 (p) consists of a family (possibly empty) or disjoint open balls
B1n , . . . , Bkn each of which is mapped by ϕ diffeomorphically onto Sn2 \ {p}.
Now suppose that fixed orientations are chosen on the two spheres. Then
the restrictions ϕ|B1n , . . . , ϕ|Bkn are called positive or negative depending on
whether they preserve or reverse orientation (i.e., their Jacobians in the cho-
sen coordinate systems are positive or negative). Then we can define the
degree deg(ϕ) of the map ϕ as the difference between the number of posi-
tive restrictions (listed above) and the number of negative ones. Intuitively
speaking, we can say that deg(ϕ) is the number of times one sphere wraps
around the other.

4.2. Incidence coefficient of two cells

Let X = i X (i) be a CW-space; let α : Dn−1 → X be an (n − 1)-cell and
S
χ = β|∂Dn : ∂Dn → X (n−1) be the attaching map of the n-cell β : Dn → X (n) .
Denote by p : X (n−1) → Sn−1 the map obtained by compressing to a single
 29

point the (n−2)-skeleton X (n−2) as well as all the (n−1) cells of X (n−1) except
α. Define the map χ b : ∂Dn → Sn−1 by setting χ b := p ◦ χ. By definition, the
incidence coefficient of the cell β to the cell α is the integer

χ) = deg(p ◦ χ) .
[β : α] := deg(b

(Here the orientations of the spheres ∂Dn and Sn−1 are those induced by the
corresponding cells.) The incidence coefficient is correctly defined if all the
maps of the form χ b are neat for all pairs of cells α and β. In that case, we say
that the CW-space X is neat. We will assume that all CW-spaces considered
in the rest of this lecture are neat.
Example. Let X be the CW-space structure on the projective plane
RP 2 consisting of three cells of dimensions 0,1,2. The 2-cell β : D2 → X is
attached to X (1) = S1 by the map χ : ∂D2 = S1 → S1 , eiϕ 7→ e2iϕ . Then the
incidence coefficient [β : α], which expresses the number of times that the
boundary of β wraps around α, is equal to 2.
Remark 1. The key property of neat CW-spaces is that for them the
incidence coefficient of two cells is correctly defined via the definition of degree
of sphere maps. There are many different ways of ensuring this property, and
other authors use, for this purpose, notions other than what I call “neatness”.
Remark 2. The general construction described above becomes unrecog-
nizable in dimension k = 1, so we will treat this case separately. The 1-disk
D1 is the segment [0, 1], which we always assume oriented from 0 to 1, and
any 1-cell σ is determined by a map of its boundary ∂D1 = {0} ∪ {1} to the
0-skeleton (the vertices) of our CW-space. Now if the point {0} is mapped to
some point p of the 0-skeleton, then the corresponding incidence coefficient is
set to be equal to −1, while if {1} is mapped to some point q, the incidence
coefficient taken to be +1; in the case p = q, the incidence coefficient will be
set equal to zero.

4.3. Definition of cellular homology

Suppose that X = ni=0 X (i) is a finite neat CW-space. Then for each
S
integer k, 0 ≤ k ≤ n, we define the group of cellular k-chains of X as the set
of all formal linear combinations of k-cells γsk : Dk → X k ,
n X o
k
Ck (X) := zs γs zs ∈ Z
all k-cells
30

with integer coefficients zs , endowed with the natural sum operation.

X X X
zs0 γ k + zs00 γ k = (zs0 + zs00 )γ k .
all k-cells all k-cells all k-cells

Under this operation Ck (X) is an Abelian group (actually, it is even a free

Z-module). We introduce the cellular boundary operator by defining it on
each cell by the formula
X
∂k (γ) := [γ : βm ] βm
all (k − 1)-cells

and then extending to the entire group Cn (X) by linearity. This definition
makes sense provided we know that the incidence coefficients are well defined;
this will be the case if the CW-space X is neat, but we have already assumed
this.
Chains c ∈ Ck (X) such that ∂k (c) = 0 are called cycles and those for
which there exists a chain c0 ∈ Ck+1 such that ∂k+1 (c) = c0 are called bound-
aries (or are said to be homologous to zero). Two cycles whose difference
is homologous to zero are said to be homologous. Being homologous is, of
course, an equivalence relation.
We claim that the boundary operator satisfies the Poincaré Lemma, i.e.,
we have
∂k+1 ◦ ∂k = 0 for all k ≥ 0.
We omit the proof of this statement.
The Poincaré lemma implies that Im∂k+1 ⊂ Ker ∂k ⊂ Ck (X), and this
allows us to define the nth (cellular) homology group of a (neat) finite CW-
space X by taking the quotient group

Ker ∂k+1
Hk (X) := for all k ≥ 0.
Im ∂k

Thus elements of the homology group Hn (X) are classes of cycles up to

homology.
Cellular homology of finite neat CW-spaces is a functor, so we must de-
fine, for every cellular map f : X → Y and every nonnegative integer n, a
homomorphism of Hn (X) to Hn (Y ). This is done in the natural way: to
 31

every n-cell χ : Dn → X (n) corresponds the cell f ◦ χ, and so (by linearity),

to each n-chain in X corresponds an n chain in Y ; we write

f∗n : Cn (X) → Cn (Y );

it is not hard to prove that

∂nY ◦ f∗n = f∗(n−1) ◦ ∂nX ,

so that cycles correspond to cycles and homologous cycles correspond to ho-

mologous cycles (thus the described correspondence is well defined on homol-
ogy classes), and preserves the sum operation (for details – see the exercise
class); the homomorphism Hn (X) → Hn (Y ) thus obtained is denoted by

f∗ : Hn (X) → Hn (Y )

and called the homomorphism induced by f .

Now suppose we are given a pair of (finite neat) CW-spaces (X, A); then
a relative chain c ∈ Cn (X, A) is a chain from Cn (X) whose coefficients at
cells of A are zero. As above, we define a boundary operator (still denoted
by ∂n ),
∂n : Cn (X, A) → Cn−1 (X, A)
which satisfies the Poincaré Lemma and so allows us to define, as above,
the relative homology group Hn (X, A) and g∗ : Hn (X, A) → Hn (Y, B), the
homomorphism induced by a cellular map g : (X, A) → (Y, B) of pairs. Note
there there is a natural identification Hn (X, ∅) ≡ Hn (X).
Remark. Here we defined cellular homology for finite CW-spaces be-
cause, at this point, we are only interested in the computation of homology
groups of compact manifolds, which all have structures of finite CW-spaces.
Actually the theory in the general case is just the same, it should only be
noted that in the definition of an n-chain, one must stipulate that the number
of nonzero integer coefficients of each chain be finite.

4.5. Some properties of cellular homology

(i) Functoriality. Cellular homology is a functor, which means that
• (f ◦ g)∗ = f∗ ◦ g∗ for all cellular maps f : X → Y and g : Y → Z:
• (idX )∗ = idHn (X) for any CW-space X and all n ∈ N.
(ii) Homology of a point. H0 (point) = Z and Hn (point) = 0 for all n ≥ 1.
32

(iii) Homotopy invariance. Homology groups are homotopy invariant (and

therefore topologically invariant). In particular, they are independent of the
CW-space structure chosen for each CW-space. We omit the proof of this
important fact.
(iv) Zero homology of connected spaces. A CW-space X is path connected
if and only if H0 (X) = Z.
(v) Exact sequence for pairs. Given a pair of CW-spaces (X, A), we
have the inclusions i : A ,→ X and j : X = (X, ∅) ,→ (X, A), which
induce homomorphisms i∗ and j∗ . Further, it is not difficult to construct a
homomorphism ∂∗ : Hn (X, A) → Hn−1 (A) (see Exercise 4.11). The three
homomorphisms defined above allow to construct a sequence similar to the
two sequences for homotopy groups studied in the previous lecture. As before,
we have the following statement.
The sequence of homomorphisms
i
∗ j∗ ∗ ∂
· · · → Hn (A) −→ Hn (X) −→ Hn (X, A) −→ Hn−1 (A) → . . .

defined above is exact.

The details concerning items (i), (ii), (iv) (including proofs) will be dis-
cussed in the exercise class. At this stage, the proofs of (iii) and (v) are
omitted, although I hope that, on the intuitive geometric level, the exactness
of the homology sequence for pairs will be understood by everyone.

4.4. Computations and applications

Here we list the values of the homology groups of some of the most popular
manifolds. The proofs can be obtained by using the simplest CW-space
decompositions of the manifolds (see Lecture 2) and the definition of Hn (·);
they will be discussed in the exercise class.
• H0 (point) = Z, Hn (point) = 0 for n ≥ 1.
• H0 (Dn ) = Z, Hk (Dn ) = 0 for k ≥ 1.
• H0 (Sn ) = Hn (Sn ) = Z, Hk (Sn ) = 0 for all k ∈
/ {0, n}.
• H0 (T2 ) = H2 (T2 ) = Z, H1 (T2 ) = Z ⊕ Z, Hk (T2 ) = 0 for all k ≥ 3.
• Hk (CP n ) = Z for k ∈ {0, 2, . . . , 2n}, Hk (CP n ) = 0 for all other k.
• Hk (RP n ) = Z for k = 0 and k = n provided n is even, Hk (RP n ) = Z2
for odd k less than n, Hk (RP n ) = 0 for all other k.
 33

The values of the homology groups of surfaces (orientable or not, with

boundary or not) will be calculated in the exercise class. Unlike homotopy
groups, homology groups “behave badly” under the Cartesian product of
spaces; however (also unlike homotopy groups), they “behave nicely” under
the wedge product, namely:
• Hn (X ∧ Y ) = Hn (X) ∗ Hn (Y ),
where ∗ denotes the free product of groups.
More facts and other applications will appear in the exercise class.
34

4.5. Problems

4.1. Compute H∗ (Mg2 ), where Mg2 is the sphere with g handles.

4.2. Compute H∗ (Ng2 ), where Ng2 is RP 2 with g handles.
4.3. Compute H∗ (CP n ).
4.4. Compute H∗ (RP n ).
4.5. Compute H∗ (Γ), where Γ is the following graph.

4.6. Let p and q be two relatively prime positive integers. Consider the
3 2
action of the group Zp with the
generator σ on
the
unit sphere S ⊂3 C
defined by σ(z, w) = exp 2πi/p z, exp 2πiq/p w . The quotient of S by
this action is a 3-manifold. It is called a lens space and is denoted by L(p, q).
Compute H∗ (L(p, q)
4.7. Compute H∗ (D), where D is the dunce hat, i.e., the triangle with the
identifications shown by arrows. Can it be retracted to its circle N M ?
4.8. Can the dunce hat be retracted to its circle N M ?

M
 35

4.9. Express Hn (A ∨ B) in terms of Hn A and Hn B, n = 0, 1, . . .

4.10. Construct the homomorphism ∂∗ : Hn (X, A) → Hn−1 (A) and show
that the homology sequence for pairs is exact.
4.11. Compute H∗ (S̄), where S̄ is the sphere S2 to which the segment [N S]
joining the North and South poles have been added.

S
36

Lecture 5
SIMPLICIAL HOMOLOGY

Simplicial homology is the oldest one of the homology theories. It is

a functor defined only on the category of simplicial spaces (which is much
narrower than the category of topological spaces or even that of CW-spaces);
its definition is quite simple and has a rather clear geometric interpretation.
However, it is not as convenient for computations as cellular homology (see
Lecture 4 and its exercises), and the proof of its main properties is much
more difficult than that of the same properties of singular homology (which
we will study in Lecture 7).
The construction of the simplicial homology functor (as of other homol-
ogy functors) is carried out in two basic steps: first, from simplicial spaces
and maps (which are geometric entities, see Sect.5.3 below), we pass to chain
complexes and morphisms (which are purely algebraic objects), and, second,
by a purely algebraic construction, we pass from chain complexes to graded
Abelian groups (the homology groups). This second step can be used with-
out any modifications in the construction of other homology theories, e.g.
singular homology, C̆ech homology, cellular homology, etc.

5.1. Chain complexes and their morphisms

A chain complex (sometimes also called graded differential group) is a
sequence of Abelian groups and homomorphims
∂n+2 ∂n+1 n ∂ ∂n−11 ∂
. . . −−−→ Cn+1 −−−→ Cn −−−→ Cn−1 −−−→ . . . −−−→ C0

satisfying the relation ∂n+1 ◦ ∂n = 0 for all n = 1, 2, . . . , or, which is the

same thing, the relation Im ∂n+1 ⊂ Ker ∂n for all n = 1, 2, . . . . The elements
of Cn are called n-chains and the homomorphisms ∂n are called boundary
operators or differentials (their subscript n is sometimes omitted). Elements
of Ker ∂ are called cycles, elements of Im ∂ are boundaries, and two cycles
in the same coset modulo Img ∂ are said to be homologous or in the same
homology class. (This terminology, which may seem strange in the abstract
algebraic context, actually comes from the geometric aspect of homology
theory, where it is quite natural: see Lecture 4.)
Chain complexes form a category whose morphisms f : C → C 0 are com-
mutative diagrams of the form
 37

∂n+1 n ∂ ∂n−1 1 ∂
. . . −−−→ Cn −−−→ Cn−1 −−−→ . . . −−−→ C0
  
  
y y y
0 0
∂n+1 ∂0 ∂n−1 ∂0
. . . −−−→ Cn0 −−−
n 0
→ Cn−1 1
−−−→ . . . −−−→ C00

The fact that this is indeed a category (i.e., the two functoriality axioms
in the definition of categories, see Lect.1, hold) follows immediately from
definitions.

5.2. Homology of chain complexes

The n-th homology group of a chain complex C = Cn , ∂n is defined as
the following quotient group

Hn (C) := (Ker ∂n+1 ) (Im ∂n )

Given any morphism f : C → C 0 of chain complexes, we can construct a

homomorphism of the corresponding homology groups

f∗ : Hn (C) → Hn (C 0 ) for all n ≥ 0

in the following way. Consider the diagram

∂n+1 ∂
n
Cn+1 −−−→ Cn −−−→ Cn−1
  
f f f
y n+1 yn y n−1
0
0
∂n+1 ∂0
Cn+1 −−−→ Cn0 −−−
n 0
→ Cn−1 ,

take some element c ∈ Ker(∂n ), and let h be its homology class. Define
c0 := fn (c), and denote by h0 the homology class of c0 . Now we can define
the induced homomorphism f∗ : Hn (C) → Hn (C 0 ) by assigning h 7→ h0 . The
fact that f∗ (here and elsewhere we omit the subscipt n) is a well-defined
homomorphism follows by an fairly easy chase in the last diagram.
It follows directly from the definitions that the assignment described
above of homology groups H∗ (C) and induced homomorphisms f∗ to chain
complexes C and their morphisms f is a functor, i.e.,


f ◦ g ∗ = f∗ ◦ g∗ and (idC )∗ = idH∗ (C) .
38

5.3. Simplicial spaces

An n-dimensional simplex (n-simplex for short) ∆n is the convex hull of
n+1 points (called vertices) in Rn , namely the origin and the endpoints of the
basis vectors; ∆n is supplied with the induced topology. Thus the 0-simplex
is the point, the 1-simplex is the closed interval I = [0, 1], the 2-simplex is
the triangle, the 3-simplex is the tetrahedron, etc. For convenience, we will
regard the empty set as a (−1)-dimensional simplex. Note that we regard the
n-simplex as a topological space (homeomorphic to the disk Dn ), but with
an additional structure: the collection of its k-faces, k = 0, . . . , n − 1, defined
as follows. The 0-faces are the vertices, each of the k-faces, 1 ≤ k ≤ n, is
obtained by choosing k + 1 vertices of ∆n and taking their convex hull. Note
that each k-face has the structure of a k-simplex.
Roughly speaking, a simplicial space is a topological space glued together
from a collection of simplices according to certain rules. It is a very par-
ticular case of a CW-space: the gluing rules for simplices are much more
restrictive than those for cells. More precisely, a simplicial space is defined
as a topological space X presented as the union of a family of simplices,
[
X= σαi σαi ≈ ∆i ,
(α,i)

and satisfying the three following conditions:

• (triangulation) any two simplices intersect in a common face (the empty
set is regarded as a face of each simplex);
• (completeness) any face of any simplex is a simplex of X;
• (local finiteness) any face is the face of only a finite number of simplices.
Simplicial spaces form a category if for morphisms we take simplicial
maps, i.e., continuous maps f : X → Y that take any simplex σ k ⊂ X
linearly onto a simplex σ l ⊂ Y , where l ≤ k.
According to the definition, any simplicial space X is supplied with a
combinatorial structure, i.e, it has a fixed decomposition into simplices. It
is sometimes useful to change this structure by subdividing the simplices of
X into smaller ones. In Figure 1 we show several ways of subdividing a
2-simplex ∆; the last of these is called the baricentric subdivision of ∆ and
is obtained by subdividing each side (1-face) of ∆ into two 1-simplices by
means of the midpoint of the side and then taking the cones with vertex g
 39

(the baricenter of ∆) over the 6 obtained 1-simplices, thus obtaining 6 new

2-simplices.
The definition of the baricentric subdivision of an n-simplex ∆n is similar:
we take the baricentric subdivision of all the faces of ∆n and construct the
cones over them with vertex g (the baricenter of ∆n ). By the baricentric
subdivision of a simplicial space X we mean the result X 0 of the simultaneous
baricentric subdivision of all the simplices of X. By iterating this procedure,
we can obtain the simplicial space X (n) consisting of the same points as X
but with simplices as small as we wish.

Figure 5.1. Subdivisions of a 2-simplex

Simplicial spaces have a topological structure and can be treated as topo-

logical spaces. The connection between the two approaches is given by the
following important theorem.
Theorem (Simplicial Approximation.) Any continuous map f : X → Y
of simplicial spaces is homotopic to a simplicial map s : X (n) → Y (n) of the
nth baricentric subdivisions of X and Y for some n. The map s may be
chosen as close to f as we wish.
The proof of this theorem is somewhat simpler than that of the cellular
approximation theorem, but uses other tools. It is omitted in this course.
Besides its topological and combinatorial structure, a simplicial space
(which is an abstract entity) can be endowed with a very concrete geometric
structure. Indeed, we have the following
Theorem (Polyhedral Realization.) Any abstract simplicial space X
of dimension n can be realized as a polyhedron P (X) in Euclidean space
R2n+1 decomposed into rectilinear simplices that bijectively correspond to
the simplices of X, with faces corresponding to faces.
This theorem is not needed further in this course, so its (not very difficult)
proof is omitted.

5.4. The chain complex of a simplicial space

40

We shall now define the chain complex C∗ (X) of an arbitrary simplicial

space X. The main protagonists in our chain complex will be oriented sim-
plices, i.e., simplices with a fixed orientation. To define the orientation of
a simplex, we write out the sequence of its vertices v0 , v1 , . . . , vn in some
order; there are n! such orders, and we call two orders equivalent if one can
be obtained from the other by an even number of transpositions; for a fixed
simplex, there are obviously two equivalence classes of orderings of its ver-
tices; each of these classes is said to be an orientation of the simplex: the
oriented n-dimensional simplex determined by the ordering v0 , v1 , . . . , vn of
its vertices, i.e., the equivalence class containing this order, will be denoted
σ n = [v0 , v1 , . . . , vn ]. Geometrically, the orientation of a 1-simplex is shown
by an arrow going from one of its vertices to the other, the orientation of a
2-simplex is a direction of rotation of the plane in which the simplex lies. As
to zero-dimensional simplices (i.e., points), let us agree that the orientation is
simply the assignment of a plus sign or a minus sign to the 0-simplex. We will
also agree to denote by −[v0 , v1 , . . . , vn ] the simplex with the same vertices,
but with the orientation opposite to that of the simplex [v0 , v1 , . . . , vn ].
Remark. Physicists distinguish positive and negative orientations, e.g.
in the plane, they consider the counterclockwise rotation positive; in 3-space
they talk about right-hand screws (which determine the positive orientation
of a 3-simplex), there is “left-hand rule” in electromagnetism and so on.
None of these distinctions make any sense mathematically, and so we will
not appeal to these mathematically meaningless preferences in the choice of
orientation.
By definition, an n-dimensional chain c is a linear combination with in-
teger coefficients (only a finite number of which are nonzero) of all oriented
n-simplices from X, X
c= zi σin ∈ Cn (X).
i

The set Cn (X) of all n-chains, called the nth chain group has an obvious
structure of an Abelian group generated by allLthe (ordered) n-simplices
(actually it is a free Z-module); the direct sum n≥0 Cn (X) of the n-chain
groups is also an Abelian group denoted C∗ (X).
To obtain a chain complex from the groups Cn (X), we must define the
boundary operators or differentials

∂n : Cn (X) → Cn−1 (X).

 41

If σ n = [v0 , v1 , . . . , vn ] is an (oriented) n-simplex, we write

[v0 , . . . , vj∨ , . . . , vn ] := [v0 , . . . , vj−1 , vj+1 , . . . , vn ]

for its (n − 1)-face obtained by throwing out its jth vertex. We now define
the boundary operator by setting
X  X X n
X
∂n c = ∂n zi σin := zi ∂n (σin ) := zi (−1)j [v0 , . . . , vj∨ , . . . , vn ].
i≥1 i≥1 i≥1 j=0

The fact that ∂n is well defined is considered in Exercise 5.6.

Lemma (Poincaré).

∂n−1 ◦ ∂n = 0 for any n ≥ 2.

Proof. Because of the linearity of ∂n , it suffices to prove the lemma for

the case in which the chain c contains only one nonzero coefficient z, and this
coefficient equals 1, so that c = σ, where σ is some n-simplex from X. But in
this case it is obvious, because
(by the definition of the boundary operators)
the (n − 2)-chain ∂n−1 ◦ ∂n (σ) consists of 2n summands, which occur in
pairs of the form
[v0 , . . . , vj∨ , . . . , vk∨ , . . . , vn ].
with opposite signs, so that the whole sum cancels out. 
Remark. Actually three different types of simplices can be used to define
simplicial homology: oriented simplices (as was done above), ordered sim-
plices, obtained by numbering all the vertices of the simplicial space and as-
signing to each geometric simplex the unique ordered simplex whose vertices
occur in ascending order, and finally ordered simplices (then each geometric
n-simplex determines n! different ordered simplices, which are the generators
of the group of n-chains). Thus we obtain three different chain complexes,
but it turns out that the resulting homology theories coincide (see Exercise
5.11 and Section 10.3 in the lecture on Poincaré duality ).
Now let us define the morphism of chain complexes corresponding to a
simplicial map f : X → Y ; this is done in the natural way, i.e., given a chain
c ∈ Ck (X) and a simplex σk appearing in it with nonzero coefficient, we take
its image f (σ k ) = τ l with the same coefficient as a summand of the image
chain, while if l < k, we do nothing (adding, so to speak, the zero summand
42

to the image chain); we do this for all simplices appearing in c with nonzero
coefficients and then combine (add together) the coefficients at like simplices
τjk ⊂ Y , obtaining the image chain, denoted by f∗k (c) ∈ Ck (Y ).
It is easy to verify that

∂nY ◦ f∗n = f∗(n−1) ◦ ∂nX ,

so that morphisms of chain complexes correspond to simplicial maps, and the

above constructions define a functor from the category of simplicial spaces
Sim to the category of chain complexes CC.
Once this is done, the definitions of the simplicial homology groups and
their induced homomorphisms
M
H∗ (X) = Hn (X), f∗ : Hn (X) → Hn (Y )
n≥0

is immediate: one simply takes (see Sec.5.2) the homology groups and the
induced homomorphisms of the chain complex constructed above.

5.5. Relative homology of simplicial spaces

A pair (X, A) of simplicial spaces is simply a simplicial space X with a
subset A consisting of simplices of X such that A inherits its own structure
of a simplicial space (i.e., the three conditions in the definition of such spaces
hold, see Sect.5.3). Simplicial maps of pairs of spaces are defined in the
natural way.
Given a simplicial pair (X, A) and a nonnegative integer n, we define
the relative chain group Cn (X, A) as the subgroup of Cn (X) all of whose
chains have zero coefficients at n-simplices from A. Further, one can define
the boundary operators ∂n : Cn (X, A) → Cn−1 (X, A) in the natural way (so
that the ordinary boundary of a chain c ∈ Cn (X, A) that lies in Cn−1 (A) is
regarded as zero) and go on to the definition of the relative homology groups
and the homomorphisms induced by simplicial maps,
M
H∗ (X, A) = Hn (X, A), f∗ : Hn (X, A) → Hn (Y, B).
n≥0

We omit the details at this stage and note without proof that we have defined
a functor from the category of pairs of simplicial spaces to the category of
graded groups.
 43

5.6. Homology with arbitrary coefficients

When we defined the chain groups Cn (X) and Cn (X, A), we used elements
of the group of integers Z for the coefficients appearing in the chains. Actu-
ally, instead of Z, one can take any other Abelian group G (e.g. the additive
group of any field, or residues mod m) and continue further constructions
exactly in the same way. One then obtains the homology groups Hn (X; G)
and Hn (X, A; G), called homology groups modulo G, with similar induced
homomorphisms and boundary operators. The theory is exactly the same as
for the integer homology groups Hn (X; Z) ≡ Hn (X), but for certain applica-
tions, the homology groups Hn (X; R) or Hn (X; Z2 ) are more convenient and
efficient than Hn (X; Z).

5.7. Zero homology and augmentation

The definition of homology given in this lecture and the previous one
is perhaps unclear in the case n = 0, because the boundary operator ∂n
was not specified for n = 0. If we define it as taking everything to zero,
∂0 : C0 (X) −→ 0, then we should define the corresponding zero homology by
setting H0 (X) := C0 (X)/Im ∂0 .
According to this definition, it is obvious that the simplicial space X is
connected if and only if H0 (X) ∼ = Z.
In order to simplify certain formulations, it is convenient to slightly mod-
ify the very end of the chain complex corresponding to a given simplicial
space, replacing the last two terms . . . −→ C1 (X) −→ C0 (X) −→ 0 by the
following three-term sequence
∂ 2 1∂ 0 ∂
. . . −−−→ C1 (X) −−−→ C0 (X) −−−→ C−1 (X) = Z −−−→ 0,

where the homomorphism ∂0 is defined by the formula

X  X
∂0 zi vi = zi .

Then we can define zero homology (now denoted by H e 0 ) on the standard

way by setting He 0 (X) := Ker ∂1 /Im ∂0 . Then the simplicial space X is
connected if and only if He 0 (X) = 0. It is also easy to see that we always
have H0 (X) = He 0 (X) ⊕ Z.
44

5.8. Problems

5.1. Consider the chain complex

∂
4 3 ∂ 2 ∂1 ∂
0 −→ Z −→ Z −→ Z −→ 0

where ∂3 (z) = 2z and ∂2 (z) = 0. Compute its homology.

5.2. Compute the homology of the singleton and the segment [0, 1] regarded
as a simplicial space with one 1-simplex.
5.3. Compute the homology of the subdivided segment (see the figure) by
using only the definition of homology groups and then verify your answer by
using their properties (such as homotopy invariance).

5.4. Compute the homology of the boundary of the triangle and the bound-
ary of the square (by using the definition of simplicial homology groups).
Compare.
5.5. Compute the homology of the boundary of the tetrahedron and the
boundary of the cube (its faces are triangulated by means of their diagonals).
Compare.
5.6. Prove that the induced homomorphism in simplicial homology theory
is well defined, i.e., does not depend on the representative of the orientation.
5.7. Prove the Poincaré lemma in detail (check that the two simplices
[v0 , . . . , v˘j , . . . , v̆i , . . . , vn ] do appear with opposite signs).
5.8. Prove that a simplicial space X is connected if and only if H0 X = Z.
5.9. Compute the homology groups H∗ (M öb, Z) of the Möbius strip directly
from the definition of simplicial homology.
5.10. Compute the homology of RP 2 modulo 2 (i.e., with coefficients in Z2 ).
5.11. Define the homology theory for ordered simplices (in which each geo-
metric n-simplex yields n! ordered simplices) and prove that it is equivalent
to the theory involving all oriented simplices and to the ordered theory, in
which the vertices of the simplicial space are ordered (so that there is only
one ordered simplex corresponding to each geometric simplex).
 45

Lecture 6
PROPERTIES OF SIMPLICIAL HOMOLOGY

The aim of this lecture is to establish some basic properties of the simpli-
cial homology groups. We begin, however, with some algebraic preliminaries
mostly related to chain complexes (short exact sequences of chain complexes
and the corresponding long homology sequences, some auxiliary statements
such as Steenrod’s five-lemma, chain homotopy, etc.). We also discuss the
notion of acyclic support, a useful geometric tool that yields important in-
formation in algebraic topology, e.g. in constructing chain homotopies.
After this is done, it turns out that a series of fundamental results are
obtained without much extra work. They are the homotopy invariance (and
hence the topological invariance) of homology groups, the exact homology
sequence for pairs, the Hurewicz theorem (which establishes a fundamental
relationship between homology and homotopy groups), the Mayer–Vietoris
sequence (which often allows to compute the homology of a space when we
know the homology of its parts).

6.1. Four algebraic lemmas

In the previous lecture, we already mentioned that a sequence of groups
and homomorphisms
hi−1 h hi+1 hi+2
. . . −−−→ Gi−1 −−−i→ Gi −−−→ Gi+1 −−−→ . . .
is called exact at the term Gi if we have Im hi = Ker hi+1 ; the whole sequence
is called exact if it is exact at all terms.
The following properties of exact sequences of groups, which immediately
follow from their definition, hold:
(i) if the sequence 0 → A → B → . . . is exact, then the homomorphism
A → B is a monomorphism;
(ii) if the sequence · · · → A → B → 0 is exact, then the homomorphism
A → B is an epimorphism;
(iii) if the sequence 0 → A → B → 0 is exact, then the homomorphism
A → B is an isomorphism.
A five-term exact sequence 0 → A → B → C → 0 starting and ending in
zero is called short exact.
46

Lemma (Short Exact Lemma). If h : G → H is an arbitrary group

homomorphism, then

i h
0 −−−→ Ker h −−−→ G −−−→ Im h −−−→ 0,

where i is the inclusion homomorphism, is a short exact sequence.

The proof of this lemma is obvious.
Lemma (Splitting Lemma.) The condition that the short exact sequence

ϕ ψ
0 −−−→ A −−−→ B −−−→ C −−−→ 0,

splits, i.e., can be rewritten in the form

i p
0 −−−→ A −−−→ A ⊕ C −−−→ C −−−→ 0,

where i is the natural inclusion and p the projection on the second factor,
is equivalent to the condition that ϕ has a left inverse (i.e., there exists a
homomorphism Φ : B → A such that ϕ ◦ Φ = idB ) as well as to the condition
ψ has a right inverse (i.e., there exists a homomorphism Ψ : C → B such
that Ψ ◦ ψ = idB ).
The proof of this lemma is an (easy) exercise.
The next algebraic lemma is used in homotopy and homology theories to
prove that different spaces have isomorphic homotopy (homology) groups. A
few instances of its application will appear in the exercise class.
Lemma (Steenrod’s Five-Lemma). In the commutative diagram

f g h i
A −−−→ B −−−→ C −−−→ D −−−→ E
    
p q r s 
y y y y yt
f0 g0 h0 i0
A0 −−−→ B 0 −−−→ C 0 −−−→ D0 −−−→ E 0
let the rows be exact and the vertical homomorphisms p, q, s, t be isomor-
phisms. Then r, the middle vertical arrow, is also an isomorphism.
The proof is a nice exercise about abstract groups and diagram chasing.
In fact, the assertion of this lemma holds under weaker conditions. This will
be discussed in the exercise class.
 47

Lemma ((3 × 3)-Lemma). Suppose that in the following commutative

diagram
0 0 0
  
p p0 p00
y y y
f g h i
0 −−−→ B −−−→ C −−−→ D −−−→ 0
  
q q0 q00
y y y
f0 g0 h0 i0
0 −−−→ B 0 −−−→ C 0 −−−→ D0 −−−→ 0
  
r 0  00
y yr yr
f 00 g 00 h00 i00
0 −−−→ B 00 −−−→ C 00 −−−→ D00 −−−→ 0
  
s 0  00
y ys ys
0 0 0
the first and second (second and third) rows and all the columns are exact;
then so is the third (first) row.
The proof is another typical simple example of diagram chasing.

6.2. Constructing long homology sequences.

We have already constructed long exact homotopy sequences in the pre-
vious lectures by ad hoc methods. The next lemma is the main algebraic
tool for constructing long homology sequences of different types in a more
“scientific way”.
Lemma (Short to Long Exact Lemma). If the short sequence of chain
complexes 0 → C → C 0 → C 00 → 0, i.e,
  
  
y y y
0 −−−→ Ci −−−→ Ci0 −−−→ Ci00 −−−→ 0
j
  
∂ ∂ 0 ∂ 00
yi yi yi
0 00
0 −−−→ Ci−1 −−−→ Ci−1 −−−→ Ci−1 −−−→ 0
j
  
  
y y y
48

is exact, then it induces the following long exact sequence in homology:

· · · → Hi (C) → Hi (C 0 ) → Hi (C 00 ) → Hi (C) → · · · → H0 (C 00 ).

Proof. Let us construct the homomorphism Hi (C 00 ) → Hi−1 (C). Pick

some c00i ∈ Ker ∂i00 . It follows from the exactness of the horizontal sequences
that c00i = j(c0i ) for some c0i , and there is an element ci−1 in Ci−1 which is
the preimage of ∂i0 (c0i ). The element c is a cycle in the complex C, and we
take the class corresponding to c in the group Hi−1 (C) to be the image of the
homology class of the cycle c00i . Thus we have constructed a map from Hi (C 00 )
to Hi−1 (C). It is easy to check that this map is a well-defined homomorphism.
The construction of the other homomorphisms in the long sequence and the
verification of its exactness is straightforward diagram chasing.

6.2. Chain homotopy

Two morphisms {fk } and {gk } between two chain complexes {Ck0 } and
00
{Ck } are called chain homotopic if there exists a family of homomorphisms
00
Dk : Ck0 → Ck+1 such that
00
∂k+1 ◦ Dk + Dk−1 ◦ ∂k0 = gk − fk ;

the family {Dk } is then called a chain homotopy between f and g.

