UNITEXT 134

K. Parthasarathy

Topology
An Invitation
UNITEXT

La Matematica per il 3+2

Volume 134

Editor-in-Chief
Alfio Quarteroni, Politecnico di Milano, Milan, Italy
École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland

Series Editors
Luigi Ambrosio, Scuola Normale Superiore, Pisa, Italy
Paolo Biscari, Politecnico di Milano, Milan, Italy
Ciro Ciliberto, Università di Roma “Tor Vergata”, Rome, Italy
Camillo De Lellis, Institute for Advanced Study, Princeton, NJ, USA
Massimiliano Gubinelli, Hausdorff Center for Mathematics, Rheinische
Friedrich-Wilhelms-Universität, Bonn, Germany
Victor Panaretos, Institute of Mathematics, École Polytechnique Fédérale de
Lausanne (EPFL), Lausanne, Switzerland
The UNITEXT - La Matematica per il 3+2 series is designed for undergraduate
and graduate academic courses, and also includes advanced textbooks at a research
level.
Originally released in Italian, the series now publishes textbooks in English
addressed to students in mathematics worldwide.
Some of the most successful books in the series have evolved through several
editions, adapting to the evolution of teaching curricula.
Submissions must include at least 3 sample chapters, a table of contents, and a
preface outlining the aims and scope of the book, how the book fits in with the
current literature, and which courses the book is suitable for.
For any further information, please contact the Editor at Springer: francesca.
[email protected]
THE SERIES IS INDEXED IN SCOPUS

More information about this subseries at https://link.springer.com/bookseries/5418

K. Parthasarathy

Topology
An Invitation
K. Parthasarathy (Emeritus)
Ramanujan Institute
University of Madras
Chennai, Tamil Nadu, India

ISSN 2038-5714 ISSN 2532-3318 (electronic)

UNITEXT
ISSN 2038-5722 ISSN 2038-5757 (electronic)
La Matematica per il 3+2
ISBN 978-981-16-9483-7 ISBN 978-981-16-9484-4 (eBook)
https://doi.org/10.1007/978-981-16-9484-4

Mathematics Subject Classification: 54Axx, 54Bxx, 54Cxx, 54Dxx, 54Exx, 54Fxx, 54Hxx, 55Pxx, 57Nxx

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Singapore Pte Ltd. 2022
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore
(Ancient Sanskrit ‘Sloka:’)
One fourth from the teacher, one fourth from
student’s own intelligence,
One fourth from co-learners, and one fourth
in course of time.
Dedicated to
my parents
Lakshmi and Krishnan
during my father’s centenary year
and to
Kannamma
Preface

Mathematics is the most beautiful and the most powerful creation of the human
spirit, said Stefan Banach. One may add: Topology is one of the most beautiful
and powerful branches of mathematics. The word ‘topology’ means ‘study of posi-
tion/place’ and is derived from two Greek words ‘topos’ and ‘logos’. From a math-
ematical point of view, a succinct description of topology may be given as ‘a study
of continuity’. Topology is indispensable in analysis and geometry. Far from being
an abstract subject, it has been growing into a versatile tool of great value in many
branches of science—physics (cosmology, quantum field theory, string theory and
condensed matter physics), biology (molecular structure, membrane biology, neural
networks, evolutionary biology, enzymes and DNA, protein folding), computer
science (topological data analysis, a ‘hot’ subject now), robotics, etc.
This book is intended to be an introduction to topology with historical perspective.
It discusses all the basics of general topology that a student of mathematics would
need and provides glimpses of the beginnings of algebraic topology and combinato-
rial methods in understanding the topology of the sphere and the Euclidean space.
For students having familiarity with elementary real analysis and some felicity with
the language of set theory and abstract mathematical reasoning, this book would
be suitable as a textbook both for a first course in topology and for self-study. The
language of sets is indispensable in modern mathematics, but we do not present or
use any axiomatic set theory, per se. The only exception to this is the use of Axiom of
Choice in the form of Zorn’s lemma in the proof of Tychonoff’s theorem on products
of compact spaces. We give a brief discussion on the Axiom of Choice and some of
its equivalents in that chapter.
This book starts with a discussion of the classical intermediate value theorem
and some of its uncommon consequences as an appetiser to kindle the interest of
the reader before the main course is served. The introduction of general topological
spaces and continuous maps on them in Chaps. 3 and 4 is preceded by a moti-
vating discussion of metric spaces and continuous maps on these (Chap. 2). Several
important special properties of spaces—compactness, connectedness, separation and
countability axioms—are presented in Chaps. 5, 8–10. In between, topologies defined
by maps—initial and final topologies—are presented in Chap. 6. The product and

ix
x Preface

quotient topologies are seen as important special cases of these. A separate chapter
(Chap. 7) is devoted to products of compact spaces. Locally compact spaces are
treated in Chap. 11 and complete metric spaces in Chap. 12. Elementary, combina-
torial treatments of some of the most important results on Euclidean spaces—the
higher-dimensional intermediate value theorem of Poincaré–Miranda, Brouwer’s
fixed point theorem, the no retract theorem, theorems on invariance of domain
and dimension, Borsuk’s antipodal theorem, the Borsuk–Ulam theorem, Lusternik–
Schnirelmann–Borusk theorem—are given in Chap. 13. The last two chapters provide
an introduction to homotopy, fundamental groups and covering spaces.
Although most of the topics discussed in general topology are standard, there
are a few that are not common in topology texts. Instances are: the intermediate
value theorem and its consequences, Bourbaki’s Mittag–Leffler theorem (from which
Baire’s theorem is deduced) and (a simple case of) continuous partitions of unity.
Nets are defined early, after limit points are introduced and are used again later as
needed—for example, to characterise continuity and compactness.
This book has several salient features. An unusual one among them is its historical
perspective. The author’s perception is that learning mathematics at any level should
be spiced up with a dash of historical development. With this in mind, original
formulations for all the basic concepts introduced and original statements of major
results are provided (all with English versions) before modern versions are given.
(The relevant original references, starting from early nineteenth century, are listed in
the bibliography.) This should give some idea of how, and through whom, concepts
developed before they took the current shape. However, this is not a book on the
history of topology—far from it. This is an introductory book on topology that also
wishes to provide a bit of historical flavour of the topics discussed. In addition to
the historical remarks interspersed throughout the text, thumbnail sketches of the
dramatis personae involved in the development of topics discussed are provided at
the end of each chapter. This is intended to enhance the reader’s understanding of
the historical development of the subject.
Poincaré’s higher-dimensional analogue of the intermediate value theorem does
not appear to be well-known even among professional mathematicians (vis-à-vis,
say, the equivalent Brouwer fixed point theorem.) The presentation of elementary
approaches to these and several stellar results like those on domain and dimension
invariance and Borsuk’s results on maps on spheres is another notable feature of the
book.
Recent (one as recent as 2020) and simple proofs of many ‘big’ theorems are
presented. Some instances are: Étienne Matheron’s proof (2020) and a 2003 version
of Paul Chernoff’s proof of Tychonoff’s theorem; Kulpa’s proof of the Poincaré
theorem and Brouwer fixed point theorem following Müger (2015); a 2005 proof
of Ky Fan’s lemma (used in the proof of Borsuk’s theorems); and Kulpa’s proof of
domain invariance theorem (also presented by Terence Tao (2014)).
Numerous exercises, of various levels of difficulty, are given at the end of each
chapter, listed topic-wise for the reader’s convenience. (If the reader is expected to
prove a statement, only the statement is given without the words ‘prove’ or ‘show’.)
Preface xi

In addition, there are some scattered, in situ exercises, following a definition or a

result. They are meant to clarify the reader’s understanding of the concept introduced.
A note to the reader
As Paul Halmos famously and laconically said, ‘The best way to learn is to do’.1
So the serious students are advised to arm themselves with paper and pencil while
reading the book and not to take any of the author’s statements for granted unless
they are convinced of its veracity. Examples and exercises are indispensable parts of
the learning process. The student should try to acquire a good repertoire of examples
and try as many exercises as possible to get a good understanding of the subject.
Something can be learned even from a partial success or a failed attempt to solve a
problem. Asking questions is an essential skill to be honed by a knowledge seeker.
Without questions, there are no answers! In fact, Georg Cantor said: In mathematics
the art of proposing a question must be held of higher value than solving it.
The author is no historian of mathematics. The sources for the historical and
biographical notes that are included in the book are: historical notes in some of the
standard books on topology (e.g. [1–5]), books devoted to the history and historical
aspects of topology listed in the bibliography, articles on historical developments [6–
15], MacTutor, ‘mathshistory’ website of St. Andrews University, UK (https://mat
hshistory.st-andrews.ac.uk) and Wikipedia entries. Many quotes of old masters used
in the book, including those of Banach and Cantor here, are from
https://mathshistory.st-andrews.ac.uk/Biographies.
The fact that the author, with zero expertise in such matters, has been able to draw
the figures himself using Latex is a big surprise. The credit goes to the generosity of
the Ubuntu/Latex community for its tutorials and illustrations freely available on the
Internet. Almost to the same degree, this holds for getting the material into a book
form.
Help in accessing original papers came from P. Aprameyan. He and P. Anutthaman
assisted in translating German passages. (The French parts were translated by the
author.) T. C. Easwaran Nambudiri chipped in with a preliminary index.

Chennai, India K. Parthasarathy

References
1. Bourbaki, N.: General Topology, Chapters 1–4. Springer (1995)
2. Bourbaki, N.: General Topology, Chapters 5–8. Springer (1998)
3. Engelking, R.: General Topology Revised Edition. Heldermann Verlag (1989)
4. Kelley, J.L.: General Topology. Affiliated East-West Press (1969)
5. Willard, S.: General Topology. Addison-Wesely (1970)
6. Andre, N., Engdahl, S., Parker, A.: An analysis of the first proofs of the Heine-Borel Theorem-
Borel’s Proof”. Loci (2003). https://doi.org/10.4169/loci003890

1[16] © Mathematical Association of America, 1975. All rights reserved. Quoted here with
permission.
xii Preface

7. Borel, A.: Twenty-Five Years with Nicolas Bourbaki (1949–1973). Not. Am. Math. Soc. 45,
373–380 (1998)
8. James, I.M.: From combinatorial topology to algebraic topology. In: James, I.M. (ed.) History
of Topology, pp. 561–574. Elsevier (1999)
9. Raman-Sundström, M.: A pedagogical history of compactness. Am. Math. Monthly 122,
619–635 (2015). https://doi.org/10.4169/amer.math.monthly.122.7.619
10. Reitberger, H.: The contributions of L. Vietoris and H. Tietze to the foundations of general
topology. In: Aull, C.E., Lowen, R. (eds.) Handbook of the History of General Topology,
vol. 1, pp. 51–72. Kluwer (1997)
11. Reitberger, H.: Leopold Vietoris, 1891–2002. Not. Am. Math. Soc. 49(10), 1232–1236 (2002)
12 Rodriguez, L.: Frigyes Riesz and the emergence of general topology. Arch. History Exact
Sci. 69, 55–102 (2015). https://doi.org/10.1007/s00407-014-0144-6
13. Russ, S.: A translation of Bolzano’s paper on the intermediate value theorem. Historia Math.
7, 156–186 (1980). https://doi.org/10.1016/0315-0860(80)90036-1
14. Sarkaria, K.S.: The topological work of Henri Poincaré. In: James, I.M. (ed.) History of
Topology, pp. 123–167. North-Holland (1999)
15. Thron, W.J. Frederic Riesz’ contributions to the foundations of general topology. In: Aull,
C.E., Lowen, R. (eds.) Handbook of the History of General Topology, vol. 1, pp. 31–40.
Kluwer (1997)
16. Halmos, P.R.: The problem of learning to teach. Am. Math. Monthly 82, 466–476 (1975).
https://doi.org/10.2307/2319737
Contents

1 Apéritif : The Intermediate Value Theorem . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Intermediate Value Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Biographical Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.1 Bolzano . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1 Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Continuity and Open Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Biographical Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.1 Fréchet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3 Topological Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.1 Topologies and Open Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 Basic Open Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3 Closed Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.4 Biographical Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.4.1 Riemann . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.4.2 Weyl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.4.3 Hausdorff . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.4.4 F. Riesz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4 Continuous Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.1 Limit Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.2 Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.3 Biographical Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.3.1 Cauchy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.3.2 Weierstrass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.3.3 Dirichlet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

xiii
xiv Contents

5 Compact Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.1 Compactness in Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.2 Compactness in Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.3 Compactness in Topological Spaces . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.4 Biographical Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.4.1 E. Borel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.4.2 Heine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.4.3 Lebesgue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.4.4 Cantor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.4.5 Vietoris . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
6 Topologies Defined by Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
6.1 Initial and Final Topologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
6.2 Product Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
6.3 Quotient Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
6.4 Biographical Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
6.4.1 R. L. Moore . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
6.4.2 Möbius . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
6.4.3 Klein . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
7 Products of Compact Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
7.1 Tychonoff’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
7.2 Appendix: Axiom of Choice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
7.3 Biographical Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
7.3.1 Bourbaki . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
7.3.2 Čech . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
7.3.3 Tychonoff . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
7.3.4 Kelley . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
8 Separation Axioms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
8.1 Hausdorff Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
8.2 Normal Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
8.3 Regular Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
8.4 Completely Regular Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
8.5 Biographical Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
8.5.1 Tietze . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
8.5.2 Urysohn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
8.5.3 Carathéodory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
8.5.4 M. H. Stone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Contents xv

9 Connected Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

9.1 Path Connected Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
9.2 Connected Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
9.3 Locally Connected Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
9.4 Biographical Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
9.4.1 Jordan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
9.4.2 Hahn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
9.4.3 Kuratowski . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
9.4.4 Knaster . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
10 Countability Axioms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
10.1 Countability Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
10.2 Urysohn Metrisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
10.3 Biographical Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
10.3.1 Lindelöf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
11 Locally Compact Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
11.1 Local Compactness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
11.2 One-Point Compactification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
11.3 Biographical Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
11.3.1 Alexandroff . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
11.3.2 Dieudonné . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
12 Complete Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
12.1 Completeness and Ascoli–Arzelà . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
12.2 Bourbaki, Baire and Banach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
12.3 Completion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
12.4 Biographical Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
12.4.1 Ascoli . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
12.4.2 Arzelà . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
12.4.3 Baire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
12.4.4 Banach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
13 Combinatorial Methods in Euclidean Topology . . . . . . . . . . . . . . . . . . 181
13.1 Convex Sets and Balls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
13.2 Cubes and Simplices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
13.3 Sperner’s Lemma, the Cubical Version . . . . . . . . . . . . . . . . . . . . . . 186
13.4 Poincaré and Brouwer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
13.5 Invariance of Domain and Dimension . . . . . . . . . . . . . . . . . . . . . . . 192
13.6 Borsuk and the Sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
13.7 Biographical Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
13.7.1 Bohl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
13.7.2 Hadamard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
xvi Contents

13.7.3 Ulam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203

13.7.4 Sperner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
13.7.5 Mazurkiewicz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
13.7.6 Brouwer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
14 Homotopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
14.1 Retracts and Deformation Retracts . . . . . . . . . . . . . . . . . . . . . . . . . . 208
14.2 Homotopy of Maps and Paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
14.3 Biographical Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
14.3.1 Borsuk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
15 Fundamental Groups and Covering Spaces . . . . . . . . . . . . . . . . . . . . . . 223
15.1 The Fundamental Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
15.2 Examples and Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
15.3 Covering Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
15.4 Biographical Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
15.4.1 Poincaré . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246

Appendix: Selected Exercises—Suggestions and Hints . . . . . . . . . . . . . . . . 249

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
About the Author

K. Parthasarathy is Former Director and Head of the Ramanujan Institute for

Advanced Study in Mathematics, University of Madras, Chennai, India. He earned his
doctoral degree from the Indian Institute of Technology Kanpur, after his schooling
and college education in Chennai (earlier Madras). His areas of research are abstract
harmonic analysis and the theory of frames. He had taught subjects ranging from
algebraic number theory to algebraic topology, differential equations to differential
geometry and linear algebra to Lie algebras for about 35 years at the postgraduate
level at different institutions. He had been Doctoral Adviser for several students and
has published a number of research papers in international journals of repute. He is a
reviewer for several research journals and for Mathematical Reviews and zbMATH.

xvii
Chapter 1
Apéritif : The Intermediate Value
Theorem

1.1 Intermediate Value Theorem

A function which is continuous either in general, or within given limits, does not
change from one of its values into another value without first having to take all
values lying between them at least once. [1].
That was how Bolzano, a mathematician (and philosopher) of Italian origin based
in Prague, stated his now-famous intermediate value theorem in 1817. The topologi-
cal concept underlying this will be taken up for discussion later. Here, we take a look
at a ‘real analysis’ proof of this theorem and some of its interesting consequences,
including a few that are not often discussed. This is meant as an hors d’oeuvre to
whet the appetite of the reader for the gourmet course to follow.

Definition 1.1 A subset I of R, the set of real numbers, is called an interval if it

contains all points between any two of its points; in symbols: if a, b ∈ I, a < b and
a < c < b, then c ∈ I.

Bolzano’s result states that a continuous function takes every value between any
two of its values. Using the term ‘interval’ it reads as follows. (Note that a one-point
set is also a (degenerate) interval, so the result holds for constant functions too.)
Theorem 1.2 (Bolzano’s Intermediate Value Theorem) Let I be an interval in R. If
f : I → R is continuous, then the image f (I ) = { f (x) : x ∈ I } is also an interval.

Proof Recall the ε-δ definition of continuity of f at x0 ∈ I : for ε > 0, there is

a δ = δ(x0 , ε) > 0 such that | f (x) − f (x0 )| < ε when |x − x0 | < δ. Let a, b ∈ I,
a < b and suppose f (a) < f (b). If f (a) < y0 < f (b), to show that y0 = f (x0 )
for some x0 ∈ I . We will, in fact, show that there is such an x0 with a ≤ x0 ≤ b.
To prove this, let E = {x ∈ I : a ≤ x ≤ b and f (x) > y0 }. Then b ∈ E ⊂ I , so E
is a nonempty set, bounded below by a. Thus, x0 = inf E exists and a ≤ x0 ≤ b
(why?), so x0 ∈ I . Claim: f (x0 ) = y0 . By definition, f (x0 ) ≥ y0 . If f (x0 ) > y0 ,
then take an ε > 0 such that f (x0 ) − ε > y0 . Use continuity of f at x0 : there is a
δ > 0 such that f (x0 ) − ε < f (x) < f (x0 ) + ε for x0 − δ < x < x0 + δ, so there
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 1
K. Parthasarathy, Topology, La Matematica per il 3+2 134,
https://doi.org/10.1007/978-981-16-9484-4_1
2 1 Apéritif : The Intermediate Value Theorem

is an x, a < x < x0 , with f (x) > y0 . This x is in E, a contradiction, as x0 is a lower

bound for E. Thus f (x0 ) = y0 . The case f (b) < f (a) is similar. 
Remark 1.3 1) A discontinuous function can have the intermediate value prop-
erty, e.g. f (x) = x, 0 < x < 1, f (0) = 1, f (1) = 0. Here is another such function:
f (0) = 0, f (x) = sin(1/x), 0 < x ≤ 1.
2) We will come to the topological content of this theorem, and give a topological
proof as well, later.

Here is the usual, ‘real analysis’ form of the result.

Corollary 1.4 If f : [a, b] → R is continuous, and if f (a), f (b) have opposite

signs, then there is a point c ∈ (a, b), such that f (c) = 0.

Remark 1.5 Geometrically, this is ‘obvious’: if one end of the graph of f lies below
and the other lies above the x-axis, then it must cut the x-axis at some point in
between. It is remarkable that Bolzano recognised the g.l.b. property of real numbers
decades before Dedekind and Weierstrass, and used it essentially as we have done.
We now discuss some simple but interesting and important consequences. The
higher dimensional analogues will be taken up later.

Theorem 1.6 If f : [a, b] → R is continuous and f (a) ≥ a and f (b) ≤ b, then

there is a c ∈ [a, b] with f (c) = c.

Proof Geometrically: a continuous curve joining a point on a side of a square to a

point on the opposite side has to cut the diagonal (Fig. 1.1). An analytical proof is
hardly more difficult.
The result holds if f (a) = a or f (b) = b; otherwise f (a) > a and f (b) < b. If
g(x) = x − f (x), then g(a) < 0 and g(b) > b and the intermediate value theorem
1.4 yields a c ∈ (a, b) with f (c) = c. 

Corollary 1.7 (Brouwer fixed point theorem, dim. 1) If f : [a, b] → [a, b] is con-
tinuous, then f has a fixed point, i.e. there is a c ∈ [a, b] such that f (c) = c.

We have deduced the fixed point theorem from the intermediate value theorem. Here
is the converse.

Proposition 1.8 If Theorem 1.6 is valid, then so is Corollary 1.4.

Proof Suppose f : [a, b] → R is continuous and f (a), f (b) have opposite signs.
Apply 1.6 to g(x) = x − f (x) if f (a) < 0 and f (b) > 0 and to h(x) = x + f (x)
if f (a) > 0 and f (b) < 0. 

Exercise: Examine what happens for other types of intervals.

Recall polar coordinates: for any z ∈ C∗ , z = |z|eıθ , θ ∈ R. Here θ is not unique
and any choice of θ is called an argument of z, written arg z. If θ, φ are any two
arguments, then θ − φ ∈ 2π Z. Any value of arg z 1 z 2 is a value of arg z 1 + arg z 2 , and
conversely. There is no continuous choice for arg z, as the following result shows.
1.1 Intermediate Value Theorem 3

(b,b)

y=f(x)

f(c)=c

(a,a)

Fig. 1.1 Fixed point theorem on [a, b]

Proposition 1.9 There is no continuous function θ : C∗ → R such that z = |z|eıθ(z) ,

z ∈ C∗ .
Proof Suppose there is such a continuous function θ and consider the continuous
function f : [0, 2π ] → R defined by f (t) = θ (eıt ) + θ (e−ıt ). Since θ (z 1 ) + θ (z 2 )
is a value of arg z 1 z 2 , the values of f lie in the set arg 1 of values of arguments
of 1. But these are just the integral multiples of 2π , and so the intermediate value
theorem ensures that it is a constant. In particular, 2θ (1) = f (0) = f (π ) = 2θ (−1).
But θ (1), being a value of arg 1, must be an even multiple of 2π , whereas θ (−1)
is a value of arg −1 and so is an odd multiple of 2π . This contradiction proves the
result. 
Have you seen a rigorous proof of the following familiar result?
Proposition 1.10 Every odd degree polynomial with real coefficients has at least
one real root.
Proof Let f (x) be a polynomial with real coefficients of odd degree. Then f is
continuous as a real valued function on R. The sign of the leading term determines
the sign of f (x) for large |x|. (Why? Intuitively clear? Proper proof?) We may
assume, without loss of generality, that the leading coefficient of f (x) is positive.
Then if a > 0 is large enough, f (a) > 0 and f (−a) < 0. The intermediate value
theorem 1.4 now implies the existence of a root: there is an x, −a < x < a, such
that f (x) = 0. 
The next application needs the concept of continuity of a real valued function
defined on the unit circle S1 = {(x, y) ∈ R2 : x 2 + y 2 = 1}. This, and continuity
of functions defined on spheres etc, would be taken for granted in the rest of this
chapter. (If you are not familiar with these from your real analysis course, see the
next chapter.)
4 1 Apéritif : The Intermediate Value Theorem

Theorem 1.11 (Borsuk–Ulam theorem, dim. 1) If f : S1 → R is continuous, then

there is a pair of antipodal points p and p ∗ = − p such that f ( p) = f ( p ∗ ).

Proof Apply the intermediate value theorem to the continuous √ function

g : [−1, 1] → R defined by g(x) = f ( p) − f (− p) where p = (x, (1 − x 2 )), not-
ing that g(−1) = −g(1). Conclude that either g(−1) = 0 = g(1) or g(1) and g(−1)
have opposite signs, in which case the intermediate value theorem 1.4 is applicable.
In either case, we get a point p with f ( p) = f (− p). 

Here is a meteorological ‘theorem’.

Corollary 1.12 At any instant of time, there are two points on any great circle of
the earth having the same temperature.

Proof Assuming that at any instant of time the temperature is a continuous function
of the position, apply Borsuk–Ulam 1.11 for the great circle under consideration in
place of S1 . 

Remark 1.13 For this amusing consequence, one may consider, in place of temper-
ature, any other meteorological quantity, e.g. pressure.

Now from meteorological to gastronomical considerations! What is usually called

the pancake theorem, we call the dosa theorem (Indian flavour!) Two thin dosas,
unfortunately not circular as they should be, are laid apart on the table. Is it possible,
with a single cut of the knife, to cut both the dosas into two equal halves? Yes, says
the theorem.

Theorem 1.14 (Dosa theorem) Let A1 , A2 be the areas enclosed by two simple
closed curves in the plane. Then there is a straight line which divides both A1 , A2
into two equal parts.

Proof (Following Chinn-Steenrod [2], Shashkin [3]) See Fig. 1.2. We use both the
intermediate value and the Borsuk–Ulam theorems. Enclose both A1 and A2 in a
sufficiently large circle C with centre at the origin. For p on C, let l p be the diametric
line of C through p and L p (x) be the chord perpendicular to l p at a distance x from
p. Then L p (x) divides A1 into two parts, say f 1 (x), g1 (x). If d is the diameter of
C, the intermediate value theorem 1.4 applied to h 1 (x) = f 1 (x) − g1 (x) on [0, d]
yields an x0 in [0, d] such that the chord L p = L p (x0 ) divides A1 into two equal
parts, say f 2 ( p), g2 ( p). Then the function h 2 ( p) = f 2 ( p) − g2 ( p) on C satisfies
h 2 ( p ∗ ) = −h 2 ( p). Apply Borsuk–Ulam 1.11 to h 2 to complete the proof. 

Exercise 1

1. Intervals. Intervals in R are of the following form:

Bounded intervals: closed intervals [a, b] (including one-point), open intervals
(a, b), half-open and half-closed intervals [a, b) and (a, b].
Unbounded intervals: [a, ∞), (a, ∞), (−∞, b], (−∞, b) and R = (−∞, ∞).
1.1 Intermediate Value Theorem 5

A1 lp

g1 (x)

f1 (x) x0

x A2

p g2 (p)
f2 (p)

Lp (x)
Lp (x0 )

Fig. 1.2 Dosa theorem

2. Intermediate values. a) If f and g are continuous on [a, b] and if f (a) > g(a)
and f (b) < g(b), then there is a c, a < c < b, such that f (c) = g(c).
b) The function f : [0, 1] → [−1, 1] defined by f (0) = 0 and f (x) = sin(1/x),
x
= 0, is not continuous, yet takes all values in [-1,1].
c) Find a, b such that there is a c, a < c < b, with f (c) = 0, where (i) f (x) =
x 3 − 7x + 23; (ii) f (x) = e−x − x 7 + 17x + sin x − 31.
d) Use the intermediate value theorem to prove that, for any a > 0, there is a
b > 0 such that b2 = a.
e) Apply the intermediate value theorem to conclude that x 4 + x − 3 has at least
two real roots.
f) Use your real analysis to prove Darboux’s theorem: If f is a real, differentiable
function on [0,1], then the derivative f  has the intermediate value property.
g) If I is an interval and f : I → R is continuous and injective, then f is mono-
tonic.
3. Fixed points. a) The identity function on [0,1] has every point as a fixed point.
Find continuous functions on [0,1] with (i) exactly 100 fixed points; (ii) countably
infinitely many fixed points; uncountably many fixed points, not the identity function.
b) Find continuous functions X → X without fixed points when X is (i) R; (ii)
[0, ∞); (iii) (0,1]; (iv) the unit circle S1 .
6 1 Apéritif : The Intermediate Value Theorem

4. Dosa theorem. a) If A is a bounded region in the plane, there are two perpen-
dicular lines dividing A into four equal parts.(See [2].)
b) Two dosas, one a perfect circle and the other a perfect square, lie on the table.
Describe a single cut dividing each into two equal parts.

1.2 Biographical Notes

1.2.1 Bolzano

Bernhardus Bolzano (1781–1848) worked in isolation in Prague, Bohemia (as the

region was called at that time). His work was not published in his life time, and it did
not get the attention it deserved. He was much ahead of his times. He made several
fundamental contributions to mathematical analysis which were recognised only
much later—the greatest lower bound property of real numbers, the ε-δ definition
of limit, the intermediate value theorem (using the glb property) and the Bolzano–
Weierstraß theorem (both rediscovered by Karl Weierstraß—written as Weierstrass
in English—decades later). He was among the first to initiate rigour in analysis, later
carried out by Weierstraß, Richard Dedekind and others.

References

1. Bolzano, B.: Rein analytischer Beweis des Lehrsatzes, dass zwischen je zwei Werthen, die
ein entgegengesetztes Resultat gewähren, wenigstens einer reelle Wurzel der Gleichung liege.
Prague, Gottlieb Haase (1817)
2. Chinn, W.G., Steenrod, N.E.: First Concepts of Topology. Mathematical Association of Amer-
ica (1966). https://doi.org/10.5948/UPO9780883859339
3. Shashkin, Y.A.: Fixed Points. American Mathematical Society-Mathematical Association of
America (1991)
Chapter 2
Metric Spaces

The classical ε-δ definition of continuity makes sense once there is a concept of
‘distance’. So, it is natural to begin a study of topology with a look at spaces with a
distance function: metric spaces. The French mathematician Maurice Fréchet defined
a metric in his thesis (1906) [1] thus:
... on peut faire correspondre à tout couple d’éléments A, B un nombre (A, B) ≥ 0
, que nous appellerons l’écart des deux éléments et qui jouit des deux propriété
suivantes: a) L’écart (A, B) n’est nul que si A et B sont identiques, b) Si A, B, C,
sont trois éléments quelconques, on a toujours (A, B) ≤ (A, C) + (C, B). (Fréchet,
1906, [1])
He used the term écart for the ‘distance’ (A, B) ≥ 0 between a pair of elements A
and B (of a set); (a) says (A, B) = 0 if A = B and (b) is the triangle inequality. The
current terminology comes from the German mathematician Hausdorff (see Chap. 3)
who used the term metrische Raumë (‘metric space’). These spaces were vigorously
studied by Hausdorff and the Polish school in the 1920s.

2.1 Metrics

Topology may be called a study of continuity. So let us begin by recalling the concept
of continuity of a function f : R → R; f is continuous at x0 ∈ R if f (x) is arbitrarily
close to f (x0 ) provided x is sufficiently close to x0 . The precise formulation of this
vague, but intuitively appealing, statement results in the famous ε-δ definition of
limits and of continuity. This is attributed to Bolzano (1817, [2]) and, independently,
to the great German mathematician Karl Weierstrass (1872). It is one of the most
fundamental concepts in mathematics.

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 7
K. Parthasarathy, Topology, La Matematica per il 3+2 134,
https://doi.org/10.1007/978-981-16-9484-4_2
8 2 Metric Spaces

Definition 2.1 A function f : R → R is said to be continuous at x0 ∈ R if, for every

ε > 0, there is a δ = δ(x0 , ε) > 0 such that | f (x) − f (x0 )| < ε when |x − x0 | < δ.
f is continuous on R if it is continuous at each x0 ∈ R.

Thus ε measures the closeness of f (x) to f (x0 ) that one is seeking and δ quantifies
the sufficient closeness of x to x0 that is required to achieve this. A rephrasing of this
local formulation is: f (x) is at a distance less than ε from f (x0 ) provided x is at a
distance less than δ from x0 . This makes sense not just on R, but wherever there is a
concept of distance. But what is a distance on a set X ? It clearly involves a pair of
points and is ‘naturally’ nonnegative: it is a map d : X × X → R+ = [0, ∞)! What
kind of a map? Intuition suggests that the distance between distinct points ‘must’ be
positive and distance ‘must’ be symmetric: distance from x to y equals that from y
to x. One also ‘knows’ that the sum of the lengths of two sides of a triangle is greater
than the length of the third side (‘triangle inequality’). Thus the following definition
is fair. It is also mathematically sound and adequate. (‘Metric’ is derived from the
Greek word for measurement.)
Definition 2.2 A metric on a (nonempty) set X is a function d : X × X → R having
the following properties:
i) d(x, x) = 0 and d(x, y) > 0, x, y ∈ X, y = x (Positivity);
ii) d(x, y) = d(y, x), x, y ∈ X (Symmetry);
iii) d(x, y) ≤ d(x, z) + d(z, y), x, y, z ∈ X (Triangle inequality.)
A set X , together with a metric d on it, is called a metric space.

Example 2.3 1. For any nonempty set X , d(x, x) = 0, d(x, y) = 1, x = y, defines

a metric on X , called the discrete metric. (Why 1, why not 1/2?)
2. On R, d(x, y) = |x − y| gives the usual distance. (Qn. What happens if we
take 2|x − y|? |x − y|1/2 ? |x − y| p , p > 0?)
3. For x ∈ (0, 1), let δ(x) = min{|x|, |x − 1|}. Define d(x, y) = |x − y|+
| δ(x)
1
− δ(y)
1
|. Then d is a metric on (0,1).
4.The standard or Euclidean metric on Rn is defined by d2 (x, y) =
(1n |xi − yi |2 )1/2 , x = (x1 , ..., xn ), y = (y1 , ..., yn ). We have metrics d p , 1 ≤ p ≤
∞, that reduce to the usual metric on R when n = 1. We only mention d2 , d1 and
d∞ : d1 (x, y) = 1n |xi − yi | and d∞ (x, y) = max |xi − yi |.
1≤i≤n
5. Fix a prime p. A rational number x ∈ Q, x = 0, can be written in the form
x = p n . ab with n, a, b ∈ Z, the set of integers, and n is uniquely determined by
the conditions a, b are not divisible by p. Define |x| p = p −n and |0| p = 0. The
p-adic metric on Q is given by d p (x, y) =| x − y | p . An interesting exercise for
the reader is the verification the triangle inequality. (In fact, the stronger inequality
|x + y| p ≤ max{|x| p , |y| p } holds.) This metric is an example of a nonarchimedean
metric.
6. Let (X, d) be a metric space and fix a P ∈ X . Define d P (x, y) = d(x, P) +
d(P, y) for x = y and d P (x, x) = 0, x, y ∈ X . Then d P is a metric on X , called the
French railways metric. (Fig. 2.1)
7. Any metric on X induces a metric on any subset E by restriction.
2.1 Metrics 9

x d(P,y)
d(x,P)
P

Fig. 2.1 French railway metric

8. If d is a metric on X , define d ∗ (x, y) = min{d(x, y), 1}, x, y ∈ X . Then d ∗ is

again a metric on X (check) and satisfies d ∗ (x, y) ≤ 1 for all x, y ∈ X . It is called
the bounded metric associated with the metric d.
9. A set E in a metric space Y is bounded if its diameter, diam(E) := supx,y∈E
d(x, y), is finite. Let X be a set and (Y, d) a metric space. Let B(X, Y ) be the set of
bounded maps f from X to Y (i.e. f (X ) is a bounded set in Y ). Then dsup ( f, g) =
supx∈X d( f (x), g(x)) defines a metric on B(X, Y ). When Y = R or C, we will just
write B(X ). In particular, ∞ , the set of bounded sequences of complex (or real)
numbers is a metric space with metric d∞ (x, y) = supn |xn − yn |.
10. The set 1 of all absolutely summable sequences (xn ) of real or complex
numbers (i.e. sequences (xn ) with |xn | < ∞) is a metric space with metric defined
by d1 (x, y) = |xn − yn |.
11. Let {(X n , dn )} be a countable family of metric spaces. Then a metric can be
defined on X = n X n as follows. Define, for x = (xn ), y = (yn ) ∈ X , d (x, y) =
sup n1 dn∗ (xn , yn ), where dn∗ (xn , yn ) = min(dn (xn , yn ), 1). It is easy to see that d is a
metric on X . The triangle inequality follows on taking the supremum in the inequality
dn∗ (xn , yn ) ≤ dn∗ (xn , z n ) + dn∗ (z n , yn )
after multiplying by n1 . (Qn. Is the factor n1 needed in the definition?)
12. On the space C[0, 1] of continuous functions on [0,1], we have the sup metric
1
dsup ( f, g) = sup0≤x≤1 | f (x) − g(x)| and the integral metric d1 ( f, g) = 0 | f (x) −
g(x)|d x. (How do you get d1 ( f, g) = 0 implies f = g? You need some real analysis.
1
If f is a nonnegative continuous function on [0,1], when is 0 f (x)d x = 0?)
Most of the examples are easy to verify. One nontrivial verification is the triangle
inequality for d2 in Rn . It is a consequence of the following important inequality
due to Cauchy (1821). It is stated here for real numbers, but then it clearly holds for
complex numbers as well.
Proposition 2.4 (Cauchy–Schwarz inequality) 1n |xi yi | ≤ (1n |xi |2 )1/2 (1n |yi |2 )1/2
for xi , yi ∈ R.
Proof We may assume xi , yi ≥ 0. The inequality holds trivially if yi = 0 for all i.
If not, consider the inequality

0 ≤ (xi − λyi )2 = xi2 − 2λxi yi + λ2 yi2

which holds for all real numbers λ. For λ = (xi yi )/(yi2 ), this gives
10 2 Metric Spaces

0 ≤ (xi2 )(yi2 ) − (xi yi )2

yielding the desired inequality on taking the last term to the left side and then taking
square roots. 

It will be convenient to use the Euclidean norm x 2 = d2 (x, 0) for x ∈ Rn . Note that
d2 (x, y) = x − y 2 . Similarly, the norms x 1 := |x j | and x ∞ := max |x j |
give rise to the metrics d1 , d∞ .
Corollary 2.5 (Triangle inequality)
x − y 2 ≤ x 2 + y 2 ,
d2 (x, y) ≤ d2 (x, z) + d2 (z, y) for x, y, z ∈ Rn .

Proof As a consequence of the Cauchy–Schwarz inequality

|xi − yi |2 = xi2 − 2xi yi + yi2
1 1
≤ xi2 + 2(xi2 ) 2 (yi2 ) 2 + yi2
1 1
= {(xi2 ) 2 + (yi2 ) 2 }2 .
Thus x − y 2 ≤ x 2 + y 2 and so x − y 2 ≤ x − z 2 + z − y 2 . 

Exercise. Find conditions for equality in the Cauchy–Schwarz and triangle

inequalities. What does equality in the triangle inequality mean?

2.2 Continuity and Open Sets

It is time for us to get back to continuity. The ε-δ definition of continuity can now
be carried over to maps between metric spaces.

Definition 2.6 If (X, d), (X

, d
) are metric spaces, f : X → X
is said to be con-
tinuous at x0 ∈ X if, for every ε > 0, there is a δ > 0 such that d
( f (x), f (x0 )) < ε
for all x ∈ X with d(x, x0 ) < δ (Fig. 2.2). The map f is continuous on X if it is
continuous at each x0 ∈ X .

Proposition 2.7 Let (X, d) be a metric space. For any fixed p ∈ X , the function
x → d(x, p), X → R, is uniformly continuous and d : X × X → R is continuous.

ε
δ f (x0 )
f
x0

Fig. 2.2 Continuity at x0

2.2 Continuity and Open Sets 11

Proof The second statement means that d(x, y) is close to d( p, q) if x is close to p

and y is close to q. The results follow from the inequalities
|d(x, p) − d(y, p)| ≤ d(x, y) and
|d(x, y) − d( p, q)| ≤ |d(x, y) − d(x, q)| + |d(x, q) − d( p, q)|
≤ d(x, p) + d(y, q). 

Continuity can be rephrased in terms of balls; in metric spaces, balls play a role
similar to that played by intervals in R.
Definition 2.8 Let (X, d) be a metric space. For x ∈ X and r > 0, write B(x; r ) =
{y ∈ X : d(y, x) < r }. It is called the open ball of radius r centred at x. When
necessary, we write B X (x; r ) or Bd (x, r ) for this ball. B(x; ε) is also referred to
as the ε-neighbourhood of x. The closed ball is defined by B̄(x; r ) = {y ∈ X :
d(y, x) ≤ r }. E ⊂ X is called a neighbourhood of x ∈ X if B(x, ε) ⊂ E for some
ε > 0 (Figs. 2.3, 2.4 and 2.5).
Thus, in terms of balls, the defining condition for continuity of a map f : X → X

at x0 ∈ X is equivalent to each one of the following:

d∞
d2

d1

Fig. 2.3 Balls in R2

Z3

Z2 Q=(1,0)
P=(0,0)

Z1

Fig. 2.4 Bd P (Q, 2) in R2

12 2 Metric Spaces

d1

d2

d∞

Fig. 2.5 Balls in R3 in metrics d1 , d2 and d∞

i) x ∈ B X (x0 ; δ) ⇒ f (x) ∈ B X
( f (x0 ); ε);
ii) B X (x0 ; δ) ⊂ f −1 (B X
( f (x0 ); ε));
iii) f (B X (x0 ; δ)) ⊂ B X
( f (x0 ); ε).
Example 2.9 1. In R, a ball is an interval: B(x; ε) = (x − ε, x + ε) and B̄(x; ε) =
[x − ε, x + ε].
2. In (R2 , d2 ), B(x; ε) is the open disc with centre x and radius ε.
3. In (R2 , d1 ), B(x; ε) is the open square centred at x with diagonals of lengths
2ε parallel to the axes.
4. In (R2 , d∞ ), B(x; ε) = (x − ε, x + ε) × (x − ε, x + ε) is the open square with
centre x and of sides of length 2ε parallel to the axes.
5. If X has the discrete metric, then B(x, r ) = {x} if r < 1 and B(x, r ) = X if
r ≥ 1.
6. If E ⊂ X , then B E (x; ε) = B X (x; ε) ∩ E.
Next we turn to formulation of continuity in terms of convergence. Sequential
convergence in metric spaces is defined exactly as in R.
Definition 2.10 A sequence {xn } in a metric space (X, d) is said to converge to
x ∈ X , written xn → x, if the sequence {d(xn , x)} of real numbers converges to
2.2 Continuity and Open Sets 13

zero; i.e. if for every ε > 0 there is a positive integer n ε such that d(xn , x) < ε for all
n ≥ n ε . This is expressed by saying that the xn eventually lie in any neighbourhood
of x.
Again, as in the case of the real line, if a sequence in a metric space has a limit,
it is unique.
Proposition 2.11 Let X be a metric space and let {xn } be a sequence in X . If xn → x
and xn → y, then x = y.
Proof If x = y, then ε = d(x, y) > 0 and B(x; ε/2) and B(y; ε/2) are disjoint, so
the xn cannot lie eventually in both. 
Example 2.12 1. Convergence in Rn with respect to any of the metrics d1 , d2 , d∞
is equivalent to coordinatewise convergence: x (k) = (x1(k) , ..., xn(k) ) converges to x =
(x1 , ..., xn ) if and only if x (k)
j → x j for all j.
2. In the discrete metric d on a set X , the only convergent sequences are the
eventually constant ones: if xn → x then xn = x for all n ≥ n 0 , for some positive
integer n 0 and conversely.
Proposition 2.13 Let X, X
be metric spaces and let x0 ∈ X . For a map f : X →
X
, the following conditions are equivalent:
1) f is continuous at x0 ;
2) the sequence { f (xn )} converges to f (x0 ) in X
for any sequence {xn } in X
converging to x0 , i.e. d(xn , x0 ) → 0 ⇒ d
( f (xn ), f (x0 )) → 0.
Proof Assume f is continuous at x0 . Let xn → x0 and ε > 0. By continuity, choose
δ > 0 such that d
( f (xn ), f (x0 )) < ε when d(xn , x0 ) < δ and then by convergence
choose n δ such that d(xn , x0 ) < δ for all n ≥ n δ . Then d
( f (xn ), f (x0 )) < ε for
n ≥ n δ , so f (xn ) → f (x0 ).
Conversely, suppose f is not continuous at x0 . This means there is an ε > 0
for which no δ ‘works’; that is, for every δ > 0 there is an x with d(x, x0 ) < δ,
but d
( f (x), f (x0 )) ≥ ε. (The reader should understand this clearly and care-
fully. Forming negations of statements is an indispensable skill in mathematics.)
Thus, for each positive integer n, there is an xn ∈ X such that d(xn , x0 ) < n1 but
d
( f (xn ), f (x0 )) ≥ ε. But then xn → x0 in X , whereas the f (xn ) lie away from
f (x0 ) and so cannot converge to f (x0 ) in X
. This completes the proof. 
Remark 2.14 The statement ‘if P is true, then Q is true’ is equivalent to the statement
‘if Q is false, then P is false’. The latter statement is called the contrapositive of the
former. Using contrapositive is an important and frequently employed strategy in
mathematical proofs.
Example 2.15 1. Let X be a metric space and let f : X → Rn . Write f (x) =
( f 1 (x), ..., f n (x)), x ∈ X . Then f is continuous at a ∈ X if and only if each coordi-
nate function f j : X → R is continuous at a.
2. For E ⊂ X , f (x) = d(x, E) := inf y∈E d(x, y) is continuous on X .
3. Any map from a discrete metric space X to any Y is continuous.
14 2 Metric Spaces

ε
y
d(x,y) r
x

Fig. 2.6 Open balls are open

So far we have been looking at local formulations of continuity. A global formu-

lation needs the concept of open sets.
Definition 2.16 A subset V of a metric space X is called an open set if it is a
neighbourhood of each of its points; i.e. for each x ∈ V there is an εx > 0 such that
B(x; εx ) ⊂ V .

Proposition 2.17 In a metric space X , the whole set X , the empty set and any ball
B(x; r ) for x ∈ X and r > 0 are open.

Proof The definition applies trivially to X and vacuously to the empty set. To prove
that B(x; r ) is open, let y ∈ B(x; r ). To find an ε > 0 such that B(y; ε) ⊂ B(x; r ).
Take ε ≤ r − d(x, y). If d(z, y) < ε then

d(x, z) ≤ d(x, y) + d(y, z) < d(x, y) + ε ≤ r

so that z ∈ B(x; r ). Thus B(y; ε) ⊂ B(x; r ). (Fig. 2.6.) 

Example 2.18 1. Let a, b ∈ R, a < b. The intervals (a, b), (a, ∞), (−∞, a) are all
open sets in R with the usual metric. Note that (a, b) = (c − r, c + r ) = B(c; r ), c =
a+b
2
and r = b−a 2
. The reader should check the assertion for the other two intervals.
Other than R itself, these are the only intervals, bounded or unbounded, that are open
sets in R. Thus, the only intervals that are open sets are the ‘open’ intervals, bounded
or unbounded.
2. The half planes {(x, y) ∈ R2 : x < a}, {(x, y) ∈ R2 : y < a} and similarly
{(x, y) ∈ R2 : x > a}, {(x, y) ∈ R2 : y > a} are all open in R2 with any of the met-
rics d1 , d2 or d∞ . So are the vertical and horizontal strips {(x, y) ∈ R2 : a < x < b}
and {(x, y) ∈ R2 : c < y < d}.

A global characterisation of continuity is in terms of open sets.

Theorem 2.19 A map f : X → Y from one metric space to another is continuous

if and only if f −1 (V ) is open in X for every open set V in Y , i.e. f pulls back open,
sets in Y to open sets in X .
2.2 Continuity and Open Sets 15

Proof If f is continuous and V is open in Y , let x0 ∈ f −1 (V ). Then f (x0 ) ∈ V and,

as V is open, BY ( f (x0 ); ε) ⊂ V for an ε > 0. Now the continuity of f at x0 yields
f (B X (x0 ; δ)) ⊂ BY ( f (x0 ); ε) ⊂ V for a δ > 0. Thus B X (x0 ; δ) ⊂ f −1 (V ) and one
part is proved.
Conversely, if f pulls back open sets to open sets, we need to prove that f
is continuous at each x0 ∈ X . If ε > 0, V = BY ( f (x0 ); ε) is open in Y and so
f −1 (V ) is open in X . Thus it contains a ball around x0 , say B X (x0 ; δ) ⊂ f −1 (V ) =
f −1 (BY ( f (x0 ); ε)) giving the continuity at x0 . 

Remark 2.20 Continuity at a point is characterised in terms of open sets thus: f

is continuous at a point x0 if and only if the pullback of every open set containing
f (x0 ) is an open set containing x0 .

The class of open sets in a metric space has the following very basic properties
with respect to unions and intersections.

Proposition 2.21 Let X be a metric space. Then

1) an arbitrary union of open sets in X is open in X ,
2) any finite intersection of open sets in X is open in X .

Proof 1) Let {Vi } be open and x ∈ V := i Vi . Then x ∈ Vi for some i and Vi is
open, so there is an ε > 0 such that B(x; ε) ⊂ Vi ⊂ V .
2) Let {U j }n1 be open and x ∈ U := U j . Then x ∈ U j , hence there is an ε j > 0
with B(x; ε j ) ⊂ U j for each j. If ε = min ε j , then B(x; ε) ⊂ B(x; ε j ) ⊂ U j for
1≤ j≤n
all j and so B(x; ε) ⊂ U . 

A consequence is that all open sets can be built up from open balls.
Corollary 2.22 A subset V of a metric space X is open if and only if it is a union of
open balls.

is open, then for each x ∈ V there is an εx > 0 such that B(x; εx ) ⊂ V .

Proof If V 
Then V = {B(x; εx ) : x ∈ V } and so V is a union of open balls. The converse is
a consequence of the facts that open balls are open sets and unions of open sets are
open. 

Example 2.23 Infinite intersections of open sets are not generally open. For instance,
look at the sequence of open intervals Un = {(− n1 , n1 )}∞
1 in R. Then ∩n Un = {0}
which certainly is not open.

Open sets in R are unions of open intervals, by Corollary 2.21. But an impressively
stronger result holds.

Theorem 2.24 Every nonempty open set in R is a countable, disjoint union of open
intervals.
16 2 Metric Spaces

Proof Let V ⊂ R be open. For each x ∈ V , let Ix be the union of all open intervals
containing x and contained in V :

Ix = {I : I is an open interval and x ∈ I ⊂ V }.

Note that the collection on the right side is nonempty since V is open. Moreover,
Ix is an interval (being a union of intervals with a common point), is open and is
a subset of V . This ensures that if z ∈ Ix , then Ix ⊂ Iz ⊂ Ix so that Iz = Ix . Thus,
if x, y ∈ V and if z ∈ Ix ∩ I y , then Ix = Iz = I y . In other words, if x, y ∈ V , then
Ix , I y are either disjoint or equal. Clearly, the union of the Ix , x ∈ V, is V itself. It
remains to prove countability. For this we use rationals (what else!). Since every open
interval contains rationals, there is a rational number qx ∈ Ix for x ∈ V . If Ix = I y ,
then Ix and I y are disjoint and so qx = q y . Thus Ix → qx is one-one map of the set
of distinct Ix into Q. Hence this collection is countable and the proof is complete.

The property of Q that we have used (‘every open interval contains a q ∈ Q’)
is shared by other countable
√ sets too, e.g. the set of dyadic rationals (those of the
form a/2k ) and {m + n 2 : m, n ∈ Z} (Proof?). Such sets, ‘dense sets’, would be
discussed later in a general setting.
As we observed earlier, convergence in all the metrics d1 , d2 , d∞ on Rn are the
same (= coordinate-wise convergence). This reflects the interesting fact that all of
them yield the same class of open sets.

Lemma 2.25 For x, y ∈ Rn √ , the following inequalities hold:

1) d2 (x, y) ≤ d1 (x, y) ≤ √n d2 (x, y);
2) d∞ (x, y) ≤ d2 (x, y) ≤ n d∞ (x, y).

Proof 1) (x j − y j )2 ≤ (|x j − y j |)2 ,

1 1 √ 1
|x j − y j | ≤ (|x j − y j |2 ) 2 (12 ) 2 = n (|x j − y j |2 ) 2 ,
the first being trivial and the second a consequence of Cauchy–Schwarz.
2) Both are straightforward and are left to the reader. 

Proposition 2.26 A subset V of Rn is d1 -open if and only if it is d2 -open if and only

if it is d∞ -open.

Proof The equivalences are consequences of (1) and (2) of the previous lemma,
respectively. The crux of the matter is that each di -ball centred at any point x contains
√ If V is d2 -open and x ∈ V ,
a d j -ball centred at x for all i, j. Here is a sample argument.
then Bd2 (x; ε) ⊂ V for some ε > 0. But d2 (x, y) ≤ n d∞ (x, y), so Bd∞ (x; √εn ) ⊂
Bd2 (x; ε) ⊂ V and V is d∞ -open. The proofs for the remaining parts are on similar
lines, and the reader should have no difficulty with them. 

There is a convenient terminology that is used in such situations.

Definition 2.27 Two metrics d, d
on a set X are said to be equivalent if each d-ball
centred at any x contains a d
-ball centred at x and conversely. This is equivalent
to saying that the class of d-open sets is the same as the class of d
-open sets, or,
2.2 Continuity and Open Sets 17

Fig. 2.7 Equivalence of d1 , d2 and d∞ on R2

equivalently, that the identity map id X : (X, d) → (X, d

) and its inverse are both
continuous.
On Rn , d1 , d2 and d∞ are pair-wise equivalent (Fig. 2.7).

Exercise 2

1. The space 2 . Let 2 = {x = (xn ) : x||2 = |xn |2 < ∞}. d2 , defined as d(x, y)
= x − y , just as in Rn .
2. Metrics. a) On (0,1), d(x, y) =| x1 − 1y | defines a metric.
b) If d is a metric on X , then the associated bounded metric d ∗ and d  (x, y) =
d(x,y)
1+d(x,y)
are both metrics and are equivalent to d. More generally, d f (x, y) =
f (d(x, y)) gives a metric for any nondecreasing f : [0, ∞) → [0, ∞) satisfying
f (x + y) ≤ f (x) +  f (y) and f (x) = 0 if and only if x = 0.
c) A finite sum ak dk is a metric if each dk is and ak > 0.
d) If f is a continuous real function a metric space (X, d), then d f (x, y) =
d(x, y) + | f (x) − f (y)| is a compatible metric.
1
e) d1 ( f, g) = 0 | f (t) − g(t) | dt, dsup ( f, g) = sup0≤x≤1 | f (x) − g(x)| both
give metrics on C[0, 1]; they are not equivalent.
f) Take the closed unit disc with the usual metric. Write down explicitly the French
railway metric (2.3) taking P as the origin.
g) For 0 < p < 1, let  p be the space of all real sequences (xn ) with |xn | p < ∞.
d((xn ), (yn )) = |xn − yn | p gives a metric on this space.
3. Unions of intervals. a) The union of a collection of intervals in R having a
common point is an interval. (An alert reader might have noticed that this fact was
used in the proof of Theorem 2.24.)
b) The decomposition in 2.24 is unique.
4. Convergence. Discuss convergence of sequences in the French railway metric
and in a product of metric spaces.
5. Isometries. A map f of a metric space X to itself is called an isometry if
d( f (x), f (y)) = d(x, y) for all x, y ∈ X .
a) Translations Ta (x) = x + a, a ∈ Rn , and orthogonal transformations, ρ A (x) =
Ax, A an orthogonal matrix, are isometries on Rn .
18 2 Metric Spaces

b) Surjective isometries on a metric space form a group under composition of

maps. (An isometry of Rn is called a rigid motion. The group Mn of such rigid
motions is generated by (the subgroup of) translations and (the subgroup of) rota-
tions.)
6. Semi-metrics. If, from the definition of a metric, the condition d(x, y) = 0
implies x = y is dropped, we get what is called a pseudometric or a semi-metric.
Define balls and open sets using a semi-metric just the way they were defined for
metric.
a) The following are semi-metrics: (i) On R2 , d(x, y) = |x1 − y1 | where x =
(x1 , x2 ), y = (y1 , y2 ); (ii) on C[0, 1], d( f, g) = | f (0) − g(0)|; (iii) d1 is defined as
in Ex. 2 d) above, but now on the space of Riemann integrable functions on [0,1].
b) Verify: arbitrary unions and finite intersections of open sets defined by a semi-
metric are open.
c) If f is a real-valued function on a set X , d f (x, y) = | f (x) − f (y)| defines a
semi-metric on X . When is it a metric? If X has a metric d, then f is d-continuous
if and only if every d f -open set is d-open.
d) This exercise shows how a semi-metric on a set gives rise, in a natural way,
to a metric (on a ‘quotient set’). Let d be a semi-metric on X and define x Ry if
d(x, y) = 0. R is an equivalence relation. Let X/R denote the set of equivalence
classes. Denoting the equivalence class of x by x̃, let d̃(x̃, ỹ) = d(x, y). Then d̃ is a
well-defined metric on X/R.
e) Identify the metric space associated with each semi-metric space given above.
f) Let c be the space of all convergent real sequences. On c, define d((xn ), (yn )) =
lim |xn − yn |. Then d is a semi-metric and the associated metric space is isometric
to R.
7. Products. Let (X 1 , d1 ), (X 2 , d2 ) be metric spaces. On X 1 × X 2
ρ1 ((x1 , x2 ), (y1 , y2 )) = d1 (x1 , y1 ) + d2 (x2 , y2 ),
ρ2 ((x1 , x2 ), (y1 , y2 )) = [d1 (x1 , y1 )2 + d2 (x2 , y2 )2 ]1/2 and
ρ∞ ((x1 , x2 ), (y1 , y2 )) = max(d1 (x1 , y1 ), d2 (x2 , y2 ))
all yield metrics. Are they equivalent?
8. Distance between sets. a) d : (X × X, d1 ) → R is continuous for any metric
d on X.
b) For subsets A, B of X , define d(A, B) = supx∈A d(x, B). Is d(A, B) =
d(B, A)? Does the inequality d(A, B) ≤ d(A, C) + d(C, B) for subsets A, B and
C of X hold? When is d(A, B) = 0?
9. Matrix space. On the set M(n, R) of all real n × n matrices, each of the
following defines a metric:
a) for X = (xi j ), Y = (yi j ) let d∞ (X, Y ) = maxi, j | xi j − yi j |;
b) d H S (X, Y ) = [trace(X t − Y t )(X − Y )]1/2 . (Trace is the usual matrix trace;
X t denotes transpose of X . HS stands for Hilbert–Schmidt.)
10. Open sets in the plane. Is it true that every open set in R2 is a disjoint union
of open rectangles (a, b) × (c, d)?
11. Metric from a norm. a) We have seen norms . p , p = 1, 2, ∞, on Rn .
Here is the general definition. A norm on a (real or complex) vector space X is a
real-valued function x → x on X such that
2.3 Biographical Notes 19

i) x > 0 for all x = 0, 0 = 0;

ii) x + y ≤ x + y , x, y ∈ X ;
iii) αx = |α| x , for x ∈ X and scalar α.
The function d defined by d(x, y) = x − y is a metric on X . Identify the metrics
we have encountered so far which come from norms.
b) If a metric d satisfies the conditions d(x + z, y + z) = d(x, y) for all x, y, z
and d(ax, ay) = |a|d(x, y) then does d come from a norm?
12. p-adic metric. The p-adic metric on Q, defined in 2.3, satisfies:
a) |x| p ≤ 1 implies |1 + x| p ≤ 1 and this implies |x + y| p ≤ max(|x| p , |y| p );
b) every point in a ball Bd p (x, r ) is a centre, i.e. if y ∈ Bd p (x, r ), then Bd p (y, r ) =
Bd p (x, r );

c) p |x| p = |x|1
, the product being over all primes p.
d) Compute |x| p for x = 1, 2, 3, 10−3 for p = 2, 3, 19.

2.3 Biographical Notes

2.3.1 Fréchet

Maurice Fréchet (1878–1973) was a well-known French mathematician, known

for his work in topology, functional analysis and probability. He was a student of
Jacques Hadamard and in his thesis (1906) introduced the concept of a metric. The
term ‘compact’ became common currency after he used it in his thesis (for what we
call limit point compact now). Fréchet space, Fréchet derivative, Fréchet distribution
are some of the many concepts named after him.

References

1. Fréchet, M.: Sur quelques point du calcul fonctionnel. Rendiconti di Palermo 22, 1–74 (1906).
https://doi.org/10.1007/BF03018603
2. Bolzano, B.: Rein analytischer Beweis des Lehrsatzes, dass zwischen je zwei Werthen, die
ein entgegengesetztes Resultat gewähren, wenigstens einer reelle Wurzel der Gleichung liege.
Prague, Gottlieb Haase (1817)
Chapter 3
Topological Spaces

We have seen that continuity in metric spaces can be expressed wholly in terms of
open sets and that the class of open sets satisfies some simple set theoretic properties.
This led to an abstract study of a set and classes of subsets with these properties so that
continuity can be defined in this setting, giving rise to general topological spaces. It
is generally accepted that Riemann was the originator of topology. He, in his work on
function theory, algebraic functions and in his famous inaugural lecture on geometry
(Über die Hypothesen welche der Geometrie zu Grunde liegen, [1]) had expressed
ideas on formulating the concept of topological spaces, subspaces of such spaces and
a general theory of such spaces. These may be considered as precursors of modern
topology.

3.1 Topologies and Open Sets

The genesis of topology in the seminal work of Riemann was followed by a huge
body of work of several great mathematicians. Cantor’s set theory and theory of
derived sets, Poincaré’s epochal work on Analysis situs, Fréchet’s metric spaces,
F.Riesz’s study of abstract spaces using a version of limit points and Weyl’s study [2]
of abstract spaces in terms of neighbourhood systems finally culminated in the work
of Hausdorff. Hausdorff’s work ushered in what is now called general topology. Here
is the definition of a topological space from Hausdorff [3]:
Unter einem topologischen Raum verstenhenwir eine Menge E, worin den Ele-
menten (Punkten) x gewisse Teimengen Ux zugeordnet sind, die wir Umgebungen
von x nennen, und zwar nach Massgabe der folgenden Umgebungsaxiome ( A topo-
logical space is a set E together with sets Ux assigned to each x ∈ E satisfying the
following neighbourhood axioms):

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 21
K. Parthasarathy, Topology, La Matematica per il 3+2 134,
https://doi.org/10.1007/978-981-16-9484-4_3
22 3 Topological Spaces

(A) Jedem Punkt x entspricht mindestens eine Umgebung Ux ; jede Umgebung

Ux enthält den Punkt x. (For each x there is at least one neighbourhood that contains
x.)
(B) Sind Ux , Vx zwei Umgebungen desselben Punktes so gibt es eine Umgebung
Wx dei Teilmenge von beiden ist (Wx ⊆ D(Ux , Vx )). (If Ux , Vx are neighbourhoods
of the same x there is a Wx ⊂ Ux ∩ Vx .)
(C) Liegt der Punkt y in Ux , so gibt es eine Umgebungen U y , die Teilmenge von
Ux ist (U y ⊆ Ux ). (If y ∈ Ux , there is a neighbourhood U y ⊂ Ux .)
(D) Für zwei verschiedene Punkt x, y gibt es zwei Umgebungen Ux , U y ohne
gemeinsamen Punkt (D(Ux , U y ) = 0). (For two distinct points x, y there correspond
disjoint neighbourhoods Ux , U y ) (Hausdorff separation axiom to be discussed in a
later chapter.)
The current ‘open set axioms’ were proposed by Tietze in 1923 [4] and adopted
by Alexandroff [5] in 1925.
(A◦ ) Jeder Punkt x (des betrachteten topologischen Raümes R) ist in mindestens
einer offenen Menge G enthalten. (Every point of the space R is in some open set
G.)
(B ◦ ) Haben zwei offenen Mengen G1 , G2 weningstens einen Punkt gemein, so ist
auch ihr Durscschnitt Menge der gemeinsamen Punkte) G1 .G2 eine offene Menge.
(The intersection G 1 .G 2 of two (nondisjoint) open sets G 1 , G 2 is an open set.)
(C ◦ ) Gibt es zu jedem in der Menge A enthaltenen Punkt x eine offene Menge
die x einthält und Teilmenge von A ist, so ist A selbst eine offene Menge. (A set A
is open if every point of A is in an open subset of A.)
(D ◦ ) Für zwei verschiedene Punkte x, y gibt es zwei ottene Mengen ohne gemein-
samen Punkt, yon denen die eine x, die andere y enthält. (For two distinct points x
and y, there are open sets without a common point, one of which contains x and the
other y.)
The modern definition does not include the last, the Hausdorff axiom. Hausdorff
himself gives the following axioms in [6], the ones used today:
(1) Der Raum E und die Nullmenge 0 sind offen. (The whole set E and the empty
set are open.)
(2) Der Durchschnitt von zwei offenen Mengen ist offen. (The intersection of two
open sets is open.)
(3) Die Summe von beliebig vielen offenen Mengen ist offen. (An arbitrary union
of open sets is open.)
The term topology was introduced by J. B.Listing (a student of Gauss) in 1847,
but the term became common currency only after Fréchet used it in his thesis in
1906. The subject was called Analysis situs earlier. Topology is a study of continuity
and we have seen that continuity can be formulated purely in terms of open sets. The
modern approach to topology takes open sets as the fundamental objects.
Definition 3.1 A topology on a nonempty set X is a collection T of subsets of X
with the following properties:
a) X, ∅ ∈ T ;
b) if {Vi } is any collection of sets in T , then ∪i Vi ∈ T ;
3.1 Topologies and Open Sets 23

c) if U1 , ..., Un are finitely many members of T , then ∩n1 U j ∈ T .

The set X together with a topology T on X is called a topological space. Members
of T are referred to as open sets. So, the conditions demand that arbitrary unions and
finite intersection of open sets are open, besides requiring that the empty set and the
whole set are open.

Example 3.2 1. The class of open sets in a semi-metric or a metric space X form a
topology on X , the topology induced by the semi-metric or metric. This has been the
motivation for our definition of a topology. A topology that is induced by a metric is
called a metric topology. Such a topological space is said to be metrisable.
2. A trivial and hardly useful example is the indiscrete topology on a nonempty
set X consisting of just the whole set and the empty set.
3. A trivial looking, but pretty useful, topology is the discrete topology on X
consisting of all subsets of X .
4. The topology on Rn given by the metric d2 (or, equivalently, by d1 or d∞ ) is
called the Euclidean or the usual topology.
5. If T is a topology on X , for any E ⊂ X , the collection of sets T E := {E ∩ V :
V ∈ T } is a topology on E known as the relative topology. E with this topology is
referred to as a subspace of X . For example, [0,1], [0,1), (0,1], (a, b), 0 < a < b < 1,
are open in [0,1].
One can list a number of examples whose utility is primarily confined to providing
examples and counter examples in specific situations involving various topological
properties. Here are a few of them.
6. On any X, Tco f consists of the empty set and any set V ⊂ X whose complement
V c is finite. It is a topology known as the cofinite topology. The cocountable topology
Tcoc is defined analogously, replacing ‘finite’ by ‘countable’. The cofinite topology
is nondiscrete when X is infinite and the cocountable topology is discrete unless X
is uncountable. Do you get a topology if ‘countable’ is replaced by ‘infinite’?
7. The set X = {0, 1} with the topology {X, ∅, {1}} is called the Sierpiński space S.
It will be useful for counter examples. {X, ∅, {0}} is ‘essentially the same’ topology.
The discrete and the indiscrete topologies are the only other topologies on X .
8. If E is a nonempty, proper subset of a set X , then {X, ∅, E} is a topology on X
and so is {X, ∅, E, E c }.
9. The collection {V ⊂ X : V = ∅ or x0 ∈ V }, for any fixed point x0 ∈ X , is a
topology on X .
10. Here is an interesting topology on Z. For a, b ∈ Z, b > 0, the set A(a, b) :=
{a + nb : n ∈ Z} is a doubly infinite arithmetic progression around a. Define a subset
V of Z to be open if it is empty or satisfies the following property: for every a ∈ V
there is a positive integer b such that A(a, b) ⊂ V . The reader is urged to carry out the
simple verification that this actually gives a nondiscrete topology Tap , the arithmetic
progression topology, on Z. (For a slightly different connected topology (a term to
be defined in Chap. 9) on Z, see [7].)

We now look at some ways to generate topologies. A straightforward and trivial

verification gives the following.
24 3 Topological Spaces

Lemma 3.3 The intersection of any nonempty collection of topologies on a nonempty

set X is again a topology on X .

Here is a very basic and important observation.

Proposition 3.4 Let X be a nonempty set and S be a family of subsets of X . Then
there is a smallest topology T S on X containing S. It is called the topology generated
by S.

Proof Just take T S to be the intersection of all topologies on X containing S. Note

that this collection is nonempty as it contains the discrete topology. It is clear that
T S has the stated properties. 

Corollary3.5 If {Ti } is a collection of topologies on a set X , then there is a smallest

topology i Ti on X that contains all the Ti .
 
Proof i Ti is the topology generated by i Ti . 

Sometimes it is useful to compare different topologies on the same set. The fol-
lowing definition will facilitate handling such situations.

Definition 3.6 If T1 and T2 are both topologies on a set X , we say that T1 is finer
or stronger than T2 if T1 ⊃ T2 ; in this case we also say that T2 is coarser or weaker
than T1 .

Example 3.7 On any set, the discrete topology is the finest and the indiscrete topol-
ogy is the weakest topology. The cofinite topology is weaker than the cocountable
topology. On the real line, the usual topology is stronger than the cofinite topology,
but is not comparable with the cocountable topology (i.e. neither is finer than the
other).

Remark 3.8 Thus the family of all topologies on a nonempty set is a partially ordered
set (with inclusion as the order relation) in which any two members have a least upper
bound (lub or sup) and a greatest lower bound (glb or inf ). Such a partially ordered
set is called a lattice. The lattice of all topologies on a set has a smallest element (the
indiscrete topology) and a largest element (the discrete topology).

To conclude the section, we observe that any subset of a topological space contains
a largest open subset.

Definition 3.9 Let X be a topological space and let E ⊂ X . The interior of E,

denoted by int E, is the union of all subsets of E which are open in X . The points
of int E are called the interior points of E.

The next result summarises the properties of the interior, all of which are almost
obvious from the definition.
3.1 Topologies and Open Sets 25

Proposition 3.10 Let E be a subset of a space X . Then

1) int E is the largest open subset of E;
2)x ∈ E is an interior point of E if and only if there is an open set V in X with
x ∈ V ⊂ E;
3) E is open if and only if E = int E.

Proof Being a union of open sets, int E is open and (3) is a consequence of (1). All
the other statements are immediate. 

Example 3.11 1. If X has the discrete topology, int E = E for E ⊂ X ; if X has the
indiscrete topology, then int E = ∅ if E = X .
2. In R, int Q = ∅ = int (R \ Q) and int [0,1] = (0,1) = int [0,1]∪Z.
3. In the space [0,1] (that is, [0,1] with the relative topology as a subspace of R),
int [0,1] = [0,1], int [0,1) = [0,1) and int (0,1) = (0,1).
4. {(x, 0) : x ∈ R}, {(x, y) : x ∈ Q, y ∈ R} and {(x, y) : y = x 2 } all have empty
interiors in R2 ; the interior of the set {(x, y) : y ≥ 0} is {(x, y) : y > 0}, {(x, y) :
x 2 + y 2 ≤ 1} has interior {(x, y) : x 2 + y 2 < 1}.
5. Consider X = {1, 2, 3} with topology T = {X, ∅, {1}, {2, 3}}. Then int {1} =
{1} = int {1, 2} = int {1, 3} and int {2} = ∅ = int {3}.

It will be convenient to call a set E, not necessarily open, a neighbourhood of

x ∈ X if there is an open set U with x ∈ U ⊂ E. Thus U is open if and only if it is
a neighbourhood of each of its points.
The following result lists properties which are all almost obvious for the family of
neighbourhoods at any point. The point of listing them here is that, these properties
determine the neighbourhood system at each point of a topological space. That is,
families of sets having these properties can be used to define a topology on the set for
which these families are neighbourhood systems at each point. These are the axioms
for a topological space in the ‘neighbourhood approach’, put forward by Hausdorff.
See the exercises at the end of the chapter.
Proposition 3.12 Let X be a topological space. For x ∈ X , let Ux be the family of
all neighbourhoods of x.
1) If U ∈ Ux , then x ∈ U ;
2) if U, V ∈ Ux then U ∩ V ∈ Ux ;
3) if U ∈ Ux and U ⊂ V , then V ∈ Ux ;
4) if U ∈ Ux , then there is a V ∈ Ux such that V is a neighbourhood of each of
its points, i.e. V ∈ U y for each y ∈ V .

Proof All the properties are almost immediate. Property (1) is obvious. If U, V ∈
Ux , then x ∈ int U ∩ int V = int (U ∩ V ), so U ∩ V is in Ux , proving (2). The fact
that U ⊂ V implies int U ⊂ int V gives (3). For (4), just take V = int U . 
26 3 Topological Spaces

3.2 Basic Open Sets

Open intervals in the real line are basic open sets in the sense that these are the
building blocks of all open sets: every open set in the real line is a union of open
intervals. In metric spaces open balls play an analogous role. Such possible building
blocks on general topological spaces would obviously be of interest and importance.

Definition 3.13 Let X be a topological space. An open base, or just a base or basis,
for (the topology on) X is a collection B of open sets such that every open set in X
is the union of members of some subcollection of B. Each B ∈ B is called a basic
open set.
A family Bx of neighbourhoods of x is a called a neighbourhood base or a local
base at x if every neighbourhood of x contains some B ∈ Bx . Sets in Bx are referred
to as basic neighbourhoods of x.

The defining condition for a base is equivalent to the following: V is open in X

and x ∈ V implies there is a B ∈ B such that x ∈ B ⊂ V .
Bx is a local base at x means that U ∈ Ux implies there is a B ∈ Bx and an open
set V such that x ∈ V ⊂ B ⊂ U .

Proposition 3.14 Suppose B is a base for the space X . Then:

1) Every x ∈ X belongs to some B ∈ B.
2) If B1 , B2 are members of B that are not disjoint, then for each x ∈ B1 ∩ B2
there is a B ∈ B such that x ∈ B ⊂ B1 ∩ B2 .

Proof Part (1) just rephrases the fact that X = ∪{B : B ∈ B}.
(2) If B1 , B2 ∈ B then B1 ∩ B2 is open and so x ∈ B1 ∩ B2 implies there is a B ∈ B
with x ∈ B ⊂ B1 ∩ B2 (Fig. 3.1). 

Now let us see if the process can be reversed. In other words, we ask: If we
start with a collection B of subsets of a set X , is there a topology on X for which
B is a base? Such a B necessarily has to satisfy conditions (1) and (2) above. It is
a pleasant fact that these conditions are also sufficient for B to form a base for a
(unique) topology on X .

B
B1 B2
x

Fig. 3.1 Condition for a base

3.2 Basic Open Sets 27

Proposition 3.15 Let B be a collection of subsets of a nonempty set X that satisfies

conditions (1) and (2) of the previous Proposition. Then there is a unique topology
T on X for which B is base.
Proof The proof is really simple and routine. It requires no great ingenuity to see
what the topology that we are seeking is: Take T to be the collection of all possible
unions of subcollections of B. Condition (1) implies X ∈ T . The empty set, being
the union of the empty subcollection of B, is in T . That T is closed under arbitrary
unions is immediate from the definition of T . To prove that T is closed under finite
intersections, it suffices to show that U1 ∩ U2 ∈ T if U1 , U2 ∈ T . This is where
condition (2) comes into play. If x ∈ U1 ∩ U2 , then x ∈ Bi ⊂ Ui for some Bi ∈ B,
i = 1, 2, so x ∈ B1 ∩ B2 and there is a B ∈ B with B ⊂ B1 ∩ B2 ⊂ U1 ∩ U2 , by
(2). Thus U1 ∩ U2 ∈ T and T is a topology. Once this is done, it is clear, from the
definition of T that B is a base for T . Uniqueness of T is obvious. 
Topologies can be compared using bases or local bases.
Proposition 3.16 Let T1 , T2 be topologies on X with bases B1 , B2 . Then T1 ⊂ T2 if
and only if for every B1 ∈ B1 and x ∈ B1 there is a B2 ∈ B2 with x ∈ B2 ⊂ B1 .
Proof Suppose T1 ⊂ T2 and let x ∈ B1 ∈ B1 . Then B1 ∈ T1 ⊂ T2 and so there is a
B2 ∈ B2 with x ∈ B2 ⊂ B1 .
Conversely, assume that the stated condition on B1 and B2 holds and let U ∈ T1 .
For each x ∈ U , there is a B1 (x) ∈ B1 such that x ∈ B1 (x) ⊂ U . So there is a B2 (x) ∈
B2 such that x ∈ B2 (x) ⊂ B1 (x). Then U = ∪x∈U B2 (x) is a union of members of
B2 and so is in T2 . 
Example 3.17 1. For R2 , each of the following is a base:
B1 = the set of all open discs in the plane;
B2 = the set of all open rectangles with sides parallel to the coordinate axes,
(a, b) × (c, d), a, b, c, d ∈ R, a < b, c < d, in the plane;
B3 = the set of all open discs in the plane with rational radii and with centres at
points with rational coordinates;
B4 = the set of all open rectangles with sides parallel to the axes and with vertices
having rational coordinates.
Note that B3 and B4 are countable. These generalise to Rn , n ≥ 2.
2. The collection {(a, b) : a, b ∈ R, a < b} of intervals is a base for R. Intervals
(a, b), a, b ∈ Q, also form a base. {(x − n1 , x + n1 ) : n ∈ N}, is a local base at x ∈ R
(Verify these assertions).
3. One-point sets form a base for a discrete space X ; it is the smallest base possible.
{x} is a local base at any x ∈ X .
4. The collection {[a, b) : a, b ∈ R} satisfies the conditions for a basis on R; the
topology T it gives is called the lower limit topology. R , R with this topology,
is a useful source of examples and counter examples. For x ∈ R , [x, x + n1 ), n =
1, 2, · · · form a local base at x. How does this compare with the usual topology?
5. For two topological spaces (X, T ), (X  , T  ) consider the collection B =
{V × V  : V ∈ T , V  ∈ T  }. Since (U × U  ) ∩ (V × V  ) = (U ∩ V ) × (U  ∩ V  )
28 3 Topological Spaces

U2 × V2

U1 × V1

Fig. 3.2 Base for a product

(Fig. 3.2), it is easy to see that B satisfies the requisite conditions to be a base
for a topology on the product set X × X  , called the product topology on X × X  . It
will be taken up for a detailed study later.
6. Arithmetic progressions A(a, b) form a base for the topology for Tap (3.2) on
Z.
7. In a metric space X , open balls form a base and open balls B(x, n1 ), n ∈ N,
form a local base at x ∈ X .
8. The collection of unbounded open intervals, i.e. intervals of the form (a, ∞)
and (−∞, b), a, b ∈ R, is not a base for any topology on R, but finite intersections
of these give all open intervals which form a base for the usual topology. To take
care of situations like these, we give the following useful definition (first introduced
by Bourbaki).

Definition 3.18 A collection S of (open) subsets of a topological space X is called

a subbase if finite intersections of sets in S form a base for the space. In this case,
the topology on X is generated by S.

Example 3.19 1. Any collection of subsets of X is a subbase for a topology on X .

2. Unbounded open intervals (a, ∞) and (−∞, b) with a, b ∈ R, form a subbase
for the usual topology on R. Describe the topology having intervals of the form
(a, ∞), a ∈ R, as a subbase.
3. Open strips (a, b) × R = {(x, y) ∈ R2 : a < x < b} and R × (c, d) = {(x, y) ∈
R : c < y < d}, a, b, c, d ∈ R, together form a subbase for the usual topology on
2

R2 .
4. For a set X , spaces Yi and a family of maps f i : X → Yi , the collection S of
all sets of the form f i−1 (Vi ), Vi ⊂ Yi open, is a subbase for a topology on X , known
as the initial topology defined by the f i . Such topologies will be studied in detail in
Chap. 6.

3.3 Closed Sets

Closed sets are complements of open sets, but the notions of ‘open sets’ and ‘closed
sets’ are not complementary!
3.3 Closed Sets 29

Definition 3.20 A subset F of a topological space X is said to be closed if its

complement F c = X \ F in X is open.

Proposition 3.21 In any topological space X , X and ∅ are closed and arbitrary
intersections and finite unions of closed sets are closed.

Proof Immediate from the definition. Look at unions and intersections of open sets
and take complements. 

Infinite unions of closed sets may not be closed: ∪n∈N [1/n, 1] = (0, 1] is not
closed. A family {E i } of subsets in a space X is locally finite if each x ∈ X has a
neighbourhood Ux which meets only finitely many E i .
Proposition 3.22 The union of a locally finite family of closed sets is closed.

Proof Let {E i } be a locally finite family of closed sets and let E = ∪E i . If x ∈ E c ,

let U be a neighbourhood of x which meets only finitely many E i , say for i ∈ F.
Then U ∩ (∩i∈F E ic ) ⊂ E c is a neighbourhood of x. So every point of E c is an interior
point. 

Example 3.23 1. In a discrete space every subset is closed and in an indiscrete space
there are no nontrivial closed sets.
2. [0,1] is closed in R (being the complement of (−∞, 0) ∪ (1, ∞)) and so are
Z, {0}; Q is not. {[n, n + 1]}∞ 0 is a locally finite family of closed sets whose union
[0, ∞) is closed.
3. In R2 , the sets {(x, 0) : x ∈ R} and {(x, y) : x 2 + y 2 = 1} are both closed;
Q × Z is not.
4. In a metric space X , single point sets {x} (hence finite sets) are closed: if y = x,
then B(y; d(x, y)) ⊂ X \ {x}, so the last set is open.
5. Every closed ball B̄(x; r ) in any metric space X is a closed set. Reason: if z ∈
X \ B̄(x; r ), then d(z, x) > r and B(z, ε) ⊂ X \ B̄(x; r ) for any 0 < ε ≤ d(z, x) −
r (check); thus B̄(x; r )c is open.
6. In the Sierpiński space S (3.2), {0} is closed, {1} is not.
7. Consider R (3.17). For real numbers a, b, (−∞, a) = ∪∞ 1 [−n, a) is open and
so is [b, ∞) = ∪∞ 1 [b, n). Thus [a, b), a < b, is closed. (Recall that such intervals
form basic open sets for R . So, in R , the intervals [a, b) are open sets which are
also closed.)

Recall that every set has a largest open subset (namely, the interior of the set).
Complementarily, every set has a smallest closed superset.

Definition
 3.24 If A is a subset of a space X , the closure Ā of A is defined by
Ā = {F ⊂ X : F ⊃ A, F is closed}.

Here are the basic, elementary properties of closure.

30 3 Topological Spaces

Proposition 3.25 Let X be a space and let A ⊂ X . Then

1) Ā is the smallest closed set in X containing A;
2)A is closed if and only if A = Ā;
3) x ∈ Ā if and only if every neighbourhood V of x intersects A;
4) Ā¯ = Ā.

Proof (1) is a consequence of the definition (the reader should explicitly spell out
the arguments) and (2) is immediate from (1). To prove (3), observe that if x ∈ / Ā,
then x ∈
/ F for some closed set F ⊃ A and V = F c is a neighbourhood of x disjoint
from A. This proves one part of (3). For the converse, retrace the steps to reverse the
arguments. Finally, (4) follows from (2) and the fact that Ā is closed. 

The class of closed sets determine the topology and closed sets are precisely those
sets which are fixed under the closure operation A → Ā, so one can expect that the
closure operation determines the topology. This is, indeed, the case; see the exercises
for a precise formulation.

Example 3.26 1. In R, the closure of (0,1) is [0,1] and the closures of Q and R \ Q
are both R. (Recall the construction of real numbers from rational numbers. What
are the properties of rationals and irrationals that are used to get the last statement?)
2. In R (3.17), the closure of (0,1) is [0,1) and the closures of Q and R \ Q are
both R .
3. Consider R with the cocountable topology. The closure of (0,1) is the whole of
R, the closure of Q is itself and that of R \ Q is R.
4. In Rn with the Euclidean metric, the closure of an open ball B(x; r ) is the
corresponding closed ball B̄(x; r ). (Warning: the assertion is not true in a general
metric space!)
5. In a discrete space Ā = A for every set A and in an indiscrete space Ā = X
unless A is empty.

Definition 3.27 A subset A of a space is locally closed at x ∈ A if there is a neigh-

bourhood U of x in X such that A ∩ U is closed in U . It is said to be locally closed
if it is locally closed at each of its points.

Exercise: Is a closed set locally closed? Converse?

Proposition 3.28 A subset A of a topological space X is locally closed if and only
if A = V ∩ F where V is open and F is closed in X .

Proof Suppose A is locally closed. For each x ∈ A, let Vx be an open neighbourhood

of x in X such that A x := A ∩ Vx is closed in Vx , so Vx \ A x is open in Vx , hence
in X . Then V = ∪x∈A Vx is open in X and V \ A = ∪(Vx \ A x ) is open. Thus A is
closed in V . Hence A = V ∩ F where F is closed in X .
Conversely, if A = V ∩ F as in the statement, then V is a neighbourhood of x for
each x ∈ A and A = F ∩ V is closed in V . 

We now look at the relation between convergence and closed sets.

3.3 Closed Sets 31

Proposition 3.29 Suppose X is a metric space and A ⊂ X . Then x ∈ Ā if and only

if there is a sequence {xn } A which converges to in x.

