Satya Algebraic

Satya Deo
Algebraic Topology
A Primer
Second Edition
123
Satya Deo
Harish-Chandra Research Institute
Allahabad
India
ISSN 2366-8725 (electronic)

Texts and Readings in Mathematics
ISBN 978-981-10-8734-9 (eBook)
https://doi.org/10.1007/978-981-10-8734-9
Library of Congress Control Number: 2018934679
This work is a co-publication with Hindustan Book Agency, New Delhi, licensed for sale in all countries
in electronic form only. Sold and distributed in print across the world by Hindustan Book Agency, P-19
Green Park Extension, New Delhi 110016, India. ISBN: 978-93-86279-67-5 © Hindustan Book Agency
2018.
1st edition: © Hindustan Book Agency (lndia) 2003;

Corrected Reprint, 2006, Hindustan Book Agency (lndia)
2nd edition: © Springer Nature Singapore Pte Ltd. 2018 and Hindustan Book Agency 2018
Preface
Algebraic Topology is an important branch of topology having several con-

nections with many areas of modern mathematics. Its growth and influence,
particularly since the early forties of the twentieth century, has been remark-
ably high. Presently, it is being taught in many universities at the M.A./M.Sc.
or beginning Ph.D. level as a compulsory or as an elective course. It is best
suited for those who have already had an introductory course in topology as
well as in algebra. There are several excellent books, starting with the first
monograph ‘Foundations of Algebraic Topology’ by S. Eilenberg and N.E.
Steenrod, which can be prescribed as a textbook for a first course on algebraic
topology by making proper selections. However, there is no general agreement
on what should be the ‘first course’ in this subject. Experience suggests that
a comprehensive coverage of the topology of simplicial complexes, simplicial
homology of polyhedra, fundamental groups, covering spaces and some of
their classical applications like invariance of dimension of Euclidean spaces,
Brouwer’s Fixed Point Theorem, etc. are the essential minimum which must
find a place in a beginning course on algebraic topology. Having learnt these
basic concepts and their powerful techniques, one can then go on in any
direction of the subject at an advanced level depending on one’s interest and
requirement.
This book is designed to serve as a textbook for a first course in algebraic

topology described above. In order to lay down the real foundation of algebraic
topology, we have also included a brief introduction to singular homology and
cohomology. Approach to the contents has been dictated by several advanced
courses taught by the author on different topics of algebraic topology at
many universities. It is necessary to mention that the subject appears a bit
abstract to begin with, but after a while it presents concrete topological as
well as geometrical results of great insight and depth. The beginning student is
expected to have some patience before appreciating the depth of these results.
Maximum care has been taken to emphasize the subtle points of the concepts
and results in a lucid manner. Examples are given at every stage to illustrate
and bring out the underlying concepts and the results. It is hoped that the
detailed explanations will help the student to grasp the results correctly and
to have a sound understanding of the subject with confidence. The prereq-
uisites for the study of this book are very little, and these have been briefly
discussed in Chapter 1 and the Appendix to make it self-contained. These are
included to maintain continuity between what the student has already learnt
and what he is going to learn. It may be either quickly reviewed or even skipped.
We introduce and study the fundamental groups and its properties in

Chapter 2. Starting with the concept of pointed spaces we show that the
fundamental groups are topological invariants of path-connected spaces. After
computing the fundamental group of the circle, we show how it can be used
to compute fundamental groups of other spaces by geometric methods. In
Chapter 3, we explain the topology of simplicial complexes, introduce the
notion of barycentric subdivision and then prove the simplicial approximation
theorem. In Chapter 4, we introduce the first classical homology theory,
viz., the simplicial homology of a simplicial complex and then proceed to
define the simplicial homology of a compact polyhedron. We provide detailed
discussion of how a continuous map between compact polyhedra induces a
homomorphism in their simplicial homology, and then prove the topological
invariance of these groups. A few classical applications of the techniques of
homology groups include the proof of the fact that the Euclidean spaces of
different dimensions are not homeomorphic, Brouwer’s Fixed Point Theorem,
Lefschetz Fixed Point Theorem and the Borsuk-Ulam Theorem, etc. In
Chapter 5, we study the theory of covering projections and its relation with
fundamental groups. Important results on lifting of a map, classification of
covering projections and the universal coverings are discussed at length.
The final topic dealing with the singular homology and cohomology has
been discussed in Chapter 6. We have presented the important properties of
singular homology including proofs of the homotopy axiom and the excision
axiom. All of this is done in a way so that singular homology gets established
as a ‘homology theory’ on the category of all topological pairs in the sense
of Eilenberg-Steenrod. Then we go on to formally explain the definitions
of an abstract homology as well as an abstract cohomology theory. Apart
from the singular homology, we show that simplicial homology is also a
homology theory on the category of all compact polyhedral pairs. Related
topics such as homology and cohomology with coefficients, the Universal
Coefficient Theorem, the Künneth Formula, the Mayer-Vietoris Sequence and
the cohomology algebra of a space etc., have been discussed and adequately
illustrated by examples. The student can test his understanding of the subject
by working out the exercises given in every chapter. All of the basic material
covered in this book should make the subject fascinating for beginners and
should enable them to pursue advanced courses in any branch of algebraic
topology.
Except for a few remarks, we have refrained from giving a detailed history
of the subject which is, of course, always very enlightening and interesting.
However, we feel that it can be appreciated adequately when the student has
learnt the subject at a more advanced level than what is presented here. The
material presented here is contained in most of the books on algebraic topology.
My contribution is basically in organizing and presenting it afresh. Definitions,
propositions, corollaries etc. are numbered by 3 digits, i.e., Theorem 4.5.1
means Theorem 1 of Section 5 of Chapter 4. A suggested guideline for teaching
a two-semester (or one academic year) course from this book is as follows:
in the first semester one can cover Fundamental Group (Sections 2.1 to 2.6),
Simplicial Complexes (Sections 3.1 to 3.4), Simplicial Homology (Sections 4.1
to 4.8) and Covering Projection (Sections 5.1 to 5.6). The chapter on Covering
Projection can be taught immediately after the chapter on Fundamental
Group. Proofs of some of the technical theorems given in Chapters 4 and 5
can be omitted. The remaining section of Chapter 4 on Simplicial Homology,
viz. Section 4.9 and the whole of Chapter 6 on Singular Homology can be
easily covered in the second semester. It is hoped that the material prescribed
for the first semester is quite adequate for acquainting the student with the
basic concepts of algebraic topology. The material for the second semester, on
the other hand, will allow him to take a deeper plunge in any of the advanced
topics of modern algebraic topology.
The author would like to express his thanks to many of his friends
and colleagues, specially to G.A. Swarup, A.R. Shastri, Ravi Kulkarni and
K. Varadarajan, all of whom have read the manuscript and made important
suggestions for improvement of the book. The author appreciates the facil-
ities made available to him by the Department of Mathematics, University
of Arkansas, where the first draft of the whole book was completed. Rajendra
Bhatia, the Managing Editor of the series ‘Texts and Readings in Mathematics’
of the Hindustan Book Agency is thanked for his keen interest in the edito-
rial work. I thank my wife Prema Tripathi for her insistence that the job of
completing the book should get preference over administrative duties. Finally,
J.K. Maitra, one of my students and a colleague, deserves special thanks for
his constant assistance in typesetting the book on LATEX.
Satya Deo
July 1, 2003 Jabalpur
Preface to the Second Edition
In this edition of ‘Algebraic Topology, A Primer’, some necessary changes

have been made in Chapters 3, 4 and 6. In the original edition, a simplicial
complex was assumed, for simplicity, to be finite, which meant that all
the polyhedra were assumed to be compact. This was indeed a sweeping
restriction, especially when one is defining singular homology on the category
of all topological pairs. Therefore, Chapter 3, has now been expanded by
including another section on general simplicial complexes. As a consequence
of this, basic results on the topology of a simplicial complex have also been
included. Then, Chapter 4, on simplicial homology has been expanded to
include simplicial homology of an arbitrary polyhedral pair. This allows us to
state the important fact later in Chapter 6 that the simplicial homology is a
homology theory on the category of all polyhedral pairs and their maps in the
sense of Eilenberg-Steenrod.
Earlier, the Chapter 6 on singular homology was really somewhat

sketchy. It has now been expanded with more details. In order that the book
remains only elementary and within size, we have not included the proof
of theorems such as the Acyclic Model Theorem, Eilenberg-Zilber Theorem
nor the properties of the cup products in singular cohomology. However, the
classical applications like the Jordan-Brouwer Separation Theorem, Jordan
Curve Theorem and the Invariance of Domain Theorem, etc. have now been
included in reasonable detail.
Many colleagues and students have pointed out typos and even a few
mathematical errors in the original edition. The author is thankful to all of
those, especially to H.K. Mukherjee, Krishnendu Gangopadhyay and Snigdha
Bharati Choudhury, who read the book carefully and made many comments.
All the corrections have now been made to the best of my knowledge. The
numbering of definitions, theorems, corollaries, propositions, lemmas, examples
and remarks have now been made in one sequence, i.e., Proposition 3.2 will
follow Lemma 3.1 and Theorem 6.4 will follow Proposition 6.3 etc. Some
exercises have been changed and the remaining ones have been properly
ordered. The author is thankful to all those who followed this book as a text
for a course on the subject and offered their valuable suggestions for improving
the book. The ‘Primer Character’ of the book, however, has not been changed.
My student Dr V.V. Awasthi of VNIT, Nagpur deserves thanks for helping me
in the typesetting of the text. Finally, the author expresses his sincere thanks
to the authorities of the Harish-Chandra Research Institute, Allahabad for
the wonderful facilities and to the National Academy of Sciences, India for the
Platinum Jubilee Senior Scientist Fellowship.
Satya Deo
Harish-Chandra Research Institute
Allahabad.
May 29, 2017
Contents
Preface v
1 Basic Topology: a review 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Euclidean Spaces and their Subspaces . . . . . . . . . . . . . 1
1.3 Continuous Maps and Product Spaces . . . . . . . . . . . . . 4
1.4 Homeomorphisms and Examples . . . . . . . . . . . . . . . . 6
1.5 Quotient Spaces . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.6 Connected and Path-Connected Spaces . . . . . . . . . . . . 17
1.7 Compact Spaces and Locally Compact Spaces . . . . . . . . 23
1.8 Compact Surfaces . . . . . . . . . . . . . . . . . . . . . . . . 27
1.9 What is Algebraic Topology? . . . . . . . . . . . . . . . . . . 30
2 The Fundamental Group 35

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.2 Homotopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.3 Contractible Spaces and Homotopy Type . . . . . . . . . . . 41
2.4 Fundamental Group and its Properties . . . . . . . . . . . . 50
2.5 Simply-Connected Spaces . . . . . . . . . . . . . . . . . . . . 67
2.6 Results for Computing Fundamental Groups . . . . . . . . . 70
3 Simplicial Complexes 83
3.1 Finite Simplicial Complexes . . . . . . . . . . . . . . . . . . . 83
3.2 Polyhedra and Triangulations . . . . . . . . . . . . . . . . . . 93
3.3 Simplicial Approximation . . . . . . . . . . . . . . . . . . . . 102
3.4 Barycentric Subdivision – Simplicial Approximation Theorem 106
3.5 General Simplicial Complexes . . . . . . . . . . . . . . . . . . 115
4 Simplicial Homology 123

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
4.2 Orientation of Simplicial Complexes . . . . . . . . . . . . . . 124
4.3 Simplicial Chain Complex and Homology . . . . . . . . . . . 127
4.4 Some Examples . . . . . . . . . . . . . . . . . . . . . . . . . 134
4.5 Properties of Integral Homology Groups . . . . . . . . . . . . 143
ix
4.6 Induced Homomorphisms . . . . . . . . . . . . . . . . . . . . 158
4.7 Some Applications . . . . . . . . . . . . . . . . . . . . . . . . 162
4.8 Degree of a Map and its Applications . . . . . . . . . . . . . 167
4.9 Invariance of Homology Groups . . . . . . . . . . . . . . . . . 174
4.9.1 Subdivision Chain Map . . . . . . . . . . . . . . . . . . 174
4.9.2 Homomorphism Induced by a Continuous Map . . . . . 179
4.9.3 Homotopy Invariance . . . . . . . . . . . . . . . . . . . 180
4.9.4 Lefschetz Fixed-Point Theorem . . . . . . . . . . . . . . 182
4.9.5 The Borsuk-Ulam Theorem . . . . . . . . . . . . . . . . 186
4.10 Homology of General Simplicial Complexes . . . . . . . . . . 192
5 Covering Projections 199

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
5.2 Properties of Covering Projections . . . . . . . . . . . . . . . 203
5.3 Applications of Homotopy Lifting Theorem . . . . . . . . . . 208
5.4 Lifting of an arbitrary map . . . . . . . . . . . . . . . . . . . 212
5.5 Covering Homomorphisms . . . . . . . . . . . . . . . . . . . . 214
5.6 Universal Covering Space – Applications . . . . . . . . . . . 219
6 Singular Homology 227

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
6.2 Singular Chain Complex . . . . . . . . . . . . . . . . . . . . . 229
6.3 One-Dimensional Homology and the Fundamental Group . . 234
6.4 Homotopy Axiom for Singular Homology . . . . . . . . . . . 245
6.5 Relative Homology and the Axioms . . . . . . . . . . . . . . 248
6.6 The Excision Theorem . . . . . . . . . . . . . . . . . . . . . . 252
6.7 Homology and Cohomology Theories . . . . . . . . . . . . . . 258
6.8 Singular Homology with Coefficients . . . . . . . . . . . . . . 264
6.9 Mayer-Vietoris Sequence for Singular Homology . . . . . . . 272
6.10 Some Classical Applications . . . . . . . . . . . . . . . . . . . 277
6.11 Singular Cohomology and Cohomology Algebra . . . . . . . . 283
7 Appendix 295
7.1 Basic Algebra – a Review . . . . . . . . . . . . . . . . . . . . 295
7.1.1 Groups and Homomorphisms . . . . . . . . . . . . . . . 295
7.1.2 Direct Product and Direct Sum . . . . . . . . . . . . . . 297
7.1.3 The Structure of a Finite Abelian Group . . . . . . . . 298
7.1.4 Free Groups and Free Products . . . . . . . . . . . . . . 300
7.1.5 Modules and their Direct Sum . . . . . . . . . . . . . . 304
7.2 Categories and Functors . . . . . . . . . . . . . . . . . . . . . 306
7.2.1 The Hom(M, N ) Functor . . . . . . . . . . . . . . . . . 310
7.2.2 Exact Sequences . . . . . . . . . . . . . . . . . . . . . . 311
7.2.3 The Tensor Product of Modules and Homomorphisms . 314
7.2.4 Chain Complexes and Homology . . . . . . . . . . . . . 324
7.2.5 Tensor Product of two Chain Complexes . . . . . . . . . 327
7.2.6 Exact Homology Sequence Theorem . . . . . . . . . . . 329
7.3 Topological Transformation Groups . . . . . . . . . . . . . . 331
7.3.1 Topological Transformation Groups . . . . . . . . . . . 333
References 339
Index 341
Chapter 1
Basic Topology: a review
1.1 Introduction
In this chapter, we briefly recall and collect some of the basic definitions
and results of point set topology which will be needed later for explaining the
concepts of algebraic topology. Important definitions are followed by some quick
examples and, if necessary, by some comments. The results are mostly stated
without proofs, but occasionally well-known proofs are indicated merely to
highlight some important routine arguments. Emphasis has been laid mostly
on those topics of set topology which appear necessary for a correct under-
standing of subsequent chapters. We assume that the student is somewhat
already familiar with the basic notions of set topology, particularly with quo-
tient spaces, connectedness and compactness, etc., which are used here freely
regardless of the order in which they occur in these preliminaries. Definitions
and results can be referred back in the book while reading the material of
algebraic topology as and when they are required.
1.2 Euclidean Spaces and their Subspaces
Let Rn denote the n-fold cartesian product of the real line R. Then an ar-
bitrary point x = (x1 , . . . , xn ) of Rn will have n real components x1 , x2 , . . . , xn .
The set Rn has several kinds of interesting and useful mathematical structures.
It is an n-dimensional vector space over the field R of reals; it has the Euclidean
inner product with respect to which it is a Hilbert space. The set of all n × n
matrices over reals corresponds to the set of all linear transformations on the
vector space Rn . In this section, we recall the usual topology on Rn . By Eu-
clidean metric on Rn , we mean the distance function d defined by the following
formula for any pair of points x = (x1 , . . . , xn ), y = (y1 , . . . , yn ) in Rn :
n
d(x, y) = { i=1 (xi − yi )2 }1/2 .
2 Chapter 1. Basic Topology: a review
This is easily seen to be a metric on Rn . For any x ∈ Rn , the distance of x from

the origin, denoted by |x|, is called the Euclidean norm of x. For any x ∈ Rn
and for any r ≥ 0, the subset B = {y ∈ Rn | d(x, y) < r} is called the open ball
(or ball) of radius r with center x. Using these balls, we define a topology τd on
Rn as follows: A subset U of Rn is a member of τd or is an open set if for each
point x ∈ U , there exists a ball of some radius r > 0 such that B(x, r) ⊆ U .
It can be easily verified that τd is indeed a topology on Rn . This topology is
called the usual topology or Euclidean topology on Rn . More generally, for any
metric space (X, ρ), the same procedure defines a topology τρ on the set X
and the resulting topology is called the topology on X induced by the metric
ρ. It is in this way that a metric space is always regarded as a topological space.
For any subset Y of Rn , we have a metric on Y induced by the Euclidean

metric of Rn . On the other hand, the Euclidean topology of Rn induces the
subspace topology on Y . It can be easily verified that the topology on Y ,
induced by the metric of Y , is identical with the above subspace topology. In
the sequel, whenever we refer to a subspace of Rn , which we will frequently
do, it will be assumed to have the induced topology from Rn , unless explicitly
stated otherwise.
Note that the subset Z of integers gets the discrete topology from R. On
the other hand, the set Q of rational numbers gets a topology from R which
is totally disconnected but is not discrete. The unit interval I = [0, 1] gets the
induced topology, which is compact as well as connected. The subspace
Sn = {x ∈ Rn+1 | |x| = 1}
is called the standard n-sphere.
X
Fig. 1.1: The torus in Euclidean 3-space
Again, it is a compact, connected subset of the Euclidean space Rn+1 .

Note that S0 = {1, −1} ⊂ R has just two points and is discrete whereas
1.2. Euclidean Spaces and their Subspaces 3
S1 = {x ∈ R2 | |x| = 1} is the set of all unimodular complex numbers. It is

a topological group with respect to multiplication of complex numbers and
is called
kthe circle group. This group is evidently abelian; its k-fold product
T k = 1 S1 is called the k-dimensional torus group, which too is a compact,
connected, abelian topological group. It is canonically a subspace of R2n = Cn ,
where C stands for the set of complex numbers. We are identifying R2n as
Cn by saying that every pair (x1 , x2 ) of real numbers can be regarded as a
complex number z = x1 + ix2 and, therefore, every 2n-tuple (x1 , . . . , x2n )
can be thought of as an n-tuple (z1 , . . . , zn ) of complex numbers where
zi = (x2i−1 , x2i ), i = 1, . . . , n. It may be mentioned here that the set C of
complex numbers has an additional structure, viz., multiplication of complex
numbers defined on the set of points of the plane R2 .
For each n ≥ 1, the compact, connected subset Dn = {(x1 , . . . , xn ) ∈

n
R | 1 x2i ≤ 1} of Rn is called an n-disk or an n-cell. Its boundary is clearly
n
S n−1
. In the 3-dimensional Euclidean space, let us consider the surface S ob-
tained by rotating the circle {(y − 3)2 + z 2 = 1, x = 0} about the z-axis. This
surface, which is a compact, connected subset of R3 , is homeomorphic to the
2-dimensional torus S1 × S1 mentioned earlier. To see why these are homeomor-
phic one has to observe that both of these spaces are compact Hausdorff spaces
and there is a continuous bijection (Exercise) from S onto S1 × S1 . Further-
more, we will prove later (see the Section 1.7) that any continuous bijection
from a compact space onto a Hausdorff space is always a homeomorphism. Our
assertion now follows immediately from these facts. We will frequently use this
type of argument in the sequel.
Simplexes
Let v0 , v1 , . . . , vn be n + 1 points of Rn , n ≥ 1, so that the set of vectors
{v1 − v0 , v2 − v0 , . . . , vn − v0 } is linearly independent. Then the subset σ n of
Rn defined by
n n
σ n = {x ∈ Rn | x = 0 αi vi , 0 ≤ αi ≤ 1 and 0 αi = 1},
is called an n-simplex.
v0
v1 v3
v2
Fig. 1.1(a): A 3-simplex

Note that the above procedure cannot be used to define a (n + 1)-simplex

in Rn because we cannot find n+1 linearly independent vectors in Rn . However,
a k-simplex σ k can be defined in Rn for all k = 0, 1, . . . , n. A 0-simplex is just
a point of Rn . Also, notice that a 1-simplex is just a line segment joining two
points whereas a 2-simplex is a triangle (with interior) obtained by taking the
convex linear combinations of three points which are not on the same straight
line. A 3-simplex is a tetrahedron, etc. An n-simplex σ n is a compact, con-
nected subset of the Euclidean space which carries a good geometric structure.
When several simplexes, possibly of different dimensions, are suitably combined
together, we get objects called simplicial complexes. These objects carry a very
nice geometric structure and possess a topology which retains the Euclidean
topology of every simplex forming the simplicial complex. A separate chapter
has been devoted to the study of these simplicial complexes and maps between
them, called simplicial maps. Here is a question which one may like to answer:
Is there a space X that admits n-simplexes for all n ≥ 0? In other words, what
is that space X in which all n-simplexes, n ≥ 0, are embedded, with each n-
simplex embedded in the next (n + 1)-simplex for all n ≥ 0? The answer will
become clear from the discussion of the concept of product spaces.
1.3 Continuous Maps and Product Spaces
A map f : X → Y between two topological spaces X and Y is said to

be continuous if the inverse image f −1 (V ) of each open set V of Y is open
in X. This is equivalent to saying that the inverse image of every closed
subset of Y is closed in X. Continuity of a map is indeed a local property. To
understand this, let x ∈ X and W be a neighbourhood of the point f (x) in
Y . This means there is an open set V of Y such that f (x) ∈ V ⊂ W . Now,
continuity of f implies that f −1 (V ) is an open set of X containing the point
x. Hence N = f −1 (V ) is a neighbourhood of x and has the property that
f (N ) = f f −1 (V ) ⊆ V ⊆ W . Thus, what we have proved is that f : X → Y
continuous means for every x ∈ X and for every neighbourhood W of f (x),
there exists a neighbourhood N of x in X such that f (N ) ⊆ W . This condition
is taken as the definition of continuity at the point x ∈ X. In other words,
we say that f is continuous at x ∈ X if for each neighbourhood W of f (x),
there exists a neighbourhood N of x such that f (N ) ⊆ W . This is quite valid
because if, according to this definition, f : X → Y is continuous at each point
x ∈ X, then f is easily seen to be continuous. To prove this last statement,
let V be an open set of Y . Suppose x ∈ f −1 (V ). This means f (x) ∈ V and
V is a neighbourhood of f (x) in Y . By the given condition, there exists a
neighbourhood N of x in X such that f (N ) ⊆ V . This means N ⊆ f −1 (V )
and so f −1 (V ) is also a neighbourhood of x. Since f −1 (V ) has just been shown
to be a neighbourhood of each of its points, it is an open set of X. Hence f is
continuous.
1.3. Continuous Maps and Product Spaces 5
In order to verify that a particular map f : X → Y is continuous, it is

frequently easy and convenient to check that f is continuous at each point
x ∈ X. It must now be mentioned here that the classical (, δ)-definition of
continuity of a function in analysis is just another way of expressing the local
definition of continuity in the context of metric spaces. It says that a map
f : (X, d) → (Y, d ) is continuous at x = x0 ∈ X if for all > 0, ∃ δ > 0
such that d(x, x0 ) < δ implies that d (f (x), f (x0 )) < . Interpreted in terms of
neighbourhoods, this definition says that for each -ball around f (x0 ), there
exists a δ-ball around x0 in X such that f (B(x0 , δ)) ⊆ B(f (x0 ), ). This is
precisely the definition of local continuity of the map f at x0 ∈ X.
There are several standard methods of constructing new topological spaces

from the given ones. Defining product spaces is one such method. Let {Xα |
∈ I} be a family of topological spaces, and
α consider the cartesian product
Xα , α ∈ I of sets Xα . For each α,
let pα : Xα → Xα denote the canonical
projection. Then the topology on Xα generated by subsets
{p−1
α (Uα ) | Uα is open in Xα , α ∈ I},

is called the product topology or the Tychonoff topology on Xα . It is impor-
tant to understand the basic open sets of the above product topology – these
are, by definition, just finite intersections of the sets of the above type. Thus,
a basic open set is
p−1 −1
α1 (Uα1 ) ∩ · · · ∩ pαn (Uαn ),
where Uαi is an open set of Xαi , i = 1, 2, . . . , n. It is necessary to realize that

the above basic open set is just α∈I Aα , where Aαi = Uαi for i = 1, 2, . . . , n
and Aα = Xα for all α
= α1 , . . . , αn . This basic open set is usually denoted by
Uα1 , . . . , U
αn . Using these basic open sets, it is easy to see that projection
maps pα : Xα → Xα are continuous, open surjections. The product topology
can also be characterized as the smallest topology on the set Xα making all
projection maps continuous. It must be pointed out here that the projection
maps need not be closed. For instance, consider the projection from the plane
R2 to either of the factors. The set F = {(x, x1 ) | x
= 0} is a closed subset of
R2 but its projection on either factor is R − {0}, which is not closed in R.
Another important observation which can be inferred from the description

of basic open sets is: An infinite product of discrete spaces {Xα | α ∈ I}, each
space having more than one point, is never discrete. Note that if the product
space was discrete, then each singleton {xα } of the product must be open and
so must contain a basic open set Uα , which is impossible!
A space X is embedded in the product space X × Y under the inclusion

map x → (x, y0 ), where y0 ∈ Y is a fixed point. The restriction of the projection
map X × Y → X to the subspace X × {y0 } is the continuous inverse of this
inclusion. If Y = {y0 } consists of a single point, then X is homeomorphic to
X × Y and we don’t get anything new. In the case of Euclidean spaces, we

note that Rn is embedded in Rn+1 under the map x → (x, 0), where the last
coordinate is zero. It is in this sense that we sometimes write:
R ⊂ R2 ⊂ R3 ⊂ · · · ⊂ Rn ⊂ · · · ,
and their union is denoted by R∞ = ∪Rn . It is also true that Rn is a vector
subspace of R n+1
for any n ≥ 1. The product R = R of countably infinite
ω
number of real lines is a space which contains Rn as a subspace for all n and so
it contains R∞ . This means R∞ has the subspace topology induced from Rω .
We can also topologize the set R∞ by taking the weak topology (see later at
the end of Section 1.5) induced by the inclusion maps Rn → R∞ , n ≥ 1. Here
we remark that the two topologies on R∞ are not the same (Exercises later
on). It now follows that either of the two spaces Rω or R∞ (with subspace
topology) can be taken as a universal space for n-simplices for all n ≥ 0.
A nice result, which is used frequently in showing that a map going into
a product space is continuous, and can be easily proved, is the following:

A map f : Y → Xα is continuous if and only if the composites pα ◦ f
are continuous for each projection map pα : X α → Xα .
1.4 Homeomorphisms and Examples
The following simple result about continuity of maps will be frequently

used while dealing with paths and homotopies in the chapter on fundamental
groups.
Lemma 1.4.1. (Continuity Lemma). Let a space X be written as X = F1 ∪ F2 ,

where F1 , F2 are closed subsets of X. Suppose f : X → Y is a map such that
f |F1 , f |F2 are continuous. Then the map f itself is continuous.
Proof. Let B ⊂ Y be a closed set in Y . Then f −1 (B) = ( f |F1 )−1 (B) ∪
( f |F2 )−1 (B). Since ( f |F1 )−1 (B) is closed in F1 and F1 is closed in X, we find
that ( f |F1 )−1 (B) is closed in X. Similarly, ( f |F2 )−1 (B) is also closed in X.
Therefore, f −1 (B), being union of two closed sets, is closed. This proves the
lemma.
One can easily generalize the above Lemma when X = ∪ni=1 Fi is a finite
union of closed sets. We must point out that if the sets F1 , F2 are not closed,
then the lemma is false. For example, consider the characteristic function
χQ : R → I of rationals Q defined by χQ (x) = 1, x ∈ Q, and 0 otherwise; being
constant on rationals as well as on irrationals, its restriction on each one of
them is continuous, but the function χQ itself is discontinuous as a function
from R → I.
1.4. Homeomorphisms and Examples 7
Recall that a map f : X → Y is said to be a homeomorphism if f is a

bijection and both f, f −1 are continuous. A space X is called homeomorphic
to a space Y if there exists a homeomorphism from one to the other and, in
topology, two homeomorphic spaces are considered to be topologically same or
equivalent. Note that the composite of two homeomorphisms is a homeomor-
phism, inverse of a homeomorphism is again a homeomorphism. It follows that
for any space X, the set of all self-homeomorphisms of X forms a group with
respect to composition of maps. It is an interesting problem to determine the
group Homeo(X) of all homeomorphisms of a given space X.
Example 1.4.2. Any two open intervals (finite or infinite) of the real line R
are homeomorphic. To see this, note that the map f : (a, b) → (0, 1), a < b,
defined by f (x) = (x − a)/(b − a) is a homeomorphism. Moreover, the map
g : R → (−1, 1) defined by g(x) = x/(1 + |x|) is a homeomorphism, whereas the
map h : (0, π/2) → (0, ∞) defined by h(x) = tan x is also a homeomorphism.
Combining all these homeomorphisms one can easily prove the result stated
above. One can, similarly, show that any two closed intervals are homeomorphic.
Example 1.4.3. For each n ≥ 1, the boundary B(σ n ) of the n-simplex σ n is

homeomorphic to the sphere Sn−1 . To see this, first we observe that if an object
in an Euclidean space is moved under some translation, rotation or scaling
(downward or upward) by a nonzero scalar, then the object changes only up
to homeomorphism. Hence, without any loss of generality, we can assume that
the sphere Sn−1 is placed inside the n-simplex σ n ( imagine the case n = 3 as
shown in the Fig.1.2).
v0
v3
v1 v2
Fig. 1.2: The radial map
Take a point P in the interior of σ n . If we draw a ray starting from P along

any direction then the ray will intersect each of the sphere and the simplex in
question, at exactly one point, say, R1 and R2 , respectively. We define a map
h : Sn−1 → B(σ n ) simply by taking R1 to R2 . It is now evident that this map
is a bijective continuous map having a continuous inverse and, therefore, h is
indeed a homeomorphism. The map h defined above is called the radial map.
An interesting generalization of the foregoing example is discussed in

Proposition 3.2.1.
Example 1.4.4. Let Dn be an n-disk, I n be an n-cube and σ n be an n-simplex.

We will now show that any two of these three spaces is homeomorphic to each
other. For simplicity of explanation, let us take the case of a 2-simplex and a
square. As remarked in the previous example, we can assume that the simplex
is inside the square (see Fig. 1.3).
Q
P
R
2 IxI
Fig. 1.3
Again, take a point P inside the simplex and consider the radial map
emanating from P . In this case, every ray defines a unique line segment P Q of
the simplex and a unique line segment P R of the square. We map the segment
P Q to the segment P R by a linear homeomorphism. It is now plain to see that
when these linear homeomorphisms along all possible directions are combined
together, we get a homeomorphism mapping the 2-simplex to the square con-
taining the simplex. The argument is clearly valid for all dimensions. The same
argument applies to the case of an n-disk and an n-cube. It may be noted that
the above homeomorphism restricted to the boundary of a disk, a simplex or
a cube gives us the radial map of the previous example.
Example 1.4.5 (Stereographic Projection). Let X = S2 − {(0, 0, 1)}, where S2

is the standard 2-sphere in R3 . Define a map f : X → R2 by the formula:
f (x1 , x2 , x3 ) = (x1 /(1 − x3 ), x2 /(1 − x3 ), 0),
where R is embedded in R3 by taking the third coordinate to be zero. It can
2
be verified that f is a continuous bijection which has a continuous inverse.

Therefore, f is a homeomorphism. This particular map is called stereographic
projection (Fig. 1.4). The same map can be generalized to give a stereographic
projection from Sn − {(0, 0, . . . , 0, 1)} → Rn , which again will be a homeo-
morphism. This fact is sometimes expressed by saying that an n-sphere, when
punctured at a point, yields the Euclidean space Rn .
Now, one can see something more: Consider S2 punctured at two points,
say, at the north pole and at the south pole (they don’t make any difference
though). Then the resulting space is homeomorphic to the hyperboloid of one
sheet as well as to a right circular cylinder having infinite length on both
sides. Again, here the length is not important – just as a finite open interval is
1.4. Homeomorphisms and Examples 9
2
S
y
Fig. 1.4: Stereographic Projection
homeomorphic to the whole real line, an infinite cylinder is homeomorphic to

a cylinder of finite height provided its boundary circles are removed; the same
thing applies to the hyperboloid of one sheet also. A similar consideration shows
that an open cylinder is homeomorphic to an open annulus in the plane. It also
follows that the plane punctured at one point is homeomorphic to an open
cylinder (Fig. 1.5).
Fig. 1.5: Homeomorphic spaces
Exercises
1. Prove that if we remove a point from the ellipse 4x2 + 9y 2 = 1, then the
remaining space is homeomorphic to the open interval (0, 1).
2. Use the (, δ)-definition of a continuous map to show that the map f : R →
R defined by

sin x1 x
= 0
f (x) =
0 x=0
is not continuous, but the map g : R → R defined by

x · sin x1 x
= 0
g(x) =
0 x=0
is continuous.
3. Give a continuous map from the interval [−1, 1] onto the boundary of
the sqare I × I, which makes exactly five rounds of the boundary of the
square.
4. A family {Aα } of subsets of a space X is called locally finite if each point x
of X has a neighbourhood N which meets only finitely many members of
{Aα }. Let {Fα } be a locally finite family of closed sets such that X = ∪Fα .
If f : X → Y is a map such that f |Fα is continuous for each α, then show
that f is continuous.
5. Show, by an example, that an injective continuous map from the half-open
interval [0, 1) into the Euclidean plane need not be a homeomorphism.
6. Show that the surface of a cone (end-circle included) is homeomorphic
to the closed 2-disk D2 , and the solid cone itself is homeomorphic to the
closed 3-disk D3 .
7. Prove that the ascending union R∞ of Euclidean spaces Rn , n ≥ 1 with
weak topology is not homeomorphic to the infinite product Rω of real
lines. (Hint: Look at the metrizability property)
1.5 Quotient Spaces
Quotient spaces are of fundamental importance in topology. The intuitive

idea of “identifying” or “pasting” is made rigorous by the concept of a quotient
space. To make the things clear, let X be a topological space and R be an
equivalence relation in X. It is well-known that R decomposes the set X
into mutually disjoint equivalence classes. For each x ∈ X, let Cx be the
equivalence class containing the element x, i.e., Cx = {y ∈ X | yRx}. Then for
any two x, y ∈ X, Cx = Cy or Cx ∩ Cy = ∅. There is a natural surjective map
ν : X → X/R, where X/R denotes the set of all distinct equivalence classes
of X, defined by ν(x) = Cx . The collection τ of all subsets V ⊂ X/R such
that ν −1 (V ) is an open set of the space X is easily seen to be a topology on
X/R. This topology on X/R evidently makes the quotient map ν : X → X/R
a continuous map. Indeed, τ is the largest topology on X/R which makes the
map ν continuous. The set X/R with the topology τ defined above is called a
quotient space of the space X. Changing equivalence relations on the set X
produces all quotient spaces of X.
Take the unit interval I = [0, 1] and consider the equivalence relation R
on I for which the equivalence classes are: C0 = {0, 1} and Cx = {x} for all
x, 0 < x < 1. Then the quotient space I/R can be easily seen to be homeo-
morphic to the circle S1 . In such a case, we normally say that S1 is obtained
from the unit interval by identifying the two end points 0 and 1 of the interval I.
1.5. Quotient Spaces 11
0 1 S
1
Fig. 1.6: Circle as the quotient space of unit interval
Interesting quotient spaces can be obtained by above kind of identifica-

tions. As another example, consider the unit square I ×I and identify the points
(0, t) with (1, t) for all t ∈ I. Then the resulting quotient space is homeomor-
phic to an Euclidean cylinder whose boundary consists of two disjoint circles.
I x I
Fig. 1.7: Cylinder as the quotient space of the unit square
In the same square I×I, let us identify the points (0, t) with (1, t) for all t ∈
I and also identify (t, 0) with (t, 1) for every t ∈ I. Then the resulting quotient
space, this time, is homeomorphic to the torus embedded in the Euclidean
space R3 (Fig.1.8).
T
I X I
Fig. 1.8: Torus as the quotient space of unit square
As yet another example, let us again start with the square I × I and
identify the point (0, t) with (1, 1 − t) for each t ∈ I. Then the resulting quo-
tient space is homeomorphic to the Möbius band M (or strip) contained in R3
(Fig.1.9).
Fig. 1.9: Möbius band as the quotient space of square

The following procedure of the same kind yields a totally different topo-
logical space: Start with I × I and for each t ∈ I, identify (t, 0) with (t, 1)
and (0, t) with (1, 1 − t). Then the resulting quotient space, being continuous
image of I × I, is a compact, connected space. It is called the Klein bottle.
Unfortunately, this space is not homeomorphic to any subspace of R3 , i.e., the
Klein bottle K cannot be embedded in R3 (Why?). However, it is an interesting
topological space which we will consider again later (Fig. 1.9a).
A
C
Fig. 1.9a: The Klein bottle as the quotient space of a square
As the final example, we have the projective plane as a quotient

space of the 2-sphere S2 . Here, we identify a point (x, y, z) of S2 with
its antipodal point, viz., (−x, −y, −z). This quotient space too, denoted by
RP2 , cannot be embedded in R3 (Why?), and hence cannot be easily visualized.
The following result is quite useful in showing that the maps from the
quotient spaces are continuous.
Proposition 1.5.1. Suppose Y is a quotient space of X with quotient map

ν : X → Y and g : Y → Z is a map to some space Z. Then g is continuous if
and only if g ◦ ν : X → Z is continuous.
Let us have some more examples of quotient spaces generalizing the idea
of projective plane RP2 , which are very important in topology as well as in
geometry.
Example 1.5.2. (Real Projective Space) Let us consider the standard n-sphere
Sn , n ≥ 1. For any x ∈ Sn , let us identify x with its antipodal point −x ∈ Sn .
Then the resulting quotient space Sn / ∼, denoted by RPn , is called the n-
dimensional real projective space. It’s easily seen that the real projective space
is a compact, connected, Hausdorff space. To see the Hausdorff property, let
us remark that the quotient map Sn → RPn is closed. This is so because
if F is a closed subset of Sn and f : Sn → Sn is the antipodal map, then
f −1 f (F ) = F ∪ f (F ). Since f is a homeomorphism, this saturation of F is
closed in S n .
There is another way to define the same real projective space RPn , n ≥ 1.
If we consider the upper hemi-sphere Dn+ ⊂ Sn , and define an equivalence re-
lation ∼ in the hemi-sphere by identifying the antipodal points lying on the
boundary of the hemi-sphere. Then there is a continuous canonical bijection
from the resulting quotient space Dn+ / ∼ and the preceding quotient space
Sn / ∼ . Since the first space is compact and the latter is Hausdorff, the canon-
ical map is a homeomorphism.
There is yet another way of defining the real projective space in practice.
Consider the set Rn − {0}, identify all points on a line passing through the
origin and let X be the quotient space. Note that every such line intersects
Sn ⊂ Rn+1 − {0} in exactly two points, which are antipodal. Now it is obvious
that there is again a canonical bijection from the space X onto Sn / ∼, which
is bicontinuous and hence a homeomorphism.
Example 1.5.3. (Complex Projective Space) Consider the n-dimensional com-

plex sphere, viz. {(z0 , . . . , zn ) ∈ Cn |zi |2 = 1}. It is evident that the above
set is, in fact the real (2n + 1)-dimensional sphere S2n+1 . Identify two points
(z0 , . . . , zn ), (w0 , . . . , wn ) if there exists a λ ∈ S1 (
= 0) such that
(w0 , . . . , wn ) = λ(z0 , . . . , zn ).
The resulting quotient space S2n+1 / ∼, denoted by CPn , is called n-dimensional
complex projective space. This space is again a compact, connected, Hausdorff
space.
To show that it is Hausdorff, it is enough to show that the quotient map
ν : S2n+1 → CPn is a closed map. For this observe that if F is a closed set
of S2n+1 , then its saturation ν −1 (ν(F )) = S1 · F , where S1 is the unit circle.
Because S1 , F both are compact, the product S1 ·F under the map S1 ×S2n+1 →
S2n+1 is the image of S1 × F , which is compact, and so is closed.
Analogous to the real projective space there is another way to define
CPn , n ≥ 1. Let us consider the set Cn \ {0} and identify all complex lines
passing through origin. Then it is easily seen that the resulting quotient space
is homeomorphic to CPn .
The above projective spaces are not only compact, connected, Hausdorff
spaces, but they have additional properties viz. these are n-dimensional mani-
folds having very nice cellular structures, and it is essentially these properties
which make projective spaces extremely useful in algebraic topology.
Note that a quotient map ν : X → X/R, which is continuous and surjec-

tive, need not be an open map or a closed map. However, if f : X → Y is a
closed (respectively, open) surjective continuous map, then the topology of Y
is the quotient topology of X. The desired equivalence relation is obtained by
identifying each point of f −1 (y) with the single point y of Y . Also, some caution
is required when we consider the subspaces of quotient spaces or the quotient
spaces of subspaces : Let Y be a subspace of X. Then an equivalence relation R
in X restricts to an equivalence relation S in Y . Then the quotient space Y /S
has two topologies on it, one as the “quotient topology“ coming from Y , and
the other as the “subspace topology“ coming from X/R. In general, the two
topologies may be different from each other. To see an example of this situation,
consider the square X = I × I ⊂ R2 . Take the subspace Y = {(t, 0) : 0 ≤ t < 1}
of X and let R be the equivalence relation in X which produces the cylinder,
i.e., identify each (0, t) with the point (1, t), t ∈ I. Note that Y , as a subspace of
X, is homeomorphic to the half-open interval [0, 1) and the induced equivalence
relation S on Y is the identity relation. This means the quotient space Y /S is
again homeomorphic to [0, 1). However, if we consider the image of Y under
the quotient map X → X/R and take the subspace topology on Y /S (bottom
circle) induced from the cylinder X/R, then the induced subspace topology on
Y /S is homeomorphic to the Euclidean circle S1 . The assertion made earlier
now follows because the two spaces [0, 1) and S1 are clearly not homeomorphic.
It may be mentioned that, in general, the quotient topology on Y /S is stronger
than the subspace topology induced from X/R because the following square is
commutative:
Y
j -X
q q
? ?
Y /S - X/R
i
Since q j = iq is continuous and Y /S has the quotient topology, the map

i is also continuous. This at once implies that the quotient topology on Y /S
is stronger than the subspace topology on Y /S. The following statement can
be easily proved: Let Y be a closed (respectively, open) subspace of X and
let X → X/R be a closed map. Then the subspace topology on Y /S coincides
with the quotient topology induced from the subspace Y of X. Similarly, if R
and S are equivalence relations in spaces X and Y, respectively, then R × S
is an equivalence relation in X × Y . There are examples to show that the
quotient space (X/R) × (Y /S) need not be homeomorphic to the quotient
space (X × Y )/(R × S) (see Dugundji [5], pp 130-131).
A quotient space of a Hausdorff space need not be Hausdorff. To see

this, take the unit interval [0, 1]. Identify all the points of the set A = [0, 1)
to one equivalence class C0 . Then this quotient space has two points C0 , C1
and the space {C0 , C1 } is not even a T1 -space, since the point {C0 } is not closed.
Recall that an equivalence relation R is just a subset of X × X. The

relation R is said to be closed if it is a closed subset of X × X. In this termi-
nology, note that X/R is Hausdorff implies that R is closed. In the case of a
compact space X, one can prove that X/R is Hausdorff if and only if R is closed.
For any space X, if we identify the top copy X × {1} of X in X × I

to a point, then the quotient space X × I/R is called cone over X. It is
denoted by C(X). In the product space X × [−1, 1], if we identify top copy
XXI S(X)
Fig. 1.10: Suspension of a space
X × {1} and bottom copy X × {−1} to two different points, then the quotient
space, denoted by S(X), is called suspension of X. It must be noticed that
constructions of cone and suspension are natural in X, i.e., any continuous
map f : X → Y induces continuous maps from C(X) to C(Y ) as well as from
S(X) to S(Y ).
The concept of a mapping cylinder is also useful. Let f : X → Y be any

continuous map. Consider the product X × I with unit interval I = [0, 1] and,
in the disjoint union (X × I) ∪ Y , identify for each x ∈ X, the points (x, 0) with
f (x) ∈ Y . Then the quotient space Mf = (X × I) ∪ Y / ∼ obtained in this way
is called the mapping cylinder of f . The space Y is embedded in Mf under the
map y → [y]. Note that whenever y = f (x) for some x ∈ X, then [y] = [f (x)].
The space X is also embedded in Mf in several ways but we normally identify
X with X ×{1} ⊂ Mf . The Figure 1.11 depicts an idea of the mapping cylinder
Mf .
X x {1}
X x I
Mf
Y
Y
Fig. 1.11: Mapping cylinder of a map
Exercises
1. If X is a regular space and A is a closed subset of X, then show that the
quotient space X/A is Hausdorff.
2. In the real line R, let us identify all rational numbers to a single point.
Then show that the quotient space is an indiscrete space.
3. Let X = [−1, 1] × {0} ∪ {(0, 1)} be the subset of the plane R2 . Define
neighbourhoods of points (x, 0), −1 ≤ x ≤ 1 to be those induced from R2 ,
but the neighbourhoods of the point (0, 1) are sets of the type [(−, ) ×
{0} − (0, 0)] ∪ (0, 1). Show hat X is locally homeomorphic to real line,
but is not Hausdorff (this is an example of a 1-manifold, which is not
Hausdorff!)
4. Let R be an equivalnce relation in a compact Hausdorff space X. Show

that the quotient space X/R is Hausdorff if and only if R is closed in
X × X.
5. Let Sn be the standard n-sphere, n ≥ 0. Prove that the suspension S(Sn )
of Sn is homeomorphic to the (n + 1)-sphere Sn+1 .
6. Let D2 be the closed 2-disk in the plane R2 . Prove that the quotient space
R2 /D2 , obtained by collapsing the disk D2 to a point, is again homeomor-
phic to the plane R2 .
7. Prove that when two copies of the disk Dn are identified along their bound-
aries, the resulting quotient space is homeomorphic to the n-sphere Sn .
8. The Möbius band is obtained by identifying an edge of a square with its
opposite edge after rotating the first edge by π radians. What will be the
quotient space when the edge is rotated by 2π instead of π ?
Separation Axioms
Recall that a space X is said to be Hausdorff or a T2 -space if given any
two distinct points x1 , x2 ∈ X, we can find two disjoint open sets U1 and U2
of X such that x1 ∈ U1 and x2 ∈ U2 . This is usually expressed by saying
that x1 , x2 can be separated in X by disjoint open sets. A condition weaker
than T2 -axiom is T1 -axiom, viz., a space X is said to be a T1 -space if given
any two distinct points x1 , x2 of X, we can find two open sets U1 and U2
(not necessarily disjoint) such that one open set contains one point but not
the other. It is easy to see that a space X is a T1 -space if and only if every
point of X is closed in X. A T1 -space X is said to be T3 -space (or regular) if
a point x ∈ X and a closed set F of X, not containing x, can be separated
by disjoint open sets. A stronger property than T3 -property is T3 12 -property
(or completely regular) which says that a point x and a closed set F , not
containing x, can be separated by real-valued continuous functions, i.e., there
exists a continuous function f : X → [0, 1] such that f (x) = 0 and f (F ) = 1.
Finally, a T1 -space X is said to be a T4 -space (or normal) if any two disjoint
closed sets of X can be separated by disjoint open sets, i.e., if F1 , F2 are two
disjoint closed subsets of X, then there exist disjoint open sets U1 , U2 of X
such that F1 ⊆ U1 and F2 ⊆ U2 . It is well known that for each i = 2, 3, 4, a
Ti -space is a Ti−1 -space, but there are examples to show that the converse of
none of these is true. There is an example (this one is not easy) of a T3 -space
which is not T3 12 . One of the most important results regarding these separation
axioms is the famous Urysohn Lemma (see Dugundji [5], for all the results on
separation axioms), which asserts that in a normal space X, any pair of disjoint
closed sets can be separated by real-valued continuous functions. This lemma,
in particular, implies that a normal space is completely regular, i.e., a T4 -space
is also a T3 12 -space whereas a T3 12 -space is easily seen to be a T3 -space. It may
also be mentioned here that on a T3 12 -space X, there exists a large family of
1.6. Connected and Path-Connected Spaces 17
real-valued continuous functions; and the study of algebraic properties of the

family of continuous functions is a beautiful topic of topology.
Weak Topology
Let X be a set and {Aα | α ∈ I} be a family of subsets of X such

that X = ∪α∈I Aα . Suppose each Aα is a topological space in its own right.
Can we define a topology on X so that, as a subspace of X, Aα gets its own
topology? In order that the answer is in affirmative, it is clearly necessary
that whenever Aα , Aβ intersect, the set Aα ∩ Aβ must get the same topology
from each of the spaces Aα and Aβ . Suppose this is the case and also assume
that for each α, Aα ∩ Aβ is closed in Aβ for all β. Now, we define a subset
U of X to be open if U ∩ Aα is closed in Aα for each α. Then the collection
of all such subsets of X forms a topology on X and this induces the same
topology on Aα as its original topology (Exercise). This topology on X is
called the weak topology or the coherent topology on X induced by the given
topologies of Aα for α ∈ I. Note that with this topology on X, the inclusion
maps iα : Aα → X are continuous. In fact, the weak topology on X is the
largest topology on X which makes all inclusion maps iα : Aα → X continuous.
One interesting aspect of the weak topology on X defined by the family

{Aα } is the following: Let f : X → Y be a map from the space X to any space
Y and suppose the restriction f |Aα : Aα → Y is continuous for each α. Then
the map f itself is continuous. Important objects of algebraic topology require
spaces like simplicial complexes and CW -complexes. The topology of these
spaces is the weak topology defined by the Euclidean topology of simplexes
and cells, respectively.
1.6 Connected and Path-Connected Spaces

A topological space X is said to be disconnected if it can be expressed
as a union of two nonempty disjoint open sets. This is equivalent to saying
that X can be expressed as a union of two nonempty disjoint closed sets.
Another way of saying that X is disconnected is to say that there exists a
subset U (
= ∅, X) of X which is both open and closed (clopen). A space X is
said to be connected if it is not disconnected.
It follows that a discrete space having more than one point is disconnected.
The space Q of rationals is disconnected; the real line R (with Euclidean topol-
ogy) is connected. In fact, the following result identifies all connected subsets
of the real line:
Example 1.6.1. A subset A of R is connected if and only if A is an interval (it

may be finite, infinite, open, closed, or half open). To see this, note that if A
is not an interval, then there is point p ∈ R − A such that A has points of R
on the left as well as right of p. This means (−∞, p) ∩ A, (p, ∞) ∩ A are two
nonempty disjoint open sets of A whose union is A and so A is disconnected.
Conversely, suppose A = [a, b], a < b is an interval. If A is not connected, then

U
= A). Let x0 ∈ A−U . Then there are points
there is a clopen set U of A (∅ =
of U which are either on the left of x0 or on the right of x0 . Suppose there are
points of U on the left of x0 and consider the set L = {x ∈ U | x < x0 }. Let
p = sup L. Then p ∈ A, since L̄ ⊂ Ū , and since U is closed, p ∈ U . But U is
also open means there is an > 0 such that [p, p + ) ⊂ U . This contradicts
the supremum property of p and so A is connected. We can similarly verify the
result for other intervals.
We have mentioned the above result just because it answers the question
“what subsets of real line are connected?”, in a most simple way. Any result
which classifies a class of mathematical objects in easily detectable terms is
considered a beautiful theorem and there are very few results of this kind.
Heine-Borel Theorem is a result of this type because it identifies all compact
subsets of the Euclidean space Rn simply as “closed and bounded” subsets
of Rn . There is yet another result of this kind which has been termed as
a “theorem par excellence”. That theorem classifies all compact, connected
2-manifolds (viz., surfaces) in terms of a 2-sphere, a connected sum of tori or
a connected sum of projective planes (see W.Massey [13] p.10 and p.33 for
details).
The following results are quite well-known:
Proposition 1.6.2. A continuous image of a connected space is connected.
Theorem 1.6.3. The product of an arbitrary family of connected spaces is con-

nected.
It follows from the latter result that the infinite cubeI ∞ , where I is the
∞
unit interval, is connected and so is the product Rω = 1 R of real lines.
Evidently, a subset of a connected space is hardly connected. But the following
is a useful result:
Theorem 1.6.4. Let A be a connected subset of a space X. Then Ā is also

connected. More generally, if B is any subset of X such that A ⊆ B ⊆ Ā, then
B is connected.
Proof. Since the closure of A in the subspace B is obviously B, the general
result follows from the special one when we take B as X. To prove the first
result, it suffices to show that any continuous map f : Ā → {0, 1}, where {0, 1}
has the discrete topology, is a constant map. For such a map, the restriction
f |A of f to A will be continuous and, since A is connected, f |A is constant.
But then, since A is dense in Ā, f itself must be constant.
Example 1.6.5 (Topologist’s Sine Curve). This example shows that certain sub-
sets of plane which visually do not appear to be connected, may indeed be
connected (see Fig.1.12).
Fig. 1.12: Topologist’s Sine Curve
Consider the subspace G = {(x, y) ∈ R2 | y = sin x1 , 0 < x ≤ π1 }. Since

the interval (0, π] is connected and G is its image under the continuous map
x → (x, sin x1 ), G is a connected set. Hence, by earlier result, its closure T = Ḡ,
i.e.,
T = G ∪ {(0, y) | |y| ≤ 1}
must be connected. This space T is popularly known as Topologist’s Sine Curve
and it has the property that any subset B satisfying G ⊆ B ⊆ T is connected.
In particular, the subset B obtained by adjoining an additional point, say,
(0, 1/2) or even a part of y-axis lying between the points (0, −1) and (0, 1), to
G is connected. Note that the set B does not appear visually to be connected;
it appears to have two pieces.
Next, we consider conditions when the union of connected subsets of a

space X is connected. We have
Proposition 1.6.6.
Let {Aα | α ∈ I} be afamily of connected subsets of a space
X such that Aα
= ∅. Then the union α∈I Aα is connected.
The above result shows, in particular, that the plane R2 (or, more gen-
erally, even Rn , n ≥ 2) is connected. In fact, the plane R2 is the union of all
straight lines in R2 passing through origin and every line is evidently connected.
Indeed, the union of any number (finite or infinite) of lines passing through a
fixed point of the plane is connected; in particular, a star-shaped subset of the
Euclidean space Rn (n ≥ 2) is connected. It seems appropriate to ask: what are
the connected subsets of the plane R2 ? Unfortunately, there is no good answer
to this question.
Example 1.6.7. Deleting a single point p from the real line R makes the re-
maining space R − {p} disconnected. However, if we remove one point from
the plane R2 or even higher dimensional space Rn , the remaining space is
still connected. In fact, we can delete even a countable number of points
A = {p1 , p2 , . . . , pn , . . .} and yet the remaining set Rn − A is connected. To
see this, let a ∈ Rn − A. It suffices to show that any other point z ∈ Rn − A

and the point a lie in some connected subset of Rn − A. For this, consider any
bisector line L of the line segment az and for each y ∈ L, let ly = ay ∪ yz.
Then, for each y ∈ L, ly is a connected set and any two of them meet only
in a and z. We claim that there exists at least one y on L such that the set
ly lies completely in the set Rn − A. If that is not the case then each ly will
contain at least one point of A, which means the set A will have at least as
many points as the line L. Since L has uncountable number of points, this is a
contradiction. Therefore, Rn − A, being union of connected sets, all having the
point a common, must be connected.
z
a
L
Fig. 1.13
Non-homeomorphic Spaces
Example 1.6.8. We have already seen that any two open (or closed) intervals are
homeomorphic. However, an open interval (a, b), a < b is not homeomorphic to
a closed interval [c, d], c < d. The question is: How to prove this? The answer is
very easy: The closed interval is compact, but the open interval is not compact.
Hence, they cannot be homeomorphic because compactness is a topological
property and so if they were homeomorphic, then one being compact will imply
that the other must be compact, which is not the case. Now, how to prove
that the open interval (a, b) is not homeomorphic to the half-open interval
(c,d]? As both are noncompact, the argument involving compactness will not
work. Here, however, we can use connectedness as follows: Let f : X → Y
be a homeomorphism. If a subset A of X is removed from X and its image
f (A) is removed from Y , then the remaining spaces X − A and Y − f (A)
must be homeomorphic (Just consider restrictions of f and f −1 to X − A
and Y − f (A), respectively). Now, it is straight to see that (c, d] cannot be
homeomorphic to (a, b) simply because removing the single point d from (c, d]
leaves it connected, whereas removing any one point from the open interval
(a, b) leaves the remaining space disconnected. The connectedness argument
can also be applied to show that (a, b) is not homeomorphic to [c, d]: removing
the two end points c and d from [c, d] leaves it connected whereas removing any
two points from (a, b) renders it disconnected.
The above argument can be applied to prove that the real line R is not
homeomorphic to the plane R2 : removal of one point from R makes it discon-
nected whereas R2 remains connected even after removing hundreds of points
(Recall the earlier example). Now, here is an interesting problem: Can we apply
connectedness argument or some other argument involving a topological prop-
erty to prove that the Euclidean spaces Rm , Rn (m
= n) are not homeomorphic?
An answer (affirmative or negative) will provide an answer to the same problem
for spheres Sm and Sn (m
= n) since these are one-point compactifications of
Rm and Rn , respectively. We can ask other questions in the same spirit: With
any of these methods, can we decide whether or not (i) The projective plane P2
is homeomorphic to the torus T 2 = S1 × S1 ?, (ii) Klein bottle is homeomorphic
to the torus? Results of algebraic topology, as we will see in due course of time,
can quickly answer these and numerous other questions of this type. These and
a few others were historically the first applications of algebraic topology.
Components
If X is a disconnected space, then the maximal connected subsets of X,
called components, are occasionally quite useful. For any point x of such a space
X, consider the union Cx of all connected subsets of X which contain the point
x. Then, since union of any family of connected subsets having a nonempty
intersection is connected, Cx is obviously connected and is a maximal connected
subset of X containing the point x. The set of all components of X forms a
partition of the set X. Since closure of a connected set is connected, every
component Cx must be closed in X. The components of a discrete space are
singleton subsets. A space X is said to be totally disconnected if the components
of X are singleton sets. In the space Q of rationals also, the components are
singleton sets and so Q is totally disconnected but, as we know, Q is not discrete.
It may be mentioned that such spaces are not of much use in algebraic topology
except for constructing pathological examples.
Paths
The concept of a path is of basic importance in algebraic topology. By a
path ω in a space X, we mean a continuous map ω : I → X, where I = [0, 1] is
the unit interval. The point ω(0) is called the initial point and ω(1) is called
the terminal point of the path ω. Sometimes, we identify the path ω with its
image set ω(I), though the two things are quite different. Since I is compact
as well as connected, the image set ω(I) is a compact and connected subset of
the space X. It must be pointed out that strange things can happen with the
image of a general path – even if ω is a 1-1 map, the image set ω(I) need not
be homeomorphic to I as, for example, when the space X is indiscrete. If X is
the Euclidean space Rn (or, more generally, any Hausdorff space), then a 1-1
path ω is indeed a homeomorphism from I to ω(I) (why?). It all depends on
the space X where the path is defined. Peano gave an example of a path ω in
R2 where the image ω(I) filled the whole square I × I (the space-filling curve)
– of course, this path cannot be 1-1.
A topological space X is said to be path-connected if any two points of

X can be joined by a path. Note that if ω1 and ω2 are two paths in a space
X such that ω1 (1) = ω2 (0), then we can define a new path ω1 ∗ ω2 , called the
product of ω1 and ω2 , in X as follows:

ω1 (2t) 0 ≤ t ≤ 1/2
(ω1 ∗ ω2 )(t) =
ω2 (2t − 1) 1/2 ≤ t ≤ 1
The continuity of the path ω1 ∗ ω2 follows from the Pasting Lemma. Also,
the initial point of ω1 ∗ ω2 is the initial point of ω1 and the terminal point of
the new path is the terminal point of ω2 . Note that the image set of the path
ω1 ∗ ω2 is just the union of the image sets of ω1 and ω2 . The inverse of a path
ω is the path ω −1 defined by ω −1 (t) = ω(1 − t). Observe that the inverse path
ω −1 travels in the direction opposite to that of ω. Using these ideas regarding
paths, one can easily prove that a space X is path-connected if and only if
there is a point x0 of X which can be joined by a path to any other point x ∈ X.
A path connected space X is connected. To see this, let x0 ∈ X. For

any other point x ∈ X, let ωx : I → X be the path joining x0 to x. Then,
∪x∈X ωx (I) = X. Since ωx (I) is connected for all x and since ∩x∈X ωx = {x0 },
it follows that X is connected. We can define path-components of a space X
analogous to components as the maximal path connected subsets of X. Though
the components are closed subsets of X, path-components need not be closed.
The following example shows this and also the fact that a connected space need
not be path-connected.
Example 1.6.9. Consider the Topologist’s Sine Curve T = S ∪ G where S =

{(0, y) | |y| ≤ 1} and G = {(x, y) ∈ R2 | y = sin x1 , 0 < x ≤ π1 }. We have
already seen (Example 1.6.5) that T is connected. We show now that T is not
path connected. In fact, we claim that the point (0, 0) ∈ T cannot be joined
to ( π1 , 0) ∈ T by a path in T . If possible, let ω : I → T ⊂ R2 be a path
in T so that ω(0) = (0, 0) and ω(1) = ( π1 , 0). Since S is a closed subset of
T, ω −1 (S) is a closed subset of I such that 0 ∈ ω −1 (S) but 1 ∈ / ω −1 (S). This
−1
means ω (S) has a maximal element, say a, 0 < a < 1. Now the restriction
ω|[a,1] is a path joining a point of S and the point ( π1 , 0) having the property
that ω(t) ∈ G, ∀ t > a, For convenience, let us assume a = 0 and write
ω(t) = (x(t), y(t)). Observe that ∀ t > 0, x(t) runs over the interval (0, π1 )
and y(t) runs along the sine curve y(t) = sin 1t . Now we show that there is a
sequence {tn } of points in I such that tn → 0, and y(tn ) = (−1)n , which will
give a contradiction to the continuity of ω. For each given n choose u such that
0 < u < x( n1 ) and sin u1 = (−1)n . Then apply the Intermediate value theorem
to find an element tn , 0 < tn < n1 such that x(tn ) = u.
A space X is said to be locally connected (respectively, locally path-

connected) if every neighbourhood N of x ∈ X contains a connected (respec-
tively, path-connected) neighbourhood. It can be easily seen that in a locally
1.7. Compact Spaces and Locally Compact Spaces 23
connected space, the components are not only closed but are open also. The
following is an interesting result.
Proposition 1.6.10. A connected, locally path-connected space is path-connected.

Proof. Let X be a connected space which is also locally path connected and
x0 ∈ X. Then we claim that the subset U = {x ∈ X | x, x0 can be joined by a
path} of X must be an open set. This follows at once because y ∈ U implies
that there is a neighbourhood N (y) which is path connected. But that means
each point of N (y) can be joined to x0 by a path and hence, N ⊆ U . Thus, U is
open. On the other hand, if z ∈/ U , then z has a path connected neighbourhood
N (z) so that no point of N (z) can be in U . This means X − U is an open set.
Since X is connected and U
= ∅, U = X, i.e., X is path-connected.
It can be easily seen that a continuous image of a path-connected space

is path-connected; a product of any family of path-connected spaces is path-
connected.
Remark 1.6.11. It must be pointed out that path-components of a space X, as

we will see later on in Chapter 6, are closely related to the zero-dimensional
singular homology groups of X. On the other hand, the components of a space
are tied up with the zero-dimensional Čech cohomology groups of a space X.
In fact, these homology and cohomology groups are the basic objects of study
in algebraic topology.
1.7 Compact Spaces and Locally Compact Spaces
We recall that a topological space X is said to be compact if every open

cover of X has a finite subcover. What this says is that given any open cover
U = {Uα | α ∈ J} of X, there are finite number of members, say, Uα1 , . . . , Uαn
of U which cover the space X. Thus, compactness is a kind of finiteness
condition on the topology of X.
Note that the unit interval I = [0, 1] is compact whereas the real line R
itself is not compact. Any n-sphere Sn , n ≥ 0, is a compact subset of Rn+1 .
The famous Heine-Borel Theorem asserts that a subset A of the Euclidean
space Rn is compact if and only if it is closed and bounded – this yields a
very easy way of verifying whether or not a subset of Rn is compact. Thus,
for instance, the ellipsoid, viz., {(x, y, z) ∈ R3 | x2 /a2 + y 2 /b2 + z 2 /c2 = 1},
where a, b, c are any three given positive constants, being closed as well
as bounded subset of R3 , is compact. On the other hand, the hyperboloid
{(x, y, z) ∈ R3 | x2 /a2 + y 2 /b2 − z 2 /c2 = 1}, being unbounded, is not a
compact space. The graph G of the sine curve y = sin x1 , 0 < x ≤ π, in the
plane R2 is not compact since it is not closed – the point (0, a) ∈ R2 , |a| ≤ 1
are limit points of G, but do not belong to the graph. But the topologist’s sine
curve T = G ∪ {(0, a) | |a| ≤ 1} is a compact space. The torus S1 × S1 ⊆ R3 is

compact but a punctured torus is not compact.
It can be directly proved from the definition that a continuous image of

a compact space is compact; a closed subspace of a compact space is compact.
Compactness, when combined with Hausdorff separation axiom, yields some
interesting results : (i) a compact subset of a Hausdorff space X is closed in
X, (ii) a compact Hausdorff space is normal. The first part at once implies
that a continuous map f : X → Y , where X is compact and Y is Hausdorff,
is always closed. For, let F be a closed subspace of X. Then, since X is
compact, F must be compact and since a continuous image of a compact space
is compact, f (F ) is compact. Finally, since a compact subset of a Hausdorff
space is closed, f (F ) must be closed in Y . With this last result, the following
fact is immediate: Any continuous bijection f : X → Y from a compact space
X to a Hausdorff space Y is a homeomorphism. This result is of common use
in identifying quotient spaces with more familiar spaces. To see an example,
take a disk D2 and identify all the boundary points to a single point. Then it
is asserted that the result is homeomorphic to S2 . How? Note that D2 − S1 is
homeomorphic to S2 − {N }, where N is the north pole of the sphere. Now, if
the homeomorphism is extended to a map f : D2 → S2 by mapping whole of S1
to the point N , then f is continuous and onto. Hence, f induces a continuous
map h : D2 /S1 → S2 , which is evidently a continuous bijection. As D2 /S1 is
compact and S2 is Hausdorff, the above result says that h is a homeomorphism.
One of the most important theorems of set topology, which has beautiful
consequences in other branches of mathematics also, is the following:
Theorem 1.7.1. (Tychonoff ). Let {Xα | α ∈

I} be any (finite or infinite) family
of compact spaces. Then the product space α∈I Xα is compact.
While dealing with covering projections, we will frequently require the

following result called Lebesgue’s Covering Lemma:
Lemma 1.7.2. (Lebesgue). Let (X, d) be a compact metric space and

U = {Uα } be an open covering of X. Then there exists a positive real num-
ber δ (called Lebesgue number for the given covering) such that any ball B(x, δ)
around any point x ∈ X is contained in some member of U.
Proof. For every x ∈ X, first let us choose a positive number r(x) such
that B(x, r(x)) is contained in some member of U. Then {B(x, r(x)/2) |
x ∈ X} is an open covering of X. Since X is compact, there exists a fi-
nite subcover, say, B(x1 , r(x1 )/2), . . . , B(xn , r(xn )/2) which covers X. Let
δ = min{r(x1 )/2, . . . , r(xn )/2}. Then, δ > 0 is the required Lebesgue num-
ber. To see this, let us consider a ball B(x, δ) for some x ∈ X. Then there is
an i = 1, . . . , n such that x ∈ B(xi , r(xi )/2). Now, if y ∈ B(x, δ), then
1.7. Compact Spaces and Locally Compact Spaces 25
d(y, xi ) ≤ d(y, x) + d(x, xi ) ≤ δ + r(xi )/2 ≤ r(xi ).

Hence, B(x, δ) ⊆ B(xi , r(xi )) ⊆ U for some U ∈ U .
Recall that a map f : (X, d) → (Y, d ) is said to be uniformly contin-

uous if given > 0, ∃ a δ > 0 (independent of the points x, x ∈ X) such
that d(x, x ) < δ implies d (f (x), f (x )) < . A continuous map need not be
uniformly continuous. However, for a compact metric space X, we have the
following useful result:
Proposition 1.7.3. Let X be a compact metric space and f : (X, d) → (Y, d )

be a continuous map. Then f is uniformly continuous, i.e., a continuous map
from a compact metric space to another metric space is uniformly continuous.
Proof. Consider the balls {B(y, /2) | y ∈ Y } and let δ > 0 be a Lebesgue
number for the open covering {f −1 (B(y, /2)) | y ∈ Y } of the compact space
X. Since each B(x, δ) lies in one of these sets, f (B(x, δ)) ⊆ B(y, /2) for some
y, and since f (x) ∈ B(y, /2), we find that for any z ∈ B(x, δ),
d(f (z), f (x)) ≤ d(f (z), y) + d(y, f (x)) < /2 + /2 = ,
which says that f (B(x, δ)) ⊆ B(f (x), ).
A generalization of the concept of compactness is local compactness. A

Hausdorff space X is said to be locally compact if every point x ∈ X has
a neighbourhood N (x) in X which is compact. It can be easily proved that
this, in fact, implies that any neighbourhood N of x will contain a compact
neighbourhood of x. The real line R is obviously locally compact whereas the
space Q of rational numbers is not locally compact. To see the latter, suppose
N (q) is a compact neighbourhood of q ∈ Q. Then N (q) must contain a closed
-neighbourhood, say, [q − , q + ] of q in Q whose end points are irrational
numbers and which contains all rationals lying between q − and q + . Being
the closed subset of a compact space N (q), the rational interval [q − , q + ] ∩ Q
must be compact. But this contradicts the Heine-Borel Theorem because it is
not closed in R. This example also shows that a subspace of a locally compact
space need not be locally compact. However, the following is an interesting
result.
Proposition 1.7.4. A subset Y of a locally compact space X, which is the inter-

section Y = F ∩ U of a closed subset F and an open subset U of X, is locally
compact. Conversely, if Y is locally compact, then there exists a closed set F
of X and an open set U of X such that Y = F ∩ U .
Discrete spaces are evidently locally compact. Since every topological

space X is a continuous image of a discrete space, it follows that a continuous
image of a locally compact space need not be locally compact. However, the
image under a continuous open map of a locally compact space is locally
compact. The infinite product Rω of real lines is not locally compact

∞ - any
neighbourhood of origin must contain a product neighbourhood 1 Ui where
all, except finitely many Ui , must
be real lines but that can never be compact.
In fact, an infinite product i∈I Xi of locally compact spaces is locally
compact if and only if all except finitely many are compact.
An important feature of a locally compact Hausdorff space X, which is

not already compact, is that it can always be compactified by just adding one
more point in the space X and by enlarging the topology of X. This idea is
due to P.S. Alexandroff and is known as one-point compactification of X. To
see this, let ∞ be a point which is not in the given locally compact space X.
Consider the set X̂ = X ∪ {∞}. We define a topology on X by defining the
neighbourhoods of every point. Declare neighbourhoods of all points of X to
be the same which were in the space X, but define neighbourhoods of ∞ to be
those subsets U of X̂ which contain the point ∞ and the complement of some
nonempty compact subset C of the space X, i.e., U = {∞} ∪ (X − C), where
C is a nonempty compact subset of X. Then one can easily verify that these
neighbourhoods define a topology on X̂. It can also be proved that the resulting
topology on X̂ is compact, Hausdorff and the original space X is an open dense
subset of X̂. The new space X̂ is called the one-point compactification of X.
Here, it is necessary to point out that the topology on X̂ defined above is, in
fact, unique, i.e., if X̂ is a compact, Hausdorff space such that X is an open
dense subset of X̂, then the topology of X̂ must be exactly the way it has been
defined above by Alexndroff. Using this uniqueness property, the following two
examples are obvious:
Example 1.7.5. The one-point compactification of the discrete space of natural

numbers N is homeomorphic to the subspace {0} ∪ { n1 | n ∈ N} of R.
Example 1.7.6. For each n ≥ 1, the one point compactification of Rn is home-

omorphic to the n-sphere Sn (see Example 1.4.5).
A locally compact space X is also completely regular. For, just take the
one-point compactification X̂ of X. Then, being compact Hausdorff, X̂ is nor-
mal, which means, X̂ is completely regular and so, X being a subspace of X̂,
must be completely regular. Indeed, given a compact set K and a closed subset
F of a locally compact space (F is disjoint from K), there exists a continuous
map f : X → R such that f (F ) = 0 and f (K) = 1. Again, this can be proved
by going to the one-point compactification of X.
Example 1.7.7 (Long Line). A locally compact space X need not be a metric
space. An interesting example of such a space is the long line L. The space L is
defined by considering the set Ω = {0, 1, 2, . . . , ω, ω + 1, . . .} of ordinals which
are less than the first uncountable ordinal ω1 . Between any two consecutive
ordinals α and α + 1, we insert a copy of the open interval (0, 1) so that α
1.8. Compact Surfaces 27
and α + 1 become the end points. The resulting set L+ is linearly ordered and,
therefore, we can give it the order topology. Take another copy L− of L+ by
giving it the opposite linear order. Then their union L = L+ ∪ L− , where the
ordinal 0 of the two spaces L+ and L− are identical, i.e., L+ ∩ L− = {0},
equipped with order topology, is called the long line. It is nonmetrizable and
locally compact. It is also countably compact. In fact, L is a nonmetrizable
one-dimensional manifold.
1.8 Compact Surfaces

The following concept is quite useful, especially in the study of surfaces.
Definition 1.8.1 (connected sum). Let us take two tori T1 , T2 (see Fig.1.15 )
and choose small closed disks D1 , D2 in T1 and T2 , respectively. Note that the
boundary of each of these disks is homeomorphic to the circle S1 . Let us remove
the interiors of the two disks from T1 and T2 and choose a homeomorphism h
from the boundary of the first disk to the boundary of the second. Let T1 and
T2 be the remaining portions of the two tori when the interiors of the disks
have been removed. Now, consider the disjoint union X = T1 ∪ T2 and identify
the boundary point x ∈ T1 with the boundary point h(x) ∈ T2 for all x in the
boundary of D1 . The identification space or quotient space X/R so obtained,
denoted by T1 #T2 , is called the connected sum of T1 and T2 .
Fig. 1.15: The connected sum of two tori
It can be shown that the connected sum T1 #T2 is independent (up to

homeomorphism) of the choice of the disks D1 , D2 and the homeomorphism
h. Note that the connected sum of two tori is a new surface. We can similarly
define the connected sum of two projective planes and, in fact, we can define
the connected sum of any two spaces X and Y each of which has two disks of
the same dimension embedded in it, e.g., we can form the connected sum of
any two n-dimensional manifolds, n ≥ 1.
Classification of Compact Surfaces

We have already mentioned a few compact surfaces like a 2-sphere, a torus,
a projective plane and a Klein bottle (we are leaving out Möbius band and cylin-
der, etc., because they are surfaces with boundary which we are not considering
at the moment). A Hausdorff topological space X, which has the property that
each point of X has an open neighbourhood homeomorphic to the Euclidean
space Rn or to an open ball Bn , is called a topological n-manifold. The 2-
dimensional manifolds are called surfaces. To avoid pathological situations, we
consider only those n-manifolds which are embedded in some Euclidean space
Rn . This is equivalent to saying that they have a countable basis. It is a deep
result of low-dimensional topology that every 2-manifold is triangulable. Here,
our objective is to know all possible compact surfaces. The ones like a 2-sphere,
etc., mentioned earlier, are all compact surfaces. The question is that besides
these, what else do we have? The answer is known and it is very interesting:
They are either a connected sum of a finite number of tori (Ti ) or they are the
connected sum of a finite number of projective planes (Pi )! This result, called
the Classification Theorem for compact surfaces, is one of the most exciting
theorems of the whole of topology. It says that any compact surface is homeo-
morphic to a 2-sphere or to a connected sum T1 #T2 · · · #Tn or to a connected
sum P1 #P2 · · · #Pn for some n ≥ 1. The theorem also asserts that each of the
above surfaces is distinct, i.e., no two of them are homeomorphic to each other.
The other fascinating result regarding these compact surfaces is that each one
of these surfaces can be obtained as a quotient of an ordinary polygon. To be
specific, let us recall how a torus was obtained from a polygon having four
sides by identifying the opposite sides of the square or how the Klein bottle
was obtained from the same 4-gon by identifying sides somewhat differently. In
the general case, we start with a n-gon and then identify various edges of this
polygon, without disturbing the interior of the polygon, in a suitable manner so
as to obtain any given compact surface (the detailed statements as well as the
proofs of these results with enough geometric insight can be found in W.Massey
[13]).
Paracompact Spaces
This concept is due to J. Dieudonné. Recall that a family

A = {Aα : α ∈ I} of subsets of a space X is said to be locally finite if
each point x ∈ X has a neighbourhood which meets only finitely many
members of A. A topological space X is said to be paracompact if every open
cover of X has a locally finite open refinement covering the space X. Every
compact space is trivially paracompact but a locally compact space need
not be paracompact, e.g., the long line mentioned earlier is locally compact,
but not paracompact. It can be proved that a paracomact, Hausdorff space
is normal and so a paracompact Hausdorff space satisfies all the separation
axioms. It is easy to see that a closed (not arbitrary) subset of a paracompact
space is paracompact, and (this one is not easy!) a closed continuous image of
a paracompact space is paracompact. One of the most important properties of
a paracompact spaces is the following: a paracompact space X has the partition
of unity subordinated to every open covering of X. What it means is that given
any open covering U of X, there exists a locally finite refinement V of U and a
1.8. Compact Surfaces 29
family of continuous functions fV : X → R, V ∈ V such that the closure of the

set {x ∈ X : fV (x)
= 0} ⊂ V for every V ∈ V and ΣV ∈V fV = 1.
Another deep result about paracompact spaces is the following theorem of

A.H.Stone: Every metric space is paracompact (see Dugundji [5] for all details).
The class of paracompact spaces has a very special status in algebraic topology
simply because every CW-complex is paracompact which implies, in particular,
that every simplicial complex (finite ones are already compact) is paracompact.
In fact, most of the important cohomology theories, e.g., the C̆ech cohomology,
singular cohomology, de Rham cohomology, etc., coincide on a paracompact
manifold; the general n-manifolds acquire interesting properties only when they
are paracompact, not otherwise. We have the following important result in this
direction:
Theorem 1.8.2. Let X be an n-manifold. Then the following are equivalent:

(1) X is paracompact (2) X is second countable (3) X is Lindeloff (4) X is
metrizable.
Exercises
1. Let X be a Hausdorff space. Show that any two disjoint compact subsets
of X can be separated by two disjoint open sets of X.
2. Let us consider the following subspace X of the Euclidean plane: X consists
of all closed line segments joining the origin and the points (1, 1/n), n ∈ N
and the line segment {(x, 0) : 0 ≤ x ≤ 1}. Prove that the weak topology
on X defined by these line segments is strictly finer than the topology of
X.
3. Let p : X × Y → X be the projection on the first factor and Y be a
compact space. Show that p is a closed map, i.e., the projection parallel
to a compact factor is a closed map.
4. Let A be any subset of X and Y be a compact space. Prove that any
neighbourhood U of the subset A × Y contains a tubular neighbourhood,
i.e., there exists a neighbourhood V of A in Y such that V × Y ⊂ U .
5. Show that an arbitrary product of path-connected spaces is again path-
connected.
6. (Hawaiian earring) For each n ≥ 1, let
Cn be the circle in plane of radius
1/n with center (1/n, 0). Let X = n≥1 Cn and consider the following
two topologies on X: one, the subspace topology from R2 , and the other,
the weak topology defined by the family of circles Cn . Show that the two
topologies on X are distinct. (The space X with subspace topology is
called the Hawaiian earring.)
7. Consider the ascending union R∞ = ∪n≥1 Rn . This set has two topologies
∞R , n ≥ 1, and the
n
on it: one, the weak topology defined by the spaces
other is the subspace topology coming from R = 1 R. Prove that the
ω
weak topology is strictly stronger than the subspace topology.

8. Consider the ascending union S∞ = n≥1 Sn of n-spheres and give it the
weak topology defined by the family of Euclidean spheres. Show that S∞
is not locally compact.
9. Suppose X is a locally compact space and Y is a locally compact subspace
of X. Show that there exists a closed subset F and an open subset U of
X such that Y = F ∩ U .
10. A subset Y of a space X is said to be locally closed if for each y ∈ Y there
exists a neighbourhood N (y) of y in X such that Y ∩ N (y) is closed in
N (y). Prove that the following conditions are equivalent: (i) Y is locally
closed (ii) Y is open in Ȳ , the closure of Y . (iii) Y is the intersection of
an open and a closed subset of X.
11. Suppose Y is a quotient space of X and Z is a locally compact space.
Show that Y × Z is a quotient space of X × Z. (This exercise indicates
that the product of two quotient spaces need not be a quotient space).
12. Prove that the connected sum R#R is homeomorphic to a hyperbola, and
S 1 #S 1 is homeomorphic to S 1 .
13. Let X, Y, Z be arbitrary surfaces (2-manifolds). Then prove that there are
the following homeomorphisms: (i) X#Y ∼ = Y #X. (ii) X#S2 ∼ = X. (iii)
(X#Y )#Z ∼ = X#(Y #Z).
1.9 What is Algebraic Topology?

There are many questions in point-set topology which are quite easy to
ask. But it is surprising that answering them is a lot difficult. To take a simple
example, let us ask whether or not the two Euclidean spaces R3 and R4 are
homeomorphic. On intuitive basis, one immediately tends to assert that these
are not homeomorphic. The focal point of the problem is: what is a proof
of this assertion? Let us observe that the same question for the Euclidean
spaces R and R4 is answered very easily. These are not homeomorphic because
if they were homeomorphic, then removing one point from both of these
spaces must leave homeomorphic spaces. However, removing one point from
R leaves the remainder space as disconnected whereas removing one point
from R4 leaves the remainder as a connected space and so they cannot be
homeomorphic. This is a proof of our assertion using a method of point-set
topology. As another example, it appears on the same intuitive grounds that
the two Euclidean spheres S3 and S4 are not homeomorphic; likewise, the
2-sphere S2 and the torus S1 × S1 are apparently not homeomorphic; the Klein
bottle and the torus also appear to be non-homeomorphic. However, using
1.9. What is Algebraic Topology? 31
only the methods of point-set topology, it is not easy to give a proof of these
facts except in the case of a 2-sphere and torus. In this case, one can argue
that there is a nontrivial circle on torus, which when removed, leaves the torus
connected, whereas removing any circle from S2 leaves it disconnected. The
general question whether or not two given spaces X and Y are homeomorphic
is known as the “Homeomorphism Problem”. Methods of algebraic topology
using the concepts of homology groups and homotopy groups of the spaces
X and Y, as we will see later, are frequently very powerful for resolving the
homeomorphism problems stated above.
The French mathematician C. Jordan (1858-1922) was the first to ask for
a proof of the following somewhat obvious statement:
Jordan Curve Theorem: A simple closed curve C in the Euclidean plane R2

separates the plane into two open connected sets with C as their common
boundary. Exactly one of these two open sets is unbounded.
In the first instance, the above statement seems obvious on intuitive

grounds, but Jordan asserted that intuition is not a proof. We just cannot
accept this as a theorem unless it has a rigorous mathematical proof. It was
Oswald Veblen (1880-1960) who used the method of algebraic topology to give
a correct mathematical proof of the above statement in 1905. This result is
now well-known as the Jordan Curve Theorem.
Another type of question which we come across in point-set topology is

the following Extension Problem. Let A ⊂ X be a subspace and f : A → Y
be a continuous map. Does there exist a continuous map F : X → Y such that
F |A = f ? In other words, can we extend f : A → Y to whole of X continuously?
f
- Y
A

*

? F
X
In the special case, when X is a normal space, A ⊂ X is a closed subspace

of X and Y = R is the real line, the Tietze’s Extension Theorem says that
any continuous map f : A → R can be extended continuously to whole of X.
However, we are interested in the general Extension Problem for arbitrary
spaces X, Y, A and an arbitrary continuous map f : A → Y. The methods of
general topology may sometimes be successful to answer an extension problem,
but only rarely. For instance, can the sine curve function f : (0, ∞) → [−1, 1]
defined by f (x) = sin 1/x be extended to (−∞, ∞)? The answer is “No”
because it is easy to see by continuity considerations that setting f (0) to be
any point in [−1, 1] can never make the extended function continuous. Now,
let us consider an interesting question of the same kind. The n-sphere Sn is
the boundary of the (n + 1) disk Dn+1 ⊂ Rn+1 . Let f : Sn → Sn be the identity

map. Can we extend this map continuously to Dn+1 ? Here, the answer could
be guessed on geometric grounds to be in negative but the proof is difficult. In
fact, once again, as we will see later, algebraic topology provides a quick proof
of this result for the case n > 0. The case n = 0 is clear from only continuity
considerations.
Yet another kind of question which we confront in point-set topology is

the important Lifting Problem. Suppose p : X̃ → X is an onto continuous map.
Let f : Y → X be a given continuous map. Does there exist a continuous map
g : Y → X̃ such that p ◦ g = f ?
g
* X̃
p
?
Y - X
f
Unlike the Tietze’s Extension Theorem, we have no result in point-set

topology which answers this Lifting Problem even in a significant special case.
However, as we will see later in this book, the methods of algebraic topology
provide a complete solution to this Lifting Problem in terms of a concept
called fundamental group of a topological space.
The concepts of fundamental group and homology group mentioned

above were the first basic notions of algebraic topology which were intro-
duced by the famous French mathematician H. Poincaré during 1895-1901.
It must be mentioned here that the subject of point-set topology did not
even exist when the concepts of algebraic topology mentioned above were
created and used in solving problems of other branches of mathematics. The
point-set topology grew out of the works done by M. Fréchet in 1906 on general
metric spaces and developed further in 1914 by the classic work of F. Hausdorff.
Given a topological space X, the basic aim of algebraic topology is to

associate some algebraic object, say, a group H(X) to the space X (it could
be an abelian group or, more generally, a module or even a sequence of mod-
ules, etc.). The algebraic object so associated depends on the topology of X
in such a manner that if f : X → Y is a continuous map between X and Y,
then it naturally induces a homomorphism H(f ) between the groups H(X)
and H(Y ). We emphasise that the ingenuity of creating the concept lies in cap-
turing the topology of the space X in defining the associated algebraic object.
Furthermore, the induced homomorphisms have the following two properties:
(i) If f : X → Y and g : Y → Z are continuous maps, then the induced
homomorphisms H(f ), H(g) and H(g ◦ f ) are related as follows:
H(g ◦ f ) = H(g) ◦ H(f ).

1.9. What is Algebraic Topology? 33
(ii) If IX : X → X is the identity map of the space X, then H(IX ) = IH(X)

is the identity map of the group H(X).
We note that H is a covariant functor from the category of topological

spaces and continuous maps to the category of groups and homomorphisms.
These two properties of the associated groups and induced homomorphisms
immediately yield the important fact that the group H(X) is a topological
invariant of the space X, i.e., if X and Y are homeomorphic spaces, then
the associated groups H(X) and H(Y ) must be isomorphic (prove it as an
easy exercise). Thus, suppose we have two spaces X and Y for which the
homeomorphism problem is to be solved. If we find that the associated groups
H(X) and H(Y ) are not isomorphic, then as a consequence of the above
topological invariance of associated groups, the spaces X and Y cannot be
homeomorphic. This is a general method of algebraic topology which solves
the Homeomorphism Problem to a large extent. We must point out that this
solution of the Homeomorphism Problem is only one way, i.e., it is successful
only when X and Y are not homeomorphic. The method does not assert that
X and Y are homeomorphic if the associated groups are isomorphic, except in
a few rare cases.
The strategy for solving the Extension Problem is now quite straightfor-
ward. Let f : A → Y be a continuous map and A ⊂ X be a subspace. If this f
can be extended to a continuous map F : X → Y, then the triangle
f
- Y
A

*
i
? F
X
must be commutative, where i : A → X is the inclusion map. Let us consider
the following diagram of associated groups and induced homomorphisms.
H(f )
H(A) H(Y )
H(i)
H(F )
H(X)
By the properties of associated groups and induced homomorphisms stated

earlier, this new diagram of groups and homomorphisms must also be commu-
tative, i.e., we must have
H(f ) = H(F ) ◦ H(i).
However, if for some reason we find that this last equation is false, then we
conclude that the map F cannot exist and so the map f cannot be extended.
This solves the Extension Problem. We remark that in case the above equation
is valid, we cannot assert that the map f can be extended. Thus, the solution
of the Extension Problem using the above method is also one way. A similar
strategy is also used to solve the Lifting Problem. This will be discussed in
detail in the chapter on covering spaces because, for Lifting Problem, the
technique of algebraic topology is surprisingly successful both ways rather
than only one way as was the case with the Homeomorphism Problem or the
Extension Problem. In fact, we obtain a necessary and sufficient condition in
terms of the associated fundamental groups and induced homomorphism for
the existence of a map g : Y → X̃ lifting the given map f : Y → X.
The subject, called Algebraic Topology, basically aims at creating and

defining various kinds of algebraic objects associated to a topological space and
homomorphisms induced by a continuous map. It is true that a particular kind
of algebraic object may be better suited for a particular problem in topology.
More often than not, computing these associated algebraic objects for various
spaces and identifying the induced homomorphisms between the associated
algebraic objects is an interesting question in its own right. The subject has
grown into a vast field as a result of developing new tools, proving theorems
for computational problems and applying them fruitfully elsewhere in various
branches of mathematics.
Chapter 2
The Fundamental Group
2.1 Introduction
Let X be a topological space. Often we associate with X an object that

depends on X as well as on a point x of X. The point x is called a base point
and the pair (X, x) is called a pointed space. If (X, x) and (Y, y) are two
pointed spaces, then a continuous map f : X → Y such that f (x) = y is called
a map between pointed spaces. Let f : X → Y be a homeomorphism and x be
a point of X. Then f is a homeomorphism between pointed spaces (X, x) and
(Y, f (x)). The composite of two maps between pointed spaces is again a map
between pointed spaces and the identity map I(X,x) : (X, x) → (X, x) is always
a homeomorphism of pointed spaces for each x ∈ X.
In this chapter, we show how to each pointed space (X, x), we can associate
a group π1 (X, x), called the fundamental group of the space X at x. Each map
f : (X, x) → (Y, y) between pointed spaces (X, x) and (Y, y) then induces a
homomorphism (denoted by f# also)
f∗ : π1 (X, x) → π1 (Y, y)
between groups π1 (X, x) and π1 (Y, y) such that the following two conditions
are satisfied:
(i) If f : (X, x) → (Y, y) and g : (Y, y) → (Z, z) are two maps of pointed
spaces, then
(g ◦ f )∗ = g∗ ◦ f∗ : π1 (X, x) → π1 (Z, z).
(ii) If I(X,x) : (X, x) → (X, x) is the identity map, then the induced group
homomorphism
I(X,x)∗ : π1 (X, x) → π1 (X, x)
is also the identity map.
36 Chapter 2. The Fundamental Group
We will also compute fundamental group of several interesting spaces and

exhibit some of their uses. The two properties of the induced homomorphism
stated above are known as functorial properties, which at once yield the follow-
ing important consequence : the fundamental group π1 (X, x) is a topological
invariant of the pointed space (X, x). The detailed meaning of this statement
is given below:
Proposition 2.1.1. If (X, x) and (Y, y) are two pointed spaces which are homeo-
morphic, then their fundamental groups π1 (X, x) and π1 (Y, y) are isomorphic.
Proof. Suppose f : (X, x) → (Y, y) is a homeomorphism of pointed spaces. Then
the inverse map f −1 : (Y, y) → (X, x) is evidently a map of pointed spaces and
has the property that
f −1 ◦ f = I(X,x) , f ◦ f −1 = I(Y,y) .
Let us consider the induced homomorphisms
f∗ : π1 (X, x) → π1 (Y, y), f∗−1 : π1 (Y, y) → π1 (X, x)
in the fundamental groups. By functorial property (i), we find that
(f −1 ◦ f )∗ = f∗−1 ◦ f∗ ,
and, by (ii), we see that I(X,x)∗ is the identity map on π1 (X, x). Thus, we
conclude that f∗−1 ◦ f∗ is the identity map on π1 (X, x). Similarly, we can
see that f∗ ◦ f∗−1 is the identity map on π1 (Y, y). Therefore, f∗ and f∗−1 are
inverses of each other, i.e., f∗ : π1 (X, x) → π1 (Y, y) is an isomorphism.
The fact that the fundamental group π1 (X, x) of a pointed space (X, x)
is a topological invariant is an interesting and a very useful result. It says
that if X and Y are two spaces such that for some x0 ∈ X, π1 (X, x0 )
is not isomorphic to any of π1 (Y, y), y ∈ Y , then the spaces X and Y
cannot be homeomorphic. For, suppose X and Y are homeomorphic and let
f : (X, x) → (Y, y) be a homeomorphism. Then f : (X, x0 ) → (Y, f (x0 ) is
a homeomorphism of pointed spaces and so, by the above proposition, the
induced group homomorphism f∗ : π1 (X, x0 ) → π1 (Y, f (x0 )) must be an
isomorphism, i.e., π1 (X, x0 ) is isomorphic to π1 (Y, f (x0 )), a contradiction.
One of the most important problems in topology, known as the classifi-

cation problem in a given class of topological spaces, is to decide whether or
not two given spaces of that class are homeomorphic. To prove that two spaces
X and Y are indeed homeomorphic, the problem is really to find out some
specific homeomorphism from X to Y , and invariably, the only method to do
this is the knowledge of point-set topology. However, to prove that X and Y
are not homeomorphic, one looks for some topological invariant possessed by
one space and not by the other. For example, when we have to show that R1
is not homeomorphic to R2 , we argue as follows : if we remove one point from
both, then the remaining spaces, first being disconnected and the second being
2.1. Introduction 37
connected, are not homeomorphic and so R1 cannot be homeomorphic to R2 .

Similarly, the circle S1 cannot be homeomorphic to the figure of eight (two
circles touching at a point) because if we remove the point of contact from the
figure of eight, then the remaining space is disconnected whereas if we remove
any point from S1 , the remaining space remains connected. To prove that a
closed interval [0, 1] is not homeomorphic to an open interval (0, 1), we say
that one is compact whereas the other is not compact and so they cannot be
homeomorphic. These methods are known as the methods of point set topology.
Now, let us ask whether or not the 2-sphere S2 is homeomorphic to the 2-torus
T = S1 × S1 (Fig. 2.1). This can also be resolved using point-set topology as
follows : Take a circle C in T as shown in Fig. 2.1. If we remove C from T, the
remaining space is clearly connected. However, if we remove any circle from
S2 , the remaining space is evidently disconnected. This says that T and S2
can not be homeomorphic. Now, let us ask whether or not the 3-sphere S3 is
homeomorphic to the 3-torus T = S1 × S1 × S1 .
S2 T
Fig. 2.1: A 2-sphere and a 2-torus
The reader is invited to discover some method of point-set topology to
show that they are not homeomorphic (they are really not homeomorphic!)
and see for himself that this can be extremely difficult. In such a case, however,
the methods of algebraic topology sometimes work very well. We will, later on,
prove that if X is a path-connected space then the fundamental group π1 (X, x)
is independent of the base point (up to isomorphism), and we denote it simply
by π1 (X). This fact, combined with the previous proposition, will say that if
X and Y are two path-connected spaces such that π1 (X) is not isomorphic
to π1 (Y ) then X and Y cannot be homeomorphic. By popular belief based
on experience, it is normally much easier to decide that two groups are not
isomorphic than to decide that two given spaces are not homeomorphic. Now,
granting that π1 (S3 ) = 0 and π1 (S1 × S1 × S1 ) ∼
= Z × Z × Z (we shall prove these
facts later), we immediately conclude that S3 cannot be homeomorphic to the
torus S1 × S1 × S1 as their fundamental groups are evidently not isomorphic.
This is just one of the several methods of algebraic topology in proving that
two spaces are not homeomorphic. The crucial point is the result that the
fundamental group π1 (X) is a topological invariant of pathr-connected spaces.
Several objects such as homology groups, Euler characteristics, etc., are other
important invariants of topological spaces. We shall come to them later.
We have only indicated that fundamental group π1 (X, x) of a pointed

space (X, x) is a topological invariant. This is a good result, but by no means
the best result. More general and interesting results, including the best ones
about fundamental groups, will be studied only after we have defined them. The
definition requires the important concept of homotopy and the generalization
of results will require the concept of homotopy type in the class of topological
spaces. We take up these notions in the next section.
2.2 Homotopy
We now proceed to define fundamental group π1 (X, x) of a given pointed

space (X, x). It will take some time to do so, but that is true of most of
the topological invariants in algebraic topology. The fundamental group is, of
course, the first such invariant we are going to deal with. There are several
important concepts which will be introduced on our way to the definition of
π1 (X, x). The first one is the concept of homotopy. We have
Definition 2.2.1. Let X, Y be two spaces and f, g : X → Y be two continuous

maps. We say that f is homotopic to g (and denote it by writing f ∼
= g) if
there exists a continuous map F : X × I → Y such that F (x, 0) = f (x) and
F (x, 1) = g(x) for all x ∈ X. The map F is called a homotopy from f to g.
We know that for each t ∈ I, the map it : X → X × I defined by it (x) =

(x, t) is an embedding. So, ft = F ◦ it : X → Y is a family of continuous maps
from X to Y, where t runs over the interval I. By the definition of homotopy F,
ft is the map f for t = 0 and for t = 1 it is the map g. Thus, a homotopy F is
simply a family of continuous maps from X to Y which starts from f , changes
continuously with respect to t and terminates into the map g. In other words, f
gets continuously transformed by means of the homotopy F and finally changes
or deforms itself into the map g. See Fig. 2.2.
ft
X x I Y
Fig. 2.2: F is a Homotopy from f = F ◦ i0 to g = F ◦ i1
Example 2.2.2. Let X = Rn = Y be the Euclidean spaces and let f, g : X → Y

be defined by f (x) = x and g(x) = 0, x ∈ X. Define the map H : Rn × I → Rn
by
H(x, t) = (1 − t)x.
2.2. Homotopy 39
Then, clearly, H is continuous and for all x ∈ X, H(x, 0) = x = f (x), H(x, 1) =

g(x). Thus, H is a homotopy from f to g and so f is homotopic to g.
Note that if we define F : Rn × I → Rn by F (x, t) = (1 − t2 )x, then F

is also a homotopy from f to g. In other words, there can be several ways of
deforming a map f into a given map g.
Example 2.2.3. This is a generalization of Example 2.2.2 above. Let X be any

topological space and Y be a convex subset of Rn , i.e., Y has the property that
whenever y1 , y2 ∈ Y, the line segment joining y1 to y2 is completely contained
in Y. Let f, g : X → Y be any two continuous maps. Then f is homotopic to g.
To see this, let us define the map H : X × I → Y by
H(x, t) = tg(x) + (1 − t)f (x).
Then we see at once that H is well-defined, continuous, it starts with f and
terminates into g. A homotopy of this kind is called a straight-line homotopy.
Example 2.2.4. Let S1 = {z ∈ C : |z| = 1} be the unit circle. We know that

we can also write S1 = {eiθ : 0 ≤ θ ≤ 2π}. Define two maps f, g : S1 → S1
by f (z) = z and g(z) = −z, z ∈ S1 . Then f is homotopic to g and the map
F : S1 × I → S1 defined by
F (eiθ , t) = ei(θ+tπ)
is a homotopy from f to g. Note that F is continuous because it is the compo-
sition of maps
S 1 × I → S 1 × S1 → S 1
(e , t) → (eiθ , eitπ ) → ei(θ+tπ) ,
iθ
where the second map is multiplication of complex numbers. Note that in this
example, the family of maps {ft : S1 → S1 } is just the family of rotations by
the angle tπ, 0 ≤ t ≤ 1.
The next result implies that the set of all maps from a space X to a space
Y can be decomposed into disjoint equivalence classes.
Theorem 2.2.5. Let X, Y be fixed topological spaces and C(X, Y ) denote the
set of all continuous maps from X to Y . Then the relation of “being homotopic
to” is an equivalence relation in the set C(X, Y ).
Proof. Note that each continuous map f : X → Y is homotopic to itself because
H : X × I → Y defined by H(x, t) = f (x) is a homotopy from f to itself. Next,
suppose H : f ∼
= g. Then the map H : X × I → Y defined by
H (x, t) = H(x, (1 − t))
is a homotopy from g to f ; to see this, note that H (x, 0) = H(x, 1) = g(x) and
H (x, 1) = H(x, 0) = f (x) for all x ∈ X. Moreover, H is continuous because
H is simply the composite of continuous maps
X ×I →X ×I →Y,
where the first map is the map (x, t) → (x, 1 − t) and the second is H. The
first map itself is continuous because its composite with the two projection
maps, viz., (x, t) → x and (x, t) → (1 − t) is continuous. Thus, the relation is
symmetric. Finally, suppose H1 : f ∼ = g and H2 : g ∼
= h. Define a map H : X ×
I → Y by

H1 (x, 2t), 0 ≤ t ≤ 1/2,
H(x, t) =
H2 (x, 2t − 1), 1/2 ≤ t ≤ 1.
Then H is continuous by the continuity lemma and H(x, 0) = f and
H(x, 1) = h. Hence H : f ∼
= h, proving the relation to be transitive.
The relation of homotopy in the set C(X, Y ) of all continuous maps,

therefore, decomposes this set into mutually disjoint equivalent classes. The
equivalence classes are called the homotopy classes of maps from X to Y and
the set of all homotopy classes is denoted by [X, Y ]. We will need the following
result very often.
Theorem 2.2.6. Let f1 , g1 : X → Y be homotopic and f2 , g2 : Y → Z be also

homotopic. Then the composite maps f2 ◦ f1 , g2 ◦ g1 : X → Z are homotopic
too, i.e., composites of homotopic maps are homotopic.
Proof. Let H1 : f1 ∼= g1 and H2 : f2 ∼ = g2 . Then, clearly, f2 ◦ H1 : X × I → Z
is a homotopy from f2 ◦ f1 to f2 ◦ g1 . Next, define a map H : X × I → Z by
H(x, t) = H2 (g1 (x), t), i.e., the map H is simply the following composite:
H2
X × I → Y × I −→ Z
(x, t) → (g1 (x), t) → H2 (g1 (x), t)
Then H is continuous and H(x, 0) = H2 (g1 (x), 0) = f2 (g1 (x)), H(x, 1) =
H2 (g1 (x), 1) = g2 (g1 (x)), i.e., H : f2 ◦ g1 ∼
= g2 ◦ g1 . Now, because f2 ◦ f1 is
homotopic to f2 ◦ g1 and f2 ◦ g1 is homotopic to g2 ◦ g1 , it follows by the
transitive property of homotopy relation that f2 ◦ f1 is homotopic to g2 ◦ g1 .
Exercises
1. Let X be a topological space and Y ⊂ S2 be the open upper hemisphere.