Remark. This definition may seem rather strange at first glance. In
order to understand what it means, the reader should consider the case in
which the chain complexes are simplicial ones and try to give a geometric
interpretation of the homomorphisms Dk .
Lemma (Chain Homotopy). If the morphisms {fk } and {gk } between two
00
chain complexes {Ck0 } and {Ck } are chain homotopic, then the corresponding
induced homomorphisms in homology {(fk )∗ } and {(gk )∗ } coincide.
Proof. Let zk ∈ Ck0 be a cycle, i.e., ∂k0 (zk ) = 0. Then
00 00
gk (zk ) − fk (zk ) = ∂k+1 (Dk (zk )) + Dk−1 (∂k0 (zk )) = ∂k+1 (Dk zk ),

which means that gk (zk ) and fk (zk ) are homologous. 

6.3. Acyclic supports

We now move away from the purely algebraic context of chain complexes
back to simplicial spaces. A simplicial space is said to be acyclic if its homol-
ogy is zero in all dimensions n > 0 and H0 (X) = Z. The support of a chain
 49

c ∈ Cn (X) is any simplicial subspace of X that contains all the simplices

which appear in the chain with nonzero coefficients.
For the statement of the next lemma, we will need a technical notion
related to 0-dimensional chains: a chain map f is called augmentation-
preserving if
X  X X X
0
f0 ai ∆i = bj ∆0j with ai = bj .
i i i i

(The motivation for introducing the notion of augmentation was discussed

in the previous lecture.)
Lemma (Acyclic Support Lemma). Suppose X and Y are simplicial
spaces and two augmentation-preserving chain maps

ϕk , ψk : Ck (X) → Ck (Y )

are given. Assume that an assignment A taking each simplex ∆ ⊂ X to a

simplicial subspace A(∆) ⊂ Y is given and it satisfies the conditions
(i) ∆0 ⊂ ∆ implies A(∆0 ) ⊂ A(∆);
(ii) A(∆) is acyclic;
(iii) the set A(∆k ) is the support of both chains ϕk (∆k ) and ψk (∆k ).
Then the maps ϕk and ψk are chain homotopic and so ϕ∗ = ψ∗ .
Proof. We will construct a chain homotopy Dk : Ck (X) → Ck+1 (Y ) by
induction on k. We begin with the case k = 0. Let ∆0 be a vertex of X. The
simplicial space A(∆0 ) supports both chains ϕ0 (∆0 ) and ψ0 (∆0 ). Since the
maps ϕ0 and ψ0 preserve augmentation, we have
X X
(ϕ0 − ψ0 )(a∆0 ) = bi ∆0i , where bi = a.

Since A(∆0 ) is acyclic, the chain bi ∆0i must be the boundary of some
P
1-chain (which we denote D0 (a∆0 )) lying in A(∆0 ), i.e.,

(ϕ0 − ψ0 )(a∆0 ) = ∂1 D0 (a∆0 ).

However, we don’t necessarily have the relation

D0 (a∆0 ) + D0 (b∆0 ) = D0 ((a + b)∆0 ),

but it will hold if we redefine D0 by setting

D0 (a∆0 ) = aD0 (1 · ∆0 ).
50

(Further on, chains of the form 1 · ∆ will be simply denoted by ∆).

To perform the induction step, suppose that the required homomor-
phisms D0 , D1 , . . . , Dk−1 have been constructed, and A(∆i ) supports the
chain Di (∆i ). We must construct a homomorphism Dk : Ck (X) → Ck+1 (Y )
which satisfies the only condition that for any k-dimensional simplex ∆ ⊂ X,

∂k+1 Dk (∆) = ck , where ck = ϕk (∆) − ψk (∆) − Dk−1 ∂k (∆).

Now all the simplices of ∂k (∆) are contained in ∆, and therefore A(∆) sup-
ports the chain ∂k (∆) and so supports the chain Dk−1 ∂k (∆). Thus A(∆)
supports the chain ck and the induction hypothesis implies that

ck = (ψk − ϕk − Dk−1 ∂k )(∆) =

= (ψk − ϕk − (ψk−1 ∂k − ϕk−1 ∂k − Dk−2 ∂k−1 ∂k ))(∆).

The support A(∆) is acyclic, therefore the cycle ck is the boundary of some
chain (which we denote Dk (∆)) supported by A(∆) and satisfying the equal-
ity ∂k+1 Dk (∆) = ck , as required. 

6.4. Homotopy invariance of homology

A simplicial space is, of course, a topological space, and as such we can
speak of its homotopy equivalence to some other space. We will not prove
that simplicial homology theory is a homotopy invariant (in this general
topological sense). It is more in the spirit of the category approach to math-
ematics to give a more combinatorial definition of homotopy and homotopy
equivalence for simplicial spaces and then prove that homology is homotopy
invariant in the combinatorial sense (often also called piecewise linear). This
is the goal of this section.
A continuous map f : X → Y of simplicial spaces is called piecewise
linear (briefly, PL) if it is simplicial in some subdivision of the simplicial
structures of X and Y . Two PL maps f, g : X → Y are called PL-homotopic
if there exists a PL-map F : X × [0, 1] → Y such that

F (x, 0) = f (x), F (x, 0) = g(x) for all x ∈ X;

here the Cartesian product X × [0, 1] is supplied with a PL-structure in the

natural way (in particular, the simplicial structure on X × {0} and X × {1}
is the same as that on X and for any simplex ∆ ⊂ X the set ∆ × I is a
simplicial subset of X × [0, 1]).
 51

Theorem. PL-homotopic PL-maps f, g : X → Y of simplicial spaces

induce the same homomorphism in homology.
Proof. Let F denote a homotopy between f and g. Consider the inclu-
sions

i0 : X ,→ X × {0} ⊂ X × I and i1 : X ,→ X × {1} ⊂ X × I.

We obviously have f = F i0 and g = F i1 , hence it suffices to prove that

i0∗ = i1∗ . Let ∆k be a simplex in X. The chains i0 (∆k ) and i1 (∆k ) have
the same support ∆k × I, which is obviously acyclic. By the Acyclic Support
Lemma, this means that there is chain homotopy joining i0∗ and i1∗ , and so
by the Chain Homotopy Lemma i0∗ = i1∗ . 
Two simplicial spaces X and Y are called PL-homotopy equivalent if there
exist simplicial subdivisions of X and Y and simplicial maps (w.r.t. these
subdivisions) f : X → Y and g : Y → X such that f ◦ g and g ◦ f are
PL-homotopic to the identity maps of Y and X, respectively. The following
corollary of the previous theorem is a direct consequence of the functoriality
of simplicial homology.
Corollary. Simplicial homology groups are invariants of PL-homotopy
equivalence.
Two simplicial spaces X and Y are called PL-homeomorphic or PL-equi-
valent if there exist simplicial subdivisions of X and Y and a simplicial
(w.r.t. these subdivisions) homeomorphism of X onto Y . The fact that PL-
equivalence (obviously) implies homotopy equivalence yields the next corol-
lary.
Corollary. Simplicial homology groups are invariants of PL-equivalence.
Remark. At this stage, with the use of the simplicial approximation
theorem and some extra technical efforts (not involving any new ideas), it
is possible to prove the (stronger) purely topological versions of the above
theorem and its corollaries. We omit these proofs mainly for aesthetic rea-
sons, referring the reader to V.Prasolov’s book Elements of Homology The-
ory, Chap.1, Sect. 2. (Part of the above exposition follows that book rather
closely, including the notation.)

6.5. Exact homology sequence for pairs

For the homology groups of pairs of simplicial spaces, we have an exact
sequence very similar to the one for homotopy groups.
52

Theorem (Homology sequence for pairs). For any simplicial pair (X, A),
we have the following exact sequence:
i ∗ j∗ ∗ ∂ ∗ i
. . . −−−→ Hn (X) −−−→ Hn (X, A) −−−→ Hn−1 (A) −−−→ Hn−1 (X)
j ∂n−1 i j
−−−→ Hn−1 (X, A) −−−→ . . . −−−→ H0 (X) −−−→ H0 (X, A).
Proof. Given a simplicial pair (X, A), consider the three chain complexes
C(A), C(X), C(X, A). There is a morphism of the first complex to the second
one (inclusion), and of the second one to the third (factorization), yielding
the sequence of chain complexes 0 → C(A) → C(X) → C(X, A) → 0, which
can be written in more detail as
  
  
y y y
0 −−−→ Cn (A) −−−→ Cn (X) −−−→ Cn (X, A) −−−→ 0
j
  
∂ ∂ ∂
y∗ y∗ y∗
0
0 −−−→ Cn−1 (A) −−−→ Cn−1 (X) −−−→ Cn−1 (X, A) −−−→ 0
j
  
  
y y y

But this is a short exact sequence of chain complexes, and it gives the required
long exact sequence by the Short to Long Exact Lemma. 

6.6. Mayer–Vietoris sequence

The exact sequence in question, known as the Mayer–Vietoris sequence,
although it was actually first discovered by M.F.Bokstein, is a relationship
between the homology of the union of two spaces and the homology of each
of the spaces and that of their intersection.
Theorem (Mayer–Vietoris). Let X1 and X2 be simplicial subspaces of
the simplicial space X such that X = X1 ∪ X2 ; denote Y = X1 ∩ X2 . Then
we have the following exact sequence:
∂ ∗ i
∗ j∗ ∗ ∂
. . . −−−→ Hn (Y ) −−−→ Hn (X1 ) ⊕ Hn (X2 ) −−−→ Hn (X) −−−→ ...,
Proof. We obviously have

C(Y ) = C(X1 ) ∩ C(X2 ) and C(X1 ) + C(X2 ) = C(X1 ) ∪ C(X2 )

 53

We have the inclusion maps i1 : Y ,→ X1 and i2 : Y ,→ X2 , as well as

j1 : X1 ,→ X and j2 : X2 ,→ X. For c ∈ C(Y ) put i(c) := (i1 (c), −i2 (c)), and
put j(c1 , c2 ) = j1 (c1 ) + j2 (c2 ) for c1 ∈ C(X1 ) and c2 ∈ C(X2 ). Thus we obtain
the following short sequence of chain complexes
i j
0 −−−→ C(X1 ∩ X2 ) −−−→ C(X1 ) ⊕ C(X2 ) −−−→ C(X1 ∪ X2 ) −−−→ 0.

Now it is easy to see that this short sequence is exact. Using the Short to Long
Exact Lemma, we obtain a long homology sequence, which is exactly the one
appearing in the statement of the theorem. 

6.7. Hurewicz theorem

The Hurewicz theorem describes an important relationship between the
n-dimensional homology and homotopy groups. For n = 1, it was actually
first discovered by Poincaré, and asserts that H1 (X, Z) is simply the abelian-
ization of the fundamental group π1 (X, p).
Let us denote the commutator subgroup of π1 by

[π1 (X, p), π1 (X, p)] := {aba−1 b−1 | a, b ∈ π1 (X, p)}.

Theorem (Poincaré–Hurewicz). The first homology group H1 (X, Z) of

a connected simplicial space X is isomorphic to the abelianization

π1 (X, p)/[π1 (X, p), π1 (X, p)]

of the fundamental group π1 (X, p) of X for any basepoint p.

Proof. To each element α ∈ π1 (X, p) let us assign an element of H1 (X).
By the simplicial approximation theorem, there is a simplicial path a ∈ α
(in some triangulation of X). Denote by h(α) the oriented chain obtained
by assigning the coefficient +1 to each oriented 1–simplex of the path a and
0 to all the other 1-simplices. Now assign the homology class of h(a) to the
element α ∈ π1 (X, p). It is not difficult to check that if we replace α by an-
other element of the coset modulo the commutator subgroup of π1 , we obtain
the same 1-homology class. Moreover, it turns out that this assignement is
an isomorphism of the abelianization of π1 (X) onto H1 (X). 
This isomorphism is called the Hurewicz isomorphism. It is a particular
case of the Hurewicz homomorphism in dimension n (n ≥ 1), which is defined
as follows.
54

Let σ : Sn → X be a spheroid in a path connected simplicial space X with

basepoint x0 ∈ X. Then, by the simplicial approximation theorem, the map
σ can be approximated by a simplicial map σ̄, which determines the same
element of πn (X, x0 ). Consider the chain cσ ∈ Cn (X; Z) in which all the
n-simplices of the image σ̄(Sn ) appear with coefficient 1, while all the other
simplices, with coefficient 0. It is not hard to verify that this chain is a cycle
that determines a well-defined homomorphism γ : πn (X, x0 ) → Hn (X, Z);
the assignment γ is known as the Hurewicz homomorphism.
Theorem (Hurewicz). Suppose X is a simplicial space such that

π0 (X) = π1 (X) = . . . πn−1 = 0, n ≥ 2.

Then there exists an isomorphism h : πn (X) → Hn (X).

Proof. To construct h, let x0 be some point of X (which is path connected
because π0 (X) = 0) and let α : (S n , s0 ) → (X, x0 ) be a spheroid. By the
simplicial approximation theorem, we can assume that α is simplicial, and so
it induces a homomorphism in simplicial homology α∗ : Hn (Sn ) → Hn (X).
But we know that Hn (Sn ) ∼ = Z, where 1 ∈ Z corresponds to the identity map
of the n-sphere. Now we can define h by putting h(α) := α∗ (id).
The fact that h is well defined is obvious, while the proof of its injectivity
and surjectivity is a mildly difficult problem for the reader (see the exercise
class). The map h is also known as the Hurewicz isomorphism.
 55

6.8. Problems

6.1. Prove the Splitting Lemma.

6.2. Prove the Five-Lemma.
6.3. Prove the strong form of the Five-Lemma, in which instead of the
commutativity of the diagram we require its commutativity up to sign, e.g.
q ◦ f = ±f 0 ◦ p, etc.
6.4. Prove the 3-by-3 Lemma.
6.5. Prove the Short-to-Long Exact Lemma.
6.6. Suppose f : (X, A) → (Y, B) is a simplicial map such that the maps
f |X and f |A are homotopy equivalences. Prove that Hn (X, A) ∼
= Hn (Y, B).
6.7. Give an example of two simplicial maps for which there is no acyclic
carrier. Give a geometric explanation of the fact that there is no acyclic
carrier in your example.
6.8. Compute the homology of the n-sphere by using the Mayer–Vietoris
sequence.
6.9. Compute the homology of the 2-torus T2 by using the homotopy invari-
ance of homology and the Mayer–Vietoris sequence.
6.10. Compute the 1-homology of the wedge of two circles knowing that
π1 (S1 ∨ S1 ) = Z ∗ Z (here ∗ stands for the free product).
6.11. Give an example of a Hurewicz homomorphism in dimension n ≥ 2
which is not injective.
6.12. Give an example of a Hurewicz homomorphism in dimension n ≥ 2
which is not surjective.
56

Lecture 7
SINGULAR HOMOLOGY

In this lecture, we develop singular homology, which is a very general

theory (it deals with arbitrary topological spaces and continuous maps) and
is as simple to construct as simplicial homology. (For this reason, many
books on homology theory start with an exposition of singular homology.)
To my mind, the main defect of that approach is that within its framework,
at first, meaningful examples of computations based on the basic definitions
are practically impossible; also, it hides the geometric meaning of homology.
We will learn, however, that singular homology satisfies a series of con-
ditions (known as the Steenrod-Eilenberg axioms) which actually uniquely
determine homology theories. The corresponding uniqueness theorem im-
plies that the singular theory yields the same homology groups and the same
induced homomorphisms as other homology theories, in particular simplicial
and cellular homology, so that the results of computations can be borrowed
from the previous lectures and exercises.

7.1 Main definitions and constructions

Let X be an arbitrary topological space. Let ∆n = [0, 1, . . . , n] be the
standard n-simplex, i.e., the convex hull of the set consisting of the origin
0 and the extremities (denoted by 1, . . . , n) of the basis vectors of Rn . We
denote the face of ∆ opposite to the ith vertex by
 
∆hii := 0, . . . , (i − 1), (i + 1), . . . , n , i = 0, . . . , n.

A singular n-simplex σ is any continuous map σ : ∆n → X. The set of all

singular n-simplices is denoted by Σn . Let G be a commutative ring with
unit. By a singular n-chain we mean any finite formal linear combination of
singular n-simplices with coefficients from G, and write
n X o
Cn (X) = c = gα σα , | gα ∈ G .
σα ∈Σn

The setP
Cn (X) has a natural
P 0 G-module structure, with the sum of two
0
chains c = gα σα and c = gα σα defined by
X
c + c0 := (gα + gα0 )σα
 57
P
and multiplication by a constant λ ∈ G by λc = (λgα )σα .
Next we define the boundary operator ∂n : Cn (X) → Cn−1 (X) on each
simplex σ by setting
n
X
∂n (σ) = (−1)k σhki,
k=0

where σhki := σ|∆hki denotes the restriction of σ to the kth face of ∆ (which
is of course a singular (n − 1)-simplex); then we extend ∂n to the whole group
Cn (X) by linearity.
Doing this for all n ≥ 0, we obtain a sequence of Abelian groups (in fact,
G-modules) and homomorphisms

∂n+1 n∂ ∂n−1 1 ∂
. . . −−−→ Cn (X) −−−→ Cn−1 (X) −−−→ . . . −−−→ C0 (X),

called the complex of singular chains of X and denoted C = (Cn , ∂n ) .
Lemma (Poincaré) The boundary operator of singular chains satisfies

∂n−1 ◦ ∂n = 0 for all n ≥ 2,

so that the complex of singular chains is a chain complex.

Proof. By linearity, it suffices to prove that ∂n−1 (∂n (σ)) = 0 for any
simplex σ. We have
n
X  X
(−1)k σhki = (−1)k ∂n−1 (σhki) = 0,


∂n−1 ∂n (σ) = ∂n−1
k=0 k,l

because each singular (n − 2)-simplex σhk, li (obtained from σ by restricting

to the (n − 2) face of ∆n which does not contain the kth and the lth vertices)
in the last sum occurs twice with opposite signs. 
From the given
topological space X, we have obtained a chain complex
C(X) = Cn (X), ∂n , so now we can define the nth homology group of X as
the nth homology group of this chain complex (see Lecture 5, Sect. 5.2), i.e.,
by setting (for any n ≥ 0)

Hn (X; G) := Hn (C(X)) = (Ker ∂n ) (Im ∂n+1 )
58

Now suppose we are given a continuous map f : X → Y of topological

spaces. Our goal is to construct homomorphisms f∗ (induced by f ) of the
corresponding homology groups. We begin by doing it on the chain level:
X  X
f∗n : Cn (X) → Cn (Y ), f∗n (c) = f∗n zk σk := zk (f ◦ σ);
k k

note that the composition f ◦ σ is of course a singular n-simplex.

Lemma (Chain Morphism Lemma). The nth boundary operator com-
mutes with the chain maps induced by f in the sense that

∂Y,n ◦ f∗n = f∗(n−1) ◦ ∂X,n .

Proof. The proof of the lemma is a straightforward verification of defini-

tions; we omit it. 
The above lemma means that we have constructed a functor from the
category of topological spaces and their continuous maps to the category
of chain complexes and their morphisms. Now we can define not only the
singular homology groups of topological spaces but also their induced ho-
momorphisms via the homology of chain complexes (Lect.5, Sect.5.2). The
functoriality of the obtained assignment T op GGr means that


(f ◦ g ∗ = f∗ ◦ g∗ and (idX )n∗ = idHn (X) .

Now let us consider pairs of topological spaces (X, A) and their maps.
By a map (X, A) → (Y, B) of such pairs we mean a continuous map f from
X to Y such that f i(A) ⊂ B; we write f : (X, A) → (Y, B) in that case.
Any pair (X, A) of topological spaces defines the
corresponding relative
singular chain complex C(X, A) = Cn (X)/Cn (A), ∂n (here by abuse of no-
tation we write ∂n for the boundary operator in C(X), which is well defined
on cosets mod Cn (A)). The homology of this chain complex is called the
relative singular homology of the pair (X, A). We write
∞
M ∞
M
H∗ (X, A; G) = Hn (X, A; G) := Hn (C(X, A)).
n=0 n=0

The relative homology group Hn (X, ∅; G), which can be identified with
Hn (X; G), is sometimes called the absolute homology group of X.
 59

Given a map of pairs f : (X, A) → (Y, B), the corresponding induced

homomorphisms f∗n of relative chains and relative homology groups (denoted
f∗ ) is defined similarily to the “absolute” induced homomorphisms. As above,
we obtain a functor from the category of pairs of topological spaces to the
category of chain complexes and, therefore, to the category of graded Abelian
groups. We omit the obvious details.

7.2 Main properties (the Steenrod–Eilenberg axioms)

We now establish the main properties of singular homology theory. It
will turn out that these properties uniquely determine the singular homology
functor, so they may be regarded as axioms for homology theories (they are
known as the Steenrod-Eilenberg axioms).
But first we summarize what was done in the previous section. To ev-
ery topological space, every pair of topological spaces, and their continuous
maps, we assigned, for each nonnegative integer n, Abelian groups (called
homology groups) and group homomorphisms, and to each pair of spaces
(X, A), a homomorphism ∂∗ of the nth homology group of the pair (X, A) to
the (n − 1)st homology group of A. In the notation introduced above, this
correspondence reads as

X 7→ Hn (X), n = 0, 1, 2, . . .
(X, A) 7→ Hn (X, A), n = 0, 1, 2, . . .
(X, A) 7→ ∂∗ : Hn (X, A) → Hn−1 (A), n = 1, 2, . . .
f :X→Y 7→ f∗ : Hn (X) → Hn (Y ), n = 0, 1, . . .
f : (X, A) → (Y, B) 7→ f∗ : Hn (X, A) → Hn (Y, B), n = 0, 1, . . .

The sum over n of the obtained Abelian groups is denoted

∞
M ∞
M
H∗ (X) := Hn (X) and H∗ (X, A) := Hn (X, A);
n=0 n=0

these objects will be called graded homology groups of X and (X, A).
Theorem. The correspondences described above define a covariant func-
tor (called singular homology) from the category T op of topological spaces to
that of graded Abelian groups, and this functor has the following properties.
(I) Dimension: H0 (pt) = G, where pt is the singleton and G is an Abelian
group, and Hn (pt) = 0 for n > 0.
60

(II) Commutation: the following square diagram

Hn (X, A) −−−→ Hn (Y, B)

f∗
 

∂∗ y

∂∗ y

Hn−1 (A) −−−→ Hn−1 (B).

(f |A )∗

is commutative.
(III) Homotopy invariance: if two maps f, g : (X, A) → (Y, B) are homo-
topic, then f∗ = g∗ , the induced homomorphisms coincide in all dimensions,
and so homotopy equivalent spaces have the same homology.
(IV) Exactness: For any pair of spaces (X,A), the following sequence is
exact:
i ∗ j∗ ∗ ∂
∗ i
. . . −−−→ Hn (X) −−−→ Hn (X, A) −−−→ Hn−1 (A) −−−→ Hn−1 (X)
j∗ ∗ ∂ ∗ i j∗
−−−→ Hn−1 (X, A) −−−→ . . . −−−→ H0 (X) −−−→ H0 (X, A).
(V) Excision: Suppose that U ⊂ X is an open subset whose closure lies in
the interior of A, where A ⊂ X. Then the inclusion (X \ U, A \ U ) ,→ (X, A)
induces an isomorphism in homology Hn (X \ U, A \ U ) → Hn (X, A) in all
dimensions n.
Proof. Property (I) is obvious.
Property (II) follows by a straightforward verification of definitions.
Let us prove (III) in the particular case A = B = ∅ (the general case is
quite similar). Given homotopic maps f, g : X → Y , we shall construct a
chain homotopy
Dk = Ck (X) → Ck+1 (Y );
this will imply f∗ ≡ g∗ by the Chain Homotopy Lemma. Let H : X × I → Y
be a homotopy between f and g. To construct Dk , consider a singular simplex
σ : ∆k → X and the Cartesian product ∆k × I =: T . The set T has a
canonical triangulation consisting of (k + 1)-simplices all of whose vertices
lie in ∆k × {0} and ∆k × {1}. For k = 1 and k = 2, the simplices are shown
in Fig.1 below.
For an arbitrary k, we denote by 0, 1, . . . k the vertices of ∆k , by 00 , 10 , . . . k0
and 01 , 11 , . . . k 1 those of ∆k ×{0} and ∆k ×{1}, respectively. Then a generic
(k + 1)-simplex of T is of the form [00 , . . . j0 , j 1 , . . . k 1 ]; note that the number
 61

j of the last vertex lying in ∆k × {0} is the same as that of the first vertex
lying in ∆k × {1}.

1 1

0 0
∆ ∆
Figure 7.1. Triangulations of ∆k × I

Now we consider the homotopy S : ∆k × I → Y , S(x, t) := H(f (x), t)

and define Dk on the singular simplex σ by setting
k
X
Dk (σ) := (−1)j S(δjk+1 ),
j=0

where δjk+1 is the linear map from the standard simplex

∆k+1 = [0, 1, . . . , k + 1]

that takes j and (j + 1) to τ (j)0 and τ (j)1 , respectively. Thus the right-
hand side of the previous displayed formula is a chain in Ck+1 (Y ). Having
constructed Dk on an arbitrary singular k-simplex of X, we extend it by
linearity to chains from Ck (X).
The fact that this construction indeed yields a chain homotopy, i.e., that

∂k+1 Dk + Dk+1 ∂k = g∗ − f∗ ,

can be established by a tedious but straightforward verification.

Finally, the isomorphism of the corresponding homology groups of ho-
motopy equivalent spaces follows from what was just proved and from the
functoriality of our construction.
To prove exactness (IV), let us consider the short sequence of chain com-
plexes
i∗ p∗
0 −−−→ C(A) −−− → C(X) −−−→ C(X, A) −−−→ 0,
62

where i∗ is the inclusion homomorphism and p∗ is the homomorphism ob-

tained by killing (i.e., supplying with zero coefficients) all singular simplices
entirely contained in A. This sequence is obviously exact. Using the Short-
to-Long Exact Lemma, we obtain the required long homology sequence,
The proof of (V) (excision) is rather technical (it is based on iterated
baricentric subdivisions of standard simplices and a kind of chain homotopy
construction); see V.Prasolov, Elements of Homology Theory, pp.198-200.
We omit it. 

7.3 Uniqueness Theorems

There are many (co)homology theories based on completely different ap-
proaches. Besides cellular, simplicial, and singular, there is Vietoris homol-
ogy (for metric spaces), Čech homology (for topological spaces), Dowker
homology (for arbitrary relations), de Rham cohomology (for smooth mani-
folds), etc. To give an answer to the natural question: (i.e., do these different
approaches yield the same groups and homomorphisms?) one would like to
have an axiomatic characterization of homology theory functors.
Such a characterization was obtained by N.Steenrod and S.Eilenberg in
the 1960ies and consists of five axioms. These axioms coincide with the five
properties of singular homology theory appearing in the theorem proved in
the previous section and are known as the Steenrod-Eilenberg axioms. Using
them, one can state and prove different uniqueness theorems, such as
• On the category of topological spaces and maps, a functor satisfying
axioms (I) – (V) is unique in the sense that it produces the same groups
H∗ (X, A) and the same induced homomorphisms f∗ as the singular homology
functor constructed in the previous section.
• On the category of simplicial spaces and maps, a functor satisfying
axioms (I) – (V) is unique in the sense that it produces the same groups
H∗ (X, A) and the same induced homomorphisms f∗ as the simplicial homol-
ogy functor described in the previous lecture.
We shall not prove these (or any other) uniqueness theorems, but let us
briefly sketch the idea of the proof of the second one for finite simplicial spaces
(i.e., simplicial spaces consisting of a finite number of simplices). Using the
dimension axiom, we can write Hn (pt) = G. Since the n-simplex can be
contracted to a point, homotopy invariance implies that its homology is the
same as that of the point. Using induction and the Mayer–Vietoris sequence,
it is not difficult to find the homology of the n-sphere. Given a finite simplicial
 63

space X, we inductively construct it by pasting together simplices of different

dimensions one after another, computing the homology along the way; in the
process, we repeatedly use the Mayer-Vietoris sequence (see below).
Remark 1. It immediately follows from the above uniqueness theorem
that simplicial homology is homotopy invariant in the topological sense, i.e.,
any homotopic continuous maps induce the same homomorphism in homol-
ogy. (Recall that in the previous lecture we proved, for simplicial homology, a
weaker form of homotopy invariance; namely, we established invariance only
for PL-homotopic maps.) Now we can assert in particular that the simplicial
homology groups do not depend on the triangulation of the given simplicial
space and homeomorphic simplicial spaces have the same homology.
Remark 2. The first of the Steenrod–Eilenberg axioms (which asserts
that the n-homology of the point is zero for n > 0 and seems rather trivial)
is actually of great importance: by replacing it by different other statements
one obtains such important theories as extraordinary homology, K-theory,
and (co)bordism theory (which are outside the scope of this course).

6.6. Mayer–Vietoris sequence in singular homology

In singular homology, we also have a Mayer–Vietoris sequence, but its
formulation is more delicate (see Problem 7.3) and requires an technical
condition concerning the subsets X1 , X2 ⊂ X, X1 ∪ X2 = X. We say that
such a pair satisfies the excision condition if the natural chain map of the
chain group C∗ (X1 ) + C∗ (X2 ) (consisting of sums of chains from X1 and X2 )
to C∗ (X1 ∪ X2 ) induces an isomorphism in homology.
Theorem (Mayer–Vietoris). Let X1 and X2 be subspaces of the topo-
logical space X whose union is X and which satisfy the excision condition
for pairs. Then we have the following exact sequence:
∂ ∗ ∗i j∗ ∗ ∂
. . . −−−→ Hn (X1 ∩ X2 ) −−−→ Hn (X1 ) ⊕ Hn (X2 ) −−−→ Hn (X) −−−→ ...,

Proof. The argument is similar to the proof in the simplicial case (it is
based on the Short-to-Long Exact Lemma), except that the third term of the
short exact sequence is C∗ (X1 ) + C∗ (X2 ) rather than C(X1 ∪ X2 ). 
64

7.4. Problems

7.1. Prove the Chain Morphism Lemma.

7.2. Give an example showing that the following version of the excision
property does not hold: suppose that U ⊂ X is an open subset that lies in
A, where A ⊂ X; then the inclusion (X \ U, A \ U ) ,→ (X, A) induces an
isomorphism in homology Hn (X \ U, A \ U ) → Hn (X, A) in all dimensions n.
7.3. Show that the Mayer–Vietoris sequence in singular homology does not
hold if we only assume that Y = X1 ∩ X2 .
7.4. Prove that H e k (Sn ; Z) = 0 for k < n and is isomorphic to Z for k = n;
here He k is the reduced singular homology group.
7.5. Prove that Hk (Dn , ∂Dn ; Z) = 0 if k 6= n and is isomorphic to Z other-
wise.
7.6. Construct and justify the suspension isomorphism Hk (Σ) ∼ =H e k−1 (X).
7.7. Prove that Hk (X, A) ∼
= Hk (X ∪ CA, CA), where CX (CA) denotes the
cone over X (over A).
7.8. Prove that Hk (X, A) ∼= Hk (X ∪ CA) for k > 0; here CA denotes the
cone over A.
7.9. Suppose that X is a connected CW-space and A is its CW-subspace.
Prove that Hk (X.A) ∼
=He k (X/Y ).
7.10*. Suppose A ⊂ Rn is closed and does not coincide with Rn ; then
e k+i (Rn+i \ A) ∼
H =He k (Rn \ A); here Rn is naturally embedded in Rn+i .
7.11. Prove the following theorem (due to Alexander): if A and B are
homeomorphic closed sets in Rn , then Hk (Rn \ A) ∼
= Hk (Rn \ B).
7.12. Prove the Jordan–Brouwer theorem: if A ⊂ Rn is homeomorphic to
the (n − 1)-sphere, then Rn \ A consists of two connected components.
7.13. Show that the excision axiom does not hold for homotopy groups.
 65

Lecture 8
APPLICATIONS OF HOMOLOGY

In this chapter, we study some simple applications of homology theory,

mostly to simplicial spaces, in particular smooth manifolds. We begin by
giving a geometric interpretation of the homology groups in the lowest (i.e.,
zero) and highest (i.e., n for n-manifolds) dimensions, namely connectedness
and orientability. We learn how to decompose integer homology groups of
finite simplicial spaces X into the direct sum of b ∈ N copies of Z (b is
the so-called Betti number) and a finite Abelian group (called the torsion
group of X). We then investigate the Euler characteristic and learn that it
is in fact a deep homological parameter of arbitrary spaces, including those
not possessing any triangulation. Finally, we prove the Lefschetz fixed point
theorem, one of the most important applications of homology theory.
In this lecture, it will be convenient for us to move from singular to
simplicial homology and back depending on the context. By the uniqueness
theorem, the results remain valid in either one of the theories (and in fact in
the other homology theories as well).

8.1. Connectedness
Homology theory gives a simple characterization of path connectedness
for arbitrary topological spaces and indicates the number of path connected
components of arbitrary topological spaces.
Theorem. A topological space X is path connected if and only if the
0-homology group H0 (X; Z) is isomorphic to Z; this condition can be replaced
by H0 (X; G) ∼= G provided that the coefficient group G is a ring with unit.
Proof. To prove necessity, choose a basepoint p in X. Then any other
point q can be joined to p by a path, this path may be regarded as a singular
simplex ∆1 ; then the 1-chain 1 · ∆1 has the boundary q − p, so that all
points (regarded as 0-chains) are homological to p, and therefore p generates
H0 (X) ∼= Z. The proof of sufficiency is left for the exercise class. 
Corollary. The number of path connected components of an arbitrary
topological space X is equal to the dimension of the linear space H0 (X; R).
Proof. Because of the previous theorem, it suffices to prove that no
two vertices located in different components are homologous. Assuming the
converse, one can easily obtain a path joining the two vertices. 
66

8.2. Orientability
Let M n be a triangulated n-dimensional manifold, i.e., a simplicial space
each point of which possesses a neighborhood homeomorphic to Rn . We
assume known that any smooth manifold has a triangulation and so does
any topological manifold of dimension two or three (the last fact is actually
a very difficult theorem proved by Edwin Moise in the 1940ies). We say that
M n is oriented if a coherent orientation of its n-simplices is given, i.e., all its
n-simplices are oriented in such a way that the two orientations induced on
any (n − 1)-simplex by the two adjacent n-simplices are opposite. If such an
orientation exists, the manifold is called orientable. It is easy to see that if
M n is connected, then there are only two possible choices of orientation on
M n.
The simplest example of a non-orientable manifold is the projective plane
RP 2 , while the n-sphere Sn is, of course, orientable.
Theorem. A connected triangulated n-dimensional manifold M n is ori-
entable if and only if its n-homology group Hn (M n ; Z) is isomorphic to Z.
Proof. To prove the “if part”, consider a coherent orientation of all the
n-simplices of M n and take the chain c1 ∈ Cn (M n ; Z) with coefficient +1 on
each of them. This chain is obviously a cycle, and so is any chain obtained
by replacing the plus ones by a fixed integer z ∈ Z. There are no other
cycles, because if the chain has non-constant coefficients, then (using the
fact that Mn is connected) we can find two adjacent n-simplices with different
coefficients; in that case the boundary of the cycle will be nonzero on their
common face. The cycle c1 clearly generates the group Hn (M n ; Z) ∼ = Z.
The “only if” part will be proved in the exercise class. 
For an orientable smooth n-manifold M , the choice of the generator of
Hn (M ) ∼= Z defines one of the two orientations on M ; this generator is called
the fundamental class of M . If M and N are both oriented and f : M → N
is a map, then the image under f∗ of the fundamental class ϕ ∈ Hn (M ) is of
the form d  ψ, where ψ ∈ Hn (N ) is the fundamental class of N and d is an
integer; this integer is called the degree of f and denoted deg(f ). The fact
that in the case M = N = Sn this definition agrees with the geometric one
for neat maps given in Lecture 4 will be proved in the exercise class.

8.3. Betti numbers and torsion subgroup

In what follows, we shall need a classical algebraic fact about Abelian
groups, which we now state without proof.
 67

Algebraic fact. Any finitely generated Abelian group G can be ex-

pressed in the form
Mb
G= Z(i) ⊕ T,
i=1

where b ∈ N is a nonnegative integer, called the rank of G, the Z(i) are copies
of Z, and T is a finite Abelian group called the torsion of G.
The above algebraic fact immediately implies the following theorem.
Theorem. If X is a finite simplicial space, then its integer homology can
be expressed as follows:
b
M
Hk (X; Z) = Z(i) ⊕ T,
i=1

where b ∈ N is a nonnegative integer, called the kth Betti number of X, the

Z(i) are copies of Z, and T is a finite Abelian group called the kth torsion
group of X.
Proof. The chain group Ck (X; Z) is finitely generated (by the k-simplices
of X) and therefore so are the group of cycles and boundaries in dimension k.
Hence Hk (X; Z) is finitely generated (and Abelian), so the assertion follows
from the previous algebraic fact. 

8.1. Euler characteristic

The Euler characteristic of a finite n-dimensional simplicial space is the
alternated sum of the number ki of its simplices in each dimension i, i.e.,

χ(X) := k0 − k1 + k2 − · · · + (−1)n kn .

This integer invariant is the oldest one in topology, it was actually first in-
vented and computed by Descartes (and only a century later by Euler) in
the case of convex polyhedra. Apparently Riemann was the first to general-
ize it to other simplicial spaces, in particular to two-dimensional manifolds
(Riemann surfaces). The reader should know that χ(M ) classifies orientable
2-manifolds M .
Consider the sequence
i ∂
0 −−−→ Zk (X; F) −−−→ Ck (X; F) −−−→ Bk−1 (X; F) −−−→ 0,
68

where F is a field and Ck , Zk , Bk−1 are the chain, cycle, and boundary groups,
respectively; its exactness implies that dim Zk + dim Bk−1 = dim Ck . On the
other hand, we obviously have dim Ck = ak . Therefore
X X X
χ(X) = (−1)k Ck = (−1)k Zk + (−1)k Bk−1 =
X
X
= (−1)k dim Zk − dim Bk−1 = (−1)k dim Hk (X; F).

because dim Hk (X; F) = Zk /Bk−1 .

The case of integer homology is similar and is left as an exercise. 
Corollary. The Euler characteristic is homotopy (and therefore topologi-
cally) invariant. In particular, it does not depend on the triangulation of the
given space.
We can now define the Euler characteristic of any topological space X
with finitely generated homology H∗ (X; R) over the field R, e.g. of any
smooth compact n-manifold, as the alternated sum of its Betti numbers (see
the displayed formula above), thus obtaining a simple but very deep integer
invariant of that space.

8.4. Lefschetz fixed point theorem

Suppose X is a finite simplicial space and f : X → X is a continuous
map. Then the induced map f∗ : Hk (X; R) → Hk (X; R) is a linear operator
on a finite-dimentional vector space over the field R and we can consider its
trace tr(fk∗ ). We now define the Lefschetz number Λ(f ) by setting
X
Λ(f ) := (−1)k tr(f∗ ). (1)
k≥0

It follows immediately from the definitions that the Lefschetz number is a

generalization of the Euler characteristic, namely the Euler characteristic is
the Lefschetz number of the identity map: Λ(idX ) = χ(X).
The above definition of the Lefschetz number involves the induced ho-
momorphism fk∗ in homology, but we can also consider the homomorphism
fk : Ck (X; R) → Ck (X; R) on the chain level; fk is a linear operator on the
space Ck (X; R), which is the vector space generated by the (finite) family
∆1 , . .. , ∆Nk of k-simplices of X; the matrix ((fij )) of this operator in the
basis ∆i has a simple geometric meaning (when f is a simplicial map): fij
equals ±1 if ∆i is mapped onto ∆j with the same (opposite) orientation and
 69

fij equals 0 if ∆i is not mapped onto ∆j . Now for a simplicial map f : X → X

we can replace f∗ by fk in definition (1), but it turns out (surprisingly!) that
the result will be the same, as the following lemma asserts.
Lemma [Hopf]. Suppose X is a finite simplicial space and a chain map
fk : Ck (X; R) → Ck (X; R) is given; then
X X
(−1)k tr(f∗ ) = (−1)k tr(fk ). (2)
k≥0 k≥0


Proof. Consider the chain complex Ck (X; R), ∂k ; denote Zk := Ker ∂k
and Bk := Im∂k+1 . For an appropriate subspace C bk of the vector space Ck ,
we can write Ck = Zk ⊕ Cbk . The operator fk maps Zk to itself and therefore
bk → C
we have a linear operator fbk : C bk such that

trfk = tr(fk |Zk ) + trfbk . (3)

For an appropriate subspace Zbk of the vector space Zk , we can write

Zk = Bk ⊕ Zbk . Then obviously Zbk ∼= Hk (X; R), and the operator induced by
fk on Hk (X; R) coincides with f∗ . Therefore we have

tr(fk |Zk ) = tr(fk |Bk ) + trfb∗ . (4)

Now the fact that ∂k : Ck → Bk−1 is a boundary operator of a chain complex

shows that we have an isomorphism Cbk → Bk−1 and that fbk actually coincides
with fk−1 . Combining this with equations (3) and (4), we obtain

tr(fk ) = tr(fk |Bk ) + tr(f∗ ) + tr(fk−1 |Bk−1 ).

Summing this over k with alternating signs, we obtain (1), because the
first and third sum in the right-hand side cancel each other out.
Theorem (Lefschetz). A continuous map f : X → X of a finite simplicial
space to itself with nonzero Lefschetz number Λ(f ) 6= 0 has a fixed point.
Proof. Suppose that the map f has no fixed points. By compactness,
there is a positive lower bound for the distance between points and their
images. Therefore, if we take the iterated baricentric subdivision of X a
sufficient number of times (denoting by X 0 the obtained simplicial space)
and consider the simplicial approximation ϕ : X 0 → X 0 of f , we can assume
that ∆0 ∩ ϕ(∆0 ) = ∅ for any simplex ∆0 ∈ X 0 . But then the diagonal of the
70

matrix of the linear operator ϕk : Ck (X; R) → Ck (X; R) consists of zeros, so

that tr(ϕk ) = 0 for all k. The theorem follows by the Hopf Lemma. 
Corollary. Any continuous map of an acyclic finite simplicial space has
a fixed point.
Note that the Brouwer fixed point theorem is a simple particular case of
this corollary.

8.6. Vector fields on spheres

In the Topology-1 course, we proved that there are no continuous vec-
tor fields without singular points on the two-dimensional sphere S2 . That
theorem has the following generalization.
Theorem. There are no continuous vector fields on the even-dimensional
sphere S2k , k ≥ 1.
Proof. Suppose that such a field v(x) on S2k exists. Regard the sphere
S as the standard sphere in R2k+1 and define the map f : S2k → S2k by
2k

assigning to each point x ∈ S2k the intersection with S2k of the ray issuing
from the origin and passing through the end point of the vector v(x).

v(x)
O

f (x)

Figure 8.11. Construction of the map f

The map f is obviously continuous and has no fixed points, is homotopic

to the identity, and therefore deg(f )=1. However, by the definition of the
Lefschetz number, we have

Λ(f ) = 1 + (−1)n deg(f )

 71

for any map f : Sn → Sn ; so in our case (n = 2k) we obtain Λ(f ) = 2 6= 0.

By the Lefschetz Theorem, this is a contradiction. 
The question of the existence of nonsingular continuous vector fields on
odd-dimensional spheres will be discussed in the exercise class.
72

8.5. Problems

8.1. Prove that a topological space with zero-dimensional homology isomor-

phic to Z is path connected.
8.2. Prove that a connected triangulated n-manifold with n-homology group
isomorphic to Z is orientable.
8.3. Prove that the rank of the Z-module Hk (X; Z) is equal to the dimension
of the vector space Hk (X; R).
8.4. Prove that the degree of sphere maps as defined geometrically in Lecture
4 for neat spaces coincides with the homological definition of degree given in
the present lecture.
8.5. Compute the Euler characteristic of the n-sphere.
8.6. Using the Mayer–Vietoris sequence, prove that

χ(X) = χ(X1 ) + χ(X2 ) − χ(X1 ∩ X2 ),

where X1 and X2 are simplicial subspaces of a simplicial space X. Prove the

same equality “by counting simplices”.
8.7. Compute χ(CP n ).
8.8. Compute χ(RP n ).
8.9. Prove that χ(X × Y ) = χ(X) · χ(Y ) for any finite simplicial complexes
X and Y .
8.10. Prove that χ(M ) = 0 for any odd-dimensional manifold M .
8.11. Prove that the Euler characteristic of the complement to a knot (or a
link) in S3 is zero.
8.12. Prove that any continuous map of RP n to itself has a fixed point if n
is even.
8.13. Do there exist continuous maps of RP n to itself with no fixed points
if n is odd?
8.14. Find a continuous vector field without any singular points on S3 .
8.15. Are there continuous vector fields without any singular points on S2k+1
for k > 1?
 73

Lecture 9
COHOMOLOGY
Cohomology groups are dual to homology groups in the same sense that
covectors are dual to vectors: they are linear functionals on homology. At
first glance, it seems useless to construct a dual theory which is, in a sense,
equivalent to the original one (in particular, it satisfies a dual version of the
same Steenrod–Eilenberg axioms). However, it turns out that in cohomology
theory there is a multiplication operation (the cup product) which has much
better properties than the corresponding operation (the cap product) in ho-
mology. Moreover, cohomology is the natural setting for other operations
(such as Steenrod squares) and for such constructions as the Poincaré iso-
morphism, and it coincides with the purely analytic cohomology of de Rham
defined on smooth manifolds by means of differential forms.
9.1. Definitions and constructions
For the sake of simplicity, we will construct cohomology theory for the
case of finite simplicial complexes, although the general case of arbitrary
topological spaces can be treated in the same way with a few modifications.
(Beware: some modifications are a bit delicate, e.g. the question of dualizing
infinite sums of Abelian groups.) You will see in what follows that in practice
the dualization of homology theory is formally quite simple, and it consists
in lifting the indices and reversing the arrows.
Let X be a simplicial space and let G be an Abelian group. We shall
write Ck (X) = Ck (X; Z) for the simplicial chain group. A homomorphism
ck : Ck → G is called a k-dimensional cochain with values in G. Cochains ob-
viously form a group (with the group structure “inherited” from G) denoted
by C k (X; G). In more compact notation, C k (X; G) := Hom(Ck (X), G).
Let ck ∈ C k (X; G) and ck ∈ Ck (X). We denote the value of the homo-
morphism ck on the chain ck by hck , ck i and define the coboundary operator
δ : C k (X; G) → C k+1 (X; G), hδck , ck+1 i = hck , ∂ck+1 i.
(Here we have omitted the index k in the notation for δ and ∂ because it is
clear from the context.) To compute the value of a k-cochain, it suffices (by
linearity) to know its value on a k-simplex [v0 , . . . , vk ]. This value is given
by the formula
k+1
X
δc([v0 , . . . , vk ]) = (−1)i hck , [v0 , . . . , vi∨ , . . . vk ]i;
i=0
74

the (easy) proof of this formula will be discussed in the exercise class.
The Poincaré Lemma ∂ ◦ ∂ = 0 implies δ ◦ δ = 0, and so we can define
the cohomology groups

Z k (X; G)
H k (X; G) := , H 0 (X; G) := Z 0 (X; G) ,
B k (X; G)

where Z k and B k are the kernel and image of the corresponding cobound-
ary operators in C k ; elements of Z k are called cocycles and those of B k ,
coboundaries.
We defined cohomology groups on the chain-cochain level, however, for
the case of fields (in that case the homology groups are finite-dimensional
linear spaces), we could have defined them directly as dual spaces to the
homology spaces. This fact is expressed by the following statements, whose
simple proofs are omitted.
Theorem (Duality) If G is the additive group of a field, then H k (X; G)
is the dual space of Hk (X; G).
Corollary. If G is the additive group of a field and the space Hk (X; G)
is finite-dimensional, then H k (X; G) ∼= Hk (X; G).
This corollary begs the following question: what is the use of cohomology
groups if they are (isomorphic to) the homology groups? The answer to
that question was given at the beginning of this lecture – because of the
multiplication operation (see Sec.9.4).
The remaining basic steps of the theory, namely the definition and/or
construction of relative cochains, relative cohomology groups, augmentation,
reduced cohomology, induced homomorphisms in cohomology for spaces and
pairs of spaces, the coboundary operator δ ∗ : H k (X, A; G) → H k (X; G) is
similar (or rather dual) to the corresponding definitions and/or constructions
in homology theory and are left as exercises for the reader.

9.2. Properties (Steenrod–Eilenberg axioms for cohomology)

We will now list the main properties of the cohomology theory constructed
above (which can also be regarded as the Steenrod–Eilenberg axioms for
cohomology) for finite simplicial spaces.
But first we summarize what was done in the previous section. To every
finite simplicial space, to every pair of such spaces, and to their simplicial
maps, we assigned, for each nonnegative integer n, Abelian groups (called
 75

cohomology groups) and group homomorphisms, and to each pair of spaces

(X, A), a homomorphism ∂∗ of the nth cohomology group of A to the (n+1)st
cohomology group of the pair (X, A). In the notation introduced above, this
correspondence reads as

X 7→ H n (X), n = 0, 1, 2, . . .
(X, A) 7→ H n (X, A), n = 0, 1, 2, . . .
(X, A) 7→ ∂ ∗ : H n (X, A) → H n−1 (A), n = 1, 2, . . .
f :X→Y 7 f ∗ : H n (Y ) → H n (X), n = 0, 1, . . .
→
f : (X, A) → (Y, B) 7→ f ∗ : H n (Y, B) → H n (X, A), n = 0, 1, . . .

The sum over n of the obtained Abelian groups is denoted

∞
M ∞
M
∗ n ∗
H (X) := H (X) and H (X, A) := H n (X, A);
n=0 n=0

these objects will be called graded cohomology groups of X and (X, A).
Theorem. The correspondences described above define a contravariant
functor (called the cohomology functor) from the category of finite simplicial
spaces to that of graded Abelian groups. The contravariance of the functor
means that
(f ◦ g)∗ = g ∗ ◦ f ∗ and (idX )∗ = idH n (X) .
This functor has the following properties.
(I) Dimension: H0 (pt) = G, where pt is the singleton and G is an Abelian
group, and Hn (pt) = 0 for n > 0.
(II) Commutation: the following square diagram

H n (X, A) −−− → H n (Y, B)

f∗
x x

∂∗ 

∂∗ 

H n−1 (A) −−−→

∗
H n−1 (B).
(f |A )

is commutative.
(III) Homotopy invariance: if two maps f, g : (X, A) → (Y, B) are homo-
topic, then f ∗ = g ∗ , the induced homomorphisms coincide in all dimensions,
and so homotopy equivalent spaces have the same cohomology.
76

(IV) Exactness: For any pair of spaces (X,A), the following sequence is
exact:

i j∗ δ∗ i
. . . ←−∗−− H n (X) ←−−− H n (X, A) ←−−− H n−1 (A) ←−∗−− H n−1 (X)
j∗ δ∗ i j∗
←−−− H n−1 (X, A) ←−−− . . . ←−∗−− H 0 (X) ←−−− H 0 (X, A).
(V) Excision: Suppose that U ⊂ X is an open subset whose closure lies in
the interior of A, where A ⊂ X. Then the inclusion (X \ U, A \ U ) ,→ (X, A)
induces an isomorphism in cohomology

H n (X, A) → H n (X \ U, A \ U )

in all dimensions n.
Proof. The proofs of all five assertions of this theorem follow by dual-
ity from the five assertions in the corresponding theorem in homology (see
Sec.7.2). For example, to prove (IV) (exactness), we dualize the short exact
sequence of chains
∗ i p∗
0 −−−→ C(A) −−−→ C(X) −−−→ C(X, A) −−−→ 0,

obtaining the dual sequence of cochains

i∗ p∗
0 ←−−− C(A) ←−−− C(X) ←−−− C(X, A) ←−−− 0,

which is also exact (this follows from the Splitting Lemma applied to the
chain sequence). The (long) exact sequence for cohomology is then obtained
by applying the Short-to-Long Exact Lemma. 
Remark. We have formulated the Steenrod–Eilenberg axioms in the
finite simplicial setting; they also hold in the topological setting, and there
are uniqueness theorems similar to those in homology theory.

9.3. Ordered homology theory

To construct the multiplication operation, we need a slight modification
of simplicial (co)homology theory, in which we order all the vertices of the
given simplicial space X and then, instead of considering all (n + 1)! ordered
simplices determined by each geometric simplex, we consider only one ordered
simplex, namely the one whose vertices are listed in ascending order (w.r.t.
 77

the given ordering of the vertices of X). It is not hard to prove that this
modification yields the same homology groups and induced homomorphisms
(see Exercise 5.11 above).
Let X be a finite simplicial space and let the coefficient group G be a
commutative associative ring
with unit (e.g. Z). Recall that we defined the
chain complex Cn (X), ∂n by using oriented simplices (which form the basis
of each G-module Cn ). We will now consider ordered simplices (v0 , . . . vn ),
where the order of the vertices vi is fixed but the vi are not necessarily

pairwise different and define the ordered chain complex C bn (X; G), ∂n by
taking Cbn (X) to be the linear combination (with coefficients in G) of ordered
simplices and using the same boundary operator ∂ as in the usual theory (see
5.4). The rest of the ordered (co)homology theory is constructed similarly to
the usual simplicial (co)homology theory. Note, however, that the chain
complex in the ordered theory is “infinite to the left” (because repetition of
vertices forces us to consider simplices of arbitrarily high “dimension”), but
this chain complex has no nontrivial homology in dimensions higher than the
dimension of X. Moreover, we have the following theorem.
Theorem. The homology of the two complexes C b∗ (X; G) and C∗ (X; G)
is canonically isomorphic.
We omit the straightforward proof of this theorem.

9.4. Multiplication (cup product)

As above, let X be a finite simplicial space and the coefficient group G be
a commutative associative ring with unit. The cup product of two cochains
cp ∈ C
bp (X; G), cq ∈ C
bq (X; G) is the cochain cp ^ cq ∈ C bp+q (X; G) given by

hcp ^ cq , (v0 , . . . , vp+q )i = hcp , (v0 , . . . , vp )i · hcq , (vp , . . . , vp+q )i.

The cup product is obviously bilinear and associative, it has a unit defined
as the cochain taking the value 1 (where 1 is the unit of the ring G) at
all vertices of X. It can be carried over to cohomology classes due to the
following beautiful (and easily verified) lemma.
Lemma (“Leibnitz rule”). δ(cp ^ cq ) = (δcp ) ^ cq + (−1)p cp ^ δ(cq ).
The proof of this lemma, as well as the fact that it allows to correctly
define the cup product on the cohomology level, is a straightforward exercise.
It is also easy to prove that

f ∗ (α ^ β) = f ∗ (α) ^ f ∗ (β).
78

Theorem. The cup product supplies the graded group H ∗ (X; G) with
the structure of a graded G-module; the cup product is skew commutative
in the sense that
α ^ β = (−1)pq (β ^ α).

9.5. De Rham Cohomology

In this section, we recall the definition of de Rham cohomology (usually
given in advanced courses in differential geometry or on analysis on smooth
manifolds) and state without proof the famous de Rham theorem asserting
that this cohomology (defined in purely analytic terms) is isomorphic (for
compact smooth manifolds) to singular (or simplicial) cohomology.
Let M n be a smooth compact n-dimensional closed (=without boundary)
manifold, let Λk M n be the (linear over R) space of differential forms on M n .
In local coordinates, a differential form ω k ∈ Λk M n is expressed as
X
ωk = ai1 ,...,ik dxi1 ∧ · · · ∧ dxik
i1 ≤···≤1k

Recall that for any two differential forms ω1 ∈ Λp M n and ω2 ∈ Λq M n , their

exterior product ω1 ∧ ω2 is defined; it is skew commutative in the sense that

ω1 ∧ ω2 = (−1)pq (ω2 ∧ ω1 )

and it supplies the graded vector space Λ∗ M n with the structure of a graded
R-algebra. Any smooth map f : M n → N m induces a linear map

f ∗ : Λk N m → Λk M n .

The main tool in the construction of de Rham cohomology is the differential

d : Λk M n → Λk+1 M n .

Given a k-form ω k in local coordinates, we can define its differential by means

of the formula
n
X ∂ϕ(x)
d(ϕ(x)dxi1 ∧ · · · ∧ dxik ) := dxm ∧ dxi1 ∧ · · · ∧ dxik
i=0
∂x m

and linearity. It is easily verified that d ◦ d = 0. A k-form ω k is called closed

if dω k = 0 and exact if there exists a (k + 1)-form λk+1 such that dλ = ω.
 79

The equality d ◦ d = 0 implies that any exact from is closed, and the quotient
space of the space of closed k-forms by exact k-forms is, by definition, the de
k
Rham cohomology HdR (M n ) in dimension k.
The exterior multiplication operation can be carried over from the level
of forms to the cohomology level, because here also the Liebnitz rule holds:
for any k-form ω1 and any l-form ω2 ,

d(ω1 ∧ ω2 ) = dω1 ∧ ω2 + (−1)k ω1 ∧ dω2 .

k ∗
Each of the HdR (M n ) is a linear space over R, their direct sum HdR (M n )
supplied with exterior multiplication is a graded R-algebra.
Theorem (de Rham). For a smooth compact closed manifold M n , the
∗
de Rham cohomology HdR (M n ) is isomorphic as a graded algebra to the
singular (or simplicial) cohomology graded algebra H ∗ (M n ; R).
There are no easy proofs of this beautiful theorem (see V.V.Prasolov’s
book Elements of Homology Theory, pp.289-300).
80

9.5. Problems
9.1. Prove that
k+1
X
δc([v0 , . . . , vk ]) = (−1)i hck , [v0 , . . . , vi∨ , . . . vk ]i;
i=0

9.2. Show that the Poincaré Lemma ∂ ◦ ∂ = 0 implies δ ◦ δ = 0.

9.3. Prove the Poincaré Lemma ∂ ◦∂ = 0 for the ordered simplicial homology
theory.
9.4. Prove the Liebnitz rule δ(cp ^ cq ) = (δcp ) ^ cq + (−1)p cp ^ δ(cq ).
9.5. Prove that f ∗ (α ^ β) = f ∗ (α) ^ f ∗ (β).
9.6. Given X = ni=1 Ai , where the Ai are contractible simplicial subspaces
S
of X, prove that the product α1 ^ · · · ^ αn vanishes for any choice of
elements αi ∈ H ki (X), where ki > 0. Use this fact to prove that the cup
product in the suspension ΣX for any finite simplicial space is trivial in the
sense that the product of any two cohomology classes of positive dimension
vanish.
9.7. Suppose that the simplicial subspace A ⊂ X is a retract of the finite
simplicial space X, i : A ,→ X is the inclusion, and r : X → A is the
retraction. Prove that H∗ (X) = Im i∗ ⊕ Ker r∗ and H ∗ (X) = Im r∗ ⊕ Ker i∗ .
Show that in the cohomology ring H ∗ (X) the subset Ker i∗ is an ideal and
Im r∗ is a subring.
9.8. Given the cup products in the algebras H ∗ (X) and H ∗ (Y ), determine
the cup product in H ∗ (X ∧ Y ).
9.9. Prove that Sn ∨ Sm is not a retract of Sn × Sm , where Sn ∨ Sm is
understood as Sn × {x0 } ∪ {y0 } × Sm and n, m ≥ 1.
9.10. Prove that the standardly embedded space RP m ⊂ RP n , n > m, is
not a retract.
 81

Lecture 10
POINCARÉ DUALITY

Poincaré duality is an isomorphism between the homology groups (of

different dimensions) of manifolds. It is a rather amazing relationship, since
it says that the inner structure of an arbitrary closed compact manifold is, in
a sense, symmetric with respect to the middle dimension: roughly speaking,
it asserts that the k-dimensional homology of an n-manifold is isomorphic
to its (n − k)-dimensional (co)homology. This symmetry is based on an
extremely simple geometric construction (see Sec.10.1) and uses the notion
of cap product, an operation in a sense dual to the cup product studied in
the previous lecture.

10.1. The dual cellular decomposition

To a given a compact closed (=without boundary) n-manifold M n with a
fixed triangulation, we will canonically associate (following Poincaré) a cel-
lular decomposition (i.e., we will supply M n with a CW-space structure). To
this end, we begin by taking the baricentric subdivision of the triangulation
and, to avoid confusion, refer to a simplex in the subdivision as a subsimplex,
saving the word “simplex” for simplices in the original triangulation. Now
to any vertex (i.e., 0-simplex) σ 0 , we associate the n-cell σ∗n consisting of all
n-subsimplices having a vertex at σ 0 . More generally, to any k-simplex σ k we
associate the (n − k)-cell σ∗n−k consisting of all (n − k)-subsimplices having
a vertex at the baricenter of σ k . For n = 2, see Figure 10.1.

σσσ111111

σ∗∗11∗∗11∗∗11
σσσ
σ

σσσ000000
σσσ∗∗22∗∗22∗∗22

Figure 10.1. The dual cellular decomposition for n = 2

82

Note that that each simplex σ intersects its dual cell transversally, their
only intersection point being the baricenter of σ.
We have not only supplied M n with a CW-space structure (which we
denote by M∗ ), but we have defined a bijection between the k-simplices σ
of the given triangulation and the dual (n − k)-cells σ∗ of M∗ . Moreover, if
the orientation of M n is fixed arbitrarily (i.e., a continuously varying basis,
which we will call positively oriented, is given at each point), then each ori-
ented simplex σ determines a corresponding orientation of the dual cell in
the natural way: we orient the dual cell so that its basis (attached to the
baricenter) added to the orienting basis of σ has the same orientation as the
positive basis of M n at the baricenter.
This allows us to construct, for any k = 0, 1, . . . n a homomorphism

δ : Ck (M ; Z) → Cn−k (M∗ ; Z)

given on simplices by δ(σ k ) = σ∗n−k and extended by linearity.

The homomorphism δ is defined for all k from 0 to n, so one might naively
think that δ is a morphism of the chain complex C(M, Z) to C(M∗ , Z) which
will give us the required isomorphism in homology. But this is completely
wrong, because the boundary operators in the chain complexes C(M, Z) and
C(M∗ , Z) act in opposite directions from the point of view of dimension, i.e.,

dim((∂ ◦ δ)(ck )) = n − k + 1 6= n − k + 1 = dim((δ ◦ ∂)(ck ))

Having in mind that coboundary operators in cochain complexes also act

“in the opposite direction”, we will overcome this difficulty by passing from
homology to cohomology; this will be done in the next section.

10.2. Homology–cohomology duality

In the rest of this lecture, we will assume that the coefficient group (in
homology and cohomology) is the same commutative associative ring R with
unit (e.g. Z); the ring R will not explicitly appear in the notation: we write,
for instance, C k (M∗ ) instead of C k (M∗ ; R).
Let us construct an homomorphism ϕk from Ck (M n ) to C n−k (M∗ ) by
assigning to each simplex σ k (regarded as a k-chain with coefficient 1 ∈ R
at σ k ) the linear functional on (n − k)-cells that assigns 1 to the cell σ∗n−k
dual to σ k and 0 to the other (n − k)-cells. Obviously, ϕk is an isomorphism
between Ck (M n ) and C n−k (M∗ ), and it readily follows from the definitions
 83

of ϕk , δ, and ∂ that for all k the diagrams

Ck (M ) −−−→ C n−k (M∗ )

ϕk
 

∂k y

δn−k y

Ck−1 (M ) −−−→ C n−k+1 (M∗ ).

ϕk−1

are commutative. This last fact implies in turn that the corresponding ho-
mology and cohomology groups are isomorphic. Thus we have proved the
following theorem.
Theorem (Poincaré Duality). Let M n be a smooth oriented compact
closed manifold and R be a commutative associative ring with unit. Then
the homology and cohomology groups of complementary dimensions are iso-
morphic, i.e., for all k = 0, 1, . . . , n,

Hk (M ; R) ∼
= H n−k (M ; R)

The theorem does not hold for the case in which the manifold M is non-
orientable, but if we replace the ring R by Z2 (the integers mod 2), then we
still have an isomorphism, namely

Hk (M ; Z2 ) ∼
= H n−k (M ; Z2 ) for all k = 0, 1, . . . , n.

The proof of this fact is an exercise.

The construction of Poincaré duality described above is very simple, but
it is defined on the chain-cochain level, it depends on the choice of the trian-
gulation of M , and does not give an effective way to perform computations
on the homology-cohomology level. Actually, the Poincaré isomorphism is
canonical, and there is a very simple formula for computing it. It requires,
however, a new notion (the cap product), which will be introduced in the
next section.

10.3. The cap product

The duality isomorphism ϕk,∗ : Hk (M ; R) → H n−k (M ; R) between the
homology and cohomology groups allows to define, for the homology of
smooth manifolds, a multiplication operation, called the homology cap prod-
uct, by simply carrying over to homology the cup product in cohomology
84

via this isomorphism. More precisely, for homology classes a ∈ Hp (M ) and

b ∈ Hq (M ), we define their cap product a _ b ∈ Hp+q (M ) by setting

a _ b := ϕ−1


p+q ϕ p (a) ^ ϕ q (b) .

From this definition and from the properties of the cup product, it imme-
diately follows that the homology H∗ (M ) of a smooth compact closed man-
ifold M supplied with the cap product is a graded R-algebra with skew-
commutative multiplication.
This statement is not, as would seem at first glance, a useless dualiza-
tion. It turns out that the cap product has a simple very visual geometrical
interpretation, which sometimes makes its computation simpler than that of
the cup product. Also, as we shall see shortly, it gives a simple description
of Poincaré’s duality isomorphism on the homology-cohomology level.
Now let us give a sketch of the geometric interpretation of the cap product.
Suppose a ∈ Cp (M ) and b ∈ Cq (M ) are cycles. To simplify the exposition,
we assume that a and b are cycles with coefficients 1 on some simplices and
zero on the others; let A be the set of all simplices with coefficient 1 for the
cycle a and B the corresponding set for b. Further assume that A and B
are in general position, which means that two (open) simplices σ p ∈ A and
τ q ∈ B either do not intersect at all or intersect in an open cell of dimension
p + q − n. Then A ∩ B is a cell space. Orient all the cells of A ∩ B as follows:
first to an orienting basis of σ p ∩ τ q (it has p + q − n vectors) add n − p vectors
to form an orienting basis of σ p , then add n − q vectors to form an orienting
basis of τ q , in such a way that the n vectors thus obtained determine the
chosen orientation of M n (Fig.10.2).

333333

2 τq
σp

Figure 10.2. Orienting the intersection of two simplices

 85

It can be shown that the (p + q − n)-cells of A ∩ B taken with coefficients

1 form a cycle a _ b ∈ Cp+q−n (M n ); now if α and β are the homology
classes determined by a and b, then the cycle a _ b determines a well-
defined homology class in Hp+q−n (M n ) denoted by α _ β. Thus we have
constructed a bilinear map

_ : Hp (M n ; R) × Hq (M n ; R) → Hp+q−n (M n ; R).

The homology cap product discussed above is more visual but also less
general than its other version, the cohomology-homology cap product (usu-
ally referred to simply as the cap-product). This operation, also denoted by
_, is defined for the homology and cohomology of path connected simpli-
cial spaces X (not only smooth manifolds) with coefficients in an associative
commutative ring R with unit. We will be using ordered simplices; to fix
an order, we number all the vertices of X once and for all; the vertices of
any simplex will always be listed in increasing order. In this setting, the
cap-product is a bilinear map

_ : H p (X; R) × Hp+q (X; R) → Hq (X; R) (1)

defined as follows.
Suppose that cp ∈ C p (X; R); as usual, we first define the cap product on
simplices; we put

cp _ [v0 , . . . , vp+q ] := hcp , [vp , . . . , vp+q ]i[v0 , . . . , vq ] ∈ Cq (X; R).

In the particular case q = 0, we put

cp _ [v0 , . . . , vp ] := hcp , [v0 , . . . , vp ]i ∈ R = C0 (X, R)

Extending this map by linearity, we obtain a bilinear map at the chain-

cochain level
_ : C p (X; R) × Cp+q (X; R) → Cq (X; R).
To obtain a map on the (co)homology level, we need the following version of
the Leibnitz rule
Lemma (Leibnitz rule for cap-products).

∂(cp _ cp+q ) = (−1q ) δcp _ cp+q + cp _ ∂cp+q .

