Elementary Fixed Point Theorems. (PDFDrive)

Forum for Interdisciplinary Mathematics
P. V. Subrahmanyam
Elementary
Fixed Point
Theorems
Editor-in-chief
P. V. Subrahmanyam, Indian Institute of Technology Madras, Chennai, India
Editorial Board
Yogendra Prasad Chaubey, Concordia University, Montreal (Québec), Canada
Jorge Cuellar, Principal Researcher, Siemens, Germany
Janusz Matkowski, University of Zielona Góra, Poland
Thiruvenkatachari Parthasarathy, Chennai Mathematical Institute, Kelambakkam, India
Mathieu Dutour Sikirić, Rudjer Boúsković Institute, Zagreb, Croatia
Bhu Dev Sharma, Jaypee Institute of Information Technology, Noida, India
The Forum for Interdisciplinary Mathematics series publishes high-quality
monographs and lecture notes in mathematics and interdisciplinary areas where
mathematics has a fundamental role, such as statistics, operations research,
computer science, financial mathematics, industrial mathematics, and
bio-mathematics. It reflects the increasing demand of researchers working at the
interface between mathematics and other scientific disciplines.
More information about this series at http://www.springer.com/series/13386

P. V. Subrahmanyam
Elementary Fixed Point

Theorems
123
P. V. Subrahmanyam (emeritus)
Department of Mathematics
Indian Institute of Technology Madras
Chennai, Tamil Nadu, India
ISSN 2364-6748 ISSN 2364-6756 (electronic)

ISBN 978-981-13-3157-2 ISBN 978-981-13-3158-9 (eBook)
https://doi.org/10.1007/978-981-13-3158-9
Library of Congress Control Number: 2018960278
© Springer Nature Singapore Pte Ltd. 2018

This work is subject to copyright. All rights are reserved 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, express 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
Preface
Fixed point theorems constitute an important and interesting aspect of applicable

mathematics and provide solutions to several linear and nonlinear problems arising
in biological, engineering and physical sciences. This book, as the title suggests,
deals with some fixed point theorems and applications. The choice of the topics is
largely guided by personal preferences, and the book will serve as a subjective
sampler of topics in fixed point theorems.
This volume neither exhausts all the important fixed point theorems nor elabo-
rates the most general formulations of the theorems presented. No attempt is made to
provide a historical/chronological background for the topics covered. Nevertheless,
it hopefully supplements the extant publications and promotes interest for further
study among the readers.
A quick run-through of the highlights of the individual chapters is not out of
place. While Chap. 1 collects the analytic and topological preliminaries needed in
the sequel, Chap. 2 describes the basic properties of iterates of real and complex
functions. It details the theorems of Thron on the rates of convergence of certain
classes of real functions. Cohen’s common fixed point theorem for commuting
continuous full surjections on a compact interval, Shield’s theorem on the existence
of a common fixed point for a commuting family of analytic functions on the closed
unit disc, an elementary proof of Sharkovsky’s theorem on periodic points of real
functions and Bergweiler’s theorem on the existence of fixed points of meromor-
phic functions are other noteworthy features of Chap. 2. Chapter 3 explores the
existence of fixed points in the setting of partially ordered sets. Knaster–Tarski
principle is formulated leading to Tarski’s theorem, and its applications to set
theory, general topology, nonlinear complementarity problem, parabolic differential
equations and formal languages are described. Also discussed are some general-
izations due to Merrifield and Stein.
Most of Chap. 4 deals with Ward’s theory of partially ordered topological spaces
culminating in Schweigert’s fixed point theorem. Manka’s fixed point theorem on
inductively and acyclically ordered posets is proved and used to deduce the fixed
v
vi Preface
point property of continuous functions on dendroids. Also highlighted is Klee’s

counterexample in fixed point theory. Contraction principle and some of its variants
are taken up in Chap. 5, including Kupka’s topological generalization and Nadler’s
extension to multifunctions. Jachymski’s proof of the converse of the contraction
principle is also elaborated. Applications to differential equations, functional
equations, algebraic Weierstrass preparation theorem, Cauchy–Kowalevsky theo-
rem and the central limit theorem are detailed in Chap. 6. Chapter 7 is devoted to
Caristi’s fixed point theorem, a generalization of the contraction principle. Separate
proofs of Caristi’s theorem due to Siegel, Penot and Kirk are provided. The
equivalence of Caristi’s theorem, Ekeland’s variational principle and Takahashi’s
minimization theorem is proved. That these three equivalent theorems characterize
metrical completeness is also established.
Chapter 8 is on fixed points of contractive and non-expansive maps. Goebel’s
proof of Browder–Gohde–Kirk fixed point theorem for nonexpansive mappings in
the setting of uniformly convex Banach spaces is provided. Pasicki’s theory of bead
and discus spaces and his theorem on the fixed points of nonexpansive maps with
an application of Matkowski to certain functional equations are other highlights.
Ishikawa’s theorems on iterates are detailed as also a fixed point theorem due to
Merrifield et al. on generalized contractions using combinatorial ideas.
Chapter 9 using the concepts of the geometry of Banach spaces establishes that
convex weakly compact subsets of nearly uniformly smooth (NUS) and non-square
Banach spaces have fixed point property for nonexpansive mappings. Brouwer’s
theorem is treated in Chap. 10. Besides an analytic proof due to Rogers, proof based
on Sperner’s lemma is provided. The existence of Walrasian and Nash equilibria for
economies is deduced. Whyburn’s proof of Stallings's generalization of Brouwer’s
fixed point theorem for connectivity maps has also been elaborated. Schauder’s
theorem and its extensions and applications constitute the major part of Chap. 11.
Besides Tychonoff’s fixed point theorem and Ky Fan–Browder–Glicksberg theo-
rem for multifunctions, Kakutani–Markov and Ryll–Nardzewski theorems are
described. The existence of Banach limits, Haar integral on compact groups
and Nash equilibria is provided besides an introduction to measures of
noncompactness. Chapter 12 describes the finite-dimensional degree theory as
presented by Heinz. Appendices A and B summarize the classical counterexamples
due to Huneke and Kinoshita, respectively. Appendix C deals with fractals and
fixed points.
The text closely follows the notation and organization of the papers quoted
extensively, within reasonable limits, to aid the readers to peruse these papers with
ease. A judicious choice of sections from Chaps. 2–12 can constitute a basic course
on fixed point theorems.
No expression of gratitude is adequate for the award of a sabbatical leave granted
by the competent authority of the Indian Institute of Technology Madras to me for
the preparation of an earlier concise version of this tract. I am thankful to my
colleagues for their support during the preparation of this tract and to my wife
Preface vii
Dr. S. Alamelu Bai for proofreading the content and for the moral support. I am
thankful to Dr. N. Sivakumar (Texas A&M University) and Dr. Antony Vijesh
(IIT Indore) for getting several papers for reference. I am grateful to Profs. M. S.
Rangachari, G. Rangan, (Late) K. S. Padmanabhan and (Late) K. N. Venkataraman
of the Madras University, Chennai, for initiating me into different aspects of mathe-
matical analysis. Thanks are due to Mr. E. Boopal for typesetting the manuscript.
Suggestions for improvement and corrections of errors and misprints are earn-
estly solicited.
Chennai, India P. V. Subrahmanyam

Contents
1 Prerequisites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Topological Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Normed Linear Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Topological Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2 Fixed Points of Some Real and Complex Functions . . . . . . . ..... 23
2.1 Fixed Points of Continuous Maps on Compact
Intervals of R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2 Iterates of Real Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3 Periodic Points of Continuous Real Functions . . . . . . . . . . . . . 31
2.4 Common Fixed Points, Commutativity and Iterates . . . . . . . . . . 35
2.5 Common Fixed Points and Full Functions . . . . . . . . . . . . . . . . 41
2.6 Common Fixed Points of Commuting Analytic Functions . . . . . 44
2.7 Fixed Points of Meromorphic Functions . . . . . . . . . . . . . . . . . . 48
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3 Fixed Points and Order . . . . . . . . . . . . . . . . .................. 51
3.1 Fixed Points in Linear Continua . . . . . .................. 51
3.2 Knaster–Tarski Principle . . . . . . . . . . . .................. 54
3.3 Tarski’s Lattice Theoretical Fixed Point Theorem
and Related Theorems . . . . . . . . . . . . . .................. 57
3.4 Some Applications . . . . . . . . . . . . . . . .................. 64
References . . . . . . . . . . . . . . . . . . . . . . . . . . .................. 70
4 Partially Ordered Topological Spaces and Fixed Points . . . . . . . . . 73
4.1 A Precis of Partially Ordered Topological Spaces . . . . . . . . . . . 73
4.2 Schweigert–Wallace Fixed Point Theorem . . . . . . . . . . . . . . . . 81
4.3 Set Theory, Fixed Point Theory and Order . . . . . . . . . . . . . . . . 86
4.4 Multifunctions and Dendroids . . . . . . . . . . . . . . . . . . . . . . . . . 90
ix
x Contents
4.5 Some Spaces with Fixed Point Property . . . . . . . . . . . . . . . . . . 92

4.6 An Example in Fixed Point Theory . . . . . . . . . . . . . . . . . . . . . 94
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5 Contraction Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.1 A Simple Proof of the Contraction Principle . . . . . . . . . . . . . . . 97
5.2 Metrical Generalizations of the Contraction Principle . . . . . . . . 100
5.3 Fixed Points of Multivalued Contractions . . . . . . . . . . . . . . . . . 105
5.4 Contraction Principle in Gauge Spaces . . . . . . . . . . . . . . . . . . . 107
5.5 A Converse to the Contraction Principle . . . . . . . . . . . . . . . . . 109
5.6 A Topological Contraction Principle . . . . . . . . . . . . . . . . . . . . 114
5.7 Another Proof of the Contraction Principle . . . . . . . . . . . . . . . . 116
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
6 Applications of the Contraction Principle . . . . . . . . . . . . . . . . . . . . 121
6.1 Linear Operator Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
6.2 Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6.3 A Functional Differential Equation . . . . . . . . . . . . . . . . . . . . . . 128
6.4 A Classical Solution for a Boundary Value Problem
for a Second Order Ordinary Differential Equation . . . . . . . . . . 129
6.5 An Elementary Proof of the Cauchy–Kowalevsky
Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
6.6 An Application to a Discrete Boundary Value Problem . . . . . . . 133
6.7 Applications to Functional Equations . . . . . . . . . . . . . . . . . . . . 135
6.8 An Application to Commutative Algebra . . . . . . . . . . . . . . . . . 140
6.9 A Proof of the Central Limit Theorem . . . . . . . . . . . . . . . . . . . 143
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
7 Caristi’s Fixed Point Theorem . . . . . . . . . ......... . . . . . . . . . . 151
7.1 Introduction . . . . . . . . . . . . . . . . . . . ......... . . . . . . . . . . 151
7.2 Siegel’s Proof of Caristi’s Fixed Point Theorem . . . . . . . . . . . . 151
7.3 Ekeland’s Variational Principle . . . . . ......... . . . . . . . . . . 156
7.4 A Minimization Theorem . . . . . . . . . ......... . . . . . . . . . . 159
7.5 An Application of Ekeland’s Principle ......... . . . . . . . . . . 161
References . . . . . . . . . . . . . . . . . . . . . . . . . ......... . . . . . . . . . . 162
8 Contractive and Non-expansive Mappings . . . . . . . . . . . . . . . . . . . 165
8.1 Contractive Mappings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
8.2 Non-expansive Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
8.3 Browder–Gohde–Kirk Fixed Point Theorem . . . . . . . . . . . . . . . 173
8.4 A Generalization to Metric Spaces . . . . . . . . . . . . . . . . . . . . . . 176
8.5 An Application to a Functional Equation . . . . . . . . . . . . . . . . . 181
8.6 Convergence of Iterates in Normed Spaces . . . . . . . . . . . . . . . . 184
Contents xi
8.7 Iterations of a Non-expansive Mapping . . . . . . . . . . . . . . . . . . 186

8.8 A Generalization of the Contraction Principle Based
on Combinatorics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
9 Geometric Aspects of Banach Spaces and Non-expansive
Mappings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
9.2 Coefficients of Banach Spaces and Fixed Points . . . . . . . . . . . . 200
9.3 Nearly Uniformly Smooth Spaces . . . . . . . . . . . . . . . . . . . . . . 204
9.4 Non-square Banach Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
9.5 An Equivalent Definition of RWða; XÞ and Fixed Points of
Non-expansive Maps in Non-square Banach Spaces . . . . . . . . . 213
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
10 Brouwer’s Fixed-Point Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
10.2 Analytic Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
10.3 Brouwer’s Fixed-Point Theorem . . . . . . . . . . . . . . . . . . . . . . . 220
10.4 A Proof of Brouwer’s Theorem from Sperner’s Lemma . . . . . . 223
10.5 Scarf’s Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
10.6 More on Brouwer’s Fixed-Point Theorem . . . . . . . . . . . . . . . . 229
10.7 A Proof of the Fundamental Theorem of Algebra . . . . . . . . . . . 235
10.8 A Generalization of Brouwer’s Theorem . . . . . . . . . . . . . . . . . 237
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
11 Schauder’s Fixed Point Theorem and Allied Theorems . . . . . . . . . 245
11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
11.2 An Application of Schauder’s Theorem to an Iterative
Functional Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
11.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
11.2.2 A Subset of CðI; RÞ . . . . . . . . . . . . . . . . . . . . . . . . . . 249
11.3 Measures of Noncompactness and Fixed Point Theorems . . . . . 253
11.4 Kakutani-Ky Fan–Glicksberg Fixed Point Theorem . . . . . . . . . . 258
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
12 Basic Analytic Degree Theory of a Mapping . . . . . . . . . . . . . . . . . . 277
12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
12.2 Heinz’s Elementary Analytic Theory of Mapping
Degree in Finite Dimensional Spaces . . . . . . . . . . . . . . . . . . . . 277
12.3 Properties of the Degree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
12.4 Some Consequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
xii Contents
Appendix A: A Counterexample on Common Fixed Points . . . . . . . . . . . 289

Appendix B: A Compact Contractible Space Without Fixed Point
Property . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
Appendix C: Fractals via Fixed Points . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
About the Author
P. V. Subrahmanyam is Professor Emeritus at the Indian Institute of Technology

Madras (IIT Madras), India. He received his Ph.D. in mathematics from IIT Madras
for his dissertation on ‘Topics in Fixed-Point Theory’ under the supervision of
(late) Dr. V. Subba Rao. He received his M.Sc. in mathematics from IIT Madras
and B.Sc. in mathematics from the University of Madras, India. He has held several
important administrative positions, such as Senior Professor and Head of the
Department of Mathematics at IIT Madras; Founder and Head of the Department of
Mathematics at the Indian Institute of Technology Hyderabad (IIT Hyderabad);
Executive Chairman of the Association of Mathematics Teachers of India (AMTI)
and President of the Forum for Interdisciplinary Mathematics (FIM). Before joining
IIT Madras, he served as Faculty Member at Loyola College, University of Madras,
and Hyderabad Central University. His areas of interest include classical analysis,
nonlinear analysis and fixed point theory, fuzzy set theory, functional equations and
mathematics education. He has published over 70 papers and served on the editorial
board of several journals including Differential Equations and Dynamical Systems
and The Journal of Analysis. He received an award for his outstanding contributions
to mathematical sciences in 2004 and the Lifetime Achievement Award from FIM
in 2016. He has given various invited talks at international conferences and com-
pleted brief visiting assignments in many countries such as Canada, the Czech
Republic, Germany, Greece, Japan, Slovak Republic and the USA. He is also Life
Member of the Association of Mathematics Teachers of India, FIM, Indian
Mathematical Society and the Society for Industrial and Applied Mathematics
(SIAM).
xiii
Chapter 1
Prerequisites
This chapter is a precis of the basic definitions and theorems used in the sequel. It is
presumed that the reader is familiar with naive set theory (see Halmos [4]) and the
properties of real numbers and real functions (see Bartle [1]). Other mathematical
concepts and theorems relevant to specific sections of a chapter will be recalled
therein.
1.1 Topological Spaces
This section collects important concepts and results from topology. For proofs and
other details, Dugundji [3], Kelley [7], Munkres [9] and Simmons [13] may be
consulted.
Definition 1.1.1 Let X be a non-empty set. A collection of J of subsets of X is
called a topology on X , if
(i) φ, X ∈ T ,
G 1 ∩ G 2 ∈ T for G 1 , G 2 ∈ T and
(ii)
(iii) G∈F G ∈ T for any F ⊆ T .
Any subset of X belonging to T is called an open set or more precisely T -open

set. The pair (X, T ) is called a topological space. Given a topological space (X, T ),
the interior of A ⊆ X , denoted by A0 is the largest open subset of A.
For a subset S of X , where (X, T ) is a topological space, T S = {G ∩ S : G ∈ T }
is a topology on S, called the relative topology (or subspace topology) on S.
Example 1.1.2 For a non-empty set X , the family {φ, X } is a topology on X called
the indiscrete topology on X , 2 X , the power set of X or the set of all subsets of X is
a topology on X called the discrete topology on X . The family of all subsets of X
© Springer Nature Singapore Pte Ltd. 2018 1
P. V. Subrahmanyam, Elementary Fixed Point Theorems,
Forum for Interdisciplinary Mathematics,
https://doi.org/10.1007/978-981-13-3158-9_1
2 1 Prerequisites
whose complements are finite sets together with the empty set is also a topology on
X called the co-finite topology on X .
Since the intersection of any collection of topologies on X is a topology on X ,
for any family F of subsets of X , there is the smallest topology on X containing F,
called the topology generated by F.
Example 1.1.3 A subset G of real numbers is called open if for each x ∈ G, an open
interval containing x lies in G. (Evidently the empty set is open.) This collection of
all open subsets of R, the real number system is a topology on R, called the usual
topology on R.
Definition 1.1.4 Let (X, T ) be a topological space. A neighbourhood of a point
x ∈ X is any subset of X containing an open subset G ∈ T , containing x. A neigh-
bourhood base or local base at x is a family Nx of neighbourhoods of x such that for
any neighbourhood N of x, there is a neighbourhood N x ∈ Nx such that x ∈ N x ⊆ N .
A topological space is called first countable if for each point there is a countable local
base. An interior point of A is a point a ∈ A such that A contains a neighbourhood
of a.
Definition 1.1.5 A subset F of a topological space (X, T ) is called a closed subset
of X if X − F is T -open. The closure of a subset A of X denoted by A is the
smallest closed set containing A. A subset S of X is said to be dense in X if S = X .
A topological space (X, T ) is called separable if it has a countable dense subset.
Remark 1.1.6 Let (X, T ) be a topological space and A, B ⊆ X . Then
(i) φ0 = φ, φ = φ, X 0 = X and X = X ;
(ii) A ⊇ A and A0 ⊆ A;
(iii) A ∪ B = A ∪ B, (A ∩ B)◦ = A◦ ∩ B ◦ ;
(iv) (A) = A and (A◦ )◦ = A◦ . Further A = {x ∈ X : every neighbourhood of x has
a non-void intersection with A}. A0 = {a ∈ A : a is an interior point of A}.
Definition 1.1.7 For a topological space (X, T ) B ⊆ T is called a base (or basis)
for T is for A1 , A2 ∈ B and x ∈ A1 ∩ A2 , there exists A3 ∈ B such that x ∈ A3 ⊆
A1 ∩ A2 . A subfamily S of T is called a subbase for T of B, if the family of
intersections of all finite subfamilies of S is a base for T . If the topology T has a
countable base, then the topological space is called second countable.
Remark 1.1.8 If S is a family of subsets of X with ∪{S : S ∈ S } = X , then S is
a subbase for a topology on X , for which B the family of subsets of X which are the
intersections of finite subfamilies of S is a base for this topology.
Remark 1.1.9 The family of all subintervals of the form [a, b), a < b, a, b ∈ R is
a base for a topology on R, called the lower limit topology on R. Similarly, the
family {(a, b] : a < b, a, b ∈ R} is a base for a topology on R called the upper limit
topology on R. The usual (standard) topology on R has the family of all open intervals
(a, b), a < b, a, b ∈ R as a base. R with the usual topology is separable and second
countable. However, R with the lower limit topology is separable and first countable
but is not second countable.
1.1 Topological Spaces 3
Definition 1.1.10 A binary relation ≤ on a non-empty subset X is called a quasi-

order if the following conditions are satisfied:
(i) x ≤ x for all x ∈ X (reflexivity);
(ii) if x ≤ y and y ≤ z, for x, y, z ∈ X , then x ≤ z (transitivity).
If, in addition a quasi-order ≤ satisfies
(iii) if x ≤ y and y ≤ x, then x = y (anti-symmetry),
then the quasi-order ≤ is called a partial order. Accordingly if ≤ is a quasi-order
on X , then (X, ≤) is called a quasi-ordered space. If ≤ is a partial order on X , then
(X, ≤) is called a partially ordered set or poset.
Definition 1.1.11 A partial order ≤ on a set X is called a linear order or total order
if for any pair of elements x, y ∈ X either x ≤ y or y ≤ x. A linearly ordered set is
also called a chain.
Definition 1.1.12 A partially ordered set (D, ≤) is called a directed set if for any
pair x, y ∈ X , there exists z ∈ D such that x ≤ z and y ≤ z.
Definition 1.1.13 A net in a topological space X is a pair (S, ≥) where S is a

function from a directed set (D, ≥) into X . A net (S, ≥) in a topological space is
said to converge to an element x ∈ X if for each open set G containing x, there is
an element m of D such that for n ≥ m, n ∈ D, S(n) ∈ G. Clearly, a sequence in
a topological space is a net directed by the set of natural numbers with the usual
ordering.
Proposition 1.1.14 A subset S of a topological space (X, T ) is closed if and only

if no net in S converges to an element of X − S. An element s ∈ S for S ⊆ X if and
only if there is a net in S converging to s.
Definition 1.1.15 Let (X i , Ti ), i = 1, 2 be topological spaces. A map f : X 1 → X 2

is said to be continuous if for each T2 -open subset G of X 2 , f −1 (G) is T1 -open in
X 1 . If f is one-one and onto X 2 and if both f and f −1 are continuous maps, then f
(as also f −1 ) is called a homeomorphism from X 1 onto X 2 (from X 2 onto X 1 ).
A map f : X 1 → X 2 is said to be continuous at x ∈ X 1 , if for each neighbourhood
N f (x) of f (x) in X 2 , there is a neighbourhood N x of x such that f (N x ) ⊆ N f (x) .
A map g : X 1 → X 2 is called open if it maps open subsets of X 1 onto open subsets
of X 2 .
The following theorem is well-known.

Theorem 1.1.16 Let (X i , Ti ), i = 1, 2 be topological spaces and f : X 1 → X 2 be
a map. The following statements are equivalent:
(i) f is continuous on X 1 ;
(ii) f is continuous at each point of X 1 ;
(iii) f −1 (F) is closed in (X 1 , T1 ) for each closed subset F of X 2 ;
(iv) if G ∈ S , a subbase for T2 , then f −1 (G) ∈ T1 ;
4 1 Prerequisites
(v) for each net (S, ≥) converging to x in X 1 , ( f (S), ≥) converges to f (x) in X 2 ;

(vi) for each subset A of X 1 , f (A) ⊆ f (A);
(vii) for each subset B of Y , f −1 (B) ⊆ f −1 (B).
Theorem 1.1.17 A topological space (X, T ) is said to be disconnected if X =
A ∪ B where A and B are non-empty disjoint proper open subsets of X . A pair of
sets A and B is said to be separated if A ∩ B = B ∩ A = φ, where A and B are non-
empty. A topological space is called connected if it is not disconnected (A connected
space is not the union of two non-void separated sets). A subset Y of X is called
connected if Y is connected in the subspace topology. A maximal connected subset
of X is called a component.
Definition 1.1.18 A topological space is called totally disconnected if the only con-
nected subsets are singletons.
Definition 1.1.19 A topological space is said to be locally connected if the family
of open connected subsets is a base for the topology.
Remark 1.1.20 A discrete topological space with more than one element is locally
connected, though totally disconnected. The set (0, 1) ∪ (2, 3) with the subspace
topology inherited from R with the usual topology is locally connected and discon-
nected though not totally disconnected.
Theorem 1.1.21 Let (X, T ) be a topological space. Then
(i) if A is a connected subset of X and A ⊆ B ⊆ A, then B is a connected subset;
(ii) the union of a family of connected subsets of X , no two of which are separated
is connected;
(iii) components of X are closed and any two components are either identical or
disjoint;
(iv) any component of an open subset of a locally connected space is open.
Definition 1.1.22 A family of open sets {G λ : λ ∈ } of a topological space (X, T )
is called an open cover for X , if X = G λ . If every open cover of X has a countable
λ∈
subcover, the topological space is said to be Lindelof. If each open cover of X has a
finite subcover, then the topological space is called compact.
Definition 1.1.23 A topological space is called locally compact, if each element has
a compact neighbourhood.
Definition 1.1.24 Let (X λ , Tλ ), λ ∈ , = φ be a family
of topological spaces.
The Cartesian product of all these sets X λ denoted by X = X λ is the set of all
λ∈
functions f : → X λ such that f (λ) ∈ X λ for each λ ∈ . The map Pλ : X →
λ∈
X λ such that Pλ ( f ) = f (λ) for each f ∈ X is called the projection of the set X into
the λth coordinate set X λ . The topology of X having {Pλ−1 (U ) : U ∈ Tλ , λ ∈ } as
a subbase is called the product topology on X and X with this topology is referred
as the product (topological) space.
1.1 Topological Spaces 5
Theorem 1.1.25 Let {(X λ , Tλ ) : λ ∈ , = φ} be a family of topological spaces

and X be the product space with the product topology T . Then
(i) Pλ , the projection of X into X λ is continuous for each λ ∈ ;
(ii) a map f : Y → X , where Y is a topological space is continuous if and only if
Pλ ◦ f : Y → X λ is continuous for each λ ∈ ;
(iii) a net S in X converges to an element s if and only if its projection in each
coordinate space converges to the projection of s.
(iv) X is connected if and only if each X λ is connected;
(v) (Tychonoff’s theorem) X is compact if and only if each X λ is compact.
Definition 1.1.26 A topological space X is said to be

(i) T1 if for each pair of distinct elements x and y, there exist neighbourhoods N x
of x not containing y and N y of y not containing x;
(ii) T2 (Hausdorff) if each pair of distinct elements has disjoint neighbourhoods;
(iii) regular, if for each x ∈ X and any closed subset F of X not containing x, there
exist disjoint open sets G 1 and G 2 with x ∈ G 1 and F ⊆ G 2 (X is called T3 if
it is T1 and regular);
(iv) normal, if for each pair of disjoint closed subsets Fi , i = 1, 2 of X , there exist
disjoint open sets G i , i = 1, 2 with Fi ⊆ G i , i = 1, 2 (X is called T4 if it is T1
and normal).
(Every T4 space is T3 and each T3 space is T2 , while a T2 space is necessarily T1 ).
Theorem 1.1.27 (Urysohn’s Lemma) If A and B are disjoint closed subsets of a

normal space X , then there is a continuous function f : X → [0, 1] such that f ≡ 0
on A and f ≡ 1 on B.
Theorem 1.1.28 Let X and Y be topological spaces and f : X → Y be a continuous

map. If X is compact, then f (X ) is a compact subset of Y . If X is connected, then
f (X ) is a connected subset of Y .
Corollary 1.1.29 If X is a compact topological space and f : X → R is a continu-

ous map, then f attains its maximum and minimum on X . If X is a connected space
and f : X → R is continuous, then f (X ) is an interval.
1.2 Metric Spaces
In this section, basic concepts and theorems from the theory of metric spaces are
recalled. For details, in addition to the references cited in Sect. 1.1, Kaplansky [6]
may be consulted.
Definition 1.2.1 Let X be a non-void set. A map d : X × X → [0, ∞) (=R+ ) is
called a metric if
6 1 Prerequisites
(i) d(x, x) = 0 for all x ∈ X ;

(ii) d(x, y) = 0 implies x = y;
(iii) d(x, y) = d(y, x), for all x, y ∈ X ;
(iv) d(x, y) ≤ d(x, z) + d(z, y) for all x, y, z (triangle inequality).
The pair (X, d) is called a metric space. A map d satisfying (i), (iii) and (iv) is called
a pseudometric and the corresponding (X, d) is called a pseudometric space.
Definition 1.2.2 If (X, d) is a metric space, the set B(x0 ; r ) = {x ∈ X : d(x0 , x) <
r } for r > 0 is called an open sphere of radius r centred at x0 , while the set {x ∈ X :
d(x0 , x) ≤ r } is referred as the closed sphere of radius r with centre x0 .
Remark 1.2.3 The family of all open spheres {B(x; r ) : x ∈ X, r > 0} is a base for
a topology on X called the metric topology on X induced by d.
Example 1.2.4 (i) If X is a non-empty set the map d : X × X → R+ defined by
d(x, y) = 1 for x = y and d(x, x) = 0 is a metric on X called the discrete
metric. The corresponding metric topology on X is the discrete topology.
(ii) On R, d(x, y) = |x − y|, the absolute value of x − y defines a metric called the
usual (or standard) metric on R and the topology induced is the usual topology
on R (with the base comprising all open intervals).
n 21

(iii) On Rn , the set of all n-tuples of real numbers, d(x, y) = |xi − yi |2 ,
i=1
where x = (x1 , x2 , . . . , xn ) and y = (y1 , y2 , . . . , yn ) defines a metric, called
the Euclidean metric on Rn .
(iv) C[a, b], the set of all continuous real-valued function on the closed interval
[a, b], where a < b, a, b ∈ R is a metric space under the metric
d( f, g) = sup{| f (x) − g(x)| : x ∈ [a, b]}
where f, g ∈ C[a, b]. This metric is called Tschebyshev or uniform metric.

(v) More generally C(X ), the set of all continuous real-valued functions on a com-
pact topological space becomes a metric space with the metric d defined by
d( f, g) = sup{| f (x) − g(x)| : x ∈ X } where f, g ∈ C(X ).
b
(vi) d( f, g) = a | f (t) − g(t)|dt also defines a metric on C[a, b] the set of all con-
tinuous real-valued functions on [a, b].
(vii) If (X, d) is a metric space and S ⊆ X , then the restriction of d to S × S is a
metric and this metric topology is precisely the topology of S relative to the
metric topology on X .
Theorem 1.2.5 A metric space is second countable if and only if it is separable.
Theorem 1.2.6 Every metric space is a Hausdorff normal space.
Definition 1.2.7 A sequence (xn ) in a metric space (X, d) is called Cauchy (fun-
damental) if d(xm , xn ) → 0 as m, n → ∞. A metric space is said to be complete if
every Cauchy sequence in X converges to an element of X .
1.2 Metric Spaces 7
Theorem 1.2.8 (Baire) No complete metric space can be written as a countable

union of closed sets having empty interior.
Definition 1.2.9 Let (X i , di ), i = 1, 2 be metric spaces. A map T : X 1 → X 2 is

called an isometry if d2 (T x1 , T x2 ) = d1 (x1 , x2 ) for all x1 , x2 ∈ X 1 .
Theorem 1.2.10 Each metric space (X, d) can be isometrically embedded in a

complete metric space (X , d) as a dense subset. Further such a complete metric
space X , called the completion of X is unique up to isometry.
Remark 1.2.11 In example 1.2.4, except the space described in (vi), the metric spaces
in examples (i)–(v) are complete.
Theorem 1.2.12 If (X, d) is a metric space, then d1 (x, y) = min{1, d(x, y)}, x, y ∈
X defines a metric on X and the topologies induced on X by these metrices are the
same.

Theorem 1.2.13 If (X n , dn ), x ∈ N is a sequence of metric spaces, then X = Xn
n∈N
is a metric space under the metric d defined by
∞
1 dn (xn , yn )
d(x, y) = ,
n=1
2 n 1 + dn (xn , yn )
where x = (xn ) and y = (yn ) are in X . Further, if each (X n , dn ) is complete, then

(X, d) is complete.
The following metrization theorem is classical.

Theorem 1.2.14 (Urysohn) A regular T1 second countable topological space is
metrizable (in the sense that there is a metric on this space whose metric topology
is the given topology).
A concept basic to the study of the metrization problem is defined below.
Definition 1.2.15 A family F of subsets of a topological space (X, T ) is called
(i) locally finite, if each point of the space has a neighbourhood that intersects only
finitely many sets in F;
(ii) discrete if each point of the space has a neighbourhood that intersects at most
one member of F;
(iii) σ-locally finite (σ-locally discrete) if it is the union of a countable collection
of locally finite (finite) subfamilies.
Theorem 1.2.16 (Metrization theorems) A topological space is metrizable if and

only if it is T1 and regular with
a σ-locally finite base (Nagata–Smirnov);
or
a σ-discrete base (Bing).
8 1 Prerequisites
Another important notion is that of paracompactness formulated below.

Definition 1.2.17 A topological space X is called paracompact if each open cover
U of X has an open locally finite refinement U ∗ (viz. U ∗ is locally finite and each
member of U ∗ is open and is a subset of some set in U).
Theorem 1.2.18 Every pseudometric space is paracompact and a paracompact T2

space is normal.
Definition 1.2.19 Let X be a topological space. A family { f λ : λ ∈ = φ} of con-

tinuous
functions mapping X into [0, 1] is called a partition of unity if for each
x ∈ X, f λ (x) = 1 and all but a finite number of f λ ’s vanish on some neighbour-
λ∈
hood of x. A partition of unity { f λ : λ ∈ = φ} in subordinate to a cover U if each
f λ vanishes outside some member of U.
Theorem 1.2.20 A regular T1 space is paracompact if and only if for each open
covering of X , there is a partition of unity subordinate to this covering. For every
compact T2 space, every open cover has a partition of unity subordinate to it.
Definition 1.2.21 A subset S of a metric space (X, d) is said to be totally bounded, if

for each > 0, there exists a finite subset {x1 , x2 , . . . , xn } (depending on ) such that
n
S⊆ B(xi ; ). A subset S is said to be bounded if S ⊆ B(x; r ) for some x ∈ X
i=1
and some r > 0.
Theorem 1.2.22 For a metric space (X, d), the following are equivalent:
(i) X is compact;
(ii) X is complete and totally bounded;
(iii) every sequence in X has a convergent subsequence;
(iv) X has the Bolzano–Weierstrass property, viz. for every infinite subset A of X
has a limit point x0 ∈ X , i.e. a point x0 such that every neighbourhood of x0
meets A.
Theorem 1.2.23 Let (X i , di ), i = 1, 2 be metric spaces and f : X 1 → X 2 be a

continuous map. If X 1 is compact, then f is uniformly continuous in the sense that
for each > 0 there exists δ > 0 depending only on so that d2 ( f (x), f (y)) <
whenever x, y ∈ X 1 and d1 (x, y) < δ.
Definition 1.2.24 For a non-void subset A of a metric space (X, d), the distance of
a point x from A is defined as D(x, A) = in f {d(x, a) : a ∈ A}.
Theorem 1.2.25 Let A be a non-void subset of a metric space (X, d). Then A =
{x ∈ X : D(x, A) = 0}. Further, |D(x, A) − D(y, A)| ≤ d(x, y) for any x, y ∈ X
and the map x → D(x, A) is a continuous map of X into R+ .
1.3 Normed Linear Spaces 9
1.3 Normed Linear Spaces
Normed linear spaces, constituting the base of Functional Analysis are metric spaces
with a richer (algebraic) structure. They provide a natural setting for mathematical
modelling of many natural phenomena. Bollabos [2], Kantorovitch and Akhilov [5],
Lyusternik and Sobolev [8], Rudin [11, 12], Simmons [13] and Taylor [14] may
be consulted for a detailed exposition of the following concepts and theorems. It is
assumed that the reader is familiar with the concepts of groups, rings and fields.
Definition 1.3.1 A linear space or vector space over a field F is a triple (V, +, ·),
where + is a binary operation (called vector addition or simply addition) and · is a
mapping from F × V into V (called scalar multiplication) satisfying the following
conditions;
(i) (V, +) is a commutative group with θ (called zero vector) as its identity element;
(ii) for all λ ∈ F, x, y ∈ V λ.(x + y) = λx + λy;
(iii) for all λ, μ ∈ F and x ∈ V (λ + μ).x = λ.x + μ.x and λ · (μ · x) = (λ · μ) · x
(where λμ is the product of λ and μ under the multiplication in the field F);
(iv) 0 · x = θ, 1 · x = x for all x ∈ V , where 0 is the additive identity and 1 the
multiplicative identity of the field F.
Often 0 is also used to represent the zero vector and the context will clarify this
without much difficulty. If F is the field of real (complex) numbers then V is called
a real (complex) vector space. In what follows, we will be concerned only with real
or complex vector spaces. Also a linear subspace V1 of V is a subset of V which is
a linear space over F with vector addition and scalar multiplication of V restricted
to V1 .
Definition 1.3.2 A subset S of a linear space V over a field F is said to be linearly

n
independent if for every finite subset {s1 , . . . , sn } of S, αi si = θ implies αi = 0
i=1

n
for i = 1, 2, . . . , n where αi ∈ F. An element of the form αi si where αi ∈ F
i=1
and si ∈ S is called a finite linear combination of members of S.
Definition 1.3.3 A subset S of a linear space V over a field F is said to span V if
every element of V can be written as a finite linear combination of elements from S.
A maximal linearly independent subset of a linear space V over F is called a basis
for V .
Any two bases of a vector space have the same cardinality.
Definition 1.3.4 The cardinality of a basis of a linear space V is called the dimension
of the linear space. If a linear space has a finite dimension, then it is called a finite-
dimensional vector space. Otherwise the linear space is infinite-dimensional.
Definition 1.3.5 Let (X, +, ·) be a linear space over F = R or C. A map · :
X → R+ is called a norm if the following conditions are satisfied:
10 1 Prerequisites
1. x = 0 if and only if x = 0;

2. x + y ≤ x + y for all x, y ∈ X (triangle inequality);
3. αx = |α|x for all α ∈ F and all x ∈ X , |α| being the modulus of α. The
pair (X, · ) is called a normed linear space.
Remark 1.3.6 If (X, · ) is a normed linear space, then d(x, y) = x − y, x, y ∈

X is a metric on X .
Definition 1.3.7 A normed linear space (X, · ) is called a Banach space if it is

complete in the metric induced by the norm.
Remark 1.3.8 The linear spaces in (ii)–(v) of Example 1.2.4 are Banach spaces
with the norms defined by x = d(x, θ) where d is the metric described in the
corresponding case, while (vi) of Example 1.2.4 is a normed linear space under the
b
norm a | f (t)|dt. However, this is not a Banach space.
Definition 1.3.9 Let (X i , · i ), i = 1, 2 be normed linear spaces over F = R or C.

A linear operator is a map T : X 1 → X 2 such that T (αx + β y) = αT (x) + βT (y)
for all α, β ∈ F and x, y ∈ X 1 . If X 2 is the base field F (=R or C with the modulus
or absolute value as a norm) which is also a normed linear space, the linear operator
is called a linear functional. If the linear operator T is continuous as a map between
the metric spaces X 1 and X 2 with metrics induced by the norms, then it is called a
continuous linear operator.
Theorem 1.3.10 Let (X i , · i ), i = 1, 2 be normed linear spaces over F = R or

C and T : X 1 → X 2 a linear operator. Then the following are equivalent:
(i) T is continuous on X 1 ;
(ii) T is continuous at some x0 ∈ X 1 ;
(iii) there exists K > 0 such that T x2 ≤ K x1 for all x ∈ X 1 .
Remark 1.3.11 A linear operator satisfying (iii) of Theorem 1.3.10 is called bounded.
In view of the theorem above, bound linear operators are precisely continuous linear
operators.
If T : X 1 → X 2 is a continuous linear operator where (X i , · i ), i = 1, 2 are
normed linear spaces, then
inf{K ≥ 0 : x ∈ X 1 and T x2 ≤ K x1 } = sup{T x2 : x1 = 1, x ∈ X 1 }
is finite and is called the norm of the linear operator and is denoted by T .
Theorem 1.3.12 For i = 1, 2, let (X i , · i ) be normed linear spaces. B(X 1 , X 2 )

the set of all bounded (continuous) linear operators is a normed linear space under
the norm described in Remark 1.3.11. If (X 2 , · 2 ) is complete so is B(X 1 , X 2 )
under this norm.
Theorem 1.3.13 (Hahn–Banach) If f is a bounded linear functional from a linear

subspace N of a normed linear space (X, · ), then there is a bounded linear
functional f ∗ on X such that f ∗ ≡ f on N and f = f ∗ .
The Hahn–Banach theorem insures the abundance of continuous linear function-
als in any nontrivial normed linear space.
Definition 1.3.14 Given a normed linear space (X, · ), the space of all continuous
linear functionals on X is called the dual or conjugate of X and is denoted by X ∗ .
The dual of X ∗ denoted by X ∗∗ is called the second dual or second conjugate of X .
Even, if X is incomplete, X ∗ and X ∗∗ are complete.
Theorem 1.3.15 Let (X, · ) be a normed linear space. The map f x defined by
f → f (x) for each x ∈ X is a bounded linear functional on X ∗ and f x = x.
The map ϕ : X → X ∗∗ defined by φ(x) = f x is one-one, isometric linear map of
X into X ∗∗ and is called the duality mapping. The duality mapping is the natural
embedding (of X into X ∗∗ ).
Definition 1.3.16 If the duality mapping ϕ maps X onto X ∗∗ , the second dual of X ,
then X is said to be reflexive.
Definition 1.3.17 If (X, · ) is a normed linear space, then the weak topology on
X is the smallest topology on X with respect to which all the functionals of X ∗ are
continuous. The weak * topology on X ∗ is the smallest topology on X ∗ such that
ϕ(x) (= f x ), ϕ being the natural embedding of X into X ∗∗ is continuous.
Theorem 1.3.18 (Alaoglu) The unit sphere S ∗ in X ∗ is compact in the weak *
topology on X ∗ .
Theorem 1.3.19 A Banach space is reflexive if and only if the closed unit sphere
S = {x ∈ X : x ≤ 1} is compact in the weak topology.
The following three theorems are basic to Functional Analysis.
Theorem 1.3.20 (Open Mapping Theorem) For i = 1, 2, let (X i , · i ) be Banach
spaces. If T : X 1 → X 2 is a continuous linear operator mapping X 1 onto X 2 , then
T is an open mapping (i.e. a function for which the image of any open set is open).
Consequently a continuous linear bijection of X 1 onto X 2 is a linear homeomorphism.
Theorem 1.3.21 (Closed Graph Theorem) For i = 1, 2, let (X i , · i ) be Banach
spaces and T : X 1 → X 2 a linear operator. T is continuous if and only if the graph of
T = {(x, T x) : x ∈ X 1 } is a closed subset of the product topological space X 1 × X 2 .
Theorem 1.3.22 (Banach–Steinhaus theorem) Let Tλ : X 1 → X 2 , λ ∈ = φ be
continuous linear operators mapping a Banach space X 1 into a normed linear space
X 2 such that for each x ∈ X 1 , {Tλ (x) : λ ∈ } is a bounded set of real numbers.
Then {Tλ : λ ∈ } is bounded.
Theorem 1.3.23 A normed linear space is finite-dimensional if and only if the unit
sphere is compact.
12 1 Prerequisites
Theorem 1.3.24 If · i , i = 1, 2 are two norms on a finite-dimensional normed

linear space then · 1 and · 2 are equivalent in the sense that there exist two
positive numbers K 1 and K 2 such that
K 1 x1 ≤ x2 ≤ K 2 x1 for all x ∈ X.
Consequently a finite-dimensional normed linear space over R or C is equivalent to

n 21

R or C with the norm given by (x1 , . . . , xn ) =
n n
|xi | 2
and is a Banach
i=1
space.
Among normed linear spaces, inner product spaces have rich geometric prop-
erties. Many features of the Euclidean spaces carry over to inner product spaces.
Parseval identity for orthogonal functions has a crisp functional analytic formula-
tion.
Definition 1.3.25 A linear space (V, +, ·) over F = R or C is called an inner product
space, if there is a map < >: V × V → F called an inner product satisfying the
following conditions:
(i) < αx + β y, z >= α < x, z > +β < y, z > for all x, y, z ∈ V and α, β ∈ F;
(ii) < x, y > = < y, x > for all x, y ∈ V , z being the complex conjugate of z ∈ C;
(iii) < x, x > ≥ 0 for all x ∈ V and < x, x > = 0 if and only if x = θ.
Proposition 1.3.26 An inner product space is a normed linear space with the
√
norm · defined by x = < x, x > (the positive square root of < x, x >), as
| < x, y > | ≤ xy for all x, y in V (Schwarz inequality).
Definition 1.3.27 A Hilbert space is an inner product space which is complete in
the norm induced by the inner product.
Example 1.3.28 (i) Rn or Cn is a Hilbert space
n in thenorm induced by the inner
n
product defined by < x, y >= xi yi xi y i for x = (x1 , . . . , xn ), y =
i=1 i=1
(y1 , . . . , yn ) ∈ Rn (Cn ).
∞

(ii) 2 , the space of complex sequences (z n ) with |z n |2 < +∞ is a Hilbert
n=1
∞

space with the inner product defined by < x, y >= xn y n for x = (xn ),
n=1
y = (yn ) ∈ 2 .
(iii) CC [a, b], the linear space of all continuous complex-valued functions on [a, b]
b
is an inner product space under the inner product < f, g >= a f (x)g(x)d x.
This is not a Hilbert space as the space is not complete in the induced norm
b 1
2
f = a | f (x)|2 d x . However, its completion is L 2 [a, b], the Hilbert space
of Lebesgue measurable complex functions which are square-integrable with
respect to the Lebesgue measure.
Theorem 1.3.29 Every incomplete inner product space can be isometrically embed-
ded as a dense subspace of a Hilbert space.
Theorem 1.3.30 A normed linear space (X, ) is an inner product space if and
only if the following parallelogram law is valid:
for x, y ∈ X, x + y2 + x − y2 = 2(x2 + y2 )
Definition 1.3.31 A subset C of a linear space V over F = R or C is called convex

if t x + (1 − t)y ∈ C for all t ∈ [0, 1], whenever x, y ∈ C.
Theorem 1.3.32 A non-empty closed convex subset of a Hilbert space contains a
unique element with least norm.
Definition 1.3.33 Let (V, <, >) be an inner product space over R or C. x ∈ V is
said to be orthogonal to y ∈ V if < x, y > = 0 and we write x ⊥ y (or y ⊥ x). For
S ⊆ V , S ⊥ = {v ∈ V : v ⊥ s, for all s ∈ S}. S ⊥ is called the orthogonal complement
of S.
Theorem 1.3.34 If M is a proper closed linear subspace of a Hilbert space H , then
(i) M ⊥ is a closed linear subspace of H ;
(ii) M ∩ M ⊥ = {θ};
(iii) each h ∈ H can be written uniquely as h = m 1 + m 2 , where m 1 ∈ M, m 2 ∈ M ⊥
and h2 = m 1 2 + m 2 2 .
(In this case, we write H = M ⊕ M ⊥ and call H the direct sum of M and its
orthogonal complement M ⊥ ).
Definition 1.3.35 A non-empty set S of a Hilbert space H is called orthogonal if
< x, y > = 0 whenever x, y ∈ S and x = y. S is called orthonormal if each element
of S has unit norm and S is orthogonal.
Theorem 1.3.36 If S = {eλ : λ ∈ = φ} is an orthonormal set in a Hilbert space
H and if x ∈ H , then {eλ :< x, eλ >= 0} is either empty or countable. Also

| < x, eλ > |2 ≤ x2 . Further, a nonzero Hilbert space has a maximal
λ∈
basis. If {eλ : λ ∈ = φ} is an
orthonormal set of vectors, called an orthonormal
orthonormal basis for H and x ∈ H , then x = aλ eλ , where aλ =< x, eλ >.
λ∈
Theorem 1.3.37(Parseval identity) If {eλ : λ ∈ = φ} is an orthonormal basis of

H then x =
2
| < x, eλ > |2 .
λ∈
Example 1.3.38 (i) L 2 [0, 2π], the space of complex-valued Lebesgue measur-
able functions f on [0, 2π] which are square-integrable in the sense that
2π
| f (x)|2 d x < +∞ is a Hilbert space under the inner product < f, g >=
02π inx
0 f (x)g(x)d x. The set √e 2π : n = 0, ±1, ±2, . . . is an orthonormal basis.
14 1 Prerequisites
∞
(ii) L 2 (R), the space of all Lebegue-measurable functions for which −∞ | f 2 (x)|d x
∞
is finite, is also a Hilbert space under the inner product < f, g >= −∞ f (x)
x2
g(x)d x. {x n e− 2 , n = 0, 1, 2, . . . } gives rise to an orthonormal basis for L 2 (R)
via the Gram–Schmidt orthogonalization process (see Simmons [13]).
Among normed linear spaces, strictly convex spaces and the more specialized
uniformly convex spaces resemble the Euclidean spaces geometrically.
Definition 1.3.39 A normed linear space (N , · ) over F = R or C is said to be

strictly convex if for x, y ∈ N with x = y = 1 and x = y, x+y
2
< 1.
Definition 1.3.40 A normed linear space (N , · ) is called uniformly convex if

there exists an increasing positive function δ : (0, 2]
→ (0, 1] such that for x, y ∈ N ,
x, y ≤ r and x − y ≥ r imply that x+y 2
< (1 − δ())r .
Remark 1.3.41 The above definition is equivalent to the requirement that for
xn , yn ≤ 1 and xn + yn → 2, xn − yn → 0 as n → ∞.
Clearly, every Hilbert space is uniformly convex. Also L p [0, 1] for p ≥ 2 is
uniformly convex. While every uniformly convex space is strictly convex, C[0, 1] is
not even strictly convex.
Hilbert spaces are isometric to their duals, in view of the following.