Prove that any two maps f, g : X → Y are homotopic.
2. Let P = {p} be a point space and X be a topological space. Show that
X is path connected if and only if the set [P, X] of homotopy classes of
maps is a singleton.
2.3. Contractible Spaces and Homotopy Type 41
3. Let X be a discrete space. Show that if a map f : X → X is homotopic to

the identity map IX : X → X, then f = IX . (Hint: The given condition
implies that there is a path joining x and f (x).)
4. Suppose X is a connected space and Y is a discrete space. Prove that the
two maps f, g : X → Y are homotopic if and only if f = g.
5. Let S1 be the unit circle of the complex plane and f, g : S1 → S1 be two
maps defined by f (z) = z and g(z) = z 2 . What is wrong in saying that
the map F : S1 × I → S1 defined by F (z, t) = z t+1 is a homotopy from f
to g?
6. Let X be a locally compact Hausdorff space and the set C(X, Y ) of all
continuous maps from X to Y be given the compact open topology. Prove
that two maps f, g ∈ C(X, Y ) are homotopic if and only if these can be
joined by a path in the space C(X, Y ).
(Hint: Use the exponential correspondence theorem. (Spanier [15] p. 6))
2.3 Contractible Spaces and Homotopy Type
The notion of a contractible space is very important and the definition

itself has some geometric appeal, as we shall see later. Recall that a map
f : X → Y is said to be a constant map provided each point of X is mapped by
f to some fixed point y0 ∈ Y . If this is the case, then it is convenient to denote
such a constant map by the symbol Cy0 i.e., Cy0 (x) = y0 , for every x ∈ X. We
have
Definition 2.3.1. A topological space X is said to be a contractible space if the

identity map IX : X → X is homotopic to some constant map Cx : X → X,
where, of course, x ∈ X. Any homotopy from IX to Cx is called a contraction
of the space to the point x ∈ X.
There are numerous examples of contractible spaces. For instance, we note

that any convex subset of an Euclidean space Rn is contractible. For, let S be
a convex subset of Rn . This means for any two points x, y ∈ S, the point
tx + (1 − t)y is also in S for all t, 0 ≤ t ≤ 1. Now let x0 ∈ S. Define a map
H : S × I → S by
H(x, t) = (1 − t)x + tx0
Then it is clear that H is a homotopy from the identity map on S to the
constant map Cx0 : S → S. Hence, H is a contraction and so S is contractible.
In particular, the Euclidean space Rn , the disk Dn are contractible spaces.
More generally, a subspace X of Rn is said to be star-shaped if there exists
a point x0 ∈ X such that the line segment joining any point of X to x0 lies
completely in X. For example, the subset X ⊂ R2 (Fig. 2.3) is star-shaped.
Fig. 2.3: A star-shaped region
Then one can easily see that X is contractible to the point x0 and the same
contraction, as defined above, works in this case also. Now, we ask the following
question: Determine whether or not the n-sphere Sn , n ≥ 1 is contractible? It
is obviously not star-shaped. That does not mean, however, that X is not
contractible (see Example 2.3.10). The answer to this question is “No”; it will
take quite sometime before we can prove it. The following concept is again
extremely basic.
Definition 2.3.2. Let f : X → Y be a continuous map. We say that f is a

homotopy equivalence if there exists a continuous map g : Y → X such that
g ◦ f is homotopic to the identity map IX on X and f ◦ g is homotopic to
the identity map IY on Y . Two spaces X and Y are said to be homotopically
equivalent or of the same homotopy type if there exists a homotopy equivalence
from one to the other.
We must observe that two homeomorphic spaces are of the same homotopy
type. For, suppose X and Y are homeomorphic and let f : X → Y be a home-
omorphism. Then the inverse map f −1 : Y → X is continuous and satisfies the
condition that f −1 ◦ f ∼
= IX and f ◦ f −1 ∼= IY . This means f is a homotopy
equivalence, i.e., X and Y are of the same homotopy type. The converse is not
true. We have
Example 2.3.3. Consider the unit disk D2 (open or closed) and a point
x0 ∈ D2 . Let i : P = {x0 } → D2 be the inclusion map and Cx0 : D2 → P
be the constant map. Then evidently Cx0 ◦ i = IP . On the other hand, the map
H : D2 × I → D2 defined by
H(x, t) = (1 − t)x + tx0
is a homotopy from ID2 to i ◦ Cx0 . Thus, D2 is of the same homotopy type as
a point space P and these are clearly not homeomorphic.
One can easily verify that the relation of “homotopy equivalence” in the
class of all topological spaces is an equivalence relation. Also, in view of the
above example, the relation of homotopy equivalence is strictly weaker than
the relation of “homeomorphism”. It is also clear from the above example that
if a space X is compact then a space Y , which is homotopically equivalent to

X, need not be compact, i.e., the compactness is not a homotopy invariant.
Similarly, the topological dimension is not a homotopy invariant because di-
mension of the plane R2 is 2 whereas a point has dimension zero. These and
several other examples show that topological invariants are, in general, not
“homotopy” invariants and so the homotopy classification of spaces is quite a
weak classification. However, it is still very important because we will later de-
fine some homotopy invariants which would be evidently topological invariants.
These invariants would also be computed for a large class of topological spaces.
The moment we notice that if any of these invariants is not the “same” for any
two given spaces X and Y , we can immediately assert that X and Y are not of
the same homotopy type and therefore cannot be homeomorphic. This is a well
established strategy of algebraic topology for proving that two given spaces are
not homeomorphic. Example 2.3.3 is a trivial case of the following:
Theorem 2.3.4. A topological space X is contractible if and only if X is of the

same homotopy type as a point space P = {p}.
Proof. Suppose X is contractible. Let H : X × I → X be a homotopy from the
identity map IX to the constant map Cx0 : X → X. Define maps i : P → X
and C : X → P by i(p) = x0 and C(x) = p, x ∈ X. Then, clearly, c ◦ i = IP .
Also, the map H is a homotopy from IX to i ◦ C because H(x, 0) = x and
H(x, 1) = Cx0 (x) = x0 = i ◦ C(x)
for each x ∈ X. Hence X and P are of the same homotopy type.
Conversely, suppose there are maps f : X → P and g : P → X such that
g◦f ∼ = IX and f ◦ g ∼= IP . Let g(p) = x0 and H : X × I → X be a homotopy
from IX to g ◦ f . Then because g ◦ f (x) = g(f (x)) = g(p) = x0 , for all x ∈ X,
g ◦ f is the constant map Cx0 : X → X. Thus, IX is homotopic to the constant
map Cx0 and so the space X is contractible.
Thus, the contractible spaces are precisely those spaces which are homo-
topically equivalent to a point space. The intuitive picture of a contractible
space X is quite interesting. A homotopy H which starts from the identity
map on X and terminates into a constant map Cx0 , x0 ∈ X, should be thought
of as a continuous deformation of the space X which finally shrinks the whole
space X into the point x0 . In other words, if we imagine the unit interval I as a
time interval then at the time t = 0, every point x ∈ X is at its original place;
as t varies from 0 to 1 continuously, x moves continuously and approaches the
point x0 ; even x0 moves accordingly and comes back to itself. Furthermore, all
of these points move simultaneously in such a way that their relative positions
do not change abruptly. Thus, if we follow the movement of an arbitrary point
x ∈ X, we note that it describes a path in X starting from x which terminates
at x0 . In particular, we intuitively see that X is path connected. We make this
statement precise:
H
a x0 a x
0
X x I X
Fig. 2.4: Contractible space is path connected
Proposition 2.3.5. If X is a contractible space, then X is path connected.

Proof. Suppose X is contractible to a point x0 and H : X×I → X is a homotopy
from IX to Cx0 . Let a ∈ X. It suffices to show that a can be joined to x0 by a
path in X. Note that H maps whole of bottom to itself whereas the entire top
to the point x0 . Define a map f : I → X × I by f (t) = (a, t) and note that it
is continuous. Then ω = H ◦ f is a path (Fig. 2.4) in X such that
ω(0) = H(f (0)) = H(a, 0) = a,
and
ω(1) = H(f (1)) = H(a, 1) = x0 .
Now suppose X is contractible. This means X can be contracted to some

point x ∈ X. Can we then contract X to an arbitrary point x0 ∈ X? The
answer to this question is “yes”. We will now explain this. Let us prove
Proposition 2.3.6. A topological space X is contractible if and only if an ar-

bitrary map f : T → X from any space T to X is homotopic to a constant
map.
Proof. Suppose X is contractible. This means the identity map IX : X → X
is homotopic to some constant map, say Cx0 : X → X. Now let f : T → X be
any map. By Theorem 2.2.6, we find that IX ◦ f is homotopic to Cx0 ◦ f . But
IX ◦ f = f and Cx0 ◦ f : T → X is the constant map.
For the converse, take T = X and the map f : T → X to be the identity
map. Then, by the given condition, we find that IX : X → X is homotopic to
a constant map, i.e., X is contractible.
It now follows from above that if X is a contractible space, then any map
f : X → X is homotopic to a constant map Cx0 : X → X. In particular, for
any x ∈ X, the constant map Cx and the identity map IX : X → X both are
homotopic to Cx0 , i.e., Cx is homotopic to IX for all x ∈ X.
Corollary 2.3.7. If X is a contractible space, then the identity map IX : X → X

is homotopic to a constant map Cx : X → X for all x ∈ X. In particular, X
can be contracted to any arbitrary point of X.
Once again, let X be a contractible space. We ask now a slightly stronger

question. Can we contract X to some point x0 ∈ X so that the point x0 does
not move at all? The answer to this question is “No”. We will give an example
of this later on (See Example 2.3.10). This question leads to the concept of
relative homotopy which is stronger than the homotopy defined earlier.
Definition 2.3.8. Let A ⊂ X be an arbitrary subset and f, g : X → Y be two

continuous maps. We will say that f is homotopic to g “relative to A” if there
exists a continuous map F : X × I → Y such that
F (x, 0) = f (x), F (x, 1) = g(x), for all x ∈ X, and
F (a, t) = f (a) = g(a), for all a ∈ A.
Note that if we take A to be null set ∅, then the concept of relative ho-
motopy reduces to that of homotopy. It is also to be noted that if f, g : X → Y
are to be homotopic relative to some subset A of X, then f and g must agree
on A to start with. The map f will change into the map g by a family of
continuous maps ht : X → Y , t ∈ I, but the points of A will remain unchanged
under ht when t varies from 0 to 1.
If A ⊆ X is a fixed subset and C(X, Y ) denotes the set of all continuous

maps f : X → Y , then following the proof of Theorem 2.2.5, one can prove
that the relation of being “relatively homotopic to” with respect to A is an
equivalence relation in the set C(X, Y ).
Ca
V U
IX
X x I X
Fig. 2.5: Strongly contractible at a is semi-locally path connected at a.
Let us now assume that a space X is contractible to a point a ∈ X

relative to the subset A = {a}, i.e., X can be contracted to the point a in
such a manner that all points of X move and finally get changed into the
point a, but the point a itself does not undergo any change. In other words, we
have a homotopy F : X × I → X which starts with the identity map IX and
terminates into the constant map Ca relative to the subset {a}. This means,
(see Fig. 2.5) under the continuous map F , the line {a} × I is mapped to the
point a ∈ X. If we take any neighbourhood U of a, the continuity of F will
give for each t ∈ I, neighbourhoods Vt (a) of a in X and W (t) of t in I such
that F (Vt (a) × W (t)) ⊂ U . The compactness of I means that the open covering
{W (t) : t ∈ I} of I will have a finite subcover, say, W (t1 ), . . . , W (tn ), such that
n
F (Vti (a) × W (ti )) ⊂ U , for all i = 1, . . . , n. Therefore, V (a) = i=1 Vti (a) is a
neighbourhood of a in X such that F (V (a) × I) ⊂ U . Now, if b ∈ V (a), then
considering the image F (V (a) × I) in U , we find that b can be joined to a by
a path which lies in U . This completes the proof of the following:
Theorem 2.3.9. If a space X is contractible to a point a ∈ X relative to the

subset {a}, then for each neighbourhood U of a in X, there exists a neighbour-
hood V of a contained in U such that any point of V can be joined to a by a
path lying completely inside U , i.e. X is semi-locally path connected.
Let us now consider the famous
Example 2.3.10. (Comb Space). We consider the following subset C (Fig. 2.6)
of the plane (shown only partly).
(1,1/3)
(0,0) (1/2,0) (1,0)

Fig. 2.6
It consists of the horizontal line segment joining (0,0) to (1,0) and vertical unit
closed line segments standing on points (1/n, 0) for each n = 1, 2, . . . together
with the line segment joining (0,0) with (0,1).
The comb space C is contractible. The projection map p : C → L, where L
is the line segment joining (0,0) with (1,0), is a homotopy equivalence. For, if
i : L → C is the inclusion map, then p ◦ i = IL and the map F : C × I → C
defined by
F ((x, y), t) = (x, (1 − t)y).
is a homotopy between IC and i ◦ p. We already know that L is homeomorphic
to the unit interval which is of the same homotopy type as a point space and
so by Theorem 2.3.1, C is contractible.
C is not contractible relative to {(0,1)}. Note that any small neighbourhood V
of (0,1) has infinite number of path components. So, if we take the neighbour-
hood U = D ∩ C of (0,1) in C, where D is the open disk around (0,1) of radius
1/2, then U cannot have any neighbourhood V each of whose points can be
joined to (0,1) by a path lying in U . This observation combined with Theorem
2.3.9, shows that C is not contractible relative to {(0,1)}.
Remark 2.3.11. Note that if f, g : X → Y are two continuous maps which are
homotopic relative to some subset A of X, then obviously f is homotopic to g.
The converse, however, is not true, i.e., there are maps f, g : X → Y which are
homotopic, and even agree on a subset A of X, yet they need not be homotopic
relative to A. For example, consider the identity map IX : X → X of the comb
space X = C and the constant map C(0,1) : X → X. Then, obviously, IX and
C(0,1) agree on (0,1) and are also homotopic by Corollary 2.3.1, since X is
contractible. However, we have just seen in the above example that IX and
C(0,1) are not homotopic relative to (0,1). This means the concept of relative
homotopy is definitely stronger than that of homotopy.
There are some fundamental concepts related to “contractible” spaces. It

is appropriate to discuss them now.
Definition 2.3.12. Let A ⊂ X. We say that A is a retract of X if there exists

a continuous map r : X → A such that r(a) = a, for all a ∈ A. The map r is
called the retraction map.
If i : A → X is the inclusion map, then the condition r(a) = a, for all

a ∈ A is equivalent to saying that r ◦ i = IA . Thus, A is a retract of X if and
only if the inclusion map i has left inverse. As an example, note that every
single point x0 ∈ X of an arbitrary topological space X is a retract of X and
the constant map Cx0 is the retraction map. Let X = [0, 1], the unit interval
and A = {0, 1}, the boundary of X. Then A cannot be a retract of X because
X is connected whereas A is disconnected. Let us ask a general question : Let
X = Dn+1 be the (n + 1)-disk and Sn be its boundary. Determine whether or
not Sn is a retract of Dn+1 . The answer is again “No” and this fact is known
as Brouwer’s No Retraction Theorem. The proof, however, will be given later
on.
Definition 2.3.13. A topological space X is said to be deformable into a subspace

A ⊂ X if there is a map f : X → A which is right homotopy inverse of the
inclusion map i : A → X, i.e., the identity map IX is homotopic to i ◦ f : X →
X.
The above definition asserts that there is a homotopy, say, D : X × I → X

such that
D(x, 0) = x, D(x, 1) = i(f (x)) = f (x).
Any such homotopy D is called a deformation of X into A and we say that X
is deformable into A. It must be observed that the homotopy D, which starts
with identity map IX : X → X, simply moves each point of X continuously,
including the points of A and finally pushes every point of X into a point of
A. In particular, if a space X is deformable into a point a ∈ X, then X is
contractible and vice versa. If we can find a deformation D which deforms X
into A but the points of A do not move at all, then the homotopy D will be
“relative to A” and we say that X is strongly deformable into A. In such a
case, note that for each a ∈ A, D(a, 1) = f (a) = a and so the map f : X → A
is automatically a retraction of X onto A. We have
Definition 2.3.14. A space X is said to be strongly deformable into a subspace

A if there is a continuous map f : X → A which is the right homotopy inverse
of the inclusion map i : A → X relative to A, i.e., the identity map IX : X → X
is homotopic to i ◦ f : X → X relative to A.
Clearly, if X is strongly deformable into a subspace A, then it is also

deformable into A. But the converse is not true even if the map f : X → A is
onto: the comb space C (Example 2.3.10) is deformable into the point {(0,1)}
because it is contractible and so the identity map IX is homotopic to the map
i ◦ C(0,1) where C(0,1) is the constant map X → X. However, we have already
seen that IX is not homotopic to i ◦ C(0,1) relative to the point {(0,1)}. Finally,
we have
Definition 2.3.15. A subspace A of a topological space X is said to be a de-

formation retract of X if X is deformable into A so that the final map is a
retraction of X onto A.
Definition 2.3.16. A subspace A of a topological space X is said to be a strong

deformation retract of X if X is deformable into A strongly (A is then auto-
matically a retract of X).
It follows from the above definitions that if A is a deformation retract

of X, then the inclusion map i : A → X has a two-sidedhomotopy inverse,
i.e., it is a homotopy equivalence, and consequently A and X are of the same
homotopy type. On the other hand, if A is a strong deformation retract of X,
then A is homotopically equivalent to X and something more is true, viz., X
can be deformed into A without moving the points of A at all. For instance,
the point (0,1) of the comb space X is a deformation retract of X but is not a
strong deformation retract of X. Quite often we will use the following:
Example 2.3.17. For n ≥ 1, Sn ⊂ Rn+1 − {(0, . . . , 0)} = X is a strong defor-

mation retract of X. Fig. 2.7 depicts the case n = 1.
'$
l L
&%
Fig. 2.7
Each infinite line segment L starting from its origin intersects the circle S1 at
exactly one point, say, l. Since origin is not a point of R2 −{(0, 0)}, these lines are
disjoint and their union is R2 − {(0, 0)}. We define a map r : R2 − {(0, 0)} → S1
by r(x) = l for all points x ∈ L. Then, clearly r is continuous and S1 becomes
a retract of R2 − {(0, 0)}. Let us define a deformation D : (R2 − {(0, 0)} × I →
R2 − {(0, 0)} by
D(x, t) = (1 − t)x + t x

x
.
Then D is clearly a strong deformation retraction of R2 − {(0, 0)} relative to
S1 into S1 . A similar argument shows that Sn is a strong deformation retract
of Rn+1 − {(0, 0)}.
Exercises
1. Let A ⊂ X be a retract of X where X is Hausdorff. Then prove that A
must be closed in X. (This implies that an open interval (0,1) can never
be a retract of any closed subset of the real line.)
2. Let X be a connected space and x0 , x1 ∈ X be two points of X which have
disjoint open neighbourhoods in X. Show that A = {x0 , x1 } can never be
a retract of X.
3. Prove that a space X is contractible if and only if every map f : X → T
to any space T is null-homotopic.
4. Show that if A is a strong deformation retract of X and B is a strong
deformation retract of A, then B is a strong deformation retract of X.
5. Prove that an arbitrary product of contractible space is again contractible.
6. Prove that a retract of a contractible space is contractible.
7. For any space X consider the cylinder X × I over X and collapse the
top X × 1 of this cylinder to a point. The resulting quotient space, called
cone over X, is denoted by C(X). Prove that C(X) is contractible for any
space X.
8. Let I 2 = [0, 1] × [0, 1] be the unit square and C ⊂ I 2 be the comb space
(Example 2.3.10). Prove that C is not a retract of I 2 . (Hint: Given any
open neighbourhood U = B((0, 12 ), 14 ) ∩ C in C, there exist a connected
neighbourhood V ⊂ U of (0, 12 ) in I × I such that r(V ) ⊂ U , but r(V ) is
disconnected.)
9. Determine which of the following spaces are contractible:
(i) Unit interval I = [0, 1].
(ii) S2 − {p}, where S2 is a 2-sphere and p is any point of S2 .
(iii) Any solid or hollow cone in R3 .
(iv) The subspace {0} ∪ {1/n : n ∈ N } of real line.
10. Consider the following subspace X of plane R2 ; X consist of all closed line
segments joining origin with points (1, n1 ), n ≥ 1 and the line {(x, 0) | 0 ≤
x ≤ 1}. Prove that X is contractible, but none of the points (x, 0), x ≥ 0,
is a strong deformation retract of X.
11. Let I be the unit interval and X be any path connected space. Prove that
the sets [I, X] and [X, I] each has only one element. (Hint: The space I
is contractible)
12. Give an example of a space X which is of the same homotopy type as a

discrete space D = {0, 1, 2, 3}, but is not homeomorphic to D.
13. Prove that a homotopy invariant is also a topological invariant. Give an
example to show that a topological invariant need not be a homotopy
invariant. (Hint: There are contractible spaces which are not compact,
not locally compact, not locally conneted etc.)
2.4 Fundamental Group and its Properties
Recall that a path in a topological space X is just a continuous map

α : I = [0, 1] → X; α(0) is called the initial point and α(1) is called the
terminal point of the path α. If α, β are two paths in X such that α(1) = β(0),
then we can define a new path (Fig. 2.8), called the product of α and β denoted
by α ∗ β, as follows:

α(2t) 0 ≤ t ≤ 1/2
(α ∗ β)(t) = (4.1)
β(2t − 1) 1/2 ≤ t ≤ 1
Note that α ∗ β : I → X is continuous by the continuity lemma, the initial
point of α ∗ β is the initial point of α and the terminal point of α ∗ β is the
terminal point of β.
α
I β
X
Fig. 2.8: Product of two paths
One can verify that if α, β, γ are three paths in X such that α(1) = β(0),
β(1) = γ(0), then (α ∗ β) ∗ γ and α ∗ (β ∗ γ) are paths in X which are not
necessarily the same paths. To see this, just apply definition (2.4.1) and look
at the image of the point t = 1/2 under both the paths. Therefore, the product
of paths is not an associative operation. Let us now fix a point x0 ∈ X and
consider the set of all closed paths at x0 , i.e., those paths whose initial and
terminal points are x0 ; such a path is also known as loop in X based at x0 . It is
clear that the product of two loops based at x0 is always defined. The difficulty
that the product of loops based at x0 need not be associative is still there. To
surmount this difficulty and to finally get a group structure, we will introduce
an equivalence relation in the set of all loops in X based at x0 ∈ X. First, we
have an important as well as general
2.4. Fundamental Group and its Properties 51
Definition 2.4.1. Let α, β be two paths in X with the same initial and terminal
points, i.e., α(0) = β(0) = x0 , α(1) = β(1) = x1 . We will say that α is
equivalent to β, and write it as α ∼(x0 ,x1 ) β, if there exists a homotopy between
α and β relative to the subset {0, 1} of I. In other words, the homotopy keeps
the end points fixed.
Thus, the path α is equivalent to β if there exists a continuous map

H : I × I → X such that
H(s, 0) = α(s), H(s, 1) = β(s)
H(0, t) = α(0) = β(0), H(1, t) = α(1) = β(1)
for all s ∈ I and for all t ∈ I. In other words, the path α changes continuously
and finally it becomes the path β, but during all this transformation the end
points remain fixed (Fig. 2.9).
(0,1) (1,1)
β
H x1
x0
α
(0,0) IXI (1,0)
Fig. 2.9: H is a path-homotopy
The relative homotopy H, which keeps the end points fixed, is sometimes
called a path homotopy, just for convenience. The next result, which is in fact
true, more generally, for homotopy relative to any subset A of the domain space
(cf. Definition 2.3.8), shows that the above relation is an equivalence relation.
Theorem 2.4.2. Let x0 , x1 ∈ X. Then the relation of “being equivalent” in

the set of all paths starting from x0 and terminating at x1 is an equivalence
relation.
Proof. Let α : I → X be any path with α(0) = x0 , α(1) = x1 . Then the map
H : I × I → X defined by H(s, t) = α(s) is a homotopy from α to itself relative
to {0,1}. Thus, the relation is reflexive. Next, suppose α, β are two paths from
x0 to x1 and H : I × I → X is a continuous map such that
H(s, 0) = α(s), H(s, 1) = β(s), ∀s ∈ I
H(0, t) = x0 , H(1, t) = x1 , ∀t ∈ I.
Define a map H : I × I → X by
H (s, t) = H(s, 1 − t).
Then H is continuous and has the property that
H (s, 0) = H(s, 1) = β(s), H (s, 1) = α(s), s ∈ I
H (0, t) = x0 , H (1, t) = x1 , ∀t ∈ I.
Thus, α ∼ β relative to {0,1} implies β ∼ α relative to {0,1}, i.e., the relation
is symmetric. To prove the transitivity of the relation, suppose α ∼ β relative
to {0,1}, β ∼ γ relative to {0,1}. Let H1 and H2 be two homotopies such that
H1 (s, 0) = α(s), H1 (s, 1) = β(s), ∀s ∈ I
H1 (0, t) = x0 , H1 (1, t) = x1 , ∀t ∈ I
and
H2 (s, 0) = β(s), H2 (s, 1) = γ(s), ∀s ∈ I
H2 (0, t) = x0 , H2 (1.t) = x1 , ∀t ∈ I.

H1 (s, 2t) 0 ≤ t ≤ 1/2
H(s, t) =
H2 (s, 2t − 1) 1/2 ≤ t ≤ 1
Then H is continuous by the continuity lemma and is indeed a homotopy
relative to {0,1} from α to γ. This completes the proof of the theorem.
When x0 = x1 , we conclude the following result.
Corollary 2.4.3. The relation “being equivalent” in the set of all loops in X
based at x0 ∈ X is an equivalence relation.
Next, we are going to deal with only loops in X based at given point
x0 ∈ X. If α, β are two loops based at x0 which are equivalent, then we will
write this as α ∼x0 β. Also, the equivalence class of a loop α based at x0 will
be denoted by the symbol [α] and called the homotopy class of the loop α.
It must be emphasized at this point that if α is treated as a map α : I → X
with α(0) = α(1) = x0 , then the homotopy class of the map α, according to
Theorem 2.2.5 is different from the path homotopy class of loop α specified by
Theorem 2.4.2. In fact, the former homotopy class is, in general, larger than
the latter path-homotopy class. Let π1 (X, x0 ) denote the set of all homotopy
classes of loops in X based at x0 , i.e.,
π1 (X, x0 ) = {[α] : α is a loop in X based at x0 }.
The next proposition implies that the product of loops induces a product in
the set π1 (X, x0 ) of all homotopy classes of loops based at x0 . Recall that if
α, β are two loops at x0 , then their product α ∗ β is also a loop at x0 ∈ X.
Proposition 2.4.4. Suppose α, β, α , β are loops in X based at x0 . If α ∼x0 α ,

β ∼x0 β then α ∗ β ∼x0 α ∗ β .
Proof. Let H1 be a homotopy from α to α , H2 be a homotopy from β to β ,
i.e., H1 , H2 are maps from I × I → X such that
H1 (s, 0) = α(s), H1 (s, 1) = α (s), ∀s ∈ I

H1 (0, t) = x0 = H1 (1, t), ∀t ∈ I
and
H2 (s, 0) = β(s), H2 (s, 1) = β (s), ∀s ∈ I
H2 (0, t) = x0 = H2 (1, t), ∀t ∈ I.

H1 (2s, t) 0 ≤ s ≤ 1/2
H(s, t) =
H2 (2s − 1, t) 1/2 ≤ s ≤ 1.
Then H is continuous by continuity lemma, and

H1 (2s, 0) 0 ≤ s ≤ 1/2
H(s, 0) =
H2 (2s − 1, 0) 1/2 ≤ s ≤ 1

α(2s) 0 ≤ s ≤ 1/2
=
β(2s − 1) 1/2 ≤ s ≤ 1
= (α ∗ β)(s).
By a similar calculation, H(s, 1) = (α ∗β )(s) and H(0, t) = x0 = H(1, t),

∀ t ∈ I. Thus, α ∗ β ∼x0 α ∗ β .
One can easily observe that the above proof yields the following general
result, viz.,
Corollary 2.4.5. Let α, α be two path homotopic paths joining x0 with x1 and
β, β be two path homotopic paths joining x1 to x2 . Then α∗β is path-homotopic
to α ∗ β joining x0 to x2 , and the path homotopy can be chosen so that the
point x1 remains fixed.
Sometimes we will need to consider the path homotopy classes of paths

joining x0 to x1 . If α is a path joining x0 with x1 , then [α] will also be used to
denote the path homotopy class represented by α. The set of all path homo-
topy classes of paths joining x0 to x1 will be denoted by π1 (X, x0 , x1 ). In this
terminology, we can define an operation
◦ : π1 (X, x0 , x1 ) × π1 (X, x1 , x2 ) → π1 (X, x0 , x2 )
by
[α] ◦ [β] = [α ∗ β].
The above corollary says that the map ◦ is well defined. In case x0 , x1 , x2 are
the same points, the map ◦ defines a binary operation in the set π1 (X, x0 ) of
all homotopy classes of loops based at x0 . Consequently, we can state the next
Definition 2.4.6. Let [α], [β] be any two elements of π1 (X, x0 ). Then we define
their product [α] ◦ [β] by
[α] ◦ [β] = [α ∗ β].
The following basic result can now be proved.
Theorem 2.4.7. The set π1 (X, x0 ) of all path homotopy classes of loops based
at x0 is a group with respect to the binary operation “ ◦” defined above.
Proof. We must prove that the operation ◦ is associative, there exists identity
element in π1 (X, x0 ) and each element of π1 (X, x0 ) has an inverse in π1 (X, x0 ).
The proof of each of these statements is achieved by constructing a suitable
path homotopy relative to {0,1} between appropriate paths and we discuss
them below one by one:
The operation is associative: Let [α], [β], [γ] be three elements of π1 (X, x0 ).
Since
([α] ◦ [β]) ◦ [γ] = [(α ∗ β) ∗ γ]
and
[α] ◦ ([β] ◦ [γ]) = [α ∗ (β ∗ γ)],
it is sufficient to show that (α ∗ β) ∗ γ ∼x0 α ∗ (β ∗ γ). By definition

⎧
⎨ α(4s) 0 ≤ s ≤ 1/4
((α ∗ β) ∗ γ)(s) = β(4s − 1) 1/4 ≤ s ≤ 1/2
⎩
γ(2s − 1) 1/2 ≤ s ≤ 1
and
⎧
⎨ α(2s) 0 ≤ s ≤ 1/2
(α ∗ (β ∗ γ))(s) = β(4s − 2) 1/2 ≤ s ≤ 3/4
⎩
γ(4s − 3) 3/4 ≤ s ≤ 1.
Thus we should define a homotopy H : I × I → X such that

H(s, 0) = ((α ∗ β) ∗ γ)(s), H(s, 1) = (α ∗ (β ∗ γ))(s)
(2.6.1)
H(0, t) = x0 = H(1, t), ∀ s, t ∈ I
Such a homotopy is given by the formula

⎧
⎨ α(4s/(t + 1)) 0 ≤ s ≤ (t + 1)/4
H(s, t) = β(4s − 1 − t) (t + 1)/4 ≤ s ≤ (t + 2)/4
⎩
γ((4s − 2 − t)/(2 − t)) (t + 2)/4 ≤ s ≤ 1
The motivation for writing this homotopy comes from Fig. 2.10.
(0,1) (1,1)
α β γ

A B

α β γ
(0,0) (1,0)
Fig. 2.10
We divide the square I × I into three quadrilaterals:

Q1 : 0 ≤ s ≤ (t + 1)/4
Q2 : (t + 1)/4 ≤ s ≤ (t + 2)/4
Q3 : (t + 2)/4 ≤ s ≤ 1, t ∈ I
For instance, the equation of the line joining (1/4,0) and (1/2,1) would
be t = 4s − 1 and so the equation of the region Q1 would be
0 ≤ s ≤ (t + 1)/4, 0 ≤ t ≤ 1.
For a fixed t ∈ I, a typical horizontal line AB would have three parts. When
t moves from 0 to 1, these three parts also change their positions. For t = 0,
we get a partition defining (α ∗ β) ∗ γ and for t = 1, we get a partition defining
α ∗ (β ∗ γ). The map H is defined by α on Q1 , β on Q2 , γ on Q3 , each of which
is continuous. On their common boundary, the two definitions match yielding
a nice map H. Hence, by the continuity lemma, H is continuous. Moreover,
conditions (2.6.1) are evidently satisfied. This completes the proof that ◦ is
associative.
Remark 2.4.8. If α is a path joining x0 with x1 , β is a path joining x1 with x2
and γ is a path joining x2 with x3 , then the same proof as above says, more
generally, that
([α] ◦ [β]) ◦ [γ] = [(α ∗ β) ∗ γ]
= [α ∗ (β ∗ γ)]
= [α] ◦ ([β] ◦ [γ]).
There exists an identity element in π1 (X, x0 ): Let us consider the constant
loop Cx0 : I → X. We claim that the class [Cx0 ] ∈ π1 (X, x0 ) is an identity
element, i.e., for each loop α in X based at x0 , we show that Cx0 ∗ α ∼x0 α
and α ∗ Cx0 ∗ ∼x0 α. For this let us consider Fig. 2.11.
t6 (1,1)
@ α
@
@
@
@ α s-
(0,0) (1,0)
Fig. 2.11

x0 0 ≤ s ≤ (1 − t)/2
H(s, t) =
α((2s + t − 1)/(t + 1)) (1 − t)/2 ≤ s ≤ 1
Note that the equation of the line joining (1/2,0) and (0,1) is 2s = 1−t. H maps
the whole triangle below this line to the point x0 and the two definitions of H
on the line 2s = 1 − t match. The continuity of H follows from the continuity
lemma. Furthermore,

x0 0 ≤ s ≤ 1/2
H(s, 0) =
α(2s − 1) 1/2 ≤ s ≤ 1
= (Cx0 ∗ α)(s)
H(s, 1) = α(s),
and H(0, t) = x0 = H(1, t). This shows that Cx0 ∗ α ∼x0 α. To prove that [Cx0 ]
is also right identity, we can write a suitable path homotopy by looking at Fig.
2.12.
t6 (1,1)
α
α Cx0 s-
(0,0) (1,0)
Fig. 2.12
This would be

α(2s/(1 + t)) 0 ≤ s ≤ (t + 1)/2
H(s, t) =
x0 (1 + t)/2 ≤ s ≤ 1
More generally, we have
Remark 2.4.9. Let α be any path joining x0 with x1 . Then the above proof
implies
(1) [Cx0 ] ◦ [α] = [Cx0 ∗ α] = [α],
(2) [α] ◦ [Cx1 ] = [α ∗ Cx1 ] = [α].
In other words, Cx0 serves as the left identity and Cx1 serves as the right
identity for any [α].
Each element of π1 (X, x0 ) has an inverse: Let [α] ∈ π1 (X, x0 ). We choose a
representative, say α, of the homotopy class [α]. For this α, we define a loop
α : I → X by
α (t) = α(1 − t).
Geometrically, α simply describes the same path as α, but in reverse direction.
We claim that the homotopy class [α ] ∈ π1 (X, x0 ) does not depend on the
choice of α from the class [α]. For, suppose β ∼x0 α by a homotopy H. Then
we can define a homotopy H : I × I → X by
H (s, t) = H(1 − s, t).

Then, obviously, H is a homotopy from β to α , i.e., [β ] = [α ]. Now we claim
that
[α ] ◦ [α] = [Cx0 ] = [α] ◦ [α ].
For this, we must construct a homotopy from α ∗ α to Cx0 relative to {0, 1}.
We now consider Fig. 2.13.
t6 Cx0 (1,1)
@
@
@
@
α α−1 @ s-
(0,0) (1/2,0) (1,0)
Fig. 2.13
Just as in earlier cases, the required homotopy H : I × I → X is given by

⎧
⎪
⎪ x0 0 ≤ s ≤ t/2
⎨
α(2s − t) t/2 ≤ s ≤ 1/2
H(s, t) =
⎪
⎪ α(2 − 2s − t) 1/2 ≤ s ≤ 1 − t/2
⎩
x0 1 − t/2 ≤ s ≤ 1.
It is easily verified that H is well defined, continuous and has the required
properties. By exactly a similar argument we can show that
[α ] ◦ [α] = [Cx0 ].
Thus, [α ] is an inverse of [α] in π1 (X, x0 ).
Remark 2.4.10. The same proof says that, more generally, if α is a path joining
x0 with x1 , then its inverse path α−1 has the following properties:
[α] ◦ [α−1 ] = [Cx0 ],
[α−1 ] ◦ [α] = [Cx1 ].
Remark 2.4.11. In the proof of the preceding theorem, it was enough to show
that π1 (X, x0 ) has a left identity and each element of π1 (X, x0 ) has a left inverse
– this is a result from elementary group theory. However, we have shown the
existence of two-sided identity and two-sided inverse only to give more practice
of writing path homotopies.
Remark 2.4.12. We also note that if α, β, γ, δ are four loops based at x0 ∈ X,

then by the associative law proved above, we find that
(α ∗ β) ∗ (γ ∗ δ) ∼x0 α ∗ (β ∗ (γ ∗ δ)) ∼x0 ((α ∗ β) ∗ γ) ∗ δ
and so the generalized associative law is valid in the sense that placing of
parentheses does not make any difference in the homotopy class. Therefore, we
can just ignore the parentheses and write the above loop only as [α ∗ β ∗ γ ∗ δ].
Definition 2.4.13. Let X be a topological space and x0 ∈ X. Then the group

π1 (X, x0 ) obtained in Theorem 2.4.7 is called the fundamental group or the
Poincaré group of the space X based at x0 ∈ X.
Remark 2.4.14. At this stage one would like to see examples of fundamental
groups of some spaces. We will give several examples somewhat later, but before
that it would be helpful to prove a few results on the behaviour of fundamental
groups so that one can have some idea about the possibilities of the nature of
fundamental group of a given space. We should also point out here that it is
in the very nature of algebraic topology that computing associated algebraic
objects is normally a long process and sometime can also be extremely difficult.
Having defined the fundamental group of a space X based at a point

x0 ∈ X, one would naturally like to ask: how important is the role of base point
x0 in the group π1 (X, x0 )? How are π1 (X, x0 ) and π1 (X, x1 ) related if x0
= x1 ?
In fact, if X is arbitrary, then a loop at x0 , being itself path connected, will
lie completely in the path component of x0 and so if x0 , x1 are in distinct path
components of X, then π1 (X, x0 ), π1 (X, x1 ) are not related at all. However, if
x0 , x1 belong to the same path component of X, then π1 (X, x0 ) and π1 (X, x1 )
are indeed isomorphic. This follows from the next
α w
x0
x1
Fig. 2.14
Theorem 2.4.15. Let X be a path connected space and x0 , x1 be any two points
of X. Then π1 (X, x0 ) and π1 (X, x1 ) are isomorphic. In fact, each path joining
x0 to x1 defines an isomorphism from π1 (X, x0 ) to π1 (X, x1 ).
Proof. Consider Fig. 2.14. Let ω : I → X be a path joining x0 to x1 and suppose
ω −1 is the inverse path of ω, i.e., ω −1 (t) = ω(1 − t) for each t ∈ I. If α is any
loop based at x0 , then it is clear that ω −1 ∗ α ∗ ω is a loop based at x1 . Thus,
we can define a map
Pω : π1 (X, x0 ) → π1 (X, x1 )
by
Pω [α] = [ω −1 ∗ α ∗ ω].
First, we check that Pω is well defined. Suppose α ∼x0 β and let H : I × I →
X be a homotopy relative to {0,1} from α to β. Then by Corollary 2.4.2,
ω −1 ∗ α ∗ ω ∼x0 ω −1 ∗ β ∗ ω, which means Pω [α] = Pω [β]. Now, let us see the
following computation:
Pω ([α] ◦ [β]) = Pω [α ∗ β]
= [ω −1 ∗ (α ∗ β) ∗ ω]
= [ω −1 ∗ α] ◦ [β ∗ ω]
= [ω −1 ∗ α ∗ Cx0 ] ◦ [β ∗ ω]
= [ω −1 ∗ α ∗ ω ∗ ω −1 ] ◦ [β ∗ ω]
= [ω −1 ∗ α ∗ ω] ◦ [ω −1 ∗ β ∗ ω]
= Pω [α] ◦ Pω [β].
If we use ω −1 instead of ω, then we get a homomorphism
Pω−1 : π1 (X, x1 ) → π1 (X, x0 ).
Now, for each [α] ∈ π1 (X, x0 ), we have
Pω−1 ◦ Pω [α] = Pω−1 [ω −1 ∗ α ∗ ω]
= [ω ∗ ω −1 ∗ α ∗ ω ∗ ω −1 ]
= [α],
which means Pω−1 ◦ Pω is identity on π1 (X, x0 ). By a similar argument,
we see that Pω ◦ Pω−1 is identity on π1 (X, x1 ). It follows that Pω is an
isomorphism.
Remark 2.4.16. It follows from the above theorem that for a path connected
space X, the fundamental group π1 (X, x) is independent of the base point x
up to isomorphism of groups. Therefore, for a path connected space X, we can
denote π1 (X, x) simply by π1 (X) ignoring the mention of the base point x, and
call it the fundamental group of the space X.
Since the isomorphism Pω depends on the path ω joining x0 with x1 , we

can examine the question as to how much Pω depends on the path ω itself. We
have,
Proposition 2.4.17. If ω, ω are two paths joining x0 to x1 which are path ho-
motopic, then the induced isomorphisms Pω and Pω are identical.
Proof. If ω, ω are path homotopic, then it is easily seen that ω −1 and (ω )−1
are also path homotopic. It follows that for any loop α based at x0 , ω −1 ∗ α is
path homotopic to (ω )−1 ∗ α and therefore ω −1 ∗ α ∗ ω is path homotopic to
(ω )−1 ∗ α ∗ ω . This means Pω [α] = Pω [α].
Proposition 2.4.18. Let X be a path connected space and x0 , x1 ∈ X. Then

π1 (X, x0 ) is abelian if and only if for each pair of paths ω, ω from x0 to x1 ,
Pω = Pω .
Proof. Assume that π1 (X, x0 ) is abelian. Since ω ∗ (ω )−1 is a loop based at x0 ,
we observe that for each [α] ∈ π1 (X, x0 ),
[ω ∗ (ω )−1 ] ◦ [α] = [α] ◦ [ω ∗ (ω )−1 ],
which means
[(ω )−1 ∗ α ∗ ω ] = [ω −1 ∗ α ∗ ω].
Conversely, suppose [α], [β] are two elements of π1 (X, x0 ). Let ω be a path in
X joining x0 with x1 . Then β ∗ ω is also a path joining x0 with x1 . Hence, by
the given condition, Pβ∗ω [α] = Pω [α]. This means
[(β ∗ ω)−1 ∗ α ∗ (β ∗ ω)] = [ω −1 ∗ α ∗ ω].
Since (β ∗ ω)−1 = ω −1 ∗ β −1 , therefore, [β −1 ∗ α ∗ β] = [α], i.e.,
[α][β] = [β][α].
Note that to each pointed topological space (X, x), we have associated
its fundamental group π1 (X, x). Next, we show that for every continuous map
f : (X, x) → (Y, y) of pointed spaces, there is an induced group homomorphism
f# between their fundamental groups.
Theorem 2.4.19. Every continuous map f : (X, x) → (Y, y) of pointed spaces

induces a group homomorphism f# : π1 (X, x) → π1 (Y, y).
Proof. Let α be a loop in X based at x. Then f ◦α is a loop based at y. Moreover,
if α ∼x0 α , say, by a homotopy H, then one can easily see that f ◦ α ∼y f ◦ α
by the homotopy f ◦ H. Hence, we define a map f# : π1 (X, x) → π1 (Y, y) by
f# ([α]) = [f ◦ α].
To see that f# is a homomorphism, observe that for any two loops α, β based
at x, we have
(f ◦ (α ∗ β))(t) = f ((α ∗ β)(t))

f (α(2t)) 0 ≤ t ≤ 1/2
=
f (β(2t − 1)) 1/2 ≤ t ≤ 1

(f ◦ α)(2t) 0 ≤ t ≤ 1/2
=
(f ◦ β)(2t − 1) 1/2 ≤ t ≤ 1
= ((f ◦ α) ∗ (f ◦ β))(t)
for each t ∈ I, which means f ◦ (α ∗ β) = (f ◦ α) ∗ (f ◦ β). Hence
f# ([α] ◦ [β]) = f# [α ∗ β]
= [f ◦ (α ∗ β)]
= [(f ◦ α) ∗ (f ◦ β)]
= f# [α] ◦ f# [β].
The two basic properties of the induced homomorphism f# are given by

the next result. In the language of categories and functors these two proper-
ties are sometimes expressed by declaring that the fundamental group π1 is
a covariant functor from the category of all pointed topological spaces and
base point preserving continuous maps to the category of all groups and group
homomorphisms.
Theorem 2.4.20. (i) If f : X → X is the identity map, then f# : π1 (X, x) →

π1 (X, x) is also the identity map for all x ∈ X.
(ii) If f : (X, x) → (Y, y) and g : (Y, y) → (Z, z) are two maps of pointed
spaces, then
(g ◦ f )# = g# ◦ f# .
Proof. (i) If f is the identity map on X, then for each loop α based at x,
f ◦ α = α and so f# [α] = [f ◦ α] = [α]. Hence f# is also identity on π1 (X, x).
(ii) For any loop α based at x, we have
(g ◦ f )# [α] = [(g ◦ f ) ◦ α]
= [g ◦ (f ◦ α)]
= g# [f ◦ α]
= g# (f# [α])
= g# ◦ f# [α].
Hence, (g ◦ f )# = g# ◦ f# .
Just before stating the above theorem, we pointed out that the above two
properties of the induced homomorphisms are basic. The reason as to why they
are important is that they imply a deep result, viz., the fundamental group of
a path connected topological space is a topological invariant. In other words,
the above properties yield the fact that if X and Y are two path connected
topological spaces which are homeomorphic, then their fundamental groups
must be isomorphic. We have already explained the proof in the introduction.
However, we present the proof once more:
Theorem 2.4.21. If X and Y are path connected spaces which are homeomor-
phic, then their fundamental groups are isomorphic.
Proof. We have already proved in Theorem 2.4.15 that for path connected
spaces, the fundamental group is independent of the base point. Let f : X → Y
be a homeomorphism and x ∈ X. Then f : (X, x) → (Y, f (x)) is a homeomor-
phism of pointed spaces and its inverse f −1 : (Y, f (x)) → (X, x) is a continuous
map which has the property that
f −1 ◦ f = IX , f ◦ f −1 = IY .
Considering the induced homomorphisms in fundamental groups, we see that
(f −1 ◦ f )# = (IX )# .
However, by (i) of Theorem 2.4.20, (IX )# = Iπ1 (X,x) and by (ii) of the same
theorem,
(f −1 ◦ f )# = (f −1 )# ◦ f# .
Therefore,
(f# )−1 ◦ f# = Iπ1 (X,x).
By a similar argument, we find that
f# ◦ (f −1 )# = Iπ1 (Y,f (x)).
Combining the last two identities, we conclude that f# : π1 (X, x) → π1 (Y, f (x))
is an isomorphism.
Corollary 2.4.22. Let A ⊂ X. Suppose r : X → A is a retraction and

i : A → X is the inclusion. Then r# : π1 (X, a) → π1 (A, a) is onto and
i# : π1 (A, a) → π1 (X, a) is 1-1 for each a ∈ A.
i r
Proof. For each point a ∈ A, the composite map (A, a) → (X, a) → (A, a) is
identity on (A, a). Consequently, by the two basic properties of the induced
i# r#
homomorphisms, the composite π1 (A, a) −→ π1 (X, a) −→ π1 (A, a) is identity
on the group π1 (A, a). This implies at once that i# is 1-1 and r# is onto.
Given two topological spaces X and Y, we are very often required to

determine whether or not X and Y are homeomorphic. This is briefly known
as the “homeomorphism problem”. Similarly, to determine whether or not
two given groups are isomorphic, is abbreviated as “isomorphism problem”
for groups. Algebraic topologists agree that the latter problem is normally
easier than the former. The fundamental group functor partially converts a
homeomorphism problem into an isomorphism problem in the sense that if
we can assert that π1 (X) is not isomorphic to π1 (Y ), then it follows from
topological invariance property of fundamental groups that X and Y cannot
be homeomorphic. To quote another example, we will later on see that for
the projective plane RP2 , π1 (RP2 ) ∼
= Z2 and π1 (S1 × S1 ) ∼
= Z ⊕ Z. Hence,
by our discussion above, it follows that the projective plane RP2 cannot be
homeomorphic to the two dimensional torus S1 × S1 . Here one might recall
that both these spaces are compact, connected, two dimensional and even
locally homeomorphic. As a matter of fact, invariants of point set topology
do not appear adequate to distinguish these two spaces. The homeomorphism
problem is only partially converted into isomorphism problem because if the
fundamental groups of X and Y are isomorphic, we cannot say that X and
Y are homeomorphic. We will later on prove that the fundamental group of a
contractible space X is isomorphic to the fundamental group of a point space
P , but X and P are never homeomorphic unless X consists of a single point.
We have some further properties of induced homomorphisms.
Theorem 2.4.23. If f, g : (X, x) → (Y, y) are continuous maps of pointed spaces

which are homotopic relative to {x}, then
f# = g# : π1 (X, x) → π1 (Y, y).
Proof. Let α be a loop in X based at x. Since f is homotopic to g and composite
of homotopic maps is always homotopic, the loops f ◦ α and g ◦ α based at y are
homotopic. This is not enough because they may not be homotopic relative to
{0,1}. But, since f is homotopic to g relative to {x}, we see that f ◦ α ∼y g ◦ α,
i.e., [f ◦ α] = [g ◦ α]. This means
f# [α] = g# [α], ∀ [α] ∈ π1 (X, x).
Using Theorem 2.4.23, we can extend Theorem 2.4.20 as follows:
Theorem 2.4.24. If f : (X, x0 ) → (Y, y0 ) is a homotopy equivalence in the rela-

tive sense, then f# : π1 (X, x0 ) → π1 (Y, y0 ) is an isomorphism of groups.
Proof. Recall that two continuous maps k, h : (X, x0 ) → (Y, y0 ) of pointed
spaces are said to be homotopic in the relative sense if there exists a homotopy
H : X × I → Y such that
H(x, 0) = k(x), H(x, 1) = h(x), ∀x ∈ X
H(x0 , t) = k(x0 ) = h(x0 ), ∀t ∈ I.
Thus, f : (X, x0 ) → (Y, y0 ) is a homotopy equivalence in the relative sense
means there exists a map g : (Y, y0 ) → (X, x0 ) of pointed spaces such that g◦f ∼
I(X,x0 ) and f ◦ g ∼ I(Y,y0 ) in the relative sense. By the two basic properties
of induced homomorphisms, we note that homomorphisms f# : π1 (X, x0 ) →
π1 (Y, y0 ) and g# : π1 (Y, y0 ) → π1 (X, x0 ) are inverse of each other. Hence, f#
is an isomorphism.
Corollary 2.4.25. Let A be a strong deformation retract of a space X. Then for

each point a ∈ A, π1 (A, a) is isomorphic to π1 (X, a).
Proof. Since X is deformable into A strongly, we find that for each a ∈ A, the
inclusion map i : (A, a) → (X, a) is a homotopy equivalence relative to {a}.
Hence, by the above theorem, i# : π1 (A, a) → π1 (X, a) is an isomorphism.
Remark 2.4.26. If the subspace A of X is a singleton, i.e., A = {a} and A is

strong deformation retract of X, then π1 (X, a) ∼ = π1 ({a}, a). We usually write
this as π1 (a) ignoring the base point. We also note that if {a} is a strong de-
formation retract of X, then X is contractible to {a} and so is path connected.
Therefore, π1 (X, x) ∼= π1 (X, a), ∀x ∈ X. Thus, π1 (X) ∼
= π1 (P ) = {e}, where P
is a point space.
We now proceed to prove our final result, viz., if two path-connected spaces
X and Y have the same homotopy type, then their fundamental groups are
isomorphic. In particular, we show that the fundamental group of a contractible
space X is trivial. First, we have
Theorem 2.4.27. Let F be a homotopy between two maps f, g : X → Y . Let

x0 ∈ X and σ : I → Y be the path joining f (x0 ) and g(x0 ) defined by σ(t) =
F (x0 , t). Then the triangle
f#
π1 (X, x0 ) - π1 (Y, f (x0 ))
HH
HH Pσ
g#
HH
j ?
π1 (Y, g(x0 ))
of induced homomorphisms is commutative.
Proof. Let α be a loop in X based at x0 . Then we know that
Pσ ◦ f# [α] = Pσ [f ◦ α] = [σ −1 ∗ (f ◦ α) ∗ σ]
and
g# [α] = [g ◦ α].
Hence, it suffices to show that the two loops σ −1 ∗ (f ◦ α) ∗ σ and g ◦ α based
at g(x0 ) are equivalent. Define a continuous map G : I × I → Y by G(s, t) =
F (α(s), t).
go α
σ σ H
Y
fo α
Fig. 2.15
Then (Fig. 2.15) for each s ∈ I, t ∈ I, we have
G(s, 0) = F (α(s), 0) = f ◦ α(s),
G(s, 1) = F (α(s), 1) = g ◦ α(s),
G(0, t) = F (α(0), t) = F (x0 , t) = σ(t),
G(1, t) = F (α(1), t) = F (x0 , t) = σ(t).
Our objective is now to construct a homotopy H : I × I → Y which starts with
the loop σ −1 ∗ (f ◦ α) ∗ σ, terminates with the loop g ◦ α and keeps the end
points fixed.
The required homotopy H is defined by
⎧ −1
⎨ σ (2s) 0 ≤ s ≤ (1 − t)/2
H(s, t) = G((4s + 2t − 2)/(3t + 1), t) (1 − t)/2 ≤ s ≤ (t + 3)/4
⎩
σ(4s − 3) (t + 3)/4 ≤ s ≤ 1.
goα
g(x0) g(x0) H
Y
−1
σ fo α σ
Fig. 2.16
It is easily verified that H is continuous and has the desired properties

(Fig. 2.16).
Corollary 2.4.28. If f : X → Y is homotopic to a constant map C : X → Y ,

then the induced homomorphism f# : π1 (X, x0 ) → π1 (Y, f (x0 )) is the zero map.
Proof. In the commutative triangle, notice that Pσ is an isomorphism and
C# : π1 (X, x0 ) → π1 (Y, C(x0 )) is a trivial map.
In case f (x0 ) = g(x0 ), the path σ is just a loop at g(x0 ) = f (x0 ) and the
automorphism Pσ : π1 (Y, f (x0 )) → π1 (Y, f (x0 )) is simply the inner automor-
phism
[α] → [σ −1 ∗ α ∗ σ] = [σ]−1 ◦ [α] ◦ [σ].
Moreover, if the homotopy F is relative to the point subspace {x0 }, then σ is
just the constant loop Cf (x0 ) and so Pσ is the identity map, i.e., f# = g# and
our result reduces to Theorem 2.4.23.
Remark 2.4.29. It is interesting to observe that if the space X is contractible

to a point x0 relative to {x0 } by a homotopy F : X × I → X, i.e., {x0 } is
a strong deformation retract of X, then any loop α based at x0 is obviously
homotopic to the constant loop Cx0 by a homotopy (s, t) → F (α(s), t) which
does not move the point x0 at all. However, if X is contractible to a point x0 ,
then any loop α based at x0 will be homotopic to the constant loop Cx0 , but as
in the case of comb space, the contracting homotopy mentioned above may not
necessarily fix the point x0 and so the loop α, it appears, may not represent the
zero element of π1 (X, x0 ). The last statement, however, is not possible. In fact,
we now show that if X is a contractible space, then for any given loop based
at a point x0 ∈ X, we can always find a homotopy which deforms the loop into
the constant loop Cx0 without moving the point x0 at all, i.e., π1 (X, x0 ) = 0.
We have
Corollary 2.4.30. Let X be a contractible space. Then each loop α based at any
point x0 ∈ X is equivalent to the constant loop Cx0 at x0 , i.e., π1 (X, x0 ) = 0.
Proof. Since X is contractible, the identity map IX is homotopic to the constant
map Cx0 by a homotopy F . Hence, if σ is the loop at x0 defined by
σ(t) = F (x0 , t),
then by Theorem 2.4.27, the triangle
(IX )#
π1 (X, x0 ) - π1 (X, x0 )
HH
H
(Cx )H
# H
Pσ
0
j
H ?
π1 (X, x0 )
is commutative, i.e.,
Pσ ◦ (IX )# [α] = (Cx0 )# [α],
for each [α] ∈ π1 (X, x0 ). But this means [σ −1 ∗α∗σ] = [Cx0 ], i.e., σ −1 ∗α∗σ ∼x0
Cx0 . This implies at once that
α ∼x0 σ ∗ Cx0 ∗ σ −1 ∼x0 Cx0 .
Now, we have our main
Theorem 2.4.31. Let X, Y be two path-connected spaces which are of the same
homotopy type. Then their fundamental groups are isomorphic.
Proof. The fundamental groups of both the spaces X and Y are independent
of the base points because each of them is path connected. Since X and Y
are of the same homotopy type, there exist continuous maps f : X → Y and
g : Y → X such that g ◦ f ∼ IX by a homotopy, say, F and f ◦ g ∼ IY by some
homotopy, say G. Let x0 ∈ X be a base point. Let
f# : π1 (X, x0 ) → π1 (Y, f (x0 ))
and
g# : π1 (Y, f (x0 )) → π1 (X, g(f (x0 )))
be the induced homomorphisms. It is evidently enough to prove that f# is a
bijective map. Let σ be the path joining x0 to gf (x0 ) defined by the homotopy
F . Then by Theorem 2.4.27, the triangle
(IX )#
π1 (X, x0 ) - π1 (X, x0 )
HH
H
(g◦f )H
# H
Pσ
j ?
H
π1 (X, gf (x0 ))
is commutative, i.e., (g ◦ f )# = Pσ ◦ (IX )# and so g# ◦ f# = Pσ . Since Pσ is

an isomorphism, we find that f# is a 1-1 map. Next, let τ be the path joining
f (x0 ) to (f ◦ g)(f (x0 )) defined by the homotopy G. Then again the triangle
2.5. Simply-Connected Spaces 67
(IY )#
π1 (Y, f (x0 )) - π1 (Y, f (x0 ))
HH
H
(f ◦g)H
# H
Pτ
j ?
H
π1 (Y, (f ◦ g)(f (x0 )))
is commutative, i.e., f# ◦ g# = Pτ . Since Pτ is an isomorphism, we conclude

that f# is also onto.
Corollary 2.4.32. Let X, Y be two path-connected spaces and f : X → Y be a

homotopy equivalence. Then f# : π1 (X) → π1 (Y ) is an isomorphism.
2.5 Simply-Connected Spaces
We have already seen that if X is a contractible space then π1 (X) is

always trivial. In fact, those spaces whose fundamental groups are trivial play
very important role in complex analysis as well as in algebraic topology. We
have
Definition 2.5.1. A space X is said to be simply connected ( or 1-connected ) if

it is path connected and π1 (X) = 0.
Notice that X being path connected, the fundamental group of X is in-

dependent of the base point. Consequently, X is simply connected if and only
if it is path connected and its fundamental group vanishes at some point of X
(hence it vanishes at every point). Corollary 2.4.30 asserts that a contractible
space is simply connected. Later we will give example to show that the converse
is not true. However, let us have
Proposition 2.5.2. A path-connected space X is simply connected if and only if

any two paths in X having the same endpoints are path homotopic.
Proof. Suppose α, β are two paths in X, joining x0 with x1 . Notice that α ∗ β −1
is a loop based at x0 . Since X is simply connected, this loop must be null
homotopic relative to x0 , i.e.,
[α ∗ β −1 ] = [Cx0 ].
Now, by Remark 2.4.12,
[(α ∗ β −1 ) ∗ β] = [Cx0 ∗ β] = [β]
and
[α ∗ (β −1 ∗ β)] = [α ∗ Cx1 ] = [α].
Hence, by the associative law of product of path homotopy classes, we find that
[α] = [β], i.e., α is path homotopic to β.
The converse part is trivial.
If a space X is expressed as the union of two simply-connected spaces,

say, X1 , X2 , then can we say that X itself is simply connected? The answer
is “no”: we shall see later that the unit circle S1 in the complex plane is not
simply connected, though S1 can be expressed as the union of two contractible
arcs. However, the next result is an important one.
Theorem 2.5.3. Let {Vi : i ∈ Λ} be an open covering of a space X, where each

Vi is simply connected. Then X itself is simply connected provided

(1) Vi
= φ
(2) for each i
= j, Vi ∩ Vj is path connected.
Proof. Since each of the open sets Vi is path connected and their intersection
is non-empty, the space X itself is path connected. Hence it suffices to show
that π1 (X, x) is trivial for some x ∈ X. We choose a point x0 ∈ ∩i Vi .
p
3 3 p
2
x0 p
1 1
Fig. 2.17: The path α is null-homotopic (case n = 3)
Let α : I → X be a loop based at x0 . We will show that [α] is the zero element
of π1 (X, x0 ). Note that {α−1 (Vi )} is an open cover of I. Since I is a compact
metric space, this open cover will have a Lebesgue number, say, > 0. This
means there will be a partition
0 = t0 < t 1 < t 2 < · · · < t n = 1
of I such that for 0 ≤ j ≤ n − 1, α([tj , tj+1 ]) will be contained in some Vi .
For convenience, let us rename the open sets Vi so that α([tj , tj+1 ]) ⊂ Vj ,
0 ≤ j ≤ n − 1. For each j, we can define a path αj in X by
αj (s) = α((1 − s)tj + stj+1 ).
Then, αj (I) is contained in the simply-connected open set Vj for each j, and
[α] = [α0 ∗ α1 ∗ · · · ∗ αn−1 ].
Look at the Fig. 2.17. Notice that α(t1 ) ∈ V0 ∩ V1 and V0 ∩ V1 is path connected
containing the base point x0 . Hence we can find a path ρ1 : I → X from x0 to
α(t1 ) such that ρ1 (I) ⊂ V0 ∩ V1 . By a similar argument, we can find a path ρj
from x0 to α(tj ) lying completely in Vj−1 ∩ Vj for j = 1, 2, . . . , n − 1. If ρ−1
j
denotes the reverse path of ρj , then we have
2.5. Simply-Connected Spaces 69
[α] = [α0 ∗ ρ−1 −1

1 ∗ ρ1 ∗ α1 ∗ ρ2 ∗ ρ2 ∗ α2 ∗ · · · ∗ ρn−1 ∗ αn−1 ].
Note that ρ−11 ∗ ρ1 is path homotopic to the constant loop based at α(t1 ), etc.
It follows that
[α] = [α0 ∗ ρ−1 −1
1 ] ◦ [ρ1 ∗ α1 ∗ ρ2 ] ◦ · · · ◦ [ρn−1 ∗ αn−1 ].
The first term in the right hand side above is a loop based at x0 and lying
completely in the simply-connected space V0 . Similarly, the second term is a
loop lying completely in the simply connected space V1 , and so on. Conse-
quently, each term is actually null homotopic in Vj for some j and hence in X.
Therefore, [α] represents the zero element of π1 (X, x0 ).
Example 2.5.4. The standard n-sphere Sn , n ≥ 2, is simply connected. For

simplicity, consider the case of a 2-sphere S2 = {(x, y, z) ∈ R3 | x2 +y 2 +z 2 = 1}.
Let N = (0, 0, 1) be the north pole of S2 , M = (0, 0, −1) be the south pole of
S2 . Then, U = S2 − {N }, V = S2 − {M } are two open sets of S2 which cover
the sphere S2 . From the results of point-set topology, one can see that
(i) each of U and V are homeomorphic to the Euclidean plane R2 , and so is
contractible. Hence these are simply connected.
(ii) U ∩ V is path connected; it follows from the fact that even if a countable
number of points are removed from S2 (here only two points have been
removed), the remaining space is still path connected.
Hence, it follows from the previous theorem that S2 is simply connected. Ex-
actly the same procedure shows that Sn is simply connected for each n ≥ 2.
We will see later on that the 1-sphere S1 is not simply connected. In fact,
we will prove that π1 (S1 ) ∼
= Z, the additive group of integers.
Exercises
1. Show that union of two contractible spaces, having nonempty path-

connected intersection, need not be contractible.
2. If P is a path component of X and x0 ∈ P , then the inclusion map
i : P → X can be regarded as a map of pointed spaces (P, x0 ) → (X, x0 ).
Prove that the induced homomorphism
i# : π1 (P, x0 ) → π1 (X, x0 )
is an isomorphism.
3. Prove that a cone C(X) over a topological space X is simply connected.
If vi is the vertex of a cone C(Xi ), i ∈ Λ, then show that their one-point
union, viz., their disjoint union with all vertices vi identified to one point,
is simply connected.
4. Let α, β be two paths in X with initial point x1 and terminal point x2 .

Prove that α ∼ β relative to {0,1} if and only if the loop αβ −1 ∼ Cx1
relative to {0,1}.
5. Let A ⊂ X be a deformation retract. Show that the inclusion map i : A →
X induces an isomorphism
i# : π1 (A, a) → π1 (X, a)
for each a ∈ A.
6. Consider two spheres Sm and Sn , m, n ≥ 2 imbedded in some Euclidean
space Rk , k sufficiently large. If Sm intersects Sn tangentially in a single
point, then show that their union is simply connected.
2.6 Results for Computing Fundamental Groups

In this section, we prove a few results which would be of help in computing
the fundamental groups of some familiar subspaces of the Euclidean space Rn ,
etc. Most importantly, we will prove in the next section that the fundamental
group π1 (S1 ) of the circle S1 is isomorphic to the additive group Z of integers.
Then, combining that result with those of this section, we will determine the
fundamental group of a number of familiar spaces.
Example 2.6.1. Let Dn (n ≥ 1) be the n-dimensional unit disk in Rn . Since

Dn is path connected, we can compute its fundamental group at any point.
For simplicity, let us choose origin as its base point. Note that the homotopy
H : Dn × I → Dn defined by
H(x, t) = (1 − t)x
shrinks every point of Dn to the origin without moving the origin at all. There-
fore, each loop in Dn based at origin is path homotopic to the constant loop.
Whence, π1 (Dn ) = 0.
More generally, if S is any convex subset of Rn and s0 ∈ S is a fixed point,

then the homotopy D defined by
D(x, t) = (1 − t)x + ts0
shrinks each point of S to the point s0 without moving the point s0 . Therefore,
each loop in S based at s0 is null homotopic, and π1 (S) = {e}. Because the
fundamental group of a contractible space is trivial, it follows directly that
the fundamental group of any star-like subspace of Rn (convex subsets, in
particular) is trivial.
Next, we proceed to give examples of spaces which have non-trivial fun-

damental groups. However, before we do so, let us prove the next useful result.
2.6. Results for Computing Fundamental Groups 71
Theorem 2.6.2. Let X, Y be two spaces with base points x0 ∈ X and y0 ∈ Y,

respectively. Then
π1 (X × Y, (x0 , y0 )) ∼
= π1 (X, x0 ) × π1 (Y, y0 ),
i.e., the fundamental group of the product space is isomorphic to the product of
fundamental groups of factors.
Proof. Suppose α : I → X ×Y is a loop based at (x0 , y0 ) and p1 : (X ×Y ) → X,
p2 : (X ×Y ) → Y are the canonical projections. Then p1 , p2 are obviously maps
of pointed spaces and hence induce group homomorphisms
p1# : π1 (X × Y, (x0 , y0 )) → π1 (X, x0 )
and
p2# : π1 (X × Y, (x0 , y0 )) → π1 (Y, y0 ).
We recall that, by definitions, p1# [α] = [p1 ◦ α] and p2# [α] = [p2 ◦ α]. Therefore,
from the definition of direct product of groups, we get a group homomorphism
ρ : π1 (X × Y, (x0 , y0 )) → π1 (X, x0 ) × π1 (Y, y0 )
defined by
ρ([α]) = ([ρ1 ◦ α], [ρ2 ◦ α]).
Now, it suffices to show that ρ is a bijection. For this, first note that given
a pair of continuous maps f : I → X, g : I → Y , there is, by definition of
product topology, a continuous map (f, g) : I → X × Y defined by (f, g)(t) =
(f (t), g(t)). Conversely, any continuous map h : I → X × Y defines a pair of
continuous maps ρ1 ◦ h : I → X, p2 ◦ h : I → Y . In fact, it is evident that this
correspondence between the set of all continuous maps h : I → X × Y and the
set of pairs (p1 ◦ h, p2 ◦ h) of continuous maps is a 1-1 correspondence. Now,
let ([α], [β]) ∈ π1 (X, x0 ) × π1 (Y, y0 ). Then the pair (α, β) of loops based at x0
and y0 respectively corresponds to the loop (α, β) : t → (α(t), β(t)) in X × Y
based at (x0 , y0 ) and defines a map
η : π1 (X, x0 ) × π1 (Y, y0 ) → π1 (X × Y, (x0 , y0 ))
given by
η([α], [β]) = [(α, β)].
To see that η is well defined, note that if F1 : α1 ∼x0 α2 , F2 : β1 ∼y0 β2 are
homotopies, then the map F : I × I → X × Y defined by
F (s, t) = (F1 (s, t), F2 (s, t))
is a homotopy from (α1 , β1 ) to (α2 , β2 ) relative to {0,1}. Now, it is clear from
definitions of ρ and η that they are inverses of each other which means ρ is a
bijection.
Note: Let {Xα : α ∈ Λ} be a family of spaces, and for each α ∈ Λ, let xα ∈ Xα

be a base point. Then the above proof easily generalizes to show that

π1 ( Xα , (xα )) ∼
= π1 (Xα , xα ).
The Fundamental Group of Circle
Recall that the unit circle S1 = {(x, y) ∈ R2 | x2 + y 2 = 1} is a

subspace of R2 . It can also be described in terms of complex numbers, viz.,
S1 = {z ∈ C : |z| = 1}. The description in terms of complex numbers at once
implies that S1 is also a topological group under the multiplication induced
by multiplication of complex numbers. We will choose (1, 0) ∈ S1 as the base
point and denote this point, hence onwards, by p0 . As remarked earlier, we
will finally prove that the fundamental group π1 (S1 , p0 ) is isomorphic to the
additive group Z of integers. A brief sketch of the proof runs as follows: We use
a map p : R → S1 , called the exponential map, defined by p(t) = e2πit , t ∈ R.
This map has the nice property that each path α : I → S1 with initial point
p0 ∈ S1 can be lifted to a unique path α : I → R such that α (0) = 0 ∈ R.
If α is a loop at p0 then α (1) must be an integer and this integer is called
the degree of the loop α. We show that if α ∼p0 β, then deg α = deg β. This
yields a map f : π1 (S1 , p0 ) → Z defined by f ([α]) = deg α. We then complete
the proof of our main result by showing that f is indeed an isomorphism.
Each of the steps outlined in the above sketch will be an interesting result
in its own right and we now proceed to furnish their complete details. The
exponential map p : R → S1 defined by p(t) = e2πit , t ∈ R is in fact a continuous
onto map which simply wraps the real line R onto the circle S1 infinite number
of times (Fig. 2.18).
2 `6 '$
6 S1
1`
0`
p- -
-1 `
eiπ p0 =(1,0)
-2 ` &%
?
Fig. 2.18
In fact, the interval [0,1] gets wrapped counterclockwise around the circle once
with p(0) = p(1) = p0 and the interval [−1, 0] also gets wrapped once but
clockwise, and so on. The map, p|(−1/2,1/2) is a homeomorphism from the open
interval (−1/2, 1/2) onto S1 − {eiπ }. We let ln : S1 − {eiπ } → (−1/2, 1/2) be
the inverse of this homeomorphism. Also, note that p(t1 + t2 ) = p(t1 ) · p(t2 ),
t1 , t2 ∈ R and p(t1 ) = p(t2 ) if and only if t1 − t2 is an integer. It follows that
p−1 (p0 ) = Z.
Definition 2.6.3. A continuous map f : X → R is called a lift or covering of

a given continuous map f : X → S1 provided the triangle is commutative, i.e.,
p ◦ f = f.
* R

f
?
p
- S1
X f
The following lemma says that any two lifts of a given map f : X → S1 ,
which agree at a single point of the connected space X, agree everywhere.
Lemma 2.6.4. Let X be a connected space and f , f : X → R be continuous

maps such that p◦f = p◦f . If f , f agree at one point x0 ∈ X, then f = f .
Proof. Since R has the additive group structure, we can define a map
g : X → R by g = f − f . Because p ◦ f = p ◦ f , we observe that for
each x ∈ X,
p ◦ g(x) = e2πig(x)

(x)−f (x)
= e2πi(f

= e2πif (x) /e2πif (x)
= p ◦ f (x)/p ◦ f (x)
= 1.
This means g(x) ∈ p−1 (1) = Z. Since X is connected and g is continuous, the
image set g(X) must be a singleton. But g(x0 ) = 0 implies that g(x) = 0, for
all x ∈ X, i.e., f (x) = f (x), for all x ∈ X.
The next two theorems are extremely basic.
Theorem 2.6.5. (The Path Lifting Property). Let p : R → S1 be the exponential

map and α : I → S1 be any path such that α(0) = p0 . Then there exists a unique
lift α of α such that α (0) = 0.
Theorem 2.6.6. (The Homotopy Lifting Theorem). Let p : R → S1 be the ex-

ponential map. Suppose F : I × I → S1 is a homotopy such that F (0, 0) = p0 .
Then there exists a unique lift F : I × I → R of F such that F (0, 0) = 0.
Theorem 2.6.5 says that any path in S1 starting at p0 can be lifted to a

unique path in R starting at the origin of R. Similarly, Theorem 2.6.6 asserts
that any homotopy between two given paths in S1 starting at p0 , can be lifted
to a unique homotopy between the two lifted paths starting at the origin of R.
We are now going to prove a general result of which the above two theorems
are obvious corollaries.
Theorem 2.6.7. Let X be a convex compact subset of the Euclidean space Rn

and x0 ∈ X. Given any continuous map f : X → S1 and any t0 ∈ R such
that p(t0 ) = f (x0 ), there exists a unique continuous map f : X → R such that
f (x0 ) = t0 and p ◦ f = f .
Proof. Note that a convex subset of Rn is always connected and so, if f , f are
two lifts of f which agree at the point x0 ∈ X, then it follows from Lemma 2.6.4
that f = f . This proves the uniqueness of the lift f satisfying the condition
that f (x0 ) = t0 .
To prove the existence of a lift f of f satisfying the condition that

f (x0 ) = t0 , we point out that the point x0 can be taken to be the origin of Rn
without any loss of generality. For, if x0 is not the origin, then we consider the
subspace L(X) of Rn where L : Rn → Rn is the translation map y → y − x0
which is clearly a homeomorphism preserving compactness and convexity of X.
The subspace L(X) now contains the origin of Rn . If the existence of a lift for
L(X) is now assumed, then let k : L(X) → R be a lift of f ◦ L−1 such that
k(0) = t0 . Here, we are treating L as a map from X onto L(X) only.
X R
PP f k1

PP
L PPP p
? q ?
P
L(X) - S1
f ◦L−1
Then the composite map k ◦ L : X → R maps x0 to t0 and has the property

that
p ◦ (k ◦ L) = (p ◦ k) ◦ L = (f ◦ L−1 ) ◦ L = f.
Hence we now assume that X is a compact convex subset of Rn containing
the origin. Since X is compact, the continuous map f : X → S1 is uniformly
continuous and so there is an > 0 such that if x−x < then |f (x)−f (x )| <
2. Here, · is the Euclidean norm of Rn and |f (x)| denotes the modulus of
the complex number f (x). Since X is bounded, we can find an integer n such
that x/n < for all x ∈ X. This means for each j = 0, 1, 2, . . . , n − 1, and all
x∈X
(j + 1)x jx
− = x/n < .
n n
Therefore,
|f ( (j+1)x
n ) − f ( jx
n )| < 2.

It follows that the quotient f (j+1)xn n ) is a point in S − {e }. Let
/f ( jx 1 iπ
gj : X → S1 − {eiπ } for 0 ≤ j ≤ n − 1 be the maps defined by
gj (x) = f ( (j+1)x
n )/f ( jx
n ).
Then for all x ∈ X, we can write f (x) as the product

f (x) = f (0) · g0 (x).g1 (x) · · · gn−1 (x).
Let us define f : X → R by
f (x) = t0 + ln g0 (x) + ln g1 (x) + · · · + ln gn−1 (x).

Then f , being the sum of n continuous functions, is continuous and has the
property that f (0) = t0 , p ◦ f = f .
Remark 2.6.8. When we consider X in the above theorem to be the unit interval
I = [0, 1] ⊂ R, we obtain the Path Lifting Property (Theorem 2.6.5) for the
exponential map p : R → S1 . On the other hand, if we take the unit square
I × I ⊂ R2 as X, we obtain the Homotopy Lifting Property for the exponential
map recorded as Theorem 2.6.6.
Now, let α : I → S1 be a loop in S1 based at the point p0 = (1, 0) ∈ S1 .

Let α : I → R be the lift of α such that α (0) = 0. Such a lift is guaranteed
by the path lifting property of the exponential map. Since α (1) ∈ R projects
to the point (1,0) under the exponential map, it must be an integer. We call
this integer α (1) to be the degree of the loop α. Since there is only one lift
of α starting at the origin of R, α (1) depends only on the loop α. If α, β are
two equivalent loops in S1 based at p0 , then how are their degrees related? The
next result answers this question.
Theorem 2.6.9. Let α, β be two equivalent loops in S1 based at p0 . Then degree

α = degree β.
Proof. Since α ∼p0 β, there is a homotopy F : I × I → S1 such that F (s, 0) =
α(s), F (s, 1) = β(s), F (0, t) = p0 = F (1, t), for all s, t ∈ I. By the Homotopy
Lifting Property for the exponential map p, there exists a homotopy F : I ×I →
R such that F (0, 0) = 0 and p ◦ F = F (Fig. 2.19).
6
`0 R
F

1
I ×I ?
PP '$
PP
F q
P ` p0
(0,0) &%
S1
Fig. 2.19
The paths α , β : I → R defined by α (s) = F (s, 0), β (s) = F (s, 1) are

evidently lifts of α and β respectively. Now, α (0) = F (0, 0) = 0, and
p ◦ F (0, t) = F (0, t) = p0 . It follows that F (0, t) is some integer for all t ∈ I.
Since I is connected and F (0, 0) = 0, we find that F (0, t) = 0, ∀t ∈ I.
In particular, F (0, 1) = 0 = β (0), i.e., α (0) = β (0) = 0. Hence, degree
α = α (1), degree β = β (1). Since F maps the vertical line {(1, t) | t ∈ I} of
I × I into the point p0 , F must map this line into a subset of p−1 (p0 ) = Z.
Moreover, since vertical line is connected, F is continuous and Z is discrete,

we find that F (1, t) is some integer, for all t ∈ I. This means degree
α = α (1) = F (1, 0) = F (1, 1) = β (1) = degree β.
The above theorem tells us that if we define a map deg : π1 (S1 , p0 ) → Z

by deg([α]) = degree α, then deg is a well-defined map. We call this map as
the degree map. Now, we will prove our final step of the main theorem.
Theorem 2.6.10. The degree map deg : π1 (S1 , p0 ) → Z is an isomorphism from

the fundamental group of circle to the additive group Z of integers.
Proof. The degree map is a homomorphism: Let [α], [β] be any two elements
of π1 (S1 , p0 ). Let α , β : I → R be the unique paths starting from origin of R
which are lifts of α and β, respectively. Define a new path ω : I → R by

α (2t) 0 ≤ t ≤ 1/2
ω(t) =
α (1) + β (2t − 1) 1/2 ≤ t ≤ 1.
Then obviously ω also begins at origin and

p ◦ α (2t) 0 ≤ t ≤ 1/2
p ◦ ω(t) =
p(α (1)) · (p ◦ β )(2t − 1) 1/2 ≤ t ≤ 1

α(2t) 0 ≤ t ≤ 1/2
=
β(2t − 1) 1/2 ≤ t ≤ 1
= (α ∗ β)(t) ∀t ∈ I.
Thus, ω is a lift of α ∗ β starting from the origin. Consequently,
deg([α] ∗ [β]) = deg([α ∗ β])

= ω(1)
= α (1) + β (1)
= deg([α]) + deg([β]).
The degree map is 1-1: Suppose [α], [β] ∈ π1 (S1 , p0 ) such that deg([α]) =
deg([β]). This means if α : I → R, β : I → R are lifts of α, β respectively
starting from origin, then α (1) = β (1). Define a homotopy H : I × I → R by
H(s, t) = (1 − t)α (s) + tβ (s), (s, t) ∈ I × I.
Then H is a homotopy between α and β relative to {0,1}. It follows that

p ◦ H : I × I → S1 is a homotopy between α and β relative to {0,1}, i.e.,
[α] = [β].
The degree map is onto: Let n ∈ Z. Define the loop γ : I → S1 by
γ(t) = e2πint
The path γ : I → R defined by γ (t) = nt, starts at the origin of R and lifts
the path γ. Therefore, deg([γ]) = γ (1) = n. Thus, we have proved that the
degree map is an isomorphism.
The above theorem is the first nice result giving a nontrivial example of
the fundamental group, viz., the fundamental group of the circle is isomorphic
to the additive group Z of integers. As we shall see next, all other examples of
fundamental groups that we discuss at this stage are based upon this example
and the results proved earlier.
Fig. 2.20: Any closed curve without self-intersection
Remark 2.6.11. We have already proved (Theorem 2.4.21) that the fundamental
groups of two homeomorphic spaces are isomorphic. Consequently, any space
which is homeomorphic to the unit circle S1 will have its fundamental group
as the additive group Z of integers. For example all closed curves which have
the subspace topology induced from R2 and, as shown above (Fig. 2.20), have
Z as their fundamental groups. In fact, take a pencil, put it at a point p in the
plane R2 , move your hand tracing some curve in any manner you like, but do
not lift your hand and do not cross the curve already traced anywhere, finally
come back to the point p. Then the curve so described is homeomorphic to the
circle S1 and will have its fundamental group isomorphic to the additive group
Z of integers.
One must take time and be convinced as to why, each of these spaces
is homeomorphic to the unit circle S1 . We also note that if we remove
just one point from any of the closed curves, then the remaining space is
homeomorphic to the open interval (0,1) and so is contractible. In other words,
the fundamental group of the remaining space is trivial. Each of this fact is
interesting, but far more interesting is the following: if two path-connected
spaces are not homeomorphic but, more generally, are of the same homotopy
type, even then they will have isomorphic fundamental groups (Theorem
2.4.31). We now consider results of this kind.
Fundamental Group of the Punctured Plane. By the punctured plane X, we

mean a topological space which is homeomorphic to the plane R2 with some
point p ∈ R2 deleted, i.e., X = R2 − {p}. It is, in fact, trivial to see that X
is homeomorphic to the punctured plane Y = R2 − (0, 0). Now, we point out
that the unit circle S1 ⊂ R2 − {(0, 0)} is a strong deformation retract of Y . The
homotopy F : Y × I → Y defined by
F (x, t) = (1 − t)x + t x
x
y
6
'$

* -x
&%
Fig. 2.21
C1
X C Y
2
Fig. 2.22(a) and 2.22(b)
is a strong deformation retraction (Fig. 2.21). This means X is of the same

homotopy type as the unit circle S1 . Therefore, π1 (X) = (Z, +).
If we consider an annulus X (Fig. 2.22(a)) surrounded by two concentric
circles C1 and C2 , then any concentric circle C lying in the annulus is a strong
deformation retract of X and hence, π1 (X) = Z.
One can give a strong deformation retraction as follows: From the centre
O draw a half line in each direction. Each such line intersects the circle C in a
unique point, say, r(x). Let r : X → C be the mapping which maps all points
of the half line to the corresponding point r(x). Now, define H : X × I → X
by H(x, t) = (1 − t)x + t · r(x). Note that the space Y of Fig. 2.22(b) is not
exactly an annulus, but has circle as its deformation retract. Hence, π1 (Y ) = Z.
Fundamental Group of Cylinder. Recall that a topological space X is said to

be cylinder if it is homeomorphic to the subspace (see Fig. 2.23)
Y = {(x, y, z) ∈ R3 : x2 + y 2 = 1, −k ≤ z ≤ k, k ∈ R}.
Let C = {(x, y, 0) ∈ R3 : x2 + y 2 = 1}. Then C is a strong deformation retract

of X under the homotopy F ((x, y, z), t) = (x, y, (1−t)z) and hence, π1 (X) = Z.
Fundamental Group of Torus. We know that a torus T (a surface) is homeo-

morphic to the product S1 × S1 of two circles S1 , whereas the n-dimensional
y
x
Fig. 2.23: Circle is a strong deformation retract
torus T n ∼= S1 × S1 × · · · × S1 (n copies). Each of them is a path-connected

space. It follows, therefore, that π1 (T ) ∼
= Z × Z and π1 (T n ) ∼
= Z × Z × ··· × Z
(n copies ). In the case of torus, two loops α, β based at a point (a, b) are the
generators of π1 (T ) (see Fig. 4.7).
The next question is about the fundamental groups of other compact sur-
faces such as the Möbius band, the projective plane, the Klein bottle, etc., which
we have not discussed so far. More generally, if we take an arbitrary compact
surface X, then it is well-known that X is homeomorphic to a connected sum
of tori or a connected sum of projective planes (see Section 1.1.7). Therefore,
the general problem is to determine the fundamental groups of all compact
surfaces, and this can be done. In order to answer these questions, however,
we first need to determine the fundamental groups of connected graphs, which
are 1-dimensional simplicial polyhedra. The solution is provided, among other
things, by the following powerful result (see W.Massey [13] for a proof and other
details), which gives the fundamental group of the union of two subspaces if we
know the fundamental groups of individual subspaces. This result also extends
Theorem 2.5.3.
Theorem 2.6.12 (Seifert-Van Kampen Theorem). Let X = U1 ∪ U2 with U1 , U2

and U1 ∩ U2 all nonempty path-connected open subsets. Then π1 (X) is the
free product of π1 (U1 ) and π2 (U2 ) amalgamated along π1 (U1 ∩ U2 ) via homo-
morphisms induced by the inclusion maps. In particular, if U1 ∩ U2 is simply
connected, then π1 (X) is the free product of π1 (U ) and π2 (U ).
Example 2.6.13. If X = S1 ∨ S1 is the one-point union of two circles (figure of

eight), then it follows from the above theorem and the fact that π1 (S1 ) ∼ =Z
that π1 (X) is the free group on two generators. This follows from the Van
Kampen theorem and the fact that the figure of eight is the union of two circles
with whiskers (open sets) having contractible intersection. Similarly, if X is
a one-point union of three circles, then π1 (X) is free group on 3-generators, etc.
Fig. 2.24: Two circles with whiskers
The fundamental groups of projective planes, Klein bottle etc. are deter-
mined using Van Kampen theorem (see amongst the exercises given below).
Exercises
1. Show that if X is an indiscrete space, then π1 (X) = 0.

2. Suppose we remove the z-axis from R3 and denote the resulting space as
X. Compute π1 (X).
3. Let f : X → Sn be a map which is not onto. Prove that f is null homo-
topic. (Hint: The image of f is contained in an Euclidean space which is
contractible.)
4. Prove that any two maps f, g : X → C(Y ) where C(Y ) is the cone over
Y , are homotopic.
5. Prove that a map f : X → Y is null homotopic if and only if f can be
extended to a map fˆ: C(X) → Y .
6. Give an example of a space which is simply connected, but not con-
tractible.
7. Let p ∈ S1 where S1 is the unit circle. Prove that S1 × {p} is a retract of
S1 × S1 , but is not a deformation retract.
8. Let X be a 2-disk punctured at two interior points. Prove that a one-point
union of two circles is embedded as a deformation retract of X and hence
π1 (X) is isomorphic to a free group on two generators.
9. Show that the figure of eight is a strong deformation retract of the unit
disc punctured at two interior points.
10. Let p, q, r be three distinct points in the plane R2 and X = R2 − {p, q, r}.
Use van-Kampen theorem to prove that π1 (R2 − {p, q, r}) is a free group
on three generators.
11. Let f : X → Y be a homotopy equivalence. Show that if we remove a

point x0 ∈ X then X − {x0 } need not be homotopically equivalent to
Y − {f (x0 )}.
12. Take the plane R2 . Remove a point p ∈ R2 . Then the remaining space
X = R2 − {p} is of the same homotopy type as circle S1 . Now remove one
point q from S1 , then the remaining space Y = S1 − {q} is contractible.
However, if we remove two distinct points p and q from R2 , then the
remaining space R2 − {p, q} is not contractible. Explain if something is
wrong.
13. Prove that the fundamental group of the punctured n-space Rn − {p},
(n ≥ 3), p ∈ Rn , is trivial. Discuss the cases n = 1, 2.
14. Compute the fundamental group of the space X = R3 − {p, q} where p, q
are two distinct points of R3 .
15. Prove that the cylinder C of height one and the Möbius band M are
not homeomorphic, but they have isomorphic fundamental groups. (Hint:
Möbius band has a circle as its deformation retract.)
16. Let X be a punctured torus. Prove that a one-point union of two circles
is embedded in X as a deformation retract. Use this exercise and van-
Kampen theorem to determine the fundamental group of torus.
17. Let X be the projective plane RP2 punctured at a point. Prove that circle
is embedded in X as a deformation retract. Use this exercise and the
van-Kampen theorem to determine the fundamental group of RP2 .
18. Let X be the Klein bottle punctured at one point. Show that a one-point
union of two circles is embedded in X as a strong deformation retract.
Hence or otherwise determine the fundamental group of the Klein bottle.
19. Let G be a topological group with identity element e. Prove that π1 (G, e)
is abelian. (Hint: In the group π1 (G, e), define another binary operation
by [α]·[β] = [α·β], where (α·β)(t) = α(t)β(t) ∈ G, ∀ t ∈ I. Then π1 (G, e)
is a group with respect to this operation also. Now verify directly that for
arbitrary loops
([α] ◦ [β]) · ([γ] ◦ [δ]) = ([α] · [γ]) ◦ ([β] · [δ]).
Putting β = γ = constant loop e, we find that [α] ◦ [δ] = [α] · [δ]. Now put
α = δ = e.)
Chapter 3
Simplicial Complexes
3.1 Finite Simplicial Complexes
The word “simplicial” is derived from the word “simplex”, and the notion
of a simplex is fundamental not only to the subject of topology, but to several
other branches of mathematics, like graph theory, approximation theory, real
analysis, etc. We will, therefore, first proceed to recall (see Section 1.1.1) the
meaning and definition of a simplex.
Let Rn denote the Euclidean space with its Euclidean topology, and recall
its properties as an n-dimensional vector space over the field of real numbers. A
subset S = {x1 , . . . , xk } of Rn is said to be linearly independent if for arbitrary
scalars α1 , α2 , . . . , αk ,

k
αi xi = 0 ⇒ αi = 0 ∀ i = 1, . . . , k.
i=1
From this notion of linear independence, we now introduce the derived notion
of geometric independence of a set of points {a0 , a1 , . . . , ak } lying in Rn as
follows:
Definition 3.1.1. A set A = {a0 , a1 , . . . , ak }, k ≥ 1, of points of Rn is said to be

geometrically independent if and only if the set S = {a1 − a0 , a2 − a0 , . . . , ak −
a0 } of vectors of Rn is linearly independent. A set having only one point will
be assumed to be geometrically independent.
As an example, consider the set {a0 , a1 , a2 } of three points of the plane

R2 . This will be a geometrically independent set if and only if {a1 − a0 , a2 − a0 }
is linearly independent. However, in R2 , we know that a set having two vectors
is linearly dependent if and only if one of the vectors is a scalar multiple of the
other vector. Thus, the set {a0 , a1 , a2 }, as shown in Fig. 3.1(a), is geometrically
84 Chapter 3. Simplicial Complexes
independent whereas the one, as shown in Fig. 3.1(b), is not geometrically

independent.
6 p
a1
p
a2 6 ar
*
2
r
a1

p ar

a0 0
- -
Fig. 3.1(a) Fig. 3.1(b)
Thus, in Rn , for n sufficiently large, a set having two points will be geometrically
independent if and only if the two points are distinct; a set having three points
will be geometrically independent if and only if they do not lie on a line; a set
having four points will be geometrically independent if and only if all the four
points do not lie on a plane (2-dimensional). The four points a0 , a1 , a2 , a3 , as
shown in Fig. 3.1(c), are geometrically independent, but the ones in Fig. 3.1(d)
are geometrically dependent.
6 6

a0
@ p @
a1
@ p @
p@
- a0 a2-
HH @ a3 @ p
a1
a3
H @ @
a2
Fig. 3.1(c) Fig. 3.1(d)
Let us understand what is a “line” in R3 , what is a “plane” in R3 , etc.? In

fact, by a line in Rn , we mean a translate of some one-dimensional subspace
of Rn . In other words, a subset L of Rn is said to be a line (or, hyperplane
of dimension one) if there exists a one-dimensional subspace V 1 of Rn and a
vector x ∈ Rn such that
L = x + V 1.
Similarly, a subset P of Rn is said to be a plane (or, hyperplane of dimension
two) if there exists a point x ∈ Rn and a two dimensional subspace V 2 of Rn
such that
P = x + V 2.
Definition 3.1.2. A subset H of Rn , n sufficiently large, is said to be a k-

dimensional hyperplane of Rn if we can find a k-dimensional subspace V k of
Rn and a vector x of Rn such that
H = x + V k.
The next result, which generalizes the observations made earlier explains
the reason as to why we use the term “geometrically independent”.
3.1. Finite Simplicial Complexes 85
Proposition 3.1.3. A subset A = {a0 , a1 , . . . , ak }, k ≥ 1, of Rn is geometrically

independent if and only if all the points of A do not lie on a hyperplane of
dimension (k − 1).
Proof. Assume A = {a0 , a1 , . . . , ak } is geometrically independent, i.e., the set
{a1 − a0 , a2 − a0 , . . . , ak − a0 } is linearly independent. If possible, let H k−1
be a hyperplane of Rn of dimension (k − 1) which contains a0 , a1 , . . . , ak . This
means, there is a vector x ∈ Rn and a subspace V k−1 of dimension (k − 1) such
that a0 , a1 , . . . , ak ∈ x + V k−1 . This is equivalent to saying that a0 − x, a1 −
x, . . . , ak − x ∈ V k−1 . By subtracting the first vector from the others, we note
that
a1 − a0 , . . . , ak − a0 ∈ V k−1 .
This is, however, a contradiction because a (k −1)-dimensional subspace cannot
contain a linearly independent set having k vectors.
Conversely, suppose the set {a1 −a0 , a2 −a0 , . . . , ak −a0 } is linearly depen-
dent. Then the subspace of Rn generated by these vectors is of dimension less
than or equal to k −1. Hence we can assume by adding some additional vectors,
if necessary, that there is a subspace V k−1 of dimension k − 1 which contains
all the vectors a1 − a0 , . . . , ak − a0 . Now, the hyperplane a0 + V k−1 = H k−1 ob-
viously contains a0 . Moreover, since ai − a0 are in V k−1 for each i = 1, 2, . . . , k,
ai ∈ H k−1 for all i = 0, 1, 2, . . . , k. Thus, all the vectors a0 , a1 , . . . , ak lie on a
hyperplane of dimension k − 1.
Remark 3.1.4. It follows from the above proposition that if the set
{a0 , a1 , . . . , ak } of points of Rn is geometrically independent, then there is a
hyperplane which contains all these points – it is just the hyperplane a0 + V k ,
where V k is the vector subspace of Rn generated by the linearly independent
set {a1 − a0 , a2 − a0 , . . . , ak − a0 }.
The geometric-independence condition on a set can be expressed alge-

braically and directly in terms of the elements of that set as follows:
Proposition 3.1.5. A subset S = {a0 , a1 , . . . , ak } of Rn is geometrically inde-

pendent if and only if for arbitrary reals αi ,
k k
(i) i=0 αi ai = 0 and (ii) i=0 αi = 0
imply αi = 0, for all i = 0, 1, 2, . . . , k.
Proof. Suppose S is geometrically independent, i.e., the set {a1 − a0 , a2 −
a0 , . . . , ak − a0 } is linearly independent. Then, putting the value of α0 from (ii)
in (i), we see that

k
(−α1 − α2 − . . . − αk )a0 + αi ai = 0.
i=1
This means

k
αi (ai − a0 ) = 0,
i=1
which implies αi = 0, for all i = 1, 2, . . . , k. But, then it also follows from (ii)
that α0 = 0.
Conversely, suppose the given condition is satisfied, and assume

k
βi (ai − a0 ) = 0,
i=1
where βi ’s are some real numbers. Then we have

k
k
− βi a0 + βi ai = 0.
i=1 i=1
k k k
Putting β0 = − 1 βi , we conclude that 0 βi ai = 0 as well as 0 βi = 0.
Hence, by our hypothesis, βi = 0, for all i = 0, 1, . . . , k. This proves the desired
result.
Any set of geometrically independent points of Rn determines a unique
hyperplane of Rn passing through all the points of that set. This follows from
the following:
Proposition 3.1.6. Let A = {a0 , a1 , . . . , ak } be a geometrically independent set

in Rn . Then there is a unique k-dimensional hyperplane which passes through
all the points of A.
Proof. Existence is immediate because, if V k is the subspace of Rn spanned
by the linearly independent set {a1 − a0 , a2 − a0 , . . . , ak − a0 }, then by the
previous remark, the hyperplane a0 +V k evidently passes through all the points
a0 , a1 , . . . , ak .
For uniqueness, suppose there are two hyperplanes H k and F k of di-
mension k which contain the points a0 , a1 , . . . , ak . This means there are two
subspaces V k and W k of dimension k each and two vectors x, y ∈ Rn such that
Hk = x + V k
F k = y + W k.
Since a0 , a1 , . . . , ak ∈ H k , it follows that a1 − a0 , a2 − a0 , . . . , ak − a0 ∈ V k .
Similarly, a1 − a0 , a2 − a0 , . . . , ak − a0 ∈ W k . Since the set {a1 − a0 , a2 −
a0 , . . . , ak − a0 } is linearly independent and V k , W k are vector spaces each of
dimension k containing that set, we must have V k = W k . Consequently, H k
and F k are two cosets of the same subspace V k . Since the intersection of two
cosets is either identical or disjoint, and here the two cosets contain the points
a0 , a1 , . . . , ak , we conclude that H k = F k .
The next proposition gives a nice representation of the points of a hyper-

plane as a linear combination of geometrically independent points. We have

in Rn . Then each point of the hyperplane passing through all the points of A
can be expressed uniquely as

k
k
h= αi ai where αi = 1.
i=0 i=0
Proof. We know that the hyperplane passing through the set A is given by the
set of all vectors h satisfying the condition

k
h = a0 + αi (ai − a0 ),
i=1
where αi ’s are real numbers which are unique. This means h can be expressed
uniquely as

k

k
h = 1− αi a 0 + αi ai
i=1 i=1

k
= α0 a0 + αi ai
i=1

k
= α i ai ,
i=0
k
where i=0 αi = 1.
Remark 3.1.8. If {a0 , a1 , . . . , ak } is a geometrically independent set in Rn ,

k ≤ n, then we can give a one-to-one correspondence between the vector space
Rk and the hyperplane H k passing through a0 , a1 , . . . , ak . The desired one-to-
one correspondence will result as follows: Any element of Rk would be a set
of k-tuples (α1 , α2 , . . . , αk ) of real numbers. Let this k-tuple correspond to the
k k
point (1 − i=1 αi )a0 + i=1 αi ai of H k . It is also not difficult to see that
this one-to-one correspondence, being a translation by some vector, is distance
preserving and consequently is a homeomorphism.
The following concept of barycentric coordinates is important.
Definition 3.1.9. Let A = {a0 , a1 , . . . , ak } be a geometrically independent set of

points of Rn , and let h be any point in the unique hyperplane passing through
a0 , . . . , ak . Write h as

k
k
h= αi ai where αi = 1.
i=0 i=0
Then, the real numbers α0 , α1 , . . . , αk , which are uniquely determined by A, are

called the barycentric coordinates of the point h with respect to the set A.
Note that barycentric coordinates of a point are real numbers which could
be positive, zero or negative. In the Euclidean space R3 , for example, let a0 =
(2, 0, 0), a1 = (0, 3, 0). Then the one-dimensional hyperplane of R3 containing
a0 , a1 is the straight line passing through a0 and a1 (Fig. 3.2(a)). All the points
h on the line, lying strictly between the two points a0 and a1 , will have positive
barycentric coordinates. The barycentric coordinates of a0 are (1, 0) whereas
that of a1 are (0, 1). If h = α0 a0 + α1 a1 , where α0 + α1 = 1, lies on the left of
a0 , then the barycentric coordinates of h are α0 , α1 where α0 > 0 but α1 < 0.
But if h lies on the right of the point a1 , then α0 < 0 and α1 > 0.
6

a -
1
a0
α0 ,α1 >0

Fig. 3.2(a)
Consider a geometrically independent set {a0 , a1 , a2 } in any Rn , n ≥ 2.

Then the hyperplane of Rn containing a0 , a1 , a2 is the unique plane passing
through a0 , a1 and a2 (Fig. 3.2(b)).
J

a2

J

J

J

J

α0 , α1 , α2 > 0 J

J

a0 a1J
Fig. 3.2 (b)
The barycentric coordinates of a0 , a1 and a2 are (1,0,0), (0,1,0) and (0,0,1)

respectively. The barycentric coordinates α0 , α1 , α2 of any point lying in
the interior of the triangle are all positive; the barycentric coordinates of
any point below the line joining a0 and a1 and opposite to a2 are such that
α0 > 0, α1 > 0, α2 < 0 etc.
As mentioned earlier, the notion of a simplex introduced below, is one of

the most fundamental concepts.
Definition 3.1.10. Let A = {a0 , a1 , . . . , ak } be a geometrically independent set

of points in Rn , n ≥ k. Then the k-dimensional geometric simplex or k-simplex
spanned by the set A, and denoted by σ k , is the set of all those points x ∈ Rn
such that
k k
x= αi ai where αi = 1 and αi ≥ 0
i=0 i=0
for each i = 0, 1, 2, . . . , k.
In other words, the k-simplex σ k consists of all those points of the hyper-
plane passing through a0 , a1 , . . . , ak whose barycentric coordinates with respect
to a0 , a1 , . . . , ak are nonnegative. The points a0 , a1 , . . . , ak are called the ver-
tices of the simplex σ k . We usually write σ k = a0 , a1 , . . . , ak just to indicate
that σ k is the k-simplex with vertices a0 , a1 , . . . , ak .
Example 3.1.11. A 0-simplex in Rn is simply a singleton set or a point. Let

a0 , a1 be any two distinct points of Rn . Then the 1-simplex determined by
{a0 , a1 } is just the closed straight line segment joining the points a0 and a1 .
a0 a1
Let a0 , a1 , a2 be any three distinct points in Rn not lying on a line. Then
the 2-simplex a0 , a1 , a2 determined by {a0 , a1 , a2 } is simply the triangle (in-
side as well as the sides) joining a0 , a1 and a2 .
Similarly, if a0 , a1 , a2 , a3 are four points which are non coplanar, then they
determine a 3-simplex a0 , a1 , a2 , a3 , which is simply the solid tetrahedron with
a0 , a1 , a2 and a3 as its vertices. It has four faces (triangles) and six edges.
a0
@
@
a1
HH @
a3
H
a2
Before we look into further properties of a simplex, let us understand the

relationship of a simplex with the so called convexity property. Recall that a
subset S of Rn is said to be convex if x1 , x2 ∈ S implies that all the points on
the line segment joining x1 and x2 also belong to S. Note that any point on
the line segment joining x1 and x2 can be expressed uniquely as tx1 + (1 − t)x2
where 0 ≤ t ≤ 1. Thus, S is convex if and only if x1 , x2 ∈ S implies that

tx1 + (1 − t)x2 ∈ S for all t, 0 ≤ t ≤ 1. One can easily see that the intersection
of any family of convex subsets of Rn is again a convex subset. If A is any
subset of Rn , then we can, therefore, consider the family of all convex subsets
of Rn containing A. The set Rn , being convex, belongs to this family, and so the
family is nonempty. The intersection of the above family will be the smallest
convex subset of Rn containing A. This intersection is called the convex hull of
the set A. The next result describes the simplex a0 , a1 , . . . , ak as the convex
hull of the geometrically independent set {a0 , a1 , . . . , ak }.

of points in Rn . Then the simplex σ k = a0 , a1 , . . . , ak is the convex hull of the
set A.
Proof. Recall that any point of the hyperplane H k containing A can be uniquely
written in terms of its barycentric coordinates. For i = 0, 1, . . . , k, let us now
consider the subset Hi of the hyperplane H k consisting of those points whose
i-th barycentric coordinate is nonnegative. In other words,
Hi = {(α0 , α1 , . . . , αk ) ∈ H k | αi ≥ 0}.
We observe that each one of these half-planes Hi is convex. For, let x1 =
(α0 , α1 , . . . , αk ), x2 = (β0 , β1 , . . . , βk ) be any two points in Hi . Then αi , βi
are nonnegative. Consequently, for each t, 0 ≤ t ≤ 1, the i-th coordinate of
tx1 +(1−t)x2 , viz., tαi +(1−t)βi , is also nonnegative, and so tx1 +(1−t)x2 ∈ Hi .
Now we observe that σ k = H0 ∩ H1 ∩ · · · ∩ Hk , which means σ k , being the
intersection of convex subsets of Rn , is convex.
Next, we show that if B is any convex subset of Rn containing A, then B
contains σ k . We prove this by induction on k. For k = 0, the result is trivial.
Assume k ≥ 1 and the result to be true for k − 1, i.e., if B contains the set
{a1 , a2 , . . . , ak } then it also contains the simplex a1 , a2 , . . . , ak . We assert that
any point of σ k can be written as ta0 + (1 − t)x1 , where x1 ∈ a1 , . . . , ak . This
will prove the result because B is convex, contains a0 , and also contains x1 by
inductive hypothesis. We know that any point x ∈ σ k can be written as

k
k
x= αi ai , αi ≥ 0, αi = 1.
i=0 0
If α0 = 1, then x = a0 ∈ B and if α0 = 0 then, obviously, x ∈ a1 , a2 , . . . , ak

k
and so again x ∈ B. If both α0 , 1 αi are nonzero, then

k
αi
x1 = k ai
1 1 αj
k
belongs to a1 , . . . , ak . Because α0 + 1 αj = 1, x = α0 a0 + (1 − α0 )x1 . This
proves the assertion and hence the result.
Definition 3.1.13. Let σ k = a0 , a1 , . . . , ak be a k-simplex. Then the set of

those points of σ k for which all barycentric coordinates are strictly positive, is
called the open k-simplex σ k .
For example, the open interval joining the points a0 and a1 of Rn will
be an open 1-simplex, the interior of the triangle joining a0 , a1 , a2 will be an
open 2-simplex, and so on. An important observation is that an open 0-simplex,
determined by a 0-simplex a0 , will be the simplex a0 itself.
Definition 3.1.14. Let σ p , σ q , p ≤ q ≤ n be two simplexes in Rn . We say that

σ p is a p-dimensional face of σ q (or merely a p-simplex of σ q ) if each vertex of
σ p is also a vertex of σ q . If σ p is a face of σ q and p < q, then σ p is a proper
face of σ q .
Any 0-dimensional face of σ q is simply a vertex of σ q . A 1-dimensional

face of a simplex is usually called an edge of that simplex. If we consider a
3-simplex σ 3 = a0 , a1 , a2 , a3 , then it has four 0-faces, six edges, four 2-faces
which are proper faces and one 3-face which is σ 3 itself.
If x ∈ σ 3 , then note that either x is a vertex of σ 3 or it lies in the interior

of some edge or it lies in the interior of some 2-simplex or it lies in the interior
of tetrahedron σ 3 itself. We also observe that any two open faces of σ 3 are
either identical or disjoint. It follows that σ 3 can be seen as the disjoint union
of all its open faces or simplexes. More generally, any k-simplex σ k can be
written as the disjoint union of all its open simplexes.
of a simplex σ = v0 , v1 , . . . , vk and we write

k
If
x is an interior point
x = αi vi , αi > 0, αi = 1, then (α0 , α1 , . . . , αk ) are the barycentric
coordinate of the point x. If one of the coordinates say αi , is zero we get an
interior point of the face of σ k opposite the vertex vi . If we put two coordinates
αi , αj to be zero, we get the interior points of a face of σ k . Finally, if we put all
coordinates to be zero except one, say αi , then we get the interior of the vertex
vi which is the point vi itself. It follows that the points of the closed simplex
|σ k | can be writtenin terms of barycentric coordinates as x = (α0 , α1 , . . . , αk )
where αi ≥ 0 and αi = 1. This representation of x ∈ |σ k | is unique.
Now we will show that the map pi : |σ k | → [0, 1], defined by

pi (α0 , α1 , . . . , αk ) = αi , called the projection on the i-th barycentric
coordinate, is continuous. We consider the unit vectors e0 , e1 , . . . , ek
of Rk+1 . Then there is a linear homeomorphism from the closed sim-
plex v0 , v1 , . . . , vk to the closed simplex e0 , e1 , . . . , ek
defined by
h(α0 v0 + α1 v1 + . . . αk vk ) = α0 e0 + α1 e1 + . . . + αk ek , αi ≥ 0, αi = 1. If
qi : Rk+1 → R is the canonical projection to the i-th component, then clearly
qi ◦ h = pi . Hence pi is continuous.
Simplexes are supposed to be the “building blocks” for a class of topo-

logical spaces considered a “nice” class of spaces. These spaces possess rich
topological properties, but the problem of their classification, as usual, is diffi-
cult. The methods and concepts of algebraic topology are such that their power
can be tested most effectively on these “nice” spaces. A variety of algebraic in-
variants have been defined for topological spaces, but often each of these is first
defined for these nice spaces and then their generalizations are considered for
arbitrary spaces. Let us now come to the main
Definition 3.1.15. A simplicial complex or a geometric complex K is a finite

collection of simplexes of Rm , m sufficiently large, which satisfies the following
conditions:
(i) If σ ∈ K, then all the faces of σ are also in K.
(ii) If σ and τ are in K, then either σ ∩ τ = ∅ or σ ∩ τ is a common face of
both σ and τ .
We also define the dimension of a simplicial complex K, denoted by dim K, to
be −1 if K = ∅ and to be n ≥ 0 if n is the largest integer such that K has an
n-simplex.
We must mention that what we have defined above is actually a finite

simplicial complex; we will define arbitrary simplicial complex later. Here, it is
also necessary to point out that an n-dimensional simplicial complex K need
not lie in Rn . Since a simplicial complex K satisfies the two conditions of the
foregoing definition, the complex could be forced to lie in Rm where m is far
greater that n. We will see examples of this later.
Example 3.1.16. Let σ 2 = a0 , a1 , a2 be a 2-simplex. Then the set K of all

faces of σ 2 , i.e.,
K = {a0 , a1 , a2 , a0 , a1 , a0 , a2 , a1 , a2 , a0 , a1 , a2 }
is a simplicial complex. This simplicial complex is denoted by Cl(σ 2 ) and is
called the closure of σ 2 . More generally, for any n, we can define the simplicial
complex Cl(σ n ) to be the collection of all faces of σ n .
Example 3.1.17. Consider the collection K of all simplexes shown in Fig. 3.3
(both pieces taken together) where shaded triangles indicate the 2-simplexes.
This is a finite collection of several 2-simplexes, 1-simplexes and 0-

simplexes satisfying the two conditions and, therefore, is a geometric complex.
On the other hand, the collection K of simplexes shown in Fig. 3.4 (both
pieces taken together) is not a simplicial complex.
Two simplexes σ k and σ l are said to be properly joined if their intersection

is either empty or is a common face of both the simplexes. Condition (ii) of
3.2. Polyhedra and Triangulations 93
V2
V1
V3
V0
V4
Fig. 3.3: The simplicial complex K
Fig. 3.4: The collection K
the definition of a simplicial complex says that all the simplexes of K must
be properly joined. In Fig. 3.4, the intersection of v0 , v1 , v2 and v3 , v4 , v5
is a point, which is a face of the latter simplex but not the former. Similarly,
the 2-simplexes a1 , a4 , a5 and a1 , a2 , a3 of the same figure are not properly
joined. The definition of a simplicial complex can be stated more intuitively
as follows: A finite collection K of properly joined simplexes is said to be a
simplicial complex (or geometric complex or complex ) if each face of a member
of K is again a member of K.
3.2 Polyhedra and Triangulations

Definition 3.2.1. Let K be a simplicial complex. Let |K| = σ∈K σ be the
union of all simplexes of K. Then |K|, being a subset of Rn for some n, will be
a topological space with the topology induced from Rn . This space |K| is called
the geometric carrier of K. A subspace of Rn , which is the geometric carrier
of some simplicial complex, is called a rectilinear polyhedron.
Definition 3.2.2. A topological space X is said to be a polyhedron if there exists

a simplicial complex K such that |K| is homeomorphic to X. In this case, the
space X is said to be triangulable and K is called a triangulation of X.
We must point out the difference between a simplicial complex K and

its geometric carrier |K|. The complex is a finite set of simplexes whereas |K|
is a subspace of some Euclidean space. Simplexes of K are simply elements of
the finite set K and not subsets of K while each simplex is a subspace of the
space |K|. Note that |Cl(σ k )| = σ k = |σ k |.
By the very definition, |K| is a polyhedron for any K and the geometric
shape of the rectilinear polyhedron |K| consists of points, edges, plane triangles,
tetrahedra, etc. Curved lines or surfaces will not be parts of |K| unless these
are contained in some higher dimensional simplexes of |K|. On the other hand,
the geometric shape of an arbitrary polyhedron could be anything from nice
tetrahedra to spheres, ellipsoids and much more curved surfaces. Let us consider
some examples for further illustrations.
Example 3.2.3. Consider the curved portion X of the unit circle defined by
(Fig.3.5)
X = {(cos θ, sin θ) ∈ R2 | 0 ≤ θ ≤ π/2}.
a1 = (0, 1) a
@ a
@ `
a2@
@
@a -
a0 = (1, 0)
Fig. 3.5
Then X is a polyhedron, but not a rectilinear polyhedron. Let a0 , a1 be

the simplex joining (1, 0) and (0, 1) and a2 be any point on this simplex.
Then the simplicial complex K = {a0 , a1 , a0 , a2 } is a triangulation of
X since |K| is homeomorphic to X. If we consider the simplicial complex
L = {a0 , a1 , a2 , a0 , a1 , a1 , a2 }, then also |L| is homeomorphic to X
and so L is another triangulation of X.
Example 3.2.4. (Cylinder) Consider the hollow prism (Fig.3.5(a)). We break

each of the side rectangles into two triangles. Then
K = {a0 , a1 , a2 , b0 , b1 , b2 , a0 , a1 , a1 , a2 , a0 , a2 ,
b0 , b1 , b1 , b2 , b0 , b2 , a0 , b2 , a1 , b2 , a0 , b1 , a0 , b0 ,
a1 , b1 , a2 , b2 , a0 , b1 , b0 , a0 , a1 , b1 , a1 , b1 , b2 ,
a1 , a2 , b2 , a0 , a2 , b2 , a0 , b0 , b2 }
is a triangulation of the cylinder.
b2 b2
b0 b1 b0 b1
a2 a2
a0 a1
a0 a1
Fig. 3.5(a)
We need the following important topological result before giving further

examples. This shows, in particular, that a square, a triangle, a pentagon etc.,
are all homeomorphic to the 2-disk D2 ; a cube, a tetrahedron, an octahedron,
etc., are all homeomorphic to the 3-disk D3 , and so on:
Proposition 3.2.5. Let U be an open set of Rn , n ≥ 1 which is bounded and

convex, and w ∈ U . Then
(i) Each half-ray L starting from w intersects the boundary of U at precisely
one point.
(ii) There is a homeomorphism of U with the unit disk Dn which carries the
boundary of U onto the unit sphere Sn−1 .
Proof. (i) Let L be a fixed half-line starting from w and consider the intersection
of L with open set U . Clearly, it is a convex, bounded and open subset of L
and so it may be put in the form
{w + t · p : t ∈ [0, a)},
where p is the unit vector along L. Then, obviously, L intersects the closure U
of U in the point x = w + a · p, say. To prove the uniqueness of the point x,
suppose y is another such point. Then y = w + b.p for some b > a and x lies
between w and y on the line L, i.e., x = (1 − t)w + ty, where t = a/b. Solving
w for this t, we have w = (x − ty)/(1 − t). Now, we can choose a sequence {yn }
of points in U which converges to y. Then putting wn = (x − tyn )/(1 − t), we
find that wn converges to w. Hence we have an n such that wn ∈ U . But this
is a contradiction because U being convex means x = (1 − t)wn + tyn must be
in U .
(ii) Without loss of generality, we can assume that w is the origin. The map
f : Rn − {0} → Sn−1 defined by f (x) = x/x is onto and continuous. By
(i) above, the restriction of f to the boundary Bd(U ) defines a bijection from
Bd(U ) to Sn−1 , which must be a homeomorphism since Bd(U ) is compact. Let
g : Sn−1 → Bd(U ) be its inverse. Now, extend g to G : Dn → U by defining

g(x/x) · x x
= 0
G(x) =
0 x = 0.
In fact, G maps the line segment lying in Dn joining origin to a point u in Sn−1
to the line segment in U joining origin to the point g(u). Then G is a bijective
continuous map, and so it is a homeomorphism whose restriction on Sn−1 is
the map g.
Example 3.2.6 (Discs). Consider the unit disc D2 in R2
D2 = {(x, y) ∈ R2 | x2 + y 2 ≤ 1}.
Let us consider the simplicial complex Cl(σ 2 ), where σ 2 = a0 , a1 , a2 is a

2-simplex. Then |Cl(σ 2 )| is the triangle Δ with vertices a0 , a1 , a2 . Since D2
is homeomorphic to the triangle Δ (see Example 1.4.4), we find that D2 is
a polyhedron and Cl(σ 2 ) is a triangulation of D2 . More generally, let Dn =
{(x1 , x2 , . . . , xn ) ∈ Rn | x21 + x22 + . . . + x2n ≤ 1} be the unit disc. Then Dn is a
polyhedron – the simplicial complex K consisting of all faces of some n-simplex
σ n is a triangulation of Dn . It now follows from Proposition 3.2.5 that every
compact convex subset of Euclidean space is a polyhedron.
Definition 3.2.7. Let K be a k-dimensional simplicial complex. For each r, 0 ≤

r ≤ k, let K r denote the set of all those simplexes of K which are of dimension
≤ r. Then K r is a simplicial complex which is known as r-dimensional skeleton
of K. The space |K r | will be a rectilinear subpolyhedron of |K|.
Example 3.2.8 (Sphere). Let σ k be a k-simplex, k ≥ 1, and K = Cl(σ k ) be

the simplicial complex consisting of all faces of σ k . Let K k−1 be the (k − 1)-
dimensional skeleton of K. Then K k−1 consists of all proper faces of σ k . It can
be shown (see Example 1.4.3) that |K k−1 | is homeomorphic to
Sk−1 = {(x0 , x1 , . . . , xk−1 ) ∈ Rk | Σk−1
0 x2i = 1},
which is a (k − 1)-dimensional unit sphere. This proves that every k-sphere Sk
is a polyhedron, and the collection of all proper faces of a (k + 1)-simplex gives
a triangulation of Sk .
It is to be observed that if K is a simplicial complex, then |K| is a compact

metric space because it is finite union of simplexes which are clearly compact.
This means any polyhedron (according to our definition in this book) is com-
pact. If K is a triangulation of a polyhedron X, then |K| is just a representation
of the space X as a rectilinear polyhedron. Very often whenever we are required
to prove some theorem about a polyhedron X, we prefer to prove that result
for |K| rather than for X because |K| has more structure than just a topology
on it. Another important point to be noted is that whenever we talk of the
n-sphere
n
Sn = {(x0 , x1 , . . . , xn ) ∈ Rn+1 | 0 x2i = 1},
this Sn is the standard unit sphere in Rn+1 . In topology, two spaces are not
distinguished if they are homeomorphic. Now, if X is a topological space which
is homeomorphic to Sn , it is a common practice to say that “X is a topological
n-sphere”. This space X may appear geometrically very different from a gen-
uine sphere allowing lot of flexibility but is still called an n-sphere. A similar
practice is also prevalent regarding an n-disc Dn . It will be very helpful if we
look at a polyhedron X with a similar meaning, i.e., a polyhedron to a recti-
linear polyhedron is the same thing as a topological sphere to a genuine sphere.
Giving examples of polyhedra as the union of a collection of simplexes

satisfying the two conditions stated earlier, is not always a convenient method.
One at once comes across complicated problems of analytic geometry for verify-
ing “properly joined”condition of the definition. It is easier to specify polyhedra
using abstract simplicial complexes, which we define as follows:
Definition 3.2.9. An abstract simplicial complex is a collection K of finite

nonempty sets such that if A is an element of K, then so is every nonempty
subset of A.
Using the above definition it is now easy to see that the following diagram
defines a simplicial complex K where the vertices are a, b, c, d, e and
f and the simplexes are a, b, d, b, c, d, c, d, e, a, c, e, a, e, f and a, f, d
along with their nonempty subsets. We also notice that K is a triangulation of
a cylinder, since |K| is just a hollow prism. This triangulation of the cylinder
is easier to describe as compared to Fig. 3.5(a).
f d e f
@ @
@ @
@ @
@ @
@ @
a b c a
Fig. 3.5(b)
In the above definition of an abstract simplicial complex K the singletons
are called vertices of K, and in this terminology, the set of vertices of an abstract
simplicial complex K may be finite or infinite. We note that each n-simplex
σ = v0 , v1 , . . . , vn can be identified with
a simplex of R . Since each∞ R is
n n
∞
embedded in R , we find that |K| = {σ | σ ∈ K} is a subset of R . The
topology on |K| is defined to be the weak topology induced by simplexes of K
(see Section 3.5 for details).
Example 3.2.10 (Möbius band). We now give a triangulation of the Möbius

band M . Recall (see Section 1.1.4) that M is a topological space obtained
as follows: take a rectangle R and identify the two opposite sides of R by
twisting one of the sides through 180◦ . In order to give a triangulation of M ,

we start with a triangulation L of the rectangle R itself (see Fig. 3.6) in which
the two opposite sides are taken as 1-simplexes of L. Remember that when
we identify the two opposite edges of R, two 1-simplexes of L are identified
yielding a triangulation K of M . The labelling given in Fig. 3.6 indicates the
identifications. The two conditions already satisfied by L continue to hold for
K also except that the number of 1-simplexes of K is one less than the number
of 1-simplexes of L. The number of 0-simplexes of K is also less than those of
L. Thus, a triangulation of M is shown in Fig. 3.6.
a3 a4 a5 a0
a0 a1 a2 a3
Fig. 3.6
Note that the edge a0 , a3 on the left of the rectangle R is identified with the
edge a0 , a3 (order reversed) on the right, and the resulting quotient space
M , the Möbius band, is homeomorphic to the geometric carrier of the simpli-
cial complex K which consists of six triangles and all of their faces as in Fig. 3.6.
At this stage we go back to an important observation indicated earlier.

Suppose we have a k-dimensional simplicial complex K. Then |K| need not lie in
some k-dimensional Euclidean space for the same k. For example, a circle is the
geometric carrier of a 1-dimensional simplicial complex, but it cannot be em-
bedded in the real line R. In fact, |K| need not be homeomorphic to a subspace
of even a (k + 1)-dimensional Euclidean space. For example, a 1-dimensional
simplicial complex (a graph) may not be embedded in 2-dimensional Euclidean
space, i.e., it need not be a planar graph (see the next example). A triangula-
tion of the projective plane is a 2-dimensional simplicial complex K, but |K|
cannot be embedded even in 3-dimensional Euclidean space. This observation
raises an interesting question: If K is a k-dimesnional simplicial complex, then
what is the least n such that |K| can be embedded in Rn ? The answer is known:
a k-dimensional finite simplicial complex can always be embedded in R2k+1 .
Of course, if K is not finite, then it has to be assumed to be locally finite
(see E.H.Spanier [18] p. 120). This is the best possible result because of the
following example:
Example 3.2.11 (Kuratowski). Consider the following simplicial complex:
K = {v0 , v1 , v2 , v3 , v4 , v5 , v0 , v3 , v0 , v4 , v0 , v5 ,
v1 , v3 , v1 , v4 , v1 , v5 , v2 , v3 , v2 , v4 , v2 , v5 }.
v0 v1 v2
v3 v4 v5
Fig. 3.6(a) : Kuratowski’s non-planar graph
Clearly, K (Fig. 3.6(a)) is a connected 1-dimensional simplicial complex. It

should be pointed out that any two edges of K intersect only at the vertices,
not at any other point. The figure is not faithful since edges like v0 , v4 and
v1 , v3 seem to intersect, but actually they don’t. A combinatorial argument
using Euler’s theorem (see Chapter 4) shows that K cannot be a planar graph,
i.e., |K| cannot be embedded in the Euclidean plane R2 .
Example 3.2.12 (Torus). A torus T is the quotient space obtained as follows

(see Section 1.1.4): Take a rectangle ABCD in the plane R2 and identify the
opposite sides, i.e., identify AB with DC and AD with BC (Fig. 3.7).
D C
A B
Fig. 3.7
Once again, we start with a triangulation L of the rectangle itself (see Fig. 3.8)
in which each side of the rectangle is broken as the union of three 1-simplexes.
The identification of opposite sides (identical labeling) collapses several pairs of
1-simplexes of L into 1-simplexes of a new complex K, and several 0-simplexes
of L collapse to 0-simplexes of K yielding a triangulation K of the torus T .
It is easily seen that conditions defining the simplicial complex L continue to
hold for K also. Thus, the following diagram with indicated labeling gives a
triangulation K of T :
a0 a1 a2 a0
! ! !
!!! ! !! !!!
a4 ! ! !
!! a8 !!! a7 !!!
a4
!! ! !
a3 ! !! !! !!
a3
! ! a5
!! a6
!
!! !! !!
a0 a1 a2 a0
Fig. 3.8
It has 9 vertices, 27 edges and 18 triangles which are joined properly as shown
in Fig. 3.8. Again note that |K| lies in R3 , not in R2 (why?), though K is
2-dimensional. We also point out that besides the above, several other trian-
gulations of T can be given and it is a good question to ask: what will be the
minimum number of vertices, edges and triangles in a triangulation of T ? This
question will be answered in Chapter 4 (see Chapter 4, pseudomanifolds) for a
large number of polyhedra.
The Klein bottle and the projective plane (see Section 1.1.4) are also
good examples of polyhedra and the reader should give a triangulation of each
of these too (see exercises at the end). These triangulations will be needed in
understanding the simplicial homology discussed in the next chapter.
Definition 3.2.13. Let K be a simplicial complex. A simplicial complex L is said

to be a subcomplex of K if L ⊂ K. The boundary of a simplicial complex K,
denoted by ∂K, is defined by
∂K = {τ | τ is a face of a simplex σ k ∈ K which belongs to a unique
(k + 1)-simplex of K}.
In other words, ∂K consists of all proper faces of maximal simplexes in K and
all of their faces. Thus ∂K is a subcomplex of K of dimension one less than
dim(K ).
For example, let K = Cl(σ 2 ) where σ 2 = a0 , a1 , a2 . Then ∂K =

{a0 , a1 , a2 , a0 , a1 , a1 , a2 , a0 , a2 }. Let K be the first piece of Exam-
ple 3.1.17. Then ∂K = {v1 , v2 , v1 , v3 , v2 , v3 , v1 , v2 , v3 , v0 , v4 }.
Definition 3.2.14. Let K and L be two simplicial complexes. By a simplicial map

f : K → L we mean a map f from the vertices of K to the vertices of L such that
if a0 , a1 , . . . , ak is a simplex of K, then f (a0 ), f (a1 ), . . . , f (ak ) are vertices of
a simplex in L. In other words, canceling repetitions, f (a0 ), f (a1 ), . . . , f (ak )
is a simplex in L whenever a0 , a1 , . . . , ak is a simplex in K.
For any simplicial complex K, the identity map IK : K → K is a simplicial

map. We also note that if f : K → L and g : L → M are two simplicial maps,
then their composite g ◦ f : K → M is also a simplicial map.
Definition 3.2.15. A simplicial map f : K → L is said to be an isomorphism if

it is bijective on vertices and has the property that v0 , v1 , . . . , vk is a simplex
in K if and only if f (v0 ), f (v1 ), . . . , f (vk ) is a simplex in L. Two simplicial
complexes K and L are said to be simplicially isomorphic (or just isomorphic)
if there is a simplicial isomorphism f : K → L.
It is easy to see that “to be simplicially isomorphic” is an equivalence

relation in the class of all simplicial complexes. As a result, the class of all
simplicial complexes is decomposed into mutually disjoint isomorphism classes.
Let K be a finite simplicial complex and suppose x ∈ |K|. Then there

is a simplex σ k = v0 , v1 , . . . , vk such that x is in the open simplex σ k . If
α0 , α 1 , . . . , α k
are the barycentric coordinates of x w.r.t. v0 , v1 , . . . , vk , then we
k
can write x = 0 αi vi where αi > 0. Now if w is any other vertex of K, not in
σ k , we put the w-th barycentric coordinate of x to be zero. Thus we find that
each x ∈ |K| has a unique representation as

x= αi vi , αi ≥ 0, αi = 1, vi ∈ K.
i
Let v be a fixed vertex of K and define a function fv : |K| → [0, 1] by fv (x) = v-

th barycentric coordinate of x. Note that fv is well defined because if x is in two
closed simplexes σ and τ , then the v-th barycentric coordinate of x depends
only on the vertices of σ ∩ τ , not on other vertices. If |σ k | is a closed simplex
of K and x ∈ |σ k |, then fv (x) is simply the projection on the v-th barycentric
coordinate of x, and is zero otherwise. Hence fv restricted to each closed simplex
σ of K is continuous. It follows that fv itself is continuous. This proves the next
Proposition 3.2.16. The vertex function fv : |K| → [0, 1] is continuous.
Next we have
Proposition 3.2.17. Let f : K →L be a simplicial map. Then there is an induced

continuous map |f | : |K| → |L| defined as follows: If x ∈ |K|, then x belongs to
k
the interior of a unique simplex v0 , v1 , . . . , vk . We write x = i=0 αi vi and
k
define |f |(x) = i=0 αi f (vi ).
Proof. First we define |f | on each closed simplex |σ| by the formula: if x =
α0 v0 + α1 v1 + . . . + αk vk , αi ≥ 0, αi = 1, then |f |(x) = α0 f (v0 ) + α1 f (v1 ) +
. . . + αk f (vk ). This shows that |f | is well-defined. If x lies in the intersection
|σ| ∩ |τ | of two closed simplexes σ and τ , then the barycentric coordinates of
x depend only on the common vertices v0 , v1 , . . . , vl of σ ∩ τ . Hence |f |(x) =
α0 f (v0 ) + α1 f (v1 ) + . . . + αl f (vl ) is unique whether x is treated as a point of |σ|
or a point of |τ |. Therefore, now it suffices to prove that |f |||σ| is continuous for
each simplex σ ∈ K. To see this note that the barycentric coordinates of |f |(x)
are αi , i = 0, 1, . . . k or the sum of these coordinates depending on whether
f (vi ) are all distinct or not. In any case, when we consider the composite
qi ◦ |f | : |σ k | → R where qi : Rk+1 → R are the canonical projection, then
this is simply the projection map (α0 , α1 , . . . , αk ) → αi or is a sum of these
projection maps. Hence qi ◦ |f | is continuous for each i = 0, 1, 2, . . . , k. This
mean |f | : |K| → |L| is continuous.
Remark 3.2.18. If we look at |K| and |L| as topological spaces, then a con-
tinuous map f : |K| → |L| may or may not be induced by some simplicial
map g : K → L. In case the map f is induced by some simplicial map, then
sometimes we abuse the language and refer f to be a “simplicial map”. In
this terminology, the set of all continuous maps from |K| to |L| is divided into
two classes, viz., the “simplicial” maps and the “non-simplicial” maps. Since
we are dealing with only finite simplicial complexes, the number of simplicial
maps from K to L is always finite, whereas the number of non-simplicial maps
is always infinite, unless, of course, |L| consists of only 0-simplexes (Why?)
Let f : K → L and g : L → M be two simplicial maps. Then it follows

from the definitions that the induced map |g ◦ f | : |K| → |M | is simply the
composite of the induced maps |f | : |K| → |L| and |g| : |L| → |M |, i.e., |g ◦ f | =
|g|◦|f |. Also, if IK is the identity map, then the induced map |IK | : |K| → |K| is
also the identity map I|K| . These observations say that K → |K| is a covariant
functor from the category of simplicial complexes and simplicial maps to the
category of topological spaces and continuous maps.
Definition 3.2.19. A simplicial complex K is said to be connected if for each

pair of vertices a, b ∈ K, there exists a sequence of 1-simplexes ai , ai+1 for
i = 0, 1, 2, . . . , p − 1 in K such that a = a0 , b = ap .
Let n ≥ 1 and K = Cl(σ n ) for some n-simplex σ n . Then the r-dimensional

skeleton K r , r ≥ 1 of K is connected whereas K 0 is not connected. It is eas-
ily seen that any triangulation of sphere, Möbius band or torus is connected.
Maximal connected subcomplexes of a simplicial complex are called the com-
binatorial components of that complex.
3.3 Simplicial Approximation

We have already noted that if f : K → L is a simplicial map, then the
induced map |f | : |K| → |L| in their geometric carriers (associated polyhedra)
is continuous. However, given a continuous map h : |K| → |L|, there may not
be any simplicial map from K to L which induces h. In fact, when |K| and |L|
are genuine geometric complexes, any induced continuous map |f | : |K| → |L|
will always take a straight line segment contained in a simplex of |K| into a
vertex or straight line segment of |L|, and in a linear fashion. A continuous map
h : |K| → |L|, on the other hand, may take a straight line segment to a curved
line segment and, therefore, need not be simplicial. Consider the unit interval
|K| = [0, 1], where K = {0, 1, 0, 1} and the unit square |L| = I × I in the
plane with the four vertices, viz., (0, 0), (1, 0), (1, 1), (0, 1), four edges and two
triangles with one common edge, as shown in Fig. 3.9.
(0,1) (1,1)

(0,0) (1,0)
Fig. 3.9
3.3. Simplicial Approximation 103
For each n ≥ 1, we define continuous maps fn : |K| → |L| by putting

fn (t) = (t, tn ). Then f1 is simplicial, but fn (n
= 1) is never simplicial. The
unit interval can also be considered as the geometric carrier |K | of some other
simplicial complex, say, K = {0, 1/3, 1, 0, 1/3, 1/3, 1}. Even then
there is no simplicial map from K to L which induces f2 . In fact, there is
no representation of [0, 1] as the geometric carrier of any simplicial complex
for which f2 or fn (n ≥ 3) is simplicial because each of these maps takes any
1-simplex, however small, to a curved arc which is not straight.
We have the following
Definition 3.3.1. Let K be a simplicial complex and v be a vertex of K. By the

star of v, denoted by st(v), we mean the following subset of K:
st(v) = {σ : σ is a simplex of K and v is a vertex of σ}.
By the open star of v, denoted by ost(v), we mean the following subset of |K|:

ost(v) = {int( σ) : σ is a simplex of K with v as a vertex of σ}.
Example 3.3.2. As an example, consider the simplicial complex K, where |K|

is shown in Fig.3.10. Then
Fig. 3.10
st(v0 ) = {v0 , v0 , v1 },
st(v1 ) = K − {v0 , v2 , v3 , v4 , v2 , v3 , v3 , v4 }.
On the other hand, ost(v0 ) = v0 ∪ intv0 , v1 , and ost(v1 ) = v1 ∪ intv0 , v1 ∪

intv1 , v2 ∪ intv1 , v3 ∪ intv1 , v4 ∪ intv1 , v2 , v3 .
It is very important to point out that the open star ost(v) of a vertex
v of a simplicial complex K is not only a subset of |K|, but also is an open
set in |K|. This follows from the fact that for any vertex v of K, the function
fv : |K| → [0, 1] given by fv (x) = the v th barycentric coordinate of x, is
continuous, and fv−1 (0, 1] = ost(v).
In this section, we are going to show that if h : |K| → |L| is any given
continuous map, then there exists a representation |K | of |K|, (i.e., there is a
simplicial complex K such that |K | = |K|) and a simplicial map g : K → L
such that for each vertex v of K , h(ost(v)) ⊂ ost(g(v)), in particular |g| :
|K | = |K| → |L| is homotopic to h. Such a result is known as a simplicial
approximation theorem. We have the following
Definition 3.3.3. Let h : |K| → |L| be a continuous map. Then a simplicial map
g : K → L is said to be a simplicial approximation of h if for each vertex v of
K, h(ost(v)) ⊂ ost(g(v)).
We will show later that if a simplicial map g : K → L is a simplicial

approximation of h, then |g| is homotopic to h.
Example 3.3.4. Let K be any simplicial complex and L =Cl(σ p ), where σ p =

a0 , . . . , ap . Suppose h : |K| → |L| is any continuous map such that h(|K|)
does not intersect the face a1 , a2 , . . . , ap . Let g : K → L be the simplicial
map which maps all vertices of K to the vertex a0 of L. Then g is a simplicial
approximation of h because for each vertex v of K, h(ost(v)) ⊂ ost(g(v)).
The following result is easily verified:
Proposition 3.3.5. Suppose g1 : K → L is a simplicial approximation of f1 :

|K| → |L| and g2 : L → M is a simplicial approximation of f2 : |L| → |M |.
The g2 ◦ g1 : K → M is a simplicial approximation to f2 ◦ f1 .
Proof. The composite map g2 ◦ g1 is clearly simplicial. We know that
for each vertex v of K, f1 (ost(v)) ⊂ ost(g1 (v)), and for each vertex w
of L, f2 (ost(w)) ⊂ ost(g2 (w)). Since g1 (v) is a vertex of L, one has
f2 (ost(g1 (v))) ⊂ ost(g2 (g1 (v))). This means f2 ◦ f1 (ost(v)) ⊂ ost(g2 ◦ g1 (v)),
which proves the result.
We now proceed to develop necessary definitions and concepts for giving

a proof of the simplicial approximation theorem mentioned earlier.
Definition 3.3.6. Let K and L be two simplicial complexes and f : |K| → |L|
be a continuous map. Then we say that K is star-related to L with respect to
f , if for each vertex v of K, there is a vertex v of L such that
f (ost(v)) ⊂ ost(v ).
The above definition will help us in constructing a simplicial map. We will

also require the following result;
vn } of vertices of a simplicial complex K

Proposition 3.3.7. A set {v0 , v1 , . . . ,
n
forms a simplex of K if and only if i=0 ost(vi ) is nonempty.
3.3. Simplicial Approximation 105
Proof. If σ n = v0 , v1 , . . . , vn is a simplex of K, then weknow that

n
int(σ n ) ⊂ ost(vi ) foreach i = 0, 1, . . . n and so φ
= int(σ n ) ⊂ i=0 ost(vi ).
n
Conversely, suppose i=0 ost(vi )
= ∅ and let x be a point of this intersection.
For each i = 0, 1, . . . , n, there is a simplex σi in K such that x ∈ int(σi ) and vi
is a vertex of σi . Recall that the set of interiors of all simplexes of a simplicial
complex K forms a partition of |K| and hence, there is a unique simplex of
K whose interior contains x, i.e., σ0 = σ1 = σ2 = . . . = σn . This implies that
v0 , v1 , . . . , vn are vertices of the simplex σ0 .
We now start by first proving a basic special case of the Simplicial Ap-
proximation Theorem as follows:
Theorem 3.3.8. Let f : |K| → |L| be a continuous map. Suppose K is star-

related to L relative to f . Then there is a simplicial map g : K → L such that
g is a simplicial approximation of f . In particular, |g| : |K| → |L| is homotopic
to f .
Proof. Since K is star-related to L relative to f , we can define a map g from
the vertices of K to the vertices of L as follows: for each vertex v of K, let g(v)
be a vertex of L such that
f (ost(v)) ⊂ ost(g(v)).
We now prove that the map g actually takes a simplex of K to a simplex of
L. For this, let v0 , v1 , . . . , vn be the vertices of a simplex of K. By Proposition
3.3.7, this means
n
ost(vi )
= φ.
i=0
Hence,
n n n
∅
= f ( i=0 ost(vi )) ⊂ i=0 f (ost(vi )) ⊂ i=0 ost(g(vi )),
n
i.e., i=0 ost(g(vi )) is nonempty, and therefore g(v0 ), . . . , g(vn ) are vertices of
some simplex of L. This proves that g : K → L is a simplicial map, and so g is
a simplicial approximation of f .
Next, we want to show that |g| : |K| → |L| is homotopic to f . Let x ∈
|K| and let σ = v0 , v1 , . . . , vk be the simplex of smallest dimension which
contains x. Let vi be any vertex of σ. Note that x ∈ ost(vi ) which means
f (x) ∈ f (ost(vi )) ⊂ ost(g(vi )). Thus f (x) ∈ intg(v0 ), . . . , g(vk ). On the other
hand if we write
x = α0 v0 + . . . + αk vk , αi > 0, ∀i,
then |g|(x) = α0 g(v0 ) + . . . + αk g(vk ), in which some of the g(vi ) s may
be equal. In any case the coefficient of g(vi ) will be strictly positive, i.e.,
|g|(x) ∈ intg(v0 ), . . . , g(vk ). This means for each x ∈ |K|, both f (x) and
|g|(x) are in the same simplex of L. Since a simplex is a convex set, we can
define a map H : |K| × I → |L| by H(x, t) = (1 − t)f (x) + t|g|(x). H is

continuous because both f and |g| are continuous and |L| is a subset of some
Euclidean space. Therefore H is a homotopy from f to |g|.
It is easy to give an example of a continuous map f : |K| → |L| and a

simplicial map g : K → L such that |g| : |K| → |L| is homotopic to f , but g is
not a simplicial approximation of f .
As remarked earlier, the above theorem is a special case of the Simplicial

Approximation Theorem (to be proved later) because for any given continuous
map f : |K| → |L|, it may be the case that K is not star-related to L relative to
f . What are we to do in that case? The answer to this question lies in creating
the concept of subdivision of a simplicial complex. Geometrically speaking,
what we will do is as follows: We will subdivide each simplex σ of K in a set of
smaller simplices thereby getting a new simplicial complex K which will have
the same geometric carrier as K, i.e, |K | = |K|, and the simplexes of K will
be so small that K will be star-related to L relative to f .
3.4 Barycentric Subdivision – Simplicial Approximation

Theorem
The concept of subdividing a simplicial complex K is an extremely im-
portant and useful method of changing the simplicial structure of K without
changing the underlying set of |K| or its topology. As an example, consider the
2-simplex σ 2 = v0 , v1 , v2 . We know that the simplicial complex
Cl(σ 2 ) = {v0 , v1 , v2 , v0 v1 , v1 v2 , v0 v2 , v0 v1 v2 }
has the simplex σ 2 as its geometric carrier, i.e., |Cl(σ 2 )| = σ 2 . Now, if we
introduce a vertex v3 in the interior of the triangle σ 2 , and consider the set
K of all triangles, edges and vertices so created, then this new set K is again
a simplicial complex (Fig.3.11) which has the same geometric carrier as the
triangle σ 2 , i.e., |K | = σ 2 .
v0
`
HH
HH

v
* 3

HH

H
HH

`

` v2
H

`
v1
Fig. 3.11
3.4. Barycentric Subdivision – Simplicial Approximation Theorem 107
In fact, K has four vertices, six edges and three triangles and is entirely dif-
ferent from K because K had only three vertices, three edges and one triangle.
Imagine what will be K when the newly introduced vertex v3 lies on one of
the edges of the simplex σ 2 . If K is a simplicial complex and we subdivide each
simplex in the same manner as indicated above, then we get a new simplicial
complex K which is called a subdivision of K. Among infinitely many possibil-
ities of subdividing a simplicial complex, there is one type of subdivision, which
is iterative, and is called the barycentric subdivision. This subdivision stands
out as the most useful, highly regular and conveniently describable subdivision.
Definition 3.4.1. Let σ = v0 , v1 , . . . , vk be a k-simplex in Rn . Then the

barycentre of σ, denoted by σ̇, is defined as the point of σ given by

k
1
σ̇ = vi .
i=0
k+1
In other words, the barycentre σ̇ of the k-simplex σ k is that point of σ k whose

k
barycentric coordinates, with respect to each of the vertices of σ k , are equal.
For example, the barycentre of a 0-simplex v0 will be the point v0

itself. The barycentre of a 1-simplex v0 , v1 is simply the midpoint of the
line segment joining v0 and v1 . Similarly, the barycentre of the 2-simplex
σ 2 = v0 , v1 , v2 is the centroid of the triangle with vertices v0 , v1 and v2 , and
so on.
Let σ k = v0 , v1 , . . . , vk and K = Cl(σ k ). Consider the set of all barycen-

tres of all faces of σ k . Note that this new set contains all the vertices of σ k and
corresponding to each edge, one new vertex; corresponding to each 2-face, one
new vertex, and so on. We now consider a new simplicial complex K whose
vertices are barycentres of all faces of σ k and whose simplexes are as follows:
σ˙0 , σ˙1 , . . . , σ˙q will be a q-simplex if σi ’s are all distinct and σi is a face of σi+1
for i = 0, 1, . . . , q − 1. For example, in the case of 2-simplex σ 2 = v0 , v1 , v2 ,
the new simplicial complex K has vertices, edges and 2-simplexes as shown in
Fig. 3.12.
v0 (v0 )
` `
@ @
@ @
@ (v0 v1 )` @ ` (v0 v2 )
@ P PP @
@ -P
P
` @
(v0 v1 v2 )
PPP @
@ PP
` @` ` ` @`
P
v1 v2 (v1 ) (v1 v2 ) (v2 )

K = Cl(σ 2 ) K
Fig. 3.12
In fact, the starting simplicial complex is
K = {v0 , v1 , v2 , v0 , v1 , v1 , v2 , v0 , v2 , v0 , v1 , v2 },
and the new simplicial complex is K , whose vertices are barycen-

tres of each simplex of K. If we use the notation (v0 v1 ) to de-
note the barycentre of v0 v1 , etc., then K has the seven vertices:
(v0 ), (v1 ), (v2 ), (v0 v1 ), (v1 v2 ), (v0 v2 ), (v0 v1 v2 ). It has 1-simplexes such as
(v0 ), (v0 v2 ), (v1 ), (v0 v1 ), . . . , (v0 v1 ), (v0 v1 v2 ), . . . etc., and 2-simplexes such
as (v0 ), (v0 v1 ), (v0 v1 v2 ), (v1 ), (v1 v2 ), (v0 v1 v2 ), . . ., and so on. The Fig. 3.12
gives a nice description of the new simplicial complex K . It is obvious that
|K| = |K |. More generally, we have
Definition 3.4.2. Let K be a simplicial complex. Let K (1) be a simplicial complex

whose vertices are barycentres of all simplexes of K, and for distinct simplexes
σ0 , σ1 , . . . , σn of K, σ˙0 , σ˙1 , . . . , σ˙n is a simplex of K (1) if and only if σi is a
face of σi+1 for each i = 0, 1, 2, . . . , n − 1. Then K (1) is a simplicial complex
and is called the first barycentric subdivision of K. By induction, we define
nth barycentric subdivision K (n) of K to be the first barycentric subdivision of
K (n−1) for each n > 1. We also put K (0) = K for convenience.
Note that |K (n) | = |K| for each n ≥ 0. We also observe that when we
take the first barycentric subdivision K (1) of a positive-dimensional complex,
the simplexes of K (1) become strictly smaller in size than those in K. In fact,
their sizes approach towards zero when we go on making further and further
subdivisions. We now make these observations precise.
Recall that if A is a subset of Rn , then
diam(A) = sup{d(x, y) : x, y ∈ A}
n
is called the diameter of A; here d(x, y) = x − y = ( i=1 (xi − yi )2 )1/2 , where
x = (x1 , . . . , xn ) and y = (y1 , . . . , yn ).
Definition 3.4.3. Let K be a simplicial complex. We define the mesh of K,

denoted by mesh(K), as follows:
mesh(K) = max{diam(σ) : σ is a simplex of K}.
Note that if K is a 0-dimensional simplicial complex, then mesh(K) =

0 = mesh(K (1) ) = mesh(K (2) ) = . . .. First, we have
Lemma 3.4.4. Let σ be any positive-dimensional simplex. Then diam(σ) = v −

w for some pair of vertices v and w of σ. In other words, the diameter of a
simplex σ is the length of its largest edge.
q
Proof. Let σ = v0 , v1 , . . . , vq and x, y ∈ σ. Let y = i=0 bi vi , where bi are
barycentric coordinates of y. Then
q q
x − y = ( i=0 bi )x − i=0 bi vi
q
= i=0 bi (x − vi )
q
≤ i=0 bi x − vi

≤ bi · max{x − vi : 0 ≤ i ≤ q}
≤ max(x − vi : 0 ≤ i ≤ q).
Replacing x by vi and y by x, we find similarly
x − vi ≤ max{vj − vi : 0 ≤ j ≤ q}.
Hence, by combining the two, we find
x − y ≤ max{vj − vi : 0 ≤ j, i ≤ q}.
This evidently implies that diam (σ) = vr − vs for some vertices vr and vs
of the simplex σ.
Now, we can prove the fact which was intuitively mentioned earlier.
Theorem 3.4.5. Let K be any positive-dimensional simplicial complex. Then

limn→∞ mesh(K (n) ) = 0.
Proof. Let us assume that dim(K) = m. First, we compare the mesh (K (1) ) with
the mesh of K. Since mesh (K (1) ) is determined by the length of its largest 1-
simplex, we consider any 1-simplex σ̇, τ̇ of K (1) where σ, τ are simplexes of
K and σ is a proper face of τ = v0 , v1 , . . . , vp . We know that
1 p
τ̇ = vi ,
(p + 1) i=0
and so, because σ̇, τ̇ both are points of τ , by the earlier lemma, there is a vertex
v of τ such that
τ̇ − σ̇ ≤ τ̇ − v.

Then
τ̇ − σ̇ ≤ τ̇ − v

1
p
= ( vi ) − v
p + 1 i=0
1
p
= (vi − v)
p + 1 i=0
1
p
≤ vi − v
p + 1 i=0
1
≤ (p · mesh(K))
p+1
m
≤ · mesh(K),
m+1
since p ≤ m. This implies that mesh (K (1) ) ≤ m+1

m
· mesh (K). By induction,
therefore, we find that mesh(K ) ≤ ( m+1 ) mesh (K). Since
(n) m n
m n
lim = 0,
n→∞ m + 1
we have the desired result.
Now, we have all the necessary ingredients to prove our general
Theorem 3.4.6. (Simplicial Approximation Theorem). Let f : |K| → |L| be

any continuous map. Then there is a barycentric subdivision K (k) of K and a
simplicial map g : K (k) → L such that g is a simplicial approximation of f . In
particular, |g| : |K (k) | = |K| → |L| is homotopic to f .
Proof. Since our simplicial complexes are finite, |K|, |L| etc., are all compact
subsets of some euclidean space Rn . We have already proved in Theorem 3.3.8
that if K is star-related to L relative to f : |K| → |L| then there is a simplicial
map g : K → L such that |g| is homotopic to f . If K is not starr-related to L
relative to f , then we proceed as follows: since |L| is a compact metric space,
we apply Lebesgue’s Covering Lemma (Lemma 1.7.2) to the open covering
{ost(v) : v is a vertex of L} of |L| to get a Lebesgue number η > 0 so that
any open ball in |L| of radius η about any point of |L| is contained in ost(v)
for some vertex v of L. Since any continuous map from a compact metric
space to any space is uniformly continuous (see Proposition 1.7.3), the map
f : |K| → |L| is also uniformly continuous. This means that we can find a
positive number δ such that x − y < δ in |K| implies that f (x) − f (y) < η
in |L|. Now, we take k so large that the mesh of the barycentric subdivision
K (k) of K is less than δ/2. If v is a vertex of K (k) , then ost(v) lies in a
ball of diameter δ, which means f (ost(v)) lies in a ball of diameter η, i.e.,
f (ost(v)) ⊂ ost(v ) for some vertex v of L. Then K (k) is evidently star-related

to L relative to f : |K (k) | = |K| → |L| and the proof follows from Theorem
3.3.8.
In the proof of the above Simplicial Approximation Theorem, the necessity

of going to the barycentric subdivision K (k) of K is an important and indis-
pensable feature. Let us explain this by an example: consider the unit circle S1
and let K be the 1-skeleton of Cl(σ 2 ) where σ 2 = a0 , a1 , a2 is a 2-simplex.
Then we know that S1 is homeomorphic to |K| by some homeomorphism, say,
h : S1 → |K|. Now, let f, g : S1 → S1 be any two continuous maps. Then since
composite of homotopic maps is homotopic, we find that f is homotopic to
g if and only if hf h−1 is homotopic to hgh−1 . In other words, the set of all
homotopy classes of maps from S1 to itself is in 1-1 correspondence with the
set of all homotopy classes of maps from |K| to itself.
S1
f - S1
h h
? ?
|K| - |K|
h◦f ◦h−1
We also note that there is a bijection between π1 (S1 ) and the set [S1 , S1 ]
of all homotopy classes of maps from S1 to S1 . Since π1 (S1 ) ∼ = Z, we see
that there are infinite number of mutually non-homotopic maps from |K| to
itself, while for any k there are only finite number of simplicial maps from
|K (k) | to itself. The Simplicial Approximation Theorem says that for any
given map f : |K| → |K|, there is a k (depending on f ) and a simplicial map
g : K (k) → K which will approximate f ; it does not say that there is a fixed
k such that every f : |K| → |K| can be approximated by some simplicial map
g : K (k) → K. One may have to go to a higher order barycentric subdivision
of K for finding out a simplicial approximation to a given continuous map
f : |K| → |K|. This fact can also be expressed by asserting that for any
given k, there exists a continuous map f : |K| → |K| which does not have a
simplicial approximation from K (k) to K.
We now present an interesting application of the Simplicial Approximation

Theorem. Recall that a space X is said to be simply connected if X is path-
connected and π1 (X) is trivial. This is the same thing as saying that any map
from S0 to X and any map from S1 to X are null-homotopic. More generally,
we have
Definition 3.4.7. A space X is said to be n-connected (n ≥ 0) if for each k ≤ n,

any map from Sk to X is null-homotopic.
Thus, 0-connectedness is the path-connectedness of X and 1-

connectedness is the same thing as simply connectedness. Now, we have
Theorem 3.4.8. For each n ≥ 1, Sn is (n − 1)-connected.

Proof. Let 0 ≤ k < n. We will prove that any map f : Sk → Sn is null-
homotopic. Let σ k+1 be a (k + 1)-simplex and σ n+1 be a (n + 1)-simplex. If K
represents a k-skeleton of Cl(σ k+1 ) and L represents the n-skeleton of Cl(σ n+1 ),
then we know that |K| is homeomorphic to Sk and |L| is homeomorphic to Sn .
Let h : |K| → Sk and h : |L| → Sn be homeomorphisms.
|K| - |L|
h h
? ?
Sk
f - Sn
Let g : K (m) −→ L be a simplicial approximation to the map

−1
(h ) ◦ f ◦ h : |K| → |L|. Now observe that dim(K (m) ) = k < dim(L) = n
and any simplicial map g : K (m) → L cannot raise the dimension of the
simplicial complex. This means g cannot be onto and hence |g| : |K (m) | → |L|
cannot be onto. Consequently, there is a point a ∈ |L| such that |g| maps |K|
into |L| − {a}. Because |L| − {a} is homeomorphic to Rn and the latter is
contractible, |g| is null-homotopic. This means (h )−1 ◦ f ◦ h : |K| → |L| is
null-homotopic and, consequently, f is null-homotopic.
We have already proved (see Example 2.5.1) the next result, but it follows
from the above theorem also.
Corollary 3.4.9. The n-sphere Sn , n ≥ 2, is simply connected.
Concluding Remarks: The basic topological spaces dealt with in this chapter
are the compact polyhedra (finite simplicial complexes) which form the most
accessible class of topological spaces. They include all graphs, compact surfaces
and compact 3-manifolds. The real projective space RPn , and the complex pro-
jective space CPn , n ≥ 2 are classical examples of polyhedra. This restricted
class of topological spaces (polyhedra) has two beautiful generalizations: the
first one is to allow a simplicial complex to have infinite number of simplexes,
in which case the topology of its geometric carrier (still called a polyhedra)
is no longer compact (see Section 3.5). However, these are paracompact and
most of the results proved in this chapter are true for them also. The second
generalization, resulting in a more powerful and appropriate class of topological
spaces from the viewpoint of algebraic topology, is known as CW -complexes.
They carry the CW -structure which is a far-reaching generalization of the sim-
plicial structure discussed here, and they encompass a remarkably large class of
nice spaces. The simplicial approximation theorem remains valid for them also.
These CW-complexes are also paracompact and become compact if the CW -
structure is finite. Chapter 2 of James Munkres [16] (for simplicial complexes)
and the book of A.T. Lundell and S. Weingram [12](for CW -complexes) are
excellent references for a study of these spaces and their extensive applications.
Exercises
1. Let σ n = v0 , v1 , . . . , vn be n-simplex. Find a formula giving the number

of faces of σ n .
2. Prove that every compact convex subset of the Euclidean space Rn is
triangulable.
3. Prove that the following rectangular strip in which the left side ada is
identified with the right side ada, and the top side abca is identified with
the bottom side abca
a b c a
! !! ! !! ! !!
! ! !
d! ! !! ! !! !d
! ! ! ! ! !!
! ! !
a b c a
is not a triangulation of torus.

4. Show that the following strip in which the left side ba is identified with
the right side ab
b d a
! ! !!
!! !!
!
a! c b
is not a triangulation of Möbius band.

5. If we take a rectangle ABCD = I × I in plane and identify all points of
D C
A B
the side AD with the corresponding points of the side BC, and also iden-
tify all the points of the side AB of the type (x, 0) with the points (1−x, 1)
of the side DC, then the resulting space is called the Klein bottle K (see
Section 1.1.4). Give a triangulation of the Klein bottle. (Hint: Modify the
triangulation of torus).
6. Take a disc D2 = {(x, y) ∈ R2 : x2 + y 2 ≤ 1}. Identify a point (cos θ, sin θ)
of the boundary of D2 with the point (− cos θ, − sin θ). Then the resulting
quotient space (Section 1.1.4), denoted by RP2 , is called the projective

plane. Prove that RP2 is a polyhedron. (Hint: start with a suitable trian-
gulation of the disc itself.)
7. Let K = Cl(σ 2 ), where σ 2 = v0 , v1 , v2 is a 2-simplex. Show that the
open simplex v0 , v1 , which is a subset of |K|, is not an open set of |K|.
However, the open simplex v0 , v1 , v2 is indeed an open set in |K|.
8. Show that any r-skeleton K (r) , r ≥ 0, of a simplicial complex K is also a
simplicial complex.
9. Prove that Kuratowski graph (Fig. 3.6(a)) is not a planar graph, i.e., it
cannot be embedded in the plane R2 .
10. Prove that a simplicial complex K is connected if and only if |K| is path-
connected.
11. Let K be a simplicial complex. Prove that for each vertex v of K, ost(v)
is an open set of |K|. Also prove that {ost(v) : v is a vertex of K} is an
open covering of |K|.
12. Let f : |K| → |L| be a continuous map. If a simplicial map φ : K → L
is a simplicial approximation of f , then show that |φ| is homotopic to f .
Give an example to show that the converse need not be true.
13. Let S = {a0 , a1 , . . . , an } be a set with n + 1 distinct points. Consider the
set
n
F (S) = {f : S → [0, 1] : i=0 f (ai ) = 1}.
Prove that the set F (S) is in one-to-one correspondence with the points
of an n-simplex σ n = v0 , v1 , . . . , vn in Rn+1 . (This exercise tells that an
abstract simplex can be given a topology from a simplex of Rn via the
above one-to-one correspondence).
14. Prove that the following compact subspace X (Topologist’s Sine Curve)
of R2 cannot be a polyhedron:
X = {(x, y) ∈ R2 : y = sin 1/x, x > 0} ∪ {(0, y) : |y| ≤ 1}.
(Hint: a polyhedron is locally contractible.)

15. A set S = {a0 , . . . , ak }, k > n of points of Rn is said to be in general
position if any n + 1 points of S are geometrically independent. Show that
for any n, we can find an infinite set of points which are in general position
in Rn .
16. Let a1 , a2 , . . . , an+1 be points of Rn and let the coordinates of ai be written
as ai = (a1i , a2i , . . . , ani ) for each i. Prove that S = {a1 , a2 , . . . , an+1 } is
geometrically independent if and only if the matrix
3.5. General Simplicial Complexes 115
⎛ ⎞
a11 a21 . an1 1
⎜ a12 a22 . an2 1 ⎟
⎜ ⎟
⎜ . . . . . ⎟
⎜ ⎟
⎝ . . . . . ⎠
a1n+1 a2n+1 . ann+1 1
is nonsingular.
3.5 General Simplicial Complexes

So far we assumed that a simplicial complex K is just a finite set of
simplexes satisfying the condition that σ ∈ K implies all faces of σ are in K,
and σ, τ ∈ K imply that either σ ∩ τ = φ or σ ∩ τ is a face of both σ and
τ . A general simplicial complex K will be defined to be an arbitrary set of
simplexes satisfying the same two conditions as above. In this case, the number
of vertices of K can be of arbitrary cardinality and the dimension of K need
not be finite. Hence, we have to understand clearly where such a simplicial
complex will lie, and how to define its topology on |K|. This can be explained
by the idea of an abstract simplicial complex defined as follows:
Definition 3.5.1. An abstract simplicial complex K consists of a set {v} of

vertices and a set {s} of nonempty finite subsets of vertices, called simplexes,
such that
(a) any set consisting of exactly one vertex is a simplex, and
(b) every nonempty subset of a simplex is a simplex.
Now, let {vα | α ∈ J} be the set of vertices of K indexed by an arbitrary

set J. Consider the set
RJ = {f : J → R | f (vα ) = 0 for all α except finitely many}.
Then, clearly RJ is a vector space over R, and it is the direct sum of as many
copies of R as the cardinality of J. There is a natural basis {α | α ∈ J} of this
vector space defined by

1 if β = α
α (β) =
0 if β
= α.
We can define a metric on RJ such that

d(f, g) = (f (vα ) − g(vα ))2 ,
α∈J
which makes RJ a metric space. Note that corresponding to each finite sub-
set {α1 , . . . , αk } of basis elements, RJ has Rk as a subspace with Euclidean
topology. Observe that if (vα0 , vα1 , . . . , vαk ) is a k-simplex of K, then we may

consider the geometric k-simplex (α0 , α1 , . . . , αk ) spanned by the unit vectors
α0 , α1 , . . . , αk , which really consists of all points f of RJ such that

k
f (vαi ) = 1, 0 ≤ f (vαi ) ≤ 1 ∀i.
0
This is indeed the closed simplex of the Euclidean space generated by the
vertices (α0 , α1 , . . . , αk ). Now, let K be the set of all geometric simplexes in
RJ defined by all simplexes of K, and let |K| be their union in RJ . Then, |K|
is called the geometric realization of the abstract simplicial complex K. Note
that K has arbitrary number of vertices and simplexes of any finite dimension.
Also, K is now a general simplicial complex satisfying both the conditions of
a simplicial complex. Since |K| ⊂ RJ , it has the subspace topology, which is a
metric topology. We denote the resulting space by |K|d .
Topology of a Polyhedron
There is yet another and more important topology on |K|. To see this,
note that the geometric realization (α0 , α1 , . . . , αk ) of each simplex s of K
is a compact subspace of the Euclidean subspace Rk and, therefore, it has a
unique topology of its own. Hence, we can consider the weak topology on |K|
defined by these closed simplexes. We call this topolgy on |K| as the coherent
topology on |K|. The resulting space, denoted by just |K|, is called a general
simplicial complex or a polyhderon. Thus, the topology of a polyhedron |K| is
the largest topology on |K| which makes all inclusions i : (α0 , α1 , . . . , αk ) →
|K| continuous. In other words, a subset F of |K| is closed iff F ∩ σ is closed
in every geometric simplex σ of |K|.
Definition 3.5.2. A topological space X is said to be triangulable (or a polyhe-

dron) if there exists a simplicial complex K and a homeomorphism f : |K| →
X, where |K| has the coherent topology defined above.
Example 3.5.3. Let K = {n, n, n + 1 | n ∈ Z}. The vertices of K are all

integers and the closed intervals [n, n + 1] are all its 1-simplexes. Then, clearly,
K is an infinite simplicial complex and |K| is homeomorphic to the real line R
with usual topology under the identity map i : |K| → R. The two topologies
on |K| are indeed identical. If F is closed in R, then, clearly, F ∩ [n, n + 1] is
closed in [n, n + 1] for each n, and so is closed in the coherent topology on |K|.
Conversely, if F ⊂ |K| such that F ∩ [n, n + 1] is closed in [n, n + 1] for all n,
then, since F = n∈Z F ∩ [n, n + 1] is a locally finite family of closed sets of R,
F itself is closed in R. This shows that the real line R is triangulable, and K
is a triangulation of R.
Example 3.5.4. By a similar argument, one can easily see that the following
figure gives a triangulation of R2 :
Y 6
(3,2)
(3,1)
X
(0,0) -
Fig.3.13
More generally, we can show that the Euclidean space Rn , n ≥ 1, is also

triangulable. To see this, consider the points of Rn having integral coordinates,
i.e., the set Zn . Define x ≤ y, x = (x1 , x2 , . . . , xn ), y = (y1 , y2 , . . . , yn ) ∈ Zn
if xi ≤ yi ∀ i = 1, 2, . . . , n. Then, clearly, Zn is a partially ordered set. Let
us take points of Zn as vertices, and let a set {x0 , x1 , . . . , xq } of vertices be
a simplex if x0 ≤ x1 ≤ . . . ≤ xq is a totally ordered chain, i.e., xqi − x0i = 0
or 1 for each i. Since a subset of totally ordered set is totally ordered, this
construction gives a simplicial complex K. Moreover, since the set of simplicies
of K is a locally finite family of closed sets of Rn , K gives a triangulation of
Rn . Thus, Rn is a polyhedron ∀ n ≥ 1.
From this latter example, one can observe that any partially ordered set
(P, ≤) always defines a simplicial complex in a natural way as described above.
Remark 3.5.5. Suppose |K| ⊂ Rn for some n. Then, |K| has two topologies:
one, the subspace topology induced from Rn which is a metric topology, and
the other is the coherent topology defined by all simplices of K. We observe
that in this case the coherent topology on |K| is always finer than the subspace
topology since F is closed in Rn means F ∩ σ will be closed in σ for all sim-
plexes σ of K, i.e., F ∩ |K| will be closed in the coherent topology of |K|. The
following example shows that the coherent topology can be strictly finer than
the subspace topology.
Example 3.5.6. Let K be a simplicial complex whose vertices are

{(0, 0), (1, 1), (1, 1/2), . . . , (1, n1 ), . . .)} ∪ {(1, 0)} in the plane R2 , and the edges
are the line segments joining (0, 0) to (1, n1 ) for all n ∈ N.
(1, 1)
(1, 12 )

..

.

(0, 0) (1, 0)
Fig. 3.14
Then, note that the set F = {(1, 1), (1, 12 ), . . . , (1, n1 ), . . .} is closed in |K| since
its intersection with each simplex is either empty or just a point and so is closed
in that simplex. However, F is not closed in the subspace topology on |K| in-
duced from the plane R2 because (1, 0) is the limit point of F which is not in F .
The above example also shows that a polyhedron |K| need not be metriz-
able. Suppose, on the contrary, the above |K| is metrizable. Then, |K| must be
first countable at the vertex v = (0, 0). Let {U1 , U2 , . . . , Un , . . .} be a countable
base for |K| at the point v. We can clearly assume that
U1 ⊇ U2 ⊇ . . . ⊇ Un ⊇ . . .
Note that v is a limit point of the open simplex ((0, 0), (1, n1 )) for each n ≥ 1.
Hence, we can choose a point an ∈ Un ∩ ((0, 0), (1, n1 )) ∀ n ∈ N. Then, we
get a sequence {a1 , a2 , . . . , an , . . .} in |K| which converges to v since in any
neighbourhood V of v there is a Un ⊆ V such that ai ∈ Un ∀ i ≥ n. But, the
sequence B = {a1 , a2 , . . . , an , . . .} is a closed set in |K| since its intersection
with each simplex of K is just a point and so is closed in that simplex. Thus, B
contains all of its limit points, a contradiction to the fact that B converges to v.
A very useful result about the coherent topology on |K| is the following:
Theorem 3.5.7. Let X be a topological space and f : |K| → X be a map. Then

f is continuous iff f |σ is continuous for each simplex σ of K.
Proof. Since restriction of a continuous map is continuous, it is enough to prove
the converse part. Hence, suppose f |σ is continuous for every σ ∈ K. Let F be
a closed set of X. Then, for each σ,
( f |σ )−1 (F ) = f −1 (F ) ∩ σ
is closed in σ. Hence, by the definition of coherent topology on |K|, f −1 (F ) is

closed in |K|. This means f is continuous.
Proposition 3.5.8. Let L be a subcomplex of K. Then |L| is a closed subspace

of |K|.
Proof. We should prove two facts. First, if F is a closed subset of |K|, then
F ∩|L| is closed in |L|; second, if A is a closed subset of |L|, then A = F ∩|L| for
some closed set F of |K|. First is obvious by definition of coherent topologies of
|K| and |L|. To prove the second, we will prove a stronger result, viz., A ⊂ |L|
is closed in |L| implies A is closed in |K|. Let σ ∈ K be any simplex. Then, for
a face τ of σ, which lies in L, A ∩ τ , is closed in τ and hence in σ. Thus A ∩ σ
is the finite union of such sets A ∩ τ , where τ is a face of σ. Being a finite union
of closed sets of σ, A ∩ σ is closed in σ. It follows that A is closed in |K|.
Theorem 3.5.9. A polyhedron |K| is Hausdorff.

Proof. Let us recall the barycentric coordinates of a point x ∈ |K|. Since x
belongs to the interior a unique simplex σ = v0 , . . . , vk , its barycentric coor-
dinates with each vertex of σ are positive, whereas its barycentric coordinates
with any other vertex of K is zero. Now, let v be a fixed vertex of K. Then, we
can define a function fv : |K| → [0, 1] such that fv (x) is equal to the barycen-
tric coordinates of x with respect to the vertex v. Note that fv |σ (x) is the v th
barycentric coordinate of x if x belongs to σ and is zero if x is not in σ. Hence,
fv |σ is continuous for each σ ∈ K. Therefore, fv is a continuous function. Now,
take two distinct points x, y ∈ |K|. Then, there exists a vertex v of K such
that fv (x)
= fv (y). Now, choose a real number r such that fv (x) < r < fv (y).
Then, clearly, the open sets fv−1 ([0, r)) and fv−1 ((r, 1]) of |K| are disjoint and
separate x and y. Hence, |K| is Hausdorff.
Proposition 3.5.10. A polyhderon |K| is compact iff K is finite. More generally,

if a subset A ⊂ |K| is compact then A ⊆ |L| for some finite subcomplex L of
K.
Proof. If K is finite, then since |K| is the union of finite number of simplexes
and each of these simplexes is compact, |K| is compact. Conversely, suppose
a compact subset A is not contained in a finite subcomplex of K. This means
there will be an infinite number of simplexes {σ1 , σ2 , . . . , σn , . . .} such that
A ∩ σi
= φ ∀ i. We choose a point xi ∈ A ∩ σi ∀ i and consider the set
B = {xi | i ∈ N}. Then, since B intersects each simplex only in finite number
of points, B ∩ σi is closed in σi . Hence, B is closed in A. In fact, every subset
of B is closed in A, which means B is an infinite discrete closed subset of A
having no limit point, a contradiction to the compactness of A.
Theorem 3.5.11. A polyhedron |K| is normal.

Proof. We will use the Tietze extension theorem to prove the normality of |K|.
Let A ⊂ |K| be a closed set of |K| and f : A → [0, 1] be a continuous function.
We will extend f continuously to the whole of |K|. For this, it is enough to
define a family {fσ | σ ∈ K} of continuous maps fσ : σ → [0, 1] such that
(a) fσ |τ = fτ whenever τ is a face of σ, and

(b) fσ |A∩σ = f |A∩σ for each σ.
Because then, we can define a map g : |K| → [0, 1] by the condition that
g|σ = fσ ∀ σ ∈ |K|. This map g will be a continuous map by Proposition 3.5.7
which will extend the given map.
We define fσ by induction on dimension of σ. If σ is a vertex, say {v},
then we define fσ (v) = f (v) if v ∈ A, and fσ (v) to be any point in [0, 1] if
v ∈/ A. Thus, fσ is defined for all 0-simplexes. Now, suppose n > 1 and fσq
is defined for all q < n which satisfies both the conditions (a) and (b). Let σ
be an n-simplex and consider the set σ ∩ A ∪ σ̇. This is a closed set of σ, and
fτ is defined for all faces τ of σ̇ (boundary of σ) satisfying (a) and (b). So we
have a continuous map f : σ ∩ A ∪ σ̇ → [0, 1] such that f |τ = fτ for all faces
τ of σ̇ and f |σ∩A = f . Now, using Tietze extension theorem, we can extend
f to a continuous map fσ : σ → [0, 1] satisfying conditions (a) and (b). This
proves the theorem.
The following result yields an example of a compact topological space

which can not be triangulated.
Theorem 3.5.12. Any polyhedron |K| is locally contractible.

Proof. Let us take a point x ∈ |K|. Assume that U is a neighbourhood of x in
K. We must show that there is an open neighbourhood V of x in U such that
V can be deformed to the point x in U . Let us assume that x = v is a vertex
of K. Then, we know that ost(v) = A is an open set in |K|, and we have a
homotopy H : A × I → |K| defined by
H(a, t) = (1 − t)a + tv.
This homotopy deforms A into the point v such that H(v × I) = {v}. Since
U is a neighbourhood of v, H is continuous and I is compact, there exists a
neighbourhood V of v in A such that H(V × I) ⊂ U . Since V is open in A and
A is open in |K|, V is an open neighbourhood of v which is deformable to the
point v in U . This proves that |K| is locally contractible at each vertex of |K|.
Next, suppose that x is not a vertex. Let σ be a simplex of smallest
dimension such that x ∈ int σ. Now, we take any subdivision K of σ̇. Then
K ∗ x is a subdivision of σ̄ and hence, we get a subdivision K of K such that
x is a vertex of K . Since |K | = |K|, and |K | is locally contractible at its
vertex x, we conclude that |K| is locally contractible at x also.
Corollary 3.5.13. The topologist sine curve

T = {(0, y) | |y| ≤ 1} ∪ {(x, sin x1 ), x > 0}
in the plane R2 is not triangulable since it is not locally path connected (a
locally contractible space must be locally path connected).
Theorem 3.5.14. Any polyhedron |K| is paracompact.

Proof. The proof will be based on the famous theorem of A.H. Stone which says
that every metric space is paracompact. We already know that the identity map
i : |K| → |K|d , where |K| has the coherent topology and |K|d has the metric
topology, is continuous. Let U be an open cover of |K|. We will prove that
U has a nbd-finite open refinement. Note that open stars of vertices of K are
always open in |K|.
First, we show that K has a subdivision Sd(K) such that the closed stars
of vertices of Sd(K) in |Sd(K)| = |K| are contained in some member of U.
We will do this by a clever induction on the nth skeleton K (n) of K for each
n ≥ 0. For each vertex v ∈ K (0) , we choose a member U (v) ∈ U such that
v ∈ U (v). Let v0 , v1 be a 1-simplex of K. By the Lebesgue covering lemma,
we can find an > 0 such that any closed interval of length is contained
in some member of U. By subdividing the interval into uniform mesh of size
/2, we get a subdivision K of v0 , v1 such that closed stars of vertices of K
lying in the open interval (v0 , v1 ) are contained in some member of U, while
closed star of vertices in K (0) are contained in the already chosen members of
U . Subdividing each 1-simplex as above, we find a subdivision K of K (1) such
that closed stars of vertices of K are contained in some member of U. This
starts the induction at n = 1.
We suppose that n > 1, and K (n−1) has a subdivision K such that closed
star of any vertex of K is contained in a member of U. Let σ n be an n-simplex,
and consider the boundary σ̇ n of σ n . This boundary, being contained in K (n−1) ,
has already been subdivided such that closed vertex stars of vertices of K ∩ σ̇ n
are contained in some member of U . For each vertex p of K ∩ σ̇ n , let us choose
a member U (p) of U such that closed star of p is contained in U (p). Let b be
the barycentre of σ, and consider the cone with vertex b over the boundary of
the new subdivision of σ̇ n induced by K . If p is such a vertex of σ̇ n , then note
that closed vertex star of p is contained in some member of U . Hence, there
is an > 0 such that the -nbd of that closed star of p is also contained in
the same member of U. Now, we take the barycentric subdivision of all open
simplexes contained in the above cone. Note that this subdivision does not
affect the subdivision of σ̇ n already obtained. Repeating this process of taking
barycentric subdivisions of open simplexes in σ n , we can find an > 0 such
that simplexes of this new subdivision of diameter < /2 are contained in some
member of U. Combining both subdivisions given above, we can find an > 0
such that closed vertex stars of new vertices are contained in some member of
U while the closed vertex stars of vertices p lying on the boundary of σ n are
contained in members U (p) of U already chosen. Thus, the new subdivision of
σ n does not disturb the subdivision of σ̇ n already obtained. We do this for each
σ n and then conclude that the nth skeleton K (n) has a subdivision K such
that the closed stars of all vertices of K are contained in some member of U.
Hence, by induction, it follows that we can find a subdivision Sd(K) of K such
that closed stars of vertices of Sd(K) form a refinement of the open cover U.
Now, consider the open stars {ost(v )} of all the vertices v of Sd(K) in
|K|d . These open stars are also open in |Sd(K)|d since the coordinate functions
fv : |Sd(K)|d = |K|d → [0, 1] are continuous functions. Since |Sd(K)|d is
paracompact, this covering has a locally finite open refinement, say V = {V }
in |K|d . Then, clearly, {i−1 (V )} is a locally finite open refinement of U in |K|.
Exercises
1. Prove that the subspace (0, 1] of the real line R is a triangulable space.
2. Let v be a vertex of any simplicial complex K. Show that ost(v) and ost(v)
are path-connected in |K|.
3. Let s be a simplex. Prove that the closed simplex |s| is homeomorphic to
the cone with base |Bd(s)|, i.e., |s| is homeomorphic to |Bd(s)| ∗ w.
4. Let U be a bounded open subset of Rn and suppose U is star-convex with
respect to the origin. Show that
(a) a ray from origin may intersect Bd(U ) in more than one point.
(b) there is an example of a U such that (U ) need not be homeomorphic
to the disc Dn .
5. A simplicial complex K is said to be locally finite if each vertex of K
is the vertex of only finitely many simplexes of K. If K is locally finite,
prove that every point of |K| has a nbd of the form |L|d where L is a finite
subcomplex of K.
6. Let K be a simplicial complex. Show that the following conditions are
equivalent:
(a) K is locally finite.
(b) |K| is locally compact.
(c) |K| → |K|d is a homeomorphism.
(d) |K| is metrizable.
(e) |K| satisfies the first axiom of countability.
7. Prove that every polyhedron |K| is a perfectly normal space.
8. Suppose K is a countable simplicial complex which is locally finite and
dim(K) ≤ n. Then prove that |K| can be linearly embedded in R2n+1 .
9. Show that the assignment K → |K| is a covariant functor from the
category of simplicial complexes and simplicial maps to the category
of topological spaces and continuous maps. More generally, show that
(K, L) → (|K|, |L|) is a covariant functor from the category of simpli-
cial pairs and simplicial maps to the category of topological pairs and
continuous maps of pairs.
10. Let (X, A) be a polyhedral pair. Prove that A is a strong deformation
retract of some nbd of A in X.
Chapter 4
Simplicial Homology
4.1 Introduction
“Homology groups” associated to a given simplicial complex K, consti-

tute the first comprehensive topic of the subject of algebraic topology. In this
chapter, we will explain how we can associate a sequence of abelian groups
{Hn (K) : n ≥ 0} to a given simplicial complex K. These groups, called ho-
mology groups of the simplicial complex K, will have interesting functorial
properties, viz., for each simplicial map f : K → L, there will be an induced
group homomorphism (f∗ )n : Hn (K) → Hn (L) for each n ≥ 0 satisfying the
following two properties:
(i) If f : K → L and g : L → M are two simplicial maps, then for each n ≥ 0,
((g ◦ f )∗ )n = (g∗ )n ◦ (f∗ )n : Hn (K) → Hn (M ).
(ii) If IK : K → K is the identity map, then for each n ≥ 0, the induced map
((IK )∗ )n is the identity map on Hn (K).
These two functorial properties, which are usual for any kind of homology
groups, yield the result that any simplicial isomorphism from K to L induces
isomorphisms between their homology groups Hn (K) and Hn (L) for each
n ≥ 0. Therefore, the homology groups Hn (K) of the simplicial complex
K are invariants of the simplicial structure of the complex K. This is a
typical phenomenon with most of the algebraic invariants which an algebraic
topologist tends to define and study. The simplicial homology groups happen
to be one of the first classical algebraic objects to be invented and studied.
The most important and interesting feature of these groups is that they lend
themselves to easy computations.
Let X be a polyhedron and K1 , K2 be two triangulations of X. We will

finally prove that Hn (K1 ) is isomorphic to Hn (K2 ) for each n ≥ 0. This allows
124 Chapter 4. Simplicial Homology
us to define the homology groups {Hn (X) : n ≥ 0} of a polyhedron X simply

by declaring Hn (X) to be Hn (K), where K is some triangulation of X. This
means that the simplicial homology groups of a simplicial complex K are not
only the invariants of simplicial structure of K, but more importantly, these
are topological invariants of the space |K|. In particular, we will conclude that
the Euler-characteristic of a polyhedron X, viz., (# of vertices − # of edges
+ # of triangles − . . .) in a triangulation of X is a topological invariant. By a
somewhat lengthy technical procedure, we will also show that any continuous
map f : X → Y between two polyhedra X = |K|, and Y = |L| induces group
homomorphisms (f∗ )n : Hn (K) → Hn (L) for each n ≥ 0, which satisfy the
analogues of the two functorial properties mentioned earlier. One of the first
uses of these homology groups will be made in solving some “classification
problems” of point set topology. For instance, it will be shown that two spheres
Sm and Sn are homeomorphic if and only if m = n, with a similar result
for Euclidean spaces Rm and Rn . We will present several other interesting
applications of these homology groups and induced homomorphisms. Detailed
proofs of the famous theorems like Brouwer’s Fixed Point Theorem, Lefschetz
Fixed Point Theorem, Borsuk-Ulam Theorem, etc. will be given. The question
of existence of tangent vector fields on spheres will also be answered.
4.2 Orientation of Simplicial Complexes
Let us consider an n-simplex σ n = v0 , v1 , . . . , vn . Suppose the vertices

n
of σ are ordered by declaring v0 < v1 < v2 < . . . < vn . Then this ordering
determines a certain direction among the vertices of σ n . Any other arrangement
of these vertices gives rise to a permutation of vertices, which is either an even
permutation or an odd permutation of the given ordering. When we consider
the simplex σ n together with the equivalence class of all even permutations of
its vertices, we say that σ n is positively oriented and write the pair as +σ n .
On the other hand, when σ n is considered with the equivalence class of all
odd permutations, we say that σ n is negatively oriented and we write this as
−σ n . Thus, by an oriented simplex σ n , we mean the simplex σ n together with
a certain order of its vertices determining the positive orientation. The same
simplex σ n , when we change the ordering of its vertices by a transposition,
will have positive orientation with respect to the new ordering whereas it has
negative orientation with respect to the earlier ordering.
As an example, consider the 1-simplex σ 1 = v0 , v1 . If we order v0 < v1 ,
then +σ 1 is v0 , v1 and −σ 1 is v1 , v0 . The two simplexes +σ 1 and −σ 1 are
now different 1-simplexes, one having the direction from v0 to v1 and the other
from v1 to v0 .
q +σ 1
- q q −σ
1
q
v0 v1 v0 v1
4.2. Orientation of Simplicial Complexes 125
As another example, consider the 2-simplex σ 2 = v0 , v1 , v2 (Fig. 4.1).
If we order the vertices as v0 < v1 < v2 , then +σ 2 = v0 , v1 , v2 =

v1 , v2 , v0 = v2 , v0 , v1 and −σ 2 = v1 , v0 , v2 = v0 , v2 , v1 = v2 , v1 , v0 . In
fact, orienting a simplex means fixing the positive direction of its vertices, and
then the negative direction is automatically fixed. Moreover, there are only two
directions which can be assigned to its vertices.
v0 v0
v1 v2 v1 v2
Fig. 4.1
A 0-simplex v0 is always taken to be positively oriented, by assumption.

Note that if an n-simplex σ n = v0 , v1 , . . . , vn is oriented by the ordering
v0 < v1 < . . . < vn , then all of its faces are automatically oriented by the
ordering induced by the above ordering. We have
Definition 4.2.1. A simplicial complex K is said to be oriented if each of its

simplexes is assigned an orientation.
There are several ways of orienting a simplicial complex K. As remarked

a while ago, if we assign a certain ordering to the set of all vertices of K, then
that ordering will induce an ordering in the vertices of every simplex of K.
Consequently, each simplex is automatically assigned an orientation and the
complex K becomes an oriented complex. Another way of orienting a complex
K is to assign an orientation to each of its simplexes arbitrarily. When we
orient by the latter method and a simplex τ of K is a face of a simplex σ of
K, it might happen that the orientation on τ induced by σ may coincide or
may be opposite to the original orientation of τ .
It may be mentioned that assigning an orientation to a simplicial complex

K is a preparation towards the main objective of associating some algebraic
structure to the complex K. We also note that the following definition depends
on the orientation of K.
Definition 4.2.2. Let K be an oriented simplicial complex and σ p , σ p+1 be two

simplexes whose dimensions differ by 1. With each such pair (σ p+1 , σ p ), we
associate its incidence number, denoted by [σ p+1 , σ p ], as follows: If σ p is not a
face of σ p+1 , we put [σ p+1 , σ p ] = 0. If σ p is a face of σ p+1 , we label the vertices
of σ p so that +σ p = v0 , v1 , . . . , vp . Let v be the additional vertex of σ p+1 .
Then v, v0 , . . . , vp is either +σ p+1 or −σ p+1 . We define

1 if v, v0 , . . . , vp = +σ p+1
[σ p+1 , σ p ] =
−1 if v, v0 , . . . , vp = −σ p+1 .
Example 4.2.3. Let the closure of σ 2 , where σ 2 = v0 , v1 , v2 , be oriented by

the ordering v0 < v1 < v2 . Then
[σ 2 , v0 , v1 ] = +1 [v0 , v1 , v1 ] = +1
[σ 2 , v1 , v2 ] = +1 [v0 , v1 , v0 ] = −1
[σ 2 , v0 , v2 ] = −1 [v0 , v1 , v2 ] = 0.
The following result on these incidence numbers is of crucial importance:
Theorem 4.2.4. Let K be an oriented simplicial complex. If σ p−2 is a

(p − 2)-face of a simplex σ p of K, then
p p−1 p−1 p−2
[σ , σ ][σ ,σ ] = 0,
where the summation is over all (p − 1)-simplexes σ p−1 of K.
Proof. Let v0 , v1 , . . . , vp−2 be the vertices of σ p−2 so that +σ p−2 =
v0 , . . . , vp−2 represents the positive orientation of σ p−2 . Let the two addi-
tional vertices of σ p be a and b. Without any loss of generality, we can assume
that +σ p = a, b, v0 , . . . , vp−2 . There are only two (p−1)-simplexes of K which
yield the nonzero terms in the above summation, viz.,
σ1p−1 = a, v0 , . . . , vp−2 , σ2p−1 = b, v0 , . . . , vp−2 .
The indicated ordering of the vertices of σ1p−1 (respectively σ2p−1 ) may give
either the positive orientation or negative orientation of σ1p−1 (respectively
σ2p−1 ). Thus, there are four distinct possibilities which we must examine:
Case I. Suppose +σ1p−1 = a, v0 , . . . , vp−2 and +σ2p−1 = b, v0 , . . . , vp−2 . Then
it is easily seen that
[σ p , σ1p−1 ] = −1, [σ1p−1 , σ p−2 ] = +1
[σ p , σ2p−1 ] = +1, [σ2p−1 , σ p−2 ] = +1.
Hence the two nonzero terms in the summation are −1 and +1, respectively,
i.e., the summation is zero.
Case II. Suppose +σ1p−1 = a, v0 , . . . , vp−2 and −σ2p−1 = b, v0 , . . . , vp−2 .
Then
[σ p , σ1p−1 ] = −1, [σ1p−1 , σ p−2 ] = +1
[σ p , σ2p−1 ] = −1, [σ2p−1 , σ p−2 ] = −1
and the desired summation is again zero.
Case III. Suppose −σ1p−1 = a, v0 , . . . , vp−2 and +σ2p−1 = b, v0 , . . . , vp−2 .
Then
4.3. Simplicial Chain Complex and Homology 127
[σ p , σ1p−1 ] = +1, [σ1p−1 , σ p−2 ] = −1

[σ p , σ2p−1 ] = +1, [σ2p−1 , σ p−2 ] = +1
and the summation is zero in this case also.
Case IV. Finally, suppose −σ1p−1 = a, v0 , . . . , vp−2 and −σ2p−1 =

b, v0 , . . . , vp−2 . Then
[σ p , σ1p−1 ] = +1, [σ1p−1 , σ p−2 ] = −1

[σ p , σ2p−1 ] = −1, [σ2p−1 , σ p−2 ] = −1
and the summation is again zero.
4.3 Simplicial Chain Complex and Homology
Let K be a simplicial complex with a fixed orientation. Let S̃q denote the
set of all oriented q-simplexes of K. Since each q-simplex, q ≥ 1, can be oriented
in exactly two distinct ways, the number of elements in the set S̃q is twice the
number of q-simplexes in K, whereas the number of oriented 0-simplexes is the
same as the number of vertices of K because 0-simplexes are assumed to be
always positively oriented.
Definition 4.3.1. Let 0 ≤ q ≤ dim K and Z be the additive group of integers.

Any map f : S̃q → Z with the property that if q ≥ 1, then f (−σ q ) = −f (σ q ) for
each σ q ∈ S̃q , is called a q-chain of K. For q = 0, a 0-chain is just a mapping
from the set of all 0-simplexes of K to Z. The set of all q-chains of K is denoted
by Cq (K). If q < 0 or q > dim K, we define Cq (K) = 0.
It is easy to see that the set Cq (K) of all q-chains is an abelian group with
point-wise operations, viz.,
(f + g)(σ q ) = f (σ q ) + g(σ q )
for all σ q ∈ S̃q . The zero map 0 : S̃q → Z defined by 0(σ q ) = 0 for all q-simplexes
σ q is the additive identity, and the inverse −f of f is defined by
(−f )(σ q ) = −f (σ q ).
The group Cq (K) is called the q-dimensional chain group of K. If we

imagine the collection of all chain groups Cq (K) arranged in descending order,
we have an infinite sequence of abelian groups
C(K) : . . . , 0, Cn (K), Cn−1 (K), . . . , Cq (K), . . . , C0 (K), 0, . . . ,
where all Cq (K) are zero except for q = 0, 1, 2, . . . , n = dim(K).

We can give a nice description of the elements of Cq (K) which is easy to

work with. For each positively oriented q-simplex σ q , define a q-chain σ̄ q (this
notation is only temporary) as follows:
⎧
⎨ +1, if τ q = σ q
q q
σ̄ (τ ) = −1, if τ q = −σ q
⎩
0, otherwise.
Then σ̄ q is indeed a q-chain, and is called an elementary q-chain of K. The next

result shows that Cq (K) is a free abelian group generated by all elementary
q-chains. Let Sq denote the set of all positively oriented q-simplexes of K.

Proposition 4.3.2. For each q ≥ 0, Cq (K) = Z · σ̄ q , σ q ∈ Sq .
Proof. It suffices to show that each f ∈ Cq (K) can be written uniquely as an
integral linear combination of the elementary q-chains σ̄ q . Suppose f (σ q ) = nq ,
where
σ q ∈ Sq . Then, by definition, f (−σ
q
) = −nq . Consider the element
q q
σ ∈Sq
q n q σ̄ . We claim that f = n q σ̄ . To see this, we must show that
f and nq σ̄ q both have the same value at each oriented q-simplex of K. If
σ q ∈ Sq , then
( nq σ̄ q )(σ q ) = nq σ̄ q (σ q ) = nq = f (σ q ).
If σ q ∈
/ Sq , then −σ q ∈ Sq and
f (σ q ) = f (−(−σ q ))
= −f (−σ q ), by definition

= −( nq σ̄ q )(−σ q ), by what we proved earlier

= ( nq σ̄ q )(σ q ).

This
proves ourclaim. To prove the uniqueness, suppose
f = nq σ̄ q =
mq σ̄ . Then (nq − mq )σ̄ = 0 in Cq (K) and so ( (nq − mq )σ̄ )(σ q ) = 0
q q q
in Z. However, this means nq − mq = 0. Since σ q is arbitrary, we find that the

corresponding coefficients of σ̄ q in the two expressions of f are identical.
If we had a different orientation of the complex K, we would have ended

up with a different set of generators for Cq (K), viz., the new generators would
be the same σ̄ q if the orientation of the simplex σ q remains the same and
would be −(σ̄ q ) if σ q gets the opposite orientation. In either case, however, the
group Cq (K) would be isomorphic to the direct sum of as many copies of Z as
there are q-simplexes in K. This also says that the group stucture of Cq (K)
depends on the simplicial complex K, not on the particular orientation of K.
The elementary q-chains σ̄ q are functions from the set of all oriented
q-simplexes of K to the additive group Z of integers. We could as well consider
q-chains as functions from the set of all oriented q-simplexes to any abelian
group G. In that case, the chain groups, denoted by Cq (K; G), would be
isomorphic to the direct sum of as many copies of G as there are q-simplexes

in K. The group G is called the coefficient group for the chain groups. In this
terminology, Cq (K) is really Cq (K; Z). We will have occasion to use coefficients
other than the group Z of integers, and sometimes their use will be preferable.
It is natural to ask how the groups Cq (K) = Cq (K; Z) and Cq (K; G) are
related. The answer is provided in terms of the so called tensor product, viz.,
C(K, G) = C(K) ⊗ G, an important well-known algebraic concept.
The elementary q-chain σ̄ q depends on the q-simplex σ q and its ori-

entation. If we fix an orientation for each q-simplex
σ q , then the group
Cq (K) = σq ∈Sq Z · σ̄ q is isomorphic to the group Z · σ q , where Z · σ q is the
infinite cyclic group generated by the simplex σ q . Since the elements of the
latter group are simply the formal integral linear combinations of q-simplexes
σ q , rather than elementary chains σ̄ q , there is no loss of generality if we
identify σ̄ q with σ q = 1 · σ q and write Cq (K) = ⊕σq ∈Sq Z · σ q . This is what
we shall do from now onwards and drop the temporary notation σ̄ q used earlier.
One might ask at this stage the following question: The q-dimensional
chain group Cq (K) associated to the oriented complex K is indeed isomorphic
to the direct sum of as many copies of Z as there are q-simplexes in K regardless
of what orientation we give to K. Then, why to introduce an orientation in K
and make the definition of Cq (K) so lengthy and complicated? The answer to
this question will become clear as we proceed to define the boundary map and
realize the geometrical interpretation of the same. In what follows, whenever
we write a q-simplex as σ q , we mean σ q is positively oriented.
Definition 4.3.3. For each q, 0 < q ≤ dim K, we now define a homomorphism

∂q : Cq (K) → Cq−1 (K), called the boundary homomorphism, as follows: on
generators σ q of Cq (K), we define

q
(∗) ∂q (σ q ) = [σ q , σiq−1 ]σiq−1 ,
i=0
where σiq−1 runs over all positively oriented (q − 1)-faces of σ q . Then we extend
it over Cq (K) by linearity, i.e., we set

∂ q ( nq σ q ) = nq ∂q (σ q ),
where ∂q (σ q ) is defined by (∗). For q ≤ 0 or q > dim K, we define ∂q to be the

zero homomorphism (which is incidently the only possible map).
We note that in the summation (∗), σiq−1 runs over all (q − 1)-faces of
σ q . If we allow σiq−1 to run over all (q − 1)-simplexes of K, the summation will
remain unchanged since [σ q , σiq−1 ] = 0 if σiq−1 is not a face of σ q . Thus, we can
write (∗) as
∂q (σ q ) = [σ q , σiq−1 ]σiq−1 .
σiq−1 ∈K
Let us observe that if σ q = v0 , v1 , . . . , vq is oriented by the ordering v0 <

v1 < . . . < vq , then σ q has (q + 1)-faces v0 , v1 , . . . , vî , . . . , vq , i = 0, 1, . . . , q,
where vî means the vertex vi has been omitted from v0 , v1 , . . . , vi , . . . , vq and
each of these faces is oriented by the induced ordering. Evidently the incidence
number [σ q , σiq−1 ] is (−1)i and, therefore, (∗) reads as:

q
(∗∗) ∂q v0 , v1 , . . . , vq = (−1)i v0 , v1 , . . . , vî , . . . , vq .
i=0
Many authors use this representation of the boundary map rather than (∗) to
avoid the concept of incidence number.
There are, however, two points which we must answer before ∂q becomes
a well-defined map by (∗∗). One, a generator v0 , v1 , . . . , vq of Cq (K) can be
written in several ways as, for example,
v0 , v1 , . . . , vq = vπ(0) , vπ(1) , . . . , vπ(q) ,
where π is any even permutation on the symbols (0, 1, 2, . . . , q). Therefore,
we must ensure that ∂q v0 , v1 , . . . , vq is independent of these representations.
Second, if π is an odd permutation, then vπ(0) , vπ(1) , . . . , vπ(q) is also a gener-
ator of Cq (K), and the relation vπ(0) , vπ(1) , . . . , vπ(q) = −v0 , v1 , . . . , vq holds
among such generators. Hence we must show that for an odd permutation π,
∂q vπ(0) , vπ(1) , . . . , vπ(q) = −∂q v0 , v1 , . . . , vq .
Since any permutation can be written as a product of transpositions, both the
points raised earlier follow from the next result.
Proposition 4.3.4. If π is a transposition on the symbols (0, 1, 2, · · · , q), then

∂q vπ(0) , vπ(1) , . . . , vπ(q) = −∂q v0 , v1 , . . . , vq .
Proof. Let σ q = v0 , v1 , . . . , vs , . . . , vt , . . . , vq , 0 ≤ s < t ≤ q and
π be the transposition which interchanges vs and vt . Suppose τ q =
v0 , v1 , . . . , vt , . . . , vs , . . . , vq is the new oriented simplex by the indicated or-
q
dering. We show that ∂q (σ q ) = i=0 [σ q , σiq−1 ]σiq−1 is the negative of ∂q (τ q ) =
q q−1 q−1
q
i=0 [τ , τi ]τi . If s
= i
= t, then observe that [σ q , σiq−1 ]σiq−1 = (−1)i σiq−1
q−1 q−1
q
and [τ , τi ]τi = (−1)i τiq−1 . Because for each such i, σiq−1 can be obtained
q−1
from τi by just one transposition, the two terms differ only in sign. If i = s,
then [σ q , σsq−1 ]σsq−1 = (−1)s v0 , . . . , vˆs , . . . , vt , . . . , vq corresponds to the term
[τ q , τsq−1 ]τsq−1 = (−1)s v0 , . . . , vt , . . . , vˆs , . . . , vq .
Note that we need (t−s−1) transpositions to change v0 , . . . , vt , . . . , vˆs , . . . , vq
into v0 , . . . , vˆs , . . . , vt , . . . , vq , i.e.,
[σ q , σsq−1 ]σsq−1 = (−1)(t+s)+(t−s−1) [τ q , τtq−1 ]τtq−1

= (−1)2t−1 [τ q , τtq−1 ]τtq−1
= −[τ q , τtq−1 ]τtq−1 .
Finally, if i = t, then a similar argument would show that

[σ q , σtq−1 ]σtq−1 = −[τ q , τsq−1 ]τsq−1 .
Summing all the terms yields the desired result.
The next result is basic in defining the simplicial homology groups of an

oriented simplicial complex K.
Lemma 4.3.5. For each q, the composite homomorphism ∂q−1 ◦ ∂q : Cq (K) →

Cq−2 (K) is the zero map.
Proof. Let σ q be a generator of chain group Cq (K). It suffices to show that
∂q−1 ◦ ∂q (σ q ) = 0. Now

∂q−1 (∂q (σ q )) = ∂q−1 [σ q , σiq−1 ]σiq−1
σiq−1 ∈K

= [σ q , σiq−1 ]∂q−1 (σiq−1 )
σiq−1 ∈K

= [σ q , σiq−1 ] [σiq−1 , σjq−2 ]σjq−2 .
σiq−1 ∈K σjq−1 ∈K
Changing the order of summation and collecting the coefficients of each (q − 2)-
simplex σjq−2 shows that the above term is equal to

[σ q , σiq−1 ][σiq−1 , σjq−2 ] σjq−2 .
σjq−2 ∈K σiq−1 ∈K
By Theorem 4.2.4,

[σ q , σiq−1 ][σiq−1 , σjq−2 ] = 0,
σiq−1 ∈K
for each σjq−2 and so ∂q−1 ◦ ∂q (σ q ) = 0.
The notations and terminology introduced below will occur frequently in

the sequel.
Definition 4.3.6. Let K be an oriented complex. A q-chain zq ∈ Cq (K) is called

a q-dimensional cycle of K (or just a q-cycle) if ∂q (zq ) = 0. The set of all
q-cycles of K, denoted by Zq (K), is the kernel of the boundary homomorphism
∂q : Cq (K) → Cq−1 (K) and, therefore, is a subgroup of Cq (K). This Zq (K)
is referred to as group of q-cycles of K. An element bq ∈ Cq (K) is said to
be a q-dimensional boundary (or just a q-boundary) of K if there exists a
c ∈ Cq+1 (K) such that ∂q+1 (c ) = bq . The set of all q-boundaries, being the
homomorphic image ∂q+1 (Cq+1 (K)), is also a subgroup of Cq (K). It is denoted
by Bq (K) and is referred to as group of q-boundaries of K.
If dim K = n, then note that there is a sequence

∂ ∂q+1 ∂q
C(K) : . . . 0 → Cn (K) →
n
. . . → Cq+1 (K) → Cq (K) → Cq−1 (K)
→ . . . → C0 (K) → 0 → . . .
of free abelian groups and group homomorphisms in which the composite of

any two consecutive homomorphisms is zero (Lemma 4.3.5). This long sequence
is called the oriented simplicial chain complex of K. Since ∂q ◦ ∂q+1 = 0 for
each q, Im ∂q+1 ⊆ ker ∂q , i.e., Bq (K) ⊆ Zq (K). Consequently, one can talk of
the factor group Zq (K)/Bq (K).
The factor group mentioned above leads us to the following basic
Definition 4.3.7. Let K be an oriented simplicial complex. Then the q-

dimensional homology group of K, denoted by Hq (K), is defined to be the
quotient group
Hq (K) = Zq (K)/Bq (K).
If we consider the chain complex C∗ (K; G) where G is any coefficient

group, then we can similarly define
Hq (K; G) = Zq (K; G)/Bq (K; G)
which will be called the homology groups of K with G as coefficients.
Observe that for q < 0 or q > dim K, all q-dimensional chain groups
Cq (K) are evidently zero groups and, therefore, Hq (K) = 0 for all such
q. Possible nontrivial homology group Hq (K) of K can occur only when
0 ≤ q ≤ dim K. Moreover, note that Bn (K) = 0 and Z0 (K) = C0 (K). Let
us also point out that for an oriented simplicial complex K, we have now
defined four kinds of groups, viz., Cq (K), Zq (K), Bq (K) and Hq (K) for each
q, 0 ≤ q ≤ dim K. It turns out, however, that it is the homology groups
Hq (K) which are of fundamental importance rather than other three groups.
Later on we will see that the homology groups are indeed invariants of the
topology of |K|, and they really detect some kind of “holes” in the space |K|
whereas other groups mentioned above do not represent anything of |K| like
this – they are only the means to define the homology groups. An element of
Hq (K) is a coset of Bq (K) in Zq (K) and is called a homology class. If this
homology class is zq + Bq (K), we normally denote it by {zq }. We note that a
q-cycle zq will represent a nontrivial element of Hq (K) if and only if it is not
a boundary. Two q-cycles zq and zq are said to be homologous if they yield
the same homology class, i.e., zq − zq is a boundary. A q-cycle is homologous
to zero if and only if it is a boundary.
An important question regarding the simplicial homology groups Hn (K)

of a simplicial complex K, which must be answered right here, is: do these
homology groups Hn (K), which have been defined with respect to a given ori-
entation of K, depend on the orientation of K? We prove below that the answer
to this question is “no”. Note that a q-simplex σ q has exactly two orientations,
and that a simplicial complex K is oriented by assigning an arbitrary orien-
tation to each of the simplexes of K. Therefore, a simplicial complex can be
oriented in several different ways. We have
Theorem 4.3.8. Let K1 and K2 denote the same simplicial complex K equipped
with different orientations. Then Hq (K1 ) ∼
= Hq (K2 ), ∀q ≥ 0.
Proof. If σ q is an arbitrary simplex of K, then it has one orientation as a
simplex of K1 and has same or different orientation with respect to K2 . If
i q
σ denotes the simplex σ q having positive orientation from Ki , i = 1, 2, then,
clearly, 1 σ q = φ(σ) · 2 σ q , where φ(σ) = +1 or −1 depending on whether σ q
has the same or different orientation received from K1 and K2 . Thus, we get
a map φ : K → {1, −1}. Now, we define a sequence fq : Cq (K1 ) → Cq (K2 ) of
homomorphisms by

fq ( gi · 1 σiq ) = gi · φ(σ q ) · 2 σiq ,

where c = gi · 1 σiq is a chain in Cq (K1 ). Next, we claim that the above
sequence of homomorphisms is a chain map. We will prove that f ∂ = ∂f by
showing that both of these maps agree on elementary chains 1 σ q . We have

fq−1 ∂(1 σ q ) = fq−1 ( [1 σ q , 1 σ q−1 ] · 1 σ q−1 )
σ q−1 ∈K

= φ(σ q−1 )[1 σ q , 1 σ q−1 ] · 2 σ q−1 )

= φ(σ q−1 )[φ(σ q ) · 2 σ q , φ(σ q−1 ) · 2 σ q−1 ] · 2 σ q−1 )

= φ(σ q−1 )φ(σ q )φ(σ q−1 )[2 σ q , 2 σ q−1 ] · 2 σ q−1 )
σ q−1 ∈K

= φ(σ q ) [2 σ q , 2 σ q−1 ] · 2 σ q−1
σ q−1 ∈K
= ∂(φ(σ ) · σ )
q 2 q
= ∂fq (1 σ q ).
We can also define a chain map gq : Cq (K2 ) → Cq (K1 ) by putting
gq (2 σ q ) = φ(σ q ) · 1 σ q
on elementary q-chains, and then extending linearly to Cq (K2 ). Then, since

φ(σ q ) · φ(σ q ) = 1, the two chain maps {fq } and {gq } are inverse of each other.
Hence the induced homomorphism f∗ : Hq (K1 ) → Hq (K2 ) is an isomorphism
for all q ≥ 0, proving the theorem.
4.4 Some Examples
Let us now have some examples to illustrate how these homology groups
can be computed in some simple cases with Z as the coefficient group. These
simple cases are not just typical illustrations of computations, but also present
interesting hints of some general results and concepts to be identified and stud-
ied later on. The reader is advised to carry out every detail of these examples
himself.
Example 4.4.1. Consider the simplicial complex K having only one vertex v,
i.e., K = {v}. There is only one possible orientation on K and with that
orientation, we have Cq (K) = 0 for all q
= 0 and C0 (K) = Z · v is the free
abelian group on the single generator v. Hence we see that for q
= 0, Hq (K) = 0
and
H0 (K) = ker ∂0 /Im ∂1 ∼
= Z.
In fact, in this case, C0 (K) = Z0 (K) = H0 (K) ∼
= Z.
Example 4.4.2. Consider a two simplex σ 2 = v0 , v1 , v2 and let K = Cl(σ 2 )

(Fig. 4.2).
Fig. 4.2
Let us orient K by the ordering v0 < v1 < v2 (it has two distinct orientations).
We will first compute the non trivial chain groups, viz., C0 (K), C1 (K), C2 (K)
and see how the boundary homomorphisms act on these groups. Since there
are three vertices in K, C0 (K) is the free abelian group on the three vertices,
i.e.,
C0 (K) = Z · v0 ⊕ Z · v1 ⊕ Z · v2 .
Similarly,
C1 (K) = Z · v0 , v1 ⊕ Z · v1 , v2 ⊕ Z · v0 , v2
and
C2 (K) = Z · v0 , v1 , v2 .
The nontrivial part of our simplicial chain complex now looks like
∂ ∂
0 → C2 (K) →2 C1 (K) →1 C0 (K) → 0.
4.4. Some Examples 135
Note that any 2-chain c is of the type c = nv0 , v1 , v2 , where n ∈ Z, and so

∂2 (c) = n∂v0 , v1 , v2 = n[v1 , v2 −v0 , v2 +v0 , v1 ], which is zero if and only if
n = 0. Hence, Z2 (K) = 0 and it follows at once that H2 (K) = ker ∂2 /Im ∂3 = 0.
Next, let us see what B1 (K) and Z1 (K) are. If c1 = m0 v0 , v1 + m1 v1 , v2 +
m2 v0 , v2 is an arbitrary 1-chain then note that
(i) ∂(c1 ) = m0 [v1 − v0 ] + m1 [v2 − v1 ] + m2 [v2 − v0 ] = −(m0 +
m2 )v0 + (m0 − m1 )v1 + (m1 + m2 )v2 which is zero if and only if m0 + m2 =
0, m0 − m1 = 0 and m1 + m2 = 0, i.e., if and only if m0 = m1 = −m2 .
Hence a 1-cycle is of the form mv0 , v1 + mv1 , v2 − mv0 , v2 , m ∈ Z. It
follows from this that Z1 (K) is isomorphic to the additive group Z of integers
with 1-cycle v0 , v1 + v1 , v2 + v2 , v0 as a generator. On the other hand, to
compute B1 (K), note that any 2-chain is of the form c2 = nv0 , v1 , v2 , n ∈ Z.
Consequently, ∂2 (c2 ) = n[v1 , v2 − v0 , v2 + v0 , v1 ]. In other words, B1 (K) is
also isomorphic to the additive group Z of integers, and with the same generator
as that of Z1 (K). Therefore, B1 (K) = Z1 (K), and hence H1 (K) = 0. Finally,
we compute H0 (K) which is slightly tricky. We already know that Z0 (K) =
C0 (K) is isomorphic to the direct sum of three copies of Z. Let us compute
B0 (K). If c0 = h0 v0 + h1 v1 + h2 v2 , where hi ’s are integers, is an element
of B0 (K), then there must be a 1-chain, say, c1 = m0 v0 , v1 + m1 v1 , v2 +
m2 v0 , v2 , where m0 , m1 , m2 are integers, such that ∂1 (c1 ) = c0 . Using (i) and
comparing the coefficients, we find
(ii) h0 = −(m0 + m1 ), h1 = (m0 − m1 ), h2 = (m1 + m2 ), i.e., h0 + h1 = −h2 .
Thus, if c0 ∈ B0 (K), then the third coefficient of c0 is completely determined
by the first two coefficients. This tells us that B0 (K) will be isomorphic to
the direct sum of two copies of Z. Now, Z0 (K) and B0 (K) are determined,
the first is isomorphic to the direct sum of three copies of Z and the second
is isomorphic to two copies of Z. One might now conjecture that the quotient
group Z0 (K)/B0 (K) will be isomorphic to one copy of Z, but this requires a
proof. For this, notice that any 0-cycle c0 = n0 v0 + n1 v1 + n2 v2 can be
written as c0 = ∂1 [n1 v0 , v1 + n2 v0 , v2 ] + (n0 + n1 + n2 )v0 . This means
that c0 is homologous to the 0-chain (n0 + n1 + n2 )v0 . Thus, any 0-cycle is
homologous to an integral multiple of v0 . On the other hand, by the criterion
(ii), any two distinct integral multiples of v0 cannot be homologous. Thus,
H0 (K) is isomorphic to the additive group Z of integers. We summarize

Z if q = 0
Hq (K) =
0 if q
= 0.
Note. The reader should compute the homology groups of the above simplicial
complex K, if it is given some other orientation and see whether or not the
final result is different.
Example 4.4.3. Let K be the 1-skeleton of the complex K = Cl(σ 2 ). Then K

has three vertices v0 , v1 , v2 and three edges v0 , v1 , v1 , v2 and v0 , v2 (Fig.
4.3) and K is a triangulation of the space |K | which is homeomorphic to the

circle S1 . Once again, we orient K by the ordering v0 < v1 < v2 . Here, the
v0
v1 v2
Fig. 4.3
nontrivial part of the simplicial chain complex C∗ (K ) looks like
0 → C1 (K ) →1 C0 (K ) → 0,
∂
where
C1 (K ) = Z · v0 , v1 ⊕ Z · v1 , v2 ⊕ Z · v0 , v2
and
C0 (K ) = Z · v0 ⊕ Z · v1 ⊕ Z · v2 .
It follows, exactly as in the previous example, that H0 (K ) ∼ = Z. Moreover,
a 1-chain c1 = n0 v0 , v1 + n1 v1 , v2 + n2 v0 , v2 is a 1-cycle if and only if
n0 = n1 = −n2 (as in the previous example). This means elements of the form
n[v0 , v1 + v1 , v2 − v0 , v2 ], n ∈ Z, generate Z1 (K ), i.e., Z1 (K ) ∼
= Z. Since
B1 (K ) = 0, H1 (K ) ∼ = Z. Therefore, we conclude that
⎧
⎨ Z, if q = 0
Hq (K ) = Z, if q = 1
⎩
0, otherwise.
Remark. We must point out that with the orientation given by the ordering
v0 < v1 < v2 , the 1-cycle v0 , v1 +v1 , v2 −v0 , v2 = v0 , v1 +v1 , v2 +v2 , v0
is indeed a “closed circuit” or cycle of K which is not a boundary, i.e., this
cycle represents a “hole” in K . The same 1-cycle in the earlier example was
not a “hole” in that complex K = Cl(σ 2 ) because there it was the boundary
of the 2-simplex v0 , v1 , v2 of that complex.
Example 4.4.4. Once again, let σ 3 = v0 , v1 , v2 , v3 be a 3-simplex and L be the

2-dimensional skeleton of K (Fig. 4.4). Then |L| is homeomorphic to a 2-sphere
S2 , and L is a triangulation of S2 . We orient L by ordering v0 < v1 < v2 < v3 .
We are going to show that

Z, if q = 0 or 2
Hq (L) =
0, otherwise.
v0
v1 v3
v2
Fig. 4.4
In this case, observe that the nontrivial part of our oriented simplicial chain
complex is
∂ ∂
0 → C2 (L) →2 C1 (L) →1 C0 (L) → 0,
where
C0 (L) = Z · v0 ⊕ Z · v1 ⊕ Z · v2 ⊕ Z · v3
C1 (L) = Z · v0 v1 ⊕ Z · v0 v2 ⊕ Z · v0 v3 ⊕ Z · v1 v2
⊕Z · v1 v3 ⊕ Z · v2 v3
C2 (L) = Z · v0 v1 v2 ⊕ Z · v1 v2 v3 ⊕ Z · v0 v1 v3 ⊕ Z · v0 v2 v3 .
To compute H0 (L), note that every 0-chain is a zero cycle. For each i =
1, 2, 3, since ∂(vi , v0 ) = v0 − vi , we find that the difference v0 − vi
is a boundary, i.e., vi is homologous to v0 for each i. Thus, any 0-cycle
c0 = n0 v0 +n1 v1 +n2 v2 +n3 v3 is homologous to (n0 +n1 +n2 +n3 )v0 . In
other words, each 0-cycle is homologous to an integral multiple of v0 . Also, it
can be seen, as in Example 4.4.2, that a 0-cycle n0 v0 +n1 v1 +n2 v2 +n3 v3
is a boundary if and only if n0 + n1 + n2 + n3 = 0. This means that the two
integral multiples mv0 and nv0 of v0 are not homologous unless m = n.
From these observations it follows at once that the map H0 (L) → Z defined by
{nv0 } → n is an isomorphism.
To compute H2 (L), observe that H2 (L) = ∼ Z2 (L). Now, if z =
av0 , v1 , v2 +bv0 , v1 , v3 +cv1 , v2 , v3 +dv0 , v2 , v3 is a 2-cycle, then ∂2 (z) = 0
i.e., a[v1 , v2 − v0 , v2 + v0 , v1 ] + b[v1 , v3 − v0 , v3 + v0 , v1 ] + c[v2 , v3 −
v1 , v3 + v1 , v2 ] + d[v2 , v3 − v0 , v3 + v0 , v2 ] = 0. Equating the coeffi-
cients of various 1-simplexes to zero, we get a = −b = −c = d. This means
z = a[v0 , v1 , v2 +v1 , v3 , v2 +v0 , v3 , v1 +v0 , v2 , v3 ]. Since a ∈ Z is arbitrary,
we conclude that H2 (L) ∼ = Z2 (L) ∼ = Z.
Finally, to prove that H1 (L) = 0, we shall show that every 1-cycle is a
boundary. Let z = a0 v0 , v1 + a1 v0 , v2 + a2 v0 , v3 + a3 v1 , v2 + a4 v1 , v3 +
a5 v2 , v3 be a 1-cycle. Note that ∂2 v1 , v2 , v3 = v2 , v3 − v1 , v3 + v1 , v2 ,
which means v2 , v3 is homologous to v1 , v3 − v1 , v2 , and so the cycle z is
homologous to a cycle in which the term v2 , v3 is replaced by v1 , v3 −v1 , v2 .
Similarly, the term v1 , v2 can also be replaced by v1 , v0 + v0 , v2 . Thus, z is

homologous to z which does not contain any terms of v2 , v3 or v1 , v2 . But,
then z cannot contain any term like v0 , v2 , otherwise ∂1 (z ) = 0 will mean
that the coefficient of v2 in ∂1 (z ) must be zero, which would be impossible.
Thus, z must be of the form z = b0 v0 , v1 + b1 v1 , v3 + b2 v3 , v0 . The fact
that ∂1 (z ) = 0 will further imply, as in Example 4.4.2, that b0 = b1 = b2 .
Consequently, z is a multiple of x = v0 , v1 +v1 , v3 +v3 , v0 = ∂2 (v0 , v1 , v3 )
which, evidently, shows that z is a boundary and therefore z is a boundary.
∼ Z, there is 2-dimensional “hole”

Note. It may be observed that since H2 (L) =
in the simplicial complex L which is a triangulation of the 2-spheres S2 ; it
has no 1-dimensional “holes”, no 3-dimensional or higher dimensional “holes”
either.
Example 4.4.5. We now consider the torus T = S1 × S1 . It has been already

observed that a triangulation K of T is given by Fig. 4.5.
v v1 v2 v0
0
v3 v v3
8 v7
v4 v v4
5 v6
v v1 v2 v0
0
Fig. 4.5
We orient triangles of K as indicated by the arrows in Fig. 4.5, not by the

ordering of its vertices, simply for computational convenience. We may orient
all edges by the linear ordering of its vertices. Our simplicial chain complex is
∂ ∂
0 → C2 (K) →2 C1 (K) →1 C0 (K) → 0,
where Ci (K) is a free abelian group of rank equal to the number of i-simplexes
in K. Evidently, Hq (K) = 0 if q > 2.
Note that each 1-simplex of K is an edge of exactly two 2-simplexes of K
and the orientation of the edge induced by one of the 2-simplexes is opposite to
that of the other 2-simplex. Now, we consider the 2-chain P which is the sum
of all positively oriented 2-simplexes of K. We claim that it is a 2-cycle. For,
the boundary ∂2 (P ) would be the sum of various 1-simplexes σ 1 of K. Any
1-simplex σ 1 will occur in the boundary of exactly two 2-simplexes, say, σ12 and
σ22 of the sum P . Since the incidence numbers [σ12 , σ 1 ] and [σ22 , σ 1 ] are opposite
to each other, the coefficient of σ 1 in ∂2 (P ) is zero. Consequently, ∂2 (P ) = 0.
On the other hand, if z ∈ C2 (K) is an arbitrary 2-cycle, then write z as a linear
combination of all 2-simplexes of K and consider the equation ∂2 (z) = 0. Since
∂2 (z) is a linear combination of various 1-simplexes of K and each 1-simplex
occurs in the boundary of exactly two 2-simplexes, the coefficient of each 1-

simplex will be zero only when the coefficients of various 2-simplexes in z are
identical, i.e, z is an integral multiple of P . Thus, H2 (K) ∼
= Z2 (K) ∼
= Z.
To compute H1 (K), note that any 1-cycle z of K is an integral linear
combination of all of its 1-simplexes. Since ∂2 (v0 , v1 , v3 ) = v1 , v3 − v0 , v3 +
v0 , v1 is a boundary, v1 , v3 can be replaced in z by v0 , v3 − v0 , v1 without
changing the homology class. Similarly, v3 , v8 can be replaced by v3 , v0 +
v0 , v1 + v1 , v8 , etc. Making such replacements, one can easily see that any
1-cycle is homologous to a cycle of the type z = a1 v0 , v3 + a2 v3 , v4 +
a3 v4 , v0 +a4 v0 , v1 +a5 v1 , v2 +a6 v2 , v0 +a7 v1 , v8 +a8 v8 , v5 +a9 v6 , v7 +
a10 v7 , v2 . Equating various coefficients of vertices in the equation ∂1 (z ) = 0
to zero, we see that a7 = 0 = a8 = a9 = a10 , a1 = a2 = a3 and a4 = a5 = a6 .
Let us put u = v0 , v1 + v1 , v2 + v2 , v0 and v = v0 , v3 + v3 , v4 + v4 , v0
and observe that each one of these is a closed circuit (hence 1-cycle) which is
not a boundary. Hence, z is homologous to au + bv ∈ Z1 (K) for some a, b ∈ Z,
i.e., Z1 (K) = Z·u+Z·v. To see that the sum is direct, we note that if au+bv is
boundary for some a, b ∈ Z, then au + bv = ∂2 (x) for some x ∈ C2 (K). Now, if
σ is an oriented 2-simplex appearing in x, then at least one edge of σ, say e, is
different from those in u and v. This means e will also be an edge of some other
2-simplex, say, τ . Since σ, τ have opposite orientations and the coefficient of e
in ∂2 (x) is zero, the coefficients of σ and τ must be identical, i.e., x = kP for
some k ∈ Z, where P is the 2-cycle used earlier. But then ∂2 (x) = k∂2 (P ) = 0.
This proves that H1 (K) ∼ = Z ⊕ Z.
By an argument parallel to Examples 4.4.2, 4.4.3 and 4.4.4, one can easily
show that H0 (K) ∼
= Z. Combining all these we have
⎧
⎪
⎪ Z, q=0
⎨
Z ⊕ Z, q = 1
Hq (K) =
⎪
⎪ Z, q=2
⎩
0, otherwise.
Example 4.4.6. Recall that the projective plane P2 is obtained by identifying

each pair of diametrically opposite points of the disc D2 . We have seen (exercises
of Chapter 3) that P2 is triangulable and a triangulation K of P2 is given in
Fig. 4.6. Once again, we orient 2-simplexes of K as indicated by the arrows
and 1-simplexes by the linear ordering of their vertices.
We will show that
⎧
⎨ Z, q=0
Hq (K) = Z/2Z, q = 1
⎩
0, otherwise.
We observe, once again, that every 1-simplex is a face of exactly two 2-simplexes
of K. Moreover, each 1-simplex v3 , v4 and v4 , v5 or v5 , v3 receives the same
v3
v4 v5
v2
v0 v1
v5 v4
v3
Fig. 4.6
orientation from both the 2-simplexes of which it is a common face whereas all
other 1-simplexes receive opposite orientations from their corresponding pairs
of 2-simplexes. Here, the simplicial chain complex is
2 ∂ 1 ∂
0 → C2 (K) −→ C1 (K) −→ C0 (K) → 0,
each Ci (K) being the free abelian group on as many generators as there are
i-simplexes of K.
Let z be a 2-cycle of K. Write z as an integral linear combination of all
2-simplexes of K and put ∂2 (z) = 0. Then we claim that the coefficients of
various 2-simplexes in z must be equal. This follows from the fact that any
2-simplex has at least a 1-face lying in the interior of the disk, which receives
opposite orientations from the other 10 2-simplexes having that 1-simplex as a
common face. This yields z = n( i=0 σi2 ), n ∈ Z and so ∂2 (z) = 2n[v3 , v4 +
v4 , v5 + v5 , v3 ] because the coefficients of all other 1-simplexes receiving
opposite orientations from the 2-simplexes must be zero. Now, ∂2 (z) = 0 means
n = 0 and so z = 0. Therefore, H2 (K) = 0.
Note that z = v3 , v4 + v4 , v5 + v5 , v3 , making a complete circuit, is a
1-cycle of K lying on the boundary of the disk. Moreover, earlier considerations
show that 2z, and so, any even multiple of z is a boundary. We use intuition and
geometry of K to assert that any 1-cycle is homologous to an integral multiple
of z. Let us see how. As an example, note that z = v3 , v2 +v2 , v0 +v0 , v3 is
also a 1-cycle. However, we can replace v3 , v2 by v3 , v4 + v4 , v2 , v2 , v0 by
v2 , v4 + v4 , v0 , v4 , v0 by v4 , v5 + v5 , v0 , v0 , v3 by v0 , v5 + v5 , v3 and
see that z is homologous to z. This tells us that H1 (K) = Z1 (K)/B1 (K) ∼ =
Z/2Z. Finally, as in the earlier examples, one can easily see that H0 (K) ∼ = Z.
This proves our claim.
Remark. In all of the above examples, the coefficient group for the simplicial
homology H∗ (K) has been the additive group Z of integers.Let G be any
abelian group. Then by defining Cq (K, G) to be the direct sum G of as many
copies of G as there are q-simplexes in K, we could have used the abelian group
G instead of Z to compute the above groups. In that case the answer would
have been different. The abelian group Z2 is frequently used as the coefficient
group and the resulting homology groups of K are known as (mod 2) homology
groups of K. We will also have occasion to use the additive group of rationals
(Q, +) as the coefficient group.
Zero-dimensional Homology Groups
From the foregoing examples, one might observe that, in each case, the
simplicial complex is connected and the zero-dimensional homology group is
isomorphic to Z. In fact, this is a special case of the general result to be proved
below. Recall that, by definition, our simplicial complexes have finite number
of simplexes.
Theorem 4.4.7. Let K be a simplicial complex. Then H0 (K) is always a free

abelian group. More precisely, if K has r combinatorial components Ki , i =
1, . . . , r and we choose
r one vertex vi from each Ki , then H0 (K) is isomorphic
to the direct sum 1 Z of r-copies of Z · {vi }.
Proof. We note that all 0-chains vi are 0-cycles. Let w ∈ Ki be any other
vertex. Since Ki is a combinatorial component of K, there is a sequence
ai , ai+1 is a 1-simplex for each
a0 , a1 , . . . , an such that w = a0 , vi = an and
i = 0, 1, . . . , n − 1. Then the boundary is ∂( ai , ai+1 ) = vi − w, which means
w is homologous to vi . This says that any 0-cycle nj wj lying in the com-
ponent Ki is homologous to ( nj )vi . More generally, this implies that any
0-chain (which is a 0-cycle) in K is homologous to a linear combination of the
elementary 0-chains vi .
{vi }i=1 are linearly independent, i.e., no

r
Next, we prove that the cycles
nontrivial linear combination c = ni vi can be a boundary. Suppose c = ∂(d)
for some 1-chain d. Since each 1-simplex of K lies in a unique
component
Ki , we can write d = di , where di ∈ Ki . Hence ∂(d) = ∂(di ), which
means ∂(di ) ∈ Ki , and therefore, ∂(di ) = ni vi , for all i = 1, 2, . . . , r. Now,
we define a homomorphism from C0 (K) to Z by putting (v) = 1 for each
vertex of K. Then it is easily seen that for each elementary 1-chain v, w,
∂(v, w) = (w − v) = 1 − 1 = 0. In particular, 0 = (∂di ) = (ni vi ) = ni , for
all i = 1, 2, . . . , r.

The first para above shows that the homomorphism f : Z → H0 (K)
defined by

f (m1 , . . . , mr ) = mi {vi }
is onto while
r the second para shows that it is 1-1. It follows that
H0 (K) ∼
= 1 Z · {vi }.
Remark. We have already seen that the combinatorial components of a

simplicial complex K are in 1-1 correspondence with the path-components of
|K|. Hence it follows from the above theorem that if we take two simplicial
complexes K and K such that |K| is homeomorphic to |K |, the number
of path components of |K| is equal to the number of path components of
|K | and so H0 (K) ∼ = H0 (K ). This says that the zero-dimensional homology
group of a simplicial complex K depends on the topology of |K|, not on
a particular triangulation of |K|. In other words, the group H0 (K) is a
topological invariant of |K|. It is worthwhile to indicate here that, though the
higher-dimensional homology groups are also topological invariants of |K|, but
the proofs, as we shall see later, are not that easy. One should also know a
better result, viz., H0 (K) is not only a topological invariant of |K|, but also a
homotopy invariant, i.e., if |K| and |K | are of the same homotopy type, then
H0 (K) ∼ = H0 (K ). This will become clear in due course.
The following version of homology groups will be quite useful in the sequel:
Definition 4.4.8. Let K be a simplicial complex. Define a homomorphism :

C0 (K) → Z, called augmentation map by putting (v) = 1 for each vertex v of
K and extending linearly to C0 (K).
Note that this augmentation homomorphism is onto. If ∂1 : C1 (K) →

C0 (K) is the boundary homomorphism, then evidently ∂ = 0.The quotient
group ker /Im ∂1 , denoted by H̃0 (K), is called the zero-dimensional reduced
homology group of K. If we put H̃p (K) = Hp (K) for each p > 1, then the
groups {H̃i (K) : i = 0, 1, . . .} are called the reduced homology groups of K.
Since Im ∂1 ⊂ ker ⊂ C0 (K), there is an induced inclusion map H̃0 (K) →
H0 (K). In fact, we have
Theorem 4.4.9. For a simplicial complex K, zero-dimensional reduced homology

group is also free abelian, and we have the following formulae:
H0 (K) ∼
= H̃0 (K) ⊕ Z, Hp (K) = H̃p (K), p ≥ 1.
Proof. The sequence

0 → ker → C0 (K) → Z → 0
is exact. This implies that
ker C0 (K)
0→ → →Z→0
Im ∂1 Im ∂1
is exact. Since Z is free, the above sequence splits and we find that
H0 (K) ∼= H̃0 (K) ⊕ Z. Moreover, if {vi }r0 is a basis of H0 (K), then {vi − v0 }r1
is a basis of H̃0 (K).
4.5. Properties of Integral Homology Groups 143
Exercises
1. Consider X as the union of two line segments AB and AC in Rn , n ≥ 2
having the point A common. Write down a triangulation K for X and
compute the homology groups Hq (K) for all q ≥ 0.
2. Consider the following simplicial complexes (vertices and edges are shown
below, and with no triangles) K1 and K2 . Compute their homology groups.
A A A
K1 : A A K2 : A
A A A
A A A
3. Give a triangulation K of the Möbius band M and compute all the ho-
mology groups Hq (K), q ≥ 0.
4. Give a triangulation of cylinder and compute all of its homology groups.
5. Give a triangulation L of the Klein bottle K and compute all of its ho-
mology groups.
6. Work out Examples 4.4.4, 4.4.5, and 4.4.6 using the coefficient group
G = Z2 , G = Q instead of Z.
7. Construct a simplicial complex K such that H0 (K) ∼
= Z, H1 (K) ∼
= Z/3Z
and Hq (K) = 0 ∀ q ≥ 2. Generalize this result, i.e., for any n find a
simplicial complex Kn such that
H0 (Kn ) ∼
= Z, H1 (Kn ) ∼
= Z/nZ, Hq (Kn ) = 0 for all q ≥ 2.
(Hint: Consider a triangle ABC, identify AB with BC, BC with CA.)

8. Let K r be the r-dimensional skeleton of an n-dimensional simplicial com-
plex K, 0 ≤ r < n. Determine, how the homology groups Hq (K r ) are
related to Hq (K n ).
9. Let K be a triangulation of the connected sum T #T of two tori. Compute
the homology groups of K.
10. Let K be a triangulation of the connected sum RP2 #RP2 of two projective
planes. Compute the homology groups of K.
4.5 Properties of Integral Homology Groups

Looking at the various examples that we have discussed so far, we
observe that direct sums of copies of Z and finite cyclic groups have occurred
as homology groups of simplicial complexes considered so far. It is, therefore,
natural to ask: what kind of groups can occur as the homology groups of an
arbitrary finite simplicial complex when we use Z as the coefficient group?
This question can be readily answered: note that, a finite simplicial complex
K has only finite number of simplexes. If dim K = n, then there are no
q-simplexes for q > n and so Hq (K) = 0 for all q > n. For each q ≤ n, there
are only finite number of q-simplexes in K and, consequently, the qth chain
group Cq (K) is a finitely generated free abelian group of rank equal to the
number of q-simplexes in K. Because Zq (K) and Bq (K) are subgroups of
Cq (K) and a subgroup of a free group is always free, the group of cycles and
boundaries are also finitely generated free abelian groups. The factor group
Hq (K) = Zq (K)/Bq (K), therefore, must be a finitely generated abelian group
which may or may not be free. In any case we conclude that the qth homology
group Hq (K) of a simplicial complex K is a finitely generated abelian group
for all q ≥ 0.
In view of the Structure Theorem for Finitely Generated Abelian Groups,

the above information about Hq (K) alone determines the nature of Hq (K)
completely. The theorem asserts that Hq (K) will be isomorphic to the direct
sum of a finite number of copies of Z and a finite number of cyclic groups of
descending orders. In other words,
k copies

(∗) Hq (K) ∼
= Z ⊕ . . . ⊕ Z ⊕Z/m1 Z ⊕ . . . ⊕ Z/mn Z,
for each q ≥ 0, where mi is divisible by mi+1 , i = 1, . . . , n − 1.
Definition 4.5.1. If we write Hq (K) ∼ = Fq (K) ⊕ Tq (K), where Fq (K) is the

direct sum of copies of Z (called the free part of Hq (K)) and Tq (K) is the
direct sum of finite cyclic groups (called torsion subgroup of Hq (K)) as in (∗)
above, then the rank of Fq (K) is called the qth Betti number (in honour of E.
Betti (1823–1892)) of the simplicial complex K and the numbers mi occurring
in the torsion subgroup Tq (K) are called the qth torsion coefficients of K.
It is interesting to emphasize that, as a consequence of the structure

theorem for finitely generated abelian groups, knowing both the qth Betti
number as well as the qth torsion coefficients of K is equivalent to knowing
the qth homology group of K itself. Hence it is one and the same thing
to determine all the homology groups Hq (K) of the simplicial complex K
or to compute all of its Betti numbers and torsion coefficients of K. As a
matter of fact, historically, it is the Betti numbers and torsion coefficients of
a complex K which were emphasized and given importance first. It is due to
the famous lady mathematician Emmy Noether that, later on, the viewpoint
changed completely and the emphasis shifted to the group structure of Hq (K)
rather than those numbers. The main reason for this transition of emphasis
was that a more useful algebra in the form of “homological algebra” was
naturally developed and introduced in the study of topological problems.
Concepts like “homology with coefficients”, “cohomology product” and “higher
order cohomology operations”, etc., to name a few, thus emerged and gave
a lot more information about arbitrary topological spaces than what could
be deduced from the knowledge of Betti numbers and torsion coefficients alone.
As pointed out in the introduction, we will finally prove (see Section

4.9) that homology groups Hq (K) of a simplicial complex K depend on the
topology of the polyhedron |K|, not on the simplicial structure of K or its
orientation. This is one of the most beautiful and exciting results of algebraic
topology. This fundamental fact at once increases our (topologist’s) respect
for homology groups enormously. First of all, the problem of computing
homology groups Hq (K) of a simplicial complex K becomes easier because
we can choose any suitable triangulation K of |K| which might be easier
to work with and then compute Hq (K ) rather than Hq (K). Secondly, any
information about the homology groups Hq (K) should be interpreted as
reflecting some interesting property of the topological space |K|. For instance,
one can say that the nontrivial homology groups of a simplicial complex
K detect a kind of “holes” in the underlying space |K|. For instance, all
positive-dimensional homology groups of a triangulation K of the 2-disk
D2 ≈ |K| are zero (Example 4.4.2) means the disk D2 has no “holes” in any
dimension; the circle S1 is homeomorphic to |K|, where K is the 1-skeleton of
the geometric complex of a 2-simplex and we have seen (Example 4.4.3) that
H1 (K) ∼= Z, which means S1 has a 1-dimensional “hole” and a generator of
Z corresponds to that hole. Moreover, S1 has no holes of higher dimension.
Similarly, the second homology group H2 (K) of a triangulation K of the
sphere S2 (Example 4.4.4) is isomorphic to Z, which means a generator of
H2 (K)(∼= Z) corresponds to a 2-dimensional hole of S2 and no holes in other
dimensions ≥ 2. Finally, for a triangulation K of the torus T , we saw (Example
4.4.5) that H1 (K) = Z · u ⊕ Z · v which means T has two ”1-dimensional holes”.
The generator u corresponds to one hole and v corresponds to the other hole
(Fig. 4.7). Since H2 (K) ∼
= Z, the torus has a “two-dimensional hole” also.
v
Fig. 4.7: The two 1-holes of torus
In case of the triangulation K (Example 4.4.6) of the projective plane

P2 , H1 (K) = Z2 and that means P2 has a “twisting” along its 1-dimensional
hole. In fact, for any complex K, the free part of Hq (K) indicates that there
are as many q-dimensional holes in |K| as there are copies of Z in Hq (K)
and the torsion part indicates that there are “twistings” around other holes
in |K|. The 0-dimensional homology group H0 (K) of K, as we have already
seen, reflects a nice topological property of |K|, viz., the rank of H0 (K) is the
number of combinatorial components of K, which is equal to the number of
connected components of |K|.
Next, we give some interesting applications of homology groups by an-

swering a few topological questions mentioned earlier. However, before coming
to these applications, we need to compute the homology groups of an n-disk
and an n-sphere for any n ≥ 1.
Simplicial Homology of Discs and Spheres

Let Dn = {(x1 , . . . , xn ) ∈ Rn | |xi |2 ≤ 1} and let Sn−1 =
{(x1 , . . . , xn ) ∈ R |
n
|xi | = 1} be its boundary. We know that if σ n
2
denotes the n-simplex v0 , . . . , vn , then K = Cl(σ n ) is the simplicial complex

consisting of all faces of σ n . Also, the collection K of all proper faces of σ n is
a simplicial complex. We have already seen that |K| ≈ Dn and |K | ≈ Sn−1 . It
follows that K is a triangulation of Dn and K is a triangulation of Sn−1 . As
mentioned in Section 4.1, later on we will prove that if K1 and K2 are any two
triangulations of a compact polyhedron X, then Hq (K1 ) ∼ = Hq (K2 ) ∀ q ≥ 0.
Therefore, we can define the simplicial homology of any polyhedron X by
putting Hq (X) = Hq (K), where K is some triangulation of X. In view of this
observation we find that Hq (Dn ) ∼ = Hq (K) and Hq (Sn−1 ) ∼ = Hq (K ) in all

dimensions and we now compute Hq (K) and Hq (K ).
Let us have
Definition 4.5.2. Suppose K is a complex in Rn for n large, and w is a point in

Rn such that each ray emanating from w intersects |K| in at most one point.
Let w ∗ K denote the collection of all simplexes of the form w, a0 , . . . , ap ,
where a0 , . . . , ap is a simplex of K, along with all faces of such simplexes. It
can be seen that w ∗ K is a simplicial complex. This complex is called a cone
over K with vertex w.
Note that if σ n = v0 , . . . , vn is a n-simplex then the complex K = Cl(σ n )

consisting of all faces of σ n is the cone v0 ∗ Clv1 , . . . , vn where v1 , . . . , vn is
the face opposite to the vertex v0 . In fact, for any vertex vi , Cl(σ n ) = vi ∗Cl(τ ),
where τ is the face opposite to vi . The following result determines the homology
groups of a n-disk Dn completely.
Theorem 4.5.3. Let K be any complex. Then the reduced homology groups of
the cone w ∗ K are trivial in all dimensions, and hence

Z, q = 0
Hq (w ∗ K) =
0, otherwise.
Proof. It suffices to show that the reduced homology groups H̃q (w∗K) vanish in
all dimensions. Since the complex w∗K is connected, the 0-dimensional reduced
homology group is evidently trivial. Now, let q > 0. Let zq be a q-cycle of w ∗K.
We will show that zq is a boundary. For any oriented simplex σ = v0 , . . . , vq of
K, let [w, σ] denote the oriented simplex w, v0 , . . . , vq of w ∗ K. Note that this
operation, called bracket operation, is well defined
and gives a homomorphism
from Cp (K) → Cp+1 (w ∗ K) by putting [w, ni σi ] = ni [w, σi ]. Also, we
note that for a simplex σ,

σ−w if dim σ = 0
∂[w, σ] =
σ − [w, ∂σ] if dim σ > 0
and so, more generally, for a chain cq ,

cq − (c0 )w, q = 0
∂[w, cq ] =
cq − [w, ∂cq ], q > 0.
Let us write zq = cq + [w, dq−1 ] where cq consists of those terms of zq which
are in the complex K and dq−1 is a chain of K. By direct calculation using the
above formulae, we get
zq − ∂[w, cq ] = cq + [w, dq−1 ] − cq + [w, ∂cq ]
= [w, cq−1 ]
where cq−1 = dq−1 + ∂cq is a chain in K. Applying the boundary operator ∂
on both sides above, and noting that zq is a cycle, we have

cq−1 − (cq−1 )w q = 1
0=
cq−1 − [w, ∂cq−1 ] q > 1.
Note that cq−1 is a chain with all its simplices in K. However, (cq−1 )w and
[w, ∂q−1 ] are chains whose all simplices, having w as a vertex, are not in K.
This is possible only when cq−1 and the latter chains are themselves 0-chains,
i.e., cq−1 = 0. But that implies
zq − ∂[w, cq ] = [w, cq−1 ] = 0
i.e., zq is a boundary. This completes the proof.
Corollary 4.5.4. Let Dn be the Euclidean n-disk. Then

n Z q=0
Hq (D ) =
0 otherwise.
Proof. The proof follows from the above theorem and the fact that K = Cl(σ n )
gives a triangulation of Dn . Furthermore, the complex K itself is a cone.
The next result determines the homology groups of the n-sphere Sn com-
pletely.
Theorem 4.5.5. Let K(n), n ≥ 1, denote the n-skeleton of the complex K =

Cl(σ n+1 ). Then
Z if q = 0 or n
Hq (K(n)) =
0 otherwise.
∼ Z if q = n and is zero otherwise.
Proof. It suffices to prove that H̃q (K(n)) =
Note that K(n) is a subcomplex of K = Cl(σ n+1 ). We consider the following
commutative diagram of chain complexes defining the homology groups of K
and K(n). The vertical maps are induced by inclusions.
∂n+1 ∂n
Cn+1 (K) Cn (K) ... C0 (K) Z
= =
0 Cn (K(n)) ... C0 (K(n)) Z

∂n
Note that both chain complexes are identical except in dimension n + 1. It

follows, therefore, that
H̃i (K(n)) ∼
= H̃i (K) = 0
for all i
= n. In dimension n, we have
Hn (K(n)) ∼
= Zn (K(n))
= ker ∂n
= Im ∂n+1 ,
since Hn (K) = 0. Now, Cn+1 (K) is infinite cyclic generated by σ n+1 and ∂n+1
is a nonzero homomorphism. Hence Im ∂n+1 is infinite cyclic because Cn (K)
has no torsion, i.e., Hn (K(n)) ∼
= Z.
Corollary 4.5.6. Let Sn , n ≥ 1, denote the n-sphere. Then

Z if n = 0 or q
Hq (Sn ) =
0 otherwise.
Proof. This follows at once from the above theorem since K(n) in the above
theorem is a triangulation of Sn .
The Euler-Poincaré Theorem

We know that boundary of a tetrahedron is homeomorphic to a 2-sphere
S2 via the radial map which takes vertices to points of S2 , edges to spherical
edges, and the triangles to spherical triangles lying on S2 . Note that for this
spherical triangulation of S2 , we have the following obvious relation:
(# of vertices − # of edges + # of faces) = 2.
L. Euler (1707–1783) proved in 1752 that if one takes an arbitrary decompo-

sition K of S2 in terms of properly joined polyhedral faces, which need not
necessarily be triangles but could be rectangles or any convex polygons, then
the same above relation holds for K also – this is the celebrated Euler’s theo-
rem, which we will prove shortly. Now, we observe that the number 2 appearing
in the right-hand side of the above relation is, in fact, the alternate sum of the
Betti numbers of S2 , and these Betti numbers, as we have pointed out earlier,
are independent of the triangulation of the 2-sphere. One can now ask whether
or not a similar relation holds for a triangulation of torus, a triangulation of
Klein bottle or a triangulation of any polyhedron. H. Poincaré made the first
use of the homology groups in proving that the answer to the above question
is in affirmative. Let us have
Definition 4.5.7. Let K be an oriented complex of dimension n, and let Rq (K)

be the (Betti numbers)
n rank of the abelian group Hq (K), q = 0, 1, . . . , n. Then
the alternate sum q=0 (−1)q Rq (K), denoted by χ(K), is called the Euler char-
acteristic of K.
We observe that, in the definition of Euler characteristic χ(K), we have

used the homology groups Hi (K) with integer coefficients. If we consider
these groups with rational coefficients Q, then each Hi (K; Q) will be a
vector space over Q, and in that case the rank Hi (K; Q) is the dimension
of the vector space Hi (K; Q). It can be seen (Exercise 1 of Section 6.10)
that rank Hi (K; Z) = dimQ Hi (K; Q) = ith Betti number of the complex
K. Hence the Euler characteristic can be defined using rational coefficients
also. The most important point to be kept in mind is that if K and K are
triangulations of a space X, then by the topological invariance of simplicial
homology, Hi (K) ∼ = Hi (K ), ∀ i ≥ 0, and hence the Euler characteristic really
depends on the topology of the space X = |K|, not on any triangulation K
of X. In other words, the Euler characteristic of a compact polyhedron is a
topological invariant.
The next theorem generalizes the Euler’s Theorem mentioned above to

arbitrary polyhedra.
Theorem 4.5.8. (Euler-Poincaré Theorem). Let K be an oriented complex of

dimension n. Suppose for each q = 0, 1, . . . , n, αq denotes the number of q-
simplexes of K. Then

n
n
(−1)q αq = (−1)q Rq (K).
0 0
Proof. Note that the right hand side of the relation to be proved is in terms
of rank of homology groups Hq (K) of K, q ≥ 0. As pointed out earlier,
rank Hq (K; Z) = dimQ Hq (K; Q) and hence, we can assume that all the
homology groups are with coefficients in Q. It follows that all the groups
Cq (K), Zq (K), Bq (K) and Hq (K) associated to the complex K are vector
spaces over Q. In what follows, we will suppress the complex K as well as
the coefficient Q while using above groups. Now, consider the following chain
complex
∂q+1
0 → Cn (K) → . . . → Cq+1 (K) −→ Cq (K) → . . . → C0 (K).
Note that αq = dim Cq (K), q ≥ 0. Also, observe that
Bq = Im (∂q+1 ) ∼
= Cq+1 / ker ∂q+1 = Cq+1 /Zq+1 .
Therefore, for all q ≥ 0, we have
dim Bq = dim Cq+1 − dim Zq+1 .

∼ Zq /Bq and so
On the other hand, Hq =
dim Hq = dim Zq − dim Bq .
Substituting the value of dim Bq from the latter equation in the former one,
we find that for all q ≥ −1,
dim Cq+1 = dim Zq − dim Hq + dim Zq+1 .
Putting q = −1, 0, . . . , n and taking the alternate sum, we find that

n
n
(−1)q dim Cq = (−1)q dim Hq .
0 0
Since dim Cq = αq and dim Hq = Rq (K), we get the desired result.
Now, we are going to derive the classical theorem due to L. Euler indicated
earlier from the Euler-Poincaré Theorem. We need the following:
Definition 4.5.9. By a rectilinear polyhedron P in the Euclidean 3-space, we

mean a solid in R3 which is bounded by properly joined 2-dimensional convex
polygons. The bounding polygons which may be triangles, rectangles or any n-
gon, are called faces of P . The intersection of any two faces is called an edge
of P and the intersection of any two edges is called a vertex. The polyhedron
is said to be simple if its boundary is homeomorphic to the 2-sphere S2 . A
rectilinear polyhedron is said to be regular if its faces are regular congruent
polygons and all polyhedral angles are equal.
In the Euler-Poincaré Theorem, if we take K to be the boundary complex

of Cl(σ 3 ), where σ 3 is a 3-simplex, then we have already proved that |K| ≈ S2
and H0 (K) ∼ = Z, H1 (K) = 0, H2 (K) ∼ = Z and Hi (K) = 0, ∀ i ≥ 3. Hence
the Euler characteristic χ(K) = 2. Moreover, since the homology groups are
independent of the triangulation, the Euler characteristic χ(K) is independent
of K, i.e., χ(K) = χ(K ) for any other triangulation K of S2 .
Theorem 4.5.10. (Euler’s Theorem). If S is any simple polyhedron with V ver-

tices, E edges and F faces, then V − E + F = 2.
Proof. Note that the boundary of S is a polyhedral complex K, i.e., it consists of
properly joined 2-dimensional convex polygons. If all these bounding polygons
are triangles, then K is a simplicial complex such that |K| is homeomorphic to
S2 and the theorem follows directly from the Euler-Poincaré Theorem because
V = α0 , E = α1 and F = α2 . Hence we should consider the case when K
consists of polygonal faces which are not triangular. Suppose τ is one such face
having n0 vertices and n1 edges. When we compute (Vertices − Edges + Faces)
for the face τ , we find that the number is n0 − n1 + 1. Note that τ can always
be subdivided into triangles by taking a vertex v in the interior of τ and joining
v with each vertex of τ . See Fig. 4.7(a).
Fig. 4.7(a): Contribution of τ is invariant
In the above triangulation of τ , one new vertex and n0 new edges have
been added. Also, the face τ has been now replaced by n0 triangles. Hence
the contribution of τ in the computation of (vertices − edges + faces) =
(n0 + 1) − (n1 + n0 ) + n0 = n0 − n1 + 1. This shows that the sum V − E + F
is not changed when we triangulate every face of K. In other words, we can
assume that K is a simplicial complex. Now, the theorem follows directly from
the Euler-Poincaré Theorem because V − E + F = α0 − α1 + α2 which is equal
to the Euler characteristic of S2 , viz., 2. Therefore, V − E + F = 2 for any
simple polyhedron.
Euler’s Theorem has a very interesting consequence. Note that a solid

tetrahedron and a solid cube both are simple polyhedron which are regular.
In the first case the faces are equilateral triangles and in the other case the
faces are squares. Now the question is: Can we have simple regular polyhedra
whose faces are regular pentagons, regular hexagons, or any n-gons, n ≥ 3?
We can also ask whether there are other simple regular polyhedra, besides
tetrahedron, whose faces are triangular? It is known since the time of Greeks
that besides tetrahedron and cube, there are also octahedron, dodecahedron
and icosahedron which are simple regular polyhedra. It is a consequence of
Euler’s Theorem that these are the only five simple regular polyhedra – these
are known as Platonic Solids. We have
Theorem 4.5.11. There are only five simple, regular polyhedra.

Proof. Suppose S is a simple regular polyhedron with V vertices, E edges and
F faces. Let m be the common number of edges which meet every vertex, and
let n be the number of edges in each face of S. Also, note that n ≥ 3. By Euler’s
theorem,
V −E+F =2
and by our assumptions,
mV = 2E = nF.
This means (nF )/m − (nF )/2 + F = 2, i.e., F (2n − mn + 2m) = 4m. Clearly,
we must have 2n − mn + 2m > 0. Since n ≥ 3,
2m > n(m − 2) ≥ 3(m − 2) = 3m − 6
which at once implies that m < 6. Geometrically, this says that in S not more
than 5 edges can meet at a vertex. Next, the relation
F (2n − mn + 2m) = 4m, n ≥ 3, m < 6,
permits the following possibilities for the values of (m, n, F ), viz.,

(m, n, F ) = (3, 3, 4), (3, 4, 6), (4, 3, 8), (3, 5, 12) and (5, 3, 20). To examine all
these cases, let n = 3. Then we should have F (6 − m) = 4m. Now, this last
equation says that m cannot be 1 or 2. The case m = 3 shows that F = 4,
and m = 4 shows that F = 8. In other words, when the faces are triangles,
we have only two possibilities, viz., tetrahedron and octahedron. Proceeding
likewise, we can similarly prove that there are only the five possibilities men-
tioned above, i.e., the only five simple regular polyhedra are tetrahedron, cube,
octahedron, dodecahedron and icosahedron (see Fig. 4.8).
Orientability and Homology

We now prove a result which is at the heart of a class of important the-
orems in algebraic topology called “duality theorems”. We will not prove here
the theorems themselves, but will explain the basic concept of orientability,
which is needed in their hypotheses and show that this is indeed a topological
concept. We have already seen what is meant by an oriented complex, viz., a
complex K in which every simplex is given a fixed but arbitrary orientation.
Here, we will discuss the idea of orientability for a class of complexes, motivated
by manifolds.
Definition 4.5.12. A simplicial complex K is said to be an n-pseudomanifold if

it satisfies the following conditions:
1. Each simplex of K is a face of some n-simplex of K.
2. Each (n − 1)-simplex is a face of exactly two n-simplexes of K.
Fig. 4.8: The five Platonic solids
3. Given a pair σ1n and σ2n of n-simplexes of K, there is a sequence of n-

simplexes τ1 , τ2 , . . . , τk such that τ1 = σ1n , τk = σ2n and for each i, τi ∩ τi+1
is a (n − 1)-face of both τi and τi+1 , i = 1, 2, . . . , k − 1.
The last condition says that K is connected. One can easily verify
that if K is the 2-skeleton of Cl(σ 3 ), where σ 3 = v0 , v1 , v2 , v3 , then K is
a 2-pseudomanifold. Similarly, the triangulation K of the projective plane
P2 (see Fig. 4.6) is a 2-pseudomanifold. Likewise, one can easily verify
that the triangulations of torus T and Klein Bottle given by Fig. 4.5 and
Exercise 4 are also 2-pseudomanifolds. An example of an n-dimensional
pseudomanifold K is obtained by letting K be the n-skeleton of Cl(σ n+1 ),
where σ n+1 = v0 , . . . , vn+1 is an (n + 1)-simplex.
Let us recall (see Chapter 1) that a connected Hausdorff topological space

X is said to be a topological n-manifold if each point x ∈ X has an open neigh-
bourhood U such that U is homeomorphic to the Euclidean space Rn . Some
people also require that X be a paracompact space or second countable or
metrizable; in fact, all these three conditions are equivalent for a manifold X
(see Theorem 1.8.2). The manifolds, which appeared in the work of G.F.B. Rie-
mann (1826-1866), form a very important class of topological spaces. There is a
nice relationship between compact n-manifolds and n-pseudomanifolds, viz., if
K is an n-pseudomanifold, then |K| is a compact, connected polyhedron, which
is not necessarily a topological n-manifold. (If we take two hollow tetrahedra
with a common base, remove interior of the triangles constituting the common
base and identify the two remaining vertices, then the resulting complex K is
a 2-pseudomanifold, but |K| is not a 2- manifold.) However, if X is a compact,
connected, topological n-manifold, which is also triangulable by a complex K,
then K is a n-pseudomanifold. In fact, any triangulation of X would be an n-
pseudomanifold. This fact is some times very useful in computing the homology
groups of a triangulable n-manifold. We are going to illustrate this somewhat
in detail. First, let us have the following result concerning compact surfaces,
every triangulation of which is a 2-pseudomanifold.
Theorem 4.5.13. Let K be a 2-pseudomanifold with α0 vertices, α1 edges and

α2 triangles. Then
(i) 3α2 = 2α1
(ii) α1 = 3(α0 − χ(K))
(iii) α0 ≥ (1/2)(7 + 49 − 24χ(K)).
Proof. A triangle has 3 edges and so there are a total of 3α2 number of edges in
the complex K. But we note that every edge has been counted twice since each
edge is a face of exactly two triangles. Therefore, α1 = (3α2 )/2. We already
know by Euler-Poincaré Theorem that
α0 − α1 + α2 = χ(K).
This means α0 − α1 + (2/3)α1 = χ(K), i.e., α1 = 3(α0 − χ(K)), which proves
(ii).
To prove the part (iii), note that α0 ≥ 4. Obviously, the maximal number of
possible edges in K is α0 (α0 −1)/2. Hence, α1 ≤ α0 (α0 −1)/2. Now, 6α2 = 4α1 ,
i.e., 2α1 = 6α1 − 6α2 or α0 (α0 − 1) ≥ 6α1 − 6α2 . This means
α02 − α0 − 6α0 ≥ 6α1 − 6α2 − 6α0

= −6χ(K)
α02 − 7α0 ≥ −6χ(K).
Therefore,
(2α0 − 7)2 ≥ 49 − 24χ(K)
and so
α0 ≥ (1/2)(7 + 49 − 24χ(K)).
The above theorem, which is a consequence of the Euler-Poincaré Theo-

rem, is quite useful in determining a triangulation of a given 2-pseudomanifold
with minimum number of vertices, edges, and triangles. Since the computa-
tion of homology groups of a complex K could be quite complicated, we can
minimize the complexity of calculation by choosing a suitable triangulation
having minimum number of simplexes. It is evidently true though that finding
a minimum triangulation itself could very well be a problem.
Example 4.5.14. Consider the 2-sphere S2 and note that χ(S2 ) = 2. Hence, by
the above theorem, we see that in any triangulation of S2 we must have
α0 ≥ 1/2(7 + 49 − 24χ(K)) = 4
α1 = 3(α0 − χ(K)) ≥ 3(4 − 2) = 6

α2 = (2/3)α1 ≥ 4.
Thus, in any triangulation of S2 , there must be at least 4 vertices, 6 edges and
4 triangles. This shows that a tetrahedron gives a minimal triangulation of S2 .
Example 4.5.15. Now, consider the projective plane P2 which is also a 2-

manifold. We have already seen that H0 (P2 ) ∼ = Z, H1 (P2 ) ∼
= Z2 and
2 2
H2 (P ) = 0. Hence, χ(P ) = 1. Therefore, by the above theorem,
α0 ≥ 6, α1 ≥ 15, α2 ≥ 10.
This shows that in any triangulation of P2 , there must be at least 6 vertices,

15 edges and 10 triangles. Hence the triangulation given in Fig.4.6 is minimal.
We can similarly try and find out possible triangulations of torus, Klein
bottle which are also minimal. Note that the lower bound for the number of
vertices α0 in the case of T #T is an irrational number. Thus, lower bounds
α0 , α1 , α2 may not actually be always attained.
Now let us consider the following concept of “orientability” of a complex.
Definition 4.5.16. Let K be an n-pseudomanifold. If σ n−1 is a (n − 1)-simplex,

then it must be the common face of two n-simplexes, say, σ1n , σ2n . An orientation
of K, which has the property that
[σ1n , σ n−1 ] = −[σ2n , σ n−1 ]
for every (n − 1)-simplex σ n−1 , is called a coherent orientation on K. An n-

pseudomanifold K is said to be orientable if it admits a coherent orientation.
The above definition can be extended to topological manifolds which are

triangulable. We will say that a n-manifold X is orientable if it has a triangu-
lation K having a coherent orientation. In fact, it can be proved (see Theorem
4.5.18) that orientability of a manifold is a topological property, i.e., if a par-
ticular triangulation of an n-manifold X has a coherent orientation, then any
other triangulation of X also has a coherent orientation. An n-manifold M
with boundary ∂M is said to be orientable if its double DM is orientable;
here DM is the n-manifold obtained from two copies of M by identifying their
boundaries. It is easily seen that 2-spheres, torus, etc., are orientable whereas
projective plane, Klein bottle and Möbius band (a manifold with boundary)
are nonorientable 2-manifolds. More generally, we have
Proposition 4.5.17. The n-sphere Sn , n ≥ 1, is orientable.

Proof. Let us consider the (n + 1)-simplex σ n+1 = v0 , . . . , vn+1 . We will prove
that its boundary complex K is orientable. For each integer j, 0 ≤ j ≤ n + 1,
let
σj = v0 , . . . , v̂j , . . . , vn+1
be the n-simplex where vj is omitted. Let us give σj the positive orientation
defined by the above ordering when j is even and give it opposite orientation
if j is odd. Then (n − 1)-simplex, positively oriented by
+σij = +v0 , . . . , v̂i , . . . , v̂j , . . . , vn+1
is a face of two simplexes σi and σj . We have, therefore, given an orientation to

all n-simplexes as well as to all (n − 1)-simplexes of the boundary complex K
of σ n+1 . Now, give arbitrary orientation to the remaining simplexes. One can
easily see that with this orientation on K,
[σi , σij ] = −[σj , σij ]
for each pair of n-simplexes σi , σj and their common face σij . Hence K has a
coherent orientation.
The above result, which should be visually verified in the case of S2 , also
follows from the next theorem which determines the orientability of a compact
n-manifold in terms of its n-th integral homology and shows that orientability
of manifolds is a topological concept.
Theorem 4.5.18. A triangulable connected n-manifold X is orientable if and

only if Hn (K) is nontrivial for some triangulation (hence, every triangulation)
K of X.
Proof. Suppose X is triangulable. This means there is a triangulation K of
X which is an n-pseudomanifold. Now, suppose K has a coherent orientation.
Thus, if σ n−1 is a (n − 1)-face of two n-simplexes σ1n and σ2n of K, then
[σ1n , σ n−1 ] = −[σ2n , σ n−1 ].
As a result, note that for any fixed integer m, the n-chain

c= m · σn
σ n ∈K
is easily seen to be a n-cycle. This means Zn (K)

= 0. Since Bn (K) = 0, we
find that Hn (K)
= 0. This proves the direct part.
Conversely, suppose for some given orientation of K, Hn (K)
= 0 and let

z = σn ∈K mi σin be an n-cycle which is nonzero. Now, using the fact that
i
∂(z) = 0, and that each (n − 1)-simplex is a face of exactly two n-simplexes
and also the fact that |K| is connected, it follows that mi = ±m0 , ∀ i
= 0
where m0 is a fixed coefficient. By changing the orientation of σin for which
mi = −m0 , we find that

mi σin = m0 ( σin )
σin ∈K σin ∈K
is an n-cycle. However, this implies that any (n − 1)-face τ of K must have

positive incidence number w.r.t. one n-simplex and negative incidence number
w.r.t. the other n-simplex. This means the changed orientation of K is
coherent, and so X is orientable.
The above theorem tells us that the projective plane P2 and Klein bottle,
which have trivial homology in dimension 2 with coefficients in Z, are nonori-
entable. Here, the integer coefficients are important because even though P2
is nonorientable, H2 (P2 ; Z2 ) ∼
= Z2 , which is nonzero. We should also mention
here that none of the even-dimensional projective spaces P2n , n ≥ 1, is ori-
entable, but all odd-dimensional ones P2n+1 are orientable. This follows from
their homology groups and the above theorem (Exercises Section 4.7).
Exercises
1. Determine which of the following surfaces is orientable:
(a) {(x, y, z) ∈ R3 : x2 /a2 + y 2 /b2 + z 2 /c2 = 1, a > b > c}
(b) Torus (c) Projective plane (d) Klein bottle (e) Möbius band M
(f) Double torus (Connected sum of two tori.)
2. Use Euler-Poincaré theorem to compute the Euler characterstic of the
spaces given in Exercise 1.
3. Show that for a simplicial complex K the Euler characterstic of K com-
puted using homology with integer coefficients is the same as that com-
puted using homology with coefficients in Q.
4. Show that torus and Klein bottle are 2-pseudomanifolds. Draw minimal
triangulations of torus and Klein bottle, and justify your answer.
5. If K is a 2-pseudomanifold then show that χ(K) ≤ 2.
6. Give an example of a 3-pseudomanifold K so that |K| is not a 3-manifold.
7. Triangulate the Möbius band in a simple way so that its centre circle is
a subcomplex. Orient the boundary circle and the central circle. Let the
resulting cycles be denoted by z1 and z. Prove that z1 is homologous to
2z or −2z.
8. Compute the homology groups of the following simplicial complexes:
(a) Three copies of the boundary of a triangle all joined at a vertex.
(b) Two hollow tetrahedra glued together along an edge.
9. Let S1 , S2 be 2 compact surfaces with triangulations K1 , K2 respectively.

Choose triangles ABC and A B C from each of |K1 |, |K2 | respectively,
remove their interiors and identify the boundaries of the two triangles
by the obvious linear homeomorphism. Then the resulting quotient space,
denoted by |K1 |#|K2 |, is called the connected sum of |K1 | and |K2 |. Prove
that this connected sum gives a triangulation of S1 #S2 , and hence deduce
that
χ(S1 #S2 ) = χ(S1 ) + χ(S2 ) − 2.
10. Using the classification theorem for compact surfaces compute the Euler
characterstic of all compact 2-manifolds.
11. Let us define a rectilinear polyhedron to be topologically regular if the
number of edges meeting at each vertex is same, e.g., a tetrahedron whose
faces are triangles, not necessarily equilateral triangles; a solid lamina
whose faces are rectangles, not necessarily squares, are examples of topo-
logical regular polyhedra. Prove that there are exactly five classes of simple
topologically regular polyhedra.
4.6 Induced Homomorphisms
Recall that a simplicial map f : K → L is a map from the set of vertices

of K to the set of vertices of L which has the property that if σ q = v0 , . . . , vq
is a q-simplex of K, then f (v0 ), . . . , f (vq ) is a simplex of L of dimension
r ≤ q: If all the vertices f (v0 ), . . . , f (vq ) are not distinct then by deleting
repetitions, we find that r < q, otherwise r = q.
Now, we have
Definition 4.6.1. Let f : K → L be a simplicial map. Then for each q ≥ 0, we

define a homomorphism f# : Cq (K) → Cq (L) by putting

f (v0 ), . . . , f (vq ), if f (vi )’s are distinct
f# (v0 , . . . , vq ) =
0, otherwise,
for each simplex v0 , . . . , vq of K and then extending it linearly on Cq (K).
It is seen at once that f# is a well-defined map because exchanging any

pair of vertices in the left hand side of the above definition results in a corre-
sponding change in the right hand side. The collection {f# : Cq (K) → Cq (L)}
of homomorphisms (we have suppressed the index q deliberately from f# and
also from ∂ because suffixing them would make the notations cumbersome) is
called the chain map induced by the simplicial map f . This is proved in the
next
4.6. Induced Homomorphisms 159
Proposition 4.6.2. If f : K → L is a simplicial map, then the induced sequence

{f# : Cq (K) → Cq (L)} of homomorphisms commutes with the boundary ho-
momorphism ∂ of the chain complex and hence, f# induces a homomorphism
f∗ : Hq (K) → Hq (L) in each dimension q.
Proof. Let us consider the following diagram q ≥ 1 :
- Cq (K) ∂ - Cq−1 (K) -
f# f#
? ?
- Cq (L) - Cq−1 (L) -
∂
We must show that f# ∂ = ∂f# , i.e.,
(∗) f# (∂v0 , . . . , vq ) = ∂(f (v0 ), . . . , f (vq ))
for each oriented q-simplex v0 , . . . , vq of K. Since there may be repetitions

in the vertices f (v0 ), . . . , f (vq ), we let τ be the simplex of L spanned by these
vertices, and note that dim τ ≤ q. We consider the following three cases:
Case I: dim τ = q. In this case all the vertices f (v0 ), . . . , f (vq ) are distinct and
the result follows directly from definitions of f# and ∂.
Case II: dim τ ≤ q − 2. In this case both sides of (∗) are zero because in the
set of vertices f (v0 ), . . . , f (vq ), at least three are identical.
Case III: dim τ = q − 1. In this case two vertices from f (v0 ), . . . , f (vq ) are iden-
tical and we can assume that f (v0 ) = f (v1 ), and f (v2 ), . . . f (vq ) are distinct.
Then, by definition, the right side of (∗) vanishes. The left side has only two
nonzero terms, viz.,
f (v1 ), f (v2 ), . . . , f (vq ) − f (v0 ), f (v2 ), . . . , f (vq ).
Since f (v0 ) = f (v1 ) these terms also cancel each other and the left side is also
zero.
Finally, to prove that for each q ≥ 0, f# induces a homomor-
phism f∗ : Hq (K) → Hq (L), we note that Hq (K) = Zq (K)/Bq (K) and
Hq (L) = Zq (L)/Bq (L). It is easily verified that f# (Zq (K)) ⊆ Zq (L) and
f# (Bq (K)) ⊆ Bq (L), i.e., f# maps cycles into cycles and boundaries
into boundaries. It follows, therefore, that f# induces homomorphism
f∗ : Hq (K) → Hq (L) defined by f∗ ({zq }) = {f# (zq )}.
The next result now follows from the definitions of induced homomor-
phism.
Theorem 4.6.3. (a) If IK : K → K be the identity simplicial map, then the

induced homomorphism (IK )∗ : Hq (K) → Hq (K) is identity for each q ≥
0.
(b) If f : K → L and g : L → M are simplicial maps, then for all q ≥ 0,
(g ◦ f )∗ = g∗ ◦ f∗ : Hq (K) → Hq (M ).
Topological Invariance of Homology Groups
Now, suppose K, L are two complexes and f : K → L is a simplicial map.

Then we know that f induces a continuous map |f | : |K| → |L|. Conversely,
given any continuous map h : |K| → |L|, we want to know whether there is
a induced homomorphism h∗ : Hq (K) → Hq (L) in homology. If h is induced
by a simplicial map, as indicated above, then the answer to the above ques-
tion is trivially, “yes”. If, however, h is not induced by a simplicial map, as
is mostly the case, then what can we do? Take a simplicial approximation
g : K (k) → L of h and note that the induced map |g| : |K| = |K (k) | → |L| is ho-
motopic to h. Now, by the previous result there is the induced homomorphism
g∗ : Hq (K (k) ) → Hq (L). If we could prove (i) there is an isomorphism
μ : Hq (K) → Hq (K (k) ) whatever be the integer k, and (ii) g : K (k) →
L, g : K (m) → L are any two simplicial approximations of h implies g∗ μ =
g∗ μ , then we could unambiguously define a homomorphism h∗ : Hq (K) →
Hq (L) by the commutativity of the following diagram:
Hq (K)
h∗ - Hq (L)

*
μ
? g∗
Hq (K (k) )
i.e., we define h∗ = g∗ ◦ μ.
Later, in Sections 4.9.1 and 4.9.2, we will show that (i) and (ii) above
can actually be proved. Hence presently we assume that any continuous map
h : |K| → |L| induces a homomorphism h∗ : Hq (K) → Hq (L) for all q ≥ 0 as
defined above. We will also assume the following in which part (i) is, of course,
obvious.
Theorem 4.6.4. (i) The identity map I|K| : |K| → |K| induces the identity
homomorphism IHq (K) : Hq (K) → Hq (K) in each dimension q.
(ii) If f : |K| → |L| and g : |L| → |M | are any two continuous maps, then
the induced homomorphism (g ◦ f )∗ = g∗ ◦ f∗ : Hq (K) → Hq (M ) in each
dimension q ≥ 0.
We have only stated the above theorem. The proof is postponed to

Section 4.9 for the simple reason that besides being lengthy, it involves several
more concepts and algebraic results, which we have not discussed so far.
The statement is given because, without interrupting continuity, we want to
4.6. Induced Homomorphisms 161
illustrate some of the classical applications of homology groups right now.
The most important result which we should mention here and which fol-
lows from the preceding theorem is that the homology groups of a polyhedron
are its topological invariants. We have
Theorem 4.6.5. Let X be a polyhedron and K, L be any two triangulations of

X. Then for each q ≥ 0, Hq (K) ∼
= Hq (L).
Proof. Since K and L are triangulations of the space X, there are homeo-
morphisms f : |K| → X, g : |L| → X. This means h = g −1 ◦ f : |K| → |L|
is a homeomorphism. Let h−1 : |L| → |K| be the inverse of h. Consider the
induced homomorphism h∗ : Hq (K) → Hq (L) and h−1 ∗ : Hq (L) → Hq (K). We
know that h−1 ◦ h = I|K| and h ◦ h−1 = I|L| . Hence, by (i) and (ii) of the
preceding theorem h−1 −1
∗ ◦ h∗ = IHq (K) and h∗ ◦ h∗ = IHq (L) . This says that
h∗ : Hq (K) → Hq (L) is an isomorphism.
Homology of a polyhedron
The above theorem says that if X is a compact polyhedron, then we

can define simplicial homology groups Hq (X), q ≥ 0, of X simply by setting
Hq (X) = Hq (K), where K is some triangulation of X. By the above result,
the simplicial homology groups of X are well-defined up to isomorphism only.
It may be noticed that if K is a triangulation of X, then |K| is a rectilinear

polyhedron whereas X is a polyhedron which may not be rectilinear, i.e., the
simplexes of X may be curved. The two spaces X and |K| are not identical,
but they are homeomorphic. The standard n-sphere Sn and the standard n-
disk Dn are all compact polyhedra, and we have already computed homology
groups of a triangulation (see Section 4.5) of Sn and a triangulation of Dn .
Thus, we can now unambiguously assert that the simplicial homology groups
(up to isomorphism) of these spaces are given by:

Z, q = 0, n
Hq (Sn ) ∼
=
0, otherwise

Z q=0
Hq (Dn ) ∼
=
0, otherwise.
Likewise we can also say (see Section 4.4 and exercises therein) that the sim-
plicial homology of the torus T , the projective plane P2 , the Möbius band M
and the Klein bottle K are given as follows:
⎧
⎨ Z, q = 0, 2
Hq (T ) ∼
= Z ⊕ Z, q = 1
⎩
0, otherwise,
⎧
⎨ Z, q=0
Hq (P2 ) ∼
= Z/2Z, q = 1
⎩
0, otherwise,

Z, q = 0, 1
Hq (M ) ∼ =
0, otherwise,
⎧
⎨ Z, q=0
Hq (K) ∼
= Z ⊕ Z/2Z, q =1
⎩
0, otherwise.
Induced homomorphism
We should also observe that if f : X → Y is a continuous map from a poly-

hedron X to another polyhedron Y , then choosing triangulations h : |K| → X
and k : |L| → Y , the continuous map k −1 f h : |K| → |L| induces a homomor-
phism (k −1 f h)∗ : Hq (K) → Hq (L) and we say that f : X → Y induces the
homomorphism f∗ : Hq (X) → Hq (Y ) by putting f∗ = (k −1 f h)∗ . This is all
right in so far as we are studying homology groups up to isomorphism only;
if we want to distinguish between two isomorphisms, then this identification is
not acceptable. To see it further, let us now choose a different triangulation,
say, h1 : |K1 | → X, k1 : |L1 | → Y . Then the induced homomorphism would
be (k1−1 f h1 )∗ : Hq (K1 ) → Hq (L1 ). We know that there are homeomorphisms
h−1 −1
1 h : |K| → |K1 |, k1 k : |L| → |L1 | which induce isomorphisms in homology
and so we have a commutative diagram
(k−1 f h)∗
|K| - |L|
(h−1
1 h)∗ (k1−1 k)∗
? ?
|K1 | - |L1 |
(k1−1 f h1 )∗
because (k1−1 k)−1 (k1−1 f h1 )(h−1

1 h) = k
−1
f h. Therefore, if we identify the
groups Hq (K) and Hq (K1 ), etc., then we can as well identify the induced
homomorphisms (k −1 f h)∗ and (k1−1 f h1 )∗ . Thus the induced homomorphisn f∗
is defined up to this identification of isomorphisms.
4.7 Some Applications
As stated earlier, having determined the homology groups of disks Dn and

spheres Sn , now we come to a few classical applications of homology groups
and induced homomorphisms. We call attention to Theorem 4.5.5 for these
applications.
4.7. Some Applications 163
Theorem 4.7.1. (Invariance of dimension). If m

= n, then
(i) Sm is not homeomorphic to Sn , and
(ii) Rm is not homeomorphic to Rn .
Proof. (i) Suppose, on the contrary, there is a homeomorphism f : Sm → Sn . Let
h : |K| → Sm and k : |L| → Sn be triangulations of Sm and Sn , respectively.
We have already computed the homology groups of K and L. Since f is a
homeomorphism, the map g = k −1 f h : |K| → |L| is also a homeomorphism.
Let g −1 : |L| → |K| be its inverse. We consider the induced homomorphisms
g∗ : Hm (K) → Hm (L) and g∗−1 : Hm (L) → Hm (K) in simplicial homology.
Since g −1 g = I|K| , we find by theorem 4.6.4, that
g∗−1 g∗ = (g −1 g)∗ = (I|K| )∗ = IHm (K) .
Similarly, we find that g∗ g∗−1 is also identity on Hm (L). Therefore,

g∗ : Hm (K) → Hm (L) is an isomorphism. But this is a contradiction, because
Hm (K) = Z whereas Hm (L) = 0, since m
= n. This proves the theorem.
(ii) Again, suppose Rm is homeomorphic to Rn . Since these are locally compact
Hausdorff spaces, their one-point compactifications, viz., Sm and Sn must
also be homeomorphic, which is a contradiction to (i) proved above. Hence, if
m
= n, Rm cannot be homeomorphic to Rn .
If Dn denotes the n-dimensional disk (closed), then its boundary is home-

omorphic to the (n − 1)-dimensional sphere Sn−1 . Thus, Sn−1 is a compact
subset of Dn . Recall that a subspace A of a space X is said to be retract of X
if there is a continuous map r : X → A such that r(a) = a for a ∈ A. Now, one
can ask the question: Is Sn−1 is a retract of Dn ? If n = 1, this is clearly impos-
sible because D1 = [−1, 1] is connected whereas S0 = {−1, 1} is disconnected.
If n ≥ 2, then also the answer to the above question is “no”. We have
Theorem 4.7.2. (No-retraction Theorem). The sphere Sn−1 cannot be a retract

of Dn , for any n ≥ 1.
Proof. If possible, suppose there is a retraction r : Dn → Sn−1 . Let i : Sn−1 →
Dn be the inclusion map. Then, clearly, ri = ISn−1 . The case when n = 1 is
clear because D1 is connected whereas S0 is not. Hence we assume that n > 1.
Let h : |K| → Dn and k : |L| → Sn−1 be triangulations of the disc and sphere.
Notice that we have continuous maps k −1 rh : |K| → |L| and h−1 ik : |L| → |K|
such that their composite is identity, i.e., (k −1 rh)(h−1 ik) = I|L| . This means
the composite
Z = Hn−1 (L) → Hn−1 (K) → Hn−1 (L) = Z
is identity on Z by theorem 4.6.4. But this last map factors through the zero
group Hn−1 (K) = 0, which is a contradiction.
Recall that a space X is said to have fixed-point property if for every

continuous map f : X → X, there exists a point x0 ∈ X such that f (x0 ) = x0 .
From the first course in real analysis, we know that the unit interval [0,1] has
the fixed-point property. The following important theorem is a far reaching
generalization of this result.
Theorem 4.7.3. (Brouwer’s Fixed-Point Theorem). Let f : Dn → Dn , n ≥ 1 be

any continuous map. Then f has at least one fixed point.
Proof. Suppose f has no fixed points. This means for every x, f (x)
= x. Now,
we define a map g : Dn → Sn−1 as follows: For any x ∈ Dn , consider the
line segment (vector) joining f (x) to x and extend it. Then, this vector when
produced will meet Sn−1 exactly at one point, which we call g(x) (see Fig. 4.9).
First, we prove that g : Dn → Sn−1 is continuous. Note that for each x, the
nonzero vector x − f (x) can be multiplied by a unique positive scalar, say λ,
such that g(x) = f (x) + λ(x − f (x)), and this λ depends on x. Since g(x) lies
on Sn−1 , its norm is 1 and so ||f (x) + λ(x − f (x))|| = 1. This implies that
||f (x)||2 + λ2 ||(x − f (x))||2 + 2λf (x) · (x − f (x)) = 1.
This is a quadratic equation in λ having only one positive real root λ.

Therefore, by the formula for the roots of a quadratic equation, we get
λ = −f||x−f
(x)·(x−f (x))
(x)||2 . This proves that λ is a continuous function of x and
therefore g is a continuous function. But g is clearly a retraction of Dn onto
Sn−1 , a contradiction to the no-retraction theorem.
g(x)
x
f(x)
Dn Sn-1
f(y)
y
g(y)
Fig. 4.9: The map g becomes a retraction

The following result gives an interesting application of the Brouwer fixed-
point theorem to a result of linear algebra. We have
Proposition 4.7.4. Let A be an n × n real matrix with positive entries. Then A

has a positive eigen value.
Proof. Consider the Euclidean space Rn , and observe that A determines a lin-
ear transformation from Rn to itself. Also, note that if x = (x1 , . . . , xn ) ∈ Rn ,
4.7. Some Applications 165
where xi ≥ 0 for all i = 1, . . . , n, then, since all entries of A are positive, Ax

also has the same property, i.e., A maps the positive octant P including its
boundary to itself. Furthermore, if one entry of x is positive, then all entries
of Ax are positive. Let us also point out that if Sn−1 denotes the unit sphere
in Rn , then Sn−1 ∩ P is homeomorphic to the (n − 1)-disk Dn−1 . Now, we
can define a map f : Sn−1 ∩ P → Sn−1 ∩ P by putting f (x) = Ax/Ax.
Then evidently f is continuous. Hence, by the Brouwer’s fixed point theorem,
there exists a non-zero vector x0 ∈ Sn−1 ∩ P such that f (x0 ) = x0 , i.e.,
Ax0 /Ax0 = x0 . This says that Ax0 = Ax0 x0 which means Ax0 (
= 0) is
an eigen value of A.
Knowing that every continuous map f : X → X has a fixed point is an

important property of the space X, and it has interesting applications. Let
us illustrate this by another example. Suppose we have a set of continuous
functions fi : Rn → R, i = 1, 2, . . . , n, each of which has a nonempty zero set,
i.e., there exists some points of Rn where fi is zero. Now, the question is: Is
there a common zero of all these functions fi ? In other words, do the following
system of simultaneous equations has a solution in Rn :
f1 (x1 , x2 , . . . , xn ) = 0
f2 (x1 , x2 , . . . , xn ) = 0
..
.
fn (x1 , x2 , . . . , xn ) = 0
This question is really a question whether the space Rn has the fixed-point
property. To see why this is so, let us consider the following continuous maps
gi : Rn → R defined by
gi (x1 , x2 , . . . , xn ) = fi (x1 , . . . , xn ) + xi
for i = 1, 2, . . . , n. We consider the map h : Rn → Rn defined by
h(x1 , . . . , xn ) = (g1 (x1 , . . . , xn ), . . . , gn (x1 , . . . , xn )).
Notice that h is continuous because each gi is continuous. Now, observe that h
has a fixed point (a1 , . . . , an ) ∈ Rn if and only if for each i = 1, . . . , n, we have
gi (a1 , . . . , an ) = ai ,
and that will happen if and only if
f1 (a1 , . . . , an ) = 0 = f2 (a1 , . . . , an ) = . . . = fn (a1 , . . . , an ),
i.e., all the fi s have a common zero. It may be remarked that the Euclidean
space Rn used in this example does not have the fixed-point property because
translations by nonzero vectors do not have fixed points. On the other hand, by
Brouwer’s theorem, the cube I n , where I = [−1, 1], has the fixed-point property
and so can very well be used for the space X to conclude that the simultaneous
equations have a solution.
Exercises
1. If a space X has the fixed-point property, then show that any space Y
which is homeomorphic to X also has the fixed-point property.
2. If A is a retract of X and if X has the fixed-point property, then show
that A also has the fixed point property.
3. Let X be a compact metric space and f : X → X be a fixed-point free
map. Prove that there is an > 0 such that d(x, f (x)) > for all x ∈ X.
4. By giving concrete examples prove that the 2-sphere, the torus and the
Klein bottle do not have the fixed-point property.
5. Prove that the projective plane P2 has the fixed-point property (this may
be bit difficult at this stage!).
6. If a polyhedron A ⊂ X is a retract of a polyhedron X, then show that for
q ≥ 0, Hq (A) is a direct summand of Hq (X).
7. Prove that an injective map between two polyhedra does not necessarily
induce an injective map between their homology groups.
8. Show that the following conditions are equivalent:
(a) Sn−1 is not a retract of Dn .
(b) Dn has the fixed-point property.
(c) The n-simplex δn has the fixed-point property.
9. Show that the map g : RP3 → RP3 defined by the linear transformation
T : R4 → R4 given by T (x1 , x2 , x3 , x4 ) = (−x2 , x1 , −x4 , x3 ) does not have
the fixed-point property. More generally, show that RP2k+1 does not have
the fixed-point property.
10. Prove that the homology groups of the real projective spaces RPn , n ≥ 2,
and that of the complex projective spaces CPn , n ≥ 1 are given by the
following (See Examples 1.1.4 (a) and 1.1.4 (b). The cases n ≥ 3 need
concepts not covered so far.)
(i) n is odd:
⎧
⎨ Z, q = 0, n
Hq (RP n ) ∼
= Z2 , 0 < q < n, q odd
⎩
0, otherwise,
(ii) n is even:
⎧
⎨ Z, q=0
Hq (RP n ) ∼
= Z2 , 0 < q < n, q odd
⎩
0, otherwise,
4.8. Degree of a Map and its Applications 167
(iii) The complex case

Z, q is even
Hq (CP n ) ∼
=
0, otherwise,
11. Show that any nonsingular linear transformation T : R3 → R3 defines a

continuous map g : RP2 → RP2 which has a fixed point. More generally,
prove that RP2k has the fixed-point property for any k ≥ 0.
12. Prove that any compact locally contractible space has the fixed-point
property.
4.8 Degree of a Map and its Applications
Recall that a homomorphism f : Z → Z of the infinite cyclic group is

completely determined by the image of its generator 1 ∈ Z under the map f ,
i.e., f is simply multiplication by the integer f (1) = n. This fact is used in the
following:
Definition 4.8.1. Let f : Sn → Sn , (n ≥ 1) be a continuous map and h : |K| → Sn

be any triangulation of Sn . Then we know that f induces a homomorphism
(h−1 f h)∗ : Hn (K) → Hn (K). Since Hn (K) ∼ = Z, there is a unique integer
d such that for every element α ∈ Hn (K), (h−1 f h)∗ (α) = dα. This unique
integer d is called the degree of f and we denote it by deg f .
We must prove that the integer d does not depend on the chosen
triangulation h : |K| → Sn . For this, let k : |L| → Sn be another triangulation
of Sn . Note that φ = k −1 h : |K| → |L| and φ−1 = h−1 k : |L| → |K| are
homeomorphisms. Therefore, (k −1 f k)∗ (α) = (k −1 h)∗ (h−1 f h)∗ (h−1 k)∗ (α) =
φ∗ (h−1 f h)∗ φ−1
∗ (α) = φ∗ (d · (φ
−1
)∗ (α)) = φ∗ (φ−1 )∗ (d · α) = d · α, which says
−1
that (k f k)∗ is again multiplication by d. This proves the claim.
The following result is a consequence of the definitions.
Proposition 4.8.2. (a) The identity map ISn : Sn → Sn has degree +1.
(b) If f : Sn → Sn , g : Sn → Sn are continuous maps, then deg(g ◦ f ) =

deg g · deg f.
(c) The degree of any homeomorphism is ±1.
Proof. (a) This follows from the fact that the identity map ISn induces the
identity map in homology.
(b) Let k : |K| → Sn be a triangulation of Sn . Suppose deg f = n1 , deg g = n2 .

Then for any α ∈ Hn (K), we have
(k −1 g ◦ f k)∗ (α) = (k −1 gk)∗ (k −1 f k)∗ (α))
= (k −1 gk)∗ (n1 (α))
= n2 (n1 α)
= (n2 n1 )α.
Hence, deg(g ◦ f ) = deg g · deg f .
(c) Let h : Sn → Sn be a homeomorphism. Then h−1 ◦ h = ISn and hence, by
(a), we have
deg(h−1 ◦ h) = 1 = deg h−1 · deg h.
Since deg h, deg h−1 both are integers, we must have deg h = deg h−1 = +1 or
−1.
Next, we are going to prove that if f, g : Sn → Sn are two homotopic maps

then deg f = deg g. First, observe that if h : |K| → Sn is any triangulation of
Sn , then h−1 f h, h−1 gh : |K| → |K| are also homotopic. We will prove later (see
Theorem 4.8.4) that two homotopic maps induce identical homomorphism in
homology. Hence the result follows at once. However, proving that homotopic
maps induce identical homomorphisms in homology is much more involved
than proving the above result on degree directly. We will, therefore, use the
classical definition of the degree of a map due to L.E.J. Brouwer, which is
more intuitive than the “homology definition” given earlier, to prove the above
result on degree. Indeed, Brouwer’s definition of degree of a map f : Sn → Sn
really means the number of times the domain sphere “wraps” around the range
sphere. We have
Definition 4.8.3. Let f : Sn → Sn be a continuous map and let h : |K| → Sn

be a triangulation of Sn . Let φ : K (k) → K be a simplicial approximation of f,
where K (k) is the kth barycentric subdivision of K. For any positively oriented
n-simplex τ of K, let p be the number of positively oriented simplexes σ of K (k)
such that φ(σ) = τ and let q be the number of negatively oriented simplexes σ
of K (k) such that φ(σ) = τ . Then the integer p − q is independent of the choice
of τ, K, K (k) and φ. This integer is called the degree of the map f .
It is true that the two definitions of the degree of a map are equivalent,
but we will not prove this fact here. The curious reader may like to see
(Hocking and Young [10]) for a detailed discussion and proofs of all the
“independent statements” made above.
It follows from Brouwer’s definition that the map f : S1 → S1 defined by

f (z) = z n has degree n; the degree of a constant map f : Sn → Sn , n ≥ 1, is
zero; the degree of identity map ISn : Sn → Sn is one, etc. Now, we can prove
Theorem 4.8.4. If two continuous maps f, g : Sn → Sn are homotopic, then

deg f = deg g.
Proof. Let H : Sn × I → Sn be a homotopy starting from f and terminating
into g. If we put H(x, t) = ht (x), x ∈ Sn , t ∈ I, then each ht : Sn → Sn is
a continuous map and h0 = f, h1 = g. Now, it suffices to show that the map
I → Z defined by t ; deg ht is constant. The last fact will follow if we can prove
that the map from I → Z is continuous because I is connected and Z has the
discrete topology. This is what we now proceed to do: Let K be a triangulation
of Sn , and consider the open cover {ost(wi ) : wi is a vertex of K}. Suppose is
a Lebesgue number for the open cover. Since H is uniformly continuous, we can
find a positive real δ such that A ⊂ Sn , B ⊂ I with diam(A) < δ, diam(B) < δ
imply that diam(H(A × B)) < . Let K (k) be a barycentric subdivision of K
with mesh less than δ/2 so that for each vertex v of K (k) , diam(ost(v)) < δ.
Select a partition
0 = t0 < t1 < . . . < tq = 1
of I for which |tj − tj−1 | < δ, ∀j = 0, 1, . . . , q. Then for each vertex vi of K, the
set h(ost(vi ) × [tj−1 , tj ]) has diameter less than , for each j. Therefore, there
is a vertex wij of K such that h(ost(vi ) × [tj−1 , tj ]) ⊆ ost(wij ). Hence, for
each t ∈ [tj−1 , tj ], we can define a simplicial map φt by putting φt (vi ) = wij ,
which is evidently a simplicial approximation to ht , for all t ∈ [tj−1 , tj ]. Since
φt is same for all t ∈ [tj−1 , tj ], it follows from the Brouwer’s definition of the
degree of a map that ht has the same degree for all t ∈ [tj−1 , tj ], showing that
t ; deg ht is a continuous map from I to Z.
We must point out here that the converse of the above theorem was also
studied by Brouwer. In fact, he had proved that if f, g : S2 → S2 are two
continuous maps such that deg f = deg g, then f and g are homotopic. It was
H. Hopf (1894-1971) who proved the converse in full generality in the year
1927. We are not including the proof of the converse here, but give only the
statement of this beautiful theorem:
Theorem 4.8.5. (Hopf ’s Classification Theorem). Two continuous maps

f, g : Sn → Sn are homotopic if and only if they have the same degree.
We will remain contented only by stating the above theorem, though

its several generalizations are also known. The most important aspect of the
above theorem is to observe that the homotopy classes of maps from Sn to
itself are completely classified by the set of integers! Since we are leaving the
above theorem without giving its proof, we can as well state the following
theorem without proof which is regarded as one of the most satisfying theorems
of topology because it accomplishes the complete classification of compact 2-
manifolds without boundary (called closed 2-manifolds or closed surfaces) up
to homeomorphisms.
Theorem 4.8.6. Two closed surfaces are homeomorphic if and only if they have
the same Betti numbers in all dimensions.
The above theorem says that any two closed 2-manifolds X and Y
are homeomorphic if and only if Hq (X) ∼ = Hq (Y ), ∀q ≥ 0. This is one in-
stance where the homology groups provide complete classification of closed
2-manifolds. Note that we have defined homology groups only for those com-
pact topological spaces which are triangulable, i.e., are polyhedra. Hence it
is natural to ask as to which manifolds are triangulable? Poincaré had asked
this question for any n-manifold, n ≥ 2, not necessarily compact – there is a
suitable definition of simplicial complexes which are not necessarily finite. It
is now well known through the works of several mathematicians that every 2-
manifold, as well as every 3-manifold, is triangulable. Hence a surface is always
triangulable whether it is compact or not. This is the reason that in the above
theorem triangulability of the surfaces is not a part of the hypothesis. We must
also mention here that the above theorem really shows the power of homology
groups. In fact, this theorem was known as early as 1890 through the works of
C. Jordan (1858–1922) and A.F. Möbius (1790–1860). As we know, C. Jordan
is known for his “Jordan Curve” Theorem, and A.F. Möbius is known for his
“Möbius band” through which he introduced the idea of “orientability”. The
modern definition of orientability in terms of homology groups was introduced
by J.W. Alexander (1888–1971). A complete proof of the above theorem in a
more precise form can be found in W.H. Massey [13].
Definition 4.8.7. Let Sn be the unit n-sphere embedded in Rn+1 . Then the map
A : Sn → Sn defined by A(x) = −x, x ∈ Sn is called the antipodal map.
Note that the antipodal map is induced by a linear transformation

T : Rn+1 → Rn+1 whose determinant is (−1)n+1 . The following result is, there-
fore, quite interesting.
Theorem 4.8.8. The degree of the antipodal map A : Sn → Sn is (−1)n+1 , n ≥ 1.

n+1
Proof. We know that Sn = {(x1 , . . . , xn+1 ) ∈ Rn+1 | 1 x2i = 1}. A map
ri : Sn → Sn defined by ri (x1 , . . . , xi , . . . , xn+1 ) = (x1 , . . . , −xi , . . . , xn+1 ) is
called a reflection map. We will show that the map r : Sn → Sn defined by
r(x1 , . . . , xn+1 ) = (x1 , . . . , −xn+1 ) has degree −1. This will mean that each
reflection ri has degree −1. To see this, note that if h is the homeomorphism of
Sn which interchanges the coordinates xi and xn+1 for a fixed i, then h−1 rh = ri
and so deg ri = deg h−1 · deg r · deg h = deg r, proving that any reflection has
degree (−1). Now, since the antipodal map A = r1 ◦ r2 ◦ . . . rn+1 , it follows
that deg A = (−1)n+1 . Next, we are going to determine a triangulation S(K)
of Sn and a simplicial map g : S(K) → S(K) which will induce the map r. To
do this, let K be a triangulation of Sn−1 embedded in Sn in the standard way.
Let w0 and w1 be two distinct vertices of the cones w0 ∗ K and w1 ∗ K. Put
S(K) = (w0 ∗ K) ∪ (w1 ∗ K). Let g be the simplicial map from S(K) to itself
which interchanges the vertices w0 and w1 and keeps all other vertices fixed. Let
h : |K| → Sn−1 be a triangulating homeomorphism. We define k : |S(K)| → Sn
by the following formula: If y = (1 − t)x + tw0 for some x ∈ |K|, then define
k(y) = ( 1 − t2 h(x), t),
and if y = (1 − t)x + tw1 , then put
k(y) = ( 1 − t2 h(x), −t).
Then k is a homeomorphism making the diagram
|S(K)| k - Sn
g r
? ?
|S(K)| - Sn
k
commutative. Hence it suffices to show that deg g = −1. Let z be an n-cycle of

S(K). Then z is a chain of the form
z = [w0 , cm ] + [w1 , dm ],
where cm and dm are chains in K and m = n − 1. Assume n > 1. Since z is a

cycle, we have
0 = ∂(z) = cm − [w0 , ∂cm ] + dm − [w1 , ∂dm ].
Restricting this chain to K, we get
0 = cm + dm .
Therefore,
z = [w0 , cm ] − [w1 , cm ].
Since g simply exchanges w0 , w1 , we find that
g∗ (z) = [w1 , cm ] − [w0 , cm ] = −z.
Hence, deg g = −1. The case n = 1 is proved similarly.
Tangent Vector Fields on Spheres

Now, we present an interesting application of the above theorem on the
existence of tangent vector fields on Euclidean spheres. We have
Definition 4.8.9. A continuous mapping f : Sn → Sn is said to be a tangent

vector field on Sn if for each vector x ∈ Sn , the two vectors x and f (x) are
perpendicular to each other.
If one thinks of the point x ∈ Sn as a vector in Rn+1 , then a vector field

f on Sn is simply a continuous map f : Sn → Rn+1 such that f (x) is another
vector in Rn+1 . If f (x) is transported in a parallel direction so that it starts
from the point x of Sn , then x and f (x) are perpendicular really means that
f (x) is tangent to Sn at the point x. For the case n = 1, it is illustrated in Fig.
4.10
f(x)
x
(0,0)
Fig. 4.10: Tangent vector field
Now, we have the following question: Given any sphere Sn , does there
always exist a nonzero tangent vector field on Sn ? The following result gives a
complete answer to this question.
Theorem 4.8.10. (Vector fields on spheres). The n-sphere Sn admits a nonzero

tangent vector field if and only if n is odd.
Proof. Suppose n = 2m + 1 is odd. Then Sn is given by the set

2m+2
{(x1 , x2 , . . . , x2m+1 , x2m+2 ) ∈ R2m+2 | x2i = 1}.
1
We can now define a map f : Sn → Sn by the formula
f (x1 , x2 , . . . , x2m+1 , x2m+2 ) = (x2 , −x1 , . . . , x2m+2 , −x2m+1 ).
Then, evidently, f is a continuous map and has the property that for each
x ∈ Sn the scalar product x · f (x) = 0. Thus, Sn has a nonzero tangent vector
field.
Conversely, suppose n is even and let g : Sn → Sn be a nonzero tangent
vector field. Let us define a homotopy F : Sn × I → Sn by
F (x, t) = x cos(πt) + g(x) sin(πt),
where x ∈ Sn and t ∈ I. Using the fact that x · g(x) = 0, it is easily verified

that F (x, t) = 1, i.e., F is a well-defined homotopy. Moreover, F (x, 0) = x
and F (x, 1) = −x, for all x ∈ Sn . This says that the identity map ISn is
homotopic to the antipodal map A : Sn → Sn . Since homotopic maps have the
same degree, it follows that 1 = (−1)n+1 , which is a contradiction because n

is even. Hence no tangent vector field on Sn can exist if n is even.
The above theorem has an amusing description for the case n = 2. Suppose
one takes a ball in which there is a nonzero vector emanating from each point of
the ball. Now, think of each vector as a hair on the ball. The word “nonzero”
means there are hairs everywhere. Finding a tangent vector field means one
can “comb the hairs” on the ball in such a way that each hair is tangent
to the 2-sphere, which is the boundary of the ball, and their directions vary
continuously. The above theorem asserts that such a “hairstyle” is impossible,
unless there is a discontinuity.
The Fundamental Theorem of Algebra

We now use the notion of “degree” to prove the fundamental theorem of
algebra which is a well-known basic result of complex analysis. We have
Theorem 4.8.11. (Fundamental Theorem of Algebra). Every polynomial of pos-

itive degree with complex coefficients has a zero in C.
Proof. It is clearly enough to prove the theorem for a polynomial of the type
p(z) = z n + a1 z n−1 + . . . + an , where n ≥ 1, and a1 , a2 , . . . , an are all complex
numbers, i.e., p(z) is a nonconstant monic polynomial. Suppose the assertion
is not true. Then z → p(z) is a mapping from the complex plane C to C − {0}.
The restriction of this mapping to various circles |z| = r, r ≥ 0 are loops in
p(z)
C − {0} and, therefore, the restriction of map z → p(z) to circles |z| = r are
p(z)/p(r)
loops in the unit circle S1 ⊆ C. This means the map fr (z) = p(z)/p(r) defined
on the circle |z| = r is a loop in S based at 1 ∈ S . Note that f0 is clearly the
1 1
constant loop and r → fr defines a homotopy from f0 to fr for any r ≥ 0. This

says that, for all r ≥ 0, the loop fr is null-homotopic, i.e., deg fr = 0.
Next, note that if we take |z| = r > max{|a1 | + |a2 | + . . . + |an |, 1}, then
we observe that
|z|n = rrn−1 ≥ (|a1 | + . . . + |an |)|z|n−1 ≥ |a1 z n−1 + . . . + an |,
which means for 0 ≤ t ≤ 1, the map
z → z n + t(a1 z n−1 + . . . + an )
is never zero on such a circle |z| = r. Since, with pt (z) = z n +t(a1 z n−1 +. . .+an ),
pt (z)/pt (r)
F (z, t) = ,
pt (z)/pt (r)
is a homotopy between the loops z → z n / z n , i.e., s → e2πins and z →
p(z)/p(r)
p(z)/p(r) , i.e., s → fr (s), we find that fr is homotopic to the loop s → e
2πins
in S1 . Since the degree of this last loop is n, which is positive, we have a

contradiction to the fact proved earlier that deg fr = 0 for all r ≥ 0.
Exercises
1. Prove that the union of two nonintersecting circles in the torus T cannot
be a retract of T. (Hint. Use set topology.)
2. Prove that the central circle in the Möbius band M is a retract of M.
3. Show that a 2-sphere S2 cannot be homeomorphic to the torus T or pro-
jective plane P2 .
4. Prove that an orientable surface cannot be homeomorphic to a nonori-
entable surface.
5. Prove that an orientable surface of genus m cannot be homeomorphic to
an orientable surface of genus n, m
= n.
6. Prove that a nonorintable surface of genus m cannot be homeomorphic to
a nonorintable surface of genus n, m
= n.
7. Let A be a nonsingular n × n matrix. Then A defines a linear homeo-
morphism from Rn → Rn , which can be extended to a map f : Sn → Sn .
Prove that deg(f) = sign Det(A) (This generalizes Theorem 4.8.4).
8. Show that the long line (Example 1.7.7) has the fixed-point property.
(This is difficult, see [18] p. 56.)
4.9 Invariance of Homology Groups
In this section, which was postponed earlier, we give detailed proofs of

two results. First, we show that the barycentric subdivision of a complex K
does not change its simplicial homology. Secondly, we prove that any continu-
ous map f : |K| → |L| between two compact polyhedra induces a well-defined
homomorphism f∗ : Hq (K) → Hq (L), ∀ q ≥ 0 from the simplicial homology
groups of K to those of L. More importantly, we go on to show that these
induced homomorphisms in homology satisfy the desired functorial properties.
The last result is then utilized to prove that if X is a compact polyhedron
and K, K are two triangulations of X, then Hq (K) ∼ = Hq (K ), ∀ q ≥ 0. This
establishes the fact that the simplicial homology of a polyhedra |K| does not
depend on the triangulation of |K|, it rather depends on the topology of |K|.
As a matter of fact we prove much more: the simplicial homology groups of
|K| are homotopy invariants of |K| – a stronger result than saying that these
are topological invariants. In sections 4.9.4 and 4.9.5 we prove the Lefschetz
fixed-point theorem and the Borsuk-Ulam theorem respectively.
4.9.1 Subdivision Chain Map
In order to show that simplicial homology groups of a complex K do not change

when we pass on from K to its barycentric subdivision, we define two chain
4.9. Invariance of Homology Groups 175
maps. The first chain map μq : Cq (K) → Cq (K (1) ), called the subdivision chain
map, will be defined using the definition of barycentric subdivision K (1) of K.
It will not arise from any simplicial map from K to K (1) which is normally the
case. The other chain map θq : Cq (K (1) ) → Cq (K), in the opposite direction,
will be defined using a simplicial map θ : K (1) → K called standard simplicial
map. Then we show that these two chain maps induce homomorphisms in the
homology which are inverse of each other, and thus we find that homology
groups of K and K (1) are isomorphic. We have
Theorem 4.9.1. Let K (1) denote the first barycentric subdivision of a simplicial
complex K. Then Hq (K) ∼ = Hq (K (1) ) for all q ≥ 0.
Proof. Recall that the vertices of K (1) are just the barycentres σ˙q of all sim-
plexes σ q of K, q ≥ 0, and σ q = σ˙0 , σ˙1 , . . . , σ˙q is a q-simplex of K (1) if there
is a permutation t of {0, 1, . . . , q} so that σt(0) is a face of σt(1) , . . . , σt(q−1) is a
face of σt(q) . It would be helpful to have the case of a 2-simplex σ 2 = v0 , v1 , v2
of K in mind (Fig. 4.11).
v0 v˙r0
r
@ @
@ @ b(v0 , v1 , v2 )
@ b(v0 , v1 )r @b(v0 , v2 )
@ P r
PP +
@
@ P

rP @
P
@ PP @
P
q @r r r PP @r
v1 v2 v˙1 b(v1 , v2 ) v˙2
Fig. 4.11
We have denoted, just for convenience, the barycentre of a vertex vi by v˙i , and
shown the barycentre of vi , vj by b(vi , vj ), etc., in the figure.
A vertex vi of K is a 0-simplex σ 0 of K. The barycentre σ˙0 of σ 0 is

evidently σ 0 itself. Hence all the 0-simplexes of K are also the 0-simplexes of
K (1) . For any q-simplex σ q of K, the elementary 1-chain 1 · σ q will be denoted
simply by σ q . Finally, if σ q = v0 , . . . , vq and v is an additional vertex such
that {v, v0 , . . . , vq } is a geometrically independent set, then the (q + 1)-simplex

v, v0 , . . . , vq will be denoted by v·σ q . More
generally, for a q-chain c = ni σiq ,
q
we will denote by v · c the (q + 1)-chain ni (v · σi ). One can see that for any
chain c in Cq (K), ∂(v·c) = c−v.∂(c). Now, we define the subdivision chain map
μq : Cq (K) → Cq (K (1) ) for q ≥ 0 by induction on q. For a vertex vi of K, we
define μ0 (vi ) = v˙i , and then extend it linearly to C0 (K). Let σ 1 = v0 , v1 be
a positively-oriented 1-simplex of K. We define μ1 (σ 1 ) = σ̇ 1 , v˙1 > − < σ̇ 1 , v˙0
and extend it linearly to C1 (K). Note that for each 1-simplex σ 1 of K,
∂μ1 (σ 1 ) = ∂(σ̇ 1 , v˙1 ) − ∂(σ̇ 1 , v˙0 )

= v˙1 − σ̇ 1 − v˙0 + σ˙1
= μ0 (v1 ) − μ0 (v0 )
= μ0 ∂(σ 1 ).
This means ∂μ1 = μ0 ∂ and so μ0 , μ1 both satisfy the condition of a chain map.
Now we assume that μq has been defined for all q < p, (p > 1), satisfying the
condition ∂μq = μq−1 ∂ for all q-chains of Cq (K). Now, let σ p be a p-simplex
of K and define μ(σ p ) = σ̇ p μp−1 ∂(σ p ), where the right hand side, as explained
before, means the sum of all p-simplexes of K (1) which are faces of the cone
over the complex μp−1 ∂(σ p ). Then we find that
∂μp (σ p ) = ∂(σ̇ p μp−1 ∂(σ p ))

= μp−1 (∂(σ p )) − σ̇ p ∂μp−1 ∂(σ p )
= μp−1 ∂(σ p ),
since, by induction hypothesis, ∂μp−1 ∂ = μp−2 ∂∂ = 0. Hence μp , satisfying

the condition ∂μp = μp−1 ∂, is also defined. Thus, by induction, μq : Cq (K) →
Cq (K (1) ) is defined for all q ≥ 0 and is a chain map. It is helpful to actually
verify the case of a 2-simplex σ 2 also to see what is happening.
Next, we define a chain map θ : C(K (1) ) → C(K) which, in fact, arises
from a simplicial map θ : K (1) → K. We are using the same symbol θ for the
induced chain map also just for convenience and this should cause no confusion.
Let σ̇ q be any vertex of K (1) . We choose a vertex v0 of σ q and define θ(σ̇ q ) = v0 .
This induces a simplicial map θ : K (1) → K. To see this, note that if σ 1 =
v0 , v1 is a 1-simplex,
r r r r r
v0 σ˙1 v1 v0 σ1 v1
then, under the map θ, σ̇ 1 goes to a vertex, say, v0 , of σ 1 , and v˙0 goes to v0 , v˙1
goes to v1 . In other words, the simplex σ̇ 1 , v1 goes to the 1-simplex v0 , v1 ,
but the 1-simplex v0 , σ̇ 1 goes to 0-simplex v0 , i.e., v0 , σ̇ 1 collapses to a
vertex v0 . Next, if σ 2 = v0 , v1 , v2 is a 2-simplex, and we put θ(σ̇ 2 ) = v0 , then
assuming that b(v1 , v2 ) goes to v2 we find that the 2-simplex v1 , b(v1 , v2 ), σ̇ 2
goes to v1 , v2 , v0 whereas the 2-simplex v2 , b(v1 , v2 ), σ̇ 2 goes to a 1-simplex
v2 , v0 , i.e., it collapses to a lower-dimensional simplex. Now, we can check that
rest of 2-simplices in σ 2 also collapse to either a 1-simplex or to a 0-simplex of
σ 2 . Now, it is clear what would be θ(1σ q ), where 1σ q is a q-simplex of K (1) .
Thus, we have defined a simplicial map θ : K (1) → K. It is, of course, evident
that θ is not uniquely defined because there are choices involved. But once
we fix our choice for each q-simplex of K (1) , q ≥ 0, then θ is a well defined
simplicial map. Hence, θ induces a chain map θ : Cq (K (1) ) → Cq (K), q ≥ 0.
In fact, the simplicial map θ is a simplicial approximation of the identity map

|K | → |K|.
We observe that the composite chain map θ ◦ μ : C(K) → C(K) is the
identity map since for any q-simplex σ q of K, out of several simplexes occurring
in μ(σ q ), there is only one q-simplex in K (1) which is mapped back to σ,
and the rest of the q-simplexes are collapsed to lower-dimensional simplexes,
contributing nothing to θμ(σ q ), i.e., θμ(σ q ) = σ q , proving our statement.
Finally, we show that the chain map μθ : C(K (1) ) → C(K (1) ) is chain
homotopic to the identity map on C(K (1) ). Once again we will define a chain
homotopy Dq : Cq (K (1) ) → Cq+1 (K (1) ), q ≥ 0, by induction on q. Given a
vertex σ̇ of K (1) , we define D0 : C0 (K (1) ) → C1 (K (1) ) such that ∂D0 (σ̇) = σ̇ −
μθ(σ̇). To see this, note that θ(σ̇) is a vertex v0 of σ. Hence, μθ(σ̇) = v˙0 = v0 .
Thus, v0 , σ̇ is a 1-simplex in K (1) and we put D0 (σ̇) = v0 , σ̇ which clearly
has the desired property. Then we extend D0 to all the zero chains in C0 (K (1) )
linearly. Having defined D0 we assume that Dq is defined for all q < p (0 < p)
satisfying the condition that
∂Dq (σ ) + Dq−1 ∂(σ ) = σ − μθ(σ )
for all q-simplexes σ of K (1) . Now, note that 1σ p − μθ(1σ p ) − Dp−1 ∂(1σ p ) is
a chain on the cone μ(σ k ) = σ̇ k · μ(∂σ k ) for some σ k ∈ K which is well known
to be acyclic. Hence any cycle of this cone will be a boundary. We compute
∂[1σ p − μθ(1σ p ) − Dp−1 ∂(1σ p )]

= ∂(1σ p ) − ∂μθ(1σ p ) − ∂Dp−1 ∂(1σ p )
= ∂(1σ p ) − ∂μθ(1σ p ) − [∂(1σ p ) − μθ∂(1σ p ) − Dp−2 ∂∂(1σ p )],
by inductive hypothesis. Since ∂∂ = 0, and since μθ is a chain map, right hand

side of the above equation is zero, i.e., the quantity 1σ p −μθ(1σ p )−Dp−1 ∂(1σ p )
is a cycle. However, this is a cycle on the cone μ(σ k ) = σ̇ k μ∂(σ k ) and, therefore,
the same must be a boundary in that cone, i.e., we can find a chain c on μ(σ k )
so that ∂(c) is the above cycle. Then we define Dp (σ p ) = c, which yields us
∂Dp (1σ p ) = 1σ p − μθ(1σ p ) − Dp−1 ∂(1σ p ).
This completes the inductive definition of the chain homotopy Dq , q ≥ 0,

satisfying the desired property. It is, therefore, clear that the homomorphisms
μ∗ : Hq (K) → Hq (K (1) ) and θ∗ : Hq (K (1) ) → Hq (K) induced by the chain
maps μ and θ are inverse of each other. Hence, Hq (K) ∼ = Hq (K (1) ) ∀ q ≥ 0.
Note that if K = K (0) , K (1) , K (2) , . . . , K (m) are the successive barycentric
subdivisions of K and μi : C(K (i) ) → C(K (i+1) ) are the subdivision chain
maps, then the composite μm−1 ◦ . . . ◦ μ1 ◦ μ0 is again a chain map which
also is denoted by the same symbol μ : C(K) → C(K (m) ) and will be called
subdivision chain map. Each of these chain maps μi has a chain homotopy
inverse θi : C(K (i+1) ) → C(K (i) ) induced by a simplicial map and so their
composite θ = θ0 ◦ θ1 ◦ . . . ◦ θm−1 : C(K (m) ) → C(K) will be a chain homotopy
inverse of μ. The composite simplicial map θ : K (m) → K will be called the
standard simplicial map. The same notation for the simplicial map θ and the
induced chain map θ should cause no confusion. Combining these results, the
following is now immediate:
Corollary 4.9.2. Let K (m) denote the mth order barycentric subdivision of the
complex K. Then Hq (K) ∼ = Hq (K (m) ), ∀ q ≥ 0.
Contiguity Classes
Let K1 , K2 be two simplicial complexes. Then the two simplicial maps

φ1 , φ2 : K1 → K2 are said to be contiguous (or close to each other) if for
each simplex s ∈ K1 , φ1 (s) ∪ φ2 (s) is a simplex of K2 . This relation in the
set of all simplicial maps from K1 to K2 is reflexive and symmetric, but not
necessarily transitive. The smallest equivalence relation generated by this
relation is contiguity relation and the corresponding equivalence classes are
called contiguity classes.
From now onwards, st(v) will stand for ost(v) for convenience. We recall
that a simplicial map φ : K1 → K2 is a simplicial approximation of f : |K1 | →
|K2 | if and only if for every vertex v of K1 , f (st v) ⊂ st φ(v).
Proposition 4.9.3. Any two simplicial approximations from K1 → K2 of the

same continuous map f : |K1 | → |K2 | are contiguous.
Proof. Let φ, φ : K1 → K2 be simplicial approximations n to f : |K1 | → |K2 |,
and let s = v0 , . . . , vn be a simplex of K1 . Then 0 st vi
= ∅ and so

∅
= f ( st vi ) ⊂ f (st vi ) ⊂ (st φ(vi ) ∩ st φ (vi )),
which shows that φ(s) ∪ φ (s) is a simplex of K2 .
Proposition 4.9.4. Let f, f : |K| → |L| be homotopic maps. Then there exists
an integer n such that f, f have simplicial approximations φ, φ : K (n) → L
and there are finite number of simplicial maps φ1 , φ2 , . . . , φk such that φ = φ1
is contiguous to φ2 , φ2 is contiguous to φ3 , . . ., φk−1 is contiguous to φk = φ .
Proof. Let F : |K| × I → |L| be a homotopy from f to f . Since |K| is compact,
there exists a sequence 0 = t0 < t1 < . . . < tk = 1 of points of I such that for
α ∈ |K|, and i = 1, 2, . . . , k there is a vertex v ∈ L such that F (α, ti−1 ) and
F (α, ti ) both belong to st v. Let fi : |K| → |L| be defined by fi (α) = F (α, ti ).
Then f = f0 , f = fk and for i = 1, 2, . . . , k, the set
Ui = {fi−1 (st v) ∩ fi−1

−1
(st v) | v ∈ L}
is an open cover of |K|. Let N be chosen sufficiently large so that K (N ) is finer

than U1 , U2 , . . . , Uk . For i = 1, 2, . . . k let φi be a vertex map from K (N ) → L
such that
fi (st v) ∪ fi−1 (st v) ⊂ st φi (v)
for each vertex v of K (N ) . Such a vertex map exists since SdN (K) is finer than
Ui . Then φi : K (N ) → L is a simplicial approximation to both fi and fi−1 . Since
φi , φi+1 are simplicial approximations to fi , these are contiguous by previous
proposition, and this completes the proof.
Proposition 4.9.5. Let φ, φ : K → L be contiguous simplicial maps. Then the

induced chain maps φ# , φ# : C(K) → C(L) in the simplicial chain complexes
are chain homotopic.
Proof. We define maps Dq : Cq (K) → Cq+1 (L), q ≥ 0, by the formula

Dq (v0 , v1 , . . . , vq ) = (−1)i φ (v0 ), . . . , φ (vi ), φ(vi ), . . . , φ(vq ),
where v0 , v1 , . . . , vq is a q-simplex of K, and then extend it to Cq (K) by

linearity. Then it is easily seen that on σ q = v0 , . . . , vq we have
∂q+1 Dq (σ q ) + Dq−1 ∂q (σ q )
= φ(v0 ), . . . , φ(vq ) − φ (v0 ), . . . , φ (vq )
= (φ# − φ# )(σ q ).
Hence, by linearity again, {Dq } is a chain homotopy from φ# to φ# .
4.9.2 Homomorphism Induced by a Continuous Map

We have already seen how a simplicial map |K| → |L| induces a homomor-
phism in simplicial homology groups Hq (K) → Hq (L) ∀ q ≥ 0. Now, let us
take a continuous map f : |K| → |L| which is not necessarily a simplicial
map. Let us choose a simplicial approximation s : |K (m) | → |L| of f , and
let μ : C(K) → C(K (m) ) be the subdivision chain map. Then, we define
the induced homomorphism f∗ : Hq (K) → Hq (L) as the composite map
s∗ μ∗ : Hq (K) → Hq (K (m) ) → Hq (L).
In the above definition of induced homomorphism f∗ : Hq (K) → Hq (L),

there is a choice involved, viz., the choice of the simplicial approximation
s : |K (m) | → |L|. Therefore, in order that f∗ is well defined, we must show
that f∗ is independent of this choice. For this, let t : |K (n) | → |L| be an-
other simplicial approximation to f . We may assume that n ≥ m. Let
μ : C(K (m) ) → C(K (n) ) be the subdivision chain map and let θ : |K (n) | →
|K (m) | be the standard simplicial map. Now, observe that sθ : |K (n) | → |L|
is also a simplicial approximation to f . Therefore, by Proposition 4.9.3 both
sθ and t are contiguous and so by Proposition 4.9.5, the induced homomor-
phisms (sθ)∗ , t∗ : Hq (K (n) ) → Hq (L) must be equal. This evidently means
(sθ)∗ ◦ (μ μ)∗ = t∗ ◦ (μ μ)∗ , i.e., s∗ μ∗ = t∗ μ∗ μ∗ since θ∗ μ∗ = Id . The last
equation says that f∗ : Hq (K) → Hq (L) is indeed well-defined.
4.9.3 Homotopy Invariance
We can now prove the desired functorial properties of the induced homomor-
phisms and much more. We have
Theorem 4.9.6. (i) If f : |K| → |K| is the identity map, then the induced
homomorphism f∗ : Hq (K) → Hq (K) is also identity for q ≥ 0.
(ii) If f : |K| → |L|, g : |L| → |M | are two continuous maps, then for all q ≥ 0,
(g ◦ f )∗ = g∗ ◦ f∗ : Hq (K) → Hq (M ).
Proof. (i) This follows at once from the definition of induced homomorphism.
(ii) Choose a simplicial approximation t : |L(n) | → |M | to g : |L| → |M |. Then
find a simplicial approximation s : |K (m) | → |L(n) | to the map f : |K| → |L| =
|L(n) |. Let μ1 : C(K) → C(K (m) ), μ2 : C(L) → C(L(n) ) be the subdivision
chain maps and let θ : |L(n) | → |L| be the standard simplicial map. Consider
the following diagram
s∗
Hq (K (m) ) Hq (L(n) )
μ1∗ μ2∗ θ∗ t∗
Hq (K) Hq (L) Hq (M ).
f∗ g∗
Note that θs is a simplicial approximation to f : |K| → |L| and ts is a simplicial
approximation to gf : |K| → |M |. Hence,
g ∗ ◦ f∗ = t∗ μ2∗ θ∗ s∗ μ1∗
= t∗ s∗ μ1∗ = (ts)∗ μ1∗
= (gf )∗ .
Theorem 4.9.7. If f, g : |K| → |L| are homotopic maps, then f∗ =

g∗ : Hq (K) → Hq (L)∀ q ≥ 0.
Proof. Since f, g are homotopic we know, by Proposition 4.9.4, that there is a
subdivision K (m) of K and simplicial maps s1 , s2 , . . . , sn : |K (m) | → |L| such
that s1 is a simplicial approximation of f , sn is a simplicial approximation of
g, and si is contiguous to si+1 , i = 1, 2, . . . , n − 1. Let μ : C(K) → C(K (m) )
be the subdivision chain map. Then, it follows from the definition of induced
homomorphism, that f∗ = s1∗ μ∗ = s2∗ μ∗ = . . . = sn∗ μ∗ = g∗ .
The above theorem has remarkable consequences. The first and foremost
says that if two polyhedra |K| and |L| are homotopically equivalent (in partic-
ular, if K and L are triangulations of the same polyhedron X), then they have
isomorphic simplicial homology, i.e., Hq (K) ∼
= Hq (L), ∀ q ≥ 0. To see this, let
f : |K| → |L| and g : |L| → |K| be two maps such that g ◦f is homotopic to I|K|
and f ◦ g is homotopic to the identity map I|L| . By the functorial properties of
induced homomorphisms, we find that the composites
f∗ g∗
Hq (K) −→ Hq (L) −→ Hq (K),
g∗ f∗
Hq (L) −→ Hq (K) −→ Hq (L)
are identity maps on Hq (K) and Hq (L), respectively, and, therefore, Hq (K) ∼ =
Hq (L) for all q ≥ 0. In particular, this says that simplicial homology groups of
a compact polyhedron X are not only topological invariants, but also they are
homotopy invariants. For instance, it follows from this result that the simplicial
homology groups of an n-disc Dn are the same as those of a point-space.
Exercises
1. Let f : X → Y be a constant map between polyhedra X and Y . Show that

the induced homomorphism f∗ : Hq (X) → Hq (Y ) is zero, for all q > 0.
What happens when q = 0?
2. Prove that the Euler characteristic of a compact contractible polyhedron

X is 1.
3. Let K be a simplicial complex, and w ∗ K denote the cone complex over

K. Define a simplicial map φ : w ∗ K → {w} by mapping every simplex of
w ∗ K to the 0-simplex w. If i : {w} → w ∗ K denotes the inclusion map,
then prove that induced chain map φ# : C(w ∗ K) → C({w}) is a chain
equivalence. Hence deduce that Hq (w ∗ K) = 0, ∀ q > 0. (This result is
stronger than Theorem 4.5.1.)
4. For any simplicial complex K, let C(K) denote the simplicial chain com-
plex of K. Then prove that C(K, G) = C(K) ⊗ G is a chain complex for
any coefficient group G and Hq (X, G) = Hq (C(K) ⊗ G), q ≥ 0. Using
this, show that for any short exact sequence 0 → G → G → G → 0 of
coefficient modules, there is a long exact sequence
. . . → Hq (K; G ) → Hq (K; G) → Hq (K; G ) → Hq−1 (K; G ) → . . .
of simplicial homology groups of K with coefficients in G , G and G

respectively.
4.9.4 Lefschetz Fixed-Point Theorem

We are now going to present one of the most important fixed-point theorems
of algebraic topology, called Lefschetz Fixed-Point Theorem, named after the
American mathematician Solomon Lefschetz (1884–1972), who discovered it
in 1926. Brouwer’s Fixed Point Theorem, proved earlier, will be a simple
corollary of this theorem. Let X be a compact polyhedron and f : X → X
be a continuous map. Associated with such a map, Lefschetz defined a
number λ(f ) and then proved the result that if λ(f )
= 0, then f must
have a fixed point. It must be pointed out that the number λ(f ), called
the Lefschetz number of f , will be an integer whose non-vanishing will im-
ply that f will have a fixed point – it will not give us the number of fixed points.
To define the Lefschetz number λ(f ), let h : |K| → X be a triangulation

of X, where dim X = n, and let us consider the simplicial homology Hq (K; Q)
of K with coefficients in the additive group of rational numbers. By result
of section 4.9.2 proved earlier, h−1 f h : |K| → |K| induces a homomorphism
(h−1 f h)∗ : Hq (K; Q) → Hq (K; Q), for all q ≥ 0. Since the homology groups
Hq (K; Q) are vector spaces, (h−1 f h)∗ is a linear transformation for each q. We
consider the alternate sum of traces of all these linear transformations for each
q ≥ 0 and define,
n
λ(f ) = (−1)q Tr (h−1 f h)∗ .
q=0
Here, (h−1 f h)∗ really means (h−1 f h)∗,q : Hq (K; Q) → Hq (K; Q) and n
denotes the dimension of the complex K, which means the sum is finite as
Hq (K) = 0 ∀ q > n.
In order that λ(f ) is well defined, we must show that λ(f ) is inde-
pendent of any triangulation h : |K| → X. For this, let k : |L| → X be
yet another triangulation of X. It suffices to show that the two linear maps
(h−1 f h)∗ : Hq (K; Q) → Hq (K; Q) and (k −1 f k)∗ : Hq (L; Q) → Hq (L; Q) have
identical traces for all q ≥ 0. Note that
k −1 f k = (k −1 h)(h−1 f h)(h−1 k)
and hence, at the homology level,
(k −1 f k)∗ = (k −1 h)∗ (h−1 f h)∗ (k −1 h)−1
∗ ,
where (k −1 h)∗ : Hq (K; Q) → Hq (L; Q) is an isomorphism. However, the above

equation shows that the matrices of linear maps (h−1 f h)∗ and (k −1 f k)∗ are
indeed similar and, therefore, they have the same traces. Now, we have
Theorem 4.9.8. (Lefschetz Fixed-Point Theorem). If X is a compact polyhedron

and f : X → X is a continuous map such that λ(f )
= 0, then f has a fixed
point.
In order to prove the above theorem, we require a result known as Hopf ’s

Trace Formula and let us explain this formula first. Let
C: 0 → Cn → Cn−1 → . . . → C1 → C0 → 0
be a finite chain complex of finite-dimensional vector spaces over Q and let

φq : Cq → Cq , 0 ≤ q ≤n be a chain map. Then, here also, we can consider the
n
well defined number i=0 (−1)q Tr (φq ). Since this chain map induces homo-
morphism φq∗ : Hq (C) → Hq (C) for all q, 0 ≤ q ≤ n, we also have the number

(−1)q Tr (φq∗ ). The Hopf’s formula is given by:
Proposition 4.9.9. (H. Hopf ). Let C be a chain complex of finite length n in

which each Cq is a finite dimensional vector space over Q. Then for any chain
map φ : C → C,

n
n
(−1)q Tr (φq ) = (−1)q Tr (φq∗ ).
q=0 q=0
Proof. We choose a suitable basis for the vector space Cq . We have sub-
space Bq ⊂ Zq ⊂ Cq of q-boundaries, q-cycles and q-chains. Start with a
basis ∂cq+1 1 , . . . , ∂cq+1
rq+1 of Bq , then extend it to a basis of Zq by adding cy-
cles z1 , . . . , zβq q , and finally extend this to a basis of Cq by adding q-chains
q
cq1 , . . . , cqrq . Thus, for each q, we work with basis of Cq :
B = {∂cq+1
1
q q q
rq+1 , z1 , . . . , zβq , c1 , . . . , crq }.
, . . . , ∂cq+1 q
The diagonal elements of the matrix of φq with respect to above basis are
obtained by taking any element w of the above basis and expressing φq (w) as a
linear combination of elements of B, and finding the coefficients of w in φq (w).
Let us denote this coefficient by λ(w). Then, clearly, the trace of φq is

λ(∂cq+1
j )+ λ(zjq ) + λ(cqj ).
Since φ is a chain map, we find that λ(∂cq+1

j ) = λ(cq+1
j ). Therefore,

n
n
βq
q
(−1) Tr (φq ) = (−1) q
λ(zjq ),
q=0 q=o j=1
since the other terms cancel away in pairs. Because {z1q }, . . . , {zβq q } is a basis
of Hq (C), we have
βq
λ(zjq ) = trace of φq∗ .
j=1
This establishes the desired formula.

Proof of the Lefschetz Fixed-Point Theorem. Let h : |K| → X be a

triangulation of X and f : X → X the given continuous map. Then
g = h−1 ◦ f ◦ h : |K| → |K| has a fixed point if and only if f has a fixed point.
We prove the theorem by contradiction. Let us assume that g : |K| → |K| has
no fixed points. Let d be the metric on |K| regarded as a subspace of some Eu-
clidean space. Since |K| is a compact space and the continuous function defined
by x → d(x, g(x)) is never zero, it has a positive minimum, say, δ > 0. By going
to sufficiently high order barycentric subdivision, we may as well assume that
mesh K is less than δ/3. Let s : |K (m) | → |K| be a simplicial approximation
of g : |K| → |K| and μ : C(K; Q) → C(K (m) ; Q) be the subdivision chain map.
Since, for all q ≥ 0, gq∗ : Hq (K) → Hq (K) is the composite gq∗ = sq∗ μq∗ , it
suffices, by Hopf’s Trace Formula, to show that trace sq ◦ μq : Cq (K) → Cq (K)
is zero, because that will imply that λ(f ) = 0, a contradiction.
Let σ be an oriented q-simplex of K and let τ be an oriented q-simplex
of K (m) which occurs in the chain μq (σ). Thus, τ is contained in σ. Now, if
x ∈ τ , d(s(x), g(x)) < δ/3 because s is a simplicial approximation of g. Hence,
by d(x, g(x)) ≥ δ, we find that
d(x, s(x)) ≥ d(x, g(x)) − d(g(x), s(x)) > δ − δ/3 = 2δ/3.
Next, let y ∈ σ, then d(x, y) < δ/3, and by what we proved just now,
d(y, s(x)) ≥ d(s(x), x) − d(x, y) > 2δ/3 − δ/3 = δ/3.
This means s(x) and y do not lie in the same simplex of K and, therefore,
s(τ ) ∩ σ = φ for each face τ ⊂ σ. Hence, in the chain sq ◦ μq (σ), the coefficient
of σ is zero. In other words, we have proved that trace sq ◦ μq = 0. This
completes the proof.
It may be pointed out that the Hopf’s Trace Formula stated earlier for a
simplicial map f : K → K is a generalization of the Euler-Poincaré Theorem
4.5.8 when f = IK . We also observe that for a simplicial map g : |K| → |K|,
which has the property that for each simplex σ ∈ K, σ ∩ g(σ) = ∅, clearly,
the Hopf Trace Formula will say that λ(g) = 0. Thus, the whole point of the
Lefschetz Fixed-Point Theorem was to prove that under the given conditions,
if f has no fixed points, then f can be approximated by a simplicial map g
which has the property stated above.
Corollary 4.9.10. (Brouwer’s Fixed-Point Theorem) Let X be a compact poly-

hedron which is contractible. Then any continuous map f : X → X will have a
fixed point.
Proof. We just note that because X is contractible, H0 (X; Q) ∼ = Q and
Hq (X) = 0, ∀ q > 0. Moreover, the induced homomorphism f∗ : H0 (X; Q) →
H0 (X; Q) is the identity map. Therefore, λ(f ) = 1
= 0, and f has a fixed point
by Lefschetz Fixed-Point Theorem.
Corollary 4.9.11. Let f : Sn → Sn be a continuous map. Then λ(f ) = 1 +

(−1)n deg f . In particular, if deg f
= ±1, then it has a fixed point.
Proof. We know that H0 (Sn ; Q) ∼ =Q∼ = Hn (Sn ; Q) and Hq (Sn ; Q) = 0 for all
other values of q. Since f∗ : H0 (S ; Q) → H0 (Sn ; Q) is the identity map and
n
f∗ : Hn (Sn ; Q) → Hn (Sn ; Q) has trace equal to the deg f , the result is now
immediate.
Exercises
1. Let f : X → X be a continuous map where X is a compact polyhedron.

Recall that the definition of Lefschetz number λ(f ) is given using homol-
ogy with the rational coefficients Q. Prove that λ(f ) must always be an
integer. Show by an example that this is not true if X is not a compact
polyhedron.
2. Let X be a compact polyhedron and f : X → X be a constant map,
i.e., f (x) = x0 , ∀x ∈ X and for some x0 ∈ X. Determine the induced
homomorphism f∗ : Hq (X) → Hq (X) in homology for all q ≥ 0. Hence or
otherwise, prove that a null homotopic map f : X → X must have a fixed
point.
3. Let the additive group R of real numbers act on a compact polyhedron X
with χ(X)
= 0 under the action θ : R × X → X (see Appendix). For each
t ∈ R, let us define a map θt : X → X by putting θt (x) = θ(t, x). Prove
that
(a) Any two maps θt , θt are homotopic to each other.
(b) Each θt has a fixed point in X.
(c) For each n ≥ 1, any fixed point of θ1/2n+1 is a fixed-point of θ1/2n .
(d) There is a point x0 ∈ X which is a fixed point of θt for all diadic
rational t = (m/2n ).
(e) There is a point x0 ∈ X which is fixed under the group R.
4. Let G be a path-connected topological group. Show that any left transla-
tion Lg : G → G defined by Lg (x) = g ·x is homotopic to the identity map.
Hence or otherwise deduce that the Euler characteristic of a compact con-
nected triangulable topological group G must be zero. (Hint: Let ωg be a
path in G joining the identity element with g and define F : G × I → G
by F (x, t) = ωx−1 (t) · x)
5. Prove that an even-dimensional sphere S2n cannot be a topological group.
6. Prove that the only compact surface which can be a topological group is
the torus T . Why not the Klein bottle?
4.9.5 The Borsuk-Ulam Theorem
We now come to another very interesting result about the maps from spheres
Sm to spheres Sn , where m and n are arbitrary nonnegative integers. Here, we
consider the n-sphere Sn as a subspace of Rn+1 defined by
!
n+1
"
Sn = (x1 , . . . , xn+1 ) | x2i = 1 .
i=1
Two points x, y ∈ Sn are said to be antipodal points if y = −x, i.e., these

are opposite to each other with respect to the origin. The map A : Sn → Sn
defined by A(x) = −x is called the antipodal map, and we have already come
across this map earlier (see Definition 4.8.7). A continuous map f : Sm → Sn ,
m, n arbitrary positive integers, is said to “preserve” antipodal points if for all
x ∈ Sm , f (A(x)) = A(f (x)). The question that we would like to ask is: What
are the values of m and n so that there is a continuous map f : Sm → Sn
preserving the antipodal points? The answer is trivial if m ≤ n. To see this,
recall that for m ≤ n, Sm is embedded into Sn in such a manner that antipodal
points of Sm are also the antipodal points of Sn . For example, we can place
S1 = {(x, y) ∈ R2 | x2 + y 2 = 1} into S2 = {(x, y, z) ∈ R3 | x2 + y 2 + z 2 = 1}
simply by taking the z-coordinate to be zero. More generally, take Sm as a
subset of Sn where the last n − m coordinates are zero. Then the inclusion
map i : Sm → Sn defined by i(x1 , . . . , xm+1 ) = (x1 , . . . , xm+1 , 0, . . . , 0) clearly
preserves the antipodal points. In particular, note that the identity map of Sn
preserves the antipodal points. It is the case when m > n, however, which is
not at all clear! In fact, the Borsuk-Ulam Theorem asserts that if m > n, then
there can be no continuous map f : Sm → Sn preserving the antipodal points.
This is what we will prove in this section. It may be mentioned that this result
was conjectured by S. Ulam and first proved by K. Borsuk in 1933.
Once we have proved the above result, several interesting consequences,

some of which are in fact equivalent to the Borsuk-Ulam Theorem, follow very
quickly. For example,
Theorem 4.9.12. Any continuous map f : Sn → Rn must map a pair of antipodal

points to the same point.
Proof. If this theorem is not true, then we find that for each x ∈ Sn , f (−x)
=
f (x), i.e., f (x) − f (−x)
= 0. Hence we can define a map g : Sn → Sn−1 by the
formula
f (x) − f (−x)
g(x) = .
f (x) − f (−x)
Then, clearly, g is continuous and for all x ∈ Sn , g(−x) = −g(x), i.e., g preserves
antipodal points, a contradiction to Borsuk-Ulam Theorem.
Corollary 4.9.13. The n-sphere Sn , n ≥ 1, cannot be embedded into Rn . In

particular, the 2-sphere S2 cannot be put into the plane R2 .
This follows at once because any embedding i : Sn → Rn is continuous,
and so by the preceding theorem, must map at least one pair of points to a
single point of Rn , i.e., it cannot be an injective map.
Here is an interesting illustration of the above form of Borsuk-Ulam The-

orem: Suppose our Earth is a 2-sphere and T , P are functions on the earth
defining temperature and barometric pressure at any point of time and place
on the Earth. Then these are continuous functions and, hence, so is the map
h : S2 → R2 defined by h(x) = (T (x), P (x)). The Borsuk-Ulam Theorem says
that at any point of time there exists a pair of opposite points on the earth
where the temperature and barometric pressure are identical!
Corollary 4.9.14. (Invariance of Dimension). If m

= n, then the Euclidean
space Rm cannot be homeomorphic to Rn .
Proof. One can assume that m > n. If possible, let f : Rm → Rn be a
homeomorphism. Then f will map the n-sphere Sn ⊂ Rm into Rn as a
continuous injective map, i.e., f restricted to Sn is an embedding into Rn ,
contradicting the earlier corollary.
There is yet another interesting consequence of the Borsuk-Ulam Theorem

which must be mentioned: Let S1 be the unit circle in the plane R2 . If we cover
S1 by two sets A1 = {eiθ | 0 ≤ θ < π} and A2 = {eiθ | π ≤ θ < 2π}, then
note that none of these sets contains a pair of antipodal points. None of these
sets, however, is closed. If we take one of them to be closed, then observe that
A1 or A2 will definitely contain a pair of antipodal points. A similar situation
is true for S2 also, viz., if we cover S2 by three closed sets, then at least one
of them will definitely contain a pair of antipodal points. The next result says
that these observations can be proved for Sn for any n. We have
Theorem 4.9.15. (Lusternik-Schnirelmann Theorem). If {A1 , A2 , . . . , An+1 } is

a covering of Sn , n ≥ 1, by closed sets, then at least one of these sets must
contain a pair of antipodal points.
Proof. Let us define a map f : Sn → Rn by the formula f (x) =
(d(x, A1 ), d(x, A2 ), . . . , d(x, An )), where d(x, Ai ) denotes the distance of point
x from the closed set Ai , i = 1, 2, . . . , n. Since each of these distance func-
tions is continuous in x, the map f itself is continuous. Hence, by the Borsuk-
Ulam Theorem, there must be an x ∈ Sn such that (d(x, A1 ), . . . , d(x, An )) =
(d(−x, A1 ), . . . , d(−x, An )), which says that d(x, Ai ) = d(−x, Ai ) for every i.
Now, if d(x, Ai ) > 0 for each i, then d(−x, Ai ) > 0 for each i also, and so both
x, −x must be in the remaining set An+1 because {A1 , A2 , . . . , An+1 } covers
Sn . On the other hand, if d(x, Ai ) = 0 for some i then, since Ai is closed, both
x and −x must be in Ai .
Remark 4.9.16. In the proof of the above theorem we used the fact that out of
n + 1 sets only n are closed.
Note that the antipodal map A : Sn → Sn is a homeomorphism of Sn of
order 2. This says that the group G = {1, −1} of order two acts on Sn for each n
via the antipodal maps. Now, if f : Sm → Sn is a continuous map preserving the
antipodal points then, in terms of group action, this means the map f preserves
the above G-action. Such a map induces a continuous map f¯: Sm /G → Sn /G
on orbit spaces, i.e., f¯ is a continuous map from the real projective space RPm
to RPn . Thus, the Borsuk-Ulam Theorem says that if m > n, then there cannot
be a continuous map from RPm to RPn induced by any equivariant map from
Sm to Sn .
Now, we return to the proof of the Borsuk-Ulam Theorem, viz., there

cannot be a continuous map f : Sm → Sn preserving the antipodal points
if m > n. (The case when m = n + 1 is Borsuk-Ulam Theorem). Suppose,
on the contrary, that m > n and there is a continuous map f : Sm → Sn
preserving antipodal points. Then, to prove the result we will, on the basis
of this assumption, arrive at a contradiction. In fact, we will prove (Theorem
4.9.18) that any g : Sn → Sn mapping antipodal points to antipodal points
must have odd degree. On the other hand, since f : Sm → Sn , m > n, preserves
antipodal points, it restricts to a map g : Sn → Sn preserving antipodal points.
We will show that this restriction map has degree zero, giving a contradiction.
The latter assertion follows rather very easily. For, the restriction g : Sn → Sn
of f : Sm → Sn clearly has an extension to Dn+1 ⊂ Sn+1 ⊆ Sm and, therefore,
g is homotopic to a constant map. Since homotopic maps have same degrees
and the constant map has degree zero, we find that deg g = 0. Thus, it is the
first part which requires a good amount of work. We now proceed to prove
that result next.
Note that for a given map g : Sn → Sn , its degree deg(g) = d is an integer

having the property that if h : |K| → Sn is any triangulation of Sn , then the
induced homomorphism (h−1 gh)∗ : Hn (K) → Hn (K) is simply multiplication
by d. We have already seen that degree d is independent of the triangulation.
Therefore, we can choose any triangulation of Sn which is convenient to us
and can work with that only. We choose a special triangulation π : |Σn | → Sn
described below: For each integer i = 1, 2, . . . , n + 1, let vi be the point of
Rn+1 having all its coordinates zero except the i-th coordinate which is 1. Let
J = {1, 2, . . . , n + 1} and let a : J → {1, −1} be any map. For each a, we define
an n-simplex σan = a(1)v1 , a(2)v2 , . . . , a(n+1)vn+1 . Then, note that the union
of the geometric carriers of all these simplexes is the following subset of Rn+1 :
!
n+1
"
Qn = (x1 , . . . , xn+1 ) ∈ Rn+1 | |xi | = 1 .
i=1
The case n = 1 is illustrated in the following figure:

x2
6
v2
@
@
@
@
@ - x1
@ v1
@
@
@
@
Note that the set of all n-simplexes described above, along with all of their
faces, gives a triangulation Σn of Sn – the homeomorphism h : |Σn | → Sn is
just the “radial projection” from the origin of Rn+1 : any ray starting from
origin meets Qn = |Σn | and Sn in exactly one point, say, x and y, respectively,
and we define h(x) = y. This triangulation is specially suited for dealing with
antipodal map A because both Sn and Qn are invariant under A. The simplicial
complex Σn can also be regarded as invariant with respect to the antipodal map
because A maps an n-simplex σan of Σn to σ−a n
. Thus, the antipodal map A is
n
also a simplicial map from Σ to itself and, therefore, induces a chain map in
the simplicial chain complex C(Σn ) of Σn . First, we have
Theorem 4.9.17. Let f : Sn → Sn be a continuous map preserving the antipodal

points. Then the Lefschetz number λ(f ) of f is an even integer.
Proof. We use the above triangulation π : |Σn | → Sn of Sn and recall that
λ(f ) = λ(π −1 f π). Let Σ(k) be a barycentric subdivision of Σn , which is star
related to π −1 f π, and let τ : |Σ(k) | → |Σn | be a simplicial approximation of
π −1 f π. We assert that τ can be chosen so that τ preserves the antipodal points
of Σ(k) and Σn . We divide the vertices of Σ(k) into two parts B and the image
A(B) of B under A so that no points of B are antipodal. For each vertex v in
B, choose a vertex τ (v) in Σ so that π −1 f π(st (v) ⊆ st (τ (v)). Then define τ (v)
for elements of A(B) by putting τ A(x) = Aτ (x). Then it is straightforward to
see, using the fact that π −1 f π maps antipodal points to antipodal points, that
the simplicial map τ so defined is a simplicial approximation of π −1 f π, and τ
preserves antipodal points.
Next, we consider the subdivision chain map μ : C(Σn ) → C(Σ(k) ) and
the orientation of Σ(k) . From the definition of μ, each q-simplex σjq of Σn ,
i q
μ(σjq ) = j αi , where the sum is over all q-simplexes of Σ(k) contained in
σj and j = +1 or −1. We say that a simplex αiq of Σ(k) contained in σjq is
q r
oriented like σjq if ij = +1 and is unlike σjq if ij = −1. We may orient Σ(k) so
that whenever αiq is contained in σjq , then αiq is oriented like σjq if and only if
A(αjq ) is oriented like A(σjq ). The purpose of this particular orientation is to
ensure that μ commutes with A# . This is easily verified since for each simplex
σjq of Σn , we have

A# μ(σjq ) = A# ( ij αiq ) = ij A(αiq ) = μA# (αjq ).

Next, we consider the Lefschetz number λ(π −1 f π) = (−1)i tr (τ# μ). Let
us consider the tr (τ# μ) in a particular dimension, say, q. Let M = (mij ) be
the matrix representing τ# μ with respect to the basis of Cq (Σn ) consisting of
elementary integral q-chains σjq ∈ B and Aσjq ∈ A(B). Then the integer mjj is
the coefficient of σjq in the chain τ# μ(σjq ) and the sum of all these numbers is
the trace (τ# μ) in dimension q. Now, we will prove that such a coefficient in
this sum appears twice. Let us denote the coefficient of σjq in τ# μ(σjq ) by the
symbol g(σjq ), i.e., mjj = g(σjq ). It is enough to show that g(A(σjq )) = g(σjq ).
We have,

τ# μ(σjq ) = g(σjq )σjq + i=j mij σiq ,
so that
τ# μ(A(σjq )) = τ# μ(A∗ (σjq )) = A∗ τ# μ(σjq )

= A∗ (g(σjq )σjq + mij σjq )
i=j

= g(σjq )(A(σjq )) + mij A(σiq ).
i=j
Since the coefficient of A(σjq ) in τ# μ(A(σjq )) is g(σjq ), we conclude that

g(A(σjq )) = g(σjq ). This shows that trace of τ# μ is even in each dimension, and
so λ(f ) = λ(π −1 f π) is even.
Now, we have
Theorem 4.9.18. Let f : Sn → Sn , n ≥ 1, be a continuous map preserving

antipodal points. Then deg f is odd.
Proof. We know that H0 (Sn ) ∼=Z∼ = Hn (Sn ), and Hq (Sn ) = 0 for other values
of q. We also know, because S is path-connected, that f∗ : H0 (Sn ) → H0 (Sn )
n
is the identity map. Hence, by the preceding theorem, λ(f ) = 1 + (−1)n deg f
is even, which implies that deg f must be odd.
Theorem 4.9.19. (Borsuk-Ulam). If m > n ≥ 0, then there cannot be a contin-

uous map f : Sm → Sn preserving the antipodal points.
Proof. Suppose, on the contrary, there is a continuous map f : Sm → Sn which
preserves the antipodal points. Let i : Sn → Sm denote the standard inclusion
map, i.e., Sn is obtained from Sm by putting last m − n coordinates equal to
zero. Then the composite i ◦ f : Sm → Sm also preserves the antipodal points.
Hence, by Theorem 4.9.18 , deg(i ◦ f ) must be odd. On the other hand, the
composite (i ◦ f )∗ = i∗ ◦ f∗ : Hm (Sm ) → Hm (Sn ) → Hm (Sm ) factors through

Hm (Sn ) = 0, which means (i ◦ f )∗ is the zero map, i.e., deg(i ◦ f ) is zero. This
is a contradiction, proving the theorem.
Corollary 4.9.20. Let f : Sn → Rn be a continuous map preserving antipodal

points. Then there exists a point x ∈ Sn such that f (x) = 0.
Proof. If f (x)
= 0, ∀ x ∈ Sn , then we can define a map g : Sn → Sn−1 by
the formula g(x) = ff (x)
(x)
. Then g is a continuous map preserving antipodal
points, a contradiction to Borsuk-Ulam Theorem.
Exercises
1. Prove that the Borsuk-Ulam theorem is equivalent to the following state-

ment: if f : Sn → Rn is a continuous map, then there exists a point x0 ∈ Sn
such that f (x0 ) = f (−x0 ).
2. Show that Sn is not homeomorphic to a proper subspace of itself.
3. Show that there is no continuous map f : Sm → Sn , for all m > n ≥ 0,
preserving antipodal points if and only if there is no continuous map
g : Sn+1 → Sn , preserving antipodal points for all n ≥ 0.
4. Suppose a map f : Sn → Sn satisfies the condition that f (−x) =
f (x), ∀ x ∈ Sn . Then prove that deg f is even. If n is even, show that
deg f = 0.
5. If f : Sn → Sn has odd degree, then prove that f (−x) = −f (x) for some
x ∈ Sn . (Hint: Suppose the contrary. Consider the map h(x) = (f (x) +
f (−x))/ f (x) + f (−x) .)
6. Let A1 , A2 , A3 be three bounded convex subsets of E3 , and define a func-
tion f : S3 → E3 w.r.t. A1 , A2 , A3 as follows: For x ∈ S 3 take a hyperplane
P (x) perpendicular to x passing through (0,0,0,1/2). Let fi (x) be the vol-
ume of that part of Ai which lies on the same side of the hyperplane P (x)
as the point x, i = 1, 2, 3. Put
f (x) = (f1 (x), f2 (x), f3 (x))
Prove that f is continuous.

(Hence, by the Borsuk-Ulam Theorem there exists an x0 ∈ S3 such
that f (x0 ) = f (−x0 ), i.e., f1 (x0 ) = f1 (−x0 ), f2 (x0 ) = f2 (−x0 ),
f3 (x0 ) = f3 (−x0 ). This result is called the Ham-Sandwich Theorem be-
cause it proves the following surprising result: Given a three-layered Ham-
sandwich it is possible to cut it by a single stroke of knife so that each of
the three pieces is divided into two equal parts!)
4.10 Homology of General Simplicial Complexes
So far we have defined and studied the finite simplicial complexes K

and their homology groups Hq (K), q ≥ 0. We know that for such complexes,
the induced topological space |K| are compact polyhedra. Using this fact,
we defined and studied the homology groups of a compact polyhedron X.
The homology of a noncompact polyhedron, however, has not been defined
yet. We note that the Euclidean space Rn , n ≥ 1, is a polyhedron and a
triangulation of Rn is an infinite simplicial complex. This means we have
not defined homology of even such nice spaces as Rn . In this section, we
will extend the definition of homology to any general simplicial complex K,
which will enable us to define the simplicial homology of any polyhedron X,
compact or noncompact. In particular, we will define Hq (Rn ), q ≥ 0, for all
Euclidean spaces, and the homology of many other nice noncompact polyhedra.
Let K be a general simplicial complex as in Section 3.5. We give an ori-

entation to K by orienting each simplex of K arbitrarily and then define the
simplicial chain complex C∗ (K) of K as follows:
∂q+1 ∂q 1 ∂
C∗ (K) : . . . −→ Cq+1 (K) −→ Cq (K) −→ Cq−1 (K) . . . −→ C0 (K) −→ 0.
Here Cq (K) is the free abelian group whose basis is the set of all q-simplexes
of K, and ∂q : Cq (K) → Cq−1 (K) is defined as before viz.,
q
∂q (v0 , . . . , vq ) = 0 (−1)i v0 , . . . , v̂i , . . . , vq
where v0 , . . . , vq is a basis element of Cq (K). We know that ∂q ◦ ∂q+1 =
0, ∀ q ≥ 0. Since the simplicial complex K is arbitrary, the above chain complex
will be of infinite length. Each Cq (K) could be a finitely generated or infinitely
generated abelian group depending on whether the number q-simplexes is finite
or infinite. Then we define the homology groups Hq (K) of K by
ker ∂q
Hq (K) = , q ≥ 0.
Im ∂q+1
We observe that, in contrast with finite simplicial complexes, now Hq (K) may
be nonzero for infinitely many values of q and each of these groups could be
finitely generated or infinitely generated.
Next, let f : K → L be a simplicial map. Then we can define the induced

chain map f# : Cq (K) → Cq (L) exactly as in finite simplicial complexes. This
will induce a homomorphism f∗ : Hq (K) → Hq (L), ∀ q ≥ 0, in homology, and
they will satisfy the functorial properties, viz., (i) the identity map IK : K → K
will induce identity map (IK )∗ : Hq (K) → Hq (K)∀ q ≥ 0, and (ii) the induced
map will preserve the composite maps, i.e., if f : K → L and g : L → M are
simplicial maps, then (g ◦ f )∗ = g∗ ◦ f∗ : Hq (K) → Hq (M ) ∀ q ≥ 0.
4.10. Homology of General Simplicial Complexes 193
Homomorphism induced by a continuous map

Let X be a polyhedron, compact or noncompact. We now wish to define
the simplicial homology Hq (X), q ≥ 0 of X. Let us take a triangulation K of
X and define
Hq (X) = Hq (K) for all q ≥ 0.
In order that the above is well defined up to isomorphism, we need to prove
that if K is any other triangulation of X, then for all q ≥ 0, Hq (K) ≈ Hq (K ).
As in the case of finite simplicial complexes, this raises the question as
to what is the homomorphism in homology induced by a continuous map
f : |K| → |L|. Let us suppose that f : |K| → |L| is a continuous map which
need not arise from a simplicial map from K to L. How this map f will induce
a homomorphism f∗ : Hq (K) → Hq (L)∀ q ≥ 0, in their simplicial homology?
For this, we need a simplicial approximation theorem for the continuous map
f : |K| → |L| for general simplicial complexes K and L.
As in the case of finite complexes, for a simplicial approximation theorem,

we require the concept of barycentric subdivision of a general simplicial complex
K. If we try to proceed as in the case of finite simplicial complexes, we confront
a basic problem viz., for general K, |K| has the coherent topology and so it need
not be a metric space. More than that, even if |K| is a metric space, it need not
lie in an Euclidean space Rn and so we cannot assert that there is a simplex of
K having the largest diameter. In order words, it is not possible to define the
mesh of a simplicial complex K. Therefore, we cannot have subdivisions of K
having arbitrary small mesh. This problem has been resolved by the concept
of generalized barycentric subdivision of K. This is a technical generalization
of the usual barycentric subdivision where we consider barycentric subdivision
of higher dimension skeleton K (n) of K without disturbing the subdivision of
K (n−1) . We refer the reader to the book of J.R. Munkres [16] for all details of
the generalized barycentric subdivision. Using this subdivision of an arbitrary
simplicial complex K, one can prove the following (see [16]).
Theorem 4.10.1. (Generalized Simplicial Approximation Theorem). Let

f : |K| → |L| be a continuous map. Consider the open cover V = {st w :
w is a vertex of L} of |L|. Then there exists a generalized barycentric subdivi-
sion SdN (K) = K of K such that the cover U = {st v : v is a vertex of K }
by closed vertex stars is a refinement of {f −1 (v) : v ∈ V}. This means |K| is
star-related to |L| w.r.t. f .
Assuming the above simplicial approximation theorem for general simpli-

cial complexes, now we can define the homomorphism induced by a continuous
map.
Definition 4.10.2. Let f : |K| → |L| be a continuous map and let g : |K | → |L|
be a generalized simplicial approximation of f . Suppose λ : C∗ (K) → C∗ (K ) is
the subdivision operator. Then we define f∗ : Hq (K) → Hq (L) as the composite

f ∗ = g ∗ ◦ λ∗ .
It can now be proved, exactly as in the case of finite simplicial complexes,

that the induced homomorphism f∗ : Hq (K) → Hq (L) is really independent
of the simplicial approximation. Hence f∗ is well-defined. Then one can easily
prove the following:
Theorem 4.10.3. (i) Let I : |K| → |K| be the identity map. Then the induced
homomorphism I∗ : Hq (K) → Hq (K) is also the identity map ∀ q ≥ 0.
(ii) Let f : |K| → |L| and g : |L| → |M | be two continuous maps. Then, ∀q ≥ 0
(g ◦ f )∗ = g∗ ◦ f∗ : Hq (K) → Hq (M ).
Corollary 4.10.4. Let h : |K| → |L| be a homeomorphism. Then h∗ : Hq (K) →

Hq (L) is an isomorphism ∀ q ≥ 0.
Proof. Let k : |L| → |K| be the inverse of h. Then k ◦ h : |K| → |K| is the
identity on |K| and h ◦ k : |L| → |L| is the identity on |L|. Hence by (i) and (ii)
above, we have ∀ q ≥ 0,
(k ◦ h)∗ = k∗ ◦ h∗ = IHq (K)
(h ◦ k)∗ = h∗ ◦ k∗ = IHq (L) .

This means h∗ : Hq (K) → Hq (L) is an isomorphism ∀ q ≥ 0.
Simplicial Homology of a Polyhedron

Now we can define the simplicial homology of a polyhedron which has
been the objective of this section.
Definition 4.10.5. Let X be a general polyhedron. We define Hq (X) =

Hq (K)∀ q ≥ 0, where K is any triangulation of X.
We remark that the above definition of homology of X is unambigu-

ous (up to isomorphism) because if K is any other triangulation of X,
then we have homeomorphisms h : |K| → X, k : |K | → X, which means
k −1 ◦ h : |K| → |K | is a homeomorphism. Therefore, by the above corollary it
induces an isomorphism (k −1 ◦ h)∗ : Hq (K) → Hq (L) ∀ q ≥ 0.
Thus we have now defined the simplicial homology of any polyhedron X

(of course up to group isomorphism). Let us be clear what we mean by saying
that ‘Hq (X) is defined up to isomorphism’. This means if we consider the
class of all simplicial complexes K which triangulate the polyhedron X, then
the homology groups of all such complexes are isomorphic to each other and
we define Hq (X) to be simply this isomorphism class. Hence if we identify all

these isomorphic groups, then Hq (X) is uniquely defined, and this is what we
do in the sequel.
Next, let X, Y be two polyhedra and f : X → Y be a continuous map. To

define the induced homomorphism f∗ : Hq (X) → Hq (Y ), we take triangulations
K and L of X and Y respectively. Then we have a commutative diagram
f
X Y
h k
|K| |L|
g
Here the maps h and k are homeomorphism and the map g is being
defined as g = k −1 f h : |K| → |L|. Now we define f∗ to be simply
g∗ : Hq (K) → Hq (L) ∀q ≥ 0. The definition of f∗ depends not only on
the map f and topology of |K|, |L|, but it also depends on the simplicial
complexes K and L. If we identify Hq (X) with Hq (K) and Hq (Y ) with Hq (L)
where K and L are triangulations of X and Y , then we can as well identify f∗
with g∗ . This means f∗ is also uniquely defined up to the isomorphism.
We can also prove the following:
Proposition 4.10.6. (i) Let I : X → X be the identity map.Then ∀ q ≥ 0 the

induced homomorphism I∗ : Hq (X) → Hq (X) is also the identity map.
(ii) If f : X → Y and g : Y → Z are continuous map of polyhedra, then ∀ q ≥ 0
(g ◦ f )∗ = g∗ ◦ f∗ : Hq (X) → Hq (Z).
The above proposition tells us that the homology groups of a polyhedron

X are topological invariants.
Example 4.10.7. Let us compute the homology groups of the real line R. We
know that K = {n, n, n + 1 : n ∈ N } is a triangulation of R. Hence the
simplicial chain complex C∗ (R) is given by
∂1
C∗ (R) : 0 ⊕Zn, n + 1 ⊕Z · n 0
C1 (R) C0 (R)
Now note that there is not a single non-zero 1-cycle since ∂1 n, n + 1 = n +

1−n = 0 means n = n+1, a contradiction. Hence H1 (R) = 0 = H2 (R) = . . ..
To compute H0 (R), note that the 0-cycle 1 is not a boundary. Moreover if we

take any other 0-cycle, say for example 3, then
∂1 (1, 2 + 2, 3) = 3 − 1,
which means the cycle 3 is homologous to 1. Therefore, there is exactly one
nontrivial homology class {1}. Hence H0 (R) ≈ Z · {1}.
By a similar consideration using a triangulation of Rn , we can prove that

Z, q = 0,
H0 (Rn ) =
0, otherwise.
We can also prove the following.
Theorem 4.10.8. Suppose X and Y are two polyhedra. If f, g : X → Y are two

continuous maps which are homotopic, then for all q ≥ 0
f∗ = g∗ : Hq (X) → Hq (Y ).
The above theorem implies that homology groups Hq (X) of a poly-

hedron are not only topological invariants, but they are also homotopy
invariants, i.e., if X and Y are two polyhedra of the same homotopy type,
then Hq (X) ≈ Hq (Y ) ∀ q ≥ 0.
In order to prove the above theorem we require a number of results which

we summarize below. Since their proofs are technically involved, we refer to
([16]) for details. First, we have
Theorem 4.10.9. If X is a polyhedron, then the product space X × I is also a

polyhedron.
Proof of Theorem 4.10.8: In fact, one can prove the following sharper result: For
any polyhedron |K|, we can always find a triangulation M of |K| × I such that
for each simplex σ ∈ K, σ ×{0} and σ ×{1} are both simplexes of M . Moreover
the subset σ × I is an acyclic subcomplex of M . Now let F : |K| × I → |L|
be a homotopy from a continuous map f : |K| → |L| to the continuous map
g : |K| → |L|. Suppose f∗ : Hq (|K|) → Hq (|L|) and g∗ : Hq (|K|) → Hq (|L|) are
the induced homomorphisms in homology. We now have the following diagram
of continuous maps and polyhedra.
|K|
CC
CC i2
CC
CC
!
|M |
F / |L|
=
{{{
{
{{
{{ i1
|K|
where i1 (σ) = σ × {0} and i2 (σ) = σ × {1} are the canonical injections. Then
F ◦ i1 = f and F ◦ i2 = g. Using the ‘acyclic model’ theorem, one can prove
that the induced chain maps i1# , i2# : C∗ (|K|) → C∗ (|L|) are chain homotopic.
This means that the induced maps in homology
f∗ = F∗ ◦ i1∗ = F∗ ◦ i2∗ = g∗ .
Next, let (|K|, |L|) be a polyhderal pair, i.e., L is a subcomplex of K. Then

we can define the simplicial homology Hq (|K|, |L|) of the pair to be simply the
homology of the quotient chain complex of the following exact sequence of chain
complexes
C∗ (K)
0 → C∗ (L) → C∗ (K) → →0
C∗ (L)
# $
C∗ (K)
In other words, we define Hq (|K|, |L|) = Hq C∗ (L) ∀ q ≥ 0. We can also
define the homomorphism induced by a continuous map of pairs f : (|K|, |L|) →
(|K |, |L |) by considering the chain map f# induced by the following diagram
of chain complexes.
0 C∗ (|L|) C∗ (|K|) C∗ (|K|, |L|) 0

( f ||L| )# ( f ||K| )# f#
0 C∗ (|L |) C∗ (|K |) C∗ (|K |, |L |) 0
Then, clearly f# induces the homomorphism
f∗ : Hq (|K|, |L|) → Hq (|K |, |L |) ∀ q ≥ 0
The above result now can be used to define the simplicial homology theory
on the category of polyhedral pairs and their morphisms (see the chapter on
singular homology). One can now easily prove the following general homotopy
theorem for pairs:
Theorem 4.10.10. Let f, g : (X, A) → (Y, B) be two maps of polyhedral pairs

which are homotopic as maps of pairs. Then the induced homomorphism in
homology of pairs are equal, i.e., f∗ = g∗ : Hq (X, A) → Hq (Y, B) ∀ q ≥ 0.
Proof. Let F : (X × I, A × I) → (Y, B) be a homotopy from the map f to g.
Define h¯0 , h¯1 : (X, A) → (X × I, A × I) by h0 (x) = (x, 0), h1 (x) = (x, 1). If
h0 , h1 : X → X × I are defined by h¯0 , h¯1 , then we know that there is a natural
chain homotopy
D : (h0 )# ∼
= (h1 )# : S∗ (X) → S∗ (X × I).
This chain homotopy restricts to a chain homotopy
D : (h0 )# ∼
= (h1 )# : S∗ (A) → S∗ (A × I).
Hence we can pass on to the quotient chain homotopy

D̄ : (h¯0 )# ∼
= (h¯1 )# : S∗ (X, A) → S∗ (X × I, A × I).
Therefore, that the induced homomorphism in homology are equal, i.e.,
(h̄0 )∗ = (h̄1 )∗ : Hq (X, A) → Hq (X × I, A × I).
Hence f∗ = F∗ ◦ (h̄0 )∗ = F∗ ◦ (h̄1 )∗ = g∗ .
Exercises
1. Let (X, A) be a polyhedral pair. Show that there is a long exact sequence
in simplicial homology:
i
∗ j∗ ∂
. . . → Hq (A) → Hq (X) → Hq (X, A) →∗ Hq−1 (A) → . . . ,
where i : A → X, j : X → (X, A) are the inclusion maps of pairs and

∂∗ : Hq (X, A) → Hq−1 (A) is the connecting homomorphism.
2. Let L be a subcomplex of K. Prove that
(a) if there is a retraction r : |K| → |L|, then Hq (K) ≈ Hq (L) ⊕
Hq (K, L).
(b) if the identity map i : K → K is homotopic to a map f : |K| → |L|,
then
Hq (L) ≈ Hq (K) ⊕ Hq+1 (K, L).
(c) if the inclusion j : |L| → |K| is homotopic to some constant map,
then
Hq (K, L) ≈ H̃q (K) ⊕ H̃q−1 (L)
3. Give an example to show that there are polyhedral pairs (X, A) and
(Y, B) such that for all q, we have Hq (X) ≈ Hq (Y ), Hq (A) ≈ Hq (B)
but Hq (X, A)
≈ Hq (Y, B).
4. Suppose we have a sequence of finitely generated abelian groups
G0 , G1 , . . ., where G0 is free and nontrivial. Then show that there is a
simplicial complex K such that Hq (K) ≈ Gi ∀ i ≥ 0.
Chapter 5
Covering Projections
5.1 Introduction
Let exp : R → S1 denote the normalized exponential map defined by

exp(t) = e2πit , t ∈ R. Recall (Chapter 2, Theorem 2.6.5, 2.6.6) that this map
has the Path Lifting Property as well as the Homotopy Lifting Property.
We now wish to generalize these results to a wider class of continuous maps
p : X̃ → X, called covering projections. The theory of covering projections
is of great importance not only in topology, but also in other branches of
mathematics like complex analysis, differential geometry and Lie groups, etc.
The concept of fundamental group, discussed in Chapter 2, is intimately
related to covering projections. Among other things, we will see that the
problem of “lifting” a continuous map f : A → X to a continuous map
f˜: A → X̃, where p : X̃ → X is a covering projection, has a complete solution
in terms of the fundamental groups of the spaces involved and the induced
homomorphisms among them.
Let p : X̃ → X be a continuous map. Then an open set U of X is said to

be evenly covered by p if p−1 (U ) is a disjoint union of open subsets of X̃ each
of which is mapped homeomorphically onto U by the map p. In such a case,
we say that U is an admissible open set of X. It is possible that p−1 (U ) can
be expressed as a disjoint union of open sets satisfying the above condition in
more than one way. However, if U is a path-connected open set, then there is
only one way of expressing p−1 (U ) as a disjoint union of open sets of X̃ such
that each open set is homeomorphic to U under the map p. These open sets are
simply path-components of p−1 (U ) – these are path-components because each
one is an open as well as a closed subset of p−1 (U ), and is path-connected as
each one is homeomorphic to the path-connected space U . It is for this reason
that, in this chapter, all of our spaces are assumed to be path-connected and
locally path-connected unless stated otherwise.
200 Chapter 5. Covering Projections
We also mention a topological fact to be kept in mind (see Proposition

1.6.10). Every connected and locally path-connected space is path con-
nected. Thus, for locally path-connected spaces, connectedness is equivalent
to path-connectedness and so all of our spaces are assumed to be locally
path-connected and connected (rather than path-connected).
The following is our basic concept:
Definition 5.1.1. A continuous map p : X̃ → X is said to be a covering pro-

jection or a covering map if each point x ∈ X has a path connected open
neighbourhood U which is evenly covered by p, i.e., U is an admissible neigh-
bourhood of x ∈ X. The space X̃ is called a covering space of X and X itself
is called the base space of the covering projection p.
Evidently, if an open set U is admissible, then any path connected open

subset of U is also admissible.
Example 5.1.2. Let exp : R → S1 be the exponential map defined by

exp(t) = e2πit . Then exp is a covering projection. To see this, recall how the
above map wraps the real line R onto the unit circle S1 . See Fig. 5.1.
−2
` −1
` 0̀ 1̀ 2̀ - R
i`
'$
(−1, 0) ` ` (1, 0)
&% `
−i S
1
Fig. 5.1
Note that the complex number 1 = 1 + i · 0 and −1 = −1 + i · 0 lie on S1 .

Consider the open set U1 = S1 − {−1} and observe that
∞
% 1 1
(exp)−1 (U1 ) = (n − , n + )
−∞
2 2
is a disjoint union of open intervals, each of which is homeomorphic to U1

under the map exp. Similarly, one can see that the open set U2 = S1 − {1}
is also evenly covered by exp map. Thus, each point of S1 has an admissible
open neighbourhood in S1 . As a matter of fact, any proper connected arc of
S1 is easily seen to be evenly covered by the given exponential map. The same
argument shows that the map p : R → S1 defined by p(t) = eiαt , where α ∈ R
is a fixed nonzero number, is also a covering projection; the case α = 1 is the
usual exponential map and will be frequently used later on.
5.1. Introduction 201
Example 5.1.3. Any homeomorphism p : X̃ → X is trivially a covering pro-

jection. More generally, let Y be a discrete space and X be any space. If
p : X × Y → X denotes the projection map on the first factor then p is a
covering projection. In fact, here the space X itself is evenly covered because
%
p−1 (X) = X × Y = X × {y}
y∈Y
is a disjoint union and X is homeomorphic to X × {y} for each y ∈ Y under

the map p.
Example 5.1.4. For any nonzero integer n, let us define a map qn : S1 → S1 by

qn (z) = z n ,
where z ∈ S1 is the unit complex number and z n is the nth power of z. Then
qn is a covering projection. As an illustration, consider the case when n = 4. If
we divide the domain
'$ A` 2 '$
q4 -
A3 ` ` A1
&% ` S1 &% S1
A4
Fig. 5.2: q4 wraps first circle four times onto second
circle into four equal arcs (Fig. 5.2), then q4 maps each arc [Ai , Ai+1 ], i =
1, 2, 3, 4 and A5 = A1 onto the whole of range circle S1 , so that the open arc
(Ai , Ai+1 ) gets mapped to S1 −{A1 } homeomorphically under the map q4 . This
means that each point of the range circle S1 −{A1 } has an open neighbourhood
which is evenly covered by q4 . If we consider mid-points Bi of arcs [Ai , Ai+1 ],
B5 = B1 , then each open arc (Bi , Bi+1 ) gets mapped homeomorphically under
q4 onto the open set S1 − {A3 } of the range circle. Thus it follows that each
point of the range circle has an open neighbourhood which is evenly covered
by qn : S1 → S1 . We may also observe here that qn wraps the domain S1 onto
the range circle exactly n times either clockwise or anticlockwise depending on
whether n is negative or positive.
Example 5.1.5. If p : X̃ → X and q : Ỹ → Y are covering projections, then the

product map p × q : X̃ × Ỹ → X × Y defined by (p × q)(x, y) = (p(x), q(y)) is
also a covering projection. To see this, let (x0 , y0 ) ∈ X × Y . Choose an open
neighbourhood U of x0 and an open neighbourhood V of y0 which are evenly
covered by p and q, respectively. Then U × V is a neighbourhood of (x0 , y0 ) in
X × Y which is easily verified to be evenly covered by p × q.
In particular we get: the map (exp, exp) : R × R → S1 × S1 is a covering
n 1In fact, for every n ∈ N, the product map (exp) : R → T , where
n n n
projection.
T = 1 S is the n-dimensional torus, is a covering projection.
n
Example 5.1.6. Let X = [0, 4], Y = S1 and p = exp|X , i.e., p is the restriction of
the usual exponential map to the space X. Then p wraps X four times on Y , but
is not a covering projection. To see this, note that each point of S1 −{(1, 0)} = U
has an open neighbourhood, viz., U itself, which is evenly covered but the point
(1,0) does not have any small neighbourhood in S1 which is evenly covered by
p, since the inverse image of any small open neighbourhood N around (1,0)
in S1 will have five pieces out of which three will be homeomorphic to N and
the remaining two, being semi-closed intervals, will not be homeomorphic to N
under any map and so, in particular, under p.
Example 5.1.7. Note that if X̃ → X is any covering map, then it follows from
definitions that there must be an open set Ũ of X̃ and an open set U of X
which are homeomorphic. Now, let I = [0, 1] denote the closed unit interval and
p : I ×I → I be the projection on one of the factors. Then p cannot be a covering
projection because no open set of I × I can be homeomorphic to any open set
of I; this last fact is clear since if that was the case then an open rectangle in
R2 must be homeomorphic to an open interval of R, a contradiction. The same
argument proves that no map from I × I to I can be a covering projection.
Example 5.1.8. Let C denote the complex plane. We claim that the exponential
∞ n
map exp : C → C − {0} defined by exp(z) = ez = 0 zn! is a covering map.
To show this, we begin by observing that the map k : C → R × R defined by
k(x + iy) = (x, y) is a homeomorphism. Next, we see that any element w of
C − {0} can be uniquely written (in terms of polar coordinates) as w = reiθ ,
0 < r and 0 ≤ θ < 2π. Since the map exp : R → R+ = {x ∈ R | x > 0} is a
homeomorphism, we can write w = ex · eiy , where x is uniquely determined by
r and y = θ. Hence it is clear that the map h : C − {0} → R+ × S1 defined by
h(w) = (ex , eiy ) is a homeomorphism. Since p : R → R+ defined by p(x) = ex is
a homeomorphism and q : R → S1 defined by q(y) = eiy is a covering projection,
it follows from Example 5.1.5 that the product map p × q : R × R → R+ × S1 is
a covering projection. Now, one can easily see that ez = h−1 ◦ (p × q) ◦ k and
hence is a covering projection from C to C − {0}. We must point out here that
in this example, we have used the obvious fact that if X̃ → X is a covering
map and X̃ is replaced by a homeomorphic space Ỹ or X is replaced by a
homeomorphic space Y , then Ỹ is a covering space of Y through a suitable
map.
Example 5.1.9. Let us observe that for any integer n

= 0, the map f : R+ → R+
defined by f (x) = xn is a homeomorphism and so is a covering projection.
Also, the map q : S1 → S1 defined by q(eiy ) = einy is a covering projection.
It follows, therefore, that p × q : R+ × S1 → R+ × S1 is a covering projection.
Then composing with the homeomorphisms h and h−1 defined in the previous
example, we conclude that the map g : C − {0} → C − {0} defined by g(z) = z n
is a covering projection for any n
= 0. Also, one can easily see that for any
n
= 0, 1, the map g is not a covering map from C to C.
5.2. Properties of Covering Projections 203
Example 5.1.10 (Figure Eight). Consider the following subset X of the product
space S1 × S1 :
X = {(z, w) ∈ S1 × S1 | z = (1, 0) or w = (1, 0)}.
Here, S1 should be thought of as the standard unit circle in the complex plane.
The space X can be thought of as two circles which are touching each other at
one point - this is the space usually known as the figure of eight. Now consider
the following subspace X̃ of the plane R2 (see Fig. 5.3);
X̃ = {(x, y) ∈ R2 | x or y is an integer}.
Define p : X̃ → X by
p(x, y) = (e2πix , e2πiy ).
6
'$
'$
` - p -
(0,0)
&%&%
X
Fig. 5.3
For each integer x, p is the usual covering map; similarly, for each integer y,
the map p is the same exponential covering map. Hence it is not difficult to see
that p is indeed a covering projection.
A mapping p : E → B from a space E to a space B is said to be a local
homeomorphism if each point e ∈ E has an open neighbourhood U such that
p(U ) is open in B, and the map p|U : U → p(U ) is a homeomorphism. If
p : X̃ → X is a covering projection, then clearly it is a local homeomorphism:
For any x̃ ∈ X̃, let U be an admissible open neighbourhood of p(x̃) and Ũ be
the component of p−1 (U ) which contains x̃ and is homeomorphic to U . Then
Ũ is the required open neighbourhood of x̃. The converse, however, need not
be true. Let us consider the following
Example 5.1.11. Suppose p : R → S1 is the exponential map p(x) = e2πix .

Consider the restriction q = p|(0,8) of p to the open interval (0,8). Then q
is a local homeomorphism but cannot be a covering map because the point
(1, 0) ∈ S1 does not have an admissible neighbourhood in S1 - there will be
nine components of the inverse image of a small open neighbourhood of (0, 1)
in which seven will be homeomorphic to the small neighbourhood whereas the
other two will not be homeomorphic to that neighbourhood under the map p.
5.2 Properties of Covering Projections

In this section, we will prove some of the basic properties of a covering
map p : X̃ → X. We have already seen that a covering map is a local homeo-
morphism. It follows from the definition itself that a covering map is an onto
map. Recall that a space X is locally path connected if and only if each path
component of an open subset of X is open.
Proposition 5.2.1. A covering projection p : X̃ → X is always an open map.

Proof. Let V be any open subset of X̃. We must show that p(V ) is open in
X. For that, let x ∈ p(V ) and choose x̃ ∈ V such that p(x̃) = x. Let U be
an admissible neighbourhood of x and W be the path component of p−1 (U )
which contains x̃. Since X̃ is locally path connected, W is open in p−1 (U )
and hence in X̃. Since p maps W homeomorphically onto U , p must map the
open set W ∩ V onto an open subset p(W ∩ V ) of U and hence of X. Clearly,
x ∈ p(V ∩ W ) ⊆ p(V ), i.e., p(V ) is a neighbourhood of each of its points and
hence is open in X.
Note that the above result combines two facts together, viz., a covering
projection is a local homeomorphism and a local homeomorphism is an open
mapping.
Proposition 5.2.2. The cardinality of all the fibres of a covering projection

p : X̃ → X are same.
Proof. By the definition of a covering projection, each point p ∈ X has an
open neighbourhood
U such that p−1 (U ) is homeomorphic to the disjoint
union {Uα : α ∈ I} such that Uα is homeomorphic to U under the map
p for each α. The space α∈I Uα is clearly homeomorphic to the product
U × I, where I has the discrete topology. Hence fibres over U have the same
cardinality as that of the indexing set I. In other words, the cardinality
of fibres over the base space X is locally constant. Since X is connected,
the cardinality of the fibres must be constant over X, which proves the result.
Because a covering map p : X̃ → X is always an open onto continuous

map, we conclude that the topology of the base space X is the quotient
topology induced from X̃ by the map p.
We have seen (Example 5.1.11) that the restriction of a covering map

p : X̃ → X to an open set U of X̃ need not be a covering map even though it
may be onto X. However, the following is a useful result:
Proposition 5.2.3. Let p : X̃ → X be a covering map and A be a connected and

locally path connected subset of X. If Ã is a path component of p−1 (A), then
the restriction map p|Ã : Ã → A is a covering projection.
Proof. For each a ∈ A, choose an admissible open neighbourhood U of a (in
X) and let {Ui } be the path components of p−1 (U ) such that each Ui is home-
omorphic to U under p. Then, clearly, U ∩ A is evenly covered by Ui ∩ p−1 (A).
Since A is locally path connected, we can find a path-connected open (in A)
neighbourhood V of a contained in U ∩ A. Then, clearly, V is evenly covered by
p. If any path component Vi of p−1 (V ) intersects Ã, then it must be contained

in Ã as Ã is path connected. Therefore, p|Ã : Ã → A is a covering projection.
Path Lifting Property
The following is an important topological problem (see Chapter 1, last

section) which one usually comes across in various situations, particularly in
topology.
Lifting Problem. Let p : X̃ → X be any continuous, onto map, not necessarily a

covering projection. Suppose a continuous map f : A → X is given. Can we find
a continuous map f˜: A → X̃ such that the following triangle is commutative?
*
X̃
f˜

p
?
A -X
f
In other words, the problem is to find a map f˜: A → X̃ such that p ◦ f˜ = f .

This question may or may not have an affirmative answer; furthermore, if there
is one f˜ satisfying the above condition, then there could be more than one such
map; any such f˜ is called a lift of f . We will be able to answer the above question
“quite satisfactorily” when p : X̃ → X is a covering projection. However, first
we will consider the case when the map f is a path in X. But, before that, we
can easily decide the uniqueness of lifts. We have
Theorem 5.2.4. Let p : X̃ → X be a covering map and A be any space (not

necessarily connected or locally path connected ). If f, g : A → X̃ are any two
continuous maps such that p ◦ f = p ◦ g, then the set of points of A, where f
and g agree, is a clopen subset of A.
Proof. Let S = {a ∈ A | f (a) = g(a)}. To prove that S is open we will show
that S contains a neighbourhood of each of its points. Let x0 ∈ S. Choose
an admissible neighbourhood U of pf (x0 ) = pg(x0 ), and let V be the path
component of p−1 (U ) containing f (x0 ) = g(x0 ). Then V is open in X̃ and
hence f −1 (V ) ∩ g −1 (V ) is an open set of A containing x0 . We will show that
f −1 (V ) ∩ g −1 (V ) is a subset of S. Let x ∈ f −1 (V ) ∩ g −1 (V ). This means f (x)
and g(x) are in V . Since p restricted to V is a homeomorphism, and we are
given that pf (x) = pg(x), it follows that f (x) = g(x) which proves the required
assertion.
Next, we prove that S is a closed subset of A. Suppose b is a limit
point of S which is not in S. Then f (b)
= g(b). Since pf (b) = pg(b), we
can find an admissible neighbourhood W of pf (b) such that the points f (b)
and g(b) must be in distinct path components V1 and V2 of p−1 (W ). Now,
b ∈ f −1 (V1 ) ∩ g −1 (V2 ) which is an open set containing b and since b is a limit
point of S, there should be point a ∈ S such that a ∈ f −1 (V1 ) ∩ g −1 (V2 ). But

this a contradiction, since V1 ∩ V2 = φ and a ∈ f −1 (V1 ) ∩ g −1 (V2 ) implies
f (a) = g(a) ∈ V1 ∩ V2 . Hence all limit points of S must be in S and so S is
closed.
The following gives the uniqueness result for the lifts of a map f : A → X
when A is a connected space.
Corollary 5.2.5. Let p : X̃ → X be a covering projection. Suppose g1 , g2 : A → X̃

are lifts of the same map f : A → X such that g1 and g2 agree at one point of
A. Then g1 = g2 .
Proof. By the given condition, pg1 = f = pg2 . Hence, by the above theorem,
the set S = {a ∈ A | g1 (a) = g2 (a)} is a clopen subset of A. The fact that g1 , g2
agree at some point of A implies that S is nonempty. Since A is connected, we
conclude that S = A, i.e., g1 = g2 .
The following theorem shows that any path in the base space X of a
covering projection p : X̃ → X can always be lifted to a path in X̃, and is
unique if the initial point is already prescribed.
Theorem 5.2.6. (The Path Lifting Property). Let p : X̃ → X be a covering

projection and α : I → X be a path beginning at some point x0 ∈ X. Then
given any point x̃0 over x0 (i.e., p(x̃0 ) = x0 ), there exists a unique path α̃ in
X̃ beginning at x̃0 which lifts α, i.e., p ◦ α̃ = α.
Proof. Suppose α(I) is contained in some admissible open neighbourhood U of
X. Choose a path component V of p−1 (U ). Then α̃ = ( p|V )−1 ◦ α : I → X̃ is
a path in X̃ which has the property that
p ◦ α̃ = p ◦ ( p|V )−1 ◦ α = α.
Hence the path α has been lifted to a path α̃. If α(I) is not contained in a single
admissible neighbourhood of X, choose an open cover {Vi } of X consisting of
admissible open neighbourhoods. Let > 0 be a Lebesgue number for the open
covering {α−1 (Vi )} of I. We can now select a partition
0 = t0 < t1 < . . . < tn = 1

of I so that |ti+1 −ti | ≤ , for all i = 0, 1, . . . , n−1. Then, obviously α[ti , ti+1 ] is
contained in some admissible open neighbourhood, say, Vi+1 , i = 0, 1, . . . , n−1.
Let V1 denote the path component of p−1 (V1 ) to which the desired point x̃0
belongs and then define α̃(t) = ( p|V1 )−1 ◦ α(t), for all t ∈ [t0 , t1 ]. Then α̃ is a
path defined on [t0 , t1 ] in X̃ which lifts α|[t0 ,t1 ] . Now, inductively, suppose k > 1
and we have defined a lift α̃ : [t0 , tk ] → X̃ of α|[t0 ,tk ] . Then α[tk , tk+1 ] ⊂ Uk+1 .
Now, we choose a path component Vk+1 of p−1 (Uk+1 ) which contains α̃(tk ).
Since α|Vk+1 is a homeomorphism, define the extension ᾱ of α̃ by putting

ᾱ(t) = ( p|Vk+1 )−1 ◦ α(t), t ∈ [tk , tk+1 ]. Then ᾱ is a path in X̃ defined on
[t0 , tk+1 ] which lifts α|[t0 ,tk+1 ] . Taking k = n − 1, we have a path in X̃ defined
on I which lifts the given path α in X and has the initial point x̃0 .
The uniqueness of α̃, which lifts α and starts at the prescribed point x̃0
in X, follows at once from the preceding corollary because I is connected and,
by assumption, any two lifts of α agree at the point a = 0 ∈ I.
Homotopy Lifting Property
Having proved the path lifting property of a covering projection p : X̃ →

X, now we ask whether or not any homotopy F : A × I → X between two maps
f, g : A → X can be lifted to a homotopy F̃ : A × I → X̃. Since any f : A → X
is homotopic to itself, lifting of this homotopy F : A×I → X will automatically
imply that any map f can always be lifted to a map f˜: A → X, which evidently
need not be true. Hence the general answer to the above question cannot be
in affirmative. However, the answer is “yes” provided a lift of f or any map
which is homotopic to f under F is already given. This result is known as the
homotopy lifting property of the covering projection p : X̃ → X and the map
F̃ is also known as covering homotopy. We have
Theorem 5.2.7. (The Homotopy Lifting Property). Let p : X̃ → X be a covering

projection and A be a compact space. Suppose f˜: A → X̃ is a continuous map
and F : A × I → X is a homotopy starting from p ◦ f˜. Then there is a homotopy
F̃ : A×I → X̃ which starts from f˜ and lifts F . Furthermore, if F is a homotopy
relative to a subset S of A, then so is F̃ .
Proof. Since F is continuous and A is compact, the subset F (A×I) of X must be
compact. Therefore, we can cover F (A × I) by finite number of admissible open
sets, say, U1 , . . . , Ur . Then the open sets F −1 (U1 ), . . . , F −1 (Ur ) cover A×I. For
each point (a, t) ∈ A × I, find a rectangular neighbourhood V × (−, ) which
is contained in F −1 (Ui ) for some i. Then, using compactness of A × I, one can
find a finite number of open connected sets {Vα }n1 of A covering A and a finite
partition 0 = t0 < t1 < . . . < tn = 1 of I such that for every α, 1 ≤ α ≤ n, and
every i, 0 ≤ i ≤ k − 1, F (Vα × [ti , ti+1 ]) ⊂ Vj for some j, 1 ≤ j ≤ r.
First, we construct a continuous map G : A×[0, t1 ] → X̃ which agrees with
f on A×{0} and lifts F |A×[0,t1 ] . For every α, we know that F (Vα ×[0, t1 ]) ⊂ Uj
for some j, and also we know that f (Vα × {0}), being connected, must be
contained in one of the components, say, Wα of p−1 (Uj ). Hence we can define
a map
−1
Gα1 = ( p|Wα ) ◦ F : Vα × [0, t1 ] → X̃.
Then, obviously, Gα 1 is continuous, G1 (a, 0) = f (a) for all a ∈ A and p ◦ G1 =
α α
F |Vα ×[0,t1 ] . Next, we prove that for every pair α, β for which Vα ∩ Vβ
= 0,
β
Gα1 , G1 will match on the common domain (Vα ∩ Vβ ) × [0, t1 ]. For this, note
β
that on (Vα ∩ Vβ ) × {0}, Gα 1 = f = G1 . Moreover, for any t, 0 ≤ t ≤ t1 and any
a ∈ Vα ∩ Vβ , the point (a, t) can be joined to (a, 0) by an obvious path lying in
(Vα ∩ Vβ ) × [0, t1 ]. This implies that Gα
1 will map (Vα ∩ Vβ ) × [0, t1 ] into Wβ
also and so into Wα ∩ Wβ . Similarly Gβ1 will also map it into Wα ∩ Wβ . Thus,
on (Vα ∩ Vβ ) × [0, t1 ],
−1
Gα
1 = ( p|Wα ∩Wβ ) ◦ F = Gβ1 .
Since {Vα × [0, t1 ]} is an open cover of A × [0, t1 ], all the Gα
1 combine to give
a continuous map G1 : A × [0, t1 ] → X̃ such that G1 (a, 0) = f (a) for all a ∈ A
and p ◦ G1 = F |A×[0,t1 ] . Now, using the map Gi |A×{t1 } in place of f , we can
construct a continuous map G2 : A×[t1 , t2 ] → X̃ such that G2 (a, t1 ) = G1 (a, t1 )
for all a ∈ A and p ◦ G2 = F |A×[t1 ,t2 ] . Proceeding inductively, we can now
construct continuous functions Gi : A × [ti−1 , ti ] → X̃, i = 2, 3, . . . , k − 1,
such that for all a ∈ A, Gi (a, ti ) = Gi+1 (a, ti ) and p ◦ Gi = F |A×[ti−1 ,ti ] .
Applying continuity lemma for functions Gi on closed sets A × [ti−1 , ti ], we
have a continuous function F̃ : A × I → X̃ such that p ◦ F̃ = F and F̃ (a, 0) =
f (a) ∀ a ∈ A. This completes the proof of the existence of F̃ satisfying the
stated conditions.
Finally, it is clear from the construction of Gi ’s that if for any a ∈ A,
F (a, t) remains constant for all t ∈ I, then so are Gi ’s. Therefore, if the
homotopy F is relative to a subset S of A, then so is the homotopy F̃ .
We may remark that the above result remains true even when A is not
compact, but the proof would be different.
5.3 Applications of Homotopy Lifting Theorem

We will now present some applications of the path lifting property and
homotopy lifting property of covering projections. First of all, we must point
out that the path lifting property is a consequence of the homotopy lifting
property as shown below.
Corollary 5.3.1. If p : X̃ → X is a covering projection and x̃0 ∈ X̃, x0 ∈ X are

such that p(x̃0 ) = x0 , then for any path w : I → X such that w(0) = x0 , there
exists a unique path w̃ : I → X̃ such that w̃(0) = x̃0 and p ◦ w̃ = w.
Proof. The uniqueness follows from the fact that I is connected. Let P = {a}
be a singleton space and consider the map f : P → X defined by f (a) = x0 .
The path w defines a homotopy F : P × I → X on P by putting F (a, t) = w(t).
By the Homotopy Lifting Property, we have a map F̃ : P × I → X̃ such that
F̃ (a, 0) = x̃0 and p ◦ F̃ = F . Then w̃ : I → X̃ defined by w̃(t) = F̃ (a, t), t ∈ I,
is a path in X̃ which starts from x̃0 and has the property that for t ∈ I
pw̃(t) = pF̃ (a, t) = F (a, t) = w(t).
5.3. Applications of Homotopy Lifting Theorem 209
Recall that any continuous map f : X → Y induces a group homomor-

phism f∗ : π1 (X, x) → π1 (Y, f (x)) for all x ∈ X, defined by f∗ [α] = [f ◦ α].
If p : X̃ → X is a covering map, then we know that it is onto. But the in-
duced map, in general, need not be onto. However, it follows from the following
theorem that p∗ : π1 (X̃, x̃0 ) → π1 (X, x0 ) is always an injective map.
Theorem 5.3.2. (The Monodromy Theorem). Let p : X̃ → X be a covering

projection. Suppose α̃, β̃ are two paths in X̃ having the same initial point x̃0 .
Then α̃, β̃ are equivalent if and only if pα̃, pβ̃ are equivalent paths in X.
Remark. By definition, two paths in the space are equivalent means they have
the same initial and terminal points. Thus, if pα̃ and pβ̃ are equivalent, then
by above theorem α̃, β̃ are equivalent. We are given that they have the same
initial point. Hence it is a consequence (not a hypothesis) of the above theorem
that α̃, β̃ must have the same terminal point.
Proof. If F : I × I → X̃ is a continuous map such that

F (s, 0) = α̃(s), F (s, 1) = β̃(s), s ∈ I
F (0, t) = x̃0 , F (1, t) = x̃1 , t ∈ I,
then p ◦ F : I × I → X is a continuous map such that for all s ∈ I, p ◦ F (s, 0) =
pα̃(s), p ◦ F (s, 1) = pβ̃(s) and for all t ∈ I, p ◦ F (0, t) = p(x̃0 ), p ◦ F (1, t) =
p(x̃1 ). This proves the direct part.
Conversely, suppose F : I × I → X is a homotopy between pα̃ and pβ̃
relative to {0, 1} ⊂ I. By the Homotopy Lifting Property, there is a unique
homotopy F̃ : I × I → X̃ such that F̃ (0, 0) = x̃0 and pF̃ = F . Restricting F̃ on
(s, 0), s ∈ I, we find a path s → F̃ (s, 0) starting from x̃0 and lifting pα̃. Clearly,
s ; α̃(s) is also a path in X̃ starting from x̃0 and lifting pα̃. Hence, by the
uniqueness property of the covering paths, F̃ (s, 0) = α̃(s) for all s ∈ I. By a
similar argument, we have F̃ (s, 1) = β̃(s). Restricting F̃ on (0, t), t ∈ I, we get
a path t ; F̃ (0, t) which projects under p to the constant path at x0 ; on the
other hand, the constant path t ; x̃0 in X̃ also starts at x̃0 and projects under
p to the constant path t ; x0 in X. Hence, again by the uniqueness theorem,
t ; F̃ (0, t) must be a constant path based at x̃0 . Similarly, the path t ; F̃ (1, t)
must be a constant path based at some point x̃1 over x1 . This proves that F̃ is
a homotopy between α̃ and β̃ relative to {0, 1}. Therefore, α̃ is equivalent to β̃.
Now we can state the remark made earlier as a useful
Corollary 5.3.3. If p : X̃ → X is a covering projection and p(x̃) = x, then the

induced homomorphism p∗ : π1 (X̃, x̃) → π1 (X, x) is a one-one map.
As yet another application of the path lifting property of a covering pro-

jection p : X̃ → X, we prove the next result regarding the cardinality of fibres
p−1 {x}, x ∈ X once again; the earlier proof used only set topology whereas
the next one uses the concept of algebraic topology with a little gain, viz., a
specific bijection can be given between any two fibres.
Proposition 5.3.4. Let p : X̃ → X be a covering map. Then for any two points
a, b ∈ X, the fibres p−1 {a} and p−1 {b} have the same cardinality.
Proof. It is enough to produce a bijection from p−1 {a} to p−1 {b}: Fix a path
w in X with w(0) = a and w(1) = b, and let x ∈ p−1 {a}. Using the unique
path lifting property, we can find a path w̃x in X̃ such that w̃x (0) = x and
pw̃x = w. Then w̃x (1) ∈ p−1 {b} is uniquely determined by x. We define a
map f : p−1 {a} → p−1 {b} by putting f (x) = w̃x (1). We assert that f is one-
one because f (x1 ) = f (x2 ) implies w̃x1 (1) = w̃x2 (1), which means w̃x1 and
w̃x2 both agree at a point of I and hence, by uniqueness of the lifted paths,
w̃x1 = w̃x2 . Therefore, x1 = w̃x1 (0) = w̃x2 (0) = x2 . Next, we claim that f
is onto. For that, let z ∈ p−1 {b} and consider the path w−1 in X defined by
w−1 (t) = w(1−t). Let w̃z be the path in X̃ such that w̃z (0) = z and pw̃z = w−1 .
Then
p(w̃z )−1 (t) = p(w̃z )(1 − t)

= w−1 (1 − t)
= w(t),
and (w̃z )−1 (0) = w̃z (1). Hence (w̃z )−1 is a lift of w starting at w̃z (1). Putting
w̃z (1) = x, we see that x ∈ p−1 {a}, since both (w̃z )−1 and w̃x start at x
and both lift w, wx = (w̃z )−1 . It follows that w̃x (1) = (w̃z )−1 (1) = w̃z (0) = z.
From the above result, we see that for any covering projection p : X̃ → X,
all the fibres have same cardinality. This common cardinal number is called
the number of sheets of X̃ over X. If this number is a natural number n,
then we also express this fact by saying that X̃ is a n-sheeted covering of X.
For example, the covering projection p : S1 → S1 defined by p(z) = z 4 is a
four-sheeted covering of S1 by itself. Similarly, the quotient map q : S2 → P2
identifying the antipodal points of S2 is a two-sheeted covering of the projective
plane P2 .
The figure 5.4 depicts a three-sheeted covering of the figure-eight. Note

that the figure-eight space is a closed curve with one self-intersection, where as
the three-sheeted covering is a closed curve with three self-intersections.
The Right Action of the Fundamental Group
The number of sheets of a covering projection has another interesting

interpretation given by the following result:
5.3. Applications of Homotopy Lifting Theorem 211
Fig. 5.4: The three-sheeted covering of the figure-eight.
Proposition 5.3.5. Let p : X̃ → X be a covering projection. Then for any x ∈ X

and for any x̃ ∈ p−1 {x}, there is a canonical one-one correspondence between
the set p−1 {x} (number of sheets) and the set π1 (X, x)/p∗ (π1 (X̃, x̃)) of all
cosets of the subgroup p∗ (π1 (X̃, x̃)) in the group π1 (X, x).
Proof. We will define a natural right action of the group π1 (X, x) on the set
p−1 {x} and show that the action is transitive with p∗ (π1 (X̃, x̃)) as the isotropy
group at the point x̃ ∈ p−1 {x}. Take an element [α] ∈ π1 (X, x) and point
x̃ ∈ p−1 {x}. By the unique path lifting property, there is unique path α̃x in X̃
which starts from x̃ and lifts α. We define
x̃ · [α] = α̃x (1). (5.3.1)
We must show that the point α̃x (1) depends on the path class [α], not on
any representative of [α]. For that, choose another loop β in X based at x
so that [α] = [β]. Let β̃x be a path in X̃ which starts at x̃ and lifts β. Since
the paths α, β are homotopic and α̃x , β̃x , respectively, are paths in X̃ lifting
them and each starting at the same point x̃, we can apply the Monodromy
Theorem to conclude that α̃x and β̃x must have the same terminal points, i.e,
α̃x (1) = β̃x (1).
To prove that (5.3.1) defines an action, let x be the constant loop in X
based at x. Then, for any x ∈ X, the constant loop ˜x will be a path in X̃
which lifts x and starts at x̃. It follows that
x̃ · [x ] = ˜x (1) = x̃. (1)
Next, let [α], [β] ∈ π1 (X, x). For any x̃ ∈ p−1 {x}, first choose a path α̃x
in X̃ starting at x̃ and lifting α and then choose a path β̃x in X̃ which lifts β
and starts at α̃x (1). Then, by definition,
(x̃ · [α]) · [β] = α̃x (1) · [β] = β̃x (1).

On the other hand, α̃x ∗ β̃x is a path in X̃ starting at x̃ and lifting α ∗ β. This
means
x̃ · ([α] · [β]) = x̃ · [α ∗ β] = (α̃x ∗ β̃x )(1) = β̃x (1).
Consequently, for each x̃ ∈ p−1 {x},
x̃ · ([α] · [β]) = (x̃ · [α]) · [β]. (2)
Thus, in view of (1) and (2), the group π1 (X, x) acts on right on the set p−1 {x}.
Finally, we prove that the above action is also transitive. For that, let
x̃0 , x̃1 be any two points of p−1 {x}. Since X̃ is path connected, there is a path
w̃ in X̃ such that w̃(0) = x̃0 and w̃(1) = x̃1 . Evidently, then pw̃ = w is a loop
in X based at x, i.e, [w] ∈ π1 (X, x), and by definition, x0 ·[w] = w̃(1) = x1 . The
isotropy group of the above action at any x̃ is obviously p∗ (π1 (X̃, x̃)) because
x̃ · [α] = x̃ if and only if there is a closed loop α̃ in X̃ which lifts α and starts
at the point x̃, i.e., [α] = p∗ ([α̃]), which means [α] ∈ p∗ (π1 (X̃, x̃)). Hence it
follows from the property of the isotropy groups and the orbits that p−1 {x}
will be in 1-1 correspondence with the set π1 (X, x)/p∗ (π(X̃, x̃)).
Corollary 5.3.6. If p∗ : π1 (X̃, x̃) → π1 (X, x) is onto, then the map p : X̃ → X

is a homeomorphism.
5.4 Lifting of an arbitrary map
Now, we return to the “lifting problem” that we mentioned earlier: Given

a covering projection p : X̃ → X and a continuous map f : A → X, can we
find a continuous map f˜: A → X̃ such that p ◦ f˜ = f ? If f is a path or a
homotopy between paths, we have seen that the answer to the lifting problem
is in affirmative. However, for any continuous map f , the answer is negative.
We have the following:
Example 5.4.1. Consider the exponential map exp : R → S1 defined by exp(t) =

e2πit . We assert that the identity map Id : S1 → S1 cannot be lifted to a con-
tinuous map φ : S1 → R making the triangle
* R

φ

2πit
?e
S1 - S1
Id
commutative. If there was such a φ, then φ must be one-one. But then, S1
being compact will mean φ is an embedding of S1 into R. This is impossible
because any compact connected subset of R must be a closed interval and that
can never be homeomorphic to S1 .
In view of the above example, the lifting problem for covering projections
now amounts to asking for a necessary and sufficient condition so that any
5.4. Lifting of an arbitrary map 213
map f : A → X can be lifted. It is very interesting to remark that the above

question can be beautifully answered in terms of fundamental groups of spaces
A, X, X̃ and the homomorphisms amongst them induced by f and p. This is
one of the several fine illustrations of the methods of algebraic topology where
the topological problem of lifting a continuous map is faithfully translated into
an easier algebraic problem regarding fundamental groups. This result is stated
in the next theorem.
Theorem 5.4.2. (Lifting Theorem). Let p : X̃ → X be a covering projection

and f : A → X be any continuous map. Then, given any three points a0 ∈ A,
x0 ∈ X and x̃0 ∈ X̃ such that f (a0 ) = x0 and p(x̃0 ) = x0 , there exists a
continuous map f˜: A → X̃ satisfying f˜(a0 ) = x̃0 such that p ◦ f˜ = f if and
only if f∗ (π1 (A, a0 )) ⊂ p∗ (π1 (X̃, x̃0 )).
Proof. First, suppose there exists a continuous map f˜: A → X̃ satisfying the
given conditions. Then the following diagram of continuous maps is commuta-
tive:
(X̃, x̃0 )
*

f˜

p
?
(A, a0 ) - (X, x0 )
f
Passing on to the fundamental groups and induced homomorphisms, we see
that the following diagram must be commutative:
π1 (X̃, x̃0 )
*

f˜∗

p∗
?
π1 (A, a0 ) - π1 (X, x0 )
f∗
This shows that f∗ (π1 (A, a0 )) = p∗ (f˜∗ (π1 (A, a0 )) ⊂ p∗ (π1 (X̃, x̃0 )), and proves
the necessity of the given condition.
~ ~α(1)
f(~α )
f(a)=
~x0
~f ~
p X
α a f f α x=f(a)
a0 x0
A X
Fig. 5.5
Conversely, suppose the given algebraic condition is satisfied. We define
f˜: A → X̃ as follows: Let a ∈ A and choose a path α in A such that α(0) = a0
and α(1) = a. Then f α is a path in X starting at x0 . By the Path Lifting
Property, there exists a path α̃ in X̃ starting at x̃0 which projects to f α. We
define f˜(a) = α̃(1). In order that f˜ is well-defined, we must show that f˜(a) is
independent of the path α that joins a0 to a. So, let β be any other path in A
joining a0 to a and suppose β̃ is a path in X̃ starting from x̃0 such that pβ̃ = f β.
It suffices to show that α̃(1) = β̃(1). Now, observe that α ∗ β −1 is a loop in A
based at a0 . Hence, by the given algebraic condition, there exists a loop δ in X̃
based at x̃0 such that the loop pδ is homotopic to f (α ∗ β −1 ) = f α ∗ (f β)−1 .
But then by the Homotopy Lifting Theorem, there is a loop w based at x̃0 such
that pw = f α ∗ (f β)−1 . We now break the loop w into the product w1 ∗ w2−1 of
paths w1 , w2 in X̃ by defining w1 (t) = w(t/2), w2 (t) = w(1 + t/2), t ∈ I. Then
for each t ∈ I, pw1 (t) = pw(t/2) = (f α) ∗ (f β)−1 (t/2) = f α(t) and, similarly,
pw2 (t) = f β(t). Since w1 and α̃ agree at x̃0 and pw1 = f α = pα̃, we can apply
the Monodromy Theorem to conclude that w1 (1) = α̃(1) and w2 (1) = β̃(1).
However, since w1 (1) = w2 (1) = w( 12 ), we find that α̃(1) = β̃(1). It is clear
from the definition that pf˜ = f .
Finally, we prove that f˜ is continuous. Let a ∈ A and U be an open
neighbourhood of f˜(a). Choose an admissible neighbourhood U of pf˜(a) =
f (a) such that U ⊂ p(U ). Let W be the path component of p−1 (U ) which
contains the point f˜(a), and let U be an admissible neighbourhood of f (a)
such that U ⊂ p(U ∩W ). Then the path component of p−1 (U ) containing f˜(a)
must be contained in U . Since f is continuous and A is locally path connected,
we can find a path connected neighbourhood V of a such that f (V ) ⊂ U . Then
one can readily verify that f˜(V ) ⊂ U .
5.5 Covering Homomorphisms
We have already proved several important results concerning a covering

projection p : X̃ → X. In this section we want to consider the following
question: Given a space X, how many distinct covering projections of X one
can find? In other words, in how many distinct ways a space X can be covered?
Before answering this question, we must specify clearly as to what is meant
by distinct covering projections of a given space X. As an example, suppose
p : X̃ → R is a covering projection. Then, by the Monodromy Theorem,
the induced homomorphism p∗ : π1 (X̃) → π1 (R) must be one-one and since
π1 (R) = 0, it must also be onto. Therefore, by Corollary 5.3.6, p : X̃ → R
must be a homeomorphism, i.e, any covering projection p of R must be a
homeomorphism. The same argument shows that if X is simply connected,
then any covering map of X must be a homeomorphism. We emphasize that
all of our covering spaces here are assumed to be connected as well as locally
path connected.
As another example, the map qn : S1 → S1 defined by qn (z) = z n is a

n-sheeted covering projection of the circle S1 by itself. Hence, by any plausible
definition, the coverings qn and qm must be treated as distinct if m
= n. It must
be mentioned here that the space X̃ and the map p : X̃ → X both are essential
5.5. Covering Homomorphisms 215
parts of a covering projection – the same space X̃ can cover X in so many

distinct ways. These considerations lead us to the following well-established
definition:
Definition 5.5.1. Let p1 : X̃1 → X, p2 : X̃2 → X be two covering projections of

a space X. A continuous map h : X̃1 → X̃2 is said to be a homomorphism of
covering spaces if the diagram
X̃1
h - X̃
2
@
p1 p2
@@
R
X
is commutative. If, in addition, the map h is a homeomorphism, then h is called
an isomorphism of the two covering projections.
If we consider the class of all covering projections of a space X, then one

can easily verify that (i) the identity map IX̃ : X̃ → X̃ is a homomorphism,
and (ii) the composite of two homomorphisms is again a homomorphism.
It follows that covering projections p : X̃ → X as objects and homomor-
phism amongst them as morphisms, form a category, and the isomorphisms in
this category are just the isomorphisms of covering projections as defined above.
The covering spaces p1 : X̃1 → X, p2 : X̃2 → X will be called comparable

if one can find a homomorphism from either one of them to the other; they
will be considered same if there exists an isomorphism from one onto the other.
An isomorphism h from a covering space p : X̃ → X to itself is called an
automorphism or a deck transformation. In this section, we will answer the
following two questions very nicely in terms of the fundamental groups and the
induced homomorphisms amongst them: When are two covering projections of
a space X (a) comparable? (b) same?
Group of Deck Transformations
Let p : X̃ → X be a covering projection. Then the set of all automor-

phisms of this covering projection is closed under the obvious operation of
composition of maps. Furthermore, since the identity map of X̃ to itself is an
automorphism and the inverse of an automorphism is again an automorphism,
the set of all automorphisms of a covering projection forms a group with
respect to the composition operation. This group, usually denoted by A(X̃, p),
is called the automorphism group of the covering projection p : X̃ → X.
The automorphisms are also known as the covering transformations or deck
transformations of the covering projection p.
Let p1 : X̃1 → X, p2 : X̃2 → X be covering projections. Suppose

g, h : X̃1 → X̃2 are any two homomorphisms. Then one can think of each of g
and h as lifts of the map p1 : X̃1 → X with respect to the covering projection
p2 : X̃2 → X.
* X̃2

g,h

p2
?
X̃ - X
1 p1
Consequently, if g and h both agree on a single point of X̃1 , then g = h

because X̃1 is connected (Corollary 5.2.1). This yields the next
Proposition 5.5.2. Let g, h : X̃1 → X̃2 be two homomorphisms of the covering

projections. If g(x̃1 ) = h(x̃1 ) for some x̃1 ∈ X̃1 , then g = h.
We already know that the group Homeo(Y ) of all homeomorphisms of a

space Y always acts on the set Y by the action defined by
h · y = h(y),
where h ∈ Homeo(Y ) and y ∈ Y . Since the group A(X̃, p) of all automorphisms
of a covering projection p : X̃ → X is a subgroup of the group of all homeo-
morphisms of X̃, A(X̃, p) also acts on the space X̃ by the above action. In this
terminology, the following result is a special case of the foregoing proposition:
Corollary 5.5.3. The group G = A(X̃, p) of all automorphisms of the covering

projections p : X̃ → X acts on X̃ freely, i.e., if g ∈ G and g(x̃) = x̃ for any
x̃ ∈ X̃, then g = IX̃ .
Since an automorphism h : X̃ → X̃ is a homeomorphism making the dia-

gram h - X̃
X̃
@ p p
@@
R
X
commutative, h must map each point of the fibre p−1 {x} to a point of itself, i.e,
h simply permutes the points of each fibre. We now wish to answer the following
question: Given any two points x̃1 , x̃2 of p−1 {x}, can we find an automorphism
h : X̃ → X̃ of the covering projection p : X̃ → X such that h(x̃1 ) = x̃2 ? In
the terminology of group actions, the question amounts to asking whether the
group G = A(X̃, p) acts on X̃ so that it is transitive on each fibre p−1 {x},
x ∈ X. We will shortly see that the question has a very simple answer in
algebraic terms involving fundamental groups. Applying the Lifting Theorem
to the map p1 : X̃1 → X with respect to the covering projection p2 : X̃2 → X,
we have
Proposition 5.5.4. Let p1 : X̃1 → X, p2 : X̃2 → X be covering projections over

X and x̃1 ∈ X̃1 , x̃2 ∈ X̃2 be such that p1 (x̃1 ) = p2 (x̃2 ). Then there exists a
homomorphism h : X̃1 → X̃2 such that h(x̃1 ) = x̃2 if and only if
p1 ∗ (π1 (X̃1 , x̃1 )) ⊆ p2 ∗ (π1 (X̃2 , x̃2 )).
5.5. Covering Homomorphisms 217
If h : X̃1 → X̃2 above is a homeomorphism then not only p1 : X̃1 → X

has a lift h but the map p2 : X̃2 → X is also lifted by h−1 : X̃2 → X̃1 for
the covering projection p1 : X̃1 → X. Hence, in the above proposition, the re-
verse inclusion also holds, i.e, p2 ∗ (π1 (X̃2 , x̃2 )) ⊆ p1 ∗ (π1 (X̃1 , x̃1 )). Conversely,
suppose the equality holds. Then, again, by the Lifting Theorem, there are ho-
momorphisms h1 : X̃1 → X̃2 and h2 : X̃2 → X̃1 such that p2 h1 = p1 , p1 h2 = p2 ,
h1 (x̃1 ) = x̃2 and h2 (x̃2 ) = x̃1 . This means h2 ◦ h1 : X̃1 → X̃1 is a homomor-
phism which maps x̃1 to itself, and so by Proposition 5.5.2, h2 ◦ h1 = IX̃1 .
Similarly, we have h1 ◦ h2 = IX̃2 . Thus, each of h1 and h2 is an isomorphism
and we have
Proposition 5.5.5. Let p1 : X̃1 → X, p2 : X̃2 → X be covering projections

and let x̃1 ∈ X̃1 , x̃2 ∈ X̃2 be points such that p1 (x̃1 ) = p2 (x̃2 ). Then there
exists an isomorphism h : X̃1 → X̃2 satisfying h(x̃1 ) = x̃2 if and only if
p1 ∗ (π1 (X̃1 , x̃1 )) = p2 ∗ (π1 (X̃2 , x̃2 )).
This immediately yields
Corollary 5.5.6. Let p : X̃ → X be a covering projection and x̃1 , x̃2 ∈ p−1 {x}.
Then there exists an automorphism h of X̃ such that h(x̃1 ) = x̃2 if and only
if p∗ (π1 (X̃, x̃1 )) = p∗ (π1 (X̃, x̃2 )). In particular, the group G = A(X̃, p) acts
transitively on the fibre p−1 {x} if p∗ (π1 (X̃, x̃1 )) = p∗ (π1 (X̃, x̃2 )) for every
x̃1 , x̃2 ∈ p−1 {x}.
Next, we note the following:
Proposition 5.5.7. Let p : X̃ → X be a covering projection. Then the set

{p∗ (π1 (X̃, x̃)) | x̃ ∈ p−1 {x}} forms a complete conjugacy class of subgroups
of the group π1 (X, x).
Proof. Let x̃1 , x̃2 ∈ p−1 {x}. Since X̃ is path connected, there is a path, say,
w in X̃ joining x̃1 to x̃2 . Then the map fw : π1 (X̃, x̃1 ) → π1 (X̃, x̃2 ) defined by
fw [α] = [w−1 ∗ α ∗ w] is clearly an isomorphism. Note that pw is a closed path
at x. Since the diagram
fw
π1 (X̃1 , x̃1 ) π1 (X̃1 , x̃2 )
p∗ p∗
π1 (X, x) π1 (X, x)
fpw
is easily seen to be commutative, we find that
p∗ (π1 (X̃, x̃2 )) = p∗ fw (π1 (X̃, x̃1 )) = fpw (p∗ (π1 (X̃, x̃1 ))).
Because fpw is simply a conjugation in π1 (X, x) by the element [pw], we con-

clude that p∗ (π1 (X, x̃1 )) and p∗ (π1 (X, x̃2 )) are conjugate in π1 (X, x) for any
two points x̃1 , x̃2 in p−1 {x}. On the other hand, suppose H is a subgroup of
π1 (X, x) which is conjugate to p∗ (π1 (X̃, x̃0 )) for some x̃0 ∈ p−1 {x}. Then there
is a closed path δ in X based at x such that
H = [δ]−1 p∗ (π1 (X̃, x̃0 ))[δ].
Let δ̃ be a lift of δ starting from x̃0 and put x̃1 = δ̃(1). Then we assert that
H = p∗ (π1 (X̃, x̃1 )). To see this, let β be any loop in X̃ based at x̃0 . Then
δ −1 ∗ pβ ∗ δ is a loop in X based at x. Since p(δ̃ −1 β δ̃) = δ −1 pβδ and δ̃ −1 β δ̃ is
a loop based at x̃1 in X̃,
[δ −1 ]p∗ (π1 (X̃, x̃0 ))[δ] ⊆ p∗ (π1 (X̃, x̃1 )).
Conversely, if α is a loop in X̃ based at the point x̃1 , then pα ∼ δ −1 δpαδ −1 δ ∼

δ −1 (pδ̃ ∗ pα ∗ (pδ̃)−1 )δ ∼ δ −1 p(δ̃ ∗ α ∗ δ −1 )δ. Because δ̃ ∗ α ∗ δ̃ −1 is a loop in X̃
based at x̃0 , p∗ (π1 (X̃, x̃1 )) ⊆ [δ −1 ]p∗ (π1 (X̃, x̃))[δ].
Corollary 5.5.8. The group G = A(X̃, p) acts transitively on a fibre p−1 {x}
if and only if the subgroup p∗ (π1 (X̃, x̃)) is normal in π1 (X, x) for every x̃ ∈
p−1 {x}.
Now, we can characterize as to when two covering projections of X are

comparable. We have
Proposition 5.5.9. Let p1 : X̃1 → X and p2 : X̃2 → X be two covering pro-

jections of the space X. Then there exists a homomorphism h : X̃1 → X̃2 of
covering projections if and only if for any x̃1 ∈ p−1 −1
1 {x} and x̃2 ∈ p2 {x},
p1 ∗ (π1 (X̃1 , x̃1 )) is conjugate to a subgroup of p2 ∗ (π1 (X̃2 , x̃2 )).
Proof. First, assume that there is a homomorphism h : X̃1 → X̃2 . Then, by the
Lifting Theorem, we must have
p1 ∗ (π1 (X̃1 , x̃1 )) ⊆ p2 ∗ (π1 (X̃2 , h(x̃1 ))).
By the previous proposition, since the subgroup p2 ∗ (π1 (X̃2 , h(x̃1 ))) is conjugate
to p2 ∗ (π1 (X̃2 , x̃2 )) in π1 (X, x), it follows that p1 ∗ (π1 (X̃1 , x̃1 )) is conjugate to a
subgroup p2 ∗ (π1 (X̃2 , x̃2 )).
Conversely, if
[α]−1 p1 ∗ (π1 (X̃, x̃1 ))[α] ⊆ p2 ∗ (π1 (X̃2 , x̃2 ))
then
p1 ∗ (π1 (X̃1 , x̃1 )) ⊆ [α]p2 ∗ (π1 (X̃2 , x̃2 ))[α]−1

= p2 ∗ (π1 (X̃2 , x̃3 ))
5.6. Universal Covering Space – Applications 219
for some x̃3 ∈ p−1

2 {x}. Hence, again by the Lifting Theorem, there exists a
homomorphism h : X̃1 → X̃2 (such that h(x̃1 ) = x̃3 ).
The question as to when are two covering spaces of a space “same” is

completely answered in terms of the fundamental groups by the following result:
Theorem 5.5.10. Let p1 : X̃1 → X, and p2 : X̃2 → X be two covering projec-

tions. Then X̃1 is isomorphic to X̃2 if and only if for any x̃1 ∈ X̃1 , x̃2 ∈ X̃2
such that p1 (x̃1 ) = p2 (x̃2 ), p1 ∗ (π1 (X̃1 , x̃1 )) is conjugate to p2 ∗ (π1 (X̃2 , x̃2 )).
Proof. First, suppose h : X̃1 → X̃2 is an isomorphism of the two covering projec-
tions. Let x̃ ∈ X̃1 be such that p1 (x̃) = p1 (x̃1 ). Then, by the Lifting Theorem,
p1 ∗ (π1 (X̃1 , x̃)) ⊆ p2 ∗ (π1 (X̃2 , h(x̃))).
Since h−1 : X̃2 → X̃1 is also an isomorphism and maps h(x̃) to x̃, we have, by
the same theorem,
p2 ∗ (π1 (X̃2 , h(x̃))) ⊆ p1 ∗ (π1 (X̃1 , x̃)).
Hence the above inclusions reduce to equality. Since, by Proposition 5.5.7,

p1 ∗ (π1 (X̃1 , x̃1 )) is conjugate to p1 ∗ (π1 (X̃1 , h(x̃))) and p2 ∗ (π1 (X̃2 , h(x̃))) is con-
jugate to p2 ∗ (π1 (X̃2 , h(x̃))), and thus we have proved the direct part.
Conversely, suppose the given groups are conjugate. Then, by Proposition
5.5.9, there is a homomorphism h1 : X̃1 → X̃2 such that h1 (x̃1 ) = x̃2 . Similarly,
there is a homomorphism h2 : X̃2 → X̃1 such that h2 (x̃2 ) = x̃1 . It follows that
h2 h1 : X̃1 → X̃1 is a homomorphism fixing the element x̃1 ∈ X̃1 , which, in
turn, says that h2 h1 = IX̃1 . Likewise h1 h2 = IX̃2 . Hence each of h1 and h2 is
an isomorphism.
Remark. Note that for any covering projection p : X̃ → X, the subgroups

{p∗ (π1 (X̃, x̃)) | x̃ ∈ p−1 {x}} form a conjugacy class of subgroups of π1 (X, x).
The above theorem says that a conjugacy class of subgroups of π1 (X, x) com-
pletely determines the covering projection p up to isomorphism. Here, we are
not saying that given any conjugacy class of subgroups of π1 (X, x), there al-
ways exists a covering projection p : X̃ → X which yields that conjugacy class,
though it is true. This is an important question which we have not considered
so far (see exercises at the end).
5.6 Universal Covering Space – Applications
The concept of universal covering space of a given space X is extremely

useful in several situations. Before we give its definition and examples, let us
prove the following result.
Proposition 5.6.1. Let p1 : X̃1 → X and p2 : X̃2 → X be two covering projec-

tions and suppose h : X̃1 → X̃2 is a homomorphism of covering spaces. Then h
itself is a covering projection.
Proof. First, we show that h is onto. Let z ∈ X̃2 . Since p1 is onto and p2 (z) ∈ X,
there is a point x1 ∈ X̃1 such that p1 (x1 ) = p2 (z). Let x2 = h(x1 ). Since
X̃2 is path connected, there is a path w2 in X̃2 which starts from x2 and
terminates into z. Then, evidently, p2 w2 is a loop based at p1 (x1 ). Hence, by
the Path Lifting Property, there is a path w1 in X̃1 starting at x1 such that
p1 w1 = p2 w2 . Let z̃ = w1 (1). Note that hw1 and w2 are both paths in X̃2
starting from h(x1 ) = x2 such that
p2 hw1 = p1 w1 = p2 w2 .
Hence, by the uniqueness of lifted paths, hw1 = w2 . This means
h(z̃) = h(w1 (1)) = w2 (1) = z.
Next, we prove that each point of X̃2 has an open admissible neighbour-
hood. Let z ∈ X̃2 . Observe that there is an open admissible neighbourhood U1
of p2 (z) for the covering p1 , and an open admissible neighbourhood U2 for the
covering p2 . Hence, if we take U to be the path component of U1 ∩ U2 which
contains p2 (z), then U is an admissible neighbourhood for p1 and p2 both. Now,
let V be the path component of p−1 2 (U ) which contains z. We claim that V is
an admissible neighbourhood for h : X̃1 → X̃2 . To see this, let x1 be any point
in X̃1 such that h(x1 ) = z. Then x1 ∈ p−11 (p2 (z)). Since p1 is a covering projec-
tion, the path component V of p−1 1 (U ), which contains x1 , is homeomorphic
to U under p1 . But this means h|V : V → V is a homeomorphism because
h|V = ( p2 |U )−1 ◦ ( p1 |V ).
Now, we have
Definition 5.6.2. A covering projection q : U → X in which the covering space

U is simply connected is called the Universal Covering space of X.
Let us observe that, since U is simply connected, π1 (U ) = 0 and so for any

covering projection p : X̃ → X, q∗ (π1 (U )) = {0} ⊆ p∗ (π1 (X̃)). Therefore, by
the Lifting Theorem, there exists a map h : U → X̃ such that p ◦ h = q. By the
previous proposition, this means h : U → X̃ is a covering projection. It follows
that the universal covering space U of a space X is big enough so that it covers
any covering space of X. It is for this reason that U is called Universal. It also
follows from above that any two universal covering spaces of X are isomorphic
and hence the universal covering space U of X is unique, provided it exists.
Example 5.6.3. (a) Since R is simply connected, the exponential map p : R →

S1 , defined by p(t) = e2πit , is the universal covering of S1 .
(b) Since R×R is also simply connected, the map q : R×R → S1 ×S1 , defined by
q(t1 , t2 ) = (e2πit1 , e2πit2 ), is the universal covering space of torus T = S1 × S1 .
(c) The quotient map f : S2 → P2 which identifies the antipodal points is a
covering projection which is universal since S2 is simply connected. More gen-
erally, since Sn (n ≥ 2) is simply connected, the quotient map which identifies
the antipodal points is a universal covering projection of the projective space
Pn , (n ≥ 2).
The next result gives an interesting interpretation of the fundamental

group of a space X, which admits the universal covering space.
Proposition 5.6.4. Let q : U → X be the universal covering of a space X. Then

the automorphism group A(U, q) is isomorphic to the fundamental group π1 (X)
of X. Moreover, the order of π1 (X) is equal to the number of sheets of the
universal covering.
Proof. The last part follows from Proposition 5.3.5, since π1 (U ) = 0. To
prove the main result, let u0 ∈ U and q(u0 ) = x0 . We define a map
T : A(U, q) → π1 (X, x0 ) as follows: If f ∈ A(U, q), then since f permutes the
points of the fibre q −1 (x0 ), the point f (u0 ) is in this fibre. Let α be a path in
U joining u0 with f (u0 ). Then qα is a loop in X based at x0 . We define
T (f ) = [qα].
If β is any other path joining u0 with f (u0 ) then, since U is simply connected,
α is equivalent to β which means [qα] = [qβ]. Then the map T is well defined.
To prove that T is a homomorphism, let f, g ∈ A(U, q) and suppose α, β
are paths in U joining u0 to f (u0 ) and to g(u0 ), respectively. Then T (f ) = [qα]
and T (g) = [qβ]. Observe that f β is a path joining f (u0 ) to f (g(u0 )) and so
α ∗ f β is a path in U joining u0 with f (g(u0 )). Therefore,
T (f ◦ g) = [q(α ∗ f β)] = [qα ∗ qf β] = [qα] ◦ [qf β].
Since qf = q, we see that T (f ◦ g) = T f ◦ T g. Next, we see that T is one-

one because T (f ) = T (g) implies [qα] = [qβ] where α, β are paths in U
starting from u0 . This means q∗ [α] = q∗ [β] and so, by the Monodromy Theorem,
α, β must have the same terminal point also, i.e, f (u0 ) = g(u0 ). Since any
two automorphisms of q : U → X which agree at a single point must agree
everywhere (Proposition 5.5.2), we must have f = g.
Finally, to show that T is onto, let α ∈ π1 (X, x0 ). Let α̃ be a lift of the
path α in U such that α̃(0) = u0 . Consider the diagram
* U

h

q
?
U -X
q
Since U is simply connected covering projection of X, there exists a homo-

morphism h : U → U such that h(u0 ) = α̃(1). By the same argument, there is
also a homomorphism k : U → U such that k(α̃(1)) = u0 . It follows that the
homomorphism k ◦ h : U → U must map u0 to itself and so, by Proposition
5.5.2, it must be the identity map; similarly h ◦ k too must be identity. This
shows that h ∈ A(U, q) and, by definition, T (h) = [q α̃] = [α].
Example 5.6.5. The above result immediately yields the fundamental group
π1 (RPn ) of the projective space RPn for each n ≥ 2. For this, consider the
universal covering q : Sn → RPn , where Sn is the standard n-sphere and the map
q identifies the antipodal points of Sn . Since q is a 2-sheeted covering projection,
we can apply the above proposition to assert that the group π1 (RPn ) must be of
order two. Hence, π1 (RPn ) ∼
= Z/2Z. The other alternative is to verify directly
that the group of automorphisms of q : Sn → RPn has only two elements.
Example 5.6.6. The covering R → S1 by exponential map is a universal cov-

ering. Consider the set of all translations of R by all integers. Each of this is
easily seen to be a deck transformation of the covering and so that group of
deck transformations is isomorphic to the additive group of integers. This says
that the fundamental group of the circle is the group Z of integers – a result
already proved directly.
Example 5.6.7. The product covering R × R → S1 × S1 is again a universal

covering and the group of Deck transformations is easily seen to be isomorphic
to Z ⊕ Z. It follows that the fundamental group of torus is Z ⊕ Z, which is also
already computed.
Recall the concepts of ‘group action’ and ‘properly discontinuous action’

as discussed in Appendix 7.3.
Proposition 5.6.8. Let G act freely and properly discontinuously on X. Then

the quotient map η : X → X/G is a covering projection.
Proof. Any element of X/G will be of the form η(x) for some x ∈ X. Since the
action is free, it follows from the definition of properly discontinuous action
that there exists an open set V of X with x ∈ V and g ∩ gV = φ for any
g
= e in G.Then W = η(V ) is an open set containing η(x) in X/G. Moreover,
η −1 (W ) = g∈G gV a disjoint union and η|gV : gV → W is a homeomorphism
for each g ∈ G.
For covering projections which arise, as above, from a free and properly
discontinuous action of a group G on a simply connected space X, it is
interesting to observe (see Bredon [2] p.149) that G is isomorphic to the

fundamental group of the orbit space X/G. This follows from Proposition
5.6.4 and the fact that if X is simply connected, then the group of deck
transformations of the covering map X → X/G is isomorphic to G itself. (See
Exercise 14 at the end of the chapter for a general result.)
As an application of the last observation, we can determine the funda-

mental group of lens spaces L(p, q).
Example 5.6.9. (Fundamental Group of lens spaces). Recall (see Appendix 7.3)
that if p, q are any two positive integers, which are relatively prime, then we
have defined an action of the finite group Zp on the space S3 . The quotient
space S3 /Zp is called the lens space L(p, q). Since the action of Zp on S3 is
free and properly discontinuous, and S3 is simply connected, it follows from
the above observation that π1 (L(p, q)) ∼
= Zp for each q.
Covering projections and fundamental groups, as we have seen so far,

are very closely related. Fundamental groups were used not only in proving
the existence of covering projections, but also in classifying them. There are
situations when both can be combined to produce interesting results of deeper
significance. Recall that, using homology groups and degree concept of a map
from Sn to Sn , we proved the Borsuk-Ulam theorem, viz., there cannot be a
continuous map from Sn to Sn−1 preserving antipodal points. This classical
result can also be proved using the ideas of covering projections and homotopy
groups. We prove below the case n = 2 of Borsuk-Ulam theorem using covering
projections and fundamental groups; the case n ≥ 3 requires, in addition, the
computation of the cohomology algebra of higher-dimensional projective spaces
which we have not dealt with so far.
Example 5.6.10. (Borsuk-Ulam Theorem for n = 2). There cannot be a con-

tinuous map S2 → S1 which preserves antipodal points. Suppose, on the con-
trary, there is a continuous map f : S2 → S1 satisfying the condition that
f (−x) = −f (x) for any x ∈ S2 . We show that this leads to a contradiction. We
pass on to the quotient spaces P2 and P1 of S2 and S1 , respectively, by iden-
tifying antipodal points. Note that the projective space P1 is homeomorphic
to S1 itself. Since f preserves antipodal points, it induces a continuous map
g : P2 → P1 which makes the following diagram commutative:
f
S2 S1
q1 q2
P2 g P1
We know that both q1 and q2 are two sheeted covering projections. Note
that π1 (P2 ) ∼= Z/2Z and π1 (P1 ) ∼ = Z and the latter group has no ele-
ments of finite order other than the identity. Hence the induced homomophism
g∗ : π1 (P2 ) → π1 (P1 ) must be a trivial homomorphism. Next, we show that g∗
is indeed nontrivial. Let α be a path in S2 joining a pair of antipodal points,
say, x0 and −x0 . Then q1 ◦ α is a closed path in P2 and f ◦ α is a path joining
the pair of antipodal points f (x0 ), f (−x0 ). Therefore, q2 ◦(f ◦α) is also a closed
path in P1 , and we assert that this is not null-homotopic. To see this, suppose
it is null-homotopic and let Cf (x0 ) be the constant path in S1 based at f (x0 ).
Then
q2 ∗ [Cf (x0 ) ] = [q2 ◦ (f ◦ α)] = q2 ∗ [f ◦ α].
Hence, by the Monodromy Theorem, f ◦ α must have the same terminal point
as Cf (x0 ) , i.e, f ◦ α(1) = f (−x0 ) = f (x0 ), a contradiction. Now, since
g∗ [q1 ◦ α] = [g ◦ (q1 ◦ α)] = [q2 ◦ f ◦ α] = q2 ∗ [f ◦ α]

= 0,
we conclude that g∗ is nontrivial.
Exercises
1. Let p : X̃ → X be a covering projection. Prove that the family of admis-

sible open neighbourhoods is a basis for the topology of X.
2. Suppose X is connected and locally path-connected and p : X̃ → X is a
covering projection. Prove that
(a) X is Hausdorff implies X̃ is Hausdorff.
(b) X is completely regular if and only if X̃ is completely regular.
(c) X is locally compact if and only if X̃ is locally compact.
3. Give an example of a Hausdorff space X and a group G acting on X
freely and properly discontinuously so that the orbit space X/G is not
Hausdorff. This shows that the converse part of 2(a) is not true. (Hint:
See Massey [13] pp. 167–169.)
4. Let p : X̃ → X be a covering projection where X is connected and locally
path-connected. Prove that X̃ is compact if and only if X is compact and
p is finite sheeted.
5. Let pn : X̃n → Xn be the
kcovering k exp : R → S , for all n ≥ 1.
k projection
1
Show that the product 1 pn : 1 X̃n → 1 Xn is a covering projection.

6. Let L(p, q) denote the lens space, where p, q are relatively prime positive
integers. Prove that if p1
= p2 , then L(p1 , q), L(p2 , q) are not homeomor-
phic.
7. Construct a 6-sheeted covering projection of the figure of eight.
8. Let RPn (n ≥ 2) denote the n-dimensional real projective space. Prove

that any map f : RPn → S1 , the unit circle, is null-homotopic. Is it true
when RPn is replaced by Sn ?
9. Let G, G̃ be connected and locally path-connected topological groups and
p : G̃ → G be a covering projection which is also a group homomorphism.
Prove that ker p ∼
= Aut(G̃, p).
Existence of universal coverings
10. A space X is said to be semi-locally simply connected if for each x ∈ X,

there exists an open neighbourhood U of x in X such that the inclusion
map U ⊂ X induces trivial homomorphism π1 (U ) → π1 (X). Prove that
a connected, locally path-connected space X has a universal covering if
and only if it is semi-locally simply connected.
11. Let X be a connected, locally path-connected and semi-locally simply con-
nected space. Show that for any given conjugacy class {H} of subgroups
of π1 (X), there exists a covering p : X̃ → X such that the conjugacy class
of the subgroup p∗ (π1 (X̃)) agrees with the class of H.
12. Determine the fundamental group of an infinite product of circles. Deduce
that the latter has no universal covering space, whereas any product of
n-spheres (n ≥ 2) always has an obvious universal covering. (Hint: Use
Exercise 10 above.)
13. Let X be the subset of the real plane consisting of circles having radius
1/n and center (1/n, 0) for n = 1, 2, . . .. Show that X has no universal
covering. (Hint: Use Ex 10 above.)
14. For a covering projection p : X̃ → X, let p(x̃0 ) = x0 and H =
p# (π1 (X̃, x̃0 )). Suppose N (H) is the normalizer of H in π1 (X, x0 ). Prove
that N (H)/H is isomorphic to the group Aut(X̃, p). In particular, if X̃ is
simply connected, then π1 (X, x0 ) ∼ = Aut(X̃, p).
Chapter 6
Singular Homology
6.1 Introduction
Having defined the simplicial homology groups of a polyhedron (space

of a general simplicial complex) now we come to another kind of homology
groups, called singular homology groups of a topological space. The first
interesting feature of these homology groups is that they are defined for all
topological spaces X, not only for polyhedra. Singular homology groups were
first defined by S. Lefschetz in 1933, and were perfected in their present form
by S. Eilenberg (1913–1998) in the beginning of the 1940’s. These turn out
to be the most important and natural generalization of simplicial homology
groups, and are most suited for the study of topological manifolds. We may
recall (Theorem 4.9.6) that it took a good amount of hard work and new ideas
to prove the topological invariance of simplicial homology in Chapter 4. This
was first done by J.W. Alexander (1888–1971). In a contrast to this, we will
see in this chapter that the topological invariance of singular homology follows
almost obviously – this is another attractive feature of singular homology.
Historically, several kinds of homology groups, e.g., Vietoris homology,

Čech homology, etc., were defined, but each was defined for a restricted class
of topological spaces. However, their basic properties were found to be the
same or at least very similar to each other. Each of these basic properties was
considered a deep theorem for the particular homology under consideration.
It was only in 1945 that S. Eilenberg, jointly with N. Steenrod (1910–1971),
succeeded in a very ingenious way, in clarifying and unifying all these homology
groups by defining a homology theory axiomatically. They recognized the
basic data underlying different homology groups and identified a set of seven
axioms to define a homology theory. Each of these seven axioms was a basic
theorem for different homologies proposed at the time. Eilenberg and Steenrod
also proved the interesting fact that any two homology theories, for which
the zero-dimensional groups of a point-space (viz., the coefficient group) were
228 Chapter 6. Singular Homology
isomorphic, gave the same result for any compact polyhedron. This was indeed
their important uniqueness theorem. The compact polyhedra, whose geometry
and topology we studied in Chapter 3, therefore, gained even better status
than before.
The most remarkable aspect which emerged out of Eilenberg-Steenrod

axiomatization was that several results, valid for other homology theories,
were derived using only these axioms, not the specific definitions of those
homologies. As a matter of fact, this axiomatization brought about a sort
of revolution in algebraic topology and the subject grew thereafter very
fast cutting deep into other areas of mathematics. It continues unabated
ever since. One of the greatest outcomes of this revolution was the birth of
“generalized homology and cohomology theories”, which include fascinating
topics like K-theory, bordism and cobordism theories, etc. The concept of
relative homology of a pair (X, A), which will be explained later, is of crucial
importance in axiomatizing a homology theory. Since our scope is quite
limited, we will not go into other aspects of singular homology except for
giving its brief introduction. A curious reader, however, may referred to any of
the advanced books such as [2], [6], [15], [18], for a fuller and comprehensive
study of this subject.
In this chapter, we explain the definition of singular homology of a topo-

logical pair, and establish that it is a homology theory in the sense of Eilenberg-
Steenrod. Our approach is quite brief and pointed towards proving those the-
orems about singular homology, which go to show that the singular homology
satisfies the seven axioms of Eilenberg-Steenrod. It is true that the real purpose
and importance of singular homology lies in proving theorems of topology and
solving open problems of topology, geometry and other areas of mathematics.
But that can be done after accomplishing this major goal. Numerous results
about singular homology become available to us as a consequence of singular
homology being a homology theory. We don’t have to prove these results as
something special to singular homology. Eilenberg-Steenrod also defined the
concept of a “cohomology theory” as a dual of a homology theory, satisfying
seven axioms analogous to those of a homology theory. After defining singular
homology, we introduce singular cohomology valid for the category of all topo-
logical pairs and continuous maps. It turns out that a cohomology theory has
more structure than a homology theory, viz., there is a natural product called
“cup product” between any two cohomology classes of a given pair (X, A). As a
result, the direct sum of cohomology groups of (X, A) of all dimensions becomes
not only a graded-ring but a graded R-algebra, where R is the ground ring.
Again, the singular cohomology is the most general example of a “cohomology
theory” in the sense of Eilenberg-Steenrod. Their classic book (see [6]) deals
with all of these aspects almost exhaustively.
6.2. Singular Chain Complex 229
6.2 Singular Chain Complex

Let Rω denote a countably infinite product of copies of the real line R.
We fix the following points of Rω :
e0 = (0, 0, . . .)
e1 = (1, 0, 0, . . .)
e2 = (0, 1, 0, . . .)
..
.
en = (0, 0, . . . , 1, 0, . . .), 1 at the nth place, etc.
Let Rn denote the n-dimensional Euclidean space. Then Rn can be
identified with a subspace of Rω under the map (x1 , x2 , . . . , xn ) →
(x1 , x2 , . . . , xn , 0, 0, . . .). With this identification, R ⊂ R2 ⊂ · · · ⊂ Rn ⊂ ···
Let Δn = e0 , e1 , . . . , en be the standard geometric n-simplex spanned by the
points e0 , e1 , . . . , en in Rω .
Given a set of q + 1 points x0 , x1 , . . . , xq in Rn , there is a unique affine

map from Δq to Rn which maps ei to xi for i = 0, 1, . . . , q. We denote this map
by the symbol (x0 , x1 , . . . , xq ). Now we have our basic concept,
Definition 6.2.1. Let X be a topological space. A continuous map σ : Δq → X

is called a singular q-simplex in X.
Note that since Δq is a compact connected space, the image set σ(Δq ) in
X must be compact and connected. Thus, a singular 1-simplex in X is simply
a path in X, whereas a singular 0-simplex in X is just a point in X; a singular
2-simplex in X will be a curved triangle or region (see Fig. 6.1) with its interior.
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxxxxx
xxxxxxxxxxxxxx
xxxxxxxxxxxxxx
xxxxxxxxxxxxxx
xxxxxxxxxxxxxx
xxxxxxxxxxxxxx
xxxxxxxxxxxxxx
Fig. 6.1: A singular 1-chain and a singular 2-chain
Now, let R be a commutative PID (principal ideal domain) with 1

= 0,
and let Sq (X) be the free R-module generated by the set of all singular q-
simplexes in X. The elements of Sq (X), which are formal linear combinations
of singular q-simplexes in X with coefficients in R, are called singular q-chains
in X. We now proceed to define boundary homomorphisms Sq (X) → Sq−1 (X),
for all q ≥ 1, so that we get a chain complex. For this, let Fqi : Δq−1 → Δq ,
0 ≤ i ≤ q, be the unique affine maps which map the vertices (e0 , e1 , . . . , eq−1 ) to
the set of points (e0 , e1 , . . . , êi , . . . , eq ), where êi means the point ei is omitted.
In other words, let

ej , j<i
Fqi (ej ) =
ej+1 , j ≥ i
The map Fqi is called the ith face operator of the q-simplex Δq , and it simply
maps the (q − 1)-simplex Δq−1 into the ith face of the q-simplex Δq , i.e., to the
face of Δq opposite to the vertex ei . Thus, F2i will map the standard 1-simplex
e0 , e1 to e1 , e2 , e0 , e2 and e0 , e1 under the face operators F20 , F21 , F22 ,
respectively. For a given singular q-simplex σ, the composite σ (i) = σ ◦ Fqi is
called the ith face of σ, i = 0, 1, . . . , q. Now, we can define the boundary ∂(σ)
of a singular q-simplex σ by the formula

q
∂(σ) = (−1)i σ (i) .
0
The right hand side is the sum of (q+1) singular (q−1)- simplexes. Since Sq (X)
is the free R-module generated by the set of all singular q-simplexes in X, we
can extend ∂ by linearity to an R-homomorphism
∂q : Sq (X) → Sq−1 (X), for
all q ≥ 1, i.e., for any singular chain c = rσ σ, we define

∂q (c) = rσ ∂(σ), rσ ∈ R.
We note that if in the standard q-simplex Δq = e0 , e1 , . . . , eq , we first

drop the ith vertex followed by the jth vertex, j < i, then the result is same as
first dropping the jth vertex and then dropping (i − 1)th vertex. Reversing this
observation and using the definition of face operators, we get the following:
j
Lemma 6.2.2. The face operators Fqi , Fq−1 , 0 ≤ j < i ≤ q, satisfy the following
condition:
j
Fqi ◦ Fq−1 = Fqj ◦ Fq−1
i−1
.
Using the above lemma, we have
Proposition 6.2.3. The composite ∂q−1 ◦ ∂q = 0, for all q ≥ 1.
Proof. It suffices to prove that the composite ∂ ◦ ∂ vanishes on generators

σ ∈ Sq (X). We have

∂∂(σ) = (−1)i ∂(σ (i) )
j
= (−1)i ( (−1)j ((σ ◦ Fqi ) ◦ Fq−1 )

q
j

q−1
j
= (−1)i+j σ ◦ (Fqi Fq−1 )+ (−1)i+j σ ◦ (Fqi Fq−1 )
j<i=1 0≤i≤j

q
(i−1)

q−1
j
= (−1)i+j σ ◦ (Fqj Fq−1 ) + (−1)i+j σ ◦ (Fqi Fq−1 )
j<i=1 0≤i≤j

q
q−1
j
= (−1)k+l+1 σ ◦ (Fqk Fq−1
l
)+ (−1)i+j σ ◦ (Fqi Fq−1 )
0≤k≤l 0≤i≤j
= 0,
since, by the foregoing lemma, everything cancels: we simply applied the lemma
in the last but two step and replaced j by k, i − 1 by l in the last but one step,
which shows that the two expressions are identical except for signs.
Remark 6.2.4. Whenever one has to prove a result of the above kind, it is always
helpful to verify the first few cases when σ is a 2-simplex, and a 3-simplex, and
notice the result how the terms cancel with each other. Then one can take up
the general case to prove it carefully.
It follows from the above proposition that S∗ (X) = {Sq (X), ∂q }q≥0 is a
chain complex of free R-modules. This chain complex is called the singular
chain complex of the space X. Its homology {Hq (S∗ (X))}q≥0 is called the
singular homology of X; each of these groups is an R-module. If we take Z for
R, the resulting homology groups are called integral singular homology groups
of the space X.
Next, we are going to prove some basic properties of the singular homol-
ogy. We begin by computing the singular homology of a point-space.
Example 6.2.5 (Dimension Axiom). Let X = {x} be a single-point space. What

is Sq (X) for each q ≥ 0? The answer is obvious, viz., for each q, there is only one
singular q-simplex σ : Δq → X which is just the constant map taking all of Δq
to the single point x. Therefore, Sq (X) = R · σ q , where σ q denotes the constant
map σ : Δq → X. Note that ∂(σ1 ) = σ0 − σ0 = 0, ∂(σ2 ) = σ1 − σ1 + σ1 = σ1 ,
etc. In fact, more generally, we have

σq−1 , if q is even > 0
∂(σq ) =
0, otherwise.
Thus, the singular chain complex, in this case is

I 0 I 0
· · · −→ Rσ3 −→ Rσ2 −→ Rσ1 −→ Rσ0 −→ 0 · · · ,
where 0 denotes the zero map and I denotes the identity map. It follows that
H0 (X) ∼ = R/R = 0. H2 (X) ∼
= R under the map r · σ0 → r. H1 (X) = Z1 /B1 ∼ =
0/0 = 0, . . . Thus, we conclude that

∼ R, if q = 0
Hq (X) =
0, otherwise.
Instead of taking the coefficient module as the ring R, we could have

taken, more generally, any R-module G as the coefficient module. Then what
we proved above can be stated, more generally, as
Proposition 6.2.6. If X is a point-space, then

∼ G, if q = 0
Hq (X; G) =
0, otherwise.
Zero-Dimensional Homology
Next, we compute the zero-dimensional homology module of a space X
with coefficients in R itself. The results are true with coefficients in any general
R-module G, and will also be suppressed unless we state it otherwise. The
zero-dimensional homology is obtained in terms of its path-components. First,
we have
Proposition 6.2.7. If X is path-connected, then H0 (X) ∼

= R.
Proof. Note that, by definition, Z0 (X) = S0 (X). To determine B0 (X), ob-
serve that
if c = ri σi is a 1-chain in X, then each σi is a path in X and
∂(c) = ri (σ(1) − σ(0)) ∈ B0 (X) and it has the property that the sum of the
coefficients of ∂(c) is zero. Note that the converse of this observation is also
k
true. For this, let x0 ∈ X be a fixed point and suppose c = i=1 ri xi is a
k
0-chain which has the property that i=1 ri = 0. Since X is path-connected,
we can find a path, say, σxi , joining x0 to xi for all i = 1, 2, . . . , k. Then

c= ri xi = ri xi − ( ri )x0
k
= i=1 ri (xi − x0 )

= ri ∂(σxi )
k
= ∂( i=1 ri σxi ).

Thus, we find that B0 (X) consists of all those 0-chains c = ri xi in X which
have the property that the sum ri of their coefficients in the ring R is zero.
Now, we define an R-homomorphism : S0 (X) → R called augmentation map,

by putting ( ri xi ) = ri . The map is clearly onto and its kernel is B0 (X).
This says that
Z0 (X) S0 (X) ∼
H0 (X) = = = R.
B0 (X) ker
Proposition 6.2.8. Let {Xi , i ∈ I} be the family of all path-components of a

space X. Then, for all q ≥ 0,
Hq (X) ∼
= ⊕i∈I Hq (Xi ).
Proof. In fact, since Δq is a path-connected space, any singular q-simplex

σ : Δq → X is indeed a singular simplex in Xi for some i ∈ I. This implies that
&
Sq (X) = Sq (Xi )
i∈I
&
Zq (X) = Zq (Xi )
i∈I
&
Bq (X) = Bq (Xi ).
i∈I

Therefore, it follows at once that for all q ≥ 0, Hq (X) ∼
= i∈I Hq (Xi ).
Corollary 6.2.9. For any space X, H0 (X) is the free R-module consisting of as
many copies of R as there are path-components of X.
Proof. Let xi ∈ Xi , i ∈ I, be a fixed point of the path component Xi of X.
Define a singular 0-simplex cxi : Δ0 → X by putting cxi (e0 ) = xi . We claim
that the 0-cycles {xi }, i ∈ I, form a basis for H0 (X). Note that any singular
0-simplex cx , x ∈ X is homologous to cxi for some i. This is because if x lies
in the path component Xi , we have a path ωx : I → X such that ωx (0) = xi
and ωx (1) = x. Then, clearly ∂(ωx ) = cx − cxi is a boundary. It follows that
any singular 0-chain will be homologous to a linear combinationof cxi , i ∈ I.
To see that these are also linearly independent, let us suppose ri cxi is a 0-
cycle,
i.e.,
it is a boundary. Then there exists a singular 1-chain d such that
∂( d) = ri cxi . Note that any singular 1-simplex lies in a path component
of X. Collecting all
1-chains di which are in the path component Xi , we find
that for all i, ∂( di ) = ri cxi . Applying the augumentation map on both
sides we find that 0 = ri .
Remark 6.2.10. It may be mentioned here, just for information, that all of
the foregoing results about zero-dimensional homology modules are valid for
singular homology only. In general, this need not be true for other homologies.
6.3 One-Dimensional Homology and the Fundamental Group
We have already seen in Corollary 6.2.1 that zero-dimensional singular

homology of a space X is closely related to the path-components π0 (X) of
X. Next, we are going to show that the one-dimensional integral singular
homology H1 (X) of a space X is also nicely related to the fundamental group
of X. In fact, the exact relationship for a path-connected space X is that
H1 (X) is isomorphic to the factor group [π1π,π
1
1]
, where π1 = π1 (X, x0 ) is the
fundamental group of the space X at some point x0 ∈ X and [π1 , π1 ] denotes
the commutator subgroup of π1 .
For the time being, therefore, we assume that X is a path-connected space

and we let H1 (X) denote the singular homology of X with integer coefficients.
We fix a point x0 ∈ X, and for each x ∈ X, we let wx be a path joining x0 to x;
there are several such paths and we choose one of them to be wx . Recall that a
path w in X is a continuous map w : I → X, where I = [0, 1], whereas a singular
1-simplex σ is a continuous map from the standard 1-simplex [e0 , e1 ] into X.
Since there is an affine homeomorphism between I and any 1-simplex in Rn , a
path in X is a singular 1-simplex in X and a continuous map from any closed
interval in Rn can be regarded as a path in X via the affine homeomorphism.
e3
e0 e1
e2
We begin with the observation that if w is a loop in X based at any

point, then the corresponding singular 1-simplex is a cycle in the singular chain
complex S∗ (X). Also, note that a constant path w in X mapping the whole
domain to the point x0 ∈ X is not only a cycle, but is a boundary. To see this,
consider a singular 2-simplex σ : Δ2 → X which maps the whole of Δ2 to that
point x0 . Then ∂(σ2 ) is the constant path w. In fact, much more is true, viz., if
w is a path joining x0 to x1 , then the 1-chain w +w−1 is a boundary. To see this
also, just consider the singular 2-simplex σ : Δ2 → X which maps [e1 , e2 ] as w
and is constant on every line of Δ2 parallel to [e0 , e2 ]. Then the restriction of σ
on [e0 , e1 ] gives w−1 . Since restriction of σ on [e0 , e2 ] is constant, we find that
∂(σ) = w − constant + w−1 . Because the constant path is already a boundary,
we conclude that w + w−1 is a boundary. Thus, we have proved the following:
6.3. One-Dimensional Homology and the Fundamental Group 235
Lemma 6.3.1. Any constant path in X is a singular 1-boundary. If w is a path

in X, then the singular 1-chain w + w−1 is also a boundary.
Now, let us define a map from π1 (X, x0 ) to H1 (X) just by sending a loop
w based at x0 to the homology class {w} of the singular 1-cycle w in X. First,
we prove that such a map is well-defined. We have
Lemma 6.3.2. Suppose w ∼ w rel {0, 1}. Then the 1-chain w − w in X is a

boundary.
Proof. Let F : I × I → X be a path-homotopy from w to w . Because F is
constant on the left vertical side of I × I, the above map factors through the
map I ×I → Δ2 , which maps the left vertical side to the vertex e0 of Δ2 . Hence
we get a singular 2-simplex σ : Δ2 → X such that σ = w on [e0 , e1 ], σ = w on
[e0 , e2 ] and σ is the constant path wx0 on [e1 , e2 ]. Then we have
∂(σ) = w + wx0 − w .
Since the constant path wx0 is already a boundary, we find that w − w is a

boundary.
In view of the above lemma, we now have a well-defined map

φ : π1 (X, x0 ) → H1 (X) given by φ[w] = {w}. Next, we prove that φ is a
homomorphism. This follows from the following result.
Proposition 6.3.3. If w, w are two loops based at x0 ∈ X, then the 1-chain

w ∗ w − w − w is a boundary in X, i.e., {w ∗ w } = {w} + {w } in H1 (X).
Proof. We may assume that w is defined on the edge [e0 , e1 ] and w is defined
on [e1 , e2 ]. Now we define a singular 2-simplex σ : Δ2 → X by declaring it to be
constant on lines parallel to the line joining e1 with mid-point of [e0 , e2 ]. Then
σ restricted to [e0 , e2 ] clearly gives the product path w ∗ w . Hence, keeping the
definition of boundary map ∂ in mind, we get
∂(σ) = w + w − (w ∗ w ).
Having shown that φ : π1 (X, x0 ) → H1 (X) is a homomorphism in which

H1 (X) is abelian, we find that φ maps the commutator subgroup [π1 , π1 ] to
zero, and so we get the induced homomorphism
π1
φ̄ : → H1 (X).
[π1 , π1 ]
We next prove that the map φ̄ is indeed an isomorphism. We have
Theorem 6.3.4. (Poincaré-Hurewicz Theorem) Assume that X is a path-

connected space. Then the map φ̄ defined above is an isomorphism.
We have only to show that the map φ is onto and has the kernel equal to
[π1 , π1 ]. We will prove both these results by producing a homomorphism from
H1 (X) to [π1π,π1
1]
which will be the inverse of φ̄. For this, let σ : Δ1 → X be a
singular 1-simplex in X. Putting σ(e0 ) = σ(0) and σ(e1 ) = σ(1), we get a loop
−1
σ̂ = wσ(0) ∗ σ ∗ wσ(1) based at x0 . Recall that wx is a path joining x0 to x and
there are several choices for wx . If μ is another path joining x0 to σ(0), then
−1
wσ(0) ∗ σ ∗ wσ(1) and μσ(0) ∗ σ ∗ μ−1
σ(1) are conjugate. This means the loop σ̂ is not
uniquely defined by σ, but note that its coset in [π1π,π 1
1]
is uniquely determined.
This follows from the fact that two conjugate elements in a group G determine
G
the same elements of [G,G] . We define ψ(σ) = [σ̂], where [σ̂] is a well-defined
π1
element of [π1 ,π1 ] . Since S1 (X) is the free abelian group generated by all such
1-simplexes σ in X, we get a homomorphism
π1
ψ : S1 (X) → .
[π1 , π1 ]
The next result shows that ψ maps the group B1 (X) of 1-boundaries of X into
the trivial element of [π1π,π
1
1]
. We have
Lemma 6.3.5. Im ψ∂ = {1}, the identity element of π1

[π1 ,π1 ] .
Proof. Since B1 (X) is generated by all elements of the type ∂(σ) where σ : Δ2 →
X is a singular 2-simplex, it suffices to prove that for any such σ, ψ∂(σ) is the
identity element. Let us put σ(ei ) = yi ; i = 0, 1, 2 and f = σ (2) , g = σ (0) , h =
(σ (1) )−1 denote the ith face of σ (see Fig.6.2).
y0
xxxxxxxxxxxxxxxxxxxxxx
wy xxxxxxxxxxxxxxxxxxxxxx
0 xxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxx h
fxxxxxxxxxxxxxxxxxxxxxx
x xxxxxxxxxxxxxxxxxxxxxx
0 wy yxxxxxxxxxxxxxxxxxxxxxx
1 1
g
y
2
wy
2
Fig. 6.2: Path f ∗ g ∗ h is null-homotopic

We have
ψ(∂σ) = ψ(σ (0) − σ (1) + σ (2) )

= ψ(f + g − h−1 )
= ψ(f ) · ψ(g) · (ψ(h−1 ))−1
' (
= [fˆ][ĝ][(ĥ−1 )−1 ]
= [fˆ ∗ ĝ ∗ (ĥ−1 )−1 ]
' (
= [wy0 ∗ f ∗ wy−1 1
∗ wy1 ∗ g ∗ wy−1
2
∗ (wy0 ∗ h−1 ∗ wy−1
2
)−1 ]
' (
= [wy0 ∗ f ∗ g ∗ h ∗ wy−1 0
] .
Since the loop f ∗ g ∗ h has an extension to the 2-disk, it is homotopic to the

constant path at y0 , which implies that the loop in bracket is path-homotopic
to the constant loop at x0 .
We need one more notation. The map x → wx is, in fact, a map from
the generators of S0 (X) into S1 (X) and so it can be extended to a ho-
momorphismw : S0 (X) → S1 (X). We use the following notation to write
w( ni xi ) = ni wxi . We have
Proposition 6.3.6. If σ is a singular 1-simplex in X, then φ̄ψ(σ) is represented

by the cycle σ + wσ(0) − wσ(1) = σ − w∂(σ) . If c is a 1-chain, then φ̄ψ(c) =
{c − w∂(c) }, and φ̄ψ(c) = {c} if c is a cycle.
Proof. We have already noted that for any path w, w + w−1 is a boundary, i.e.,
{w} = −{w−1 } in BS1 (X)
1 (X)
. Now, for any singular 1-simplex σ, we have
−1
φ̄ψ(σ) = φ̄[wσ(0) ∗ σ ∗ wσ(1) ]
−1
= {wσ(0) } + {σ} + {wσ(1) }
= {wσ(0) + σ − wσ(1) }
= {σ − w∂(σ) }.
The next part is obvious from the definition whereas the last part follows from
the fact that a constant path is a boundary.
Proof of the Poincaré-Hurewicz Theorem. We have seen that φ̄ : [π1π,π 1

1]
→
H1 (X) defined earlier is a homomorphism. The last lemma clearly defines a
homomorphism ψ̄ : H1 (X) → [π1π,π 1
1]
by putting ψ̄{z} = ψ(z). Moreover, for
each path-class [α] based at x0 , we have
ψ̄ φ̄[[α]] = ψ̄({α})
= [[wx0 ∗ α ∗ wx−1
0
]]
= [[α]]
in π1
[π1 ,π1 ] . Conversely, for each {z} ∈ H1 (X), the last proposition shows that
φ̄ψ̄{z} = φ̄(ψ(z)) = {z}.
This completes the proof that φ̄ is an isomorphism.
The geometric idea of cycles and boundaries in the case of simplicial

homology, as we saw in Chapter 4, is quite apparent. In the case of singular
homology, however, the pictures underlying cycles, boundaries, etc., get highly
complicated and very often even lost. The proof of the main result elaborated
above certainly gives us a glimpse of how 1-cycles and 1-boundaries look like in
singular theory. In this respect, the above relation between the 1-dimensional
singular homology and the fundamental group of a space X is quite revealing.
Sometimes, the Poincaré-Hurewicz isomorphism theorem is expressed by saying
that the 1-dimensional integral singular homology of a space is “its fundamental
group abelianized”.
Example 6.3.7. Since the fundamental groups of several spaces are already
determined (see Chapter 2), using Poincaré-Hurewicz Theorem, we can quickly
tell their one-dimensional singular homology with integer coefficients. As an
illustration of this approach, we get the following results where symbols have
their usual meaning:
H1 (S1 ) ∼
= Z
n
H1 (S ) = 0 for all n ≥ 2
H1 (T ) ∼
= Z⊕Z
H1 (RP n ) ∼
= Z/2Z, n ≥ 2
H1 (L(p, q))) ∼
= Zp .
If X denotes the figure of eight (see Example 5.1.10), then we know that its
fundamental group is the free group on two generators (nonabelian). Hence,
abelianizing this free group, we conclude that H1 (X) ∼
= Z ⊕ Z.
Homomorphism Induced by a Continuous Map
Let f : X → Y be a continuous map. If σ : Δq → X is a singular simplex

in X then, clearly, f ◦ σ : Δq → Y is a singular simplex in Y . Thus, for all
q ≥ 0, we can define a homomorphism f# : Sq (X) → Sq (Y ) simply by putting
f# (σ) = f ◦ σ on generators of Sq (X) and then extending it to Sq (X) by
linearity. Now, we have
Proposition 6.3.8. The sequence f# : Sq (X) → Sq (Y ), q ≥ 0, of homomor-

phisms is a chain map.
Proof. We must prove that for all q ≥ 1, the following squares are commutative:
- Sq (X) ∂ - S (X)
q−1
-
f# f#
? ?
- Sq (Y ) ∂ - S (Y )
q−1
-
However, this is immediate because for a generator σ ∈ Sq (X), we have:

f# ∂(σ) = f# ( (−1)i σ i )

= (−1)i f# (σ ◦ Fqi )

= (−1)i (f ◦ σ) ◦ Fqi
= ∂(f ◦ σ)
= ∂f# (σ).
It follows from the above proposition that a continuous map f : X → Y

induces an R-homomorphism f∗ : Hq (X) → Hq (Y ) for all q ≥ 0, in singular
homology. The actual definition of f∗ is the following: if z ∈ Sq (X) is a q-cycle,
then f∗ {z} = {f# (z)}. With this definition, we can easily verify the following
functorial properties of induced homomorphisms:
Proposition 6.3.9. (Identity Axiom). If I : X → X is the identity map, then

the induced homomorphism I∗ : Hq (X) → Hq (X) is also identity map, for all
q ≥ 0.
Proposition 6.3.10. (Composition Axiom). If f : X → Y, g : Y → Z are con-

tinuous maps then, for all q ≥ 0, the induced homomorphisms in homology
satisfy
(g ◦ f )∗ = g∗ ◦ f∗ : Hq (X) → Hq (Z).
The above two propositions yield the following topological invariance of

singular homology groups:
Corollary 6.3.11. If f : X → Y is a homeomorphism then, for all q ≥

0, f∗ : Hq (X) → Hq (Y ) is an isomorphism.
Proof. Let f −1 denote the inverse of the homeomorphism f . Then, clearly

f −1 ◦ f = IX and f ◦ f −1 = IY . Considering the induced homomorphisms
f∗ : Hq (X) → Hq (Y ) and f∗−1 : Hq (Y ) → Hq (X) in homology, we find from
Proposition 6.3.10 and Proposition 6.3.9, that f∗ is the inverse of f∗−1 . This
means f∗ is an isomorphism.
Remark 6.3.12. The above corollary says that the singular homology modules
are topological invariants. At this point, we must recall what we presented
earlier in Chapter 4 (Theorem 4.9.6), viz., the proof of the topological invariance
of the simplicial homology. That proof was indeed quite long and surprisingly
difficult. Even the proof of the functorial property for simplicial homology,
analogous to the composition axiom for singular homology proved above, took
considerable effort. This is in great contrast with the corresponding results for
the singular homology where they have simple and obvious proofs.
Proposition 6.3.13. Let X, Y be path-connected spaces and f : X → Y be a

continuous map. Then the induced homomorphism f∗ : H0 (X) → H0 (Y ) in
singular homology is an isomorphism. In particular, when X = Y , any map
f : X → X induces the identity map f∗ : H0 (X) → H0 (X).
Proof. To see this, note that the augmentation map : S0 (X) → R defined by

( ri xi ) = ri gives us the isomorphism H0 (X) = SIm0 (X)
∂1 → R. The same
argument for the space Y gives that H0 (Y ) = SIm
0 (Y )
∂1 → R is an isomorphism.
The chain map f# : S0 (X) → S0 (Y ) clearly maps B0 (X) into B0 (Y ) making
the following diagram commutative:

S0 (X) R
f# Id
S0 (Y ) R
This clearly proves that f∗ : H0 (X) → H0 (Y ) is an isomorphism. Forthe case

when X = Y , recall from Proposition 6.2.3 that a singular 0-chain
ri xi in
X is a boundary
if and only if ri = 0. For any 0-chain
c = r x
i i , we find
that f# (c) = ri f (xi ). But the 0-chain c − f# (c) = ri xi − ri f (xi ) has
the property that the sum of their coefficients is zero. This means c − f# (c) is
a boundary, i.e., {c} = {f# (c)} = f∗ {c} in H0 (X). This proves that f∗ is the
identity map.
Proposition 6.3.14. If f : X → Y is a constant map, then the induced homo-

morphism f∗ : Hq (X) → Hq (Y ) in homology is zero, for all q > 0.
Proof. To see this, suppose f (x) = y0 ∈ Y for every x ∈ X. Then the map
c i
f : X → Y factors as X −→ {y0 } −→ Y where c is the obvious constant map
and i is the inclusion. Hence the induced homomorphism f∗ : Hq (X) → Hq (Y )
c∗ i∗
factors as Hq (X) −→ Hq ({y0 }) −→ Hq (Y ), for all q ≥ 0. Since Hq ({y0 }) = 0,
for all q > 0, f∗ = i∗ ◦ c∗ = 0.
Reduced Homology Groups
Let X
= ∅ and let S∗ (X) denote the singular chain complex of the space
X. We consider an abstract chain complex Cq such that C0 = R, the R-module
R, andCq = 0 forall q
= 0. Then there is a chain map : S∗ (X) → C∗ defined
by ( ri xi ) = ri at the zero level and = 0 in positive dimensions. This
chain map is clearly an onto map. The ker , which is a sub chain complex of
S∗ (X), is denoted by S̃∗ (X) and is called the reduced singular chain complex
of X. Since S∗ (X) is a chain complex of free R-modules, the augmentation map
splits, and we find that S0 (X) = S̃0 (X) ⊕ R and Sq (X) = S̃q (X) for all q > 0.
We define the reduced singular homology modules of X, denoted by H̃q (X),
to be the homology of the chain complex S̃∗ (X). Since (Im ∂1 ) = 0, we find
that Im ∂1 ⊆ ker = S̃0 (X). Therefore,
S0 (X) S̃0 (X)

H0 (X) = = ⊕ R = H̃0 (X) ⊕ R
Im ∂1 Im ∂1
and Hq (X) = H̃q (X) for all q > 0. This last result says that the reduced
singular homology modules are completely determined in terms of the usual
singular homology modules of X and vice-versa. The reduced groups are very
useful in computations of homology groups.
What is a basis for the reduced group H̃0 (X) with integer coefficients?
By definition, all the elements of S˜0 (X)
are reduced 0-cycles, and any element
of S˜0 (X)
is a singular 0-chain c = ni xi in X which satisfies the condition
that ( ni xi ) = 0. Here we are identifying a singular 0-simplex Δ0 → {xi }
by the point xi of X.
Let us fix some point x0 ∈ X. Then for any x ∈ X, x − x0 is reduced 0-

cycle, since the sum of their coefficients is zero. If x and x0 both lie in the same
path components of X, then there is a path ω : I → X such that ω(0) = x and
ω(1) = x0 . But this means ∂1 ω = x0 − x is a boundary, and hence it represents
the zero element of H̃0 (X). It is also now clear that if x and x0 lie in different
path components of X, then the reduced 0-cycle x − x0 is not a zero element
of H̃0 (X). Let us choose one point xi (i
= 0) in each path component Xi of X.
Then we claim that the set {xi − x0 } forms
a basis of H̃0 (X). To see this, let
c= ni xi be a reduced 0-cycle. Then ni = 0. Hence

c= n i xi = ni x i − n i x0

= ni (xi − x0 ),
shows that the set {xi − x} generates the group H̃0 (X). Also, this set is linearly
independent, since
ni (xi − x0 ) = 0

in H̃0 (X)
means ni (xi − x0 ) is a boundary, i.e., there is a chain d such
that ∂( d) = ni (xi − x0 ). Since a singular 1-simplex is simply a path and
hence it lies in a single
path component, say Xi of X. Collecting all paths in
Xi , we find that ∂( di ) = (xi − x0 ) for each i. Applying on both sides, we
get 0 = ni . Hence, the set {xi − x} is also linearly independent. Thus, these
elements of H̃0 (X) form a basis for H̃0 (X).
Let f : X → Y be a continuous map between two nonempty spaces and

consider the induced homomorphism f# : S0 (X) → S0 (Y ). Since f# maps
B0 (X) into B0 (Y ), we find that f# induces a homomorphism f˜# : S̃0 (X) →
S̃0 (Y ) which, in turn, induces a homomorphism f∗ : H̃0 (X) → H̃0 (Y ). Now, it
can be easily verified that these induced homomorphisms in reduced singular
homology satisfy the two functorial properties analogous to Propositions 6.3.9
and 6.3.10.
Definition 6.3.15. A space X is said to be acyclic if H̃q (X) = 0 for all q ≥ 0.
A point-space is evidently acyclic. After proving the homotopy axiom for

singular homology, we will see that all contractible spaces are also acyclic. Let
us compute the singular homology of convex subsets of the Euclidean space,
like an n-disc, n-simplex and a cone etc. First we have a little general concept.
Definition 6.3.16. A subset X of some Euclidean space Rn is said to be star-

convex with respect to a point x0 ∈ X if each point x ∈ X can be joined to x0
by a line segment which lies in X.
For example, a disc Dn is star convex with respect to any of its points.
The following subspace of the plane R2 is not convex, but is star convex with
respect to x0 ∈ X.
.x0
Let us have the following
Definition 6.3.17. Let X ⊂ Rn be star convex relative to x0 ∈ X. We define a

bracket operation [ · , x0 ] : Sq (X) → Sq+1 (X) as follows: Let σ be a singular
q-simplex in X. We define [σ, x0 ] ∈ Δq+1 (X) to be the linear map which carries
the line segment joining x ∈ Δq to eq+1 ∈ Δq+1 to the line segment joining

σ(x) to x0 . Then we extend this map linearly, i.e., if c ∈ Sq (X) and c = ri σ i ,
then we put
[c, x0 ] = ri [σi , x0 ].
Note that the map [c, x0 ] when restricted to the face Δq of Δq+1 gives the
chain c itself. Also, notice that if σ is a linear map
σ(e0 , . . . , eq ) → (y0 , . . . , yq )
then [σ, x0 ] is also the linear map taking (e0 , . . . , eq , eq+1 ) to (y0 , . . . , yq , x0 ).
We can now prove the following
Lemma 6.3.18. The map [σ, x0 ] : Δq+1 → X is continuous.

Proof. Clearly, there is the quotient map Δq × I → Δq+1 defined by H(x, t) =
(1 − t)x + t · eq+1 which collapses the top Δ1 × {1} to a single point eq+1 and
is 1-1 otherwise. The continuous map f : Δq × I → X defined by
f (x, t) = (1 − t)σ(x) + tx0
is constant on Δq × {1} and hence induces a continuous map Δq+1 → X,

which is nothing but the map [σ, x0 ].
Now we have the following
Proposition 6.3.19. Suppose X is a star convex with respect to x0 ∈ X and c

is a singular q-chain in X. Then

[∂c, x0 ] + (−1)q+1 c, q>0
∂[c, x0 ] =
(c)σx0 − c, q=0
where σx0 is the singular 0-simplex mapping Δ0 to x0 .

Proof. If σx is a singular 0-simplex, then [σx , x0 ] maps the simplex Δ1 linearly
onto the line segment from σx (Δ0 ) to x0 . Then ∂[σx , x0 ] = σx0 − σx , and the
second part of the formula is verified.
Suppose q > 0. It suffices to check the formula for singular q-simplexes σ.
Let us compute

q+1
∂[σ, x0 ] = (−1)i [σ, x0 ] ◦ F i ,
0
where F : Δq → Δq+1 denotes the linear map

i
(e0 , . . . , eî , . . . , eq+1 ) → (e0 , . . . , 0, . . . , eq+1 ).
The map F q+1 is simply the inclusion map Δq → Δq+1 . Since restriction of
[σ, x0 ] to Δq equals σ itself, the last term in above summation is (−1)q+1 σ.
To complete the proof, let us consider [σ, x0 ]◦F i where i < q +1. The map
i
F carries Δq homeomorphically onto the ith face of Δq+1 , it carries (e0 , . . . , eq )
to (e0 , . . . , ei−1 , ei+1 , . . . , eq+1 ) respectively. Therefore, the restriction of F i to
Δq−1 = (e0 , . . . , eq−1 ) carries this simplex by a linear map onto the simplex
spanned by (e0 , . . . , ei−1 , ei+1 , . . . , eq ) since i < q + 1. Thus

F i Δq−1 = F i (e0 , . . . eî , . . . , eq )
Now we compute [σ, x0 ] ◦ F i : Δq → X. Let x be a general point of Δq−1 .

Since F i : Δq → Δq+1 is a linear map, it carries the line segment from x to
eq linearly onto the line segment from F i (x) to eq+1 . Since F i (x) ∈ Δq , the
map [σ, x0 ] : Δq+1 → X carries the line segment linearly onto the line segment
joining σ(F i (x)) to x0 . Therefore, by definition,

[σ, x0 ] ◦ F i = [σ ◦ ( F i Δq−1 ), x0 ]

Putting the value of F i Δq−1 obtained earlier, we get

q
∂[σ, x0 ] = (−1)i [σ0 F (e0 , . . . , eî , . . . , eq ), x0 ] + (−1)q+1 σ
0
= [∂σ, x0 ] + (−1)q+1 σ.
Theorem 6.3.20. Let X be a star convex subset of Rn with respect to some point
x0 ∈ X. Then X is acyclic in singular homology.
Proof. First we prove that H̃0 (X) = 0. Recall that a singular 0-chain in X
is a 0-cycle in reduced singular homology H̃0 (X) iff the augumentation map
carries it to zero. Let c be a singular 0-chain such that (c) = 0. Then by
previous proposition
∂[c, x0 ] = (c)σx0 − c = −c.
Hence c is a boundary, i.e., H̃0 (X) = 0. To show that Hq (X) = 0 for q > 0, let
z be a singular q-cycle in X. By the previous proposition again
∂[z, x0 ] = [∂z, x0 ] + (−1)q+1 z = (−1)q+1 z,
which is a boundary.
Corollary 6.3.21. Any simplex Δq , q ≥ 0 is acyclic in singular homology. The

product Δq × I is also acyclic.
We will compute the singular homology of n-spheres Sn , n ≥ 0 and of

several other interesting spaces later. In order to compute the singular homology
of various other topological spaces, it will be quite cumbersome if we apply
the definition of singular homology and make the computation. Hence, first we
6.4. Homotopy Axiom for Singular Homology 245
should prove some general results or theorems which could be effectively used to
make the actual computation a pleasant task. It is very similar to computing the
differential coefficients of various differentiable functions in differential calculus.
With this objective in our mind, we will now prove some basic theorems about
singular homology – in fact these theorems will later become basic axioms of
‘a homology theory’ so that we can compute the singular homology using only
these axioms and avoid the definition completely. We call these theorems as
axioms so that their importance is indicated right in the beginning.
Exercises
1. Prove that the n-simplex Δn is acyclic for all n ≥ 0. (Hint: Regard Z as a

chain complex C∗ where C0 = Z and Cq = 0, for all q
= 0. Prove that the
augmentation homomorphism : S∗ (Δn ) → Z is a chain equivalence. To
do so, fix a point v0 ∈ Δn and define a homomorphism τ : Z → S0 (Δn ) by
putting τ (1) = v0 . Then show that ◦ τ = IZ and τ ◦ is chain homotopic
to the identity map IS∗ (Δn ) .)
2. Compute the singular homology with integer coefficients of the Topolo-
gist’s Sine curve.
3. Compute the singular homology groups of the space X if
(a) X is the discrete space having countable number of points.
(b) X is the space of all rational numbers.
(c) X is the space of all irrational numbers.
(d) X is the classical Cantor set.
6.4 Homotopy Axiom for Singular Homology
Here, we first prove a special case (the absolute version) of the homotopy
axiom. The general case (the relative version) will be proved later. Note that
if f : X → Y , g : Z → W are two continuous maps, then the product map
f × g : X × Z → Y × W defined by (f × g)(x, z) = (f (x), g(z)), is continuous.
Theorem 6.4.1. (Homotopy Axiom). Let f, g : X → Y be two maps which are

homotopic. Then for all q ≥ 0, f∗ = g∗ : Hq (X) → Hq (Y ), i.e., the homotopic
maps induce identical maps in singular homology.
Proof. Let F : X × I → Y be a homotopy from f to g so that F (x, 0) = f (x)

and F (x, 1) = g(x) for all x ∈ X. Let i0 , i1 : X → X × I be the bottom and
top inclusion maps. Then, clearly, f = F ◦ i0 and g = F ◦ i1 . Therefore, it is
enough to prove that i0∗ = i1∗ : Hq (X) → Hq (X × I) for all q ≥ 0. Since chain
homotopic maps induce identical maps in homology, we will prove that the two
chain maps i0# , i1# : S(X) → S(X × I) are chain-homotopic: For this, we will
define chain-homotopy simultaneously for all spaces X and by induction on q:
- S1 (X) - S0 (X) - 0

i0# , i1# i0# , i1#
D0
?
+ ?
- S1 (X × I) - S0 (X × I) - 0
To start, let us define D0 : S0 (X) → S1 (X × I) by D0 (σ) = σ × I, where σ is

a singular 0-simplex in X and σ × I denotes the singular 1-simplex in X × I.
Extend D0 to all 0-chains by linearity. Then, for all x ∈ X, we have
∂D0 (x) = ∂(x × I)

= (x, 1) − (x, 0)
= i1 (x) − i0 (x).
This implies ∂D0 = i1# − i0# .

Note that D0 is a natural map, i.e., if h : X → Y is a continuous map,
then the diagram
D0
S0 (X) S1 (X × I)
h (h × I)#
S0 (Y ) S1 (Y × I)
D0
is commutative because for any singular 0-simplex σ in X, we have
(h × I)# ◦ D0 (σ) = (h × I) ◦ (σ × I)
= (h ◦ σ) × I
= D0 (hσ)
= D0 h# (σ).
Having defined D0 for all spaces, we now assume that Dj : Sj (X) →

Sj+1 (X × I) has been defined for all j ≤ q − 1 and for all spaces having the
following two properties:
(a) ∂j+1 Dj + Dj−1 ∂j = i1# − i0# .
(b) For each map h : X → Y, Dj h# = (h × I)# Dj .
Let us observe that in view of (a) for j = q − 1, we have
∂q (i1# − i0# ) = (i1# − i0# )∂q

= (∂q Dq−1 − Dq−2 ∂q−1 )∂q
= ∂q Dq−1 ∂q .
6.4. Homotopy Axiom for Singular Homology 247
This means
∂q (i1# − i0# − Dq−1 ∂q ) = 0
for all spaces X.
To construct Dq for all spaces X, let us assume that δq : Δq → Δq denotes
the identity map. Then taking X = Δq , we find that
(i1# − i0# − Dq−1 ∂q )(δq )
is a q-cycle in Sq (Δq × I). Since the space Δq × I is acyclic, there is a w ∈
Sq+1 (Δq × I) such that
∂q+1 (w) = (i1# − i0# − Dq−1 ∂q )(δq ).
We define Dq (δq ) = w. This yields
∂q+1 Dq (δq ) + Dq−1 ∂q (δq ) = (i1# − i0# )(δq ).
Then, for a singular q-simplex σ in Sq (X), we define
Dq (σ) = (σ × I)# (Dq (δq )),
and extend it by linearity to Sq (X) → Sq+1 (X × I). Now, we will show that
Dq also satisfies both (a) and (b). Condition (b) follows at once because
Dq h# (σ) = Dq (hσ)
= (hσ × I)# Dq (δq )
= (h × I)# (σ × I)# (Dq (δq ))
= (h × I)# Dq (σ).
To verify (a), we compute
(i) ∂q+1 Dq (σ) = ∂q+1 (σ × I)# Dq (δq )
= (σ × I)# ∂q+1 Dq (δq ).
Writing σ = σ# (δq ) and applying ∂q yields
∂q σ = σ# ∂q (δq ).
Therefore,
(ii) Dq−1 ∂q σ = Dq−1 σ# ∂q (δq )
= (σ × I)# Dq−1 ∂q (δq ), by (b).
Combining (i) and (ii) together, we have
(∂q+1 Dq + Dq−1 ∂q )(σ)
= (σ × I)# (∂q+1 Dq + Dq−1 ∂q )(δq )
= (σ × I)# (i1# − i0# )(δq )
= (i1# − i0# )σ# (δq ), because (σ × I)ij = ij σ; j = 0, 1.
= (i1# − i0# )(σ).
This completes the induction and hence the proof.
Corollary 6.4.2. Let f, g : X → Y be two homotopic maps. Then the induced

chain maps f# , g# : S∗ (X) → S∗ (Y ) are chain-homotopic.
Proof. We have already shown that the bottom and top inclusion maps
i0 , i1 : X → X × I induce chain-homotopic maps S∗ (X) → S∗ (X × I).
Since f = F ◦ i0 , g = F ◦ i1 , we find f# = F# ◦ i0# is chain homotopic
to F# ◦ i1# = g# , because composites of chain-homotopic maps are again
chain-homotopic.
The following result is of deeper consequence and establishes the homo-

topy invariance of singular homology.
Corollary 6.4.3. If f : X → Y is a homotopy equivalence, then for all q ≥ 0,

f∗ : Hq (X) → Hq (Y ) is an isomorphism.
Proof. Let g : Y → X be a continuous map such that g ◦ f is homotopic to IX
and f ◦ g is homotopic to IY . Then, by Proposition 6.3.9, Proposition 6.3.10
and Theorem 6.4.1, the induced homomorphisms f∗ : Hq (X) → Hq (Y ) and
g∗ : Hq (Y ) → Hq (X) have the property that g∗ ◦f∗ = (g ◦f )∗ = (IX )∗ = IHq (X)
and f∗ ◦ g∗ = IHq (Y ) , for all q ≥ 0. This proves the desired result.
Remark 6.4.4. The proof of homotopy invariance theorem is indeed a special

case of a very powerful theorem called “acyclic model theorem”. We refer the
interested reader to [18] p. 164–167 for details. Also, the above corollary has
a very interesting consequence: it says that any two spaces having the same
homotopy type will have isomorphic homology groups. In particular, if we find
that two spaces X, Y are such that Hq (X) is not isomorphic to Hq (Y ) for some
q ≥ 0, then the spaces X, Y cannot be of the same homotopy type, and so they
cannot be homeomorphic in the first place. We frequently use this observation
to distinguish between two non-homeomorphic spaces.
6.5 Relative Homology and the Axioms
By a topological pair (X, A) we mean a topological space X along with

a subspace A of X. When A = ∅, we identify the pair (X, ∅) with the space
X and thus the concept of a pair can be considered as generalization of
topological spaces. If (X, A) and (Y, B) are two such pairs, then a continuous
map f : X → Y is called a map of pairs if f (A) ⊆ B; we write such maps as
f : (X, A) → (Y, B). It is clear that the composite of two maps of pairs is again
a map of pairs. It follows that the class of all pairs, with morphisms as maps
of pairs, is a category which includes the category of all topological spaces as
a subcategory.
6.5. Relative Homology and the Axioms 249
We will now define the singular homology of a pair (X, A). Note that
the singular chain complex S∗ (A) of the space A is a subcomplex of the
singular chain complex S∗ (X). Hence we get the quotient chain complex
S∗ (X)/S∗ (A), which we denote by S∗ (X, A). This chain-complex is called the
singular-chain complex of the pair (X, A) and its homology modules, denoted
by Hq (X, A), q ≥ 0, are called the relative singular homology modules of
the pair (X, A). In the case A = φ, the relative homology modules are called
absolute homology modules, distinguishing it from the relative case when
A
= φ.
Let us consider the relative chain complex S∗ (X, A) a bit further. The
, denoted by Zq (X, A), is the
S (X) S (X)
kernel of the homomorphism Sqq (A) → Sq−1 q−1 (A)
Sq+1 (X)
module of all relative q-cycles. The image under the homomorphism Sq+1 (A) →
Bq (X, A),
Sq (X)
Sq (A) ,
denoted by is the module of all relative q-boundaries of the
pair (X, A). Each of these is a submodule of the quotient module Sq (X)/Sq (A)
and so their elements are cosets c + Sq (A), where c ∈ Sq (X). The relative
homology modules Hq (X, A) = Zq (X, A)/Bq (X, A) are, therefore, quotients of
the quotient modules. We make these simpler by defining
Zq (X, A) = {c ∈ Sq (X) | ∂(c) ∈ Sq−1 (A)},
Bq (X, A) = {b ∈ Sq (X) | b − ∂(c ) ∈ Sq (A), for some c ∈ Sq+1 (X)}.

Then it is clear that Zq (X, A) = Zq (X, A)/Sq (A) and Bq (X, A) =
Bq (X, A)/Sq (A). Hence it follows that Hq (X, A) = Zq (X, A)/Bq (X, A) ∼
=
Zq (X, A)/Bq (X, A). With this interpretation, we have a better hold on the
homology classes of the pair (X, A).
Induced Homomorphism for Pairs
Let f : (X, A) → (Y, B) be a map of pairs. The map f : X → Y induces

a chain map f# : S∗ (X) → S∗ (Y ) and its restriction f |A : A → B induces a
chain map ( f |A )# : S∗ (A) → S∗ (B). Hence there is an induced chain map in
their quotients, i.e., we get a chain map f# : S∗ (X, A) → S∗ (Y, B). We are
using the same notation f# for the induced map in the relative as well as the
absolute chain complexes, but this should cause no confusion. The above chain
map clearly induces a homomorphism f∗ : Hq (X, A) → Hq (Y, B) for all q ≥ 0,
in their singular homology modules. Strictly speaking, f∗ should also have the
index q attached to it, but we have suppressed it because homology modules
always indicate the correct dimension. Now, the following functorial properties
of these induced homomorphisms can be easily verified:
Proposition 6.5.1. (i) (Identity Axiom): If I : (X, A) → (X, A) is the identity

map, then the induced homomorphism I∗ : Hq (X, A) → Hq (X, A) is also
the identity map for all q ≥ 0.
(i) (Composition Axiom): If f : (X, A) → (Y, B) and g : (Y, B) → (Z, C) are

maps of pairs then, for all q ≥ 0, the induced homomorphisms in homology
satisfy the following:
(g ◦ f )∗ = g∗ ◦ f∗ : Hq (X, A) → Hq (Z, C).
Next, we want to show that homotopic maps from (X, A) to (Y, B) induce
identical maps in relative homology. To see this, recall that the two continuous
maps f, g : (X, A) → (Y, B) are said to be homotopic if there exists a map
F : (X × I, A × I) → (Y, B) of pairs such that F (x, 0) = f (x), F (x, 1) = g(x)
for all x ∈ X. With this definition, we have
Theorem 6.5.2. (Homotopy Axiom). Let f, g : (X, A) → (Y, B) be homotopic

maps. Then the induced homomorphisms in relative homology are identical,
i.e., for all q ≥ 0,
f∗ = g∗ : Hq (X, A) → Hq (Y, B).
Proof. To prove this theorem we go back to the proof of this result (see Theorem
6.4.1) in the absolute case. Let i0 , i1 : (X, A) → (X × I, A × I) be the bottom
and top inclusion maps and let F be a homotopy from f to g. As in the absolute
case, it evidently suffices to show that these bottom and top inclusion maps
induce chain-homotopic maps S∗ (X, A) → S∗ (X × I, A × I). We have already
proved in the previous section that there is a chain homotopy Dq : Sq (X) →
Sq+1 (X × I), q ≥ 0, between the chain maps induced in the absolute case and
that the same is natural. We now consider the following part of the diagram of
chain complexes:
0 Sq (A) Sq (X) Sq (X, A) 0
Dq Dq D̄q
0 Sq+1 (A × I) Sq+1 (X × I) Sq+1 (X × I, A × I) 0
By the naturality of the chain homotopy {Dq }, the first square is com-
mutative for all q ≥ 0. Therefore, we get induced homomorphisms
D̄q : Sq (X, A) → Sq+1 (X × I, A × I) for all q ≥ 0. This gives us the
chain homotopy between the chain maps i0 # , i1 # induced in the relative
chain complexes S∗ (X, A) → S∗ (X × I, A × I). Since f = F ◦ i0 , g = F ◦ i1 , we
have f∗ = F∗ ◦ i0 ∗ and g∗ = F∗ ◦ i1 ∗ , by the functorial properties of induced
homomorphisms in homology. Now the theorem follows because i0∗ = i1∗ .
Next, let us consider a pair (X, A). Let i : A → X and j : (X, φ) → (X, A)
denote the inclusion maps. These maps induce homomorphisms i∗ : Hq (A) →
Hq (X) and j∗ : Hq (X) → Hq (X, A) for all q ≥ 0. With these in mind, we have
the following important result:
6.5. Relative Homology and the Axioms 251
Theorem 6.5.3. (Exactness Axiom). For any pair (X, A), there is the following
long exact sequence in homology:
∗i j∗ ∗ ∂ i
(∗) · · · −→ Hq (A) −→ Hq (X) −→ Hq (X, A) −→ Hq−1 (A) −→ ··· ,
where ∂ : Hq (X, A) → Hq−1 (A) is the connecting homomorphism and other
homomorphisms are induced by inclusion maps.
Proof. The proof follows from the fundamental theorem of homological algebra
(Theorem 7.2.22). Note that, for the given pair (X, A), the two inclusion maps
i : A → X and j : X → (X, A) give rise to the following exact sequence of
singular chain complexes
i# j#
0 −→ S∗ (A) −→ S∗ (X) −→ S∗ (X, A) −→ 0.
Therefore, by the fundamental theorem mentioned above, we get the long
exact sequence (∗) in homology, where ∂ is the connecting homomorphism.
We also have the following
Theorem 6.5.4. (Commutativity Axiom). The connecting homomorphism

∂ : Hq (X, A) → Hq−1 (A) is natural in the pair (X, A), i.e., if f : (X, A) →
(Y, B) is a map of pairs, then the squares in the second diagram below are
commutative for every q ≥ 0.
Proof. The map f : (X, A) → (Y, B) of pairs clearly gives continuous maps
f |A : A → B and f : X → Y . Hence we get induced homomorphisms in their
singular chain complexes making the following diagram of chain-complexes
commutative in which, by definition, the rows are exact:
i# j#
0 S∗ (A) S∗ (X) S∗ (X, A) 0
( f | A )# f# f#
0 S∗ (B) S∗ (Y ) S∗ (Y, B) 0
i# j#
Now, if we write the two long exact homology sequences for pairs (X, A) and
(Y, B) then, because the connecting homomorphism is natural in short exact
sequences of chain complexes, the following diagram must be commutative for
all q ≥ 0:
j∗ ∂ i∗
Hq (X) Hq (X, A) Hq−1 (A) Hq−1 (X)
f∗ f∗ ( f | A )∗ f∗
Hq (Y ) Hq (Y, B) Hq−1 (B) Hq−1 (Y )

j∗ ∂ i∗
The exact homology sequence of a pair (X, A) yields some immediate

consequences which are worth mentioning. We have
Proposition 6.5.5. For any space X, Hq (X, X) = 0 for all q ≥ 0.

Proof. This follows at once when we write the exact homology sequence for
the pair (X, X) and note that the inclusion map i : X → X, which is identity,
induces the identity homomorphism i∗ : Hq (X) → Hq (X) for all q ≥ 0.
Proposition 6.5.6. If A ⊂ X is a retract of X, then for all q ≥ 0,
Hq (X) ∼
= Hq (A) ⊕ Hq (X, A),
and so, Hq (A) is a direct summand of Hq (X) for all q ≥ 0.

Proof. Let r : X → A be a retraction map. Then r ◦ i : A → A is the identity
map. This means for all q ≥ 0, r∗ ◦ i∗ : Hq (A) → Hq (X) → Hq (A) is also
identity. Now, consider the exact homology sequence of the pair (X, A)
i
- Hq (A) ∗ - Hq (X) j∗ -
Hq (X, A) ∂ - H (A) -
q−1
r∗
Since r∗ ◦ i∗ is identity, i∗ is a 1-1 map, i.e., ker i∗ = 0 = Im ∂. But this says

j∗ : Hq (X) → Hq (X, A) is onto for all q ≥ 0. In other words, for all q ≥ 0, we
have a short exact sequence
i∗ - j∗ -
0 - Hq (A) Hq (X) Hq (X, A) - 0,
r∗
which splits. Therefore, for all q ≥ 0, Hq (X) ∼

= Hq (A) ⊕ Hq (X, A).
Proposition 6.5.7. If A ⊂ X is a deformation retract, then Hq (X, A) = 0 for

all q ≥ 0.
Proof. If A is a deformation retract, then the inclusion map i : A → X is a
homotopy equivalence and the retraction r : X → A is its homotopy inverse.
Hence, by Corollary 6.4.2, i∗ : Hq (A) → Hq (X) is an isomorphism for all q ≥ 0.
Now it follows from the exact sequence of the pair (X, A) that j∗ : Hq (X) →
Hq (X, A) must be a zero map, and so, by exactness, 0 = Im j∗ = ker ∂ =
Hq (X, A).
6.6 The Excision Theorem

Now, we come to the last theorem about singular homology which is, like
the homotopy axiom, a bit difficult to prove. Since the proof involves several
interesting ideas about the singular homology, it is all worthwhile. The final
result, known as the excision theorem for singular homology, will complete our
6.6. The Excision Theorem 253
main objective of showing that the singular homology is a homology theory

in the sense of Eilenberg-Steenrod. The theorem says that if U ⊆ X is such
that Ū ⊆ int A, then the inclusion map ρ : (X − U, A − U ) → (X, A), called
excision, induces isomorphism ρ∗ : Hq (X − U, A − U ) → Hq (X, A) for all q ≥ 0
in homology. To ‘excise’ means to ‘cut’ or ‘remove’ the subset U from the pair
(X, A) making the pair smaller without affecting the homology groups of the
pair (X, A). This theorem is a powerful tool not only in making computations
of relative homology groups of the pair (X, A), but also in other contexts like
homology manifolds, orientation of manifolds, etc.
Before we proceed any further, let us show that the foregoing hypothesis
on the set U , which is to be excised from the pair (X, A), is really needed.
We give an example to show that, in general, an excision map need not induce
isomorphism in singular homology. We have
Example 6.6.1. Let us take the 2-sphere S2 and let A = {(x, y, 0) ∈ S2 | x ≥ 0},
U = P = (1, 0, 0). We will later show that H2 (S2 ) ∼= Z for singular homology
also, though we already know that this is true for the simplicial homology. Note
that A is a part of the great circle of S2 lying in the (x, y)-plane and so it is
homeomorphic to the closed interval [0,1]. Since the sequence
· · · → H2 (S2 − U ) → H2 (S2 − U, A − U ) → H1 (A − U ) → · · ·
for the pair (S2 − U, A − U ) is exact, and since S2 − U is homeomorphic to R2 ,

we find that H2 (S2 − U, A − U ) = 0. On the other hand, the exact sequence
· · · → H2 (A) → H2 (S2 ) → H2 (S2 , A) → H1 (A) → · · ·
shows that H2 (S2 , A) ∼= H2 (S2 ) ∼

= Z. Thus we find that H2 (S2 − U, A − U )
2
is not isomorphic to H2 (S , A), i.e., the excision theorem is not valid in this
case. The reason is also obvious, viz., int A = φ and so the hypothesis that
Ū ⊆ int A is not satisfied.
Singular Chains of Small Sizes

Though the proof of the excision theorem is, in fact, motivated by the
results of simplicial homology, it uses new ideas about singular homology not
discussed so far. One important new result which we will prove says, roughly
speaking, that singular homology of a space X can be computed using only
chains of smaller sizes, rather than all chains. To make this point precise, let U
be an open covering of X. A singular simplex σ : Δq → X is said to be small
of order U if σ(Δq ) ⊂ U for some U ∈ U . Now, observe that if σ is small of
order U, then so is every face σ (i) of σ. A singular chain c in X is said to be
small of order U if every singular simplex occuring in c is small of order U . Let
Sq (X, U ) denote the free abelian group generated by all singular q-simplexes of
X which are small of order U. Then it is now clear that {Sq (X, U)}q≥0 forms a
sub chain complex of the singular chain complex {Sq (X)}q≥0 . The interesting
result that we will prove is that the inclusion chain map S∗ (X, U) ⊂ S∗ (X)
induces isomorphism in homology Hq (X, U ) → Hq (X) for all q ≥ 0. We
also observe that both the chain complexes S∗ (X, U ) and S∗ (X) consist of
free abelian groups, it follows from a result in homological algebra (see E.H.
Spanier [18] Theorem 10 on p. 192) that the inclusion map S∗ (X, U) → S∗ (X)
is a chain homotopy equivalence.
Now, let us consider singular simplexes in a fixed n-simplex

Δn . A
singular q-simplexσ : Δq → Δn is said to be affine if σ( ti pi ) = ti σ(pi ),
where ti ∈ I and ti = 1. Note that if σ is affine then each of its faces σ (i)
is also affine, i = 0, 1, . . . , q. The set of all affine simplexes in Δn generate a
subcomplex S∗ (Δn ) ⊂ S∗ (Δn ).
It is evident that an affine simplex σ in Δn is completely determined by

the images of its vertices: If x0 , x1 , . . . , xq are points in Δn , then let us denote
by (x0 , x1 , . . . xq ) the affine simplex which maps the vertex pi of Δq into the
point xi , 0 ≤ i ≤ q. Then it is clear that

∂(x0 , x1 , . . . , xq ) = (−1)i (x0 , . . . , x̂i , . . . , xq ).
Also, notice that the identity map ξn : Δn → Δn is a linear simplex (p0 , . . . , pn );

this ξn will play a fundamental role.

Now, let bn denote the barycentre of Δn , i.e., bn = 1
n+1 pi . For q ≥ 0,
we define a homomorphism
βn : Sq (Δn ) → Sq+1

(Δn )
by the formula
βn (x0 , . . . , xq ) = (bn , x0 , . . . , xq ).
Then, as we have seen in the case of simplicial homology, βn gives a chain
contraction of the chain complex S (Δn ). An epimorphism : C0 → Z, such
1 ∂
that the composite C1 −→ C0 −→ Z is trivial, is called an augmentation for
a chain complex C. A nonnegative chain complex C with an augmentation is
called an augmented chain complex.
Now, if τ : Z → S0 (Δn ) is defined by τ (1) = (bn ), then one can easily
see by direct computation that βn : IS (Δn ) ∼= τ ◦ is a chain homotopy

equivalence, i.e., ∂βn + βn ∂ = Id on S∗ (Δn ). In particular, ∂βn = Id on all
cycles.
Using the chain contraction βn , we now define an augmentation preserving

subdivision chain map Sd : S(X) → S(X) in singular theory, simultaneously
along with a chain deformation D : S(X) → S(X) from Sd to IS(X) , which
will be natural in X, i.e., for each continuous map f : X → Y , we will have

commutative squares:
Sd D
Sq (X) Sq (X) Sq (X) Sq+1 (X)
f# f# f# f#
Sq (Y ) Sq (Y ) Sq (Y ) Sq+1 (Y )
Sd D
Both will be defined by induction on q-chains, q ≥ 0. If c is a 0-chain, we put

Sd(c) = c and D(c) = 0. Now, let n ≥ 1 and assume that Sd and D both
have been defined for 0 ≤ q < n. Then we define both of these on the singular
n-simplex ξn : Δn → Δn by
Sd(ξn ) = βn (Sd(∂ξn )) (a)
D(ξn ) = βn (Sd(ξn ) − ξn − D∂(ξn )). (b)

Then, for every singular n-simplex σ : Δ → X, let us define
n
Sd(σ) = σ# (Sd(ξn ))
D(σ) = σ# (D(ξn )).
Now we have the following relations:
Proposition 6.6.2. (i) ∂ Sd = Sd ∂, (ii) ∂D = Sd − Id − D∂.

Proof. (i) We have
Sd ∂(σ) = Sd ∂(σ# (ξn ))

= Sd σ# (∂ξn )
= σ# Sd(∂ξn ) (by naturality)
= σ# ∂Sd(ξn ) (by (a) and the fact that Sd(∂ξn ) is a cycle)
= ∂ Sd(σ) (by definition).
(ii) Here, we have
D∂σ = D∂σ# (ξn )

= Dσ# (∂ξn )
= σ# D∂(ξn )
and
∂Dσ = ∂σ# (Dξn )

= σ# ∂D(ξn ),
so that for each σ,
(D∂ + ∂D)(σ) = σ# ((D∂ + ∂D)(ξn ))

= σ# ((Sd − Id )(ξn )) by (b)
= (Sd − Id )(σ).
Note that the above properties of the chain homotopy have been estab-
lished using the already known similar properties for affine simplexes. Yet
another point which must be observed is that image of any σ under the maps
Sd and D lie in σ(Δn ).

Now, if X is a metric space and c = ni σi is a singular q-chain in X,
then we can define
mesh c = sup{diam σi (Δq ) | ni

= 0}.
With this definition one can easily prove the following by induction on q:
Lemma 6.6.3. Let Δn be the Euclidean n-simplex and c be an affine q-chain in

Δn . Then
q
mesh (Sd(c)) ≤ mesh (c).
q+1
We can also define, by induction on m, the subdivision chain maps

Sdm : S(X) → S(X) of higher order m. We put Sd0 = IS(X) and Sdm =
Sd(Sdm−1 ), m ≥ 1. From the remarks made earlier, it also follows that the
subdivision operator Sdm is also chain homotopic to the identity map IS(X) .
Since Δn is a metric space, it follows from the preceding lemma that if c is any
affine q-chain in Δn , then for all m ≥ 1, mesh (Sdm (c)) ≤ ( q+1q m
) mesh (c).
Next, we have
Theorem 6.6.4. Let {z} ∈ Hn (X, A). Then, given any open covering V of X,
we can find an r > 0 such that Sdr (z) is a linear combination of singular
n-simplexes in X which are small of the order of V.
Proof. Let σ be a singular n-simplex occurring in the relative cycle z. It
suffices to show that there exists an r > 0 such that each singular simplex
occurring in Sdr (σ) is contained in V for some V ∈ V. Note that σ : Δn → X
is continuous and Δn is a compact metric space. Therefore, the open cover
{σ −1 (V ) | V ∈ V} will have a Lebesgue number, say, λ. Now, we choose r so
q r
large that ( q+1 ) diam(Δn ) < λ, which means mesh (Sdr (ξn )) < λ, i.e., each
singular simplex occurring in σ# (Sdr (ξn )) = Sdr (σ) will be contained in V
for some V ∈ V.
Now, we prove what we stated in the beginning.

Theorem 6.6.5. Let U be any open covering of X. Then the inclusion map
S(X, U) → S(X) induces isomorphism HqU (X) → Hq (X) for all q ≥ 0.
Proof. To show that the above map is a monomorphism, let {z} ∈ HqU (X)
which goes to zero under the above homomorphism. This means z = ∂(e) for
some e ∈ Sq+1 (X). It suffices to show that z = ∂(e ) for some e ∈ Sq+1 (X, U ).
Note that we can find an integer k such that Sdk (e) ∈ Sq+1 (X, U) and
Sdk (e) − e = Dk ∂(e) + ∂Dk (e)

= Dk (z) + ∂Dk (e),
which means
∂Sdk (e) − ∂(e) = ∂Dk (z)
and, therefore,
z = ∂(Sdk (e) − Dk (z)) ∈ ∂(Sq+1 (X, U )),
proving the result.

To prove the onto part, let z ∈ Sq (X) such that ∂(z) = 0. We find an
integer k such that Sdk (z) ∈ Sq (X, U). Then
Sdk (z) − z = Dk ∂(z) + ∂Dk (z)

= ∂Dk (z).
This means Sdk (z), being a cycle in Sq (X, U), is homologous to z in Sq (X),
i.e., the class {Sdk (z)} goes to the class {z} under the above homomorphism.
Now, we can prove
Theorem 6.6.6. (Excision Theorem). Let (X, A) be any pair and U ⊆ X such
that Ū ⊆ int A. Then the excision map ρ : (X − U, A − U ) → (X, A) induces
an isomorphism
ρ∗ : Hq (X − U, A − U ) → Hq (X, A) for all q ≥ 0.
Proof. Let {z} ∈ Hq (X, A), where z is a q-cycle in X mod A. Note that {X −
Ū , int A} is an open covering of X (see Fig. 6.3).
X
A U
Fig. 6.3: U can be excised

n
Hence we can write z = 1 ni σi , where each σi is small of the order of
the covering {X − Ū , int A}. This means any σi which is not in int A will
lie in X − Ū ⊆ X − U . Dropping all those σi ’s which lie in int A, we find
that z = i=1 ni σi , where m ≤ n, is a q-cycle in X − U mod A − U . Since
m
the inclusion map takes z to itself and {z} = {z } in Hq (X, A), we find that
ρ∗ {z } = {z}. This shows ρ∗ is onto.
To prove that it is one-one, let {z} ∈ Hq (X − U, A − U ) such that z ∼ 0

in Hq (X, A). Thus, z = z + ∂(w), where z is a q-chain in A and w is (q + 1)-
chain in X. Now, we choose r so large that Sdr (w) is a linear combination
of simplexes which are small of order {X − Ū , int A}. Thus, we can write
Sdr (w) = w1 + w2 , where all singular simplexes occurring in w1 map in X − U
and those occurring in w2 map in A. Then we get
Sdr (z) − ∂(w1 ) = Sdr (z ) + ∂(w2 ).
Since the left hand side is a chain in X − U , the right hand side is a chain
in A, we find that each is a chain in A − U . This shows that Sdr (z) ∼ 0 in
(X − U ) mod (A − U ). But this means z ∼ 0 in X − U mod A − U , i.e., ρ∗ is a
1-1 map.
6.7 Homology and Cohomology Theories
Let us now summarize what we have done so far. Our ground ring R
is a commutative principal ideal domain. For every pair (X, A), we defined
a sequence Hq (X, A), q ≥ 0, of R-modules called relative homology mod-
ule of the pair (X, A); the topological space X is identified with the pair
(X, ∅) in which case the relative homology modules Hq (X, ∅) = Hq (X) are
called absolute homology modules of X. Given a map f : (X, A) → (Y, B) of
pairs, we showed that this induces a sequence f∗ : Hq (X, A) → Hq (Y, B) of
R-homomorphisms. Finally, we showed that for each pair (X, A) there exists
a sequence ∂ : Hq (X, A) → Hq−1 (A) of R-homomorphisms, called connecting
homomorphisms. Among other things, we proved that all of the above satisfy
the following properties:
1. If f : (X, A) → (X, A) is the identity map, then the induced homomor-

phism f∗ : Hq (X, A) → Hq (X, A) is also the identity map for all q ≥ 0.
2. If f : (X, A) → (Y, B) and g : (Y, B) → (Z, C) are maps of pairs then, for
all q ≥ 0,
(g ◦ f )∗ = g∗ ◦ f∗ : Hq (X, A) → Hq (Z, C).
3. For any map f : (X, A) → (Y, B) of pairs, the square

6.7. Homology and Cohomology Theories 259
∂(X,A)
Hq (X, A) Hq−1 (A)
f∗ ( f | A )∗
Hq (Y, B) Hq−1 (B)

∂(Y,B)
is commutative for all q ≥ 1. This says that for each pair (X, A), the con-
necting homomorphism ∂(X,A) : Hq (X, A) → Hq−1 (A) is a natural trans-
formation from the functor Hq (X, A) to the functor Hq−1 (A) for all q ≥ 1.
4. For each pair (X, A), there is a long exact sequence of homology modules,
viz.,
i∗ j∗ ∂(X,A)
· · · −→ Hq (A) −→ Hq (X) −→ Hq (X, A) −→ Hq−1 (A) −→ · · · .
5. If f, g : (X, A) → (Y, B) are homotopic maps of pairs then, for all q ≥ 0,
f∗ = g∗ : Hq (X, A) → Hq (Y, B).
6. Given a pair (X, A), let U ⊂ X be such that Ū ⊆ int A. Then the excision
map ρ : (X − U, A − U ) → (X, A) induces isomorphism ρ∗ : Hq (X − U, A −
U ) → Hq (X, A) for all q ≥ 0.
7. If X = {p} is a point-space then Hq (X) = 0 for all q
= 0. The R-module
H0 (X) = G is the coefficient module.
All the results summarized above express the deep fact that the singular
homology, defined for all pairs (X, A) of topological spaces, is a “homology
theory” in the sense of Eilenberg and Steenrod. The seven properties, proved
earlier and stated above, simply say that singular homology satisfies all the
seven axioms of Eilenberg and Steenrod on the category of all topological pairs
with morphisms as maps of these pairs. In fact, Eilenberg and Steenrod defined
a homology theory, more generally, on a suitable admissible category. Accord-
ing to them, a Homology Theory (H, ∂) on an admissible category C of pairs
consists of the following:
• A sequence Hq (X, A), q ∈ Z, of R-modules for each pair (X, A) in cate-
gory C.
• For each morphism f : (X, A) → (Y, B) in C, there is a sequence of R-
homomorphisms f∗ : Hq (X, A) → Hq (Y, B), q ∈ Z.
• For each pair (X, A), there is a natural transformation
∂(X,A) : Hq (X, A) → Hq−1 (A), q ∈ Z.
In addition, these are required to satisfy the seven axioms stated above.
Similarly, a Cohomology Theory (H, δ) on an admissible category C consists of
the following:
• A sequence H q (X, A), q ∈ Z, of R-modules for each pair (X, A) in cate-
gory C.
• For each morphism f : (X, A) → (Y, B) in C, there is a sequence of R-
homomorphisms f ∗ : H q (Y, B) → H q (X, A), q ∈ Z.
• For each pair (X, A), there is a natural transformation δ(X,A) : H q (A) →
H q+1 (X, A), q ∈ Z.
And these are required to satisfy the seven analogous axioms of Eilenberg and
Steenrod (see [6] for details) for a cohomology, viz., they satisfy the identity
axiom, composition axiom, exactness axiom, commutativity axiom, homotopy
axiom, excision axiom and the dimension axiom.
On the basis of these axioms, Eilenberg and Steenrod derived interesting

properties of any homology theory. As a simple illustration of these ideas, we
now compute the homology groups of the n-sphere Sn using only these axioms.
This will mean, in particular, that these are also the singular homology
modules of Sn . We state the results for the coefficient R = Z, the additive
group of integers.
First, let us prove the following version of the excision property for a
homology theory.
Proposition 6.7.1. Let A ⊂ X and V ⊂ U ⊂ A. Suppose V can be excised and

(X − U, A − U ) is a deformation retract of (X − V, A − V ). Then U can also
be excised.
Proof. Since (X − U, A − U ) is a deformation retract of (X − V, A − V ), the
inclusion map i : (X − U, A − U ) → (X − V, A − V ) is a homotopy equivalence.
This means i∗ : Hq (X − U, A − U ) → Hq (X − V, A − V ) is an isomorphism
by the functorial properties and the homotopy axiom of the given homology
theory. Since (X − U, A − U ) → (X, A) is just the composition of the inclusion
maps i and the given excision map (X − V, A − V ) → (X, A), it follows that U
can be excised from the pair (X, A).
Homology of Spheres
Next, we are going to compute the integral homology groups of an n-

sphere, n ≥ 0, using only the axioms of a homology theory.
Proposition 6.7.2. Let En+ and En− denote the northern and southern hemi-
spheres of the n-spheres Sn , n ≥ 1, so that Sn−1 = En+ ∩ En− is their common
boundary. Then the excision map (En+ , Sn−1 ) → (Sn , En− ) induces an isomor-
phism Hq (En+ , Sn−1 ) → Hq (Sn , En− ) for all q ≥ 0.
Proof. Note that if we could excise the open set U = Int(En− ) from the pair
(Sn , En− ), then the result would be obvious (see Fig. 6.4).
En+
Sn
−
En S n−1
V
Fig. 6.4
Since Ū is not contained in Int(En− ), the excision property cannot be
applied directly. Let us define V = {x ∈ Sn | xn+1 < −1/2}. Then V satisfies
the hypothesis of the excision axiom. On the other hand one can readily verify
that (Sn − U, En− − U ) is a deformation retract of (Sn − V, En− − V ) and so, by
the preceding proposition, U can be excised proving the assertion.
Proposition 6.7.3. Let S0 = {1, −1} denote the 0-sphere. Then

0 Z ⊕ Z, q = 0
Hq (S ) =
0, otherwise.
Proof. Let P1 = {1}, P2 = {−1} be two point-spaces. Then S0 = P1 ∪ P2 is
a discrete space in which P1 and P2 both are clopen sets. This means we can
excise the set P1 from the pair (S0 , P1 ) so that the excision k : P2 → (S0 , P1 )
induces an isomorphism k∗ : Hq (P2 ) → Hq (S0 , P1 ), for all q ≥ 0. Now consider
the exact homology sequence for the pair (S0 , P1 ):
i 1∗ j∗
· · · −→ Hq (P1 ) −→ Hq (S0 ) −→ Hq (S0 , P1 ) → Hq−1 (P1 ) −→ · · · ,
where i1 : P1 → S0 , i2 : P2 → S0 , j : S0 → (S0 , P1 ) denote the inclusion maps.
Note that k = j ◦ i2 . Since k∗ is an isomorphism, (j ◦ i2 )∗ k∗−1 is identity on
Hq (S0 , P1 ), i.e., the sequence
i1∗ j∗
Hq (P1 ) −→ Hq (S0 ) −→ Hq (S0 , P1 )
splits for all q ≥ 0. This means Hq (S0 ) ∼
= Hq (P1 ) ⊕ Hq (P2 ) for all q ≥ 0. By
dimension axiom, we know that for i = 1, 2

Z, if q = 0
Hq (Pi ) =
0, otherwise
and that proves the proposition.
Remark 6.7.4. If X = {a1 , a2 , . . . , an } is a discrete space having n-points, then

it follows from the above result that
n
1 Z, q = 0
Hq (X) =
0, otherwise.
Remark 6.7.5. The above consideration also yields the result that if a space
X = X1 ∪ X2 is the disjoint union of two clopen sets then, for all q ≥ 0,
Hq (X) ∼
= Hq (X1 ) ⊕ Hq (X2 )
for any homology theory (Hq , ∂).
Definition 6.7.6. If X
= ∅ then we can define the reduced homology groups
H̃q (X) for any homology theory as follows: Let : X → P be the constant map,
where P is a point-space. Then, if i : P → X is any map, we find that ◦ i = IP
so that the induced homomorphism ∗ : Hq (X) → Hq (P ) is onto. We define
H̃q (X) = ker ∗ and call it the reduced homology group of the space X.
It follows from the above definition that the following sequence is exact:
0 → H̃0 (X) → H0 (X) → H0 (P ) → 0.
Since the group H0 (P ) ∼ = Z is just the coefficient group of the homology
theory, we find that H0 (X) ∼ = H̃0 (X) ⊕ Z. We also have H̃q (X) = Hq (X) for
all q
= 0. Moreover, if A
= ∅, we define H̃q (X, A) = Hq (X, A) for all q ≥ 0.
Now, it can be verified that the reduced homology groups of a nonempty pair
(X, A) also satisfies the exactness axiom. This last fact is of frequent occurrence.
From the above remarks, we have
Theorem 6.7.7. If X is a nonempty contractible space then H̃q (X) = 0 for all
q ≥ 0.
Now we have:
Theorem 6.7.8. The reduced homology group of the n-sphere Sn , n ≥ 0, are

given by
Z, if q = n
H̃q (Sn ) =
0, otherwise.
Proof. We observe that the theorem is true for n = 0 by Proposition 6.7.3.
Also, we note that each of the hemispheres En+ and En− defined earlier is home-
omorphic to the n-disk Dn , which is contractible. Thus, by the homotopy and
the dimension axiom, H̃q (En+ ) = 0 = H̃q (En− ) for all q ≥ 0. Writing out the
exact homology sequence for the pair (Sn , En− ), we have
· · · → H̃q (En− ) → H̃q (Sn ) → H̃q (Sn , En− ) → H̃q−1 (En− ) → · · · .
From this exact sequence, we conclude that H̃q (Sn ) ∼

= H̃q (Sn , En− ). However, by
the excision axiom, we have already proved that H̃q (Sn , En− ) ∼ = H̃q (En+ , Sn−1 )
for all q ≥ 0. Then writing the following exact sequence
· · · → H̃q (En+ ) → H̃q (En+ , Sn−1 ) → H̃q−1 (Sn−1 ) → H̃q (En+ ) → · · ·
for the pair (En+ , Sn−1 ), we find that H̃q (En+ , Sn−1 ) ∼
= H̃q−1 (Sn−1 ). Combining
the two results, we see that, for all q ≥ 0, H̃q (Sn ) ∼
= H̃q−1 (Sn−1 ). Taking q = n,
we find that
H̃n (Sn ) ∼
= H̃n−1 (Sn−1 ) ∼
= ··· ∼
= H̃0 (S0 ) = Z,
and taking q
= n, we find that
H̃q (Sn ) ∼
= H̃n−q (S0 ) = 0.
Corollary 6.7.9. For any homology theory (Hq , ∂),

n ∼ Z, if q = 0 or n
Hq (S ) =
0, otherwise.
Exercises
1. Use the axioms of the homology theory to show that the relative homology
of the pair (Rn , Sn−1 ), n ≥ 2, is given by

Z, q = n
Hq (Rn , Sn−1 ) =
0, otherwise.
2. If (X, A) is a pair such that both X and A are path-connected, then

show that H0 (X, A) = 0, and mention an example where under the same
conditions, Hq (X, A) need not vanish in higher dimensions.
3. Give examples to show that, in general, for a pair (X, A)
(i) Hq (X, A)
= Hq (X − A)
(ii) Hq (X, A)
= Hq (X/A), where X/A is the quotient space of X ob-
tained by collapsing A to a point.
4. If i : A ⊂ X is such that for all q ≥ 0 the induced map i∗ : Hq (A) → Hq (X)
is an isomorphism, then compute the homology groups Hq (X, A), q ≥ 0.
5. Let X be a topological n-manifold, n ≥ 1. For each point x ∈ X, show
that
Z, if q = n
Hq (X, X − x) =
0, otherwise.
6.8 Singular Homology with Coefficients
Let R be a PID and G be an R-module. We are now going to discuss the

singular homology with coefficients in the module G. The two special cases of
this coefficient module G are: (i) when R is the ring Z of integers so that G is
simply an abelian group, and (ii) when R is a field so that G is a vector space
over R. The first case practically covers everything whatever is true for an
R-module G with R a principal ideal domain. In fact, this is one reason why
we always tend to call “homology modules” simply as “homology groups”.
This happens because of the well-known algebraic result which says that a
submodule of a free R-module is free provided R is a principal ideal domain.
On the other hand, the case of a field gives the result in the simplest possible
form. We consider the case of a general R-module G so that both the cases,
mentioned above, are covered.
For a pair (X, A), recall that there is the following short exact sequence
of singular chain complexes of free R-modules, where chain maps are induced
by the inclusion maps i : A → X, j : X → (X, A):
i# j#
0 −→ S∗ (A) −→ S∗ (X) −→ S∗ (X, A) −→ 0.
If we take the tensor product of the above exact sequence by the R-module G
(on left or right, it makes no difference since tensor product is commutative),
then since the chain complexes are free, we get the new short sequence of chain
complexes and chain maps
i# ⊗1 j# ⊗1
0 −→ S∗ (A) ⊗ G −→ S∗ (X) ⊗ G −→ S∗ (X, A) ⊗ G −→ 0 (∗)
which remains exact. The chain complex S∗ (X, A)⊗G, denoted by S∗ (X, A; G),
is called the singular chain complex of the pair (X, A) with coefficients in the
R-module G. Its homology modules Hq (S∗ (X, A; G)), denoted by Hq (X, A; G),
q ≥ 0, are called the singular homology modules of (X, A) with coefficients
in G. The case, when G is the R-module R, is the original case of singular
homology modules Hq (X, A) of the pair (X, A). The special case, when R = Z
and the group G ∼ = Z/2Z, is known as the (mod 2) homology of the pair
(X, A) and it has been historically very important.
It may be remarked that the homology with coefficients is really a concept

derived from the integral homology. To see this, note that if S∗ (X) denotes the
integral singular chain complex of X then the elements of the group Sq (X)
are integral linear combinations like k1 f1 + · · · + kn fn of singular q-simplexes
f1 , . . . , fn in X. We already know that an abelian group G is isomorphic to
the group Z ⊗ G under the inverse of the map k ⊗ a → k.a, a ∈ G. Hence any
element of G ⊗ Sq (X) will be of the type a1 ⊗ (k1 f1 ) + a2 ⊗ (k2 f2 ) + · · · + an ⊗
(kn fn ) = b1 f1 + b2 f2 + · · · + bn fn , where bi = ki ai ∈ G, i = 1, . . . , n. Observe
6.8. Singular Homology with Coefficients 265
that the result of taking tensor product could become different, e.g., ki fi may
be nonzero while (ai .ki )fi could very well be zero, depending on the abelian
group G. Furthermore, if G is an R-module, then G is an abelian group with
a ring R of endomorphisms of G acting on G. Each of these endomorphisms
induces an abelian group endomorphism of G ⊗ Sq (X), and so the R-module
S∗ (X, G) is simply the abelian group G ⊗ S∗ (X) with R acting on it as a ring of
endomorphisms. Thus, a typical element of the R-module G ⊗ S∗ (X) is again
of the type c1 f1 + · · · + cn fn , where ci ∈ G for all i.
Bockstein Homomorphism
Homology with coefficients has several interesting properties. First of all, let us
note that if
f g
0 −→ G −→ G −→ G −→ 0
is an exact sequence of R-modules then, since S∗ (X, A) is a free chain complex,
the following sequence of chain complexes with different coefficients remains
exact:
1⊗f 1⊗g
0 −→ S∗ (X, A) ⊗ G −→ S∗ (X, A) ⊗ G −→ S∗ (X, A) ⊗ G −→ 0.
Hence, by the fundamental theorem of homological algebra, we get the

following long exact sequence connecting the singular homology with differ-
ent coefficients:
f∗ g∗ ∂q
→ Hq (X, A; G ) → Hq (X, A; G) → Hq (X, A; G ) → Hq−1 (X, A; G ) →
The connecting homomorphism ∂q : Hq (X, A; G ) → Hq−1 (X, A; G ), occur-

ring in the above homology exact sequence, is quite important and it is known
as the Bockstein homology operator. The special case of Bockstein operator,
arising from the exact sequence 0 → Z → Z → Zp → 0, where p is a prime,
has found an interesting interpretation and several applications so much so
that a symbol βq : Hq (X, Zp ) → Hq−1 (X, Z) has been almost reserved for it.
We may recall that homology groups (with integer coefficients) are the
simplest tools for distinguishing between two non-homeomorphic spaces. In
this context, we will see later that by varying the coefficients, we will not be
able to distinguish between those spaces which are not already distinguished
by their integral homologies and yet the homology with coefficients is very
interesting – the main reason for this is the fact that very often homology with
particular coefficients arises very naturally. Properties of tensor products are
crucial to the study of homology with coefficients. We have
Proposition 6.8.1. On the category of all topological pairs, the singular

homology with coefficients in a fixed R-module G is a homology theory in the
sense of Eilenberg-Steenrod.
Proof. The proof consists of easy verifications. Let f : (X, A) → (Y, B) be a map
of pairs. Suppose f# : S∗ (X, A) → S∗ (Y, B) denote the induced chain map in
singular chain complexes. Then the tensor product f# ⊗ 1 : S∗ (X, A) ⊗ G →
S∗ (Y, B) ⊗ G, where 1 : G → G is the identity map on the R-module G, induces
homomorphism f∗ : Hq (X, A; G) → Hq (Y, B; G) for all q ≥ 0. The verifica-
tion of the identity axiom and the composition axiom now is plain from the
properties of tensor product. The exactness axiom follows from the short exact
sequence (∗) defining the singular homology of the pair (X, A) with coefficients
in G. The commutativity axiom follows from the following commutative dia-
gram of chain complexes and chain maps in which the rows are exact:
i# j#
0 S∗ (A) ⊗ G S∗ (X) ⊗ G S∗ (X, A) ⊗ G 0
( f | A )# ⊗ 1 f# ⊗ 1 f# ⊗ 1
0 S∗ (B) ⊗ G S∗ (Y ) ⊗ G S∗ (Y, B) ⊗ G 0
i# j#
To prove the homotopy axiom, let f, g : (X, A) → (Y, B) be homotopic

maps of pairs. These induce chain maps f# , g# : S∗ (X, A) → S∗ (Y, B)
which are chain homotopic by the homotopy axiom of singular theory. Tak-
ing the tensor product by the R-module G, we find that the chain maps
f# ⊗ 1, g# ⊗ 1 : S∗ (X, A) ⊗ G → S∗ (Y, B) ⊗ G are also chain homotopic,
by Proposition 7.2.21. Therefore, the induced homomorphism in homology
are equal, i.e., f∗ = g∗ : Hq (X, Y ; G) → Hq (Y, B; G) for all q ≥ 0. Fi-
nally, to prove the excision axiom, let U ⊆ A such that Ū ⊂ int A. The
excision map ρ : (X − U, A − U ) → (X, A) evidently induces chain map
ρ# : S∗ (X − U, A − U ) → S∗ (X, A). We know that the induced homomorphism
ρ∗ : Hq (X − U, A − U ) → Hq (X, A) in singular homology is isomorphism for
all q ≥ 0, by the excision theorem for singular homology. Since singular chain
complexes S∗ (X −U, A−U ) and S∗ (X, A) are free and ρ# induces isomorphism
in homology, it follows from Exercise 4, Section 7.2.6 that ρ# is indeed a chain
equivalence. Taking tensor product with G, we find that
ρ# ⊗ 1 : S∗ (X − U, A − U ) ⊗ G → S∗ (X, A) ⊗ G
is also a chain equivalence. Hence the induced homomorphism ρ∗ :

Hq (X − U, A − U ; G) → Hq (X, A; G) in homology is an isomorphism
for all q ≥ 0. Since the dimension axiom is evidently true, this completes the
proof.
In Chapter 4, we defined simplicial homology of a polyhedral pair. Now,

we prove the following theorem.
Theorem 6.8.2. Simplicial homology, defined on the category of all polyhedral

pairs, is a homology theory in the sense of Eilenberg-Steenrod.
Proof. Recall that a topological pair (X, A) is a polyhedral pair if there exists
a pair (K, L) of polyhedra and a homeomorphism h : (|K|, |L|) → (X, A) of
topological pairs giving a triangulation of (X, A). We define the simplicial ho-
mology Hq (X, A) of the pair (X, A) by putting Hq (X, A) := Hq (C∗ (K, L)) for
all q ≥ 0; here, C∗ (K, L) is the quotient chain complex C∗ (K)/C∗ (L) defined
by the following exact sequence of simplicial chain-complexes:
0 → C∗ (L) → C∗ (K) → C∗ (K, L) → 0.
We have already proved in Chapter 4 that the groups Hq (X, A) defined
above are independent of the choice of a triangulation (K, L) of the pair
(X, A). Now, let f : (X, A) → (Y, B) be a map of polyhedral pairs. Suppose
h : (|K|, |L|) → (X, A) and k : (|K1 |, |L1 |) → (Y, B) are triangulations of (X, A)
and (Y, B). Let s : (K (m) , L(m) ) → (|K1 |, |L1 |) be a simplicial approximation
of the map k −1 f h : (|K|, |L|) → (|K1 |, |L1 |). Then we define the induced ho-
momorphism f∗ : Hq (X, A) → Hq (Y, B) for all q ≥ 0 to be the composite ho-
momorphism (k −1 f h)∗ ◦ μ : Hq (K, L) → Hq (K (m) , L(m) ) → Hq (K1 , L1 ), where
μ : Hq (K, L) → Hq (K (m) , L(m) ) is the subdivision homomorphism. Following
Proposition 4.10.6, one can prove that these induced homomorphisms in homol-
ogy satisfy the functorial properties. For any pair (K, L) of simplicial complexes,
the exact sequence
0 → C∗ (L) → C∗ (K) → C∗ (K, L) → 0
of simplicial chain-complexes yields the exactness axiom for a polyhedral pair
(X, A) ≈ (|K|, |L|). The commutativity axiom follows easily from the Simplicial
Approximation Theorem and the results about chain complexes of R-modules.
The homotopy axiom follows from Section 4.10.10. For excision, let (X, A) be
a polyhedral pair and U ⊂ A such that Ū ⊂ int A and that the inclusion map
(X − U, A − U ) → (X, A) is a map of polyhedral pairs. Then we must show that
this inclusion induces isomorphism in simplicial homology. Then we must show
that this inclusion induces isomorphism in simplicial homology. Let (K1 , L1 )
be a triangulation of the pair (X − U, A − U ) and (K, L) be a triangulation
of (X, A). There is a subtle point involved here. The simplicial pair (K1 , L1 )
need not be a subcomplex of the pair (K, L). First we assume that this is the
case. Then consider the chain map C∗ (K1 ) → C∗ (K) → C∗ (K)/C∗ (L) which
is the composite of the inclusion and the quotient maps. This is clearly an
onto map since U ⊂ |L|. Its kernel is obviously C∗ (L1 ). Therefore we have the
following isomorphism of chain complexes: C∗ (K)/C∗ (L) ∼ = C∗ (K1 )/C∗ (L1 ).
This implies that
C∗ (|K| − U, |L| − U ) = C∗ (K1 , L1 ) ∼
= C∗ (K, L) = C∗ (|K|, |L|).
Since chain maps commute with the boundary maps, we find that the excision
map induces isomorphism in homology Hq (X −U, A−U ) → Hq (X, A), ∀q ≥ 0.
In case the simplicial pair (K1 , L1 ) is not a subpair of (K, L), we refer to [16]
for details.
Remark 6.8.3. Note that when we excise the subset M from the simplicial
pair (K, L), we really excise, geometrically speaking, not only |M | but also the
open cones over simplexes of M whose vertices may be in L. Thus, in terms of
geometric carriers, we may be excising more than what is permitted in singular
homology - this is an advantage of simplicial homology over singular homology.
For example, from the double cone (S2 , D2− ) over the boundary of a triangle,
when we excise the lower vertex v4 (see Fig. 6.5) from the pair (S2 , D2 ), we
really excise the open lower hemisphere, viz., Int(D2− ) to get the pair (D2+ , S1 ).
v0
v v2
1
v3
v4
Fig. 6.5: The bottom vertex to be excised
Uniqueness of Homology Theory

As mentioned in the introduction, Eilenberg-Steenrod also proved that
any two homology theories, which are isomorphic for point-spaces, are actually
isomorphic on all compact polyhedra (see Spanier [18], p. 203). By now we have
shown that on the category of compact polyhedra, both simplicial homology
as well as the singular homology are homology theories. Therefore, by the
Eilenberg and Steenrod Uniqueness Theorem, we find the nice result that if
X is a compact polyhedron, then singular homology of X is the same as its
simplicial homology. For instance, if K is a triangulation of Sphere, Torus,
Klein bottle or Möbius band, then simplicial homology of |K|, viz., Hq (K) is
isomorphic to the singular homology of the space |K|. We will use this result
from now onwards without mentioning it explicitly.
Universal Coefficient Theorem for Singular Homology

We continue our assumption that R is a PID and G is an R-module.
The Universal Coefficient Theorem for singular homology gives an exact
relationship between the homology modules Hq (X, A) and the homology
modules Hq (X, A; G) with coefficients in G. The exact relationship, given
later, is determined in terms of the tensor product and torsion product of G
and the modules Hq (X, A).
For a pair (X, A), we consider the singular chain complex S∗ (X, A) defined
earlier. This is a chain complex of free R-modules. Let Zq (X, A) and Bq (X, A)
denote the submodules of Sq (X, A) consisting of relative q-cycles and relative
q-boundaries. Now, note that if z ∈ Zq (X, A) is a q-cycle then for any g ∈ G,
z ⊗ g is a q-cycle in Sq (X, A) ⊗ G. Similarly, if b ∈ Bq (X, A) is a boundary

then b ⊗ g ∈ Sq (X, A) ⊗ G is also a boundary. Now, we can define a map
f : Hq (X, A) × G → Hq (X, A; G) by putting f ({z}, g) = {z ⊗ g}. It is
quickly seen that this map is an R-bilinear map and hence it defines an R-
homomorphism μ : Hq (X, A) ⊗ G → Hq (X, A; G) by the formula μ(z ⊗ g) =
{z ⊗ g} on generators. Using this homomorphism, one can prove the following
Universal Coefficient Theorem for singular homology (see Spanier [18] for a
proof) which says that the above homomorphism μ is a monomorphism with
cokernel Tor(Hq−1 (X, A); G):
Theorem 6.8.4. For a topological pair (X, A) and for any R-module G, there is
the following split exact sequence of R-modules for all q ≥ 0:
μ
0 −→ Hq (X, A) ⊗ G −→ Hq (X, A; G) −→ T or(Hq−1 (X, A); G) −→ 0.
The exact sequence is natural in (X, A) and in the coefficient module G, but
the splitting is not natural.
The splitting of the above sequence implies that Hq (X, A; G) is the direct
sum of Hq (X, A) ⊗ G and Tor(Hq−1 (X, A); G). Note that the naturality of the
sequence in (X, A) implies, in particular, that if f : (X, A) → (Y, B) is a map
which induces isomorphism in integral singular homology, then it induces iso-
morphism in singular homology with coefficients in an arbitrary abelian group
G. This follows from the Five Lemma because, by naturality, the following
diagram is commutative:
0 Hq (X, A) ⊗ G Hq (X, A; G) Tor(Hq−1 (X, A); G) 0

∼
= f∗ ⊗ 1 f∗ ∼
= Tor(f∗ , 1)
0 Hq (Y, B) ⊗ G Hq (Y, B; G) Tor(Hq−1 (Y, B); G) 0
Thus, if two spaces cannot be distinguished by integral homology groups

then they cannot be distinguished by homology with any coefficients.
Example 6.8.5. We already know the integral singular homology groups of the
projective plane P2 as follows: H0 (P2 ) ∼
= Z, H1 (P2 ) ∼
= Z2 and Hq (P2 ) = 0 for all
other values of q. Hence, by the Universal Coefficient Theorem, the nontrivial
(mod 2) homology of P2 is given by:
H0 (P2 ; Z2 ) ∼
= Z ⊗ Z2 ∼
= Z2 ,
H1 (P2 ; Z2 ) ∼
= Z2 ⊗ Z2 ⊕ Tor(Z, Z2 ) ∼
= Z2 ,
H2 (P2 ; Z2 ) = 0 ⊕ Tor(Z2 , Z2 ) ∼
= Z2 .
Example 6.8.6. If K is the Klein bottle then we know the integral singular
homology as H0 (K) ∼ = Z, H1 (K) ∼= Z ⊕ Z2 and Hq (K) = 0 for all other values
of q. Hence, by the Universal Coefficient Theorem, the (mod 2) homology is
given by H0 (K; Z2 ) ∼
= Z2 , H1 (K; Z2 ) ∼
= (Z ⊕ Z2 ) ⊗ Z2 ∼
= Z2 ⊕ Z2 , H2 (K; Z2 ) =
Tor(Z ⊕ Z2 , Z2 ) ∼ Z
= 2 .
Note that (mod 3) homology of the projective plane P2 is given by

H0 (P2 ; Z3 ) ∼
= Z3 , Hq (P2 ; Z3 ) = 0, q
= 0, i.e., the projective plane is like a
2-disk with respect to the ( mod 3) homology. Similarly, the Klein bottle K has
the same homology as torus with respect to the Z2 -coefficients. Thus, though
the integral homology distinguishes the Klein bottle from torus, the two spaces
cannot be distinguished by the (mod 2) homology.
Example 6.8.7. Let L(p, q) denote the lens space, where (p, q) = 1. Then the
integral homology is given by H0 (L(p, q)) ∼= Z, H1 (L(p, q)) ∼ = Zp , H3 (L(p, q)) =
Z, Hi (L(p, q)) = 0 for other values of i. Hence, by the Universal Coefficient
Theorem, H0 (L(p, q); Zq ) ∼
= Zq , H1 (L(p, q); Zq ) = 0 and H3 (L(p, q); Zq ) ∼ = Zq
whereas H0 (L(p, q); Zp ) ∼
= pZ , H 1 (L(p, q); Z )
p =
∼ Z p , H 3 (L(p, q); Z ) ∼
p = Zp .
Thus, with coefficients in Zq , L(p, q) has the same homology as 3-sphere, but
with respect to Zp -coefficients, the two are different.
Künneth Formula for Singular Homology

We are now going to state a result which not only generalizes the
Universal Coefficient Theorem for singular homology, but also gives the
singular homology of the product of two spaces X and Y in terms of the
singular homologies of the spaces X and Y . For convenience and simplicity, we
state the result for integral coefficients but the theorem is true for arbitrary
coefficients over a PID. The special case, when the ground ring is a field, must
always be noted.
Let S∗ (X) and S∗ (Y ) denote the integral singular chain complexes of the
spaces X and Y, respectively. First of all, let us state the important theorem
due to Eilenberg-Zilber (see Spanier [18] p. 232), which asserts that the singular
chain complex S∗ (X ×Y ) of the product space X ×Y is indeed chain equivalent
to the tensor product S∗ (X) ⊗ S∗ (Y ) of the singular chain complexes of X and
Y . In fact, this result of Eilenberg-Zilber shows how the geometric situation
of the cartesian product has a nice algebraic representation in terms of the
tensor product of singular chain complexes of factor spaces. As a result, we
immediately see that for all q ≥ 0, Hq (X ×Y ) ∼= Hq (S∗ (X)⊗S∗ (Y )). Thus, the
problem of computing the homology groups of the product space X ×Y is really
the problem of computing the homology of the tensor product of two free chain
complexes. However, we have discussed this question for two abstract chain
complexes in Appendix, and the same can be applied to obtain the following
basic result:
Theorem 6.8.8. (Künneth). Let X, Y be any two spaces. Then, for all n ≥ 0,
there is the following split short exact sequence of abelian groups:
μ
0→ Hp (X) ⊗ Hq (Y ) → Hn (X × Y ) →
p+q=n

Tor(Hp (X), Hq (Y )) → 0.
p+q=n−1
The above sequence is natural in X, Y but the splitting is not natural.
The monomorphism μ above is called the homology cross product. Let us

remark that if G is an abelian group, we can define an abstract chain complex
C∗ as follows: We put Cq = 0 for all q
= 0 and C0 = G with obvious homo-
morphisms. Since the Künneth theorem is true for abstract chain complexes,
we can replace S∗ (Y ) by the above chain complex C∗ . Then the tensor product
S∗ (X) ⊗ C∗ is just the chain complex S∗ (X) ⊗ G. Since Hq (C∗ ) = G for q = 0,
and zero otherwise, the above Künneth Theorem reduces to the following exact
sequence for all n ≥ 0:
0 → Hn (X) ⊗ G → Hn (X, G) → T or(Hn−1 (X), G) → 0.
This is exactly the Universal Coefficient Theorem for singular homology.
If we take the singular homologies of X and Y with coefficients in abelian

groups G and G , respectively, we find the following theorem:
Theorem 6.8.9. (Künneth). For any two spaces X and Y and for any two
abelian groups G and G such that Tor(G, G ) = 0, there is the following split
exact sequence, for all n ≥ 0, of singular homology groups:

0→ Hp (X, G) ⊗ Hq (Y, G ) → Hn (X × Y, G ⊗ G )
p+q=n

→ Tor(Hp (X, G), Hq (Y, G )) → 0.
p+q=n−1
Moreover, the sequence is natural in spaces X, Y as well as in the coefficients
G and G , but the splitting is not natural.
Example 6.8.10. The integral singular homology of the n-sphere Sn is given by:
H0 (Sn ) ∼
=Z∼= Hn (Sn ), and Hq (Sn ) = 0 for other values of q. Hence, by the
Künneth Theorem, we get for m
= n,

Z, q = 0, m, n or m + n
Hq (S × S ) =
m n
0, otherwise.
But, for m = n, we have
⎧
⎨ Z, q = 0, 2n
Hq (Sn × Sn ) = Z ⊕ Z, q=n
⎩
0, otherwise.
Note that P2 is a two-dimensional polyhedron which means P2 × P2 is 4-

dimensional polyhedron. As mentioned below, the integral homology of P2 × P2
in dimensions higher than three vanishes but, as can be seen easily, the (mod
2) homology of P2 × P2 does not vanish up to dimension 4.
Example 6.8.11. Let P2 denote the projective plane. Then, by the

Künneth Formula, we have
H0 (P2 × P2 ) ∼
= Z,
H1 (P2 × P2 ) ∼
= Z2 ⊕ Z 2 ,
H2 (P2 × P2 ) ∼
= Z2 ∼= H3 (P2 × P2 ), and
Hq (P × P ) =
2 2
0 for all q ≥ 4.
6.9 Mayer-Vietoris Sequence for Singular Homology

Let X1 , X2 be two subspaces of a space X. We would like to know how
the singular homology groups of X1 ∪ X2 can be computed if we know the
homology groups of X1 , X2 and X1 ∩ X2 . The Mayer-Vietoris exact sequence
provides a nice answer to this question. If X1 , X2 both are closed (respectively
open) in X and X1 ∩X2 = φ then, obviously, both X1 , X2 are open (respectively,
closed) in X1 ∪ X2 and, in such a case, we have already seen (Remark 6.7.5)
that for a general homology theory, Hq (X1 ∪ X2 ) ∼ = Hq (X1 ) ⊕ Hq (X2 ) for all
q ≥ 0. However, if any one of the sets is only closed and the other is only
open, then X1 ∩ X2 = φ does not mean that such a relation will be valid. In
case X1 ∩ X2
= φ, then Mayer-Vietoris exact sequence connects the homology
modules of X1 , X2 , X1 ∩ X2 and X1 ∪ X2 under a suitable condition on X1
and X2 . We explain it precisely: Let i1 : X1 ∩ X2 → X1 , i2 : X1 ∩ X2 → X2 ,
j1 : X1 → X1 ∪ X2 and j2 : X2 → X1 ∪ X2 be the obvious inclusion maps. These
inclusion maps induce homomorphisms in singular chain complexes which are
also inclusion maps. This evidently gives us the following sequence of free chain
complexes and chain maps, which is easily seen to be exact:
(i1# , i2# )
0 S∗ (X1 ∩ X2 ) S∗ (X1 ) ⊕ S∗ (X2 )
j1# − j2#
S∗ (X1 ) + S∗ (X2 ) 0.
The above chain maps are defined by:

(i1# , i2# )(c) = (i1# (c), i2# (c)),
(j1# − j2# )(c1 , c2 ) = j1# (c1 ) − j2# (c).
We can now apply the long exact homology sequence theorem to the above
short exact sequence of chain complexes to get the following exact sequence in
homolgy.
· · · → Hq (X1 ∩ X2 ) → Hq (X1 ) ⊕ Hq (X2 ) → Hq (S∗ (X1 ) + S∗ (X2 )) →
6.9. Mayer-Vietoris Sequence for Singular Homology 273
Hq−1 (X1 ∩ X2 ) → · · · .
In the above sequence we wish to replace Hq (S∗ (X1 )+S∗ (X2 )) by Hq (X1 ∪
X2 ) for each q ≥ 0. To achieve this let us say that a pair (X1 , X2 ) of subspaces
of a space X is excisive if the inclusion map S∗ (X1 ) + S∗ (X2 ) ⊂ S∗ (X1 ∪ X2 )
induces isomorphisms in homology. In view of Theorem 6.6.2, we know that
the above inclusion map is a chain homotopy equivalence if X = X1 ∪ X2 =
int X1 ∪ int X2 . Thus, if we assume that X = X1 ∪ X2 = int X1 ∪ int X2 ,
then the pair (X1 , X2 ) will be excisive and the long exact homology sequence
mentioned above gives us the following long exact sequence in singular homol-
ogy:
(i1∗ , i2∗ ) j1∗ − j2∗
(∗) ··· Hq (X1 ∩ X2 ) Hq (X1 ) ⊕ Hq (X2 )
∂
Hq (X1 ∪ X2 ) Hq−1 (X1 ∩ X2 ) ···
The above sequence is known as the Mayer-Vietoris Exact Sequence for

singular homology. This sequence is frequently useful in computing the
homology of union of two subspaces if we know the homology of individual
subspaces and their intersection.
Note that all the singular chain complexes involved in the preceding
short exact sequence of chain complexes are free and hence the sequence will
remain exact even after we tensor it by an R-module G. Thus, the resulting
short exact sequence of chain complexes gives us a similar Mayer-Vietoris
exact sequence for singular homology with coefficients.
For simplicial homology we have a stronger result. Observe that if K1 , K2

are any two subcomplexes of a simplicial complex K then, by definition,
C∗ (K1 ∪ K2 ) = C∗ (K1 ) + C∗ (K2 ). This directly gives us the following short
exact sequence of simplicial chain complexes:
0- C∗ (K1 ∩ K2 ) -) C∗ (K1 ) ⊕ C∗ (K2 ) j1# − j-

(i1# , i2# 2#
C∗ (K1 ∪ K2 )- 0.
This, in turn, yields the following Mayer-Vietoris exact sequence in simplicial

homology (for arbitrary K1 and K2 ):
→ Hq (K1 ∩ K2 ) → Hq (K1 ) ⊕ Hq (K2 ) → Hq (K1 ∪ K2 ) → Hq−1 (K1 ∩ K2 ) →
Mayer-Vietoris Sequence for Reduced Homology

There is also a Mayer-Vietoris exact sequence for reduced singular ho-
mology for those subsets X1 , X2 of a space X which satisfy the condition
that X1 ∩ X2
= ∅ and X1 ∪ X2 = int X1 ∪ int X2 . To prove this note that
i j
the sequence 0 −→ Z −→ Z ⊕ Z −→ Z, where the maps i and j are defined
by i(n) = (n, n), j(m, n) = m − n, is clearly exact. Now, let S∗ (X1 ) → Z,

S∗ (X2 ) → Z and S∗ (X1 ∩ X2 ) → Z be the augmentation homomorphisms.
Then, with U = {int X1 , int X2 }, we have the following commutative diagram
of chain complexes and chain maps in which the rows are exact and the vertical
maps are onto.
(i1# , i2# ) j1# − j2#

0 S∗ (X1 ∩ X2 ) S∗ (X1 ) ⊕ S∗ (X2 ) S∗ (X1 ) + S∗ (X2 ) 0
0 Z Z⊕Z Z 0
i j
It follows from the above diagram that the kernel sequence of the augmentation
homomorphisms is also exact, viz., the sequence
0- S̃∗ (X1 ∩ X2 ) -) S̃∗ (X1 ) ⊕ S̃∗ (X2 ) j1# − j-

(i1# , i2# 2#
S̃∗ (X1 ) + S̃∗ (X2-
) 0
is exact. Then, as in the proof of the unreduced case, this last exact sequence
of chain complexes and chain maps gives us the following Mayer-Vietoris exact
sequence in reduced singular homology under the conditions stated above:
→ H̃q (X1 ∩ X2 ) → H̃q (X1 ) ⊕ H̃q (X2 ) → H̃q (X1 ∪ X2 ) → H̃q−1 (X1 ∩ X2 ) →
We have a similar reduced Mayer-Vietoris exact sequence for simplicial

homology and for homology with coefficients.
Mayer-Vietoris exact sequence is a very powerful and convenient tool for

computing homology groups. We have
Example 6.9.1. (One-point union of n-circles). Let X = X1 ∪ · · · ∪ Xn denote

the one-point union of n-spaces Xi , each of which is homeomorphic to the unit
circle. Then X (sometimes called n-leaved rose) has a triangulation, viz., it is
homeomorphic to the union of n triangles T1 , T2 , . . . , Tn all having one vertex,
say v, in common (see Fig. 6.6).
Fig. 6.6: The case n = 3 of n-leaved rose

6.9. Mayer-Vietoris Sequence for Singular Homology 275
Note that for each i = 1, 2, . . . , n, H0 (Ti ) ∼

= Z ∼
= H1 (Ti ), Hq (Ti ) = 0 for
all q > 1. Since X is path-connected, H0 (X) ∼ = Z. Applying reduced Mayer-
Vietoris exact sequence to complexes K1 and K2 , where |K1 | ≈ T1 , |K2 | ≈ T2 ,
K1 ∩ K2 = {v} and |K1 ∪ K2 | = T1 ∪ T2 , we find the nontrivial part of the
exact sequence as
0 → H̃1 (K1 ) ⊕ H̃1 (K2 ) → H̃1 (K1 ∪ K2 ) → 0.
Thus, we get H̃1 (K1 ∪ K2 ) = Z ⊕ Z and for q ≥ 2, H̃q (K1 ∪ K2 ) = 0. Applying

induction and using the reduced Mayer-Vietoris sequence of simplicial homol-
ogy, we find that H̃1 (X1 ∪ X2 ∪ · · · ∪ Xn ) = ⊕n1 Z and Hq (X1 ∪ · · · ∪ Xn ) = 0
for all q > 1. Consequently, we have
⎧
⎨ Z,
n
q=0
Hq (X1 ∪ · · · ∪ Xn ) ∼
= Z, q =1
⎩ 1
0, otherwise.
Remark 6.9.2. We may mention here that in the above example, the Mayer-
Vietoris sequence for singular homology does not apply directly – one has to
extend the triangles at the vertices by a small amount, use homotopy axiom,
etc., before applying the Mayer-Vietoris exact sequence. The above compu-
tation using simplicial homology is quite convenient; since X is a compact
polyhedron, the final result is valid for any homology.
The above result also holds for the one-point union of a finite number of
n-spheres for any n ≥ 1. As a matter of fact, a close examination of the above
example tells us that if we take a space X as the one-point union of n connected
polyhedra K1 , K2 , . . . , Kn , then the simplicial homology of X is clear: it is Z
in dimension zero and is isomorphic to the direct sum Hq (K1 ) ⊕ · · · ⊕ Hq (Kn )
in dimension q > 0.
Remark 6.9.3. There is also a Mayer-Vietoris exact sequence for pairs of sub-
spaces {(X1 , A1 ), (X2 , A2 )} of a topological pair (X, A) under suitable hypoth-
esis on X1 , X2 and A1 , A2 . The formulation of the statement, etc., is left as an
exercise.
Remark 6.9.4. One can also compute the homology groups of the n-sphere
Sn using Mayer-Vietoris exact sequence. For this, let N be the north pole
and S be the south pole of Sn (see Fig. 6.7). Then X1 = Sn − {N },
X2 = Sn − {S} are two contractible subspaces of Sn such that Sn = X1 ∪ X2 ,
int X1 ∪ int X2 is of the same homotopy type as Sn−1 . Hence, applying the
reduced Mayer-Vietoris sequence for X1 , X2 for singular homology, we find that
H̃q (Sn ) ∼
= H̃q−1 (Sn−1 ) ∼
= H̃q−n (S0 ) ∼
= Z if q = n, and 0 otherwise. For simpli-
cial homology, we can imagine Sn as the union of two cones X1 , X2 over Sn−1 ,
i.e., X1 ∪ X2 = Sn , X1 ∩ X2 = Sn−1 . Since cones are acyclic, we can apply
the reduced Mayer-Vietoris sequence for simplicial homology to conclude that

H̃q (Sn ) ∼
= H̃q−1 (Sn−1 ), etc. Note that, unlike the case of singular homology,
homotopy axiom is not necessary in computing the simplicial homology of Sn .
n
S
S n−1
S
Fig. 6.7
Mayer-Vietoris Sequence for Abstract Homology
There is an important point about Mayer-Vietoris exact sequence

which must be mentioned. This sequence is exact not only for simpicial
or singular homology, but it is also valid for any homology theory. This
appears rather too much in the first instance because a homology theory
does not arise from any chain complex, and so the first question will be: how
can a Mayer-Vietoris sequence be even formulated for an abstract homol-
ogy theory? It is interesting to see that it is possible, though it does require
some additional work to be done using only the axioms of the homology theory.
Let X1 , X2 be two subspaces of a space X. We say that the

pair (X1 , X2 ) is an excisive couple or (X, X1 , X2 ) is a proper triad
if the inclusion maps k1 : (X1 , X1 ∩ X2 ) → (X1 ∪ X2 , X2 ) and
k2 : (X2 , X1 ∩ X2 ) → (X1 ∪ X2 , X1 ) induce isomorphisms in homology,
i.e., for all q ≥ 0, k1∗ , k2∗ : Hq (Xi , X1 ∩ X2 ) → Hq (X1 ∪ X2 , Xj ) is an
isomorphism for i = 1, 2 and j = 2, 1 respectively. Examples of proper triads
are (i) any two open sets X1 , X2 in a space X – this follows from the excision
axiom and, (ii) any two subcomplexes |K1 |, |K2 | of a simplicial complex |K| –
this is left as an exercise.
For a given proper triad (X, X1 , X2 ), we put A = X1 ∩ X2 and X =

X1 ∪ X2 . Let us write the long exact sequences for the pairs (X1 , A) and
(X, X2 ) and then consider the vertical maps arising from the excision map
(X1 , A) → (X, X2 ). Then every square of the following diagram is easily seen
to be commutative:
6.10. Some Classical Applications 277
- Hq (A) i1 ∗- j1 ∗- ∂ - H (A) -
Hq (X1 ) Hq (X1 , A) q−1
k1 ∗ k1 ∗ k1 ∗ k1 ∗
? i1∗- ?
j1∗ ? ?
- Hq (X2 ) Hq (X) - Hq (X, X2 ) ∂ - H (X ) -
q−1 2
Let us define maps φq , ψq and ηq as indicated in the following sequence,

where η is given by η(z) = ∂ ◦ k∗−1 ◦ j1∗

(z):
(∗) · · · -) Hq (X1 ) ⊕ Hq (X2 )ψq = (k ∗ −i

- Hq (A) φq = (i1∗ , k1∗ -1∗ )
ηq - - ··· .
Hq (X) Hq−1 (A)
Then a diagram chase shows that the above sequence is exact (this is left as an
exercise). The above sequence is, in fact, the Mayer-Vietoris exact sequence of
the proper triad (X, X1 , X2 ). We emphasize again that the sequence has been
derived using only the axioms of a homology theory.
6.10 Some Classical Applications
We have seen (see the chapter on Simplicial Homology) that using the
simplicial homology of polyhedra, we could prove that Sm is not homeomorphic
to Sn , (m
= n); Rm is not homeomorphic to Rn , (m
= n); the discs Dn have
the fixed point property (Brouwer’s fixed point theorem); the spheres Sn , n ≥ 2
do not have nonzero tangent vector fields, etc. All these results can also be
proved, rather more easily, using the singular homology in place of simplicial
homology. In this section we are going to present a few deeper applications of
singular homology, proved by Brouwer as early as in 1911.
Proposition 6.10.1. Let α ∈ Hq (X). Then there exists a compact subset K of

X such that α lies in the image of i∗ : Hq (K) → Hq (X), where i : K → X is
the inclusion map.
Proof. Note that if σ : Δq → X is a singular simplex, then σ(Δq ) is a compact

subset of X, called the minimal carrier of σ. Clearly, the minimal carrier of a
finite sum of singular simplexes is simply the union of their minimal carriers.
Hence, if c = ri σi ∈ Sq (X) is a singular q-chain in X such that α = {c}, then
the minimal carrier of c is a compact subset, say K of X. We can consider c = c
as a singular q-chain in K. Then i : K → X maps c to c. Since ∂c = ∂c, if c is
a cycle in X, it is also a cycle in K, i.e., i∗ {c } = {c} = α, where {c } ∈ Hq (K).
The above proposition is usually stated by saying that singular homology

has compact supports.
Proposition 6.10.2. Let K ⊂ X be a compact subset and α ∈ Hq (K) is such

that i∗ (α) = 0 in Hq (X), where i : K → X is the inclusion map. Then, there
exists a compact subset K containing K such that j∗ (α) = 0 in Hq (K ), where
j : K → K is the inclusion map.

Proof. If α is represented by the singular q-chain c = ri σi , where σi are
singular q-simplexes, then i(c) = c = ri σi is a chain in X, which is a
boundary by assumption. Therefore, there exists a (q + 1)-chain d in X such
that ∂(d) = c . Let K be the minimal carrier of d and put K = K ∪ K . Then
c becomes a boundary in K , i.e., if j : K → K is the inclusion map, then
j(c) = c is a boundary in K . This says that j∗ {c} = {c } = 0 in Hq (K ).
The classical Jordan Curve theorem says that if we take any simple closed
curve C (homeomorphic copy of the unit circle) in the plane R2 , then C de-
composes the plane into two components, one inside of the curve and the other
outside of the curve, whose common boundary is the curve C itself.
Fig. 6.5
In this section, we will prove this theorem and much more. We emphasize
that we need a proof even in the case of plane since the shape of a simple
closed curve in the plane could be very complicated, and beyond imagination;
it need not look like the one in Fig. 6.5. Note also that if we take the one-point
compactification R2 ∪ {∞} (homeomorphic to S2 ) of R2 , then the Jordan curve
theorem says that any simple closed curve C in S2 decomposes the sphere into
two components, whose common boundary is the curve C itself. We will prove
a more general result: Let n ≥ 1 be a positive integer and C be a subset of Sn
which is homeomorphic to the sphere Sn−1 . Then Sn − C has two components
and C is the common boundary of both the components. Here again, C need
not look like Sn−1 embedded in Sn in the canonical fashion; it could be visually
unimaginable.
Let us start with the following observation: If we remove one point

from the n-sphere Sn , n ≥ 1, the remaining space is homeomorphic to
the Euclidean space Rn . What happens if we remove a subset of Rn which
looks like (is homeomorphic to) a closed interval (1-disc) or a 2-disc or

more generally, a k-disc, k ≤ n? The next theorem answers this question
in terms of singular homology. We do not know about the homeomorphism
type of the remaining space, but we prove below that the remaining space
is homologically trivial, i.e., it is path connected and much more. Next,
we also want to know what happens when we remove a subset A of Sn
n+1 S , 0 ≤ k < n. Note that if we write
k
which is homeomorphic to a k-sphere
Sn = {(xi , . . . , xn+1 ) ∈ Rn+1 | i x2i = 1} and put xn+1 = 0, then we get
Sn−1 embedded in Sn canonically. Removing this Sn−1 from Sn clearly leaves
two open hemispheres Dn+ and Dn− whose common boundary is Sn−1 itself.
However, if we remove a subset C of Sn which is homeomorphic to Sn−1 , it
is not clear what is the remaining space. Jordan-Brouwer Separation theorem
asserts that even in this case, we get the same result - this is indeed remarkable!
Let us prove the following :
Theorem 6.10.3. Let D be a subset of Sn homeomorphic to the closed k-ball

Dk , 0 ≤ k ≤ n. Then Sn − D is acyclic, i.e., H̃q (Sn − D) = 0 ∀ q ≥ 0.
Proof. We prove the theorem by induction on k. When k = 0, Sn − D is
simply a point in case n = 0, otherwise it is the punctured n-sphere and so is
homeomorphic to Rn . We know that Rn is acyclic and the theorem is true for
k = 0. Now let us suppose that the theorem is true for (k − 1), k ≥ 1, and
D is a homeomorphic to a k-disc Dk . Let h : I k → D be a homeomorphism.
Let D1 = h(I k−1 × [0, 12 ]) and D2 = h(I k−1 × [ 12 , 1]) be subsets of Sn and
C = h(I k−1 × { 12 }) be the disc of dimension (k − 1).
Sn
C
D1
D2
Fig. 6.6
Suppose α ∈ H̃q (Sn − D) is a nonzero element. There are inclusion maps

i1 : (Sn − D) → (Sn − D1 ) and i2 : (Sn − D) → (Sn − D2 ). Let X = Sn − C.
Then by induction hypothesis, X is acyclic. Clearly, X = X1 ∪ X2 where
X1 = Sn − D1 and X2 = Sn − D2 . Since X1 , X2 are open in Sn , these are
also open in X. Hence by the reduced Mayer-Vietoris sequence, the following
sequence is exact:
H̃q+1 (X) → H̃q (A) → H̃q (X1 ) ⊕ H̃q (X2 ) → H̃q (X).
Here A = X1 ∩ X2 = Sn − D. Since X is acyclic, the middle map is an isomor-

phism. We know the middle map which takes α ∈ Hq (A) to (i1∗ (α), −i2∗ (α)).
Hence, at least one of the two elements i1∗ (α) or i2∗ (α) must be zero.
Suppose i1∗ (α)
= 0. Then, we write D1 as the union of h(I k−1 × [0, 14 ]) and
h(I k−1 × [ 14 , 12 ]). By the same argument as above, we find that the image of
α in H̃q (I k−1 × [0, 14 ]) or H̃q (I k−1 × [ 14 , 12 ]) is nonzero. We continue like this
and find a sequence of closed intervals [a1 , b1 ] ⊃ [a2 , b2 ] ⊃ · · · ⊃ [am , bm ] of
lengths 12 , 14 , . . . such that the image of α in H̃q (I k−1 × [am , bm ]) is nonzero
for each m ≥ 1. Let x0 be the unique point in the intersection of all these
intervals and put E = h(I k−1 × {x0 }). By induction hypothesis, the group
H̃q (Sn − E) is trivial. Hence the image of α in this group is also zero. Since
the singular homology has compact supports (see Proposition 6.10.1), there
is a compact subset A of Sn − E such that the image of α in H̃q (A) is
zero. Since Sn − E is the union of open sets Sn − D1 , Sn − D2 , . . . , where
Di = h(I k−1 × [ai , bi ]) and A is compact, A ⊂ Sn − Dm for some m. This
means image of α in H̃q (Sn − Dm ) must be zero, which contradicts the asser-
tion proved earlier. Hence α ∈ H̃q (Sn −D) must be zero, i.e., Sn −D is acyclic.
We have computed the singular homology of spaces obtained by removing

k-discs from the n-sphere, 0 ≤ k ≤ n. Now let us compute the singular homol-
ogy of those spaces which are obtained by removing homeomorphic copies of
k-spheres Sk from Sn , 0 ≤ k < n. We have
Corollary 6.10.4. Let n > k ≥ 0 and h : Sk → Sn be an embedding. Then

Z, q =n−1−k
H̃q (Sn − h(S k )) =
0, otherwise.
Proof. We will prove this Corollary again by induction on k. When k = 0, h(S0 )

consists of two points, say a and b of Sn . Since Sn − {a} − {b} is homeomorphic
to Rn − {0}, and the latter is of the same homotopy type as Sn−1 , we see that
H̃q (Sn − h(S0 )) ≈ Z if q = n − 1, and zero otherwise. Suppose now that the
Corollary is true for (k − 1), k ≥ 1. Let D+ k
and D−k
be the upper and lower
closed hemispheres of S and put X1 = S − h(D+ ), X2 = Sn − h(D−
k n k k
).
h(Sk ) h(D+
k
)
h(Sk−1 ) h(D−
k
)
Fig. 6.7
Let X = X1 ∪ X2 and A = X1 ∩ X2 = Sn − h(Sk ). Since X1 , X2 are open sets

of X, we have the following part of the reduced Mayer-Vietoris exact sequence
H̃q+1 (X1 ) ⊕ H̃q+1 (X2 ) → H̃q+1 (X) → H̃q (A) → H̃q (X1 ) ⊕ H̃q (X2 ).
Both X1 , X2 are acyclic by the previous theorem, and therefore H̃q+1 (X) →
H̃q (A) is an isomorphism, i.e., ∀ q
Hq+1 (Sn − h(Sk−1 )) ≈ Hq (Sn − h(Sk )).
Hence, by induction hypothesis

Z, q + 1 = n − 1 − (k − 1)
H̃q (Sn − h(S k )) =
0, otherwise.
Now we can prove our
Theorem 6.10.5. (Jordan-Brouwer Separation Theorem). Let C be a subset of

n-sphere Sn , n ≥ 1, which is homeomorphic to Sn−1 . Then Sn − C has exactly
two components of which C is the common boundary.
Proof. By the previous theorem we know that H̃0 (Sn − C) ≈ Z. Hence Sn − C
has exactly two path components. Since Sn − C is locally path connected open
sets, path components and components are same. Now the use of algebraic
topology in this proof is over. Let W1 and W2 be the two components. These
are open in Sn , since Sn − C is an open set. Hence the boundary of each of
these is W 1 − W1 and W 2 − W2 respectively.
C2
x U
b
C1 a y S2
W1 C
W2
Fig. 6.8
Since no point of W2 can be a limit point of W1 , W 1 − W1 ⊆ C. Similarly,

W 2 − W2 ⊆ C. Hence it is enough to prove that C ⊆ W i − Wi , i = 1, 2. Let
us show that C ⊆ W 1 − W1 . Let x ∈ C. We show that x is a limit point of
W 1 − W1 , and since W 1 − W1 is closed, x will belong to W 1 − W1 . To prove
that x is a limit point of W 1 − W1 , let U be an open neighbourhood of x in
Sn . We consider C as Sn−1 and divide C into closed discs (upper hemisphere
and lower hemisphere) C1 and C2 . We can assume that C1 is small enough to

be contained in U . Now since Sn − C2 is connected, we take a path ω joining
a point a ∈ W1 and a point b ∈ W2 . This path ω will intersect W 1 − W1
because otherwise the image of ω (connected set) will lie in the union of W1
and Sn − W 1 , which are disjoint open sets, a contradiction. Let y ∈ W 1 − W1
be the point where the path ω intersects W 1 − W1 . Since y ∈ C and y ∈ / C2 ,
we find that y ∈ C1 . But C1 ⊆ U by our assumptions. Hence y ∈ U . Thus U
intersects W 1 − W1 and the proof is complete.
The Jordan Curve Theorem is classically stated for simple closed curves in
the plane R2 , rather than for simple closed curves in S2 . The same is also true
for the more general Jordan Brouwer separation theorem. This can be proved
as follows.
Theorem 6.10.6. Let C be a subset of Rn , n ≥ 2, which is homeomorphic

to Sn−1 . Then Rn − C has two path components of which C is the common
boundary.
Proof. We can consider Rn as Sn − {N }, where N is the north pole of Sn . Then
C is a subset of Sn homeomorphic to Sn−1 . Hence by the Jordan-Brouwer
separation theorem, Sn − C has two path components of which C is the
common boundary. Let W1 and W2 be the two components and assume that
N ∈ W1 . We claim that W1 − {N } is connected. This follows from a general
fact that if U is an open connected set of Sn and a point a ∈ U is removed, then
U − {a} remains connected: to see this suppose U − {a} is not connected and
consider its components (all of them are open in Sn ). There is a neighbourhood
V of a which is homeomorphic to Rn , and so V − {a} ≈ Rn − {0} will be
connected, since n > 1. Now let A be the component of U − {a} which
contains the connected set V − {a} and B be the union of other components
of U − {a}. Then A ∪ {a} and B give a decomposition of the connected set
U , a contradiction. Therefore, it follows that W1 − {N } and W2 are two
components of Rn − C, whose common boundary is C.
Next we are going to use the Jordan-Brouwer separation theorem to prove

a remarkable result, called “invariance of domain” theorem. In set topology, we
already know that a continuous bijective map need not be homeomorphism.
However, it is interesting to see that if U is an open set of Rn or Sn and f : U →
Rn or Sn is an injective continuous map, then f is indeed a homeomorphism.
We have
Theorem 6.10.7. (Invariance of Domain). Let U ⊂ Rn be an open set and

f : U → Rn be a continuous injective map. Then f (U ) is an open set of Rn and
f : U → f (U ) is an embedding.
Proof. Let x ∈ U and y = f (x). We will show that f (U ) is a neighbourhood of
each of its points y and is, therefore, open in Rn .
6.11. Singular Cohomology and Cohomology Algebra 283
U W1 W2
B f
x y
S
f(S )
Fig. 6.9
Let us take an open ball B of radius around x whose closure lies in U . Put
S = B − B and note that S is a (n − 1)-sphere. Since f : S → f (S ) is a
continuous bijection from a compact space to a Hausdorff space, f (S ) is also
an (n − 1)-sphere. Since f : S → f (S ) is an embedding, by Jordan-Brouwer
separation theorem f (S ) decomposes Rn into two components say W1 and W2
which have f (S ) as the common boundary. Note that the set f (B ) is connected
and is disjoint from f (S ). Hence it lies either in W1 or in W2 . Suppose f (B ) ⊆
W1 . We claim that f (B ) = W1 because otherwise f (B ) = f (B )∪f (S ), being
a disc, will separate Sn − f (B ) into two components viz., W1 − f (B ) and W2 .
This will give a contradiction to the fact that when a disc is removed from
Sn , the remaining space remains connected. Hence f (B ) = W1 and f (U ) is
an open neighbourhood of y in Rn . Now f is an embedding because the map
f : U → f (U ) is easily seen to be an open map, and it is already bijective.
6.11 Singular Cohomology and Cohomology Algebra

In this section, we define the singular cohomology of a topological pair
(X, A) and for this we basically require the properties of the contravariant
Hom(. , .) functor discussed in Appendix. For the sake of generality, we again
assume that R is a PID. For a topological space X, let
∂q+1 ∂q
S∗ (X) : −→ Sq+1 (X) −→ Sq (X) −→ Sq−1 (X) −→ · · · −→ S0 (X) −→ 0,
denote the singular chain complex of X. Here, each Sq (X) is an R-module and
∂q ◦∂q+1 = 0 for all q ≥ 0. If we apply the Hom(. , R) functor to the above chain
complex, we get the following long sequence going in the opposite direction:
Hom(S∗ (X), R) : 0 - Hom(S0 (X), R) - · · · - Hom(Sq−1 (X), R)
Hom(∂q , R)
- Hom(∂q+1 , R)
-
Hom(Sq (X), R) Hom(Sq+1 (X), R) -
If, for all q ≥ 0, we put Hom(Sq (X), R) = S q (X) and Hom(∂q , R) = δ q , then
the above sequence of R-modules may be written as
δq δ q+1
S ∗ (X) : 0 −→ S 0 (X) −→ · · · −→ S q−1 (X) −→ S q (X) −→ S q+1 (X) −→
which has the property that for all q ≥ 0,
δ q+1 ◦ δ q = Hom(∂q+1 , R) ◦ Hom(∂q , R)

= Hom(∂q ◦ ∂q+1 , R)
= Hom(0, R) = 0.
Thus, S ∗ (X) is indeed a cochain complex of R-modules, which is called the

δ q+1
singular cochain complex of X. Its homology, viz., Hq (S ∗ (X)) := ker Im δ q ,
denoted by H q (X), q ≥ 0, is called the singular cohomology module of the
space X. For later use, let us recall that δ q : S q−1 (X) → S q (X) is defined by
δ q (α) = α ◦ ∂q for any α ∈ Hom(Sq−1 (X), R). Hence, onwards we write δ q−1
for δ q for consistency of notation.
Given a continuous map f : X → Y , we have the induced chain map

f# : S∗ (X) → S∗ (Y ). Applying the functor Hom(. , R), we get a cochain
map Hom(f# , R) : Hom(S∗ (Y ), R) → Hom(S∗ (X), R), which we denote by
f # : S ∗ (Y ) → S ∗ (X). Then, this cochain map obviously induces a homomor-
phism in homology, i.e., we get induced homomorphism f ∗ : H q (Y ) → H q (X)
for all q ≥ 0, in singular cohomology modules of X and Y . With these defini-
tions, one can easily verify the following functorial (contravariant) properties:
Proposition 6.11.1. (i) If IX : X → X is the identity map, then the induced

homomorphism (IX )∗ : H q (X) → H q (X) is also the identity map for all
q ≥ 0.
(ii) If f : X → Y, g : Y → Z are two maps then, for all q ≥ 0,
(g ◦ f )∗ = f ∗ ◦ g ∗ : H q (Z) → H q (X).
The following result should be compared with the corresponding homology

result (see Proposition 6.2.8).
Proposition 6.11.2. Let {X

i , i ∈ I} be the path-components of a space X. Then,
for all q ≥ 0, H q (X) ∼
= i∈I H q (Xi ), i.e., the singular cohomology of X is
isomorphic to the product of the singular cohomology of its path-components.
Proof. Since the q-simplex Δq is path-connected, any singular q-simplex
σ : Δq → X maps Δq in some path-component Xi of X. Therefore, for all
q ≥ 0, Sq (X) ∼= i∈I Sq (Xi ). Since the Hom functor converts the direct sum
into direct product, we find that
)
Hom(Sq (X), R) ∼
= Hom(⊕Sq (Xi ), R) ∼
= Hom(Sq (Xi ), R).
i∈I
Since the group of q-cocycles and q-coboundaries of X are also the

direct products of q-cocycles and coboundaries of Xi , i ∈ I, we get the desired
result.
The above proposition tells us, in particular, that the singular cohomology
in dimension zero of a discrete space X will be isomorphic to the direct product
of as many copies of R as there are points in X and will be zero for higher
dimensions.
Reduced Cohomology
If X
= φ, then we can define an augmentation map η : R →
Hom(S0 (X), R) as thedual of the augmentation
: S0 (X) → R by putting
η(k)( ni xi ) = k · ( ni xi ) = k · ( ni ). Using this, we define the reduced
singular cohomology H̃ q (X) of the space X as the homology of the cochain
complex:
η δ0 δ1 δq
0 −→ R −→ S 0 (X) −→ S 1 (X) −→ · · · −→ S q (X) −→ S q+1 (X) −→ · · · ,
where we have δ 0 η = 0. It follows from this definition that
H̃ 0 (X) ⊕ R ∼
= H 0 (X), H̃ q (X) = H q (X), for all q > 0.
Relative Cohomology and the Axioms

Let us now extend the definition of singular cohomology to topological
pairs (X, A). Recall that
0 → S∗ (A) → S∗ (X) → S∗ (X, A) → 0
is an exact sequence of singular chain complexes of R-modules in which each

complex consists of free R-modules. Since the above sequence of chain com-
plexes splits, by applying the Hom(., R) functor, we find that the induced se-
quence of singular cochain complexes
j# i#
0 −→ S ∗ (X, A) −→ S ∗ (X) −→ S ∗ (A) −→ 0
is also exact. We define the cohomology of the cochain complex S ∗ (X, A) to

be the singular cohomology of the pair (X, A), which we denote by H q (X, A),
q ≥ 0. Since the preceding sequence of cochain complexes is exact, we get the
following long exact sequence for singular cohomology:
j∗ ∗ i ∂
· · · −→ H q (X, A) −→ H q (X) −→ H q (A) −→ H q+1 (X, A) −→ · · · .
This proves the Exactness Axiom for singular cohomology. The Commutativity
Axiom for a map f : (X, A) → (Y, B) also follows by writing the long exact
cohomology sequences for pairs (X, A) and (Y, B) arising from the following
commutative diagram of singular cochain complexes and cochain maps:
0 - S ∗ (X, A) - S ∗ (X) - S ∗ (A) - 0

6
f# 6
f# 6
( f |A ) #
0 - S ∗ (Y, B) - S ∗ (Y ) - S ∗ (B) - 0
To prove the Homotopy Axiom for singular cohomology, let

i0 , i1 : (X, A) → (X × I, A × I) denote the bottom and top inclu-
sion maps. Then we have already proved that the induced chain maps
i0# , i1# : S∗ (X, A) → S∗ (X × I, A × I) are chain homotopic. Conse-
quently, applying the Hom(. , R) functor, we find that the cochain
∗
maps i# 0 , i#
1 : S (X × I, A × I) → S ∗ (X, A) are cochain homo-
topic. Therefore, the induced homomorphisms in singular cohomology
i∗0 , i∗1 : H q (X × I, A × I) → H q (X, A) are identical. But this means, as in
the case of singular homology, that if f, g : (X, A) → (Y, B) are homotopic
maps of pairs then the induced homomorphisms in singular cohomology
f ∗ , g ∗ : H q (Y, B) → H q (X, A) are identical for all q ≥ 0.
By a similar argument, one can easily prove the Excision Axiom for sin-
gular cohomology. The Dimension Axiom being evidently true, we have the
following result:
Theorem 6.11.3. The singular cohomology defined on the category of all topo-
logical pairs is a cohomology theory in the sense of Eilenberg-Steenrod.
One can similarly define the simplicial cohomology of a polyhedral pair

(X, A), using the simplicial chain complex C∗ (X, A) in place of S∗ (X, A) ev-
erywhere. If G is an R-module, then singular cohomology of a pair (X, A), with
coefficients in G, can also be easily defined. Instead of the functor Hom(. , R),
we apply the functor Hom(. , G) to the singular chain complex S∗ (X, A) and
get the cochain complex S ∗ (X, A; G) := Hom(S∗ (X, A), G). The corresponding
homology modules, viz., Hq (S ∗ (X, A; G)), denoted by H q (X, A; G), are called
the singular cohomology modules of the pair (X, A) with coefficients in G. One
can again easily prove that H q (X, A; G), q ≥ 0, which are contravariant in the
pair (X, A), are indeed covariant functors in the coefficient module G. Thus,
an exact sequence 0 → G → G → G → 0 of coefficient R-modules gives us
the following long exact sequence
δq
→ H q (X, A; G ) → H q (X, A; G) → H q (X, A; G ) → H q+1 (X, A; G ) →
of cohomology of R-modules. The connecting homomorphism δ q :

H q (X, A; G ) → H q+1 (X, A; G ) is called the Bockstein cohomology op-
erator and they are known to have better applications in cohomology as
against homology.
Universal Coefficient Theorem for Cohomology

Having defined the singular homology and singular cohomology of a space
X, it is natural to ask as to how are the homology and cohomology groups of X
related. If our ground ring R is a PID, there is a short exact sequence connecting
the singular homology with singular cohomology of X showing that cohomology
groups are, to a great extent, dual to the homology groups. To appreciate the
exact relationship, we need to recall the properties of the Ext(. , .) functor
discussed in Appendix. The precise result is given by the following theorem:
Theorem 6.11.4. (Universal Coefficient Theorem for Cohomology). For an R-

module G, where R is a PID, and for any pair (X, A) of topological spaces,
there is the following split short exact sequence of R-modules for all q ≥ 0:
0 → Ext(Hq−1 (X, A), G) → H q (X, A, G) → Hom(Hq (X, A), G) → 0.
The above sequence is natural in (X, A) as well as in the coefficient module G,

but the splitting is not natural with respect to (X, A).
The proof of the above theorem follows from the corresponding result of
abstract chain and cochain complexes because here the singular chain com-
plexes are free (see Spanier [18] p. 243). Note that, because the above se-
quence splits, we find that for all q ≥ 0, H q (X, A; G) ∼
= Hom(Hq (X, A), G) ⊕
Ext(Hq−1 (X, A), G). This result is quite nice, at least for computational pur-
poses. We must keep the two special cases of the above theorem in mind: first,
when we take R = Z and secondly, when R is a field. When R is a field, the
Ext-term above vanishes and we see that cohomology is simply the dual of
homology. The following interesting result, which we state only for R = Z, is
also true for any PID.
Corollary 6.11.5. Suppose a continuous map f : X → Y induces isomorphism

f∗ : Hq (X) → Hq (Y ) in integral singular homology for all q ≥ 0. Then the
induced homomorphisms
f∗ : Hq (X; G) → Hq (Y ; G)
and
f ∗ : H q (Y ; G) → Hq (X; G),
in homology and cohomology with coefficients in G, are isomorphisms for any
abelian group G and every q ≥ 0.
Proof. Let S∗ (X), S∗ (Y ) denote the integral singular chain complexes of X
and Y and let f# : S∗ (X) → S∗ (Y ) denote the induced chain map. Since these
are chain complexes of free abelian groups and the induced homomorphism
f∗ : Hq (X) → Hq (Y ) in homology by the chain map f# is an isomorphism,
a well-known result of homological algebra (see Ex. 4, Section 7.2) says that
f# is a chain equivalence. This means f# ⊗ 1 : S∗ (X) ⊗ G → S∗ (Y ) ⊗ G

is also a chain equivalence of R-modules. Hence the induced homomor-
phism f∗ : Hq (X; G) → Hq (Y ; G) in homology, with coefficients in G, must
be an isomorphism. Next, we apply the Hom(. , G) functor to the chain
equivalence f# : S∗ (X) → S∗ (Y ) to infer that the induced homomorphism
f # : S ∗ (Y, G) → S ∗ (X, G) in cochain complexes is also a cochain equivalence.
This implies that the induced homomorphism f ∗ : H q (Y ; G) → H q (X; G) in
cohomology is also an isomorphism.
Note that the above result says that if two spaces cannot be distinguished
by integral homology groups then they cannot be distinguished by homology
or cohomology groups with any coefficient.
Example 6.11.6. Applying the universal coefficient theorem for cohomology,

one can now easily compute the integral singular cohomology groups of the
n-sphere Sn , the torus T , the Möbius band M , the projective plane P2 and
the Klein bottle K, etc. Notice that for an abelian group G, the Universal
Coefficient Theorem says that
H q (Sn ; G) ∼
= Hom(Hq (Sn ), G) ⊕ Ext(Hq−1 (Sn ), G).
Since Hq (Sn ) ∼
= Z for q = 0 or n, and is zero otherwise, we find

G, q = 0, n
H q (Sn ; G) =
0, otherwise.
Similarly, for the projective plane P2 , we have H 0 (P2 ) ∼

= Z, H 1 (P2 ) ∼
=
2 2 ∼
Hom(Z2 , Z) ⊕ Ext(Z, Z) = 0, H (P ) = Ext(Z2 , Z) = Z2 , H (P2 ) = 0 for all
q
q ≥ 3.
For coefficients in Z2 , we have H 0 (P2 ; Z2 ) ∼ = Z2 , H 1 (P2 ; Z2 ) =
Hom(Z2 , Z2 ) = Z2 , H (P ; Z2 ) = Hom(0, Z2 ) ⊕ Ext(Z2 , Z2 ) ∼
∼ 2 2
= Z2 ,, H q (P2 ) = 0
for all q ≥ 3.
Remark 6.11.7. We emphasize that cohomology with coefficients in G is ob-

tained by taking the Hom and Ext of integral homology, not of the homology
with coefficients in G. This can be checked by the example of projective plane
P2 computed above.
The fact that H 1 (P2 ) = 0 shows that the fundamental group of a space
is not related to the 1-dimensional cohomology of the space, notwithstanding
the fact that it is nicely related to its 1-dimensional homology, viz., π1 (P2 ) ∼
=
H1 (P2 ).
We also point out that the integral cohomology of Sn is exactly the dual of
its homology whereas the integral cohomology of the projective plane P2 is not
the dual of its integral homology. We observe at the same time that, though P2
is not orientable, yet its second integral cohomology is nonzero. This says that
orientability, which is related to the nonvanishing of nth integral homology of
a n-manifold, is not related to its cohomology in the same way.
We also have the following Mayer-Vietoris exact sequence for singular

cohomology.
Theorem 6.11.8. Let X1 , X2 be two subspaces of a space X such that X1 ∪

X2 = int X1 ∪ int X2 . Then the following Mayer-Vietoris sequence for singular
cohomology is exact for any R-module G (coefficients are suppressed ):
δ q−1
-H q (X1 ∪ X2 ) (j1∗ , j2∗-
) i∗1 − i∗2-
H q (X1 ) ⊕ H q (X2 ) H q (X1 ∩ X2 )
δ q - H q+1 (X ∪ X ) -
1 2
Subject to some conditions, the Künneth Formula for singular cohomology

is also valid and is given by the next theorem:
Theorem 6.11.9. (Künneth). Let X, Y be any two spaces. Then, for all n ≥ 0,
there is the following split short exact sequence for any coefficient groups G and
G satisfying the condition that Tor(G, G ) = 0, and (a) both H ∗ (X), H ∗ (Y )
are of finite type or, (b) H ∗ (Y ) is finite type and G is finitely generated,

0→ H p (X; G) ⊗ H q (Y ; G ) → H n (X × Y, G ⊗ G )
p+q=n

→ Tor(H p (X; G), H q (Y ; G )) → 0.
p+q=n+1
Furthermore, the above sequence is natural in spaces X, Y and the coefficient

groups G, G , but the splitting is not natural.
It is interesting to observe the case of Künneth formula when G = G = R

is a field. In this simplest situation, everything being free, we find that the
cohomology of the product space agrees exactly with the tensor product of the
cohomologies of the factor spaces because the term involving the Tor functor
vanishes. The proofs of the preceding theorems are omitted and can be found
in ([18] p. 239, 247) in all their details.
Cohomology Algebra
Let X be a topological space. We continue to assume that R is a

PID and that all homology, cohomology modules in this section are with
coefficients in the R-module R. Note that we have an obvious R-isomorphism
m : R ⊗ R → R defined by m(r ⊗ s) = r · s, where r · s is the multiplication
of the ring R. The cohomology cross product H p (X) × H q (X) → H p+q (X)
of the Künneth formula can be used qto define a product in the graded
cohomology module H ∗ (X) := q≥0 H (X) as follows: Let d : X → X × X
denote the diagonal map given by d(x) = (x, x). Clearly, d induces a ho-
momorphism d∗ : H n (X × X) → H n (X), n ≥ 0, of cohomology modules.
Observe that for any u ∈ H p (X), v ∈ H q (X), the cohomology product
u × v ∈ H p+q (X × X, R ⊗ R) and the multiplication isomorphism m de-
fines an element m∗ (u × v) ∈ H p+q (X × X, R). Therefore, the element
d∗ (m∗ (u × v)) ∈ H p+q (X, R). We define the cup product of elements
u ∈ H p (X), v ∈ H q (X), denoted by u ∪ v or just by the juxtaposition uv,
by putting u ∪ v = d∗ (m∗ (u × v)). Then, using the bilinearity, it is extended
to define the cup product in the whole graded cohomology module H ∗ (X)
of X. This product has several interesting properties; it is associative, it
has an identity element, viz., 1 ∈ H 0 (X, R), where 1 is represented by the
singular 0-cochain of X which maps all points of X (singular 0-chains) to
the unity element 1 ∈ R. The cup product is also commutative up-to a sign,
i.e., uv = (−1)deg(u)·deg(v) vu for all u, v ∈ H ∗ (X). Thus, H ∗ (X) becomes
an R-algebra with respect to this cup product. This algebra is called the
cohomology algebra of the space X. The cup product has one more striking
property, viz., it is natural in X, i.e., if f : X → Y is a map then the induced
homomorphism f ∗ : H ∗ (Y ) → H ∗ (X) in cohomology preserves the cup
product. In other words, f ∗ (u ∪ v) = f ∗ (u) ∪ f ∗ (v) for all u, v ∈ H ∗ (Y ). This
says, in particular, that the cohomology algebra H ∗ (X) of X is a topological
invariant of the space X.
In order to give an idea of the structure of the cohomology algebra of

a space, we will just state a few examples without going into details. The
interested reader can find the full details in any of the advanced books on
algebraic topology such as Bredon [2] or Spanier [18], etc.
Example 6.11.10. The cohomology algebra H ∗ (Sn ) of the n-sphere Sn , n ≥ 1,

is trivial, i.e., if a ∈ H n (Sn ) ∼
= Z is the generator, then a2 = 0 and this
says that the product of any two elements of positive degree will be trivial.
On the other hand, the cohomology algebra of Sn × Sn is nontrivial, i.e., if
u, v ∈ H ∗ (Sn × Sn ) ∼= Z ⊕ Z are the generators, then their cup product uv ∈
H 2n (Sn ×Sn ) ∼= Z is again a generator. In fact, it can be shown that u2 = 0 = v 2
and uv
= 0.
For any field F , let F [x] denote the polynomial ring over F in one indeter-
minate x. Let (xn+1 ) denote the ideal of the polynomial ring generated by the
element xn+1 . Then the quotient ring F [x]/(xn+1 ), which is an algebra over F ,
consists of polynomials in x of degree at most n. Such an algebra is called a
truncated polynomial algebra of height n. Furthermore, if F [x] is regarded as
a graded ring with indeterminate x of degree k, then F [x]/(xn+1 ) is called a
truncated polynomial algebra of degree k and height n. This concept is useful

in understanding the cohomology algebra of some familiar spaces.
Example 6.11.11. Let RPn denote the real projective space of dimension n. This
space is a compact, connected topological manifold of dimension n which is
triangulable. Its cohomology with coefficients in Z2 is given by H q (RPn , Z2 ) ∼ =
Z2 for q = 0, 1, 2, . . . , n and is trivial for other values of q. It can be shown that
the nonzero element w ∈ H 1 (RPn , Z2 ) has the property that w, w2 , w3 , . . . , wn
are all nonzero and wn+i = 0 for all i ≥ 1. Thus, the cohomology algebra of
RPn is a truncated polynomial algebra over Z2 of degree 1 and height n.
Example 6.11.12. Let CPn denote the complex projective space. It is a simply
connected, compact, triangulable manifold of real dimension 2n. The integral
cohomology of CPn is given by H q (CPn ) ∼ = Z, q = 0, 2, . . . , 2n and is trivial in
other dimensions. If w ∈ H 2 (CPn , Z) is a generator then it can be proved that
w2 , w3 , . . . , wn are all nonzero, and they generate the corresponding cohomol-
ogy groups and wn+1 = 0. It follows that the cohomology algebra of CPn over
Z is the truncated polynomial algebra over Z of degree 2 and height n.
It must be remarked here that the cup product can sometimes distinguish
between two nonhomeomorphic spaces when they cannot be distinguished by
homology or cohomology groups. We have
Example 6.11.13. Let X = Sp × Sq , p

= q. In the first example above,
we computed the cohomology algebra of X. Let us now consider a different
space Y = Sp ∨ Sq ∨ Sp+q , which is the one-point union of the three spheres
Sp , Sq , Sp+q . All these spheres are triangulable and therefore Y can be so
triangulated that all the spheres have a vertex v in common. Then, using the
Mayer-Vietoris exact sequence for simplicial homology, we find that

Z, i = 0, p, q, p + q
Hi (Y, Z) =
0, otherwise.
Using the universal coefficient theorem for homology and cohomology, one can
easily verify that the two spaces X and Y have isomorphic homology as well
as isomorphic cohomology groups with all coefficients. It is obvious that the
space X is not homeomorphic to the space Y because, for instance, any point
of Sp − Y has a neighbourhood which is homeomorphic to Rp whereas there
is no such point in the space X. In fact, the space X is a (p + q)-manifold
and Y is not a manifold. This example exhibits the limitation of the additive
structure of the homology and cohomology groups. However, it is interesting
to observe that the additional structure of the cup product in cohomology does
distinguish between X and Y . If a ∈ H p (Sp × {y}), b ∈ H q ({x} × Sq ) are
the generators, then the product ab ∈ H p+q (Sp × Sq ) is nonzero whereas the
product ab ∈ H p+q (Y ) is zero. In fact, this gives even a stronger result, viz.,
X, Y are not even of the same homotopy type – a result which is not at all
obvious by considerations of set topology.
Exercises

1. Prove that the Euler characteristic χ(X) = (−1)i rkHi (X, F ) of a
space X is independent of the coefficient field F used in defining the Eu-
ler
characteristic. Show, in particular, that χ(X) = (−1)i rkHi (X) =
(−1) rkHi (X, Q).
i
2. Let (X, A) be a pair. If any two of the groups H∗ (A), H∗ (X) and H∗ (X, A)
are finitely generated, then show that the third group is also finitely gen-
erated, and furthermore χ(X) = χ(A) + χ(X, A).
3. Let X1 , X2 be a pair of excisive couple of subspaces in a space X such
that H∗ (X1 ), H∗ (X2 ) are finitely generated. Prove that H∗ (X1 ∪ X2 ) is
finitely generated if and only if H∗ (X1 ∩ X2 ) is finitely generated, and if
that happens, show that χ(X1 ∪ X2 ) = χ(X1 ) + χ(X2 ) − χ(X1 ∩ X2 ).
4. Let X be the space obtained by puncturing a 2-sphere at two distinct
points. Prove that H0 (X) ∼
= Z, H1 (X) ∼ = Z and Hq (X) = 0 for all q ≥ 2.
(Hint: X is homotopy type of a circle.)
5. Consider the torus T as the quotient of the square I × I by identifying
the opposite edges AB with DC and AD with BC respectively (Fig. 6.8).
Let q : I × I → T denote the quotient map. Prove that
(i) The sets q(AB) and q(AD) are singular 1-simplexes in T which are
cycles.
(ii) The sets q(ADB) and q(ACD) are singular 2-simplexes whose dif-
ference is a boundary.
(iii) If X is a circle lying in the interior of I × I, then show that q(X) is
a singular one simplex in T which is a boundary.
A D q (AD)
B C q(AB)
Fig. 6.8: The two nonzero cycles in torus T .
6. Compute the singular homology of the following spaces:
(a) The Topologist’s Sine Curve, i.e., X = {(x, y) ∈ R2 : y =
sin 1/x, x > 0} ∪ {(0, y) : |y| ≤ 1}. (Hint: Proposition 6.2.4).
(b) The Warsaw Circle Y = {(x, y) ∈ R2 : y = sin 1/x, 0 < x < 3π 1
}∪
A ∪ B, where A is an arc joining ( 3π , 0) to (0, 0) lying in the lower
1
plane and B = {(0, y) | |y| ≤ 1} (Fig. 6.9).

Fig. 6.9: The Warsaw Circle
7. Show that the boundary ∂M of the Möbius band M is not a retract of M .

Give a map f from M to the middle circle C of M which is a deformation
retraction, i.e., there is a homotopy starting from the identity map of M
and ending with the retraction f of M onto C.
8. Let Δq denote the q-simplex and for 0 ≤ m < q, let Δq (m) denote the
m-skeleton of Δq . Compute the singular homology groups of Δq (m) (see
Ex. 8, Section 4.4).
9. Let X be the quotient space obtained from the triangle ABC by identi-
fying AB with BC, BC with CA and CA with AB by linear homeomor-
phisms. Compute the singular homology groups of the space X (see Ex.
7, Section 4.4).
10. Use the above exercise to show that given any m ≥ 1 and any n > 0,
there is a compact polyhedron X such that the reduced singular homology
H̃q (X) = Zm when q = n, and is 0, otherwise.
11. Use the preceding exercise to prove that given any finite sequence
{G0 , G1 , . . . , Gn } of finitely generated abelian groups in which G0 is free,
there exists a compact polyhedron X such that the reduced singular ho-
mology groups H̃q (X) ∼ = Gq for all q ≥ 0.
12. Let X be the space of a 2-sphere with a circle touching the sphere at a
point (see Fig. 6.10). Compute the singular homology of X.
X:
Fig. 6.10: A 2-sphere with a tangent circle
13. Compute the singular homology of a two-holed torus, i.e., a 2-sphere with
two handles (see Fig. 6.11) or the connected sum of two tori.
14. Prove that for any k with 0 < k < n:
Fig. 6.11: A two-holed (double) torus
(i) Sk ⊂ Sn cannot be a retract of Sn .

(ii) RPk ⊂ RPn cannot be a retract of RPn .
(iii) CPk ⊂ CPn cannot be a retract of CPn .
Chapter 7
Appendix
7.1 Basic Algebra – a Review
Here we present some basic definitions and a few results from elemen-
tary algebra and homological algebra which are required in introducing and
discussing the concepts of algebraic topology. We assume that the reader is
familiar with the fundamentals of groups and modules from where we recall
only essential portions for completeness.
7.1.1 Groups and Homomorphisms
A set G with an associative binary operation is called a semigroup. A

semigroup G is said to be a group if there is a two-sided identity element e in
G, and every element a ∈ G has an inverse a−1 ∈ G. It is readily verified that
identity element of a group G is unique, and that inverse of each element in
G is also unique. We will, for convenience only, denote the product a.b of two
elements of G just by ab. A group G is called abelian if ab = ba for each a, b ∈ G.
A map f : G → G from a group G to another group G is said to

be a homomorphism if it preserves binary operations in G and G , i.e.,
f (ab) = f (a)f (b) for all a, b ∈ G. One can easily prove that f preserves the
identity element and also the inverse of each element, i.e., f (e) = e , where
e ∈ G and e ∈ G are identity elements and (f (a))−1 = f (a−1 ) for each
a ∈ G. A bijective homomorphism is called an isomorphism. A homomorphism
f : G → G from a group G to itself is called an endomorphism and a bijective
endomorphism is said to be an automorphism.
A subset H of a group G, which is closed under the binary operation of

G, is called a subgroup of G if H is a group with respect to this induced binary
operation. A group G is said to be finitely generated if there exists a finite
subset S = {g1 , g2 , . . . , gk } of G such that every element of G can be written
296 Chapter 7. Appendix
as a product of powers of gi ∈ S, i.e., for each a ∈ G, there exist integers

n1 , n2 , . . . , nk and a permutation σ on the set {1, 2, . . . , k} such that
n1 n2 nk
a = gσ(1) gσ(2) . . . gσ(k) .
We point out that when the group G is abelian, the above expression for
a ∈ G is usually written as a = n1 gσ(1) + n2 gσ(2) + . . . + nk gσ(k) . If S generates
G, we write G = [g1 , . . . , gk ]. A group G generated by a single element is said
to be a cyclic group. In other words, a group G is cyclic if there exists a g ∈ G
such that G = [g]. It is well-known that a cyclic group G is isomorphic to
either the additive group Z of integers or to the additive group Zm of residue
classes modulo m for some m ∈ N.
A subgroup N of a group G is called a normal subgroup if gng −1 ∈ N

for each g ∈ G and for each n ∈ N . This is equivalent to say that gN g −1 ⊆ N
for each g ∈ G. If N is normal in G then, for each g ∈ G, left cosets and right
cosets of N in G are identical, i.e., gN = N g for all g ∈ G. Let N be normal
in G and let G/N denote the set of all cosets (right or left – it is the same
thing) of N in G. Then we can define a product in the set G/N by setting
(g1 N )(g2 N ) = (g1 g2 )N . Note that this product in G/N is, in fact, induced by
the product of G and is well-defined since N is normal. Then one can easily
verify that, with this operation, the set G/N becomes a group with eN = N as
the identity element. This group is called a factor group (or quotient of G by
N ) of G by the normal subgroup N . The following is a basic result connecting
normal subgroups of G and kernels of homomorphisms emanating from G:
Theorem 7.1.1. (Fundamental Theorem of Homomorphism). Let f : G → G

be a surjective homomorphism and K = ker f . Then K is a normal subgroup
of G and G/K ∼ = G .
The map f : R → S1 , where S1 is the unit circle in the complex plane,

defined by f (x) = e2πix , is a surjective homomorphism with group Z of in-
tegers as its kernel. Hence, by the Fundamental Theorem of Homomorphism,
R/Z ∼ = S1 . As another interesting example, let GL(n, R) denote the multi-
plicative group of all n × n nonsingular real matrices, n ≥ 1. Then the map
f : GL(n, R) → R∗ = R − {0} (the multiplicative group of nonzero reals), de-
fined by f (A) = det A, is a surjective homomorphism whose kernel is the group
SL(n, R) of real matrices with determinant 1. Hence, by the Fundamental The-
orem of Homomorphism,
GL(n, R)/SL(n, R) ∼
= R∗ .
Exercises
1. Let n ≥ 1 and GL(n, R) be the set of all n × n non-singular matrices over

reals. Prove that GL(n, R) is a group with respect to multiplication of
matrices. Identify the group GL(1, R).
7.1. Basic Algebra – a Review 297
2. Show that the additive group Q of rationals is not finitely generated.

3. Let G be a semigroup. Suppose it has a left identity, and every element of
G has a left inverse in G. Then show that G is a group. (A similar result
also holds assuming right identity and right inverses.)
7.1.2 Direct Product and Direct Sum

Let G1 , G2 be two groups and consider the cartesian product G1 × G2 of the
two sets G1 and G2 . Let us define a binary operation in G1 × G2 as follows:
(a1 , b1 ) · (a2 , b2 ) = (a1 a2 , b1 b2 ).
Then G1 × G2 is a group with (e1 , e2 ), where e1 ∈ G1 , e2 ∈ G2 are identity ele-

ments, as the identity element. The resulting group is called the direct product
of G1 and G2 . We have projection maps p1 : G1 × G2 → G1 , p2 : G1 × G2 → G2
defined by p1 (a, b) = a, p2 (a, b) = b for (a, b) ∈ G1 × G2 which are
surjective group homomorphisms. We have also the inclusion maps
i1 : G1 → G1 ×G2 , i2 : G2 → G1 ×G2 defined by i1 (a) = (a, e2 ), i2 (b) = (e1 , b).
These are injective homomorphisms. One can easily see that all these homo-
morphisms satisfy the following properties: p1 ◦i1 = IG1 , p2 ◦i2 = IG2 , ker p1 =
{e1 } × G2 , ker p2 = G1 × {e2 }. It follows that ker p1 = Im i2 , ker p2 = Im i1
and G2 ∼ = (G1 × G2 )/Im i1 , G1 ∼ = (G1 × G2 )/Im i2 . One can clearly extend
all of the above and define the direct product G1 × · · · × Gn of a finite
numbern of groups. The product here will also be coordinate-wise. Projections
pk : i=1 Gi → Gk , defined by pk (a1 , a2 , . . . , an ) = ak , and the inclusions
n
ik : Gk → 1 Gi defined by ik (ak ) = (e1 , . . . , ak , . . . , en ) (the element ak at
the kth place and identities elsewhere), are homomorphisms and these will
also satisfy appropriate analogous conditions.

Let {Gi | i ∈ I} be
a family of groups. Recall that an element f ∈ i∈I Gi
is just a map f : I → i∈I Gi such that f (i) ∈ Gi for all i ∈ I. Therefore, one
can define a product in Gi by putting for all i ∈ I,
(f · g)(i) = f (i) · g(i),

where f, g ∈ Gi . Note that
this is again a coordinate-wise product and, if
we write the element f ∈ Gi as f = (f (i))i∈I as an I-tuple, then the above
product can be expressed as
(f (i)) · (g(i)) = (f (i) · g(i)).

The new group i∈I Gi with the above operation is called the direct product
of the groups {Gi | i ∈ I}. In this general situation also we have projection
and inclusion homomorphisms which satisfy the analogous properties stated
earlier for finite products.
Now, we come to the important concept of direct sum of a given family

{Gi | i ∈ I} of groups and we assume that ei ∈ Gi denotes the identity
element
of Gi . For
this we consider the following subset, denoted by i∈I G i , of the
product i∈I Gi ;

⊕Gi = {(gi ) ∈ Gi | gi = ei ∈ Gi for all i except finitely many indices}.
Note that ⊕i∈I Gi defined above is a subgroup of the direct product. This
subgroup is called the direct sum of the family {Gi | i ∈ I} of given groups.
It must also be noted that if the indexing set I is finite, then the direct sum
⊕Gi coincides with direct product. Another important observation is that if
elements of each group Gk ,k ∈ I, are identified with their images under the
inclusion maps ik : G k → k∈I Gk , i.e., gk ∈ Gk is treated as the element
(. . . , el , gk , em , . . .) ∈ Gk where el ∈Gl is the identity element of Gl for all
l
= k, then an arbitrary element g ∈ Gk can be uniquely written as
g = gk1 gk2 . . . gkp ,
where gki ∈ Gki are the nonidentity components of the element g. Using the
additive notation for each Gk and for Gk , the above expression is written as:
g = gk1 + gk2 + . . . + gkp , gki ∈ Gki .
n
The case of a finite product k=1 Gk of n groups G1 , G2 , . . . , Gn (n ≥ 2)
requires some further
n discussion: for each k = 1, 2, . . . , n, we have the inclusion
maps ik : Gk → 1 Gk defined by ik (g) = (e1 , . . . , g, . . . , en ), where g is at the
kth place. Let us denote the image set ik (Gk ) as Ḡk . Then it can be proved
that these image sets satisfy the following conditions:

(i) each Ḡk is a normal subgroup of Gk .
n n
(ii) the product group 1 Gk can be written as 1Gk = Ḡ1 · Ḡ2 · · · Ḡn , i.e.,
the union of subgroups Ḡk generate the group Gk .
Ḡk ∩ [Ḡ1 ∪ · · · ∪ Ḡk−1 ∪ Ḡk+1 ∪ · · · ∪ Ḡn ] = {e}, the identity
(iii) for each k,
element of Gk , i.e., each Ḡk intersects the subgroup generated by the
rest of them only in the identity element.
It is interesting to observe that the converse of above result is also true.

To be precise, suppose a group G has n subgroups H1 , H2 , . . . , Hn such that
these subgroups satisfy the three conditions stated above for Ḡk . Then it can
be proved that G is isomorphic to the direct product H1 × · · · × Hn of these
subgroups. We come across the above situation quite often and the above result
then better clarifies the structure of the group G.
7.1.3 The Structure of a Finite Abelian Group
As an important illustration of the foregoing result, let us consider a finite

abelian group G of order n. In this case, the structure of G can be clarified
completely. We write n = pk11 · · · pkr r as a product of positive primes. By Sylow’s

Theorem, the group G has a pi -Sylow subgroup Pi of order pki i for all i =
1, 2, . . . , r. The group G being abelian, each Pi is a normal subgroup of G, and
clearly, Pi ∩ [P1 ∪ · · · ∪ Pi−1 ∪ Pi+1 ∪ · · · ∪ Pr ] = {e} for each i. Hence, from what
we said earlier, G is isomorphic to the direct product of its pi -Sylow subgroups
Pi . The case when all the exponents are 1 is interesting. Since any group of
prime order is cyclic, it follows from the above decomposition that an abelian
group of order p1 p2 · · · pr , where each pi is prime, is cyclic. For understanding
the general case, let us recall (I.N. Herstein, Topics in Algebra, p. 113) that
if P is a p-group, p a prime, then P is isomorphic to the direct sum of cyclic
groups, viz.,
P ∼
= Z(pl1 ) ⊕ Z(pl2 ) ⊕ . . . ⊕ Z(plm ) ,
where l1 ≥ l2 ≥ . . . ≥ lm are uniquely determined by the group P ; here, Z(pl )
denotes the cyclic group of order pl . Uniqueness means if P is also isomorphic
to the direct sum
Z(pr1 ) ⊕ Z(pr2 ) ⊕ · · · ⊕ Z(prn )
with r1 ≥ r2 ≥ · · · ≥ rn , then n = m and l1 = r1 , . . . , ln = rn . In fact, these
numbers {l1 , l2 , . . . , ln } are complete invariants of the p-group P . Now, let us
go back to the abelian group G with its pi -Sylow subgroups Pi ; i = 1, 2, . . . , r.
Then we have already noted that
G∼
= P1 ⊕ . . . ⊕ Pr .
Next, we write each Pi ∼ = Zpi r1 (pi ) ⊕ . . . ⊕ Zpi rt (pi ) , r1 ≥ r2 ≥ . . . ≥ rt ≥ 0, for

i = 1, 2, . . . , r; each ri depends on pi . If we put
d1 = pr11 · pl21 · . . . · psr1
d2 = pr12 · pl22 · . . . · psr2

..
.
dt = pr1n · pl2n · . . . · psrn ,
where some of these exponents could be zero, then each di is divisible by di+1
and the group G itself can be written as a direct sum of cyclic groups uniquely.
To see this, we note that pr11 , . . . , psr1 are all relatively primes and, since the
direct sum of finite cyclic groups of relatively prime orders is again cyclic, we
find that
G∼ = Zd1 ⊕ Zd2 ⊕ . . . ⊕ Zdt ,
where d1 is divisible by d2 , d2 is divisible by d3 , . . . , dt−1 is divisible by dt .
These numbers d1 , d2 , . . . , dt , which are uniquely determined by the abelian
group G, are called torsion coefficients of G. In fact, two finite abelian groups
are isomorphic if and only if they have the same torsion coefficients. In this
way the structure of a finite abelian group is completely determined.
7.1.4 Free Groups and Free Products
The notion of a free group is basic to the study of fundamental groups discussed
in Chapter 2. We prefer to introduce a free group by its universal property.
Defining a concept using the universal property is not only quicker but is also
efficient in proving other results about the concept itself. However, it has the
disadvantage that nothing is clear about the inner structure of the object being
defined: in fact, all of that insight is relegated to the existence theorem of the
object itself. We have
Definition 7.1.2. By a free group F on a given set S (called a set of generators)

we mean a group F and a function φ : S → F which satisfies the following
universal property: For any group H and any function ψ : S → H, there exists
a unique group homomorphism f : F → H such that the following diagram is
commutative:
φ -F
S
HH
HH
ψ H f
j ?
H
H
i.e., f ◦ φ = ψ.
Let F = [x] be the infinite cyclic group generated by a single element x. In

this case, we know that F consists of all powers xn , n ∈ Z, of x. Then it is readily
seen that F is free group on the set S = {x} with inclusion map i : S → F as the
map φ. On the other hand, the cyclic group Zm of order m can never be a free
group on any set S. Note that, up to isomorphism, there is a unique free group
on a given set S. To see this, let F and F be two groups on S via the maps
φ : S → F and φ : S → F . Then, by the defining universal property, we have
a unique homomorphism f : F → F and a unique homomorphism g : F → F
which make the following two triangles commutative, i.e., f ◦ φ = φ , g ◦ φ = φ.
F
φ
φ
*
f S -F
HH
?
φ - F
HH
φ H IF
S
HH Hj ?
HH F
g
φ H
H
j ?
F
It follows that the homomorphism g ◦ f : F → F makes the bigger triangle
commutative, i.e., g ◦ f ◦ φ = φ. But the identity map IF : F → F already
makes the bigger triangle commutative. Hence, by the uniqueness of the
defining homomorphisms, we must have g ◦ f = IF . By a similar argument,
we find that g ◦ f = IF . Hence, it follows that each of the maps f and g is an
isomorphism and so F ∼
= F .
In order to discuss the question of existence of free groups, we need the

concept of free products of a family {Gi | i ∈ I} of groups. This too can be
defined by its universal property. We have
Definition 7.1.3. Let {Gi | i ∈ I} be a family of groups. Then a group G together

with a family of homomorphisms φi : Gi → G, i ∈ I is called free product of
the family {Gi | i ∈ I} if the following universal property is satisfied : for any
group H and any family of homomorphisms ψi : Gi → H, there exists a unique
homomorphism f : G → H such that every triangle
φi -G
Gi
HH
HH
ψi H f
j ?
H
H
is commutative, i.e., f ◦ φi = ψi for all i ∈ I.
The uniqueness of the free product of a given family {Gi | i ∈ I} of groups

follows from the universal property defining the free product. We omit proof of
the following existence theorem (see W. Massey [13], p. 98).
Theorem 7.1.4. Given any collection {Gi | i ∈ I} of groups, their free product
exists.
It is also true that each homomorphism φi : Gi → G occurring in the

definition of free product is indeed a monomorphism and the free product is
generated by the union of their images in G. Wedenote the free product of
∗
groups G1 , G2 , . . . , Gn by G1 ∗ G2 ∗ . . . ∗ Gn = 1≤i≤n Gi whereas the free
∗
product of the family {Gi | i ∈ I} is denoted by i∈I Gi .
Example 7.1.5. Let G1 = {1, x1 }, G2 = {1, x2 } be two cyclic groups, each of

order 2. Then their free product consists of the following elements
x 1 , x 1 x 2 , x 1 x 2 x1 , . . . or, x2 , x2 x1 , x2 x1 x2 , . . . .
The elements x1 x2 and x2 x1 are different and each one is of infinite order. Note
that the direct product G1 × G2 of the two groups consists of only 4 elements
whereas the free product G1 ∗ G2 consists of infinite number of elements, each
of infinite order.
Now, we come back to the question of existence of a free group on a given

set S. Let us index the set of elements of S as S = {xi | i ∈ I} and let Fi = [xi ]
be the free group on one generator
∗ xi – this would be the infinite cyclic group
generated by xi . Now, if F = i∈I Fi denotes the free product of these groups
Fi , then it can be seen that F is indeed a free group on the set S. This proves
that given any set S, there always exists a free group on S and S is a set of
generators for F . As an example, let S = {x1 , x2 } have just two elements. Then
the free group F [S] on S will have elements of the type
xn1 1 xn2 2 xn1 3 xn2 4 . . . xn1 k x2nk +1 or, xm 1 m2 m3 m4 ml ml +1
2 x 1 x 2 x 1 . . . x 2 x1 ,
where ni and mi are nonzero integers (compare with the preceding example).
These elements are called reduced “words” in symbols x1 and x2 . The crucial
point is that there are no relations among the powers of x1 and x2 , i.e.,
none of the elements written above ever equals identity element. Also, note
that the free group on a set S having just one element is abelian whereas
the free group on a set S having more than one element is never abelian.
Furthermore, it follows from the universal property that two free groups on the
sets S1 and S2 are isomorphic if and only if S1 and S2 have the same cardinality.
Let G be a group with S as a set of generators. Consider the free group

F on the set S with respect to a map φ : S → F and let ψ : S → G, where G is
seen as a group, be the inclusion map. Then, by the universal property of F ,
there exists a homomorphism f : F → G such that f ◦ φ = ψ. This means f is
onto and so, by the Fundamental Theorem of Homomorphism, G ∼ = F/ ker f .
This proves the useful fact that every group G is isomorphic to a quotient of
a free group.
When we consider a group G as the quotient of a free group F , as

explained above, the elements of ker f in F are finite products of powers of
elements of S (a reduced word). But when the same element is seen in the
group G via the inclusion map ψ then each of them reduces to an identity of
G. Each of the elements in the ker f is called a relation defining the group
G. If the group G is described as the quotient of a free group as above, we
say that G is given in terms of generators and relations. The elements of
S are generators and the elements of ker f are relations for G. It may be
observed that if r1 , r2 are relations in G, then their product, their inverses,
their conjugates in G are all relations for G. A set R of relations which
generates the ker f as the smallest normal subgroup containing ker f is called
a complete set of relations for G. A group G is completely described by a set
S of generators and complete set R of relations.
A group G is said to be finitely presented if the sets S and R describing

G as above are finite. We write this as G = S : R.
Example 7.1.6. 1. A cyclic group Zn of order n has a presentation x : xn .

2. A free abelian group of rank two, Z × Z, has a presentation x, y :
xyx−1 y −1 .
3. The dihedral group D2n having 2n elements has a presentation a, b :
an , b2 , (ab)2 .
Free Abelian Groups
The concept of a free abelian group is needed in defining various kinds of

homology groups. The definition uses the universal property analogous to the
one already given for free groups. We have
Definition 7.1.7. Let S be a set. By a free abelian group F on the set S we mean
an abelian group F together with a map φ : S → F which has the following
universal property: For any abelian group A and for any map ψ : S → A, there
exists a unique homomorphism f : F → A such that the triangle
φ -F
S
HH
HH
ψ H f
j ?
H
A
is commutative, i.e., f ◦ φ = ψ.
The question of uniqueness of the free abelian group F on S is answered

exactly as in the case of free groups discussed earlier. However, the existence
question in the case of free abelian group is answered by the concept of direct
sum instead of free product: Let S = {xi | i ∈ I} be the given set and let
Fi = Zxi be the infinite cyclic group on the generators xi , i ∈ I. Then the
direct sum F = i∈I Fi , with the map φ : S → F as the obvious inclusion
map, is the free abelian group on S. As pointed out in the definition of direct
sum, every element x ∈ F can be expressed (in additive notation) uniquely:

x= n i xi , xi ∈ S, ni ∈ Z,
as a finite sum. This explains the easy structure of a free abelian group on a
set S of generators of F .
It is easily shown that every abelian group is isomorphic to a quotient of

a free abelian group. If F is free abelian on the set S, then the cardinality of
the set S is called rank of F . It is then readily verified that two free abelian
groups are isomorphic if and only if both have the same rank.
Exercises
1. Let G = Zp be a cyclic group of order p, p a prime. Show that the group

Aut(G) ∼
= Zp−1 .
2. Let Q = (Q, +) be the additive group of rationals. Prove that Aut(Q) ∼ =
Q∗ , the multiplicative group of non-zero rational numbers. Hence conclude
that the only nontrivial finite subgroup of Aut(Q) is Z2 .
3. Show that a subgroup of a free group is free (This is the important Nielsen-
Schreier Theorem).
4. Show that a subgroup H of a free abelian group F is free abelian. Also,
prove that rank H ≤ rank F (This result need not be true in the case of
free group, Ex 3).
5. Let G = Zm and G = Zn (m, n > 1). Show that the free product G ∗ G is
not finite and the only elements of finite order in G ∗ G are the elements
of G, G or their conjugates.
7.1.5 Modules and their Direct Sum
Let R be a commutative ring with unity element 1

= 0. An abelian group M
with a multiplication R × M → M (called scalar multiplication) is said to be
an R-module if the following conditions are satisfied:
(i) (r1 + r2 ) · m = r1 · m + r2 · m
(ii) r · (m1 + m2 ) = r · m1 + r · m2
(iii) (r1 r2 ) · m = r1 · (r2 · m)
(iv) 1 · m = m
for each r1 , r2 , r ∈ R and m1 , m2 , m ∈ M . If R = F is a field, then a
module over R is just a vector space over F . An abelian group A is, clearly, a
Z-module. An ideal I of a ring R is an R-module. In particular, a ring R is a
module over itself. Submodules and quotient modules are defined analogous to
subgroups and quotient groups.
If M, M are two modules over R, then a homomorphism f : M → M

of abelian groups is said to be an R-homomorphism if f (r · m) = r · f (m)
for all r ∈ R, m ∈ M , i.e., f preserves scalar multiplication. Isomorphisms,
endomorphisms, automorphisms of modules are defined analogous to the
corresponding notions in groups. The fundamental theorem of homomorphism
for R-modules is also valid.
Let {Mα | α ∈ I} be a family of R-modules. Consider the product

Mα as the direct product of abelian groups and
α∈I define scalar multipli-
cation in this product as r · (mα ) = (r · mα ). Then α∈I Mα is an R-module
and is called the direct productof the given family of R-modules. Now, let us
consider a subset, denoted by α∈I Mα or Σα∈I Mα , of this product module
defined by

Mα = {(xα ) ∈ α∈I Mα | xα = 0 for all but finitely many indices α ∈ I}.

Then it is evident that Mα is a submodule of the product module α∈I Mα .
This module is called the direct sum of the given family {Mα | α ∈ I} of
R-modules. If the indexing set I is finite, then note that the direct sum

Mα coincides with the direct product Mα . A module M is said to
be free if M ∼ = ⊕α∈I Mα , where for each α ∈ I, Mα ∼ = R as R-module. It
is not difficult to prove that any R-module M is the quotient of a free R-module.
A module M over R is said to be finitely generated if there exists a finite

set X = {x1 , x2 , . . . , xn } of elements of M such that every element x ∈ M
can be written as a linear combination of elements of X, i.e., given m ∈ M
there are elements r1 , r2 , . . . , rn of R such that m = r1 x1 + . . . + rn xn . A
module M is said to be cyclic if it is generated by a single element x ∈ M ,
i.e., M = R · x = (x). This generalizes the concept of cyclic groups. A finite
set X = {x1 , x2 , . . . , xn } of elements of M is said to be a basis for M if
every element of M can be expressed as a linear combination of elements
x1 , x2 , . . . , xn in a unique way. The set B = {e1 , . . . , en } of elements of
Rn = R ⊕ . . . ⊕ R, where ei = (0, . . . , 1, . . . , 0) (1 at the ith position) for
i = 1, 2, . . . , n, is a basis of the R-module Rn . A finitely generated R-module
can be expressed as a quotient of the free R-module M = Rn for some n ≥ 1.
An R-module, even though it is finitely generated, may not have a basis,
e.g., the abelian group Zm (m > 1) is cyclic over Z but does not have a basis
because the element 2̄ = 2 · 1̄ = (m + 2).1̄ can be expressed in different ways.
Let M be an R-module, where R is a domain. An element m ∈ M

is called a torsion element of M if there exists an r(
= 0) ∈ R such that
r · m = 0. The set of all torsion elements of M form a submodule of M
and is called the torsion submodule of M . An R-module M is said to be
torsion free if torsion M = {0}. Note that a torsion free module need not
be free: the abelian group Q of rationals is torsion free over Z, but is not
free over Z. What about a finitely generated torsion free module? A module
over a field R, i.e., a vector space is always free and has a basis, finite or infinite.
Let M1 , M2 , . . . , Mn be submodules of an R-module M . Suppose these

modules satisfy the condition that every element of M can be uniquely ex-
pressed as
m = m1 + m2 + · · · + mn ,
where mi ∈ Mi , i = 1, 2, . . . , n. Then one can easily prove that M = ∼ M1 ⊕
M2 ⊕ · · · ⊕ Mn . The following fundamental result is very important. For a proof
of this one can see (See N. Jacobson [11] p. 187).
Theorem 7.1.8. (Structure Theorem for finitely generated Modules). Let R

be a PID (Principal Ideal Domain). Then a finitely generated R-module M is
isomorphic to the direct sum of a finite number of cyclic R-modules.
Structure Theorem for Finitely Generated Abelian Groups

As a corollary to the above theorem, it follows that a finitely generated
abelian group A can be expressed as a direct sum of finite number of cyclic
groups. More precisely, for such an A, there exist integers n ≥ 0, k ≥ 0 and

r1 , r2 , . . . , rk such that
n copies

∼
A = Z ⊕ · · · ⊕ Z ⊕Zr1 ⊕ Zr2 ⊕ · · · ⊕ Zrk ,
where r1 is divisible by r2 , r2 is divisible by r3 , etc. If A has such a decom-

position then the number n is called the rank or Betti number of A and the
elements of the set {r1 , r2 , . . . , rk } are called the torsion coefficients of A. In
fact, the rank and the torsion coefficients of A completely characterize the
finitely generated abelian groups A. The subgroup T = Zr1 ⊕ · · · ⊕ Znk of A
is called the torsion subgroup of A. Note that the factor group A/T is free
abelian group and A ∼ = (A/T ) ⊕ T . The factor group A/T is called the free
part of A. It is now clear that the two finitely generated abelian groups A and
A are isomorphic if and only if their free parts as well as their torsion parts are
isomorphic. This will happen if and only if their ranks and torsion coefficients
are the same.
7.2 Categories and Functors
The concept of a category, functor and several accompanying ideas were

discovered and formalized by S. Eilenberg and S. Mac Lane around the year
1942. They have provided a powerful language to mathematics in achieving
unity of concepts and expression. These have proved to be of immense impor-
tance in stating and formulating diverse ideas involving definitions and results,
which appeared very different from each other, into a nice whole. Some of these
have even brought out new and deeper interpretations of known results which
were earlier hidden underneath. Using this new language, dozens of known the-
orems can be stated and proved as a single result. Here, we will give only the
definition of categories and functors and a few examples. The interested reader
is referred to (Homology : S. Mac Lane [17]) for a fuller account.
Categories
A category C consists of three items:

(a) a class of objects, to be denoted by letters A, B, C, . . ., etc., which can be
thought of as sets with some additional structures on them,
(b) for each pair (A, B) of objects in C, a set Mor(A, B), called the set of
morphisms f : A → B, which may be thought of as maps preserving
additional structures, and
(c) a product Mor(A, B) × Mor(B, C) → Mor(A, C) for each triple (A, B, C)
of objects in C, called composition and taking the pair (f, g) to g ◦ f ∈
Hom(A, C).
7.2. Categories and Functors 307
These are required to satisfy the following two axioms:

(i) For any three morphisms f : A → B, g : B → C and h : C → D, we have
h ◦ (g ◦ f ) = (h ◦ g) ◦ f,
i.e., the composition of morphisms is associative.

(ii) For each object A in C, there exists a morphism IA : A → A, called identity
morphism, which has the property that for all f : A → B and all g : C →
A, we have
f ◦ IA = f, IA ◦ g = g.
Some examples of categories are given below:

(1) If we take objects as topological spaces, morphisms as continuous maps
and the product to be composition of continuous maps, then we get a
category, called the category of topological spaces and continuous maps.
(2) Let us take groups as objects, homomorphisms from one group to another
as morphisms and the composition of homomorphisms to be the com-
position of morphisms. Then we have the category of groups and group
homomorphisms.
(3) The category of sets and maps has objects as the class of all sets, mor-
phisms as just maps and the product as the composition of maps.
(4) If we let the objects be the class of all simplicial complexes, the morphisms
to be the set of all simplicial maps and the product to be the composition
of simplicial maps, then we have the category of simplicial complexes and
simplicial maps.
(5) If we take objects as finite groups, morphisms as group homomorphisms
and the product to be the composition of homomorphisms, then again
we get a category. This category is in fact a subcategory of the category
of all groups and homomorphisms – one can define, in a natural way, a
subcategory of a category.
There are numerous important categories which we often come across,

especially in algebraic topology. The category of all R-modules and module
homomorphisms for a given ring R is an important category for homological
algebra as well as for commutative algebra. The class of all sets with morphisms
as injective maps is a subcategory of the category of sets and maps. Note that
in this category, the set Mor(A, B) between sets A and B are clearly small as
compared to the set Mor(A, B) in the category of sets. Thus, one can see that
if C is a subcategory of D, then the class of objects of C may be quite small
compared to that of D. Also, the Mor(A, B) between any two objects A and
B of C could be smaller or even equal to that for the category D.
A morphism f : A → B is called an equivalence (or isomorphism) if there

is a morphism g : B → A such that g ◦ f = IA and f ◦ g = IB . Note that an
equivalence in the category of sets is just a bijective map; an equivalence in the
category of topological spaces is a homeomorphism and an equivalence in the
category of R-modules is a module isomorphism.
Functors
A covariant functor F from a category C to a category D is a pair of func-
tions (both denoted by the same letter F ) which maps objects of C to the
objects of D and, for any pair (A, B) of C, it maps the set Mor(A, B) to the
set Mor(F (A), F (B)) and is required to satisfy the following two conditions:
(i) F (IA ) = IF (A) for every A ∈ C.
(ii) F (g ◦ f ) = F (g) ◦ F (f ) : F (A) → F (C) for any two morphisms f : A → B
and g : B → C in the category C.
A contravariant functor F from a category C to the category D is similarly

defined. The only difference is that the contravariant functor maps a morphism
f : A → B to a morphism F (f ) : F (B) → F (A) in the opposite direction to
that of f. Thus, the two conditions in this case will be
(i) F (IA ) = IF (A) for all A ∈ C.
(ii) F (g ◦ f ) = F (f ) ◦ F (g) : F (C) → F (A).
Example 7.2.1. Let C be the category of finite sets and maps between them as
morphisms. Let D be the category of free abelian groups having a finite bases
and morphisms as group homomorphisms between them. For each object X
in C, let F (X) be the free abelian group with X as a basis and, for any map
f : X → Y in the category C, let F (f ) : F (X) → F (Y ) be the abelian group
homomorphism defined by the map f (any set theoretic map f from a basis of
a free abelian group G defines a unique homomorphism from the free abelian
group G). Then it can be easily verified that F is a covariant functor from the
category C to the category D.
Example 7.2.2. For each topological space X, let C(X) denote the ring of con-
tinuous real-valued functions on X with point-wise operations. For any contin-
uous map f : X → Y, we get a ring homomorphism f ∗ : C(Y ) → C(X) defined
by f ∗ (α) = α ◦ f. Then one readily verifies that we get a contravariant functor
from the category of topological spaces and continuous maps to the category
of rings and homomorphisms.
Natural Transformations
Let F and G be two covariant functors from a category C to the category
D. Then a map η, which assigns to each object A ∈ C a unique morphism
ηA : F (A) → G(A), is called a natural transformation from F to G, if for each
f : A → B in C, the diagram
F (f )
F (A) F (B)
ηA ηB
G(A) G(B)
G(f )
is commutative. A natural transformation between contravariant functors

can be similarly defined. A natural transformation η : F → G is called an
equivalence if ηA is an equivalence in the category D for every object A in C.
Example 7.2.3. Let C be the category of all vector spaces (over a field F ) and
linear transformations as morphisms. Let S and T be two functors from C to
C defined by
S = IC and T (V ) = V ∗∗ ,
where V ∗∗ is the double dual of V. For each linear map f : V → W in V, let
T (f ) = f ∗∗ : V ∗∗ → W ∗∗
be the homomorphism induced in the double dual, i.e.,
f ∗∗ (φ)(α) = φ(α ◦ f ) for all α ∈ W ∗ and φ ∈ V ∗∗ .
Now, if for any V ∈ C, we define ηV : V → V ∗∗ by
ηV (x)(α) = α(x),
∗
where x ∈ V and α ∈ V , then we get a natural transformation from the
functor S to the functor T. This follows since, for any linear map f : V → W,
the diagram
f
V W
ηV ηW
V ∗∗ W ∗∗
f ∗∗

is easily seen to be commutative. If we take the subcategory C of C having
only finite-dimensional vector spaces, then ηV : V → V ∗∗ is a vector space
isomorphism and hence η is, in fact, a natural equivalence from the identity

functor S to the double dual functor T on C .
We come across numerous examples of categories, functors and natural

transformations when we study different topics of algebraic topology. In fact,
it is sometimes said “algebraic topology is the study of a few functors!!”.
7.2.1 The Hom(M, N ) Functor
Here, we begin with the remark that the material being discussed from this
point onward is indeed a portion of elementary Homological Algebra. Let M, N
be R-modules and let HomR (M, N ) be the set of all R-homomorphisms from
M to N . Then we can define an addition f + g and a scalar multiplication r · f
in the set HomR (M, N ) as follows:
(f + g)(m) = f (m) + g(m)
(rf )(m) = r · f (m)
for all m ∈ M and for all r ∈ R. It is easy to verify that the set HomR (M, N )
becomes an R-module w.r.t. the above operations. This module will be
denoted by just Hom(M, N ) whenever there is no confusion. Note that if
α : M → M is any R-module homomorphism, then this induces an R-module
homomorphism α∗ : Hom(M , N ) → Hom(M, N ), in a direction opposite to α,
by the formula α∗ (f ) = f ◦ α. Similarly, a homomorphism β : N → N induces
an R-homomorphism β∗ : Hom(M, N ) → Hom(M, N ), in the direction of β,
by the formula β∗ (f ) = β ◦ f .
With above definitions, one can easily verify the following properties of
induced homomorphisms:
(i) If α1 : M → M and α2 : M → M are any two homomorphisms, then
(α2 ◦ α1 )∗ = α1∗ ◦ α2∗ .
(ii) If IM : M → M is the identity homomorphism, then the induced map
∗
IM : Hom(M, N ) → Hom(M, N ) is also the identity map.
Likewise, if β1 : N → N and β2 : N → N are two homomorphisms,

then (β2 ◦ β1 )∗ = β2 ∗ ◦ β1 ∗ and if IN : N → N is the identity map, then
(IN )∗ : Hom(M, N ) → Hom(M, N ) is also the identity map.
Recall that if f : M → N is an R-homomorphism, then ker f = {x ∈ M |

f (x) = 0}, Im f = f (M ) and coker f = N/Im f are all R-modules. Using the
Fundamental Theorem of Homomorphism for R-modules, one can easily prove
the following facts:
(i) If L ⊇ M ⊇ N are R-modules, then (L/N )/(M/N ) ∼
= L/M .
(ii) If M1 , M2 are submodules of M , then (M1 + M2 )/M1 ∼
= M2 /(M1 ∩ M2 )
(Noether Isomorphism Theorem).
7.2.2 Exact Sequences
Let us consider the following sequence (finite or infinite) of R-modules and

R-homomorphisms
fn+1 fn
. . . −→ Mn+1 −→ Mn −→ Mn−1 −→ . . . .
We say that this sequence is exact at Mn if Im fn+1 = ker fn . It is said to be
exact if it is exact at Mn for each n. One can easily verify that
(i) A homomorphism f : M → M is 1-1 if and only if the sequence 0 →
f
M → M is exact.
f
(ii) A homomorphism f : M → M is onto if and only if the sequence M →
M → 0 is exact.
α β
(iii) A sequence 0 → M → M → M → 0 is exact if and only if α is 1-1, β is
onto and Im α = ker β, i.e., M ∼
= M/α(M ).
α β
An exact sequence 0 → M → M → M → 0 is said to be split exact
if there exists a map γ : M → M such that βγ = IM . In such a case
M ∼ = M ⊕ M . This, of course, already implies that M ∼
= M/α(M ) and is
indeed stronger than (iii). A sequence of the kind (iii) above, consisting of five
terms is called a short exact sequence.
The Hom functor is left exact
Concerning the exactness of the contravariant functor Hom(. , N ), we have
Proposition 7.2.4. Let

α β
M → M → M → 0 (∗)
be a sequence of R-modules and homomorphisms. Then it is exact if and only
if for any R-module N , the induced sequence
β∗ α∗
0 → Hom(M , N ) → Hom(M, N ) → Hom(M , N ) (∗∗)
is exact.
Proof. Suppose (∗) is exact. To show that (∗∗) is exact, we must show that
β ∗ is 1-1 and Im β ∗ = ker α∗ . Suppose β ∗ (α) = 0, where α : M → N is a
homomorphism. Then we claim that α = 0. If α
= 0, then note that there exists
an element x ∈ M such that α(x )
= 0. But, since β is onto, there exists
an element x ∈ M such that x = β(x) which means β ∗ (α)(x) = (α ◦ β)(x) =
α(x )
= 0, i.e., β ∗ (α)
= 0, a contradiction. Thus, β ∗ is 1-1. Next, note that
α∗ ◦ β ∗ = (β ◦ α)∗ = 0∗ = 0. Hence Im β ∗ ⊆ ker α∗ . Conversely, suppose
f ∈ ker α∗ , i.e., α∗ (f ) = f ◦ α = 0. Then f maps α(M ) into zero which
means f maps ker β = α(M ) to zero. Then f defines a unique homomorphism
g : M → N satisfying the condition that g ◦ β = f , i.e., f = β ∗ (g). Thus,

ker α∗ ⊆ Im β ∗ .
Conversely, suppose the sequence (∗∗) is an exact sequence for each

R-module N . Then we will prove that the sequence (∗) is exact. Assume
that the map β is not an onto map. This means M /β(M )
= 0. Hence
the quotient map ν : M → M /β(M ) is nonzero. Since β ∗ is 1-1, taking
N = M /β(M ), we find that β ∗ (ν)
= 0. But clearly νβ = 0, a contradiction.
Next, take N = M and consider the identity map IM ∈ Hom(M , M ). By
the exactness of (∗∗) for this choice of N , we must have (α∗ ◦ β ∗ )(IM ) = 0.
But this means (β ◦ α)∗ (IM ) = IM ◦ (β ◦ α) = β ◦ α = 0. This shows that
Im α ⊆ ker β. Conversely, take N = M/Im α and let φ : M → M/Im α be
the quotient map. Then α∗ (φ) = φ ◦ α = 0 which implies that there exists a
ψ : M → M/Im α such that φ = ψ ◦ β. It follows that ker β ⊆ ker φ = Im α.
Thus, (∗) is exact.
A similar result stated below is also true for the covariant functor
Hom(M, .). A proof analogous to the above is left as an exercise.
Theorem 7.2.5. Let

f g
0 → N → N → N
be a sequence of R-modules. Then it is exact if and only if for any R-module

M , the induced sequence
f∗ g∗
0 → Hom(M, N ) → Hom(M, N ) → Hom(M, N )
is exact.
The basic fact contained in the above two propositions is frequently ex-
pressed by saying that the functor HomR (., .) is left exact in each variable. The
following example shows that the functor Hom(., N ) is not exact on the right,
i.e., we have examples of injective homomorphisms of modules α : M → M
but the corresponding map α∗ : Hom(M, N ) → Hom(M , N ) are not injective.
It is the important fact which gives rise to the functors Ext1 (., .), Ext2 (., .)
etc.
Example 7.2.6. Note that Hom(Z, Zn ) ∼

= Zn , n ≥ 2. Let f : Z → Z be the
1-1 map defined by f (x) = 2x. Then the induced map f∗ : Hom(Z, Z2 ) →
Hom(Z, Z2 ) is indeed the zero map.
In dealing with homomorphisms in homology and cohomology groups, one

frequently needs the following simple result:
Lemma 7.2.7. (Five Lemma). Let
-M α - β - - 0
0 M M
f g h
? γ - ? ?
0 -N N
δ -
N - 0
be a commutative diagram of R-modules and homomorphisms in which the two

rows are exact. If the two vertical maps f and h are isomorphisms, then g is
also an isomorphism.
Proof. The map g is 1-1. For, suppose g(m) = 0. This means hβ(m) = β g(m) =
0. Since h is 1-1, we find that β(m) = 0, i.e., m ∈ ker β. The exactness of
first row implies that there exists an m ∈ M such that α(m ) = m. Thus,
α (f (m )) = gα(m ) = 0. Since α is 1-1, we see that f (m ) = 0. Since f is 1-1,
we find that m = 0. Therefore, m = α(m ) = 0.
To show that g is onto, let n ∈ N . Then β (n) ∈ N . Since h is onto,
there is an m ∈ M such that h(m ) = β (n). But, since β is onto, there is
an m ∈ M such that m = β(m), i.e., h(β(m)) = β (n) = β g(m). This means
β (g(m)−n) = 0. Therefore, g(m)−n = α (n ) for some n ∈ N . Since f is onto,
there is an m ∈ M such that n = f (m ). Thus, g(m) − n = α f (m ) = gα(m )
which implies g(m − α(m )) = n.
Exercises
1. For any R-module M , show that Hom(R, M ) ∼

= M.
2. Let A be a finite abelian group. Prove that Hom(A, Z) = 0.
3. Identify the following modules: HomZ (Zm , Zm ), HomZ (Q, Q) and
HomQ (Q, Q).
4. Suppose
i1 p2
0 → M M M → 0
is a direct sum diagram, i.e., p1 ◦ i1 = IM , p2 ◦ i2 = IM , i1 ◦ p1 + i2 ◦
p2 = IM . Then show that, for any module P, the diagram induced by
Hom( , P ), viz.,
p∗
2 i∗
1
0 → Hom(M , P ) Hom(M, P ) Hom(M , P ) → 0
is also a direct sum diagram. State and prove a similar result for the
functor Hom(P, ) also.
5. Let
f1 - f2 - f3 - f4 -
A1 A2 A3 A4 A5
α β γ δ
? g1 ? g2 ? g ? g4 ?
B1 -B2 - B3 3 -
B4 - B5
be a commutative diagram of R-modules and R-homomorphisms with

exact rows. If β and δ are isomorphisms, α is onto and is 1-1, then show
that γ is also an isomorphism (This is a generalization of five lemma).
f g
6. A short exact sequence 0 → A → B → C → 0 of R-modules is said to
split if there exists a homomorphism k : C → B such that gk = IC . Prove
that the following two conditions are equivalent:
(i) The above exact sequence splits.
(ii) There exists a homomorphism l : B → A such that l ◦ f = IA .
In either of the above cases, show that B ∼
= A ⊕ C.
f1 α β f2
7. Given two sequences A → An → A → 0 and 0 → A → An−1 → A ,
let us define a map f : An → An−1 by putting f = β ◦ α. Show that the
f1 f f2
sequence A → An → An−1 → A is exact if and only if the two given
sequences are exact. Conclude that a long exact sequence can be broken
into a sequence of short exact sequences and vice-versa.
7.2.3 The Tensor Product of Modules and Homomorphisms
Tensor product of R-modules is crucial in explaining homology and cohomology

with general coefficients as well as for introducing “product” in cohomology
groups of a topological space. We continue to assume that R is a commutative
ring with unity element 1
= 0. We define tensor product of two R-modules,
not by the universal property, as we did in the case of free groups and free
products, but by directly showing its existence.
Let M and N be two R-modules. Consider the free R-module C on the set
M × N . Thus, the elements of this free module are finite linear combinations
of elements of M × N with coefficients in R, i.e., the elements are of the form
n
i=1 ri (xi , yi ), where ri ∈ R and (xi , yi ) ∈ M × N . Let D be the submodule
of C generated by all elements of C of the type:
(x + x , y) − (x, y) − (x , y),
(x, y + y ) − (x, y) − (x, y ),
(rx, y) − r(x, y),

(x, ry) − r(x, y).
The quotient module C/D, denoted by M ⊗R N , is called the tensor product
of M and N . Observe that if x ⊗ y denotes the image of (x, y) ∈ M × N under
the quotient map C → D then, in the tensor product M ⊗ N, the following
relations must be true:
(x + x ) ⊗ y = x ⊗ y + x ⊗ y,
x ⊗ (y + y ) = x ⊗ y + x ⊗ y ,
(rx) ⊗ y = r(x ⊗ y) = x ⊗ (ry).
Furthermore, the set {x ⊗ y | (x, y) ∈ M × N } is, clearly, a set of generators
for the tensor product M ⊗R N .
Let M, N, P be three R-modules. Recall that a map f : M × N → P is

said to be R-bilinear if for all y ∈ N , the map x → f (x, y) from M to P and
for all x ∈ M the map y → f (x, y) from N to P are R-linear maps. We make
the following observations:
(1) The map M × N → M ⊗R N, taking (x, y) → x ⊗ y, is an R-bilinear map
– this follows from the relations mentioned above.
(2) If f : M × N → P is any R-bilinear map, where P is an R-module, then f
can be extended by linearity to a module homomorphism f (denoted by
the same symbol) from the free R-module C to P which maps generators
of D into zero. Hence f induces a well-defined R-homomorphism f : M ⊗
N = C/D → P such that f (x ⊗ y) = f (x, y). In fact, the map f , so
defined by f , will be unique.
From the above two properties, we find that the tensor product M ⊗R N
of two R-modules M and N satisfies the following universal property:
Universal Property: Tensor product of two R-modules M and N is an R-module

M ⊗R N along with an R-bilinear map g : M ×N → M ⊗N such that whenever
there is an R-bilinear map f : M × N → P , where P is any R-module, then
there exists a unique R-homomorphism f : M ⊗ N → P making the following
triangle commutative:
g - M ⊗N
M ×N
HH
HH
f H f
j ?
H
P
As usual, it is the above universal property of tensor product, rather than
its definition, which is used in proving results about tensor products. In fact, the
construction used in defining tensor product was needed to show the existence
of tensor product: Once that is done, one will hardly use the construction again
and the universal property will suffice. A few basic results illustrating the point
just made are included in the following proposition:
Proposition 7.2.8. Let M, N, P be three R-modules. Then there exist unique

isomorphisms given below:
(i) M ⊗ N ∼= N ⊗ M.
(ii) (M ⊗ N ) ⊗ P ∼
= M ⊗ (N ⊗ P ).
∼ M.
(iii) R ⊗ M =
(iv) (M ⊕ N ) ⊗ P ∼
= M ⊗ P ⊕ N ⊗ P.
Proof. In each case, we will use the universal property of tensor product to prove
the existence of a homomorphism and then we will find a homomorphism using
the same or some other property which will be its inverse.
(i) In this case, consider the map f : M × N → N ⊗ M taking (x, y) → y ⊗ x,
i.e., f (x, y) = y ⊗ x. Then observe that
f (x + x , y) = y ⊗ (x + x )
= y ⊗ x + y ⊗ x
= f (x, y) + f (x , y).
Similarly, we prove that f (x, y + y ) = f (x, y) + f (x, y ), f (rx, y) = rf (x, y) =

f (x, ry) for each r ∈ R. Thus, f is indeed a bilinear map. Hence f defines a
unique homomorphism f : M ⊗ N → N ⊗ M satisfying f (x ⊗ y) = y ⊗ x. Note
that in proving bilinearity of f , we used the bilinearity of tensor product in N ⊗
M . Using the same procedure, we find a homomorphism h : N ⊗ M → M ⊗ N
such that h (y⊗x) = x⊗y. Thus, it is now evident that h ◦f (x⊗y) = x⊗y and,
similarly, f ◦ h (y ⊗ x) = y ⊗ x. Then uniqueness of these homomorphisms will
yield that each of h ◦f and f ◦h is the identity map, i.e., f : M ⊗N → N ⊗M
is an isomorphism.
(ii) This requires more work. The whole idea is use the universal property of
tensor product and find a homomorphism from (M ⊗ N ) ⊗ P to M ⊗ (N ⊗ P )
taking (x ⊗ y) ⊗ z to x ⊗ (y ⊗ z) for each x ∈ M , y ∈ N , z ∈ P . Then, one also
finds a homomorphism from M ⊗ (N ⊗ P ) → (M ⊗ N ) ⊗ P taking x ⊗ (y ⊗ z)
to (x ⊗ y) ⊗ z. We leave the details as an exercise. Once this has been done, it
is then clear that each of the above two homomorphisms is an isomorphism.
(iii) The map f : R × M → M defined by f (r, x) = r · x is, clearly, bilinear and
hence defines an R-homomorphism f : R ⊗ M → M such that f (r ⊗ x) = r · x.
The map g : M → R ⊗ M, defined by g(x) = 1 ⊗ x, is a homomorphism which
is the inverse of f because
gf (r ⊗ x) = g(r · x) = 1 ⊗ rx = r ⊗ x,
showing that gf is identity. Similarly, f g is clearly the identity map. Hence

f is an isomorphism.
(iv) This one is also straightforward. We find a homomorphism
(M ⊕ N ) ⊗ P → M ⊗ P ⊕ N ⊗ P taking (x, y) ⊗ z to (x ⊗ z, y ⊗ z).
Conversely, we find a homomorphism from M ⊗ P to (M ⊕ N ) ⊗ P tak-
ing x ⊗ z to (x, 0) ⊗ z and also find a homomorphism taking y ⊗ z to
(0, y) ⊗ z. Then we use the definition of direct sum of R-modules to find a
homomorphism M ⊗ P ⊕ N ⊗ P → (M ⊕ N ) ⊗ P taking (x ⊗ z, y ⊗ z) to
(x, 0) ⊗ z + (0, y) ⊗ z = (x, y) ⊗ z. The two homomorphisms are clearly inverse
of each other.
Suppose we have homomorphisms f : M → M and g : N → N . Then

we can use the universal property of tensor products to define the tensor
product of two homomorphism, viz., f ⊗ g : M ⊗ N → M ⊗ N . For this, just
consider the map M × N → M ⊗ N taking (x, y) to f (x) ⊗ g(y). It is readily
verified that this map is R-bilinear and so we have unique homomorphism, de-
noted by f ⊗ g, from M ⊗ N → M ⊗ N defined by (f ⊗ g)(x ⊗ y) = f (x) ⊗ g(y).
The tensor product of homomorphisms preserves composites also, viz., if

f : M → M and g : N → N are further two homomorphisms, then it is
true that (f ⊗ g ) ◦ (f ⊗ g) = (f ◦ f ) ⊗ (g ◦ g). To verify this, just notice that
both the homomorphisms agree on the generators x ⊗ y ∈ M ⊗ N .
Example 7.2.9. It will follow from the next proposition that tensor product of
two epimorphisms is an epimorphism. It is, however, not true that the tensor
product of two monomorphisms is again a monomorphism. To see an example
of this, consider the map f : Z → Z defined by f (x) = 2x. If we tensor by
Z2 , we get the induced map f ⊗ 1 : Z ⊗ Z2 → Z ⊗ Z2 , and this map is not
1-1 because 1 ⊗ 1 ∈ Z ⊗ Z2 is a generator of Z ⊗ Z2 ∼ = Z2 which goes to
(f ⊗ 1)(1 ⊗ 1) = 2 ⊗ 1 = 2(1 ⊗ 1) = 1 ⊗ 2 = 0.
Tensor Product is Right Exact

Let
f g
0 −→ M −→ M −→ M −→ 0 (∗)
be an exact sequence of R-modules. If we tensor the above sequence by an
R-module N , then we get an induced sequence
f ⊗1 g⊗1
0 −→ M ⊗ N −→ M ⊗ N −→ M ⊗ N −→ 0 (∗∗).
As we have seen in the earlier example, even though f is a monomorphism,

f ⊗ 1 need not be a 1-1 map, i.e., the sequence (∗∗) need not be exact on left.
However, the induced sequence is exact at all other terms as the following result
shows!
Proposition 7.2.10. If
f g
M −→ M −→ M −→ 0
is an exact sequence of R-modules then, for any R-module N, the induced se-
quence
f ⊗1 g⊗1
M ⊗ N −→ M ⊗ N −→ M ⊗ N −→ 0
is exact.
Proof. Clearly, g ⊗ 1 is onto. For, if m ⊗ n is a generator of M ⊗ N then, since
g is onto, we can find an m ∈ M such that (g ⊗ 1)(m ⊗ n) = g(m) ⊗ n = m ⊗ n.
Next, since g ◦ f = 0, we find that (g ⊗ 1) ◦ (f ⊗ 1) = (g ◦ f ) ⊗ 1 = 0 which
means Im (f ⊗ 1) ⊆ ker(g ⊗ 1).
For converse, let ν : M ⊗ N → (M ⊗ N )/Im (f ⊗ 1) = L, say, be the
quotient map. By the inclusion just stated, we have an obvious homomorphism
μ : L → M ⊗ N . On the other hand, for each (m , n) ∈ M ⊗ N , we
can choose an element m ∈ M such that g(m) = m . Then the element
ν(m ⊗ n) ∈ L is independent of the choice of m and this defines a bilinear map
(m , n) → ν(m ⊗ n). Therefore, we have a homomorphism ω : M ⊗ N → L
which is just the inverse of μ. Hence M ⊗N ∼
= L and we find that Im (f ⊗1) =
ker(g ⊗ 1).
The fact that tensoring is not exact on the left is very important because
it is this observation which gives rise to the functors Tor1 (., .), Tor2 (., .) etc.
We must remark that if the sequence (∗) above is split exact, then tensoring
by a module will make (∗∗) exact on the left also. For example, if M is free,
then (∗∗) will be exact everywhere. It is interesting to observe that the two
functors, viz., Hom and Tensor product are closely related to one another. We
have
Proposition 7.2.11. For any R-modules M , N and P, there is a canonical iso-

morphism
Hom(M ⊗ N, P ) ∼
= Hom(M, Hom(N, P )).
Proof. We know that any bilinear map f : M × N → P uniquely defines a
homomorphism f : M ⊗ N → P which, in fact, gives a 1-1 correspondence
between the set of all above bilinear maps and the set Hom(M ⊗ N, P ). On
the other hand, given a bilinear map f : M × N → P , we have a linear map
fx : N → P for each x ∈ M defined by fx (y) = f (x, y). Therefore, f yields a
map φ : M → Hom(N, P ) defined by f (x) = fx . Note that for any y ∈ N ,
fx+x (y) = f (x + x , y) = f (x, y) + f (x , y) = (fx + fx )(y).
Similarly, frx = r·fx . Thus, the map φ is indeed a linear map. Conversely, given
a linear map φ : M → Hom(N, P ), we can define a bilinear map g : M × N → P
by g(x, y) = φ(x)(y). This establishes a 1-1 correspondence between the set of
all bilinear maps M × N → P and the set Hom(M, Hom(N, P )). Combin-
ing the two 1-1 correspondences, we find that there is 1-1 correspondence be-
tween the two sets Hom(M × N, P ) and Hom(M, Hom(N, P )). To conclude the
proof, we just observe that the above 1-1 correspondence η : Hom(M ⊗N, P ) →
Hom(M, Hom(N, P )) is defined by the formula
[(ηf )(x)](y) = f (x ⊗ y)
for f ∈ Hom(M ⊗ N, P ) and this η is, in fact, a linear map.
Exercises
1. If m, n are relatively prime, then show that Zm ⊗ Zn = 0. More generally,

for any two positive integers m and n, prove that Zm ⊗ Zn ∼ = Zd , where
d = (m, n) is the gcd of m and n.
2. Use Proposition 7.2.10 to prove that the tensor product of two epimor-
phisms is an epimorphism. Give two distinct examples to show that the
tensor product of two monomorphisms need not be a monomorphism.
3. Compute Zm ⊗Z Q, Q ⊗Z Q and Q ⊗Q Q.
4. Let G be a finite abelian group. Prove that G ⊗Z Q = 0.
Ext and Tor Functors
Suppose A is an abelian group, i.e., A is a Z-module. We can generalise

the results that follow when A is any R-module, R a PID. We say that A is
injective if the following diagram of abelian groups, with exact row, can always
be completed to make it a commutative diagram:
i
0 G G’
h
h’
A
In other words, the abelian group A is injective if any homomorphism h : G → A

can always be extended to a homomorphism h : G → A. Similarly, we say that
an abelian group B is projective if the following diagram, with exact row, can
be completed to a commutative diagram
h’ h
G q
G’’ 0
It is well-known that a Z-module A can always be embedded into an injective

Z-module. Similarly, a Z-module B can always be written as the quotient of
a projective module. It is also true that a quotient of an injective Z-module
is again injective and a submodule of a projective Z-module is again projective.
Now let us define Ext(A, B) for any two abelian groups A and B (i.e., for
Z-modules) as follows:
Definition 7.2.12. Embed B into an injective Z-module I0 and let I1 be the

quotient module. Thus we have an exact sequence
i q
0 −→ B −→ I0 −→ I1 −→ 0
of abelian groups. Apply the covariant functor Hom(A, .) on the above exact
sequence to get an exact sequence
∗ i q∗
0 −→ Hom(A, B) −→ Hom(A, I0 ) −→ Hom(A, I1 ).
Note that q∗ need not be onto. We define
Ext(A, B) = Coker q∗
Thus Ext(A, B) is an abelian group, which makes the sequence
0 −→ Hom(A, B) −→ Hom(A, I0 ) −→ Hom(A, I1 ) −→ Ext(A, B) −→ 0
exact. It can be proved that Ext(A, B), defined as above, is really independent
of the choice of the injective group I0 . In case, P is a projective group (a free
abelian group is always projective), then the induced sequence
0 −→ Hom(P, B) −→ Hom(P, I0 ) −→ Hom(P, I1 ) −→ 0
is always exact, by the definition of projective group. Therefore, it follows that

if A is projective, then
Ext(A, B) = 0
for any group B.
We can also define Ext(A, B) as follows: Write down A as the quotient

of a projective group P0 and let P1 be the kernel. Then we have a short exact
sequence
q1 q0
0 −→ P1 −→ P0 −→ A −→ 0
of abelian groups. We now apply the contravariant functor Hom(., B) to the
above sequence. Then we know that the induced sequence
0 q∗ 1 q∗
0 −→ Hom(A, B) −→ Hom(P0 , B) −→ Hom(P1 , B)
is exact and the map q1∗ need not be onto. We define
Ext(A, B) = Coker q1∗ .
In other words, Ext(A, B) is the abelian group which makes the following
sequence exact :
0 q∗ 1 q∗
0 −→ Hom(A, B) −→ Hom(P0 , B) −→ Hom(P1 , B) −→ Ext(A, B) −→ 0.
Again it can be proved that this definition of Ext(A, B) is independent

of the choice of the projective group P0 and is, in fact, equivalent to the one
given earlier using injective groups.
If we regard C : 0 → B → I0 → I1 → 0 as an injective resolution of B,

and consider the induced cochain complex
C ∗ : 0 → Hom(A, B) → Hom(A, I0 ) → Hom(A, I1 ) → 0

∼ Hom(A, B) and H 1 (C ∗ ) ∼
then, by definition, H 0 (C ∗ ) = = Ext(A, B). This
interpretation yields the following interesting result.
Proposition 7.2.13. Let 0 → A → A → A” → 0 be an exact sequence of abelian

groups. Then for any abelian group B there is an induced exact sequence of
abelian groups
0 −→ Hom(A , B) −→ Hom(A, B) −→ Hom(A , B) −→ Ext(A , B)
−→ Ext(A, B) −→ Ext(A , B) −→ 0.
Proof : Let 0 → B → I0 → I1 → 0 be an injective resolution of B. Consider the
following commutative diagram in which the horizontal rows are exact (because
I0 , I1 both are injective)
0 0 0
↓ ↓ ↓
0 → Hom(A , B) → Hom(A, B) → Hom(A , B) → 0
↓ ↓ ↓
0 → Hom(A , I0 ) → Hom(A, I0 ) → Hom(A , I0 ) → 0
↓ ↓ ↓
0 → Hom(A , I1 ) → Hom(A, I1 ) → Hom(A , I1 ) → 0
↓ ↓ ↓
0 0 0
Applying the homology exact sequence theorem (ker-coker theorem) to the

two bottom horizontal rows of the above diagram (see later in this section)
and interpreting the Hom and Ext terms as the kernel and cokernel of vertical
maps, the result follows immediately.
Using projective resolution of the first variable in Ext(A, B), one can easily
derive the following analogous result for the second variable:
Proposition 7.2.14. Let 0 → B → B → B → 0 be an exact sequence of abelian

groups. Then for any abelian group A, there is the following exact sequence of
abelian groups
0 → Hom(A, B ) → Hom(A, B) → Hom(A, B ) → Ext(A, B )
→ Ext(A, B) → Ext(A, B ) → 0.
Examples
(i) Ext(Z, A)=0 for each abelian group A. This follows since Z is free (and
hence projective).
∼ B/nB. In particular Ext(Zn , Z) ∼
(ii) Ext(Zn , B) = = Zn .
n
Observe that 0 → Z → Z → Z/nZ → 0 is a free resolution of Z/nZ.
Hence the following sequence is exact
0 → Hom(Z/nZ, B) → Hom(Z, B) → Hom(Z, B) → Ext(Z/nZ, B) → 0.
Since Hom(Z, B) ∼
= B and the middle map is just multiplication by n,
Ext(Z/nZ, B) ∼
= B/nB.
(iii) Ext(Zn , Zm ) = Zd , where d =gcd(n, m).
Consider the exact sequence
m
0 → Zm → Q/Z −→ Q/Z → 0,
where the map on the right is the multiplication by m. Then we get the
following exact sequence
∗ m
0 → Hom(Zn , Zm ) → Hom(Zn , Q/Z) −→
Hom(Zn , Q/Z) → Ext(Zn , Zm ) → 0
in which one checks easily that m∗ is also multiplication by m.
Since Hom(Zn , Q/Z) ∼= Zn , we see that the groups at both ends should be
cyclic. Exactness of the sequence yields that both end groups must also be of
the same order, say d. Now we claim that ker m∗ = d =gcd(n, m). Let n = dp
and m = dq where p, q are relatively prime. Then
r ∈ ker m∗ ⇔ n|rm ⇔ dp|rdq ⇔ p|rq ⇔ p|r.
Therefore r = p works and is the smallest such positive integer. This proves
our assertion. Hence Ext(Zn , Zm ) ∼
= Zd ∼
= Ext(Zm , Zn ), where d =gcd(n, m).
Tor Functor
i
To define Tor(A, B) for any two abelian groups A, B, let 0 → P1 −→
P0 → A → 0 be a projective resolution of A. Then the sequence
i⊗B
0 → P1 ⊗ B −→ P0 ⊗ B → A ⊗ B → 0
is exact everywhere except possibly on the left. We define A ∗ B (A torsion B)

or Tor(A, B) to be just the kernel of the homomorphism i ⊗ B. In other words,
A ∗ B is an abelian group which makes the following sequence exact
0 → A ∗ B → P1 ⊗ B → P0 ⊗ B → A ⊗ B → 0.
The torsion product A ∗ B is also symmetric in a natural way. Furthermore,

given any exact sequence 0 → A → A → A → 0 of abelian groups, there is
an induced long exact sequence for any abelian group B viz.,
0 → A ∗ B → A ∗ B → A ∗ B → A ⊗ B → A ⊗ B → A ⊗ B → 0.
Now one can easily verify the following facts using the same techniques
which were adopted earlier for computing the Ext(A, B).
Examples
(i) Tor(A, Z) = 0 ∀ abelian group A.

(ii) Tor(A, Zn ) ∼
= ker{n : A → A}, where n is the multiplication by n.
(iii) Tor(Zn , Zm ) = Zd = Tor(Zm , Zn ), where d =gcd (n, m). To see the last
fact, note that the exact sequence
n
0 → Z −→ Z → Zn → 0
induces an exact sequence
0 → Tor(Z, Zm ) → Tor(Z, Zm ) → Tor(Zn , Zm ) →
Z ⊗ Zm → Z ⊗ Zm → Zn ⊗ Zm → 0.
Since Z is free, this reduces to
∗ n
0 → Tor(Zn , Zm ) → Zm −→ Zm → Zn ⊗ Zm → 0
where n∗ is the multiplication by n. This means both of the end groups

are cyclic and are of the same order, say d. Then, as before, d =gcd (n, m).
Exercises
1. An abelian group A is said to be divisible if given any a and an n(

= 0) ∈ N,
there exists a b ∈ A such that nb = a. Prove that A is divisible if and only
if A is injective as a Z-module.
2. Prove that the quotient of an injective abelian group is again injective.
3. Show that a subgroup of a projective abelian group is again projective.
4. Prove that for a given abelian group A, Ext(A, B) = 0 for all B if and
only if A is projective.
5. Prove that for a given B, Ext(A, B) = 0 for all A if and only if B is
injective.
6. For an abelian group A, show that Tor(Zn , A) = {a ∈ A | na = 0}.
7. Prove that A is torsion free if and only if Tor(A, B) = 0 for every B.
7.2.4 Chain Complexes and Homology
A sequence M = {(Mn , ∂n ), n ∈ Z}, of R-modules Mn with R-homomorphisms

∂n : Mn → Mn−1 , called boundary homomorphisms, viz.,
∂n+1 ∂
M : . . . −→ Mn+1 −→ Mn −→
n
Mn−1 −→ . . . ,
is called a chain-complex if ∂n ◦ ∂n+1 = 0 for all n ∈ Z. The chain complexes

of practical importance which we would come across will have all the negative
terms zero – these are called positive chain complexes.
Let Zn = ker ∂n and Bn = Im ∂n+1 . Then Zn and Bn are submodules

of Mn . The elements of Zn are called n-cycles and the elements of Bn
are are called n-boundaries. The elements of Mn are called n-chains. Since
∂n ◦ ∂n+1 = 0, we find that Bn ⊆ Zn for each n. The quotient module Zn /Bn ,
denoted by Hn (M ), is called the n-dimensional homology module of the chain
complex M∗ . The elements of Hn = Zn /Bn are called homology classes and
are denoted by {z}, where z ∈ Zn .
If we take a sequence M = {(M n , δ n )}, viz.,
δ n−1 δn
M : . . . −→ M n−1 −→ M n −→ M n+1 −→ . . .
of R-modules and R-homomorphisms such that δ n ◦ δ n−1 = 0 for each n ∈ Z,

then M is called a cochain complex of R-modules. Occasionally, a chain
complex M is denoted by M∗ and a cochain complex M by M ∗ . The indices
in (M n , δ n ) are superscripts rather than subscripts just to distinguish it from
a chain complexe M – it is true that except for the position of indices, there
is no conceptual difference between a chain complex and a cochain complex.

We call the elements of Z n = ker δ n cocycles, elements of B n = Im δ n−1
coboundaries, elements of M n cochains and the elements of the quotient group
H n (M ) = Z n /B n cohomology classes. The quotient group itself is called
n-dimensional cohomology module of the cochain complex M . Note that if the
homology groups Hn (M ) = 0, then the chain complex becomes exact at Mn .
Thus, the homology groups of a chain complex M measure its deviation from
exactness of the sequence.
Let M and N be two chain complexes of R-modules. Then a sequence

{fn : Mn → Nn } of R-homomorphisms is called a chain map from M to N
if these homomorphisms commute with boundary homomorphisms, i.e., the
square
-Mn ∂n - -
Mn−1
fn fn−1
? ∂ ?
-Nn n -
Nn−1 -
is commutative for each n ∈ Z, i.e., fn−1 ◦ ∂n = ∂n ◦ fn for all n ∈ Z. The

definition of a chain isomorphism (or chain equivalence) is now evident. We
readily verify that fn maps cycles into cycles and boundaries into boundaries
and, therefore, it defines a homomorphism (fn )∗ : Hn (M ) → Hn (N ), n ∈ Z by
the formula (fn )∗ {z} = {fn (z)}, where {z} ∈ Hn (M ). These are called induced
homomorphisms in homology. If there is no confusion, we write (fn )∗ just as
f∗ : Hn (M ) → Hn (N ). Let us observe that these induced homomorphisms are
functorial, viz.,
(i) If f : M → N and g : N → P are two chain maps, then their composite
g ◦ f is a chain map and
(g ◦ f )∗ = g∗ ◦ f∗ : Hn (M ) → Hn (P ).
(ii) If I : M → M is the identity chain map, then the induced homomorphism

I∗ : Hn (M ) → Hn (M ) is also the identity map.
A chain complex M is said to be acyclic if Hn (M ) = 0 for all n ∈ Z, i.e.,

all the homology groups of M are trivial. When we dualize each of the notions
defined above, we get cochain map, cochain isomorphisms, etc. Next, we have
Definition 7.2.15. Let f, g : M → N be two chain maps. We say that f is chain

homotopic to g, written as f ≈ g, if there is a sequence {Dn : Mn → Nn+1 } of
homomorphisms such that for each n ∈ Z,
fn − gn = ∂n+1 Dn + Dn−1 ∂n .
We now observe that if we consider the set of all chain maps from M to
N , then the relation of chain-homotopy is an equivalence relation in that set.
The following result is noteworthy:
Proposition 7.2.16. If two chain maps f, g : M → N are chain homotopic,

then the homomorphism in homology induced by them are identical, i.e., for all
n ∈ Z,
f∗ = g∗ : Hn (M ) → Hn (N ).
Proof. Let {Dn : Mn → Mn+1 } be a chain homotopy from f to g and
{z} ∈ Hn (M ). This means ∂n (z) = 0. Since fn − gn = ∂n+1 Dn + Dn−1 ∂n , we
find that fn (z) − gn (z) = ∂n+1 Dn (z) is a boundary. Hence {fn (z)} = {gn (z)},
i.e., f∗ ({z}) = g∗ ({z}).
Similarly, one can define a cochain-homotopy between two cochain maps.

Then it follows from the above proposition that any two cochain homotopic
maps induce identical maps in cohomology.
Let us recall the Hom and Tensor functors discussed earlier. Note that if
M∗ : . . . → Mn+1 → Mn → Mn−1 → . . .
is a chain complex, and if for any R-module N we apply the Hom( , N ) functor
to this chain complex, then we get the induced sequence Hom(M∗ , N ):
. . . ← Hom(Mn+1 , N ) ← Hom(Mn , N ) ← Hom(Mn−1 , N ) ← . . .
of R-modules going in opposite direction. In other words, we get a cochain

complex Hom(M∗ , N ). Here, we must point out that if the chain complex M∗
of R-modules is acyclic, i.e., Hn (M∗ ) = 0 for all n ∈ Z and we take the induced
cochain complex Hom(M∗ , N ), it is not necessary that this cochain complex
will also be acyclic - this is clear because Hom(. , N ) is only left exact, it is not
an exact functor. This observation is used in defining the concept of higher or-
der Hom( , ) functors called Extn (. , ) functors. Similarly, when we tensor the
chain complex M∗ , which is acyclic, by another R-module N , it is not necessary
that the resulting complex M∗ ⊗ N would be acyclic. This phenomenon gives
rise to higher order tensor products called Torn (. , ) functors (see S. MacLane
[17] for all the details). The special case when N = R is of frequent occur-
rence in algebraic topology. In this case, the homology of the cochain complex
Hom(M∗ , R) is called the cohomology of the chain complex M∗ . Now, it is perti-
nent to ask: What is the relationship between the homology groups of the chain
complex M∗ and the homology groups of the cochain complex Hom(M∗ , R)?
In other words, how is the homology of a chain complex M∗ related with its
cohomology? The answer to this question is given by the following theorem
known as the Universal Coefficient Theorem for Cohomology.
Theorem 7.2.17. Let K be a chain complex of free R-modules and G be an R-

module, where R is a PID (principal ideal domain). Then, with H n (K, G) =
Hn (Hom(K, G)), there exists a split exact sequence
β α
0 → Ext(Hn−1 (K), G) → H n (K, G) → Hom(Hn (K), G) → 0.
The homomorphism β is natural with respect to the chain map as well as with
respect to the coefficient module, but the splitting is not natural with respect to
the chain map.
Another significant situation arises when we take the tensor product of a

given chain complex M∗ by an R-module N . We get the new chain complex,
viz., M∗ ⊗ N . Again, the question is: How is the homology of the chain complex
M∗ related to the homology of the new chain complex M∗ ⊗ N ? The answer
is known as the “Universal Coefficient Theorem for Homology” (See Spanier,
E.H. [18] for this and the preceding result).
Theorem 7.2.18. (Universal Coefficient Theorem for Homology). Let K be a

chain complex of free R-modules, R a PID, and let G be an R-module. Then
there exists a split exact sequence
α β
0 → Hn (K) ⊗ G → Hn (K ⊗ G) → Tor1 (Hn−1 (K), G) → 0,
which is natural with respect to the chain complex K and coefficient module G,
but the splitting is not natural with respect to the chain maps.
7.2.5 Tensor Product of two Chain Complexes
Let (K, ∂ ) and (L, ∂ ) be two chain complexes of R-modules which are positive,
i.e., Kn = 0 = Ln for all negative integers n. The tensor product of K and L
is a new chain complex, denoted by K ⊗ L, whose nth term is given by the
module (K ⊗ L)n = p+q=n Kp ⊗ Lq and the boundary operator
∂n : (K ⊗ L)n → (K ⊗ L)n−1
is defined using ∂ , ∂ as follows: for any generating element a ⊗ b ∈ Kp ⊗ Lq ,
∂n (a ⊗ b) = ∂ a ⊗ b + (−1)deg a a ⊗ ∂ b.
Here, deg a means degree of the element a, where deg a = p if a ∈ Kp . We have
Proposition 7.2.19. The homomorphism ∂ : (K ⊗ L)n → (K ⊗ L)n−1 satisfies

the property ∂ ◦ ∂ = 0, i.e., (K ⊗ L, ∂) is a chain complex.
Proof. Note that (K ⊗ L)n consists of the direct sum of (n + 1) terms of the
type Kp ⊗ Lq , where p + q = n. If a ⊗ b is a generator of Kp ⊗ Lq , then
∂ ◦ ∂(a ⊗ b) = ∂(∂ a ⊗ b + (−1)deg a a ⊗ ∂ b)

= ∂ (∂ a) ⊗ b + (−1)deg ∂ a (∂ a) ⊗ ∂ b + (−1)deg a ∂ a ⊗ ∂ b
+(−1)deg a+deg a a ⊗ ∂ (∂ b)
= 0,
since deg a and deg(∂ a) differ by 1, and this proves the assertion.
One can similarly define the tensor product of two positive cochain com-
plexes. Once again, it may now be asked as to how the homology groups of the
new chain complex K ⊗ L are related to the homologies of K and L. There is
a well-known answer to this question. It is known as the “Künneth Formula”,
which provides an exact sequence connecting the homology of K ⊗ L in terms
of the tensor product and torsion product of homology groups of K and L.
The Künneth formula is extremely useful in computing the homology (respec-
tively, cohomology) of the product topological space in terms of the homology
(respectively, cohomology) groups of the individual spaces (see E.H. Spanier
[18]).
Theorem 7.2.20. (Künneth Formula for Homology). Let K and L be two chain
complexes of R-modules, at least one of which is a chain complex of free modules
and R is a PID. Then there exists a split exact sequence
α
0→ Hi (K) ⊗ Hj (L) → Hn (K ⊗ L)
i+j=n
β
→ Tor1 (Hi (K), Hj (L)) → 0.
i+j=n−1
The homomorphisms α and β are natural with respect to the chain maps, but
the splitting is not natural.
At this point, it is worth mentioning that the tensor product respects

chain homotopies between chain maps. To explain this, let us denote the chain
homotopy between chain maps f and f as D : f ∼ = f . In this notation, we
have
Proposition 7.2.21. Let f1 , f2 : K → L be two chain-homotopic chain maps

and g1 , g2 : K → L be another chain-homotopic chain maps. Then the chain
maps f1 ⊗ g1 , f2 ⊗ g2 : K ⊗ K → L ⊗ L are also chain-homotopic.
Proof. Let S : f1 ∼
= f2 and T : g1 ∼
= g2 be two chain homotopies. Then we can
define a chain homotopy H : f1 ⊗g1 ∼= f2 ⊗g2 by the formula H = S⊗g1 +f2 ⊗T ,
i.e., we define
H(x ⊗ y) = S(x) ⊗ g1 (y) + (−1)deg x f2 (x) ⊗ T (y).

In fact, S ⊗ 1 gives a chain homotopy K ⊗ L → K ⊗ L from f1 ⊗ 1 to f2 ⊗ 1,
and 1 ⊗ T gives a homotopy between 1 ⊗ g1 ∼ = 1 ⊗ g 2 : K ⊗ L → K ⊗ L .
Composing these homotopies yields the desired result.
7.2.6 Exact Homology Sequence Theorem
The next result, called Exact Homology Sequence Theorem, which we are
going to state and discuss, is sometimes known as the Fundamental Theo-
rem of Homological Algebra. It is indeed a very powerful result which has
numerous applications in the development of all kinds of homology groups
and is of frequent occurrence elsewhere also. We will not prove the result
as the proof is not only technical but also requires lengthy verifications. We
will, however, explain the crucial steps involved in the statement of the theorem.
Let C , C and C be three chain complexes of R-modules and suppose

there are chain maps f : C → C and g : C → C which yield a short exact
sequence
f g
0 −→ C −→ C −→ C −→ 0
of chain complexes. Since f and g are chain maps, they induce homomorphisms
f∗ g∗
Hn (C ) −→ Hn (C) −→ Hn (C )
for each n ∈ Z. The Exact Homology Sequence Theorem asserts the existence
of a sequence of homomorphisms δ : Hn (C ) → Hn−1 (C ), n ∈ Z, such that
the resulting long sequence
f∗ g∗
→ Hn (C ) → Hn (C) → Hn (C ) → Hn−1 (C )
δ
f∗ g∗
→ Hn−1 (C) → Hn−1 (C ) →,
connecting the homology groups of C , C and C , is exact. The homomorphism
δ : Hn (C ) → Hn−1 (C ) which connects the nth level to the (n − 1)th level of
homology groups is called the connecting homomorphism. The crucial aspect
of this homomorphism is that it is natural in the given short exact sequence of
chain complexes, i.e., if we have two short exact sequences
E: 0 → C → C → C → 0
and
F : 0 → D → D → D → 0
which are related by chain maps α : C → D , β : C → D, γ : C → D
making various squares commutative, then the connecting homomorphism
δE : Hn (C ) → Hn−1 (C ) and δF : Hn (D ) → Hn−1 (D ) make the resulting

squares in the diagram
- Hn (C ) δE - -
Hn−1 (C )
γ∗ α∗
? δF - ?
- Hn (D ) Hn−1 (D ) -
commutative for each n ∈ Z. We can now state
Theorem 7.2.22. Let

f g
0 → C → C → C → 0
be an exact sequence of chain complexes of R-modules. Then the long sequence

induced in their homology, viz.,
g∗ f∗
· · · → Hn (C) → Hn (C ) → Hn−1 (C ) → Hn−1 (C) → · · ·
δ
is exact, where δ is the connecting homomorphism.
We can easily indicate the definition of δ as follows: take a homology class

{z } ∈ Hn (C ). Using ontoness of g, find a c ∈ Cn such that gn (c) = z . Then
∂(c) ∈ Cn−1 . It can be shown that ∂c ∈ ker gn−1 = Im fn−1 . Hence we can find

a z ∈ Cn−1 such that fn−1 (z) = ∂c. Then it can be proved that z is also a
cycle of C and we define δ{z } = {z }, i.e., δ{z } = {fn−1 ∂gn−1 }. The technical

details are needed to show that δ is well defined in the sense that the definition
is independent of all the choices which we have made in going from z ∈ Cn to
z ∈ Cn−1

. Once it is proved that δ is a well-defined homomorphism, then the
exactness of the long sequence is proved by “diagram chase”. That completes
the proof of the long exact homology sequence theorem – the fundamental
theorem of homological algebra (see G.E. Bredon [2] p. 178–180 for details).
Exercises
1. Let f, g : C → C ; f , g : C → C be chain maps and ∼ = denote chain

homotopy. Prove that f ∼ = g and f ∼ = g implies f ◦ f ∼
= g ◦ g.
2. Let
0 → C → C → C → 0
be an exact sequence of chain complexes. If any two of these are acyclic,

then show that the third one is also acyclic.
3. Let
7.3. Topological Transformation Groups 331
0 - C - C - C - 0
? ? ?
0 - D - D - D - 0
be a commutative diagram of chain complexes in which the two rows are

exact. Show that if any two of the vertical chain maps induce isomorphisms
in homology then so does the remaining third chain map.
4. Let f : C → C be a chain map between two nonnegative chain complexes
of free abelian groups such that the induced homomorphism f∗ : Hq (C) →
Hq (C ) is an isomorphism for all q ≥ 0. Then prove that the chain map f
is a chain equivalence.
7.3 Topological Transformation Groups
The concepts, e.g., a topological group acting on a topological space con-

tinuously, a Lie group acting on a smooth manifold smoothly, a discrete group
acting on a simplicial complex simplicially, etc., are all similar and can be de-
fined in an analogous manner with obvious changes. Here, we deal with only
continuous actions as they provide the best examples of topological spaces in
the guise of orbit spaces. Before we define it formally, let us see a very general
concept which is always present in the background of everything mentioned
above.
Group Acting on a Set
Let X be a set and G be group. A map G × X → X, (g, x) → g.x, is called

an action of G on the set X if the two conditions stated below are satisfied for
all g1 , g2 ∈ G and x ∈ X:
(i) (g1 .g2 ).x = g1 .(g2 .x),
(ii) e.x = x.
Here, e ∈ G is the identity element of G. Fix a point x ∈ X and consider the
set Gx = {g ∈ G | g.x = x}. Then Gx is a subgroup of G and is called the
isotropy subgroup of the point x ∈ X. Next, the subset G(x) = {g.x | g ∈ G}
of X is called the orbit of x in X. One can easily verify that any two orbits of
X are either identical or disjoint. Hence these orbits decompose the set X into
mutually disjoint orbits of X. One can easily prove that, for any x ∈ X, the map
g → g.x induces a 1-1, onto map f : G/Gx → G(x) defined by f (g.Gx ) = g.x.
Thus, the set of all cosets of the isotropy subgroup Gx of x in G is in bijection
with the set of all elements in the orbit G(x). Now, if we choose exactly one
xi from each orbit G(xi ), i ∈ I, then since orbits {G(xi ) | i ∈ I} provide a
partition of X, we find an important formula, viz.,

|X| = |G(xi )|,
i∈I
where |X| denotes the number of elements in the set X. It may also be noted
that if x, y are in the same orbit, then the isotropy subgroups Gx and Gy are
conjugate in G, i.e., Gx = gGy g −1 for some g ∈ G. The converse of this is also
true, viz., if H is conjugate to an isotropy subgroup G(x), say, H = gGx g −1 ,
then there exists a point y ∈ G(x), viz., y = gx such that Gy = H.
We now illustrate some interesting actions of a group G on different sets.
Example 7.3.1. Let us consider the group G acting on the set G which is the
underlying set of the group itself defined by g.x = gxg −1 , where the right-hand
side is the conjugate of x by the element g.
In the above example, we readily see that the isotropy subgroup Ga of an

element a ∈ G is simply the normalizer N (a) of {a} in G. Thus, a ∈ Z(G),
the centre of the group if and only if G(a) = {a} or Ga = G. The orbit of an
element a ∈ G is just the conjugacy
class of
the element a. It follows that if G
is a finite group then |G| = |G/N (a)| = |G|/|N (a)|. Now, the important
formula mentioned earlier gives us the following equation, called class equation
of the finite group G,

|G| = |G/N (a)|
|G|
= |Z(G)| + .
|N (a)|
N (a)=G
There is yet another useful case of a group acting on a set:
Example 7.3.2. Let us consider the set L(G) of all subgroups of G. Consider
the group G acting on the set L(G) under the action defined by g.H := gHg −1 .
Here, we should note that the point H of L(G) is a fixed point under the
above action if and only if H is normal in G. The orbit of H is just the set
of all subgroups of G which are conjugate to H, i.e., the conjugacy class of
H. The isotropy subgroup of a subgroup H of G is simply the normalizer
N (H) of H in G. Hence it follows from the general facts about group actions
that |G/N (H)| must be equal to the number of distinct conjugacy classes of
H in G. This concept of conjugacy classes of subgroups of a group G plays a
fundamental role in the classification of covering projections.
7.3.1 Topological Transformation Groups
Now, we will discuss the important notion of a “topological transformation

group”. Let G be a topological group, i.e., G is a group with a topology on it
so that the group operations G×G → G and G → G, defined by (g1 , g2 ) → g1 .g2
and g → g −1 , respectively, are continuous. Let X be a topological space. Then a
continuous map G×X → X, denoted by (g, x) → g.x, is said to be a topological
action (or just an action) of G on X if the following two conditions (same as
mentioned earlier) are satisfied:
(i) (g1 .g2 ).x = g1 (g2 .x)
(ii) e.x = x
for every g1 , g2 ∈ G and for all x ∈ X. Here, e ∈ G is the identity element.
The pair (G, X) along with the given action is called a topological
transformation group. If we change any one of the three items, viz., G, X or
the action, then we get a different transformation group. If, in the definition of
a transformation group, we forget the topologies from the space X, the group
G and the action, then we revert back to the notion of “a group G acting on a
set X” discussed earlier.
Recall the definitions of isotropy group and orbit given earlier. Note
that if the group G is compact and the space X is Hausdorff, then the orbits
will be closed sets of X. In fact, in this case, the coset space G/Gx will be
homeomorphic to the orbit G(x). The action of G on X is said to be free
if the isotropy group of each point is trivial, i.e., Gx = {e} for all x ∈ X.
The set F (G, X) = {x ∈ X | Gx = G} is called the fixed-point set of the
transformation group (G, X) and is also denoted by X G . We have already
noted that any two orbits in X are either identical or disjoint. The set of all
distinct orbits of X, denoted by X/G, with the quotient topology induced
from X, is called the orbit space of the transformation group. We will come
across many spaces which are simply the orbit spaces of a space X under a
suitable action of a topological group G.
Before giving examples, let us remark that if a topological group G acts

on a space X then, for each g ∈ G, we can define a map
fg : X → X
by setting
fg (x) = g.x.
One can, using conditions (i) and (ii) above, easily verify that for any pair
g1 , g2 ∈ G, fg1 .g2 = fg1 ◦ fg2 and fe = IX , where e ∈ G is the identity element
of G. We also note that, since the action is continuous, each fg : X → X is a
continuous map. Now, taking g1 = g and g2 = g −1 , we find that
fg ◦ fg−1 = fg.g−1 = fe = IX .
Similarly, we have
fg−1 ◦ fg = IX .
This shows that each of the maps fg : X → X is indeed a homeomorphism.
Furthermore, the map η : G → Homeo(X) defined by η(g) = fg is a homomor-
phism of groups. We say that the given action of G on X is effective if η is 1-1,
i.e., each g(
= e) yields a nonidentity homeomorphism of X. In case the action
is effective, we can clearly identify every element g ∈ G with a homeomorphism
x → g.x of X and thus, G can be viewed as a group of homeomorphism of the
space of X. With this viewpoint, the definitions of isotropy groups, orbits, free
actions, etc., have nice geometric flavour, e.g., the action of G on X is free if
and only if each g(
= e) ∈ G moves every point of X.
Conversely, given a group G of homeomorphism of X, we can always

define an action of G on X by putting g.x = g(x). The only question which is
not immediately obvious is: what is the topology on G? Since G ⊆ Homeo(X),
any topology on Homeo(X) (for instance, the compact open topology), will
induce a topology on G. But, then the question will be whether or not the
map G × X → X defined above will be continuous. It is well known that if
X is a locally compact Hausdorff space then the compact open topology does
give a continuous action of the desired type (see Dugundji [5] p. 259).
An important case, which occurs frequently, is the case when we take

the discrete topology on a group G of homeomorphisms of a space X. In this
case, the map G × X → X, defined above, is always continuous and we have
a nice action of G on X. In the special case of a finite group G, the topology
on G is always assumed to be discrete; we even do not mention this explicitly.
The finite group of homeomorphisms of a space X have been studied and used
extensively in the study of fixed-point sets.
Example 7.3.3 (Real Projective Space). Let Sn be the standard n-sphere, n ≥ 1.

Let A : Sn → Sn be the antipodal map, i.e., A(x) = −x. Then A2 is identity.
Thus, G = {I, A} is a group of homeomorphisms of Sn and so it acts on Sn .
The action is free and the orbit space Sn /G, denoted by RPn , is called the real
projective space. The space RPn is a compact, Hausdorff, connected manifold
of dimension n.
Example 7.3.4 (Complex Projective Space). Let S1 ⊂ C be the circle group.

Then it is easily seen that S1 is a topological group. Let

n
S2n+1 = {(z0 , z1 , . . . , zn ) ∈ Cn : |zi |2 = 1},
i=1
be the (2n + 1)-dimensional unit sphere. Then S1 acts on S2n+1 continuously

under the action defined by
z.(z0 , z1 , . . . , zn ) = (zz0 , zz1 , . . . , zzn ).
This again is a free action. The orbit space S2n+1 /S1 , denoted by CPn , is called
the complex projective space. This space is a compact, Hausdorff, connected
manifold of real dimension 2n.
Example 7.3.5 (Torus). Let R be the real line and define a map h : R → R by
h(x) = x+1. Then h is a homeomorphism and for each integer n, hn : R → R is
also a homeomorphism. These are just translations by integer amounts. Thus,
we get the infinite cyclic group Z = [h] generated by the homeomorphism
h of R. Giving Z the discrete topology, we find that Z acts as a group of
homeomorphisms on R. Again, the action is free and the orbit space, denoted
by R/Z, can be seen to be homeomorphic to the circle group S1 .
We can take the product action of the discrete group Z × Z on R × R by
considering two homeomorphisms h and k of R and defining the action of (h, k)
on R × R by putting
(h, k)(x, y) = (x + 1, y + 1).
Then
(hm , k n )(x, y) = (x + m, y + n)
for every pair of integers (m, n). Again, the action is free and the orbit space
is S1 × S1 , which is the 2-torus. One can similarly obtain n-torus as an orbit
space of Rn for all n ≥ 2 under the action of Zn .
Example 7.3.6 (Klein bottle). A group may act on a space in many different
ways. To see a simple example of this situation, consider the torus T as a
subspace of R3 given by rotating the circle (x − 3)2 + z 2 = 1 about z-axis.
Define maps f, g and h as follows:
f (x, y, z) = (x, −y, −z),

g(x, y, z) = (−x, −y, z),
h(x, y, z) = (−x, −y, −z).
Each of these is a homeomorphism of T of order 2. Hence, the group G =

Z2 = {1, a} can act on T in three distinct ways. The interesting point to be
verified and observed is that orbit spaces corresponding to all the three actions
defined by homeomorphisms f , g, h on T are distinct. In the first case, it is
homeomorphic to the sphere; in the second case, it is homeomorphic to the
torus itself and in the last case, it is homeomorphic to the Klein bottle.
Example 7.3.7 (Lens Spaces). Let us take the 3-sphere

4
S3 = {(x1 , x2 , x3 , x4 ) ∈ R4 | xi 2 = 1}.
i=1
We can also describe this sphere in terms of pairs of complex numbers z1 =

x1 + ιx2 , z2 = x3 + ιx4 as
S3 = {(z1 , z2 ) | |z1 |2 + |z2 |2 = 1}.
Let us take two positive integers p and q which are relatively prime. Now, we
2πιq
consider the complex number e p , whose modulus is clearly 1. Let us define
a map h : S3 → S3 by
2πι 2πιq
h(z1 , z2 ) = (z1 .e p , z2 .e p ).
Then, clearly, h is a continuous map. Furthermore, since hp is identity on S3 ,

h is in fact a homeomorphism of S3 of order p. It follows, as before, that the
group Zp = [h] acts on S3 . Again, the action is free and so the orbit of every
point has p distinct points. The orbit space S3 /Zp , denoted by L(p, q), is called
the lens space. The question is: how to visualize the lens space? Since S3 itself
cannot be easily visualized, the quotient space L(p, q) is still worse to imagine.
Note that the special case L(2, 1) is evidently RP3 , which contains RP2 . We
have already pointed out that RP2 cannot be embedded in R3 . Thus, we should
forget about visualizing the lens space and must rather study it on the basis of
its definition only. Being the orbit space of S3 under a free action, these lens
spaces have several interesting properties. One question which crops up right
now is the following: The lens space L(p, q) clearly depends on the pair (p, q)
of relatively prime integers. Then, is it true that L(p, q) are all distinct spaces
for different values of pairs (p, q)? Note that if (p, q1 ) and (p, q2 ) are two such
2πιq1 2πιq2
pairs so that q1 − q2 = kp, then e p = e p and hence, the two actions of
Zp on S3 with respect to the pairs (p, q1 ) and (p, q2 ) are same. This means that
L(p, q1 ) = L(p, q2 ). In fact, it is well-known that L(p, q1 ) ∼
= L(p, q2 ) if and only
if q1 ≡ ±q2 (mod p) or q1 q2 ≡ ±1(mod p) (see [16] p. 242).
Now, suppose a topological group G acts continuously on a space X. Fix

a point x ∈ X and consider the orbit G(x) of x. If Gx is the isotropy group of
the point x, then we can define a map
f : G/Gx → G(x)
simply by putting
f (gGx ) = g.x.
This is well defined since g1 Gx = g2 Gx implies that (g1 )−1 .g2 ∈ Gx , i.e.,
((g1 )−1 .g2 ).x = x, i.e., g1 .x = g2 .x. Clearly, the map f is continuous and
is a bijection. However, the map f need not be a homeomorphism. (see one

of the exercises later). In case the group G is compact and X is Hausdorff,
then, clearly, f is a homeomorphism. In other words, for compact groups G
acting on a Hausdorff space X, the orbits are simply homeomorphic to the
coset spaces of the type G/H, where H ⊂ G is a closed subgroup of G. An
action of G on X is said to be transitive if given any two points x, y ∈ X,
there exists a g ∈ G such that g.x = y. This really means that the space X
is just a single orbit and so X is homeomorphic to the coset space G/H for
some closed subgroup H of G. Such spaces are known as “homogeneous spaces”.
In our study of the covering projections and fundamental groups, we come

across the following special kind of group actions.
Definition 7.3.8. A group G of homeomorphisms of a space X is said to act on

X properly discontinuously if each point x ∈ X has a neighbourhood U such
that U ∩ g(U )
= ∅ implies that g = e, where e ∈ G is the identity element of G.
One can easily verify that any action of a finite group on a Hausdorff space
is properly discontinuous. Thus, except for the linear actions (see following ex-
ercises) of GL(n, R), O(n) and the irrational flow mentioned later in Exercises,
the rest of the examples are properly discontinuous.
Exercises
1. Suppose a connected topological group G acts continuously on a space X.

Prove that X is connected if and only if the orbit space X/G is connected.
2. Find an action of Z on R×[0, 1] such that the orbit space is homeomorphic
to the Möbius band.
If a noncompact group G acts on a topological space X, the orbit

G(x) of a point x ∈ X need not be homeomorphic to the quotient space
G/Gx . Even the orbit space X/G may have unexpected topology. The
following exercise provides a good example to show these pathological
situations.
3. (Irrational Flow) Let α be a fixed irrational number. We define an action

of the additive group R of real numbers on the torus T = S1 × S1 using
α as follows: Let (e2πix , e2πiy ) be an arbitrary point of T , and for t ∈ R,
put
t.(e2πix , e2πiy ) = (e2πi(x+t) , e2πi(y+αt) ).
Prove that the above is a free action of R on T and that every orbit in
T is a proper subset of T which is dense in T . Also show that the orbit
space T /R has the indiscrete topology.
(Hint: Imagine S1 as the quotient space R/Z so that the torus T = S1 ×S1
can be thought of as the unit square I × I in the plane R × R, where the
opposite sides of the square have been identified. Now, chase the orbit
of the point (1/2, 0) in the square to conclude that the orbit cannot be
locally connected whereas R is locally connected).
4. An n × n real matrix can be thought of as a point in the Euclidean space
2
Rn . Prove that the set GL(n, R) of all nonsingular n × n real matrices
2
with the Euclidean subspace topology from Rn forms a topological group
with respect to multiplication of matrices. Show that this group acts on
the Euclidean space Rn by multiplication, i.e., A · X is the product of
n × n matrix A ∈ GL(n, R) with the n × 1 column vector X ∈ Rn .
5. Prove that the group O(n) of all n × n orthogonal real matrices acts on
the unit disk Dn as well as on the unit sphere Sn under the action of
preceding exercise. Show that this action is transitive on Sn . Hence or
otherwise, deduce that O(n)/O(n − 1) ∼= Sn−1 .
References
[1] Agoston, M.K., Algebraic Topology: A First Course, Marcel Dekker, 1967.
[2] Bredon, G.E., Topology and Geometry, Springer-Verlag, 1997.
[3] Chinn, W.G., and Steenrod, N., First Concepts of Topology, Random
House, 1966.
[4] Dieudonné J., A History of Algebraic and Differential Topology (1900–
1960), Birkhäuser, 1989.
[5] Dugundji, J., Topology, Allyn and Bacon, 1966.
[6] Eilenberg, S. and Steenrod, N., Foundations of Algebraic Topology, Prince-
ton University Press, 1952.
[7] Greenberg, M. and Harper, J., Algebraic Topology- A First Course, Ben-
jamin/Cummings, 1981.
[8] Herstein, I.N., Topics in Algebra, John Wiley and Sons, 1975.
[9] Hilton, P.J. and Wylie, S., Homology Theory, Cambridge University Press,
1960.
[10] Hocking, J. and Young, G.S., Topology, Addison Wesley, 1961.
[11] Jacobson, N., Basic Algebra I, W.H. Freeman and Co., 1985.
[12] Lundell, A. and Weingram, S., The Topology of CW-Complexes, Van Nos-
trand Reinhold, 1969.
[13] Massey, W.S., Algebraic Topology: An Introduction, Springer-Verlag, 1967.
[14] Maunder, C.F., Algebraic Topology, Van Nostrand Reinhold, 1970.
[15] Munkres, J.R., Topology: A First Course, Prentice-Hall, 1975.
[16] Munkres, J.R., Elements of Algebraic Topology, Addison Wesley, 1984.
[17] MacLane, S., Homology, Academic Press, Springer-Verlag, 1963.
[18] Spanier, E.H., Algebraic Toplogy, McGraw-Hill, 1966.
[19] Seifert, H., and Threlfall, W., A Textbook of Topology (Translation from
German Edition of 1947), Academic Press, 1980.
[20] Whitehead, G.W., Elements of Homotopy Theory, Springer-Verlag, 1978.
Index
Abelian group, 295 Categories, 306

Abstract simplicial complex, 97 Chain
Acyclic, 242 n-chain, 324
Acyclic Model Theorem, 248 Chain complex, 324
Alexander, J.W., 170 Chain contraction, 254
Antipodal map, 170 Chain group, 127
Augmentation, 254 Class equation, 332
Augmentation map, 142 Coherent orientation, 155
Augmented chain complex, 254 Cohomology algebra, 289
Automorphism, 295 Cohomology theory, 260
Axiom Comb space, 46
Commutativity, 251 Combinatorial components, 102
Composition , 239, 250 Complex Projective Space, 13
Exactness , 251 Complex projective space, 335
Homotopy , 245, 250 Components, 21
Identity , 239, 249 Connected, 17
Excision, 252 n-connected, 111
Connected sum, 27
Barycentre, 107 Connecting homomorphisms, 258
Barycentric coordinates, 88 Contiguity classes, 178
Barycentric subdivision, 106, 178 Contiguity relation, 178
nth barycentric subdivision, Contiguous maps, 178
108 Continuous, 5
first barycentric subdivision, Continuous maps, 4
108 Contractible, 49
Betti number, 144 Contraction, 41
Betti, E., 144 Contravariant functor, 308
Bockstein homomorphism, 265 Convex, 89
Borsuk-Ulam theorem, 186, 191 Convex hull, 90
Boundary Convex set, 90
n-boundary, 324 Covariant functor, 61, 308
q-dimensional, 132 Covering homomorphism, 214
Boundary homomorphism, 129 Covering projection, 199, 200
Boundary map, 129 Covering transformations, 215
Brouwer’s fixed point theorem, 164 Cup product, 228
Brouwer’s fixed point theorem, 184 CW-complexes, 112
342 Index
CW-structure, 112 Finite Simplicial Complexes, 83

Cycle Finitely presented group, 302
n-cycle, 324 Free abelian group, 303
q-dimensional, 132 Free group generated by a set, 300
Free product of groups, 300
Deck transformations, 216 Functors, 308
Deformable, 47 Fundamental group, 35, 58, 59
Deformation, 47 of a circle, 72
Degree map, 76 Fundamental theorem of algebra,
Degree of a loop, 75 173
Degree of a map, 167 Fundamental theorem of homomor-
Dimension, 92 phism, 296
Dimension Axiom, 231, 232, 260
Direct product of groups, 297 General Simplicial Complex, 115,
Direct sum, 298 192
Direct sum of groups, 297 Generalized Barycentric Subdivi-
Dodecahedron, 152 sion, 193
Generalized homology theory, 228
Edge, 91 Generalized Simplicial Approxima-
Eilenberg and Steenrod axioms, 228 tion, 193
Eilenberg, S., 227 Geometric Realization, 116
Elementary q-chain, 128 Geometrically independent, 83
Endomorphism, 295 Group, 295
Euclidean
norm, 2, 74 Ham-Sandwich theorem, 191
space, 1 Homeomorphism, 6, 7
topology, 83 Homologous cycles, 133
topology, 2 Homology
Euler characteristic, 149 n-dimensional homology mod-
Euler’s theorem, 151 ule, 324
Euler, L., 149 Čech, 227
Euler-Poincaré theorem, 150 relative, 228
Exact homology sequence theorem, simplicial, 227
329 singular, 227
Exact sequence, 311 Vietoris, 227
Excision axiom, 257 Homology and Cohomology theo-
Excision theorem for singular ho- ries, 258
mology, 252 Homology cross product, 271
Exponential map, 75 Homology group, 123
Extension problem, 31 of a disc, 161
of a sphere, 161
Face of an n-disc, 146
p-dimensional face, 91 of an n-sphere, 148
Face operator, 230 of the Klein bottle, 162
Figure of eight, 203 of the Möbius band, 162
Index 343
of the projective plane, 162 Kuratowski’s non-planar graph, 99

of the torus, 162 Kuratowski, K., 98
reduced, 142
topological invariance, 160 Lebesgue’s covering lemma, 24
Homology group with coefficients, Lefschetz fixed point theorem, 182
132 Lefschetz number, 182
Homology of spheres, 260 Lefschetz Solomon, 182
Homology theory, 227, 259 Lemma
Homology with coefficients, 264 Continuity, 6
Homomorphism, 295 Five, 312
R-homomorphism, 304 Lens space, 336
Homotopy, 38 Lifting problem, 32, 205
path, 51 Lifting theorem, 213
relative, 51 Linearly independent, 83
Homotopy lifting property, 207 Locally compact space, 25
Homotopy class of a loop, 52 Locally connected space, 22
Homotopy equivalence, 42 Long line, 26
Homotopy lifting property, 73, 75, Lusternik-Schnirelmann theorem,
207 187
Homotopy type, 48, 50, 64
Möbius band, 11
Hopf’s classification theorem, 170
Möbius, A.F., 170
Hopf’s trace formula, 183
Manifold
Hopf, H., 169
topological n-manifold, 28, 154
Hyperplane
Map of pairs, 248
k-dimensional, 84
Mapping cylinder, 15
1-dimensional, 84
Mayer-Vietoris exact sequence, 273
Mayer-Vietoris sequence, 272
Icosahedron, 152
Mesh of a complex, 108
Incidence number, 125
Module
Integral singular homology, 231
R-module, 304
Invariance of dimension, 163
Monodromy theorem, 209
Invariance of domain theorem, 282
Irrational flow, 337 Natural transformation, 309
Isomorphism, 295 No-retraction theorem, 163
Noether, Emmy, 145
J.W. Alexander, 227 Normal subgroup, 296
Jordan curve theorem, 31, 278, 282 Null homotopic, 80
Jordan, C., 170
Jordan-Brouwer separation theo- Octahedron, 152
rem, 281 One-point compactification, 26
Open star of a vertex, 103
Künneth formula, 270 Orbit space, 333
Künneth theorem, 271 Orientability, 153
Künneth, K., 271 Orientable complex, 155
Klein bottle, 12, 335 Orientation, 124
344 Index
Oriented complex, 125 k-dimensional geometric sim-

Oriented simplicial chain complex, plex, 88
132 k-simplex, 88
open k-simplex, 91
Paracompact space, 28 Simplicial approximation theorem,
Path, 21 110
Path lifting property, 73, 75, 205, Simplicial approximation, 104
206 Simplicial approximation theorem,
Planar graph, 98 105, 106
Platonic solids, 152 Simplicial chain complex, 127
Poincaré group, 58 Simplicial complex, 92
Poincaré H., 149 boundary of, 100
Poincaré-Hurewicz theorem, 235 Simplicial Homology, 123
Polyhedra, 93 Simplicial isomorphism, 100
Polyhedron, 93 Simplicial map, 100
rectilinear, 151 Simply connected, 67, 111
rectilinear, 93 Singular
Product topology, 5 q-chain, 229
Projection map, 5 q-simplex, 229
Projective plane, 62, 114, 139 chain complex, 231
Properly discontinuous, 337 Singular chain complex, 229
Properly joined, 92 Singular chains of small sizes, 253
Pseudomanifold Singular cochain complex, 284
n-pseudomanifold, 153 Singular cohomology, 284
Punctured torus, 81 Singular homology, 231
Skeleton
Quotient space, 10 r-dimensional, 102
Quotient topology, 10 Space
contractible, 41
Radial map, 7 pointed, 35
Rank of a free abelian group, 303 semi-locally simply connected,
Real Projective Space, 12 225
Real projective space, 334 Star of a vertex, 103
Reduced cohomology, 285 Star related, 104
Reduced homology group, 241 Star-shaped, 41
Reduced singular homology mod- Steenrod, N., 227
ule, 241 Stereographic projection, 8
Retract, 47, 49 Strongly Deformable, 47
deformation, 48 Structure of a finite abelian group,
strong deformation, 48, 63 298
Retraction map, 47 Structure theorem for finitely gen-
Riemann, G.F.B., 154 erated modules, 305
Subdivision chain map, 175, 254
Separation axioms, 16 Surfaces, 28
Simplex, 3, 83 Suspension, 15
Index 345
Tangent vector field, 172 of the Klein bottle, 113

Tensor product, 315 of the Möbius band, 97
Theorem of the Torus, 99
Borsuk-Ulam, 124
Brouwer’s fixed point, 124 Uniformly continuous, 25
Lefschetz fixed point, 124 Uniqueness of homology theory, 268
Topological invariant, 33, 61 Universal coefficient theorem, 268
Topological pair, 248 Universal coefficient theorem for
Topological transformation group, cohomology, 287
333 Universal covering
Topologist’s sine curve, 19, 114 existence of, 225
Torsion coefficients, 144 Universal covering space, 220
Torsion subgroup, 144 Universal property, 315
Torus, 11, 37, 335
n-dimensional, 201 Van Kampen theorem, 79
Triangulable space, 93 Vector field, 172
Triangulation, 93
Weak topology, 17
of a disc Dn , 96
of a sphere Sk , 96

Satya Algebraic

Uploaded by

Copyright:

Available Formats

Satya Algebraic

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Satya Algebraic

Uploaded by

Copyright:

Available Formats

Satya Deo

ISSN 2366-8725 (electronic)

1st edition: © Hindustan Book Agency (lndia) 2003;

Algebraic Topology is an important branch of topology having several con-

This book is designed to serve as a textbook for a ﬁrst course in algebraic

We introduce and study the fundamental groups and its properties in

In this edition of ‘Algebraic Topology, A Primer’, some necessary changes

Earlier, the Chapter 6 on singular homology was really somewhat

1 Basic Topology: a review 1

2 The Fundamental Group 35

4 Simplicial Homology 123

5 Covering Projections 199

6 Singular Homology 227

Basic Topology: a review

1.2 Euclidean Spaces and their Subspaces

This is easily seen to be a metric on Rn . For any x ∈ Rn , the distance of x from

For any subset Y of Rn , we have a metric on Y induced by the Euclidean

Again, it is a compact, connected subset of the Euclidean space Rn+1 .

S1 = {x ∈ R2 | |x| = 1} is the set of all unimodular complex numbers. It is

For each n ≥ 1, the compact, connected subset Dn = {(x1 , . . . , xn ) ∈

Fig. 1.1(a): A 3-simplex

Note that the above procedure cannot be used to deﬁne a (n + 1)-simplex

1.3 Continuous Maps and Product Spaces

A map f : X → Y between two topological spaces X and Y is said to

In order to verify that a particular map f : X → Y is continuous, it is

There are several standard methods of constructing new topological spaces

where Uαi is an open set of Xαi , i = 1, 2, . . . , n. It is necessary to realize that

Another important observation which can be inferred from the description

A space X is embedded in the product space X × Y under the inclusion

X × Y and we don’t get anything new. In the case of Euclidean spaces, we

1.4 Homeomorphisms and Examples

The following simple result about continuity of maps will be frequently

Lemma 1.4.1. (Continuity Lemma). Let a space X be written as X = F1 ∪ F2 ,

Recall that a map f : X → Y is said to be a homeomorphism if f is a

Example 1.4.3. For each n ≥ 1, the boundary B(σ n ) of the n-simplex σ n is

Fig. 1.2: The radial map

Take a point P in the interior of σ n . If we draw a ray starting from P along

An interesting generalization of the foregoing example is discussed in

Example 1.4.4. Let Dn be an n-disk, I n be an n-cube and σ n be an n-simplex.

Example 1.4.5 (Stereographic Projection). Let X = S2 − {(0, 0, 1)}, where S2

be veriﬁed that f is a continuous bijection which has a continuous inverse.

Fig. 1.4: Stereographic Projection

homeomorphic to the whole real line, an inﬁnite cylinder is homeomorphic to

Fig. 1.5: Homeomorphic spaces

is not continuous, but the map g : R → R deﬁned by

1.5 Quotient Spaces

Quotient spaces are of fundamental importance in topology. The intuitive

Fig. 1.6: Circle as the quotient space of unit interval

Interesting quotient spaces can be obtained by above kind of identiﬁca-

Fig. 1.7: Cylinder as the quotient space of the unit square

Fig. 1.9: Möbius band as the quotient space of square

Fig. 1.9a: The Klein bottle as the quotient space of a square

As the ﬁnal example, we have the projective plane as a quotient

Proposition 1.5.1. Suppose Y is a quotient space of X with quotient map

Example 1.5.3. (Complex Projective Space) Consider the n-dimensional com-

Note that a quotient map ν : X → X/R, which is continuous and surjec-

Since q j = iq is continuous and Y /S has the quotient topology, the map

A quotient space of a Hausdorﬀ space need not be Hausdorﬀ. To see

Recall that an equivalence relation R is just a subset of X × X. The