86

The proof is a mildly difficult exercise. An immediate consequence of this

lemma is the definition of the bilinear map (1), i.e., that of the cap product
at the (co)homology level.
The cap product possesses the following properties:
(1) for any map f of simplicial spaces f∗ (f ∗ a _ b) = a _ f∗ (b);
(2) haq , bp _ cp+q i = haq _ bp , cp+q i;
(3) ap _ (bq _ cp+q+r ) = (ap ^ bq ) _ cp+q+r );
here a, b are cohomology classes, c is a homology class.

10.4. The Poincaré isomorphism

Poincaré duality was defined above on the chain-cochain level, and that
definition did not give an effective way to perform computations on the
(co)homology level. However, the Poincaré isomorphism can be described
directly in (co)homology by a simple formula involving the cap product.
Theorem (Poincaré Isomorphism). The assignment

H n−k (M n ; R) 3 αn−k 7→ αn−k _ [M n ] ∈ Hk (M n ; R),

where [M n ] is the fundamental class of the smooth orientable manifold M n ,

determines the Poincaré duality isomorphism H n−k (M n ; R) ∼ = Hk (M n ; R).
Proof. Let K be some triangulation of M n and K 0 be its baricentric
subdivision. We will be dealing with ordered homology, so we order all the
vertices of K 0 as follows: first we enumerate the vertices of K, then the
baricenters of the 1-simplices of K, then the baricenters of the 2-simplices
of K, and so on. For the cycle representing the fundamental class of M n let
us take the chain with coefficients 1 on all the n-simplices (v0 , . . . , vi ) of K 0
written so that the vertices appear in increasing order.
For this ordering, each of the ordered simplices σ i = (v0 , . . . , vi ) of K 0 is
contained in exactly one ordered simplex τ i of K and the latter is determined
by the last vertex vi of σ i . The corresponding (n−i)-cell τ∗i (see Sec.10.1) can
be represented as the union of simplices ±(vi , . . . , vn ) with vertices written
in increasing order.
zi σik we
P
Under Poincaré duality (see Sec.10.1), to each chain ck =
n−k n−k
assign the cochain c such that hc , σi∗ i = zi . Therefore,
X X
cn−k _ ±(v0 , . . . , vn ) = hcn−k , ±(vk , . . . , vn )i(v0 , . . . , vk )
X
= hcn−k , σi∗ iσik = zi σik = ck ,
 87

as required. 

10.5. The Gauss linking number

The Gauss linking number is an integer invariant of pairs of smooth curves
embedded in R3 which says how many times one curve “wraps around” the
other. Originally defined by Gauss as a double integral in the context of elec-
trodynamics, it has a simple homological interpretation that we now present.
The Gauss linking number lk(C, D) of two disjoint circles C1 and C2 em-
bedded in the sphere S3 is the integer lk(C1 , C2 ) defined as follows: assume
(without loss of generality) that C1 and C2 are composed of 1-simplices (ori-
ented along the circles) and consider the two chains c1 , c2 ∈ C1 (S3 ; Z) with
coefficients 1 on these simplices and 0 on the others; obviously, these chains,
in particular c2 , are cycles; since S2 has trivial 1-homology, c2 is a boundary,
i.e., there exists a 2-chain d ∈ C2 (S2 ) such that ∂(d) = c2 ; we put

lk(C1 , C2 ) := [d _ c1 ] ∈ H0 (S3 ) = Z.

The fact that the linking number is well defined (does not depend on the
choice of d) is the subject matter of an exercise.
Actually lk(C1 , C2 ) is a so-called finite-type invariant in the sense of Vas-
siliev and can be computed in an elementary way from the diagram of the
link C1 ∪ C2 ⊂ S3 by counting the “signs” of the crossing points of the two
curves (as shown in Fig.3)

+ −
+

+
−
− lk(c1 , c2 ) = −2
−
Figure 10.3. Counting the Gauss linking number
88

10.6. Problems
10.1. Let ϕ : Ck (M ) → C n−k (M∗ ) be the natural homomorphism between
the simplicial chain group of a smooth closed manifold M n and the cochain
group of the dual cell decomposition M∗ of M n . Prove the following commu-
tation relation δn−k ◦ ϕk = ϕk−1 ◦ ∂k , where ∂ and δ are the boundary and
coboundary operators.
10.2. Prove that if H1 M 3 = 0 for a closed 3-manifold M 3 , then M 3 is a
homology 3-sphere, i.e., its homology groups as S3 .
10.3∗ . Give an example of a homology 3-sphere which is not homeomorphic
to the 3-sphere.
10.4. Let the torus be presented as a CW-space with four cells, let c1 be
the cochain taking one of the 1-cells to 1 and the other to 0, while c1 is the
chain with coefficient 1 at the second 1-cell and 0 at the other. Compute
c 1 _ c1 .
10.5. Prove the Leibnitz rule for the cap product.
10.6. Prove the following properties of the cap product:
(i) f∗ (f ∗ (b) _ a) = b _ f∗ (a), where f : X → Y is a simplicial map,
a ∈ H∗ X, and b ∈ H ∗ Y ;
(ii) haq , bp _ cp+q i = haq _ bp , cp+q ;
(iii) ap _ (bq _ cp+q+r ) = (ap ^ bq ) _ cp+q+r .
10.7. Give an example of two simplicial complexes with the same homology
but with different cap products.
10.8. Prove that the GaussP linking number of two oriented circles embedded
3
in R given by lk(C1 , C2 ) = εi , where ε = ±1 is as in Fig.3, is well defined.
10.9. Prove the Poincaré duality theorem for nonoriented manifolds.
 89

Lecture 11
OBSTRUCTION THEORY

In the first lecture of this course, we considered the extension problem:

given a map f : A → Y , where A is a subset of X, to construct a map
F : X → Y that coincides with f on A. If X is a simplicial space (or a
CW-space), it is natural to attack this problem by induction: first extend
f to the vertices of the triangulation of X, then to the 1-simplices, to the
2-simplices, etc. It turns out that each inductive step can be carried out if
and only if a certain cohomology class vanishes. In a sense, this cohomology
class “obstructs” our construction, and so the inductive approach described
above is called obstruction theory.
This theory has numerous applications (not only to the extension of
maps), in particular to the classification of homotopy classes of various fam-
ilies of maps by means of the appropriate cohomology groups and to the
construction of the so-called Eilenberg–MacLane spaces. The latter, in turn,
play a key role on other homotopy classification theorems.

11.1. The obstruction cocycle

Suppose that X is a simplicial space, A is its simplicial subspace, X n the
n-dimensional skeleton of X, and we are given a map f : X b n → Y , where
Y is a path connected topological space and X b n := X n ∪ A. Our goal is to
extend this map to X b n+1 . We will assume that n ≥ 1, because for a path
connected Y the construction of the extension is obvious. Further, we will
assume that Y is n-simple, i.e., that the action the π1 (Y ) on its homotopy
groups of dimension n is trivial (this will allow us to add elements of the
group πn (Y, y) with different basepoints y).
To the map f : X b n → Y we assign the cochain

cn+1 (f ) ∈ C n+1 (X; πn (Y ))

as follows. As usual, it suffices to define cn+1 on (n + 1)-simplices; since each

simplex ∆n+1 is oriented, its orientation defines an orientation on ∂(∆n+1 )
and so the restriction of f to ∂(∆n+1 ) can be regarded as a spheroid in Y ;
the homotopy class of this spheroid is, by definition, the element of πn (Y )
that we assign to f . Thus:

cn+1 (f )(∆n+1 ) := f |∂(∆n+1 ) ∈ πn (Y ).

 
90

Note that if ∆n+1 belongs to A, the spheroid corresponding to it is obviously

trivial, so we can regard (and will regard) cn+1 (f ) as a relative cochain, i.e.,
an element of C n+1 (X, A; πn (Y )).
Lemma (Obstruction Cocycle) . The cochain cn+1 (f ) is a relative cocy-
cle, i.e.,
δ cn+1 (f ) = 0.

Proof. We must prove that cn+1 (f )(∂∆n+2 ) = 0 for any (n + 2)-simplex

n+2
∆ in X. Note that the n-skeleton of the simplicial space ∂ ∆n+2 ≈ Sn+1
is, obviously, an (n − 1)-connected space (i.e., all its homotopy groups up to
the (n − 1)st are trivial), because it is in fact homotopy equivalent to the
sphere Sn+1 with several punctures in it (at the baricenters of the (n+1)-faces
of ∆n+2 ).
Let ∂(∆n+2 ) = ∆in+1 . The element cn+1 (f )(∆n+1
P

i ) of the group πn (X)
n+2
corresponds to the map f : ∂ ∆ → X. Denote by B n the n-skeleton of
∂(∆n+2 ). Let αi be the element of πn (B n ) corresponding to the orientation-
preserving homeomorphism Sn → ∂(∆n+1 i ). Obviously, the homomorphism

f∗ : πn (B n ) → πn (X)

takes αi to cn+1 (f )(∂∆n+1i ).

Since the space B n is (n − 1)-connected, the Hurewicz homomorphism
h : πn (B n ) → Hn (B n ) is an isomorphism (by the Hurewicz Theorem). Hence
we can consider he sequence of homomorphisms
h−1 f∗
Hn (B n ) −−−→ πn (B n ) −−−→ πn (X);

Their composition f∗ ◦ h−1 takes the homology class determined by the cycle
∂(∆n+1 ) =: z to cn+1 (f )(∂∆n+1
P
i i ). But
X
∂(∆n+1
i ) = ∂∂∆n+2 = 0,

i.e., the cycle z is zero and therefore its image by the homomorphism f∗ ◦h−1 is
also zero as claimed. 

11.2. The obstruction cohomology class

The obstruction cocycle cn+1 (f ) determines a cohomology class that we
denote by Γn+1 (f ).
 91

b n → Y can be extended to X
Theorem (Eilenberg) The map f : X b n+1 if
and only if Γn+1 (f ) = 0.
We omit the detailed proof of this theorem, which is rather straightfor-
ward. Indeed, in order to extend the map f , we must be able to extend it
to each (n + 1)-simplex from its boundary (on which f is already defined).
Roughly speaking, the condition Γ = 0 says that the image of the boundary
of each such simplex is homotopy trivial in Y , so that the extension of the
map to the (n + 2)-simplex is possible.

11.3. The Hopf–Whitney Theorem

The homotopy classification of maps from one topological space to an-
other is one of the fundamental problems of topology. We denote by [X, Y ]
the set of homotopy classes of maps from X to Y . The next theorem gives
a solution of the homotopy classification problem for a wide class of spaces
in terms of cohomology of the source space whose coefficients are in the
homotopy group of the target space.
Theorem (Hopf–Whitney). For any n-dimensional simplicial space X
and any (n − 1)-connected space Y , the following bijection exists:

X, Y ←→ H n (X; πn (Y )).
 

For the particular case in which Y = Sn , this theorem classifies up to

homotopy maps of simplicial complexes to the n-sphere: they correspond bi-
jectively to elements of H n (X; Z) because πn (Sn ) = Z. Another consequence
of this theorem is the following nice geometrical interpretation of the nth
cohomology group of an arbitrary n-dimensional simplicial space.
Corollary. Any element of the cohomology group H n (X, Z), where X
is an n-dimensional simplicial space, can be realized by maps to the sphere,
i.e., it can be represented in the form f ∗ (s), where s is the generator of the
group H n (Sn ; Z) and f : X → Sn is a map unique up to homotopy.

11.4. The Eilenberg Mac Lane spaces K(π, n)

These are spaces whose topology is concentrated, in a certain homotopic
sense, in a single dimension. They possess some beautiful properties and are
useful in homotopy classification problems as well as in the construction of
92

the s–called cohomology operations. By definition, an Eilenberg–MacLane

space K(π, n) is a topological space X satisfying the condition
(
π if k = n;
πk (X) =
0 if k = n.

Examples:
• the circle S1 is a K(Z, 1) space;
• any surface M 2 other than S2 and RP 2 is a K(π1 (M 2 ), 1) space;
• RP ∞ is a K(Z2 , 1) space;
• CP ∞ is a K(Z2 , 2) space;
• the infinite-dimensional lens space L∞ m is a K(Zm , 1) space.
The validity of these specific examples will be discussed in the exercise
class. The existence of many other K(π, n) spaces will be established in the
next theorem.
Theorem (Existence of K(π, n) spaces). For any finitely presented group
π and any n there exists a K(π, n) space.
Proof. The proof is by a direct construction, in which we first construct
a space with zero homotopy groups in all dimensions up to n−1, then supply
it with the necessary n-homotopy group (isomorphic to π), and finally kill
all the higher homotopy inductively by “chasing them away to infinity”.
Let hg1 , . . . , gr |R1 = · · · = Rs = 1i be a presentation of π. Denote by K n
the wedge of r copies of the (triangulated) n-sphere. Take s copies of the
(n + 1)-disk and glue it to the wedge of spheres in accordance to the relations
R1 , . . . , Rs , obtaining a simplicial space that we denote by K n+1 . Clearly,
Hn (K n+1 ) ∼ = π. By the Hurewicz Theorem, we have πn (K n+1 ) ∼ = π.
However, our construction is not finished because K n+1 can have nontriv-
ial higher (than n) homotopy groups. We kill these groups inductively by
gluing disks onto their generators. 
Theorem (Uniqueness of K(π, n) spaces). Two K(π, n) spaces with the
same π and n are homotopy equivalent.
Proof. It suffices to prove that any K(π, n) space is homotopy equivalent
to the one constructed in the proof of the Existence Theorem. This is done
inductively on the dimension of the skeletons of the given space, the key
point is the application of obstruction theory, which is easy because the
corresponding obstructions vanish. 
 93

Theorem (Maps to K(π, n) spaces). If Y is a simplicial K(π, n) space

and X is any simplicial space, then there is a bijection

X, Y ←→ H n (X; π).
 

Proof. Let us denote by Fπ ∈ H n (X, π) the fundamental class of the

space Y (i.e., the class corresponding to the inverse Hurewicz isomorphism
h−1 : Hn (X) → πn (X), see the proof of the Existence Theorem). To any map
f : X → Y we assign the cohomology class f ∗ (Fπ ) ∈ H n (X; π); clearly this
assignment depends only on the homotopy class of f . Its bijectivity follows
from the Hopf–Whitney Theorem. The details are left to the reader. 
Corollary. Maps from one Eilenberg-MacLane space to another (with
the same n but possibly different π’s ) are classified up to homotopy by the
homomorphisms of their groups:

K(π, n), K(π 0 , n) ←→ Hom(π, π 0 ).

 

Proof. The so-called universal coefficients formula, which will be dis-

cussed in the next lecture, implies that

H n (K(π, n); π 0 ) ∼
= Hom(Hn (K(π, n); π 0 )),

so that the corollary follows from the previous theorem. 

94

11.5. Problems
11.1. Prove that Sp × Sq = (Sp ∧ Sq ) f Dp+q , where f is some map
S
f : Sp+q → Sp ∧ Sq non homotopic to the constant map.
11.2. In the notation of Sec.11.2 of the lecture, let f, g : K b n → Y be
two maps that coincide on K b n−1 . Denote by dn (f, g) ∈ C n (K n , A; πn (Y ))
the cochain (called distinguishing) defined on a simplex ∆n as follows: take
two copies of ∆n , glue then along their boundary, map one of the copies via
f and the other via g to Y , thus determining an element of πn (Y ). Prove
that
dn (f, g) = cn (g) − cn (f ).
Use this fact to prove the Eilenberg Theorem.
11.3. Prove that dn (f, g) + dn (g, h) + dn (h, f ) = 0, where dn is the
distinguishing cochain defined in Problem 11.2.
11.4. Prove that S1 is a K(Z, 1) space.
11.5. Let M 2 be a closed surface other than S2 or RP 2 . Prove that M 2
is a K(π1 (M 2 ), 1) space.
11.6. Prove that RP ∞ is a K(Z2 , 1) space.
11.7. Prove that CP ∞ is a K(Z, 2) space.
11.8. Prove that L∞ m is a K(Zm , 1) space, where the infinite-dimensional
lens space L∞ m is defined as follows: ∞ ∞
P 2 consider S as the subspace of C of all
points (z1 , z2 , . . . ) such that |zi | = 1; introduce the equivalence relation
(z1 , z2 , . . . ) ∼ (εz1 , εz2 , . . . ), where ε = exp(2πi/m); then L∞ ∞
m := S / ∼.
11.9. Prove that
(
Zm for odd n;
Hn (L∞ ) =
0 for even n.

11.10. Let M 3 be a closed 3-manifold such that π1 (M 3 ) is infinite and

π2 (M 3 ) = 0. Prove that M 3 is a K(π, 1) space for some π.
11.11. Let X be a finite-dimensional simplicial K(π, 1) space. Prove
that π has no finite-order elements.
 95

Lecture 12
VECTOR BUNDLES AND G-BUNDLES

In this lecture, we study and classify vector bundles (a key notion in dif-
ferential topology, the main example being the tangent bundle of a smooth
manifold). and principal G-bundles (which play an important role in geomet-
ric topology, K-theory, and theoretical physics). Vector bundles are classified
over paracompact spaces by using the partition of unity, a generalization of
the Feldbau Theorem, and the Gauss map to the canonical Grassmann bun-
dle, while G-bundles are classified via the beautiful Milnor construction and
the notion of classifying space of a topological group.

12.1. The category of vector bundles

A vector bundle p : E → B is a (locally trivial) fiber bundle (see Sec.2.3)
whose fiber F is a vector space of fixed dimension n. A morphism of vector
bundles (ϕ, Φ) : p1 → p2 is a pair of maps Φ : E1 → E2 and ϕ : B1 → B2
such that ϕ ◦ p1 = Φ ◦ p2 and the restriction of Φ to any fiber, i.e.,

Φ|p−1
1 (b)
: p−1 −1
1 (b) → p2 (ϕ(b))

is a linear map for any b ∈ B1 .

Given a vector bundle p : E → B and a subset X ⊂ B, one defines the
restriction of p to B1 , denoted p|X in the natural way; more generally, if
f : X → B is a map, then the pullback f ∗ (p) : E1 → X of p to X is defined
as follows:

and f ∗ (p)(b, e) := b.

E1 := (b, e) ∈ X × E : f (b) = p(e)

In that situation, there is a canonical morphism f ∗ (p) → p given by the

formulas ϕ(b) = f (b), Φ((b, e)) = e.
A morphism of vector bundles ϕ : p1 → p2 is said to be an isomorphism
if there exists a morphism ψ : p2 → p1 such that we have Ψ ◦ Φ = idB1 and
Φ ◦ Ψ = idB2 .
Lemma. A morphism of fiber bundles over the same base is an isomor-
phism if and only if its restriction to any fiber is an isomorphism of linear
spaces.
The proof is an exercise.
96

Vector bundles are usually considered over paracompact bases (because

partition of unity is needed to develop the theory). Vector bundles over all
paracompact spaces form a category that will be denoted by Vect. We shall
also consider the categories of vector bundles over fixed paracompact bases;
they are denoted Vect(B), where B is the (paracompact) base, and Vectk (B),
when the dimension k of the fiber is fixed.
Examples:
• the trivial bundles, or product bundles pr1 : R × B → B, pr1 (r, b) = b;
• the tangent bundle τ : T M n → M n of a smooth manifold;
• the normal bundle ν : N M n → M n of a smooth manifold smoothly
embedded in RN ;
• the canonical Grassmann bundle γkm : Ekm → Gm m
k , where Gk is the
Grassmann manifold (i.e., the set, supplied with the natural topology, whose
points are the k-dimensional linear subspaces L of the vector space Rm ),

Ekm := (L, r) ∈ Gm m

k ×R : r ∈ L ,

and γkm is the natural projection (L, r) 7→ L. There are obvious inclusions
m+1
Gmk ⊂ Gk which allow to define G∞ ∞
k (called the Grassmannian) and γk by
passing to the inductive limit.

12.2. Classification of vector bundles over a given base

In this section, we study the category Vectk (B) of k-vector bundles over
a fixed paracompact base B. The main result is the following.
Theorem (Classification). The isomorphism classes of k-vector bundles
p : E → B over a paracompact base B are in bijective correspondence with
the homotopy classes of maps of the base to the Grassmannian G∞k ,

B, G∞
 
k ←→ Vectk (B) ,

where the correspondence assigns to each map f : B → G∞

k the pullback
bundle f ∗ (γ ∞k ).
To prove the theorem, we must establish three facts.
(i) The correspondence is well defined, i.e., any two homotopic maps
∗ ∞ ∼ ∗ ∞
f, g : B → G∞k have isomorphic pullbacks f (γk ) = g (γk ).
 97

(ii) The correspondence is injective, i.e., if the bundles f ∗ (γ ∞k ) and

g ∗ (γ ∞k ) are isomorphic, then the maps f and g, f, g : B → G∞ k , are ho-
motopic.
(iii) The correspondence is surjective, i.e., for any bundle p ∈ Vectk (B)
there is a map f : B → G∞ k such that the pullback bundle f (γ
∗ ∞k
) is
isomorphic to p.
Let us comment on the proofs of these facts.
Item (i) is a particular case of the following more general theorem.
Theorem (Homotopy Invariance). For any vector bundle γ : E → B
with paracompact base B, homotopic maps f, g of a paracompact space B1
to B have isomorphic pullback bundles f ∗ (γ) ∼ = g ∗ (γ).
This theorem has a corollary, which is a far-going generalization of the
Feldbau Theorem (Lecture 2, Sec.2.3).
Corollary (Triviality). Any vector bundle over a contractible paracom-
pact space is trivial.
The proof of the Homotopy Invariance Theorem is based on the following
lemma.
Lemma (Multiplication by [0, 1]). For any vector bundle p with basis
B × [0, 1] there exists an open covering {Uα } of B such that the restriction
of the bundle to each of the sets Uα × [0, 1] is trivial.
Item (ii) is proved by using a version of the Gauss map, which we first
define in the case of a smooth manifold M k embedded in RN , as the vector
bundle γkN : T M k → GN k obtained by the parallel transport of all the tangent
planes Tx (M k ), x ∈ M k , to the origin 0 (so that x is taken to 0).
In the general case, the construction is somewhat more difficult, and
requires using the infinite-dimensional Grassmannian and the notion of Gauss
inclusion of a vector bundle p : E → B, defined as any map g : E → R∞
which is a linear monomorphism on the fibers.
The proof of (ii) is based on the two following lemmas.
Lemma (Local Triviality). For any vector bundle p : E → B with
paracompact base B there exists a countable open covering {Ui } such that
the restriction of p to each Ui is trivial.
Lemma (Gauss Inclusion). For any vector bundle p : E → B with
paracompact base B there exists a Gauss inclusion g : E → R∞ .
Item (iii) of the classification theorem is proved, in the general case, by
using a sophisticated argument based on presenting R∞ as the linear sum
98

Rev ⊕ Rod of the subspaces generated by the basis vectors with even and odd
coordinates. For the particular case in which the given bundle is the tangent
bundle τ of a smooth manifold M k , the proof is easy: we embed M k in some
RN , regard GNk as consisting of k-dimensional linear subspaces of R
N
and
k
apply the Gauss map, assigning to each point x ∈ M the linear subspace
obtained by parallel transport of the tangent plane Tx M k to the origin.
A good reference for the proofs is the book Fiber Bundles by D.Husemoller
(Rassloennye prostranstva, M., Nauka, 1970).

12.3. The category of G-bundles

A topological group is a topological space G with a group structure com-
patible with its topology, i.e., such that the maps G × G → G, (s, t) 7→ st,
and G → G, s 7→ s−1 , are continuous. We say that topological group G acts
from the right on a topological space X if we are given a continuous map
X × G → X, (x, g) 7→ xg (xg is called the image of the point x under the
action of the element g) such that
• x(gh) = (xg)h for all x ∈ X and all g, h ∈ G;
• x1 = x for all x ∈ X, 1 being the unit (neutral element of G).
A left action of a topological group G on a topological space X is defined
similarly. If G is Abelian, then left actions are also right actions and vice
versa (but the notations remain different).
Examples:
(1) the (natural) left actions of GL(n) and of O(n)) on Rn ;
(2) the (natural) right action of the group of orthogonal matrices O(k)
on the Stiefel manifold Vkn of orthonormal k-frames in Rn ;
(3) the left (=right) action of the nonzero real numbers on Rn by multi-
plication of the coordinates of points by these numbers.
A G-bundle p : E → B is the projection of a topological space E, supplied
with a right action of a topological group G, onto the orbit space B = E/G
of this action. Note that unlike vector bundles, G-bundles are not necessarily
locally trivial. Thus a G-bundle doesn’t have to be a fiber bundle, and of
course for a fiber bundle p there doesn’t have to exist a group G such that p
is a G-bundle.
A morphism of G-bundles is defined in the natural way. The class of all
G-bundles forms a category denoted BunG ; by fixing the base B, we obtain
one of its subcategories, denoted BunG (B).
 99

A G-bundle p : E → B is called principal if its fiber F = p−1 (b) is

(homeomorphic to) the group G for all b ∈ B. The class of all principal
G-bundles forms a category denoted PBunG ; by fixing the base B, we obtain
one of its subcategories, denoted PBunG (B).
Examples.
(1) The identification of opposite points on the n-dimensional sphere is a
principal Z2 bundle over RP 2 .
(2) The natural projection of the Stiefel manifold Vkn onto the Grassmann
manifold Gnk is a principal O(k)-bundle.

12.4. The Milnor construction

Let G be any topological group; denote by EG (n) := G ∗ G ∗ · · · ∗ G
(n factors) the n-fold iterated join of the group G with itself. Obviously

G ⊂ EG (2) ⊂ EG (3) ⊂ · · · ⊂ EG (n) ⊂ · · · ⊂ EG ,

where EG is the inductive limit of EG (n) as n → ∞.

Consider the action of G in EG by right shifts. The corresponding bundle

ωG : EG → BG = EG /G

is called the universal G-bundle, its base is called the classifying space of the
n
group G. In a similar way, one defines the bundle ωG : EGn → BG n
called
(briefly) the n-universal G-bundle, its base is called the n-classifying space
of the group G (in less shortened form, the classifying space up to dimension
n).
Examples:
(1) The classifying space of the group S1 is CP ∞ and ES1 = S∞ . The
k-classifying space of S1 is CP k and ESk1 = S2k+1 .
(2) The classifying space of the group Z2 is RP ∞ and EZ2 = S∞ . The
k-classifying space of Z2 is RP k and ESk1 = Sk .

12.5. Classification of principal G-bundles

The classification of principal G-bundles is similar, but more complicated,
than the one for vector bundles. The role of the canonical Grassmann bundle
here is played by Milnor’s universal G-bundle ωG : EG → BG , and the main
idea is the same: to take the pullback f ∗ (ωG ) of the universal bundle for
maps f : B → BG .
100

Theorem (Homotopy Invariance). If two maps f, g : B → BG of the

same space B to the classifying space BG are homotopic, then the corre-
sponding pullbacks f ∗ (ωG ) and g ∗ (ωG ) are isomorphic.
Corollary (Triviality). Any principal bundle over a contractible para-
compact space is trivial.
Theorem (Principal Bundle Classification). For any paracompact space
B, the assignment [B, BG ] 3 f 7→ f ∗ (ωG ) determines a bijection

[B, BG ] ←→ PBunG (B)

between the homotopy classes of maps of B into the classifying space BG and
the isomorphism classes of principal G-bundles over B.
We will not prove this theorem in the present course (see the book by
Hosemueller, loc.cit).

12.6. Bundles associated with principal G-bundles

Given a topological group G, a principal G-bundle ξ : E → B, and a
topological space F with a left action of G on it, one can construct in a
canonical way the associated a fiber bundle denoted by ξ[F ]; E[F ] → B by
“replacing the fiber G of ξ by F ”; more precisely, ξ[F ] is defined as follows: in
the space E×F , consider the right action of G defined by (e, f )g = (eg, g −1 f ),
and denote
E[F ] := (E × F )/G, and ξ[F ] := ξ ◦ pr1 ,
where pr1 is the projection pr1 (e, f ) = e.
Bundles associated with principal G-bundles are sometimes called bundles
with fiber F and structural group G or briefly (G, F )-bundles. Note that such
bundles are not necessarily locally trivial. A morphism of two bundles ξ[F ]
and ξ 0 [F ] is defined as a pair of maps ϕ̄ : B → B 0 , Φ̄ : E[F ] → E 0 [F ] forming
a commutative square with the bundle projections provided there exists a
morphism of the principal G-bundles (Φ, ϕ) : (E, B) → (E 0 , B 0 ) such that
Φ̄(e, f ) = Φ(e) and ϕ̄(b) = ϕ(b).
Thus, if a left action of a topological group G on a topological space F is
given, then we can consider the category Bun(G,F ) and, if the (paracompact)
base B is fixed, the category Bun(G,F ) (B).
The very rich theory of these categories is beyond the scope of this course.
 101

Lecture 13
MISCELLANY

In this very eclectic lecture, I have gathered some important constructions

and theorems (presented here without proofs) which should appear in any
serious introductory course in algebraic topology, but which I was unable to
treat in detail in the previous lectures.

13.1. The functors Hom, Tor, and Ext

Given two Abelian groups A, B, the group Hom(A, B) is defined as the
set of homomorphisms of A to B with the natural group structure. Using
Hom, we define the groups Tor(A, B) and Ext(A, B) as follows. Let
i p
0 −−−→ R −−−→ F −−−→ A −−−→ 0,
be the exact sequence which is the free resolution of the group A (so F is
the free group in the generators of A and R is the free group generated by
the relations of A). Tensoring this sequence with B and taking Hom(·, B) of
this sequence, we obtain two exact sequences
i⊗1 p⊗1
R ⊗ B −−−→ F ⊗ B −−−→ A ⊗ B −−−→ 0,
ī β
Hom(R, B) ←−−− Hom(F, B) ←−−− Hom(A, B) ←−−− 0,


Then we set Tor(A, B) := Ker(i ⊗ 1) and Ext(A, B) := Coker ī . It is not
difficult to check that Tor and Ext are well defined, i.e., do not depend on
the choice of the free resolution.
Examples.
(1) Tor(Z, B) = 0, Ext(Z, B) = 0;
(2) Tor(Zm , Z) = 0, Ext(Zm , Z) = Zm ;
(3) Tor(Zp , Zq ) = Z(p,q) , Ext(Zp , Zq ) = Z(p,q) , where (p, q) is the greatest
common divisor of p and q.
(4) Ext(A, T ) = T if A = Zk ⊕ T , where T is a finite group.
(5) Tor(G, B) = 0, Ext(A, G) = 0 if G is the additive group of one of the
fields Q, R, C.
(6) Tor(A, B) = 0 if A is an Abelian group such that Ker(µn ) = 0 for any
n ∈ N, where the homomorphism µn is given by µn (a) = na .
102

13.2. Universal coefficient formula

The universal coefficient formula allows to express the (co)homology groups
with any coefficient group G in terms of its (co)homology with integer coef-
ficients. We have the following theorem.
Theorem (Universal Coefficients) For any Abelian group G and any
simplicial space X, there exist two exact sequences

0 −−−→ Hk (X) ⊗ G −−−→ Hk (X; G) −−−→ Tor(Hk−1 (X), G) −−−→ 0,

0 ←−−− Hom(Hk (X), G) ←−−− H k (X; G) ←−−− Hom(Hk−1 (X), G) ←−−− 0;

these sequences split and therefore

Hk (X; G) ∼



= Hk (X) ⊗ G ⊕ Tor Hk−1 (X), G

H k (X; G) ∼



= Hom Hk (X), G ⊕ Ext Hk−1 (X), G .

Corollary. If G is the additive group of one of the fields Q, R, C, then

Hk (X; G) ∼
= Hk (X) ⊗ G,

and
H k (X; G) ∼


= Hom Hk (X), G .

13.3. Künneth formula

The Künneth formula expresses the (integer) homology of the Cartesian
product of two simplicial spaces in terms of the homology of the factors,
namely

Hk (X × Y ) ∼
M
M
= Hl (X) ⊗ Hm (Y ) ⊕ Tor(Hl (X), Hm (Y )) .
l+m=k l+m=k−1

For the case in which the coefficient group is G, the additive group of one
of the fields Q, R, C, then we have the simpler relation

Hk (X × Y ) ∼
M

= Hl (X) ⊗ Hm (Y ) .
l+m=k
 103

13.4. Alexander duality

Alexander duality expresses the reduced (co)homology of the complement
to a submanifold in Sn in terms of the manifold’s (co)homology.
Theorem (Alexander Duality). If M is a submanifold (not necessarily
smooth) in the n-sphere Sn , then, for any k,

e k (M ) ∼
H =H e k (M ) ∼
e n−k−1 (Sn \ M ), H =He n−k−1 (Sn \ M ) .

Corollary (Jordan-Brouwer Theorem). Any (n − 1)-sphere Sn−1 embedded

in Rn splits Rn into two connected components.
13.5. The Poincaré–Hopf theorem
This generalizes Poincaré’s theorem on vector fields on surfaces.
Theorem (Poincaré–Hopf). The index of a generic vector field on a
closed smooth orientable manifold equals the manifold’s Euler characteristic.
13.6. C̆ech homology
C̆ech homology, which is defined for arbitrary topological spaces, was
invented (before singular homology) by P.S.Alexandrov (and not by C̆ech).
We do not describe it in detail here, only indicating the two main ideas
underlying its definition.
Given an arbitrary covering of ω = {Uα } of a topological space X, let us
define the nerve Nω of this covering as the family of simplices corresponding
to this covering: the 0-simplices (vertices)

are the open sets of the covering ω,
the 1-simplices are pairs Uα1 , Uα2 such that Uα1 ∩ Uα2 6= ∅, the 2-simplices
are triplets of vertices having at least one common point, etc.

Figure 13.1 Nerve of a covering

104

Chain groups (say with integer coefficients) Ck (Nω ) are defined as the sets
of all linear combinations of k-simplices, the boundary operator is defined as
in simplicial homology, and it yields the homology groups Hk (Nω ).
It should be intuitively clear that if X is a “nice” enough and the covering
ω is “sufficiently fine”, then the nerve Nω is a “good approximation” of the
topology of X so that Hk (Nω ) is a good approximation of the k-homology of
the space X.
Alexandrov’s second main idea, roughly speaking, is based on the fact
that if one covering is inscribed in another, there is a natural projection of
the homology of the inscribed nerve onto the homology of the other nerve,
and the definition of the homology groups is obtained by taking the projective
limit of these groups.
It can be proved that the C̆ech homology functor satisfies the Steenrod-
Eilenberg axioms and therefore coincides, say, with the singular homology
functor.