Proof If x ∈ Ā, then A ∩ B(x; n1 ) is nonempty for each positive integer n, so there is
an xn ∈ A ∩ B(x; n1 ) and xn → x. Conversely, if x ∈ / Ā, then there is a ball B(x; ε)
disjoint from A, that is d(y, x) ≥ ε for all y ∈ A, and no sequence in A can converge
to x. 

Sequential convergence in topological spaces is defined as in metric spaces, but

using neighbourhoods in place of balls.

Definition 3.30 A sequence {xn } in a topological space X is said to converge to

x ∈ X if for every neighbourhood V of x there is a positive integer n V such that
xn ∈ V for all n ≥ n V .

Limits of sequences from a closed set are always in the set. But a set with this
property may not be closed in a topological space, as the following example shows.

Example 3.31 Consider the space R with the cocountable topology. The only closed
sets different from the whole space are countable. So if I = R \ Q is the set of
irrational numbers, then Ī = R. If {xn } is a sequence in R with xn → x, then V =
{x} ∪ R \ {xn : n = 1, 2, ...} is a neighbourhood of x and so must contain the xn for
all but finitely many n. Clearly, this is possible only when the sequence is eventually
the constant x. In other words, the only convergent sequences are those which are
eventually constant. In particular no rational number is a limit of a sequence of
irrationals. Thus, the previous proposition fails in this case. We conclude that for
arbitrary topological spaces, sequences are inadequate to describe the topology.

This leads us to the following definition.

Definition 3.32 A subset E of a topological space X is sequentially closed if for

every sequence in E which converges in X , the limit is in E. If E is sequentially
closed, E c is said to be sequentially open.

Clearly, any closed set is sequentially closed, so open sets are sequentially open.
The previous example shows that the set of irrationals is sequentially closed but
not closed in the cocountable topology of R and so the reverse implication is not
generally true. But in metric spaces, the class of closed sets is the same as the class
of sequentially closed sets, as we have seen earlier (3.29).

Proposition 3.33 Let (X, T ) be a topological space and let Tseq be the collection
of all sequentially open sets in X . Then Tseq is a topology on X that is stronger than
T.

Proof The only property that requires verification is that the intersection of two
sequentially open sets is sequentially open, or, equivalently, that the union of two
sequentially closed sets is sequentially closed. So, let E, F be sequentially closed in
X and let A = E ∪ F. Suppose A is not sequentially closed. Then there is a sequence
32 3 Topological Spaces

{xn } in A converging to x ∈ Ac . But at least one of E and F must contain xn for

infinitely many n, say E contains a subsequence {xn k }. Then this subsequence con-
verges to x ∈ Ac ⊂ E c , contradicting the assumption that E is sequentially closed.
This contradiction proves A is sequentially closed and completes the proof that Tseq
is a topology. The last part is clear since open sets are sequentially open. 

Definition 3.34 A set D in a topological space X is said to be dense in X if its

closure is the whole space: D̄ = X .

Proposition 3.35 A subset D of a topological space X is dense in X if and only if

D intersects every nonempty open set V in X .

Proof Suppose D intersects every nonempty open set. Then for any x ∈ X , any
neighbourhood V of x intersects with V and so x ∈ D̄. Thus D̄ = X . On the other
hand, if there is a nonempty open set V disjoint from D, then x ∈
/ D̄ for x ∈ V and
hence D is not dense. 

Example 3.36 1. Q, the set of rationals, and R \ Q, the set of irrationals are both
dense in R.
2. In an indiscrete space, every nonempty set is dense and in a discrete space no
proper subset is dense.
3. In the Sierpiński space S (3.2), {1} is dense, but {0} is not.
The next two examples were mentioned in passing earlier.
4. Here is a ‘small’ set of rationals that is also dense in R. A rational number of the
form m/2n , m, n ∈ Z is called a dyadic rational. It is an often used fact that the set
of dyadic rationals is dense in R. Proof : a < b implies 2n (b − a) > 1 for large n, so
the interval (a.2n , b.2n ) has length greater than one and hence contains an integer m:
a.2n < m < b.2n or a < m/2n < b. In other words, every interval contains a dyadic
rational. Hence the assertion. √
5. Now a ‘small’ dense set of irrationals: {m + n 2 : m, n ∈ Z} is a (countable)
dense set in R. This is an exercise for the reader.

Exercise 3

1. Topologies. a) Determine all the topologies on {1, 2, 3}.

b) The union of two topologies need not be a topology.
c) Find the lub and the glb of pairs of topologies on R seen so far.
d) Tbd , the set of complements V = K c of those subsets K of R which are closed
and bounded with the usual metric on R, is a topology. How does it compare with
the usual topology on R?
e) Define a set V in N to be open if n ∈ V implies every divisor of n is in V . This
gives a topology on N.
f) Write R̄ = [−∞, ∞] for R ∪ {±∞}. Let T∞ be the topology generated by
sets of the form (a, ∞] = (a, ∞) ∪ {∞}, [−∞, b) = {−∞} ∪ (−∞, b), a, b ∈ R.
R̄, with this topology, is called the extended real line. The relative topology on R ⊂ R̄
is the usual topology.
3.3 Closed Sets 33

g) Describe the topology of a straight line in the plane as a subspace of Rd × R

(Rd is the real line with the discrete topology.).
h) If (X, T X ), (Y, TY ) are spaces, is T X × TY a topology on X × Y ?
2. Hausdorff metric. Let (X, d) be a metric space and let Bc (X ) be the
class of bounded, closed subsets of X . Recall (A, B) = supx∈A d(x, B). Define
d H (A, B) = max{d(A, B), d(B, A)}, A, B ∈ Bc (X ). For ε > 0, let
A(ε) := ∪{B(x, ε) : x ∈ A}. Then
(a) d H (A, B) = inf{ε : A ⊂ B(ε), B ⊂ A(ε)}.
(b) d H is a metric on Bc (X ), called the Hausdorff metric.
3. Interiors. a) Examine the truth or falsity of the following statements, for subsets
A, B of a topological space X :
i) int (A ∪ B) = int A ∪ int B;
ii) int (A ∩ B) = int A ∩ int B.
b) Find the interiors of:
i) R \ {1/n : n = 1, 2, ...} in R;
ii) R \ N in R;
iii) R2 \ (Z × Q) in R2 ;
iv) [0, 1], (0, ∞) in R with T = {V ⊂ R : V = ∅ or 0 ∈ V}.
4. Bases. a) Do intervals of the form (−∞, a), a ∈ R, form a base for a topology?
If so, compare it with other topologies on R given earlier.
b) If B, B  are bases for (X, T ), (X  , T  ) respectively, B × B  is a base for the
product topology on X × X  .
c) The product topology on R × R (usual topology on R) coincides with the usual
topology of R2 . The same holds for Rn , n ≥ 2.
d) Do intervals of the form [a, b), a, b ∈ Q, form a base for R (3.17)?
e) Sets of the form U ( f ; x1 , · · · , xn ; ε) = {g :| g(xi ) − f (xi ) |< ε} is a basis for
a topology on C[0, 1]. Here, of course, f ∈ C[0, 1], x1 , · · · , xn finitely many points
in [0,1] and ε > 0 are all arbitrary.
f) A collection B of open sets in a space X is a base for X if and only if Bx =
{B ∈ B : x ∈ B} is a local base at x for each x ∈ X .
g) On N, all subsets which contain Tn = {k : k ≥ n} for some n form a base for a
topology.
h) All intervals (a, b), a, b ∈ R, together with one-point sets {n}, n ∈ N, form a
base for a topology on R.
i) Let Z (P) be the set of zeros of a real polynomial in n variables, Z (P) =
{(x1 , · · · , xn ) ∈ Rn : P(x1 , · · · , xn ) = 0}. As P varies over all such polynomials,
the complementary sets Z (P)c form a base for a topology on Rn , called the Zariski
topology, in which one-point sets are closed, but distinct points cannot be separated
(see Chap. 8 on separation properties). On R, it is the cofinite topology, but in higher
dimensions it is not. Clearly, it makes sense in a more general setting and is of great
importance in Algebraic Geometry.
5. Products. Recall the metrics ρ1 , ρ2 and ρ∞ on a finite product of metric spaces.
(See Exercises in Chap. 2). In each case, the metric topology is the product topology.
6. Subbases. a) Any arbitrary nonempty collection S of subsets of a nonempty
set X is a subbase for some topology on X .
34 3 Topological Spaces

b) Describe the topology on R2 having as a subbase i) all straight lines; ii) all
straight lines parallel to x-axis.
7. Order topology. Ordered spaces were first defined and studied by Eilenberg,
[8]. Let X be a totally ordered set and define intervals as in the real line. The order
topology has as a subbase all sets of the form L a := {x ∈ X : a < x} and Ub := {x ∈
X : x < b}, a, b ∈ X .
a) On R, the usual order gives the usual topology.
b) Write down a base for the order topology on X . (Warning: X may have a
smallest element or a largest element or both or neither.)
c) On the product X × Y of ordered sets, (x, y) < (x  , y  ) if x < y or x = y and
x < y  gives a total order, called the dictionary order. The order topology on N is


the discrete topology, but the order topology on N × N with the dictionary order is
not the discrete topology.
d) What is the order topology on R × R with the dictionary order?
e) In a space X with order topology, is [a, b] the closure of (a, b)?
f) On [0, 1] × [0, 1] with the dictionary order topology, describe neighbourhoods
of (0,0) and (1,1).
g) Compare the dictionary order topology on [0, 1] × [0, 1] with the subspace
topology of [0, 1] × [0, 1] as a subspace of R × R with the dictionary order topology.
h) The order topology induced by the induced order may not be the same as the
relative order topology. Make this precise.
8. Subspaces. a) A subspace of a subspace is a subspace, that is, if A ⊂ B ⊂ X ,
then the relative topology on A as a subspace of X is the same as its relative topology
as a subspace of the subspace B of X .
b) Let X be a space and A ⊂ E ⊂ X . Discuss the relation between:
i) interior of A in E and interior of A in X ;
ii) closure of A in E and closure of A in X .
c) Let E be a subspace of X . Does a basis (or a subbasis) of X give rise to a basis (or
a subbasis) for E? For example, if B is a basis for X , is E ∩ B := {E ∩ B : B ∈ B}
a basis for E?
9. Topology from neighbourhood bases. a) Obtain a result on comparing two
topologies in terms of local bases, analogous to (3.16).
b) Let X be a set and suppose a collection Ux of subsets of X is assigned
for each x ∈ X satisfying the properties (i)–(iv) listed in Proposition 3.12. Define
T = {U ⊂ X : x ∈ U ⇒ U ∈ Ux }. Then T is a topology on X and Ux is precisely
the neighbourhood system at each x ∈ X in this topology. Obtain an analogous char-
acterisation for a collection Bx to form a neighbourhood base at x ∈ X .
c) Fix a prime p. For n ∈ Z and k ∈ N, let Unk = {n + mp k : m ∈ Z}. Then Un =
{Unk : k ∈ N} is a local base at n ∈ Z for a topology on Z.
10. Closures and interiors. a) Examine the validity of the following:
i) the closed ball B̄(x, r ) (in a metric space) is a closed set;
ii) the interior of B̄(x, r ) is the open ball B(x, r );
iii) the closure B(x.r ) of B(x, r ) is the closed ball B̄(x.r ).
b) A ∪ B = Ā ∪ B̄. What about infinite unions? Intersections?
3.3 Closed Sets 35

c) If E is a subset of the real line which is bounded above, then sup E ∈ Ē. Is it
true that sup E is a limit point of E?
d) Find the closure of the graph of f (x) = sin 1/x, 0 < x ≤ 1.
e) True or false? i) Ē = int E; ii) int E=int Ē;
iii) ( Ē)c = int E c ; iv) (int E)c =E c .
f) If ∪ Āi is closed, then ∪ Āi = ∪Ai .
g) If (X, d) is a metric space and A, B ⊂ X , then diam Ā= diam A and d( Ā, B̄) =
d(A, B).
h) Any infinite closed subset of R is the closure of a countable set. Does the result
hold for any Rn ?
i) What is the closure of {1/n : n ∈ N} in R with the cofinite topology? cocount-
able topology?
j) In R (3.17), x ∈ Ā√ if and√
only if there is√
a sequence
√ xn ≥ x with |xn − x| → 0.
Find the closures of (1, 2), ( 2, 2) and (− 2, 2).
k) Let X be the subspace of R2 consisting of the unit circle together with the
diameter on the x-axis. For p = (0, 0), determine the open ball B X ( p, 1), its closure
and the closed ball B̄ X ( p, 1).
11. Closure of the interior vs interior of the closure. (See Bourbaki, [9].) If
X is a topological space and A ⊂ X , let α(A) =int Ā and β(A) = int A.
a) A ⊂ α(A) if A is open and β(A) ⊂ A if A is closed.
b) α(α(A)) = α(A) and β(β(A)) = β(A) for any A.
c) If U and V are disjoint open sets, then so are α(U ) and α(V ).
d) Find a subset A of the real line such that A, int A, Ā, α(A), β(A), β( Ā) and
α(int A) are all distinct.
12. Dense sets. a) D is dense if and only if D c has empty interior.
b) If A, B are dense subsets, A ∩ B may not be dense. (In fact, they can be
disjoint.) However if A, B are also open, then A ∩ B is dense.
c) If D is dense in X , is D ∩ E dense in E for any E? open E?
d) If a subset D intersects all sets in a subbasis, is D necessarily dense? What
happens if it intersects all sets in a basis?
e) Find an explicit countable dense subset in R \ Q.
f) If D is dense in A then it is dense in Ā.
g) When does a subset meet every dense subset?
h) Every dense subset of R contains a countable dense subset.
13. Arithmetic progressions. a) The arithmetic progression topology Tap (3.2)
is a topology on Z and each nonempty open set is infinite.
b) In (Z, Tap ), the basic open sets  A(a, b) are also closed sets.
c) Observe that Z \ {1, −1} = {A(0, p) : p is a prime number} and deduce that
there are infinitely many primes. (This topological proof of the infinitude of primes
is due to Hillel (Harry) Furstenberg, [10].)
14. Matrices. In the space M(n, R) of all real n × n matrices,
a) the subset G L(n, R) of nonsingular matrices is open;
b) the subset S L(n, R) of matrices determinant one is closed;
c) the set Symm(n, R) of symmetric matrices is closed.
36 3 Topological Spaces

15. G δ sets and Fσ sets. A countable intersection of open sets is called a G δ -set
and a countable union of closed sets is called an Fσ -set.
a) E is a G δ if and only if E c is an Fσ ;
b) In a metric space, any closed set is a G δ and any countable set is an Fσ (In
particular, Q is an Fσ ; it is not a G δ as we will we see in the chapter on locally
compact spaces);
c) A closed interval [a, b] is both a G δ and an Fσ .
16. Regular open sets and regular closed sets. An open set V is regular open
if it is the interior of its closure: V = int V̄ and a closed set E is regular closed if it
is the closure of its interior: E = int E.
a) The closure of an open set is regular closed and the interior of a closed set is
regular open.
b) V is regular open if and only if V c is regular closed.
c) Find examples of open sets and closed sets that are not regular.
17. Kuratowski closure operation: a) Let X be a set and suppose there is a
mapping, denoted by E → cl(E), from the power set P(X ) to itself satisfying the
following properties:
K1) cl(∅) = ∅;
K2) E ⊂ cl(E);
K3) cl(cl(E)) = cl(E);
K4) cl(E ∪ F) = cl(E) ∪ cl(F) for E, F ⊂ X .
T = {E c : E ⊂ X, cl(E) = E} is a topology on X and cl(E) is just the closure Ē
in this topology. Any map satisfying the conditions K1 - K4 is called a Kuratowski
closure operation on X . [11]
b) What is an analogous operation for the interior of a set?
c) Find the Kuratowski closure operation for the cofinite topology.
d) Fix A ⊂ X and define cl(E) = E ∪ A for any nonempty E ⊂ X (and cl(∅) =
∅). This gives a closure operation. Describe the associated topology. Discuss the
cases when A = ∅, A = {a} and A = X .
18. Uniform spaces. A uniformity for a set X is a nonempty family U of subsets
of X × X with the following properties:
i) every member of U contains the diagonal = {(x, x) : x ∈ X };’
ii) if U ∈ U , then U −1 ∈ U , where U −1 = {(x, y) : (y, x) ∈ U };
iii) if U ∈ U , then V.V ⊂ U for some V ∈ U where, for V, W ∈ U , V.W =
{(x, y) : ∃z, (x, z) ∈ V, (z, y) ∈ W };
iv) if U, V ∈ U , then U ∩ V ∈ U ;
v) if U ∈ U and V ⊃ U , the V ∈ U .
A set with a uniformity is called a uniform space. This was introduced by A.Weil
[12]. For detailed treatments, see [9, 13, 14].
a) If X is a metric space, U = {U ⊂ X × X : ∃r > 0, Dr ⊂ U } is a uniformity,
where Dr = {(x, y) : d(x, y) < r }.
b) If (X, U ) is a uniform space, {E : ∀x ∈ E, ∃U ∈ U , U [x] ⊂ E} where
U [x] = {x ∈ X : (x, y) ∈ U }, is a topology on X (U -topology).
3.3 Closed Sets 37

c) The class of all subsets of X × X containing is a uniformity which gives

the discrete topology. The single set X × X forms a uniformity giving the indiscrete
topology.
d) B ⊂ U is a base for a uniformity U if each member of U contains a member
of B. In this case {B[x] : B ∈ B} is a local base at x for x ∈ X . Dr , r > 0, form a
base in the metric uniformity of a).

3.4 Biographical Notes

3.4.1 Riemann

Bernhard Riemann (1826–1866) was one of the all-time greats in mathematics. He

was mentored by the formidable genius Karl Friedrich Gauss himself at Göttingen
in Germany. He did not publish much, but almost each of his papers in mathe-
matics created a new area in the subject - they were truly seminal. The genesis of
modern topology can be traced to his work on function theory, abelian integrals
etc. He was the first to apply topology to analysis. Real analysis, Fourier series,
function theory, number theory and differential geometry are the main areas of his
contributions, besides mathematical physics. The mathematical objects named after
him are too many to be listed, but here are some of them: Riemann sum, Riemann
integral, Riemann rearrangement theorem, Riemann localisation theorem, Riemann
sphere, Riemann surface, Riemann zeta function, Riemann hypothesis, Riemannian
geometry, Riemannian connection, Riemannian symmetric space, Cauchy–Riemann
equations, Riemann–Roch theorem, Riemann–Hilbert problem.

3.4.2 Weyl

Hermann Weyl (1885–1955) was a student of the great David Hilbert at Göttingen
and is well-known for his work in several areas of mathematics, including differential
operators, Riemann surfaces, Lie theory and representation theory, number theory,
foundations, relativity. He was the first to give a rigorous definition of Riemann
surfaces (in which process he used neighbourhood systems to define an abstract
space). We come across Weyl transform, Weyl character formula, Weyl integration
formula, Weyl chamber, Weyl group, Weyl basis, Peter–Weyl theorem, Schur–Weyl
duality, etc. in mathematics.
38 3 Topological Spaces

3.4.3 Hausdorff

Felix Hausdorff (1868–1942) was a mathematician and philosopher from Germany.

He published his work on philosophy under the pseudonym Paul Mongré. His doc-
toral dissertation was on astronomical light refraction, but he later worked in many
areas of mathematics—topology, set theory, measure theory, functional analysis,
function theory, etc. His book Grundzüge der Mengenlehre (1914), and a revised
edition, Mengenlehre, (1927) had great influence in the development of general topol-
ogy. The first definition of topological space and a large part of what we study in the
subject now are found here. Hausdorff committed suicide (along with his wife and her
sister) in 1942 due to Nazi persecution. Hausdorff space, Hausdorff measure, Haus-
dorff dimension, Hausdorff metric, Hausdorff moment problem, Hausdorff maximal
principle (an equivalent of Axiom of Choice), Hausdorff–Young inequality are some
of the things named after him.

3.4.4 F. Riesz

Frigyes Riesz (1880–1956) was a Hungarian mathematician and was one of the
founders of Functional Analysis. Besides his fundamental work in functional anal-
ysis, he contributed to integral equations, ergodic theory and topology. He made
significant contributions to axiomatic development of general topology. His main
suggestions (made in 1906 and 1908) for possible formulation were overlooked
at that time. According to historians of mathematics, his contributions came to be
appreciated only much later. For a discussion of F. Riesz’s role in the development of
general topology, see [15, 16]. The definition of connected spaces that we use today
is essentially the one given by him. Riesz Lemma, Riesz Representation Theorem,
Riesz-Fischer theorem, Riesz-Schauder theory (of compact operators), Riesz spaces,
F. and M.Riesz theorem (named after F.Riesz and his younger brother Marcel Riesz,
who is also a famous analyst) are some of the things that stand testimony to his con-
tributions to analysis. It is said that many of the central theorems of modern analysis
(e.g. Spectral theorem, Ergodic theorem) have proofs due to F. Riesz.

References

1. Riemann, B.: Über die Hypothesen, welche der Geometrie Grunde liegen. Abhandlungen der
Königlichen Gesellschaft der Wissenschaften zu Göttingen 13, 133–150 (1868)
2. Weyl, H.: Die Idee der Riemannschen Flaäche, Teubner, 1913. (Translated into English as
‘The Concept of a Riemann Surface’, (Latest: Dover 2019)
3. Hausdorff, F. Grundzüge der Mengenlehre. Leipzig (1914)
4. Tietze, H.: Beiträge zur allgemeinen Topologie I. Math. Ann. 88, 290–312 (1923). https://
doi.org/10.1007/BF01579182
References 39

5. Armstrong, M.A.: Basic Topology. Undergraduate Texts in Mathematics, Springer Interna-

tional (2004)
6. Hausdorff, F.: In: de Gruyter (ed.), 3rd edn. English Translation by J.R. Aumann et al: Set
Theory, Chelsea (1957)
7. Golomb, S.W.: A connected topology for the integers. Am. Math. Monthly 66, 663–665
(1959). https://doi.org/10.2307/2309340
8. Eilenberg, S.: Ordered topological spaces. Am. J. Math. 63, 39–45 (1941). https://doi.org/10.
2307/2371274
9. Bourbaki, N.: General Topology, Chapters 1–4. Springer (1995)
10. Furstenberg, H.: On the infinitude of primes. Am. Math. Monthly 62, 353 (1955). https://doi.
org/10.2307/2307043
11. Kuratowski, Sur l’opération Ā de l’Analysis Situs, Fund. MathK., 3, 182–199 (1922). https://
doi.org/10.4064/fm-3-1-182-199
12. Weil, A.: Sur les Espaces à Structure Uniforme et sur la Topologie Générale. Hermann, Paris
(1937)
13. Kelley, J.L.: General Topology. Affiliated East-West Press (1969)
14. Willard, S.: General Topology. Addison-Wesely (1970)
15. Rodriguez, L.: Frigyes Riesz and the emergence of general topology. Arch. History Exact Sci.
69, 55–102 (2015). https://doi.org/10.1007/s00407-014-0144-6
16. Thron, W.J. Frederic Riesz’ contributions to general topology. In: Aull, C.E., Lowen, R. (eds.)
Handbook of the History of General Topology, vol. 1, pp. 31–40. Kluwer (1997)
Chapter 4
Continuous Maps

Continuity is the heart of topology. Here are the first proper definitions of continuity
(on the real line).
According to a correct definition, the expression that a function f (x) varies
according to the law of continuity for all values of x inside or outside certain limits
means only that, if x is any such value the difference
f (x + ω) − f (x) can be made smaller than any given quantity, provided ω can be
taken as small as we please. (Bolzano, 1817) [1]
. . . f (x) sera fonction continue, si . . . la valeur numérique de la différence
f (x + α) − f (x) décroît indéfiniment avec celle de α . . .(Cauchy, [2] 1821).
(. . . f (x) will be (called) a continuous function, if . . . the numerical values of the
difference f (x + α) − f (x) decrease indefinitely with those of α . . .)
Wir nennen dabei eine Grösse y eine stetige Function von x, wenn man nach
Annahme einer Grösse ε die Existenz von δ beweisen kann, sodass zu jedem Wert zwis-
chen x0 − δ...x0 + δ der zugehörige Wert von y zwischen y0 − ε...y0 + ε(Weierstrass
[3, 4])
(We will call a quantity y a continuous function of x, if after choosing a quantity
ε the existence of δ can be proved, such that for any value between x0 − δ...x0 + δ
the corresponding value of y lies between y0 − ε...y0 + ε.)
Cauchy’s is an acceptable qualitative definition. The quantitative version of this
is the ε-δ definition due to Bolzano (1817) [1] and Weierstrass (1874) [4], discussed
in the chapter on metric spaces. (Some historians of mathematics feel that Cauchy
might have got the idea from Bolzano whom he met earlier in Prague during his exile
from France, see [5].)

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 41
K. Parthasarathy, Topology, La Matematica per il 3+2 134,
https://doi.org/10.1007/978-981-16-9484-4_4
42 4 Continuous Maps

4.1 Limit Points

Continuity is obviously tied to the idea of limit. We consider that later and here start
with the related concept of limit points. But first here is an example.

Example 4.1 Intuitively, the closure Ā of a set A is got by adding to A points which
are close to A. Points of A may or may not be ‘close’ to A. If A = (0, 1) ∪ {3} in
R, then Ā = [0, 1] ∪ {3}. The points 0 and 1 are not points of A but are ‘close’ to
A—all neighbourhoods of these points meet A whereas 3 ∈ A, but 3 is not close to
A—the neighbourhood (2,4) of 3 meets A in only one point, namely 3 itself.

Definition 4.2 An x ∈ X is called a limit point or an accumulation point of E ⊂ X

if V ∩ (E \ {x}) = ∅ for each neighbourhood V of x (In words?); if x ∈ E is not a
limit point, it is an isolated point of E. E  , the set of limit points of E, is called the
derived set of E.

Proposition 4.3 For any subset E of a space X , Ē = E ∪ E  and E is closed if and

only if E  ⊂ E. In words, the closure of a set is made up of points of the set and the
limit points of the set, and a set is closed if and only if it contains all its limit points.

Proof The characterisation of Ē given earlier and the definition of E  make it clear
that E  ⊂ Ē and so E ∪ E  ⊂ Ē. On the other hand, if x ∈ Ē \ E, then, for any
neighbourhood V of x, V ∩ E = ∅, so V ∩ (E \ {x}) = ∅, and x is a limit point of
E. Now the second assertion is immediate from the fact that E is closed if and only
if E = Ē. 

Example 4.4 1. If E = (0, 1) in R, then E  = [0, 1] = Ē.

2. Every point in R is a limit point of Q and also of R \ Q.
3. In R, every point of Z is an isolated point.
4. In a discrete space, every point is an isolated point (What happens in an indis-
crete space?).
5. Let E = {( n1 , y) : n ∈ N, y ∈ [0, 1]}, a union of vertical segments in R2 . Then

E = Ē is the union of E and the vertical segment {(0, y) : 0 ≤ y ≤ 1}.

In metric spaces, limit points satisfy much stronger properties than the ones demanded
by the definition.

Proposition 4.5 If x is a limit point of a set E in a metric space X , then every

neighbourhood of x contains infinitely many points of E and there is a sequence {xn }
of distinct points of E converging to x.

Proof Suppose there is a neighbourhood V of x with V ∩ E is finite, say V ∩ E \

{x} = {y1 , ..., yk }. Let ε = min d(x, y j ), choose δ > 0 such that B(x; δ) ⊂ V and
1≤ j≤k
let η = min(ε, δ). Then B(x; η) is disjoint from E \ {x} and so x is not a limit
4.1 Limit Points 43

point of E, proving the first part. Hence, if x is a limit point of E, for each positive
integer n, B(x; n1 ) ∩ E is infinite. Thus xn can be chosen inductively such that xn ∈
B(x; n1 ), xn = xk , 1 ≤ k ≤ n − 1. The sequence {xn } satisfies the requirements of
the last assertion. 

Corollary 4.6 A finite set in a metric space has no limit points and hence is closed.

Example 4.7 In R with the cocountable topology, any rational number q is a limit
point of the set R \ Q of irrational numbers, but no sequence of irrationals can
converge to q. So sequential limits are not sufficient to get limit points in general
topological spaces.

In spite of this ‘set back’, limit points and closures in arbitrary topological spaces
can be characterised in terms of convergence. What is needed is a concept of ‘gen-
eralised sequences’. A sequence in a set X is a map N → X where N has the natural
order. A ‘generalised sequence’ in X is got by replacing N by a set with suitable
order properties.

Definition 4.8 A directed set is a set D with an order relation ≤ which is reflexive,
transitive and has the property that any two elements have an upper bound: (i) α ≤
α; (ii) α ≤ β, β ≤ γ , α, β, γ ∈ D implies α ≤ γ ; and iii) for any pair of elements
α, β ∈ D there is a γ ∈ D with α ≤ γ , β ≤ γ . A generalised sequence or a net in a
set X is a map x : D → X from a directed set to X . Following standard convention
we write xα for x(α) and write a net x as (xα ) (just as for sequences).

Example 4.9 1. N, Q, R with their natural orders are all directed sets. Hence,
sequences are nets. (Is every E ⊂ R a directed set?)
2. The class P(X ) of all subsets of a nonempty set X is a directed set, ordered
by inclusion.
3. Let X be a topological space and let x ∈ X . The set Ux of all neighbourhoods
of x ordered by reverse inclusion (i.e. U ≤ V if V ⊂ U ) is a directed set.
4. The Cartesian product D = D1 × D2 of two directed sets, with coordinate-wise
order (α1 , α2 ) ≤ (β1 , β2 ) if α j ≤ β j , j = 1, 2.
5. Let P be the set of subdivisions S = {a = t0 < t1 < · · · < tn = b} of a closed
interval [a, b] in R. For two subdivisions S1 , S2 define S1 ≤ S2 to mean S2 is a
refinement of S1 , that is, S2 is got by adding more points of subdivisions to S1 :
S1 ⊂ S2 . This makes P a directed set.

Now convergence of nets is defined by mimicking that of sequences.

Definition 4.10 A net (xα )α∈D in a topological space X converges to a point x ∈ X

if, for every neighbourhood V of x, there is an αV ∈ D such that xα ∈ V for all
α ≥ αV (i.e. αV ≤ α); this is expressed by saying that the xα ’s are eventually in any
neighbourhood of x.
44 4 Continuous Maps

Remark 4.11 Note that, unlike the case of sequences, ‘eventually’ may not be the
same as ‘all but finitely many’.

This convergence is also called ‘Moore–Smith convergence’, after the work (1922)
of American mathematicians Moore and Smith [6, 7]. They considered only scalar
valued ‘generalised sequences’, motivated by the example of Riemann sums, 4.12.
The concept of directed sets was independently introduced in 1921 by Vietoris [8],
who considered a general set-theoretic set up. ‘Whereas Moore and Smith con-
sidered only generalized sequences with numerical values, Vietoris studied right
from the start ... systems of sets under Zermelo’s axioms, indexed by a directed set,
and gave the definition: ... So Vietoris developed in parallel today’s theory of con-
vergence for generalized set sequences (nets) and filter bases through comparison
with the directed set of neighborhoods...Vietoris was thus the father of the modern
convergence concepts’. (Quoted with the permission of the American Mathemat-
ical Society.) [9] (For ‘filters’, another theory of convergence, see the exercises.)
Net convergence was applied to topology by the American mathematician Garret
Birkhoff (1937), [10]. The term ‘net’ is due to Kelley [11]. We will see that nets can
be used in general topological spaces just as sequences are used in metric spaces -
to characterise limit points, continuity and compactness (to be studied later).
Example 4.12 1. A net in a discrete space is convergent if and only if it is eventually
a constant. In an indiscrete space, every net converges to every point. Atrocious as
the statement sounds, it is true. Thus the indiscrete topology is anything but pleasant
or useful.
2. Take the directed set Ux of neighbourhoods of a point x in a space X and let
x V ∈ V for V ∈ Ux . The net (x V ) converges to x.
3. Let f be a bounded, real-valued function on the interval [a, b]. For any partition
P of [a, b], consider the Riemann sum s( f, P) (recall?) of f over the subdivision S.
b
The net {(s( f, P)) : P ∈ P[a,b] } converges to the Riemann integral a f (t) dt if f
is Riemann integrable. In the same way, nets of lower and upper Riemann sums can
be considered, converging to the lower and upper Riemann integrals, respectively.
Limit points and closure in metric spaces are characterised using sequences. In
topological spaces this can be done using nets.

Proposition 4.13 Let E be a subset of a space X . Then

1) x ∈ X is a limit point of E if and only if there is a net in E \ {x} which converges
to x;
2) x ∈ Ē if and only if there is a net in E converging to x;
3) E is closed if and only if no net in E converges to an x ∈ E c .

Proof If x is a limit point of E, then for each V ∈ Ux , there is an x V ∈ V ∩ E \ {x}.

We get a net (x V ) converging to x. Conversely, suppose (xα ) is a net in E \ {x}
converging to x. If V ∈ Ux , then there is an α0 such that xα ∈ V for α ≥ α0 . In
particular, V ∩ E \ {x} is nonempty and x is a limit point of E and 1) is proved.
4.1 Limit Points 45

2) Let x ∈ Ē. If x ∈ E, then the constant net (xα = x) (defined on any directed
set) converges to x. Otherwise, part (1) applies. Conversely, if x is a limit of a net
(xα ) in E, then every neighbourhood of x contains some xα and so x ∈ Ē.
The last assertion (3) follows from (2). 

We conclude by looking at the (topological) boundary of a set.

Definition 4.14 For a subset E of a topological space X , define the boundary ∂ E

of E by ∂ E = Ē ∩ X \ E. Thus, x ∈ ∂ E if and only if every neighbourhood of x
intersects both E and its complement.

The basic elementary properties of the boundary are spelt out here.

Proposition 4.15 Let E be a subset of a space X . Then

1) ∂ E = ∂ E c ;
2) Ē is the disjoint union Ē = int E ∪ ∂ E;
3) E is closed if and only if ∂ E ⊂ E;
4) ∂ E is empty if and only if E is both open and closed.

Proof 1) is a consequence of the fact that the definition of boundary of E is symmetric

in E and E c , while (3) follows from (2).
For 2), the inclusion int E ∪ ∂ E ⊂ Ē and the fact that int E and ∂ E are disjoint
are obvious. The inclusion Ē ⊂ int E ∪ ∂ E holds because if x ∈ Ē\ int E, then
every neighbourhood of x meets both E and E c , so x ∈ ∂ E.
Now ∂ E = ∅ = ∂ E c ⇔ E and E c are both closed ⇔ E is both closed and open,
proving (4). 

Example 4.16 1. In R, ∂[0, 1] = {0, 1}, ∂Q = R, ∂Z = Z.

2. In a discrete space ∂ E is empty for all subsets E.
3. If E = {(x, 0) : x ∈ R} ⊂ R2 , then ∂ E = E.
4. The boundary of the unit ball Bn = {(x1 , ..., xn ) ∈ Rn :
x 2j < 1} is the unit
sphere Sn−1 = {(x1 , ..., xn ) ∈ Rn :
x 2j = 1}.
5. If C is the Cantor (ternary/middle-third) set, ∂C = C.
6. In R , ∂[a, b) is empty.
7. In the Sierpiński space S, ∂{1} = {0} = ∂{0}.

4.2 Continuity

Continuity is the central concept in topology. We defined continuity of maps on

topological spaces in Chap. 2, but used it only to introduce initial and final topologies
as examples of topologies. A detailed discussion of continuous maps is imperative
and is being taken up now. Various formulations of continuity are obtained, and the
basic concept of topological equivalence is introduced. Recall the definition.
46 4 Continuous Maps

Definition 4.17 A map f : X → Y from one space to another is said to be contin-

uous if it pulls back open sets to open sets: f −1 (V ) is open in X for every open set
V in Y .
Theorem 4.18 Let X, Y be topological spaces. The following conditions are equiv-
alent for a map f : X → Y :
1) f is continuous.
2) f −1 (F) is closed in X for every closed set F in Y .
3) f −1 (V ) is open in X for every V in some base B for Y .
4) f −1 (U ) is open in X for every set U in a subbase S for Y .
5) For every net (xα ) in X converging to some x ∈ X , the net ( f (xα )) in Y
converges to f (x).
Proof The equivalence of (1) and (2) is a consequence of the definition of conti-
nuity and the fact that taking inverse images preserve complements: X \ f −1 (F) =
f −1 (Y \ F); that of (1) and (3) follows because any open set is a union
 −1 of basic open
sets and unions are preserved under inverse images: f −1 ( Vi ) = f (Vi ). Since
finite intersections of subbasic open sets form a base and intersections are preserved
on taking inverse images, the equivalence of (3) and (4) follows.
To complete the proof, we shall show that (1) implies (5) and (5) implies (2).
Suppose f is continuous and (xα ) is a net in X converging to x ∈ X . If V is a
neighbourhood of f (x) in Y , then f −1 (V ) is a neighbourhood of x in X , and so the
xα eventually lie in f −1 (V ). This means that the f (xα ) ∈ V eventually, so ( f (xα ))
converges to f (x). If (2) fails, f −1 (F) is not closed in X for some closed set F in
Y . Then there is a net (xα ) in f −1 (F) converging to some point x ∈ X \ f −1 (F).
Thus f (xα ) is a net in F and f (x) ∈ / F, so f (xα ) cannot converge to f (x), as F is
closed. Hence (5) fails and the proof is complete. 
Remark 4.19 In general topological spaces, continuity cannot be characterised using
sequential convergence; nets are needed.
Observe that we have defined continuity in topological spaces ‘globally’ unlike
the case of metric spaces where a ‘local’ definition was given. But it is not difficult
to obtain a local formulation in the general setting.
Definition 4.20 Let X, Y be topological spaces. We say that a map f : X → Y is
continuous at x ∈ X if for every neighbourhood V of f (x) in Y , there is a neighbour-
hood U of x in X such that f (U ) ⊂ V , or equivalently, f −1 (V ) is a neighbourhood
of x in X .
Proposition 4.21 Let X, Y be topological spaces, f : X → Y .
1) f is continuous at x ∈ X if and only if f (xα ) → f (x) for every net (xα ) in X
converging to x;
2) f is continuous (on X ) if and only if it is continuous at each x ∈ X .
Proof Part (1) is proved essentially as in the proof of equivalence of (1) and (4) of
the theorem above. Part (2) is easy if one observes that a set is open if and only if it
is a neighbourhood of each of its points. 
4.2 Continuity 47

Example 4.22 Consider R with the cocountable topology Tcc and let Tseq denote
the strictly larger topology of sequentially open sets on R. So the identity map
idR : (R, Tcc ) → (R, Tseq ) is not continuous. However, a convergent sequence in
Tcc is eventually constant and so converges in Tseq as well (and, in fact, in any
topology). Thus idR preserves sequential convergence although it is not continuous.

Here are some expected results whose proofs are all trivial. Nevertheless, these
are recorded here as they are used almost all the time.

Proposition 4.23 1) The identity map on a space is continuous.

2) Restrictions of continuous maps are continuous: if a map f : X → Y between
topological spaces is continuous and E ⊂ X , then the restriction f | E : E → Y is
also continuous.
3) Compositions of continuous maps are continuous: if f : X → Y and g : Y →
Z are continuous maps between topological spaces, then so is g ◦ f : X → Z .

Proof (1) is trivial. Recalling the definition of the subspace topology, (2) follows as
( f | E )−1 (V ) = E ∩ f −1 (V ) for any (open) set V in Y . Lastly, (3) is a consequence
of the equality (g ◦ f )−1 (W ) = f −1 (g −1 (W )) for any (open) set W in Z . 

Often one needs to ‘patch up’ together continuous maps defined on pieces, to get
a continuous map on a bigger domain. Here is one such way of gluing continuous
maps that is very useful.

Lemma 4.24 (Pasting lemma) Let A, B be subsets of a topological space X . Sup-

pose f : A → Y and g : B → Y are continuous maps into a space Y such that
f (x) = g(x) for x ∈ A ∩ B. Define

f (x), x ∈ A
f  g(x) =
g(x), x ∈ B

Then h := f  g : A ∪ B → Y is a well defined map that is continuous provided

either both A and B are open or both are closed.

Proof Since f = g on A ∩ B, h is well defined. Assume now that both A and

B are open and let W be an open set in Y . Then, for x ∈ A ∪ B, the point h(x)
is in W if either x ∈ A and f (x) ∈ W or x ∈ B and g(x) ∈ W . Thus h −1 (W ) =
f −1 (W ) ∪ g −1 (W ). By continuity of f and g, f −1 (W ) is open in A, hence in X
since A is open. Similarly g −1 (W ) is open in X . So their union h −1 (W ) is open as
well. The case when both A and B are closed is treated in the same way, replacing
‘open’ by ‘closed’. 

Example 4.25 1. Consider the function defined on R by


x, x ≤ 0
h(x) =
x 2, x ≥ 0
48 4 Continuous Maps

Then h = f  g where f (x) = x on A = (−∞, 0], g(x) = x 2 on B = [0, ∞), A ∩

B = {0} and f (0) = 0 = g(0). Thus h is continuous on R. The graph of f , the lower
half of the straight line y = x and the graph of g, the right half of the parabola y = x 2
meet at the origin.
2. Let x0 , x1 , x2 be three points in a space X . A path in X from x0 to x1 is
a continuous map α : [0, 1] → X with α(0) = x0 , α(1) = x1 . If we have a path α
from x0 to x1 and β from x1 to x2 , we can combine them to get a path γ joining x0
to x2 defined by 
α(2t), 0 ≤ t ≤ 1/2
α ∗ β(t) =
β(2t − 1), 1/2 ≤ t ≤ 1

Note that α  β = α0  β1 where α0 (t) = α(2t), 0 ≤ t ≤ 1/2, and β1 (t) = β(2t −

1), 1/2 ≤ t ≤ 1. The pasting lemma shows that γ is a path from x0 to x2 . This will
be used in the chapter on homotopy.

Definition 4.26 A homeomorphism is a bijective continuous map f : X → Y

between spaces with a continuous inverse f −1 : Y → X . X is homeomorphic to
Y if there is a homeomorphism from X to Y . Thus two mutually homeomorphic
spaces are the ‘same as topological spaces but for labelling’. A homeomorphism is
also called a topological equivalence.

Proposition 4.27 ‘Homeomorphism’ is an equivalence relation.

Proof The identity map on any space, the inverse of a homeomorphism and the
composition of two homeomorphisms are all homeomorphisms. 

Example 4.28 1. If either X is discrete or if Y is indiscrete, any map f : X → Y is

continuous. If X is not discrete (or Y is not indiscrete), then there is a discontinuous
f : X → Y for a suitable Y (or X , respectively).
2. Suppose T1 and T2 are topologies on X . Then the identity map (X, T1 ) →
(X, T2 ) is continuous if and only if T1 is stronger than T2 ; it is a homeomorphism if
and only if T1 = T2 .
3. Any constant map f : X → Y is continuous.
4. The map x → ax + b is a homeomorphism on R for a, b ∈ R, a = 0.
5. For a, b, c, d ∈ R, the map R2 → R2 , (x, y) → (ax + by, cx + dy) is contin-
uous; it is a homeomorphism if ad − bc = 0.
6. f : R → (−1, 1), f (x) = 1+|x| x
, is a homeomorphism. It also gives a homeo-
morphism of the extended real line R̄ onto [-1,1].
7. In Rn , the three sets {(x1 , ..., xn ) :
x 2j = 1}, {(x1 , ..., xn ) :
|x j | = 1} and
{(x1 , ..., xn ) : max |x j | = 1} are mutually homeomorphic. Note that these are the
4.2 Continuity 49

unit spheres with metrics d2 , d1 and d∞ respectively. (See pictures in Chap. 2.) The
corresponding open balls Bd2 (0, 1), Bd1 (0, 1) and Bd∞ (0, 1) are also homeomorphic.
8. Consider an ellipse in R2 centred at the origin. Let A(x) denote the area enclosed
by the ellipse in the half plane {(ξ, η) ∈ R2 : ξ ≤ x}. A : R → R is a continuous
function.
9. For real numbers a, b, a < b, (a, b) = ∪∞ 1 [a + n , b). Consequently, (a, b) is
1

open in R also. So the standard topology is smaller than the lower-limit topology
on R. The identity map R → R is therefore a continuous bijection that is not a
homeomorphism.
10. Let A be an n × n real matrix. The map x → Ax on Rn given by matrix
multiplication is continuous. It is a homeomorphism if and only if A is nonsingular.
If A is orthogonal it gives a self-homeomorphism of the sphere Sn−1 , by restriction.
11. The determinant and trace are continuous functions on the matrix space
2
M(n, R) (with the Euclidean metric of Rn ) and the transpose map A → At is an
auto-homeomorphism of M(n, R).

Exercise 4

1. Directed sets. a) Let d be a metric on X and fix p ∈ X . Define x < y if

d(y, p) < d(x, p). This makes X \ p a directed set. In particular, Rn \ 0 is a directed
set with x < y if y < x.
2. Uncountable sets of real numbers. Let E ⊂ R be uncountable. Then a) E
has a limit point (how many limit points does it have?);
b) there is a sequence of distinct points converging to a point in E;
c) E ∩ (−∞, a), E ∩ (a, ∞) are both uncountable for some a ∈ R.
3. Derived sets. a) Find E  , E  and all the successive derived sets of E in R
when E = { m1 + n1 : m, n ∈ N}, E = {m + nπ : m, n ∈ Z}.
b) (A ∪ B) = A ∪ B  .
4. Boundaries. a) int (A ∪ B) = int A ∪ int B if ∂ A ∩ ∂ B = ∅.
b) int E = E \ ∂ E, E\ int E = ∂ E, and X = int E ∪ ∂ E ∪ int E c .
c) In R, find a subset E such that ∂ E = [0, 1].
d) The boundary of an open set has empty interior and conversely, every closed
set with empty interior is the boundary of an open set.
e) In Rn , every closed set is the boundary of some set.
5. Continuous functions. a) If f, g are real continuous functions on a space X ,
so are f + g, f g, α f, α ∈ R, max{ f, g} and min{ f, g}.
b) Matrix product M(n, R) × M(n, R) → M(n, R) continuous.
c) Find a continuous bijection of [0, 1) onto S1 that is not a homeomorphism. Can
there be a homeomorphism between them?
d) Let U be a nonempty, proper, open subset of Rn . Find a real function f on Rn
such that f is continuous at x if and only if x ∈/ U.
e) A real valued function on R that is right continuous is R → R continuous. Is
the converse true?
50 4 Continuous Maps

f) If d is a metric on X , then the d-topology is the smallest topology on X that

makes d continuous.
g) In any space X , the set of points of discontinuity of χ E is ∂ E.
h) Consider the area enclosed between the x-axis and the Gaussian curve y =
e−x , x ∈ R. Let A(x) denote the part of this area that lies in the half plane {(ξ, η) ∈
2

R : ξ ≤ x}. A : R → R is continuous.
2

i) Consider the unit sphere S2 = {(x, y, z) ∈ R3 : x 2 + y 2 + z 2 = 1}. Let A(t) be

the area of the region obtained by the projection, on the y-z plane, of the spherical
cap of S2 cut by the plane x = t, 0 ≤ t ≤ 1. A : [0, 1] → R is continuous.
6. Continuous maps. a) f : X → Y is continuous if and only if f ( Ā) ⊂
f (A) ∀A ⊂ X if and only if f −1 (B) ⊂ f −1 ( B̄) for all B ⊂ Y .
c) If f : X → Y and g : X → Z are continuous, then f × g : X → Y × Z
defined by f × g(x) = ( f (x), g(x)) is continuous. (Note that parametric equations
of curves in the plane are of this form, e.g. γ (t) = (a cos t, b sin t) is the parametric
equation of the ellipse.)
d) If f : X → Y , the set G f = {(x, f (x)) : x ∈ X } ⊂ X × Y is called the graph
of f . If f is continuous, the map x → (x, f (x)) is a homeomorphism of X onto
G f . Is the converse true?
e) Let E ⊂ X. If f : X → Y is a map such that the restriction f | E is continuous,
is it continuous at each point x ∈ E (as a map on X )?
f) Let X, Y be metric spaces, P ∈ X and d P the French railway metric. A map
f : X → Y is d P - continuous if and only if f is d X -continuous at P.
g) For any (X, T ), C(X, S) = {χU : U ∈ T }. (S = Sierpiǹski space.)
h) Let X be a vector space with a norm (see Exercise in Chap. 2). A linear map
T on X is continuous if and only if there is a constant c such that T x ≤ cx for
all x ∈ X .
i) If f, g : X → Y are continuous and f (x) = g(x) for all x in a dense set D ⊂ X ,
is it true that f = g?
j) Determine all continuous maps on X when X is i) any set with the cofinite
topology; ii) Z with Tap ; iii) N as in Ex. 1 d) in Chap. 3.
k) Let {E i } be a collection of open sets in X with X = ∪E i . If f : X → Y is such
that f | Ei is continuous for each i, then f is continuous. The corresponding statement
is false for closed sets. However, if {E i } is a locally finite family (3.22) of closed
sets, then the continuity of each f | Ei implies continuity of f .
l) If every map f from X to any Y is continuous, then X is discrete. What is the
analogous statement for indiscrete spaces?
7. Homeomorphisms. a) f : X → Y is a homeomorphism if and only if it is
bijective and f ( Ā) = f (A) for every A ⊂ X if and only if f −1 (B) = f −1 ( B̄) for
all B ⊂ Y .
b) Classify all intervals in R up to homeomorphisms.
c) The open ball B(0; 1) in Rn is homeomorphic to Rn .
d) Determine all the homeomorphisms f of R that satisfy the condition f (x +
y) = f (x) + f (y) for all x, y ∈ R.
e) Any two nonempty open convex sets in R2 are homeomorphic.
4.2 Continuity 51

f) Find homeomorphisms between i) the punctured plane and the exterior of the
closed unit disc; ii) the open unit disc and the open upper half plane. (Complex
numbers may make this less complex!)
g) For any space X , the set Aut(X ) of all homeomorphisms X onto itself is a
group under composition. If X and Y are homeomorphic, then Aut(X ) and Aut(Y )
are isomorphic.
h) If X = [0, 1] and Y = (0, 1), then the restriction map h → h|Y is an isomor-
phism Aut(X )  Aut(Y ).
i) Find Aut(X ) for the discrete space {0, 1} and the space S (3.2).
j) If there is a bijective continuous map from X onto Y and another from Y onto
X then, X and Y need not be homeomorphic.
k) X is homeomorphic to the diagonal (X ) = {(x, x) : x ∈ X }.
l) Suppose X, Y are countable disjoint unions of open sets, X = X n , Y = Yn .
If X n is homeomorphic to Yn for each n, then X  Y .
m) Does a homeomorphism of metric spaces preserve Cauchy sequences?
n) Any homeomorphism of the sphere Sn−1 extends to a homeomorphism of the
ball Bn .
8. Isometries. a) An isometry (on a metric space) is a homeomorphism of X onto
f (X ). Find a homeomorphism that is not an isometry.
b) The set of all isometries of a metric space X onto itself is a group Iso(X ) under
composition.
c) Any isometry of R is of the form x → ax + b, a = ±1, b ∈ R.
d) On Rn , orthogonal transformations and translations are isometries. (It is a fact
that Iso(Rn ) is generated by these; this group is called the group of rigid motions of
Rn or the motion group of Rn .)
e) Let X be a set and (Y, dY ) be a metric space. Suppose σ : X → Y is a bijection.
Define dσ (x1 , x2 ) = dY (σ x1 , σ x2 ), x1 , x2 ∈ X . Then dσ is a metric on X and σ is an
isometry. If X already had a metric d X , when is dσ equivalent to d X ?
9. Semi-continuity. a) If { f i } is a family of bounded real-valued continuous
functions on a space X , then supi f i and inf i f i may not be continuous. This leads
to the concept of semi-continuous functions (due to Baire). A function f : X →
(−∞, ∞] is lower semi-continuous (lsc) if f −1 ((a, ∞]) is open in X for all a ∈ R;
f is upper semi-continuous (usc) if − f is lsc. The supremum (infimum) of any
family of lsc (usc) functions is lsc (usc).
b) If E is a subset of a space X , when is the characteristic function χ E continuous?
(χ E (x) = 1 if x ∈ E and χ E (x) = 0 if x ∈ E c .) Are there such sets in R? When is
it lower (upper) semi-continuous?
c) Every lsc function is continuous on a dense G δ [12].
10. Sequences. a) All sequences in the Sierpiǹski S (3.2) converge to 0 and the
constant sequence (1, 1, 1, · · · ) converges to both 0 and 1.
52 4 Continuous Maps

b) Discuss convergence of sequences in R with the lower-limit topology, cofi-

nite topology, cocountable topology and the topology generated by sets of the form
(a, ∞), a ∈ R.
11. Topological groups. A group G with a topology on it is called a topological
group if the group multiplication G × G → G and the group inversion G → G are
continuous. (For a detailed treatment, see [13].)
a) Examples: (Rn , +); R∗ and T = {z ∈ C : |z| = 1} with multiplication; G L(n,
R) with matrix multiplication; any subgroup of a topological group; any group with
the discrete topology.
b) Let G be a topological group with identity e and a ∈ G. Then
i) the left (right) translation x → ax (x → xa) are homeomorphisms;
ii) if A, B are open subsets, so is AB = {ab : a ∈ A, b ∈ B};
iii) AB may not be closed if A, B are closed;
iv) U is a neighbourhood of e if and only if xU , U x are those of x;
v) if U is a neighbourhood of e then so is U −1 = {x −1 : x ∈ U } and V = U ∩ U −1
is a symmetric neighbourhood of e, i.e. V = V −1
c) Let H be a subgroup of a topological group. Then
i) H̄ is a subgroup and is a normal subgroup if H is;
ii) if int H is nonempty, then H is open;
iii) H is closed if it is open;
iv) if V is a symmetric neighbourhood of e, then H = ∪∞ n
1 V is an open subgroup.
(Here V = V V and inductively, V
2 n+1
= V V .)
n

d) For any A ⊂ G, Ā = ∩{AU : U ∈ Ue }.

e) A homomorphism between two topological groups is continuous if it is con-
tinuous at e. Any continuous homomorphism R → R is of the form f a (x) = ax for
some a ∈ R, so is either 0 or is an automorphism.
f) The centre of G is a closed normal subgroup.
g) Check whether the following are topological groups: i) R with the usual
addition of real numbers; ii) R with the cofinite topology; iii) Z with the arithmetic
progression topology.
12. Subgroups of R. a) Suppose H is an additive subgroup of R.
i) If H = {0} has a smallest positive element a then H is the infinite cyclic group
aZ = {na : n ∈ Z} generated by a and so is discrete;
ii) if H is not discrete , then H is dense in R (e.g. Q);
iii) if H is closed, either it is R or is of the form aZ for some a ∈ R.
b) What are the open subgroups of R?
13. Uniformities on topological groups. Let G be a topological group with local
base Be at the identity. For B ∈ Be , let L B = {(x, y) : x ∈ y B} and R B = {(x, y) :
x ∈ By}. These sets form bases for uniformities Ul and Ur , called the left and right
uniformities. G has equivalent uniform structures if Ul = Ur . (e.g. G is abelian or
compact.)
14. Filters and their convergence. Henri Cartan of France introduced and
developed the theory of filters in 1937 [14] as an alternative approach to convergence
4.2 Continuity 53

in topological spaces. (Filter bases were earlier considered by Vietoris (1931) and
Garrett Birkhoff (1935)).
A filter on a set X is a nonempty collection F of nonempty subsets of X such
that E ∩ F ∈ F whenever E, F ∈ F and F ∈ F , F ⊂ E implies E ∈ F . A filter
base E in X is a subcollection of F such that F = {F ⊂ X : ∃E ∈ E , E ⊂ F}.
(References for filters: [15, 16].)
a) Examples. i) For any x in a topological space X , the neighbourhood system
Ux at x is a filter, called the neighbourhood filter a x. A local base Bx is a filter base
for Ux . (This is the motivating example.) More generally, for any fixed E ⊂ X , the
neighbourhood filter U E at E consists of subsets A of X with E ⊂int A.
ii) For any set X and E ⊂ X , {F ⊂ X : E ⊂ F} is a filter with filter base {E}, called
the principal filter generated by E.
iii) On an infinite set X , subsets of X with finite complements form a filter. When
X = N, this is called the Fréchet filter. For n ∈ N, if Fn = {k ∈ N : k ≥ n}, then {Fn }
is a filter base for this filter.
iv) Find all filters on {1, 2, · · · , n}.
b) Convergence. We say that a filter F in a space X converges to x ∈ X , written
F → x, if Ux ⊂ F .
i) In a metric space, a convergent filter has a unique limit.
ii) In a discrete space X , F → x if and only if x ∈ F for all F ∈ F .
iii) If F is the filter formed by complements of finite sets in X , for what x does
F → x in the cofinite topology of X ?
iv) Let T1 , T2 be topologies on a set X . Then T1 is finer than T2 if and only if for any
filter F → x in T1 , F → x in T2 .
c) Closure. Let E be a subset of a space X . Then x ∈ Ē if and only if there is a
filter F with E ∈ F such that F → x.
d) Continuity and filters. If f : X → Y and F is a filter on X , let f (F ) denote
the filter with filter base all sets of the form f (F), F ∈ F. f is continuous if and
only if a filter F in converges to x ∈ X implies f (F ) → f (x) in Y .
15. Nets vs Filters. Convergence theories of net and filters are ‘equivalent’.
If (xα ) is a net in X , the associated filter is the filter generated by the filter base
consisting of sets Bβ := {xα : α ≥ β}. Conversely, if F is a filter on X , then DF =
{(x, F) : x ∈ F ∈ F } is a directed set with order (x1 , F1 ) ≤ (x2 , F2 ) if F2 ⊂ F1 .
The net ν in X defined by ν(x, F) = x is called the net based on F .
a) A net (xα ) converges to x if and only if the associated filter converges to x.
b) A filter F converges to x if and only if the net based on F converges to x.
(See, for instance, [16, 17]).
54 4 Continuous Maps

4.3 Biographical Notes

4.3.1 Cauchy

Augustin-Louis Cauchy (1789–1857), the French mathematician, was one of the

all time greats in Mathematics. He was one of the most prolific mathematicians of
all times, perhaps next only to Euler. He published more than 800 papers and several
books. His collected works run into 27 volumes. He studied engineering, started his
career as a civil engineer and took up a career in mathematics later. But he began his
mathematical research even when he was an engineer. Areas where he made signifi-
cant contributions include complex function theory, real analysis, group theory, dif-
ferential equations, astronomy, hydrodynamics, optics, elasticity, etc. He was the first
to introduce some form of rigour into mathematics. The concept of limit, continuity,
convergence of series, multiplication of infinite series, the first adequate definition of
definite integrals, improper integrals, the Jacobian, are among the concepts in analysis
introduced by him. Convergence criterion for series, tests for convergence, existence
of lim(1 + 1/n)n , the first rigorous proof of Taylor’s theorem, a mean value theorem
are some of his contributions to real analysis. His Cours d’Analyse had a major impact
in the understanding of Mathematical Analysis. Abel called it “an excellent work
which should be read by every analyst who loves mathematical rigour.” He can justi-
fiably be called the founder of Complex Analysis. The Cauchy–Riemann equations,
contour integrals, Cauchy’s fundamental integral theorem for analytic functions, the
integral formula, residue at a singularity, the residue theorem lie at the heart of the
subject. He also gave proofs of the fundamental theorem of algebra. His work on
multi-valued functions “shed a clear light of understanding upon those functions”
until Riemann’s work on Riemann surfaces appeared. He made important contri-
butions to determinants, group theory, Fermat’s conjecture on polygonal numbers,
differential equations and the mathematical theory of elasticity. Results and concepts
named after him are too numerous to be fully listed here. Here are some of them:
Cauchy criterion for convergence, Cauchy’s root test, Cauchy’s condensation test,
Cauchy’s integral test, Cauchy product (of series), Cauchy’s mean value theorem.
Cauchy’s principal value integral, Cauchy–Schwarz inequality, Cauchy–Riemann
equations, Cauchy–Hadamard formula for power series, Cauchy’s integral theorem,
Cauchy’s integral formula, Cauchy’s residue theorem, Cauchy’s argument principle,
Cauchy–Euler equation, Cauchy problem (in PDE), Cauchy–Kovalevskaya theorem,
Cauchy momentum equation, Cauchy–Peano theorem (ODE), Cauchy’s theorem on
groups, Cauchy distribution. Cauchy also was infamously known for rejecting and
losing the epoch-making work of the teen-aged French supernova Evariste Galois
on polynomial equations and that of the young Norwegian great Niels Henrik Abel
on elliptic functions submitted to the Paris Academy of Sciences. (By the time their
manuscripts were found and their worth recognised, both Galois (at 20) and Abel (at
26) had died.)
4.3 Biographical Notes 55

4.3.2 Weierstrass

Karl Weierstraß (1815–1897), usually written in English as Weierstass, of Ger-

many is widely accepted as the father of modern analysis, one of those responsible
for introducing rigour into analysis. He was one of the founding triumvirate of com-
plex function theory and developed the subject by means of power series. (The other
two are Cauchy, with his integrals, and Riemann with his geometric approach). He
had his schooling in several towns, his father being a government official. He had to
take up a part time job to augment the family income, yet he excelled in his studies
at the gymnasium at Paderborn. In 1834 his father put him at the University of Bonn
to study law, finance and economics, in preparation for a government job. This was
in conflict with his mathematical interests and affected him deeply, resulting in his
quitting the University after four years without a degree. Later he qualified as a sec-
ondary school teacher, teaching in several places for 14 years. A heavy work load
of teaching mathematics, physics, botany, geography, history, German, calligraphy,
gymnastics, etc. could not curtail his enthusiasm for mathematics and he continued
to study the works of Laplace, Jacobi, Abel and others. He got interested in elliptic
functions from the lectures of Christoph Gudermann at Münster. With no one to dis-
cuss with, with no access to a library and lacking the means to correspond with other
mathematicians, he still carried out research on his own without publishing. His ill-
health continued to affect him, with frequent attacks of severe dizziness. Finally, he
published a part of his research on abelian functions in Crelle’s Journal in 1854. This
instantly catapulted him into stardom and the University of Königsberg conferred an
honorary doctorate on him. He had offers of positions from several universities and,
in 1856 (the year in which the full version of his work on abelian functions appeared
in Crelle’s Journal), took up a chair at the University of Berlin. Serious attacks on his
health forced him to lecture sitting down December 1861 onwards, with a student
writing for him on the blackboard. He published very little, but most of his work
was disseminated through his lectures at Berlin. He gave several courses there—on
analysis, analytic functions, mathematical physics, Fourier series, elliptic functions,
abelian functions and integrals, calculus of variations, etc., and these attracted stu-
dents from far and wide. (e.g. Georg Cantor, Ferdinand Georg Frobenius, Kurt
Hensel,Otto Hölder,Adolf Hurwitz,Wilhelm Killing, Felix Klein, Sophus Lie,
Hermann Minkowski, Gösta Mittag-Leffler, Carl Runge, Arthur Schönflies,
Friedrich Schottky, Hermann Schwarz, Friedrich Engel, Sofia Kovalevskaya)
His lectures were published based on the notes of students—Killing, Hurwitz and
others. The first two volumes of his collected work were published under his own
supervision in 1894 and 1895. The other volumes were published after his death,
the seventh appearing in 1927. All the volumes have been republished since. He was
confined to a wheel chair during the last three years of his life and he died of pneu-
monia in 1897. Weierstrass’ definition of limit and continuity, Weierstrass Approx-
imation Theorem, Weierstrass M-Test for series of functions, Bolzano–Weierstrass
theorem, Weierstrass continuous nowhere differentiable function, Weierstrass the-
orem on analytic functions, Weierstrass factorisation theorem on entire functions,
56 4 Continuous Maps

Casorati–Weierstrass theorem, Weierstrass preparation theorem, Weierstrass elliptic

functions are all famous.
“Ce qui me frappe dans la carrière mathématique de Weierstrass, c’est la remar-
quable unité de la pensée, persistant à travers l’étendue et la variété de son oeuvre”,
says Poincaré, [18]. (“What strikes me about Weierstrass’s mathematical career is
the remarkable unity of thought, persisting through expanse and the variety of his
work.”)

4.3.3 Dirichlet

Lejeune Dirichlet (1805–1859) is considered as one of the nineteenth century greats

in mathematics. He studied in Paris for some time, but spent most of his career in Ger-
many. His academic mentors included Joseph Fourier, Siméon Lapalace and Karl
Friedrich Gauss. His main area of interest was Number Theory, and his favourite
work was Gauss’s Disquisitiones Arithmeticae. His doctoral students included Got-
thold Eisenstein and Leopold Kronecker. His first research was on Fermat’s Last
Theorem for n = 5 on which he lectured at the Paris Academy of Science at the
age of 20! Analytic Number Theory was essentially his creation. His theorem on
arithmetic progressions, Dirichlet characters, Dirichlet L-functions, Dirichlet series,
Dirichlet’s unit theorem, Dirichlet’s Diophantine approximation theorem, the class
number formula, biquadratic reciprocity law are some of his important basic contri-
butions to number theory. Dirichlet kernel and Dirichlet’s convergence theorem on
Fourier series, Dirichlet’s test for series convergence, Dirichlet problem and Dirich-
let principle are his notable contributions to analysis and differential equations. He
is also considered as one of the first to give a proper definition of a ‘function’. He
introduced the concept of uniform continuity and proved, in his lectures, that a con-
tinuous function on a closed interval is uniformly continuous (a result later published
by Heine). Dirichlet distribution and process in probability theory are named after
him. His lectures were known for their clarity and attracted many mathematicians,
including Richard Dedekind, Riemann and Heine.

References

1. Bolzano, B.: Rein analytischer Beweis des Lehrsatzes, dass zwischen je zwei Werthen, die
ein entgegengesetztes Resultat gewähren, wenigstens einer reelle Wurzel der Gleichung liege.
Prague, Gottlieb Haase (1817)
2. Cauchy, A.L.: Cours d’Analyse de l’École Royale Polytechnique Paris (1821)
3. Weierstrass, K.: Über continuirliche Functionen eines reellen Arguments, die für keinen Werth
des letzteren einen bestimmten Differentialquotienten besitzen, Königl. Akad. der Wiss.,
Berlin, Werke 2, pp. 71–74 (1872)
References 57

4. Weierstrass, K.: Theorie der analytischen Funktionen, Vorlesung an der Univ. Berlin,
1874, manuscript (ausgearbeitet von G. Valentin), Math. Bibl. Humboldt Universität Berlin,
Valentin) (1874)
5. Freudenthal, H.: Did Cauchy plagiarize Bolzano? Arch. Hist. Exact Sci. 7(5), 375–392 (1971).
https://doi.org/10.1007/BF00327099
6. Moore, E.H., Smith, H.L.: A General Theory of Limits. Am. J. Math. 44, 102–121 (1922)
7. Moore, E.H.: General Analysis I. Part II. Mem American Philosphical, Society (1939)
8. Vietoris, L.: Stetige Mengen. Monatsh Math. und Physik 31, 173–204 (1921). https://doi.org/
10.1007/BF01702717
9. Reitberger, H.: Leopold Vietoris, 1891–2002. Not. Am. Math. Soc. 49(10), 1232–1236 (2002)
10. Birkhoff, G.: Moore-Smith convergence in general topology. Ann. Math. 38, 39–56 (1937).
https://doi.org/10.2307/1968508
11. Kelley, J.L.: General Topology. Affiliated East-West Press (1969)
12. Engelking, R.: General Topology Revised Edition. Heldermann Verlag (1989)
13. Hewitt, E., Ross, K.A.: Abstract Harmonic Analysis I. Springer (1963)
14. Cartan, H.: Théorie des filtres. C.R. Acad. Sci. Paris 205, 595–598 (1937)
15. Bourbaki, N.: General Topology, Chapters 1–4. Springer (1995)
16. Willard, S.: General Topology. Addison-Wesely (1970)
17. Bartle, R.G.: Nets and filters in topology. Am. Math. Monthly 62, 551–557 (1955). https://
doi.org/10.1080/00029890.1955.11988688
18. Poincaré, H.: L’oeuvre mathématique de Weierstrass. Acta Math. 22, 1–18 (1899). https://
doi.org/10.1007/BF02417867
Chapter 5
Compact Spaces

Compactness is, doubtless, the most important topological property not only for
analysis (see, e.g., [1]) but also in topology and geometry. Fréchet [2], who was the
first to define and study compactness (in metric spaces) expressed the importance of
compactness this way:
‘Nous avons déjà signalé et nous reconnaîtrons dans tout le cours de ce Livre
l’importance des ensembles compacts. Tous ceux qui ont eu à s’occuper d’Analyse
générale ont vu qu’il était impossible de s’en passer.’ (‘We have already noted, and
will recognize throughout this book, the importance of compact sets. All those who
are involved with general analysis will see that it is impossible to do without them’.)
For general topological spaces, Vietoris [3] gave the first definition thus: Eine
Menge M heiße lückenlos, wenn in ihm jede geordnete Teilmenge ohne letztes Ele-
ment mindestens einen zu M gehörigen Grenzpunkt hat. (A space is compact (lück-
enlos = ‘without gaps’) if every filter base (‘kranz’) has a cluster point, in present
terminology.)

5.1 Compactness in Rn

We begin with the classical setting of the real line, and thence to the analogous
Euclidean space setting, before moving to metric spaces and general topological
spaces. Recall the basic properties of real continuous functions f on a closed inter-
val [a, b]: (i) f is bounded and has a maximum and minimum; (ii) f is uniformly
continuous. Both these properties are of central importance. What is the underly-
ing attribute of [a, b] that leads to these properties? The fundamental characteristic
behind this nice behaviour is manifested in Borel’s remarkable result in his thesis
[4]:
Si l’on a sur une droite une infinitè d’intervalles partiels, tels que tout point de la
droite soit intérieur à l’un au moins des intervalles, on peut déterminer effectivement
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 59
K. Parthasarathy, Topology, La Matematica per il 3+2 134,
https://doi.org/10.1007/978-981-16-9484-4_5
60 5 Compact Spaces

un nombre limité d’intervalles choisis parmi les intervalles donnés et ayant la même
propriété- tout point de la droite est intérieur à au moins l’un d’eux. (“If on a line one
has an infinite number of subintervals, such that every point of the line is interior to
at least one of the intervals then one can determine effectively a bounded number of
intervals from among the given intervals that have the same property -every point of
the line is interior to at least one of them.” [5]) This result is remarkable, not only for
its content but also for its formulation. Its conclusion, in current terms, has become
the modern definition of compactness.

Theorem 5.1 (Borel’s theorem, 1894) Let {Vi } be a collection of open sets in [a, b]
with the property that every point of the interval is in some Vi . Then there is a finite
subcollection of {Vi } with the same property : every point of [a, b] belongs to at
least one of them. In other words, every open cover of [a, b] has a finite subcover.

Proof We present a classical proof by the method of bisection. Assume that, contrary
to the assertion of the theorem, there is an open cover {Vi } of [a, b] having no
finite subcollection that also covers I0 = [a, b]. Then this is also an open cover for
each of the intervals [a, c], [c, b], where c = a+b 2
is the midpoint of the interval. A
finite subcover for each together would give a finite subcover for the whole interval,
so for at least for one of these, call it I1 = [a1 , b1 ], there is no finite subcover.
Repeat the argument for I1 in place of I0 and choose one of the halves of I1 for
which {Vi } has no finite subcover; call it I2 = [a2 , b2 ]. Induction yields a sequence
Ik = [ak , bk ] of intervals with the following properties: Ik+1 ⊂ Ik , a ≤ ak ≤ ak+1 ≤
bk+1 ≤ bk ≤ b and bk − ak = b−a 2k
for every k. Then {ak }, {bk } are both monotonic,
bounded sequences and so have limits α = lim a k = sup ak ≤ β = lim bk = inf bk .
But bk − ak → 0 and so α = β. In other words, Ik = {α}. Now, α ∈ V j for some
j, V j is open and α = lim ak = lim bk . Thus the ak and bk eventually belong to V j ;
that is, there is a k0 such that Ik ⊂ V j for k ≥ k0 . This means a single V j covers these
Ik , contradicting the choice of the Ik . This contradiction completes the proof. 

Remark 5.2 i) Here the method of bisection has been used and completeness of the
real line is the crucial ingredient, appearing in the form of Cantor’s nested interval
theorem which has been obtained in the course of the proof. This theorem says that
a decreasing sequence of closed intervals in R has nonempty intersection, and so if
the lengths of the intervals decrease to zero, then the intersection is a single point. A
proof using the equivalent lub, glb properties can be given (as was done in the proof
of the intermediate value theorem; bisection method can be used to prove that result
too.)

ii) In the theorem, it is immaterial whether the Vi are assumed to be open in [a, b]
or in R (Why?).
Definition 5.3 A topological space X is said to be compact if every collection of
open sets covering it has a finite subcollection which also covers it; i.e. every open
cover for X has a finite subcover.
5.1 Compactness in Rn 61

The history of compactness and related properties in the real line is complicated and
we will mention only a few salient points. (For more details, see Manya Raman-
Sundström [5] and Andre et al. [6].) Borel stated his result 5.1 only for countable
covers and the general form is usually credited to the famous French mathematician
Henri Lebesgue, who gave a short elegant proof of the general result. Almost at the
same time as Borel, Pierre Cousin (of France, 1867–1933) had also proved (1895)
the general form of Borel’s theorem (5.1). Perhaps the result may be called the
Borel–Cousin–Lebesgue theorem, but there are several other names involved; see
the references mentioned above.
The proof of compactness
 of intervals [a, b] easily extends to give compactness
of ‘rectangular boxes’ n1 [a j , b j ], in Rn .

Theorem 5.4 Any box n1 [a j , b j ] in Rn is compact.

Proof Here is a sketch; details to be filled in by the reader, imitating

 the proof for
[a, b]. Assume that there is an open cover {Vi } of the box I0 = [a j , b j ] having no
finite subcover. Bisect each [a j , b j ] giving rise to 2n subboxes at least one of which,
say I1, is not covered by any finite
 subfamily. Inductively get a decreasing sequence
Ik = [a kj , bkj ] with ‘volume’ (bkj − a kj ) decreasing to zero, none of which can be
covered by any finite subcollection. The intersection of all these boxes is a single
point x = (x1 , ..., xn ) and any Vi containing x covers all but finitely many Ik , giving
the sought after contradiction. 

The stage is now set for presenting the famous characterisation of compact subsets
of Rn . It is essentially a consequence of Borel’s theorem. Heine’s name had been
mistakenly associated with the theorem. To quote from [5]: “While Heine is credited
with a theorem he did not prove, it appears that Cousin was largely overlooked for
a lemma he did prove.” (© Mathematical Association of America, 2015. All rights
reserved. Quoted here with permission.) But the name Heine–Borel theorem has now
become entrenched, though many French texts call it Borel–Lebesgue theorem. As
a still incorrect compromise, we combine all the three names.

Theorem 5.5 (Heine–Borel–Lebesgue theorem) A subset K of Rn is compact if

and only if it is closed and bounded.

Proof Boundedness is with respect to the Euclidean metric d2 (or, equivalently, the
‘box’ metric d∞ ). Thus a set K in Rn is bounded if and only if it is contained in some
Euclidean ball or in some box.
If K is not bounded, there are points of K outside any ball and so the open cover
{B(0; k) : k = 1, 2, ...} does not have a finite subcover. Thus K is not compact. On
the other hand, if K is not closed, there is a limit point x0 of K with x0 ∈ / K . Then
every ball B(x0 ; k1 ), k = 1, 2, ... contains points of K and the complements of the
closed balls B̄(x0 ; k1 ) form an open cover for K without any finite subcover. Thus
again K is not compact.
Conversely, suppose K is closed and bounded. It is then a closed subset of some
large enough box B. If {Vi } is an open cover for K , then the Vi together with the
62 5 Compact Spaces

open set K c is an open cover for the set B which is compact by the earlier result.
A finite subcover obtained from this compactness also covers K and the theorem is
proved. 

Example 5.6 1. R, (0, 1), (a, ∞), (−∞, b] are not compact.
2. Closed balls and spheres in Rn are compact; half planes in R2 and hyperboloids
in R3 are not.
3. The Cantor ‘middle-third’ (or ‘ternary’) set in R is compact.
4. The set {(a,
 b,  c, d) ∈ R4 : ad − bc = 1} is noncompact, whereas
a b
{(a, b, c, d) ∈ R4 : is an orthogonal matrix} is compact.
cd
5. The previous example easily generalises. The space S L(n, R) of real n × n
matrices of unit determinant is noncompact and the set O(n) of orthogonal matrices
is closed and bounded, hence is compact.

5.2 Compactness in Metric Spaces

Here is the first ever definition of compactness, although it was given in the context
of metric spaces. Fréchet’s definition gives what we now call ‘limit point compact
spaces’.
Nous dirons qu’un ensemble est compact lorsqu’il ne comprend qu’un nombre
fini d’éléments ou lorsque toute infinité de ses éléments donne lieu à au moins un
élément limite. (Fréchet, [7])
(‘We say that a set is compact if it is finite or any infinitely many of its elements
has at least one limit point.’)
The open cover definition of compactness is perhaps a bit mysterious and is
hardly appealing to intuition. In the case of Rn , the Heine–Borel–Lebesgue theorem
gives a more tangible characterisation of compact sets. In the case of metric spaces,
Fréchet’s formulation and other, intuition friendly and simpler to apply, formulations
are equivalent to the open cover definition.
Recall: (i) Every bounded sequence of real numbers has a convergent subsequence;
(ii) every bounded infinite set of real numbers has a limit point. If a subset K of R
is compact, then it is bounded and closed and so (1) every sequence in K has a
subsequence convergent in K and (2) every infinite subset of K has a limit point in
K . It turns out that each of these properties characterises compactness in arbitrary
metric spaces, not just in R. Thus we are led to the following convenient definitions.

Definition 5.7 A topological space X is said to be (i) sequentially compact if every

sequence in X has a convergent subsequence; (b) Bolzano–Weierstrass compact or
limit point compact if every infinite subset of X has at least one limit point. A space
has the Bolzano–Weierstrass property if any infinite bounded set has a limit point.
5.2 Compactness in Metric Spaces 63

Bolzano’s result on the greatest lower bound property enabled Weierstrass to prove
that infinite bounded sets of real numbers have limit points (Bolzano–Weierstrass
theorem). Fréchet used this property to define limit point compactness for metric
spaces.
Our first objective is to prove the equivalence of each of these with compactness in
metric spaces. It is not difficult to get the mutual equivalence of these two properties
and the implication that these properties are implied by compactness. The hard part is
to prove that sequentially compact metric spaces are compact. In the process we also
encounter another important and interesting property for subsets of metric spaces that
is stronger than boundedness. We take this up now, beginning with some examples.
Example 5.8 1. Consider an interval (a, b), a, b ∈ R, a < b. It is of course bounded,
but there is something more. To see this, let ε > 0 and choose a positive integer n
such that b−an
< ε and subdivide the interval into n equal subintervals. Then each
subinterval is of length less than ε. Another way of stating this is that there are finitely
many points (namely, the points of subdivision) such that every point of the whole
interval is at a distance less than ε from at least one of these points. (It is immaterial
whether the interval is open, closed, half-open or half-closed for this discussion.)
2. On the other hand, consider an infinite set with discrete metric. The distance
between two distinct points is always one and the phenomenon described in the
previous example is far from true here.
3. For a nontrivial space of the kind encountered in the previous example, consider
the space ∞ of all bounded (real or complex) sequences with the metric d(x, y) =
sup |xn − yn |, x = (xn ), y = (yn ) ∈ ∞ . Let
E = {e1 = (1, 0, 0, ...), e2 = (0, 1, 0.0, ...), e3 = (0, 0, 1, 0, ...), ...}.
Then d(en , em ) = 1, n = m. E is bounded, but the property described in Example 1
fails miserably here too. What happens in 1 and 2 ?
This discussion leads us to a strong form of boundedness that is related to com-
pactness and is especially useful in analysis.

Definition 5.9 A subset E of a metric space X is said to be totally bounded if, for
every ε > 0, there is a finite set of points x1 , ..., xk in X such that for any x ∈ E
there is an x j with d(x, x j ) < ε. This just means that finitely many balls of any given
radius cover E. The set of centres of such a collection of balls, {x j }k1 , is called an
ε-mesh for E.

The examples above show that any bounded interval (and hence any bounded set)
in R is totally bounded (what happens in Rn ?), whereas an infinite set in a discrete
metric space and the set of unit vectors in ∞ are bounded, but far from totally
bounded. In fact, we have seen that for any metric d on a set X , there is an associated
equivalent bounded metric d ∗ .
For any ε > 0, a metric space can be covered by ε-balls, and if the space is compact,
a finite subcollection of these balls cover the space. Thus there appears to be some
resemblance between compactness and total boundedness. However, [a, b], (a, b)
64 5 Compact Spaces

are both totally bounded, whereas only the former is compact. What makes the
difference? An obvious difference between the two intervals is that [a, b] is closed in
R and so is complete but (a, b) is not. This turns out to be the crucial difference. In
fact, total boundedness and completeness, in tandem, yield compactness. To obtain
this last mentioned implication, we shall invoke an interesting covering result, also
called Lebesgue Number Lemma, that is of independent importance.
Theorem 5.10 (Lebesgue covering lemma) Let {Vi } be an open cover of a sequen-
tially compact metric space X . There is a δ > 0 (called a Lebesgue number of the
cover {Vi }) such that every subset E of X with diameter at most δ is contained in
some Vi .

Proof Suppose, on the contrary, that there is no such δ. This means that for every
δ > 0 there is a set E with diam E < δ which is not contained in any Vi . Take δ = n1
and denote the corresponding E by E n , n = 1, 2, .... Choose a sequence (xn ) such
that xn ∈ E n for every n. By the assumed sequential compactness of X , there is a
subsequence (xn k ) which converges to some x ∈ X . Since {Vi } is a cover for X , there
is a j with x ∈ V j and so V j contains a ball B(x; ε). Choose k large enough so that
1
nk
< 2ε and xn k ∈ B(x; 2ε ). Then, for y ∈ E n k we have
ε ε
d(y, x) ≤ d(y, xn k ) + d(xn k , x) < 2
+ 2
= ε.
Thus E n k ⊂ B(x; ε) ⊂ V j , contrary to our assumption on the E n . So our assumption
is untenable and the proof is complete. 
The definition of compactness that each open cover has a finite subcover can be recast
in terms of closed sets, by taking complements: every collection of closed sets having
empty intersection has some finite subcollection also having empty intersection. This
will be very useful.
A sequence {xn } in a metric space (X, d) is called a Cauchy sequence (or a d-
Cauchy sequence) if for every ε > 0 there is an n ε > 0 such that d(xm , xn ) < ε for
m, n > n ε . X is said to be complete (or d-complete) if every Cauchy sequence in X
has a limit in X . Completeness is a very important property and a later chapter is fully
devoted to it. For now, we use it in the following immensely useful characterisation.
Theorem 5.11 For a metric space X , the following are equivalent:

1) X is compact;
2) X is Bolzano–Weierstrass compact;
3) X is sequentially compact;
4) X is totally bounded and complete.
Proof 1) ⇒ 2). Suppose there is an infinite set E in X having no limit point. Then
for each x ∈ E, E x := E \ {x} also has no limit point and so is closed. Every finite
subcollection of the collection {E x : x ∈ E} of closed sets has nonempty intersection
as E is infinite. But, clearly, ∩x∈E E x = ∅. The conclusion is that X is not compact.
2) ⇒ 3). Assume that every infinite subset of X has a limit point, and let {xn }
be a sequence in X . If the sequence has only finitely many distinct points, then it
5.2 Compactness in Metric Spaces 65

has a subsequence that is constant and hence convergent. Otherwise the set of points
of the sequence is infinite and so has a limit point x, by hypothesis. So, there is an
xn k ∈ B(x; k1 ) for each k with n 1 < n 2 < · · · . Then {xn k } converges to x.
3) ⇒ 4). If a Cauchy sequence has a convergent subsequence, then the original
sequence itself converges. This shows that sequential compactness of X immediately
implies completeness. We assume X is not totally bounded, so there is an ε > 0 such
that X has no finite ε-mesh. We prove that it is not sequentially compact. For any
x1 ∈ X there is an x2 with d(x1 , x2 ) ≥ ε. By induction, there is a sequence {xn }
satisfying the lower bound d(xm , xn ) ≥ ε whenever m = n. Clearly no subsequence
of {xn } can be convergent.
4) ⇒ 3). Assume that X is totally bounded and complete. To prove sequential com-
pactness, it suffices to show that every sequence {xn } in X has a Cauchy subsequence.
Cover X by a finite number of balls of radius 1. One of these balls, B1 , contains xn
for infinitely many n, so contains a subsequence, denoted by {x1n }. Repeating the
argument for balls of radius 21 with the sequence {x1n } in place of {xn }, we conclude
that there is a ball B2 of radius 21 containing a subsequence {x2n } of {x1n }. Inductively,
for each positive integer k > 1 we get a ball Bk of radius k1 containing a subsequence
{xkn } of {x(k−1)n } (so d(xkn , xkm ) < 2/k for all n, m). Then the ‘diagonal’ subse-
quence {xnn } is a Cauchy sequence. (Why?) (The method employed here exemplifies
what is usually referred to as ‘Cantor’s diagonal argument’. It is a simple, yet pow-
erful tool that is useful in diverse situations and the reader will do well to master it.)
3) + 4) ⇒ 1). Here is where we need to invoke the Lebesgue covering lemma.
Since 3) and 4) are equivalent, this implication would complete the proof. Suppose
X is sequentially compact, hence is totally bounded. Let {Vi } be an open cover for
X and let δ > 0 be a Lebesgue number for this cover. By total boundedness, a finite
number of balls B(x j ; 2δ ), 1 ≤ j ≤ k, cover X . Since the diameters of these balls
are less than δ, for each j there is an i j with B(x j ; 2δ ) ⊂ Vi j . Hence Vi j , 1 ≤ j ≤ k,
form a cover for X . The proof is complete. 

The equivalence (for metric spaces) of sequential compactness with limit point com-
pactness and with the open cover definition were obtained by Fréchet and Hausdorff,
respectively.
Remark 5.12 Using this characterisation, another proof of the Heine–Borel–
Lebesgue theorem can be given: Any bounded sequence of real numbers has a conver-
gent subsequence. So, if K is a closed and bounded set in Rn , any sequence from K
will give n bounded sequences of real numbers, one for each coordinate, and each of
these will have convergent subsequences. Convergence in Rn is coordinate-wise con-
vergence and we can get a convergent subsequence of the original sequence. Thus K
is sequentially compact. Alternately, the Bolzano–Weierstrass theorem for real num-
bers implies that any closed and bounded set in Rn is Bolzano–Weierstrass compact.
An important property of a continuous function on [a, b] is that it is uniformly
continuous. Dirichlet had observed this in his lectures in 1852. Heine ([8], 1870)
mentions uniform continuity as follows.
66 5 Compact Spaces

Es scheint aber noch nicht bemerkt zu sein, dass . . . diese Continuität in jedem
einzelnen Punkte . . . nicht diejenige Continuität ist . . . die man gleichmässige Conti-
nuität nennen kann, weil sie sich gleichmässig Über alle Punkte und alle Richtungen
erstreckt.
(‘It does not seem to have been observed yet, that . . . continuity at each single
point . . . is not that continuity . . . which can be called uniform continuity, because
it extends uniformly to all points and in all directions.’
We give a brief ab initio discussion on uniform continuity here. We look at a
couple of examples first, to motivate and clarify the concept.
Example 5.13 1. Recall the ε-δ definition of continuity. Consider the following
function: f : R → R, f (x) = sin x. Observe that
| sin x − sin y| = |2 cos( x+y
2
) sin( x−y
2
)| ≤ 2| sin( x−y
2
)| ≤ 2 |x−y|
2
= |x − y|
if |x − y| is small enough. Hence, | sin x − sin y| < ε for any sufficiently small ε,
for any two real numbers x, y with |x − y| < δ, where δ is any positive real number
with δ ≤ ε. In other words, we get, for a given ε, a δ independent of the points being
considered.
2. On the other hand, consider the function f (x) = x1 on (0,1). Let us try to
prove its continuity. Take x0 ∈ (0, 1) and let ε > 0. We want to find a δ > 0 such
that | f (x) − f (x0 )| < ε whenever |x − x0 | < δ. To begin with, choose 0 < δ <
x0
2
. If |x − x0 | < δ, then x20 < x < 3x20 and so x1 < x20 . Thus | x1 − x10 | = |x−x
x x0
0|
<
x02
2δ
x02
. The last term is smaller than ε provided δ < 2
ε. Hence |x − x0 | < δ implies
x02
| f (x) − f (x0 )| < ε if 0 < δ ≤ min{ x20 , 2 ε}. The important thing to observe is that
this δ gets smaller and smaller as x0 comes closer and closer to 0. In fact, it is not
possible to find a δ which is independent of x0 . For instance, for any 0 < δ < 1, we
δ δ2 δ
have 0 < δ − 1+δ = 1+δ < δ whereas | f (δ) − f ( 1+δ )| = | 1δ − 1+δ
δ
| = 1. Another
way is to observe that n − 2n = 2n is small if n is large whereas | f ( n ) − f ( 2n
1 1 1 1 1
)| = n
becomes unbounded, so a choice of a uniform δ is impossible.
Note, however, if η is any fixed positive real number, then
| x1 − x10 | = |x−x
x x0
0|
< |x−x
η2
0|
for x, x0 ∈ (η, 1),
so that if δ > 0 is chosen with δ ≤ η2 ε, then | f (x) − f (x0 )| < ε whenever
|x − x0 | < δ whatever x0 ∈ (η, 1) is.
The point of these examples is to show that δ depends, in general, not just on ε,
but also on the point x0 being considered. When it is possible to choose a uniform δ
for all points of the space, depending only on ε, we have a strong form of continuity.

Definition 5.14 A map f : (X, d) → (X , d ) between metric spaces is said to be

uniformly continuous if, for every ε > 0, there is a δ > 0 such that x, y ∈ X and
d(x, y) < δ imply d ( f (x), f (y)) < ε.
The examples above show that sin x is uniformly continuous on the whole of R, 1/x
is uniformly continuous in any interval bounded away from 0, but not on (0,1). The
5.2 Compactness in Metric Spaces 67

reason for the uniform continuity of functions on [a, b] is its compactness. Now we
can justify this statement.
Uniform continuity of continuous functions on [a, b] was first proved by Dirichlet
‘(with a more explicit use of coverings and subcoverings than in Heine’s theorem’
[5] (© Mathematical Association of America, 2015. All rights reserved. Quoted here
with permission.)) as early as 1852 in his lectures, which appeared in print only in
1904. In the meanwhile Heine, who studied under Dirichlet, published a proof in
1872 [9].
Theorem 5.15 (Uniform continuity theorem) Let (X, d) be a compact metric space
and (X , d ) be an arbitrary metric space. Then any continuous map f : X → X is
uniformly continuous.
Proof 1. This is an easy consequence of the Lebesgue covering lemma. Let ε > 0
and cover X by open balls Bi of radius 2ε . Then the pull-backs f −1 (Bi ) form an open
cover for X . If δ is a Lebesgue number for this cover, then x, y ∈ X , d(x, y) < δ
imply x, y ∈ f −1 (B j ) for some j. Hence f (x), f (y) ∈ B j , so d ( f (x), f (y)) ≤
diam B j < ε. 
Proof 2. Assume f : X → X is continuous, but not uniformly. Then there is an ε >
0 such that for every positive integer n there are points xn , yn ∈ X with d(xn , yn ) <
1
n
, but d ( f (xn ), f (yn )) ≥ ε. By sequential compactness, the sequences {xn } and
{yn } have convergent subsequences {xn k } and {yn k }, say xn k → x, yn k → y in X .
(Why can we choose a common subsequence {n k }?) But d(xn k , yn k ) → 0, hence
x = y. Then f (xn k ) → f (x) and f (yn k ) → f (x) in X and so d ( f (xn k ), f (yn k )) →
d( f (x), f (x)) = 0. This is impossible because d ( f (xn k ), f (yn k ) ) ≥ ε for all k. The
proof is complete. 
Corollary 5.16 If X is a compact metric space (e.g. X = [a, b]) and f : X → R is
continuous, then f is uniformly continuous.
The second proof provides us with a useful sequential criterion.
Proposition 5.17 Let f : (X, d) → (X , d ) be a map from one metric space to
another. The following conditions are equivalent:
i) f is uniformly continuous;
ii) d ( f (xn ), f (yn )) → 0 whenever d(xn , yn ) → 0.
Proof Let f be uniformly continuous and d(xn , yn ) → 0 where xn , yn ∈ X . Let
ε > 0 and choose a corresponding δ from the uniform continuity of f . Choose n δ
such that n ≥ n δ implies d(xn , yn ) < δ. Thus n ≥ n δ implies d(xn , yn ) < δ implies
d ( f (xn ), f (yn )) < ε.
If (i) fails, then for some ε > 0 and for every n, there are xn , yn ∈ X with
d(xn , yn ) < n1 and d ( f (xn ), f (yn )) ≥ ε, so (ii) fails. 
Exercise. Prove that a uniformly continuous map takes Cauchy sequences to
Cauchy sequences. Is the converse true? (Hint: If {xn } is a Cauchy sequence in
R, what about {xn2 }?) What if f is continuous?
Exercise. Use the previous exercise to show that the function 1/x on (0,1) is not
uniformly continuous.
68 5 Compact Spaces

5.3 Compactness in Topological Spaces

Although the first compactness result of Borel [10, 11] (5.1) was formulated in terms
of open covers in the 1880’s, the general definition was proposed by Tietze [12] and
adopted by Alexandroff and Urysohn [13, 14] (who used the term ‘bicompact’) only
much later, in the 1920’s. This modern open cover definition appeared in Alexan-
droff and Hopf [15]. Ein topologischer Raum heißt bikompakt, wenn jede seiner
offenen Überdeckungen eine endliche Überdeckung enthält. (‘A topological space is
bicompact if each of its open cover contains a finite cover.’)

Definition 5.18 A topological space X is said to be compact if every open cover of

X has a finite subcover.

Warning: In Bourbaki [16], ‘compact’ means our compact+ Hausdorff separation

axiom (see Chap. 6 below); quasi-compact is the term used for the usual ‘compact’.
For compactness, it is sufficient if basic open covers have finite subcovers. This is
easy. The corresponding result for subbasic open covers is also true, but lies deeper.
See the exercises in Chap. 7.
Proposition 5.19 Let X be a topological space and suppose there is a base B for X
with the property that every cover of X by members of B has a finite subcover. Then
X is compact.

Proof Let {Vi } be any open cover for X . Writing each Vi as a union of members of
B, we get a cover of X by members of B. By hypothesis, this has a finite subcover
{B1 , ..., Bk } of basic open sets. Choosing, for each j, 1 ≤ j ≤ k, a Vi j such that
B j ⊂ Vi j we end up with a finite subcover {Vi1 , ..., Vik } of the original cover {Vi }.

Here is a definition that will enable us to state the already mentioned reformulation
of compactness in terms of closed sets.

Definition 5.20 A collection {E i } of subsets of X is said to have finite intersection

property (f.i.p. for short) if the intersection of any finite subcollection is nonempty.

Proposition 5.21 (F. Riesz) A topological space is compact if and only if every
family of closed subsets having finite intersection property has nonempty intersection.

Proof As already observed, just take complements in the open cover definition.
Write out the details. 

Corollary 5.22 If X is compact, then so is any closed subset F.

Proof If {E i } is a family of closed sets in F with finite intersection property, then it

is such a family in X . Use compactness of X . 
5.3 Compactness in Topological Spaces 69

Exercise. Prove the corollary from the open cover definition directly.
Exercise. Prove the following variant of the proposition: X is compact if and only

if Āi is nonempty for every family {Ai } of subsets of X with finite intersection
property.
Exercise. A subset K of a space X is compact if and only if every collection of
sets open in X covering K has a finite subcover. Thus it is immaterial whether sets
of the cover are open in K or in X .
Here is a simple, widely used property of compact spaces.

Proposition 5.23 A continuous image of a compact space is compact: if f : X → Y

is continuous and X is compact, then so is f (X ).

Proof If {Vi } is an open cover for f (X ), then the pull-backs f −1 (Vi ) form an open
cover for the compact space X and so finitely many of these { f −1 (Vi j )}k1 cover X .
Then the Vi j cover f (X ). 

The classical theorem on maxima and minima of real continuous functions on closed
intervals is now an easy consequence.

Corollary 5.24 A real valued continuous function f on a compact space X is

bounded and has a maximum and a minimum. In particular, if f is positive, there is
a δ > 0 such that f (x) ≥ δ > 0.

Proof f is bounded means its range is bounded. K = f (X ) is compact in R, so is

closed and bounded. Thus f is bounded and hence m := inf f (x) and M := sup f (x)
are in R. As K is closed, m, M ∈ K and so m = min f, M = max f . For the last
part, take δ = m. 

The case X = [a, b] is a basic result in Real Analysis and was first published by
Weierstrass in 1877, although he must have proved and used it in his lectures much
earlier. The ideas underlying his proof go back to the proof of a fundamental property
of real numbers, namely, the greatest lower bound property, which Bolzano published
in 1817.
Corollary 5.25 If f is a continuous bijection from a compact space X onto a metric
space Y , it is actually a homeomorphism.

Proof To show that f −1 , which exists as f is bijective, pulls back closed sets to
closed sets, i.e. to show that f maps closed sets to closed sets. If C is closed in X ,
then it is compact and so f (X ) is compact. As Y is a metric space, the compact set
f (C) is closed. 

The proof hinges on the fact that compact sets are closed in metric spaces. So, the
result holds if the range space has this property. Hausdorff spaces are such more
general spaces. See 8.10.
Recall that a metric space X is compact if and only if every sequence {xn } has a
convergent subsequence {xn k }. The statement that {xn } has a subsequence xn k → x
can be reformulated as follows: the xn are frequently in every neighbourhood U of
70 5 Compact Spaces

x, i.e. for any N there is an n > N such that xn ∈ U . In this case we say that x is a
cluster point of the sequence {xn }. So x is a cluster point of {xn } if and only if there
is a subsequence converging to x. Hence a metric space X is compact if and only if
every sequence in X has a cluster point.
This is no longer true for a general topological space. But is there an analogous
result for arbitrary topological spaces? Indeed, we have an exact, useful analogue
in terms of nets. The concept of a cluster point of a net poses no problems. But
what is the analogue of a subsequence? A subsequence of the sequence n → xn
is got by composing this map (i.e. the sequence) with a strictly increasing map
k → n k , N → N, to get the sequence k → xn k . This leads us to subnets. One last
point to be remembered - the domains of definition of a sequence and a subsequence
are always the same set N. But for a net and a subnet they can be different directed
sets.
Definition 5.26 A point x ∈ X is a cluster point of a net {xα } in X if xα lies frequently
in every neighbourhood Ux of x. This means {α : xα ∈ Ux } is a cofinal subset. Here,
a subset E of a directed set D is said to be cofinal if for every α ∈ D there is a
β ∈ E with β ≥ α. A subnet of a net x : D → X is a net x ◦ ν : E → X , where E is
a directed set and ν : E → D is a map which satisfies the condition: for each β ∈ D
there is an η ∈ E such that ξ ≥ η implies ν(ξ ) ≥ β.
In particular, if ν is increasing (ν(η) ≤ ν(ξ ) if η ≤ ξ ) and ν has cofinal range, then
x ◦ ν is a subnet. In consonance with our notation for subsequences, we write {xαη } for
the subnet that is the composition of the net α → xα and the map η → αη , E → D.
The condition on ν says that ν(ξ ) is ‘large’ if ξ is ‘large’. Kelley [17] coined the term
net and introduced subnets in 1950. Sequences are nets, subsequences are subnets
and the cofinality condition is automatic in the case of sequences. (Caution: a subnet
of a sequence need not be a subsequence.)
As early as 1921, Vietoris [3] introduced the concept of a cluster point (‘Gren-
zpunkt’): “Definition: Ein Punkt r heisst teilweiser Grenzpunkt einer orientierten
Menge zweiter Ordnung M ... ohne letztes Element wenn es zu jeder Umgebung
V von r und zu jedem Element Q β von M noch Elemente Q α > Q β gibt, welche
Punkte von V enthalten.” (A point r is a cluster point of M if for any neighbourhood
V of r and any Q β ∈ M, V contains points Q α > Q β .)

Proposition 5.27 If a net (xα ) in a topological space X converges to a point x ∈ X ,

then any subnet (xαη ) also converges to x.

Proof If (xα ) converges to x ∈ X , then for any neighbourhood U of x, there is an

α0 such that xα ∈ U for α ≥ α0 . By definition, there is an η0 ∈ E such that η ≥ η0
implies ν(η) ≥ α0 . Thus xνη ∈ U for η ≥ η0 and xνη → x. 

Proposition 5.28 A point x in a space X is a cluster point of a net {xα } if and only
if it is a limit of some subnet of {xα }.

Proof If x is not a cluster point of {xα }, then there is a neighbourhood U of x such that
{xα } is not frequently in U , so is eventually in U c . Then any subnet is also eventually
5.3 Compactness in Topological Spaces 71

in U c and hence cannot converge to x. Conversely, if x is a cluster point of {xα },

then {xα } is frequently in every U ∈ Ux . Let E := {(α, U ) ∈ D × Ux : xα ∈ U }. It
is easy to check that it is a directed set with order (α, U )  (β, V ) if α ≤ β and
V ⊂ U . Define ν(α, U ) = α. Then ν is increasing and clearly has cofinal range.
Hence x ◦ ν is a subnet. If V ∈ Ux and if β ∈ D is such that xβ ∈ V , then for
any (α, U )  (β, V ), x ◦ ν(α, U ) = xα ∈ U ⊂ V and so the subnet is eventually
in V . 
Example 5.29 There is a bijection N → Q, so we can write the rationals as a
sequence q1 , · · · , qn , · · · . For this sequence every real number is a cluster point.
The sequence {n} has no cluster point in R.
Theorem 5.30 A topological space X is compact if and only if every net in X has
a cluster point.
Proof Suppose every net in X has a cluster point. Let {E i }i∈Ibe a family of closed
sets with finite intersection property.  We have to show that i∈I E i is nonempty.
Now, for each finite F ⊂ I, let E F := i∈F E i , so that E F is a nonempty, closed set.
Choosing a point xF ∈ E F for each finite F ⊂ I, we get a net (xF ), finite subsets of
the index set I being ordered by inclusion. Observe that if F ⊃ F , then E F ⊂ E F
and so xF ∈ E F ⊂ E F . This means that the net (xF ) is eventually in each E F . But,
by hypothesis, (xF ) has a cluster point and this cluster point belongs to each of the
closed sets E F . Thus the intersection E F over all finite subsets of I is nonempty.
Since i∈I E i contains this intersection, one implication is proved.
Conversely, assume that X is compact and let {xα : α ∈ D} be a net in X . For
each α ∈ D, let Aα := {xβ : β ≥ α}. Then {Aα } is a collection of sets with finite
intersection property, because any finite set in D has an upper bound.Then { Āα }
also has finite intersection property and, by compactness, ∩α Āα is nonempty, say
x̄ ∈ ∩ Āα . If x̄ is not a cluster point of {xα }, there is a neighbourhood U of x̄ such
that {xα } is not frequently in U . So there is a β with xα ∈ / U for α ≥ β. Thus U and
Aβ are disjoint and so x̄ ∈ / Āβ , a contradiction. 
Exercise 5
1. Compact sets. a) A compact set need not be closed, the closure of a compact
set may not be compact and the intersection of two compact sets may not be compact.
(See Kelley, [18])
b) If A is a compact subset of a metric space X , then there are points a, b, c ∈ A
such that d(x, A) = d(x, c), diam A = d(a, b).
c) If {K i } is a family of closed compact sets and V is an open set with ∩i K i ⊂ V ,
then ∩n1 K i j ⊂ V for some finitely many indices i j .
d) Let { f k } be a sequence of real continuous functions on Rn converging uniformly
to f 0 . If K is compact, then so is C = ∪∞ 0 f n (K ).
e) Find infinitely many mutually nonhomeomorphic compact sets in the real line.
(In fact, there are uncountably many such sets, [19].)
f) In an ordered space X with order topology, all closed, order bounded sets are
compact if and only if X is order complete.
72 5 Compact Spaces

g) What are the compact sets in R with cofinite topology?

2. Cluster points. a) If a net converges, its limit is a cluster point. If a net has a
unique cluster point, is the net necessarily convergent?
b) Find a net with exactly three cluster points.
3. Isometries. a) An isometry of a compact metric space is onto.
b) Find a metric space X and a nonsurjective isometry on X .
4. Continuous injections. a) No continuous injection S1 → R exists. (Give a
heuristic reasoning. This rightfully belongs to the chapter on connectedness. Come
back to this at that stage.)
b) Find two continuous injections R → {(x, y, z) : x 2 + y 2 = 1}.
5. Locally finite refinements. Any open cover {Un } of Rn has a locally finite
refinement {Vn }, i.e. {Vn } is an open cover such that Vn ⊂ Un and any compact set
intersects only finitely many Vn .
6. Countable compactness. A space X is said to be countably compact if every
countable open cover of X has a finite subcover.
a) X is countably compact if and only if every sequence in X has a convergent
subnet.
b) A countable compact metric space is (sequentially) compact.
c) Characterise countable compactness using f.i.p.
d) A space is countably compact if and only of every decreasing sequence of
closed sets has nonempty intersection.
e) Every sequentially compact space is countably compact. (The converse is not
true, but we will not discuss this. See, e.g., [20])
f) The product of a compact space and a countably compact space is countably
compact.
7. Pseudocompact spaces. A space is pseudocompact if every real continuous
function on it is bounded.
a) Find an example of a noncompact, pseudocompact space.
b) A countably compact space is pseudocompact.
8. Noncompact metric spaces. a) On any noncompact metric space, there is an
unbounded real continuous function (i.e. a pseudocompact metric space is compact).
b) A noncompact metric space admits a compatible unbounded metric (i.e. a metric
space is compact if (and, clearly, only if) every admissible metric is bounded).
9. σ -compact spaces. A space is σ -compact if it is a countable union of compact
subsets.
a) Rn and any countable discrete space are σ -compact.
b) The product of two σ -compact spaces is σ -compact.
10. Sequentially compact spaces. a) Compactness and sequential compactness
are independent - neither implies the other.
[0, 1][0,1] is compact (by Tychonoff’s theorem, see Chap. 7) but is not sequentially
compact. The standard example of a noncompact, sequentially compact space is the
5.3 Compactness in Topological Spaces 73

ordinal space. Since we have not discussed ordinal numbers, we refer to our standard
references [18, 19, 21, 22] for examples involving ordinal spaces.
11. Noncompact balls. The closed unit balls in the sequence spaces 1 , 2 and
∞
 and in C[0, 1] are not compact.
12. Semi-continuity. If f is a real lower (upper) semi-continuous on a compact
space, then f has a minimum (maximum) on X .
13. Bolzano–Weierstrass. a) On N, consider the topology generated by {Un =
{2n − 1, 2n}}n∈N as a subbase. This space is Bolzano–Weierstrass compact, but not
compact nor sequentially compact.
b) Does the Bolzano–Weierstrass theorem hold in R ?
14. Compact set of eigenvalues. The set of eigenvalues of a compact set K of
matrices is compact: i.e. if σ (A) is the set of eigenvalues of an A ∈ M(n, R), then
σ (K) := ∪{σ (A) : A ∈ K} is compact.
15. Ultranets. A net {xα } in X is called an ultranet or a universal net if for any
E ⊂ X , {xα } is either eventually in E or eventually in E c . See Kelley [18] or Willard
[19] for this exercise and the next.
a) The image of an ultranet under any map is an ultranet.
b) Every net has a subnet which is an ultranet.
c) A subnet of an ultranet is an ultranet.
d) A space X is compact if and only if each ultranet in X converges.
16. Ultrafilters. A filter F on X is called an ultrafilter if it is a maximal filter,
i.e. there is no filter F ∗ with F  F ∗ . [23]
a) A filter F on X is an ultrafilter if and only if for any E ⊂ X, either E ∈ F or
Ec ∈ F .
b) Every filter is contained in an ultrafilter. (Needs Zorn’s lemma.)
c) Any map takes an ultrafilter to an ultrafilter.
d) X is compact if and only if every ultrafilter in X converges if and only if every
filter in X has a cluster point. (A point x is a cluster point of a filter F if each F ∈ F
intersects every U ∈ Ux .)
e) Obtain results on the relation between cluster points of nets and filters and
between ultranets and ultrafilters. (see e.g. [19])

5.4 Biographical Notes

5.4.1 E. Borel

Émil Borel (1871–1956) was a French mathematician and politician. His main math-
ematical work was in analysis, measure theory and probability. Borel’s theorem on
the compactness of closed intervals, Heine–Borel–Lebesgue theorem, Borel sum-
mation, Borel sets, Borel measure, Borel algebra, Borel–Cantelli lemma, Borel’s
Law of Large Numbers, Borel–Carathéodory theorem, Borel distribution bear his
74 5 Compact Spaces

name. The famous mathematician Lebesgue was his student. There is the Centre
Émil Borel at the Institut Henri Poincaré (the institute of which he was a founder).
A moon crater is named after him. He was a cabinet minister for a brief period when
the mathematician Paul Painlevé was the Prime Minister.

5.4.2 Heine

Eduard Heine (1821–1881) was a German mathematician known for his work on
Real Analysis and special functions. He dedicated his thesis (1842), on differential
equations, to Dirichlet. His lectures were known for clarity and as a teacher he was
greatly liked. He was the first to publish a proof of uniform continuity of continuous
functions on closed intervals. He is famous because of the Heine–Borel theorem.

5.4.3 Lebesgue

Henri Lebesgue (1875-1941) was a French analyst whose theory of integration is

the most commonly used in modern mathematics. His doctoral thesis Intégrale,
longueur, aire, written under Borel’s supervision and published in 1902, was a
remarkable, seminal work. It is generally considered as one of the finest mathe-
matical dissertations of all times. Lebesgue’s first paper (1898) was on the Weier-
strass approximation theorem. In addition to his work in analysis, he also did some
work in topology and geometry. His attempts to prove dimension and domain invari-
ance led to a concept of dimension of a topological space. His doctoral students
include the analyst Paul Montel, the topologist Zygmunt Janiszewski and the
famous differential topologist George de Rham. Lebesgue measure, Lebesgue inte-
gral, Lebesgue constant, Lebesgue number, Lebesgue decomposition, Lebesgue den-
sity, Lebesgue differentiation theorem. Lebesgue’s dominated convergence theorem,
Lebesgue–Radon–Nikodym theorem, Lebesgue covering dimension, Lebesgue cov-
ering lemma, Riemann–Lebesgue lemma (in Fourier analysis), Walsh–Lebesgue the-
orem (on harmonic functions) are some things mathematical named after Lebesgue.

5.4.4 Cantor

Georg Cantor (1845–1918) was the creator of Set theory on which the structure
of modern mathematics has been raised. His theories of ‘infinity of infinities’ and
transfinite numbers and their arithmetic were vehemently opposed by many leading
mathematicians of his time, including his teacher Leopold Kronecker. This hos-
5.4 Biographical Notes 75

tility greatly affected his work and is thought to be the reason for several bouts of
mental depressions he suffered from 1884 until his death. This criticism was matched
by later encomium (e.g. David Hilbert’s famous quotes: ‘No one shall expel us from
the paradise that Cantor has created.’ Cantor’s set theory is ‘The finest product
of mathematical genius and one of the supreme achievements of purely intellectual
human activity.’ https://mathshistory.st-andrews.ac.uk/Biographies/Cantor/). The
Royal Society’s highest honour was bestowed on him. Initially, he wrote several
papers on number theory, before he turned to Trigonometric Series. He solved the
long-standing problem of the uniqueness of trigonometric series representation of
functions and, in the course of this work, discovered transfinite ordinals. Cantor’s
1874 paper on algebraic numbers marked the beginning of set theory. He showed that
algebraic numbers are countable, real numbers are not and so deduced the existence
of (uncountably many) transcendental numbers, without producing even one tran-
scendental number. This kind of proofs were also criticised. He proved the existence
of many kinds of infinities, developed the theory of transfinite numbers, cardinal
numbers, ordinal numbers, well-ordering principle, introduced the concept of one–
one correspondence (His proof that ‘the unit interval contains as many points as
the unit square’ shocked the mathematical world.), the diagonal argument (which
is simple, powerful and very useful; we have used it.), got the real numbers as the
completion of rationals (before Richard Dedekind developed his ‘cuts’ to get real
numbers), formulated the Continuum Hypothesis, which he tried to prove in vain
and was the first of the famous 23 problems that Hilbert listed, at the Paris Inter-
national Congress (1900); it has since been shown that it can neither be proved nor
disproved in the standard ZFC-set theory, as Kurt Gödel (1940) and the harmonic
analyst turned logician Paul Cohen (1963) showed. Kronecker and Weierstrass
were his thesis advisers. Cantor set, Cantor intersection theorem, Cantor function
are well-known in analysis and topology. Cantor was instrumental in the formation
of the German Mathematical Society and also in convening the first International
Congress of Mathematicians.

5.4.5 Vietoris

Leopold Vietoris (1891–2002) (Look at the years! No typo!) was an Austrian mathe-
matician, mainly known for his contributions to topology, both general and algebraic,
(and for his longevity! His last paper appeared when he was 103!). He obtained his
first results when he was fighting in World War I, was an Italian prisoner of war
and, soon after his release, submitted his thesis to the University of Vienna in 1919.
He spent three semesters in Amsterdam with Brouwer; Alexandroff and Menger
were also there at that time. During his stay there, he obtained his most significant
results in Algebraic Topology—Mayer–Vietoris sequence for homology groups, etc.
He defined regularity of topological spaces (slightly different from the currently
used definition due to Tietze) and the modern notion of compact spaces (with the
76 5 Compact Spaces

name ‘spaces without gaps’). As mentioned earlier (after Remark 4.11), he came up
with the idea of ‘directed sets’ in a general setting (1921) even before the Moore-
Smith paper on nets (1922) and also the concept of a filter base. He had essentially
developed a theory of convergence parallel to the modern theories of net and filter
convergence and so may be considered as the originator of modern convergence
theory (see Heinrich Reitberger [24]). He essentially proved (1921) the normality
of compact Hausdorff spaces. He made attempts to formulate the definition of a
manifold. The ‘Vietoris topology’ that he defined on the class of closed sets in a
topological space coincides with the Hausdorff metric on compact metric spaces.
Mayer–Vietoris sequences, Vietoris complexes, Vietoris cycles, Vietoris mapping
theorem are some other concepts in algebraic topology. Functional differential equa-
tions, probability theory, orientation in mountainous terrain by differential geometric
methods, the strength of the alpine ski and the physics of block glaciers are other
areas to which he made contributions.

References

1. Hewitt, E.: The role of compactness in analysis. Am. Math. Monthly 67, 499–516 (1960).
https://doi.org/10.2307/2309166
2. Fréchet, M.: Espaces abstraits. Gauthier-Villars, Paris (1928)
3. Vietoris, L.: Stetige Mengen. Monatsh Math. und Physik 31, 173–204 (1921). https://doi.org/
10.1007/BF01702717
4. Borsuk, K.: Sur les rétractes. Fund. Math. 17, 152–170 (1931). https://doi.org/10.4064/fm-17-
1-152-170
5. Raman-Sundström, M.: A pedagogical history of compactness. Am. Math. Monthly 122, 619–
635 (2015). https://doi.org/10.4169/amer.math.monthly.122.7.619
6. Andre, N., Engdahl, S., Parker, A.: An analysis of the first proofs of the Heine-Borel Theorem-
Borel’s Proof”. Loci (2003). https://doi.org/10.4169/loci003890
7. Fréchet, M.: Sur quelques point du calcul fonctionnel. Rendiconti di Palermo 22, 1–74 (1906).
https://doi.org/10.1007/BF03018603
8. Heine, E.: Über trigonometrische Reihen. J. Reine und Angew. Math. 71, 353–365 (1870).
https://doi.org/10.1515/crll.1870.71.353
9. Heine, E.: Die Elemente der Funktionenlehre. J. Reine und Angew. Math. 74, 172–188 (1872).
https://doi.org/10.1515/crll.1872.74.172
10. Borel, E.: Sur quelques points de la théorie des fonctions. Gauthier-Villars (1894). https://doi.
org/10.24033/asens.406
11. Borel, E.: Leçons sur la Théorie des Fonctions. Paris (1898)
12. Tietze, H.: Beiträge zur allgemeinen Topologie I. Math. Ann. 88, 290–312 (1923). https://doi.
org/10.1007/BF01579182
13. Alexandroff, P., Urysohn, P.: Zur theorie der topologischen Räume. Math. Ann. 92, 258–266
(1924). https://doi.org/10.1007/BF01448008
14. Alexandroff, P., Urysohn, P.: Memoire sur les espaces topologiques compactes. Verh. Akad.
Wetensch. Amsterdam 14, 1–96 (1929)
15. Alexandroff, P., Hopf.: Springer, H. Topologie I (1935)
16. Bourbaki, N.: General Topology, Chapters 1–4. Springer (1995)
17. Kelley, J.L.: Convergence in topology. Duke Math. J. 17, 277–283 (1950). https://doi.org/10.
1215/S0012-7094-50-01726-1
18. Kelley, J.L.: General Topology. Affiliated East-West Press (1969)
References 77

19. Willard, S.: General Topology. Addison-Wesely (1970)

20. Steen, L.A., Seebach Jr., J.A.: Counter Examples in Topology. Dover (1995)
21. Dugundji, J.: Topology. Prentice Hall of India (1975)
22. Munkres, J.R.: Topology, 2nd edn. Pearson Education, Asia (2001)
23. Cartan, H.: Filtres et Ultrafiltres. C.R. Acad. Sci. Paris 205, 777–779 (1937)
24. Reitberger, H.: Leopold Vietoris, 1891–2002. Not. Am. Math. Soc. 49(10), 1232–1236 (2002)
Chapter 6
Topologies Defined by Maps

The product topology is a very important special case of an important way of defining
topologies using maps: defining a topology on a set X using maps from X to topo-
logical spaces. In the other direction, getting a topology on a set Y from maps into Y
from spaces has quotient topology as a particular case. We discuss both these ways of
defining topologies, (called the initial topology and the final topology, respectively,
by Bourbaki) before specialising to the product and quotient topologies.
Tychonoff’s famous 1930 definition of the product topology is given only for
arbitrary products of copies of [0,1]:
Es sei {Jα } eine Menge von der Mächtigkeit τ von abstrakt gegebenen zueinan-
der fxemden Einheitsstrecken 0 ≤ t ≤ 1. Ein Punkt x des Raumes Rτ isr definition-
sgemäss der Inbegriff {t1 , t2 , . . . , tα , . . .} von “Koordinaten” tα , wobeit tα ein Punkt
vonJα , also eine reelle Zahl 0 ≤ tα ≤ 1 ist. Die Umgebungsdefinition geschieht fol-
gendermassen: Es sei x0 = {t10 , t20 , . . . , tα0 , . . .} ein Punkt yon Rτ . Wir wählen beliebig
endlichviele Jα1 , Jα2 , . . . , Jαk und auf jedem dieser Intervalle Jαi zwei rationale
Zahlen t  αi , < tα0i < tαi ; eine Umgebung von x0 = {t10 , t20 , . . . , tα0 , . . .} besteht dann
definitionsgemass aus allen Punkten x = {t1 , t2 , . . . , tα , . . .} die den Bedingungen
tα i < tαi < t"αi genügen. (Tychonoff, [1])
Tychonoff says: Jα is a collection of copies of [0,1], Rτ is the set of points x
with coordinates {t1 , . . . , tα , . . .}, tα ∈ Jα . If we choose finitely many Jα1 , . . . , Jαk
and rational numbers tα i < tα0i < tαi , a neighbourhood of x0 consists, by definition,
of points x of Rτ with tα i < tαi < tαi . Čech [2] points out that the definition carries
over to the general case. The general definition is given thus:
Let {Si } be a family of sets; the subscript runs over an arbitrary given set I . The
cartesian product Pi Si of the family {Si } is the set of all families x = {xi }, each
xi belonging to X i . The xi are called coordinates of x. If every Si is a topological

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 79
K. Parthasarathy, Topology, La Matematica per il 3+2 134,
https://doi.org/10.1007/978-981-16-9484-4_6
80 6 Topologies Defined by Maps

space, we introduce a topology into S = Pi Si by means of the following open base

B: the elements of B are sets of the form Pi G i , where (1) each G i is an open subset
of Si , (2) G i = Si except for a finite number of subscripts i.

6.1 Initial and Final Topologies

We now take up ‘initial topology’, the first of the two ways, mentioned above, of
defining topologies using maps.

Proposition 6.1 Let X be a set and let {Yi } be a family of topological spaces. For
each i, let a map f i : X → Yi be given. There is the smallest topology on X making
each f i continuous.

Proof The required topology is the intersection of all topologies making each f i
continuous. This family is nonempty as the discrete topology is in it. Explicitly, the
topology is generated by the family of all sets of the form f i−1 (Vi ) as Vi varies over
all open sets in Yi and i runs through all the indices under consideration. 

The topology on X given by the proposition is the initial topology on X induced

by the family { f i } of maps. The term weak topology generated by the functions f i is
also used, but we avoid this terminology.

Example 6.2 1. The relative topology on a subset E of a space X is the initial

topology generated by the inclusion map ι : E → X .
2. The indiscrete and discrete topologies are, respectively, generated by all con-
stant real-valued functions and all real functions.
3. For any space (X, T ), the initial topology on X given by the identity map
id X : X → (X, T ) is T itself.
4. Let Ti be a collection of topologies on X and let idi : X → (X, Ti ) be the
identity
 map for all i. Then the initial topology on X generated by the family {idi }
is Ti = sup Ti , the topology generated by i Ti .

Basic open sets for initial topologies and continuity for maps from a space with initial
topology are easily described.
Proposition 6.3 Let X be endowed with the initial topology generated by a family
f i : X → Yi of maps. Then
1) the sets of the form ∩n1 f i−1
k
(Uik ), where i 1 , . . . , i n are any finite number of
indices and Uik are open sets in Yik , form a basis for X ;
2) for any topological space Y , a map g : Y → X is continuous if and only if
f i ◦ g : Y → Yi is continuous for all i.
6.1 Initial and Final Topologies 81

Proof 1) is immediate: sets of the form f i−1 (Ui ), Ui ⊂ Yi open, is a subbasis for the
initial topology, by definition, and hence finite intersections of these form a basis.
2) The f i are continuous and so if g is continuous, so are f i ◦ g for all i. Conversely,
suppose all the f i ◦ g are continuous. Then, for any open Ui ⊂ Yi , g −1 ( f i−1 (Ui )) is
open in Y . This means that g pulls back all the subbasic open sets f i−1 (Ui ) to open
sets and so is continuous. 

We will look at the important instance of the product topology in the next section.
Here is the dual construction, the ‘reverse method’.

Proposition 6.4 Let Y be a set and let {X i } be a family of topological spaces. For
each i, let a map gi : X i → Y be given. Then there is the largest topology on Y
making each gi continuous.

Proof The indiscrete topology on Y makes all the gi continuous. So the collection
T
 of all topologies on Y making all the gi continuous
 is nonempty. Define T{gi } :=
{T : T ∈ T }, the topology generated by {T : T ∈ T }. This is the required
topology. 

The topology given by the proposition is called the final topology on Y given by
the maps {gi }. An important instance of this topology, the quotient topology, will be
considered later in the chapter.

Proposition 6.5 Let {X i } be a collection of spaces and let a set Y be given the final
topology given by a set of maps gi : X i → Y . Then
1) V is open in Y if and only if gi−1 (V ) is open in X i for every i;
2) for any topological space Z , a map h : Y → Z is continuous if and only if
h ◦ gi : X i → Z is continuous for all i.

Proof 1) holds just by the definition of final topology.

2) Again, one implication is trivial: if gi and h are continuous, then so is h ◦ gi .
Conversely, suppose h ◦ gi is continuous for each i. Then for any open set W in Z ,
gi−1 (h −1 (W )) is open in X i for all i and so, h −1 (W ) is open in Y , by the definition
of the final topology. 

Example 6.6 1. Let {Ti } be a collection of topologies on a set X and let idi :
(X, Ti ) → X be the identity
 map for all i. Then the final topology on X generated
by the family {idi } is Ti = inf Ti .
2. The function p(x) = e2πı x , x ∈ [0, 1], maps [0,1] onto the unit circle S1 , is one-
one on (0,1) and p(0) = p(1) = 1. The set of fibres p −1 (z) for z ∈ S1 are singletons
{x}, x ∈ (0, 1) for z = 1 and p −1 (1) = {0, 1}. Let X̃ denote the set of fibres: X̃ :=
{ p −1 (z) : z ∈ S1 }. Note that X̃ is got by ‘identifying’ the end points of [0,1]. Give X̃
the final topology defined by the map g : [0, 1] → X̃ , g(x) = x̃ := p −1 ( p(x)). Then
X̃ , a continuous image of the compact space [0,1], is compact. The map h : X̃ →
S1 defined by h(x̃) = p(x) is well-defined and is continuous because h ◦ g = p :
[0, 1] → S1 is continuous. Thus h is a bijective continuous map from the compact
82 6 Topologies Defined by Maps

space X̃ onto S1 ⊂ R2 and so is a homeomorphism. Thus, all that we are saying

is that by identifying the end points of [0,1] we get a circle. This is ‘geometrically
obvious’: by joining the end points of a string, we get a circle!

The last example is a special instance of ‘quotient spaces’ that will be discussed in
the final section of the chapter.

6.2 Product Topology

While discussing basic open sets (Chap. 3), we mentioned the product of two topo-
logical spaces. Let us recall that definition. If X, Y are two topological spaces, the
collection B of subsets of X × Y of the form U × V , where U ⊂ X, V ⊂ Y are
open, satisfies the needed conditions to form a basis for a topology on X × Y .
This topology with B as a basis is called the product topology on X × Y . Let
us look at this topology from a different perspective now. Consider the projec-
tion maps p : X × Y → X, q : X × Y → Y given by p(x, y) = x, q(x, y) = y. For
open sets U ⊂ X, V ⊂ Y , p−1 (U ) = U × Y and q−1 (V ) = X × V and so p, q
both pull back open sets to open sets, showing that they are continuous. More-
over, if T is any topology on X × Y such that both p and q are continuous
and U ⊂ X, V ⊂ Y are open, then p−1 (U ) = U × Y, q−1 (V ) = X × V ∈ T , hence
U × V = (U × Y ) ∩ (X × V ) is in T . This means B ⊂ T and so T contains the
product topology. We have proved that the product topology is the smallest topology
on X × Y that makes both the projections p, q continuous. Thus, it is the initial
topology defined on X × Y by the projections p and q. We take cue from this to
define the product topology for any collection of spaces.
To undertake a full fledged discussion of product spaces, which is a desideratum
for a basic understanding of topology, it is essential to first understand the meaning
of the cartesian product set of an arbitrary collection of sets. Let us first look at a
finite product X 1 × X 2 × . . . X n of sets. This, by definition, is the set of n-tuples
(x1 , x2 , . . . , xn ) with x j ∈ X j for each j ∈ {1, 2, . . . , n}. But what is an n-tuple?
Writing x( j) for x j , we see that a we have a map x : {1, 2, . . . , n} → ∪ j X j with
x( j) ∈ X j for each j and the n-tuple is just the listing out of all the values of this
map on the set {1, 2, . . . , n} as (x(1), x(2), . . . , x(n)). With this in mind, we can now
give the definition of the cartesian product of an arbitrary indexed family {X i }i∈I
of nonempty sets. (The index set I will  not be usually mentioned explicitly, except
when necessary.) The product X  = i X i is, by definition, the set of all maps x on
the index set I taking values in X i with x(i) ∈ X i for every indexi ∈ I. Thus,
for example, a sequence (xn ) in a set X is an element of the product n X n where
X n = X for each n ∈ N. In consonance with the notation in this case, conventionally
one writes x(i) as xi and x = (xi ) ∈ i X i . When all the X i are equal, say X i = X
for all i, we also write X I for X i . With this notation, a sequence in X is an element
of X N .
6.2 Product Topology 83


Definition 6.7 Take the product X = i X i of an indexed family {X i } of nonempty
sets. For each i, the i-th projection pi : X → X i is the map given by pi (x) = xi
taking x = (xi ) ∈ X to its ‘i-th coordinate’. .
The product topology can be described in many ways. We formulate our definition
in terms of the projections, as explained above.

Definition 6.8 For a family {X i } of spaces, the product topology on X i is the
initial topology defined by the pi : X i → X i ; i.e. it is the smallest topology that
makes all the projections pi continuous.
As noted, the product topology was defined by Tychonoff. According to J.L.Kelley,
the formulation given here is due to Bourbaki.

Proposition 6.9 The product topology on the cartesian product X = i X i of an
 i , i ∈ I} of topological spaces has as a base the collection B of
indexed family {X
sets of the form pi−1 (Ui ) with Ui ⊂ X i open and the intersection is over finite sets
F ⊂ I.

Proof The product topology has as subbase all sets of the form pi−1 (Ui ), Ui ⊂ X i
open, and finite intersections of these give B. 
 
Remark 6.10 1. In X i , a subset of the form Ui is open if and only if each Ui
is open and Ui = X i for all but finitely many i.
2. Thus, an infinite product of discrete spaces is never discrete.

As a space with initial topology, for a product space continuity of maps into it is
characterised in terms of projections.

Proposition 6.11 A map f : Y → X i , from a space to a product space, is con-
tinuous if and only if its composition with each projection, pi ◦ f : Y → X i , is con-
tinuous.

Proof Recall that the product topology is the initial topology defined by the family
pi of projections and the earlier characterisation of continuity of maps into a space
with initial topology 6.3. 

A projection not only pulls back open sets to open sets, it also takes open sets to
open sets. A map which takes any open set (closed set) to an open set (a closed set)
is called an open map (a closed map).

Proposition 6.12 Each projection pi : X i → X i is an open map; that is, it maps
open sets to open sets.

Proof Let U be open in X , x = (xi ) ∈ U and B = j∈F p−1 j (U j ) be a basic open
set with x ∈ B ⊂ U , where F is finite. If i ∈ F, then xi ∈ pi (B) = Ui ⊂ pi (U ); if
i∈/ F, then pi (B) = X i = pi (U ). In any case, pi (U ) is a neighbourhood of xi and
the proof is complete. 
84 6 Topologies Defined by Maps

Exercise. Is a projection a closed map? e.g. consider p1 : R2 → R.

Convergence in a product space is just coordinate-wise convergence.

Proposition 6.13 A net {xα } in a product X = X i converges to an x ∈ X if and
only if pi (xα ) converges to pi (x), i.e. xα (i) → x(i) for each i.

Proof xα → x in X implies pi (xα ) → pi (x) = xi for every i, by continuity of pi .

Conversely,
 if {xα } does not converge to x, then there is a basic neighbourhood
B = j∈F p−1 j (V j ) of x such that {x α } is ‘frequently’ outside B. Then, for j ∈ F,
p j (xα ) is frequently outside the neighbourhood V j of p j (x), so p j (xα ) does not
converge to p j (x). 

Example 6.14 1. The usual topology on Rn is the same as the product topology and
continuity of f : X → Rn , f (x) = ( f 1 (x), ..., f n (x)), is equivalent to the continuity
of each coordinate function f j .
2. If (X, d), (X  , d  ) are metric spaces, metrics ρ1 , ρ2 and ρ∞ (analogous to the
metrics d1 , d2 and d∞ on R2 ) were introduced in Chap. 2, Exercises. The topology
given by each of this metric is nothing but the product topology on X × X  .
3. If d is a metric on X , then d : X × X → R is continuous.
4. Let {(X
n , dn )} be a countable collection of metric spaces and recall the metric
d on X = X n (see 2.10 (6)). Then the associated metric topology on X is just
the product topology.

6.3 Quotient Topology

A quotient space of a given space is obtained by ‘identifying certain subsets of

the space to points’. Hence these are also called ‘identification spaces’. Quotient
topology, in a specific case, was studied by the American topologist R.L.Moore in
1925 [3].

Definition 6.15 A topological space X̃ is called a quotient space of a topological

space X if there is a surjective map q : X → X̃ such that the topology of X̃ is the
final topology defined by q, i.e. a set V ⊂ X̃ is open if and only if q −1 (V ) is open in
X . In this case the (continuous) map q is known as a quotient map and the topology
of X̃ is called the quotient topology (induced by q).

We start with a set theoretic observation and use it to realise a quotient space as an
‘identification space’.
6.3 Quotient Topology 85

Lemma 6.16 For sets X and Y , the following are equivalent.

1) There is a surjection q : X → Y .
2) There is an equivalence relation ∼ on X such that Y is in one-one correspon-
dence with the set X̃ := X/ ∼ of equivalence classes.
3) There is a partition X into disjoint subsets and there is a bijection h from Y to
the set of these subsets.
Proof 1) ⇒ 2): Suppose q : X → Y is a surjection. For x, x  ∈ X , define x ∼ x 
if q(x) = q(x  ). This is an equivalence relation on X and the set X̃ of equivalence
classes is formed by the fibres q −1 (y), y ∈ Y . Define h : Y → X̃ by h(q(x)) = x̃ =
p(x), where p : X → X̃ is the natural map p(x) = x̃, the equivalence class of x.
This is well-defined since q(x) = q(x  ) implies x̃ = x̃  . This is clearly surjective and
is injective as x̃ = x̃  implies q(x) = q(x  ).
2) ⇔ 3): An equivalence relation on X gives a partition of X whose partitioning
subsets are the equivalence classes, and conversely a partition yields an equivalence
relation whose equivalence classes are the subsets giving the partition. So it is clear
that 2) ⇔ 3).
2) ⇒ 1): If h : Y → X̃ is a bijection, the map q : X → Y defined by q(x) =
h −1 (x̃) is surjective. 
Proposition 6.17 With notation as in the previous lemma, if X is a topological space,
then Y and X̃ are homeomorphic with the quotient topologies given by q and p. In
fact, h is a homeomorphism.
Proof This is easy. Assuming that p, q are quotient maps, we have to show that the
bijection h : Y → X̃ defined by h(q(x)) = p(x) satisfies the condition that Ũ ⊂ X̃
is open if and only if h −1 (Ũ ) is open in Y . Recalling that p, q are quotient maps and
that h ◦ q = p, we have
Ũ is open ⇔ p −1 (Ũ ) is open
⇔ q −1 (h −1 (Ũ )) is open
⇔ h −1 (Ũ ) is open. 
So we can identify Y and X̃ and think of a quotient space of X to be got by identi-
fying certain subsets of X (the equivalence classes) to points. Thus quotient topology
is also called identification topology. A quotient space is completely determined by
the quotient map.
When does a continuous map from X drop to a map from X̃ ?
Proposition 6.18 Let q : X → X̃ be a quotient map and Z be any topological space.
1) A map g : X̃ → Z is continuous if and only if g ◦ q is continuous.
2) If f : X → Z is continuous and is constant on the fibres q −1 (x̃), then there is
a unique continuous map f˜ : X̃ → Z such that f = f˜ ◦ q.
Proof 1) A quotient topology is a final topology and 6.5 applies.
2) f˜(q(x)) = f (x), x ∈ X, gives a well-defined map as f is constant on fibres.
Continuity follows from i), and uniqueness is clear. 
86 6 Topologies Defined by Maps

What are some conditions for a map to be a quotient map?

Proposition 6.19 A surjective continuous map q : X → Y is a quotient map if q is
either an open map or a closed map.

Proof By definition, q is a quotient map means: E is open (closed) in Y if and only

if q −1 (E) is open (closed) in X . Note that surjectivity implies E = q(q −1 (E)) for
any E ⊂ Y .
If q is an open map, V ⊂ Y and q −1 (V ) is open in X imply V = q(q −1 (V )) is
open in Y . Conversely, continuity of q ensures that V is open implies q −1 (V ) is open.
In view of the opening remarks of the proof, the same argument holds if q is a
closed map just by replacing ‘open’ by ‘closed’. 

Corollary 6.20 Any continuous map q of a compact space X onto a metric space
Y is a quotient map. The associated quotient space X̃ is homeomorphic to Y .

Proof Under the hypotheses, q is a closed map: if C is closed in X, it is compact,

hence q(C) is compact and so is closed in Y . The map h : x̃ → q(x) is a bijec-
tive continuous map of the compact space X̃ onto the metric space Y and so is a
homeomorphism by 5.25. 

The result and the proof are valid if compact sets in Y are closed; e.g. Y is a Hausdorff
space; see 8.10.

Proposition 6.21 If a continuous map q : X → Y has a right inverse; i.e. if there

is a continuous map p : Y → X such that q ◦ p = idY , then q is a quotient map.

Proof Suppose V ⊂ Y and q −1 (V ) is open. To show V is open. But if q −1 (V ) is

open, then so is p −1 (q −1 (V )) = (q ◦ p)−1 (V ) = V. 

Corollary 6.22 If E is a subspace of X and there is a continuous map r : X → E

such that r | E = id E , then r is a quotient map.

Proof If ι : E → X is the inclusion map, then r ◦ ι = id E and the proposition

applies. 

Such a map r is called a retraction and the set E is called a retract of X . These
will be studied in some detail in Chap. 14.

Example 6.23 1. In Section 1, as an example of final topology, we saw that the circle
S1 = T := {z ∈ C : |z| = 1}, which is also the one dimensional torus, is a quotient
space of the closed interval [0,1] got by identifying the end points. This also follows
from 6.20 taking q(t) = ex p(2πıt).
2. A compact cylinder is got as a quotient space of a square patch by identifying
corresponding points of one pair of opposite edges. For example, in [0, 1] × [0, 1],
identify (x, 0) with (x, 1) for each x ∈ [0, 1]. The reader is urged to write down the
details, using Fig. 6.1.
6.3 Quotient Topology 87

a2 c2 b2

U2 W2 V2

U1 W1 V1 Ũ W̃ Ṽ
ã b̃
c̃
a1 c1 b1

Fig. 6.1 Cylinder as a quotient of the square: ã = q(a1 ) = q(a2 ); b̃ = q(b1 ) = q(b2 ); c̃ = q(c1 ) =
q(c2 ).

a2 b2 a3

U2 V2 U3
Ṽ b̃

U1 U4
V1
Ũ
a1 b1 a4 ã

Fig. 6.2 Torus as a quotient of the square: ã = q(a j ), b̃ = q(bi ); Ũ is made up of the U j and Ṽ
of the Vi .

3. As in example 1 above, q : [0, 1] × [0, 1] → T2 , q(s, t) = (e2πıs , e2πıt ) is a

quotient map and the torus is a quotient of the square. Geometrically, the torus
is got by identifying each pair of opposite edges of the square. The equivalence
classes are: the class {(0, 0), (0, 1), (1, 1), (1, 0)} formed by the four corners and
for each 0 < x, y < 1, we have two-element classes, {(x, 0), (x, 1)}, {(0, y), (1, y)}.
The torus is thus got by joining together the two circular ends of the cylinder (Fig. 6.2).
4. Define an equivalence relation on R by x ∼ y if x − y ∈ Z. The quotient is
denoted by R/Z (this is just the quotient group) and q(x) = x + Z = {x + n : n ∈ Z}
gives the quotient map. The map p : R → T, p(t) = e2πıt is surjective, continuous
and open, so p is a quotient map. Thus giving R/Z the quotient topology, the map
h : R/Z → T defined by h(x + Z) = h(q(x)) = p(x) is a homeomorphism.
5. (Pinching a subspace) Let E be a subspace of a space X . Let X E denote the
set of equivalence classes {E} and the singleton sets {x} for x ∈ E c . The quotient
space X E is got by ‘pinching the subspace E to a point’. (An often used notation is
X/E. But we avoid this because it will be in conflict with more standard usage, like
88 6 Topologies Defined by Maps

X × {1}
V

[0,1]

X × {0}

Fig. 6.3 Cone (X ), got by pinching X × {1} to a point V

R/Z in the previous example.) As an example, the identification of the circle as a

quotient of [0,1] is got pinching the end-point set {0, 1} to a point.
6. (Cone) Let X be any space and let Cone(X ) denote the quotient space of
X × [0, 1] got by pinching X × {1} to a point. It is called the cone over X . If x̃t
denotes the equivalence class of (x, t), then X can be identified with the subspace of
Cone(X ) consisting of equivalence classes of (x, 0) via the homeomorphism x → x̃0 .
For example, if X ⊂ Rn is compact, we can give a concrete, geometric description
of Cone(X ). For this, identify X with the subspace X × {0} of Rn+1 . Let en+1 be the
standard unit vector (0, 0, . . . , 0, 1) in Rn+1 and X cone := {ten+1 + (1 − t)(x, 0) :
0 ≤ t ≤ 1, x ∈ X } be formed by all the line segments joining each point of X to
en+1 . This is the familiar geometric cone with ‘vertex’ en+1 and ‘base’ X . The map
p : X × [0, 1] → X cone defined by p(x, t) = ten+1 + (1 − t)(x, 0) is continuous,
surjective, and it is a quotient map by 6.20. Further, p(x1 , t1 ) = p(x2 , t2 ) if and only
if t1 = t2 = 1 and x1 , x2 ∈ X arbitrary, or 0 ≤ t1 = t2 < 1, x1 = x2 . This means that
the partition defined by p is the same as that defining Cone(X ). Hence, with their
respective quotient topologies, Cone(X ) and X cone are homeomorphic. So, we get
the familiar geometric description of the cone (Fig. 6.3).
7. As another example, we show that starting with the closed unit disc B2 :=
{(x, y) ∈ R2 : x 2 + y 2 ≤ 1} and pinching its boundary circle S1 to a point, we get the
sphere S2 . For this, take homeomorphisms h 1 : R2 → S2 \ (0, 0, 1) and h 2 : D → R2
(e.g. h 1 =stereographic projection, h 2 (x, y) = √1 2 2 (x, y), i.e. h 2 (z) = 1−|z|
z
.).
1− x +y
Define q : B2 → S2 by

h 1 (h 2 (x, y)), (x, y) ∈ D
q(x, y) =
(0, 0, 1), (x, y) ∈ S1

Then q is a quotient map, being a continuous, closed surjection. Now q −1 (0, 0, 1) =

S1 and all the other fibres are singletons. Thus, q just pinches S1 to a point. The
induced map B2S1 → S2 is a homeomorphism. The method extends to higher dimen-
sions.
8. An important way that quotient spaces arise in topology and geometry is as orbit
spaces via group actions. A group G acts on a space X if there is a homomorphism of
G into the group of homeomorphisms of X ; i.e. each g ∈ G gives a homeomorphism,
6.3 Quotient Topology 89

written x → g.x, and (gh).x = g.(h.x), x ∈ X, g, h ∈ G. Ox := G.x := {g.x : g ∈

G}, x ∈ X, the orbits, give a partition of X . The corresponding quotient space is
called the orbit space X/G. We will not give a general discussion on this, but merely
confine ourselves to a simple example.
Any nonsingular n × n real matrix A has a natural action on Rn : x → Ax. If A
is orthogonal, then A preserves the Euclidean norm: Ax = x for all x. Hence
x → Ax yields an action of the group O(n) of all orthogonal matrices on the unit
 to check that, for A ∈
sphere Sn−1 . If e1 is the standard first basis vector, itis easy
1 0
O(n), Ae1 = e1 if and only if A is of the form A = where B ∈ O(n − 1).
0B
The subgroup of O(n) formed by such matrices will be identified with O(n − 1).
The map q : O(n) → Sn−1 , A → Ae1 , is continuous and surjective (Prove!). It is
a quotient map because O(n) is compact. Moreover, q −1 (e1 ) = O(n − 1) and the
fibres are just the left cosets of O(n − 1) in O(n). Hence the quotient space is just the
coset space O(n)/O(n − 1) with quotient topology and is homeomorphic to Sn−1 ,
QO(n − 1) → Qe1 being a homeomorphism.

Exercise 6

1. Product topology. a) Let {X i } be a collection of spaces and let E i ⊂ X i for

each i. The product  topology on (each E i given the subspace topology) and
Ei 
the topology on E i as a subspace of X i are the same. (‘Product of subspaces =
subspace  of product’.)
  
b) Ai = Āi for Ai ⊂ X i . Hence
 Ai is closed if each Ai is and Ai is
dense in X i if Ai is dense in X i for each i. 
c) Fix a = (ai ) in a product space X i and  let Da be the set of all (xi ) ∈ X i
with {i : xi = ai } finite. Then Da is dense in X i .  
d) If f i : X i → Yi is a homeomorphism for each i, is the map X i → Yi , (xi ) →
( f i (xi )), a homeomorphism? 
e) Let (X n , dn ) be a sequence of metric spaces. If d is the metric on X n as in
2.3, then the d -topology is the product topology.

 2. Evaluation map. Let X, Yi be spaces, f i : X → Yi be maps and fi : X →
Yi be
 the map x  → ( f i (x)); it is the evaluation map e.
a) f i is continuous if and only if each f i is continuous.
b) e is one to one if and only if the collection { f i } separates points of X ; i.e. x = y
implies f i (x) = f i (y) for some i. 
c) e is an embedding (i.e. a homeomorphism into Yi ) if and only if { f i } separates
points of X and the topology of X is the initial topology generated by { f i }.
3. Associative law for products. a) If {X i }, {Y j }, {Z k } are families of spaces with
products X, Y, Z , respectively, then X × (Y × Z ) is homeomorphic to (X × Y ) × Z .
b) More generally, the following holds. Let I, J be index sets and suppose there is
a bijection σ : I → J . Let {X i }i∈I , {Y j } j∈J be topological
 spaces. Suppose
 there is
a continuous map f i : X i → Yσ (i) for each i. Then f i : X i → Yσ (i) defined

by (xi ) → ( f i (xi )) is continuous. If each f i is a homeomorphism, then so is fi .
90 6 Topologies Defined by Maps

 
c) Hence, if σ is a permutation of I , then X i is homeomorphic to X σ (i) ; e.g.
if I is countably infinite X I  X N for any space X .

4. Box topology. If {X i } is a family of
spaces, the product Ui of open subsets
Ui in X i form a base for a topology on X i , called the box topology. This equals
the product topology only for finite products.
a) Take RN with the box topology. Then f : R → RN , f (x) = (x, x, . . . , ) is not
continuous. What about the map g(x) = (x/n)?
b) What is the closure in RN of the set of all sequences whose terms are eventually
0 in i) the product topology, ii) the box topology.?
5. Quotients. a) The projection p1 : R2 → R is a continuous, open surjection,
but is not a closed map.
b) Let E = {(x, y) ∈ R2 : x ≥ 0} ∪ {(x, 0) ∈ R2 : x ∈ R}. Then p1 : E → R is
a quotient map which is neither open nor closed.
c) If f : X → Y is continuous, then f = g ◦ q where q : X → Z is a quotient
map and g : Z → Y is a continuous injection, for a space Z .
d) A subspace of a quotient may not be the quotient of a subspace. Make the
statement precise and justify.
6. Möbius band. Here is a ‘twisted’ version of a (compact) cylinder. Take [0, 1] ×
[0, 1] and identify (x, 0) with (1 − x, 0) for each 0 ≤ x ≤ 1. The resulting quotient is
a ‘one-sided’ surface called the Möbius band, named after the German mathematician
August Ferdinand Möbius. Elaborate on the following picture, Fig. 6.4.
7. Klein bottle. Form the quotient of [0, 1] × [0, 1] by identifying (x, 0) with
(x, 1) and (0, y) with (0, 1 − y) for all 0 ≤ x, y ≤ 1. The resulting space is a ‘twisted
torus’ called the Klein bottle, after the well-known German mathematician Felix
Klein. Draw the square and mark the identifications and try to draw a picture of the
surface.
8. Projective space. On X = Rn+1 \ {0}, define x Ry if 0 lies on the straight line
joining x and y. This is an equivalence relation and the quotient space X/R is called
the projective space RPn .
a) RPn is homeomorphic to the quotient of the sphere Sn got by identifying any
x with its antipodal point −x.
b) RP1 , the ‘projective line’, is homeomorphic to S1 .
9. Suspension. For a space X , the suspension S(X ) is the quotient X × [0, 1]/R
where R is the equivalence relation given by (x, 0)R(y, 0) and (x, 1)R(y, 1) for
x, y ∈ X , (i.e. it is obtained by pinching each of X × 0 and X × 1 to a point (Fig. 6.5).
(Picturise a double cone with base X .) What is the relation between Cone(X ) and
S(X )?
S({0, 1}) is (homeomorphic to) the circle S1 .

10. The upper half-plane. G = S L(2, R), the group of real 2 × 2 matrices of
determinant one, acts on H := {z ∈ C : I m(z) > 0}, the upper half-plane, as Möbius
transformations z → az+b
cz+d
. Check:
6.3 Quotient Topology 91

b2 c2 a2

U2 W2 V2

U1 W1 V1
a1 c1 b1

Fig. 6.4 Möbius band as a quotient of a rectangle

X × {1}
V1

[0,1]

V0
X × {0}

Fig. 6.5 Suspension S (X ) got by pinching X × {0} and X × {1}

a) if z ∈ H, then A.z ∈ H and A.(B.z)

 = (AB).z for A, B ∈ G;
cos t sin t
b) K := {A ∈ G : A.ı = ı} = { : t ∈ R};
− sin t cos t
c) the orbit O(ı) := G.ı = H;
d) the map G/K → H, given by AK → A.ı is well-defined, is a bijection and is
homeomorphism. (G/K is given the quotient topology.)
11. Disjoint unions. a) For a collection {X i } of sets, ‘make them disjoint’ and
take their union to get their disjoint union. How to make them disjoint? For example,
take X i := X i ∪ {i} for each i, so that the X i are mutually disjoint and X i may be
identified with X i via x → (x, i). ∪X i is called the disjoint union of the X i and is
written X i . If X i are all topological spaces, a topology on this can be defined by
declaring a set V ⊂ X i to be open if V ∩ X i is open in X i for each i. Check that
this makes X i into a topological space, denoted
X i .
b) If X i = X for all i ∈ I , then
X i is homeomorphic to X × I where the index
set I is given the discrete topology.
12. Attaching spaces. Let X, Y be disjoint spaces and let E ⊂ X be closed.
Suppose f : E → Y is continuous. In X + Y , pinch the set f −1 (y) ∪ {y} to a point
for each y ∈ E. The quotient space of X + Y thus obtained is denoted by X ∪ f Y
or X + f Y and is called the space got by attaching X to Y by f ; f is called the
attaching map.
92 6 Topologies Defined by Maps

a) If q is the quotient map, then the restrictions of q to E c and Y are both home-
omorphisms, q(E c ) is open and q(Y ) is closed in X + f Y .
b) If f is the constant map f (E) = y0 , then X + f {y0 } is the space X E obtained
from X by pinching the set E to a point.
c) Cone (X ) is got by attaching X × [0, 1] to a point y0 by the map f (X × {1}) =
y0 .
d) Let X, Y be disjoint compact intervals in R, say X = [0, 1] and Y = [3, 4].
Take E = {0, 1} and define f (0) = 3, f (1) = 4. Then X + f Y is homeomorphic to
the circle S1 .
13. Topological groups—products and quotients. a) The product of any
nonempty collection of topological groups is a topological group.
b) Let G be a topological group and H a subgroup. Equip the left coset space
G/H with the quotient topology. H is closed if and only if singletons are closed in
G/H and H is open if and only if G/H is discrete.

6.4 Biographical Notes

6.4.1 R. L. Moore

Robert Lee Moore (1882–1974) was an American topologist who spent most of his
career at the University of Texas, a student of the famed topologist Oswald Veblen.
He is best known for his unusual teaching method, now called the Moore method.
He had 50 doctoral students, including many well-known topologists.

6.4.2 Möbius

August Ferdinand Möbius (1790–1868) was German mathematician. He is best

known for Möbius band, a nonorientable ‘one-sided’ surface (which was indepen-
dently discovered by J. B. Listing a little earlier). He studied astronomy under Gauss
and Johann Pfaff and was a Professor of Astronomy and Higher Mechanics at
Leipzig. Homogeneous coordinates in projective geometry were introduced by him.
Möbius transformations of complex numbers, Möbius function and Möbius inversion
formula in number theory are well-known.
6.4 Biographical Notes 93

6.4.3 Klein

Felix Klein (1849–1925) was a German mathematician. His doctoral adviser at Bonn
was the geometer Julius Plücker. He was appointed as a Professor at Erlangen when
he was only 23. He moved to München, to Leipzig and, finally, to Göttingen. His
famous Erlangen Programme was initiated in his inaugural lecture at Erlangen and
was completed and published at Göttingen. It presents geometry as the study of the
properties of a space that is invariant under a group of transformations, thus unify-
ing geometry and encompassing both Euclidean and nonEuclidean geometries. The
work had had a profound influence on the development of mathematics and its ideas
have since been ingrained in general mathematical thinking. He sought to re-establish
Göttingen’s primacy in world mathematics and succeeded in bringing David Hilbert
to Göttingen. His notable contributions were in geometry, group theory, invariant the-
ory, number theory, elliptic modular functions, automorphic functions and reforms to
mathematical education in secondary schools. Klein bottle, his nonorientable surface,
cannot be embedded in the three-dimensional Euclidean space.

References

1. Tychonoff, A.: Über die Topologische Erweiterung von Räumen. Math. Ann. 102, 544–561
(1930). https://doi.org/10.1007/BF01782364
2. Čech, E.: On bicompact spaces. Ann. Math 2(38), 823–844 (1937). https://doi.org/10.2307/
1968839
3. Moore, R.L.: Concerning upper semi-continuous collections of continua. Trans. Am. Math. Soc.
27, 416–428 (1925). https://doi.org/10.2307/1989234
Chapter 7
Products of Compact Spaces

Tychonoff’s product theorem is one of the most useful and important results of
topology. Here is how Tychonoff stated it in his celebrated paper.
Betrachten man unter den Elementen ... nur diejenigen Funktlonen f(x), die der
Ungleichung −k ≤ f (x) ≤ k, 0 ≤ x ≤ 1, für eine feste natürliche Zahl k genügen,
so erhält man eine Teilmenge Fk ... , die ihrerseits als topologischer Raum (als Rela-
tivraum ...) betrachtet werden kann. Von diesen Räumen Fk beweisen wir nun, daßsie
bikompakte Hausdorffscke Räume sind. (Tychonoff, 1930, [1]).
He says that ‘the subset of those functions f such that −k ≤ f (x) ≤ k, 0 ≤ x ≤ 1,
for a fixed natural number k, can be viewed as a topological space (relative subspace).
We now show that these spaces are bicompact, Hausdorff.’ In modern notation, this
says that the product space [−k, k][0,1] is compact (and Hausdorff). Later, Tychonoff
remarked that his proof was valid for products of arbitrary compact spaces.
Recall (Chap. 6) that Tychonoff defined the product topology only on [0, 1][0,1]
in [2] and this was carried over to products of arbitrary spaces by Čech, [3]. Čech
also gave the proof of the general form of Tychonoff’s theorem. He stated the result
simply as follows.
The Cartesian product S = Pi Si of any family of bicompact spaces is a bicompact
space.

7.1 Tychonoff’s Theorem

We begin with Kuratowski’s charming characterisation of compactness in terms of

projections.

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 95
K. Parthasarathy, Topology, La Matematica per il 3+2 134,
https://doi.org/10.1007/978-981-16-9484-4_7
96 7 Products of Compact Spaces

Proposition 7.1 (Kuratowski)

A topological space X is compact if and only if the projection q : X × Y → Y is a
closed map for each space Y .
Proof Let X be compact, F be closed in X × Y and (yα ) be a net in q(F) converging
to y0 ∈ Y . Choose xα ∈ X with (xα , yα ) ∈ F. Since X is compact, (xα ) has a conver-
gent subnet xαη → x0 . Then (xαη , yαη ) → (x0 , y0 ) ∈ F, y0 = q(x0 , y0 ) ∈ q(F) and
so q(F) is closed.
Conversely, suppose X is not compact. Then there is a net (xα )α∈D in X without
cluster points. Let D ∗ = D ∪ {∞} for some point ∞ ∈ / D. Topologise D ∗ so that D
has the discrete topology and neighbourhoods of ∞ are of the form Vα := {β ∈ D :
β ≥ α} ∪ {∞} for α ∈ D. Take F = Ā where A = {(xα , α) : α ∈ D} ⊂ X × D ∗ .
If (x, ∞) ∈ F for some x ∈ X , then for any neighbourhood U of x and β ∈ D
there is an α ≥ β such that xα ∈ U . Thus x would be a cluster point of (xα ), a
contradiction. Hence (x, ∞) ∈ / F for any x ∈ X and so q(F) ⊂ D. But q(A) = D
and so q(F) = D, which is not closed in D ∗ . 
The projection in Kuratowski’s theorem is what is known as a ‘proper map’. A
continuous map f : X → Y is called a proper map (or a perfect map) if it is a closed
map and each fibre f −1 (y) is compact.
Because of the importance of the product theorem, we give proofs with different
flavours, first in the finite case and then in full generality. Our first proof of the next
result uses Kuratowski’s result which we proved using nets. The reader should try to
give a direct proof using nets for the finite case.
Theorem 7.2 If X and Y are compact spaces, then so is X × Y .
Proof 1. This is almost immediate from Kuratowski’s result. We have to show that,
for any space Z , the projection X × Y × Z → Z is a closed map. This projection is
the composition of the projections X × Y × Z → Y × Z and Y × Z → Z . These
are both closed maps by compactness of X and Y , respectively, by Kuratowski. 
Proof 2. (Using finite intersection property)
Let A be a collection of subsets of X × Y having finite intersection property. To
show that ∩{ Ā : A ∈ A} is nonempty. Let B be the collection of finite intersections
of members of A and let C be the collection of subsets of X × Y which intersect every
member of A. Then {E i } = E := A ∪ B ∪ C has the finite intersection property.
Let p, q be the projections from X × Y to X , Y respectively. If Ai = p(E i ), Bi =
q(E i ), then {Ai }, {Bi } have finite intersection property and, by compactness, there are
x0 ∈ ∩ Āi , y0 ∈ ∩ B̄i . Take a basic neighbourhood U0 × V0 of (x0 , y0 ). Then p−1 (U0 )
and q−1 (U0 ) meet each E i and so E i intersects p−1 (U0 ) ∩ q−1 (U0 ) = U0 × V0 for
all i. Thus (x0 , y0 ) ∈ Ē i for all i and ∩ Ē i = ∅, hence ∩{ Ā : A ∈ A} = ∅ too. 
Any finite product of compact spaces is compact, either using analogous proofs or by
induction. None of the proofs, however, carries over to arbitrary products. We give
two approaches—open covers and nets. More proofs are indicated in the exercises.
Here is a direct, simple ‘open cover’ proof, based on Étienne Matheron’s presentation,
[4].
7.1 Tychonoff’s Theorem 97

Theorem 7.3(Tychonoff’s theorem)

The product i∈I X i of any collection of compact spaces is compact.


Proof (via open covers) First we set up some needed notation. Let X = i∈I X i ;
recall that it is the set of all maps x with domain  I such that x(i) ∈ X i for i ∈ I (See
appendix below). For J ⊂ I, we call X J = j∈J X j a partial product. An element
u = u J ∈ X J is a map with domain J such that u( j) ∈ X j for j ∈ J and will be
called a partial map. If u, u are partial maps with disjoint domains J, J , then write
u  u ∈ X J × X J for the partial map with domain J ∪ J whose restrictions to
J, J are, respectively, u, u . Then x ∈ X may be identified with x| J  x| I \J . Note
that elements of X J are precisely the restrictions x| J , x ∈ X = X I and x → x J is
just the projection p J : X → X J . Observe also that when J = ∅, the only partial
map is the empty map and so X ∅ = {∅}.
We can now begin the proof. Suppose X is not compact and {Ua }a∈A is an open
cover of X with no finite subcover. Let P be the set of all partial maps u such that
for any neighbourhood V J of u in X J , the pull back p−1 J (V J ) cannot be covered by
finitely many Ua , where J = domain u. Note that the assumption on the open cover
means that the empty partial map ∅ ∈ P. To complete the proof, we show that there
is an x ∈ P with domain I , i.e. the partial map x is a full map: x ∈ X . We get this x
by applying Zorn’s lemma to P.
Step 1: Partial order on P. Define the partial order as the natural extension of
maps: u  u if domain u = J ⊂ J = domain u and u = u | J . In this step, we show
that if u ∈ P and v  u, then v ∈ P. Write domain u = J and domain v = K . If
V is a neighbourhood of v in X K and if q = q J K is the projection X J → X K , then
q−1 (V ) is a neighbourhood of u in X J . Since u ∈ P, p−1 −1
J (q (V )) cannot be covered
by a finite number of Ua . But p K = q ◦ p J and so p J (q (V )) = p−1
−1 −1
K (V ) and we
conclude v ∈ P.
Step 2: In this step, we apply Zorn’s lemma to get a maximal element in P. To
ensure the applicability of Zorn’s lemma, we prove that any chain C in P has an upper
bound in P. To get an upper bound, take J0 to be the union of the domains of all partial
maps u ∈ C. Then u 0 on J0 given by u 0 ( j) = u( j) if j ∈ J =domain u, u ∈ C, is well
defined as C is a chain. To see that u 0 ∈ P it is enough to check finitely many Ua do
not cover p−1 n −1
J0 (U ) for any basic neighbourhood U of u 0 in X J0 , U = ∩1 p jk (W jk ),
jk ∈ J0 , 1 ≤ k ≤ n, W jk ⊂ X jk open. Write F = { j1 , · · · , jk } and u F = u| F . Since
F is finite and C is a chain, there is a v0 ∈ C  with u F  v0 . But then, by Step 1,
u F ∈ P. Note that U = q−1 J0 F (W ) where W = k W jk is a neighbourhood of u F in
X F . Hence p−1 F (W ) = p −1
J0 (U ) cannot be covered by finitely many Ua . Thus u 0 ∈ P,
C has an upper bound in P and Zorn’s lemma yields a maximal element x0 ; let
domain x0 = I0 .
Step 3: We complete the proof by proving that I0 = I and x0 ∈ X .
Assume there is an i ∈ I \ I0 . For a u ∈ X i consider the partial map x0  u. Write
x0u = x0  u and domain x0u = I0i = I0 ∪ {i}. We claim that x0u ∈ P for some u ∈ X i .
Suppose not. Then, for each u ∈ X i , there is a neighbourhood Vu = Wu × Nu of x0u
98 7 Products of Compact Spaces

in X I0i , with Wu a neighbourhood x0 in X I0 and Nu a neighbourhood of u in X i , such

that p−1Ii
(Vu ) can be covered by finitely many Ua . Since X i is compact, the open cover
0
{Nu : u ∈ X i } has a finite subcover, {Nu j }k1 . Then Wx0 = ∩k1 Wu j is a neighbourhood
of x0 in X I0 and
p−1 k −1 k −1
I0 (W x0 ) = ∪1 p I i (W x0 × Nu j ) ⊂ ∪1 p I i (Vu j )
0 0

can be covered by finitely many Ua since each set on the right can be thus covered.
This contradicts the maximality and shows that I0 = I. Thus x0 ∈ X and so x0 ∈ Ua
for some a. Since p I = id X , p−1
I (Ua ) = Ua is covered by a single Ua , a contradiction
as x0 ∈ P. Thus the assumption that {Ua }a∈A has no finite subcover is untenable. 
A simple proof of Tychonoff’s theorem using nets was given by Paul Chernoff
[5] in 1992. We now give a variation of this proof, following Kenneth Ross [6]. The
concluding, clinching part of this proof is separated as a little lemma due to Charles
Pugh (2003), see [6].
Lemma 7.4 Let X be any topological space and Y a compact space. Let {(xα , yα )}α∈D
be a net in X × Y . If (xα ) has a cluster point x in X , then (x, y) is a cluster point of
{(xα , yα )} for some y ∈ Y .
Proof Let D  := {(α, Ux ) : α ∈ D, Ux ∈ Ux , xα ∈ Ux }. It is a directed set with
order defined by (α, Ux )  (β, Vx ) if α ≤ β and Vx ⊂ Ux . Define η : D  → Y by
η(α, Ux ) = yα . This is a net in the compact space Y , so has a cluster point y ∈ Y .
We prove that (x, y) is a cluster point of the net {(xα , yα )}. For this, take a basic
neighbourhood U × V of (x, y). Since x is a cluster point of (xα ), for any α ∈ D
there is a β ≥ α such that xβ ∈ U and since y is a cluster point of η, there is a
(γ , Vx )  (β, U ) with yγ = η(γ , Vx ) ∈ V .This means that γ ≥ α and (xγ , yγ ) ∈
U × V proving that (x, y) is a cluster point of the net {(xα , yα )}. 
Proof (Tychonoff’s Theorem via nets.)

Let X = i∈I X i be a product of compact spaces. It suffices to show that every
net (xα ) in X has a cluster point.
A partial net of (xα ) is a net of the form (xα | J ) in X J for some J ⊂ I . A cluster
point of a partial net is a partial cluster point of (xα ). When J = { j}, the partial net
(xα | J ) is a net in X j and, by compactness, has a cluster point. Hence the set P of all
partial cluster points of (xα ) is nonempty. P has a natural order: x J ≤ x K if J ⊂ K
and x K | J = x J . Our strategy is to get a maximal partial cluster point in P via Zorn’s
lemma and prove that it is really a cluster point.
To be able to invoke Zorn’s lemma, we show that every totally ordered subset
T ⊂ P has an upper bound in P. To do that, let J0 = ∪{J : x J ∈ T } and define
y0 ( j) = x J ( j) for j ∈ J with x J ∈ T . This is well-defined since T is totally ordered.
We have to show that y0 ∈ P, i.e. that y0 is a cluster point of (xα | J0 ). For this, take
a basic neighbourhood U0 of y0 in X J0 , U0 = ∩ j∈F p−1 j (U j ) where F ⊂ J is finite
and U j is open in X j for j ∈ F. As T is totally ordered, there is an x K ∈ T with
F ⊂ K . Since x K is a cluster point of (xα | K ), for any β there is an α ≥ β such that
xα ( j) ∈ U j for j ∈ F. This means xα | J0 ∈ U0 . We have thus proved that the xα | J0
7.1 Tychonoff’s Theorem 99

is frequently in every neighbourhood of y0 . Thus y0 is a cluster point of (xα | J0 ) and

y0 ∈ P. Thus Zorn’s lemma applies to give a maximal element x0 := x I0 ∈ P.
Finally, we complete the proof by proving that I0 = I and x0 is a cluster point
of the net (xα ). Suppose that there is an ι ∈ I \ I0 and let I = I ∪ {ι}. Then (xα | I )
is a partial net in X I . Since (xα | J0 ) has a cluster point y0 and X ι is compact, the
lemma shows that (xα | I ) has a cluster point in X I = X I0 × X ι , contradicting the
maximality of x I0 . Hence I0 = I and the proof is complete. 

Example 7.5 1. Products of copies of [0,1] (‘cubes’) are compact. In particular, the
infinite dimensional cube [0, 1]N is compact.
2. Infinite products of finite discrete spaces are compact (and nondiscrete!). For
example {0, 1}N is compact. Observe that {0, 1}N is the space of sequences whose
terms are either 0 or 1.

The Axiom of Choice was used in both the proofs, in the form of Zorn’s lemma.
It is unavoidable. In fact, S.Kakutani conjectured in 1935 and Kelley proved in 1950
[7] that Tychonoff’s theorem is equivalent to the Axiom of Choice. Here is Kelley’s
simple proof.

Theorem 7.6 (Kelley, 1950)

Tychonoff’s Theorem implies the Axiom of Choice.

Proof Assume Tychonoff’s theorem. We prove the Axiom of Choice in one of its
equivalent formulations: the product of any nonempty collection of nonempty sets
is nonempty.
So let {Si : i ∈ I } be such a collection. When the index set is finite, the choice
axiom is not needed, so we assume this case and suppose that I is infinite. Let X i be
the disjoint union X i = Si  {i}. Equip X i with the topology Ti = Tco f ∪ {i}, i.e. the
 set {i}.
open sets in X i are the empty set, complements of finite sets and the one-point
Each X i is compact and Si is closed in X i , hence so are pi−1 (Si ) in X = X i , where
pi are the projections. We show: {pi−1 (Si )} has finite intersection property. Let F be
a finite subset of I . Take s j ∈ S j for j ∈ F (No Axiom of Choice!). Define x ∈ X
by x( j) = s j for j ∈ F and x(i) = i for i ∈ I \ F. Then x ∈ ∩ j∈F p−1 j (S j ). Thus
the family {pi−1 (Si )} of closed sets in the compact  space X has finite intersection
property and hence ∩i∈I pi−1 (Si ) is nonempty. But i∈I Si = ∩i∈I pi−1 (Si ) and the
proof is complete. 

7.2 Appendix: Axiom of Choice

Suppose C is a nonempty collection of nonempty sets of natural numbers. Each

C ∈ C has a smallest element n C . Define χ (C) = n C . This gives a function with
domain C such that χ (C) ∈ C. The definition of this function rests on the property
that every nonempty set of natural numbers has a smallest element. If members
100 7 Products of Compact Spaces

of C are arbitrary nonempty sets, can we still get such a function? Does such a
function exist? Such a function, if it exists, is called a choice function. If each C ∈ C
is well-ordered, then we can define χ as before. Otherwise, it is not clear that a
choice function always exists. One may feel that since each C is nonempty, one can
‘choose’ an element n C ∈ C and define χ as before. But is there a way of ‘choosing’
an n C for each C ∈ C ? As observed earlier, if every set can be well-ordered, then our
argument works. However, the statement that every nonempty set can be well-ordered
is equivalent to the existence of a choice function.
The edifice of modern mathematics rests on the foundations of set theory initiated
by the work of Cantor and mathematical logic. It was realised that indiscriminate use
of set theory led to contradictions and paradoxes. These logical inconsistencies led
to an axiomatic development of set theory. There are several axiomatic approaches:
Zermelo–Fraenkel theory [8, 9]; see also [10], Hilbert–Bernays–von Neumann sys-
tem, etc. Most mathematicians take a naive point of view of set theory and we do
the same in this book. We are not going to present any axiomatic treatment of set
theory. (See Halmos, [11] for a nice, accessible treatment.) In fact, we are not going
to discuss set theory at all. Basic concepts, standard notation and results of ‘set
theory for a working mathematician’ are all assumed. In the proof of Tychonoff’s
theorem, Axiom of Choice is used in its equivalent form of Zorn’s lemma, and con-
versely, Tychonoff’s theorem implies the Axiom of Choice. Our purpose here is to
merely mention some of the related formulations of one of the most famous axioms
in modern mathematics.
A partial order on a nonempty set X is a relation  that is reflexive, transitive and
antisymmetric. A set with a partial order is called a partially ordered set. A partial
order is a total order if for any two x, y ∈ X , either x  y or y  x. A totally ordered
set is also called a chain. A partially ordered set X is well-ordered if every nonempty
subset of X has a least element. An element M of a partially ordered set X is maximal
if M  x ∈ X implies M = x. A minimal element is defined analogously. An upper
bound for A ⊂ X is an element u ∈ X such that a  u for all a ∈ A.
Examples: R, with the usual order, is totally ordered, but not well-ordered. The
same is true for Q and Z, but N is well-ordered. The set P(X ) of all subsets of a set
X is partially ordered, but not totally ordered, by set inclusion.
Zorn’s Lemma: If X is a partially ordered set such that every totally ordered
subset has an upper bound in X, then X has a maximal element.
Axiom of Choice: For every nonempty collection of nonempty sets, there is a
choice function.
Theorem: The following are equivalent:
1) Axiom of Choice
2) Given a collection C of disjoint nonempty sets, there is a set A consisting of
exactly one element from each C ∈ C . (Zermelo Postulate)
3) The Cartesian product of any nonempty collection of nonempty sets is nonempty.
4) If E is a collection of sets and C is a chain in E , then there is a maximal chain
M in E containing C . (Hausdorff Maximal Principle)
5) On any nonempty set X, there is a well-ordering. (Well-ordering Principle)
7.2 Appendix: Axiom of Choice 101

See Kelley [12] or Moore [13], for example. Kelley takes the Hausdorff maximal
principle as an axiom and proves the others. Of these, Zorn’s lemma is the most
commonly used formulation these days. We have used it in the proof of Tychonoff’s
theorem. For its use in algebra, see the exercises. Kuratowski (1922) and Salomon
Bochner (1922) had given versions of Zorn’s lemma much before Max Zorn (1935).
Axiom of Choice and its equivalents have a tangled history. It has a large number
of equivalents in various fields. There are whole books devoted exclusively to such
matters, e.g. G.H Moore’s tome, [13].
One final word. The place of Axiom of Choice in set theory had been the object
of long, vigorous study. Kurt Gödel showed (1938) that the negation of the Axiom is
consistent with Zermelo–Fräenkel set theory. Paul Cohen (Fields medallist) proved
(1963) that Axiom of Choice itself is consistent with Zermelo–Fräenkel set theory.
Thus Axiom of Choice is independent of Zermelo–Fräenkel set theory, [14, 15].

Exercise 7

1. Proper maps. a) A space X is compact if and only if the map of X to a

one-point space is proper.
b) For a continuous map f : X → Y , the following are equivalent:
i) f is a proper map.
ii) f × id Z : X × Z → Y × Z is a proper map for any space Z .
iii) f × id Z : X × Z → Y × Z is a closed map for any space Z .
c) A proper map pulls back compact sets to compact sets.
2. Cube. The product space [0, 1][0,1] is compact, but is not sequentially compact.
3. Tube Lemma. Let A ⊂ X, B ⊂ Y be compact. If W is open in X × Y and
A × B ⊂ W , there are open sets U ⊂ X, V ⊂ V with A × B ⊂ U × V ⊂ W . Can
compactness be dropped?
4. Tychonoff via Alexander subbase lemma.(Adapted from [12]).
a) A purely set-theoretic lemma: Let X be a nonempty set. Suppose S, B are collec-
tions of subsets of X with the property that every member of B is a finite intersection
of members of S. If every cover of X by members of S has a finite subcover, then
the same is true for the collection B.(Suppose, on the contrary, that every cover from
S has a finite subcover and that there is a cover C ⊂ B without a finite subcover. By
Zorn’s lemma, there is a maximal collection C ∗ containing C with the property. Then
S ∩ C ∗ itself is not a cover for X . This leads to a contradiction. )
b) Apply a) to a subbasic open cover S of a space X and the corresponding basic
open cover B to get that if every subbasic open cover of X has a finite subcover, then
so does every basic open cover.
c) Alexander subbasis theorem: X is compact if every subbasic open cover of a
space X has a finite subcover.
102 7 Products of Compact Spaces


d) Apply the preceding considerations to a product X = X i of compact spaces,
the subbase S consisting of sets of the form pi−1 (Vi ) (pi being the projections and
Vi open in X i ) and the associated base to get a proof of Tychonoff’s theorem.
5. Tychonoff using f.i.p.(Bourbaki) a) Let X be a nonempty set and let F be a
collection of subsets of X with f.i.p. Then there is a family E ⊃ F with f.i.p. property
and satisfying the following conditions: i) if E ∈ E and F ⊃ E, then F ∈ E; ii) if
E 1 , E 2 ∈ E then E 1 ∩ E 2 ∈ E; iii) if a subset E 0 of X intersects every set in E, then
E 0 ∈ E.(P be the family of all collections of subsets of X with f.i.p. that contain F,
partially ordered by inclusion. Apply Zorn’s lemma to get a maximal collection E;
this collection
 satisfies the requirements.) 
b) { Ē : |E ∈ E} is nonempty, hence so is { F̄ : F ∈ F}, completing the proof
of Tychonoffs theorem.
6. Alaoglu Theorem. a) Let X be a normed space with closed unit ball B.
Let B ∗ denote the set of scalar-valued linear maps f on X satisfying the property
| f (x)| ≤ x for all x ∈ X . If D is the closed unit disc in the complex plane, then
B B∗ = { f | B : f ∈ B ∗ } is a subset of D B since | f (x)| ≤ 1 for x ∈ B and f ∈ B ∗ . It
is a closed subset of the product D B and hence is compact by Tychonoff’s theorem.
b) Suppose X as in a) is also an algebra with identity satisfying x y ≤ xy
for all x, y ∈ X (so X is normed algebra). Then the set (X ) := { f | B : f ∈
B ∗ , f (x y) = f (x) f (y)∀x, y ∈ X } ⊂ D B is also compact. (Actually, for any linear
functional f on X satisfying the condition f (x y) = f (x) f (y) for all x, y (multi-
plicative linear functional), the condition f ∈ B ∗ is automatic. The topology of B ∗
as a subspace of D B is called the weak ∗ topology in functional analysis.)

7.3 Biographical Notes

7.3.1 Bourbaki

Nicolas Bourbaki is the pseudonym of a group of leading French mathematicians

(named after a nineteenth-century French general). Henri Cartan and André Weil
were to teach calculus at Strasbourg jointly. They found no suitable text book, espe-
cially a satisfactory treatment of Stokes’ formula. To avoid such repeated problems,
Weil came up with idea of writing a book themselves. Thus was born Bourbaki. For
the story in Weil’s own words, see [16]. Thus the initiative to form a group came from
André Weil (1906–1998), one of the brightest luminaries of the twentieth century
mathematics, when another giant Henri Cartan (1904–2008) complained about the
lack of a suitable calculus book. The founding members were Weil, Cartan, Claude
Chevalley (1909–1984) (a very influential figure in the mathematics of the century),
Jean Dieudonné (1906–1992) (the scribe for the group) and Jean Delsarte (1903–
1968) (known for his work in Analysis). Although the initial plan was only to write a
book on calculus, later the project became more ambitious in scope, seeking to write
7.3 Biographical Notes 103

out definitive tomes on all of pure mathematics. The group met periodically and had
fierce, heated, thorough discussions on each one’s draft on the subject under consid-
eration until unanimous agreement was reached for the final draft. Armand Borel,
a later Bourbaki talks of ‘a truly unselfish, anonymous, demanding work by people
striving to give the best possible exposition of basic mathematics’ and says ‘the activ-
ity within Bourbaki was a tremendous education, a unique training ground, obviously
a main source of the breadth and sharpness of understanding... The requirement to
be interested in all topics clearly led to a broadening of horizon’, [17]. (Quoted with
the permission of the American Mathematical Society.) These sentiments are shared
by many members of the group (e.g. Dieudonné [18] and Cartan [19], p xix). The
proposed series was called Éléments de Mathématique and volumes on Theory of
Sets, General Topology, Real Analysis, Algebra, Commutative Algebra, Topological
Vector Spaces, Spectral Theory, Integration, Lie Groups and Lie Algebras, History of
Mathematics and, after a gap of several decades, Algebraic Topology have appeared
so far. The group is ‘dynamic’: older members are retired and new, younger mem-
bers are drafted. The books are known for their rigour, generality and emphasis on
mathematical structure. They have had considerable influence on the development
of mathematics. In its periodic seminars, Séminaires Bourbaki, significant recent
developments are discussed by experts and the proceedings are published.

7.3.2 Čech

Eduard Čech (1893–1960) was a Czech mathematician known for his work on
topology and geometry. Čech cohomology and Stone-Čech compactification are
well-known. The first published proof (1937) of the general form of Tychonoff’s
theorem is due to him.

7.3.3 Tychonoff

Andrey Tychonoff (1906–1993) was a Soviet mathematician and geophysicist. He

was a student of Alexandroff at the Moscow State University. The product topology
is also called the Tychonoff topology because he was the one to define it. He is
best known for the theorem on product of compact spaces. He first proved, in 1930,
that products of copies of [0, 1] are compact and later gave the general result in
1935. The embedding theorem for completely regular spaces in cubes was proved
by him and completely regular spaces are also called Tychonoff spaces. Stone-Čech
compactification was defined by him and Čech studied its properties. He also proved
the Urysohn metrisation theorem that regular second countable spaces are metrisable
(1926). (Urysohn’s paper on this topic, assuming normality in place of regularity,
was posthumously published in 1925.) He founded the theory of asymptotic analysis
104 7 Products of Compact Spaces

for differential equations and proved uniqueness results for the heat equation. His
work on ill posed problems is also well-known.

7.3.4 Kelley

John Kelley (1916–1999) was an American mathematician who spent most of his
academic life at Berkeley. He mainly worked in topology and functional analysis.
The concept of a subnet was introduced by him in 1950 and, in fact, the term ‘net’
itself was coined by him. His 1955 book General Topology, one of the earliest in
English, was a classic. Its Appendix on set theory is an oft-cited short and succinct
presentation of the foundations, now called Morse–Kelley Set Theory. He also proved
the equivalence of the Tychonoff product theorem and the Axiom of Choice. His
other books include Linear Topological Spaces (with I. Namioka) and Measure and
Integral (with T.P.Srinivasan). He was visiting the Indian Institute of Technology,
Kanpur, India, (the current author’s alma mater) during the formative years of the
Institute. This author first learned the subject from Kelley’s book. (The course was
taught by the author’s teacher at Kanpur, U.B.Tewari, an alumnus of UC, Berkeley.)
Solving the exercises was as much excitement as reading the text was a pleasure.

References

1. Tychonoff, A.: Über einen Funktionenraum. Math. Ann. 111, 762–766 (1935). https://doi.org/
10.1007/BF01472255
2. Tychonoff, A.: Über die Topologische Erweiterung von Räumen. Math. Ann. 102, 544–561
(1930). https://doi.org/10.1007/BF01782364
3. Čech, E.: On bicompact spaces. Ann. Math 2(38), 823–844 (1937). https://doi.org/10.2307/
1968839
4. Matheron, E.: Three proofs of Tychonoff’s theorem. Am. Math. Monthly 127, 437–443 (2020).
https://doi.org/10.1080/00029890.2020.1718951
5. Chernoff, P.R.: A simple proof of Tychonoff’s theorem via nets. Am. Math. Monthly 99, 932–
934 (1992). https://doi.org/10.1080/00029890.1992.11995956
6. Ross, K.A.: Appendix to Informal Introduction to Set Theory (2003). https://pages.uoregon.
edu/math/people/ross/SetTheoryAppendix.pdf
7. Kelley, J.L.: The Tychonoff product theorem implies the axiom of choice. Fund. Math. 37,
75–76 (1950). https://doi.org/10.4064/fm-37-1-75-76
8. Zermelo, E.: Untersuchungen über die Grundlagen der Mengenlehre I. Math. Ann. 65, 261–281
(1908). https://doi.org/10.1007/BF01449999
9. Fränkel, A.: Zu den Grundlagen der Cantor-Zermeloschen Mengenlehre. Math. Ann. 86, 230–
237 (1922). https://doi.org/10.1007/BF01457986
10. Fränkel, A.: Abstract Set Theory. North-Holland (1953)
11. Halmos, P.R.: Naïve Set Theory. Affiliated East-West Press (1960)
12. Kelley, J.L.: General Topology. Affiliated East-West Press (1969)
13. Moore, E.H.: General Analysis I. Part II. Mem American Philosphical, Society (1939)
14. Cohen, P.J.: Independence Results in Set Theory, Theory of Models, Proceedings of Sympo-
sium, Berkeley, 1963, North Holland, pp. 39–54 (1965)
References 105

15. Gödel, K.: The consistency of the axiom of choice and the generalized continuum hypothesis
with the axioms of set theory. Uspehi Mat Nauk (N.S.) 3, 96–149 (1948)
16. Weil, A.: The Apprenticeship of a Mathematician (Translated from the French by J. Gage.)
Birkhäser (1992)
17. Borel, A.: Twenty-Five Years with Nicolas Bourbaki (1949–1973). Not. Am. Math. Soc. 45,
373–380 (1998)
18. Dieudonné, J.A.: Une Généralisations des Espaces Compacts. J. Math. Pures. Appl. 23, 65–76
(1944)
19. Cartan, H.: Oeuvres, vol. 1. Springer (1979)
Chapter 8
Separation Axioms

‘Separation’ here refers to separation of points and closed sets by open sets. Haus-
dorff, [1], states the separation axioms as follows:
Trennungsaxiome.
Ist x1 = x2 , so gibt es eine offene Menge G 1 die x1 , aber nicht x2 enthält.
Ist x1 = x2 , so gibt es zwei disjunkte offene Mengen G 1 , G 2 mit x1 ∈ G 1 , x2 ∈ G 2 .
Ist F2 eine den Punkt x1 nicht enthaltende abgeschlossene Menge, so gibt es zwei
disjunkte offene Mengen G 1 , G 2 mit x1 ∈ G 1 , F2 ⊂ G 1 .
Sind F1 , F2 zwei disjunkte abgeschlossene Mengen, so gibt es zwei disjunkte
offene Mengen G 1 , G 2 mit F1 ⊂ G 1 , F2 ⊂ G 2 .
(‘Separation axioms:
If x1 = x2 , there is an open neighbourhood G 1 of x1 not containing x2 .
If x1 = x2 , there are disjoint open sets G 1 , G 2 with x1 ∈ G 1 , x2 ∈ G 2 .
If F2 is a closed set not containing x1 , there are two disjoint open sets G 1 , G 2
with x1 ∈ G 1 , F2 ⊂ G 2 .
If F1 , F2 are disjoint closed sets, there are two disjoint open sets G 1 , G 2 with
F1 ⊂ G 1 , F2 ⊂ G 2 .’)
Alexandroff-Hopf [2] call these: Fréchetsches, Hausdorffsches, Vietorissches and
Tietzesches Trennungsaxiome; these are now called T1 , T2 (Hausdorff), T3 (regularity)
and T4 (normality) axioms, respectively. Regularity and normality were defined by
Vietoris [3] and Tietze [4], respectively. T1 -spaces (Riesz spaces or Fréchet spaces)
were first considered by Riesz [5] and by Fréchet [6]. The German term Trennungsax-
iome, meaning ‘separation axioms’, was first used by Tietze [4] in 1923.
Separation axioms Tk , for each k ∈ {0, 1, 2, 3, 4, 5, 3 12 } have been studied (you
know where the T comes from!). We discuss only some of the important ones. A
word of warning on terminology: the use of regular, normal etc. vary considerably
in different references. For instance, in Kelley [7] normality means separation of
closed sets and T4 means normal+T1 . The reverse usage can also be seen in some

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 107
K. Parthasarathy, Topology, La Matematica per il 3+2 134,
https://doi.org/10.1007/978-981-16-9484-4_8
108 8 Separation Axioms

references. Similar remarks apply to regularity and complete regularity. So, while
looking at any reference the reader should check the usage.

8.1 Hausdorff Spaces

We begin with the separation axiom that was given by Hausdorff initially as a part of
the definition of a topological space. For most purposes in analysis and differential
geometry, Hausdorff spaces are the most basic.
Definition 8.1 A space X is called a Hausdorff space or a T2 -space if distinct points
can be separated by disjoint open sets, i.e. if x, y ∈ X, x = y, there are disjoint open
sets U, V with x ∈ U , y ∈ V .

Proposition 8.2 A metric space is a Hausdorff space.

Proof Let (X, d) be a metric space and x, y ∈ X, x = y. Then the open balls B(x; ε)
and B(y; ε) are disjoint if ε < d(x, y)/2. 

We look at some characterisations and properties of these spaces. We have

observed that a sequence or a net may converge to more than one point. A nice
thing about Hausdorff spaces is that this pathological behaviour is impossible in
these spaces, just as for metric spaces.
Proposition 8.3 In a space X , the following are equivalent:
1) X is Hausdorff.
2) If x ∈ X , then for any y ∈ X, y = x, there is a neighbourhood U of x such
that y ∈
/ Ū . 
3) For any x ∈ X , {Ū : U ∈ Ux } = {x}.
4) A convergent net in X has a unique limit.
5) The diagonal  = {(x, x) : x ∈ X } is closed in X × X .

Proof 1) ⇒ 2): If x = y, then there are disjoint open sets U, V such that x ∈ U, y ∈
V . Then U ⊂ V c , so Ū ⊂ V c and hence y ∈ / Ū .
2) ⇒ 3): If y = x, 2) says that there is a neighbourhood U of x such that y ∈ / Ū
and so y is not in the intersection mentioned in 3).
3) ⇒ 1): Let x, y ∈ X, x = y. By 3) there is an open set U with x ∈ U but y ∈ / Ū .
So V = Ū c is open, is disjoint from U and y ∈ V .
1) ⇒ 4): If x = y, disjoint open sets U, V separate them, so no net can converge
to both the points, as it cannot be eventually in both.
4) ⇒ 5): If  is not closed, there is a net {(xi , xi )} converging to (x, y) with
x = y. But then {xi } converges to both x and y.
5) ⇒ 1): If x = y, then (x, y) ∈
/ . So, if  is closed, then are open sets U, V in
X such that (x, y) ∈ U × V ⊂ c . But this means that U, V are disjoint neighbour-
hoods of x, y respectively. 
8.1 Hausdorff Spaces 109

The property that one-point sets are closed is equivalent to the T1 axiom, the first in
the list above.
Example 8.4 1. In an infinite set X with the cofinite topology, finite sets are closed,
but X is not Hausdorff; in fact, no two nonempty open sets in X are disjoint. But it
is T1 .
2. Z with arithmetic progression topology is Hausdorff.

Proposition 8.5 In a Hausdorff space, one-point sets are closed and so the space is
T1 .

Proof This is immediate from the definition. Let X be Hausdorff and x ∈ X . If

y ∈ X \ {x} and if U, V are disjoint neighbourhoods of x, y respectively, then V ⊂
X \ {x} and so X \ {x} is open. 

Proposition 8.6 In a Hausdorff space, a point and a compact set can be separated
by disjoint open sets. That is, if K is a compact subset of a Hausdorff space X and
if x ∈ X , x ∈
/ K , then there are disjoint open sets V, W with x ∈ V, K ⊂ W .

Proof Using separation for x ∈ K c and each y ∈ K , there is a pair Vy , W y of disjoint

open sets with x ∈ Vy , y ∈ W y . Thus {W y : y ∈ K } is an open cover of K and, by
compactness, there are finitely many points y1 , . . . , yn in K such that W y j , j =
1, . . . , n, cover K . The sets V := ∩n1 Vy j , W := ∪n1 W y j satisfy the requirements. 

In Hausdorff spaces, compact sets behave like points. We will see many examples
of this. The last result and the next two are instances.

Corollary 8.7 In a Hausdorff space, compact sets are closed.

Proof In the proof above, for x ∈ K c , V ⊂ K c is a neighbourhood of x. So each

point of K c is an interior point. 

The next result is essentially due to Vietoris, see 8.15. Notice how we are able to
pass from separation of points to separation of a point and a compact set, thence to
separation of compact sets.

Proposition 8.8 In a Hausdorff space, disjoint compact sets can be separated by dis-
joint open sets. In other words, if K 1 , K 2 are disjoint compact subsets of a Hausdorff
space X , there are disjoint open sets V1 , V2 such that K j ⊂ V j , j = 1, 2.

Proof We apply the previous proposition to K 1 and each z ∈ K 2 to get disjoint

open sets Vz , Wz with K 1 ⊂ Vz and z ∈ Wz . Since K 2 is compact, there is a finite
subset {z 1 , . . . , z m } of K 2 such that {Wz j }m
1 is a cover for K 2 . Consequently, V1 :=
∩m1 Vz j
, V2 := ∪ m
1 W y j
are disjoint open sets containing K 1 , K 2 respectively. 

Here are some interesting properties of maps into Hausdorff spaces.

110 8 Separation Axioms

Proposition 8.9 If f, g : X → Y are continuous maps from a space X into a Haus-

dorff space Y , then the following hold:
1) The set {x ∈ X : f (x) = g(x)} is closed;
2) if f = g on a dense set D, then f ≡ g;
3) G f , the graph of f , is closed in X × Y .

Proof 1) If {xi } is a net converging to x in X and if f (xi ) = g(xi ) for all indices i,
then f (x) = lim f (xi ) = lim g(xi ) = g(x). Note that we have used uniqueness of
limits in a Hausdorff space.
2) is a consequence of 1).
3) If {xi } is a net such that (xi , f (xi )) converges to (x, y) in X × Y , then {xi }
converges to x in X and { f (xi )} converges to y in Y . But continuity of f implies
f (xi ) → f (x) and so, by uniqueness of limits, y = f (x) and the proof is complete.

In general bijective continuous maps need not be homeomorphisms. Here is a

nice setting where they are. 

Proposition 8.10 If f is a continuous bijection from a compact space X onto a

Hausdorff space Y , then f is a homeomorphism.

Proof Since compact sets in a Hausdorff space are closed, repeat the proof of 5.25.

Hausdorff spaces are well behaved vis à vis subspaces and products. 

Proposition 8.11 1) If a space is Hausdorff, so is any subspace;

2) Products of Hausdorff spaces are Hausdorff.

Proof 1) Distinct points x, y of a subspace E of a Hausdorff space X have disjoint

neighbourhoodsU, V in X and so E ∩ U, E ∩ V are disjoint neighbourhoods in E.
2) Let X = i X i where each X i is a Hausdorff space. If x = (xi ) and y = (yi )
are distinct points in X , there is an index i such that xi = yi . If Ui , Vi are disjoint
neighbourhoods of these in X i , then pi−1 (Ui ), pi−1 (Vi ) are disjoint neighbourhoods
of x, y in X , where pi : X → X i is the projection. 

8.2 Normal Spaces

What is the normal meaning of the word ‘normal’? Ordinary, usual, common, stan-
dard are some of the synonyms that come to mind. However, in mathematics, nor-
mally’normal’ means ‘not normal’, not ordinary, something special or abnormal.
Normal operators on Hilbert spaces, normal subgroups and normal topological spaces
are examples. (Having said that, one can also argue that these are the ones that ‘nor-
mally’ arise in practice!)
Let us begin by observing some special properties for disjoint closed sets in
metric spaces. Recall the continuous distance function d(x, E) for any subset E of a
metric space X . d(x, E) = 0 precisely when x ∈ Ē. If A, B are disjoint and closed,
8.2 Normal Spaces 111

the function f = f A,B defined by f (x) = d(x,A)+d(x,B)

d(x,A)
, x ∈ X , is continuous on X
with 0 ≤ f ≤ 1, f (x) = 0 for x ∈ A and f (x) = 1 for x ∈ B. The disjointness of
A, B ensures that the denominator never vanishes. Further, U = f −1 ([0, 1/2)) and
V = f −1 (1/2, 1]) are disjoint open sets with A ⊂ U, B ⊂ V .
Thus we have proved the following result.

Proposition 8.12 Let A, B be disjoint closed sets in a metric space X .

1) There is a continuous function f = f A,B : X → [0, 1] with f (x) = 0 for
x ∈ A and f (x) = 1 for x ∈ B.
2) There are disjoint open sets U, V such that A ⊂ U, B ⊂ V .

The first property is separation of closed sets by continuous functions and the
second one is separation of closed sets by open sets. What we have observed is
that they are equivalent in metric spaces. In general, it is clear that the latter is a
consequence of the former. In fact, it turns out that the two properties are equivalent
in a general setting, but the reverse implication is deep, as we will see. Note also that,
by considering bd(x,A)+ad(x,B)
d(x,A)+d(x,B)
, we can replace 0 and 1 by any two real numbers a, b.
Separation of closed sets is stronger than the separation points in a T1 -space. This
explains the T1 assumption in the next definition. This separation axiom was given
by Tietze in 1923 [8].

Definition 8.13 A normal space or a T4 space is a T1 topological space X in which

any two disjoint closed sets can be separated by disjoint neighbourhoods: if A, B are
disjoint closed sets in X , there are disjoint open sets U, V such that A ⊂ U, B ⊂ V .

The initial discussion, which motivated this definition, shows that metric spaces
are normal.
Proposition 8.14 The following are equivalent for a T1 space X :
1) X is normal.
2) If A is closed and U is open in X with A ⊂ U , then there is an open set V
such that A ⊂ V ⊂ V̄ ⊂ U . (‘Every neighbourhood of a closed set contains a closed
neighbourhood.’)
3) For each pair A, B of closed sets in X , there is a pair of open neighbourhoods
V, W such that V̄ , W̄ are disjoint.

Proof 1) ⇒ 2): If A is closed and U is open with A ⊂ U , then A, U c are disjoint

closed sets. By normality, there are disjoint open sets V, W such that A ⊂ V, U c ⊂
W . Then V ⊂ W c and W c is closed and so V̄ ⊂ W c ⊂ U .
2) ⇒ 3): Suppose A, B are disjoint closed sets and apply 2) with B c in place of
U . So there is an open set V such that A ⊂ V ⊂ V̄ ⊂ B c . Then B ⊂ X \ V̄ and,
again by 2), there is an open set W such that B ⊂ W ⊂ W̄ ⊂ X \ V̄ . Thus we have
V, W with the asserted properties.
3) ⇒ 1) is clear. 
112 8 Separation Axioms

(−a, a)

R
A
R

Fig. 8.1 R × R has an uncountable discrete subset A, hence is not normal

Here is another large class of normal spaces, besides metric spaces. Vietoris proved
this (even before normal spaces were defined!), [3], Satz 27. It was also later obtained
by Alexandroff and Urysohn [9].

Proposition 8.15 (Vietoris, 1921)

Every compact Hausdorff space is normal.

Proof Since closed sets in a compact space are compact, this is a consequence of
the result (8.8) that in a Hausdorff space, disjoint compact sets can be separated. 

Example 8.16 Product of two normal spaces may not be normal. R is normal,
but R × R is not. To see the latter assertion, first note that the subspace A =
{(x, −x) : x ∈ R} is an uncountable discrete subspace of R × R (Fig. 8.1). D =
Q × Q is a countable dense subspace. Suppose R × R is normal. For each S ⊂ A,
S and A \ S are disjoint closed sets and so there are disjoint open sets U S , VS with
S ⊂ U S , A \ S ⊂ VS . Since D is dense, D ∩ U S is nonempty. Note that U S ∩ VT is
open and nonempty if S = T . Since D is dense D ∩ U S ∩ VT is nonempty, whereas
D ∩ UT ∩ VT is empty. It follows that D ∩ U S = D ∩ UT if S = T . This means that
S → U S is an injective map from P(A) to P(D). But this is impossible because A
is uncountable whereas D is countable. Thus R × R is not normal.

Remark 8.17 Subspaces of normal spaces may not be normal. [0, 1][0,1] is compact
Hausdorff and hence is normal. But it is a fact that the subspace (0, 1)[0,1] is not
normal. Thus a subspace of a normal space may not be normal. For an example
involving ordinals, see [7, 10, 11].

However, we have a positive result.

Proposition 8.18 Closed subspaces of normal spaces are normal.
8.2 Normal Spaces 113

Proof Let X be normal and C ⊂ X be closed. One-point sets are closed in X and
so are closed in C. If A, B are disjoint closed sets in C, they are also closed in X . If
U, V are disjoint open sets in X separating A, B, then U ∩ C, V ∩ C are open in C
and separate them. 
The deepest, fundamental results on normal spaces are Tietze’s result on extension
of continuous functions and Urysohn’s characterisation of normality by separation
of closed sets by continuous functions. We take up the latter result first. We have
seen an easy proof of this for metric spaces in the opening discussion of this section.
As observed earlier, separation of closed sets by continuous functions yield sepa-
ration by open sets. Hence the converse of the following theorem holds for T1 -spaces.
The theorem, obtained by Urysohn [12], is one of the fundamentally important results
of general topology.
Theorem 8.19 (Urysohn’s Lemma, 1925)
Let X be a normal space and A, B be disjoint closed subsets. Then there is a continu-
ous function f : X → [0, 1] such that f (x) = 0 for x ∈ A and f (x) = 1 for x ∈ B.

Proof The set DQ = { 2mn : m, n ∈ Z, 0 ≤ m ≤ 2n } of dyadic rationals in [0,1] is

countable and dense in [0,1]. We enumerate
DQ = {0, 1, 21 , 41 , 43 , 18 , 38 , 58 , 78 , . . . , 21n , 23n . . . 2 2−1
n
n , . . .}.

We first construct open sets Uq for each q ∈ DQ with the properties

A ⊂ Uq ⊂ Ūq ⊂ Ur ⊂ Ūr ⊂ U1 = B c
for q < r in DQ and then use these to define the sought after function.
Step 1: Construction of the Uq .
Take U1 = B c and observe that A ⊂ U1 . Since A is closed and U1 is open, by
normality there is an open set V such that A ⊂ V ⊂ V̄ ⊂ U1 . Take U0 = V . Noting
that Ū0 ⊂ U1 , there is an open set U1/2 with Ū0 ⊂ U1/2 ⊂ Ū1/2 ⊂ U1 . Then there are
open sets U1/4 and U3/4 with
Ū0 ⊂ U1/4 ⊂ Ū1/4 ⊂ U1/2 ⊂ Ū1/2 ⊂ U3/4 ⊂ Ū3/4 ⊂ U1 .
At the next step we get appropriate open sets such that
Ū0 ⊂ U1/8 ⊂ Ū1/8 ⊂ U1/4 ⊂ Ū1/4 ⊂ U3/8 ⊂ Ū3/8 ⊂ U1/2 ⊂ Ū1/2 ⊂ U5/8 ⊂ Ū5/8
⊂ U3/4 ⊂ Ū3/4 ⊂ U7/8 ⊂ Ū7/8 ⊂ U1 .
It is now clear how to proceed to define a Uq for each q ∈ DQ recursively,
using normality multiple times in each step. Step n will yield open sets Uq for
q ∈ {0, 1, 21 , 14 , 43 , . . . , 21n , 23n , . . . , 2 2−1
n
n } such that

A ⊂ U0 ⊂ Ū0 ⊂ U1/2n ⊂ Ū1/2n ⊂ U1/2n−1 ⊂ Ū1/2n−1 ⊂ U3/2n ⊂ · · · ⊂ U(2n −3)/2n

⊂ Ū(2n −3)/2n ⊂ U(2n −1)/2n ⊂ Ū(2n −1)/2n ⊂ U1 = B c .
Thus we get, as promised, open sets Uq for each q ∈ DQ such that
A ⊂ Uq ⊂ Ūq ⊂ Ur ⊂ Ūr ⊂ U1 = B c for q, r ∈ DQ , q < r.
Step 2: Construction of the function f .
Define f on X by
114 8 Separation Axioms


inf{q ∈ DQ : x ∈ Uq }, x ∈ U1 = B c
f (x) =
1, x∈B

Then f (x) = 0 for x ∈ A since A ⊂ Uq for all q ∈ DQ and f (x) = 1 for x ∈ B.

Now we come to the nontrivial part, the continuity of f . Since intervals of the form
[0, a) and (b, 1] form a subbase for [0,1], it suffices to show that the inverse images
of such intervals are open in X .
If q ∈ DQ and f (x) < q, then x ∈ Uq ; for, there is an r such that x ∈ Ur with
f (x) < r < q and so x ∈ Uq . Thus if f (x) < a, we can choose q such that f (x) <
q< a and then x ∈ Uq . If y ∈ Uq , then f (y) ≤ q < a. Hence, f −1 ([0, a)) =
{Uq : q ∈ DQ , q < a} is open.
Next, we observe that f (x) > q, q ∈ DQ , implies x ∈ / Ūq . For, if f (x) > q,
choosing r ∈ DQ with f (x) > r > q, we get x ∈ / Ur and so x ∈ / Ūq ⊂ Ur . We
show that every x ∈ W := f −1 (b, 1] is an interior point. Choosing q ∈ DQ with
b < q < f (x), note that V = Ūqc is a neighbourhood of x. Further V ⊂ W ; for,
y ∈ V and f (y) < q would imply y ∈ Uq ⊂ Ūq , contradicting the opening line of
the paragraph. Thus f (y) ≥ q > b, proving y ∈ W . 

Remark 8.20 There is nothing special about the values 0 and 1. They can be
replaced by any two distinct real numbers, say a < b. In fact, defining f a,b (x) =
(b − a) f (x) + a, we get a continuous function f a,b : X → [a, b] taking values a on
A and b on B. The reader should also check whether T1 axiom was used in the proof.

Next we turn to the question of continuous extensions. Suppose f is a real contin-

uous function on [0,1]. Does f extend continuously to R? The answer is easy: just
define the function to be the constant f (0) to the left of 0 and the constant f (1) to
the right of 1 (Fig. 8.1). Of course there are any number of other choices. Note that
the continuous function f (x) = 1/x on the open interval (0,1) has no such extension
to R (Why?). If the closed interval is replaced by an arbitrary closed set in R, can
you think of some way of getting an extension?
Suppose X is a Hausdorff space having the extension property: for any closed set
E in X , every continuous function f : E → [a, b] extends to a continuous function
g : X → [a, b]. If A, B are disjoint closed sets in X , take E = A ∪ B and define f
on E by f = a on A and f = b on B. Then f is continuous (by the pasting lemma)
and so has a continuous extension g : X → [a, b]. Thus, disjoint closed sets can be
separated by continuous functions. In other words, we have shown that if X has the
stated extension property, then X is normal. The theorem of Tietze shows that this
extension property characterises normality: any normal space has such an extension
property. This was proved by Tietze [8] for metric spaces and by Urysohn [12] for
the general case (Fig. 8.2).
This is a nontrivial result and we prove it using the Urysohn lemma.
8.2 Normal Spaces 115

a b

Fig. 8.2 Extension of continuous function from [a, b] to R

Theorem 8.21 (Tietze Extension Theorem)

Let E be a closed subset of a normal space X . Then any continuous function f on E
taking values in R, or in an interval [a, b], has a continuous extension g to X taking
values in R, respectively, [a, b].

Proof The case when f is a constant is trivial, so assume that f is not a constant. First
consider the case when f is bounded, with range in [a, b]. Taking c = sup{| f (x)| :
x ∈ E} and composing with a homeomorphism [a, b] → [−c, c], we may replace
[a, b] by [−c, c].
We will make repeated use of Urysohn’s lemma (8.19) and construct the sought
after function as the uniform limit of continuous functions.
Let c0 := sup{| f (x)| : x ∈ E} and consider the disjoint closed sets
A0 = {x ∈ E : f (x) ≤ −c0 /3}, B0 = {x ∈ E : f (x) ≥ c0 /3}.
Urysohn’s lemma yields a continuous function g0 : X → [−c0 /3, c0 /3] with g0 =
−c0 /3 on A0 and g0 = c0 /3 on B0 . Note that |g0 | ≤ c0 /3 and | f − g0 | ≤ 2c0 /3
on E. Repeat this argument with f replaced by f 1 := f − g0 on E, i.e. take c1 =
sup{| f 1 (x)| : x ∈ E}, note that c1 ≤ 2c0 /3 and apply Urysohn’s lemma to the disjoint
closed sets
A1 = {x ∈ E : f (x) ≤ −2c0 /32 }, B1 = {x ∈ E : f (x) ≥ 2c0 /32 }
to get a continuous function g1 : X → [−2c0 /32 , 2c0 /32 ] such that g1 = −2c0 /32
on A1 , g1 = 2c0 /32 on B1 . Take f 2 := f − g0 − g1 and note that | f 1 (x)| ≤ 22 c0 /32
for x ∈ E. Induction gives sequences of continuous functions { f n }∞ ∞
1 on E and {gn }0
on X satisfying

|gn | ≤ 2n c0 /3n+1 and | f n | ≤ 2n+1 c0 /3n+1 , f n := f − n0 gk on E.

Weierstrass M-test shows that ∞ 0 gn converges absolutely and uniformly to a
continuous function g on X . On E, f − g = lim f n = 0 uniformly since | f n | ≤
2n+1 c0 /3n+1 → 0. Moreover
 
|g| ≤ |gn | ≤ c0 /3 ∞ 0 (2/3) = c0 .
n

The case when f is unbounded reduces to the bounded case on composing with a
homeomorphism R → (−c, c) (e.g. x → 1+|x|
cx
), . 
116 8 Separation Axioms

Remark 8.22 The theorem easily extends to complex valued functions and functions
taking values in Rn .

8.3 Regular Spaces

Regularity was defined by Vietoris in 1921 [3] as follows. (The name ‘regular’ was
introduced later by Alexandroff.)
(E) Eine Umgebung Ux eines Punktes x enthalt immer eine Umgebnng Wx von
x, so daßjeder Punkt von CUx (Komplementgrmenge von Ux ) samt einer seiner
Umgebungen in C Wx liegt. (Vietoris [3])
(‘(E) A neighbourhood Ux of a point x always contains a neighbourhood Wx of
x, so that each point of Uxc including one of its neighbourhoods lies in Wxc .’)
The slightly different current definition was introduced by Tietze, [8]. Regular
spaces lie ‘between’ Hausdorff and normal spaces: separation of a point and a closed
set.

Definition 8.23 A regular space or a T3 space is a T1 space X in which a point and

a closed set can separated, i.e. if F is a closed set and x ∈
/ F, there are disjoint open
sets U, V with x ∈ U and F ⊂ V .

Proposition 8.24 In a T1 space X , the following are equivalent:

1) X is regular.
2) If x ∈ X , U is open and x ∈ U , there is an open set V such that x ∈ V ⊂
V̄ ⊂ U ; i.e. every neighbourhood of a point contains a closed neighbourhood.
3) If A is closed and x ∈/ A, there is an open sets V such that x ∈ V and V̄
and A are disjoint.

Proof The argument must be familiar by now.

1) ⇒ 2): If U is open and x ∈ U , then U c is a closed set not containing x. So, by
regularity, there are disjoint open sets V, W with x ∈ V and U c ⊂ W . Then V ⊂ W c
and hence x ∈ V ⊂ V̄ ⊂ W c ⊂ U .
2) ⇒ 3): If A is closed and x ∈ / A, applying 2) with Ac in place of U , we get an
open set V with x ∈ V ⊂ V̄ ⊂ Ac and so A ∩ V̄ = ∅.
3) ⇒ 1): If A is closed and x ∈ / A, let V be as in 3). Then U = X \ V̄ and V are
open sets containing A and x, respectively. 

Unlike normality, regularity behaves well in respect of subspaces and products.

Proposition 8.25 1) A subspace of a regular space is regular.

2) An arbitrary product of regular spaces is regular.

Proof 1) Let X be a regular space and let E be a subspace. Let C be a closed set
in E so that C = E ∩ A where A is closed in X . If x ∈ E \ C, then x ∈ / A and by
regularity of X , there are disjoint open sets U, V in X such that x ∈ U, A ⊂ V . Then
8.3 Regular Spaces 117

E ∩ U, E ∩ V are disjoint open sets in E, x ∈ E ∩ U and C ⊂ E ∩ V . Moreover,

E is also a T1 space
 being a subspace of a T1 space, and i) is proved.
2) Let X = i X i be a product of regular spaces, let U be open in X and let
x = (xi ) ∈ U . Let ∩n1 pi−1 j
(Ui j ) be a basic open neighbourhood of x contained in U .
Then xi j ∈ Ui j , so by regularity there are open sets Vi j with xi j ∈ Vi j ⊂ V̄i j ⊂ Ui j .

Then V = ∩n1 pi−1 j
(Vi j ) is an open set in X , x ∈ V ⊂ V̄ ⊂ U since V̄ = i V̄i where
Vi = X i for i = i j . 

Example 8.26 R is regular, hence so is R × R . But we have see that R × R is

not normal, see 8.16.

8.4 Completely Regular Spaces

Separation of disjoint closed sets by disjoint open sets and by continuous functions
turned out to be equivalent. So it is tempting to believe that an analogous result holds
for separation of a point and a closed set, i.e. in a regular space a closed set E and
an x ∈/ E can be separated by continuous functions. But this is not true. Separation
by continuous functions is stronger than separation by open sets in this case. This
necessitates the introduction of a new class of spaces.
Definition 8.27 A T1 space X is said to be completely regular or a Tychonoff space
if for a closed set E in X and a point x0 ∈ X, x0 ∈
/ E, there is a continuous function
f : X → [0, 1] with f (x0 ) = 1, f (x) = 0 for x ∈ E.
Urysohn [12] introduced these spaces in 1925. They are also called Tychonoff
spaces because of Tychonoff’s work [13]; e.g. see 8.32. A completely regular space
is regular (U = f −1 [0, 1/2) and V = f −1 (1/2, 1] are disjoint open sets containing
E, x0 , respectively,), but there are regular spaces on which the only continuous real
functions are constants (Hewitt [14]). A normal space is completely regular. So,
completely regular spaces lie between T3 spaces and T4 spaces (and are therefore
called T3 21 spaces!).

Proposition 8.28 A T1 space X is completely regular if and only if the following

property holds: if x0 ∈ U and U is open in X , there is a continuous function f :
X → [0, 1] with f (x0 ) = 1 and f = 0 on U c .

Proof If X is completely regular, U is open in X and x0 ∈ U , there is an open set V

such that x0 ∈ V ⊂ V̄ ⊂ U . Complete regularity ensures that there is a continuous
function f : X → [0, 1] such that f (x0 ) = 1 and f (x) = 0 for x ∈ V c ⊃ U c (the
last condition is also expressed by saying that f is supported in U or supp f ⊂ U ).
Conversely, suppose the stated condition holds, E ⊂ X is closed and x0 ∈/ E. Then
U = E c is open and x0 ∈ U . Hence there is a continuous function f : X → [0, 1]
with f (x0 ) = 1, f = 0 on U c = E. 
118 8 Separation Axioms

Normality is the only the separation property considered in the chapter that has
bad behaviour vis à vis subspaces and products.

Proposition 8.29 Subspaces of completely regular spaces and products of com-

pletely regular spaces are completely regular.

Proof Let X be completely regular and E ⊂ X . Since X is a T1 space, so is E. If A is

closed in E and x0 ∈ E \ A, then A = E ∩ B, B is closed in X and x0 ∈ / B. As X is
completely regular, there is a continuous function f : X → [0, 1] with f (x0 ) = 1 and
f (B) = 0. Thus f | E : E → [0, 1] separates x0 and A, proving complete regularity
of E. 
Let X = i X i where each X i is completely regular. Each X i is a T1 space and
so is their product X . Let x = (xi ) ∈ X and let U = ∩n1 pi−1 j
(Ui j ) be a basic open set
containing x. By complete regularity of the X i , there are continuous functions f i j :
X i j → [0, 1] such that f i j (xi j ) = 1 and f i j = 0 on Uicj . Letting f = min j f i j ◦ pi j
we see that f : X → [0, 1] is continuous, f (x) = 1 and f = 0 on U c . 

A completely regular space embeds into a product of (possibly uncountably many)

copies of [0,1]. A few definitions will be helpful.

Definition 8.30 Let X be a space and let F be a family of continuous maps f from
X to a space Y f . F separates points of X if for x = y in X , there is an f ∈ F
such that f (x) = f (y). F separates points and closed sets if for each closed set E
in X and x ∈ E c , there is an f ∈ F such that f (x) ∈
/ f (E). The evaluation map
e : X → f ∈F Y f is defined by e(x)( f ) = f (x).

The next lemma gives the basic properties of the evaluation map. It will also be used
in the proof of the Urysohn metrisation theorem in the next chapter.
Lemma 8.31 (Embedding lemma)
Let notation be as above. Then
1) The evaluation map e is continuous.
2) e is one-one if F separates points of X .
3) e is an open map onto e(X ) if F separates points and closed sets.

Proof 1) e is continuous as p f ◦ e = f is continuous for all f .

2) For x = y in X , if there is an f ∈ F with f (x) = f (y), then e(x)( f ) =
e(y)( f ), so e(x) = e(y) and hence e is injective.
3) Let U be open in X and x ∈ U . By hypothesis,
 there is an f with f (x) ∈
/ f (U c ).
The set W = {(y f ) : y f ∈/ f (U c )} is open in Y f and e(x) ∈ W ∩ e(X ) ⊂ e(U ),
proving that e(U ) is open in e(X ). 

Theorem 8.32 (Embedding in a cube—Tychonoff 1925)

A topological space is completely regular if and only if it can be embedded in a
product of copies of [0, 1].
8.4 Completely Regular Spaces 119

Proof One part is easy in view of Tychonoff’s product theorem. Any product of
copies of [0,1] is a compact Hausdorff space, hence is normal. In particular, it is
completely regular and so is any subspace.
Conversely, suppose X is a completely  regular space. For each continuous
f : X → [0, 1], let I f = [0, 1] and Y = f I f . (So Y = [0, 1]C(X,[0,1]) .) Take Y f =
[0, 1] for all f ∈ F = C(X, [0, 1]) in the lemma (8.31) to show that the evalua-
tion map given by e(x) = ( f (x)) is a homeomorphism of X onto e(X ). Since X is
completely regular, F separates points and closed sets. This also gives separation of
points as X is T1 . The proof is complete on invoking the lemma (8.31). 

(0,1) is not compact but is a dense subspace of the compact space [0,1]. For what
spaces X is such a situation possible?

Definition 8.33 A compactification of a space X is a pair (X  , h) where X  is a

compact space, h is a homeomorphism of X onto a dense subspace h(X ) of X  .
More succinctly, a compactification of X is a compact space in which X embeds as
a dense subspace.

We often say that X  is a compactification of X , with no explicit mention of h,

e.g. [0,1] is a compactification of (0,1), [0,1) and (0,1]. Compactifications were first
studied by Carathéodory, [15], for open subsets of the plane in the context of analytic
functions.

Corollary 8.34 (Stone-Čech compactification)

With notation as in the proof of the theorem, the closure β X of the image e(X ) in Y
is a compactification of X .

β X is called the Stone-Čech compactification of X . Its construction appeared in

1937 in independent papers by Eduard Čech, a Cech topologist [16] and American
mathematician Marshall Stone [17]. Actually, Tychonoff in [13] shows that any
completely regular space X is densely embedded in a compact Hausdorff space β X
such that any bounded real continuous function on X has a continuous extension to
β X . This is pointed out by Čech himself in [16]. Considering this and the historical
remarks made about Tychonoff’s theorem earlier, one might say that, if properly
understood, there is some truth in the statement that ‘Tychonoff’s theorem was proved
by Čech and the Čech compactification was obtained by Tychonoff’. Čech studies
properties of β X in [16].
Here is the extension property of β X .

Theorem 8.35 If Y is a compact Hausdorff space, then any continuous map ϕ :

X → Y has an extension to β X ; more precisely, there is a continuous map  :
β X → Y such that ϕ =  ◦ e.

Proof Y is compact, Hausdorff and hence is normal. Inparticular, it is completely

regular and so has an embedding in a cube eY : Y → Ig , where Ig = [0, 1] for
each g ∈C(Y, [0, 1]). We have the embedding e : X → I ffor f ∈  C(X, [0, 1]).
For u ∈ I f , let (u)(g) = u(g ◦ ϕ). This gives a map : I f → Ig . If qg is
120 8 Separation Axioms

the projection on Ig , we have qg ◦ (u) = pg◦ϕ (u), p f being the projection on I f .

Thus qg ◦ is continuous for all g and so is continuous. Moreover, (e(x))(g) =
g(ϕ(x)) = eY (ϕ(x))(g) so that (e(x)) = eY (ϕ(x)) for all x ∈ X and ◦ e = eY ◦
ϕ. Thus (e(X )) = eY (ϕ(X )) ⊂ eY (Y ). Since e(X ) is dense in β X , so is (e(X ))
in (β X ). On the other hand, eY (Y ) is compact and so is closed, proving eY (Y ) ⊃
(β X ). Define  = e−1Y ◦ ( |β X ). Then  : β X → Y is continuous and, for x ∈ X ,
(e(x)) = e−1Y ( (e(x))) = ϕ(x). The proof is complete. 

The extension property embodied in the theorem determines β X uniquely and β X

is the ‘largest’ compactification of X . These and more results on the Stone-Čech
compactification are given in the exercises.

Example 8.36 [0,1] is a compactification of (0,1), but it is not the Stone-Čech com-
pactification as the continuous function ϕ(x) = sin x1 on (0,1) taking values in [−1,1]
has no continuous extension to [0,1].

In general, β X is huge. For example, βN has the cardinality of the set of all subsets
of R (See [10] or [18]). For more on β X , see [19].

Exercise 8

1. T0 -spaces. Recall the Sierpiǹski space S = {0, 1} with the only proper open
set {1}. Thus {1} is an open set containing 1 but not 0. However there is no open
set containing 0 but not 1, as the only open set containing 0 is the whole space.
Such spaces are called T0 spaces. Thus, X is a T0 space if, for every pair x, y of
distinct points, either there is a neighbourhood of x not containing y or there is a
neighbourhood of y not containing x (but not necessarily both). (If both hold, the
space is a T1 space mentioned earlier.) These spaces are hardly interesting and we
will have no further discussion about them.
Find another T0 space.
2. T1 -spaces. a) The following are equivalent: i) X is a T1 space; ii) one-point
sets are closed in X ; iii) E = ∩{V : E ⊂ V, V is open} for any E ⊂ X ; iv) id X :
(X, T ) → (X, Tco f ) is continuous.
b) In a T1 space, if x is a limit point of E, every neighbourhood of x has infinitely
many points of E. So, finite sets have no limit points.
c) A T1 -space X is countably compact if and only if it is Bolzano–Weierstrass
compact.
d) On any set X , there is a smallest T1 topology.
e) The derived set E  of any set in a T1 space is closed.
3. Hausdorff spaces. a) X is Hausdorff if {(x, x  ) : f (x) = f (x  )} is closed
whenever f : X → Y is an open surjective map.
b) If T is a Hausdorff topology on X , any finer topology is also Hausdorff. What
are the corresponding statements for other properties like compactness and other
separation properties?
c) Every infinite Hausdorff space contains a countably infinite discrete subspace.
8.4 Completely Regular Spaces 121

d) If X is Hausdorff, then so is Cone(X ).

e) A totally ordered set with order topology is Hausdorff.
f) If x1 , . . . , xn are distinct points in a Hausdorff space X , there are mutually
disjoint open sets U1 , . . . , Un such that x j ∈ U j for all j.
g) If (X, T ) is compact and Hausdorff, then no strictly bigger topology T   T
on X is compact.
4. Regularity. a) On R, take neighbourhoods of any x = 0 as usual, but basic
neighbourhoods of 0 as sets of the form (−ε, ε) \ {1/n : n ∈ N}. This defines a
topology on R which is Hausdorff, but is not regular.
b) In a regular space, any closed set is the intersection of all the open sets containing
it.
c) Z with Tap is not regular.
d) In a regular space, every pair of distinct points have neighbourhoods with
disjoint closures.
e) Find a non-regular space in which every closed set F and points x ∈ / F have
disjoint neighbourhoods.
5. Completely regular spaces. a) A T1 space is completely regular if and only
if the complements of zero sets of real continuous functions form a base for the
topology of X .
b) If f is a lower semi-continuous function (See Chap. 4, Ex. 9) on a completely
regular space X , then f is the supremum of all continuous functions g ≤ f on
X .(Bourbaki).
c) Every completely regular space is regular and every subspace of a normal space
is completely regular. (For examples of a regular space that is not completely regular
and a subspace of a normal space that is not normal, see [7, 10, 11, 20, 21].)
d) Let X be completely regular and let Cb (X ) be the space of real, bounded continu-
ous functions on X . Then sets of the form U ( f, x, ε) =
{y ∈ X : | f (x) − f (y)| < ε}, ε > 0, x ∈ X, f ∈ Cb (X ), form a subbase for the
topology of X .
6. Stone-Čech compactification. See [7, 11, 22]. a) The extension property of
Theorem (8.35) determines β X uniquely, i.e. any other compactification of X having
the extension property is homeomorphic to β X by a homeomorphism leaving points
of X fixed.
b) β X is a maximal compactification: more precisely, any compactification of X
is a quotient space of β X .
7. Normal spaces. a) R is normal, but R × R is not. For example, the disjoint
closed sets A = {(x, −x) : x ∈ Q} and B = {(x, −x) : x ∈ / Q} cannot be separated
by open sets. [10], p.137
b) Deduce Urysohn’s lemma from the Tietze extension theorem.
c) If X is normal and A, B ⊂ X are disjoint closed sets. If A is a G δ set, then
there is a continuous function f : X → [0, 1] with f −1 (0) = A and f = 1 on B. If
both A and B are G δ sets, there is such an f with f −1 (0) = A and f −1 (1) = B.
122 8 Separation Axioms

d) There are normal spaces in which every closed set is a G δ ; such spaces are
called perfectly normal spaces and in such a space every subspace is normal. Metric
spaces are perfectly normal.
e) Every Fσ set in a normal space is normal.
f) If X is normal and E 1 , . . . , E n are closed sets with ∩n1 F j = ∅. Then there are
open sets V j ⊃ E j with ∩n1 V̄ j = ∅.
8. Paracompact spaces. A cover U of a space X is called a refinement of a
cover V if every U ∈ U is a subset of some V ∈ V .
a) X is compact if and only if every open cover of X has a finite refinement that
covers X .
A space X is called a paracompact space if every open cover of X has a locally
finite refinement that is a cover.
These spaces were introduced by Dieudonné in 1944 [23]. They are very useful
in topology and are important in geometry because of the existence of partitions of
unity. A.H.Stone is one of the most important contributors to the study of paracom-
pactness in topology.
b) Every paracompact Hausdorff space is normal. (Theorem of A.H.Stone.) ( See,
e.g., Dugundji [10], for the results given here.)
c) Every metric space is paracompact.
9. Quotients. a) Let R be an equivalence relation on a space X and let p : X →
X/R be the canonical map.
i) If p is open and R is closed in X × X , then X/R is Hausdorff.
ii) X/R is T1 if and only if each equivalence class is closed in X .
iii) Suppose p is open and closed. If X is normal, then so is X/R.
b) Let X, Y be compact Hausdorff spaces and let f : X → Y be a continuous
surjection. Let R be the equivalence relation whose classes are the fibres of f , i.e.
x Ry if f (x) = f (y). Then the quotient space X/R is homeomorphic to Y . In fact,
h which takes the fibre f −1 (x) to f (x) is well defined and is a homeomorphism. We
have seen examples of this in Chap. 6: S1 as a quotient of [0,1] and the torus as a
quotient of the square.
10. Cantor set. Let C be the Cantor ternary set in [0,1]. Every x ∈ C has a ternary
(or triadic) expansion x =  a3nn , where (an ) ∈ {0, 2}N .
a) ϕ : {0, 2}N → C, (an ) →  a3nn , is a bijection.
b) If C has the subspace topology from R and {0, 2}N is given the product topology,
ϕ is continuous.
c) ϕ is a homeomorphism.
d) Construct a continuous function on [0,1] whose zero set is C.
11. Space filling curves: existence. a) ψ : {0, 2}N → [0, 1] defined by
ψ((an )) =  2an+1
n
is a continuous surjection.
b) There is a continuous map of the Cantor set C onto [0,1].
c) C is homeomorphic to C × C.
d) There is a continuous surjection of C onto [0, 1] × [0, 1].
8.4 Completely Regular Spaces 123

e) Any continuous C → [0, 1] extends continuously to [0, 1] → [0, 1].

f) There is a continuous map of [0,1] onto [0, 1] × [0, 1]. This gives a continuous
curve passing through every point of the square [0, 1] × [0, 1]! Such curves are called
space filling curves or Peano curves, after the Italian mathematician who shocked
the mathematical world by first constructing such curves in 1890, [24].
g) A map as in f) cannot be injective.
h) Adopt the arguments of f) to show that there is a continuous surjection [0, 1] →
[0, 1]n for any n.
12. Separation in topological groups. If a topological group is T1 , then it is T2
and even T3 .
13. Partitions of unity.
a) Let X be normal. If X = ∪n1 Vk , Vk open, then
i) there are closed sets E k ⊂ Vk such that X = ∪E k (induction);
ii) there are continuous f k : X → [0, 1] with  f k = 1 on X and supp f k ⊂ Vk
(Use Urysohn).
b) If the Vk in a) cover only a closed set E, then we get f k as above but with
 f k = 1 on E.
c) Let X be a paracompact Hausdorff space. For every open cover {Vi } of X , there
is a locally finite refinement {Ui } that is an open cover of X and a partition of unity
subordinate to {Ui }, i.e. there is a family f i of non-negative real valued continuous
functions f i on X such that i) supp f i ⊂ Ui and ii)  f i (x) = 1 for all x. (Note that
the sum is a finite sum for each x.) See [10].
d) For any locally finite open cover of a nor mal space, there is a partition of unity
subordinate to it.

8.5 Biographical Notes

8.5.1 Tietze

Heinrich Tietze (1880–1964) was an Austrian mathematician, best known for his
work in topology. He was the one who defined normal spaces [4], 1923 and is famous
for the Tietze Extension Theorem. He is considered as one of the first ‘topology
specialists’ and carried forward Poincaré’s work on topology. In a long 1908 paper,
he settled many outstanding questions. For example, he produced a finite presentation
for the fundamental group and invented the now well-known Tietze transformations to
show that fundamental groups are topological invariants. He gave the first reference to
the isomorphism problem for groups: if two groups are defined by finite presentations,
is there an algorithm to decide whether they are isomorphic or not? He also completed
the relationship between the fundamental group and what is now called the first
homology group. For more details, see the account of I.M.James [25]. He also worked
124 8 Separation Axioms

on other areas like geometry, knot theory, map colouring, continued fractions, prime
numbers, ruler-compass constructions, partitions.

8.5.2 Urysohn

Pavel Urysohn (1898–1924) was an Ukranian mathematician and was a student of

Luzin in Moscow. His first paper (1915) was on Physics and his Habilitationsschrift
was on integral equations. Egoroff turned Urysohn’s attention to topology, asking
him two difficult questions in the subject. In response, Urysohn came up with his
dimension theory of metric spaces and lectured on it in Moscow; it was published by
the French Academy of Sciences. Later he met Brouwer and learned about his theory
of topological dimension. Alexandroff showed that Urysohn’s definition of dimen-
sion was equivalent the one Menger had given for topological spaces. Urysohn’s
Lemma on normal spaces is one of the most fundamental results in general topology.
His metrisation theorem is now part of standard courses in topology. Completely reg-
ular spaces were introduced by Urysohn. One of his last results was the construction
of a universal metric space, i.e. a metric space containing an isometric image of any
metric space. The open cover definition for compactness was adopted to topological
spaces by Alexandroff and Urysohn. Alexandroff was a friend and collaborator for
him. The two went on their second European trip in 1924, met Hilbert, Hausdorff and
Brouwer. During one of their swims in the sea in Brittany, France, Urysohn drowned.
He was not even 26 at that time!

8.5.3 Carathéodory

Constantin Carathéodory (1873–1950) was a Greek mathematician, born in Ger-

many, had his education in Belgium and Germany and spent most of his academic
career in Germany (with short stints in Greece and visits to USA). He studied engi-
neering in Brussels and served as a military engineer. He got his doctoral degree
in mathematics in 1902 from Göttingen, served in various universities in Germany
and retired from Munich. His doctoral students included Hans Rademacher and Paul
Finsler. Calculus of variations, real and complex analysis, measure theory, convex
geometry, optics, thermodynamics are some of the areas of his significant contri-
butions. In measure theory, his extension theorem is fundamental. He published
several celebrated books on real functions, calculus of variations, function theory
and geometric optics. Einstein consulted him on the Hamilton–Jacobi equation and
the canonical transformations. He was considered as one of the great mathematicians
of his era.
8.5 Biographical Notes 125

8.5.4 M. H. Stone

Marshall Stone (1903–1989), was an American mathematician, best known for his
work on Functional Analysis. His thesis adviser at Harvard was George D.Birkhoff,
a leading mathematician at that time. He is also well-known for rebuilding the Math-
ematics Department at Chicago into one of eminence again after World War II. Sev-
eral analysts of repute in the twentieth century were among his students. Stone-Čech
compactification that we have discussed, Stone–Weierstrass Theorem generalising
the classical Weierstrass Approximation Theorem, Stone–von Neumann theorem on
the canonical commutation relations in Quantum Mechanics/ representations of the
Heisenberg group, Stone’s theorem on one-parameter groups of unitary operators,
Banach–Stone theorem and Stone’s representation theorem for Boolean algebras are
all famous (and all of them were discovered in the 1930s!). His monograph Linear
transformations on Hilbert spaces (1932, incidentally the same year as Banach’s
classic Théorie des opérations linéaires appeared) was one of the earliest on the
topic and is considered a classic. Of interest in India is the fact that he visited India
frequently and died in Madras (now called Chennai).

References

1. Hausdorff, F. Grundzüge der Mengenlehre. Leipzig (1914)

2. Alexandroff, P., Hopf.: Springer, H. Topologie I (1935)
3. Vietoris, L.: Stetige Mengen. Monatsh Math. und Physik 31, 173–204 (1921). https://doi.org/
10.1007/BF01702717
4. Tietze, H.: Beiträge zur allgemeinen Topologie I. Math. Ann. 88, 290–312 (1923). https://doi.
org/10.1007/BF01579182
5. Riesz, F.: Stetigkeitsbegriff und Abstrakte Mengenlehre Atti Contr. Internat. Mat. Roma 2,
18–24 (1908)
6. Fréchet, M.: Espaces abstraits. Gauthier-Villars, Paris (1928)
7. Kelley, J.L.: General Topology. Affiliated East-West Press (1969)
8. Tietze, H.: Über Funktionen, die auf einer abgeschlossenen Menge stetig sind. J. Rein. Angew.
Math. 145, 9–14 (1915). https://doi.org/10.1515/crll.1915.145.9
9. Alexandroff, P., Urysohn, P.: Zur theorie der topologischen Räume. Math. Ann. 92, 258–266
(1924). https://doi.org/10.1007/BF01448008
10. Dugundji, J.: Topology. Prentice Hall of India (1975)
11. Willard, S.: General Topology. Addison-Wesely (1970)
12. Urysohn, P.: Über die Mächtigkeit der zusammenhängenden Mengen. Math. Ann. 94, 262–295
(1925). https://doi.org/10.1007/BF01208659
13. Tychonoff, A.: Über die Topologische Erweiterung von Räumen. Math. Ann. 102, 544–561
(1930). https://doi.org/10.1007/BF01782364
14. Hewitt, E.: On two problems of Urysohn. Ann. Math. 2(47), 503–509 (1946). https://doi.org/
10.2307/1969089
15. Carathéodory, C.: Über die Begrenzung einfach zusammenhangender Gebiete. Math. Ann. 73,
323–370 (1913). https://doi.org/10.1007/BF01456699
16. Čech, E.: On bicompact spaces. Ann. Math 2(38), 823–844 (1937). https://doi.org/10.2307/
1968839
17. Stone, M.H.: Applications of the theory of boolean rings to general topology. Trans. Am. Math.
Soc. 41, 375–481 (1937). https://doi.org/10.1090/S0002-9947-1937-1501905-7
126 8 Separation Axioms

18. Wilder, R.L.: Evolution of the topological concept of “connected". Am. Math. Monthly 85,
720–726 (1978). https://doi.org/10.1080/00029890.1978.11994684, Correction and Adden-
dum, Am. Math. Monthly 87 31–32 (1980). https://doi.org/10.1080/00029890.1980.11994947
19. Gillman, L., Jerison, M.: Rings of Continuous Functions. Van Nostrand (1960)
20. Munkres, J.R.: Topology, 2nd edn. Pearson Education, Asia (2001)
21. Steen, L.A., Seebach Jr., J.A.: Counter Examples in Topology. Dover (1995)
22. Engelking, R.: General Topology Revised Edition. Heldermann Verlag (1989)
23. Dieudonné, J.A.: Une Généralisations des Espaces Compacts. J. Math. Pures. Appl. 23, 65–76
(1944)
24. Peano, G.: Sur une Courbe qui ramplit Toute une Plane. Math. Ann. 36, 157–160 (1890). https://
doi.org/10.1007/BF01199438
25. James, I.M.: From combinatorial topology to algebraic topology. In: James, I.M. (ed.) History
of Topology, pp. 561–574. Elsevier (1999)
Chapter 9
Connected Spaces

A satisfactory formulation for the concept of connectedness was obtained only after
many inadequate starts. The evolution took a long time—through Bolzano (1817
[1]), Cantor (1883 [2]), C.Jordan (1893 [3]), Schönfliesz (1904 [4]) and others—
and culminated in the modern formulation in 1906, due to F.Riesz and N.J.Lennes,
independently.
‘Das mathematische Kontinuum heisse zusammenhangende wenn es nicht in zwei
offene Teilmengen zerlegt werden kann, die Komplementarmengen fur einander sein.’
(F. Riesz, 1906, [5])
(‘The mathematical continuum is called connected if it cannot be decomposed into
two complementary open subsets.’)
‘A set of points is a “connected set" if at least one of any two complementary
subsets contains a limit-point of points in the other set.’ (Lennes, 1906, [6])
Riesz’s formulation says that the space cannot be decomposed into two com-
plementary open sets and is equivalent to that of Lennes. It was rediscovered by
Hausdorff in 1914. For the history of evolution of the concept of connectedness, see
[7].
We begin with an important, strong, geometric form of connectedness before
moving on to a more abstract form and then on to local versions of these properties.

9.1 Path Connected Spaces

An interval, the real line, the graph of a continuous real function on the real line,
the punctured plane, an annulus, a circle, the surface of a sphere—all these share an
important geometric property: any two points in any of these spaces can be connected
by a path which lies entirely in the space. We now take up this geometric form of
connectedness. This concept is of considerable importance in geometric aspects of

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 127
K. Parthasarathy, Topology, La Matematica per il 3+2 134,
https://doi.org/10.1007/978-981-16-9484-4_9
128 9 Connected Spaces

topology and is more intuition friendly than the more abstract property to be taken
up in the next section.
Definition 9.1 A path in a topological space X is a continuous map γ : [0, 1] → X .
For such a path, x0 = γ (0) is called the initial point and x1 = γ (1) is called the
terminal point of γ . In this case we also say that γ is a path from x0 to x1 or a path
joining x0 and x1 . The space X is said to be path connected if, for every pair x, y of
points in X , there is a path in X from x to y. A path connected set in X is a subset
C that is path connected in the relative topology.
The concept of paths and path connected sets have been around for a long time,
going back at least to the time of Weierstrass. Here are some simple properties of
path connected spaces.

Proposition 9.2 A continuous image of a path connected space is path connected.

Proof Let X be path connected and let f : X → Y be continuous and surjective. Let
y0 = f (x0 ), y1 = f (x1 ) ∈ Y . If γ is a path in X from x0 to x1 , then then γ f := f ◦ γ
is a path in Y from y0 to y1 . 

Dealing with products is easy for path connected spaces, in contrast to products
of connected spaces considered in the next section.

Proposition
 9.3 If {X i } is a collection of path connected spaces, the product space
X = X i is also path connected.

Proof Let x = (xi ), y = (yi ) ∈ X and let, for each i, γi be a path in X i from xi to
yi in X i . Then γ (t) = (γi (t)), 0 ≤ t ≤ 1. gives a path in X from x to y. (Each X i is
the image of X under the projection pi and so the converse is obvious.) 

Here are some easy criteria for path connectedness.

Proposition 9.4 Each of these implies X is path connected:

1) for every pair of points x, y ∈ X , there is a path connected subset Px,y con-
taining both x and y;
2) every x ∈ X can be joined to a fixed x0 ∈ X by a path in X ;
3) X = A ∪ B where A, B are non-disjoint path connected subsets;
4) X is the union of a chain {Ci } of path connected subsets;
5) X is the union of an increasing sequence of path connected sets.

Proof 1) is trivial since a path in Px,y is also a path in X .

2) Let x, y ∈ X and suppose α, β are paths from x0 to x, y respectively. Then
← ←
α defined by α (t) = α(1 − t), t ∈ [0, 1], is the ‘reverse’ path from x to x0 . A path
←
from x to y is the product α ∗ β given by

←
← α (2t), 0 ≤ t ≤ 21
α ∗β=
β(2t − 1), 21 ≤ t ≤ 1
9.1 Path Connected Spaces 129

3) Let x, y ∈ X . If both are in A, there is a path in A, which is also a path in X ,

from x to y. Same considerations hold when both are in B. If x ∈ A, y ∈ B, take
x0 ∈ A ∩ B. The product of a path from x to x0 in A and a path from x0 to y in B is
a path in X from x to y.
4) If x, y ∈ X then, as {Ci } is totally ordered, there is an i such that x, y ∈ Ci .
Now since Ci is path connected, the proof is complete.
5) is a special case of 4). 
Every space decomposes into path connected pieces.
Proposition 9.5 Let X be a space. For x ∈ X , let Px be the union of all path con-
nected subsets of X containing x. Then
1) Px is path connected;
2) Px is the largest path connected subset containing x.
3) For x, y ∈ X , either Px = Py or Px , Py are disjoint.
Proof 1) is a consequence of 2) of the previous proposition and 2) is clear from
the definition. If there is a z ∈ Px ∩ Py , then z ∈ Px , x ∈ Pz and Px = Pz . Similarly
Py = Pz . 
Definition 9.6 Px is called the path component (short for path connected compo-
nent) of x in X .
Observe that the path components are the equivalence classes in X under the
equivalence relation defined by x ∼ y if there is a path in X from x to y. The closure
of a path connected set may not be path connected and so path components may not
be closed, as we shall see.
Example 9.7 1. Any convex set in Rn is path connected. (Recall that a set C is
called convex if it contains the entire line segment joining any two of its points.)
More generally, if S ⊂ Rn has a point x such that the line segment joining x to any
y ∈ S lies wholly in S, then S is path connected. (Such sets are said to be star-
shaped.) (See Figs. 9.1, 9.2, 9.3) Note: Convex ⇒ Star-shaped ⇒ Path connected.
In particular, balls in Rn are path connected.
2. For n > 1 the punctured space Rn \ {0} is path connected. In fact, any two
points can be joined by at most two line segments.
3. The sphere Sn , being the image of Rn+1 \ {0} under the continuous map x →
x/x, is path connected. A more direct, geometric way to see this is to observe that
there is a great circle joining any two points a, b on the sphere. The reader should
explicitly write down a parametric equation of this circle γa,b , at least for S2 .
4. The torus S1 × S1 and the cylinder S1 × R are path connected.
5. The subset {(x, y) ∈ R2 : x ∈ Q or y ∈ Q} is path connected.
6. S O(2), the space of 2 × 2 real orthogonal matrices of unit determinant is path
connected. In  fact, it is easy to see that
cos t sin t
S O(2) = { : t ∈ R} = {γ (t) : t ∈ R} so S O(2) is the image of
− sin t cos t
R under the continuous map γ : R → M(2, R). (Note also that SO(2) is homeomor-
phic to the unit circle S1 and group-isomorphic to S1 , the group of complex numbers
of modulus 1.
130 9 Connected Spaces

Fig. 9.1 Path connected

Fig. 9.2 Star shaped

Fig. 9.3 Convex

7. The unitary group U(n) of n × n unitary matrices is path connected. We use a

bit of linear algebra. First, eigenvalues of a unitary matrix are of the form eıα , α ∈ R.
Second, any unitary matrix can be diagonalised: if u ∈ U(n), there is a v ∈ U(n)
such that u = vD(α1 , . . . , αn )v∗ where D(α1 , . . . , αn ) is the diagonal matrix diag
(eıα1 , . . . , eıαn ). Define γ (t) = vD(tα1 , . . . , tαn )v∗ for 0 ≤ t ≤ 1. Then γ is a path
in U(n) from the identity matrix in to u. Thus, there is a path joining the identity
matrix to any element of U(n). Thus U(n) is path connected.
9.2 Connected Spaces 131

9.2 Connected Spaces

There is a concept of ‘connectedness’ more general, more abstract and less tangible
than the geometric property of path connectedness discussed in the previous section.
It is an important concept in geometry and analysis. Roughly speaking, these are
spaces that are in ‘one piece’. But to give a mathematical shape to this vague notion
of connectedness, we start with a proposition.
Proposition 9.8 For a space X , the following are equivalent:
1) If A, B are nonempty subsets and if X = A ∪ B, then either A ∩ B̄ or Ā ∩ B
is nonempty.
2) If X = A ∪ B, A, B disjoint open sets, then either A or B is empty.
3) If X = A ∪ B, A, B disjoint closed sets, then either A or B is empty.
4) If A ⊂ X is both open and closed then either A is empty or A = X .
5) If f : X → ({0, 1}, Td ) is continuous, it is a constant.

Proof 1) ⇒ 2): If X = A ∪ B with A, B disjoint nonempty open sets, then Ā =

A, B̄ = B. So both A ∩ B̄ and Ā ∩ B are empty.
2) ⇔ 3) as in each of them both A and B are both open and closed.
3) ⇒ 4) because if A is both open and closed, then B = Ac is also open and closed
and X = A ∪ B.
4) ⇒ 5): If f : X → {0, 1} is continuous, then f −1 {0} is both open and closed.
If it is X , then f ≡ 0 and if it is empty then f ≡ 1.
5) ⇒ 1): If 1) fails, there are nonempty A, B with X = A ∪ B, but A ∩ B̄ = ∅ =
Ā ∩ B. So A, B are disjoint, open and if we define f | A = 0 and f | B = 1 then f is
nonconstant, continuous (Pasting lemma). 

We are now ready to give what is essentially the Riesz–Lennes–Hausdorff definition

mentioned in the introduction.
Definition 9.9 A topological space X is said to be connected if it it is not the union
of two disjoint nonempty open sets (or satisfies any of the equivalent conditions
listed above). A subset C of a space X is said to be connected if it is connected in
the relative topology. Thus X is not connected if and only if it has a separation or a
disconnection, i.e. it has a decomposition X = A ∪ B with A and B nonempty and
both open (or, equivalently, both closed). If A ∩ B̄ and Ā ∩ B are both empty, A and
B are said to be separated.

Example 9.10 1. Intervals in R are connected, as we shall soon see. This may look
obvious, but the proof is actually not all that trivial.
2. A discrete space with more than one point is not connected, whereas any
indiscrete space is connected.
3. An infinite set with cofinite topology is connected (since no two nonempty
open sets are disjoint!).
4. [0,1] ∪ [2,3] is not connected.
132 9 Connected Spaces

2
5. The space O(n) of real n × n orthogonal matrices (as a subspace of Rn ) is
not connected since the determinant function is a continuous function mapping O(n)
onto (check!) the discrete space {1, −1}.

More work is needed before important, nontrivial examples can be given. The fol-
lowing is almost trivial, but is very important.

Proposition 9.11 Any continuous image of a connected space is connected: If X is

a connected space and there is a continuous map of X onto a space Y , then Y is
connected.

Proof A disconnection Y = A ∪ B for Y gives a disconnection X = f −1 (A) ∪

f −1 (B) for X . 

Exercise. Give a proof of 9.11 using the characterisation (5) of 9.8. Note also that,
conversely, 9.11 implies (5) of 9.8.
As the most basic examples, let us begin by looking at connected subsets of the
real line. In R, intervals are the only connected sets.

Theorem 9.12 A subset E of R is connected if and only if it is an interval (bounded

or unbounded).

Proof Recall: E ⊂ R is an interval if a, b ∈ E and a < c < b imply c ∈ E, i.e. it

contains every point between any two of its points.
Suppose E is not an interval. This means there are points a, b, c with a < c <
b, a, b ∈ E, c ∈
/ E. This yields a disconnection of E: E = (E ∩ (−∞, c)) ∪ (E ∩
(c, ∞)).
Conversely, suppose E has a disconnection E = A ∪ B. Take a ∈ A, b ∈ B. Then
a = b since A and B are disjoint. We may assume, changing notation if necessary,
that a < b. Let c = sup(A ∩ [a, b]). Then c ∈ Ā. Let us now see what happens in the
two cases c ∈ E, c ∈ / E. If c ∈
/ E, then a < c < b and E is not an interval. If c ∈ E,
then c ∈ A (reason: c ∈ Ā and A is closed in E). But then, since A is open in E,
there is an ε > 0 with E ∩ (c − ε, c + ε) ⊂ A. Take d ∈ (c, c + ε). Then d ∈ / A by
definition c and so d ∈/ E. Thus we have a < d < b, d ∈ / E and we again conclude
that E is not an interval. 

Thus the list of connected subsets of R is: {a}; [a, b], [a, b), (a, b], (a, b); (−∞, a],
(−∞, a), [a, ∞), (a, ∞); (−∞, ∞) = R, a, b ∈ R, a < b.
Here is the intermediate value theorem with which the book started.
Corollary 9.13 (Intermediate value theorem)
If f is a continuous, real valued function on an interval in R, then f takes every
value between any two of its values.

Proof This is a consequence of two facts: 1. the continuous image of a connected

space is connected; 2. connected sets of the real line are precisely intervals. 

Connectedness is preserved by taking closure.

9.2 Connected Spaces 133

Proposition 9.14 If a subset C of a space X is connected, then so are the closure

C̄ and any E with C ⊂ E ⊂ C̄.
Proof If C̄ = A ∪ B is a disconnection for C̄, then there is a disconnection C =
(C ∩ A) ∪ (C ∩ B) for C. For, it is clear that C ∩ A and C ∩ B are disjoint and open
in C. If, for instance, C ∩ A is empty, then C ⊂ B and so C̄ ⊂ B (because B is
closed in C̄), hence A is empty. Thus if A and B are nonempty, so are (C ∩ A) and
(C ∩ B).
The closure of C in E is E itself and so the first part applied to the subspace E
(with the relative topology) gives the second part. 
The next result has rather useful consequences.
Proposition 9.15 Let {Ci } be a family of connected sets in a space X . If Ci , C j are
not separated for i = j, then ∪i Ci is connected.
Proof Suppose ∪i Ci = A ∪ B is a disconnection. For each i, Ci = (A ∩ Ci )∪
(B ∩ Ci ) is a disjoint union of sets open in Ci and since Ci is connected, either
Ci ⊂ A or Ci ⊂ B for each i. However, it cannot happen that Ci ⊂ A and C j ⊂ B
for some i = j. This is because A, B are separated whereas Ci , C j are not. Thus,
either Ci ⊂ A for all i or Ci ⊂ B for all i. This means that either A or B is empty.
Corollary 9.16 If each Ci is connected and if ∩i Ci is nonempty, then ∪i Ci is con-
nected.
Proof If ∩i Ci is nonempty, then no two Ci are separated. 
Corollary 9.17 If Cn is a sequence of connected sets, then ∪n Cn is connected in
each of the following cases: 1) Cn ⊂ Cn+1 ; 2) Cn ∩ Cn+1 is nonempty for each n.
Proof Case 1) is an immediate consequence of the previous corollary. To prove
case 2), observe that E 1 := C1 is connected by hypothesis and E 2 := C1 ∪ C2 is
connected by the previous corollary. By induction E n := ∪n1 Ck is connected for each
n. An application of case a) to the sequence E n now yields the connectedness of
∪n E n = ∪n Cn . 

As an application, we can deal with products of connected spaces. If X = i X i is
a product space, observe that each X i is the image of X under the projection map pi .
Thus if the product X is connected, so is each X i by Proposition 9.11. The converse
is more interesting and is nontrivial. We first consider finite products.
Theorem 9.18 If X and Y are connected spaces, then so is X × Y .
Proof Fix a point (x0 , y0 ) ∈ X × Y . X × {y0 } is the homeomorphic image of X
under the map x → (x, y0 ), so is connected. Similarly, the space {x} × Y is connected
for every x ∈ X . The point (x, y0 ) is common to these two sets and so their union
E(x) is connected, by the earlier corollary. The point (x0 , y0 ) is in E(x) for all x ∈ X
and hence the union of all the sets E(x) is connected. But this union is none other
than X × Y itself! (Fig. 9.4). 
134 9 Connected Spaces

{x} × Y {x } × Y

E(x) E(x )

X × {y0 }
(x, y0 ) (x0 , y0 ) (x , y0 )

Fig. 9.4 Product of two connected spaces is connected

 9.19 If {X 1 , . . . , X n } are finitely many connected spaces, then the product

Corollary
space n1 X k is connected.

Proof Use the theorem and induction on n. 

We can handle the infinite product case utilising the finite product case and the
following lemma,

Lemma 9.20  Let {X i : i ∈ I } be a family of topological

 spaces and fix a point
a = (ai ) ∈ X i . Then the set D = Da of all (xi ) ∈ X i for which {i ∈ I : xi = ai }
is finite, is dense in the product space X i .

Proof It suffices to show that D intersects any basic open set B = ∩i∈F pi−1 (Vi ),
where Vi is open in X i and F is a finite subset of I . But this is all but obvious: taking
xi = ai , i ∈
/ F and xi ∈ Vi arbitrary for i ∈ F, we see that x = (xi ) ∈ B ∩ D. 

Theorem 9.21 If {X i : i ∈ I } is a family of connected spaces, the product space


X i is connected.

Proof Since the closure of a connected set is connected, it is enough to prove

that the set D of the lemma is connected. For each finite setF ⊂ I , let C F :=
{(xi ) ∈ X i  : xi = ai , i ∈
/ F} ⊂ D. For any point (xi )i∈F ∈ i∈F X i , let x F be
the point of i∈I X i defined by x F (i)  = xi , i ∈ F and x F (i) = ai , i ∈
/ F. Then
(xi )i∈F → x F is a homeomorphism of i∈F X i onto C F . Thus, by the finite product
case, each C F is connected. But (ai ) ∈ C F for each F and the union of all the C F is
just the set D. The theorem is thus proved. 

Any space can be split up into disjoint connected pieces.

Theorem 9.22 Let X be a topological space. For each x ∈ X , let C x denote the
union of all connected subsets of X containing x. Then
9.2 Connected Spaces 135

1) C x is connected and is the largest connected set containing x.

2) If z ∈ C x , then C z = C x .
3) For x, y ∈ X , either C x = C y or C x and C y are disjoint.
4) C x is closed in X for every x ∈ X .
Proof 1) For x ∈ X , the collection Ex of all connected subsets E x with x ∈ E x is
nonempty since {x} ∈ Ex . Since x ∈ ∩Ex , C x = ∪Ex is connected (9.16). Maximality
of C x is clear since each E x ⊂ C x .
2) If z ∈ C x , then C x ⊂ C z as C x is connected and C z is maximal connected. This
gives x ∈ C z and, as before, C z ⊂ C x . This proves 2).
3) is immediate from 2): if z ∈ C x ∩ C y , then C x = C z = C y .
4) C x is maximal connected and C̄ x is connected, so C x = C̄ x . 
Definition 9.23 For any x in a topological space X , the largest connected set C x
containing x obtained above is called the connected component of x. If C x = {x} for
every x ∈ X (i.e. the only connected subsets are one-point sets) then X is said to be
totally disconnected.
Remark 9.24 C x are the equivalence classes under the equivalence relation x ∼ y
if there is a connected set containing x and y both. Totally disconnected spaces were
considered in Knaster–Kuratowski [8]. Connected components were introduced by
Hausdorff [9].
Example 9.25 1. X is connected ⇔ there is only one component.
2. [0,1]∪ (2,3) has two connected components, [0,1] and (2,3).
3. A discrete space is totally disconnected.
4. Q is totally disconnected (but is not discrete). Note that the connected compo-
nents in this case are not open.
5. The Cantor ternary set is totally disconnected.
6. Let I be an interval in R. If f : I → R is continuous, then the graph of f ,
G( f ) := {(x, f (x)) : x ∈ I }, is a connected subset of the plane R2 , being the image
of I under the continuous map x → (x, f (x)).
7. The unit circle S1 is a connected set in the plane, being the image of the
interval [0,1] under the continuous map t → (cos 2π t, sin 2π t). Consequently, the
torus S1 × S1 and the cylinder S1 × R are connected.
8. The set G L(n, R) of all nonsingular n × n real matrices, considered as a sub-
2
space of Rn , is not connected because its image under the continuous determinant
function, A → det A, is the disconnected space R \ {0}. In fact, it can be shown that
it has two connected components G L + (n, R) and G L − (n, R), consisting of matrices
with positive and negative determinants, respectively.
9. The space O(n) also has two connected components, the sets of those orthogonal
matrices of determinant ±1.
10. For any space X , the space X̃ of equivalence classes C x with the quotient
topology is totally disconnected.
The form of connectedness discussed in the previous section, i.e. path connect-
edness, is stronger than that of the present section.
136 9 Connected Spaces

Fig. 9.5 Comb space

Proposition 9.26 If, for every pair x, y of points in a space X , there is a connected set
C = C x,y ⊂ X containing both, then X is connected. In particular, a path connected
space is connected.

Proof A disconnection X = A ∪ B of X leads to a disconnection C x,y =

(A ∩ C x,y ) ∪ (B ∩ C x,y ) for C x,y , x ∈ A, y ∈ B. Alternately, y ∈ C x,y ⊂ C x for all
y ∈ X , so X = C x for any x ∈ X .
For x, y ∈ X , any path from x to y is a connected set containing both x and y, so
the first part applies to yield the second part. 

The converse is not true. A connected space may not be path connected. Here are
two examples.

Example 9.27 1. Let C be the following subset of the plane: the segment (0,1] on
the horizontal x-axis together with vertical line segments of unit length at each point
( n1 , 0), n = 1, 2, . . . (Fig. 9.5). Thus C is the union (0, 1] × {0} ∪ (∪∞
1 { n } × [0, 1]).
1

Then C is path connected. In fact, any two points in C can be joined by a path in C
consisting of at most three line segments. Hence C̄ = C ∪ {0} × [0, 1] is connected.
However, C̄ is not path connected. For example, there is no path in C̄ from the point
(0,1) to the point (1,0). Thus a path connected space may not have a path connected
closure and a connected space may not be path connected. Observe also that C̄ has a
single connected component and two path components, namely C and the segment
{0} × [0, 1]. C is usually referred to as the comb space.
2. Another example with the same properties is the topologist’s sine curve.
Consider G f = {(x, sin(1/x)) : 0 < x ≤ 1}, the graph of the continuous function
f (x) = sin(1/x), 0 < x ≤ 1 (Fig. 9.6). It is path connected, but it closure Ḡ f =
G f ∪ {0} × [0, 1] is not path connected. Here again there is no path from (0,1) to
(1,0) in Ḡ f .
9.2 Connected Spaces 137

sin(1/x)

Fig. 9.6 Topologist’s sine curve

However, we have the following positive result; it is useful, for example, in com-
plex analysis (e.g. Cauchy’s integral theorem).

Proposition 9.28 An open, connected set in Rn is path connected.

Proof Let C ⊂ Rn be open and connected, x ∈ C and Px be the path component

of x in C. For x ∈ C, C contains a ball Bx around x. So Bx ⊂ Px as Bx is path
connected. Thus Px is open. On the other hand, the complement Q x of Px in C is the
union of all the path components of C other than Px . Each of these path components
is open just as Px is. So Q x is open. Thus Px is both open and closed in C. As C is
connected, Px = C and the proof is complete. 

Exercise. 1. Any two points in an open connected set C in Rn can actually be con-
nected by a polygonal arc in C. For example, in the plane there is such a polygonal
path made up of horizontal and vertical line segments.
2. Any product of discrete spaces is totally disconnected.
3. For x ∈ X, y ∈ Y , is the connected component of (x, y) in X × Y the product
C x × C y of the connected components of x and y?

9.3 Locally Connected Spaces

Local analogues of the properties discussed in the last two sections are important in
geometric considerations in topology and analysis.

Definition 9.29 A space X is said to be locally (path) connected if every neighbour-

hood contains a (path) connected neighbourhood; more elaborately, for every x ∈ X
and every neighbourhood U of x there is a (path) connected neighbourhood V of x
such that V ⊂ U .

Locally connected spaces were introduced by Hahn (1914), [10], who proved 9.30
for metric spaces.
Note that discrete spaces (with more than one point) are not connected but are
locally connected. The comb space C and topologist’s sine curve G f are locally
connected, but their closures are not locally connected; for instance, the defining
property fails for the point (0,1).
138 9 Connected Spaces

Proposition 9.30 (Hahn-Tietze), [10, 11]

A space X is locally (path) connected if and only if (path) connected components of
open sets are open.
Proof We write down the proof of the result on local connectedness. The proof for
local path connectedness is exactly similar.
Let C be a connected component of an open set U in a locally connected space X.
If x ∈ C, V ⊂ C for any connected neighbourhood V ⊂ U of x. Thus every point
of C is an interior point and C is open.
Conversely, if the connected components of open sets in X are all open and
x ∈ U ⊂ X, U open, then C x is a connected neighbourhood of x, C x ⊂ U . Thus X
is locally connected. 
Corollary 9.31 If X is locally path connected, then every connected component is
a path component.
Proof Let C x be a connected component of a locally path connected space X . Then
Px ⊂ C x and if the equality does not hold, let Q x be the union of all the path compo-
nents that intersect C x . Then Px , Q x are open and C x = Px ∪ Q x is a disconnection,
contradicting the connectedness of C x . Thus Px = C x and the proof is complete. 
Exercise 9
1. Path connectedness. a) If n > 1 and E is a countable subset of Rn , then
R \ E is path connected. In particular, the set of all points in Rn with at least one
n

irrational coordinate is path connected.

b) Find an uncountable subset E of R2 with E c path connected.
c) The Sierpiński space S (3.2) is path connected.
d) Check for path connectedness/connectedness:
i) the extended real line;
ii) any X with the cofinite topology;
iii) R (3.17), the real line with the lower limit topology;
iv) the complement of any straight line in R3 .
e) A space X is path connected if and only if it is connected and locally path
connected.
2. Connectedness. a) Let f : R → R. True or false?: f is continuous if and
only if its graph G f is connected.
b) Use connectedness to conclude that the following pairs of spaces are not home-
omorphic:
i) R and Rn , n > 1;
ii) S1 and Sn , n > 1;
iii) [0,1] and S1 ;
iv) The closed unit disc and the unit circle in the plane.
c) Let E, C be subsets of a space X and suppose C is connected. If C intersects
both E and E c , then it intersects ∂ E.
d) If C is a connected set which is both open and closed, it must be a connected
component.
9.3 Locally Connected Spaces 139

e) If A, B are proper subsets of connected spaces X, Y respectively, then (X ×

Y ) \ (A × B) is connected.
f) Is the intersection of a decreasing sequence of connected sets necessarily con-
nected?
g) RN is not connected in the box topology: the set A of all bounded sequences
and the set B of all unbounded sequences give a disconnection RN = A ∪ B.
h) If E is a connected set, are int E and ∂ E also connected? What about the
converse?
3. Embedding an interval in the plane a) A homeomorphic image of [0,1] in
R2 has no interior.
b) The image of a continuous injective map [0, 1] → R2 has empty interior. R2
can replaced by Rn , n ≥ 2, in both a) and b).
4. Components. a) If f : X → Y is continuous, show that f (C x ) ⊂ C f (x) for
all x in X and equality holds if f is a homeomorphism.
b) In particular, if X and Y are homeomorphic, then every component in X is
homeomorphic to some component in Y and conversely. There is a bijection between
the components of X and Y . Is the converse of the last statement true?
c) Let A, B be the following subsets of in R. A = ∪∞ n=0 An where An =
∞
(3n, 3n + 1) ∪ {3n + 2} and B = ∪Bn=0 where B0 = (0, 1], Bn = (3n, 3n + 1) ∪
{3n + 2}, n > 0. A and B are not homeomorphic, but there are continuous bijections
f : A → B and g : B → A.[12, 13]
d) Find the connected components and path components of R (3.17) and deter-
mine all the continuous maps R → R .
e) An open set in Rn has only countably many connected components. Give an
example of a set in Rn having uncountably many connected components.
f) A connected component  may not be open.

g) In a product space X = X i , C x = C xi for x = (xi ) ∈ X .
h) {(x, y) ∈ R2 : x y = 1} is not connected. Find its components.
i) If K is a compact subset of R2 , the connected components of K c are all (there
may be infinitely many, but only countable!) path connected and there is exactly one
unbounded component. (See [12].)
j) Find the path components of the comb space (9.27).
5. Quotient map. Let p : X → Y be a quotient map. If X is connected then Y is
connected. Conversely, if Y is connected, then so is X , provided each fibre p −1 (y)
is connected.
6. Totally disconnected spaces. a) Any product of totally disconnected spaces
is totally disconnected.
b) If a space is totally disconnected, so is any subspace.
c) If Z2 is the discrete two element group, ZN 2 is a compact, totally disconnected
topological group.
d) A countable metric space is totally disconnected.
7. Components in a topological group. a) The component Ce of the identity
element in a topological group G is a closed normal subgroup.
b) The quotient G/Ce is totally disconnected.
c) The coset aCe is the component Ca .
140 9 Connected Spaces

9.4 Biographical Notes

9.4.1 Jordan

Camille Jordan (1838–1922) was a French mathematician well-known for his work
on group theory and analysis. Jordan curve theorem (a fundamental theorem in plane
topology), Jordan measure, Jordan decomposition, Jordan matrix, Jordan canonical
form for matrices, Jordan–Höder theorem for groups are important contributions
from him. His Cours d’Analyse was a popular analysis text book in the nineteenth
century. His work in group theory helped to bring Galois theory (discovered by the
French teenager genius Evariste Galois) to the mainstream of mathematics.

9.4.2 Hahn

Hans Hahn (1879–1934) was a lawyer turned mathematician from Austria, who was
also a philosopher. He is best known for his work in functional analysis. His most
famous contributions include the Hahn–Banach theorem (a fundamental theorem
of functional analysis) and the Hahn decomposition theorem in measure theory.
Vitali–Hahn–Saks theorem and Hahn–Mazurkiewicz theorem are also well-known.
He made important contributions to real analysis, calculus of variations and topology
(the concept of local connectedness is due to him). His famous students include Karl
Menger (dimension theory), Witold Hurewicz (higher homotopy groups, dimension
theory) and Kurt Gödel (set theory and logic, the continuum hypothesis, the famous
Gödel incompleteness theorem).

9.4.3 Kuratowski

Kazimierz Kuratowski (1896–1980), usually written Casimir Kuratowski in English,

was one of the founders and a leading member of the famed topology group from
Poland. Kuratowski was largely responsible for reviving Polish mathematics after
the Second World War. This was reflected in the fact that topology was one of the
most active fields of research in Poland during that period. (See [14].) His main math-
ematical contributions were in set theory, topology and measure theory. Kuratowski
closure axioms (see Chap. 3), developed in his thesis (1921), his characterisation of
planar graphs, Kuratowski–Zorn lemma (commonly called Zorn’s lemma—proved
by Kuratowski in 1922 and by Max Zorn in 1935), the theory of Polish spaces
(developed by Kuratowski, Wacław Sierpiǹski and Alfred Tarski) are some his
well-known contributions. His work on connectedness, along with his collaborators
(Knaster et al.), is well-known. For example, Knaster and Kuratowski constructed
the ‘Knaster–Kuratowski fan’, a connected space that becomes totally connected on
9.4 Biographical Notes 141

removing a single point! His monographs on topology are famous. During World War
2, he lectured at the ‘underground university’ and after the war played an important
role in the revival of mathematics and science in Poland.

9.4.4 Knaster

Bronisław Knaster (1893–1980) of Poland is known for his work in Set Topology.
He was a student of Mazurkiewicz. His hereditarily indecomposable continuum,
the Knaster continuum, Knaster–Kuratowski fan (mentioned above), Knaster–Tarski
theorem on lattices, his work with Banach and Steinhaus on ‘fair division’ are well-
known. He was one of the trio to prove Brouwer’s theorem from Sperner’s lemma. It
is interesting to note that he took to mathematics when he was past 20; he had studied
medicine for three years at Paris earlier. He had a passion for mathematical editorial
work and took up editorial work for the Polish mathematical journals Studia Math-
ematica and Colloquium Mathematicum. He was the one who translated Banach’s
celebrated monograph Théorie des opérations linéaires into French.

References

1. Bolzano, B.: Rein analytischer Beweis des Lehrsatzes, dass zwischen je zwei Werthen, die
ein entgegengesetztes Resultat gewähren, wenigstens einer reelle Wurzel der Gleichung liege.
Prague, Gottlieb Haase (1817)
2. Cantor, G.: Über unendliche, lineare Punktmannigfaltigkeiten, Math. Ann. 15 1–7 (1879);
17, 355–388 (1880); 20, 113–121 (1882); 21 51–58 (1883), 545–591; 23, 453–488
(1884). https://doi.org/10.1007/BF01444101, https://doi.org/10.1007/BF01446232. https://
doi.org/10.1007/BF01443330, https://doi.org/10.1007/BF01442612. https://doi.org/10.1007/
BF01446819, https://doi.org/10.1007/BF01446598
3. Jordan, C.: Cours d’Analyse de l’École Polytechnique, vol. I., 2nd edn. Paris (1893)
4. Schoenflies, A.: Beiträge zur Theorie der Punktmengen. Math. Ann. 58, 195–234 (1903).
https://doi.org/10.1007/BF01447784
5. Riesz, F.: Die Genesis des Raumbegriffs. Math. Naturwiss. Ber. Ungarn. 24, 309–353 (1906)
6. Lennes, N.J.: Curves in non-metrical analysis situs with an application in the calculus of
variations. Am. J. Math. 33, 287–326 (1911)
7. Wilder, R.L.: Evolution of the topological concept of “connected". Am. Math. Monthly 85,
720–726 (1978). https://doi.org/10.1080/00029890.1978.11994684, Correction and Adden-
dum, Am. Math. Monthly 87 31–32 (1980). https://doi.org/10.1080/00029890.1980.11994947
8. Knaster, B., Kuratowski, K.: Sur les ensembles connexes. Fund. Math. 2, 206–255 (1921).
https://doi.org/10.4064/fm-2-1-206-255
9. Hausdorff, F. Grundzüge der Mengenlehre. Leipzig (1914)
10. Hahn, H.: Über der allgemeinste eben Punktmenge, die stetiges Bild einer Strecke ist Jahresber.
Deut. Math. Ver. 23, 318–322 (1914). https://doi.org/10.1007/978-3-7091-6601-711
11. Tietze, H.: Beiträge zur allgemeinen Topologie III, Monatsh.f. Math. und. Phys. 32, 15–17
(1923). https://doi.org/10.1007/BF01705586
12. Dugundji, J.: Topology. Prentice Hall of India (1975)
13. Kelley, J.L.: General Topology. Affiliated East-West Press (1969)
14. James, I.M.: From combinatorial topology to algebraic topology. In: James, I.M. (ed.) History
of Topology, pp. 561–574. Elsevier (1999)
Chapter 10
Countability Axioms

Most of us are more comfortable dealing with countable sets than with uncountable
sets. In this chapter we deal with various properties of topological spaces that involve
countable sets in some way. Hausdorff, in his Grundzüge [1] (1914), gives two
countability axioms (which are now called the first and second countability axioms.):
Abzählbarkeitsaxiome: (E) Für jeder Punkt x ist die Menge seiner verschiedenen
Umgebungen Ux höchsten abzählbar.
(F) Die Menge aller verschiedenen Umgebungen U ist abzählbar.
(Countability axioms: (E): For each point x, the set of all different neighbourhoods
Ux is countable.
‘(F): The set of different neighbourhoods U is countable’.)
Fréchet [2] defined another countability condition thus:
“Nous appellerons ensuite classe séparable une classe qui puisse être considérée
d’au moins une façon comme l’ensemble dérivé d’un ensemble dénombrable de ses
propres éléments.” (Fréchet, [2], p. 23) (“We will call a class separable if it can be
considered in at least one way as the derived set of a countable set of its elements.”)

10.1 Countability Properties

Definition 10.1 If a topological space X has a countable local neighbourhood base

Bx at each point x ∈ X, it is called a first countable space (or, more elaborately, a
space satisfying the first countability axiom).
Example 10.2 1. In R, (x − n1 , x + n1 ) and, in general, the balls B(x, 1/n), n ∈ N,
in a metric space, form a countable local base at x.
2. A discrete space is (trivially) first countable.
3. R is first countable.
4. R with the co-countable topology is not first countable.
In first countable spaces, as in metric spaces and unlike general topological spaces,
sequences suffice to describe continuity and limit points.
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 143
K. Parthasarathy, Topology, La Matematica per il 3+2 134,
https://doi.org/10.1007/978-981-16-9484-4_10
144 10 Countability Axioms

Proposition 10.3 Let X be a first countable space.

1) If A ⊂ X and x is a limit point of A, then there is a sequence {xn } in A that
converges to x.
2) Let Y be any space. A map f : X → Y is continuous provided f (xn ) → f (x)
in Y for every sequence {xn } converging to x in X .

Proof 1) Let x be a limit point of A and let {Un } be a local base at x. Choose
xn ∈ Un ∩ A, xn = x for each n. Then xn → x.
2) It suffices to show that f ( Ā) ⊂ f (A) for any A ⊂ X . If x ∈ Ā, by (i), choose
xn ∈ A, xn → x. Then f (xn ) → f (x), so f (x) ∈ f (A). 

Here is the stronger, second countability axiom introduced by Hausdorff.

Definition 10.4 A topological space X having a countable base for its topology is
called a second countable space.

Example 10.5 1. R is second countable: open intervals (a, b), with rational a, b,
form a countable base.
2. Every second countable space X is first countable.
3. R with the discrete topology is first, but not second, countable.
4. R is not second countable (but is first countable, as already noted). Let B be
any base for R . For any x, the interval [x, x + 1) is an open set and so there is a
Bx ∈ B such that x ∈ Bx ⊂ [x, x + 1). Since Bx = B y for x = y (a is min Ba for
every a), the collection {Bx : x ∈ R } is uncountable. Thus B is uncountable. We
have shown that any base for R is necessarily uncountable. We can also conclude
that R × R is not second countable from the fact that it contains an uncountable
discrete subspace (see 8.16).

Proposition 10.6 Let E be a subspace of a space X . If X is first countable or second

countable, then so is E.

Proof If Bx is a countable local base in X at x ∈ E, then the sets E ∩ B, B ∈ Bx

form a countable local base in E at x.
If B is a countable base in X , then the sets E ∩ B, B ∈ B form a countable base
in E. 

Here is another countability property of second countable spaces.

Theorem 10.7 (Lindelöf’s theorem)
Every open cover of a second countable space has a countable subcover.

Proof Let X be a space with a countable base {Bn } and let {Ui }i∈I be an open cover
for X . Let J = {n : Bn ⊂ Ui f or some i ∈ I }. For each n ∈ J , let i n ∈ I be such
that Bn ⊂ Uin . If x ∈ X , there is a Ui with x ∈ Ui and then there is a k such that
x ∈ Bk ⊂ Ui . This means that k ∈ J and x ∈ Uik . In other words, {Uin : n ∈ J } is a
cover for X and is a countable subcover that we are looking for. 
10.1 Countability Properties 145

In view of this theorem [3], a space for which every open cover has a countable
subcover is called a Lindelöf space. Ernst Lindelöf was a mathematician from Finland
and Lars Ahlfors’ teacher.
Example 10.8 1. Every compact space is a Lindelöf space. R is an example of a
Lindelöf space that is noncompact.
2. R is Lindelöf.
Definition 10.9 A topological space is said to be separable if it has a countable
dense subset. Fréchet introduced this property.
Proposition 10.10 A second countable space is separable.
Proof Suppose {Bn } is a countable base for a space X . For each n, let xn ∈ Bn . The
countable set D of all the xn is dense in X . For, every nonempty open set U contains
a Bn and so contains the point xn . Thus D intersects every nonempty open set and
hence is dense. 
Example 10.11 R is separable (Q is dense in R !), but is not second countable.
Thus R satisfies all the countability properties considered in this chapter except
second countability.
In a metric space, separability is equivalent to second countability.
Theorem 10.12 For a metric space X , the following are equivalent:
1) X is second countable.
2) X is Lindelöf.
3) X is separable.
Proof We have seen that 1) implies both 2) and 3) for any space X . For a metric
space X , we show 2)⇒ 3)⇒ 1).
2) ⇒ 3): For each n, the balls B(x, 1/n), x ∈ X, form an open cover Bn of X , so
has a countable subcover Bmn = {Bmn = B(xmn , 1/n)}m . The countable set {xmn } is
dense in X .
3) ⇒ 1): Let D = {xn } be a countable dense set in a metric space X . We prove
that the countable collection {B(xm , 1/n) : m, n ∈ N} of open balls form a base
for X . If U is open and x ∈ U , then B(x, 1/n) ⊂ U for some n and then there
is an xm ∈ B(x, 1/2n) ∩ D, as D is dense. So x ∈ B(xm , 1/2n) ⊂ B(x, 1/n) ⊂ U ,
completing the proof. 

10.2 Urysohn Metrisation

Recall that a topological space X is metrisable if there is a metric on X which induces

the given topology. We know that a metric space is normal. In the presence of second
countability, normality is also sufficient for metrisability. In fact, regularity and sec-
ond countability together yields normality and the Tychonoff–Urysohn metrisation
theorem says that a regular, second countable space is metrisable.
146 10 Countability Axioms

Proposition 10.13 (Tychonoff) A regular space having a countable base is normal.

Proof Let A, B be disjoint closed subsets of a regular space X with a countable
base {Bn }. B c is a neighbourhood of any x ∈ A. By regularity there is an open set
Ux such that x ∈ Ux ⊂ Ūx ⊂ B c . Since {Bn } is a basis, there is an n(x) such that
x ∈ Bn(x) ⊂ Ux . Then {Bn(x) : x ∈ A} is a countable open cover for A such that
 
B̄n(x) ⊂ B c . Denote this open cover of A by {Un }. Then Ūn ⊂ B c for all n. Similarly,
   
we get an open cover {Vn } of B such that V̄n ⊂ A for all n. Define Un := Un \ ∪n1 V̄k
c
 
and Vn := Vn \ ∪n1 Ūk . Then U := ∪∞ ∞
1 Un , V := ∪1 Vn are open, A ⊂ U and B ⊂ V .

If x ∈ U ∩ V , then x ∈ U j ∩ Vk for some j, k, say j ≤ k. Then x ∈ U j ⊂ U j and
 
x ∈ Vk = Vk \ ∪k1 Ūi , a contradiction, proving that U and V are disjoint. 
If X is normal, U is open and x0 ∈ U , then {x0 } and U c are disjoint closed sets, so
by Urysohn’s lemma, there is a continuous function f = f x0 ,U : X → [0, 1] such that
f (x0 ) = 1 and f = 0 outside U . In the presence of a countable basis, the following
lemma shows that there is a countable family of functions such that for any choice
of x0 and U some member of the family has this property.
Lemma 10.14 Let X be a normal space having a countable base. There is a count-
able collection of continuous functions f n : X → [0, 1] such that for any x0 ∈ X and
any open neighbourhood U of x0 , there is an n with f n (x0 ) > 0 and f n = 0 on U c .
Proof Let S = {(m, n) : B̄m ⊂ Bn } where {Bn } is a countable basis for X . For each
(m, n) ∈ S, by normality there is a continuous function gmn : X → [0, 1] such that
gmn = 0 on Bnc and gmn = 1 on B̄m . Claim: the collection {gmn } satisfies the required
condition. To see this, let x0 ∈ U , U open. Then there is a Bn 0 such that x0 ∈ Bn 0 ⊂ U .
Using regularity, twice successively, get an open V and a Bm 0 such that
x0 ∈ Bm 0 ⊂ B̄m 0 ⊂ V ⊂ V̄ ⊂ Bn 0 ⊂ U .
Then gm 0 n 0 = 0 on U c ⊂ Bnc0 and gm 0 n 0 = 1 on Bm 0 . 
Theorem 10.15 (Urysohn metrisation theorem)
A regular, second countable space embeds in [0, 1]N and is metrisable.
Proof Let X be a regular space with countable basis. To get the embedding of X in
[0, 1]N we make use the two preceding results. The first one (10.13) shows that X
is normal and so the lemma applies. Let { f n } be a countable family of continuous
functions as in the preceding lemma. We apply the Embedding Lemma 8.31 with
F = { f n }, Y fn = [0, 1]. Proposition 10.13 implies X is normal and so the previous
lemma shows that F separates points and closed sets (and hence separates points).
Invoking Lemma 8.31, we conclude that the evaluation is an embedding of X in
[0, 1]N . But [0, 1]N , being a countable product of metric spaces, is metrisable, hence
so is its subspace e(X ). Hence X is also metrisable. 
Remark 10.16 i) Vietoris already had proved that compact second countable spaces
are metrisable in 1922 [4]. Urysohn’s paper [5] on metrisation appeared posthu-
mously. He had assumed normality and asked whether it could be weakened to reg-
ularity: “Nun fragt es sich ... ob nicht schon die Regularität allein für die Metrisier-
barkeit der das II. Abzählbarkeitsaxiom erfüllenden Räume genügt.” (“Now the
10.2 Urysohn Metrisation 147

question arises whether ... regularity alone is not enough for the metrisability of
spaces satisfying the second axiom of countability”.) Tychonoff answered this affir-
matively in [6] by proving that a second countable regular space is, in fact, normal.
Tychonoff was 19 at that time!
ii) The general metrisation theorem, giving necessary and sufficient conditions
for metrisation of a topological space, was obtained, independently, by three math-
ematicians in three different countries: Bing (USA), Nagata (Japan) and Smirnov
(Soviet Union) [7–9]. See [10–13] for details.

Exercise 10

1. Examples. a) Define a set U ⊂ R2 to be open if U \ (R × {0}) is open in the

usual topology. This defines a topology T on R2 . Discuss the countability axioms
for (R2 , T ).
b) Which countability axioms does the cofinite topology satisfy?
2. Separable spaces.
a) A continuous image of a separable space is separable.
b) A subspace of a separable metric space is separable.
c) An open subspace of a separable space is separable.
d) A subspace of a separable space need not be separable: Look at the ‘antidiag-
onal’ {(x, −x) : x ∈ R } of the separable space R × R .
e) C[0, 1] with the sup metric is separable.
f) The sequence spaces 1 and 2 are separable. But ∞ is not separable. More
generally B(X ), the space of bounded functions on a set X , is not separable when X
is infinite.
g) A countable product of separable spaces is separable.
h) The product space [0, 1][0,1] is separable, but not first countable.
i) If X is separable, any family of disjoint open sets in X is countable. Is the
converse true?
3. First countable spaces. a) Subspaces of, and countable products of, first
countable spaces are first countable.
b) Let X be first countable. A subspace V of X is open if and only if any convergent
sequence in X with limit in V eventually lies in V .
c) [0, 1] × [0, 1] with the dictionary order topology is first countable but is not
separable.
4. Second countable spaces. a) Any base in a second countable space contains
a countable subfamily which is also a base.
b) Any compact metric space is second countable.
c) Let X denote {(x, y) ∈ R2 : y ≥ 0} = E ∪ F the closed upper half-plane,
where F stands for the x-axis and E = X \ F. Define a topology on X as fol-
lows. For point in E take the usual neighbourhoods and for a point p ∈ F, the basic
neighbourhoods are of the form (B( p, r ) ∩ E) ∪ { p} (i.e. open upper half-discs with
centre p together with the point p). X is separable and first countable, but not second
countable.
148 10 Countability Axioms

d) A second countable Hausdorff space may not be metrisable. Here is a standard

example: R with basic neighbourhoods of 0 as (−ε, ε) \ {1/n : n ∈ N} and for all
other points the usual neighbourhoods.
e) A second countable space is sequentially compact if and only if it is compact.
5. Products. The product of two second countable spaces is second countable.
What about infinite products?
6. Lindelöf spaces. a) R × R is not Lindelöf.
b) If X is compact and Y is Lindelöf, then X × Y is Lindelöf.
c) We say that a collection {E i } of subsets has the countable intersection property
if every countable subcollection has nonempty intersection. A space X is Lindelöf if
and only if for every collection {E i } of subsets having countable intersection property,
∩i Ē i is nonempty.
d) [0, 1] × [0, 1] with the dictionary order topology is compact, but the subspace
(0, 1) × [0, 1] is not Lindelöf.
e) A regular space is Lindelöf if and only if every open cover has a countable
subcollection whose closures form a cover.
f) A regular, Lindelöf space is normal.
g) A regular Lindelöf space is paracompact. (Not easy. See [12].)
7. Usrysohn metrisation. Here is an alternative proof of the Urysohn metri-
sation theorem. Observe that replacing f n by n1 f n , we may assume the f n takes

values in [0, n1 ]. Consider the Hilbert cube H = ∞ 1 [0, n ], endowed with the uni-
1

form metric defined by d(x, y) = sup |xn − yn |. Define  : X → H as before:

(x) = ( f 1 (x), f 2 (x), · · · ).  is a homeomorphism of X onto its image (X ) in
H.
8. Countability in groups. a) If a topological group G is T1 and second countable
, it is metrisable (it is enough to assume first countability).
b) If G is separable and first countable, it is second countable.

10.3 Biographical Notes

10.3.1 Lindelöf

Ernst Lindelöf (1870–1946) of Finland worked in a number of areas in mathematics

including topology, complex analysis, differential equations, real analysis. The result
that any open set in the real line is a countable union of open intervals, Lindelöf’s
theorem and Lindelöf spaces that we have encountered in this chapter, Lindelöf’s
theorem and Phragmén–Lindelöf Principle that give conditions for an analytic func-
tion in unbounded domains to be bounded and generalises the Maximum Modulus
Principle, Picard–Lindelöf theorem on the existence and uniqueness of solutions of
first order ordinary differential equations (often called Picard’s theorem), Lindelöf
10.3 Biographical Notes 149

hypothesis on the rate of growth of the Riemann zeta function on the critical line are
some of his significant contributions. Lars Ahlfors, the great complex analyst who
was one of the first two Fields medallists (1936), was his student at Helsinki.

References

1. Hausdorff, F. Grundzüge der Mengenlehre. Leipzig (1914)

2. Fréchet, M.: Sur quelques point du calcul fonctionnel. Rendiconti di Palermo 22, 1–74 (1906).
https://doi.org/10.1007/BF03018603
3. Lindelöf, E.: Sur Quelques Points de la Théorie des Ensembles. C. R. Acad. Sci. Paris 137,
697–700 (1903)
4. Vietoris, L.: Bereiche zweiter Ordnung. Monatsh. Math. und Physik 32, 258–280 (1922).
https://doi.org/10.1007/BF01696886
5. Urysohn, P.: Zum Metrisationproblem. Math. Ann. 94, 309–315 (1925). https://doi.org/10.
1007/BF01208661
6. Tychonoff, A.: Über einen Metrisationsatz von P. Urysohn. Math. Ann. 95, 139–142 (1926).
https://doi.org/10.1007/BF01206602
7. Bing, R.H.: Metrization of topological spaces. Can. J. Math. 3, 175–186 (1951). https://doi.
org/10.4153/CJM-1951-022-3
8. Nagata, J.: A Necessary and Sufficient Condition for Metrizability. J. Inst. Polytech. Osaka
City University 1, 93–100 (1950)
9. Smirnov, Y.M.: A necessary and sufficient condition for metrizability of a topological space.
Dokl. Akad. Nauk SSSR (NS) 7, 197–200 (1951). (Russian)
10. Dugundji, J.: Topology. Prentice Hall of India (1975)
11. Kelley, J.L.: General Topology. Affiliated East-West Press (1969)
12. Munkres, J.R.: Topology, 2nd edn. Pearson Education, Asia (2001)
13. Willard, S.: General Topology. Addison-Wesely (1970)
Chapter 11
Locally Compact Spaces

Not all important spaces are compact—even the real line is not. Locally compact
spaces, introduced, independently, by Tietze [1] and Alexandroff [2], are nice enough
and sufficiently abundant to be very important in analysis and geometry.

11.1 Local Compactness

Alexandroff’s definition is as follows:

Ein topologischer Raum R heißt im Punkte ξ bikompakt (bzw. kompakt), falls
eine gewisse Umgebung U (ξ ) dieses Punktes existiert, so daft die abgeschlossene
Menge Ū (ξ ), als Relativraum betrachtet, bikompakt (bzw. kompakt) ist. Ein in jeder
seiner Punkte bikompakter (bzw. kompakter) topologischer Raum heißt im Kleinen
bikompakt (bzw. kompakt). [2] 1924
This definition says: a space R is bicompact at ξ if the point has a neighbour-
hood U (ξ ) whose closure is bicompact and the space is ‘Kleinen’ bicompact if it
is bicompact at each point. (Bicompact = compact). ‘Kleinen bicompakt’ (=‘small
compact’) is now called ‘locally compact’.
The real line R is not compact, but for any x ∈ R, every neighbourhood of x
contains a neighbourhood of the form [x − δ, x + δ], which is compact. (Recall our
convention: E is a neighbourhood of x if x is an interior point of E.) If a space is
not compact, the next best thing is to have a local property like this. Such spaces are
extremely important in various areas of mathematics.
Definition 11.1 A topological space X is said to be locally compact if every point
x ∈ X has a compact neighbourhood; i.e. there is an open set V such that x ∈ V and
V̄ is compact.

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 151
K. Parthasarathy, Topology, La Matematica per il 3+2 134,
https://doi.org/10.1007/978-981-16-9484-4_11
152 11 Locally Compact Spaces

Sets with compact closure are called relatively compact or precompact sets. Thus X is
locally compact means that every point of X has a precompact open neighbourhood.

Example 11.2 Any compact space is locally compact and so is any discrete space.
The Euclidean space Rn and an infinite discrete space are locally compact and non-
compact. Q is not locally compact.
Theorem 11.3 In a locally compact Hausdorff space X , every neighbourhood of
any x ∈ X contains a compact neighbourhood.
Proof Let x ∈ X and let V be an open neighbourhood of x with V̄ compact. Then
V̄ is a compact Hausdorff space, hence is normal and is therefore regular. If U is
a neighbourhood of x, U ∩ V̄ is a neighbourhood of x in this regular space and so
there is a relative neighbourhood E of x with U ∩ V̄ ⊃ Ē V̄ := the closure of E
in V̄ = Ē ∩ V̄ . But E = V̄ ∩ W , with W an open neighbourhood of x in X . Now
V ∩ W is a neighbourhood of x in X with compact closure, being a closed subset of
V̄ and is contained in U . 
This property extends to compact sets in place of points: every neighbourhood of
a compact set contains a compact neighbourhood.
Corollary 11.4 Let X be a locally compact Hausdorff space, K be a compact subset
and U be an open subset such that K ⊂ U . Then there is a precompact open set V
such that K ⊂ V ⊂ V̄ ⊂ U .
Proof For each x ∈ K there is a precompact open set Vx such that x ∈ Vx ⊂ V̄x ⊂ U .
By compactness of K , the open cover {Vx : x ∈ K } of K has a finite cover {Vx j }m 1 . If
V = ∪m 1 Vx j
, then V̄ = ∪m
1 V̄x j
and so V̄ , being a finite union of compact sets, is com-
pact. Thus V is a precompact set with K ⊂ V ⊂ V̄ ⊂ U , and the result is proved. 
Corollary 11.5 A locally compact Hausdorff space has a basis of open precompact
sets.
Proof This is immediate from the theorem. 
If the space is also second countable, we can improve on this.
Proposition 11.6 A locally compact Hausdorff space X with a countable basis has
a countable base consisting of precompact sets.
Proof Let {Bn } be a countable basis for X . Cover each Bn by precompact open sets,
{Wx : x ∈ Bn }. As X is second countable, so is Bn , hence is a Lindelöf space. So the
open cover {Wx : x ∈ Bn } has a countable subcover, say {Wn,k }k∈N . The countable
collection {Wn,k : k, n ∈ N} of open precompact sets form a basis for X . 
Proposition 11.7 A subspace E of a locally compact Hausdorff space X is locally
compact if and only if E is locally closed in X , or equivalently, E = V ∩ F where
V is open and F is closed in X . In particular, open sets and closed sets in X are
locally compact.
11.1 Local Compactness 153

Proof If E is locally closed (3.27), for each x ∈ E there is neighbourhood U of

x in X such that E ∩ U is closed in U . Since X is locally compact, there is a
compact neighbourhood V ⊂ U of x. E ∩ V is a neighbourhood of x in E, E ∩ V =
U ∩ E ∩ V is closed in V and hence is compact. Thus every point of E has a compact
neighbourhood.
Conversely, if E is locally compact, then every x ∈ E has a neighbourhood U in
X such that E ∩ U is compact, hence closed in U .
The mentioned equivalence is 3.28. 

Example 11.8 The space M(n, R) of matrices, with the usual Euclidean topology, is
locally compact and Hausdorff. Hence so is the open subset G L(n, R) of nonsingular
matrices. The space S L(n, R) of matrices of determinant one is closed and so is
locally compact too.

The intersection of two dense sets may not be dense (it can even be empty!), but
if the sets are also open, the intersection (of even countably infinitely many of them)
is dense. The next two results are valid also in complete metric spaces; see 12.14 and
12.15.
Theorem 11.9 (Baire’s theorem)
If {Vn }∞
1 is a countable collection of open, dense subsets in a locally compact Haus-
dorff space X , then ∩Vn is also dense.

Proof Let W0 ⊂ X be nonempty, open. Then V1 ∩ W0 is open, nonempty as V1

is dense. So, by local compactness, there is a nonempty open precompact set W1
with W̄1 ⊂ V1 ∩ W0 . Inductively, we get open, precompact sets Wn such that W̄n ⊂
Vn ∩ Wn−1 for all n ≥ 1. W̄n is a decreasing sequence of compact sets, so has finite
intersection property and hence K = ∩∞ 1 W̄n is nonempty and compact. W0 ∩ (∩Vn )
is nonempty and in fact contains K since K ⊂ W0 and K ⊂ Vn for all n. Thus ∩Vn
intersects every nonempty open set and so is dense. 

Corollary 11.10 (Baire Category Theorem)

A locally compact Hausdorff space X cannot be a countable union of closed sets Fn
with empty interiors.

Proof Fn is closed and int Fn is empty

⇔ Vn = Fnc is open and dense, and.
X = ∪Fn ⇔ ∩Vn is empty. 

A set E whose closure has empty interior is called a nowhere dense set. Note that E
is nowhere dense if and only if E c is dense. For remarks on the term ‘category’ see
remarks after 12.15

Example 11.11 Consider R and let Vq = {q}c , q ∈ Q. Each Vq is open and dense
and ∩q Vq is the set of irrational numbers, which is dense. This also shows that the
set of irrational numbers is a G δ . Now Baire’s theorem implies that Q is not a G δ .
For, if Q = ∩Un is a countable intersection of open sets Un , then Un ⊃ Q for each
154 11 Locally Compact Spaces

n and so Un is dense. But then {Vq } ∪ {Un } is a countable family of open dense sets
in R with empty intersection, contradicting Baire’s theorem. The same arguments
show that no countable dense set is a G δ in R.

In the presence of compactness, we have seen that Hausdorff spaces are normal. If
only local compactness is assumed, we can still get complete regularity of Hausdorff
spaces.

Theorem 11.12 A locally compact Hausdorff space is a Tychonoff space.

Proof Let a ∈ X , B be closed and a ∈ / B. B c is a neighbourhood of a and so there

is a precompact open V with a ∈ V ⊂ V̄ ⊂ B c . Repeating this argument, we get a
precompact open set W such that a ∈ W ⊂ W̄ ⊂ V ⊂ V̄ ⊂ B c . Since V̄ is compact
and Hausdorff, it is normal. Urysohn’s lemma gives a continuous function f : V̄ →
[0, 1] such that f (a) = 1 and f = 0 on V̄ ∩ W c . Defining f to be 0 on W c , the
pasting lemma ensures the continuity of f on X . This f satisfies the conditions
f (a) = 1 and f = 0 on B and the proof is complete. 

We now look at products of locally compact spaces.

Proposition 11.13 If X, Y are locally compact Hausdorff spaces, then so is their

product X × Y .

Proof We have already seen that products of Hausdorff spaces are Hausdorff, so
we need to show that X × Y is locally compact. This is easy. Let x ∈ X, y ∈ Y .
Then there are open sets U ⊂ X, V ⊂ Y with Ū and V̄ compact. Then U × V is an
open neighbourhood of (x, y) in X × Y with compact closure Ū × V̄ . The proof is
complete. 

Remark 11.14 The result, clearly, extends to a finite product.

 Consider RN . A basic
neighbourhood of 0 = (0, 0, . . . , ) is of the form V = n Bn where Bn is an interval
around 0 for finitely many n and Bn = R for all other n. So V̄ is not compact. So
RN is not locally compact. The argument shows that the product of infinitely many
noncompact, locally compact spaces is not locally compact. in fact, we have the
following result.

Theorem 11.15 A product X i of locally compact spaces is locally compact if and
only if all but finitely many X i are compact.

Proof Let X = X i . Suppose X i are compact for i ∈ / F for a finite set F of indices.
Let x = (xi ) ∈ X and take Vi =  X i for i ∈
/ F and Vi be a compact neighbourhood
of xi in X i for i ∈ F. Then V = Vi is a compact neighbourhood of x in X . Thus,
X is locally compact.
Conversely, suppose X is locally compact. Let x ∈ X and let B = ∩i∈F pi−1 (Vi )
be a compact basic neighbourhood of x, where F is a finite set of indices and Vi is
a compact neighbourhood of xi . Then, for i ∈ / F, X i = pi (B) is compact. 
11.1 Local Compactness 155

Remark 11.16 In the converse part, from the local compactness of X we can actually
conclude that the X i are locally compact from the facts that the projections are open
surjections.

Proposition 11.17 An open continuous image of a locally compact space is locally

compact.

Proof Let X be locally compact and f : X → Y be a continuous, open surjection.

Let y ∈ Y and write y = f (x). Choose a compact neighbourhood U of x in X . Then
f (U ) is compact by continuity of f , and it is a neighbourhood of f (x) = y since f
is open. 

11.2 One-Point Compactification

If S1 = {(x, y) ∈ R2 : x 2 + y 2 = 1}, the map σ : S1 \ {(0, 1)} → R defined by

σ (x, y) = 1−y
x
is a homeomorphism of the punctured circle onto the real line.
(Of course, this is just the stereographic projection and extends to higher dimen-
sions, giving a homeomorphism from the punctured sphere Sn \ {(0, . . . , 0, 1)} to
the Euclidean space Rn .) Identifying R with S1 \ {(0, 1)} = σ −1 (R), we observe that
the noncompact space R has a compactification S1 got by adding just one more point.
This kind phenomenon holds for any locally compact Hausdorff space.
Removing a point from any compact Hausdorff space yields a locally compact
space: If Y is compact, Hausdorff and p ∈ Y , then Y \ { p} is locally compact. We
have a converse, due to Alexandroff [2].
Theorem 11.18 (One-point Compactification)
Let X be a noncompact locally compact Hausdorff space.
1) There is a compactification (X ∞ , h ∞ ) of X such that X ∞ \ h ∞ (X ) is a single
point.
2) Such a compactification is uniquely determined, up to a homeomorphism. In
fact, if (Y j , h j ) are two compactifications of X with Y j \ h j (X ) a singleton, j = 1, 2,
there is a homeomorphism η : Y1 → Y2 ‘which fixes points of X ’; more precisely, η
is such that h 2 = η ◦ h 1 .

Proof 1) Consider an ‘object’ that is not an element of X and, following usual

convention, denote it by ∞. Write X ∞ := X ∪ {∞}, let K be the class of compact
subsets of X and take
T∞ = T ∪ {X ∞ \ K : K ∈ K} ∪ {X ∞ },
where T is the topology of X ; i.e. T∞ the collection consists of open sets in X ,
complements in X ∞ of compact subsets of X and X ∞ itself.
a) T∞ is a topology on X ∞ . For this we have to show that T∞ is closed
under arbitrary unions and finite intersections. So let {Ui } ⊂ T∞ and let U = ∪i Ui .
If ∞ ∈/ Ui for any i, then Ui ∈ T and so U ∈ T ⊂ T∞ . If ∞ ∈ U j for some
j, then U j = X ∞ \ K j for a compact K j ⊂ X . Take K i = X ∞ \ Ui , i = j, and
156 11 Locally Compact Spaces

Fig. 11.1 Uniqueness of h1

one-point compactification X Y1
η
h2
Y2

C j = K j ∩ (∩i= j K i ). Then U = X ∞ \ C j . C j , being a closed subset of a compact

space is compact and so U ∈ T∞ .
Now consider finite intersections. Let V1 , . . . , Vn ∈ T∞ and let V = ∩n1 Vk . If
∞ ∈ Vk for all k, then ∞ ∈ V and X \ V = ∪n1 (X \ Vk ) is a finite union of compact
sets and so is compact. On the other hand, if ∞ ∈ / Vk for some k, then ∞ ∈ / V and
V ∈T.
b) The relative topology on X is nothing but the given topology T itself. It is also
clear that X is an open, dense subspace of X ∞ .
c) X ∞ is compact. For, if {Ui } is an open cover of X ∞ , choose i 0 with ∞ ∈ Ui0 .
Then Uic0 is compact and so finitely many Ui , say Ui1 , . . . , Uin cover Uic0 . Thus {Uik }n0
is a cover for X ∞ .
d) X ∞ is Hausdorff. Since X is Hausdorff, distinct points of X can be separated
by disjoint open sets in X , hence in X ∞ . To show that any point in X and ∞ can
be separated. This is easy. Let x ∈ X and let U be any neighbourhood of x in X .
By local compactness, there is a precompact open set V such that x ∈ V ⊂ V̄ ⊂ U .
Then V and X ∞ \ V̄ are disjoint neigbourhoods of x and ∞, respectively.
Note that, in our construction, X ⊂ X ∞ and h ∞ is nothing but the inclusion map
ι : X → X ∞ .
2) Suppose (Y j , h j ), j = 1, 2, are two compactifications of X as in b). Write
Y j \ h j (X ) = {∞ j }. Define η : Y1 → Y2 by η(h 1 (x)) = h 2 (x) for x ∈ X and
η(∞1 ) = ∞2 (Fig. 11.1). Then h 1 = η ◦ h 2 , by the definition of η. To complete
the proof, we show that η is a homeomorphism.
If y = lim yα in Y1 , to show that η(y) = lim η(yα ).
Case 1: y = ∞1 ; to show that lim η(yα ) = η(∞1 ) = ∞2 .
Take any neighbourhood of ∞2 in Y2 , say Y2 \ h 2 (K ), where K ⊂ X is compact.
Since lim yα = ∞1 , there is an α0 such that yα belongs to the neighbourhood
Y1 \ h 1 (K ) of ∞1 for α ≥ α0 . Since η is bijective, this shows η(yα ) ∈ η(Y1 ) \
η(h 1 (K )) = Y2 \ h 2 (K ) for all α ≥ α0 . This just means that the η(yα ) lie eventually
in any neighbourhood of ∞2 in Y2 , or, equivalently, lim η(yα ) = ∞2 = η(∞1 ) =
η(y).
Case 2: y = h 1 (x) ∈ h 1 (X ); to show lim η(yα ) = η(y) = h 2 (x).
So take any neighbourhood h 2 (U ) of h 2 (x), where U is a neighbourhood of x in X .
Then h 1 (U ) is a neighbourhood of h 1 (x) = y = lim yα in Y1 . There is an α0 such
that yα ∈ h 1 (U ) for all α ≥ α0 . This shows that η(yα ) ∈ η((h 1 (U ))) = h 2 (U ) for all
α ≥ α0 . Thus lim η(yα ) = η(y).
We have thus proved the continuity of η. But, the roles of h 1 and h 2 can be
interchanged and, from the relation η−1 ◦ h 2 = h 1 , we get the continuity of η−1 . The
proof is complete. 
11.2 One-Point Compactification 157

P*

P
Q*

Fig. 11.2 Sn , the one-point compactification of Rn via stereographic projection: N = north pole,
P, Q ∈ Rn ,P ∗ , Q ∗ images on Sn

This is called the one-point compactification of X . (The definite article is justi-

fied by the uniqueness statement.) It is useful in analysis. For example, in complex
analysis, the extended complex plane is the Riemann sphere, the one-point compact-
ification of the complex plane.
Example 11.19 The circle S1 is the one-point compactification of R and the sphere
Sn that of Rn for n > 0, Fig. 11.2. Any open interval, e.g. (−1, 1), is homeomorphic
to R and so has one-point compactification S1 . [0,1] is the one-point compactification
of [0,1).
Exercise The one-point compactification of the discrete space N is {0}∪
{ n1 : n ∈ N}, as a subspace of the real line.
Exercise Discuss X ∞ when X itself is compact.
Exercise If X is discrete, what things can you say about its one-point compactifi-
cation?
The result that a locally compact Hausdorff space is completely regular, 11.12,
now has an easy proof using the one-point compactification.
Corollary 11.20 A locally compact Hausdorff space is a completely regular space.
Proof Let X be a locally compact Hausdorff space. X ∞ is a compact Hausdorff
space, hence is normal.Thus X ∞ is completely regular and so is its subspace X. 
This yields separation of points and closed sets by continuous functions on the
space. But now we can get a stronger separation. This is again easy by going to the
one-point compactification.
158 11 Locally Compact Spaces

(a,1) (b,1)

(c,0) (a,0) (b,0) (d,0)

Fig. 11.3 A trapezoidal function—Urysohn lemma

Theorem 11.21 (Urysohn for Locally Compact Spaces)

Let X be locally compact, Hausdorff, K , C be disjoint subsets with K compact, C
closed. Then there is a function f : X → [0, 1], continuous with compact support,
such that f = 1 on K and f = 0 on C.

The support, supp f , of a continuous (real or complex) function f on a space X is

the closure of the set {x ∈ X : f (x) = 0}. It is the smallest closed set outside which
f vanishes.
Proof K and C are disjoint closed subsets of X . X ∞ is compact, Hausdorff, hence is
normal and Urysohn’s lemma yields a continuous function g : X ∞ → [0, 1] such
that g = 1 on K and g = 0 on C. Then the restriction f := g | X satisfies the
requirements. 
The conclusion can be stated as f = 1 on K and supp f ⊂ U = C c .
Example 11.22 Take X = R and K = [a, b], a < b. Let c, d be real numbers such
that c < a < b < d. Consider the trapezoidal function f defined by f = 1 on K ,
f (x) = 0 for x > d and x < c and linear on [c, a] and [b, d]. See Fig. 11.3. The
reader should explicitly write down the function. Note that f = 1 on [a, b] and supp
f = [c, d].

Theorem 11.23 (Partition of Unity)

Let K be a compact subset of a locally compact Hausdorff space X and let {Vk }n1
be a finite open cover of K . Then there are continuous functions f 1 , . . . , f n on X
with compact support such that 0 ≤ f k ≤ 1, supp f k ⊂ Vk and 1n f k (x) = 1 for all
x ∈ K.

Proof Each x ∈ K is in some Vk and so there is an open precompact set Ux with

x ∈ Ux ⊂ Ūx ⊂ Vk . Since K is compact, there are x1 , . . . , xm such that K ⊂ ∪m 1 Ux j .
For each k, write Ck for the union of those Ūx j which are contained in Vk . Then each
Ck is a compact subset of Vk . By Urysohn’s lemma, there is a continuous function
gk on X with 0 ≤ gk ≤ 1 such that gk = 1 on Ck and supp gk ⊂ Vk , 1 ≤ k ≤ n.
Take f 1 = g1 and f k = (1 − g1 ) · · · (1 − gk−1 )gk for k > 1. It is clear that each
f k is continuous, 0 ≤ f k ≤ 1 and supp f k ⊂ Vk . Now f 1 + f 2 = g1 + g2 − g1 g2 =
1 − (1 − g1 )(1 − g2 ) and, inductively, f 1 + · · · + f n = 1 − (1 − g1 ) · · · (1 − gn ).
It is now clear that the fact that C1 , . . . , Cn cover K yields that 1n f k (x) = 1 for
x ∈ K. 
11.2 One-Point Compactification 159

We have looked at partitions of unity in Chap. 6 too, in the exercises on paracom-

pact spaces. One also has ‘smooth partitions of unity’ on spaces where smoothness
makes sense (Rn or ‘manifolds’). These are important tools in ‘patching up’ local
constructions to yield objects (e.g. smooth functions, vector fields, connections) glob-
ally. Hence, they are important in analysis and geometry and even in ‘applied’ areas
like signal processing.

Exercise 11

1. Examples. Test for local compactness: R

,
2 , R with cocountable topology.
2. One-point compactification. a) Discuss the one-point compactification of an
infinite discrete space.
b) Let X, Y be locally compact Hausdorff spaces. Then every homeomorphism
f : X → Y extends to a homeomorphism of their one-point compactifications
X ∞ → Y∞ .
c) The one-point compactification of a connected space is connected. Is the con-
verse true?
d) If X is compact Hausdorff and K is a closed subset, the quotient got by pinching
K to a point is homeomorphic to the one-point compactification of X \ K .
3. Proper maps and local compactness. Recall the definition of proper maps
from Chap. 7. For this exercise and more on proper maps, see Bourbaki [3]; also
Engelking [4].
a) Let X, Y be Hausdorff and Y locally compact. A map f : X → Y is proper if
and only if it pulls back compact sets to compact sets. If f is proper, X is locally
compact.
b) Let X, Y be locally compact, Hausdorff spaces. A continuous map f : X → Y
is proper if and only if its extension X ∞ → Y∞ to the one-point compactifications
with f (∞) = ∞ is continuous.
c) Let X be a compact Hausdorff space, R an equivalence relation on X and
q : X → X/R the canonical map. Then q is a proper map ⇔ X/R is Hausdorff ⇔
R is closed ⇔ q is a closed map.
d) In c) suppose X is locally compact with one-point compactification X ∞ and let
R∞ = R ∪ {(∞, ∞)}. Then q is proper ⇔ the canonical map q∞ : X ∞ → X ∞ /R∞
is closed ⇔ q is closed and the equivalence classes of R are compact.
e) Let f : X → Y be a proper map, X Hausdorff. Then X is (locally) compact if
and only if f (X ) is (locally) compact Hausdorff.
4. Smooth Urysohn. On R (and on Rn ), the smooth analogue of Urysohn Lemma
(8.19) holds: If K ⊂ V ⊂ R where K is compact and V is open, find a smooth (=
C ∞ , infinitely differentiable) function with f = 1 on K and f has compact support
in V .
5. Smooth partitions of unity. Get a C ∞ partition of unity on Rn .
6. Locally compact images. Let f : X → Y be open, continuous surjective. If
X is locally compact Hausdorff, then for any compact C in Y , there is a compact K
in X with f (K ) = C.
160 11 Locally Compact Spaces

7. Locally compact quotients. a) Let R be an equivalence relation on a locally

compact metric space X such that each equivalence class is compact. If the quotient
map q : X → X/R is closed, then X/R is locally compact.
b) The quotient space of R obtained by identifying all points of Z is Hausdorff,
but is not locally compact.
8. Points of continuity. a) For the pointwise limit of a sequence of continuous
functions on R, the set of points of discontinuity is a set of first category. In particular,
the points of continuity is uncountable.
b) For f : R → R, the set of points of continuity is a G δ . (Conversely every G δ
in R is the set of points of continuity of some f , Kim [5].)
c) Let D be a countable dense set in R. Use b) and Baire to show that there is no
continuous f : R → R whose points of continuity is D.
d) Find a function whose set of points of continuity is R \ Q.

11.3 Biographical Notes

11.3.1 Alexandroff

Pavel (Paul) Alexandroff (1896–1982) was a famous Soviet topologist whose

monograph Topologie, with the German topologist Heinz Hopf, is a classic and
was one of the first in the subject. He was a student of Dmitri Egoroff and Nikolai
Luzin in Moscow. His first result, proved when he was 18, was that any uncountable
Borel set in R contains a copy of the Cantor set. Another famous result: any abso-
lute G δ is homeomorphic to a complete metric space. The ‘open set’ definition of a
topology, the open cover definition of compactness, the one-point compactification,
homological dimension theory, the nerve of a covering, the inverse spectrum of a
space, locally finite families, the concept of an essential map, Alexandroff–Urysohn
metrisation theorem are some of the things credited to him. H. Hopf, Urysohn and the
great Andrei Kolmogorov were his closest friends. Lev Pontrayagin and Tychonoff
were his students.

11.3.2 Dieudonné

Jean Dieudonné (1906–1992) was a French mathematician and was one of the
founders of the Bourbaki group. His thesis, as a student of Paul Montel, was on
classical analysis. He worked in a wide variety of mathematical areas including gen-
eral topology, topological vector spaces, algebraic geometry, invariant theory and the
classical groups. He was the scribe for Bourbaki: Armand Borel mentions ‘the super-
human efficiency of Dieudonné’ and says ‘he took care of the final drafts, exercises,
and preparation for the printer of all the volumes’ [6]. (Quoted with the permission of
11.3 Biographical Notes 161

the American Mathematical Society.) He wrote several books, including a 9-volume

Treatise on Analysis. As a historian of mathematics he has written books on the
history of functional analysis, algebraic and differential topology, algebraic geome-
try, etc. He was also involved in the production of the famous, multi-volume EGA,
Éléments de géométrie algébrique of Alexander Grothendieck that revolutionised
algebraic geometry.

References

1. Tietze, H.: Beiträge zur allgemeinen Topologie II. Über die Einführung uneigentlicher Elemente.
Math. Ann. 91, 210–224 (1924). https://doi.org/10.1007/BF01556079
2. Alexandroff, P.: Über die Metrisation der im Kleinen kompakten topologische Räume. Math.
Ann. 92, 294–301 (1924). https://doi.org/10.1007/BF01448011
3. Bourbaki, N.: General Topology, Chapters 1–4. Springer (1995)
4. Engelking, R.: General Topology Revised Edition. Heldermann Verlag (1989)
5. Kim, S.S.: A characterization of the set of points of continuity of a real function. Am. Math.
Monthly 106, 258–259 (1999). https://doi.org/10.1080/00029890.1999.12005038
6. Borel, A.: Twenty-Five Years with Nicolas Bourbaki (1949–1973). Not. Am. Math. Soc. 45,
373–380 (1998)
Chapter 12
Complete Metric Spaces

Completeness, already mentioned earlier, is an important property in analysis.

A brief introduction to the topic is presented, including a discussion of comple-
tion and the theorems of Ascoli–Arzelà [1, 6], Bourbaki [2], Baire [3] and Banach
[4]. Here is the first definition, given by Fréchet.
Nous dirons alors qu’une classe (V) admet une généralisation du théorème de
CAUCHY si route suite d’éléments de cette classe, qui satisfait aux conditions de
CAUCHY, a un élément limite (nécessairement unique). Fréchet [5, p. 23]
(‘We say a class V admits a generalisation of Cauchy’s theorem if every sequence
satisfying the Cauchy condition has a limit element (necessarily unique)’.) Class V
is what we now call a metric space. Note that he does not use the word ‘complete’.

12.1 Completeness and Ascoli–Arzelà

Arzelà–Ascoli theorem is about precompact sets in spaces of continuous functions.

We present an elementary version. We begin with recalling the concept of complete-
ness.
Definition 12.1 A Cauchy sequence in a metric space (X, d) is a sequence {xn } the
terms of which are ‘eventually close to each other’; in precise terms, {xn } is a Cauchy
sequence if, for every ε > 0, there is a positive integer n ε such that d(xn , xm ) < ε
for all m, n > n ε . The space or the metric is said to be complete if every Cauchy
sequence in X has a limit in X .

Proposition 12.2 Any convergent sequence is a Cauchy sequence.

Proof If the terms of a sequence are eventually close to some point, they are also
eventually close to each other. The reader should write out a formal proof (which
takes hardly more space). 

So, the reverse implication is the one of interest and importance. Recall that the
property of R on which most of the basic results of real analysis is based is its
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 163
K. Parthasarathy, Topology, La Matematica per il 3+2 134,
https://doi.org/10.1007/978-981-16-9484-4_12
164 12 Complete Metric Spaces

completeness. In fact, the chief reason for the construction of real numbers from
rational numbers is the incompleteness of Q and R was obtained by ‘completing’ Q.

Example 12.3 The basic fact on which completeness of function spaces and sequence
spaces rests is the completeness of R.
1. Rn and Cn are complete for all n with each of the metrics d1 , d2 and d∞ (which
are defined on Cn just as for Rn ).
2. For any metric space X , the space Cb (X ) of bounded (real or complex) continu-
ous functions on X is complete with the sup metric: dsup ( f, g) =
supx∈X | f (x) − g(x) | is complete. (The basic underlying reason: ‘uniform limits
of continuous functions are continuous’.)
3. The sequence spaces 1 , 2 and ∞ given earlier are all complete.
4. A discrete metric space is complete: in a discrete metric space, a Cauchy
sequence is eventually a constant and hence is convergent.
1 ∞
5. In (0,1), { n+1 }1 is a Cauchy sequence having no limit.
Which subspaces of complete spaces are complete?
Proposition 12.4 Let X be a metric space and let E ⊂ X .
1) If E is complete (with the induced metric), then it is closed in X (i.e. a complete
subspace of a metric space is closed.)
2) If X is complete and E is closed in X , then E is complete (i.e. a closed subspace
of a complete metric space is complete).

Proof 1) Suppose E is complete. If xn → x, xn ∈ E, x ∈ X , then it is a Cauchy

sequence in E. By completeness of E, it has a limit say y = lim xn ∈ E. But then
x = lim xn = y ∈ E, hence E is closed.
2) If {xn } is a Cauchy sequence in E, then it is so in X . X is complete and
x = lim xn ∈ X exists. Since xn ∈ E for all n and E is closed, x ∈ E. Thus every
Cauchy sequence in E has a limit in E. 

Here is an intrinsically interesting result; it will be used in 11.9.

Proposition 12.5 Let U be a nonempty open set in a metric space (X, d). Define
dU (x, y) = d(x, y)+ | d(x,U1
c ) − d(y,U c ) |, x, y ∈ U. Then dU is a metric on U equiv-
1

alent to d and (U, dU ) is complete if (X, d) is.

Proof To get the triangle inequality, note that, for x, y, z ∈ U

dU (x, y) = d(x, y) + | d(x,U
1
c ) − d(y,U c ) |
1

≤ d(x, z) + d(z, y) + | d(x,U

1
c ) − d(z,U c ) | + | d(z,U c ) − d(y,U c ) |
1 1 1

= dU (x, z) + dU (z, y).

The other properties of a metric are clear.
12.1 Completeness and Ascoli–Arzelà 165

To show that d and dU are equivalent on U , we need to show that convergence

in d and dU are the same. Since d(x, y) ≤ dU (x, y) for x, y ∈ U , dU -convergence
implies d-convergence. To see the converse, note that for xn , x ∈ U and z ∈ U c ,
d(xn , z) ≤ d(xn , x) + d(x, z), so d(xn , U c ) ≤ d(xn , x) + d(x, U c ).
Thus d(xn , x) → 0 implies d(xn , U c ) → d(x, U c ), and hence implies that
dU (xn , x) → 0.
Suppose (X, d) is complete and let {xn } be a Cauchy sequence in (U, dU ). Then
{xn } is d-Cauchy and so has a limit x ∈ X . We need to prove that x ∈ U . On the
contrary, suppose x ∈ U c . Then 0 = d(x, U c ) = lim d(xn , U c ) and so d(xn1,U c ) →
∞. However, {xn } is dU -Cauchy gives that { d(xn1,U c ) } is a Cauchy sequence of real
numbers and so is bounded. This contradiction completes the proof. 
Recall Cantor’s Nested Interval Theorem from real analysis: Any decreasing
sequence {In } of closed intervals in R has nonempty intersection; if the lengths
of In → 0, then the intersection is a one-point set. (This was obtained in the course
of the proof Theorem 5.1.) The underlying reason for the result to hold is the com-
pleteness of R.
Theorem 12.6 (Cantor’s intersection theorem)
If a metric space X is complete, then for any decreasing sequence {Fn } of (nonempty)
closed sets in X with diam Fn → 0 as n → ∞, we have F := ∩Fn has exactly one
point.
Proof Form a sequence {xn } by picking xn ∈ Fn . We show that this sequence is a
Cauchy sequence in X . To do this, let ε > 0 and choose n ε such that diam Fn < ε for
all n > n ε . If m, n > n ε , then Fm , Fn are subsets of Fn ε , so xm , xn ∈ Fn ε and hence
d(xm , xn ) ≤ diam Fn ε < ε, proving that {xn } is a Cauchy sequence. By completeness,
it has a limit x. For any m, xn ∈ Fm for all n ≥ m and, Fm being closed, the limit x
also belongs to Fm . Hence x ∈ F. If y ∈ F, then y ∈ Fn for all n, so d(x, y) ≤ diam
Fn → 0, showing that y = x. 
Example 12.7 The example X = R and Fn = (0, n1 ) shows that the result fails if
the Fn are not closed. The result also fails without the condition on diam Fn ; e.g.
Fn = [n, ∞) in R.
Exercise The result fails without completeness (Look in Q.)
In fact, conversely, the nested sequence property gives completeness.
Proposition 12.8 Let X be a metric space. Suppose that every nested sequence {Fn }
of nonempty closed sets with diam Fn → 0 has nonempty intersection. Then X is
complete.
Proof Suppose X has the stated property and {xn } is a Cauchy sequence in X . Let Fn
be the closure of the set of terms of the tail sequence Tn = {xk : k ≥ n} of the given
sequence. Then {Fn } is a decreasing sequence of closed sets and diam Tn → 0 as {xn }
is a Cauchy sequence and so diam Fn → 0. By hypothesis, there is an x ∈ F := ∩Fn .
Then d(xn , x) ≤ diam Fn → 0 and so xn → x. The proof is complete. 
166 12 Complete Metric Spaces

Exercise Without completeness, Cantor’s theorem holds if each Fn is closed and

compact. In fact, the result is true in any topological space. If {K n } is a decreasing
sequence of nonempty closed, compact sets in a topological space X , then ∩K n is
nonempty.
Discussion. Suppose { f n } is a sequence in C[0, 1] which converges uniformly
to f 0 on [0,1]. Then f 0 is continuous, of course. Let ε > 0. Each f n , n ≥ 0, is uni-
formly continuous and so there a δn > 0 such that | f n (x) − f n (y)| < ε/3 whenever
|x − y| < δn . Since f n → f 0 uniformly, there is a positive integer N such that
| f n (x) − f (x)| < ε/3 for n > N and for all x ∈ [0, 1]. Now, from the inequality
| f n (x) − f n (y)| ≤ | f n (x) − f 0 (x)| + | f 0 (x) − f 0 (y)| + | f 0 (y) − f n (y)|
and using an ε/3-argument, we see that | f n (x) − f n (y)| < ε whenever |x − y| < δ0
and for all n > N . Take δ = min{δ0 , δ1 , . . . , δ N }. Then it is easy to see that
| f n (x) − f n (y)| < ε for |x − y| < δ and for all n. In other words, given ε, we are
able to get a δ not only independent of the points x ∈ [0, 1] but also independent of
n. This leads us to the following definition.

Definition 12.9 A family F of maps between metric spaces (X, d) and (X , d ) is

said to be equicontinuous if for any ε > 0, there is a δ > 0 such that
x, y ∈ X, d(x, y) < δ implies d ( f (x), f (y)) < ε for all f ∈ F . Note that each
member of such a family is uniformly continuous.

Our discussion shows that any uniformly convergent sequence of continuous func-
tions on [0,1] yields an equicontinuous family. The next result, important in many
applications, is in the reverse direction. It is a simple modern version of results first
proved by Ascoli [6] in 1883 and Arzelà [1] in 1895. It is used in many parts of anal-
ysis, e.g. normal families in complex analysis. Our formulation is slightly different
from the usual one.
Theorem 12.10 (Ascoli–Arzelà theorem)
Let X be a separable metric space and let Y be a complete metric space having
the Bolzano–Weierstrass property. If F is an equicontinuous, point-wise bounded
family (i.e. the set { f (x) : f ∈ F } is bounded in Y for each x ∈ X ) of maps from
X into Y, then every sequence in F has a subsequence converging uniformly on
compact subsets of X .

Proof Let {xn } be a countable dense set in X and let { f n } be a sequence in F . Then
{ f n (x1 )} is a bounded sequence in Y and so, by the assumption on Y , there is a
subsequence { f n1 (x1 )} which converges in Y . Next consider the bounded sequence
{ f n1 (x2 )} and get a convergent subsequence { f n2 (x2 )}. Note that both { f n2 (x1 )} and
{ f n2 (x2 )} converge. Proceed inductively to get subsequences { f nk } with { f nk (x j )}
convergent for each 1 ≤ j ≤ k such that each of them is a subsequence of the pre-
ceding one. The ‘diagonal’ subsequence {gn = f nn } has the property that {gn (xk )}n
converges for all k. (Cantor’s diagonal argument again!)
Let K ⊂ X be compact. Now let us invoke the equicontinuity. Let ε > 0 and
choose δ > 0 such that d X (x, y) < δ implies dY ( f n (x), f n (y)) < ε/3 for all n.
Compactness of K ensures that it can be covered by a finite number of balls
12.1 Completeness and Ascoli–Arzelà 167

B1 , . . . , Bm of radii δ/2. Since {xn } is dense, there is an i j with xi j ∈ B j for

1 ≤ j ≤ m. Then {gn (xi j )}n is convergent for all j and so there is an n ε such that
dY (gn (xi j ), gm (xi j )) < ε/3 for all j provided m, n > n ε . If x ∈ K , then x ∈ B j for
some j and so d X (x, xi j ) < δ. We have
dY (gn (x), gm (x))
≤ dY (gn (x), gn (xi j )) + dY (gn (xi j ), gm (xi j )) + dY (gm (xi j ), gm (x))
≤ 2ε/3 + dY (gn (xi j ), gm (xi j )) < ε if m, n > n ε .
Hence {gn (x)} is a Cauchy sequence in Y and so has a limit, say g(x). In the last
inequality, keeping n > n ε fixed and letting m → ∞, we get dY (gn (x), g(x)) ≤ ε
for n > n ε for all x ∈ K . Thus the subsequence { f nn } is uniformly convergent on K
and the proof is complete. 
Corollary 12.11 Let X be a compact metric space and let F be an equicontinuous,
point-wise bounded family of Rn -valued functions on X . Then every sequence in F
has a uniformly convergent subsequence.
Proof A compact metric space is separable, and Theorem (12.10) applies with Y =
Rn . 
Corollary 12.12 Let X be a compact metric space and C(X, Rn ) be the space
of continuous Rn -valued functions on X with the sup metric: dsup ( f, g) = supx∈X

f (x) − g(x)
. Then any equicontinuous, point-wise bounded family F is precom-
pact in C(X, Rn ).
Proof If F is equicontinuous, point-wise bounded, then so is its closure F¯ in
C(X, Rn ). It is sequentially compact by the theorem. 

12.2 Bourbaki, Baire and Banach

Baire’s theorem for complete metric spaces is an important tool to obtain some basic
theorems of Functional Analysis and has other applications as well. Banach’s fixed
point theorem is used in differential equations and other parts of analysis. We take
a rather unusual route to Baire’s theorem via the following result of Bourbaki [2].
Theorem 12.13 (Bourbaki’s Mittag-Leffler Theorem)
Let (X n , dn ) be a complete metric
 space for each n and f n : X n+1 → X n be con-
tinuous maps. Let X  = {(xn ) ∈ n X n : xn = f n (xn+1 )}. If the range f n (X n+1 ) is
dense in X n for each n, then p1 (X  ) is dense in X 1 , where p1 is the projection on
X 1.
Proof First let us define metrics d̃n on the X n inductively as follows. Let d̃1 =
d1 and d̃n+1 (x, y) = dn+1 (x, y) + d̃n ( f n (x), f n (y)) for x, y ∈ X n+1 , n ≥ 1. Observe
that dn+1 (x, y) ≤ d̃n+1 (x, y) and d̃n ( f n (x), f n (y)) ≤ d̃n+1 (x, y). Moreover,
i) d̃n gives the same topology as dn because dn (xk , x) → 0 if and only if
d̃n (xk , x) → 0;
168 12 Complete Metric Spaces

ii) d̃n is complete; for if {xk } is d̃n -Cauchy, then it is dn -Cauchy, hence by dn -
completeness, dn (xk , x) → 0 for some x ∈ X n and so d̃n (xk , x) → 0.
Let x1 ∈ X 1 and let ε > 0. Since f 1 (X 2 ) is dense in X 1 and f 2 (X 3 ) is dense in X 2
there are x2 ∈ X 2 such that d̃1 (x1 , f 1 (x2 )) < ε/2 and x3 ∈ X 3 with d̃2 (x2 , f 2 (x3 )) <
ε/22 . Then d̃1 ( f 1 (x2 ), f 1 ( f 2 (x3 )) < ε/22 as well. Inductively, choose {xn } with the
following properties:
a) dn (xn , f n (xn+1 )) < ε/2n and
b) d̃k ( f k,n−1 (xn ), f kn (xn+1 )) < ε/2n , n ≥ 2, k < n, where f kn is the composition
fk ◦ · · · ◦ fn .
For each k, { f k,n−1 (xn )}n≥k is a Cauchy sequence in X k and so has a limit ξk . Since
f k,n−1 (xn ) = f k ( f k+1,n−1 (xn )). continuity of f k yields ξk = f k (ξk+1 ), so (ξk ) ∈ X  .
We complete the proof by showing that d̃1 (x1 , ξ1 ) ≤ ε by the following estimates:

d̃1 (x1 , ξ1 ) = lim d̃1 (x1 , f 1,n−1 (xn ))

≤  d̃n ( f 1,n−1 (xn ), f 1n (xn+1 ))
≤ ε/2n = ε.


A system (X n , f n ) as in the theorem is called a projective system or an inverse
system, and X  is called the projective limit or the inverse limit of the system. The
theorem (in a much more general form) is given in Bourbaki’s tome [2]. Bourbaki
calls it Mittag-Leffler’s theorem and derives the classical Mittag-Leffler’s theorem on
meromorphic functions from it. Our presentation follows those in Esterle [7] (where
it is used in the theory of Banach algebras) and Runde [8].
Corollary 12.14 (Baire’s theorem, [3]) 
If {Un } is a sequence of open, dense subsets in a complete metric space X , the n Un
is also dense in X .
Proof Let U, V be open dense sets. Then for any nonempty open set W , V ∩ W
is nonempty as V is dense. It is also open and denseness of U implies W ∩ U ∩ V
is nonempty. Thus U ∩ V intersects each nonempty open set and hence is dense.
Induction gives the result that any finite intersection ∩n1 Uk of open dense sets is
dense.
Consider now the general case. Let {Un } be a sequence of open, dense subsets in
X andlet Vn = ∩1 Uk . Then {Vn } is a decreasing sequence of dense ∞
n
open sets and
U = Vn = Un . We apply the theorem to the sequence {(X n , dn )}0 of complete
spaces, where (X 0 , d0 ) = (X, d) and (X n , dn ) = (Vn , dVn ), n > 0, with dVn as in
12.5, and with the maps f n as the inclusion maps ιn : X n → X n−1 . Clearly the
condition of the theorem is satisfied,

X  = {(xn ) ∈ X n : xn = xn−1 , n ≥1} and
p0 (X  ) = {x0 ∈ X 0 : x0 ∈ Vn ∀n} = Vn . 
The proof is complete.
12.2 Bourbaki, Baire and Banach 169

Corollary 12.15 (Baire category theorem)

space. If {Fn } is a countable collection of closed subsets
Let X be a complete metric 
with empty interiors, then Fn also has empty interior. In particular, X is not a
countable union of closed sets without interiors.
Note that Fn is closed and has empty interior means Un = Fnc is open and
Proof 
dense; Un is dense means ∪Fn has empty interior. 
Baire’s theorem can be proved directly, without having to invoke Mittag-Leffler
theorem, as we did for locally compact Hausdorff spaces.
Exercise Look at the proof of (11.9) and give a proof of Baire’s theorem for
complete metric spaces along the same lines, using Cantor’s intersection theorem in
place finite intersection property used there.
A set whose closure has empty interior is called a nowhere dense set. A set of first
category is a countable union of nowhere dense sets. A set is of second category if
it is not of first category. The theorem says that a complete metric space is of second
category. Hence the term ‘category’ in the theorem, (a term used by Baire himself.)
This result is of importance in functional analysis. It is also used for proving the
existence of continuous, nowhere differentiable functions, without constructing such
a function (and for other similar purposes). See the exercises.
Completeness is a nice property and especially important in analysis. Here is a
simple fixed point theorem [4] that is useful in analysis. An f as in the theorem
is called a contraction. Note that the fixed point x0 obtained is independent of the
choice of the initial point x.
Theorem 12.16 (Banach contraction mapping theorem)
Let (X, d) be a complete metric space. Suppose f : X → X is a map satisfying the
condition d( f (x), f (y)) ≤ c.d(x, y) for some 0 < c < 1 and all x, y ∈ X . Then f
has a unique fixed point.
Proof Take any x ∈ X and consider the sequence {x, f (x), . . . , f n (x), . . .}, the orbit
of x under f . Here f n = f ◦ f n−1 is the n-fold composition of f with itself. We have
d( f (x), f 2 (x)) ≤ cd(x, f (x)) and inductively d( f n (x), f n+1 (x)) ≤ cn d(x, f (x))
for all n. It follows that
d( f n (x), f m (x)) ≤ d( f n (x), f n+1 (x)) + · · · + d( f m−1 (x), f m (x))
≤ (cn + · · · + cm−1 )d(x, f (x)) for m > n
cn
≤ 1−c d(x, f (x)) for m > n.
It follows that { f n (x)} is a Cauchy sequence and hence has a limit, x0 = lim f n (x) ∈
X. Observe that the hypothesis gives (uniform) continuity of f and so applying f
on both sides we get f (x0 ) = lim f n+1 (x) = x0 .
If f (y0 ) = y0 , then d(x0 , y0 ) = d( f (x0 ), f (y0 )) ≤ cd(x0 , y0 ) and so y0 = x0 and
uniqueness follows. 
Example 12.17 Let f (x) = 21 (x + x1 ), x ∈ [1, ∞). Note that X = [1, ∞) is com-
plete and f maps X to itself. It is easy to check that f is a contraction with c = 1/2
and 1 is the fixed point of f .
170 12 Complete Metric Spaces

12.3 Completion

We now show that an incomplete metric space can always be ‘completed’.

Definition 12.18 A complete metric space ( X̄ , d̄) is called a completion of a metric
space (X, d) if there is an isometric embedding of X in X̄ as a dense subspace; i.e.
there is an isometry i : X → X̄ such that the image i(X ) is dense in X̄ .
Every metric space has a (unique!) completion—mimick Cantor’s construction of
real numbers as equivalence classes of Cauchy sequences of rational numbers. The
construction was given by Hausdorff [9].
Let (X, d) be a metric space that is not complete. Let X be the set of all Cauchy
sequences in X . The first step is to observe that, for (xn ), (yn ) ∈ X, lim d(xn ), yn )
exists. To see this, note that
d(xn , yn ) ≤ d(xn , xm ) + d(xm , ym ) + d(ym , yn ),
| d(xn , yn ) − d(xm , ym ) |≤ d(xn , xm ) + d(ym , yn ) → 0 as m, n → ∞.
This shows that {d(xn , yn )} is a Cauchy sequence of real numbers and so has a limit.
(Completeness of real numbers!).
Define a relation R on X by declaring, for x = (xn ), y = (yn ) ∈ X, x Ry if
limn d(xn , yn ) = 0. It is easy to see that R is an equivalence relation. Let X̄ = X/R
be the quotient set of all equivalence classes. Let x̄ = [(xn )] be the equivalence class
of (xn ) ∈ X. For x̄, ȳ ∈ X̄ , let d̄(x̄, ȳ) = lim d(xn , yn ). If (xn )R(an ) and (yn )R(bn ),
then
| d(xn , yn ) − d(an , bn ) |≤ d(xn , an ) + d(bn , yn ) → 0,
showing that lim d(xn , yn ) = lim d(an , bn ). This means that d̄ is well-defined. We
will show that it is a metric on X̄ . It is clear that d̄ ≥ 0 and d̄(x̄, ȳ) = d̄( ȳ, x̄).
If d̄(x̄, ȳ) = 0, then lim d(xn , yn ) = 0 and so (xn )R(yn ), hence x̄ = ȳ ∈ X̄ . With
obvious notation, we also have
d̄(x̄, ȳ) = lim d(xn , yn ) ≤ lim d(xn , z n ) + lim d(z n , yn ) = d̄(x̄, z̄) + d̄(z̄, ȳ),
yielding the triangle inequality. Thus d̄ is a metric.
Next we embed X isometrically in X̄ . To do this, for any x in X , let x denote the
constant sequence (x, x, . . .) in X. Then d̄([x], [y]) = d(x, y), so that ϕ(x) = [x]
defines an isometric embedding X into X̄ .
To prove that ϕ(X ) is dense in X̄ , we show that every ball Bd̄ (x̄, δ) in X̄ contains
an element of ϕ(X ). Let {xn } ∈ x̄ and choose n δ such that d(xm , xn ) < δ/2 for all
m, n ≥ n δ . Write xδ for xn δ and consider the constant sequence xδ = (xδ , xδ , . . .) =
ϕ(xδ ) ∈ ϕ(X ). Now estimate d̄(x̄, xδ ) = limn d(xn , xn δ ) ≤ δ/2 < δ, so xδ ∈ Bd̄ (x̄, δ).
To show that ( X̄ , d̄) is complete, let {x̄n } be a Cauchy sequence in X̄ . First consider
the case when x̄n ∈ ϕ(X ), say x̄n = [xn ] for all n. Then d(xn , xm ) = d̄([xn ], [xm ]) =
d̄(x̄n , x̄m ), so {xn } is a Cauchy sequence in X and x̄ = [(xn )] ∈ X̄ . Note that
d̄(x̄n , x̄) = d̄([xn ], x̄) = limm d(xn , xm ),
limn d̄(x̄n , x̄) = limn limm d(xn , xm ) = 0,
since {xn } is a Cauchy sequence. Hence {x̄n } converges to x̄ in X̄ .
12.3 Completion 171

Finally, let {x̄n } be any Cauchy sequence in X̄ . As ϕ(X ) is dense in X̄ , there is

[xn ] ∈ ϕ(X ) with d̄(x̄n , [xn ]) < n1 for each n. Then
d̄([xn ], [xm ]) ≤ d̄([xn ], x̄n ) + d̄(x̄n , x̄m ) + d̄(x̄m , [xm ]) < n1 + m1 + d̄(x̄n , x̄m ).
Thus {[xn ]} is a Cauchy sequence. By the special case above, x̄ = lim[xn ] exists
in X̄ . But then
d̄(x̄n , x̄) ≤ d̄(x̄n , [xn ]) + d̄([xn ], x̄) < n1 + d̄([xn ], x̄) → 0,
so that {x̄n } converges to x̄ in X̄ , yielding the following theorem.
Theorem 12.19 Let (X, d) be a metric space which is not complete. There is an
isometric embedding ϕ of X into a complete metric space X̄ such that ϕ(X ) is dense
in X̄ . 
We give a ‘soft’ approach to get a completion in the exercises. It is good to see
the ‘hard’ construction once.
Exercise What does the construction give when X is complete?
Thus, every metric space has a completion. Completion of a metric space is unique
‘up to isometries’, as we now show. Thus if we can obtain some completion, then it
is the completion.
But before taking this up, let us take a look at the following extension question: if
f is continuous on A does it extend to Ā? Not always! There is the familiar example
of 1/x on (0,1). This function is not uniformly continuous and that is the obstruction.

Proposition 12.20 Let (X, d X ), (Y, dY ) be metric spaces with Y complete. If A ⊂ X

then any uniformly continuous f : A → Y has a unique continuous extension f¯ :
Ā → Y . If f is an isometry, so is f¯.

Proof Uniqueness is clear since g, h : Ā → Y are continuous and g = h on A imply

g = h on Ā.
To define an extension, let x ∈ Ā and let {xn } be a sequence in A with x = lim xn .
Uniform continuity of f ensures that { f (xn )} is a Cauchy sequence in Y and, by
completeness, y = lim f (xn ) exists. We want to define f¯(x) = y, but before we
can do that we have to be sure that y is independent of the choice of xn → x. Let
us check this. Suppose {xn } also converges to x in X and y = lim f (xn ). Given
ε > 0, by uniform continuity, there is a δ > 0 such that dY ( f (x1 ), f (x2 )) < ε when
d X (x1 , x2 ) < δ. Both {xn }, {xn } converge to x, so d X (xn , xn ) < δ for large n in which
case dY ( f (xn ), f (xn )) < ε. But
dY (y, y ) ≤ dY (y, f (xn )) + dY ( f (xn ), f (xn ) + dY ( f (xn ), y ) < 3ε
if n is large enough. This forces y = y.
Thus we can legitimately define f¯(x) = y if x = lim xn and y = lim f (xn ).
We show that f¯ is uniformly continuous. Take x = lim xn , x = lim xn and y =
lim f (xn ), y = lim f (xn ). Then
dY (y, y ) ≤ dY (y, f (xn )) + dY ( f (xn ), f (xn )) + dY ( f (xn ), y )
The first and the last terms are small if n is large. If x, x are close, then f (xn ), f (xn )
are close for large n and so the middle term is small. Conclusion: if x, x are close
172 12 Complete Metric Spaces

i
X X̄

j σ

X

Fig. 12.1 Uniqueness of completion

enough, then y = f¯(x), y = f¯(x ) are close and thus f¯ is uniformly continuous.
The reader can make these precise with ε, δ etc.
Finally, suppose f is an isometry. Then, with notation as before,
dY ( f¯(x), f¯(x )) = lim dY ( f (xn ), f (xn )) = lim d X (xn , xn ) = d X (x, x ). 
Corollary 12.21 A uniformly continuous map f on a dense subspace of a metric
space X into a complete metric space has a unique continuous extension to X and
the extension is an isometry if f is.
Now it is easy to get the uniqueness of completion.
Theorem 12.22 Let (X, d) be a metric space. If i : (X, d) → ( X̄ , d̄) and j : (X, d)
→ (X  , d  ) are both completions of (X, d), then there is a surjective isometry σ :
( X̄ , d̄) → (X  , d  ) such that σ ◦ i = j.
Proof Define σ on i(X ) by σ (i(x)) = j(x), x ∈ X, Fig. 12.1. This is well-defined as
x = y if i(x) = i(y) and is clearly an isometry. Now we extend σ to an isometry on
X̄ by the previous corollary. The image σ ( X̄ ) is complete, being isometric to the
complete space X̄ , and hence is closed. On the other hand, it contains the dense set
j(X ), so itself is dense. Conclusion: σ ( X̄ ) = X  and σ is surjective. 
The completion of Q with the standard metric is R. The completion of Q with
the p-adic metric is denoted by Q p . It is locally compact and totally disconnected.
It also carries the structure of a field, called the p-adic number field. It is of central
importance in number theory.
We conclude by looking at an important space of continuous functions.
Definition 12.23 Let X be a locally compact Hausdorff space. A continuous (com-
plex valued) function f on X is said to vanish at infinity if, for any ε > 0, there is a
compact set K in X such that | f (x) |< ε for all x ∈ K c . Let C0 (X ) denote the space
of all continuous functions on X vanishing at infinity and let Cc (X ) be the subspace
of continuous functions with compact support.
Theorem 12.24 Let X be a locally compact Hausdorff space. Then
1) every function in C0 (X ) is bounded;
2) C0 (X ) is complete in the sup metric dsup ;
3) Cc (X ) is dense in C0 (X ).
12.3 Completion 173

Proof 1) Let f ∈ C0 (X ). If K is a compact set with | f (x)| < 1 for x ∈ K c and

M = maxx∈K | f (x)|, then | f | ≤ max(1, M) on X .
2) Let { f n } be a dsup -Cauchy sequence in C0 (X ). For ε > 0, choose n ε such that
| f n (x) − f m (x)| < ε for m, n ≥ n ε , and for all x ∈ X .(*)
Thus { f n (x)} is a Cauchy sequence of scalars for each x ∈ X and so f (x) = lim f n (x)
exists, yielding a function f on X . To show that f ∈ C0 (X ). In (*), keep n ≥ n ε fixed
and let m → ∞ to get | f n (x) − f (x)| ≤ ε for n ≥ n ε and for all x ∈ X . Thus f n → f
uniformly on X . To check continuity of f , take x0 ∈ X and choose, by continuity, a
neighbourhood U of x0 such that | f n ε (x) − f n ε (x0 )| < ε for x ∈ U . Then, continuity
of f at x0 is got from
| f (x) − f (x0 )| ≤ | f (x) − f n ε (x) + | f n ε (x) − f n ε (x0 )| + | f n ε (x0 ) − f (x0 )|
< 3ε,
for x ∈ U . To see that f vanishes at infinity, write
| f (x)| ≤ | f (x) − f n ε (x)| + | f n ε (x)| < ε + | f n ε (x)| for all x.
So, if K is a compact set such that | f n ε (x)| < ε for x ∈ K c , then | f (x)| < 2ε for
x ∈ K c completing the proof that f ∈ C0 (X ).
3) We use Urysohn’s lemma. Let f ∈ C0 (X ) and, given ε > 0, choose a compact
set C such that | f (x)| < ε for x ∈ / C. There is a precompact open set V such that
C ⊂ V . By Urysohn’s lemma there is a continuous function g on X such that 0 ≤
g ≤ 1, g = 1 on C and supp g ⊂ V . If h = f g, then h = f on C, h has compact
support in V and | f (x) − h(x)| ≤ 2| f (x)| < 2ε off C. We thus get that h ∈ Cc (X )
and | f (x) − h(x)| < 2ε for all x ∈ X . Thus Cc (X ) is dense in C0 (X ). 
Corollary 12.25 C(X ) is complete with the sup metric for any compact Hausdorff
space X .
Exercise A continuous function f on R vanishes at infinity if and only if lim|x|→∞
f (x) = 0.
Exercise 12
1. Cauchy sequences. If {xn } is a Cauchy sequence in a metric space, there is
a subsequence {xn k } such that d(xn k , xn k+1 ) < ∞.
2. Completion. a) Let σ : Sn \ {(0, 0, 1)} → Rn be the stereographic projection.
Define ρ(x, y) = d(σ −1 x, σ −1 y) for x, y ∈ Rn and d is the Euclidean metric in
Rn+1 . Then ρ is a metric, is totally bounded, not complete. What is the completion?
Is it equivalent to the usual metric?
b) What is the completion of C[−1, 1] with the metric d1 ?
3. Space of maps. Recall the metric space B(X, Y ) of bounded maps from a set
X to a metric space Y with sup metric. It is complete if and only if Y is complete. If
X is also a metric space, the space Cb (X, Y ) of bounded continuous maps is a closed
subspace of B(X, Y ).
4. Products. a) The completion of the product of two metric spaces is the com-
pletion of the product. Obtain a precise statement. 
b) If (X n , dn ) is a sequence of complete metric spaces, then X n is complete in
the metric d (see 2.3).
174 12 Complete Metric Spaces

5. Completeness a) Let (X, d) be a metric space, P ∈ X and let d P be the

associated French railway metric, 2.3. Then (X, d P ) is complete.
b) Find an incomplete metric that is equivalent to a complete metric. Thus com-
pleteness is not a topological invariant.
c) Metrics d1 , d2 on X are said to be uniformly equivalent if the identity ι X :
(X, d1 ) → (X, d2 ) and its inverse are both uniformly continuous. If one of them is
complete, so is the other.
d) If X is complete, then so is the space Bc (X ) of bounded closed subsets with
the Hausdorff metric d H .
6. Compactness. a) A compact metric space is complete.
b) If a complete metric space (X, d) is complete in every equivalent metric, then
it is compact. (See [10], 4.3.E(d))
c) A metric space is totally bounded if and only if every sequence has a Cauchy
subsequence.
d) If X is totally bounded, then so is Bc (X ) with the Hausdorff metric d H . Thus
Bc (X ) is compact if X is.
7. Countable complete metric spaces. If X is a complete metric space and X
is countable, the set of isolated points in X is dense. Give a nondiscrete example.
8. Topologies on spaces of maps. This exercise concerns different topologies
on the space Y X of all maps from a topological space X to a topological space Y .
C(X, Y ) is the subset of continuous maps. (References: [11–13].)
a) First, there is the topology of point-wise convergence. This is just the product
topology on Y X , in which convergence is point-wise convergence. Write down a
subbase for this topology.
b) Let B( f, ε) = {g : g(x) ∈ B( f (x), ε)∀x ∈ X } for f : X → Y and ε > 0 where
Y is a metric space. These sets form the basis for a topology on Y X in which conver-
gence is uniform convergence on X . The topology is called the topology of uniform
convergence.
c) Let B(K , f, ε) = {g : g(x) ∈ B( f (x), ε)∀x ∈ K } where Y is a metric space,
f : X → Y , K is a compact subset of X and ε > 0. Sets of this type form the basis
for a topology on Y X in which convergence is uniform convergence on compact
subsets of X . The topology is called the topology of local uniform convergence.
d) Let X be a locally compact Hausdorff space and f : X → Y. If f | K is contin-
uous for each compact subset K of X , then f is continuous.
e) C(X, Y ) is closed in Y X in the topology of local uniform convergence if X is
locally compact and Hausdorff.
f) The compact-open topology on Y X has sets of the form S(K , V ) :=
{ f : f (K ) ⊂ V } as a subbase, where K ⊂ X is compact and V ⊂ Y is open. Here
Y is any topological space.
g) When Y is a metric space, the compact-open topology equals the topology of
local uniform convergence on C(X, Y ).
h) If X is locally compact Hausdorff and C(X, Y ) is given the compact-open
topology, then the evaluation map e : X × C(X, Y ) → Y , defined by e(x, f ) = f (x)
is continuous.
12.3 Completion 175

i) For x ∈ R, define χx (t) = eı xt and let R = {χx : x ∈ R}. Then x → χx is a

bijection R ↔  R and is a homeomorphism if  R ⊂ C(R, T) is given the compact-
open topology.
9. Equicontinuity. a) Let X be a metric space and let F be a uniformly bounded,
equicontinuous family of real functions on X . Then the function defined by g(x) =
sup{ f (x) : f ∈ F } is continuous.
b) Suppose X is compact and { f n } is a sequence with f n ∈ F for all n. Define
gn = max1≤k≤n f k . Then {gn } is equicontinuous and converges uniformly.
c) Any uniformly convergent sequence of continuous functions constitute an
equicontinuous family.
d) Find conditions for a family in C0 (X ) to be precompact, where X is locally
compact and Hausdorff.
10. Fixed points. a) Let X be a complete metric space. If f : X → X is such
that f 2 = f ◦ f satisfies the condition of Banach’s theorem, then f has a unique
fixed point. Can f 2 be replaced by f n for some n?
b) If X is compact and the condition d( f (x), f (y)) < d(x, y) holds for all x = y,
the same conclusion holds. The result does not hold if we assume only completeness
of X in place of compactness.
c) Illustrate (a) and (b) with examples.
d) Let X be a complete metric space and consider an open ball B = B(x0 , r ). Sup-
pose f : B → X be a contraction with constant c < 1. If d(x0 , f (x0 )) <
(1 − c)r , then f has a fixed point.
11. A soft approach to completion. Let B(X, R) be the space of bounded real
functions on X and let dsup ( f, g) = supx∈X | f (x) − g(x)|.
a) dsup is a complete metric on B(X, R).
b) Fix x0 ∈ X and define f p (x) = d(x, p) − d(x, x0 ), x ∈ X, for each p ∈ X .
f p ∈ B(X, R), dsup ( f p , f q ) = d( p, q) and p → f p is an isometry ι of X into
B(X, R).
c) The closure X̄ of ι(X ) in B(X, R) is a completion of X .
12. Continuous, nowhere differentiable functions. It is easy to construct func-
tions on the real line which are continuous, but are not differentiable at any finite set
of points. Weierstrass (in 1872, [14]) shocked the mathematical world of his times
by constructing a continuous function that is nowhere differentiable. (It appears
that Bolzano (1831) (see [15–18]) and a Swiss mathematician Charles Cellérier
(1860) [19] had constructed such functions earlier, but they were published only
much later.) Baire’s theorem can be used to show (e.g. [20]) that ‘most’ of the
functions in C[0, 1] are nowhere differentiable! More precisely, the set of func-
tions in C[0, 1] which are nowhere differentiable is dense. This can be seen as fol-
lows. Write h f (x) = | f (x + h) − f (x)|/ h for h > 0 and x, x + h ∈ [0, 1]. Let
Un , n > 1, be the set of those f such that h f (x) > n for all 0 ≤ x ≤ 1 − 1/n for
some 0 < h ≤ 1/n. Then Un is open, dense and every f ∈ U = Un is nowhere
differentiable. Baire’s theorem ensures that U is dense.
176 12 Complete Metric Spaces

14. Complementary dense sets. In R, the rationals Q and the complementary

set Qc of irrationals are both dense. Find other examples of spaces where there are
complementary dense sets.
15. Vanishing at infinity. For a locally compact Hausdorff space X , {g ∈ C(X ∞ ) :
g(∞) = 0} can be identified isometrically with C0 (X ) via the map g → g | X . (Thus,
‘vanishing at infinity’ holds literally!)
16. Polish spaces. A topological space X is called a Polish space if it is homeo-
morphic to a complete, second countable metric space. These spaces were extensively
studied by the Polish school (e.g. Sierpiński. Kuratowski ), hence the name. These
are very much used in descriptive set theory, measure and probability theory, etc.
Consult the books of Bourbaki [2, 21] and Kuratowski [22] for more on these spaces.

a) R, Cantor set, [0,1], C[0, 1], 1 , 2 , NN are all Polish spaces.

b) (0,1),[0, ∞), [0,1) are Polish; a closed subset of a Polish space, a compact
metric space, a countable discrete space are Polish.
c) A countable product of Polish spaces, a closed subset and an open subset a
Polish space are Polish. (May use 12.4 for the last one.)
d) A G δ subset of a Polish space is Polish. (In particular, the space of irrational
numbers is Polish.)
e) A locally compact, σ -compact, metrisable space is Polish.
17. Banach contraction mapping. Banach’s fixed point theorem 12.16 has sev-
eral generalisations, converses and applications (to various fields: differential equa-
tions, analysis, economics, electrical engineering, etc.) Here is one. Let g : [0, 1] →
1
R and k : [0, 1] × [0, 1] → R be continuous. Assume that max0≤x≤1 0 |k(x, y)|dy
< 1. Then there is a unique continuous function f : [0, 1] → R with
1
f (x) − 0 k(x, y) f (y)dy = g(x).
18. Space-filling curves: construction. For existence of such curves, see the
exercises in Chap. 7. After Peano’s construction [23], there have been many construc-
tions of space-filling curves. One of the most commonly discussed is the one due to
Hilbert (See, e.g., [24]). It is interesting to note that while Peano’s paper contains no
pictures, most later writings on the topic give a pictorial understanding of the con-
struction. There are generalisations—for example, curves filling a cube etc. There
are also applications in several areas—fractals, for example. There is a monograph
on the topic where several constructions are detailed and further references can be
found—see Hans Sagan [25].
For the first stage, intervals [0,1/2] and [1/2,1] are mapped to the two line segments
forming a ‘tent’, Fig. 12.2a. For the next stage, each of these two line segments is
mapped to four segments as in Fig. 12.2b. Repeat this for each tent, Fig. 12.2c, d. Let
f 0 , f 1 , . . . denote the sequence of functions [0, 1] → [0, 1]2 thus obtained.
a) { f n } converges uniformly on [0,1] and so the limit function f is continuous.
b) The range of f is dense in the square.
c) f maps [0,1] onto [0, 1]2 . Can f be injective?
12.4 Biographical Notes 177

Fig. 12.2 Four stages of the construction of a square-filling curve

12.4 Biographical Notes

12.4.1 Ascoli

Giulio Ascoli (1843–1896) was an Italian mathematician who made contributions

to analysis, partial differential equations and Fourier series. The concept of equicon-
tinuity and a weak form of the Arzelà–Ascoli theorem is due to him (1884).

12.4.2 Arzelà

Cesare Arzelà (1847–1912) was also an Italian analyst. He studied in Pisa under
Enrico Betti and Ulisse Dini. In 1889 he generalised and clarified Ascoli’s result to
178 12 Complete Metric Spaces

give the Arzelà–Ascoli theorem. Leonid Tonelli was his student. (The formulation
of the theorem for functions on compact metric spaces was given by Fréchet.)

12.4.3 Baire

René-Louis Baire (1872–1934) was a French mathematician. He worked in set

theory and real analysis. He is best known for his definition of a nowhere dense set,
Baire category theorem and Baire classes of functions (successive limits, starting
with continuous functions). Baire set, Baire measure, Baire function, Baire space are
all named after him. Arnaud Denjoy was his student.

12.4.4 Banach

Stefan Banach (1892–1945) was the brightest star in the Polish mathematical firma-
ment and was a founding father of the subject of Functional Analysis in the early part
of the twentieth century. Normed spaces which are complete in the metric induced
by the norm are called Banach spaces. These were studied by Banach in his thesis
(completed in 1920, published in 1922), which laid the foundations of functional
analysis. His fixed point theorem that we have discussed also appeared in his thesis.
His 1932 monograph, Théorie des opérations linéaires, was the definitive work on
functional analysis at that time. Banach space, Banach algebra, Hahn–Banach theo-
rem, Banach–Steinhaus theorem, Banach’s closed graph theorem and open mapping
theorem, Banach–Mazur theorem, Banach–Schauder theorem, Banach–Alaoglu the-
orem are all of great importance in functional analysis. Banach–Tarski paradox,
Banach fixed point theorem, Banach–Stone theorem are also well-known. Banach’s
teacher Hugo Steinhaus ‘discovered’ him in a park and took him under his wings.
Steinhaus posed some problems on Fourier series that he was finding difficult and
Banach solved them in a week. Steinhaus used to say that his greatest mathematical
discovery was the discovery of Banach. Several talented youngsters soon gathered
around Banach leading to the golden era of Polish mathematics. Stanislaw Mazur
was a doctoral student of Banach at the University of Lwòw, and there were several
others who were inspired by Banach. During the Nazi occupation of Poland, he was
employed as a lice feeder. He died of cancer in 1945, before the war ended.

References

1. Arzela, C.: Funzioni di Linee. Atti della Reale Academia dei Lincei Rendiconti 5, 342–348
(1889)
2. Bourbaki, N.: General Topology, Chapters 1–4. Springer (1995)
References 179

3. Baire, R.: Sur les fonctions de variables réelles. Ann. di Mat. 3(3), 1–123 (1899). https://doi.
org/10.1007/BF02419243
4. Banach, S.: Sur les opérations dans les ensembles abstraits et les applications aux equations
intégrales. Fund. Math. 3, 133–181 (1922). https://doi.org/10.4064/fm-3-1-133-181
5. Fréchet, M.: Sur quelques point du calcul fonctionnel. Rendiconti di Palermo 22, 1–74 (1906).
https://doi.org/10.1007/BF03018603
6. Ascoli, G.: Le curve limit di una varietà data di curve. Mem. Accad. Lincei 18, 521–586 (1883)
7. Esterle, J.: Mittag-Leffler methods in the theory of Banach algebras and a new approach to
Michael’s problem. In: Greenleaf, F., Gulick, D. (eds.) Proceedings of the Conference on
Banach Algebras and Several Complex Variables, Contemp. Math. Am. Math. Soc. 32 (1984)
pp. 107–129 https://doi.org/10.1090/conm/032
8. Runde, V.: A Taste of Topology. Universitext, Springer (2005)
9. Hausdorff, F. Grundzüge der Mengenlehre. Leipzig (1914)
10. Engelking, R.: General Topology Revised Edition. Heldermann Verlag (1989)
11. Dugundji, J.: Topology. Prentice Hall of India (1975)
12. Kelley, J.L.: General Topology. Affiliated East-West Press (1969)
13. Willard, S.: General Topology. Addison-Wesely (1970)
14. Weierstrass, K.: Über continuirliche Functionen eines reellen Arguments, die für keinen Werth
des letzteren einen bestimmten Differentialquotienten besitzen, Königl. Akad. der Wiss.,
Berlin, Werke 2, pp. 71–74 (1872)
15. Jarnìk, V.: Bolzano and the Foundations of Mathematical Analysis. Society of Czech Mathe-
matics and Physics, Prague (1981)
16. Kowalewski, G.: Über Bolzanos Nichtdifferenzierbare Steige Funktion. Acta Math. 44, 315–
319 (1923). https://doi.org/10.1007/BF02403926
17. Russ, S.: The Mathematical Work of Bernard Bolzano. Oxford University Press (2004)
18. Rychlìk, K.: Theorie der reellen Zahlen im Bolzanos handschriftlichen. Nachlasse, Prague
(1962)
19. Cellérier, C.: Note sur les principes fondamentaux de l’analyse. Bull. des Sci. Math., 2nd series
14, 142–160 (1890)
20. Banach, S.: Über die Baire’sche Katagorie gewisser Functionenmengen. Studia Math. 3, 174–
179 (1931). https://doi.org/10.4064/sm-3-1-174-179
21. Bourbaki, N.: General Topology, Chapters 5–8. Springer (1998)
22. Kuratowski, K.: Topology. Academic Press (1966)
23. Peano, G.: Sur une Courbe qui ramplit Toute une Plane. Math. Ann. 36, 157–160 (1890). https://
doi.org/10.1007/BF01199438
24. Munkres, J.R.: Topology, 2nd edn. Pearson Education, Asia (2001)
25. Sagan, H.: Space-Filling Curves. Springer (1994)
Chapter 13
Combinatorial Methods in Euclidean
Topology

Combinatorial tools and methods have become very powerful in modern mathemat-
ics. Elementary proofs of deep topological results ate obtained here using these.
A central piece is the following higher dimensional intermediate value theorem of
Poincaré.
Soient ξ1 , . . . , ξn fonctions continues de n variables x1 , . . . , xn ; la variable xi est
assujettie à varier entre les limites +ai et −ai . Supposons que, pour xi = ai , ξi soit
constamment positif, et pour xi = ai , constamment négatif; je dis qu’il existera un
système de valeurs des x pour lequel tous les ξ ’s annuleront. (Poincaré 1883) [1, 2]
(“Let ξ1 , . . . , ξn be n continuous functions of n variables x1 , . . . , xn : the variable
xi is allowed to vary between the limits +ai and −ai . Suppose that for xi = ai , ξi is
constantly positive, and that for xi = −ai , ξi is constantly negative; I say there will
exist, a system of values of x for which all the ξ ’s vanish.")
That is the n-dimensional intermediate value theorem of Poincaré. Just as in the
one-dimensional case, this will give the higher dimensional fixed point theorem of
Brouwer.
Some of the most fundamental and hard theorems of topology concern ‘famil-
iar’ geometric objects in the usual Euclidean spaces: curves, balls, spheres and the
Euclidean spaces themselves. Jordan curve theorem in plane topology (that we will
not discuss) is an example of an ‘intuitively obvious’ basic result that is far from
easy to prove. Among such classical results, we consider Poincaré’s higher dimen-
sional intermediate value theorem, Brouwer’s theorems on fixed points and invariance
and Borsuk theorems on maps of the sphere. We present elementary combinatorial
approaches to these and some related results.
Fixed point theorems are important for applications to many areas in mathematics,
game theory, mathematical biology, economics, etc. (e.g. John Nash, Nobel Prize
winner in Economics, 1994, used Kakutani fixed point theorem in his work on game
theory and economics; incidentally, Nash also was a recipient of Abel Prize, 2015, in

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 181
K. Parthasarathy, Topology, La Matematica per il 3+2 134,
https://doi.org/10.1007/978-981-16-9484-4_13
182 13 Combinatorial Methods in Euclidean Topology

Mathematics for his work on differential topology.) The simplest of these is that any
continuous map f : [0, 1] → [0, 1] has a fixed point (see 1.7), an easy consequence
of the intermediate value theorem. Just as in the one-dimensional case, Poincaré’s
higher dimensional intermediate value theorem is equivalent to a higher dimensional
Brouwer fixed point theorem. The case n = 3 of the fixed point theorem was proved
in 1904 by the Latvian mathematician Bohl [3] (that went unnoticed), and by the
famous Dutch topologist L. E. J. Brouwer in 1909 [4]. The general case was proved
in 1910 by both Hadamard [5] and Brouwer [6]; Brouwer made systematic use
of the recently evolving methods of simplicial complexes introduced by Poincaré.
The standard text book proofs use homology theory or degree theory. The Polish
trio Knaster–Kuratowski–Mazurkiewicz gave a simple proof in 1929 [7], based on
Sperner’s 1928 combinatorial lemma [8]. For some recent developments related to
Sperner’s lemma, see [9, 10] and for those on Brouwer’s theorem, see [11].

13.1 Convex Sets and Balls

In this chapter, we are mainly concerned with cubes and balls. Topologically these
are the same, but in some situations, e.g. combinatorial methods, we need to con-
sider cubes rather than balls. On the other hand, one may prefer balls for geometric
purposes. So we first show that their topological equivalence. More generally, we
show that any compact convex set with interior is homeomorphic to the ball.
A set E in Rn is convex if the line segment joining any two points of E lies
entirely in E. An equivalent condition is that if p1 , . . . , pk are in E, then so is any
convex combination t j p j with t j ≥ 0, t j = 1. The intersection of any collection
of convex sets is convex and so, for any set E, the intersection conv(E) of all convex
sets containing E is the smallest convex set containing E. It is called the convex hull
of E and consists of all convex combinations of points of E.
Lemma 13.1 Let K be a compact convex set in Rn with nonempty interior. Then ∂ K
is homeomorphic to Sn−1 . In fact, if p is an interior point of K , then h : x → x−
x− p
p
is a homeomorphism of ∂ K onto Sn−1 .
Proof By applying the homeomorphism x → x − p, it suffices to consider the case
p = 0; this will give us notational simplification. So assume that 0 ∈ int K and
define h by h(x) = x x
. This is a continuous map Rn \ {0} → Sn−1 . We show that
its restriction to ∂ K is bijective.
To show that h is onto, let y ∈ Sn−1 and τ = sup{t ≥ 0 : t y ∈ K }. This is finite
since K is bounded. τ y ∈ K as K is closed. Since 0 is an interior point of K , some
ball B(0, r ) ⊂ K , and so t y ∈ K for any 0 ≤ t < r , proving τ > 0. By definition,
for every ε > 0, there is a t > τ − ε with t y ∈ / K . This shows that every ball around
τ y contains points not in K . Thus τ y ∈ ∂ K , and clearly h(τ y) = y. This proves that
h : ∂ K → Sn−1 is surjective. See Fig. 13.1.
To prove injectivity, note that the ray joining 0 to y meets the boundary of K at τ y.
Injectivity of h just means that such a point is unique. A ray R from 0 is of the form
13.1 Convex Sets and Balls 183

{t y0 : t ≥ 0}, y0  = 1, which is a closed set and so R ∩ K is a compact, convex set,

hence is a line segment [0, t0 y0 ] for some t0 ≥ 0. If t0 = 0 then R does not meet ∂ K
since 0 is an interior point of K . If t0 > 0, we show that sy0 ∈ int K for 0 < s < t0 .
Choose 0 < s < r ≤ t0 and take a small closed ball B in int K with radius < r . The
cone over B with vertex r y0 is contained in K , by convexity and contains sy0 as an
interior point. Thus sy0 ∈ / ∂ K , so that R can not meet ∂ K in more than one point.
We have shown that h is a continuous bijection between the compact Hausdorff
spaces ∂ K and Sn−1 and hence is a homeomorphism. 

Theorem 13.2 Let K be a compact, convex set in Rn with nonempty interior.

1) For any p ∈ int K , there is a homeomorphism ĥ : Bn → K such that ĥ(Sn−1 ) =
∂ K (i.e. ĥ maps the boundary of Bn to the boundary of K ) and ĥ(0) = p.
2) If K is symmetric (i.e. −K = K ), then ĥ can be chosen to be symmetric, i.e.
ĥ(−x) = −ĥ(x).

Proof 1) We assume, without loss of generality, that 0 ∈ int K , as in the previ-

ous lemma. Let h : ∂ K → Sn−1 be the homeomorphism obtained there. For x ∈ Bn ,
define ĥ(0) = 0 and ĥ(x) = x.h −1 (x/x) for x
= 0. What does ĥ do geometri-
cally? See Fig. 13.1.
The map ĥ is bijective and is clearly continuous at any x
= 0. To check continuity
at x = 0, note that boundedness of K yields an M > 0 such that x ≤ M for all
x ∈ K , and in particular, h −1 (y) ≤ M for all y ∈ Sn−1 . Thus ĥ(x) ≤ Mx for
all x ∈ Bn . This gives the continuity at x = 0. Being a continuous bijection between
compact Hausdorff spaces, ĥ is a homeomorphism.
2) Since K is convex, the symmetry of K implies 0 ∈ K . If p is an interior point
of K , so is − p; in fact, if B ⊂ K is a ball around p, then −B ⊂ −K = K is a ball
around − p. The union of all line segments with one end in B and the other in −B
(i.e. the cylinder with faces B, −B) is a subset of K and contains 0 in its interior.
Thus 0 is an interior point of K . Hence h is symmetric and so is ĥ. 

Corollary 13.3 Let . be any norm on Rn and let Bn and Sn−1 be the corresponding
closed unit ball and the unit sphere, respectively. Then there is a homeomorphism of
Bn onto Bn mapping Sn−1 onto Sn−1 and which preserves antipodal points.

τy = x
t0 y 0
sy0 y=h(x)
ry0 y0 B
K
0 Sn−1

Fig. 13.1 Convex sets with interior are homeomorphic to the ball
184 13 Combinatorial Methods in Euclidean Topology

We are interested in the norms x1 = |x j | and x∞ = max j |x j |. For .1 , the
closed unit ball n = {x ∈ Rn : |x j | ≤ 1} is the convex hull of ±e1 , . . . , ±en . Its
boundary sphere is denoted by ♦n−1 . ♦0 = {±1} and ♦1 is the square with vertices
{(±1, 0), (0, ±1)} in R2 (the shape of our symbol ♦). ♦2 is the octahedron with
equatorial face ♦1 and is made up of eight triangular faces got by joining ±e3 to the
four vertices ±e1 , ±e2 of ♦1 in the (x, y)-plane of R3 . It is got by pasting together the
bases of two pyramids. It will also be convenient to consider ♦n+ = {x ∈ ♦n : xn+1 ≥
0} and ♦n− = {x ∈ ♦n : xn+1 ≤ 0}, the ‘upper’ and ‘lower’ hemispheres. For k < n,
we identify the k-sphere as the subset {x ∈ ♦n : x j = 0, j > k + 1} of the n-sphere.
Note that the (n − 1)-sphere is the ‘equator’ of the n-sphere: ♦n−1 = ♦n+ ∩ ♦n− .
In .∞ , we write n for the ball and n−1 for the sphere. n is the convex hull
of ±e1 ± · · · ± en and is the product [−1, 1]n see Fig. 2.5.
Actually, we will not need the general result 13.2 and we will only need the
previous corollary for .1 and .∞ . These are simple to obtain directly. Recall that
Bn and Sn−1 are the ball and sphere in the Euclidean norm given by x22 = |x j |2 .
A map h on a ball is antipode preserving if h(−x) = −h(x). (±x are said to be
antipodes.)

Proposition 13.4 There are antipode preserving homeomorphisms

1) h ♦ : Bn → n with h ♦ (Sn−1 ) = ♦n−1 and
2) h  : Bn → n with h  (Sn−1 ) = n−1 .
x2
Proof It is simple to write down the maps explicitly. Define h ♦ (x) = x1
x and
x2
h  (x) = x∞
x for x
= 0, h ♦ (0) = 0 = h  (0.) Then h ♦ and h  are bijective con-
√
tinuous maps. (Continuity at 0 follows since x2 ≤ nx∞ and x2 ≤ x1 .)
Since all the spaces are compact and Hausdorff, both the maps are homeomorphisms.
The other properties are clear from the definitions. 

Exercise: a) If K is a compact convex set with nonempty interior in Rn , then

K − K := {x − y : x, y ∈ K } spans Rn .
b) A nonempty compact convex set in Rn is homeomorphic to Bm for some m ≤ n
(e.g. a line segment is a one-dimensional ball).

13.2 Cubes and Simplices

We collect a minimal amount of combinatorial settings—cubes and their subdivisions

into simplices—that we need for our proofs of the fundamental theorems of Euclidean
topology. Fix n ∈ N. For a < b, [a, b]n is called an n-cube. It will be convenient to
consider I n = [0, 1]n or the symmetric cubes I (a)n = [−a, a]n . There is a boundary
preserving homeomorphism between an n-cube and the ball Bn , 13.4.
The sets I j− = {x ∈ I n : x j = 0} and I j+ = {x ∈ I n : x j = 1} are the j-th faces
of I n , 1 ≤ j ≤ n. Similarly, write I j+ (a) and I j− (a) for the j-th faces of I (a)n . For
13.2 Cubes and Simplices 185

n = 2, the faces are the four sides of the square. For n = 3, the faces are the six
squares bounding the cube.
We need to consider subdivisions of the cube into smaller subcubes and sub-
cells. For example, the midpoints of the sides of the square and the four vertices
together give a subdivision of the square into 16 triangular cells. We formalise such
subdivisions for any cube.
Consider I n . Let e j be the standard unit n-vectors and write, for k ∈ N, say n >
1, ekj = k1 e j . The combinatorial n-cube is the product C(k) = {0, k1 , . . . , k−1 k
, 1}n .
± k
For brevity, we generally suppress n in the notation for I j , e j , C(k) etc. The faces
of C(k) are C ±j (k) = C(k) ∩ I j± and their union ∂C(k) = ∪ j (C +j (k) ∪ C −j (k), is
the boundary of C(k). Combinatorial subcubes of C(k) are sets of the form Q k =
{c +  j a j ekJ : a j ∈ {0, 1}} ⊂ C(k), c ∈ C(k). Analogous definitions and notation
are used for I (a)n : ekj (a) = ak e j , Ca (k) = {0, ±a k
, . . . , ±ka
k
}n , etc.
A combinatorial n-simplex is an ordered set S = v0 , v1 , . . . , vn , with v1 − v0 =
eσk (1) , v2 − v1 = eσk (2) , . . . , vn − vn−1 = eσk (n) , for a permutation σ of {1, . . . , n};
v0 , . . . , vn in C(k) are its vertices. For example, 0, e1k , e1k + e2k , . . . , e1k + · · · + enk 
is such an n-simplex with σ = id and one with σ = (1, 2) (the transposition) is
0, e2k , e1k + e2k , . . . , e1k + · · · + enk . The associated geometric n-simplex is the con-
vex hull of the v j :
S = [v0 , v1 , . . . , vn ] := {0n t j v j : t j ≥ 0, t j = 1}.
Note that {v1 − v0 , . . . , vn − v0 } is a basis for Rn and so every x ∈ R is a unique
linear combination x = 0n l j (x)v j with 0n l j = 0. The l j are linear functionals on Rn
and hence are continuous. (If l is a linear functional, l(x) = xi f (ei ) and |l(x)| ≤
lx where l is the norm of (l(e1 ), . . . , l(en )) ∈ Rn . l is uniformly continuous as
|l(x) − l(y)| = |l(x − y)| ≤ lx − y.) C(k) gives rise to a subdivision the cube
into smaller cubes and into n-simplices which become smaller as k gets bigger. The
subset F j (S) = v0 , . . . , v̂ j , . . . , vn , the j-th face of S, is got by omitting the j-th
vertex v j from S. It is not an (n − 1)-simplices as per our definition, unless j = 0
or j = n.
Geometric simplices are points, line segments, (closed) triangular regions, solid
tetrahedra and their higher dimensional analogues. A simplicial complex is a col-
lection of simplices arranged in a natural, nice way. Their study and applications to
geometry and topology was initiated by Poincaré (who else!) in 1899 in his second
paper on Analysis situs. We will not study simplicial complexes, but will only use
cubes and their subsimplices.

Proposition 13.5 A geometric n-simplex is convex and compact.

Proof  := {(t0 , . . . , tn ) ∈ Rn+1 : λi ≥ 0, λi = 1} is closed and bounded and

hence is compact. The map, (t0 , . . . , tn ) → t j v j , Rn+1 → Rn , is continuous
and the simplex S is the image of  under this map. It is trivial to check that S
is convex. 
186 13 Combinatorial Methods in Euclidean Topology

13.3 Sperner’s Lemma, the Cubical Version

Sperner’s Lemma concerns simplices and their subdivisions. To avoid the technical-
ities about simplicial complexes and their subdivisions, we use a form of Sperner’s
Lemma for cubes. The Cubical Sperner Lemma was first given by Kuhn [12] in 1960.
We follow Kulpa’s proof, [13, 14].

Lemma 13.6 1) If S is an n-simplex in C(k), for every j ∈ {0, . . . , n}, there is a

unique ‘neighbour’ n-simplex S[ j] such that S ∩ S[ j] = F j (S).
2) S[ j] ⊂ C(k) holds if and only if F j (S) is not contained in the boundary ∂C(k).

Proof 1) If S = v0 , v1 , . . . , vn , define S( j) explicitly for each j:

i). S(0) = v1 , . . . , vn , w0  where w0 = vn + v1 − v0 ;
ii). For 0 < j < n, S( j) = v1 , . . . , v j−1 , w j , v j+1 , . . . , vn  where w j = v j−1 +
v j+1 − v j = v j−1 + eσ ( j+1) ; Fig. 13.2.
iii). S(n) = wn , v0 , . . . , vn−1  where wn = v0 − vn + vn−1 .
It is clear that S ∩ S[ j] = F j (S) for all j. The differences between the successive
vertices are eτk (1) , . . . , eτk (n) where, in the three cases,
i). τ is the cycle (σ (1), . . . , σ (n));
ii). τ is the transposition (σ ( j), σ ( j + 1));
iii). τ is the cycle (σ (n), . . . , σ (1))
and hence S[ j] is an n-simplex for each j.
Uniqueness. The case j = 0. If S  is an n-simplex such that S ∩ S  = F0 (S) =
v1 , . . . , vn , then S  is of the form v1 , . . . , vn , w for some w and then, by the
requirement on the successive differences, w = vn + (v1 − v0 ). An analogous argu-
ment takes care of S[n].
In case 2, 0 < j < n, S ∩ S  = F j (S) = v0 , . . . , v̂ j , . . . , vn , and it is clear that
S  is of the form v0 , . . . , v j−1 , w, v j+1 , . . . , vn . If τ is the permutation associated
to S  , writing the successive differences, we get
τ (1) = σ (1), . . . , τ (i) = σ (i) for 1 ≤ i ≤ n, i
= j, j + 1.

v1 S[0] v2
S
S[2] S[1]
v0

Fig. 13.2 Neighbours of a simplex

13.3 Sperner’s Lemma, the Cubical Version 187

Hence, τ
= σ forces τ ( j) = σ ( j + 1) and τ ( j + 1) = σ ( j), so w = w j as in the
definition of S[ j] and S  = S[ j]. Uniqueness is thus proved in all the cases.
2) Again, we consider the three cases.
Case i). S = v1 − eσk (1) , v1 , . . . , vn , S[0] = v1 , . . . , vn , vn + eσk (1)  and F0 (S) ⊂
C(k). Then the coordinates of all the vertices lie between 0 and 1. To conclude that
j-th coordinate of the vr can not all be 0 or 1 for any j. For j = σ (1) it is clear
that v1 ( j)
= vn ( j) since 0 ≤ v1 ( j) − 1/k, vn ( j) + 1/k ≤ 1. For other coordinates,
since vi+1 − vi = eσk (i) , it follows that vi+1 (σ (i))
= vi (σ (i)). Thus, σ (i) coordinate
of all of v1 , . . . , vn can not be the same and so no coordinate of all the vr can be the
same. In particular, F0 (S)  I j± for any j.
Conversely, if F0 (S) is contained in some C ±j , either the j-th coordinates of
v1 , . . . , vn are all 0 or all of them are 1. In either case, from the difference relations
of the vertices, we get that the j-th coordinate of eσk (i) is 0 for i
= 1, so σ (i)
= j
for i
= 1. Hence σ (1) = j. If vi ( j) = 0 for all i, the j-th coordinate of v1 − eσk (1) =
v1 − ekj is −1/k and if vi ( j) = 1 for all i, the j-th coordinate of vn + eσk (1) = vn + ekj
is 1 + 1/k. In any case, not all vertices in S or S[0] have all coordinates in [0,1].
This proves 2) in case i)). The other cases are similar. 

Corollary 13.7 An (n − 1)-face of an n-simplex in C(k) is a face of exactly one or

two simplices in C(k), according as it lies on the boundary or not.

A labelling or colouring of points (‘vertices’) of C(k) is a map λ : C(k) →

{0, 1, . . . , n}. An (s + 1)-element subset S of C(k) is said to be fully coloured if
λ(S) = {0, 1, . . . , s}. We call a colouring λ a Sperner colouring if (i) λ(v) < j if
v ∈ C −j (k) and (ii) λ(v)
= j − 1 if v ∈ C +j (k). See Figs. 13.3 and 13.4

Theorem 13.8 (Cubical Sperner’s Lemma)

If λ is a Sperner colouring of C(k), the number of fully coloured n-simplices in C(k)
is odd for each positive integer k. In particular, there is at least one fully coloured
n-simplex in C(k).

Proof Let Nk = Nk (n) be the number of fully coloured n-simplices in C(k) =

Cn (k). We show that Nk is odd by induction on n.
The case n = 1. C(k) = {0, k1 , . . . , k−1k
, 1} and the definition of Sperner colouring
j
forces λ(0) = 0, λ(1) = 1. Each λ( k ) is either 0 or 1. Every change of value of λ
gives a fully coloured 1-simplex and there are an odd number of such changes (since
we start with 0 and end with 1). So the number N1 of fully coloured 1-simplices is
odd.
Let n > 1 and, for an n-simplex S, let N (S) denote the number of fully coloured
(n − 1)-faces of S. If S = v0 , . . . , vn  is full, then the only fully coloured (n − 1)-
face of S is got by omitting the v j with λ(v j ) = n and N (S) = 1. If S is not fully
coloured, then there are two possibilities: (a) λ(S) = {0, 1, . . . , (n − 1)}, in which
case vi = v j for some i, j, i
= j and omission of either vi or v j gives a full face,
so N (S) = 2; (b) otherwise, when λ(S)
= {0, 1, . . . , (n − 1)}, N (S) = 0 because
omission of any v j will not yield a full face. So, if a is the number of fully coloured
188 13 Combinatorial Methods in Euclidean Topology

1 0 2
F NF
NF NF
1 2 0
F NF
F F
0 0
1

Fig. 13.3 Labelling: F-full, NF-not full

0 2 0 2

0 1 2 1

0 2 1 2

0 1
1 0

Fig. 13.4 Sperner coloured

n-simplices in C(k) and b is the number of n-simplices S with λ(S) = {0, 1,

. . . , (n − 1)}, then the total number of fully coloured (n − 1)-faces in C(k) is a + 2b.
Now let us count this in a different way. The previous corollary shows that a
fully coloured (n − 1)-face F lies in two n-simplices in C(k) if it does not lie in the
boundary ∂C(k) and any boundary fully coloured (n − 1)-face F is a face of only
one n-simplex. Hence the total number of fully coloured (n − 1)-faces is 2 p + q,
where p, q are the number of nonboundary and boundary n-simplices, respectively.
Thus 2 p + q = a + 2b.
If F is a full (n − 1)-face in ∂C(k), then F ⊂ Cn− (k). (Condition (ii) in the
definition of Sperner colouring precludes superscript + and condition (i) rules
out the superscript – when the subscript is < n.) We can identify Cn− (k) = C(k)
∩In− with Cn−1 (k), and under this identification F is a full (n − 1)-simplex in
{0, k1 , . . . , k−1
k
, 1}n−1 . Thus q = Nk−1 . This is odd, by induction hypothesis.
Since a + 2b = 2 p + q and q is odd, a has to be odd as well and the induction
is complete. 

Corollary 13.9 Under the assumptions of the cubical Sperner lemma, there is a
fully coloured subcube Ck .

Proof If Sk = v0 , . . . , vn  is a fully coloured n-simplex, then the subcube Ck :=

{v0 + a j ekj : a j ∈ {0, 1}} ⊃ Sk is fully coloured. 
13.4 Poincaré and Brouwer 189

13.4 Poincaré and Brouwer

The introduction of qualitative methods in nonlinear analysis (and, in particular,

use of fixed point theorems) by Poincaré is considered as one of his most profound
contributions (e.g. see [15]).
Brouwer’s statement of the theorem: Eines eindeutige und stetige Transformation
eines n-dimensionalen Elementes in sich besitzt sicher einen Fixpunkt. (‘A continu-
ous transformation of an n-dimensional element into itself has a fixed point’.) (For
Brouwer, an n-dimensional element is a one-one continuous image of an n-simplex
in a Euclidean space).
In 1883 Poincaré [1, 2] gave a higher dimensional analogue of the intermediate
value theorem and indicated a proof in [16]. This result is surprisingly not as familiar
in mathematical circles as it should be. The result was rediscovered by Miranda [17]
in 1940, who proved that it is equivalent to the widely known Brouwer’s fixed point
theorem. We give proofs of both of these results following Kulpa [13]. We begin
with a simple, noncombinatorial lemma.
Lemma 13.10 In a metric space (X, d), if {Sk } is a sequence of subsets with diam
Sk → 0 and {K i }i∈I is a family of compact subsets such that Sk ∩ K i is nonempty
for all k and all i, then ∩i K i is nonempty.

Proof First consider the case when I is finite, so let {K 1 , . . . K r } be compact

 sets
such that K j ∩ Sk is nonempty for all j, k. Consider the product space K = r1 K j ,
say with the metric d1 (x, y) =  j d(x j , y j ), where x = (x j ), y = (y j ). Choose
xk j ∈ K j ∩ Sk for each j, k and take xk = (xk1 , . . . , xkr ). By compactness, there
is a convergent subsequence {xkm } of {xk }, say limm xkm = a = (a1 , . . . , ar ). Then
d(ai , a j ) ≤ d(ai , xkm i ) + d(xkm i , xkm j ) + d(xkm j , a j )
≤ 2d1 (a, xkm ) + diam(Sk ) → 0 as m → ∞.
So ai = a j for all i, j and (a1 , . . . , a1 ) ∈ ∩ j K j .
The general case is not needed for us but, by the finite case, {K i } has f.i.p., hence
has nonempty intersection, by compactness. 

Müger [14] extracted and explicitly stated the following result from Kulpa’s proof
of the Poincaré-Miranda theorem. He called it the higher connectedness of the cube.
The reader may like to ponder over this designation and see what the theorem says
when n = 1.
Theorem 13.11 Let H j± be closed sets in I n such that H j± ⊃ I j± and H j+ ∪ H j− =
I n for each 1 ≤ j ≤ n. Then ∩n1 (H j− ∩ H j+ )
= ∅.

Proof Define F0 = I n and F j := H j+ \ I j− for 1 ≤ j ≤ n. Define λ(x) := max{k :

x ∈ ∩k1 F j } for x ∈ I n . Now I j− ∩ F j is empty, so λ(x) < j for x ∈ I j− . On the other
hand, let x ∈ I j+ and suppose λ(x) = j − 1. Then x ∈ ∩0 Fi and x ∈
j−1
/ F j , i.e. x ∈
/
H j+ or x ∈ I j− . This is not possible because I j+ is disjoint from I j− and I j+ ⊂ H j+ .
190 13 Combinatorial Methods in Euclidean Topology

Conclusion: λ(x)
= j − 1 for x ∈ I j+ . Thus, λ|C(k) is a Sperner colouring and, by
the cubical Sperner lemma (13.8), there is a fully coloured n-simplex Sk in C(k) for
each k. Claim: Sk meets each H j± . For, any x ∈ Sk with λ(x) = j ∈ {0, 1, . . . , n}
belongs to F j ⊂ H j+ . If x ∈ Sk and λ(x) = j − 1 ∈ {0, 1, . . . , (n − 1)}, then x ∈/ Fj
/ H j+ or x ∈ I j− . In either case, x ∈ H j− and the claim is proved.
or, equivalently, x ∈
√ previous lemma with X = I , and
n
Now the proof is completed on invoking the
±
{K i } = {Hi }, since diam Sk ≤ diam Ck = n/k → 0. 

For example, if H j+ = {x ∈ I n : x j ≥ 1/2}, H j− = {x : x j ≤ 1/2}, then

∩n1 (H j− ∩ H j+ ) = {(1/2, . . . , 1/2)}.
The Poincaré theorem now is a surprisingly easy consequence.
Corollary 13.12 (Poincaré-Miranda theorem)
Let f = ( f 1 , . . . , f n ) : I n → Rn be continuous. If f j (I j− ) ⊂ (−∞, 0] and f j (I j+ ) ⊂
[0, ∞) for all j, then there is an x ∈ I n such that f (x) = 0.

Proof Write R+ = [0, ∞), R− = (−∞, 0] and apply the theorem (13.11) with
H j± = f j−1 (R± ) to get an x ∈ ∩ j H j± . Then f (x) = 0. 

Example 13.13 Let 0 ≤ c ≤ 1, n = 2. The function f (x1 , x2 ) = ((x1 − c)x2 ,

x2 − x1 ). f satisfies the conditions and f (c, c) = (0, 0).

These results hold for any cube [a, b]n in place of I n . For n = 1, the corollary is
precisely the intermediate value theorem.
Corollary 13.14 If g, h : I n → I n are continuous maps and if h(I j± ) ⊂ I j± for all
j, there is a p ∈ I n such that g( p) = h( p).

Proof Poincaré-Miranda applies to f (x) = h(x) − g(x). 

Corollary 13.15 If h : I n → I n is continuous and if h(I j± ) ⊂ I j± , then h is surjec-

tive.

Proof If b ∈ I n , take g to be the constant function g(x) = b in the last corollary. 

Corollary 13.16 (Brouwer fixed point theorem)

If g : I n → I n is continuous, there is a point x ∈ I n such that g(x) = x.

Proof Apply Poincaré-Miranda (13.12) for f (x) = x − g(x). 

As mentioned earlier, Brouwer actually stated his theorem for an n-simplex. This is
equivalent to the result for a cube or a ball since there are boundary preserving home-
omorphisms between an n-simplex, an n-cube and an n-ball. The same remarks apply
to Borsuk’s no retract theorem and Bohl’s theorem given below. (See Proposition
13.21.)
For the next corollary, we need a definition. A subset E of a space X is a retract
of X if there is continuous map r : X → E such that r (x) = x for x ∈ E. Such a
13.4 Poincaré and Brouwer 191

map r is called a retraction. This concept, introduced by Borsuk [18], will be studied
in detail in the next chapter. The next corollary is a special case of a result due to
Borsuk [18]; see also 3.6 of [19]. (A main result of [18] is that if B is a retract of Rn ,
the components of Rn \ B are unbounded (Théorème 25, [18]).)
Corollary 13.17 (Borsuk No Retract Theorem, 1931)
The boundary ∂ I n is not a retract of I n .

Proof Corollary 13.15 applies to a such a retraction. 

Corollary 13.18 (Bohl’s Theorem, 1904)

There is no continuous map f : I n → Rn \ {0} such that f is the identity on the
boundary of I n .

Proof Suppose there is such a map f and let ρ(x) = x

x∞
for x
= 0. Then ρ◦ f is
a retraction of I n onto ∂ I n . 

Exercise We have proved Bohl’s theorem from Borsuk’s theorem and Borsuk’s the-
orem from Brouwer’s theorem. Obtain both the reverse implications. Before we
conclude, we make some simple observations on fixed points and give an application
to matrices.
Definition 13.19 A space X is said to have the fixed point property if every contin-
uous map on X (i.e. X → X ) has a fixed point.
For an extensive treatment of modern fixed point theory, see the thick monograph [20]
of Granas and Dugundji. Shashkin [21] contains a delightful, elementary discussion.

Example 13.20 1. Every point in X is a fixed point for id X .

2. On R, the translation x → x + 1 has no fixed point.
3. We have seen, as a consequence of the intermediate value theorem, that [0,1]
has the fixed point property.
4. The rotation ρ on S1 , ρ(eı x ) = eı(x+π/2) has no fixed point. More generally, on
any sphere Sn−1 , n > 1, nontrivial rotations leave no point fixed. For example, n × n
⎛
matrices of the form ⎞
cos γ − sin γ 0
⎝ sin γ cos γ 0 ⎠ , γ ∈ R,
0 0 I
where I is the identity matrix (of the appropriate size) map the sphere to itself, but
fix no point for γ ∈ / 2π Z.

Proposition 13.21 If X and Y are homeomorphic spaces and if X has the fixed point
property, then so does Y .

Proof Suppose X has the fixed point property and ϕ : X → Y is a homeomorphism.

If g is a continuous map on Y and if x is a fixed point for f = ϕ −1 ◦ g ◦ ϕ on X ,
then ϕ(x) is a fixed point for g. 
192 13 Combinatorial Methods in Euclidean Topology

Example 13.22 The hemisphere {x ∈ Sn : xn+1 ≥ 0} and the positive ‘octant’ Sn+ :=
{x ∈ Sn : x j ≥ 0, 1 ≤ j ≤ n + 1} of the unit sphere are both homeomorphic to the
n-ball Bn and hence have the fixed point property.

Corollary 13.23 A real (n + 1) × (n + 1) matrix A has a positive eigenvalue in

each of the following cases:
1) all the entries of A are positive or
2) all the entries of A are nonnegative and A is nonsingular.

Proof A acts linearly on Rn+1 in the canonical way: x → Ax and either of the
assumptions of the corollary ensures that Ax is nonzero and has nonnegative coor-
dinates for x ∈ Sn+ . Thus x → Ax/Ax maps Sn+ to itself and so has a fixed point.
Thus there is an x0 ∈ Sn+ with Ax0 = Ax0 x0 and so Ax0  is a positive eigenvalue
of A. 

13.5 Invariance of Domain and Dimension

Here are Brouwer’s statements of the results:

Satz 2. Bine m-dimensionale Mannigfaltigkeit kann nicht das eineindeutige
und stetige Abbild eines Bereiches höherer Dimensionenzahl enthalten. (‘An m-
dimensional manifold cannot contain the one-one continuous image of a region of
higher dimension’.)
Satz 3. In einer m-dimensionalen Mannigfaltigkeit ist das eineindeutige und stetige
Abbild eines Bereiches geringerer Dimensionenzahl eine nirgends dichte Punkt-
menge. ‘In an m-dimensional manifold, an image of a lower dimensional region
is nowhere dense’.)
Die Sätze 2 und 3 enthalten beide die Invarianz der Dimensionenzahl als unmit-
telbare Folgerung. (‘Theorems 2 and 3 have dimension invariance as an immediate
consequence’.) (Brouwer [6])
Poincaré had formulated and used without proof the invariance of dimension in
his work on automorphic functions in 1883 [1]. Apparently he had also suggested
proofs much later. Brouwer must have been inspired by Poincaré’s works (see [22])
and his meeting with Poincaré.
The concept of dimension for Euclidean spaces is familiar to most, even if its
precise formulation is not: R, R2 , R3 have dimensions 1, 2, 3, respectively. So no
one has too much difficulty in accepting that Rn has dimension n. A bit of exposure
to Linear Algebra also shows that Rn has ‘vector space dimension’ n. In particular,
Rn and Rm are not ‘the same’ as vector spaces if m
= n. We are concerned here
with the corresponding topological question. When are Rn and Rm homeomorphic?
As we have observed earlier, using connectedness it is easy to conclude R is not
homeomorphic Rn , n
= 1. But beyond this, it is far from easy to show even that
R2 and R3 are not homeomorphic. Clearly, it is a very basic problem. The theorems
13.5 Invariance of Domain and Dimension 193

on invariance of domains and invariance of dimension are among the most basic
theorems in topology and geometry. These were both proved by Brouwer, no doubt
inspired by Poincaré and his methods, in 1911 [6].
One can define dimension for more general spaces. As simple examples, look at
the sphere Sn locally. Points in S1 have neighbourhoods which are arcs and arcs have
dimension one! In fact, an arc is homeomorphic to an interval. Similar arguments
can be given for any Sn and one can say Sn has dimension n. There are theories of
dimension for topological spaces, but we will not discuss them. We will be solely
concerned with Euclidean spaces and show that Rn and Rm are ‘not the same’ as
topological spaces if m
= n. This is the dimension invariance theorem. The domain
invariance theorem is more general.
In this section, first we present a proof due to Müger [14], of a result on dimension
invariance, based on Kulpa’s proof of the cubical Sperner lemma. Then, Kulpa’s
elementary proof of the domain invariance theorem, [22], is given.
A closed set F separates closed sets A and B in X , if there are disjoint open sets
U and V such that A ⊂ U, B ⊂ V and U ∪ V = F c .
We use the earlier notation and results on cubes. Here is a consequence of the
Theorem on Higher Connectedness 13.11.
Proposition 13.24 If a closed set F j in the cube I n separates I j+ and I j− for each
j ∈ {1, . . . , n}, then ∩n1 F j is nonempty.
Proof For each j, there are disjoint open sets U ± ± ± +
j with I j ⊂ U j and F j = U j ∪
c

U− ± ± ± ∓ +
j . The sets H j := U j ∪ F j is closed since X \ H j = U j . For each j, H j ∪
− + − + −
H j = I and 13.11 shows ∩ j (H j ∩ H j ) is nonempty. But H j ∩ H j = F j and so
n

the proof is complete. 

Lemma 13.25 Let D be dense in R. If A, B are disjoint closed sets in I n , there is a
closed set F ⊂ {(x1 , . . . , xn ) ∈ I n : ∃ j, x j ∈ D} separating A and B.

Proof There is a basic neighbourhood Ux = I n ∩ (a j , b j ) of x for each x ∈ A,
with a j , b j ∈ D such that Ūx is disjoint from B. By compactness there are finitely
many points x1 , . . . , xk in A with U := Ux1 ∪ . . . Uxk ⊃ A. Then F = ∂U satis-
fies the requirements. For, F is closed and F c = U ∪ V with V = I n \ Ū , A ⊂ U,
B⊂ V , so F separates A and B. One coordinate of a boundary point of
I n ∩ (a j , b j ) is a j or b j and so each point of ∂U ⊂ ∪∂Ux j has a coordinate
in D. 
Proposition 13.26 If {(A j , B j )}n+1
1 are pairs of disjoint closed sets in I n , there are
closed sets F j separating A j , B j such that ∩ j F j = ∅.
Proof Let {Dk } be a sequence of mutually disjoint dense subsets of R. (Example:
Dk = kπ + Q, k ∈ N.) By the lemma, for each j there is a closed set F j separating
A j and B j with every point of F j having some coordinate in D j . If there is a point
in x ∈ ∩ j F j then, for each j, there is a k j such that the k j -th coordinate xk j ∈ D j .
Thus there are j
= j  such that k j = k j  (Pigeon hole principle!). This contradicts
the fact that D j and D j  are disjoint and so the proof is complete. 
194 13 Combinatorial Methods in Euclidean Topology

Theorem 13.27 If [0, 1]n is homeomorphic to [0, 1]m , then n = m. Equivalently, Bn

and Bm are not homeomorphic for m
= n.
Proof Proposition 13.26 shows that, for any n, sets separating any n + 1 pairs of
disjoint closed sets in I n have empty intersection, whereas by 13.24 this is not true
for any fewer number of pairs. For I m , the first two sentences hold with m in place
of n. If I m and I n are homeomorphic, these numbers coincide, i.e. n = m. 
Now we turn to Kulpa’s proof of the deeper theorem of invariance of domain.
It will be convenient to work with cubes I (a)n = [−a, a]n , a > 0. We use terms
and notation as before. Kulpa’s proof [22] was presented by Terence Tao in his
blog and later in his book [23] on Hilbert’s fifth problem. Kulpa based his proof on
the following topological lemma. He also pointed out that analytical arguments can
be used in place of the topological lemma. Tao uses this analytical method in his
presentation of Kulpa’s proof.
Lemma 13.28 (Kulpa)
Let X, Y be compact subsets of Rn with Y having no interior. Suppose f : X →
Rn \ {0} is a continuous, nonvanishing function. Then, for any ε > 0, there is a non-
vanishing continuous function F : X ∪ Y → Rn \ {0} such that  f (x) − F(x) < ε
for x ∈ X .
Proof Let ε > 0. Choose a, δ > 0 such that 0 < 2δ < ε, X ∪ Y ⊂ I (a)n and the
closed ball B̄(0, 2δ) is disjoint from f (X ). By Tietze’s theorem, there is a continuous
extension of f to I (a)n , still denoted by f for convenience. Cover Rn by cubes Q i
of diameter < δ. By uniform continuity of f , there is a subdivision of I (a)n into
small n-simplices S j such that each f (S j ) is contained in some Q i (take n-simplices
in C(k) for large k).
Define a map h on I (a)n so that it is affine linear on each S j : if v0 , . . . , vn are the
vertices  of S j any point x ∈ S j is a unique convex combination x = tr vr . Define
h(x) = tr f (vr ). Each f (S j ) ⊂ Q i for some i and so h(S j ) ⊂ Q i by convexity.
In particular,  f (x) − h(x) < δ.
Now Y has dense complement and so does h(Y ). This is seen as follows. If the
f (v j ) are not convex independent (i.e. { f (v1 ) − f (v0 ), . . . , f (vn ) − f (v0 )} is lin-
early dependent) then h(S j ) lies in an (n − 1)-dimensional hyperplane in Rn and so
its complement is dense. On the other hand, if the f (v j ) are convex independent, both
the sets {v1 − v0 , . . . , vn− v0 } and { f (v1 ) −
f (v0 ), . . . , f (vn ) − f (v0 )} are bases
for Rn . Hence h(w) = tr f (vr ) for w = tr vr defines an affine linear homeo-
morphism of Rn . Each S j ∩ Y has dense complement and hence so does h(S j ∩ Y ).
Thus, h(Y ) is a finite union of compact sets with dense complements, so is a compact
set with dense complement. For x ∈ X
h(x) ≥  f (x) −  f (x) − h(x) ≥ δ
i.e. B(0, δ) ⊂ h(X )c , where x = x∞ = max |x j |. Further, h(Y )c is dense, so
meets B(0, δ). Hence there is a v ∈ B(0, δ) \ h(X ∪ Y ). Define F(z) = h(z) − v
for z ∈ X ∪ Y . As v ∈ / h(X ∪ Y ), F never vanishes and
 f (x) − F(x) ≤  f (x) − h(x) + v < 2δ < ε for x ∈ X . 
13.5 Invariance of Domain and Dimension 195

Now we are ready to prove the very important invariance theorem.

Theorem 13.29 (Domain Invariance Theorem)
Let U ⊂ Rn be open (and nonempty!). If h : U → Rn is continuous and injective,
the image h(U ) is open in Rn .
Proof Fix a p ∈ U . It suffices to prove that h( p) is an interior point of h(U ). By
a translation, we may assume that p = 0. Choose a cube I (a)n ⊂ U ; to show that
u = h(0) is an interior point of h(I (a)n ).
Now, I (a)n is compact and Rn is Hausdorff and so h is a homeomorphism of
I (a)n onto h(I (a)n ). Thus h −1 : h(I (a)n ) → I (a)n is continuous and hence has a
continuous extension g : Rn → Rn , by the Tietze extension theorem. If u = h(0),
then g(u) = 0 and so there is a δ > 0 such that g(y) < a when y − u < 2δ.
Suppose u is not an interior point of h(I (a)n ). Then there is a v ∈ / h(I (a)n ) with
v − u < δ. This will lead to a contradiction.
δ
For z ∈ h(I (a)n ) ∪ B ∗ (v, δ), let ϕ(z) = v + max{ z−v , 1}(z − v). Then
∗
ϕ(z) − v ≥ δ for all z, ϕ(z) − v = δ for z ∈ B (v, δ) = B(v, δ) \ {0} and
for z ∈ h(I (a)n ) \ B ∗ (v, δ), ϕ(z) = z. Thus ϕ : h(I (a)n ) ∪ B ∗ (v, δ) → X ∪ Y is a
continuous function, where
X = {x ∈ h(I (a)n ) : x − v ≥ δ} and Y = {x ∈ Rn : x − v = δ}.
Note that g is nonzero on the compact set X , and so there is a η > 0 such that
g(x) > η for x ∈ X . We can take η < a. By Kulpa’s lemma, with X, Y as above
and g = h −1 in place of f , there is a continuous function F : X ∪ Y → Rn \ {0}
with F(x) − g(x) < η for x ∈ X . Hence ψ = F ◦ ϕ ◦ h : I (a)n → Rn \ {0} is a
nonvanishing continuous function.
We show that it satisfies the conditions of the Poincaré-Miranda theorem. Let
y ∈ I j (a)+ . Then h(y) ∈ X ; for if h(y) − v < δ, then h(y) − u < 2δ and so
y = g(h(y)) < a and y ∈ int I (a), a contradiction. Thus h(y) ∈ X hence
ϕ(h(y)) = h(y).
ψ(y) − y = F(h(y)) − g(h(y)) < η < a,
|ψ j (y) − a| = |ψ j (y) − y j | ≤ ψ(y) − y < a.
and so ψ j (y) > 0 and ψ j (I j (a)+ ) ⊂ (0, ∞) for all j. Similarly, we can show that
ψ j (I j (a)− ) ⊂ (−∞, 0) for all j. On invoking Poincaré-Miranda, we conclude that
ψ vanishes at some point, giving the sought after contradiction. Hence the theorem
is proved. 

Corollary 13.30 (Dimension Invariance Theorem)

If m < n, there is no continuous injection Rn → Rm . In particular, Rn and Rm are
not homeomorphic if m
= n.
Proof If f : Rn → Rm is a continuous injection, take the inclusion map ι : (x1 , . . . ,
xm ) → (x1 , . . . , xm , 0, . . . , 0) of Rm in Rn . Then h = ι ◦ f : Rn → Rn is a contin-
uous injection. But clearly h(Rn ) is not open in Rn as the last n − m coordinates of
any point in it are 0. This contradicts the domain invariance theorem (13.29). 
196 13 Combinatorial Methods in Euclidean Topology

The dimension invariance (13.30) was proved by Brouwer (1911). Lebesgue also
gave proofs of the invariance results. These turned out to be incorrect, but his work
led to the concept of covering dimension.

13.6 Borsuk and the Sphere

Borsuk, in 1933, obtained three remarkable results on the sphere:

Satz I. Jede antipodentreue Abbildung von Sn is wesentlich.
Satz II. Ist f ∈ RnSn (d.h. bildet f die Sphäre Sn auf einen Teil von Rn ab) so gibt
es einen derartigen Punkt p ∈ Sn dass f ( p) = f ( p ∗ ) ist.
Satz III. Sind A0 , A1 , . . . , An in sich kompakte der Sphäre Sn enthälte, so enthälte
die Summe 0n Ai die Sphäre Sn nicht [24].
These may be stated thus: 1. Any antipodal map of Sn is essential. 2. If f maps Sn
into Rn , there is a point p ∈ Sn such that f ( p) = f ( p ∗ ). ( p ∗ = − p is the antipode
of p.) 3. If A0 , A1 , . . . , An are compact sets in Sn none of which contains a pair of
antipodal points, they can not cover Sn . The second one is usually called the Borsuk–
Ulam theorem (13.34). The only reference to Ulam is Borsuk’s foot-note: “Dieser
Satz wurds als Vermutung von St.Ulam aufgestallt." (“This theorem was raised as
a conjecture by St.Ulam.”) Satz III is the Lusternik–Schnirelmann–Borsuk theorem
(13.35).
We obtain results on maps of spheres à la Borsuk employing a famous combina-
torial result known as Tucker’s Lemma [25]. This, in turn, is deduced from a more
general result: Ky Fan’s Lemma [26]. Our presentation of Ky Fan’s Lemma is based
on [27].
It will be convenient to work with the octahedron ♦n , which is the unit sphere in the
norm .1 . A finite family T of simplices in ♦n covering ♦n is called a triangulation
or subdivision of ♦n if (1) any two intersecting simplices in T intersect in a common
face, (2) each face of any σ ∈ T is in T and (3) each (n − 1)-simplex in T is the
common face of exactly two n-simplices in T. For example, taking the midpoint of
each edge as a new vertex of the square forming ♦1 gives a triangulation. In the same
way, subdividing each triangular face of the octahedron ♦2 yields a triangulation of
♦2 .
The barycentre of a simplex with vertices v0 , . . . , vk is, by definition, the point
1
k+1
v j . Starting with ♦n and taking barycentres of all its faces as vertices, we get a
subdivision called the (first) barycentric subdivision, (Figs. 13.5, 13.6 and 13.7a–d).
Clearly, we can form successive barycentric subdivisions of ♦n .
The reflection ρ, ρ(x) = −x, is a homeomorphism of Rn+1 and of ♦n called
the antipodal map. Each ♦k , k < n, is ρ-invariant. A triangulation T is symmetric
if σ ∈ T implies ρ(σ ) = −σ ∈ T. e.g.: the barycentric subdivision of ♦n . A k-
hemisphere in ♦n is a subset Hk ⊂ ♦n that is homeomorphic to to the k-ball k . The
upper and lower hemispheres ♦k± are k-hemispheres.
13.6 Borsuk and the Sphere 197

2+

2−

Fig. 13.5 2

Fig. 13.6 Barycentric subdivision of 2+

A flag of hemispheres in ♦n is a sequence of H0 ⊂ · · · ⊂ Hn ⊂ ♦n such that

H0 , −H0 are antipodal points, Hn ∪ −Hn = ♦n and ∂ Hk = ∂(−Hk ) = Hk ∩ H−k =
Hk−1 ∪ (−Hk−1 )  ♦k−1 . E.g. Hk = ♦k+ .

A symmetric triangulation is said to be aligned with hemispheres if there is a flag

of hemispheres with Hk contained in the union of k-simplices of the triangulation
(= the k-skeleton of the triangulation). The barycentric subdivision is an example.
The support of a simplex σ in such a triangulation is the smallest Hk or −Hk that
contains it.
For a triangulation T of ♦n , let V (T) denote the set of vertices in T. A map
 : V (T) → Z is called a labelling or colouring of T (by integers.) A labelling of
a symmetric triangulation is called an antipodal labelling if (−v) = −(v) for
each vertex. Given a labelling, a simplex [v0 , . . . , vk ] is +-alternating if (vi ) =
(− ji )i−1 with 0 < j0 < j1 < · · · , < jk , that is, its labels have the form 0 < j0 , − j1 ,
. . . , (−1)k jk . It is −-alternating if the labels are of the form − j0 , j1 , − j2 , . . .
(−1)k−1 jk . It is almost alternating if it has a (k − 1)-face (= a facet) that is ±-
alternating, i.e. removal of a vertex results in an alternating simplex. Consider 2-
simplices with labels (1, −2, 3), (1, 1, −2) and (1, 2, −1): the first is alternating, the
198 13 Combinatorial Methods in Euclidean Topology

(a) (b)

(c) (d)
Fig. 13.7 a–d Octahedron ♦2 , stages of its barycentric subdivision

second is almost alternating and the last is not even almost alternating. In a symmetric
triangulation aligned to a flag of hemispheres, an alternating or an almost alternating
simplex is admissible if its sign matches that of its support hemisphere.
In the proof of Ky Fan’s Lemma we construct a graph. A (finite) graph G consists
of a finite set V (G) and a subset E(G) ⊂ V (G) × V (G). Points v ∈ V (G) are called
the vertices and elements (v, w) ∈ E(G) are called the edges of G. Geometrically,
an edge is thought of as the line segment joining two vertices (=points). Two vertices
forming an edge are adjacent. They are complementary (for a given labelling ) if
(v) = −(w). The number of edges at a vertex v is called its degree. [v1 , . . . , vr ]
is a path in G if vi , vi+1 are adjacent for 1 ≤ i ≤ r − 1.
13.6 Borsuk and the Sphere 199

Theorem 13.31 (Ky Fan Lemma)

Let T be a symmetric triangulation of ♦n aligned with hemispheres and let  :
V (T) → {±1, . . . , ±m} be an antipodal labelling with no complementary edge.
Then the number of +-alternating n-simplices is odd with an equal number of
−-alternating n-simplices. In particular, there is at least one n-simplex that is +-
alternating and one that is −-alternating and so m ≥ n + 1.

Proof Let H0 ⊂ . . . ⊂ Hn be a flag of hemispheres aligned to T. Define a graph G

as follows. The vertex set V (G) is made up of three types of simplices: the alter-
nating k-simplices, the admissible almost alternating k-simplices and the admissible
alternating (k − 1)-simplices. Simplices σ, τ ∈ V (G) are defined to be adjacent if
one is a facet of the other—say, τ is a facet of σ , the smaller simplex τ is alternating
and the sign of the support hemisphere of the bigger one σ matches that of τ .
We show that the degree is 1 or 2 for each vertex of σ ∈ G and deg σ =1 if and
only if its support is ±H0 (i.e. it is a 0-simplex) or it is an alternating n-simplex. We
analyse the degrees for various σ ∈ V (G).
1. σ is an alternating k-simplex carried by ±Hk .
It has one alternating facet τ whose sign agrees with the sign of the support
hemisphere of σ . Removal of which vertex from σ yields τ ? If v, w are the vertices
with the maximum or minimum of the absolute values of the labels of σ , removal of
one of them would yield τ with the correct sign (=sign of the support hemisphere)
and removal of the other would give the opposite sign. In any case, σ and τ would
be adjacent. Removal of any other vertex would not yield an alternating simplex.
Thus σ has a unique facet τ adjacent to it in G. Further, σ is a facet of two (k + 1)-
simplices, one each in Hk+1 , −Hk+1 , but only one of them will be adjacent to σ in
G, depending on the sign of σ . Conclusion: deg σ =2 for 0 < k < n. If k = 0, σ has
no facets and deg σ =1. If k = n, then σ is not the facet of any simplex and deg σ =1.
The other two cases are easy.
2. σ is an admissible, almost alternating k-simplex with support ±Hk . It is adjacent
to the two admissible alternating facets, deg σ = 2.
3. σ is an admissible, alternating (k − 1)-simplex supported in ±Hk . It is the
facet of exactly two k-simplices; each of these is admissible alternating or almost
alternating with the same support. The adjacency conditions are satisfied and deg
σ =2.
To summarise, except ±H0 and the alternating n-simplices, all the vertices in G
have degree 2. In other words, G is made up of disjoint paths with end points at ±H0
and alternating n-simplices.
Let us do some counting. Note that if σ, τ are adjacent in G, then so are −σ, −τ .
Hence if a set of simplices of T form a path P in G, then their antipodes form a path
-P. Since σ
= −σ and σ, −σ are not adjacent for σ ∈ G, it is easy to see that no
path P in G is the same as its antipodal path -P. If the number of vertices of P is odd,
P=-P would force (by pairing ±σi ) σ = −σ for some σ ; in the case of even number
of vertices, we would get adjacent vertices ±σ, untenable conclusions in both cases.
Hence the number of paths is even, 2r , and the number of end points is 4r . ±H0 are
two of these, so the number of alternating n-simplices is 4r − 2 = 2(2r − 1). The
200 13 Combinatorial Methods in Euclidean Topology

antipode of a +-alternating n-simplex is a −-alternating n-simplex and conversely.

Thus the number of each type of alternating n-simplices is 2r − 1. 
Corollary 13.32 (Tucker’s lemma)
Notation as in Ky Fan Lemma. If  : V (T) → {±1, . . . , ±n} is antipodal, then (v) =
−(w) for some pair of adjacent vertices v, w.
Proof Immediate from the last assertion of Fan’s Lemma. 
Borsuk’s theorem and other topological results are stated below for the ball Bn+1
and the sphere Sn , but the proofs are written out for n+1 and ♦n . This is legitimate
as there is an antipode and boundary preserving homeomorphism between Bn+1 and
n+1 (13.2,13.4).
Theorem 13.33 (Borsuk’s theorems)
1) There is no continuous map f : Bn+1 → Sn that is antipode preserving on Sn ,
i.e. such that f (−x) = − f (x) for all x ∈ Sn .
2) There is no continuous antipode preserving map Sn → Sn−1 .
3) If f : Sn → Rn is continuous and antipode preserving, it vanishes at some
point.
Proof 1. Let f : n+1 → ♦n be antipode preserving on ♦n . Choose a δ > 0 such
that  f (x) − f (y)1 < n2 for x − y1 < δ (uniform continuity) and a symmet-
ric triangulation T of ♦n with diam(s) < δ for s ∈ T. Define jv = min{ j : | f (v) j |
≥ 1/n}, v ∈ V (T) and (v) = sign( f (v) jv ). jv . Since f is antipode preserving on
♦n , so is . Tucker’s lemma yields adjacent vertices v, w such that j = (v) =
−(w) > 0. Then f (v) j ≥ 1/n and f (w) j ≤ −1/n and hence  f (v) − f (w)1
≥ 2/n. This is a contradiction as v − w1 < δ implies  f (v) − f (w)1 < 2/n.
2. h : (x1 , . . . , xn+1 ) → (x1 , . . . , xn ) is a homeomorphism of Sn+ onto Bn . If f :
Sn → Sn−1 is antipodal, then so is f ◦ h −1 : Bn → Sn−1 on Sn−1 .
3. If f : Sn → Rn is antipodal and nonvanishing, then an antipodal map Sn →
Sn−1 is given by x → f (x)/ f (x). 
These statements and the ones below are mutually equivalent, see [20], [28]. The
following famous and important result was proved by Borsuk in [24]. According to
him, it was conjectured by Ulam.
Theorem 13.34 (Borsuk–Ulam Theorem, Borsuk—1933)
If f : Sn → Rn is continuous, there is an x such that f (x) = f (−x).
Proof Apply part 3. of the previous result (13.33) to the antipode preserving map
g(x) = f (x) − f (−x). 
The next result proved by Lusternik–Schnirelmann [29, 30] and Borsuk [24] has
spawned many generalisations and applications.
Theorem 13.35 (Lusternik–Schnirelmann–Borsuk)
In any set of n + 1 closed sets C1 , . . . , Cn+1 covering Sn , at least one of them contains
a pair of antipodal points.
13.6 Borsuk and the Sphere 201

Proof Define f : Sn → Rn by f (x) = ( f 1 (x), . . . , f n (x)) where f j (x) = dist

(x, C j ). By Borsuk–Ulam (13.34), there is a point x such that f (x) = f (−x).
If f j (±x) = 0 for some 1 ≤ j ≤ n, then x, −x ∈ C j . If no f j (x) = 0, then ±x
∈
/ C j , 1 ≤ j ≤ n and so ±x ∈ Cn+1 . 

Remark 13.36 Look at a triangle and the inscribed circle to conclude n + 1 is the
best possible.

Theorem 13.37 (Open set LSB theorem)

In any set of n + 1 open sets U1 , . . . , Un+1 covering Sn , at least one of them contains
a pair of antipodal points.

Proof Suppose {U1 , . . . , Un+1 } is an open cover. For x ∈ Sn , there is a j such that
x ∈ U j . Then there is an open ball B(x, r ) with closure B̄(x, r ) ⊂ U j . A finite
number of these open balls cover Sn , by compactness. For each j, let C j be the union
of those closed balls that are contained in U j . Then C j is closed and C j ⊂ U j for
each j and {C1 , . . . , Cn+1 } cover Sn . By the L-S-B theorem above (13.35), some C j
contains a pair of antipodal points and then so does U j . 

The results of this section (Borsuk et al.) have several applications in various fields.
See, e.g. [28]. For example, there is a higher dimensional analogue of the dosa
theorem (idli theorem!) usually called the ham sandwich theorem [28, 31, 32].

Exercise 13

1. Fixed points. a) Find a continuous map on X without fixed points when X is

(0,1), [0,1), [0, 1] ∪ [2, 3].
b) If f : [0, 1] → [0, 1] is continuous and f ◦ f is the identity, then f has a
unique fixed point.
c) If X is a compact metric space and a continuous f : X → X has no fixed point,
then d(x, f (x)) > δ for all x and some δ > 0.
2. Brouwer’s theorem. If X has the fixed point property, then so has any retract
of X . Deduce the no retract theorem from the fixed point theorem. Thus Brouwer’s
theorem (13.16) and the No Retract Theorem (13.17) are equivalent. There are several
other equivalent topological, combinatorial statements and results from economics—
e.g. Poincaré-Miranda theorem (13.12), Bohl’s theorem (14.25), noncontractibility
of the sphere (14.24). References [20, 33, 34].
3. Knaster–Kuratowski–Mazurkiewicz. Let X ⊂ Rn and let κ : X → P(X )
be a KKM map, i.e. for every finite F ⊂ X , conv(F) ⊂ κ(F). If κ(x) is closed for
1 κ(x j ) is empty, let K be the convex
each x ∈ X , then {κ(x) : x ∈ X } has f.i.p. (If ∩m
202 13 Combinatorial Methods in Euclidean Topology


hull of the x j , d j := d(x j , κ(x j )), d = κ(x j )d j > 0. Apply Brouwer on K to
f (x) = t j x j , t j = d j /d.)
4. Subdivisions of a 2-simplex. Let S = S0 be an open 2-simplex in R2 and let
Sn+1 be the open 2-simplex with vertices at the midpoints of the sides of Sn , n ≥ 1.
Find ∩Sn .
5. A result of Hadamard, 1910 [5]. Let f : Bn → Rn be continuous and let
f = ( f 1 , . . . , f n ). If  j x j f j (x) ≥ 0 for x ∈ Sn−1 , then f (x) = 0 for some x ∈ Bn .
Deduce this from Brouwer’s theorem (13.16). [35]
6. Generalised L-S-B theorem. If Sn is covered by A1 , . . . , An+1 each of which
is either open or closed, then ±x ∈ A j for some x and some j.
7. Real functions on the sphere. If f j , 1 ≤ j ≤ n are real continuous functions
on Sn , there is an x ∈ Sn such that f j (x) = f j (−x) for all j.
8. Integer lattices. Are Zn and Zm homeomorphic for any n, m?
9. Homeomorphisms of balls and spheres. If either Sn and Sm are homeo-
morphic or if Bn and Bm are homeomorphic, then m = n.
10. Dimension invariance from Borsuk–Ulam. Use Borsuk–Ulam (13.34) to
prove that Rn and Rm are not homeomorphic for m
= n. In fact, there is no injective
continuous map Rn → Rm for n > m.
11. Interior and boundary. If E ⊂ Rn and h : E → Rn is a homeomorphic
embedding, then x∈ int E implies h(x)
∈ int h(E) and x ∈ ∂ E implies h(x) ∈ ∂h(E).
12. Embedding a sphere in Euclidean space. (a) Sn has no embedding in Rm
for n ≥ m.
b) Sn is not homeomorphic to any proper subset of itself.

13.7 Biographical Notes

13.7.1 Bohl

Piers Bohl (1865–1921) was a Latvian mathematician who worked in topology,

differential equations, quasi-periodic functions and uniform distribution. He proved
the no retract theorem and the Brouwer fixed point theorem, in his study of quasi-
linear differential equations, seven years before Brouwer. He was ahead of his times
in this and the results did not attract attention they deserved. He was a Professor at
Riga Polytechnic Institute. He was a highly rated chess player and introduced the
‘Riga Variation’ of the Ruy Lopez. It is said that he had no family and lived only for
science, indifferent to fame.
13.7 Biographical Notes 203

13.7.2 Hadamard

Jacques Hadamard (1865–1963) was a doyen of French mathematics in the last

part of the 19th century and in the first half of the 20th century. His teacher was
the renowned Émile Picard and his students included several stars—André Weil,
Maurice Fréchet, Paul Lévi, Szolem Mandelbrojt. It is impossible here to convey
the influence he had on the mathematics of his time or the range of his contribu-
tions. His thesis of 1892 was on complex function theory and ‘was one of the first
to examine the general theory of analytic functions, in particular his thesis con-
tained the first general work on singularities’. His work on entire functions and zeta
functions (1892), the proof of the Prime Number Theorem (1896) (which was also
proved independently by Charles de la Vallée Poussin in the same year), the
paper on dynamic trajectories on surfaces and geodesics (1896), his determinant
inequality (1893), his work on partial differential equations of outstanding impor-
tance, his work on geodesics on surfaces of negative curvature (1898) that led to
the development of symbolic dynamics, his introduction of well-posed problems,
his work on calculus of variations (1910) are all of great value. He also contributed
to elasticity, geometrical optics, hydrodynamics, probability and Markov chains.
His famous book Lectures on Cauchy’s problem in linear partial differential equa-
tions was published in 1922. He wrote the book The psychology of invention in
the mathematical field in 1945. Hadamard Three Circle Theorem, Hadamard Three
Lines Theorem, Hadamard Product Theorem on entire functions, Hadamard deter-
minant inequality, Hadamard matrices, Hadamard product of matrices, Hadamard
code, Hadamard dynamical system, Hadamard space, Hadamard transform, Cartan-
Hadamard manifold, Cartan-Hadamard theorem on complete Riemannin manifolds,
Cauchy-Hadamard theorem on radius of convergence of power series, Hadamard’s
method of descent in PDE, Hadamard regularisation are some that bear his name.
Another important work of his was to edit the collected works of Poincaré, not an
easy task. Laurent Schwartz, Fields medalist, said on Hadamard’s centenary “
I believe that he had a fantastic influence on his time, and that all living analysts
were shaped by him, directly or indirectly." (https://mathshistory.st-andrews.ac.uk/
Biographies/Hadamard/)

13.7.3 Ulam

Stanisław Ulam (1909–1984) Polish mathematician and nuclear scientist got his
doctoral degree in mathematics under Kuratowski. The famous Scottish Café in
Lwów was the place where many famous Polish mathematicians met to discuss
mathematics. Ulam was a member of the Lwów School of Mathematics and an active
contributor to the Scottish Book of problems. In mathematics, he contributed to
several areas like set theory, measure theory, topology, ergodic theory, group theory,
number theory, combinatorics and graph theory. He visited the USA many times on
204 13 Combinatorial Methods in Euclidean Topology

the invitation of von Neumann. During World War II, he worked on the Manhattan
Project at the Los Alamos Laboratory. After the war, he solved the problem of
how to initiate fusion in thermo-nuclear weapons (Teller-Ulam design) and devised
the ’Monte-Carlo method’ widely used in solving mathematical problems using
statistical sampling. Nuclear propulsion, joint work with Enrico Fermi et al. that led
to the creation of nonlinear science, contributions to theoretical biology are some
of the other things he is known for. In mathematics, Borsuk–Ulam theorem, Mazur-
Ulam theorem, Kuratowski-Ulam theorem, Ulam’s packing conjecture are among
his contributions.

13.7.4 Sperner

Emmanuel Sperner (1905–1980) was a German mathematician. His famous

Lemma on colourings of subdivisions of a simplex that led to a simple proof (by
Knaster–Kuratowski–Mazurkiewicz) of the Brouwer fixed point theorem. Sperner
himself used it to a give a proof of a result of Lebesgue in dimension theory. Another
important contribution is Sperner’s theorem in discrete mathematics. It describes the
largest possible families of finite sets none of which contain any other sets in the
family. It is considered as a central result in extremal set theory. Both were published
in 1928 when he was 22.

13.7.5 Mazurkiewicz

Stefan Mazurkiewicz (1888–1945) was a mathematician from Poland who worked

in Topology, Analysis and Probability. His mentor was Wacław Sierpiński and
he had a galaxy of famous students—Karol Borsuk, Bronisław Knaster, Kaz-
imierz Kuratowski, Stanisław Saks and Antoni Zygmund. His paper on inde-
composable continua is considered one of the most elegant pieces of work in gen-
eral topology. Hahn-Mazurkiewicz theorem is a basic result on curves. He, along
with Kuratowski and Knaster, gave a simple proof of Brouwer’s fixed point theo-
rem using Sperner’s lemma.

13.7.6 Brouwer

Luitzen Egbertus Jan Brouwer (1881–1966) was a Dutch mathematician and

philosopher based in Amsterdam. He was one of the great topologists in the early
part of the twentieth century. Later he turned to ‘constructive’ mathematics, what he
called intuitionism, as opposed to formalism. Just as 1905 was the annus mirabilis for
Einstein, the two years between 1910–1912 were miraculous years for Brouwer (and
topology). He obtained a series of remarkable, fundamental results—fixed point theo-
rem, invariance of dimension, singularities of vector fields, Jordan theorem in higher
13.7 Biographical Notes 205

dimensions, etc.- and introduced new methods and tools (e.g. simplicial approxi-
mation, degree of a map) that were often compared to a magic wand. His home
near Amsterdam became the focal point where leading topologists (Vietoris, Hopf,
Hurewicz, Alexandroff et al.) met and discussed topology. This burst of brilliance
was perhaps inspired and triggered by Brouwer’s meeting with Poicaré in Paris in
1910 where he also met Borel and Hadamard. For these and more on the remarkable
years, see Freudenthal’s writing in [36] and the article of James [37].
In addition to proving ground-breaking theorems, his methods have also become
standard tools in the subject. He was killed in 1966 in an accident, struck by a vehicle
while crossing the street in front of his house. Brouwer fixed point theorem, Brouwer
degree, Brouwer theorems on invariance of domain and dimension are among those
giving him mathematical immortality.

References

1. Poincaré, H.: Sur certaines solutions particulieres du probleme des trois corps. C. R. Acad.
Sci. Paris 97, 251–252 (1883)
2. Poincaré, H.: Sur certaines solutions particulieres du probleme des trois corps. Bull.
Astronomique 1, 63–74 (1884)
3. Bohl, P.: Über die Bewegung eines mechanischen Systems in der Nähe einer Gleichgewicht-
slage. J. Reine Angew. Math. 127, 179–276 (1904). https://doi.org/10.1515/crll.1904.127.
179
4. Brouwer, L.E.J.: Zur Analysis Situs. Math. Ann. 68, 422–434 (1910) (Collected Works, vol.
2, 354–366.). https://doi.org/10.1007/BF01475781
5. Hadamard, J.H.: Sur quelques applications de l’indice de Kronecker, Introduction to: Tannery.
J. La Theorie des Fonctions d’une Variable, Hermann, Paris (1910)
6. Brouwer, L.E.J.: Beweis der Invarianz der Dimensionenzahl. Math. Ann. 70, 161–165 (1911)
(Collected Works, vol. 2, 430–434.). https://doi.org/10.1007/BF01461154
7. Knaster, B., Kuratowski, K., Mazurkiewicz, S.: Ein Beweis des Fixpunksatzes für n-
dimensionale Simplex. Fund. Math. 14, 132–137 (1929). https://doi.org/10.4064/fm-14-1-
132-137
8. Sperner, E.: Neuer Beweis für die Invarianz der Dimensionszahl und des Gebiets Math. Sem.
Hamburg. Univ. 6, 265–272 (1928). https://doi.org/10.1007/BF02940617
9. De Loera, J.A., Goaoc, X., Meunier, F., Mustafa, N.H.: The discrete yet ubiquitous theorems
of Carathéodory, Helly, Sperner, Tucker and Tverberg Bull. Am. Math. Soc. 56, 415–511
(2019). https://doi.org/10.1090/bull/1653
10. Ivanov, N.V.: Sperener’s lemma, the Brouwer fixed point theorem and cohomolgy (2019).
arXiv: 09065193v2
11. Park, S.: Ninety years of the Brouwer fixed point theorem. Vietnam J. Math. 27, 187–222
(1999)
12. Kuhn, H.W.: Some combinatorial lemmas in topology. IBM J. Res. Devel. 4, 518–524 (1960).
https://doi.org/10.1147/rd.45.0518
13. Kulpa, W.: The Poincaré-Miranda Theorem. Am. Math. Monthly 104, 545–550 (2018). https://
doi.org/10.2307/2975081
14. Müger, M.: A remark on the invariance of dimension. Math. Semesterber. 62, 59–68 (2015)
15. Browder, F.E.: Fixed point theory and nonlinear problems. Bull. Am. Math. Soc. 9, 1–39
(1983). https://doi.org/10.1090/S0273-0979-1983-15153-4
206 13 Combinatorial Methods in Euclidean Topology

16. Poincaré, H.: Sur les courbes définies par une équation difféerentielle IV. J. Math. Pures Appl.
85, 151–217 (1886)
17. Miranda, C.: Un’ osservazione su una teorema di Brouwer. Boll. Unione Mat. Ital. 3, 527
(1940)
18. Borsuk, K.: Sur les rétractes. Fund. Math. 17, 152–170 (1931). https://doi.org/10.4064/fm-
17-1-152-170
19. Borsuk, K.: Theory of Retracts. Monografie Mat, Warsaw (1967)
20. Granas, A., Dugundji, J.: Fixed Point Theory, Monographs in Mathematics. Springer (2003)
21. Shashkin, Y.A.: Fixed Points. American Mathematical Society-Mathematical Association of
America (1991)
22. Kulpa, W.: Poincaré and domain invariance theorem. Acta Univ. Carol. Math. et Phys. 39,
127–136 (1998)
23. Tao, T.: Hilbert’s Fifth Problem and Related Topics. Graduate Studies in Mathematics, vol.
153. Americian Mathematical Society (2014)
24. Borsuk, K.: Drei Sätz über die n-dimensional euklidische Sphäre. Fund. Math. 20, 177–190
(1933). https://doi.org/10.4064/fm-20-1-177-190
25. Tucker, A.W.: Some topological properties of disk and sphere. In: Proceedings of First Cana-
dian Mathematical Congress, Montreal, pp. 285–309. University of Toronto Press, Toronto
(1946)
26. Fan, K.: A generalization of Tucker’s combinatorial lemma with topological applications.
Ann. Math. 2(56), 431–437 (1952). https://doi.org/10.2307/1969651
27. Prescott, T., Su, F.E.: A constructive proof of Ky Fan’s generalization of Tucker’s lemma. J.
Combin. Theory Ser. A 111(2), 257–265 (2005). https://doi.org/10.1016/j.jcta.2004.12.005
28. Matoušek, J.: Using the Borsuk-Ulam Theorem. Springer (2008)
29. Lusternik, L., Schnirelmann, L.: Topological Methods in Variational Problems, (in Russian)
Moscow (1930)
30. Lusternik, L., Schnirelmann, L.: Topological methods in variational problems and their appli-
cations to the differential geometry of surfaces. Uspekhi Mat. Nauk 2(1), 166–217 (1947).
(in Russian)
31. Hatcher, A.: Algebraic Topology. Cambridge University Press (2003)
32. Spanier, E.: Algebraic Topology. Tata-McGraw-Hill (1966)
33. Idzik, A., Kulpa, W., Maćkowiak, P.: Equivalent forms of the Brouwer fixed point theorem I.
Top. Methods Nonlinear Anal. 44, 263–276 (2014)
34. Kuratowski, K.: Topology. Academic Press (1966)
35. Mawhin, J.: Simple Proofs of the Hadamard and Poincaré-Miranda theorems using the
Brouwer fixed point theorem. Am. Math. Monthly 126, 260–263 (2019). https://doi.org/10.
1080/00029890.2019.1551023
36. Brouwer, L.E.J.: In: Freudenthal, H. (ed.) Collected Works, vol. 2, North-Holland (1976)
37. James, I.M.: From combinatorial topology to algebraic topology. In: James, I.M. (ed.) History
of Topology, pp. 561–574. Elsevier (1999)
Chapter 14
Homotopy

Homotopy, a concept that was introduced by Poincaré, may be described as con-

tinuous deformation. It is extremely important in modern mathematics. Some basic
ideas are presented here. Here is how Brouwer describes homotopy in a footnote in
his paper [1].
Under a continuous modification of a univalent continuous transformation we
understand in the following always the construction of a continuous series of univa-
lent continuous transformations, i.e. a series of transformations depending in such
a manner on a parameter, that the position of an arbitrary point is a continuous
function of its initial position and the parameter.
Topology, as an individual discipline, hardly made its presence felt prior to the
work of Poincaré. Experts generally agree that topology, as a subject, ‘had arrived’
after the appearance of his Analysis situs. Among other things, homotopy, fundamen-
tal group, simplicial methods and homology made their first appearance in Analysis
situs. For more on these aspects, see [2, 3].
Showing that R and R2 are not homeomorphic is easy, as we have seen—just use
connectedness. But what about R2 about R3 ? Such simple arguments do not work. In
fact, as pointed out in the last chapter, the question is far from simple for general Rn
and Rm . What about the spheres Sn and Sm or tori Tn and Tm or even the sphere S2
and the torus T2 ? If even for familiar, concrete spaces, the question is so deep, what
about general spaces? How to tackle the fundamental problem in topology: given
two spaces X and Y , how to decide whether or not they are homeomorphic?
Tools to tackle such questions were introduced and developed first by the versatile
French genius Henri Poincaré in a series of papers [4–10] on Analysis situs (the old
name for topology) beginning in 1895. Homotopy is one of them. It is a part of
the subject called Algebraic Topology whose founding father was Poincaré. The
underlying philosophy of the subject is to associate an algebraic object (e.g. a group)
to a topological space in such a way that homeomorphic spaces have isomorphic
associated algebraic objects. This way, a topological problem is transformed into an
algebraic one, which, hopefully (at least in some cases) would be easier to handle.
Various ways to get such an association have been developed—homotopy, homology,
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 207
K. Parthasarathy, Topology, La Matematica per il 3+2 134,
https://doi.org/10.1007/978-981-16-9484-4_14
208 14 Homotopy

etc. (It is interesting to note that study of general topology came into being much later;
e.g. Hausdorff’s classic Grundzüg der Mengenlehre [11] appeared only in 1914; the
definition of a metric appeared first in 1906 in Fréchet’s thesis [12].
We will have a glimpse of algebraic topology (in the form of homotopy) in this
and the next chapter. In the previous chapter we had looked at a bit of combinatorial
topology. For a delightful, lucid invitation to differential topology from the master,
see Milnor [13].

14.1 Retracts and Deformation Retracts

Before looking at homotopy, we discuss the related, useful concepts of retracts and
deformation retracts.
Étant donné un ensemble B ⊂ A, j’appelle fonction rétrancte l’ensemble A en B,
toute fonction f intrinsèque et continue dans A, telle que B = f (A) et qui remplit
l’équation fonctionelle f ( f (x)) = f (x) Borsuk [14].
‘Given a set B ⊂ A, I call a retraction function of A in B every function defined
and continuous on A such that B = f (A) and which satisfies the functional equation
f ( f (x)) = f (x)’.
Thus did Borsuk introduce the concept of retraction in his thesis in 1931. (The
term ‘retract’ was suggested by his thesis advisor Mazurkiewicz.) This and related
concepts have turned out to be quite important in topology. He also obtained most of
the basic properties of retracts that we present here. (See also Borsuk’s monograph
[15].)
Consider the function f defined on R by
⎧
⎨ 0, x ≤ 0
f (x) = x, 0 ≤ x ≤ 1
⎩
1, x ≥ 1

Then f : R → [0, 1] is continuous (look at the graph of f , or use pasting lemma!)

and f is identity on [0,1].
Exercise: Find another function with the same properties. How many such func-
tions can you find?
Such sets serve as a link between set topology and algebraic topology and are
important enough to warrant a name. We recall the definition given in the last chapter
in the context of no retract theorem.
Definition 14.1 E, a subset of a space X , is called a retract of X if there is a
continuous map r : X → E such that r | E = id E (i.e. id E has a continuous extension
to X ); such an r is called a retraction of X on E.
Exercise. A retract of a Hausdorff space is closed. Find an example of a retract
that is not closed. (Look at the examples listed below.)
14.1 Retracts and Deformation Retracts 209

r(x)
x r(x) r(x) x
r(x)

Fig. 14.1 Annulus

Fig. 14.2 Disc with two holes

Example 14.2 1. A single point set in any space is a retract.

2. A closed interval [a, b] is a retract but {0, 1} and [0, 1] ∪ [2, 3] are not retracts
of R. (Why?)
3. An annulus has a circle as a retract and two externally touching circles is a
retract of a disc with two holes (Figs. 14.1 and 14.2).
4. The sphere Sn is a retract of the punctured space Rn+1 \ (0), but not of the
closed ball Bn . (Do you think Sn is a retract of Rn+1 ?)
5. The closed unit ball in Rn is a retract of Rn . (Find a retraction!)
210 14 Homotopy

6. S L(n, R) is a retract of G L + (n, R)

7. In a discrete space, every nonempty set is a retract.
8. There are nontrivial spaces in which every closed set is a retract. For instance,
the Cantor set C. (This needs a proof.)
9. If rn is a retraction of R on [−1/n, 1/n], then (xn ) 
→ (rn (xn )) is a retraction
of the real sequence space 2 onto the Hilbert cube H = [−1/n, 1/n].

Retracts are closely tied up with the problem of extensions of maps.

Proposition 14.3 E ⊂ X is a retract of a space X if and only if any continuous
f : E → Y has a continuous extension to X for any Y.

Proof Suppose r : X → E is a retraction and let f : E → Y be continuous. Then

f ◦ r : X → Y is a continuous extension of f . Conversely, if the stated extension
property holds, then an extension of the identity map id E : E → E is a retraction. 

A stronger, useful concept is the following.

Definition 14.4 : Let X be a space and let E be a subset. Then E is called a
deformation retract of X if there is a continuous map, a deformation retraction,
H : X × [0, 1] → X such that
i) H (x, 0) = x for all x ∈ X ;
ii) H (y, t) = y for all y ∈ E, t ∈ [0, 1];
iii) H (x, 1) ∈ E for all x ∈ X .

Remark 14.5 : In this case, r (x) = H (x, 1) defines a retraction.

Example 14.6 1. A singleton set in a space X is not always a deformation retract.

For example, if X = {0, 1} with the topology T = {∅, {0}, X } then {0} is a retract,
but is not a deformation retract.

2. Write X = Rn+1 \ (0). Define H : X × [0, 1] → X defined by H (x, t) =

(1 − t)x + t xx
is a deformation retraction of Rn+1 \ (0) on Sn . (Geometrically,
what does H do?)
3. If E is a retract of Rn , then it is a deformation retract. In fact, if r : Rn → E is a
retraction, then H defined by H (x, t) = (1 − t)x + tr (x) is a deformation retraction.
Thus [a, b] is a deformation retract of R and the closed unit ball Bn in Rn is a
deformation retract of Rn .
4. Any compact convex subset is a deformation retract of Rn (Proof?).
5. The punctured plane is a retract of R3 \ z-axis (Fig. 14.3).
14.2 Homotopy of Maps and Paths 211

H(t,x)=(1-t)x+tr(x)

r(x)

r(x)

H(t,x)=(1-t)x+tr(x)

Fig. 14.3 Punctured plane: deformation retract of R3 \ {z − axis}

14.2 Homotopy of Maps and Paths

‘We shall say that two transformations belong to the same class, if they can be
transformed continuously into each other.’ Brouwer [1]
Homotopy of two maps was thus expressed explicitly for the first time, by
Brouwer. (‘Class’ in Brouwer’s definition is, of course, homotopy class.) Homo-
topy was introduced by Poincaré in his Analysis situs in 1895, but a definition had
to wait till 1912.
Historians of mathematics say that roots of homotopy of maps and paths can
be traced back to Lagrange’s work on calculus of variations and Cauchy’s work
on complex integration, see, e.g., [16]. The term ‘homotopie’ occurs first in Dehn–
Heegard [17]. Brouwer’s definition, in its modern garb:
Definition 14.7 Let X, Y be spaces. If f, g : X → Y are continuous, we say that f
is homotopic to g, written f ∼ g, if there is a continuous map F : X × [0, 1] → Y
such that f (x) = F(x, 0), g(x) = F(x, 1) for every x ∈ X . Such an F is called a
homotopy from f to g and we write F : f ∼ g. F gives a continuous deformation
of f to g.
Another, equivalent, way of looking at homotopy is this: Consider for each
t ∈ [0, 1], the continuous map f t : X → Y defined by f t (x) = F(x, t) for all
212 14 Homotopy

x ∈ X . This gives a continuous one parameter family f t , 0 ≤ t ≤ 1, of continuous

maps X → Y such that f 0 = f, f 1 = g.

Example 14.8 Let f, g : Rn → Rn be any two continuous maps. Define F : Rn

×[0, 1] → Rn by F(x, t) = (1 − t) f (x) + tg(x), which is the line segment joining
f (x) and g(x) for each x. Then F : f ∼ g and F is, naturally, called the straight
line homotopy. In particular, any continuous f : Rn → Rn is homotopic to a constant
map.

Proposition 14.9 On the set of continuous maps from a space X to a space Y ,

homotopy (i.e. ∼) is an equivalence relation.

Proof Let f, g, h : X → Y be continuous. Then

i) F : f ∼ f, where F : X × [0, 1] → Y is defined by F(x, t) = f (x) for every
x ∈ X and t ∈ [0, 1].
ii) F : f ∼ g ⇒ G : g ∼ f where G(x, t) = F(x, 1 − t) for x ∈ X, t ∈ [0, 1].
iii) F : f ∼ g, G : g ∼ h implies H : f ∼ h, where

F(x, t), 0 ≤ t ≤ 21
H (x, t) =
G(x, 1 − t), 21 ≤ t ≤ 1

for x ∈ X . Note that F(x, 1) = g(x) = G(x, 0), so continuity of H is a consequence

of the pasting lemma. 

Lemma 14.10 Let X, Y, Z be topological spaces, f, g : X → Y and h, k : Y → Z

be continuous maps. If f ∼ g, then h ◦ f ∼ h ◦ g; if h ∼ k, then h ◦ g ∼ k ◦ g.

Proof If F : f ∼ g, then, check that h ◦ F : h ◦ f ∼ h ◦ g. If H : h ∼ k, then define

G(x, t) = H (g(x), t), x ∈ X, t ∈ [0, 1]. Check the continuity of G and the condi-
tions to get that G : h ◦ g ∼ k ◦ g. 

Corollary 14.11 With the notation of the Lemma above, if f ∼ g and h ∼ k, then
h ◦ f ∼ k ◦ g.

Proof Immediate from the transitivity of ∼ and the lemma. 

Two homeomorphic spaces are ‘topologically equivalent’ or have the ‘same topo-
logical type’. Here is the homotopy counterpart.

Definition 14.12 Two spaces X, Y are said to be homotopically equivalent, or of the

same homotopy type if there exist two continuous maps f : X → Y and g : Y → X
such that g ◦ f ∼ id X and f ◦ g ∼ idY .

Exercise. Homotopy equivalence is an equivalence relation on topological spaces.

Exercise. Homotopy equivalence is preserved by products: If X and Y have the

same homotopy type and X  and Y  have the same homotopy type, then so do X × X 
and Y × Y  .
14.2 Homotopy of Maps and Paths 213

Remark 14.13 If X and Y are homeomorphic, then they are homotopically equiva-
lent (but not conversely, as we will see ere long).

Homotopically trivial spaces have an important role.

Definition 14.14 A space X is said to be contractible if it has the same homotopy

type as a point, i.e. a single point space. A map which is homotopic to a constant
map is said to be null homotopic.

Proposition 14.15 A space X is contractible if and only if id X is null homotopic.

Proof Suppose X is contractible, so there exist f : X → {y0 } and g : {y0 } → X

with g ◦ f ∼ id X . One part is proved because g ◦ f is the constant map x → x0 =
g(y0 ). Conversely, assume F : id X ∼ c where c is a constant map, say, c(x) ≡ x0 .
If f : X → {x0 } is the constant map and g : {x0 } → X is the inclusion map, then
g ◦ f = c ∼ id X and f ◦ g = id{x0 } . Thus X has the homotopy type of {x0 }. 

Corollary 14.16 A space X is contractible if and only if any two continuous maps
f, g : Y → X are homotopic for every space Y .

Proof If the stated condition holds then, in particular, id X is homotopic to a constant

map on X . Conversely, suppose X is contractible. Let Y be any space and let f, g :
Y → X be continuous maps. If id X ∼ c, where c is a constant map on X , then
f = id X ◦ f ∼ c ◦ f = c ◦ g ∼ id X ◦ g = g. 

Corollary 14.17 Every contractible space is path connected.

Proof Let X be contractible and let x0 , x1 ∈ X . Let c be the constant map c(x) ≡ x1
and let H : id X ∼ c. Then α(s) = H (x0 , s) defines a path in X joining x0 to x1 . 

Example 14.18 1. Two null homotopic maps need not be homotopic. In fact, even
two constant maps need not be homotopic. For instance, suppose x0 , x1 belong to two
different path components in X . Then the constant maps c0 (x) = x0 and c1 (x) = x1
are not homotopic.
2. Rn is contractible (Example 14.8). Thus, homotopically speaking, Rn , for n ≥ 1,
is indistinguishable from a point! Thus, homotopy equivalence is a rather crude
relation. It is highly valuable nevertheless.
3. The arguments used for Rn remain valid for any convex set in Rn yielding the
result any convex set in Rn is contractible.
4. Recall the comb space
C̄ = {(1/n, y) ∈ R2 : n ∈ N, 0 ≤ y ≤ 1} ∪ {0} × [0, 1] ∪ [0, 1] × {0}. Define F : C̄
× [0, 1] → C̄ by F((x, y), t) = (x, (1 − t)y). Then F : idC̄ ∼ p1 , the projection on
the x-axis and p1 is homotopic to the constant map 0. So idC̄ is homotopic to 0. Geo-
metrically, push each vertical segment to the x-axis and then push every point on the
x-axis segment to the origin along the x-axis. Thus the comb space is contractible.
214 14 Homotopy

The next couple of results are about homotopy of maps into the sphere and from the
sphere.
Proposition 14.19 Let X be any space. If f, g : X → Sn are continuous maps such
that f (x) and g(x) are not antipodal points for any x ∈ X , then f is homotopic to
g.
(1−t) f (x)+tg(x)
Proof Define F : X × [0, 1] → Sn by F(x, t) = (1−t) f (x)+tg(x)
. Note that the
assumption f (x), g(x) are never antipodal ensures that the denominator never van-
ishes. Then it is clear that F : f ∼ g. 

Corollary 14.20 If f : X → Sn is not surjective, then f is null homotopic.

Proof Take p ∈ Sn \ f (X ) and let c p be the constant map defined by c p (x) = − p.

Then f and c p are never antipodal. 

Proposition 14.21 A continuous map f : Sn → Y is null homotopic if and only if

f has a continuous extension F : Bn+1 → Y .

Proof Suppose H : f ∼ c where c is a constant map. Any point in x ∈ Bn+1

uniquely determines a t ∈ [0, 1] and an ξ ∈ Sn such that x = (1 − t)ξ . Then F(x) =
F((1 − t)ξ ) := H (ξ, t) defines the sought after extension of f . Note that F(0) =
H (ξ, 1) = c for all ξ .
Conversely, assume that f has an extension F to the unit ball. Define the map H :
Sn × [0, 1] → Y by H (ξ, t) := F((1 − t)ξ ). Then H (ξ, 0) = F(ξ ) = f (ξ ) and
H (ξ, 1) = F(0), hence H : f ∼ c0 where c0 is the constant map c0 (x) = F(0). 

Corollary 14.22 Any continuous map from Sn into a contractible space has a con-
tinuous extension to the closed unit ball Bn+1 .

Proof A consequence of 14.16 and 14.21 

Proposition 14.23 A continuous map Sn−1 → Sn−1 that is antipode preserving is

not nullhomotopic.

Proof If f : Sn−1 → Sn−1 is null homotopic, then it has a continuous extension

F : Bn → Sn−1 . Identifying Bn with Sn+ , F extends to Sn by defining it to be
−F(−x) for x ∈ Sn− . Note that this is well defined on Sn+ ∩ Sn− and gives a continuous
function Sn → Sn−1 that is antipode preserving, contradicting Borsuk’s theorem, 3)
of 13.33. 

Corollary 14.24 The identity map Sn → Sn is not null homotopic. Hence Sn is not
contractible.

Proof The identity map is antipodal and so the previous proposition applies. Alter-
natively, if idSn is null homotopic, it has a continuous extension Bn+1 → Sn , contra-
dicting the no retract theorem. 
14.2 Homotopy of Maps and Paths 215

Corollary 14.25 (Bohl’s theorem)

If f : Bn+1 → Rn+1 is continuous, then either f has a fixed point or f (x) = λx for
some x ∈ Sn and some λ > 1.
Proof Suppose f (x) = x holds for no x ∈ Bn+1 and f (y) = λy holds for no y ∈ Sn ,
λ > 1. Then y = t f (y) holds for no y ∈ Sn and 0 ≤ t ≤ 1, and F : Sn × [0, 1] → Sn
defined by
 y−2t f (y)
, 0 ≤ t ≤ 21
F(y, t) = y−2t f (y)
(2−2t)y− f ((2−2t)y)
(2−2t)y− f ((2−2t)y)
, 21 ≤ t ≤ 1.

gives a homotopy of the identity map on Sn with a constant. 

There is a relation between homotopy and deformation retracts.
Proposition 14.26 If E is a deformation retract of X , then E is homotopically
equivalent to X .
Proof The deformation retraction gives a retraction r : X → E. If ι : E → X is the
inclusion, then r ◦ ι = id E and ι ◦ r ∼ id X . 
Paths are maps, but a path homotopy is stronger than a map homotopy: it keeps the
end points fixed throughout the homotopy.
Definition 14.27 Let X be a space and let α, β : [0, 1] → X be paths in X joining
two points x0 , x1 in X . A path homotopy from α to β is a continuous map H :
[0, 1] × [0, 1] → X with the following properties: H (s, 0) = α(s), H (s, 1) = β(s)
for all s ∈ [0, 1] and H (0, t) = x0 , H (1, t) = x1 for all t ∈ [0, 1]. If such an H exists,
α and β are said to be path homotopic and we write α ∼β. ˙
The definition just means that H deforms the path α to the path β through a
continuous family αt of paths joining x0 and x1 , where αt (s) = H (s, t) so that α0 = α
and α1 = β (Fig. 14.4a, b).
Proposition 14.28 Path homotopy is an equivalence relation on the set of all paths
in X joining x0 and x1 .
Proof The proof is essentially the same as that for map homotopy. Additionally one
has to check that the end points are fixed by the homotopies. This offers no difficulties
and is left to the reader. 
Example 14.29 1. In Rn , any two paths α, β joining a point x0 to a point x1 are path
homotopic. In fact, the straight line homotopy given by H (s, t) = (1 − t)α(s) +
tβ(s) is a path homotopy.
2. Consider the paths joining (1,0) to (−1,0) in R2 \ (0) defined by α(s) =
(cos π s, sin π s), β(s) = (cos π s, − sin π s), 0 ≤ s ≤ 1. It is intuitively clear that α
can not be deformed into β keeping the end points fixed—the origin is an ‘obstruc-
tion’, we cannot get past the hole at the origin. The reader should get back to this
example after the next chapter to prove that α and β are, in fact, not path homotopic.
216 14 Homotopy

a
[0, 1] × {1} b
β

β
x1
[0, 1] × {t} x0
αt
αt
[0, 1] × {0} α
α

Fig. 14.4 Path homotopy

Two paths α from x0 to x1 and β from x1 to x2 can be combined in a natural way

to get a path from x0 to x2 : just follow α from x0 to x1 and then β from x1 to x2 . Here
is the mathematical definition:

Definition 14.30 Let X be a space and let x0 , x1 , x2 ∈ X . Suppose α, β are paths

in X joining x0 to x1 and x1 to x2 , respectively. Define α
β : [0, 1] → X to be the
path from x0 to x1 given by

α(2s), 0 ≤ s ≤ 21
α
β(s) =
β(2s − 1), 21 ≤ s ≤ 1.

Remark 14.31 Noting that at s = 21 , α(2s) = α(1) = x1 = β(0) = β(2s − 1), α

β is continuous by the pasting lemma. α

β describes α from x0 to x1 and β from
x1 to x2 , both at ‘double speed’.

Example 14.32 α(s) = (cos π s, sin π s), 0 ≤ s ≤ 1, is a path from (1,0) to (-1,0)
and β(s) = (cos π(s + 1), sin π(s + 1)), joins (-1,0) to (0,1). They are the upper
and lower semicircles, respectively, described in the counter clockwise. In this case,
α
β is a loop at (1,0), i.e. α
β(0) = (1, 0) = α
β(1) and is given by α
β(s) =
(cos 2π s, sin 2π s), 0 ≤ s ≤ 1. It describes a complete circle, starting from (1,0).

Where do the 2s in α and 2s − 1 in β come from? Simple: s → 2s is the ‘linear’

homeomorphism of [0, 21 ] onto [0,1] taking the end points to the respective end
points, and similarly for s → 2s − 1 and the intervals [ 21 , 1] and [0,1]. Variations
of this idea will be used again and again in the sequel and it will be convenient to
formalise it. For any two intervals [a, b] and [c, d] in R, with a < b, c < d there is a
unique homeomorphism ϕ = ϕabcd : [a, b] → [c, d] of the form ϕ(t) = pt + q with
ϕ(a) = c, ϕ(b) = d. Its graph is just the line joining (a, c) to (b, d). (Write down the
exact expression for ϕ!) We call ϕ the oriented affine map from [a, b] to [c, d]. We
write ϕab for ϕab01 mapping [a, b] to [0,1]. Thus α
β is given by α
β = α ◦ ϕ0,.5
on [0, 21 ] and = β ◦ ϕ.5,1 on [ 21 , 1]. (In fact, ϕ0,.5 (s) = 2s, ϕ.5,1 (s) = 2s − 1.)
Is the operation of composition of paths compatible with deformation? ‘Yes’, says
the following lemma.
14.2 Homotopy of Maps and Paths 217

x2

id × ϕ.5,1
β βt β
G x1

α αt α

F
ϕ0,.5 × id x0

Fig. 14.5 Path homotopy respects products

Lemma 14.33 Let α, α  be paths in a space X joining x0 x1 and β, β  be paths

˙  and β ∼β
joining x1 to x2 . If α ∼α ˙  , then α
β ∼α
˙ 
β
˙  and G : β ∼β
Proof If F : α ∼α ˙  , define H : [0, 1] × [0, 1] → X by

F(2s, t) = F(ϕ0,.5 (s), t), 0 ≤ s ≤ 21 , 0 ≤ t ≤ 1
H (s, t) =
G(2s − 1, t) = G(ϕ.5,1 (s), t), 21 ≤ s ≤ 1, 0 ≤ t ≤ 1

The pasting lemma guarantees the continuity of H . Moreover, H (0, t) = F(0, t) =

x0 , H (1, t) = G(1, t) = x2 and H (s, 0) = α
β(s) as

F(2s, 0) = α(2s), 0 ≤ s ≤ 21
H (s, 0) =
G(2s − 1, 0) = β(2s − 1), 21 ≤ s ≤ 1

Similarly H (s, 1) = α 
β  (s). All these together mean that H : α
β ∼α
˙ 
β .
The genesis of the proof is that ϕ0,.5 × id : (s, t) → (2s, t) expands the rectangle
[0, 21 ] × [0, 1] to the square [0, 1] × [0, 1] and then F acts on the square; similarly
for [ 21 , 1] × [0, 1] and G (Fig. 14.5). 
Now consider three paths in X , α from x0 to x1 , β from x1 to x2 and γ from x2
to x3 . Combining these would yield a path from x0 to x3 . But there are two ways of
combining these: (α
β)
γ and α
(β
γ ). Which is the ‘correct’ way to combine?
Are these related? Let us first write out these explicitly for our edification.

α
β(2s), 0 ≤ s ≤ 21
(α
β)
γ (s) =
γ (2s − 1), 21 ≤ s ≤ 1
⎧
⎨ α(4s), 0 ≤ s ≤ 14
= β(4s − 1), 41 ≤ s ≤ 21
⎩
γ (2s − 1), 21 ≤ s ≤ 1

α(2s), 0 ≤ s ≤ 21
(α
β)
γ (s) =
β
γ (2s − 1), 21 ≤ s ≤ 1
218 14 Homotopy
⎧
⎨ α(2s), 0 ≤ s ≤ 21
= β(4s − 2), 21 ≤ s ≤ 43
⎩
γ (4s − 3), 34 ≤ s ≤ 1

How are the expressions got? Look at α

β(2s) in [0, 21 ], for example:

α(ϕ0,.5 (2s)) = α(4s), 0 ≤ 2s ≤ 21
α
β(2s) =
β(ϕ.5,1 (2s)) = β(4s − 1), 21 ≤ 2s ≤ 1

Similarly for others. Manifestly, (α

β)
γ and α
(β
γ ) are not the same.
Observe that, in (α
β)
γ , γ is described at ‘double speed’, whereas α, β are
described at ‘quadrupled speed’; on the other hand, in α
(β
γ ), α is traversed at
‘double speed’ and the other two at ‘quadrupled speed’. In spite of this ‘set back’,
there is some good news: the two ways of combining three paths are the same, homo-
topically speaking! Before proving this, let us make a simple general observation that
would be helpful in other situations as well.
Lemma 14.34 Let α be a path in X joining x0 to x1 and let σ, τ : [0, 1] → [0, 1] be
continuous with σ (0) = τ (0) = s0 and σ (1) = τ (1) = s1 . If H : [0, 1] × [0, 1] →
X is defined by H (s, t) = α((1 − t)σ (s) + tτ (s)), then H = Hσ τ : α ◦ σ ∼α ˙ ◦ τ.
(Both are paths joining α(s0 ) and α(s1 ).)
Proof This is just a simple verification. Observe that H is continuous and then check
the required homotopy conditions. 
Remark 14.35 The proof works because (s, t) → (1 − t)σ (s) + tτ (s) is a path
homotopy from σ to τ .
Lemma 14.36 If α, β and γ are paths in a space X from x0 to x1 , x1 to x2 and x2
to x3 , respectively, then (α
β)
γ ∼α
˙
(β
γ ).
Proof Consider the map η : [0, 1] → [0, 1] got by pasting together the oriented
affinemaps [0, 14 ] → [0, 21 ], [ 41 , 21 ] → [ 21 , 43 ] and [ 21 , 1] → [ 43 , 1]:
⎧
⎨ 2s, 0 ≤ s ≤ 21
η(s) = s + 4 , 4 ≤ s ≤ 21
1 1
⎩ s+1
2
, 21 ≤ s ≤ 1.
⎧
⎨ α(4s), 0 ≤ s ≤ 41
α
(β
γ )(η(s)) = β(4s − 1), 41 ≤ s ≤ 21
⎩
γ (2s − 1), 21 ≤ s ≤ 1

=(α
β)
γ (s)
The last lemma yields the required homotopy Hid,η (Fig. 14.6a). 
Definition 14.37 For any point x in a space X , the constant path x at x is defined
by x (s) = x for all s ∈ [0, 1].
14.2 Homotopy of Maps and Paths 219

a b c
α β γ α ex0 α ←
α

α β γ α ex0

Fig. 14.6 a–c Product of classes: associativity, identity, inverse

Clearly, if we travel along x0 (i.e. we do not move at all, but stand still at x0 ), and
then travel from x0 to x1 along α, effectively we have travelled only along α. Path
homotopy respects this naive description.

Lemma 14.38 Let α be a path in a space X from x0 to x1 . Then x0

α ∼α
˙ and
α
x1 ∼α.
˙

Proof Observe that x0

α(s) = α(σ (s)), where

0, 0 ≤ s ≤ 21
σ (s) =
2s − 1, 21 ≤ s ≤ 1

Noting that σ (0) = 0, σ (1) = 1, apply 14.34 taking τ = id. In fact, if H (s, t) =
α((1 − t)σ (s) + ts), then H : x0
α ∼α.
˙ Write out the analogous arguments show-
˙ (Fig. 14.6b).
ing α
x1 α 

For any path from x0 to x1 , there is a natural reverse path from x1 to x0 ‘described
at the same speed’.
←
Definition 14.39 For a path α from x0 to x1 in a space X , the reverse path α from
←
x1 to x0 is defined by α (s) = α(1 − s), s ∈ [0, 1].

The expected result holds: homotopy respects reversing of paths.

← ←
˙ then α ∼
Lemma 14.40 For paths α, β in a space X , if α ∼β, ˙ β.

Proof This is easy and predictable: just reverse each member of the family of paths
← ←
deforming α to β - the resulting continuous family of paths deforms α to β . In other
← ← ← ←
˙ then H : α ∼
words, if H : α ∼β, ˙ β , where H (s, t) = H (1 − s, t) for all s, t ∈ [0, 1].


←
Travelling from x0 to x1 along α and then back from x1 to x0 along α is equivalent
to standing still at x0 !
220 14 Homotopy

← ←
Lemma 14.41 If α is a path from x0 to x1 in a space X , then α
α ∼
˙ x0 and α

α ∼
˙ x1 .
←
Proof α
α (s) = α(σ (s)) and x0 (s) = α(τ (s)) where τ ≡ 0 and σ : [0, 1] →
[0, 1] is given by 
2s, 0 ≤ s ≤ 21 ,
σ (s) =
2 − 2s, 21 ≤ s ≤ 1,

←
hence H = Hσ τ : α
α ∼˙ x0 (14.34). Explicitly, H (s, t) = α((1 − t)σ (s)). Simi-
larly write down the other part (Fig. 14.6c). 

Exercise 14

1. Retracts. a) A retract of a retract is a retract: if A ⊂ B ⊂ X , if A is a retract

of B and B is a retract of X , then A is a retract of X .
b) If E is retract of a Hausdorff space, then E is closed. Find a retract that is not
closed.
c) True or false? Any closed convex set in Rn is a retract of Rn .
d) A retract of a locally compact Hausdorff
 space is locally
 compact.
e) If E i is a retract of X i for each i, E i is a retract of X i .
f) Any line in the plane is a retract of R2 .
g) Find retractions of Rn on Rk , k < n, and of 2 onto the subspace M = {(xn )
∈ 2 : x2n = 0, n ∈ N}.
h) Any vector subspace of Rn is a retract of Rn .
2. Null homotopy. a) Null homotopic maps may not be homotopic.
b) A continuous map f : X → Y is null homotopic if and only if it has a contin-
uous extension to Cone (X ) → Y .
3. Loops. a) A loop α at x0 ∈ X can be considered as a map f α : S1 → X with
f α (1) = x0 and conversely. They are related via α(s) = f α exp(2πıs). Two such
loops α, β are path homotopic if and only if f α , f β are homotopic by a homotopy F
with F(1, t) = x0 for all t.
b) Write down explicitly the loop α at 1 winding around the unit circle S1 thrice,
twice in the positive sense and once in the negative sense. Find a loop homotopy
from α to the simple loop β given by β(s) = ex p(2πıs), s ∈ [0, 1].
4. a) Paths in open sets. Let α be a path in an open set  ⊂ Rn with δ :=
inf 0≤s≤1 d(α(s), ∂) > 0. If β is a path with the same end points and d(β(s), α(s)) <
δ for all s, then α, β are path homotopic.
b) Every path in an open set in Rn is path homotopic to a polygonal path, i.e. a
path made up of finitely many line segments.
5. Space of paths Let X be a compact metric space, a, b ∈ X and P(a, b) be the
set of all paths in X from a to b. ρ(α, β) := sup{d(α(t), β(t)) : 0 ≤ t ≤ 1} defines
14.3 Biographical Notes 221

a metric on P(a, b). Paths α, β ∈ P(a, b) are path homotopic if and only if they lie
in the same path component of P(a, b) (from [18]).
6. Homotopy and fixed points. Is the fixed point property a homotopy invariant?

7. Map homotopy. a) If f, g : X → Y are homotopic then, for E ⊂ X , f | E is

homotopic to g | E
.
b) f, g : X → Yi are homotopic if and only if pi ◦ f is homotopic to pi ◦ g for
every i.
8. Homotopy classes. For spaces X, Y , let [X, Y ] denote the homotopy classes
[ f ] of continuous maps f from X to Y .
a) Let ψ : Y → Z . Then, for any space X , ψ induces maps ψ : [X, Y ] → [X, Z ]
defined by ψ [ f ] = [ψ ◦ f ] and ψ  : [Z , X ] → [Y, X ] given by ψ  [g] = [g ◦ ψ].
b) With obvious notation, (ψ ◦ φ) = ψ ◦ φ and (ψ ◦ φ) = φ  ◦ ψ  .
c) ψ = φ and ψ  = φ  if ψ ∼ φ.
d) [X, Y ] is a single element set if either Y is contractible or X is contractible and
Y is path connected.

14.3 Biographical Notes

14.3.1 Borsuk

Karol Borsuk (1905–1982) was a famous topologist from Poland. Retracts were first
defined by him in his thesis in 1931. He also introduced absolute neighbourhood
retracts, which have turned out to be rather important. During World War 2, he
taught at the underground university (along with Kuratowski) in Warsaw. During the
period 1931–1932 he did postdoctoral studies with leading European topologists of
the time, Menger in Wien(=Vienna), Heinz Hopf in Zürich and Vietoris in Innsbruck.
He was the leader of the Warsaw school of topology which was greatly influenced by
his wonderful insight. See [2]. He founded the modern branch of topology known as
shape theory with his celebrated paper of 1968 on homotopic properties of compacta.
He had spent time in leading American universities like Princeton and University of
California, Berkeley. Borsuk–Ulam theorem, Borsuk antipodal theorem and Borsuk
conjecture are well-known in topology. His teacher was the well-known topologist
Mazurkiewicz and the outstanding topologist Samuel Eilenberg was his student.

References

1. Brouwer, L.E.J.: Continuous one-one transformations of surfaces in themselves. Proceedings

of Koninklijke Nederlandse Akademie van Wetenschappen 15, 352–360 (1912) (Collected
Works, Vol. 2, 527–535)
222 14 Homotopy

2. James, I.M.: From combinatorial topology to algebraic topology. In: James, I.M. (ed.) History
of Topology, pp. 561–574. Elsevier (1999)
3. Sarkaria, K.S.: The topological work of Henri Poincaré. In: James, I.M. (ed.) History of Topol-
ogy, pp. 123–167. North-Holland (1999)
4. Poincaré, H.: Analysis situs. J. de l’École Polytech. 2(1) 1–123, Oeuvres; vol. VI, 193–288
(1895)
5. Poincaré, H., Complèment à l’Analysis situs, Rend. Circ. Math. d. Palermo, 13, 285–343;
Oeuvres, vol. VI, 290–337 (1899)
6. Poincaré, H.: Second Complèment à l’Analysis situs. Proc. Lond. Math. Soc. 32, 277–308
(1900). https://doi.org/10.1112/plms/s1-32.1.277
7. Poincaré, H., Cinquième Complément à l’Analysis Situs, Rend. Circ. Math. d. Palermo, 18,
45–110; Oeuvres, Vol. VI, 435–498 (1904)
8. Poincaré, H.: Analyse des travaux scientifiques de Henri Poincaré faite par lui-meme (1901).
Acta Math. 38, 1–135 (1921). https://doi.org/10.1007/BF02392063
9. Poincaré, H., Oeuvres, Tome VI, Garniers, R., Leray, J. (eds.) Gauthiers-Villars (1953)
10. Poincaré, H.: Papers on Topology: Analysis Situs and Its Five Supplements. Translated by J.
Am. Math. Soc, Stillwell (2010)
11. Hausdorff, F. Grundzüge der Mengenlehre. Leipzig (1914)
12. Fréchet, M.: Sur quelques point du calcul fonctionnel. Rendiconti di Palermo 22, 1–74 (1906).
https://doi.org/10.1007/BF03018603
13. Milnor, J.: Topology from the Differentiable Viewpoint. University of Virginia Press (1965)
14. Borsuk, K.: Sur les rétractes. Fund. Math. 17, 152–170 (1931). https://doi.org/10.4064/fm-17-
1-152-170
15. Borsuk, K.: Theory of Retracts. Monografie Mat, Warsaw (1967)
16. Dieudonné, J.A.: The work of Nicholas Bourbaki. Am. Math. Monthly 77, 134–145 (1970).
https://doi.org/10.1080/00029890.1970.11992437
17. Dehn, M., Heegaard, P.: Analysis situs. Encyclopedia der mathematischen Wissenschaften mit
Einschluss ihrer Anwendungen, vol. 3, pp. 153–220. Halfte I, Teubner, Leipzig, Teil I (1907)
18. Gamelin, T.W., Greene, R.E.: Topology. Dover (1999)
Chapter 15
Fundamental Groups and Covering
Spaces

The fundamental group is a most basic invariant in topology and is of fundamental

importance. It was defined and studied by Poincaré in his Analysis situs [1] in 1895
and its fifth supplement [2]. His preamble before he introduces the fundamental
group goes thus:
“... Géométrie est l’art de bien raisonner sur des figures mal faites; encore ces
figures, pour ne pas nous tromper, doivent-elles satisfaire à certaines conditions ;
les proportions peuvent être grossièrement altéré, mais les positions relatives des
diverses parties ne doivent pas être bouleversées.”
“La Géométrie, en effet, ... est avant tout l’étude analytique d’un groupe...”
“Le groupe G peut donc servir á définir la forme de la surface et s’appeler le
groupe de la surface. Il est clair que si deux surfaces peuvent se transformer l’une
dans l’autre par voie de déformation continue, leurs groupes sont isomorphes.”
(Poincaré, 1895, [1])
“... geometry is the art of reasoning well from badly made figures; however, these
figures, if they are not to deceive us, must satisfy certain conditions; the proportions
can be altered grossly, but the relative positions of the various parts must not be
upset.”
“Geometry, in fact, ... is, above all, the analytic study of a group...”
“The group G can then help to define the form of the surface and it is called
the group of the surface. It is clear that if two surfaces can be transformed into
each other by continuous deformations, then their groups are isomorphic.” (Author’s
translations)
As mentioned earlier, this group is of central importance in mathematics. For a
recent manifestation of this importance, see the note on the famous Poincaré conjec-
ture on the sphere in the biographical sketch at the end of the chapter.

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 223
K. Parthasarathy, Topology, La Matematica per il 3+2 134,
https://doi.org/10.1007/978-981-16-9484-4_15
224 15 Fundamental Groups and Covering Spaces

15.1 The Fundamental Group

The appearance of Analysis situs marks, by general consensus, the birth of Algebraic
Topology. It contained Poincaré’s construction of homology theory and the funda-
mental group. (For translations of all the parts of Analysis situs, with commentary,
see [3].) The fundamental group, with its remarkable abstract structure, has turned
out to be really of central importance in topology and elsewhere. Here is how the
first definition appeared.
Cela posé, il est clair que l’on peut imaginer un groupe G satisfaisant aux con-
ditions suivantes:
1. A chaque contour fermé M0 B M0 correspondra une substitution S du groupe;
2. La condition nécessaire et suffisante pour que S se réduise à lu substitution
identique, c’est que M0 B M0 ≡ 0;
3. Si S et S  correspondent aux contours C et C  et si C  ≡ C + C  , la sub-
stitution correspondant á C  sera SS  . Le groupe G s’appellera le groupe
fondamental de la variété V .
“La Science dont l’objet est l’étude de ce groupe et de quelques autres analogues
a reçu le nom d’Analysis situs.” [1]
Poincaré’s definition may be translated as follows.
“Given this, it is clear that we can think of a group G satisfying the following
conditions:
1. Each closed contour M0 B M0 corresponds to a substitution S of the group.
2. The necessary and sufficient condition for S to reduce to the identity substitution
is that M0 B M0 ≡ 0.
3. If S and S  correspond to contours C and C  and if C  ≡ C + C  then the
substitution corresponding to C  is SS  . The group G is called the fundamental
group of the manifold V.”
“The science whose object is the study of this group and other analogues get the
name Analysis situs.”
We are going to define the fundamental group essentially as above—as classes of
loops at a point. All the ground work needed for this has been carried out in the last
chapter. We need to just specialise the results on path homotopy obtained there to the
case of closed paths or loops. This immediately leads us to the fundamental group.
Definition 15.1 A path α in a space X is called a loop at x0 if α(0) = α(1) = x0 .
π1 (X, x0 ) is the set of loop homotopy classes at x0 .
Let us examine what we get if the lemmas of the last chapter are applied to loops at
x0 . An alert reader might have already noticed the emergence of a familiar structure
there. Note that if α, β are loops at x0 , then so is α  β. For a loop α at x0 , let [α]
denote its loop homotopy class. Thus π1 (X, x0 ) = {[α] : α is a loop at x0 }.
   
Lemma 14.33 says that if [α] = [α ] and [β] = [β ], then [α  β] = [α  β ]. This
assures the validity of the following natural definition.
15.1 The Fundamental Group 225

Definition 15.2 Let X be a space and let x0 ∈ X . For [α], [β] in π1 (X, x0 ), define
[α]  [β] = [α  β].
Let us see what the remaining lemmas say about this binary operation on
π1 (X, x0 ): Lemma 14.36 says that the binary operation  is associative. Lemma
14.38 implies that x0 , the constant loop at x0 , acts as the identity element. Finally,
←
the content of Lemma 14.40 is that [ α ], the class of the reverse path of α, is the
inverse of [α] for the operation . The following basic result paraphrases all these
observations.

Theorem 15.3 π1 (X, x0 ) is a group under . 

π1 (X, x0 ) is called the fundamental group of X with base point x0 or of the pointed
space (X, x0 ). The subscript in π1 indicates that perhaps there are more things—the
πn —to come. This is indeed true—one has the homotopy groups πn (X ) and the
fundamental group is just the first of these sequence of groups. However, we will not
be discussing the higher homotopy groups. See [4–6].
One usually considers only path connected spaces while studying fundamental
groups. The following simple result tells us why.

Proposition 15.4 Let x0 , x1 be any two points in X . If there is a path in X from x0 to

x1 , then π1 (X, x0 ) is isomorphic to π1 (X, x1 ). More explicitly, any path σ in X from
x0 to x1 induces an isomorphism σ  : π1 (X, x0 ) → π1 (X, x1 ), defined by σ  ([α]) :=
←
[ σ  α  σ ], of π1 (X, x0 ) onto π1 (X, x1 ). In particular, if X is path connected, then
π1 (X, x0 ) is isomorphic to π1 (X, x1 ) for any pair of points x0 , x1 ∈ X .
←
Proof The proof is routine. Note first that σ  α  σ is a loop at x1 and so σ  maps
π1 (X, x0 ) in to π1 (X, x1 ). For loops α, β at x0
σ  ([α])  σ  ([β])
← ←
= [σ  α  σ ]  [σ  β  σ ]
← ←
= [ σ ]  [α]  ([σ ]  [ σ ])  [β]  [σ ]
←
= [ σ ]  ([α]  [ex1 ])  [β]  [σ ]
←
= [ σ ]  ([α]  [β])  [σ ]
= σ  ([α]  [β])
 ←
so σ is a homomorphism. Writing τ for the reverse path σ , it is easy to see that
τ  : π1 (X, x1 ) → π1 (X, x0 ) is the inverse of σ  , so that σ  is an isomorphism. The
proof is complete. 

Thus, for a path connected space we can speak of the fundamental group π1 (X )
of the space, without any reference to the base point. Note, however, that the given
isomorphism is not natural and, in general, depends very much on the choice of the
path σ . In fact, it can be shown that the isomorphism σ  is independent of σ if and
only if π1 (X, x0 ) is abelian. This last condition puts a severe restriction on X .
Examples of fundamental groups are not easy to come by. Here is the simplest of
these: Any two loops at x0 ∈ Rn are path homotopic and so π1 (Rn , x0 ) is the trivial
one element group. Spaces of this type are important and so have a special name.
226 15 Fundamental Groups and Covering Spaces

Definition 15.5 A simply connected space is a path connected space having a trivial
fundamental group.

Example 15.6 Any (nonempty) convex set in Rn is simply connected—it is clearly

path connected and any two paths are path homotopic via the straight line homotopy.

The fundamental group gives us a way that leads to groups from spaces: π1 :
(pointed) topological spaces  groups. Having a method of passing from one class
of objects (topological spaces) to another (groups), one would like to take along
appropriate maps between objects of the first class to relevant maps between objects
of the second class—in this case, continuous maps between topological spaces to
homomorphisms between their fundamental groups. These are easily got and the
procedure has all the nice properties one would expect.
For convenience, we just write f : (X, x0 ) → (Y, y0 ) to mean that x0 ∈ X,
y0 ∈ Y , f : X → Y and f (x0 ) = y0 .

Proposition 15.7 Let f : (X, x0 ) → (Y, y0 ) be continuous. If α is a loop in X at x0 ,

then f ◦ α is a loop in Y at y0 and the map f ∗ : π1 (X, x0 ) → π1 (Y, y0 ) defined by
f ∗ ([α]) := [ f ◦ α] is well-defined and is a group homomorphism.

Proof Observe that if α, β are two loops at x0 and if α ∼β, ˙ then f ◦ α ∼ ˙ f ◦ β. In

fact, if F : α ∼β,
˙ then it is easy to see that f ◦ F : f ◦ α ∼ ˙ f ◦ β. This means that
f ∗ is well-defined. Now it is an easy verification to see that f  is a homomorphism.
Note first that if α, β are two loops at x0 , then f ◦ (α  β) = ( f ◦ α)  ( f ◦ β). Thus
we have

f ∗ ([α]  [β]) = f ∗ ([α  β])

= [ f ◦ (α  β)]
= [( f ◦ α)  ( f ◦ β)]
= [( f ◦ α)]  [( f ◦ β)]
= f ∗ ([α])  f ∗ ([β]).

The proof is complete. 

The homomorphism f  is called the homomorphism induced by the continuous

map f . Note that it is in the same direction as f : if f is a map from X to Y , then f 
is from the fundamental group of X to that of Y . Induced homomorphisms provide
a passage from topology to algebra and behave nicely with respect to compositions,
as the next result shows. Topologists call these properties functorial and say that the
fundamental group is a (covariant) functor from the category of (pointed) topological
spaces to the category of groups. (Covariant just refers to the fact that the induced
homomorphisms f  go in the same direction as the maps f .)
15.1 The Fundamental Group 227

Proposition 15.8 Homomorphisms induced by maps satisfy:

1) If f : (X, x0 ) → (Y, y0 ), g : (Y, y0 ) → (Z , z 0 ) are continuous maps, then
(g ◦ f )∗ = g∗ ◦ f ∗ : π1 (X, x0 ) → π1 (Z , z 0 ).
2) If id X : (X, x0 ) → (X, x0 ) is the identity map on X , then (id X )∗ = idπ1 (X,x0 ) ,
the identity map on π1 (X, x0 ).
Proof 1) presents no difficulties. For any loop α at x0 ,
(g ◦ f )∗ ([α]) = [(g ◦ f ) ◦ α)] = [g ◦ ( f ◦ α)] = g∗ ([ f ◦ α]) = g∗ ( f ∗ ([α])).
2) is even more trivial: (id X )∗ ([α]) = [id X ◦ α] = [α]. 
Corollary 15.9 (Topological Invariance)
If f : (X, x0 ) → (Y, y0 ) is a homeomorphism, then the induced map f ∗ : π1 (X, x0 )
→ π1 (Y, y0 ) is an isomorphism.
Proof If g = f −1 : Y → X is the inverse homeomorphism, then g ◦ f = id X and
f ◦ g = idY and so
g∗ ◦ f ∗ = (g ◦ f )∗ = (id X )∗ = idπ1 (X,x0 )
and similarly f ∗ ◦ g∗ = idπ1 (Y,y0 ) . Thus f ∗ and g∗ are inverses of each other and hence
each is an isomorphism. 
Thus the fundamental group is a topological invariant. It is, in fact, a homotopy
invariant. If f, g : X → Y are homotopic, is f ∗ = g∗ ? Not quite, since possibly
f (x0 )
= g(x0 ). However, the two differ only by the isomorphism induced by a path
joining the two points.
Lemma 15.10 Let X, Y be path connected, x0 ∈ X and continuous maps f, g :
X → Y be homotopic. Then f ∗ : π1 (X, x0 ) → (Y, f (x0 )) and g∗ : π1 (X, x0 ) →
(Y, g(x0 )) ‘differ by an inner automorphism of π1 (Y )’. More precisely, if F : f ∼ g,
←
then g∗ = σ  ◦ f ∗ , i.e. [g ◦ α] = [ σ  ( f ◦ α)  σ ], where σ is the path in Y from
y0 = f (x0 ) to z 0 = g(x0 ) defined by σ (t) = F(x0 , t).
Proof If F : f ∼ g, then G : f ◦ α ∼ g ◦ α where α is a loop at x0 in X and
G(s, t) = F(α(s), t). Write γt (s) = G(s, t) and σt (s) = σ (ts) for 0 ≤ s, t ≤ 1.
Then γ0 = f ◦ α, γ1 = g ◦ α and σ0 = e y0 , σ1 = σ . Geometrically, σt is the path
from f (x0 ) to σ (t) defined by σ |[0,t] and γt is a loop at σ (t), ‘the position of
f ◦ α at time t in the process of deforming f ◦ α to g ◦ α’. Thus κt := σt 
←
(γt  σt ), 0 ≤ t ≤ 1, gives a continuous family of loops at f (x0 ) = y0 that yields
←
a path homotopy between f ◦ α and σ  (g ◦ α)  σ ). In fact, if K (s, t) := κt (s),
then K : [0, 1] × [0, 1] → Y is continuous,
K (0, t) = σt (0) = F(x0 , t) = f (x0 ) = y0 ,
←
K (1, t) = σ (1) = f (x0 ) = y0 ,
←
K (s, 0) = σ0  (γ0  σ0 ) = e y0  (( f ◦ α)  e y0 ),
←
K (s, 1) = σ  ((g ◦ α)  σ ).
← ←
This shows that K : σ  ((g ◦ α)  σ )∼e ˙ y0  (( f ◦ α)  e y0 )∼˙ f ◦ α. Hence g ◦ α ∼ ˙σ
 (( f ◦ α)  σ ). In other words, g∗ ([α]) = σ  ( f ∗ ([α]) for any loop α at x0 in X . This
is the assertion of the lemma. 
228 15 Fundamental Groups and Covering Spaces

Corollary 15.11 1) With the notation of the lemma, if f  is trivial (respectively,

injective or surjective), then so is g .
2) If f : (X, x0 ) → (Y, y0 ) is homotopic to the constant map c, c(x) = y0 for all
x ∈ X , then f  is the trivial homomorphism.

Proof This is immediate since σ  is an isomorphism. 

Theorem 15.12 (Homotopy invariance)

Two path connected spaces which are homotopically equivalent have isomorphic fun-
damental groups. In fact, if X, Y are path connected spaces then any homotopy equiv-
alence f : (X, x0 ) → (Y, y0 ) induces an isomorphism f ∗ : π1 (X, x0 ) → π1 (Y, y0 ).

Proof Let g : Y → X be a homotopy inverse of f , so that we have homotopies

F : g ◦ f ∼ id X and G : f ◦ g ∼ idY . Then σ (t) = F(x0 , t) defines a path in X
from x1 := g(y0 ) to x0 and the lemma shows that
g∗ ◦ f ∗ = (g ◦ f )∗ = σ  ◦ (id X )∗ = σ  ◦ idπ1 (X,x0 ) = σ  .
where g∗ : π1 (Y, y0 ) → π1 (X, x1 ).
Next, let y1 = f (x1 ) and η(t) = G(y0 , t). Then η is a path in Y from y1 to y0
and f ∗ ◦ g∗ = η where f ∗ : π1 (X, x1 ) → π1 (Y, y1 ) is the homomorphism induced
by f : (X, x1 ) → (Y, y1 ).
We have thus shown that g∗ ◦ f ∗ and f ∗ ◦ g∗ are both isomorphisms. From the
first we conclude that g∗ is surjective and we infer its injectivity from the second.
Thus g∗ is an isomorphism. But then f ∗ is an isomorphism as well, because g∗ ◦ f ∗
is also an isomorphism.(Note that g∗ may not be the inverse of f ∗ .) 

Corollary 15.13 A contractible space is simply connected.

Proof A contractible space is path connected and has the homotopy type of a point,
whose fundamental group is clearly trivial. 

The sphere S2 is simply connected as we will see a little later, but it is a fact that it
is not contractible.

Corollary 15.14 If E is a deformation retract of a path connected space X , then E

and X have isomorphic fundamental groups.

Proof We have proved that a space and its deformation retract are homotopically
equivalent (14.26). 

Exercise. If r : (X, x0 ) → (E, x0 ) is a retraction and ι : E → X is the inclusion map,

then r∗ is surjective and ι∗ is injective.
15.2 Examples and Applications 229

15.2 Examples and Applications

In general, the task of computing the fundamental group of a given space is a difficult
one. As a very simple example, we have seen that Rn has trivial fundamental group.
Determining the fundamental groups of even simple spaces like the circle or the
sphere is nontrivial. With minimal machinery we will try to discuss some simple
examples.

Theorem 15.15 The sphere Sn is simply connected for n > 1.

Proof Let α be a loop in Sn . By uniform continuity there is a subdivision of [0,1],

0 = s0 < s1 < . . . < sk = 1, such that α(s j ) − α(s j−1 ) < 1/2 for all j, Let β be
the polygonal curve formed by the successive line segments [α(s j−1 ), α(s j )], j =
1, . . . , k and let γ (s) = β(s)/β(s). Then γ is the curve on Sn formed by the
arcs got by radial projections of the line segments forming β. In particular, γ is
not surjective. If p ∈ Sn \ (image γ ), then γ is a loop in Sn \ { p} Rn and so
(1−t)α(s)+tγ (s)
is homotopic to the constant loop. On the other hand H (s, t) = (1−t)α(s)+tγ (s)
defines a homotopy from α to γ . Conclusion: α is null homotopic and the proof is
complete. 

We will revisit this result later in a more general set-up. The case n = 1 is more
involved and we begin with an informal, heuristic discussion about S1 . It is, of course,
path connected. It is intuitively clear that it cannot have trivial fundamental group: the
simple, circular loop (at, say, 1) cannot be shrunk to a point loop. It is also reasonably
‘clear’ that the loop that makes m revolutions is ‘different’ from the one that makes n
revolutions unless m = n. Of course, revolutions can be positive (anti-clockwise) or
negative (clockwise). For example, making a positive revolution and a negative one
will amount to making no revolution at all, i.e., is equivalent to the constant loop.
Thus one can feel that there is a close connection between π1 (S1 ) and integers. We
now proceed to make these intuitive observations precise and rigorous, and identify
the fundamental group of the circle.
The first example of a nontrivial fundamental group that we will see is that of
the circle. The methods involved in this are important and general. However, we
postpone all the technical definitions and concepts involved to the next section. Here
we directly and explicitly work with circle and eschew generalities and jargon.

Theorem 15.16 (Basic lifting theorem)

Let p : R → S1 be the map p(x) = e2πı x and let [a, b] a ≤ b, be a compact interval in
R. If F : [a, b] × [0, 1] → S1 is continuous and F(a, 0) = 1, then there is a unique
continuous map F̃ : [a, b] × [0, 1] → R such that F̃(a, 0) = 0 and p ◦ F̃ = F.

Proof Note that p restricted to (−1/2, 1/2) is a homeomorphism of this interval

onto S1 \ {−1}. Let q be the inverse homeomorphism. Write Z = [a, b] × [0, 1] for
230 15 Fundamental Groups and Covering Spaces

notational convenience. Uniform continuity of F ensures that there is a δ > 0 such

that
|F(z) − F(z  )| < 1 for z, z  ∈ Z with |z − z  | < δ.
Observe then that F(z)
= −F(z  ) for |z − z  | < δ. Fix a positive integer N > d :=
max|z − z 0 |/δ, where z 0 = (a, 0). For z ∈ Z , divide the line segment [z 0 , z] into N
z∈Z
equal parts by points z 0 , z 1 , . . . , z N = z. Define
F(z )
F̃(z) = 1N q( F(z j−1
j
)
).
This yields a function F̃ : Z → R which is clearly continuous, noting that z j =
z 0 + Nj (z − z 0 ), 0 ≤ j ≤ N . Note that F̃(z 0 ) = q(1) = 0 and
F(z ) F(z j ) F(z N )
p ◦ F̃(z) = 1N p(q( F(z j−1
j
)
)) = 1N F(z j−1 )
= F(z 0 )
= F(z).
F̃ has the stated properties. To prove the uniqueness, suppose G̃ : Z → R is contin-
uous and satisfies the conditions G̃(z 0 ) = 0, p ◦ G̃ = F. Then
p( F̃(z) − G̃(z)) = p( F̃(z))p(G̃(z)) = F(z)F(z) = 1,
and so F̃(z) − G̃(z) is an integer for all z ∈ Z . Now the continuity of F̃ − G̃ and
connectedness of Z forces F̃ − G̃ to be a constant. But F̃(z 0 ) = 0 = G̃(z 0 ) and so
F̃ = G̃. 

F̃ is called the lift of F to R. We now apply this to lift loops in S1 .

Corollary 15.17 (Path lifting theorem)
If α is a loop at 1 in S1 , there is a unique path α̃ in R with initial point 0 such that
p ◦ α̃ = α. Thus loops in S1 lift to paths in R.

Proof Take a = b = 0, identify {0} × [0, 1] with [0,1] and take F = α in the basic
lifting lemma. More elaborately, let F : {0} × [0, 1] → S1 be defined by F(0, t) =
α(t) and let F̃ : {0} × [0, 1] → R be its lift. Then α̃(t) = F̃(0, t) gives the required
lifting with α̃(0) = F̃(0, 0) = 1. 

Note that even though α is a loop, the lifted path α̃ may not be a loop. The choice
of the base point as 1 is only a convenience. If α is a loop at z 0 = e2πıt0 , it has a
unique lift α̃ as a path in R with initial point t0 and end point t0 + n for some n ∈ Z,
so p ◦ α̃ = α.
The next agendum is to see that homotopies can also be lifted.

Corollary 15.18 (Homotopy lifting theorem)

Let α, β be loops at 1 in S1 . Suppose there is a path homotopy F : α ∼β. ˙ Then
there is a unique continuous map F̃ : [0, 1] × [0, 1] → R such that p ◦ F̃ = F and
F̃ : α̃ ∼
˙ β̃, α̃, β̃ being the lifts of α, β, respectively, i.e. a homotopy of loops in S1 lifts
to a homotopy of the lifted paths in R.

Proof We show that the lift F̃ : [0, 1[×[0, 1] → R of F, given by the basic lifting
lemma, is a path homotopy from α̃ to β̃. Observe that s → F̃(s, 0) is a path in R with
initial point F̃(0, 0) = 0 such that p( F̃(s, 0)) = F(s, 0) = α(s). The uniqueness of
15.2 Examples and Applications 231

lifting shows that F̃(s, 0) = α̃(s). Since p( F̃(0, t)) = F(0, t) = 1 for 0 ≤ t ≤ 1,
the continuous function t → F̃(0, t) is integer valued, hence, by connectedness,
F̃(0, t) = F̃(0, 0) = 0, 0 ≤ t ≤ 1. In particular, F̃(0, 1) = 0 and so s → F̃(s, 1) is
the unique lift of β with initial point 0, as argued above for α. That is, F̃(s, 1) = β̃(s).
Moreover, F̃(1, t) is a constant function of t, just as we have proved for F̃(0, t).
Conclusion: F̃ is actually a path homotopy, F̃ : α̃ ∼ ˙ β̃, as asserted. 

All these ‘lifting’ results are valid in more generality and these important results will
be taken up in the next section. Now we are ready to enjoy the fruits of our labour in
proving these results: to reach our first goal of identifying the fundamental group of
the circle.
Theorem 15.19 Let α be any loop at 1 in S1 and let α̃ be its unique lift with α̃(0) = 0.
Then
1) deg α := α̃(1) is an integer that depends only on the loop homotopy class of
α, and is called the degree of α.
2) The map deg: π1 (S1 , 1) → Z, deg([α]) := deg α = α̃(1), is an isomorphism
of groups.

Proof 1) α̃(1) is an integer because p(α̃(1)) = α(1) = 1. If β is a loop at 1 in S1

with α ∼β,˙ then we have seen that α̃ ∼ ˙ β̃, in particular α̃(1) = β̃(1). This proves 1).
2) To see that deg is a homomorphism, we need to identify the lift of α  β for
any two loops α, β at 1. Let deg α = m, deg β = n. A path η in R from m to m + n
is defined by η(s) = β̃(s) + m, 0 ≤ s ≤ 1 and α̃  η is a path in R from 0 to m + n.
Noting that p(η) = p(β̃) = β, we have p(α̃  η) = p(α̃)  p(η) = α  β. Thus α̃  η
is the lift of α  β. Hence
deg([α]  [β]) = deg([α  β]) = α̃  η(1) = m + n = deg([α]) + deg([β]).
Thus deg is a group homomorphism.
If n is any integer, εn (s) = e2πıns defines a loop at 1 in S1 whose lift is given by
ε̃n (s) = ns. So, deg εn = n, hence the map deg is surjective.
To prove its injectivity, suppose α is a loop at 1 in S1 with deg α = 0. Then α̃
is a loop at 0 in R and by contractibility of R, there is a path homotopy F̃ from α̃
to ε˜0 , the constant loop at 0 in R. Thus F̃(s, 0) = α̃(s), F̃(s, 1) = 0 and F̃(0, t) =
0 = F̃(1, t), 0 ≤ s, t ≤ 1. Define F = p ◦ F̃. Then F(s, 0) = α(s), F(s, 1) = 1 and
F(0, t) = 1 = F(1, t), 0 ≤ s, t ≤ 1. This means that F : α ∼ε ˙ 0 , the constant loop at
1 in S1 . This proves that deg is injective and the proof is complete. 

The 1-dimensional case of the following theorem was given in Chap. 1, as a conse-
quence of the Intermediate Value Theorem. The general case was given in Chap. 13,
Sect. 6. Here the 2-dimensional case is presented as an application the results of this
section.
Theorem 15.20 (Borsuk-Ulam Theorem, dim 2)
If f : S2 → R2 is continuous, there exists x ∈ S2 such that f (x) = f (−x); i.e. f
takes the same value at some pair of antipodal points.
232 15 Fundamental Groups and Covering Spaces

Proof If the result is false, the function g(x) = f (x) − f (−x) never vanishes and
g(x)
so h(x) = g(x) is a well-defined continuous map S2 → S1 . Define α : [0, 1] → S2
by α(s) = (cos 2π s, sin 2π s, 0); it describes the equatorial circle in S2 . The loop
σ = h ◦ α lifts to a path σ̃ in R. Since h(−x) = −h(x), σ (s + 1/2) = −σ (s) for
0 ≤ s ≤ 1/2, and so σ̃ (s + 1/2) = σ̃ (s) + (2k + 1)/2 for some integer k = k(s).
But k is continuous on [0,1/2] and so must be a constant. Hence
σ̃ (1) = σ̃ (1/2) + (2k + 1)/2
= σ̃ (0) + (2k + 1)/2 + (2k + 1)/2
= σ̃ (0) + (2k + 1).
This shows that σ is not null homotopic. On the other hand, α is null homotopic in
S2 , with null homotopy H say; then h ◦ H is a null homotopy of h ◦ α = σ . This
contradiction completes the proof. 

Proposition 15.21 Let X, Y be path connected spaces and take points x0 ∈ X,

y0 ∈ Y . Then π1 (X ×Y, (x0 , y0 )) is isomorphic to the direct product π1 (X, x0 ) ×
π1 (Y, y0 ).

Proof Let p : X × Y → X and q : X × Y → Y be the projections. We repeatedly

use the fact that, for any space Z , a map f : Z → X × Y is continuous if and only
if p ◦ f and q ◦ f are both continuous.
Let α, β be loops at x0 in X and y0 in Y , respectively. Then α × β defined by
α × β(s) = (α(s), β(s)) is a loop at (x0 , y0 ) in X × Y . This direct product of loops
preserves homotopy: if α, α  are path homotopic loops at x0 and β, β  are such loops
at y0 , then α × β is path homotopic to α  × β  . In fact, if F : α ∼β,
˙ G : α  ∼β˙  , then
 
it is straightforward to check that F × G : α × β ∼α ˙ × β , where F × G(s, t) :=
(F(s, t), G(s, t)). We thus have a well-defined map
ϕ : π1 (X, x0 ) × π1 (Y, y0 ) → π1 (X × Y, (x0 , y0 )), ϕ([α], [β]) = [α × β].
Direct verification shows that (α  α  ) × (β  β  ) = (α × β)  (α  × β  ). This means
that ϕ is a homomorphism of groups.
To see that ϕ is injective, suppose H is a path homotopy from α × β to the constant
loop ε(x0 ,y0 ) . Then p ◦ H and q ◦ H are path homotopies from α to εx0 and from β
to ε y0 , respectively.
Finally, to show that ϕ is surjective, let η be a loop at (x0 , y0 ) in X × Y . Then
α := p ◦ η is a loop at x0 and β := q ◦ η is a loop at y0 . Further, η = α × β, so
[η] = ϕ([α], [β]). The proof is complete. 

Let us now take a re-look at the sphere. It is intuitively clear that any loop at, say,
x0 = (1, 0, 0) on the sphere S2 can be deformed on the sphere to the single point x0 .
This is also not difficult to prove: Divide such a loop α into small pieces so that each
piece αi lies wholly in either in U = S2 \ {(0, 0, 1)} or V = S2 \ {(0, 0, −1)}. Each
end of αi can be joined to x0 by a path that lies in U or V , as the case may be; these,
together with αi , gives a loop at x0 , lying in U or V . As both U and V are simply
connected (homeomorphic to R2 !), each of these loops is null homotopic. But the
product of all these loops, homotopically, is just α. Conclusion: every loop at x0 is
homotopic to the trivial loop and so S2 is simply connected. These arguments can
15.2 Examples and Applications 233

carried out in a general setting with hardly any extra effort and so we write down the
details in this general context. We begin with the observation that ‘a path is the sum
of its pieces’. We state it as a lemma. The idea of the proof is as in the associative
law for path products.

Lemma 15.22 Let α : [0, 1] → X be a path in a space X . Consider a subdivision

0 = t0 < t1 < · · · < tn = 1 of [0, 1]. Let αi be the path defined by α|[ti−1 ,ti ] . Then
˙ 1  (α2  (· · · )  αn ).
α ∼(α

Proof The case n = 1 being trivial, let us consider the case n = 2. So we have 0 =
t0 < t1 < t2 = 1. Note that α j = α ◦ η j where η1 : [0, 1] → [0, t1 ], η1 (s) = t1 s and
η2 : [0, 1] → [t1 , 1], η2 (s) = (1 − t1 )s + t1 are the respective oriented affine maps.

α1 (2s) = α(η1 (2s)), 0 ≤ s ≤ 21 ,
α1  α2 (s) =
α2 (2s − 1) = α(η2 (2s − 1)), 21 ≤ s ≤ 1.

Thus α1  α2 (s) = α(τ (s)) where


η1 (2s), 0 ≤ s ≤ 21 ,
τ (s) =
η2 (2s − 1), 21 ≤ s ≤ 1.

Since τ (0) = η1 (0) = 0, τ (1) = η2 (1) = 1, 14.34 gives α1  α2 = α ◦ τ ∼ ˙ α.

Assume, inductively, that the result holds for a product of n − 1 paths. Thus, for
˙  , with the notation of the lemma, where α  is the path
n > 2, α1  (· · · )  αn−1 ∼α
given by α|[0,tn−1 ] . Thus
˙   αn ∼(α
α ∼α ˙ 1  (α2  (· · · )  αn ).
This completes the induction and the lemma is proved. 

Corollary 15.23 Take subdivisions 0 = t0 < t1 < · · · < tn = 1 and 0 = s0 < s1 <
· · · < sm = 1 of [0, 1] and let αi and β j be the paths given by α|[ti−1 ,ti ] and α|[s j−1 ,s j ] ,
respectively. Then
(α1  (α2  (· · · ))αn ∼(β
˙ 1  (β2  (· · · ))βm .

Proof Each is path homotopic to α. 

Theorem 15.24 (Seifert-van Kampen Theorem, [7, 8])

Let X be a path connected space. Suppose X = U1 ∪ U2 where U1 , U2 are over-
lapping open sets and U1 , U2 , U1 ∩ U2 are path connected. Let ι j : U j → X be the
inclusion maps. Then, for x0 ∈ U1 ∩ U2 , π1 (X, x0 ) is generated by the images of
(ι j ) : π1 (U j , x0 ) → π1 (X, x0 ). In particular, if U1 and U2 are simply connected,
then so is X .

Proof Let x0 ∈ U1 ∩ U2 and let α be a loop at x0 in X . Let ε > 0 be a Lebesgue

number for the open covering {α −1 (U1 ), α −1 (U2 )} of [0,1]. Choose a subdivision 0 =
t0 < t1 < · · · < tm = 1 of [0,1] such that each subinterval has length less that ε. Then,
for each i = 1, . . . , m, α([ti−1 , ti ]) is contained in either U1 or U2 . Thus if αi denotes
the path got by the restriction α|[ti−1 ,ti ] (i.e. αi = α ◦ ϕi ,with ϕi (s) = (ti − ti−1 )s +
ti−1 ), then αi lies entirely in U1 or U2 . This is not a loop; we get a loop from this as
234 15 Fundamental Groups and Covering Spaces

follows. Take a path σi joining x0 to α(ti ) that lies in U1 , U2 or U1 ∩ U2 according as

αi lies in U1 , U2 or U1 ∩ U2 ; this is where the path connectedness of the U j and their
←
intersection come in to play. This way we get loops at x0 : η1 = α1  σ1 , η2 = σ1 
← ←
(α2  σ2 ), . . . , ηm−1 = σm−2  (αm−1  σ m−1 ), ηm = σm−1  αm , each lying in either
U1 or U2 . These satisfy
α ∼α
˙ 1  α2  · · ·  αm
∼η
˙ 1  · · ·  ηm

where we may put the parenthesis at convenient places, because of homotopy. We

can rewrite this as
[α] = [η1 ]  · · ·  [ηm ]
where each [ηk ] belongs to the image of either (ι1 ) or (ι2 ) . This proves the first
assertion of the theorem that π1 (X, x0 ) is generated by these images. If U1 and U2
are both simply connected, then the images are trivial, and so the second assertion
is immediate from the first. 
Note that the hypothesis of path connectedness of X is redundant as it follows
from the assumptions on the U j . The theorem given here is only a special case of the
full Seifert-van Kampen theorem which involves free products of groups. See [5, 6,
9, 10].
Corollary 15.25 Sn is simply connected for n ≥ 2.
Proof The sphere is path connected since any two points on it can be joined by a great
circular arc. If you are not convinced that this is an ‘actual’ proof, note that Rn+1 \ {0}
is path connected for n ≥ 1 and Sn is its image under the continuous map x → x/x.
Apply the theorem with U1 = Sn \ {(0, . . . , 0, 1)} and U2 = Sn \ {(0, . . . , 0, −1)}.
Note that {U1 , U2 } is an open cover for Sn , each U j is homeomorphic to Rn and U1 ∩
U2 is homeomorphic to Rn \ {0}, all under stereographic projections. In particular,
U1 , U2 are simply connected and U1 ∩ U2 is path connected for n ≥ 2. Now (15.24)
yields the corollary. 
The only examples of fundamental groups that we have seen so far are those of
Rn and Sn , n ≥ 1. With these in our kitty, the product result and the topological and
homotopy invariance can be used to produce more examples. Here are some of these.
Example 15.26 1. The unit circle S1 is a deformation retract of the punctured plane
R2 \ {0} and of the punctured disc D \ {0} (via radial projections). So each of the
latter is homotopically equivalent to the circle and hence have fundamental groups
isomorphic to Z.
2. The fundamental groups of the cylinders S1 × R and S1 × [0, 1] are also isom-
rphic to Z, by the product theorem. Another way to see this is to observe that the
cylinders are homotopically equivalent to the circle, just by collapsing along each
generator.
3. Consider X = R3 \ {(0, 0, z) : z ∈ R}, the three dimensional Euclidean space
with the z-axis removed. As in the previous example, every vertical line can be
15.2 Examples and Applications 235

collapsed to a single point on the x-y-plane to end up with the punctured plane.
H ((x, y, z), t) := (x, y, (1 − t)z) defines a deformation retraction of X to the x-y-
plane punctured at the origin (see Fig. 14.3) So π1 (X ) is isomorphic to Z.
4. The n-dimensional torus Tn , which is the product of n copies of T := S1 , has
fundamental group Zn , the free abelian group of rank n.
5. The product of two simply connected spaces is simply connected. In particular,
Rn−1 × Sm and Sn × Sm are simply connected for m, n > 1.
6. Rn \ {0} → Sn−1 × (0, ∞), x → (x/x, x), is a homeomorphism. This
shows that Rn \ {0} simply connected for n ≥ 3. For n = 2 we again get that
π1 (R2 \ {0}) π1 (S1 ) Z.
7. Two circles touching externally is a deformation retract of the plane or disc
with two holes (Fig. 14.2). So all these have isomorphic fundamental groups. A circle
together with a diameter (called the theta space) has the same homotopy type, and
hence the same fundamental group, as each of these. But what is this group? We have
not computed it! We only make some heuristic remarks. Think of the picture of two
circles touching externally. Take this point of contact x0 as the base point. You can
form loops at x0 by going around either circle any number of times. These give loops
of the form (with obvious notation) ρ1m 1  ρ2n 1  ρ1m 2  ρ2n 2  · · · ρ1m k  ρ2n k , m j , n j
∈ Z. It is true that any loop at x0 is path homotopic to a loop of this form. Thus
the fundamental group consists of elements of the form [ρ1 ]m 1  [ρ2 ]n 1  [ρ1 ]m 2 
[ρ2 ]n 2  · · · [ρ1 ]m k  [ρ2 ]n k , m j , n j ∈ Z (see [9]). This is the free group on two gen-
erators and is the first nonabelian fundamental group that we have encountered.

Corollary 15.27 R2 is not homeomorphic to Rn , n

= 2.

Proof The fundamental group of Rn \ {0} is trivial for n > 2 and is nontrivial for
n = 2. 

Example 15.28 Let n ∈ Z and consider f : S1 → S1 , f (z) = z n . What is f ∗ ? ε1 (s)

= e2πıs is a loop at 1 and f ∗ ([ε1 ]) = [εn ], (where εn (s) = e2πıns ), has lift ε̃n (t) =
nt. Thus deg( f ∗ ([ε1 ])) = n and identifying π1 (S1 , 1) with Z, f ∗ : Z → Z is the
homomorphism k → nk.

Here are special cases of the general results proved in Chap. 13, but now proved
using fundamental groups.

Theorem 15.29 (The no retract theorem-dim 2)

The unit circle S1 is not a retract of the closed unit disc B2 = D.

Proof Suppose there is a retraction r : B2 → S1 and let ι : S1 → B2 be the inclu-

sion map. Then r ◦ ι = idS1 and so r ◦ ι = idπ1 (S1 ,1) . In particular, r : π1 (B2 , 1) →
π1 (S1 , 1) is surjective. But this is absurd since π1 (B2 , 1) is the trivial group and
π1 (S1 , 1) is not. 
236 15 Fundamental Groups and Covering Spaces

Theorem 15.30 (Brouwer fixed point theorem—dim 2)

Every continuous map f : B2 → B2 has a fixed point.

Proof Suppose f : B2 → B2 is continuous and has no fixed point. Thus f (z)

= z
for all z ∈ B2 . For each z ∈ B2 , the oriented line segment from f (z) to z, when
extended, meets S1 in a unique point, say ρ(z). Note that ρ(z) = z for z ∈ S1 . This
yields a map ρ : B2 → S1 such that ρ(z) = z for any z ∈ S1 . But since the continuity
of f ensures that of ρ, we get a retraction ρ of B2 on S1 , a contradiction. (It is easy to
compute ρ in terms of f . Assume |z| < 1 and write z = x + ı y, f (z) = u + ıv.. Then
ρ(z) = 1−t1
z − 1−tt
f (z), where t = t (z) is given as follows: t = (1 − xu − yv +
[(1 − xu − yv) − (1 − x 2 − y 2 )(1 − u 2 − v 2 )]1/2 )/(1 − u 2 − v 2 ) when | f (z)|
=
2

1 and t = (1 − x 2 − y 2 )/2(1 − xu − yv) if | f (z)| = 1. Note that |xu + yv| < 1 by

Cauchy-Schwarz inequality and so the denominator is nonzero.) 

Here is a topological proof of a familiar result.

Theorem 15.31 (Fundamental theorem of algebra)

Every nonconstant complex polynomial has a complex root.

Proof It suffices to prove the result for monic polynomials. Consider such a polyno-
mial p(z) = z n + a1 z n−1 + · · · + a0 and suppose it has no zeros in C. We can then
define h t (z) := p(t z)/| p(t z)| to get a map h t : S1 → S1 for each t ≥ 0. Note that h 0
is a constant and F(z, s) = h ts (z) gives a continuous map F = Ft : S1 × [0, 1] → S1
for any t > 0. Then F(z, 0) = h 0 and F(z, 1) = h t (z). Thus F is a homotopy from
h 0 to h t . Since h 0 is a constant, it induces the trivial homomorphism on π1 (S1 , 1)
and so h t also induces the trivial homomorphism for all t > 0.
On the other hand, we now show that for sufficiently large t > 0, h t is homotopic to
the map z → z n . We know that this last map induces the nontrivial homomorphism
k → nk and so h t would also induces a nontrivial homomorphism, leading to a
contradiction.
To see the opening assertion of the previous paragraph, observe first that lim h t (z)
t→∞
= z n for each z ∈ S1 . In particular, (1 − s)z n + sh t (z)
= 0, 0 ≤ s ≤ 1, for all large
t. If we let
H (z, s) := (1 − s)z n + sh t (z)/|(1 − s)z n + sh t (z)|,
then H : S1 × [0, 1] → S1 is continuous, H (z, 0) = z n and H (z, 1) = h t (z), thus
giving a homotopy H = Ht from the function z n to h t for each sufficiently large t.
Hence our assertion is proved yielding a contradiction and completing the proof. 

15.3 Covering Spaces

The idea of a covering space goes back to Riemann’s discussion of ‘sheets’ of a

Riemann surface. Poincaré was very likely aware of the lifting theorem even though
he did not state it explicitly. For Riemann surfaces, it was given by Weyl(1913). In
15.3 Covering Spaces 237

their book [11] of 1934, Seifert and Threlfall presented, probably for the first time,
a thorough discussion of the relation between the fundamental group, path lifting
and covering spaces. For more on this, see Dieudonné’s history, [12]. For a more
detailed discussion of fundamental groups and covering spaces, see [5, 9, 10, 13,
14].
Let us begin by looking back at the procedure leading to the identification of
the fundamental group of the circle. It relied on the fact that R and S1 are tied
up topologically in a special way. This relation is implemented by the exponential
map p : R → S1 , p(x) = e2πı x , that was critically used throughout. This surjective
continuous map has some interesting properties. For instance, if U = S1 \ {−1} is the
circle punctured at −1, it is a neighbourhood of 1 in S1 , p−1 (U ) is the disjoint union
∪ (n − 21 , n + 21 ) and p|(n− 21 ,n+ 21 ) is a homeomorphism of this interval onto U for
n∈Z
each n. (For n = 0 this homeomorphism was used in our discussion of π1 (S1 , 1).)
If V is the open semicircular arc in the right half plane joining ı and −ı, then
p−1 (V ) = ∪Vn∗ where Vn∗ = (n − 41 , n + 41 ) and p is a homeomorphism of V onto
Vn∗ . These properties are encapsuled in the concept of covering spaces that is of
fundamental importance in topology and geometry.

Definition 15.32 A covering space of a space X is a pair ( X̃ , p) where X̃ is a

space and the covering map p : X̃ → X is a continuous surjection with the property:
each x ∈ X has an open neighbourhood U of x such that p−1 (U ) a disjoint union
∪Ũi of open sets Ũi in X̃ and p|Ũi is a homeomorphism onto U (Fig. 15.1). Such
neighbourhoods U are called admissible neighbourhoods. The Ui are referred to as
the sheets or slices over U and we say that U is evenly covered by these sheets. A
map f˜ : X̃ → Y is called a lift of f : X → Y if p ◦ f˜ = f (Fig. 15.2).

X is thought of as the base space and several (finitely or infinitely many) copies of
U stacked above U . What the covering map p does is to just squash down the entire
stack on U . Note that p is a local homeomorphism. (What should the term mean?)

Example 15.33 1. By the opening discussion, p : R → S1 , p(x) = e2πı x is an ‘infi-

nite cover’ (cover with infinitely many sheets).
2. The identity map on any space is a covering map, trivially.
3. For any positive integer n, S1 → S1 , z → z n is an ‘n-fold’ cover.
4. If p j : X˜ j → X j is a covering for j = 1.2, then we have a covering p1 × p2 :
X˜1 × X˜2 → X 1 × X 2 , where p1 × p2 (x˜1 , x˜2 ) = (p1 (x˜1 ), p2 (x˜2 )).
5. The projective space RPn is got by identifying the antipodal points of Sn . The
quotient map p : Sn → RPn is a ‘double’ cover.

Here are some basic properties of covering maps.

Proposition 15.34 Every covering map is an open map.

238 15 Fundamental Groups and Covering Spaces

Ũ3

Ũ2

Ũ1 p

Fig. 15.1 Covering map

X̃

f˜
p

f
Y X

Fig. 15.2 Lifting

Proof Let p : X̃ → X be a covering map, Ṽ be an open set in X̃ and let x = p(x̃)

∈ p(Ṽ ), x̃ ∈ Ṽ . Choose an admissible neighbourhood U of x and the sheet Ũ of
p−1 (U ) containing x̃, so that p|Ũ is a homeomorphism onto U . Ṽ ∩ Ũ is open in Ũ
and so p(Ṽ ∩ Ũ ) is open in U , hence in X . Thus every point of p(Ṽ ) is an interior
point. 

Lifts, when they exist, are unique.

Lemma 15.35 (Uniqueness of lifts)
Let p : X̃ → X be a covering map and let Y be a connected space. If f˜, g̃ : Y → X̃
are continuous maps such that p ◦ f˜ = p ◦ g̃, then either f˜ = g̃ or f˜ and g̃ agree
nowhere on Y . So, if f has a lift, it is unique.

Proof We show that A = {y : f˜(y) = g̃(y)} and B = Ac are both open. Let y ∈ Y
and choose an open neighbourhood U of f (y) that is evenly covered and sheets
Ũ , Ṽ of p −1 (U ) such that f˜(y) ∈ Ũ , g̃(y) ∈ Ṽ . Then Ũ = Ṽ if y ∈ A and they are
disjoint if y ∈ B. By continuity, there is an open neighbourhood V of y such that
V ⊂ f˜−1 (Ũ ) ∩ g̃ −1 (Ṽ ).
Suppose y ∈ A. Then Ũ = Ṽ and for y  ∈ V , p( f˜(y  )) = f (y  ) = p(g̃(y  )).
Since p is injective on Ũ = Ṽ , we get f˜(y  ) = g̃(y  ). So V ⊂ A and hence A is
open. If y ∈ B, then Ũ and Ṽ are disjoint, so f˜(y  )
= g̃(y  ) for y  ∈ V and V ⊂ B,
hence B is open. 
15.3 Covering Spaces 239

x̃0 Ũ11 x̃1 Ũ21

Rij F̃11
p F̃21 p
R21 F (R11 ) x F (R21 )
z1 0 U11 x1 U21
F
R11 R12 F
z0

Fig. 15.3 Basic lifting—construction

The underlying idea of the proof of lifting in the general case is the same as in the
special case of R and S1 , but the details are more involved. In that case, we could write
down everything explicitly, but now we have to work with admissible neighbourhoods
and sheets. Cover F([a, b] × [0, 1]) by finitely many admissible open sets Ui and
subdivides [a, b] and [0,1] into small enough intervals I j and Jk such that each
F(I j × Jk ) is contained in a single Ui ; F̃ is then carefully defined piece-wise on
each I j × Jk to give a lift of F| I j ×Jk so that these pieces could be patched up using
the pasting lemma. Once we get the basic lifting, path and homotopy liftings follow.

Theorem 15.36 (Basic Lifting Theorem)

Let p : X̃ → X be a covering map. Suppose a, b ∈ R, a ≤ b and F : [a, b] ×
[0, 1] → X is continuous. Let z 0 = (a, 0), x0 = F(z 0 ) and let x̃0 ∈ X̃ be such
that p(x̃0 ) = x0 . Then there is a unique lifting F̃ : [a, b] × [0, 1] → X̃ , that is,
F̃(z 0 ) = x̃0 and p ◦ F̃ = F.

Proof By uniqueness, we only need to find a lift F̃. If X itself is admissible, take
F̃ = (p|Ũ0 )−1 ◦ F, Ũ0 being the sheet containing x̃0 .
Otherwise, let {Uα } be a cover of X by admissible open sets and let δ > 0
be a Lebesgue number of the open cover {F −1 (Uα )} of Z := [a, b] × [0, 1]. Let
a = s0 < s1 < · · · sm = b, 0 = t0 < t1 < · · · tn = 1 be partitions so that each Ri j =
[si−1 , si ] × [t j−1 , t j ] has diameter less than δ. This ensures that each Ri j is mapped
by F into a single admissible neighbourhood: F(Ri j ) ⊂ Uα for some α. Now F̃ is
defined on Z by first defining it on each Ri j and then pasting all the pieces together.
Begin with the bottom left corner rectangle R11 . F(R11 ) lies in an admissible
neighbourhood U11 of x0 . If Ũ11 is the sheet over U11 with x̃0 ∈ Ũ11 and p(x̃0 ) = x0 ,
then p11 := p|Ũ11 is a homeomorphism onto U11 and F̃11 := (p11 )−1 ◦ F is the lift of
F on R11 with F̃11 (z 0 ) = x̃0 .
Next consider R21 , the rectangle above R11 in the first column. Write z 1 =
(a, t1 ), x̃1 = F̃11 (z 1 ). Choose an admissible neighbourhood U21 of x1 = F(z 1 ) with
F(R21 ) ⊂ U21 and a sheet Ũ21 above it that contains x̃1 . F̃21 := (p21 )−1 ◦ F is the lift
of F on R21 with F̃21 (z 1 ) = x̃1 . Note that x1 ∈ U11 ∩ U21 and if z ∈ F −1 (U11 ∩ U21 ),
then F̃11 (z) = p−1 −1
11 (F(z)) = p21 (F(z)) = F̃21 (z). Thus F̃11 = F̃21 on F (U11 ∩
−1
240 15 Fundamental Groups and Covering Spaces

U21 ) ⊃ R11 ∩ R21 . The pasting lemma yields a continuous map F̃2 with F̃2 = F̃i1
on Ri1 for i = 1, 2. This F̃2 is the lift of F on R11 ∪ R21 with F̃2 (z 0 ) = x̃0 . It is now
clear how to proceed: take the next rectangle R31 in the first column, using admissi-
ble open sets as before, get a lift F̃31 of F on R31 with F̃31 (z 2 = (a, s2 )) = F̃21 (z 2 ).
Paste this with F̃2 to get the lift F̃3 of F on ∪31 Ri1 with F̃3 (z 0 ) = x0 . After m steps
(induction) we get a lift F̃m of F on ∪m 1 Ri1 with F̃m (z 0 ) = x 0 . (See Fig. 15.3.) (U11
and U21 overlap; they both contain the image of the common boundary of R11 and
R21 . For clarity of pictures, we have drawn them separately.)
What we have done so far is this: starting with the bottom left corner rectangle
R11 we have inductively constructed Fi1 on each of the rectangles Ri1 above R11 in
the first column. Similarly, starting with R11 we can move right to get the lifts in each
rectangle R1 j in the bottom row. Again, starting with each bottom rectangle, we can
go up the column and cover all the Ri j . Note that, at each stage Ri j , the lift is chosen
to ‘match with’ the lifts in the adjacent rectangles Ri−1, j and Ri, j−1 , immediately
below and to the left of Ri j . This enables the pasting to be done at each stage. Finally,
after going through the patching at each stage, we get the required lift F̃. 
Corollary 15.37 (Path Lifting)
Let p : X̃ → X be a covering map. Let x0 ∈ X and let x̃0 ∈ X̃ be such that p(x̃0 ) =
x0 . If α is a path in X with α(0) = x0 , then there is a unique path α̃ in X̃ with initial
point x̃0 such that p ◦ α̃ = α.
Proof The proof of 15.17 is valid with hardly any changes. 
Corollary 15.38 (Homotopy Lifting)
Let p : ( X̃ , x̃0 ) → (X, x0 ) be a covering map. Let α, β be paths in X with α(0) =
x0 = β(0) and α(1) = β(1). Let α̃, β̃ be their lifts with initial point x̃0 . If there is
a path homotopy F : α ∼β, ˙ then there is a unique lift F̃ : [0, 1] × [0, 1] → X̃ of F
with F̃ : α̃ ∼˙ β̃.
Proof Let F̃ be the unique lift of F with F̃(0, 0) = x̃0 . Then
p( F̃(s, 0)) = F(s, 0) = α(s), p( F̃(s, 1)) = F(s, 1) = β(s).
By uniqueness of liftings, F̃(s, 0) = α̃(s), F̃(s, 1) = β̃(s). Further,
p( F̃(0, t)) = F(0, t) = x0 and p( F̃(1, t)) = F(1, t) = α(1) = β(1).
Thus F̃(0, t) is in the discrete space p−1 {x0 }, so F̃(0, t) ≡ F̃(0, 0) = x̃0 , by conti-
nuity and F̃ : α̃ ∼
˙ β̃. In the same way, F̃(1, t) is a constant. 
Proposition 15.39 Let p : X̃ → X be a covering map and let x0 ∈ X , x̃0 ∈ X̃ with
p(x̃0 ) = x0 . Then p : π1 ( X̃ , x̃0 ) → π1 (X, x0 ) is injective; in particular, if X is sim-
ply connected, then so is X̃ .
Proof Suppose α̃ is a loop at x̃0 in X̃ and F : p ◦ α̃ ∼e
˙ x0 . Let F̃ be the lift of F such
that F̃(0, 0) = x0 . Then F̃ : α̃ ∼e
˙ x̃0 . 
Proposition 15.40 Let p : ( X̃ , x̃0 ) → (X, x0 ) be a covering map. If X̃ is path con-
nected, then [α] → α̃(1) is a well-defined map η : π1 (X, x0 ) → X̃ , where α̃ is the
lift of α with α̃(0) = x̃0 . Further
15.3 Covering Spaces 241

1) η : π1 (X, x0 ) → p−1 (x0 ) is surjective;

2) η is bijective if X̃ is simply connected.

Proof Note that η is well-defined because α̃ ∼ ˙ β̃ whenever α ∼β. ˙

∼
1) Clearly p(η([α])) = x0 , so η maps π1 (X, x0 ) into p (x0 ). If x1 ∈ p−1 (x0 ), let
−1
∼
α̃ be a path in X̃ from x̃0 to x1 . Then α := p ◦ α̃ is a loop at x0 in X and, by definition,
∼
η([α]) = x1 .
2) If X̃ is simply connected, α, β are loops at x0 and α̃(1) = β̃(1), there is a
homotopy F̃ : α̃ ∼ ˙ β̃. Then p ◦ F̃ : α ∼β
˙ and η is injective. 

Corollary 15.41 The fundamental group of the real projective space is the cyclic
group of order 2: π1 (RPn ) Z2 , n > 1.

Proof p : Sn → RPn , x → [x] = {±x}, is a covering. Each p−1 (x0 ) has two ele-
ments and Sn is simply connected; 2) applies. 

Proposition 15.42 Let p : ( X̃ , x̃0 ) → (X, x0 ) be a covering map and let

H = p (π1 ( X̃ , x̃0 )). Let α, β be loops in X at x0 with lifts α̃, β̃.
1) If α̃(0) = x̃0 , then α̃(1) ∈ p−1 (x0 ).
2) α̃(1) = β̃(1) if and only if [β] ∈ H  [α].
3) H  [α] → α̃(1) is a well defined injective map from the set of right cosets of
H to p−1 (x0 ), which is bijective if X̃ is path connected.

Proof 1) is clear: p(α̃(1)) = α(1) = x0 .

2) Suppose [β] = [p ◦ σ̃ ]  [α] for some loop σ̃ at x̃0 in X̃ . So β and (p ◦ σ̃ )  α
are path homotopic and hence their lifts β̃ and σ̃  α̃ are path homotopic. In particular,
β̃(1) = σ̃  α̃(1) = α̃(1).
←
Conversely, if α̃(1) = β̃(1), then σ̃ := β̃  α̃ is a loop at x̃0 and
← ←
(p [σ̃ ])  [α] = [p ◦ (β̃  α̃ )]  [α] = [β  ( α  α)] = [β], so [β] ∈ H  [α].
3) That H  [α] → α̃(1) is a well-defined injective map from the set of right
cosets of H and p−1 (x0 ) is a restatement of 2). If X̃ is path connected, x̃1 ∈ X̃
and p(x̃1 ) = x0 , take a path α̃ in X̃ from x̃0 to x̃1 . Then α := p ◦ α̃ is a loop at
x0 = p(x̃0 ) = p(x̃1 ) in X , α̃ is its lift with α̃(0) = x̃0 and α̃(1) = x̃1 , showing that
the map is surjective. 

The lift of a loop may not be a loop. When is it a loop?

Corollary 15.43 With notation as before, a loop α at x0 in X lifts to a loop in X̃ if
and only if [α] ∈ H := p (π1 ( X̃ , x̃0 )).

Proof [α] ∈ H ⇔ [α] ∈ H  [ex0 ] ⇔ α̃(1) = ẽx0 (1) = ex̃0 (1) = x̃0 . 

Any two fibres have the same cardinality.

Corollary 15.44 If X̃ is path connected, then there is a bijective correspondence
between any two fibres p−1 (x0 ) and p−1 (x1 ), x0 , x1 ∈ X .
242 15 Fundamental Groups and Covering Spaces

∼
Proof Let x̃0 ∈ p−1 (x0 ), x1 ∈ p−1 (x1 ) and let H j = p (π1 ( X̃ , x̃ j )). It suffices, by
2) of 15.42, to give a bijection between the right coset spaces π1 (X, x0 )\H0 and
π1 (X, x1 )\H1 . If σ̃ is a path from x̃0 to x̃1 in X̃ , then σ := p ◦ σ̃ is a path in X from
←
x0 to x1 . If σ  ([α]) = [ σ  α  σ ] and if [α] = [p ◦ α̃] ∈ H0 , then
∼ ∼ ∼
← ← ←
σ  ([α]) = [p ◦ σ  p ◦ α̃  p ◦ σ̃ ] = [p ◦ ( σ  α̃  σ̃ )] = p ([ σ  α̃  σ̃ ]) ∈ H1 .
←
Thus σ  maps H0 to H1 and its inverse σ maps H1 to H0 , hence σ  (H0 ) = H1 and
H0  [α] ↔ H1  σ  ([α]) is the required bijection. 

Corollary 15.45 If x̃0 , x̃1 ∈ p−1 (x0 ), then H0 := p (π1 ( X̃ , x̃0 )) and H1 :
= p (π1 ( X̃ , x̃1 )) are conjugate subgroups in π1 (X, x0 ).

Proof Taking x1 = x0 in the previous corollary, σ is a loop at x0 in X and σ  is

conjugation by [σ ] taking H0 to H1 . 

Definition 15.46 Two coverings p : X̃ → X and q : Ỹ → X are isomorphic or

equivalent if there is a homeomorphism h : X̃ → Ỹ such that p = q ◦ h; h is called
an isomorphism. This defines an equivalence relation on coverings of X . A deck
transformation or a covering transformation of a covering p : X̃ → X is an isomor-
phism X̃ → X̃ , that is, an automorphism of X̃ . The set G( X̃ ) = G( X̃ , p) of all deck
transformations form a group (of homeomorphisms of X̃ ).

Example 15.47 1. For the standard covering p : R → S1 , p(x) = e2πı x , if h is a

deck transformation, then e2πı h(x) = e2πı x , x ∈ R, and so n(x) := h(x) − x ∈ Z. By
connectedness, n(x) is a constant. Thus h(x) = x + n for all x ∈ R. Conversely, each
h = h n is a deck transformation and the group of deck transformations is isomorphic
to Z.
2. The two coverings p(z) = z n , q(z) = z m are isomorphic if and only if m = n.
What are the deck transformations for p?
3. For the cover id X : X → X , the only deck transformation is id X .
4. Let p : X̃ := X × {1, . . . , n} → X be the projection. If h̃ : X̃ → X̃ , (x, j) →
(h(x), σ ( j)), is a homeomorphism, then σ ∈ Sn and h is a homeomorphism of X .
If h̃ is a deck transformation, x = h(x). Thus the group of deck transformations is
the permutation group Sn .
5. Any homeomorphism p of X is a covering and any two such coverings p, q are
isomorphic, q−1 ◦ p : X → X being an isomorphism.

Proposition 15.48 Suppose X has a simply connected covering ( X̃ , p). Then every
x ∈ X has a neighbourhood U such that each loop at x in U is path homotopic in
X to the constant loop ex .

Proof Let x ∈ X and let U be an admissible neighbourhood of x. Let α be a loop

at x in U . Let Ũ be a sheet above U and let x̃ ∈ Ũ be such that p(x̃) = x. Let α̃ be
the lift of α in Ũ with α̃(0) = x̃. Since X̃ is simply connected, there is a homotopy
F̃ : α̃ ∼e
˙ x̃ in X̃ and so we have the homotopy F := p ◦ F̃ : α ∼e˙ x in X . 
15.3 Covering Spaces 243

The condition is also sufficient for a simply connected cover to exist.

Definition 15.49 A space X is semilocally simply connected if every x ∈ X has a

neighbourhood U such that each loop at x in U is null homotopic in X , or, equiva-
lently, the homomorphism induced by ι : U → X is trivial.

We conclude the chapter and the book with the basic result on the existence of
simply connected cover. For proofs, see [5, 9, 14].

Theorem 15.50 Let X be a space which is path connected, locally path connected
and semilocally simply connected and let x0 ∈ X .
1) For any subgroup H of π1 (X, x0 ), there is a path connected covering p :
( X̃ , x̃0 ) → (X, x0 ) such that p (π1 ( X̃ , x̃0 )) = H . This covering is unique: If q :
(Ỹ , ỹ0 ) → (X, x0 ) is a covering such that q (π1 (Ỹ , ỹ0 )) = H , then there is an iso-
morphism h : ( X̃ , x̃0 ) → (Ỹ , ỹ0 ).
2) In particular, X has a unique simply connected covering space.

The simply connected covering space X̃ guaranteed by the theorem is called the
universal covering space of X . For example R is the universal cover for S1 , Sn for
RPn , n ≥ 2, and Rn for Tn , n ≥ 1.

Exercise 15

1. Homotopy classes of the circle. If X is path connected and π1 (X ) is abelian,

there is a bijection between π1 (X ) and [S1 , x].
2. Contractible spaces. a) The comb space is contractible.
b) The product of any collection of [0,1] is contractible.
c) X is contractible if and only if X is a retract of Cone(X ).
d) X is contractible if it has a one-point set as a deformation retract.
e) X × Y has the same homotopy type as Y if X is contractible.
f) Cone(X ) is contractible for any space X .
g) A retract of a contractible space is contractible.
3. Abelian fundamental groups. If x0 , x1 ∈ X and X is path connected, then
π1 (X, x0 ) is abelian if and only if for any two paths σ and τ in X joining x0 to x1 ,
the induced isomorphisms are equal: σ  = τ  .
4. Homotopic maps. a) If f : S1 → S1 is not homotopic to the identity map,
then f (z) = −z for some z ∈ S1 .
b) Any two maps f, g : X → Cone(Y ) are homotopic.
c) The inclusion S1 → R2 \ {0} and idS1 are not null homotopic.
5. Vector fields. A vector field in S ⊂ Rn is a continuous function v : S → Rn .
Geometrically, we think of v(x) as the oriented line segment from x to v(x). If v is
a vector field in B2 , then there are points x, y ∈ B2 with v(x), v(y) point directly
inwards and directly outwards, respectively.
6. Homotopy equivalents. Spaces in each of the following sets are homotopically
equivalent (see pictures in Chap. 14 (14.1, 14.3)):
244 15 Fundamental Groups and Covering Spaces

a) a punctured plane, a punctured disc, an annulus, a circle;

b) the plane or a disc with two holes, two circles touching externally.
7. The fundamental group. a) Any star-shaped set S in Rn is path connected
and simply connected.
b) X is simply connected if and only if each continuous map S1 → X extends to
B2 → X .
c) Determine the fundamental group of the following spaces:
i) exterior of the open (closed) unit disc;
ii) a circle with a radius segment;
iii) a circle together with a tangent at a point;
iv) R2 \ {(x, y) : x y = 0, x 2 + y 2
= 0};
v) R3 with a line (two lines) removed;
vi) R3 \ (D × R).
8. Homotopy type. a) A subspace E of X is homotopically equivalent to X if
and only if there is a map r : X → E such that r ◦ ι is homotopic to id E , where
ι : E → X is the inclusion map.
b) If X and X  have the same homotopy type as Y and Y  respectively, then X × X 
has the same homotopy type as Y × Y  .
9. Retracts and deformation retracts. a) Any compact convex subset of Rn is
a deformation retract of Rn .
b) E × Y is a retract of X × Y for any Y if E is a retract of X .
10. Loop homotopy in the circle. Any loop at 1 ∈ S1 is path homotopic to a
unique εn , n ∈ Z, where εn (s) = e2πıns .
11. Covering spaces. a) p : C → C∗ , p(z) = e z , is a covering map.
b) Find a covering space and a simply connected covering for the annulus Aa,b =
{z ∈ C : a < |z| < b} for 0 < a < b. What is π1 (Aa,b )?
12. Locally simply connected, not simply connected. Give an example of a
path connected space X such that every point x ∈ X has a simply connected neigh-
bourhood, but X itself is not simply connected.
13. Deck transformations. Find the deck transformations: Sn over RP n .
14. Fundamental groups of topological groups. Let G be a topological group
and consider the fundamental group π(G, e) of G at the identity element e. Let
L(G, e) be the set of loops at e in G.
a) L(G, e) is a group with multiplication: (αβ)(s) = α(s)β(s).
b) This product respects homotopy: α ∼α ˙  , β ∼β
˙  imply αβ ∼α
˙  β  . The loop classes
form a group under this multiplication [α][β] = [αβ].
c) This group is abelian: [α][β] = [β][α].
d) [α][β] = [α]  [β], hence π1 (G, e) is abelian.
e) G, H are topological groups and p : G → H is a covering map and a homo-
morphism. The fibre p −1 (e H ) is a subgroup and π1 (H, e H ) p −1 (e H ).
15.4 Biographical Notes 245

15.4 Biographical Notes

15.4.1 Poincaré

Henri Poincaré (1854–1912) was a versatile, multi-faceted French genius. He

was a mathematician, theoretical physicist, a philosopher of science and a mining
engineer—a veritable polymath and a universalist. In mathematics, he is considered
as the father of Algebraic Topology and originator of the theory of several complex
variables. He published a sequence of papers on Analysis situs, as it was called then,
from 1895, introduced and studied homotopy, fundamental group, homology etc.
He created the Qualitative theory of differential equations, discovered and studied
automorphic functions from 1880s (Automorphic forms lie at the heart of current
profound developments in Number Theory.), showed that three body problem—the
problem of relative motions of the sun, moon and the earth (that led him to the
creation of Algebraic Topology)—is not integrable and wrote two monographs on
celestial mechanics, which contained the genesis of chaos theory, bifurcation and
dynamical systems. He introduced group theory in Physics and was the first to study
the Lorentz group, proposed gravitational waves (1905), discovered mass-energy
relation (1900), was one of the discoverers of Relativity. Einstein considered him a
pioneer in relativity.
Gaston Darboux says “Partout, quand on lui demandait de résoudre une diffi-
culté, sa réponse partait avec la rapidité de la flèche. Lorsqu’il écrivait un Mémoire,
il le rédigeait tout d’un trait, se bornant à quelques ratures, sans revenir sur ce qu’il
avait écrit’, [15]. (“When asked to solve a difficulty, his answer always came out with
the speed of an arrow. When he wrote a Memoir, he wrote it all at once, restricting
himself to a few cross outs, not going back to what he had written.”) Yet, lucidity
and clarity of exposition shine through all his writings.
Interestingly, on Mittag-Leffler’s request, Poincaré himself wrote a review of his
own work upto 1901, [16]. In this, about his work on topology he says: “Quant à
moi, toutes ]es voies diverses oh je m’étais engagé successivement me conduisaient
à l’Analysis Sitûs. J’avais besoin des données de cette science pour poursuivre mes
études sur les courbes définies par les équations différentielles et pour les étendre
aux équations différentielles d’ordre supérieur et en particulier à celles du problème
des trois corps. J’en avais besoin pour l’étude des fonctions non uniformes de 2
variables. J’en avais besoin pour l’étude des périodes des intégrales multiples et
pour l’application de cette étude au développement de la fonction perturbatrice. Enfin
j’entrevoyais dans l’Analysis Sitûs un moyen d’aborder un probléme important de la
théorie des groupes, la recherche des groupes discrets ou des groupes finis contenus
dans un groupe continu donné. C’est pour routes ces raisons que je consacrai à
cette science un assez long travail.” (“As for me, all the diverse paths on which I
was successively engaged have led me to Analysis Situs. I had need of the ideas of
this science to pursue my studies on curves defined by differential equations and for
extending these to higher order differential equations and in particular to those of
the three body problem. I had need of it for the study of multivalued functions of 2
246 15 Fundamental Groups and Covering Spaces

variables. I had need of it for the study of periods of multiple integrals and for the
application of this study to the development of the perturbation function. Finally I
glimpsed in Analysis Situs a means of attacking an important problem in the theory
of groups, the search for discrete or finite groups contained in a given continuous
group. It is for all these reasons that I devoted to this science a fairly long work.”)
His collected works in topology is [17]; English translation may be found in [3]. His
writings had to keep pace with the constant pouring of ideas in various areas. He
relied more on his great intuition than rigorous proofs. See [18].
Poincaré conjecture, Poincaré duality, Poincaré-Brirkhoff-Witt theorem, Poincaré-
Hopf theorem, Poincaré metric, Poincaré upper half-pane, Poincaré group, Poincaré-
Bendixon theorem, Poincaré recurrence theorem, Poincaré-Hilbert series are some
of the mathematical objects named after him. In France, a mathematical institute, a
mathematical journal, a university are all named after him. A moon crater and an
asteroid are also given his name.
It would be nice to conclude with a mention of the Poincaré conjecture. In Analysis
situs [19] he conjectured that if a 3-manifold M has the same homology groups as the
sphere S3 , then it is homeomorphic to S3 . But later, in his fifth complement [2], he gave
a counter example—a homology 3-sphere with a nontrivial fundamental group) and
gave the revised (correct) conjecture: Every simply connected closed 3-manifold
is homeomorphic to S3 . After almost a century, this was solved in the affirmative by
the Russian mathematician Grigori Perelman in 2002–2003 (using analytic methods
of Ricci flow developed by Richard Hamilton). The higher dimensional conjectures
had been solved earlier: Steve Smale for n ≥ 5 in 1961 and for n = 4 by Michael
Freedman in 1982. Smale, Freedman and Perelman were chosen for Fields medal—
Perelman declined it.
Here is how Poincare stated it: (V is the 3-manifold under consideration.) “One
question remains to be dealt with: Is it possible for the fundamental group of V
to reduce to the identity without V being simply connected” (i.e. without being
homeomorphic to S3 ?) He concludes with the words: “However, this question would
carry us too far away.”
We sign off with the original in French: “Il resterait une question à traiter: Est-il
possible que le groupe fondamental de V se réduise à la substitution identique, et que
pourtant V ne soit pas simplement connexe? ... Mais cette question nous entraînerait
troploin.”

References

1. Poincaré,H., Sur l’analysis situs, C. R. Acad. Sci. 118, 663–666, Oeuvres; vol. VI, 189–192
(1892)
2. Poincaré, H., Cinquième Complément à l’Analysis Situs, Rend. Circ. Math. d. Palermo, 18,
45–110; Oeuvres, Vol. VI, 435–498 (1904)
3. Poincaré, H.: Papers on Topology: Analysis Situs and Its Five Supplements. Translated by J.
Am. Math. Soc, Stillwell (2010)
4. Gamelin, T.W., Greene, R.E.: Topology. Dover (1999)
References 247

5. Hatcher, A.: Algebraic Topology. Cambridge University Press (2003)

6. Spanier, E.: Algebraic Topology. Tata-McGraw-Hill (1966)
7. Seifert, H.: Konstruktion drei dimensionaler geschlossener Raume, Berichte Sachs. Akad.
Leipzig, Math.-Phys. Kl. 83, 26–66 (1931)
8. van Kampen, E.R.: On the connection between the fundamental groups of some related spaces.
Am. J. Math. 55, 261–267 (1933)
9. Massey, W.S.: Algebraic Topology. Springer, An Introduction (1990)
10. Munkres, J.R.: Topology, 2nd edn. Pearson Education, Asia (2001)
11. Seifert, H., Threlfall, W.: A Textbook of Topology (Translated from German by M.A. Goldman).
Academic Press (1980)
12. Dieudonné, J.A.: A history of algebraic and differential topology. 1900–1960. Birkhäuser
(1989)
13. Armstrong, M.A.: Basic Topology. Undergraduate Texts in Mathematics, Springer International
(2004)
14. Singer, I.M., Thorpe, J.A.: Lecture Notes on Elementary Topology and Geometry. Springer
(2003)
15. Darboux, G., Éloge historique d’Henri Poincaré, Académie des Sciences (1913). https://fr.
wikisource.org/wiki/Eloge historique d’Henri Poincaré par Gaston Darboux.)
16. Poincaré, H.: Analyse des travaux scientifiques de Henri Poincaré faite par lui-meme (1901).
Acta Math. 38, 1–135 (1921). https://doi.org/10.1007/BF02392063
17. Poincaré, H., Oeuvres, Tome VI, Garniers, R., Leray, J. (eds.) Gauthiers-Villars (1953)
18. Kulpa, W.: Poincaré and domain invariance theorem. Acta Univ. Carol. Math. et Phys. 39,
127–136 (1998)
19. Poincaré, H.: Analysis situs. J. de l’École Polytech. 2(1) 1–123, Oeuvres; vol. VI, 193–288
(1895)
Appendix: Selected Exercises—Suggestions
and Hints

To the reader: Do not look at these until you have tried hard enough. Sufficient
investment of time and thought on a problem is important. Even if you fail to get a
solution, the effort you have put in will yield its own fruits. Suggestions given here
may not be the only, nor even the best, line of attack. You can possibly come up with
a different (and perhaps better) way of doing things. Doing is learning! So, get back
to thinking!
Chapter 1
2(d) Apply IVT to x 2 − a on a suitable interval.
2(f) Suffices to show: if b lies between f  (0) and f  (1), say f  (0) < b < f  (1),
then there is an a between 0 and 1 such that f  (a) = b. g(t) = bt − f (t) has a
maximum on [0,1] and the maximum cannot be attained at 0 or 1. So there is 0 <
a < 1 with g  (a) = 0.
3(a) Draw graphs.
3(b) Think geometrically.
Chapter 2

1. First extend Cauchy–Schwarz to infinite sums.

2(b) Apply the mean value theorem to f (x) = x/(1 + x) to get f (x + y) ≤
f (x) + f (y). This yields the triangle inequality for d  .
2(c) It is sufficient if some a j > 0 and the other ak ≥ 0. It is also enough if one
d j is a metric, a j > 0, the other dk are semi-metrics and ak ≥ 0.
2(d) d-convergence is equivalent to d f -convergence.
2(e) Convergence in dsup implies convergence in d1 . Find a sequence that is d1 -
convergent, but not dsup -convergent.
8. What is d(A, B) if A ⊂ B?
2
9. If each matrix is identified with a point in Rn , how are these metrics related to
2
familiar metrics on Rn ?
10. If a collection of open rectangles cover a disc, take one of the rectangles R1 .
Look at a boundary point p of R1 . If p is in R2 , can R1 and R2 be disjoint?
© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer 249
Nature Singapore Pte Ltd. 2022
K. Parthasarathy, Topology, La Matematica per il 3+2 134,
https://doi.org/10.1007/978-981-16-9484-4
250 Appendix: Selected Exercises—Suggestions and Hints

12(a) Is |x y| p = |x| p |y| p ?

12(c) Take a positive integer x first and use its prime factorisation.
Chapter 3

1(g) Consider vertical, horizontal and inclined straight lines.

2. Triangle inequality for d H : d(x, z) ≤ d(x, y) + d(z, y), x ∈ A, y ∈ B and z ∈
C. Taking inf over y ∈ B, we get
d(x, z) ≤ d(x, B) + d(z, B) ≤ d(A, B) + d(C, B)
≤ d H (A, B) + d H (B, C).
Taking inf over z ∈ C and then sup over x ∈ A, we get d(A, C) ≤ d H (A, B) +
d H (B, C). Similarly, d(C, A) ≤ d H (A, B) + d H (B, C) follows from d(z, x) ≤ d H
(A, B) + d H (B, C). Together, these yield
d H (A, C) ≤ d H (A, B) + d H (B, C).
6(b) What are the finite intersections of these subbasic open sets?
7(b) If X has a smallest element a0 , then [a0 , b) is open for any b ∈ X . Similarly,
consider the case when X has a largest element b0 .
7(c) and (d) If a, b, c ∈ X and a < b < c, then the interval (z, w) is open, where
z = (a, b), w = (a, c). If a < c and z = (a, b), w = (c, d), what is the interval
(z, w) in the product?
7(f) Note that 0 is the smallest and 1 is the largest element.
7(g) The dictionary order topology on [0, 1] × [0, 1] is smaller than the subspace
topology from R × R with the dictionary order. { 21 } × [0, 21 ) is open in the subspace
topology, but not in the order topology.
7(h) Look at g).
10(h) For each interval I = (q − 1/k, q + 1/k), q ∈ Q, k ∈ N, that meets E (an
infinite subset of R), choose a point qk ∈ E ∩ I . Call Ak the set of all such qk and
let A = ∪Ak .√ Then Ā = E.
12(e) q + 2 is irrational for all q ∈ Q
12(g) When is E c dense?
12(h) A dense set meets all basic open sets.
14. Determinant function is continuous!
15(b) E is closed if and only if E = {x ∈ X : d(x, E) = 0}.
Chapter 4
2(c) Let α = sup{a : E ∩ (−∞, a) is countable} and β = inf{a : E ∩ (a, ∞)
is countable}. Consider the cases α < β, α = β and α > β.
4(e) Let E be a closed set in Rn and write D = Qn . Take A = E \ (D ∩ int E).
Then Ā = E (D c ∩ int E is dense in int E) and int A is empty (E c is dense in
(int E)c and D c ∩ int E is dense in int E). Hence ∂ A = E. (Note: the proof uses
only the fact that D and D c are both dense.)
5(d) f (x) = e2πı x . What is the inverse image of a small arc around (1,0)? 1 − 1/n
has no limit in [0,1), but z n = f (1 − 1/n) has limit (1,0) in S1 . Come back to this
after studying compactness and connectedness.
5(e) f (x) = d(x, U c ). χQn .
Appendix: Selected Exercises—Suggestions and Hints 251

6(l) If X is not discrete, find Y and a discontinuous f : X → Y .

7(d) What is f (ax) when a ∈ N, a ∈ Z and a ∈ Q?
7(n) What about the radial extension?
9(c) For q ∈ Q, let Aq = {x : f (x) > q}, G q = X \ Āq . G = ∩q∈Q G q is the
required dense G δ . √
11(b)(iii) What about Z + 2Z in R?
11(c)(iii) If H is open, what about cosets of H ?
12(a)(ii) If H is not cyclic, it has no smallest positive element.

Chapter 5
1(a) Take R with the topology T = {V : 0 ∈ V } ∪ {∅}. Then {0} is compact, but
what is its closure?
Take Z with the usual (discrete) topology and X = Z ∪ {±1/2}. Consider the
topology on X so that neighbourhoods of 1/2 are X and Z ∪ {1/2}. Similarly for −1/2.
Then K + = Z ∪ {1/2} and K − = Z ∪ {−1/2} are compact, but their intersection is
not.
1(c) Use f.i.p. for the collection {K i ∩ V c }.
1(d) Show that C is closed and bounded. For boundedness, note that uniform
convergence implies f (K ) ∪ (∪∞ n f (K m )) is bounded for large n. To show that C
is closed, take yk ∈ C, yk → y. If infinitely many yk lie in a single f n (K ), we are
done. If not, going to subsequences wherever needed, yk = f n k (xk ) with xk → x.
Then y = f (x).
3(a) δ = d(x0 , f (X )) > 0, ∃xn , d(xn , xm ) ≥ δ, m = n, xn+1 = f (xn ).
4(b) Map a line to a generator or wind it around as a helix.
7(a) Find a space in which no two nonempty open sets are disjoint.
7(c) Here is an elementary proof. If X is not compact, there is a sequence {xn } of
distinct elements with no limit points. Suppose the xn are separated: inf{d(xn , xm ) :
n = m} > δ > 0. For any x, there is at most one n with d(x, xn ) < δ/2. Define
f (x) = n(1 − 2d(x, xn )/δ) if there is such an n. Otherwise define f (x) = 0. Then
f is continuous and f (xn ) = n.
Suppose inf{d(xn , xm ) : n = m} = 0. Then there is a subsequence {xn k } with
limk,l d(xn k , xnl ) = 0, hence g(x) = limk d(x, xn k ) exists and nonzero for all x. But
g(xnl ) → 0. Thus f (x) = 1/g(x) is continuous and unbounded.
The more common proof defines f (xn ) = n for all n and invokes Tietze’s exten-
sion theorem (Chap. 12) to extend f to X .
7(d) d(x, y) + | f (x) − f (y)| with f as in (c).
11. Try finding a convergent subsequence of the sequence {en } of standard unit
vectors.
14. Heine–Borel–Lebesgue again. K is bounded implies the set σ (K) is bounded.
That it is closed follows from two facts: (1) every sequence in K has a convergent
subsequence; (2) the set of singular matrices is closed.
252 Appendix: Selected Exercises—Suggestions and Hints

Chapter 6

1(e) d convergence is the same as point-wise convergence.

5(a) Think of a hyperbola.
6. c1 is identified with c2 in the quotient and the shaded half discs together form
a neighbourhood of c1 = c2 in the quotient. Similar remarks for a1 , a2 and b1 , b2 . A
small disc around an interior point of the rectangle remains a disc (albeit bent) in the
quotient.
7. It is a fact that the Klein bottle cannot be embedded in R3 . So drawing a picture
of the surface on paper is difficult.
8(a) The quotient map Rn+1 \ {0} → RPn restricted to Sn identifies antipodal
points on the sphere.
8(b) How do you picturise this? It is to be shown that the quotient S1 /R is home-
omorphic to S1 , where z Rw if and only if w = −z. The map f (z) = z 2 , S1 → S1 ,
is surjective, continuous and satisfies f (z) = f (−z). So it lifts to a continuous map
f˜ : S1 /R → S1 , and there is a continuous bijection h : S1 /R → S1 . h is a homeo-
morphism.
9. S(X ) is a quotient of Cone(X ) got by pinching the ‘base’ X of the cone to a
single point. More generally, S(Sn ) is Sn+1 . (And Cone(Sn ) is the n + 1-ball.)
12. A little explanation about ‘attaching spaces’ may be helpful. To simplify
things, suppose X, Y are disjoint to start with so that we don’t have to ‘disjointify’
them. On X ∪ Y , we take an equivalence relation as follows. For z, w ∈ X ∪ Y , z Rw
if and only if one of the following holds: (a) z = w; (b) z ∈ E and w = f (z); (c)
w ∈ E and z = f (w); (d) z, w ∈ E and f (z) = f (w). The equivalence classes may
be described like this: For x ∈ / E, the class [x] is the one-point set {x}. Similarly, for
points not in the range of f : [y] = {y} for y ∈ Y \ f (E). For x ∈ E, [x] = { f (x)} ∪
f −1 ( f (x)) = { f (x)} ∪ {x  ∈ E : f (x  ) = f (x)}. In the same way, for y ∈ f (E),
[y] = {y} ∪ f −1 (y)
If X, Y are not disjoint, let X  = X × {1} and Y  = Y × {2}, so that X  , Y 
ar disjoint. Take E  = E × {1} and f  (x, 1) = ( f (x), 2). Proceed as above with
X  , Y  , E  , f  in place of X, Y, E, f .
Chapter 7
1(b) iii) implies i). Let y ∈ Y . To show that f −1 (y) is compact. Consider
the restriction map f y := f f −1 (y) . f y × id Z = ( f × id Z ) f −1 (y)×Z : f −1 (y) × Z →
{y} × Z of the closed map f × id Z is also closed. If p : {y} × Z → Z is the projec-
tion, p f y is also closed. But p f y is the projection f −1 (y) × Z → Z . The compactness
of f −1 (y) is now a consequence of Kuratowski’s theorem. 
2. Define f n : [0, 1] → [0, 1] by f n (x) = xn if x = n≥1 xn /2n , xn ∈ {0, 1}. If
f n k is any subsequence, define yn = 1 if n = n 2k for some k and yn = 0 otherwise.
Then f n k (y) = (0, 1, 0, 1 . . .) does not converge. Thus no subsequence of { f n } is
convergent, in the product topology.
3. First consider A = {a}. Think of a region containing the x-axis in the plane
which thins out asymptotically to the x-axis.
Appendix: Selected Exercises—Suggestions and Hints 253

Chapter 8

2(c) If E = {xn } has no limit points, then E is closed. For each n there is a
neighbourhood Vn of xn , with xk ∈ / V, k = n. The Vn and E c form a countable open
cover without a finite subcover.
If there is a countable open cover {Vn } without a finite subcover, choose xn ∈ / ∪n1 Vk .
Can the set of xn have a limit point?
3(c) X be infinite, Hausdorff and E the set of isolated points. If E is finite,
inductively get {xn } in E c and open Vn with xn ∈ Vn and xk ∈ / Vn for k < n. xn ∈ /
{xk : k = n} and {xn } is infinite discrete.
5(a) If V is open, x ∈ V , to show: there is a continuous f with x ∈ {y : f (y) =
0} ⊂ V . This is almost immediate from the definition.
5(b) Assume f ≥ 0. To show: for each x0 ∈ X with f (x0 ) > 0 and each 0 < a <
f (x0 ), there is a continuous g ≤ f such that g(x0 ) ≥ a. There is a neighbourhood U
of x0 such that f (x) ≥ a on U . Choose a continuous function g such that g(x0 ) = a,
g = 0 on U c and 0 ≤ g ≤ a.
7(c) Write A = ∩Vn , Vn open, V n+1 ⊂ Vn with B ⊂ V1 . Apply Urysohn for An , B
c

for each n to get an f n . Take f = f n /2 . For the second part, get a g for B as in
n

the first part. Take h = f /( f + g).

10(d) For x in the removed middle-third interval of [ 3kn , k+1 3n
], define f (x) =
1
2.3n
− |x − k+1/2
3n
| for each k, n and f (x) = 0 otherwise. You can check that f is
actually Lipschitz, | f (x) − f (y)| ≤ |x − y|.
11(a) Use dyadic (=binary) expansions in [0,1].
11(b) With ϕ as in Exercise 10, ψ ◦ ϕ −1 is one such map.
11(c) {0, 2}N is homeomorphic to {0, 2}N × {0, 2}N - both are products of count-
ably infinite copies of {0, 2} and so are homeomorphic, Ex. Chapter 6 (‘associativity
of products’).
11(d) Let f be a continuous map of C onto [0,1], as in b) and let h be a homeo-
morphism of C onto C × C as in d). Take ( f × f ) ◦ h.
11(e) Easy. Extend by linearity on each middle-third interval removed in the
construction of C. (Want a sledgehammer? Tietze!)
11(f) Take a continuous map g of C onto [0, 1] × [0, 1] as in d). Write p1 , p2 for
the projections of [0, 1] × [0, 1] onto [0,1]. Extend each p j ◦ g to g j : [0, 1] → [0, 1]
and take g1 × g2 : [0, 1] → [0, 1] × [0, 1].
11(g) If it is, it is a homeomorphism and [0,1]; but [0, 1] × [0, 1] are not homeo-
morphic.
12. Suppose x, y ∈ G, x = y. If {x} is closed, there is a neighbourhood V of e
such that V = V −1 and yV V ⊂ {x}c . Then x V, yV are disjoint neighbourhoods of
x, y, respectively.
Every neighbourhood of e contains a closed neighbourhood. In fact, if U is a
neighbourhood of e choose a symmetric neighbourhood V of e such that V V ⊂ U .
Then V̄ ⊂ U . Translations show that every neighbourhood of any point in G contains
a closed neighbourhood.
254 Appendix: Selected Exercises—Suggestions and Hints

Chapter 9
2(a) Is {(x, sin(1/x)) : 0 < x < 1} ∪ {(0, 0)} connected?
2(e) Take a ∈ Ac , b ∈ B c and (ξ, η) ∈ X × Y \ (A × B). Find a connected set
containing (a, b) and (ξ, η). Use the following facts: (i) X × y and x × Y are con-
nected for all x ∈ X, y ∈ Y . (ii) The union of connected sets with nonempty inter-
section is connected.
3. A punctured disc is connected.
4(c) (0,1] is a component of B, not homeomorphic to any component of A.
f (1) = 2, f (x) = x otherwise. Map (0, 1] ∪ (3, 4) in B to (0,1) in A and translate
all the other Bn to Bn − 3 to get g.
4(d) If a, b ∈ A, a < b, take a < c < b. A = ((−∞, c) ∩ A) ∪ ([c, ∞) ∩ A) is
a disconnection.
A continuous image of a connected space is connected.
6(d) Let C be a connected subset of a countable metric space X . Take a ∈ C. The
function f (x) = d(x, a) is a continuous function on C and so f (C) is a countable
connected set of real numbers.

Chapter 10
1(a) What is the subspace topology on the x-axis?
1(b) Separable. First countable, even second countable if X is countable and
not first countable if X is uncountable. Compact, hence Lindelöf. What about the
co-countable topology?
2(e) Recall Weierstrass approximation theorem.
2(f) For ∞ , observe that the set of sequences of zeros and ones is uncountable
and the distance between any two of these is 1. For the last part, what can you say
about the set of characteristic functions of subsets of X ?
2(h) The set D of all functions with finite range is dense. Find a countable subset
of D which is also dense.
2(i) Spaces in which no two nonempty open sets are disjoint?
3(d) Can you find an uncountable discrete subset? (It is also compact. Can you
prove this using the result that in any totally ordered set with lub property, any closed
interval with the order topology is compact (which is proved in [1]).)
4(d) The closed set {1/n : n ∈ N} and 0 cannot be separated.
4(e) As X is second countable, we need only to prove: X is sequentially compact
if and only if every countable open cover has a finite subcover. Suppose {Vn } is an
open cover with no finite subcover. Take xn ∈ X \ ∪n1 Vk . {xn } has no convergent
subsequence. For the converse, if a sequence {xn } has infinitely many distinct points,
it has a limit point and there is a subsequence converging to this limit point.
6(f) X regular, Lindelöf, A, B disjoint, closed. {U : U open, Ū ∩ B = ∅} is a
cover for A and a similar open cover for B, These and X \ (A ∪ B) form an open
cover for X . So there is a countable subcollection Un , Ūn ∩ B = ∅, covering A and
an analogous collection {Vn } covering B. Make them disjoint as in the proof of 10.13
and take the unions.
Appendix: Selected Exercises—Suggestions and Hints 255

Chapter 11
1. Is [a, b) compact in R ? Is the unit ball in 2 compact?
2(c) What is the one-point compactification of [0, 1) ∪ (1, 2]?
4. Use the function f (x) = e−1/(a−x) for x > a and 0 otherwise to get a smooth
function on R that vanishes outside [a, b]. Use this, in turn, to get a smooth function
that is 1 for x < a and 0 for x > b (a < b). This yields a smooth function on Rn that
is 1 on a ball and vanishes outside a bigger ball. Use such functions.
5. Use smooth Urysohn to get smooth partitions of unity, just as we did in the con-
tinuous case. (Reference: suitable advanced calculus books or differential topology
books.)
6. For x ∈ f −1 (K ), let Vx be a precompact neighbourhood. f (Vx ) form an open
cover of K , so finitely many V j = Vx j cover K . A = ∪V̄ j is compact and K ⊂ f (A).
C = A ∩ f −1 (K ) is compact and f (C) = K .
7(b) What are the neighbourhoods of the point [Z] in the quotient?
8(a) Let f (x) = lim f n (x), f n continuous. For each m, n, let Vmn be the interior
of the set ∩k≥m {x : | f m (x) − f k (x)| ≤ 1/n}, Wn = ∪m≥1 Vmn and let E = ∩n≥1 Wn .
f is continuous at each point of E, E c is of first category and the points of continuity
of f is dense. (E is open, dense.)
8(b) f is continuous at x if and only if f −1 (In (x)) is an open set containing x
for all n, where In (x) = ( f (x) − 1/n, f (x) + 1/n). Thus the set C( f ) of points of
continuity of f is ∩Vn , where Vn is the union of all open intervals In with length
(In ) < 1/2n.
8(c) No countable dense set in R is a G δ .

Chapter 12
2(b) C[−1, 1] is d1 -dense in L [−1, 1].
1

5(d) Suppose X is complete and let {An } be a Cauchy sequence of closed bounded
sets in the Hausdorff metric d H . Let A be the set of limits of all sequences {an } in
X with an ∈ An for all n. Show that A is the d H -limit of {An }. For this, it is to be
shown that, for any ε > 0, An ⊂ A(2ε) and A ⊂ An (2ε) for large n.
Take ε > 0, fix N with d H (An , Am ) < ε m, n ≥ N . A ⊂ A N (2ε).
Next, choose Nk < Nk+1 for all k such that d H (Am , An ) < ε/2k for m, n ≥ Nk .
Observe that for each a ∈ An and k with n ≥ Nk , there is a bm ∈ Am such that
d(a, bm ) < ε/2k if m ≥ n ≥ Nk .
Using the observation, for each a ∈ An with n ≥ N1 , inductively construct a
sequence {xl } with xl ∈ Al such that d(xi , x j ) < ε/2k for Nk ≤ i, j ≤ Nk+1 and
d(a, x N2 ) < ε/2. Then {x j } is a Cauchy sequence in X and converges to a b with
d(a, b) < 2ε. An ⊂ A(2ε) if n ≥ N1 .
6(b) If (X, d) is not compact, it is not countably compact and there is a decreasing
sequence {Fn } of closed sets with
 empty intersection. Define f n (x) = d(x, Fn ) for
each n and define ρ(x, y) = 2−n ρn (x, y) where ρn (x, y) = [| f n (x) − f n (y)| +
256 Appendix: Selected Exercises—Suggestions and Hints

(min{ f n (x), f n (y)})d(x, y)]. ρ is a metric that is equivalent to d and diam ρ (Fn ) ≤
2−n . If xn ∈ Fn , {xn } is a ρ-Cauchy sequence and has no limit as ∩Fn is empty.
6(d) Let ε > 0 and let {x1 , . . . , xn } be an ε-mesh for X . Let {A j }21 −1 be the
n

family of all nonempty subsets of {x1 , . . . , xn }. Show that {A j } is an ε-mesh for

(Bc (X ), d H ).
7. If A is the set points which are not an isolated, Da = X \ {a} is open, dense
for each a ∈ A and ∩Da is the set of isolated points.
9(b) Every sequence has a convergent subsequence.
10(d) Apply Banach to the closed ball {x : d(x, x0 ) ≤ δ}, δ < r .
17. The required f is the fixed point of T : C[0, 1] → C[0, 1] defined by T f (x) =
1
g(x) + 0 k(x, y) f (y)dy.

Chapter 13
5. Let r be the retraction of Rn onto the unit ball and take g = r − f ◦ r . Then g
maps a suitable ball into itself. Apply Brouwer to g. (See Mawhin [2].)
9. Stereographic projection.
10. Assume n > m and h : Rn → Rm is continuous and injective. Consider the
restriction h|Sn−1 .
12(a) Borsuk–Ulam.
12(b) Sn \ { p} is homeomorphic to Rn .
Chapter 14
1(b) Let X be Haudorff and r : X → A be a retraction. If x ∈ Ā, x ∈ / A, there are
disjoint neighbourhoods U of x and V of r (x) with r (U ) ⊂ V . There is a y ∈ A ∩ U
and a = r (a) ⊂ V , a contradiction.
2(a) When are two constant maps homotopic?
2(b) If f has an extension F to Cone(X ), use the quotient map q and F to get a
null homotopy. For the converse, use the homotopy and the fact that q restricted to
X × {0} is a homeomorphism.
4. What does the given condition say about the line segment [α(s), β(s)]?
Chapter 15
2(f) Move each point of the cone to the vertex along a generator.
4(b) Move f (x) to g(x) via the vertex, along generators.
5. Use 4(c).
Remark on vector fields on spheres. A vector field v on Sn is a tangent vector
field if v(x) ⊥ x for all x ∈ Sn . It is very easy to see that there are nonvanishing
tangent vector fields on S2n−1 for all n. A famous result due to Poincaré (1885) says
this is not the case on S2 . This was later extended to all even dimensional spheres.
This is known as the hairy ball theorem. It says that any tangent vector field on
Appendix: Selected Exercises—Suggestions and Hints 257

S2n necessarily vanishes at some point. (‘You can’t comb a porcupine’!) Milnor’s
beautiful proof of this is in [3]. For S2 , a simple proof using Sperner’s lemma is given
in [4].
7(c) iv) Is it simply connected?
7(c) vi) Plane with a hole.
9(b) What about outward radial projection?
11. What is the relation between the annulus and S1 × (a, b)?

References

1. Munkres, J.R.: Topology, 2nd edn. Pearson Education, Asia (2001)

2. Mawhin, J.: Simple Proofs of the Hadamard and Poincaré-Miranda theorems using the
Brouwer fixed point theorem. Am. Math. Monthly 126, 260–263 (2019). https://doi.org/10.
1080/00029890.2019.1551023
3. Milnor, J.: Analytic proofs of the “Hairy Ball Theorem⣞ and the Brouwer fixed point Theorem.
Am. Math. Monthly 85, 521–524 (1978). https://doi.org/10.2307/2320860
4. Jarvis, T., Tanton, J.: The Hairy ball theorem via Sperner’s lemma. Am. Math. Monthly 111,
599–603 (2004). https://doi.org/10.2307/4145162
Index

Symbols ♦n , unit n-sphere in 1-norm, 184

β X , Stone-Čech compactification, 119 1 , space of absolutely summable sequences,
k-hemisphere, 197 9
p-adic metric, 19 2 , space of square summable sequences, 10
Čech, Eduard, 103, 119 ∞ , space of bounded sequences, 9
(a, b), open interval in R, 4 Bn , the open Euclidean unit ball in Rn , 183
(a, b], half-closed interval in R, 4 C, the set of complex numbers, 2
+-alternating simplex, 198 C∗ , nonzero complex numbers, 2
−-alternating simplex, 198 I, the set of irrationals, 31
B(x, r ), open ball, 12 N, the set of natural numbers, 33
E  , the derived set of E, 42 R, the set of real numbers, 1
E c , the complement of E ⊂ X , 23 R+ = [0, ∞), the set of nonnegative real
Fσ set, 36 numbers, 8
G L(n, R); the space of real n × n nonsingu- Rn , Euclidean n-space, 8
lar matrices, 35 R × R is not normal, 112
G δ set, 36 S1 , the unit circle, 3
M(n, R), the space of n × n real matrices, Sn , the Euclidean unit n-sphere, 51
18 Z, the set of integers, 2
S L(2, R) action, 91 Bn , the closed Euclidean unit n-ball in Rn ,
51
S L(n, R), the space of n × n real matrices
C, Cantor ternary set, 45
of determinant 1, 35
C [0, 1], the space of continuous functions on
S O(n) real orthogonal n × n matrices of unit
[0,1], 17
determinant, 129
U (n) is path connected, 130
T1 -space, 107
U (n), the unitary group, 130
T2 -space, 108
P (X ), the power set of X , 43
T3 -space, 116
∂ E, the boundary of E, 45
T4 -space, 111
n , unit n-sphere in ∞-norm, 184
T3 1 -space, 117 ε- mesh, 63
2
[a, b), half-open interval in R, 4 ε-δ definition of continuity, 10
[a, b], closed interval in R, 2 d ∗ , bounded metric associated to a metric d,
α  β, product of paths α and β, 48 9
Ā, the closure of A, 29 k-hemisphere, 196
B̄(x, r ), closed ball, 12 k-skeleton, 197
Stone-Čech compactification, 119, 121 p-adic metric, 8
n , closed unit n-ball in 1-norm, 184 p-adic number field, 172
n , closed unit n-ball in ∞-norm, 184 Analysis situs, 208
© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer 259
Nature Singapore Pte Ltd. 2022
K. Parthasarathy, Topology, La Matematica per il 3+2 134,
https://doi.org/10.1007/978-981-16-9484-4
260 Index

of Poincaré, 207, 223, 224 on Cauchy’s Cours d’Analyse, 54

Trennungsaxiome, 107 Absolute neighbourhood retract, 221
O(n), the orthogonal group, 135 Accumulation point, 42
Figures Adjacent vertices, 198
Bd(0,0) ((1, 0), 2), 8 Admissible neighbourhood, 237
2 and barycentric subdivision of 2+ , Admissible simplex, 198
197 Ahlfors, Lars, 145, 149
Sn , one-point compactification of Rn : Alaoglu Theorem, 102
stereographic projection, 157 Alexander subbase lemma, 101
Cylinder: quotient of a square, 86 Alexandroff, Pavel, 22, 68, 75, 103, 107, 112,
Torus: quotient of a square, 87 116, 124, 151, 155, 160
Balls in R3 , 11, 12 Algebraic topology, 208
Barycentric subdivision of the octahe- Almost alternating simplex, 198
dron, 197 Alternating simplex, 198
Basic open sets, 28 Antipodal labelling, 198
Comb space, 135, 136 Antipodal map, 196
Cone (X ), 87 Antipodal point, 90
Continuous, 114 Antipode preserving map, 184
Convex sets with interior are homeomor- Antipodes/Antipodal points, 3, 184
phic to the ball, 183 Arithmetic progression topology, 23
Covering map, 237 Arzelà, Cesare, 166, 177
Deformation retracts of an annulus and Ascoli, Giulio, 166, 177
of a disc with two holes, 209 Ascoli-Arzelà theorem, 166
Dosa/Pancake theorem, 4 Associative law for products, 90
Equivalent metrics, 16 Attaching space, 91
Fixed point theorem on [a, b], 2 Axiom of Choice, 99
French railways metric, 8 and Cohen, 101
Lifting, 238 and Gödel, 101
Lifting construction, 239 equivalents, 101
Möbius band, 90 from Tychonoff’s theorem, 99
independence of, 101
Neighbours of a simplex, 186
Open balls are open, 14
Path homotopy, 215 B
Path homotopy respects products, 216 Baire category theorem, 153
Product of homotopy classes— for complete metric spaces, 169
associativity, identity, inverse, 218 for locally compact spaces, 153
Product of two connected spaces, 133 Baire’s theorem, 153
Punctured plane: deformation retract of for complete metric spaces, 168
R3 \ {z, 210 for locally compact spaces, 153
Space-filling curve, construction, 176 Baire, René-Louis, 153, 154, 163, 167, 175,
Sperner colouring, 188 178
Suspension S (X ), 90 Banach contraction mapping theorem, 169,
Topologist’s sine curve, 136 176
Trapezoidal function, 158 Banach, Stefan, 125, 140, 163, 167, 169,
Uniqueness of completion, 172 176, 178
Figures Uniqueness of one-point compact- Barycentre, 196, 197
ification, 156 Barycentric subdivision, 197
int E, the interior of E, 24 Base, Basis, 26
Čech, Eduard, 103 Basic Lifting Theorem, 229, 239
Basic open set, 26
Betti, Enrico, 178
A Bing-Nagata-Smirnov metrisation theorem,
Abel, Neil Henrik, 54 147
Index 261

Birkhoff, Garrett, 53 Cantor’s Nested Interval Theorem, 60, 165

Bochner, Salomon, 101 Cantor, Georg, 21, 54, 74, 100, 170
and Zorn’s lemma, 101 Carathéodory, Constantin, 119
Bohl’s Theorem, 191, 214 Cartan, Henri, 53, 103
Bohl, Piers, 182, 191, 202 Cartesian product, 82
Bolzano’s Intermediate Value Theorem, 1 Category, 153, 169, 226
Bolzano, Bernhardus, 1, 2, 6, 7, 41, 42, 69, Cauchy sequence, 64, 67, 163, 170, 173
127, 175 Cauchy, Augustin-Louis, 41, 42, 54
Bolzano-Weierstrass compact, 63, 65, 73, and Abel, 54
120 and Galois, 54
Bolzano-Weierstrass property, 63, 166 Cauchy-Schwarz inequality, 9
Bolzano-Weierstrass theorem, 6, 65, 73 Chain, 100
Borel’s Theorem, 60 Characteristic function, 51
statement becomes a definition, 60 Chernoff, Paul, 98
Borel, Émil, 60, 61, 68, 73, 74, 205 Chevalley, Claude, 103
on compactness of intervals, 60 Choice function, 100
Borel, Armand, 103 Closed ball, 11
on Bourbaki meetings, 103 Closed map, 83
Borel-Cousin-Lebesgue Theorem, 61 Closed sets, 28
Borel-Lebesgue Theorem, 61 Closure of a subset, 29
Borsuk No Retract Theorem, 191 Cluster point, 70
dimension 2, 235 of a filter, 73
Borsuk’s theorems, 200 of a net, 70, 98
Borsuk, Karol, 181, 191, 196, 200, 204, 208, of a squence, 70
221, 235 partial, 98
Borsuk-Ulam Theorem, 196, 200, 204 Vietoris’ definition, 70
dim 1, 3 Co-ordinatewise convergence, 13
dim 2, 231 Coarser topology, 24
Boundary of a set, 45 Cocountable topology, 23
Boundary of a simplex, 185 Cofinal subset, 70
Bounded metric associated to a metric, 9 Cofinite topology, 23
Bourbaki’s Mittag-Leffler Theorem, 167 Cohen, Paul, 75, 101
Bourbaki, Nicolas, 28, 79, 83, 101, 102, 160, and Axiom of Choice, 101
167, 257 and Continuum Hypothesis, 75
birth of, 103 Colouring, 187, 198
Box topology, 90 Comb space, 136, 213
Brouwer fixed point theorem, 181, 182, 190, figure, 136
202, 204, 205 Combinatorial n-cube, 185
Brouwer’s statement, 189 Combinatorial n-simplex, 185
dimension 1, 2 Combinatorial topology, 208
dimension 2, 236 Compact spaces, 60
equivalents, 201 and filters, 73
Brouwer, Luitzen Egbertus Jan, 124, 189, and finite intersection property, 68
190, 192, 193, 196, 202, 204, 207, and nets, 71
211 F.Riesz’s formulation, 68
and topology, 205 Fréchet definition, 62
open cover definition, 68
Vietoris definition, 59
C Compact topological space, 68
Cantor ternary set, 45, 62, 122, 135, 160, Compact-open topology, 174
176, 210 Compactification, 119
Cantor’s diagonal argument, 65, 166 one-point, 155
Cantor’s intersection theorem, 165, 169 Stone-Čech, 119
262 Index

Complementary vertices, 198 Deformation retraction, 210

Complete metric space, 64, 163 Degree, 231
Fréchet’s definition, 163 Degree of a vertex, 198
Completely regular space, 117 Delsarte, Jean, 103
Completion of a metric space, 170 Denjoy, Arnaud, 178
uniqueness, 172 Dense subset, 16, 32
Cone(X ), 88 de Rham, George, 74
Connected component, 135 Derived set, 42, 49
Connected sets in R, 132 Dictionary order, 34
Connected space, 131 Dieudonné, Jean, 103, 160
F.Riesz’s definition, 127 Differential topology, 208
Lennes’ definition, 127 Dimension invariance
locally, 137 Brouwer’s statements, 192
not path connected, 136 Dimension Invariance Theorem, 195
path, 128 Dini, Ulisse, 178
Constant path, 218 Directed set, 43
Continuity, 46 Dirichlet, Lejeune, 56
Bolzano’s defintion, 41 and uniform continuity, 65
Cauchy’s defintion, 41 Disconnection/separation, 131
in terms of convergence, 13, 46, 47 Discrete metric, 8
Weierstrass’ defintion, 41 Discrete topology, 23
Continuous deformation, 208 Disjoint union, 91
Continuous function, 8 Domain Invariance Theorem, 195
compactly supported, 172 Dosa theorem/Pancake theorem, 4
global formulation, 14 Dyadic rational, 32
local formulation, 10
nowhere differentiable, 175
vanishing at infinity , 172 E
Continuum Hypothesis, 75 Edge, 198
Contractible space, 213, 243 Egoroff, Dmitri, 160
Contraction mapping, 169 Eilenberg, Samuel, 34, 221
Contrapositive, 13 Eisenstein, Gotthold, 56
Convex combination, 182 Embedding, 89
Convex hull, 182 Embedding lemma, 118
Convex set, 182 Engel, Friedrich, 55
Convex sets and balls, 182 Equator, 184
Countable compactness, 72 Equicontinuity, 175
Cousin, Pierre, 61 Equicontinuous family, 166
and compactness, 61 Equivalent metrics, 16
Covariant functor, 226 Esterle, Jean, 168
Covering map, 237 Euclidean metric, 8
Covering spaces, 237 Euclidean norm, 10, 184
equivalence of, 242 Euclidean topology, 23
Covering transformation, 242 Evaluation map, 89, 118, 174
Cube, 184 Evenly covered, 237
Cubical Sperner’s Lemma, 186, 187 Eventually, 13, 43, 44
Extension from [a, b] to R, 114

D
Darboux’s theorem, 5 F
Deck transformation, 242 Face, 185
Dedekind, Richard, 6, 56, 75 Facet, 198
Deformation retract, 210 Fermat, Pierre, 56
Index 263

Filter, 53 Geometric n-simplex, 185

convergence, 53 Gödel, Kurt, 75, 101, 140
Fréchet, 53 and Axiom of Choice, 101
principal, 53 and Continuum Hypothesis, 75
Filter base, 53 Graph, 198
Filters vs nets, 53 Grothendieck, Alexander, 161
Final topology, 81 Group actions on a space, 89
Finer topology, 24 Group of rigid motions of Rn , 51
Finite intersection property, 96
First category, 169
First countable space, 143 H
Hausdorff’s definition, 143 Hadamard, Jacques, 19, 182, 202, 203, 205
Fixed point, 2, 167, 201 Hadamard’s theorem, 202
Fixed point property, 191, 221 Hahn, Hans, 137, 140, 204
Fixed point theorem, 2, 181, 189, 201 Hairy Ball Theorem, 257
Banach, 167, 169, 176, 178 Hausdorff, Felix, 7, 21, 22, 38, 65, 108, 124,
Brouwer, 2, 181, 182, 189, 190, 204, 205, 127, 131, 135, 143, 144, 170, 208
236 and general topology, 21
Kakutani, 181 definition of a topological space, 22
Flag of hemispheres, 196, 197 defintion of a topological space, 22
Fourier, Joseph, 56 separation axiom, 22, 108
Fréchet, Maurice, 7, 19, 21, 59, 62, 65, 107, Hausdorff Maximal Principle, 100
143, 145, 178, 203, 208 Hausdorff metric, 33
defintion of a metric, 7 Hausdorff space, 108
defintion of separable space, 143 Heine, Eduard, 56, 61, 74
limit point compactness for metric and Hein-Borel Theorem, 61
spaces, 63 Heine-Borel-Lebesgue theorem, 61
on compact spaces, 59 Hemisphere, 192
Free group on two generators, 235 Hensel, Kurt , 56
Hewitt, Edwin, 117
French railways metric, 8, 17
Higher connectedness theorem, 189
Frobenius, Ferdinand Georg, 54
Higher homotopy groups, 225
Fully coloured simplex, 187
Hilbert-Bernays-von Neumann set theory,
Functorial, 226
100
Fundamental group, 224, 225
Hilbert, David, 37, 75, 93
homotopy invariance, 228
Hólder, Otto, 56
nonabelian, 235
Homeomorphism, 48
of a topological group, 244
Homotopy, 48, 208
of the circle, 231
Brouwer’s definition, 207
of the cylinder, 234
equivalence, 212
of the punctured disc, 234
invariance, 227, 228
of the punctured plane, 234
lifting theorem, 230, 240
of the torus, 235 of maps, 211
Poincaré’s definition, 224 of paths, 215
topological invariance, 227 type, 212
Fundamental theorem of algebra, 236 Hopf, Heinz, 160, 221
Furstenberg, Hillel (Harry), 35 Hurewicz, Witold, 140
Hurwitz, Adolf, 56

G
Galois, Evariste, 54, 140 I
Gauß/Gauss, Karl Friedrich, 22, 37, 56 Identification topology, 85
Generalised sequence, 43 Idli theorem/ham sandwich theorem, 201
General topology, 21 Inadequacy of sequences, 31, 43
264 Index

Indiscrete topology, 23 Lie, Sophus, 56

Induced homomorphism, 226 Lift, 230
Induced metric, 8 Limit point/accumulation point, 42
Infinitude of primes, Furstenberg’s topolog- Limit point compact, 63
ical proof, 35 Lindelöf, Ernst, 144, 145, 148
Initial topology, 28, 80 Lindelöf space, 145, 148
Integer lattice, 202 Lindelöf’s theorem, 144
Interior of a subset, 24 Listing, J. B., 22
Interior point, 24 Listing, J.B., 92
Intermediate value theorem, 1, 132 Local base., 26
Interval, 1 Locally closed set, 30, 152
Intervals in R, 4 Locally compact space, 151
Inverse limit, 168 Alexandroff’s definition, 151
Inverse system, 168 Locally connected space, 137
Isolated point, 42 Locally finite family, 29, 50
Isometry, 17 Locally finite refinement, 72
Loop, 224
Loop homotopy class, 224
J Lower hemi-sphere, 184
Janiszewski, Zygmunt, 74 Lower limit topology, 27
Jordan, Camille, 140 Lusternik-Schnirelmann-Borsuk Theorem,
Jordan curve theorem, 181 196, 200
Luzin, Nikolai, 160

K
Kelley, John, 44, 70, 83, 99, 104 M
Killing, Wilhelm, 56 Matheron, Étienne, 96
KKM map, 202 Matrix space, 18
Klein bottle, 90, 93 Maximal element, 100
Klein, Felix, 56, 90, 93 Mazurkiewicz, Stefan, 141, 182, 204, 208,
Knaster, Bronisław, 135, 141, 182, 204 221
Knaster-Kuratowski-Mazurkiewicz theo- Mazur, Stanislaw, 178
rem/KKM theorem, 202 Menger, Karl, 140
Kolmogorov, Andrei, 160 Method of bisection, 60
Kovalevskaya, Sofia, 55 Metric, 8
Kronecker, Leopold, 56, 74 on a product, 9, 89, 173
Kulpa, Władysław, 193 Metric from a norm, 18
Kuratowski closure operation, 36 Metric space, 8
Kuratowski, Kazimierz/Casimir, 36, 96, 135, compactness, 64
140, 141, 176, 182, 204, 221 Metric topology, 23
and Zorn’s lemma, 101 Metrisable space, 23, 146
characterisation of compactness, 95 Milnor, John, 208, 257
Ky Fan Lemma, 198 Minimal element, 100
Minkowski, Hermann, 56
Mittag-Leffler, Gösta, 56, 167
L Mittag-Leffler’s Theorem, 168
Labelling, 187, 198 Möbius, August Ferdinand, 90, 92
Lattice, 24 Möbius band, 90, 92
Lebesgue covering lemma, 64 Montel, Paul, 74, 161
Lebesgue, Henri, 61, 64, 67, 74, 196, 204 Moore, Robert Lee, 92
Lebesgue number, 64 Moore-Smith convergence, 44
Lebesgue number lemma, 64 Motion group of Rn , 51
Lennes, N.J., 127, 131 Müger, Michael, 193
definition of connectedness, 127 Multiplicative linear functional, 102
Index 265

N Pinching a subspace, 88
Nash, John, 182 Plücker, Julius, 93
Neighbourhood, 11, 25 Poincaré conjecture, 223, 246
Neighbourhood axioms, 22 Poincaré, Henri, 21, 208, 245
Neighbourhood base, 26 and fixed point theorems, 189
Net, 43 and Relativity, 245
Nets vs filters, 53 definition of fundamental group, 224
Nonabelian fundamental group, 235 Gaston Darboux on, 245
Non-archimedean metric, 8 Hairy Ball Theorem, 257
Noncompact balls, 73 higher dimensional intermediate value
Noncontractibility of the sphere, 214 theorem, 181
No Retract Theorem, 191, 201, 202, 214, 235 inspiration to Brouwer, 192
Norm, 10, 19 on geometry and fundamental group, 223
Normal space, 111 on his own work, 245
Normed algebra, 102 on Weierstrass, 56
Nowhere dense set, 153, 169 Poincaré-Miranda theorem, 190
Null homotopic, 213 Pointed topological space, 226
Points of continuity, 160
Polish space, 176
O Pontryagin, Lev, 160
Octahedron ♦n , 184 Positive eigenvalue, 192
Odd degree polynomials, 3 Positive octant of the sphere, 192
One-point compactification, 155 Precompact/Relatively compact, 152
Open ball, 11 Product topology, 28, 83
Open base, 26 Čech’s definition, 79
Open map, 83 Tychonoff’s defintion, 79
Open sets, 22, 23 Projection, 83
Orbit, 89 Projective limit, 168
Orbit space, 89 Projective line, 90
Projective space, 90
Order topology, 34
Projective system, 168
Oriented affine map, 216
Proper map, 96, 101, 159
Orthogonal group O(n), 89
Pseudocompact space, 72
Orthogonal matrix, 89
Pugh, Charles, 98

P Q
Paracompact space, 122 Quotient map, 84, 86
Partial cluster point, 98 Quotient topology, 84
Partial map, 97
Partial net, 98
Partial order, 97, 100 R
Partial product, 97 Raman-Sundström, Manya, 61
Partition of unity, 123, 158 Regular closed set, 36
Pasting lemma, 47 Regular open set, 36
Path, 128 Regular space, 116
Path connected component, 129 Relative topology, 23
Path connected space, 128 Retract, 86, 191, 208
Path homotopy, 215 Retraction, 86, 208
Path Lifting Theorem, 230, 240 Reverse path, 128, 219
Peano curve, 123 Riemann, Bernhard, 21, 37, 56
Perfectly normal space, 122 and topology, 21
Pfaff, Johann, 92 creator of topology, 21
Picard, Émile, 203 Riesz, Frigyes, 21, 38, 68, 107, 127, 131
266 Index

and connectedness, 127 discovery of Banach, 178

and development of general topology, 38 Stereographic projection, 157, 234
Riesz, Marcel, 38 Stone, A.H., 122
Right continuous, 49 Stone, Marshall, 103, 104, 119, 125, 178
Rigid motion, 18 Straight line homotopy, 212
Ross, Kenneth, 98 Stronger topology, 24
Rotation, 191 Subbase, 28
Runde, Viktor, 168 Subcell, 185
Runge, Carl, 56 Subcube, 185
Subdivision, 196, 197
Subgroups of R, 52
S Subnet, 70
Saks, Stanisław, 204 Support hemisphere, 197
Schönflies, Arthur, 55 Support of a function, 158
Schottky, Friedrich, 55 Suspension, 90
Schwarz, Hermann, 55 Symmetric neighbourhood, 52
Second category, 169 Symmetric triangulation, 196, 197
Second countable space, 144
Hausdorff’s definition, 143
Seifert-van Kampen Theorem, 233 T
Semi-continuity, 51 Tao, Terence, 194
Semi-continuous, 73 Tarski, Alfred, 141
Semilocally simply connected, 243 Tietze Extension Theorem, 114, 121
Semi-metric/pseudo-metric, 18 Tietze, Heinrich, 22, 68, 107, 111, 116, 123,
Separable space, 145 137, 151, 237
Fréchet’s definition, 143 Tonelli, Leonid, 178
Separated sets, 131 Topological equivalence, 48
Separating closed sets, 193 Topological group, 52, 92
Separation axioms, 107 equivalent uniform structures, 52
Sequences are inadequate, 31, 43 uniform structures, 52
Sequential compactness, 63 Topological invariance, 227
Sequential convergence, 12, 13, 31 Topological space, 22
Sequentially closed subset, 31 Hausdorff’s defintion, 22
Sequentially open subset, 31 Tietze-Alexandroff definition, 22
Shape theory, 221 Topologist’s sine curve, 136
Sheet, 237 figure, 136
Sierpiński space, 23 Topology, 22
Sierpiński, Wacław, 141, 204 compact-open, 174
σ -compact space, 72 generated by S , 24
Simply connected, 226 induced by a metric/semi-metric, 23
Sn , n > 1, is, 229, 234 of local uniform convergence, 174
Slice, 237 of point-wise convergence, 174
Smooth partitions of unity, 159 of uniform convergence, 174
Smooth Urysohn Lemma, 159 Totally bounded subset, 63
Space filling curve, 122 Totally disconnected space, 135
Sperner colouring, 187 Total order, 100
Sperner, Emmanuel, 141, 182, 186–188, Triangle inequality, 10
190, 193, 204 Triangulation, 196, 197
Sphere, 184 Triangulation aligned with hemispheres, 197
is not contractible, 214 Tube Lemma, 101
is simply connected, 229, 234 Tychonoff space, 117
Star-shaped set, 129 Tychonoff’s Theorem, 97
Steinhaus, Hugo, 141, 178 via Alexander subbase lemma, 101
Index 267

via f.i.p., 101, 102 V

via nets, 98 Veblen, Oswald, 92
via open covers, 97 Vector field, 243, 257
as sated by Čech, 95 Vertices, 198
as stated by Tychonoff, 95 Vietoris, Leopold, 44, 53, 70, 107, 112, 116,
implies Axiom of Choice, 99 147
Tychonoff, Andrey, 79, 103, 117–119, 146, and compactness, 59
147, 160 father of modern convergence, 44

U
Ulam, Stanisław, 203 W
Ultrafilter, 73 Weak* topology, 102
Ultranet/Universal net, 73 Weaker topology, 24
Uniform continuity, 65, 66
Weierstraß/Weierstrass, Karl, 6, 7, 41, 42, 55
and Dirichlet, 65
Poincaré on, 56
and Heine, 65
Weil, André, 36, 102
Uniform continuity theorem, 67
Well-ordered set, 100
Uniformity, 36
Base for a, 37 Well-ordering, 100
Topology given by a, 36 Well-ordering Principle, 100
Uniformly equivalent metrics, 174 Weyl, Hermann, 21, 37
Uniform space, 36
Uniqueness of lifts, 238
Universal cover, 243
Upper bound, 100 Z
Upper half-plane, 91 Zariski topology, 33
Upper hemi-sphere, 184 Zermelo-Fräenkel set theory, 100, 101
Urysohn metrisation theorem, 146 Zorn’s lemma
Urysohn, Pavel, 68, 103, 112, 113, 117, 124, and Bochner, 101
146, 147, 160 and Kuratowski, 101
Urysohn’s Lemma, 113 Zorn, Max, 101, 141
for locally compact spaces, 158 Zorn’s lemma, 97–101, 141
Usual topology, 23 Zygmund, Antoni, 204