Theorem 1.3.42 (Riesz Representation Theorem) Let H be a Hilbert space over R
or C and f ∈ H ∗ , the dual of H . Then there exists a unique element y f ∈ H such
that f (x) =< x, y f > for each x ∈ H and f = y f .
For f y ∈ H ∗ defined by f y (x) = < x, y > the correspondence Ty = f y maps H
onto H ∗ so that T (y) = y, T (y1 + y2 ) = T y1 + T y2 and T (αy) = αT (y) for
all y ∈ H .
Theorem 1.3.43 Every Hilbert space is reflexive.
In view of the above theorems for a bounded linear operator T : H → H , H being

a Hilbert space over R or C, there is a unique bounded linear operator T ∗ : H → H
such that < T x, y > = < x, T ∗ y > for all x, y ∈ H .
Definition 1.3.44 Let H be a Hilbert space and T : H → H a bounded linear

operator. A linear operator T ∗ : H → H satisfying, < T x, y > = < x, T ∗ y > for
all x, y ∈ H is called an adjoint operator of T .
Theorem 1.3.45 If T ∈ B(H ), the space of all bounded linear operators mapping
H into itself, then T ∗ the adjoint of T is uniquely defined. Further,
(i) (T1 + T2 )∗ = T1∗ + T2∗ ,
(ii) (αT )∗ = αT ∗ ,
(iii) (T1 T2 )∗ = T2∗ T1∗ ,
(iv) (T ∗ )∗ = T ∗ ,
(v) T ∗ 2 = T 2 = T ∗ T .
Definition 1.3.46 A linear operator T ∈ B(H ), the space of all bounded linear oper-
ators on a Hilbert space H is said to be self-adjoint if T = T ∗ .
Definition 1.3.47 For T ∈ B(H ), the spectrum of T is the set {λ ∈ C : T − λI is
not invertible}, I being the identity operator. An eigenvalue of T is a number λ ∈ C
such that there exists a nonzero vector x0 ∈ H with T x0 = λx0 and in this case x0 is
called an eigenvector (corresponding to the eigenvalue λ).
Theorem 1.3.48 For T ∈ B(H ), the space of all bounded linear operators on a
Hilbert space H is self-adjoint if and only if < T x, x > is real for all x ∈ H . So
the eigenvalues of a self-adjoint operator are real. Further σ(T ), the spectrum of
T lies in [m, M], where m = inf{< T x, x >: x ∈ H and x = 1} and M = sup{<
T x, x >: x ∈ H and x = 1}. Also, m, M ∈ σ(T ).
Definition 1.3.49 A linear operator P in B(H ) is called a projection if P is self-
adjoint and P 2 = P.
Remark 1.3.50 If P is a projection on a Hilbert space H , then P = M ⊕ M ⊥ where
M = {P x : x ∈ H }, the range of P and M ⊥ , the range of I –P. Further, every rep-
resentation of H as the orthogonal sum M + M ⊥ defines a unique projection of H
onto M.
Theorem 1.3.51 For any self-adjoint operators T in B(H ), there is a family {Pλ :
λ ∈ R} of projections on H satisfying the following conditions:
(i) if T C = C T for C ∈ B(H ), then Pλ C = C Pλ for all λ ∈ R;
(ii) Pλ Pμ = Pλ , if λ < μ;
(iii) Pλ−0 = lim Pμ = Pλ (i.e. Pλ is continuous from the left with respect to λ);
μ→λ−0
(iv) Pλ = 0 if λ ≤ m and Pλ = I for λ > M.
(Such a family of projections Pλ is called a resolution of identity generated by
T ).
Theorem 1.3.52 (Spectral theorem) For every self-adjoint operator T ∈ B(H ) and
any > 0,
M+
T = λd Pλ
m
where the Stieltjes integral is the limit of (appropriate) integral sums in the operator-
norm topology.
Definition 1.3.53 A linear operator T : N1 → N2 where N1 and N2 are normed
linear spaces is said to be a compact operator if T (U ) is compact in N2 for each
bounded subset U of N1 .
Theorem 1.3.54 Let T : B → B be a compact linear operator on a Banach space
B. σ(T ), the spectrum of T is finite or countably infinite and is contained in
[−T , T ]. Every nonzero number in σ(T ) is an eigenvalue of T . If σ(T ) is
countably infinite, then 0 is the only limit point of σ(T ).
16 1 Prerequisites
1.4 Topological Vector Spaces
It is convenient to recall the definition of a topological group and list some of its
properties (see Kelley [7], Rudin [11] and Royden [10]).
Definition 1.4.1 Let (G, ·) be a group with the identity element e and for each
x ∈ G, x −1 denote the inverse of x (with respect to the binary operation ·). The triple
(G, ·, T ) is called a topological group where T is a topology on the group G with
the binary operation · such that the map (x, y) → x y −1 mapping G × G into G is
continuous. (Here G × G carries the product topology.)
If (G, ·) is a group and A, B ⊆ G, we write A · B = {a · b : a ∈ A, b ∈ B}.
Theorem 1.4.2 Let (G, ·, T ) be a topological group with the identity e. Then
(i) the map x → x −1 mapping G into G and the map (x, y) → x y mapping G × G
into G are continuous. Conversely if T1 is a topology on a group (G, ·) such
that x → x −1 and (x, y) → x y are continuous on G with the topology T1 , then
(G, ·, T1 ) is a topological group.
(ii) the inversion map i, defined by i(x) = x −1 is a homeomorphism of G onto
G; for each a ∈ G, L a (Ra ) called the left (right) translation by a, defined by
L a (x) = ax (Ra (x) = xa) are homeomorphisms;
(iii) a subset S of G is open if and only if for each x ∈ S, x −1 S (or equivalently
Sx −1 ) is a neighbourhood of e;
(iv) the family N of all neighbourhoods of e has the following properties:
(iv-a) for U, V ∈ N , U ∩ V ∈ N ;
(iv-b) for U ∈ N , V.V −1 ⊆ U for some V ∈ N ;
(iv-c) for U ∈ N and x ∈ G, x.U.x −1 ∈ N ;
(v) the closure of a (normal) subgroup of G is a (normal) subgroup of G;
(vi) every subgroup G 1 of G with an interior point is both open and closed and G 1
is closed or G 1 − G 1 is dense in G 1 ;
(vii) G is Hausdorff if it is a T0 space in the sense that for every pair of distinct
points, there is a point for which some neighbourhood does not contain the
other point.
A topological vector space can be defined in analogy with a topological group.
Definition 1.4.3 The quadruple (X, +, ·, T ) where (X, +, ·) is a vector space over
F = R or C and T is a topology on X is called a topological vector space (linear
topological space) if the following assumptions are satisfied:
(i) (X, T ) is a T1 -space;
(ii) the function (x, y) → x + y mapping X × X into X is continuous and
(iii) the function (α, x) → α.x mapping F × X into X is continuous.
Often, we simply say that X is a topological vector space (or t.v.s for short) when
the topology T on X and the vector space operations are clear from the context.
1.4 Topological Vector Spaces 17
Definition 1.4.4 A subset S of a topological vector space X is said to be bounded

for every neighbourhood V of θ in X , there is a real number s such that S ⊆ t.V for
every t > s. S ⊆ X is called
balanced if α.S ⊆ S for all α ∈ F with |α| ≤ 1. S is
called absorbing if X = t.S.
t>0
Theorem 1.4.5 Let X be a t.v.s. For each a ∈ X and λ = 0 ∈ F define the transla-
tion operator Ta and the multiplication operator Mλ by the rules Ta (x) = x + a and
Mλ (x) = λ.x respectively for each x ∈ X . Then, Ta and Mλ are homeomorphism of
X onto X .
Further G ⊆ X is open if and only if Ta (G) is open for each a ∈ X . So the local
base at 0 completely determines the local base at any x ∈ X and hence the topology
on X .
Remark 1.4.6 Every normed linear space is a t.v.s.
Definition 1.4.7 A function p mapping a vector space X over F(=R or C) into F
is called a seminorm if
(i) p(x + y) ≤ p(x) + p(y) for all x, y ∈ X and
(ii) p(αx) = |α| p(x) for all x ∈ X and all α ∈ F.
A seminorm is a norm if p(x) = 0 for x = θ. A family P of seminorms is sepa-
rating if for each x = y, there is a seminorm p ∈ P with p(x − y) = 0.
Theorem 1.4.8 If P is a separating family of seminorms on a vector space V , then
V ( p, n) = {x ∈ X : p(x) < n1 }, p ∈ P is a local base of convex sets for a topology
T on X . Thus, (X, T ) is locally convex and each p is continuous. Also, E is bounded
if and only if p(E) is bounded for each p ∈ P.
Definition 1.4.9 For an absorbing subset A of a t.v.s. X , the map μ A : X → R
defined by μ A (x) = inf{t > 0 : t −1 x ∈ A} is called the Minkowski functional of A.
Listed below are some of the basic properties and features of a topological vector
space.
Theorem 1.4.10 Let X be a topological vector space
(i) if S ⊆ X , S = ∩{S + V : V is a neighbourhood of 0};
(ii) if S1 , S2 ⊆ X , S1 + S2 ⊆ S1 + S2 ;
(iii) if C ⊆ X is convex, so are C 0 and C;
(iv) if B ⊆ X is balanced, so is B and if in addition 0 ∈ B 0 , B 0 is balanced;
(v) the closure of a bounded set is also bounded;
(vi) every neighbourhood of 0 also contains a balanced neighbourhood of 0 and
so X has a balanced local base;
(vii) every convex neighbourhood of 0 contains a balanced convex neighbourhood
of 0;
∞

(viii) if V is a neighbourhood of 0 and rn ↑ +∞ where r1 > 0, X = rn V ;
n=1
18 1 Prerequisites
(ix) if V is a bounded neighbourhood of 0 and δn ↓ 0, δ1 > 0, {δn V : n ∈ N} is a

local base at 0;
(x) if X is first countable, then it is metrizable and the metric is translation invari-
ant;
(xi) if X is locally compact, then X is finite dimensional.
Theorem 1.4.11 If A is a convex absorbing subset of a vector space X , then

(i) μ A (x + y) ≤ μ A (x) + μ A (y) for all x, y ∈ X ;
(ii) μ A (t x) ≤ tμ A (x) for t ≥ 0;
(iii) μ A is a seminorm, when A is balanced;
(iv) B = {x : μ A (x) < 1} ⊆ A ⊆ C = {x : μ A (x) ≤ 1} and μ A = μ B = μC .
Theorem 1.4.12 If B is a local base for a t.v.s. (X, J ) comprising convex balanced
neighbourhood, then {μV : V ∈ B} is a family of continuous seminorms that are
separating (i.e. for x, y ∈ X , then there is a μV such that μV (x) = μV (y)). Further,
the topology having a local base generated by these seminorms of the form {x :
μV (x) < n1 }, V ∈ B, n ∈ N coincides with the topology on X .
Definition 1.4.13 A t.v.s is said to be locally convex if it has a local base of convex
sets. It is called an F-space if the topology is generated by complete translation-
invariant metric. A locally convex F-space is called a Frechet space.
Theorem 1.4.14 If P = { pi : i ∈ N} is a countable separating family of seminorms

on a vector space X , then the topology on X induced by P is metrizable and this
metric d is given by
∞
pi (x, y)
d(x, y) = i (1 + p (x, y))
i=1
2 i
is translation invariant.
Theorem 1.4.15 (Kolmogorov) A topological vector space is normable if and only

if the origin has a convex balanced neighbourhood.
Example 1.4.16 Let be the union of a sequence of compact sets K n ⊆ R m

for n = 1, 2, . . . with K n ⊆ K n+1 o
, n = 1, 2, . . . . Define for each f ∈ C(), the
set of all complex-valued functions on , pn ( f ) = sup{| f (x)| : x ∈ K n }. Then,
{ pn , n = 1, 2, . . . } is a separating family of continuous seminorms defining a com-
plete translation-invariant metric on C(). As the origin has no bounded neighbour-
hood, C() is non-normable. Since C() is locally convex, it is a Frechet space.
If is any non-empty open subset C, then H (), the set of all complex functions
analytic on is a closed subspace of C(). H () too is not normable.
Example 1.4.17 Let be a non-void open set in Rn . A multi-index α is an

ordered n-tuple of the form α = (α1 , . . . , αn ) where αi are non-negative inte-
gers. For each multi-index, the differential operator D α associated is defined by
α1 αn
D α = ∂x∂ 1 . . . ∂x∂ n whose order is |α| = α1 + · · · + αn and for |α| = 0,
D α f = f . A complex-valued function defined on is said to belong to C ∞ () if

∞

D α f ∈ C() for every multi-index α. Let = K m where each K m is compact
m=1
and K m ⊆ K m+1
0
, m = 1, 2, . . . . Define the seminorms φm on C ∞ (), m = 1, 2, . . . ,
by φm ( f ) = sup{|D α f (a)| : x ∈ K m , |α| ≤ m}. Then, C ∞ () is a Frechet space
under the topology generated by the seminorms φm . Although every closed bounded
subset of C ∞ () is sequentially compact (and hence compact in this case), C ∞ ()
is not locally bounded and hence not normable.
Example 1.4.18 For 0 < p < 1, let L p [0, 1] be the linear space of all Lebegue-
1
measurable functions f on [0, 1] for which δ( f ) = 0 | f (a)| p d x < +∞. Then d,
defined by d( f, g) = δ( f − g) defines a translation-invariant metric on L p [0, 1] and
this metric is complete. Thus L p [0, 1] is an F-space. However, it is not locally
convex. Indeed L p [0, 1] is the only open convex set. So, 0 is the only continuous
linear functional on L p [0, 1] for 0 < p < 1 (See Rudin [12]).
Definition 1.4.19 Let X be a topological vector space. The dual of X , denoted by

X ∗ is the set of all continuous linear functionals on X .
Theorem 1.4.20 If X is a locally convex t.v.s, then X ∗ separates points in X .
Definition 1.4.21 Let K be a non-empty subset of a vector space X . A point s ∈ K

is called an extreme point of K if s = t x + (1 − t)y for t ∈ (0, 1) for some x, y ∈ K
implies x = y = s. The convex hull of a set E ⊆ X is the smallest convex set in X
containing E. The closed convex hull of E is the closure of its convex hull.
Theorem 1.4.22 (Krein-Milman [11]) If X is a topological vector space on which

X ∗ separates points. Every compact convex set in X is the closed convex hull of the
set of its extreme points. So in a locally convex t.v.s X every compact convex set in
X is the closed convex hull of the set of its extreme points.
In this context, it is pertinent to recall Riesz Representation theorem (see Rudin

[12]).
Theorem 1.4.23 (Riesz-Representation) Let X be a locally compact T2 space and L
be a positive linear functional on CC (X ) the linear space of all continuous complex-
valued functions with compact support and the supremum norm. Then, there exists
a σ-algebra S on X containing all the Borel subsets of X and a unique positive
measure μ on S representing L according to the formula

Lf = f dμ for f ∈ CC (X )
X
with the following properties:

(i) μ(K ) < +∞ for each compact subset of X ;
(ii) for each E ∈ S , μ(E) = inf{μ(G) : G ⊇ E and G is open in X };
20 1 Prerequisites
(iii) μ(E) = sup{μ(K ) : K ⊆ E and K compact} is true for each open set E and
for any E ∈ S with μ(E) < +∞;
(iv) for E ∈ S with μ(E) = 0, A ∈ S for any A ⊆ E and μ(A) = 0.
When X is compact, μ can be chosen so that μ(X ) = 1, i.e. a Borel probability
measure.
Remark 1.4.24 In a Frechet space, for the convex hull H of a compact set, H is
compact and in a finite-dimensional space Rn , H itself is compact. Also if an element
x lies in the convex hull and a set E ⊆ Rn , then it lies in the convex hull of a subset
of E that contains at most n + 1 points.
We now proceed to define vector-valued integrals. Rudin [11] may be consulted

for further details.
Definition 1.4.25 Let (Q, J, μ) be a measure space, X a t.v.s for which X ∗ separates
points and f : Q → X be a function such that f is integrable with respect to μ for
each ∈ X ∗ (here ( f )(q) = ( f (q)) for q ∈ Q). If there exist y ∈ X such that

y = f dμ
Q
for each ∈ X ∗ , then we define

f dμ = y.
Q
Theorem 1.4.26 Let X be a t.v.s such that X ∗ separates points and μ be a Borel
probability measure on a compact Hausdorff space Q. If f : Q → X is continuous
and if the convex hull H of f (Q) has compact closure H in X , then the integral

y= f dμ
Q
exists (as per Definition 1.4.25).
Theorem 1.4.27 Let X be a t.v.s such that X ∗ separates points and Q, a compact
subset of X and H , the closed convex hull of Q be compact.
y ∈ H if and only if there is a regular Borel probability measure μ on Q such that

y= xdμ.
Q
When X is a Banach space we also have

Theorem 1.4.28 Let Q be a compact T2 space, X a Banach space, f : Q → X a
continuous map and μ a positive Borel probability measure on Q. Then

f dμ ≤ f dμ.
Q Q
Indeed vector-valued integrals can also be defined more directly as limits of

(integral) sums.
References
1. Bartle, R.G.: The Elements of Real Analysis, 2nd edn. Wiley, New York (1976)
2. Bollobas, B.: Linear Analysis, An Introductory Course, 2nd edn. Cambridge University Press,
Cambridge (1999)
3. Dugundji, J.: Topology. Allyn and Bacon Inc., Boston (1966)
4. Halmos, P.R.: Naive Set Theory. D. Van Nostrand Co., Princeton (1960)
5. Kantorovich, L.V., Akilov, G.P.: Functional Analysis, Translated from the Russian by Howard
L. Silcock, 2nd edn. Pergamon Press, Oxford-Elmsford (1982)
6. Kaplansky, I.: Set Theory and Metric Spaces, 2nd edn. Chelsea Publishing Co., New York
(1977)
7. Kelley, J.L.: General Topology. D. Van Nostrand Company, Inc., Toronto (1955)
8. Lusternik, L.A., Sobolev, V.J.: Elements of Functional Analysis. Hindustan Publishing Corpo-
ration, Delhi (1974)
9. Munkres, J.R.: Topology: A First Course. Prentice-Hall, Inc., Englewood Cliffs (1975)
10. Royden, H.L.: Real Analysis, 3rd edn. Macmillan Publishing Company, New York (1988)
11. Rudin, W.: Functional Analysis. Tata McGraw-Hill Publishing Co. Ltd., New Delhi (1974)
12. Rudin, W.: Real and Complex Analysis, 3rd edn. McGraw-Hill Book Co., New York (1987)
13. Simmons, G.F.: Introduction to Topology and Modern Analysis. McGraw-Hill Book Co., Inc.,
New York (1963)
14. Taylor, A.E.: Introduction to Functional Analysis. Wiley, New York; Chapman Hall, Ltd.,
London (1958)
Chapter 2
Fixed Points of Some Real and Complex
Functions
This chapter highlights some fixed point theorems for certain real and complex
functions.
2.1 Fixed Points of Continuous Maps on Compact Intervals

of R
The following definitions are well-known.
Definition 2.1.1 Let f, g : X → Y be maps, X and Y being non-empty sets. An

element x0 ∈ X is called a coincidence point of f and g if f (x0 ) = g(x0 ). If f : X →
X is a map and if for some x0 ∈ X , f (x0 ) = x0 , then x0 is called a fixed point (fix
point) of f . If f, g : X → X are maps such that for some x0 ∈ X , x = f (x0 ) = g(x0 ),
then x0 is called a common fixed point of f and g.
Definition 2.1.2 Let f : X → X be a map on a non-void set X . The sequence

{ f n (x)} called the sequence of f iterates is defined recursively by : f 0 (x) = x,
f 1 (x) = f (x), f n+1 (x) = f ( f n (x)), n = 0, 1, 2, . . . ,. This sequence is called a
sequence of ( f ) iterates generated at x. We also call the set { f k (x) : k = 0, 1, 2, . . . }
the orbit of x under f and denote it by O f (x). f m (x) is called the mth iterate of f
at x.
Definition 2.1.3 For a map f : X → X , x0 ∈ X is called a periodic point of period

m if f m (x0 ) = x0 and f n (x0 ) = x0 for n < m.
The classical intermediate value theorem for real functions due to Bolzano is
equivalent to Brouwer’s fixed point theorem for real functions on intervals of real
numbers. In a sense, Bolzano’s theorem can be viewed as the harbinger of fixed point
theory.
https://doi.org/10.1007/978-981-13-3158-9_2
24 2 Fixed Points of Some Real and Complex Functions
Theorem 2.1.4 (Bolzano’s Intermediate Value Theorem) If g : [a, b] → R is a con-

tinuous function then for every real number r between g(a) and g(b), there is an
element c = c(r ) between a and b such that g(c) = r .
Proof Without loss of generality, we can assume that g(a) = g(b). Since g is contin-
uous, g[a, b] is a connected subset of R containing g(a) and g(b). Since connected
subsets of R are intervals, the interval with g(a) and g(b) as endpoints is in the range
of g. Hence if r lies between g(a) and g(b), there is an element c = c(r ) between a
and b such that g(c) = r .
As an immediate consequence, we have
Theorem 2.1.5 (Brouwer’s fixed point theorem in R) If f : [a, b] → [a, b] is a
continuous function, then f has a fixed point.
Proof If f (a) = a or f (b) = b, then the theorem is true. So without loss of gen-
erality we assume that f (a) = a and f (b) = b. Since function g : [a, b] → R
defined by g(x) = f (x) − x is continuous on [a, b] and g(a) = f (a) − a > 0 and
g(b) = f (b) − b < 0 (as f (a), f (b) ∈ (a, b)) by Theorem 2.1.4, there is a point
c ∈ [a, b] such that g(c) = 0 ∈ [g(b), g(a)]. Thus c is a fixed point of f .
Remark 2.1.6 The above fixed point theorem, a consequence of the intermediate
value theorem, is indeed equivalent to this theorem.
Let g : [a, b] → R be continuous. Without loss of generality let g(a) < r < g(b).
Define the map f : [−1, 1] → [−1, 1] by
⎛ ⎞
(1−t)a (1+t)b
{r − g 2
+ 2
f (t) = ρ ⎝t − ⎠
g(b) − g(a)
where ρ(x) = −1 for x < −1 and ρ(x) = 1 for x > 1 and ρ(x) = x for other real
numbers. Since g is continuous and ρ is continuous on R, clearly f is continuous and
t0 ∈ [−1, 1]. Further
maps [−1, 1] into itself. So by Theorem 2.1.5, f has a fixed point
(1−t0 )a (1+t )b
r −g + 20
t0 is neither −1 nor 1 and −1 < t0 < 1. So t0 = f (t0 ) = t0 −
2
g(b)−g(a)
.

Hence r = g (1−t20 )a + (1+t2 0 )b . In short, g has the intermediate value property.
The following is another useful fixed point theorem.

Theorem 2.1.7 Let f : [a, b] → R be a continuous map such that f [a, b] ⊇ [a, b].
Then f has a fixed point.
Proof Since f [a, b] ⊇ [a, b], [a, b] = [ f (c), f (d)] for some interval with end
points c and d lying [a, b]. If c ≤ d, then f (c) ≤ a ≤ c ≤ d ≤ b ≤ f (d). Thus
f (x) − x changes sign in [c, d] and hence by Theorem 2.1.4 has a zero, which is a
fixed point of f . If c ≥ d, then f (d) ≤ d ≤ c ≤ f (c). Thus again f (x) − x changes
sign in [d, c] and so has a fixed point.
2.1 Fixed Points of Continuous Maps on Compact Intervals of R 25
Remark 2.1.8 Theorem 2.1.4 is not true if the interval is not compact. the map x →
x + 1 is continuous but has no fixed point in (−∞, ∞) or [0, ∞). The continuous
map x → 1+x 2
on [0, 1) has no fixed point in [0, 1). Theorem 2.1.4 fails even if f is
continuous everywhere on [a, b] except at a single point. For instance f : [0, 1] →
[0, 1] defined by
x
, x = 0
f (x) = 2
1, x = 0
has no fixed point, and x = 0 is the only point of discontinuity of f .
Remark 2.1.9 F f , the set of fixed points of a continuous map on [a, b] is closed.
Indeed F f = {x ∈ [a, b] : f (x) = x} = g −1 (0) where g : [a, b] → R is defined by
g(x) = f (x) − x. Since {0} is a closed set and g is continuous g −1 {0} is a closed
subset. So F f is a closed subset of [a, b] ([a, b] being compact, F f is also compact).
Remark 2.1.10 Indeed we can prove that for each closed subset F of [0, 1] there
is a continuous map f : [0, 1] → [0, 1] for which F is the set of fixed points of
f . For proving this we can, without loss of generality, assume that 0, 1 ∈ F. So
[0, 1] − F = G is open and is a countable union of disjoint open intervals (ai , bi ),
i ∈ N. Now we consider the case when this collection is countably infinite, leaving
the case of finite collection as an exercise.
For n ∈ N define f n : [0, 1] → [0, 1] by
⎧ ∞
⎪
⎨x, x ∈ F ∪ i=n (a , b ),
ai +bi i i
f n (x) = ai , if x ∈ ai , 2 for i < n,
⎪
⎩ ai +bi
2x − bi , if x ∈ 2 , bi for i < n.
It can be seen that the sequence of continuous functions ( f n ) converges uniformly to

a continuous function f for which f (x) = x when x ∈ F and f (x) = x if x ∈ / F.
In fact, the result is true for any non-empty closed subset of R.
2.2 Iterates of Real Functions
In this section, some theorems on the behaviour of iterates of real functions are
discussed. First, Krasnoselskii’s theorem on the convergence of special iterates of
non-expansive maps of [a, b], following Bailey’s [2] proof using elementary prop-
erties of subsequential limits is discussed in detail. Theorems 2.2.6–2.2.8 detail the
rates of convergence of iterates of special class of functions and are due to Thron
[30].
Theorem 2.2.1 (Krasnoselskii [20], Bailey [2]) Let f : I (= [a, b]) → I be a map
such that | f (x) − f (y)| ≤ |x − y| for all x, y ∈ I . For any x ∈ I , the sequence (xn )
defined recursively by xn+1 = 21 (xn + f (xn )), n = 1, 2, . . . , converges to some fixed
point of f .
Proof Suppose that (xn ) does not converge to a fixed point. We show that this leads
to a contradiction. To this end, the proof is divided into several steps.
Step I. If (xn ) converges to z ∈ I , then (xn+1 ) also converges to z. As xn+1 =

1
(x + f (xn )), and f is continuous, xn+1 converges to f (z)+z
2 n 2
. So z = f (z).
Step II. No subsequence of (xn ) converges to a fixed point of f . For, if (xni ) converges
to z and f (z) = z, then |z − xni +1 | ≤ |z − 21 (xni + f (xni )| ≤ 21 |z − xni | + 21 | f (z) −
f (xni )| (as z = 21 (z + f (z))) ≤ |z − xni | (since | f (x) − f (y)| ≤ |x − y|). This
shows that (xn ) itself converges to z, a fixed point of f , contradicting our assumption
that (xn ) does not converge to a fixed point of f .
Step III. Since (xn ) lies in the compact interval I = [a, b], it has a subsequential limit
p for which f ( p) > p. Otherwise for all subsequential limits p of (xn ), f ( p) ≤ p.
Let z be the infimum of all subsequential limits. Then z itself is a subsequential limit
of (xn ). So f (z) ≤ z. If f (z) < z, then f (z) < 21 ( f (z) + z) < z and 21 ( f (z) + z)
is a subsequential limit of (xn ) smaller than z, the smallest subsequential limit of
(xn ), we get a contradiction, unless f (z) = z. But by Step II above, f (z) cannot be
z. Thus, there is a subsequential limit p of (xn ) for which f ( p) > p.
Step IV. By Step II, there exists > 0 such that | f (x) − x| ≥ for all subsequential
limits x of (xn ). Otherwise, there is a sequence (wn ) of subsequential limits of (xn )
with |wn − f (wn )| < n1 for all n. This in turn implies that any subsequential limit
of (wn ), which is also a subsequential limit of (xn ) is a fixed point of f , contrary to
Step II.
Step V. Let w be the largest subsequential limit of (xn ) such that f (w) > w so
f (w) > Q = 21 ( f (w) + w) > w. Since Q is a subsequential limit exceeding w,
f (Q) < Q.
By Step IV, there is the least subsequential limit R of (xn ) such that f (R) < R
and w < R < f (w) (at least Q satisfies these conditions). Now f (R) < w.
Otherwise for A = 21 [R + f (R)], w < A < R. If f (R) ≥ w, then A = 21 (R +
f (R)) ≥ 21 (R + w) > 21 (w + w) = w and A = 21 (R + f (R)) < 21 (R + R) = R.
Since A is a subsequential limit greater than w, the largest subsequential limit less
than f (w), f (A) ≤ A. As A < R and R is the least subsequential limit with f (R) <
R, A ≤ f (A). Hence A = f (A) and this contradicts our assumption that no subse-
quential limit can be a fixed point of f . Hence f (R) < w. Consequently f (R) <
w < R < f (w) and |w − R| = R − w < | f (R) − f (w)| = f (w) − f (R). This is
a contradiction to the assumption on the map f that | f (x) − f (y)| ≤ |x − y| for all
x, y ∈ I . Hence (xn ) converges to a fixed point of f .
Remark 2.2.2 However, for any continuous map of I into itself, the sequence of iter-
ates defined in Theorem 2.2.1 may not converge. Let f : [0, 1] → [0, 1] be defined
by
2.2 Iterates of Real Functions 27
⎧
⎪ 3
⎨4 for 0 ≤ x ≤ 14

f (x) = 3 21 − x for 41 < x ≤ 21
⎪
⎩
0 for 21 ≤ x ≤ 1.
Clearly x = 38 is a fixed point of f . For x1 = 41 , x2 = 21 (x1 + f (x1 ) = 21 , x3 =

1
(x + f (x2 )) = 41 and so on. This shows that xn does not converge.
2 2
In this context, the following result due to Cohen and Hachigian [10] is pertinent.
Theorem 2.2.3 Let f : [−1, 1] → [−1, 1] be a continuous map such that f (−1) =
−1 and f (1) = 1. Then for each m = 0, 1, 2, . . . , f m+1 − I ≥ f m − I . Here I
denotes the identity map and g = sup{|g(x)| : x ∈ [−1, 1]) for any g ∈ C[−1, 1].
Proof If f ≡ I , the conclusion is obvious. So suppose that f = I . Let F = {x ∈
[−1, 1] : f (x) = x}. Since F is closed, the complement of F is open and so can be
written as a disjoint union of open subintervals Sα of [−1, 1]. For x ∈ Sα , f (x) <
x or f (x) > x. Clearly the conclusion is true for m = 0. Suppose the inequality
f k+1 − I ≥ f k − I is true for k = 1, 2, . . . , m. As [−1, 1] is compact and f m
is continuous, there exists p in [−1, 1] such that | f m ( p) − p| = f m − I .
Suppose without loss of generality f m ( p) > p. We claim that f ( p) > p. Clearly
f ( p) = p. If f ( p) < p, then for q = f ( p),
f m−1 − I ≥ | f m−1 (q) − q| = | f m ( p) − q|

= f m ( p) − q (as q f m ( p) − p = f m − I .
As this is a contradiction f ( p) > p. Let p ∈ Sα = (a, b). So for x ∈ Sα , f (x) > x.

As a, b ∈/ Sα , a = f (a) x in Sα
and r ∈ Sα , f (r ) = p > r . Now
f m+1 − I > | f m+1 (r ) − r | = f m ( p) − r

> f m ( p) − p = f m − I .
Thus for f different from I , the identity map
f m+1 − I ≥ f m − I , m = 0, 1, 2, . . . .
Cohen and Hachigian [10] have constructed an example of a continuous self-map
on the closed unit disc for which every point on the unit circle is a fixed point, with
the property that I − f > I − f k for some iterate f k of f .
For special real functions Thron [30] had obtained some interesting results on
the rates of convergence of iterates. Some of these are relevant to the solution of
Schroder’s functional equation. They provide useful estimates in approximating fixed
points by iterates.
Definition 2.2.4 A map g : R → R is said to belong to the class H (a1 , k) if for some
x0 > 0, 0 < g(x) < x for x ∈ (0, x0 ] and g(x) = a1 x + x k+1 h(x) for x ∈ [0, x0 ]
where 0 ≤ a1 ≤ 1, k is a positive number and k is a continuous function on [0, x0 ]
with |h(x)| < M in [0, x0 ].
Remark 2.2.5 Clearly for g ∈ H (a, k), 0 is the unique fixed point of g and every
sequence (xn ) of g-iterates defined by xn+1 = g(xn ), n ∈ N and x1 ∈ (0, x0 ] con-
verges to 0.
Theorem 2.2.6 Let g ∈ H (a1 , k) where 0 < a1 < 1. Then for the sequence (xn ) of
g-iterates, there exists a constant K 1 (g, x) such that
xn
lim = K1
n→∞ a1n
Proof From the definition of g and xn+1
xn+1 a1 xn + xnk+1 h(xn )

= = a1 + xnk h(xn )
xn xn
As (xn ) decreases to zero, there exists x0 ∈ N such that for x ≥ n 0
1 − a1
0 < xnk M <
2

Sor xxn+1 < 1+a 2
1
< 1. Hence xn and xnk h(xn ) converge. So, the infinite product
∞
n
x k h(xn )
1+ n converges to a number L (say). Writing u n = axnn it follows that
a 1 1
n=1
u n+1 x k
h(x )
un
= ax1n+1
xn
= 1+ n a n .
n
xmk h(xm )
Since u n+1 = u 1 1+ , u n+1 converges to u 1 L. Hence u n = axnn
m=1
a 1 1
converges to u 1 L (= K 1 (g, x1 )).

Theorem 2.2.7 If g ∈ H (a1 , k) for a1 = 0 and (xn ) is the sequence of iterates gen-
erated at x1 ∈ (0,n x0 ], then there is a constant K 2 (g, x1 ) with 0 < K 2 < 1 such that
0 < xn < K 2(k+1) for all n after some stage. If additionally lim inf h(x) > 0, then
x→0
−n
for some K 3 (g, x1 ) with 0 < K 3 < 1, lim xn(k+1) = K3.
x→∞
Proof Since a1 = 0 and xn+1 = xnk+1 h(xn ),

log xn+1 = (k + 1) log xn + log h(xn ).
Define vn = (k + 1)−n log xn . We obtain for n ≥ n 0
vn+1 = vn + (k + 1)−(n+1) log h(xn )

n
= vn 0 + (k + 1)−(m+1) log h(xm ). (2.2.1)
m=n 0
∞

If lim inf h(x) > 0, then (k + 1)−(m+1) log h(xm ) converges to a number
x→0
m=n 0
−n
K 3 (g, x1 ) − vn 0 , say. So (vn ) converges to log K 3 as n → ∞ or lim (xn )(k+1) =
n→∞
K3.
Suppose 0 < h(x) < M and that log h(xn ) could approach −∞ so that the series
(2.2.1) might not converge. Nevertheless, we have from (2.2.1)

n
vn+1 < (k + 1)−(m+1) log M + (k + 1)−n 0 log xn 0
m=n 0

−(n 0 +1) 1 − (k + 1)−n+n 0 +1
= log M(k + 1) + (k + 1)−(n 0 +1) log xnk+1
1 − (k + 1)−1 0
(2.2.2)
If log M < 0, choosing x0 such that xn 0 < 1, we get from (2.2.2)
vn < (k + 1)−(n 0 +1) log M < 0. (2.2.3)
If log M ≥ 0, (2.2.2) gives

1+k
vn < (k + 1)−(n 0 +1) log M k xnk+1
0
. (2.2.4)
For large n 0 , the right-hand side of (2.2.3) or (2.2.4) as the case may be is negative
and is set as log K 2 (g, x1 ).
Now vn < log K 2 for n ≥ n 0 . So 0 < xn < K 2(k+1) .
n

Theorem 2.2.8 Let g ∈ H (a1 , k) for a1 = 1. Then B1 = lim inf

+
−h(x) ≥ 0, B2 =
x→0
lim sup −h(x) ≤ M. Given > 0 for the sequence (xn ) of iterates in (0, x0 ] there
x→0+
exists N (, g, x1 ) so that
xn > [(B2 + )kn]− k for n > N .

1
If B1 > 0 and 0 < < B1 , then for some N (, g, x1 )
xn < [(B1 − )kn]− k for n > N

1
Proof Since g(x) = x + x k+1 h(x), g(x) < x and |h(x)| < M, 0 ≤ −h(x) < M
for x ∈ [0, x0 ]. Hence B1 ≥ 0 and B2 ≤ M. Writing −h(xn ) = dn , xn+1 = xn +
xnk+1 h(xn ) becomes, for k = 1
xn+1 = xn (1 − xn dn )
and so
1 1 1
= .
xn+1 xn (1 − xn dn )
∞

Choose n 1 (g, x1 , ) so that xn dn < 1, dnm xnm−1 < and B1 − < dn < B2 + .
m=2
3 3 3
For n ≥ n 1
∞
1 1
= + dn + dnm xnm−1 (by Binomial theorem)
xn+1 xn m=2
1 2
< + B2 + . (2.2.5)
xn 3
1
So xn 1 +m > .
2 1
m B2 + +
3 xn1
So for n ≥ n 1
1
xn >
n 1− n1
n
B2 + 2
3
+ 1
nxn 1
1
>
n B2 + 2
3
+ 1
nxn 1

Choose n 1 ≥ n 1 so that 1
nxn 1
< 3
for n ≥ n 1 . So we have for n ≥ n 1 ,
1
xn > .
n(B2 + )
From (2.2.5) for n ≥ n 1 , we get
1 1
> + B1 − .
xn+1 xn 3
So when B1 − > 0, for n > n 1
1 1
> + (n − n 1 )(B1 − ) or
xn xn1
1
xn < . (2.2.6)
n 1 − n (B1 − ) + (nxn 1 )−1
n1
Choose N > n 1 ≥ n 1 , such that for n > N ,

n1
1− B1 − > B1 − .
n 3
So for n ≥ N , we get from (2.2.6)
1
xn < .
n(B1 − )
For the case k = 1, define wn = xnk then xn+1 = g(xn ) = xn (1 + xnk h(xn )). So
1 k 1 k
wn+1 = g wnk = wn 1 + wn h wnk
= wn [1 + wn h 1 (wn )].
1
Since [g(wnk )]k is a function of wn , say g1 , it follows that g1 (w) ∈ h 1 (1, 1) for 0 ≤
w ≤ w0 = x0k . Also lim inf +
h 1 (w) = k B1 , lim sup −h 1 (w) = k B2 . The discussion
w→0 w→0+
now reduces the case k = 1 to the case k = 1 for g1 ∈ H (1, 1). It follows from the
previous discussion that for B1 > 0 and 0 < < B1 , there exists N ∈ N such that
for n > N
xn < [(B1 − )kn]− k
1
and for > 0, there exists N0 ∈ N such that for n > N0 ,
xn > [(B2 + )kn]− k .

1

x
Remark 2.2.9 Since g(x) = sin( x2 ) ∈ H ( 21 , 2) in (0, 1), lim (2n sin n ( )) converges
n→∞ 2
for each x ∈ (0, 1) by Theorem 2.2.6.
Remark 2.2.10 Theorem 2.2.7 can be applied to g(x) = sin(x 1+ ) for any > 0 in
−n
(0, 1) to conclude that for any sequence (xn ) of iterates of sin(x 1+ ), lim (xn(1+) )
n→∞
converges.
2.3 Periodic Points of Continuous Real Functions
This section treats Sharkovsky’s theorem on the existence of periodic points of con-
tinuous self-maps on a compact interval I ⊆ R. Sharkovsky published a fundamental
paper [27] on the existence of periodic points of continuous self-maps on compact
intervals in 1964, when he was about 27 years old. He introduced a new (total) order
on the set of natural numbers, often called Sharkovsky order. Interestingly, if a con-
tinuous map has a periodic point of period m, in the compact interval I (which it
maps into itself) it has periodic points of all periods ‘bigger than’ m (with respect to
this order). The smallest natural number in this order is 3 and so it turns out that if
a continuous function mapping [a, b] into itself has periodic point of period 3, then
it has periodic points of all periods. Another implication of Sharkovsky’s theorem is
that if such a map has an odd periodic point then it has periodic points of all even
periods.
The more remarkable feature of Sharkovsky’s theorem is that its proof is essen-
tially based on the ingenious applications of the intermediate value theorem. The
paper by Li and Yorke [21] in 1975 proving a special case of Sharkovsky’s theorem
as well as May’s paper [22] highlighted the complicated behaviour of iterates of
simple functions and brought to limelight Sharkovsky’s work. The ‘simple proof’ of
Sharkovsky’s theorem presented below is due to Bau-Sen Du [14].
In the following, we assume that f : I → I is a continuous map, where I is a
compact interval in R. The following total ordering in N, the set of natural numbers
is called Sharkovsky’s ordering ≺. m ≺ n in the following ordering:
3 ≺ 5 ≺ 7 ≺ · · · ≺ 2.3 ≺ 2.5 · · ·
≺ 22 .3 ≺ 22 .5 ≺ 22 .7 ≺ · · · ≺ 23 .3 ≺ 23 .5 ≺ · · ·
≺ · · · ≺ 2n .3 ≺ 2n .5 ≺ · · ·
≺ · · · ≺ 23 ≺ 22 ≺ 2 ≺ 1
Sharkovsky’s theorem states that if f : I → I has an m-periodic point then f has

an n-periodic point precisely when m ≺ n.
Lemma 2.3.1 Let a and b be points of I such that either f (b) < a < b ≤ f (a) or
f (b) ≤ a z and f 2 (x) < x for y < x ≤ v.
Proof Whether f (b) < a < b ≤ f (a) or f (b) ≤ a < b < f (a), f (x) − x changes
sign in (a, b) and hence has a zero in (a, b). In other words, f has a fixed point z in
(a, b). As b ≤ f (a), a < z < b, and f (z) = z, there exists v ∈ [a, z) with f (v) = b.
If f (x) > z when min I ≤ x ≤ v, let u = min I ; otherwise let u = max{x : min I ≤
x ≤ v, f (x) = z}. Then f 2 (u) ≥ u and f (x) > z for u < x ≤ v. Since f 2 (v)(=
f (b)) ≤ a < v, f 2 has a fixed point in [u, v) or f has a 2 periodic point in [u, v).
If y is the largest 2-periodic point, then u ≤ y < v < z < f (y). Since f 2 (v) < v,
f 2 (x) < x for each x in (y, v].
Remark 2.3.2 Let P be a period-m orbit of f with m ≥ 3. Let p, b ( p < b) be

points in P such that f ( p) ≥ b and f (b) ≤ p. So f has a fixed point in [ p, b]. Let
a ∈ [ p, b) be such that f (a) = b. Since f (b) < a (< b = f (a)), the hypotheses of
Lemma 2.3.1 are satisfied. Also b, as a point in P, has least period m.
Theorem 2.3.3 If f has a periodic point of least period m with m ≥ 3 and odd then
f has periodic points with least period n for each odd integer n ≥ m.
2.3 Periodic Points of Continuous Real Functions 33
Proof Let P be a periodic orbit of f with period m. By Lemma 2.3.1 and Remark
2.3.2. f has a fixed point z, a 2-periodic point y and a point v with y < v < z < f (y)
such that f (v) lies in P and f (x) > z and f 2 (x) < x when y < x ≤ v. Define
pm = v. As m is odd and y is a 2-periodic point of f , f m+2 (y) = f (y) > y and
because f 2 ( pm )(= f 2 (v)) is a period-m point of f , f m+2 ( pm ) = f 2 ( pm ) < pm .
So pm+2 = min{x : y ≤ x ≤ pm , f m+2 (x) = x} is well-defined and is an (m + 2)
periodic point of f . Since f m+4 (y) = f (y) > y and f m+4 ( pm+2 ) = f 2 ( pm+2 ) <
pm+2 (and it be noted that f 2 ( pm+2 ) cannot be pm+2 ). So pm+4 = min{x : y ≤ x ≤
pm+2 , f m+4 (x) = x} exists and is a periodic point of f with period (m + 4). Thus
proceeding, we obtain a decreasing sequence of points pm , pm+2 , . . ., pm+2k , . . . with
y < · · · < pm+2k+2 < pm+2k < · · · < pm+2 < pm = v
such that pm+2k is a periodic point of f with period m + 2k (k = 1, 2, . . . ).

Theorem 2.3.4 If f has a periodic point of least period m with m ≥ 3 and odd,
then f has periodic points of all even periods. Further, there exist disjoint closed
subintervals I0 and I1 of I such that f 2 (I0 ) ∩ f 2 (I1 ) ⊇ I0 ∪ I1 .
Proof Let P be an m-orbit of P. By Lemma 2.3.1 and Remark 2.3.2, there is a fixed
point z of f , a 2-periodic point y of f and a point v such that f (v) = b ∈ P,
max{ f 2 (v), y} < v < z z for x in (y, v]. Write g = f 2 and let z 0 = min{t :
v ≤ t ≤ z, g(t) = t}. Then y and z 0 are fixed points of g such that y < v <
m+1
z 0 ≤ z z for y < x < z 0 . If g(x) <
z 0 for min I ≤ x ≤ z 0 , then g([min I, z 0 ]) ⊆ [min I, z 0 ] and this contradicts that
m+1
g 2 (v) = b > z 0 . Hence d = max{x : min I ≤ x ≤ y, g(x) = z 0 } is well defined
and f (x) > z > z 0 > g(x) for all x in (d, z 0 ). Define s = min{g(x) : d ≤ x ≤ z 0 }.
m+1
If s ≥ d, then g([d, z 0 ]) ⊆ [d, z 0 ]. But this contradicts that g 2 (v) = b > z 0 .
So s < d, [s, d] ∪ [d, z 0 ] are non-overlapping closed subintervals and f 2 [s, d] ∩
f 2 [d, z 0 ] ⊇ [s, d] ∪ [d, z 0 ]. Let
g : [d, z 0 ] → [d, z 0 ] be the map defined by g (x) =
max{g(x), d}. Clearly, g is continuous and onto and let t = min{x : d ≤ x ≤ z 0 ,
g(x) = d}. For each n ∈ N, define cn = min{x : d ≤ x ≤ t, g (x) = x}. It is not dif-
ficult to note that d < · · · < c4 < c3 < c2 < c1 ≤ y and that cn generates an n-period
orbit Q n ⊆ (d, z 0 ) of g . Clearly Q n is also an n-period orbit of g = f 2 . Since
x < z 0 ≤ z < f (x) for x in Q n , Q n ∪ f (Q n ) is 2n-period orbit of f . Thus f has
periods of all even orders.
Theorem 2.3.5 (Sharkovsky) Let f : I → I be a continuous map, where I is a
compact interval of real numbers. Then
(1) if f has a periodic point of period m and if m ≺ n (in the Sharkovsky order),
then f has also a periodic point of period n;
(2) for each positive integer n, there exists a continuous map g : I → I that has a
periodic point of period n but no point of period m ≺ n;
(3) there is a continuous map h : I → I having a 2i -periodic point for 0, 1, 2, . . . ,

and has no other periodic point.
Proof If f has j-periodic point with j ≥ 3 and odd, then by Theorem 2.3.3 f has
( j + 2) periodic point and by Theorem 2.3.4, f has a periodic point of period (2.3).
If f has (2. j) periodic point with j ≥ 3, and odd, f 2 has j-periodic point. So by
Theorem 2.3.3, f 2 has ( j + 2) periodic point and so f has either ( j + 2) periodic
point or period 2( j + 2) points. If f has ( j + 2) periodic point, then by Theorem
2.3.4, f has 2( j + 2) periodic point. In any case f has 2( j + 2) periodic point. If f 2
has j-periodic point, by Theorem 2.3.4, f 2 has 2.3 periodic point. So f has (22 .3)
k−1
periodic point. So if f has 2k . j periodic point, j ≥ 3 and odd and if k ≥ 2, then f 2
2k−1
has period 2. j points. So from what we have proved, we see that f has period
2( j + 2) points and period 22 .3 points. It follows that f has period (2k .( j + 2))
points and period (2k+1 .3) points, with j ≥ 3. If f has (2i . j) periodic points, j ≥ 3
i i −i
and odd and if i ≥ 0, then f 2 has j-periodic point. For ≥ i f 2 = ( f 2 )2 has
2
period j points. So by Lemma 2.3.1, f has period 2 points. So f has period 2+1
k−2
points for ≥ i. Finally when f has 2k -periodic points for some k ≥ 2, then f 2
2k−2
has 4 periodic point. Again by Lemma 2.3.1 f has 2 periodic points implying
that f has 2k−1 periodic points. Hence (1) is true.
For proving (2) and (3), without loss of generality, we can assume that I = [0, 1]
and T (x) = 1 − |2x − 1|, a map with a triangular graph having vertices at (0, 0),
( 21 , 1) and (1, 0). Then for each n ∈ N, T n (x) = x has exactly 2n distinct solutions
in I . So T has finitely many n-periodic orbits. Among these let Pn be an orbit
of the least diameter (= max Pn − min Pn ). Define Tn on I by Tn (x) = max Pn , if
T (x) ≥ max Pn , Tn (x) = min Pn , if T (x) ≤ min Pn and Tn (x) = T (x) for min Pn ≤
T (x) ≤ max Pn . Clearly Tx is continuous on I and Tx has exactly one-period n orbit,
i.e. Pn but has no m-periodic orbit for any m ≺ n.
Let Q 3 be any 3-periodic orbit of T of minimal diameter. Then [min Q 3 , max Q 3 ]
contains finitely many 6-periodic orbits of T . If Q 6 is one with smallest diameter,
then [min Q 6 , max Q 6 ] contains finitely many 12-periodic orbits of T . We choose
one, say Q 12 of minimal diameter and continue this process inductively. Define
q0 = sup{min⎧ Q 2i .3 : i ≥ 0} and q1 = inf{max Q 2i .3 : i ≥ 0}. Define T : I → I by
⎪
⎨q0 if T (x) ≤ q0
T (x) = q1 if T (x) ≥ q1 . Clearly T is continuous and has 2i -periodic
⎪
⎩
T (x) if q0 ≤ T (x) ≤ q1
point for i = 0, 1, 2, . . . but has no other periodic point. Thus (2) and (3) are
true.
Remark 2.3.6 Lemma 2.3.1 has interesting consequences. Let x0 ∈ I and n ≥ 2 be a

natural number such that f n (x0 ) < x0 < f (x0 ). Let X = { f k (x0 ) : 0 ≤ k ≤ n − 1}
(a finite set), a = max{x ∈ X : q0 ≤ x < f (x)}, and b ∈ {x ∈ X : a < x ≤ f (a)}
with f (b) < a. From these conditions on a, b, x0 , f (x0 ) and X it is clear that f (b) <
a 1) then f has n-periodic
points.
2.3 Periodic Points of Continuous Real Functions 35
If in addition O f (c) contains both a fixed point z and a point different from z, then
f has periodic points with all even periods. Arguments similar to those in Theorems
2.3.3 and 2.3.4 can be used.
Remark 2.3.7 Sharkovsky’s theorem cannot be generalized to continua (compact
2πi
connected subsets) of the plane. On the unit disc, the map z → ze 3 has 0 as the
only fixed point and all the other points are 3-periodic points. For each n ∈ N, the
2πi
map z → ze n has only one fixed point and the rest of the points are n-periodic
points. No point of fundamental period greater than n exists.
Sharkovsky’s result is definitely and unalterably one-dimensional (See
Ciesielski and Pogoda [8].) Nevertheless, there has been appropriate generaliza-
tion of Sharkovsky’s theorem to general topological spaces and more general maps
than continuous functions. See Schirmer [25].
2.4 Common Fixed Points, Commutativity and Iterates
It is natural to find out if two continuous real functions f, g : I (= [a, b]) → I have
a common fixed point. The maps x → x2 and x → 1 − x on [0, 1] have the only fixed
points 0 and 21 respectively. Since their compositions are 1−x2
and 1 − x2 , they do not
commute. If f, g : I → I have a common fixed point x0 , then x = f (x0 ) = g(x0 ) =
g f (x0 ) = f g(x0 ) and thus f and g commute at least on {x0 }. Ritt [24] showed that if
f and g are polynomials that commute, then they are within certain homeomorphisms
iterates of the same function, both power of x or both must be Chebyshev polynomials
and in both these cases, the commuting polynomials have a common fixed point. So
Dyer conjectured that if f, g : I (= [a, b]) → I are continuous real functions that
commute, then f and g have a common fixed point. However, Boyce [5] and Huneke
[17] had disproved the conjecture independently by constructing counter-examples
to point out that commuting continuous self-maps on a compact real interval may not
have a common fixed point. Isbell [18] first recorded this problem in a more general
form.
This section discusses some results that ensure the existence of common fixed
points of two commuting continuous functions f, g : I → I under suitable additional
assumptions. We recall the following definitions.
Definition 2.4.1 Let F be a family of maps from a topological space X into a
metric space (X, d). It is said to be equicontinuous at x0 ∈ X , if for each > 0,
there exists an open set O in X containing x0 such that for each x ∈ O and f ∈ F ,
d( f (x0 ), f (x)) < . F is said to be equicontinuous on X , if it is equicontinuous at
each x ∈ X .
Definition 2.4.2 If f : X → X is a map, a subset A ⊆ X is said to be f -invariant
or invariant (under f ) if f (A) ⊆ A.
An elementary proposition on invariant subsets of continuous maps on compact
intervals is given below.
Proposition 2.4.3 If f : I = [a, b] → I is a continuous map on the compact inter-

val I of real numbers, then every non-empty closed invariant subset C of I contains
a minimal closed invariant non-empty subset C .
Proof Let C be a non-empty closed invariant subset of I and C be the family of all
closed invariant subsets of C. Clearly C ⊂ C . Let F be a chain of sets in C . Since F
is a subfamily of non-empty closed subsets of C which are indeed compact subsets of
I , F0 = ∩{F : F ∈ F } is non-empty and compact. Further f (F0 ) ⊆ f (F) ⊆ F for
all F ∈ F and hence f (F0 ) ⊆ ∩{F : F ∈ F } = F0 . Thus, F0 is an invariant closed
subset which is contained in each F ∈ F . Thus F0 is the least element of F in C.
So by Zorn’s Lemma, C has a minimal element C0 , which is a non-empty minimal
closed invariant subset of C.
Remark 2.4.4 Indeed if f : X → X is a continuous map of a compact connected

T2 space, then every non-empty closed invariant subset A of X contains a minimal
closed invariant subset of A.
Proposition 2.4.5 If Y is a minimal non-empty closed invariant subset of I a

compact interval of R, then for y ∈ Y , Y = O f (y) where O f (y) = { f n (y) : n =
0, 1, 2, . . . } is the orbit of y, under f .
Proof If y ∈ Y , then O f (y) ⊆ Y as f (Y ) ⊆ Y . Since Y is closed, O f (y) ⊆ Y . Now

by the continuity of f , O f (y) ⊆ Y . By the minimality of Y , Y ⊆ O f (y). So Y =
O f (y).
Theorem 2.4.6 (Schwartz [26]) Every non-void closed invariant minimal subset of
the continuous function f : I → I is contained in the closure of P f , where P f =
{x ∈ I : f k (x) = x for some k ∈ N}, the set of periodic points of f .
Proof Let Y be a non-empty minimal closed invariant subset of I . If Y is the orbit

of a periodic point, obviously it is finite and closed and the conclusion is true.
Suppose Y is not a periodic orbit. Let c = inf Y . As Y is closed, c ∈ Y . As Y is
minimal closed invariant subset, by Proposition 2.4.5, Y = O f (c). So given > 0,
we can find k ∈ N with |y − f k (c)| < 2 . Also we can find M, N ∈ N such that
c < f N +M (c) < f N (c) < c + , as c = inf Y = O f (c). As Y is minimal and is
not a periodic orbit, f M (c) > c. Thus, the continuous map f M maps [c, f N (c)] into
itself and so has a fixed point d. Since c < f M (c) < f M+N (c), d ∈ (c, f N (c)). Thus
f M (d) = d is a periodic point and |c − d| < f N (c) − c < .
As f k is continuous at c, for > 0 we can find δ > 0 with > δ such that
| f (x) − f k (c)| < 2 for |x − c| < δ. Since |y − f k (d)| ≤ |y − f k (c)| + | f k (d) −
k
f k (c)|, choosing = δ, we see that |y − f k (d)| < . As f M (d) = d, it is clear

that z = f k (d) is a periodic point of f which is within (> 0) distance from y. So
Y ⊆ P( f ).
Corollary 2.4.7 If Y is a non-empty minimal closed invariant subset of f then Y is

nowhere dense.
2.4 Common Fixed Points, Commutativity and Iterates 37
Proof Let x0 be an interior point of Y . Then for some > 0, [x0 − , x0 + ] ⊆ Y .

If [x0 − , x0 + ] contains a periodic point y of Y , then O f (y) is finite and is closed.
Since y ∈ Y , Y = O f (y) = O f (y) and this contradicts that Y is uncountable (since
it has an interior point). So [x0 − , x0 + ] has no periodic point. As x0 ∈ Y , by
Theorem 2.4.6, [x0 − , x0 + ] must contain a periodic point, contradicting the
preceding assertion. Hence Y is nowhere dense.
Theorem 2.4.8 (Cano [6]) Let F = F1 ∪ F2 be a collection of continuous func-

tions mapping a compact interval I = [a, b] ⊆ R into itself, satisfying the following
assumptions:
(i) for f ∈ F1 , F f the set of fixed points of f in I is a compact interval [a f , b f ];
(ii) for f ∈ F2 , every periodic point of f is a fixed point of f ;
(iii) for f, g ∈ F , f (g(x)) = g( f (x)) for all x ∈ I ( f and g commute).
If h : I → I is a continuous function that commutes with each f ∈ F , then F ∪ {h}
has a common fixed point in I .
Proof Let C1 ∪ {h} be any finite subset of F ∪ {h} of the form { f 1 , . . . , f n } ∪ {h} ∪
{g1 , . . . , gm } where f i , i = 1, 2, . . . , n ∈ F1 and {g1 , . . . , gm } ⊆ F2 . Since F fi is
n
a compact interval and f i ’s commute F fi is a non-empty compact interval, say
i=1
[c, d]. As h commutes with each f i ∈ C1 , h maps [c, d] into itself and so has a
fixed point z ∈ [c, d]. Now g1n (z) has a limit point z 1 in Pg1 by Theorem 2.4.6. As
Pg1 = Fg1 (by hypothesis (ii), and Fg1 is closed, Pg1 = Pg1 . Similarly g2n (z 1 ) has a limit
point z 2 in Pg2 = Fg2 = Pg2 and as Fg2 is closed z 2 ∈ Fg2 . Thus z 1 , z 2 ∈ [c, a]. Thus
proceeding, we see that {g nj (z j−1 )} has a limit point z j in Pg j for j = 2, . . . , m which
is fixed for f 1 , . . . , f n , h, g1 , . . . , gm . So ∩F f = φ for all f ∈ C1 ∪ {h}. It is also
easily seen that for any finite subset C2 of F1 , F f = φ as also F f = φ for
f ∈C2 f ∈C3
any finite subset C3 of F2 . Thus, the family of closed subsets {F f : f ∈ F ∪ {h}} of
[a, b] has finite intersection property and hence ∩{Fr : f ∈ F ∪ {h}} is non-empty,
in view of the compactness of [a, b].
Theorem 2.4.9 (Cano [6]) Let f : I (= [a, b]) → I be a continuous function such
that { f n : n ∈ N} is an equicontinuous family at each x ∈ I . Then
(1) F p , the fixed point set of f is a compact subinterval of I ;
(2) if F f is a non-degenerate interval, then F f = P f (P f being the set of periodic
points of f ).
Proof As f : I → I is continuous, F f = φ. If F f is a singleton, the theorem is true.

Suppose a0 , b0 ∈ F f and a0 < b0 . Assume that for no x ∈ (a0 , b0 ), x0 = f (x0 ).
Then for all x ∈ (a0 , b0 ), f (x) > x or f (x) < x. Assume that f (x) > x for all
x ∈ (a0 , b0 ).
Case (i) If f (x) < b) for all x ∈ (a0 , b0 ) then f n (x) ∈ (a0 , b0 ) for all n ∈ N and
f n (x) < f n+1 (x) < b0 and so it converges to a fixed point of f , which cannot
be in (a0 , b0 ) and hence has to be b0 . So given > 0, by the equicontinuity of
{ f n } at a0 , there exists δ > 0 such that |a0 − x0 | < δ such that for |a0 − x0 | < δ,
| f n (a0 ) − f n (x0 )| < . Since f n (a0 ) = a0 , for all n, this contradicts that f n (x0 )
converges to b0 .
Case (ii) Suppose for some x0 ∈ (a0 , b0 ), f (x0 ) ≥ b0 . Then there is a least num-
ber z in (a0 , b0 ) with f (z) ≥ b0 . In fact f (z) = b0 . Otherwise, there exists z < z
with f (z ) ≥ b0 by the continuity of f and this contradicts the definition of z. Thus
proceeding, we can find a non-increasing sequence (xn ) in (a0 , z] such that (xn ) con-
verges to a0 , x1 = z and f (xn ) = xn−1 , n = 2, 3, . . . . Since f n (xn ) = f n−1 (xn−1 ) =
· · · f (x1 ) = f (z) = b0 for all n, f n cannot be equicontinuous at a0 . (Note that as
(xn ) is non-increasing in (a0 , z) it converges to a number z ≥ a0 . z > a0 is a con-
tradiction as z = f (z ) and by assumption f has no fixed point in (a0 , b0 ).)
Suppose f (x) < x for all x ∈ (a0 , b0 ). We consider
Case (i) Suppose f (x) > a0 for all x ∈ (a0 , b0 ). Then for all x ∈ (a0 , b0 ), f n (x) >
f n+1 (x), n ∈ N and ( f n (x)) as in Case (i) converges to a0 . However the family of f
iterates cannot be equicontinuous at b0 .
Case (ii) If for some x ∈ (a0 , b0 ), f (x) ≤ a0 . Then there is a greatest element z
in (a0 , b0 ) with f (z ) ≤ a0 . In fact f (z ) = a0 . By this process, a non-decreasing
sequence (yn ) can be chosen in (z , b0 ] with y1 = z , f (yn ) = yn−1 , n = 2, 3, . . . .
So f n (yn ) = f (z ) = a0 . If (yn ) converges to w, then f (yn ) (= yn−1 ) converges
to f (w) and so w = f (w). As w ∈ / (a0 , b0 ), (yn ) converges to b0 . Since f n (yn ) =
f n−1
(yn−1 ) · · · = f (z ) = a0 . As yn converges to b0 , there is a contradiction to the
equicontinuity of f n at b0 .
Thus we have shown that F f is a non-void compact interval. If F f is non-
degenerate let F f = [a0 , b0 ] where a0 < b0 . Let f n (x) = x for some n and x ∈
[a, a0 ). (If x ∈ (b0 , b], then a similar argument can be provided). Since f n has
a fixed point and its iterates are equicontinuous at each point, f n (y) = y for
all y ∈ [x, a0 ] by what has been proved in (i) so far. Since f (y) > y for all
y ∈ [a, a0 ) and f (a0 ) = a0 , we can choose y from (x, a0 ) close to a0 , such that
a0 − k1 < y < f (y) · · · < f n−1 (y) < a and this implies f n (y) > y, a contradic-
tion. So a0 + k1 > f (y) > a0 > y > a0 − k1 . Then f (y) is a fixed point for f .
So f (y) = f 2 (y) and f n (y) = f n−2 ( f 2 (y)) = f n−1 (y). Thus proceeding, y =
f n (y) = f n−1 (b) · · · = f (y) contradicting f (y) > a > y. Thus if F f = [a0 , b0 ],
[a0 , a) has no periodic point. Similarly (b0 , b] has no periodic point.
This leads to the following.
Theorem 2.4.10 (Jachymski [19]) Let g : I → I be a continuous map and I , a

compact interval [a, b] of real numbers. Then the following are equivalent:
(i) Fg the set of fixed points of g is a compact subinterval of I ;
(ii) either Fg is a singleton or the family {g n : n ∈ N} of iterates is equicontinuous

on Fg ;
(iii) g has a common fixed point with each continuous map f : I → I that commutes
with g on Fg .
Proof (i) =⇒ (ii). Suppose Fg is not a singleton and is [a1 , b1 ] where a1 <
b1 . Since for a1 < x < b1 , g n (x) = x for all n ∈ N, the continuity of g at x
implies that given > 0 with b − a > , there is a δ() > 0 such that (x − δ, x +
δ) ⊆ (a1 , b1 ) and |g(x) − g(x )| < for x ∈ (x − δ, x + δ). So |g n (x) − g n (x )| =
|g(x) − g(x )| < for x ∈ (x − δ, x + δ), proving the equicontinuity of {g n } on
(a1 , b1 ). We now show that {g n } is equicontinuous at a1 . Since g is continuous at
a1 , there exists δ() > 0 with > δ() for a given > 0 such that for a1 − δ < x <
a1 + δ, |g(x) − g(a1 )| = |g(x) − a1 | < . We now show by the principle of finite
induction that a1 − < g n (x) < a1 + for all x ∈ (a1 − δ, a1 + δ) for all n ∈ N.
Clearly, the inequality is true for n = 1. Suppose it is true for n = 1, 2, . . . , k.
Let x ∈ (a − δ, a). If a1 ≤ g k (x) < a1 + , then g k (x) ∈ Fg and so |g k+1 (x) −
a1 | = |g k+1 (x) − g k+1 (a1 )| = |g k (x) − g k (a1 )| = |g k (x) − a1 | < . If g k (x) < a1 ,
then g i (x) < a1 for i = 1, 2, . . . , k. Otherwise by induction hypothesis for some i,
1 ≤ i ≤ k and a1 ≤ g i (a) < a1 + or g i (x) ∈ Fg and so g k (x) ∈ Fg or g k (x) ≥ a1 ,
a contradiction. Since Fg = [a1 , b1 ], g(x) > x for x ∈ [a, a1 ). So g i (x) > g i−1 (x)
for i = 1, 2, . . . , k, implying that g k (x) > g k−1 (x) > · · · > x. As a1 − δ < x and
g k (x) < a1 , it follows that g k (x) ∈ (a1 − δ, a1 ). So |g(g k (x)) − g(a1 )| = |g k+1 (x) −
a1 | < . For x ∈ (a1 , a1 + δ) ⊆ [a1 , b1 ], |g n (x) − g n (a1 )| = |x − a1 | < . Thus g n
is equicontinuous at a1 . By a similar reasoning, (g n ) is equicontinuous at b1 .
(ii) =⇒ (i). This follows from the proof of Theorem 2.4.9 (i). In fact to prove
(i) of Theorem 2.4.9, it suffices to assume that { f n } is equicontinuous on F f .
(i) =⇒ (iii). If f commutes with g on Fg then Fg is invariant under f . Since Fg

is a compact interval by (i), f has a fixed point in Fg which is a common fixed point
of f and g.
(iii) =⇒ (i). If Fg is not an interval, then there exists a1 , b1 ∈ Fg such that

(a1 , b1 ) ∩ Fg = φ. Define f : [a, b] → [a1 , b1 ] by
⎧
⎪
⎨b1 for x ∈ [a, a1 ]
f (x) = b1 + a1 − x for x ∈ (a1 , b1 ]
⎪
⎩
a1 for x ∈ (b1 , b]
f is continuous on I . Let x ∈ Fg . Then x ∈ [a, a1 ] or [b1 , b]. If x ∈ [a, a1 ], then

f g(x) = f (x) = b1 = g f (x) = g(b1 ). If x ∈ [b1 , b], then f g(x) = f (x) = a1 =
g(a1 ) = g f (x). Thus, f and g commute on Fg but F f ∩ Fg = φ. Hence the theo-
rem.
Example 2.4.11 The continuous map g : [0, 1] → [0, 1] defined by g(x) = 1 on

[0, 41 ], 23 − 2x for x ∈ ( 41 , 43 ] and 0 on ( 43 , 1] has the only fixed point x = 21 . But
g n ( 21 + δ) = (−2)n δ + 21 for 0 < δ < 14 , as long as 2n δ < 41 or δ < 2n+2 1
. Suppose
g is equicontinuous at x = 2 . Then for = 4 , there exists δ > 0 such that |g n (1 +
1 1
δ) − g n ( 21 )| < for all n. Since g( 21 ) = 21 and choosing least n 0 such that 2n 0 δ > 14 ,
it follows that g n ( 21 + δ) = 0 for all n ≥ n 0 and |g n ( 21 + δ) − g n ( 21 )| = |0 − 21 | =
| 21 | < 41 , a contradiction. So (g n ) is not equicontinuous.
If f : [0, 1] → [0, 1] commutes with g at 21 , then f g( 21 ) = g( f ( 21 )) = f ( 21 ) (as
g( 2 ) = 21 ). Since f ( 21 ) is a fixed point of g and g has the unique fixed point 21 , f ( 21 ) =
1
1
2
. Thus, f and g have a common fixed point, even though {g n } is not equicontinuous.
This example points out that the hypothesis Fg is a singleton cannot be dropped
in Theorem 2.4.10.
The next theorem on the convergence of iterates, due to Coven and Hedlund [12],
was also obtained independently by Chu and Moyer [7].
Theorem 2.4.12 If f : I = [a, b] → I is continuous and P f = F f , then for each

x ∈ I , there exists p ∈ F f such that { f n (x)} converges to p.
Proof If { f n (x)} converges to p, it follows from the continuity of f , that p ∈ F f .

Thus it suffices to prove the convergence of { f n (x)} for each x ∈ I . If f n (x) ∈ P f
for some n ≥ 0, the conclusion is obvious. Suppose that f n (x) is not a periodic point
of f for any n ≥ 0. Let Cn be the component of N P f , the set of non-periodic points
of f in I containing f n (x). Let ξn = +1 if f is completely positive on Cn (i.e.)
( f (x) > x on Cn ) and ξn = −1 if f is totally negative on C (i.e. f (x) < x on Cn ).
Since f is continuous and Cn is connected, f (x) − x cannot take both positive and
negative values on Cn as Cn has no fixed point.
If for some N ≥ 0, ξn = +1 for n ≥ N , then f N (x) < f N +1 (x) and so f n (a)
converges. Similarly if ξn = −1 for all n ≥ N1 , then { f n (x)} converges.
Suppose +1 and −1 appear infinitely many times in the sequence (ξn ), n ≥ 0. Let
A = {n ≥ 0 : ξn = +1} = { p1 < p2 < · · · } and B = {n ≥ 0 : ξn = −1} = {m 1 <
m 2 < · · · }. { f pi (x)} is increasing while { f m i (x)} is decreasing in I and hence these
subsequences of { f n (x)} converge to p and q respectively in I . Now we can find a
subsequence ki ∈ A such that ki + 1 ∈ B. Since { f ki (x)} converges to p { f ki +1 (x)}
converges to q and f is continuous f ( p) = q. By a similar reasoning we find that
f (q) = p. Thus f 2 ( p) = f (q) = p and f 2 (q) = f ( p) = q. Thus p ∈ P f = F f .
So p = f ( p) = q. Hence the theorem.
Corollary 2.4.13 If f : I = [a, b] → I is continuous and the set of least periods

or periodic points is finite, then for each x ∈ [a, b], there exists p ∈ P f such that
| f n (x) − p| converges to zero as n → ∞.
Proof Let N be the least common period of the periodic points. Apply Theorem
2.4.12 to f N and that P f N = F f N . (It is to be observed that N must be a power of 2,
as can be seen from Sharkovsky’s theorem.)
Our next theorem characterizes functions f : I → I that are continuous and for
which P f = F f .
Theorem 2.4.14 (Jachymski [19]) Let g : I = [a, b] → I be a continuous function.

Then the following are equivalent:
(i) Fg = Pg ;
(ii) {g n : n ∈ N} is pointwise convergent on I ;
(iii) g has a common fixed point with every continuous map f : I → I that com-
mutes with g on F f .
Proof (i) =⇒ (ii) is precisely Theorem 2.4.12.
(ii) =⇒ (iii). Let x ∈ F f . By the commutativity of f and g on F f , F f is g-

invariant. So g n (x) ∈ F f for all n ≥ 1. Since {g n (x)} converges to z ∈ I by (ii) and
F f is closed z ∈ F f and as g is continuous z = g(z). Thus z = f (z) = g(z).
(iii) =⇒ (i). Let C be a non-empty g-invariant closed subset of I . We show that

C ∩ Fg = φ. For such a set, there is a continuous map f : I → I such that F f = C.
If x ∈ F f , then g( f (x)) = g(x) and f (g(x)) = g(x), since C is g-invariant. So f
and g commute on F(g). By assumption (iii) F f ∩ Fg = C ∩ Fg = φ. Let p be a
periodic point of least period M for g. Then C = { p, g( p), . . . , g M−1 ( p)} is closed
and invariant under g. So from what we have shown, C has a fixed point of g. If
for 1 ≤ i ≤ M − 1, g(g i ( p)) = g i ( p), g i ( p) = g (M) ( p) = p, contradicting p is a
periodic point of g with least period M. So i = 0 gives g( p) = p or p is a fixed point
of g. Thus Pg = Fg .
2.5 Common Fixed Points and Full Functions
In this section, an existence theorem on the common fixed points for two commuting
continuous self-maps on a compact real interval, due to Cohen [9] is proved. This
supplements the theorems in Sect. 2.4. Without loss of generality we take I = [0, 1].
We need the following lemmata and definitions.
Lemma 2.5.1 Let f, g : I → I be continuous maps and h : I → J = [c, d] be a

homeomorphism onto J . f and g commute on I and have a common fixed point if
and only if h f h −1 and hgh −1 commute on J and have a common fixed point.
Proof Let h : I → J be a homeomorphism onto J and f, g : I → I be contin-

uous functions. Let h f h −1 : J → J and hgh −1 : J → J be commutative and y0
be a common fixed point. Then y0 = h f h −1 (y0 ) = hgh −1 (y0 . Since h is a home-
omorphism from I onto J , so h −1 is a homeomorphism of J onto I . So h −1 y0 =
h −1 (h f h −1 (y0 )) = h −1 hgh −1 (y0 ). Thus h −1 (y0 ) = f (h −1 (y0 )) = g(h −1 (y0 )) or
x0 = h −1 (y0 ) belongs to I and is a common fixed point for f and g in I . Also
by the commutativity of h f h −1 and hgh −1 we get h f gh −1 = (h f h −1 ) ◦ hgh −1 =

(hgh −1 ) ◦ (h f h −1 ) = hg f h −1 whence f g = gh on I .
If f (g(x)) = g( f (x)) for all x ∈ I and h −1 : J → I is a homeomorphism, for
each y ∈ J , f gh −1 (y) = g f h −1 (y) and so f h −1 hgh −1 y = gh −1 h f h −1 y for y ∈ J .
Premultiplying by h we get for y ∈ J
(h f h −1 )(hgh −1 )y = (hgh −1 )(h f h −1 )y.
Thus h f h −1 and hgh −1 commute. If for x0 ∈ I x0 = f (x0 ) = g(x0 ), then h(x0 ) =

h f (x0 ) = hg(x0 ). But x0 = h −1 (y0 ) for some y0 ∈ J . So y0 = h f h −1
(y0 ) = hg f −1 (y0 ). Thus h f h −1 and hgh −1 have a common fixed point.
Lemma 2.5.2 If f, g : I → I are commuting continuous functions without a com-

mon fixed point, then there are commuting functions mapping I onto I without a
common fixed point.
Proof Let a1 = max{inf f, inf g} and b1 = min{sup f, sup g}. Since f and g com-
I I I I
mute, f [0, 1] ∩ g[0, 1] = φ both f and g map [a1 , b1 ] into itself. Otherwise for
some x ∈ [a1 , b1 ], f (x) > b1 would imply that for some y ∈ [0, 1], g(y) = x and
g( f (y)) = f g(y) = f (x) > b1 . This implies that b1 < min{sup f, sup g}. Similarly
I I
f (x) < a1 for some x ∈ [a1 , b1 ] would imply that there exists y ∈ [0, 1] with g(y) =
x and g( f (y)) = f g(y) = f (x) < a1 . This means that a1 > max{inf I f, inf I g}, a
contradiction. Writing f 1 and g1 as the restrictions of f and g on J1 = [a1 , b1 ] respec-
tively, we can inductively define ai , bi and f i by ai = max{inf Ji−1 f, inf Ji−1 g} and
bi = min{sup Ji−1 f, sup Ji−1 g} where Ji−1 = [ai−1 , bi−1 ], i = 2, 3, . . . , and f i is the
restriction of f i−1 to Ji−1 . Since [ai , bi ], i = 1, 2, . . . , form a nested sequence of
compact subsets of [0, 1], they have a non-void intersection. If this intersection is
a singleton, then f and g have a common fixed point contrary to the assumption.
∞
Hence, [ai , bi ] is a non-degenerate compact interval [a, b] and the restriction f
i=1
and g of f and g respectively map [a, b] onto itself. If h is a homeomorphism of
[a, b] onto I = [0, 1]. Then, the continuous maps h f h −1 and hgh −1 map [0, 1] onto
itself but have no common fixed points by Lemma 2.5.1.
Lemma 2.5.3 If f, g : I → I are commuting continuous functions, so are f and

g f . f and g have a common fixed point if and only if f and g f have a common fixed
point.
Proof f (g f ) = g f ◦ f as f g = g f . If x0 = f (x0 ) = g(x0 ), then x0 = f (x0 ) =

g(x0 ) = g( f (x0 )). If x1 = f (x1 ) = g( f (x1 )), then x1 = f (x1 ) = g(x1 ).
Definition 2.5.4 A continuous function f : I → I is said to be full if there is a

partition P f = {x0 = 0 < x1 < x2 · · · < xn = 1} of I such that f on [xi , xi+1 ] is a
homeomorphism on [0, 1] for each i = 0, 1, . . . , n − 1.
2.5 Common Fixed Points and Full Functions 43
Definition 2.5.5 A partition P f = {0 = x0 < x1 < · · · < xn = 1} is regular if the

length of the subintervals xi+1 − xi is the same for all i = 0, 1, . . . , n − 1. A partition
Pg refines a partition P f uniformly if each subintervals in P f formed by consecutive
partition points of P f is the union of partitioning subintervals of g.
Lemma 2.5.6 If f 1 , g1 are commuting full functions on [0, 1] without a common
fixed point, there are functions f and g with the same properties and additionally
f (0) = g(1) = 0 and f (1) = g(0) = 1, P f , Pg and P f g are regular and Pg refines
P f uniformly.
Proof If f 1 (0) = g1 (0) = 0, then f 1 and g1 have a common fixed point contrary
to the assumption. So essentially two cases arise: (i) f 1 (0) = 0, g1 (0) = 1 and (ii)
f 1 (0) = 1 = g1 (0). In case (i) f 1 (1) = f 1 g1 (0) = g1 f 1 (0) = 1 and so g1 (1) = 0, as
otherwise g1 (1) = 1 would imply that f 1 and g1 have 1 as a common fixed point. In
this case let f 2 = f 1 and g2 = g1 .
For case (ii), f 1 (1) = f 1 g1 (0) = g1 f 1 (0) = g1 (1). So f 1 (1) = g1 (1) = 0 as other-
wise 1 would be a fixed point. In this case let f 2 = f 1 g1 and g2 = g1 , g2 (0) = g(0) =
1, f 2 (1) = f 1 g1 (1) = f 1 (0) = 1 and g2 (1) = g1 (1) = 0. In either case let f 3 = f 2
and g3 = g2 f 2 . Clearly Pg3 refines P f2 uniformly. Let h be any order preserving
homeomorphism on [0, 1] taking P f3 g3 into the corresponding regular partition of [0,
1]. Define f = h f 3 h −1 and g = hg3 h −1 . As f 3 and g3 have no common fixed point,
by Lemma 2.5.1 f and g do not have a common fixed point. Also P f , Pg , P f g are
regular and as Pg3 refines P f3 uniformly. Pg refines P f uniformly.
Theorem 2.5.7 (Cohen) Commuting continuous full functions mapping [0, 1] onto
[0, 1] have a common fixed point.
Proof Let f 1 , g1 : I → I be two commuting full functions without a common fixed
point. So using Lemma 2.5.6, we can find commuting full functions f 1 , g1 map-
ping [0, 1] onto itself such that f (0) = g(1) = 0, f (1) = g(0) = 1, P f , Pg and
P f g regular partitions with Pg refining P f uniformly. Let P f = {0, n1 , n2 , . . . , 1} and
Pg = {0, m1 , m2 , . . . , 1} and P f g = {0, mn
1
, mn
2
, . . . , 1} where m and n are odd. Let f i
and gi be restrictions of f to [ n , n ] and g to [ i−1
i−1 i
m
, mi ], respectively. Let r = n+1
2
and s = 2 . Suppose r is odd and s is even. If D( f i , g j ) is the domain of f i g j for
m+1
each i and j then it is a subinterval of P f g . In particular

r −1 r −1 1
D(g1 fr ) = , +
n n mn

r −1 1 r −1 2
D(g2 fr ) = + , + ...,
n mn n mn

r −1 s−1 r −1 s
D(gs fr ) = + , +
n mn n mn

mn − 1 mn + 1
= ,
2mn 2mn
Similarly

s−1 s−1 1
D( f 1 gs ) = , +
m m mn

s−1 1 s−1 2
D( f 2 gs ) = + , + ...,
m mn m mn

s−1 r −1 s−1 r
D( fr gs ) = + , +
m mn m mn

mn − 1 mn + 1
= ,
2mn 2mn
Thus D( fr gs ) = D(gs fr ). Since gs is continuous and onto [0, 1], its graph must
intersect the diagonal of I × I and gs has a fixed point z 1 . As D(gs ) ⊆ D( f 0 ),
z 1 ∈ D( fr ) and thus z 1 ∈ D( fr gs ) = D(gs fr ). So gs fr (z 1 ) = fr gs (z 1 ) = fr (z 1 ) and
z 2 = fr (z 1 ) is a fixed point of gs . Thus proceeding, we get a sequence z p of fixed
points of gs with z p+1 = fr (z p ). Since fr is monotone the sequence z p converges to
z 1 a fixed point of both f and g. The case when r is even and s is odd can be handled
similarly.
Remark 2.5.8 One can show that f is full if and only if f maps [0, 1] onto [0, 1]
and is an open map. For related work, Baxter and Joichi [3] may be referred.
2.6 Common Fixed Points of Commuting Analytic

Functions
We prove a theorem of Shields [28] on the common fixed points of analytic functions
in this section. We denote by G, a non-void bounded open connected set in the
complex plane. Let FG be the family of all analytic functions mapping G into itself.
Clearly FG is a semigroup under composition of mappings. We can consider H (G)
the linear space of all functions analytic on G and continuous on G, with the topology
of uniform convergence on compact subsets of G. This topology is a metric topology
and indeed it arises from a complete metric and so FG will inherit this metric topology.
The following lemma implies that FG is a topological semigroup (i.e. the composition
map is a continuous function from G × G into G).
Lemma 2.6.1 Let f n , gn ∈ FG and f n → f , gn → g in the topology of uniform con-
vergence on compact subsets of G. Then f n (gn ) → f (g) and so FG is a topological
semigroup.
Proof Let K be a compact subset of G and let U be an open set containing g(K )
with U compact and lying in G. Since gn → g uniformly on K , gn (K ) ⊂ U for all
n ≥ n 0 for some n 0 ∈ N. Now for all n
| f (g(z)) − f n (gn (z))| ≤ | f (g(z)) − f (gn (z))| + | f (gn (z)) − f n gn (z)|
Since g(z), gn (z) ∈ U for z ∈ K for all n ≥ n 0 and f is uniformly continuous

on the compact set U and f n → f uniformly on U , the above inequality implies
2.6 Common Fixed Points of Commuting Analytic Functions 45
that f (gn (z)) → f g(z) and | f n (gn (z)) − f (gn (z))| ≤ sup | f n (w) − f (w)| → 0 as
w∈U
n → ∞. Hence ( f n gn ) converges uniformly on K to f g. Thus FG is a topological
semigroup.
A few facts from the theory of topological semigroups will be needed in the
sequel. For proofs and other details Numakura [23], Wallace [31] and Ellis [15] may
be consulted.
Definition 2.6.2 Let (S, ·) be a semigroup. An element e of S is called an idempotent
if e.e = e2 = e. An element 0 is termed zero if 0.x = 0 for all x ∈ S. 1 is called an
identity of S if 1.x = x = x.1 for all x ∈ S. In a semigroup S if ax = ay (xa = ya)
implies x = y for all a, x, y in S then S is called a semigroup satisfying the left
(right) cancellation law. If S satisfies both the left and right cancellation laws, it is
called a semigroup satisfying cancellation law.
The following is a basic result in the theory of topological semigroups and the
proof is essentially from Ellis [15].
Lemma 2.6.3 Let S be a compact Hausdorff topological semigroup. Then S has an
idempotent element.
Proof Let F be the family of all compact subsets K of S such that K 2 ⊆ K . F = φ,
as S ∈ F . F is partially ordered by set inclusion. As every chain in F has a lower
bound F has a minimal element A in F . If r ∈ A, then r A is a non-void compact
subset of S as r A is the image of the compact set A under the continuous map
x → r.x. So r A ∈ F and r A ⊆ A. Since A is minimal r A = A. So there exists
p ∈ A such that r p = r . Define L = {a ∈ A : ra = r }. Clearly p ∈ L and L is a
compact subset of A. Let 1 , 2 ∈ L. Then r 1 2 = r 2 = r and hence 1 ◦ 2 ∈ L.
So L 2 ⊆ L. Hence L ∈ F . As L ⊆ A and A is minimal L = A. Since r ∈ A = L,
r 2 = r from the definition of L. Thus S has an idempotent element.
We skip the proof of the following.
Lemma 2.6.4 Let S be a compact T2 topological semigroup which is commutative.
For x ∈ S and (x) = cl{x, x 2 , . . . , }, we have
(i) (x) contains exactly one idempotent;
(ii) if e is an identity for (x), then (x) is a group and x has an inverse in (x);
(iii) if e is a zero for (x), then xn → e.
The following lemma makes use of the basic properties of analytic functions.
Lemma 2.6.5 If the analytic function e ∈ FG is idempotent, then e(z) ≡ z on e(z)
is constant for all z ∈ G.
Proof If e(z) is constant for all z ∈ G, clearly it is an idempotent. Suppose e is a
non-constant analytic function on G, then f is an open mapping. So G 1 = e(G) is an
open set. Since e2 (z) = e(z), e(z) = z on G 1 . As G 1 is uncountable, and the analytic
functions, viz. identity function and e coincide on G 1 , e(z) must be z at each z
in G.
We also recall some classical results from complex analysis (see Conway [11]
and Ahlfohrs [1].
Theorem 2.6.6 (Montel) Let H (G) be the linear space of analytic functions on the
open region G. A family F in H (G) is normal in the sense that every sequence in
F has a convergent subsequence if and only if F is locally bounded in H (G) (i.e.
for each compact subset K of G, there is a positive constant Mk with | f (z)| ≤ Mk
for all f ∈ F and z ∈ K ).
Theorem 2.6.7 (Hurwitz) Let A(G) be the linear space of all analytic functions
with the topology of uniform convergence on compact subsets of G. If ( f n ) converges
to f in H (G) and f n never vanishes on G for each n, then f ≡ 0 or f is non-zero
throughout G.
Lemma 2.6.8 Let D be the open unit disc in the complex plane C and f : D → D be
a bilinear (Mobius) transformation of D onto D. Then there arise three possibilities:
(i) f (z) = z on D;
(ii) f has exactly one fixed point in the closed unit disc;
(iii) f has two distinct fixed points in the unit circle and the iterates of f converge
to one of these fixed points.
(z−a)
Proof The general form of such a bilinear transformation is f (z) = α (1−a)z where
|α| = 1, |a| < 1.
If f is not the identity function the fixed points z = f (z) are given by
az 2 − (1 − α)z − αz = 0
As this equation is invariant under z → 1z , the fixed points of f (z) are inverses of
each other with respect to the unit circle. So there is a fixed point inside and another
outside the circle or there is a ‘double fixed point’ or two distinct fixed points on the
unit circle.
Lemma 2.6.9 Let f ∈ FG , be the subset of H (G) containing all analytic functions
mapping G into itself. Suppose f is not a homeomorphism of G onto itself. Then
there is a point z 0 in G and a subsequence { f ni } of f -iterates such that f ni (z) → z 0
uniformly on compact subsets of G.
Proof Write ( f ) = cl{ f n } in H (G). If ( f ) ⊆ FG , then ( f ) is a compact semi-
group under composition of functions and contains an idempotent element e(z) by
Lemma 2.6.3.
By Lemma 2.6.5 e(z) ≡ z for all z ∈ G or is a constant z 0 for all z ∈ G. If the
identity map belongs to ( f ), then by Lemma 2.6.4, ( f ) is a group and f ∈ ( f ) ⊆
FG would be invertible in F(G) contradicting that f is not a homeomorphism. Hence
e(z) ≡ z 0 , for all z ∈ G and is thus a zero for ( f ). So again by Lemma 2.6.4 f n (z)
converges to z 0 in the topology of FG .
Suppose g ∈ ( f ) does not belong to FG . Since f n (G) ⊆ G, g(G) ⊆ G. As
g∈/ FG , there is a point z ∈ G with g(z ) = z 0 ∈ / G. We claim that g(z) ≡ z 0 .
2.6 Common Fixed Points of Commuting Analytic Functions 47
As g ∈ ( f ), we can find f n k , a subsequence of f iterates converging to g in

H (G). Now f n k (z) − z 0 never vanishes in G as z 0 ∈ G and converges to g(z) − z 0 .
So by Lemma 2.6.7 (Hurwitz Theorem), g(z) − z 0 is identically zero in G or never
vanishes in G. But already for z = z ∈ G, g(z ) − z 0 = 0. So g(z) ≡ z 0 for all z ∈ G.
Lemma 2.6.10 Let f ∈ FG and suppose f is not a homeomorphism of G onto itself.

Let z 0 be the element of G such that f ni converges to z 0 in H (G). Then z 0 is a common
fixed point for all continuous g on G that map G into itself and commute with f .
Proof By Lemma 2.6.9, there exists z ∈ G with lim f ni (z) = z 0 in FG . For g ∈ C(G),
g(z 0 ) = g(lim f ni (z)) = lim f ni (g(z)) = z 0 .
The following remarks are relevant.
Remark 2.6.11 If f is a bilinear map of the open unit disc D onto itself with two
distinct fixed points on the boundary, consider p a bilinear map, mapping D onto
the upper half-plane and taking these fixed points into 0 and ∞. For g = p f p −1 ,
0 and ∞ are fixed points of g and g maps the upper half-plane onto itself. Hence
g is a dilatation and is of the form g(z) = az, a > 0 and a = 1 as f (z) ≡ z. So
g n (z) = a n z tends to zero or to ∞. Thus the iterates of f converge to one of the fixed
points of f .
Remark 2.6.12 Wolff [32] and Denjoy [13] have shown independently in 1926 that
if f is analytic in D and f (D) ⊆ D, then either f is a bilinear map of D onto itself
with exactly one fixed point or f n converges to a constant C ∈ D.
We are now in a position to prove a theorem of Shields [28] on the fixed points
of commuting family of analytic functions on D.
Theorem 2.6.13 (Shields [28]) Let F be a commuting family of continuous functions

on D which are analytic in D. Then there is a common fixed point z 0 for all functions
in F.
Proof If F contains a constant function then that constant is the common fixed point.
Suppose it contains only non-constant continuous functions on D which are analytic
in D. So by the Maximum Modulus Theorem f (D) ⊆ D for each f ∈ F. Suppose
not all functions of F are bilinear maps of D onto D. So there exists f , different
from the identity map in F. Then Lemma 2.6.10 can be invoked to conclude that
there is a common fixed point for each f ∈ F. On the other hand if all the members
of F are bilinear, then if one of them has just one fixed point, then it is a common
fixed point for all. In case these have two fixed points then by Remark 2.6.11, the
iterates converge to one of the two fixed points and so invoking Lemma 2.6.10, we
conclude that for each f in F there is a common fixed point.
Remark 2.6.14 Theorem 2.6.13 due to Shields has been generalized to Hilbert spaces
by Suffridge [29].
2.7 Fixed Points of Meromorphic Functions
In this section, an interesting theorem on the fixed points of meromorphic functions,

due to Bergweiler [4] is detailed. Bergweiler’s short proof is elementary, though it
invokes Picard’s theorem. We recall
Theorem 2.7.1 (Picard (see Conway [11])) Suppose an analytic function f has an
essential singularity at a. Then in each neighbourhood of a, f assumes each complex
number, with one possible exception, infinitely many times.
Corollary 2.7.2 An entire function which is not a polynomial assumes every complex
number, with one exception infinitely many times.
In response to a question raised by Gross [16], Bergweiler [4] proved the following.
Theorem 2.7.3 (Bergweiler [4]) Let f be a meromorphic function that has at least
two different poles and let g be a transcendental entire function. Then the composite
function f ◦ g has infinitely many fixed points.
The theorem above makes use of the following lemmas.
Lemma 2.7.4 Let f be a meromorphic function and z 0 be a pole of order p. Then

there is a function h, defined and analytic in a neighbourhood of 0 such that h(0) = 0
and f (h(z) + z 0 ) = z − p for z = 0.
Proof The function k defined as k(z)− p = f (z + z 0 ) is analytic in a neighbourhood

of 0 and k (0) = 0. So k(z) is invertible in a neighbourhood of 0 and this inverse h(z)
is analytic in a neighbourhood of 0. Now k(0) = 0. So h(0) = 0 and f (h(z) + z 0 ) =
z − p for z = 0.
Lemma 2.7.5 Let f and g be meromorphic functions. Then f ◦ g has infinitely many
fixed points if and only if g ◦ f does.
Proof If x0 = f g(x0 ), then gx0 = g f (g(x0 )) so that g(x0 ) is a fixed point of g f .

If x0 = f g(x0 ) and x1 = f g(x1 ), then g(x0 ) = g(x1 ) would imply that f g(x0 ) =
f g(x1 ) so that x0 = x1 . Thus g maps the set of fixed points of f ◦ g injectively into
the set of fixed points of g ◦ f . Indeed if x ∗ is a fixed point of g ◦ f , then f (x ∗ ) is
a fixed point of f ◦ g. Similarly f maps the set of fixed points of g ◦ f injectively
into the set of fixed points of f ◦ g. Thus the sets of fixed points of f ◦ g and g ◦ f
have the same cardinality. (Indeed g maps the set of fixed points of f ◦ g bijectively
onto the set of fixed points of g ◦ f ).
Now we provide the proof of Theorem 2.7.3.
Proof Let z 1 and z 2 be poles of f of order p1 and p2 . Using Lemma 2.7.4 choose the
functions h j for j ∈ {1, 2}. Let k1 (z) = h 1 (z p2 ) + z 1 and k2 (z) = h 2 (z p1 ) + z 2 . Now
f (k1 (z)) = f (k2 (z)) = z − p1 p2 for z = 0 in a neighbourhood of 0. Define u(z) =
2.7 Fixed Points of Meromorphic Functions 49
g(z − p1 p2 ). Then 0 is an essential singularity of u and in a punctured neighbourhood

of 0, u(z) = g( f k1 (z)) = g f (k2 (z)).
If f ◦ g has only finitely many fixed points, then so has g ◦ f only finitely many
fixed points by Lemma 2.7.5. So u(z) = k j (z) for j = 1, 2 in a punctured neigh-
bourhood of 0, since k1 (0) = z 1 = z 2 = k2 (0). Define
u(z) − k1 (z)
v(z) = .
k2 (z) − k1 (z)
0 is an essential singularity for u and v does not take the values 0, 1 and ∞ in a
punctured neighbourhood of 0. This contradicts Picard’s Theorem 2.7.1. Hence the
theorem.
Remark 2.7.6 It can be similarly shown that if f and g are transcendental meromor-
phic functions and if either f or g has at least three poles, then f ◦ g has infinitely
many fixed points.
References
1. Ahlfohrs, L.V.: Complex Analysis an Introduction to the Theory of Analytic Functions of One
Complex Variable, 3rd edn. McGraw-Hill Book Co., New York (1978)
2. Bailey, D.F.: Krasnoselski’s theorem on the real line. Am. Math. Mon. 81, 506–507 (1974)
3. Baxter, G., Joichi, J.T.: On functions that commute with full functions. Nieuw Arch. Wiskd.
3(XII), 12–18 (1964)
4. Bergweiler, W.: On the existence of fix points of composite meromorphic functions. Proc. Am.
Math. Soc. 114, 879–880 (1992)
5. Boyce, W.M.: Commuting functions with no common fixed point. Trans. Am. Math. Soc. 137,
77–92 (1969)
6. Cano, J.: Fixed points for a class of commuting mappings on an interval. Proc. Am. Math. Soc.
86, 336–338 (1982)
7. Chu, S.C., Moyer, R.D.: On continuous functions commuting functions and fixed points. Fun-
dam. Math. 59, 91–95 (1966)
8. Ciesielski, K., Pogoda, Z.: On ordering the natural numbers or the Sharkovski theorem. Am.
Math. Mon. 115, 159–165 (2008)
9. Cohen, H.: On fixed points of commuting function. Proc. Am. Math. Soc. 15, 293–296 (1964)
10. Cohen, H., Hachigian, J.: On iterates of of continuous functions on a unit ball. Proc. Am. Math.
Soc. 408–411 (1967)
11. Conway, J.B.: Functions of One Complex Variable, Springer International Student Edition.
Authorized reprint of the original edition published by Springer, New York, Narosa Publishing
House Reprint (2nd edn.) ninth reprint (1990)
12. Coven, E.M., Hedlund, G.A.: Continuous maps of the interval whose periodic points form a
closed set. Proc. Am. Math. Soc. 79, 127–133 (1980)
13. Denjoy, A.: Sur l’iteration des fonctions analytiques. C.R. Acad. Sci. Paris 182, 255–257 (1926)
14. Du, B.S.: A simple proof of Sharkovsky’s theorem revisited. Am. Math. Mon. 114, 152–155
(2007)
15. Ellis, R.: Distal transformation groups. Pac. J. Math. 8, 401–405 (1958)
16. Gross, F.: On factorization of meromorphic functions. Trans. Am. Math. Soc. 131, 215–222
(1968)
17. Huneke, J.P.: On common fixed points of commuting continuous functions on an interval.
Trans. Am. Math. Soc. 139, 371–381 (1969)
18. Isbell, J.R.: Commuting mappings of trees, research problem # 7. Bull. Am. Math. Soc. 63,
419 (1957)
19. Jachymski, J.: Equivalent conditions involving common fixed points for maps on the unit
interval. Proc. Am. Math. Soc. 124, 3229–3233 (1996)
20. Krasnoselskii, M.A.: Two remarks on the method of sequential approximations. Usp. Mat.
Nauk 10, 123–127 (1955)
21. Li, T.Y., Yorke, J.A.: Period three implies chaos. Am. Math. Mon. 103, 985–992 (1975)
22. May, R.B.: Simple mathematical models with very complicated dynamics. Nature 261, 459–
467 (1976)
23. Numakura, K.: On bicompact semigroups. Math. J. Okayama Univ. 1, 99–108 (1952)
24. Ritt, J.F.: Permutable rational functions. Trans. Am. Math. Soc. 25, 399–448 (1923)
25. Schirmer, H.: A topologist’s view of Sharkovsky’s theorem. Houst. J. Math. 11, 385–394 (1985)
26. Schwartz, A.J.: Common periodic points of commuting functions. Mich. Math. J. 12, 353–355
(1965)
27. Sharkovsky, A.N.: Coexistence of cycles of a continuous mapping of the line into iteslf. Ukr.
Math. J. 16, 61–71 (1964)
28. Shields, A.L.: On fixed points of analytic functions. Proc. Am. Math. Soc. 15, 703–706 (1964)
29. Suffridge, T.J.: Common fixed points of commuting holomorphic maps of the hyperball. Mich.
Math. J. 21, 309–314 (1974)
30. Thron, W.J.: Sequences generated by iteration. Trans. Am. Math. Soc. 96, 38–53 (1960)
31. Wallace, A.D.: The structure of topological semigroups. Bull. Am. Math. Soc. 61, 95–112
(1955)
32. Wolff, J.: Sur l’iteration des functions holomorphe. C.R. Acad. Sci. Paris 182, 42–43 (1926).
200-201
Chapter 3
Fixed Points and Order
This chapter deals with fixed points of mappings on partially ordered sets (vide
Definition 1.1.10) under diverse hypotheses.
3.1 Fixed Points in Linear Continua
In this section, some elementary fixed point theorems on linear continua due to
Andres et al. [3] are discussed.
Definition 3.1.1 A partially ordered set (or a poset, for short) (X, ≤) is said to be
linearly ordered or totally ordered or a chain, if for any pair of elements x, y ∈ X ,
x ≤ y or y ≤ x.
Definition 3.1.2 Let S be a nonempty subset of a poset (X, ≤). x0 ∈ X is called

an upper bound (lower bound) for S if s ≤ x0 (x0 ≤ s) for all s in S. An element
x0 ∈ X is called least upper bound, (l.u.b. for short) or supremum (greatest lower
bound (g.l.b. for short) or infimum) of S if x0 is an upper bound for S and x0 ≤ x for
every upper bound x for S (if x0 is a lower bound for S and x ≤ x0 for every lower
bound x of S). A maximal element (minimal element) m of X is an element m of
X for which m ≤ x(∈ X ) implies x = m (an element m of X for which x ≤ m of
x ∈ X implies x = m).
Definition 3.1.3 A linearly ordered set (X, ≤) with more than one element is called
a linear continuum, if
(i) it is densely ordered or without gaps if for x, y ∈ X with x < y, there exists
z ∈ X such that x < z < y (a < b if a ≤ b and a = b) and
(ii) for each nonempty subset S of X bounded above there is a least upper bound
(l.u.b) in X (called l.u.b property).
https://doi.org/10.1007/978-981-13-3158-9_3
52 3 Fixed Points and Order
Remark 3.1.4 If (X, ≤) is a linearly ordered set, then the sets of the form {x ∈ X :
x < a}, {x ∈ X : a < x}, a ∈ X is a subbase for a topology on X , called the order
topology on X .
X is connected and Hausdorff in the order topology if and only if X is a linear con-
tinuum in the sense of Definition 3.1.3. Moreover, the order topology is the smallest
topology on X under which < is continuous (see Kelley [10], pp. 57–58). Also in a
linearly ordered set with the order topology every closed and (order) bounded subset
of X is compact if and only if it has the l.u.b property (see Kelley [10], p. 162).
Definition 3.1.5 Let X be a linear continuum and 2 X denote the set of all subsets
(power set) of X . A map ϕ : X → 2 X − {φ} is called a multimap. An element x0 ∈ X
such that x0 ∈ ϕ(x0 ) is called a fixed point of ϕ.
Theorem 3.1.6 (Andres et al. [3]) Let (X, ≤) be a linear continuum and I =
[a, b] = {x ∈ X : a ≤ x ≤ b}, where a, b ∈ X . Let ϕ : I → X be a multimap on
I with a connected graph. If either I ⊆ ϕ(I ) or ϕ(I ) ⊆ I , then ϕ has a fixed point.
Proof Let G ϕ = {(x, y), y ∈ ϕ(x), x ∈ I } be the graph of ϕ, a subset of I × X ⊆

X 2 . Let D = {(x, x) : x ∈ X }, P1 = {(x, y) ∈ X 2 : x < y} and P2 = {(x, y) ∈ X 2 :
y < x}. Suppose ϕ has no fixed point. Then Fϕ = {x ∈ I : x ∈ ϕ(x)} is empty. Sup-
pose I ⊆ ϕ(I ). Then there exists c, d ∈ I such that a ∈ ϕ(c) and b ∈ ϕ(d). Since
Fϕ = φ, a < c and d < b. As a(∈ φ(c)) < c, (c, a) ∈ P2 ∩ G ϕ and as b ∈ ϕ(d) and
d < b, (d, b) ∈ P1 ∩ G ϕ . Since G ϕ ⊆ P1 ∪ P2 and P1 and P2 are separated open sets
and G ϕ is connected in X 2 , G ϕ ⊆ P1 or G ϕ ⊆ P2 . This is a contradiction to the fact
that (c, a) ∈ P1 ∩ G ϕ and (d, b) ∈ P2 ∩ G ϕ . Hence ϕ has a fixed point in I .
If ϕ(I ) ⊆ I , then a < p for all p ∈ ϕ(a) as we have assumed that Fϕ = φ. For
similar reasons, q < b for all q ∈ ϕ(b). So (a, p) ∈ P1 ∩ G ϕ for all p ∈ ϕ(a) and
(b, q) ∈ P2 ∩ G ϕ for all q ∈ ϕ(b). As G ϕ ⊆ P1 ∪ P2 and P1 and P2 are separated,
G ϕ ⊆ P1 or G ϕ ⊆ P2 . In any case, this is a contradiction as G ϕ contains points of
P1 as well as P2 . So, in this case also ϕ has a fixed point in I .
Theorem 3.1.6 has a simple corollary generalizing Theorems 2.15 and 2.1.7.
Corollary 3.1.7 Let f : I → X be a map with a connected graph where I =

[a, b] ⊆ X , a linear continuum. If f (I ) ⊆ I or I ⊆ f (I ), then f has a fixed point
in I .

sin x1 , x = 0
Example 3.1.8 Define f : [−1, 1] → [−1, 1] by f (x) = . f is not
1, x =0
continuous (at x = 0). But it has a connected
graph and
has infinitely many fixed
points as it maps continuously In = (2n+3) π , (2n+1) π into [−1, 1] (containing In )
1 1
2 2
for each n (and so has a fixed point in In ).
Proposition 3.1.9 Let f : L → 2 L − {φ} be a multimap with a connected graph

G f . If f has an n-orbit for some x then f has a fixed point in L.
3.1 Fixed Points in Linear Continua 53
Proof Let {x1 , x2 , . . . , xn } be an n-orbit of f . Thus xi+1 ∈ F(xi ), i = 1, . . . , n −

1 and xn ∈ F(x1 ). Let F f = {x ∈ L : x ∈ f (x)}, the set of fixed points of f be
empty. Let a = min{x1 , . . . , xn } and b = max{x1 , x2 , . . . , xn }. So there exist xk , x
in the orbit of f such that xk ∈ f (a) and x ∈ f (b). Then a < xk and x < b since
/ F f . As before (a, xk ) ∈ P1 = {(x, y) ∈ G f with x < y} and (b, x ) ∈ P2 =
a, b ∈
{(x, y) ∈ G f with y < x}. Since G f = P1 ∪ P2 and P1 and P2 are disjoint open
subsets of G f and G f the graph of f is connected, G f ⊆ P1 or G f ⊆ P2 . This is a
contradiction as G f ∩ P1 , G f ∩ P2 = φ. Hence f has a fixed point.
Example 3.1.10 Define f : [0, 1] → 2[0,1] − {φ} as follows:

{ 1 , 1 }, x ∈ [0, 1)
f (x) = 16 13
[ 6 , 3 ], x = 1
f has a connected graph, though f 2 (x) = { 61 , 13 } does not have a connected graph.
However f has a fixed point in view of Theorem 3.1.6 as f (I ) ⊆ I and so f 2 also
has a fixed point.
Corollary 3.1.11 Let f : L → 2 L − {φ} be a map on a linear continuum L with a

connected graph. If for some n ∈ N, for the nth iterate f n of f , there exists a closed
interval I ⊆ L such that I ⊆ f n (I ) or f n (I ) ⊆ I , then f has a fixed point.
Proof As f n has a connected graph, by Theorem 3.1.6, f n has a fixed point. If this
is not a fixed point of f , then there is a nontrivial k ∈ N such that k factors n. So by
Proposition 3.1.9 f has a fixed point.
Example 3.1.12 Define the multimap f on [0, 1] by

⎧ 1
⎪
⎪ [ , 1], x =0
⎪ 2
⎨ 1 − x, x ∈ (0, 1) − { 21 }
f (x) = (3.1)
⎪
⎪ 0, x = 21
⎪
⎩ 1
[0, 2 ], x =1
Now
⎧
⎪
⎨[0, 2 ], x = 0
1
f 2 (x) = x, x ∈ (0, 1) − { 21 }
⎪
⎩ 1
[ 2 , 1], x ∈ { 21 , 1}
and every point of [0, 1] is a fixed point of f 2 but f has no fixed point. Further the
graph of f is not connected. Hence in Corollary 3.1.11 the hypothesis that the graph
of f is connected cannot be dropped.
3.2 Knaster–Tarski Principle
A set-theoretical fixed point theorem for maps in the power set of a set proved in
1927 by Knaster [11] and improved by Tarski germinated into the following theorem
referred as the Knaster–Tarski principle in the literature.
Definition 3.2.1 A map f : X → X , where (X, ≤) is a poset is said to be isotone

if f (x) ≤ f (y) for all x, y ∈ X with x ≤ y.
Theorem 3.2.2 Let (X, ≤) be a poset and f : X → X be an isotone map such that
(i) b ≤ f (b) for some b ∈ X ;
(ii) every chain in X 1 = {x ∈ X ; b ≤ x} has a supremum.
Then F f , the set of fixed points of f is nonempty and contains a maximal fixed point.
Proof Consider the subset P = {x ∈ X : b ≤ x ≤ f (x)}. P is nonempty as b ∈ P.

Every chain C in P, being a subset of X 1 has a supremum u in X 1 and for c ∈ C,
c ≤ u and by the isotonicity of f and the definition of C ⊆ P, c ≤ f (c) ≤ f (u). So
f (u) is an upper bound for C and clearly b ≤ u ≤ f (u). So u ∈ P. Now by Zorn’s
Lemma [20] since every chain in P has an upperbound (indeed a supremum), P has
a maximal element x0 in X . Since x0 ≤ f (x0 ), f (x0 ) ≤ f ( f (x0 )) and so f (x0 ) ∈ P.
If x0 = f (x0 ), there is a contradiction to the maximality of x0 . So x0 = f (x0 ). x0
being a maximal element of P is a maximal fixed point and b ≤ x0 .
In this context the following theorem due to Bourbaki [5], also called Zermelo’s
theorem [19] is stated and the proof is left as an exercise.
Theorem 3.2.3 (Bourbaki–Zermelo) Let (X, ≤) be a poset in which every chain has
an upper bound. Let f : X → X be a map such that x ≤ f (x) for all x ∈ X . Then
f has a fixed point.
In fact Abian [1] and Moroianu [13] have pointed out that it suffices to assume
the completeness of each well-ordered subset of X instead of the completeness of a
chain in Bourbaki’s theorem. First, we recall
Definition 3.2.4 A subset S of a poset (X, ≤) is called well-ordered if every

nonempty subset S0 of S has a least element in S0 .
Remark 3.2.5 The set of natural numbers with the ‘natural order’ is well-ordered,
though the real number system is not well-ordered. While every nonempty subset of
[0, 1] has a least and a greatest element in [0, 1], these need not belong to the subset
concerned. So [0, 1] is not well-ordered. The Axiom of Choice is equivalent to the
hypothesis that every nonempty set can be well-ordered (see Kelley [10]).
We now prove the following version of Bourbaki’s theorem, due to Moroianu [13]
which is equivalent to the Axiom of Choice.
3.2 Knaster–Tarski Principle 55
Theorem 3.2.6 If every well-ordered subset A of a poset (X, ≤) has an upper bound
in X and f : X → X is a map such that x ≤ f (x) for all x ∈ X , then f has a fixed
point in X .
At this stage, it is convenient to introduce the following remark and a definition.
Remark 3.2.7 A map f : X → X , where (X, ≤) is a poset is called progressive or
expansive if x ≤ f (x) for all x ∈ X .
Definition 3.2.8 Let (X, ≤) be a poset and W be the set of well-ordered subsets of
X . If A ∈ W and x ∈ A the initial segment defined by x in A is the set A x = {y ∈
A : y < x}.
Definition 3.2.9 Let F : W → X be a map, where (X, ≤) is a poset and W , the set
of all well-ordered subsets of X . A well-ordered subset A of X is called an F-chain
if for each x ∈ A which is not the least element of A, f (A x ) = x.
We prove Theorem 3.2.6, making use of the following lemmata.
Lemma 3.2.10 For two distinct well-ordered subsets A and B of a poset (X, ≤),
the following statements are equivalent:
(i) one of A, B is an initial segment of the other;
(ii) if x ∈ A and y ∈ B are such that A x = B y then x = y.
Proof Let A be an initial segment of B. Let y1 = infimum B − A. Then A = B y1 . Let
x ∈ A and y ∈ B such that A x = {a ∈ A : a < x} = {b ∈ B : b < y} = B y . Since
A = B y1 , A x = {b ∈ B : b < x and b < y1 } = {b ∈ B : b < x ∗ } = Bx ∗ = B y . So
x ∗ = y. Hence x ∗ = min(x, y1 ) = y. Since each a ∈ A is less than y1 ∈ B − A and
x ∈ A, x ∗ = x. Thus x = y.
Suppose for two distinct well-ordered sets A and B there exist x in A and y
in B such that A x = B y . Clearly A and B have the same least element. Define
S = {x ∈ A ∩ B : A x = B y }. S is nonempty as it contains the least element of A and
B. Further, for any x ∈ S, S contains A x as also Bx . So S = A or A x for some x in
S. Since A and B are distinct, one can suppose that S = A. So S = A x for some x
in A. For similar reasons S = B or B y for some y in B. Thus S = A x = B y . Now
by hypothesis (ii) x = y and this means that A x = Bx and by the definition of S,
x ∈ S = A x implying that x < x. So A x = S = B. Thus (i) is true.
From the proof that (ii) implies (i) in Lemma 3.2.10 we have
Corollary 3.2.11 If A and B are two F-chains with the same least element then one
of them is an initial segment of the other.
Lemma 3.2.12 If τ (a) is the family of F-chains, having a ∈ X as the least element,
then C = ∪{A : A ∈ τ (a)} is an element of τ (a).
Proof By Corollary 3.2.11, C is a well-ordered subset of X with a as the least ele-
ment. If x ∈ C, then a ∈ A for some A ∈ C(a) and A x = C x . So f (C x ) = x from
the definition of A as an F chain. This implies that τ itself is an f chain.
Corollary 3.2.13 If F : W → X is a mapping such that for each A ∈ W , F(A) is

an upper bound of A, then F(C) ∈ C.
Proof If F(C) ∈ / C, let C = C ∪ {F(C)}. For x ∈ C , C x = C x if x = F(C) and
C x = C if x = F(C). This is always F(C x ) and C ∈ τ (a). Since C contains C as
a proper subset this contradicts the definition of τ .
Theorem 3.2.14 Let X be an ordered set such that every well-ordered subset A of X
has an upper bound in X . Let f : X → X be a (progressive) map such that x ≤ f (x)
for all x ∈ X . Then f has a fixed point.
Proof As A is well-ordered subset of X , so is f (A). Define a map F : W → X such

that (i) F(A) is an upper bound for each A ∈ W when A does not contain the greatest
element and (ii) F(A) = f (x), when A contains the greatest element x. Clearly F
satisfies all the hypotheses of Corollary 3.2.13 and so there is a well-ordered subset
C of X such that F(C) is the greatest element of C. In view of (ii) F(C) = f (F(C)).
Thus F(C) ∈ X is a fixed point of f .
Remark 3.2.15 Since each A ∈ W has an upper bound, U the set of upper bounds
of A is nonempty. For each A ∈ W , we can choose one from U A using the axiom
of choice. In case each A ∈ W has a supremum in X , then F(A) = f (sup A) is a
well-defined map of W into X (and there is no need to invoke the axiom of choice).
However, by invoking the Axiom of Choice to X it is clear that X has a maximal
element which will, of course, be a fixed point of f , the last element of C. Theo-
rem 3.2.14 is equivalent to the Axiom of Choice, as observed by Abian [2].
Theorem 3.2.16 (Abian [2]) The Axiom of Choice is equivalent to the statement
that every (progressive) map f : X → X , where (X, ≤) is a poset wherein every
well-ordered subset has an upper bound and x ≤ f (x) for all x ∈ X , has a fixed
point.
Proof The use of the Axiom of Choice in the proof of the fixed point theorem,
i.e. Theorem 3.2.14 was already noted. (Zorn’s Lemma, equivalent to the Axiom of
Choice can be invoked to obtain a maximal element z 0 and as z 0 ≤ f (z 0 ), z 0 is a
fixed point of f under the hypotheses of Theorem 3.2.14).
We now show that this fixed point theorem implies the Axiom of Choice. Let S
be any set and (W, ≤) be the set of all well-ordered subsets of S which are partially
ordered initial segment wise (i.e. W1 ≤ W2 if W1 is an initial segment of W2 , where
W1 , W2 ∈ W ). Let (N, ≤) be the set of all natural numbers with the usual order. Let
(W × N, ≤) be the cartesian product of W with N partially ordered lexicographically
(i.e. (W1 , n 1 ) ≤ (W2 , n 2 ) if W1 ≤ W2 and W1 = W2 and if W1 = W2 , n 1 ≤ n 2 . Define
f : W × N → W × N by f (w, n) = (w, n + 1). Clearly x ≤ f (x) for all x ∈ W ×
N and f has no fixed point. In view of Theorem 3.2.14 (W × N, ≤) has a well-ordered
subset without an upper bound (I).
However, every well-ordered subset of W has an upper bound. On the other hand,
not every (well-ordered) subset of N has an upper bound. (I) is impossible unless
(W, ≤) has a maximal element. This implies that S itself is a well-ordered set.
3.2 Knaster–Tarski Principle 57
We conclude this section by proving two theorems on the existence of a common

fixed point for a family of mappings.
Theorem 3.2.17 Let F be a commuting family of mappings of a poset (X, ≤) into

itself. Suppose x ≤ f (x) for all f ∈ F and all x ∈ X . If each well-ordered set in X
has an upper bound then there is a common fixed point for each f ∈ F .
Proof For f ∈ F , in view of Remark 3.2.15 f has a fixed point x0 ∈ X , which is

a maximal element. If g ∈ F , then g(x0 ) = g( f (x0 )) = f (g(x0 )) ≥ g(x0 ) ≥ x0 =
f (x0 ) using the commutativity of f and g and the progressive nature of f and g. As
x0 is a maximal element, g(x0 ) = x0 = f (x0 ). Thus, x0 is a common fixed point for
all members of F .
Our next theorem supplements the above theorem.
Theorem 3.2.18 Let F be a commuting family of isotone maps of a poset (X, ≤)

into itself. Suppose for some fixed element a ∈ X , a ≤ f (a) for all f ∈ F . If each
chain in X containing a has a supremum in X , then F has a common fixed point.
Proof The set A = {x ∈ X : a ≤ x ≤ f (x) for all f ∈ F } is a poset with the partial
order inherited from (X, ≤) and every chain in A has a supremum in X . Since a ≤ x
for x ∈ A and f is isotone, f (a) ≤ f (x). As a ≤ f (a), we have a ≤ f (a) ≤ f (x)
for all x ∈ A and f ∈ F . For g ∈ F , a ≤ x ≤ f (x) implies a ≤ g(a) ≤ g(x) ≤
g( f (x)) = f g(x) for all f ∈ F and x ∈ A. So each g ∈ F maps A into itself.
Further every chain C in A has a supremum s in X and as a ≤ x ≤ s for x ∈ C,
a ≤ f (a) ≤ f (x) ≤ f (s) and x ≤ f (x) ≤ f (s) for each f ∈ F and x ∈ C. So
sup C = s ≤ f (s). Thus C has an upper bound in A. So there is a maximal element
x0 in A (by Zorn’s Lemma). Thus a ≤ x0 ≤ f (x0 ) for all f ∈ F . By the maximality
of x0 , it follows that x0 = f (x0 ) for all f ∈ F . Hence, there is a common fixed point
for all the functions in F .
3.3 Tarski’s Lattice Theoretical Fixed Point Theorem and

Related Theorems
The following well-known definitions figure in the original version of Tarski’s fixed
point theorem.
Definition 3.3.1 A poset (X, ≤) is called a lattice if it contains the minimum and
maximum of every pair of elements.
A lattice is said to be complete if every nonempty subset S of X has a supremum
and an infimum in X .
Example 3.3.2 (a) R, the set of all real numbers is a lattice in the usual order.
Though it is totally ordered, R is not a complete lattice. While N, the set of
natural numbers is well-ordered, R is not well-ordered.
(b) C[0, 1], the set of all continuous real-valued functions on [0, 1] with the partial
order defined by f ≺ g if and only if f (x) ≤ g(x) for all x ∈ [0, 1] is a lattice.
For f, g ∈ C[0, 1], min{ f, g} or max{ f, g} need not coincide with either f or
g. For instance for f (x) ≡ x, g(x) ≡ 21 for x ∈ [0, 1],

x, x ∈ [0, 21 ]
min{ f, g}(x) = 1
2
, otherwise

1
,x ∈ [0, 21 ]
max{ f, g}(x) = 2 .
x, otherwise
However, these are different from both f and g; That C[0, 1] is not a complete
lattice is left as an exercise.
(c) Let I be the closed unit interval in R and for n ∈ N, n > 1, let I n be the n-
fold cartesian product of I with itself. Define the partial order ≺ on I n be x =
(x1 , x2 , . . . , xn ) y = (y1 , y2 , . . . , yn ) for x, y ∈ I n if and only if xi ≤ yi for
all i = 1, 2, . . . , n. Though I n is not totally ordered it is a complete lattice for
n > 1.
(d) For a nonempty set X , 2 X , the power set of X with set-inclusion as a partial order
is a complete lattice.
Theorem 3.3.3 (Tarski [18]) Let (X, ≤) be a complete lattice and f : X → X an

isotone map. Then F f , the set of all fixed points of f is a nonempty complete lattice
under the partial order induced by ≤. In particular sup F f = sup{x ∈ X : f (x) ≥ x}
and inf F f = inf{x ∈ X : f (x) ≤ x} belong to F f .
Proof Let 0 = inf X and 1 = sup X and these exist as X is a complete lattice. So
v = sup{x ∈ X : f (x) ≥ x} is well-defined as f (0) ≥ 0. For any x ∈ X with x ≤
f (x), x ≤ u and by the isotonicity of f , f (x) ≤ f (u) and so x ≤ f (x) ≤ f (u).
As x ≤ u, for x ≤ f (x), u ≤ f (u). Again by the isotonicity of f , f (u) ≤ f ( f (u)).
So f (u) ∈ {x ∈ X : x ≤ f (x)}. Since u = sup{x ∈ X : x ≤ f (x)}, f (u) ≤ u. Thus
u = f (u) and u ∈ F f . Since for every fixed point x of f , f (x) ≥ x, u = sup F f .
The set X with the partial order , defined by x y if y ≤ x, called the dual-
order of ≤ is a complete lattice on which f is again an isotone map (in the dual
order). Further for any subset S of X , inf (≤) S = sup( ) S and sup(≤) S = inf ( ) S
where is the dual-order of the partial order ≤. So, by the first part of the theorem
proved so far, inf{x ∈ X : f (a) ≤ x} = sup{x ∈ X : x f (x)} is in F f . Clearly
inf F f = inf{x ∈ X : f (x) ≤ x}.
Let Y be any subset of F f . Then ([sup Y, 1], ≤) is a complete lattice, where
[a, b] = {x ∈ X : a ≤ x ≤ b}. If x ∈ Y , then x ≤ sup Y and so x = f (x) ≤
f (sup Y ) as f is isotone and Y ⊆ F f . So sup Y ≤ f (sup Y ). If sup Y ≤ z, then
sup Y ≤ f (sup Y ) ≤ f (z). So f , the restriction of f to [sup Y, 1] maps the com-
plete lattice [sup Y, 1] into itself and is isotone. So v = inf of all fixed points of f
(and indeed f ) in [sup Y, 1] is itself a fixed point of f by the preceding part of
the theorem. This is, indeed, the least fixed point of f , which is an upper bound of
3.3 Tarski’s Lattice Theoretical Fixed Point Theorem and Related Theorems 59
all elements of Y . Similarly, consider the complete lattice [0, inf Y ]. If x ∈ Y , then
inf Y ≤ x and f (inf Y ) ≤ f (x) ≤ x, as f is isotone and so f (inf Y ) ≤ inf Y . Thus
f ∗ , the restriction of f to [0, inf Y ] maps this complete lattice into itself. So u = sup
of all the fixed points of f ∗ in [0, inf Y ] is itself a fixed point and indeed the greatest
lower bound of all fixed points of f in Y . Thus, inf Y and sup Y are also fixed points
of f . Thus, (F f , ≤) is a complete lattice.
Remark 3.3.4 The existence of a fixed point in Tarski’s theorem can be proved along
the lines of Theorem 3.2.2 as well.
The next theorem is also due to Tarski [18] and its proof is left as an exercise.
Theorem 3.3.5 Let (X, ≤) be a complete lattice and G a commuting family of

isotone maps of X into itself. Then FG the set of common fixed points of all the
functions f in G is nonempty and (FG , ≤) is a complete lattice. Further sup FG =
sup{x ∈ X : x ≤ f (x) for all f ∈ G} and inf FG = inf{x ∈ X : f (x) ≤ x for all
f ∈ G}.
Remark 3.3.6 A decreasing map on a complete lattice may not have a fixed point.
For example, the map f : [0, 1] → [0, 1] defined by

0, if x = 0
f (x) =
1, if x = 0
is decreasing (i.e. f (x) ≥ f (y) for x ≤ y) and has no fixed point.
Remark 3.3.7 Theorem 3.3.5 is not true for a non-commutative family of isotone
maps. For instance, define f : [0, 1] → [0, 1] by f (x) = 1+x
2
and g : [0, 1] → [0, 1]
by g(x) = 2 for all x ∈ [0, 1]. While both f and g are isotone and non-commuting,
1
f and g do not have a common fixed point.
Remark 3.3.8 Any non-decreasing map of [0, 1] into itself (being isotone) will
always have a fixed point, even if it is not continuous. Contrast this with Theo-
rem 2.1.5 and Corollary 3.1.7.
Remark 3.3.9 The fixed point guaranteed by Tarski’s theorem is not unique. For the
map f : [0, 1] → [0, 1] defined by

x, x ∈ [0, 21 )
f (x) =
1, x ≥ 21 .
F f = [0, 21 ) ∪ {1} is the set of fixed points. Although 21 = sup[0, 21 ) in [0, 1], as a
subset Y = [0, 21 ) of the set of fixed points F f of f the supremum is 1!. F f is a
complete lattice.
Davis [7] had proved that the converse of Tarski’s theorem that a lattice in which
every isotone map has a fixed point is complete.
Based on Tarski’s observation that inf{x ∈ X : f (x) ≤ x} is a fixed point for f ,
an isotone self-map on a complete lattice X , Merrifield and Stein and Stein [12, 17]
proved some fixed point theorems supplementing/generalizing Tarski’s fixed point
theorem in complete lattices. A few of these results are highlighted in the sequel, as
these embellish the crux of the proof of Tarski’s theorem.
Theorem 3.3.10 (Stein [16]) Let (X, ≤) be a complete lattice and T : X → X , a

map. Suppose there exist positive integers k and n such that for x, y ∈ X
(i) T n (x) ≤ y implies T k (T n (x)) ≤ T k (y) and
(ii) x ≤ T n (y) implies T k (x) ≤ T k (T n (y)).
Then T k has a fixed point.
Proof Define H = {x ∈ X : x ≥ T nk (x)} H is non-void as 1 = sup X ∈ H . For x ∈

H , T nk (x) ≤ x and so T n (T (k−1)n (x)) ≤ x. So by (i) T k (T nk (x)) ≤ T k (x). Thus
T nk (T k (x)) ≤ T k (x). So T k (x) ∈ H , for x ∈ H . Consequently T jk (x) ∈ H , for
j = 1, 2, . . ., whenever x ∈ H . Also for x ∈ H , x ≥ T k (x) ≥ T 2k (x) · · · ≥ T jk (x),
j ≥ 2.
Since n and k are fixed natural numbers, there exists a natural number of
sufficiently large such that (qn − 1)k ≥ n. For x ∈ H , T (qn−1)k (x) ∈ H in view
of the preceding paragraph, as qn > 1. Let h = inf H . Then h ≤ T (qn−1)k (x) =
T n (T (qn−1)k−n (x)). Using (ii), we get that T k (h) ≤ T k T (qn−1)k (x) = T qnk (x). As
x ∈ H , T qnk (x) ∈ H and T qnk (x) ≤ x, T k (h) ≤ T qnk (x) ≤ x. Since x is an arbi-
trary element of H , T k (h) ≤ h. So by definition of H , h ∈ H and so T k (h) also is
in H . As h = inf H and T k (h)(∈ H ) ≤ h, h = T k (h). Thus T k has a fixed point in
X.
Remark 3.3.11 The transposition T (0) = 1 and T (1) = 0 on the lattice {0, 1} has
no fixed point. But T satisfies all the conditions of Theorem 3.3.10 for k = 2, n = 1
and hence T 2 has a fixed point.
Next is a theorem on common fixed points.
Theorem 3.3.12 (Stein [12]) Let F be a commuting family of maps on a complete

lattice X into itself satisfying the conditions for S, T ∈ F (i) x ≤ S(y) implies
T x ≤ T S(y) and (ii) S(x) ≤ y implies T S(x) ≤ T (y). Then the family F has a
common fixed point.
Proof Define H = {x ∈ X : x ≥ T (x), for all T ∈ F }. Clearly 1 = sup X ∈ H . Let

h = inf H .
Let x ∈ H and T ∈ F . Then it follows from (ii) that for any S ∈ F and y = x,
T S(x) = ST (x) ≤ T (x) and hence T (x) ∈ H . Since T (x) ∈ H for x ∈ H , h ≤
T (x) from the definition of h. Invoking (i) for h ≤ T x, we get T h ≤ T (T (x)) ≤
T (x) ≤ x for all x ∈ H . So T (h) ≤ h. From the definition of H , it follows that
h ∈ H . Since T (x) ∈ H for x ∈ H , T (h) ∈ H . Since h = inf H , h ≤ T (h). Hence

h = T (h). Thus for each T ∈ F , h is a fixed point. In other words F has a common
fixed point.
The following definition is needed for the next theorem.
Definition 3.3.13 Let (X, ≤) be a complete lattice and (xn ) a sequence in X . Then
lim inf(xn ) = sup {inf(xm : m ≥ n)}

n→∞ n=1,2,...
lim sup(xn ) = inf {sup(xm : m ≥ n)}.

n→∞ n=1,2,...
If lim inf(xn ) = lim sup(xn ) = x, then (xn ) is said to converge to x (with respect
n→∞ n→∞
to the partial order ≤).
Theorem 3.3.14 Let (X, ≤) be a complete chain and T : X → X be a map such
that x ≤ y, x, y ∈ X implies that there exists a natural number N = N (x, y) such
that for all n ≥ N , T n (x) ≤ T n (y). Then T has a fixed point.
Proof Suppose T has no fixed point. Then the sets A = {x ∈ X : x ≥ T (x)} and
B = {x ∈ X : x ≤ T (x)} are disjoint and nonempty as 1 ∈ A and 0 ∈ B.
Let x ∈ A. If for some natural number p, T p (x) ≤ T p+1 (x), then by hypothe-
sis there exists N1 such that for n ≥ N1 , T p+n (x) ≤ T p+n+1 (x). As T x ≤ x, there
exists N2 such that for n ≥ N2 , T n+1 (x) ≤ T n (x). So for n ≥ max{N1 + p, N2 },
T n+1 (a) ≤ T n (x) ≤ T n+1 (x) or T n+1 (x) = T n (x). Thus T has a fixed point, viz.,
T n (a). Since we have assumed that T has no fixed point, for x ∈ A, T p (x) ≥ T p+1 (x)
for all p so that lim inf T n (x) ≤ x. By a similar reasoning lim inf T n (x) ≥ x for
n→∞ n→∞
x ∈ B. Let a = inf A. So for x ∈ A, a ≤ x. Hence by hypothesis, there exists N
such that for all n ≥ N , T n (a) ≤ T n (x). So lim inf T n (a) ≤ lim inf T n (x) ≤ x
n→∞ n→∞
for each x ∈ A. So lim inf T n (a) ≤ a = inf A.
If a ∈ B, then lim inf T n (a) = a implies a = T (a) = T 2 (a) . . . and T has a
n→∞
fixed point.
If T (a) < a, then T (a) ∈ A. Since a > T (a) ≥ T 2 (a) . . . , contradicting that a
is a lower bound for A. So T (a) = a. Thus T has a fixed point.
Remark 3.3.15 In a complete chain an operator T such that lim inf T n (x) ≤
n→∞
lim inf T n (y) whenever x ≤ y may not have a fixed point. The map T : [0, 1] →
n→∞
[0, 1] defined by
x
, for x = 0
T (x) = 2
1, for x = 0
has no fixed point though T n (x) → 0 as n → ∞.

The next theorem, also due to Merrifield and Stein [12] insures a common fixed
point for a pair of isotone maps S and T for which, at each point of X , an iterate of
S is majorized by an iterate of T and vice versa.
Theorem 3.3.16 (Merrifield and Stein [12]) Let (X, ≤) be a complete lattice and
S, T : X → X be isotone maps. Suppose for each x ∈ X , there exist positive integers
p = p(x), q = q(x), i = i(x) and j = j (x) (depending on x) such that S p (x) ≤
T q (x) and T i (x) ≤ S j (x). Then S and T have a common fixed point.
Proof Define H = {x ∈ X : S(x), T (x) ≤ x}. H is nonempty as 1 ∈ H . Let u =

inf{S k (x), T k (x) : k = 0, 1, 2, . . .}. As u ≤ T n (h) for h ∈ H for all non-negative
integers n, since T is isotone, it follows from the definition of H that T (u) ≤
T n+1 (h) ≤ T n (h) for all n = 0, 1, 2, . . ., for all h ∈ H .
By hypothesis, there exist non-negative integers p1 , q1 such that T p1 (h) ≤ S q1 (h)
for h ∈ H . Suppose we have chosen p1 < p2 · · · < pk , q1 < q2 · · · < qk integers
such that T p j (h) ≤ S q j (h) for 1 ≤ j ≤ k. By hypothesis there exist positive integers
a and b such that T a (T pk (h)) ≤ S b (S qk (h)) (by the isotonicity of S). Let pk+1 =
pk + a and qk+1 = qk + b. Then pk < pk+1 and qk < qk+1 and T pk+1 (h) ≤ S qk+1 (h).
For n ≥ 0 choose k such that qk ≥ n. Then u ≤ T pk −1 (h) and so T (u) ≤ T pk (h) ≤
S qk (h) ≤ S n (h) and the fact that for h ∈ H , S m (h) ≤ S m−1 (h) ≤ · · · ≤ S(h) ≤ h for
all m ≥ 1. Thus T (u) ≤ T n (h), S n (h) for all n = 0, 1, . . . and all h ∈ H . So T (u) ≤
v. By a similar argument S(u) ≤ u. It follows from this that u ∈ H . Since u ∈ H ,
and u ≤ T (h), S(h) for all h ∈ H , u ≤ T (u), S(u). Hence u = T (u) = S(u).
Corollary 3.3.17 Let (X, ≤) be a complete lattice and S, T : X → X be isotone

maps. If (ST )k = T n or (ST )k S = T n for some natural numbers k and n, then S
and T have a common fixed point.
Proof If (ST )k S = T n , then (ST )k ST = (ST )k+1 = T n+1 . So it suffices to prove

the first part of the corollary.
If (ST ) j = T m for natural numbers j and m, then (ST ) and T satisfy the
hypotheses of Theorem 3.3.16 and hence they have a common fixed point u.
Thus u = T (u) = S(T (u)). Since T (u) = u, S(T (u)) = S(u). Thus u = T (u) =
S(u).
The following is a result akin to the corollary above, also due to Merrifield and
Stein [12].
Theorem 3.3.18 Let S and T be isotone maps on the complete lattice (X, ≤) into
itself. If ST S n = T for some natural number n, then S and T have a common fixed
point.
Proof As before let H = {x ∈ X : S(x), T (x) ≤ x} H is non-void as 1 ∈ H . Since

S is isotone, for h ∈ H , S(S(h)) ≤ S(h) and also T (S(h)) = ST S n (Sh) ≤ ST (h) ≤
S(h), as T (h) ≤ h. Hence S(h) ∈ H . Let u = inf H . Then u ≤ h for h ∈ H , S(u) ≤
S(h) ≤ h and T (u) ≤ T (h) ≤ h by the isotonicity of S and T . Hence S(u), T (u) ≤ u
and so u ∈ H . Since S(h) ∈ H for h ∈ H , S(u) ∈ H . Since u = inf H , u ≤ S(u).
Hence u = S(u) ST S n (u) = T (u) and so ST (u) = T (u) ≤ u. So T (u) ∈ H . As
u = inf H , u ≤ T (u). Hence u = T (u). Thus u = T (u) = S(u).
Corollary 3.3.19 Let (X, ≤) be a complete lattice and S, T : X → X be isotone

maps with ST S n = T S p where n > p and n, p are non-negative integers. Then S
and T have a common fixed point.
Proof In Theorem 3.3.18, replace T by T S p and n by n − p. We now conclude that

T S p and S have a common fixed point which is a common fixed point of S and T as
well.
Merrifield and Stein [12] have provided a number of sufficient conditions to

ensure that u = inf{x ∈ X : S(x), T (x) ≤ x} is a common fixed point for a pair of
isotone maps S and T in a complete lattice. The following theorem embodying such
conditions is a sample and the proof is left as an exercise.
Theorem 3.3.20 Let (X, ≤) be a complete lattice and S, T : X → X be isotone

maps. S and T have a common fixed point, if for each x ∈ X , sup{S n (x) : n =
1, 2, . . .}, is equal to any one of the following:
(a) sup{T n (x) : n = 1, 2, . . .};
(b) inf{T n (x) : n = 1, 2, . . .};
(c) lim inf{T n (x)};
n→∞
(d) lim sup{T n (x)}.
n→∞
While a decreasing or antitone self-map on a complete lattice may not have a fixed
point (vide Remark 3.3.6) Roth [14] noted that such a map has a fixed point under
additional assumptions. We conclude this section by providing two such results for
such maps, using the following.
Definition 3.3.21 Let (X, ≤) be a complete lattice map f : X → X is called

antitone (or decreasing) if f (y) ≤ f (x) whenever x ≤ y, x, y ∈ X . A function
f : X → X is called join antimorphism if f (sup A) = inf f (A) for each nonempty
subset A of X . f : X → X is called a meet antimorphism if f (inf A) = sup f (A)
for each nonempty subset A of X .
Remark 3.3.22 Clearly a join or meet antimorphism is antitone.
Theorem 3.3.23 (Blair and Roth [4]) Let h, g : X → X be maps on a complete

lattice (X, ≤). Suppose h is isotone and g maps Fh , the set of fixed points of h in X
into Fh . Then there exists x ∈ X such that x = h(x) and x ≤ g(x).
Proof Since h is an isotone map on the complete lattice X , Fh the set of fixed points
of h in X is non-void and x0 = inf{x ∈ X : h(x) ≤ x} is in Fh , as this is the smallest
fixed point of h, in view of Tarski’s Theorem 3.3.3. Since x0 = inf Fh and g maps
Fh into itself h(x0 ) = x0 = inf Fh ≤ g(x0 ) as x0 ∈ Fh and x0 ≤ g(x) for all x ∈ Fh .
Thus, there is an element x0 with x0 = h(x0 ) and x0 ≤ g(x0 ).
Corollary 3.3.24 (Roth [14]) If f : X → X is a join antimorphism on a complete

lattice (X, ≤), then there exists x ∈ X such that x = f 2 (x) and x ≤ f (x).
Proof As f is join antimorphism, f is antitone and f 2 is isotone. Setting h = f 2

and g = f in Theorem 3.3.23 and noting that f maps F f 2 into itself, it follows that
f 2 has a least fixed point x0 and x0 ≤ f (x0 ).
Theorem 3.3.25 (Blair and Roth [4]) Let (X, ≤) be a complete lattice and u 1 , u 2 :
X → X be maps such that f = u 1 ◦ u 2 and g = u 2 ◦ u 1 are isotone. Then there exist
x, y ∈ X such that x = f (x) ≤ u 1 (y) and y = g(y) ≤ u 2 (x).
Proof Since f and g are isotone and X is a complete lattice F f and Fg the sets of fixed
points of the maps f and g, respectively, are nonempty and x = inf F f ∈ F f and
y = inf Fg ∈ Fg , by Tarski’s fixed point Theorem 3.3.3. So x = u 1 u 2 (x). So u 2 (x) =
u 2 u 1 (u 2 (x)) = g(u 2 (x)). So u 2 (x) ∈ Fg , as g = u 2 ◦ u 1 . By definition of y and g,
y = g(y) = u 2 u 1 (y) ≤ u 2 (x). Similarly, as y = u 2 u 1 (y), u 1 (y) = u 1 (u 2 u 1 (y)) =
(u 1 u 2 )(u 1 (y)) = f (u 1 (y)). Hence, u 1 (y) ∈ F f , f being u 1 ◦ u 2 . From the definition
of x, x = f (x) ≤ u 1 (y).
Remark 3.3.26 Theorem 3.3.25 and Corollary 3.3.24 have found applications in
Game Theory.
3.4 Some Applications
Tarski’s fixed point theorem central to lattice theoretical fixed point theorems is of
wide applicability. In this section, we use it to prove Schroder–Bernstein theorem a
basic result in set-theory, Cantor–Bendixon theorem, a representation theorem for
closed sets in general topology and existence theorems for a Cauchy problem for
a parabolic partial differential equation, a nonlinear complementarity problem and
certain formal languages.
Theorem 3.4.1 (Schroder–Bernstein) If f is a one-to-one mapping of a set A into

a set B and g is a one-to-one mapping from B into A, then there is a bijection (i.e.
a one-to-one mapping which is onto) of A onto B.
Proof Let 2 A and 2 B be the power sets (sets of all subsets) of A and B, respectively.
2 A (and for that matter 2 B as well) is a complete lattice under set-inclusion as partial
order (see Example 3.3.2(d)). Define T : 2 A → 2 B by T (S) = A − g(B − f (S)) for
each subset S of A. For A1 , A2 ⊆ A and A1 ⊆ A2 , g(B − f (A1 )) ⊇ g(B − f (A2 ))
and so A − g(B − f (A1 )) ⊆ A − g(B − f (A2 )). In short T (A1 ) ⊆ T (A2 ) or T is
isotone on 2 A . So by Tarski’s fixed point Theorem 3.3.3, T has a fixed point S ∗ ⊆ A.
Thus S ∗ = T (S ∗ ) = A − g(B − f (S ∗ )).
Define h : A → B by

f (x), if x ∈ S ∗ ,
h(x) =
g (x), if x ∈ A − S ∗
−1
3.4 Some Applications 65
Clearly h is well-defined on S ∗ (as f ). If x ∈ A − S ∗ , then x ∈ A − T (S ∗ ) = g(B −

f (S ∗ )) and so g −1 (x) is well-defined as g −1 is 1-1 on g(B) and hence on g(B −
f (S ∗ )). Further h is one-to-one and
h(A) = h(S ∗ ∪ A − S ∗ )
= h(S ∗ ) ∪ h(A − S ∗ )
= f (S ∗ ) ∪ g −1 (g(B − f (S ∗ ))
= f (S ∗ ) ∪ B − f (S ∗ )
= B.
Hence h is a bijection of A onto B.
Remark 3.4.2 Schroder–Berstein theorem is quite useful in proving that two sets
have the same cardinality, for example (0, 1) ∪ (2, 3) and [−1, 0].
For the statement of Cantor–Bendixon theorem, we recall the definitions of a

derived set, perfect set and a scattered set.
Definition 3.4.3 For a subset A of X where (X, T ) is a topological space, x0 ∈ X

is called an accumulation point or a cluster point of A if every neighbourhood of x0
contains a point of A other than x0 . The set of all accumulation points of A is called
the derived set of A and is denoted by A .
Definition 3.4.4 Let (X, T ) be a topological space. A subset A of X is said to be

perfect if A = A (the derived set of A). (In other words, every point of A is an
accumulation point of A and every accumulation point of A is in A).
A subset A of X is called scattered if the only subset B of A for which B ⊇ B is
the empty set.
Example 3.4.5 For R with the usual topology, [0, 1] is a perfect set. So is the Cantor-
ternary subset of [0, 1]. N, the set of natural numbers is scattered. If X is a T1 -space,
then the derived set (of a set) is closed.
Theorem 3.4.6 (Cantor–Bendixon) Every closed subset of a topological space

(X, T ) is a disjoint union of a perfect set and a scattered set.
Proof Let A be closed subset of X . Define B = ∪{S ∈ 2 X : A ∩ S ⊇ S} and C =

A − B. Clearly, the map defined by T (S) = A ∩ S , S ∈ 2 X is an isotone map on the
complete lattice 2 X . So from Tarski’s Theorem 3.3.3 and its proof B = sup{S ∈ 2 X :
T (S) ≥ S} is a fixed point for T . So B = T (B) = A ∩ B or B ⊇ B. Thus every
point of B is an accumulation point of itself. Since A is closed and contains B and
B, B ⊆ B ⊆ A. Hence A ∩ B = B . Thus B = B and B is perfect.
Clearly B is disjoint from C = A − B. For S ⊆ C and S ⊇ C, A ∩ S ⊇ S since
C ⊆ A. By the definition of B, B ⊇ S. So S = S ∩ C ⊆ B ∩ C = φ. Hence S is
empty. Thus C is scattered. The proof is complete.
Finally, an application of Tarski’s fixed theorem for the solution of a Cauchy

problem for a parabolic differential equation due to Schäfer [15] is described.
Theorem 3.4.7 (Schäfer [15]) Let E be a real Banach space and g : R × [0, T ] →
E be a bounded continuous function such that
g(x, t) − g(y, t) ≤ L|x − y|
for x, y ∈ R for each t ∈ [0, T ], an interval of real numbers, for some positive
constant L.
Then the function u defined by
t ∞
1 (x−ξ )2
u(x, t) = √ e− 4(t−τ ) g(ξ, t)dξ dt
0 −∞ 4π(t − τ )
for x ∈ R, t ∈ [0, T ] is a solution of the Cauchy problem
u t (x, t) − u x x (x, t) = g(x, t), x ∈ R, t ∈ [0, T ]

u(x, 0) = 0, x ∈ R.
Remark 3.4.8 Schäfer [15] observed that the above theorem can be proved along the
lines of proof of an existence theorem in Friedman [8].
We will use the above theorem to prove the following application of Tarski’s
theorem to the Cauchy problem for a parabolic equation.
Theorem 3.4.9 Let f : R × [0, T ] × ∞ (A) → ∞ (A) be a function with the fol-
lowing properties, A being a nonempty set, and ∞ (A) the Banach space of bounded
real functions on A.
(i) f is continuous;
(ii) for a constant L 1 > 0, for all (x, t, z), (y, t, z) ∈ R × [0, T ] × ∞ (A)
f (x, t, z) − f (y, t, z) ≤ L 1 |x − y|;
(iii) for some L 2 > 0, for all (x, t, z 1 ), (x, t, z 2 ) in R × [0, T ] × ∞ (A)
f (x, t, z 1 ) − f (x, t, z 2 ) ≤ L 2 z 1 − z 2 ;
(iv) for z 1 , z 2 ∈ ∞ (A) for (x, t) ∈ R × [0, T ] z 1 (a) ≤ z 2 (a) for all a ∈ A implies
f (x, t, z 1 ) ≤ f (x, t, z 2 );
(v) for a constant M > 0, for all (x, t, z) in R × [0, T ] × ∞ (A), f (x, t, z) ≤
M.
Then there exists a continuous function u : R × [0, T ] → ∞ (A) with
u t (x, t) − u x x (x, t) = f (x, t, u(x, t)) for (x, t) ∈ R × [0, T ] and

u(x, 0) = 0 for x ∈ R
Proof Define as the set of functions {w : R × [0, T ] → ∞ (A) with w(x, t) −

w(y, s) ≤ L 3 |x − y| + L 4 |t − s|, for x, y ∈ R and t, s√∈ [0, T ], with − (a) ≤
w(a) ≤ (a) for all a ∈ A, where (a) ≡ T M, L 3 = 2M T and L 4 = M + (L 1 +
L 2 L 3 )2 πT }.
As ∈ , is non-void. For each w ∈ define
t ∞
1 (x−ξ )2
φ(w)(x, t) = √ e− 4(t−τ ) f (ξ, τ, w(ξ, τ ))dξ dt
0 −∞ 4π(t − τ )
for (x, t) ∈ R × [0, T ].

Since the integrand is continuous φ(w) is continuous on R × [0, T ]. Further, from
the definition of and the hypotheses (ii) and (iii), it follows that for (x, t), (y, t) ∈
R × [0, T ].
f (x, t, w(x, t)) − f (y, t, w(y, t)) ≤ f (x, t, w(x, t)) − f (y, t, w(x, t))
+ f (y, t, w(x, t)) − f (y, t, w(y, t))
≤ L 1 |x − y| + L 2 w(x, t) − w(y, t)
≤ (L 1 + L 2 L 3 )|x − y|. (vi)
So by Theorem 3.4.7, for w ∈ , (x, t) ∈ R × [0, T ]
φt w(x, t) − φx x w(x, t) = f (x, t, w(x, t)) and φ(w)(x, 0) = 0 (vii)
are satisfied.
We now show that is a complete lattice and φ maps into itself and φ is isotone.
Since ∈ , is nonempty. Let S be a subset of which is non-void. For each s ∈ S,
−T M ≤ s(x, t)(a) ≤ T M for all a ∈ A and (x, t) ∈ R × [0, T ]. sup s(x, t)(a) for
a∈A
(x, t) ∈ R × [0, T ] lies between −T M and T M. Further, as s(x, t) − s(y, t ) ≤
L 3 |x − y| + L 4 t − t |, sup s(x, t) − sup s(y, t ) ≤ L 3 |x − y| + L 4 t − t |.
S S
Thus sup S ∈ . By a similar reasoning inf S ∈ . Hence is a complete lattice.
For w ∈ , since − ≤ w(x, t) ≤ it follows that − ≤ φ(w)(x, t) ≤ .
Further for s < t, s, t ∈ [0, T ]⎡φ(w(x, t)) − φ(w(x, s)) ≤⎤J1 + J2 , where J1 =
(x−ξ )2 (x−ξ )2
s ∞
1 e− 4(t−τ ) e− 4(s−τ ) ⎦
√ f (ξ, τ, w(ξ, τ ) ⎣ √ −√ dξ dτ and J2 =
0 −∞ π 4π(t − τ ) 4π(s − τ )
t ∞
1 (x−ξ )2
√ e− 4(t−τ ) f (ξ, τ, w(ξ, τ ))dξ dτ . Clearly J2 ≤ M|t − s| as
s −∞ 4π(t − τ )
| f (ξ, τ, w) ≤ M. Now

s ∞
1 √ √
√ e−α f (x − 2α t − τ , τ, w(x − 2α t − τ , τ )
2
J1 ≤
0 −∞ π
√ √
− f (x − 2α s − 0, 0, w(x − 2α s − τ , τ )dτ dα
s √ √ ∞
1
e−α 2|α|dαdτ using (vi)
2
≤ √ (L 1 + L 2 L 3 ) ( t − τ − s − τ)
π 0 −∞
1 2 3 3 3
≤ √ (L 1 + L 2 L 3 ) (t 2 − s 2 − (t − s) 2 )2
π 3
1 √
≤ √ (L 1 + L 2 L 3 ) T |t − s|.
π
So J1 + J2 ≤ {M + (L 1 + L 2 L 3 )2 T
π
}|t − s|. Thus φ(w)(r, t) − φ(w)(x, s) ≤
L 4 |t − s|. Also for each a ∈ A, φ(w(a))(x, s) − φ(w(a))(y, s) = ∂∂x φ(w)(a)(η, s)
(x − y) by the mean value √theorem, where η lies between x and y. So |φ(w)(a)(x, s)
− φ(w(a))(y, s)| ≤ 2M T |y − s| = L 3 |y − s|. Thus φ(w)(x, t) − φ(w)(y, s)
≤ L 3 |x − y| + L 4 |t − s|. Clearly − ≤ φ(w)(a) ≤ for w ∈ . So φ maps into
itself and is isotone. As is a complete lattice by Tarski’s fixed point Theorem 3.3.3
φ has a fixed point u. Clearly u satisfies (vii) and is thus a solution to the Cauchy
problem.
Chitra and Subrahmanyam [6] obtained an interesting application of Tarski’s fixed

point theorem to a nonlinear complementarity problem supplementing the solution of
a class of nonlinear complementary problems in the space of real-valued continuous
functions on a compact Hausdorff space due to Fujimoto [9]. A brief description of
their main theorem is described below.
Let L p [a, b] be the Banach space of all real-valued Lebesgue measurable func-
1p
b
tions f : [a, b] → R with the norm f = a | f (x)| p d x , 1 ≤ p < ∞. Let
K be the cone of functions in this space which are non-negative almost every-
where. L p [a, b] is partially ordered by the relation x ≥ y (x, y ∈ L p [a, b]) if
x(t) ≥ y(t) a.e. Of course, functions which are almost everywhere equal are con-
sidered equal.
The nonlinear complementarity problem (NLCP) is to find x ∈ K (⊆ L p [a, b])
such that
x − T x − b ≥ 0 and min(x, x − T x − b) = 0
where b is a given element in L p [a, b] and T maps K (the cone in L p [a, b]) into
L p [a, b]. We have the following:
Theorem 3.4.10 ([6]) Suppose

(i) T : K → L p [a, b] is monotone with respect to K (i.e. T x ≥ T y, if x ≥ y, x, y ∈
K );
(ii) there exists an x0 in K with x0 − T x0 − b ≥ 0.

Then the above NLCP has a solution.
Proof Consider the set D = {x ∈ K : x ≤ x0 } where x0 satisfies (ii). The mapping
F defined on D by F(x) = max(T x + b, 0) is non-negative and F x ≤ x0 for all
x ∈ D, by the monotonicity of T and (ii). Thus F is a monotone map of D into
itself. From the separability of L p [a, b] and the fact that every Cauchy sequence in
L p [a, b] has a subsequence which converges pointwise a.e in L p [a, b], it follows
that D is a complete lattice. So by Tarski’s theorem, the monotone operator F has a
fixed point x ∗ in the complete lattice D. Clearly for x ∗ ∈ D, x ∗ − T x ∗ − b ≥ 0 and
min(x ∗ , x ∗ − T x ∗ − b) = 0. Further x ∗ ≤ x0 .
Remark 3.4.11 The above fixed point x ∗ can be obtained as the limit of the monotonic
sequence of iterates defined by x1 = max(T x0 + b, 0) xn+1 = max(T xn + b, 0)
n ∈ N. Define E = {x ∈ K : x − T x − b ≥ 0}. As inf E = inf{x ∈ K , x ≤ x 0 , x −
T x − b ≥ 0} = inf{x ∈ D : F x ≤ x}, from Tarski’s theorem inf E is a solution to
the NLCP.
Corollary 3.4.12 Let f : [0, 1] × R → R be a function such that
(i) f (s, u) is continuous with respect to u for almost all s ∈ [0, 1] and measurable
with respect to a for all u ∈ R;
(ii) | f (s, u)| ≤ a0 (s) + b0 |u| where b0 > 0 as a0 (s) ∈ L p [0, 1]; and
(iii) for almost all s, f (s, u) is increasing with respect to u and there exists
x0 (s), b(s) ∈ L p [0, 1] such that x0 (t) − f (t, x0 (t)) − b(t) ≥ 0 for almost all
t ∈ [0, 1].
Then the problem of finding x ∈ K ⊆ L p [0, 1] such that x(t) − f (t, x(t)) − b(t) ≥
0 and min(x(t), x(t) − f (t, x(t)) − b(t)) = 0 has a solution.
Proof Define T : K → L p[0, 1] by T x(t) = f (t, x(t)). From (i) and (ii), it follows
that T x ∈ L p [0, 1] for x ∈ L p [0, 1]. Using (iii), it can be seen that conditions (i)
and (ii) of Theorem 3.4.10 are fulfilled by T . So by Theorem 3.4.10 F has a fixed
point, where F = max{T x + b, 0} has a fixed point. In other words the NLCP has a
solution.
√
Remark 3.4.13 For the choice f (s, u) = cos 1+us , x0 ≡ 1 and b ≡ 0, the nonlinear
√
s
complementarity problem x(s) ≥ 0, x(s) − cos 1+x(s) ≥ 0 and min (x(s), x(s)−
√
s
cos 1+x(s) = 0 has a solution. It can also be proved that this solution is unique.
It may be noted that this operator is not completely continuous.
Finally follows an application of Tarski’s theorem for proving the existence of
certain formal languages. Our alphabet consists of just two letters α and β. Words
are formed by concatenation of these letters. For example, αβ, ααβ are words and
words have finite length. The empty word denoted by φ is also allowed. There is
also a grammar to this formal language. More specifically, it has three grammatical
terms:
(i) the term A - consisting of all words containing α and φ;

(ii) the term B - comprising all words containing β and φ;
(iii) the term C - made of all words belonging to A or B.
The rules of the grammar are clear:
(i) for η ∈ A, ηα ∈ A;
(ii) for η ∈ B, ηβ ∈ B;
(iii) for η ∈ C, if and only if η ∈ A or η ∈ B;
(vi) φ ∈ A, B, C.
Theorem 3.4.14 There exists a formal language fulfilling the above conditions.
Proof Let W be the set of all words and X the set of all triples x = (A, B, C) where
A, B and C are subsets of words as described above. Define a partial order ≤ on X
by (A1 , B1 , C1 ) ≤ (A2 , B2 , C2 ) if A1 ⊆ A2 , B1 ⊆ B2 and C1 ⊆ C2 . It can be seen
that X is a complete lattice. Define f : X → X by f (A, B, C) = (Aα, Bβ, A ∪ B).
Clearly f is isotone and by Tarski’s theorem f has a least fixed point. Indeed for
∞
∞ ∞
A= An , B = Bn and C = Cn , where A0 = B0 = φ and An+1 = An α,
n=0 n=0 n=0
Bn+1 = Bn β and Cn+1 = An ∪ Bn , n = 0, 1, 2, . . . (A, B, C) corresponds to this
∞
∞
∞

least fixed point where A = An , B = Bn and C = Cn .
n=0 n=0 n=0
References
1. Abian, A.: Fixed point theorems of the mappings of partially ordered sets. Rend. Circ. Mat.
Palermo 20, 139–142 (1971)
2. Abian, A.: A fixed point theorem equivalent to the axiom of choice. Arch. Math. Log. 25,
173–174 (1985)
3. Andres, J., Pastor, K., Snyrychova, P.: Simple fixed point theorems on linear continua. Cubo
10, 27–43 (2008)
4. Blair, C., Roth, A.E.: An extension and simple proof of a constrained lattice fixed point theorem.
Algebra Universalis 9, 131–132 (1979)
5. Bourbaki, N.: Sur le theoreme de Zorn. Arch. Math. 2, 434–437 (1949–1950)
6. Chitra, A., Subrahmanyam, P.V.: Remarks on a nonlinear complementarity problem. J. Optim.
Theory Appl. 53, 297–302 (1987)
7. Davis, A.: A characterization of complete lattices. Pac. J. Math. 5, 311–319 (1955)
8. Friedman, A.: Partial Differential Equations of Parabolic Type. Prentice-Hall Inc., Englewood
Cliffs (1964)
9. Fujimoto, T.: Nonlinear complementarity problems in a function space. SIAM J. Control Optim.
18, 621–623 (1980)
10. Kelley, J.L.: General Topology. Springer, Berlin (1975)
11. Knaster, B.: Un theoreme sur les functions d’ensembles. Ann. Soc. Polon. Math. 5, 133–134
(1927–1928)
12. Merrifield, K., Stein, J.D.: Common fixed points of two isotone maps on a complete lattice.
Czechoslov. Math. J. 49(124), 849–866 (1999)
References 71
13. Moroianu, M.: On a theorem of Bourbaki. Rend. Circ. Math. Palermo 20, 139–142 (1971)
14. Roth, A.E.: A lattice fixed-point theorem with constraints. Bull. Am. Math. Soc. 81, 136–138
(1975)
15. Schäfer, U.: An existence theorem for a parabolic differential equation in ∞ (A) based on the
Tarski fixed point theorem. Demonstr. Math. 30, 461–464 (1997)
16. Stein, J.D.: Fixed points of inequality preserving maps in complete lattices. Czechoslov. Math.
J. 49(124), 35–43 (1999)
17. Stein, J.D.: A systematic generalization procedure for fixed point theorems. Rocky Mt. J. Math.
30, 735–754 (2000)
18. Tarski, A.: A lattice theoretical fix point theorem and its applications. Pac. J. Math. 5, 285–309
(1955)
19. Zermelo, E.: Neuer Beweis fur die Moglichkeit einer Wohlordnung. Math. Ann. 15, 107–128
(1908)
20. Zorn, M.: A remark on method in transfinite algebra. Bull. Am. Math. Soc. 41, 667–670 (1935)
Chapter 4
Partially Ordered Topological Spaces
and Fixed Points
A partial order on a set induces a natural topology on this set, and special properties of
the partial order influence this topology significantly. These aspects lead to new and
interesting fixed point theorems. The interconnections among partial order, topology
and fixed point property were systematically investigated by Wallace [11], Ward [12]
and Manka [6]. This chapter highlights these contributions to fixed point theory and
supplements the theorems detailed in the preceding chapter.
4.1 A Precis of Partially Ordered Topological Spaces
This section is a precis of the fundamental contributions of Ward Jr. [12], although
Wallace [11] had already pointed out the importance of partial order and the induced
topology for fixed point theorems. Using the definition of a quasi-ordered set (see
Definition 1.1.10), we define certain sets as in the following.
Definition 4.1.1 Let (X, ≤) be a quasi-ordered set (viz. ≤ is a reflexive, transitive

binary relation on X ). For A ⊆ X , we define
L(A) = {y ∈ X : y ≤ x for some x ∈ A},

M(A) = {y ∈ X : x ≤ y for some x ∈ A},
E(A) = L(A) ∩ M(A).
L(A) is called the set of predecessors of A and M(A), the set of successors of A.
Remark 4.1.2 Clearly A ⊆ E(A), as A ⊆ L(A), M(A).
Definition 4.1.3 Let (X, ≤) be a quasi-ordered set and A ⊆ X . If A = L(A)(M(A)),

A is said to be monotone decreasing (monotone increasing) or simply decreasing
(increasing).
https://doi.org/10.1007/978-981-13-3158-9_4
74 4 Partially Ordered Topological Spaces and Fixed Points
Remark 4.1.4 If (X, ≤) is a quasi-ordered set and A ⊆ X , and A is decreasing

(increasing), then X − A is increasing (decreasing). Also, the union (intersection) of
a family of increasing (decreasing) sets is increasing (decreasing). Further, for each
subset A of a quasi-ordered set X , M(A) (L(A)) is increasing (decreasing).
Definition 4.1.5 Let (X, T ) be a topological space with a quasi-order ≤. This quasi-
order is said to be lower semicontinuous (upper semicontinuous) provided for a ≤ b
(b ≤ a) in X , there exists an open set U with a ∈ U such that for x ∈ U , x ≤ b
(b x). The quasi-order is said to be semicontinuous if it is both lower and upper
semicontinuous. The quasi-order is called continuous, if whenever a b, there are
open sets U and V in X with a ∈ U , b ∈ V and for x ∈ U and y ∈ V , x y.
A quasi-ordered topological space QOTS for short is a topological space with a
semicontinuous quasi-order. If the quasi-order is a partial order, then the correspond-
ing quasi-ordered topological space is called a partially ordered topological space or
POTS, for short.
Remark 4.1.6 Birkhoff [1] may be referred for a detailed discussion of quasi-order.
The definitions and remarks given above are from Ward [12] and Wallace [11].
For (x1 , y1 ), (x2 , y2 ) belonging to the Euclidean space R2 a partial order can
be defined by (x1 , y1 ) (x2 , y2 ) if x1 ≤ x2 and y1 ≤ y2 . (R2 , ) with the Euclidean
topology is a partially ordered topological space (POTS).
The proof of the following proposition is straightforward and is left as an exercise.
Proposition 4.1.7 Let (X, ≤) be a quasi-ordered set. It is QOTS if and only if for
each x ∈ X . L(x) and M(x) are closed subsets of X . Hence in a QOTS for each x,
E(x) is a closed set.
The following lemmata are also needed in the sequel.
Lemma 4.1.8 Let X be a topological space with a quasi-order. Then the following
are equivalent:
(i) the quasi-order is continuous;
(ii) the graph of the quasi-order is closed in X × X ;
(iii) for a b in X , there exist disjoint neighbourhoods N and N of a and b, respec-
tively, such that N is increasing and N is decreasing.
Proof (i) =⇒ (ii). Let G be the set {(a, b) ∈ X 2 : a ≤ b}. Suppose (a, b) ∈ / G.
Since a b and the quasi-order is continuous by Definition 4.1.5, there are open
sets U and V with a ∈ U , b ∈ V such that for (x, y) ∈ U × V , x y. So U × V ⊆
X 2 − G and (a, b) ∈ U × V , an open set containing an arbitrary element (a, b) of
X 2 − G. So X 2 − G is open and G is closed in X 2 .
(ii) =⇒ (i). Let G, the graph of the quasi-order be closed in X 2 . Let a b,
where a, b ∈ X . Since (a, b) ∈ X 2 − G and X 2 − G is open, there exist open neigh-
bourhoods U and V of a and b, respectively, in X such that U × V ⊆ X 2 − G.
For x ∈ U and y ∈ V , x ≤ y would imply that (x, y) ∈ G, contradicting that
(x, y) ∈ U × V ⊆ X 2 − G. Hence x y. So the quasi-order is continuous.
4.1 A Precis of Partially Ordered Topological Spaces 75
(i) =⇒ (iii). Let a, b ∈ X and a b. Then N = X − L(b) and N = X − M(a)

are open sets containing a and b, respectively. While N is increasing, N is decreasing.
If x0 ⊆ N ∩ N , then x0 x0 , a contradiction. So N ∩ N = φ.
(iii) =⇒ (i). Let a b. Then there exist disjoint neighbourhoods N and N of
a and b such that N is increasing and N is decreasing. Let x ∈ N and y ∈ N . If
y ∈ N ∩ N contradicting that N and N are disjoint. So x y.
Lemma 4.1.9 A continuous quasi-order is semicontinuous. A partially ordered

topological space is a T1 space and a partially ordered topological space with con-
tinuous partial order is a Hausdorff space.
We merely prove the last part of this lemma, leaving the other parts as exercises.
Proof Let (X, ≤) be a POTS with a continuous partial order. Let a and b be two
distinct points. If a b, by Lemma 4.1.8, there are disjoint neighbourhoods of a and
b. If a ≤ b then b a and again by Lemma 4.1.8, there are disjoint neighbourhoods
of b and a.
Lemma 4.1.10 If X is a QOTS with a linear quasi-order, then the quasi-order is

continuous.
Proof Since X is a QOTS, the quasi-order is semicontinuous. To prove the continuity

of the quasi-order, let a, b ∈ X and a ∈
/ b. So b ≤ a. So b ∈
/ E(a). If there exists c
such that b ≤ c ≤ a, then a ∈ U = X − L(c) and b ∈ V = X − M(c) and U and
V are open sets by Proposition 4.1.7. If x ∈ U and y ∈ V , then x c or c ≤ x and
c y or y ≤ c. So y ≤ x or x y.
If there is no such c with b ≤ c ≤ a, then U = X − L(b) and V = X − M(a)
are open sets containing a and b, respectively, and for x ∈ U and y ∈ V , x b or
b ≤ x and a y or y ≤ a. If x ≤ y, then b ≤ a and this contradicts the assumption
that there is not c with b ≤ c ≤ a. Hence x y.
The existence of maximal chains (maximal linearly ordered subsets) in a quasi-

ordered set was proved by Wallace [11], using Zorn’s lemma. Wallace [11] also
proved
Theorem 4.1.11 (Wallace [11]) Every maximal chain in a QOTS is a closed set.
Proof Let C be a maximal chain in a QOTS. Let x ∈ C. Then x ∈ L(x) ∪ M(x).

For c ∈ C, x ≤ c or c ≤ x. If x ≤ c, then c ∈ M(x) and
if c ≤ x, then c ∈ L(x). So
C ⊆ L(x) ∪ M(x) for each x ∈ C. Hence C ⊆ D = {L(x) ∪ M(x)}. If x ∈ D,
x∈C
then x ≤ x or x ≤ x for each x ∈ C. If x ∈
/ C, then x ∪ C is a chain containing C
properly, contradicting the maximality of C. Hence x ∈ C or D ⊆ C. Thus C = D.
By Proposition 4.1.7 L(x) and M(x) and hence L(x) ∪ M(x) are closed sets. Hence
D (= C) is closed.
Definition 4.1.12 Let (X, ≤) be a quasi-ordered set. y ∈ X is called minimal (max-

imal) if x ≤ y (y ≤ x) in X implies y ≤ x (x ≤ y).
For the following basic theorem due to Wallace [11], a proof due to Ward is
sketched.
Theorem 4.1.13 Let X be a non-empty compact space with a lower (upper)

semicontinuous quasi-order ≤. Then X has a minimal element.
Proof Let L = {L(x) : x ∈ X }. L can be partially ordered with respect to set inclu-
sion. Clearly, if {L(xλ ) : xλ ∈ X, λ ∈ } is a chain in L, L(xλ ) is the intersection
λ∈
of
a family of closed sets with finiteintersection property in the compact space X . So
L(xλ ) is non-empty. Let x0 ∈ λ∈ L(xλ ). So L(x0 ) is a lower bound for each
λ∈
L(xλ ), λ ∈ . So by Zorn’s Lemma L has a maximal element L(x0 ). If x ≤ x0 and
x = x0 , then L(x ) ≥ L(x0 ) contradicting the maximality of L(x0 ). So x ∈ L(x0 )
or x0 ≤ x . Thus x0 is a minimal element. The proof for the upper semicontinuous
case is similar and left as an exercise, with the hint that one has to consider the closed
set M(x) instead.
The above theorem leads to an interesting proof of a theorem, due to Moore, on the
existence of non-cutpoints of a continuum. A few requisite definitions are recalled.
Definition 4.1.14 A continuum is a compact connected Hausdorff space. A non-

degenerate continuum (viz. containing more than a point) is called indecomposable if
it is not a union of two of its proper subcontinua. A property is said to be hereditary for
a continuum if each of its non-degenerate subcontinua has this property. A continuum
is called unicoherent if for each representation of it as a union of two subcontinua,
the common part of these subcontinua is connected.
Definition 4.1.15 Let X be a connected topological space. A point c ∈ X is called

a cutpoint if X − {c} is disconnected. A non-cutpoint of X is an element which is
not a cutpoint.
Remark 4.1.16 Generally, metrizable continua are studied in detail. Apart from those
continua defined in Definition 4.1.14, there are various classes of continua such as
tree-like continua, Peano continua and so on. Fixed point property for continuous
functions on such continua is a topic of active research.
For [a, b] in R, every interior point is a cutpoint while no point of the rectangle
[a, b] × [c, d] or a circle is a cut point.
The machinery of POTS developed so far can be used to prove the following
theorem due to Moore. The proof is due to Ward [12].
Theorem 4.1.17 (Moore, See Wilder [16]) A non-degenerate continuum has at least
two non-cutpoints.
Proof For the non-degenerate continuum X let N be the set of non-cut points. Sup-
pose N has at most one point. So there exists x0 ∈ X − N . As x0 is a cutpoint,
X − {x0 } = A ∪ B where A and B are non-empty separated sets. Without loss of
generality we can assume that N ⊆ B. So each point of A is a cut point. For each
x ∈ A, there is a decomposition X − {x} = A(x) ∪ B(x) where A(x) and B(x) are
non-void separated sets with x0 ∈ B(x) and so A(x) ⊆ A. We can define a partial
order on A by x ≤ y if and only if A(x) ⊆ A(y). Since A ∩ B(x) = φ, x ∈ A − A
cannot be a non-cutpoint different from x0 , possibly the only non-cutpoint lying in B.
So L(x) = A(x). Hence this partial order is lower semicontinuous. Since X is com-
pact, A is compact. So by Theorem 4.1.13, there is a minimal element p ∈ A. Since
p is minimal, A( p) is empty. Otherwise suppose q ∈ A( p). Then q ∈ A( p) = L( p).
So q ≤ p. Since p is a minimal element of A, q = p. So p ∈ A( p) and this contra-
dicts the fact that X − { p} = A( p) ∪ B( p). So A( p) is empty. This again contradicts
the construction that A(x) is non-empty for all x ∈ A. Hence our assumption, that
there is only one non-cutpoint, viz. x0 is wrong. So a non-degenerate continuum has
at least two non-cut points.
The concept of convexity given below is useful for the study of fixed point theo-
rems in QOTS.
Definition 4.1.18 A subset A of a QOTS is called convex if A = E(A). X is said to
be quasi-locally convex if for x ∈ X and E(x) ⊆ U , an open set, there is a convex
open set V such that E(x) ⊆ V ⊆ U . X is called locally convex, provided, for x ∈ X
and U an open set, there is a convex open set V with x ∈ V ⊆ U .
The next theorem due to Ward [12], an extension of a theorem of Nachbin on
compact POTS, is stated without proof.
Theorem 4.1.19 Let X be a compact QOTS with a continuous quasi-order and
b a, where a, b ∈ X . Then we can find a continuous order-preserving map f :
X → [0, 1] such that f (a) = 0 and f (b) = 1.
Using the above theorem, Ward proved
Theorem 4.1.20 A compact Hausdorff QOTS with a continuous quasi-order is
quasi-locally convex.
Proof Let X be a compact Hausdorff QOTS, x ∈ E(x) ⊆ U , where U is an
open subset of X . For t ∈ X − U , t x or x t. If t x, then by Theorem
4.1.19 above there is a continuous order-preserving map f t : X → [0, 1] with
f t (x) = 0 and f t (t) = 1. Let Ut = {y ∈ X : f t (y) < 21 }. Then Ut is a decreasing
open set containing x and t ∈ / U t . If x t, then Ut = {y ∈ X : 21 < gt (y)} is an
increasing open set containing x and t ∈ / U t , where gt : X → [0, 1] is a contin-
uous order-preserving map with gt (t) = 0 and gt (x) = 1, as guaranteed by The-
n
orem 4.1.19. Thus X − U ≤ U {X − U t : t ∈ X − U } with X − U ≤ X − U ti .
i=1

n
U= Uti is an open set containing x and disjoint from X − U . Since E(x) ⊆ Uti
i=1
as Uti is either increasing or decreasing, x ∈ E(x) ⊆ V ⊆ U . So X is quasi-locally
convex.
Noting that a POTS with a continuous partial order is Hausdorff, we have
Corollary 4.1.21 A compact POTS with a continuous partial order is locally convex.
We proceed to study the convergence and clustering of nets in QOTS.
Definition 4.1.22 Let (X, ≤) be a quasi-ordered set and {xλ , λ ∈ D} be a set in X

where (D, ≤1 ) is a directed set (xλ ) is said to be monotone increasing (decreasing)
if xλ ≤ xμ for λ ≤1 μ (μ ≤1 λ) in D.
Definition 4.1.23 Let X be a topological space. A net (xλ , λ ∈ D), where (D, ≤1 )
is a directed set, is said to cluster at x0 ∈ X if for any open set U containing x0 and
λ ∈ D, there exists μ ∈ D with λ ⊆ μ such that xμ ∈ U .
Lemma 4.1.24 Let X be a compact Hausdorff QOTS with a continuous order ≤.

Then every monotone net in X clusters and the set of cluster points is contained in
E(x0 ) for some x0 ∈ X .
Proof Without loss of generality, we assume that the net is monotone increasing.
Since X is compact, every net in X has a subset converging to some x0 ∈ X , and
hence clusters at x0 . This is also true of a monotone increasing net (xλ ). By Theorem
4.1.20 X is quasi-locally convex. So given an open set U containing E(x0 ), there is
an open convex set V such that E(x0 ) ⊆ V ⊆ U . Since (xλ ) clusters at x0 , xλ0 ∈ V
for some λ0 . For λ0 ≤1 λ, there exists λ with λ ≤1 λ such that xλ ∈ V (Here the net
is indexed over the directed set (D, ≤1 )). Since V is convex and xλ ∈ V , xλ ∈ V for
all λ ≥ λ . Suppose x ∈/ E(x0 ). Then there are disjoint open sets U1 and U2 such that
x ∈ U1 and x0 ∈ U2 , as X is Hausdorff. So U1 and V ∩ U2 are disjoint and hence (xλ )
cannot cluster at x as otherwise U1 and V ∩ U2 would intersect. Hence the cluster
points of (X λ ) are contained in E(x0 ).
Corollary 4.1.25 If X is a compact POTS with continuous order, then every mono-
tone net in X is convergent.
The following lemmata can be proved easily.
Lemma 4.1.26 Let X be a topological space and f : X → X be a continuous map.

If for some x ∈ X , { f n (x) : n ∈ N} clusters at some x0 ∈ X , then it clusters at f (x0 ).
Lemma 4.1.27 Let f : X → X be a continuous map on a topological space X and

{xn : n ∈ N}, a sequence in X such that xn = f (xn+1 ). If {xn } clusters at x0 , then
{xn } clusters at f (x0 ).
The following theorem gives a necessary and sufficient condition for a continuous
order-preserving map f on a Hausdorff QOTS so that there exists x in X that can
compare with f (x). This theorem is quite useful in obtaining fixed point theorems
in the setting of QOTS.
Theorem 4.1.28 (Ward [12]) Let X be a Hausdorff QOTS with compact maximal
chains and f : X → X , a continuous order-preserving map. A necessary and suf-
ficient condition that there exists a non-empty compact set K ⊆ E(x0 ) for some
x0 ∈ X such that f (K ) = K is that there exists x in X such that x and f (x) are
comparable.
Proof The necessity is obvious. Suppose x and f (x) are comparable. Then { f n (x) :
n ∈ N} is a chain, as f is order-preserving and is therefore contained in a maximal
chain, which by hypothesis is compact. So by Lemma 4.1.24 it clusters at some x0 and
all its cluster points are contained in E(x0 ). By Lemma 4.1.26 f (E(x0 )) ⊆ E(x0 ).
Define K = ∩{ f n (E(x0 )) : n ∈ N}. Then K is a non-empty compact subset of E(x0 )
and f (K ) = K .
Corollary 4.1.29 If X is a POTS with compact maximal chains and f : X → X is
a continuous order-preserving map, then a necessary and sufficient condition that f
has a fixed point is that there is an x ∈ X for which x and f (x) are comparable.
Proof If x = f (x), then x and f (x) are comparable. Conversely if x ≤ f (x) or
f (x) ≤ x, then { f n (x) : n ∈ N} is a chain which is contained in a maximal chain
which also clusters at some x0 ∈ X . Since f (E(x0 )) ⊆ E(x0 ) and E(x0 ) = {x0 }, X
being partially ordered x0 = f (x0 ).
It is possible to define the concept of boundedness in quasi-ordered spaces, as in
the following.
Definition 4.1.30 Let (X, ≤) be a quasi-ordered set with an element e ∈ X such
that e ≤ x for all x ∈ X . A subset A of X is said to be bounded away from e if there
is y ∈ X − E(e) with A ⊆ M(y).
Using this concept, some results on fixed points can be obtained.
Theorem 4.1.31 (Ward [12]) Let X be a Hausdorff QOTS with compact maximal
chains and suppose there exists e ∈ X such that e ≤ x for all x ∈ X . Let f : X → X
be a continuous order-preserving map satisfying
(i) for some x ∈ X − E(e), x and f (x) are comparable;
(ii) for x satisfying (i), either { f n (x) : n ∈ N} is bounded away from e or there
exists y ∈ X with x ∈ E( f (y)) and f (y) ≤ y.
Then there exists x0 ∈ X − E(e) and a non-void compact set K ⊆ E(x0 ) such that
f (K ) = K .
Proof Suppose (i) is satisfied for some x ∈ X − E(e). If { f n (x) : n ∈ N} is bounded
away from e, define K = ∩{ f n (E(x0 )) : n ∈ N} where { f n (x) : n ∈ N} clusters at
x0 and all its cluster points are in E(x0 ) (by Lemma 4.1.24) as in Theorem 4.1.28.
For this choice of K , f (K ) = K and K ⊆ E(x0 ). Clearly x0 ∈ / E(e).
If for no x satisfying (i), { f n (x) : n ∈ N} is bounded away from e, by (ii) there
is a y1 such that for x ∈ E( f (y1 )). f (y1 ) ≤ y1 . Inductively, (yn ) can be chosen for
n ≥ 2 by
x ≤ f (y1 ) ≤ y1 ≤ f (y2 ) ≤ y2 ≤ · · ·
where each yn ∈ E( f (yn+1 )), n ≥ 1. Since {yn : n ∈ N} is a chain, it is contained in

a maximal chain which by hypothesis is compact. So by Lemma 4.1.24, {yn } clusters
at x0 and all its cluster points are contained in E(x0 ) for some x0 ∈ X − E(e). By
Lemma 4.1.27, f (E(x0 )) ⊆ E(x0 ). For K , as defined in the proof of Theorem 4.1.28,
by K = ∩{ f n (E(x0 )) : n ∈ N}, f (K ) = K , K ⊆ E(x0 ) with x0 ∈ X − E(e).
Corollary 4.1.32 Let X be a POTS with compact maximal chains such that for some
e ∈ X , e ≤ x for all x ∈ X and f : X → X be a continuous order-preserving map
satisfying the following:
(i) for some x ∈ X − E(e), x and f (x) are comparable and
(ii) if x satisfies (i), then { f n (x) : n ∈ N } is bounded away from e or there exists
y ∈ X such that x ∈ E( f (y)) and f (y) ≤ y.
Then f has a fixed point different from e.
Proof By Theorem 4.1.31, for some x0 = e, f maps a compact subset K of E(x0 )

into K . Since f is partially ordered E(x0 ) = {x0 } and thus f has a fixed point
different from x0 .
The proof of the following corollaries is easy and left as exercises.
Corollary 4.1.33 Let (X, ≤) be a POTS which is Hausdorff and having compact
maximal chains and f : X → X an order-preserving continuous map. Suppose (i)
for some u ∈ X , L(u) = X and (ii) for x, y ∈ X , there exists z with x ≤ z, y ≤ z.
Corollary 4.1.34 Let X be a compact Hausdorff QOTS such that

(i) L(u) = X for some u ∈ X ;
(ii) for x, y ∈ X , there exists z ∈ X with x ≤ z, y ≤ z;
(iii) for some e ∈ X , e ≤ x for all x ∈ X and X = E(e).
If f : X → X is an order-preserving continuous surjection, then there exists a non-
void compact set K ⊆ E(x0 ) for some x0 ∈ X − E(e) for which f (K ) = K .
Corollary 4.1.35 Let X , f be as in Corollary 4.1.34. If X is partially ordered, then

f has a fixed point different from e.
Remark 4.1.36 Ward Jr. [12] has given an example to show that in a compact POTS
an order-preserving continuous surjection may not have a fixed point different from
e, without an additional hypothesis such as (ii) in Theorem 4.1.31.
4.2 Schweigert–Wallace Fixed Point Theorem 81
4.2 Schweigert–Wallace Fixed Point Theorem
This section continues further the theory of partially ordered topological spaces
developed by Ward Jr. In particular, the concepts of end point and end element
and non-alternating maps and theorems relating to these are described. The section
culminates in the proof of Schweigert–Wallace fixed point theorem [9, 11] for home-
omorphisms on certain locally connected continua.
Definition 4.2.1 An element e of a topological space X is called an end point if,

whenever e ∈ U , an open set, there exists an open set V such that e ∈ V ⊆ V ⊆ U
and V − U is a singleton.
Two subsets P and Q of a connected topological space X are said to be separated
by a set K ⊆ X if X − K = A ∪ B and P ⊆ A, Q ⊆ B and A and B are separated.
If two points p and q of a topological space are not separated by any point, we write
p ∼ q.
Definition 4.2.2 A prime chain is a continuum which is either an endpoint, a cutpoint

or a non-degenerate set E containing a distinct pair of elements a and b with a ∼ b
with the condition that E = {x ∈ X : a ∼ x and x ∼ b}. An end element is a prime
chain E with the property that if E ⊆ U , an open set, then there is an open set V
such that E ⊆ V ⊆ U and V − V is a singleton.
Remark 4.2.3 R, with the usual topology has no endpoint, while [a, b] of real num-
bers (a < b) has a and b as endpoints. On the other hand, the unit circle has no
endpoint.
Lemma 4.2.4 Let X be a connected locally connected Hausdorff space. If E is an

end element of X , then E contains at most one cutpoint of X .
Proof Suppose E has two distinct cutpoints x1 and x2 of X . Then for i = 1, 2,

X − {xi } = Ai ∪ Bi , where Ai and Bi are separated and E − {xi } ⊆ Ai . Since X
is locally connected, we can take Ai to be connected with x2 ∈ A1 and x1 ∈ A2 .
Also Bi − B j = φ if i = j. Let yi ∈ Bi − B j (i = 1, 2, i = j) and let Ci be the
component of X − {xi } such that yi ∈ Ci . If C1 ∩ C2 = φ, C1 ∪ C2 is connected. As
x2 C1 , C1 ⊆ B2 contradicting that y1 ∈ B1 − B2 . So C1 ∩ C2 = φ.
Choose an open set U such that E ⊆ U and U intersects both C1 and C2 but
contains neither of these sets. Since E is an end element, there is an open set V such
that E ⊆ V ⊆ U and V − V is a singleton. So x1 ∈ V or x2 ∈ V . Otherwise both
x1 , x2 ∈ V − V , a contradiction. It readily follows that V − V meets both C1 and
C2 , a contradiction since C1 ∩ C2 = φ.
Lemma 4.2.5 If X is a connected, locally connected Hausdorff space and E is an

end element of X with a cutpoint x, then E − {x} and X − E are separated.
Proof Since E is a continuum, there is a component C0 in X − {x} containing E −

{x}. Since X is locally connected, it suffices to prove that C0 = E − {x} otherwise
there exists y ∈ C0 − E. Let C be any component of X − {x} distinct from C0 and P,

a connected open set such that x ∈ P, y ∈ / P and C − P = φ. Then C0 ∪ P − {y}
is an open set containing E. Since E is an end element, there is a connected open set
V (in view of local connectedness of X ) such that E ⊆ V ⊆ C0 ∪ P − {y} where
V − V is a singleton {z}. Since y ∈/ V and y ∈ C0 − E, z ∈ C0 and so x ∈ V . Since
x ∈ P, C − P = φ and C is a component of X − {x} distinct from C0 , the component
of X − {x} containing E − {x}, V − V ∩ C = φ. So z ∈ C. This contradicts that
z ∈ C0 and C and C0 are distinct (and hence disjoint) components. So C0 = E − {x}.
Further E − {x} and X − E are separated sets as E − {x} ⊆ E and so E − {x} ∩ X −
E = φ. If x0 ∈ X − E ∩ E − {x}, then every neighbourhood of x0 meets X − E.
Since X is locally connected, there is an open connected neighbourhood N of x0
(= x) in E containing points of X − E other than x. So N ∪ E − {x} would be a
connected subset of X − {x} containing E − {x} as a proper set contradicting that
E − {x} is a component in X − {x}. So E − {x} and X − E are separated.
Another useful concept in this context is that of a non-alternating map, defined

below.
Definition 4.2.6 A continuous map f : X → Y where X and Y are Hausdorff topo-

logical spaces is called non-alternating if for every decomposition, X − f −1 (y0 ) =
M ∪ N where M and N are separated in X , there is no y in Y such that f −1 (y)
intersects both M and N . Further f (X ) = Y .
A continuous surjection f : X → Y where X and Y are Hausdorff spaces is called
monotone if f −1 (y) is connected for each y in Y .
Remark 4.2.7 Clearly every monotone map is non-alternating.
We state below without proof, two results due to Wallace [10] for subsequent use.
Lemma 4.2.8 Let X and Y be Hausdorff topological spaces and f : X → Y , a

continuous surjection. The following are necessary and sufficient conditions for f
to be non-alternating for each decomposition X − f −1 (y0 ) = M ∪ N where M and
N are separated
(1) M = f −1 ( f (M)) and N = f −1 ( f (N ));
(2) f (M) ∩ f (N ) = φ;
(3) f −1 ( f (M)) ∩ f −1 ( f (N )) = φ.
Lemma 4.2.9 If f : X → Y is a closed and non-alternating map of X onto Y and

f −1 (y0 ) separates P and Q in X , then y0 separates f (P) and f (Q).
In a locally connected space with an end element, a quasi-order can be defined in

a natural way. This is captured in the following lemma and part of its proof can be
found in Whyburn [15].
Lemma 4.2.10 Let X be a locally connected continuum with an end element. Then
the relation ≤ defined on X by x ≤ y for x ∈ E, x = y or x separates E and y in X
is a semicontinuous quasi-order. If E is a singleton, then ≤ is a partial order.
Proof The proof that ≤ is a quasi-order and that L(x) is closed for each x ∈ X can
be found in Whyburn [15].
If x ∈ E, M(x) = X and is closed. For x ∈ X − E, M(x) = {x} ∪ {y : x sep-
arates E and y in X }. Let C be the component of X − {x} containing E, then
M(x) = X − C. As X is locally connected and X − {x} is open, C is open. So
M(x) is closed.
Lemma 4.2.11 Let X be a locally connected T2 continuum with end element E. If

f : X → X is non-alternating and f (E) ⊆ E, then f is order-preserving.
Proof Let x ≤ y, x, y ∈ X . For x = y, x ∈ E. Clearly f (x) ≤ f (y), since f (x) ∈

E. If x ∈/ E and x ≤ y, then x separates E and y. Since f (E) ⊆ E, f (x) ∈ X −
f (E) and f (x) = f (y). Then f −1 f (x) separates E and y. So by Lemma 4.2.9 f (x)
separates E and f (y) or f (x) ≤ f (y).
The next lemma is an important step towards proving Schweigert–Wallace Fixed

point theorem.
Lemma 4.2.12 Let X be a locally connected T2 continuum with an end element E.

If f : X → X is non-alternating and f (E) = E, then there is a cutpoint x of X such
that x and f (x) are comparable. Further x can be so chosen that for some y ∈ X
with x < y and x < f (y).
Proof If x ∈ E is a cutpoint of X , then as f (x) ∈ E, x and f (x) are comparable.

Since f (X ) = X and f (E) ⊆ E, there is an element y ∈ X − E such that f (y) ∈
X − E. So by Lemma 4.2.5, x < y and x < f (y).
If E has no cutpoint of X , choose y ∈ X − E with f (y) ∈ X − E. Since E is an
end element, there is an open set A with E ⊆ A ⊆ X − {y, f (y)} and A − A = {x}, x
a cutpoint, since X − {x} = X − A ∪ A. Clearly x separates E and {y} as also E and
f (y). So x < y, x < f (y). Also X − {x} = A ∪ B where A and B are separated with
E ⊆ A and {y, f (y)} ⊆ B. Also f (y) ∈ B ∩ f (B). If f (x) ∈ B, then x ≤ f (x) and
if f (x) ∈ E, then f (x) ≤ x. In both these cases the lemma is valid.
Suppose f (x) ∈ / B ∪ E. So f (x) ∈ A − E. In this case there are two possibilities.
Case (i): f (x) is not a cutpoint. So by Lemma 4.2.9, f −1 f (x) does not separate X .
So f −1 f (x) contains A or B. If B ⊆ f −1 f (x), then f (y) ∈ A, a contradiction and
when f −1 f (x) contains A, then f (x) ∈ E, a contradiction.
Case (ii): Suppose f is a cutpoint. In this case we prove that X − { f (x)} = f (A) −
{ f (x)} ∪ f (B) − { f (x)}, where f (A) − { f (x)} and f (B) − { f (x)} are non-void
and separated. Since E ⊆ A and f (x) ∈ X − E, f (E) ⊆ f (A) − { f (x)} = φ.
As f (x) = f (y) (∈ f (B)), f (B) − { f (x)} = φ. If t ∈ f (A) − { f (x)} ∩ f (B) −
{ f (x)}, then t ∈ f (A) ∩ f (B). Since f is non-alternating, t = f (x), a contradic-
tion. For similar reasons, f (A) − { f (x)} ∩ f (B) − { f (x)} is empty. If x ∈ f (B),
then f (x) ≤ x. If x ∈ f (A) − { f (a)}, then f (A) − { f (x)} intersects both A and B
and as f (x) ∈ A and B is connected, it follows that B ⊆ f (A) − { f (x)}. But this
implies that f (y) ∈ f (A) − { f (x)}, a contradiction.
Before proving another interesting theorem, similar to Theorem 4.1.31, we define

a concept of convergence for a sequence of subsets of a given set, treated for instance
in Whyburn [15].
Definition 4.2.13 ([15]) A sequence {An } of subsets of a topological space X is

said to cluster at a point x in X , if given any open neighbourhood of x infinitely
many members of the sequence {An } meet the given open neighbourhood. The set
of cluster points is denoted lim sup An . The sequence (An ) is said to converge to
x if all but a finite number of the An meet any given open neighbourhood of x.
The set of convergence points of An is denoted by lim inf An . (Clearly lim inf An ⊆
lim sup An .) If lim sup An = lim inf An , then this set is denoted lim An .
Theorem 4.2.14 Let X be a locally connected continuum with an end element E

and f : X → X be a monotone continuous map with f (X ) = X and f (E) = E.
Then X contains a non-empty subcontinuum K such that f (K ) = K and either K
is a cutpoint or K ⊆ X − E. Further, no point separates any pair of points of K in
X.
Proof By Lemma 4.2.10, X is a QOTS and by Lemma 4.2.11, f is order-preserving.

Lemma 4.2.12 insures that there is a cutpoint x ∈ X such that x and f (x) are
comparable. If x ∈ E, then by Lemma 4.2.5, X − {x} = E − {x} ∪ X − E and
X − E and E − {x} are separated. If f (x) = x, the theorem is proved; other-
wise f (x) ∈ E − {x} and so by Lemma 4.2.4 f (x) is not a cutpoint. So either
E ⊆ f −1 f (x) or X − E ⊆ f −1 f (x) as f is monotone. But E ⊆ f −1 f (x) contra-
dicts f (E) = E and X − E ⊆ f −1 f (x) contradicts f (X ) = X .
In the other case for consideration, x ∈ X − E. If x ≤ f (x), then by Theorem
4.1.31, there is a non-empty set K ⊆ E(x0 ), x0 ∈ X − E such that f (K ) = K . Since
E(x0 ) = {x0 }, the theorem is proved. If f (x) < x, let X − {x} = A ∪ B, where A and
B are separated, A being the component of X − {x} containing E. Then f (x) ∈ A. By
Lemma 4.2.12, x can be so chosen that there exists b ∈ X with b ∈ B, f (b) ∈ B. Note
that for any positive integer n, X − f −n (x) = f −n (A) ∪ f −n (B) where f −n (A)
and f −n (B) are separated. Since f is monotone f −1 (B) is connected and because
x ∈ f −1 (A), b ∈ B, f (b) ∈ B, f −1 (B) ⊆ B. So A ⊆ f −1 (A) and so for n < m,
f −n (A) ⊆ f −m (A). Since X is compact, lim sup f −n (x) = φ.
(i) lim sup f −n (x) = lim inf f −n (x). Otherwise there is x0 ∈ lim sup f −n (x) and
an open connected set U containing x0 such that if N is a positive integer with
f −N (x) ∩ U = φ, then there is m > N with f −m (x) ∩ U = φ. So U ⊂ f −m (A)
or U ⊆ f −m (B); since f −N (x) ∩ U = φ and f −N (x) ⊆ f −m (A), it follows that
U ⊆ f −m (A). So for p > m U ⊆ f −m (A) ⊆ f − p (A) so that U ∩ f − p (x) = φ,
contradicting the assumption that x0 ∈ lim sup f −n (x).
(ii) lim f −n (x) is a continuum. Clearly f −n (x) is closed in X and hence compact.
Suppose lim f −n (x) = P ∪ Q where P and Q are separated. Since X is normal
there exist open sets U and V which are disjoint and for which P ⊆ U and Q ⊆ V .
Suppose P = φ = Q, we can choose a positive integer N such that for all m ≥ N .
f −m (x) ∩ U = φ = f −m (x) ∩ V.
As each f −m (x) is connected, we can choose a sequence yn , n ≥ N with yn ∈

f −n (x) − U ∪ V . Clearly yn clusters at some y0 ∈ X − U ∪ V , contradicting that
lim f −n (x) ⊆ U ∩ V . So lim f −n (x) is connected and hence is a continuum.
(iii) No point of X separates any pair of points of lim f −n (x) in X . Otherwise there
exists a ∈ X such that the pair { p, q} ⊆ lim f −n (x) is separated by a. Thus X −
{a} = P ∪ Q where P and Q are separated with p ∈ P and q ∈ Q. Since f −n (x)
converges to both p and q, there is a positive integer M such that f −m (x) meets
both P and Q for m ≥ M. In view of (ii), a ∈ ∩{ f −m (x) : m ≥ M}. So f N +1 (a) =
f (x) = x. This contradicts our assumption that f (x) < x. Consequently, it follows
that f (lim f −n (x)) ⊆ lim f −n (x), so that

K = f i (L), where L = lim f −n (x)
i∈N
is a non-empty continuum and f (K ) = K . Further no point of X separates any

pair of points of K . As A ⊆ f −n (A), for each n = 1, 2, . . ., we conclude that K ⊆
X − A ⊆ X − E.
Corollary 4.2.15 In addition to the hypotheses of Theorem 4.2.14 above if E is a
singleton, then K ⊆ X − E.
The following fixed point theorem is due to Schweigert and Wallace.
Theorem 4.2.16 (Schweigert [9] and Wallace [11]) Let X be a locally connected
continuum with an end element E. If f (X ) = X and f : X → X is a homeomorphism
with f (E) = E, then f has a fixed point not in E.
Proof By Lemma 4.2.11, f and f −1 are order-preserving and if X has a cutpoint p,
then p and f ( p) are comparable by Lemma 4.2.12. If p ∈ E, then f ( p) is a cutpoint.
So, by Lemma 4.2.4, p = f ( p). If p ∈ X − E, then p ≤ f ( p) or f ( p) ≤ p. If
p ≤ f ( p), then by Theorem 4.1.31 there exists a non-empty compact set K with
K ⊆ E(x0 ), x0 ∈ X − E such that f (K ) = K . Since E(x0 ) = x0 , the theorem is
proved. If f ( p) ≤ p, by the same reasoning applied to f −1 , f −1 (and hence f ) has
a fixed point in X − E.
Remark 4.2.17 Ward Jr. has noted that for non-alternating maps, the above theorem
may not be true as shown by the following example. Let X be the locally connected
continuum of points (x, y, t) in R3 , satisfying x = 0 = y for 1 ≤ t ≤ 2 and x 2 +
y 2 = 1 − t 2 for 0 ≤ t ≤ 1. e = (0, 0, 2) is an endpoint. The map f : X → X defined
by
⎧
⎪
⎪ (1−2t)x (1−2t)y
⎨ − (x 2 +y 2 ) 21 , − (x 2 +y 2 ) 21 , 2t , 0 ≤ t ≤ 2 ,
1
f (x, y, t) =
⎪
⎪ (0, 0, t), 1
≤ t ≤ 1,
⎩ 2
(0, 0, 2), 1≤t ≤2
is continuous and monotone with e as the only fixed point. K = {(x, y, t) ∈ X : t =

0} is fixed under f and is a subcontinuum. Of course f is not one-to-one.
Remark 4.2.18 Schweigert [9] originally proved the theorem under the assumption
that X was separable and semi-locally connected. Wallace [11] noted that these
hypotheses could be dropped in preference to local connectedness and also pointed
to the use of quasi-order in the proof. Wallace, in fact, proved a fixed point theorem
for homeomorphisms T such that T and T −1 preserve a transitive, reflexive binary
relation on the space X .
Remark 4.2.19 Wallace [11] has shown that if X has a cutpoint, then it has an end
element. He also noted that prime chains in Peano space are precisely cyclic elements,
considered by Whyburn [15]. Wallace [11] has further proved that if the end element
E of a continuum X has no cutpoint of X , then X − E is connected and if P is the
union of all the end elements of X not containing a cutpoint of X , then X − P is
connected and each component of P is an end element.
Remark 4.2.20 It was already stated that fixed points theorems generally prove that
for certain topological spaces every continuous self-map has a fixed point. Bing (see
[8]) has noted that for the proof of numerous fixed point theorems depend on the
‘dead-end method’ or the ‘dog-chase rabbit argument’. Roughly speaking for a given
mapping f , as x moves in X , f (x) moves ‘ahead’ of x ‘relative to some hidden order
structure’ till a special feature of the underlying space is exploited to locate a point
x below f (x) in the order and corner f (x) in a dead end. Ward Jr. [14] has captured
this ‘dead-end method’ in the following theorem, which is related to Theorem 4.1.31
and its corollaries.
Theorem 4.2.21 Let X be a compact Hausdorff space with a lower semicontinuous

partial order , such that every maximal chain is compact. Suppose that if C is
a maximal chain and a ∈ C then a(= {x : (a, x) ∈ }) ∩ C is a closed set. If f
is a continuous, order-preserving self-map on X , then a necessary and sufficient
condition that for some x0 , X contains a fixed point of f is that there exists x ∈ x0
such that x ≤ f (x).
Proof The necessity is obvious. Suppose x0 ≤ x ≤ f (x). Since f is order-preserving,

clearly x ≤ f (x) ≤ f 2 (x) ≤ · · · ≤ f n (x) ≤ · · · , and so { f n (x)}n∈N lies is a com-
pact maximal chain C. So { f n (x)} has a cluster point y ∈ C. As f n (x) ∩ C is
closed for each n ∈ N, f n (x) ≤ y for n ∈ N. If for some z, f n (x) ≤ z < y for all
n ∈ N, then X − z is an open neighbourhood of y, which fails to have any f n (x).
Hence y is the supremum of { f n (x)}n∈N which converges to y. Since f is continuous
and the space is Hausdorff, y = f (y).
4.3 Set Theory, Fixed Point Theory and Order
In an insightful paper, Manka [6] described a connection between set theory and
fixed point theory via partial order. In this section, Manka’s approach is described,
based on the following definitions.
4.3 Set Theory, Fixed Point Theory and Order 87
Definition 4.3.1 A partial ordered set X is called inductive if for every totally
ordered subset of X there exists a least upper bound of this subset in X . Clearly
an inductive (partially ordered) set is non-empty. A partially ordered set X is called
acyclically ordered if for every p, r ∈ X with p ≤ r , the segment [ p, r ] = {x ∈ X :
p ≤ x ≤ r } is totally ordered.
Manka obtained a fixed point theorem for a class of mappings on an acyclically

inductively ordered sets.
Theorem 4.3.2 (Manka [6]) Let (X, ≤) be an inductively and acyclically ordered
poset and f : X → X be a map satisfying the following two conditions:
(i) p < f ( p) implies that for some q ∈ ( p, f ( p)] with q ≤ f (q);
(ii) q ≤ f (q) for all q ∈ Y ⊆ X implies that sup Y ≤ f (sup Y ).
Proof Since (X, ≤) is inductively ordered, P f = {q ∈ X : q ≤ f (q)} is also induc-

tively ordered. Since X is acyclically ordered, for each p ≤ f ( p) in X , there exists in
P f supremum of the set [ p, f ( p)] ∩ P f . Thus the map ϕ( p) = sup([ p, f (A)] ∩ P f )
sends P f into itself. Since p ≤ ϕ( p) for all p ∈ P f , it follows that there exists a fixed
point of ϕ.
We note that for each p ∈ P f , p = ϕ( p) implies p = f ( p). If p = f ( p), then
p < f ( p) since p ∈ P f . By (i) and the definition of P f , there exists q ∈ ( p, f ( p)] ∩
P f . So p < q and so q ≤ ϕ( p). So p < ϕ( p) for p ∈ P f and ϕ cannot have a fixed
point, in P f a contradiction.
Manka used the above theorem to prove that certain compact connected topologi-
cal spaces have fixed point property for continuous functions. We need the following
concepts.
Definition 4.3.3 A non-empty connected compact Hausdorff space is called a con-

tinuum.
(Generally, metrizable space is considered.)
Definition 4.3.4 A continuum having exactly two points which do not disconnect it
is called an arc, including continua with only one point.
Definition 4.3.5 A continuum X is said to be arcwise connected if for any pair of

points p, q ∈ X , there exists an arc joining these points in X . X is called one-arcwise
connected, if this arc is unique in X and will be denoted by pq.
Definition 4.3.6 Let X be an arbitrary one-arcwise connected continuum such that

for every monotone family of arcs apτ ⊆ X , τ ∈ T1 , there exists b ∈ X with
∪apτ = ab.
(Such a continuum X is some times called one-arcwise connected nested continuum.)

For the sake of completeness, we provide a list of definitions of various kinds of

continua.
Definition 4.3.7 A continuum X is said to be irreducible between the points a and

b if X contains a and b and no other subcontinuum of X contains both these points;
a (and b as well) is called a point of irreducibility.
Remark 4.3.8 a is a point of irreducibility for a continuum X if and only if X is not

the union of two proper subcontinua both of which contain a.
Definition 4.3.9 A continuum X is called unicoherent if the intersection of every

pair of continua whose union is X , is a continuum.
Definition 4.3.10 A continuum X is said to be decomposable, if it is a union of two

continua not contained in one another.
Definition 4.3.11 A property of a continuum X is called hereditary if every non-

trivial subcontinuum has that property (thus we have hereditarily unicoherent and
hereditarily decomposable continua).
Definition 4.3.12 A hereditarily unicoherent and arcwise connected metric contin-

uum is called a dendroid. By a λ-dendroid is meant a hereditarily unicoherent and
hereditarily decomposable metric continuum.
Example 4.3.13 A cone over an arbitrary hereditarily indecomposable plane contin-

uum is a one-arcwise connected nested continuum which is not hereditarily unico-
herent.
In what follows, we will show that arcwise connected hereditarily unicoherent

continua have fixed point property for continuous functions (see Manka [6], using
Theorem 4.3.2).
Let X be an arcwise connected nested continuum and a ∈ X . The binary relation
≤a defined by p ≤a q for p, q ∈ X whenever the arc ap ⊆ aq, is indeed a partial
order. With respect to this partial order a is the smallest element of X and a subset
{ pt ∈ X : t ∈ T } is totally ordered if and only if the family of arcs {apt : t ∈ T } is
monotone.
Lemma 4.3.14 Let X be an arcwise connected nested continuum. For a ∈ X ,

(X, ≤a ) is inductively ordered.

Proof Let {apt : t ∈ T } be a monotone family of arcs in X . So apt = ab. Clearly
t∈T
pt ≤a b. Suppose pt ≤a c for all t ∈ T . Then apt ⊆ ac for all t ∈ T . So apt ⊆ ac.
t∈T

As ac is closed, apt = ab ⊆ ac. So by definition of ≤a , b ≤ c. Thus { pt : t ∈ T }
t∈T
has c as an upper bound. Thus b is the supremum of { pt : t ∈ T } under this order.
4.3 Set Theory, Fixed Point Theory and Order 89

Remark 4.3.15 In Lemma 4.3.14, b = sup{ pt : t ∈ T } if and only if apt = ab.
t∈T

Clearly apt ⊆ ab for all t ∈ T so that apt ⊆ ab. But the closure of apt is a
t∈T t∈T
subarc ac of ab and apt ⊆ ac. So ac ⊆ ab. Hence c is an upper bound for { pt ∈ X :
t ∈ T }. Since b is the last upper bound, c = b.
In an arcwise connected continuum X , for every pair of arcs pq, pr with the same
initial point P, a binary relation ≺ can be defined by pq ≺ pr if pq ∩ pr = { p}.
In other words pq ≺ pr if pq ∩ pr is an arc non-degenerate to the point p. The
following lemma is left as an exercise.
Lemma 4.3.16 The binary relation ≺ is an equivalence relation in the family of all
non-degenerate arcs with the same initial point in X . If K is an arcwise connected
continuum of X and q, r ∈ K , then for p ∈/ K , pq ≺ pr as qr ⊆ K . Further ap ⊆
aq and pq ≺ pr imply ap ⊆ ar .
(For further details and related ideas Manka [5] may be consulted.)
Remark 4.3.17 ap ⊆ aq implies that p ∈ aq and ap ∪ pq = aq so that ap ⊆ aq

implies pq ⊆ aq.
The above ideas lead to the following.
Theorem 4.3.18 (Manka [6]) Every one arcwise connected nested continuum X with
the partial order ≤a is an inductively and acyclically ordered set. If f : X → X is
a map such that (a) f ( pq) is an arcwise connected continuum for each pq and (b)
if for each p = f ( p) there is { p} = pq ⊆ p f ( p) with pq ∩ f ( pq) = φ, satisfying
conditions (i) and (ii) of Theorem 4.3.2 in the order ≤a , then f has a fixed point.
Proof By Lemma 4.3.14, (X, ≤a ) is inductively ordered. If p ≤a q, then the interval

[ p, q] is the arc pq by the one-arcwise connectedness of X and the order ≤a in pq
is the natural order of the arc pq (defined in pq as ≤ p ). So (X, ≤a ) is acyclically
ordered.
We will verify that (i) and (ii) of Theorem 4.3.2 are satisfied. If ap a f ( p) so
that p = f ( p), by (b) there exists q ∈ X such that { p} = pq ⊆ p f ( p) and pq ∩
f ( pq) = φ. So ap aq ⊆ a f ( p). Now we show that aq ⊆ a f (q). Since p, q ∈ /
f ( pq) and f ( pq) is an arcwise continuum by (a), q f ( p) ≺ q f (q) and p f ( p) ≺
p f (q) (by Lemma 4.3.16). Since pq ≺ p f ( p) (by (b)), pq ≺ p f (q) by transitivity
of ≺. So ap ⊆ a f (q) by transitivity of ≺. So ap ⊆ a f (q) and consequently p f (q) ⊆
a f (q). Clearly, these lead to pq ⊆ a f (q). Therefore pq ⊆ a f (q). Thus q ∈ a f (q)
and aq ⊆ a f (q). Thus (i) of Theorem 4.3.2 is true.
Let {apt : t ∈ T } be a monotone family of arcs in X with apt ⊆ a f ( pt ), t ∈ T . Let
b ∈ X be the supremum in X with the order ≤a of the set { pt : t ∈ T }. So by Remark

4.3.15, let ab = apt . Suppose ab ⊆ a f (b) is not true. Clearly pt = b for all t ∈ T
t∈T
as apt ⊆ a f ( pt ). Then b ∈
/ a f (b) implying that ba ≺ b f (b) by Lemma 4.3.16. As

b = f (b) and bpt = {b}, bpt ⊆ b f (b) for bpt with sufficiently small diameter,
t∈T
there exists t ∈ T with bpt ∩ f (bpt ) = φ. So we can conclude that ab ∩ f ( pt b) = φ.
In the other case, since pt ∈ / f ( pt b), apt ∩ f ( pt b) = φ. Take the first point p
in the arc apt belonging to f ( pt b), the image being a continuum by (a). We would
have pt ∈/ ap. So pt ∈ / ap ∪ f ( pt b). But a f ( pt ) ⊆ ap ∪ f ( pt b). Hence pt ∈
/ a f ( pt )
contrary to the assumption that apt ⊆ a f ( pt ). So apt ∩ f ( pt b) = φ. Since the con-
tinua ab and f ( pt b) are disjoint, by the one-arcwise connectedness of X , there
exists in X a unique arc joining them so that ab ∩ pq = { p} and pq ∩ f ( pt b) = {q}
and any arc joining an arbitrary point of ab with an arbitrary point of f ( pt b) con-
tains pq. In particular p ∈ a f (b). As b ∈ / a f (b), p = b. However, ap ∪ pb = ab.
Taking an arc pt1 b ⊆ ab of smaller diameter than pt b and a suitable q1 ∈ f ( pt1 b)
instead of q we can suppose that ap ∩ pt b = φ without change of notation. But
pq ∩ pt b = φ from the choice of pq. Since q f ( pt ) ⊆ f ( pt b), q f ( pt ) ∩ pt b = φ.
Thus (ap ∪ pq ∪ q f ( pt )) ∩ pt b = φ. As a f ( pt ) ⊆ ap ∪ pq ∪ q f ( pt ), we infer that
pt b ∩ a f ( pt ) = φ, implying that pt ∈ / a f ( pt ), contradicting the definition of apt .
Thus (ii) of Theorem 4.3.2 is also satisfied. So f has a fixed point.
Corollary 4.3.19 An arcwise connected hereditarily unicoherent continuum has the
fixed point property for continuous maps.
Proof The corollary follows from the fact that a continuous self-map on such a
continuum satisfies (a) and (b) of Theorem 4.3.18.
The following example due to Manka [6] shows that Theorem 4.3.18 holds even
for discontinuous maps.
∞

Example 4.3.20 Let X be the plane continuum defined as In , where I0 is the
n=0
line segment joining (0, 0) to (1, 0) and In is the line segment joining (0, 0) to
(1, n1 ), n = 1, 2, . . .. Define the map f : X → X by f ( p) = p for p ∈ In , n ∈ N
p for p ∈ {0} × [0, 21 ]
and f ( p) = . Clearly, f satisfies all the conditions of
(n, 0) for p ∈ {0} × ( 21 , 1]
Theorem 4.3.18, though f is not continuous.
4.4 Multifunctions and Dendroids
In this section, we prove a fixed point theorem for dendroids following the techniques
developed by Manka, as described in Sect. 4.3. In fact, Manka’s fixed point theorem
subsumes that of Ward [13] and is proved in [7]. To this end, we need the following.
Definition 4.4.1 Let X be a topological space. A multifunction F : X → 2 X − {φ}
is called upper semicontinuous if F(x) is a closed set for each x ∈ X and F −1 (A) =
{x ∈ X : F(x) ∩ A = φ} is a closed subset of X for each closed subset A of X . F is
called lower semicontinuous, if F −1 (A) is open for each open subset A of X .
4.4 Multifunctions and Dendroids 91
Remark 4.4.2 For a compact metric space X , upper semicontinuity of F means that
limF(xn ) ⊆ F(lim xn ) for each convergent sequence (xn ) in X . Similarly, lower
semicontinuity of F : X → 2 X − {φ} means that F(lim xn ) ⊆ lim inf F(xn ) for
n n→∞
each sequence (xn ) converging in the compact metric space X . F is called continuous
if it is both lower and upper semicontinuous. Using the following version of Brouwer
reduction theorem, Manka [5] obtained an alternative proof of Ward’s fixed point
theorem for upper semicontinuous closed valued multifunctions on a dendroid.
Theorem 4.4.3 (Manka [5]) Every non-empty family P of closed subsets of a com-
pact metric space X which is closed with respect to the operation of closure of a
union of increasing sequences contains a maximal element.
Proof As X is a compact metric space, it has a countable base {Bn }. Take P0 ∈ P

and define a sequence Pn ∈ P inductively as follows: Take P1 ∈ P containing P0 and
intersecting B1 if such a P1 does not exist take P1 = P0 . Suppose P1 ⊂ P2 · · · ⊆ Pk−1
have been defined take Pk as an element of P containing Pk−1 and intersecting Bk .
If such a Pk does not exist, set Pk = Pk−1 .
The sequence Pk ∈ P defined thus has the property that for each Q ∈ P Bk ∩ Q =
φ and Pk−1 ⊆ Q imply that Bk ∩ Pk = φ. As {Pk } is increasing P = ∪Pk ∈ P by
definition of P. We claim that P is a maximal set in P. If for some Q ∈ P, Q contains
P properly, then there exists Bk , a basic set such that P ∩ Bk = φ and Bk ∩ Q = φ.
Since Pk−1 ⊆ Q, Bk ∩ Pk = φ. This contradicts that P ∩ Bk = φ. So P is maximal
in P.
We also recall the following results and their proofs are left as exercises.
Remark 4.4.4 Let X be an arcwise connected hereditarily unicoherent metric contin-

uum (dendroid) and F : X → 2 X − {φ} be an upper semicontinuous
multifunction
for which F(x) is a continuum for each x ∈ X . Then F(K ) = F(x) is a contin-
x∈K
uum, whenever K is a continuum. If F is continuous and F(x) is a closed subset of
X for each x ∈ X , then F(x) intersects every component of F(K ) whenever K is a
continuum, in X .
We use the following lemmata and assume that X is a hereditarily unicoher-

ent arcwise connected metric continuum (dendroid) and F : X → 2 X − {φ} is a
continuum-valued upper semicontinuous multifunction. We define Pa as the set of
all arcs a, b ⊆ X such that for p ∈ ab − {b} and each q ∈ F( p), the relation pq ≺ pb
holds (see the paragraph preceding Lemma 4.3.16 for definition of ≺).
Lemma 4.4.5 If a ∈
/ F(a), then for every d ∈ F(a), there exists ab ∈ Pa such that
ab ⊂ ad.
Proof Suppose d ∈ F(a) and that an arc ab ⊂ ad satisfies, by the upper semicon-
tinuity of F, ab ∩ F(ab) = φ. Now for each p ∈ ab − {b} and each q ∈ F( p), we
have p ∈ / F(ab) and d, q ∈ F(ab). Since F(ab) is a continuum (see Remark 4.4.4,
pq ≺ pd.
Lemma 4.4.6 If ab ∈ Pa , b ∈
/ F(b) and d ∈ F(b), then ab ⊂ ad.
Proof As b ∈/ F(b), it follows from the upper semicontinuity of F that for some p ∈
ab − {b}, p b ∩ F( p b) = φ. For each p ∈ p b − {b}, p ∈ ab − {b}. Hence ap ⊂

ab and pb ≺ pq for each q ∈ F( p) by the definition of Pa , By Lemma 4.3.16, ap ⊂

aq. Now p ∈ / F( p b) and q, d ∈ F( p b). Hence pq ≺ pd. So by Lemma 4.3.16,
ap ⊂ ad for every p ∈ p b − {b}. Since the union of all such arcs ap is ab − {b},
we have ab − {b} ⊂ ad. So ab ⊂ ad.
Lemma 4.4.7 If ab ∪ ac = ac, ab ∈ Pa and bc ∈ Pb , then ac ∈ Pa .
Proof For p ∈ ab − {b}, pc ≺ pb, since ab ⊂ ac. Since ab ∈ Pa , pb ≺ pq for each
q ∈ F( p). So pc ≺ pq, by the transitivity of ≺. If p ∈ bc − {c}, then pc ≺ pq, since
bc ∈ Pb .

Lemma 4.4.8 If ab = abn and abn ∈ Pa , n ∈ N, then ab ∈ Pa .
n∈N
Proof For p ∈ ab − {b}, there exists n ∈ N such that p ∈ abn − {bn }. So pbn ≺ pb
since abn ⊂ ab and pq ≺ pbn for every q ∈ F( p), in view of abn ∈ Pa . From the
reflexivity of ≺, we have pq ≺ pb.
Theorem 4.4.9 (Ward Jr. [13]) If F : X → 2 X − {φ} is a continuum-valued upper
semicontinuous multifunction on a dendroid X , then F has a fixed point.
Proof (As in Manka [7]) If ab1 ⊂ ab2 ⊂ · · · is an increasing sequence of arcs in the

dendroid X , then abn is an arc ab for some b. Since abn is a continuum, for
n∈N n∈N
a proper subcontinuum this continuum containing a some b j will not be a member.

So abn is a continuum which is not the union of two proper subcontinua both
n∈N

containing a. So a is a point of irreducibility of abn . Thus abn = ab for some
n∈N n∈N
b. So by Lemmata 4.4.5, 4.4.8 and Theorem 4.4.3, for a ∈
/ F(a), there exists an arc ab
maximal in Pa . Now from Lemmata 4.4.5–4.4.7 it follows that b ∈ F(b). Otherwise,
if b ∈
/ F(b) then ab ⊂ ad for each d ∈ F(b) and by Lemma 4.4.5 there exists bc ⊂
bd such that bc ∈ Pb . So ab ∪ bc = ac ∈ Pa by Lemma 4.4.7 contradicting the
maximality of ab in Pa .
Corollary 4.4.10 (Borsuk [2]) Every dendroid has fixed point property for contin-
uous functions.
4.5 Some Spaces with Fixed Point Property
In this section, elementary methods of constructing spaces with fixed point property
are described. Relevant concepts are also presented.
4.5 Some Spaces with Fixed Point Property 93
Definition 4.5.1 Let X be a topological space and Y ⊆ X . Y is said to be a retract

of X if there is a continuous map T from X onto Y such that r (y) = y for all y ∈ Y .
r is called a retraction of X onto Y .
Proposition 4.5.2 If X has fixed point property (for continuous functions), then any
retract of X also has fixed point property.
Proof Let g : Y → Y be continuous and r : X → Y be a retraction of X onto Y .

Then g ◦ r maps X into Y ⊆ X . Since g ◦ r is continuous and X has the fixed point
property for continuous functions, there exists x0 ∈ X such that g(r (x0 )) = x0 and
x0 ∈ Y . Since x0 ∈ Y and r is a retraction of X onto Y r (x0 ) = x0 . So g(x0 ) = x0 ∈ Y .
So Y has fixed point property for continuous functions.
Proposition 4.5.3 If X is a disconnected topological space, then X does not have

the fixed point property.
Proof Since X is disconnected X = A ∪ B where A and B are non-empty disjoint

proper closed subsets of X . Let a ∈ A and b ∈ B. The map f : X → X defined by
f (x) = b for x ∈ A and f (x) = a for x ∈ B is a continuous map without a fixed
point.
Remark 4.5.4 The unit circle S 1 in R2 with the usual topology is a connected, locally
connected compact metric space without fixed point property. For example, (x, y) →
(−x, −y) on S 1 has no fixed point.
Theorem 4.5.5 Let (X, d) be a compact metric space. Suppose for each > 0,
there is a continuous map f : X → X , where X is a subset of X with fixed point
property. If d( f (x), x) < for each x ∈ X , then X has the fixed point property.
Proof Let f : X → X be a continuous map. For each > 0, f ◦ f maps X into

itself and is continuous. Since X has fixed point property, f ◦ f (x ) = x for some
x ∈ X . Now d( f (x ), f ( f (x )) < by hypothesis. Setting = n1 , n ∈ N, it follows
from the compactness of X , that (xn k ) converges to an element x ∗ of X for some
subsequence (xn k ) of (xn ). From the continuity of f , and d( f (xn k ), xn k ) < n1k . We
conclude that f (x ∗ ) = x ∗ . Thus X has the fixed point property.
The proof of the following proposition is left as an exercise.
Proposition 4.5.6 Let X and Y be topological spaces such that Y is homeomorphic

to X . If X has the fixed point property, so has Y .
∞

Remark 4.5.7 The subset K = Ik ∪ I0 , where I0 is the line segment joining (0, 0)
n=1
and (0, 1) and In is the line segment joining (0, 1) to ( n1 , 0), n ∈ N in R2 has the fixed
point property. Similarly the sine circle {(x, sin x1 ) ∈ R2 0 < x ≤ 1} ∪ {(0, y) ∈ R2 :
−1 ≤ y ≤ 1} has the fixed point property. Theorem 4.5.5 may be used to prove these
conclusions.
Yet another useful idea is that of wedge of two spaces, defined below
Definition 4.5.8 Let X and Y be two disjoint spaces and let p ∈ X and q ∈ Y . The
wedge of X and Y at p and q, denoted by X ∨ p,q Y (or simply X ∨ Y ) is the quotient
space of X ∪ Y obtained by identifying p with q. (Clearly X ∨ Y has a natural copy
of X in X ∨ Y .)
Theorem 4.5.9 Let X and Y be T1 spaces with fixed point property. Then X ∨ Y
has the fixed point property.
Proof Let w denote the point at which X and Y intersect in X ∨ Y . Let f : X ∨ Y →

X ∨ Y be a continuous map. Define r : X ∨ V → X by
p, p ∈ X
r ( p) =
w, p ∈ Y
X and Y are closed in X ∨ Y and r is continuous. Since X has the fixed point property,
r ◦ f has a fixed point, p say. Now r f ( p) = p. If w = f (w), then there is nothing to
prove. So let w = f (w). So p = w. So r f ( p) = w. So by definition of r , f ( p) ∈ X .
So r f ( p) = f ( p) = p. (Note the continuity of r is based on the T1 -hypothesis).
For these and similar results and examples, Nadler [8] may be consulted.
4.6 An Example in Fixed Point Theory
Connell [3] had given examples of noncompact plane sets U, V and W each having
fixed point property such that cl W , U 2 lack fixed point property and V is locally
contractible. Klee [4] had given an example of a space combining all these features.
For the sake of completeness, we give the following definitions.
Definition 4.6.1 Let X and Y be topological spaces and I , the closed unit interval
[0, 1]. A homotopy in a continuous map h : X × I → Y . We write h t to denote the
map from X into Y defined by h t (x) = h(x, t) for all x ∈ X for any fixed t ∈ I . A
continuous map f : X → Y is said to be homotopic to a continuous map g : X → Y ,
if there exists a homotopy h : X × I → Y such that h 0 = f and h 1 = g.
Definition 4.6.2 A continuous map f : X → Y is null-homotopic or inessential if

f is homotopic to a constant map. A continuous map f : X → Y is essential if it is
not null-homotopic.
Definition 4.6.3 A topological space X is called contractible if the identity map on

X is null-homotopic.
We now briefly present Klee’s [4] example.

4.6 An Example in Fixed Point Theory 95
Example 4.6.4 Let Y be the set of real sequences (yn ) in the Hilbert space 2 such
that yi is non-zero for at most one i and 0 ≤ yi ≤ 1. If θ is the zero sequence in
2
and δn the sequence in 2 which is one in the nth place and zero elsewhere then
∞
Y = σn , where σn is the line segment joining θ to δn (and so σn = [θ, δn ]). Clearly,
n=1
Y is both contractible and locally contractible.
For each n, let rn be the retraction of Y onto σn which is identity in σn and maps
Y − σn onto θ . Let f : Y → Y be a continuous map. Suppose f (θ ) = θ . So for
some n, f (θ ) ∈ σn − {θ }. As rn f maps σn onto itself and rn f (θ ) = f (θ ) = θ and
σn being essentially a compact real interval has the fixed point property, rn f ( p) = p
for some p ∈ σn − {θ }. Since rn f ( p) = θ , f ( p) ∈ σn and so rn f ( p) = f ( p). So
f ( p) = p. Thus Y has the fixed point property.
In the space 2 × 2 , let P be the infinite polygon with vertices in the order
(θ, δ1 ), (δ1 , θ ), (θ, δ2 ), (δ2 , θ ), . . . , (θ, δn ), (δn , θ ), . . .. Clearly, P is closed in Y ×
Y and P is homeomorphic with [0, ∞). As Y × Y would admit a retraction onto P,
and P lacks the fixed point property, Y × Y cannot have the fixed point property, in
view of Proposition 4.5.2.
For each t ∈ [0, π ] and n ∈ N consider τn the arc consisting of all points
(xn (t), yn (t)) where xn (t) = (−1)n (1 + nt ) cos t and yn (t) = (1 + nt ) sin t. Each arc
τn has (1, 0) as an end point and X , the union of all the arcs τn is a homeomorphic
of y. But cl X contains the unit circle C and has a retraction onto C. But C does not
have the fixed point property. So cl X does not enjoy the fixed point property.
References
1. Birkhoff, G.: Lattice Theory. American Mathematical Society, Providence (1948)

2. Borsuk, K.: A theorem on fixed points. Bull. Acad. Pol. Sci. 2, 17–20 (1954)
3. Connell, E.H.: Properties of fixed point spaces. Proc. Am. Math. Soc. 10, 974–979 (1959)
4. Klee, V.: An example related to the fixed point property. Niew arch. voor Wiskunde 8, 81–82
(1960)
5. Manka, R.: Association and fixed points. Fundam. Math. 91, 105–121 (1976)
6. Manka, R.: Connection between set theory and the fixed point property. Colloq. Math. 43,
177–184 (1987)
7. Manka, R.: On fixed point theorems for multifunctions in dendroids. Colloq. Math. 42, 185–192
(1987)
8. Nadler, S.B.: The fixed point property for continua. Apartaciones Matematicas 30 (2005)
9. Schweigert, G.E.: Fixed elements and periodic types for homeomorphisms on S.L.C continua.
Am. J. Math. 66, 229–244 (1944)
10. Wallace, A.D.: Monotone transformations. Duke Math. J. 9, 487–506 (1943)
11. Wallace, A.D.: A fixed point theorem. Bull. Am. Math. Soc. 51, 413–416 (1945)
12. Ward Jr., L.E.: Partially ordered topological spaces. Proc. Am. Math. Soc. 5, 144–161 (1954)
13. Ward Jr., L.E.: A characterization of the fixed point property for a class of set-valued mappings.
Fundam. Math. 50, 159–164 (1961)
14. Ward Jr., L.E.: Monotone surjections having more than one fixed point. Rocky Mt. J. Math. 4,
95–106 (1974)
15. Whyburn, G.T.: Analytic Topology. American Mathematical Society, Providence (1942)
16. Wilder, R.L.: Topology of Manifolds. American Mathematical Society, Providence (1949)
Chapter 5
Contraction Principle
The contraction mapping principle proved independently by Banach [1] and Cac-
ciopoli [7] is a fundamental fixed point theorem, with an elementary proof. This
theorem has a wide spectrum of applications and is a natural choice in approximat-
ing solutions to nonlinear problems. According to Rall [18], the applications of the
contraction principle would fill volumes and Bollabos [4] calls it a doyen of fixed
point theorems. Charmed by both the simplicity and utility of this theorem, many
authors have generalized it in diverse directions. This chapter samples a few of these.
5.1 A Simple Proof of the Contraction Principle
A simple proof of the contraction principle due to Palais [16] is given below. This is
preceded by a few definitions and remarks.
Definition 5.1.1 Let (X, d) be a metric space. A map T : X → X is said to be
a Lipschitz map with Lipschitz constant M, if for some M ∈ R+ , d(T x, T y) ≤
Md(x, y) for all x, y ∈ X . In this case M is called a Lipschitz constant for the map
T . If M < 1, then T is called a contraction (mapping) with contraction constant M.
If M = 1, T is called a non-expansive map. If d(T x, T y) = d(x, y) for all x, y ∈ X .
T is a distance-preserving map and is called an isometry.
Remark 5.1.2 Every Lipschitz map T : (X, d) → (X, d) is uniformly continuous.

Also an isometry is non-expansive.
Definition 5.1.3 Let (X, d) be a metric space. A map T : X → X is called contrac-

tive if d(T x, T y) < d(x, y), whenever x, y ∈ X and x = y.
Palais [16] proved the contraction principle, using the following contraction
inequality.
https://doi.org/10.1007/978-981-13-3158-9_5
98 5 Contraction Principle
Lemma 5.1.4 (Contraction-Inequality) Let T : X → X be a contraction mapping

on the metric space (X, d) with contraction constant k. Then for any x, y ∈ X
1
d(x, y) ≤ [d(x, T x) + d(y, T y)]
1−k
Proof For x, y ∈ X , by triangle inequality
d(x, y) ≤ d(x, T x) + d(T x, T y) + d(T y, y)

≤ d(x, T x) + kd(x, y) + d(T y, y)
(as T is a contraction)
So (1 − k)d(x, y) ≤ d(x, T x) + d(y, T y) for all x, y ∈ X . Since 0 ≤ k < 1,

d(x, y) ≤ 1−k
1
[d(x, T x) + d(y, T y)] for all x, y ∈ X .
Corollary 5.1.5 A contraction mapping can have at most on fixed point.
This corollary follows at once from Lemma 5.1.4 by choosing x and y as fixed
points of T .
Lemma 5.1.6 (Estimate for iterates) Let (X, d) be a metric space and T : X → X ,
a contraction mapping with contraction constant k. Then for any x ∈ X ,
kn + km
d(T n x, T m x) ≤ d(x, T x)
(1 − k)
Proof Replacing x and y by T n x and T m x, respectively, in the contraction inequality

(vide Lemma 5.1.4), we get
d(T n x, T n+1 x) + d(T m x, T m+1 x)

d(T n x, T m x) ≤
(1 − k)
For j ∈ N,
d(T j x, T j+1 x) ≤ kd(T j−1 x, T j x)

≤ k j d(x, T x)
for all x ∈ X . It follows that
kn + km
d(T n x, T m x) ≤ d(x, T x)
(1 − k)
for all x ∈ X . As 0 < k < 1, it follows that {T n x} is a Cauchy sequence.
Theorem 5.1.7 (Contraction Principle) Let (X, d) be a complete metric space and
T : X → X , a contraction mapping with contraction constant k. Then, T has a
5.1 A Simple Proof of the Contraction Principle 99
unique fixed point x ∗ in X and every sequence T n x of T -iterates generated at any

element x ∈ X , converges to x ∗ . Further, for x ∈ X
kn
d(x ∗ , T n x) ≤ d(x, T x).
1−k
Proof For any x ∈ X , {T n x}, the sequence of T -iterates generated at x is a Cauchy

sequence, in view of Lemma 5.1.6. As (X, d) is a complete metric space (T n x)
converges to an element x ∗ . Since T is continuous, {T (T n (x))} = {T n+1 x} converges
to T x ∗ , {T n+1 x} being a subsequence of {T n x} must converge to x ∗ . Since the limit of
a convergent sequence is unique in a metric space x ∗ = T x ∗ . Thus every sequence of
iterates converges to a fixed point of T . By Corollary 5.1.5 this fixed point is unique.
For x ∈ X , according to Lemma 5.1.4,
km + kn
d(T m x, T n x) ≤ d(x, T x).
(1 − k)
Proceeding to the limit in the above inequality as m tends to ∞, and noting that
{T m x} converges to x ∗ = T x ∗ , we get
kn
d(x ∗ , T n x) ≤ d(x, T x).
1−k

∗
Remark 5.1.8 In order that T x is at a distance less than (> 0) from x the fixed
n
kn
point of T , it suffices to choose n such that 1−k d(x, T x) < . In other words for
log +log(1−k)−log d(x,T x) ∗
N> log k
, d(x , T x) < . Thus in a specific situation, the fixed
N
point of a contraction T can be approximated by T N x within an error of a pre-assigned

positive number by choosing N appropriately.
Remark 5.1.9 Since the closed sphere B(a : r ) = {x ∈ X : d(a, x) ≤ r } of a com-

plete metric space (X, d) is complete with respect to the restricted metric, a contrac-
tion T : X → X with contraction constant k maps B(a : r ) into itself and hence has
a unique fixed point in B(a : r ), provided d(a, T a) ≤ (1 − k)r .
We also have the following useful:

Corollary 5.1.10 Let (X, d) be a complete metric space and T : X → X be a map
such that T n is a contraction for some n ∈ N with n > 1. Then T has a unique fixed
point.
Proof Since T n is a contraction on the complete metric space (X, d), it has a unique
fixed point x ∗ , say. Now T n+1 (x ∗ ) = T (T n x ∗ ) = T (x ∗ ) = T (T n x ∗ ) = T n (T x ∗ ).
Thus T x ∗ is a fixed point of T n . Since T n has the unique fixed point x ∗ , x ∗ = T x ∗ .
If y ∗ is another fixed point of T , then T n (y ∗ ) = y ∗ will be a fixed point of T n and
hence y ∗ = x ∗ . Thus T has a unique fixed point.
Remark 5.1.11 A contractive map T which is not a strict contraction may not have
a fixed point in a complete metric space. For example, the map x → x + x1 maps
[2, ∞) into itself has no fixed point in [2, ∞) which is complete with respect to the
usual metric. For 2 ≤ x < y, 0 < (y + 1y ) − (x + x1 ) = (y − x)(1 − x1y ) < y − x
and consequently this map is contractive.
5.2 Metrical Generalizations of the Contraction Principle
It is natural to explore if maps satisfying inequalities similar to the Lipschitz condi-

tion have fixed points in a complete metric space. Kannan [13] proved a fixed point
theorem in this direction for a class of mappings which are not necessarily contin-
uous, though they satisfy a metrical inequality similar to the contraction condition.
Interestingly such (Kannan) maps have unique fixed points to which sequences of
iterates always converge. Hardy and Rogers [9] later extended Kannan’s theorem
to obtain a common generalization of the contraction principle and Kannan’s fixed
point theorem.
Theorem 5.2.1 (Hardy and Roger [9]) Let (X, d) be a complete metric space and
T : X → X a map satisfying the condition
d(T x, T y) ≤ a1 d(x, y) + a2 d(x, T x) + a3 d(y, T y) + a4 d(x, T y) + a5 d(y, T x)

5
for all x, y ∈ X , where ai ≥ 0 for i = 1, 2, . . . , 5 and ai < 1.
i=1
Then T has a unique fixed point and every sequence of T -iterates converges to
the unique fixed point.
Proof Interchanging the roles of x and y in the inequality satisfied by T , we get
d(T x, T y) ≤ a1 d(y, x) + a2 d(y, T y) + a3 d(x, T x) + a4 d(y, T x) + a5 d(x, T y).
So we get
a2 + a3
d(T x, T y) ≤ a1 d(x, y) + [d(x, T x) + d(y, T y)]
2
a4 + a5
+ [d(x, T y) + d(y, T x)] (5.2.1)
2
Letting y = T x in (5.2.1), we have
5.2 Metrical Generalizations of the Contraction Principle 101
a2 + a3
d(T x, T 2 x) ≤ a1 d(x, T x) + [d(x, T x) + d(T x, T 2 x)]
2
a4 + a5
+ [d(x, T 2 x)]
2
a2 + a3 + a4 + a5
≤ a1 d(x, T x) + d(x, T x)
2
a2 + a3 + a4 + a5
+ d(T x, T 2 x) (5.2.2)
2
upon using d(x, T 2 x) ≤ d(x, T x) + d(T x, T 2 x).

(a2 +a3 +a4 +a5 )
a1 + d(x,T x)
Thus d(T x, T 2 x) ≤ 2
(a +a +a +a ) .
1− 2 3 2 4 5
(a +a +a +a )
a1 + 2 3 2 4 5
Writing k = (a2 +a3 +a4 +a5 ) we get
1− 2
d(T x, T 2 x) ≤ kd(x, T x) for all x ∈ X (5.2.3)

5
Since 0 ≤ ai < 1, 0 ≤ k < 1. Further, d(T n x, T n+1 x) ≤ k n d(x, T x) for n ∈ N
i=1
and x ∈ X . For n, j ∈ N and x ∈ X

j
n
d(T x, T n+ j
x) ≤ d(T n+i−1 x, T n+i x)
i=1

j
≤ k n+i−1 d(x, T x)
i=1
n
k
≤ d(x, T x).
1−k
So {T n x} is a Cauchy sequence in the complete metric space X and so it converges

to an element x ∗ ∈ X .
Now
d(x ∗ , T x ∗ ) ≤ d(x ∗ , T n+1 x) + d(T n+1 x, T x ∗ )

a2 + a3
≤ d(x ∗ , T n+1 x) + a1 d(x ∗ , T x ∗ ) + [d(x ∗ , T x ∗ ) + d(T n x, T n+1 x)]
2
a4 + a5
+ [d(x ∗ , T n+1 x) + d(T x ∗ , T n x)]
2
a2 + a3
≤ d(x ∗ , T n+1 x) + a1 d(x ∗ , T x ∗ ) + [d(x ∗ , T x ∗ ) + d(T n x, T n+1 x)]
2
a4 + a5
+ [d(x ∗ , T n+1 x) + d(x ∗ , T x ∗ ) + d(T x ∗ , T n x)] (5.2.4)
2
Allowing n to tend to ∞ in (5.2.4) and noting that x ∗ = lim T n x = lim T n+1 x, we

get
(a2 + a3 + a4 + a5 + 2a1 )
d(x ∗ , T x ∗ ) ≤ d(x ∗ , T x ∗ )
2

5
Since 0 ≤ ai < 1, it follows that x ∗ = T x ∗ .
i=1
If y ∗ is also a fixed point of T , then
d(x ∗ , y ∗ ) = d(T x ∗ , T y ∗ ) (5.2.5)

∗ ∗ ∗ ∗ ∗ ∗
≤ a1 d(x , y ) + a2 d(x , T x ) + a3 d(y , T y )
+ a4 d(x ∗ , T y ∗ ) + a5 d(y ∗ , T x ∗ ) (5.2.6)
∗ ∗
≤ (a1 + a4 + a5 )d(x , y ). (5.2.7)

5
Since 0 ≤ a1 + a4 + a5 ≤ ai < 1, x ∗ = y ∗ . Thus T has a unique fixed point and
i=1
every sequence of T -iterates converges to the unique fixed point.
Corollary 5.2.2 (Kannan [13]) If T : X → X is a map on a complete metric space
(X, d) such that
d(T x, T y) ≤ k1 d(x, T x) + k2 d(y, T y)
for all x, y ∈ x with k1 , k2 ≥ 0 and k1 + k2 < 1, then T has a unique fixed point and
Proof Set a1 = a4 = a5 = 0, a2 = k1 and a3 = k2 in Theorem 5.2.1.
Remark 5.2.3 A mapping satisfying the conditions of Corollary 5.2.2 (or Theo-
rem 5.2.1) need not be continuous, as seen from the following example.

x
, 0 ≤ x ≤ 21
The map T : [0, 1] → [0, 1] defined by T x = x4 1 is a discontinuous
5
, 2 <x ≤1
map with 0 as the unique fixed point. For x, y ∈ [0, 1], it can be shown that
3
|T x − T y| ≤ [|x − T x| + |y − T y|].
8
Corollary 5.2.4 Theorem 5.1.7 (Contraction Principle).
Proof Set a1 = k, a2 = a3 = a4 = a5 = 0 in Theorem 5.2.1.
In another direction, Boyd and Wong [6] generalized the contraction principle by
majorizing d(T x, T y) by ψ(d(x, y)) instead of kd(x, y), imposing suitable assump-
tions on the real-valued function ψ of the real variable. In this context we recall the
following.
Definition 5.2.5 Let ψ : [a, ∞) → R be a function, where a ∈ R. ψ is said to be
upper semicontinuous from the right at c ∈ [a, ∞) if lim+ sup ψ(t) ≤ ψ(c).
t→c
Theorem 5.2.6 Let (X, d) be a complete metric space and T : X → X , a map such
that for all x, y ∈ X
d(T x, T y) ≤ ψ(d(x, y))
where ψ : P → [0, ∞) is upper semicontinuous from the right on P and ψ(t) < t
for all t ∈ P and t = 0, P being the range of d and P its closure in R+ . Then, T
has a unique fixed point and every sequence of T -iterates converges to this unique
fixed point.
Proof For x ∈ X , define cn = d(T n x, T n−1 x), n ∈ N with T 0 x = x. Clearly cn is
non-increasing and non-negative and hence converges to c ≥ 0. Since cn+1 ≤ ψ(cn )
for all n ∈ N, for c > 0,
c = lim cn = lim sup cn ≤ lim+ sup ψ(t) ≤ ψ(c),

t→c
a contradiction. Thus d(T n−1 x, T n x) converges to zero as n → ∞ for each x ∈ X .

We now show that {T n x} is a Cauchy sequence. Suppose {T n x} is not a Cauchy
sequence. Then, for some 0 > 0 and each k ∈ N, we can find natural numbers m(k)
and n(k) with m(k) > n(k) ≥ k such that for all k ∈ N
dk = d(T m(k) x, T n(k) x) ≥ 0
and
d(T m(k)−1 x, T n(k) x) < 0 .
This can be done by choosing m(k) as the least natural number exceeding n(k) for
which dk ≥ 0 . Now
dk = d(T m(k) x, T n(k) x) ≤ d(T m(k) x, T m(k)−1 x) + d(T m(k)−1 x, T n(k) x)

≤ cm(k) + 0 .
Thus 0 ≤ dk ≤ 0 + cm(k) . Consequently 0 ≤ lim inf dk ≤ lim sup dk ≤ lim

k→∞ k→∞ k→∞
sup 0 + cm(k) = 0 . Thus lim dk = 0 . Indeed dk → +
0 as k → ∞.
k→∞
Further,
dk = d(T m(k) x, T n(k) x) ≤ d(T m(k) x, T m(k)+1 x) + d(T m(k)+1 x, T n(k)+1 x)

+ d(T n(k)+1 x, T n(k) x)
≤ cm(k) + ψ(d(T m(k) x, T n(k) x)) + cn(k)
≤ 2ck + ψ(dk ) (as (c j ) is non-increasing).
Allowing k to tend to +∞ in the above inequality, it follows that
lim sup dk = lim dk = +

0 ≤ lim sup ψ(dk ) ≤ ψ(0 ).
k→∞ k→∞ k→∞
Since 0 > 0, this contradicts that 0 > ψ(0 ). Hence {T n x} is a Cauchy sequence
in X . As X is complete, it converges to an element x ∗ in X . Since for all x, y ∈ X ,
d(T x, T y) ≤ φ(d(x, y)) ≤ d(x, y), T is continuous. Since {T n+1 x} converges to
T x ∗ and is also a subsequence of {T n x}, it follows that x ∗ = T x ∗ . Since φ(t) < t
for all t > 0, it follows that the fixed point of T is unique.
Remark 5.2.7 Let X be (−∞, −1] ∪ [1, ∞) with the usual metric. Define T : X →
X by
− (x+1) , if x ≥ 1
T x = (1−x)2
2
, if x ≤ −1
and ψ : (0, ∞) → (0, ∞) by

t
, if t < 2
ψ(t) = 2
1+ t
2
, if t ≥ 2
Clearly |T x − T y| ≤ ψ(|x − y|). T has no fixed point, as ψ(2) = 2 < 2. Thus,

ψ(t) < t is not true for all t > 0 even though ψ is upper semicontinuous from the
right on (0, ∞). On the other hand, −T has both −1 and 1 as fixed points with
| − T x − (−T y)| ≤ ψ(|x − y|) for the same ψ. (In this case uniqueness of the fixed
point is lost).
Corollary 5.2.8 (Rakotch [17]) Let (X, d) be a complete metric space and T : X →
X an operator such that for all x, y ∈ X d(T x, T y) ≤ α(d(x, y))d(x, y), where
α : (0, ∞) → [0, 1) is a montonic decreasing function such that 0 ≤ α(t) < α(s)
for 0 < t < s. Then, T has a unique fixed point and every sequence of T -iterates
converges to the unique fixed point.
Proof Set ψ(t) = α(t)t for all t > 0 in Theorem 5.2.6. Since all the assumptions of
Theorem 5.2.6 are satisfied, the corollary follows.
In another direction, Jungck [12] obtained an extension of the contraction principle

as a common fixed point theorem.
Theorem 5.2.9 (Jungck [12]) Let f : X → X be a continuous mapping, (X, d)
being a complete metric space. Let g : X → X be a map such that f commutes with
g, g(X ) ⊆ f (X ) and for all x, y ∈ X there exists α ∈ (0, 1) such that
d(gx, g y) ≤ αd(x, y).
Then f and g have a unique common fixed point.
Proof As f is continuous and d(gx, g y) ≤ αd( f x, f y) for all x, y ∈ X , it follows

that g is continuous on X . As g(X ) ⊆ f (X ), given x0 ∈ X , we can find x1 ∈ X such
that f x1 = gx0 . Inductively, we can define a sequence xn ∈ X such that gxn−1 = f xn
for all n ∈ N. So d( f xn+1 , f xn ) = d(gxn , gxn−1 ) ≤ αd( f xn , f xn−1 ) ≤ αn d( f x1 ,
f x0 ). As 0 < α < 1, { f xn } = {gxn−1 } is a Cauchy sequence in X . Since X is com-

plete { f xn } converges to some x ∗ ∈ X . So {gxn } converges to x ∗ as gxn = f xn+1 .
As f and g are both continuous, {g f xn } converges to gx ∗ , while { f gxn } converges
to f x ∗ . Since g f xn = f gxn , for all n, f x ∗ = gx ∗ . Now
d(gx ∗ , ggx ∗ ) ≤ αd( f x ∗ , f gx ∗ )

= αd(gx ∗ , ggx ∗ )
Since 0 < α < 1, gx ∗ = ggx ∗ = g f x ∗ = f gx ∗ . Clearly gx ∗ is a common fixed point

of f and g. If a and b are two common fixed points of f and g, then
d(a, b) = d(ga, gb) ≤ αd( f a, f b) = αd(a, b).
Since 0 < α < 1, a = b. Thus f and g have a unique common fixed point.
Corollary 5.2.10 Let f and g be commuting mappings on a complete metric space
(X, d) into itself. Suppose f is continuous and g(X ) ⊆ f (X ). If for some α ∈ (0, 1)
and a positive integer k, d(g k x, g k y) ≤ αd(x, y) for all x, y ∈ X , then f and g have
a unique common fixed point.
Proof By Theorem 5.2.9, g k and f have a unique common fixed point a, say. Then
a = g k (a) = f a. So ga = g k ga = g f a = f ga, showing that ga is also a fixed point
for g k and f . By the uniqueness of the common fixed point for f and g k , it follows
that a = ga = f a.
Remark 5.2.11 In fact if f has a fixed point, then we can find a commuting map g with
a unique fixed point common with f , g(X ) = f (X ) and d(gx, g y) ≤ αd( f x, f y)
for all x, y ∈ X for some α ∈ (0, 1). This is readily seen by setting gx ≡ a, a fixed
point of f , and choosing any α ∈ (0, 1).
Example 5.2.12 f (x) = x 2 and g(x) = x 4 on [0, 21 ] satisfy all the assumptions of
Theorem 5.2.9 and 0 is the unique common fixed point.
5.3 Fixed Points of Multivalued Contractions
Nadler [15] generalized the contraction principle for multivalued functions, involving
the Hausdorff metric. We need the following
Definition 5.3.1 Let (X, d) be a metric space and C B(X ) be the set of all non-empty
closed bounded subsets of X . For C ∈ C B(X ) define
N (C, ) = {x ∈ X : d(x, c) < for some c ∈ C}

= B(c; ).
c∈C
For A, B ∈ C B(X ), define
H (A, B) = inf{ ∈ R : > 0, A ⊆ N (B, ) and B ⊆ N (A, )}.
Remark 5.3.2 H defines a metric on C B(X ), called the Hausdorff distance on the
space C B(X ). Further for x, y ∈ X , H ({x}, {y}) = d(x, y).
Nadler [15] proved a generalization of the contraction principle for mappings of
X into C B(X ), using the following definition and a lemma.
Definition 5.3.3 Let (X, d) be a metric space. A map F : X → C B(X ) is called
a multivalued contraction if there exists α ∈ (0, 1) such that for all x, y ∈ X ,
H (F x, F y) ≤ αd(x, y).
Lemma 5.3.4 Let (X, d) be a metric space and A, B ∈ C B(X ). Given > 0 and
a ∈ A, we can find b ∈ B such that d(a, b) ≤ H (A, B) + .
Proof Let r = H (A, B). For r = 0, the lemma is clear. For r > 0, by definition of
H (A, B), A ⊆ N (B, r + ). So for a ∈ A, there exists b ∈ B such that a ∈ B(b, r +
) or d(a, b) < r + = H (A, B) + .
Theorem 5.3.5 (Nadler) Let (X, d) be a complete metric space and F : X →
C B(X ) be a multivalued contraction with contraction constant α ∈ (0, 1). Then
F has a fixed point in X (i.e. an element x0 ∈ X with x0 ∈ F x0 ).
Proof For p0 ∈ X , since F( p0 ) ∈ C B(X ) for any p1 ∈ F( p0 ), for = α, it follows
from Lemma 5.3.4 above that there exists p2 ∈ F( p1 ) such that
d( p1 , p2 ) ≤ H (F( p0 ), F( p1 )) + α.
Since F( p1 ), F( p2 ) ∈ C B(X ) and p2 ∈ F( p1 ) there is a point p3 ∈ F( p2 ) insured

by Lemma 5.3.4 for = α2 with
d( p2 , p3 ) ≤ H (F( p1 ), F( p2 )) + α2
Thus inductively we can define a sequence of points { pi : i ∈ N} such that
d( pi , pi+1 ) ≤ H (F( pi−1 ), F( pi )) + αi
for i ≥ 1. Since F is a multivalued contraction, for all i ≥ 1,
d( pi , pi+1 ) ≤ H (F( pi−1 ), F pi ) + αi

≤ αd( pi−1 , pi ) + αi
≤ α[H (F pi−2 , F pi−1 ) + αi−1 ] + αi
≤ α2 d( pi−2 , pi−1 ) + 2αi
≤ αi d( p0 , p1 ) + iαi
5.3 Fixed Points of Multivalued Contractions 107
So

j
d( pi , pi+ j ) ≤ d( pi+k−1 , pi+k )
k=1

j
≤ (αi+k−1 d( p0 , p1 ) + [i + k − 1]αi+k−1 ).
k=1
∞
∞

As αn and nαn converge, it follows that { pn } is a Cauchy sequence in the
1 1
complete metric space (X, d) converging to an element p ∗ in X . Writing d(a, B) =
inf{d(a, b) : b ∈ B}, it follows that
d( p ∗ , F p ∗ ) ≤ d( p ∗ , F pn ) + H (F p ∗ , F pn )
≤ d( p ∗ , pn+1 ) + αd( p ∗ , pn )
(as pn+1 ∈ F pn and F is a
multivalued contraction).
This implies that d( p ∗ , F p ∗ ) = 0 as { pn } converges to p ∗ . Since F p ∗ is closed,

p ∗ ∈ F p ∗ . Thus F has a fixed point.
Corollary 5.3.6 The contraction principle (Theorem 5.1.7).
Proof The map x → {T x} maps the complete metric space (X, d) into C B(X )
and is a multivalued contraction, whenever T : X → X is a contraction, in view of
Remark 5.3.2. Hence by Nadler’s Theorem 5.3.5, this map has a fixed point, which
is clearly a fixed point T . The uniqueness can be proved independently.
Remark 5.3.7 A fixed point of the multivalued contraction, insured by Nadler’s the-
orem need not be unique. For example, for the map x → [0, 1] of R into C B(R),
every point of [0, 1] is a fixed point. This map, being a constant map, is clearly, a
contraction.
5.4 Contraction Principle in Gauge Spaces
We recall the following definition of a pseudometric on a non-empty set.

Definition 5.4.1 A map d : X × X → R+ is called a pseudometric if it satisfies the
following conditions:
(i) d(x, x) = 0 for all x ∈ X ;

(ii) d(x, y) = d(y, x) for all x, y ∈ X ;
(iii) d(x, y) ≤ d(x, z) + d(z, y) for all x, y, z ∈ X .
Remark 5.4.2 The map T : X → X on a metric space (X, d) induces a pseudometric

dT defined by dT (x, y) = d(T x, T y) on X in a natural way. dT is a metric on X if
and only if T is one-to-one mapping.
A family of pseudometrics on X defines a topology on X in a natural way, leading

to the concept of a gauge space.
Definition 5.4.3 ([8]) Let X be a non-empty set and G = {dλ : λ ∈ } be a family of
pseudometrics on X , where = φ. G is called a gauge on X . The family {Bdλ (x, ) :
dλ ∈ G, x ∈ X, > 0} is a sub-base for a topology on X , where Bdλ (x, ) = {y ∈ X :
dλ (x, y) < }. X with this topology is called a gauge
space. (Every neighbourhood of
x in this gauge space contains a set of the form Bdλ (x, λ ), where F is a non-void
λ∈F
finite subset of and λ > 0.)
Tan [20] generalized the contraction principle to gauge spaces, using the following
concepts.
Definition 5.4.4 A gauge {dλ : λ ∈ , = 0} is said to be separating on X if for
each pair x, y ∈ X with x = y, there exists μ ∈ with dμ (x, y) > 0.
Remark 5.4.5 Clearly a gauge space is Hausdorff if and only if the gauge is sepa-
rating.
Definition 5.4.6 Let D = {dλ : λ ∈ , = 0} be a gauge on X . A sequence (xn )

is called a Cauchy sequence in (X, D) if lim dλ (xm , xn ) = 0 for each λ ∈ . A
m,n→∞
gauge space X is said to be sequentially complete if every Cauchy sequence in X
converges to an element of X .
The following is Tan’s generalization [20] of the contraction principle to sequen-

tially complete Hausdorff gauge spaces.
Theorem 5.4.7 (Tan [20]) Let D = {dλ : λ ∈ , = φ} be a separating gauge
on X . Let T be a self-map on X such that for each λ ∈ , there exists cλ with 0 ≤
cλ < 1 such that dλ (T x, T y) ≤ cλ dλ (x, y) for all x, y ∈ X . If (X, D) is sequentially
complete, then T has a unique fixed point and every sequence of T -iterates converges
to the fixed point.
Proof Define a sequence {xn } iteratively in X by setting x1 = T x0 , for an arbitrary

x0 in X and defining xn+1 = T xn , n ∈ N. For each λ ∈ , n, p ∈ N,
5.4 Contraction Principle in Gauge Spaces 109

p
dλ (xn , xn+ p ) ≤ dλ (xn+k−1 , xn+k )
k=1
p
≤ dλ (T n+k−1 x0 , T n+k x0 )
k=1

p
≤ cλn+k−1 dλ (x0 , x1 )
k=1
j−1
(as dλ (x j−1 , x j ) ≤ cλ d(x0 , x1 ))
cλn
≤ dλ (x0 , x1 ).
1 − cλ
Thus lim dλ (xm , xn ) = 0 for each λ ∈ and so (xn ) is a Cauchy sequence in X .

m,n→∞
As (X, d) is sequentially complete, (xn ) converges to an element x ∗ ∈ X . Clearly
lim T (an ) = T (a), whenever (an ) converges to a in (X, D). So (T xn ) converges
n→∞
to T x ∗ . (T xn ) being (xn+1 ), (T xn ) must also converge to x ∗ . As D is a separating
family, X is Hausdorff in the topology induced by the gauge D. Since the limit
of a convergent sequence in a Hausdorff space is unique, x ∗ = T x ∗ . Thus T has
a fixed point. If y ∗ = T y ∗ , then dλ (x ∗ , y ∗ ) = dλ (T x ∗ , T y ∗ ) ≤ cλ dλ (x ∗ , y ∗ ). Since
0 ≤ cλ < 1 for all λ ∈ , it follows that x ∗ = y ∗ . Thus, the fixed point is unique and
Remark 5.4.8 The topology of a gauge space need not be metrizable. For example, let
X be the space of all mappings of R into itself. Then for f, g ∈ X (= RR ), dx defined
by dx ( f, g) = | f x − gx| is a pseudometric on X for each x ∈ R. D = {dx : x ∈ R}
is a separating family of pseudometrics on X and (X, D) is sequentially complete.
It can be seen that X is not first countable. Hence it cannot be metrized.
Corollary 5.4.9 The contraction principle (Theorem 5.1.7).
Proof If (X, d) is a complete metric space and T : X → X a contraction, then
T has a unique fixed point, as {d} is a separating sequentially complete gauge
on X .
5.5 A Converse to the Contraction Principle
In this section, a converse to the contraction principle due to Bessaga [3] is proved
following a simplified approach due to Jachymski [11].
Theorem 5.5.1 (Bessaga) Let X be a non-empty set and T : X → X be a map and
k ∈ (0, 1). Then
(a) there exists a metric d on X such that d(T x, T y) ≤ kd(x, y) for all x, y ∈ X ,
whenever T n has at most one fixed point for each n ∈ N ;
(b) if in addition some T n has a fixed point, X has a complete metric d such that
d(T x, T y) ≤ kd(x, y) for all x, y ∈ X .
Jachymski’s proof [11] is based on the following
Lemma 5.5.2 Let T : X → X be a map and k ∈ (0, 1). The following statements
are equivalent:
(i) there exists a complete metric d on X such that d(T x, T y) ≤ kd(x, y) for all
x, y ∈ X ;
(ii) there exists a function ϕ : X → R+ such that ϕ(T x) ≤ kϕ(x) for all x ∈ X and
ϕ−1 {0} is a singleton.
Proof (i) ⇒ (ii). Since by the contraction principle T has a fixed point x ∗ , the map ϕ
defined by ϕ(x) = d(x, x ∗ ) is such that ϕ(T x) = d(T x, T x ∗ ) ≤ kd(x, x ∗ ) = kϕ(x)
for all x ∈ X and ϕ−1 {0} = {x ∗ }, a singleton. Thus (ii) is true.
(ii) ⇒ (i). Define d : X × X → R+ by

ϕ(x) + ϕ(y) for x = y
d(x, y) =
0, for x = y.
Clearly d is a metric on X . d(T x, T y) = ϕ(T x) + ϕ(T y) ≤ kϕ(x) + kϕ(y) =

kd(x, y). So T is a contraction on X with Lipschitz constant k. Consider a Cauchy
sequence (xn ) in X . Without loss of generality, suppose that {xn : n ∈ N} is an infinite
set. Otherwise, (xn ) would have a constant subsequence and being Cauchy would
converge in X . So there is an infinite subsequence (xn k ) of distinct elements such
that
d(xn k , xn m ) = ϕ(xn k ) + ϕ(xn m ) for k = m.
So ϕ(xn k ) → 0 as k → ∞. Since ϕ−1 ({0} is a singleton {z} by hypothesis (ii),

ϕ(z) = 0. So d(xn k , z) = ϕ(xn k ) + ϕ(z) tends to zero as k → ∞. Thus lim xn k = z.
This implies that the entire Cauchy sequence (xn ) converges to z. In other words
(X, d) is complete.
We first prove part (b) of Theorem 5.5.1.

Proof (of part (b) of Theorem 5.5.1).
By assumption T n has a unique fixed point x ∗ . By uniqueness of the fixed point of
T n , x ∗ = T x ∗ . So by part (a) of Theorem 5.5.1 T m has the unique fixed point x ∗ for
each m ∈ N. Define
= {ϕ : ϕ is a map with domain Dϕ ⊆ X into R+ with ϕ−1 {0} = {x ∗ } and

T (Dϕ ) ⊆ Dϕ and ϕ(T x) ≤ ϕ(x) for all x ∈ Dϕ }
Clearly is non-empty, for ϕ1 : {x ∗ } → R+ defined by ϕ1 (x ∗ ) = 0 and Dϕ1 = {x ∗ },

belongs to . We can partially order by ϕ1 ϕ2 ⇔ Dϕ1 ⊆ Dϕ2 and ϕ2 | Dϕ1 = ϕ1 .
5.5 A Converse to the Contraction Principle 111

If 0 ⊆ is a chain under , then the set D = Dϕ is T -invariant and ψ : D →
ϕ∈0
R+ defined by ψ(x) = ϕ(x), x ∈ Dϕ is an upper bound for 0 . Thus by Zorn’s
Lemma there exists a maximal element ϕ0 : D0 → R+ in (, ). We now show that
D0 = X . Otherwise we can find x0 ∈ X − D0 . Set O(x0 ) = {T n−1 x0 : n ∈ N}.
Suppose O(x0 ) ∩ D0 = φ. Then the elements T n−1 (x0 ) for n ∈ N are all distinct.
Otherwise, T has a periodic point which necessarily is the unique fixed point x ∗ ,
implying that x ∗ ∈ O(x0 ), a contradiction as x ∗ ∈
/ D0 . Define Dϕ = O(x0 ) ∪ D0
and ϕ : Dϕ → R+ by

ϕ(x), x ∈ D0
Dϕ (x) = n−1
k , x = T n−1 (x0 ) ∈ O(x0 ), for n ∈ N
Clearly ϕ ∈ , ϕ0 ≤ ϕ and ϕ0 = ϕ, contradicting the maximality of ϕ0 . So O(x0 ) ∩

D is non-empty. Now, define m = min{n ∈ N : T n (x0 ) ∈ D}. Clearly T m−1 (x0 ) ∈ /
D0 . Define Dϕ = {T m−1 (x0 )} ∪ D0 . Then T (Dϕ ) = {T m x0 } ∪ T (D0 ) ⊆ D0 . Thus
Dϕ is T -invariant. Consider ϕ : Dϕ → R+ defined by ϕ|D0 = ϕ0 and ϕ(T m−1 (x0 )) =
1, if T m−1 (x0 ) = x ∗ and ϕ(T m−1 (x0 )) = k1 ϕ0 (T m (x0 )) for T m−1 (x0 ) = x ∗ . In both
these cases ϕ ∈ , ϕ0 ϕ and ϕ = ϕ0 , contradicting the maximality of ϕ0 . So by
Lemma 5.5.2, the proof of part (b) of Theorem 5.5.1 is complete.
The next lemma, not only helps to prove part (a) of theorem but even extends it.
Lemma 5.5.3 Let X be a non-void set and T : X → X , a map, with α ∈ (0, 1). The
following statements are equivalent.
(i) T has no periodic point;
(ii) the Schroder functional equation ϕ(T x) = αϕ(x) has a solution ϕ : X →
(0, ∞).
Proof We prove that (i) implies (ii). Define = {ϕ : Dϕ → (0, ∞)|Dϕ = φ, Dϕ ⊆

X, T (Dϕ ) ⊆ Dϕ and ϕ(T x) = αϕ(x) for x ∈ Dϕ }. Let x0 be a fixed element of X
and Dϕ1 = O(x0 ), ϕ1 being the map ϕ1 : O(x0 ) → (0, ∞) defined by ϕ1 (T n−1 (x0 )) =
αn−1 , n ∈ N. Clearly T is 1-1 on Dϕ1 as T has no periodic point in X by hypothesis.
Further ϕ1 (T x) = αϕ1 (x) on Dϕ1 . So ϕ1 ∈ and is non-empty. We can repeat
the argument used to prove part (b) of Theorem 5.5.1 involving Zorn’s Lemma to
ensure that there is a maximal element ϕ0 in (, ). Here, is the partial order
in defined by ϕ1 ϕ2 if Dϕ1 ⊆ Dϕ2 and ϕ2 | Dϕ1 = ϕ1 . We claim that Dϕ0 = X .
Otherwise there exists x0 ∈ X − D0 . If O(x0 ) ∩ Dϕ0 is empty, then T n−1 (x0 ) are all
distinct, as otherwise T would have a periodic point x0 . Define Dϕ = O(x0 ) ∪ Dϕ0 ,
ϕ| Dϕ0 = ϕ0 and ϕ(T n−1 (x0 )) = αn−1 for n ∈ N. Clearly ϕ ∈ , ϕ = ϕ0 and ϕ0 ϕ,
a contradiction. Hence O(x0 ) ∩ Dϕ0 is non-empty.
Define m = min{n ∈ N : T n (x0 ) ∈ Dϕ0 }. So T m−1 (x0 ) ∈ / Dϕ0 . Define Dϕ =
{T m−1
x0 } ∪ Dϕ0 . Clearly T (Dϕ ) ⊆ Dϕ . Define ϕ : Dϕ → (0, ∞) by ϕ| D0 = ϕ0 and
ϕ(T m−1 (x0 )) = ϕ0 (T α (x0 )) . It may be noted that T m−1 (x0 ) = T m (x0 ). Further ϕ ∈ ,
m
ϕ = ϕ0 and ϕ0 ϕ, contradicting the maximality of ϕ0 once again. So Dϕ0 = X

and ϕ0 (T (x)) = αϕ0 (x) for all x ∈ X .
To prove that (ii) implies (i), suppose that x0 = T k (x0 ) for some x0 ∈ X , k ∈
N. By hypothesis (ii) ϕ(x0 ) = ϕ(T k (x0 )) = αϕ(x0 ) and this implies ϕ(x0 ) = 0, a
contradiction.
We now indicate the proof of (a) of Theorem 5.5.1, below.
Proof of Theorem 5.5.1 (a) If T m has a fixed point for some m ∈ N, either T has a
fixed point or a periodic point. So part (b) applies and there exists a complete metric
d on X under which T is α-contraction.
If T m has no fixed point for all m ∈ N, T has no periodic point. By Lemma 5.5.3,
there exists ϕ : X → (0, ∞) such that ϕ(T (x)) = αϕ(x). Now d(x, y) = ϕ(x) +
ϕ(y) for x = y and d(x, y) = 0 for x = y is a metric on X with d(T x, T y) =
αd(x, y). Further, the open sphere centred at x and radius ϕ(x) contains only x. So
this metric topology is indeed discrete.
It is natural to enquire if there exist incomplete metric spaces in which every
contraction has a fixed point. The example due to Borwein [5] describes such a
space.
Example 5.5.4 Let L k be the line segment in R2 joining A = (0, 0) to Bk = (1, 21k )

for each k ∈ N. Consider C = L k with the usual euclidean metric in R2 . Clearly
k∈N
each L k is a connected subset of C for k ∈ N and C itself is connected. As C contains
the line segment joining A = (0, 0) with (1, 0), C is not closed in R2 . So C with the
euclidean metric is incomplete. Let T be a contraction mapping C into itself. If each
T n (C) contains (0, 0), then T (0, 0) = (0, 0) (= A) as diameter T n (C) tends to zero.
Clearly A = (0, 0) is the fixed point of T . If T n (C) does not contain A = (0, 0) for
some n ∈ N, then T n (C) being a connected subset not containing A must lie properly
in some L k . So T n maps L k (⊆ C) into L k . Being a contraction on the complete space
L k into itself, T n and hence T has a fixed point in this L k . Thus every contraction of
C into itself has a fixed point, even though C is not complete.
On the other hand, the following theorem ensures the completeness of a metric
space (X, d) under the assumption that a class of self-maps satisfying certain metrical
inequalities have fixed points (see Subrahmanyam [19]).
Theorem 5.5.5 Let (X, d) be a metric space in which every map T : X → X satis-
fying the following conditions has a fixed point.
(i) for some λ > 0 and x, y ∈ X
d(T x, T y) ≤ λ max{d(x, T x), d(y, T y)};
(ii) T (X ) is countable.
Then (X, d) is complete.
Proof Let, if possible, A = {xn } be a non-convergent Cauchy sequence of distinct
elements of X . For any x ∈ X , d(x, A) = inf{d(x, a) : a ∈ A} > 0. Since {xn } is
Cauchy, there exists a least positive integer N (x) such that for m, n ≥ N (x)
5.5 A Converse to the Contraction Principle 113
d(xm , xn ) < λd(x, A) ≤ λd(x, x ), ∈ N.
In particular
d(xm , x N (x) ) < λd(x, x ), ∈ N, m ≥ N (x).
By a similar reasoning for n ∈ N, there exists a least positive integer n = n (n) > n
such that
d(xm , xn ) < λd(xn , xn ), m ≥ n .
Define T : X → X by

x N (x) , if x ∈
/ A
T (x) =
xn , if x = xn ∈ A.
Clearly T has no fixed point, though it satisfies conditions (i) and (ii) above. Indeed
for T (x) = xn , T (y) = xm ,

λd(y, A − {y}), n ≥ m
d(xm , xn ) <
λd(x, A − {x}), n < m
and consequently (i) is true. This contradiction shows that (X, d) must be
complete.
Corollary 5.5.6 (Converse to Kannan’s fixed point Theorem (Corollary 5.2.2)) Let
(X, d) be a metric space. If for each λ ∈ (0, 21 ) every map T : X → X satisfying the
condition
d(T x, T y) ≤ λ[d(x, T x) + d(y, T y)], x, y ∈ X
has a fixed point, then (X, d) is complete.
Corollary 5.5.7 (Converse to Theorem 5.2.1) Let (X, d) be a metric space. If for

5
each {a1 , a2 , a3 , a4 , a5 } ⊆ [0, 1] with ai < 1, each map T : X → X satisfying
i=1
the inequality
d(T x, T y) ≤ a1 d(x, y) + a2 d(x, T x) + a2 d(y, T y) + a4 d(x, T y) + a5 d(y, T x), x, y ∈ X
has a fixed point, then (X, d) is complete.
Corollary 5.5.7 follows from Corollary 5.5.6.

Indeed, the following theorem can be proved using the argument for Theo-
rem 5.5.5.
Theorem 5.5.8 ([19]) Let (X, d) be a metric space in which every mapping T sat-
isfying either
(a) d(T x, T y) ≤ λ max{inf k∈N d(x, T k x), inf k∈N d(y, T k y)} x, y ∈ X
or
(b) d(T x, T y) ≤ λ max{inf k∈N d(x, T k y), inf k∈N d(y, T k x)} x, y ∈ X for fixed
λ > 0
and
(c) T (X ) is countable
has a periodic point.
Then (X, d) is complete.
Earlier Hu [10] proved the following converse of the contraction principle using
this argument.
Theorem 5.5.9 ([10]) Let (X, d) be a metric space. If for every closed non-empty
subset C of X , any contraction T : C → C has a fixed point, then (X, d) is complete.
5.6 A Topological Contraction Principle
In this section, a topological version of the contraction principle due to Kupka [14]
is presented. This fixed point theorem is proved for multifunctions which are feebly
topologically contractive, without involving any concept of completeness.
Definition 5.6.1 Let (X, T ) be a topological space. By a multifunction on X , we
mean a mapping of X into the set of all non-empty subsets of X . The graph of a
multifunction F on X is the set G ◦ F = {(x, y) : y ∈ F x, x ∈ X } and is denoted
by Gr F.
Definition 5.6.2 Let (X, T ) be a topological space. A multifunction F : X → 2 X −

{φ} is said to be feebly topologically contractive if for each open cover G of X and
for any pair of points a, b ∈ X there exists k ∈ N such that for some open set G ∈ G,
F k (a) ⊆ G and F k (b) ∩ G = φ. If f : X → X is (a single-valued) map, then f is
feebly topologically contractive if for any cover G of X and any pair a, b ∈ X , there
exists k ∈ N such that for some G ∈ G, f k (a), f k (b) ∈ G.
Kupka obtained the following generalization of the contraction principle.

Theorem 5.6.3 (Kupka [14]) Let X be an arbitrary topological space and F : X →
2 X − {φ}, a feebly contractive multifunction, with a closed graph. Then F has a fixed
point. If, X is in addition a T1 topological space, then the fixed point is unique and
F(z) = {z}, where z is the fixed point of F.
Proof Suppose F has no fixed point. So x ∈ / F x. Since Gr F is a closed subset of

X × X with the product topology, O = X × X − Gr F is an open set containing
the diagonal {(x, x) : x ∈ X } of X 2 . Let G = {v ∈ T : V × V ⊆ O}. Clearly G is
5.6 A Topological Contraction Principle 115
an open cover for X as (x, x) ∈ O for each x ∈ X . For a ∈ X and b ∈ F(a), by

the feeble-topological contractivity of F, there exists k ∈ N and an open set G ∈
G such that F k (a) ⊆ G and F k (b) ∩ G = φ. Since b ∈ F(a), F k (b) ⊆ F k+1 (a).
So F k+1 (a) ∩ G ⊇ F k (b) ∩ G is non-void. Since F k (a) ⊆ G, F(G) ∩ G = φ. This
implies that Gr F ∩ (G × G) = φ. So Gr F ∩ O = φ, a contradiction to the choice
of O. So F has a fixed point in X .
Suppose further that X is a T1 -space. Let x ∗ be a fixed point of F. Suppose b = x ∗
also belongs to F x ∗ . Clearly G = {X − {x ∗ }, X − {b}} is an open cover for X . As F is
feebly topologically contractive, there is an open set G ∈ G such that for some k ∈ N,
F k (x ∗ ) ⊆ G. As x ∗ ∈ F x ∗ , x ∗ ∈ F k−1 (x ∗ ) and hence {z, b} ⊆ F(x ∗ ) ⊆ F k (x ∗ ) ⊆
G. But this is impossible as G either lacks x ∗ or b. Thus F(x ∗ ) = {x ∗ }.
If F has two distinct fixed points a and b, then from the above argument F(a) =
{a} and F(b) = {b}. Now G = {X − {a}, X − {b}} is an open cover for X . Since
F is feebly topologically contractive, there exists k ∈ N such that for some G ∈ G,
F k (a) ⊆ G and F k (b) ∩ G = φ. Since {a} = F(a) and {b} = F(b), it implies that
G must contain both a and b. This is a contradiction. Hence F has a unique fixed
point.
Corollary 5.6.4 Let T : X → X be a feebly contractive map with a closed graph.

Then T has a fixed point. Further, if X is T1 , then T has a unique fixed point.
Proof Every contraction has a closed graph and is feebly topologically contractive
(as T has a unique fixed point x ∗ in a complete metric space X for a given pair
a, b ∈ X , G can be chosen as an open set containing x ∗ ).
Example 5.6.6 Define G : R → 2R − {φ} by G(x) = {0, n}, if x = n1 , n ∈ N = n >

1 and G(x) = {0}, otherwise. G has closed graph and is feebly topologically con-
tractive, the topology on R being the usual topology. Clearly 0 is the fixed point of
G.
Example 5.6.7 Let X be { n1 : n ∈ N} ∪ {0} with the usual metric. Define f : X → X

by f (0) = 1 and f ( k1 ) = k+1
1
, k ∈ N. While X is a complete metric space and f is
feebly topologically contractive, f has no fixed point. This is because the graph of
f is not closed in X 2 . Thus the assumption that F is closed cannot be dropped in
Theorem 5.6.3.
Kupka [14] had also introduced a stronger notion of contractivity, as in.

Definition 5.6.8 Let (X, T ) be a topological space and T : X → X , a map. T is
called topologically contractive if for each open cover G of X , given a pair of points
a and b in X , there exists n ∈ N such that for all k ≥ n, there exists G ∈ G such that
f k (a), f k (b) ∈ G.
Clearly every topological contraction is topologically feebly contractive. In fact
in Example 5.6.7, the map f is topologically contractive.
Example 5.6.9 Let X be {0, 1, 3} and T be {φ, {0, 1}, X }. Define f : X → X by

f (x) = x for x = 3 and f (3) = 1. f is feebly contractive with 0 and 1 as fixed
points. Evidently (X, J ) is not T1 .
5.7 Another Proof of the Contraction Principle
In this section, Baranga’s proof [2] of the contraction principle using Kleene’s fixed
point theorem is presented. To this end, we need the following definitions and theo-
rems.
Definition 5.7.1 Let (P, ≤) be a partially ordered set. For an increasing sequence
(xn : n ∈ N), we denote the supremum of this sequence by ∨{xn : n ∈ N}. (P, ≤) is
said to be ω-complete if every increasing sequence (xn ) in P has a supremum in P.
Definition 5.7.2 Let (P, ≤) and (Q, ≤) be two partially ordered sets. A map f :
P → Q is said to be ω-continuous if for every increasing sequence (xn ) in P, such
that ∨{xn : n ∈ N} exists in P, also ∨{ f xn : n ∈ N} exists in Q and f (∨{xn : n ∈
N}) = ∨{ f (xn ) : n ∈ N}. (Clearly any ω-continuous function is increasing).
We now state Kleene’s fixed point theorem.

Theorem 5.7.3 (Kleene’s fixed point theorem) Let (P, ≤) be an ω-complete par-
tially ordered set and f : P → P an ω-continuous function. If x ∈ P is such that
x ≤ f (x), then x ∗ = ∨{ f n (x) : n ∈ N} has the following properties:
(i) f (x ∗ ) = x ∗ ; (ii) x ∗ ≥ x and for each y in P with y ≥ x and f (y) ≤ y, x ∗ ≤ y
(x ∗ is the least fixed point of f in {y ∈ P : x ≤ y}.
The proof is left as an exercise.
Given a metric space (X, d), we can define a partially ordered space in a natural
way. Consider X × R+ = X + , R+ being [0, ∞), the set of non-negative real numbers
with the usual order. Define the binary relation ≤ on X + by (x, a) ≤ (y, b) by
d(x, y) ≤ a − b. Clearly this is a partial order on X + .
The following propositions relating convergence in X and convergence in X + ,
lead to the contraction principle as a consequence of Kleene’s fixed point theorem.
Proposition 5.7.4 Let (xn , kn ) be an increasing sequence in (X + , ≤) where X + =
X × R+ and (X, d) is a metric space. Then:
(i) the sequence (kn ) is a decreasing convergent sequence in R+ ;
∞
(ii) d(xn , xn+1 ) converges;
n=1
(iii) (xn ) is a Cauchy sequence in X .
Proof Since (xn , kn ) ≤ (xn+1 , kn+1 ) for n ∈ N, d(xn , xn+1 ) ≤ kn − kn+1 for all n ∈
N. So (kn ) decreases in R+ and so converges to a non-negative number in R+ , say
5.7 Another Proof of the Contraction Principle 117

n
k. Clearly d(x j , x j+1 ) ≤ k1 − kn+1 . Allowing n to tend to +∞, it follows that
j=1
∞
∞

d(x j , x j+1 ) ≤ k1 − k. So d(xn , xn+1 ) converges. Clearly (xn ) is a Cauchy
j=1 1
sequence.
Proposition 5.7.5 Let (xn , kn ) be an increasing sequence in (X + , ≤), X + = X ×

R+ , (X, d) being a metric space. The least upper bound of (xn , kn ) exists in X + if and
only if (xn ) converges in (X, d). Further, the least upper bound of {(xn , kn ) : n ∈ N}
is (x, k) where x = lim xn and k = lim kn .
Proof (Necessity) Let x = lim xn and k = lim kn . For m, n ∈ N with m ≥ n, it fol-

lows from the increasing nature of (xn , kn ), d(xn , xm ) ≤ kn − km . Allowing m to tend
to +∞, it follows that (x, k) is an upper bound for {(xn , kn ) : n ∈ N}. If (x , k ) is
another upper bound, then allowing n to tend to +∞ in d(xn , x ) ≤ kn − k , we get
d(x, x ) ≤ k − k implying that (x, k) ≤ (x , k ). So (x, k) is the least upper bound
of {(xn , kn ) : n ∈ N}.
(Sufficiency) Suppose (x, k ) is the least upper bound of {(xn , kn ) : n ∈ N}. By
Proposition 5.7.4, (kn ) decreases to some k ∈ R+ . Since kn ≥ k for each n ∈ N,
k = lim kn ≥ k . If k = k , then kn − k ≥ d(xn , x) would imply that lim xn = x.
Suppose k = k . We claim that for each > 0, there exists x ∈ X such that
d(xn , x ) ≤ + kn − k for each n ∈ N. Otherwise for some 0 > 0, for each y ∈ X
we can find n y ∈ N such that
d(xn y , y) > 0 + kn y − k.
For n > n y , we have

n−1
0 + kn y − k < d(xn y , y) ≤ d(xi , xi+1 ) + d(xn , y)
i=n y
≤ kn y − kn + d(xn , y)
≤ kn y − k + d(xn , y).
So d(xn , y) > 0 for each n > n y .

As (xn , kn ) is increasing, it follows from Proposition 5.7.4 that (xn ) is Cauchy.
So (xn ) converges to some x ∗ in X ∗ , the completion of X . Since d(xn , y) > 0 for
each n ≥ n y . Allowing n (≥ n y ) to tend to +∞, it follows that d(x ∗ , y) ≥ 0 , for
each y ∈ X , contradicting that d(x ∗ , xm ) does not tend to zero in X ∗ . Hence for
each > 0, there exists x ∈ X such that d(xn , x ) ≤ + kn − k for each n ∈ N.
Choose 0 < < k − k . It follows that for this choice of , (x , k − ) is an upper
bound for (xn , kn ). As (x, k ) is the least upper bound of {(xn , kn ) : n ∈ N}, we have
(x, k ) ≤ (x , k − ) on d(x, x ) ≤ k − (k − ) = + k − k < 0, a contradiction.
Hence (xn ) converges to x, and kn converges to k.
Corollary 5.7.6 If the metric space is (X, d) is complete, then (X + , ≤) is ω-

complete.
A Lipschitz map on a metric space defines a ω-continuous map in a natural way,

as described in the following.
Proposition 5.7.7 Let (X, d) be a metric space and f : X → X be a map such that
for some c > 0, d( f x, f y) ≤ cd(x, y) for all x, y ∈ X . Then the map f + : X + =
(= X × R+ ) → X + defined by f + (x, a) = ( f (x), ca) is ω-continuous.
Proof Let (xn , kn ), n ∈ N be an increasing sequence in X + with (x, k) = ∨{(xn , kn ) :

n ∈ N}. For (y1 , a1 ), (y2 , a2 ) ∈ X + with (y1 , a1 ) ≤ (y2 , a2 ), d(y1 , y2 ) ≤ a1 − a2 .
Now d( f y1 , f y2 ) ≤ cd(y1 , y2 ) ≤ ca1 − ca2 . So ( f y1 , ca1 ) ≤ ( f y2 , ca2 ). So f + (y1 ,
a1 ) ≤ f + (y2 , a) and f + is increasing. Also f + (x, k) = ( f (x), ck) = ∨{( f (xn ),
ckn ) : n ∈ N} by Proposition 5.7.4. Thus f + (x, k) = ∨{ f + (xn , kn ) : n ∈ N} and f +
is ω-continuous.
We are now in a position to provide Baranga’s proof [2] of the contraction principle
5.1.7 via Kleene’s fixed point Theorem 5.7.3.
Proof of Theorem 5.1.7 (Baranga [2]).
Let (X + , ≤) and f + : X + → X + be defined as in Proposition 5.7.7. For x0 ∈ X ,
we can find a > 0 such that (1 − c)a > d(x0 , f (x0 )) so that (x0 , a) ≤ f + (x0 , a).
By Kleene’s Theorem 5.7.3, (x̄, 0) = ∨{( f + )n (x0 , a) : n ∈ N} is a fixed point of f + .
So x̄ = lim f n (x0 ) is a fixed point of f . Clearly x̄ is independent of the choice of a.
Also f + (y, b) ≤ (y, b) if and only if b = 0 and f (y) = y. If y is a fixed point of f ,
one can choose a > 0 so that (x0 , a) ≤ (y, 0) and (x0 , a) ≤ f + (x0 , a). By the least
fixed point property it follows that (x̄, 0) ≤ (y, 0). This implies that d(x̄, y) = 0 or
f has a unique fixed point x̄.
References
1. Banach, S.: Sur les operations dans les ensembles abstraits et leur application aux equations
integrales. Fund. Math. 3, 133–181 (1922)
2. Baranga, A.: The contraction principle as a particular case of Kleene’s fixed point theorem.
Discret. Math. 98, 75–79 (1991)
3. Bessaga, C.: On the converse of Banach fixed point principle. Colloq. Math. 7, 41–43 (1959)
4. Bollabos, B.: Linear Analysis: An Introductory Course. Cambridge University, Cambridge
(1999)
5. Borwein, J.M.: Completeness and the contraction principle. Proc. Am. Math. Soc. 87, 246–250
(1983)
6. Boyd, D., Wong, J.S.W.: On nonlinear contractions. Proc. Am. Math. Soc. 20, 458–469 (1969)
7. Cacciopoli, R.: Un teorema generale sullesistenzadi elementi uniti in una transformazione
funzionale. Rend. Accad. Lincei 6(11), 749–799 (1930)
9. Hardy, G.E., Rogers, T.D.: A generalization of a fixed point theorem of Reich. Can. Math. Bull.
16, 201–206 (1973)
10. Hu, T.K.: On a fixed point theorem for metric spaces. Am. Math. Mon. 74, 436–437 (1967)
References 119
11. Jachymski, J.: A short proof of the converse to the contraction principle and some related
results. Topol. Methods Nonlinear Anal. 15, 179–186 (2000)
12. Jungck, G.: Commuting mappings and fixed points. Am. Math. Mon. 83, 261–263 (1976)
13. Kannan, R.: Some results on fixed points. Bull. Calcutta Math. Soc. 60, 71–76 (1968)
14. Kupka, I.: Topological conditions for the existence of fixed points. Math. Slovaca 48, 315–321
(1998)
15. Nadler, S.B.: Multi-valued contracting mappings. Pac. J. Math. 30, 475–488 (1969)
16. Palais, R.S.: A simple proof of the Banach contraction principle. J. Fixed Point Theory Appl.
2, 221–223 (2007)
17. Rakotch, E.: A note on contractive mappings. Proc. Am. Math. Soc. 13, 459–465 (1962)
18. Rall, L.B.: Computational Solution of Nonlinear Operator Equations. Wiley, New York (1969)
19. Subrahmanyam, P.V.: Fixed points and completeness. Monat. Math. 80, 325–330 (1975)
20. Tan, K.K.: Fixed point theorems for nonexpansive mappings. Pac. J. Math. 41, 829–842 (1972)
Chapter 6
Applications of the Contraction Principle
This short chapter offers a few samples of applications of the contraction principle.
It was already pointed out that the evergrowing list of applications of this fixed point
theorem would fill volumes.
6.1 Linear Operator Equations
We have the following simple result.
Theorem 6.1.1 Let (X, · ) be a Banach space and T : X → X , a continuous

linear operator. If T < 1, then the equation x = T x + a has a unique solution in
X for each fixed a ∈ X . Further {T1n (x0 )} converges to the solution of this equation
for each x0 ∈ X , T1 (x0 ) being T (x0 ) + a for x0 ∈ X .
Proof Define T1 (x) = T (x) + a for each x ∈ X . Clearly T1 maps X into itself and
T1 (x) − T1 (y) = T x − T y ≤ kx − y, where k = T < 1. Thus T1 is a con-
traction mapping X into itself. Since X is complete, T1 has a unique fixed point
x ∗ . Thus T x + a = x has the solution x = x ∗ . Again, by the contraction principle
{T1n (x0 )} converges to the unique solution of the equation x = T x + a.
Corollary 6.1.2 Let A = (ai j ) be an n × n real matrix and b = (b1 , . . . , bn ) be a

vector in Rn . Then the system of linear equations
n
xi = ai j x j + bi , i = 1, 2, . . . , n
j=1

n
n
n
has a unique solution (i) provided sup |ai j | < 1 or (ii) ai2j < 1.
i j=1 i=1 j=1

https://doi.org/10.1007/978-981-13-3158-9_6
122 6 Applications of the Contraction Principle
Proof Choose X = Rn and T : X → X as the operator x → Ax where

n
x = (x1 , . . . , xn ) ∈ Rn and Ax = (y1 , . . . , yn ) where yi = ai j x j , and
j=1
a = (b1 , . . . , bn ).
(i) Let x = sup |xi | where x = (x1 , . . . , xn ). Clearly Ax = sup |yi | ≤
i=1,...,n i=1,...,n

n
n
n
sup |ai j |x. Since sup |ai j | < 1, xi = ai j x j + bi has a unique solution
i j=1 i j=1 j=1
in view of the Theorem
n 6.1.1. 21

(ii) Let x = |xi |2 of x = (x1 , . . . , xn ). Then Rn with this Euclidean
i=1
norm is complete. Now
n 21

Ax = yi2
i=1
⎛⎛ ⎞2 ⎞ 21

n
n
=⎝ ⎝ ai j x j ⎠ ⎠
i=1 j=1
⎧ ⎡ ⎤⎞ 21
⎨n
n
n
≤ ⎣ ai2j x 2j ⎦⎠
⎩
i=1 j=1 j=1
(by Cauchy-Schwarz inequality)

⎛ ⎞ 21

n
n
≤⎝ ai2j ⎠ x
i=1 j=1

n
n
n
So A is a contraction on Rn , since ai2j < 1. Thus xi = ai j x j + bi ,
i=1 j=1 j=1
i = 1, 2, . . . , n has a unique solution, by Theorem 6.1.1.
Corollary 6.1.3 Let a, b ∈ R and a < b. Suppose K : [a, b] × [a, b] → R is con-

tinuous and is not zero everywhere and g : [a, b] → R is also continuous. Then the
Fredholm integral equation
b
x(t) = λ K (s, t)x(s)dt + g(t), t ∈ [a, b]
a
has a unique solution x ∈ C[a, b] for |λ| < 1

M(b−a)
where M = Sup{|K (s, t)| :
s, t ∈ [a, b]}.
6.1 Linear Operator Equations 123
b
Proof Define the linear operator T : X → X by T (x(t)) = λ a K (s, t)x(s)ds, t ∈
[a, b] for x ∈ X = C[a, b], with this norm is a Banach space. That, T x(t) is a
continuous function on [a, b] , can be proved, using the uniform continuity of K on
[a, b] × [a, b]. Let M = Sup{|K (s, t)| : s, t ∈ [a, b]}. Since K is continuous on the
compact set [a, b] × [a, b], 0 < M < +∞. Now for x1 , x2 ∈ C[a, b] and t ∈ [a, b]
b
|T x1 (t) − T x2 (t)| ≤ |λ| |K (s, t)||x1 (s) − x2 (s)|ds
a
≤ M|λ|x1 − x2 (b − a)
So T x1 − T x2 ≤ M(b − a)|λ|x1 − x2 and is a contraction as |λ| < M(b−a)

1
. So
by Theorem 6.1.1, T1 has a unique fixed point, where T1 x = T x + g, x ∈ X . Thus
the Fredholm integral equation has a unique solution.
Corollary 6.1.4 For a, b ∈ R with a < b, let K : [a, b] × [a, b] → R and g :

[a, b] → R be continuous. Then the Volterra integral equation
t
x(t) = λ K (s, t)x(s)ds + g(t), t ∈ [a, b]
a
has a unique continuous solution on [a, b] for all λ ∈ R.
Proof Let tX be the Banach space C[a, b] with the supremum norm. Then for each x ∈
C[a, b], a K (s, t)x(s)ds defines a continuous function on [a, b]. Further x(t) →
t
T (x(t)) = λ a K (s, t)x(s)ds is a linear operator on X .
For x1 , x2 ∈ X , λ ∈ R and t > 1
t
T x1 (t) − T x2 (t) = λ K (s, t)(x1 (s) − x2 (s))ds
a
So
t
|T x1 (t) − T x2 (t)| ≤ |λ| Mx1 − x2 ds
a
≤ |λ|Mx1 − x2 (t − a)
where M = Sup{|K (s, t)| : s, t ∈ [a, b]} and x1 − x2 = Sup{|x1 (t) − x2 (t)| : t ∈
[a, b]}}. Now for n ∈ N and t > a.
|T n+1 x1 (t) − T n+1 x2 (t)| ≤ |λ|MT n x1 (t) − T n x2 (t)(t − a)
Inductively it can be shown that
|λ|k M k
|T k x1 (t) − T k x2 (t)| ≤ (t − a)k x1 − x2
k!
for k ∈ N and t ∈ (a, b]. So
|λ|n+1 M n+1
T n+1 x1 − T n+1 x2 ≤ (b − a)n+1 x1 − x2
(n + 1)!
|λM(b − a)|n+1
= 0, for some n 0 ∈ N, |λM(b−a)|
n
Since lim < 1 for n ≥ n 0 . Thus
n→∞ (n + 1)! n!
n0
T is a contraction on X and for T1 = T + g, T1 is also a contraction on X . So T1
n0
has a unique fixed point in X = C[a, b] which is the solution of the given integral
equation. Thus the Volterra integral equation has a unique solution in C[a, b] for all
λ ∈ R.
Remark 6.1.5 We can supplement Corollary 6.1.3 on the eigen-value problem for
Fredholm integral equations. Suppose K : [a, b] × [a, b] → R and g : [a, b] → R
bb
are Lebesgue measurable functions such that g ∈ L 2 [a, b] and 0 < a a K 2 (s, t)
dsdt < +∞. Then we can show that the equation
b
x(t) = λ K (s, t)ds + g(t)
a
bb
has a unique solution in L 2 [a, b] for |λ| a a K 2 (s, t)dsdt < 1. For the proof we
use the Hilbert space L 2 [a, b] instead of C[a, b].
The next result insures that certain mappings on Banach spaces are surjections
and indeed homeomorphisms.
Theorem 6.1.6 Let (X, · ) be a Banach space and T : X → X be a bounded

linear transformation of X onto X with a bounded inverse. Let G : X → X be map
such that Gx − Gy ≤ αx − y for all x, y ∈ X . Suppose β = T −1 and αβ <
1. Then x → T x − Gx defines a homeomorphism of X onto itself.
Proof Let y ∈ X . Consider the map F defined by F x = T x − Gx, x ∈ X and the

map H defined by H x = T −1 (Gx + y) for each x ∈ X . Then H x = T −1 (T x −
F x + y) or H x = x − T −1 F x + T −1 y. The equation F x = y has a unique solution
in X if and only if H has a unique fixed point. Now H x1 − H x2 = T −1 (Gx1 −
Gx2 ) ≤ αβx1 − x2 for x1 , x2 ∈ X . Since 0 ≤ αβ < 1, H is a contraction on the
complete space X and hence has a unique fixed point. Thus F is a continuous map
of X onto X . Also for x1 , x2 ∈ X
F T −1 x1 − F T −1 x2 = x1 − GT −1 x1 − x2 + G −1 x2
≥ x1 − x2 − GT −1 x1 − GT −1 x2
≥ (1 − αβ)x1 − x2
So F x1 − F x2 ≥ (1 − αβ)T x1 − T x2
or x1 − x2 ≥ (1 − αβ)T F −1 x1 − T F −1 x2
6.1 Linear Operator Equations 125
This shows that T F −1 is continuous and so F −1 is continuous. Hence F is a home-

omorphism.
Corollary 6.1.7 Let (X, · ) be a Banach space and G : X → X a (strict) con-

traction. Then (I − G) : X → X is a homeomorphism.
Proof In Theorem 6.1.6, set T = I , the identity operator.
Example 6.1.8 For each (y1 , y2 ) ∈ R2 , we can find a unique (x1 , x2 ) ∈ R2 such that

x1 + x2 |x1 | + |x2 |
(2x1 − 3x2 , x1 − 2x2 ) − cos , = (y1 , y2 ).
50 50(1 + |x1 | + |x2 |
6.2 Differential Equations
We prove below an existence theorem for the solution of an initial value problem for
a system of first-order ordinary differential equations.
Theorem 6.2.1 Let G be an open set in Rn+1 containing the point (t0 , x10 , x20 , . . . , xn0 )
and f i : G → R be continuous functions satisfying the Lipschitz condition

n
| f i (t, y1 , . . . , yn ) − f i (t, z 1 , . . . , z n )| ≤ M |y j − z j |
j=1
for each i = 1, 2, . . . , n for all (t, y1 , y2 , . . . , yn ) and (t, z 1 , z 2 , . . . , z n ) ∈ G. Then

we can find h > 0 such that there exist continuously differentiable functions x1 , . . . , xn
mapping [t0 − h, t0 + h] into R satisfying the conditions
d xi
= f i (t, x1 (t), . . . , xn (t)) for i = 1, 2, . . . , n
dt
and xi (t0 ) = xi0 for i = 1, 2, . . . , n. Further, these are unique solutions and (t, x1 (t),
x2 (t), . . . , xn (t)) ∈ G for all t ∈ [t0 − h, t0 + h].
Proof Since G is open and (x10 , . . . , xn0 , t0 ) ∈ G, we can find a closed rectangle R =
[x10 − a, x10 + a] × . . . × [xn0 − a, xn0 + a] × [t0 − a, t0 + a], a > 0 in G. As each f i
is continuous on R and R is compact, we can find K > 0 such that | f i ( p)| ≤ K for
all i = 1, 2, . . . , n for all p ∈ R. In view of the continuity of each f i , i = 1, 2, . . . , n,
it is clear that xi (t) is a solution of
d xi
= f i (t, x1 (t), . . . , xn (t))
dt
xi (t0 ) = xi(0) , i = 1, 2, . . . , n
if and only if t
xi (t) = xi (0) + f i (τ , x1 (τ ), . . . , xn (τ ))dτ .
t0
Choose h > 0 such that K h < a. Let (X, ρ) be the metric space of vector-valued
continuous real functions (x1 (t), . . . , xn (t)) defined on I = [t0 − h, t0 + h] with
n
the metric ρ(x, y) = sup |xi (t) − yi (t)|, where x = (x1 (t), . . . , xn (t)) and y =
t∈I i=1
(y1 (t), . . . , yn (t)).
Clearly (X, ρ) is complete. For x = (x1 , . . . , xn ) ∈ X , define T x = (y1 , . . . , yn )
where t
yi (t) = xi0 + f i (τ , x1 (τ ), . . . , xn (τ ))dτ .
t0

n
n
T (x) ∈ X whenever x ∈ X . Further, ρ(T x, T x ) = |yi − yi | ≤ Mρ(x, x )
i=1 i=1
h = n Mρ(x, x )h. If we further choose h such that n Mh < 1 then T is a contraction
on X and hence has a unique fixed point x = (x1 , . . . , xn ) which is the solution
of the initial value problem in [t0 − h, t0 + h] and |xi0 − xi (t)| ≤ K h ≤ a for all
t ∈ [t0 − h, t0 + h] and so lies in R ⊆ G. Thus for h < min{ Ka , n1M }, a unique local
solution exists in [t0 − h, t0 + h] for the initial value problem.
Corollary 6.2.2 Let G be an open set in R2 containing the point (t0 , x0 ) and f :
G → R be a continuous function satisfying the condition that | f (t, x1 ) − f (t, x2 )| ≤
M|x1 − x2 | for all (t, x1 ), (t, x2 ) ∈ G. Then there exists h > 0 such that the initial
value problem
dx
= f (t, x(t)), x(t0 ) = x0
dt
has a unique solution x(t) on [t0 − h, t0 + h] such that (t, x(t)) ∈ G for t ∈ [t0 −
h, t0 + h].
(This pertains to the case n = 1 in Theorem 6.2.1.
Next, we prove a theorem on the existence of analytic solutions for a differential

equation. To this end, we recall a few definitions.

Definition 6.2.3 A power series of the form ai1 ...im x1i1 . . . xmim where ai1 i2 ...im
i 1 ,i 2 ,...≥0
are real numbers is called real-analytic in m variables x1 , . . . , xm inside the sphere

m
S(a; ρ) = {(x1 , . . . , xm ) ∈ Rm : (xi − ai )2 < ρ2 (ρ > 0)} if it converges therein,
i=1
a being (a1 , . . . , am ).
Remark 6.2.4 If the power series in m-real variables converges in S(a; ρ), then it
converges in S(a; ρ ) for 0 < ρ < ρ.
6.2 Differential Equations 127
Remark 6.2.5 The uniform limit of a convergent sequence of real-analytic functions

is real-analytic in any domain interior to the domain of uniform convergence of the
sequence. (This result is due to Weierstrass).
We have the following theorem due to Cauchy.
Theorem 6.2.6 (Cauchy) Consider the initial value problem
du
= f (x, y), y(x0 ) = y0 ,
dx
where f is real-analytic in {(x, y) ∈ R2 : |x − x0 | < h, |y − y0 | < h}. Then there

exists a unique solution y = φ(x) for this initial value problem which is a power
series in [x0 − , x0 + ] for some > 0 with < h.
Proof As before the initial value problem is equivalent to solving

x
y(x) = y0 + f (t, y(t))dt.
x0
Since f is a power series in x and y in {(x, y) : |x − x0 |, |y − y0 | < h}, ∂∂ yf exists and

is a power series in x and y and is continuous in K ={(x, y) : |x − x0 | ≤ h , |y −

y0 | ≤ h } where 0 < h < h. Let M = Sup{ ∂∂ yf (x, y) : (x, y) ∈ K }. Without loss
of generality we may assume that M > 0.
Let X be the set of real-analytic function, real-analytic in {(x, y) : (x − x0 )2 +
(y − y0 )2 ≤ h 2 } with the supremum metric, d. Define T : X → X by T (φ(x)) =
x
y0 + x0 f (t, φ(t))dt for all x with |x − x0 | ≤ h . Clearly T φ is well-defined and is
real-analytic in |x − x0 | ≤ h . Further for φ1 , φ2 ∈ X
x
|T φ1 (x) − T φ2 (x)| ≤ | f (t1 φ1 (t) − f (t, φ2 (t))|dt
x0
≤ h Md(φ1 , φ2 )
If we choose 0 < < h such that < M1 , clearly d(T φ1 , T φ2 ) ≤ αd(φ1 , φ2 ), where
α = M < 1. Thus T is a contraction on the complete metric space (X, d) and hence
has a unique fixed point, which is the unique solution of the initial value problem in
[x0 − , x0 + ]. Further, this solution is a power series in x in this interval.
Remark 6.2.7 Clearly this theorem can be extended to a system of differential equa-
tions involving real-analytic functions of several real variables.
6.3 A Functional Differential Equation
Utz [12] raised the problem of determining conditions for the existence of a real
function y(x), not identically zero for which y (x) = ay(g(x))) where a is a given
constant and g(x) a given real function. Ryder [7] gave a solution to this problem
under suitable assumptions.
Theorem 6.3.1 (Ryder [7]) Let g : Dg → Dg be a continuous function on a interval

in R containing the origin with |g(x)| ≤ k for all x in Dg . Then there exists a unique
solution f (x) : Dg → Rn to the initial value problem
y (x) = Ay(g(x)), x ∈ Dg
y(0) = f 0
with A is a given n × n real matrix such that Ak < 1, f 0 ∈ Rn .
Proof Let S be the set of all functions f : Dg → Rn such that f (0) = f 0 and
f (x) − f 0 < L|x| for all x ∈ Dg for some L > 0. Define ρ : S × S → R+ by
ρ( f 1 , f 2 ) = inf{L : f 1 (x) − f 2 (x) ≤ L|x|, x ∈ Dg }
If can be shown that ρ is a metric on S and in fact (S, ρ) is a complete metric space.
Define the operator T : S → S by
x
T ( f )(x) = f 0 + A f (g(s))ds.
0
If f ∈ S, then f (g(x)) is well-defined and continuous on DG and T f (0) = f 0 .

Further,
x

T ( f (x)) − f 0 ≤ A [ f 0 + L|g(s)|]ds
0
≤ A( f 0 + Lk)|x|, x ∈ DG .
So T ( f ) ∈ S for f ∈ S.
If f 1 , f 2 ∈ S, then f 1 (x) − f 2 (x) ≤ L|x| for x ∈ DG . Now for x ∈ Dg ,

x
T ( f 1 )(x) − T ( f 2 )(x) ≤ A L|g(s)|ds
0
≤ Ak L|x|
Consequently
ρ(T f 1 , T f 2 ) ≤ Akρ( f 1 , f 2 )
6.3 A Functional Differential Equation 129
Since Ak < 1, it follows that T is a contraction on the complete

x metric space and
hence has a unique fixed point f (x). Since f (x) = f 0 (x) + A 0 f (g(s))ds, and f
and g are continuous, if follows that f (x) = A f (g(x)) and f (0) = f 0 (0).
Example 6.3.2 Let g : [− 21 , 1] → [− 21 , 1] be the continuous function defined by

−x
g(x) = x+1 and |g(x)| ≤ 1 for all x ∈ Dg = [− 21 , 1]. The functional differential
−x
equation f (x) = 21 f x+1 , f (0) = 1 has a unique solution in [− 21 , 1] in view of
√
Theorem 6.3.1, A being 2 . In fact the unique solution is f (x) = x + 1.
1
6.4 A Classical Solution for a Boundary Value Problem for

a Second Order Ordinary Differential Equation
We consider the following boundary value problem
d2x
= α f (t, x(t)), t ∈ (0, 1)
dt 2
x(0) = x(1) = 0.
Theorem 6.4.1 Let f : [0, 1] × [−a, a] → R be a continuous function satisfying

the Lipschitz condition | f (t, x1 ) − f (t, x2 )| ≤ M|x1 − x2 | for all (t, xi ) ∈ [0, 1] ×
[−a, a], i = 1, 2. Then the above boundary value problem has a unique solution for
sufficiently small values of α.
Proof The Green’s function for the problem G(t, s) is defined by

(t − 1)s, 0 ≤ s ≤ t
G(t, s) =
t (s − 1), t ≤ s ≤ 1
Let M = {x ∈ C[0, 1] : x ≤ a} where x = Sup{|x(t)| : t ∈ [0, 1]}. As f is

continuous, | f (s, 0)| ≤ K for some K > 0 for all s ∈ [0, 1]. Also for x ∈ M,
s ∈ [0, 1]
| f (s, x(s))| ≤ | f (s, 0)| + | f (s, 0) − f (s, x(s))|

≤ K + Mx ≤ k + Ma.
The operator T : M → C[0, 1] defined by

1
T (x)(t) = α G(t, s) f (s, (x))ds
0
maps M into itself provided

α α
|T (x)(t)| ≤ f (s, x(s)) ≤ (K + Ma) < a
8 8
That is for α < 8a

K +Ma
. Further, for x1 , x2 ∈ M, we have for all t ∈ [0, 1]

|T (x1 )(t) − T (x2 )(t)| = |α G(t, s)( f (s, x(s)) − f (s, x(s))ds|
0
|α|M
≤ x1 − x2
8

So if we choose |α| < min M8 , K +Ma8a
, then T being a contraction, has a unique
fixed point, which indeed is the unique solution to the boundary value problem.
6.5 An Elementary Proof of the Cauchy–Kowalevsky

Theorem
Walter [13] gave an elementary proof of the Cauchy–Kowalevsky theorem using

the contraction principle. First the linear version of this theorem is proved using
Nagumo’s lemma and then the proof for the quasi-linear case is given, as the nonlinear
case can be reduced to the quasi-linear case. So we merely discuss the proof of the
Cauchy–Kowalevsky theorem for the linear case following Walter [13].
Let G be a non-empty open subset of Cn+1 or R × Cn , an open set in Cn with
non-empty boundary = ∂. Define d(z) = dist (z, ) measured in the maximum
norm |z| = max{|z i | : i = 1, 2, . . . , n}. We further restrict G by requiring that it
consists of all points (t, z), with z ∈ and |t| < ηd(z) where η > 0 is to be specified.
The linear (Cauchy) problem is to solve the initial value problem

n
u t = A(t, z)u + B j (t, z)u z j + c(t, z) for (t, z) ∈ G (6.5.1)
j=1
with u(0, z) = φ(z) in , where u = (u 1 , . . . , u m ) has values in Cm . A and B, are

complex m × m matrices, c and φ being complex-valued column vectors. Further,
u t = ∂u
∂t
∂u
, u z j = ∂z j
. We seek a solution of the above problem which is continuous in
G and analytic in z for fixed t (real case) and analytic in t and z (complex case). We
m
also write A = max |ai j | for A = (ai j ).
j
i=1
For proving the existence of analytic solution of (6.5.1). We need the following
lemma due to Nagumo [6].
Lemma 6.5.1 Let f : → Cm be analytic and p ≥ 0. Then
C C
| f (z)| ≤ implies | f z j (z)| ≤ C p
d p (z) d p+1 (z)
6.5 An Elementary Proof of the Cauchy–Kowalevsky Theorem 131
where C p = (1 + p)(1 + 1p ) p < e( p + 1), C0 = 1.
Proof If g is analytic function of single complex variable in |z − z | ≤ r , then
1
|g (z)| ≤ max |g(z )|.
r |z−z |=r
This follows from Cauchy’s integral formula

g(z )
g (z) = dz .
|z−z |=r 2πi(z − z)2
Applying this for z = z j we get the inequalities
1 C 1 C
| f z j (z)| ≤ max | f (z )| ≤ max p ≤
r |z−z |=r r d (z ) r (d − r ) p
where 0 < r < d = d(z) and d(z ) ≥ d − r . The choice r = d

p+1
leads to the esti-
mate stated in the lemma.
Remark 6.5.2 The Cauchy problem (6.5.1) can be written as the equivalent integral
equation.
t
n
u(t, z) = g(t, z)+ [A(s, z)u(s, z) + B j (s, z)u z j (s, z)]ds (6.5.2)
0 j=1
t
where g(t, z) = φ(z) + c(s, z)ds (6.5.3)
0
Theorem 6.5.3 Suppose

(i) the functions A(t, z), B j (t, z), C(t, z) are continuous in G and analytic in z for
fixed t, the function φ(z) being analytic in z;
(ii) there exist positive constants α, β j , η, δ and p such that
α
|A(t, z)| ≤ , |B j (t, z)| ≤ β j ,
d(t, z)
η δ
|C(t, z)| ≤ p+1 , |φ(z)| ≤ in G;
d (t, z) d p (z)
α
(iii) p
+ (1 + 1p ) p+1 β j < η1 .
Then the Eq. (6.5.2) has a unique solution u in G satisfying |u(t, z)| ≤ C
d p (t,z)
for
(t, z) ∈ G.
Proof Let X be the Banach space of all functions u = (t, z) which are continuous
on G and taking values in Cm and analytic in z with the finite norm
u = Sup |u(t, z)|d p (t, z)

G
Clearly convergence in the norm is uniform convergence on compact subsets of

G. So the limit of a convergent sequence in X is analytic and X is complete in this
norm. Equation (6.5.2) can be written as
u = g + T (u)
where T is the linear operator defined by

t
T u(t, z) = [A(s, z)u(s, z) + B j (s, z)u s j (s, z)]ds]
0 j
Since
|t|
t
ds ds η
= p+1 < , g(t, z) ∈ X.
d p+1 (s, z) pd p (t, z)
0 0
d(s) − ηs
From the definition of the norm

u
|u(t, z)| ≤ .
d p (t, z)
Applying Nagumo’s Lemma 6.5.1 to t with distance d(t, z) instead of and d(z)
we get,
u
|u z j (t, z)| ≤ C p p+1
d (t, z)
Assumption (ii) and these estimates give
αu u
|Au| ≤ , |B j u z j | ≤ p+1 βjCp
d p+1 (t, z) d (t, z)

So for β = βj

t
ds
|T u(t, z)| ≤ u(α + βC p )
0 d (s, z)
p+1
1 ηu
≤ (α + βC p ) p .
p d (t, z)
So T u ≤ qu, where q = 1p [α + β(1 + 1p ) p+1 ]η < 1. Since 0 < q < 1, u →

g + T u is a contraction on the Banach space X and has a unique fixed point. Thus
6.5 An Elementary Proof of the Cauchy–Kowalevsky Theorem 133
(6.5.2) has a unique solution. Thus the Cauchy problem (6.5.1) with analytic data
has a unique solution.
Note 6.5.4 In Theorem 6.5.3, η > 0 can be chosen sufficiently small so that (iii) of
Theorem 6.5.3 holds.
6.6 An Application to a Discrete Boundary Value Problem
In this section a solution to a discrete boundary value problem obtained by Tisdell

[10], using the contraction principle is described.
Let f : [0, N ] × R → R be a continuous function. The discrete boundary value
problem is to solve
xi
= f (ti , xi ), i = 0, 1, . . . , n, ux0 + vxn = w, u + v = 0 (6.6.1)
h
where 0 < h < Nn < N and ti = i h (i = 0, 1, . . . , n) are the grid points with xi =
xi+1 − xi , i = 0, . . . , n − 1 and u, v, w are constants. The problem is to find a vetor
x̃ = (x0 , x1 , . . . , xn ) ∈ Rn+1 satisfying (6.6.1).
Theorem 6.6.1 (Tisdell [10]) Let f : [0, N ] × R → R be a continuous function

and u + v = 0. Suppose there exist constants L > 0, δ ∈ (0, N ), p, q > 1 such that
(i) | f (t, y) − f (t, z)| ≤ L|y − z| for t ∈ [0, N ], y, z ∈ R;
p
1p
n
Lδ
(ii) |u+v| i=0 [|u|q
i + |v|q
(n − i)] q < 1, (with 1p + q1 = 1);
Then the boundary value problem has a unique solution for 0 < h ≤ δ.
n 1
Proof Consider X = Rn+1 with the norm x = i=0 |x i |
p p
for p > 1. Clearly
X is a Banach space. The problem (6.6.1) is equivalent to the system of equations

n−1
w
xi = h G(i, j) f (t j , x j ) + , i = 0, . . . , n
j=0
u+v
where
u
, 0≤ j ≤i −1
G(i, j) = u+v
−u
u+v
, i ≤ j ≤n−1
Define T : X → X by

n−1
w
(T x̃)i = h G(i, j) f (t j , x j ) + , i = 0, . . . , n
j=0
u+v
for x̃ = (x0 , . . . , xn ) ∈ X .
For x̃, ỹ ∈ X = Rn+1

n−1
|T (x̃)i − T ( ỹ)i | ≤ h |G(i, j)|| f (t j , x j ) − f (t j , y j )|
j=0

n−1
≤ Lh |G(i, j)||x j − y j |
j=0
⎛ ⎞ q1 ⎛ ⎞ 1p

n−1
n−1
≤ Lh ⎝ |G(i, j)|q ⎠ ⎝ |x j − y j | p ⎠
j=0 j=0
(by Holder’s inequality)
From the definition of G(i, j) we have

n−1
i|u|q + (n − i)|v|q
|G(i, j)|q = , i = 0, . . . , n.
j=0
|u + v|q
So
Lhx̃ − ỹ 1
|T (x̃)i − T ( ỹ)i | ≤ [i|u|q + (n − i)|v|q ] q , i = 0, . . . , n.
|u + v|
So
p
Lh p
|(T x̃)i − (T ỹ)i | ≤
p
x̃ − ỹ p [i|u|q + (n − i)|v|q ] q , i = 0, . . . , n.
|u + v|
So
n 1p

T x̃ − T ỹ = |(T x̃)i − (T ỹ)i | p
i=0
n 1p
Lh p
≤ x̃ − ỹ [i|u| + (n − i)|v| ]
q q q
|u + v| i=0
≤ αx̃ − ỹ
n 1p
Lδ p
where α = [i|u|q + (n − i)|v|q ] q < 1. So T is a contraction on X
|u + v| i=0
and so has a unique fixed point, which is a solution to the discrete boundary value
problem.
6.6 An Application to a Discrete Boundary Value Problem 135
Corollary 6.6.2 ([10]) Let f : [0, N ] → R be continuous and for 0 ≤ t ≤ N and

x, y ∈ R, | f (t, x) − f (t, y)| ≤ L|x − y| for some L > 0 and u + v = 0. Then for
0 < h ≤ δ, the √boundary value problem has a unique solution provided
√
L N 2 + δ N < √2|u+v|u 2 +v 2
.
The above inequality insures that the conditions of Theorem 6.6.1 are fulfilled.
Example 6.6.3 Set u = 1, v = 2, w = 0, N = 1, f (t, y) = 45 (y + t + cos y). The

corresponding boundary value problem satisfies the conditions of Corollary 6.6.2 for
L = 85 and has a unique solution for 0 < h ≤ 41 .
6.7 Applications to Functional Equations
Contraction principle has been a handy tool in the solution of a variety of functional
equations. As a matter of fact the implicit function theorem is a consequence of the
contraction principle.
Theorem 6.7.1 (Implicit Function Theorem) Let U be an open neighbourhood of

(x0 , y0 ) in R2 . Let f : U → R be a continuous function such that ∂∂ yf exists in U and
is continuous at (x0 , y0 ). Suppose (i) ∂∂ yf (x0 , y0 ) = 0 and (ii) f (x0 , y0 ) = 0. Then
there exists a unique continuous function g defined in a neighbourhood N (x0 ) of x0
such that f (x, g(x)) = 0 for all x ∈ N (x0 ).
Proof Define D = ∂∂ yf (x0 , y0 ). Since f is continuous on U and ∂∂ yf is continuous

at (x0 , y0 ), we can find a closed rectangle
I = [x0 − , x0 + ] × [y0 − , y0 + ]

inside U such that D1 ∂∂ yf (x, y) − 1 < 21 for all (x, y) ∈ I and D1 f (x, b) < 2δ for
x0 − ≤ x ≤ x0 + .
Define X = {y : I → R is continuous and y(x0 ) = y0 with |y(x) − y0 | ≤ δ for
all x ∈ I }. Clearly X is a closed subset of C(I ), the Banach space of all continuous
real functions on I with the supremum norm. Hence X is complete. Define the map
T : X → C(I ) by
1
(T y)(x) = y(x) − f (x, y(x)), x ∈ I.
D
Clearly T y is continuous for each y ∈ X . If y(x0 ) = y0 , then T y(x0 ) = y(x0 ) = y0
as f (x0 , y(x0 )) = f (x0 , y0 ) = 0. Now
1
T y0 − y0 = f (x, y0 )
D
1
< δ for |x − x0 | <
2
(by construction of I )
Also for y1 , y2 ∈ M, for x with |x − x0 | ≤

1 ∂ 1 ∂

|T y1 (x) − T y2 (x)| = y1 (x) − f (x, y1 (x)) − y2 (x) + f (x, y2 (x))
D ∂y D ∂y
1
< |y1 (x) − y2 (x)|, since
2
∂
(y − 1 f (x, y)) = (1 − 1 ∂ f (x, y) < 1
∂y D D ∂y 2
by choice of I.
So T y1 − T y2 ≤ 21 y1 − y2 . Also for y ∈ X ,
T y − y0 ≤ T y − T y0 + T y0 − y0
1
≤ y − y0 + T y0 − y0
2
1 1
< δ+ =δ
2 2
Thus T maps X into itself and is a contraction (with constant 21 ). Since X is complete
T has a unique fixed point y0 (x) which is a solution of f (x, y0 (x)) = 0 for x ∈
[x0 − , x0 + ].
Remark 6.7.2 The above theorem can be extended to functions taking values in a
Banach space after suitable modifications.
The problem of finding a curve which is invariant under a continuous transforma-

tion can also be reduced to the solution of a functional equation. Here, we consider a
relatively simple planar situation wherein the transformation F is differentiable with
a fixed point at (0, 0) such that F (0, 0) is invertible. Indeed we can transform
the
λ0
axes to coincide with the eigen vectors of F (0, 0) so that F (0, 0) = A = .
0μ
Under suitable assumptions, we can prove the existence of Lipschitzian solutions to
the problem of finding curves invariant under F in a neighbourhood of (0, 0). This
problem has been considered by Hadamard, Lattes and Montel (see [4]).
Theorem 6.7.3 Let F = ( f, g) be a continuously

differentiable mapping of a neigh-
λ0
bourhood U of (0, 0) in R2 with F = A = where 0 < |λ| < |μ| and |λ| < 1
0μ
or 0 < |μ| < |λ|, |λ| > 1. Then for each L > 0, we can find a c > 0 and a unique
function ϕ : [−c, c] → R satisfying
(i) |ϕ(s) − ϕ(t)| ≤ L|s − t|, s, t ∈ [−c, c] and
(ii) ϕ(0) = 0.
In fact ϕ is differentiable at zero and ϕ (0) = 0.
6.7 Applications to Functional Equations 137
Proof Without loss of generality we shall assume that 0 < |λ| < |μ| and |λ| < 1.
(In the second case F is replace by F −1 which exists in some neighbourhood of (0,
0)).
By the inverse function theorem there exists a continuously differentiable mapping
h defined in a neighbour V of(0, 0) satisfying h(x, g(x, y)) = y for (x, y) ∈ V . Also
h(x, y) = (0, μ1 ).(x, y) + o( x 2 + y 2 ), (x, y) → (0, 0).
Let L be any non-negative real number and c > 0 be such that D = {(x, y) ∈ R2 :
|x| ≤ c, |y| ≤ L|x|} ⊆ U ∩ V . Let X be the set of all functions ϕ : I = [−c, c] → R
such that ϕ(0) = 0 and |ϕ(s) − ϕ(t)| ≤ L|s − t| for all s, t ∈ I . It can be seen that
ρ : X × X → R+ defined by

ϕ(x) − ψ(x)
ρ(ϕ, ψ) = Sup
x∈I −{0} x
is a metric on X and that (X, ρ) is a complete metric space. Define T : X → X

by (T ϕ)(x) = h(x, ϕ( f (x, ϕ(x))), x ∈ I . By taking c small, we can insure that
f (x, ϕ(x)) ∈ I for x ∈ I and ϕ ∈ X .
For ϕ ∈ X , (T ϕ)(0) = 0. For fixed > 0 from the properties of f and h we have
| f (x, y) − f (u, v)| ≤ (|λ| + )|x − u| + |y − v|
and
1
|h(x, y) − h(u, v)| ≤ |x − u| + + |y − v|
|μ|
for all (x, y), (u, v) ∈ D.

So for ϕ ∈ X and (s, t) ∈ I × I , we have

1
|T (ϕ)(s) − T (ϕ)(t)| ≤ + + L((|λ| + ) + L)|s − t|
|μ|

As μλ < 1, this shows that we can choose > 0 small such that
|T (ϕ)(s) − T (ϕ)(t)| ≤ L|s − t|, s, t ∈ I.
So T maps X into itself. Again we can show that for ϕ, ψ ∈ X .

1
ρ(T ϕ, T ψ) ≤ + (|λ| + + 2L)ρ(ϕ, ψ)
|μ|
Decreasing and hence c we can choose the Lipschitz constant in the above inequality
to be less than 1. So T is a contraction and hence has a unique fixed point. ϕ(x) ∈ X .
Thus
ϕ(x) = h(x, ϕ( f x, ϕ(x)) = T ϕ(x).
Since h(x, g(x, φ(x)) = φ(x) and is inverse to g with respect to the second variable,
ϕ( f x, ϕ(x)) = g(x, ϕ(x)). Hence the existence of a Lipschitzian solution follows.
Finally we prove the existence of a unique solution to an operator equation using

Tan’s extension [9] of the contraction principle (see Theorem 5.4.7) as treated in
Subrahmanyam [8].
Let (Y, d) be a complete metric space, X a set and s : X → X , a function. For an
operator P on a sequentially closed subset S0 of Y X the operator equation f (x) =
P f (s(x)), x ∈ X is studied in the following. The solutions of this operator equation
are fixed points of the map P on S0 defined by P = P( f ◦ s). Theorem 6.7.4 below
can therefore provide a set of sufficient conditions for the existence of a unique
solution for this operator equation.
Theorem 6.7.4 Let (Y, d) be a complete metric space, X a set and s : X → X a

function. For any function f 0 ∈ Y X the set
S = { f ∈ Y X : Dx ( f 0 , f ) < +∞ for all x ∈ X }
is a sequentially complete Hausdorff gauge space under the family F of all pseudo-
metrics Dx defined by
Dx ( f, g) = sup{d( f (s k (x)), g(s k (x))) : k = 0, 1, 2, . . .}
(s 0 : X → X being the identity map).
The proof of Theorem 6.7.4 is omitted. The next result is a fixed point theorem
obtained from Tan’s Theorem (see [9]) and Theorem 6.7.4.
Theorem 6.7.5 (see [8]) Let (Y, d) be a complete metric space and s : X → X a
function. Let P : S0 → S0 be an operator on a sequentially closed subset S0 of Y X
with the topology of pointwise convergence such that
(i) D X ( f 0 , P( f 0 )) < +∞ for some f 0 ∈ S0 and all x ∈ X where Dx is as defined
in Theorem 6.7.4;
(ii) for each x ∈ X there is an a(x) such that given f, g ∈ S0 there exists a non-
negative integer n = n(x) with d(P f (x), Pg(x)) ≤ a(x)d( f (s n (x)), g(s n (x))).
Further, for each x ∈ X ,
0 < sup{a(s k (x)) : k = 0, 1, 2, . . .} < 1.
Then P has a unique fixed point in S0 .
Proof Define S = { f ∈ S0 : Dx ( f 0 , f ) < +∞ for each x ∈ X }. S = φ, since f 0 ∈

S. Also S is a sequentially complete Hausdorff gauge space under the family F =
{Dx : x ∈ X } of pseudometrics, in view of Theorem 6.7.4 and the assumption that
S0 is sequentially closed in Y X .
6.7 Applications to Functional Equations 139
For the operator P mapping S0 into itself, (i) implies that P( f 0 ) ∈ S. Further, for
f ∈ S and x ∈ X ,
Dx ( f 0 , P( f )) ≤ Dx ( f 0 , P( f 0 )) + Dx (P( f 0 ), P( f ))
≤ Dx ( f 0 , P( f 0 )) + A(x)Dx ( f 0 , f ) (by (ii))
< +∞
where A(x) = sup{a(s k (x)) : k = 0, 1, 2, . . .}. So P maps S into itself. By (ii) again
Dx (P( f ), P(g)) ≤ A(x)Dx ( f, g)
for each x ∈ X and f, g ∈ S. As 0 < A(x) < 1, P is strictly contractive and it maps
the Hausdorff sequentially complete gauge space (S, F) into itself. Therefore P has
a unique fixed point g0 ∈ S, by Theorem 5.4.7. Also, as g0 ∈ S, for each x ∈ X
sup{d( f 0 (s k (x)), g0 (s k (x))) : k = 0, 1, 2, . . .} < +∞.
The proof is complete.

An application of Theorem 6.7.5 is now given.
Corollary 6.7.6 (see [8]) Let B X be the space of all functions mapping X into a
Banach space (B, · ) over the real or complex field K . Let s : X → X , a : X → K ,
b : X → B and h : X × B → B be given functions. Suppose that for each x ∈ X
(i) given y1 , y2 ∈ B, h(x, y1 ) − h(x, y2 ) ≤ y1 − y2 ;
(ii) A(x) = sup{|a(s k (x))| : k = 0, 1, 2, . . .} < 1;
(iii) sup{b(s k (x)) : k = 0, 1, 2, . . .} < +∞ and sup{h(s k (x), 0) : k =
0, 1, 2, . . .} < +∞.
Then the functional equation
f (x) = a(x)h(x, f (s(x)) + b(x) (6.7.1)
has a unique solution f in B X .

Proof Set Y = (B, · ), S0 = B X and f 0 = 0 in Theorem 6.7.5. Then the class S
is simply the set
{ f ∈ B X : sup f (s k (x)) < +∞, k = 0, 1, 2, . . . , for all x ∈ X }.
The operator P of Theorem 6.7.5 is defined by
P( f (x)) = a(x)h(x, f (s(x))) + b(x), x ∈ X.
For this choice of the operator P and the space S, all the condition of Theorem 6.7.5
are satisfied. So the operator P has a unique fixed point in S. As any other solution
of the functional equation (6.7.1) is in S in view of (i), (ii) and (iii), (6.7.1) has a
unique solution in B X .
Example 6.7.7 The functional equation
f (x) = g(x) cos(x + f (x 2 )) (6.7.2)
has a unique solution in R R , the space of all real valued functions of a real variable,
where the function g is the one defined below:
⎧
⎪
⎨1/x, |x| > 1
g(x) = x, |x| ≤ 1, x irrational
⎪
⎩
x/2, otherwise.
Choosing B = R, a(x) = g(x), b = 0, h(x, y) = cos(x + y) and s(x) = x 2 , the

conditions of Corollary 6.7.6 are readily satisfied.
6.8 An Application to Commutative Algebra
In this section a short proof of the algebraic Weierstrass Preparation theorem due
to Gersten [2] is highlighted. To this end a few basic definitions are described. For
these Lang [5] and Zariski and Samuel [14] may be consulted.
Definition 6.8.1 A ring R is called a local ring if it is commutative and has a unique
maximal ideal.
Remark 6.8.2 If R is a local ring with the unique maximal ideal M, then x ∈ A − M
is a unit.
∞

Definition 6.8.3 For a ring R and an ideal I , suppose that I n = {0}. We can
n=1
define I n as a neighbourhood of 0 for each n. I n is the ideal generated by elements
of the form a1 a2 . . . an , ai ∈ I for i = 1, 2, . . . , n. In the same way we can say that a
sequence {xn } in R is Cauchy if given some power I k of I , there exists an integer M
such that for all m, n ≥ M, xm − xn ∈ I k . A sequence (xn ) in R is said to converge
to x in R if for each I k , there exists an integer M such that (xn ) ∈ x + I k for all
n ≥ M. R is said to be complete in the I -adic topology if every Cauchy sequence in
R converges.
Definition 6.8.4 Let R be a local ring and I = M the maximal ideal. R is called a
complete local ring if R is complete in the M-adic topology and we assume that the
M-adic topology is Hausdorff.
6.8 An Application to Commutative Algebra 141
Remark 6.8.5 Let k be a field and R = k[[X 1 , . . . , X n ]] the power series ring in n-
variables. Then R is a complete local ring. If M is the ideal generated by X 1 , . . . , X n
then R/M is isomorphic with k and so M is a maximal ideal. Any power series of
the form f (X ) = c0 − f 1 (X ) where c0 = 0 ∈ k and f 1 (X ) ∈ M is invertible as

f 1 (X ) ( f 1 (X ))2
(c0 − f 1 (X )) −1
= c0−1 1+ + + ··· .
c0 c02
So M is the unique maximal ideal and R is local. It can be readily verified that R is
complete.
For a ring A, the power series ring in n variables for n > 1 can be viewed
as the ring of power series in variable X n over the ring of power series in (n −
1) variables X 1 , . . . , X n−1 . Thus we have the identification A[[X 1 , . . . , X n ]] =
A[[X 1 , . . . , X n−1 ]][[X n ]]. When A is a field A[[X 1 , . . . , X n ]] is a complete local
ring. Moreover if R is a complete local ring, the power series ring R[[X ]] is a com-
plete local ring with the maximal ideal [M, X ], M being the maximal ideal of R. If
the power series an X n has unit constant a0 ∈ R − M, then the power series is a

unit in R[[X ]] as (a0 + h)−1 is a0−1 1 − ah0 + ah 2 − · · · .
2
Proposition 6.8.6 If A is a complete local ring with maximal ideal M, then it is a

metric space under the metric defined by

0 if a = a
d(a, a ) = −s
2 if a = a with a − a ∈ Ms − Ms+1 , M0 being A
The routine proof is omitted.
Proposition 6.8.7 ([2]) Let A be a complete local ring with maximal ideal M which
is Hausdorff in the M-adic topology. Let B = A[[t]] be the power series ring in
∞

one variable t over A. Define d1 ( f, f ) = sup d(ak , ak ) for f = ak t k and f =
k∈Z + k=0
∞

ak t k where ak , ak ∈ A and d is as in Proposition 6.8.6. Then (B, d1 ) is a complete
k=0
metric space.
Proof Since f ∈ B is uniquely identified by the sequence (a0 , a1 , . . . , . . . ) for

∞ ∞

f = ak t k , ai ∈ A, given a Cauchy sequence f n = ank t k in B, (ank ) is uni-
0 0
formly Cauchy in n. As (A, d) is complete, ank → ak ∈ A for each k > 0. Further
∞

d1 ( f n , f ) → 0 as n → ∞ where f = ak t k ∈ B. Thus (B, d1 ) is complete.
0
Definition 6.8.8 Let A be a complete local ring with maximal ideal M and B =
A[[t]]. A distinguished polynomial in B is of the form p0 + p1 t + · · · + pn−1 t n−1 +
t n where pi ∈ M.
The following is
∞

Theorem 6.8.9 (Algebraic Weierstrass Preparation Theorem) If f = ak t k ∈ B,
k=0
where ak ∈ A and if there exists n ∈ N such that ak ∈ M for k < n and an ∈ / M,
then f = up where u is a unit of B and p is a distinguished polynomial of degree n.
Also u and p are uniquely determined.
The proof of the algebraic preparation theorem is based on the following division
theorem for which Gersten [2] has given a proof using the contraction principle.
Theorem 6.8.10 ([2]) If f, b ∈ B and b ∈ M[t] and if n ∈ N, then f = q(t n +
b) + r where q, b ∈ B and r is a polynomial in t of degree < n. Further q and r are
uniquely determined.
Proof Define the operator E : B → B by x = p + E(x)t n , where p is a polynomial
in t of degr ee < n. Define T : B → B by T x = E( f − xb) for x ∈ B. So f −
xb = p + E( f − xb)t n . If x ∈ B, then f − x b = p + E( f − x b)t n . Noting that
T y = E( f − yb), and subtracting the latter equation from the former, we get −b(x −
x ) = p + (T x − T x )t n where p is a polynomial of degree < n. Note that the
coefficients of T x − T x involve only the coefficients of the left-hand side of degree
≥ n. So
1
d1 (T x, T x ) ≤ d1 (bx, bx ) ≤ d1 (x, x )
2
since b ∈ M[[t]]. Thus T is a contraction on B and hence has a unique fixed point
q ∈ B such that T q = q or f − qb = r + qt n where r is a polynomial of degree
< n. Thus f = q(t n + b) + r .
We can now deduce the algebraic Weierstrass preparation Theorem 6.8.9 from the
Division Theorem 6.8.10
∞

Proof Let f ∈ B with f = ak t k , where ai ∈ A and for some n ∈ N,ai ∈ M for
k=0

n−1
i < n with an ∈
/ M. Let b = ak t k . As ak ∈ M for k < n, b ∈ M[t]. Now by
k=0
Division Theorem 6.8.10, f = q(t n + b) + r where q, r ∈ B and r is a polynomial
in t of degree < n. Since b ∈ M[t], the distinguished polynomial p = t n + b is a
unit in B. So ( f − r ) p −1 = q or f p −1 = q + r p −1 = q ∈ b. So f = q p = pq
∞
with p being a distinguished polynomial. If q = bn t n , then a0 = b0 a0 so q is a
0
unit. The proof of uniqueness is left as an exercise.
6.9 A Proof of the Central Limit Theorem 143
6.9 A Proof of the Central Limit Theorem
Central limit theorem, a classical theorem of probability theory can be proved using
the contraction principle. Trotter [11] gave a proof based on Lindberg condition and
invoking the properties of certain linear non-expansive operators leading to a con-
traction. While Trotter’s proof completely avoids the use of characteristic functions,
Hamedani and Walter [3] gave a proof of the central limit theorem using the properties
of characteristic functions, for sub-independent identically distributed random vari-
ables, again using the contraction principle. Before describing the proof of Hamedani
and Walter [3], a few basic concepts of probability theory are stated below. See Kai
Lai Chung [1].
Definition 6.9.1 A measure space (, S , P) where S is a σ-algebra on the non-

empty and P is a measure on S is called a probability measure space if P() = 1.
A map X : → R is called a random variable if {x ∈ : X (x) ≤ r } ∈ S for each
r ∈ R. (Thus a random variable on is simply a measurable function).
Definition 6.9.2 A function F : R → R that is increasing and right continuous

with F(−∞) = lim F(t) = 0 and F(+∞) = lim F(t) = 1 is called a distribu-
t↓−∞ t↑+∞
tion function.
Definition 6.9.3 If X is a random variable on (, S , P), then μ defined by

μ(−∞, x] = P{ p ∈ : X ( p) ≤ x} = F(x) induces a measure on the Borel subsets
of R called the probability distribution measure of X and F is called the distribution
function of the random variable X . A family of random variables having the same
distribution is said to be identically distributed.
Definition 6.9.4 For r > 0 and a ∈ R given a random variable X with distribution
function F, the moment of X of order r about a is defined as
∞
E(X − a)r = (x − a)r μ(d x) = (x − a)r d F(x)
R −∞
μ being the probability distribution measure of X , provided the integral exists for
a = 0, r = 1 E(X ) is called the mean of X . The moments about the mean are called
central moments. The central moment of order 2 is called the variance of X and
is denoted by V ar (X ). The positive square root of V ar (X ) is called the standard
deviation of X , denoted by σ(X ).
Definition 6.9.5 The random variables {X i : 1 ≤ i ≤ n} are said to be

independent if
n
n
for any Borel sets Bi , i = 1, 2, . . . , n in R P (x : X i (x) ∈ Bi ) = i=1 P{x :
i=1
X i (x) ∈ Bi }.
Remark 6.9.6 The above definition of (stochastic) independence can be extended to

i = 1, 2,
any family of random variables in a natural way. If X i , . . . , nn are indepen-
!
n !
dent random variables with finite expectations, then E Xi = E(X i ).
i=1 i=1
Definition 6.9.7 For a random variable X with induced probability measure μ and
distribution function F, the characteristic function f (t) is defined as

f (t) = E(eit X ) = eit X (w) P(dw)

∞
= e μd(x) =
it x
eit x d F(x).
R −∞
Remark 6.9.8 For all t ∈ R, | f (t)| ≤ 1 = f (0) and f (−t) = f (t) (z being the com-
plex conjugate of z). f is uniformly continuous on R and for a, b ∈ R f a X +b (t) =
f X (at)eitb where f X denotes the characteristic function of X . For independent ran-
!
m !n
dom variables X i , i = 1, 2, . . . , n, E(e ) =
it Sn
E(e ) or f Sn =
it X i
f X i , Sn being
i=1 i=1

n
Xi .
i=1
Definition 6.9.9 The convolution of two distribution functions F1 and F2 is defined

to be the distribution function F such that
∞
F(x) = F1 (x − y)d F2 (y) for x ∈ R
−∞
and is written F = F1 ∗ F2 .
Remark 6.9.10 If X 1 and X 2 are independent random variables with distribution

function F1 and F2 , then X 1 + X 2 has the distribution function F1 ∗ F2 .
For these and the following theorem, Kai Lai Chung [1] may be consulted.
Theorem 6.9.11 If the distribution function F has a finite absolute moment of posi-
tive integral order k ≥ 1, then its characteristic function f has a bounded continuous
derivative of order k given by
∞
(k)
f (t) = (i x)k eit x d F(x)
−∞
Further

k
(i) j θk (k) k
f (t) = m ( j) t j + μ |t|
j=0
j! k!
where m ( j) is the moment of order j, μ(k) the absolute moment of order k and |θk | ≤ 1.
Definition 6.9.12 Two random variables X and Y on (, S , P) are said to be sub-
independent if the distribution of their sum is given by
FX +Y (t) = (FX ∗ FY )(t).
In terms of characteristic functions it is described by
φ X,Y (t, t) = φ X (t)φY (t) for t ∈ R
where φ X,Y (t, s), φ X (t), φY (t) are characteristic functions corresponding to (X, Y ),
X and Y respectively. The random variables X 1 , X 2 , . . . , X n are said to be sub-
independent if for each subset {X i1 , . . . , X ik } of {X 1 , . . . , X n }
!
k
φ X i1 ,...X ik (t, t, . . . , t) = φ X i (t) for t ∈ R
i=1
Remark 6.9.13 The concept of sub-independence is more general than that of inde-
x 2
−t /2
e√
pendence. Further the random variable with distribution function F(x) = −∞ dt
2π
is called the standard normal distribution
For the subsequent discussion culminating in the proof of central limit theorem
we introduce the following class of random variables
Definition 6.9.14 Let Rλ , λ ≥ 0 be the set of all random variables X on (, S , P)

such that
(i) E(|X |λ ) < +∞;
(ii) E(X λ ) = m k for k = 1, 2, . . ., [λ] where m k is the k-th moment of Z , the stan-
dard normal variable.
Let Mλ denote the set of distributions of X ∈ Rλ .
i Xt
e − eiY t
Definition 6.9.15 Define dλ : Mλ × Mλ → R+ by dλ (F, G) = sup E
t∈R |t|λ
where F and G are the distribution functions respectively of the random variables X
and Y .
In order to deduce the central limit theorem from the contraction principle we
need to set up appropriate complete metric spaces.
Proposition 6.9.16 (Mλ , dλ ) is a metric space.
Proof Clearly the main issue in the proof is to show that dλ (F, G) is finite-valued
for F, G ∈ Mλ . If φ1 and ψ1 are the real parts of the characteristic functions of F
and G in Mλ , then for n = [λ].
|φ1 (t) − ψ1 (t)| |φ(n) (n)

1 (ξ) − ψ1 (ξ)| 1
λ−n
= . λ−n
|t ||t|
n n! |t|
|φ(n) (n)
1 (ξ) − m n | |m n − ψ1 (ξ)|
≤
n!|t|λ−n n!|t|λ−n
by Theorem 6.9.11, in view of φ(k) (k)

1 (0) = ψ1 (0) for k ≤ n.
Now

|φ(n) (n)
1 (ξ) − ψ1 (0)| |eiξx − 1| n
≤ |x| d F x
n!|t|λ−n |t|λ−n
iξx
|e − 1| λ
≤ |x| d F(x)
|ξx|λ−n
≤ K E(|X |λ ) for some K ∈ R+
F being the distribution function if X . A similar estimate holds for the imaginary
parts of F and G. Thus dλ (F, G) < +∞.
Proposition 6.9.17 (Mn , dn ) is a complete metric space when n is an integer. For

n < λ, M λ ⊆ Mn where M λ is the completion of Mλ .
Proof Let (Fk ) be a Cauchy sequence of distributions in Mn with the corresponding

characteristic function φk . For each > 0 there is a positive integer K such that for
k, m ≥ K
|φk (t) − φm (t)|
<
|t n |
So φ(t) = lim φk (t) exists for each t ∈ R. Now h m defined by h m (t) =

⎧ k→∞ ⎫
⎨ n
(it) j ⎬ −n
φ (t) − mj t is a uniformly Cauchy sequence of continuous func-
⎩ m j! ⎭
j=0
tions converging uniformly to a continuous function h(t). Further h(0) = 0 as
h m (0) = 0 by definition of Mn . Thus

m
(it) j
φ(t) − mj = t m h(t)
j=0
j!
Since Fk has the same jth moment for U = 0, 1, . . . , n, and φ is continuous at zero
Fk converges weakly to F in the sense that E(X k Y ) → E(X Y ) for each bounded
random variable Y , X k being the random variable with distribution function Fk and X
the random variable with distribution function X . Using Fatou’s lemma the existence
of jth moment of F follows and φ(t) is j times differentiable at 0. So its jth derivative
for j = 0, 1, 2, . . . n must be the same as that of φk . Then F ∈ Mn . That a Cauchy
sequence in Mλ for λ > n is a Cauchy sequence in Mn is left as an exercise.
Definition 6.9.18 Let X, Y ∈ Rλ be sub-independent and have identical distribu-

tions. For α > 0 define Tα : Mλ → M0 as the map taking the distribution function
of X into the distribution function of X +Y . (Tα takes the characteristic function φ(t)
α
into φ2 αt ).
Proposition 6.9.19 Let αλ > 0 be such that Tα : Mλ → Mλ . Then

(i) Tα is non-expansive on (Mλ , dλ ) for αλ ≥ 2;
(ii) Tα is strictly contractive for αλ > 2.
Proof
2t
φ − ψ2 t
α α
dλ (Tα F, Tα G) = sup
t∈R |t|λ
t t
φ − ψ t sup φ +ψ t
α α t∈R α α
≤ sup
t∈R |α|λ | αt |λ
≤ 2α−λ dλ (F, G).
For αλ ≥ 2, dλ (Tα F, Tα G) ≤ dλ (F, G), while

for αλ > 2, dλ (Tα F, Tα G) ≤ μdλ (F, G)
where μ = 2α−λ < 1.
Pursuing a suggestion of J. Blum that central limit theorem ‘could be interpreted
and proved as a fixed point theorem’, Hamedani and Walter [3] proved the following
version of the central limit theorem.
Theorem 6.9.20 ([3]) Let (X m ) be a sequence of sub-independent identically dis-
tributed (s.i.i.d) random variables with mean 0 and variance 1 such that E(|X |λ ) <
∞ for some λ > 2. Then
1
2n
Xi → Z
2n/2 i=1
in distribution as n → ∞ and their distribution functions converge in the metric of

Mλ . Moreover the rate of convergence is governed by
λ
dλ (T√n 2 F, φ) < 2n(1− 2 ) (E(|X |λ ) + E(|Z |λ ))
Proof By Proposition 6.9.19, T = T√2 is a contraction on (Mλ , dλ ). So the iter-

ates of T n must converge to the unique fixed point in the completion of Mλ .
Since Tα takes the distribution function φ(t) of X into φ2 ( αt ), and φ2 ( √t 2 ) = φ(t)
t s2
for φ(t) = −∞ e− 2 ds, the distribution function of standard normal variate. So
|ψn (t) − e−t /2 |
2
dλ (φ, T F) → 0 as n → ∞. Thus lim sup

n
= 0, ψn being the char-
n→∞ t∈R |t|λ
acteristic function of T n F Thus (ψn (t)) converges to e−t /2 for all t in R. So by the
2
well-known Levy-Cramer theorem of probability theory the corresponding distri-

2n
bution functions converge. Thus 2−n/2 i=1 X i → Z in distribution. In view of the
error estimate from the contraction principle.
λ
dλ (φ, T n+1 F) ≤ 2(1− 2 )(n+1) dλ (φ, F)
λ
≤ 2(1− 2 )(n+1) dλ (φ, F)
λ
≤ 2(1− 2 )(n+1) [E(|x|λ ) + E(|z|λ )]
as dλ (φ, F) ≤ [E(|x|λ ) + E(|z|λ )]
(from the proof of Proposition 6.9.16).
The next proposition leads to the central limit theorem for s.i.i.d random variables.
Proposition 6.9.21 ([3]) Let the random variables X i , Yi , i = 1, 2 in Rλ have the

distribution functions Fi , G i , i = 1, 2 correspondingly and α > 0. If X 1 and X 2 are
sub-independent as also Y1 and Y2 , then
dλ (F, G) ≤ α−λ [dλ (F1 , G 2 ) + dλ (F2 , G 2 )

F1 +F2 G 1 +G 2
where F and G are distribution functions of α
and α
respectively.
Proof If φi is the characteristic function of X i and ψi is the characteristic function

of Yi , i = 1, 2, then

φ1 ( t )φ2 ( t ) − ψ1 ( t )ψ2 ( t ) φ 1 ( t ) φ 2 ( t ) − ψ 2 ( t )
α α α α α α α
≤
|t|λ tλ

ψ 2 ( t ) φ 1 ( t ) − ψ 1 ( t )
α α α
+
tλ
So

φ1 ( t )φ2 ( t ) − ψ1 ( t )ψ2 ( t )
α α α α
dλ (F, G) = sup
t∈R |t|λ

t −λ t −λ

≤ sup φ1 ( ) e dλ (F2 , G 2 ) + sup ψ2 ( ) e dλ (F1 , G 1 )
t∈R α α
t∈R
≤ α−λ [dλ (F1 , G 1 ) + dλ (F2 , G 2 )]
Theorem 6.9.22 (Central limit theorem [3]) Let {X n } be a sequence of sub-

independent random variables in Rλ for some λ > 2 whose distribution functions
1
n
belong to a bounded set in Mλ . Then √ X i → Z in distribution as n → ∞.
n i=1
Proof Let (Yn ) be an s.i.i.d sequence in Rλ . Repeated application of Proposi-

n+1
Xi
n+1
Yi
tion 6.9.21 for √ and √ gives
i=1
n+1 i=1
n+1
n+1
Xi
n+1
Yi
dλ √ , √
i=1
n + 1 i=1 n + 1
λ n
n Xi
2 n Yi λ
≤ √ dλ , √ + (n + 1)− 2 d λ (X n+1 , Yn+1 )
n+1 n + 1 n + 1
i=1 i=1
⎛ m ⎞
m λ
2 2m
n+1
2 2 X Y λ
dλ ⎝ ⎠ + (n + 1)− 2
i i
≤ , d λ (X i , Yi )
n+1 2 m/2 2 m/2
i=1 i=1 m i=2 +1
λ
n+1
≤ (n + 1)− 2 d λ (X i , Yi )
i=1
d λ being the metric in Mλ . As {X n } and {Yn } are both sequences whose distribution
functions are bounded in Mλ , the right-hand side of the last inequality converges to
zero as n tends
n+1to infinity. n+1
Xi Yi
So d λ √ , √ → 0 as n → ∞. By Theorem 6.9.20
n + 1 i=1 n + 1
2m i=1

Zi 2m
Yi
dλ , → 0 as m → ∞ provided each Z i is a standard normal
i=1
2m/2 i=1 2m/2
variate. Choosing m to be the largest integer with 2m < n + 1, we get
λ

n+1
λ
(n + 1)− 2 d λ (Z i , Yi ) ≤ (λ + 1)− 2 (n − 2m )C
i=2m +1
≤ 2−m ( 2 ) (2m+1 − 2m )C
λ
(C, a constant)
As λ > 2, the right-hand side

(and hence theleft-hand side) of the last inequality
n+1
Yi
tends to 0 as n → ∞. So d λ Z , √ → 0 as n → ∞. Now
i=1
n+1
n+1 n+1
Xi Xi
n+1
Yi
dλ √ ,Z ≤ dλ √ , √
i=1
n+1 i=1
n + 1 i=1 n + 1
n+1
Yi
+ dλ √ ,Z
i=1
n+1

n+1
Xi
So √ converges to Z in distribution.
i=1
n+1
References
1. Chung, K.L.: A Course in Probability Theory. Academic, Cambridge. An imprint of Elsevier

Indian Reprint 2007
2. Gersten, S.M.: A short proof of the algebraic Weierstrass preparation theorem. Proc. Am. Math.
Soc. 88, 751–752 (1983)
3. Hamedani, G.G., Walter, G.G.: A fixed point theorem and its application to the central limit
theorem. Arch. Math. 43, 258–264 (1984)
4. Kuczma, M., Choczewki, B., Ger, R.: Iterative Functional Equations. Encyclopedia of Mathe-
matics and its Applications, vol. 32. Cambridge University, Cambridge (1990)
5. Lang, S.: Algebra. Addison-Wesley Publishing Company Inc, Boston (1995)
6. Nagumo, M.: Uber da anfangswertproblem partieller differentialgleichungen. Jpn. J. Math. 18,
41–47 (1942)
7. Ryder, G.H.: Solutions of a functional differential equation. Am. Math. Mon. 76, 1031–1033
(1969)
8. Subrahmanyam, P.V.: An operator equation in a Tychonoff space. Aequationes Math. 20, 166–
169 (1980)
9. Tan, K.K.: Fixed point theorems for nonexpansive mappings. Pac. J. Math. 41, 829–842 (1972)
10. Tisdell, C.C.: A note on improved contraction methods for discrete boundary value problems.
J. Differ. Equ. Appl. 18, 1773–1777 (1972)
11. Trotter, M.F.: An elementary proof of the central limit theorem. Arch. Math. 10, 226–234
(1959)
12. Utz, W.R.: The equation f (x) = a f (g(x)), research problems. Bull. Am. Math. Soc. 71, 138
(1965)
13. Walter, W.: An elementary proof of the Cauchy-Kowalevsky theorem. Am. Math. Mon. 92,
115–126 (1985)
14. Zariski, O., Samuel, P.: Commutative Algebra, vol. II. Van Nostrand Company Inc, New York
(1969). East-West Edition, New Delhi
Chapter 7
Caristi’s Fixed Point Theorem
7.1 Introduction
In 1976, Caristi [4] published a novel generalization of the contraction principle.

Using transfinite arguments which was also later simplified by Wong [15]. Brondstedt
[3] provided an alternative proof by introducing an interesting partial order. On the
other hand, Ekeland [6] established a variational principle whence deducing Caristi’s
theorem. Brezis and Browder [2] proved an ordering principle also leading to this
fixed point theorem. Subsequently Altman [1], Turinici [14, 15] and others have
extended this principle. In this chapter, we discuss some of these as well as proofs
of Caristi’s theorem by Kirk [8], Penot [10] and Seigel [11]. That both Ekcland’s
principle and Caristi’s theorem characterize completeness is also brought out.
7.2 Siegel’s Proof of Caristi’s Fixed Point Theorem
Caristi’s fixed point theorem is the following.
Theorem 7.2.1 Let (X, d) be a complete metric space and φ : X → R+ be a lower

semicontinuous function (i.e. φ(x) ≤ lim inf φ(xn ) whenever lim xn = x). If T :
n→∞ n→∞
X → X is a map such that d(x, T x) ≤ φ(x) − φ(T x) for all x ∈ X , then T has a
fixed point.
Siegel’s proof [11] is detailed below, based on a few definitions and lemmata,
φ, T and X being as in Theorem 7.2.1 above.

Definition 7.2.2 Let = { f : X → X with d(x, f x)) ≤ φ(x) − φ( f (x)) for x ∈
X }. Define T = { f ∈ : φ( f ) ≤ φ(T )}.
Lemma 7.2.3 Both and T are closed under compositions. If φ is lower

semicontinuous, then these classes are closed under countable compositions.
https://doi.org/10.1007/978-981-13-3158-9_7
152 7 Caristi’s Fixed Point Theorem
Proof For f 1 , f 2 ∈ and x ∈ X , d(x, f 2 ( f 1 (x))) ≤ d(x, f 1 (x)) + d( f 1 (x),

f 2 ( f 1 (x))) ≤ φ(x) − φ( f 1 (x)) + φ( f 1 (x)) − φ( f 2 ( f 1 (x)) = φ(x) − φ( f 2 ( f 1 (x))).
So f 1 ◦ f 2 ∈ . If f 1 , f 2 ∈ T , then
φ( f 2 ( f 1 (x)) ≤ φ( f 1 (x)) (since f 2 ∈ T )

≤ φ(x) (since f 1 ∈ T ).
Thus f 1 ◦ f 2 ∈ T .
Let ( f n ) be a sequence in . For x ∈ X , define xn = f n ( f n−1 . . . ( f 1 x)) . . . ) =
f 1 ◦ f 2 · · · ◦ f n (x) for n ∈ N. Since d(xi , xi+1 ) ≤ φ(xi ) − φ(xi+1 ) for all i ∈ N and φ
is non-negative, {φ(xi )} is a non-increasing sequence of non-negative numbers which
is convergent. Consequently (xn ) is a Cauchy sequence in the complete metric space
(X, d) and so converges to some element x in X . Now d(xn , x) = lim d(xn , xk ) ≤
k→∞
φ(xn ) − lim inf φ(xk ) ≤ φ(xn ) − φ(x) (in view of the lower semicontinuity of φ).
k→∞
Thus for f n ∈ φ, n ∈ N, lim xn = lim f 1 ◦ . . . f n (x) = x exists for each x ∈ X .
n→∞ n→∞

n
Let x = f (x) = lim f i (x). Now
n→∞
i=1
d(x, f (x)) ≤ d(x, xn ) + d(xn , f (x))

≤ φ(x) − φ(xn ) + d(xn , x)
≤ φ(x) − φ(xn ) + φ(xn ) − φ(x)
(by the above argument)
≤ φ(x) − φ( f (x)) (as x = f (x))
Thus f ∈ and is closed under countable compositions.

n
For f n ∈ T , n ∈ N, let x = f (x) lim xn = lim f i (x) as before. Then
n→∞ n→∞
i=1
φ( f (x)) = φ(x) = φ( lim xn ) ≤ lim inf φ(xn ) (by the lower semicontinuity of φ)
n→∞ n→∞

n
≤ φ(T (x)) as φ(xn ) = φ( f i (x)) ≤ φ(T (x)) for each n ∈ N and x ∈ X , since
i=1

n
f i ∈ T . Thus T is also closed under countable compositions.
i=1
Definition 7.2.4 For A ⊆ X , define

(i) D(A) = Diameter of A = Sup{d(x, y) : x, y ∈ A};
(ii) r (A) = inf{φ(x) : x ∈ A};
(iii) Sx = { f (x) : f ∈ } for ⊆ .
Clearly for B ⊆ A, d(B) ≤ d(A) while r (B) ≥ r (A).
Lemma 7.2.5 For x ∈ X , D(Sx ) ≤ 2(φ(x) − r (Sx )).

7.2 Siegel’s Proof of Caristi’s Fixed Point Theorem 153
Proof For f 1 (x), f 2 (x) ∈ Sx ,
d( f 1 (x), f 2 (x)) ≤ d(x, f 1 (x)) + d(x, f 2 (x))

≤ φ(x) − φ( f 1 (x)) + φ(x) − φ( f 2 (x))
= 2φ(x) − (φ( f 1 (x)) + φ( f 2 (x))
≤ 2(φ(x) − r (Sx )).
Theorem 7.2.6 Let ⊆ be closed under compositions. Suppose x0 ∈ X .

(a) If is also closed under countable compositions, then there exists f ∈ such
that x = f (x0 ) and g(x) = x for all g ∈ .
(b) If the members of are continuous functions, then there exists a sequence of
functions ( f n ) ∈ and x = lim f n f n−1 . . . f 1 (x0 ) such that g(x) = x for all
n→∞
g ∈ .
Proof From the definition of r (Sx0 ), it follows that we can find f 1 ∈ such
that 0 ≤ φ( f 1 (x0 )) − r (Sx0 ) < 21 . Write x1 = f 1 (x0 ). As is closed under com-
positions, Sx1 ⊆ Sx0 and D(Sx1 ) ≤ 2(φ(x1 ) − r (Sx1 )) ≤ 2(φ( f 1 (x0 )) − r (Sx0 )) < 1.
Thus proceeding inductively we can obtain a sequence of functions f n such that
xn+1 = f n (xn ), Sxn+1 ⊆ Sxn and D(Sxn ) < n1 .
∞
k
If (a) is true, define f = f n and x = f (x0 ). Since x = lim f j (x j ),
k→∞
n=1 j=i+1
∞

x ∈ Sxi for each i. Since lim D(Sxn ) = 0, x ∈ Sxn . We claim that for g ∈ ,
n→∞
⎛ ⎞ n=1
∞

g(x) = x. Since g(x) = g ⎝ f i (xn )⎠, g(x) ∈ Sxi for each i ∈ N. g(x) = x as
j=i+1
lim D(Sxn ) = 0.
n→∞
Suppose (b) is true and x = lim f n f n−1 . . . f 1 (x0 ) = lim xn . Since xk ∈ Sn for
n→∞ n→∞
∞

k ≥ x, x ∈ S n for all n. As D(Sxn ) = D(S xn ), {x} = S xn . For g ∈ , g(xn ) ∈ Sxn
n=1
for each n and by the continuity of g, lim g(xn ) = g(x). So for any given > 0 we
n→∞
can find n 0 such that B(g(x), ) ∩ Sxn = φ for n > n 0 . So for n > n 0 , d(g(x), x) <
+ n1 . Thus d(g(x), x) ≤ . As > 0 is arbitrary g(x) = x.
Remark 7.2.7 Caristi’s theorem 7.2.1 follows upon setting = {T n : n ∈ N} the

set of finite iterates of T and T itself.

Proof Let T : X → X be a contraction with d(T (x), T (y)) ≤ kd(x, y) for x, y ∈ X

where 0 < k < 1. Define φ : X → R+ by φ(x) = d(x,T 1−k
x)
. Clearly φ is continuous.
2
Since d(x, T x) + φ(T x) = d(x, T x) + d(T1−k
x,T x)
≤ d(x, T x) + 1−k
k
d(x, T x) =
d(x,T x)
1−k
= φ(x). Thus d(x, T x) ≤ φ(x) − φ(T x) for all x ∈ X . Hence by Caristi’s
theorem T has a fixed point.
Brondstedt’s proof [3] of Caristi’s theorem involved a partial order and an applica-
tion of Zorn’s lemma. Penot’s constructive proof [10] of Caristi’s theorem exploiting
this order is given below.
Remark 7.2.9 Penot’s proof of Caristi’s Theorem [10].
Define the binary relation ≤ on X by x ≤ y if d(x, y) ≤ φ(y) − φ(x). Clearly this
defines a partial order on X . Define M(x) = {y ∈ X : y ≥ x}. Define an increasing
sequence {xn } inductively as in the following. Choose x arbitrarily and when x1 , . . . xn
are given choose xn+1 ∈ M(xn ) with φ(xn+1 ) ≤ inf{φ(x) : x ∈ M(xn )} + n1 . So
xn+1 ≥ xn and for each x ∈ M(xn+1 ) ⊆ M(xn ) we have φ(x) ≥ inf φ{M(xn )} >
φ(xn+1 ) − n1 .
d(x, xn+1 ) ≤ φ(xn+1 − φ(x)
So the diameter of M(xn ) ≤ n2 . As φ is lower semicontinuous, each M(x) is closed.

Since {M(xn )} is a decreasing sequence of closed sets with diameter of M(xn ) tending
∞

to zero, by the completeness of X , M(xn ) = {x}, (invoking the Cantor intersection
n=1
theorem). Clearly x is a maximal element of X under this order. If y (= x) ≥ x then
∞

y ≥ xn for all n and y ∈ M(xn ) for all n and this would contradict M(xn ) = {x}.
n=1
Since T (x) ≥ x, by hypothesis, x = T (x).
Making use of Brondstedt order Kirk [8] gave a proof using Zorn’s lemma.
Remark 7.2.10 Kirk’s proof of Caristi’s Theorem [8].
For x, y ∈ X , define the partial order ≤ by x ≤ y if d(x, y) ≤ φ(x) − φ(y) as
before. Consider all totally ordered subsets of X (with respect to this order), with
respect to set-inclusion. Since every chain in this collection has an upper bound,
by Zorn’s lemma, there is a maximal totally ordered subset E = {xα : α ∈ I } of
X where I is totally ordered by xα ≤ xβ if and only if α ≤ β (α, β ∈ I ). Since
{φ(xα ) : α ∈ I } is a decreasing net of non-negative real numbers converging to r as α
increasing. So for > 0, there exists α0 ∈ I such that r ≤ φ(xα ) < r + for α ≥ α0 .
So for β ≥ α ≥ α0 , d(xα , xβ ) ≤ φ(xα ) − φ(xβ ) < . Thus {xα }α∈I is a Cauchy net
in the complete metric space X and hence converges to some x ∈ X . φ being lower
semicontinuous φ(x) ≤ r . Since d(xα , xβ ) ≤ φ(xα ) − φ(xβ ) letting β ↑, d(xα , x) ≤
φ(xα ) − r ≤ φ(xα ) − φ(x). Since E is maximal, x ∈ E. Also by hypothesis φ(x) −
φ(T (x)) ≥ d(x, T (x)). Thus xα ≤ x ≤ T (x) for all α ∈ I . By the maximality of x,
x = T (x).
As has already been pointed out, the direct converse of the contraction principle
is not true. However Kirk [8] has pointed out that if in a metric space every map
7.2 Siegel’s Proof of Caristi’s Fixed Point Theorem 155
satisfying the inequality of Caristi’s theorem has a fixed point, then the space is
complete.
Theorem 7.2.11 (Kirk [8]) Let (X, d) be a metric space and φ : X → R+ , any
lower semicontinuous map. If every map T : X → X satisfying d(x, T x) ≤ φ(x) −
φ(T (x)) has a fixed point, then (X, d) is complete.
Proof Let (xn ) be a non-convergent Cauchy sequence in (X, d). For each x ∈ X ,
define φ(x) = limn→∞ d(x, xn ). (Note that φ(x) is well-defined as d(x, xn ) is a
Cauchy sequence of real numbers and hence is convergent.) Given x ∈ X , define
the least natural number n(x) such that 0 < 21 d(x, xn ) < φ(x) − φ(xn(x) ). Define
T : X → X by T (x) = xn(x) and ψ : X → R+ by ψ(x) = 2φ(x), for x ∈ X . Since
|φ(x) − φ(y)| ≤ d(x, y), both φ and ψ are continuous. Clearly from the definition
of ψ, 0 < d(x, T x) ≤ ψ(x) − ψ(T x) for all x ∈ X . Clearly T has no fixed point.
Thus if X is not complete we can find a fixed point free map T : X → X satisfying
d(x, T x) ≤ ψ(x) − ψ(T (x)) for some lower semicontinuous map ψ : X → R+ .
Thus Caristi’s theorem is characteristic of completeness. The following simple
example shows that Caristi’s theorem applies even when T is not continuous.
7.2.12 Let X be [0, 1] with the usual metric and T : X → X be defined as

Example
x
if x = 1
Tx = 2 . Then |x − T x| ≤ φ(x) − φ(T (x)) for φ(x) = x. Clearly T is
1 if x = 1
discontinuous and has 0 and 1 as fixed points.
Remark 7.2.13 It may be clarified that while proofs of Caristi’s theorem by Caristi
[4], Wong [16], Kirk [8] and Brondsted [3] invoke some form of the Axiom of
Choice such as Zorn’s Lemma, the constructive proofs such as those by Penot [10]
and Siegel [11] are indeed based on the axiom of choice for countable families. On
the other hand, Manka showed that Caristi’s theorem can be proved without choice
using Zermelo’s fixed point theorem for special posets. (See Kirk [9] and Jachymski
[7] for detailed comments and references.)
Jachymski [7] noted that Nadler’s fixed point Theorem 5.3.5 can be deduced from
Caristi’s Theorem 7.2.1.
Theorem 7.2.14 (Jachymski [7]) Let (X, d) be a complete metric space and T :
X → cl B X (the set of all bounded non-empty closed subsets of X ) such that for some
α ∈ (0, 1). H (T x, T y) ≤ αd(x, y), x, y ∈ X . Then T admits a selection g : X → X
which satisfies the conditions of Caristi’s Theorem 7.2.1 and hence has a fixed point.
So does T .
Proof Let β ∈ (0, 1) be such that α < β. For each x ∈ X , the set {y ∈ T x :
β(x, y) ≤ d(x, T x)} is non-empty. By the Axiom of Choice there exists a map
g : X → X such that gx ∈ T x and d(x, gx) ≤ d(x, T x). So d(gx, T x) ≤ H (T x,
T (g(x))) ≤ αd(x, g(x)). So
1
d(x, g(x)) ≤ (βd(x, g(x)) − αd(x, g(x))
(β − a)
1
≤ [d(x, T x) − d(g(x), T g(x))]
(β − a)
≤ φ(x) − φ(g(x))
d(x, T x)
where φ(x) =
(β − a)
Thus g is a Caristi map, a selection from T and hence has a fixed point in X . So does
(T x,T y)
T . Further |φ(x) − φ(y)| ≤ d(x,y)+Hβ−α
≤ β−a
1+α
d(x, y).
Remark 7.2.15 Nadler noted that the multivalued map of his theorem may not admit
a selection which is a contraction. Let X be the unit circle in the complex plane. For
α α
each z = eiα , α ∈ [0, 2π], let T (z) = {ei 2 , ei(π+ 2 ) }, the set of square roots of z. For
z i = eiαi , i = 1, 2,

2 sin(α42 −α1 if α2 − α1 ≤ π
H (T z 1 , T z 2 ) =
2 cos(α42 −α1 if α2 − α1 > π
√ √
and H (T z 1 , T z 2 ) ≤ 22 d(z 1 , z 2 ) = 22 |2 sin(α42 −α1 |. Following Jachymski [7] if T has
a selection g which is a contraction and if g(1) = 1, then by the continuity of g at
α
1, for some α0 > 0, g(eiα ) = ei 2 for all α ∈ [0, α0 ). By the continuity of g at α∗ =
α α∗
sup{α0 ∈ (0, 2π) : g(eiα ) = ei 2 for all α ∈ [0, α0 )}. For α∗ < 2π, we get ei 2 =
α∗
ei( 2 +π) so that α∗ = 2π. While g(ei(2π− n ) ) → eiπ as n → ∞, by the continuity of g
1
α
at 1, ei(2π− n ) → −1 and g(ei 2 ) → 1. This is a contradiction. If g(1) = −1, a similar
1
contradiction results. Thus T has no contractive selection.
7.3 Ekeland’s Variational Principle
Ekeland [6] obtained a theorem in the setting of metric spaces that finds wide use
in solving optimization problems and partial differential equations. His proof uses a
partial order employed earlier by Brondsted and Rockafeller and Bishop and Phelps.
Theorem 7.3.1 (Ekeland [6]) Let (X, d) be a complete metric space and F : X →
R ∪ {+∞}, a lower semicontinuous function ≡ +∞ bounded from below. For each
> 0 and u ∈ X satisfying
inf F(x) ≤ F(u) ≤ inf F(x) +

x∈X x∈X
and every λ > 0 there exists v ∈ X such that F(v) ≤ F(u), d(u, v) ≤ λ and for
w = v, F(w) > F(v) − λ d(v, w).
7.3 Ekeland’s Variational Principle 157
The proof is based on the following lemma.
Lemma 7.3.2 Let (X, d) be a complete metric space. Define the binary relation
on X × R by (x1 , a1 ) (x2 , a2 ) iff (a2 − a1 ) + αd(x1 , x2 ) ≤ 0 where α is any given
positive number. Then is a partial order on X × R such that {(x, a) : (x1 , a1 ) ≤
(x, a)} is closed in X × R for each (x1 , a1 ) in X × R. Further if S is a closed subset
of X × R such that for all (x, a) ∈ S, there exists m ∈ R such that a ≥ m. Then for
every (x1 , a1 ) ∈ S, there exists for the ordering an element (x, a) which is maximal
and greater than (x1 , a1 ).
Proof That is a continuous partial order is easily verified. Define inductively

(xn , an ) in S, n ∈ N beginning (x1 , a1 ). If (xn , an ) is known, define
Sn = {(x, a) ∈ S : (xn , an ) (x, a)}

m n = inf{a ∈ R : (x, a) ∈ Sn }
Indeed m n ≥ m. Choose (xn+1 , an+1 ) ∈ Sn such that an − an+1 ≥ 21 (an − m n ). Now

Sn are closed and non-empty and Sn+1 ⊆ Sn for all n. Also
1 1
|an+1 − m n+1 | ≤ |an − m n | ≤ n |a1 − m|.
2 2
So for (x, a) ∈ Sn+1 we get
1
|an+1 − a| ≤ |a1 − m|
2n
1 1
d(xn+1 , x) ≤ n |a1 − m|
2 α
Thus Diam(Sn ) → 0 as n → ∞.
∞

As X × R is complete, Sn = {(x, a)}. From the definition of (x, a), (xn , an )
n=1
(x, a) for each n and in particular for n = 1. If (x0 , a0 ) ∈ S is greater than (x, a),
∞

then by transitivity (xn , an ) (x0 , a0 ) for each n. So (x0 , a0 ) ∈ Sn and hence
n=1
(x0 , a0 ) = (x, a). Thus (x, a) is maximal.
Proof of Theorem 7.3.1. Let S = {(x, a) : x ∈ X, a ≥ F(x)}. S, the epigraph of

F is a closed subset of X × R as F is lower semicontinuous. Set α = λ and
(x1 , a1 ) = (u, F(u)) and apply Lemma 7.3.2 to get a maximal element (v, a)
in S satisfying (u, F(u)) (v, a). The maximality relation can be rephrased as
(v, a) ≺ (w, b) ∈ S ⇒ αd(w, v) < a − b. As (v, a) ∈ S is maximal, a = F(v).
If w = v, a = F(v) > b ≥ F(w) implying that (v, a) is not maximal. So for
w = v, and w = v, (w, F(w)) (v1 , F(v)). Or F(w) − F(v) + αd(v, w) ≥ 0
or − λ d(v, w) < F(w) − F(v)) If for this > 0, u ∈ X is such that inf{F(x) :
x ∈ X } ≤ F(u) ≤ inf{F(x) : x ∈ X } + , then (u, F(u)) (v, F(v)). This implies

λ
d(u, v) < F(u) − F(v) ≤ F(u) − inf{ f (x) : x ∈ X } ≤ or d(u, v) ≤ λ and for
u = v, F(u) > F(v) − λ d(u, v).
Corollary 7.3.3 Let (X, d) be a complete metric space and F : X → R ∪ {+∞}

a lower semicontinous map which is bounded below and not identically +∞. Let
> 0 be given and u be a point in X such that
F(u) ≤ inf{F(x) : x ∈ X } +
Then there exists v ∈ X such that (i) F(v) ≤ F(u); (ii) d(u, v) ≤ 1 and for all w = v,
F(w) + d(v, w) > F(v).
Proof In Theorem 7.3.1 set λ = 1.
Corollary 7.3.4 Under the hypotheses of Corollary 7.3.3 given > 0, there exists
x ∈ X with f (x) < inf f + .
Corollary 7.3.5 (Caristi’s fixed point theorem)
Proof Let T : X → X be an operator such that d(x, T x) ≤ φ(x) − φ(T (x)) for all
x ∈ X , where φ is a non-negative lower semicontinuous function.
Proof By Corollary 7.3.3 for = 1, there exists v ∈ X such that for all u = v
φ(v) < φ(u) + d(v, u)
If v = T (v), then φ(T (v)) − φ(u) < d(u, T v) a contradiction. So v = T (v). Thus
T has a fixed point.
Indeed we can even prove a multivalued version of Caristi’s theorem.
Theorem 7.3.6 (Multivalued version of Caristi’s Theorem) Let (X, d) be a complete

metric space and φ : X → R ∪ {+∞} be a lower semicontinuous function which is
bounded below. Let T : X → 2 X be a multivalued map such that T (x) = φ for each
x ∈ X and d(x, y) ≤ φ(x) − φ(y) for all y ∈ T x. Then there exists x0 ∈ X with
x0 ∈ T x0 .
Proof As before set = 1 in Corollary 7.3.3. So there exists v ∈ X such that for all
y = v, φ(v) < φ(y) + d(v, y). We claim that v ∈ T (v). Otherwise for all y ∈ T (v),
d(v, y) > 0 and φ(v) − φ(y) < d(v, y). But by hypothesis d(v, y) ≤ φ(v) − φ(y).
Thus φ(v) − φ(y) < φ(v) − φ(y), a contradiction. So for some y ∈ T (v), d(v, y) =
0 or v ∈ T (v).
7.4 A Minimization Theorem 159
7.4 A Minimization Theorem
Takahashi [13] proved a minimization theorem which can be deduced from Ekeland’s
variational principle.
Theorem 7.4.1 (Takahashi [13]) Let (X, d) be a complete metric space and F :
X → R ∪ {+∞} a lower semicontinuous map bounded below. Suppose F satisfies
the (Takahashi) condition: there exists α0 > 0 such that for x ∈ X − Z , there exists
y = x with α0 d(y, x) ≤ F(x) − F(y) Z being the set {z ∈ X : F(z) = inf F} (the
X
set of possible minima of F}. Then there exists x0 ∈ X with F(x0 ) = inf F (or Z is
X
non-void).
Proof For 0 < α < α0 , by Ekeland’s principle (Corollary 7.3.4) there exists x ∈ X
such that F(x) < F(y) + αd(y, x), ∀ y = x. By Takahashi condition for z ∈ X −
Z , there exists y = z such that F(y) + αd(y, z) ≤ F(z). If x ∈/ Z , the last inequality
for the choice z = x contradicts its preceding inequality. So x ∈ Z is a minimizer of
F over X .
We now show that Takahashi’s minimization theorem implies Ekeland’s principle

(Theorem 7.3.1).
Theorem 7.4.2 Theorem 7.4.1 implies Theorem 7.3.1.
Proof Let x0 be the minimum guaranteed by Theorem 7.4.1. Define X 0 = {x ∈

X : F(x) ≤ F(x0 ) − λ d(x0 , x)}. Clearly x0 ∈ X 0 and X 0 = φ. Since F is lower
semicontinuous and λd is continuous, X 0 is closed. Further for x ∈ X 0 .

d(x0 , x) ≤ F(x0 ) − F(x) = F(x0 ) − inf f ≤
λ X
So d(xλ0 ,x) ≤ 1 or d(x0 , x) ≤ λ. We also have F(x0 ) ≤ F(x). We claim that for
x ∈ X 0 and y = x, F(y) > F(x) − λ d(x, y). Otherwise for some y ∈ X , y = x,
F(y) ≤ F(x) − λ d(x, y). Then

d(y, x0 ) ≤ d(x0 , x) + d(x, y)
λ λ λ
≤ F(x0 ) − F(x) + F(x) − F(y)
= F(x0 ) − F(y).
So y ∈ X 0 . So by Theorem 7.4.1 there exists x ∈ X such that F(x) = inf x∈X F(x).
This contradicts that F(y) < F(x0 ), a contradiction.
Theorem 7.4.3 Caristi’s Theorem 7.2.1 implies Ekeland’s Variational principle, viz.
Corollary 7.3.4.
Proof Let the hypotheses of Corollary 7.3.4 hold. Clearly d1 (x, y) = d(x, y)
defines an equivalent metric on X . For x ∈ x, define T (x) = {y ∈ X : F(x) ≥
F(y) + d1 (x, y), y = x}. If the conclusion of Corollary 7.3.4 is false, then Clearly
T (x) = φ for all x ∈ X . T is a multivalued map of X into 2 X − {φ} satisfying
F(y) ≤ F(x) − d1 (x, y) for y ∈ T x. Since (X, d1 ) is complete, by Caristi’s theo-
rem, T has a fixed point x0 . However for x0 ∈ T x0 the definition of T (x) is violated.
This contradiction shows that Corollary 7.3.4 is true.
Remark 7.4.4 Thus both Caristi’s fixed point theorem and Takahashi’s minimization
theorem are equivalent to Ekeland’s variational principle. In other words Caristi’s
theorem, Ekeland’s principle and Takahashi’s theorem are equivalent.
Since Carisit’s theorem is characteristic of completeness, both Takahashi’s the-
orem and Ekeland’s principle are also characteristic of completeness. Sullivan [12]
proved that Ekeland’s principle implies the completeness of the metric space.
Theorem 7.4.5 (Sullivan [12]) Let (X, d) be a metric space and F : X → R ∪
{+∞}, F ≡ +∞ be any continuous map bounded below such that for each >
0, there exists x0 ∈ M satisfying F(x0 ) ≤ inf F + and for all x = x0 , F(x) >
F(x0 ) − d(x, x0 ).
Proof Suppose that (xn ) is a non-convergent Cauchy sequence in X . For each x ∈
F, the x → F(x) = lim d(x, xn ) is well-defined, continuous on X and bounded
n→∞
below by 0. Since (xn ) is Cauchy inf F = 0 and F ≡ +∞. Let 0 < < 1 then by
assumptions on F, we can find x0 ∈ X such that F(x0 ) ≤ and F(x) > F(x0 ) −
d(x, x0 ) for all x = x0 . As (xn ) is non-convergent F(x0 ) > 0.
Let η be any positive number less than (1−)F(x 4
0)
. Since (xn ) is Cauchy we can
find x N0 such that d(x N0 , xn ) ≤ η for all n ≥ N0 so that F(x N0 ) = lim (x N0 , xn ) ≤ η.
n→∞
Setting x = x N0 in the inequality
0 < F(x0 ) < F(x) + d(x0 , x)
we get
0 < F(x0 ) < F(x N0 ) + d(x0 , x N0 )

< F(x N0 ) + [d(x0 , xn ) + d(xn , x N0 )] for n ≥ N0
≤ F(N0 ) + lim [d(x0 , xn ) + d(xn , x N0 )]
n→∞
≤ F(x N0 ) + [F(x0 ) + η]
0 < F(x0 ) ≤ η + η + F(x0 )
or
(1 − )F(x0 )
0 < (1 − )F(x0 ) < 2η <
2
This contradiction shows that F(x) cannot be positive for all x ∈ X . In other words
(xn ) must converge in X or (X, d) is complete.
7.4 A Minimization Theorem 161
The next theorem shows that Takahashi’s minimization theorem also characterizes
completeness.
Theorem 7.4.6 A metric space (X, d) is complete if for every uniformly continuous
function f : X → R ∪ {+∞}, f ≡ +∞ and every x ∈ X with inf X f < f (x )
there exists z ∈ X with z = x and f (z) + d(x , z) ≤ f (x ), there exists x0 with
f (x0 ) = inf f .
X
Proof Let (xn ) be a Cauchy sequence in X . Define f : X → R ∪ {+∞} by f (x) =

lim d(x, xn ) for all x ∈ X . Clearly f is well-defined and uniformly continuous. Also
n→∞
f (x0 )
inf f (x) = 0. Let f (x0 ) > 0. Then there exists xm ∈ X with xm = x0 , f (xm ) < 3
x∈X
and d(xm , x0 ) − f (x0 ) < f (x0 ). Thus we have 3 f (xm ) + d(xm , x0 ) < f (x0 ) +
2 f (x0 ) = 3 f (x0 ). So there exists x ∈ X with f (x) = inf f (x) = 0. Thus
lim d(xm , x) = 0 or (xn ) converges in X . Thus (X, d) is complete.
7.5 An Application of Ekeland’s Principle
We begin with some basic ideas of calculus.
Definition 7.5.1 For locally convex linear topological spaces X and Y and U ⊆ X
open, the map F : X → Y is said to be Gateaux differentiable at u ∈ U if for each
F(u + th) − F(u)
h ∈ X , lim exists. This limit, denoted by d F(u, h) is called the
t→0 t
Gateaux differential at u and is homogeneous.
Remark 7.5.2 If the Gateaux differential is linear and continuous, it is called Gateaux
derivative. Since the Gateaux differential of a discontinuous linear function is itself,
it follows that a Gateaux differential can be linear without being continuous.
Definition 7.5.3 A map F : X → Y where X and Y are normed linear spaces is said
to be Frechet differentiable at u if there exists a bounded linear function L such that
F(u + h) = F(u) + L(h) + o(h).
Remark 7.5.4 If F is Frechet differentiable then it is Gateaux differentiable and both

the Gateaux and Frechet derivatives coincide. While a Frechet differentiable function
is continuous, a Gateaux differentiable function need not be continuous. Indeed even
if the Gateaux differential is linear and continuous, the Frechet derivative may not
exist.
We proceed to describe an application of Ekeland’s principle.
Theorem 7.5.5 Let X be a Banach space and φ : X → R, a lower semicontinuous

function bounded below. Suppose φ has Gateaux derivative on X . Then for each
> 0 there exists x ∈ X such that φ(x ) ≤ inf φ + and Dφ(x ) ≤ .
X
Proof From Ekeland’s Principle (Theorem 7.3.1), there exists x ∈ X such that
φ(x ) ≤ φ(x) + x − x for all x ∈ X . Let h ∈ X and t > 0. Setting x = x + th
in the above inequality we get
1
[φ(x ) − φ(x + th)] ≤ h
t
As t → 0 we get −Dφ(x )(h) ≤ h. Writing −h for h we get Dφ(x )h ≤ h.

Thus |Dφ(x )h| ≤ h. So Dφ(x ) is bounded and Dφ(x ) ≤ .
We can even ensure the existence of a critical point for φ which is a minimum
under an additional condition.
Definition 7.5.6 Let X be a Banach space and φ : X → R be a function which is

continuously differentiable. φ is said to satisfy the Palais–Smale condition if each
sequence (xn ) in X for which φ(xn ) is bounded and φ (xn ) → 0 in X has a convergent
subsequence.
Theorem 7.5.7 Let X be a Banach space and φ : X → R be a C 1 function bounded

below satisfying the Palais–Smale condition. Then there exists x0 ∈ X such that
φ(x0 ) = inf φ and φ (x0 ) = 0.
X
Proof Setting = 1
n
in Theorem 7.5.5 we get a sequence (xn ) ∈ X such that
1 1
φ(xn ) ≤ inf φ + and φ (xn ) ≤
X n n
As φ satisfies the Palais–Smale condition (xn ) has a subsequence (xn k ) converging

to x0 ∈ X . Clearly φ(x0 ) = inf φ and φ being continuous, φ (xn k ) = φ (x0 ) and
X
φ (x0 ) = 0.
For other applications De Figueiredo [5] may be consulted.
References
1. Altman, M.: A generalization of the Brezis-Browder principle on ordered sets. Nonlinear Anal.
6, 157–165 (1982)
2. Brezis, H., Browder, F.E.: A general principle on ordered sets in nonlinear functional analysis.
Adv. Math. 21, 355–364 (1976)
3. Brondstedt, A.: On a lemma of Bishop and Phelps. Pac. J. Math. 55, 335–341 (1974)
4. Caristi, J.: Fixed point theorems for mappings satisfying inwardness conditions. Trans. Am.
Math. Soc. 125, 241–251 (1976)
5. De Figuiredo, D.G.: Lectures on the Ekeland Variational Principle with Applications and
Detours. T.I.F.R (1972)
6. Ekeland, I.: On the variational principle. J. Math. Anal. Appl. 47, 324–353 (1974)
References 163
7. Jachymski, J.R.: Caristi’s fixed point theorem and selections of set-valued contractions. J. Math.
Anal. Appl. 229, 55–67 (1998)
8. Kirk, W.A.: Caristi’s fixed point theorem and metric convexity. Colloq. Math. 36, 81–86 (1976)
9. Kirk, W.A.: Contraction mappings and extensions. In: Kirk, W.A., Sims, B. (eds.) Handbook
of Metric Fixed Point Theory. Springer, Berlin (2001)
10. Penot, J.P.: A short constructive proof of the Caristi fixed point theorem. Publ. Math. Univ.
Paris 10, 1–3 (1976)
11. Siegel, J.: A new proof of Caristi’s fixed point theorem. Proc. Am. Math. Soc. 66, 54–56 (1977)
12. Sullivan, F.: A characterization of complete metric spaces. Proc. Am. Math. Soc. 83, 345–346
(1981)
13. Takahashi, W.: Existence theorems generalizing fixed point theorems for multivalued mappings.
In: Baillon, J.B., Thera, M. (eds.) Fixed Point Theory and Applications, vol. 252, pp. 387–406.
Longman Scientific & Technical, Harlow (1991)
14. Turinici, M.: A generalization of Altman’s ordering principle. Proc. Am. Math. Soc. 90, 128–
132 (1984)
15. Turinici, M.: Remarks about a Brezis-Browder principle. Fixed Point Theory 4, 109–117 (2003)
16. Wong, C.S.: On a fixed point theorem of contractive type. Proc. Am. Math. Soc. 57, 282–284
(1976)
Chapter 8
Contractive and Non-expansive
Mappings
In this chapter, fixed points of contractive and non-expansive mappings are studied,
as also the convergence of their iterates.
8.1 Contractive Mappings
The following definition is recalled.
Definition 8.1.1 A mapping T : X → X where (X, d) is a metric space is called

contractive if d(T x, T y) < d(x, y) for all x, y ∈ X with x = y.
A contractive mapping need not have a fixed point in a complete metric space as
seen from the following example.
Remark 8.1.2 The map x → e−x mapping R+ into itself has no fixed point. For 0 ≤
x < y, 0 < e−x − e−y = (= |e−x − e−y |) = e−ξ (y − x) < y − x = |x − y| by the
Mean-value Theorem.
In this connection the theorem below leads to the existence of a fixed point for
contractive mappings under a set of suitable conditions.
Theorem 8.1.3 Let (X, d) be a metric space and T : X → X a continuous map.

Suppose
(i) d(T x, T 2 x) ≤ d(x, T x) for all x ∈ X ;
(ii) for x = T x, there exists a natural number n = n(x) such that
d(T n x, T n+1 x) < d(x, T x);
(iii) for some x ∈ X , xn k = T n k (x ), a subsequence of T -iterates at x converges to

u ∈ X.

https://doi.org/10.1007/978-981-13-3158-9_8
166 8 Contractive and Non-expansive Mappings
Then u is a fixed point of T .
Proof As T is continuous, the map φ(x) = d(x, T x) is a continuous map of X

into R+ . As xn k converges to u, φ(xn k ) → φ(u). Since φ(T x) ≤ φ(x), the sequence
{φ(xn )} is a non-increasing sequence of non-negative real numbers, xn being T n (x ).
So {φ(xn )} converges to a non-negative real number r . φ(xn k ) being a subse-
quence converging to φ(u), φ(u) = r . If r = φ(u) > 0, then by (iii), there exists
m = m(u), a natural number such that φ(T m (u)) < φ(u) = r . As {xn k } converges
to u, {T m (xn k )} converges to T m (u) by the continuity of T m . So {φ(T m (xn k ))}
converges to φ(T m (u)). Since this is a subsequence of φ(xn ) which converges to
φ(u), φ(T m (u)) = φ(u) = r contradicting that φ(T m (u)) < φ(u). So r = 0. Or
u = T (u).
Corollary 8.1.4 (Edelstein [6]) Let (X, d) be a metric space and T : X → X a

contractive map. If for some x ∈ X , {xn k = T n k x } is a subsequence of the T -iterates
at x converging to u ∈ X , then u is a unique fixed point of T . Further {T n x }
converges to u.
Proof While the existence follows from Theorem 8.1.3 the proof of the uniqueness is
left as an exercise. Since (xn k ) converges to u, given > 0, there exists N (), a posi-
tive integer such that d(u, xn k ) < . For all k ≥ N (). Now for n > n N () then for p =
n − n N () , d(u, xn ) = d(T p u, T p xn N ) ≤ d(T p−1 u, T p−1 xn N ) < d(u, xn N ) < . So
(xn ) converges to u.
Corollary 8.1.5 Let (X, d) be a compact metric space. Then every contractive self
map on X has a unique fixed point to which every sequence of iterates converges.
Remark 8.1.6 Reduction of the problem of finding the fixed point of the operator T
to that of finding the zero of the map x → d(x, T x) in the metric setting has been
suggested for instance in Dieudonne [3].
Even as Kannan’s Corollary 5.2.2 or the more general fixed point Theorem 5.2.1
apply to mappings which are not necessarily continuous, Theorem 8.1.3 has an
analogue for operators which need not be continuous.
Theorem 8.1.7 Let (X, d) be a compact metric space and T : X → X a mapping

satisfying the following conditions:
(i) for some non-negative real numbers a1 , a2 , a3 , a4 and a5 for x, y ∈ X with
x = y
d(T x, T y) < a1 d(x, T x) + a2 d(y, T y) + a3 d(x, T y) + a4 d(y, T x) + a5 d(x, y);
a1 +a3 +a5
(ii) a2 + a3 < 1 and 1−a2 −a3
= 1.
Then T has a fixed point. If further a4
1−a3 −a5
< 1, then the fixed point is unique.
8.1 Contractive Mappings 167
Proof Since d(x, T x) ≥ 0 for x ∈ X , let r = inf{d(x, T x) : x ∈ X }. Let xn ∈ X be

such that lim d(xn , T xn ) = r . As X is compact, T (xn ) has a subsequence yk =
n→∞
T (xn k ) converging to y.
Now d(y, T y) ≤ d(y, T (xn k )) + d(T xn k , T y). But
d(T xn k , T y) ≤ a1 d(xn k , T xn k ) + a2 d(y, T y) + a3 d(xn k , T y) + a4 d(y, T xn k ) + a5 d(xn k , y)
d(xn k , T y) ≤ d(xn k , T xn k ) + d(T xn k , y) + d(y, T y) and d(xn k , y) ≤ d(xn k , T xn k )

+ d(T xn k , y)
Using these inequalities we get
d(y, T y) ≤ (a1 + a3 + a5 )d(xn k , T xn k ) + (a2 + a3 )d(y, T y) + (1 + a4 )d(y, T xn k )

+ (a3 + a5 )d(T xn k , y)
So
(1 − a2 − a3 )d(y, T y) ≤ (a1 + a3 + a5 )d(xn k , T xn k ) + (1 + a4 )d(y, T xn k )

+ (a3 + a5 )d(T xn k , y)
Proceeding to the limit as k → ∞ in the above, we get
(1 − a2 − a3 )d(y, T y) ≤ (a1 + a3 + a5 )r
1 +a3 +a5
As a1−a 2 −a3
= 1, it follows that d(y, T y) = r . If y = T y, then (ii) gives (1 − a2 −
a3 )d(T y, T y) < (a1 + a3 + a5 )d(y, T y). So d(T y, T 2 y) < d(y, T y) = r , contra-
2
dicting that r = inf{d(x, T x) : x ∈ X }. Thus y = T y.

If x and y are two fixed points of T , then
d(x, y) = d(T x, T y) < (a3 + a5 )d(x, y) + a4 d(y, x)
Thus (1 − a3 − a5 )d(x, y) < a4 d(y, x). Since a4

1−a3 −a5
< 1, x = y.
Corollary 8.1.8 (Reich [21]) Let (X, d) be a compact metric space and T : X → X
be a map such that for x = y, x, y ∈ X
1
d(T x, T y) < [d(x, T x) + d(y, T y)]
2
Then T has a unique fixed point.
Edelstein’s result (Corollary 8.1.4) may not be true for contractive maps on closed
bounded sets which are not compact, as shown by the following.
Example 8.1.9 (Ira Rosenholtz [22]) Let c0 be the Banach space of all null real
sequences with the supremum norm and S, the closed unit ball in c0 . Define T :
S → S by
T (x1 , x2 , . . . , xn , . . . ) = (1, a1 x1 , a2 x2 , a3 x3 , . . . , an xn , . . . )
where an is a sequence of positive real numbers such that each ak is less than 1 and
n
n
+1
Pn = a j is bounded away from zero. For example an = 22n +2 . Clearly T maps S
j=1
into the boundary of S as T x = 1 for x ∈ S. Also if x = (x1 , x2 , . . . , xn , . . . ) is a
fixed point of T , then T x = (y1 , . . . , yn , . . . ) where y1 = 1 and yn = an−1 xn−1 for
(1 + 1k )
n−1
1 + 2n−1
1
1
n > 1, so that yn = Pn−1 = 2
≥ ≥ . As (yn ) ∈
/ c0 , T cannot
k=1
(1 + 2k−1 ) 1 2 2
have a fixed point in S.
Edelstein’s theorem can be used to solve a classical problem of geometry. Given

a triangle ABC with sides a (= BC), b (= CA) and c (= AB), the problem is to con-
struct the triangle with prescribed lengths of angle bisectors. Originally the problem
required the solution to be constructed using ruler and compass. This problem has
been independently posed by several persons including Brocard and Terquem. See
[4, 16] for a detailed history of this problem. Indeed the internal bisectors of the
angles of a triangle are concurrent while the external bisectors form a triangle. Van
den Berg [24] showed that given three positive numbers m, n, p the necessary and
sufficient conditions for the existence of a triangle with lengths of exterior bisectors
of angle m, n, p is that the maximum of {mn, np, pm} is larger than the sum of the
remaining two of this triplet. He also noted that in this case the problem admits of
two solutions. While Korselt showed that it is impossible to construct the triangle by
ruler and compass, given the lengths of external bisectors of angles, Neiss proved the
impossibility of construction by ruler and compass the triangle for which the lengths
of the internal bisectors of the angles of the triangle are given.
In what follows we consider the problem of constructing the triangle, given the
lengths of internal bisectors of angles. Let ABC be the triangle with sides a, b, c and
let m, n and p be the lengths of internal bisectors of angles A, B and C respectively.
A
N
P
p
n m
B M C
Area of ABC = Area of ABM + Area of AMC. So
1 1 A 1 A
bc sin A = cm sin + bm sin
2 2 2 2 2
or
2bc A
m= cos .
b+c 2
Since
a 2 = b2 + c2 − 2bc cos A

A
= b + c − 2bc 2 cos
2 2 2
−1
2
we get
⎫
m = (b+c)
1
bc[(b + c)2 − a 2 ⎪⎬

n = (c+a)
1
ca[(c + a)2 − b2 ] (8.1.1)
⎪
⎭
p = (a+b)
1
ab[(a + b)2 − c2 ]
This can be equivalently written as
bc[(b + c)2 − a 2 ] − (b + c)2 m 2 = 0

ca[(c + a)2 − b2 ] − (c + a)2 n 2 = 0
ab[(a + b)2 − c2 ] − (a + b)2 p 2 = 0
Thus the problem is reduced to solving for a, b, c in terms of m, n and p using if

necessary elimination theory.
Mironescu and Panaitopol [16] viewed it as a problem in fixed point theory. This
is described in the following theorem.
Theorem 8.1.10 (Mironescu and Panaitopol [16]) The problem of finding the trian-
gle, given the lengths of internal angle bisectors is equivalent to a problem of finding
the fixed point of a suitable mapping.
Proof The set of Eq. (8.1.1) can be rewritten suitably. For instance the first equation
in (8.1.1) is
bc[(b + c)2 − a 2 ] − (b + c)2 m 2 = 0
It can be written as
(b + c)2 − (b − c)2
4m 2 = [(b + c)2 − a 2 ]
(b + c)2
or
(b − c)2 a 2
4m 2 = (b + c)2 + − [a 2 − (b − c)2 ]
(b + c)2
2
(b − c)a
= (b + c) ± − [a ± (b − c)]2
(b + c)
(b−c)a
Eliminating (b+c)
between those two equations one gets

2(b + c) = 4m 2 + (c + a − b)2 + 4m 2 + (a + b − c)2
Define x, y, z by
a = y + z, b = z + x, c = x + y
whence
b+c−a c+a−b a+b−c
x= , y= , z=
2 2 2
Using these in the above expression for m we get
1 2 1
x= [ m + y 2 − y] + [ m 2 + z 2 − z]
2 2
1 2 1
y = [ n + z 2 − z] + [ n 2 + x 2 − x]
2 2
1 2 1
z = [ p + x 2 − x] + [ p 2 + y 2 − y]
2 2
Since m, n, p > 0, x, y, z > 0. Further x < m, y < n and z < p. Thus the map
F : K → K where K = [0, m] × [0, n] × [0, p] defined√by F(x, y, z) = ( f m (y) +
f m (z), f n (z) + f n (z), f p (x) + f p (y)) where f α (t) = 21 [ α 2 + t 2 − t] on [0, α] for
α > 0 has a fixed point if and only if the system of Eq. (8.1.1) has a solution. a, b, c
in terms of m, n and p.
Remark 8.1.11 Mironescu and Panaitopol invoked Brouwer’s fixed point theorem
9.1 - to conclude that F has a fixed point which is the solution to the three internal
angle bisectors problem.
On the other hand Dinca and Mawhin [4] deduced it from the contraction principle
though it can also be obtained from Edelstein’s theorem (Corollary 8.1.4).
Theorem 8.1.12 (Dinca and Mawhin [4]) The map F : K → K defined by F(x, y,
z) = ( f m (y) + f m (z),√f n (z) + f n (x), f p (x) + f p (y)) where K = [0, m] × [0, n] ×
[0, p] and f α (t) = 21 [ α 2 + t 2 − t] is contractive and has a unique fixed point and
every sequence of F-iterates converges to the unique solution of the three internal
angle bisectors problem.
√
Proof f α (t) = 21 [ α 2 + t 2 − t] is continuous and f α (t) = 21 √α2t +t 2 − 1 is neg-
ative. Further, | f α (t)| < 21 . Now K is compact in R3 with the norm (x, y, z) =
max{|x|, |y|, |z|}. Further, | f α (t1 ) − f α (t2 )| < 21 |t1 − t2 | for t1 = t2 . So for (x, y, z)
= (x1 , y1 , z 1 ) F(x, y, z) − F(x1 , y1 , z 2 ) < max{|x − x1 |, |y − y1 |, |z − z 1 |}.
Further as f α (t) < α2 and F maps K into itself. Thus F is a contractive self-map on
the compact space K . So by Edelstein’s theorem (Corollary 8.1.4) F has a unique
fixed point to which every sequence of F iterates converges. In other words there
is a unique triangle up to isometry having with prescribed lengths of internal angle
bisectors.
Remark 8.1.13 Dinca and Mawhin [4] proved Theorem 8.1.12 applying the contrac-
tion principle to a sequence of contractions converging pointwise to F.
8.2 Non-expansive Maps
In this section, some elementary results on the fixed points of non-expansive maps
are discussed.
Definition 8.2.1 A map T : X → X , where (X, d) is a metric space is called

non-expansive if d(T x, T y) ≤ d(x, y) for all x, y ∈ X . T is called an isometry if
d(T x, T y) = d(x, y) for all x, y ∈ X .
Remark 8.2.2 A non-expansive mapping on a complete metric space may not have
a fixed point as is evident by considering the map x → x + a, a = 0 on R. Indeed
themap eit → ei(t+α) t ∈ [0, 2π ) where α ∈ (0, 2π ) in the unit circle in the complex
plane is an isometry without a fixed point on the compact connected locally connected
space S 1 .
For x = (xn ) in the closed unit sphere of c0 , the Banach space of all null sequences
x → T x = (1, x1 , x2 , . . . ), x = (xn ) is a fixed point free isometry.
Dotson Jr [5] proved a fixed point theorem for non-expansive maps in the setting
of star-shaped subsets of normed linear spaces.
Definition 8.2.3 A non-empty subset S of a linear space X is called star-shaped

if there exists an element a ∈ S such that ta + (1 − t)s ∈ S for all s ∈ S for all
t ∈ [0, 1]. In this case a is called a star-centre of S.
Remark 8.2.4 While all convex sets are star-shaped about every one of its points a
star-shaped subset may not be convex as in evident by considering S = {0} × [0, 1] ∪
[0, 1] × {0} in R2 .
Theorem 8.2.5 (Dotson [5]) Let S be a compact star-shaped subset of a normed

linear space and T : S → S, a non-expansive map. Then T has a fixed point in S.
Proof Without loss of generality we can assume that S is star-shaped about 0 (i.e. 0 is
a star-centre of S). Then the map Tn defined by Tn (x) = (1 − n1 )T x (= n1 .0 + (1 −
1
n
)T x) maps S into itself for each n ∈ N. As Tn is a contraction on S and S being
compact is complete, Tn has a unique fixed point xn , say. So Tn (xn ) = (1 − n1 )T xn =
xn for each n ∈ N. As S is compact and xn ∈ S for each n ∈ N, there is a subsequence
{xn k } of {xn } converging to x ∗ ∈ S. Since T is continuous, xn k (= (1 − n1k )T xn k )

converges to x ∗ , T xn k = (1 − n1k )−1 xn k converges to T x ∗ = x ∗ . Thus x ∗ is a fixed
point of T .
Corollary 8.2.6 Every non-expansive self-map on a compact convex subset of a

normed linear space has a fixed point.
In this context an application to approximation theory is described, based on a

few concepts and results.
Definition 8.2.7 Let (X, d) be a metric and S, a non-empty subset and x0 ∈ X − S.

If there exists an element s0 ∈ S such that d(x0 , s0 ) = inf d(x0 , s), then s0 is called
s∈S
a best approximation to x0 in S.
While every element of X − S of a metric space X need not have a best approxima-
tion in S ