Mathematics

AN ILLUSTRATED
INTRODUCTION TO

TOPOLOGY and HOMOTOPY

AN ILLUSTRATED INTRODUCTION TO

TOPOLOGY and HOMOTOPY

TOPOLOGY

AN ILLUSTRATED INTRODUCTION TO
An Illustrated Introduction to Topology and Homotopy explores the beauty of
topology and homotopy theory in a direct and engaging manner while illustrating
the power of the theory through many, often surprising, applications. This self-
contained book takes a visual and rigorous approach that incorporates both and
HOMOTOPY
extensive illustrations and full proofs.
The first part of the text covers basic topology, ranging from metric spaces and
the axioms of topology through subspaces, product spaces, connectedness,
compactness, and separation axioms to Urysohn’s lemma, Tietze’s theorems,
and Stone-Čech compactification. Focusing on homotopy, the second part starts
with the notions of ambient isotopy, homotopy, and the fundamental group. The
book then covers basic combinatorial group theory, the Seifert-van Kampen
theorem, knots, and low-dimensional manifolds. The last three chapters discuss
the theory of covering spaces, the Borsuk-Ulam theorem, and applications in
group theory, including various subgroup theorems.
Requiring only some familiarity with group theory, the text includes a large number
of figures as well as various examples that show how the theory can be applied.
Each section starts with brief historical notes that trace the growth of the subject
and ends with a set of exercises.
Features
• Provides a comprehensive treatment of basic topology and homotopy that
uses a combination of rigorous proofs and extensive visualization

KALAJDZIEVSKI
• Gives full details of most proofs
• Contains nearly 600 figures that help clarify difficult concepts
• Presents historical notes that outline the growth of the subject
• Includes about 750 exercises, many of which are relatively new

K12146
SASHO KALAJDZIEVSKI

w w w. c rc p r e s s . c o m

K12146_cover.indd 1 2/19/15 11:01 AM

 AN ILLUSTRATED
INTRODUCTION TO

TOPOLOGY
and
HOMOTOPY

K12146_Book.indb 1 2/10/15 4:51 PM

This page intentionally left blank
 AN ILLUSTRATED
INTRODUCTION TO

TOPOLOGY
and
HOMOTOPY

SASHO KALAJDZIEVSKI
University of Manitoba
Winnipeg, Canada

K12146_Book.indb 3 2/10/15 4:51 PM

CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
© 2015 by Taylor & Francis Group, LLC
CRC Press is an imprint of Taylor & Francis Group, an Informa business

No claim to original U.S. Government works

Version Date: 20141014

International Standard Book Number-13: 978-1-4822-2081-0 (eBook - PDF)

This book contains information obtained from authentic and highly regarded sources. Reasonable efforts have been
made to publish reliable data and information, but the author and publisher cannot assume responsibility for the valid-
ity of all materials or the consequences of their use. The authors and publishers have attempted to trace the copyright
holders of all material reproduced in this publication and apologize to copyright holders if permission to publish in this
form has not been obtained. If any copyright material has not been acknowledged please write and let us know so we may
rectify in any future reprint.

Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or uti-
lized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopy-
ing, microfilming, and recording, or in any information storage or retrieval system, without written permission from the
publishers.

For permission to photocopy or use material electronically from this work, please access www.copyright.com (http://
www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923,
978-750-8400. CCC is a not-for-profit organization that provides licenses and registration for a variety of users. For
organizations that have been granted a photocopy license by the CCC, a separate system of payment has been arranged.

Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for
identification and explanation without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com
and the CRC Press Web site at
http://www.crcpress.com
 To Timjan, Damjan, Darja and Nina

K12146_Book.indb 5 2/10/15 4:51 PM

This page intentionally left blank
 Table of Contents

Preface xi
Acknowledgment xv

Part I Topology

Chapter 1 n Sets, Numbers, and Cardinals 3

1.1 Sets and Numbers 3
1.2 Sets and Cardinal Numbers 9
1.3 Axiom of Choice and Equivalent Statements 12

Chapter 2 n Metric Spaces: Definition, Examples, and Basics 17

2.1 Metric Spaces: Definition and Examples 17
2.2 Metric Spaces: Basics 23

Chapter 3 n Topological Spaces: Definition and Examples 33

3.1 The Definition and Some Simple Examples 33
3.2 Some Basic Notions 39
3.3 Bases 48
3.4 Dense and Nowhere Dense Sets 57
3.5 Continuous Mappings 61

Chapter 4 n Subspaces, Quotient Spaces, Manifolds, and CW-Complexes 71

4.1 Subspaces 71
4.2 Quotient Spaces 75
4.3 The Gluing Lemma, Topological Sums, and Some Special
Quotient Spaces 81
4.4 Manifolds and CW-Complexes 90

vii

K12146_Book.indb 7 2/10/15 4:51 PM

 viii   ◾    Table of Contents

Chapter 5 n Products of Spaces 101

5.1 Finite Products of Spaces 101
5.2 Infinite Products of Spaces 110
5.3 Box Topology 117

Chapter 6 n Connected Spaces and Path Connected Spaces 121

6.1 Connected Spaces: Definition and Basic Facts 121
6.2 Properties of Connected Spaces 128
6.3 Path Connected Spaces 133
6.4 Path Connected Spaces: More Properties and Related
Matters 139
6.5 Locally Connected and Locally Path Connected Spaces 147

Chapter 7 n Compactness and Related Matters 153

7.1 Compact Spaces: Definition 153
7.2 Properties of Compact Spaces 159
7.3 Compact, Lindelöf, and Countably Compact Spaces 165
7.4 Bolzano, Weierstrass, and Lebesgue 170
7.5 Compactification 173
7.6 Infinite Products of Spaces and Tychonoff Theorem 179

Chapter 8 n Separation Properties 185

8.1 The Hierarchy of Separation Properties 185
8.2 Regular Spaces and Normal Spaces 191
8.3 Normal Spaces and Subspaces 195

Chapter 9 n Urysohn, Tietze, and Stone–Čech 199

9.1 THE Urysohn Lemma 199
9.2 The Tietze Extension Theorem 205
9.3 Stone–Čech Compactification 209

Part 2 Homotopy

Chapter 10 n Isotopy and Homotopy 217

10.1 Isotopy and Ambient Isotopy 218
10.2 Homotopy 228
10.3 Homotopy and Paths 232
10.4 The Fundamental Group of a Space 238

K12146_Book.indb 8 2/10/15 4:51 PM

 Table of Contents   ◾    ix

Chapter 11 n The Fundamental Group of a Circle and Applications 247

11.1 The Fundamental Group of a Circle 247
11.2 Brouwer Fixed Point Theorem and The Fundamental
Theorem of Algebra 255
11.3 The Jordan Curve Theorem 261

Chapter 12 n Combinatorial Group Theory 269

12.1 Group Presentations 269
12.2 Free Groups, Tietze, Dehn 275
12.3 Free Products and Free Products with Amalgamation 283

Chapter 13 n Seifert–van Kampen Theorem and Applications 291

13.1 Seifert–van Kampen Theorem 291
13.2 Seifert–van Kampen Theorem: Examples 299
13.3 Seifert–van Kampen Theorem and Knots 310
13.4 Torus Knots and Alexander’s Horned Sphere 319
13.5 Links 327

Chapter 14 n On Classifying Manifolds and Related Topics 333

14.1 1-Manifolds 333
14.2 Compact 2-Manifolds: Preliminary Results 336
14.3 Compact 2-Manifolds: Classification 343
14.4 Regarding Classification of CW-Complexes and Higher
Dimensional Manifolds 356
14.5 Higher Homotopy Groups: A Brief Overview 365

Chapter 15 n Covering Spaces, Part 1 371

15.1 Covering Spaces: Definition, Examples, and Preliminaries 371
15.2 Lifts of Paths 375
15.3 Lifts of Mappings 379
15.4 Covering Spaces and Homotopy 388

Chapter 16 n Covering Spaces, Part 2 397

16.1 Covering Spaces and Sheets 397
16.2 Covering Transformations 400
16.3 Covering Spaces and Groups Acting Properly
Discontinuously 405

K12146_Book.indb 9 2/10/15 4:51 PM

 x   ◾    Table of Contents

16.4 Covering Spaces: Existence 415

16.5 The Borsuk–Ulam Theorem 421

Chapter 17 n Applications in Group Theory 429

17.1 Cayley Graphs and Covering Spaces 429
17.2 Topographs and Presentations 436
17.3 Subgroups of Free Groups 442
17.4 Two Subgroup Theorems 448

Bibliography 455

List of Symbols 459

Index 463

K12146_Book.indb 10 2/10/15 4:51 PM

 Preface

Chronology. In the year 2000 I was given two courses to teach: analysis and topology.
The Analysis course was assigned to me due to an emergency redistribution; Topology
was thrown in as an award for accepting to teach Analysis on such short notice. Algebraic
topology, which covered only homotopy theory in its curriculum, was (re)introduced into
the undergraduate-graduate program of the Mathematics Department (University of
Manitoba) in 2004. I have taught these two courses almost continuously ever since. This
book is primarily an upshot of my teaching lifestyle over these years.

The level of the book, and the target audience. The book has two clearly differentiated
parts: Part 1 is topology and Part 2 is homotopy. As indicated by the title of the book, we
do not have any pretense to go very deeply into these subjects. On the other hand, neither
could the content be described as too breezy, for we include most of the important theo-
rems of the basic theory.
The material covered in this book is on the fuzzy boundary between the undergradu-
ate and the graduate level. One may tentatively state that Part 1 is mostly at the advanced
undergraduate level, Part 2 is mostly at the early graduate level. Indeed, the Algebraic
topology (Homotopy theory) course that I have been teaching has always been cross-listed,
and has almost always been taken by both graduate and advanced undergraduate students.
It is my hope that this book will bring this beautiful theory closer to the undergraduate
curriculum.

Content of the Book

Part 1. Part 1 begins with a brief introductory chapter on sets and (cardinal) numbers,
followed by an equally short chapter about metric spaces. Metric spaces, and metrizable
spaces, are discussed (along the way) in the rest of Part 1. The proper topology theory starts
with the axioms given at the beginning of Chapter 3. The development of the theory from
that point on follows the classical pattern. The basic notions are introduced in Chapter 3,
followed by increasingly more complex ideas and properties (Chapters 4–7). The theory
given in Part 1 culminates with Tychonoff’s theorem (7.6) and a few other significant results
(Chapters 8 and 9). Various applications are discussed as they emerge from the theory.
The connection between Part 1 and Part 2. The basic concepts (as are continuity, bases,
connectedness, and compactness) from Part 1 are used extensively in Part 2. There are a
number of simple results (for example, that the components of open sets in locally path
xi

K12146_Book.indb 11 2/10/15 4:51 PM

 xii   ◾    Preface

connected spaces are open), and more significant theorems (for example, Tietze’s theorem
on extending functions) that are invoked in Part 2. However, the intellectual flavors of the
two parts of this book are rather different, primarily because of the extensive link in Part 2
between topology theory and group theory.
Part 2. The central theory in this part begins with the concept of homotopy. The introduc-
tory discussion of homotopy covers most of the first two chapters in Part 2; this is where
we see the introduction of group theory through the notion of the fundamental group of a
space. This initial part peaks with the proof that the fundamental group of a circle is infi-
nite cyclic. It is rather amazing that this one, seemingly obvious fact yields such a diverse
array of applications, as are, among others, the Brouwer fixed point theorem, the funda-
mental theorem of algebra, and the Jordan curve theorem.
In order to access the core theory of Part 2 one needs combinatorial group theory. Group
presentations and the underlying basic theory are introduced in Chapter 12.
The statement and the proof of the crucial Seifert–van Kampen theorem given in Chapter
13 rely on Chapter 12. The rest of Chapter 13 is devoted to applications of this theorem.
The classification theorem for compact connected 2-manifolds is proven in Chapter
14. The proof is incomplete: due to technical reasons, and rather reluctantly, I have
decided against my initial plan to include a proof of Radó’s theorem on triangulability of
2-manifolds in an appendix of the book. This chapter ends with an overview of higher-
dimensional manifolds and higher homotopy groups.
Two chapters are devoted to covering spaces and their immediate applications. The
interaction between groups and topological spaces is the most interesting in this last part
of the book (Chapters 15–17). It is remarkable (but certainly not unique) that, through the
theory of covering spaces, one needs groups to prove such a glaringly topological statement
as is the Borsuk–Ulam theorem.
Applications of covering spaces theory in group theory are given in the last chapter (17).
This part reflects my own mathematical affinities.

Remarks Regarding Using This Book as a Textbook

Topology (Part 1). The material in Part 1 roughly fits a 1.5-semester course. Depending
on the audience, it could easily fit into a full year course. Typically, during these years of
teaching topology as a one-semester course, I cover the following in my lectures: the entire
Chapter 3, Sections 4.1 and 4.2, small parts of 4.3 and 4.4, 5.1, 5.2 and the definition of Box
topology in 5.3, Chapter 6, 7.1, 7.2 and large parts of 7.3, 7.4 and 7.5, and, finally, 8.1 and 8.2.
For the remaining teaching time, if any, I choose one, or at most two of 7.6 (Tychonoff), 8.3
(Tychonoff plank), 9.1 (Urysohn), 9.2 (Tietze) and 9.3 (Stone–Čech).
Homotopy (Part 2). Part 2 also has more material than what would fit in a reasonably paced
one-semester course. Here is one recommendation for a one-semester course based on my
personal experience: 10.1 (briefly), 10.2–10.4, 11.1 and 11.2, 11.3 (briefly or time permitting),
12.1–12.3, 13.1 and 13.2, then 13.3 (through a couple of examples and a short overview)
13.4 (one example of a torus link; I usually have time only to mention Alexander’s horned

K12146_Book.indb 12 2/10/15 4:51 PM

 Preface   ◾    xiii

sphere), 13.5 (one example); 14.2 and 14.3 (I never bother much with 14.1 and 1-manifolds),
14.4 (overview) and 14.5 (overview); the entire sections 15.1–15.3, and parts of 15.4; 16.1–
16.3 (almost entirely); I rarely have enough time to prove the existence theorem in 16.4;
16.5 is usually done completely. By the end, there may be time to show one application in
group theory. I often choose the claim that subgroups of free groups are free (17.3).
Parts 1 and 2. A year-long Topology-Homotopy course is also an option, but I have never
done it.
The various kinds of beauty of the subject. One does not have to be over-enthusiastic to
declare that the mathematical subjects we cover here are of exquisite and diverse beauty.
This aesthetic richness does not end within the intellectual realm. There are about 500
figures or illustrations in the book. With that we justify the word ‘illustrated’ in the title.
The inclusion of so many figures is partly a consequence of my personal affinity for visual-
ization, partly in order to make some of the arguments, ideas and concepts more evident,
and finally, to hint at that “other beauty” of the theory we cover. Most often the first two
motives dominate. In a rather small number of instances (say, in case of fractals, Section
2.2) some graphics are included mainly because of their perceived aesthetic merit.
Exercises in the book. There are a total of 756 problems in this book, 435 in Part 1, and
321 in Part 2. Most of the exercises came my way as I was writing the theory, so some of
them should be new. The exercises are rather roughly ordered according to their level of
difficulty in each section. Occasionally the exercises at the end of a section follow the chro-
nology of the topics of that section, in which case easier exercises may appear later.
Historical notes in the book. There is also one more somewhat unorthodox feature in
this book: I preface almost every section with a brief historical remark. The aim of these
remarks is to indicate the chronology of the development of the theory, and also to men-
tion the names of some remarkable mathematicians who contributed in important ways.
These encapsulated historical remarks are very laconic, but my hope is that they serve the
indicated purpose.
Regarding the Style. I have not chosen the style of this book; it chose me. It manifests, for
example, through my fondness of pictures, and through my resentment of overly condensed
proofs, sometimes called “short proofs”, where it may be hard to easily separate obvious
implicit claims from the gaps in the argument of various sizes. Consequently, I have tried
to avoid short proofs of such kind. I can see both the advantages and the disadvantages, but
I would rather say a few words about the former. I try, most of the time, to give complete
proofs of the claims we set out to prove completely. Along the way, we will inevitable encoun-
ter obvious, repetitive, or easy arguments. Sometimes the latter are relegated to the set of
exercises. Most importantly, when we skip a step during an argument, we will announce it
explicitly most of the time. This slightly—but not steeply—diminishes in frequency along
the way, after some of the often-mentioned results become sufficiently folklorized.
Technical Aspects. Most of the statements in this book are labeled, and prefixed in the
standard way. Thus we usually use the word “theorem” as an attribute to a more difficult,

K12146_Book.indb 13 2/10/15 4:51 PM

 xiv   ◾    Preface

or (mathematically or historically) more important statement. Most of the time we prefix

the statements with the neutral term “proposition”. The proofs of the labeled claims will
alternatively end with filled (black) squares ◼. The examples we cover more thoroughly
will also be clearly separated from the rest of the text, and the end of the related discussion
will usually be indicated with the white square ☐.
I used PovRay to make most of the 3D graphics, sometime in combination with
Mathematica; exceptionally I used EI Animator. For 2D graphics and for processing 3D
graphics I worked with Adobe Illustrator. Labels within the graphics were mostly done
with MathType. Images from Art Explosion 525 000 were used in Figures 1.2.2, 1.3.1, 2.1.9,
4.3.2, 4.3.3, and 6.5.2.

I will maintain a web page dedicated to this book:

http://home.cc.umanitoba.ca/~sasho/sk/topology_homotopy.html
The web page contains the figures from the book, as well as additional graphics that did not
make the final cut. One advantage is that many of the figures, rendered gray in the book
because of cost-related issues, could now be viewed in their original color versions. Posting
the figures here also makes them available for usage in lectures. I will eventually add short
animations. In the web page I also include various, ostensibly non-mathematical tidbits
(say, an insert of Calvino’s All At One Point).

K12146_Book.indb 14 2/10/15 4:51 PM

 Acknowledgment

The very first course I taught as a teaching assistant was topology, under the mentorship
of Smilka Zdravkovska. My own approach in teaching topology during the last decade or
so was also influenced by Murray Bell. I am indebted to both of them.
Thanks to Chris Camfield for a review of the project in its early stages, to Yemon
Choi for pointing to me an application of Tychonoff’s theorem in Ramsey Theory, and to
David Gunderson for sharing his library and for drawing my attention to a solution of
a combinatorial problem. Many thanks to Adam Clay for pointing some inaccuracies as
well for suggesting a few exercises.
I am grateful to the following people for helping me with the mathematical content
of the book. Grant Woods read a large portion of Part 1 of the book; his comments and
suggestion were most valuable to me. Jaydeep Chipalkatti was very kind to read Part 2 of
the book: his lucid remarks impacted the treatment of the whole book.
I am very grateful to Allan Edmonds for his extensive review of both Part 1 and Part 2
of the book. Moreover, he was extremely gracious to allow me to be in contact with him
during the completion of the project, replying thoroughly and promptly to many questions
and doubts that I had.
Special thanks to Nina Zorboska, who was a discreet contributor of many ideas and
exercises. My in-home editor Darja Kalajdzievska was invaluable to me during this project.
Thank you very much Dar!
Last, but not least, I thank my many students, whose collective input made the most
profound impact on this project. Thanks to James Requeima, Clint Enns, Trevor Wares,
Adam Bookatz, Iian Smythe, Max Bennett, Blake Madill, Andrii Arman, Mykhailo
Akhtariiev, and special thanks to David Gabrielson (who was kind to give me Latex-ed
notes of my early topology lectures), Karen Johannson (for sharing with me her carefully
handwritten notes of my early algebraic topology lectures), Ivan Levinski (for pointing
to me some results I was not aware of), Xiangui Zhao (who found a counterexample
to a false claim), Danylo Radchenko (for a few excellent proofs and exercises), Ievgen
Bilokopytov (who profoundly influenced one section of this book), Angel Daniel Barría
Comicheo (for his commentaries on some exercises), and Damjan Kalajdzievski (for his

xv

K12146_Book.indb 15 2/10/15 4:51 PM

 xvi   ◾    Acknowledgment

contributions and for his illuminating and uncompromising critique of “visual proofs”
in general). Thanks to Timana too.
Bob Stern, Taylor & Francis editor, was very supportive and helpful from the start to
almost the end of this project. Also many thanks to Suzanne Lassandro for her help and
patience during the proofreading stages of the project.
Jos Leys is the author of the artwork in Figures 14.3.24 and 16.3.2. He was also very
generous to share with me some of his PovRay codes, and some of the tricks of his trade.
I’ve made use of a few crop circles. Thanks to their creators.

K12146_Book.indb 16 2/10/15 4:51 PM

 I
Topology

K12146_Book.indb 1 2/6/15 5:11 PM

This page intentionally left blank
 Chapter 1

Sets, Numbers, and Cardinals

W e start with a brief review of some basic notions of set theory. This will be done in a
mostly informal manner, for a precise axiomatic set-theoretical approach requires
much more than what could be covered in one preliminary chapter. However, it should be
understood that these set-theoretical notions lie in the deeper basis of topology or, for that
matter, of virtually every mathematical theory. A good introductory book on the subjects
in this chapter is [29].

1.1 Sets and Numbers

A brief historical note: The notions of sets and numbers are ancient. The first rigorous defini-
tion of the set of real numbers (1871) is attributed to Georg Cantor (1845–1918).

The term “set” will be occasionally replaced by “family,” “class,” or “collection,” especially
when there is a hierarchy of sets under inclusion, and for the purpose of avoiding the self-
referential, paradox-generating notions such as “sets of all sets.” The reader may assume
that these four terms are interchangeable throughout this book.
We will use the standard notation of basic set theory. Thus, ∅ (the empty set), ⊆ (subset),
 (proper subset), ∈ (belongs to), ∪ (union), ∩ (intersection), and Ac (the complement
X \ A of A in X) all have the usual meaning. The only exception is ⊂, which will take the
role of ⊆ in this book (thus, making the latter unnecessary). This is justified by the fact that
we will rarely be encountering . In the few cases where we will want to emphasize that a
set A is a proper subset of a set B we will indicate it.

K12146_Book.indb 3 2/6/15 5:11 PM

 4   ◾    Sets, Numbers, and Cardinals

Basic results in set theory, such as de Morgan’s laws, are assumed to be known to the reader.
We will consistently use the following notation:
 = {0,1,2,…} is the set of natural numbers.
 = {…, −2, −1,0,1,2,…} is the set of integers.
+ = {1,2,…} is the set of positive integers, or the set of positive natural numbers.
Given two sets A and B, their set product is the set A × B = {(a, b): a ∈ A, b ∈B} . Instead
of A × A, we will sometimes
n
write A2 . The set product of the sets X1 , X 2 ,… , Xn , denoted
X1 × X 2 ×  × Xn , or Π Xi , is the set {( x1 , x 2 , … , xn ) : xi ∈ Xi , i = 1,2, … , n}. As in the case
i =1 n
for two sets, if X1 = X 2 = … = Xn = X , then we will sometimes use X n as short for Π Xi .
i =1
A mapping f : X → Y is a subset S of X × Y such that for every x ∈ X, there is exactly one
y ∈Y, such that ( x , y ) ∈S. We will use the standard notation and write y = f ( x ) instead of
( x , y ) ∈S. Surjection (onto mapping), injection (one-to-one mapping), and bijection will have
the usual meaning. Occasionally we will use the word “map” instead of “mapping.” If Y is the
set of real numbers (to be defined further below), then we will say function, instead of mapping.
A relation R on the set A is a subset of A × A. Instead of ( x , y ) ∈R we will more often write
x R y.
We list the standard terminology regarding a relation R on a set X:

• reflexive, if for every x ∈ X , ( x , x ) ∈R,

• irreflexive, if for every x ∈ X , ( x , x ) ∉R,
• symmetric, if for every x , y ∈ X , if ( x , y ) ∈R then ( y , x ) ∈R,
• antisymmetric, if for every x , y ∈ X, if ( x , y ) ∈R and x ≠ y then ( y , x ) ∉R,
• transitive, if for every x , y , z ∈ X , if ( x , y ) ∈R and ( y , z ) ∈R, then ( x , y ) ∈R.

An order of a set X is a relation that is antisymmetric and transitive; the set X is then
ordered by such a relation. Order relations will usually be denoted by ≤. A strict order,
denoted <, is an irreflexive order; this means that x < y if and only if x ≤ y and x ≠ y . We
will often omit the attribute “strict,” since the notation implies it. If x < y , we will say
that x is smaller than y. Given an element a in an ordered set X, a predecessor of a is any
x ∈ X such that x < a. A linear order is a strict order such that for every distinct x , y ∈Y,
either x < y or y < x . For example, 1 < 2 < 3 < 4 <  < n < n + 1 <  is the standard linear
order of +.
The following notation is common for an ordered set X:

• (a, b) is the set of all elements x in X such that a < x < b,

• [a, b) is the set of all elements x in X such that a ≤ x < b,
• (a, b] is the set of all elements x in X such that a < x ≤ b,
• [a, b] is the set of all elements x in X such that a ≤ x ≤ b.

K12146_Book.indb 4 2/6/15 5:12 PM

 1.1 Sets and Numbers   ◾    5  

The set (a, b) is called an open interval in X, the sets [a, b) and (a, b] are semi-open or half-
open intervals in X, and the set [a, b] is a closed interval in X. We notice the ambiguity:
(a, b) denotes both an open interval, and a pair of numbers. Which of these two meanings
is used should be clear from the context.
An equivalence relation on a set X is a relation that is reflexive, symmetric, and transi-
tive. For equivalence relations R we will often use the symbol ~, and in that case, instead of
( x , y ) ∈R we will write x ~ y.

Example 1
Let X and Y be any sets and f : X → Y be a mapping. Define a relation ~ on X as
follows: x ~ y if f ( x ) = f ( y ). It is easy to see that this is an equivalence relation: it is
reflexive since for every x ∈ X we have f ( x ) = f ( x ), it is symmetric since if f ( x ) = f ( y )
then f ( y ) = f ( x ), and it is transitive since f ( x ) = f ( y ) and f ( y ) = f ( z ) imply that
f ( x ) = f ( z ). We will see this equivalence relation several times later on. ☐

Example 2
Let X be any nonempty set. A family P = { Xi : i ∈I } of nonempty subsets of X is a
partition of X if X j ∩ X k = ∅ for every j ≠ k, and if ∪ Xi = X .
i∈I

X3
X
X2
X1
X4

X5
X8 X9
X6
X7 Illustration 1.1 We show a partition of X into
X11
X10 11 disjoint sets. Most of the time further on,
the sets X and P will be infinite.

Every partition of a set X defines an equivalence relation ~ as follows: x ~ y if both x and y

belong to the same Xi ∈P . ☐

The converse of the last claim in Example 2 is also true, as we explain. Given an equiva-
lence relation on a set X, the equivalence class [x] of an element x ∈ X is the subset [x ] =
{ y ∈ X : x ~ y} of X. The family of all equivalence classes of the elements of X is a partition of
X (Exercise 4). Thus, every equivalence relation generates a partition of X into its equivalence
classes, and every partition defines an equivalence relation as in Example 2. Hence we can
view equivalence relations as being the same as partitions of the underlying set X.
The set of all equivalence classes of the elements of X will be called the quotient set, and
will be denoted by X . We may think of X as being obtained from X by shrinking every
~ ~
equivalence class to a point.

K12146_Book.indb 5 2/6/15 5:12 PM

 6   ◾    Sets, Numbers, and Cardinals

Illustration 1.2

X X X3 X~ X3
X2 X1 X2
X1 X4 X4

X5 X5
X8 X9 X8 X9
X6 X6
X7 X7
X10 X10
X11 X11

Start with a set X. Partition X into disjoint Shrink each disjoint subset of
subsets. the partition into a point to
get the quotient set X ~.
Aiming at a certain traditional notation, we temporarily denote pairs (a, b) ∈ × ( \ {0})
by ba . The set of rational numbers, denoted by  , is the set of equivalence classes [ ba ],
a, b ∈, b ≠ 0, where the equivalence is defined by ba ~ dc if ad = bc. As is customary, instead
of ~ we will simply write =, and instead of [ ba ] we will use the standard ba . With that we can
write  = { ba : a, b ∈, b ≠ 0}.
We will now outline one construction of the set of real numbers. Since we will not refer
to this construction, the reader may safely skip it, all the way to the Least Upper Bound
Property a few paragraphs further below.
An infinite sequence a1 , a2 ,…, an ,… of rational numbers is a Cauchy sequence if for
every rational number ε > 0 there is an integer N such that if n, m > N , then an − am < ε .
We define an equivalence relation on the set of all Cauchy sequences: two Cauchy sequences
a1 , a2 ,…, an ,… and b1 , b2 ,…, bn ,… are equivalent if for every rational number ε > 0 there is
an integer N such that if n > N , then an − bn < ε . It is straightforward to show that this is
indeed an equivalence relation.
The set of real numbers, denoted by  from now on, is the set of equivalence classes
of Cauchy sequences of rational numbers (under the just defined equivalence). Each real
number can be thus represented by any Cauchy sequence in the corresponding equivalence
class. Integers, and more generally, rational numbers can be represented by constant Cauchy
sequences. For example, a Cauchy sequence representative of the integer 2 is 2,2,…,2,… .
Loosely speaking, a Cauchy sequence of rational numbers converges toward the real
number it defines, where we are using the intuitive meaning of the term “converge” (and
where we will evade explaining any meaning of the word “intuitive”).
The skeleton of the construction of the basic structure of  is outlined in the next list of
steps where we denote by [an ] the real number (equivalence class) defined by the Cauchy
sequence (an ) = a1 , a2 ,…, an ,… .

• Define addition, multiplication, and subtraction in a natural way. For example

[an ] + [bn ] = [an + bn ] .
• Define a real number [an ] to be positive if there is a positive rational number q such that,
from some point on, the members of the Cauchy sequence (an ) are all larger than q.
• Define an order < as follows: [an ] < [bn ] if [bn ] − [an ] is positive.

K12146_Book.indb 6 2/6/15 5:12 PM

 1.1 Sets and Numbers   ◾    7  

The last property gives us the usual, linear order of the set of real numbers. We will say a
few more things about this order in the next section.
Each positive real number r = [an ], as well as r = 0 , can be represented in a unique way
as an infinite decimal (sequence) n . d1d2 …dn … where n ∈ , and 0 ≤ di ≤ 9 for all i ∈+ ,
such that no right-hand side portion of that infinite sequence consists only of 9s. Here is the
idea of how that is done (without going into details): Start with a real number r. There exists
the largest integer n such that n ≤ r . Next, consider r − n: by construction it is a number in
the interval [0,1), and so 0 ≤ 10(r − n) < 10 . Let d1 be the largest integer between 0 and 9 such
that d1 ≤ 10(r − n). So we have 0 ≤ 10(r − n) − d1 < 1, and thereby 0 ≤ 10(10(r − n) − d1 ) < 10.
Now find the largest integer d2 between 0 and 9 such that d2 ≤ 10(10(r − n) − d1 ) . Keep going!
This procedure gives the representation of positive real numbers as (what we call) decimal
numbers. For example, the decimal expression of the real number [9/10, 99/100, 999/1000,
…] is 1.000….
The set of positive real numbers will be denoted by  + .
Once we have the positive real numbers represented as decimals, it is easy to do the same
to the negative numbers. This gives the well-known decimal representations of real numbers.
In this context, rational numbers are real numbers that are expressible as either termi-
nating decimals (the decimal digits are all 0 from some point on, as in 3.24300000…) or as
periodic decimals (where the digits of the decimal from one point on are made of a repeat-
ing finite sequence of digits (for example, 1.23245454545(45)…). Irrational numbers are
the real numbers that are not rational.
We will often invoke the standard visualization of  as a line equipped with dis-
tance to a fixed point on that line. As a consequence,  2 is a plane with a coordinate
system, etc.

Illustration 1.3 A part of  3 (a solid cube) cutting out a planar area (a part of  2 ) and a line
segment (a part of ), all of them embedded in a space called Old Mills Beach. And yes, there
will be unnecessary illustrations in this book!

K12146_Book.indb 7 2/6/15 5:12 PM

 8   ◾    Sets, Numbers, and Cardinals

In what follows in this section we draw attention to the Least Upper Bound Property,
and its twin, the Greatest Lower Bound Property, since they are behind virtually every topo-
logical property of the set . The set , endowed with increasingly richer structure as we
go, will be the blueprint model in the development of our theory. The Least Upper Bound
Property is sometimes given as an axiom in the formal theory of real numbers, and that is
how we should take it.
A subset A of an ordered set X is bounded from above if there is an x ∈ X , such that for
every a ∈ A, we have a ≤ x ; if there is a y ∈ X , such that for every a ∈ A, y ≤ a, then we say
that A is bounded from below.

The Least Upper Bound Property. Every subset of  that is bounded from above has the
least upper bound.

Explicitly, this means that for every subset A of , if there is a real number r such that a ≤ r for
every a in A, then there is a number c in  such that c is the smallest of all such numbers r.
The number c in the statement of the Least Upper Bound Property is called the supremum
of the set A and will be denoted by sup(A).

Here is the Least Upper Bound Property’s twin:

The Greatest Lower Bound Property. Every subset of  that is bounded from below has
the greatest lower bound.

This greatest lower bound is called the infimum of the associated set A, and it is denoted
by inf (A).

Exercises
1. Find a set A such that for every a ∈ A, we have a ⊂ A.
2. Given a set X, show that the relation ⊂ is an order of the set of all subsets of X. For
which sets of X is this order linear?
3. Describe a linear order on (a) the set  2 , and (b) the set  2.
4. Show that if ~ is an equivalence relation on a set X, then every two equivalence
classes are either disjoint or equal.
5. Which of the following relations are equivalence relations on the set ? If a relation
is an equivalence relation, describe the equivalence classes.
(a) x ~ y if x − y is rational.
(b) x ~ y if x − y is irrational.
(c) x ~ y if x − y is divisible by 5 (that is, if x− y
5
is an integer).

K12146_Book.indb 8 2/6/15 5:12 PM

 1.2 Sets and Cardinal Numbers   ◾    9  

6. Is there a relation R on the set , such that R is different from the relation =, and such
that R is both an order and an equivalence relation?
7. Let X be a nonempty set and let f : X → Y be any mapping. Show that “u ~ v if and
only if f (u) = f (v )” defines an equivalence relation on X.

1.2 Sets and Cardinal Numbers

A brief historical note: The first proof that the set of real numbers has more elements than the
set of integers was given by Georg Cantor (1874). The claim that there is no set with cardinality
strictly between ℵ0 and 2ℵ0 is known as the Continuum Hypothesis.

Two sets A and B have the same cardinality if there is a bijection (one-to-one and onto
mapping) from one set onto the other set. The cardinality of a set A, denoted by A , is the
class of all sets of the same cardinality as the set A.
A set A is finite if there is an n ∈+ such that A has the same cardinality as the set {1,2,…, n},
or if it is the empty set. In the former case we denote A = n , and we will say that A has n ele-
ments; in the case when A = ∅, we write A = 0 . An infinite set is any set that is not finite,
and when A is infinite we will write A = ∞ . A set A is countably infinite if it is of the same
cardinality as the set  . In that case we will write A =ℵ0 (aleph zero). A set is countable if it
is either finite or countably infinite. An uncountable set is a set that is not countable.

Example 1
The set  is countably infinite because the mapping that sends n ∈  to the element
that appears at position number (n + 1) in the infinite list (specified by the following
pattern) 0, −1,1, − 2, 2, − 3, 3,… is a bijection from  onto . ☐

As this example indicates, in order to show that a set A is countably infinite, it suffices to
exhibit a well-defined sequence a0 , a1 , a2 ,…, an ,… that has no repetitions and consists of all
elements of A (in which case, obviously, n  an is a bijection from  onto A). It follows
that every infinite subset of a countably infinite set is also countably infinite.
We define a linear order on the class of all cardinal numbers as follows: A ≤ B if there
exists a one-to-one mapping from A into B. It is straightforward to show that this relation
is well defined, reflexive and transitive. Proving that it is antisymmetric is not so elemen-
tary and we will skip it.

Theorem 1. (Cantor–Bernstein1) If A ≤ B and B ≤ A , then A = B .

Our definition of the relation ≤ among cardinals was given in terms of one-to-one mappings.
Onto mappings could have been used to the same effect:

Proposition 2. A ≤ B if and only if there is an onto mapping g : B → A.

1
Also known as Schröder-Bernstein theorem.

K12146_Book.indb 9 2/6/15 5:12 PM

 10   ◾    Sets, Numbers, and Cardinals

Proof. (Warning: in the second part of the following argument we use a seemingly harm-
less claim, which we will identify following the proof.)

⇒ Suppose A ≤ B , so that there is a one-to-one mapping f : A → B. The mapping f −1 :

f ( A) → A is then well defined. Define g : B → A to be f −1 over f (A) and arbitrary over
B \ f ( A). Then g is obviously onto.

⇐ Suppose there is an onto mapping g : B → A. We will show that there is a one-to-one

mapping f : A → B, thus establishing that A ≤ B . Define f in terms of g as follows: for
every a ∈ A , choose precisely one element ba from the subset g −1 (a) of B, and set f (a) = ba.
Since we are assuming that g is onto, we have that g −1 (a) ≠ ∅ for every a ∈ A , and so this
definition of f makes sense. It is evident that f is one-to-one, and so A ≤ B .

We have deduced what we claimed—by closing our eyes for a moment and claiming the
obvious—that we can choose one element from each nonempty set g −1 (a). This claim is the
Axiom of Choice. We will come back to it in the next section. For the time being we merely
note that by using it we are acknowledging our acceptance of it.
In Illustration 1.4, we use Proposition 2 to exhibit the standard justification of the claim
that the set of positive rational numbers has cardinality ≤ than ℵ0 . Since the opposite
inequality is obvious, it follows that this set of positive rational numbers is countably infi-
nite. One consequence (via a short argument) is that the set  of all rational numbers is
also countably infinite (i.e., has cardinality ℵ0 ).

1 1 1 1
1 2 3 4
2 2 2
1 2 3 Illustration 1.4 Following the arrows we
3 3 en­cou­nter all positive rational numbers +
1 2
(with rep­etition), listed according to the sum
of the numbers in the numerator and denomi-
4
nator. This defines a mapping from the set 
1
onto + showing that + ≤  .

The power set of a set A is the set of all subsets of A; it will be denoted by P ( A). The
following simple but important proposition establishes a way of producing large cardinals.

Proposition 3. For every set A, we have A < P( A) .

Proof. Since a  {a} is a one-to-one mapping A → P ( A), it follows that A ≤ P( A) .

Suppose | A| = |P ( A)|, i.e., suppose there is a bijection f : A → P ( A). Consider the subset
B = {b ∈ A : b ∉ f (b)} of A. Since f is a bijection, there exists a unique x ∈ A such that f ( x ) = B.
Is x ∈B? If yes, then it follows by the definition of x that x ∉ f ( x ) = B; this is obviously not
possible. On the other hand, if x ∉B, then the definition of B implies that x ∈B, again yielding
a contradiction.

K12146_Book.indb 10 2/6/15 5:12 PM

 1.2 Sets and Cardinal Numbers   ◾    11  

It is easy to verify by induction that for every n ∈ , if A = n then |P ( A)| = 2n . By analogy,

if A = m (infinite cardinals included), then |P ( A)| will be denoted by 2m . For example,
P ( ) = 2ℵ0 . Since P ( ) = (0,1), and since (0,1) =  (Exercise 4), it follows that  = 2ℵ0.
One way to verify that P ( ) = (0,1) is to express the real numbers in the interval (0,1) in
binary notation and associate to every subset A of  the real number in (0,1) whose n-th
decimal is 1 only if n ∈ A .
The following proposition is a short summary of the results on cardinals that we will
invoke later on.

Proposition 4.
(a) If J is countable and if each A j , j ∈ J , is countable, then so is ∪ A j .
j∈J
(b) If for every i ∈{1, 2, …, n} the set Xi is countable, then so is the set product
X1 × X 2 ×  × X n .

Proof. Exercise 5.
Exercises
1. Let X be an infinite set. Show that for every finite subset A of X, X \ A = X . Show that
there is subset B of X such that B =ℵ0 and such that X \ B = X .
The Hilbert Hotel, with countably many rooms, is a popular rendering of this exer-
cise in the case where X =ℵ0; see Illustration 1.5 for a special subcase.

Illustration 1.5

K12146_Book.indb 11 2/6/15 5:12 PM

 12   ◾    Sets, Numbers, and Cardinals

2. Show that if A and B are finite sets and if S is the set of all mappings A → B , then
A
S=B .

3. Let A = A1 , B = B1 , let S be the set of all mappings A → B , and let S1 be the set of
all mappings A1 → B1. Show that S1 = S .
4. Show that  = (0,1) .
5. Prove Proposition 4.

6. Is the set of all mappings  → {a, b} countable or uncountable? Justify!

7. Prove that if A = B , then P ( A) = P ( B) .

8. If f : X → Y , define f # : P (Y ) → P ( X ) by f # ( A) = f −1 ( A). Prove that f is one-to-one

if and only if f # is onto, and that f is onto if and only if f # is one-to-one.
9. Let A = n, let B = m (where n and m are any cardinal numbers). Define m + n to be
A1 ∪ B1 , where A1 and B2 are any two disjoint copies of A and B, respectively. Show
that this operation is well defined (i.e., show it does not depend on the choice of A1
and B1 ). Show that 2 + 3 = 5 .
10. Let A = n, let B = m (where n and m are any cardinal numbers). Define m ⋅ n to be
A× B .
(a) Show that 2 ⋅ 3 = 6.
(b) Show that ℵ0 ⋅ℵ0 =ℵ0 .
11. (a) Prove that if n ≥ m, then 2n ≥ 2m .
(b) Prove that if 2n ≥ℵ0 , then 2n ≥ 2ℵ0 .

1.3 Axiom of Choice and Equivalent Statements

A brief historical note: The Axiom of Choice was first explicitly formulated in 1904 by Ernst
Zermelo (1871–1953).

We accept the following axiom and, consequently, all of its consequences (a few of which
we will mention further below).

Axiom of Choice. For every family { Ai : i ∈ I } of pairwise disjoint sets there exists a map-
ping f : I → ∪ Ai such that f (i ) ∈ Ai .
i∈I

K12146_Book.indb 12 2/6/15 5:12 PM

 1.3 Axiom of Choice and Equivalent Statements   ◾    13  

Illustration 1.6 If each Ai is a (possibly infinite) set of eggs in a basket, and if each of the
(possibly infinitely many) baskets contains eggs, then we use f to choose exactly one egg f (i)
from each of the baskets.

Informally speaking, we assert in this tame-looking axiom that we can choose, by means of f,
exactly one element from each of the sets Ai (see Illustration 1.6). The Pandora’s box of unex-
pected, and often counter-intuitive, consequences is opened by the stipulation of the existence
of the “choice-function” f, without specifying exactly how that mapping is to be constructed.
We have already defined the set product of finitely many sets. More generally, the set
product of the sets Ai , i ∈ I , denoted ∏ Ai , is the set of all mappings I → ∪ Ai . Very infor-
i ∈I i∈I
mally, this is the set of all | I | -tuples of elements in ∪ Ai , one element per each set Ai .
i∈I
We will usually denote a generic element of ∏ Ai by (ai )i ∈I . When I is finite (with n
i ∈I
elements) or countable infinite we will sometimes denote the elements of the respective
set-products by (a1 , a2 ,… , an ) and (a1 , a2 ,… , an ,…).

The Axiom of Choice is equivalent to the following statement (Exercise 1).

Axiom for Products. Let { Ai : i ∈ I } be a collection of sets. If each Ai is nonempty, then so

is ∏ Ai .
i ∈I

Some other equivalent versions of the Axiom of Choice are not so simple to deduce. In
order to state them we need a few preliminaries.

K12146_Book.indb 13 2/6/15 5:12 PM

 14   ◾    Sets, Numbers, and Cardinals

The usual order of  was mentioned in Section 1.1; it is a linear order. The restriction of
that linear order to  , , and  yields the usual linear order of these sets.
Let A be an ordered set. An element c is a maximal element in A if for every x ∈ A, if
c ≤ x or x ≤ c , then x ≤ c . In other words, c is maximal if x ≤ c for every x ∈ A that is com-
parable via ≤ to c. Maximal elements need not be unique. For example, A = {a, b, c} with
an order < defined by a < b, a < c , has two maximal elements, b and c. A chain in A is any
subset B of A that inherits a linear order from the order of A. For example, with the partial
order on A = {a, b, c} as above, the set {a, b} is a chain in A. An element u ∈ A is an upper
bound of subset B of A if b ≤ u for every b ∈B . Considering again the set A = {a, b, c}
with the given order, we notice that every element of A is an upper bound for {a}, and that
{b, c} has no upper bound. A lower bound of B is defined analogously.

We now state two versions of Zorn’s Lemma (the first one without a proof). We will
invoke the second, stronger version several times later on.

Zorn’s Lemma 1. If every chain in an ordered nonempty set A has an upper bound in A, then
A contains a maximal element.

Zorn’s Lemma 2. If every chain in an ordered nonempty set A has an upper bound in A,
then for every a ∈ A there is a maximal element m ∈ A such that a ≤ m.
Proof. (Based on Zorn’s Lemma 1.)
Take any a ∈ A and consider the set Y = { y ∈ A : a ≤ y} . Clearly a ∈Y , so Y ≠ ∅ . By assump-
tion, every chain of elements in Y (being at the same time a chain of elements in A) has an
upper bound in A. It follows from the definition of Y that this upper bound is also in Y. By
Zorn’s Lemma 1 (applied to the set Y), Y contains a maximal element m. Again invoking
the definition of Y, it follows that the element m is also maximal in A, and, since a ≤ m , our
search for a maximal element in A comparable to a is over.

The following example illustrates a typical usage of Zorn’s Lemma.

Example 1
Let X be a set with X ≥ 2 and ordered with ≤. Fix an element x ∈ X , and con-
sider the collection P of all subsets A of X such that A is linearly ordered by ≤ and
such that x ∉ A. The relation ⊂ (“subset of”) is an order of the collection P . We show
that there exists Y ∈ P , such that Y is maximal with respect to ⊂.

Since X ≥ 2 , we have that P ≠ ∅ . Let Q be any chain in P (with respect to ⊂): this
means that for every A, B ∈Q , either A ⊂ B or B ⊂ A. Consider the set C = Z∪ ∈Q
Z . It
is obvious that x ∉C. It is very easy to show that C is a chain with respect to ≤. (Prove
it! That is, show that for every a, b ∈C, either a ≤ b or b ≤ a.) Finally, it is also clear that
Z ⊂ C for every Z ∈Q . This shows that C is an upper bound for Q . The conclusion we
wanted now follows from Zorn’s Lemma 1. ☐

Our acceptance of Zorn’s Lemma is justified by the following.

K12146_Book.indb 14 2/6/15 5:13 PM

 1.3 Axiom of Choice and Equivalent Statements   ◾    15  

Theorem 1. The Axiom of Choice and Zorn’s Lemma are equivalent statements (in a set
theory with standard basic axioms).

Suppose A is ordered by ≤, and let B be a subset of A. An element b ∈B is the least ele-

ment for B if for every x ∈B, b ≤ x. The definite article in the preceding sentence (i.e., the
uniqueness of the least element, if it exists) is easy to justify. The greatest element c ∈B is
such that x ≤ c for every x ∈B .
A set A is well ordered if it is linearly ordered (by ≤) and if every nonempty subset of
A has the least element. In that case we say that ≤ is a well ordering of A. For example,
the usual order of  makes it well ordered. However, the usual linear order of  does not
make it well ordered, since, for example,  itself does not have the least element. It is easy
to make  well ordered: for example, 0 ≤ −1 ≤ 1 ≤ −2 ≤ 2 ≤ −3 ≤ 3 ≤ … will do. The problem
of explicitly well ordering  turned out to be deeper and more difficult. The existence
part is established by Theorem 2 below, but only because we have accepted the Axiom of
Choice.

Well Ordering Principle. Every nonempty set has a well ordering.

Theorem 2. The Axiom of Choice and the Well Ordering Principle are equivalent state-
ments (in a set theory with standard basic axioms).

Consequently, in the set theory based on the Axiom of Choice and the other standard
axioms, the set  of all real numbers allows well ordering. (Observe that the usual linear
order of  is not a well ordering.)
The following example is important, since it is a part of a structure that we will explore
later on.

Example 2
Let < be a well ordering of  ; we know nothing about this ordering beyond its
bare existence (based on the Well Ordering Principle). Keep in mind that this is not
the usual linear order of  . Let ∗ be a well ordered copy of + disjoint from  ;
for example, we may take ∗ = {1 = (1,0), 2 = (2,0), 3 = (3,0),…} with 1 < 2 < 3 <….
Choose z ∉∗ ∪  and extend the order < (defined so far) on ∗ ∪  ∪ {z } as follows:
1 < 2 < 3 < … < x < z for every x ∈  . Then this is a well ordering on ∗ ∪  ∪ {z }. To
justify the last statement, take any nonempty subset A of ∗ ∪  ∪ {z }; if A ∩ ∗ ≠ ∅ ,
then the least element of A ∩ ∗ is also the least element of A; if A ∩ ∗ = ∅ , then
A ⊂  ∪{z }, and the least element of A in the well ordered set  ∪{z } is also the least
element of A in ∗ ∪  ∪ {z }.

It follows that the nonempty set {x ∈∗ ∪  ∪ {z } : { y : y < x } = ∞} must

have the least element; denote this element by ω 0 . Similarly, the set
{x ∈∗ ∪  ∪ {z } : { y : y < x } is uncountable} is also nonempty (since z is

K12146_Book.indb 15 2/6/15 5:13 PM

 16   ◾    Sets, Numbers, and Cardinals

there—this is the reason we introduced z), and so it also contains the least ele-
ment, which we denote by ω. Observe that ω 0 < ω , and that ω 0 ≤ x for every
x ∈ (so that ω 0 is the smallest element in  with respect to the well order
which we assumed exists).
The well ordered set {x ∈∗ ∪  ∪ {z } : x ≤ ω}, endowed with more structure as
we go, will be important to us. In anticipation of that development—the “Tychonoff
Plank” will come out of it—we denote TP = {x ∈∗ ∪  ∪ {z } : x ≤ ω}. In Illustration 1.7,
we depict the order structure of TP.
TP

1 2 3 4 ω0 ω z

N* R

Illustration 1.7 The well-ordered set TP: if x is to the left of y, then x < y.
☐

Exercises
1. Prove that the Axiom of Choice and the Axiom for Products are equivalent.
2. Let X be an ordered nonempty set such that every chain in X has a maximal element.
Prove that for every x ∈ X , there exists a maximal element y ∈ X such that x ≤ y.
3. Let Y ⊂ X , and let f : X → Y be onto. Use the Axiom of Choice to show that for every
mapping g : X → Y , there is a mapping h : X → X , such that g = f  h. Identify where
in your solution you have used the Axiom of Choice. (Remark: We express the state-
ment that g = f  h for some h in the commutative diagram shown in Illustration 1.8;
we will encounter more commutative diagrams later on.)

h
X X

g f Illustration 1.8 The double arrow indicates a

surjection; the dashed line is where we claim
Y
existence.

4. Prove that if X is linearly ordered by ≤ , and if every countable subset of X is well

ordered by ≤, then X is well ordered by ≤. (Hint: Start by assuming that X is not well
ordered, and construct a countable linearly ordered set without the least element.)
5. Prove that for every order R of any set A there exists a linear order Q of A such that
R ⊂ Q . [Hint: 1. For every chain R1 ⊂ R2 ⊂  of orders on the set A, notice that
∞
∪ R is a partial order containing all of Ri − s . 2. Show that if M is a maximal order,
i =1 i
then M must be linear. Use Zorn’s Lemma.]

K12146_Book.indb 16 2/6/15 5:13 PM

 Chapter 2

Metric Spaces: Definition,

Examples, and Basics

A brief historical note: The historical start of topology is fuzzy, since , functions  →  ,
complex numbers and other related traditional mathematical objects, had been studied for
centuries before the word topology was introduced.

The “topological” structure of the classical models is determined by the notion of distance
between their elements. Metric spaces, introduced in this chapter, are a rather straight-
forward generalization of the ordinary, Euclidean spaces. They set the stage for further
natural generalizations. We start our journey with them.

2.1 Metric Spaces: Definition and Examples

The following is an important example; we will encounter it many more times.

Example 1: The Two-Dimensional Euclidean Metric Space

Consider for a moment the plane  2: the (Euclidean) distance between the points A = (a1 , a2 )
A = (a1 , a2 ) and B = (b1 , b2 ), denoted d( A, B),
a1 A(a1, a2)
is d( A, B) = (a1 − b1 )2 + (a2 − b2 )2 . (See
Illustration 2.1.) The distance d can be
(a1 – b1)2 + (a2 – b2)2
viewed as a mapping from  2 into  a2 – b2
with the following properties: B(b1, b2) a2
b1 a1 – b1

(a) For all points A and B, d( A, B) ≥ 0

b2
and d( A, B) = 0 if and only if A = B.
(b) For all points A and B, d( A, B) = d( B, A)
d( A, B) = d( B, A).
(c) For all points A, B and C = (c1 , c2 ),
d( A, B) ≤ d( A, C ) + d(C , B). Illustration 2.1 Euclidean distance in  2 .

17

K12146_C002.indd 17 2/6/15 5:09 PM

 18   ◾    Metric Spaces: Definition, Examples, and Basics

Property (c) is sometimes called the triangle inequality. Properties (a) and (b) are
obvious. After replacing xi = ai − ci and yi = ci − bi , squaring and simplifying, prop-
erty (c) reduces to the following inequality, which in turn is a particular case of the
(more general) Cauchy inequality (which will be stated and proven in Example 3).

(♦)   x1 y1 + x 2 y 2 ≤ x12 + x 22 y12 + y 22 .

We justify (♦) in pictures (Illustrations 2.2 and 2.3); an explanation is provided in the
paragraph that follows. Since we will prove (♦) in a more general setting, the reader
may safely skip the rest of this example.

E
F

y2 x2
A
x1
y1
D
B
y2 y1
G
C

Illustration 2.2

A b z

B
a
D

Illustration 2.3

The inequality (♦) is equivalent to the claim that the sum of the areas of the two
smaller right-angled shaded triangles shown in Illustration 2.2 is less than the area of
the larger shaded triangle. This is the same as proving that (see Illustration 2.2) Area
(BEFG)  ≤  Area(ABCDE). Observe  t hat Area( BEFG ) = 12 ( x1 + y 2 )( x 2 + y1 ), and that

K12146_C002.indd 18 2/6/15 5:09 PM

 2.1 Metric Spaces: Definition and Examples   ◾    19  

the same formula holds for Area(ABCDE) when A, B, and C happen to be collinear
(when we get a trapezoid). We can achieve collinearity of A, B, and C by rotating
the triangle BCD around the point B (Illustration 2.3; only the side ED changes in
size). Such rotation will only affect the area of the triangle BDE. The Heron func-
tion f (z ) = s(s − a)(s − b)(s − z ) , s = 12 (a + b + z ), gives the area of the triangle BDE
(during the rotation). Using basic calculus we establish that this function achieves
its maximum precisely when z = a 2 + b 2 , that is, at the moment depicted in
Illustration 2.2. Consequently, Area(ABCDE) at this moment (Illustration 2.2) is at
least as much as Area(ABCDE) when A, B, and C are collinear, which in turn is the
same as Area(BEFG). ☐

The notion of metric is a straightforward b

generalization of the Euclidean distance d

of Example 1 through its three properties a
d(a, b) = 1983

listed there. Given a nonempty set X, we say d(b, c) = 1991

that a mapping d : X × X →  is a metric d(a, c) = 1988
if it satisfies the properties (a), (b), and (c) c
for all elements A, B, and C in X. The pair
(X, d) is called a metric space. If the met- Illustration 2.4 Extending d by setting d(x, x) =
ric d is understood from the context we 0 and d(x, y) = d(y, x) for all x,  y ∈ {a,  b,  c} gives a
metric space over {a, b, c}.
will often refer to X as being a metric space.
In Example 1 we have defined the Euclidean metric d, and the Euclidean metric space
( 2 , d ). A simple example of a metric space over the set {a, b, c} is shown in Illustration 2.4.

b
Example 2: The Discrete Metric Space
Let X be any nonempty set and define a mapping d(a, b) = 1
a
 0 if x = y
d(b, c) = 1
d : X × X →  by d( x , y ) =  .
1 if x ≠ y
 d(a, c) = 1
It is easy to check that (a), (b), and (c) are satis- c

fied. We call this metric the discrete metric on Illustration 2.5 We modify the metric
X, and (X, d) is the discrete metric space over shown in Illustration 2.4 to get the dis-
X (see Illustration 2.5). ☐ crete metric over {a, b, c}.

Example 3: The General Euclidean Metric Space

In this important example we will generalize the Euclidean metric space ( 2 , d ). The
underlying set is now n (n ≥ 1) and the Euclidean metric is defined as follows: for
A = (a1 , a2 ,…, an )and B = (b1 , b2 ,…, bn ), d( A, B) = (a1 − b1 )2 + (a2 − b2 )2 +  + (an − bn )2 .
Since the conditions (a) and (b) are obviously satisfied in this case too, we focus on pro­
ving (c). We need to show that for every (a1 , a2 ,…, an ), (b1 , b2 ,…, bn ), (c1 , c2 ,…, cn ) ∈n ,
n n n
Σ (ai − bi )2 ≤ Σ (ai − ci )2 + Σ (ci − bi )2 . In order to see more clearly ahead of us,
i =1 i =1 i =1

K12146_C002.indd 19 2/6/15 5:10 PM

 20   ◾    Metric Spaces: Definition, Examples, and Basics

we shorten the notation via xi = ai − ci and yi = ci − bi ; the last inequality now becomes
n n n
Σ ( xi + yi )2 ≤ Σ xi2 + Σ yi2 . Since the terms within the roots are positive, squaring
i =1 i =1 i =1
n n n n n
both sides yields the equivalent inequality Σ ( xi + yi )2 ≤ Σ xi2 + Σ yi2 + 2 Σ xi2 Σ yi2 ,
i =1 i =1 i =1 i =1 i =1
n n n
which, after some cancellation, reduces to Σ xi yi ≤ Σ xi2 Σ yi2 . Squaring again
i =1 i =1 i =1
yields another equivalent inequality, called the Cauchy inequality:

2
 n
  n
 n

(*)   

∑
i =1
x i yi  ≤ 
 
∑i =1
2
x 
i

∑ i =1
yi2  .


The following is one of many standard proofs of the Cauchy inequality. We con-
sider the function f :  →  defined by f (ω ) = Σ ( xi + ωyi )2 = Σ xi 2 + 2  Σ xi yi  ω
n n n

i =1 i =1  i =1 
+  Σ yi 2  ω 2 . It is a concave-up parabola, which, since Σ ( xi + ωyi )2 ≥ 0 , never reaches
n n

 i =1  i =1
2
below the x-axis. Consequently, the discriminant 2 Σ xi yi  − 4  Σ xi2   Σ yi2  is
 n n n

 i =1   i =1   i =1 
always ≤ 0, which, as can be readily seen, yields (*).
The metric space (n , d ) is called the Euclidean metric space. It is the prototype of
the theory of metric spaces and, further, of topology. This being the case, and unless we
state otherwise, we will assume from now on that n is equipped with the Euclidean
metric. ☐

Example 4: The City Street Metric or the Taxi Cab Metric

The set X is  2 again, but the met-
ric is now defined as follows: for any
two points A = (a1 , a2 ) and B = (b1 , b2 ), A
d( A, B) = a1 − b1 + a2 − b2 (see Illustration
2.6). Properties (a) and (b) are again obvious.
Property (c) reduces to proving the follow-
ing inequality: a1 − c1 + a2 − c2 ≤ a1 − b1 + B
a2 − b2 + b1 − c1 + b2 − c2 . This is a conse-
quence of a + b ≤ a + b . ☐

It is evident that the metrics in Examples 3

and 4 are different. That notwithstanding,
from the topological, or structural point of
Illustration 2.6 The city metric in  2 : the
view, the associated metric spaces are basi- shortest distance is not the usual distance.
cally the same. We will explain this more
precisely after we introduce some important notions.

K12146_C002.indd 20 2/6/15 5:10 PM

 2.1 Metric Spaces: Definition and Examples   ◾    21  

Let (X, d) be a metric space. A ball centered at a ∈ X and with radius r (r > 0),
denoted B(a, r), is the set {x ∈ X : d(a, x ) < r }. The associated closed ball is defined to be
{x ∈ X : d(a, x ) ≤ r } . A subset U of X is open if for every element x ∈ U there is a ball B(x, r)
contained in U. Closed subsets of X are the complements in X of open subsets of X.

Open Closed
Illustration 2.7 Open sets will often be shown
with dashed bounding lines, closed sets will
be with full lines.

The following property will later be used to generalize the notion of metric spaces to the
more general concept of topological spaces.

Proposition 1. The following are true for every metric space X:

(i) The sets ∅ and X are open.

(ii) Unions of open sets are open.
(iii) Intersections of finitely many open sets are open.

Proof.
The sets X and ∅ are plainly open, as are unions of U1 U2
open sets. We confirm (iii): assume U1 ,U 2 ,…,U n
are open sets and consider their intersection
n
V = ∩U i . Choose a point x ∈ V. Then, x ∈U i
i =1
for i =1, 2, … , n . Since the sets U1 ,U 2 ,…,U n are
open, there are balls B( x , r1 ), B( x , r2 ),…, B( x , rn )
contained in these sets, respectively. Choose Illustration 2.8 The smaller of the two
disks corresponding to U1 and U2 tells
r = min{r1 , r2 ,…, rn } . Then B( x , r ) ⊂ V and so V is us that the intersection U1 ∩ U 2 is open.
open too (Illustration 2.8). ◼

The interconnectedness and the mutual relationship of the open sets in a metric space,
and, later, in a topological space, are what constitutes the inner (topological) structure of
the space. Generally speaking, our primary object of study will be this deeper anatomy of
the spaces, as opposed to the exterior, geometrical considerations, which in this theory will
play a marginal role.
Two metrics d1 and d2 on a set X are said to be equivalent if they generate the same
classes of open sets. The two metric spaces defined in Examples 3 and 4 are equivalent
(Exercise 2), but neither is equivalent to the discrete metric space. On the other hand, every
metric space over a finite set X is equivalent to the discrete metric space (Exercise 3). As a

K12146_C002.indd 21 2/6/15 5:10 PM

 22   ◾    Metric Spaces: Definition, Examples, and Basics

consequence, finite metric spaces are not interesting from the structural point of view; we
have seen them in Illustrations 2.4 and 2.5 and we will not mention them again.

Proposition 2. Let X be equipped with two metrics, d1 and d2 . Assume that for every ball
B(a,r) in any one of the metric spaces ( X , d1 ), ( X , d2 ) , there is a ball B(a, r′ ) in the other
metric space such that B(a, r ′ ) ⊂ B(a, r ). Then, ( X , d1 ) and ( X , d2 ) are equivalent.

Proof. Straightforward.

In the following example we describe a metric space over  2 that is not equivalent to the
Euclidean metric space.

Example 5: The Post Office Metric

We define a metric d p over  2 as follows: A
we fix a point p in  2 and set d p ( x , y ) =
d( x , p) + d( y , p) when x ≠ y, and d p ( x , x ) = 0,
where d is the Euclidean metric (Illustration 2.9).
It is easy to check that this is indeed a metric.
Here, every set A that does not contain the B
point p is open, while only some of the sets
Illustration 2.9 The post office metric:
containing p are open (Exercise 5). It is evident
The shortest path leads through a fixed
that this space is not equivalent to the Euclidean point (the post office, represented by a
metric space over  2 . ☐ mailbox).

Exercises
1. Show that for every metric space X and for every x ∈ X the singleton {x } is closed.
2. Prove that the Euclidean metric space and the city metric over  2 are equivalent.
3. Show that every finite metric space is equivalent to the discrete space over the same set.
4. Show that for every two distinct points x and y in a metric space X, there exist disjoint
open sets U and V such that x ∈ U and y ∈ V.
5. Consider  2 with the topology induced by the post office metric with respect to a
fixed point p.
(a) Show that if p ∉ A ⊂  2 then A is open.
(b) Show that if p ∈ A ⊂  2 then A is open if and only if there is a Euclidean ball
B(p, r) (that is, a ball with respect to the Euclidean metric) that is contained in A.
6. Suppose d1 and d2 are two metrics over a set X. Show that d3 defined by
d3 ( x , y ) = min{d1 ( x , y ), d2 ( x , y )} is also a metric over X.

K12146_C002.indd 22 2/6/15 5:10 PM

 2.2 Metric Spaces: Basics   ◾    23  

7. Is the intersection of countably many open sets necessarily open? Give a proof or a
counterexample. What happens when you try to generalize the proof of Proposition
1 (iii)?
8. Let F be a closed subset of a metric space (X, d). For every point x ∈ X, define
d( x , F ) = inf{d( x , y ) : y ∈ F }. Show that x ∈ X \ F if and only if d( x , F ) > 0 .
9. Let A and B be two subsets of a metric space (X, d). Define d( A, B) =
inf{d ( a, b ) : a ∈ A, b ∈B}. Find a metric space X and two disjoint closed subsets A and
B of X such that d( A, B) = 0.
10. Show that if F and G are two disjoint closed sets in a metric space X, then there are
two disjoint open sets U and V such that F ⊂ U and G ⊂ U . [Hint: Exercise 8.]
11. Consider a set R of all open rectangles in  2 above the x-axis, bounded by edges of
slopes 1 and −1, with one corner at the x-axis, and such that every point (x, 0) on the
x-axis is a corner of exactly one member Rx of R (see Illustration 2.10). Show that
there is a rational number q and an irrational number z such that Rq ∩ Rz ≠ ∅. (We
will refer to this exercise in Section 8.2.)

Ra
1

Ra
2

Illustration 2.10 We show a few rectangles

a2 a1
from the set R.

2.2 Metric Spaces: Basics

A brief historical note: The French mathematician Augustin-Louis Cauchy (1789–1857) is con-
sidered to have been one of the founders of Analysis.

Let (X, d) be a metric space, and let A be a nonempty subset of X. It is very easy to verify
that the restriction d A × A of d to A × A is a metric on A. We say that ( A, d A × A ) is a metric
subspace of (X, d). We will prefer the less accurate but simpler (A, d) instead of ( A, d A × A ).
When d is understood, then we say that A is a (metric) subspace of X. The following propo-
sition establishes the principal link between the open sets in X and the open sets in A.

K12146_C002.indd 23 2/6/15 5:10 PM

 24   ◾    Metric Spaces: Definition, Examples, and Basics

Proposition 1. Let (A, d) be a metric subspace of (X, d). For every V ⊂ A,V is open in A if
and only if there is an open subset U of X such that V = A ∩U .
We will eventually prove Proposition 1 in a more general setting of topological spaces; at
this stage we leave it as an exercise (Exercise 3).

The relationship between the closed subsets of A and X is similar: F is closed in A if and
only if there is a closed subset G of X such that F = A ∩ G (Exercise 4).
A sequence ( xn ) of elements in a metric space X converges to x, denoted lim xn = x , if
n→∞
for every ε > 0, there is an integer N such that for every n > N , xn ∈ B( x , ε ).

xN
x
xN+2
xN+1
x2

x3
x1

Illustration 2.11

The following theorem will be needed.

Theorem 2 (Cantor’s Nested Intervals Theorem). Let {[an , bn ] : n = 1,2,…} be a set of closed
∞
intervals in  such that for every n ∞
∈{1,2,…}, [an+1 , bn+1 ] ⊂ [an , bn ]. Then, ∩[an , bn ] ≠ ∅. In
n=1
addition, if lim(bn − an ) = 0 , then ∩[an , bn ] contains exactly one element.
n→∞ n=1

Proof. Our assumptions imply that for every n ∈ {1, 2, …}, an ≤ b1 and so the set
{an : n = 1,2,…} is bounded from above. By the Least Upper Bound Property, there is the
least upper bound; denote it by a. Notice that a ≤ bn for every n ∈+ , for if a > bm , then
bm would be an upper bound for {an : n = 1,2,…} that is smaller than a. Similarly, the set
{bn : n = 1,2,…} has the largest lower bound b, for which we have an ≤ b for every n ∈+.
This set∞of inequalities, together with a ≤ bn for every n ∈+ , implies that a ≤ b , and that
[a, b] = ∩ [an , bn ]. This completes the proof of the first part of the theorem.
n=1
Since an +1 ≥ an and bn +1 ≤ bn for all n =1, 2, … , it follows easily that lim an = a and
n→∞
lim bn = b. If we further assume that lim(bn − an ) = 0, then 0 = lim(bn − an ) = lim bn − lim an =
n→∞ n→∞ n→∞ n→∞ n→∞
∞
b − a, so that a = b and hence [a, b] = ∩ [an , bn ] is a singleton.
n =1

A sequence ( xn ) in a metric space (X, d) is a Cauchy sequence if for every ε > 0, there is an
integer N such that if n, m ≥ N , then d( xm , xn ) < ε.
It is obvious that every convergent sequence is a Cauchy sequence. The converse is not
always true. For example ( n1 ) is a Cauchy sequence in the interval (0, 1) (considered as a

K12146_C002.indd 24 2/6/15 5:10 PM

 2.2 Metric Spaces: Basics   ◾    25  

metric subspace of ), but it does not converge there. Metric spaces in which every Cauchy
sequence converges are called complete.

Example 1:  Is Complete
We show in this example that  is a complete metric space. There will be three small
gaps in the argument: one will be called “Exercise,” and we will alert the reader of
the other two by following them with “Why?” These three exercises deserve some
thought.
∞
Let ( xn ) be a Cauchy sequence in . Then so is ( xn )n = k , k ≥1. Every Cauchy
sequence is (obviously) bounded. Denote the least upper bound of ( xn )n∞= k by bk .
Then {bk } is a nonincreasing sequence (i.e., bn +1 ≤ bn for all n), and it is bounded from
below. Hence, it converges (Exercise 2); denote lim bk = b. Symmetrically, if the great-
k→∞
est lower bound of ( xn )n∞= k is denoted by ck , then (ck ) is nondecreasing and bounded
from above, and so it converges; denote lim ck = c. Observe that ci ≤ b j all the time,
k→∞
so that c ≤ b.
Suppose c < b. The interval (c, b) must have members of ( xn ) , or else the sequence
fails to be Cauchy. (Why?) On the other hand, it follows from our construction of c
and b that x k ≥ b or x k ≤ c for every k. (Why?) This clearly implies that x k ∉(c , b) for
every k. Our assumption c < b led us to a contradiction. Hence, b = c. It is now easy to
confirm that ( xn ) converges to b = c. ☐

Let ( X , d1 ) and (Y , d2 ) be metric spaces. We say that a mapping f : X → Y is conti­nuous

at a point x ∈ X if for every B( f ( x ), ε) in Y, there is B( x , δ) in X such that f ( B( x , δ )) ⊂
B( f ( x ), ε). The reader should notice that in the case where X = Y =  and d1 = d2 =
the Euclidean metric, the above definition of continuity reduces to the standard “ε −δ
definition.”

Proposition 3. Let ( X , d1 ) and (Y , d2 ) be metric spaces. A mapping f : X → Y is continu-

ous at x ∈ X if and only if for every sequence ( xn ) converging to a in X, the sequence
( f ( xn )) converges to f (a) in Y.

X
f (x1)
xn x2 x1 f Y f (x2)
a
f (a)

f (xn)

Illustration 2.12 X is a metric space; f is continuous if and only if images of sequences con-
verging to a are sequences converging to f (a).

K12146_C002.indd 25 2/6/15 5:10 PM

 26   ◾    Metric Spaces: Definition, Examples, and Basics

Proof. ⇒ Suppose ( xn ) converges to a in X, and choose any B( f ( x ), ε), ε > 0. Then, by

assumption, there is B(a, δ) such that f ( B(a, δ)) ⊂ B( f (a), ε). Since lim xn = a there
n→∞
exists N such that for every n > N , xn ∈ B(a, δ ). It is then evident that for every n > N ,
f ( xn ) ∈ f ( B(a, δ )) ⊂ B( f (a), ε ), showing that lim f ( xn ) = f (a).
n→∞
⇐ Suppose f : X → Y is not continuous at a. Then there is B( f (a), ε) such that for every
B(a, δ), we have f ( B(a, δ )) ⊂/ B( f (a), ε). Choose xn in B ( a, n1 ) such that f ( xn ) ∉ B( f (a), ε),
n =1, 2, … . Then lim xn = a , and lim f ( xn ) ≠ f (a).
n→∞ n→∞

Remark: In the proof of Proposition 3 we have used the symbols ⇒ and ⇐. We will do
this consistently throughout this book. If a statement is of type “P if and only if Q,” then ⇒
indicates that we are focusing on “P implies Q,” while ⇐ means that we are dealing with
“Q implies P.”

A mapping f : X → Y between two metric spaces is continuous if it is continuous at every

point of X. If, in addition, f is a bijection with a continuous inverse f −1 : Y → X , then we
say that f is a homeomorphism, and that the metric spaces X and Y are homeomorphic. If
X and Y are homeomorphic then we write X ≅ Y .

The following proposition will metamorphose into a definition in Chapter 3.

Proposition 4. Let ( X , d1 ) and (Y , d2 ) be metric spaces. A mapping f : X → Y is continu-

ous if and only if for every open subset U of Y, f −1 (U ) is open in X.

Proof. ⇒ Let f be continuous, and let U be an open subset of Y. Showing that f −1 (U ) is

open in X is equivalent to showing that every x ∈ f −1 (U ) is the center of a ball that is con-
tained in f −1 (U ) . Consider f ( x ) ∈ U . Since U is open, there is a ball B( f ( x ), ε) such that
B( f ( x ), ε) ⊂ U . Since f is continuous, it is continuous at x, which in turn implies that there
is a ball B( x , δ) such that f ( B( x , δ)) ⊂ B( f ( x ), ε). It is then evident that B( x , δ ) ⊂ f −1 (U ) .
⇐ Assuming f −1 (U ) is open in X for every open subset U of Y, we need to show that f
is continuous at every point x ∈ X . Start with a ball B( f ( x ), ε). Then, since B( f ( x ), ε)
is open in Y, f −1 ( B( f ( x ), ε )) is open in X. Choose a ball B centered at x and such that
B ⊂ f −1 ( B( f ( x ), ε )). Then, f ( B) ⊂ f ( f −1 ( B( f ( x ), ε))) = B( f ( x ), ε).

Example 2
Let A be a subset of a set X. The inclusion in : A → X is defined by in(a) = a for every
a ∈ A. If X is a metric space, then A becomes a metric subspace of X. In this case the
inclusion in is necessarily continuous (Exercise 6). If f : X → Y is a continuous mapping
between two metric spaces, then the restriction f A : A → Y is also continuous. ☐

n
Let ( Xi , di ), i = 1, 2, …, n, be metric spaces and consider the set product X = ∏ Xi . We
i =1
notice two obvious types of mappings: the projections pi : X → Xi , i ∈{1, 2, …, n}, defined

K12146_C002.indd 26 2/6/15 5:10 PM

 2.2 Metric Spaces: Basics   ◾    27  

by pi ( x1 , x 2 ,…, xn ) = xi , and the coordinate mappings cia : Xi → X (i ∈ {1, 2, … , n}) defined

by cia ( xi ) = (a1 , a2 ,…, ai−1 , xi , ai+1 ,…, an ), where a = (a1 , a2 ,…, an ) is a fixed element of X.
n
We want a metric over X = ∏ Xi such that all projections pi and all coordinate mappings
i =1
cia , for all choices of a, are continuous. One natural choice is the product metric d p , defined
as follows: for every ( x1 , x 2 ,…, xn ), ( y1 , y 2 ,…, yn ) ∈ X , d p (( x1 , x 2 ,…, xn ), ( y1 , y 2 ,…, yn )) =
(d (x , y ) )
1
2
1 1 1 + d2 ( x 2 , y 2 )2 +  + dn ( xn , yn )2 2 .

The product metric is a generalization of the Euclidean metric for n. Confirming that d p
is indeed a metric follows along the lines of the argument given in Example 3 of 2.1. As was
the case there, it is the triangle inequality that needs primary attention:

n n n

∑d (x , z )
i =1
i i i
2
≤ ∑d (x , y ) + ∑d ( y , z ) .
i =1
i i i
2

i =1
i i i
2

This is sometimes called the Cauchy–Schwartz inequality. Simple manipulation reduces

n n n
it to proving the following: Σ di ( xi , yi )di ( yi , zi ) ≤ Σ di ( xi , yi )2 Σ di ( yi , zi )2 , which we
i =1 i =1 i =1
recognize as the more general version of the Cauchy inequality. The proof of the Cauchy
inequality provided in 2.1 needs only a slight modification to justify the inequality shown
here. We leave the details to the reader (Exercise 11).
The space ( X , d p ) is called the product metric space. That it works as advertised is left
as an exercise (Exercise 9).
n
If ( Xi , di ) = (Y , d∗ ) for all i = 1, 2, …, n, then we will abbreviate ∏ Xi to Y n .
i =1
The product metric d p is certainly not the only metric over X making the projections
and the coordinate mappings continuous. For example, dm (( x1 , x 2 ,…, xn ), ( y1 , y 2 ,…, yn )) =
max{d1 ( x1 , y1 ), d2 ( x 2 , y 2 ),…, dn ( xn , yn )} is another such metric over X. In fact, these two
metrics over X are equivalent (Exercise 12). So, from the point of view of the structure
of the set of open subsets of X, it really does not matter which of the two we choose to
work with.

Example 3: Continuity of the Metric

As usual, let (X, d) denote a metric space. Consider the product metric space ( X 2 , d p )
and the metric d : X 2 →  . We will now show that d is continuous.
Suppose (( xi , yi )) is a sequence converging to a point (a, b) ∈ X 2 . This implies that
lim xi = a and that lim yi = b (Exercise 14). So, for every ε > 0, there is an integer N,
i →∞ i →∞
such that for every i > N , d( xi , a) < 2ε and d( yi , b) < 2ε . Hence, by the triangle inequal-
ity, d( xi , yi ) ≤ d( xi , a) + d(a, yi ) ≤ d( xi , a) + d(a, b) + d(b, yi ) ≤ d(a, b) + ε, for all i > N .
Similarly, d(a, b) ≤ d(a, xi ) + d( xi , yi ) + d( yi , b) ≤ d( xi , yi ) + ε, for all i > N . It follows
that, for all i > N , d(a, b) − ε ≤ d( xi , yi ) ≤ d(a, b) + ε. All this tells us that the sequence
(d( xi , yi )) converges to the number d(a,b).

K12146_C002.indd 27 2/6/15 5:10 PM

 28   ◾    Metric Spaces: Definition, Examples, and Basics

It follows from Proposition 3 that d, considered as a mapping X 2 → , where X 2 is

equi­pped with the product metric d p , is always continuous. ☐

We now strengthen the definition of continuity. Let ( X , d1 ) and (Y , d2 ) be metric spaces. A

mapping f : X → Y is uniformly continuous if for every ε > 0, there is a δ > 0 such that for
every x1 , x 2 ∈ X , if d1 ( x1 , x 2 ) < δ, then d2 ( f ( x1 ), f ( x 2 )) < ε.
It is obvious (by temporarily fixing one of x1 , x 2) that uniform continuity implies con-
tinuity at every given point; thus uniform continuity implies continuity. A standard coun-
terexample showing that the converse fails is the function f ( x ) = x1 from the interval (0,1)
(considered as a metric subspace of ) into . The conditions under which continuity
implies uniform continuity will be studied later.
Continuous mappings do not always carry a Cauchy sequence into a Cauchy sequence
(Exercise 7). However, uniformly continuous mappings do!

Proposition 5. Let ( X1 , d1 ) and ( X 2 , d2 ) be two metric spaces, let ( xn ) be a Cauchy sequence

in X1, and let f : X1 → X 2 be a uniformly continuous mapping. Then ( f ( xn )) is a Cauchy
sequence in X 2 .

Proof. Take any ε > 0. Since f is uniformly continuous, there is a δ > 0 such that if
d1 ( y1 , z1 ) < δ, then d2 ( f ( y1 ), f (z1 )) < ε. Since ( xn ) is a Cauchy sequence, there exists an
integer N such that if n, m > N , then d1 ( xn , xm ) < δ. Hence, for the same n and m, we have
d2 ( f ( xn ), f ( xm )) < ε, establishing that ( f ( xn )) is a Cauchy sequence.

Given two metric spaces ( X1 , d1 ) and ( X 2 , d2 ) a mapping f : X1 → X 2 is a dilation if there is

a positive number α such that for every x1 , x 2 ∈ X1 we have d2 ( f ( x1 ), f ( x 2 )) = αd1 ( x1 , x 2 ).
A dilation for which α =1 is called an isometry. Every dilation, hence every isometry, is
one-to-one and uniformly continuous. We leave the justification as an exercise (Exercise 19).
A dilation with 0 < α < 1 is called a contraction by a fixed factor α. If there is α ∈(0, 1)
such that d2 ( f ( x1 ), f ( x 2 )) ≤ αd1 ( x1 , x 2 ) for every x1 , x 2 ∈ X1 , then we say that f is a contrac-
tion. It is very easy to see that contractions must be continuous.
A mapping f : X → X has a fixed point if there is an element x ∈ X such that f ( x ) = x .

Theorem 6 (Banach fixed point theorem). Let X be a complete metric space. Every con-
traction X → X has a unique fixed point.

Proof. Let f : X → X be a contraction. So, there is α ∈ (0, 1) such that d( f (u), f (v )) ≤

αd(u, v ) for every u, v ∈ X .
The sequence x , f ( x ), f 2 ( x ), … is Cauchy, since

d( f m ( x ), f m+k ( x )) ≤ α md( x , f k ( x )) ≤ α m (d( x , f ( x )) + d( f ( x ), f 2 ( x )) +  + d( f k−1 ( x ), f k ( x ))) ≤

≤ α m (d( x , f ( x )) + αd( x , f ( x )) + α 2d( x , f ( x )) +  + α k−1d( x , f ( x ))) =
1 − αk 1
= α md( x , f ( x ))(1 + α +  + α k−1 ) = α md( x , f ( x )) ≤ α md( x , f ( x )) ,
1− α 1− α

K12146_C002.indd 28 2/6/15 5:10 PM

 2.2 Metric Spaces: Basics   ◾    29  

so that d( f m ( x ), f m+ k ( x )) can be made arbitrarily small > 0 by choosing m sufficiently

large, for every k ≥ 0. Since we have assumed that the space X is complete, it fol-
lows that x , f ( x ), f 2 ( x ), … converges to some z ∈ X . Since applying f to the sequence
x , f ( x ), f 2 ( x ), … produces the sequence f ( x ), f 2 ( x ), f 2 ( x ),…, converging to the same
point, and since f is continuous, it follows that f (z ) = z .
The uniqueness of the fixed point is immediate, for if there is an element z1 ∈ X such
that f (z1 ) = z1, then d(z , z1 ) = d( f (z ), f (z1 )) ≤ αd(z , z1 ), which, since 0 < α < 1, is true only
if d(z , z1 ) = 0 . Hence. z = z1 .

Example 4: Fractals
We briefly and tangentially mention fractals; we will not see them further in this
book. A metric space (X, d) is a fractal if there is a contraction f : X → X by a fixed
factor (called the stretching factor), such that f ( X ) X .
Intervals in  (considered as metric subspaces of ), except the degenerate one-
point intervals and except  = (−∞, ∞) itself, are simple and not very interesting
fractals. The only not entirely obvious part of this claim is that  is not a fractal
(Exercise 22). Notice that the interval (0,1) is a fractal and  is not, even though they
are homeomorphic. Hence, being a fractal is a geometric property per se, not recog-
nized by the structure of the open subsets of the metric space.
A couple of nontrivial fractal metric subspaces—one of  2 , the other of  3—are
shown in Illustrations 2.13 and 2.14, respectively; we justify their inclusion on the
merits of their visual beauty only.

Illustration 2.13 Pentaflake. Illustration 2.14 Menger sponge.

☐
We will encounter more properties of metric spaces as we go.

K12146_C002.indd 29 2/6/15 5:11 PM

 30   ◾    Metric Spaces: Definition, Examples, and Basics

Exercises
1. Show that if lim xn = a and if lim xn = b in a metric space, then a = b.
n→∞ n→∞

2. Let (an ) be a nondecreasing sequence in  that is bounded from above. Show that
(an ) converges.
3. Prove Proposition 1.
4. Let (A, d) be a metric subspace of (X, d). Prove that F is closed in A if and only if there
is a closed subset G of X such that F = A ∩ G .
5. Show that a subset A of a metric space X is closed if and only if every sequence (an ) of
elements in A that converges in X also converges in A.
6. Let A be a subset of a metric space X.

(a) Consider A as a metric subspace of X. Show that the inclusion in : A → X is

continuous.
(b) Let d1 be a metric over A, and let d be the metric over X. Is the following true: if
the inclusion in : A → X is continuous, then A is a metric subspace of X?
7. Show that a continuous image of a Cauchy sequence need not be a Cauchy sequence.
8. Show that if ( xn ) is a Cauchy sequence that has a convergent subsequence, then ( xn )
is also convergent.
n
9. Let ( Xi , di ), i =1, 2, … , n, be metric spaces, and denote their product metric space ∏ Xi
i =1
by X. Show that every projection pi : X → Xi and every coordinate mapping cia : Xi → X
is continuous.
n
10. Let ( Xi , di ) , i =1, 2, … , n, be metric spaces, and consider the set product X = ∏ Xi .
i =1
Define d # (( x1 , x 2 ,…, xn ), ( y1 , y 2 ,…, yn )) = d1 ( x1 , y1 ) + d2 ( x 2 , y 2 ) +  + dn ( xn , yn ) , for
every ( x1 , x 2 ,…, xn ), ( y1 , y 2 ,…, yn ) ∈ X . Show that d # is a metric on the set X, and
show that it is equivalent to the product metric over X.
11. Let ( Xi , di ), i =1, 2, … , n, be metric spaces. Prove that
n n n

∑ d (x , y )d ( y , z ) ≤ ∑ d (x , y ) ∑ d ( y , z ) .
i i i i i i i i i
2
i i i
2

i =1 i =1 i =1

12. Show that the product metric d p and the metric dm as defined in this section (follow-
ing Example 2) are equivalent.
13. Show that n × m ≅ n+m.
14. Let ( X1 , d1 ) and ( X 2 , d2 ) be metric spaces and let (( xi , yi )) be a sequence in the product
metric space X1 × X 2 . Show that lim( xi , yi ) = (a, b) in X1 × X 2 if and only if lim xi = a
i →∞ i →∞
in X1 and lim yi = b in X 2 .
i →∞

K12146_C002.indd 30 2/6/15 5:11 PM

 2.2 Metric Spaces: Basics   ◾    31  

15. (a) Find a set {(an , bn ) : n = 1,2,…} of open intervals in  such that for every
∞
n ∈{1,2,…}, (an+1 , bn+1 ) ⊂ (an , bn ), and such that ∩ (an , bn ) = ∅ .
n=1

(b) Let {(an , bn ) : n = 1,2,…} be a set of open ∞intervals in  such that for every
n ∈{1,2,…}, an < an+1 < bn+1 < bn . Show that ∩ (an , bn ) ≠ ∅ .
n=1

16. Let f :[0, 1] → [0, 1] be a continuous function. Show that the function g :[0, 1] → [0, 1]
defined by g (t ) = sup{ f ( x ): x ∈[0, t ]} is continuous and nondecreasing. Draw a sketch
comparing the graphs of f and g if f ( x ) = − sin( 2πx ).
17. Let S be a set, and let (X, d) be a metric space. We say that a sequence ( f j ) of func-
tions f j : S → X , j = 1, 2, 3, …, is a Cauchy sequence of functions if for every ε > 0
there exists an integer N such that for every n , m > N , and for every x ∈S , we have
d( fn ( x ), fm ( x )) < ε. The same sequence of functions converges uniformly to a func-
tion f : S → X if for every ε > 0 there exists an integer N such that for every n > N
and for every x ∈S , we have d ( fn ( x ), f ( x )) < ε. Show that if X is complete, then for
every Cauchy sequence ( f j ) of functions f j : S → X there exists a function f : S → X
such that ( f j ) converges uniformly to f.
18. Let ( X , d1 ) and ( X , d2 ) be two metric spaces, let X = A ∪ B and let A ∩ B = C ≠ ∅. Assume
also that d1 (c1 , c2 ) = d2 (c1 , c2 ) for every c1 , c2 ∈C . Define d : X × X →  as follows:

 d1 ( x , y ) if x , y ∈ A

d( x , y ) =  d2 ( x , y ) if x , y ∈ B
 inf{d ( x , z ) + d ( y , z ) : z ∈C} otherwise.
 1 2

Show that d is a metric.

19. Show that every dilation is one-to-one and uniformly continuous.
20. Find an example of homeomorphic metric spaces which have no dilations from one
onto the other.
21. Let ( X1 , d1 ) and ( X 2 , d2 ) be metric spaces. Show that if f : X1 → X 2 is a dilation, then
f −1 : f ( X1 ) → X1 is continuous.
22. Show that  is not a fractal.
23. Let f :  2 →  2 be a dilation. Prove the following:
(a) f maps lines to lines.
(b) f sends angles (pairs of rays emanating from a single point) to angles of equal size.
(c) f maps circles to circles.
(d)  2 is not a fractal.
[Note: Justifying that n is not a fractal for all, n ≥ 3 can be done in steps similar to
the ones indicated in this exercise.]

K12146_C002.indd 31 2/6/15 5:11 PM

This page intentionally left blank
 Chapter 3

Topological Spaces:
Definition and Examples

O ur short theory of metric spaces will now be used as a springboard to launch us into
the study of the theory of topological spaces.

3.1 The Definition and Some Simple Examples

A brief historical note: Historically, topology evolved from analysis. Indeed, in its early stages
it was called Analysis Situs (positional analysis; “Analysis Situs” is the title of the seminal
paper by Henri Poincaré, published 1895). The term topology was coined in 1914 by Felix
Hausdorff (1868–1942), abstract spaces with topological structures were introduced in 1906
by Maurice Fréchet (1878–1973), and the modern, axiomatic setup of topology—the one we
will deal with in this book—was introduced in 1922 by Kazimierz Kuratowski (1896–1980).

Proposition 1, Section 2.1 does not explicitly involve the notion of metric, hence we can
abstract the following generalization from it.
A topological space is a set X and a set τ of subsets of X satisfying the following axioms:

(i) ∅ and X are in τ.

(ii) If U i , i ∈ I are all in τ, then so is ∪U i .
i∈I
n
(iii) If U1 , U 2 ,… ,U n are all in τ, then so is ∩U i .
i =1

It follows from (i) that τ is never empty. In the case when the set X is empty τ must be
{∅}—the most boring topological space one can imagine. We do not want to deal with it
and so:

We will assume now and forever that the set X in this definition is not empty.
33

K12146_C003.indd 33 2/12/15 3:40 PM

 34   ◾    Topological Spaces: Definition and Examples

The collection τ is called a topology over X. Formally, a topological space is the pair (X, τ),
where the topology τ satisfies the above axioms. We will sacrifice formality for brevity
and, when the topology τ is understood from the context, we will often refer to X itself as
a topological space.
The sets contained in τ are called open sets. Thus, according to axiom (i), the sets ∅ and
X are always open; axiom (ii) stipulates that unions of open sets must be open, while in
axiom (iii) we request that finite intersections of open sets be open.

Example 1: Metric Spaces

Before we see more exotic spaces, we back up for a moment to our starting models:
the metric spaces. Proposition 1 in 2.1, as announced, makes a topological space out
of every metric space. Importantly, the Euclidean metric space n now becomes the
Euclidean topological space (or the usual topological space). ☐

We reiterate the following convention (stated for the first time in 2.1):

Unless otherwise stated, n stands for the Euclidean topological space.

We will remind the reader of this a few more times. In special cases,  is the Euclidean
line,  2 is the Euclidean plane. At times we will use the all-encompassing term space even
when n = 1 or n = 2.

Example 2: The Indiscrete Topology

If ∅ and X happen to be the only members of τ, then we get one of the two
extreme topologies over a set X: the indiscrete topology. Since in this case the sets
U i mentioned in the axioms can only be ∅ or X, all of the axioms are satisfied.
The indiscrete topological space is the simplest, and in a way, the least interesting
of all spaces. ☐

Example 3: The Discrete Topology

Suppose now that τ is the set P (X) of all subsets of a set X. It is again plain that
the three axioms are satisfied so that τ is a topology over X. This is the other most
extreme example of a topology, called the discrete topology over X. So, not surpris-
ingly, this space possesses some extreme properties. ☐

Example 4: A Finite Topological Space

Take X = {a , 1, $} and τ = {∅ , {a , 1, $}, {a}, {a , 1}}. Since ∅ and X are given to be in

τ, the first axiom is satisfied. The union of any members of τ is the largest of them
according to the linear order ∅ ⊂ {a} ⊂ {a , 1} ⊂ {a , 1, $}, and so it is also in τ. This
shows that the second axiom is fulfilled. Similarly, any (finite) intersection of members

K12146_C003.indd 34 2/12/15 3:40 PM

 3.1 The Definition and Some Simple Examples   ◾    35  

of τ is the smallest of them with respect to ∅ ⊂ {a} ⊂ {a , 1} ⊂ {a , 1, $}, and so it is in τ.

Consequently the last axiom holds too and thereby τ is a topology. ☐

Finite topological spaces will rarely be encountered in this book.

Example 5: Finite Complement Topology or Co-finite Topology

With X being any (nonempty) set, define open sets (elements of τ) to be complements
in X of finite sets, as well as the empty set. We get the finite complement topology, or,
more succinctly, the co-finite topology (Illustration 3.1).

Illustration 3.1 An open set in the finite complement topology over a plane: they are obtained
by removing finitely many points in the plane (indicated by white disks in the picture).

This time we will prove that τ is indeed a topology.

(i) The set ∅ is open since we explicitly included it in τ, while X is open since it is
the complement of the finite set ∅.
(ii) Suppose {U i : i ∈ I } is a nonempty family of open sets. We want to show that
∪U i is also open. Since U i , i ∈ I , are open, each U i is a complement of some
i∈I

( ∩V ) . Since ∩V
assumption de Morgan law c
∪ (Vi )
c
finite Vi . Then we have ∪U i = = i∈I
i i is a
i∈I i∈I i∈I
subset of each Vi it must be finite too, and so ∪U i is a complement of a finite
i∈I
set. Thereby, it is open. n
(iii) Suppose that U1 , U 2 ,… ,U n are open. We show that ∩U i is open. By
i∈I
assumption, for every i = 1, 2, …, n , U i = (Vi )c for some finite set Vi . So

(∪V ) . Since all V , V ,…,V are finite, so is their

n assumption n de Morgan law n c
∩U i = ∩(Vi )c = i 1 2 n
i =1 i =1 i =1
n
union. Thereby, ∩U i is a complement of a finite set, and so it is indeed open. ☐
i∈I

Observe that infinite intersections of open sets in the cofinite topology are not
necessarily open.

K12146_C003.indd 35 2/12/15 3:40 PM

 36   ◾    Topological Spaces: Definition and Examples

Example 6: Countable Complement, or Co-countable Topology

Let X be any (always nonempty) set, and let τ be the collection of all complements in
X of (finite or infinite) countable sets. In addition we take that the empty set is in τ.
Proving that τ is a topology is left as an exercise. This is the countable complement
topology, sometimes also called the co-countable topology (Illustration 3.2).

Illustration 3.2 The countable complement topology over a plane: the open sets are obtained
by removing countably many points in the plane (indicated by white disks in the picture).

Note that justifying one of the three axioms requires Proposition 4, Section 1.2. ☐

Example 7: Nested Topology

{
For any (nonempty) space X, let τ = ∅, X , U1 ,U 2 ,…,U n ,… , ∪U i
∞

i =1
} be a family of
subsets of X such that U1 ⊂ U 2 ⊂  ⊂ U n ⊂ U n +1 ⊂  . (Recall again that ⊂ includes =,
and so τ could be finite.) Then, τ is a topology over X, called a nested topology over
X. The finite topological space given in Example 4 is an example of a nested topol-
ogy. Here is one more example: X =  = {…, − 2, − 1, 0, 1, 2, …}, the set of integers,
τ = {∅ , {−1, 1}, {−2, − 1, 1, 2}, {−3, − 2, − 1, 1, 2, 3}, … , X }. It is easy to show (we leave it
to the reader) that nested topologies τ satisfy our axioms of topological spaces. ☐

One more example of a nested topology is shown in Illustrations 3.3 and 3.4.

Illustration 3.3

K12146_C003.indd 36 2/12/15 3:40 PM

 3.1 The Definition and Some Simple Examples   ◾    37  

   Illustration 3.4

The set X (Illustration 3.4) consists of all of the points in the set of infinitely many pen-
tagonal stars positioned as shown in the picture. A topology τ over X consists of the empty
set, the set X itself, and the members of the infinite sequence of finitely many concentric
stars as indicated in Illustration 3.3 (only the first three members of that sequence are
shown). The topological structure of the set X equipped with τ is obviously different from
what one might expect just by viewing the object as a metric subspace of  2 .

Example 8: Order Topology

Let X be equipped with a linear order <. Let τ consist of X and of all unions of inter-
vals of the following type: open intervals, semi-open intervals of type [a, b) where a is
the smallest element of X (if it exists), and semi-open intervals of type (a, b] where b is
the largest element of X (if it exists). Then, we get a topology over X, called the order
topology over X. The proof of the last claim is relegated to the set of exercises. ☐

The last construction of a topological space—where open sets are characterized as

unions of some family of subsets of X—is fairly standard and will be utilized often to define
new topological spaces. We will consider the idea in more detail in Section 3.3.

Example 9: The Euclidean Topology over  , Again

The usual linear ordering of  generates an order topology in which open sets are
unions of open intervals (a, b). This topology coincides with the metric space topol-
ogy (or the Euclidean topology) over  (Exercise 13). ☐

If ( X , τ) is a topological space and if there is a metric over a space X such that the
metric space topology coincides with τ , then we say that ( X , τ) is metrizable. Hence, 
equipped with the linear order topology is metrizable. As we develop the theory further
we will consider various criteria implying metrizability.

K12146_C003.indd 37 2/12/15 3:40 PM

 38   ◾    Topological Spaces: Definition and Examples

Exercises
1. Find all topologies over X = {1, 2}.
2. Find all topologies over X = {1, 2, 3}.
3. Show that the number of topologies over a finite set X grows exponentially with
respect to number of elements of X. More precisely, show that there are at least 2n
topologies over a set with n elements.
4. Let τ be the collection of all subsets U of  satisfying the following: for every
n ∈ , if n ∈U , then n + 1 ∈U . Show that τ is a topology.

5. Let X be a set, Y is space, and f : X → Y a mapping. Show that τ = { f −1 (V ) : V ⊂ Y }

open
is a topology over X.

6. Fix an element x ∈ X and let τ be the collection of all subsets of X containing x,

together with the empty set. Show that τ is a topology.
7. Show that the countable complement topology over a set X is indeed a topology.
8. Let X be a countable set equipped with the countable complement topology. Show that
in this case the countable complement topology coincides with the discrete topology.
9. Let X be any (always nonempty) set, and let τ consist of a collection of subsets of X lin-
early ordered by ⊂, together with the sets ∅ and X. Show that τ need not be a topology.
10. Let Ω be a family of subsets of a (nonempty) set X such that the following is true:
(a) ∅ and X are in Ω.
(b) Every intersection of sets in Ω is a set in Ω.
(c) Every finite union of sets in Ω is a set in Ω.
Show that {U ⊂ X : U c ∈Ω} is a topology over X.
11. Let τ1 and τ 2 be topologies over a set X. Show that τ1 ∩ τ 2 is also a topology over X.
More generally, show that if {τi : i ∈ I } are topologies over X, then so is ∩τi .
i∈I

12. Let X be a space, let Y be a set, and let f : X → Y be a mapping. Define a subset U of Y
to be open if and only if f −1 (U ) is open in X. Show that this defines a topology over Y.
Generalize: Let Xi be spaces, let Y be a set and let fi : Xi → Y , i ∈I , be mappings.
{ }
Show that T = U ⊂ Y : for every i ∈I , fi −1 (U ) ⊂ Xi is a topology over Y.
open

13. Show that the (Euclidean) metric space topology and the (usual) ordered topology
over  are the same.
14. Show that the order topology defined in Example 8 is indeed a topology.
15. Let τ be the collection defined in Example 8, except that we do not require that X be
in τ. Show that in that case τ is not necessarily a topology.

K12146_C003.indd 38 2/12/15 3:40 PM

 3.2 Some Basic Notions   ◾    39  

3.2 Some Basic Notions

A brief historical note: Bernard Bolzano (1781–1848) proved Theorem 5 in the 1830s. His
definition of real numbers was less rigorous than that of Karl Weierstrass (1815–1897), who
independently proved the same theorem.

We saw in Section 2.3 that in a metric space X a subset U is open if and only if for every
element x ∈U there is an open ball B(x, r) contained in U. This can be easily modified as a
criterion for a set to be open in any topological space.
Let X be a topological space and let A be a subset of X. A point a in A is an interior point
for A if there is an open set U such that a ∈U ⊂ A .

Illustration 3.5 The point a in this figure is an

b
interior point of the set S that is enclosed by the
curve, while the point b is not an interior point
for S. This claim is true under our convention
a
that two-dimensional illustrations refer to the
Euclidean 2-space (the plane)—it is precisely
there, in the Euclidean spaces, where most of
our terminology meets the meta-mathematical
(“intuitive”) meaning of the words we use. That
convention notwithstanding, there is a topol-
ogy over  2 such that b is an interior point for
S and such that a is not an interior point for S
(Exercise 1).

Proposition 1. A subset A of a topological space X is open if and only if every point of A is

an interior point for A.

Proof. ⇒ This is obvious.

⇐ Suppose every point of a subset A of X is an interior point for A. Then, for every a in
A there is an open set Ua such that a ∈Ua ⊂ A . Observe that A = ∪ Ua , and so it is open
a∈A
(being a union of open sets).

Given a subset A of a space X, the interior of A, denoted int A, is the subset of A consisting
of all interior points for A. It is clear that int A ⊂ A, and it follows from Proposition 1 that
int A = A if and only if A is open.

Example 1
Consider the square A = {( x , y ) : x ≤ 1, y ≤ 1} as a subset of  2 . If  2 is with the
usual topology, then int A = {( x , y ) : x < 1, y < 1}; considering the usual meaning of
the word interior, that is what we would expect from the Euclidean  2 . However,
this may not be true for other topologies. Indeed, it is true that for every subset B

K12146_C003.indd 39 2/12/15 3:40 PM

 40   ◾    Topological Spaces: Definition and Examples

of A, there is a topology over  2 such that int A = B (Illustration 3.6). That is easy to
achieve: simply define a topology over  2 by declaring open sets to be ∅,  2 , and B.

A intA A intA

Usual topology Other topology

Illustration 3.6
☐

Proposition 2. For every subset A of a space X, intA is the largest open subset of the set A.

Proof. We already know that intA is open, and that it is a subset of A. It is the largest such
subset since if int A ⊂ B ⊂ A with B open, then every point of B is an interior point for
A and so B ⊂ intA.

Example 2
Consider the interval A =[0, 1) as a subset of . We will find intA if the topology over
 is
(i) usual,
(ii) indiscrete,
(iii) finite complement.
(i) The only point of A that is not an interior point for A is 0. So, int A = (0, 1).
(ii) The only subset of A that is open in  is ∅. So, int A = ∅.
(iii) In this case nonempty open sets avoid only finitely many elements of . So, the
only subset of A that is open in  is again the empty set and thus int A = ∅. ☐

We have encountered closed sets when considering metric spaces. The following is a
straightforward generalization.
A subset of a topological space X is closed if it is the complement in X of an open set.
The empty set and the whole set X are closed, since both are open and are mutual com-
plements. So, these two sets are both open and closed. These may be the only such sets in a
topological space (we will deal with such spaces in a later chapter), but that does not hap-
pen in general. For example, in the discrete topology over X all subsets of X are both open
and closed. Open and closed subsets are sometimes called clopens.
The closed intervals [a, b] = {x ∈ : a ≤ x ≤ b} are closed (as the name suggests), since
their complements are open sets of type ( −∞ , a ) ∪ (b , ∞ ). Since open sets in  are unions
of open intervals, it follows from de Morgan laws that closed sets in  are (finite or infinite)
intersections of complements ( −∞ , a]∪[b , ∞ ) of open intervals.
Since closed sets are complements of open sets, the following theorem comes as no
surprise.

K12146_C003.indd 40 2/12/15 3:40 PM

 3.2 Some Basic Notions   ◾    41  

Theorem 3. Let (X, τ) be a topological space. Then the following is true.

(1) ∅ and X are closed sets.
(2) Intersections of closed sets are closed sets.
(3) Finite unions of closed sets are closed sets.

Proof.
Part (1) is obvious.
( ) (de Morgan)
c c
(2): Suppose {Fi : i ∈ I } is a family of closed sets. Then, ∩ Fi ∪( F ) , and since
i
i∈I = i ∈I
the right-hand side of this equality is open (being a union of open sets), so is the left-
hand side. Thereby, ∩ Fi is closed.
i∈I
(3) Exercise 3 below.

If a family of subsets of a set X (no topology yet) satisfies the properties (1), (2), and (3) in
Theorem 3, then the complements of these subsets of X define a topology over X. This claim
is left as an exercise (Exercise 10, Section 3.1). It shows that we can access the notion of topo-
logical space through closed sets via the three statements of Theorem 3 taken as axioms.

Example 3
The semi-line (0, ∞) is open in . The same semi-line considered as a subset of the
usual topological space  2 (that is, the set {( x , y ) ∈ 2 : y = 0 and x > 0}) is not open
there. Is it closed in  2 ? ☐

Example 4
The set { 1n ∈ : n = 1,2,3,…} is not closed in  since its complement is not open.
The complement is not open since the point 0 is not an interior point for the com-
plement; that is so since any open ball around 0 must contain elements of the set
{ 1n ∈ : n = 1, 2, 3, …} . ☐

The point 0 is an accumulation point for the set { 1n ∈ : n = 1,2,3,…}. We define the con-
cept properly in the next two paragraphs.
An open neighborhood of a point x in a space X is an open subset U of X that contains x.
A point x is an accumulation point (or a limit point) for the subset A of a space X if
every open neighborhood of x contains other elements of A. That is, for every open neigh-
borhood Ux of x, we have U x ∩ ( A\{x }) ≠ ∅. The set of all accumulation points for a subset
A of a space X will be denoted by A′.

y
x

A
Illustration 3.7 A subset A of  2: the points x,
z
y, and z are in A′.

K12146_C003.indd 41 2/12/15 3:40 PM

 42   ◾    Topological Spaces: Definition and Examples

Note that we do not stipulate that the accumulation point for a set A must be in A. Indeed,
A′ \ A ≠ ∅ in Example 4. In that example, the set A failed to be closed since it did not con-
tain its only accumulation point. In the next theorem we link these two notions explicitly.

Theorem 4. A subset A of a space X is closed if and only if it contains all of its accumula-
tion points.

Proof.
⇒ Let A be a closed subset of X and suppose it does not contain an accumulation point
x. Then x ∈ Ac , and so Ac is an open neighborhood of x that is disjoint from A. Hence x is
not accumulation point for A, and we have a contradiction.
⇐ Suppose A contains all of its accumulation points and take a point x in Ac . Since x is
not an accumulation point for A there is an open neighborhood of x that does not contain
any points from A. That makes x an interior point for Ac . So, every point of Ac is an interior
point, and so Ac is open. That makes A closed.

Example 5
If  is equipped with the usual topology, then ′ =  . This is true since every open
set in  (with the usual topology) contains as a subset an open interval, and every
open interval contains a rational number (in fact, infinitely many of them). If, on the
other hand,  has the countable-complement topology, then ′ = ∅ since for every
x in  , the set {x } ∪ ( \ ) is an open neighborhood of x that has an empty intersec-
tion with  \{x }. ☐

A subset of n is bounded if it is contained in some ball B(a, r), where we remind the reader
that, unless otherwise stated, we are assuming that n comes with the usual, Euclidean
(metric) topology.
The following theorem, as well as Proposition 6, will be generalized in Section 7.4.

Theorem 5 (Bolzano–Weierstrass). Every bounded and infinite subset A of  has an accu-

mulation point.

Proof. Without loss of generality we can assume that A is a subset of [0, 1]. Subdivide [0, 1]
into [0, 1 2] and [ 1 2 ,1]. Since A is infinite, at least one of these two intervals contains infi-
nitely many points. Subdivide that interval into two closed subintervals of equal length,
and repeat the argument in the previous sentence. Iterate this procedure to get an infi-
nite sequence of subintervals [0,1] ⊃ I1 ⊃ I 2 ⊃ I 3 ⊃ … . It follows from the Cantor’s Nested
∞ ∞
Intervals Theorem (Theorem 2, Section 2.2) that ∩ I j ≠ ∅ . Take a ∈∩ I j ≠ ∅ and an open
j =1 j =1
ball (a − ε , a + ε ) around a. Since, by construction, the lengths of the intervals I j tend to 0 as
j tends to infinity, and since a belongs to each of these intervals I j , there is a large enough
j such that a ∈ I j ⊂ (a − ε , a + ε ). By construction, each I j contains infinitely many points

K12146_C003.indd 42 2/12/15 3:40 PM

 3.2 Some Basic Notions   ◾    43  

of A, and thus, so does (a − ε , a + ε ). So, the set (a − ε , a + ε ) contains a point from A other
than a, implying that a ∈ A′.

A similar argument justifies the following.

Proposition 6. Every infinite and bounded subset of n has an accumulation point.

The closure of a subset A of a space X, denoted A, is the smallest closed subset of X such
that A ⊂ A.

Theorem 7. (a) If A ⊂ B then A′⊂ B′.

(b) For every subset A of a space X we have that A = A ∪ A′.

Proof. (a) This is left as an exercise (Exercise 7).

(b) First we show that A ∪ A′ is closed. According to Theorem 4, it suffices to show that
A ∪ A′ contains all of its accumulation points. Suppose otherwise; that is, suppose
that x is an accumulation point for A ∪ A′ and x ∉ A ∪ A′. If every open neighbor-
hood of x contains points from A, then x ∈ A′ and so x ∈( A ∪ A′ ) ; contradiction.
So, there is an open neighborhood U of x that contains points of
U
A′ and does not contain points of A. Let y ∈U ∩ A′ . Then (see x
Illustration 3.8) U would be an open neighborhood of y ∈ A′ that y
does not contain any points of A, arriving again at a contradic-
tion. We have exhausted all possibilities and we can conclude that
Illustration 3.8
A ∪ A′ contains all of its accumulation points. So, it is closed.
Since A is the smallest closed set containing A, we have that A ⊂ A ∪ A′ . If the
inclusion is proper then there must be some x ∈ A′ such that x ∉ A . Since A is
closed, Theorem 4 implies that ( A)′ ⊂ A . So x ∉( A)′ , and thereby A′ ⊂ ( A)′ . The
last statement contradicts part (a).

A sequence x1 , x 2 ,… , xn ,… of points in a topological space X converges to a point x if for

every open neighborhood U of x, there is a positive integer N such that xn ∈U for every
n > N . This definition is compatible with the definition of the same concept given for metric
spaces.
We will, as before, use the shorter notation ( xn ) for a sequence x1 , x 2 ,… , xn ,…. If the
sequence ( xn ) converges to x we will write either lim xn = x or ( xn ) → x . So, ( xn ) → x if
n→∞
for any open neighborhood of x, from some point on, all of the members of the sequence
are in that neighborhood.

Proposition 8. Suppose X is a space and x1 , x 2 ,…, xn ,…∈ A ⊂ X . If ( xn ) → x then x ∈ A .

K12146_C003.indd 43 2/12/15 3:41 PM

 44   ◾    Topological Spaces: Definition and Examples

Proof. If x = xm for some m, then x ∈ A and so x ∈ A .

Otherwise, it is obvious that x ∈ A′, and by Theorem 7(b) y
it follows that x ∈ A . ◼

The converse of Proposition 8 is true if X is a metric space, A

but not true in general (Exercise 9).
z
A point x in a subset A of a space X is a boundary point
for A if for every open neighborhood Ux of x, we have Illustration 3.9 A subset A of
that Ux ∩ A ≠ ∅ ≠ Ux ∩ Ac (Illustration 3.9). The set of all  2: the points y and z are in ∂A;
boundary points of A will be denoted by ∂A, and will be compare with Illustration 3.7.
called the boundary of A.

Example 6
For subsets A of the Euclidean space n the notion of boundary is compatible with
the usual meaning of the word, at least when the sets A are simple enough. For exam-
ple, if A = {( x , y ) : x 2 + y 2 ≤ 1} (the closed unit disk), then ∂ A = {( x , y ) : x 2 + y 2 = 1}
(the unit circle). ☐

Our intuition and expectations should be restrained when dealing with more exotic spaces
or sets. For example, in Exercise 15 the boundary of a set is the “interior” (the dictionary
meaning) of that set.

Theorem 9. For every subset A of a space X the following is true.

(a) ∂A is a closed set.
(b) A = int A ∪ ∂ A and int A ∩ ∂A = ∅ (and so ∂ A = A \ int A = A ∩ (int A)c ).
(c) A is open and closed (i.e., a clopen) if and only if ∂A = ∅.

Proof.
(a) Every point in (∂ A)c is either in A\ ∂ A or in Ac \ ∂ A . Suppose x ∈ A\ ∂ A. Then there
must be an open neighborhood of x that is entirely in A\ ∂ A because otherwise
every open neighborhood must contain points from ∂A or from Ac \ ∂ A , which
would force x to be in ∂A. (Why is it true that if every open neighborhood of x
contains points from ∂A then x is in ∂A?) This makes every point of A\ ∂ A an inte-
rior point. So A\ ∂ A is open. By symmetry, so is Ac \ ∂ A . So (∂ A)c is open. So ∂A is
closed.
(b) It is obvious that intA ∩ ∂A = ∅ (just compare the definitions of interior and bound-
ary points). It is also obvious that every point of A is either an interior point or a
boundary point. So, A ⊂ int A ∪ ∂ A.
If x is an accumulation point for A that is not in A, then the definitions of accu-
mulation points and boundary points imply that x ∈∂A . If x is an accumulation

K12146_C003.indd 44 2/12/15 3:41 PM

 3.2 Some Basic Notions   ◾    45  

point for ∂A then x is in ∂A; this again follows from the definition of boundary
points. We have thus shown that int A ∪ ∂A contains all of its accumulation points
and so it is closed. Since A is the smallest closed set containing A, it follows that
A ⊂ int A ∪ ∂ A.
Take any point x ∈intA ∪ ∂ A . If x ∈ A , then x ∈ A . If x ∉ A , then x ∈∂A . It fol-
lows directly from the definition of ∂A that in this case x ∈ A′. Hence, in all cases
x ∈intA ∪ ∂ A implies x ∈ A ∪ A′ = A . So int A ∪ ∂ A ⊂ A.
(c) It follows from part (b) that intA = A if and only if ∂A = ∅ . Since intA ⊂ A ⊂ A we
have that int A = A = A if and only if ∂A = ∅, so that A is both open and closed if
and only if ∂A = ∅.

We end this section with a notion that will be encountered frequently as we proceed.
A space X is Hausdorff if for every distinct point x , y ∈ X , there are disjoint open neigh-
borhoods U and V of x and y, respectively (see Illustration 3.10).

U V
y
x
Illustration 3.10 In a Hausdorff space every
two points can be separated with two disjoint
open sets.

Every metric space is Hausdorff (Exercise 4, Section 2.1). The indiscrete space with more
than two elements is clearly not Hausdorff.

Exercises
1. Consider Illustration 3.5 and its caption. Find a topology over  2 such that b is an
interior point for S and at the same time a is not an interior point for S.
2. Consider the square A = {( x , y ) : x − 2 < 1, y − 2 ≤ 1} as a subset of the set  2 . Find
intA if  2 is equipped with
(i) the discrete topology,
(ii) the usual topology,
(iii) the topology induced by the post office metric (Section 2.3) with (0, 0) being the
special point (the mailbox).
3. Show that unions of finitely many closed sets are closed. Give an example of a family
of closed sets in  (usual topology), such that their union is not closed.
4. Find A′ if A is as in Exercise 2.

K12146_C003.indd 45 2/12/15 3:41 PM

 46   ◾    Topological Spaces: Definition and Examples

5. Find ′ and ( \ )′ if  is equipped with

(i) discrete topology,
(ii) indiscrete topology,
(iii) finite complement topology.
6. (a) Give an example of a nonempty subset A of  2 (usual topology) such that
A′ = ∅ .
(b) Give an example of a countable subset A of  2 (usual topology) such that A′ =  2 .
(c) Show that there is no nonempty and bounded subset A of  2 (usual topology)
such that A′ = A and ∂A = ∅ .
7. Let A and B be subsets of a space X and assume A ⊂ B in parts (a), (b), and (c). Prove
the following.
(a) A′ ⊂ B′ .

(b) int( A ) ⊂ int( B ).

(c) A ⊂ B .
(d) ( A ∪ B )′ = A′ ∪ B′.
8. Let A be a subset of a metric space X. Show that ( A′ )′ ⊂ A′ . Give an example of a sub-
set A of  2 (always usual topology unless otherwise stated!) for which the inclusion
is proper, and give an example of A ⊂  2 for which ( A′ )′ = A′ .
9. Let X be a metric space and let A be a subset of X. Show that if x ∈ A′ then there is a
sequence ( xn ) of elements in A that are distinct from x such that ( xn ) → x . Show that
that is not necessarily true in the case where X is not a metric space.
10. Show that the following is true for every subset A of a space X.

(a) int(int A) = int A.

(b) ∂( ∂A ) ⊂ ∂A .
(c) ( A) = A .
11. Consider the space  and let  1 and  2 be two disjoint copies of that space.
Define a topology over  1 ∪  2 by declaring open sets to be all of the sets of type
A1 ∪ A2 where A1 and A2 are copies in  1 and  2 , respectively, of a single open set
A in .
(a) Show that this defines a topology over  1 ∪  2.
(b) Find a subset A of  1 ∪  2 (with the above topology) such that ∂(∂A) is a proper
subset of ∂A.

K12146_C003.indd 46 2/12/15 3:41 PM

 3.2 Some Basic Notions   ◾    47  

12. In Illustration 3.11 we depict a set F of points in

 2 (only the black dots count).
Find a topology over  2 such that int F = F and
F ′ = F . Find another topology over  2 such that
int F = F and F′ = ∅ .
13. Find a topology over  2 such that the bound-
ary of the square A = {( x , y ) : x ≤ 1, y ≤ 1} is
{( x , y ) ; x < 1, y < 1}.
14. Which of the following topologies over  is
Hausdorff? Illustration 3.11
(a) Co-finite topology.
(b) Co-countable topology. B

15. Define a topology over  2 as follows. Open sets are

the empty sets, {( x , y ) ∈ 2 : x 2 + y 2 = 1} , and all
sets of type A ∪ B where A is any subset of the open
disk {( x , y ) ∈ 2 : x 2 + y 2 < 1} and B is any subset of A
the complement of that open disk that contains a
ring {( x , y ) ∈ 2 : 1 ≤ x 2 + y 2 ≤ r } for some r >1 (see
Illustration 3.12). Call it the In–Out topology over
2. Illustration 3.12
(a) Prove that the In–Out topology over  2 is indeed a topology.
(b) Show that in the In–Out topology the boundary of the closed unit disk
D 2 = {( x , y ) ∈ 2 : x 2 + y 2 ≤ 1} is the open unit disk {( x , y ) ∈ 2 : x 2 + y 2 < 1}.
What is the interior of D 2 ?
16. Consider  2 with the post office topology with (0, 0) being the designated point (the
post office). For A = {( x , y ) ∈ 2 : ( x − 1)2 + y 2 < 1} find intA, A, A′, and ∂A.
17. Show that for every two subsets A and B of a space X the following are true.

(a) int( A ∩ B ) = (int A ) ∩ (int B ).

(b) (int A ) ∪ (int B ) ⊂ int( A ∪ B ) (and give an example where the inclusion is
proper).

(c) A ∪ B = A ∪ B .

(d) A ∩ B ⊂ A ∩ B .
18. Suppose X is a metric space and A is a subset of X. Show that if x ∈ A , then there is a
sequence ( xn ) of elements in A converging to x.

K12146_C003.indd 47 2/12/15 3:41 PM

 48   ◾    Topological Spaces: Definition and Examples

19. Show that every sequence of infinitely many elements in a bounded subset of  con-
tains a convergent subsequence.
20. (a) Suppose a space X is a Hausdorff space. Show that if a is an accumulation point
for a subset A of X, then every open neighborhood of a contains infinitely many
points of A.
(b) Show that every metric space is Hausdorff.

21. (a) Find three distinct open subsets A, B, and C of  2 such that ∂A = ∂B = ∂C .
(b) Find three pairwise disjoint subsets A, B, and C of  2 such that A ∪ B ∪ C =  2
and such that ∂A = ∂B = ∂C .
(c) Find four pairwise disjoint subsets A, B, C, and D of  2 such that A ∪ B ∪ C ∪ D =
 2 and such that ∂A = ∂B = ∂C = ∂D .
(d) Find infinitely many pairwise disjoint subsets of  2 that share a common
boundary.
22. Denote the usual (Euclidean) topology over  by τ , and the set of irrational num-
bers by J. Define σ = {U ∪ T : U ∈ τ, T ⊂ J }.
(a) Prove that σ is a topology over  and that τ ⊂ σ.
(b) A point x in a space X is isolated if {x} is open in X. Find all isolated points in 
equipped with σ.

3.3 Bases

A brief historical note: The topological space introduced in Example 6 was first described by
the American mathematician Robert Sorgenfrey (1915–1995); the spaces given in Examples 7
and 8 were found independently by Robert Lee Moore (1882–1974) and Viktor Niemytzki (also
spelled Nemytskii) (1900–1967).

A basis for a topological space X is a family B of open sets such that every open subset of
X is a union of some members of B. Sometimes such a set B is called an open basis for
X: this finesse in the terminology emphasizes the presence of a topology over X before
the basis is identified (compare with Theorem 2 further below). The empty set ∅, which
is open in every space, should be considered as the empty union of the elements of the
basis.

Proposition 1. A collection B of open sets in a space X is a basis for the space X if and only
if for every open subset U of X, and every x ∈U , there is a B ∈B such that x ∈B ⊂ U .

K12146_C003.indd 48 2/12/15 3:41 PM

 3.3 Bases   ◾    49  

B1
x1 U

B3 x3
B2
x2
Illustration 3.13 Every point xi in U is in
some Bi ∈B , such that Bi ⊂ U .

Proof. ⇒ If B is a basis for the space X, then every open subset U in X is a union of mem-
bers of B , and so for every x ∈U, there is a B ∈B such that x ∈B ⊂ U .
⇐ Let B be a collection of open sets such that, for every open set U, and for every x ∈U ,
there is Bx ∈B such that x ∈ Bx ⊂ U . Then U = ∪ Bx .
x ∈U

Example 1
The set of all open sets is surely a basis for the underlying topological space. This is
the most uninteresting example of a basis. The point of introducing the concept of a
basis is precisely in getting smaller families of open sets, used as building blocks to
assemble the whole topology. ☐

Example 2
The set of open intervals is a basis for the usual topology over . The set of all open
balls B(a, r), a ∈n , r ∈ + , is a basis for the usual topology over n . More gener-
ally, the set of open balls B(a, r) in a metric space X , a ∈ X , r ∈ + , is a basis for the
metric space topology over X. All of these three claims follow directly from the
definitions. ☐

Example 3
The set {X} is a basis for the indiscrete topology over X. A minimal (under inclusion)
basis for the discrete topology consists of all one-element sets (singletons). ☐

Now we turn to a related problem: if X is merely a set, and if B is a collection of subsets X,

when is B a basis for some topology over the set X? One answer is: the set of unions of ele-
ments in B must satisfy the axioms of topological spaces. Before we state a better criterion,
we take a look at a couple of negative examples.

Example 4

The set D = {{1}, {2}} is not a basis for a topology over the set X = {1, 2, 3}. This is true
since X is must be an open set in any topology over X, and since it is not a union of
the elements of D. ☐

K12146_C003.indd 49 2/12/15 3:41 PM

 50   ◾    Topological Spaces: Definition and Examples

Example 5
The set of intervals D = {(n , n + 2): n ∈} is not a basis for a topology over . The
set S of all unions of elements of D does not satisfy all of the axioms of topological
spaces: the intervals (1,3) and (2,4) are obviously in S (they are in D ), but their inter-
section is not (and so the last axiom of topological spaces fails). ☐

Theorem 2. A collection B of subsets of a set X is a basis for some topology over X if and
only if the following two conditions are satisfied.

(1) ∪ B = X (that is, X is the union of the elements of B).

B∈B
(2) For every B1 and B2 in B and for every x ∈B1 ∩ B2 , there is B ∈B such that
x ∈B ⊂ B1 ∩ B2 (see Illustration 3.12).

B1 B
x B2

Illustration 3.14

Proof.
⇒ Suppose B is a basis for a topology τ over X. This means that the set of unions of
elements of B satisfies the axioms for topological spaces. In particular the first axiom is
satisfied, and so ∪ B = X .
B∈B
Since the elements of B are clearly open in the topology τ, so are their finite intersec-
tions. Consequently, for every B1 and B2 in B , B1 ∩ B2 must also be a union of elements of
B . So, every element in that intersection must be in some member of B that comprises
that union. So (2) is true too.
⇐ Suppose (1) and (2) are true. We need to show that the set τ of all unions of elements
of B is a topology over X. So, we need to check the axioms for topological spaces. The empty
set is the empty union of elements of B , while it follows from (1) that X is also a union of ele-
ments of B. It is plain that unions of elements of τ are also in τ. It remains to be shown that
finite intersections of elements of τ are also in τ. We will do that for intersections of pairs
of elements of τ; the rest follows by induction. Take any U1 and U 2 in τ; so U1 = ∪ Bi and
i∈I
U 2 = ∪ B j , where all Bi ’s and B j ’s are members of B. Since A ∩ ( ∪ B j ) = ∪ ( A ∩ B j ) for every
( )
j∈J j ∈J j ∈J
 
sets A, B j , j ∈ J , we have, U1 ∩ U 2 = ∪ Bi ∩  ∪ B j  = ∪ ( Bi ∩ B j ). So, it suffices to show that
i∈I  j∈J  i ∈I
j∈J
each Bi ∩ B j is a union of elements of B. Assumption (2) implies that for every x ∈ Bi ∩ B j
there is a member Bx of B such that x ∈ Bx ⊂ Bi ∩ B j . This implies that Bi ∩ B j = ∪ Bx .
x ∈Bi ∩B j

According to the last theorem, if B satisfies (1) and (2) in that theorem, then the family τ
of all unions of elements of B is a topology over the underlying set X. The topology τ is
called the topology generated by the basis B.

K12146_C003.indd 50 2/12/15 3:41 PM

 3.3 Bases   ◾    51  

Remark. Let B be a collection of open subsets of a space X satisfying the conditions (1) and
(2) of Theorem 2. Then B is not necessarily a basis for the given topology over X. Prove this
claim through a simple example.
We say that a topology τ1 is weaker or coarser than a topology τ 2 over the same set X if
τ1 ⊂ τ 2 . At the same time we say that τ2 is stronger or finer than τ1 . The topology generated
by a basis B is the weakest topology that contains the elements of B (Exercise 7).
Defining a topology by specifying a basis that generates that topology is an efficient
approach that is often used when introducing new topological spaces.
The next few examples are important and will be invoked many times.

Example 6: Sorgenfrey Line

Consider the set of all half-open intervals of type [a, b). The union of all such half-
open intervals is , and the intersection of any two non-disjoint intervals of that type
is an interval of the same type. Consequently the set of all half-open intervals of that
type is a basis for a topology over . This topology is called the right half-open inter-
val topology over . The left half-open interval topology over  is defined similarly,
using intervals of type (a, b] as a basis. Since these two topologies have essentially
the same properties, we will deal only with the right half-open interval topology, also
called the Sorgenfrey line. We will use this latter terminology from now on. ☐

Example 7: Half-Disk Topology

Consider the upper closed half plane  2up = {( x , y ) : y ≥ 0} . Let B consist of all open
balls B( x , r ) that are entirely in  2up (as is the set A in Illustration 3.15) and of sets of
type B( x , r )up ∪ {x} where B( x , r )up is an open half ball {( x , y ) : ( x − a)2 + y 2 < r , y > 0}
centered at the point x = (a,0) on the x-axis (as is the set B in Illustration 3.15).

Illustration 3.15

It is very easy to see that B satisfies the requirements of Theorem 2. The topology
generated by B is called half-disk topology. ☐

Example 8: Tangent Disk Topology

The underlying set is again the upper closed half plane  2up = {( x , y ) : y ≥ 0} . Let B
consist of all open balls B( x , r ) that are entirely in  2up and are not tangential to
the x-axis (as is the set A in Illustration 3.16), and of sets of type B(a , r ) ∪ {x } where
B(a , r ) is an open ball touching the x-axis at the point x (depicted as the set B in
Illustration 3.16).

K12146_C003.indd 51 2/12/15 3:41 PM

 52   ◾    Topological Spaces: Definition and Examples

x
Illustration 3.16

It is again easy to show that B is indeed a basis (Exercise 3). The topology over  2up
generated by B is called the tangent disk topology (introduced independently by
Niemytzki and Moore). ☐

Example 9: Order Topology, Again

Let X be a set equipped with a linear order <, and let B be the collection of all inter-
vals of the following types:

• open intervals (a , b ), a , b ∈ X , a < b ,

• semi-open intervals of type [l, b) where l is the least element of X, if such an element
l exists, b ∈ X , l < b ,
• semi-open intervals of type (a, m] where m is the largest element of X, if such an ele-
ment m exists, a ∈ X , a < m .

Then B is a basis for the order topology over X. ☐

Example 10: Lexicographic Order Topology over  2

Define < on  2 as follows: ( x1 , y1 ) < ( x 2 , y 2 ) if y1 < y 2 or if y1 = y 2 and x1 < x 2 (where
when we compare elements of , < stands for the usual linear order on ). This order
is shown in Illustration 3.17. The associated basis of open intervals yields the lexico-
graphic order topology over  2 . As we will notice later, this topology is very different
from the Euclidean topology over  2 .

c
Illustration 3.17 We show a part of the plane
a b  2 . The higher the point, the larger it is. The
horizontal lines inherit the usual linear order
of the real line. In this figure a < b < c .
☐

Example 11
Recall the well-ordered set TP = [1, ω] defined in Example 2, Section 1.3. The order
topology makes it a topological space (to be used later, in Section 8.3). ☐

K12146_C003.indd 52 2/12/15 3:41 PM

 3.3 Bases   ◾    53  

Occasionally it is convenient to define a topology by specifying only a carefully chosen

subset of a basis for that topology. The next definition introduces such a subset.
A set S of open subsets of a space X is a subbasis for the topology over X if the set of
finite intersections of the elements of S is a basis for that topology.

Example 12
The set of all open rays (intervals of type) ( −∞ , a ) and (b , ∞ ) is a subbasis for the usual
topology over . ☐

A set S of subsets of a set X is a subbasis for some topology over X if the set of unions of
finite intersections of the elements of S is a topology over X. We say that this topology is
generated by the subbasis S. It follows from Theorem 2 that a set S of subsets of a set X
is a subbasis for some topology over X if and only if the set B of finite intersections of the
elements of S satisfy the two conditions of that theorem. However, the second condition
will in this case be automatically satisfied since the intersection of two finite intersections
of elements of S is again a finite intersection of elements of S , and thus it belongs to B.
Hence S is a subbasis for some topology over X if and only if the union of the elements of
S is the set X. We have deduced a simple but useful proposition that we state separately.

Proposition 3. A set S of subsets of a set X is a subbasis for a topology over X if and only
if the union of the elements of S is the set X.

A space X is second countable if it has a countable basis.

Example 13
The space  (usual) is second countable: the set of all open intervals with rational
ends is a countable basis for . We provide a short justification. Take any point x in
an interval (a, b). Then, there is a rational number r between a and x, and there is a
rational number p between x and b, so that x ∈(r , p ). It follows that every open inter-
val is a union of smaller open intervals with rational ends and so every open set is
a union of open intervals with rational ends. So, the countable set of open intervals
with rational endpoints is a basis for . ☐

Similar argument shows that the Euclidean space n is also second countable (Exercise 8).

The following two examples will be used later on.

Example 14
Let { i : i ∈} be a family of pairwise disjoint copies of . Define a linear order
< over X = ∪  i as follows. On each  i the order is the usual linear order; if
i∈
x ∈ i and y ∈ j for i ≠ j , then x < y if and only if i < j . The space X equipped
with the order topology fails to be second countable (Exercise 16). We note in

K12146_C003.indd 53 2/12/15 3:41 PM

 54   ◾    Topological Spaces: Definition and Examples

passing that this space and the lexicographic order space over  2 (Example 10) are
in essence the same (or, they are homeomorphic, a notion that will be introduced
in Section 3.5). ☐

Example 15
The Sorgenfrey line S (Example 6 above) is not second countable. Assume otherwise,
and let B be a countable basis for S , and consider the family F = {[ x , x + 1): x ∈}
of open sets in S . Since B is a basis, for each [ x , x +1) there exists Bx ∈B such that
x ∈Bx ⊂ [x , x + 1). This choice defines a mapping f :[x , x + 1)  Bx from F into B.
This mapping is not one-to-one, since B has cardinality smaller than F . So, there is
a set B0 in B such that x1 ∈B0 ⊂ [x1 , x1 + 1) and x 2 ∈B0 ⊂ [x 2 , x 2 + 1) for some x1 ≠ x 2 .
By symmetry we can assume that x1 < x 2 . In that case x1 ∉[x 2 , x 2 + 1) , and so, since
B0 ⊂ [x 2 , x 2 + 1), we have that x 1 ∉B0, a clear contradiction. ☐

A set B x of open neighborhoods of a point x in a space X is a local basis at the point

x if for every open neighborhood U of x there is an element Bx ∈B x such that Bx ⊂ U
(Illustration 3.18).

Illustration 3.18 Local basis: for every open

set U containing x there is a member Bx of the
%x
local basis B x at x such that Bx ⊂ U .

This is the first local notion that we have encountered so far (we will see more of local
notions and properties).
If B is a basis for a space X, then the set of all elements of B that contain a fixed point
of X is a local basis for that point. That claim is very easy to justify. It is not much more dif-
ficult to show the converse (next proposition).

Proposition 4. If {B x : x ∈ X } is a collection of local bases in a space X, then ∪ B x is a

x ∈X
basis for X.

Proof. Exercise 9.

A space X is first countable if for every point x in X there is a countable local basis at x.
Metric spaces are first countable (Exercise 13); in particular n is first countable. (However,
there are metric spaces which are not second countable.) In the following example we show
that uncountable sets with cofinite topology are not first countable.

K12146_C003.indd 54 2/12/15 3:41 PM

 3.3 Bases   ◾    55  

Example 16
Let X be any uncountable set, equipped with the co-finite topology. Suppose X is first
countable. This means that, given a point x ∈ X , there is a countable local basis
c
{B j : j ∈+ } at x. Consider ∩ B j : since  ∩ B j  = ∪ Bcj , and since each Bcj is finite
∞ ∞ ∞

j =1  j =1  i=1
∞
(recall that each B j is open), the set ∪ Bcj is countable (Proposition 4 in 1.2). Hence
∞ i =1 ∞
its complement ∩ B j is uncountable. Choose any point y in ∩ B j other than x. Then
j =1 j =1
∞
∩ B j is not a subset of the open neighborhood X \{y} of x, and so neither is any B j .
j =1
We have a contradiction. So X is not first countable. ☐

Second countable implies first countable, but the converse is false (see Exercise 10).

Exercises
1. Consider the set B = {{(c , y ) ∈ 2 : a < y < b} : a, b, c ∈ , a < b} of all vertical line seg-
ments in  2 . Show that B is a basis for a topology over  2 .
2. Show that every interval (a, b) is open in the Sorgenfrey topology over . Is (a, b) closed
there? Show that the Sorgenfrey topology is finer than the Euclidean topology over .
3. Show that the family B in the definition of the tangent disk topology is indeed a basis
for a topology over  2up .
4. [Fürstenberg, 1955] Let B be the family of all arithmetic progressions in ;
Ba ,b = {… , a − 2b, a − b, a, a + b, a + 2b, …} is a generic element of B, for some a, b ∈ .
(a) Show that B is a basis for a topology over . [Hint: if x ∈ Ba ,b ∩ Bc ,d , then show
that x ∈ Bx ,bd ⊂ Ba ,b ∩ Bc ,d .]
(b) Show that there are infinitely many primes. [Hint: Show that for every m ∈,
B 0,m is closed. Assuming the set P of primes is finite, it would follow that ∪ B0, p
p ∈P
is also closed. What is the complement of this set?]
5. Show that the family B in the definition of the order topology (Example 9) is indeed
a basis.
6. Show that if {Bn : n ∈+ } is a local basis at a point x ∈ X , and if xn ∈ Bn for every
n ∈+ , then the sequence ( xn ) converges to x.
7. Show that the topology generated by a basis B is the weakest topology that contains
the sets of B.
8. Show that n with the usual topology is second countable.
9. Show that if {B x : x ∈ X } is a collection of local bases, then ∪ B x is a basis for X.
x ∈X

10. Show that every second countable space is first countable, and find a simple counter-
example refuting the converse.

K12146_C003.indd 55 2/12/15 3:42 PM

 56   ◾    Topological Spaces: Definition and Examples

11. Let X be a second countable space. Show that every basis for X contains a countable
subset that is also a basis for X.
12. Show that if X is first countable, then for every x ∈ X there is a countable local basis
{Bn : n ∈+ } at x such that B1 ⊃ B2 ⊃ B3 ⊃… .
13. Show that every metric space is first countable.
14. Show that every open set in the usual topology over  is also open in the Sorgenfrey
topology over .
15. Let P be the set of all polynomials on n variables with coefficients in . For a poly­
{
nomial p( x1 , x 2 ,…, xn ) ∈P , define Z ( p) = ( x1 , x 2 ,… , xn ) ∈n : p( x1 , x 2 ,… , xn ) = 0 . }
Show that the collection {n \ Z ( p) : p ∈P } is a basis for a topology over n . This is
the Zariski topology over n . Show that for n = 1 the Zariski topology coincides with
the cofinite topology over .
16. Show that the space defined in Example 14 is not second countable.

17. Let ( X , τ1 ) and (Y , τ 2 ) be topological spaces such that X ∩ Y = ∅ . Show that τ1 ∪ τ 2

is a basis for a topology over X ∪ Y . Show that this claim is in general false when X
and Y are not disjoint.

18. Let Xi , i ∈I , be a collection of spaces. Show that the collection B of all sets of type
∏ U i , where U i is open in Xi , and for all but finitely many i ∈I , U i = Xi , is a basis for
i ∈I
some topology over the set product ∏ Xi .
i ∈I
19. Let X be a space with a nested topology, and denote, as usual, the power set of X by
P (X). For every two open subsets U, and V of X, denote (U ,V ) = W ⊂ X : U  W  V .
{ open }
(a) { }
Show that the collection (U , V ) : U , V ⊂ X ∪{∅} ∪ {P ( X )} is a basis for a topo­
open
logy over P (X). Call it the power topology.
(b) Show that if X ≥ 2 , then the power topology over P (X) is not metrizable.
20. Given a space (X, τ), show that the set of all 〈U1 ,U 2 ,…,U n 〉 =
n
+
{ A ⊂ X : A is closed, A ⊂ ∪U i and each A ∩ U i ≠ ∅}, U i ∈τ, n ∈ , is a basis for a
i =1
topology over the power set P (X).

21. Suppose f : X → Y is a mapping from a set X into a space Y. Show that the set
{ f −1 (U ) : U is open in Y } is a basis for some topology over X.

{ }
22. Show that the family ( x − n1 , x + n1 ) \ {x } ⊂  : x ∈ , n ∈+ of punctured intervals
is a basis for a topology τ over  . What is the relationship between τ and the usual
topology over ?

K12146_C003.indd 56 2/12/15 3:42 PM

 3.4 Dense and Nowhere Dense Sets   ◾    57  

23. (a) Show that if X is first countable and x ∈ X is an accumulation point for a subset A
of X, then there exists sequence x1 , x 2 ,… of elements in A\{x} that converges to x.
(b) Show that the co-countable topology over  is not first countable.
(c) Find a subset A of the space  equipped with the co-countable topology, and a
point x ∈ such that x ∈ A′ but no sequence of elements of A converges to x.

3.4 Dense and Nowhere Dense Sets

A brief historical note: René-Louis Baire (1874–1932) proved Theorem 4 in his dissertation, 1899.

A subset D of a space X is dense in X if D ∩ U ≠ ∅ for every nonempty open subset U of X.

Illustration 3.19 The dense set D is repre-

sented by the grainy texture; every open sub-
set U of X has a nonempty intersection with D.

Example 1
For every space X the set X is always dense in itself. If X is equipped with the discrete
topology, then X is the only dense set. On the other hand if X is equipped with the indis-
crete topology then every subset of X is dense in X. These claims are almost trivial. ☐

Example 2
The set of rational numbers is dense in the usual space . Similarly, the set  ×  is
dense in  2. In general n is dense in n . All of these claims follow from the fact that
there is a rational number between any two distinct numbers. ☐

Here is an alternative criterion for dense sets.

Proposition 1. A subset D of a space X is dense in X if and only if D = X .

Proof.
⇒ Suppose D  X . Then U = X \ D is a nonempty open subset of X such that D ∩ U = ∅ .
So, D is not dense in X.
⇐ Suppose D = X , take a nonempty open subset U of X and choose a point y in it. If y ∈D
then certainly D ∩ U ≠ ∅. Otherwise, since D = D ∪ D ′ and D = X , it must be that y ∈D ′. But
then it follows from the definition of accumulation points that D ∩ U ≠ ∅. So, D is dense.

K12146_C003.indd 57 2/12/15 3:42 PM

 58   ◾    Topological Spaces: Definition and Examples

We notice that the dense sets in Example 2 are countable. It follows that the Euclidean
spaces n are separable. A space X is separable if it contains a countable dense set.
Being separable and being second countable both involve existence of countable sets. So
it is not a surprise that there is a close link between these two concepts.

Proposition 2. Every second countable space is separable.

Proof.
Suppose X is second countable and let B be a countable basis for X. Choose an element x
from each nonempty member of B. This gives a countable set D . It is now easy to see that
D is a dense set: if U is a non-empty open set then it is a non-empty union of members of
B, and so it contains all of the elements in D corresponding to these members of B .

The converse of Proposition 2 is not true in general (Exercise 4). However, in the case of
metric spaces, it is true:

Proposition 3. Every separable metric space is second countable.

Proof. Let (X, d) be a separable metric space; hence it contains a countable dense set
D = {d1 , d2 ,…, dn ,…} . Consider the family B = {B(di , q) : i ∈+ , q ∈} of open balls in X.
By Proposition 4 in 1.2, B is countable. We prove it is a basis for the metric topology over X.

B (x, r)

x B
d
Bx
q

Illustration 3.20

Let B(x, r) be any ball in X (Illustration 3.20). Take a small concentric ball, say
B = B ( x , 10r ) . Since D is dense, there exists d ∈B ∩ D ; pick any rational number q such
that 2 10r < q < 3 10r . Then x ∈B(d , q ) ⊂ B( x , r ). Denoting Bx = B(d , q) , we conclude that
for every ball B(x, r) there is a ball Bx ∈B , such that x ∈ Bx ⊂ B( x , r ). It follows from
Proposition 1 in 3.3 and the definition of basis that B is a basis for X.

A subset A of X is nowhere dense in X if int( A) = ∅ . In a way, being a nowhere dense subset

A of X is the opposite of being a dense subset D of X: in the latter case nonempty open sets
must intersect D, while in the former case nonempty open sets contain nonempty open
subsets that completely avoid A (Exercise 11). In that sense one can say that nowhere dense
sets are thinly distributed within the space X.

K12146_C003.indd 58 2/12/15 3:42 PM

 3.4 Dense and Nowhere Dense Sets   ◾    59  

U Illustration 3.21 The nowhere dense set A

is represented by the points-pattern: every
open set U has a nonempty open subset V
V
avoiding A.

Example 3
Every proper subset of the indiscrete space is nowhere dense. The only nowhere dense
subset of the discrete space is the empty set. The subset  of  is nowhere dense in .
So is any finite subset of . If we consider m as a subset of n , n > m (via the map-
ping ( x1 ,…, xm )  ( x1 ,…, xm , am+1 ,…, an ), am+1 ,…, an fixed elements in ), then m is
nowhere dense in n. ☐

A space X is of the first category if it is a union of countably many nowhere dense sets.
Otherwise we say that X is of the second category. Intuitively, spaces of the first category
are thin spaces, being countable unions of thin sets, and spaces of the second category are
thick spaces.

Example 4
Since every singleton is nowhere dense in , every countable subset of  is a union
of countably many nowhere dense sets in . The same argument applies to n ,
n ∈+ , in place of . Note that it is not true that a countable subset X of  (or
n ), considered as metric subspace of  (or n, respectively), is necessarily of the
first category, since although singletons are nowhere dense in , they need not be
nowhere dense in X (Exercise 14). ☐

The following theorem tells us that complete metric spaces are always thick (i.e., of second
category).

Theorem 4. (Baire Category Theorem) Every complete metric space is of the second
category.

Proof. Let X be a complete metric space, and suppose it is of the first category. This means
∞
that X = ∪ An , where each An is nowhere dense in X. It follows that for every open set
n=1
U in X, there is an open ball Bn such that Bn ⊂ U and such that Bn ∩ An = ∅ (Exercise 11).
We now choose these balls carefully. First we choose B1 to be any open subset such that
B1 ∩ A1 = ∅. Then we choose B2 to be of radius less than 1 2 and such that B2 ⊂ B1 and
B2 ∩ A2 = ∅ . Recursively, if Bn is chosen, we choose Bn+1 such that its radius is less than
1
n +1
, such that Bn +1 ⊂ Bn , and such that Bn +1 ∩ An +1 = ∅ . This gives a nested sequence

K12146_C003.indd 59 2/12/15 3:42 PM

 60   ◾    Topological Spaces: Definition and Examples

( Bn )n∞=1 of balls. Choose xn ∈ Bn , n = 1, 2,… to get a sequence ( xn )n∞=1 . Since the radius of
Bn is less than 1n , the sequence ( xn )n∞=1 is a Cauchy sequence. By the completeness of X, it
converges to some x ∈ X .
Assume for a moment that x ∉ Bn for some n. Then x ∉ Bm for every m ≥ n. There is a ball
B around x and disjoint from Bn (Exercise 8 in 2.1); so B ∩ Bm = ∅ for every m ≥ n. This con-
tradicts the fact that ( xn )n∞=1 converges to x. Hence x ∈ Bn for every n. Since Bn ∩ An = ∅, it
∞
follows that x ∉ An for every n. This contradicts the assumption that X = ∪ An . Hence, X
n=1
could not be of the first category.

Exercises
1. Find a space X such that X′ = ∅ . True or false: If a space X is such that X′ ≠ ∅ then
X′ is dense in X.
2. Show that the Sorgenfrey line is first countable and separable.
3. Show that the Sorgenfrey line is Hausdorff but not metrizable.
4. Show that both the tangent disk and half-disk topologies are separable but not
second countable.
5. Find reasonable sufficient and necessary conditions for a space equipped with the
countable complement topology to be separable.
6. Find a nonseparable metric space. (Hint: Consider the metric spaces we have intro-
duced so far.)
7. Show that the only closed and dense subset of a space X is X itself.
8. The set D of dyadic numbers is the subset of the rational numbers consisting of all
fractions of the type m2i , where m is an integer and i is a natural number. Show that D
is dense in .
9. Show that a finite union of nowhere dense sets is nowhere dense.
10. Is the subspace  \  of  (consisting of all irrational numbers) of the first category?
Justify!
11. (a) Show that A is nowhere dense in X if and only if for every nonempty open subset
U of X, there is a nonempty open set V such that V ⊂ U and V ∩ A = ∅ .
(b) Show that A is nowhere dense in a metric space X if and only if for every
nonempty open subset U of X, there is a ball B such that B ⊂ U and such that
B∩A = ∅.
12. A subset B of a space X is discrete if for every b ∈B there is an open neighborhood U
such that U ∩ B = {b}. Show that if X is a metric space without isolated points, then
the closure of any discrete subset of X is nowhere dense in X.

K12146_C003.indd 60 2/12/15 3:42 PM

 3.5 Continuous Mappings   ◾    61  

13. Show that if A is nowhere dense in X, then ( A ) is dense in X.

c

14. Find countably infinite subspaces A and B of  such that A is of the second category
and B is the first category.

3.5 Continuous Mappings

A brief historical note: The first formal definition (the “ε − δ ” definition) of a continuous func-
tion was given in 1817 by Bolzano. In 1910 Fréchet introduced continuous mappings of abstract
topological spaces.

Example 1

Consider the space X = {1, 2, 3} and the space Y = {a , b , c}, both with the discrete topo­
logy. It is evident that there is no substantial difference between these two spaces—
the only difference is in the names of the elements. ☐

In order to formally compare two topological spaces we need a concept that will naturally
relate the open sets of the two spaces. The main goal of this section is the introduction of
such a concept.
We start with a local notion. Recall the definition of continuity in metric spaces: a func-
tion f :  →  is continuous at a point a ∈  if for every open ball B( f (a ), ε ) there is an
open ball B(a, δ) such that f ( B(a , δ )) ⊂ B( f (a ), ε ). Open balls comprise the standard basis
for the usual topology over . We use that observation to generalize the last definition to
general topological spaces.
Let X and Y be topological spaces. A mapping f : X → Y is continuous at a point a ∈ X
if there is a local basis B a at a and if there is a local basis B f (a ) at f (a) such that for every
B f (a ) ∈B f (a ) there is Ba ∈B a such that f ( Ba ) ⊂ B f (a ) (see Illustration 3.22).

Y
X
Ba
a
f(a)
f
Bf(a)

Illustration 3.22

If f is not continuous at a ∈ X , then we say that it is discontinuous at a.

K12146_C003.indd 61 2/12/15 3:42 PM

 62   ◾    Topological Spaces: Definition and Examples

Proposition 1. A mapping f : X → Y is continuous at a point a ∈ X if and only if for

every open neighborhood V f (a ) of f(a) there is an open neighborhood Ua of a, such that
f (Ua ) ⊂ V f (a ) .

Y
X
Ua
a f(a)
f
Vf(a)

Illustration 3.23

Proof. ⇒ Let f : X → Y be continuous at a point a ∈ X . Let B a , B f (a ) be as in the defini-

tion of continuity and let V f (a ) be an open neighborhood of f (a). Since B f (a ) is a local basis,
there is B f (a ) ∈B f (a ) such that B f (a ) ⊂ V f (a ). Since f is continuous, there is Ba ∈B a such that
f ( Ba ) ⊂ B f (a ). Now take Ua = Ba.
⇐ This is clear: just take B a , B f (a ) to be the sets of all open neighborhoods of a and f (a),
respectively.

Example 2
 2 if x ≥ 0
The mapping f  :  →  defined by f ( x ) =  is continuous at every
 –2 if x < 0
point except 0. The mapping f is not continuous at 0 since no open neighborhood
of 0 is mapped into the open neighborhood (1, 3) of f (0) = 2 . ☐

Proposition 3 in 2.2 gives an alternative criterion for continuity in the case when f : X → Y
is a mapping between two metric spaces. The same criterion holds when only X is a metric
space. The following proposition generalizes all of that.

Proposition 2. Suppose X is a first countable space and Y is any space. A mapping f : X → Y

is continuous at a point a ∈ X if and only if for every sequence ( xn ) converging to a in X,
the sequence ( f ( xn )) converges to f (a) in Y.

Proof. ⇒ Virtually the same as the corresponding proof of Proposition 3 in 2.2.

⇐ Suppose for every sequence ( xn ) converging to a in X, the sequence ( f ( xn )) converges
to f (a) in Y. Let V be an open neighborhood of f (a). We want to show that there is an open
neighborhood U of a such that f (U ) ⊂ V . Consider the subset W = f −1 (V ) of X. Clearly, it
contains the point a. It suffices to show that a is an interior point for W. We will now show
that the assumption that a is not an interior point leads to a contradiction.
Let B a = {Bi : i ∈+ } be a countable local basis at a. Since for every n ∈+
B1 ∩ B2 ∩ ... ∩ Bn is an open set, and since it is nonempty (a is there), our assumption
that a is not an interior point for W implies that B1 ∩ B2 ∩ ... ∩ Bn ∩ W c ≠ ∅ . Choose
xn ∈ B1 ∩ B2 ∩ ... ∩ Bn ∩ W c . This defines a sequence ( xn ) . The construction of this sequence

K12146_C003.indd 62 2/12/15 3:42 PM

 3.5 Continuous Mappings   ◾    63  

and the fact that B a = {Bi : i ∈+ } is a local basis at a easily imply that ( xn ) converges to a.
On the other hand, it is evident that each f ( xn ) is out of V, so that the sequence ( f ( xn ))
does not converge to f (a).

Looking at Example 2 again, we see that the mapping f is not continuous at 0 since
(− n1 ) → 0 while ( f (− n1 )) is the constant sequence −2, − 2, − 2, …, and so it does not con-
verge to f (0) = 2.

Example 3
Denote the space  with the usual topology by X and denote the space  with the
Sorgenfrey topology by Y. Let f : X → Y and g : Y → X be the identity mappings
( f ( x ) = g ( x ) = x for every x). Then, g is continuous at every point while f is discontinu-
ous at every point. The former statement is true since every open set in X is also open in
Y (Exercise 2, Section 3.3), while f is discontinuous at every a ∈ since no open neigh-
borhood of a in X is mapped inside the open neighborhood [a , a +1) of a in Y. ☐

In the last example we have encountered a function that is continuous at every point of the
domain space. We say that a mapping f : X → Y between two spaces is continuous if it is
continuous at every point of X.
The following proposition summarizes a few important criteria for continuity.

Proposition 3. Let X and Y be two spaces and f : X → Y be a mapping. The following three
claims are equivalent.
(a) The mapping f : X → Y is continuous.
(b) For every open set U of Y, the set f −1 (U ) is open in X.
(c) For every basis BY of the space Y and for every B ∈BY , the set f −1 ( B) is open in X.

Proof. (a ) ⇒ ( b) Assume that f is continuous and that U is an open subset of Y. Since f is

conti­nuous, for every a ∈ f −1 (U ) there is an open subset Va of X such that f (Va ) ⊂ U . Observe
that since f (Va ) ⊂ U it has to be that Va ⊂ f −1 (U ). It follows that f −1 (U ) = ∪ Va , and so
a∈f −1 (U )
f −1 (U ) is open (being a union of open sets).
( b) ⇒ (c ) Since basis sets are open subsets, this part is trivial.
(c ) ⇒ (a ) Assume BY is a basis for Y such that for every B ∈BY , the set f −1 ( B) is open
in X. We want to show that f is continuous at every point a ∈ X . As we have observed in
Section 3.3, the set B f (a ) = {B ∈BY : f (a) ∈ B} is a local basis at f (a). Hence for every open
set V containing f (a), there is B ∈B f (a ) , such that f (a ) ∈B ⊂ V . By our assumption f −1 ( B)
is an open neighborhood of a. Clearly f ( f −1 ( B)) = B ⊂ V . It follows from Proposition 1
that f is continuous at a.

Most of the time we will be using criterion (b) for continuity: f is continuous if and only if
inverse images of open sets are open sets.
The next proposition shows that we can use closed sets in place of open sets to the same
effect.

K12146_C003.indd 63 2/12/15 3:42 PM

 64   ◾    Topological Spaces: Definition and Examples

Proposition 4. A mapping f : X → Y between two topological spaces is continuous if and

only if inverse images of closed sets are closed.

Proof.
⇒ Suppose f : X → Y is continuous and suppose F is a closed subset of Y. Then F c is open
and so f −1 ( F c ) is also open. We conclude that ( f −1 ( F c ))c is closed. Now, it is easy to show
that the sets ( f −1 ( F c ))c and f −1 ( F ) are the same, so that f −1 ( F ) is also closed.
⇐ Similar; we leave it as an exercise.

Example 4
As an application of Baire’s theorem (Theorem 4, Section 3.4), we show that there
is no continuous f :  →  such that f () ⊂  \  , and f ( \ ) ⊂  . Suppose
otherwise. It follows that  = f −1 () ∪  = { f −1 ( x ) : x ∈} ∪ ∪ {x }. Since f is con-
x ∈

( )
tinuous, the set f ( x ) is closed for every x ∈ , and so int f −1 ( x ) = int ( f −1 ( x )) .
−1

Since f −1 ( x ) is a subset of the set  \  of irrationals, we have that int ( f −1 ( x )) = ∅.

( )
Consequently, int f −1 ( x ) = ∅ , and so each f −1 ( x ) is nowhere dense. Hence
 = { f −1 ( x ) : x ∈} ∪ ∪ {x } exhibits  as a space of first category. Since  is a
x ∈
complete metric space, this contradicts Baire’s theorem. ☐

A mapping f : X → Y between two topological spaces is open if for every open subset U
of X, the set f (U) is open in Y. The mapping f is closed if for every closed subset F of X, the
set f (F) is closed in Y.
It is clear from the definition that if f : X → Y is a bijection then f is open if and only if the
inverse mapping f −1 : Y → X is continuous. Similarly, it follows from Proposition 4 that if
f is a bijection, then f is closed if and only if the inverse mapping f −1 : Y → X is continu-
ous. It follows that in the case of bijections, being an open mapping is equivalent to being
a closed mapping.

Example 5
The mapping f in Example 3 is open and closed, while the mapping g in the same
example is not open (even though g is both continuous and a bijection). Note that
open mappings need not be closed (Exercise 7) and closed mappings need not be
open. ☐

A mapping f : X → Y between two spaces is a homeomorphism if f is an open, continuous

bijection. If such a mapping exists then we say that the spaces X and Y are homeomorphic,
and we write X ≅ Y . If X ≅ Y then we will say that each of X and Y is a copy of the other.
Observe that if f : X → Y is a homeomorphism then the inverse mapping f −1 : Y → X
is also a homeomorphism.

K12146_C003.indd 64 2/12/15 3:42 PM

 3.5 Continuous Mappings   ◾    65  

Now we can say exactly what we meant when we stipulated in Example 1 that there was
no substantial difference between the spaces X = {1, 2, 3} and Y = {a , b , c}, both with the
discrete topology: they are homeomorphic. One homeomorphism is 1  a , 2  b , 3  c .
From a topological point of view there is no difference between homeomorphic spaces.
We will sometimes say that such spaces are the same up to homeomorphism.

Example 6
The subspace {( x , y ) : x 2 + y 2 = 1, ( x , y ) ≠ (0,1)} of  2 is homeomorphic to , and
a homeomorphism is described in Illustration 3.24. This homeomorphism is called
the stereographic projection from {( x , y ) : x 2 + y 2 = 1, ( x , y ) ≠ (0,1)} onto  .

NP

P1 P4

P2
P3

f (P1) f (P2) f (P3) f (P4)

Illustration 3.24 The stereographic projection (as indicated by the arrows) is a homeomor-
phism from S1 \{NP} onto , where S1 = {(x, y) ∈  2 : x2 + y2 = 1} is the unit circle.
☐

Example 7
In Illustration 3.25, we depict the stereographic projection S 2 \{NP} (the two-sphere
S 2 = {( x , y , z ) : x 2 + y 2 + z 2 = 1} without the point NP = (0,0,1)) onto  2 . It is clearly
a homeomorphism. The homeomorphisms in Examples 6 and 7 could be described
analytically by computing the intersections of the sphere and the plane with rays ema-
nating from NP and thus could easily be generalized to stereographic projections from
S n \{NP} onto n where Sn = {(x1, x2, … , xn, xn+1) ∈ n +1 : x12 + x22 + … + xn2 + xn+1 2
= 1}
is the unit n-sphere, and where NP = (0, 0, …, 0, 1) in this context.
NP

P1

P2

f (P2)
f(P1)
Illustration 3.25 We see the images f ( P1 ) and
f ( P2 ) of the points P1 and P2 .
☐

K12146_C003.indd 65 2/12/15 3:42 PM

 66   ◾    Topological Spaces: Definition and Examples

Example 8
If two spaces are of different cardinality, then they cannot be homeomorphic. For
example, no topology over  will make the resulting space homeomorphic to  with
any topology. The Euclidean  is not homeomorphic to the Euclidean  2 despite
having the same cardinality; we will postpone the justification of that claim until
the chapter on connectedness. For Euclidean spaces it is true that n is not homeo-
morphic to m if n ≠ m. This is a consequence of the Invariance of Domain Theorem,
which we will encounter further in this book. ☐

Properties preserved by homeomorphisms are called topological properties. We have

noticed earlier that being a fractal metric space is not a topological property. Virtually all
properties we have seen so far or we will deal with are topological.

Proposition 5. Let f : X → Y be a homeomorphism and let A be a subset of X.

(i) If X is first countable, then so is Y.
(ii) If X is second countable, then so is Y.
(iii) f (int A ) = int f ( A ).
(iv) f ( A) = f ( A).
(v) f(∂A) = ∂(f(A)).
(vi) f ( A′ ) = ( f ( A))′ .

Proof. We will prove (ii) and (v); the rest will be left as an exercise.

(ii) Suppose X is second countable and suppose B is a countable basis for X. Then,
the set { f ( B) : B ∈B} is a countable set of open sets. If U is an open set in Y, then
f −1 (U ) is an open set in X, and so f −1 (U ) = ∪ Bi for some elements Bi ∈B . But then

( )
i∈I
U = f ( f −1 (U )) = f ∪ Bi = ∪ f ( Bi ) and so { f ( B) : B ∈B} is a basis for Y.
i∈I i∈I
(v) Suppose y ∈ f ( ∂A ). Then y = f ( x ) for some x ∈∂A . Since x ∈∂A , we have that
Ux ∩ A ≠ ∅ and Ux ∩ Ac ≠ ∅ for every open neighborhood Ux of x. Consequently
f (Ux ∩ A) ≠ ∅ and f (Ux ∩ Ac ) ≠ ∅. Since f is a bijection, f (Ux ∩ A) = f (Ux ) ∩ f ( A)
and f (Ux ∩ Ac ) = f (Ux ) ∩ f ( Ac ). So f (Ux ) ∩ f ( A) ≠ ∅ and f (Ux ) ∩ f ( Ac ) ≠ ∅ for
every open neighborhood Ux of x. Since f is a homeomorphism (so, it is continuous
and open), every open neighborhood of y = f ( x ) is f (Ux ) for some open neighbor-
hood Ux of x. So, as Ux ranges through all open neighborhoods of x, the set f (Ux )
ranges through all open neighborhoods of y. Consequently, since f ( Ac ) = ( f ( A))c ,
the statements f (Ux ) ∩ f ( A) ≠ ∅ and f (Ux ) ∩ f ( Ac ) ≠ ∅ imply that y ∈∂( f ( A)), so
that f ( ∂A ) ⊂ ∂( f ( A )).

K12146_C003.indd 66 2/12/15 3:43 PM

 3.5 Continuous Mappings   ◾    67  

Since f −1 is also a homeomorphism, it follows from what we have proven that

f −1 (∂ B) ⊂ ∂( f −1 ( B)), for every B ⊂ Y. Replacing B = f ( A) in the last equality we get
f −1 (∂( f ( A))) ⊂ ∂( f −1 ( f ( A))), which simplifies to f −1 (∂ f ( A)) ⊂ ∂ A. After applying
f on both sides we get ∂( f ( A)) ⊂ f (∂ A), establishing the other inclusion, and, as a
consequence, showing that ∂( f ( A)) = f (∂ A).

Let X be a set, let Y be a space, and let f : X → Y be a mapping. The weakest topology
over X that makes f continuous is called the topology induced by f. More generally, if
fi : X → Yi , i ∈I , are mappings from a set X into spaces Yi , then the topology over X
induced by { fi : i ∈ I }, is the weakest topology over X making all fi continuous.

Proposition 6. Let fi : X → Yi , i ∈I , be mappings from a set X into spaces Yi . The topology

{
over X induced by { fi : i ∈ I }, is generated by the subbasis S = fi −1 (U ) : U ⊂ Yi , i ∈ I . In
open
particular, if the topology over X is induced by f : X → Y , then that subbasis is the family
}
{ f −1 (U ) : U ⊂ Y .
open }
Proof. Exercise 22.

Here is the dual story: Let X be a space, let Y be a set, and let fi : X → Y , i ∈I , be mappings.
The strongest topology over Y that makes all fi continuous is called the topology induced
by { fi : i ∈ I }.

Proposition 7. Let X be a space, let Y be a set, and let fi : X → Y , i ∈I , be mappings. The

topology over Y induced by { fi : i ∈I } is { }
U ⊂ Y : for every i ∈ I , fi −1 (U ) ⊂ X .
open

Proof. Exercise 12, Section 3.1.

Example 9
The product metric space X1 × X 2 × … × Xn of the metric spaces Xi , i = 1, 2, …, n , was
defined in Section 2.1. It is obvious from its construction that its topology is induced
by the projections pi : X1 × X 2 × … × Xn → Xi , i = 1, 2, …, n . Dually, the product met-
ric topology is also induced by the coordinate mappings ci : Xi → X1 × X 2 × … × Xn ,
i = 1, 2, …, n . ☐

Exercises
1. (a) Let X be equipped with the discrete topology and let Y be any topological space.
Show that every mapping f : X → Y is continuous.
(b) Let Y be equipped with the indiscrete topology and let X be any topological
space. Show that every mapping f : X → Y is continuous.

K12146_C003.indd 67 2/12/15 3:43 PM

 68   ◾    Topological Spaces: Definition and Examples

2. Let X, Y, and Z be topological spaces and let f : X → Y and g : Y → Z be continuous

mappings. Show that the composition g  f : X → Z is also continuous.
3. Let f : X → Y be continuous and onto.
(a) Show that if D is a dense subset of X, then f(D) is a dense subset of Y.
(b) Show that if X is separable then so is Y.

4. Let f : X → Y be a bijection between two spaces. Show that f is open if and only if the
inverse mapping f −1 is continuous.

5. Show that if f : X → Y is a bijection, then f is open if and only if it is closed.

6. Show that f : X → Y is open if and only if there is a basis B for the space X such that
for every B ∈B, f (B) is open in Y.
7. (a) Let X be a space with topology τ, let p be a point not in X, and denote Y = X ∪ { p}.
Show that τ ∪ {Y } is a topology over Y.

(b) Find a mapping f : X → Y that is open but not closed.

8. Let f : X → Y be a continuous, one-to-one mapping, and let A be a subset of X.

Show the following.

(i) f (intA ) ⊃ int f ( A).

(ii) f ( A′ ) ⊂ ( f ( A))′ .

9. Let f : X → Y be a homeomorphism and let A be a subset of X. Show the following.

(i) If X is first countable, then so is Y.
(ii) f (int A ) = int f ( A ).
(iii) f ( A) = f ( A).
(iv) f ( A′ ) = ( f ( A))′ .
10. Let A and B be homeomorphic subsets of . Does it follow that A and B are
homeomorphic?
11. Let (X, d) be a metric space, and let x ∈ X . Show that the mapping f : X →  defined
by f ( y ) = d ( x , y ) is continuous.
12. Let X and Y be spaces, and let S be a subbasis for a space Y. Show that a mapping
f : X → Y is continuous if and only if f −1 (S ) is open for every S ∈S .

K12146_C003.indd 68 2/12/15 3:43 PM

 3.5 Continuous Mappings   ◾    69  

13. Let A be the subset of  2 consisting of the black points in Illustration 3.26.

   Illustration 3.26

(a) Let A be equipped with the discrete topology. Is A homeomorphic to the discrete
space over  2 ?
(b) Let Y be the topological space over A with open sets being the empty set and the
intersections of A with open disks centered at the center of A. Let Z be the topo-
logical space over  2 consisting of the empty set and all open disks centered at a
fixed point. (Why are Y and Z indeed topological spaces?) Prove that the spaces
Y and Z are not homeomorphic.

14. Show that every homeomorphism f : X → Y induces a bijection from the set of open
subsets of X onto the set of open subsets of Y.

15. Show that if f : X →  and g : X →  are continuous functions, then the function
h : X → , as defined below, is also continuous:
(a) h( x ) =| f ( x )|.
(b) h( x ) = min{ f ( x ), g ( x )}.
(c) h( x ) = max{ f ( x ), g ( x )}.
16. Show that f is continuous with domain X if and only if f ( A) ⊂ f ( A) for every subset
A of X.
17. Show that the fixed-point property (every continuous f : X → X has a fixed point) is a
topological property.
18. Let D be a dense set in a space X, let Y be Hausdorff, and let f : X → Y and g : X → Y
be continuous. Show that if f |D = g |D , then f = g . (Recall that f |D denotes the
restriction of f on D.)
19. (a) Show that if f : X → X is continuous and satisfies f  f = identity , then f is a
homeomorphism.
(b) Show that if f : n → n is a homeomorphism without fixed points, then for
every x ∈n there is an open neighborhood U of x such that f (U ) ∩ U = ∅.

K12146_C003.indd 69 2/12/15 3:43 PM

 70   ◾    Topological Spaces: Definition and Examples

20. Let A be a subspace of a metric space X, and let r : X → A be continuous and such that
f (a ) = a for every a ∈ A. Show that A is closed in X.
21. Consider the space  with the co-countable topology, and the space  with the co-
finite topology. Then  is not first countable (Exercise 23 (b) in 3.3).
(a) Show that the only convergent sequences in  are of type x1 , x 2 ,…, xn , x , x , x ,…
(i.e., sequences that eventually stabilize).
(b) Define f :  →  by f ( x ) =  x  (the largest integer ≤ x). Show that if ( x j ) is a
convergent sequence in , then ( f ( x j )) is a convergent sequence in .
(c) Show that f is not continuous.
(With this exercise we establish that the first countability condition in Proposition 2
is necessary.)
22. Let X be a set, let Y be a space, and let f : X → Y be a mapping. Show that
 −1 
 f (U ): U ⊂ Y  is the topology over X induced by f . Generalize: show that the
 open 
topology over X induced by fi : X → Yi , i ∈I , is the same as the topology over X
 
generated by the subbasis S = ∪  fi −1 (U i ) : U i ⊂ Yi  .
i ∈I  open 

23. Let X be a space, let Y be a set, and let fi : X → Y , i ∈I, be mappings. Show that
 −1 
U ⊂ Y : for every i ∈I , fi (U ) ⊂ X  is the topology over Y induced by { fi : i ∈I }.
 open 

K12146_C003.indd 70 2/12/15 3:43 PM

 Chapter 4

Subspaces, Quotient Spaces,

Manifolds, and CW-Complexes

I n this chapter we present a few important constructions of new spaces from given
spaces.

4.1 Subspaces
In the following proposition we describe a simple and natural construction through which
every space X passes its topological structure to each nonempty subset.
Proposition 1. Let X be a topological space and let A be a nonempty subset of X. The set
τ A = { A ∩ U : U is an open subset of X } is a topology over A.

Proof. The sets A and ∅ are in τ A since A = X ∩ A and ∅ = ∅ ∩ A. If follows from

i∈I
( ) that unions of elements of τ are also in τ . Similarly
∪( A ∩ U i ) = A ∩ ∪U i
i∈I
A A

∩( A ∩ U ) = A ∩ ( ∩U ) , and so finite intersections of elements of τ are in τ .

n n
i i A A
i =1 i =1

The topology τA is called the subspace topology over A; with that A becomes a subspace
of X.
X
U
A

Illustration 4.1 An open set in the subspace A

is an intersection A ∩ U where U is open in
X. Observe that A ∩ U need not be open in X.
Closed sets in subspaces are generated similarly:
71

K12146_C004.indd 71 2/6/15 5:18 PM

 72   ◾    Subspaces, Quotient Spaces, Manifolds, and CW-Complexes

Proposition 2. A subset F of a subspace A of a space X is closed in A if and only if there is a

closed subset G in X such that F = A ∩ G .

Proof.

⇒ Suppose F is closed in A. Then A \ F is open in A. So, there is an open set W in X such

that A \ F = A ∩ W . This, together with the assumption that F is a subset of A, implies that
F = A \ ( A ∩ W ). It is easy to see that A \ ( A ∩ W ) = A ∩ W c , and so F = A ∩ W c . Since W c is
closed, we have what we have looked for.
⇐ Suppose there is a closed subset G in X such that F = A ∩ G. Then G c is open in X
and so A ∩ G c is open in A. Since A \ ( A ∩ G c ) = A ∩ G , it follows that F is closed in A.

Given a space X, we will generally assume that subsets of X are equipped with the subspace
topology. For example, it will be understood, unless otherwise stated, that the subsets of 
are equipped with the subspace topology.
As indicated earlier, when we see n and unless otherwise stated, we will assume that we deal
with the Euclidean topology over n. We will carry this convention over to subspaces of n : unless
otherwise stated they are to be regarded as being equipped with the Euclidean subspace topology.

Example 1
Consider the subspace {( x , y ): y = 0} of  2 (the x-axis). Since intersections of open balls
with this line are open intervals, it follows that {( x , y ): y = 0} is homeomorphic to .
So, up to a homeomorphism,  is a subspace of  2. Similarly, if m ≤ n, then, up to a
homeomorphism, m is a subspace of n.

Illustration 4.2 A plane in the space (i.e.,  2 as a subspace of  3). ☐

A mapping f : X → Y between two spaces is an embedding if f is a homeomorphism from

X onto f(X) (with the latter considered as a subspace of Y). A special simple case of an
embedding is the inclusion in : A → X , where A is a subspace of X, and where in is defined
by in(a) = a for every a ∈ A.

K12146_C004.indd 72 2/6/15 5:18 PM

 4.1 Subspaces   ◾    73  

Another simple concept is the restriction f A of a mapping f : X → Y to a subspace A of X.

Proposition 3. If f : X → Y is continuous, then so is f A : A → Y , for every subspace A

of X.

Proof. Suppose U is open in Y. Since ( f )

−1
A (U ) = A ∩ f −1 (U ) and since A ∩ f −1 (U ) is open
in A, it follows that ( f )
−1
A (U ) is also open in A, and so f A is continuous.

A property is hereditary if it is always preserved in subspaces. Some important properties

are not hereditary (Example 3), while some are (Proposition 4).

Example 3
The tangent disk space over  2up is separable (see Example 8, Section 3.3 and
Exercise 4, Section 3.4). The ­x-axis L considered as a subspace of the tangent disk
space inherits the discrete topology; that is true since every one-element subset {x}
of L is an intersection of an open set of type B (a , r ) ∪ {x} and L. So L, being a discrete
uncountable space, is not separable. Thus, separability is not hereditary. ☐

Proposition 4. (a) Second countability is hereditary.

(b) First countability is hereditary.

Proof. (a) Suppose Y is second countable and X is a subspace of Y. Let B be a count-

able basis for Y. Consider the set B X = { X ∩ B : B ∈B }. It is countable and made of
open subsets of X. If U is any open subset of X, then U = X ∩ V for an open subset V
of Y. Since V is open in Y and B is a basis for Y, V = ∪ Bi for some Bi in B . But then
( )
i∈I
U = X ∩ V = X ∩ ∪ Bi = ∪ ( X ∩ Bi ), and so every open subset U of X is a union of some
i∈I i∈I
members of B X . Hence, B X is a countable basis for X.

(b) Exercise 12.

Exercises
1. Let Z be a space and let X ⊂ Y ⊂ Z . Show that the subspace X of Z is the same as the
subspace X of the subspace Y of Z. (This exercise shows that it does not matter if we
consider X as a subspace of Y or as a subspace of Z.)
2. Let X be a space and let A be an open subset of X. Show that every subset of A that is
open in the subspace A is also open in the space X.
 π π
3. Show that (0,1) ≅  − ,  ≅  (usual topology).
 2 2
4. Consider the subspace  of all rational numbers in . Show that  is not homeo-
morphic to . Show that the subspace of irrational numbers \ is also not
homeomorphic to .

K12146_C004.indd 73 2/6/15 5:18 PM

 74   ◾    Subspaces, Quotient Spaces, Manifolds, and CW-Complexes

5. Let Z be a subspace of Y. Show that if f : X → Z is continuous, then g : X → Y defined

by g ( x ) = f ( x ), for every x ∈ X , is also continuous. Show that this statement is not
always true if the word “continuous” is replaced by “open” (or “closed”).
6. Find a continuous bijection f from the interval (0,1] onto the unit circle
S1 = {( x , y ) ∈ : x 2 + y 2 = 1} such that f is not an open mapping.
7. Let X be a subspace of Y. Show that the inclusion mapping in : X → Y is
continuous.
8. Show that there is no embedding e : X → Y , where X is  with the discrete topology,
and where Y is the open unit disk B2 = {( x , y ) ∈ 2 : x 2 + y 2 < 1} considered as a
subspace of  2 .
9. Show that if B is a basis for X and A is a subspace of X, then the set {B ∩ A : B ∈B} is
a basis for the subspace A.

10. Denote X =  (usual topology) and Y =  with the countable complement topology.
Show that the subspace of Y obtained by deleting a point from the set Y is homeo-
morphic to Y, while the subspace of X obtained by deleting a point from X is not
homeomorphic to X.
11. Let X be the subset of  2 defined by the (uncountably many) black points in Illustration 4.3.

(a) Show that X (as a subspace of  2 ) is not homeomorphic

to  2 .
(b) Consider Y =  2 with the topology induced by the post office
metric (Section 2.1), where the designated point (the post
office) is a white point in  2 . Show that in that case X (as
a subspace of Y) is homeomorphic to  2 equipped with the
Illustration 4.3 discrete topology.

12. (a) Show that being first countable is hereditary.

(b) Show that being metrizable is hereditary.
13. Show that the property of being Hausdorff is hereditary.
14. Let X be a space and let p ∉ X. Find a topology over Y = X ∪ { p} such that X with
the given topology is the same as the subspace topology of X in Y, and such that no
nonempty open set in X is open in Y.
 quip X = {a1 , a2 , a3 } with the topology generated by the discrete metric dt
15. (a) E
(Section 2.1). Find and visualize an embedding f from the space X into  2 such
that dt ( x , y ) = d( f ( x ), f ( y )) for every x , y ∈ X (d stands for the Euclidean metric).
(b) Generalize (a) for X = {a1 , a2 , a3 ,…, an }, and with n−1 in place of  2 .

K12146_C004.indd 74 2/6/15 5:18 PM

 4.2 Quotient Spaces   ◾    75  

16. Let X be a subspace of Y and let A be a subset of X. Denote the closure of A in X by

clX ( A), and denote the closure of A in Y by clY ( A). Show that clX ( A) = clY ( A) ∩ X .
17. Let X = A ∪ B and suppose U ⊂ A ∪ B is open (closed) in both A and B. Show that U
is open (closed, respectively) in X.

4.2 Quotient Spaces

A brief historical note: Quotient spaces were introduced in 1925 by Robert L. Moore and,
independently, by Pavel Alexandroff (1896–1982).

Let X be a topological space and let ∼ be an equivalence relation on X, generating a parti-

tion of X and the quotient set X ~ (see Section 1.1). The topology over X ~ induced by the
quotient map q : X → X ~ (x [x ],[x ] is the equivalence class of x) is called the quotient
topology (or the factor topology). Explicitly, a subset U of X ~ is open in X ~ if and only if
the set ∪ [x ] is open in X (see Illustration 4.4).
[ x ]∈U

X3 X3
X/~ X
X2 X2
X1 X1
X4 X4
U
X5 X5
X8 X9 X6 X8 X9
X6
X7 X7
U*
X10 X10
X11 X11

Illustration 4.4 The subset U of X ~ in the first picture is made of the elements X5 , X 7 , and X 8 .
We declare it to be open in X ~ if and only if the set U∗ consisting of the union of the subsets
X5 , X 7 , and X 8 of X is open in the space X.

If the equivalence ~ corresponds to a partition P = { Xi : i ∈ I } of X, then a subset U of X ~

is open in X ~ if and only if ∪ Xi is open in X.
Xi ∈U

Closed sets in X ~ adhere to a similar criterion:

Proposition 1. A subset F of X ~ is closed in X ~ if and only if the set ∪ [x ] is closed in X.

[ x ]∈F

Proof. F is closed in X ~ , if and only if its complement F c in X ~ is open, if and

( )
c
only if ∪ c[x ] is open in X, if and only if ∪ c[x ] is closed in X. Now notice that
[ x ]∈F [ x ]∈F

( ∪ [x]) = ∪ [x].
c
[ x ]∈F c [ x ]∈F

K12146_C004.indd 75 2/6/15 5:18 PM

 76   ◾    Subspaces, Quotient Spaces, Manifolds, and CW-Complexes

Example 1
Partition the space  into the interval [a, b], and singletons disjoint from this interval.
The associated equivalence ~ is defined by x ~ y if and only if either x = y or x , y ∈[a, b].
Then  ~ is the space obtained from  by shrinking [a, b] to a point.

 /~
a b

Illustration 4.5 The interval [a, b] has shrunk to a point. The space  ~ looks like . It is, in
fact, homeomorphic to  (Exercise 10). ☐

Example 2
Suppose we use the open interval (a, b) in place of [a, b] in the previous example. So,
in this case x ~ y if and only if either x = y or x , y ∈(a, b). Then, we get something
different.

 /~
a b

Illustration 4.6 This time the open interval (a, b) has shrunk to a point. The space  ~ again
looks like . However, this time it is not (homeomorphic to)  (Exercise 11). ☐

Example 3
Consider the interval [0,2] as a subspace of  and define ~ on [0,2] by x ∼ y if and
only if x = y or x , y ∈{0,2}. Then [0,2] ~ is (homeomorphic to) a circle (as a subspace
of  2). We animate this in Illustration 4.7 below, where all of the spaces shown are
subspaces of  2 .

0 2 0~2

0
0 2 2

Illustration 4.7

The circle itself is obtained at the last moment of the four-frames animation; that
it is not homeomorphic to the starting space [0,2] will be easily justifiable later
in our theory.
Expanding slightly on the last example, suppose the equivalence ~ on [0,2] is
defined by 0 ~1 ~ 2, and x ~ y only if x = y otherwise. Then [0,2] ~ becomes a
bouquet of two circles as shown in Illustration 4.8.

K12146_C004.indd 76 2/6/15 5:18 PM

 4.2 Quotient Spaces   ◾    77  

0 2
1
0 2 1
0 1 2
0~1~2

Illustration 4.8 ☐

Example 4
What do we get if we glue all of the integers in ? That is, what is  ~ if ~ is defined
by x ~ y and x ≠ y, only if x and y are integers? The answer is in Illustration 4.9.

Illustration 4.9 The space  ~ in this example consists of infinitely many (countably many) circles
in a bouquet (with a single common point).

Despite the appearance, the space  ~ is not a subspace of  3. We will come back to
this problem later on. ☐

The definition of the quotient space implies immediately that the quotient mapping
q : X → X ~ is continuous. This mapping need not be open (Exercise 6), and it is almost
never one-to-one.

Proposition 2. Let X and Y be spaces and let f : X → Y be continuous. Consider the equiva-
lence relation ~ f on X defined by: x1 ~ f x 2 if and only if f ( x1 ) = f ( x 2 ). Then the following
is true.
(a) The mapping g : X ~ → Y defined by g ([x ]) = f ( x ) is well defined, one-to-one, and
f
continuous.
(b) If f is open or closed, then g is a homeomorphism X ~ → f ( X ).
f

Proof.
(a) If x1 ∈[x ] then x ~ x1 and so f ( x ) = f ( x1 ). Consequently, g([x]) does not depend on
the choice of the element in [x].
It is clear that g is one-to-one.

K12146_C004.indd 77 2/6/15 5:18 PM

 78   ◾    Subspaces, Quotient Spaces, Manifolds, and CW-Complexes

In order to check that the mapping g is continuous, first notice that f = g  q, where
q : X → X ~ is the quotient map. Since f is continuous, if U is an open subset of Y then
f
f −1 (U ) is open in X. Since q −1 ( g −1 (U )) = f −1 (U ), it follows that q −1 ( g −1 (U )) is open in
X, which in turn means that g −1 (U ) is open in X ~ .
f
(b) We have shown in (a) that g is continuous and one-to-one.
Assume that f is open. According to the definition of the quotient topology, any open
set U in X ~ is of the type V ~ , where V is open in X. The definition of g implies that
f f
g (U ) = f (V ), and so, since f is open, this set is open in Y. Consequently, g (U ) ∩ f ( X )
( )
is open in f ( X ) = g X ~ . So g is open, and thereby it is a homeomorphism onto f (X).
f
Assume that f is closed. Choose a closed subset F of X ~ . It follows from Proposition 2
f
that F is of the type G ~ , for some closed subset G of X. Since g ( F ) = f (G ), and since by
f
assumption f (G) is closed in Y, it follows that g (F) is also closed in Y, and so g ( F ) ∩ f ( X )
( )
is closed in f ( X ) = g X ~ . So, g is closed. On the other hand if g is closed and onto
f
then it must also be open and onto (since we have observed in Section 3.5 that a bijection
is open if and only if it is closed). So, g is a homeomorphism.

The equivalence relation ~ f defined in Proposition 2 is sometimes called equivalence modulo f.

The second part of Proposition 2 tells us that if f : X → Y is onto and open (or closed), then we
can get (a homeomorphic copy of) the space Y by identifying those points in X that have the same
images via f. In order to illustrate this we will again consider the quotient space from Example 1.

Example 5 0 if x ∈[0,1]

Let f :  →  be defined by f ( x ) =  x if x < 0 . Notice that f is onto.
 x − 1 if x > 1

It is also easy to see that f is closed. It follows from Proposition 2 that  ~ is homeo-
f
morphic to . As we saw in Example 1, the space  ~ is obtained by shrinking the
f
closed interval [0,1] in  to a point. ☐

Suppose now that X is a space, Y is just a set, and f : X → Y is a mapping. The space X ~
f
is called the quotient space determined by f (Illustration 4.10).

f
Y
X

Illustration 4.10 The quotient space X ~ is determined by the mapping f.

f

K12146_C004.indd 78 2/6/15 5:18 PM

 4.2 Quotient Spaces   ◾    79  

Recall that specifying a subset U of Y to be open if and only if f −1 (U ) is open in X, yields

the topology over Y induced by f (Exercise 12, Section 3.1). The mapping f : X → Y then
becomes continuous. Proposition 2 implies that the mapping g : X ~ → Y defined by
f
g ([x ]) = f ( x ) is well defined and continuous. In this case, as we see further below, it is
automatically a homeomorphism onto f(X). We show that g is open; the rest is obvious.
Let V be an open subset of X ~ . That means that ∪ [v] is open in X. We want to show
f [v ] ∈V
−1  
that g(V) is open in Y, or, equivalently, that f ( g (V )) is open in X. Now, g (V ) = f  ∪ [v] , so
 [v ] ∈V 
−1    
that f ( g (V )) = f  f ∪ [v]  . It follows from the definition of [v] that f ([v]) = { f (v )},
−1
  [v ]∈V  
  
and this in turn implies that f −1  f  ∪ [v]  = ∪ [v]. So, g(V) is indeed open. In light
  [v ] ∈V   [v ]∈V
of Proposition 2, we proved the following:

Proposition 3. Given any space X and any set mapping f : X → Y such that f is onto,
the quotient space X ~ is homeomorphic (via the mapping g) to the space Y induced
f
by f.

Exercises
1. Give a sufficient and necessary criterion for a quotient map to be a homeomorphism.
2. Suppose X =  2. Describe (visualize) the space X ~ if ~ is the smallest equivalence
relation satisfying the following conditions.

(a) ( x , y ) ~ ( x ′ , y ′ ) if and only if x = x ′ − 1.

(b) ( x , y ) ~ ( x ′ , y ′ ) if and only if x = x ′ − 1 and y = y ′ − 1.
(c) ( x , y ) ~ ( x ′ , y ′ ) if and only if x = x ′ − 1 or y = y ′ − 1.

3. Find an equivalence relation ~ on  2 such that

2
(a)  ~ is (homeomorphic to) a circle (equipped with the usual topology).
2
(b)  ~ is (homeomorphic to) a sphere (equipped with the usual topology).

{
4. Denote, as usual, S 2 = ( x , y , z ) ∈ 3 : x 2 + y 2 + z 2 = 1 } (the unit sphere), and
{ }
D 3 = ( x , y , z ) ∈ 3 : x 2 + y 2 + z 2 ≤ 1 (the closed unit ball).
2
(a) Define f : S → S by f ( x , y , z ) = ( x , y , − z ). Describe the space S ~ .
2 2
f
3
(b) Define g : D → D to be f over S and the identity elsewhere. Describe D ~ .
3 3 2
g

K12146_C004.indd 79 2/6/15 5:19 PM

 80   ◾    Subspaces, Quotient Spaces, Manifolds, and CW-Complexes

5. Show that the quotient mapping q : X → X ~ is closed if and only if for every closed
A ⊂ X, the set ∪ [x ] is closed in X. Show that the same is true if “closed” is
[ x ]∩ A≠∅
replaced by “open.”
6. Show that a quotient mapping q : X → X ~ need not be open.
7. Let ~ be an equivalence relation on a space X, and let f : X → Y be a continu-
ous mapping such that f ( x1 ) = f ( x 2 ) whenever x1 ~ x 2 . Show that the mapping
g : X ~ → Y defined by g ([x ]) = f ( x ) is well defined and continuous.

8. (a) Define an equivalence relation ~ on the circle S1 = {( x , y ) : x 2 + y 2 = 1} by

1
( x , y ) ~ (− x , − y ). Show through pictures that S ~ is homeomorphic to S1.
(b) Define an equivalence relation ~ on the sphere S 2 = {( x , y , z ) : x 2 + y 2 + z 2 = 1} by
2
( x , y , z ) ~ (− x , − y , − z ). Show through pictures that S ~ is homeomorphic to S 2.
(c) Define an equivalence relation ~ on the disk D 2 = {( x , y ) : x 2 + y 2 ≤ 1} by
( x , y ) ~ (− x , − y ) if x 2 + y 2 = 1 and (x, y) is only equivalent to itself in other cases.
2
Show through pictures that D ~ is homeomorphic to S 2 .
2
9. Consider the upper half-plane up = {( x , y ) : y ≥ 0} equipped with the tangent-disk
2
topology. Partition up into equivalence classes, one of which consists of all ele-
ments of the x-axis, while the others are singletons; denote the resulting equiva-
lence relation by ~.
∞
(a) Is the sequence  (−1)n , 1  convergent in up
2
? Justify your answer.
 
n n=1
2
(b) Find a countable local basis of [(0,0)] ∈ up ~ . (Remark: [(0,0)] is the equivalence
class of (0,0).)
10. Define an equivalence relation ~ on  as follows: x ~ y if and only if either x = y or
x , y ∈[0,1]. Show directly that  ~ is homeomorphic to .
11. Define an equivalence relation ~ on  as follows: x ~ y if and only if either x = y
or x , y ∈[0,1). Show that  ~ is not homeomorphic to  . Show that the same
con­clusion holds if [0,1) is replaced by (0,1] or by (0,1).
12. Define ~ on  as follows: x ~ y and x ≠ y if and only if one of x, y is in (0,1] and the
other is larger by exactly 1. Show that  ~ is not homeomorphic to  .
13. Let ~ be the smallest relation on  defined by x ~ y if x and y are integers. Then the
quotient space  ~ is a wedge of countably many circles (countably many circles with
one common point).
(a) Show that  ~ is not embeddable in  2 .
(b) Show that  ~ is not first countable (hence, neither is it second countable).

K12146_C004.indd 80 2/6/15 5:19 PM

 4.3 The Gluing Lemma, Topological Sums, and Some Special Quotient Spaces   ◾    81  

14. Identify all rational numbers in  (x ~ y and x ≠ y only if x and y are rational num-
bers). Describe the open subsets in  ~.
15. Define ~ on  as follows: x ~ y if x − y ∈ . Confirm that this is an equivalence rela-
tion and describe the open sets in the quotient space.
16. Let X be a space, and let G be a subgroup of the group of all homeomorphisms X → X
(where the operation in the group is the composition of homeomorphisms). Define a
relation ~G on X as follows: x1 ~G x 2 if there is f ∈G such that f ( x1 ) = x 2 .
(a) Show that ~G is an equivalence relation.

(b) Denote the quotient mapping by q : X → X ~ . Show that q is open.

G

4.3 The Gluing Lemma, Topological Sums,

and Some Special Quotient Spaces

A brief historical note: The notion of sums of general topological spaces seems to appear
for the first time in the first edition (1940) of the book General Topology by Nicolas Bourbaki
(a collective pseudonym for a group of mathematicians).

The following useful lemma will be encountered many times.

Lemma 1. (The Gluing Lemma) Let A and B be two closed subsets of a space X, let
X = A ∪ B, and let f : A → Y , g : B → Y be continuous mappings such that f ( x ) = g ( x ) for
 f ( x ) if x ∈A
every x ∈ A ∩ B. Then the mapping h( x ) =  is a continuous mapping
 g ( x ) if x ∈B
X →Y. 

Proof. Let V be a closed subset of Y. We show that h is continuous by proving that h −1 (V ) is

closed (Proposition 4, Section 3.5).
It is easy to see that h −1 (V ) = f −1 (V ) ∪ g −1 (V ). Since f is continuous, f −1 (V ) is a closed
subset of A. Proposition 2, Section 4.1 implies that f −1 (V ) = F ∩ A for some closed subset
F of X. Since A is closed in X it follows that f −1 (V ) = F ∩ A is closed in X. By symmetry,
−1 −1 −1
g −1 (V ) is also closed in X. Hence, h (V ) = f (V ) ∪ g (V ) is closed in X , too.

Lemma 1 holds if we replace “closed” by “open” (Exercise 1). The mapping h : A ∪ B → Y

defined in the last proposition is called the union of the mappings f and g.

We now turn to other methods of constructing new spaces from old spaces.

K12146_C004.indd 81 2/6/15 5:19 PM

 82   ◾    Subspaces, Quotient Spaces, Manifolds, and CW-Complexes

Let Xi , i ∈I be a collection of pairwise disjoint (nonempty) spaces. The family B of sub-

sets of ∪ Xi that are open in some Xi is a basis for a topology over ∪ Xi. The space over the set
i∈I i∈I
∪ Xi generated by B is called the discrete sum (or the topological sum) of the spaces Xi , i ∈I,
i∈I
and is usually denoted by ⊕ Xi . For the discrete sum of two spaces X and Y we will use X ⊕ Y .
i∈I

Proposition 2. A subset U of ∪ Xi is open in ⊕ Xi if and only if U ∩ Xi is open in Xi for

i∈I i∈I
every i ∈I .

Proof. Exercise 7.

It follows from the gluing lemma that if fi : Xi → Y is continuous for every i ∈I, then so is
the mapping f : ⊕ Xi → Y defined by f ( x ) = fi ( x ) if x ∈ Xi .
i∈I

Example 1
Let X1 = {a, b} be equipped with the discrete topology, and let X 2 = {c , d} be equipped
with the indiscrete topology. The open subsets of the discrete sum X1 ⊕ X 2 are of type
U1 ∪ U 2 , where U1 is open in X1 and U 2 is open in X 2. We list them all: {∅,{a},{b},{a, b},
{c , d},{a, c , d},{b, c , d},{a, b, c , d}}. ☐

The discrete sum topology over ∪ Xi is induced by the inclusions Xi → ⊕ Xi , i.e., it is the
i∈I i∈I
finest topology making all the inclusions Xi → ⊕ Xi continuous (Exercise 8).
i ∈I

Example 2
Take pairwise disjoint copies Xi , i ∈I, of the circle S1 = {( x , y ) : x 2 + y 2 = 1} (viewed as a
subspace of  2). The space ⊕ Xi can be viewed as I -many separate, unrelated circles, as
i ∈I
shown in Illustration 4.11. (This space is homeomorphic to soon-to-be-defined product
space I × S1, where I is equipped with the discrete topology; Exercise 3, Section 5.1.) ☐

Illustration 4.11

K12146_C004.indd 82 2/6/15 5:19 PM

 4.3 The Gluing Lemma, Topological Sums, and Some Special Quotient Spaces   ◾    83  

The more general case, when the spaces in the collection A = { Xi : i ∈ I } are not neces-
sarily disjoint, can be approached through the criterion given in Proposition 2: the set
T ( A ) = {U ⊂ ∪ Xi : U ∩ Xi is open in Xi for every i ∈I } is a topology over ∪ Xi . It is tradi-
i∈I i∈I
tionally called the weak topology over ∪ Xi determined by A . As was the case when the
i∈I
sets Xi were pairwise disjoint, it is again true that this topology is the finest topology
making the inclusions Xi → ∪ Xi continuous.
i∈I

Example 3
Let X1 = {a, b} be with the discrete topology and let X 2 = {b, c} be with the indis-
crete topology. The weak topology over {a, b, c} determined by A = { X1 , X 2 } is
{∅,{a},{b, c},{a, b, c}}. ☐

Let X and Y be two spaces, and suppose X ∩ Y = A. Choose disjoint copies X ′ and Y ′ of X
and Y respectively, and for each a ∈ A denote by a1 and a2 the corresponding elements in
X ′ and Y ′, respectively. Partition X ′ ∪ Y ′ into {a1 , a2 } for every a ∈ A, and into singletons
for other elements of X ′ ∪ Y ′ . This defines an equivalence ~ over X ′ ∪ Y ′ . It is straight-
forward to show that X ′ ∪ Y ′ ~ and the weak topology over X ∪ Y determined by { X ,Y }
define homeomorphic spaces. (Exercise: prove it!) We now pay more attention to spaces
obtained in a similar way as X ′ ∪ Y ′ ~ .
Let A be a nonempty subset of a space X, let Y be any space disjoint from X, and let
f : A → Y be any mapping. Denote B = f ( A), and partition X ∪ Y into classes of type
{ y} ∪ f −1 ({ y}) when y ∈B, and singletons otherwise. This defines an equivalence relation
~ over X ∪ Y . The quotient space X ⊕ Y ~ is called the space obtained from X and Y
by identifying (or by gluing) the subspaces A and B along f; we will denote it by X ∪ f Y .
When the rest is clear from the context, we will simply say that the new space is obtained
by identifying (or gluing) A and B.
In a special case when both X and Y are disjoint subspaces of a fixed space Z, and
when A coincides with X, then the above partition of X ∪ Y , extended by the singletons
{{z } : z ∈Z \ ( X ∪ Y )} , determines an equivalence relation ~ on Z. In this case the space Z ~
will again be referred to as being obtained from Z by identifying (gluing) the subspaces
A and B (along f). We will denote it by Z f .
Summarizing the somewhat superficial difference between the above two definitions:
X ∪ f Y is obtained by identifying subspaces of two disjoint spaces X and Y, while in the
case of Z f we identify (glue) two disjoint subspaces of a single space Z.

Example 4
The space X is a truncated ball {( x , y , z ) : x 2 + y 2 + z 2 ≤ 1, z ≤ 3 4} while Y is the
truncated part {( x , y , z ) : x 2 + y 2 + (z − 1)2 ≤ 1, z ≥ 7 4}. The mapping f sends
every point in the disk {( x , y , z ) : x 2 + y 2 + z 2 ≤ 1, z = 3 4} to the point in the disk
{( x , y , z ) : x 2 + y 2 + (z − 1)2 ≤ 1, z = 7 4} with the same first two coordinates. The
space X ∪ f Y is a whole ball. Illustrations 4.12 and 4.13 repeat the whole story.

K12146_C004.indd 83 2/6/15 5:19 PM

 84   ◾    Subspaces, Quotient Spaces, Manifolds, and CW-Complexes

Illustration 4.12 The space X is the lower truncated ball, Y is the upper truncated ball. The
map f is indicated by the arrow.

Illustration 4.13 After gluing the two disks we get the space X ∪ f Y : a whole ball. ☐

Example 5
Denote X = [0,1] × [1,3], A = [0,1] × [1,2], Y = [0,2] × {0} (both subspaces of  2 ; see
Illustration 4.14), and let f : A → Y be the projection f ( x , y ) = ( x ,0 ). A copy of the
space X ∪ f Y is shown in Illustration 4.15.
X
3

A
1
f

1 2
Y

Illustration 4.14

K12146_C004.indd 84 2/6/15 5:19 PM

 4.3 The Gluing Lemma, Topological Sums, and Some Special Quotient Spaces   ◾    85  

1
X ∪f Y

Illustration 4.15 The space A gets flattened into a line segment within X ∪ f Y . ☐

Example 6
Let X =[0,1] and Y =[2,3]. If A = {0,1} ⊂ X and g : A → Y is the embedding defined by
g (0) = 2, g (1) = 3, then X ∪ g Y is homeomorphic to a circle (Illustration 4.16).

0 [0, 1]
[0, 1] 1
0 1

2 3 2 3
[2, 3] [2, 3]

Illustration 4.16 Illustration 4.17

2 if x < 1
If A = X and f is defined by f ( x ) =  (Illustration 4.17), then the topological
3 if x = 1
structure of X ∪ f Y is different from any subspace of n (Exercise 15). ☐

inclusion quotient map inclusion quotient map

Neither of the compositions X → X ⊕ Y → X ∪ f Y and Y → X ⊕Y →
X ∪ f Y in the last example (Illustration 4.17) is an embedding. In that sense, the structure of
the space X is very easy to destroy: if f is not one-to-one, then the first of the two compositions
is certainly not an embedding. However, the space Y is much more resilient, and, as we see
inclusion quotient map
in Proposition 3, the composition Y → X ⊕Y → X ∪ f Y could be an embedding
even when f is not one-to-one.

Proposition 3. Let X and Y be two spaces and let A be a closed subspace of X. If f : A → Y

inclusion quotient map
is continuous, then the composition Y → X ⊕Y → X ∪ f Y is an embedding.

Remark. Recall that X ∪ f Y is a quotient space of X ⊕ Y , so that we may talk about

the quotient map X ⊕ Y → X ∪ f Y . Also recall that the inclusion Y → X ⊕ Y (sending
each y ∈Y to y ∈ X ⊕ Y ) is an embedding, and consequently we can assume that Y is a

K12146_C004.indd 85 2/6/15 5:19 PM

 86   ◾    Subspaces, Quotient Spaces, Manifolds, and CW-Complexes

subspace of X ⊕ Y . Thereby it suffices to show that the restriction of X ⊕ Y → X ∪ f Y to

Y is an embedding.

Proof. We want to show that the mapping g : Y → X ∪ f Y defined by g ( y ) = [ y] is an

embedding.

The equivalence class [y] of y ∈Y is { y} ∪ f −1 ( y ). If y1 ≠ y 2 , then ({ y1 } ∪ f −1 ( y1 ))

∩ ({ y 2 } ∪ f −1 ( y 2 )) = ∅ , and so [ y1 ] ≠ [ y 2 ]. Hence, g is one-to-one.

The mapping g is continuous since it is a restriction of a continuous map.

We complete the proof by showing that g : Y → g (Y ) is a closed mapping. Let B be a

closed subset of Y. Then g ( B) = {[ y] ∈ X ∪ f Y : y ∈B} = {{ y} ∪ f −1 ( y ) : y ∈B}. It follows
from Proposition 1 in 4.2 that {{ y} ∪ f −1 ( y ) : y ∈B} is closed in X ∪ f Y if and only if
∪ ({ y} ∪ f −1 ( y )) is closed in X ⊕ Y . Notice that ∪ ({ y} ∪ f −1 ( y )) = B ∪ f −1 ( B) and that
y ∈B y ∈B
−1
f ( B) is closed in A since f is continuous. Since A is equipped with the subspace topology of
X, we have that f −1 ( B) = C ∩ A for some closed subset C of X. So, B ∪ f −1 ( B) = B ∪ (C ∩ A),
and this is closed in X ⊕ Y since B is closed in Y, and since C ∩ A is closed in X (we have
just used our assumption that A is closed in X). So ∪ ({ y} ∪ f −1 ( y )) is closed in X ⊕ Y , and
y ∈B
so g(B) is closed in X ∪ f Y . Consequently, g(B) is closed in g(Y).

Under the assumptions of Proposition 3, in addition to Y “surviving” within X ∪ f Y ,

the subspace X \ A of X also remains intact within X ∪ f Y (Exercise 16).

Because of this proposition, the notion of attaching a space to a space via a mapping is
sometimes defined under the additional assumptions that f is continuous and A is closed
in X. Indeed, as will be evident, most of the time we will be attaching spaces under such
conditions.

Example 7: Sphere Obtained by Identifying Subspaces of a Disk

Start with a closed disk D, and subdivide its boundary into two closed arcs as
in the first frame of Illustration 4.18 (where both arcs are labeled by a). Then
identify these two segments via the mapping f that projects the points on the
top segment vertically down to the points of the bottom segment. We get a space
denoted D f .
The sequence of frames shown in Illustration 4.18 animates the construction of
D f . As we see, the resulting quotient space D f is a sphere.

Alternatively, we can introduce a new closed interval Y disjoint from D, directly

below the horizontal diagonal of D, then project the boundary S1 of D vertically onto Y.
Denoting this projection by g, the resulting space is D ∪ g Y . It is clear that this space is
homeomorphic to D f (Illustration 4.19).

K12146_C004.indd 86 2/6/15 5:19 PM

 4.3 The Gluing Lemma, Topological Sums, and Some Special Quotient Spaces   ◾    87  

a
a
a

a a
a

a
a
a

Illustration 4.18 From top left to bottom right: Start with a disk and identify the boundary
points according to the arrows (i.e., according to the mapping f ).

Illustration 4.19 Y = [−1, −1] × {−2} is a subspace of  2 , and the mapping f : S1 → Y is the pro-
jection defined by f ( x , y ) = ( x , −2). ☐

Exercises
1. (This exercise generalizes the gluing lemma.) Let X and Y be spaces, and let f : X → Y
be a mapping.
(a) Let X = ∪U i , where each U i is open in X, and suppose for every i ∈I, the restric-
i∈I
tion f Ui is continuous. Prove that f is continuous.
n
(b) Let X = ∪ Fi , where each Fi is closed in X, and suppose for every i ∈{1,2,…, n}, the
i =1
restriction f Fi is continuous. Prove that f is continuous.
(c) Does (b) remain true if we replace {Fi : i = 1,2,…, n} by an infinite collection of
closed subsets of X?

K12146_C004.indd 87 2/6/15 5:19 PM

 88   ◾    Subspaces, Quotient Spaces, Manifolds, and CW-Complexes

n
2. Let { Ai : i = 1,2,3,…, n} be a family of closed subsets of X such that ∪ Ai = X and such
i =1
that Ai ∩ A j = ∅ for j ≠ i + 1, i = 1, 2, 3, …, n − 1, and let { fi : Ai → Y : i = 1, 2, 3, …, n} be
a set of continuous mappings such that fi and fi+1 agree on Ai ∩ Ai+1 , i = 1, 2, 3, …, n − 1.
Prove that the mapping g : X → Y defined by g ( x ) = fi ( x ) if x ∈ Ai is well defined and
continuous.

3. (a) Let { Ai : i = 1,2,3,…} be a family of closed subsets of a metric space X such

 ′ ∞
that A j ∩  ∪ Ai  = ∅ , j = 1, 2, 3, …, such that ∪ Ai = X , and such that
 i− j ≥ 2  i =1

Ai ∩ A j = ∅ for | i − j| ≥ 2 . Let { fi : Ai → Y : i = 1,2,3,…} be a set of continuous

mappings such that fi and fi+1 agree on Ai ∩ Ai +1 , i = 1, 2, 3, … . Then the map-

ping g : X → Y defined by g ( x ) = fi ( x ) if x ∈ Ai is well defined and continuous.
[Hint: Show that inverse images of closed sets are closed.]

 ′
(b) Show that the condition A j ∩  ∪ Ai  = ∅ , j = 1, 2, 3, … is necessary.
 i− j ≥ 2 

4. Show that if we omit the requirement that both A and B must be closed or open
in X in the statement of the gluing lemma, then the union of f and g need not be
continuous.

5. Suppose X1 = {0, a} (a is not a number) is equipped with the indiscrete topology, and
suppose X 2 = . Describe the open subsets of the discrete sum X1 ⊕ X 2.
6. Show that if fi : Xi → Y is an embedding for every i ∈I, such that fi ( Xi ) ∩ f j ( X j ) = ∅
for every i ≠ j, then so is the mapping f : ⊕ Xi → Y defined by f ( x ) = fi ( x ) if x ∈ Xi .
i∈I
7. Prove Proposition 2.
8. Show that the discrete sum topology over the disjoint union ∪ Xi of the spaces Xi ,
i∈I
i ∈I, is the finest topology such that all of the inclusions ini : Xi → ∪ Xi (defined by
i∈I
ini ( x ) = x ) are continuous. Show that the inclusions ini : Xi → ⊕ Xi are embeddings.
i∈I

9. Let A = { Xi : i ∈I } be a collection of (not necessarily disjoint) spaces and consider the

{ }
family S = U ⊂ ∪ Xi : U ⊂ Xi for some i ∈I . Show that S is a subbasis for a topology
i∈I open

τ over ∪ Xi , and show that τ is finer than the weak topology over ∪ Xi determined by
i∈I i∈I
A. Show that the inclusions ini : Xi → ∪ Xi need not be continuous with respect to τ.
i∈I

10. Find two disjoint subsets A and B of the unit circle S1 and a bijection f : A → B
such that S1f is homeomorphic to S1. Do the same for the unit sphere S 2 in place
of S1.

K12146_C004.indd 88 2/6/15 5:20 PM

 4.3 The Gluing Lemma, Topological Sums, and Some Special Quotient Spaces   ◾    89  

11. Let X be the closed unit ball {( x , y , z ) ∈ 3 : x 2 + y 2 + z 2 ≤ 1}. Denote A = {( x , y , z ) ∈ X :

x 2 + y 2 + z 2 = 1, z > 0} and B = {( x , y , z ) ∈ X : x 2 + y 2 + z 2 = 1, z < 0}. Define f : A → B
by f ( x , y , z ) = ( x , y , − z ). Show that the space X f , obtained by identifying A and B along
f, is homeomorphic to the three-sphere S 3 = {( x , y , z , u) ∈ 4 : x 2 + y 2 + z 2 + u 2 = 1}.
(Compare with Example 7.)
12. Let X and Y be two disjoint copies of the solid torus S1 × D 2 (as a product of two
metric spaces), and denote by A and B the subspaces of X and Y corresponding to the
subspace S1 × S1 of S1 × D 2. Define a mapping f : A → B by f ( x , y ) = ( y , x ) for every
( x , y ) ∈ A ≅ S1 × S1. Show visually and algebraically that X ∪ f Y is homeomorphic to
the three-sphere S 3 = {( x , y , z , u) ∈  4 : x 2 + y 2 + z 2 + u 2 = 1}.

13. Let f : A1 → A2 and g : B1 → B2 be homeomorphisms. Let C be a subspace of A1, let

D be a subspace of B1 and let h : C → D be a homeomorphism. Show that A1 ∪h B1 is
homeomorphic to the space A2 ∪ g h f −1 B obtained by identifying A2 and B2
( D ) ( f (C ) ) 2
(
along ( g D )  h  f f (C ) .
−1
)
14. Let X ∪ f Y be the space obtained by identifying along a mapping f : A → Y , A ⊂ X .
For z ∈ X ∪ Y denote its equivalence class in X ∪ f Y by [z].
(a) Show that if f is an embedding, then the mapping k : X → X ∪ f Y defined by
k( x ) = [x ] is also an embedding.
(b) Show that if A is a closed subset of X and f is a closed mapping then k(X),
k defined as in part (a), is a closed subset of X ∪ f Y .
15. Let A = X =[0,1], let Y =[2, 3] and let f be the mapping A → Y defined by
2 if x < 1
f (x ) =  . Show that the space X ∪ f Y is not embeddable in any n.
 3 if x = 1
16. Let X and Y be two spaces, let A be a closed subspace of X, and let f : A → Y be con-
tinuous. Show that the restriction of the quotient map X ⊕ Y → X ∪ f Y to X \ A is an
embedding.

17. Let X and Y be two spaces, let A be an open subspace of X, and let f : A → Y be con-
tinuous. Show that the restriction of the quotient map X ⊕ Y → X ∪ f Y to Y is an
embedding.

18 (a) Find two disjoint subspaces A and B of , find (nonempty) A1 ⊂ A, B1 ⊂ B, and

find a homeomorphism f : A1 → B1, such that the space ( A ∪ B ) f is homeomor-
phic to .

(b) Consider f :[0,1) → [1, 2) defined by f ( x ) = x + 1. Show that the space  f is

not homeomorphic to .

K12146_C004.indd 89 2/6/15 5:20 PM

 90   ◾    Subspaces, Quotient Spaces, Manifolds, and CW-Complexes

4.4 Manifolds and CW-Complexes

A brief historical note: Low dimensional manifolds (line, plane, circle, sphere, etc.) have been
mathematical folklore for a long time. The study of higher dimensional manifolds was initiated
by Henri Poincaré (1854–1912). The Invariance of Domain theorem (Theorem 1) and Corollary
2 were first proved in 1912 by Luitzen Egbertus Jan Brouwer (1881–1966), whom we will
encounter many times. CW-complexes were introduced in 1940 by John Henry Constantine
Whitehead (1904–1960).

In this section we introduce manifolds and CW-complexes. Further analysis will be pro-
vided later on.

A space X is an n-manifold (n ∈+ ) if the following conditions are satisfied.

(i) X is Hausdorff.
(ii) X is second countable.

(iii) For every x ∈ X there exists an open neighborhood U of x such that U is homeomor-
phic to the open n-ball Bn = {( x1 , x 2 ,…, xn ) ∈n : x12 + x 22 +  + xn2 < 1}.

The positive integer n in this definition is called the dimension of the manifold X. It is well
defined since n ≅ m only if n = m. This is a consequence (Corollary 2) of the following
theorem that we will not prove.

Theorem 1. (Invariance of Domain) If U is an open subspace of n and if f : U → n is an

embedding, then f(U) is open in n.

In the old terminology, we assert in Theorem 1 that domains (as open sets in n used to be
called) are sent to domains via embeddings, explaining the origin of the name “Invariance
of Domain.” It is one of these innocent-looking theorems that appear deceptively easy, yet
no elementary proof is known.

Corollary 2. n ≅ m if and only if n = m.

Proof. The only nontrivial part of the claim is ⇒. Suppose n < m, and suppose there is a
homeomorphism g : n → m. Take a nonempty open subset U of m . Then g −1 (U ) is
a nonempty open subset of n. Fix an+1 , an+2 ,…, am ∈ and let c : n → m be defined by
c( x1 , x 2 ,…, xn ) = ( x1 , x 2 ,…, xn , an+1 ,…, am ). Since c is an embedding and g is a homeomor­phism,
the composition c  g −1 : m → m is also an embedding. Theorem 1 implies that c  g −1 (U ) is
open in m . This means that c sends the nonempty open subset V = g −1 (U ) of n onto a non-
empty open subset of m . But this is not possible since c(V ) ⊂ n × {an+1 } × {an+2 } × × {am },
and no subset of n × {an +1 } × {an + 2 } ×  × {am } has any interior points.

K12146_C004.indd 90 2/6/15 5:20 PM

 4.4 Manifolds and CW-Complexes   ◾    91  

The condition (iii) in the definition of n-manifolds is crucial. The other two condi-
tions are included in order to avoid pathological manifolds; see Exercise 1. In the
examples covered in this section we will focus only on part (iii): the first two condi-
tions will be evident in all cases and so we will not pay attention to them. In Chapter 14,
where manifolds will be covered in more detail, we will make full use of all of these
conditions.

Example 1
The Euclidean space n is an n-manifold. This is trivial. ☐

Example 2
The circle S1 = {( x , y ) ∈ 2 : x 2 + y 2 = 1} is a 1-manifold. We will see later that, in a
sense, S1 is the only 1-manifold other than 1. ☐

The family of 2-manifolds is much richer. In the next example we show only a few of
infinitely many fundamentally different 2-manifolds.

Example 3
In Illustration 4.20, we show three 2-manifolds.

Illustration 4.20 From left to right: a simple sphere, two unbounded cylindrical surfaces capped
with half-spheres and joined with three short tubes, and a “dog-bone” (surface only). ☐

Example 4: Torus
Start with a closed filled rectangle Z (as a subspace of  2 ) bounded by two verti-
cal and two horizontal edges. Then identify the two vertical edges (labeled by a in
Illustration 4.21) along the function f that projects every point on the left-hand side
edge horizontally to a point on the right-hand side edge. The resulting space Z f is
a cylindrical surface; the two horizontal edges in the starting square became two
circles, both denoted by b (Illustration 4.21).

K12146_C004.indd 91 2/6/15 5:20 PM

 92   ◾    Subspaces, Quotient Spaces, Manifolds, and CW-Complexes

b
a a

a
b b

Illustration 4.21 We get a cylinder by identifying the edges labeled a along the horizontal
projection.

Proceeding further, we identify these two circles along the function that projects each
point on the top circle vertically onto the point on the bottom circle. We animate the
process in Illustration 4.22.

a
b a

Illustration 4.22 We identify the bounding circles along the vertical projection.

The resulting quotient space is a torus. The last frame of the visualization
shown in Illustration 4.21 represents the torus faithfully in the sense that if we
consider it as a subspace of  3, it is homeomorphic to the torus as the quotient
space of the rectangle Z.
In Illustration 4.23 we summarize the above procedure: we get a torus from a
closed rectangle by identifying two pairs of edges as indicated. Identifying edges
in filled closed polygons will be our main route for introducing 2-manifolds
later on.

b
a a
a

Illustration 4.23 A torus as a quotient space of a rectangle.

The torus is an important 2-manifold, and we will encounter it again. ☐

K12146_C004.indd 92 2/6/15 5:20 PM

 4.4 Manifolds and CW-Complexes   ◾    93  

Example 5: Torus via a Bouquet of Two Circles

Observe that the four edges of the starting rectangle Z in Example 4 (Illustration 4.23)
became two circles in the torus, one labeled by a, the other by b, sharing a point v (as
in Illustration 4.24).
b v
a

Illustration 4.24

Denote the subspace of the rectangle Z consisting of the four boundary edges by A,
and denote the space shown in Illustration 4.24 by B. Define f : A → B by winding
each of the four edges once around the circle in B with the same label, starting at v,
and in the direction of matching arrows. Since ~ f yields precisely the same partition
of Z as in Example 4, the space Z ∪ f B is again a torus. ☐

Example 6: Klein Bottle

The Klein bottle is the quotient space of the square shown in Illustration 4.25 obtained
by identifying the edges with the same labels so that the arrows match.
b

a a

Illustration 4.25 Klein bottle from a square.

The only difference here is that the arrows of the horizontal edges point in opposite
directions. Formally, that part of the identifying mapping is defined by first rotating
one of the two horizontal edges around the midpoint through 180°, then projecting
vertically to the other edge. An animation of this step and the resulting quotient space
is shown in Illustration 4.26.

Illustration 4.26 Identifying the a-edges of the square gives a cylindrical surface, as for the
torus. However, this time the bounding circles are identified differently, corresponding to the
orientation of the b-edges.

K12146_C004.indd 93 2/6/15 5:20 PM

 94   ◾    Subspaces, Quotient Spaces, Manifolds, and CW-Complexes

The Klein bottle is not embeddable in 3; this claim is beyond the scope of this book.
So, the animation shown in Illustration 4.26 has an unavoidable deficiency: we must
put up with self-intersections of the subspaces of 3 representing the Klein bottle.
Nevertheless, it is a beautiful object that deserves even an imperfect visualization. ☐

In the next example we describe and depict a 3-manifold.

Example 7: A 3-Manifold
The interesting 3-manifolds, with the possible exception of  3, are not embeddable in
 3. That notwithstanding, all of them can be described as quotients of spaces that are
embeddable in  3. That is what we will do in this example. Another formulation of
the space in this example will be provided in the next section (Exercise 17, Section 5.1).
A solid torus T (defined in Exercise 12, Section 4.3) is the closed subspace of  3
bounded and enclosed by a torus T 2 . For convenience we will assume that the
largest (equatorial) circle of T 2 is in the plane z = 0. Split T 2 into two closed sets:
F = T 2 ∩ {( x , y , z ): z ≥ 0} and G = T 2 ∩ {( x , y , z ) : z ≤ 0}. Project the points of F verti-
cally onto the points of G, then identify these two sets along that projection. In plain
terms, we identify every point of the upper half-torus with the point straight below
it on the lower half-torus, leaving the equatorial circles untouched. The resulting
space is a 3-manifold. A visual justification is given in Illustration 4.27 and in the
next paragraph.

A
B

Illustration 4.27 Every point in the quotient space has a neighborhood homeomorphic to the
open 3-disk.

Illustration 4.27 depicts the solid torus T, the boundary of which is to be identified as
indicated above to get the space T ~. The points in the interior of T are unaffected in
the sense that small open 3-disks around such point give rise to open neighborhoods of
the corresponding points in T ~. One such point is B in Illustration 4.27. Open 3-disks
also become neighborhoods in T ~ of what used to be the boundary points of T. We
illustrate that claim for the two points labeled A in Illustration 4.27 (becoming one
point after identification): the two half-disks in T as shown will fit into one open 3-disk
in T ~ around the single point we get by identifying the pair of points A. A similar

K12146_C004.indd 94 2/6/15 5:20 PM

 4.4 Manifolds and CW-Complexes   ◾    95  

argument works for the points on the two equatorial circles. Hence, every point in T ~
has a neighborhood homeomorphic to an open 3-disk, and so T ~ is a 3-manifold. ☐
Recall that we have chosen not to pay attention to Hausdorff-ness and second countability
in our examples. The readers are encouraged to devote a few moments and check that these
two properties are satisfied in all of the above examples.

We turn our attention to CW-complexes. A 0-cell is a point; a 1-cell is the interval [−1,1], and,
in general, an n-cell is the closed unit n-ball D n = {( x1 , x 2 ,… , xn ) ∈n : x12 + x 22 +  + xn2 ≤ 1}.
All of these together will be called cells.
A CW-complex X is defined recursively as follows. We start with any discrete space; it
constitutes the 0-skeleton X 0 of X, and it can be viewed as a discrete sum of 0-cells. Then
we attach copies of 1-cells via any mapping that sends the boundary {−1,1} of each 1-cell
into X 0 . Note that, since {−1,1} is a discrete subspace of [−1,1], such mappings are always
continuous. The resulting space is the 1-skeleton of X and is denoted by X 1. Assuming
we have the n-skeleton X n, n ∈, the (n + 1)-skeleton is obtained by attaching copies of
D n+1 via continuous mappings sending ∂(D n +1 ) = S n into X n. If this procedure terminates,
then X = X n for some smallest n ∈, and the finite number n is the dimension of the
∞
CW-complex X. Otherwise, X = ∪ X n, equipped with the weak1 topology determined by
n=0
{ Xi : i ∈I } (that is, U ⊂ X if and only if U ∩ X n is open for every n ∈). In this case we
open
take that the dimension of the CW-complex X is infinity.

Remark. X n as the n-skeleton and X n as an n-fold product of the space X with itself is an over-
lapping notation. It should be clear from the context which of the two meanings applies.
A process of building a simple CW-complex of dimension 2 is shown in Illustration 4.28.

Illustration 4.28 From left to right: X 0 , X 1 , and X 2 .

∞
It follows from Proposition 3, Section 4.3 that each skeleton X n of a CW-complex X = ∪ X n
n=0
is embeddable in X, and so it could be considered a subspace of X.
1
The word “weak” is the source of W in CW; C stands for “closure finite.”

K12146_C004.indd 95 2/6/15 5:20 PM

 96   ◾    Subspaces, Quotient Spaces, Manifolds, and CW-Complexes

If D n is attached to the (n − 1) -skeleton of X via a continuous map f : ∂(D n ) → X n −1, then

the image of D n under the mapping D n inclusion → X n −1 ⊕ D n quotient
 map
→ X n −1 ∪ f D n will
be called an n-cell in X. Since X n survives intact in X, the terminology makes sense.
A CW-complex is finite if it has finitely many cells. Otherwise, it is infinite. We will
mostly deal with finite complexes.
A subcomplex Y of a CW-complex X is a subspace of X obtained in such a way that for every
n ≥ 0 the (n + 1)-st skeleton of Y is obtained from the n-th skeleton of Y by attaching a subset
of the (n + 1)-cells used to construct X in the same way they have been attached in X. For
example, every n-skeleton of X is a subcomplex of X.

Example 8: n as a CW-Complex
The Euclidean space  is an infinite CW-complex of dimension 1. It is homeomor-
phic to the complex we get by attaching copies of 1-cells between each two consecutive
integers. A similar construction shows that n is a CW-complex of dimension n
(Exercise 7). With respect to this construction  is a subcomplex of  2 . Exercise 9
tells us that this is not always true. ☐

A CW-complex X of dimension 1 is also called a graph. In this setting the points in the
0-skeleton are called the vertices of X, and the images of the 1-cells under the quotient map-
ping are called the edges of the graph X. A special graph is introduced in the next example.

Example 9: Bouquets of Circles (Roses)

A bouquet of circles (or a rose) is a graph consisting of only one vertex and a non-
empty set of edges. The edges (circles) in the bouquet are sometimes called petals.
We will mostly be concerned with bouquets with finitely many petals. (Note that, in
Illustration 4.29, when there are infinitely many petals, one should not assume from
the figure that the topology is inherited from  3; we will come back to this point
again in 13.2.)

Illustration 4.29 A bouquet of circles. ☐

K12146_C004.indd 96 2/6/15 5:20 PM

 4.4 Manifolds and CW-Complexes   ◾    97  

Observe that it follows from Example 5 that a torus is a CW-complex of dimension 2 obtained
from a bouquet of two circles by attaching one 2-cell. We will eventually confirm that every
2-manifold is a CW-complex of dimension 2.

Exercises
1. (a) Find a space X such that the conditions (i) and (iii) in the definition of the n-mani-
fold are satisfied, and the condition (ii) fails.
(b) Find a space X such that the conditions (ii) and (iii) in the definition of the n-mani-
fold are satisfied, and the condition (i) fails.
2. Identify the edges of the filled polygon according to the labels and arrows shown in
Illustration 4.30, as explained in Examples 4 and 6; for example, the a-edges are iden-
tified along the obvious (linear) homeomorphism that follows the arrows. Describe
(or visualize) the resulting 2-manifold.
a

c b

c a

b
   Illustration 4.30
3. A Möbius band is the quotient space obtained from a filled rectangle by identifying
the two edges labeled a according to the arrows (top figure, Illustration 4.31). Start
with two copies of a Möbius band and identify the boundaries according to the labels
and arrows in the second row of figures in Illustration 4.31. Which 2-manifold is the
resulting quotient space?

a a

b b

a1 a1 a2 a2

c c
Illustration 4.31

K12146_C004.indd 97 2/6/15 5:20 PM

 98   ◾    Subspaces, Quotient Spaces, Manifolds, and CW-Complexes

4. (a) Define an equivalence relation ~ on the three-dimensional unit ball

D 3 = {( x , y , z ) : x 2 + y 2 + z 2 ≤ 1} by partitioning it into classes consisting of sin-
gletons or of sets of type {( x , y , z ),( x , y , − z )} whenever x 2 + y 2 + z 2 = 1. Visualize
3
a homeomorphism between D and the three-sphere S 3 . [Hint: Slice D 3 with
~
planes and see what happens to these slices after the identification.]
(b) Let f : D 3 → X and g : D 3 → Y be two homeomorphism (so that X and Y are two
copies of D 3). Identify the points f ( x , y , z ) and g ( x , y , z ) for each ( x , y , z ) ∈S 2.
What is the resulting quotient space?
5. Let O be a subset of  3. Denote the rotation around the x-axis through α radians
(α ∈ ) by fα , and denote the rotation around the z-axis through γ radians ( γ ∈ ) by
hγ . The configuration space C(O) of O with respect to the rotations fα and hγ (in that
2
order) is the quotient space  ~ of  2 , where ~ is defined as follows: (a, b) ~ (c , d ) if
hb ( fa (O )) = hd ( fc (O )). Describe the configuration space C(O) if

(a) O = {(0, 0, z ): − 1 ≤ z ≤ 1}.

1 1
(b) O = {(0, 0, z ): −1 ≤ z ≤ 1} ∪ {(0, y ,1) : 0 ≤ y ≤ } ∪ {(0, y , −1) : 0 ≤ y ≤ } . (See Illus­t­
2 2
ration 4.32.)

x
   Illustration 4.32

6. (a) Let G be the group of symmetries of  2 consisting of all compositions of transla-

tions along integer multiples of the vectors u = (0,1) and v = (1,0). Partition  2
into sets of type { g ( x ): g ∈G}, x ∈ 2 (these are the orbits of the elements of  2
under the action of G), and let ~ be the corresponding equivalence relation.
Describe the quotient space  2 .
~
(b) Same question, but G generated by a translation along u and the glide reflection
that is the composition of the translation along v followed by the reflection with
respect to the x-axis.

K12146_C004.indd 98 2/6/15 5:20 PM

 4.4 Manifolds and CW-Complexes   ◾    99  

7. Show that n is a CW-complex of dimension n.

8. Let X be the CW-complex obtained by attaching the closed disk via the mapping that
sends the boundary of the disk D to the only edge of the graph Γ as indicated by the
labels (see Illustration 4.33). Let Y be the space obtained from two open cones sharing
a slant height by identifying the two base circles labeled c, as indicated in Illustration
4.34. Explain why X is homeomorphic to Y.

a Y
c
Γ

a D a
a

a c
      
Illustration 4.33 Illustration 4.34

9. Find a CW complex Y of dimension 2 such that Y is homeomorphic to  2 and such

that no subcomplex of Y is homeomorphic to .
10. Show that every finite CW-complex is Hausdorff.
11. Show that the k-skeleton X k of a finite CW-complex X is closed in X.
2
1 1
12. Consider the subspace X = ∪ Ck of  2 , where Ck = {( x , y ) ∈ :  x −  + y 2 = 2 },
∞

k=1  k  k
k ∈ (see Illustration 4.35). Why is X not a CW-complex?

1.0

0.5

0.5 1.0 1.5 2.0

–0.5

–1.0
   Illustration 4.35

13. Show that the Klein bottle is a CW-complex of dimension 2.

K12146_C004.indd 99 2/6/15 5:20 PM

This page intentionally left blank
 Chapter 5

Products of Spaces

T here are a couple of straightforward and reasonable ways to extend the notion of
products of metric spaces (Section 2.2) in the setting of topological spaces. The pre­
ference for one of these two choices will be clearer after the next chapter.

5.1 Finite Products of Spaces

A brief historical note: Fréchet was the first to study finite products of abstract (topological)
spaces (1910).

The product space of two spaces X and Y is the set X × Y, together with the topology gener­
ated by all of the sets of type U × V , where U is open in X and V is open in Y. This topology
over X × Y is called the product topology.
It is easy to verify that the subsets of the set X × Y of type U × V , where U ⊂ X and
open
V ⊂ Y , indeed constitute a basis for a topology over X × Y . That particular basis will be
open
called the standard basis for X × Y We will denote both the product space and the product
set by X × Y, and, unless otherwise stated, we will always assume that X × Y denotes the
product space.
This definition is natural in the sense of the following proposition.

Proposition 1. The product topology over X × Y is induced by the projections p1 : X × Y → X

and p2 : X × Y → Y .

Proof. Denote the product topology by τ1 , and denote the topology induced
by { p1 , p2 } by τ 2 . By Proposition 6 in 3.5, τ 2 is generated by the subbasis

{ }{ }
S = p1−1 (U ) : U ⊂ X ∪ p2−1 (V ) : V ⊂ Y . Since p1−1 (U ) = U × Y and p2−1 (V ) = X × V , we
open open

101

K12146_C005.indd 101 2/6/15 5:21 PM

 102   ◾    Products of Spaces

have that p1−1 (U ) ∩ p2−1 (V ) = U × V , and so τ1 ⊂ τ 2 . On the other hand it is clear that
S ⊂ τ1, and so τ 2 ⊂ τ1.

We will now provide an alternative interpretation of product spaces, more suitable for
visualization.
Start with two spaces, X and Y. For every x ∈ X choose a copy Yx of Y, such that
if x1 ≠ x 2 , then Yx1 ∩ Yx 2 = ∅, and let φ x : Y → Yx be the associated homeomorphism.
Consider the set ∪ Yx . You may think of it as being obtained by replacing each point x of
x ∈X
X with a copy Yx of the space Y. Introduce a topology over that set as follows: it is generated
by the basis consisting of all subsets of the set ∪ Yx of type ∪ Vx , where U is an open
x ∈X x ∈U
subset of Y, and where for every x ∈U , Vx = φ x (V ) for some fixed open subset V of Y. It is
fairly easy to show that these types of sets indeed constitute a basis. We get a topological
space over ∪ Yx that we temporarily denote by P(X,Y). That temporary notation will be all
x ∈X
but extinguished by the following proposition.

Proposition 2. The product space X × Y and the space P(X,Y) are homeomorphic.

Proof. Define f : X × Y → P ( X , Y ) by f ( x , y ) = φ x ( y ). It is easy to see that f is a bijec­

( )
tion. The continuity of f follows from the observation that f −1 ∪ Vx = U × V , and from
x ∈U
Proposition 3(c) of Section 3.5. Similarly, since ∪ Vx = f (U × V ), the same proposition
x ∈U
implies that f −1 is continuous, and so f is open.

Corollary 3. For each fixed y ∈Y , the subspace {φ x ( y ) : x ∈ X } of P(X, Y) is homeomor­

phic to X.

Proof. The restriction of f to X × { y} is a homeomorphism onto {φ x ( y ) : x ∈ X } . On the

other hand it is obvious that X ≅ X × { y}.

In the view of Proposition 2 we may regard P(X, Y) as being the same as X × Y . So, in
order to visualize P(X, Y) or X × Y we need to replace each point of X with a copy Yx of
the space Y. Corollary 3 then asserts that for each y ∈Y the sets {φ x ( y ) : x ∈ X } could be
viewed as a family of pairwise disjoint copies of X.

Example 1
Let X and Y both be homeomorphic copies of the circle S1 = {( x , y ) : x 2 + y 2 = 1}
(viewed as a subspace of  2). Then X × Y, or P(X,Y), is depicted in Illustrations 5.1
and 5.2.

K12146_C005.indd 102 2/6/15 5:21 PM

 5.1 Finite Products of Spaces   ◾    103  

Illustration 5.1 This is an illustration of the

space X × Y = S1 × S1. It is, of course, incom­
plete, mainly because only finitely many
points of X have been replaced by copies of
the space Y. A better illustration is given in
Illustration 5.2. However, in more complicated
cases, when X × Y is not embeddable in our
visualizing space  3, the complete visualiza­
tion as in Illustration 5.2 will often obscure the
view of the main features of the space, and so
the restricted visualization as in Illustration 5.1
will sometimes be preferable.

Illustration 5.2 This is a better view of S1 × S1.

The spaces Yx are assembled continuously
within the product space X × Y . The topo­
logy is determined by the standard basis, and
a basis element U × V , where Vx (a copy of
V in Yx ), is shown in this figure. We get a
standard depiction of the space S1 × S1 . It is
fairly faithful, since S1 × S1 is embeddable in
U  3 . Note, however, that even  3 is not good
enough to properly depict the symmetry that
Vx
exists in X × Y between X and Y.

Example 2
Suppose now that Y is the circle S1 = {( x , y ) : x 2 + y 2 = 1} as a subspace of  2 , and
suppose X is the same circle, but with the discrete topology. Then the structure of the
space X × Y is entirely different from the space X × Y in Example 1. See Illustration 5.3
and its caption for further explanation.

Illustration 5.3 The topology of this space is

different from the space shown in illustration.
One member of the standard basis of X × Y
is U × V, where Vx is a copy of an open arc
V in some Yx , while U is just the point x.
There is really nothing that holds the circles
U Yx together in this space; X × Y is now just a
Vx bunch of circles, one per each point in X. See
also Exercise 3 below.
☐

K12146_C005.indd 103 2/6/15 5:22 PM

 104   ◾    Products of Spaces

The product space T 2 = S1 × S1 , where both copies of S1 are with the usual topology (inher­
ited from  2), is homeomorphic to the torus (as defined in Section 4.41). According to
the following proposition, higher dimensional manifolds can easily be generated through
products of spaces.

Proposition 4. Let X be an n-manifold, and let Y be an m-manifold. Then X × Y is an

n + m -manifold.

Proof. Take a point ( x , y ) ∈ X × Y . Since X is an n-manifold, there is an open neighborhood

U of x such that U ≅ n . Similarly, there is an open neighborhood V of y, such that V ≅ m.
Then U × V ≅ n × m (Exercise 1), and n × m ≅ n +m (Exercise 13, Section 2.2).

Example 3
By Proposition 4, the space T 2 × S 2 , where T 2 is a torus and S 2 is the unit sphere is
a 4-manifold. It is embeddable (and visualizable) in  6 (Exercise 5). Nevertheless,
Proposition 2 allows us to depict the main features of that space in Illustration 5.4;
the caption explains some of the limitations of that illustration.

Illustration 5.4 All of the points of the torus are replaced by disjoint copies of the
sphere—for obvious, technical reasons, our illustration only shows finitely many
spheres. But that is just one of the limitations of our rendering of T 2 × S 2 . We also show
three, semi-transparent, concentric tori in this illustration; each one of the tori is made
of points on the spheres, one point per a copy of the sphere S 2 , and each of these points
correspond to a fixed point of the original space S 2 . For example, the north poles of all
spheres together make one torus.

According to Corollary 3, a properly depicted T 2 × S 2 should show infinitely many

copies of T 2 , too. It is difficult to show more than one, since the largest will obscure
the view of the smaller tori. We use three transparent tori in Illustration 5.4 to stress

1
As in the case of S 2 or  2, the superscript in T 2 is to indicate that we deal with a two-dimensional manifold.

K12146_C005.indd 104 2/6/15 5:22 PM

 5.1 Finite Products of Spaces   ◾    105  

that point. We are reaching the limits of visual depiction of manifolds as a way of
gaining intuitive understanding. ☐

Example 4
In Illustration 5.5 we depict T 2 × K , where T 2 is a torus and K is a Klein bottle. In
Illustration 5.6 we zoom in and slice to show the texture of the space, consisting of
small copies of a Klein bottle, one per each point of the torus.

Illustration 5.5 T × K.  Illustration 5.6 We zoom in and cut to see

better.

This is a standard visualization procedure used to model the space we sometimes call the
universe. For example, the Calabi–Yau model in string theory is the ten-dimensional
manifold  4 × CY , where  4 accounts for  3 (the three “spatial dimensions”) and
for linear time as the fourth dimension, and where each point of  4 is replaced by a
small (and relatively complicated) 6-manifold CY (Celabi–Yau space) according to
our procedure P. ☐

We enumerate a few basic properties of products of spaces.

Proposition 5. The mapping f : X × Y → Y × X defined by f ( x , y ) = ( y , x ) is a home­

omorphism.

Proof. It is clear that f is a bijection. Since f −1 (U × V ) = V × U for every set U × V in the

standard basis for X × Y , f is continuous. It follows by symmetry that f is open.

Example 5
In the case of relatively simple topological spaces we can literally see that
X × Y ≅ Y × X . For example, in Illustrations 5.7 and 5.8 we show what we get when
the space X is a copy of the circle S1 and the space Y is the interval [0,1] (depicted as
a vertical line segment).

K12146_C005.indd 105 2/6/15 5:22 PM

 106   ◾    Products of Spaces

Illustration 5.7 In the space S1 × [0,1] we replace Illustration 5.8 … while in the space [0,1] × S1
1
each point of S by a (vertical) line segment … we replace each point of [0,1] by a circle.

We can see in Illustrations 5.7 and 5.8 that X × Y is homeomorphic to Y × X . Both

illustrations, as we have indicated several times by now, are incomplete skeletons
of the full objects. Both X × Y and Y × X are cylinders. ☐

Example 6

In the case of subspaces of n , n ≥ 4, visually justifying that X × Y ≅ Y × X is less

convincing. For example, the spaces T 2 × S 2 and S 2 × T 2 , as depicted in Illustrations
5.4 and 5.9, are homeomorphic.

Illustration 5.9 S 2 × T 2 .

The space T 2 × S 2 is shown in Illustration 5.4: every point of the torus is replaced by
a (small) sphere. In Illustration 5.9, we show the space S 2 × T 2 where each point on
the sphere is replaced by a small torus. The difference between the two spaces is geo­
metrical rather than topological: in Illustration 5.4 the tori are large and the spheres
are small, while in Illustration 5.9 it is the other way around. ☐

The product space of the spaces X, Y, and Z (in that order) is defined to be ( X × Y ) × Z . The
product space of the spaces Xi , i = 1,2,… , n , (in that order) is (…(( X1 × X 2 ) × X 3 ) × … × Xn ).

K12146_C005.indd 106 2/6/15 5:22 PM

 5.1 Finite Products of Spaces   ◾    107  

Proposition 6. The collection of sets of type (U × V ) × W = {(( x , y ), z ): x ∈U , y ∈V , z ∈W },

where U, V, and W range through the open sets in X, Y, and Z, respectively, is a basis for the
product space ( X × Y ) × Z .

Proof. If B1 and B 2 are bases for the spaces X and Y, respectively, then the collection of sets
{U × V : U ∈B1 , V ∈B 2 } is a basis for the product space X × Y (Exercise 6). That, together
with the fact that the collection of open sets U × V = {( x , y ): x ∈U , y ∈V }, U is open in X
and V is open in Y, is a basis for X × Y , implies the statement in the proposition.
The following simple proposition will allow us to further simplify the notation.

Proposition 7. For any three spaces X, Y, and Z, the mapping h : (( x , y ), z )  ( x ,( y , z ))

defines a homeomorphism from ( X × Y ) × Z onto X × (Y × Z ).
Proof. The mapping h is a bijection. The argument given in Proposition 6
needs only a minor modification in order to conclude that the collection
U × (V × W ) = {( x , ( y , z )): x ∈U , y ∈V , z ∈W }, U, V, and W range through the open sets in
X, Y, and Z, respectively, is a basis for X × (Y × Z ). Since h −1 (U × (V × W )) = ((U × V ) × W ),
it follows that h is continuous. By symmetry, h is open.

Proposition 7 (together with a straightforward induction) allows us to disregard the paren­

theses. Hence we will denote the product space of X, Y, and Z by X × Y × Z, and wen will
denote the product space of Xi , i = 1,2,…n, by X1 × X 2 × … × Xn , or, more succinctly, by ∏ Xi .
i =1
Example 7
The two images in Illustration 5.10 depict the space (S 2 × T 2 ) × S 2 , where, as before,
S 2 is the unit sphere and T 2 is a torus.

Illustration 5.10 The space (S 2 × T 2 ) × S 2 is obtained by first replacing each point in S 2 by a

copy of T, and then replacing each point in each copy of T 2 by a copy of S 2 . The right-hand-
side illustration is a magnification of a part of the left-hand-side picture: we see the small
spheres distributed along the corresponding tori. We repeat again that there are uncountably
many small tori and small spheres within the space ( X × Y ) × X and that we can only depict
finitely many of them. The space ( X × Y ) × X is a 6-manifold. ☐

K12146_C005.indd 107 2/6/15 5:22 PM

 108   ◾    Products of Spaces

Almost all of the properties of topological spaces that we have introduced so far are pre­
served in (finite) product spaces.
n
Proposition 8. (a) If Xi , i = 1,2,… , n , are first countable, then so is ∏ Xi .
n i =1
(b) If Xi , i = 1,2,… , n , are second countable, then so is ∏ Xi .
n i =1
(c) If Xi , i = 1,2,… , n , are separable, then so is ∏ Xi .
i =1
n
(d) If Xi , i = 1,2,… , n , are metrizable, then so is ∏ Xi .
i =1

n
Proof. (a) Take a point x = ( x1 , x 2 ,… , xn ) in ∏ Xi . There is a countable local basis Bi for xi ,
i =1 n
i = 1, 2, …, n. Consider the collection B x of sets of type ∏ Bi , where Bi ∈Bi . Proposition 4
i =1
in 1.2 implies that B x is countable. We show it is a local basis at x, thus justifying (a). Take
n n
an open subset U of ∏ Xi . Then there is a standard basis set ∏ U i containing x and such
n i =1 i =1
that ∏ U i ⊂ U . Since each U i is an open set containing xi and since Bi is a local basis at xi ,
i =1 n n
it follows that
n
there is an element Bi of Bi such that xi ∈ Bi ⊂ U i . But then x ∈∏ Bi ⊂ ∏ U i ,
i =1 i =1
and so x ∈∏ Bi ⊂ U .
i =1
We leave the proofs of (b) and (c) as exercises.
(d) Suppose Xi , i = 1,2,… , n are metrizable. So, there is a metric di for Xi such that the
metric space generated
n
by di is the same as the original topological space. Consider the pro­
duct metric over ∏ Xi (as defined in 2.2: for every x = ( x1 , x 2 ,… , xn ) and y = ( y1 , y 2 ,… , yn )
n i =1
in ∏ Xi , we have d( x , y ) = d1 ( x1 , y1 )2 + d2 ( x 2 , y 2 )2 +  + dn ( xn , yn )2 ). Then it is straight­
i =1
forward to check
n
that the topology generated by that metric is the same as the product
topology over ∏ Xi (Exercise 10).
i =1

Exercises
1. Show that if X1 ≅ X 2 and Y1 ≅ Y2 , then X1 × Y1 ≅ X 2 × Y2 .
2. T = {{n, n + 1, n + 2, …} : n ∈+ } ∪ {∅} is a topology over + (Exercise 4, Section 3.1),
and Ω = {{…, −n − 2, −n − 1, −n, n, n + 1, n + 2, …} : n ∈+ } ∪ {∅} is a topology
over \{0}. Show that (  \ {0}, Ω ) is homeomorphic to the product space X = + × {0,1},
where + is equipped with T , and {0,1} is with the indiscrete topology. Show that 
equipped with the topology Ω1 = {{…, −n − 2, −n − 1, −n, n, n + 1, n + 2,…} : n ∈} ∪ {∅}
is not homeomorphic to X.
3. Show that if X is a discrete space, then the product space X × Y is homeomorphic to
the discrete sum ⊕ Yx , where Yx , x ∈ X , are pairwise disjoint copies of the space Y.
x ∈X

4. (a) Prove that projections are open.

(b) Show that projections are not always closed.
5. Let A be a subspace of X and let B be a subspace of Y. Show that A × B with the pro­
duct space topology is homeomorphic to A × B considered as a subspace of X × Y.

K12146_C005.indd 108 2/6/15 5:22 PM

 5.1 Finite Products of Spaces   ◾    109  

6. Let B1 and B 2 be bases for the spaces X and Y, respectively. Show that the collection
of sets {U × V : U ∈B1 , V ∈B 2 } is a basis for the product space X × Y .
7. Visualize the space X × Y if
(a) X is the number 8 (viewed as a subspace of  2 ) and Y is the (unit) circle S1.
(b) The spaces X and Y are copies of the number 8 (viewed as a subspace of  2 ).
8. Visualize the space X × Y × X where
(a) X is the unit interval [0,1] and Y is the (unit) circle S1.
(b) X is the circle S1 and Y is a torus.
n
9. (a) Show that if each Xi , i = 1,2,… , n , is separable, then so is ∏ Xi .
i =1 n
(b) Show that if each Xi , i = 1,2,… , n , is second countable, then so is ∏ Xi .
i =1
10. Show that if Xi , i = 1,2,… , n , are metric spaces, then the product space and the prod­
n
uct metric space over ∏ Xi are homeomorphic.
i =1

11. Let f : X → Y be continuous, and let Z be the subspace of X × Y defined as follows:

Z = {( x , y ) ∈ X × Y : y = f ( x )} .
(a) Show that Z is homeomorphic to X.
(b) Show that if Y is Hausdorff, then Z is closed in X × Y .
12. Let f : Y → X and g : Z → X be homeomorphisms, let p1 : Y × Z → Y , p2 : Y × Z → Z
be the projections, and denote W = {( y , z ) ∈Y × Z : f ( y ) = g ( z )} . Show that the
restrictions pi W , i = 1, 2 are both homeomorphisms onto Y and Z respectively.
n n
13. Let Ai be a subset of the space Xi , i = 1, 2, …, n (so that ∏ Ai is a subset of ∏ Xi ).
n n i =1 i =1
Show that ∏ Ai = ∏ Ai . (Hint: Do it for n = 2 and then use induction.)
i =1 i =1 n
14. Let Xi , i = 1, 2, …, n, Y be spaces, and let f : Y → ∏ Xi be a mapping. Show that f is
i =1
continuous if and only if the compositions p j  f are continuous for every projection
n
map p j : ∏ Xi → X j j = 1, 2, …, n.
i =1
15. Show that if Y is Hausdorff and f : X → Y is continuous, then {( x1 , x 2 ) ∈ X × X :
f ( x1 ) = f ( x 2 )} is a closed subset of X × X .
16. Show that if Z is Hausdorff and f : X → Z , g : Y → Z are continuous, then
{( x , y ) ∈ X × Y : f ( x ) = g ( y )} is a closed subset of X × Y .
17. Describe the 3-manifold introduced in Example 7, Section 4.4, as a product of known
topological spaces.
18. Starting with the space X = {( x , y , z ) ∈ 3 : 1 ≤ x 2 + y 2 + z 2 ≤ 4} identify the bound­
ing spheres along the mapping f that sends a point (x, y, z), x 2 + y 2 + z 2 = 1 , to the

K12146_C005.indd 109 2/6/15 5:22 PM

 110   ◾    Products of Spaces

point (2x, 2y, 2z). We get the quotient space X f . Explain why X f is a 3-manifold and
describe it as a product of known spaces.
19. (a) Find a subset U of  2 such that the intersection of U with any horizontal or
vertical line is open in that line (considered as a subspace of  2 ), yet U is not
open in  2 .

(b) Show that the usual topology over X × Y is, in general, strictly weaker than the
topology induced by all coordinate mappings c1z0 : X → X × Y , c2z0 : Y → X × Y ,
where z 0 = ( x 0 , y0 ) ranges through X × Y , and where we define c1z0 ( x ) = ( x , y0 ),
and c2z0 ( y ) = ( x 0 , y ) .

5.2 Infinite Products of Spaces

A brief historical note: Infinite products of spaces were first introduced in 1930 by Andrey
Tychonoff (also spelled Tikhonov; 1906–1993). The product space topology is sometimes
called Tychonoff topology.

In this section we introduce a topology over the set-product ∏ Xi , where each Xi is a (non-
i∈I
empty) topological space, and I is any non-empty set, thus generalizing the finite case.
⋅
Recall that, formally, the elements of the set ∏ Xi are mappings f : I → ∪ Xi , such that
i∈I i∈I
f (i ) ∈ Xi for every i ∈I (the dot over the union symbol indicates that the sets Xi , i ∈I are
assumed to be pair-wise disjoint). We will more often use a simpler notation: instead of
⋅
f : I → ∪ Xi , such that f (i ) ∈ Xi for every i ∈I , we will use x = ( xi )i ∈I , where xi = f (i ) . In
i∈I
case I is countable, instead of x = ( xi )i ∈I we may write x = ( x1 , x 2 ,…, xn ,…), or, occasion­
ally, x = ( xn ) The elements xi are called the coordinates or components of x.
As noted earlier, we accept the validity of the Axiom of Choice, and so we take that
∏ Xi ≠ ∅ when all Xi ≠ ∅ and I ≠ ∅.
i ∈I

The projections p j : ∏ Xi → X j are defined by p j (( xi )i ∈I ) = x j . The product topology

i ∈I
over the set ∏ Xi is induced by the projections p j : ∏ Xi → X j , j ∈I ; this means that it
i ∈I i ∈I
is the weakest topology making the projections continuous. The product space will be
denoted by the same symbol ∏ Xi . When the index set I is the finite set {1, 2, …, n}, we
n i ∈I
will use the notation ∏ Xi , or X1 × X 2 ×  × Xn , as introduced in the preceding section.
i =1 ∞
When I is countably infinite, then we will denote the product space by ∏ Xi . When all
n i =1
Xi , i = 1, 2, …, n are copies of a single space X, then instead of ∏ Xi we will more often
i =1
write X n . Similarly, if all of the spaces Xi , i = 1, 2, …, n, …, are copies of a single space
∞
X, then instead of ∏ Xi we will often write X ∞ . The countably infinite interpretation
i =1
of the symbol ∞ in this context will be clear from the setup.

K12146_C005.indd 110 2/6/15 5:22 PM

 5.2 Infinite Products of Spaces   ◾    111  

The following theorem describes the product topology more explicitly.

Theorem 1. The product topology over ∏ Xi is generated by all sets of type ∏ U i where for
i ∈I i ∈I
every i ∈I , U i is an open subset of Xi , and where U i is Xi for all but finitely many i ∈I .

(In other words, the collection B of all sets of type ∏ U i , U i is open in Xi , and for all
i ∈I
but finitely many i ∈I , U i = Xi , is a basis for the product topology.)

Proof. Showing that B is a basis for some topology τ over ∏ Xi is straightforward

i ∈I
(Exercise 18 in Section 3.3). It is evident that the projections are continuous with respect
to the topology τ , so that the product topology is a subset of τ. It remains to show that
the elements of B are open in the product topology over ∏ Xi . Since, by assumption, each
i ∈I
projection p j is continuous, for every open subset V of X j , the set p −j 1 (V ) is open in the
product space ∏ Xi . Since p −j 1 (V ) = ∏ U i where U j = V and where U i = Xi for i ≠ j , and
i ∈I i ∈I
since all finite intersections of sets of such type constitute the collection B , it follows that
the elements of B are open in ∏ Xi .
i ∈I

As in the case of finite product spaces, the basis B will be called the standard basis for the
product space ∏ Xi .
i ∈I

Example 1
Suppose for each i ∈ , Xi is a copy of the discrete space over the set {0, 2}. The set
∏ Xi has the same cardinality as the set  . The nonempty open sets in the product
i ∈
space ∏ Xi are unions of sets of type ∏ U i where in all but finitely many cases U i is
i ∈ i ∈
Xi and in the remaining cases U i is a singleton. We will see this space again in this
section. ☐

Note that the collection of sets of type ∏ U i , U i is open in Xi , is also a basis for a topol­
i ∈I
ogy over the set ∏ Xi : if I is infinite, then this last topology, called the box topology, is dif­
i ∈I
ferent from the product topology. Even though one may argue that this generating basis
is more natural in some way than the standard basis, the box topology fails to satisfy
many properties that the product topology fulfills. The main culprit for the shortcom­
ings of the box topology is the overabundance of open sets. We will say more about the
box topology in the next section.
In the next example we will introduce the Hilbert metric space H. We will prove in
Section 7.6 that (the countable) product  ∞ is homeomorphic to a subspace of H. This
implies that  ∞ is metrizable. In fact, it is true that countable products of metric spaces
are always metrizable (we will not prove that). However, as we will show in the Example
3, this fails in case of uncountable products of metric spaces.

K12146_C005.indd 111 2/6/15 5:23 PM

 112   ◾    Products of Spaces

Example 2: The Hilbert Space

( )
∞
Let H be the set of all ( xn ) ∈ ∞ for which the sequence x12 + x 22 +  + xn2 con­
n=1
verges in . We will denote the point of con­vergence of a convergent sequence

( )
∞ ∞
x12 + x 22 +  + xn2 by Σ xi2 .
n=1 i =1
If {xn } and { yn } are in H, then so is {xn − yn } . In order to justify that claim, we need to
( ( y + y +  + y ) converge, then so does
)
∞ ∞
2 2 2
show that if x12 + x 22 +  + xn2 and 1 2 n
n=1 n=1

( (x − y ) + (x − y ) +  + (x − y ) ) . The latter sequence converges if and

∞
2 2 2
1 1 2 2 n n
n=1

( )
∞
only if the sequence ( x1 − y1 ) + ( x 2 − y ) + … + ( xn − yn )2 2
2
2
n =1
converges, and since
the last sequence is an increasing sequence of positive numbers, it suffices to show
that (( x1 − y1 )2 + ( x 2 − y 2 )2 +  + ( xn − yn )2 )n∞=1 is bounded from above. We compute:
n n n n n n n n n n
Σ ( xi − yi )2 = Σ xi2 − 2 Σ xi yi + Σ yi2 = Σ xi2 − 2 Σ xi yi + Σ yi2 ≤ Σ xi2 + 2 Σ xi yi + Σ yi2 .
i =1 i =1 i =1 i =1 i =1 i =1 i =1 i =1 i =1 i =1
n n ∞ ∞
The numbers Σ x 2
i and Σ y are not larger than the numbers Σ x and Σ yi2 ,
2
i
2
i
i =1 i =1 i =1 i =1
∞
2
respectively (where the latter two series converge since, by assumption, Σ x and i
i =1
∞
Σ yi2 converge). By the Cauchy–Schwarz inequality, for every n we have that
i =1
n n n n n n ∞
Σ xi yi ≤ Σ xi2 Σ yi2 . Hence we have that Σ xi2 + 2 Σ xi yi + Σ yi2 ≤ Σ xi2 +
i =1 i =1 i =1 i =1 i =1 i =1 i =1
∞ ∞ ∞
2 Σ xi2 ⋅ Σ yi2 + Σ yi2 , and since the expression to the right defines a (finite) num­
i =1 i =1 i =1
ber, we have justified our claim.
∞
We can now define a metric d over H by d( x , y ) = Σ ( xi − yi )2 for every x = ( xn )
i =1
and y = ( yn ) in H. An argument similar to the one given for the Euclidean metric
over  ∞ shows that this is indeed a metric. The metric space generated by this metric
is the Hilbert space.
Notice that the metric that defines the Hilbert space is a straightforward extension
of the usual metric over n . The set H can thus be viewed as consisting of all points
in  ∞ at a finite distance from the origin. ☐

Example 3
Let X be the discrete space over {0,1}. Then the product space ∏ X is not metrizable.
i ∈
We will show this by exhibiting a subset A of ∏ X and a point x ∈ ∏ X such that
i ∈ i ∈
x ∈ A and such that no sequence of elements in A converges to x. That is not possible
in metric spaces (Exercise 20, Section 3.2).

K12146_C005.indd 112 2/6/15 5:23 PM

 5.2 Infinite Products of Spaces   ◾    113  

Let A be the subset of ∏ X consisting of the elements of the type ( xi )i ∈ where

i ∈
x j = 0 for every j in a finite subset J ⊂  , and where xi = 1 for i ∈  \ J. Denote
x = ( 0 )i ∈ (each coordinate is 0). Every element in the standard basis for ∏ X con­
i ∈
taining x must be of the type ∏ U i where finitely many U i are {0} while the rest of
i ∈
them are equal to X. Such an open set ∏ U i will contain the elements of A that have
i ∈
0’s at the same coordinates where U i ’s are {0}. It follows that every open set contain­
ing x must contain points from A. So, x is an accumulation point for A, and thus
x ∈A .
Let s be a sequence of elements in A. We show that s does not converge to x. Each
element of A has 0 in finitely many coordinates, and since every sequence has count­
ably many members, the cardinality {k ∈ : x k = 0 for some member ( xi )i∈ of s} is
countable (as a countable union of finite sets). So, there exists a number j ∈ such
that x j = 1 for every member ( xi )i ∈ of the sequence s. Consider the following
open neighborhood of x: ∏ U i , where U j = {0} and U i = X if i ≠ j . It does not
i ∈
contain any member of s, and so s does not converge to x.
It follows from this example that ∏  is not metrizable, too (Exercise 12). ☐
i ∈

∞
In the case where the index set I is countably infinite, the product space ∏ Xi can also
i =1
be described by means of the procedure P that we have utilized to introduce the finite
products: we start with the space X1 , then we replace each of its points by disjoint copies
of X 2 , then we replace each point in each copy of X 2 by disjoint copies of X 3 ; continue
∞
this procedure recursively. In this way a point ( x1 , x 2 ,..., xi ,...) ∈∏ Xi can be viewed as cor­
i =1
responding to a sequence of such steps. The topology is the same: a basic neighborhood of
the point ( x1 , x 2 ,…, xi ,…) consists of a sequence of open sets, all but finitely many of them
being the whole spaces Xi . The procedure P allows us again to provide crude visualizations
of products of countably many spaces.

Example 4

Suppose the spaces X 2n+1, n = 0, 1, 2,… are (disjoint) copies of a torus, and suppose
the spaces X 2n, n = 1, 2,… are (disjoint) copies of a sphere. A very rudimentary illus­
∞ ∞
tration of the space ∏ Xi is provided in Illustration 5.11. In order to get ∏ Xi we start
i =1 i =1
with a torus X1 , then replace each of its points by a copy of a sphere (only 4 such
points are shown), then we replace each point of each copy of the sphere by a copy of
the torus (only 16 such points are shown) and iterate countably many times. In the
picture we show only 5 steps of that iteration.

K12146_C005.indd 113 2/6/15 5:23 PM

 114   ◾    Products of Spaces

Illustration 5.11 An oversimplified illustration of the space T 2 × S 2 × T 2 × S 2 × T 2 × S 2 × … . ☐

The product space depicted in Illustration 5.11 is (homeomorphic to) a fractal (Exercise 16).

Example 5: The Cantor Set

The Cantor set we are about to construct is a subset of the closed interval [0,1] (with
the usual topology). We delete from [0,1] the open middle interval ( 13 , 23 ) to get C1
(the first step in the construction; see Illustration 5.12). Then we delete the middle
intervals ( 19 , 92 ) and ( 79 , 89 ) from the remaining two intervals to get C2 . Iterate: We get
Cn+1 from Cn by deleting the open middle thirds of the closed intervals whose union
constitutes Cn . The Cantor set C is the set we get after performing this procedure
∞
infinitely many times; that is, C = ∩ Cn (with the subspace topology).
n=1

C1
0 1 2 1
3 3

C2
0 1 2 1 2 7 8 1
9 9 3 3 9 9

C3

Illustration 5.12 The first three steps in the construction of the Cantor set C.

It appears at first glance that after removing so many intervals not much is left in C.
However, C is still uncountable. Moreover, C is homeomorphic to the space ∏ Xi
i ∈
given in Example 1 (the product of uncountably many two-element discrete spaces).

K12146_C005.indd 114 2/6/15 5:23 PM

 5.2 Infinite Products of Spaces   ◾    115  

It is fairly simple to provide a justification for both of these two claims. Write the
numbers in the interval [0,1] as decimals 0. x1 x 2 … xn … using ternary base (using
digits 0,1, and 2; note that 1 = 0.222… ). The digits we use are the numerators in the
sum x31 + x322 +  + 3xnn + . When we remove the first middle subinterval to get C1 we are
removing the set of real numbers for which x1 = 1 (i.e., the numbers that can be writ­
ten as 0.1∗∗∗ in ternary base, where ∗∗∗ substitutes the rest of the decimal number),
except the number 0.1. Since 0.1 = 0.0222… , this number allows decimal representa­
tion without using the digit 1. Similarly, the step from C1 to C2 removes all decimals
for which x 2 = 1 , except the number 0.01 and 0.21. These two can also be represented
without using the digit 1 as follows : 0.01 = 0.00222…, and 0.21 = 0.20222… . We see
by induction that the numbers of the set C are precisely those that can be written
without using the digit 1 in decimal notation in ternary basis. It is now clear that C is
uncountable: the correspondence that associates to the number 0. x1 x 2 … xn … in C
the subset {n ∈+: xn = 2} of the set of positive integers is a bijection from C onto the
power set P (+ ), and, since by Proposition 3 in 1.2 P (+ ) is uncountable, so is C.
There is now a very straightforward candidate for a homeomorphism
f : C → ∏ Xi , where ∏ Xi is as in Example 1: writing the elements of C as ternary
i ∈ i ∈
decimals as in the previous paragraph, we set f (0. x1 x 2 … xn …) = ( xi )i∞=1 . We leave
the justification that f is indeed a homeomorphism as an exercise (Exercise 13).
We digress to note that the Cantor set gives rise to some exotic but important
spaces. In Illustration 5.13, we show a Cantor interval, an interesting homeomorphic
copy of the closed interval [0,1]. We will use this idea in the construction of the
Alexander’s Horned Sphere (to be seen in Section 13.4).

CS

1
32

1
16

Illustration 5.13 The Cantor interval. The dotted lines are only to indicate the stages of the
construction of the Cantor set. In each stage of this recursive construction we shrink slightly
so that no line in the Cantor interval is vertical. The vertical projection is then a homeomor­
phism onto the base interval. ☐

Let f : Y → ∏ Xi be a mapping. For every fixed j ∈I, the component (or coordinate) map-
i ∈I
ping f j : Y → X j of f is defined by f j ( y ) = p j ( f ( y )), where p j : ∏ Xi → X j is the projection.
i ∈I
We end this section with a useful property.

Proposition 2. A mapping f : Y → ∏ Xi is continuous if and only if the component map­

i ∈I
ping f j is continuous for every j ∈I .

K12146_C005.indd 115 2/6/15 5:23 PM

 116   ◾    Products of Spaces

Proof.
⇒ If f is continuous, then, since each component f j is a composition of two continuous
functions, it is also continuous.
⇐ Suppose each component f j is continuous. In order to show that f is continuous it suf­
fices to show that the inverse image of every member of the standard subbasis for ∏ Xi is
i ∈I
open in Y (Exercise 12, Section 3.5). A member U of the standard subbasis of ∏ Xi is of
i ∈I
the type ∏ U i , where U j0 is open in X j0 for some fixed j0 ∈ I , and U i = Xi otherwise. It
i ∈I
is easy to check that f −1 ∏U i  = f −j01(U j0 ). Since f j0 is continuous by assumption, the set
i∈I 
f j0 (U j0 ) is open in Y, and thereby, so is f −1 ∏ U i  .
−1
i ∈I 

Exercises
1. Show that each projection p j : ∏ Xi → X j is open.
i ∈I
2. Show that for every element z 0 = ( xi )i ∈I the coordinate mapping c zj 0 : X j → ∏ Xi
i ∈I
defined by c zj 0 ( x ) = ( yi )i ∈I , where yi = xi for i ≠ j , and y j = x , is an embedding.
3. Show that it is not true that if each of the spaces Xi is separable (first countable,
second countable), then the space ∏ Xi must be separable (first countable, second
i ∈I
countable, respectively).
4. (a) Let Y j be a closed subset of X j , for every j ∈ J . Show that ∏ Y j is a closed subset
j ∈J
of ∏ X j .
j ∈J
(b) Show that if Xi , i ∈I , are topological spaces and Ai ⊂ Xi are nonempty subsets,
then ∏ Ai = ∏ Ai  , where the closure to the right is taken in ∏ Xi .
i ∈I i ∈I  i ∈I
(c) In the setting of part (b), is it necessarily true that int ∏ Ai  = ∏ (int Ai )? Justify
i ∈I  i ∈I
your answer.
5. Show that if I is infinite, and if for every i ∈I , Xi is homeomorphic to a fixed space X,
then for any fixed j ∈I , the spaces ∏ Xi and ∏ Xi are homeomorphic.
i ∈I i ∈I
i≠ j
6. (a) Show that if Xi is homeomorphic to Yi , i ∈I , then ∏ Xi is homeomorphic to
i ∈I
∏ Yi .
i ∈I
(b) Show that if Xi is a subspace of Yi , i ∈I , then ∏ Xi is a subspace of ∏ Yi .
i ∈I i ∈I
7. Show that if I is infinite then the set ∏ U i , where each U i is a nonempty proper open
i ∈I
subset of Xi , is not open in ∏ Xi .
i ∈I
8. Show that if each Xi , i ∈I , is Hausdorff, then so is ∏ Xi .
i ∈I

K12146_C005.indd 116 2/6/15 5:23 PM

 5.3 Box Topology   ◾    117  

9. (a) Let X be any space over the set of natural numbers. Show that X ∞ is separable.
∞
(b) Show that if Xi , i = 1, 2, …, n , …, are spaces such that Xi =ℵ0 , then ∏ Xi is
i =1
separable.
10. Show that if Di is dense in Xi , i ∈ I , then the set ∏ Di is dense in the space ∏ Xi .
i ∈I i ∈I

11. Let ( xi )i ∈I be a fixed element in ∏ Xi . For every finite subset T of I, denote C(T ) = ∏ Ai ,
i ∈I i ∈I
where Ai = {xi } if i ∉T , and Ai = Xi if i ∈T . Show that the set Y = ∪ C(T ) is dense
T ⊂ I
in ∏ Xi . finite

i ∈I

12. Show that the product space ∏  is not metrizable. (Hint: Use Example 3.)
i∈
13. Show that the mapping f (0. x1 x 2  xn ) = ( xi )i∞=1 from the Cantor set C to product
space ∏ Xi (with each Xi a copy of the discrete space over {0, 2}, as in Example 5) is
i ∈I
indeed a homeomorphism.
14. Show that if X is second countable, then so is X ∞ .
15. (a) Show that the Cantor set C, considered as a subspace of , is a fractal (see
Example 4, Section 2.2).
(b) Visualize the product spaces C × C and C × C × C .
(c) Show that if X is a fractal then the product space X n is a fractal.
16. Consider the space X = T 2 × S 2 × T 2 × S 2 × T 2 × S 2 × … , as depicted in Illustration 5.11,
as a subspace of the metric space  ∞ (which you may assume is homeomorphic to the
Hilbert metric space H). Show that X is (homeomorphic to) a fractal.

5.3 Box Topology

A brief historical note: The box topology was first introduced in 1923 by Heinrich Tietze (1880–1964).

In this section we will briefly discuss the box topology mentioned in Section 5.2 (following
Example 1 there).
Given a set of spaces Xi , i ∈I , the box space is the space over the set product ∏ Xi
i ∈I
generated by the basis consisting of all sets of type ∏ U i , where each U i is open in Xi .
i ∈I
That these sets indeed make a basis for a topology is left as an exercise (Exercise 1). We
Box
denote the box space by ∏ Xi . The basis we have used to define the box space will be
i ∈I Box
called the standard basis for the box space ∏ Xi . The corresponding topology is called
i ∈I
the box topology over the set product ∏ Xi .
i ∈I

K12146_C005.indd 117 2/6/15 5:23 PM

 118   ◾    Products of Spaces

Box
If the index set I is finite, then the box space ∏ Xi coincides with the product space
i ∈I
∏ Xi . Otherwise the box topology is strictly finer than the product topology. For example,
i ∈I
if each of the spaces Xi , i ∈+ is equipped with the discrete topology over {0,1}, then
Box
{(0, 0, 0, …, 0, …)} is an open subset of ∏ Xi (it is an element of the standard basis), but it is
i ∈I
not open in the product space ∏ Xi .
i ∈I
The main problem with the box space is that it has too many open sets. As a conse­
quence, some important properties which are preserved in the product space are in general
lost in the box spaces. For example, the analogue of Proposition 2, Section 5.2 is no longer
true (Exercise 2). Another example is given in the next proposition.
∞
Proposition 1. The box topology over the set ∏  is not first countable.
i =1

∞
Proof. Take a point x = ( x1 , x 2 ,…, xn ,…) in ∏  and suppose B x = {B j : j = 1,2,…} is a count­
i =1
able local basis at x. Then there is a countable local basis at x consisting of sets in the standard
basis (with respect to the box topology). Hence, we may assume that the elements of B x are
∞
all from the standard basis of the box space, so that each of them is of the form B j = ∏ U j , i ,
i =1
where each U j , i is nonempty and open in  (with ∞
the usual topology). Let Vi be a proper
open subset of U i ,i containing xi . Then the set ∏ Vi is an open neighborhood of x. Hence
∞ i =1
there is a member of B x that is a subset of ∏ Vi . However our construction prevents that from
i =1
happening (see Illustration 5.14 and its caption for further explanation). So, B x could not be
a local basis at x.

U1,1 U1,2 U1,3 U1,n

U2,1 U2,2 U2,3 U2,n

Illustration 5.14 We use the standard diag­
U3,1 U3,2 U3,3 U3,n onal argument: by shrinking the indicated
open subsets of  (shown in frames in the
array to the left) to proper subsets Vi , we get
∞
a new open neighborhood ∏ Vi of x such
Un,1 Un,2 Un,3 Un,n i =1
that no listed neighborhood of x is a subset
∞
of ∏ Vi .
i =1

∞
Corollary 2. The box space over ∏  is not second countable and it is not metrizable.
i =1

Proof. The first part follows from Proposition 1 here and Exercise 10 in 3.3. The second part
is a consequence of Proposition 1 here and Exercise 13, Section 3.3.

The main reason why the product topology is preferable to the box topology will be seen in
Chapter 7: the former preserves compactness, the latter does not.

K12146_C005.indd 118 2/6/15 5:23 PM

 5.3 Box Topology   ◾    119  

Exercises
1. For a collection of spaces Xi , i ∈I , show that the set of all sets of type ∏ U i , U i is
i ∈I
open in Xi , is a basis for a topology over the set ∏ Xi .
i ∈I
Box
2. Find a mapping f :  → ∏  such that each coordinate function fi , i = 1, 2,… is
i∈+
continuous, but f is not continuous. (Compare with Proposition 2, Section 5.2.)
Box Box
3. Show that ∏ Ai = ∏ Ai .
i ∈I i ∈I
Box
4. Show that if each Xi , i ∈I , is Hausdorff, then so is ∏ Xi .
i ∈I

5. Consider the set X = ∏  of all functions  → ; let B be the set of all bounded
i ∈
functions in X; so f ∈B if there are numbers u , v ∈, such that u ≤ f ( x ) ≤ v for
every x ∈ .

(a) Show that the sets of type ∏ ( f (i ) − ε , f (i ) + ε) , f ∈B , ε > 0, comprise a basis for
i ∈
a topology over B. [Here ( f (i ) − ε , f (i ) + ε) denotes an interval; two basis sets
are shown in Illustration 5.15.]
g + ε1
g
g – ε1

f + ε2

f – ε2

  
Illustration 5.15 Two sets from our basis of bounded real valued functions. What is
their intersection? Careful!

(b) Define ρ : B × B →  by ρ( f , g ) = sup{ f (t ) − g (t ) : t ∈} . Show that ρ is a metric.

(c) Show that the topology T generated by the basis in part (a) is the same as the
topology of the metric space (B, ρ).
(d) Show that the topology T is strictly coarser than topology over B considered as
a subspace of X = ∏  equipped with the box topology.
i∈

K12146_C005.indd 119 2/6/15 5:23 PM

This page intentionally left blank
 Chapter 6

Connected Spaces and

Path Connected Spaces

6.1 Connected Spaces: Definition and Basic Facts

A brief historical note: The notion of connectedness of some subsets of  2 was introduced
in 1893 by Camille Jordan (1838–1922). A systematic study of connected topological spaces
started in 1914 with a book by Felix Hausdorff.

The definition of connectedness is fairly simple.

A space X is disconnected if the set X is a union of two disjoint, nonempty, open subsets A
and B of X. Otherwise we say that X is connected. The set {A, B} is called a separation of
the space X (Illustration 6.1).

A B

Illustration 6.1 A space X is disconnected if it is a union of two disjoint, nonempty, open

subsets A and B.

121

K12146_C006.indd 121 2/12/15 4:40 PM

 122   ◾    Connected Spaces and Path Connected Spaces

Example 1
Suppose X = {a, b} is with the discrete topology. Then X is disconnected, since
X = {a} ∪ {b}, where {a} and {b} are two disjoint, nonempty, open subsets of X. If the same
set X is equipped with the indiscrete topology over the same set, then it is connected. The
indiscrete space over any set is always connected. ☐

Example 2
The Euclidean space  is connected. We will show that  is not a union of two disjoint,
nonempty subsets. Suppose otherwise:  = A ∪ B, where A and B are two disjoint, non-
empty, open subsets of . Since A ≠ ∅, there is an element a in A. Since B ≠ ∅, there is
an element b ∉ A such either a < b or b > a. These two cases are symmetric, so we con-
sider one of them, say a < b. Look at the set S = {x ∈ A : x < b} : it is nonempty (since
a ∈S ), and it is bounded from above by b. By the Least Upper Bound Property, there
exists the least upper bound for S. This means that there is an element u ∈ such that
x < u for every x ∈S , and such that u is the smallest element with that property. Where
is this least upper bound u? Since  = A ∪ B, it must be either in A or in B. We will show
that in both cases we get a contradiction and so  cannot be disconnected.
A
B

a u b

Illustration 6.2 Where is u: in A or in B?

Suppose u ∈A. Since A is open, u must be an interior point. So, there is an interval
(u − r , u + r ) such that (u − r , u + r ) ⊂ A. Since b is not in A, it is not in (u − r , u + r ).
Since u ≤ b (u being the least upper bound), it follows that b must be larger than all
r r
of the elements in (u − r , u + r ). In particular, u + < b . But, that would put u + in
r 2 2
S, and since u < u + , u is not an upper bound for S. Contradiction.
2
Suppose u ∈B. Because B is open, there is an interval (u − r , u + r ) such that
(u − r , u + r )⊂ B. Virtually the same argument as in the preceding paragraph implies that
r
u − is a smaller upper bound for S, contradicting the choice of u. ☐
2
Since the last example is important for us, we promote its conclusion to a proposition.

Proposition 1. The Euclidean space  is connected.

Suppose now that we have an order topology over a set X (Example 8, Section 3.1) and sup-
pose the space X satisfies the Least Upper Bound Property with respect to that linear
order on X. Would a slight modification of the proof of Proposition 1 be sufficient to
show that X is connected? The word “slight” refers primarily to the use of the interval

K12146_C006.indd 122 2/12/15 4:40 PM

 6.1 Connected Spaces: Definition and Basic Facts   ◾    123  

(u − r , u + r ) around u in  : in the setting of the ordered topology, it should be replaced

with an open interval (c, d) containing the point u. We see a problem: in the above proof
of Proposition 1 we have found an element in (u − r , u + r ) that is between u and u + r
(the number u + 2r ). There need not be an element in an interval (x, y) in the linearly
ordered space X, and any modification of the above proof to the more general setting
must account for that difference. It is easy to see that this would be sufficient: if a
linear order in X is such that for every x < y in X, there is an element z ∈ X such that
x < z < y , then a minor modification of the above proof of Proposition 1 will work for
ordered spaces (with the two extra conditions). We summarize this in the following defi-
nition and in Proposition 1ʹ.

An ordered space X is a linear continuum if the linear order satisfies the following two
properties.

(i) (The Least Upper Bound Property) If A is a subset of X that is bounded from above,
then it has the least upper bound in X.
(ii) For every x , y ∈ X such that x < y , there is an element z ∈ X such that x < z < y .

Proposition 1′. Every linear continuum is connected.

Connected subsets of  are easy to classify.

Proposition 2. A subset A of  is connected if and only if it is an interval.

Proof. ⇒ Suppose A is connected. Suppose the largest lower bound x for A and the least
upper bound y for A exist. If x = y, then A = [x , x ], a singleton. Suppose x < y. There is no
element c strictly between x and y that is out of A, since otherwise { A ∩ (−∞, c ), A ∩ (c , ∞)}
would be a separation of A. It follows that A is ([x , y]), where ([ and ]) mean “open or closed”
to the left and right, respectively. If the largest lower bound x for A does not exist (but
y exists) then the set A is the interval (−∞, y]) . Similarly, if the least upper bound for A
does not exist, while x exists, then A is ([x , ∞). Otherwise A = (−∞, ∞), and we are back to
Proposition 1.

⇐ Repeat the argument given in Proposition 1.

Example 3
Recall that the Sorgenfrey (half-open) topology over  is generated by all half-
open intervals of type [a, b). This topology is a refinement of the Euclidean topology
(Exercise 2 in 3.3), and the additional open sets cause its disconnectedness. For
exa­mple,  = (−∞,1) ∪ [1, ∞), where the two intervals are open in the Sorgenfrey
topo­logy. If one is to visualize the Euclidean  as a smooth, unbounded rod,

K12146_C006.indd 123 2/12/15 4:40 PM

 124   ◾    Connected Spaces and Path Connected Spaces

then the Sorgenfrey space  is the same rod with infinitely many cracks in it (see
Illustration 6.3).

Illustration 6.3 In order to depict the structure of the (infinitely thin) Sorgenfrey line  we
fatten it up to the space  × S1 (that shares many properties related to connectedness with
the Sorgenfrey line ). Roughly speaking, there are natural cracks in the fabric of the space
 × S1 (left-hand-side picture), and we can split the space into two, disjoint subspaces without
damaging the “connecting fiber” (the right-hand-side picture). What we mean by “splitting
the space without damaging the fiber” is clarified by Proposition 3. ☐

Proposition 3. Let A and B be open subspaces of a space X. {A, B} is a separation of the

space X if and only if X = A ⊕ B.

Proof.
⇒ Assume {A,B} is a separation for X. We prove that the topology of X is the same as the
topology of the discrete sum A ⊕ B.
If U is open in X, then U ∩ A is open in A and U ∩ B is open in B, so that U is open in A ⊕ B.
Conversely, if U is open in A ⊕ B, then, by Proposition 2 in 4.3, U ∩ A is open in A and U ∩ B is
open in B, and since A and B are open in X, it follows that (U ∩ A ) ∪ (U ∩ B ) = U is open in X.
⇐ Suppose X = A ⊕ B , where A and B are equipped with the subspace topology. Then
both A and B are open in X , X = A ∪ B , and A and B are nonempty, so that { A, B} is a
separation of X.

The following is a simple criterion for connectedness.

Theorem 4. A space X is connected if and only if the only open and closed sets (clopens)
are X and ∅.

Proof.
⇒ Suppose A is a proper nonempty subset of X that is both open and closed. Then Ac is
also open (since A is closed), and it is nonempty and not equal to X (since A is such). So,
X = A ∪ Ac and X is disconnected.
⇐ Suppose X is disconnected. So, X = A ∪ B for some nonempty, disjoint, open subsets
⋅
A and B. Since A and B are nonempty and disjoint, they are not X or ∅. Since X = A ∪ B,

K12146_C006.indd 124 2/12/15 4:40 PM

 6.1 Connected Spaces: Definition and Basic Facts   ◾    125  

the sets A and B are mutual complements of each other and so they are both open and
closed.

A subset A of a space X is connected if it is connected as a subspace of X. So the attribute

“connected” applies to subsets and subspaces in the same way.

The following observation will be used without explicitly referring to it: if A ⊂ X ⊂ Y

and Y is a space, then A is a connected subset of Y if and only if A is a connected subset of
the space X considered as a subspace of Y.

We will refer to the next theorem quite a few times later on.
Theorem 5. If A is a connected subset of a space X, then so is A.

Proof. Suppose A is disconnected, so that A = B ∪ C for some nonempty, disjoint, open

subsets of A (considered as a subspace of X). It does not matter whether we consider A as a
subspace of X or as a subspace of A (see Exercise 1, Section 4.1) and so we will consider A as
a subspace of A. Suppose A ⊂ B. Then C ⊂ A \ A, and so C is disjoint from A. This is not pos-
sible since C is a non-empty open set and since the only points of A \ A are the accumulation
points of A. So, A is not a subset of B. By symmetry, A is not a subset of C. Consequently,
A ∩ B and A ∩ C are two nonempty and disjoint subsets of A. They are open in A (since B
and C are open in A). So, A is disconnected.

The following proposition allows us to build larger connected spaces by taking unions
of connected subspaces.

Proposition 6. (a) Let Ai , i ∈I , be connected sub-

sets of a space X, and suppose ∩ Ai ≠ ∅. Then ∪ Ai A4
i∈I i∈I A5
is a connected subset of X (see Illustration 6.4). A1

(b) Let A0 and Ai , i ∈I , be connected subsets of

a space X and suppose A0 ∩ Ai ≠ ∅ for every i ∈I . A2

( )
Then, A0 ∪ ∪ Ai is a connected subset of X.
i∈I
A3

∩ Ai
Proof. (a) Suppose ∪ Ai is disconnected. By i∈I

i∈I
Theorem 4, there is a nonempty, proper subset B of
∪ Ai such that B is a clopen in ∪ Ai . Since B ≠ ∅ , Illustration 6.4 The intersection of the conn­
i∈I i∈I
ected sets A (shaded part) is not empty.
there is some j ∈I , such that B ∩ A j ≠ ∅. The only i

clopens in Aj are ∅ and Aj ; hence, B ∩ A j = A j. So Aj ⊂ B.We proved that if B ∩ Ai ≠ ∅ then

Ai ⊂ B . It follows that B = ∪ Ai for some proper subset J of I.
i∈J

K12146_C006.indd 125 2/12/15 4:40 PM

 126   ◾    Connected Spaces and Path Connected Spaces

Since ∅ ≠ ∩ Ai ⊂ A j ⊂ B , it follows that Ai ∩ B ≠ ∅ A0

i ∈I A1
for every i ∈ I . Hence Ai ⊂ B for every i ∈ I . Consequently
∪ Ai ⊂ B , and so ∪ Ai = B . We proved that the only non- A2
i ∈I i ∈I
empty clopen in ∪ Ai is ∪ Ai itself, thus establishing its A3
i ∈I i ∈I
connectedness. A4
(b) Illustration 6.5 depicts the setting of this part of A5
the propo­sition. An argument that is very similar to the
one provided in part (a) suffices. We leave the details as A6
an exercise (Exercise 2). ◼
Illustration 6.5
Example 4

( ∞
)
Consider the subspace X = ∪ {( x , 1 n) : x ∈} ∪ {( x ,0) : x ∈} ∪ {(0, y ) : y ∈} of
2
n=1

 (Illustration 6.6). The vertical line and all horizontal lines are connected (being
copies of ). Since each of the horizontal lines intersects with the vertical line, it
follows from Proposition 6(b) that this space is connected. ☐

Illustration 6.6 This space is connected; the dots at the ends of the lines are there only to
indicate that the lines are unbounded.

Example 5
Suppose now that we delete the point (0,0) from the space X described in Example
4. The line X 0 = {( x ,0) : x ∈} does not intersect the vertical line any more and so
Proposition 6 does not apply. Is the set X \ {(0,0)} still connected? Perhaps, somewhat
surprisingly, the answer is affirmative! The closure of X \ X 0 in X \ {(0,0)} is X \ {(0,0)}.
This is so since all of the points in X 0 \ {(0,0)} are accumulation points in X \ {(0,0)}
for the set X \ X 0. Since X\X0 is connected (Proposition 6(b)), it follows from Theorem 5
that X \ X 0 (closure taken in X \ {(0,0)}) is connected, and since X \ X 0 is X \ {(0,0)}, it
follows that X \ {(0,0)} is connected.
In fact, we can delete as many points from X0 as we wish; a similar argument
applies and the remaining set will still be connected (Illustration 6.7).

K12146_C006.indd 126 2/12/15 4:40 PM

 6.1 Connected Spaces: Definition and Basic Facts   ◾    127  

Illustration 6.7 We have deleted some points from the x-axis. ☐

Remark. Example 5 suggests that our definition of connectedness might not match the
intuitive (dictionary) meaning of that word: even though the set X \ {(0,0)} appears to be
visually disconnected (Illustration 6.7), there are internal links that make it connected.
(A better match for the dictionary meaning of connectedness might be the path connected-
ness. We will introduce this later in this chapter.)
A component C in a space X is a connected subset of X such that if C ⊂ D and D is con-
nected, then C = D .
Components are maximal (under inclusion) connected subsets. The component containing
an element x in X is the union of all connected subsets of X containing x (Exercise 5). This union
is connected by Proposition 6(a).

Proposition 7. Components are closed.

Proof. Suppose A is a component in X. Suppose A is not closed. Then A properly con-

tains A and, by Theorem 5, it is connected. That contradicts the assumption that A is a
component.

Components need not be open (Exercise 8).

Proposition 8. A space is connected if and only if it has only one component.

Proof. Easy, and we leave it as an exercise.

Observe that the class of all components of a space is a partition of the space.

Exercises
1. Show that X is disconnected if and only if there are two nonempty open subsets A and
B of X, such that X = A ∪ B and A ∩ B = ∅ = A ∩ B.
2. Prove Proposition 6(b).

K12146_C006.indd 127 2/12/15 4:40 PM

 128   ◾    Connected Spaces and Path Connected Spaces

3. Let Ai , i = 1,2, … , be a collection of connected subsets of a space X such that

∞
Ai ∩ Ai+1 ≠ ∅ for every i = 1,2, …. Show that ∪ Ai is connected.
i =1

( )
4. Show that the set { x ,sin 1 x : x ∈ + } ∪ {(0, y ) : −1 ≤ y ≤ 1} is connected.
5. Show that for every x in a space X, the component of X containing x is the union of all
connected subsets of X containing x.
6. A space is totally disconnected if its components are singletons. Show that  is a
union of two totally disconnected subspaces.
7. Find three distinct connected subsets A, B, and C, of  2 such that ∂ A = ∂ B = ∂C .
8. (a) Show that if X has finitely many components, then each of them is both open and closed.
(b) Show that, in general, components need not be open.
9. Describe the components of the space X.
(a) X is the subspace of  consisting of all irrational numbers.
(b) X is the Sorgenfrey line.
10. Show that  2up = {( x , y ) : y ≥ 0} with the tangent disk topology is connected.

11. Find a subspace of  2 with uncountably many components, all of them uncountable.
12. Show that if there is a dense connected subset of a space X, then X is connected.
13. Show that every countable subset A of a metric space such that | A| > 1 is
disconnected.
14. (a) Find a subspace X of  2 such that for every points u, v ∈ X , the set X \ {u, v} is
connected, and there exist points x , y , z ∈ X, such that the set X \ {x , y , z} is
disconnected.
(b) Find a subspace X of  2 such that for every n ∈+ and every points x1 , x 2 ,…, xn ∈ X ,
the set X \ {x1 , x 2 ,…, xn } is connected, and there exist points y1 , y 2 ,…, yn , yn+1 ∈ X ,
such that X \ { y1 , y 2 ,…, yn , yn+1 } is disconnected.
15. Show that no closed interval in  is a union of m many pairwise disjoint closed sets,
where 2 ≤ m ≤ℵ0.

6.2 Properties of Connected Spaces

A brief historical note: The finite case of Theorem 5 was proven by David van Dantzig
(1900–1959) in 1930.

We start with a property that is very easy to prove, but, nonetheless, is very useful.

K12146_C006.indd 128 2/12/15 4:40 PM

 6.2 Properties of Connected Spaces   ◾    129  

Proposition 1. If X is connected and f : X → Y is continuous then f ( X ) is a connected subset of Y.

Proof. Suppose f ( X ) is disconnected and A, B is a separation of f ( X ). Then, f −1 ( A),

f −1 ( B) is a separation of X: these two sets are disjoint (since A and B are disjoint), their
union is X, they are nonempty (since A and B are nonempty) and, finally, f −1 ( A) and
f −1 ( B) are open since f is continuous and since A and B are open.

Corollary 2. Connectedness is a topological property.

Corollary 3. (The Intermediate Value Theorem) If f : [a, b] →  is a continuous function

and x is between f (a) and f (b) , then there is a number c ∈ [a, b] such that f (c ) = x .

Proof. Let x be between f (a) and f (b), and suppose that there is no c ∈ [a, b] such that
f (c ) = x . Then {(−∞, x ) ∩ f ([a, b]),( x , ∞) ∩ f ([a, b])} is a separation of f ([a, b]) contradict-
ing Proposition 1 and the fact that [a, b] is a connected subset of .

We illustrate the usage of Proposition 1 with a couple of examples.

Example 1
The subspace X = {( x , y ) : x 2 + y 2 = 1} ∪ {( x , y ) : x 2 + ( y − 2)2 = 1} of  2 (a bouquet
of two circles) is not homeomorphic to the subspace S1 = {( x , y ) : x 2 + y 2 = 1} of  2 .
Suppose otherwise and let f : S1 → X be a homeomorphism.

S1
f
(0, 1)

P
Illustration 6.8

So, there is a point P of S1 such that f ( P ) = (0,1). The space S1 \ {P} is con-
nected (since it is homeomorphic to  via a stereographic projection), and the
restriction of f to S1 \ {P} is a homeomorphism from S1 \ {P} onto X \ {(0,1)}. By
Proposition 1, we must have that f (S1 \ {P}) = X \ {(0,1)} is also connected, which is
clearly false. So, we have a contradiction. ☐

Example 2
The semi-open interval (0,1] is not homeomorphic to the open interval (0,1) (both as
subspaces of ), since (0,1]\ {1} is connected while (0,1)\ { p} is disconnected for every
p ∈(0,1). ☐

K12146_C006.indd 129 2/12/15 4:41 PM

 130   ◾    Connected Spaces and Path Connected Spaces

Example 3
There does not exist a continuous one-to-one mapping from the closed disk D2 onto the
circle S1, and so D2 is not homeomorphic to a circle. Otherwise, if f : D 2 → S1 is con-
tinuous, one-to-one, and onto, then f D 2 \{ x , y } , where x and y are any two distinct points
in D2, is also continuous, one-to-one, and onto S1 \ { f ( x ), f ( y )}. But this could not be
true, since D 2 \ {x , y} is connected, while f (D 2 \ {x , y}) = S1 \ { f ( x ), f ( y )} is not. ☐

Continuous mappings can be used to give yet another criterion for connectedness.

Proposition 4. A space X is connected if and only if there is no continuous, onto mapping

f : X → {0,1}, where {0,1} is equipped with the discrete topology.

Proof. ⇒ Suppose there is a continuous, onto mapping f : X → {0,1}. Then f −1 (0) is both
open and closed (since {0} is both open and closed and since f is continuous), nonempty
(since f is onto), and not equal to X (again because f is onto, so that some points are mapped
to 1). It follows by Theorem 4, Section 6.1, that X is disconnected.
⇐ Suppose X is disconnected and A, B is a separation of X. Then the mapping
0 if x ∈ A
f : X → {0,1} defined by f ( x ) =  is continuous and onto.
 1 if x ∈B
Connectedness is preserved in product spaces; this will require slightly more effort to justify.

Theorem 5. Suppose Xi , i ∈I are connected spaces. Then ∏ Xi is also connected.

i∈I

Proof. In order to (literally) see what happens we will first prove the case of the product of
two spaces. Then we will consider arbitrarily many spaces.
Suppose X and Y are connected and consider the product space X × Y . Obviously,
X × Y = ∪ X × { y}. Each of the subspaces X × { y} is homeomorphic to X. It follows from
y ∈Y
Corollary 2 that the spaces X × { y}, y ∈Y , are connected. By a symmetric argument, for
any fixed x 0 , the space {x 0 } × Y is connected. Observe now that for every y ∈Y , we
have ({x 0 } × Y ) ∩ ( X × { y}) = {( x 0 , y )} ≠ ∅ . So, by Proposition 6(b) in 6.1, the space
X × Y = ∪ X × { y} is also connected. Illustration 6.9 summarizes the argument.
y ∈Y
X×Y

X × {yk }

X × {yi}

X × {yj}

{x0} × Y

Illustration 6.9

K12146_C006.indd 130 2/12/15 4:41 PM

 6.2 Properties of Connected Spaces   ◾    131  

A straightforward induction shows that finite products of connected spaces are connected.
Now, consider the product ∏ Xi of connected spaces, where the set I is infinite. Fix an ele-
i∈I
ment x = ( xi )i ∈I in ∏ Xi . For every finite subset T of I, denote C(T ) = ∏ Ai , where Ai = {xi }
i∈I i∈I
if i ∉T , and Ai = Xi if i ∈T . Then C(T) is homeomorphic to ∏ Xi , and, by the first part
i ∈T
of the proof, it is connected. Since x ∈ ∩ C(T ) , it follows by Proposition 6(a) in 6.1 that
T ⊂ I
finite

Y = ∪ C(T ) is connected. The set Y is dense in ∏ Xi (Exercise 11, Section 5.2). It follows
T ⊂ I i∈I
finite

from Theorem 5 in 6.1 (or Exercise 12 in 6.1) that ∏ Xi is connected.

i∈I

Example 4

An immediate consequence of this theorem and of Proposition 1 in 6.1 is that the

product space ∏  for any index set I is connected. In particular, the Euclidean
i∈I
spaces n and the Hilbert space are connected. ☐

The analogue of Theorem 5 for the box topology fails.

Example 5: Box Products of Connected Spaces Need Not Be Connected

Consider the set product ∏  and recall the formal definition of the set products:
i ∈
every element of ∏  is a function  → . Some of these functions are bounded
i ∈
(their graphs are between two horizontal lines), and some are unbounded. Let B be
the subset of ∏  containing all bounded functions and let U be the subset of ∏ 
i ∈ i ∈
containing all unbounded functions. Obviously, B ∩ U = ∅ and B ∪ U = ∏  . We
i∈
will now (literally) see that both B and U are open in the box topology over ∏  ,
box i ∈
so that the box space ∏  is disconnected.
i ∈

f f

  
    Illustration 6.10 A bounded function.   Illustration 6.11 An unbounded function.

If f is a function in B, then ∏ ( f (i ) − ε, f (i ) + ε) is an open neighborhood of f in the box space

box i∈
∏  (because each of the intervals ( f (i ) − ε, f (i ) + ε) is open in ). Since f is bounded it
i ∈

K12146_C006.indd 131 2/12/15 4:41 PM

 132   ◾    Connected Spaces and Path Connected Spaces

follows that every function in ∏ ( f (i ) − ε, f (i ) + ε) is bounded (see Illustration 6.12, left).

i∈
So, every point of B is an interior point, and so B is open. Similar argument (see Illustration
6.12, right) shows that U is open.

f–ε
f–ε

f
f
f+ε
f+ε

Illustration 6.12 Every function in the open neighborhood (in the box topology) of a bounded
(unbounded) function is a bounded (respectively, unbounded) function. The neighborhoods of
the functions are shaded. (Warning: Do not misinterpret the above illustration. The elements
in the shaded neighborhoods are the functions with graphs within the shaded neighborhood,
not merely points in the shaded part of  2.) ☐

Exercises
1. Show that in each case below X is not homeomorphic to Y.
(a) X is the subspace of  made of all irrational numbers, Y is .
(b) X is , Y is  2.
(c) X is the letter X, Y is the letter Y (both considered as subspaces of  2).
(d)

X= Y=

(e)

X= Y=

(f)

X= Y=

Illustration 6.13

2. Let X be obtained from  ×[0,1] and  ×[2,3] by gluing the pairs of points (z, 1) and
( z ,2), z ∈. Show that X is not homeomorphic to a (closed or open) disk in  2 .

K12146_C006.indd 132 2/12/15 4:41 PM

 6.3 Path Connected Spaces   ◾    133  

3. Let A be a component of X and let f : X → Y be continuous. Show through a counter­

example that f (A) need not be a component of the subspace f (X) of Y.
4. Let { Xi : i ∈I } be a class of pairwise disjoint spaces such that ∏ Xi is a connected space

{ }
i∈I
and let ( xi ) ∈ ∏ Xi . The partition {{ xi : i ∈I }} ∪ {x } : x ∈∪ Xi , x ≠ xi for every i ∈I
i∈I i∈I
of ∪ Xi defines an equivalence relation ~ on ∪ Xi . Show that the quotient space
i∈I i∈I
⊕ Xi
i∈I~ is connected.
5. Let X and Y be spaces, and let A ⊂ X , B ⊂ Y . Show that A is a component in X and B
is a component in Y if and only if A × B is a component in the product space X × Y .
6. Let X be connected and consider the subset D = {( x , x ) : x ∈ X } of X × X. Show that
D is connected.
7. Use properties of connectedness to show that there is no continuous mapping  → 
that sends the rational numbers to irrational numbers and irrational numbers to ratio-
nal numbers. (Another solution was given in Example 4, Section 3.5.)
8. Show that if A is a countable subset of  2 , then  2 \A is connected.
9. Show that if f :  2 →  is continuous and if both 1 and –1 are images under f of some
points in  2 , then f −1 (0) is uncountable.
10. Show that if f :[0,1] → [0,1] is continuous and onto then f has a fixed point.
11. An involution f : X → X is a continuous map such that f ° f = identity. (So, involu-
tions are self-inverse homeomorphisms.) Show that if f :  →  is an involution, then
f must have a fixed point.
12. Let A = {0} ∪ [1,2] ∪ {3} and B = [0,1] ∪ {2} ∪ {3} (as subspaces of  ). Show that A
and B are homeomorphic but there is no homeomorphism from  onto  taking
A onto B.
13. (a) Show that if f : S1 → S1 is antipode-preserving, then f is onto. (A mapping
f : S1 → S1 is antipode preserving if f ( x ) = − f (− x ) for every x ∈S1.)
(b) Show that if f : S1 → S1 is an embedding than it is a homeomorphism.
14. Let U be a subset of  2 with a nonempty interior. Show that there is no embedding
f : U → S1.

6.3 Path Connected Spaces

We start by introducing the main notions of this section; we will encounter them fre-
quently, especially in Part 2 of this book.
A path in a space X is a continuous mapping α :[0,1] → X . The image α([0,1]) will some-
times be called an arc in X. The point α(0) is called the initial point of the path α, and the

K12146_C006.indd 133 2/12/15 4:41 PM

 134   ◾    Connected Spaces and Path Connected Spaces

point α(1) is the terminal point α. We also say that α starts at α(0) and ends at α(1), or,
more succinctly, that α is a path from α(0) to α(1). The points α(0) and α(1) are jointly
called the endpoints of the path α. A space X is path connected if for every two points in X
there is a path starting at one of these two points and ending at the other.

X
α(0)
α(1)
α

0 1

Illustration 6.14 A path in X from α(0) to α(1).

The main reason we choose the closed unit interval [0, 1] to be the domain of every path α is
convenience. Occasionally we will encounter continuous mappings λ :[a, b] → X . The path
α λ :[0, 1] → X associated to the mapping λ is defined by α λ (t ) = λ ((b − a)t + a ) .
From now on, and unless otherwise stated, we will use the letter I to designate the
unit interval [0,1] (as a subspace of ). We will repeat this convention several times as
we go.
Since all closed intervals are homeomorphic, the existence of a path in a space X starting
at x and ending at y is equivalent to the existence of a continuous map g from any closed
interval [a, b], a < b into X such that g (a) = x and g (b) = y .

Example 1
The Euclidean space  is path connected. Given any a, b ∈ the mapping α :[0,1] →  ,
defined by g (t ) = a + t (b − a) is a path from a to b. Similar argument gives that n is
path connected for every n ∈+. ☐

Example 2: A Space That Is Connected but Not Path Connected

Illustration 6.15 depicts a space called the topological comb: it consists of the
vertical line segment {( x , y ) : x = 0, 0 ≤ y ≤ 1} and the horizontal line segments {( x , y ) :
y = 1n , 0 ≤ x ≤ 1}, n = 1, 2, 3,…, and {( x , y ) : y = 0,0 ≤ x ≤ 1}, considered as a subspace
of  2 (Illustration 6.15). Illustration 6.16 is the same space, with the point A at the
lowest horizontal segment deleted (we denote the resulting space by X). Its topological
structure is in many ways similar to the structure of the space described in Example
5, Section 6.1. Specifically, the argument used there to show that the space in that
example is connected could be used almost without any change to show that the space
X is also connected. Is the space X path connected? Our intuition suggests a negative

K12146_C006.indd 134 2/12/15 4:41 PM

 6.3 Path Connected Spaces   ◾    135  

answer: there should be no way to continu-

ously bridge the gap at the point A with a
path starting at B and ending at C. We will,
however, resist the temptation to use obvi-
ousness as an argument in this case and we
will prove that the space X is indeed not path
connected. Obvious claims are the ultimate
test of a theory; if they are provable then we
can say that the theory indeed performs
the way we want it to perform. Illustration 6.15 A topological comb.

We will show that there is no path in X starting at B and ending at C.

( )
Suppose A = (a,0), a > 0, and consider the points a, 1 n , n = 1,2,3,… (above A; see
( )
Illustration 6.16). If a point a, 1 m is not in f ([0,1]), then no point to the right of
it (along the horizontal segment) is in f ([0,1]), for otherwise the pair of open sets
f −1 (U ), f −1 (V ) , U, and V, as in Illustration 6.17, would be a separation of [0, 1].

B A C

Illustration 6.16 A topological comb with one point deleted.

U
1/m

B A C

Illustration 6.17

( )
Suppose that only finitely many points a, 1 n are in f ([0,1]). Let m be the largest
( ) (
integer such that a, 1 m is in f ([0,1]); take any m if no point a, 1 n is in f ([0,1]). )
−1 −1
Then, f (U ), f (V ), where U and V are as in Illustration 6.18, would be a separation of
[0,1]. Note that both f −1 (U ) and f −1 (V ) are not empty since the former contains the point

K12146_C006.indd 135 2/12/15 4:41 PM

 136   ◾    Connected Spaces and Path Connected Spaces

f −1 (C ) while the latter contains the point f −1 ( B). Also note that [0,1] = f −1 (U ) ∪ f −1 (V )
(even though U ∪ V is not X).

1/m

B A C U

Illustration 6.18

(
Suppose now that infinitely many points a, 1 n are in f ([0,1]). Denote )
{( ) }
S = a, 1 n : n = 1,2,… . So the set f −1 (S ) is an infinite set in the closed (and bounded)
interval [0,1]. By the Bolzano–Weierstrass Theorem (Theorem 5, Section 3.2), there must
be an accumulation point P for f −1 (S ) and, since [0,1] is closed, that accumulation point
(
is in [0,1]. Since f is continuous, f (P) must be an accumulation point for f f −1 (S ) (this )
( )
is an easy exercise). However, f f −1 (S ) is a subset of the set a, 1 n : n = 1,2,… , {( ) }
and the latter set has no accumulation points in X (precisely because A is missing). We
have a contradiction. ☐

The last example shows that, in general, connectedness does not imply path connectedness.
However, the converse is true.

Proposition 1. If a space X is path connected, then it is connected.

Proof. Suppose X is disconnected and suppose that A, B is a separation of X. Since both A

and B are not empty, there are points P and Q in A and B, respectively.

X
A
B

α
α–1(A)
0 1
α–1(B)

Illustration 6.19 A separation of X would generate a separation of the interval [0,1].

There is no path α starting at P and ending at Q since otherwise α −1 ( A), α −1 ( B) would be a

separation of the interval [0,1].

K12146_C006.indd 136 2/12/15 4:41 PM

 6.3 Path Connected Spaces   ◾    137  

Next, we introduce an algebraic structure over the

set of paths. Let α :[0,1] → X and β :[0,1] → X . be two
paths in X such that α(1) = β(0). It is clear that con-
catenating these two paths as in Illustration 6.20 β
gives another path in the space. We make it for-
mal as follows: consider the mapping γ :[0,1] → X α
defined by 0 1
Illustration 6.20
 α(2t ) if 0 ≤ t ≤ 12
γ (t ) =  .
β(2t − 1) if 1 ≤ t ≤1
 2

We call it the product of the paths α and β, and denote it by αβ.

Proposition 2. The product of two paths is a path.

Proof. We only need to show that αβ is continuous. This follows from the gluing lemma with
the roles of f, g, A, and B in that lemma taken by α, β, 0, 1 2  , and  1 2 ,1, respectively.
The algebraic structure we have introduced so far (a partial groupoid) is rather weak.
We will expand it further in Part 2 of this book.
A subset A of a space X is a path connected subset if A is path connected as a subspace of X.

Proposition 3. Let Ai , i = 1,2, …, be path connected subsets of a space X such that

n
Ai ∩ Ai+1 ≠ ∅ for every i = 1,2, … . Then for every n = 1,2, …, ∪ Ai is path connected.
i =1

Proof. We use induction, with the first step being our assumption that A1 is path con-
k k +1
nected. Assume ∪ Ai is path connected, and let P and Q be two points in ∪ Ai . The only
i =1 k i =1
nontrivial case is when P ∈∪ Ai and Q ∈ Ak +1.
i =1

R
Q

Illustration 6.21
k
Since, by assumption, Ak ∩ Ak+1 ≠ ∅ there is a point R in Ak ∩ Ak+1. Since ∪ Ai is path
k i =1
connected, there is a path α in ∪ Ai from P to R. Since Ak+1 is path connected, there
i =1 k +1
is a path β in Ak+1 from R to Q. The product path αβ is then a path in ∪ Ai from P
i =1
to Q.

In the case of Euclidean spaces n, connectedness plus a bit more implies path
connectedness.

K12146_C006.indd 137 2/12/15 4:41 PM

 138   ◾    Connected Spaces and Path Connected Spaces

Proposition 4. Every connected and open subset of n is path connected.

Proof. Suppose A is a connected and open subset of n and let P and Q be two points in A.
Consider the collection O of all open balls W such that W ⊂ A. Since A is open, A = ∪ W .
W ∈O
Construct a sequence of open sets as follows: U1 is an open ball in O containing P, U2 is the
union of all open balls in O having nonempty intersections with U1, and, inductively, Un + 1 is
∞
the union of all open balls in O having nonempty intersections with Un. Denote VP = ∪U i .
i =1
Assume that there is an open ball in O not used in the construction of VP . Then the
union U of all such balls is a nonempty open set disjoint from VP , and A = U ∪ VP . That
contradicts the connectedness of A.
So, Q ∈VP , and so there is a sequence of balls Bi such that Bi ∩ Bi+1 ≠ ∅, P ∈ B1, and Q ∈ Bm
for some m. Since every ball in n is path connected, the claim of this proposition now
follows from Proposition 3.

Exercises
1. Let Z be a space and let X ⊂ Y ⊂ Z. Show that X is path connected as a subset of Z if
and only if it is path connected as a subset of the subspace Y of Z.
2. Show that every convex subset of n is path connected.

–2 –1 1 2

–1

–2
   Illustration 6.22

3. In Illustration 6.22 we show the subspace X of n consisting of the points on the spi-
ral r = 1 + 10 ϕ in polar coordinates, ϕ ∈[2π , ∞), together with the points on the circle
r = 1. Show that X is connected but not path connected.
4. True or false: If A is an open path connected subset of n then A is also path
connected.

5. (a) Let Ai , i ∈I , be path connected subsets of a space X and suppose ∩ Ai ≠ ∅. Show

i∈I
that ∪ Ai is a path-connected subset of X.
i∈I

K12146_C006.indd 138 2/12/15 4:41 PM

 6.4 Path Connected Spaces: More Properties and Related Matters   ◾    139  

(b) Let A0 and Ai , i ∈I , be path connected subsets of a space X and suppose A0 ∩ Ai ≠ ∅

( )
for every i ∈I . Show that A0 ∪ ∪ Ai is path connected.
i∈I

(c) Let Ai , i = 1, 2, …, be path-connected subsets of a space X such that Ai ∩ Ai+1 ≠ ∅

∞
for every i = 1, 2, …. Show that ∪ Ai is path connected.
i =1

6. Find a path-connected space that is not second countable.

7. An ambient path in a space X starting from P ∈ X and ending at Q ∈ X is a map-

ping h: X × I → X such that the restriction h X ×{t } is a homeomorphism for every t
in I, such that h X ×{0} : ( x ,0)  x for every x ∈ X, and such that h X ×{1} sends P to Q.
A space is ambient path connected if for every two points P and Q in X there is an
ambient path in X from P to Q.
(a) Show that every ambient-path-connected space is path connected.
(b) Show that the (path-connected) subspace of  2 depicted in Illustration 6.23 below
is not ambient path connected.

   Illustration 6.23

8. Let X be the subspace of  2 consisting of two intersecting line segments, such that
the intersection point is not an end point of any of the segments. Show that there is
no ambient path from the intersection point to any other point in X.

6.4 Path Connected Spaces: More Properties and Related Matters

A path component in a space X is a maximal (under inclusion) path connected subset
of X.
The path component of X containing a point x is the set of all points y in X for which
there is a path starting at x and ending at y.

K12146_C006.indd 139 2/12/15 4:41 PM

 140   ◾    Connected Spaces and Path Connected Spaces

Example 1 X
There are two path components in the topological
comb, with the point A = (a,0), 0 ≤ a < 1, deleted: the
line segment {(t ,0) : a < t ≤ 1} is one path component,
the rest of X is another path component.
☐
A
The following is an observation.
Illustration 6.24

Proposition 1. A space is path connected if and only if it has one path component.

Proof. Obvious.

Example 2: A Connected Space with Infinitely Many Path Components

Let Xi , i ∈ , be a copy of the interval [0, ∞) (thus we have  copies of [0, ∞)).
Assume these copies of [0, ∞) are pair-wise disjoint. Define a linear order on
Y = ∪ Xi as follows: the order within each single copy of [0, ∞) is the usual linear
i∈
order of real numbers, while for every i < j (usual linear order of real numbers),
each element of Xi is less than each element of Xj. The linear order we get in such
a way generates the order topology over Y. It is easy to check that Y is a linear
continuum. Consequently, it is a connected space (Proposition 1′, Section 6.1).

Xi Xj

Illustration 6.25 Each Xi is a copy of [0, ∞ ). The copies of Xi with larger index (with respect to
the usual order in ) are to the right and contain larger elements in Y.

We will now see that Y is not path connected. Take any distinct i , j ∈ with i < j
and suppose x ∈ Xi and y ∈ X j . Suppose there is a path α :[0,1] → Y from x to y. Then
every point z between x and y (in the linear order of Y) is in α([0,1]) for otherwise
α −1 ({w ∈Y : w < z }) and α −1 ({w ∈Y : w > z }) would be a separation of [0, 1].
Choose any open interval in each X k , i < k < j ; for example, choose a copy Z k in
X k, i < k < j , of the open interval (0,1). Since α is continuous, each α −1 ( Z k ) is open in
[0, 1], and since the elements of Z k are between x and y, the open sets α −1 ( Z k ) are not
empty. So, each of these sets α −1 ( Z k ), i < k < j , contains (as a subset) an open interval
in [0, 1]. Choose one such interval Uk for each α −1 ( Z k ). Since the sets Z k, i < k < j ,
are pairwise disjoint, so are the sets α −1 ( Z k ), i < k < j , and so are the intervals U k,
i < k < j.
Define a mapping f : (i , j ) → [0,1] (where (i, j) denotes the open interval in ) as
follows: for every k ∈(i , j ), choose any rational number qk in U k and set f (k ) = qk .
Since U k ’s are intervals, they contain rational numbers, and so f is well defined.
Since the intervals U k , i < k < j , are pairwise disjoint, the mapping f is one-to-one.

K12146_C006.indd 140 2/12/15 4:41 PM

 6.4 Path Connected Spaces: More Properties and Related Matters   ◾    141  

Xk Xl
( ) ( )
0 1 0 1

α–1

( ) ( )
0 qk ql 1
Illustration 6.26 f sends k to the rational number qk , and it sends l to the rational number ql.

So, summarizing, we got a one-to-one mapping f from the uncountable set of real
numbers in the interval (i , j ) into a subset of the countable set of rational numbers
in [0,1]. That is not possible (see Section 1.2). So, Y is not path connected.
In fact, we proved more: we showed that if a subset A of Y intersects (non-emptily)
both Xi and X j , i ≠ j , then it is not path connected. Consequently, the path components
of Y are the sets Xi , i ∈ , each of them a copy of [0, ∞). So, Y has uncountably many
path components and only one component. ☐

The next proposition tells us that path connectedness is preserved under continuous
maps (and so it is a topological property).

Proposition 2. If a space X is path connected and f : X → Y is continuous, then f ( X ) is a

path connected subset of Y.

Proof. Suppose P and Q are two points in f ( X ). Let p, q ∈ X be such that f ( p) = P and
f (q ) = Q .

X Y
q
f (X)
α p Q
f P

Illustration 6.27

Since X is path connected, there is a path α in X starting at p and ending at q. The composi-
tion f α is then a path in f ( X ) starting at P and ending at Q.

Example 3
Every quotient space of a path connected space is path connected since quotient maps
are continuous and surjective. ☐

Example 4
Suppose X is path connected and consider the diagonal D = {( x ) j ∈ J : x ∈ X } as a sub-
space of ∏ X . Then, D is path connected, too. It is clear that the mapping f : X → D
j∈ J

K12146_C006.indd 141 2/12/15 4:41 PM

 142   ◾    Connected Spaces and Path Connected Spaces

defined by f ( x ) = ( x ) j ∈ J is onto. In order to see that it is continuous take a stan-

dard basis neighborhood ∏ U j of ( x ) j ∈ J in ∏ X ; so all but finitely many U ’s are
j∈ J j∈ J
the entire space X, and the remaining {U j1 ,U j2 ,…,U jk } are open neighborhoods
k
of x in X. It is visible that f −1 ( ∏ U j ) = ∩ U j = ∩U ji , an open set. So, f is con-
j ∈J j ∈J i =1

tinuous. Since D = f ( X ) Proposition 2 applies, and D is path connected. ☐

Proposition 3. If X j , j ∈J are path-connected spaces, then so is ∏ X j .

j∈ J

Proof. The idea of the proof is outlined in Illustration 6.28, where we take J = {1,2}, and
where X1 and X2 are copies of [0, ∞).

X2
]

(α,β)(t) = (α(t),β(t))
[

(α,β)
[ ]
X1
β
α

Illustration 6.28

Formally, suppose X j , j ∈ J are path-connected spaces, and let P , Q ∈ ∏ X j . Then for every
j ∈J
fixed i ∈ J there is a path α i from pi ( P ) to pi (Q ) in Xi , where, as usual, pi denotes the
projection ∏ X j → Xi . Consider now the mapping ∏ α j :[0,1] → ∏ X j defined by
j ∈J j∈ J j∈ J

( )
∏ α j (t ) = ∏ α j (t ) . By Proposition 2 in Section 5.2, ∏ α j is a (continuous) path, and it
j ∈J j ∈J j∈ J
starts at P and ends at Q.

In the rest of this section we explore the notion of travel in a space through some natural
generalizations of path connectedness. We will not use these generalizations in the sequel.
However, we will come close to them in Section 10.1 where we introduce ambient isotopy.
In a path-connected space X a point can travel between any two positions in the
space. If path connectedness emulates movement of points, then we need a suitable gen-
eralization to correspond to movement of other objects in a space. The next definition
does that to a certain degree: we allow the starting object to be deformed along a path
as long as it does not change up to a homeomorphism.

K12146_C006.indd 142 2/12/15 4:42 PM

 6.4 Path Connected Spaces: More Properties and Related Matters   ◾    143  

Let X be a space and S a subspace of X. We say that X is S-connected if for every two embed-
dings f : S → X and g : S → X there is a continuous map F : S × [0,1] → S (where S ×[0,1] is
equipped with the product topology) such that F (s ,0) = f (s ), F (s ,1) = g (s ) for every s in S, and
such that F (s , t0 ): S × {t0 } → X is an embedding for every t0 ∈I.
As in the definition of path connectedness, the usage of the interval I = [0,1] in
the last definition is for the purpose of standardizing the domain and is not essen-
tial: we could replace it by any other closed intervals without affecting the intended
meaning of the definition. We may view I in this context as denoting time, so that traveling
from position F (s ,0) = f (s ), s ∈S to position F (s ,1) = g (s ), s ∈S is always executed in one
unit of time.

The definition of S-connectedness generalizes the notion of path connectedness: if S is a

one-point space, then being S-connected is the equivalent to being path connected.

Example 5
In Illustration 6.29 we depict a hollow box B with a point-thin slit at the top face.
Keeping the notation used above, we take X =  3 \ B.

X Y Z

Illustration 6.29

The space Y is a stickman with a flat head, and Z is a stickman with a spherical head,
both as subspaces of X. The space X is Y-connected, but not Z-connected. The first
claim is clear, and the second is equivalent to the statement that there is no subspace
V of X such that V = V1 ∪  V2 ∪
 V3 , V1 is inside B, V2 is in the slit (i.e., V2 is homeomor-
phic to a subspace of a line segment), and V3 is outside B, and such that V ≅ S 2 . This
merits attention (Exercise 10). ☐

Example 6
Let X be a cylinder (see Illustration 6.30). Then X is D2-connected (where we recall
that D 2 = {( x , y ) ∈  2 : x 2 + y 2 ≤ 1}). However, X is not S 1-connected ( S1 = {( x , y ) ∈ 2 :
x 2 + y 2 = 1} ). There is no way to travel from the homeomorphic image S11 of the
circle S1 to the homeomorphic image S21 (Illustration 6.30). A simple formal justifica-
tion of this statement will come out of the theory we will develop in Part 2.

K12146_C006.indd 143 2/12/15 4:42 PM

 144   ◾    Connected Spaces and Path Connected Spaces

D2

D1

S21

S11

Illustration 6.30 It is not possible to travel from position S11 to position S21 as shown in this picture.
This is true even if we drop the requirement that at each moment of time the image of S1 on
the cylinder is homeomorphic to the original. ☐

S-connectedness is a topological property but it is not necessarily preserved under con-

tinuous mappings.
Finally, we will model another type of movement, whereby we deform the whole space
X in order for a given portion of it to reach a predetermined position.
Let X be a space and let S be a subspace of X. We say that X is isotopically
S­- connected, if for every two embeddings f : S → X and g : S → X there is a continuous
mapping F : X × I → X such that F X ×{t } is a homeomorphism for every t ∈I, such that
F ( x , 0) = x for every x ∈ X , and such that F ( f (s ),1) = g (s ) for every s ∈S. (The map-
ping F here is called an ambient isotopy within X from f (S ) to g (S ); we will encounter
ambient isotopies in Part 2 of this book.)

Proposition 4. If a space X is isotopically S­-connected, then it is S-connected.

Proof. Exercise 13.

Example 7
The double cone in Illustration 6.31 is neither S1-connected nor isotopically
S1-connected. These claims constitute Exercise 14. The space X depicted in Illustration
6.32 (obtained by deleting two open disks on two spheres and then joining the
boundaries of these disks with a cylindrical surface) is both S1-connected and iso-
topically S1-connected. Can you visualize that claim? Finally, the space Y depicted
in Illustration 6.33 (obtained by joining two solid balls with a cylindrical surface) is
S1-connected but not isotopically S1-connected (Exercise 15).

K12146_C006.indd 144 2/12/15 4:42 PM

 6.4 Path Connected Spaces: More Properties and Related Matters   ◾    145  

Illustration 6.31 A double-cone. Illustration 6.32 The space X.

Illustration 6.33 The space Y. ☐

Exercises
1. Describe the path components in  with the half open interval topology.
2. (a) Find a connected subset A of  2 such that ∂A has three path components.
(b) Find a connected subset A of  2 such that ∂A has infinitely many path
components.
3. Show that path components of open subsets of  2 are open.
4. Show that each path component is contained in a component.

5. Consider the set  2 equipped with the post office metric (Section 2.3), where we
assume that the designated point (the post office) is the origin. Denote the metric
space generated by that metric by X.

K12146_C006.indd 145 2/12/15 4:42 PM

 146   ◾    Connected Spaces and Path Connected Spaces

(a) Which of the following three mappings from [0,1] into X is a path in X from the
point (1,0) to the point (0,1)?


 (1 − 2t ,0) if 0 ≤ t ≤ 12
α(t ) = 
 (0,2t − 1) if 1 ≤ t ≤1
 2

β(t) = (1, 0) + t(−1, 1)

 (1,0) if t=0

γ (t ) =  (0,0) if 0<t <1
 (0,1) if t =1


(b) Describe the path components of X.

6. The ordered square topology over the set {( x , y ) ∈ 2 : 0 ≤ x ≤ 1, 0 ≤ y ≤ 1} is generated
by the lexicographic linear order < defined as follows: ( x , y ) < ( z , w ) if y < w or if
y = w and x < z . Show that the ordered square is connected but not path connected.
7. Show that every co-finite space over an infinite countable set is connected but not
path connected.
8. Show that there is no continuous mapping f from the closed disk D 2 = {( x , y ) :
x 2 + y 2 ≤ 1} onto the circle S1 such that for every x ∈S1 the set f −1 ( x ) is finite.
9. Show that there are no two disjoint nonempty subsets A and B of , and a bijection
f : A → B, such that the space  f (obtained by identifying A and B along f) is homeo-
morphic to .
10. Show that the space X in Example 5 (and Illustration 6.29) is not Z-connected.
11. Find a space X that is
(i) Path-connected but not disk-connected.
(ii) Circle-connected but not sphere-connected.
(iii) Sphere-connected but not torus-connected.
12. Given a space S, a space X is weakly S-connected if for every two points x, y in X, there
exist two embeddings f : S → X and g : S → X such that x ∈ f (S ) and y ∈ g (S ), and
there exists a continuous map F : S × [0,1] → X such that F (s ,0) = f (s ), F (s ,1) = g (s )
for every s in S, and such that F (s , t ): S × {t } → X is an embedding for every fixed t. Find
a subspace X of  3 such that X is weakly circle-connected but not circle-connected.
13. Prove Proposition 4.
14. Prove that the double cone in Illustration 6.31 is not S1-connected, and not isotopi-
cally S1-connected.

K12146_C006.indd 146 2/12/15 4:42 PM

 6.5 Locally Connected and Locally Path Connected Spaces   ◾    147  

15. Prove that the space Y depicted in Illustration 6.33 is S1-connected, but not isotopi-
cally S1-connected.
16. Let X be the space obtained from the cylinder S1 × [0, 1] by deleting at most one
point from each S1 × {t } , t ∈[0, 1]. Is X necessarily path connected? [Remark: I have
no solution; a justified answer may require more that what we have here.]

6.5 Locally Connected and Locally Path Connected Spaces

A space X is locally connected at x if for every open neighborhood U of x, there is an open

connected neighborhood V of x, such that V ⊂ U . A space X is locally connected if it is
locally connected at every x ∈ X .

Example 1
The space  is locally connected since open intervals are connected and since every
open neighborhood of a point contains an open interval containing the same point.
Similar argument (using the fact that (open) balls are connected) verifies the local
connectedness of n (n ∈+ ). ☐

Example 2

x V
U

Illustration 6.34 The space to the left (the subspace of  2 made of the lines y = n, n ∈) is
locally connected. The topological comb to the shown to the right (the subspace of  2 made
of the lines y = 1n , n ∈+ , the line y = 0, and of one vertical line as shown) is not locally
connected. ☐

Example 2 implies that connectedness does not imply local connectedness.

Our definition of local connectedness may look somewhat peculiar. However, it is per-
fectly natural: the condition that there are open connected sets within arbitrary open sets
guarantees the existence of arbitrarily small open connected neighborhoods of points.
If we define a space to be locally connected when every point has an open connected

K12146_C006.indd 147 2/12/15 4:42 PM

 148   ◾    Connected Spaces and Path Connected Spaces

neighborhood, then the topological comb itself would be locally connected (since the whole
space is a connected open neighborhood)—not meeting our expectations of the notion.
Sometimes it is easier to access a space through its local topological properties. Prime
examples are manifolds, as well as covering spaces (to be introduced later). We will need the
following two propositions when we continue our study of manifolds and covering spaces.

Proposition 1. A space X is locally connected if and only if the components of open sub-
spaces of X are open in X.

Proof.
⇒ Let X be a locally connected space, let U be an open subspace of X, and let C be a compo-
nent of U. We will show that ∂(C ) ∩ C = ∅ (where the boundary is taken in U) and it will
follow that C is open in U (Theorem 9, Section 3.2). Suppose x ∈∂(C ) ∩ C . Since X is locally
connected, and since U is an open subset of X, there is an open connected subset V of X,
such that x ∈V ⊂ U . The set V is clearly open in U, too, and so, since it is an open neighbor-
hood of the boundary point x, it must contain points in U \C. So, C is a proper subset of the
set C ∪ V .

V
x

Illustration 6.35 If C has a boundary point x then C ∪ V would be a larger connected set in
U, contradicting the assumption that C is a component.

On the other hand, it follows from Proposition 6(a), Section 6.1 that C ∪ V is also con-
nected, contradicting our assumption that C is a component of U. So, ∂(C ) ∩ C = ∅, and
thus C is open in U.
By the definition of subspace topology, C = U ∩ W for some open subset W of X. So,
since both U and W are open in X, so is C.
⇐ Suppose that the components of open subspaces of a space X are open in X. Take a
point x in X and an open neighborhood U of x. Let Cx be the component of U (viewed as a
subspace of X ) that contains x. It follows from our assumption that Cx is open in X. So, X is
locally connected.

A space X is locally path connected at x ∈ X if for every open neighborhood U of x, there

is an open path connected neighborhood V of x, such that V ⊂ U . A space X is locally path
connected if it is locally path connected at every x ∈ X .

K12146_C006.indd 148 2/12/15 4:42 PM

 6.5 Locally Connected and Locally Path Connected Spaces   ◾    149  

The first of the two spaces shown in Example 2 is visibly locally path connected, and the
second is visibly not locally path connected.
Since path connectedness implies connectedness, if a space is locally path connected then it
is locally connected. The converse is not true. We justify the last claim with the next example.

Example 3
Start with the linear continuum of Example 2, Section 6.4, consisting of  pair-
wise disjoint copies of Xi = [0, ∞), and modify it as follows: replace the index
set  with the open interval (0, ∞), so that we are dealing with the ordered set
Y = ∪ Xi (with the inherited linear ordering). Now we add a point y to the left
i∈(0,∞ )
of all of the elements in Y ( y < x for every x ∈Y ) thus obtaining a new linearly
ordered set Y ′ = Y ∪ { y}. The order topology over Y′ is obviously locally connected
and locally path connected at every point x ≠ y .
Since Y′ is a linear continuum, so is every interval [ y , b), b ∈Y . Since the set
of all such intervals is a local basis at y, and since every linear continuum is con-
nected (Proposition 1ʹ in 6.1), the space Y′ is locally connected at y. However, Y′ is
not locally path connected at y because every open neighborhood of the point y is
not path connected. Specifically, every open neighborhood of y contains x ∈ Xi ,
i ∈(0, ∞), and there is no path connecting y and x. For a detailed justification of the
last claim it suffices to repeat the argument of Example 2, Section 6.4. ☐

Proposition 1 has an analogue for locally path-connected spaces.

Proposition 2. A space X is locally path connected if and only if the path components of
open subspaces of X are open in X.

Proof. Similar to the proof of Proposition 1, so we leave it as an exercise.

We now know that path connectedness implies connectedness, and that the converse is
false. However …

Proposition 3. Every connected and locally path connected space is path connected.

Proof. Suppose X is connected and locally path connected. Let Ci , i ∈J, be the set of all
path components of X. By Proposition 2 each of Ci is open in X. If J > 1 then {Ci , Cic } is a
separation of X, contradicting our assumption that X is connected. So, J = 1 and thus X
is path connected.

K12146_C006.indd 149 2/12/15 4:42 PM

 150   ◾    Connected Spaces and Path Connected Spaces

Example 4: Local (Path) Connectedness Is Not a Continuous Invariant

The space depicted below is an image of the locally connected and locally path-con-
nected space (0,1], but it is not locally connected or locally path connected.

sin x1 , 0 < x ≤ 0.5

f (0.5)

f (1)

Illustration 6.36 The mapping f sends (0,1] into  2 as shown in this illustration,
where f (1) = (0,0). Every sufficiently small open neighborhood of f (1) = (0,0) has infinitely
many (path) components, and so this space is not locally (path) connected at that point. ☐

Example 5: Local (Path) Connectedness Is Not Always Preserved in Product Spaces

Let I be an infinite set and let each space Xi , i ∈I , be a copy of {0,1} with the discrete
topology. Each of the spaces Xi is locally connected and locally path connected, but the
space X = ∏ Xi is not locally connected (and, so, not locally path connected, too). In
i∈I
fact, we show more: that every nonempty open subset of X is disconnected. If suffices
to show that every member of the standard basis of X is disconnected. Standard basis
sets are of the type V = ∏ Vi , where all but finitely many Vi are Xi. Since I is infinite, there
i∈I
exists k ∈I such that Vk = X k . Set V 1 = ∏ Vi1, where Vi1 = Vi for every i ≠ k and Vk1 = {0} ,
i∈I
and setV 2 = ∏ Vi 2 , whereVi 2 = Vi for every i ≠ k and Vk2 = {1} . Then V = V 1 ∪ V 2, and the
i∈I
sets V 1 and V 2 are open, nonempty, and disjoint. ☐

Local (path) connectedness is preserved in finite product spaces (Exercise 7).

Exercises
1. Show that each of the following spaces is both locally connected and locally path
connected.
(i) Indiscrete space
(ii) Discrete space
(iii) Half-disk topology.

K12146_C006.indd 150 2/12/15 4:42 PM

 6.5 Locally Connected and Locally Path Connected Spaces   ◾    151  

2. Show that the following spaces are not locally connected and not locally path
connected.
(i) Sorgenfrey topology over .
(ii) The Fortissimo space over an uncountable set X ∪{ p}, defined as follows: a set
U ⊂ X is open if either U c is countable or if p ∈U c.
3. Prove that if U is an open subset of  2 , then the components and path components
of U coincide.
4. Prove Proposition 2.
5. (a) Show that local (path) connectedness is not hereditary.
(b) Show that if X is a locally (path) connected space, and A is an open subspace of
X, then A is also locally (path) connected.
6. Prove that if a space X is locally path connected then the path components of X are
both open and closed in X.
7. Show that if I is finite and each of the spaces Xi is locally connected (locally path
connected), then the space ∏ Xi is locally connected (locally path connected,
i∈I
respectively).
8. Show that every co-finite space over a countable set is locally connected but not
locally path connected.

9. Let X ~ be a quotient space, and let q : X → X ~ be the quotient map.

(a) Show that if C is a component of X ~ then q −1 (C ) is a union of components of X.
[Hint: Show that for every x ∈q −1 (C ) the component of X containing x is a subset
of q −1 (C )].

(b) Show that if X is locally connected then X ~ is locally connected.

K12146_C006.indd 151 2/12/15 4:42 PM

This page intentionally left blank
 Chapter 7

Compactness and
Related Matters

7.1 Compact Spaces: Definition

A brief historical note: Heinrich Heine (1821–1881) used finite subcovers in a paper (1872)
on uniformly continuous functions. Émile Borel (1871–1956) proved in 1894 that countable
open covers of closed intervals have finite subcovers. Borel’s paper is considered to be the
origin of the idea of compactness. The general case was settled in 1900 by Arthur Schönflies
(1853–1928) and in 1898 (published in 1904) by Henri Lebesgue (1875–1941). The outcome is
the Heine–Borel theorem (Theorem 4). The criterion for compactness given in terms of closed
sets with the finite intersection property (Proposition 5) was introduced in 1921 by Leopold
Vietoris (1891–2002; 111 years!). The definition of compact spaces we use here was introduced
(under the name of bicompact spaces) in the period 1920–1923 by Pavel (Paul) Alexandroff,
and Pavel (Paul) Urysohn (1898–1924).

We remind the reader of our convention that, unless otherwise stated, n stands for the
Euclidean topological space.
Consider for a moment our blueprint model, the Euclidean space . Images of closed
subsets of  under continuous mappings  → X need not be closed when X is a met-
ric space. For example, the mapping f = in  tan −1 ( x ) , where in :  →  2 is an embedding
defined by in( x ) = ( x , 0), sends  (closed within itself) to the open interval (−π , π ), and
the latter is not closed in  2 . However, images of closed and bounded subsets of  under
continuous mappings  → X are closed and bounded:

Proposition 1. If A is a closed and bounded subset of , and if f :  → X is a continuous

mapping into a metric space X, then f (A) is a closed and bounded subset of X.

153

K12146_C007.indd 153 2/13/15 2:10 PM

 154   ◾    Compactness and Related Matters

Proof.
f

Illustration 7.1 If there is a bounded set A such that f (A) is not bounded, then there is a con-
vergent sequence that is mapped to a divergent sequence. The latter is not possible, since
continuous maps preserve convergence of sequences. The details are given below.

We assume that A is closed and bounded and that f :  → X is a continuous mapping.

Suppose f (A) is not bounded. Then there is a sequence ( yi )i∞=1 of elements in f (A) such that
yi ∉ B(a, i ), where a is some fixed element in X and i ∈+ . It follows that ( yi )i∞=1 diverges.
However, if xi ∈ f −1 ( yi ) , then ( xi )i∞=1 is a sequence of infinitely many elements in the
bounded set A, and so it must have a least upper bound b in . So, there is a subsequence
( zi )i∞=1 of ( xi )i∞=1 that converges to b (see also Exercise 20, Section 3.2). Since ( zi )i∞=1 converges
to b and f is continuous, ( f ( zi ))i∞=1 converges to f (b) (Proposition 2 in 3.5). On the other
hand, ( f ( zi ))i∞=1 is a subsequence of the sequence ( yi )i∞=1 and so it must diverge. (We are not
claiming here that subsequences of divergent sequences must diverge.) Contradiction. So,
f (A) is bounded.
Next, we prove that f (A) is closed by showing that it contains all of its accumulation
points. Suppose y is an accumulation point for f (A). Then there is a sequence ( yi )i∞=1 of
elements in f (A) converging to y (Exercise 20, Section 3.2). Choose ( xi )i∞=1 as in the first
part of the proof ( xi ∈ f −1 ( yi )). Since A is bounded there is a subsequence ( zi )i∞=1 of ( xi )i∞=1
converging to some x in  (Exercise 19 in 3.2). Since A is closed, x must be in A. Then,
( f ( zi ))i∞=1 converges to f (x). However, since ( f ( zi ))i∞=1 converges to y, and since every con-
vergent sequence in a metric space converges to only one point, it follows that y = f ( x )
and so y ∈ f ( A).

We proved in Proposition 1 that bounded and closed sets have an interesting property.
Later, we will show other interesting and important properties of such sets. However, our
immediate goal is to find a generalization of the notion of closed and bounded sets in the
setting of general topological spaces, so that at least Proposition 1 holds. Since the notion
of boundedness makes sense only within metric spaces, our generalization will not be
straightforward.
Let Y be a subspace of a space X. A cover of Y in X is a family O = {Oi : i ∈ I } of sub-
sets of X such that Y ⊂ ∪ Oi . An open cover of Y in X is a family O = {Oi : i ∈I } of open
i∈I
subsets of X such that Y ⊂ ∪ Oi . A subcover of a cover O of Y is a subset of O that is
i∈I
still a cover of Y.
Here is an important definition: A subspace Y of a space X is compact if every open
cover of Y in X has a finite subcover.

K12146_C007.indd 154 2/13/15 2:10 PM

 7.1 Compact Spaces: Definition   ◾   155  

X X

Y Y

Illustration 7.2 Y is a compact subspace of X if every cover made of open sets in X allows a
finite subcover.

The reference to the space X in the definition of compactness is not important in the fol-
lowing sense:

Proposition 2. Suppose X is a subspace of Z. A subspace Y of the space X is compact if and

only if it is compact as a subspace of Z.

Proof. We leave it as an exercise (Exercise 3).

In particular, a subspace Y of a space X is compact as a subspace of X if and only if it is

compact as a subspace of itself.
In view of the last proposition there is not much danger in simplifying the terminology:
rather than using “cover of Y in X,” we will simply say “cover of Y,” and instead of “open
cover of Y in X,” we will say “open cover of Y.” Finally, rather then saying that “Y is a com-
pact subspace of a space X,” we will often simply say that “Y is compact.”
The next theorem shows that, in the setting of the space  , compactness is equivalent
to being bounded and closed, and so it is a reasonable generalization of that notion. This is
the primary reason why we state and prove Theorem 3.

Theorem 3. (Heine–Borel Theorem) A subset A of  is closed and bounded if and only if

it is compact.

Proof. ⇒ Let A be a nonempty, closed, and bounded subset of , and let U = {U j : j ∈ J }

be an open (in  ) cover of A. Suppose initially that A = [a, b1 ] and that U is the set of open
intervals {(c j , d j ) : j ∈ J }. As we will see by the end of this part of the proof, these restric-
tions are easy to overcome.
Since b1 ∈ A, and since U covers A, the set C1 = {c j : j ∈ J , b1 ∈(c j , d j )} is not empty. If there
is some c j ∈C1 such that c j < a , then A ⊂ (c j , d j ) and {(c j , d j )} covers A. Otherwise c j ≥ a for
every c j ∈C1 . Denote b2 = inf C1 . The last assumption implies that b2 ≥ a , and since b2 < b1 ,
we have that b2 ∈ A. Since b2 = inf C1 , there is a nonincreasing sequence c j1 , c j2 ,… , c jn ,…

K12146_C007.indd 155 2/13/15 2:10 PM

 156   ◾    Compactness and Related Matters

of elements in C1 that converges to b2 . It follows that {(c j1 , d j1 ),(c j2 , d j2 ),… ,(c jn , d jn ),…} is
a countable subset of U that covers (b2 , b1 ]. Since b2 ∈ A , there is (ck , dk ) ∈U such that
b2 ∈(ck , dk ). Observe that no interval in U that contains b2 also contains b1 , since other-
wise b2 would not be the infimum of C1 . Consequently, dk ≤ b1. If dk = b1 , then for every
c j ∈C1, {(c j , d j )} ∪ {(ck , dk )} covers [b2 , b1 ]. On the other hand, if dk < b1 , then it has to be
that dk ∈(c jm , d jm ) for some m ∈+ , in which case we have that {(c jm , d jm )} ∪ {(ck , dk )} cov-
ers [b2 , b1 ] . If b2 = a , then in both cases we have found a finite subcover of A.
The remaining case is b2 > a . Repeat the procedure from the previous paragraph, with
the role of b1 now taken by b2 : we get b3 = inf {c j : j ∈ J , b2 ∈(c j , d j )} , and a subset of U
consisting of at most two intervals covering [b3 , b2 ]. As in the first step, none of the inter-
vals in U containing b3 also contains b2 . If b3 ≤ a , then the union of the two finite covers
of [b3 , b2 ] and [b2 , b1 ] is a finite cover of A. Otherwise b3 > a and we iterated the procedure.
In step k we get a subset of U consisting of at most two intervals, covering [bk +1 , bk ], where,
as in the first step, none of the intervals in U containing bk+1 contain bk .
If after k iterations we have bk +1 ≤ a for the first time, then the union of the finite covers
of [bi , bi−1 ], i = 2, 3, … , k + 1 is a finite subcover of U that we were looking for.
Otherwise bk > a for every k ∈+ . Thus, we have a decreasing sequence (b j ) of ele-
ments in A. Since A is bounded this sequence converges to some b. Since A is closed, b ∈ A.
Hence, there is an interval in U ∈U containing b. It follows that, from some point on, all
the members of the sequence (b j ) are in U. In particular, there will be two consecutive
bk , bk +1 ∈U . This is a contradiction (see the last sentence two paragraphs earlier). So, the
preceding cases apply, and we have established that A is compact.
Finally, we eliminate the two initial restrictions on the cover and on the set A.
Let U be any open cover. We replace it with the family of all intervals that are subsets
of the open sets in U , then we use the preceding argument to find a finite subcover of that
cover made of intervals, and, at the end, we replace each interval of that finite subcover
with one element of U that contains it as a subset.
The second restriction was that A was a closed interval. Suppose now that A is any
closed set. Then, since it is bounded, it is a subset of an interval [c, d]. Hence, U ∪{ \ A}
is an open cover of [c, d]. By the above argument, there is a finite subclass of U ∪{ \ A}
that covers [c, d]; we simply eliminate {\ A} from this subclass to get a finite subcover
of U that covers A.
⇐ Suppose now that every open cover of a subset A of  has a finite subcover. First
we show that A is bounded. Otherwise there is a sequence ai , i = 1,2, … , of elements of A
such that for every positive integer n there is an element of that sequence out of the inter-
val (−n, n). Then {(−n, n) : n ∈+ } would be an open cover of A that does not have a finite
subcover.
Now we show that A is closed. Suppose otherwise. Then there is an accumulation point
for A that is not contained in A. Denote it by x. Take any a, a < x , and any b, x < b, and con-
sider the following cover of A made of open intervals in  (where ε is any positive number).

 1   1 
{(−∞, a)} ∪  a − ε, x −  : n = 1,2, … ∪  x + , b + ε  : n = 1,2, … ∪ {(b, ∞)}.
  n     n  

K12146_C007.indd 156 2/13/15 2:11 PM

 7.1 Compact Spaces: Definition   ◾   157  

Extracting a finite subcover

1 1
{(−∞, a)} ∪  a − ε, x −    
 : n = 1,2, …, p  ∪  x + , b + ε  : m = 1,2, …, q  ∪ {(b, ∞)}
 n   m 

( )
of this cover is not possible, since otherwise the interval x − p1+1 , x + q1+1 would contain no
elements of A, contradicting the assumption that x is an accumulation point for A.

The Heine–Borel theorem holds for n in place of  (Theorem 6, Section 7.2).

Theorem 3 cannot be generalized to all metric spaces. For example, the Hilbert space
H is a metric space having closed and bounded subsets that are not compact [Exercise 10,
Section 7.4].

Generalizations often offer views from afar that reveal properties that are hard to see
from within a specific model of a theory. Compare, for example, the proof of Proposition 1
and the proof of the following generalization; the advantages of the latter are obvious.

Proposition 4. Compactness is preserved under continuous mappings.

That is, if A is a compact subspace of a space X and f : X → Y is continuous, then f (A)
is a compact subspace of Y.

Proof. Suppose U = {U i : i ∈ J } is an open cover of f (A). Then, since f is continuous,

{ f −1 (U i ) : i ∈ J } is an open cover of A. Since A is compact, that cover possesses a finite sub-

n
−1
n
n
cover { f −1 (U i ) : i = 1,2, … , n} of A. Hence, A ⊂ ∪ f −1 (U i ) . Then f ( A) ⊂ f ∪ f −1 (U i ) =
i =1
( n

i =1
)
∪ f ( f (U i )) = ∪U i , and so {U i : i = 1,2, … , n} is a finite subcover of f (A).
i =1 i =1

As a consequence, compactness is a topological property (i.e., it is preserved under

homeomorphisms).

We finish this section by extracting an easy alternative criterion for compactness.

A family F = {Fi : i ∈I } of subsets of X has the finite intersection property if every finite
intersection of elements of F has a nonempty intersection.

Proposition 5. A space X is compact if and only if every family of closed subsets of X with
the finite intersection property has a nonempty intersection.

Proof. ⇒ Let X be compact and let F = {Fi : i ∈I } be a family of closed subsets of X with the
finite intersection property. We need to show that ∩ Fi ≠ ∅ . Otherwise ∩ Fi = ∅ , and so,
i∈I i∈I
by de Morgan laws, ∪ Fic = X , so that the family {Fjc : i ∈I } of open sets covers X. Since X is
i∈I n
compact, there is a finite subcover {F1c , F2c ,… , Fnc } , so that ∪ Fi c = X . Applying de Morgan
n i =1
laws again, we have ∩ Fi = ∅ , contradicting our assumption that F has the finite intersec-
i =1
tion property.

K12146_C007.indd 157 2/13/15 2:11 PM

 158   ◾    Compactness and Related Matters

⇐ Suppose every family of closed subsets of X with the finite intersection property has a
nonempty intersection. Let {U i : i ∈I } be an open cover for X, so that ∪U i = X . Then, by de
i∈I
Morgan laws, ∩U ic = ∅ . Since all U ic are closed sets, our assumption implies that for some
i∈I
finite subset J of I, ∩U ic = ∅ . Applying de Morgan laws again we get ∪U i = X .
i∈J i∈J

Here is a variant of the last proposition.

Proposition 6. A space X is compact if and only if for every family { Ai : i ∈ I } of subsets of

X satisfying the finite intersection property, we have ∩ Ai ≠ ∅ .
i∈I

Proof. Exercise 16.

Exercises
1. When is a subspace of a discrete space X compact? What if X is indiscrete?
2. Let ( X ,T1 ) and ( X ,T 2 ) be topological spaces such that T1 ⊂ T 2 . Show that if ( X ,T 2 )
is compact, then so is ( X ,T1 ) .
3. Prove Proposition 2.
4. Show that the Cantor set is compact.
5. Show that the Sorgenfrey line is not compact.
6. Show that a quotient space of a compact space is compact.
7. (a) Show that the interval [0, ∞) is not compact.
(b) Show that [0, ∞) is not homeomorphic to the circle.
8. Show that if K is a compact subspace of  2 , then  2 \ K has exactly one unbounded
path component.
9. Prove that if X is a compact space, then the diagonal {( x , x ) : x ∈ X } is a compact
subspace of X × X .
10. Show that a space X is compact if and only if for every basis B of the space X, every
cover of X consisting of elements of B has a finite subcover.
11. Show that if a set X is well ordered, and if there exists the largest element m in X, then
X with the order topology is a compact space.
12. Show that if X is a compact metric space then every sequence has a convergent
subsequence.
13. Show that every compact metric space is complete.
14. Show that the topological sum ⊕ Xi of compact spaces is compact if and only if the
i ∈I
index set I is finite.

K12146_C007.indd 158 2/13/15 2:11 PM

 7.2 Properties of Compact Spaces   ◾   159  

15. (a) Show that there is no homeomorphism f :  2 →  2 such that f ( A) = B (see

Illustration 7.3).

A B
Illustration 7.3 The subspace A of  2 is shown to the left (all black points); B is to the right.

(b) Describe (visually or otherwise) a homeomorphism g :  2 →  2 such g ( A) = B

(see Illustration 7.4).

A B
   Illustration 7.4

16. Prove Proposition 6: A space X is compact if and only if for every family { Ai : i ∈I } of
subsets of X satisfying the finite intersection property, we have ∩ Ai ≠ ∅ .
i∈I

17. Show that if { A j : j ∈1,2,…} is a family of compact nonempty subsets of a Hausdorff

∞
space X such that A1 ⊃ A2 ⊃  ⊃ A j ⊃ , then ∩ A j ≠ ∅ . Show that if “Hausdorff”
j =1
is omitted than the new statement can fail.
18. (a) Prove that compact subspaces of a metric space are closed and bounded.
(b) Show that in the Hilbert space (Example 2, Section 5.2) there are closed and
bounded sets that are not compact.
19. Show that every finite CW-complex is compact.

7.2 Properties of Compact Spaces

A brief historical note: Proposition 2 is due to Vietoris (1921).

Compactness is obviously not hereditary; for example, no nonempty, open, proper sub-
space of the closed interval [0,1] (with the usual topology) is compact. In the next proposi-
tion we postulate an extra condition for heredity of compactness.

K12146_C007.indd 159 2/13/15 2:11 PM

 160   ◾    Compactness and Related Matters

Proposition 1. Every closed subspace of a compact space is compact.

Proof. Let X be a compact space and A be a closed subset of X. Note that Proposition 2,
Section 7.1 allows us the freedom to choose a cover made of open subsets of A or a cover
made of open subsets of X. We choose the latter. The proof is contained in Illustrations
7.5–7.8 and the associated captions.

X X
A A

Illustration 7.5 Let U = {U i : i ∈I } be an open Illustration 7.6 Then U ∪ { Ac } is an open

cover of A. cover of X.

X X
A A

Illustration 7.7 Since X is compact, there is a

finite subcover U ∗ of U ∪ { Ac }. Illustration 7.8 Then U ∗ \{ Ac } is a finite sub-
cover of U covering A.

Example 1
That the Sorgenfrey line S (generated by all intervals of type (a, b]) is not compact
is an easy exercise (Exercise 5, Section 7.1). It is equally easy to show that any open
interval [a, b) (considered as a subspace of S ) also fails to be compact. Since these
intervals are also closed in the Sorgenfrey topology, it follows from Proposition 1 that
any subspace of S that contains some [a, b) as a subset cannot be compact. In particular,
closed intervals [a, b] are not compact in S . ☐

K12146_C007.indd 160 2/13/15 2:11 PM

 7.2 Properties of Compact Spaces   ◾   161  

Compact subspaces of a compact space X need not be closed in X (Exercise 3). For the com-
pact subspaces to be closed in X it is sufficient to assume that the space X is Hausdorff. We
have defined that notion earlier; it is now time to raise its prominence.

Proposition 2. Every compact subspace of a Hausdorff space is closed.

Proof. Let X be a Hausdorff space and let A be a compact subspace of X. We show that Ac
is open. It suffices to show that every point x ∈ Ac is interior for Ac . The proof is given in
Illustrations 7.9–7.12 and in the captions.

X X

A y Uxy

Vy A x
x

Illustration 7.9 Since X is Hausdorff, for x and Illustration 7.10 Repeat that for every y ∈ A.
for every y ∈ A there are two disjoint open neigh- We get an open cover {Vy : y ∈ A} for A and a
borhoods U xy and Vy of x and y, respectively. class of open neighborhoods {U xy : y ∈ A} of x.

X X

A A
x x

n
Illustration 7.11 Since A is compact, there is Illustration 7.12 The finite intersection ∩ U i is
i =1
a finite subcover {V1 ,V2 ,…,Vn } of {Vy : y ∈ A} then an open neighborhood of x having empty
for A. Choose the associated neighborhoods n n
inter­section with ∪ Vi . So, ∩ U i is entirely
{U1 ,U 2 ,…,U n } of x. i =1 i =1
with­in Ac.

K12146_C007.indd 161 2/13/15 2:11 PM

 162   ◾    Compactness and Related Matters

The following easy consequence will be used later on.

Corollary 3. Let X be a compact space, let Y be Hausdorff, and let f : X → Y be continuous

and bijective. Then f is a homeomorphism.

Proof. Exercise 7.

We now turn our attention to product spaces: Is the product of a collection of compact
spaces a compact space? The affirmative answer to that question—which we will establish
in due course—is the primary reason the product topology is regarded to be “more natu-
ral” or “better behaved” than the box topology.
The case of finite products of spaces is much more accessible; the general case is post-
poned for a while.

Theorem 4. A finite product of compact spaces is a compact space.

Proof. We prove the claim for two spaces; the rest follows by induction. So, given two
compact spaces X and Y, we need to prove that X × Y is also compact.
Start with any open cover U of X × Y .
In Illustrations 7.13–7.17, the space X is represented by a horizontal ellipse, while Y is
depicted as a vertical line segment. The first part of the proof is given in the captions of
these illustrations.

x0 × Y

X×Y

x0

Illustration 7.13 For any point x 0 of X the

subspace x 0 × Y of X × Y is compact (being Illustration 7.14 So, we can extract a finite
homeomorphic to the space Y). subcover of U for the space x 0 × Y .

K12146_C007.indd 162 2/13/15 2:11 PM

 7.2 Properties of Compact Spaces   ◾   163  

Illustration 7.16 Project these open sets over

Illustration 7.15 Remove X × Y to see more X to get finitely many open sets in X. Take any
clearly the “skewer” of open sets surround- open subset V of their intersection containing
ing x0 × Y. x 0 (we may take the intersection itself).

Illustration 7.17 The open tube V × Y con-

tains (as a subset) x0 × Y (but is not necessar-
ily a member of U ).

Repeating the above procedure for every element x 0 of X we get a collection C of open
neighborhoods V covering X and a collection of tubes V × Y covering all X × Y . Since X
is compact, there is a finite subset of C covering X. The associated finite set of tubes will
therefore cover all of X × Y . Now replace each of these finitely many tubes with the asso-
ciated “skewer” (see Illustrations 7.15 and 7.16) consisting of finitely many open subsets
of U whose union contains the tube as a subset. We get a finite subcover of U covering
X × Y.

K12146_C007.indd 163 2/13/15 2:11 PM

 164   ◾    Compactness and Related Matters

Denote the unit closed interval [0,1] by I. The Euclidean unit n-cube is

{ }
I n = ( x1 , x 2 ,… , xn ) ∈n : 0 ≤ x1 ≤ 1, 0 ≤ x 2 ≤ 1,… , 0 ≤ xn ≤ 1 .

Corollary 5. The Euclidean cube I n is compact.

Proof. The unit interval I is compact, since it is bounded and closed. The claim follows
from Theorem 4.

It is now easy to generalize Theorem 3, Section 7.1.

Theorem 6 (Heine–Borel). A subset A of the space n is compact if and only if it is closed

and bounded.

Proof. ⇒ Suppose A is a compact subspace of n. Since n is Hausdorff, it follows from

Proposition 2 that A is closed. The set A could not be unbounded since otherwise, for every
x ∈n , the open cover consisting of the open balls B( x , m), m = 1, 2, … would not have a
finite subcover (and A would not be compact).
⇐ Suppose A is a closed and bounded subset of n. Since A is bounded, there is a large
cube C = [a1 , b1 ] × [a2 , b2 ] ×  × [an , bn ] in n containing A as a subset. Since I n is compact
(Corollary 5) and since any two cubes in n are homeomorphic, it follows that C is com-
pact. So, A is a closed subset of the compact space C. It follows from Proposition 1 that A
is compact.

Exercises
1. Show that a finite union of compact subspaces of a space X is a compact subspace
of X.
2. Recall that for two subsets A and B of a metric space (X, d), d(A, B) stands for
inf {d(a, b) : a ∈ A, b ∈ B}. Is there a metric space X and two disjoint closed subsets
A and B of X, at least one of them compact, such that d( A, B) = 0? (Compare to
Exercise 9, Section 2.1.)
3. Show that a compact subspace of a compact space X need not be closed in X.
4. Let X be a Hausdorff space.n Prove that for every collection {F1 , F2 ,… , Fn } of closed
subsets of X, the subspace ∪ Fj of X is compact if and only if Fj is compact for every
j =1
j = 1, 2, … , n.
5. Show that the closure of a compact subspace of a space X need not be compact.
(Hint: Consider the Particular-point topology over an infinite set X, defined by stipu-
lating that open sets are only the sets containing a fixed point of X.) What if X is
Hausdorff?
6. Let X be a compact space, let Y be a Hausdorff space, and let f : X → Y be continu-
ous mapping. Show that for every A ⊂ X , f ( A) = f ( A) .

K12146_C007.indd 164 2/13/15 2:11 PM

 7.3 Compact, Lindelöf, and Countably Compact Spaces   ◾   165  

7. Let X be a compact space, let Y be Hausdorff, and let f : X → Y be a bijection.

(a) Show that if f is continuous that it is open.
(b) Show that if f is open then it is continuous.
8. (a) Show that if X is Hausdorff then intersections of compact subsets are compact.
(b) Find a space X and two compact subspaces of X such that their intersection is not
compact.
9. (a) Show that if A is a compact subspace of a Hausdorff space X and if x ∈ X \ A, then
there are two disjoint, open subsets U and V of X such that A ⊂ U and x ∈V .
(b) Show that for every two disjoint compact subspaces A and B of a Hausdorff space
X, then there are two disjoint, open sets U and V such that A ⊂ U and B ⊂ V .
10. Show that there does not exist a class of infinitely many pairwise disjoint open sub-
sets of a closed interval [a, b] ⊂  , a < b.
11. Denote X = {( x ,0) : 0 ≤ x ≤ 1} ∪ ∪ ({s} × [0,1]) , where J is a totally disconnected subset
s ∈J
of the interval [0,1] (see the following illustration). Show that if f :[0, 1] → X is a
continuous mapping, then f ([0,1]) ∩ ∪ ({s} × {1}) is finite.
s ∈J

   Illustration 7.18

7.3 Compact, Lindelöf, and Countably Compact Spaces

A brief historical note: Lindelöf spaces are named after the Finnish topologist Ernst Lindelöf
(1870–1946), who in 1903 proved that open subsets of n are (what was later called)
Lindelöf spaces.

We have discovered in Proposition 4, Section 7.1, that compactness is preserved under con-
tinuous mappings. Other relationships between compactness and continuity that we note
in this section are consequences of that proposition.

Proposition 1. Every continuous mapping from a compact space into a Hausdorff space
is closed.

K12146_C007.indd 165 2/13/15 2:11 PM

 166   ◾    Compactness and Related Matters

Proof. Suppose X is compact, Y is Hausdorff, and f : X → Y is continuous. Choose a closed

subset F of X. Then, by Proposition 1, Section 7.2, F is compact. So, by Proposition 4, Section 7.1,
f (F) is also compact. Finally, since Y is Hausdorff, it follows from Proposition 2, Section 7.2
that f (F) is closed.

Recall that if f : X → Y is a mapping, then ~ f is the equivalence relation defined by x ~ f y

if f ( x ) = f ( y ).

Proposition 2. If f : X → Y is a continuous mapping from a compact space X onto a

Hausdorff space Y, then X ~ is homeomorphic to Y.
f

Proof. This is a consequence of Proposition 1 here and Proposition 2, Section 4.2.

Example 1
Consider the mapping f : S 2 →  6 defined by f ( x , y , z ) = ( x 2 , y 2 , z 2 , xy , xz , yz ) .
Since the component functions are continuous, so is f (Exercise 14, Section 5.1). By
2
Proposition 2, f (S 2 ) is homeomorphic to S ~ . In this case, for every x , y ∈S 2 such
f
that x ≠ y , we have f ( x ) = f ( y ) if and only if x and y are two antipodal points in S 2
(i.e., if and only if x = − y ). So, f (S 2 ) is homeomorphic to the quotient space obtained
from S 2 by identifying all pairs of antipodal points. This quotient space is the projec-
tive plane P 2 ; it will be important to us later on. We visualize it in Illustrations 7.19
and 7.20.

–x
x

Illustration 7.20 The space we get is called

Illustration 7.19 We identify each pair of the projective plane. It is not embeddable in
antipodal points x and −x on the sphere S 2 .  3 , so we cannot see it.
☐

K12146_C007.indd 166 2/13/15 2:11 PM

 7.3 Compact, Lindelöf, and Countably Compact Spaces   ◾   167  

The real-valued functions are important to us since they link general spaces with the well-
studied blueprint model .

Proposition 3. If X is compact and f : X →  is continuous, then there are points a, b ∈ X

such that f ( X ) ⊂ [ f (a), f (b)].

Proof. Suppose X is compact. Then f (X) is also compact. By Theorem 3, Section 7.1, f (X) is
a closed and bounded subset of . Since f (X) is bounded it has the least upper bound and
the greatest lower bound. Since f (X) is closed, these two are elements of f (X).

As a consequence every continuous mapping f :[c , d] →  attains its absolute maximum

and absolute minimum (i.e., we got the Extreme Value Theorem from basic calculus).

Example 2
Let (X, d) be any compact metric space, let A be a closed subset of X and let b ∈ X \ A
(see Illustration 7.21). Define a mapping f : A →  by f ( x ) = d( x , b) for every x in
A. Then, by Example 3, Section 2.2, and by simple considerations, f is continuous at
every point x of A. Since A is a closed set in a compact space X, it is itself compact. By
Proposition 3, the mapping f attains the minimal value at some point amin ∈ A , and it
attains the maximal value at some point amax ∈ A . Since b ∉ A, we have d(amin , b) > 0
and d(amax , b) > 0.

amax

A
amin
b
Illustration 7.21 There is a point in the closed
X set A that is the closest to b and there is a
point in A that is the farthest from b. ☐

The following proposition concerning compact metric spaces is important in basic analysis:

Proposition 4. Let ( X , d1 ) be a compact metric space, and let (Y , d2 ) be a metric space. A

mapping f : X → Y is continuous if and only if it is uniformly continuous.

Proof. We noticed in Section 2.2 that ⇐ holds. For the other implication assume that
f : X → Y is continuous, and choose any ε > 0. The continuity of f implies that for every
x ∈ X , there is a δ x > 0, such that if x ′ ∈ X is such that d1 ( x , x ′ ) < δ x , then d2 ( f ( x ), f ( x ′ )) < 2ε .
{( ) }
The family of balls B x , δ2x : x ∈ X is an open cover of X. Since X is compact, there exists

{( ) }
a finite subcover B x j , δ2x j : j = 1,2,…, n . Choose δ = min { δx j
2 }
: j = 1,2,…, n .
We will complete the proof by showing that if u, v ∈ X are such that d1 (u, v ) < δ , then
δ
(
d2 ( f (u), f (v )) < ε . So, take u, v ∈ X , such that d1 (u, v ) < δ. Some B x k , x2k contains u, )

K12146_C007.indd 167 2/13/15 2:11 PM

 168   ◾    Compactness and Related Matters

δ
and we have: d( x k , v ) ≤ d1 ( x k , u) + d1 (u, v ) < δ x2k + x2k = δ x k. So, d( x k , v ) < δ x k and, by sym-
metry, d( x k , u) < δ x k . Hence, by assumption, d2 ( f ( x k ), f (v )) < 2ε and d2 ( f ( x k ), f (u)) < 2ε .
Consequently, d2 ( f (u), f (v )) ≤ d2 ( f (u), f ( x k )) + d2 ( f ( x k ), f (v )) < 2ε + 2ε = ε .

The following two concepts are weaker than the notion of compactness.
A space X is Lindelöf if every open cover of X has a countable subcover. A space X is
countably compact if every countable open cover of X has a finite subcover.
Obviously a space X is compact if and only if it is both Lindelöf and countably compact.

Example 3
We mentioned in Example 1, Section 7.2, that the space X =  equipped with the
Sorgenfrey line topology S is not compact. However, this space is Lindelöf. In this
example we justify this claim.
Start with any open cover O of X and consider the set A of all intervals of the type
(a, b) such that (a, b) ⊂ O for some O ∈O . Obviously A is a cover of the subspace
X • = ∪ (a, b) of X.
( a ,b )∈A
For every x in X • choose (a,b) that contains it, then find two rational numbers, q1x
between a and x and q2x between x and b, so that x ∈(q1x , q2x ) ⊂ (a, b). There are clearly
countably many such intervals (q1x , q2x ), and it follows from the definition of A that
they cover X •. For each of them choose one (a, b) that contains it as a subset to get a
countable subcover B of A. Now, for each of the members of B choose an open set
from O that contains it as a subset (which we can do by the definition of A ) to get a
countable subset C of O that covers X •.
To complete our justification it suffices to find a countable subcover D of O that
covers X \ X • , in which case C ∪ D will then be a countable subcover of O that cov-
ers all of X.
Since the intervals of the type [a, b) make a basis for the Sorgenfrey topology,
for every x in X \ X • and every Ox ∈O that contains it, there is [ax , bx ) such that
x ∈[ax , bx ) ⊂ Ox . Then, x ∈[x , bx ) ⊂ Ox . For every x1 , x 2 ∈ X \ X ∗, x1 < x 2 , we have
that [x1 , bx1 ) ∩[x 2 , bx 2 ) = ∅, for otherwise ( x1 , bx1 ) would contain x 2 , and since
( x1 , bx1 ) ⊂ Ox1 ∈O , this would imply that x 2 is in X • . So, all of these intervals [x , bx )
are pairwise disjoint as x ranges through X \ X ∗. Since each of them contains a ratio-
nal number, there are countably many of them. So, there are countably many ele-
ments in X \ X • . Now choose one open set from O for each element of X \ X • to get
a countable subcover of O that covers all of X \ X ∗ . That is the desired D . ☐

Proposition 5. (Lindelöf Lemma) Every second countable space is Lindelöf.

Proof. Let O be an open cover of a second countable space X, and let B be a countable basis
for X. Each member of O is a union of elements of B . Replace each member of O by the

K12146_C007.indd 168 2/13/15 2:12 PM

 7.3 Compact, Lindelöf, and Countably Compact Spaces   ◾   169  

elements of B comprising that union. We get a new cover of X, and since each member of that
cover is a member of B , this is a countable cover. Now replace each member of that new cover
by any element of O containing it as a subset. We get a countable subcover of O .

Example 4
The Euclidean line  is Lindelöf. This is now a consequence of Proposition 5 and the
fact that  is second countable. ☐

Observe that the argument in Example 4 is not transferable to the case when  is equipped
with the Sorgenfrey topology, since that topology is not second countable (Example 15,
Section 3.3).

Exercises
1. Let f :[a, b] →  be a continuous function. Show that f ([a, b]) is a closed and
bounded interval.
2. Show that among all Hausdorff topologies over a fixed set X, compact topologies are
minimal. In other words, show that if ( X ,T1 ) is a Hausdorff space, ( X ,T 2 ) is a com-
pact space and T1 ⊂ T 2 , then T1 = T 2 .
3. (a) P
 rove that a space X is countably compact if and only if for every countable
collection {Fj : j = 1,2,…} of closed subsets of X that has the finite intersection
∞
property, ∩ Fj ≠ ∅.
j =1

(b) Let Y be a countably compact subspace of a space X, and let {Fj : j = 1,2,3,…}
be a collection of closed subsets of X such that F1 ⊃ F2 ⊃ F3 ⊃  . Show that
∞
∩ Fj ≠ ∅.
j =1

4. Show that if f : X → Y is continuous and onto and X is Lindelöf, then so is Y.

5. (a) Show that closed subspaces of Lindelöf spaces are Lindelöf.
(b) Let X =  be equipped with the Sorgenfrey line topology. Show that X × X is
not Lindelöf. (Hence products of Lindelöf spaces need not be Lindelöf.)
6. Show that if X is compact and Y is Lindelöf, then X × Y is Lindelöf.
7. Show that the Lindelöf property is not hereditary.
8. Define a sequence ( f j ) ( j ∈+ ) of functions I → I × I through the frames shown in
the Illustration 7.22.

K12146_C007.indd 169 2/13/15 2:12 PM

 170   ◾    Compactness and Related Matters

Illustration 7.22 From left to right: f1, f 2 , and f3 ; iterate to get each member of the sequence ( f j ).

The sequence ( f j ) is obviously a Cauchy sequence of functions (“obvious” in this sen-

tence means “as seen by the naked eye”—the visually introduced functions f j do not allow
any rigor). According to Exercise 17, Section 2.2, it converges uniformly to a continuous
function f : I → I × I ; f is the space filling Hilbert curve. Does it follow from Corollary 3,
Section 7.2, that I and I × I are homeomorphic? Justify!

7.4 Bolzano, Weierstrass, and Lebesgue

A brief historical note: Bolzano proved in 1817 that every bounded sequence in n has a
convergent subsequence. This is a special case of Corollary 2 below. Bolzano’s result was later
generalized by Weierstrass. The notion of a Lebesgue number stems from Lebesgue’s work on
measure theory, starting with his dissertation in 1902.

We continue investigating properties of compact spaces, or of spaces that are close to being
compact spaces.
A space X is a Bolzano–Weierstrass space, or a BW-space, if every infinite subset of X
has an accumulation point in X (Illustration 7.23).

Illustration 7.23 BW-space: infinite sets have accumulation points.

Proposition 1. Suppose X is a Hausdorff space. Then, X is a BW-space if and only if it is

countably compact.

K12146_C007.indd 170 2/13/15 2:12 PM

 7.4 Bolzano, Weierstrass, and Lebesgue   ◾   171  

Proof. ⇒ Assume X is a BW-space and let {Fi : i = 1,2,3,…} be a family of closed subsets of X
∞
satisfying the finite intersection property. It suffices to show that ∩ Fi ≠ ∅ (Exercise 3,
n i =1
Section 7.3). Take an element xn ∈∩ Fi , n ∈+ . If the sequence ( xn ) stabilizes at x, then
i =1
this x is in each Fi , and so it is in the intersection. Otherwise ( xn ) has infinitely many ele-
ments. Since X is a BW-space there is an accumulation point x for ( xn ). Every open set U
around x contains infinitely many elements from ( xn ) (Exercise 20, Section 3.2) and so it
intersects every Fi . Consequently, x is an accumulation point for each Fi . Since each Fi is
∞ ∞
closed, x ∈Fi for every i = 1,2,3,…. Hence, x ∈∩ Fi and we proved that ∩ Fi ≠ ∅ .
i =1 i =1
⇐ Suppose X is countably compact, and let B be an infinite set. Then, B has a count-
ably infinite subset A = {a1 , a2 ,…}. If A does not have an accumulation point in X, then
neither does any of the sets Ak = {ak , ak +1 ,…} , k = 2, 3, …. Consequently, each Ak is closed.
However these sets clearly satisfy the finite intersection property, and it is equally obvious
∞
that ∩ Ai = ∅ , contradicting the assumption that X is countably compact (and Exercise 3,
i =1
Section 7.3).

We did not need the assumption that X is Hausdorff to prove that countably compact
implies BW. Since every compact space is countably compact, the following is then an
immediate consequence.

Corollary 2. Every compact space is a BW-space.

We will see in a couple of propositions that for metric spaces the converse of Corollary 2
is also true.

Proposition 3. For every open cover O of a metric BW-space (X, d) there is a positive
number ε such that, for every x ∈ X , the open ball B(x, ε) is a subset of some member
of O .

Proof. Let X be a metric BW-space, and let O be an open cover of X. Assume that the
conclusion of the proposition fails. This means that, however small a number δ we choose,
there is a point x ∈ X , such that B( x , δ ) is not a subset of a member of O. In particular, for
every δ = n1 , n ∈+ , there is a point xn ∈ X such that B ( xn , n1 ) is not a subset of a member
of O. The set {xn : n ∈+ } cannot be finite; otherwise, there is a number m ∈+ such that
xn = xm for every n ∈{n1 , n2 ,… , nk ,…}, m < n1 < n2 < … < nk < … , which leads to the state-
ment that B ( xm , n1 ) is not a subset of a member of O for every n ∈{n1 , n2 ,…, nk ,…}. This
{ }
is not possible since xm must be in some open set in O , and since B ( xm , n1 ) : n ∈+ is a
local basis at xm . So {xn : n ∈+ } is infinite.
The assumption that X is a BW-space implies that {xn : n ∈+ } has an accumu­lation
point x ∈ X . Since O is a cover, there exists a U ∈O such that x ∈U . Since U is open, there

K12146_C007.indd 171 2/13/15 2:12 PM

 172   ◾    Compactness and Related Matters

is a ball B(x, r) that is a subset of U. Look at the ball

B ( x , 2r ) (see Illustration 7.24). Since X is a metric space,
B ( x , 2r ) must contain infinitely many members of the
r
set {xn : n ∈+ }. Choose any s such that 1s < 2r . Then
(see Illustration 7.24), B ( x s , 1s ) ⊂ B( x , r ) ⊂ U ∈O , and x
xs
we have a contradiction. ◼

The number λ = 2ε, where ε is as in Proposition 3, is

U
called a Lebesgue number for the covering O of the
compact space X. Explicitly, λ is a Lebesgue number
for O if every subset of X with a diameter less than λ
is contained within some member of O . Illustration 7.24 The radius 1
sof the
We will make use of the existence of the Lebesgue ball centered at x s is small enough
number several times in Part 2 of this book. so that this ball stays within B(x, r).

Proposition 4. A metric space is compact if and only if it is a BW-space.

Proof. ⇒ This was settled in Corollary 2.

⇐ Let X be a BW-space, and let U be an open cover of X. By Proposition 3 there exists
a Lebesgue number λ associated to U : every open ball B(x, λ) is a subset of some member
U x of U . In the next two paragraphs we will find a finite subset {x1 , x 2 ,… , xm } of X such
that { B( xi , λ ) : i = 1,2,…, m} is a cover of X. Then {U xi : i = 1,2,… , m} is a finite subcover of
U , and so X is compact.
Choose any x1 ∈ X . If there are no points in X out of B( x1 , λ ) then we can take
m = 1 and we are done. Otherwise there is an x 2 ∉ B( x1 , λ ) ; so x1 and x 2 are at dis-
tance > λ . Suppose we have chosen {x1 , x 2 ,… , xn } such that for every j ≠ k , d( x j , x k ) > λ . If
n
{B(xi , λ ) : i = 1,2,…, n} is a cover for X, we are done. Otherwise there is an xn+1 ∉∪
i =1
B( xi , λ ).
Suppose this procedure never ends. Then we have an infinite sequence ( xi ) such
that, for every j ≠ k , d( x j , x k ) > λ . Since X is a BW-space, there must be an accumula-
tion point x for ( xi ) . Consider the ball B ( x , λ 2 ) : it contains infinitely many members
of ( xi ). Any two of them are at distance less than λ, yielding a contradiction. Hence,
the procedure must end after finitely many steps, and we have the desired finite cover
of open balls.

Corollary 5. A metric space is compact if and only if it is countably compact.

Proof. That compact implies countably compact is clear. For the converse, suppose X
is a countably compact metric space. Since metric spaces are Hausdorff, it follows from
Proposition 1 that X is a BW-space. Proposition 4 now implies that X is compact.

K12146_C007.indd 172 2/13/15 2:12 PM

 7.5 Compactification   ◾   173  

Exercises
1. Show that every closed subspace of a BW-space is a BW-space.
2. (a) Check that the collection B = {{−n, n} : n ∈} is a basis for a topology T over .
(b) Show that ( ,T ) is not a BW-space.
3. Let X be an uncountable space equipped with the co-countable topology, and
let Y = [0, 1] be with the indiscrete topology. Show that X × Y is a noncompact
BW-space.
4. Show that every uncountable space X equipped with co-countable topology is not
a BW-space.
5. Let X be a BW-space, and let ( Fi ) be a sequence of closed, nonempty subsets of X such
∞
that Fi +1 ⊂ Fi for every i ∈+ . Show that ∩ Fi ≠ ∅ .
i =1

6. Show that every compact metric space is complete.

7. A space X is sequentially compact if every sequence has a convergent subsequence.
(a) Show that if X is sequentially compact then it is a BW-space.
(b) Show that if X is first countable and a BW-space then X is sequentially compact.
8. A mapping p : X → Y is perfect if it is continuous, closed, surjective, and if for every
y ∈Y , p −1 ( y ) is a compact subset of X. Prove that if p : X → Y is a perfect mapping
and Y is compact, then X is compact.
9. Show that every compact subspace of the Sorgenfrey line is countable. [Hint: You may
want to start with the identity map from the Sorgenfrey line onto the Euclidean line .]
10. Let H be the Hilbert metric space. Show that the unit closed ball
 ∞ 
D = ( x j )∞j =1 ∈ ∞ : ∑ xi2 ≤ 1 in H is not compact (despite being closed and
 i =1 
{ }
bounded). [Hint: Show that the set ( x j )∞j =1 ∈ ∞ : x j0 = 1, and x j = 0 for j ≠ j0 does
not have an accumulation point.]

7.5 Compactification

A brief historical note: In 1913 Constantin Carathéodory (1873–1950) was the first to study
compactifications of spaces (he worked on open subsets of  2).

We saw in Sections 7.1–7.4 that compact spaces have important properties. Hence, given
a space X it makes sense to try to compactify it by embedding it into a compact space.

K12146_C007.indd 173 2/13/15 2:12 PM

 174   ◾    Compactness and Related Matters

Achieving just that is rather simple; in fact we can do more without much effort. As the
following example shows we can easily extend X to a compact space Y such that X is
dense in Y.

Example 1
Start with a space X and choose a point p not in X. Define a topology over Y = X ∪ { p}
as follows: open sets in Y are Y and the open sets in X. Then Y is a compact space: any
cover of Y must contain Y as a member (since that is the only open set containing p),
and so {Y} would be a finite subcover. It is also clear that X (with the original topology)
is a dense subspace of Y. ☐

The plainness of this example hints at the necessity of upgrading the properties of the tar-
get space Y. Observe that Y in Example 1 is never Hausdorff: recall that we have assumed
that our spaces are not empty, and so X has a point that cannot be separated from p with
two disjoint open sets. We do not want such an ugly compact extension Y of X. However,
it is a good start.
A compact space Y is a compactification of a space X if the following two conditions
are satisfied:

(i) There is an embedding e : X → Y .

(ii) e(X) is dense in Y.

We are looking for a procedure that will associate Hausdorff compactifications under some
reasonable conditions. Before we present one, let’s look at our blueprint model again.

Example 2
The space X is the circle without the NP

point NP (the north pole). Consider

the homeomorphism from X onto  P4
P1
defined by the stereographic projec-
tion f (Illustration 7.25). If we simply
P2
add the north pole to X we clearly get
P3
a compactification S1 = X ∪ {NP}
of X. We want to add a new point f (P1) f (P2) f (P3) f (P4)
to  such that that the extension
f : S1 →  ∪ {the new point} of f is Illustration 7.25

also a homeomorphism (so that  ∪ {the new point} also becomes a compact space).
Since we may think of the mapping f as sending the north pole to “infinity,” we
will denote that new point by ∞ and we will call it an infinity point. The set  ∪ {∞}
will be denoted by  ∞ . As we have pointed out, we want it to be homeomorphic to
the circle via f . So, the set of open subsets of  ∞ should be { f (U ) : U open in S1 }.

K12146_C007.indd 174 2/13/15 2:12 PM

 7.5 Compactification   ◾   175  

There are two types of open sets in S1 : open sets containing NP and open sets not
containing NP. In the latter case, f operates the same way as f, so that the open sets
in  are also open in  ∞ . In the former case, as is easy to see, the complement of an
open set containing NP is mapped via f onto the complement in  of a closed and
bounded subset of  , so that f maps such sets into complements in  ∞ of closed
and bounded subsets of  . Consequently, open subsets  ∞ containing the infinity
point are complements in  ∞ of compact subsets of  . ☐

We now promote Example 2 into a general recipe for obtaining a special compactification.
Suppose X is noncompact, and ∞ a point not in X. The one-point compactification of X,
denoted X ∞ , is the space over X ∪ {∞} with topology specified as follows:

(a) ∅ and X ∪ {∞} are open sets.

(b) Every open subset of X is open in X ∪ {∞}.
(c) A subset A of X ∪ {∞} containing the point ∞ is open if and only if X \ A is a closed
and compact subset of X.

If X is Hausdorff, then stipulating that X \ A is closed is redundant, since compact subsets

of Hausdorff space are closed.

Example 3
The one-point compactification of  2 is homeomorphic to the sphere S 2 , and the ste-
reographic projection (see Example 7, Section 3.5) extended by sending the north pole
of the sphere to the ∞-point is a homeomorphism from S 2 onto  2∞ . As emphasized in
Illustration 7.26 the open sets in S 2 containing the North Pole correspond to comple-
ments in  2∞ of compact (and closed) subsets of  2 .

U
S2 \ U

f (S 2 \ U)

f (U)

Illustration 7.26 The dark-shaded open set U in S 2 contains the North Pole. The correspond-
ing open set in  2∞ of matching shade is the complement of the compact set f (S 2 \ U ), where
f is the stereographic projection.

K12146_C007.indd 175 2/13/15 2:12 PM

 176   ◾    Compactness and Related Matters

The one-point compactification of an open disk in  2 is also homeomorphic to S 2 (see

Exercise 2). ☐

The following proposition justifies the terminology used in the last definition.

Proposition 1. For every noncompact space X, the one-point compactification X ∞ is a

compactification of X.

Proof. We leave the proof that X ∞ is indeed a topological space as an exercise (Exercise 1).
X is a subspace of X ∞ (i.e., the inclusion X → X ∞ is an embedding): We check that a
subset U of X is open in X if and only if it is V ∩ X for some open subset V of X ∞ . It fol-
lows from the definition of the topology over X ∞ that if U is open in X then it is open
in X ∞ , and so we may take V = U . Conversely, we show that for every open subset V of
X ∞ , V ∩ X is open in X. Open subsets V of X ∞ are either open subsets of X (and so their
intersection with X is open in X), or they are such that X \V is closed (and compact) in X
(and so X \( X \ V ) = X ∩ V is open in X ).
X is dense in X ∞ : The set {∞} is the only nonempty subset of X ∞ that does not intersect
X. Since X is not compact, {∞} is not open in X ∞ . It follows that X is dense in X ∞ .
X ∞ is compact: Let O be an open cover of X ∞ . So, some member O of O contains the
point ∞. By assumption O c is compact in X. By Proposition 2, Section 7.1, O c is compact
in X ∞ . So, the open cover O of O c has a finite subcover O ′ . Then, O ′ ∪ {O} is a finite
subcover of O covering X ∞ .

In order to achieve Hausdorff-ness of the one-point compactification we need to pay atten-

tion to a local version of compactness.
A space X is locally compact at a point x ∈ X , if there is an open neighborhood U of x
such that the closure U is compact. A space X is locally compact if it is locally compact at
every point of X.
Obviously every compact space is locally compact. (Why?) The converse is obviously not
true. (Why? Exercise 9.)

Proposition 2. If a space X is locally compact at a point x then there is a compact set K and
an open set U such that x ∈U ⊂ K . For Hausdorff spaces X the converse is also true.

x U
K Illustration 6.27 Locally compact: every point
x within a compact set K possesses an open
neighborhood U that is a subset of K.

K12146_C007.indd 176 2/13/15 2:12 PM

 7.5 Compactification   ◾   177  

Proof. ⇒ Suppose that there is an open neighborhood of x such that its closure is compact.
Take K to be that closure.
⇐ Let K be a compact set containing x and let U be open in X such that x ∈U ⊂ K . Since
K is compact in X and since X is Hausdorff, K must be closed. So the closure U of U is a
subset of K. U is compact since it is a closed subset of the compact space K.

Example 4
The space  is locally compact. However, the subspace  is not. Suppose other­
wise. Take any point x in . Then there is an open subset U of  and a compact
subspace F of  such that x ∈U ⊂ F . Since U is open in  , there must be an open
interval (a, b) in  such that x ∈(a, b) ⊂ U . Take any two rational numbers p < r
in (a, b) and consider the closed interval [p, q] (with only rational numbers in it).
Since it is closed in  and since F contains [p, q], it follows that [p, q] is closed in
F. But F is compact, and closed subsets of compact spaces are compact, so that
[p, q] is compact. We now show that this is not true: take any irrational point y
in [p, q] and consider all closed intervals [ y − ε, y + ε] that are completely within
[p, q] (so, we should take ε to be small enough). This set clearly satisfies the finite
intersection property. But the intersection of all of them is empty (keep in mind
that y is irrational). So, [p, q] is not compact and we have arrived at a contradiction.
Consequently,  is not locally compact. ☐

We now prove that in order to obtain a compactification that is a Hausdorff space, the
starting space X must be Hausdorff and locally compact.

Proposition 3. A space X is locally compact and Hausdorff if and only if the one-point
compactification X ∞ is Hausdorff.

Proof. ⇒ Suppose X is Hausdorff and locally compact. Take any two points x and y in X ∞ .
The only interesting case is when one of them is ∞; so, assume y = ∞. Since X is locally
compact there is an open neighborhood U of x such that U is a compact subset of X. Then
U and U c ∪ {∞} are two disjoint and open (in X ∞ ) neighborhoods of x and ∞, respectively.
⇐ Suppose X ∞ is Hausdorff. Since being Hausdorff is hereditory, it follows that X is
Hausdorff, too. We check that it is locally compact. Let x ∈ X . Since X ∞ is Hausdorff, we
can separate x and ∞ with two open sets, U x and U ∞ , respectively. Then X \ U ∞ is compact
in X (by the definition of open sets around ∞), and it contains U x . The conclusion now fol-
lows from Proposition 2.

Example 5
The one-point compactification  ∞ of  is not Hausdorff since  is not locally
compact (Example 4). ☐

In Chapter 8 we will see another important compactification procedure: the Čech–Stone

compactification.

K12146_C007.indd 177 2/13/15 2:12 PM

 178   ◾    Compactness and Related Matters

Exercises
1. Show that the one-point compactification X ∞ of a space X is a topological space.
2. Show that if X and Y are homeomorphic, then so are their one-point compacti­fi­cations.
3. Let X be the subspace of the unit circle S1 consisting of the points in S1 with rational
polar angles (in degrees), and let Y = X \ {(0, 1)}. Is X homeomorphic to the one-
point compactification of Y?
4. (a) W
 hat familiar space is (homeomorphic to) the one-point compactification of the
interval [0,1)?
(b) Visualize the one-point compactification of the union of two intervals
(0,1) ∪ (1, 2).
(c) Visualize the one-point compactification of the quotient space of (0,2) ∪ (3,5)
obtained by identifying the points 1 and 4.
(d) Visualize and describe the structure of the one-point compactification of the
subspace of  obtained from  by deleting all integers.
5. Find the one-point compactification of a discrete space X with countably many ele-
ments. Identify a subspace of  which is isomorphic to the one-point compactification
of  .
6. Denote the set of nonnegative real numbers by  ≥0 , and its one-point compactifica-
tion by  ≥∞0. Attach the line segment {x } × [0, 1] at every x ∈ ∪ {∞} by identifying x
with {x } × {0}. Is the resulting space homeomorphic to the topological comb? Justify
your answer.
7. Show that if X is a subspace of a compact space Y, then there is a compact subspace Z
of Y such that X ⊂ Z and such that X is dense in Z.
8. (a) Show that if X is a closed subspace of Y, and the point ∞ is not in Y then X ∞ is a
subspace of Y∞ .
(b) Show that if the assumption that X is closed in Y is dropped then the conclusion
fails.
9. Show that a locally compact space need not be compact.
10. (a) Show that each compact space is locally compact.
(b) Show that finite products of locally compact spaces are locally compact.
11. (a) S how that a closed subset of a locally compact space is locally compact. [Hint:
You might need Exercise 16, Section 4.1.]
(b) Show that an open subset of a locally compact space need not be locally compact.

K12146_C007.indd 178 2/13/15 2:12 PM

 7.6 Infinite Products of Spaces and Tychonoff Theorem   ◾   179  

7.6 Infinite Products of Spaces and Tychonoff Theorem

A brief historical note: Tychonoff proved in 1930 that for every n ∈+ , the cube [0,1]n is
compact. The statement of Theorem 2 appears for the first time in his paper published in 1936.

Our goal in this section is to show that (any) product of compact spaces is com-
pact (Tychonoff’s theorem). The following theorem, which will be used in the proof of
Tychonoff’s theorem, is interesting in its own right.

Theorem 1. (The Alexander Subbasis Theorem) A space X is compact if and only if there
is a subbasis S for X such that every cover of X by (some of the) elements in S has a finite
subcover.

Proof. ⇒ If X is compact, then we can take S to be the whole topology over X.

⇐ Suppose there is a subbasis S for X such that every cover of X by elements in S has a
finite subcover. Suppose X is not compact. Then, there is a collection O of open sets cover-
ing X such that no finite subcollection of O covers X.
Let U be the family of all open covers O ′ of X such that O ⊂ O ′ and such that O ′
possesses no finite subcovers of X. Since O ∈U ,U is not empty. Suppose now that K is
a chain of elements of U (ordered under inclusion); so, each element C of K is a cover
of X that has no finite subcover, and such that O ⊂ C . Then the union of the elements of
K is a collection of open sets that certainly covers X. We now show that it is also in U .
Assume that union has a finite subcover {V1 , V2 ,… , Vm } . Then each Vk is in some member
of Cik ∈K . Since the members of K are ordered under inclusion, the set {V1 , V2 ,… , Vm }
must be a subset of one of Ci1 , Ci2 ,… , Cik , which would thus have a finite subcover. We
have a contradiction. Consequently, the union of the elements of K does not have a finite
subcover, and so it is in U . So, each chain of elements of U is bounded from above by a
member of U . By Zorn’s lemma, there exists an M ∈U such that O ⊂ M , and such that
M is a maximal (under inclusion) element of U .
Suppose now that M ∈M , and x ∈ M . Since M is an open set and since S is a sub-
basis, there are elements S1 , S2 ,… , Sn of S such that x ∈S1 ∩ S2 ∩ … ∩ Sn ⊂ M . Suppose
that each of S1 , S2 ,… , Sn is not a member of M. Then M ∪{Sk } is a proper overset of
M for every k = 1, 2, …, n. Since M is a maximal cover that has no finite subcover, it
follows that M ∪{Sk } must have a finite subcover. So, for every such k, there is a finite
subcollection M k of M such that M k ∪{Sk } is a finite cover of X. We claim that
M1 ∪ M 2 ∪ … ∪ M k ∪ … ∪ M n ∪ {S1 ∩ S2 ∩ … ∩ Sk ∩ … ∩ Sn } covers X. This is true since
any element that is not in Sk must be in the union of the elements of M k (since M k ∪{Sk }
covers X) and so the elements out of S1 ∩ S2 ∩ … ∩ Sn are in the union of the sets in
M1 ∪ M 2 ∪ … ∪ M k ∪ … ∪ M n .
Since M1 ∪ M 2 ∪ … ∪ M k ∪ … ∪ M n ∪ {S1 ∩ S2 ∩ … ∩ Sk ∩ … ∩ Sn } covers X and since
S1 ∩ S2 ∩… ∩ Sn ⊂ M , it follows that M1 ∪ M 2 ∪ … ∪ M k ∪ … ∪ M n ∪{ M } covers X, too. We

K12146_C007.indd 179 2/13/15 2:13 PM

 180   ◾    Compactness and Related Matters

get a finite subcover of M , contradicting our assumption. So, at least one of S1 , S2 ,… , Sn is

a member of M . We have shown that every element in X is contained in a set Sk ∈M ∩ S .
So, M ∩ S is an open cover of X. Since it is a part of our subbasis, it follows that it has a
finite subcover. That finite subcover is also a finite subcover of M . This is a contradiction.
So, X is compact.

Theorem 2. (Tychonoff Theorem) If X j , j ∈ J , is a family of compact spaces, then ∏ X j

j ∈J
is also compact.

Proof. Let X j , j ∈ J , be compact spaces, and denote X = ∏ X j . According to the

j ∈J
Alexander Subbasis Theorem, in order to show that X is compact it suffices to choose one
subbasis and extract a finite cover out of any cover of X made of some of the elements of
that subbasis. Not surprisingly we choose the standard subbasis S , consisting of the sets of
type p −j 1 (U j ), where U j is open in X j and p j : X → X j is the projection. So we start with
a cover O of X consisting of such sets.
Assume that O does not allow extraction of a finite subcover. This will lead to contra-
diction. Consider the set O j = {U j ⊂ X j : p −j 1 (U j ) ∈O } . It follows from the definition of O
that the sets of O j are all open in X j .
Assume O j covers X j for some j ∈ J . Then, since X j is compact, we can extract a
{ }
finite subcover O jfin . It is then clear that p −j 1 (U j ) : U j ∈O jfin covers X, and that it is a
finite subcover of O . This contradicts the assumption that O does not allow extraction of
a finite subcover.
Assume O j does not cover X j for every j ∈ J . So, there is an x j in each X j that is not
covered by O j . This implies that x = ( x j ) j ∈J is an element in X = ∏ X j out of the cover O ,
j ∈J
contradicting the assumption that O covers X.

The Tychonoff theorem is one of the primary reasons why the product topology is in a way
superior to the boxed product topology; the boxed product topology of compact spaces
need not be compact (Exercise 2).
As ∞the first application of this theorem we will now show that the product space
∞
 = ∏  is metrizable.
i =1

Example 1:  ∞ Is Metrizable
Since subspaces of metric spaces are metric spaces (through the inherited metric), it
suffices to show that  ∞ can be embedded into a metric space. Since  is homeo-
morphic to ∞the interval (0,1), it follows that  ∞ can be embedded into the product
space I ∞ = ∏[0,1], called the Hilbert cube. So, it suffices to embed I ∞ into a metric
i =1
space.
Recall the Hilbert space H, introduced in Example 2, Section 5.2, consists of all
( )
∞
sequences ( xn ) for which the sequence x12 + x 22 +  + xn2 converges in  . We
n=1
∞
2
showed there that d( x , y ) = ∑ ( xi − yi ) is a metric over H.
i =1

K12146_C007.indd 180 2/13/15 2:13 PM

 7.6 Infinite Products of Spaces and Tychonoff Theorem   ◾   181  

 x x 
Define f : I ∞ → H by f ( x1 , x 2 ,… , xn ,…) =  x1 , 2 ,… , n ,… . This is well defined
∞  2 n 
 2 x2 2
since  x1 + 2 +  + n 

xn 2 
( )
 n=1
( )
converges when xi ∈[0,1], i = 1, 2, …. It is clear
∞
that f ( I ∞ ) = ∏ [ 0, 1i ] and that f is one-to-one. Since f is continuous (next paragraph),
i =1
since H is Hausdorff, and since I ∞ is compact (Tychonoff’s theorem), it follows from
Corollary 3, Section 7.2 that f is a homeomorphism from I ∞ onto a subspace of H,
and so I ∞ is metrizable.
It remains to be shown that f is continuous. Choose any open set U in f ( I ∞ ) and
( )
any point a = ann in U. Notice that f (a1 , a2 ,… , an ,…) = a . Since U is open, there is an
open ball B(a , r ) contained in U. It consists of all z1 , z22 ,…, znn ,… , zi ∈[0,1] such that( )
( )
∞ 2 ∞
ai
∑ i − zii < r . Choose m large enough such that ∑ 1
i2
< 2r . This can be done
i =1 ∞ i =m
since ∑ i12 converges. Choose intervals (ai − εi , ai + εi ) , i = 1, 2, … , m − 1, small enough
i =1

( )
m−1 2
ai
such that ∑ i − zii < 2r for all z i ∈(ai − εi , ai + εi ) , i = 1, 2, … , m − 1. This is doable
i =1
since the last sum is finite. Then for every ( z1 , z 2 ,… , zn ,…) ∈(a1 − ε1 , a1 + ε1 ) ×

( ) ( )
∞ 2 m −1 2 ∞
ai
× (am−1 − εm−1 , am−1 + εm−1 ) × I × I  we have ∑ ai
i − zii ≤ ∑ i − zii +∑ 1
i2
i =1 i =1 i =m
< 2r + 2r = r . We proved that the open neighborhood (a1 − ε1 , a1 + ε1 ) ×
× (am−1 − εm−1 , am−1 + εm−1 ) × I × I  of (a1 , a2 ,… , an ,…) is sent into B(a , r ) ⊂ U
via f, thus establishing that f is continuous at (a1 , a2 ,… , an ,…) . ☐

The versatility of Tychonoff’s theorem is illustrated in the next example, where we enter the
territory of pure combinatorics.

Example 2. Ramsey Theorem (Finite Version)

A k-coloring of a set A by a set of colors {c1 , c2 ,…, ck } is a mapping c : A → {c1 , c2 ,…, ck }.
A subset B of A is monochromatic if all elements of B are colored with one color.
Precisely, B is monochromatic if there exists j ∈{1, 2, … , k}, such that c( B) = {c j }.
For a set A and a fixed integer r, 0 ≤ r ≤ A , we will denote the set of all subsets C of
A such that C = r by A(r ) . In this example we discuss colorings of A(r ) . Our starting
point is the infinite version of Ramsey’s theorem.
Theorem 3. (Ramsey Theorem; infinite case) For every k-coloring of (r ) , there is an
infinite subset M of  such that the subset M (r ) of (r ) is monochromatic.
It is not difficult to prove Theorem 3 using induction; see, for example [5],
page 157. ◼

In the finite version of Ramsey’s theorem we consider k-colorings of sets {1,2,… , n}(r ) .
Given n, k, r, and m, there may be a coloring of {1,2,…, n}(r ) such that for no subset M
of {1, 2, … , n} with M = m , the subset M (r ) of {1,2,… , n}(r ) is monochromatic. One

K12146_C007.indd 181 2/13/15 2:13 PM

 182   ◾    Compactness and Related Matters

1
such example, for k = 2, n = 5, m = 3 and r = 2, is given in
Illustration 7.28: the elements of {1,2,3,4,5}( 2) are repre-
sented by “colored” edges, one set of edges colored with
a fill line, and the rest of the edges colored with a dashed 5 2

line. It is easy to see that for no three-element subset M of

{1, 2, 3, 4, 5}, M ( 2) is monochromatic. In terms of the col-
ored edges, no subgraph of the graph shown in Illustration
7.28 that is spanned by three vertices, has all the edges in 4 3
one color (where “spanned” means that we include all of Illustration 7.28
the edges joining the chosen vertices). The finite version of
Ramsey’s theorem tells that, for fixed k, r, and m, there is a number n which precludes
such a case.
1 2
Theorem 4. (Ramsey Theorem; finite case): For
every k, r, and m, there is an n such that, for every
k-coloring of {1,2,… , n}(r ) , there is a subset M of
{1,2,…,n} such that M = m , and such that M (r )
is a monochromatic subset of {1,2,… , n}(r ).
6 3

For example, when k = 2 (two colors), r = 2

(we are coloring subsets of size 2) and m = 3, we
can take n = 6 : however we color the edges of the
graph Γ shown in Illustration 7.29, there must be a
5 4
three-element subset M of {1, 2, 3, 4, 5, 6} such that
the set of edges of the subgraph Ζ of Γ spanned by Illustration 7.29
the vertices in M is monochromatic. In the coloring we show in Illustration 7.29 we
can choose Ζ to be with the set of vertices being either {2, 3, 5} or {1, 2, 4}.

Proof (based on [5]). In the first two paragraphs we introduce the notation and make
some preliminary observations. The actual proof of the theorem is in the last paragraph.
A k-coloring of (r ) is a mapping (r )  {c1 , c2 ,… , ck }, which in turn is an element
of X = ∏ Xi , where Xi = {c1 , c2 ,… , ck } for every i ∈(r ). Equip {c1 , c2 ,… , ck } with the
i ∈( r )
discrete topology. Since it is a finite space, it is compact. By Tychonoff ’s theorem, so is
the space X.
Let A be a finite subset of , such that A ≥ r . The set
(r ) (r )
C A = {ϕ ∈ X : A is a monochromatic subset of  with respect to the coloring ϕ} is
     
the set   ∏ {c1 } × ∏ Xi  ∪ ∪   ∏ {ck } × ∏ Xi  , where all A(r )-coordinates
 i ∈A( r )  i ∈( r )  i ∈A( r )  i ∈( r )
 i ∉A( r )
  i ∉A( r )

are equal. Clearly, each C A is open in X.
Suppose the statement of Theorem 4 fails. This means that there are
k , r , m ∈+ such that for every n ∈+ , n ≥ r , there exists a k-coloring of
{1,2,… , n}(r ) such that for every M ⊂ {1, 2, … , n}, M = m , the set M (r ) is

K12146_C007.indd 182 2/13/15 2:13 PM

 7.6 Infinite Products of Spaces and Tychonoff Theorem   ◾   183  

not monochromatic. It follows that (for such k , r , m ∈+ ), for every finite
subset F of  with at least m elements, F (r ) is not monochromatic.
Consequently, for every A1 , A2 ,…, Ap such that A1 = A2 =  = Ap = m , the
set C A1 ∪ C A2 ∪… ∪ C Ap cannot be the entire space X. Since X is compact, it
follows that ∪ C A ≠ X , and hence ∩ (C A )c ≠ ∅ . Notice that (C A )c is the set
|A|=m |A|=m
{ϕ ∈ X : A(r ) is not a monochromatic subset of (r ), with respect to the coloring ϕ}.
Then, for every coloring ψ ∈ ∩ (C A )c ≠ ∅ of (r ) , there is no subset B of 
|A|=m
such that | B| = m and such that B(r ) is monochromatic. This contradicts
Theorem 3.
☐

Exercises
1. Prove the converse of the Tychonoff theorem: If ∏ X j is compact, then so is every X j ,
i ∈J
j ∈J.
Box
2. Find compact spaces X j , j ∈ J , such that ∏ Xi is not compact.
i ∈I
3. Show that an infinite and countable discrete space X is never equal to the product
space ∏ Y j , where each Y j is not homeomorphic to X.
j ∈J
4. Use the Tychonoff theorem to prove that the Cantor set is compact.
5. Consider the Tychonoff cube X = ∏ I j , where each I j is a copy of I = [0, 1]. Let H
j ∈I
be the subspace of X consisting of all nondecreasing functions I → I (H is the Helly
space). Show that H is compact. [Hint: Exhibit H as f (X) for a continuous f : X → X .]
6. Consider the space ∏ X j , where each X j is a fixed compact space X. Prove that the
j ∈J
diagonal D = {( x ) j ∈J : x ∈ X } is compact in ∏ X j .
j ∈J
7. Let Xi , i ∈ I , be nonempty sets, let yi ∉ Xi for every i ∈ I , and define topolo-
gies over Yi = Xi ∪ { yi } , i ∈ I , by declaring open sets to be {∅, Xi , { yi }, Yi } . Denote
Y = ∏ Yi , and let pi : Y → Yi , be the projections.
i ∈I

{ }
(a) Use the Tychonoff theorem to show that C = pi−1 ( yi ) : i ∈I is not a cover of Y.
(b) Deduce the Axiom of Choice; that is, show that there exists f : I → ∪ Xi , such
i∈I
that f (i ) ∈ Xi for every i ∈ I .

K12146_C007.indd 183 2/13/15 2:13 PM

This page intentionally left blank
 Chapter 8

Separation Properties

I n a metric space every two disjoint closed subsets are respective subsets of two disjoint
open sets (Exercise 10, Section 2.1). Using alternative terminology, in metric spaces
every two closed subsets can be separated by two disjoint open subsets. Spaces satisfy-
ing that property (with some additional constraints to be discussed later) are called
normal spaces. As we will see, not every normal space is metrizable. Nevertheless, nor-
mal spaces have fine enough structure and possess important properties. For example,
we will prove later that any continuous map into  defined on a closed subset of a
normal space can be extended over the whole space (Tietze extension theorem).
In this chapter we will consider spaces in which various types of disjoint sets can be
separated (in some way) by open sets. We will start from rough topologies and introduce
more and more refining properties leading to near-metric spaces.

8.1 The Hierarchy of Separation Properties

A brief historical note: Andrey Kolmogorov (1903–1987) introduced T0 -spaces around 1930;
T1 spaces were first defined in 1907 by Frigyes Riesz (1880–1956). These classes of spaces are
sometimes—but lately rather rarely—named after their inventors. On the other hand, T2 -spaces
are commonly known as Hausdorff spaces; Hausdorff introduced them in 1914. Regular spaces
were defined by Vietoris in 1921; normal spaces were introduced by Tietze in 1923, and inde-
pendently by Alexandroff and Urysohn in 1929.

We start by listing the (tentatively called) main separation properties. The first one will be
used infrequently.
A space X is said to be a T0 -space if for every x , y ∈ X there are open sets U and V such
that x ∈U , y ∈V and such that y ∉U or x ∉V .

185

K12146_C008.indd 185 2/9/15 3:01 PM

 186   ◾    Separation Properties

y
V
Illustration 8.1 T0 -space: x ∈U , y ∈V ; how-
x ever, at least one of U and V does not contain
y or x. In this illustration we have x ∉V .

The discrete topology over any X will be a universal—albeit, rather uninteresting—

example of a space possessing this, and every other separation property in our emerging
list. A less trivial example follows.
Example 1

Let X = {a , b} and T = {∅ , {a}, X }. Then, as is very easy to see, this is a simple T0 -space. ☐

A space X is said to be a T1-space if for every x , y ∈ X there are open sets U and V such that
x ∈U , y ∈V and such that y ∉U and x ∉V .

V
y
x Illustration 8.2 T1 -space: x ∈U , y ∈V , x ∉V , y ∉U . Note
that U and V need not be disjoint.

It is obvious that every T1-space is a T0 -space. On the other hand, the T0-space described in
Example 1 is not a T1-space. Consequently, the class of T1-spaces is a proper subclass of the
class of T0 -spaces.

Example 2
Let X be an infinite set equipped with the co-finite topology. Then X is a T1-space: for
every x , y ∈ X , the sets U = X \{ y } and V = X \{ x } perform as in the above definition
of a T1-space. ☐

T1-spaces are important because of the following simple result.

Proposition 1. A space X is a T1-space if and only if each singleton in X is closed.

Proof. ⇒ Suppose X is a T1-space and take any x ∈ X . Then it follows straight from the
definition of T1-spaces that every point in { x }c is an interior point for { x }c , so that { x }c is
open, and thus {x} is closed.
⇐ Assume all singletons are closed and take any two distinct elements x , y ∈ X . Then, { x }c
and { y }c are open, x ∈{ y}c , y ∈{x }c , and y ∉{ y}c and x ∉{x }c .

K12146_C008.indd 186 2/9/15 3:01 PM

 8.1 The Hierarchy of Separation Properties   ◾    187  

Next, we meet an old acquaintance of ours, under a different name.

A space X is said to be a T2 -space if for every x , y ∈ X there are disjoint open sets U and
V such that x ∈U , y ∈V .
What we are now calling a T2 -space was earlier called a Hausdorff space. We will mostly
stick with the old terminology except where it is convenient to talk about Ti -spaces in gen-
eral, for various choices of the index i.

Illustration 8.3 T2 -space: x ∈U , y ∈V , and

U
V U and V are disjoint.
x
y

It is plain that every T2 -space is a T1 -space. That the converse is not true is shown in
Example 2 above. Hence, the class of T2 -spaces is a proper subclass of the class of T1 -spaces.
(For another T1 -space that fails to be a T2 -space see Exercise 5.)
The next example will be of a T2 (Hausdorff) space which is not a “ T3 ”-space (the defini-
tion follows soon).

Example 3
Consider the half-disk topology over  2up = {( x , y ) : y ≥ 0} (Example 7, Section 3.3). It
is clear that it makes  2up a Hausdorff space. ☐

It is convenient at this stage to introduce the following terminology: Given two disjoint
subsets A and B of a space X, we will say that two disjoint open sets U and V separate A
and B, respectively, if A ⊂ U and B ⊂ V . This terminology is compatible with the one used
in relation to connectedness of spaces (Exercise 3).
A space X is a regular space if for every x ∈ X and every closed subset F of X, there are
disjoint open sets U and V such that x ∈U and F ⊂ V . It is a T3-space if it is regular and
a T1-space.1

U V

x F
Illustration 8.4 T3 -space or regular space:
separating points and closed sets with dis-
joint open sets.

1
There is not much consistency in the literature regarding this terminology; sometimes “regular = T3,” sometimes “regu-
lar = T3 + T1”; in this book we will stick with “regular +T1 = T3.”

K12146_C008.indd 187 2/9/15 3:01 PM

 188   ◾    Separation Properties

Regular spaces allow us to separate every point and every closed set not containing that
point with two disjoint open sets. The stipulation that X be T1 is needed to establish the
following proposition.

Proposition 2. Every T3 -space is a T2 -space.

Proof. Left as an easy exercise (Exercise 6).

As announced, Example 3 furnishes a space that is T2 but not T3 . It is not T3 since a point
(a, 0) on the x-axis and the closed set {( x ,0) ∈ 2 : x ≠ a} cannot be separated by two dis-
joint open sets. This, together with Proposition 2, implies that the class of T3 -spaces is a
proper subclass of the class of T2 -spaces.
The next example is of a T3 -space.

Example 4
The Sorgenfrey topology over  makes it a T3 -space. This claim is easy to justify.
First note that this topology makes  a T1-space. For the rest, start by taking a closed
subset F of  and a point a out of F. There must be an open set (open in the Sorgenfrey
topology, of course) [a, b) that is disjoint from F, for otherwise we would have a ∈ F ′,
and so a ∈ F = F. The complement [a, b)c is ( −∞ , a ) ∪[b , ∞ ), and it is open in the
Sorgenfrey topology. Then U =[a , b ) and V = ( −∞ , a ) ∪[b , ∞ ) is a separation of a and
F by two disjoint open sets. ☐

At the top of our hierarchy of Ti -spaces are the T4 -spaces.

A space X is said to be a normal space if for every two disjoint closed subsets F and
G of X, there are disjoint open sets U and V such that F ⊂ U and G ⊂ V . The space X is a
T4 -space, if in addition to being normal, it is a T1 -space.

U V

F G
Illustration 8.5 T4 -space or normal space:
separating disjoint closed sets with disjoint
open sets.

The stipulation that T4 -spaces are T1 makes it obvious that every T4 -space is a T3 -space. As
we saw in Example 4, the Sorgenfrey topology makes  a T3 -space. We will prove later that
it is also a T4 -space. For the time being we will not show an example of a T3 -space that is
not a T4 -space; for that we need to work harder (Example 2, Section 8.2).
Illustration 8.6 depicts the hierarchy of the various types of spaces introduced in this
section; the question marks indicate instances where strict inequality has not yet been
established.

K12146_C008.indd 188 2/9/15 3:01 PM

 8.1 The Hierarchy of Separation Properties   ◾    189  

T0

T1

T2
T3 ?
T4 metric
?

Illustration 8.6

The next two theorems establish the fact that Ti -properties, i = 0, 1, 2, 3, are hereditary and
preserved under products of spaces. Observe that T4 spaces are excluded; we will justify
that exclusion in the next section.

Theorem 3. Subspaces of Ti -spaces are Ti -spaces, i = 0, 1, 2, 3.

Proof. We leave it as an exercise (Exercise 7).

Theorem 4. If X j , j ∈ J are Ti -spaces (i ∈{0, 1, 2, 3}), then so is ∏ X j .

j ∈J

Proof. We deal only with i = 3; the rest is left as an exercise (Exercise 8).

Suppose X j , j ∈ J are all regular spaces, and consider X = ∏ X j . It follows from Exercise 8
j∈J
that, since each of X j is T1 ( j ∈ J ), the same is true for X. Let x ∈X , and let F be a closed
subset of X not containing x. Then F c is an open set containing x, and hence there is a stan-
dard basis element U for X such that x ∈U ⊂ F c . The idea of the rest of the proof is outlined
in Illustration 8.7 and its caption.

V1 V2

x
F
Illustration 8.7 We will separate x and U c with
the disjoint open sets V1 and V2 , thus achiev-
ing a separation of x and F by the same open
U Uc
sets.

The open set U is of the type U = ∏ U j where all but finitely many U j coincide with the
j ∈J
respective spaces X j . Since F ∩ U = ∅, U is not all of X, and so there is j0 ∈ J such that
U jo ≠ X j0 . Since X j0 is regular, we can separate the j0 -th coordinate x j0 of x and U cj0
with two disjoint open sets W1 and W2 , respectively. Denote, as usual, the projection

K12146_C008.indd 189 2/9/15 3:02 PM

 190   ◾    Separation Properties

X → X j0 by p j0 . Then V1 = p −j01 (W1 ) and V2 = p −j01 (W2 ) are two disjoint open sets separat-
ing x and U c ( x ∈V1 , U c ⊂ V2 ), and since F ⊂ U c , these two open sets separate x and
F, respectively.

As hinted by the last two theorems, T4 -spaces are exceptional compared to the other
Ti -spaces, i = 0,1, 2, 3. We will examine them more carefully in the following sections.

Exercises
1. Show that X is a T0 space if and only if for every distinct x , y ∈ X , {x } ≠ { y}.
2. Show that the number of T0 -topologies over a set with n elements is at least 2n .
3. Let A and B be two disjoint subsets of a space X. Show that there are U and V that
separate A and B if and only if {A, B} is a separation of the subspace A ∪ B of X.
4. Show that every finite T1 space is discrete.
5. Show that the Zariski topology over n is T1 but not T2 .
6. Prove Proposition 2: Show that every T3 -space is Hausdorff.

7. Prove Theorem 3: Subspaces of Ti -spaces are Ti -spaces, i = 0, 1, 2, 3.

8. Prove Theorem 4: Show that if X j , j ∈ J are Ti-spaces (i ∈{0 , 1, 2}), then so is ∏ X j .

j ∈J
9. (a) Find a normal space X and a quotient space Y = X ~ such that Y is not Hausdorff.

(b) Find a normal space X and a quotient space Y = X ~ such that Y is not a T0 -space.
10. Show that every subspace of a space X is normal if and only if every open subspace of
X is normal.
11. (a) Show that if for every two distinct elements x , y ∈ X there is a continuous func-
tion f : X →  such that f ( x ) = 0 and f ( y ) = 1 then the space X is Hausdorff.
(b) Show if X is a T1 -space and if for every x ∈ X and every closed subset F of X not
containing x there is a continuous function f : X →  such that f ( x ) = 0 and
f ( F ) = {1}, then X is a T3 -space.
(c) Show that if X is a T1-space and if for every two disjoint closed subsets F and G of
X there is a continuous function f : X →  such that f ( F ) = {0} and f (G ) = {1}
then X is a T4 -space.
[Note: The converses of (a) and (b) are not true. The converse of (c) is a major
theorem—Urysohn’s lemma—and will be proven later.]

K12146_C008.indd 190 2/9/15 3:02 PM

 8.2 Regular Spaces and Normal Spaces   ◾    191  

8.2 Regular Spaces and Normal Spaces

A brief historical note: The name T3 1 -spaces for completely regular spaces, introduced by
2
Tychonoff in 1930, was a joke, that somehow became semi-standardized terminology.

The following useful theorem gives us two new criteria for normality.

Theorem 1. The following statements are equivalent.

(i) X is a normal space.

(ii) For every closed subset F of X and every open subset U of X such that F ⊂ U , there is
an open subset W of X such that F ⊂ W ⊂ W ⊂ U .
(iii) For every two disjoint closed subsets F and G of X there are two open sets U and V
such that F ⊂ U and G ⊂ V , and such that U ∩ V = ∅ .

U
X
F
W

W Illustration 8.8 Part (ii) of Theorem 1.

Proof. (i) ⇒ (ii) Assume X is a normal space, and choose a closed subset F of an open
set U. Then F and U c are disjoint closed sets. By the normality of X there are two disjoint
open sets V and W such that F ⊂ W and U c ⊂ V (see Illustration 8.9).

W V

U Uc
Illustration 8.9

Since W ∩ V = ∅, we have that W ⊂ V c . Hence, W ⊂ V c . Since V c is closed V c = V c , and

so F ⊂ W ⊂ W ⊂ V c ⊂ U , where V c ⊂ U since U c ⊂ V .

K12146_C008.indd 191 2/9/15 3:02 PM

 192   ◾    Separation Properties

(ii) ⇒ (iii) Take any two disjoint closed sets F and G. By (ii) there is an open set U such
that F ⊂ U ⊂ U ⊂ G c . The inclusion U ⊂ G c implies G ⊂ (U )c , and we can use (ii) again to
claim existence of an open set V such that G ⊂ V ⊂ V ⊂ (U )c . Since V ⊂ (U )c , it follows
that U and V are disjoint, and so the open sets U and V we have found satisfy the require-
ments of (iii).
(iii) ⇒ (i) is obvious.

An analogous statement is true for regular spaces. In the next theorem we make this
more explicit.

Theorem 2. A space is regular if and only if for every x ∈ X , and for every open subset U of
X containing x, there is an open subset W of X such that x ∈W ⊂ W ⊂ U .

Proof. The proof of this theorem is similar to the corresponding part of the proof of
Theorem 1; so we leave it as an exercise (Exercise 2).

In the next theorem we establish that the difference between being regular and normal is
not more than being Lindelöf.

Theorem 3. If a space is regular and Lindelöf, then it is normal.

Proof.
Start with two disjoint closed subsets F and G of a regular and Lindelöf space X. Since X is
regular, for every point x ∈G there are two disjoint open sets Vx and U x such that x ∈Vx
and F ⊂ U x . The set V = {Vx : x ∈G} is an open cover of G, and so V ∪{G c } is an open cover
of X. The space X being Lindelöf, we can extract a countable open cover of X from V , which,
after throwing out G c , becomes a countable open cover {Vx1 ,Vx 2 ,…,Vxn ,…} of G. Easing the
notation by relabeling each Vxi by Vi , we get a countable open cover {V1 , V2 ,… , Vn ,…} of G.
Since Vx ⊂ (U x )c , we have that Vx ⊂ (U x )c = (U x )c ⊂ F c , and so the closures of the mem-
bers of the cover V stay away from F. Consequently, the countable cover {V1 , V2 ,… , Vn ,…}
of G has the property that Vi ∩ F = ∅ for every i = 1, 2, ….
Symmetric argument yields an open cover {U1 ,U 2 ,… ,U n ,…} of F such that U i ∩ G = ∅
for every i = 1, 2, …. ∞ ∞
The problem we need to fix is that ∪ Vi and ∪U i need not be disjoint. We will do it by
i =1 i =1
using countability to recursively modify the two families of open sets.
Define V 1• = V1 \ U1 , Vn• = Vn \ (U1 ∪ U 2 ∪ … ∪ U n ), n = 2, 3, …. Note that since V •n =
( )
Vn ∩ (U1 )c ∩ (U 2 )c ∩ ... ∩ (U n )c all these sets are open. Since U i ∩ G = ∅ for every
{ }
i = 1, 2, …, the family of open sets V1• , V2• ,… , Vn• ,… is still a cover of G. Symmetrically,
U 1• = U1 \ V1 , U n• = U n \ (V1 ∪ V2 ∪ … ∪ Vn ), n = 2, 3,… defines an open cover of F. It follows
∞ ∞
from our construction that U n• ∩ V •m = ∅ for every n , m ∈{1, 2, …}, so ∪ V i• and ∪U •i are
i =1 i =1 ∞
disjoint. Summarizing, we have found what we wanted: a pair of disjoint open sets ∪ V •i
∞ ∞ ∞ i =1
and ∪U •i such that G ⊂ ∪V •i and F ⊂ ∪U •i .
i =1 i =1 i =1

K12146_C008.indd 192 2/9/15 3:02 PM

 8.2 Regular Spaces and Normal Spaces   ◾    193  

Example 1
We know that the Sorgenfrey line is a regular space (Example 4, Section 8.1) and that it
is Lindelöf (Example 3 in Section 7.3). Theorem 3 thus implies that the Sorgenfrey line is
normal. ☐

Corollary 4. Every second countable and regular space is normal.

Proof. This is a consequence of Theorem 3 and Lindelöf’s lemma (Proposition 5, Section 7.3).

We are ready for the promised regular space that fails to be normal.

Example 2. A Regular, but Not Normal Space

The Sorgenfrey plane is the product space S × S of two Sorgenfrey lines. The rectan-
gles of type [a , b ) × [c , d ) constitute a basis for S × S . Since S is regular (Example 4,
Section 8.1), it follows from Theorem 4, Section 8.1, that S × S is also regular. However,
the Sorgenfrey plane is not normal: take the line y = − x (or, for that matter, any line
of negative slope) and consider it as a subspace E of S. Let Q be the set of all points in
E of rational distance from (0,0), and let P consist of points in E of irrational distance
from (0, 0). Then both Q and P are closed subsets of S × S (see Illustration 8.10).

E S×S

x P

Illustration 8.10 The three generic cases

z
of points x, y, and z not in Q showing
that Q c is open (and that Q is closed).

These two closed sets cannot be separated with two disjoint open sets (Exercise 11,
Section 2.1). ☐

The last example confirms that the class of T4 -spaces is a proper subclass of the class of
T3 -spaces. There is one more important class of spaces squeezed between these two; it is
sometimes called the class of T3 1 -spaces. The related notion is that of completely regular
2
spaces.
A space X is completely regular if for every x ∈ X and every closed subset F of X that
does not contain x, there exists a continuous function f : X → [0, 1] such that f ( x ) = 0 and
f ( F ) = {1}. If in addition X is T1 , then we say that it is a T3 1 -space.1
2

1
A completely regular Hausdorff space is called a Tychonoff space; the terminology (completely regular, T3 1 , Tychonoff)
2
varies from author to author.

K12146_C008.indd 193 2/9/15 3:02 PM

 194   ◾    Separation Properties

Every completely regular space is regular (Exercise 11, Section 8.1). A direct consequence
of Urysohn’s lemma (to be stated and proven in Section 9.1) is that every normal T1-space is
completely regular. We will show in 8.3 that the class of normal T1-space is a proper subclass of
the class of the class of completely regular spaces. For that we need the following simple result.
Proposition 5. Every subspace of a completely regular space is completely regular.
Proof. Let X be a completely regular space, and let Y be a subspace of X. Consider a closed
subset F of Y and a point x ∈Y \ F . Then F = G ∩ Y for some closed subset G of X not con-
taining x. By assumption there exists a continuous mapping f : X → [0, 1] such that f ( x ) = 0
and f (G ) = {1}. The restriction of f to Y confirms that Y is completely regular.

There are regular spaces that fail to be completely regular. We will not provide such
examples here (see, for example, [71]).

Exercises
1. (a) Let X and Y be normal spaces. Show that X ⊕ Y is a normal space.
(b) Let X and Y be completely regular spaces. Show that X ⊕ Y is a completely regular space.
2. Prove Theorem 2; that is, show that a T1-space is regular if and only if for every
x ∈ X and every open subset U of X containing x, there is an open subset W of X
such that x ∈W ⊂ W ⊂ U .
3. Show that every closed subspace of a normal space is normal.
4. Use Theorem 3 to show that a product of two Lindelöf spaces need not be Lindelöf.
5. Show that if X and Y are completely regular spaces, then so is X × Y .
6. Let X be a Hausdorff space. Prove that for every compact subset K of X and every
point x ∈ X \ K , there are open disjoint subsets U and V such that K ⊂ U and x ∈V .
7. Show that every locally compact Hausdorff space is regular.
8. Let A be a compact subspace of a T3 -space X. Show that for every closed set F such that
A ∩ F = ∅, there exist open sets U , V ⊂ X such that A ⊂ U , F ⊂ V , and U ∩ V = ∅.
9. Let X be a normal separable space, and let E be a subset of X such that E ≥ 2ℵ0. Assume
that E has no accumulation points.
(a) Show that for every Y ⊂ E , Y and E \Y are disjoint and closed subsets of X.
(b) Define a mapping f : P ( E ) → P ( D ) as follows: for every Y ⊂ E choose two dis-
joint open sets UY and VE \Y such that Y ⊂ UY and E \ Y ⊂ VE\Y , and set f (Y ) = UY .
Show that f is well defined and one-to-one.
(c) Deduce from (a) and (b) that if a subset E of a normal and separable space, satis-
fies E ≥ 2ℵ0 , then it must have an accumulation point.
10. Show that any countable, completely regular, Hausdorff space with at least two
elements is disconnected.

K12146_C008.indd 194 2/9/15 3:02 PM

 8.3 Normal Spaces and Subspaces   ◾    195  

8.3 Normal Spaces and Subspaces

A brief historical note: The space in Example 1 (the Tychonoff plank) was constructed in 1930,
not surprisingly, by Tychonoff.

If we weaken the requirement of regularity to Hausdorff-ness and strengthen the

requirement of Lindelöf-ness to compactness in Theorem 3, Section 8.2, then the con-
clusion is the same:

Theorem 1. Every compact Hausdorff space is normal.

Proof. In view of Theorem 3 in 8.2 it suffices to show that compact Hausdorff spaces are
regular. Let F be a closed subset of a compact Hausdorff space X, and let a be a point in
X \ F. Since X is Hausdorff, for every x ∈F there are open disjoint neighborhoods U x and
U ax around x and a, respectively. The class {U x : x ∈ F } is an open cover of F. Since closed
subsets of a compact space are compact, F is compact, and hence there is a finite subcover
n n
{U x1 ,U x 2 ,… ,U xn } of {U x : x ∈ F } . Then ∩U xi and ∪U axi is a separation of F and a by two
i =1 i =1
open sets.

The examples that follow will give us a nonhereditary normal space, showing, as promised,
that Theorem 3, Section 8.1 fails for normal spaces.

Example 1
Recall the well-ordered set TP defined in Example 2, Section 1.3: briefly, it is the
closed interval [1, ω], containing a copy of + , followed (with respect to the order
in TP) by a left segment of a well ordered copy of the set of real numbers bounded
on the right by the least real number ω that is preceded by uncountably many ele-
ments. The order topology (Example 9, Section 3.3) makes it a topological space.
Explicitly, this means that the intervals of type [1, b), (a, b) and (a, ω] constitute a
basis for TP; in this section we will call this basis a standard basis for TP. We will
now show that TP is a compact Hausdorff space. Hence, by Theorem 1, TP is a
normal space.
Hausdorff-ness: Consider any a, b ∈[1, ω]; we may suppose a < b. If there is no
x ∈[1, ω] such that a < x < b (i.e., if a is the immediate predecessor of b), then the
sets [1, b) and (a, ω] are two disjoint (basic) open sets separating a and b. If there is an
x ∈[1, ω] such that a < x < b , then [1, x ) and ( x , ω] are two disjoint (basic) open sets
separating a and b.
Compactness: First we show that TP is a Bolzano–Weirstrass space. It is easy to
see that ω is an accumulation point of every uncountable subset of TP (Exercise 3).
Thereby it suffices to show that every infinite countable set has an accumulation point.
Let A be an infinite countable set in the space TP, and let x be the least element of

K12146_C008.indd 195 2/9/15 3:02 PM

 196   ◾    Separation Properties

TP that is preceded by infinitely many elements of A. We show that x ∈ A′. Take any
standard basis neighborhood U of x in TP. If U = [1, b) for some b, then it is obvious
that U ∩ A ≠ ∅. Consider the case when U = (a , b ) or U = (a , ω ] for some a and b.
There must be an element y < x such that y ∈U , or else a would be an immediate
predecessor of x larger than infinitely many elements of A, contradicting the choice
of x. If U ∩ A = ∅ then y would be an upper bound of infinitely many elements of A
smaller than x, again contradicting the choice of x. So, U ∩ A ≠ ∅ for every open set
U containing x, showing that x ∈ A′ , and establishing that TP is a Bolzano–Weirstrass
space.
This result, together with Proposition 1, Section 7.4, implies that TP is count-
ably compact. In order to show compactness it suffices to extract a countable sub-
cover from any cover O consisting of standard basis sets for TP = [1, ω]. Since ω is
covered, some (a,ω] is in O. Then [1, a] = TP \ (a, ω] is countable, or else a would
have uncountably many predecessors, contradicting the choice of ω. For every
x ∈[1, a] choose Ox ∈O containing it. Then {Ox ∈O : x ∈[1, a]} ∪ {(a, ω]} is a count-
able subcover of O . ☐

We are ready to give the example we promised.

Example 2. Tychonoff Plank; A Normal Space with a Non-normal Subspace)

Let ω 0 be the smallest element in TP larger than all natural numbers (so that
[1, ω 0 ] = {1, 2 , 3,… , n ,… , ω 0 }). Since [1, ω 0 ] is a closed subspace of the compact space
TP = [1, ω], it is also compact. It is clear that [1, ω 0 ] is Hausdorff. The product space
X = [1, ω] × [1, ω 0 ] is called the Tychonoff plank (see Illustration 8.11). Since a product
of Hausdorff and compact spaces is Hausdorff and compact, the space X is Hausdorff
and compact. By Theorem 1, X is normal. However, the subspace Y = X \ {(ω , ω 0 )}—the
deleted Tychonoff plank—is not normal.
{ω} × [1,ω0)

ω0 [1,ω] × {ω0}
(ω,ω0)

n
an

3 a3
2
a2
1
1 2 3 4 ω0 a1 x ω

Illustration 8.11 Tychonoff plank; the points ai and x will emerge further in our argument.

The sets A1 = {ω} × [1, ω 0 ) and B1 = [1, ω] × {ω 0 } (see Illustration 8.11) are closed in X,
and so A = A1\{(ω , ω 0 )} and B = B1\{(ω , ω 0 )} are closed in Y. We will show that A and
B cannot be separated with two disjoint open sets.

K12146_C008.indd 196 2/9/15 3:02 PM

 8.3 Normal Spaces and Subspaces   ◾    197  

Let U be an open subset of Y such that A ⊂ U (see Illustration 8.12). Since

(ω , n ) ∈ A ⊂ U , we have that U ∩ ([1, ω] × {n}) ≠ ∅ , for every n ∈{1, 2 ,…} . Denote the
least element of U ∩ ([1, ω] × {n}) by an , and the image of an under the projection
[1, ω] × {n} → [1, ω] by bn . The set Z = {bn : n = 1,2,…} is a countable set. Observe that
none of bn is ω, since otherwise U ∩ ([1, ω] × {n}) consists only of the pair (ω , n ) ,
contradicting the fact that U ∩ ([1, ω] × {n}) is open in [1, ω] × {n}.
Since ω > bn for every n ∈+ , the set of all elements in [1, ω] larger than all ele-
ments in Z is not empty, and thus has the least element x. We note that x ≠ ω since
∞
x is preceded by the elements in the countable set ∪[1, bi ], while ω has uncountably
i =1
many predecessors.
We proved that there exists an x ∈[1, ω ) such that ( x , ω ) × n ⊂ U for every n (where,
to avoid misunderstanding, we point out that here (x, ω) denotes an open interval).
This means that ( x , ω ) × [1, ω 0 ) is a subset of U (as indicated in Illustration 8.12). Our
argument will be complete once we show that ( x , ω ) × [1, ω 0 ) has a nonempty intersection
with every open subset V of Y such that B ⊂ V .

{y} × {u}
ω0
V

n
U

3
2
1
1 2 3 4 ω0 x ω

Illustration 8.12

Let V be an open subset of Y such that B ⊂ V . Choose an element y in the inter-

val (x, ω). The set ({ y} × [1, ω 0 ]) ∩ V is open in { y} × [1, ω 0 ] and by construction
it contains the point { y} × {ω 0 } . So, there is an open interval ( z , ω 0 ] in [1, ω 0 ]
such that { y} × {ω 0 } ∈{ y} × ( z , ω 0 ] ⊂ ({ y} × [1, ω 0 ]) ∩ V . Finally, notice that for every
u ∈( z , ω 0 ], the point { y } × {u} is in both V and U, the latter being true since
y > x . We proved that U ∩ V ≠ ∅ and that Y is not normal. ☐

We have established the truth of the following claim.

Proposition 2. There exists a normal space with a non-normal subspace.

Remark: The whole construction of Tychonoff’s plank and the subsequent examples and
counterexamples are all based on a well ordering of . The existence of the latter is a
consequence of the Axiom of Choice. However, there is no known (and relatively simple)
explicit well ordering of , and so our constructs are existential only.

K12146_C008.indd 197 2/9/15 3:02 PM

 198   ◾    Separation Properties

Exercises
1. Give an example of a compact space that is not normal.
2. Let X be a compact and Hausdorff space and ~ an equivalence relation over X. Show
that the quotient mapping q : X → X~ is closed if and only if the set {( x , y ): x ~ y } is
a closed subset of X × X .
3. Show that ω is an accumulation point for TP, and that there is no sequence in TP \{ω}
converging to ω. [Hint for the second part: The argument is essentially the same as
the one provided for the set Z in Example 2.]
4. Show that the point (ω, ω0 ) in the Tychonoff plank is not an intersection of countably
many open sets. [Hint: Exercise 3.]
5. A space X is perfectly normal if it is normal and each closed subset is a countable
intersection of open sets.
(a) Show that being perfectly normal is hereditary.
(b) Show that the Tychonoff plank is not perfectly normal.

K12146_C008.indd 198 2/9/15 3:02 PM

 Chapter 9

Urysohn, Tietze,
and Stone–Čech

I n this chapter we focus on a few important theorems.

9.1 The Urysohn Lemma

A brief historical note: Pavel (Paul) Urysohn and Pavel (Paul) Alexandroff were mentioned
together several times earlier in the book. They were together one summer day in 1924 when
they decided to take a swim in the rough waters of the Atlantic Ocean, off the coast of Brittany,
France. Urysohn drowned. He was only 26.

{ }
Recall that D = m2i : m ∈, i ∈ is the set of dyadic numbers. According to Exercise 8,
Section 3.4, this set is dense in . It is then easy to show that D ∩[0, 1] is dense in [0,1].

Theorem 1. (The Urysohn Lemma) Let X be a T1 -space. Then X is normal if and only if for
every two disjoint closed sets A and B there is a continuous mapping f : X → [0, 1] such
that f ( A) = {0} and f ( B) = {1}.

Proof.
⇐ Take any two disjoint closed subsets A and B of X. By assumption there is a continu-
ous f : X → [0,1] with f ( A) = {0} and f ( B) = {1}. Let a and b be any two numbers such that
0 < a < b < 1. Then, f −1 ([0, a)) and f −1 ((b,1]) are two disjoint open sets containing A and B,
respectively.
⇒ Suppose that X is normal, and let A and B be two disjoint closed subsets of X. Then
A ⊂ Bc , and since Bc is open, we can use Theorem 1, Section 8.2 to conclude that there is
an open set U 1 2 satisfying A ⊂ U 1 2 ⊂ U 1 2 ⊂ Bc . (The reason we use dyadic numbers as sub-
sripts will become clearer as we go.) Denote A = U 0 and Bc = U1, and define f1 : X → [0,1]
as follows: f1 ( x ) = inf{t ∈{0, 1 2 ,1} : x ∈U t } if x ∈U1, and f1 ( x ) = 1 if x ∈ B. This means that
199

K12146_C009.indd 199 2/12/15 4:43 PM

 200   ◾    Urysohn, Tietze, and Stone–Čech

f1 sends the points of A to 0, sends the points of U 1 2 \ A to 1 2 , and sends the rest of the
points of X to 1. The graph of f1 is shown below, and the caption explains our intention.

B
B
X U½ \ A X
A A

Illustration 9.1 We start with the closed subsets A and B of X and with a function shown to the
left (sending A to 0 and B to 1). Then we build a rough stairway from A to B via the function
f1, the graph of which is shown to the right.

Apply Theorem 1, Section 8.2 to U 0 ⊂ U 1 2 and to U 1 2 ⊂ U1 to establish the existence of open

sets U 1 4 and U 3 4 such that U 0 ⊂ U 1 4 ⊂ U 1 4 ⊂ U 2 4 and U 2 4 ⊂ U 3 4 ⊂ U 3 4 ⊂ U1 (where we
have written the index 1 2 as 2 4 ). We get: U0 ⊂ U 1 4 ⊂ U 1 4 ⊂ U2 4 ⊂ U 2 4 ⊂ U 2 4 ⊂ U 3 4 ⊂ U1 .
Now define f 2 ( x ) = inf{t ∈{ i 22 : i = 0,1,2,3,4} : x ∈U t } if x ∈U1, and f1 ( x ) = 1 if x ∈ B. The
graph of f 2 is shown in Illustration 9.2.
Proceeding inductively, in the k­-th step we get U 0 ⊂ U 1 k ⊂ U 1 ⊂ U 2 k ⊂ U 2 ⊂ 
2 2k 2 2k
⊂ U(2k −1) ⊂ U(2k −1) ⊂ U1 and f k ( x ) = inf{t ∈{ i 2k : i = 0,1,…,2k } : x ∈U t } if x ∈U1, f k ( x ) = 1
2k 2k
if x ∈ B.

B
B U⁷⁄ ₈ \U⁶⁄ ₈
U⁶⁄ ₈ \U⁵⁄ ₈
U³⁄ ₄ \U²⁄ ₄ U⁵⁄ ₈ \U⁴⁄ ₈
U⁴⁄ ₈ \U⅜
U¹⁄ ₄ \ A U²⁄ ₄ \U¹⁄ ₄ U¹⁄ ₈ \ A U²⁄ ₈ \U¹⁄ ₈ U⅜ \U²⁄ ₈
A A

Illustration 9.2 We show the graphs of the functions f2 (left) and f3 (right); our stairway becomes
more refined.

At this point we observe the following:

(i) If U t1 and U t2 are two distinct open sets we get in this procedure and if U t1 ⊂ U t2 ,
then t1 < t 2 .
(ii) For each dyadic number t in the interval [0,1], there is an open set U t introduced in
some step of the above procedure.

Define f : X → [0, 1] by f ( x ) = lim k→∞ f k ( x ) . We will now show that f is the promised
smooth transition from the ­A-plateau at altitude 0 to the ­B-plateau at altitude 1 (see
Illustration 9.3).

K12146_C009.indd 200 2/12/15 4:43 PM

 9.1 The Urysohn Lemma   ◾    201  

A
Illustration 9.3 The function f provides a continuous transition from A (at altitude 0) to B
(at altitude1).

It follows from (i) that for every x ∈ X , the sequence ( f k ( x ))∞k =1 is nonincreasing. Since it is
bounded from below it must converge. So the mapping f is well defined. Since f k (a) = 0 for
every a ∈ A and for every k =1, 2, … , it follows that f ( A) = {0}. Similarly, since f k (b) = 1
for every b ∈B, and for every k =1, 2, … , it follows that f ( B) = {1}.
It remains to be shown that f is continuous. First, observe (directly from the definitions
of f k and f ) that f ( x ) = inf{t ∈D : x ∈U t } if x ∈U1, and f ( x ) = 1 if x ∈ B. The continuity
of f follows from the following two properties, and from Exercise 12, Section 3.5 (since the
intervals of types [0, a) and (b,1] form a subbasis for the topology over [0,1]):

(1) f −1 ([0, a)) = ∪ U t

t <a

( )
c
(2) f −1 ((b,1]) = ∪ U t .
t >b

(1) Take any x ∈ f −1 ([0, a)) . This means that f ( x ) ∈[0, a) . Since D is dense, there is
an element t0 ∈D with f ( x ) < t0 < a. If x ∉U t0 then inf{t ∈D : x ∈U t } ≥ t0 , and
so f ( x ) ≥ t0 . So, it must be that x ∈U t0 . Hence, x ∈∪ U t . We have proven that
t <a
f −1 ([0, a)) ⊂ ∪ U t .
t <a
  Conversely, suppose x ∈∪ U t . It follows that x ∈U t0 for some t0 < a . The observa-
t <a
tion (i) further above implies that x ∈U t for every t > t0 . So inf{t ∈D : x ∈U t } ≤ t0 , so
f ( x ) ≤ t0 , and since t0 < a , we get that f ( x ) ∈ [0, a). Consequently, f −1 ([0, a)) ⊃ ∪ U t .
t <a
So f −1 ([0, a)) = ∪ U t and (1) is proven.
t <a
(2) Take any x ∈ f −1 ((b,1]) , so that b < f ( x ) ≤ 1. Since D is dense in [0,1], there are two
dyadic numbers t1 and t 2 such that b < t1 < t 2 < f ( x ). The last part of this inequal-
ity implies that x ∉U t2 . The inequality t1 < t 2 gives U t1 ⊂ U t1 ⊂ U t2 , and so x ∉U t1 .
Consequently, x ∈(U t1 )c , and we conclude that f −1 ((b,1]) ⊂ ∪(U t )c .
c t >b
  Conversely, suppose x ∈∪(U t )c . Hence x ∈U t0 for some to > b , which means
t >b
that x ∉U t0 . This in turn implies that x ∉U t0 , and so f ( x ) > t0 . Recalling that
to > b we conclude that f ( x ) ∈ (b, 1], i.e., that x ∈ f −1 ((b,1]) . We have proven that
∪(U t )c ⊂ f −1 ((b,1]) and so we have established that (2) is true.
t >b

K12146_C009.indd 201 2/12/15 4:43 PM

 202   ◾    Urysohn, Tietze, and Stone–Čech

Here is a simple consequence.

Corollary 2. There is a completely regular space that is not normal.

Proof. Since the Tychonoff’s plank TP is normal, the Urysohn lemma (applied to a singleton—
which is a closed set since TP is T1—and to a disjoint closed set) implies that it is completely
regular. Consider the subspace Y of TP defined in Example 2, Section 8.3. We showed there that
Y is not normal. On the other hand, by Proposition 5, Section 8.2, Y is completely regular.

The following result, based on the Urysohn lemma, shows that small regular T1 spaces,
those that have countable bases, are metrizable.

Theorem 3. (Urysohn)1 Every second countable T3 space is metrizable.

Proof. First we give a short synopsis of the proof. Start with a countable basis for X, then
extract pairs ( B∗ , B) of basis sets such that B∗ ⊂ B. By the Urysohn lemma, each such pair
gives rise to a mapping onto {0,1} separating B∗ and Bc . Countably many such mappings
will give rise to an embedding from X into the Hilbert spaces H, and since H is a metric
space, so is X.
Let B = {Bi : i ∈+ } be a countable basis for a regular space X. We may suppose that
∅ ∉B , so that there exists an element xi ∈ Bi for each i ∈+ . By Theorem 2, Section
8.2, for every xi and Bi , there exists an open set U i such that xi ∈U i ⊂ U i ⊂ Bi . Since
B is a basis, for each such U i there exists a set Bx∗i ∈B such that xi ∈ Bx∗i ⊂ U i , and so,
since Bx∗i ⊂ U i , we have established that for every Bi ∈B , and for every xi ∈ Bi , there
exists some Bx∗i ∈B such that xi ∈ Bx∗i ⊂ Bx∗i ⊂ Bi . The set of all such pairs Bx∗i , Bi is ( )
countable, and so, after reindexing 2 if necessary, we can denote the elements of that
(
set by Bi∗ , Bi , i ∈+ . )
By Corollary 4, Section 8.2, the space X is normal. By the Urysohn lemma,
the mapping that sends Bi∗ to {0} and Bic to {1} can be extended to a continu-
ous mapping fi : X → [0,1], i ∈+ . Choose an arbitrary x ∈ X and consider the
f 2 ( x ) f3 ( x ) fn ( x )
sequence of real numbers f1 ( x ), 2 , 3 ,… , n ,… . Since 0 ≤ fi ( x ) ≤ 1 , we have
( ) ( )
∞
fn ( x ) 2 fn ( x ) 2
n ≤ n12 and so the series ∑ n converges. This means that the sequence

( ( f ( x )) + ( ) + ( ) +  + ( ) )
n =1 ∞
f2 ( x ) 2 f3 ( x ) 2
( f (x ),
2 fn ( x ) 2 f2 ( x )
1 2 3 n converges, establishing that 1 2 ,
n =1
f3 ( x )
3 ,…, fn ( x )
n )
,… is an element of the Hilbert space H (Example 2, Section 5.2).
(
So far we have shown that F : x  f1 ( x ), 2 2 , 3 3 ,… , nn ,… is a mapping from X
f (x ) f (x ) f (x )
)
into H. We will show that it is an embedding; that will make X homeomorphic to a sub-
space of H. Since H is a metric space, so is every subspace of H, and so we would thus
establish that X is metrizable.

1
Theorem 3 is sometimes called the Urysohn–Tychonoff Metrization Theorem.
2
In order to unburden the notation, we are allowing some ambiguity here: Bi as a member of B , and Bi as the second
( )
coordinate of Bi∗ , Bi may be different sets; which of these we are using should be clear from the context.

K12146_C009.indd 202 2/12/15 4:43 PM

 9.1 The Urysohn Lemma   ◾    203  

Step 1. (F is continuous.) Choose an open neighborhood V of some F (x). Since H is a

metric space, there is an ε > 0 such that the open ball B = {y ∈H : d( F ( x ), y ) < ε} is a subset
of V. We are searching for an open neighborhood U of x such that F (U ) ⊂ B.
∞ 2 ∞
Find N such that ∑ 1
i2
< ε4 ; this can be done since the series ∑ i12 converges. Since
i = N +1 i =1
each of f1 , f 2 ,…, f N is a continuous mapping X →[0, 1], for each i ∈ {1, 2, … , N }
there is an open neighborhood U i of x such that fi (U i ) is a subset of the interval
N
( fi ( x ) − 2εN , fi ( x ) + 2εN ) ∩[0,1] , that is open in [0, 1] ⊂ . Denote U = ∩U i ; then, for every
i =1
y ∈ U , and for every i ∈ {1, 2, … , N }, we have fi ( y ) − fi ( x ) < 2εN .
We prove that F (U ) ⊂ B by establishing that d( F ( x ), F ( y )) < ε for every y ∈ U .
Deno­t ing F ( x ) = ( )
fi ( x ) ∞
i
i =1
and F ( y ) = ( ) fi ( y ) ∞
i
i =1
, our target-inequality d( F ( x ), F ( y )) < ε

( )
∞
fi ( x ) fi ( y ) 2
can thus be written as follows: ∑ i − i < ε . Focusing on the left-hand side of
i =1
this inequa­l ity, we have:
∞ 2 N 2 ∞ 2

∑ i =1
 fi ( x ) fi ( y ) 

i
−
i 
 ≤ ∑i =1
 fi ( x ) fi ( y ) 

i
−
i 
 + ∑
i = N +1
 fi ( x ) fi ( y ) 

i
−
i
 ≤

N 2 ∞ 2
≤ ∑ i =1
 fi ( x ) fi ( y ) 

i
−
i 
 + ∑
i = N +1
 fi ( x ) fi ( y ) 

i
−
i 
 =

N
fi ( x ) − fi ( y )
∞
( f i ( x ) − f i ( y ))2 N ∞
( f i ( x ) − f i ( y ))2 .
= ∑ i =1
i
+ ∑
i = N +1
i2
≤ ∑ f (x ) − f ( y ) + ∑
i =1
i i
i = N +1
i2
N
By assumption fi ( y ) − fi ( x ) < ε
2N for every i ∈ {1, 2, … , N }, so that ∑ fi ( x ) − fi ( y )
i =1
≤ N 2εN = 2ε . On the other hand, since fi (z ) ∈[0,1] for every z ∈X , we have that
∞ ∞
( fi ( x )− fi ( y ))2
( fi ( x ) − fi ( y ))2 ≤ 1, i ∈ {N , N + 1, …}, and so ∑ i2
≤ ∑ 1
i2
. We have assumed
i = N +1 i = N +1
∞ ∞ N
2 ( fi ( x )− fi ( y ))2 ε2
that ∑ 1
i2
≤ ε4 , hence ∑ i2
≤ 4 = 2ε . Summarizing all this: ∑ fi ( x ) − fi ( y )
i = N +1 i = N +1 i =1

( )
∞ ∞
( fi ( x )− fi ( y )) 2
fi ( x ) fi ( y ) 2
+ ∑ i2
≤ 2ε + 2ε = ε , and so ∑ i − i < ε for every y ∈U. We proved that
i = N +1 i =1
F is continuous.
Step 2. ( F : X → F ( X ) is open.) Let U be an open subset of X. For every x ∈U there is
Bn ∈B such that x ∈Bn ⊂ U . We observed in the first paragraph of the proof of this theorem
that there exists some Bn∗ ∈B such that x ∈ Bn∗ ⊂ Bn∗ ⊂ Bn ⊂ U . The mapping fn : X → [0,1] was
defined earlier; it is such that fn ( Bn∗ ) = {0} and fn ( Bnc ) = {1}. We will now show that the open
ball B( F ( x ), n1 ) in F (X) is a subset of F (U), thus establishing that F (U) is open in F (X).

( ) ( )
∞
fi ( x ) fi ( y ) 2 fn ( x ) fn ( y ) 2
For every y ∈ X \ U , we have d( F ( x ), F ( y )) = ∑ i − i ≥ n − n =
i =1
∗ c
= n = n1 , where fn ( x ) = 0 since x ∈ Bn , and fn ( y ) = 1 since y ∈Bn . We proved
fn ( x )− fn ( y )
0 −1
n
that d( F ( x ), F ( y )) ≥ n1 for every y ∈ X \ U .

K12146_C009.indd 203 2/12/15 4:44 PM

 204   ◾    Urysohn, Tietze, and Stone–Čech

Finally, take F ( z ) ∈ B( F ( x ), n1 ); this means that d( F ( x ), F ( z )) < n1 . By the last statement

in the preceding paragraph, z ∉ X \ U . Hence z ∈ U , and so F (z ) ∈ F (U ). We proved that
B( F ( x ), n1 ) ⊂ F (U ), as claimed.

Step 3. (F is one-to-one.) Let x and y be two distinct elements of X. Since X is T1 ,{ y}c is open,
( )
and there exists a pair Bn∗ , Bn of sets in B such that x ∈ Bn∗ ⊂ Bn∗ ⊂ Bn ⊂ y c. Then y ∉Bn ,
fn ( x ) = 0 and fn ( y ) = 1 , and so F ( x ) ≠ F ( y ).

Corollary 4. Every second countable regular space is metrizable.

Proof. By Corollary 4, Section 8.2, second countable and regular imply normal.

Exercises
1. Show that a space is normal if and only if every three closed pairwise disjoint sets can
be separated with three open sets.
2. Let X be a compact Hausdorff space. Prove that for every x , y ∈ X , x ≠ y , there is a
continuous function f : X →  such that f ( x ) ≠ f ( y ).
3. Show through a counterexample that we cannot replace the requirement that the sets
A and B are closed in the statement of the Urysohn lemma with the stipulation that
these two sets are open.
4. Show that if X is normal then for every two distinct real numbers a and b, and for
every disjoint closed sets A and B there is a continuous function f : X → [a, b] such
that f ( A) = {a} and f ( B) = {b}.
5. Prove that X is normal if and only if for every three pairwise disjoint closed sets A,
B, and C, and every three real numbers a, b, and c, there is a continuous function
f : X →  such that f ( A) = {a}, f ( B) = {b} and f (C ) = {c}.
6. Prove that X is normal if and only if for every n pairwise disjoint closed sets
A1 , A2 ,… , An and every (not necessarily distinct) real numbers a1 , a2 ,… , an there is a
mapping f : X →  such that f ( Ai ) = {ai } for every i ∈ {1, 2, … , n}.
7. Find an example of a normal space with countable many closed pairwise disjoint
subsets Ai in it, and find countably many points ai in  such that there is no con-
tinuous mapping f : X →  such that f ( Ai ) = {ai } for every i.
8. Denote the set of all continuous bounded functions  →  by C( ,  ). Show that
the usual topology over the domain space  is the coarsest topology that makes all
functions in C( ,  ) continuous. (In order to avoid any misinterpretation we point
again that the topology of the range space  in this problem is fixed to be the usual
topology.)

K12146_C009.indd 204 2/12/15 4:44 PM

 9.2 The Tietze Extension Theorem   ◾    205  

9.2 The Tietze Extension Theorem

A brief historical note: Tietze proved Theorem 1 in 1915, for the case when X is a metric
space. The general version appeared for the first time in an article by Urysohn, published in
1925.

Theorem 1. (The Tietze Extension Theorem) Let X be a normal space. For every closed
subset A of X, and every continuous function f : A →  there is a continuous exten-
sion fˆ : X →  of f. Moreover, if f : A → [a, b], then fˆ could be chosen to be such that
fˆ ( X ) ⊂ [a, b].

Proof.
In a nutshell, we start with f and then we truncate the extremities of its graph. The horizon-
tal plateaus we get by slicing down the peaks and those that we get by raising up the depres-
sions in the graph give rise to two closed subsets of X, and after applying the Urysohn
lemma to these two closed set, we get a continuous mapping fˆ1 : X →  that extends f and
approximates it over A. We repeat this procedure, where in all subsequent steps we trun-
cate successively less and less significant differences over A between the approximations
that we get and the starting function f. Details follow.
Case 1: f is bounded. We may (and we do) assume that f : A → [−1, 1] (Exercise 1). Since
f is continuous, the sets A1 = f −1 ([−1, − 1 3]) and B1 = f −1 ([ 1 3 ,1]) are closed in A, and so,
since A is closed in X, they are closed in X, too. By the Urysohn’s lemma (and Exercise 4,
Section 9.1) the mapping g 0 : A1 ∪ B1 → {− 1 3 , 1 3} defined by g 0 ( A1 ) = {− 1 3}, g 0 ( B1 ) = { 1 3}
can be extended to a mapping f1 : X → [− 1 3 , 1 3] . Our construction of f1 guarantees that
f ( x ) − f1 ( x ) ≤ 2 3 for every x ∈ A. For the sake of introducing a notation consistent with
what will come in this proof, we double-denote f1 by fˆ1.
This step and the next step in our construction are visualized in Illustration 9.4. The sets
A1 and B1 are the domains corresponding to the green (lower flat) and blue (upper flat)
parts of the graph of f1 , respectively. Our goal is, as we have stated in the theorem, to get a
continuous function over X that matches f over A. So far we are short of fulfilling the latter.
Since both f and fˆ1 A are continuous, so is the restriction of g 1 = f − fˆ1 over A, and
since f ( x ) − fˆ1 ( x ) ≤ 2 3 over A, the map g 1 sends A into [− 2 3 , 2 3]. Denote A2 = g 1−1[− 2 3 , − 2 9]
and B2 = g 1−1[2 9 , 2 3]. These are two closed disjoint subsets of A (and of X) over which f and
f1 (or fˆ1 ) differ by at most 2 9 . By Urysohn’s lemma, the mapping A2  {− 2 9} , B2  {2 9}
can be extended to a continuous mapping f 2 : X → [− 2 9 , 2 9]. Our construction implies that
g 1 ( x ) − f 2 ( x ) ≤ 4 9 for every x ∈ A.
Bearing in mind that the mapping f 2 , when restricted to A, approximates the differ-
ence between f and f1 , we define fˆ2 = f1 + f 2 . We compute: f ( x ) − fˆ2 ( x ) = f ( x ) − f1 ( x )
− f 2 ( x ) = g 1 ( x ) − f 2 ( x ) ≤ 4 9 over A, so that fˆ2 is a better (than f1 ) approximation of f
over A. (See Illustration 9.4.)

K12146_C009.indd 205 2/12/15 4:44 PM

 206   ◾    Urysohn, Tietze, and Stone–Čech

1.0 1.0 1.0

0.5
2 0.5 2
0.5
2
0.0 0.0 0.0
–0.5 –0.5 –0.5
–1.0 0 –1.0 0 –1.0 0
–2 –2 –2
0 0 0 –2
–2 –2
2 2
2

Illustration 9.4 The graph of a continuous function f (defined over A = [−3, 3] × [−3, 3]) is
shown to the right; the restrictions of the mappings f1 (= fˆ1 ) and fˆ2 to that square are shown
in the left and center graphics, respectively (their domain is the entire space X ). We see that
fˆ2 is an improved approximation of f. The green (bottom) and blue (top) plateaus correspond
to A1 and B1 (graphics to the left), and A2 and B2 (center graphics), respectively.

Next, we proceed recursively and define new mappings fn , g n , and fˆn in terms of the
continuous mappings fi : X →  − 23i , 23i  , fˆi = f1 + f 2 +  + fi−1, and g i = f − fˆi , satisfying
i −1 i −1

f ( x ) − fˆ ( x ) ≤ 2 i , i = 1, 2, … , n − 1.
i
i 3

Since f ( x ) − fˆn −1 ( x ) ≤ 2 3n−1 over A, the map g n−1 = f − fˆn−1 sends A into  − 2 3n−1 , 2 3n−1  .
n−1 n−1 n−1

Denote An = g n−−1 1  − 2 3n−1 , − 2 3n  and Bn = g n−−1 1  2 3n , 2 3n−1  . These are two closed dis-
n−1 n−1 n−1 n−1

joint subsets of A and of X over which f and fˆ differ by at most 2 n . By the Urysohn
n−1
n−1 3
lemma, the mapping An  − 3n , Bn  { 2n−1
}3n { 2n−1
}
can be extended to a continuous map-
ping fn : X →  − 3n , 3n  . Our construction implies that g n −1 ( x ) − fn ( x ) ≤ 2n 3n for
n−1 n−1
2 2

every x ∈ A..
Set fˆn = f1 + f 2 +  + fn−1, and compute: f ( x ) − fˆn ( x ) = f ( x ) − f1 ( x ) − f 2 ( x ) − − fn ( x ) =
n
g n−1 ( x ) − fn ( x ) ≤ 2 3n over A.
In Illustration 9.5 we show a short subsequence of { fˆn }.

1.0 1.0 1.0

0.5 0.5 0.5
0.0 2 , 0.0 2 , 0.0 2 ,
–0.5 –0.5 –0.5
–1.0 0 –1.0 0 –1.0 0
–2 –2 –2
0 –2 0 –2 0 –2
2 2 2

1.0 1.0 1.0

0.5 0.5 0.5
0.0 2 , 0.0 2 , 0.0 2
–0.5 –0.5 –0.5
–1.0 0 –1.0 0 –1.0 0
–2 –2 –2
0 –2 0 –2 0 –2
2 2 2

Illustration 9.5 From top left to bottom right corner: fˆ1 , fˆ5, fˆ9, fˆ13, fˆ17, and f. The maximal dif-
ference between fˆ17 and f over A is less than 216
317
≈ 0.0005.

K12146_C009.indd 206 2/12/15 4:44 PM

 9.2 The Tietze Extension Theorem   ◾    207  

n−1
Since fn ( x ) ≤ 2 for every x ∈ X , we have that f1 ( x ) + f 2 ( x ) +  + fn ( x ) ≤ f1 ( x ) +
3n
n
f 2 ( x ) +  + fn ( x ) ≤ ∑ 2 3i = 1 − ( 2 3 ) < 1 over X. It follows that lim fˆn ( x ) exists for
i −1 n

i =1 n→∞
∞
every x ∈ X , or, to say the same thing differently, the series ∑ fn ( x ) converges
∞ n =1
(and it converges absolutely). We define fˆ ( x ) = ∑ f ( x ) (= lim fˆ ( x )). Observe that n n
n =1 n→∞
f1 ( x ) + f 2 ( x ) +  + fn ( x ) < 1 implies that fˆ ( x ) ≤ 1 for every x ∈ X , so that fˆ ( X ) ⊂ [−1,1].
n −1
The inequality f ( x ) − fˆn ( x ) ≤ 2 3n , x ∈ A means that f ( x ) − ∑ fi ( x ) ≤ 2n 3n , and this
n

implies that fˆ ( x ) = f ( x ) for every x ∈ A. i =1

We will prove that fˆ is continuous, and, with that, the case when f is bounded
will be completed. We will do that by showing that fˆ is continuous at every point of
X. Let x ∈ X , and choose an open ball B( fˆ ( x ), ε) in  . Since each fi is continuous
(at x), for every positive integer N there is an open neighborhood Vi of x such that
N
f (V ) ⊂ B( f ( x ), ε ). Setting V = ∩ V , we have the following for every y ∈V : fˆ ( y ) − fˆ ( x ) =
i i i 2N i
i =1
∞ ∞ (1) ∞ ∞
∑ fi ( y ) − ∑ fi ( x ) = ∑ ( fi ( y ) − fi ( x )) ≤ ∑ fi ( y ) − fi ( x ) , where step (1) is true since the
i =1 i =1 i =1 i =1
∞ N ∞
series are convergent. Continuing: ∑ fi ( y ) − fi ( x ) = ∑ fi ( y ) − fi ( x ) + ∑ fi ( y ) − fi ( x ) .
i =1 i =1 i = N +1

Since we have assumed that fi ( y ) − fi ( x ) < ε 2 N for all y ∈V , i = 1, 2, …, N , it follows that

N
∑ fi ( y ) − fi ( x ) ≤ N ε 2 N = ε 2 . On the other hand,
i =1

∞ ∞ ∞ ∞

∑
i = N +1
fi ( y ) − fi ( x ) ≤ ∑(
i = N +1
fi ( y ) + fi ( x ) = ) ∑(
i = N +1
) ∑ ( f (x ) )
fi ( y ) +
i = N +1
i

∞ ∞ ∞

≤ ∑
i = N +1
2i −1
i
3
+ ∑
i = N +1
2i −1
i
3
= ∑
i = N +1
2i
3i
=3 ( 2 N +1
3N +1 ).

Note that this is true for every N. At this point we simply choose N such that
( )
N ∞
2 N +1
3 N +1 < 2 . With that we have ∑ f i ( y ) − f i ( x ) + ∑ fi ( y ) − fi ( x ) < ε 2 + ε 2 = ε , which,
ε
3
i =1 i = N +1
backing up a bit, means that fˆ ( y ) − fˆ ( x ) < ε , for every y ∈ V . We showed that fˆ sends V
into B( fˆ ( x ), ε) , and so we established the continuity of fˆ at x.
The case when f : A → [a, b], a < b, can now be handled easily. Take any homeomorphism
h :[a, b] → [−1, 1]. By what we have done so far, g = h  f : A → [−1, 1] can be extended to
gˆ : X → [−1,1] . Then, h −1  gˆ : X → [a, b] is the desired extension of f.
Case 2: f : A →  is not bounded. The mapping h( x ) = tan −1 ( x ) is a homeomorphism
from  onto the interval (−1, 1). By Case 1, the mapping g = h  f : A → (−1, 1) can be
extended to gˆ : X → [−1,1] . This means that ĝ A = h  f , and so h −1  gˆ A = f . If we can
choose ĝ so that its range is (−1, 1) rather than [−1, 1], then that would fit with the domain
of h −1 , and we could simply take h −1  gˆ = fˆ : X →  . In the last part of the proof we show
that we can make that choice without loosing any generality.

K12146_C009.indd 207 2/12/15 4:44 PM

 208   ◾    Urysohn, Tietze, and Stone–Čech

If gˆ −1 ({−1,1}) = ∅ then there is nothing to do, and we can take fˆ = h −1  gˆ . Suppose

gˆ −1 ({−1,1}) = B ≠ ∅ . Then B is a closed set, and, since gˆ( A) ⊂ (−1,1), it is disjoint from A. By
the Urysohn lemma and its proof there is a continuous mapping k : X → [0.5, 1] sending B to
0.5 and A to 1 and the rest of X to (0.5, 1). Then the mapping kgˆ defined by kgˆ( x ) = k( x ) gˆ( x )
agrees with ĝ over A, since for every a ∈ A, kgˆ(a) = k(a) gˆ(a) = gˆ(a). Further, kgˆ sends X
into (−1, 1) as we explain if x ∈ A, then kgˆ(a) = gˆ(a) ∈(−1,1). If x ∈ X \ A, then k( x ) < 1, so
that, since gˆ( x ) ≤ 1, we have that k( x ) gˆ( x ) < 1. We have achieved what we wanted, and we
may now take the desired extension to be h −1  (kgˆ ).

The following generalization of Theorem 1 is easy to prove (Exercise 5).

Theorem 2. If X is a normal space, and if A is a closed subset of X, then every continuous

mapping f : A → n can be extended to a continuous fˆ : X → n .

As an illustration of the applicability of the Tietze extension theorem we will now give
another justification of the claim that the Sorgenfrey plane is not normal. Later (in
Section 11.3) we will again use the Tietze extension theorem to establish another important
classical result, the Jordan curve theorem.

Example 1
Suppose the Sorgenfrey plane is normal, and consider the set of points A on any
line of slope −1. We have seen earlier that A inherits the discrete topology from the
Sorgenfrey plane. Consequently, any mapping from A into  is continuous. In par-
ticular, that is true for f : A →  that sends the set Q of all points at rational distance
from (0,0) to 2, and the set P of all points at irrational distance from (0,0) to –2. By
the Tietze extension theorem, this is extendable to a continuous mapping fˆ :  2 → ,
where, of course,  2 is equipped with the Sorgenfrey topology. Consider the open
sets U = fˆ −1 (1,3) and V = fˆ −1 (−3, −1); since the intervals (1, 3) and (−3, − 1) are dis-
joint, so are U and V.
Since 2 ∈ (1, 3), the open set U contains the set Q; similarly V contains P. Since
the Sorgenfrey plane topology is generated by rectangles of type [a, b) ×[c , d ), both
U and V are unions of such rectangles, the first one containing Q as a subset, the
second one containing P as a subset. That such two unions must have a nonempty
inter­section was an exercise a long time ago: Exercise 11, Section 2.1. Consequently,
U ∩ V cannot be empty, leading to an obvious contradiction. Hence the Sorgenfrey
plane is not normal. ☐

Exercises
1. Let X be a normal space and let Y be any space homeomorphic to the interval [0,1].
Show that for every closed subset A of X, and every continuous function f : A → Y
there is a continuous extension fˆ : X → Y .

K12146_C009.indd 208 2/12/15 4:44 PM

 9.3 Stone–Čech Compactification   ◾    209  

2. Show that for every proper closed subset A of a normal space X, and for every continu-
ous mapping f : A →  , there are at least  -many continuous extensions fˆ : X →  .
3. Deduce the Urysohn lemma from the Tietze extension theorem.
4. Find a normal space X, a closed subset A of X, and a continuous mapping f : A → S ,
S ⊂  , such that f cannot be extended to a continuous mapping fˆ : X → S .
5. (a) Show that if X is a normal space, and if A is a closed subset of X, then every con-
tinuous mapping f : A → n can be extended to a continuous fˆ : X → n .
(b) Generalize (a) as follows: Let Y be a space such that for every normal space X, for
every closed subset A of X, and for every continuous mapping f : A → Y , there is
a continuous mapping fˆ : X → Y extending f. Show that the same is true for the
product space Y n in place of Y.
6. The mapping S1 ∪ {(0,0),(2,2)} →  2 that is the identity over S1 and interchanges
(0,0) and (2,2) is continuous. According to 5(a) it can be extended to a continuous
mapping  2 →  2 . Describe or visualize one such extension.
7. Show that if X is a T1 space such that for every closed subset A of X and every continu-
ous mapping f : A →  there is an extension fˆ : X →  of f, then X is normal.

9.3 Stone–Čech Compactification

A brief historical note: What is now called Stone-Čech compactification occurs in a paper
published by Tychonoff in 1930. Eduard Čech (1893–1960) and Marshal Stone (1903–1989)
published their articles independently in 1937.

We defined the one-point compactification X ∞ of a space X in 6.4; it turned out to be

Hausdorff, provided X is locally compact and Hausdorff. In this section we define and
briefly consider the important Stone-Čech compactification of X, standardly denoted by
β(X), and possessing the important additional property of extendability of bounded con-
tinuous functions X →  to bounded and continuous functions β( X ) → . Example 1
illustrates the main difference between the one-point compactification and the Stone–
Čech compactification.

Example 1
Recall that the one-point compactification of the open disk B2 = {( x , y ): x 2 + y 2 < 1}
is (homeomorphic to) the sphere S 2 = {( x , y , z ): x 2 + y 2 + z 2 = 1} (see Example 3,
Section 7.5). The mapping f : B2 →  defined in polar coordinates by
( )
f (ρ, θ) = sin 1−ρ
1
, 0 ≤ θ < 2 π, 0 ≤ ρ < 1, and graphed in Illustration 9.6, cannot be

K12146_C009.indd 209 2/12/15 4:44 PM

 210   ◾    Urysohn, Tietze, and Stone–Čech

extended to a con­tinuous mapping S 2 →  . This is an easy consequence of, say,

Proposition 2, Section 3.5. However f can be extended to a continuous mapping
β( B2 ) → .

0.5

1.0
0.0 0.5
0.0
–0.5
–1.0
–0.5

–0.5

0.0

0.5

Illustration 9.6 The graph of f (ρ, θ) = sin ( ), 0 ≤ θ < 2π, 0 ≤ ρ < 1.

1
1−ρ ☐

The extendability property of X →  to β( X ) →  is significant and has many appli-

cations in topology, analysis, and elsewhere.

Consider the metric space C ∗ ( X ) of all bounded and continuous functions X →  ,

where X is, for the time being, any space, and where the metric is again the supremum
metric ρ( f , g ) = sup{ f ( x ) − g ( x ) : x ∈ X }. The Least Upper Bound Property of the set of
real numbers, and the fact that we consider bounded functions X →  , easily imply that ρ
is indeed a metric. An open ball around a bounded function f :  →  with respect to this
metric is shown in Illustration 9.7.

f+ε
f
f–ε

Illustration 9.7 The ball B( f , ε ) in the met-

ric space of all continuous and bounded
functions  → .

K12146_C009.indd 210 2/12/15 4:44 PM

 9.3 Stone–Čech Compactification   ◾    211  

Assume a j < b j for every j ∈ J. A set S = { f j : X → [a j , b j ] : j ∈ J } of continuous functions

separates points from closed sets if for every x ∈ X and every closed subset F of X that
does not contain x, there exists f j ∈S such that f j ( x ) = a j and f j ( F ) = {b j } .
The concept of complete regularity can now be restated in terms of the set C ∗ ( X ).

Proposition 1. A space X is completely regular if and only if the set C ∗ ( X ) separates points
from closed sets.

Proof. ⇒ This follows directly from the definition of complete regularity.

⇐ Assume that for every F ⊂ X and x ∈ X \ F , there exists a bounded function
closed
f : X → [a, b] such that f ( x ) = a and f ( F ) = {b}. Let g :[a, b] → [0, 1] be the function defined
by g (t ) = bt −−aa . Then the composition h = g  f is continuous and such that h( x ) = 0 and
h( F ) = {1}, confirming that X is completely regular.

The preliminaries will be completed with the following simple result.

Proposition 2. If X is completely regular, then the weak topology of X generated by C ∗ ( X )

is the same as the original topology of X.

Proof. Recall that the weak topology over X generated by C ∗ ( X ) has the family
{ f −1 (U ) : U is open in  , f ∈C ∗ ( X )} as a subbasis (Proposition 6, Section 3.5). Denoting this
topology by T ′, and the original topology over X by T , we observe that, since the functions
C ∗ ( X ) are continuous with respect to T , these subbasis-sets are in T , showing that T ′ ⊂ T .
We show that T ⊂ T ′. Choose any open set U ∈T and a point x ∈ U . Then, by
Proposition 1, x and U c can be separated by some f ∈C ∗ ( X ); this means that f ( X ) = [a, b],
f ( x ) = a and f (U c ) = {b}. Now, for every c ∈ (a, b), f −1 ([a, c )) is an open set containing x
and disjoint form U c . The latter makes f −1 ([a, c )) a subset of U, showing that x is an interior
point for U with respect to T ′. So, U ∈T ′ .

Constructing the Stone-Čech Compactification

Before we begin with the construction we urge the reader not to be intimidated by
the apparent largeness of the elements of β(X)—even when X is a singleton, the ele-
ments of β(X) have continuum many coordinates—for the construction of β(X) is
very short.
We will need Proposition 2, and so our starting space X will now be assumed to be
completely regular. We will also assume that X is a T1 -space (thus becoming a T3 1 -space).
2
For every f ∈C ∗ ( X ) denote any closed and bounded interval that contains f (X)
by I f (we may take I f to be the smallest interval containing f (X)). Since each I f is
compact and Hausdorff, by Tychonoff’s theorem1 and by Theorem 4, Section 8.1, so
is the product space ∏ I f . Recall the formal definition of the set ∏ I f noted
f ∈C ∗ ( X ) f ∈C ∗ ( X )

1
Here is where the Axiom of Choice comes into play: as we have noted earlier, Tychonoff theorem is equivalent to the
Axiom of Choice.

K12146_C009.indd 211 2/12/15 4:45 PM

 212   ◾    Urysohn, Tietze, and Stone–Čech

way back in Section 0.1: the elements of this set are mappings γ : C ∗ ( X ) → ∪ If
f ∈C ∗ ( X )
such that γ ( f ) ∈I f .
Define a mapping e : X → ∏ I f as follows: e(x) is that γ ∈ ∏ I f for which
f ∈C ∗ ( X ) f ∈C ∗ ( X )
γ ( f ) = f ( x ), or, in a bit more compressed form, e( x )( f ) = f ( x ).
Since the mapping e is crucial in our construction, we will describe it more explic-
itly. Assume for a moment (only until the end of the next sentence) that C ∗ ( X ) is
countable, so that we can write it as follows: C ∗ ( X ) = { f1 , f 2 ,… , fn ,…} . Then e(x),
defined as above, is simply ( f1 ( x ), f 2 ( x ),… , fn ( x ),…). In general, e( x ) = ( f ( x )) f ∈C∗ ( X ),
i.e., it is a “ C ∗ ( X ) -tuple” whose coordinate corresponding to any f ∈C ∗ ( X ) is f (x).
Our construction of the Stone–Čech compactification β(X) needs one more very
short step: we define β( X ) = e( X ).

Theorem 3. (Stone–Čech) Let X be a completely regular T1 -space. The following is true.

(a) β(X) is a compactification of X.
(b) β(X) is Hausdorff.
(c) Every bounded continuous function f : X →  has a bounded and continuous exten-
sion F : β( X ) →  , in the sense that F (e( x )) = f ( x ) for every x ∈ X .

Proof.
(a) This part consists of three claims:
(i) e is an embedding.
(ii) e(X) is dense in β(X).
(iii) β(X) is compact.

The statements (ii) and (iii) are justifiable virtually for free: e(X) is dense in β(X) because
β( X ) = e( X ), and β(X) is compact since it is the closed subset e( X ) of the compact space
∏ I f . Part (i) merits three separate paragraphs.
( )
f ∈C ∗ ( X )
Denote the projection ∏ I f → I f0 by p f0 . Then ( p f0  e )( x ) = p f0 ( f ( x )) f ∈C∗ ( X ) = f0 ( x ) .
f ∈C ∗ ( X )
Since each f0 is continuous, so is p f0  e . Since this is true for every projection p f , f ∈C ∗ ( X ),
it follows from Proposition 2, Section 5.2 that e is continuous.
Showing that e is one-to-one is equally easy; it is here that we will need our assumption
that X is completely regular. Suppose x1 ≠ x 2. Since X is T1, singletons are closed. Since X is
completely regular, there is a function g : X → [0, 1] separating x1 and x 2 . Then, obviously,
e( x1 ) and e( x 2 ) have different g-coordinates, and so e( x1 ) ≠ e( x 2 ).
To complete the proof of part (i) we need to show that e : X → e( X ) is open, or, equiv-
alently, that e −1 : e( X ) → X is continuous. It suffices to show that e(S) is open in e(X) for
every S in some subbasis S of X (see Exercise 12, Section 3.5). We noticed in the proof
of Proposition 2 (which we may use, since X is assumed to be completely regular) that
{ f −1 (U ) : U is open in  , f ∈C ∗ ( X )} is a subbasis of X. It is obvious in Illustration 9.9 that
the diagram shown in Illustration 9.8 commutes for every f0 ∈C ∗ ( X ).

K12146_C009.indd 212 2/12/15 4:45 PM

 9.3 Stone–Čech Compactification   ◾    213  

X
e
П* If
x
e
П* f (x)
f C (X ) f C (X )

Pf0 Pf0
f0 f0

If f0(x)
0

Illustration 9.8 Illustration 9.9

Consequently, e( f0−1 (U )) = p −f01 (U ) ∩ e( X ) , and since p −f01 (U ) ∩ e( X ) is open in e(X), so

is e( f0−1 (U )).

(b) We have noticed earlier that ∏ I f is Hausdorff. Since Hausdorff-ness is heredi-

f ∈C ∗ ( X )
tary, it follows immediately that β(X) is also Hausdorff.
(c) Given any continuous and bounded f0 : X →  , the commutativity of the dia-
gram in Illustration 9.8 implies that the restriction of p f0 to β(X) is the desired extension
p f0 : β( X ) →  .

When the (completely regular) space X is compact, then e(X) is also compact. Since
∏ I f is Hausdorff, it follows from Proposition 2, Section 7.2 that e(X) is closed, and
f ∈C ∗ ( X )
so e( X ) = e( X ) . Consequently, when X is compact, then β(X) is homeomorphic to X. So,
unlike one-point compactification, the Stone–Čech compactification is more efficient in
the sense that it does not alter the original space when it is compact.

Example 2: β(
)
β(  ) is one of the most studied Stone–Czech compactifications, both from the topo-
logical and from the analytical point of view. We devote a few paragraphs to it.
View X =  as a subspace of  . Then C ∗ ( ) consists of all bounded sequences1
( f (n))n∞= 0 in  . Since C ∗ ( ) has 2ℵ0 elements, each element of ∏ I f has 2ℵ0
f ∈C ∗ (  )
coordinates. Each of these coordinates is a term of some bounded sequence in  . The
elements of e( ) are of type ( f (n)) f ∈C∗ (  ) , where n is a fixed natural number. Since
 is not compact, neither is e( ). Consequently e( ) is not closed, and thereby
there are accumulation points for e( ) out of e( ) (which means that e( ) is a
proper subset of β( )).
Note that we have only claimed existence of the accumulation points of e( ); we
do not have a simple explicit description of a single such point. One of the difficulties
in describing elements of (e( ))′ is that no point of (e( ))′ \ e( ) is a point of con-
vergence of a sequence of points in e( ).
We prove the last claim as follows.

1
This space, usually denoted by l ∞, is important in analysis.

K12146_C009.indd 213 2/12/15 4:45 PM

 214   ◾    Urysohn, Tietze, and Stone–Čech

Suppose otherwise: this means that there is a sequence s1 , s2 ,… of elements in 

such that e(s1 ), e(s2 ),…, converges to some element in (e( ))′ \ e( ).
The sequence s1 , s2 ,… could not stabilize at some fixed number, since that would
force e(s1 ), e(s2 ),… to converge to an element in e(). Since a sequence made of finitely
many elements could not converge in a Hausdorff space unless it stabilizes, the set of
terms of the sequence s1 , s2 ,… is infinite. We can delete the terms that repeat to get
a subsequence m1 , m2 ,… that still converges to an element in (e( ))′ \ e( ). Denote
M = {m1 , m2 ,…} .
By looking at the coordinates of e(m1 ), e(m2 ),… we see that the convergence of
that sequence implies that, for every bounded function f :  →  , the sequence
f (m1 ), f (m2 ),… converges. In other words, the subsequence ( f (m j ))∞j =1 of every
bounded sequence ( f (n))n∞= 0 converges to some a f . We now show that this is not
possible.
Note that M must be a proper subset of , since there are bounded but divergent
sequences ( f (n))n∞= 0 . Consider the bounded function h : M →  by h(m j ) = (−1) j .
Then (h(m j )) is the infinite sequence −1,1, −1,1,…, which obviously diverges. Now
look at the sequence (h(n))n∞= 0 that agrees with (h(m))m∈M over M and such that
h(r ) = 2 for every r ∈  \ M . It is bounded, and its restriction to M is not conver-
gent. We have a contradiction. ☐

Exercises
Note: In the exercises here the spaces X, X1 , X 2 are completely regular T1 -spaces.
1. Show that if X is connected, then so is β(X).
2. Show that if X1 ≅ X 2 then β( X1 ) ≅ β( X 2 ).

3. Show that if f : X → [0,1] is a continuous function, then there is a continuous

extension F : β( X ) → [0,1].
4. Show that for every completely regular space X and every n ∈+ there is a compacti-
fication β∗ ( X ) and an embedding e ∗ : X → β∗ ( X ), such that β∗ ( X ) is Hausdorff and
such that every bounded continuous function f : X → n can be extended to a con-
tinuous function F : β∗ ( X ) → n (in the sense that F (e ∗ ( x )) = f ( x ) for every x ∈ X ).

5. Let A and B be subsets of a space X, and let f : X → [0,1] be a continuous function

such that f ( A) = {0} and f ( B) = {1}. Show that the closures A and B in β(X) are
disjoint.
6. Show that β( ) is uncountable.
7. Show that β(  ) is not metrizable.

K12146_C009.indd 214 2/12/15 4:45 PM

 2
Homotopy

215

K12146_Book.indb 215 2/9/15 3:08 PM

This page intentionally left blank
 Chapter 10

Isotopy and Homotopy

A knot in  3 is the image of S1 under an embedding f : S1 →  3. Consider the trefoil knot

shown (twice) in Illustration 10.1: it is depicted diagrammatically to the left and tubularly
to the right (where we gain the 3-dimensionallity of the view at the expense of thickening the
image of S1). For a while (until we give the precise definition further in this section) we will take
that an unknot is the image of S1 under the embedding f ( x , y ) = ( x , y , 0). Is the trefoil knot
deformable to the unknot?

Illustration 10.1 Two ways to depict the trefoil knot.

Under the intuitive meaning of the word deformable, the most natural answers seem to
be the following two:

(a) It is clear that the trefoil knot can be deformed to the unknot, since there is nothing
to deform; it is homeomorphic to the unknot anyway.
(b) It is clear that the trefoil knot cannot be deformed to the unknot, since any such
deformation would produce self-intersections changing its structure.

Our initial goal in this chapter is to clear this glaring ambiguity and, in the process, to
introduce some important ideas in our transition from pure topology to algebraic topology.
The main goal in the first chapter is the introduction of the concept of the fundamental
group of a topological space, the principal object of study in the subsequent chapters.
217

K12146_Book.indb 217 2/9/15 3:08 PM

 218   ◾    Isotopy and Homotopy

10.1 Isotopy and Ambient Isotopy

A brief historical note: The isotopy-related concepts arose with the development of knot theory.
Braid theory was introduced by Emil Artin (1898–1962) in 1926.

In this section we introduce the notion of ambient isotopy within a larger space. Braids
(Example 5) are given as a thematic context; they will rarely be seen in the reminder of the
book.
We remind the reader of the convention announced in Section 6.3: the letter I is reserved
for the unit interval [0,1].
Unless otherwise stated, the letters X, Y, and Z will denote topological spaces. Given a
continuous mapping H : X × I → Y , we will denote by Ht , t ∈ I , the mapping Ht : X → Y
defined by Ht ( x ) = H ( x , t ).
Two embeddings f , g : X → Y are isotopic if there is a continuous mapping H : X × I → Y
such that the mappings Ht : X → Y , t ∈ I , satisfy the following two conditions:
(i) Ht is an embedding for every t ∈ I ,
(ii) H 0 = f and H1 = g .
In this context one may think of I as representing time during which we continuously
change f (at time t = 0 ) to g (at time t = 1 ).
The relation of “being isotopic” is an equivalence relation on the set of all embeddings
X → Y (Exercise 1). The mapping H is sometimes called an isotopy from f to g.

Example 1
Let X = S1 and let Y be the bounded double cone, as shown in Illustration 10.2. We
depict the images of four embeddings of X into Y. Any embeddings associated with Ai
and A j , i , j ∈{1, 2, 3}, and of the same orientation, are isotopic. On the other hand it is
clear that no pair ( Ai , A4 ), i ∈{1, 2, 3}, is associated to isotopic embeddings. For example,
deforming A1 to A4 would require passing through the point joining the two cones,
where we would be violating the condition (i) in the definition of isotopy. ☐

A1 A3

A2

A4

Illustration 10.2 Various embeddings of a circle.

K12146_Book.indb 218 2/9/15 3:08 PM

 10.1 Isotopy and Ambient Isotopy   ◾   219  

Example 2
Consider again the trefoil knot (see Illustration 10.1) and the unknot f (S1 ), where
f ( x , y ) = ( x , y , 0). As in the previous example, we can assume that the trefoil knot is
defined in some natural way by an embedding from g : S1 →  3. For example, we can take
g (cos2πt , sin2πt ) = (sin2πt + 2sin4 πt , cos2πt − 2cos4 πt , − sin6πt ), t ∈I .
Are f and g isotopic? Somewhat surprisingly, the answer is affirmative: it is possible
to deform f to g without disturbing the inner structures of their images. The isotopy is
described (visually) in Illustration 10.3 and in the text that follows.

Illustration 10.3 Bachelor’s unknotting: an isotopy joining the (mapping that defines the) tre-
foil knot and (the mapping that defines) the unknot.

Each of the four objects shown in Illustration 10.3 describes an embedding from the
circle into  3 (with only the first and the last explicitly described earlier). The short
sequence of spaces outlines the steps of the isotopy. The knotted part shrinks and
becomes a point “at the last moment.” ☐

In the preceding example the actual formulas for the embeddings f and g did not matter.
In such cases, and when the clarity of the argument is not adversely affected, we will be
satisfied with visualizations.

Example 2 indicates that in order to match the expectation of non-deformability of the

trefoil knot into the unknot, it is necessary to strengthen the notion of isotopy.
Let f , g : X → Y be two embeddings. We say that f is ambient isotopic to g within
Y if there is a continuous mapping H : Y × I → Y such that the mappings Ht : Y → Y ,
t ∈ I , satisfy the following two conditions:

(i) Ht is a homeomorphism for every t ∈ I ,

(ii) H 0 = idY and H1  f = g .

In this case we say that H is the ambient isotopy from f to g.

We can now define an unknot in  3 precisely: it is the image of any embedding e : S1 →  3,
such that e is ambient isotopic to the mapping f : S1 →  3 defined by f ( x , y ) = ( x , y , 0).
The difference between isotopy and ambient isotopy is that in the former case we
change the image of X within Y, transforming   f to g in the process, while in the latter case
we maneuver the points of the whole space Y without affecting Y up to homeomorphism,
achieving the same effect of changing f to g.

K12146_Book.indb 219 2/9/15 3:08 PM

 220   ◾    Isotopy and Homotopy

If H : Y × I → Y is an ambient isotopy from f : X → Y to g : X → Y , then the mapping

K : X × I → Y defined by K ( x , t ) = Ht ( f ( x )) is an isotopy from f to g (Exercise 3). So, the
notion of ambient isotopy is indeed a strengthening of the notion of isotopy.

Example 3
Recall that the simple embedding f indicated by A1 in Illustration 10.2 is isotopic to
the embedding g corresponding to A2. However, these two are not ambient isotopic
within the double cone Y, since otherwise, if H is such an ambient isotopy, then the
homeomorphism H1 would send the points in f (S1 ) = A1 to the points in A2, which
is not possible. The details are left for Exercise 1. ☐

In the examples so far we have mainly focused our attention on the images of embeddings,
rather than on the embeddings themselves. In order to see the difference between the two
concepts more clearly we need the following definition.
Two subsets A and B of a space Y are ambient isotopic within Y if there is a continuous
mapping H : Y × I → Y such that the mappings Ht : Y → Y , t ∈ I , satisfy the following two
conditions:

(i) Ht is a homeomorphism for every t ∈ I ,

(ii) H 0 = idY and H1 ( A) = B.

The mapping H in this definition is called an ambient isotopy from A to B within Y, or an

ambient isotopy between A and B within Y. The latter terminology suggests that the rela-
tion of ambient isotopy is an equivalence relation on the set of all subsets of Y. This claim is
true if the underlying set Y is n (Exercise 6). Since the ambient isotopies we will consider
in this book will always be within n (most often within  3), our usage of ambient isotopy
as an equivalence relation of sets is justified. I do not know under which general conditions
on Y the ambient isotopy of subsets of Y is an equivalence relation.
The specification “within Y” will be omitted when it is clear from the context.

Proposition 1. If the mappings f , g : X → Y are ambient isotopic, then so are the sub-
sets f (X) and g(X) of Y.

Proof. If H is an ambient isotopy joining f and g, then the same H is an ambient isotopy
joining f (X) and g (X).

Consequently, if f(X) and g(X) are not ambient isotopic, then neither are f and g. However,
the converse of Proposition 1 is false (Exercise 10).

Example 4
Is the trefoil knot ambient isotopic to the unknot? We will see in Chapter 13 that the
associated subsets of  3 are not ambient isotopic within  3 (but the main reason for

K12146_Book.indb 220 2/9/15 3:08 PM

 10.1 Isotopy and Ambient Isotopy   ◾   221  

this will be indicated after Proposition 2 below); it follows from Proposition 1 that
the associated embeddings (as defined precisely in Example 2) are also not ambient
isotopic within  3. If the mappings defining the trefoil and the unknot are composed
with any embedding  3 →  4, then it makes sense to ask the question if these two, or
the associated subsets of  4, are ambient isotopic within  4 . The answer to that ques-
tion is affirmative, and a visual justification is provided in Illustrations 10.4 and 10.5.

Illustration 10.4 We change the crossing pointed by the arrow; the procedure changing the
knot to the left to the one to the right is described in Illustration 10.5.

The knot depicted on the right-hand side of Illustration 10.4 can be changed via an
ambient isotopy within  3 (and so, within  4 ) into the unknot.

R4 R4 R4
2 2 2
1 1 1
R3 R3 R3

Illustration 10.5 Focusing on the neighborhood of the crossing indicated in Illustration 10.4,
we first lift the interior of the line segment labeled 2 in  4 and out of  3 (middle picture), then
drop it back into  3, but on the other side of the segment labeled by 1.

Similar argument shows that any two knots are ambient isotopic within  4 and,
therefore, ambient isotopic within n for every n ≥ 4. ☐

So,  3 is the only Euclidean space where ambient isotopy of knots is a meaningful notion.
The problem of classifying the isotopy classes of knots is difficult. Algorithms recognizing
distinct isotopy classes of knots do exist.1 When such an algorithm is inputted with, say, the
knot shown in Illustration 10.6, then the eventual output would be: “It is an unknot!”
There is no intrinsic topological difference between two knots K1 and K 2, be they ambient
isotopic within  3 or not: as we have noted several times, K1 ≅ K 2 ≅ S1 . What makes them
not ambient isotopic is their position within  3 , or more specifically, their complements
 3 \ K1 and  3 \ K 2 : when K1 and K 2 are not ambient isotopic, their complements in  3 are
not homeomorphic. The last statement is not easy to prove.2 What is easy to prove is that
ambient isotopy of subsets and ambient isotopy of their complements in the underlying
space are equivalent notions.

1
See, for example, Corollary 6.1.4 in [48]
2
See [25]

K12146_Book.indb 221 2/9/15 3:08 PM

 222   ◾    Isotopy and Homotopy

Illustration 10.6 Gordon knot (ignore the “irises of the eyes”); it is ambient isotopic within
 3 to the unknot!

Proposition 2. Two subsets A and B of a space Y are ambient isotopic if and only if their
complements Ac and Bc in Y are ambient isotopic.

Proof. Let H be an ambient isotopy joining A and B. Then the same H is an ambient
isotopy joining Ac and Bc . The converse follows by symmetry.

Corollary 3. Two knots K1 and K 2 in  3 are ambient isotopic if and only if their comple-
ments  3 \ K1 and  3 \ K 2 are ambient isotopic.

Further refinement of the definition of ambient isotopy of sets is given in the next definition.

Two subsets A and B of a space Y are ambient isotopic (within Y) relative to a set
C ⊂ A ∩ B if there is a continuous mapping H : Y × I → Y such that the mappings
Ht : Y → Y , t ∈ I , satisfy the following conditions:

(i) Ht is a homeomorphism for every t ∈ I ,

(ii) H 0 = idY and H1 ( A) = B,
(iii) Ht (c ) = c for every c ∈C and every t ∈ I .

In this case H is called an ambient isotopy relative to C joining A and B, and the sets A
and B are ambient isotopic relative to C. If C = ∅, then ‘ambient isotopy relative to C’ is
the same as ‘ambient isotopy’.
The intuitive meaning of this definition is obvious: A and B are ambient isotopic relative
to C if we can homeomorphically deform the whole space Y without moving the points of
C in such a way that the points of A are shifted onto the set B.

K12146_Book.indb 222 2/9/15 3:08 PM

 10.1 Isotopy and Ambient Isotopy   ◾   223  

In the following longer example we introduce and briefly investigate one of many links
between topological spaces and groups. It will not be needed later, so it may be skipped.

Example 5: Braids
Informally, a braid on n strings is a set of n strings in  3, connecting n points with a
disjoint sets of n points, such that no single string is knotted.
First we give a formal definition of a braid on 3 strings. Let p : {1, 2, 3} → {1, 2, 3}
be a permutation. For every j = 1, 2, 3, let β j : I →  3 be a one-to-one path starting at
the point (j, 0, 0) and ending at (p(j), 0, 1), such that β j ( I ) is ambient isotopic, within
 3 and relative to {(j, 0, 0), (p(j), 0, 1)}, to the line segment joining (j, 0, 0) and ending
at (p(j), 0, 1). In addition we require that β j ( I ) ∩β k ( I ) = ∅ for every j ≠ k , and that
p3 β j is monotonic, where p3 :  3 →  is the projection onto the third coordinate,
3
j = 1, 2, 3. Then ∪β j ( I ) is a braid on 3 strings.
i =1

The stipulation that β j ( I ) be ambient isotopic, relative to {(j, 0, 0), (p(j), 0, 1)}, to the
line segment joining (j, 0, 0) and (p(j), 0, 1), together with the assumption that p3 β j
is monotonic, prevents each of the strings from being self-knotted, so that the differ-
ence of these spaces is solely a consequence of the manner the strings are mutually
entangled (or braided) within it. Omitting the monotonicity of p3 β j allows knotting
of a particular string (Exercise 14).
In Illustration 10.7 we show (twice) a braid on three strings, where the permutation
p is the identity permutation. The planar diagrams as the one to the right, even though
they lack the three-dimensionality of the diagram to the left, are often a clearer (and
relatively standard) way of depicting braids.

(3,0,1)
(2,0,1) (1,0,1) (2,0,1) (3,0,1)
(1,0,1)

(2,0,0) (3,0,0)
(1,0,0)
(1,0,0) (2,0,0) (3,0,0)

Illustration 10.7 One braid on three strings, depicted in  3 (left) and through a planar diagram (right).
n
A braid on n strings is ∪β j ( I ), where the conditions on β j ’s generalize in a straight-
i =1
forward way of the ones given in the definition of a braid on 3 strings.
n
Braids can be multiplied in a rather natural way: informally, if x = ∪ β j ( I ) and
n i =1
y = ∪ γ j ( I ) are two braids, then their product xy is the braid obtained by placing
i =1
n n n
∪ γ j ( I ) on top of ∪ β j ( I ). Formally, the product of these two is ∪ δ j ( I ), where
i =1 i =1 i =1

K12146_Book.indb 223 2/9/15 3:08 PM

 224   ◾    Isotopy and Homotopy

 1
h  β j (2t ) for 0 ≤ t ≤ 2

δ j (t ) =  and h( x , y , z ) = ( x , y ,0.5z ).
h  γ  1 1
β j (1) (2t − 1) +  0,0,  for ≤ t ≤1
  2 2

In Illustration 10.9 we show the product of the braid depicted in Illustration 10.7 with
the one shown in Illustration 10.8. We plainly see that after shrinking the braids along
the z-axis by a factor of 0.5—executed by the mapping h in the formal definition given
above—we simply put the second braid on top of the first.

(1,0,1) (2,0,1) (3,0,1) (1,0,1) (2,0,1) (3,0,1)

(2,0,0.5) (3,0,0.5)
(1,0,0.5)

(1,0,0) (2,0,0) (3,0,0)

(1,0,0) (2,0,0) (3,0,0)

Illustration 10.8 One more braid on 3 strings. Illustration 10.9 Multiplication of braids.

The simplest braid is the trivial braid depicted in Illustration 10.10. We denote it by ε.

(1,0,1) (2,0,1) (3,0,1) (n,0,1)

(1,0,0) (2,0,0) (3,0,0) (n,0,0)

Illustration 10.10 The trivial braid.

If x is a braid on n strings, then x ε = x if and only if x = ε. We do not want such rigid-

ity, for it is clear that the structure of any braid β is not affected by attaching the trivial
braid on top of it (xε) or at the bottom of it (εx).
This shortcoming can easily be rectified by replacing the strings in a braid x with their
isotopy classes within  3 relative to their endpoints. This defines an equivalence relation
on the set of braids on n strings. If we denote the equivalence class of the braid x by [x],
then it is fairly straightforward to show that for every two braids x and y, [x ][ y] = [xy]
is a well-defined operation. It turns out that with this operation this set of equivalence

K12146_Book.indb 224 2/9/15 3:08 PM

 10.1 Isotopy and Ambient Isotopy   ◾   225  

classes becomes a group. We have arrived at the concept of the braid group Bn on n
strings, an object of extensive research since its introduction in 1925 by Emil Artin.
We will not study the group Bn here. We only mention that Bn possesses a natural
and simple set of generators, each of them with precisely two strings crossing each
other (as is any one of the two braids shown in Illustration 10.11).
(1,0,1) (2,0,1) (3,0,1) (1,0,1) (2,0,1) (3,0,1)

(1,0,0) (2,0,0) (3,0,0) (1,0,0) (2,0,0) (3,0,0)

Illustration 10.11 Two mutually inverse braids.

The two braids shown in Illustration 10.11 are mutually inverse in the group. This special
case indicates how to construct an inverse of an element in Bn (Exercise 15). ☐

This correlation between (equivalence classes of) braids (hitherto purely topological
objects) and groups allows many topological problems concerning braids to be translated
into purely group-theoretical language. Starting with the following section we will build
another intriguing and productive link between topological spaces and groups.

Exercises
1. Consider Illustration 10.2. Explain why any embedding of the circle S1 onto the
space A1 is not ambient isotopic within the double cone to any embedding of S1 onto
the space A2 . Prove the same for the spaces A2 and A4.
2. Show that the relation of isotopy of embeddings X → Y is an equivalence relation on
the set of all embeddings X → Y .
3. Show that every ambient isotopy joining the embeddings f and g from X into Y is an
isotopy between the same two embeddings.
4. (a) Show that every two mappings f , g : {a} → X are isotopic if and only if X is path
connected.
(b) Show that every two embeddings f , g : S → X are isotopic if and only if X is
S-connected. (Note: “S-connected” was defined in Section 6.4.)
5. Let Y be a Hausdorff, compact space. Show that the relation of ambient isotopy of
embeddings X → Y is an equivalence relation on the set of all embeddings X → Y .
6. Show that the relation of ambient isotopy of subsets A and B of n relative to a set
C ⊂ A ∩ B is an equivalence relation on the class of subsets of n containing C as a
subset.

K12146_Book.indb 225 2/9/15 3:08 PM

 226   ◾    Isotopy and Homotopy

7. Determine which of the knots shown in the illustration below are ambient iso-
topic in  3 to the unknot (circle), which are ambient isotopic to the trefoil knot
shown in Illustration 10.1. Describe the ambient isotopy in all cases.

Illustration 10.12

8. Show that the identity map from the double cone DC (shown in Illustration 10.2) is
not ambient isotopic within DC to the homeomorphism that reflects the cones with
respect to the common point.
9. Show that for any subspace S of a discrete space X the ambient isotopy class of S
within X is {S}.
10. Show that the converse of Proposition 1 is false: find two embeddings f and g of the
circle S1 into a space X such that f (S1 ) and g (S1 ) are ambient isotopic within X, but f
and g are not ambient isotopic within X.
11. Consider the spaces X and Y shown on the left and right hand side (respectively) of
the arrow in Illustration 10.13.

Illustration 10.13

(a) Visualize an ambient isotopy within  3 joining X and Y.

(b) Visualize an ambient isotopy within  2 joining X and Y.
12. An n­-times twisted band Tn is obtained from the square S = {( x , y , 0): 0 ≤ x ≤ 1,
0 ≤ y ≤1} by rotating the edge {(1, y , 0): 0 ≤ y ≤ 1} nπ radians around the y-axis (and
twisting S in the process), then identifying it with {(0, y , 0) : 0 ≤ y ≤ 1} via a trans-
lation. For example, T0 is a simple cylindrical surface, and T1 is a Möbius band.

K12146_Book.indb 226 2/9/15 3:09 PM

 10.1 Isotopy and Ambient Isotopy   ◾   227  

A 3-times twisted band is shown in Illustration 10.14 (the two vertical edges of S
are identified). Assuming S is embedded into a space X (any space X!), show that
there are at least 2 classes of ambient isotopy classes of twisted bands within X.

Illustration 10.14

13. Visualize an ambient isotopy relative to the endpoints between the product (in any
order) of the braids shown in Illustration 10.11 and the trivial braid.
14. Visualize an ambient isotopy within  3 and relative to {A,B} between the string on
the left and the string on the right (see Illustration 10.15).

B
B

A A

Illustration 10.15

15. Find the inverse of the equivalence class of the braid shown in Illustration 10.7. (Hint: you
need to successively multiply that braid by a sequence of generating braids, slowly untangling
the strings after each step; one possible first step is shown in Illustrations 10.8 and 10.9.)
16. Show that B2 is an infinite cyclic group.

K12146_Book.indb 227 2/9/15 3:09 PM

 228   ◾    Isotopy and Homotopy

10.2 Homotopy

A brief historical note: The definition of homotopy of mappings given here is due to Jan
Brouwer, 1911, but the term (with a somewhat different meaning) was introduced for the first
time in 1907 in a joint paper by Max Dehn (1878–1952) and Poul Heegaard (1871–1948).

As the title of this book suggests, the following definition is fundamental.

Let f , g : X → Y be two continuous mappings agreeing over a subset A of X. We say that f

and g are homotopic relative to A if there is a continuous mapping H : X × I → Y such that
H ( x , 0) = f ( x ) and H ( x , 1) = g ( x ) for every x ∈ X , and such that H (a, t ) = f (a) = g (a) for
every a ∈ A and for every t ∈ I . The mapping H is then a homotopy from f to g relative
to A. If A = ∅ then we say that f and g are homotopic.
If f : X → Y is homotopic to a constant mapping (sending all elements of X to a fixed
element in Y), then we say that f is nullhomotopic.
We write f  g when f is homotopic to g, and f  A g when f is homotopic to g relative
to A.
The visual meaning of the definition is evident: f  A g if we can continuously change f to
g keeping their action on the set A unchanged in the process. The difference between homo-
topy and isotopy is that in the former case the mappings f, g, and H are continuous, while
in the latter case f and g, as well as the mappings Ht as defined in 10.1, are all embeddings.
Consequently, if two embeddings are isotopic, then they are homotopic too.

Example 1
We revisit the embeddings discussed in Example 1 and Illustration 10.2, Section 10.1. It is
easy to see that each pair of such mappings is a pair of homotopic mappings. For example,
the embedding f associated with the circle f (S1 ) is homotopic to the embedding g associ-
ated with g (S1 ) via a homotopy shown in Illustration 10.16, where we see four intermedi-
ate steps Hti ( x ) = H ( x , ti ), i = 1, 2, 3, 4 , of the homotopy. In this case nothing prevents us
from shrinking the mapping (and the circle associated with it) to a constant (to a point,
respectively)—something that cannot be done in the case of isotopy. The embedding h
associated with h(S1 ) (as shown in Illustration 10.16) is homotopic to g as well (Exercise 1).
If we change the double cone Y to the cylinder C shown in Illustration 10.17, then
the story changes profoundly. The embeddings f , g : S1 → C , as shown in Illustration
10.17, are not homotopic; neither are f and f ′, where the latter is reversing the orienta-
tion of f (that is, f ′( x , y ) = f ( x , − y )). Neither of these two claims can be easily proven
at this stage. The theory that we are developing starting with this section will eventu-
ally yield very simple justifications.

The first step in the development of our theory is the following analogue of the similar
statements concerning isotopy.

K12146_Book.indb 228 2/9/15 3:09 PM

 10.2 Homotopy   ◾   229  

f´(S1)
f (S1) = H0 (S1)

h(S1)

Ht1(S1)

Ht2(S1) f(S1)

Ht3(S1)
g(S1) = H1(S1) g(S1)
Ht4(S1)

Illustration 10.16 Homotopic embeddings of a Illustration 10.17 Homotopic embeddings

circle into a double cone. of a circle into a cylinder. ☐

Proposition 1. The relation  A of “being homotopic relative to A” is an equivalence relation

on the set of continuous mappings X → Y .

Proof. For every continuous f : X → Y , the mapping F ( x , t ) = f ( x ), t ∈ I , yields

f  A f . Further, if F is a homotopy from f to g relative to A, then G( x , t ) = F ( x , 1 − t )
furnishes homotopy from g to f relative to A. Finally, if f  A g via a homotopy F, and
 F ( x ,2t ) for 0 ≤ t ≤ 0.5
g  A h via a homotopy G, then H ( x , t ) =  makes f  A h ,
 G( x ,2t − 1) for 0.5 ≤ t ≤ 1
where the continuity of H is a consequence of the continuity of F and G and of the
gluing lemma.

A continuous mapping f : X → Y is a homotopy equivalence relative to A, if there is a

continuous mapping g : Y → X such that g  f A id X and f  g  f ( A ) idY , where for every
space Z, idZ is the identity mapping Z → Z. If in the above setup A = ∅ , then we say that
f : X → Y is a homotopy equivalence.
Homeomorphisms have inverses; as we see from the last definition, homotopy equiva-
lences have inverses, up to homotopy.
Two spaces X and Y are homotopically equivalent, if there is a homotopy equivalence
f : X → Y . We will occasionally write X  Y for such spaces.
As the name suggests, this is indeed an equivalence relation on the family of all spaces. The
equivalence class containing a space X will be occasionally called the homotopy class of X.

Example 2
The inclusion in : {( x , y ,0): x 2 + y 2 = 1} → {( x , y , z ): x 2 + y 2 = 1, 0 ≤ z ≤ 1} establishes a
homotopy equivalence between the circle S1 = {( x , y ,0): x 2 + y 2 = 1} and the cylinder
C = {( x , y , z ): x 2 + y 2 = 1, 0 ≤ z ≤ 1}; the inverse of the inclusion, up to homotopy, is
the projection p : ( x , y , z ) ↔ ( x , y , 0). This is easy to justify: p  in = idS1 , and the

K12146_Book.indb 229 2/9/15 3:09 PM

 230   ◾    Isotopy and Homotopy

mapping F : C × I → I defined by F (( x , y , z ), t ) = ( x , y ,(1 − t )z ) is a homotopy from

idC to in ° p. The mapping F is visualized in Illustration 10.18. ☐

Illustration 10.18

Example 3
Here, we make the following point: at this stage of our theory it is not easy to exhibit
a single pair of spaces which are not homotopically equivalent, and fully justify that
they are indeed not homotopically equivalent. For example, consider S1 and a single-
ton {x}. It is intuitively clear that these are not homotopically equivalent. A heuristic
argument goes as follows: otherwise there are continuous mappings f : S1 → {x } and
g : {x } → S1 such that g  f  idS1 via some homotopy H; since g  f maps S1 onto
some y ∈S1, this means that H is a continuous transformation from idS1 to the con-
stant map S1 → { y}. Thinking of the identity mapping over S1 as an elastic circular
string wrapped around the circle, we continuously deform the string always staying
within the circle. Obviously, to get to the constant map at y, either we need to cut the
string at one moment, or we must momentarily collapse it to a point (in both cases H
is discontinuous); the hole of the circle cannot be bridged otherwise.
Our theory will detect such holes easily; once it matures, the justification that S1 and
{x} are not homotopically equivalent will then be so obvious that we will barely stop
to note it. Through this example we merely express our anticipation of the theory that
will match our expectations.

{y} × {1}

S1 × {t1} Ht1 Illustration 10.19 One of two reasons

S1 is not homotopically equivalent
S1 × {0} to a point: at some moment t1 the
elastic string must be cut in order to
     shrink it to a point y on the circle. ☐

K12146_Book.indb 230 2/9/15 3:09 PM

 10.2 Homotopy   ◾   231  

A subspace A of a space X is a retract of X if there is a continuous mapping r : X → A

such that r  in = id A, where in : A → X is the inclusion mapping. The mapping r is then
called a retraction. If, in addition, r satisfies in  r  id X , then we say that A is a deformation
retract of X, and in that case the mapping r is a deformation retraction.

Example 4
For every space X and every point x ∈ X , {x } is a retract of X, since the constant
mapping X → {x } is a retraction. However, it follows from our (still not fully justi-
fied) claims in Example 3, that if X = S1, then no point of X is a deformation retract
of X. The circle at a base of a cylinder is a deformation retract of the cylinder; a
justification was provided in Example 2. ☐

It follows immediately from the definitions that if A is a deformation retract of X, then A

and X are in the same homotopy class. The converse is not true (Exercise 8).

Exercises
1. (a) Denote the double cone in Illustration 10.16 by Y. Visualize a homotopy between
an embedding h : S1 → Y associated with h(S1 ) and an embedding g : S1 → Y
associated with g (S1 ) (as in Illustration 10.16).
(b) Visualize a homotopy between an embedding f : S1 → Y associated with the
circle f (S1 ) (as in Illustration 10.16), and the embedding k : S1 → Y defined
by k( x , y ) = f ( x , − y ) ( f and k wind in opposite directions).
2. Show that {0} is a deformation retract of .
3. Show that the line y = 0 considered as the subspace of  2up with the tangent-disk
topology is not a retract of  2up. (Recall that  2up = {( x , y ) : y ≥ 0}.)
4. (a) Let X and Y     be two homotopically equivalent infinite spaces. Is it true that for every
point x ∈ X , there is a point y ∈Y such that X \ { x } and Y \ { y } are homotopically
equivalent?
(b) Let X and Y be two homotopically equivalent infinite spaces and let x ∈X. Is it true
that there exists a subset B of Y such that {x }  B and such that X \ {x }  Y \ B ?
5. Show that {x} and the interval (0, 1) are homotopically equivalent.
6. Show that if A is a deformation retract of X and if X is a deformation retract of Y, then
A is a deformation retract of Y.
7. Show that if A is a deformation retract of X and if there are mappings f : X → Y and
g : Y → X such that g  f A id A and f  g  idY , then f(A) is a deformation retract of Y.

K12146_Book.indb 231 2/9/15 3:09 PM

 232   ◾    Isotopy and Homotopy

8. Find a space X and a subspace A of X such that A and X are homotopically equiv-
alent, but A is not a deformation retract of X. (You may use the claims stated in
Example 1)
9. Consider the bouquet X = (S1 × { y})∪ ({x } × S1 ) of two circles as a subspace of the
punctured torus Y = T 2 \ {P}, where P is a point on the torus T 2 = S1 × S1 not con-
tained in (S1 × { y})∪ ({x } × S1 ) . Visualize a deformation retraction r : Y → X .
10. (a) Denote the z-axis of  3 by Z, and denote by W the subspace {(x, y, z) ∈ 3 :
x 2 + y 2 < 1} of  3. Show that  3 \ W is a deformation retract of  3 \ Z .
(b) Let K be a knot in  3 defined by a smooth embedding k : S1 →  3. For every
x ∈K denote by B( x , rx ) the open disk centered at x, of radius rx , and in the
plane perpendicular to the tangent at x to the curve k. Suppose K and these open
disks satisfy the following two properties:
(i) for every x ∈K , B( x , rx ) ∩ K = {x }.
(ii) For every distinct x , y ∈ K , B( x , rx ) ∩ B( y , ry ) = ∅.
Let T be the tunnel ∪ B( x , rx ) around K. Show that  3 \ T is a deformation
x ∈K
retract of  3 \ K .
11. Show that if f : S n → S n has no fixed points, then f is homotopic to g : S n → S n
defined by g ( z ) = − z . [Hint: find an explicitly formula for a homotopy.] (Recall that
all the mappings are assumed to be continuous, unless otherwise stated.)

10.3 Homotopy and Paths

A brief historical note: The multiplication of homotopy classes of compatible paths (as in
Corollary 2 here) was introduced by Henri Poincaré in 1895.

Recall that a path in a space X is a continuous mapping I → X . The basic terminology

and the multiplication of compatible paths have already been introduced (Section 6.3).
We will show in this section that this algebraic operation induces multiplication of
homotopy classes of paths (Corollary 2), eventually yielding a richer algebraic structure.
The homotopies of paths that we will encounter from now on will almost exclusively
be relative to {0,1}. Because of this, and in order to make the notation less burdened with
symbols, from this point on we accept the following:
For paths α , β : I → X , α  β is short for α  {0,1} β.
On rare occasions, when we violate this convention, we will explicitly announce it.

Proposition 1. For every paths α , α ′ , β, and β′ such that α(1) = α ′(1) = β(0) = β′(0), if
α  α ′ and β  β′, then αβ  α ′β′.

K12146_Book.indb 232 2/9/15 3:09 PM

 10.3 Homotopy and Paths   ◾   233  

Proof. Let F be a homotopy relative to {0,1} from α to α′ and let G be a homotopy relative to
{0,1} from β to β′. The following formula for a homotopy H relative to {0,1} from αβ to α ′β′
can easily be decoded from the diagram in Illustration 10.20:

 F (2t , s ) if 0 ≤ t ≤ 0.5, 0 ≤ s ≤ 1
H (t , s ) = 
G(2t − 1, s ) if 0.5 ≤ t ≤ 1, 0 ≤ s ≤ 1


α´ (0.5,1) β´
(0,1)

F G

t
α (0.5,0) β (1,0)

Illustration 10.20 A homotopy from αβ to α′β′.

H is continuous since F and G are, and since F (1, s ) = α(1) = β(0) = G(0, s ); that is, F and G
agree over the intersection of their domains in the definition of H. Next,

 F (2t ,0) if 0 ≤ t ≤ 0.5  α(2t ) if 0 ≤ t ≤ 0.5

H (t ,0) =  = = αβ , and
 G(2t − 1,0) if 0.5 ≤ t ≤ 1  β(2t − 1) if 0.5 ≤ t ≤ 1

 F (2t ,1) if 0 ≤ t ≤ 0.5  α ′(2t ) if 0 ≤ t ≤ 0.5

H (t ,1) =  = = α ′β′.
G(2t − 1,1) if 0.5 ≤ t ≤ 1 β′(2t − 1) if 0.5 ≤ t ≤ 1
 

Finally we show that the homotopy H is relative to {0, 1} : H (0, s ) = F (0, s ) = α(0) for every
s ∈ I (since F is relative to {0}), and H (1, s ) = F (1, s ) = β(1) for every s ∈ I (since G is relative
to {1}).

We will consistently denote the homotopy class (relative to {0,1}) of a path α by [α].
Proposition 1 aimed at the following corollary.

Corollary 2. Let α and β be two paths in X such that α(1) = β(0). Then [α][β] = [αβ] is a
well-defined operation.

K12146_Book.indb 233 2/9/15 3:09 PM

 234   ◾    Isotopy and Homotopy

With [α][β] = [αβ] we have defined a partial operation over the set of all homotopy classes
of paths. For paths not satisfying that condition α(1) = β(0), the operation [α][β] = [αβ] has
no meaning since αβ is not well defined. This is one of the shortcomings of the algebraic
structure we have so far. Our goal in this and in the next section is to extract better behaving
substructures from it.
The multiplication of compatible homotopy classes satisfies the associative law:

Proposition 3. Let α, β, and γ be paths in X such that α(1) = β(0) and β(1) = γ (0). Then
([α][β])[γ ] = [α]([β][γ ]).

Proof. ([α][β][γ ] = [α][β][γ ]) means (αβ)γ  α(βγ ), where, by our earlier convention, the
homotopy is relative to {0,1}. Such a homotopy can again be extracted from the diagram in
Illustration 10.21; all we need is the equation of the lines l1 and l2 in the (t,s)-coordinate
system. The line l1 is defined by t = s+41 ; the line l2 by t = s+42 .

s
l1 l2
α (0.5,1) β γ
(0,1)

t
α (0.25,0) β (0.5,0) γ (1,0)

Illustration 10.21 Proving the associative law.

The following is then a homotopy from (αβ)γ to α(βγ):

  4  s +1
 α t for 0≤t ≤
  s + 1  4
 s +1 s+2
for every 0 ≤ s ≤ 1, F (t , s ) =  β(4t − s − 1) for ≤t ≤
 4 4
  4 s + 2 s+2
 γ  2 − s t + s − 2  for 4
≤ t ≤1


The computational details are similar to those done in Proposition 1 and are left as an exercise
(Exercise 7).

K12146_Book.indb 234 2/9/15 3:09 PM

 10.3 Homotopy and Paths   ◾   235  

In view of this proposition we may unambiguously write [α][β][γ], and (by induction)
[α1 ][α 2 ]…[α n ] whenever the multiplication is well defined.
Let δ be a path in a space X starting at x and ending at y. The inverse of the path δ,
denoted δ −1, is a path defined by δ −1 (t ) = δ(1 − t ), t ∈ I . Clearly, δ −1 is a path starting at y and
ending at x. Observe that the innocuous statement “is a path” needs at least a glance, since
with that we claim that δ −1 is a continuous mapping I → X .

Remark. We have encountered again a traditional notation with two distinct meanings:
when δ −1 stands for the inverse of δ, then δ−1 (t ) = δ (1 − t ). On the other hand, according to
an earlier standard notation, for every mapping δ, we have δ−1 (t ) = {x : δ( x ) = t }. In the case
of paths, the former meaning (δ −1 (t ) = δ(1 − t )) should be assumed, unless otherwise stated.
In general, the context should clarify which of the two applies.
We digress to state and prove the following proposition, which we will need later on.

Proposition 4. Let α be a loop at x ∈ X , let β be a loop at y ∈ X , and let F : I × I → X

be a homotopy from α to β such that F (0, t ) = F (1, t ) for all t ∈ I (not necessarily relative
to {0,1}). Then there is a path δ from x to y such that the loops δβδ −1 and α are homotopic
relative to {0,1}.

F1 = β

F⅔

F⅓ y

F0 = α
X

Illustration 10.22

The stipulation that F (0, t ) = F (1, t ) for all t ∈ I means that for every t ∈ I , Ft = F I ×{t } is
a loop (Illustration 10.22).

Proof. Define δ(t ) = F (0, t ). A homotopy G from α to δβδ −1 is described visually in

Illustration 10.23, and algebraically as follows.

 t
 F (0,3s ) if 0 ≤ s ≤
2t + 1

  3t  t t +1
G(s , t ) =  F  (2t + 1)s − t ,  if ≤s≤
 2t + 1  2t + 1 2t + 1

 F (0,3 − 3s ) t + 1
if ≤ s ≤ 1.
 2t + 1

K12146_Book.indb 235 2/9/15 3:09 PM

 236   ◾    Isotopy and Homotopy

F F

t t

Illustration 10.23 The thick lines to the left and right describe the mappings F(s,t), and G(s,t),
respectively; when the second variable is a fixed.

We may assume that each of the three paths in δβδ −1 takes one third of the time (see Exercise 3),
that is, that

 1
 δ(3s) if 0≤s≤
3
 1 2
δβδ −1 (s ) =  β(3s–1) if ≤s≤
 3 3
 δ −1 (3s–2) 2
if ≤ s ≤1.
 3

The justification of the claim that the mapping G performs as advertised is left as an exercise
(Exercise 9).

Exercises
1. Let X be the space over  equipped with the co-finite topology. Show that if α and
β are two paths in X, homotopic relative to {0,1}, then α = β. [Hint: you may need
Exercise 15 of 6.1.]
2. (a) Show that if X is homotopically equivalent to a point, then every two paths
α , β : I → X with common endpoints are homotopic.
(b) Find a formula for a homotopy showing that the double cone depicted in
Illustration 10.16 is homotopically equivalent to a point. Visualize the homotopy.
3. Show that for every three paths α, β, and γ for which the product (α β)γ is well defined,
 1
 α(3s ) if 0 ≤ s ≤
3

 1 2
the paths (αβ)γ and αβγ (s ) =  β(3s − 1) if ≤ s ≤ are homotopic relative to {0,1}.
 3 3
 γ (3s − 2) if 2 ≤ s ≤ 1
 3

4. Consider the loop ( γαγ−1 )(δβδ−1 ) in the space  3 \ (T1 ∪ T2 ), where T1 and T2
are the tori shown in Illustration 10.24, and where α, γ, β, and δ are the paths as
shown in the same illustration. Show by drawing a few frames of an animation that

K12146_Book.indb 236 2/9/15 3:09 PM

 10.3 Homotopy and Paths   ◾   237  

( γαγ−1 )(δβδ−1 ) is homotopic in  3 \ (T1 ∪ T2 ), relative to the interval [0,0.5], to the

loop ( γαγ −1 )((δ1β1δ 1−1 )( γαγ −1 )) (shown in Illustration 10.25).

T1

T2

γ δ

Illustration 10.24 Illustration 10.25

5. Let α be a path in a space X.
(a) Show that there are paths 1r and 1l in X such that [α][1r ] = [α] and [1l ][α] = [α].
(b) Show that there are paths α r−1 and α l−1 in X such that [α][α r−1] = [1l ] and [α l−1][α] = [1r ].
6. We have seen in Section 6.4 that if α i : I → Xi , i = 1, 2, … , n, are paths, then
n
α(t) = (α1 (t ), α 2 (t ), … , α n (t )) , t ∈ I , defines a path in the product space X = ∏ Xi .
i=1
Show that if β(t ) = (β1 (t ), β 2 (t ), … , βn (t )) is a path in X and if α i ~ βi , i = 1, 2, … , n, then
α ~ β.
7. Complete the details of the proof of Proposition 3.
8. Let { Ai : i ∈} be a family of open pairwise disjoint subintervals of I, and let α be a path
in X. Denote α i = α | Ai , and suppose βi is homotopic to α i relative to the endpoints of
Ai . Show that if β is a path obtained by replacing each α i with βi , then α ~{0,1} β.
9. Let α be a loop at x ∈ X , let β be a loop at y ∈ X, and let F : I × I → X be a homotopy
from α to β, not necessarily relative to {0,1}, such that F (0, t ) = F (1, t ) for all t ∈ I .
Show that there is a path δ from x to y such that the loops
 1
 δ(3s ) if 0≤s≤
3
 1 2
δβδ −1 (s ) =  β(3s − 1) if ≤s≤
 3 3
2
 δ −1 (3s − 2) if ≤ s ≤1
 3
and α are homotopic relative to {0,1}. (That is, fill in the details of the proof of
Proposition 4.)
10. Let α be a loop at a point x in a space X, and suppose there is a homotopy (always
relative to {0,1}, unless otherwise stated) from α to the constant loop cx (defined by
cx (t ) = x for every t ∈ I).

K12146_Book.indb 237 2/9/15 3:09 PM

 238   ◾    Isotopy and Homotopy

(a) Define β : I → X by β(t ) = α( t 2 ), t ∈ I , and define γ : I → X by γ (t ) = α −1 ( t 2 ),

t ∈ I . Note that both β and γ start at x and end at α( 1 2 ). Prove that β and γ are homo-
topic (relative to {0,1}) by finding an explicit homotopy between them.
(b) Choose any r ∈ I and define β to be associated to the restriction of α to [0,r] (see
the definition of ‘the path associated to a mapping’ at the beginning of Section
6.3) and γ to be associated to the restriction of α −1 to [0, 1 − r ]. Prove that β and γ
are homotopic (relative to {0,1}) by finding an explicit homotopy between them.

10.4 The Fundamental Group of a Space

A brief historical note: The idea of the fundamental group of a space was introduced in 1895
by Henri Poincaré in his influential paper “Analysis Situs”.

Starting from this section we will assume knowledge of basic group theory. Here is a glos-
sary of some elementary group theoretical terms (and some of the definitions): group, the
identity, the inverse of an element, homomorphism ( f : G → H is a homomorphism if
f ( xy ) = f ( x ) f ( y ) for every x , y ∈G ), cyclic group, subgroup, monomorphism (one-to-
one homomorphism), epimorphism (onto homomorphism), isomorphism (one-to-one
and onto homomorphism; the notation for isomorphic groups is G ≅ H).
The main shortcoming of the algebraic structure over the set of homotopy classes of
paths introduced so far is that the multiplication of homotopy classes of paths is defined
only for compatible homotopy classes of paths. We will now repair that shortcoming by
reducing the associated set of homotopy classes of paths. The first step is simple: instead of
paths, we will consider loops.
A loop at x ∈ X is a path α: I → X such that α(0) = α(1) = x . The notion of multi-
plication of paths carries over to multiplication of loops at a fixed point x ∈ X , and, by
Proposition 1 and Corollary 2 of 10.3, it induces a well-defined operation on homotopy
classes of loops at x. The set of all homotopy classes of loops at x together with the operation
of multiplication of such homotopy classes of loops at x is called the fundamental group
of X at x, or the first homotopy group of X at x, and is standardly denoted by π1 ( X , x ) .
The terminology (the word group) is justified in Theorem 1. The meaning of first will be
discussed later. In this context we say that point x is the base point for the space X.

Theorem 1. π1 ( X , x ) is a group.
Proof. We have already proven that the operation is well defined and associative. We need
to establish the existence of the identity and of inverses.
The role of the identity in π1 ( X , x ) is played by the constant loop c x : I → X defined by
cx (t ) = x for every t ∈ I . (If x is clear from the context, we will write c instead of cx .) A jus-
tification of that claim is essentially in Illustration 10.26, where we indicate why [α][c] = [α]
(that is, αc  α relative to {0,1}) for every loop α.

K12146_Book.indb 238 2/9/15 3:10 PM

 10.4 The Fundamental Group of a Space   ◾   239  

The equation of the line l in Illustration 10.26 is t = s+21 ; from that, and from the diagram
in Illustration 10.26, it is easy to decode an algebraic formula for a homotopy (relative to
{0,1}) F from αc to α:
s
l
α
(0,1)

t
α (0.5,0) c (1,0)

Illustration 10.26 Showing that [c] is the identity in π1 ( X , x ) .

  2  s +1
 α t for 0 ≤ t ≤
  s + 1  2
For every 0 ≤ s ≤ 1, F (t , s ) = 
 x s +1
for ≤ t ≤1
 2
Checking that this satisfies all of the requirements is a straightforward computation.
Similar argument is needed to show that [c][α] = [α].
The proof will be completed once we show the existence of inverses. Not surprisingly, the
inverse [α]−1 of [α] is [α −1 ]. We show that [α][α −1 ] = [c], and it follows that [α −1 ][α] = [c].
s
l2 l1
c c
s=1
X
x

s = s2 X
x

s = s1 X
x

s=0 t α α–1 X
α (0.5,0) α–1 (1,0) x

Illustration 10.27 The four frames to the right indicate the paths corresponding to the restric-
tion of the homotopy to I × {s} for four choices of s. The paths α and α −1 follow the same
trajectory (with reversed orientation); we separate them to make the homotopy clearer.

K12146_Book.indb 239 2/9/15 3:10 PM

 240   ◾    Isotopy and Homotopy

The homotopy from αα −1 to c is visualized in Illustration 10.27. Here is an algebraic

description of the homotopy, decoded from the equations of the lines l1 and l2 shown in
Illustration 10.27:

 s
 x for 0 ≤ t ≤
 2
 s 1
 α ( 2t − s ) for ≤t ≤
2 2
For every 0 ≤ s ≤ 1, F (t , s ) = 
 α −1 ( 2t − 1 + s ) 1 2−s
for ≤t ≤
 2 2
 2−s
 x for ≤ t ≤ 1.
 2

It is again straightforward to check that the homotopy F is continuous (gluing lemma!) and
that it performs as advertised.

The constant loop cx at x, whose homotopy class is the identity of π1 ( X , x ), will sometimes be
called the trivial loop. If α is a loop at x homotopic to cx , then we will say that α is a contract-
ible loop or a nullhomotopic loop.

Example 1
If X is a singleton {x}, then it is apparent that π1 ( X , x ) is the trivial group (consisting
only of the identity [cx ] ). ☐

Proposition 2. If x and y are in the same path component of X, then the groups π1 ( X , x )
and π1 ( X , y ) are isomorphic.

Proof. Let γ be a path from x to y. Then for every loop α at x, the path γ −1αγ is a loop at
y. We show that the mapping f :[α]  [γ −1αγ ] is an isomorphism π1 ( X , x ) → π1 ( X , y )
(Illustration 10.28).

X
α

γ–1 y
x
γ

Illustration 10.28

K12146_Book.indb 240 2/9/15 3:10 PM

 10.4 The Fundamental Group of a Space   ◾   241  

It is easy to show that f is well defined. It is a homomorphism since for every two
loops α and β at x, we have f ([α]) f ([β]) = [γ −1αγ ][γ −1βγ ] = [γ −1α][γγ −1 ][βγ ] = [γ −1αβγ ]
= f ([αβ]) = f ([α][β]), where we have twice used the associativity of multiplication of
paths.
The mapping f is a bijection since g : π1 ( X , y ) → π1 ( X , x ) defined by g ([β]) = [γβγ −1 ] is its
inverse. For example, f ( g ([β])) = f ([γβγ −1 ]) = [γ −1 γβγ −1 γ ]] = [β].

Corollary 3. If X is path connected, then the groups π1 ( X , x ), x ∈ X , are mutually isomorphic.

On the other hand, different path components of a single space X have, in general, unrelated
fundamental groups. Consequently it makes sense to investigate only the fundamental groups
of path connected spaces. Hence, from now on, unless otherwise stated, we will assume that
all spaces we consider are path connected.
Occasionally, we will repeat this, either for emphasis or to make sure the convention has not
been missed. Along these lines, we will at times disregard the base point x and say “the fun-
damental group of a space X,” rather than “the fundamental group of a space X at a point x.”
Next, we will see that continuous mappings of spaces induce homomorphisms of their
fundamental groups. First we define what we mean by “induce.” If f : X → Y is a continu-
ous mapping such that f ( x ) = y , then we write f : ( X , x ) → (Y , y ). Such a mapping induces
a mapping f∗ : π1 ( X , x ) → π1 (Y , y ) defined as follows: for every loop α at x, f∗ ([α]) = [ f  α]
(Illustration 10.29). Note that this is well defined, since if α is a loop at x, then f α is a
loop at y, and since α  β implies f  α  f  β (we leave the proof of the last statement to
the reader). The next proposition shows that f∗ is more than just a mere mapping.

X
α(I) f(α(I))
f
I α Y
x y

Illustration 10.29

Proposition 4. For every continuous f : ( X , x ) → (Y , y ) the induced mapping

f∗ : π1 ( X , x ) → π1 (Y , y ) is a homomorphism.

Proof. For every two loops α, β, at x, and for every continuous mapping f : X → Y , easy
computation shows that f  αβ = ( f  α )( f  β). Then we have: f∗ ([α][β]) = f∗ ([αβ]) =
[ f  αβ] = [( f  α )( f  β)] = [ f  α][ f  β] = f∗ ([α]) f∗ ([β]).
One should be warned that the assumption that the mapping f is one-to-one in general
does not imply that f∗ is one-to one (an example is the inclusion S1 →  2; this will be jus-
tified in the next section). However, many properties of f do carry over to f∗ . A few such

K12146_Book.indb 241 2/9/15 3:10 PM

 242   ◾    Isotopy and Homotopy

properties will be noted in the consequences further below. First, we need to show that the
mapping f  f∗ behaves well.

Proposition 5. For every continuous f : ( X , x ) → (Y , y ) and g : (Y , y ) → ( Z , z ), we have

( g  f )∗ = g ∗  f∗ .

Proof. ( g  f )∗ ([α]) = [( g  f )  α] = [ g ( f  α )] = g ∗ ([ f  α]) = g ∗ ( f∗ ([α])) = g ∗  f∗ ([α]).

The following few results will be used repeatedly.

Theorem 6. Let f : ( X , x ) → (Y , y ) and g : ( X , x ) → (Y , y ) be continuous mappings,

homotopic relative to {x}. Then the induced homomorphisms f∗ and g ∗ are the same.

Proof. Assume F is a homotopy from f to g relative to {x}. We need to show that for every
loop α at x, f∗ ([α]) = g ∗ ([α]). The last equality means f  α  {0,1} g  α. A homotopy mak-
ing this happen is the following: G(t , s ) = F (α(t ), s ). We check: G(t,0) = F(α(t),0) = f(α(t));
G(t,1) = F (α(t ), 1) = g (α(t )); finally, if a ∈{0,1}, then G(a, s ) = F (α(a), s ) = F ( x , s ) = y ,
the last equality being true since F is a homotopy relative to {x}.

g
α (x,1)
X × {1}
Y
α (x,s) F
I X × {s}
y
I
α (x,0) X × {0}
I f

Illustration 10.30

Illustration 10.30 is a visualization of the proof: after identifying I × {s} and X × {s} with I
and X, respectively, and by means of the obvious homeomorphisms, we simply extend the
domain of α to I × I as shown, then follow that extension by F.

Theorem 7. If f : ( X , x ) → (Y , y ) is a homotopy equivalence, then f∗ : π1 ( X , x ) → π1 (Y , y )

is an isomorphism.

Proof. Let f : ( X , x ) → (Y , y ) be a homotopy equivalence. Then there is a continuous map-

ping g : (Y , y ) → ( X , x ′ ) such that g  f  id X and f  g  idY . If x = x ′ and if the homotop-
ies are relative to x, then the result follows from Theorem 6 and Exercise 14. Since this is
not what happens in general, we need to do some work.
Let F be a homotopy from id X to g  f . Then δ(t ) = F ( x , t ) defines a path from
x to x′. By the proof of Proposition 2 the mapping k : π1 ( X , x ′ ) → π1 ( X , x ) defined by
[β]  [δβδ −1 ] is a group isomorphism. Denoting for shortness h = g  f , we claim
that k  h∗ : π1 ( X , x ) → π1 ( X , x ) is the identity isomorphism. We need to show that

K12146_Book.indb 242 2/9/15 3:10 PM

 10.4 The Fundamental Group of a Space   ◾   243  

[δ(h α )δ −1 ] = [α] for every loop α at x, or, equivalently, that δ(h α )δ −1  {0,1} α for every
such loop α. A justification is given in Illustration 10.31 and in the caption.

x
h
I
α
I
γ1–1 F
δ
γ1
x x x´
I id
α X

Illustration 10.31 We map I × I into X × I as shown: at level s , 0 ≤ s ≤ 1, γ 1 sends the thick hori-
zontal line segment on the left side of I × I linearly onto the vertical segment from {x } × {0} to
{x } × {s}; we follow it by a loop that works like α, and then by γ 1−1.

It is easy to see that the composition of the mapping I × I → X × I , as defined in

Illustration 10.31, followed by F gives a homotopy from α to δ(h α )δ −1 , as claimed. (See
also Exercise 3 in 10.3.) We proved that k  h∗ = idπ1 ( X ,x ).
Since k and idπ1 ( X , x ) are isomorphisms, so is h∗ = ( g  f )∗ = g ∗  f∗. Hence f∗ is one-to-
one and g ∗ is onto. By symmetry, f∗ is onto and g ∗ is one-to-one. Hence both f∗ and g ∗
are isomorphisms.

A particular case of Theorem 7 is that homeomorphic path-connected spaces have

isomorphic fundamental groups.

Corollary 8. If A is a deformation retract of X, then A and X have isomorphic fundamental

groups.

Proof. We have observed in 10.3 that if A is a deformation retract of X, then A and X are
homotopic. The result now follows from Theorem 7.

The problem with the theory developed so far is that it does not provide methods for
computing the fundamental groups of spaces, beyond the bare definitions. In the subse-
quent chapters we will establish such methods.
A path connected space is simply connected if its fundamental group is trivial.

Example 2
The unit disk is simply connected, since any point in it is its deformation retract. Because
of the same reason, the double cone seen in Example 1 of 10.2 is also simply connected.
This explains why all of the paths mentioned in that example are mutually homotopic.
In general, if a point in a space is a deformation retract of the space, then that space is
simply connected. ☐

K12146_Book.indb 243 2/9/15 3:10 PM

 244   ◾    Isotopy and Homotopy

Recall that, for G and H groups, G × H is their direct product, with the multiplication
× defined to be component-wise: ( g 1 , h1 ) × ( g 1 , h1 ) = ( g 1 g 2 , h1h2 ). When both G and H are
commutative, then ⊕ is usually used instead of ×, and G ⊕ H is called the direct sum of the
commutative groups G and H.

Proposition 9. For all spaces X and Y, the groups π1 ( X × Y ,( x , y )) and π1 ( X , x ) × π1 (Y , y )

are isomorphic.

Proof. Denote the two projections X × Y → X and X × Y → Y by pX and pY, respectively.

Define f : π1 ( X × Y ,( x , y )) → π1 ( X , x ) × π1 (Y , y ) by f ([α]) = (( pX )∗ ([α]),( pY )∗ ([α])) . Then
f is a homomorphism:
( )
(1)
f ([αβ]) = (( pX )∗ ([αβ]),( pY )∗ ([αβ])) = (( pX )∗ ([α]))(( pX )∗ ([β])) , (( pY )∗ ([α]))(( pY )∗ ([β]))

( ) ( )
= (( pX )∗ ([α])) , (( pY )∗ ([α])) × (( pX )∗ ([β])) , (( pY )∗ ([β])) = f ([α]) × f ([β]),

where in (1) we have used the fact that both ( pX )∗ and ( pY )∗ are homomorphisms
(Proposition 4).
The simplest way to show that f is a bijection is by finding its inverse. We leave it as an
exercise (Exercise 11).

It is true that π1 ( ∏ X j , x ) ≅ × π1 ( X j , p j ( x )) , where pk : ∏ X j → X k is the projection.

j∈ J j∈ J j∈ J
A slight modification of the proof of Proposition 9 suffices as a justification (Exercise 12).

Exercises
[The first two exercises are a diversion; we will be concerned primarily with spaces closely
related to Euclidean spaces.]

1. What is π1 ( X , x ) if X is a space equipped with (a) the indiscrete topology or (b) the
discrete topology?

2. Find π1 ( ,1), where  is equipped with the co-finite topology.

3. Let α and β be two loops at x ∈ X , and let γ be a path in X starting at x. Show that
αβ  {0,1} αγγ −1β .
4. Show that if X is a simply connected space and α, β are two paths in X such that
α(0) = β(0), α(1) = β(1), then α  {0,1} β .
5. Let α be a loop in a path connected space X, and let α(0) = α(1) = x . Let F : X × I → X
be an ambient isotopy within X from α(I) to F (α( I ) × {1}) ⊂ X , and define a loop β
at y = F (α(0) × {1}) in X as follows: β(t ) = F (α(t ) × {1}). Denote the restriction of F to

K12146_Book.indb 244 2/9/15 3:10 PM

 10.4 The Fundamental Group of a Space   ◾   245  

{x } × I by δ. Then δ is a path from x to y in X. Show that α and δβδ −1 are loops at x

homotopic in X relative to {0,1}.
6. (a) Show for each path α : I → S 2 = {( x , y , z ) ∈ 3 : x 2 + y 2 + z 2 = 1} there exists a path
β : I → S 2 such that α  β and such that β( I ) ≠ S 2.
(b) Show that S 2 is simply connected.
7. Prove that if A and B are path connected and simply connected open subspaces of
a space X, and if A ∩ B is path connected and non-empty, then A ∪ B is simply
connected.
8. A tree is a 1-complex T such that no subspace of T is homeomorphic to S1 . Show that
every finite tree is simply connected.
9. (a) Show that the (path disconnected) space {( x , y ) ∈  2 : y = sin x1 , 0 < x ≤ 1}
∪ {(0, y ) ∈ 2 : − 1 ≤ y ≤ 1} is simply connected (see Illustration 6.36 in 6.5).
(b) Find an open, simply connected and connected subspace A of  2 such that ∂(A)
is not path connected.

10. Let α : I → S1 × S1 be the loop at x = ((1,0),(1,0)) defined by α(t ) = ((cos 2πt ,sin 2πt ),
(1,0)), and let β : I → S1 × S1 be the loop at x = ((1,0),(1,0)) defined by
β(t ) = ((1,0),(cos 2πt ,sin 2πt )). Recalling that S1 × S1 is a torus, visualize α and β. Show
that the subgroup of π1 (S1 × S1 , x ) generated by [α] and [β] is commutative by visualizing
a homotopy between αβ and βα.
11. Complete the proof of Proposition 9.

12. Show that π1 ( ∏ X j , x ) ≅ × π1 ( X j , p j ( x )), where pk : ∏ X j → X k is the projection.

j∈ J j∈ J j∈ J
13. Exhibit infinitely many pairwise non-homeomorphic, path connected, and simply
connected spaces. Justify!

14. Show that if f : X → Y is a homotopy equivalence relative to {x} then f∗ : π1 ( X , x ) →

π1 (Y , f ( x )) is an isomorphism.

15. Let α : I → X be a loop at x ∈ X . Then f : S1 → X defined by f ((cos 2πt ,sin 2πt )) = α(t )
 and continuous. Show that α  cx if and only if there is a continuous
is well defined
extension f : D 2 → X of f.

16. The space X shown in Illustration 10.32 consists of a hollow ball {( x , y , z ) ∈  3 :

1.5 ≤ x 2 + y 2 + z 2 ≤ 2} through which we drill 6 tunnels leading from the hollow center
to the space outside the large sphere. Does the homotopy class of the loop α (see fig-
ure below) belong to the subgroup of the fundamental group of X at x generated by
the homotopy classes of the loops β1 , β 2 , β3 , β 4 , β5 ? If yes, then express [α ] in terms
of [β1 ] , [β 2 ] , [β3 ] , [β 4 ] , and [β5 ].

K12146_Book.indb 245 2/9/15 3:10 PM

 246   ◾    Isotopy and Homotopy

17. Recall that a subset X of n is convex if for every x , y ∈ X , the line segment
{(1 − t )x + ty : t ∈[0,1]} is a subset of X. Prove that every convex subset X of n is simply
connected.

β1

β5
x β2

β4
β3

Illustration 10.32

18. Let f : ( X , x ) → ( X , x ) be continuous and denote the subgroup f∗ (π1 ( X , x )) of π1 ( X , x )

by H. Show that for every homotopy F : X × I → X relative to {x} and such that
F X ×{0} = f , we have F∗ (π1 ( X × I ,( x ,0))) = H .

K12146_Book.indb 246 2/9/15 3:10 PM

 Chapter 11

The Fundamental Group of

a Circle and Applications

W e will go the extra mile here to show that the fundamental group of a circle is infi-
nite cyclic. The payoff is plentiful: it is amazing how many relevant, varied and deep
results can be extracted from π1 (S1 , x ) =  .

11.1 The Fundamental Group of a Circle

A brief historical note: The history of the proof of the claim that π1(S1 , x ) =  is rather fuzzy. It
is one of these statements that were too “obvious” to deserve a formal proof. The first textbook
where one can find a rigorous proof of the claim appears to be Seifert–Threlfall’s seminal work
[68].

Our goal in this section is to compute the fundamental group of a circle.

We start by introducing the notion of a Lebesgue subdivision of a continuous mapping
f : A → X , where A is the rectangle I × [0, a], 0 ≤ a (as a subspace of  2 ), and X is any space
covered with a family of open sets U = {U j : j ∈ J }. The set { f −1 (U j ) : U j ∈U } is an open
cover of A. Since A is compact, there is a Lebesgue number ε associated with this cover.
We subdivide I into finitely many subintervals [t0 = 0, t1 ], [t1 , t 2 ], … , [ti , ti+1 ], … , [tn−1 , tn = 1],
and subdivide [0, a] into [s0 = 0, s1 ], [s1 , s2 ], … , [sk , sk+1 ], … , [sm−1 , sm = a] , such that each
little rectangle Ri , k = [ti , ti+1 ] × [sk , sk+1 ] is of diameter ≤ ε. Then each Ri , k is completely in
one of f −1 (U j ) ,

247

K12146_C011.indd 247 2/12/15 12:03 PM

 248   ◾    The Fundamental Group of a Circle and Applications

or, equivalently, each f ( Ri ,k ) , i ∈{0,1,…, n − 1}, k ∈{0,1,…, m − 1}, is completely in

one of U j ∈U . Denoting fi ,k = f Ri ,k , we say that the set of continuous mappings
{ fi ,k : i ∈{0,1,…, n − 1}, k ∈{0,1,…, m − 1}} is a Lebesgue subdivision of the mapping f (rela-
tive to the cover U ); see Illustration 11.1.

f0,1(R0,1) f1,1(R1,1)

f0,1 f1,1

R0,1 R1,1 f U3

R0,0 R1,0 A
f1,0 U2
f0,0 f0,0(R0,0) U1

f1,0(R1,0)

Illustration 11.1 A Lebesgue subdivision.

When a = 0 then f yields a path in X. In this case each of f ([ti , ti+1 ]) , i ∈{0,1,…, n − 1},
(t0 = 0 , tn = 1), is a path which is completely in one of U j ∈U . After relabeling the sets
from the open cover U we may assume that f ([ti , ti +1 ]) ⊂ U i +1 , i ∈{0,1,…, n − 1} (where we
do not assume U1 ,U 2 ,… ,U n are distinct). Denoting fi = f [ti−1 ,ti ], we have f = f1 f 2 … fn .
The last expression for f as a product of paths will be referred to as a Lebesgue subdivision
of the path f (relative to the cover U); see Illustration 11.2.

I = [0,1] f U3

U1 U2

Illustration 11.2 A Lebesgue subdivision of a path.

The following notational abbreviation is standard: for any loop α, we will use [α]n to
denote the product [α][α]…[α], n times.
The fundamental group π1 ( X , x ) of a space X is, in a way, a measure of the num-
ber of “certain holes” of the space. More than that: it describes how these “holes” are
interconnected within the space. An archetype of a “one-holed” space is the circle S1 ;
so it is prudent to begin our computations of fundamental groups of more complicated
spaces by computing π1 (S1 , x ) . We will start doing that in a moment. We must point
out that Theorem 1, the proof of which will take the rest of this section, will eventually
be a very simple corollary of the theory that we will develop independently (Example 2
in 16.2). Hence, it is logically sound to accept the truth of Theorem 1 without proving it
at this stage. However, in the forthcoming sections we will present an array of amazing

K12146_C011.indd 248 2/12/15 12:03 PM

 11.1 The Fundamental Group of a Circle   ◾    249  

applications of this theorem. So, it seems psychologically necessary to place the theorem
on solid ground first. Hence … .

Theorem 1. π1 (S1 , x ) is an infinite cyclic group.

It follows from Proposition 2 of 10.4 that the theorem is not affected by the choice of
x ∈S1; for convenience in the following argument we choose x = (1,0).
We will split the proof of the theorem into two parts.

Part 1: π1 (S1, x ) is cyclic.

Part 2: π1 (S1, x ) is infinite.

The proofs of both parts are fairly technical, and, so, in both cases we will start by outlining
the main ideas of the proofs.

Proof of Part 1:
Here is an outline of the basic idea of this part of the proof: We split S1 into two simply
connected sets, and express every loop in S1 as a product of paths, each in one of these two
sets. We will use the fact that every two paths between any two fixed points in a simply
connected set are homotopic, and deduce from it that that every loop is homotopic to the
simplest loop that winds several times around the circle.
We prove that π1 (S1 , x ) is generated by [α], where α : I → S1 is the loop defined
by α(t ) = (cos2πt , sin2πt ),0 ≤ t ≤ 1 (Illustration 11.3). Notice that α = α1α 2, where α1 (t ) =
(cos πt ,sin πt ), 0 ≤ t ≤ 1, and α 2 (t ) = (cos(t + 1)π ,sin(t + 1)π ), 0 ≤ t ≤ 1.

S1
0 1 x = (1,0)
Illustration 11.3

Denote U1 = {( x , y ) ∈ 2: x 2 + y 2 = 1, y > −0.25} and U 2 = {( x , y ) ∈ 2: x 2 + y 2 = 1, y < 0.25}.

Then {U1 ,U 2 } is an open cover of S1 (Illustration 11.4). Observe that both U1 and U 2 are
simply connected.

U1

S1

y x

U2 Illustration 11.4 x = (1,0); y = (−1,0).

K12146_C011.indd 249 2/12/15 12:04 PM

 250   ◾    The Fundamental Group of a Circle and Applications

Let β be a loop at x in S1, giving rise to the element [ β] ∈π1 (S1, x ). Let β1β 2 …βm be a
Lebesgue subdivision of the path β relative to the cover {U1 ,U 2 } . This means that for every
i = 1,2,…, m, βi ([ti −1 , ti ]) ⊂ U j for some j ∈{1,2}. If both βi ([ti −1 , ti ]) and βi +1 ([ti , ti +1 ]) are
contained in a single U j , then we can replace βi and βi+1 by their product in the Lebesque
subdivision. Repeating this procedure we can guarantee that no two βi ([ti −1 , ti ]) and
βi +1 ([ti , ti +1 ]) are in a single U j . After relabeling, we have a Lebesgue subdivision β1β 2 …βm
of β satisfying that property.
Denote y = (−1,0). There are two path components of U1 ∩ U 2 : one contains x, the
other y. We denote the former by U x , and the latter by U y . The modification of the
original Lebesgue subdivision of β that we have performed in the preceding paragraph
guarantees that the endpoints βi (ti −1 ) and βi (ti ) of each βi are either in U x or in U y .
If βi (ti ) ∈U z , z ∈{x , y}, then we choose a path γ i in U z from βi (ti ) to z; this could be
done since U z is path connected. Since γ i γ i−1 is homotopic to the constant path at βi (ti ) ,
it follows from the associativity of the operation of multiplication of homotopy classes
of paths that [β1β 2β3 …βm−1βm ] = [β1γ 1 ][γ 1−1β 2 γ 2 ][γ 2−1β3 γ 3 ]…[γ m−1−2βm−1γ m−1 ][γ m−1−1βm ], where
each path in square parenthesis leads from x or y to x or y (it may be a loop) within one
single U j , j ∈{1,2}.
If some γ i−−11βi γ i is a loop at z ∈{x , y}, then, since U1 and U 2 are simply connected, it is
homotopic to the constant map at z, and so [γ i−−11βi γ i ] could be omitted without affecting
[β]. So, we may suppose that every γ i−−11βi γ i is a path from one of x, y to the other. Then, if
necessary, we repeat the same argument as above to ensure again that no two consecutive
γ i−−11βi γ i and γ i−1βi +1 γ i +1 are paths in a single U j . We relabel again (if needed) to end up with
[β] = [β1γ 1 ][γ 1−1β 2 γ 2 ][γ −2 1β3 γ 3 ]…[γ m−1−2βm−1γ m−1 ][γ m−1−1βm ] where the paths in the square paren-
theses join x and y (in some order), and each two consecutives paths are in different U1 and U 2 .
It is easy to confirm that all of the above changes of the original Lebesgue subdivision
β1β 2 …βm of β had no effect whatsoever on the fact that the first path of the product starts
at x, and that the last path of the product ends at the same point, i.e., that we are dealing
with a loop at x.
We are almost done: β1 γ 1 is a path from x to y in U j ( j ∈{1,2}). Suppose it is in U1 . Since
α1−1 is a path from y to x in U1, and since U1 is simply connected, it follows that [β1 γ 1 ] = [α1 ]
(Exercise 4 in 10.4). Hence [β] = [α1 ][γ 1−1β 2 γ 2 ][γ 2−1β3 γ 3 ]…[γ m−1−2βm−1 γ m−1 ][γ m−1−1βm ]. The next
path γ 1−1β 2 γ 2 must start at y and end at x, and it must be in U 2 . Arguing similarly as
for β1 γ 1 we conclude that [γ 1−1β 2 γ 2 ] = [α 2 ]. So, [β] = [α1 ][α 2 ][γ 2 −1β3 γ 3 ]…[γ −1m−2βm−1γ m−1 ]
[γ −1m−1βm ]. Repeating this argument m many times (and using induction), we conclude
that [β] = [α1 ][α 2 ][α1 ][α 2 ]…[α1 ][α 2 ]; the last homotopy class in this product must be [α 2 ]
since γ m−1−1βm ends at x. At the same time we observe that m is even. Since [α1 ][α 2 ] = [α]
m
it follows that [β] = [α] 2 .
If in the initial step β1 γ 1 is a path from x to y in U 2 , then only slight modifications of the
−m
above argument are needed to conclude that [β] = [α] 2 .
Summarizing: we showed in this part that every [β] ∈π1 (S1 , x ) is a power of [α] or its
inverse [α]−1 , confirming that π1 (S1 , x ) is a cyclic group generated by [α] and completing
the proof of Part 1.

K12146_C011.indd 250 2/12/15 12:04 PM

 11.1 The Fundamental Group of a Circle   ◾    251  

Proof of Part 2:
Here is a sketch of the basic idea of the rest of the proof: We associate with every loop at
x = (1,0) in S1 a path in  that we call a lift of the loop, and then we show that two loops
in S1 are homotopic if and only if their lifts starting at a fixed point must end at the same
point. At the end we merely notice that, for every m ≠ n, the lifts of α n and α m starting
at 0 ∈ end at different points, so that [α ] and [α ] must be different. It follows that
n m

π1 (S1 , x ) is an infinite cyclic group.

It suffices to show that [α]j ≠ [α]k for j ≠ k. The proof we will give is a special case of the
theory of covering spaces that we will develop in a few chapters.
Consider the continuous mapping p :  → S1 defined by p(s ) = (cos2π , sin 2πs ) and
depicted in Illustration 11.5.
 

2 2
2

1 1
1
p–1(U1) p–1(U2)

0 0
0

p p
p
U1
S1 (1,0)
S1 (1,0) S1 (1,0)
U2

Illustration 11.5 Illustration 11.6 Illustration 11.7

If U j ∈{U1 ,U 2 } , where {U1 ,U 2 } is the cover of S1 as in the proof of Part 1, then p −1 (U j )

( j ∈{1,2}) is a disjoint union of its path components, each of which is homeomorphic to U j
(Illustrations 11.6 and 11.7), and each containing precisely one integer. Denote the compo-
nent of p −1 (U j ) containing the integer a by Vj ,a , and the restriction of p to Vj ,a by p j ,a ;
note that each p j , a is a homeomorphism. Observe also that if C and D are path components
of p −1 (U1 ) and p −1 (U 2 ), respectively, then C ∩ D is either empty or it is path connected.
Given a continuous mapping f : I × [0, c] → S1 , 0 ≤ c , such that f (0, s ) = f (1, s ) = x
for every s ∈[0, c], we will construct a continuous mapping f : I × [0, c] →  such that
p  f = f . We call f the lift of f; this lift will satisfy additional properties, which we will
identify along the way. Before we proceed, we point out that we are primarily concerned
with two cases: c = 0, in which case f is a loop at x, and c = 1 in which case f is a homotopy
relative to {0,1} from the loop f I ×{0} at x to the loop f I ×{1} at x, where we identify I with
I × {0} and I × {1} via the evident homeomorphisms.
Let { fi ,k : i ∈{0,1, … , n − 1}, k ∈{0,1, … , m − 1}} be a Lebesgue subdivision of the mapping
f (relative to the cover {U1 ,U 2 }) of S1. The domain of each fi .k is the rectangle Ri ,k as in
Illustration 11.8.

K12146_C011.indd 251 2/12/15 12:04 PM

 252   ◾    The Fundamental Group of a Circle and Applications

We construct a lift  fi ,k : Ri ,k →  of each fi ,k , so that these lifts agree over the intersec-
tions of their domains. By the gluing lemma, the mapping f : I × [0, c] →  defined to be

fi ,k over Ri ,k (i ∈{0,1,…, n − 1}, k ∈{0,1,…, m − 1}), would then be a continuous.
In the first step we construct  f0,0 as follows: since f0,0 ( R0,0 ) ⊂ U j for some j ∈{1,2},
define f0,0 = p j ,0  f0,0 . Observe that 
 −1
f0,0 ( R0,0 ) is a subset of the component of p −1 (U j )
containing the integer 0.
Proceed recursively, and assume that we have constructed, row by row, from the bottom to the
top and from left to right, the lifts corresponding to all shaded rectangles in Illustration 11.8 agree-
ing over their intersections. The union of these mappings is a continuous mapping g defined
over these shaded rectangles. We now pay attention to Ri ,k . We lift fi ,k : Ri ,k → S1 to a continu-
ous  fi ,k : Ri ,k →  as follows: By construction fi ,k ( Ri ,k ) is a subset of some U j ( j ∈{1,2}). Let us
assume, to simplify the indexing, that j = 1; the case when j = 2 is symmetric. So, fi ,k ( Ri ,k ) ⊂ U1 .
We will distinguish two cases: (1) both fi −1,k ( Ri −1,k ) and fi ,k −1 ( Ri ,k −1 ) are subsets of U1 , and (2)
at least one of them is a subset of U 2 .
Case (1): Since Ri −1,k ∩ Ri ,k −1 ≠ ∅ , it follows that  fi−1,k ( Ri−1,k ) ∩ 
fi ,k−1 ( Ri ,k−1 ) is also non-
 
empty. Hence the sets fi−1,k ( Ri−1,k ) and fi ,k−1 ( Ri ,k−1 ) are contained in the same path component
of p −1 (U1 ) . Suppose that path component contains the integer a. Then define  fi ,k = p1,−1a  fi ,k .
Since the same p1,−1a has been used to construct all three of  fi−1,k and 
f i ,k ,  fi ,k−1 , it follows that
they agree over the intersections of their domains, and so the extension of g by the mapping 
fi , k
is also continuous.

a+1

R0, m–1 R1, m–1 Ri–1, m–1 Ri, m–1 Rn–1, m–1 fi,k a

a−1
fi–1,k
R0, k R1, k Ri–1, k Ri, k Rn–1, k
p
R0, k–1 R1, k–1 Ri–1, k–1 Ri, k–1 Rn–1, k–1
fi,k
Ri−1,k Ri,k
R0, 1 R1, 1 Ri–1, 1 Ri, 1 Rn–1, 1 x = (0,1)
Ri−1,k−1 Ri,k−1
R0, 0 R1, 0 Ri–1, 0 Ri, 0 Rn–1, 0 fi−1,k
S1

Illustration 11.8 Illustration 11.9

Case (2): We may suppose that fi −1,k ( Ri −1,k ) ⊂ U 2 . By the recursive step, there is an integer
a such that  fi−1,k = p2,−1a  fi−1,k . Pay attention to the vertical edge E common to Ri −1,k and Ri ,k :
since fi ,k ( Ri ,k ) ⊂ U1 , it follows that fi −1,k ( E ) ⊂ U1 ∩ U 2 . Consequently 
fi−1,k ( E ) is a subset
of the intersection of a path component of p −1 (U1 ) with a path component of p −1 (U 2 ).

K12146_C011.indd 252 2/12/15 12:04 PM

 11.1 The Fundamental Group of a Circle   ◾    253  

There are two components of p −1 (U1 ) intersecting p2,−1a (U 2 ); one contains the integer a,
the other contains the integer a −1. In both cases the intersection of the component of
p −1 (U1 ) and p2,−1a (U 2 ) contains  fi−1,k ( E ). Let b ∈{a − 1, a} be the (only) integer in  fi−1,k ( E ).
(In Illustration 11.9 we depict the case when b = a.) Define  − 1
fi ,k = p1,b  fi ,k . It follows imme-
 
diately that fi−1,k and fi ,k agree over E (the intersection of their domains).
Since Ri −1,k ∩ Ri ,k −1 ≠ ∅ , and because g 
(introduced just before Case 1) is well defined
(that is, since  fi−1,k and  fi ,k−1 agree over their
intersection), we have that either  fi ,k−1 = 2
p2,−1b  fi ,k −1, or fi ,k −1 = p1,−1b  fi ,k −1 (depending
on whether f ( Ri ,k −1 ) is in U2 or in U1, respec-
tively). It is evident that in both cases  fi ,k−1 1

and fi ,k agree over the intersection of their
domains.
α2 0
We have shown that the extension of g by p
the mapping  fi ,k is continuous.
We have constructed a continuous map-
ping f : I × [0, c] →  for every c ≤ 0. When α2
c = 0, the mapping f is a loop at x, and its
lift f is a path starting at 0. In Illustration 0 1
S1
2
11.10 we show the lift of the loop α , where x = (1,0)
α is the generator of π1 (S1 , x ) introduced
in the first part of the proof of the theorem. Illustration 11.10
It is clear from our construction that the
lift of α i must end at i. So, if i ≠ k, then the lifts of the loops α i and α k are paths that
end at different integers.
The proof of our theorem will be completed once we make the following observations.
If c = 1, then, as we have observed above, f is a homotopy relative to {0,1}. We claim that in
that case f is also a homotopy relative to {0,1}. This has already been confirmed, except
for the claim that the homotopy is relative to {0,1}. Note that since (by definition) f sends
{0} × I to x, our construction guarantees that f sends the same set to 0, and since f sends
{1} × I to x, f must send the same set to the terminal point of the lift of the path f I ×{0} .
(In particular, the end-points of the lifts of the paths f I ×{0} and f I ×{1} must be the same.)
So, two homotopic loops at x in S1 are lifted to homotopic paths in  starting at 0 and
ending at the same point. Consequently, if j ≠ k , then [α]j ≠ [α]k , proving that [α] gener-
ates an infinite cyclic group. Hence π1 (S1 , x ) is an infinite cyclic group.

Exercises
1. (a) Find the fundamental group of a cylinder.
(b) Find π1 (T 2 , x ) , where T 2 is a torus.
(c) Find π1 ( 2 \{(0,0)}, x ).

K12146_C011.indd 253 2/12/15 12:04 PM

 254   ◾    The Fundamental Group of a Circle and Applications

2. Show that the unit open disk B2 = {( x , y ) ∈ 2 : x 2 + y 2 < 1} is not homeomorphic to

B2 ∪ Y , where Y is any non-empty subset of S1 .
3. Denote X = S 2 \{(0,0, −1),(0,0,1)}. Define the homeomorphism f : X → X by f ( x , y , z ) =
(− x , y , z ). Show that f is not ambient isotopic within X to the identity mapping id X .
[Hint: π1 ( X ) =  .]
4. (a) Show that the equatorial circle C in the torus T 2 (shown in Illustration 11.11) is a
retract of T 2.
(b) Show that C is not a deformation retract of T 2.

T2

Illustration 11.11

5. True or false: if A and B are path connected and simply connected subspaces of a
space X, and if A ∩ B is nonempty, path connected, and simply connected, then so is
A ∪ B.
6. Prove that every bouquet of n circles is a 1-skeleton of a simply connected 2-complex.
7. Let C be a thick cylinder {( x , y ) ∈ 2 : 0.5 ≤ x 2 + y 2 ≤ 1} × I , and let A and B be the two
spirally subspaces of C shown in Illustration 11.12. (A winds exactly once around the
outer surface of C; B winds exactly twice.) Show that there is no homeomorphism
f : C → C such that f ( A) = B and such that f ( x ) = x (x is shown in the Illustration
11.12).

C
B
C
A

x x

Illustration 11.12 To avoid clutter, we show C twice, first with A, then with B.

K12146_C011.indd 254 2/12/15 12:04 PM

 11.2 Brouwer Fixed Point Theorem and The Fundamental Theorem of Algebra   ◾    255  

8. Show that the sphere S 2 is not homeomorphic to the

disk D 2 .
9. The space X depicted in Illustration 11.13 is constructed
as follows: starting with two concentric spheres, remove
four open disk-like subspaces, D11 and D12 from the
first and D21 and D22 from the second sphere; join
the circular boundaries of ∂(D11 ) and ∂(D21 ) , and of
∂(D12 ) and ∂(D22 ) with two tubes, respectively. Find
the fundamental group of X.
Illustration 11.13

11.2 Brouwer Fixed Point Theorem and The

Fundamental Theorem of Algebra

A brief historical note: Jan Brouwer proved the fixed point theorem in 1911. During his early
studies his main interests were split between mathematics and philosophy, so much so that his
Ph.D. advisor refused an early version of Brouwer’s dissertation (eventually defended 1907)
because it incorporated substantial philosophical components. In 1806 (republished in 1813)
Jean-Robert Argand (1768–1822) produced the first rigorous proof of the fundamental theorem
of algebra.

Let A be a retract of a space X, and let a ∈ A. This means that there is a continuous
r : ( X , a) → ( A, a) such that r  i = id A , where i : A → X is the inclusion. Hence, (r  i )∗ = (id A )∗,
and so, since (r  i )∗ = r∗  i∗, and since (id A )∗ = idπ1 ( A ,a ) , we have r∗  i∗ = idπ1 ( A ,a ). In particu-
lar, r∗  i∗ must be a bijection. It is easy to verify that if f  g is a bijection, then g must be
one-to-one, and f must be onto. We conclude from this that i∗ must be a monomorphism
(= one-to-one homomorphism) and that r∗ must be an epimorphism (= onto homomor-
phism). We have proven the following initial result.

Proposition 1. If r : ( X , a) → ( A, a) is a retraction then r∗ : π1 ( X , a) → π1 ( A, a) is an epimor-

phism, and the inclusion i : A → X induces a monomorphism i∗ : π1 ( A, a) → π1 ( X , a) .

Example 1
The circle S1 is not a retract of the disk D 2 . Otherwise, it would follow from
Proposition 1 that there would be an epimorphism from the trivial group π1 (D 2 , a)
onto the cyclic group π1 (S1 , a) , which is clearly impossible. More generally, the same
argument shows that if a subspace A of a simply connected space X is not simply con-
nected, then A is not a retract of X. ☐

In the following examples we will see that π1 ( A, a) ≅ π1 ( X , a) is not a guarantee that a

subspace A of X is a retract of X.

K12146_C011.indd 255 2/12/15 12:04 PM

 256   ◾    The Fundamental Group of a Circle and Applications

Example 2
We have introduced the Möbius band in Exercise 3 of 4.4; it is the quotient space
obtained from a filled rectangle by identifying two opposite edges (see Illustration 4.31
in 4.4). Illustration 11.14(a) shows a Möbius band, denoted M. The middle circle C of
M is a deformation retract of M: the deformation retraction is indicated by the arrows
in Illustration 11.14a. It follows from Corollary 8 of 10.4 that π1 ( M , x ) is infinite
cyclic; it is generated by the homotopy class of the loop α shown in Illustration 11.14b.

B
M

C β

Illustration 11.14a Illustration 11.14b

The boundary B of M (see Illustration 11.14a) is obviously homeomorphic to S1, so its

fundamental group is also infinite cyclic.
We will now show that the boundary B of M is not a retract of M. Assume otherwise,
i.e., assume that there is a retraction r : M → B. By Proposition 1, r∗ : π1 ( M , x ) → π1 ( B, x )
is an epimorphism from an infinite cyclic group to another infinite cyclic group. Since
every such epimorphism sends a generator to a generator, it follows easily that r∗
must be an isomorphism. That, together with r∗  i∗ = idπ1 ( B , x ), implies that i∗ is also
an isomorphism (where, as before, i : B → M is the inclusion). However, one of the two
generators of π1 ( B, x ) is the homotopy class of the loop β (see Illustration 11.14b),
and, since β  α 2 , we have that i∗ ([β]) = [α]2 . Hence, i∗ is not onto, and so not an
isomorphism. We have a contradiction. ☐

Recall that D n = {( x1 , x 2 ,…, xn ) ∈n : x12 + x 22 +  + xn2 ≤ 1} is the closed n-disk.

Theorem 2. (Brouwer fixed-point theorem) Every continuous mapping D n → D n has a

fixed point.

The case n = 1 is Exercise 10 in 6.2. The case n = 2 is within the scope of our theory so
far and will be proven below. We will not prove the statement for n > 2; an explanation will
be provided in Section 14.5 when we touch on higher homotopy groups.

Proof (n = 2).
Suppose there is a continuous f : D 2 → D 2 such that f ( x ) ≠ x for every x ∈D 2 . This
implies that there is a unique semi-line lx in  2 starting at f ( x ) and passing through x
(Illustration 11.15).

K12146_C011.indd 256 2/12/15 12:04 PM

 11.2 Brouwer Fixed Point Theorem and The Fundamental Theorem of Algebra   ◾    257  

g(x) S1

x
f(x)

D2
Illustration 11.15

Define a mapping g : D 2 → S1 as follows: g sends every x ∈D 2 to the intersection of lx

and the circle S1 (Illustration 11.15). It is obvious that g is continuous, and that it fixes every
x ∈S1, hence, it is a retraction D 2 → S1 . By Example 1, that is not possible, hence the initial
hypothesis in this proof fails.

Readers who do not like visual arguments are invited to provide an algebraic justification
(Exercise 2). Since “having a fixed point” is a topological property (Exercise 17 in 3.5), we could
replace D n in the statement of the Brouwer fixed point theorem with any homeomorphic space.

If every continuous f : X → X has a fixed point then we say that X has the fixed-point
property.
Here are a couple of nonmathematical versions of instances of Theorem 2:

(a) (Crumpling Map Theorem) Crumpling a map and throwing it over a copy of the
same map will always have two corresponding points of the maps on the same verti-
cal line (case n = 2).
(b) (Cup of Coffee Theorem) Stirring a cup of coffee will always leave a point unmoved
(case n = 3).

The following example provides a method for establishing existence of solutions of sys-
tems of n equations with n unknowns. Our incomplete proof of the Brouwer fixed point
theorem provides justification only when n = 2.

Example 3
We want to establish that a system of n equations with n unknowns is consistent (i.e.,
that a solution exists).

h1 ( x1 , x 2 ,… , xn ) = 0
h2 ( x1 , x 2 ,… , xn ) = 0


hn ( x1 , x 2 ,… , xn ) = 0

K12146_C011.indd 257 2/12/15 12:05 PM

 258   ◾    The Fundamental Group of a Circle and Applications

Suppose that each of h1 , h2 ,… , hn is a continuous function; then the function f : n → n

defined by f ( x ) = (h1 ( x ) + x1 , h2 ( x ) + x 2 ,… , hn ( x ) + xn ) is also continuous. Suppose we can
identify any subset A of n such that A ≅ D n, and such that f ( A) ⊂ A. Then, by the Brouwer
fixed point theorem, there is an element a = (a1 , a2 ,… , an ) ∈n such that f (a ) = a . This a is
then a solution of the original system, and our goal would thus be accomplished. ☐

Example 4 has a similar flavor.

Example 4
Let us call a subspace C of  3 a conic subspace if it consists of all of the points on a set
of rays originating at (0, 0, 0), and if the intersection of all rays in C and the sphere S 2 is
homeomorphic to the closed disk D 2 (Illustration 11.16). For example, the first octant of
 3, consisting of all points (x, y, z) such that x ≥ 0, y ≥ 0 and z ≥ 0 , is a conic subspace.

~
= D2

Illustration 11.16 In this figure we show the

boundary of a conic space C and the inter-
section of C with a sphere.

Let f :  3 →  3 be continuous and such that f (C \{(0,0,0)} ⊂ C \ {(0,0,0)} for some

conic subspace C. Then, we claim that there is a point x 0 ∈C such that (0, 0, 0), x 0
and f ( x 0 ) are collinear. This means that we claim that there is an element x 0 ∈C
and a number k ∈ + such that f ( x 0 ) = kx 0 . We justify it as follows. Denote the
restriction of f to D = S 2 ∩ C by g. Denote the coordinate functions of g by g 1 , g 2 , and
g 3 , and define g by g ( x , y , z ) = g 1 ( x , y , z )2 + g 2 ( x , y , z )2 + g 3 ( x , y , z )2 . Observe
g
that maps D into itself. Since D is homeomorphic to D 2 , there must be a fixed
g
g
point x 0 for . That gives g ( x 0 ) = g ( x 0 ) x 0 , and so f ( x 0 ) = f ( x 0 ) x 0 , that is,
g
f ( x 0 ) = kx 0 for k = f ( x 0 ) .
Here is a seemingly unrelated simple consequence: every 3 × 3 matrix A with posi-
tive real entries has a positive eigenvalue. Every such matrix maps the first octant with-
out the origin into itself, and the above argument establishes existence of a vector x
such that Ax = A( x ) x . Hence, A( x ) is a positive eigenvalue for A, as asserted. ☐

K12146_C011.indd 258 2/12/15 12:05 PM

 11.2 Brouwer Fixed Point Theorem and The Fundamental Theorem of Algebra   ◾    259  

We are still at the early stage of the development of our theory; yet what we have so far
is sufficient to deduce a major theorem in algebra.

Theorem 3. (The Fundamental Theorem of Algebra) Every polynomial a0 + a1 z + a2 z 2 +

 + an z n with complex coefficients and of degree ≥1 has a complex root.

Proof. Let an ≠ 0 , so that the degree of p( z ) = a0 + a1z + a2 z 2 +  + an z n is n. For what fol-

lows, there is no loss in generality to take that an = 1 .
Assume p( z ) = 0 has no complex solution. Hence, p maps  2 into  2 \{(0,0)}. Consider
the restriction of p to the circle rS1 = {r (cos2πt ,sin2πt ) : 0 ≤ t ≤ 1} of radius r ≥ 1, and
let α : I → rS1 be the loop defined by α(t ) = r (cos2πt , sin2πt ). Then β = p  α is a loop in
 2 \{(0,0)} (Illustration 11.17).

p
S1
α
I

zn

Illustration 11.17

Since the restriction of p to the closed disk bounded by rS1 is an extension of p rSit , it fol-
lows that the loop β is homotopic to the trivial loop (Exercise 15, Section 10.4). On the
other hand, the loop δ we get by composing α with the mapping z n winds n times around
the circle r n S1 (Illustration 11.17), and since we have assumed that n ≥ 1, that loop defines a
non-trivial element in the fundamental group of r n S1 . Since r n S1 is a deformation retract
of  2 \{(0,0)} , it follows that δ defines a non-trivial element in the fundamental group of
 2 \{(0,0)} . We will get a contradiction by showing that a conjugate of β and δ are homo-
topic in  2 \{(0,0)} .
We restrict p to rS1, and rearrange slightly: p( z ) = a0 + a1 z + a2 z 2 +  + an−1 z n−1 + z n =
1
z n (1 + q( z )) where q( z ) = n ( a0 + a1z + a2 z 2 +  + an−1z n−1 ). We claim that for sufficiently
z
large r, the mapping F : rS1 × I →  2 \{(0,0)} defined by F (z , t ) = z n (1 + tq(z )) is a homo-
topy from z n to p(z) (both restricted to rS1). The only non-obvious part of this claim is that
the range of F avoids the origin (0, 0), and we handle it as follows:
( ∗)
 
 a0 a1 an −1  1
1 + tq( z ) ≥ 1 − tq( z ) ≥ 1 − q( z ) ≥ 1 −  n + n −1 +  + ≥ 1 − ( a0 + a1 +  + an −1 ) ,
z z z  r

and since (∗) could be made as small as we wish by taking sufficiently large r, it follows that
for such r we have 1 + tq(z ) > 0.

K12146_C011.indd 259 2/12/15 12:05 PM

 260   ◾    The Fundamental Group of a Circle and Applications

We have proved that the loops β and δ are homotopic in  2 \{(0,0)}. It follows from
Proposition 4 in 10.3 that a conjugate γβγ −1 of β, and δ, are homotopic relative to {0,1},
and so they define the same element in the fundamental group of  2 \{(0,0)}. We have a
contradiction.

Exercises
1. Let x 0 ∈S 2 and let B be the (3-dimensional, open) ball B( x 0 , ), where  ∈(0,2).
Show that there is no continuous one-to-one mapping S 2 → S 2 such that f ( x ) = x
for every x ∈S 2 ∩ B , and f ( x ) ≠ x for every x ∈S 2 \ B .
2. Provide an algebraic proof of the Brouwer fixed-point theorem (Theorem 2) for n = 2.
[Hint: Find algebraically the intersection of the lines through x and f ( x ) and the circle S1.]
3. Let X be any nonempty subset of S1 . Find a continuous mapping f : D 2 \ X → D 2 \ X
such that f does not have any fixed points.
4. Prove that the torus S1 × S1 is not a retract of the solid torus S1 × D 2.
5. Prove that the boundary B of the triply-twisted band T shown in Illustration 11.18 is
not a retract of T.

B
   Illustration 11.18

6. Let A be a subspace of X, let a ∈ A. Show that if π1 ( A, a) > π1 ( X , a) then A is not a

retract of X.
7. Let A be a retract of a space X, and let X have the fixed-point property. Show that A
has the fixed-point property.
8. Show that the subspace of  2 depicted in Illustration 11.19 has the fixed-point
property.
Illustration 11.19 The space is not homeo-
morphic to a disk: each of the three objects
juggled by the silhouette is touched at a sin-
gle point.

K12146_C011.indd 260 2/12/15 12:05 PM

 11.3 The Jordan Curve Theorem   ◾    261  

9. Show that every tree with finitely many vertices has the fixed-point property.
10. Let X be the space obtained from a sphere by identifying two distinct points. Prove
that X is not simply connected. [Note: Prove it!]
11. Let f : D 3 → D 2 , f ( x , y , z ) = ( f1 ( x , y , z ), f 2 ( x , y , z )) be continuous and such that
( f1 ( x , y , z ), f 2 ( x , y , z )) ≤ 1 − z 2 for every ( x , y , z ) ∈D 3 . Prove that there are
uncountably many points ( x , y , z ) ∈D 3 such that f ( x , y , z ) = ( x , y ).
12. An infinitely tall cup {( x , y , z ) : x 2 + y 2 = 1, z ≥ 0} ∪ D 2 is filled with coffee. Show that
stirring the coffee always leaves uncountably many points such that for each of them
its original position and its position after the stirring is on the same vertical line. [For
those who do not like the often loose jargon of recreational mathematics, you are asked
to prove that for every continuous mapping f : C → C , C = {( x , y , z ) : x 2 + y 2 ≤ 1, z ≥ 0},
there are uncountably many points x ∈C such that x and f ( x ) are on the same vertical
line.]
13. Let f be a continuous mapping T → T , where T = S1 × D 2 is a solid torus. Show that
there are uncountably many points x ∈T such that x and f ( x ) are on the same lati-
tudinal circle S1 × { y} , for some y ∈D 2 .
14. Show that the following system is consistent.

sin( xy + 1) − x = 0

cos( x + 2 y 2 + 4) − y = 0

15. Show that the following system is consistent.

1
e xyz = 2 xyz
x
e +1
1
cos( x + y + z ) = 2 xyz
y
e +1
1
sin( x + y + z ) = 2 xyz
z
e +1

11.3 The Jordan Curve Theorem

A brief historical note: Bernard Bolzano was the first to state and try to prove Theorem 1
(1808–1813). Jordan’s proof in 1887 contained some gaps, and the first rigorous proof is usually
credited to Oswald Veblen (1880–1960), who published it in 1905.

K12146_C011.indd 261 2/12/15 12:05 PM

 262   ◾    The Fundamental Group of a Circle and Applications

The main goal in this section is to give a proof of the Jordan curve theorem and to present
(without proofs) some important related results. We start with a couple of notions and the
statement of the main theorem.
A Jordan curve in  2 is the image f (S1 ) of the unit circle S1 under an embedding
f : S1 →  2 . A simple arc in  2 is the image f ([a, b]) under an embedding f :[a, b] →  2 .
There is no difference between path components and components of open subsets of  2
(Exercise 3 in 6.5). Hence, when we deal with open subsets of  2 (which is what we will do
in this section), we will use the word component to also mean path component.

Theorem 1. (Jordan Curve Theorem) For every Jordan curve C in  2 , the subspace  2 \ C
has two components, one bounded, the other unbounded.
The proof we give is based on Maehara [46]. Another proof can be found in Moise’s
textbook [55].
The following result is interesting in its own right.

Lemma 2. Denote the square {( x , y ) ∈ 2 : x ≤ 1 and y ≤ 1} by S, let f :[−1,1] → S be

a continuous function satisfying f (−1) ∈{( x , −1) : x ≤ 1} and f (1) ∈{( x ,1) : x ≤ 1} ,
and let g :[−1,1] → S be a continuous function satisfying g (−1) ∈{(−1, y ) : y ≤ 1} and
g (1) ∈{(1, y ) : y ≤ 1} . Then f ([−1,1]) ∩ g ([−1,1]) ≠ ∅.
In other words, under the assumptions of the lemma we assert that the curves defined
by f and g must intersect (see Illustration 11.20).
y
f (1)

g (–1)

g (1)

f (–1)
Illustration 11.20 The curves f and g intersect.

Proof. Let f1 and f 2 be the coordinate functions of f, so that f (t ) = ( f1 (t ), f 2 (t )), t ∈[−1,1].

Similarly, let g (t ) = ( g 1 (t ), g 2 (t )), t ∈[−1,1]. We want to prove that there are s, t ∈[−1, 1]
such that f (s ) = g (t ). Assume otherwise. So, for every s , t ∈[−1,1], either f1 (s ) ≠ g 1 (t )
{ }
or f 2 (s ) ≠ g 2 (t ). Hence, the function H (s , t ) = max f1 (s ) − g 1 (t ) , f 2 (s ) − g 2 (t ) is such
 f (t ) − g 1 (s ) g 2 (t ) − f 2 (s ) 
that H (s , t ) > 0 all the time. Consequently K (s , t ) =  1 , is a well-
 H (s , t ) H (s , t ) 
f (t ) − g 1 (s ) g 2 (t ) − f 2 (s )
defined function for every s , t ∈[−1,1]. Since −1 ≤ 1 , ≤ 1, it follows
H (s , t ) H (s , t )
that K (S ) ⊂ S. Since the square S is homeomorphic to D 2 , it follows from the Brouwer’s fixed

K12146_C011.indd 262 2/12/15 12:05 PM

 11.3 The Jordan Curve Theorem   ◾    263  

point theorem that K must have a fixed point. Hence, there is a pair (s0 , t0 ) ∈S such that
f (t ) − g 1 (s ) g 2 (t ) − f 2 (s )
K (s0 , t0 ) = (s0 , t0 ). Notice that, for every s , t ∈[−1,1], at least one of 1 ,
H (s , t ) H (s , t )
is in the set {−1,1}. Hence, either s0 = 1 or t0 = 1. By symmetry we may assume that s0 = 1.
f (t ) − g 1 (1)
If s0 = 1 , then K (1, t0 ) = (1, t0 ). This and the definition of K imply that 1 0 = 1,
H (1, t0 )
which is not possible since f (t0 ) − g 1 (1) = f (t0 ) − 1 ≤ 0 and (as we have noticed earlier)
f (t ) − g 1 (−1)
H (s , t ) > 0. If s0 = −1 , then K (−1, t0 ) = (−1, t0 ) and so 1 0 = −1 which is also
H (1, t0 )
impossible since f1 (t0 ) − g 1 (−1) ≥ 0 .

Lemma 3. Let C be a Jordan curve in  2 . For every component U of  2 \ C , we have

∂(U ) = C .
Proof. Since C is compact, it is closed in  2 , and so  2 \ C is open. It follows from
Proposition 2 in 6.5 that the components of  2 \ C are also open. Consequently, for every
component U of  2 \ C , ∂(U ) ⊂ C .
Suppose there is a component U of  2 \ C such that ∂(U )  C . Hence, there is a point in
c ∈C \ ∂(U ). There is an open neighborhood of c avoiding ∂(U ), or else c ∈∂(∂(U )), which
could not be, since ∂(∂(U )) ⊂ ∂(U ) (Exercise 10(b) in 3.2). It follows that ∂(U ) is a subset
of a closed arc A in C (where a closed arc in C is the image of a closed arc in S1 under the
embedding sending S1 onto C).
Let D be any closed disk such that C ⊂ D . Since the closed simple arc A is homeomor-
phic to any closed interval on  , the Tietze Extension Theorem (Theorem 1 in 9.2; see also
Exercise 1 in Section 9.2) implies that the identity mapping A → A can be extended to a
continuous g : D → A.
There are two cases: (i) U is bounded, and (ii) U is unbounded.
 g ( x ) if x ∈U
In case (i) we define a mapping h : D → D as follows: h( x ) =  c
. Observe
 x if x ∈U ∩ D
that U ⊂ D or else there will be points in ∂U \ D . Since U and U c are closed, and since g is
the identity over ∂(U ) = U ∩ U c , it follows from the gluing lemma that h is continuous. Since
U ⊂ D, it follows that h ∂ D = id ∂ D . Hence, ∂D is a retract of D, contradicting Example 1 in 11.2.
 x if x ∈U
Case (ii) requires a similar argument: define h : D → D by h( x ) =  c
.
 g ( x ) if x ∈U ∩ D

This time U c must be a subset of D, and the same argument as in part (i) brings us to
the conclusion that ∂D is a retract of D, again achieving a contradiction.
Hence, it is not possible that ∂(U )  C . Thereby ∂(U ) = C .
We are ready for the proof of the main theorem.
Proof of Theorem 1.
Step 1. In the first part of the proof we will find a point in a bounded component of  2 \C.
(That this point is indeed in a bounded component of  2 \C will be shown in Step 2.)
Since C is closed, there are two points a, b ∈C such that d(a, b) is maximal. We may
assume that d(a, b) = 2, and that a = (−1,0), b = (1,0). Hence, C lies within the rectangle
R = {( x , y ) ∈ 2 : −1 ≤ x ≤ 1, − 1.1 ≤ y ≤ 1.1} , and a and b are the only points in ∂ R ∩ C .

K12146_C011.indd 263 2/12/15 12:05 PM

 264   ◾    The Fundamental Group of a Circle and Applications

The points a and b split C into two simple curves Cn and Cs (sharing the endpoints a and b).
By Lemma 2 both of these two simple curves intersect the vertical line l = {( x , y ) ∈ 2 : x = 0,
−1.1 ≤ y ≤ 1.1}. Identify the point cn in C ∩ l that is closest to the point p = (0,1.1), and
let Cn be the one of the two curves Cn and Cs that contains cn . Let dn be the point in Cn ∩ l
farthest from p; it could happen that dn = cn . We illustrate all this in Illustration 11.21.
R p

cn

Cn

dn

a b

Cs

Illustration 11.21
From now on, for convenience, we will use the following notation. For distinct x , y ∈R we
will denote the path from x to y along the line segment between x and y by xy ; we will use the
same notation for the image of that path (an arc in  2 ). If f is a path in  2 and if f (s ) = x and
,
f (t ) = y for some s , t ∈I , then we will denote the restriction f [s , t ] , as well as its image by xy
where f will either be clear from the context or will be mentioned explicitly.
Denote q = (0, −1.1). It has to be that dn q ∩ Cs ≠ ∅, for otherwise the product of paths
pcn c n dn dn q and the path Cs do not intersect, contradicting Lemma 2. Let cs ∈Cs ∩ dn q be
the closest to q, and ds ∈Cs ∩ dn q farthest from q. Finally, let z be the point in the middle
of the line segment dn ds . See Illustration 11.22.
R p

cn

Cn

dn
z b
a
ds

Cs ds
cs
l

q
Illustration 11.22

K12146_C011.indd 264 2/12/15 12:05 PM

 11.3 The Jordan Curve Theorem   ◾    265  

Step 2. Let U be the component of  2 \C containing z. In this step we show that U is bounded.
Suppose U is the only unbounded component of  2 \C (see Exercise 1). Then there is a
path λ in U leading from z to a point outside R. Since ∂ R splits  2 into two components,
there must be a point w ∈∂ R where λ meets ∂ R for the first time. Since a, b ∈C, w is neither
a nor b. By symmetry, we may assume that w is on the vertical line segment containing b, and
below b (see Illustration 11.23).
R p

cn pcn

Cn
cndn
dn
dnz
a z b
ds zw

Cs w λ
cs
l
δ
q
Illustration 11.23

Then the path δ from w vertically down to the corner (1,−1.1), and then horizontally to q
does not touch a and b. Look at the path pcn c  
n dn dn z zw δ , where cn dn is along Cn and zw

is along λ: it does not intersect Cs , contradicting Lemma 2.
Step 3. We show that U is the only bounded component of  2 \ C , and the theorem will
be proven.
Let W be another bounded component of  2 \ C . It follows from the argument given in
Step 2 that W ⊂ R \ C . The path α = pcn c 
n dn dn ds ds c s c s q (Illustration 11.24) does not inter-
sect W since each of the indicated sub-paths is either in U, or in the unbounded component
of  2 \C , or it is a part of C.
R p

cn

Cn

dn
z
a b1 b
a1
ds
Ua Ub

Cs
cs
α
l β

Illustration 11.24

K12146_C011.indd 265 2/12/15 12:05 PM

 266   ◾    The Fundamental Group of a Circle and Applications

Since a and b are not points on α, they have open neighborhoods U a and U b , respec-
tively, avoiding α. By Lemma 3, we have that a, b ∈∂W . Hence, there are a1 ∈W ∩ U a and
b1 ∈W ∩ U b . Let β be any path in W from a1 to b1 . Then aa1 β b1b is path from a to b
which does not intersect α (Illustration 11.24). This contradicts Lemma 2, and the proof
is completed.

The following is the version of Jordan curve theorem for S 2 in place of  2.

Corollary 4. Let f : S1 → S 2 be an embedding. Then S 2 \ f (S1 ) has two components, and

each of these two components has f (S1 ) as its boundary.

Proof. Since f (S1 ) ≠ S 2 , there is x ∈S 2 \ f (S1 ) . We may assume that x is the north pole
of the sphere. The stereographic projection s sends S 2 \{x } homeomorphically onto  2 ,
and so s( f (S1 )) is a Jordan curve. By Theorem 1, its complement in  2 has two compo-
nents, one bounded, the other unbounded. By Lemma 2, the boundary of each of them
is s( f (S1 )) . Denote that unbounded component by U and the bounded one by V. Then
s −1 (U ) ∪ {x } and s −1 (V ) are the components of S 2 \ f (S1 ) , and f (S1 ) is the boundary of
each of them.

The Jordan curve theorem has been generalized in many directions. Theorem 5 is the
higher dimensional version; it was proven by Brouwer in 1912. Its proof is beyond the scope
of this book.

Theorem 5. (Jordan–Brouwer Separation Theorem) If f : S n → n +1 is an embedding,

then n +1 \ f (S n ) has two components, and f (S n ) is the boundary of each of these two
components.

A consequence is the Invariance of Domain Theorem, stated and used in Section 4.4.
For a proof see, for example, [31], page 172.
Another important generalization of the Jordan curve theorem is the following result.
For a proof, see, for example, [73].

Theorem 6. (Jordan-Schönflies Theorem) Every embedding f : S1 →  2 can be extended

to a homeomorphism ϕ :  2 →  2 .

Theorem 6 does not allow a straightforward generalization to higher dimensions.

Specifically, there exist embeddings f : S 2 →  3 that do not have extensions ϕ :  3 →  3 .
We will exhibit two of them in Section 13.4.

K12146_C011.indd 266 2/12/15 12:05 PM

 11.3 The Jordan Curve Theorem   ◾    267  

Exercises
1. Show that if C is a Jordan curve, then  2 \ C has exactly one unbounded component.
2. Show that if f :  →  2 is an embedding such that f () \ D has at least two compo-
nents for every disk D intersecting f (), then  2 \ f ( ) has two components.
3. Prove that the Jordan–Schoenflies theorem (Theorem 6) implies the Jordan curve
theorem (Theorem 1).
4. Let X = S 2 \{(0,0,1),(0,0, −1)}, where S 2 is the unit sphere. Show that the mapping
f : X → X defined by f ( x , y , z ) = (− x , − y , − z ) is not ambient isotopic within X to id X .
[Hint: transfer the statement of the problem from X to  2 \{(0,0)} by means of the
stereographic projection, then use the Jordan curve theorem.]
5. [This problem is a commentary on the Invariance of Domain Theorem, but that theo-
rem is not needed to solve the problem.]
(a) Find a subset A of  2 , not closed in  2 , and find a continuous, bijective
f : A→  2 such that f is not open.
(b) Show that if K is a compact subset of  2, and if f : K→  2 is continuous and
bijective, then f is open.
(c) Find an example of a continuous and bijective mapping f : K→  2 over closed
subspace K of  2 , such that f is not open.
6. (a) Let A, B, C and D be points on S1 in clockwise (or counterclockwise) order. Show
that if α is a path from A to C in D 2, and if β is a path from B to D in D 2, then
α( I ) ∩β( I ) ≠ ∅.
(b) Use the results given in this section to show that the Möbius band is not embed-
dable in  2 .
7. Show that if f :[0,1] →  2 is an embedding, then  2 \ f ([0,1]) is connected. (That is,
show that arcs do not separate  2 ).
8. Prove that the open Möbius band (= Möbius band without the points on the
boundary) is not homeomorphic to the open cylinder S1 × (0,1).

K12146_C011.indd 267 2/12/15 12:06 PM

This page intentionally left blank
 Chapter 12

Combinatorial Group Theory

I n order to strengthen the π1 ( X , x )-connection between groups and spaces we need more
knowledge of group theory. A natural group theoretical setup in this context is combi-
natorial group theory, and its branch, geometric group theory. In this chapter we give a brief
introduction to this area of group theory.
In addition to the basic group theoretical terminology mentioned at the beginning of
10.4, the reader is expected to be familiar with the following notions: subgroup, normal
subgroup (N is a normal subgroup of G, denoted N  G, if gNg −1 = N for every g ∈ G ),
index of a subgroup in a group, coset, and quotient group or factor group. Other basic
terms will be encountered (and defined) along the way.
There will not be many illustrations in the chapter, since visualizing this part requires
an approach that would take us too far away from the mainstream exposition.

12.1 Group Presentations

A brief historical note: The concept of defining groups through group presentation is due to
Walther von Dyck (1856–1934); the corresponding paper was published in 1882.

For any nonempty set A = {a, b,…}, we will denote the set of formal inverses of the elements
of A by A−1 = {a −1 , b −1 ,…}, and A ± will stand for A ∪ A−1. A word on A ± is a finite sequence
of elements of A ± . For every x ∈ A ± we denote the word xxx…x, n-times, by x n , and we
denote the word ( x −1 )n by x − n. For example, b −2a3bcc −4 denotes a word over the set A ± ,
where A = {a, b, c}. The number of appearances of elements of A ± in a word W is called
the length of the word W, denoted W . For example, b−2a3bcc −4 = 11 . The empty word is
the word of length 0. If W = x1x 2 … xm−1xm , we denote W −1 = xm−1x m−1−1 … x 2−1x1−1, where we
write ( x −1 )−1 = x for every x ∈ A. For every two words U and V, UV is the word we get by
concatenating the word V after the word U.

269

K12146_C012.indd 269 2/9/15 4:28 PM

 270   ◾     Combinatorial Group Theory

Given any set R = {P , Q ,…} of words on A ± , we now proceed to define a group which we
will denote by A ; R , or by a, b,…; P , Q ,… (where, despite what the notation may sug-
gest, we do not put any constraints on the cardinalities of the sets {a, b,…} and {P , Q ,…}).
Step 1. Define an equivalence relation ~ on the set of all words on A ± with respect to R as
follows: U ~ V if there is a sequence of words W1 = U ,W2 ,…,Wk−1 ,Wk = V such that for every
j = 1,2,… , k − 1, Wj+1 is obtained from Wj by applying one of the following operations:
(a) Insert or delete xx −1, x ∈ A ±, somewhere in the word Wj ;
(b) Insert or delete a word X, X ∈{P , Q ,…} ∪ {P −1 , Q −1 ,…} somewhere in the word Wj .
It is straightforward to show that in this way we indeed get an equivalence relation. Denote
the equivalence class of the word U on A ± by [U]. The set of all such equivalence classes
will be denoted by 〈 A ; R 〉.

Step 2. Define an operation over the set 〈 A ; R 〉 by [U ][V] = [UV]. This turns 〈 A ; R 〉 into
a group. We outline an argument justifying this claim; the reader is invited to supply the
details (Exercise 1).
(i) The operation is well defined; this means that if U1 ~ U 2 and V1 ~ V2 , then
[U1 ][V1 ] = [U 2 ][V2 ]. One way to prove this is by induction on the total number of
steps of types (a) and (b) (as defined in Step 1) needed to make U1 ~ U 2 and V1 ~ V2 .
(ii) Associativity, or ([U ][V ])[W ] = [U ]([V ][W ]) : use induction on W .
(iii) Denote the equivalence class of the empty word by []. For every word U,
[U ][] = [][U ] = [U ], and so [] plays the role of the identity, which we will usually
denote by 1.
(iv) [U ][U −1 ] = [U −1 ][U ] = [ ] for every word U. This is easy.
With this we have established that 〈 A ; R 〉 = 〈a, b, … ; P , Q ,…〉 , together with the operation
just defined, is a group. It is generated by A = {a, b,…}.
It is obvious that [P] = 1,[Q] = 1,… in 〈 A ; R 〉. If U and V are two words on A ± such that
[U ] = [V ] is satisfied in 〈 A ; R 〉, then we will say that [U ] = [V ] is a relation in 〈 A ; R 〉, and
that it is a consequence of the relations [P] = 1,[Q] = 1, … . The relations [P] = 1,[Q] = 1, … are
called the defining relations for the group. If [W ] = 1 is a relation in 〈 A ; R 〉, then 〈 A ; R 〉
and 〈 A ; R ∪ {W }〉 are the same group (Exercise 2).
In order to ease the notation, and in the context of the group 〈a, b, … ; P , Q ,…〉 , we
will often write U for [U ]. Thus, for example, we will often write P = 1, Q = 1,…, instead
of [P] = 1,[Q] = 1,… . Along these lines, we will sometimes write 〈a, b, … ; P = 1, Q = 1,…〉
instead of 〈a, b, … ; P , Q , …〉 .

Example 1
Consider the group 〈a ; a 2 〉 : it follows by induction on the length of words that every
word on {a}± is equivalent to either 1 or a. That 1 and a are distinct in the group ­follows

K12146_C012.indd 270 2/9/15 4:28 PM

 12.1 Group Presentations   ◾    271  

from the observation that the operations (a) and (b) of Step 1 do not change the parity
of the length of the word. We conclude that 〈a ; a 2 〉 is a cyclic group of order 2. ☐

In the preceding example we needed a paragraph to state the “obvious”: that 〈a ; a 2 〉 is cyclic of
order 2. To increase the efficiency of our theory we develop it a little further.
Let G be a group, and let f : A → G be any mapping. Define a mapping, also denoted by
f, from the set of all words on A ± into G as follows:
(1) f ( x1α1 x 2αn … xnαn ) = ( f ( x1 )) ( f (x 2 ))α … ( f (xn ))α , xi ∈ A, αi ∈{1, −1} .
α1 2 n

We would like to extend it further and define a homomorphism f : 〈 A ; R 〉 → G by

(2) f ([x1α1 x 2αn … xnαn ]) = ( f ( x1 )) ( f (x 2 ))α … ( f (xn ))α , xi ∈ A, αi ∈{1, −1} .

α1 2 n

However, this last mapping f need not be well defined. The question of when the last f is a
well-defined homomorphism is settled by the following proposition.

Proposition 1. For every group G, the extension f : 〈a, b,… ; P , Q ,…〉 → G of f : A → G ,

as defined in (2), is a well-defined homomorphism if and only if f ( X ) = 1 in G for every
X ∈{P , Q ,…}.

Proof. ⇒ This is clear, since the identity element is mapped to the identity element by any
homomorphisms.

⇐ Suppose [U ] = [V ] in 〈a, b,… ; P , Q ,…〉 . This means that U ~ V with respect to {P , Q ,…},
which in turn means that we can get V from U by means of the operations (a) and (b) as in
Step 1 above. It is easy to see that if one such operation is required to get U ~ V, then f (U ) =
f (V ), hence f ([U ]) = f ([V ]). For example, if U = U1U 2 and V = U1PU 2 , then f (U ) =
f (U1 ) f (U 2 ) and f (V ) = f (U1 ) f ( P ) f (U 2 ), and so, since we have assumed in this part of
the proof that f ( P ) = 1, we have that f (V ) = f (U ) in G. Induction on the number of opera-
tions (a) and (b) needed to get V from U implies that f (U ) = f (V ). This shows f is well
defined. That f is a homomorphism is an immediate consequence of its definition.

Remark. We have used the same f with three different meanings: a mapping f : A → G , a
mapping f : A ± → G defined by (1), and the homeomorphism defined f : 〈 A ; R 〉 → G by
(2). We will almost always deal with the first and the last meanings. Occasionally, we will
want to explicitly distinguish them. Otherwise, we will consistently denote them by the
same letter; the context will resolve any ambiguity.
When the extension 〈 A ; R 〉 → G of f : A → G , is an isomorphism, then we say that
〈 A ; R 〉 is a presentation of the group G with respect to f : A → G . Alternatively, we say
that G is presented by 〈 A ; R 〉 with respect to f. If the set of generators and defining
relations is finite, then we say that F is finitely presented by 〈 A ; R 〉 (with respect to f ).
Every group G can be presented by 〈{x : x ∈ G} ; {xyz −1 : x , y , z ∈ G and xy = z in G }〉
(with respect to the identity mapping G → G ). One of the primary objectives in this

K12146_C012.indd 271 2/9/15 4:28 PM

 272   ◾     Combinatorial Group Theory

context is to find presentations with small (possibly minimal) sets of generators and
defining relations.
We will be particularly interested in finitely presented groups.

Example 2
Recall the cycle notation for permutations: ( x1 x 2 … xn ) denotes the permutation
sending xi to xi+1, for i = 1,2,…, n − 1, sending xn to x1 and fixing the other elements of
the underlying set. In the context of group presentations it is convenient to make an
exception to our longstanding convention and accept that compositions (products)
p1 p2 of permutations are from left to right.
It is easy to confirm that the group G of permutations of the set {1,2,3} is generated
by (1 2 3) and (1 2). We claim that 〈a, b ; a3 , b 2 , aba −2b〉 is a presentation of G with
respect to f : a  (1 2 3), b  (1 2).
Computation shows that (1 2 3)3 = id , (1 2)2 = id and that (1 2 3)(1 2) =
(1 2)(1 2 3)2 confirming that f (a3 ) , f (b 2 ) , and f (aba −2b) are all the identity in G.
Proposition 1 implies that the extension f : 〈a, b ; a3 , b 2 , aba −2b〉 → G is a well defined
epimorphism. It remains to be shown that it is one-to-one. The identity aba −2b = 1
implies that ab = b −1a 2 . It follows that every word in the presentation is equivalent to
some of {anbm : n, m ∈}. Further, the identities a3 = 1 and b 2 = 1 allow us to fur-
ther reduce the class of all equivalence classes of words in the presentation to
{a0 = 1, a, a 2 , b, ab, a 2b}. So, 〈a, b ; a3 , b 2 , aba −2b〉 has no more than 6 elements. Since
we have already established that f : 〈a, b ; a3 , b 2 , aba −2b〉 → G is onto, and since G
has 6 elements, it follows that f must also be one-to-one. ☐

We recapitulate: in order to establish that 〈 A ; R 〉 is a presentation of a group G with

respect to f : A → G we need to
(i) show that f : 〈 A ; R 〉 → G is a well-defined homomorphism,
(ii) show that f is onto, and
(iii) show that f is one-to-one.
Part (i) needs Proposition 1 and (ii) is equivalent to the claim that f(A) generates G. It is part
(iii) that often needs more attention.
Recall the following fact in basic group theory: a group homomorphism f : H → K is one-to-
one if and only if Ker ( f ) = {1}, where Ker( f ) = {x ∈H : f ( x ) = 1 in K } is the kernel of f.
Here is a simple procedure for establishing that a homomorphism f : 〈 A ; R 〉 → G is
one-to-one: find a set of words representing all elements of A ; R and such that for only
one element W of that set of words, f (W ) = 1. Then Ker ( f ) = {1} and so f must be one-to-one.
The case when G is finite is simpler: if (i) and (ii) are checked, then, since an onto
mapping between two finite sets of equal cardinality must also be one to one, the only
thing we need to do to confirm (iii) is to find G many words representing all elements
of A ; R . This is what we did in Example 2. An example where G is infinite follows.

K12146_C012.indd 272 2/9/15 4:28 PM

 12.1 Group Presentations   ◾    273  

Example 3
A symmetry of a subspace A of n is an isometry of n mapping the subspace A onto
itself. Consider the group G of all symmetries (operation is composition) of the Frieze
pattern F shown in Illustration 12.1, regarded as a subspace of  2.
A–2 A–1 A0 A1 A2

B–1 B0 B1 B2

Illustration 12.1 The Frieze pattern F.

It is easy to identify the elements of the group G of the symmetries of F, and then to
show that the group G is generated by the translation t along the vector v (shown in
Illustration 12.1) and the rotation r around A0 through 180°. For example, the rotation
around B1 through 180° is r  t −1. We show that 〈 x , y ; xyx −1 y = 1, x 2 = 1〉 is a presenta-
tion of G under the mapping x → r , y → t .
Checking that rtr −1t = 1 and r 2 = 1 are satisfied in G is straightforward. It remains
to be shown that the extension f : 〈 x , y ; xyx −1 y = 1, x 2 = 1〉 → G of x  r, y  t is
one-to-one. The relation xyx −1 y = 1 can be written as xy = y −1x. This, together with
x 2 = 1 implies that every word in 〈 x , y ; xyx −1 y = 1, x 2 = 1〉 is equivalent to some of
y n x ε , n ∈, ε ∈{0,1}. The only such word that is sent to 1 in G is y 0 x 0 = 1, and so f
is one-to-one. ☐

Example 4
Let G be the subgroup of the group of symmetries of the plane generated by the trans-
lations t u along the vector u = (1,0), and t v along the vector v = (0,1) . We claim that
P = a, b ; aba−1b−1 = 1 is a presentation of G with respect to a  t u , b  t v . It is easy to
check that the extension f : P → G is a well-defined epimorphism. To prove that f is
one-to-one, first observe that aba −1b −1 = 1 tells us that ab = ba , that is, that the group
P is commutative. It follows that the set {anbm : n, m ∈} contains all of the represen-
tatives of the equivalence classes of the words in P . Of all the elements of that set only
1 = a0b0 is mapped to 1 ∈ G , and so f is indeed one-to-one. ☐

There is something deeper going on in Example 4 than what may appear at first sight.
We showed in this example that the group G was isomorphic to  ⊕ , a free commuta-
tive group on two generators. We already know that the fundamental group of the torus
is isomorphic to  ⊕ . On the other hand, the quotient space of the plane  2 obtained
by identifying all orbits { g ( x ) : g ∈ G} of the elements x ∈ 2 under the action of G is
precisely the torus (see Exercise 6 in 4.4). As we will see in Example 3, 16.3, this is not a
coincidence.

K12146_C012.indd 273 2/9/15 4:28 PM

 274   ◾     Combinatorial Group Theory

Exercises
1. Complete the details of Step 2, showing that 〈a, b,… ; P , Q ,…〉 is indeed a group.
2. (a) Show that if [U ] = 1 is a relation in a, b,…; P , Q ,… , then the groups 〈a, b,… ;
P , Q ,…〉 and 〈a, b,… ; P , Q ,… , U 〉 are identical.

(b) Let U be any word on {a, b,…}± and let x ∉{a, b,…}± . Show that the groups
〈a, b,… ; P , Q ,…〉 and 〈a, b,… , x ; P , Q ,… , Ux −1 〉 are isomorphic.
3. Let G be the group of symmetries of the Frieze pattern shown in Illustration 12.2.
First, guess a presentation for G, then prove your guess.

   Illustration 12.2

4. Show that the group of symmetries G of the Frieze pattern F shown in Illustration
12.1, Example 3, is generated by the translation t and the rotation r ′ centered at B0
and through 180°, find a presentation of G on the generating set {t , r ′}, and prove that
it is indeed a presentation of G.
5. Let G be the subgroup of the group of all symmetries of  2 generated by the translation
t v along the vector v = (0,1) and the composition g of the translation along the vector
u = (1,0) followed by the reflection with respect to the x-axis. Guess a presentation for
G, then justify your guess.
 ε k 
6. ([47]) Let G be the group of matrices of the type   over the ring of integers
 0 1 
modulo n, where ε ∈{−1,1}, k ∈{0,1,… , n − 1} (the operation is multiplication of
matrices). Show that 〈a, b ; an , b 2 , abab −1 〉
l
is a presentation of G under the mapping
 1 1   −1 0 
a  , b   .
 0 1   0 1 

7. The group G of all symmetries of the sub-

space of  2 depicted in Illustration 12.3
is generated by the translation tb along the
vector b and the glide reflection g that is the
a
composition of the translation f = t a 2 along
the vector a 2 followed by the reflection with
respect to the vertical line l. Show that the
group G has a presentation 〈a, b ; aba−1b〉 b
under the mapping a  g, b  f. Illustration 12.3

K12146_C012.indd 274 2/9/15 4:28 PM

 12.2 Free Groups, Tietze, Dehn   ◾    275  

8. Show that the group 〈 x , y ; yx −1 y −1 x 2 , y −1 x −1 yxyx 〉 is trivial.

9. Show that the group 〈a, b, c ; a3, b 2 , c 3, aba 2b, a 2c 2ac , bcbc −1 〉 has exactly 18 elements.
(Note: it is easy to show that the group has ≤ 18 elements; showing that the number of
elements is exactly 18 requires a strategy.)
10. Show that 〈a, b ; a 4 , b 2 , (ab)2 〉 is a presentation for the group of symmetries of a
square.
11. Show that the groups 〈a, b, c ; ababab 2 〉 and 〈 x , y ; ∅〉 are isomorphic. [Hint: consider
a  x 4 , b  x −3 , c  y.]
12. Show that if N is a normal subgroup of G = 〈a, b, … ; P , Q , …〉 generated by {x j : j ∈ J },
then 〈a, b, … ; P , Q , … , {x j : j ∈ J }〉 is a presentation of the group G N .

12.2 Free Groups, Tietze, Dehn

A brief historical note: The group-theoretical work of both Heinrich Tietze and Max Dehn
was inspired by Poincaré’s introduction of the fundamental group of a space. Tietze intro-
duced (what were later called) Tietze transformations in a paper published in 1908, and
Dehn stated his three fundamental problems in 1911, in a paper dedicated entirely to group
presentations.

A special and important class of groups is the class of free groups. A group F is free if it is iso-
morphic to a group of type 〈a, b,… ; ∅〉, that is, if it could be presented with the empty set of
defining relations. In this setting, the set {a, b,…} is called the set of free generators of F, or we
say that F is freely generated by that set. The free group that is freely generated by the empty
set is the trivial group. The simplest nontrivial free group is the infinite cyclic group 〈a ; ∅〉.
If a free group is freely generated by the sets A and B, then A = B (Exercise 13). The
cardinality {a, b,… } of the set of free generators is the rank of F.
It follows directly from the definition of presentations that free groups satisfy only the
relations satisfied in all groups, namely those that are consequences of the relations of the
type xx −1 = 1, x ∈{a, b,… }± .
Denote G = 〈a, b,… ; P , Q , …〉 , and F = 〈a, b,… ; ∅〉. Proposition 1 in 12.1 implies that
the identity mapping over {a, b,…} can be extended to a homomorphism f : F → G . Since
f is onto, it follows from the first isomorphism theorem for groups that the group G is
isomorphic to the factor group F Ker ( f ). So, every group is a quotient group of the free
group on the same generating set. The normal subgroup Ker(f) of F is the normal closure
of the set {P , Q ,…} of elements in F (Exercise 1), where the normal closure NC(A) of a
subset A of a group H is the smallest normal subgroup of H containing A as a subset. The
elements of NC(A) are products of conjugates in H of the elements of A (the last claim is
an easy exercise). Hence we have the following:

K12146_C012.indd 275 2/9/15 4:28 PM

 276   ◾     Combinatorial Group Theory

Remark. If G = 〈a, b,… ; P , Q ,…〉, and if F = 〈a, b,… ; ∅〉, then G ≅ F / NC({P , Q ,…}).

A group presented with a nonempty set of defining relations may be free. So, it makes sense
to ask the following questions: if we know that a presentation defines a free group, is there a
procedure that will transform any presentation of that group into a relations-free presentation?
More generally, if two presentations define the same group, is there a procedure to change one
of them into the other without altering (up to isomorphism) the group they define in each step.
The affirmative answer was established by Tietze in 1908 in the following theorem.

Theorem 1. If 〈a1 , a2 ,… ; P1 , P2 , …〉 and 〈b1 , b2 ,… ; Q1 , Q2 , …〉 are two presentations of a

single group, then the presentation 〈b1 , b2 ,… ; Q1 , Q2 , …〉 can be obtained from the pre-
sentation 〈a1 , a2 ,… ; P1 , P2 , …〉 by applying finitely many steps of the following type, where
each step yields a presentation of the same group.

(T1) If Y1 ,Y2 ,… are consequences of X1 , X 2 ,…, then adjoin Y1 ,Y2 ,… in the set of defining
relations of 〈 x1 , x 2 ,… ; X1 , X 2 , …〉 to get 〈 x1 , x 2 ,… ; X1 , X 2 , … , Y1 , Y2 ,… 〉 .
(T2) [Reverse (T1)] If Y1 ,Y2 ,… are consequences of X1 , X 2 ,…, then delete Y1 ,Y2 ,… from
the set of defining relations of 〈 x1 , x 2 ,… ; X1 , X 2 , … , Y1 ,Y2 ,…〉 to get 〈 x1 , x 2 ,… ; X1 , X 2 , …〉 .
(T3) If W1 ( x1 , x 2 ,…),W2 ( x1 , x 2 ,…), … are words on {x1 , x 2 ,…}± , and y1 , y 2 ,… are new
symbols, then change 〈 x1 , x 2 ,… ; X1 , X 2 ,…〉 to 〈x1 , x 2 ,… , y1 , y 2 ,… ; X1 , X 2 , … , y1 =
W1 ( x1 , x 2 ,…), y 2 = W2 ( x1 , x 2 ,…) ,…〉.
(T4) [Reverse (T3)] Change to 〈 x1 , x 2 ,… , y1 , y 2 ,… ; X1 , X 2 , … , y1 = W1 ( x1 , x 2 ,…), y 2 =
W2 ( x1 , x 2 ,…) ,…〉 to 〈 x1 , x 2 ,… ; X1 , X 2 , … 〉.

Remark. We do not assume in the statement of the theorem and in the proof below that
the sets of generators and defining relations are countable, as the notation might suggest.

Proof. (In a nutshell, we use (T1) and (T3) to change 〈a1 , a2 ,… ; P1 , P2 , … 〉 to a presentation
that is symmetric with respect to the two generating sets, then we apply (T2) and (T4) to
arrive at 〈b1 , b2 ,… ; Q1 , Q2 , …〉 .)
Suppose 〈a1 , a2 ,… ; P1 , P2 , … 〉 is a presentation of G with respect to a1  g 1 , a2  g 2 ,…,
and suppose 〈b1 , b2 ,… ; Q1 , Q2 , …〉 is a presentation of the same group with respect to
b1  h1 , b2  h2 ,…. Since the set { g 1 , g 2 ,…} generates G, each of h1 , h2 ,… can be written as
a word on { g 1 , g 2 ,…}± . Suppose h1 = W1 ( g 1 , g 2 ,…), h2 = W2 ( g 1 , g 2 ,…), …. It is easy to check
(see Exercise 3) that 〈a1 , a2 ,… , b1 , b2 ,… ; P1 , P2 , … , b1 = W1 (a1 , a2 ,…), b2 = W2 (a1 , a2 ,…),… 〉
is also a presentation of G under the mapping a1  g 1 , a2  g 2 ,…, b1  h1 , b2  h2 ,…. We
have used (T3).
Each of Q1, Q2,… is a word on {b1 , b2 ,…}± ; to emphasize that, we write Qi = Qi (b1 , b2 ,…).
Since 〈b1 , b2 ,… ; Q1 (b1 , b2 ,…), Q2 (b1 , b2 ,…), …〉 is a presentation of G with respect to b1  h1 ,
b2  h2 ,… , it follows that Q1 (h1 , h2 ,…) = 1, Q2 (h1 , h2 ,…) = 1,… in G. Consequently,
adjoining Q1 , Q2 ,… to the set of defining relations of 〈a1 , a2 ,… , b1 , b2 ,… ; P1 , P2 , … ,
b1 = W1 (a1 , a2 ,…), b2 = W2 (a1 , a2 ,…),…〉 , and geting 〈a1 , a2 ,… , b1 , b2 ,… ; P1 , P2 , … , Q1 , Q2 ,…,

K12146_C012.indd 276 2/9/15 4:29 PM

 12.2 Free Groups, Tietze, Dehn   ◾    277  

b1 = W1 (a1 , a2 ,…), b2 = W2 (a1 , a2 ,…),…〉 as the result, again yields a presentation of G under
a1  g 1 , a2  g 2 ,… , b1  h1 , b2  h2 ,… . We have used (T1) in this step.
Since h1 , h2 ,… also generate G, we can write g 1 = U1 (h1 , h2 ,…), g 2 = U 2 (h1 , h2 ,…), …
for some words U1 ,U 2 ,…. Consequently, a1 = U1 (b1 , b2 ,…), a2 = U 2 (b1 , b2 ,…), … are
true in 〈a1 , a2 ,…, b1 , b2 ,…; P1 , P2 , …, Q1 , Q2 ,…, b1 = W1 (a1 , a2 ,…), b2 = W2 (a1 , a2 ,…),…〉
and adding them as relations will not affect the group. Thus, we get the follow-
ing presentation of the same group with respect to the same mapping a1  g 1 ,
a2  g 2 ,… , b1  h1 , b2  h2 ,…: 〈a1 , a2 ,…, b1 , b2 ,…; P1 , P2 , …, Q1 , Q2 , …, b1 = W1 (a1 , a2 ,…),
b2 = W2 (a1 , a2 ,… ),… , a1 = U1 (b1 , b2 ,… ), a2 = U 2 (b1 , b2 ,…),… 〉 We have used (T3) again.
In the last presentation there is symmetry between {a1 , a2 ,…} and {b1 , b2 ,…}; so trac-
ing the steps backwards (thus employing (T2) and (T4)), and interchanging the roles of
{a1 , a2 ,…} and {b1 , b2 ,…} in the process, leads us to 〈b1 , b2 ,…; Q1 , Q2 , …〉 as a presentation of
G with respect to b1  h1 , b2  h2 ,… .

The transformations (T1)–(T4) are called Tietze transformations.

Theorem 1 can be rephrased as follows.

Theorem 2. The presentations 〈a1 , a2 ,… ; P1 , P2 , … 〉 and 〈b1 , b2 ,… ; Q1 , Q2 , …〉 define iso-

morphic groups if and only if one of them can be obtained from the other by means of
Tietze transformations.
Proof.
⇒ Let f : 〈a1 , a2 ,… ; P1 , P2 , … 〉 → 〈b1 , b2 ,… ; Q1 , Q2 , … 〉 be an isomorphism. Then 〈a1 , a2 ,…;
P1 , P2 , …〉 is a presentation of the group 〈b1 , b2 ,… ; Q1 , Q2 , …〉 with respect to f. On the other
hand 〈b1 , b2 ,… ; Q1 , Q2 , …〉 is certainly a presentation of itself (eith respect to the identity
mapping). The result now follows from Theorem 1.
⇐ This is an immediate consequence of the fact that Tietze transformations do not affect
the isomorphism class of the group, as noted in the proof of Theorem 1.

We will say that two presentations are equivalent if one can be obtained from the other by
means of Tietze transformations.
Corollary 3. 〈a1 , a2 ,… ; P1 , P2 , … 〉 and 〈b1 , b2 ,… ; Q1 , Q2 , …〉 are presentations of a single
group if and only if they are equivalent.

Proof. ⇒ is established in Theorem 1.

⇐ By Theorem 2, equivalent presentations define isomorphic groups. On the other hand, it is
indicated in the proof of Theorem 2 that two isomorphic presentations are presentations of a
single group.

Example 1
In a few chapters the presentations 〈a, b, d ; ab = ba, d 2 = a, da = ad , db = b −1d 〉 and
〈b, d ; dbd −1b〉 will arise as the fundamental group of a single space, using two dif-
ferent methods. By Theorem 1 there must be a sequence of Tietze transformations

K12146_C012.indd 277 2/9/15 4:29 PM

 278   ◾     Combinatorial Group Theory

changing one to the other. The proof of Theorem 1 actually establishes a generic—
but usually rather inefficient—sequence of Tietze transformations. Here is a shorter
sequence of Tietze transformations, leading from one to the other presentation in
this example:

〈a, b, d ; ab = ba, d 2 = a, da = ad , db = b −1d 〉 (T 4)

→
〈d , b ; d 2b = bd 2 , dd 2 = d 2d , db = b −1d 〉 (T
2) 〈d 2 2 −1
→ , b ; d b = bd , db = b d 〉 (T
2)
→
〈d , b ; db = b −1d 〉 = 〈d , b ; dbd −1b〉

In the last step above we have eliminated d 2b = bd 2 ; we can do that since it is

a consequence of db = b −1d, as follows: d 2b = d(db) = d(b −1d ) = (db −1 )d = (bd )d = bd 2
(and note that bd = db −1 is a consequence of db = b −1d ). ☐

Example 2
Consider the group 〈a, b, c ; a3 , b 2 , c 3 , aba 2c , acac 2 〉 . We use Tietze transformations to
show that this group is (isomorphic to) the trivial group. First we use the third relation
to get c −1 = aba 2, and then c = a −2b −1a −1 = aba 2. Hence c = c −1 , which yields c 2 = 1. This
together with c 3 = 1 implies that c = 1. We can then eliminate c using (T1) and (T2)
to get 〈a, b, c ; a3 , b 2 , c 3 , aba 2c , acac 2 〉 → 〈a, b ; a3 , b 2 , aba 2 , a 2 〉 . Since a3 = 1 = a 2 implies
that a = 1, we can remove a in the same way to get 〈a, b ; a3 , b 2 , aba 2 , a 2 〉 → 〈b ; b 2 , b〉 ,
and this is clearly the trivial group. ☐

In Example 2 we managed to show that the presentation defines the trivial group. Could
determining if any finitely presented group is trivial be done algorithmically? The answer
is an emphatic “No!” More importantly, we have arrived at Dehn’s fundamental problems1.
We first state these problems in terms of fundamental groups. Recall that two elements x,
and y in a group G are conjugate if there is an element a ∈ G such that x = aya −1.
(D1)∗ Is there an algorithm to decide if, in a connected space with a known presentation
of its fundamental group, a loop is homotopic to the trivial loop?
(D2)∗ Is there an algorithm to decide if, in a connected space with a known presentation
of its fundamental group, the homotopy class of a loop at the base point is conjugate to the
homotopy class of another loop at the base point?
(D3)∗ Is there an algorithm to decide if two connected spaces with known presentations
of their respective fundamental groups have isomorphic fundamental groups?
A special case of (D3)∗ is the following: Is there an algorithm to decide if a space is simply
connected?
The general, group-theoretical versions of the same three problems follow:

1
Max Dehn is widely regarded as the founder of combinatorial/geometric group theory.

K12146_C012.indd 278 2/9/15 4:29 PM

 12.2 Free Groups, Tietze, Dehn   ◾    279  

(D1) (The Word Problem for Groups) Is there an algorithm to decide if in a given pre-
sentation a word determines the trivial element?
(D2) (The Conjugacy Problem for Groups) Is there an algorithm to decide if in a given
presentation two words are conjugate?
(D3) (The Isomorphism Problem for Groups) Is there an algorithm to decide if two pre-
sentations determine isomorphic groups?

There is a specific and finite presentation for which the word problem (D1) is unsolvable
(Novikov 1954, Boone 1959)1; see, for example, [66], page 362. This answers (D1) in the nega-
tive, and as a consequence, (D2) and (D3) as well (Exercise 8).
The Word Problem is solvable for some specific classes of groups. For example, that is the
case with the class of groups presentable with not more than one defining relation (Wilhelm
Magnus, 1930; see, for example, [45]). This is relevant, since, as we will see, the fundamental
groups of compact, connected 2-manifolds are presentable with one defining relation.

Example 3
We will solve the word problem for the group 〈a, b ; a 2 , b 2 〉. This example will be
eclipsed by Corollary 4 in 12.3.
It is easy to see that every word W on {a, b}± is equivalent to a word of type
U = a εbaba…babδ , where ε, δ ∈ {0,1}. This class of words includes the empty word. The
set of all such words of length 1 is {a,b}; the set of all such words of length 2 is {ab,ba},
etc. The change from W to U can be done effectively by repeatedly reducing all even
exponents of a and b to empty words (a consequence of {a 2 = 1, b 2 = 1} in the group).
Now we show that the only word of type U that is equivalent to the identity in the
group is the empty word. Let P(x) be the ring of all polynomials on the variable x with
coefficients in the field  2 = {0,1}. Let G be the group of all 2 × 2 matrices with entries
  1 0   1 x  
in P(x) and generated by   ,    (the operation is multiplication
  x 1   0 1    1 0 
of matrices). Consider the mapping f : {a, b} → G defined by f (a) =   and
 x 1 
 1 x 
f (b) =   . Denote these two matrices by A and B, respectively. Our goal is
 0 1 
to show that none of ( AB)m, ( AB)mA (m ∈+ ) is the identity matrix.

1
The chain of results leading to the final solution is interesting. Here is a brief summary: In 1936 Turing produced the
notion of (what is now called) a Turing machine. Roughly, a Turing machine consists of a string of squares containing 0 or
1, and a finite program determining what should be done when a particular square is being observed, the options being:
(a) leave or change the entry of the observed square and (b) move forward, move backward, or stay on the same square of
the string of squares. Since Turing machines (i.e., the corresponding finite programs) can be effectively enumerated, the
following problem makes sense (The Halting Problem): Is there a Turing machine such that if given a number n and the
string consisting only of 0-s, it will determine if the Turing machine numbered n will stop or run forever. The standard
diagonal argument (similar to the one used to prove that the set  is not countable) shows that the answer to the Halting
Problem is negative. In 1947 Emil Post (and, independently, A. A. Markov, same year) proved that the Word Problem for
semigroups is unsolvable; the basic idea was to associate semigroups to Turing machines and to rely on the unsolvability
of the Halting Problem. Novikov’s and Boone’s results for groups is based on Post’s solution for semigroups.

K12146_C012.indd 279 2/9/15 4:29 PM

 280   ◾     Combinatorial Group Theory

 1 0   1 0 
Since, as is easy to check, ( f (a)) = A2 =   and ( f (b)) = B = 
2 2 2
,
 0 1   0 1 
it follows from Proposition 1 in 12.1 that f is extendible to a homomorphism
f : 〈a, b ; a 2 , b 2 〉 → G . This homomorphism is onto, since { A, B} generate G.
 1 x 
We compute AB =  2  , and use induction to show that the second entry
 x x + 1 
of the second row in ( AB)m is a polynomial of degree 2m, and that this is the largest

degree of all four polynomial entries of ( AB)m, m ∈+. The case m = 1 is obvious.
 P11 P12 
Inductively, assuming ( AB)k =  2k
 satisfies the inductive assump-
 P21 Q + x 
tions (that is, assuming all P11 , P12 , P21, and Q have degrees less than 2k), we have

 P11 P12   1 x 
( AB)k+1 =  2k
  2 
 P21 Q+x   x x +1 

 P11 + xP12 xP11 + ( x 2 + 1)P12 

= .
 p21 + xQ + x 2 k+1 xP21 + Q + x 2 k + x 2Q + x 2 k+2 
 

It is now easy to confirm that, since the degrees of P11 , P12 , P21, and Q are smaller than
2k, the degree of xP21 + Q + x 2 k + x 2Q + x 2 k + 2 is indeed 2k + 2, and that the other three
polynomial entries of the last matrix have degrees smaller than 2k + 2. Similar argu-
ment shows that the entry in the second row, first column of ( AB)m A is a polynomial
of degree 2m + 1, and it is the largest degree of all polynomial entries of that matrix.
We have proven what we were aiming at: none of ( AB)m, ( AB)m A (m ∈+) is the
identity matrix. Consequently, no word of type (ab)m, (ab)m a (m ∈+) is equivalent to
the empty word in the presentation.
By symmetry, no word of the type (ba)m , (ba)m b (m ∈+) is equivalent in the
presentation to the empty word.
We proved that no word U = a εbaba…babδ defines the identity in the presentation
unless U is the empty word.
±
Summarizing, here is an algorithm to check if a word W on {a, b} is the same as the
identity in the presentation: change each appearance of a −1 and b −1 to a and b, respectively
(using a 2 = 1 and b 2 = 1, respectively), then reduce W using {a 2 = 1, b 2 = 1} and decreas-
ing the length of the word in each step. Stop when no further reductions are possible, and
check if the word obtained is the empty word. W =1 if and only if that happens. ☐

The group 〈a, b ; a 2 , b 2 〉 is the free product of two cyclic groups of order 2: we will define
and briefly cover such groups in the next section, where we will also show that, under some
natural hypotheses, the word problem of such groups is solvable.

K12146_C012.indd 280 2/9/15 4:29 PM

 12.2 Free Groups, Tietze, Dehn   ◾    281  

The isomorphism problem for group presentations is, in general, more difficult than
the word problem. For example, at the time of writing this, the isomorphism problem
for one-relation groups is still open. That is, it is not known if there is an algorithm to
determine if any two presentations, each with one defining relation, are isomorphic. Such
algorithms exist for many classes of one-relation presentations, including for almost all
that we will encounter here.
Tietze’s transformations is one path to follow in order to confirm that two presentations
are isomorphic. As Examples 4 and 5 illustrate, these transformations also come in handy
when our goal is to show that two (groups given by their) presentations are not isomorphic.

Example 4
We show that the free group F2 = 〈a, b ; ∅〉 of rank 2 is not isomorphic to the free group
F3 = 〈 x , y , z ; ∅〉 of rank 3. Suppose otherwise, and let f : F3 → F2 be an isomorphism.
Then f (z ) is a nonidentity element of F2 , and it is easy to deduce directly from the
definition of presentations that G = 〈a, b ; f (z ) = 1〉 is not freely generated by {a,b}. That
G is not freely generated by any set of two generators is an exercise left to the reader.
However, the group H = 〈 x , y , z ; z = 1〉 is a free group on two generators. Consequently
G and H are not isomorphic. On the other hand, it follows from the remark at begin-
ning of this section that G ≅ F2 NC ({ f ( z )}) and H ≅ F3 NC ({ z }). Since f ( NC({z }) = NC({ f (z )}),
a simple group-theoretical exercise implies that G ≅ H , giving a clear contradiction.
(The exercise we referred to is the following: If f : G1 → G2 is an isomorphism, and if N
is a normal subgroup of G1, then f ( N ) is normal in G2 and G1 N ≅ G2 f ( N )). ☐

Similar argument can be used to show that, in general, the rank of a free group is preserved
under isomorphisms.
For tackling more complicated presentations we need to develop a mini-machine.
Let S = { X1 , X 2 ,… , Xn } be a set of symbols, and let W ( X1 , X 2 ,… , Xn ) be any word over
the set S ±. For any group G we will denote the set {W ( g 1 , g 2 ,… , g n ) : g 1 , g 2 ,… , g n ∈G}
by {W ( X1 , X 2 ,… , Xn )}G . For example, if S = { X , Y } and if W ( X , Y ) = XYX −1Y −1, then
{ XYX −1Y −1 }G = { ghg −1h −1 : g , h ∈G}.

Proposition 4. If G = 〈a1 , a2 ,… ; P1 , P2 , …〉 and H = 〈b1 , b2 ,… ; Q1 , Q2 , …〉 are two isomorphic

groups, then for every word W ( X1 , X 2 ,… , Xn ), the groups G1 = 〈a1 , a2 ,… ; P1 , P2 , … , {W ( X1 ,
X 2 ,…, Xn )}G 〉 and H1 = 〈b1 , b2 ,… ; Q1 , Q2 , … ,{W ( X1 , X 2 ,… , Xn )}H 〉 are also isomorphic.

Proof. Let f : G → H be an isomorphism. It is easy to see that f ({W ( X1 , X 2 ,…, Xn )}G ) =

{W ( X1 , X 2 ,… , Xn )}H . It is very easy to show that the subgroup of any group G gener-
ated by the elements {W ( X1 , X 2 ,…, Xn )}G is normal in G. The claim now follows from the
group-theoretical exercise mentioned at the end of Example 4, and from Exercise 2.

Remark. From now on the capital letters X ,Y , X1 , X 2 ,… within words appearing as defin-
ing relations in a presentation have the same meaning as in the setup of Proposition 4.

K12146_C012.indd 281 2/9/15 4:29 PM

 282   ◾     Combinatorial Group Theory

Example 5
Here is an alternative way to show that the free groups F2 and F3 discussed in Example
4 are not isomorphic. Assume they are isomorphic. Then, by Proposition 4, so are
the groups H = 〈a, b ; { XYX −1Y −1 }H 〉 and G = 〈 x , y , z ; { XYX −1Y −1 } G 〉 . Each relation in
{ XYX −1Y −1 }H is a consequence of aba −1b −1; hence, by Tietze, H = 〈a, b ; aba−1b−1 〉 .
Similarly G = 〈 x , y , z ; xyx −1 y −1 , xzx −1z −1 , yzy −1z −1 〉. It is then easy to see that H is a
presentation of the group  ⊕ , and G is a presentation of  ⊕  ⊕  . These two
groups are not isomorphic (by a simple exercise), and we have a contradiction. ☐

The last step in Example 5 is also a consequence of the following basic theorem of group theory,
which will come handy at times later on. A proof can be found in any textbook on basic algebra.
(In the statement below  m stands for the finite group of integers mod(m) under addition.)

Theorem 5. Every finitely generated abelian group G is isomorphic to  m1 ⊕  m2 ⊕… ⊕  mk ⊕

⊕ ⊕ … ⊕  where m1 , m2 ,…, mk are positive integers such that m1 > 1 and m1 | m2 | … | mk .
Moreover, the sequence m1 , m2 ,…, mk (hence, the number k) and the number of infinite cyclic
groups that appear as factors in this direct sum are uniquely determined by the group G.

The group  ⊕  ⊕ … ⊕ , n copies of  , is called free abelian group of rank n.

The following example will be needed in Chapter 13, where we consider the problem of
classification of certain 2-manifolds.

Example 6

Denote Gm = 〈a1 , b1 , a2 , b2 ,… , am , bm ; a1b1a1−1b1−1a2b2a2−1b2−1 … ambmam−1bm−1 〉 , m ≥1, and

Hn = 〈d1 , d2 ,… , dn ; d12d22 … dn2 〉 , n ≥1. We show that for every m and n the group Gm
is not isomorphic to the group Hn . Suppose otherwise. Proposition 4 then imp­
lies that Gm∗ = 〈a1 , b1 , a2 , b2 ,… , am , bm ; a1b1a1−1b1−1a2b2a2−1b2−1 … ambmam−1bm−1 , { XYX −1Y −1 }〉
is iso­morphic to Hn∗ = 〈d1 , d2 ,… , dn ; d12d22 … dn2 , { XYX −1Y −1 }〉 . The group Gm∗ is a free
abelian group of rank 2m (i.e., Gm∗ ≅  ⊕  ⊕ … ⊕, 2m times). Turning our atten-
tion to Hn , introducing c = d1d2 … dn by means of Tietze, changes the given pre-
sentation of Hn to Hn∗ = 〈c , d1 , d2 ,… , dn ; d12d22 … dn2 , c = d1d2 … dn , { XYX −1Y −1 }〉 .
Using Tietze transformations two more times (in one of which we eliminate d1), we
get Hn∗ = 〈c , d2 ,… , dn ; c 2 , { XYX −1Y −1 }〉 . We see that Hn∗ ≅  2 ⊕  ⊕ … ⊕  (n − 1
appearances of the group  ). It now follows from Theorem 5 that Gm∗ and Hn∗ are
not isomorphic. Contradiction! ☐

Exercises
1. Prove that if f : 〈a, b,… ; ∅〉 → 〈a, b,… ; P , Q , …〉 is the extension of the identity map
{a, b,…} → {a, b,…}, then Ker ( f ) = NC({P , Q ,…}).
2. Show that if H is a normal subgroup of G = 〈a, b,… ; P , Q , …〉 generated by {h j : j ∈ J },
then 〈a, b,… ; P , Q , … , {h j : j ∈ J }〉 is a presentation for G H .

K12146_C012.indd 282 2/9/15 4:29 PM

 12.3 Free Products and Free Products with Amalgamation   ◾    283  

3. Let H1 = 〈a1 , a2 ,… ; P1 , P2 ,…〉 be a presentation of a group G with respect to

f1 : a1  g 1 , a2  g 2 ,… . Choose any h ∈G. Since { g 1 , g 2 ,…} generates G,
h = W ( g 1 , g 2 ,…) for some word W on { g 1 , g 2 ,…}± . Choose any b ∉{a1 , a2 ,…}± . Show
that H 2 = 〈a1 , a2 ,… , b ; P1 , P2 ,… , b = W (a1 , a2 ,…)〉 is a presentation of the same group
G with respect to f 2 : a1  g 1 , a2  g 2 ,…, b  h.
4. Use Tietze transformations (and Theorem 2) to show that 〈a, b, c ; b3 , (bc )3 〉 and
〈 x , y , z ; x 3 , y 3 〉 are isomorphic.
5. Use Tietze transformations to transform the group 〈a, b ; a 4 , b7 , ab3 = ba3 〉 to 〈d ; d 2 〉.
Conclude that the order of the former is also 2. (Hint: d = ab.)
6. The groups 〈a, b, c ; a 2 , b3 , c11 , aba 2b 2 , acac 3 〉 and 〈 x ; x 3 〉 are isomorphic. Prove it by
means of Tietze transformations.
7. Prove that the groups < a, b ; ab = ba > and < x , y , z ; xy = yx , xyz = zxy > are not
isomorphic.
8. Starting from the fact that the Word Problem for Groups is unsolvable, prove that
the Conjugacy Problem and the Isomorphism Problem are also unsolvable.
9. Solve the word problem for the following group 〈a, b ; a3 , b3 〉 . (Specifically, determine
in finitely many steps if a word on the generators of the group defines the identity in
the group.) [Hint: Modify the argument in Example 3.]
10. Show that if n and m are positive, distinct integers, Fn is a free group of rank n, and
Fm is a free group of rank m, then Fn ≅ Fm .

11. Show that 〈a, b ; an = bn+1 , XYX −1Y −1 〉 is an infinite cyclic group for every n ≥ 0.
12. Use Tietze transformations to change 〈a, b ; a 2b 2 〉 to 〈 x , y ; xyxy −1 〉 .
13. Generalize Exercise 10: prove that if F is isomorphic to 〈a, b,… ; ∅〉 and to
〈 x , y ,… ; ∅〉 , then {a, b,…} = {x , y ,…} . [Hint: add X 2 and XYX −1Y −1 to the rela-
tions in both presentations.]

12.3 Free Products and Free Products with Amalgamation

A brief historical note: The first results on free products of groups, including Proposition 3,
were obtained by Emil Artin and Otto Schreier (1901–1929) in 1926 or earlier. Free products
with amalgamation were introduced by Schreier in 1927.

We start with an important definition.

Let {a, b,…}± and {x , y ,…}± be disjoint sets. The free product of the groups G1 = 〈a, b,…;
P , Q , …〉 and G2 = 〈 x , y ,…; R , S , …〉 , denoted G1 ∗ G2, is the group G1 = 〈a, b,…, x , y ,… ;

K12146_C012.indd 283 2/9/15 4:30 PM

 284   ◾     Combinatorial Group Theory

P , Q , … , R , S , …〉 . More generally, if A ±j , j ∈ J , are pairwise disjoint sets, then the free

product of the groups G j = 〈 A j ; R j 〉, j ∈ J , is the group ∗ G j = ∪ A j ; ∪ R j . The groups
j∈J j∈J j∈J
G j are factors of ∗ G j .
j ∈J
If the index set J is clear from the context, then we will sometimes write ∗G j instead
of ∗ G j .
j ∈J

Example 1
If G1 = 〈u, v ; u 2 , uv 2 = vu 2 〉 and if G2 = 〈 x , y ; xy = yx 〉 , then G1 ∗ G2 = 〈u, v , x , y ; u 2 ,
uv 2 = vu 2 , xy = yx 〉.
The group 〈a, b ; a 2 , b 2 〉 , introduced in Example 3, 12.2, is the free product
〈a ; a 2 〉∗〈b ; b 2 〉 .
Every free group 〈 A ; ∅〉 is the free product ∗ 〈a ; ∅〉 of A -many infinite cyclic
a∈ A
groups. ☐
In the next proposition we assert that the groups G j survive intact within ∗ G j .
j ∈J

Proposition 1. Let G be the free product of the groups G j = 〈 A j ; R j 〉 , j ∈ J . For every

fixed k ∈ J , the mapping ik : Gk → ∗ G j , determined by ik (a) = a for every a ∈ Ak , is a well
j ∈J
defined monomorphism of groups.

Proof. Proposition 1 in 12.1 implies that ik is a well-defined homomorphism. The same

argument implies that the mappings pk : ∗ G j → Gk , determined by pk (a) = a for every
j∈ J
a ∈ Ak, and by pk ( x ) = 1 for every x in the set ∪ A j of the remaining generators of ∗ G j , is
j∈J j∈J
j≠k
also a well-defined homomorphism. Since pk  ik = idGk it follows that ik is one-to-one.

In view of Proposition 1 we may consider G j’s to be subgroups of their free product ∗ G j ,

j ∈J
and i j -s to be inclusions.
The following statement links free products with a universal property of homomorphisms.
Free products are sometimes defined in terms of this universal property.

Proposition 2. Let G be a group, let G j , j ∈ J , be a set of subgroups of G such that Gi ∩ Gk = {1}

for every i , k ∈ J , i ≠ k. Further, let i j : G j → G denote the inclusions, and let G be generated by
∪ G j. Then G is the free product of G j , j ∈ J , if and only if for every group H and every set
j∈ J
of homomorphism f j : G j → H, there is a unique homomorphism f : G → H making the
following diagram commutative for every j ∈ J .
ij
Gj G

fj f

H Illustration 12.4

K12146_C012.indd 284 2/9/15 4:30 PM

 12.3 Free Products and Free Products with Amalgamation   ◾    285  

Proof. ⇒ Let G j = 〈 A j ; R j 〉 and G = ∗ G j . Given the homomorphisms f j : G j → H ,

j ∈J
j ∈ J , let f : G → H be determined by f ( x ) = f j ( x ) for every x ∈ A j , and for every j ∈ J. Then
it is visible that f is a well-defined homomorphism and that the above diagram commutes.
The commutativity of the diagram implies that f ( x ) must be f j ( x ) for every x ∈ A j , every
j ∈ J ; hence f is unique.
⇐ Let G j, j ∈ J , be subgroups of G as in the statement of the proposition, and let i j : G j → G
be the inclusions. Suppose that for every H and homomorphisms f j there is a unique
homomorphism f such that the above diagram commutes for every j ∈ J . In particular
this is true for H = ∗ G j , where each f j is the inclusion established in Proposition 1 (see
j ∈J
Illustration 12.5).
ij
Gj G

fj f

∗ Gj
j J

Illustration 12.5

The proof will be completed once we establish that the homomorphism f is an isomorphism.
It is onto since each generator g k ∈Gk ⊂ ∗ G j is f (ik ( g k )). Finally, f is one-to one, since
j ∈J
g  f = idG , where g : ∗ G j → G is the homomorphism defined over the set ∪ A j of generators
j∈ J j∈J
of ∗ G j by g ( x ) = i j ( x ) for every x ∈ A j .
j ∈J

A sequence g 1 , g 2 , … , g n of elements in ∗ G j is reduced if either n = 0 (and the

j ∈J
sequence is empty) or n > 0, each g k is a non-identity element in some G jk , and no two
consecutive g k , g k+1 are in the same group G j . It is obvious that every element of ∗ G j
j ∈J
can be written as g 1 g 2 … g n where g 1 , g 2 , … , g n is a reduced sequence; the identity
element of the group is obtained when n = 0. More is true:

Proposition 3. For each element x of G = ∗ G j there is a unique n and a unique reduced

j ∈J
sequence g 1 , g 2 , … , g n of elements ∪ G j such that x = g 1 g 2 … g n in G.
j∈J

Proof 1. For every j ∈ J and every a ∈G j , a ≠1, we associate a permutation pa of the set of
all reduced sequences as follows:

 (a, g 1 , g 2 , … , g n ) if a ∈G j1 , g 1 ∈G j2 , j1 ≠ j2

pa ( g 1 , g 2 , … , g n ) =  (ag1 , g 2 , … , g n ) if a ∈G j , g 1 ∈G j , ag1 ≠ 1
 ( g 2 , … , gn ) if a ∈G j , g 1 ∈G j , ag 1 = 1.


1
Artin and van der Warden; see [45], page 176.

K12146_C012.indd 285 2/9/15 4:30 PM

 286   ◾     Combinatorial Group Theory

(In the last case, ( g 2 , … , g n ) is the empty sequence if n =1.) We set pa to be the identity
permutation if a =1 in G j . The mapping pa is indeed a permutation for every a ∈∪ G j ,
j∈J
since pa−1 is its inverse. It is easy to confirm that for every a, b ∈G j , pab = pa  pb , so that for
every j ∈ J , ϕ j : a  pa is a homomorphism from G j into the group P of all permutations of
the set of reduced sequences of elements in G. By Proposition 2, we have a homomorphism
ϕ : G → P making the corresponding diagrams (with ϕ j-s in place of f j -s) commutative.
It follows from the construction of φ that if g 1 , g 2 , … , g n is a reduced sequence, then
ϕ( g 1 g 2 … g n ) = ϕ j1 ( g 1 )  ϕ j2 ( g 2 ) … ϕ jn ( g n ) = pg1  pg 2  …  pgn. Observe that the last com-
position sends the empty sequence to ( g 1 , g 2 , … , g n ). Consequently, products g 1 g 2 … g n
corresponding to different reduced sequences are sent by the homomorphism φ to dif-
ferent elements of the group P, and hence, since φ is well defined, two different reduced
sequences give rise to different elements in G.

Corollary 4. If the word problem is solvable for each G j , j ∈ J , then it is solvable for ∗ G j .
j ∈J

Proof. The obvious algorithm (Exercise 1) expresses each x ∈ ∗ G j as g 1 g 2 … g n , where

j ∈J
( g 1 , g 2 , … , g n ) is reduced. If n ≥ 2, it follows from Proposition 3 that x ≠1 in ∗ G j . If n =1,
j ∈J
we can use the solvability of the factor groups to check if x = 1.

For example, since 〈a ; a 2 〉 is finite, it follows from Corollary 4 that the group 〈a, b; a 2 , b 2 〉
has a solvable word problem. Compare this with the work done in Example 3, 12.2.

The following proposition is, in a way, a converse of Proposition 3.

Proposition 5. Let G be a group and let {G j : j ∈ J } be a family of subgroups of G such

that ∪ G j generates G, and such that G j ∩ Gk = {1} whenever j ≠ k. Suppose further that
j∈J
for every two sequences g 1 , g 2 , … , g n and h1 , h2 , … , hm which are reduced with respect to
{G j : j ∈ J } , g 1 g 2  g n = h1h2 hm in G implies that the two sequences are equal. Then G is
isomorphic to the free product ∗ G j .
j∈ J

Proof. Consider the mapping φ : ∗ G j → G determined by φ( g ) = g for every g ∈∪ G j .

j ∈J j∈J
It follows plainly that ϕ is well defined; it is onto since ∪ G j generates G. In order to show
j∈J
it is one-to-one, assume φ( g ) = φ(h), for some g , h ∈ ∗ G j. By Proposition 3 we can write
j ∈J
g = g 1 g 2  g n and h = h1h2 hm where the sequences g 1 , g 2 , … , g n and h 1 , h2 , … , hm are
reduced in ∗ G j . According to our definition of ϕ, φ( g ) = g 1 g 2  g n and φ(h) = h1h2 hm,
j ∈J
viewed as products of the generators ∪ G j of G. Our assumptions in the statement of this
j∈J
proposition, together with φ( g ) = φ(h), force m = n and g k = hk in G jk . This implies that
g = h in ∗ G j .
j ∈J

In the last part of our brief encounter with combinatorial group theory we will
generalize the concept of free product.

K12146_C012.indd 286 2/9/15 4:30 PM

 12.3 Free Products and Free Products with Amalgamation   ◾    287  

Let G1 = 〈a, b,… ; P , Q , …〉 and G2 = 〈 x , y ,… ; R , S , …〉 be two groups such that {a, b,…}±
and {x , y ,…}± are disjoint, let H be a group and let φ1 : H → G1 and φ2 : H → G2 be homo-
morphisms. The free product of the groups G1 and G2, amalgamating the subgroups
φ1 ( H ) and φ2 ( H ) , denoted G1 ∗ G2 , is the group 〈a, b, … , x , y , … ; P , Q , … , R , S , … ,
H
{φ1 (h) = φ2 (h) : h ∈ H }〉 .
Before we generalize to amalgamated products of more than two groups, we pause for a
moment to explain in plain words what we are doing. In the amalgamated product of two
groups G1 and G2 we first take G1 ∗ G2, then we identify, element by element, the two images
φ1 ( H ) and φ2 ( H ) .
Let G j = 〈 A j ; R j 〉, j ∈ J , be groups with pairwise disjoint sets A ±j , let {H jk : j , k ∈ J }
be a class of groups, and let φ jjk : H jk → G j , φkjk : H jk → Gk be homomorphisms for each
j , k ∈ J , j ≠ k. The free product of the groups G j , j ∈ J , amalgamating the subgroups
φ jjk( H jk ), φ kjk ( H jk ), j , k ∈ J , denoted ∗ G j, is the group ∪ A j ; ∪ R j ∪ ∪ {φ jjk (h) =
H jk j ∈J j ∈J j , k ∈J
φkjk (h) : h ∈ H jk } . In this context the groups G j are again called factors, and the groups
φ jjk( H jk ) , φ kjk ( H jk ) are amalgamated subgroups.
If the groups H jk are all equal to a group H, then we will write ∗ G j instead of ∗ G j. In
H H jk
order to avoid even more cluttered notation, we have not included the homomorphisms
φ jjk: H jk → G j , φkjk : H jk → Gk in the notation ∗ G j . Almost all of the time φ jjk ’s and φkjk ’s
H jk
will be clear from the context.
If B jk is a generating set for the group H jk , j , k ∈ J , then we can reduce the set of rela-
tions {φ jjk(h) = φ kjk (h) : h ∈ H jk } in the last presentation given above to its subset
{φ jjk (b) = φ kjk (b) : b ∈ B jk }, without affecting the group defined by the presentation. This is true
by Tietze’s Theorem 1 in 12.2, since the assumption that φ jjk and φkjk are homomorphisms
implies that if φ jjk(b) = φ kjk (b) holds for every b ∈ B jk , then φ jjk(h) = φ kjk (h) for every h ∈H jk .
Summarizing, if B jk is a set of generators for H jk , j , k ∈ J , then we can write more efficiently:
∗ G j = ∪ A j ; ∪ R j ∪ ∪ {φ jjk (b) = φkjk (b) : b ∈ B jk } .
H jk j ∈J j ∈J j , k ∈J

If the groups H jk are trivial, then, obviously, ∗ G j = ∗G j , so that free product with
H jk
amalgamation indeed generalizes the concept of a free product. Similarly, if each φ jjk , φkjk
is trivial (sending each element to the identity element of the corresponding G j ), then again
∗ G j = ∗G j .
H jk

Example 2
Let G1 = 〈a ; a3 〉 , G2 = 〈b ; b 4 〉 , H = 〈c ; ∅〉 , φ1 (c ) = a 2 , φ2 (c ) = b3. Then G1 ∗ G2 = 〈a, b ;
H
a3 , b 4 , a 2 = b3 〉 . Since 1 = a6 = (a 2 )3 = (b3 )3 = b9 = b8b = b , the relation b =1 is a conse-
quence of the relations in the presentation. This easily implies that a =1, and so G1 ∗ G2 is
H
the trivial group. ☐

K12146_C012.indd 287 2/9/15 4:30 PM

 288   ◾     Combinatorial Group Theory

Example 3
Let G1 = 〈a1 , a2 ; a1a2 = a2a1 〉, G2 = 〈b1 , b2 ; b1b2 = b2b1 〉, G3 = 〈c1 , c2 ; c1c2 = c2c1 〉, H =
〈h ; ∅〉, φ112 (h) = a2 , φ12 2
(h) = b1 , φ 223 (h) = b2 , φ 323 (h) = c1 , φ331 (h) = c2 φ131 (h) = a1. Then
∗ G j = 〈a1 , a2 , b1 , b2 , c1 , c2 ; a1a2 = a2a1 , b1b2 = b2b1 , c1c2 = c2c1 , a2 = b1 , b2 = c1 , c2 = a1 〉.
H
We eliminate b1 , b2 and c2 using Tietze transformations to get: H∗ G j = 〈a1 , a2 , c2 ;
a1a2 = a2a1 , a2c1 = c1a2 , c1a1 = a1c1 〉 , which is isomorphic to  ⊕  ⊕  . ☐

We say that G1 ∗ G2 is a proper free product of the groups G1 and G2 amalgamating φ1 ( H )

H
and φ2 ( H ) if φ1 ( H ) and φ2 ( H ) are proper nontrivial subgroups of G1 and G2 , respec-
tively. Example 3 implies that a proper free product with amalgamation can be a non-
trivial abelian group. (Compare this claim with Exercises 3(a) and 10.)
Remark. In the remainder of this section we will discuss a few properties of the free
products with amalgamation. We will not invoke or need these properties in the rest of
the theory that we will develop.
We see from Example 2 that in free products with amalgamation both factor groups
could be annihilated. The culprits for this “misbehavior” are the homomorphisms φ1 and
φ2 : neither is one-to-one in this example.
When the homomorphisms φ1 and φ2 are not one-to-one, then the group G1 ∗ G2 could
H
behave rather pathologically. For example, it is easy to construct a finitely presented G1 ∗ G2
H
with unsolvable word problem, while at the same time the associated presentations of G1 and
G2 have solvable word problems: Let N = 〈 A ; R 〉 be a finitely presented group with unsolvable
word problem; let G1 = 〈 A ; ∅〉, G2 = 〈z ; ∅〉 and H = 〈{hr : r ∈ R } ; ∅〉. Define φ1 : H → G1 by
φ1 (hr ) = r , and set φ2 : H → G2 to be the trivial homomorphism. Both G1 and G2 are free, and so
the word problem in these two groups is solvable (Corollary 4). However, G1 ∗ G2 = 〈 A ∪ {z } ; R 〉
H
does not have a solvable word problem, because N does not have a solvable word problem.
If, however, the homomorphisms φ j , j ∈ {1,2}, are all monomorphisms, then the word
problem in G1 ∗ G2 is solvable if the associated presentations of G1 and G2 have solvable word
H
problems. Exercise 11 contains the steps needed to prove that claim.
Stipulating that all φ jjk’s and φkjk’s (as in the general definition) are monomorphisms does
not guarantee that the homomorphisms G j → ∗ G j, j ∈ J , induced by the inclusion of
H jk
generators are also monomorphisms (Exercise 12). However, in the case of the free prod-
uct with amalgamation G1 ∗ G2 involving only two groups, the corresponding statement is
H
true, and the groups G1 and G2 survive within G1 ∗ G2 . We give the formal statement below,
H
but we do not provide a proof (which can be found in, say, [47], starting on page 199).

Proposition 6. If ϕ1 : H → G1 and ϕ 2 : H → G2 are both monomorphisms, then the homo-

morphisms in1 : G1 → G1 ∗ G2 and in2 : G2 → G1 ∗ G2 induced by the respective inclusions of
H H
the generators are both monomorphisms.
The following proposition generalizes Proposition 2. It is not in the mainstream of our
theory and we will not use it. We give it here since it is sometimes used as a route to the
Seifert-van Kampen theorem (next chapter) bypassing group theory.

K12146_C012.indd 288 2/9/15 4:31 PM

 12.3 Free Products and Free Products with Amalgamation   ◾    289  

Proposition 7. Let G j , j ∈ J , be subgroups of a group G such that G j ∩ Gk = {1} when j ≠ k ,

and such that ∪ G j generates G. Further, let {H jk : j , k ∈ J } be a set of groups, and let
j∈J
j k
φ : H jk → G j , φ jk : H jk → Gk , j , k ∈ J , be group homomorphisms. The group G is the free
jk
product of the groups G j, j ∈ J , amalgamating the subgroups φ jjk( H jk ), φ kjk ( H jk ), j , k ∈ J , if
and only if for every group K and every set {ψ j : j ∈ J } of homomorphisms making the dia-
gram in Illustration 12.6 commutative for every j , k ∈ J , there is a unique homomorphism
f : G → K such that, for every j , k ∈ J the diagram in Illustration 12.7 commutes (where
i j , ik are the inclusions).

φjkj Gj
ψj G
Hjk φ j ij
Hjk K jk ik f
k Gj ψj
k φjk
φjk ψk K
Gk Gk ψk

Illustration 12.6 Illustration 12.7

Proof. Exercise 13.

Exercises
1. Use induction to give a precise proof of the claim that every x ∈ ∗ G j is equal to some
j ∈J
g 1 g 2 … g n , where ( g 1 , g 2 , … , g n ) is a reduced sequence.
2. Let G be the free product of at least two non-trivial groups. Show that the center
Z (G ) = {x ∈ G : xy = yx for every y ∈ G} of G must be trivial.
3. (a) Show that no non-trivial abelian group is the free product of two non-trivial groups.
(b) Show that the group 〈a, b ; W 〉 , where W is any cyclically reduced word on both a
and b not equal to 1 in 〈a, b ; ∅〉 , is not a free product of two non-trivial factors.
4. Show that 〈b, c ; b3 , cb 2cb 2 〉 is a free product of two nontrivial groups.
5. Show that the group 〈a ; an 〉∗ 〈b ; bm 〉 is isomorphic to the group 〈 x ; x p 〉∗ 〈 y ; y q 〉 if
and only if {n, m} = { p, q}.
6. Let φ1 : H → G1 × H , φ2 : H → G2 × H be group homomorphisms defined by,
φ1 (h) = (1G1 , h), φ2 (h) = (1G2 , h) for every h ∈ H , respectively. Show that
(G1 × H ) ∗(G2 × H ) ≅ (G1 ∗ G2 ) × H.
H

7. The braid group on three strings has a presentation a, b ; aba = bab . Show that it is
a non-trivial free product with amalgamation. [Hint: start by introducing c = ab via
Tietze transformations.]
8. Consider the group G1 ∗ G2 where the amalgamation is with respect to the homo-
H
morphisms φ j : H → G j , j ∈ {1,2}. Prove that if the mappings i j : G j → G1 ∗ G2 ,

K12146_C012.indd 289 2/9/15 4:31 PM

 290   ◾     Combinatorial Group Theory

i ∈{1,2}, defined by i j ( x ) = x , for every x ∈G j , are monomorphisms, then both

φ1 and φ2 are monomorphisms.
9. Let H, G1, and G2 be groups, let G3 be the trivial group, and let φ1 : H → G1 and
φ2 : H → G2 be two homomorphisms. Recalling that NC(A), the normal closure of a
subset A of a group G, is the subgroup of G generated by all conjugates of the elements
in A, show that ( G1 NC (φ1 ( H )) ) ∗ ( G2 NC (φ2 ( H )) ) is isomorphic to the free product of the groups
G j , j ∈ {1,2,3}, amalgamated along H.
10. Let G = G1 ∗ G2 be a proper free product of the groups G1 and G2 amalgamating φ1 ( H )
H
and φ2 ( H ). Suppose that NC(φ1 ( H )) ≠ G1 and NC(φ2 ( H )) ≠ G2. Prove that the center
Z (G ) = {x ∈ G : xy = yx for every y ∈ G} of G must be trivial. Deduce that no abelian
group is a proper free product of two groups with amalgamation.
11. Let G1 = a, b,… ; P , Q , … , G2 = 〈 x , y ,… ; R , S , …〉 have solvable word problems, let H
be a group, and let φ1 : H → G1 and φ2 : H → G2 be monomorphisms.
(i) Show that each element x ∈ G can be written as ha1a2 … an (n ≥ 0), where
h ∈ φ1 ( H ), a j ∈(G1 \ φ1 ( H )) ∪ (G2 \ φ2 ( H )) for every j, where no two consecutive
a j , a j+1 belong to the same G1, G2, and where n is minimal with the above properties.
(ii) Show that an application of one of the operations (a) and (b) of Step 1 in the
definition of the equivalence relation on the set of words in presentations
(described in Section 12.1) does not affect n.
(iii) Deduce that if n > 0 then the product ha1a2 … an (as in part (a)) is not equal to the
identity in G = G1 ∗ G2 .
H

(iv) Prove that the word problem for G is solvable. (Note: only this part requires the
assumption that φ1 : H → G1 and φ2 : H → G2 are monomorphisms.)

12. Let G1 = 〈a1 , a2 ; ∅〉, G2 = 〈b1 , b2 ; ∅〉, G3 = 〈c1 , c2 ; ∅〉 , and H = H12 = H13 = H 23 = 〈d1 ,
d2 ; ∅〉 . Find monomorphisms ϕijk : Hij → Gk , i , j ∈{1,2,3}, i < j , k ∈{i , j}, such that the
corresponding free product with amalgamation ∗ G j satisfies the following: the
H
homomorphism G1 → ∗ G j determined by a1  a1, a2  a2 is not one-to-one.
H

13. Prove Proposition 7.

14. Find two groups G1 and G2 such that G1 × G2 is not isomorphic to any G1 ∗ G2 .
H

K12146_C012.indd 290 2/9/15 4:31 PM

 Chapter 13

Seifert–van Kampen
Theorem and Applications

I n the previous chapter we laid the groundwork for further expansion of the homotopy
theory. The Seifert–van Kampen theorem is a major step forward; it provides a way to
express the fundamental group of a space in terms of the fundamental groups of open
subspaces forming a cover. We will give a proof of one of the generalizations of the original
statement of the theorem, then illustrate the power of the theorem through many examples.

13.1 Seifert–van Kampen Theorem

A brief historical note: In 1932 Egbert van Kampen (1908–1942) published a proof of Theorem
1 below. At about the same time, and independently, Herbert Seifert (1907–1996) produced a
proof of Theorem 2 when the space X belongs to a class of CW complexes.

In order to clearly see the basic idea of the Seifert–van Kampen theorem (which we will
refer to as the SvK theorem), we first state it in its simplest version. It is the version that we
will use most often in our applications.

Theorem 1 (Seifert–van Kampen) Let {U1 ,U 2 } be an open cover of a path connected space
X, such that U1 , U 2 and U1 ∩ U 2 are non-empty and path connected. Let x ∈U1 ∩ U 2 , and
denote the inclusions U1 ∩ U 2 → U1 and U1 ∩ U 2 → U 2 by i1 and i2 , respectively. Then
π1 ( X , x ) = π1 (U1 , x ) ∗ π1 (U 2 , x ), amalgamating the subgroups (i1 )∗ (π1 (U1 ∩ U 2 , x ))
π1 (U1 ∩U 2 , x )
and (i2 )∗ (π1 (U1 ∩ U 2 , x )) .

291

K12146_C013.indd 291 2/12/15 12:11 PM

 292   ◾    Seifert–van Kampen Theorem and Applications

U2 U1
x

U1 U2
Illustration 13.1

With only slightly more effort we will prove a more general variant of the SvK theorem:

Theorem 2. (Seifert–van Kampen). Let X be a path connected space, let x ∈ X, and let
U = {U j : j ∈ J } be an open cover of X, such that each U j is path connected, and such that
each intersection of four or fewer members of U is path connected and contains x. Denote
the inclusion U j ∩ U k ⊂ U j by i jjk . Then π1 ( X , x ) is a free product of the groups π1 (U j , x ) ,
( ) ( )
amalgamating the subgroups i jkj (π1 (U j ∩ U k , x )) and i kjk (π1 (U j ∩ U k , x )).
∗ ∗

Remark 1. The source of the mysterious stipulation regarding the intersection of four or
fewer members of U is the fact that in a rectangular net every vertex is a corner of four or
fewer rectangles. The precise connection will be established in the proof.

Remark 2. It follows from the definition of free products with amalgamation that Theorem 1
can be reformulated as follows: if π1 (U j , x ) is presented by < Y j ; R j > , j ∈ J , and if Z jk is a
generating set for π1 (U j ∩ U k , x ), j , k ∈ J , then the group π1 ( X , x ) is presented by

j ∈J j ∈J
( ) ∗
( )
< ∪ Y j ; ∪ R j , { i jkj ( z ) = i jkj ( z ) : z ∈ Z jk , j , k ∈ J } > .
∗ (∗)

Remark 3. The proliferation of indexes in the statement of Theorem 2 was virtually

forced upon us by the nature of the statement. Nevertheless, it is rather distracting and
could make the proof that follows unnecessarily saturated. We will, therefore, sacrifice
some of the precision, trusting that the domains and the ranges of the mappings we will
encounter will be clear from the context. Thus, we will use i j : U j ∩ U k → U j instead of
i jkj : U j ∩ U k → U j , and use ik : U j ∩ U k → U k instead of i kjk : U j ∩ U k → U k . Further, we will
use i ∗j : π1 (U j ∩ U k , x ) → π1 (U j , x ) and ik∗ : π1 (U j ∩ U k , x ) → π1 (U k , x ) for the homomor-
( ) ( )
phisms induced by i j and ik , respectively (in place of i jkj and i kjk , respectively). With this,
∗ ∗
and in the light of Remark 2, we can reformulate the Seifert–Van Kampen theorem as follows.

Theorem 2, formulation 2. (Seifert–van Kampen). Let X be a path connected space, let

x ∈ X , and let U = {U j : j ∈ J } be an open cover of X, such that each U j is path connected,
and such that each intersection of four or fewer members of U is path connected and
contains x. Denote each inclusion U j ∩ U k ⊂ U j by i j , and the induced homomorphism by
i ∗j : π1 (U j ∩ U k , x ) → π1 (U j , x ). Let π1 (U j , x ) be presented by < Y j ; R j > , and let Z jk be a
generating set for π1 (U j ∩ U k , x ). Then < ∪ Y j ; ∪ R j , {i ∗j ( z ) = ik∗ ( z ) : z ∈ Z jk , j , k ∈ J } > is a
j ∈J j ∈J
presentation for the group π1 ( X , x ).

K12146_C013.indd 292 2/12/15 12:11 PM

 13.1 Seifert–van Kampen Theorem   ◾    293  

Proof. Let g : < ∪ Y j ; ∪ R j , {i ∗j ( z ) = ik∗ ( z ) : z ∈ Z jk , j , k ∈ J } > → π1 ( X , x ) be the extension

j ∈J j ∈J
of the following mapping defined over the generating set: the homotopy class in Y j of a
loop α at x in U j is sent to the homotopy class of the same loop considered as a loop in
X. We will show that g is a isomorphism by proving that it is well defined (Step 1), onto
(Step 2), and one-to-one (Step 3). The first two steps will be easy, the third will require
some work.
Step 1 (g is well defined). By Proposition 1 in 12.1, we need to check that for every
r ∈ R j , j ∈ J , g (r ) is the identity element in π1 ( X , x ), and that g (i ∗j ( z )) = g (ik∗ ( z )) for
every z ∈ Z jk , j , k ∈ J .
Let r ∈ R j for some j ∈ J . This means that r is the homotopy class of a loop at x that
is entirely in U j , and that this loop is homotopic to the constant loop at x by means of a
homotopy ranging entirely within U j . This clearly implies that r is homotopic to the con-
stant loop at x within X, and so g(r) is the trivial element in π1 ( X , x ) .
The assumption that i ∗j ( z ) = ik∗ ( z ) means that the homotopy class in U j of a loop α at
x in U j ∩ U k representing z is the same as the homotopy class in U k of the same loop. It
follows plainly (from the fact that we deal with a single loop) that α defines a unique homo-
topy class in π1 ( X , x ) , so that g (i ∗j ( z )) = g (ik∗ ( z )).
Step 2 (g is onto). Let f be a loop at x, so that [f] is an element of π1 ( X , x ). Let f = f1 f 2 … fn
be the Lebesgue subdivision of the path f relative to the cover U . This means that each of
f ([ti , ti+1 ]), i ∈{0,1, … , n − 1} ( t0 = 0 ), is in some U i ∈U . As in the first part of the proof
of Theorem 1 in 11.1, if f ([ti , ti+1 ]) and f ([ti +1 , ti + 2 ]) are both in a single U i , then we join
[ti , ti+1 ] and [ti +1 , ti + 2 ] into a single interval. We do that until there are no such cases.
After relabeling the intervals we may assume that f ([ti , ti+1 ]) and f ([ti +1 , ti + 2 ]) are always
in different U i ’s. This implies that for every i ∈{0,1, … , n − 1}, f (ti ) ∈U i −1 ∩ U i . Since
U i −1 ∩ U i is a path connected set containing x, there is a path δi from f (ti ) to x within
U i −1 ∩ U i . Since f = f1 f 2 … fn is homotopic to ( f1 δ1 )(δ1−1 f 2δ 2 )(δ 2 −1 f3 δ 3 ) … (δn −1 fn ) ,
and since each of the paths in parentheses is a loop at x entirely within an open set
belonging to our cover, it follows that [ f ] = [( f1 δ1 )(δ1−1 f 2 δ 2 )(δ 2 −1 f3 δ 3 ) … (δn−1 fn )] =
[ f1 δ1 ][δ1−1 f 2 δ 2 ][δ 2 −1 f3 δ 3 ] … [δn−1 fn ], where each of the homotopy classes to the right-
hand side of the equality is an element of π1 (U i , x ), some U i ∈U . This shows that the
union of the sets π1 (U j , x ) , j ∈ J , generates π1 ( X , x ) , and so g is onto.

Step 3. It remains to be shown that g is one-to-one, or, equivalently, that the kernel of g
consists only of the identity element in the group defined by the presentation. This is the
crucial part of the proof. The basic idea is to construct a step-by-step homotopy between
a loop in X and the constant loop at x, so that in each step the range of the homotopy is
confined within a single member of the cover U.
Take a product [ f1 ][ f 2 ] … [ fn ] of elements in {Y j : j ∈ J }. Consider [ f1 ], [ f 2 ], … , [ fn ] to
be elements in π1 ( X , x ) (via g), so that f = f1 f 2 … fn is a loop at x in X. Assuming this loop
is homotopic to the trivial loop in X at x, we need to show that [ f1 ][ f 2 ] … [ fn ] is the identity
element in the presentation, or, more explicitly, that it is equivalent to the empty word in
the above presentation.

K12146_C013.indd 293 2/12/15 12:11 PM

 294   ◾    Seifert–van Kampen Theorem and Applications

Since f = f1 f 2 … fn is homotopic to the constant loop at x, there is a homotopy

F : I × I → X relative to {0,1} from f to the constant loop at x (so, {0} × I and {1} × I are
mapped to x). Since I × I is compact, we can split it into a net so that each rectangle in this
net is mapped into some U j in the cover of X. Further subdivision of this net will ensure
that the endpoints of the domains of each of f1 , f 2 ,… , fn are on the vertical lines of the
net (see Illustration 13.2).

... ... ...

... ... ...

... ... ...

g(i+1)j

Rij hij
...

...

...

...
...

...

hi(j–1)
gij

... ... ...

g21 ... g2k ... ... g2m

h10
R11 h11 . . .
R1k h1k
... ... R1m h1m
g11 g1k ...... g1m Illustration 13.2 I × I subdivided into rect-
f1
angles, each of which is mapped into some
fn
U j via F.

Focusing at I ×{0} we see (in Illustration 13.2) that f1 f 2 … fn = g 11 g 12 … g 1m, where g 1 j is

the restriction of F to the j-th segment of I ×{0} in the net.
More generally, the restriction of F to any side of any small rectangle in the net defines
a path in X. We denote the restrictions of F to the edges of the rectangle Rij as follows: the
restriction of F to the bottom edge is g ij , and to the top edge is g (i +1) j ; the restriction of F
to the left edge is hi( j −1) , and to the right edge it is hij (see Illustration 13.2). The range of
any of these paths is in one member of U . They need not be loops at x in X. We will replace
them with loops in the next step.

fi fn

Illustration 13.3 We add closed line segments at some vertices in the net.

Assuming that the rectangular net is in a horizontal plane, we attach a vertical closed
line segment I v at most of the vertices v of the net, as shown in Illustration 13.3. A vertex v
at the bottom of a vertical line segment is mapped by F to a point in the intersection of the
open sets in U containing the images of the rectangles adjacent to this vertex. There are

K12146_C013.indd 294 2/12/15 12:11 PM

 13.1 Seifert–van Kampen Theorem   ◾    295  

either two such adjacent rectangles (when the vertex is at the bottom or at the top edge of
I × I ) or four of them (for the other vertices). So, bearing in mind that the associated open
sets in U need not be distinct, F (v) is contained in one, two, three, or four open sets in U .
Since we have assumed that the open sets in U are path connected, and that the intersec-
tions of two, three, or four open sets in U are path connected, and since each of these
intersections contains the base point x, it follows that there is a path τ F (v ) from F (v) to x
within the intersection of these open sets. We extend F on the vertical line segments I v as
follows: identifying I v with I, the restriction of the extension of F on I v is τ F (v ) . Take this
path to be the constant at x if it happens that F (v) coincides with x. Denote this extension
of F by F̂ . It is clear that F̂ is continuous (gluing lemma).
For convenience, we modify the notation for τ F (v ): if sij denotes the vertical line seg-
ment at the right bottom corner of the rectangle Rij , 1 ≤ j < m, then we denote the path
F̂ sij by λ ij . In Illustration 13.4 we show this notation pertaining to the bottom edge of
I × I extended by the vertical segments.

λ11 λ1k λ1(m–2) λ1(m–1)

g11 g1k g1(m–1) g1m

Illustration 13.4

−1 −1
Observe that f = f1 f 2 … fn = g 11 g 12 … g 1m  g 11 λ11 λ11 g 12 λ12 λ12 … g 1(m−1)λ1(m−1)λ1−(1m−1) g 1m =
(
( g 11λ11 ) λ11 −1
) ( )( )
g 12 λ12 … λ1−(1m−2) g 1(m−1)λ1(m−1) λ1−(1m−1) g 1m , where the parentheses in the last step
are only for emphasis. Each path in the parentheses is, by construction, a loop at x entirely
within one member of the starting cover U of X.
Focus for a moment on f1 : recall that it is a loop within some U j ∈U defin-
ing an element in the generator set Y j of the presentation < Y j ; R j > of
π1 (U j , x ) . As we see in Illustration 13.2, f1 = g 11 g 12 … g 1k , and so g 11 , g 12 ,… , g 1k
are also paths in U j . We achi­eved above that f1 = g 11 g 12 … g 1k is homotopic to
( ) ( )( )
( g11λ11 ) λ11−1 g12λ12 … λ1−(1k−2) g1(k−1)λ1(k−1) λ1−(1k−1) g1k , where the paths λij in this product
are also in U j , and where the corresponding homotopy shrinks each λ1i λ1−i1 to a point
within U j . Moreover, all the paths in the parentheses are loops in U j . All this implies
that f1 and ( g 11λ11 ) λ11 ( −1
) ( )( )
g 12 λ12 … λ1−(1k−2) g 1( k−1)λ1( k−1) λ1−(1k−1) g 1k define the same element
in π1 (U j , x ). Consequently, [ f1 ] = ( g 11λ11 ) λ11
 ( −1
) ( )( )
g 12 λ12 … λ1−(1k−2) g 1( k−1)λ1( k−1) λ1−(1k−1) g 1k 

in < Yj ; R j >, hence this is also true in the presentation (∗) of π1 ( X , x ). Since
 (
( g 11λ11 ) λ11 −1
) ( )
g 12 λ12 … λ1−(1k−1) g 1k  =  g 11λ11   λ11 g 12 λ12  …  λ1( k−1) g 1k  , we showed that

−1 −1

−1
 f1  =  g 11λ11   λ11 g 12 λ12  …  λ1−(1k−1) g 1k  in the presentation (∗).
Similar argument holds for f 2 , f3 ,… , fn ; so we can state that [ f ] = [ g 11λ11 ]
 λ11−1
g 12 λ12 …  λ1−(1m− 2) g 1(m−1)λ1(m−1)   λ1−(1m−1) g 1m  holds in the presentation (∗). Note again
that the homotopy classes to the right are associated to loops at x, each of them entirely
within one member of U , so that they also represent elements in the presentation (∗).

K12146_C013.indd 295 2/12/15 12:11 PM

 296   ◾    Seifert–van Kampen Theorem and Applications

Now it suffices to show that the product of the loops in the right-hand side of the last
equality is equal to the identity element in the presentation (∗). We will prove it by chang-
ing that product in steps, each involving one small rectangle in our net.
In the first step we consider the square we have denoted by R11, together with the two
vertical line segments we have attached to it.

λ21
g21
t = 0.8
R11 t = 0.6
h10 t = 0.4
h11
t = 0.2 λ11

g11
Illustration 13.5

A simple homotopy (within an open set U j containing R11) between the loops h10 g 11λ11 and
−1
( g 21λ 21 )(λ 21 h11λ11 ) is indicated in Illustration 13.5. Recall that h10 is the constant loop at
x. Since this happens within a single U j , we have that [ g 11λ11 ] = [ g 21λ 21 ]  λ 21
−1
h11λ11  in the
−1
presentation of our theorem. Hence, we can change  g 11λ11   λ11 g 12 λ12  …  λ1−(1m−1) g 1m 
to ( g 21λ 21 )(λ 21−1h11λ11 )  λ11−1 g 12 λ12  …  λ −11(m−1) g 1m  without changing the element in
the presentation. This completes the first step of our induction.
...

...

...

...

...

...

Rij

... ... ...

... ... ...

R11 ... R1k ... ... R1m

g11 g1k g1m
......
f1 fn
Illustration 13.6

Assume we have established that [ f ] and the product of homotopy classes of the
loops associated to the edges of the rectangles along the thick line segments shown
in Illustration 13.6 (extended by the vertical segment which we do not show in this
figure), define the same element in the presentation (∗). Roughly speaking, we now
change the path along the two thick edges of Rij (left and bottom) to a path over the
other two edges of Rij (top and right). The argument is similar to the one used in the
first step of the induction.

K12146_C013.indd 296 2/12/15 12:11 PM

 13.1 Seifert–van Kampen Theorem   ◾    297  

λ(i+1)(j–1) λ(i+1)j
gi(j+1)
t = 0.8
t = 0.6
hi(j–1) t = 0.4
Rij
hij
λi(j–1) t = 0.2 λij

gij

Illustration 13.7

(
A homotopy between the (products of) loops λ(−i1+1)( j −1)hi( j −1) λ i( j −1) λ i−(1j −1) g ij λ ij )( )
( )( )
and λ(−i1+1)( j −1) g i( j +1) λ(i +1) j λ(−i1+1) j hij λ ij is indicated in Illustration 13.7. Since the
loops in parentheses are at x, and since these loops, as well as the homotopy shown
in Illustration 13.7, are entirely within an open set U j containing Rij , we have that
 λ −1 h λ  λ g λ  =  λ −1 g λ   λ −1 h λ  in the presentation
 (i +1)( j −1) i( j −1) i( j −1)   i( j −1) ij ij   (i+1)( j −1) i( j +1) (i+1) j   (i+1) j ij ij 
(∗). Thus, we can extend our homotopy over the rectangle Rij without changing the ele-
ment in the presentation (∗).
It follows by induction that the element in (∗) defined by f is equal to the element in the
presentation defined by a path that goes along the left edge of I × I , then the top edge, and
then along the right edge. The last path is the constant loop at x; hence, f is the same as the
trivial element in the presentation (∗), completing the proof of the theorem.

Question: Where were the identities {i ∗j ( z ) = ik∗ ( z ) : z ∈ Z jk , j , k ∈ J } in the presentation

given in the statement of the theorem used in the above proof?
Answer: Consider, for example, the loop z = λ (−i1+1)( j −1) g i ( j +1)λ (i+1) j used in the last step
of the proof. There, we treat it is as a loop in the open set Uj containing the image of the
rectangle Rij . When we deal with the rectangle R(i+1) j that is just above Rij , we will take
this loop to represent an element in the fundamental group of the open set U k contain-
ing R(i+1) j . However z is in the intersection of Uj and Uk, and consequently, since i ∗j ( z )
and ik∗ ( z ) have been equated in the presentation, we can replace the homotopy class z in
Uj (used when we deal with the rectangle Rij ) with the homotopy class of the same loop
(used when we deal with the rectangle R(i+1) j ), without changing the element in the pre-
sentation. So, i ∗j ( z ) = ik∗ ( z ) was utilized when switching from the loops made of the edges
of one rectangle to the loops made of the edges of the adjacent rectangles.
We remind the reader that in what follows we will often refer to Theorem 2 as the SvK
theorem.

Corollary 3. Suppose the conditions of Theorem 2 are satisfied. If each U j ∩ U k is simply

connected, then π1 ( X , x ) is the free product of the fundamental groups π1 (U j , x ), j ∈ J .

Proof. This follows immediately from Theorem 2.

The following statement is a slight generalization of Corollary 3.

K12146_C013.indd 297 2/12/15 12:11 PM

 298   ◾    Seifert–van Kampen Theorem and Applications

Corollary 4. Let the conditions of Theorem 2 be satisfied and let the inclusion
U j ∩ U k → U j and U j ∩ U k → U k induce the trivial homomorphisms of the fundamental
groups for every j , k ∈ J . Then π1 ( X , x ) is the free product of the fundamental groups
π1 (U j , x ) , j ∈ J .

Proof. Using the notation of Theorem 2, the assumptions imply that both i ∗j ( z ) and ik∗ ( z )
are the identity element, and so i ∗j ( z ) = ik∗ ( z ) can be removed by Tietze transformations,
ending up with the free product of the groups π1 (U j , x ), j ∈ J .

Corollary 5. Suppose the conditions of Theorem 2 are satisfied. If each of the groups
π1 (U j , x ) , j ∈ J , is trivial, then so is π1 ( X , x ) .

Proof. This is evident.

Here is a specific application of the SvK theorem. We will see many more in the coming
sections.

Corollary 6. The sphere S n, n ≥ 2, is simply connected.

Proof. Let A denote the north pole (0,0,…,0,1) of S n and let B denote the south pole
(0, 0, … , 0, − 1) of S n . Take U1 = S n \{ A} and U 2 = S n \{B}. The north-pole and south-
pole stereographic projections show (respectively) that the open sets U1 and U 2 are both
homeomorphic to n , and so they are both simply connected. Since U1 ∩ U 2 is path con-
nected for n ≥ 2, the conditions of the SvK theorem are fulfilled and Corollary 5 applies.
So, S n is also simply connected.

Exercises
1. Show that there does not exist a cover of the torus T 2 = S1 × S1 consisting of two
open subsets U and V such that both U and V are not simply connected and such that
U ∩ V is simply connected.
2. Let U1 , U 2 be an open cover of a path connected space X, and let V1 , V2 be an
open cover of a path connected space Y, such that both covers satisfy the condi-
tions of the SvK theorem (Theorem 1). Show that U1 × V1 , U1 × V2 , U 2 × V1 , U 2 × V2
is an open cover of X × Y satisfying the conditions of Theorem 1.
3. Consider the statement of Theorem 2. Show through a counterexample that if the sets
in the cover U = {U j : j ∈ J } are assumed to be closed (while keeping the other condi-
tions of Theorem 2 unchanged), then the conclusion of Theorem 2 fails.
4. Cover a sphere with two open sets as indicated in Illustration 13.8 and its caption. Use
the SvK theorem to find a presentation for the fundamental group of the sphere, then
use Tietze transformations to reduce that presentation to the trivial one.

K12146_C013.indd 298 2/12/15 12:11 PM

 13.2 Seifert–van Kampen Theorem: Examples   ◾    299  

Illustration 13.8 U is made of the red (dark

gray) two quarters, the yellow (light) strips
included, and V is made of the green (light
gray) two quarters, the yellow (light) strips
included. The intersection consists of the yel-
low (light) thin strips along two perpendicu-
lar large circles.

5. (a) Show that for every x ∈ 3 , the space  3 \{x } is simply connected.
(b) Show that  2 is not homeomorphic to  3.
(c) Show that for every x1 , x 2 ,… , xn ∈ 3 , the space  3 \{x1 , x 2 ,… , xn } is simply
connected.
6. Show that  3 \ D 3 is simply connected (recall that D 3 = {( x , y , z ) ∈ : x 2 + y 2 + z 2 ≤ 1}).
k
7. (a) Denote Sn2 = {( x , y , z ) : ( x − 2n)2 + y 2 + z 2 = 1} , n =1, 2, … , k. Show that ∪ S n2
n=0
(considered as a subspace of  3 ) is simply connected.
(b) Use the SvK theorem to find a presentation of the fundamental group of a neck-
lace of n (touching) spheres (see Illustration 13.9).

   Illustration 13.9

13.2 Seifert–van Kampen Theorem: Examples

A brief historical note: Herbert Seifert stated and proved his version of the SvK theorem in
his dissertation in 1930, when he was only 23 years old. His dissertation was about compact
3-manifolds, and he needed the theorem to compute the fundamental groups of some spaces
that emerged in his work.

K12146_C013.indd 299 2/12/15 12:12 PM

 300   ◾    Seifert–van Kampen Theorem and Applications

The Seifert–van Kampen Theorem substantially expands the class of spaces whose funda-
mental group can be relatively easily determined.

Example 1
Consider the space X depicted in Illustration 13.10. It is obtained by drilling a
Y-shaped tunnel within a ball (open or closed, it does not matter). We use the SvK
theorem directly to compute the fundamental group of X (there is a somewhat shorter
computation, as indicated in Exercise 4). The open sets U1 and U 2 are depicted in
Illustrations 13.11 and 13.12. They form a cover of X. The intersection U1 ∩ U 2 is
shown in Illustration 13.13. We chose the base point x for all of the spaces mentioned
here to be in this intersection.

Illustration 13.10 The space X. Illustration 13.11 The subspace U1.

Illustration 13.12 The subspace U 2 . Illustration 13.13 The subspace U1 ∩ U 2 .

Since, as we see, U1 ∩ U 2 is simply connected, it follows from Corollary 3 in 13.1 that

π1 ( X , x ) = π1 (U1 , x ) ∗π1 (U 2 , x ). The spaces U1 and U 2 are homeomorphic. Further, both

K12146_C013.indd 300 2/12/15 12:12 PM

 13.2 Seifert–van Kampen Theorem: Examples   ◾    301  

of them are homotopically equivalent to the space we get by drilling a tube through
a ball, which in turn is homotopically equivalent to a circle. So, π1 ( X , x ) =  ∗  .
Specifically π1 ( X , x ) = [α],[β] ; ∅ , where α is a loop at x that winds once around
the branch of the Y-shaped tunnel entirely in U1 , and β is a loop at x that winds once
around the branch of the Y-shaped tunnel entirely in U 2 . ☐

Example 2
Consider the space X depicted in Illustration 13.14 (two tunnels) and Illustration
13.15 (n tunnels): it is obtained by drilling tunnels out of the hollow ball
{( x , y , z ) ∈ 3 : 1.5 ≤ x 2 + y 2 + z 2 ≤ 2}. In the case when n = 2, we cover the space with
the open sets U and V, where U encompasses a bit more than the right half of the
space X, and V a bit more than the left half of X, so that the intersection U ∩ V is a
washer-like space.
Since every loop in U ∩ V is contractible in each of U and V, it follows that the inclu-
sion U ∩ V → X induces the trivial homomorphism π1 (U ∩ V , x ) → π1 ( X , x ). Hence,
Corollary 4 in 13.1 applies, and consequently, π1 ( X , x ) is the free product of the fun-
damental groups of U and V. Each of these two open sets is homotopically equivalent
to a ball with a tunnel along (say) a diagonal drilled out of it, which in turn is homo-
topically equivalent to a circle.

Illustration 13.14 A hollow ball with two tunnels. Illustration 13.15 A hollow ball with n tunnels.

Consequently, π1 (U , x ) = < a ; ∅ > , π1 (V , x ) = < b ; ∅ > , and π1 ( X , x ) = < a, b ; ∅ >

is a free group of rank 2.
In the case of n tunnels (Illustration 13.15) the fundamental group of that space
is free of rank n. This follows by simple induction, since Corollary 4 of 13.1 applies
if we choose a cover with two open sets, one encompassing one tunnel, the other the
remaining n −1 tunnels. ☐

The following is a simple but important example.

K12146_C013.indd 301 2/12/15 12:12 PM

 302   ◾    Seifert–van Kampen Theorem and Applications

Example 3: A Bouquet of Circles

Recall that we have defined a bouquet X to be a 1-dimensional CW-complex, whose
0-skeleton consists of a single point (denoted by x below). We cover X with open sets
as shown in the middle figure of Illustration 13.16, where we depict a typical open set
in a cover {U j : j ∈ J } of X. We also show the intersection of any two distinct open
sets in the cover (the right-most figure in Illustration 13.16).

Uj

Uj ∩Uk

x x x

Illustration 13.16 From left to right: the space X, one typical member of the open cover, and
the intersection of any two such open sets.

Clearly π1 (U j ∩ U k , x ) = 1 for every j , k ∈ J . Since each U j is homotopically equiva-

lent to a circle, we have that π1 (U j , x ) = 〈a j ; ∅〉 . It follows from Corollary 3 of 13.1
that π1 ( X , x ) = 〈{a j : j ∈ J } ; ∅〉. ☐

In the following example we contrast the CW-complex in Example 3 with the corre-
sponding subspace of  3.

Example 4: A Bouquet of Circles—Subspace Topology

Let X be the space consisting of infinitely many circles C j , j ∈ J , of radius 1, where
C j is obtained by rotating the circle C = {(0, y , z ) ∈ 3 : ( y − 1)2 + z 2 = 1} around the
z-axis through j radians (0 ≤ j < 2π ). Hence, X is a space of circles sharing a com-
mon point x = (0, 0, 0). We consider it as a subspace of  3 . If J is finite, then the
argument in Example 3 applies without changes, since the subspace topology and the
CW-complex topology over X coincide. However, if J is infinite, and if X is a closed
subset of  3 , then the topologies are different, and, in particular, at least one of the
sets U j described in Example 3 is not open in this space X; this claim follows from
the fact that the space X is compact (hence, Bolzano–Weierstrass). Thereby, the argu-
ment used in Example 3 does not apply. Nevertheless, we show in this example that
the fundamental group of X is also a free group.
Denote the set of points in the circles C j antipodal to the origin by A, and let
{Bk : k ∈ K } be the set of all path components of the set A. For each k ∈ K denote
Dk = ∪ C j . Then X is the union of the spaces Dk , k ∈ K , where the intersection
C j ∩Bk ≠∅
of any two Dk1 and Dk2 is the singleton {x = (0, 0, 0)} (see Illustration 13.17).

K12146_C013.indd 302 2/12/15 12:12 PM

 13.2 Seifert–van Kampen Theorem: Examples   ◾    303  

Dj
Dk
Illustration 13.17 A set Bs could be an arc
(as is Bk in the figure), or it could be a point
Bj Bk
( B j in the figure), in which case the corre-
sponding D j is a circle.

We prove that π1 ( X , x ) is a free group of rank K . We will use Propositions 3 and 5,

Section 12.3, as well as the finite case of Example 3 above.
Observe first that each Dk is homotopically equivalent to a circle, and so
π1 (Dk , x ) = 〈α k ; ∅〉 , where α k is the homotopy class of a loop at x that winds once
around Dk .
Next, we show that every [α ] ∈π1 ( X , x ) can be expressed uniquely as a
product β1β 2 …βn , where β1 , β 2 , … , βn is a sequence reduced with respect to
{π1 (Dk , x ) : k ∈K } (the terminology comes from 12.3). Since the interval I is com-
pact, there is a small open neighborhood U x of x such that α( I ) ∩ (Ck \ {U x }) ≠ ∅
only for finitely many k ∈{k1 , k2 ,… , km } ⊂ K . Following the method used in
m
Example 3, the fundamental group of the subspace ∪ Dks is freely generated
s =1
by {ak1 , ak2 ,… , akm } . Consequently [α ] can be expressed uniquely in terms of
m
{ak1 , ak2 ,… , akm } , such that the corresponding sequence of elements in ∪ π1 (Dks , x )
s =1
is reduced (Proposition 3 in 12.3). This element [α ] does not allow other expres-
t
sion in terms of the elements in some ∪ π1 (D pr , x ) such that the associated
r =1
sequence is reduced with respect to {π1 (D pr , x ) : r = 1,2,…, t }, for otherwise these
two products would correspond to different sequences reduced with respect to
{π1 (Dks , x ): s = 1,2,…, m} ∪ {π1 (D pr , x ): r = 1,2,…, t }, contradicting Example 3 and
Proposition 3 in 12.3.
We have proven that [α] can be expressed uniquely as a product β1β 2 … βn , where
β1 , β 2 , … , βn is a sequence reduced with respect to {π1 (Dk , x ) : k ∈ K } . It follows from
Proposition 5 in 12.3 that π1 ( X , x ) is the free product of {π1 (Dk , x ) : k ∈ K } , which, in
this case, implies that π1 ( X , x ) ≅ 〈{ak : k ∈ K } ; ∅〉 . ☐

∞
{ }
Remark. The subspace ∪ ( x , y ) ∈ 2 : ( x − n1 ) + y 2 = n12 of  2 , called Hawaiian Earrings,
n=1
2

also consists of circles sharing a common point. However, its fundamental group is not
free. This is not an easy result (see, for example, [70]). The pathological behavior of the
Hawaiian Earrings will also be noted in Example 3, Section 15.4 (see Illustration 15.28
there).

K12146_C013.indd 303 2/12/15 12:12 PM

 304   ◾    Seifert–van Kampen Theorem and Applications

Example 5: Torus, Again

We regard the torus T 2 as the quotient space described in Illustration 4.22,
Section 4.4. We cover it with two open sets as shown in Illustration 13.18.
b

Illustration 13.18 U1 is inside the dashed

a a
rectangles, and U2 is outside the smallest
x U1 U2 (closed) rectangle (so, both U1 and U2 are
open). The intersection is as indicated in the
figure. The base point x is in the intersection,
b
as required.

U1 is simply connected, and U1 ∩ U 2 is homotopically equivalent to a circle. Hence,

π1 (U1 , x ) is trivial and π1 (U1 ∩ U 2 , x ) is infinite cyclic generated by the homotopy class
of the loop α shown in Illustration 13.20. A bouquet of two circles (extended by a
line segment to reach the base point, as shown in the right-most figure of Illustration
13.19) is a deformation retract of U 2 (Illustration 13.19). It follows by Example 3 that
π1 (U 2 , x ) = 〈[γaγ −1 ],[γbγ −1 ] ; ∅〉 , where γ is the path shown at the right-most figure of
Illustration 13.19.
U2 b b b b
a
γ x
a a a a a a a a=
x x x x b

b b b b

Illustration 13.19 A bouquet of two circles as a deformation retract of U2.

Denoting the inclusions by i1 : U1 ∩ U 2 → U1 and i2 : U1 ∩ U 2 → U 2 , we have that

(i1 )∗ ([α]) must be the identity in π1 (U1 , x ) , while we see in Illustration 13.20 that
(i2 )∗ ([α]) = [γbab −1a −1γ −1 ] = [γbγ −1 ][γaγ −1 ][γb −1γ −1 ][γa −1γ −1 ] . (Do not forget that the
edges labeled by a and b are identified in the torus, forcing all four of the vertices of
the large rectangle to be identified into a single point.)

b b b

a a a a a a
x α x x
α
γ γ α
b b b

Illustration 13.20 The loop α is homotopic within U2 to the loop ( γbγ −1 )( γaγ −1 )( γb −1 γ −1 )( γa −1 γ −1 ).

K12146_C013.indd 304 2/12/15 12:12 PM

 13.2 Seifert–van Kampen Theorem: Examples   ◾    305  

Denoting for brevity x = [γaγ −1 ] and y = [γbγ −1 ], the SvK theorem gives (again)
〈 x , y ; yxy −1 x −1 = 1〉 as a presentation for π1 (T 2 , x ). ☐

Example 6: Graphs
Recall that a graph is a CW-complex of dimension 1; it is obtained by identifying the
endpoints of closed intervals (1-cells) to the points of a discrete space (made of 0-cells
or isolated points).
Recall that the 0-cells in a graph X are called vertices, while the 1-cells in X give
rise to the edges in the graph. If the 1-cell I is attached to the 0-skeleton of X by
identifying 0 and 1 with u and v, respectively, then we say that the edge in X that
corresponds to that 1-cell has u and v as its end-vertices, and that it is an edge from
u to v. It is, of course, possible that these two end-vertices coincide in X. We will
denote the set of edges in a graph X by E(X), and the set of vertices in X by V(X).
Since we are interested in fundamental groups, we will consider only path con-
nected graphs.
A tree is a path connected and simply connected graph. A maximal tree in a graph
X is a tree in X that is maximal in X (i.e., for which there is no tree of X that properly
contains it as a subcomplex.)

Proposition 1. Every tree in a graph X is a subtree of a maximal tree in X.

Proof. The set of all trees in X is partially ordered with respect to ≤, where T1 ≤ T2
means that the tree T1 is a subcomplex of the tree T2 . Intending to use Zorn’s Lemma
(Section 1.3), we consider a chain of trees T1 ≤ T2 ≤ … ≤ Tn ≤ … in X. It is obvious that
T = ∪+ Tj is a subcomplex of X of dimension 1. Since each Tj is path connected, it
j∈
follows from Proposition 3 in 6.3 that T is also path connected.
If it is not simply connected, then there is a loop α at a vertex in V(T) that is not
homotopic to the trivial (constant) loop at that vertex. If α(I) does not contain the
entire interior of a 1-cell in T, then α is homotopic to a loop that does not intersect
the interior of that 1-cell. Hence, we may assume that if a 1-cell in T intersects α(I),
then it is a subset of α(I). Consider the set S of all 1-cells in X that are subsets of α(I).
Suppose S is infinite. Let U s be a 1-cell s ∈ S, possibly extended slightly at the two
end-vertices to get an open set. Then the family {U s : s ∈S} is an open cover of α(I)
containing no finite subcover. This is not possible since α(I) is compact. So, S must
be finite. Since every 1-cell in S is in ∪+ Tj , it must be in Tk , for some k ∈+ . This
j∈
forces α(I) to be in Tk . Since Tk is simply connected, α is homotopic to the trivial
loop within Tk , and so, it is homotopic to the trivial loop in T, contradicting our ini-
tial assumption. Hence, T is a tree.
We have proven that every chain of trees in X is bounded by a tree. The statement
of the proposition now follows from Zorn’s Lemma (Section 1.3).

K12146_C013.indd 305 2/12/15 12:12 PM

 306   ◾    Seifert–van Kampen Theorem and Applications

Let u and v be two vertices in a graph X. A geodesic between u and v in the graph X is
a minimal tree in X containing u and v. Every geodesic has finitely many vertices and
is homeomorphic to I =[0, 1] (Exercise 6).
Every maximal tree T in X must contain all of the vertices of X. Otherwise, if G is a
geodesic joining a vertex out of T with a vertex in T, then T ∪ G will be a tree properly
containing T (contradicting the maximality of T).

Proposition 2. Let X be a path connected graph containing exactly one edge that does
not belong to a maximal tree T in X. Then π1 ( X , x ) is an infinite cyclic group.

Proof. Let e be the edge in X out of a maximal tree T, and let G be a geodesic in T join-
ing the end-vertices of e. Choose two open sets U and V covering X as in Illustration
13.21 (the figure to the right) and its caption.

T G
x

G
x
G
x

Illustration 13.21 The graph X is shown to the left, with the geodesic G shown thicker. We
split it into two open sets: U is shown up right, V is down right. The intersection U ∩ V is G
extended by a few whiskers made out of small parts of the adjacent edges.

Then T and G are deformation retracts of U and U ∩ V , respectively, and so both

U and U ∩ V are simply connected. Observe that G ∪ e is a deformation retract of
V. Since G is homeomorphic to a proper closed interval in  , it follows that G ∪ e
is homeomorphic to S1, and so π1 (V , x ) is infinite cyclic. It now follows from the
simple version of the SvK theorem described in Corollary 3 of 13.1 that π1 ( X , x ) is
an infinite cyclic group.

Proposition 3. For every path connected graph X, π1 ( X , x ) is a free group freely

generated by a set of cardinality E( X ) − E(T ) , where T is any maximal tree in X.

Proof. Choose an open cover of X as in Illustration 13.22 and its caption.

K12146_C013.indd 306 2/12/15 12:12 PM

 13.2 Seifert–van Kampen Theorem: Examples   ◾    307  

X Uj
T T
x x
ei

ek

ej ej

Illustration 13.22 For every e j ∈ E( X )\ E(T ), the open set U j is the union of an open set slightly
larger than T and the edge e j (compare X and U j around the edge ek ).

For every j , k ∈ J , T is a deformation retract of U j ∩ U k , and so U j ∩ U k

is simply connected. The result now follows from Proposition 2 (above) and
Corollary 4 of 13.1.
☐

Exercises
1. Show that {( x , y , z , w ): x 2 + y 2 + z 2 + w 2 ≥ 1} is simply connected.

2. Find two homotopic continuous mappings f , g : ( X , x ) → (Y , y ) such that the

induced homomorphisms f∗ : π1 ( X , x ) → π1 (Y , y ) and g ∗ : π1 ( X , x ) → π1 (Y , y ) are
not equal. (Compare with Theorem 6 in 10.4.)
3. Find a presentation of the fundamental group of the punctured torus (S1 × S1 )\ {( x , y )}.
4. Visualize a homotopy equivalence between the space X in Example 1 (Illustration 13.10)
and the bouquet of two circles. Conclude again that the fundamental group of X is a
free group of rank 2.
5. Show that A = (S1 × {x }) ∪ ({x } × S1 ), x ∈S1 , is not a retract of X = S1 × S1 .
6. Show that every geodesic joining two vertices in a connected graph X has finitely
many edges, and that it is isomorphic to the unit closed interval I.
7. Let X be a tree, and let T1 and T2 be trees in X such that T1 ∩ T2 ≠ ∅ . Show that
T1 ∩ T2 is also a tree.
8. Use the SvK theorem to find the fundamental group of a sphere together with one of
its diameters.
9. Find the fundamental group of the space made of two concentric spheres connected
with
(a) one tunnel (Illustration 13.23),
(b) two tunnels (Illustration 13.24),
(c) n tunnels along rays emanating from the common center of the spheres.

K12146_C013.indd 307 2/12/15 12:12 PM

 308   ◾    Seifert–van Kampen Theorem and Applications

Illustration 13.23 Illustration 13.24

10. Use the SvK theorem to compute the fundamental group of the space in Illustration
13.25 (a tunnel drilled out of a solid ball).

Illustration 13.25 Illustration 13.26

11. Let X be the space consisting of a ball, with finitely many disjoint smaller balls removed
from its interior and with finitely many disjoint tunnels connecting each pair of the
smaller balls. Assume in addition that X could be projected in  2 such that both the
set of small balls and the set of tunnels are pairwise disjoint in the projection. (This
somewhat mysterious condition is given to avoid any knotting or linkage of the tun-
nels; without it the conclusion is false in general.) An illustration of one such X, with 3
small balls removed and 6 tunnels connecting them, is provided in Illustration 13.26.
Show that π1 ( X , x ) is a free group of rank equal to the number of tunnels.
12. Let B be an open ball containing the object C shown in Illustration 13.27, and let X be
B \C. Use SvK to compute the fundamental group of the space X.

   Illustration 13.27

K12146_C013.indd 308 2/12/15 12:12 PM

 13.2 Seifert–van Kampen Theorem: Examples   ◾    309  

13. Let B be a ball containing the solid knotted object K shown in Illustration 13.28, and
let X be B \ C. Compute the fundamental group of the space X. (Hint: you do not have
to use the SvK theorem! But, if you do, then it is a good idea first to change the space
via an ambient isotopy within  3.)

   Illustration 13.28

14. Find the fundamental group of the space obtained from the hollow solid ball
{( x , y , z ) ∈ 3 : 1 ≤ x 2 + y 2 + z 2 ≤ 2} by drilling n many non-overlapping tunnels out
of it along rays emanating from the origin.

15. Let X be the space consisting of any finite number n ≥ 3 of closed (open) half-disks
sharing a common diameter (Illustration 13.29). Show that X is not homeomor-
phic to the disk D 2 (to B2 , respectively). [Hint: take out a point from the common
diameter.]

   Illustration 13.29

16. (a) Find the fundamental group of the space depicted in Illustration 13.30 (a torus
and a line segment joining two points on the surface of the torus).
(b) Find the fundamental group of the space depicted in Illustration 13.31 (n tori
touching a circle)

K12146_C013.indd 309 2/12/15 12:13 PM

 310   ◾    Seifert–van Kampen Theorem and Applications

T1 T2

T3
Tn

Illustration 13.30

Illustration 13.31

17. (a) Show that the fundamental group of the subspace X of  3 depicted in

Illustration 13.32 is a free group (circles converging on both ends to two
points, A and B, both of which are excluded from X). Find a set of free
generators of π1 ( X ) .
(b) Show that π1 ( 3 \({ A, B} ∪ X )) is also a free group and find a set of free
generators.

A B

Illustration 13.32

13.3 Seifert–van Kampen Theorem and Knots

A brief historical note: In 1905 Wilhelm Wirtinger (1865–1945) invented a method for comput-
ing the fundamental group of the complement of a tame knot in  3. Another early procedure
for describing such groups was published in 1910 by Max Dehn.

If two knots K1 and K 2 are ambient isotopic within  3, then they are called equivalent
knots. For such knots their  3 -complements K1c and K 2c are also ambient isotopic within
 3 (Corollary 3 in 10.1). Since ambient isotopic sets are necessarily homeomorphic, it
follows that K1c and K 2c have isomorphic fundamental groups. Consequently, if π1 ( K1c , x )
and π1 ( K 2c , x ) are not isomorphic, then K1 and K 2 are not ambient isotopic within  3 .
(The converse is false; more about that later in this section.)
The fundamental group π1 ( K c , x ) of the  3 -complement K c of a knot K is called the
knot group of K. In this section we will investigate some knot groups.
Let B = {( x , y , z ) ∈ 3 : x 2 + y 2 + z 2 < a } be an open ball enclosing a knot K. Then the
knot group of K is isomorphic to π1 ( B \ K , x ) (Exercise 1). Hence, in order to obtain a pre-
sentation of the knot group it suffices to investigate the complement of the knot in B. This
is what we will do in this section1.
1
Usually the knots are enclosed in the three sphere S3 (viewed as the one-point compactification of  3 ) instead of the ball
B. The advantage of S3 is that it is a fixed compact manifold containing all knots. For the purposes of what we cover in
this book, there is no difference between the two. We choose the ball B since it is readily visualizable in  3.

K12146_C013.indd 310 2/12/15 12:13 PM

 13.3 Seifert–van Kampen Theorem and Knots   ◾    311  

The following lemma tells us how to thread a loop around crossings of planar representa-
tions of knots. It will be needed throughout the rest of this section.

Lemma 1. With the setup shown in Illustration 13.33, αβα −1  δ in B \ K, where B is an

open ball enclosing the knot K.

Proof. The proof we give is a six-frame animation, as shown in Illustrations 13.33 to 13.38.

δ δ δ

α α
β β αβ

Illustration 13.33 The setup. Illustration 13.34 First we homo- Illustration 13.35 Continuing
topically change αβ. homotopy of αβ.

δ δ δ

αβα–1

α
αβα–1
αβ

Illustration 13.36 We
did Illustration 13.37 Showing the Illustration 13.38 It is now
what we wanted to αβ. Now result after performing a sim- evident that αβα −1  δ .
we show α again, and start ple homotopy.
deforming αβα–1.

This lemma facilitates the usage of the SvK theorem, as will be illustrated in the following three
examples in which we apply the SvK theorem directly on the complements of some knots. By the
end of this section we will provide a general method for computing knot groups.

Example 1: The Trefoil Knot Group

Consider the trefoil knot (as a subspace of an open ball B of radius, say, 2, as shown in
Illustration 13.39). We cover its complement X in B with two open sets, as shown in
Illustrations 13.40 and 13.41, and the corresponding captions.

K12146_C013.indd 311 2/12/15 12:13 PM

 312   ◾    Seifert–van Kampen Theorem and Applications

α1
α2
x
α3

Illustration 13.39 The trefoil knot; the space Illustration 13.40 The open set U: it consists of
we consider is its complement X in B. the points in X out of a closed ball of radius 1.

β1 β2
γ1

γ2
x β6
x β3
β4
β5

γ3

Illustration 13.41 The open set V: it consists of Illustration 13.42 The intersection U ∩ V
the points in X in an open ball of radius 1.5. contains the base point x. We also depict
(here and in Illustration 13.40 and 13.41) a
few loops that we will need.

The group π1 (U , x ) is free on a1 = [α1 ], a2 = [α 2 ], and a3 = [α 3 ] (Example 2 in 13.2).

The space V is homotopically equivalent to a bouquet of three circles, and so its fun-
damental group is also a free group; it is freely generated by c1 = [γ 1 ], c2 = [γ 2 ], and
c3 = [γ 3 ]. Finally π1 (U ∩ V , x ) is generated by b1 = [β1 ], b2 = [β 2 ], b3 = [β3 ], b4 = [β 4 ],
b5 = [β5 ], and b6 = [β6 ]. (In fact, π1 (U ∩ V , x ) is generated by any 5 of the 6 shown
elements; see Exercise 16 in 10.4.)
Denoting the inclusions by i : U ∩ V → U and j : U ∩ V → V , we observe (by staring
at Illustration 13.40–13.42) the following: i∗ (b1 ) = a1 , i∗ (b2 ) = a2−1 , i∗ (b3 ) = a2 , i∗ (b4 ) = a3−1,
i∗ (b5 ) = a3 , and i∗ (b6 ) = a1−1 . Three of the following six equalities need Lemma 1 as indi-
cated; the other three are obvious: j∗ (b1 ) = c1 , j∗ (b2 ) = c1−1c3−1c1 (Lemma 1), j∗ (b3 ) = c2 ,
j∗ (b4 ) = c2−1c1−1c2 (Lemma 1), j∗ (b5 ) = c3 , and j∗ (b6 ) = c3−1c2−1c3 (Lemma 1).
The SvK theorem then gives us the following presentation for the trefoil knot group:
〈a1 , a2 , a3 , c1 , c2 , c3 ; a1 = c1 , a2−1 = c1−1 c3−1 c1 , a2 = c2 , a3−1 = c2−1c1−1c2 , a3 = c3 , a1−1 = c3−1c2−1c3 〉.
Tietze transformations can reduce the redundancies in this presentation (for exam-
ple, we could immediately eliminate all of the ai’s), eventually ending up with
〈 x , y ; xyx = yxy 〉, one of the standard presentations for this group (Exercise 5a). ☐

K12146_C013.indd 312 2/12/15 12:13 PM

 13.3 Seifert–van Kampen Theorem and Knots   ◾    313  

Example 2: A Truncated Trefoil Knot

Consider the space Y depicted in Illustration 13.43, obtained from the space X in
Example 1 by removing a solid spherical cap from the enclosing ball.

γ2

γ1

Illustration 13.43 The space Y; the loops γ 1 and γ 2 , giving rise to the generators c1 and c2 of
its fundamental group are shown to the right.

With U and V the same as in Example 1, U ∩ Y and V ∩ Y make an open cover of Y,

and their intersection is path connected (so that the SvK theorem applies). The analysis
employed in Example 1 requires only slight modifications: compared to U , U ∩ Y is
lacking one tunnel, and its fundamental group has one less generator ( α1 ) ; compared to
U ∩ V , U ∩ V ∩ Y has a couple of tunnels missing, and hence has two fewer generators
( β1 and β6 ); V ∩ Y is the same as V. The computation of π1 (Y , x ) now follows on the
footsteps of the one done in Example 1, the only difference being the exclusion of the rela-
tions i∗ (b1 ) = j∗ (b1 ) , and i∗ (b6 ) = j∗ (b6 ) (corresponding to β1 and β6 ), and the exclusion
of the generator a1 (corresponding to the loop α1 ). Consequently the following is a pre-
sentation of π1 (Y , x ): a2 , a3 , c1 , c2 , c3 ; a2−1 = c1−1c3−1c1 , a2 = c2 , a3−1 = c2−1c1−1c2 , a3 = c3 .
This could be reduced to c1 , c2 ; c1c2c1−1 = c2−1c1c2 (Exercise 5c), where c1 and c2 are
the homotopy classes of γ 1 and γ 2 , respectively, as shown in Illustration 13.43. We will
use this result in Example 3 below. ☐

Example 3: The Square Knot Group

The Square Knot is shown in Illustration 13.44; we are after its knot group.

Illustration 13.44 The square knot.

Illustration 13.45 The square knot tunneled
out of an open disk.

K12146_C013.indd 313 2/12/15 12:13 PM

 314   ◾    Seifert–van Kampen Theorem and Applications

δ1 γ1 λ1
δ2 γ2

λ2

x x
x

Illustration 13.46 The open set Illustration 13.47 The open set Illustration 13.48 U ∩ V .
U. We also show two loops. V (and two loops).

We put the square knot in an open ball B (Illustration 13.45) and denote its comple-
ment in the ball by X, so that the associated knot group is π1 ( X , x ). Choose two open
sets, U and V, covering X as in Illustrations 13.46 and 13.47 ; notice that both U and V
are the same as the space denoted by Y and shown in Illustration 13.43.
It follows from Example 2, that c1 , c2 ; c1c2c1−1 = c2−1c1c2 is a presentation of
π1 (V , x ) , and d1 , d2 ; d1d2d1−1 = d2−1d1d2 is a presentation of π1 (U , x ) , where ci is a
homotopy class of the loop denoted γ i and shown in Illustration 13.47 (i = 1, 2), and
where di is a homotopy class of the loop denoted δi and shown in Illustration 13.46
(i = 1, 2). The group π1 (U ∩ V , x ) is (freely) generated by the homotopy classes of the
loops λ1 and λ 2 (Illustration 13.46).
As usual we denote the inclusions by i : U ∩ V → U and j : U ∩ V → V . We observe
that i∗ ([λ1 ]) = d1 , and that j∗ ([λ1 ]) = c1 , so that c1 = d1 will appear as one of the rela-
tions in the presentation stemming from the SvK theorem.
We now analyze the relation i∗ ([λ 2 ]) = j∗ ([λ 2 ]) in the SvK presentation. A straight-
forward approach would be to express i∗ ([λ 2 ]) and j∗ ([λ 2 ]) in terms of the genera-
tors of π1 (U , x ) and π1 (V , x ), respectively. For example, threading loops through V
by means of Lemma 1, it can be seen that λ 2 is homotopic to γ −2 1 γ 1−1 γ 2 γ 1 γ 2 , implying
that j∗ ([λ 2 ]) = c2−1c1−1c2c1c2 . Symmetrically, i∗ ([λ 2 ]) = d2−1d1−1d2d1d2 .
Hence, by SvK theorem, c1 , c2 , d1 , d2 ; d1d2d1−1 = d2−1d1d2 , c1c2c1−1 = c2−1c1c2 , c1 = d1 ,
c2 c1 c2c1c2 = d2−1d1−1d2d1d2 is a presentation for π1 ( X , x ) . With that our computation
−1 −1

is completed.
Remark. There is a shorter route that makes it evident that it is completely irrel-
evant what i∗ ([λ 2 ]) and j∗ ([λ 2 ]) are in terms of the generators of π1 (U , x ) and
π1 (V , x ). The loop λ 2 is homotopic within V to the loop λ1−1 : stare at Illustrations
13.47 and 13.48 and simply widen λ 2 enough so that we can thread the whole
knotted tunnel through it, emerging on the other side as λ1−1 . So, j∗ ([λ 2 ]) = c1−1
holds in the presentation for π1 (V , x ) . By symmetry, δ 2 is homotopic to δ1−1
within U, and so i∗ ([λ 2 ]) = d1−1 holds in the presentation of π1 (U , x ). On the other
hand we showed above that c1 = d1 holds in the SvK presentation of π1 ( X , x ).

K12146_C013.indd 314 2/12/15 12:13 PM

 13.3 Seifert–van Kampen Theorem and Knots   ◾    315  

Thereby i∗ ([λ 2 ]) = j∗ ([λ 2 ]) is a consequence of the other relations in the SvK pre-
sentation of π1 ( X , x ), and thus, by Tietze, it could be omitted. We conclude that
c1 , c2 , d1 , d2 ; d1d2d1−1 = d2−1d1d2 , c1c2c1−1 = c2−1c1c2 , c1 = d1 is a presentation of π1 ( X , x ).
Eliminating d1 (Tietze), we get c1 , c2 , d2 ; c1d2c1−1 = d2−1c1d2 , c1c2c1−1 = c2−1c1c2 , which in
turn is equivalent to c1 , c2 , d2 ; d2c1d2 = c1d2c1 , c2c1c2 = c1c2c1 , or 〈a, b, c ; cac = aca,
bab = aba〉, which is one of the standard presentations of the square knot group. ☐

Example 4: The Granny Knot Group

The Granny Knot is depicted in Illustration 13.49. The knot group of the granny knot
is the same as the knot group of the square knot (Exercise 7).

Illustration 13.49 The granny knot.

However, the square knot is not ambient isotopic within  3 to the granny knot. This
is a consequence of the fact that the complements of these two knots in  3 are not
homeomorphic [20]. ☐

We end this section with the promised general procedure for computing knot groups.

A Method for Computing Knot Groups Using Knot Diagrams

For practical reasons we usually represent knots in the drawing plane, rather than in  3 ,
where they belong. More precisely, our illustrations represent p(K), where K is the knot placed
above the drawing plane  2 , and p :  3 →  2 is the vertical projection. In doing so, we intro-
duce intersections: these are points x ∈ p( K ) such that | p −1 ( x )| ≥ 2. We consider only knots
for which it is possible to choose the drawing plane such that | p −1 ( x )| is either 1 or 2 for
every x, and such that when | p −1 ( x )| = 2 then the intersection is transverse. Briefly, transverse
crossing is when the two parts of the knot in the diagram cross in the manner the x-axis and
the y-axis do (but not similar to the way y = x 2 and the x-axes cross). Precise definition and
technical details can be found in books on knot theory. If p(K) satisfies all of the above con­
ditions, then we call it a knot diagram. The points x in a diagram p(K) where | p −1 ( x )| = 2 are
called crossings. For example, the diagram for the trefoil knot (Illustration 13.50) has three
crossings.
The method we are about to describe applies to knots allowing knot diagrams with
finitely many crossings.
 
Let D be K \  ∪ p −1 ( x ) . Denote the components of D by C1 , C2 ,…, Cm . Each com-
 x is a crossing 
ponent Ci of D projects injectively via p onto a component of the diagram of the knot with

K12146_C013.indd 315 2/12/15 12:13 PM

 316   ◾    Seifert–van Kampen Theorem and Applications

the crossing points deleted. For example, the standard diagram of the trefoil knot is such
that D has 6 components (Illustration 13.51).

x3 C2 C1
x3

C4 C5
α1
x1 x2
x1 C6
x2 α5

α3

C3
Illustration 13.50 Three crossing points
in the standard diagram for the trefoil Illustration 13.51 The 6 components of D here are C1
knot. to C3. The loop α i winds once around Ci. To avoid
clutter, only the loops α1 , α 3 and α 5 are shown.

Choose a loop α i at a base point and winding once around exactly one component Ci
(see Illustration 13.51). Then the homotopy classes of these loops generate the knot group.
The last claim can be justified by repeated usage of the SvK theorem; we indicate how to
do this below.

αi3
αi2

αi4
αi1

Illustration 13.52 The neighborhood U.

Illustration 13.53 The relevant (homotopy classes

of) loops in U are α i1 , α i2 , α i3 and α i4 .

Choose an open ball U containing the inverse image p −1 ( x1 ) of a crossing x, such that
U is small enough to avoid the inverse images of any other crossing points (Illustration
13.52). Then  3 \ K is covered by U, and by the complement V of a smaller ball U1 ⊂ U
encompassing p −1 ( x1 ). The intersection U ∩ V is a spherical shell with four tunnels, simi-
lar to the space depicted in Illustration 13.42. The four tunnels give rise to four (homo-
topy classes of) loops α i1 , α i2 , α i3 , and α i4 . Careful usage of the SvK theorem, together
with Lemma 1, introduces the following two relations: α i1 = α i2 and α i4 = α i1 α i3 α i−11

K12146_C013.indd 316 2/12/15 12:14 PM

 13.3 Seifert–van Kampen Theorem and Knots   ◾    317  

(see Illustration 13.53). If A ; R is a presentation of π1 (V , x ), then SvK produces

A ; R , α i1 = α i2 , α i4 = α i1 α i3 α i−11 as a presentation for the knot group π1 ( 3 \ K , x ).
If there are no other crossings then π1 (V , x ) is a free group, and we end up with
A ; α i1 = α i2 , α i4 = α i1 α i3 α i−11 as the final presentation of the knot group. Otherwise we
repeat the above procedure, working with V in place of  3 \ K . Each step will bring rela-
tions of the type α i = α j or α i = α j α k α −j 1 . After finitely many steps (corresponding to the
number of crossings), and after eliminating one half of the generators using the relations
of type α i = α j , we end with an explicit presentation 〈 A ; {α is = α js α ks α −js1 : s = 1,2,…, n}〉
of the knot group π1 ( 3 \ K , x ) . This is the Wirtinger presentation of the knot group; for
more details see, for example, [72], page 144.
The following example makes it clear what should be done to precisely identify the rela-
tions in the presentation of a knot group.

Example 5: Trefoil Knot Group, Again

The last procedure allows us to essentially read a presentation of the trefoil knot group
(or any other knot group) directly from the associated knot diagram. In Illustration
13.54, we assume that the base point of the space is close to the location of the viewer’s
eyes above the plane containing the figure, and that the generating loops are winding
in the indicated direction (to make this orientation of the loops consistent, we have
introduced an orientation along the knot).

α2 = α1
α4 = α2α5α–1
2

α5

α2 α4
α1
α6

α5 = α6 α3 = α4
α2 = α6α3α –1 α6 = α4α1α 4–1
6

α3
Illustration 13.54

The six relations that immerge are shown in the figure. Consequently, we have the
following presentation of the trefoil knot group: α1 , α 2 , α 3 , α 4 , α 5 , α 6 ; α 2 = α1 , α 4 =
α 2α 5α −2 1 , α 5 = α 6 , α 2 = α 6α 3α 6−1 , α 3 = α 4 , α 6 = α 4α1α 4−1 . Using Tietze trans-
formations we eliminate α2, α4 and α6 to end up with the Wirtinger presentation
〈α1 , α 3 , α 5 ; α 3 = α1α 5α1−1 , α1 = α 5α 3α 5−1 , α 5 = α 3α1α 3−1 of the trefoil knot group. That
what we get is equivalent to the earlier presentations for this knot group is left as an
exercise (Exercise 8). ☐

K12146_C013.indd 317 2/12/15 12:14 PM

 318   ◾    Seifert–van Kampen Theorem and Applications

Exercises
1. Let K be a knot and let a be a positive number such that K ⊂ B = {( x , y , z ) ∈ 3 :
x 2 + y 2 + z 2 < a} . Show that the knot group of K is isomorphic to π1 ( B \ K , x ).
2. View S 3 as the one-point compactification of  3 so that the latter can be considered as
a subspace of the former (via the embedding equal to the inverse of the stereographic
projection). Show that the knot group of a knot K (in  3 ) is isomorphic to π1 (S 3 \ K , x ).
3. Find the knot group of an unknot.
4. Prove that each generator in every Wirtinger presentation defines a non-trivial loop
in the complement of the knot.
5. (a) S how that the presentation for the trefoil knot group obtained in Example 1 can
be reduced by means of Tietze transformations to 〈 x , y ; xyx = yxy 〉. [Hint, first
eliminate ai ’s, then, in the next step, c1 ]
(b) Show that 〈a, b ; a3 = b 2 〉 is also a presentation of the trefoil knot group.
(c) Show that the presentation of the fundamental group of the space Y
from Example 2 can be reduced by means of Tietze transformations to
〈 x , y ; xyx −1 = y −1 xy 〉 .
6. Consider the open cover of the complement of the trefoil knot, as shown in
Illustration 13.55.

Illustration 13.55 The open sets U (left) and V (middle), and U ∩ V (right).

Use the SvK theorem on this cover, and again find the trefoil knot group, then employ
Tietze transformations to change that presentation to 〈 x , y ; xyx = yxy 〉 .
7. (a) Use the SvK theorem to find a presentation for the granny knot shown in
Illustration 13.49.
(b) Use Tietze transformations to show that a presentation solving (a) is equivalent
to a, b, c ; cac = aca, bab = aba , concluding that the granny knot and the square
knot have the same knot groups.
8. Use Tietze transformations to confirm that the presentation of the trefoil knot group
obtained in Example 5 is equivalent to 〈 x , y ; xyx = yxy 〉.
9. Show that the knot group of a knot representable by a knot diagram with finitely
many crossing points is infinite.

K12146_C013.indd 318 2/12/15 12:14 PM

 13.4 Torus Knots and Alexander’s Horned Sphere   ◾    319  

10. Find a presentation of the knot group of the knot shown in Illustration 13.56.

Illustration 13.56
11. Find a presentation of the knot group of the Figure-Eight Knot shown in Illustration 13.57.

Illustration 13.57

12. The notion of a connected sum of two knots is explained in Illustration 13.58 and its
caption. (For a precise definition see, for example, [12].)
K1 K2 K1 # K2

Illustration 13.58 We cut both knots between two crossing points, then glue them as in
the right figure.

Show that if one of the knot groups of the knots K1, K 2 is not cyclic, then every con-
nected sum of these two knots is not an unknot.

13.4 Torus Knots and Alexander’s Horned Sphere

A brief historical note: James Waddell Alexander (1888–1971) presented his wild embedding
of the sphere in  3, or the “horned sphere,” in an article published in 1924.

Torus knots are knots along the surface of a torus. We will classify them completely
through their knot groups. The Fox–Artin wild arc, as we will see, is related to torus knots.

K12146_C013.indd 319 2/12/15 12:14 PM

 320   ◾    Seifert–van Kampen Theorem and Applications

It is one of the two examples mentioned in this section where we indicate that a sphere can
be properly knotted in  3. The other example is the Alexander’s horned sphere.

Torus Knots
A torus knot of type (p,q), where p and q are relatively prime positive integers, is a
knot on the surface of a torus that winds p times in the latitudinal direction and q times
in the meridinal (or longitudinal) direction. Alternatively, the knot k p , q : I → S1 × S1 ⊂  3,
k p ,q (0) = k p ,q (1), is a torus knot of type (p,q), for p and q relatively prime positive integers,
kp , q
if I  → S1 × S1 p 1
→ S1 ( p1 is the projection onto the first S1) winds p times around
k
S1, and I  p,q
→ S1 × S1 p2
→ S1 ( p2 is the projection onto the second S1) winds q times
1
around that S . There is yet a third equivalent definition: if we consider the torus to be
the quotient space of  2 under the equivalence defined by ( x , y ) ~ ( x ′ , y ′ ) if both x − x ′
p
and y − y ′ are integers (Exercise 2 in 4.2), then the line y = q x in the plane, p and q
relatively prime, becomes a torus knot of type (p,q).
An example of a torus knot is given in Illustrations 13.59 and 13.60. It is evident that the
knot is of type (4,5).

Illustration 13.60 The same torus knot of

Illustration 13.59 A torus knot of type (4,5). type (4,5) is shown on the surface of a torus.

Our first goal is to find the knot group of a torus knot of type (p,q). The illustrations we
show are of the (4,5) knot, but the computation we give below is for the fundamental group
of any torus knot of type (p,q).
First we replace the torus knot with a tubular torus knot—that is, with the product of the
torus knot with a small open disk, small enough to avoid self intersections. (This change is
now not only for esthetic reasons.) Since these two are isotopic within  3 (Exercise 10 in
10.2), so are their complements. Consequently, we can compute the knot group of the torus
knot by computing the fundamental group of the complement of this tubular torus knot.
Finally, we can substitute that underlying space  3 with any large open ball B containing
the knot. Our target now becomes finding the fundamental group of the complement B \ K
of the tubular (p,q)-knot K.
We cover B \ K with two open sets, U and V. The open set U is the grooved solid torus
(S1 × B2 ) \ K, where, we recall, B2 is the open unit disk (see Illustration 13.61). The tube
around the torus knot cuts a tubular groove on the surface of U, and so U has the same

K12146_C013.indd 320 2/12/15 12:14 PM

 13.4 Torus Knots and Alexander’s Horned Sphere   ◾    321  

fundamental group as the solid torus S1 × B2 . Hence π1 (U , x ) is infinite cyclic, freely

generated by the homotopy class a of the loop that follows the middle circle S1 × {(0,0)} ,
winding once around the torus.

Illustration 13.61 The open Illustration 13.62 The open Illustration 13.63 The open
set U. set V. set U ∩ V .

The open set V is the complement in B \ K of ( B \ K ) ∩ (S1 × B∗2 ), where B∗2 is an open disk
of radius slightly smaller than 1 (“slightly” means “less than one half of the diameter of the
tube in the tubular knot we used above”). The open set V is shown in Illustration 13.62. The
difference between V and the complement of the solid torus S1 × B∗2 in B is inconsequen-
tial: V has tubular carvings on its walls, while B \(S1 × B∗2 ) does not. Consequently their
fundamental groups are the same. Since the fundamental group of B \(S1 × B∗2 ) is infinite
cyclic (see Exercise 3 in 13.3), so is π1 (V , x ). A generator is the homotopy class b of any loop
that winds once around the solid torus S1 × B∗2 (passing through the central hole of the torus).
The intersection U ∩ V is shown in Illustration 13.63: it is a strip that winds p times
in the latitudinal direction and q times in the meridinal direction. Since U ∩ V is clearly
homotopically equivalent to a circle, the group π1 (U , x ) is freely generated by the homotopy
class of a loop δ winding along the strip. It is evident that δ defines a p in π1 (U , x ), and that
it defines bq in π1 (V , x ) . It follows from the SvK theorem that π1 ( B \ K , x ) = 〈a, b ; a p = bq 〉.
Summarizing: the knot group of a torus knot of type (p, q) is 〈a, b ; a p = bq 〉.
If ( p1 , q1 ) and ( p2 , q2 ) are two distinct pairs of relatively prime positive integers, such
that p j < q j , j ∈ {1, 2}, then 〈a, b ; a p1 = bq1 〉 and 〈a, b ; a p2 = bq2 〉 are not isomorphic
(Exercise 3). Consequently, torus knots of distinct types ( p, q), p < q, are not ambient
isotopic within  3 .

Alexander’s Horned Sphere

According to the Jordan-Schönflies theorem (Theorem 6 in 11.3), every embedding
f : S1 →  2 can be extended to a homeomorphism fˆ :  2 →  2 that sends the open unit
disk B2 onto the bounded component of  2 \ f (S1 ), and sends  2 \ D 2 onto the unbounded
component of  2 \ f (S1 ) . For a while people believed that the analogue statement should
hold a dimension higher, with S 2 in place of S1 and with  3 in place of  2 . At one point
Alexander announced that he had a proof. Upon discovering a gap in his argument he came

K12146_C013.indd 321 2/12/15 12:14 PM

 322   ◾    Seifert–van Kampen Theorem and Applications

up with a counter example, now called the Alexander’s horned sphere. It is depicted in
Illustration 13.64.

CS
1
32

1
16

Illustration 13.64 Alexander’s horned sphere AHS: it is homeomorphic to the sphere. The
precise template for the “horns” is shown to the right and it comes from Illustration 5.13
(the Cantor interval). In the figure to the left we widen up the “horns” to show the linkage
more clearly.

We show that the unbounded component X of  3 \ AHS is not simply connected.

Consequently,  3 \ D 3 (i.e., the unbounded component of  3 \ S 2 ), being simply connected,
is not homeomorphic to X, hence the once-expected extension of the Jordan–Schönflies
theorem in three dimensions fails.
First observe that π1 ( X , x ) is generated by the homotopy classes of the loops in
A = {a11 , a12 } ∪ {a21 , a22 , a23 , a24 } ∪ {a31 , a32 , a33 , a34 , a35 , a36 , a37 , a38 } ∪ … as shown in Illu­st­­
ration 13.65. We assume that they all wind clockwise.
We use the SvK theorem to find an incom-
plete presentation of the fundamental group of
AHS over the set A.
a31 Σ3
Step 1. We split X into two open sets, U1 and
U 0 : U 0 is bounded from above by a plane that
a22 a23 a24
is just above the plane ∑1 and U1 is bounded a21 Σ2
from below by a plane that is just below ∑1 , as
shown in Illustration 13.65. Using the notation a11
Σ1
a12
shown in Illustration 13.65 it is easy to compute
that π1 (U 0 , x ) = a11 , a12 ; a12 = a11−1 , and that
π1 (U 0 ∩ U1 , x ) = 〈a11 , a12 ; ∅〉 , where aij -s denote
both the loops and their homotopy classes in
the respective open sets. As for U1 , for the time
being we merely state that π1 (U1 , x ) = 〈 A ; R 〉
for some set of relations R. The amalgamation Illustration 13.65

K12146_C013.indd 322 2/12/15 12:14 PM

 13.4 Torus Knots and Alexander’s Horned Sphere   ◾    323  

produces equating a11 and a12 with the corresponding generators in π1 (U 0 , x ) and
π1 (U1 , x ) which we have denoted by the same symbols. Hence, these relations could be
ignored, and the SvK theorem gives π1 ( X , x ) = A ; a12 = a11−1 , R .

Step 2. Now we focus on U1 and cover it with two open sets U 2 and U 3 , where U 2 is bounded
from above by a horizontal plane just above ∑ 2 , and U 3 is bounded from below by a horizon-
tal plane just below ∑ 2 as in Illustration 13.65. We compute: π1 (U 2 , x ) = 〈a21 , a22 , a23 , a24 ; ∅〉,
π1 (U 2 ∩ U 3 , x ) = 〈a21 , a22 , a23 , a24 ; ∅〉, and π1 (U 3 , x ) = 〈 A \ {a11 , a12 }; R1 〉 , where R1 is a
set of relations not involving a11 and a12 . Using the SvK theorem, this gives π1 (U1 , x ) =
〈 A \ {a11 , a12 }; R1 〉 . Now we back up and modify the outcome of Step 1: since the last pre-
sentation of π1 (U1 , x ) does not use a11 and a12 , the amalgamation of π1 (U 0 ∩ U1 , x ) =
〈a11 , a12 ; ∅〉 in Step 1 requires that we express a11 and a12 in terms of A \ {a11 , a12 }. This
is easy to achieve: it is obvious that a12 = a22a24 , and by Lemma 1 in 13.3 we have a11 =
a21a22a23a22 −1 . So, Step 1 and 2 together produce the following: π1 ( X , x ) = A ; a12 = a11−1 , a11 =
a21a22a23a22 −1 , a12 = a22a24 , R1 where the set R1 does not have relations involving a11 and a12 .
To see the iterative procedure more clearly we do one more step.

Step 3. Here we take a look at U 3 . We split it into open sets U 4 and U 5 as suggested by
Steps 1 and 2: U 4 is bounded from above by a horizontal plane just above ∑3 , and U 5 is
bounded from below by a horizontal plane just below ∑3 as in Illustration 13.65. It is
again easy to compute that π1 (U 4 , x ) = 〈a31 , a32 , a33 , a34 , a35 , a36 , a37 , a38 ; ∅〉, π1 (U 4 ∩ U 5 , x ) =
〈a31 , a32 , a33 , a34 , a35 , a36 , a37 , a38 ; ∅〉, and π1 (U 5 , x ) = 〈 A \ {a11 , a12 , a21 , a22 , a23 , a24 }; R2 〉, where
R2 does not contain relations involving the generators that we have deleted from the set A.
SvK gives π1 (U 3 , x ) = 〈 A \ {a11 , a12 , a21 , a22 , a23 , a24 }; R2 〉 . We again back up to Step 2 and use
this presentation there; the only thing we need to do is to express {a11 , a12 , a21 , a22 , a23 , a24 }
in terms of the other generators of π1 (U 3 , x ). Since a11 , a12 were already given in terms of
a21 , a22 , a23 , a24 , we focus on these last four generators. Fortunately (and for our limited
purposes) we do not need to worry about all of them: just focus on a21 (the first one), and
−1
a24 (the last one). We compute (using Lemma 1 in 13.3) that a21 = a31a32a33a32 , and we
easily see that a24 = a37a38 . The remaining two generators, a22 and a23 are clearly express-
ible in terms of a32 , a33 , a34 , a35 , a36 , a37 ; this little piece of information is sufficient for our
−1
purposes. We have π1 ( X , x ) = A ; a12 = a11−1 , a11 = a21a23 −1
a22a23 , a12 = a22a24 , a21 = a31a32a33a32 ,
a24 = a37a38 , Q1 , R2 , where Q1 involves only middle generators—those not of type ai1 , and
not of type ai 2i —and where R2 contains relations not involving generators encountered
so far.
Continuing this recursive construction and using induction, one gets the following:
π1 ( X , x ) = A ; a12 = a11−1 , a11 = a21a22a23a22 −1
, a21 = a31a32a33a32 −1
, a31 = a41a42a43a42 −1
,
−1
a41 = a51a52a53a52 ,…a12 = a22a24 , a24 = a37 a38 , a38 = a4 15a4 16 ,… , Q ,
where Q has only relations not involving generators of types ai1 and ai 2i .

K12146_C013.indd 323 2/12/15 12:14 PM

 324   ◾    Seifert–van Kampen Theorem and Applications

Now add a relation of type y =1 for each generator y ∈ A not of type ai1 and not of
type ai 2i . The resulting group os a quotient group of π1 ( X , x ) . After applying Tietze
transformations to eliminate redundant relations, we change this new presentation to

A ; a12 = a11−1 , a11 = a21 , a21 = a31 , a31 = a41 , a41 = a51 ,… ,

a12 = a24 , a24 = a38 , a38 = a4 16 ,…,

By means of evident Tietze transformations we can further reduce this presentation to

〈a12 ; ∅〉. Since 〈a12 ; ∅〉 is obviously not the trivial group, and since it is a quotient group of
π1 ( X , x ), it follows that π1 ( X , x ) ≠ 1, and we have established what we were after.
A description of the fundamental group of the complement of the Alexander’s horned
sphere in  3 can be found in, say, [31], page 171.

Exercises
1. Explain why the trefoil knot is a (2,3) torus knot. Conclude that 〈 x , y ; x 2 = y 3 〉 is a
presentation for the knot group of the trefoil knot. Use Tietze transformations to show
that this presentation is equivalent to 〈a, b ; aba = bab〉 (see Example 1 in Section 13.3).
2. Show that 〈 x , y ; x 4 = y 5 〉 is not isomorphic to 〈a, b ; a5 = b6 〉. Deduce that the
(4,5)-knot (Illustration 13.66) is not ambient isotopic within  3 to the (5,6)- knot
(Illustration 13.67).

A (4,5)-strip. A (5,6)-strip.
Illustration 13.66 Illustration 13.67

3. Generalize Exercise 2 as follows. Let ( p1 , q1 ) and ( p2 , q2 ) be two distinct pairs of relatively

prime positive integers such that p j < q j , j ∈ {1, 2}. Denote G = x 1 , y1 ; x 1p1 = y 1q1 and
H = x 2 , y 2 ; x 2p2 = y 2q2 .
(a) Show that the centralizer of G is the subgroup of G generated by x1p . (Recall that
the centralizer of a group G is {x ∈ G : xy = yx for every y ∈ G}.) [Hint: The sub-
group N of G generated by x1p is normal in G and G N has trivial center.]
(b) Show that G and H are not isomorphic. Deduce that a torus knot of type ( p1 , q1 )
is not ambient isotopic within  3 to a torus knot of type ( p2 , q2 ) .

K12146_C013.indd 324 2/12/15 12:15 PM

 13.4 Torus Knots and Alexander’s Horned Sphere   ◾    325  

4. Take a torus knot K of type (p,q), where p and q are relatively prime. Entangle the
torus so it becomes a thick trefoil knot (as in Illustration 13.68). The torus knot K
changes and it becomes a trefoil knot of type (p,q). Compute its knot group.

Illustration 13.68

5. Embed the Möbius band in  3 in the usual way (see Illustration 14.33). Observe
that its boundary forms the unknot. Confirm (with naked eye) that the bound-
ary of the triple twisted band (embedded in  3 by rotating an edge of a rectangle
by 3π radians, then identifying it with the opposite edge) is the trefoil knot.
Draw knot diagrams for the boundaries of (2n + 1)-times twisted bands. Find
presentations for their knot groups.

6. In Illustration 13.69, we show a subspace H of  3 ; it is a solid object consisting of

all points within the sphere and the knotted part, together with the apparent point
of convergence at the top of the figure. Find the fundamental group of  3 \ H.

1
16
1
8
1
4

1
2

Illustration 13.69

K12146_C013.indd 325 2/12/15 12:15 PM

 326   ◾    Seifert–van Kampen Theorem and Applications

7.* The Fox-Artin Embedding of S 2 in  3

We describe another embedding f : S 2 →  3 that cannot be extended to a homeomor-
phism f :  3 →  3 such that the unbounded component  3\ D 3 of  3\ S 2 is sent to the
unbounded component of  3\ f (S 2 ). It is based on the so-called Fox-Artin wild arc
(see Illustration 13.70).

Illustration 13.70 Fox-Artin wild arc, or a special tubular copy of S2.

The Fox-Artin wild arc, gives rise to an embedding f of S 2 in  3 as follows: view the
knot as a tube, the diameter of which tapers off to a point as we move away to the left
or to the right from the middle part. We close that tube with the two accumulation
points shown to the right and to the left of the knot, thus obtaining a sphere embedded
in  3.
  It is known that the unbounded component of  3\ f (S 2 ) is not simply connected
([21]). We will indicate how to justify this claim based on the method we have used to
compute the knot groups of torus knots.
  We exhibit the Fox-Artin embedding of S 2 as a kind of a “torus knot,” where the
role of the torus is now played by a special connected sum of tori (as shown in the
Illustrations 13.71 and 13.72).

Illustration 13.71 Exhibiting the space from the illustration above as a “torus knot”.

Illustration 13.72 Zooming in to see it more clearly.

Denote the unbounded component of  3\ f (S 2 ) by X. Cover X with two open sets

in the same way as with the torus knots. One set, U, is the intersection of X and the
chain of open solid tori, the boundary of which contains the central curve of the tube.
Illustrations 13.71 and 13.72 apply to U too, except that we should view the tori in the
chain as being solid, with the knotted sphere along the Fox-Artin wild arc canalling
a ridge on the surface of the object. The other open set, V, is the complement of a
slightly smaller concentric chain of closed solid tori. We point out again that U and V

K12146_C013.indd 326 2/12/15 12:15 PM

 13.5 Links   ◾    327  

are subsets of X, and so the two accumulation points at the extreme right and left of
the arc are neither in U nor in V.
(a) Apply the SvK theorem to the cover {U, V} of X to get a presentation for π1(X).
(b) Show that π1(X) is not the trivial group.
(c) Deduce that the embedding f : S 2 →  3 cannot be extended to a homeomorphism
f :  3 →  3.

13.5 Links

A brief historical note: As objects of mathematical studies, links appeared for the first time
in a note (1833) by Carl Friedrich Gauss (1777–1855), and in relation to his work on elliptic
integrals.

In this short section we will define links and show a few examples.
A link in  3 is a disjoint union of two knots. The link group of a link is the fundamental
group of the complement of the link in  3 (or the complement of the link in an open disk
that encloses it). Two links L1 and L2 are equivalent if they are ambient isotopic within
 3. Two disjoint knots K1 and K 2 are unlinked (or, K1 and K 2 form an unlink) if there is
an ambient isotopy within  3 between the link K1 ∪  K 2 and a link K1′ ∪ K 2′ , so that the
knots K1′ , K 2′ can be separated by a plane. Otherwise they are linked. The two unknots in
Illustration 13.73 are unlinked.

Illustration 13.73 Two unlinked unknots!

It follows from the SvK theorem that if K1 and K 2 are unlinked, then the associated link
group is the free product of the knot groups of K1 and K 2 . In particular, if K1 and K 2 are
unknots forming an unlink, as in Illustration 13.73, then the link group is the free product
 ∗  = 〈a, b ; ∅〉 .

Example 1: The Hopf Link

The Hopf link is depicted in Illustration 13.74. An easy application of the SvK theo-
rem—for example, by employing a slight modification of the knot-diagram method
described in Section 13.3—establishes that the link group for the Hopf link is

K12146_C013.indd 327 2/12/15 12:15 PM

 328   ◾    Seifert–van Kampen Theorem and Applications

 ⊕  = a, b; ab = ba (Exercise 1). Since the groups  ∗  and  ⊕  are not iso-

morphic, we have an algebraic justification of what is visually clear: that the Hopf link
is not equivalent to the unlink.

Illustration 13.74 The Hopf link.

☐

Example 2: The Generalized Hopf Link

Let 〈 A ∪ {α} ; P 〉 be a presentation of the knot group of a knot K1 , and let 〈 B ∪ {β} ; R 〉
be a presentation of the knot group of a knot K 2 ; the generators α and β—or rather
the corresponding loops—are shown in Illustration 13.76 below; each of them is not
trivial in the corresponding knot group (Exercise 4 in 13.3). We link the knots K1
and K 2 as in Illustration 13.75. Our goal is to find a presentation of the link group of
this link in terms of the given presentations of the two knot groups.

K1
K2

Illustration 13.75 A generalized Hopf link.

In Illustration 13.76, we show the link again, together with four loops: α1, α, β1,
and β. It is visible that β1 and β are homotopic in the complement of the shown link:
we merely thread the knot K 2 through one of these two loops, as indicated by the dot-
ted loop at the right of Illustration 13.76. Similarly, the loops α1 and α are homotopic
in the complement of the link. So, we can write β1 = β, and α1 = α in the link group.

K1 α1 β1 K2

α β

Illustration 13.76 The dotted loop indicates how to thread K 2 through loops homotopic to β1.

At this stage, in order to compute the link group, we can either use the general
procedure for computing knot groups (as described at the end of Section 13.3),

K12146_C013.indd 328 2/12/15 12:15 PM

 13.5 Links   ◾    329  

modified slightly for links, or use the SvK theorem. The former is faster, and we
see that there are two relations introduced by the two crossings: α = βαβ −1 and
β = αβα −1 . Both tell us that αβ = βα. Consequently, the link group has a presentation
〈 A ∪ {α} ∪ B ∪ {β}; P ∪ R , αβ = βα〉. It follows from the assumption that α and β are
not equal to the trivial elements in the respective presentation, from Proposition 3 in
12.3, and from the fact that αβα −1β −1 = 1 in this group, that this presentation is not
equivalent to the free product of the two knot groups. Consequently, the knots K1
and K 2 are necessarily linked in the generalized Hopf link. ☐

A knot K1 is weakly unlinked 1 from the knot K 2 if every loop in K1 is homotopically

trivial in  3 \ K 2 . Clearly, if the knots in a link are unlinked, then each of the knots is
weakly unlinked from the other. The following example shows that in linked knots one of
the knots could be weakly unlinked from the other.

Example 3: The Whitehead Link

Illustration 13.77

K2

K2 K2

K1

Illustration 13.78

In Illustration 13.77 we depict the Whitehead link. The unknot K1 is weakly unlinked
from the unknot K 2 since any generating loop of π1 ( K1 ) is contractible in  3 \ K 2
(see Illustration 13.78; do not forget that self-intersection of a loop during a homotopy
is allowed), and so every loop in K1 is contractible in  3 \ K 2 . ☐

1
Alternative terminology: if a knot K1 is weakly unlinked from a link K 2, then it is said that the former is link-homotopic
to the unknot.

K12146_C013.indd 329 2/12/15 12:15 PM

 330   ◾    Seifert–van Kampen Theorem and Applications

Exercises
1. (a) Find the link group of an unlink.
(b) Show that the link group of the Hopf link is isomorphic to  ⊕ .
2. Compute the link group of the Whitehead link. (Perhaps the view of the
Whitehead link shown in Illustration 13.79 is better suited for computational
purposes.)

Illustration 13.79

3. Compute the link group of the link shown in Illustration 13.80, then show that it is
not isomorphic to a free group of rank 2.

Illustration 13.80

4. Find the link group of the link shown in Illustration 13.81.

Illustration 13.81

K12146_C013.indd 330 2/12/15 12:15 PM

 13.5 Links   ◾    331  

5. Show that the unknots K1 and K 2 in Illustration 13.82 are linked, show that K1 is
weakly unlinked from K 2 and show that K 2 is weakly unlinked from K1 . [Hint: for
the first part: compute the link group and show that it is not isomorphic to the link
group of the unlink.]

K1

K2

Illustration 13.82

Illustration 13.83

6. Compute the fundamental group of the complement of the link depicted in Illustration
13.83.

7. Exhibit a link of three unknots K1 , K 2 , and K 3 such that for every i , j ∈{1,2,3}, i ≠ j ,
the knots K i , K j are not unlinked, and K i is weakly unlinked from the knot K j .
8. Prove or find a counterexample: for every embedding f : D 2 →  3 , if a knot K is linked
with f (∂D 2 ) = f (S1 ), then f (D 2 ) ∩ K ≠ ∅ (see Illustration 13.84 below).

K
f (D2)

Illustration 13.84 If K is linked with f (S1 ), does it follow that K perforates f (D 2 )?

K12146_C013.indd 331 2/12/15 12:16 PM

This page intentionally left blank
 Chapter 14

On Classifying Manifolds
and Related Topics

A brief historical note: The notion (and the name) of manifold was introduced by Bernhard
Riemann (1826–1866).

14.1 1-Manifolds
It is relatively easy to classify all connected 1-manifolds, and to provide a justification that,
up to homeomorphism, there are only two of them: S1 (Theorem 1) and  (Exercise 3).

Theorem 1. Every compact, connected 1-manifold is homeomorphic to S1 .

Proof (based on Gale, [23]).

Let X be a connected, compact 1-manifold, and let U = {U j : j ∈ J } be an open cover of

X such that for each i ∈ J there is a homeomorphism ϕ j : U j → (0,1). The set {ϕ j : j ∈ J }
is called a chart for X; the existence of this set is a consequence of the definition of a
1-manifold.

Step 1. Let U j , U k ∈ U be such that U j \ U k, U j ∩ U k and U k \ U j are not empty. Then for
every component C of U j ∩ U k , ϕ j (C ) is one of (0, a), (b, 1), for some a, b ∈(0,1), and ϕ k (C )
is one of (0, a′ ), (b′ ,1), for some a′ , b′ ∈(0,1).

Proof of Step 1. Since U j ∩ U k is open and locally path connected (easy), every component
C is open in U j ∩ U k (Proposition 1 in 6.5), and hence, open in each of U j , U k . So, ϕ j (C )
and ϕ k (C ) are open subintervals of (0,1).

333

K12146_C014.indd 333 2/12/15 2:49 PM

 334   ◾  On Classifying Manifolds and Related Topics

Assume ϕ j (C ) = (c , d ) , 0 < c < d < 1. The points ϕ −j 1 (c ) and ϕ −j 1 (d ) are obviously in U j,

and none of them is in U k , or else one of [c, d), (c, d] would be a connected subset of U j ∩ U k
larger than C.
Since ϕ k (C ) is an open connected subinterval of (0,1), it has to be that ϕ k (C ) = (e , f ), for
some e , f , 0 ≤ e < f ≤ 1. Since U k \ U j ≠ ∅ (assumption), we have that U j ∩ U k ≠ U k , and
so either 0 < e or f < 1. We may assume the former (the latter case is symmetric). Then
ϕ −k 1 (e ) ∈U k and, arguing as above, ϕ −k 1 (e ) ∉U j .

{ }
∞
1 −1
The sequence ϕ −k1 (e + ) of elements in C converges to ϕ k (e ) ∈U k \ U j . Observe
n n=1
{ }
∞
1
also that, since ϕ j and ϕ k are homeomorphisms, the sequence ϕ j  ϕ −k1 (e + ) con-
n n=1
sists of pairwise distinct elements in ϕ j (C ) = (c , d ). So, by Corollary 2 in 7.4, this
sequence must have an accumulation point x ∈[c , d], and hence, it has a subsequence

{ }
∞
1 −1
{xn } of ϕ j  ϕ −k 1 (e + ) that converges to x. It follows that the sequence {ϕ j ( xn )},
n n =1
{ }
∞
1
which is a subsequence of ϕ −k1 (e + ) , converges to ϕ −j 1 ( x ) ∈U j . (At this point we
n n=1
have used the claim that ϕ −j 1 ([c , d]) is well defined; as we saw above, this is a consequence
of our assumption that 0 < c < d < 1 .) However, we already know that every subsequence

{ }
∞
1
of ϕ −k 1 (e + ) converges to ϕ −k 1 (e ) ∈U k \ U j . So ϕ −k 1 (e ) and ϕ −j 1 ( x ) are two distinct
n n =1
points of convergence of {ϕ −j 1 ( xn )}, contradicting the Hausdorff-ness of X Hence either
c = 0 or d = 1. This proves Step 1.

It follows from Step 1 that for every U j , U k ∈U , U j ∩ U k has at most two

com­p o­nents.

Step 2. If there exist U j , U k ∈ U such that U j ∩ U k has two components, then X is a

circle.

Proof of Step 2. Let U j , U k ∈ U be such that U j ∩ U k has two components. It follows that
X = U j ∪ U k (Exercise 2).

Denote the components of U j ∩ U k by C and D. By Step 1, we have that {ϕ j (C ), ϕ j (D )} =

{(0, a), (b,1)} and {ϕ k (C ), ϕ k (D )} = {(0, a′ ), (b′ ,1)} for some a, b, a′ , b′ ∈(0,1), a < b and a′ < b′.
We may assume that (ϕ j (C ), ϕ j (D )) = ((0, a), (b,1)) and that (ϕ k (C ), ϕ k (D )) = ((0, a′), (b′,1)).
It is easy to see that the only accumulation points in (U j )′ \ U j are ϕ −k 1 (a′ ) and ϕ −k 1 (b′ ). (Why?
Exercise 1.) Consequently, U j = U j ∪ {ϕ k−1 (a′ ), ϕ k−1 (b′ )} . We may now extend the homeo-
morphism ϕ j : U j → (0,1) to a homeomorphism ϕ j : U j → [0,1] by ϕ j (ϕ −k 1 (a′ )) = 0 and
j (ϕ −k 1 (b′ )) = 1. (Why is this a homeomorphism? Exercise 1.)
ϕ

K12146_C014.indd 334 2/12/15 2:49 PM

 14.1 1-Manifolds   ◾    335  

0 1

A
B

g
a´ b´

Illustration 14.1

In Illustration 14.1 we describe two embeddings: f :[0,1] → S1 and g :[a′ , b′] → S1 (where
f (0) = g (a′ ) = A and f (1) = g (b′ ) = B). We use them to define a homeomorphism
µ : X → S1 as follows:

 f ϕ j ( x ) for x ∈U j
 .
µ( x ) = 
 g  ϕ k ( x ) for x ∈U k \ U j

It is clear that μ is one-to-one and onto. Continuity follows from the gluing lemma. Finally,
Corollary 3 in 7.2 is now applicable, and it establishes that μ is a homeomorphism. This
completes the proof of Step 2.

Step 3. If, for some U j , U k ∈ U , U j ∩ U k is connected, then U j ∪ U k is homeomorphic to

(0, 1).

Proof of Step 3. Let U j , U k ∈ U be such that U j ∩ U k is connected. The claim is obvious if

U j \ U k or U k \ U j is empty. Otherwise, Step 1 applies, with C = U j ∩ U k . We will consider
the case when ϕ j (C ) = (b,1) and when ϕ k (C ) = (0, a′ ) (see Illustration 14.2). The other three
cases require similar or symmetric arguments.

Uj
Uk
j

0 1 κ

0 a´ 1

Illustration 14.2

K12146_C014.indd 335 2/12/15 2:49 PM

 336   ◾  On Classifying Manifolds and Related Topics

 ϕ j ( x ) if x ∈U j
As in Step 2, we have that U j = U j ∪ ϕ k−1 (a′ ), and ϕ  ( x ) =  is a
j
 1 if x = ϕ k −1 (a′ )
homeomorphism from U j onto (0,1]. On the other hand, the restriction of ϕ k to U k \ U j
is a homeomorphism onto [a′ ,1).
Let f : (0,1] → (0, 0.5] and g :[a′ ,1) → [0.5,1) be the linear homeomorphisms (defined
 t − a′ 
by f (t ) = 0.5t and g (t ) = 0.5 + 0.5   ; we do not need these formulas beyond the
 1 − a′ 
observation that f (1) = 0.5 and g (a′ ) = 0.5 ). Define µ : U j ∪ U k → I as follows:

  (x )
 f ϕ j if x ∈ U j
µ( x ) =  .
 g  ϕ k ( x ) if x ∈ U k \ U j

Then μ is a homeomorphism from U j ∪ U k onto (0, 1), completing the proof of Step 3.
Since X is compact, we can extract a finite subcover of U. Step 3 allows us to modify this
finite cover and the chart for X by taking the union of each pair satisfying the assumptions
of that step, one by one. After finitely many steps we end with two open sets as in Step 2,
hence establishing that X is homeomorphic to S1 .

Exercises
1. Answer the two why’s in the proof of Theorem 1.
2. Let X be a connected 1-manifold, and let U and V be two open subsets of the space
X such that each of them is homeomorphic to the interval (0,1) and such that U ∩ V
has two components. Show that X = U ∪ V .
3. Modify the proof of Theorem 1 to show that, up to a homeomorphism, the only
non-compact, connected 1-manifold is .
4. A 1-manifold with boundary is a Hausdorff, second countable space, such that each
point has a neighborhood homeomorphic to (0,1) or to [0,1). Show that, up to homeo-
morphism, the only 1-manifolds with boundaries which are not 1-manifolds are [0,1)
and [0,1]. (Hint: [23].)

14.2 Compact 2-Manifolds: Preliminary Results

A brief historical note: The modern definition of 2-manifolds appeared for the first time in
1913 in a paper by Hermann Weyl (1885–1955), The first proof that every 2-manifold can be
triangulated was published in 1925 by Tibor Radó (1895–1965).

In this section and in the next we will classify, up to homotopy, all connected compact
2-manifolds.

K12146_C014.indd 336 2/12/15 2:49 PM

 14.2 Compact 2-Manifolds: Preliminary Results   ◾    337  

We fix a triangle in  2, say T = {( x , y ): 0 ≤ x ≤ 1, 0 ≤ y ≤ − x + 1} (considered as a sub-

space of  2). The edges of T will be denoted by e1, e2 and e3 (Illustration 14.3).

(0,1)

e3 e2
T

(0,0) e1 (1,0)

Illustration 14.3 The triangle T.

Let X be a space and let f : T → X be an embedding. The subspace f (T) of X will be called a
curved triangle in X, or simply a triangle, when X is clear from the context; f (e1 ), f (e2 ) , f (e3 )
are the edges, or the curved edges of the triangle f (T), and f (v1 ) , f (v 2 ), f (v3 ) are its vertices.
A triangulation is a cover of X with triangles such that every two (curved) triangles are either
disjoint, or share precisely one vertex or precisely one edge. A trian­gulation is finite if it con-
tains finitely many triangles. For example, in Illustration 14.4 we see a finite triangulation of
a sphere.

Illustration 14.4

The foregoing definition is not particularly useful in the setup of general topological
spaces. For example, there are compact connected subspaces of  3 that allow triangu-
lation with infinitely many pairwise disjoint triangles. (Can you find one such space?)
However, it is a very natural concept in case of 2-manifolds—inasmuch as it was taken
for granted for several decades that every 2-manifold can be triangulated, and that every
compact 2-manifold possesses a finite triangulation. The first one who produced a proof
was Radó in 1925. It turned out—just as in the case of the Jordan curve theorem—that
the obvious was not easy at all. To date there does not exist a simple—a subjective term
indeed—self-contained proof.

Theorem 1. (Radó, 1925). Every compact 2-manifold possesses a finite triangulation.

K12146_C014.indd 337 2/12/15 2:49 PM

 338   ◾  On Classifying Manifolds and Related Topics

We will not prove it. A condensed proof of Radó’s theorem, based on a generalization
of Jordan-Schönflies theorem, can be found in [73]. In our central narrative regard-
ing the classification of compact, connected 2-manifolds, Theorem 1 is our starting
point.

Proposition 2. Let T be a finite triangulation of a 2-manifold. The following is true.

(a) Every edge is adjacent to exactly two triangles in T .
(b) The triangles in T adjacent to a fixed vertex can be arranged in a sequence T0 , T1 , …,Tn ,
such that the pairs of triangles Ti, Ti+1, i = 0,1, …, n − 1, and T0, Tn have precisely one
edge in common.

Proof.
(a) The alternative is that an edge is adjacent to more than two triangles. Take a point
in the linear interior of that edge. On one hand, since we are dealing with a 2-manifold,
there is a neighborhood of that point that is homeomorphic to the open disk D 2 .
On the other hand, since the edge is adjacent to more than two triangles, there is a
neighborhood of the point that is homeomorphic to finitely many open half-disks
sharing a common diagonal. The result now follows from Exercise 15 in 13.2.
(b) Denote the vertex by v, and start with any one of the (finitely many) triangles adja-
cent to v; call that triangle T0. Choose an edge e1 of T0 . By part (a) there is a unique tri-
angle T1 sharing e1 with T0. Denote the other edge of T1 that contains v by e2 . Repeat
the procedure with T1 and e2 : there is a unique triangle T2 sharing e2 with T1. If T2 = T0
, then go to the next paragraph. Otherwise keep going to end up with a pyramid C
made of finitely many triangles T0 , T1 , …,Tn , all adjacent to v, and each pair Ti, Ti+1,
i = 0,1, …, n − 1, and T0, Tn, have precisely one edge in common (Illustration 14.5).

v v

ei ei+1 ei+1

Ti–1 Ti+1
Ti Ti T

Illustration 14.5 Illustration 14.6

K12146_C014.indd 338 2/12/15 2:49 PM

 14.2 Compact 2-Manifolds: Preliminary Results   ◾    339  

We now show that the above procedure must exhaust all triangles adjacent to v. There are
two cases to eliminate:

(i) There is a triangle T different from T0 , T1 , …,Tn , and adjacent to an edge in C contain-
ing v (Illustration 14.6). This contradicts part (a).
(ii) There is a triangle T different from T0 , T1 , …,Tn , and adjacent to v. We repeat the proce-
dure in the first paragraph, starting with T and ending up with another cone sharing v
with C. Since the triangulation is finite, and because of part (a), we must eventually end
with the situation depicted in Illustration 14.7. In that case the point v will not have a
neighborhood homeomorphic to the open disk, contradicting the assumption that the
underlying space is a 2-manifold. (The last claim is a simple exercise in connectedness.)

Illustration 14.7

Recall that, say, the torus is a quotient space of a rectangle as described in Section 4.4. The follow-
ing proposition tells us that every compact 2-manifold can be obtained in a similar way.

Proposition 3. Every compact connected 2-manifold X is homeomorphic to the quotient

space of a filled polygon with 2n edges, where the edges are identified in pairs.

Proof. By Theorem 1, there is a finite triangulation for X. For every triangle Tj in the tri-
angulation choose a triangle Tj∗ in  2 and a homeomorphism f j : Tj∗ → Tj sending edges to
edges and vertices to vertices. We will construct our polygon from the planar triangles Tj∗.
Choose any T1∗, and let e1 be one of its three edges. By Proposition 2(a), the edge f1 (e1 ) is
common to precisely two triangles in the triangulation: one of them is T1 . Denote the other
by T2 . Identify e1 in T1∗ with the edge f 2−1 ( f1 (e1 )) in T2∗. (Formally, we identify along the
restriction of f 2−1  f1 to the edge e1 to get P2 = T1∗ ∪ f −1  f T2∗.) If all of the edges in the result-
2 1
ing quadrilateral P2 are to be identified in pairs, then the procedure stops. Otherwise,
continue recursively, always identifying the new triangle Tn∗+1 to precisely one edge in the
previously obtained polygon Pn. After each step the quotient space of the polygon Pn ,
obtained by identifying some of the pairs of edges on its boundary that correspond to a
common edge of a pair of triangles in T1 , T2 , … , Tn from the triangulation of X, is homeo-
morphic to the subspace T1 ∪ T2 ∪ … ∪ Tn of X (see Exercise 13 in 4.3).

K12146_C014.indd 339 2/12/15 2:49 PM

 340   ◾  On Classifying Manifolds and Related Topics

After finitely many steps we get a polygon Pm with all edges on its boundary to be identified
in pairs, yielding a subspace Y of X. We claim that Y = X . Otherwise, there must be a triangle
in the triangulation of X sharing only one vertex v with Y. All of the triangles adjacent to v
would then yield the situation depicted as in Illustration 14.7, which we saw is not possible by
Proposition 2(b).

We will use the term identifying polygon for a filled polygon with an even number of edges
with pairs of edges to be identified. The 2-manifold that arises from identifying polygons can
be uniquely encoded with a cyclic word (= word written circularly), describing which edges and
in which direction they are to be identified. Because of technical reasons we will not write the
cyclic words circularly but rather linearly. For example, the cyclic word aba −1cbc encodes the
following polygon (Illustration 14.8) with edges to be identified in pairs as indicated, traversing
them clockwise.
a
b

c
a

b
c

Illustration 14.8

The edges oriented opposite to the clockwise direction come with a negative exponent
in the encoding word. Such is one of the two a-edges in Illustration 14.8.
The road in front of us is clear: we want to see which kinds of 2-manifold arise from iden-
tifying polygons. Along the way we will discover that every identifying polygon, no matter
how we identify the edges in pairs, gives rise to a compact, connected, 2-manifold.

Example 1
We have already encountered a few compact 2-manifolds as quotient spaces of iden-
tifying polygons. For example, the sphere is the quotient space of a polygon with two
edges and with aa −1 as the associated cyclic word (see Example 7 in 4.3). The torus was
seen as a quotient space of a rectangle with the identification according to aba −1b −1
(Example 4 in 4.4); and the Klein bottle corresponds to aba −1b (Example 6 in 4.4). The
projective plane (introduced and depicted in Section 7.3) is a rectangle with edges
identified according to abab (Illustration 14.9).
b

a a

Illustration 14.9 The projective plane.

K12146_C014.indd 340 2/12/15 2:49 PM

 14.2 Compact 2-Manifolds: Preliminary Results   ◾    341  

Observe that, by considering ab to be one edge (denoted c), the projective plane is also
associated to the identifying polygon with two edges, identified according to the cyclic
word cc. ☐

Example 2: Connected Sums of 2-Manifolds

Start with any identifying polygon P,1 giving rise to a compact 2-manifold M1. Taking
an open disk out of the interior of P1 leaves boundary points along a circle C1 in P1 (and
a homeomorphic copy of C1 in M1; see Illustrations 14.10 and 14.11 where we illustrate
this when M1 is a torus).

a a

Illustration 14.10 Illustration 14.11

Each of these boundary points in C1 has a neighborhood that is homeomorphic to

{ }
the half disk D = ( x , y ) ∈ 2 : x 2 + y 2 < 1, y ≥ 0 .
Doing the same procedure to another identifying polygon P2 and the associated
manifold M 2 yields another punctured manifold. Identify the associated boundary
circles C1 and C2 along the ‘identity’ homeomorphism. The fact that the points in
both C1 and C2 have neighborhoods homeomorphic to the half disk D implies that
after the identification the corresponding points have neighborhoods homeomorphic
to an open disk. Since this is already given for the other points in the two punctured
2-manifold, the resulting quotient space is again a manifold. We call it the connected
sum of M1 and M 2, and we denote it by M1 # M 2. An Illustration of M1 # M 2 when both
M1 and M 2 are tori is given in Illustration 14.12.

Illustration 14.12

Note that if we do the identification of C1 and C2 at the level of identifying poly-

gons, then we do not get an identifying polygon. This is a small problem, which will

K12146_C014.indd 341 2/12/15 2:49 PM

 342   ◾  On Classifying Manifolds and Related Topics

be rectified by carefully choosing a disk-shaped hole within the identifying polygon.

We will do that in the next section.
The choice of the position of the open disk in the interior of the identifying poly-
gons does not affect the homeomorphic class of the resulting connected sums. Also
note that the operation # is associative, up to a homeomorphism. This allows us to
write M1 # M 2 # # Mn without worrying about placing parentheses. ☐

Exercises
1. (a) Considering the Klein bottle K as the quotient space of rectangles with edges
identified in pairs as described earlier, exhibit a triangulation of K.
(b) The projective plane P2 can be viewed as the quotient space of the rectangle as in
Illustration 14.9 above. Why is the subdivision of the P2 into region as shown in
Illustration 14.13 not a triangulation of the P2? Draw a triangulation of P2 on its
representation in Illustration 14.9.
b

a a

Illustration 14.13 The projective plane again.

2. Consider the circle C = {( x , y ,0) ∈ 3 : x 2 + ( y − 1)2 = 1} . Rotate C around the x-axes

through an angle θ, and at the same time rotate it around its diameter containing the
origin, through the angle θ2 . In order to avoid self-intersection perform the rotation in
 4 (for example, by setting the fourth coordinate of the points of the circle obtained
after rotating through θ around the x-axis to be θ sin θ). Let X be the 2-manifold
(depicted in Illustration 14.14) of all points attained by C as θ ranges from 0 to 2π .
Which of the spaces mentioned so far is homeomorphic to X? Justify.

Illustration 14.14 We rotate the circle as

indicated. Which space do we get?

K12146_C014.indd 342 2/12/15 2:50 PM

 14.3 Compact 2-Manifolds: Classification   ◾    343  

3. Describe two triangulations Tr = {T1 ,T2 ,… ,Tm } and Tr ′ = {T1′ ,T2′ ,… ,Tn′ } of the
sphere, such that Tr ∪Tr ′ cannot be extended to a triangulation of the sphere by
subdividing the regions of type Ti ∩ Tj′ into curved triangles.
4. Show that every triangulation of a compact 2-manifold must necessarily be finite.
5. The Euler characteristic e ( X ) of a compact 2-manifold X is # t − # e + # v , where #t
is the number of curved triangles, #e is the number of curved edges, and #v is the
number of vertices in some triangulation of X. It is true that e( X ) does not depend
on the triangulation. (Not obvious, precisely because of the problem identified in
2
Exercise 3.) Find the Euler characteristic of the connected sum T  # T 2
# #T 2
 of n
2 2 2 n
P 
tori, n ≥ 1, and of the connected sum  P
# #
 P of n projective planes, n ≥ 1.
#
n
6. A 2-manifold with a boundary is a second countable Hausdorff space such that
every point has an open neighborhood homeomorphic either to  2 or to the upper
half-plane {( x , y ) ∈ 2 : y ≥ 0}. Prove that if every compact, connected 2-manifold
with a boundary can be triangulated, then Radó’s theorem (Theorem 1) is true.

14.3 Compact 2-Manifolds: Classification

A brief historical note: The first rigorous classification of compact 2-manifolds, except for
Rado’s theorem, was established jointly in 1907 by Max Dehn and Poul Heegaard.

We are ready to state and prove the main theorem in this chapter.

Theorem 1. (The Classification Theorem). Every compact, connected 2-manifold is homeo-

morphic to a sphere, to a connected sum of finitely many tori, or to a connected sum of
finitely many projective planes.
Illustration 14.15 visualizes the possibilities stated in Theorem 1.

...

...

Illustration 14.15 Note the ellipses to the right: the procedure continues. The projective plane
is not embeddable in  3; hence, we see self-intersections where such should not be.

K12146_C014.indd 343 2/12/15 2:50 PM

 344   ◾  On Classifying Manifolds and Related Topics

Proof.
Our starting point is Proposition 3 in 14.2, allowing us to represent every compact,
connected 2-manifold as the quotient space of an identifying polygon. We will carefully
modify this polygon, without changing the underlying 2-manifold, ending up with an
identifying polygon that clearly gives rise to one of the manifolds listed in the theorem.
We begin by classifying the edges in the identifying polygons into two types: the edges
of the first kind 1 are those that have the same label and are of opposite cyclic orientation
(i.e., those whose labels come with different exponents in the associated cyclic word); the
edges that have the same label and have the same cyclic orientation are edges of the second
kind. For example, in the identifying polygon giving rise to the Klein bottle (Illustration
4.26), the a-labeled edges are of the first kind, the two b-labeled edges are of the second
kind (the associated cyclic word is aba −1b). We note in passing that the Klein bottle is not
explicitly mentioned in the statement of Theorem 1—but, as we will see, it is there.

Step 1. (2-gons)
If the identifying polygon has two edges, then the associated quotient space is either a
sphere (cyclic word aa −1), or it is a projective plane (cyclic word aa).
We now consider identifying polygons with at least four edges.

Step 2. (Eliminating spheres)

If the identifying polygon contains two adjacent edges of the first kind, then we can
eliminate them by simply identifying them (Illustration 14.16). We assume that this step is
done every time there is such a pair of edges, so that we are left with an identifying polygon
with no such edges, or with a polygon with two edges as in Step 1.
a a
a

P
P´

Illustration 14.16 Eliminating adjacent edges of the first kind.

Step 3. (Grouping edges of the second kind)

(i) In Illustration 14.17 we show how to ensure that every two edges of the second kind
and with the same label are adjacent.
a
c c
c
c
=
a a
c

Illustration 14.17 Making the edges of the second kind adjacent.

1
The terminology follows Massey’s book [48]. Other authors use different names for the two classes of edges.

K12146_C014.indd 344 2/12/15 2:50 PM

 14.3 Compact 2-Manifolds: Classification   ◾    345  

As we see in Illustration 14.17, first we cut along c, temporarily separating the polygon
into two components, then identify the edges labeled a, bringing the two components
back together and producing another identifying polygon corresponding to the same
underlying 2-manifold. This cut-and-paste technique will be used in the subsequent steps.
Following this step we may assume that the edges with the same label and of the second
kind are adjacent. If new non-adjacent pairs of edges of the second kind emerge in the
ongoing procedure we back up and apply this step before we proceed.

(ii) Consider two pairs of edges of second kind, grouped according to the preceding step.
Now we show how to make any such two pairs of edges adjacent in the identifying poly-
gon (see Illustration 14.18). Observe in this illustration that the a-labeled edges at the
starting polygon are of the second kind, and that the b-labeled edges are of the first kind.
a b b b
c c c
a c
c a
A A A
d
=
b b b

B B

d
c c d

A
d
=
b
d c

B c
B
A

Illustration 14.18 Moving pairs of edges of second kind closer together.

The net effect of this step is a kind of commutativity of pairs of edges of the second
kind with certain edges of the first kind; comparing with the starting polygon, observe
that one of the c-labeled edges of the first kind in the last frame is now on the other side
of the pair of d-labeled edges of the second kind.
Also observe that the mutual relationship of the edges in each of the dotted parts
denoted by A and B in Illustration 14.18 are not disturbed by these moves. However,
some of the pairs of edges of first kind, one edge in A the other in B, may have become
pairs of edges of the second kind (compare the arrows of A and B in the first and last
frame of Illustration 14.18). In this case we alternate steps 3(i) and 3(ii) until all edges of
the second kind are adjacent, and until all pairs of edges of the second kind are adjacent.
After performing Step 3 as many times as possible we end up with an identifying polygon
where all of the edges of the second kind, if any, form a connected subset of the circular bound-
ary of the polygon, and the rest of the boundary is covered by the edges of the first kind.

Step 4. (Carefully grouping edges of the first kind). Assume that there are edges of the first
kind in the polygon we get after Step 3. By Steps 1 and 3, there must be at least four such edges,

K12146_C014.indd 345 2/12/15 2:50 PM

 346   ◾  On Classifying Manifolds and Related Topics

and there is a pair of adjacent edges of the first kind with different labels. Our immediate goal
is to change the identifying polygon (without affecting the underlying manifold) so that we
end up with edges of the first kind arranged according to the word cdc −1d −1 (without spoiling
what we have done so far). We do it in two subcases.

(i) Four edges of the first kind are positioned as in the first (upper left) frame of
Illustration 14.19.
b
a a
a c
c
c
a a d
a
=
b c

b
c

c d
c
d c

d a d

c
=

Illustration 14.19 Bringing together quadruples of edges of the first kind.

Observe that the arrangement of the edges in the part of the boundary of the polygon
indicated by dashed lines is not affected by this step. Consequently, we have not spoiled
anything we have achieved in the first 3 steps. Note that in the last polygon we have
four edges of the first kind coming in an arrangement corresponding clockwise to the
−1
word d −1cdc −1, or, after replacing the label d by f , fcf −1c −1, which is what we wanted to
achieve.

(ii) In the starting polygon, the labels a and b of the two adjacent edges are interchanged
(Illustration 14.20, left).
a a
b c
A b
A

a B a
B

C C
b b

Illustration 14.20

This can be reduced to the preceding case as follows. The parts of the boundary polygon
corresponding to A and B must contain some edges; else we are violating Step 2. Step 3 then
guarantees that at least one of A and B contains only edges of the first kind. By symmetry,

K12146_C014.indd 346 2/12/15 2:50 PM

 14.3 Compact 2-Manifolds: Classification   ◾    347  

we may suppose that A contains only such edges. Label the edge in A following the a-edge
in the clockwise direction by c (see Illustration 14.20, right). If the other c-edge is in A, then
repeat this procedure to the edges labeled a and c; after finitely many such steps we end up
with four edges as at the beginning of Step 4 (i). On the other hand, the other c-edge is not
in A, then we are immediately within Step 4 (i).
Assume that the previous steps have been completed, resulting in groups of quadruples
of edges of the first kind and pairs of edges of the second kind. Notice that all of the vertices
in such groups of edges are identified in a single vertex.

(iii) All the edges of the first kind are grouped in quadruples except for an extra pair of edges
of the first kind, resulting in the identifying polygon depicted in Illustration 14.21.
c Y
X
B

A
c W
Z

Illustration 14.21
Since the identifications happen only among pairs of edges in the part A or pairs of edges
in the part B (see Illustration 14.21), it follows that the vertex X is not identified with
the vertex Y, and the vertex Z is not identified with the vertex W in the quotient space.
Hence the rectangle XYZW in Illustration 14.21 becomes a cylinder in the quotient space
(Illustration 14.22, left frame). This cylinder is homotopically equivalent to a circle, and
the homotopy (shown in Illustration 14.22) does not affect the rest of the space.
B B
c B
c
A A A

Illustration 14.22

This allows us to simply delete such pairs of extra edges of the first kind, without affecting
the resulting quotient space (up to homotopy equivalence) and leaving only quadruples of
edges of the first kind grouped as per steps 4 (i) and (ii), if there are edges of the first kind
at all (see Illustration 14.23).
X

B
B

A = A

Illustration 14.23 It follows from the remark at the end of Step 4 (ii) that the vertices X and Z
are identified in the quotient space.
(An alternative argument is suggested in Exercise 4.)

K12146_C014.indd 347 2/12/15 2:50 PM

 348   ◾  On Classifying Manifolds and Related Topics

We summarize what has been achieved so far: we showed in Steps 1–4 that we can replace
the starting identifying polygon with one where the edges of the first and the second kind
occupy two components of the boundary of the polygon, respectively, where the edges of the
second kind and of the same label are adjacent, and where the edges of the first kind come in
quadruples corresponding to words of type aba −1b −1. All this is summarized in Illustration
14.24.
x1 x1
bm
am

bm
xn
am
xn

a1

b1
b1 a1

Illustration 14.24

Step 5. (Completing the procedure) We will now see that if there is a single pair of edges of
the second kind in the identifying polygon after completing Steps 1 to 4, then we can replace

a d b
b c
a d b
c a d
d d e
b
b b
=
c c c

c c
d d e
e d
e
f f d
b f
e = d
c e
d c c
e

f
f g f
d g
f f
g d g
d g

e e
= e
=
e e
e

Illustration 14.25

K12146_C014.indd 348 2/12/15 2:50 PM

 14.3 Compact 2-Manifolds: Classification   ◾    349  

the identifying polygon with another one (without affecting the underlying 2-manifold)
containing only edges of the second kind. The justification is in Illustration 14.25: specifically,
we show that edges corresponding to aabcb −1c −1 could be changed—without affecting
the mutual position of the other edges on the boundary—to 6 edges of the second kind
corresponding to f −1f −1gge−1e−1. (An alternative approach is indicated in Exercise 8.)
With this we showed that the identifying polygon can be reduced to one containing only
edges of the first kind as in Illustration 14.26 or to one containing only edges of the second
kind as in Illustration 14.27.
a1
bm b1

am a1 x1
x1

bm b1
xn

am
xn

Illustration 14.26 Illustration 14.27

We end the proof of Theorem 1 by showing that identifying the edges of the polygon in
Illustration 14.26 according to the labels gives rise to a connected sum of tori, while the result
of doing the same to the identifying polygon in Illustration 14.27 yields a connected sum of
projective planes. This is shown in Illustrations 14.28–14.29 and 14.30–14.31, respectively.
b
b c
c
a a = a
a

b b

Illustration 14.28 A torus with a hole: the two end-vertices of the c-edge in the pentagon to the
right are forcibly identified through the identification of the edges corresponding to aba −1b −1 .

b1 a2
a1 b2

b1 a2
a1 b2

Illustration 14.29 A connected sum of two tori, realized as an identifying polygon.

In Illustrations 14.28 and 14.29 and the associated captions we show that the identifying
polygon in Illustration 14.26 is indeed a connected sum of m tori.

K12146_C014.indd 349 2/12/15 2:50 PM

 350   ◾  On Classifying Manifolds and Related Topics

x c
c =

x x

Illustration 14.30 A projective plane with a hole: the two end vertices of the c-edge in the
triangle to the right are identified through the identification of the x-edges.

x y

x y

Illustration 14.31 A connected sum of two projective planes, realized as an identifying polygon.

Illustrations 14.30 and 14.31 and the captions imply that the identifying polygon shown in
Illustration 14.27 is a connected sum of n many projective planes.

An orientation of a triangle T is a mapping oT : {T } → {−1,1}. One may visualize the orien-

tation as a set of arrows along the boundary of the triangle, pointing (say) clockwise if oT (T ) = 1,
counter-clockwise if oT (T ) = −1. A triangulation of a 2-manifold is orientable if there is an
orientation of the triangles in the triangulation such that for every two triangles T1 and T2
sharing a common edge, we have oT1 (T1 ) = oT2 (T2 ). For connected 2-manifolds, this reduces
to the claim that oT1 (T1 ) = oT2 (T2 ) for every two triangles in the triangulation. In terms of
arrows, this means that the orientations of the boundaries of the triangles are opposite along
the common edge (Illustration 14.32). A compact, connected 2-manifold is orientable if every
triangulation of that manifold is orientable. Otherwise the 2-manifold is non-orientable.

–1 1

–1 –1

Illustration 14.32 Left: two triangles with matching orientations (the arrows go in opposite
directions at the common edge). Right: two triangles with distinct orientations.

It is true that if one triangulation of a compact, connected 2-manifold X is orientable,

then so is every other. Similarly, if one triangulation of X is non-orientable, then so is

K12146_C014.indd 350 2/12/15 2:50 PM

 14.3 Compact 2-Manifolds: Classification   ◾    351  

every other. We will not prove these two statements ([55], page 163). It follows that (non)
orientability is an intrinsic property of the associated 2-manifold, that is, it does not
depend on a specific triangulation.
It is easy to exhibit orientable triangulations for S 2 and for connected sums of tori.
Consequently, these are orientable manifolds. Observe that each of these spaces has two
orientations.
It is also evident—through raw, straightforward enumeration of all possible orientations—
that the triangulation of the Möbius band as shown in Illustration 14.33 is non-orientable.
Since the Möbius band is a subspace of connected sums of projective planes, it follows that
the latter spaces are all non-orientable.
v4

v1
v3

v6

v2

v5

Illustration 14.33 A triangulation of a Möbius band: the vertices are labeled; some of the
edges of the triangles are along the boundary-curve of the band. Notice that all pairs of verti-
ces are joined by a simple curve in this way (including v5 and v6! ). However, not every three
vertices are corners of a curved triangle. Can you find one such triple?

Turning our attention back to Theorem 1, there is one obvious question that deserves atten-
tion: is the provided classification efficient? Is it possible that some pairs of 2-manifolds listed
in that theorem are homotopically equivalent? In order to answer that question we will first
compute the fundamental groups of these 2-manifolds; this can now be done in one fell swoop.

Example 1: The Fundamental Groups of Compact 2-Manifolds

Spheres are simply connected. We pay attention to connected sums of tori and to con-
nected sums of projective planes.
Start with any identifying polygon representing a 2-manifold X (a connected
sum of tori as in Illustration 14.34, or a connected sum of projective planes as in
Illustration 14.35). Split X into two open sets as in Illustrations 14.34 and 14.35.
The open set U is the part of the identifying polygon outside the smaller circle; V is
inside the larger circle; the intersection U ∩ V is the ring as shown. We choose a point
y ∈U ∩ V .
Evidently π1 (V , y ) is trivial, and π1 (U ∩ V , y ) = 〈α ; ∅〉, where α is the (homotopy
class) of a loop that winds once around the ring U ∩ V . Hence, α = 1 in π1 (V , y ).

K12146_C014.indd 351 2/12/15 2:50 PM

 352   ◾  On Classifying Manifolds and Related Topics

a1 x1
bm b1 x1

am a1

xn
bm b1

U V U V
am
xn

Illustration 14.34 Illustration 14.35

In order to compute π1 (U , y ) first observe that the vertices in the boundary graph
of the identifying rectangle for the torus (aba−1b−1 ) are identified to a single vertex
in the manifold X. Consequently, the same is true for the identifying polygon corre-
sponding to a connected sum of tori. The corresponding claim is even plainer for the
identifying polygon for a connected sum of projective planes.
The boundary of the identifying polygon is a deformation retract of U. After
identifying the pairs of edges, the boundary becomes a rose of circles, one per
every label (2m circles for Illustration 14.34, n circles for Illustration 14.35). Hence,
π1 (U , y ) is a free group on the set of labels of the edges. The loop α is obviously
homotopic within U to the loop winding once around the boundary. This means that
α = a1b1a1−1b1−1 … ambmam −1bm −1 in the case of Illustration 14.34, and α = x1 x1 … xn xn in
the case shown in Illustration 14.35.
Consequently, by the SvK theorem we have the following:
Connected sum of m tori: π1 ( X , y ) = 〈a1 , b1 , a2 , b2 ,… , am , bm ; a1b1a1−1b1−1  ambm am−1bm−1 = 1〉.
Connected sum of n projective planes: π1 ( X , y ) = 〈x1 , x 2 ,… , xn ; x12 x 22  xn2 = 1〉 . ☐

The efficiency of the classification theorem (Theorem 1) is established by the following theorem.

Theorem 2. Let T1 ,T2 ,… be copies of a torus and P1 , P2 ,… be copies of a projective

plane. If X and Y are any two distinct elements of the set M = {S 2 } ∪ {T1 # T2 # 
# Tm : m ∈+ } ∪ {P1 # P2 #  # Pn : n ∈+ }, then X and Y are not homotopically equivalent.

Proof. It follows from the argument and the conclusion of Example 6 in 12.2 that the funda-
mental groups of X and Y are not isomorphic, and the statement of Theorem 2 follows.

Exercises
1. The edges of the filled polygons shown in Illustrations 14.36 and 14.37 are to be
identified in pairs according to their orientation and according to their labels. The
resulting two spaces are compact, connected 2-manifold. Recognize each of them
as one of the manifolds in the classification theorem. Justify your answer.

K12146_C014.indd 352 2/12/15 2:50 PM

 14.3 Compact 2-Manifolds: Classification   ◾    353  

a a

c b c b

c c
b b

a a

Illustration 14.36 Illustration 14.37

2. The two filled polygons depicted in Illustration 14.38 and 14.39, with pairs of edges
to be identified as shown, define two 2-manifolds. Recognize them (as manifolds
in the classification theorem).

c c
a b a b

d d
d d

a a
c c
b b

Illustration 14.38 Illustration 14.39

3. (a) W
 hich 2-manifold is represented by a filled hexagon with the edges identified in
pairs according to the word a1a2a3a1a2−1a3−1 (Illustration 14.40)?
(b) Which 2-manifold is represented by a filled 2n-gon with the edges identified
in pairs according to the word a1a2 … an −1an a1a2−1 … an−−11 an−1 (Illustration 14.41)?
Justify your answer.

a1
a1
an a2
a2

a3

a3

a2 an
a2
a1 a1

Illustration 14.40 Illustration 14.41

K12146_C014.indd 353 2/12/15 2:50 PM

 354   ◾  On Classifying Manifolds and Related Topics

4. The identifying polygon is shown in Illustration 14.42. Assume that the edges of the
second kind are grouped in pairs, and that the edges of the first kind, other than
the two edges labeled c, are grouped in quadruples. Show that the resulting quotient
space is X # Y, where X is the 2-manifold obtained by identifying the edges in the part
A, and where Y is the 2-manifold obtained by identifying the edges in the part B.

c
B

A
c

Illustration 14.42

5. Let X be the 2-manifolds obtained by identifying the edges of the planar object shown
in the Illustration 14.43 as indicated by the labels, and let Y be the 2-manifold obtained
from a filled hexagon with edges identified in pairs according to the word aacbbc −1.
Show that X and Y are homeomorphic. Find the 2-manifold listed in Theorem 1 that is
homeomorphic to both X and Y.
a

Illustration 14.43
6. The solid Klein bottle is the quotient space of D 2 × I made by identifying ((a,b), 0) with
((a, −b),1), for every (a, b) ∈D 2. Show that the solid Klein bottle is not homeomorphic
to the solid torus D 2 × S1, and that these two are homotopically equivalent.
7. Show that P 2 # P 2 is a Klein bottle.
8. Complete Step 5 in the proof of Theorem 1 as follows: start by cutting and pasting as
indicated in Illustration 14.18, use the fact that edges that are both in parts A and B
and have the same label change their kind, and do a couple of more cut-paste moves.
9. (a) Choose two distinct points on the boundary of a Möbius band, thus splitting
the boundary into two closed arcs. How can you glue these two arcs so that the
resulting space is a projective plane?
(b) Starting with the space M # T 2 as shown in Illustration 14.44 (a connected sum
of a Möbius bend and a torus; note that the c-labeled circles are to be identified),
draw a few intermediate frames of an animation showing that it is homotopi-

K12146_C014.indd 354 2/12/15 2:50 PM

 14.3 Compact 2-Manifolds: Classification   ◾    355  

cally equivalent to M # K, where K is a Klein bottle (Illustration 14.45: note the

c-labeled circles are identified differently). Deduce that P 2 # T 2 ≅ P 2 # P 2 # P 2.

c c

c
c

Illustration 14.44 A connected sum of Illustration 14.45 A connected sum of a

a projective plane and a torus: the pro- projective plane and a Klein bottle; notice
jective plane is represented here as a the reversal of the orientation of one of
Möbius band, with its boundary to be the circles labeled by c.
identified. The torus is a cylinder with the
boundary circles identified as indicated.

10. If T 2 is a torus, K is a Klein bottle, and P 2 is a projective plane, show that T 2 × K and
T 2 × P 2 are not homeomorphic.
11. Use the SvK theorem to compute the fundamental group of the 2-manifold depicted in
Illustration 14.46. Which of the 2 manifolds listed in Theorem 1 is homeomorphic to this
space?

Illustration 14.46 A triple Klein bottle (by Jos Leys). The surface is spanned by the rings (so,
ignore the discontinuities), and, of course, there are no self-intersections.

K12146_C014.indd 355 2/12/15 2:50 PM

 356   ◾  On Classifying Manifolds and Related Topics

12. Find a presentation for the space depicted in Illustration 14.47 (two Klein bottles and
a circle touching both of them).

Illustration 14.47

14.4 Regarding Classification of CW-Complexes

and Higher Dimensional Manifolds

A brief historical note: The results mentioned in this section were obtained by Hellmuth Kneser
(1898–1973), Edwin E. Moise (1918–1998), Stephen Smale (born 1930), John Milnor (born 1931),
Michael Freedman (born 1951), and Grigori Perelman (born 1966).

The first subsection of this section, as well as parts of the next one, will be presented in a
somewhat informal manner. The purpose is not to fully justify the claims that we state, but
rather to trace the boundaries of the scope of our introductory material and to stimulate the
interest for the subject.

Compact, Connected 3-Manifolds: A Brief Overview

Unless otherwise stated, we assume here that the 3-manifolds we discuss in this subsec-
tion are compact and connected. We will give the bare bones of the classification problem
for such manifolds: we will state a couple of older theorems and note the recent affirma-
tive solution of the classification problem for 3-manifolds, using the occasion to exhibit a
menagerie of some interesting, good-looking, and important 3-manifolds.
The simplest compact 3-manifold is the 3-sphere S 3 = {( x , y , z , u) ∈ 4 : x 2 + y 2 + z 2 + u 2 = 1}.
The 3-sphere is a simply connected space (Corollary 6 in 13.1). It is homeomorphic to
the quotient space of the solid 3-ball D3, along the homeomorphism that sends each
point ( x , y , z ) ∈S 2 to the point ( x , y , − z ). The process is depicted in Illustration 14.48.
Bear in mind here and in the other attempts to visualize 3-manifolds that there is not
a shred of hope to faithfully depict any of them in  3. More precisely, no 3-manifold
we consider here is embeddable in  3. We will only justify the non-embeddability of
the 3-sphere S 3 in  3; it is a consequence of the Borsuk–Ulam theorem, Section 16.5.

K12146_C014.indd 356 2/12/15 2:50 PM

 14.4 Regarding Classification of CW-Complexes and Higher Dimensional Manifolds   ◾    357  

Illustration 14.48

On the left side of Illustration 14.48 we see a closed 3-ball cut with planes as shown. If we
identify the boundary circle in each cross section, as described in Example 7, Section 4.3,
each cross section becomes a 2-sphere. The common intersection of these spheres is the circle
C obtained by identifying the endpoints A and B of the line segment shown in the figure to
the right. The “pumpkin” to the right represents the resulting 3-sphere—there is not enough
space in  3 (even less in the drawing plane  2) to show that the spheres without C are pair-
wise disjoint. The points A and B In Illustration 14.48 are yet to be identified to get C and S 3.
It turned out that, up to homotopy, S 3 is the only compact, connected, and simply con-
nected 3-manifold, affirming the long-standing Poincaré conjecture. A very long chain of
difficult theorems led to the confirmation of that conjecture; it culminated with the results
by Grigori Perelman (2002, 2003; see, for example, [56]).
Another basic 3-manifold is the projective 3-space P 3 . It could be obtained from a closed
3-ball by identifying the antipodal points on its boundary. A depiction of P 3 is provided in
Illustration 14.49.
It has now been confirmed that the so-called Hamilton–Perelman program leads to
a justification of an expected classification of all compact and connected 3-manifolds.
Merely describing the list of such 3-manifolds exceeds the scope of this book. We will be
satisfied with a rather cursory overview.

Illustration 14.49 The ball to the left is cut with planes: each of these crosscuts becomes the pro-
jective plane P2 after the identification of the antipodal boundary points. The projective planes
are positioned symmetrically around the circle C in P3, where C is their common intersection,
as in Illustration 14.48. The result is depicted to the right, where we make one of the projective
planes opaque to see it more clearly. The self-intersections are again due to the fact that P3 is
not embeddable in  3.

K12146_C014.indd 357 2/12/15 2:50 PM

 358   ◾  On Classifying Manifolds and Related Topics

A promising theorem was established in 1952 by Moise: every compact 3-manifold X

allows a finite triangulation. In this case, “a finite triangulation” means a cover of X made
of finitely many homeomorphic copies of a (Euclidean) tetrahedron T, so that each pair of
copies of T adjacent in X share either a face, an edge, or a vertex.
It follows from Moise’s theorem that every compact, connected 3-manifold is the quo-
tient space of a solid polyhedron with faces identified in pairs. One example is shown in
Illustration 14.50, where we show a solid cube with the opposite faces to be identified so
that the edges of the cube match according to labels and orientation. Pay attention to the
edges labeled d and e. The resulting space is K × S1, where K is a Klein bottle.

e
c

c
e
a a
a
a
d b
b
d

Illustration 14.50 Klein × S1. Illustration 14.51 K × {x}.

The copies of Klein bottles that we see in Illustration 14.51 are K × {x } for various x ∈S1.
Each two such Klein bottles are disjoint in K × S1, and so they should be viewed as being
disjoint in Illustration 14.51. Each point in K × S1 has a standard basis neighborhood of
type D × J , where D is a copy of D2 and J is an interval within S1. In particular, there are
no boundary points, despite what appears in the depiction. We make one slice of the
space gradually more transparent, and one K × {x } opaque, so that we can take a better
peek at the structure of the space.
In an analogy to the 2-dimensional case, one avenue of attacking the problem of classify-
ing 3-manifolds is through analyzing connected sums. A connected sum M1 # M 2 of 3-mani-
folds M1 and M 2 can be defined analogously to the case of 2-manifolds: take out an open ball
from each of the 2-manifolds, then identify the spherical boundaries via a homeomorphism.
Two obstacles appear at this early stage: first, the spherical boundaries of balls in the
3-dimensional case can be pathologically complicated, as was the case with Alexander’s
horned sphere. One way to avoid such cases is to impose smoothness (differentiability) of
the manifolds. We will not get into technical aspects of this notion.
Another issue that also needs attention is the type of homeomorphism that is used when
we identify the boundary 2-spheres of the deleted balls: if we orient the spheres, then a homeo-
morphism between them may preserve, or may reverse, the orientation. In the construction
of M1 # M 2 it is usually assumed that this homeomorphism is orientation reversing. Again, we
refrain from going further and discussing the case of orientation-preserving homeomorphism
in the construction of the connected sums.

K12146_C014.indd 358 2/12/15 2:50 PM

 14.4 Regarding Classification of CW-Complexes and Higher Dimensional Manifolds   ◾    359  

If a 3-manifold X is such that if X = M1 # M 2 for some 3-manifolds M1 and M 2, neither

homeomorphic to the 3-sphere S 3, then we say that X is not prime. Otherwise, X is a prime.
The following result is due to Kneser (1928).

Theorem 1. (Kneser’s Theorem). Every compact, orientable 3-manifold X is a connected

sum M1 # M 2 # # Mn of prime 3-manifolds, and this decomposition is unique up to inser-
tion or deletion of S 3 in the connected sum.

In 1962 Milnor [53] strengthened Kneser’s theorem to obtained uniqueness of the connected
sum, up to order and isomorphism of the summands Mi , where two manifolds are isomorphic
if there is a piecewise linear, orientation preserving homeomorphism between them.
This reduced the classification problem to classifying prime 3-manifolds. One problem
with prime 3-manifolds—illustrating the richness of 3-manifolds—is that, despite their
primeness, they could be homeomorphic to connected sums of non-spherical 3-manifolds.
The point is that the connected sum could be taken along, say, a solid torus, as we now
explain. Every 3-manifold has a solid torus as a subspace; if we take out the interiors of two
such filed tori T1 and T2 in two 3-manifolds X1 and X 2, and identify their boundaries via
a homeomorphism, then we get a new 3-manifold, X1 # ∂T1 ≅∂T2 X 2, a connected sum along
tori, or a torus sum of X1 and X 2.
Lens spaces are examples of such connected sums along tori: they are S 3 # ∂T1 ≅∂T2 S 3,
where the homeomorphisms identifying the boundaries of the deleted solid tori are of
special types that we now describe. Split S 3 into two solid tori sharing a common boundary
(Exercise 1). Delete the interior of one of these two solid tori from S3; what remains is the
other solid tori. S 3 # ∂T1 ≅∂T2 S 3 is then the space we get by identifying two solid tori along the
homeomorphism h between their boundaries that maps the meridian circles of one torus
to (p, q) torus-knots in the other, with p and q being relatively prime integers. (Recall that
a (p, q) torus knot winds p-times longitudinally and q times latitudinally along the torus.
The relative primeness of p and q guarantees that we can cover the torus with disjoint copies
of such knots.) Each such homeomorphism gives rise to the lens space denoted by L(p, q).
This is visualized in Illustration 14.52 in the case p = 4 and q = 5.

Illustration 14.52 Lens space L(4,5), made of two solid tori.

Lens spaces include the projective 3-space P 3 (Exercise 3).

K12146_C014.indd 359 2/12/15 2:50 PM

 360   ◾  On Classifying Manifolds and Related Topics

Another class of important 3-manifolds are torus bundles, which we construct as follows.
Start with a torus T 2 and the product space T 2 × I . A torus bundle is obtained by identifying
T 2 × {0} and T 2 × {1} along a homeomorphism h that preserves the orientation (assuming the
orientations of T 2 × {0} and T 2 × {1} are the same). The simplest case is when h is the “identity”
(( x , 0)  ( x ,1)), in which case we get S1 × S1 × S1, or the 3-torus T 3 (Illustration 14.53).

Illustration 14.53 S1 × S1 × S1 or the 3-torus T 3.

Discussing details of the classification problem for compact, connected 3-manifolds, includ-
ing proposed specific classifications, is beyond the scope of this book.

Finite CW-Complexes of Dimension 2

We show that finite CW-complexes of dimension 2—hence, of any dimension ≥ 2—are
not classifiable in the following sense: there is no algorithm that, when inputted with two
such CW-complexes, will decide if they are homeomorphic. The idea behind the justifica-
tion of this claim will be modified in the next subsection to show that (compact and con-
nected) n­-manifolds for n ≥ 4 are also not classifiable.

Proposition 2. For every finitely presented group G there exists a CW-complex X of

dimension 2 such that π1 ( X , x ) ≅ G .

Proof. Let G = 〈a1 , a2 ,… , am ; R1 , R2 ,… , Rs 〉. Assume s = 1 and consider the bouquet B made

of m oriented circles labeled a1 , a2 ,… , am. Its fundamental group is isomorphic to the free group
〈a1 , a2 ,… , am ; ∅〉. Let R1 = a1ε11a1ε22 a1εkk, where each a1i is in {a1 , a2 ,… , am }, and where each εi
is in {−1,1}. Let P be a (filled) polygon with k-edges labeled a11 , a12  a1k in counter-clockwise
order, and oriented counter-clockwise if εi = 1, clockwise if εi = −1. The labels and the orienta-
tion of these edges define the obvious continuous, label-preserving, and orientation-preserving
mapping h : ∂ P → B (so that each edge in P is wound once around the corresponding circle in
X via a mapping based on t  (cos2πt , sin2πt ), 0 ≤ t ≤ 1 ). We identify P and B along h to get a
space X. This procedure is shown in Illustration 14.54, where we take a1 = a, a2 = b, m = 2, and
R1 = aba.

K12146_C014.indd 360 2/12/15 2:50 PM

 14.4 Regarding Classification of CW-Complexes and Higher Dimensional Manifolds   ◾    361  

b
a =
a b a b a
Illustration 14.54 As usual, the resulting space X (to the right) should be viewed without the
self-intersections.
Split X into two open sets: one is a 2-disk in the interior of the polygon P (which stays
intact after the identification of the edges of P), the other is the complement in X of an
open 2-disk contained in the first 2-disk (as in Illustrations 14.34 and 14.35, Section 14.3).
The SvK theorem (repeating the argument that we used in Example 1, Section 14.3) yields
π1 ( X , x ) ≅ 〈a1 , a2 ,… , am ; R1 〉.

The rest follows by induction on s, where the inductive step is virtually the same as the
case s = 1.

Corollary 3. There is no algorithm to decide if two arbitrary finite CW-complexes of a fixed

dimension n ≥ 2 are homotopically equivalent.

Proof. For n = 2 this follows from Proposition 2, since any algorithm deciding if two
CW-complexes as constructed in the proof of Proposition 2 are homotopically equivalent will,
at the same time, decide if the associated fundamental groups are isomorphic. This contradicts
the nonsolvability of the isomorphism problem for finitely presented groups (D3, Section 12.2).

The case n ≥ 3 follows from the case n = 2 since attaching the n-sphere S n by identifying a
point in S n to the vertex of a CW-complex of dimension 2 gives a CW-complex of dimension n
with the same fundamental group.

Compact Connected n-Manifolds, n ≥ 4

The 4-sphere S 4 is one of infinitely many compact, connected, and simply connected
4-manifolds: another one is S 2 × S 2 (Illustration 14.55; the only 4-manifold we will attempt
to partially visualize). Even though these two 4-manifolds are simply connected, we will
see in the next section that they are not homeomorphic.

Illustration 14.55 S 2 × S 2 : recall that it can be viewed as the space obtained by replacing each
point of S2 by S2.

K12146_C014.indd 361 2/12/15 2:50 PM

 362   ◾  On Classifying Manifolds and Related Topics

The generalized Poincaré conjecture is the statement that for every n, if an n-manifold is
homotopically equivalent to the n-sphere S n, then it is homeomorphic to S n. For n = 1 and
n = 2, this is a consequence of the classification of such manifolds. Confirmations of the
conjecture for higher dimensions came in chronological order that did not follow the usual
order of the dimensions: the case n ≥ 5 was done in 1961 (Smale), n = 4 was settled in 1986
(Freedman), and, as indicated above, the affirmative solution for n = 3 is the result men-
tioned earlier (Perelman, 2003).
However, a complete classification of all compact, connected n-manifolds is not pos-
sible for any fixed n ≥ 4. We will prove this for n = 4; the cases n ≥ 5 are left as an exercise.
Let M and N be two n-manifolds, n ≥ 2, let B1n ⊂ M and B 2n ⊂ N be copies of the
open (unit) n-ball B n within the two manifolds. The connected sum M # N of these
two n-manifolds is the space M \ B1n ∪ f N \ B 2n where f is the “identity” mapping between
the two (n − 1)-spheres that are the boundaries of B1n and B 2n . It is easy to show that
π1 ( M # N ) ≅ π1 ( M ) ∗π1 ( N ) (Exercise 5).

Proposition 4. For every finitely presented group G there exists a compact and connected
4-manifold X, such that π1 ( X , x ) ≅ G.

Proof 2 . Let G = 〈a1 , a2 ,… , am ; R1 , R2 ,… , Rs 〉.

Step 1. Let X be a connected n-manifold, n ≥ 4, and let c be a simple (i.e., not self-intersecting)
closed curve in X. We will denote by the same letter the image of that curve. Further, let
B be a small open 3-ball such that c × B ⊂ X , and such that X \ ( c × B ) is a deformation
retract of X \ c. Denote T = c × B, and let x be any point in X \ T. In this step we prove that
π1 ( X , x ) ≅ π1 ( X \ c , x ) ≅ π1 ( X \ T , x ).
Since X \ T is a deformation retract of X \ c, it follows that π1 ( X \ c , x ) ≅ π1 ( X \ T , x ) .
In order to prove that π1 ( X , x ) ≅ π1 ( X \ c , x ) we apply the SvK theorem to the cover of X
consisting of T (which is open in X), and of the complement V in X of c × D1, where D1 is a
closed 3-ball smaller than B. We choose a point y in T ∩ V .
The group π1 (T , y ) is a cyclic, freely generated by the loop c (possibly appended by a
simple path to reach y).
Since T ∩ V is homeomorphic to S1 × S 2 × J , where J is an open interval, it follows that
π1 (T ∩ V , y ) is generated by (the homotopy class of) one loop l (winding once around
S1 × {a} × {b}, a ∈S 2, b ∈ J ). It is obvious that the homomorphism π1 (T ∩ V , y ) → π1 (T , y )
induced by the inclusion T ∩ V → T sends l to c or to c–1.
The other inclusion T ∩ V → V induces a homomorphism π1 (T ∩ V , y ) → π1 (V , y ) that
sends l to some element z ∈π1 (V , y ).
The SvK theorem implies that π1 ( X , y ) is the free product of π1 (V , y ) and π1 (T , y ) with
amalgamation along the subgroup π1 (T ∩ V , y ), where the amalgamation relations reduce to
only one: c = z or c = z −1. If π1 (V , y ) = 〈 A ; R 〉, this means that π1 ( X , y ) = 〈 A, c ; R , c = z ε 〉,
where ε ∈ {1, −1}, and where z is a word over A. A single Tietze transformation eliminating c

2
Following Seifert–Threlfall [68].

K12146_C014.indd 362 2/12/15 2:51 PM

 14.4 Regarding Classification of CW-Complexes and Higher Dimensional Manifolds   ◾    363  

yields π1 ( X , y ) = 〈 A ; R 〉 and so π1 ( X , y ) = π1 (V , y ) . Finally, since V is a deformation retract

of X \ c, it follows that π1 (V , y ) = π1 ( X \ c , y ), and so π1 ( X , x ) = π1 ( X \ c , x ), as claimed.

Step 2. Although tunneling T through X did not alter the fundamental group of X, it
destroyed its structure as a 4-manifold. We will now restore it to good effect.
The boundary of T in X is c × S , where S = ∂ B; notice that S is a 2-sphere. Consider
the space D 2 × S (recall that D 2 is the unit 2-dimensional closed disk) and identify
∂D 2 × S with c × S along the mapping that sends ∂D 2 homeomorphically onto c, and
which is the identity on the second coordinate. The resulting space is now a 4-mani-
fold. Moreover, it is rather easy to show by means of the SvK theorem that this space
has a presentation that extends a presentation of π1 ( X , x ) by the relation c = 1. We
leave the last claim as an exercise (Exercise 6).

Step 3. This step will complete the proof of the proposition.

 n 

( 1
)(
3 1
) ( )
We start by observing that Y = S × S # S × S 3 ## S1 × S 3 is a 4-manifold, and that
its fundamental group is 〈a1 , a2 ,… , an ; ∅〉, where ai is (the homotopy class of) a loop that
winds once around i-th appearance of S1 × S 3 in Y, extended by a simple path to reach the
base point, if necessary. This is a consequence of Proposition 9 in 10.4 and Exercise 5.
We now follow the idea used in the subsection on CW-complexes. The space Y will play
the role of the bouquet that we used in the construction of a CW-complex with a given fun-
damental group. As in that case, we orient each appearance of S1 × S 3 in Y by choosing one
orientation for S1.
Let’s assume that R1 = a1ε11a1ε22 a1εkk , where each a1i is in {a1 , a2 ,…, an }, and where each εi is
in {−1,1}. Let P be a (filled) polygon with k-edges labeled a11 , a12 , , a1k in counter-clock-
wise order, and oriented counter-clockwise if εi = 1, clockwise if εi = −1. Choose a simple
curve c1 in Y that winds, in turn, once around 1i -th appearance of S1 × S 3 in Y, following
its orientation if ε1i = 1, and going in the opposite direction if ε1i = −1. Note that there is
enough space in Y to construct such a curve without self-intersections, and such that, as
in Step 1, X \ (c1 × B) is a deformation retract of X \ c1. The loop we get in such a way corre-
sponds precisely to the word R1 in the presentation of Y. We show this in Illustration 14.56.

Illustration 14.56 We represent Y = (S1 × S 3 ) #(S1 × S 3 ) as a connected sum of two solid tori, and
then drill a tunnel through it with no self-intersection corresponding to, say, R1 = a1a2−2 in the
presentation.

K12146_C014.indd 363 2/12/15 2:51 PM

 364   ◾  On Classifying Manifolds and Related Topics

Now we utilize Steps 1 and 2. First we delete c1 × B; by Step 1, this will not affect the fundamental
group 〈a1 , a2 ,… , an ; ∅〉 of Y. Then we attach D × S as in Step 2. According to Step 2, the funda-
mental group of the resulting 4-manifold is 〈a1 , a2 ,… , am ; R1 〉.
Simple repetition of this step and induction gives us a 4-manifold X with
π1 ( X , x ) = 〈a1 , a2 ,… , am ; R1 , R2 ,… , Rs 〉.

Corollary 5. For every fixed n ≥ 5, the class of compact and connected n-manifolds cannot
be classified up to homotopy equivalence.

Proof. Exercise 7.

Exercises
1. Show that the 3-sphere S 3 is a union of two solid tori (D 2 × S1) sharing a common
boundary.
2. Is there a compact, connected 2-manifold X such that π1 ( X ) ≅ 〈 x ; x 5 〉? Construct a
path connected space X such that π1 ( X ) ≅ 〈 x ; x 5 〉; then use the SvK theorem to verify
that π1 ( X ) ≅ 〈 x ; x 5 〉.
3. Show that the Lens space L(2,1) (see Illustration 14.57) is the projective 3-space P 3
(Illustration 14.49). [Hint: What happens when we take a disk out of the projective
plane P 2?]

Illustration 14.57 We identify (1,2) torus knots on the torus to the left, with meridians on the torus
to the right.

4. Find the fundamental groups of K × S1 (K is a Klein bottle), the projective 3-space

P 3, and the 3-torus.
5. Show that for every n ≥ 3 and for every two n-manifolds M and N, π1(M # N ) ≅
π1 (M)∗π1 (N ).
6. Let X be a connected n-manifold, n ≥ 4, and let 〈 A ; R 〉 be a presentation of its fun-
damental group. Let c be a simple curve in X, and let B be a small open 3-ball such
that c × B is in X and such that c × B has no self-intersections (so that X \ (c × B)
is a deformation retract of X \ c). Denote S = ∂ B. Identify the boundary ∂(D 2 × S ) of
D 2 × S with the boundary c × S of c × B to get a new space Y. Explain why Y is also
an n-manifold. Use the SvK theorem to prove that 〈 A ; R , [c ] = 1〉 is a presentation of

K12146_C014.indd 364 2/12/15 2:51 PM

 14.5 Higher Homotopy Groups: A Brief Overview   ◾    365  

the fundamental group of the space Y. (As before, [c] denotes the homotopy class of
the loop c, possibly extended to reach the base point.)
7. (a) Prove that the class of compact, connected 5-manifolds cannot be classified up
to homotopy equivalence.
(b) Prove that the class of compact, connected n-manifolds, n ≥ 6, cannot be classi-
fied up to homotopy equivalence.

14.5 Higher Homotopy Groups: A Brief Overview

A brief historical note: Čech introduced the notion of higher homotopy groups in 1932, but
gave credit to Dehn as the one who had thought of it. Our definition is closer to that of Witold
Hurewicz (1904–1956); his article was published in 1935. However, there are earlier results on
higher homotopy groups, via different routes. For example, that π n (S n ) ≅  was first proven
in 1925 by Heinz Hopf (1894–1971). With his definition, Hurewicz managed to give a much
simpler proof of the same result. Hopf, in his paper, also proved the unexpected π 3 (S 2 ) ≅ .
The surprises regarding the homotopy groups of spheres were extended in 1937 with the proof
by Hans Freudenthal (1905–1990) that π 4 (S 3 ) ≅  2.

We will outline the standard construction of the higher homotopy groups π n ( X , x ), n ≥ 2

and the early stages of the theory of these groups.
One natural way to generalize the fundamental group π1 ( X , x ) is to consider continuous
mappings I n → X , sending the boundary ∂( In ) to x. The set {[ f ] : f : I n → X , f (∂( In )) = {x }}
of all homotopy classes of such mappings, where the homotopy is relative to ∂( In ), is denoted
by π n ( X , x ). Define an operation over the set of all mappings f , g : I n → X , ∂( In )  {x }, as
follows:

 1
 f (2s1 , s2 ,… , sn −1 , sn ), 0 ≤ s1 ≤
2
( f + g )(s1 , s2 ,… , sn −1 , sn ) =  .
 g (2s − 1, s ,… , s , s ), 1
≤ s1 ≤ 1
 1 2 n −1 n
2

The case n = 2 is shown in Illustration 14.58. The usage of the additive sign + will be
explained in the next paragraph.

g
I I
f f+g

Illustration 14.58 For f : I × I → X and g : I × I → X , we illustrate here ( f + g ) : I × I → X .

K12146_C014.indd 365 2/12/15 2:51 PM

 366   ◾  On Classifying Manifolds and Related Topics

It can be easily seen, by an argument very similar to that for the case n = 1, that [ f ] + [ g ] = [ f + g ]
is a well defined operation, and that, with this, the set π n ( X , x ) becomes a group: the n-th homo-
topy group of X at x (Exercise 1). When n ≥ 2, we will refer to these groups as higher homotopy
groups. The fact that ∂( In ) is disconnected only when n = 1 generates important (algebraic) dif-
ferences between π1 ( X , x ) on one side and the groups π n ( X , x ), n ≥ 2, on the other. Notably,
for n ≥ 2 the group π n ( X , x ) is always commutative; this is the reason we have used the addi-
tive symbol for the group operation. Justifying commutativity of π n ( X , x ) , n ≥ 2, is rather sim-
ple, and a visual explanation for n = 2 is given in Illustration 14.59.

g f

g
I I
f

Illustration 14.59 We depict a homotopy F : I × I × I → X relative to ∂( I 2 ) from f + g to

g + f : rotate the respective domains of f and g (by choosing suitable homeomorphic copies
of I × I for each of f and g) through the angles of s3 π radians, 0 ≤ s3 ≤ 1.

The initial basic setup of our theory for π1 ( X , x ) carries over to π n ( X , x ). The following
propositions confirm that point.

Proposition 1. Let ω : X → Y be continuous and such that ω( x ) = y . Then ω ∗ : π n ( X , x ) →

π n (Y , y ) defined by ω ∗ ([ f ]) = [ω  f ] is a well-defined group homomorphism.

Proof. Mimic the proof of Proposition 4 in 10.4 (Exercise 2).

In the setup of this proposition, we will again say that ω ∗ : π n ( X , x ) → π n (Y , y ) is the

induced homomorphism.

Proposition 2. Let ω : ( X , x ) → (Y , y ) and ψ : (Y , y ) → ( Z , z ) be continuous. Then

(ψ  ω )∗ = ψ ∗  ω ∗.

Proof. Same as the proof of Proposition 5 in 10.4.

Proposition 3. If the space X is path connected, then for every x , y ∈ X and every n ≥ 1, the
groups π n ( X , x ) and π n ( X , y ) are isomorphic.

K12146_C014.indd 366 2/12/15 2:51 PM

 14.5 Higher Homotopy Groups: A Brief Overview   ◾    367  

Proof. Choose a path γ from x to y, and define ϕ : π n ( X , x ) → π n ( X , y ) as indicated in

Illustrations 14.60 and 14.61, and in the text that follows.

I×I×I

γ´
I f´

I f´ γ´
            
Illustration 14.60 Defining ϕ([ f ]), n = 2.     Illustration 14.61 Defining ϕ([ f ]), n = 3.

In Illustration 14.60 we show how to define ϕ([ f ]), where [ f ] ∈ π 2 ( X , x ). The middle square
is twice smaller than I × I , and the two squares share a common center. Roughly speaking,
φ([ f ]) is the homotopy class (relative to ∂(I 2)) of the mapping g: I × I → X that along any
radial line segment is the product of f and γ. We elaborate what we mean by this. The domain
of the mapping f ′ visualized in Illustration 14.60 is the smaller square, and f ′ is defined
  1  1 1 3 1 3
as follows: f ′(s1 , s2 ) = f  2  s1 −  , 2  s2 −   , ≤ s1 ≤ , ≤ s2 ≤ . Along any radial
  4   4   4 4 4 4
segment (connecting the common center of the squares with the boundary of I × I ) we define
g to be the product of two paths: the first one is the path associated3 with the restriction of f ′
to the part of the segment within the smaller square, and the second path in the product is
that mapping γ ′ over the part of the segment between the two squares, such that the associ-
ated path is γ.
Since the boundary of the smaller square is sent to x via f ′, and since the initial point of γ
is also x, g is a well-defined continuous mapping I × I → X . Since γ ends at y, the mapping
g sends the boundary of I × I to y, so that the homotopy class of g determines an element of
π 2 ( X , y ).
It remains to be shown that φ is a well-defined isomorphism. The bijectivity fol-
lows from the existence of inverses of φ on both sides. The details of this claim and the
proof that φ is well-defined are left as an exercise (Exercise 3). We show below why φ is a
homomorphism.
A homotopy between representatives of the homotopy classes of ϕ([ f ])ϕ([ g ]) and ϕ([ fg ])
is visible in Illustration 14.62, confirming that ϕ([ f ])ϕ([ g ]) = ϕ([ fg ]) holds in π 2 ( X , y ).

3
The definition of “path associated with a mapping [a, b]→ X ” is given in 6.3.

K12146_C014.indd 367 2/12/15 2:51 PM

 368   ◾  On Classifying Manifolds and Related Topics

([fg])

I ([g])

([f ])

Illustration 14.62 Showing that φ is a homomorphism.

The definition of φ when n = 3 is indicated in Illustration 14.61. Proving the claim for
n ≥ 3 requires only minor modifications.

Corollary 4.
(a) If X and Y are homotopically equivalent, then π n ( X , x ) and π n ( X , y ), n ≥ 2, are
isomorphic.
(b) If X is a deformation retract of Y then π n ( X , x ) and π n ( X , y ) , n ≥ 2, are isomorphic.

Proof. Immediate consequence of Propositions 2 and 3, and of the definitions of homotopi-

cally equivalent spaces and deformation retracts.

Since π n ({x }, x ) is the trivial group, Corollary 4 implies that π n ( X , x ) = 1 in case X is an

n-ball, or when it is n , n ≥ 1.
As we see, the initial part of our theory developed in relation to the fundamental group
carries over, with some obvious changes, to higher homotopy groups. However, that is about
as far as it goes. The crucial Seifert van Kampen theorem has no obvious generalization.
Consider, for example, the four spaces depicted in Illustrations 14.63 through 14.66.

Illustration 14.63 A circle

with a 2-sphere.

K12146_C014.indd 368 2/12/15 2:51 PM

 14.5 Higher Homotopy Groups: A Brief Overview   ◾    369  

Illustration 14.64 The real line with a 2-sphere touching at every integer point.

Illustration 14.65 A circle with four touching 2-spheres.

Illustration 14.66 A 2-sphere with two antipodal points identified.

It may come as a surprise that the spaces shown in the illustrations above have the same
higher homotopy groups! This claim will be a relatively easy exercise (Exercise 10 in 15.4),
after we develop the theory of covering spaces.
The lack of a straightforward generalization of the SvK theorem for higher homotopy
groups is one of the reasons computing these groups turns out to be rather difficult, even
in the case of the most basic spaces. For example, our knowledge of the homotopy groups
of the n-spheres S n is still incomplete. It has been known for a while that π n (S n ) = , n ≥ 1
and that π i (S n ) = 0 for i < n. However, the pattern for π i (S n ), i > n , is still not fully under-
stood. For example, π 3 (S 2 ) = , π 4 (S 2 ) =  2 , π 6 (S 2 ) = 12 , π14 (S 2 ) =  84 ⊕  2 ⊕  .2

K12146_C014.indd 369 2/12/15 2:51 PM

 370   ◾  On Classifying Manifolds and Related Topics

The fact that π n (S n ) =  , n ≥ 1, fills a gap which we mentioned in Section 11.2: Brouwer’s
fixed-point theorem for higher dimensions. Since π n (D n , x ) is trivial for every n ≥ 2, the
argument used to prove Brouwer’s theorem for n = 2 needs only minor modifications to
justify the general claim as stated in Theorem 2 of 11.2.

Exercises
1. Show that π n ( X , x ) with the operation + is indeed a group. Visualize the homotopy
that confirms the associative law and the existence of inverses.
2. Prove Proposition 1.
3. Prove that the mapping φ, as defined in the proof of Proposition 3, is well-defined
and bijective.
4. Show that any space homeomorphic to the unit ball D n has the fixed-point property.
5. The space X shown in Illustration 14.67 below consists of three closed disks sharing a
common diameter. Show that X has the fixed-point property.

Illustration 14.67

6. Prove that for all spaces X and Y, and for every n ≥ 1, the groups π n ( X × Y ,( x , y )) and
π n ( X , x ) × π n (Y , y ) are isomorphic.
7. Use the results stated in this section to show that S 4 and S 2 × S 2 are not homotopically
equivalent (yet π1 (S 4 ) = π1 (S 2 × S 2 )).

K12146_C014.indd 370 2/12/15 2:52 PM

 Chapter 15

Covering Spaces, Part 1

I n this and the remaining chapters we will develop and apply the theory of covering
spaces. In a nutshell (and rather informally), we unwind spaces to reveal some aspects of
their structures, and, by extension, the structures of the associated fundamental groups.

15.1 Covering Spaces: Definition, Examples, and Preliminaries

A brief historical note: The notion of a (universal) covering space is implicit in Poincaré’s paper
published in 1883. He made it explicit in a paper published in 1907.

 , p), where
A covering space, or, briefly, a cover (or a covering) of a space X is a pair ( X

 is a path connected and locally path connected space, and p : X → X is a continuous
X
onto mapping satisfying the following condition: for every x ∈ X there is an open, path
connected neighborhood U of x such that the restriction of the mapping p on every path
component of p−1 (U ) is a homeomorphism onto U (Illustration 15.1).

~
X

p–1(U)

U X

Illustration 15.1 Covering space and an elementary neighborhood.

371

K12146_C015.indd 371 2/12/15 4:47 PM

 372   ◾    Covering Spaces, Part 1

The mapping p in this definition is the covering map. Occasionally, when p is understood,
 itself is a covering space, or a cover of X. The space X in this context is the
we will say that X
base space. The neighborhood U in the definition of covering space is called an elementary
neighborhood of x ∈ X .
Since p is onto, and since X  is path connected, it follows that X is path connected
(Proposition 2, in 6.4). Moreover, the local path connectedness of X  forces X to be locally
path connected as well (Proposition 2 below).

 , p) be a covering space of X. If U is an elementary neighborhood, then

Proposition 1. Let ( X
every path component of p −1 (U ) is open in X .

Proof. This is a consequence of Proposition 2 in 6.5.

 , p) be a covering space of X. Then X is locally path connected.

Proposition 2. Let ( X

Proof. Let V be an open neighborhood of x ∈ X , and let U be an elementary neighborhood

for x. If C is a path component of p −1 (U ), then the restriction p C is a homeomorphism onto
U. Denote C ′ = C ∩ p −1 (V ∩ U ), and y = p −1 ( x ) ∩ C (Illustration 15.2).

~
X

C´ C

x
U X
V

Illustration 15.2

 is locally path connected, there

Then C ′ is an open (in C) neighborhood of y. Since X

exists an open (in X ) path connected neighborhood W of y such that y ∈W ⊂ C ′. Then
(since p C is a homeomorphism) p(W) is an open (in U) path connected neighborhood of x,
such that p(W ) ⊂ V ∩ U . Since V ∩ U is open in X, p(W) is open in X. Clearly, p(W ) ⊂ V ,
and so we have achieved our goal.

In view of Proposition 2, only path connected and locally path connected spaces have
covering spaces, according to our definition. Hence, unless otherwise stated, we will
assume the following:

The base spaces are path connected and locally path connected.

K12146_C015.indd 372 2/12/15 4:47 PM

 15.1 Covering Spaces: Definition, Examples, and Preliminaries   ◾    373  

Example 1
(X, id) is a covering space of X. This is the trivial covering space. We have seen non-trivial
covering spaces earlier, and a notable example was encountered in Part 2 of the proof of
Theorem 1, Section 11.1: (, p), where p :  → S1 is defined by p(s ) = (cos 2πs , sin 2πs ),
is a covering space of the circle S1. See Illustration 11.5 in 11.1 for a visualization of this
covering space. ☐

Note that the cover  of S1 is simply connected. Such covers will play an important role
in the theory we are building, and we introduce them formally: A cover ( X  , p) of a space

X is a universal cover of the space X if the space X is simply connected. In a few sections
we will prove that universal covers are unique up to certain special homeomorphisms. In
light of that uniqueness we will, at times, refer to a simply connected cover as the univer-
sal cover.
There are, of course, covers that are neither trivial nor universal.

Example 2

~
X

Illustration 15.3

The cover space X  and the base space X are both S1 (Illustration 15.3). For every k ∈,
k ≠ 0, p(cos 2 πs , sin 2πs ) = (cos 2 kπs , sin 2kπs ), s ∈[0, 1], defines a covering map. For exam-
ple, when k = 2, the mapping p wraps the covering circle X  = S1 twice around the base
1
space X = S . This last covering space is different from the trivial cover of S1. In precisely
what sense it is different will be explained later. ☐

Example 3
We show a few coverings of the bouquet of two circles. The associated covering maps
are defined through the labels of the edges.

v
b a

Illustration 15.4 The base space X is a bouquet of two circles.

K12146_C015.indd 373 2/12/15 4:47 PM

 374   ◾    Covering Spaces, Part 1

b
a
a
a
b a
b b
a a a a
b b
b a
a a
b
a
b

Illustration 15.5 Here is a universal cover of the bouquet X.

a a a a
b

Illustration 15.6 A compact cover of the bouquet X.

b b b

a a a a a
b b b

Illustration 15.7 A non-compact, non-simply connected cover of the bouquet X.

Covering spaces of bouquets of circles are much more than cute graphs: as we will see
later they have some interesting applications in group theory. ☐

We end this section with the following property of covering maps.

Proposition 3. Covering maps are always open.

 , p) be a covering space of X, and let V be an open subset of X

Proof. Let ( X  . Choose points
−1
x ∈ p(V ) and y ∈p ( x ) ∩ V. Let U be an elementary neighborhood of x with respect to
 , p), and let C be the path component of p −1 (U ) containing y. Since C ∩ V is open in
(X
C, and since p C is a homeomorphism onto U, p(C ∩ V ) is open in U. Since U is open in
X, p(C ∩ V ) is open in X. Obviously x ∈ p(C ∩ V ) ⊂ p(V ), hence x is an interior point for
p(V).

K12146_C015.indd 374 2/12/15 4:47 PM

 15.2 Lifts of Paths   ◾    375  

Exercises
 , p) be a covering space of X and let U be an elementary neighborhood of x ∈ X .
1. Let ( X
Show that if V is a path connected open set such that x ∈V ⊂ U , then V is also an
elementary neighborhood of x.
2. Describe a universal cover of X equipped with the indiscrete topology. When does X
with the discrete topology possess a universal cover?
 , p) be a covering of X. Prove that for every x ∈ X , p−1 ( x ) is a discrete subspace of X.
3. Let ( X 

4. Prove that if (Y, p) is a covering space of a Hausdorff space X, and if Y is a BW-space, then for
every x ∈ X , p−1 ( x ) is finite. Deduce the same conclusion if Y is assumed to be compact.
5. Find a universal cover of the space depicted in Illustration 14.65, Section 14.5 (a circle
with four attached spheres).
6. Describe a universal covering space, and two more covering spaces of the space shown
in Illustration 14.66, Section 14.5 (a sphere with two antipodal points identified).
7. Suppose we do not require that the cover X  be locally path connected. Show that if X
  is locally path
is locally path connected, and ( X , p) is a covering space of X, then X
connected.
8. Find a universal cover of  2 \ B2, where B2 = {( x , y ): x 2 + y 2 < 1}.
 , p) be a covering of X.
9. Let ( X

(a) Show that if X Hausdorff, then so is X.
(b) Show that if for every x ∈ X there exists an open neighborhood U x such that

U x is homeomorphic to m , for a fixed positive integer m, then for every x ∈ X
m
there exists an open neighborhood Vx such that Vx is homeomorphic to  .
 , p) be a covering of X. Prove that if X
10. Let ( X  is an n-manifold then X is an n-manifold.
1 , p1 ) is a covering space of a (path connected and locally path connected)
11. Suppose ( X
space X1, and suppose ( X 2 , p2 ) is a covering space of a (path connected and locally
path connected) space X 2. Show that ( X 1 × X
2 ,( p1 , p2 )) is a covering space of X1 × X 2.

15.2 Lifts of Paths

A brief historical note: Propositions 1 and 2 were proven for 2-manifolds in 1913 by Hermann
Weyl. The formal study of covering spaces and their connection with the fundamental groups,
based on the path-lifting theorems stated in this section, was inaugurated by Seifert and his
teacher William Threlfall (1888–1949) through their influential book A Textbook of Topology,
published in 1934.

K12146_C015.indd 375 2/12/15 4:47 PM

 376   ◾    Covering Spaces, Part 1

The lifting properties that we will encounter in this section are at the core of the theory of
covering spaces.
Let f : Y → X be a continuous mapping and let ( X  , p) be a covering space of X. A lift f

of f is a continuous mapping Y → X making the diagram in Illustration 15.8 commutative:
~
X
~
f p

Y X
f

Illustration 15.8

We start by showing that paths can be lifted.

Theorem 1. Let ( X , p) be a covering space of X, and let α be a path starting at x ∈ X . For

−1
every y ∈ p ( x ) there exists a unique lift of α starting at y.

Proof.
Step 1 (Uniqueness). Let α 1 and α 2 be two distinct lifts of α starting at y. Denote
1 (s ) = α
z = sup{t ∈ I : α 1 (0) = α
2 (s ) for every s ≤ t }. Since α 2 (0) we have that z ≥ 0, and the
 
assumption that α1 and α 2 are distinct implies that z <1. It follows from the continuity
of the mappings α 1 and α 2 that α 1 (z ) = α
2 (z ). Notice that in every open neighborhood of
 
z there are points where α1 and α 2 disagree.

Let U be an elementary neighborhood of w = α(z ) (Illustration 15.9).

~
X

C
~
α1
~ p
α 2
U X
I z α
α(z)

Illustration 15.9

Let C be the path component of p −1 (U ) containing α 1 (z ) = α

2 (z ) . Then C is open, and
−1 −1
so α1 (C ) ∩ α 2 (C ) is an open neighborhood of z. For every t ∈α 1 −1 (C ) ∩ α 2 −1 (C ), we
have α 1 (t ) = ( p C )−1  α(t ) = α
2 (t ). Hence α
1 and α
2 agree over the open neighborhood
−1 −1
1 (C )∩α
α 2 (C ) of z, and we have a contradiction. Hence, z = 1 and each lift of α starting
at y is unique.
Step 2 (Existence). Let U be a cover of X consisting of elementary neighborhoods,
and let α = α1α 2 α n be a Lebesgue subdivision of the path α relative to the cover U .
This means that the interval I is split into [t0 = 0, t1 ], [t1 , t 2 ], … , [t j , t j +1 ], … , [tn −1 ,1 = tn ]

K12146_C015.indd 376 2/12/15 4:47 PM

 15.2 Lifts of Paths   ◾    377  

so that the domain of each α j is [t j −1 , t j ], and so that α j ([t j−1 , t j ]) is a subset of some
member U j of the cover U, j = 1, 2, …n.
Start with U1 and let C1 be the component of p −1 (U1 ) containing y. Then p |C1 is a homeo-
morphism onto U1, so that p −1 α1 is the lift of α1 starting at y. Denote the terminal point of
the path p −1 α1 by y1, and set x1 = p( y1 ). Notice that x1 is the terminal point of α1 and the
starting point of α 2, so that x1 ∈U1 ∩ U 2. Denote the component of p −1 (U 2 ) containing y1
by C2 . Then p −1 α 2 is the lift of α 2 starting at y1. It follows that ( p−1  α1 )( p−1  α 2 ) is the lift
of α1α 2 starting at y (Illustration 15.10).
~
X

~ α~2 α~3
α α~4
1

X
I α
α1 α
2 α3 α4

Illustration 15.10

Proceeding by induction we obtain a lift ( p−1  α1 )( p−1  α 2 )( p−1  αn ) of α = α1α 2 α n

starting at y.

By considering paths we are orienting the theory of covering spaces in the direction of
fundamental groups of spaces. We proceed in that direction with the following result.

 , p) be a covering space of X, let α and β be two paths in X starting

Theorem 2. Let ( X
at x1 and ending at x 2. Choose any y ∈ p −1 ( x1 ) and let α  and β be the unique lifts of α
 
and β, respectively, starting at y. If α  {0, 1} β, then α  {0, 1} β.

Proof. (The idea behind the proof that follows is rather simple: we will lift the homotopy
between α and β, to a homotopy between α 
 and β.)

Let F : I × I → X be a homotopy relative to {0,1} from α to β. Following the idea of

the proof of Theorem 1, we choose a cover U of X made of elementary neighborhoods.
Let {Fi ,k : i ∈{0,1, … , n − 1}, k ∈{0,1, … , m − 1}} be a Lebesgue subdivision of the mapping F
(relative to the cover U ). This means that Fi ,k = F |Ri ,k , where Ri ,k is a rectangle in the net of
n × m rectangles subdividing I × I , and Fi ,k ( Ri ,k ) ⊂ U i ,k for some U i ,k ∈ U , i = 0, 1, … , n − 1,
j = 0, 1, … , m − 1.
First, we lift F0,0 as follows: the rectangle R0,0 is mapped by F0,0 into the elementary neigh-
borhood U 0,0. Choose the component C0,0 of p −1 (U 0,0 ) containing y (where y is given in
the statement of the proposition), and set F −1
0,0 = ( p|c0,0 )  F0,0 . Observe that the entire left
edge of R0,0 is mapped to y by F 0,0 ; the following steps will eventually ensure that the lift of
the homotopy F we are in process of constructing is also relative to {0, 1}.

K12146_C015.indd 377 2/12/15 4:47 PM

 378   ◾    Covering Spaces, Part 1

In the next step, we consider F1,0 and proceed similarly to get a lift F −1
1,0 = ( p |C 1,0 )  F1,0 . The
only minor nuisance that needs to be addressed is the choice of a component C1,0 of p −1 (U1,0 ):
we choose the one that contains the image of the right vertical edge of R0,0 under the mapping
F
0,0 . This can be done since F0,0 and F1,0 agree over that edge.

~
X

~ ~
F F(Ri,k)

p
Ri,k

X
F
F(Ri,k)

Illustration 15.11

We proceed inductively: suppose we have a lift F i −1, k of the restriction Fi −1, k of F over the
shaded part of I × I (Illustration 15.11). We extend this lift over the rectangle Ri ,k as follows.
Denote the union of the left vertical and the lower horizontal edges of Ri ,k by Bi ,k. The map-
pings Fi −1,k and Fi ,k agree over Bi ,k. Hence, we can choose that component Ci ,k of p −1 (U i ,k )
containing Fi −1,k ( Bi ,k ) . This gives a lift F −1
i , k = ( p |Ci ,k )  Fi , k of Fi , k . By the gluing lemma, the
union of the mappings F   
i −1, k and Fi , k is a continuous extension Fi , k of Fi −1, k , and by our

construction Fi ,k is a lift of the restriction Fi ,k of F over the union of the shaded part of I × I
together with the rectangle Ri ,k.

Repeating this procedure over all of the smaller rectangles yields the desired homotopy

F between α  The construction is such that the homotopy F is relative {0, 1}.
 and β.

The following consequence of the proof of Theorem 2 should be noted.

 and β have a common terminating
Corollary 3. In the setup of Theorem 2, the unique lifts α
point.

Exercises
1. Consider the torus T 2 = S1 × S1, and define p : T 2 → T 2 by p ((cos t , sin t ), (cos s , sin s )) =
((cos t ,sin t ),(cos2s ,sin2s )) , t , s ∈[0, 2 π].

(a) Show that (T 2 , p) is a covering space of T 2 .

(b) Describe the lifts of the loop α(t ) = ((cos2 πt ,sin2 πt ),(cos a,sin a)) , t ∈[0, 1], a is
fixed, and of the loop β(t ) = ((cos a,sin a),(cos2 πt ,sin2 πt )), t ∈[0, 1], a is fixed.
(c) Which loops in T 2 have lifts with respect to (T 2 , p) that are also loops in the cover T 2.

K12146_C015.indd 378 2/12/15 4:48 PM

 15.3 Lifts of Mappings   ◾    379  

2. Consider the map p :  2 →  2 \ {(0,0)}} defined by1 p( x , y ) = e x (cos y ,sin y ).

(a) Show that ( 2 , p) is a covering space of  2 \ {(0,0)}. Describe the map p visually:
where do vertical lines go under p?
(b) Let α : I →  2 \ {(0,0)} be the path defined by α(t ) = (1 − t )(1, 0) + t (0, 1). Find the
lift of α starting at (0,0).
 , p) be a covering space of X and fix x ∈ X . Let α be a loop at x, let β be the lift
3. Let ( X
of α starting at some y ∈ p −1 ( x ), and let γ be the lift of α starting at β(1). Show that
βγ is the lift of α 2 starting at y.
 , p) be a covering space of X and let α be a loop at x ∈ X defining the trivial ele-
4. Let ( X
ment in π1 ( X , x ). Show that the unique lift α  of α starting at some y ∈ p −1 ( x ) must be a
loop, and that α  , y ).
 also defines the trivial element in π1 ( X
5. Let ( X , p) be a covering space of X, let x ∈ X and x ∈ p −1 ( x ) . Let α be a path in
X from x to some y ∈ X . Let α be the lift of α starting at x (and ending at some

y ∈ p −1 ( y )), and let α −1
be the lift of α −1 starting at y . Show that (α )−1 = α −1
.
6. View the projective plane P 2 as the quotient space of the sphere S 2 with the antipodal
points identified.
(a) Describe the mapping p that makes (S 2 , p) a universal covering space of P 2.
(b) Show that there is no mapping q that makes ( P 2 , q) a covering space of S 2 . (The
solution, of course, should be based on what we have learned so far.)
 , p1 ) and ( X
7. Let ( X  , p2 ) be two covering spaces of X such that p ( y ) = p ( y ) for some
1 2
 Assume further that for every path α in X starting at p1 ( y ) = p2 ( y ), we have that
y ∈ X.
every lift α 1 of α with respect to ( X , p1 ) starting at y is the same as the lift α 2 of α with
respect to ( X , p2 ) starting at y. Show that p1 = p2.
 , p) be a covering of a space X, and let Y be a subspace of X containing x such
8. Let ( X
that the inclusion Y → X induces a surjection π1 (Y , x ) → π1 ( X , x ). Show that p −1 (Y )
is path connected.

15.3 Lifts of Mappings
In this section we start paying more attention to the relationship between the fundamental group of
spaces and their coverings. The following simple result is important and will be used many times.
 , p) be a covering space of X. Choose any x ∈ X and x ∈ p −1 ( x ). The
Proposition 1. Let ( X
induced homomorphism p∗ : π1 ( X  , x ) → π1 ( X , x ) is always a monomorphism.
Proof.
Let α and β be loops at x, and assume p∗ ([α]) = p∗ ([β]). This means that pα and p β are
homotopic loops at x. By Theorem 2 in 15.2 their unique lifts starting at x are homotopic
1
This p is the complex map z  e z .

K12146_C015.indd 379 2/12/15 4:48 PM

 380   ◾    Covering Spaces, Part 1

relative to {0,1}. Since the loops α and β are, respectively, the lifts of pα and p β starting
at x , it follows that [α] = [β].

Proposition 1 allows us to visualize subgroups of the fundamental group π1 ( X , x ) of the

base space through the covering of spaces of X. We will see later that, under some simple
conditions, all subgroups of π1 ( X , x ) can be visualized in such a way. This is of consequence,
since in some cases it turns out that certain problems in group theory are more approachable
by considering coverings of spaces. We will discuss this in more detail in Chapter 17.
In the setup of Proposition 1, if p∗ ( π1 ( X  , x )) = H then we will say that the covering space
(( X , x ), p) of ( X , x ) corresponds to a subgroup H of π1 ( X , x ). When x is clear from the context,
or when the choice of x ∈ p −1 ( x ) is inconsequential, we will say that the covering space ( X , p) of
X corresponds to a subgroup H of π1 ( X , x ). For example, the trivial cover of X corresponds to
π1 ( X , x ), and every universal cover of X corresponds to the trivial subgroup of π1 ( X , x ).
We now take a look at lifts of arbitrary continuous mappings Y → X . The problem of
when such mappings can be lifted is solved by the following theorem.
Theorem 2. Let ( X  , p) be a cover of X, let Y be a path connected and locally path con-
nected space, and let f : Y → X be a continuous mapping. Fix y0 ∈Y , denote x 0 = f ( y0 ),
 of f with f ( y0 ) = x 0 if and
and choose any x 0 ∈ p −1 ( x 0 ). There exists a unique lift f : Y → X

only if f∗ (π1 (Y , y0 )) is a subset of p∗ (π1 ( X , x 0 )).
 , x 0 )) if
Note that, by elementary group theory, the group f∗ (π1 (Y , y0 )) is a subset of p∗ (π1 ( X
 , x 0 )).
and only if it is a subgroup of p∗ (π1 ( X
Proof. ⇒ Suppose there exists a (unique) lift f : Y → X  of f satisfying f ( y0 ) = x 0 . Then,
 , x 0 ), we have f (π1 (Y , y0 )) = p∗ ( f (π1 (Y , y0 ))) ⊂ p∗ (π1 ( X
since f ∗ (π1 (Y , y0 )) ⊂ π1 ( X  , x 0 )).
∗ ∗

 , x 0 )) , we construct f : Y → X
⇐ Assuming f∗ (π1 (Y , y0 )) is a subset of p∗ (π1 ( X  as fol-
lows. For every y ∈Y choose a path α starting at y0 and ending at y. By Theorem 1, 15.2,
the path λ = f  α has a unique lift λ starting at x 0. Let z be the endpoint of λ;  define
f ( y ) = z (see Illustration 15.12).

z ~
X
~
λ=f °α
x~0

~
f
p

y0 α x0
Y x
y f X
λ=f °α

Illustration 15.12

K12146_C015.indd 380 2/12/15 4:48 PM

 15.3 Lifts of Mappings   ◾    381  

First, we show that f is well defined. Suppose β is another path starting at y0 and ending at
y. Then αβ −1 is a loop at y0, and so f∗ ([αβ −1 ]) ∈ f∗ (π1 (Y , y0 )). By assumption, there is a loop
δ at x 0 such that p∗ ([δ]) = f∗ ([αβ −1 ]) . This means that the loops pδ and f  (αβ −1 ) are
homotopic. Denote the lift of f β −1 starting at z by  f  β −1 . Since δ is the lift of p δ start-
 
ing at x 0 , and since λ f  β = 
−1 
f  α f  β is the lift of f  (αβ −1 ) starting at the same point,
−1

Theorem 2, Section 15.2, implies that δ and λ  f  β −1 are homotopic, and, by Corollary 3
in 15.2, they end at the same point, which must be x 0 since δ ends there. It follows that

f  β −1 (which starts at z) must end at x 0 . Since f  β −1 = ( f  β)−1 – the inverse stands for
path-inverse – it follows from Exercise 5 in 15.2 that  f  β −1 = ( 
f  β)−1 . Consequently f β
(which starts at x 0), must end at z.
It is obvious that p  f = f . The proof will be completed once we establish that f is con-
tinuous and unique.
Continuity: Let W be an open neighborhood of z. We will find an open neighborhood
V of y such that f (V ) ⊂ W .

W
W´ ~
z
X
U
~
x0
~
f
p

Y y0 x0
V y f U´ x Ux X

Illustration 15.13

Choose an elementary neighborhood U x of x = f ( y ). Denote the path component of

p −1 (U x ) containing z by U. As we have seen several times now, U (being a path component
of an open set in a locally path connected space X  ) is also open. Consequently, W ∩ U
is an open neighborhood of z. Let W ′ be the path component of W ∩ U containing z. Then
W ′ is open in X  , and, since p |U is a homeomorphism, U ′ = p(W ′ ) is open in Ux . Hence
U ′ is open in X too. Let V be the path component of f −1 (U ′ ) containing y. Since we have
assumed that Y is locally path connected, it follows that V is open. We now show that
f (V ) ⊂ W .
( )
−1
Take any y ′ ∈V and a path α from y to y ′ in V. Then both p W ′  f  α and
f α are lifts of α starting at z. The former clearly ends at a point in W ′ . Hence, so
does the latter. The end point of f α is f ( y ′ ), and so f ( y ′ ) ∈W ′ . We proved that
f (V ) ⊂ W ′ . Since W ′ ⊂ W , we have f (V ) ⊂ W , which establishes continuity of f at
every point y ∈Y .

K12146_C015.indd 381 2/12/15 4:48 PM

 382   ◾    Covering Spaces, Part 1

Uniqueness: We prove that f is the only lift of f satisfying f ( y0 ) = x 0 . Otherwise, sup-

pose g : Y → X  also satisfies g ( y0 ) = x 0 . Choose any y ∈Y and assume g ( y ) ≠ f ( y ). Let
α be a path in Y from y0 to y, and let λ = f  α . Then both f  α and g α are lifts of λ
starting at x 0. Hence, by Theorem 1 in 15.2, f  α = g  α, and thus the end points of f  α
and g α are the same. However, our assumption that g ( y ) ≠ f ( y ) forces them to end at
different points.

Example 1
Consider the mapping f described in Illustration 15.14a: the circle Y is mapped
homeomorphically onto the a-labeled circle in the bouquet X of two circle, and
f ( y ) = x . The covering map p in Illustration 15.14a is determined by the labels of the
edges. The map f can easily be lifted to a map f so that f ( y ) = x : f sends Y onto the
thicker a-labeled loop in the covering.
~ ~
a a X a a a a X a a

.... .... .... ....

~ ~
b b x b b b b x b b
~ ~
f p f p
?
Y a b
a f a
y b x y b x
f Y
X X

Illustration 15.14a Illustration 15.14b

On the other hand, the mapping f described in Illustration 15.14b—the circle Y goes
homeomorphically to the b-labeled circle of the bouquet X—cannot be lifted, since
 , x )).
f∗ (π1 (Y , y )) is obviously not a subset of p∗ (π1 ( X ☐

We can now justify the usage of the definite article in “the universal cover.” Let ( X 1 , p1 )
 −1 −1
and ( X 2 , p2 ) be two universal covers of a space X. Fix x ∈ X , x1 ∈ p1 ( x ) and x 2 ∈ p2 ( x ). Since
1 → X
1 , x1 )) is trivial, Theorem 2 applies and there is a unique lift of p1 to f : X
( p1 )∗ (π1 ( X 2
such that f ( x1 ) = x 2 (see Illustration 15.15). Symmetrically, there is a lift g : X 2 → X  1
of p2 such that g ( x 2 ) = x1 (Illustration 15.15). Then g  f : X 1 → X
1 is also a lift of p1 and
g  f ( x1 ) = x1. By Theorem 2 (the uniqueness part), we have g  f = id X1. By symmetry, we

~ g ~
X1 X2
x1 f x2

p1 p2

x X

Illustration 15.15

K12146_C015.indd 382 2/12/15 4:48 PM

 15.3 Lifts of Mappings   ◾    383  

have f  g = id X2 . It follows that both f and g are homeomorphisms. We have proven the
following corollary.

Corollary 3. If a space has a universal cover, then it is unique up to a homeomorphism f

that makes the diagram in Illustration 15.15 commutative.

1 → X
If f : X 2 is a homeomorphism between two covering spaces X 1 and X 2 of X such
that the diagram in Illustration 15.15 commutes, then we say that f is a covering homeo-
2 → X
morphism. In that case f −1 : X 1 is also a covering homeomorphism.
The following generalization of Corollary 3 can be justified by a slight modification of
the argument used for Corollary 3. The details are left to the reader (Exercise 7).

2 , p2 ) are two covering spaces of a space X corresponding to the

1 , p1 ) and ( X
Corollary 4. If ( X
same subgroup of π1 ( X , x ), then there is a covering homeomorphism f : X 1 → X2 .

In particular, if a space X is simply connected, then the trivial covering space is the only
covering space of X, up to a covering homeomorphism.

Example 2 : An Application in Combinatorics

We digress to give a minor application in combinatorics.

Proposition 5. The vertices of a triangulated sphere can be colored with three differ-
ent colors such that the vertices in each triangle are of different colors if and only if
every vertex is incident to an even number of triangles.

Proof.1

⇒ If a vertex is incident with an odd number of (triangular) faces (at least three of
them), then the other vertices in the adjacent triangles must be colored with two col-
ors, alternating as we traverse the triangles cyclically. This is not possible if there are
odd many such faces.
⇐ Suppose each vertex in a triangulation of a sphere is incident to an even num-
ber of triangles. For every triangle in the triangulation we choose 6 copies, each
with a specific coloring of the vertices, exhausting all six possibilities. We say that
any two such colored triangles are neighbors if their originals on the sphere have
exactly one common edge. We identify the pairs of edges in the neighboring col-
ored triangles whenever this pair corresponds to a single edge in the triangulation
of the sphere and if the edges in the pair have endpoints of compatible colors. We get
a space Y (see Illustration 15.16 and the caption).

1
Maxim Kontsevitch, Mathematical Intelligencer, Volume 19, Number 5, 1997, pp. 48–49.

K12146_C015.indd 383 2/12/15 4:48 PM

 384   ◾    Covering Spaces, Part 1

R R
G B
B B
G B G a
R R
R G R
B G
G b G
B G R
GB G
R B B
R b B B
B
R G
a
R a
R
G
B
R
G
G
b
G R

B
R b
B
B

R a

Illustration 15.16 There are six colored replicas above each triangle in the triangulation of
the sphere. We identify only pairs of corresponding edges with compatible colors of the ver-
tices. For example, the two edges shown with label a in this figure correspond to one edge
of the original triangulation and are of compatible colored vertices. We identify them. We do
likewise for the two edges labeled b, etc. (This is emphasized in the figure to the right.)  3 is
insufficient to visualize Y without self-intersections.

Clearly Y is locally path connected. The assumption that there are evenly many
triangles around each vertex in the triangulation of S 2 allows a fit of the same number
of triangles with colored vertices around each vertex in Y. Consequently, Y is triangu-
lated such that each triangle has vertices colored in three different colors.
Define p : Y → S 2 to be the mapping that sends each triangle in Y to its original
in S 2 via the corresponding homeomorphism. By construction p is well defined,
and by the gluing lemma it is continuous. Our construction implies that for every
x ∈S 2, be it in the interior of a triangle, on an edge, or at a vertex, there is a small
open neighborhood U of x, such that the path components of p −1 (U ) are each
mapped via p homeomorphically onto U.
It follows that for each component C of Y, (C , p |C ) is a covering space of S 2. Since S 2
is simply connected Corollary 4 implies that there is a homeomorphism f : C → S 2 such
that id  f = p C , i.e., such that f = p C . This implies that p C is a homeomorphism. We
can now color each vertex v in the triangulation of S 2 with the color of the unique vertex
( p |C )−1 (v ). That will produce the desired coloring of the triangulation of S 2 .

It is evident that the argument we have used to justify Proposition 5 does not
depend on any intrinsic property of the sphere other than its being a simply con-
nected 2-manifold:  2 (in place of S 2) allows similar claims and justifications. Other
examples are in the set of exercises. ☐

K12146_C015.indd 384 2/12/15 4:48 PM

 15.3 Lifts of Mappings   ◾    385  

If the mapping f in the statement of Theorem 2 is a covering map, then more can be said
about its lift.

Proposition 6. Let ( X1 , p1 ) and ( X 2 , p2 ) be covering spaces of X. Fix x ∈ X , x1 ∈ p1−1 ( x )

and x 2 ∈ p2−1 ( x ), and suppose ( p1 )∗ (π1 ( X1 , x1 )) ⊂ ( p2 )∗ (π1 ( X 2 , x 2 )). Then the unique lift
f : ( X1 , x1 ) → ( X 2 , x 2 ) of p1 with respect to ( X 2 , p2 ) is a covering space of X 2. (See
Illustration 15.17.)
f
X1 X2

p1 p2

Illustration 15.17

Proof. Observe that Theorem 2 ensures that f : ( X1 , x1 ) → ( X 2 , x 2 ) exists. We only need to

show it is a covering map.

First we show that f is onto. Choose any y ∈ X 2 , denote x = p2 ( y ) , and choose z ∈ p1−1 ( x ). The
commutativity of the diagram in Illustration 15.17 implies that f (z ) ∈ p2−1 ( x ). Choose any
path λ from f(z) to y in X 2 . Then α = p2  λ is a loop at x. Lift it to a unique path δ starting at
z ∈ X1, and denote the endpoint of δ by w (see Illustration 15.18). Since δ starts at z, it follows
that the path

f
X1 w y X2
δ
z f(z) λ

p1 p2

x α X

Illustration 15.18

f ° δ starts at f(z). It is easy to see (Illustration 15.18) that f δ is the lift of α starting at f(z).
On the other hand, it is plain that λ is also such a lift. It follows from Theorem 1, 15.2, that
f δ = λ ; in particular their endpoints are the same. Hence, f (w ) = y and we have proven
what we wanted.
Next, we establish existence of elementary neighborhoods around points in X 2 with
respect to ( X1 , f ). Let y be any element in X 2 and choose z ∈ f −1 ( y ) ; we can do that since,
by the first part of the proof, f −1 ( y ) ≠ ∅ . Then the point x = p2 ( y ) has an elementary
neighborhood U1 with respect to ( X1 , p1 ), and an elementary neighborhood U 2 with respect
to ( X 2 , p2 ). The path component U of U1 ∩ U 2 containing x is then an elementary neigh-
borhood with respect to both ( X1 , p1 ) and ( X 2 , p2 ). Choose V to be the path component of
p2−1 (U ) containing y and choose W to be the path component of p1−1 (U ) containing z. Let Z
be the path component of f −1 (V ) containing z (see Illustration 15.19).

K12146_C015.indd 385 2/12/15 4:48 PM

 386   ◾    Covering Spaces, Part 1

Z
f
X1 W z y V X2

p1 p2

x U X

Illustration 15.19

We show that W = Z . For every a ∈ Z \ W , we have p2  f (a) ∈U , and so p1 (a) ∈U . So, if

Z \ W ≠ ∅, then, since Z ∪ W is path connected (Proposition 3 in Section 6.3), Z ∪ W would
be a larger path connected subset of p1−1 (U ). This violates the assumption that W is a path com-
ponent of p1−1 (U ). So Z \ W = ∅, i.e., Z ⊂ W . Suppose Z  W (see Illustration 15.20).

δ f f°δ
X1 W V X2
Z

p1 p2

U X
p2 ° f ° δ

Illustration 15.20

Since W is path connected, there is a path δ in W starting at a point in Z and ending at a

point in W \ Z . Then f δ is a path starting at a point in V and ending at a point out of V.
Since V is a path component of p2−1 (U ), there is a point on f δ that is out of p2−1 (U ). Hence,
the path p2  f δ starts at a point in U and is partially out of U. However, this is not pos-
sible since p2  f  δ = p1  δ, and p1 δ is a path entirely in U since δ is a path entirely in W.

We proved that f ( Z ) = f (W ) = V . Now it is easy to confirm that f |W = ( p1 |W )  ( p2 |V )−1 .

Since ( p1 |W ) and ( p2 |V )−1 are homeomorphisms, it follows that so is f |W : Z → V .

In the next corollary we state that the universal cover is a cover of any other cover. This
is the reason why simply connected covering spaces are called universal.

Corollary 7. Let ( X1 , p1 ) be a universal cover of X, and let ( X 2 , p2 ) be any cover of X.

Choose x1 ∈ X1 and x 2 ∈ p2−1 ( p1 ( x1 )), and let p1 be the unique lift of p1 sending x1 ∈ X1 to
x 2 ∈ X 2. Then ( X1 , 
p1 ) is a universal cover of X 2.

Proof. Follows immediately from Proposition 6.

Lifts of paths are paths (Theorem 1, 15.2), but lifts of loops need not be loops. Theorem 2
implies the following criterion for lifts of loops to be loops.

K12146_C015.indd 386 2/12/15 4:49 PM

 15.3 Lifts of Mappings   ◾    387  

Corollary 8. Let ( X  , p) be a covering of X, let α be a loop at x ∈ X , and let x ∈ p −1 ( x ). The

 , x )).
lift α starting at x is a loop at x if and only if [α] ∈ p∗ (π1 ( X

Proof. We leave it as an exercise (Exercise 5).

Exercises
1. Flip a few pages forward and consider the space X and its covering ( X  , p) described in
Example 3 and the bottom of Illustration 15.28, Section 15.4 (the Hawaiian earrings).
Let the space Y be the subspace of  2 as shown in Illustration 15.21 below: it consists
of the line segments sk connecting the origin and the point (1, π 3(2k−1 ) ) (in polar coordi-
nates), k = 1, 2, 3,… . Let f be the mapping Y → X that wraps sk once around the k-the
circle of X counting from the largest inwards, and sending the horizontal line segment
to the point common to all circles in X. Describe all lifts of f with respect to ( X  , p).

(1,π/3)

(1,π/6)

(1,π/12)

(1,π/24)

Illustration 15.21

2. Let (( X  , x 0 ), p) be a covering space of ( X , x 0 ) and let f : (Y × Z ,( y0 , z 0 )) → ( X , x 0 )

be a continuous mapping which can be lifted to f : (Y × Z ,( y0 , z 0 )) → ( X  , x 0 ).
Define f1 : (Y , y0 ) → ( X , x 0 ) by f1 ( y ) = f ( y , z 0 ), and define f 2 : ( Z , z 0 ) → ( X , x 0 ) by
f 2 (z ) = f ( y0 , z ). Show that f1 can be lifted to  f1 satisfying 
f1 ( y0 ) = x 0 , and show that
 
f 2 can be lifted to f 2 satisfying f 2 (z 0 ) = x 0.

3. The space X is a torus T 2 with one point deleted. Show that there exists no continuous
mapping p : T 2 → X such that (T 2 , p) is a covering space of X.
4. View the Klein K bottle as P 2 # P 2, and view P 2 # P 2 as a filled rectangle with edges
identified according to the cyclic word a 2b 2. We know that the fundamental group of
K is 〈a, b ; a 2b 2 〉. Find a covering space of the Klein bottle corresponding to the sub-
group generated by ab. Deduce that this subgroup is infinite cyclic.
5. (Corollary 8) Let ( X  , p) be a covering space of X, and let α be a loop at x ∈ X .
−1
Choose x ∈ p ( x ). Show that the lift α starting at x is a loop at x if and only if
 , x )).
[α] ∈ p∗ (π1 ( X

K12146_C015.indd 387 2/12/15 4:49 PM

 388   ◾    Covering Spaces, Part 1

6. Show that every continuous f : S 2 → S1 is nullhomotopic. Generalize: show that if X is sim-

ply connected (and path connected), then every continuous f : X → S1 is nullhomotopic.
7. Prove that if (Y, p) and (Z, q) are two covering spaces of a space X corresponding to the
same subgroup H of π1 ( X ), then there is a covering homeomorphism f : Y → Z .
8. Find a simple criterion such that if a triangulation of each of the spaces X given below
satisfies it, then there is a coloring of the vertices in the triangulation with every two
vertices on a single edge colored with distinct colors. Justify!
(a) X is the chain of finitely many, or countably infinitely many spheres, each two
consecutive sphere touching at a point (as in the Illustration 15.22).

Illustration 15.22 Illustration 15.23

 
(b) X =  2 ∪  ∪ Sn,m  , where Sn ,m is the sphere ( x − n)2 + ( y − m)2 + ( z − 1 4 )2 = 116
 n, m∈ ×  
(see Illustration 15.23).
9. Show that if a triangulation of a sphere is such that every vertex is incident to an even
number of edges, then we can color the edges in three colors so that every vertex is
incident to edges of precisely two colors.
10. Find a triangulation of the torus such that every vertex is incident to an even number of
triangles, yet there is no coloring of the vertices such that the vertices on each triangle
are of different colors.

15.4 Covering Spaces and Homotopy

We take a further look at the problem of how changes of the base space affect coverings.
The properties and examples we encounter in this section are also helpful in visualizing
more complicated covering spaces.
 , p) be a covering space of X, let A be a path connected and locally
Proposition 1. Let ( X
path connected subspace of X, and let C be a path component of p −1 ( A). Then (C , p |C ) is a
covering space of A.

K12146_C015.indd 388 2/12/15 4:49 PM

 15.4 Covering Spaces and Homotopy   ◾    389  

Proof. First we prove that p |C is onto A. Choose c ∈C , and denote a = p(c ). Since A is path
connected, there is a path α from a to any other b ∈ A. Since C is a path component of p −1 ( A),
the lift α of that path starting at c must be entirely in C. In particular α (1) ∈C . It follows that
p(α (1)) = b, and so p |C is onto A.
Now we establish existence of elementary neighbourhoods of points in A with respect
to p |C. Choose an elementary neighborhood U of an element x ∈ A and with respect to the
cover ( X , p). Let V be the path component of U ∩ A containing x. Since we have assumed
that A is locally path connected, and since U ∩ A is open in A, it follows from Proposition
2 in 6.5 that V is open in A.

~
X
W
C
D

U X
A

Illustration 15.24

We now show that V is an elementary neighborhood of x with respect to the cover (C , p |C ). Since
p |C is onto A, ( p |C )−1 (V ) ≠ ∅. Let D be one path component of ( p |C )−1 (V ). Let W be a path
component of p −1 (U ) in X  having a nonempty intersection with D (see Illustration 15.24).
Since D is path connected, and since V ⊂ U , it follows that D ⊂ W . Since p |w is a homeo-
morphism, so is p |D : D → p(D ) . Clearly, p(D ) ⊂ V . Assume for a moment that there is
v ∈V \ p(D ), and consider any path β from a point p(z ) ∈ p(D ), z ∈D , to v. Then, since D
is the path component of p −1 (V ) containing z, the lift of β starting at z must be entirely in
D. In particular, so is the end point w of that lift. But then p(w ) = v , and this tells us that
v ∈ p(D ), contradicting the assumption that v ∈V \ p(D ). We proved that p(D ) = V , so
that p D is a homeomorphism D → V .

The last thing to note is that C is locally path connected (Exercise 7 in 15.1).

Example 1: The Plane and the Torus

The (unbounded) cylindrical surface X  , together with the covering map p as
indicated in Illustration 15.25 (wrapping the cylinder around the torus), is a
covering space of the torus X. The plane, together with the mapping q (wrapping
the plane around the cylinder X  as indicated in Illustration 15.25) is in turn a

covering space of X .

K12146_C015.indd 389 2/12/15 4:49 PM

 390   ◾    Covering Spaces, Part 1

~
p
X = 2

A ~
X
X

Illustration 15.25

It is then visible that ( 2 , p  q) is a covering space of the torus X.

Pay attention to the subspace A of X consisting of the bouquet of two circles shown
on the surface X (see Illustration 15.25). It follows from Proposition 1 that every cover-
ing (Y , r ) of the torus X has a subspace A  such that ( A , p | ) is a covering of A. This is vis-
A
ible in Illustration 15.25: A  = p ( A) is a line with attached circles and A
−1  = ( p  q)−1 ( A)
is a rectangular net of lines. Notice that despite ( 2 , p  q) being the universal cover of
X, the space A  is not the universal cover of the bouquet A of two circles.
If we regard X as the quotient space of I × I with the boundary edges identified as shown
in Section 4.4, then the mapping p  q sends every pair (u, v ) ∈ 2 to the point ( x , y ) ∈ X
satisfying ( x , y ) − (u, v ) ∈  × . Similarly, if we view the 3-torus T 3 = S1 × S1 × S1 as
the quotient space of the unit cube I × I × I with opposite faces identified in the obvi-
ous way, then the Euclidean space  3, together with the mapping that sends every triple
(u, v , w ) ∈ 3 to the point ( x , y , z ) ∈T 3 satisfying ( x , y , z ) − (u, v , w ) ∈  ×  × , makes
a universal cover of the 3-torus. Analogous claim holds for higher dimensions. ☐

Example 2: Torus and Euclidean Geometry (Side Note)

One side effect of the fact that n is the universal cover of the n-torus T n = S
1
S1
×
 × S1
×

n
is that the Euclidean geometry over n can be transferred through the covering map to
a Euclidean geometry over the n-torus (see Illustrations 15.26 and 15.27).

Illustration 15.26 The view in a 2-torus (represented as  2 with points to be identified according
to Example 1).

K12146_C015.indd 390 2/12/15 4:49 PM

 15.4 Covering Spaces and Homotopy   ◾    391  

Illustration 15.27 The view in a 3-torus; looking straightforward and seeing one’s back infinitely often.
☐

Example 3: Covering of a Covering May Not Be a Covering

The reader should not infer from Example 1 that if ( X  , p) is a covering space of X, and
 , q) is a covering space of X
if ( X  , p  q) is necessarily a covering space of X. That
 , then ( X
claim is false, and a standard counterexample is introduced in Illustration 15.28.

~
X
…
…

Illustration 15.28 The base space X consists of an infinite sequence of circles of decreas-
ing radii touching at a single point (Hawaiian earrings, encountered earlier). The cover-
ing space X  consists of a line () touching an infinite sequence of copies of the base
space (the touching points correspond to the integers).

The covering map p : X  → X wraps the line around the largest circle (via t 
(cos 2 πt , sin 2 πt ), t ∈), and sends each Hawaiian earring touching the line in X  homeo-
morphically (and in the obvious way) onto the part of X without the largest circle. The
space X  has a two-sheeted covering ( X  , q) (meaning, q−1 ( x ) = 2 for every x ∈ X
 ), such
 , p  q) is not a covering space of X (Exercise 8). The deficiency of the space X will be
that ( X
pinpointed in Section 16.4. ☐

We now pay attention to the problem of how homotopy of the base space affects its covering
spaces. We start with two lemmas.

K12146_C015.indd 391 2/12/15 4:49 PM

 392   ◾    Covering Spaces, Part 1

Lemma 2. Let f : ( X , x ) → (Y , y ) be continuous. For every homotopy F : X × I → Y relative

to {x} and such that F X ×{0} = f , we have that F∗ (π1 ( X × I ,( x ,0))) = f∗ (π1 ( X , x )).

Proof. Since, for every [α] ∈π1 ( X , x ) , f∗ ([α]) = F∗ ([(α , c0 )]) , where we recall that c0 is
the constant loop at 0, it follows that f∗ (π1 ( X , x )) ⊂ F∗ (π1 ( X × I ,( x ,0))).  We now prove
that F∗ (π1 ( X × I ,( x ,0))) ⊂ f∗ (π1 ( X , x )). Denote H = f∗ (π1 ( X , x )) and let p0 be the projec-
tion X × I → X × {0}. Take a loop α : I → X × I based at (x, 0). It suffices to prove that it is
homotopic to the loop p0 α relative to {0, 1}, since then F∗ ([α]) = F∗ ([ p0  α]) = f∗ ([ p0  α]),
and f∗ ([ p0 α]) ∈ H . Proving that p0 α and α are homotopic relative to {0,1} is easy. The
homotopy G : I × I → X × I that will do the work is defined as follows: G(s , t ) = Pt (α(s )),
 ( x , t ) if r ≥ t
where Pt ( x , r ) =  ; that is, Pt projects the part of X × I above or at X × {t }
 ( x , r ) if r < t
onto X × {t }, and keeps the rest unmoved. The mapping G(s , t ) is continuous. It is clear that
G(s , 1) is α(s) (since P1 is the identity), that G(s,0) is p0 α , and that the homotopy is relative
to {0, 1}.

Lemma 3. Let ( X , p) be a cover of X , x ∈ X , x ∈ p −1 ( x ), and let H = p∗ (π1 ( X , x )). Suppose

further that f , g : X → X are such that f  { x } g , and f∗ ( H ) ⊂ H. Then the (unique) lifts ϕ
and ψ of ϕ = f  p and ψ = g  p , respectively, satisfying ϕ ( x ) = x = ψ ( x ), are homotopic
relative to {x }.

Proof. By Theorem 6 in 10.4 f∗ and g ∗ are the same, so that g ∗ ( H ) ⊂ H . The existence
and uniqueness of the lifts ϕ and ψ starting at x follows from the previous theory. We
are looking at the following diagram (Illustration 15.29).

~
~ ~
X X
~
ψ
p p
f
X X
g

Illustration 15.29

Suppose now that F : X × I → X is the assumed homotopy from f to g relative {x}. Then, by
Lemma 2 we have that F∗ (π1 ( X × I ,( x ,0))) = f∗ ( H ) ⊂ H .
Define a mapping ω : X × I → X × I as follows: ω(z , t ) = ( p(z ), t ) . It is obvious that ω
is continuous, and that ω( X  × I ) = p( X ) × I . Consider F  ω : X  × I → X . It is obviously
continuous. Notice that F∗ (ω ∗ (π1 ( X  × I ,( x ,0)))) ⊂ F∗ (π1 ( X × I ,( x ,0))) ⊂ H . Hence the
 × I → X has a unique lift G that sends ( x ,0) to x.
mapping F  ω : X  We have arrived at the
following diagram (Illustration 15.30).

K12146_C015.indd 392 2/12/15 4:49 PM

 15.4 Covering Spaces and Homotopy   ◾    393  

~ G ~
X×I X

ω p

X×I F X

Illustration 15.30
We see: F  ω( x , t ) = F (( p( x ), t )) = F ( x , t ) = x . Hence the restriction of F ω on {x } × I is
the constant path at x. Since the restriction G { x }× I is the unique lift of that path starting at x , it
must be the constant path, and hence G is a homotopy relative to {x } from G X ×{0} to G X ×{1} .
Observe that G X ×{0} is the lift of F  ω X ×{0} sending ( x ,0) to x , and that G X ×{1} is
the lift of F  ω X ×{1} sending ( x ,0) to x . Finally, since F ω(z ,0) = F (( p(z ),0)) = f ( p(z )) ,
we have F  ω X ×{0} = f  p . Symmetrically, we have F  ω X ×{1} = g  p . This gives that G is a
homotopy from ϕ  to ψ as in the statement of this lemma. That is what we wanted.

Proposition 4. Let H1 be a subgroup of π1 ( X1 , x1 ) and let H 2 be a subgroup of π1 ( X 2 , x 2 ). Further,

 1 , p1 ) be a covering of X1 corresponding to H1, and let ( X
let ( X  2 , p2 ) be a covering of X 2 corre-
sponding to H2. If f : X1 → X 2, h : X 2 → X1 are homotopy equivalences such that h  f  { x1 } id X1,
f  h  { x 2 } id X2 , and such that f∗ ( H1 ) = H 2, then there is a homotopy equivalence g : X 1 → X  2 mak-
ing the diagram in Illustration 15.31 commute.
~ g ~
X1 X2

p1 p2

f
X1 X2

Illustration 15.31

Proof. Since f is a homotopy equivalence, we have that f∗ : π1 ( X1 , x1 ) → π1 ( X 2 , x 2 ) is an

isomorphism. By symmetry, so is h∗. Since f∗ and h∗ are mutual inverses, h∗ ( H 2 ) = H1.
It follows from our assumptions that ( f  p1 )∗ (π1 ( X 1 , x1 )) = p2 (π1 ( X 2 , x 2 )), and so there is a
unique lift g of f  p1 sending x1 to x 2 . For the time being we have that g is continuous and the
diagram in Illustration 15.31 commutes. We will now show that g is a homotopy equivalence.
Symmetrically, there is a unique lift k of h  p2 sending x 2 to x1. We have the commuta-
tive diagram shown in Illustration 15.32a. In Illustration 15.32b we also show the mappings
id X1 and id X1, the latter being the lift of id X1  p1 sending x1 to x1.

idX~
1

~ g ~ k ~ ~ ~
X1 X2 X1 X1 X1
k g

p1 p2 p1 p1 p1

f h h f
X1 X2 X1 X1 X1

idX1

Illustration 15.32a Illustration 15.32b

K12146_C015.indd 393 2/12/15 4:50 PM

 394   ◾    Covering Spaces, Part 1

We are within the hypotheses of Lemma 3, with the additional provision that both h  f and
id X1 send H1 bijectively onto H1. It follows from Lemma 3 that id X 1 and k  g are homotopic.
Symmetry forces id X 2 and g  k to be homotopic as well, thus completing the proof.

Corollary 5. If the base spaces X1 and X 2 in Proposition 4 are homotopically equivalent and
if ( X 1 , p1 ) and ( X 2 , p2 ) are universal coverings of X1 and X 2, respectively, then these coverings
must be homotopically equivalent.

Proof. In this case the other conditions are satisfied automatically, since the associated groups
H1 and H 2 are both trivial.

Corollary 6. If X1 is a deformation retract of X 2 and if the inclusion i : X1 → X 2 is such that

i∗ ( H1 ) = H 2 (H1 and H2 are as in the statement of Proposition 4), then their coverings are
homotopic.

Proof. The inclusion and the associated retraction are (in case of deformation retracts)
homotopy equivalences.

Example 4
Consider the Klein bottle and its two coverings as depicted in Illustration 15.33.

~
X = 2
p

~
X
X

Illustration 15.33 The covering maps p and q are evident, paying due attention to the orienta-


 and X.
tion of the lines in the grid for both X

Observe that the plane is a universal cover of the Klein bottle X. Since the connected
sum P 2 # P 2 of two projective planes is homotopically equivalent to a Klein bottle, it
follows from Corollary 5 that the plane is a universal cover of P 2 # P 2. This is faintly
surprising, since the universal cover for P 2 is the sphere (Exercise 6, 15.2). ☐

Example 5: Klein Bottle and Euclidean Geometry (Side Note)

Keep denoting the Klein bottle by X. It follows from Example 4 above and from Exercise 11
in 15.1 that  3 is a universal cover of the compact 3-manifold X × S1. Consequently, as was

K12146_C015.indd 394 2/12/15 4:50 PM

 15.4 Covering Spaces and Homotopy   ◾    395  

the case with torus and 3-torus (Example 2), the Euclidean geometry over  2 and  3 can be
transferred to a model of Euclidean geometry over X and X × S1, respectively, (where lines
in  2 and  3 are mapped via the covering maps onto “lines” in X and X × S1 , respectively).
In Illustration 15.34 we illustrate a view of X × S1 from within.

Illustration 15.34 The view of oneself in Klein × S1. Notice the raised hand: the orientation
undergoes alternating changes at a distance.

Example 6: Torus Minus a Point

Let X be the space obtained by removing a point from the torus. Then X is homo-
topically equivalent to the space depicted in Illustration 15.35 which in turn is
homotopically equivalent to a bouquet R of two circles. The universal cover Y of
the bouquet R is shown in Illustration 15.5. It follows from Corollary 5 that the
universal cover of X should be “similar looking” to Y. What this exactly means is
left as an exercise (Exercise 2).

Illustration 15.35

Exercises
1. Describe the universal cover of the projective plane P 2 minus a point.
2. Let X be the space obtained by removing a point from the torus. Depict and describe
the universal cover of X.

K12146_C015.indd 395 2/12/15 4:50 PM

 396   ◾    Covering Spaces, Part 1

3. Describe the universal cover of S 2 \ {(0,0, −1),(0,0,1)}.

4. Find a cover of T 2 \ B (torus with an open disk deleted, Illustration 14.11 in Section
14.2) corresponding to a free group of countably infinite rank.
5. Describe the universal covering space of (a) the complement of a point in  3, (b) the
complement of a line in  3, and (c) the complement of S1 in  3.
6. Describe the universal cover of the complement of the unknot in an open 3-ball that
contains it. [Hint: Do not forget to homotopically deform first; use Exercise 6 in 15.1
and Corollary 5.]
7. Describe (and sketch) the universal cover of the connected sum of two tori.
8. Show that the space X 
 in Illustration 15.28 has a 2-sheeted covering ( X
 , q ) such that

 −1
( X , p  q ) is not a covering of X. (A covering is n-sheeted if | p ( x )| = n for every
x∈X.)
9. Generalize Theorem 2 in 15.2 as follows. Let ( X  , p ) by a covering space of X , x ∈ X ,
−1  , x )). Let Y be a path connected space, and let y ∈Y .
x ∈ p ( x ), and H = p∗ ( π1 ( X
Suppose f , g : Y → X are such that f ( y ) = g ( y ) = x and f  { y } g . Finally, assume in
addition that f∗ ( H ) ⊂ H . Prove that the unique lifts f and g sending y to x are
homotopic relative to { y }. [Hint: you may want to use Lemma 2 in this section and
Theorem 2 in 15.3.]
 , p ) be a covering space of X.
10. Let ( X
 , x ) → π ( X , x ), n ≥1 is always
(a) Prove that the induced homomorphism p∗ : πn ( X n
one-to-one (or monomorphism).
(b) Prove that the induced homomorphism p∗ : πn ( X , x ) → π ( X , x ) , n ≥ 2 is always
n
onto (or epimorphism). Deduce that the higher (n ≥ 2) homotopy groups of spaces
and their coverings are always isomorphic.
(c) Show that the spaces depicted in Illustration 14.63–14.66 of Section 14.5 have isomor-
phic higher (n ≥ 2) homotopy groups. (This settles the claim made at the end of Section
14.5.)
(d) Let X be either the torus or the Klein bottle. Prove that for every n ≥ 2, πn ( X ) is the
trivial group.
11. The projective n-space by P n is obtained by identifying each pair of antipodal points
in the unit n-sphere Sn. Show that for every m ≥ 2, πm ( P n ) ≅ πm ( Sn ).

K12146_C015.indd 396 2/12/15 4:50 PM

 Chapter 16

Covering Spaces, Part 2

W e will further explore the structure of covering spaces and base spaces through
their connection with the associated fundamental groups, uncovering along the
way a useful link between the fundamental groups of the base spaces and the groups of
covering transformations. By the end of this chapter we will give necessary and sufficient
conditions for the existence of (universal) coverings. Along the way we will encounter a few
consequences. One such notable example is the Borsuk–Ulam Theorem.

16.1 Covering Spaces and Sheets

A brief historical note: The first thorough elaboration of the connections between fundamental
groups and covering spaces was given in the book that we have mentioned a few times by now:
Seifert’s and Threlfall’s A Textbook in Topology.

We start this section by describing the effect of changing the base point x 0 of a covering
space ( X , p) of X on the group p∗ (π1 ( X  , x 0 )). Bear in mind that, by Proposition 1 in 15.3,
 , x 0 )) is always isomorphic to π1 ( X
p∗ (π1 ( X  , x 0 ).

Proposition 1. Let ( X , p) be a covering space of X. Fix x ∈ X and

x1 , x 2 ∈ p −1 ( x ). Then for every path α from x1 to x 2 , we have x2 ~
X
 , x 2 )) = [ p  a]−1 p∗ (π1 ( X
p∗ (π1 ( X  , x1 ))[ p  a]. α

 , x1 )), H = p (π ( X  , x 2 )), and β = ( p α ) x1

Proof. Denote H1 = p∗ (π1 ( X 2 ∗ 1
(see Illustration 16.1). We claim that H 2 = [β ] H1 [β ]. Choose [γ ]∈H 2
−1

p
so that [γ ] = p∗ ([δ]) for some [ δ ] ∈π1 ( X  , x 2 ). Then, αδα −1 is a
−1
loop at x1 , and p∗[αδα ] ∈ H1 . On the other hand, p∗[αδα −1 ] = x
[( p  α )( p  δ )( p  α −1 )] = [β ][ γ ][β ] , and so [β ][ γ ][β ] ∈ H1 . Hence β X
−1 −1

[ γ ] ∈[β ]−1−1H1 [β ], and thus −1H 2 ⊂ [β ]−1 H1 [β ]. By symmetry, H1 ⊂

[β ] H 2 [β ] , which yields [β ] H1 [β ] ⊂ H 2. Hence [β ] H1 [β ] = H 2 . ◼
−1 Illustration 16.1

397

K12146_C016.indd 397 2/12/15 12:32 PM

 398   ◾    Covering Spaces, Part 2

Proposition 2. Let ( X  , p) be a covering space of X. Fix x ∈ X , x 0 ∈ p −1 ( x ), and

denote G = π1 ( X , x ), H = p∗ (π1 ( X  , x 0 )) . Then { gHg −1 : g ∈G} = { p∗ (π1 ( X
 , x )) : x ∈ p −1 ( x )}.
In words, the family of all conjugates of p∗ (π1 ( X  , x 0 )) in π1 ( X , x ) is equal to
 , x )) : x ∈ p ( x )}.
{ p∗ (π1 ( X −1

Proof. Proposition 1 implies that { gHg −1 : g ∈G} ⊃ { p∗ (π1 ( X  , x )) : x ∈ p −1 ( x )} . To prove

the opposite inclusion choose a loop β at x, so that [β ] H [β ] is an arbitrary element
−1

of { gHg : g ∈G} . Lift β to a path α starting at x 0 . Then α(1) ∈ p −1 ( x ) . Denote α(1) = x1.
−1

We have [β ] H [β ] = [ p  α]−1 p∗ (π1 ( X

−1
 , x 0 ))  p  α  , and it follows from Proposition 1 that
 
−1
[ p  α] p∗ (π1 ( X  , x1 )). Since x1 ∈ p −1 ( x ) we have that [β ]−1 H [β ] ∈
 , x 0 ))  p  α  = p∗ (π1 ( X
 
 , x )) : x ∈ p −1 ( x )}.
{ p∗ (π1 ( X

Our next immediate goal is to express the cardinal number p −1 ( x ) explicitly in terms
 , x 0 )) and π1 ( X , x ). Proposition 2 does not give us an answer to that
of the groups p∗ (π1 ( X
problem since, in general, { p∗ (π1 ( X  , x )) : x ∈ p −1 ( x )} ≤ p −1 ( x ) .

 , p) be a covering space of X. Fix x ∈ X and x ∈ p −1 ( x ) . Let α and β

Proposition 3. Let ( X
be two loops at x, and denote their respective lifts starting at x by α and β. Then α and
β end at the same point if and only if [α ][β ] ∈ p∗ (π1 ( X
−1
 , x )) .

Proof. By Corollary 8 in 15.3, the lift αβ  −1

of αβ −1 is a loop at x if and only if
( )
 , x ) , which in turn is equivalent to [α][β]−1 ∈ p∗ π1 ( X
[αβ −1 ] ∈ p∗ π1 ( X
 = αβ
(
 , x ) . On the other)
hand, by uniqueness of lifts, αβ −1   , where the lift β starts at the end of α . By exer-
−1 −1

  −1
cise 5 in 15.2, β = β , and so we showed that [α][β]−1 ∈ p∗ π1 ( X
−1
( )
 , x ) if and only if αβ 
−1

is a loop at x , and the latter happens if and only if α and β end at the same point.

Corollary 4. With X and ( X , p) as above, for every x ∈ X , and for every x ∈ p −1 ( x ), p −1 ( x )

 , x )) in π1 ( X , x ) .
is equal to the index of p∗ (π1 ( X
(Recall that the index of a subgroup H of G, denoted [G : H ], is the number of right (or the
number of left) cosets of H in G.)

Proof. Denote for simplicity H = p∗ (π1 ( X  , x )). Consider the mapping f that sends a coset
[α ] H of H in π1 ( X , x ) to the end point of the lift α starting at x . Then f is well defined
and, by Proposition 3, it is one-to-one. On the other hand, it is obvious that f is onto,
since the path connectedness of X  allows a path λ from x to any other point y in p −1 ( x ),
establishing that, f ([ p λ]H ) = y .

 , x )) in π1 ( X , x ) does not depend on the

Observe that, by Corollary 4, the index of p∗ (π1 ( X
−1
choice of x in p ( x ).
At this stage we can already extract purely group-theoretical exercises (Exercise 5). In Chapter 17
we will use the theory of covering spaces to derive more complex group theoretical results.

Corollary 5. The cardinality p −1 ( x ) does not depend on x.

K12146_C016.indd 398 2/12/15 12:32 PM

 16.1 Covering Spaces and Sheets   ◾    399  

(We remind the reader that we are assuming that the base spaces and their covering
spaces are path connected and locally path connected.)

Proof. Let x1 , x 2 ∈ X , and let λ be any path from x1 to x 2 . Then, ϕ :[α] → [λαλ −1 ] is an iso-
morphism from π1 ( X , x 2 ) onto π1 ( X , x1 ) (Proposition 2, in 10.4). Choose x1 ∈ p −1 ( x1 ) and
let λ be the lift of λ starting at x1 . Since λ ends at x 2 , the path λ ends at some x 2 ∈ p −1 ( x 2 ).
Since π1 ( X , x 2 ) and π1 ( X
 , x1 ) are isomorphic and since p∗ is a monomorphism, the

groups p∗ (π1 ( X , x 2 )) and p∗ (π1 ( X  , x1 )) are isomorphic too. However, that is not sufficient
to get what we want (since two isomorphic subgroups of a group G do not always have the
same index in G – can you find a simple counterexample?) In the next paragraph we will
prove that the restriction of φ on p∗ (π1 ( X  , x 2 )) is an isomorphism onto p∗ (π1 ( X  , x1 )). That
would suffice: since it is obvious that if ϕ : G1 → G2 is an isomorphism and H is a subgroup of
G1 , then [G1 : H ] = [G2 : ϕ( H )]. Hence [π1 ( X , x 2 ) : p∗ (π1 ( X  , x 2 ))] = [π1 ( X , x1 ) : p∗ (π1 ( X
 , x1 ))],
−1 −1
and by Corollary 4, we would have p ( x 2 ) = p ( x1 ) .

~
X
~ x~2
x~1 λ

x1 x2 X
λ
Illustration 16.2

We show, as we have stated, that ϕ( p∗ (π1 ( X  , x 2 ))) = p∗ (π1 ( X

 , x1 )) : Take any loop

α at x 2 , so that [ pα] is an arbitrary element of p∗ (π1 ( X , x 2 )) . We have ϕ([ p α]) =
[λ( p  α )λ −1 ] = [ p  λ ( p  α ) p  λ −1 ] = [ p  (λαλ
  −1 )] = p [λαλ
∗
  −1 ], where in the first step we used
the definition of φ, while the rest follows from the definitions of the operations we see. Since
  −1 is a loop at x , we have proven that ϕ( p (π ( X
λαλ  , x 2 ))) ⊂ p∗ (π1 ( X
 , x1 )). By symmetry,
1 ∗ 1
−1   
ϕ ( p∗ (π1 ( X , x1 ))) ⊂ p∗ (π1 ( X , x 2 )), hence, after applying φ, p∗ (π1 ( X , x1 )) ⊂ ϕ( p∗ (π1 ( X  , x 2 ))),
and we have achieved the equality.

 , p) of X is with n sheets (or, it is n-sheeted) if p −1 ( x ) = n for every x ∈ X .

The covering ( X
Corollary 5 makes this well defined. Note that we are not assuming that n is finite.

Exercises
1. Describe algebraically an n-sheeted covering of S1 for all n ∈+ . Describe all
2-sheeted coverings of the bouquet of two circles.
 , p) be a universal cover of a bouquet (as a CW-complex) made of countably
2. Let ( X
infinite many circles sharing a common vertex v. Is p −1 (v ) countable or uncountable?

K12146_C016.indd 399 2/12/15 12:33 PM

 400   ◾    Covering Spaces, Part 2

3. A group G acts on a set S if every g ∈G defines a permutation (also denoted by)

g : S → S satisfying the following two properties: (i) for every x ∈S ,1( x ) = x
(where 1 is the identity of G), and (ii) for every g 1 , g 2 ∈G and for every x ∈S ,
 , p) be a cover of X; fix an element x ∈ X , and choose
( g 1 g 2 )( x ) = g 1 ( g 2 ( x )). Let ( X
−1
x ∈ p ( x ). Show that the mapping defined by
[α ]  the terminal point of the lift of α starting at x
is an action of the group π1 ( X , x ) on the set p −1 ( x ).
4. Prove (using the theory in this section) that for every group G, for every subgroup H
of G, and for every g ∈G , [G : H ] = [G : gHg −1 ]. You may use the following result that
we will establish in a few sections: for every group G and every subgroup H of G, there
is a space X with π1 ( X , x ) = G and a covering X corresponding to H.

5. (a) Find a 2-sheeted covering space of the Klein bottle.

(b) Prove that the group 〈a, b ; a 2b 2 〉 has a subgroup of index 2 isomorphic to  ⊕ .
[Hint: do (a) first, but consider (b) as a hint for (a).]
6. Construct a 2-dimensional CW-complex that is not simply connected and that does
not have an n-sheeted cover for every integer n ≥ 2. (You may use the fact that, say,
the group  with addition has no subgroups of finite index.)
7. Prove that for every integer n ≥ 3 the sphere S 2 is not an n-sheeted cover of some
Hausdorff space. [Hint: Exercise 10 in 15.1.]
8. Let X be the space consisting of two touching spheres. Show that there is no space Y
and no mapping p: X → Y such that (X, p) is an n-sheeted cover of Y, n ≥ 2.

16.2 Covering Transformations

A brief historical note: The definition of covering transformation appears in 1913 in an article
by Hermann Weyl.

In this section we introduce covering transformations and relate them to the fundamental
groups of the associated spaces.
Throughout this section ( X , p) will denote a covering space of a space X.
A continuous homeomorphism f : X →X  is a covering transformation if the diagram
shown in Illustration 16.3 commutes.
~ f ~
X X

p p

Illustration 16.3

K12146_C016.indd 400 2/12/15 12:33 PM

 16.2 Covering Transformations   ◾    401  

Denote the set of all covering transformations X →X  by A( X  , p), or, when p is under-
 ). For every f1 , f 2 ∈ A( X
stood from the context, denote it by A( X  ), the composition f1  f 2 ,

is also in A( X ).

 ) is a group under composition of covering transformations.

Proposition 1. A( X

Proof. Exercise 1.

The following properties of covering transformations are simple consequences of our

theory so far.

Proposition 2.
 ) operates without fixed points. That is, if f ( y ) = y for some f ∈ A( X
(a) A( X  ) and some

y ∈ X , then f is the identity.
 , x1 )) = p∗ (π1 ( X
(b) For every x ∈ X , and for every x1 , x 2 ∈ p −1 ( x ) , p∗ (π1 ( X  , x 2 )) if and

only if there is a (unique) f ∈ A( X ) such that f ( x1 ) = x 2 .

Proof.
(a) This is a consequence of Theorem 2 in 15.3, the uniqueness part, since the elements of
A( X ) are lifts of the covering map p (see Illustration 16.3).
(b) ⇒ Suppose p∗ (π1 ( X  , x1 )) = p∗ (π1 ( X
 , x 2 )). By Theorem 2 in 15.3 this implies exis-
 
tence of unique lifts f , g : X → X of p with respect to ( X  , p) such that f ( x1 ) = x 2
and g ( x 2 ) = x1 . Since f  g ( x 2 ) = x 2 and g  f ( x1 ) = x1 it follows from part (a) that
f  g = id X = g  f , and so f ∈ A( X  ).

⇐ If f ∈ A( X  , x1 )) ⊂ p∗ (π1 ( X
 ) is such that f ( x1 ) = x 2 , then p∗ (π1 ( X  , x 2 )) (Theorem 2
−1
in 15.3). Since f ( x 2 ) = x1, the same argument gives p∗ (π1 ( X  , x 2 )) ⊂ p∗ (π1 ( X , x1 )).

The normalizer N(H) of a subgroup H of a group G is the largest subgroup of G such that
H  N ( H ). It is an easy exercise to show that N ( H ) = { g ∈G : g −1Hg = H }.

Remark. Do not confuse the normalizer N(H) of a subgroup H of a group G with the nor-
mal closure NC(H) of a set H as defined in 12.2.
Let’s explore further the condition p∗ (π1 ( X  , x1 )) = p∗ (π1 ( X  , x 2 )), x1 , x 2 ∈ p −1 ( x ) .
According to Proposition 1 in 16.1 we have that p∗ (π1 ( X  , x 2 )) = [ p  α]−1 p∗ (π1 ( X
 , x1 ))[ p  α],
for every path α in X  from x1 to x 2 . Hence, the condition p∗ (π1 ( X  , x1 )) = p∗ (π1 ( X
 , x 2 ))
is equivalent to p∗ (π1 ( X , x1 )) = [ p  α ] p∗ (π1 ( X , x1 ))[ p  α ] for every path α in X from
−1
  
x1 to x 2 . This is equivalent to the claim that [ pα] is in the normalizer of the subgroup
 , x1 )) of π1 ( X , x ). In the light of Proposition 2(b), we have proven the following.
p∗ (π1 ( X

Proposition 3. An element [ g ] ∈π1 ( X , x ) is in the normalizer of p∗ (π1 ( X  , x1 )) if and only if

there is a covering transformation f ∈ A( X  ) such that f ( x1 ) = x 2 , where x 2 is the terminal
point of the lift of the loop g starting at x1 .

K12146_C016.indd 401 2/12/15 12:33 PM

 402   ◾    Covering Spaces, Part 2

Example 1
 , p) of X is shown in Illustration 16.4 (p is determined through
The covering space ( X
the labels and the orientation of the edges). The points x ∈ X and x 0 ∈ p −1 ( x ) are as
shown, and we denote H = p∗ (π1 ( X  , x 0 )) .

a a a
~
X
b y–1 b b y0 b b y1
x–1 x0 x1
a a a

X
x
a b

Illustration 16.4
It is evident that every covering transformation of X  maps x 0 to one of the ele-
ments of the set {x j : j ∈} (see Illustration 16.4). Hence a loop at x in X determines a
homotopy class that is in N(H) if and only if its lift starting at x 0 ends at some point
in {x j : j ∈}. For example, the loop at x that winds once around the ­a-circle is such,
and so is the loop that winds twice around the b-circle. On the other hand, say, the
loop that winds once around the b-circle is not in N(H), since its lift starting at x 0
does not end at {x j : j ∈}. ☐

Before we state the main theorem of this section, we remind the reader that for every
 , p) of a space X, every covering transformation f of ( X
covering ( X  , p) is a lift of the cover-
ing map p. By Theorem 2 in 15.3, the action of f is uniquely specified by the image of a single
 . According to the proof of that theorem, if f ( z ) = w , then for every z1 ∈ X
point in z ∈ X ,
and for every path α from z to z1, f (z1 ) must be the terminal point of the lift of pα
starting at w.

Theorem 4. Let ( X  , p) be a covering of X, let x be in p −1 ( x ), and let N(H) be the

(
normalizer of H = p∗ π1 ( X )
 , x ) in G = π1 ( X , x ). Define a mapping ϕ : N ( H ) H → A( X
 ) as
follows: ϕ([ g ]H ) is the covering transformation f that sends x to the terminal point
of the lift of the loop g starting at x . Then φ is a group isomorphism.
Remark. Bear in mind that the stipulation that the covering transformation f sends x to
the terminal point of the lift of the loop g starting at x,  does not mean that, for some other
−1
u ∈ p ( x ) , f (u) is necessarily the end point of the lift of g starting at u. The paragraph pre-
ceding the statement of this theorem tells us what we should do to find f (u).

Proof of Theorem 4.
Step 0. Since [ g ] ∈N ( H ), Proposition 3 guarantees that the covering transformation f exists.
Step 1 (φ is well-defined). Suppose [ g 1 ]H = [ g 2 ]H . This means that [ g 1 ]−1[ g 2 ] ∈H , and
so, by Proposition 3 in 16.1, the lifts of g 1 and g 2 starting at x must end at the same point.

K12146_C016.indd 402 2/12/15 12:33 PM

 16.2 Covering Transformations   ◾    403  

Step 2 (φ is a group homomorphism). We need to show that ϕ(([ g 1 ]H )([ g 2 ]H )) =

ϕ([ g 1 ]H )  ϕ([ g 2 ]H ).
For a loop g at x, we will denote by g y its unique lift starting at some y ∈ p −1 ( x ) .
Notice that ϕ(([ g 1 ]H )([ g 2 ]H )) = ϕ([ g 1 ][ g 2 ]H ) = ϕ([ g 1 g 2 ]H ). By definition ϕ([ g 1 g 2 ]H ) is
that f ∈ A( X  ) which maps x to the terminal point of ( g 1 g 2 )x (see Illustration 16.5).

~
f (x)
~ f ~
X (g1 g2)x~ X
~ ~
x° x

p p
g2
x
g1
X
Illustration 16.5

On the other hand ϕ([ g 1 ]H )  ϕ([ g 2 ]H ) is the composition f1  f 2 where f1 ( x ) = (

g 1 )x (1)

and f 2 ( x ) = ( g 2 )x (1) (Illustration 16.6).

~ ~
f1 f1 (x) ~ ~ f2 f2 (x)
~ X X ~
X X
~ ~
(g1)x~ ~ ~
(g2)x~
x° x x° x

p p p p
g2 g2
x
g1
X x
g1
X

Illustration 16.6

~
~ f1 (f2 (x) )
f2 (x) f1
~ ~
f1 (x) f1°(g2)x~ ~
X ~ (g2)x~ X
x

p p
g2
x
g1
X

Illustration 16.7

Pay attention to Illustration 16.7: the path f1  ( g2 )x shown in the right-hand copy of X 

obviously ends at f1 ( f 2 ( x )) . On the other hand, since both ( g 2 )x and f1 commute with
p, this same path is the lift of g 2 starting at f1 ( x ). Looking at Illustration 16.5 again, we
observe that f ( x ) = (
g 1 g 2 )x (1), where, in order to get ( g 1 g 2 )x (1), we first lift g 1 to (
g 1 )x ,

then we lift g 2 to the path starting at ( g 1 )x (1) = f1 ( x ). Hence f ( x ) = f1 ( f 2 ( x )), and by the
  
uniqueness of lifts we have that f = f1  f 2 as mappings.

K12146_C016.indd 403 2/12/15 12:33 PM

 404   ◾    Covering Spaces, Part 2

Step 3 (φ is one-to-one). Suppose ϕ([ g 1 ]H ) = ϕ([ g 2 ]H ). This means that the lifts  g1
and g2 starting at x end at the same point. By Proposition 3 in 16.1, [ g 1 ][ g 2 ]−1 ∈H . So
H[ g 1 ] = H[ g 2 ], and since [ g 1 ],[ g 2 ] ∈ N ( H ), it follows that [ g 1 ]H = [ g 2 ]H .
Step 4 (φ is onto). Take any f ∈ A( X ). If f ( x ) = y , then y ∈ p −1 ( x ). Let γ be a path from
x to y. Then p  γ is a loop at x ∈ X and ϕ([ p  γ ]H ) = f .

The following corollary is an important consequence, and we will refer to it frequently.

Corollary 5. If p∗ (π1 ( X  , x )) is a normal subgroup of π1 ( X , x ), then A( X

 ) is isomorphic to
π1 ( X ,x )  
 ,x )) . In particular, if ( X , p ) is the universal covering of X, then A( X ) is isomor-
p∗ ( π1 ( X
phic to π1 ( X , x ).

Proof. When p∗ (π1 ( X  , x )) is a normal subgroup of π ( X , x ), then the normalizer of

1

p∗ (π1 ( X , x )) is π1 ( X , x ), and the result follows by Theorem 4.

Example 2
The universal cover of S1 is ( , p), where p(t ) = ( cos2πt ,sin2πt ), and the correspond-
ing group of covering transformations is the group { f j :  →  : j ∈, f j ( x ) = j + x }
of translations through integer values. This is an infinite cyclic group generated by
f1. It follows from Corollary 5 that π1 (S1 ) is also an infinite cyclic group. (Compare
this argument with the lengthy proof of Theorem 1 in Section 11.1.) ☐

Example 3.
We recall that the projective plane P 2 is the quotient space of the sphere S 2 obtained
by identifying all pairs of antipodal points. The universal cover of the projective plane
is (S 2 , p) , where p sends each x ∈S 2 to its equivalence class [x ] ∈P 2 . The only non-
identity covering transformation is the one that sends each point x of S 2 to its antipo-
dal point − x . It follows that the group of all covering transformations in this case is
isomorphic to  2 . Consequently, π1 ( P 2 ) is also isomorphic to  2 . ☐

As the last two examples suggest, the fundamental groups of spaces with relatively simple
universal covers can be accessed through the groups of covering transformations of their
universal covers. See Exercises 6–7.

Exercises
 ), with composition, is a group.
1. Prove Proposition 1: A( X
2. Find the group of covering transformations of the covering space of the bouquet of
two circles shown in Illustration 16.4.
3. Let k ∈+ . Find the group of covering transformations of the covering space (S1 , p)
of S1 , where p is defined by p(cos2πt , sin2πt ) = (cos2kπt , sin2kπt ), t ∈I .

K12146_C016.indd 404 2/12/15 12:33 PM

 16.3 Covering Spaces and Groups Acting Properly Discontinuously   ◾    405  

4. Consider the group G = 〈a, b ; bab −1 = a 2 〉 .

(a) Let H be the subgroup of G generated by a, and let K be the subgroup of G gen-
erated by a2 . Show that K is a proper subgroup of H, and show that H and K are
conjugates in G.
(b) Let X be a space such that π1 ( X , x ) ≅ G and let (( X , x1 ), p) be a covering of
(X, x) corresponding to the subgroup K. (For example, we may take X to be
the 2-dimensional CW-complex as constructed in the proof of Proposition 2,
14.4; the existence of covering spaces corresponding to subgroups of G for such
spaces X will be established in 16.4.) Show that there exists x 2 ∈ p −1 ( x ) such that
 , x 2 ), p) corresponds to the subgroup H of G.
(( X
 , x1 ) → ( X
(c) Show that there is a unique lift f : ( X  , x 2 ) of p.

(d) Show that the lift f in part (c) is not a one-to-one, and hence conclude that the
commutativity of the diagram in Illustration 16.3, together with the assumption
that f is continuous, does not imply that f is a homeomorphism.
5. Show that if f is a continuous mapping making the diagram in Illustration 16.3 com-
mutative, and if X is a cover of X with finitely many sheets, then f is a homeomorphism.
6. (a) Describe the group of covering transformations of  2 , considered as the universal
cover of the torus as described in Example 1, 15.4.
(b) Find a simple set of generators of this group and a presentation on that set of
generators.
7. (a) The plane  2 together with the covering map described in Example 4, 15.4, is
the universal covering space of the Klein bottle. Describe the group of covering
transformations of this covering space.
(b) Find directly a simple set of generators of this group, and a presentation on that
set of generators.
(c) Use Tietze transformations to change the presentation found in (b) to any previ-
ously found presentation of the fundamental group of the Klein bottle.

16.3 Covering Spaces and Groups Acting

Properly Discontinuously

A brief historical note: Groups acting properly discontinuously have been used for a long
time. For example, the action of the group G on  2 in Example 3 was used by Gauss in the
early stages of his theory of elliptic functions. Constructing new spaces from the actions of
such groups by identifying the elements in each orbit—what we will do in this section—can
be found in the 19th century works by Henri Poincaré, Felix Klein (1849–1925), and Hermann
Schwartz (1843–1921).

K12146_C016.indd 405 2/12/15 12:33 PM

 406   ◾    Covering Spaces, Part 2

 , x )) is a normal subgroup of π1 ( X , x )
 , p) of X is regular if p∗ (π1 ( X
A covering space ( X
−1
for some x ∈ X and some x ∈ p ( x ) . Alternatively, regular covering spaces correspond
to normal subgroups of π1 ( X , x ) . Exercise 6 implies that if ( X  , p) is a regular cover for
−1
some x ∈ X and some x ∈ p ( x ), then it is a regular cover for every x ∈ X , and for every
x ∈ p −1 ( x ).

Example 1
The universal cover of any space is obviously regular. For example, such is the cover
of the bouquet of two circles X depicted in Illustration 15.5 in Section 15.1. The next
two covering spaces shown there (illustrations 15.6 and 15.7 in Section 15.1) are also
regular covers of X. To which normal subgroups of the free group π1 ( X , x ) do these
two covers correspond? (Exercise 7). ☐

The following consequence of the results in Section 16.2 will help us recognize regular
covers more clearly.

Proposition 1. If ( X  , p) is a regular cover of X if and only if for every x ∈ X , and for every
−1
x1 , x 2 ∈ p ( x ), there is a covering transformation f ∈ A( X  ) such that f ( x1 ) = x 2 .

Proof. ⇒ This is a direct consequence of Proposition 3 in 16.2, since for any path λ from
x1 to x 2 , p λ is a loop at x which, by the hypothesis of this proposition, must define an
element in the normalizer of p∗ (π1 ( X  , x1 )) in π1 ( X , x ).
⇐ Under the premise of the implication, it follows from Proposition 3 in 16.2 that
 , x1 )) in π1 ( X , x ); hence,
every homotopy class in π1 ( X , x ) is in the normalizer of p∗ (π1 ( X
 , x1 ))  π1 ( X , x ).
p∗ (π1 ( X

Informally speaking, Proposition 1 implies that regular coverings are symmetric in the
foll­owing sense: for every fixed z ∈ X , the view on a regular cover of X from every point in
p −1 (z ) is the same. For example, the cover of the bouquet of two circles in Illustration 16.4
is not regular (compare with the one shown in Illustration 15.7, Section 15.1).

So far we have discussed procedures for describing and analyzing covering spaces of a
given base space. We will now explore a procedure that will go in the other direction: given
a space Y, we will construct another space Z, such that Y is the cover of Z.
Let G be a subgroup of the group of homeomorphisms Y → Y , and define a relation ~G
as follows: y1 ~G y 2 if there is f ∈G such that f ( y1 ) = y 2 . This is an equivalence relation
on Y (Exercise 16 in 4.2). The quotient space Y ~G will be denoted by Y G . Denote the quo-
tient mapping Y → Y G by p. If Y is path connected and locally path connected, then so is
G (Exercise 2). The following question emerges on its own: when is (Y, p) a covering space
Y

of Y G ? The following example tells us that in order to achieve this we need to impose some
condition on G.

K12146_C016.indd 406 2/12/15 12:33 PM

 16.3 Covering Spaces and Groups Acting Properly Discontinuously   ◾    407  

Example 2
Take Y = S1 and G1 to be the cyclic group of homeomorphisms Y → Y consisting of
the identity mapping and the reflection with respect to the x-axis. Then Y G1 is homeo-
morphic to a closed interval, and the quotient mapping p1 : Y → Y G1 is not a covering
map (since, for example, p1−1 ( x ) is not a fixed number). On the other hand, if we take
G2 to be the cyclic group (of order 2) generated by the reflection with respect to the
origin (i.e., ( x , y )  (− x , − y ) for all ( x , y ) ∈S1), then Y G2 is a circle and p2 : Y → Y G2
wraps Y twice around Y G2 . Hence, (Y , p2 ) is a covering space of Y G2 . ☐

It is relatively easy to identify what prevented (Y , p1 ) to be a cover of Y G1 in Example 2: it

follows directly from the definition of Y G that if (Y, p) is a cover of Y G , then each g ∈G
must be a covering transformation. Since covering transformations are lifts (of p), their
uniqueness implies that the only one that has fixed points is the identity. In the case of
the group G1 , the reflection with respect to the x-axis has two fixed points, yet it is not an
identity.
However, stipulating that the non-identity elements of G do not have fixed points is not
enough to guarantee that (Y, p) is a cover of Y G (Exercise 9). What we need is provided in
the following definition. An explanation is given in the theorem that follows it.
A subgroup G of the group of all homeomorphisms Y → Y acts properly discontinu-
ously on Y, if for every y ∈Y , there exists an open neighborhood U of y, such that for every
distinct g 1 , g 2 ∈G , we have that g 1 (U ) ∩ g 2 (U ) = ∅ .

Theorem 2. Let Y be a path connected and locally path connected space, let G be a group of
homeomorphisms Y → Y acting properly discontinuously on Y, and denote the quotient
mapping Y → Y G by p. Then (Y, p) is a regular covering space of Y G and G is the group of
covering transformations of (Y, p).

Proof.
Step 1. First we prove that (Y, p) is a covering space of Y G . We will employ the standard
notation for equivalence classes: [ y] will stand for a generic element of Y G . Given such
an element, we are searching for an elementary neighborhood with respect to (Y, p). Since
G acts properly discontinuously, there exists an open neighborhood V of y in Y, such that
g 1 (V ) ∩ g 2 (V ) = ∅ for every distinct g 1 , g 2 ∈G . Denote the path component of V contain-
ing y by W. By Proposition 2 in 6.5 W is also open. It is then evident that W also satisfies
g 1 (W ) ∩ g 2 (W ) = ∅ for every distinct g 1 , g 2 ∈G . Denote U = p(W ). We claim that U is the
desired elementary neighborhood of [ y]. We split this claim into (i)-(ii).
(i) The restriction of p on every path component of p −1 (U ) is a homeomorphism.
Notice first that W is a path component of p −1 (U ) . This is true since a path λ in p −1 (U )
having at least one point in W must be entirely in W, or else {λ −1 (W ), ∪ λ −1 ( g (W ))}
g ∈G \{1}
would be a disconnection of the interval I. That p W is continuous and onto is given from

K12146_C016.indd 407 2/12/15 12:33 PM

 408   ◾    Covering Spaces, Part 2

Y
the construction of W and U. The fact that g 1 (W ) ∩ g 2 (W ) = ∅
for every distinct g 1 , g 2 ∈G , implies that p W is one-to-one. g1(W)
W
g2(W)
This map is open because of the following short argument. For
every open subset V of W, p −1 ( p(V )) = ∪ g (V ) . Since each g is a p
g ∈G
homeomorphism, it follows that ∪ g (V ) is open in Y. The defini- Y/
g ∈G U G
tion of open sets in quotient spaces then forces p(V) to be open.
All this implies that p W is a homeomorphism. Other com- Illustration 16.8
ponents of p −1 (U ) must be of the type g(W), for some g ∈G
(Illustration 16.8). Hence p g (W ) = ( p W )  g −1, and so it is a homeomorphism, since it is the
product of two homeomorphisms. This settles (i).
(ii) U is path connected and open.
It is path connected since W is such, and since U = p(W) (Proposition 2 in 6.4). Its open-
ness follows directly from the definition of quotient topology, since p −1 (U ) = ∪ g (W ) and
g ∈G
since all g(W) are open in Y.
Step 2. We prove that G = A(Y ). It is obvious that G ⊂ A(Y ). For the opposite inclusion
take any f ∈ A(Y ), and any point y ∈Y . Since p( y ) = p  f ( y ) = p( f ( y )), it follows that
there exists some g ∈G such that g ( y ) = f ( y ). Since both g and f are lifts of p, the unique-
ness of lifts implies g = f . Hence f ∈G .
Step 3. Finally we prove that (Y , p) is a regular covering space of Y G . Fix y ∈Y . We need
to show that p∗ (π1 (Y , y )) is a normal subgroup of π1 (Y G , p( y )) , that is, that every element
[α] ∈π1 (Y G , p( y )) is in the normalizer of p∗ ( π1 (Y , y )) in π1 (Y G , p( y )). By Proposition 3 in
16.2 this is equivalent to showing that there is a covering transformation of (Y  , p) sending
y to the terminal point of the lift of α starting
at y. This is clear: the lift of any loop starting
at y ends at some point z in p −1 ( p( y )) , hence
p( z ) = p( y ), hence there is a g ∈G such that
g ( y ) = z , hence (by Step 2) there is a covering
transformation sending y to z.

Example 3
In Illustration 16.91, we show a tiling of
X =  2. The group G of all translational
symmetries of this tiling is generated by
the translations tu and tv along the vec-
tors u and v (as shown). Then a, b ; ab = ba
is a presentation of G under the mapping
Illustration 16.9

1
Artwork by Jos Leys (vectors extra).

K12146_C016.indd 408 2/12/15 12:34 PM

 16.3 Covering Spaces and Groups Acting Properly Discontinuously   ◾    409  

a  tu , b  tv (Example 4 in 12.1). Observe that G acts properly discontinuously on X,

and so (X, q), where q : X → X G is the quotient map, is a universal cover of X G . It follows
from Theorem 2 that π1 ( X G ) is isomorphic to G. Since X G is a torus, this shows again
that a, b ; ab = ba is a presentation for the fundamental group of the torus. ☐

Example 4.
The space X depicted in Illustration 16.10 (the wide arrows) l
is a subspace of  2 . The group G of symmetries of X is
generated by the translation tb along the vector b, and the
glide reflection g that is the composition of the translation
f = t a along the vector a 2 followed by the reflection with
2
respect to the vertical line l. The group G has a presentation a
a, b ; aba −1b under the mapping a  g , b  f (Exercise
7 in 12.1). Since X G is in this case a Klein bottle, we have
b
found yet another presentation for that 2-manifold.
Further, notice that g 2 is the translation ta along the Illustration 16.10
vector a (Illustration 16.10). The subgroup H of G gener-
ated by ta and tb is of index 2 in G, hence H  G . It follows (see Exercise 16) that
the cyclic group G H of order 2 acts properly discontinuously on the quotient space
X , giving a covering X
H of
X . We saw in Example 3 that X
H G H is a torus; hence the
torus is a two sheeted covering of the Klein bottle. Tracing carefully how the original
action of the group of symmetries filters into the action of G H on the torus yields the
covering map X H → X G described in Illustration 16.11 and in its caption.

Illustration 16.11 The action of G

H on the
f torus: m denotes any fixed meridian circle,
and A and B range through the pairs of antip-
A
e odal points on m. The only non-identity ele-
m ment gH sends A to B, and sends the points on
the equatorial circle e passing through A to
the points on the equatorial circle f through B
B
in the opposite direction.
☐

Here is an important type of groups that act properly discontinuously on a space.

Proposition 3. Let (Y, p) be a covering space of X.

(a) The group A(Y, p) of covering transformations of (Y, p) acts properly discontinuously on
Y, so that Y, together with the quotient map q : Y → Y A(Y , p ), is a regular covering space.
(b) Moreover, if (Y, p) is a regular cover, then there is a homeomorphism f : X → Y A(Y , p )
such that the diagram in Illustration 16.12 commutes.

K12146_C016.indd 409 2/12/15 12:34 PM

 410   ◾    Covering Spaces, Part 2

p q

X Y
f A(Y, p)
Illustration 16.12

Remark. Informally, we claim in Proposition 3(b) that there is no essential difference

between the coverings (Y, p) and (Y,q), if the former is regular.

Proof.

(a) Start with y ∈Y , and denote x = p( y ). Let U be an elementary neighborhood of X,

and denote by V the path component of p −1 (U ) . Then V is open (by the ubiquitous
Proposition 1 in 6.5).
We claim that for every non-identity ϕ ∈ A(Y , p), ϕ(V ) ∩ V = ∅. That would suffice
since if ϕ1 (V ) ∩ ϕ 2 (V ) ≠ ∅ for some distinct ϕ1 and ϕ 2 , then ϕ −2 1  ϕ1 (V ) ∩ V ≠ ∅.
We know that each non-identity covering transformation has no fixed points. Now
suppose ϕ(V ) ∩ V ≠ ∅, ϕ ≠ idY . Choose z ∈ϕ(V ) ∩ V , so that z is in V and z = ϕ(v )
for some v ∈V . Since φ has no fixed points, z ≠ v. On the other hand, since p  ϕ = p,
we have that p( z ) = p(ϕ(v )) = p(v ). This contradicts the fact that the restriction of p
on V should be a homeomorphism (hence, one-to-one).
(b) Define f : X → Y A(Y , p ) by f ( x ) = q( p −1 ( x )) for every x ∈ X . If y1 , y 2 ∈ p −1 ( x ) , then,
by Proposition 1 (which is applicable since (Y, p) is regular), there is g ∈ A(Y , p)
such that g ( y1 ) = y 2 . Hence q( y1 ) = q( y 2 ) , and so f is well defined. It follows imme-
diately that the above diagram commutes. It is equally easy (and requires a similar
short argument) to check that f is one-to-one and onto. If U is an open subset of
A(Y, p), then f −1 (U ) = p(q −1 (U )) , and this set is open since q is continuous and p is
open. Hence f is continuous. Similarly, f is open.

Since the space X and its covering space (Y, p) are a part of the setup of Proposition 3,
the usage of Theorem 2 in this context is not in merely constructing a pair consisting of
a covering space and a base space. The point here is that (Y,q) (where q : Y → Y A(Y , p ) is the
quotient mapping) is a regular covering space of the new base space Y A(Y , p ) , even when
(Y, p) is not a regular covering space. We illustrate this in the following example.

Example 5
Consider again the covering space of Example 1 in 16.2 (shown again in Illustration
16.13; the covering map p is defined by the labels of the edges). The group
A( X ) = A( X
 , p) is the infinite cyclic group generated by the translation sending x 0 to
x1 . The quotient space X A( X ) is shown in Illustration 16.14.

K12146_C016.indd 410 2/12/15 12:34 PM

 16.3 Covering Spaces and Groups Acting Properly Discontinuously   ◾    411  

a a a
~
X
a ~ ~ a
b y–1 b b y0 b b y1 X/A(X)
x–1 x0 x1
a a a b

x
X u
v
a b b

Illustration 16.13 Illustration 16.14

In this example, we can confirm the regularity of the covering ( X  , q) , where


q : X → A( X ) is the quotient map, in many ways. Here are three of them:

X

( )
(a) (Direct check) The group G = π1 X A( X ) , u has a presentation 〈a, b 2 , bab −1 ; ∅〉
(see Proposition 3 in 13.2). It is clear (using the same Proposition 3 in 13.2 and
the definition of q) that H = q∗ (π1 ( X  , x 0 )) is the subgroup of G generated by
2n −2n 2n+1 −3− 2n
{b ab : n ∈} ∪ {b ab : n ∈} . In order to prove that H is normal in G, it
−1
suffices to show that c dc ∈ H for every generator c ∈{a, b 2 , bab −1 } in the above pre-
sentation for G, and for every generator d ∈{b 2nab −2n : n ∈} ∪ {b 2n+1ab −3−2n : n ∈}
for H. Since a ∈H , the claim is obvious when c = a. For c = b 2 , it is evident that
c −1dc ∈ H when d = b 2n ab −2n , as well as when d = b 2n +1ab −3− 2n . Finally, conjugating
a generator in H by c = bab −1 again gives an element in H, since bab −1 = (bab −3 )b 2 ,
since bab −3 ∈H , and by the previous case.
(b) (Relying on Proposition 1) It is easy to see that for every two elements in q −1 (u), i.e., for
any two vertices labeled by some x j in Illustration 16.13, there is a covering transformation
sending one to the other. The regularity of the covering now follows from Proposition 1.
(c) Proposition 3 and Theorem 2 immediately imply that X  together with the quotient

mapping q : X → A( X ) is a regular covering of the space X A( X ) .

X

Observe that X A( X ) is also a covering of the bouquet X of two circles shown in

Illustration 16.14 (where, again, the covering map is defined through the labels of
the edges in the graph). This is not a coincidence, as we will see in the next section
(Proposition 1 in 16.4). ☐

Exercises
1. Show that if π1 ( X , x ) is abelian, then every covering space of X is regular.
2. Show that if Y is path connected and locally path connected, and G is a group of
homeomorphisms Y → Y , then Y G is also path connected and locally path connected.
 , p) is a cover of X such that p −1 ( x ) = 2 for some x ∈ X , then ( X
3. Show that if ( X  , p)
is a regular covering space of X.

K12146_C016.indd 411 2/12/15 12:34 PM

 412   ◾    Covering Spaces, Part 2

4. Consider the subspace X = {( x , y ,0) : x , y ∈} ∪ {( x ,0, z ) : x , z ∈} of  3 (two per-

pendicular planes). Let G be the group generated by the translation along the vector
(1,0,0) . Describe the space Y = X G and use the theory of covering spaces to identify
explicitly the group π1 (Y , y ).
5. Let X be the bouquet of two circles as in Illustration 15.4. Hence, π1 ( X , v ) = 〈a, b ; ∅〉 .
(a) Sketch a 3-sheeted non-regular covering space of X. [In this part, as well as in
parts (b) and (c), describe the covering map p by labeling and orienting the edges
of the covering.]
(b) Sketch a 3-sheeted regular covering space of X.
(c) Sketch a covering space of X corresponding to the subgroup of π1 ( X , v )
generated by {a 2 , a 2b 2 } . (More precisely, sketch a covering space ( X , p) of X
 
and indicate there a point v ∈ X such that p∗ (π1 ( X , v )) is generated by the set
{a 2 , a 2b 2 }. )
n n
(d) A word a1b δ1 a2 b δ2 an b δn is balanced if Σ  j = 0 and Σ δ j = 0 . It is easy to
j =1 j =1
show that the subset of 〈a, b ; ∅〉 consisting of all elements that can be rep-
resented by balanced words is a subgroup of 〈a, b ; ∅〉 . Describe the covering
(X , p) of X corresponding to this subgroup. Is ( X  , p) regular? Fund the group
 ) of all covering transformations of ( X
A( X  , p).
 , x ), p) be a regular covering space of (X,  x) for some x ∈ X and some x ∈ p −1 ( x ).
6. Let (( X
Show that ( X  , p) is a regular covering space for every x ∈ X , and for every x ∈ p −1 ( x ).

7. Let X be a bouquet of two circles (Illustration 15.4 in 15.1), and let ( X1 , p1 ) and ( X 2 , p2 )
be the two covering spaces of X shown in Illustrations 15.6 and 15.7, Section 15.1, respec-
tively. Describe the normal subgroups ( p1 )∗ (π1 ( X1 , x1 )) and ( p2 )∗ (π1 ( X 2 , x 2 )) of
π1 ( X , v ), where v is the only vertex of X (as shown in Illustration 15.4, Section 15.1), and
where x1 ∈ p1−1 (v ), x 2 ∈ p2−1 (v ).
8. Let (Y , p) be the universal cover of the bouquet X of two circles (labeled a and b;
see Illustrations 15.4 and 15.5 in Section 15.1). Hence A(Y ) ≅ π1 ( X ) . Describe or
depict Y H if H is the subgroup of A(Y) corresponding to the following subgroup G
of π1 ( X , v ) = 〈a, b ; ∅〉 :
(a) G is the cyclic group generated by a.
(b) G is the normal closure of a in π1 ( X ) ; that is, the elements of G are all conjugates
in π1 ( X ) of a.
9. Find a space Y and a group G of homeomorphisms Y → Y such that each g ∈G , g ≠ 1,
acts without fixed points, and such that (Y, p), p is the quotient mapping, is not a
covering of Y G .

K12146_C016.indd 412 2/12/15 12:34 PM

 16.3 Covering Spaces and Groups Acting Properly Discontinuously   ◾    413  

10. Let (Y, p) be a covering space of X. Show that for every elementary neighborhood U in X,
for every path component V of p −1 (U ), and for every non-identity ϕ ∈ A(Y , p), ϕ(V )
is a path component of p −1 (U ) distinct (so, disjoint) from V.
11. A continuous mapping f : X → X is an involution if f  f = id X . Prove the following:
(a) If f :  2 →  2 is an involution without fixed points then the group {identity, f }
acts properly discontinuously.
(b) Every involution f :  2 →  2 must have a fixed point.
(c)  2 is never a 2-sheeted covering.
(d)  2 is never a 2n -sheeted covering.
12. Let (Y , p) be a covering space of X. Define h : Y A(Y , p ) → X by h([ y]) = p( y ), for every y ∈Y .
(a) Show that ( Y
A (Y , p ) )
, h is a covering space of X.
(b) Show that (Y, p) is a regular covering if and only if h is a homeomorphism.
13. The projective n-space P n is obtained from the sphere S n by identifying each pair of
antipodal points. Use the theory in this section to compute π1 ( P n ) .
14. The group G of symmetries of the 3-dimensional design shown in the following illus-
tration acts properly discontinuously on X =  3 . Show that X G is K × S1, where K is
Klein bottle, and use this to find a presentation for K × S1 (visualized in Illustration
14.51, Section 14.4). Show that S1 × S1 × S1 (together with a certain mapping) is a
two-sheeted covering space of K × S1 .

Illustration 16.15 Copies of the plane, as in Illustration 16.10, are arranged in  3 in an infinite
column of equidistant vertical planes.

K12146_C016.indd 413 2/12/15 12:34 PM

 414   ◾    Covering Spaces, Part 2

15. (a) Represent X = T 2 # T 2 as follows: start with the union of the sphere S 2 and two
vertical cylindrical surfaces intersecting the xy-plane at circles of radius 1 4 and
centered at the points ( − 1 2 , 0, 0 ) and ( 1 2 , 0, 0 ) respectively; then remove the four
open sets within the two cylinders from S 2 (see Illustration 16.16). Let g : X → X
be the reflection with respect to the origin. Then the group G = {id , g } acts prop-
erly discontinuously on X. Show that X G ≅ P 2 # P 2 # P 2 and conclude that T 2 # T 2
is a 2-sheeted covering of P 2 # P 2 # P 2 .
(b) Use the idea in part (a) to show that T 2 #T 2 # #T 2 is a 2-sheeted cover of

P 2 # P 2 # # P 2 . n−1

n

(c) Show that for every n ∈+ the group a1 ,a2 ,… , an ; a12a22 …an2 has a normal sub-
group of index 2 that is isomorphic to b1 ,c1 ,b2 ,c2 …,bn−1 ,cn−1 ; b1c1b1−1c1−1b2c2b2−1c2−1
bn−1cn−1bn−−11c n−−11 .

Illustration 16.16 A connected sum of two tori.

16. Let G be a group of homeomorphisms of a (connected and locally path con-

nected) space Y, acting properly discontinuously on it, and let H be a normal
subgroup of G.
(a) Show that if p : Y → Y H , then (Y, p) is a covering space of the quotient
quotient map

space Y H .
(b) Show that the following defines a properly discontinuous action of the group G H
on the space Y H : for every gH ∈G H , and for every z ∈Y , ( gH )([z]) = [ g ( z )] (where
[z ] = p( z ) ).
Y
(c) Show that H
G is the same as (homeomorphic to) Y
G .
H

Y
(d) Deduce that Y
H projection
 →Y G = H
G
H
is also a covering map.

K12146_C016.indd 414 2/12/15 12:34 PM

 16.4 Covering Spaces: Existence   ◾    415  

16.4 Covering Spaces: Existence

A brief historical note: The first textbook containing a proof of existence of covering spaces
under certain conditions seems to be Seifert and Threlfall’s book mentioned earlier.

We start with the following preliminary result.

Proposition 1. Let (X, p) be a covering space of Z, and let (X, q) be a covering space of Y, and
let r : Y → Z be a continuous mapping such that the diagram in Illustration 16.17 com-
mutes. Then (Y, r) is a covering space of Z.
X
Proof. Since p = r  q, and since p is onto, it follows that r is onto. Now q
we search for elementary neighborhoods with respect to (Y, r). Take p
z ∈Z and let U z be an elementary neighborhood with respect to p. Y
−1 1
Choose an element x ∈ p (z ) and consider the path component U x r
of p −1 (U z ) containing x. Since p −1 (U z ) is open, and since X is locally Z

path connected, U 1x is also open. Consider y = q( x ) : since q is a cov- Illustration 16.17

ering there is an elementary neighborhood U y of y. Denote the path
component of q −1 (U y ) that contains x by U x2 . By the same argument we have used for
U 1x , the set U x2 is also open. Hence, U 1x ∩ U x2 is open too. Take the path component U of
U 1x ∩ U x2 containing x: since X is locally path connected, the set U is open. Since U ⊂ U 1x ,
V = p(U ) is homeomorphic to U. Similarly, W = q(U ) is also homeomorphic to U. So W
is homeomorphic to V via the restriction of r over W (we have used the commutativity of
the diagram). Moreover, again because of the commutativity of the diagram, W ⊂ r −1 (V ) .
It remains to be shown that W is a path component of r −1 (V ).
Since V ⊂ U z , each path component of p −1 (V ) is contained in a path component of
−1
p (U z ). So the path component of p −1 (V ) containing x is in U x . Since p −1 (V ) ∩ U x = U
and since U is path connected, U is the path component of p −1 (V ) containing x. Since q
is continuous, W = q(U ) is path connected. Suppose it is not a path component of r −1 (V ).
Then there is an element a ∈r −1 (V ) such that a ∉W and such that there is a path α in
r −1 (V ) starting at y and ending at a. The lift of that path with respect to q starting at x
ends at some point out of U. This terminal point must be in q −1 (r −1 (V )) simply because
a ∈r −1 (V ). So, the terminal point of that lift is in p −1 (V ) (we have used the commutativ-
ity of the diagram again). But that would make U not a (maximal) path component of
p −1 (V ); contradiction.

The following related result will be needed later.

Proposition 2. Let (Y,  p) be a regular cover of X, and let H be a normal subgroup of the
group A(Y, p) of covering transformations. Define r : Y H → X by r ([ y]) = p( y ) , for every
y ∈Y . Then (Y H ,r ) is a regular cover of X.

K12146_C016.indd 415 2/12/15 12:34 PM

 416   ◾    Covering Spaces, Part 2

Proof. It is clear that r is well defined, since if [ y1 ] = [ y 2 ] then h( y1 ) = y 2 Y

q
for some h ∈H , which in turn implies that p( y1 ) = p( y 2 ).
Recall that the quotient map q : Y → Y H is a covering map (Theorem 2, p Y
H
and Proposition 3 in 16.3). The definition of r implies that r  q = p, and
r
so Illustration 16.18 is a commutative diagram. Hence, for every open X
subset U of X, we have r −1 (U ) = q( p −1 (U )). Since p is continuous, and
since q is open (Proposition 3, 15.1), it follows that q( p −1 (U )) is open, Illustration 16.18
and hence r is continuous. Proposition 1 now implies that (Y H ,r ) is a
covering space of X. It remains to be shown that it is regular.
Since (Y , p) is regular, we have that A(Y, p) and π1 ( X ) p∗ ( π1 (Y )) are isomorphic under the
isomorphism defined in Theorem 4 in 16.2. Let H′ be the image of H under this isomor-
phism. Since it is given that H  A(Y , p), it follows that H ′  π1 ( X ) p∗ ( π1 (Y )) . By one of the iso-
morphism theorems of group theory (see, for example, [64], page 50), there exists a normal
subgroup M of π1 ( X ) such that H ′ = M p∗ ( π1 (Y )).
We will show that r∗ ( π1 (Y H )) = M , and, hence that (Y H ,r ) is regular. Let α be a loop in X
such that [α] ∈ M . Recall the isomorphism ϕ : π1 ( X ) p∗ ( π1 (Y )) → A(Y , p) from Theorem 4 in 16.2
maps α to the covering transformation in A(Y, p) that sends every point y ∈Y to the terminal
point of the lift β of α starting at y. The covering transformations corresponding to the elements
in M are in H. Hence, there exists an element h ∈H such that h(β(0)) = β(1). This means that
q(β) is a loop in Y H . It follows from the commutativity of the diagram in Illustration 16.18 that
r∗ ([q∗ (β)] = [α], and so we have that M ⊂ r∗ ( π1 (Y H )) . Conversely, start with [α ] ∈r∗ ( π1 (Y H )) .
Then [α ] = r∗ [ δ ] for some loop δ in Y H . Hence, the lift β of δ with respect to (Y , q) starting
at some point β(0) = y ends at h(y), for some h ∈H . By Theorem 4 in 16.2 (since ϕ( H ) = H ′,
and since H ′ = M p∗ ( π1 (Y )) ) we have that p∗ (β) ∈M , and by the commutativity of the diagram
in Illustration 16.18, it follows that [α] = r∗[δ] ∈M . Hence, r∗ ( π1 (Y H )) ⊂ M .

The existence of a universal covering space has many important consequences. The next
two claims are such examples.

Theorem 3. If a space X has a universal cover, then for every x ∈ X , and for every subgroup
H of π1 ( X , x ), the space X possess a cover corresponding to H.

Proof. Let ( X  , p) be the universal cover of X, let x ∈ p −1 ( x ) , and let H be a subgroup of

π1 ( X , x ). By Corollary 5 in 16.2, the groups A( X  , p) and π1 ( X , x ) are isomorphic under
the isomorphism φ defined in Theorem 4, 16.2. Let H′ be the subgroup of A( X  , p) corre-
sponding to H under that isomorphism. By Proposition 3 in 16.3 the group A( X  , p) acts
properly discontinuously on X  ; hence so does its subgroup H′. By Theorem 2 in 16.3, ( X  , q ),
where q : X  → X H ′ is the quotient mapping, is a regular covering of X , and H ′ = A( X  , q).
H′
Since X 
 is simply connected, it follows from Corollary 5 in 16.2 that A( X , q) is isomorphic
to π1 (  H ′ , y ), where y = q( x ); hence π1 ( X H ′ , y ) is isomorphic to H′.
X

We now show that the space X H ′ , together with a covering map that we are about to con-
struct, is the desired covering space of X. Denote the element of X H ′ by [z] (the equivalence
class of z ∈ X  ). Define r : X H ′ → X by r ([z ]) = p( z ).

K12146_C016.indd 416 2/12/15 12:35 PM

 16.4 Covering Spaces: Existence   ◾    417  

First we show that r is well defined. Let [z1 ] = [z 2 ]. Then there is a covering transforma-
tion h ∈ H ′ of ( X  , q) such that h(z1 ) = z 2 . Since H′ is a subgroup of A( X  , p) , it follows from
the definition of covering transformations (and the commutativity of the diagram shown
in Illustration 16.19) that p(z1 ) = p(z 2 ).
The mapping r is continuous because p is continuous, and because ~
X
q is open (Proposition 3 in 15.1). The details are easy or obvious, so we q

leave them as an exercise (Exercise 3). p ~

X
Proposition 1 implies that r : X H ′ → X is also a covering map. H´
( 
X
)
It remains to be shown that r∗ π1 ( H ′ , y ) = H . Given [α ] ∈π1 ( H ′ , y ) ,

X
X
r

the loop α has a unique lift α (with respect to the covering ( X  , q))
starting at x . Denote the terminal point of α by z. By Corollary 3 in Illustration 16.19
15.2, z does not depend on the choice of the representative of the homo-
topy class [α]. Since q(z ) = q( x ) , there exists an element h ∈ H ′ such that h( x ) = z . Since
H′ is a subgroup of A( X  , p), h( x ) = z implies that p( x ) = p(z ) . Hence, p α is a loop at x,
and α is the lift of p α starting at x . It follows from the definition of the isomorphism φ
in Theorem 4, 16.2, that ϕ([ p  α ]) = h. Hence, [ p  α ] ∈H . Now notice that to get [ p  α ] we
started from [α] and traveled the diagram in Illustration 16.19 counterclockwise. By the
commutativity of that diagram, we have r∗ ([α]) = [ p  α ] , hence r∗ ([α]) ∈H . This proves that
( )
r∗ π1 ( X H ′ , y ) ⊂ H . The converse inclusion requires a similar argument: If [β ] ∈ H , then
ϕ ([β]) is equal to that h ∈ H ′ which sends x to the terminal point of the lift (with respect to
p) of β starting at x . It follows that q  β is a loop at y in X H ′ , and by the commutativity of the
( )
diagram in Illustration 16.19, r∗ ([q  β]) = [β], so that [β ] ∈r∗ π1 ( X H ′ , y ) .

Here is one more reason why the existence of the universal cover is important. Compare
with Example 3 in 15.4.

Proposition 4. Let X be a space that has a universal cover. If (Y, p) is a covering space of X,
and if (Z, q) is a covering space of Y, then ( Z , p  q) is a covering space of X.

 , r ) be the universal cover of X. By Proposition 6 in 15.3, the lift r : X → Y of

Proof. Let ( X
r with respect to (Y, p) is a covering map. By the same proposition, the lift r : X → Z with
respect to (Z, q) is also a covering map. Since p  q : Z → X satisfies r  ( p  q) = r , the result
now follows from Proposition 1.

Virtually the entire theory in the last two chapters is conditioned by the existence of the
covering spaces we consider. By what we have seen in this section so far, that, in turn, is
determined by the existence universal covering spaces. We close our discussion of covering
spaces by establishing sufficient and necessary conditions for the existence of the universal
cover of a space.
A space is semilocally simply connected if for every x ∈ X there exists an open neighbor-
hood U of x such that every loop at x with range in U defines the trivial element in π1 ( X , x ).

K12146_C016.indd 417 2/12/15 12:35 PM

 418   ◾    Covering Spaces, Part 2

This is equivalent to the statement that the inclusion in : U → X induces the trivial homo-
morphism in∗ : π1 (U , x ) → π1 ( X , x ) . (Yet another equivalent statement is given in Exercise 1.)
If X is assumed to be locally path connected—which we usually do—than the path compo-
nent of U containing x is open, and so, in that case, there is no loss in generality if we take
U to be path connected.
An example of a space that is not semilocally simply connected is
{ }
∞
X = ∪ ( x , y ) : ( x − n1 ) + y 2 = n12 (the Hawaiian earrings), as a subspace of  2 . It is
2

n=1
depicted in Illustration 15.28, Section 15.4 (see also Exercise 8, 15.4), where we have noted
 of X
 of X, and X
that it allows coverings, X  is not a covering of X. It follows
 , such that X
from Proposition 4 that X does not possess the universal cover.
The main theorem follows.

Theorem 5 (Existence of Universal Covers). Suppose X is a path connected, locally path

connected, and semilocally simply connected space. Then there exists a universal cover
for X.

Proof. First we construct the covering space, then we prove that it is indeed a universal
covering space.
Step 1 (The construction). Fix x 0 ∈ X . The set X  is the set of all homotopy classes of
paths in X starting at x 0 (and recall that, unless otherwise stated, our homotopies of paths
are always relative to {0,1}). We introduce a topology over X  by describing a basis that will
generate it.
The family B of all non-empty open, path connected subsets U of X, such that every
loop at a point in U is contractible in X, is a basis for the given topology over X. This fol-
lows from Proposition 4 in 3.3, and Exercise 1 below. For every [α] ∈ X  ending at x ∈ X ,
and for every U x ∈B containing x, define U  [α ] = {[αλ] ∈ X
 : λ a path in U x starting at x }.
(See Illustration 16.20)

x0
α
 [α ] of X
Illustration 16.20 The subset U  con-
X x Ux
λ sists of the homotopy classes of paths that
extend α within U x .

We now prove that the family B  of all such U

 [α ] is a basis for a topology over X .
Following Theorem 2 in 3.3, there are two conditions to check: (a) X = ∪ U  [α ], and
  U [α ] ∈B

(b) for every U  [α ] , V

 [β] ∈B  , and for every [δ] ∈U
 [α ] ∩ V
 [β] , there exists W
 [ γ ] ∈B such that
[δ] ∈W [γ ] ⊂ U [α ] ∩ V  [β ] .
Part (a) is a consequence of the fact that B covers X. To prove (b) assume U  [α ] , V 
 [β] ∈B
and [δ] ∈U  [α ] ∩ V [β] . This means that δ is homotopic to a path of type αλ1 , where α ends at
a point x1 in some U x1 ∈B , and λ1 is entirely within U x1 . At the same time δ is homotopic

K12146_C016.indd 418 2/12/15 12:35 PM

 16.4 Covering Spaces: Existence   ◾    419  

to a path of type βλ 2 , where β ends at a point x 2 in some Vx 2 ∈B , and λ 2 is entirely within

Vx 2 . Hence, αλ1 and βλ 2 are homotopic (and thus ending at the same point y). It follows
that y ∈U x1 ∩ Vx 2 . Since B is a basis, there exists Wy ∈B such that y ∈Wy ⊂ U x1 ∩ Vx 2 . It is
 [α ] ∩ V
 [δ ] ⊂ U
clear that [δ] ∈W  [β] (see Illustration 16.21).

x0
α Wy

Ux1
λ1
β X
λ2
Vx2
Illustration 16.21 Keep in mind that αλ1 and
βλ 2 are homotopic.

We have established that B  is indeed a basis for a topology over X

 . With this, X

becomes a topological space.
Step 2 (The covering map). Define p : X  → X by p ([α]) = α(1) . This is clearly a well-
defined mapping.
Denote the set of all homotopy classes of paths from x 0 to x by Γ. Keeping the notation
established in Step 1, we claim that p −1 (U x ) = ∪ U
 [α ] . We prove that p −1 (U x ) ⊂ ∪ U
 [α ] ;
[α ]∈Γ [α ]∈Γ
the opposite inclusion can be proven similarly. Let [β] ∈ p −1 (U x ). Then β(1) ∈U x , and if λ
is a path in Ux from x to β(1) (which exists since U x was taken to be path connected), then
[β ] = [βλ −1λ]. Since [βλ −1 ] ∈Γ (simply because βλ −1 ends at x), we have that [β] ∈U [βλ−1 ].
Consequently p −1 (U x ) is open in X  (as the union of open sets). Hence, p is
continuous.
Next, we prove that p is a covering map. We will do it in two mini-steps: first, we show
that sets U  [α ] , are pairwise disjoint, and then we
 [α ] , [α] ∈Γ , appearing in p −1 (U x ) = ∪ U
[α ]∈Γ
prove that each of them is homeomorphic to U x via p.
Let [α],[β] ∈Γ be distinct (both starting at x 0 and ending at x ∈U x ). Then
  [β] = ∅. Otherwise, suppose [ αλ1 ] = [βλ 2 ] ∈U
U [α ] ∩ U  [α ] ∩ U
 [β] . Then the paths λ1 and
λ 2 start at x and end at the same point. Since U x ∈B , these two paths must be homo-
topic. It then follows easily from that, and from [αλ1 ] = [βλ 2 ] , that [α] = [β], which gives
a contradiction. This ends the first mini-step. We will complete the proof that p is a
covering map by showing that each U  [α ] , [α] ∈Γ is homeomorphic to U x via p. The
only part of that claim that is not entirely obvious is that the restriction p |U [α ] is open.
The restriction is open, since, by the definition of the topology over X  , it is open over
  
every member of B that is a subset of U [α ], and since U [α ] is a union of such sets in B .
We have completed the justification that p is a covering map, and along the way have
proven that U x is an elementary neighborhood of x with respect to p.
 is path connected and locally path connected). Each member of the basis B
Step 3. ( X
is path connected. We have proven in Step 2 that each member of the basis B  for X


K12146_C016.indd 419 2/12/15 12:35 PM

 420   ◾    Covering Spaces, Part 2

is homeomorphic to a member in B. Hence the members of the basis B  are all path

connected sets. It follows that X is locally path connected.
We now prove the path connectedness of X  . Denote the homotopy class of the
constant path at x 0 ∈ X by [cx0 ]. In order to prove that X  is path connected it suf-
fices to show that for every [α] ∈ X there is a path f : I → X
 from [cx ] to [α]. We do
0

it as follows: extend the constant path at x 0 , moving farther and farther along α; the
associated homotopy classes of paths in each moment constitute the desired path
(Illustration 16.22); a precise argument follows.

α(1)

α(t3) α(t3)
[α]= f (1)
α(t2) α(t2) α(t2)
f (t3)
α(t1) f (t2) α(t1) α(t1)
α(t1)
f (0) f (t1)

x0 x0 x0 x0 x0

Illustration 16.22 A path between (homotopy classes of) paths: at the first moment we are at [c x ], then 0

we take (homotopy classes of) larger and larger chunks of α, until at the last moment we reach [α ].

Define f : I → X  by f (t ) = [βt ], where βt : I → X is a path satisfying βt (s ) = α(st ). Observe

that βt starts at x 0 and ends at α(t), in harmony with
α(1)
Illustration 16.22. It is evident that f (0) = [cx0 ], and that
f (1) = [α]. It remains to be shown that f is continuous. Let βt(1) = α(t)

U f (t ) ∈B be a neighborhood of f (t ) = [β ] . It consists of all
t Uβt(1)
homotopy classes of the paths that extend the path βt by a
path within U βt (1) , where, by definition, βt (1) = α(t ) . Since
βt
α is continuous, α −1 (U βt (1) ) is an open subset of I. Clearly,
x0
for every r ∈α −1 (U βt (1) ), we have α(r ) ∈U βt (1) . So, for every
r ∈α −1 (U βt (1) ), βr is an extension of βt by a path in U βt (1) Illustration 16.23
(Illustration 16.23). This means that [βr ] ∈U  f (t ) for every
r ∈α −1 (U βt (1) ), which in turn means that f (r ) ∈U  f (t ) for every r ∈α −1 (U β (1) ). Hence
( )
t
 f (t ) , and we have proven that f is continuous at any t ∈I .
f α −1 (U βt (1) ) ⊂ U
In Steps 2 and 3 we have proven that ( X  , p) is a covering space of X. To complete the
 , p) is a universal covering space of X.
proof we need to establish that ( X
Step 4. ( X is simply connected.)
By Proposition 1, 15.3, it suffices to show that p∗ (π1 ( X  ,[cx ])) = 1. By Corollary 8, 15.3,
0
this is equivalent to showing that, for every loop α at x 0 in X, if its lift α at [cx0 ] is a loop, then
α is contractible. Most of the work is done in Step 3: the path f constructed in Step 3 starts at
[cx0 ], and it is a lift of the path α. The latter claim is true since p  f (t ) = p([βt ]) = βt (1) = α(t ).

K12146_C016.indd 420 2/12/15 12:35 PM

 16.5 The Borsuk–Ulam Theorem   ◾    421  

If f is a loop, then f (1) = [cx0 ]. However, by construction, f (1) = [α]. So, [α] = [cx0 ], and α
is contractible.

For path connected and locally path connected spaces, the converse of Theorem 4 is also
true (Exercise 4).

Exercises
1. Show that X is semilocally simply connected if and only if for every x ∈ X , the class
of all open (path connected) neighborhoods U of x such that every loop at x in U is
contractible in X is a local basis at x.
2. A space X is locally simply connected if for every x ∈ X there is an open simply con-
nected neighborhood of x. Prove that there is a semilocally simply connected space
that is not locally simply connected. Deduce that there are spaces that are not locally
simply connected yet possess a universal cover.
3. Prove that the mapping r defined in the proof of Theorem 3 is continuous.
 , p) is the universal cover of X, then X is semilocally simply connected.
4. Show that if ( X
5. The space X shown in Illustration 16.24 consists of the interval [0,1] along the x-axis,
and a sequence {Cn } of circles parallel to the yz-plane, such that Cn is of radius n1 and
touching [0,1] at n1 . Describe a cover ( X  , q) of X
 , p) of X and a covering ( X  such that

( X , p  q) is not a cover of X.

Illustration 16.24

16.5 The Borsuk–Ulam Theorem

A brief historical note: The statement of the Borsuk–Ulam theorem is attributed to Stanislaw
Ulam (1909–1984). A proof was found in 1933 by Karol Borsuk (1905–1982).

In this section we exhibit a few applications of our theory.

A mapping f : S n → m is antipode-preserving if f ( x ) = − f (− x ) for every x ∈S n
(Illustration 16.25).

K12146_C016.indd 421 2/12/15 12:35 PM

 422   ◾    Covering Spaces, Part 2

f (–x )

(0,0)

f (x) = –f (–x) Illustration 16.25 An antipode preserving

mapping f : S1 →  2 .

Theorem 1 (Borsuk–Ulam). Let f : D n → n be continuous and such that, f Sn−1

is antipode-
preserving. Then there exists y ∈D n such that f ( y ) = (0, 0, …, 0).

Proof (only for n = 2). Suppose no such y exists.

f ((cos2 πt ,sin2 πt ))
Define a loop β : I → S1 by β(t ) = f ((cos2 πt ,sin2 πt )) . Since (the range of) f avoids (0,0), β
f ((1,0))
is a well-defined loop in S1 based at z = f ((1,0)) ∈S1. Because of the same reason, it follows
f (x )
that the mapping ϕ : D 2 → S1 defined by ϕ( x ) = f ( x ) is well defined and continuous. It is
an extension of β′: S1 → S1 defined by β′((cos 2πt ,sin 2πt )) = β(t ). It follows that β is homo-
topic to the trivial loop at z (see Exercise 15, section 10.4).
We will now show that our assumption (that no y as in the statement of the theorem exists)
implies that β is not homotopic to the trivial loop, and this would give us a contradiction.
Since f S1 ( x ) = − f S1 (− x ) for every x ∈S1, it follows that β(t ) = −β( 1 2 + t ) for every
t ∈[0, 1 2]. Denote β1 (t ) = β ( 2t ) , and β 2 (t ) = β ( 12 + 2t ) , for t ∈I . Hence, β = β1β 2 . It follows
from β(t ) = −β( 1 2 + t ) for every t ∈[0, 1 2] that β1 (t ) = −β 2 (t ) for every t ∈I . Observe that if
β1 starts at z then it ends at − z ; at the same time β 2 starts at − z , and it ends at z.
Assume that z = (1, 0); we will see at the end of the argument what needs to be mod-
ified if z ≠ (1, 0). Let (, p) be the universal cover of S1, where, as earlier, p is defined
by p(t ) = ( cos 2πt ,sin 2πt ) . Let β 1 be the lift of β1 starting at 0; this is well defined since
p(0) = z . Since β1 starts at z and ends at − z , the lift β 1 must end at n + 1 2 , for some n ∈.
We may assume that n ≥ 0 (the case n ≤ 0 is symmetric).
Consider the path λ = β 1 + n + 1 2 in  defined by λ(t ) = β 1 (t ) + n + 1 2 . Since β 1 starts at
0, the path λ starts at n + 1 2 , and since β 1 ends at n + 1 2 , λ ends at 2n + 1. The following
computation shows that λ is the lift of β 2 starting at n + 1 2 :

( ) ( ) ( ( ) (
p β 1 (t ) + n + 1 2 = p β 1 (t ) + 1 2 = cos2π β 1 (t ) + 1 2 ,sin2π β 1 (t ) + 1 2 ))
( )
= cos2πβ 1 (t )cos π − sin2πβ 1 (t )sin π ,sin2πβ 1 (t )cos π + cos2πβ 1 (t )sin π

= ( − cos2πβ (t ), − sin2πβ (t )) = − ( cos2πβ (t ),sin2πβ (t )) = −β (t ) = β (t ).

1 1 1 1 1 2

K12146_C016.indd 422 2/12/15 12:36 PM

 16.5 The Borsuk–Ulam Theorem   ◾    423  

It follows that β 1λ is the lift of the loop β = β1β 2 starting at 0 (and ending at 2n + 1 ).
Since 2n + 1 > 0, the path β 1λ is not a loop. By Corollary 8 in 15.3, β = β1β 2 is not in
p∗ (π1 ( ,0)). Since p∗ (π1 ( ,0)) is the trivial group, this means that β is not contractible.
We have a contradiction.
If z ≠ (1, 0) then compose each of the paths β, β1 , and β 2 with a rotation of S1 that sends
z to (0,1), and repeat the above argument for the resulting paths.

The next three results are easy consequences of Theorem 1. We call them variants of
Theorem 1, because the four statements are equivalent (Exercise 3).

Theorem 2 (Borsuk–Ulam variant 2). Let f : S n → n be such that f ( x ) = − f (− x ) for

every x ∈S n . Then there exists y ∈S n such that f ( y ) = (0, 0, … , 0).

Proof. Restrict f on the closed upper hemisphere of S n and apply Theorem 1.

Theorem 3 (Borsuk–Ulam variant 3). There is no continuous antipode-preserving map

S n → S n−1 , n ≥ 1.

Proof. Suppose there is such an antipode-preserving mapping f : S n → S n−1 . This means that
for every x ∈S n , f ( x ) = − f (− x ) . Now, if in : S n−1 → n is the inclusion, then in  f : S n → n
is antipode preserving and for every x ∈S n , in  f ( x ) = f ( x ) = 1 , which violates the
conclusion of Theorem 2.

Theorem 4. (Borsuk–Ulam variant 4) If f : S n → n is continuous, then there exists y ∈S n

such that f ( y ) = f (− y ).

–y y –y y –y = y

Illustration 16.26 The Borsuk–Ulam Theorem.

Proof. Suppose there is no y ∈S n such that f ( y ) = f (− y ). Then the mapping h : S n → S n−1,

defined by h( y ) = ff (( yy ))−− ff (( −− yy )) , is well-defined, continuous, and antipode preserving, contra-
dicting Theorem 3.

A consequence of Theorem 4 is that S n is not embeddable into n .

Even though we have deduced Theorems 2–4 as stated, it is fair to emphasize that,
since we have proven Theorem 1 only in the special case when n = 2, the same is true for
the other three versions. Proving the general result is not very difficult (see, for example,
[49]). However, since the theory of covering spaces cannot be generalized in a straightfor-
ward manner for higher homotopy groups (see Section 14.5), the argument in the proof of
Theorem 1 presented here for n = 2 does not allow a simple modification for n > 2.

K12146_C016.indd 423 2/12/15 12:36 PM

 424   ◾    Covering Spaces, Part 2

The Brouwer fixed point theorem, proven in Theorem 2 in 11.2, is an easy consequence
of Theorem 1.

Corollary 5 (Brouwer fixed point theorem). Every continuous mapping D n → D n has a

fixed point.

Proof (n = 2). Suppose otherwise. As in the first proof of this theorem (Theorem 2 in 11.2),
this hypothesis implies the existence of a mapping f : D 2 → S1 that is the identity over ∂D 2 . If
in : S1 →  2 is the inclusion, then g = in  f : D 2 →  2 clearly satisfies g ( x ) = − g (− x ) for every
x ∈S1 . It is also evident that there is no y ∈D 2 such that g ( y ) = 0, contradicting Theorem 1.
The general case (any n) is deducible mutatis mutandis from the full strength of
Theorem 1.

The following corollary of the Borsuk–Ulam theorem is stated in general terms. We will
prove it only for n = 2, because we have proven only the case n = 2 of the Borsuk–Ulam
theorem, and since this is the highest dimension that allows a visualization of the argu-
ment. However, deducing Corollary 6 in its full strength from Theorem 4 requires only a
modification of the argument we provide (see, for example, [49]).

Corollary 6. Given n objects in n with well-defined, positive hypervolumes, there exists a

hyperplane in n that cuts all of them into halves (hypervolume-wise).

Proof (n = 2). We deal with two planar objects with well-defined positive areas. We do not
assume that the regions are bounded. Our goal is to show that there is a line halving these
two objects.
First, we associate a line la ,b ,c in the xy-plane to every point (a, b, c ) ∈S 2 \{(0,0,1),(0,0, −1)}:
la ,b ,c is defined by ax + by = c. Since the points in (a, b, c ) ∈S 2 \{(0,0,1),(0,0, −1)} satisfy
a 2 + b 2 ≠ 0 , the vector v = (a, b) is not the zero-vector. The line la ,b ,c is perpendicular to that
vector, hence to the line x = at , y = bt . It intersects this line a2c+b2 units away from the origin
(in the direction of v if c > 0, in the opposite direction to v if c < 0; see Illustration 16.27).
Observe that every line in the plane is la ,b ,c for some (a, b, c ) ∈S 2 \{(0,0,1),(0,0, −1)} .

y7 y6

y8 y5

y9 y4 l4
l5
y
l1 = l7 l6
l2 = l8 l = l y3
3 9

y2
x y1

Illustration 16.27 We show 9 points on a meridian circle. The line associated to the point yi is
denoted in the figure by li . The closer a point yi is to a pole, the further away from the origin is the
associated line. In the illustration we have chosen y1 and y7 , y 2 and y8, y3 and y9 to be antipodal
in pairs. The lines associated to antipodal points coincide.

K12146_C016.indd 424 2/12/15 12:36 PM

 16.5 The Borsuk–Ulam Theorem   ◾    425  

For every line of type la ,b ,c denote the half-plane ax + by ≥ c by l a+,b ,c .

Denote the two planar objects in  2 (as in the statement of the corollary) by R1
and R2 , and define a mapping f : S 2 \ {(0,0,1),(0,0, −1)} →  2 as follows: f (a, b, c ) =
( )
Area( R1 ∩ l a+,b ,c ), Area( R2 ∩ l a+,b ,c ) ; see Illustration 16.28. It is obvious that f is continuous:
small changes in (a,b,c) result in small changes in la ,b ,c and l a+,b ,c , which in turn result in
(
small changes in Area( R1 ∩ l a+,b ,c ), Area( R2 ∩ l a+,b ,c ) . ) c
Notice that as (a,b,c) approaches (0,0,1), the distance a2 +b2 between la ,b ,c and the
origin tends to ∞. This means that the distance between the half plane l a+,b ,c and the
( )
origin tends to infinity too, and so Area( R1 ∩ l a+,b ,c ), Area( R2 ∩ l a+,b ,c ) tends to (0,0). It
follows that defining f (0, 0,1) = (0, 0) continuously extends f over S 2 \{(0,0, −1)} (we have
c
used Proposition 3 in 2.2). Similarly, as (a,b,c) approach (0, 0, − 1), a2 +b2 tends to −∞,
and the distance from the origin to the half-plane  2 \ l a+,b ,c tends to ∞. Consequently,
( )
Area( R1 ∩ l a+,b ,c ), Area( R2 ∩ l a+,b ,c ) tends to ( Area( R1 ), Area( R2 )), and extending f by
f (0,0, −1) = ( Area( R1 ), Area( R2 )) gives a continuous mapping f : S 2 →  2.

y7 y6

y8 y5
+
y9 y4 l4 l5+
+
l7 l +
1
+ + y3
l8 l2+ l9 l3+
y2
+
l6
y1

Illustration 16.28 The half plane associated to the point yi is denoted in the figure by li+ . The
half planes associated to a pair of antipodal points are complementary, except for the common
boundary line.

By the Borsuk–Ulam theorem (Theorem 4), there exists y = (a0 , b0 , c0 ) ∈S 2 such that
f ( y ) = f (− y ). Since ( Area( R1 ), Area( R2 )) ≠ (0,0) , this point y is not in {(0, 0,1), (0, 0, − 1)}.
( ( ) ( )) ( ( ) ( ))
Hence, Area R1 ∩ l a+0 ,b0 ,c0 , Area R2 ∩ la+0 ,b0 ,c0 = Area R1 ∩ l−+a0 ,− b0 ,− c0 , Area R2 ∩ l−+a0 ,− b0 ,− c0
for some (a0 , b0 , c0 ) ∈S 2 \ {(0,0,1),(0,0, −1)}. Finally, notice that l a+,b ,c and l +− a ,− b ,− c are the two
half planes determined by the line la ,b ,c , and so the line la0 ,b0 ,c0 halves both Area( R1 ) and
Area( R2 ).

More formal statement and justification of this corollary (in terms of Borel measure
in place of “hypervolume”) can be found in the book by Matoušek [49], where there is a
wealth of other interesting results related to the Borsuk–Ulam theorem.
Corollary 6 is sometimes called the Ham Sandwich Theorem, because of the fol-
lowing pretty instance when n = 3 : a sandwich made of bread, cheese, and ham can
be cut once so that the bread, the cheese, and the ham are all halved volume-wise
(Illustration 16.29).

K12146_C016.indd 425 2/12/15 12:36 PM

 426   ◾    Covering Spaces, Part 2

Illustration 16.29 Cutting Illustration 16.30 The objects that we halve

a sandwich fairly. could be distant.

Corollary 7. If two of A, B, and C are closed subsets of S 2 and if S 2 = A ∪ B ∪ C , then at

least one of these three sets contains a pair of antipodal points.

Proof. We may assume that A and B are closed in S 2 . Define a mapping f : S 2 →  2

by f ( x ) = (d( x , A) − d(− x , A), d( x , B) − d(− x , B)), where d(x, D) is the distance
from a point x ∈S 2 to the set D ⊂ S 2 . Since f is a continuous mapping, Theorem 4
implies that there exists an element y ∈S 2 such that f ( y ) = f (− y ). This means that
(d( y , A) − d(− y , A), d( y , B) − d(− y , B)) = (d(− y , A) − d( y , A), d(− y , B) − d( y , B)), which easily
yields d( y , A) = d(− y , A) and d( y , B) = d(− y , B). Hence, (using the assumption that A is
closed) y ∈ A if and only if − y ∈ A; by symmetry we have that y ∈B if and only if − y ∈B.
It follows that y ∈S 2 \ ( A ∪ B) if and only if − y ∈S 2 \( A ∪ B), and so both y and − y are in
at least one of A, B, and C.

Exercises
1. Is it true that for every continuous f : S1 × S1 →  2 there exists ( x , y ) ∈S1 × S1 such
that f ( x , y ) = f (− x , − y )?
2. Show that for every continuous f : D 3 →  2 there are uncountably many x ∈D 3 such
that f ( x ) = f (− x ).

3. Deduce Theorem 1 from every Theorem n, n = 2, 3, 4.

4. Give a direct proof of Theorem 3 for n = 2. (Hint: Suppose f : S 2 → S1 is an antipode
preserving continuous mapping, and let ( , p) be the universal cover of S1 . Let f
be a lift of f with respect to ( , p) . Find a loop α: I → S 2 → S1 such that f α is not a
loop in , thus obtaining a contradiction.)
5. Prove that there does not exist a continuous one-to-one mapping  2 →  .

K12146_C016.indd 426 2/12/15 12:36 PM

 16.5 The Borsuk–Ulam Theorem   ◾    427  

6. Show that for every two objects in  3 with well-defined volumes there exists a verti-
cal plane simultaneously halving the volumes of these two objects.
7. Denote the area of a planar object S by A(S), and consider the unit disk D 2 . Given
any r ∈ + , find a planar object B of positive area, such that if a line l divides D 2
into C1 and C2 satisfying A(C1 ) = rA(C2 ), then B does not intersect l. (This exercise
shows that the ratio 1:1 in the statement of Corollary 6, n = 2, cannot be replaced by
another number.)
8. Let X be a subspace of  3 having precisely one point on each ray ra ,b ,c defined
by x = at , y = bt , z = ct , t ∈(0, ∞), a 2 + b 2 + c 2 = 1. Suppose in addition that
f : (a, b, c )  X ∩ ra ,b ,c defines a homeomorphism S 2 → X . Show that there are two
antipodal points in X equidistant from the origin.
9. Show that if A and B are two closed subsets of S 2 such that A ∪ B = S 2 then at least
one of them contains a pair of antipodal points. Show that the analogous statement
for 4 closed subsets of S 2 fails. Show that Corollary 7 fails if we do not assume that
two of the sets A, B, and C are closed.

K12146_C016.indd 427 2/12/15 12:36 PM

This page intentionally left blank
 Chapter 17

Applications in Group Theory

S o far we have used group theory as a tool to investigate various properties of topological
spaces. In this chapter we will reverse the roles, and use homotopy theory to discover
and deduce pure group-theoretical results.

17.1 Cayley Graphs and Covering Spaces

A brief historical note: In 1878 Arthur Cayley (1821–1895) introduced (what were later called)
“Cayley graphs” through an example. In his paper he sketched the (Cayley) graph of the group
A4, generated by a pair of elements different from the ones we use in Example 2.

Let G be a group and let A be a set of generators of G. We construct a graph as follows: The
vertices are the elements of G. For every vertex g ∈G , and for every a ∈ A, we introduce an
oriented edge starting at g and ending at ga. We label that edge by a, and call it an a-edge.
The graph constructed in such a way is called the Cayley graph of the group G on the gen-
erating set A. It will be denoted by Γ(G ,A ), or by Γ G when A is clear, or by Γ when both G
and A are understood from the context.
We now show a few pictures of Cayley graphs.

Example 1: The Cayley Graph of a Cyclic Group

Consider the cyclic group G = 〈 x ; x 4 〉 of order 4, generated by x. We depict Γ (G ,{ x })
in Illustration 17.1.

429

K12146_C017.indd 429 2/12/15 12:38 PM

 430   ◾    Applications in Group Theory

1 x x

x x

x3 x x2

Illustration 17.1 ☐

In order to avoid clutter, we will sometimes omit labeling the vertices of Cayley
graphs.
Recall that the symmetric group Sn on the set {1,2,…,n} is the group of all permuta-
tions (bijections) of the elements of this set, where the operation is composition. The
1 2 3 … n
permutation sending k to ik, k = 1,2,…, n, is often denoted by  .
 i1 i2 i3 … in 
A k-cycle in Sn is a permutation of the type j1  j2  j3 …  jk  j1 , for some distinct
j1 , j2 , j3 … , jk ∈{1,2, … , n}, fixing the other elements of {1,2, … , n} . Such a k-cycle is usually
denoted by ( j1 j2 j3 … jk ). For example, the 3-cycle (1 2 3) in S4 is the permutation 1  2,
2  3, 3  1, 4  4. The group Sn is generated by the set {(i j ) : i , j ∈{1,2, … , n}, i ≠ j} of all
2-cycles, also called transpositions. A permutation in Sn is odd if it is a composition of an
odd number of 2-cycles; otherwise, it is even. Even permutations in Sn make a subgroup of
Sn called the alternating group, commonly denoted by An.

Example 2: The Cayley Graph of an Alternating Group

In this example we will depict a Cayley graph of the alternating group A 4 .

The group A 4 is generated by x = (1 2 3) and the product (composition) y = (1 2)(3 4)

of two transposition. We point out again that in case of compositions of permutations
the left mapping acts first. This is not of consequence for the product defining y (since
the transpositions commute), but it matters in general.
The Cayley graph Γ of A 4 on the generating set {x , y} is shown in Illustration 17.2.

As we see, the vertices and the edges of Γ fit nicely along the vertices and the edges
of a truncated tetrahedron. That this is not merely an aesthetic observation will be
clearer later (Exercise 3 in 17.2).

K12146_C017.indd 430 2/12/15 12:38 PM

 17.1 Cayley Graphs and Covering Spaces   ◾    431  

1 2 3 4 1 2 3 4
x=
2 3 1 4 3 1 2 4

1 2 3 4
1 2 3 4

1 2 3 4 1 2 3 4
1 4 2 3 4 2 1 3

1 2 3 4 1 2 3 4
3 4 1 2 4 3 2 1

1 2 3 4
y=
2 1 4 3 1 2 3 4
1 2 3 4
2 4 3 1 4 1 3 2

1 2 3 4 1 2 3 4
1 3 4 2 3 2 4 1

Illustration 17.2 The edges corresponding to the generator x are depicted in full lines, those
corresponding to the generator y are shown in dashed lines. Recall that every oriented
edge labeled z, where z ∈ {x, y}, joins a vertex g ∈ A4 with the vertex gz, where in the composition
g acts first. Thus, for example, the oriented edge at the bottom of the graph joins the vertex
 1 2 3 4   1 2 3 4   1 2 3 4 
 3 2 4 1  with the vertex  1 3 4 2  =  3 2 4 1  x . ☐

Cayley graphs of infinite groups have infinitely many vertices and edges. Here are a couple
of them.

Example 3:  ⊕  and  ∗
The Cayley graph Γ ⊕ of  ⊕  on the generating set {(1,0),(0,1)} is partially shown in
Illustration 17.3, and in Illustration 17.4 we show a part of the Cayley graph Γ ∗ of the
free group  ∗  on the set of generators consisting of two copies of 1 (one for each ).

Illustration 17.3 The edges corresponding to (1,0) ∈ ⊕  are in dashed lines, and the edges
corresponding to (0,1) ∈ ⊕  are in full lines.

K12146_C017.indd 431 2/12/15 12:38 PM

 432   ◾    Applications in Group Theory

Illustration 17.4

We recognize Γ ∗ as the universal cover of the bouquet of two circles. This is not a
coincidence, as we explain in the next few paragraphs. ☐

Let Γ be the Cayley graph of a group G on a generating set A. Each edge of Γ is oriented,
and it is labeled by an element in A. Let R be the bouquet of A-many circles, labeled
bijectively by the elements in A, and with the only vertex denoted by v. Then the labels of
the edges define a continuous mapping l : Γ → R (sending the interiors of the edges in Γ
homeomorphically onto the interiors of the loops in R, and sending the vertices of Γ to v).
It is evident that l is a covering map, hence (Γ , l ) is a covering space of R. We will call l the
labeling map.
Each g ∈G acts as a homeomorphism of Γ as follows: if e is an oriented edge of Γ labeled
by a ∈ A and staring at g 1 ∈G (and ending at g 1a ∈G ), then g(e) is the edge also labeled by
a, and starting at gg 1 and ending at gg 1a; Illustration 17.5.

g1a gg1a
Γ Γ
g
a a

g1 gg1

Illustration 17.5

It is clear that the group G, considered as a group of homeomorphisms of Γ with the

action just described, acts properly discontinuously on Γ. Moreover, since each g ∈G
(regarded as a homeomorphism of Γ) maps every a-edge onto an a-edge, it follows
that l  g = l , and so each g becomes a covering transformation of (Γ , l ). Finally, observe
that for every g 1 , g 2 ∈G (considered as vertices in Γ) there is an element g ∈G (con-
sidered as a covering transformation) such that gg 1 = g 2 : simply take g = g 2 g 1−1. This
implies that every covering transformation of (Γ , l ) agrees at a vertex with one covering

K12146_C017.indd 432 2/12/15 12:39 PM

 17.1 Cayley Graphs and Covering Spaces   ◾    433  

transformation in G. It follows from Proposition 2(a) in 16.2 that G is isomorphic to the

group A(Γ , l ) of covering transformations of (Γ , l ) . (As before, the “A” in A(Γ , l ) stands
for the first letter of the word automorphism, and it should not be confused with the
generating set A.)
By Proposition 1 in 16.3, ( Γ ,l ) is a regular cover of R. Proposition 3(b) in 16.3 then tells
us that the covers (Γ , l ) and (Γ , q), where q is the quotient mapping Γ → Γ G , are the same
covering spaces (up to a homomorphism that makes the diagram in Proposition 3, 16.3
com­mutative). By Corollary 5 in 16.2, the group G is isomorphic to the quotient group
π1 ( R )
l∗ ( π1 ( Γ )).
We have proven the following.

Proposition 1. Let Γ be the Cayley graph of a group G on a generating set B, and let R be the
bouquet of B circles labeled by the elements of B. Then (Γ , l ) , where l is the labeling map,
is a regular covering of R, and G ≅ π1 ( R ) l∗ ( π1 ( Γ )) ≅ A( Γ , l ) .

In other words, the Cayley graph of a group G generated by B is a covering of the bouquet
of B circles corresponding to a normal subgroup N of the group F freely generated by B,
such that G ≅ F N . The converse is also true:

Proposition 2. If Γ is a graph and if (Γ , p) is a regular cover of the bouquet R of circles, then

it is a Cayley graph of the group π1 ( R ) p∗ ( π1 ( Γ )).

Proof. (Exercise 3)

The assumption in Proposition 2 that Γ is a graph will be made redundant in Section 17.3
where we will show that coverings of graphs must be graphs.
The following is an immediate consequence of Proposition 1 here, and Exercise 12 in
Section 12.1.

Corollary 3. Let Γ, G, B, R, and (Γ , l ) be as in Proposition 1. Then 〈 B ; l∗ (π1 (Γ ))〉 is a presen-

tation of the group G.

The link from groups to Cayley graphs and covering spaces allows conversion of group-
theoretical statements and proofs into corresponding claims within the setup of the theory
of covering spaces. This is one route of visualizing group theory, and in some cases it gives
a better view on the intricacies and interconnectedness of various theorems.
As an illustration of the wide scope of the theory of covering spaces, we prove the following
classical theorem. The proof we give is somewhat longer than the standard group-theoretical
proof, partly because we did not aim the theory at such statements, partly because we give the
details.

Theorem 4 (Cauchy). If G is a group of order n and if p is a prime dividing n, then there is

an element in G of order p.

K12146_C017.indd 433 2/12/15 12:39 PM

 434   ◾    Applications in Group Theory

Proof. Let G be a group of order pm, where p is a prime and m ∈+. Let A be a minimal (under
inclusion) generating set for the group G. The case when G is a cyclic group is easy, and so
we may assume that A > 2 . Consider the Cayley graph Γ of G on the generating set A. By
Proposition 1, (Γ , l ), where l is the labeling map, is a covering space of the bouquet R with A
circles (labeled by the elements of A). Since A is minimal generating set, deleting from Γ all
edges labeled by a fixed element a ∈ A, disconnects Γ.
If the order of a is a multiple of p, then we are finished. Suppose otherwise, and denote
the graph we get by deleting all a-edges from Γ by Δ. The restriction of l on every com-
ponent Δc of Δ is a covering mapping of the rose R without the a-labelled circle. Since
Γ is regular, so is Δc, and each two components of Δ are homeomorphic (via a covering
transformation). There are two cases.
Case 1. The number of vertices of any (hence every) component Δc of Δ is divisible by p. Since
each Δc is a Cayley graph of a subgroup L of G, and since | L | is the number of vertices of Δc, it
follows that p divides L. Since K < G we may use induction on the number of elements of the
underlying group to conclude that there is an element b in L ⊂ G such that b p = 1.
Case 2. The prime p does not divide the number of vertices of the components of Δ; in
this case, since the number of vertices in Δ is the same as the number of elements in G, the
number p divides the number s of components in Δ.
The subgroup H of G generated by A \ {a} acts properly discontinuously on Γ as a group of
covering transformations (Proposition 1). Then Γ, together with the quotient map q : Γ → Γ H ,
is a regular covering space of Γ H . The action of H on Γ is such that the orbit of an edge consists
only of edges with the same label (see Illustration 17.5). Consequently, labeling an edge e of Γ H
with the label of q −1 (e ) is well defined. We get a graph Γ H with edges labeled by A, where each
edge labeled by A \ {a} is a loop. The labeling map from Γ H onto the bouquet R is a covering
map. By Exercise 4 (or, if shortcuts tempt us, by symmetry), Γ H with the labeling map is a regu-
lar cover of R. By Proposition 2, Γ H is the Cayley graph of a certain group, which we denote by
K. The cardinality of K is s (the number of vertices in Γ H ). Since p divides s, and since s < G we
can use our (implicit) induction to conclude that there is an element in K of order p. Since the
elements in A \ {a} are all trivial in K, this element must be of type al for some l. So alp = 1 in
K, which means that every path in Γ H labeled by the word alp is a loop α in Γ H . Lift α to a path
in Γ starting at a vertex x1. If the end vertex of this path is x 2 , then x1 and x 2 are in the same
component of Δ (so that q sends this path to a loop in Γ H; see Illustration 17.6).
∆1
x1
x2 ∆lp

∆2

∆3
∆4

Illustration 17.6 The vertices x1 and x 2 must be in the same component ∆1 of Δ. We do not
assume the components ∆ i are different.

K12146_C017.indd 434 2/12/15 12:39 PM

 17.1 Cayley Graphs and Covering Spaces   ◾    435  

∆1 x1 = xm

∆lp

∆2

∆3
∆4

Illustration 17.7 There are no earlier overlapping of vertices along the path, else we would
contradict the assumption that the order of a in K is lp.

Now lift α starting at x 2; if the end point is x3, then all x1 , x 2 , and x 3 are in the same component
of Δ. Iterate. Since we are dealing with finite graph, at some point we must have xi = xi+m−1,
hence x1 = xm, for some smallest m ∈+ (Illustration 17.7). It follows that (alp )m = 1 in G
and that at ≠ 1 for t < l pm. So, alm has order p.

Exercises
1. Sketch the Cayley graphs of the following groups.
(a) 〈a, b, c ; ∅〉

(b) 〈a, b, c ; a 2 , b 2 , c 2 〉

(c) 〈a, b, c ; a 2 , b 2 , c 2 , aba1b−1 , aca−1c −1 , bcb−1c −1 〉

2. Let Γ be the Cayley graph of a group G on a generating set A, and let Γ′ be a disjoint
copy of Γ.
(a) Identify the corresponding pairs of vertices in Γ and Γ′. Do you get a Cayley
graph of a group? If yes, describe the group in terms of G.
(b) Attach two edges joining each v ∈Γ to the corresponding v ′ ∈Γ ′. Do you get a
Cayley graph of a group? If yes, describe that group in terms of G and some other
known groups.
3. Prove Proposition 2.
4. Let X be a bouquet of circles (labeled by the elements of a set A), and let (Y, p)
be a regular covering space of X. Let H be a subgroup of the group of covering
transformations of Y with respect to p, and let q : Y → Y H be the quotient map
(so that (Y, q) is a regular cover of Y H ). Define r : Y H → X by r ([ y]) = p ( y ), where,
as always, for y ∈Y ,[ y] is the equivalence classes of y in Y H . Show that r is well
defined and that (Y H , r ) is a regular covering of X. (You may assume that covers
of graphs are graphs.)

K12146_C017.indd 435 2/12/15 12:39 PM

 436   ◾    Applications in Group Theory

5. Let Γ be the Cayley graph of a group G on a generating set A, so that (Γ, l), where l is
the labeling map, is a regular covering of the bouquet X with A circles. An element
of A(Γ , l ) is an a-sliding transformation if it moves a vertex v to the vertex va. (Note
that v  va fully specifies the images of the other points in Γ; see Theorem 2 in 15.3.)
Show that for every a ∈ A, the subgroup H of A(Γ , l ), generated by all a-sliding trans-
formations is a normal subgroup of A(Γ , p).
6. Find Γ H if Γ is the Cayley graph of the alternating group A4 as shown in Illustration 17.2,
and H is the subgroup of A (Γ , l ) generated by all y-sliding transformations (y as in
Example 2). Show that H is a proper subgroup of A (Γ , l ) and use Exercise 5 to deduce
that A4 is not a simple group.
7. Let X be a space and let (Y, p) be its universal cover. Let R be a subspace of X such that
π1 ( R , v ) generates π1 ( X , v ) and such that R is homeomorphic to a bouquet of circles.
Show that p −1 ( R ) is homeomorphic (as a topological space) to the Cayley graph of
π1 ( X , v ) on the generating set π1 ( R , v ).
8. Let X be a bouquet of two circles. (a) Find a regular cover (Y, p) of X and a regular
cover (Z, q) of Y such that ( Z , p  q) is not a regular cover of X. (b) Find groups G, H
and K such that G  H , H  K , and G ⋪ K.

17.2 Topographs and Presentations

A brief historical note: The origin of the idea of “reducing peaks” in Cayley graphs can be
found in two papers (1936–1937) by John Henry Constantine Whitehead, where he established
the existence of an algorithm deciding if a cyclic word in a free group can be carried to another
via an automorphism of the free group.

In this section we describe a method for obtaining efficient group presentations.

Let Γ be a graph. We will consistently denote its set of vertices by V (Γ). A topograph is
a graph Γ together with a mapping t : V (Γ ) →  , which we call topography of Γ. For every
v ∈V (Γ ), the image t(v) will be called the altitude of v.
A peak in a topograph Γ is a subgraph of Γ with two edges e1 and e2, configured as in one
of Illustrations 17.8, 17.9, and 17.10.
v e2 w v u e1 v
e1 e1 e2 e2

u u w w
      
Illustration 17.8 t (u) < t (v ) = t (w ). Illustration 17.9 t (u) < t (v ) > t (w ). Illustration 17.10 t (u) = t (v ) > t (w ).

We do not assume that the vertices u, v, and w in these illustrations are all distinct.
If the edges e1−1 (considered as the inverse of the path from u to v along e1) and e2

K12146_C017.indd 436 2/12/15 12:39 PM

 17.2 Topographs and Presentations   ◾    437  

coincide in Illustration 17.9 we will say that the peak is trivial. Trivial peaks will not
interest us.
A peak from u to w in a topograph Γ (with the notation as in Illustrations 17.8–17.10) is
reducible if there is a path in Γ from u to w such that t (z ) < t (v ) for every vertex z in that
path other than u and w (Illustrations 17.11–17.13). Such a path is then called a reduction
of the corresponding peak.
v

e1 e2

v e2 w u e1 v
u w
e1 eḱ eḱ e2
e1́ e1́
e2́

u e2́ w
e1́
e2́ Illustration 17.12 In case u = w, eḱ
k is 0, and the reducing path is
Illustration 17.11 the constant at u. Illustration 17.13     

If every peak in Γ is reducible then we say that Γ is a reducible topograph.

If Γ is the Cayley graph of a group G on a generating set A, and if it is equipped with a
topography, then we say it is a (Cayley) topograph of G on A. Since in this case V (Γ ) = G
the topography is a mapping G → . We will continue to denote the labeling mapping with
respect to which Γ is a covering space of the associated bouquet R of labeled circles by l. For
every path e1e2 … en in Γ the image l(e1 )l(e2 )  l(en ) is an element in the free group π1 ( R ). In
particular, if e1e2 is a peak and if e1′e2′ … ek′ is its reduction, then l(e1′ )l(e2′ ) … l(ek′ )l(e2 )−1 l(e1 )−1
is called a reducing element in π1 ( R ).
Recall that a subgraph Λ of a graph Γ is spanned by a set S of vertices of Γ if it consists
of S and of all edges in Γ joining the vertices in S.

Proposition 1. Let Γ be a reducible Cayley topograph of a group G on a generating set A, let

the topography mapping t attain its minimum, and let Λ be the subgraph of Γ spanned by the
set of all vertices of minimal altitude. Assume that Λ is connected and let R be the set of all
reducing elements of the Cayley topograph Γ. Then 〈 A ; l∗ (π1 ( Λ )), R 〉 is a presentation of G.

The subgraph Λ in Proposition 1 will be called the floor of Γ. The group π1(Λ) in the
statement of Proposition 1 may be replaced by any generating set of that group without
altering the meaning of the statement. Before we prove Proposition 1 we illustrate its usage
with the following basic example.

Example 1: A Dihedral Group

The group D8 of all symmetries of the square {( x , y ) ∈ 2 : x + y ≤ 1} is generated by
the rotation ρ through π2 (counterclockwise) and centered at (0,0), and by the reflec-
tion r with respect to x = 0. The Cayley graph of D8 on these two generators is shown
in Illustration 17.14. We make it a topograph in the indicated natural way, by simply

K12146_C017.indd 437 2/12/15 12:39 PM

 438   ◾    Applications in Group Theory

interpreting Illustration 17.14 as roughly depicting the edges of a cube with one face at
altitude 0, and the other at altitude 1.
1 ρ 1

ρ ρ
r
1 ρ 1 r

r r
r r
ρ
ρ
r 0 r 0
ρ

0 ρ 0

Illustration 17.14

The relevant reducing elements come from the peaks shown in Illustrations 17.15 and
17.16 and from their reducing paths in Illustrations 17.16 and 17.17.
1 e2 ρ 1
v

e1

Illustration 17.15

1 e2 ρ 1
1 v
v

r r r r
e2
e1 e1

0 0 ρ 0

Illustration 17.16      Illustration 17.17

Hence, R = {ρr ρ−1r −1 , r 2 }. It is evident from Illustration 17.14 that Λ is the bottom
rectangular graph. Hence, π1 ( Λ ) is a cyclic group, and l∗ (π1 ( Λ )) is generated by ρ4. It
follows from Proposition 1 that 〈r , ρ ; ρr ρ−1r −1 , r 2 , ρ4 〉 is a presentation of D8. ☐

Proof of Proposition 1. By Proposition 1 in 17.1 and by Exercise 12 in 12.1, 〈 A ; l∗ (π1 (Γ , x ))〉 is a

presentation of G, where, for convenience, we choose x ∈V ( Λ ). The proof will be complete if we
establish that every W ∈l∗ (π1 (Γ , x )) is a consequence of l∗ (π1 ( Λ )) ∪ R . The lift δ of W starting
at x is a loop. If e1e2 is a peak in δ, then we replace it with its reduction e1′e2′ … ek′ . This corre-
sponds to changing W from Ul(e1 )l(e2 )V to W1 = Ul(e1′ )l(e2′ ) … l(ek′ )V . Since this change can
be realized by inserting l(e1′ )l(e2′ ) … l(ek′ )l(e2 )−1 l(e1 )−1 immediately following the subword U in
W, we have that W = W1 is a consequence of the relation l(e1′ )l(e2′ ) … l(ek′ )l(e2 )−1 l(e1 )−1 = 1, and

K12146_C017.indd 438 2/12/15 12:39 PM

 17.2 Topographs and Presentations   ◾    439  

so a consequence of R . Since we have chosen x to be in V(Λ), iterating this procedure changes

δ to a path δ′ in Λ, and changes W to W ′ = l(δ ′ ). By induction, W = W ′ is a consequence of R .
Since δ′ is in Λ, l(δ ′ ) is in l∗ (π1 ( Λ )), hence, W ′ = 1 is a consequence of l∗ (π1 ( Λ )). We proved that
W = 1 is a consequence of l∗ (π1 ( Λ )) ∪ R .

Example 2: The Symmetric Group

We noticed earlier that the symmetric group (also known as the group of permuta-
tions) Sn on the set {1,2, … , n} is generated by the set T = {(i j ) : i , j ∈{1,2,… , n}, i ≠ j}
of all transpositions. An inversion in an element α ∈Sn is an occurrence of a pair
(α(i1 ), α(i2 )) such that i1 < i2 and α(i1 ) > α(i2 ) . We define a topography t on the Cayley
graph Γ of Sn on the generating set T as follows: for every α ∈Sn , t (α ) is the num-
ber of inversions in α. For example, if α is the 3-cycle (2 1 3) ∈S3 , then t (α ) = 1. In
Illustration 17.18 we show the topograph Γ when n = 3.
3
(2 3
)
)
(1 2

2
2)
(1

(1 3)
)

(2
(1 3

3)

1 (1 3) 1
(2

2)
3)

(1

id
0

Illustration 17.18 The topograph Γ of S3: only the bottom vertex is labeled (by the identity
permutation); the numbers by the vertices are their altitudes. We indicate the transposi-
tions corresponding to the edges. Notice that since each transposition is of order 2, there
are pairs of edges connecting every pair of vertices.

Let a, b, c, and d be distinct elements of {1,2, … , n}. There are three types of peaks
in Γ: type 1 peaks correspond to a word of type (a b)(a b), type 2 peaks corre-
spond to (a b) (c d), and type 3 peaks correspond to (a b)(b c). The first type of
peak allows the trivial (constant) reduction, and the associated reducing element
is (a b)2 (accounting again for the fact that the transpositions are of order 2).
Since (a b) (c d) = (c d) (a b), the type 2 peaks allow a reduction with the asso-
ciated reducing element (a b) (c d) (a b) (c d). The third kind of peak requires
more care, but the work is rather straightforward, and we leave the details to the
reader (Exercise 1). The reducing elements stemming from this case are of type
{α (a b) α (α(a) α(b)) : α , (a, b) ∈T }. This set contains the reducing elements we got

K12146_C017.indd 439 2/12/15 12:39 PM

 440   ◾    Applications in Group Theory

in the second case. Since the graph spanned by the set of vertices of minimal altitude
contains only one vertex, it follows from Proposition 1 that the following is a pre­
sentation of Sn : 〈T ; {α 2 = 1 : α ∈ T } ∪ {α (a b) α (α(a) α(b)) : α, (a, b) ∈ T }〉 . ☐

Example 3: A Free Group

The general linear group of all invertible n × n matrices with integer entries is com-
monly denoted by GL(n, ). Consider the subgroup G of GL(2, ) generated by the
  1 2   1 0  
set   ,   , and its Cayley graph Γ on that set. We make Γ a
  0 1   2 1  
 a a12 
2

topograph by defining t  
11
 =
  a21 a22  i , j =1
 
∑
aij2 . A case-by-case careful (but

tedious) inspection shows that there are no nontrivial peaks in Γ. For example,
 
if the edges e1 and e2 in a path e1e2 are both labeled by  1 2 , if e1 starts at
 a b   0 1 
u=  ∈ , and if it ends at v with
G t ( v ) > t (u ), then the end vertex w of e2 is
 c d 
such that t (w ) > t (v ), and hence the edges e1 and e2 do not make a peak. Justifying this
 a b   a b  1 2 
claim is equivalent to proving that if t   <t    ,
 c d   c d  0 1 
2
  a b   1 2   
a b  1 2 
then t     < t      , which in turn
  c d   0 1    c d   0 1  
reduces to a straightforward implication involving two inequalities. Since there are no
non-trivial peaks, the topograph is plainly reducible and the set of reducing elements
is empty.

The identity 2 × 2 matrix is the only matrix in G at minimal altitude 2, for if there
is another matrix M in G of that altitude, then the path in Γ starting at the identity
matrix and ending at the matrix M must have non-trivial peaks. Hence, the floor of Γ
consists of one vertex only, and so it is simply connected.
  1 2   1 0  
It follows from Proposition 1 that G is freely generated by   , . ☐
  0 1   2 1  
 1 0   1 0 ,
The group GL(2, ) itself is generated by the matrices x =  , y =  
 1 −1   0 −1 
 0 1  2 2 2 4 2 3
and z =   . It can be shown that 〈 x , y , z ; x , y , z , (zy ) , (zy ) (zx ) 〉 is a presen-
 1 0 
tation of GL(2, ) ([15], page 23). However, the Cayley topograph over the generating set
{x , y , z } and with topography as in Example 3 is not reducible (Exercise 5).

K12146_C017.indd 440 2/12/15 12:39 PM

 17.2 Topographs and Presentations   ◾    441  

The group GL(2, ) has bearing on the spaces we have encountered so far, as we
will now indicate. The mapping class group MCG(M) of a manifold M is the group
of all isotopy classes of homeomorphisms M → M . Consider, for example, the torus T 2 . If
f is a self-homeomorphism of T 2, then it induces an isomorphism f∗ : π1 (T 2 ) → π1 (T 2 ),
and ϕ : f  f∗ is a homomorphism from the group of all self-homeomorphisms of T2
into the group of all automorphisms (self-isomorphisms) of π1 (T 2 ). It is true that φ
is onto and that its kernel is the class of all homeomorphisms of T 2 isotopic to the
identity (see, for example, [63], pages 26–28). It follows that MCG(T 2 ) is isomor-
phic to the group of automorphisms of π1 (T 2 ) . On the other hand, we know that
π1 (T 2 ) = 〈 x1 , x 2 ; XYX −1Y −1 〉 ≅  ⊕  , and basic group theory tells us that the group
of all automorphisms of  ⊕  is GL(2, ), the group of all invertible 2 × 2 matrices
with integer entries. So MCG(T 2 ) ≅ GL(2, ), and so we have exhibited above (following
Example 3) a presentation for the mapping class group of the torus.
More generally, MCG(T n ) ≅ GL(n, ), where T n = S 1
S1
×
 ×
×S1 is the n-torus. An
n
explicit finite presentation of GL(n, ) is given in [69]. It is not known if there is a finite set
of generators for GL(n, ) such that the Cayley topograph, with topography as in Example
3, is reducible with a finite set of reducible elements. Such a topograph, besides generating
another finite presentation of MCG(T n ) ≅ GL(n, ), would deepen our understanding of
the structure of this group (see, for example, [25]).
The method outlined in the above examples has been used extensively to discover
and investigate presentations and other related properties of more complicated groups.
One such group is AutFn , the group of all automorphisms of a free group Fn of rank
n. Indeed, this is where the method originated. In the inaugurating articles, J. H. C.
Whitehead ([76],[77)]) showed, using (what we call here) reducible topographs, that
there is an algorithm to check if for two elements in x , y ∈ Fn there is an automorphism
f ∈ AutFn such that f ( x ) = y . Hence, this technique of reducing peaks is sometimes called
the Whitehead’s method.

Exercises
1. Complete the details in Example 2. In particular, find reductions of the peaks of type
(a b)(b c).
2. Consider the group of symmetries of the frieze pattern shown in Illustration 12.2,
Section 12.1. Sketch the Cayley graph over the generating set consisting of a rotation
and a glide reflection. Assign a simple (natural) topography to that graph, and use it
to find a presentation for that group of symmetries.
3. Use the Cayley graph of the alternating group given in Illustration 17.2, Section 17.1
and the topography that is associated with that representation of the Cayley graph
(higher vertices are of higher altitude) to show that the topograph is reducible and to
find a presentation for that group.

K12146_C017.indd 441 2/12/15 12:39 PM

 442   ◾    Applications in Group Theory

4. The dihedral group D2n is the group of symmetries of a regular polygon with n
edges. Generalize the method used in Example 1 to find a presentation for this
group.
 1 0 
5. Show that the group GL(2, Z ) is generated by the matrices x =  ,
 1 −1 
 1 0   0 1 
y=  , and z =   . Show that the Cayley graph of GL(2, ) over
0 − 1
   1 0 
this generating set and with topography as in Example 3 is not reducible.
6.∗ Let Fm be a free group freely generated by X = {x1 , x 2 , … , xm }. The group ESm of extended
symmetries of Fm is a group of automorphisms of Fm generated by the set T of all tran­
spositions( xi x j ) of type x j  xi , xi  x j (i ≠ j , and i , j ∈{1,2,… , m}), x k  x k otherwise,
and by the set IN of all inversions ri of type xi  xi−1, for a fixed i ∈{1,2,… , m},
x k  x k if k ≠ i. Use topographs to show that ESm ≅ 〈T , IN ; R , {ri 2 , ( xi x j ) ri =
rj ( xi x j ) : i , j ∈{1,2, … , m}〉 , where R is a set of relations for the group of symmetries
(generated by the set T) as found in Example 2.
7.∗ Prove by means of topographs and Proposition 1 that subgroups of finitely generated
free groups are free.

17.3 Subgroups of Free Groups

A brief historical note: That subgroups of free groups are free was first proven by Otto
Schreier in 1927.

In this section we briefly examine a few properties of subgroups of free groups.

Proposition 1. If (Y, p) is a cover of a graph then Y is a graph.

Proof. Let X be a graph and let (Y, p) be a covering space of X. We may assume that X is
connected. Subdivide each edge in X into three closed intervals, corresponding to the sub-
intervals [0, 1 4 ], [ 1 4 , 3 4 ], and [ 3 4 ,1] of I = [0,1]. We call these parts subedges of edges, and
we will refer to the subedges corresponding to [ 1 4 , 3 4 ] as middle subedges. Each vertex
v of X, together with the adjacent subedges is a simply connected subspace of X, which
we will denote by Zv . The space Zv can be given a structure of a graph by declaring
the endpoints of the subedges to be vertices. By Proposition 1 in 15.4, each compo-
nent of p −1 ( Zv ) is a covering space of Zv . Since Zv is simply connected, each of these

K12146_C017.indd 442 2/12/15 12:39 PM

 17.3 Subgroups of Free Groups   ◾    443  

components is homeomorphic to Zv , and hence each of them is a graph. Similarly, if em

is a middle subedge, then each components of p −1 (em ) is homeomorphic to I.

b
a c
b
a c
b
a c
b

a
b

Illustration 17.19 To the left we show a graph X (with two vertices and three edges), and its
covering space Y, where the covering maps is determined by the labels. To the right we see
what we do to recognize that Y is also a graph.

It is clear that Y is the space obtained by identifying the end-vertices of the components of
p–1 (em ) with vertices of ∪ Z ν (see Illustration 17.19). Hence, Y is also a graph.
ν∈Vert ( X )

An immediate consequence is the following classical result.

Theorem 2. Every subgroup of a free group is free.

Proof. Let F be a free group and let H be a subgroup of F. Represent F as the fundamental
group of a bouquet R of rank(F) many circles. Since R has a universal cover, there is a cov-
ering space Y of R corresponding to H. By Proposition 1, Y is a graph. By Proposition 3 in
13.2, π1 (Y ) is a free group, and since H ≅ π1 (Y ), H is a free group too.

We will now aim our mini-theory in the direction of exploring intersections of subgroups
of free groups.
Let p1 : Y1 → X and p2 : Y2 → X be continuous mappings. The pullback of these two
mappings is the triple ( Z , q1 , q2 ), where q1 : Z → Y1 and q2 : Z → Y2 are continuous map-
pings such that the following two conditions are satisfied.

(a) The diagram in Illustration 17.20 commutes.

(b) For every space W and continuous mappings r1 : W → Y1 and r2 : W → Y2 such that
the diagram in Illustration 17.21 (without the dashed line) commutes, there is a unique
continuous mapping s : W → Z such that the whole diagram in Illustration 17.21
commutes.

K12146_C017.indd 443 2/12/15 12:39 PM

 444   ◾    Applications in Group Theory

W
r1
s
q1 q1
Z Y1 Z Y1
r2
q2 p1
q2 p1

Y2 X Y2 X
p2 p2
   
Illustration 17.20 Illustration 17.21

The uniqueness condition in part (b) above guarantees the uniqueness of Z up to

homeomorphism.

Lemma 3. The pullback of p1 : Y1 → X and p2 : Y2 → X is ( Z , q1 , q2 ) , where

Z = {( y1 , y 2 ) ∈Y1 × Y2 : p1 ( y1 ) = p2 ( y 2 )} (as a subspace of Y1 × Y2 ), and q1 , q2 , are the rest­
rictions of the projections.

Proof. Part (a) of the definition of pullbacks is plain. If W, r1 and r2 are as in part (b) of
the definition, then the continuous mapping s defined by s(w ) = (r1 (w ), r2 (w )) makes the
diagram in Illustration 17.21 commutative. Its uniqueness is clear from the commutativity
of the diagram.

We are interested in pullbacks of covering spaces of bouquets of circles. For such spaces
the following is true.

Proposition 4. Let (Y1 , p1 ) , (Y2 , p2 ) be covering spaces of a bouquet X of circles. Let

( Z , q1 , q2 ) be the pullback of p1 and p2, and let C be a path component of Z. Then (C , p1  q1 )
is a covering of X. Moreover, if ((Y1 , z1 ), p1 ) corresponds to a subgroup H1 of π1 ( X , x ) , if
((Y2 , z 2 ), p2 ) corresponds to a subgroup H2 of π1 ( X , x ) , and if C is the component of Z
containing ( z1 , z 2 ), then the covering (C , p1  q1 ) of X corresponds to the subgroup H1 ∩ H 2
of π1 ( X , x ).

Proof. By Lemma 3 the pullback Z is {( y1 , y 2 ) ∈Y1 × Y2 : p1 ( y1 ) = p2 ( y 2 )}, and q1 , q2 are the

restrictions of the projections. By Proposition 1, both Y1 and Y2 are graphs. We now show
that Z is a graph. The set of vertices of Z is ( p1  q1 )−1 (v ) , where v is the only vertex of X.
Let a be an oriented edge in X and let ( z1 , z 2 ) be a vertex in Z. The path α winding once
along a has a unique lift α 1 with respect to (Y1 , p1 ) starting at z1; it ends at some w1 ∈Y1 .
Similarly α has a unique lift α 2 with respect to (Y2 , p2 ) starting at z2 which ends at some
w 2 ∈Y2 . It follows that {(α 1 (t ), α 2 (t )) : t ∈ I } is a subset of Z homeomorphic to the edge
we get by attaching the points 0 and 1 of the interval I to the vertices ( z1 , z 2 ) and (w1 , w 2 )
respectively. Since every point of Z is either a vertex or is in some {(α 1 (t ), α 2 (t )) : t ∈ I } cor-
responding to some oriented edge a as in the above procedure, it follows that Z is a graph.

K12146_C017.indd 444 2/12/15 12:40 PM

 17.3 Subgroups of Free Groups   ◾    445  

The preceding argument also establishes that for every vertex ( z1 , z 2 ) in Z and for every
oriented edge a in X, there exists a unique edge starting at that vertex and mapped by
( p1  q1 ) onto a. Consequently, if C is the path component of Z containing ( z1 , z 2 ), then
(C , p1  q1 ) is a covering space of X.
It remains to be shown that (C , p1  q1 ) of X corresponds to the sub-
group H1 ∩ H 2 of π1 ( X , x ). Since q2 is the lift of p1  q1 with respect to
(Y2 , p2 ) such that q2 ( z1 , z 2 ) = z 2 , it follows from Theorem 2 in 15.3 that
( p1  q1 )∗ π1 (C , ( z1 , z 2 )) ⊂ H 2 . By symmetry, ( p2  q2 )∗ π1 (C , (z1 , z 2 )) ⊂ H1. Recalling that
p1  q1 = p2  q2 , we have proved so far that ( p1  q1 )∗ π1 ((C , ( z1 , z 2 )) ⊂ H1 ∩ H 2 . In order to
establish that ( p1  q1 )∗ π1 ((C , ( z1 , z 2 )) = H1 ∩ H 2 it suffices to show that every loop α at v
(the only vertex of X), representing an element in H1 ∩ H 2 , lifts to a loop in C starting
at ( z1 , z 2 ) (Corollary 8, 15.3). There is a unique lift α 1 of α at z1 with respect to (Y1 , p1 ) ,
and since α ∈ H1 , this lift is a loop. Symmetrically, there is a unique lift α 2 of α at z2 with
respect to (Y2 , p2 ) , and α 2 is also a loop. Then β(t ) = (α 1 (t ), α 2 (t )) is a loop at ( z1 , z 2 ) that
is a lift of α with respect to ( Z , p1  q1 ).

Example 1
In Illustration 17.22, we show the pullback Z of coverings of a bouquet X with two circles.

Y2 q2 q1 Y1

p1 p2
b a
X
Illustration 17.22 The covering maps are such that only dashed edges go to dashed edges. We
also introduce arrows to make the covering map unambiguous.

The vertices of Z are p1−1 ( x ) × p2−1 ( x ) , where x is the only vertex of X, and we have an
edge from (u1 , v1 ) to (u2 , v2 ) if there is an edge e1 in Y1 from u1 to u2 and an edge e2
in Y2 from v1 to v2 such that p1 (e1 ) = p2 (e2 ). According to Proposition 4 the covering
Z of X corresponds to the intersections H1 ∩ H 2 of the subgroups of π1 ( X ) associ-
ated with the coverings Y1 and Y2. Observe that the pullback structure allows us to

K12146_C017.indd 445 2/12/15 12:40 PM

 446   ◾    Applications in Group Theory

obtain a set of free generators for H1 ∩ H 2 from the fundamental group of Z (via
Proposition 3 in 13.2). ☐

The case when Y1 and Y2 correspond to finitely generated subgroups of the fundamental
group of the bouquet is interesting and we explore it further.

The core of a graph Y, denoted Core(Y), is the minimal (under inclusion) subgraph Z
of Y such that π1 ( Z , z ) ≅ π1 (Y , z ), z ∈ Z . An example is given in Illustration 17.23.

Illustration 17.23 The core of the graph to the left is the one to the right: to get the latter
we simply trim the subtrees of the former that do not contribute to the computation of the
fundamental group.

It follows directly from Proposition 3 in 13.2 that for every graph Y , the subgraph Core(Y)
is finite if and only if the free group π1 ( X ) is of finite rank. Consequently, if X is a bouquet
with finitely many circles, a covering Y of X corresponds to a finitely generated subgroup of
π1 ( X ) if and only if Core(Y) is finite.

If Z, together with the mappings q1 and q2, is the pullback of the coverings (Y1 , p1 ) and
(Y2 , p2 ) of the bouquet X, then Core(Z) is the pullback of (Core(Y1 ), p1 Core(Y1 ) ) and
(Core(Y2 ), p2 Core(Y2 ) ) (Exercise 5). Hence, the intersection of two finitely generated subgroups
of a free group of finite rank is also finitely generated (Exercise 6)1, and we have an algorithm
for determining a set of free generators of the intersection, and hence, for establishing its rank.
We illustrate it in the following example.

Example 2
We only show the cores of the two coverings of X (Illustration 17.24); the rest of the
coverings is then uniquely determined (by attaching trees where needed). The core
of the pullback is a finite graph, and its fundamental group is free on 3 generators
(Proposition 3 in 13.2).

1
This is not so easy to establish using pure group theoretical concepts only; see [35].

K12146_C017.indd 446 2/12/15 12:40 PM

 17.3 Subgroups of Free Groups   ◾    447  

Z
(u3, v1) b (u3, v2)
a a

(u2, v1) a a (u2, v2)

a a
(u1, v1) (u1, v2)
b u2 b
Y1 a a Y2
q1 q2
a
a
u3
a u1
v1 v2
p1 b
p2
X
b
b a

Illustration 17.24

Denote the subgroups of π1 ( X ) corresponding to the coverings (Y1 , p1 ) and (Y1 , p2 )

by H1 and H2, respectively. Hence ( Z , p1  q1 ) corresponds to H1 ∩ H 2 . Counting the
ranks, we see that rank ( H1 ) = 3, rank ( H 2 ) = 2, and rank ( H1 ∩ H 2 ) = 3. ☐

Observe that in Example 2 we have rank ( H1 ∩ H 2 ) − 1 ≤ (rank ( H1 ) − 1) (rank ( H 2 ) − 1).

That this inequality is true for all non-trivial finitely generated subgroups H1 and H2 of a
free group was conjectured by Hanna Neumann in 1956–1957 [59, 60]. This long-standing
open problem has now been solved (Mineyev, 2012, [54]), and the proof utilizes pullback
diagrams of graphs.

We do not know of any formula expressing explicitly rank ( H1 ∩ H 2 ) in terms of rank (H1)
and rank (H2) .

Exercises
1. Find two covering spaces of a bouquet of circles such that the pullback space is not
path connected.
2. Prove that each path component of the pullback of two regular coverings is a regular
covering.
3. Let F be a free group of finite rank. Prove that there are finitely many finitely gener-
ated subgroups of F of finite index in F.
4. Let F be a free group and let N1 and N 2 be two normal subgroups of F, both of infinite
ranks. Prove (using covering spaces and pullbacks) that N1 ∩ N 2 is either trivial or of
infinite rank.

K12146_C017.indd 447 2/12/15 12:40 PM

 448   ◾    Applications in Group Theory

5. Let X be a graph and let (Y1 , p1 ) and (Y2 , p2 ) be coverings of X. The pullback of
(Core(Y1 ), p1 Core(Y1 ) ) and (Core(Y2 ), p2 Core(Y2 ) ) is defined in the same way as the pull-
back of coverings. Show that this pullback is the same as the core of the pullbacks of
(Y1 , p1 ) and (Y2 , p2 ).
6. Show that the intersection of two finitely generated subgroups of a free group of finite
rank is also finitely generated.

17.4 Two Subgroup Theorems

A brief historical note: Theorem 1 was one of the results published in 1949 in a paper authored
jointly by Graham Higman (1917-2008), Bernhard Neumann (1909–2002), and Hanna Neumann
(1914–1971). Alexander Kurosh (1908–1971) proved his theorem in two papers published in
1933 and 1934.

In this section we use the theory of covering spaces to prove two subgroup-theorems.

Theorem 1 (Higman, Neumann, Neumann). Every countable group G can be embedded

into a group H generated by two elements.

Proof. We start by outlining the plan of the proof: The group G can be realized as the
group of covering transformations of a regular covering space Γ of a countable bouquet
of circles. We will explain how to modify the graph Γ to a graph Γ1 in such a way that Γ1
can be seen to be a regular covering space of a bouquet of just two circles. The group of
covering transformations of Γ1 contains G as a subgroup; on the other hand it is a quotient
group of the fundamental group of the bouquet of two circles, hence it is generated by two
elements.

Let M = + if G is countably infinite, and let M = {1,2,… , n} if G = n + 1.

Let A = { g j : j ∈ M } be the list of all nonidentity elements of G, and let Γ be the Cayley
graph of G on the generating set A. By Proposition 1 in 17.1, the graph Γ, together with the
mapping p defined by the labels of the edges, is a regular covering space of the bouquet R
of A circles labeled by the elements of A, and G is isomorphic to π1 ( R ) p (π (Γ,w )), for any
∗ 1
vertex w in Γ (we may take w to be the identity element of G). By Theorem 4, 16.2, G is also
isomorphic to the group of covering transformations of Γ.
We change Γ to another graph Γ1 as follows (see Illustrations 17.25 and 17.26): First we
replace each vertex u ∈V (Γ ) = G with a graph Γ u , where V (Γu ) = {ui : i ∈ M }, and where for
each ui , ui +1 ∈V ( Γ u ), i ≠ sup M, there is an edge labeled by a new symbol a ∉ A starting
at ui and ending at ui +1. Then we stipulate that for every edge in Γ labeled g j , starting at a
vertex u, and ending at a vertex v, we have an edge in Γ1 labeled by b ∉ A ∪{a}, starting at

K12146_C017.indd 448 2/12/15 12:40 PM

 17.4 Two Subgroup Theorems   ◾    449  

uj and ending at vj. Observe that, since Γ is a deformation retract of Γ1 , these two graphs
have isomorphic fundamental groups. The graph Γ1 can be viewed as a core of a covering of
the bouquet R1 with two circles, labeled a and b, where the covering map is defined by the
labels of the edges in Γ1. In order to extend this to a covering of R1 corresponding to the
same subgroup of π1 ( R1 ) we need to attach trees where needed. The outcome is a covering
(Δ, q) of R1.
In Illustrations 17.25 and 17.26, we illustrate this procedure with two simple examples:
In Illustration 17.25, we choose G to be the cyclic group 〈 x ; x 3 〉; in Illustration 17.26, we
deal with G =  2 ⊕  2 . The theorem applies trivially to G in both cases; we use these groups
only to visualize the idea of the proof. The smallest group that is not generated by ≤ 2
elements is  2 ⊕  2 ⊕  2; it has 8 elements and an illustration of our embedding procedure
using this group would be too large and cluttered.

a a
x2 b b
2
x b b b
x2 x x2 b b
x b
x a b
x x a a b
x b
a b a
a b
b b b
x2 x2

Illustration 17.25 From left to right: We start with the Cayley graph of G on the set A of gen-
erators consisting of all non-identity elements (in this case, A = {x , x 2 }). Then we replace each
vertex v with a linear graph with A -many vertices vi , i ∈M (two vertices in this example); an
edge labeled gj starting at a vertex u and ending at a vertex v in the initial Cayley graph, now
starts at uj and ends at vj . In the third step we re-label all old edges by b, and all edges in the
linear graphs Γ v by a. Lastly, we attach trees where needed to get a covering of the bouquet Δ
with two edges, labeled a and b.

b b b b
a a
b
b b a a
b
b a b
a b a b
b a b
b a b a a
b
a b a b b
a b a
g1 b
b b b a
a a
b b b
g2 g3 g3 b
g2 a a w1
b a b a
w g1 w1 b b b b
b

Illustration 17.26 From Γ to Δ in case of the group G =  2 ⊕  2 . We denote g 1 = (1,0) , g 2 = (0,1)

and g 3 = (1,1).

K12146_C017.indd 449 2/12/15 12:40 PM

 450   ◾    Applications in Group Theory

We claim that G is embeddable in the two-generator group H = π1 ( R1 ) NC (q∗ ( π1 ( ∆ ,w1 ))), where we
recall that NC(B) stands for the normal closure of the subset B in the underlying group.
More precisely, we prove that the mapping f : π1 ( R ) p∗ ( π1 ( Γ ,w )) → π1 ( R1 ) NC (q∗ ( π1 ( ∆ ,w1 ))) defined by
f ( g i ) = ai −1ba − (i −1) is a monomorphism. (For brevity we write g here instead of the more
precise gK for an element of the quotient group G K ).
The group π1 (Γ , w ) is generated by the set L of all (homotopy classes of) loops at w that
involve 2 or 3 edges (specifying the multiplication table of G) and by the conjugates of such
loops. Hence <A; L> is a presentation of π1 ( R ) p∗( π1 )( Γ , w )). Observe that if e is an edge in Γ
labeled by g j , starting at a vertex v and ending at a vertex u, then the path in Δ correspond-
ing to the word a j −1ba − ( j −1) and starting at v1, must end at u1 . It follows that if g i1 g i2 g i3 is the
element of G corresponding to a loop at w of edge-length 3 (i.e., a loop along three edges),
then the word (ai1 ba−i1 )(ai2 ba−i2 )(ai3 ba−i3 ) defines a loop at w1. Hence, by Proposition 1 in
12.1, f extends to a well-defined homomorphism f : π1 ( R ) p∗ ( π1 ( Γ ,w )) → π1 ( R1 ) NC (q∗ ( π1 ( ∆ ,w1 ))).
That f is one-to-one is also evident: since g i ≠ 1 for every i ∈ M , the edge in Γ labeled by g i
and starting at w ends at some other vertex u. By the observation in the middle of the previ-
ous paragraph, the path corresponding to the word a j −1ba − ( j −1) and starting at w1 must end
at u1. Hence it is not a loop, and so f ( g i ) = ai −1ba − (i −1) does not belong to NC(q∗ (π1 ( ∆ , w1 ))) ,
implying that ai −1ba − (i −1) NC(q∗ (π1 ( ∆ , w1 ))) is not the identity in H.

Now we turn our attention to the Kurosh subgroup theorem. To obtain a proof of this
theorem we need a few preliminary lemmas. The proofs of the first two are left as exercises.

Lemma 2. If T1 and T2 are subtrees of a tree T, and if T1 ∩ T2 ≠ ∅, then T1 ∩ T2 is a tree.

A forest in a graph is a disjoint union of trees.

Lemma 3. Every forest in a connected graph is a subgraph of a maximal tree.

Lemma 4. If X is a path connected 2-dimensional CW-complex and if Y is its 1-skeleton,

then π1 (Y ) generates π1 ( X ) .

Proof. This follows from the observation that every loop in a 2-cell is homotopic to a loop
on its boundary.

 , p) is a covering space of X, then

Lemma 5. If X is a 2-dimensional CW-complex and if ( X

X is a 2-dimensional CW-complex.

Sketch of proof.
Denote the 1-skeleton of X by V. Then, p −1 (V ) is a graph (Proposition 1 in 17.3). If in the
process of constructing the CW-complex X we attach a 2-cell D by identifying its boundary
with a loop α in R at x, then the homotopy class of that loop is trivial in π1 ( X , x ) , and so
its lift at any point y is also a loop. Then p −1 (D ) (where we view D as a subspace of X) is the
result of attaching a 2-cell by identifying the boundary of D with the lift α at y.

K12146_C017.indd 450 2/12/15 12:40 PM

 17.4 Two Subgroup Theorems   ◾    451  

Theorem 6 (Kurosh Subgroup Theorem). Let G be the free product j ∈∗J G j and let H be a
subgroup of G. Then H = F ∗ ( k ∈∗K H k ), where F is a free group and for every k ∈ K there is
j ∈ J such that Hk is a conjugate of a subgroup of G j .

Proof. Follow the construction described in Proposition 2, Section 14.4, and rep-
resent each G j with a 2-dimensional CW-complex Xj such that π1 ( X j , x j ) ≅ G j . In
particular, each X j has exactly one vertex x j . Choose a point x ∉ ∪ X j and, for each
j ∈J
j ∈ J , attach a 1-cell [–1, 1] via the mapping –1  x,1  x j. We end up with a space X
depicted in Illustration 17.27.

Xk
Xj
xk
xj Xi
Xs
xi
xs
x

Illustration 17.27

An easy application of the SvK theorem yields π1 ( X , x ) ≅ G . We may identify these two
groups, so that we may view each G j as the fundamental group at x of the space U j consisting
of X j together with the edge from x to x j . The space X possesses the properties guaranteeing
the existence of the universal cover. Consequently, there is a cover (( X  , x ), p) , x ∈ p −1 ( x ) , of
X corresponding to the subgroup H of G.
Denote the subspace of X consisting of the vertex x, the edges adjacent to it, and the ver­
tices of x j by S (see Illustration 17.27). Then S is a simply connected tree; hence S = p −1 (S ) is a
forest. Denote the 1-skeleton of X by X 1 . By Proposition 1 in 17.3, p −1 ( X 1 ) is a graph. Observe
that p −1 (S ) is a subgraph of p −1 ( X 1 ). It follows by Lemma 3 that S can be extended to a maxi-
mal tree Tm in p −1 ( X 1 ). Since p −1 (S ) contains all of the vertices of p −1 ( X 1 ), to get Tm we need
to add new edges, where each of these new edges is in the 1-skeleton of some p −1 ( X j ) .
We say that a thin neighborhood of the 1-skeleton of a 2-dimensional CW-complex is
obtained by deleting the disk {( x , y ) : x 2 + y 2 ≥ 0.5} from every 2-cell during the process of
attaching 2-cells. Notice that the 1-skeleton of a 2-dimensional CW-complex is a defor-
mation retract of its thin neighborhood. Hence these two have the same fundamental
group.
We are now ready to describe an open cover of X  that satisfies the condition of the SvK

theorem. Let U be S together with the thin neighborhoods of the 1-skeletons of p −1 ( X j ),
j ∈ J (Illustration 17.28). Notice that this is an open set in X.  Let Vj∗ be the union of the
k

shortest geodesic along the tree Tm leading from the point x to a point in a component
Wjk of some p −1 ( X j ), and the component Wjk itself; extend Vj∗k to Vjk by adding by short
paths to get an open set (Illustration 17.29). Then clearly U together with all Vjk -s, j ∈ J , is
a cover of X. 

K12146_C017.indd 451 2/12/15 12:40 PM

 452   ◾    Applications in Group Theory

Wst

Wjk

~
x

 widened by its thin neighbor-

Illustration 17.28 The open set U consists of the 1-skeleton of X
hood in each Wj . k

Wjk

x~

Illustration 17.29 An open set Vjk ; the short trees sticking out of Wjk are included so that Vjk
.
is open in X

It follows from Lemma 2 that the intersection of any two distinct Vjk and Vst is a tree
containing the point x , and so it is simply connected. On the other hand, U ∩ Vjk is the
union of a geodesic in Tm (used in the construction of Vjk) together with the thin neigh-
borhood of the 1-skeleton of Wjk . Hence, the fundamental group of this intersection is a
free group, and, by Lemma 4, its generators also generate π1 (Vjk , x ).
Observe that π1 (U , x ) is isomorphic to the fundamental group of the 1-skeleton of X. 
Hence, it is also a free group. Since Wjk together with the restriction of p is a covering space
of X j , it follows that for every jk , π1 (Vjk , x ) is isomorphic to a conjugate of a subgroup of
π1 ( X j ) .
It follows from the SvK theorem that π1 ( X  , x ) is the free product of the groups π1 (U , x )
and π1 (Vk , x ) (k in some index set K), amalgamating certain groups. The groups along
which we are amalgamating are trivial in all cases except when arising from intersections
of type U ∩ Vk . These last intersections bring about the following types of relations into the
presentation of π1 ( X  , x ) :

free generator of π1 (U , x ) = generator of π1 (Vk , x ) ,

K12146_C017.indd 452 2/12/15 12:40 PM

 17.4 Two Subgroup Theorems   ◾    453  

where we have used Lemma 4 to claim that we indeed deal with generators of π1 (Vk , x ) .
Using Tietze transformations we can eliminate those free generators of π1 (U , x ) appearing
in such relations, as well as the relations themselves. The rest of the generators of π1 (U , x )
freely generate a subgroup F of π1 (U , x ), and the presentation reduces to the free product
( )
F ∗ ∗ π1 (Vk , x ) . The claim of Kurosh’s theorem now follows, since, as we have observed,
k∈K
each π1 (Vk , x ) is isomorphic to a conjugate of a subgroup of π1 ( X j ) .

Exercises
1. Let X be the space obtained by identifying a pair of points in two projective planes.
Describe 5 covering spaces of X and the associated 5 subgroups of π1 ( X ) =  2 ∗  2 .
2. Show that if a subgroup A of a free product ∗ H j intersects each conjugate of H j
j∈J
trivially, then A is a free group. Deduce that if A is a normal subgroup of G which
intersects each H j trivially, then A is free.
3. Define the epimorphism ϕ: G ∗H → G × H, by ϕ(g) = (g,1) if g ∈G and ϕ(h) = (1, h) if
h∈H. Show that ker(ϕ) is a free group.
4. Let N be the smallest normal subgroup of G ∗H containing H. Prove that N is the free
product of the subgroups gHg −1, g∈G.

K12146_C017.indd 453 2/12/15 12:40 PM

This page intentionally left blank
 Bibliography

The long bibliographical list that follows does not show the subjective differences in the
influences of these sources on the book at hand. I note the following three books from the
bibliography that stand out for me: Croom’s Principles of Topology (simply because I inher-
ited it when I started teaching topology, and I used it as the primary course textbook dur-
ing the first few years), Massey’s Algebraic Topology (my principal textbook when I studied
the subject as a student), and Magnus, Karrass, and Solitar’s Combinatorial Group Theory
(probably the most influential book in my own mathematical life). The influence of these
three books may be evident.
The articles in the list are indicated by []*.
[1] C. W. Baker, Introduction to Topology, Wm. C. Brown Pub., 1991.
[2] G. Baumslag, Topics in Combinatorial Group Theory, Birkhäuser Verlag, Basel, 1993.
[3] G. Baumslag and C. F. Miller III, Algorithms and Classifications in Combinatorial Group Theory,
Springer-Verlag, New York, 1992.
[4] K. G. Binmore, Topological Ideas, Cambridge University Press, Cambridge, 1981.
[5] B. Bollobás, Combinatorics, Cambridge University Press, London, 1986.
[6] N. Bourbaki, General Topology: Part 1, Addison-Wesley, Reading, MA, 1966.
[7] D. Bushaw, Elements of General Topology, John Wiley & Sons, New York, 1963.
[8] G. L. Cain, Introduction to General Topology, Addison-Wesley, Reading, MA, 1994.
[9] S. C. Carlson, Topology of Surfaces, Knots, and Manifolds, John Wiley and Sons, New York, 2001.
[10] B. Chandler and W. Magnus, The History of Combinatorial Group Theory: A Case Study in the
History of Ideas, Springer-Verlag, New York, 1982.
[11] D. E. Cohen, Combinatorial Group Theory: A Topological Approach, Cambridge University
Press, Cambridge, 1989.
[12] P. Cromwell, Knots and Links, Cambridge University Press, Cambridge, 2004.
[13] F. H. Croom, Principles of Topology, Sounders College, Philadelphia, 1989.
[14] S. W. Davis, Topology, McGraw-Hill, Boston, 2005.
[15] W. Dicks and M. J. Dunwoody, Groups Acting on Graphs, Cambridge University Press,
Cambridge, 1989.
[16] J. Dieudonné, A History of Algebraic and Differential Topology, 1900–1960, Springer, 2009.
[17] R. Engelking, General Topology, Heldermann Verlag, Berlin, 1989.
[18] R. Engelking, Outline of General Topology, North-Holland, Amsterdam, 1968.
[19]* S. Friedl, An Introduction to 3-Manifolds, http://www.mi.uni-koeln.de/~stfriedl/papers/
muenster.pdf
[20]* R. H. Fox, On the complementary domains of a certain pair of inequivalent knots, Proc. Konink.
Nederl. Akad. Wetensch. Ser. A 55 (1952) pp. 37–40; Indag. Math. 14 (1952) pp. 37–40.

455

K12146_C018.indd 455 2/10/15 4:47 PM

 456   ◾    Bibliography

[21]* R. H. Fox and E. Artin, Some wild cells and spheres in three-dimensional space, Ann. Math.
Second Series, 49(4): 979–990, October 1948.
[22] W. Fulton, Algebraic Topology: A First Course, Springer, New York, 1995.
[23]* D. Gale, The classification of 1-manifolds: A take-home exam, Am. Math. Monthly, 94(2): 170–
175, February 1987.
[24] T. W. Gamelin and R. E. Greene, Introduction to Topology, Sounders College, Philadelphia, 1983.
[25]* S. M. Gersten: Isoperimetric and isodiametric functions of finite presentations, in Geometric
Group Theory, Volume 1, London Mathematical Society, Cambridge University Press,
Cambridge 1993, pp. 79–96.
[26] S. E. Goodman, Beginning Topology, Thomson Brooks/Cole, Belmont, CA, 2005.
[27]* C. McA. Gordon and J. Luecke, Knots are determined by their complements, Bull. Am. Math.
Soc. 20(1), January 1989.
[28] J. Greever, Theory and Examples of Point-Set Topology, Brooks/Cole, Belmont, CA, 1968.
[29] A. G. Hamilton, Numbers, Sets and Axioms, Cambridge University Press, Cambridge, 1982.
[30] P. de la Harpe, Topics in Geometric Group Theory, The University of Chicago Press, Chicago, 2000.
[31] A. Hatcher, Algebraic Topology, Cambridge University Press, Cambridge, 2002.
[32] K.P. Hart, J. Nagata, and J. E. Vaughan (Editors), Encyclopedia of General Topology, Elsevier,
Amsterdam, 2004.
[33]* G. Higman, B. H. Neumann, and H. Neumann, Embedding theorems for groups, J. London
Math. Soc. s1-24(4): 247–254, 1949.
[34] J. G. Hocking and G. S. Young, Topology, Dover, New York, 1988.
[35] A. G. Howson, On the intersection of finitely generated free groups, J. London Math. Soc. 29:
428–434, 1954.
[36] I. M. James (Editor), History of Topology, Elsevier, Amsterdam, 1999.
[37] D. L. Johnson, Presentations of Groups, Cambridge University Press, Cambridge, 1990.
[38] K. D. Joshi, Introduction to General Topology, John Wiley & Sons, New York, 1983.
[39]* S. K., Topographs and groups, J. Pure Appl. Algebra 80: 47–57, 1992.
[40] I. Kaplansky, Set Theory and Metric Spaces, Allyn & Bacon, Boston, 1972.
[41] R. H. Kasriel, Undergraduate Topology, R. E. Krieger Publ., Malabar, FL, 1986.
[42] T. Lawson, Topology: A Geometric Approach, Oxford University Press, New York, 2003.
[43] J. M. Lee, Introduction to Topological Manifolds, Springer, New York, 2000.
[44] S. Lipschutz, Theory and Problems of General Topology, Schaum, New York, 1965.
[45] R. C. Lyndon, P. E. Schupp, Combinatorial Group Theory, Springer-Verlag, Berlin, 1977.
[46]* R. Maehara, The Jordan curve theorem via Brouwer fixed point theorem, Am. Math.
Monthly (Mathematical Association of America) 91(10): 641–643, 1984.
[47] W. Magnus, A. Karass, and D. Solitar, Combinatorial Group Theory: Presentations of Groups in
Terms of Generators and Relations, Interscience Publishers, New York, 1966.
[48] W. S. Massey, Algebraic Topology: An Introduction, Springer-Verlag, New York, 1967.
[49] J. Matoušek, Using the Borsuk-Ulam Theorem, Springer-Verlag, New York, 2003.
[50] S. Matveev, Algorithmic Topology and Classification of 3-Manifolds, Springer-Verlag, Berlin, 2007.
[51] R. Messer and P. Straffin, Topology Now! Mathematical Association of America, 2006.
[52]* C. F. Miler III, Combinatorial Group Theory, 2004, http://www.ms.unimelb.edu.au/~cfm/
notes/cgt-notes.pdf.
[53]* J. Milnor, A unique decomposition theorem for 3-manifolds, Am. J. Math. 84(1): 1–7, January 1962.
[54]* I. Mineyev, Submultiplicativity and the Hanna Neumann conjecture, Ann. Math. 175(1): 393–
414, 2012.
[55] E. E. Moise, Geometric Topology in Dimensions 2 and 3, Springer-Verlag, New York, 1977.
[56]* J. W. Morgan and G. Tian, Ricci Flow and the Poincaré Conjecture, http://arxiv.org/pdf/
math/0607607v2.pdf
[57] J. R. Munkres, Topology, Prentice-Hall. Englewood Cliffs, NJ, 2000.

K12146_C018.indd 456 2/10/15 4:47 PM

 Bibliography   ◾    457

[58]* R. Myers, End sums of irreducible open 3-manifolds, Quart. J. Math. Oxford Ser. (2), 50(197):
49–70, 1999.
[59]* H. Neumann, On the intersection of finitely generated free groups, Publ. Math. Debrecen 4:
186–189, 1956.
[60]* H. Neumann, On the intersection of finitely generated free groups; Addendum, Publ. Math.
Debrecen 5: 128, 1957.
[61] C. W. Patty, Foundations of Topology, Jones & Bartlett, Boston, 2009.
[62] H. Poincaré, (Translated by J. Stillwell), Papers on Topology: Analysis Situs and 5 Five
Supplements, AMS and London Math Society, 2010.
[63] D. Rolfsen, Knots and Links, Publish or Perish, Wilmington, DE, 1976.
[64] J. S. Rose, A Course in Group Theory, Cambridge University Press, Cambridge, 1978.
[65] J. J. Rotman, An Introduction to Algebraic Topology, Springer-Verlag, New York, 1988.
[66] J. J. Rotman, An Introduction to the Theory of Groups, Allyn & Bacon, Boston, 1984.
[67] A. W. Schurle, Topics in Topology, Elsevier, North Holland, 1979.
[68] H. Seifert and W. Threlfall, A Textbook of Topology, Academic Press, New York, 1980.
[69]* M. Siegmund, Generalized Torelli Groups, Dissertation, Düsseldorf Univ., 2007.
[70]* B. de Smit, The fundamental group of the Hawaiian earring is not free, Int. J. Algebra Comput.
2(1): 33–37, 1992.
[71] L. A. Steen and J. A. Seebach, Jr., Counterexamples in Topology, Springer-Verlag, New York, 1978.
[72] J. Stillwell, General Topology and Combinatorial Group Theory, Springer-Verlag, New York, 1980.
[73]* C. Thomassen, The Jordan-Schönflies theorem and the classification of surfaces, Am. Math.
Monthly, 99(2):116–131, 1992.
[74] W. P. Thurston, Three-Dimensional Geometry and Topology, Princeton University Press,
Princeton, NJ, 1997.
[75] O. Ya. Viro, O. A. Ivanov, N. Yu. Netsvetaev, and V. M. Kharlamov, Elementary Topology:
Problem Book, American Mathematical Society, Providence, RI, 2008.
[76]* J. H. C. Whitehead, On certain elements in a free group, Proc. London Math. Soc. 41: 48–56, 1936.
[77]* J. H. C. Whitehead, On equivalent sets of elements in a free group, Ann. Math. (2) 37:
pp. 782–800, 1936.
[78] S. Willard, General Topology, Dover, Mineola, NY, 2004.

In Russian:

[79] А. В. Архангелъский, В. И. Пономарев, Основы общей топологии в задачах и

упражнения, Издателъство “Наука,” Москва, 1974.
[80] В. В. Прасолов, Наглядная топология, МЦНМО, 1995.

K12146_C018.indd 457 2/10/15 4:47 PM

This page intentionally left blank
 List of Symbols

SYMBOL MEANING PAGE

∅ empty set 3
⊂ subset (may not be proper) 3
⊊ proper subset 3
∈ belongs to 3
∪ union 3
∩ intersection 3
Ac complement of a set A 3
A\ B A ∩ Bc 3
 set {0,1,2,…} of natural numbers 4
 set {…, −1, −1,0,1,2,…} of integers 4
+ set {1,2,…} of positive integers 4
A×B product of the sets A and B 4
n

∏X
i =1
i product of the sets Xi , i = 1,2,…, n 4

Xn product of the set X = Xi , i = 1,2,…, n 4

< relation of strict order 4
≤ relation of order 4
(a,b) open interval 5
(a,b] left half-open interval 5
[a, b) right half-open interval 5
[a, b] closed interval 5
~ equivalence relation 5
[x] equivalence class of the element x 5
X quotient set 5
~
 set of rational numbers 6
 set of real numbers 6
 +
set of positive real numbers 7
sup(A) supremum of A 8
inf (A) infimum of A 8
A cardinality of the set A 9

ℵ0 countably infinite cardinality 9

a↦b a is mapped to b 9
+ set of positive rational numbers 10

459

K12146_C019.indd 459 2/13/15 2:32 PM

 460   ◾    List of Symbols

SYMBOL MEANING PAGE

P( A) power set of A 10
2 m
carinality of P( A) if A = m 11
13
∏A i∈I
i set product of the sets Ai , i ∈I

(ai )i∈I element of Π

i∈I
Ai 13

(a1 , a2 ,..., an ) element of A1 × A2 ×  × An 13

(a1 , a2 ,..., an ,...) element of A1 × A2 ×  × An × … 13
d Eucledean metric, unless otherwise stated 17
(X,d) metric space X with metric d 19
B(a,r) open ball centered at a with radius r 21
lim xn limit of the sequence ( xn ) 24
n→∞

X ≅Y metric spaces X and Y are homeomorphic 26

dp product metric 27
(X, τ) topological space 34
n Euclidean n-space (unless otherwise stated) 34
intA interior of the set A 39
A′ set of the accumulation points of the set A 41
A closure of the set A 43
∂A boundary of the set A 44
 2
up
2
set {( x , y ) ∈ : y ≥ 0} 51
X≅Y spaces X and Y are homeomorphic 64
S1 3 2
unit circle {( x , y ) ∈ : x + y = 1} 2 65
S2 unit 2-sphere {( x , y , z ) ∈ 3 : x 2 + y 2 + z 2 = 1} 65

Sn unit n-sphere {( x1 , x 2 ,…, xn+1 ) ∈n+1 : x12 + x 22 +  + xn2+1 = 1} 65

g∘ f composition of f and g, in that order 68
f A restriction of f on the set A 73

B 2 2 2
open unit disk {( x , y ) ∈ : x + y < 1} 2 74
X quotient space 75
~
~f equivalence modulo a mapping f 77
X quoteint space determined by f : X → Y 77
~f

⊕X i discrete or topological sum of spaces 82

i∈I

X ∪f Y space obtained by gluing along f : A → B, A ⊂ X , B ⊂ Y 83

Zf space obtained by gluing along f : A → B, A, B ⊂ Z , A ∩ B = ∅ 83
T 2 1
torus S × S 1 94
D 2 2 2
closed unit disk {( x1 , x 2 ) ∈ : x + x ≤ 1}
1
2
2
95
n n
D 2 2 2
closed unit ball {( x1 , x 2 ,..., xn ) ∈ : x1 + x 2 +  + xn ≤ 1} 95
n
X n-skeleton of a CW-complex X 95
X  ×  Y product space of spaces X and Y 101

K12146_C019.indd 460 2/13/15 2:32 PM

 List of Symbols   ◾    461

SYMBOL MEANING PAGE

n

∏X
i =1
i
product space of spaces Xi , i = 1,2,…, n 107

∏X
i∈I
i
product space of spaces Xi , i ∈I 110

Xn product space of n copies of X 110

∞
∞
X ∏ Xi where each Xi is X 110
i =1
Box

∏X
i∈I
i box product of the spaces Xi , i ∈ I 117

A∪ B
⋅
A ∪ B where A ∩ B = ∅ 124

I unit interval [0, 1] 134

αβ product of the paths α and β, in that order 137
P2 projective plane 166
X∞ one-point compactification of the space X 175
A (r )
the set of all subsets C of A such that C = r 181

fˆ : X →  the extension of f : A →  , A ⊂ X 205

C*(X ) the metric space of all bounded functions X →  210

β(X) Stone-Čech compactification of X 212
Ht The mapping defined by Ht ( x ) = H ( x , t ) 218
Bn braid grop on n strings 225
f≃g mappings f and g are homotopic 228
f A g mappings f and g are homotopic relative to the set A 228
id X the identity mapping x  x for every x ∈ X 229
α≃β paths α and β are homotopic relative {0,1}, unless otherwise stated 232
δ −1
inverse of the path δ 235
π1 ( X , x ) the fundamental group of the space X at the point x, or the first 238
homotopy group of X at x
cx constant loop, or trivial loop at x 238
f :(X, x) → (Y, y) mapping f : X → Y such that f ( x ) = y 241
f∗ : π1 ( X , x ) → π1 (Y , y ) homomorphism induced by f : ( X , x ) → (Y , y ) 241
G×H direct product of the groups G and H 244
G ⊕H direct sum of the commutative groups G and H 244
xy line segment between the points x and y 264

xy ark between the points x and y 264
N⊲G N is a normal subgroup of G 269
A −1 −1 −1
set {a , b ,...} of formal inverses of A = {a, b,…} 269
A ±
A ∪ A −1 269
W length of the word W 269
〈a, b,...; P , Q ,...〉 group presentation, with generators a, b,… and relation P , Q ,… 270

K12146_C019.indd 461 2/13/15 2:32 PM

 462   ◾    List of Symbols

SYMBOL MEANING PAGE

U~V words U and V are equivalent 270
( x1 x 2 ... xn ) permutation x1  x 2 , x 2  x 3 , ... , xn  x1 272
Ker ( f ) kernel of a homomorphism 272
NC(A) normal closure of A in a group 275
G1 ∗ G2 free product of the groups G1 and G2 283
∗ Gj free product of the groups G j , j ∈ J 284
j ∈J

G1 ∗ G2 free product of G1 and G2 with amalgamation 287

H

∗ Gj free product of G j , j ∈ J , with amalgamation 287

H jk

(p, q) type of a torus knot, p and q are relatively prime positive integers 320
M1 # M 2 connected sum of two n-manifolds 341
M1 # M 2 #  # Mn connected sum of the n-manifolds M1 , M 2 , ... , Mn 342
X1 # ∂T1 ≅∂T2 X 2 connected sum along tori 359
L( p, q) lens space 359
πn ( X , x ) n-th homotopy group 365
 , p)
(X covering space of the space X 371
f a lift of a mapping f 376

Tn
1
n-torus S S1
× ×
 S1
× 390
n

A( X , p) or A( X )  , p)
group of covering transformations of ( X 401
N(H) normalizer, largest subgroup in which H is normal 401
Γ (G , A ) or Γ G or Γ Cayley graph of the group G generated by A 429
Sn symmetric group on a set with n elements 430
An alternating group 430
V(Γ) set of vertices of a graph Γ 436
t(v) altitude of a vertex v 436
GL(n, ) general linear group of matrices 440
Core(Y) core of the graph Γ 446

K12146_C019.indd 462 2/13/15 2:32 PM

This page intentionally left blank
 Mathematics
AN ILLUSTRATED
INTRODUCTION TO

TOPOLOGY and HOMOTOPY

AN ILLUSTRATED INTRODUCTION TO

TOPOLOGY and HOMOTOPY

TOPOLOGY

AN ILLUSTRATED INTRODUCTION TO
An Illustrated Introduction to Topology and Homotopy explores the beauty of
topology and homotopy theory in a direct and engaging manner while illustrating
the power of the theory through many, often surprising, applications. This self-
contained book takes a visual and rigorous approach that incorporates both and
HOMOTOPY
extensive illustrations and full proofs.
The first part of the text covers basic topology, ranging from metric spaces and
the axioms of topology through subspaces, product spaces, connectedness,
compactness, and separation axioms to Urysohn’s lemma, Tietze’s theorems,
and Stone-Čech compactification. Focusing on homotopy, the second part starts
with the notions of ambient isotopy, homotopy, and the fundamental group. The
book then covers basic combinatorial group theory, the Seifert-van Kampen
theorem, knots, and low-dimensional manifolds. The last three chapters discuss
the theory of covering spaces, the Borsuk-Ulam theorem, and applications in
group theory, including various subgroup theorems.
Requiring only some familiarity with group theory, the text includes a large number
of figures as well as various examples that show how the theory can be applied.
Each section starts with brief historical notes that trace the growth of the subject
and ends with a set of exercises.
Features
• Provides a comprehensive treatment of basic topology and homotopy that
uses a combination of rigorous proofs and extensive visualization
KALAJDZIEVSKI
• Gives full details of most proofs
• Contains nearly 600 figures that help clarify difficult concepts
• Presents historical notes that outline the growth of the subject
• Includes about 750 exercises, many of which are relatively new

K12146
SASHO KALAJDZIEVSKI

w w w. c rc p r e s s . c o m

K12146_cover.indd 1 2/19/15 11:01 AM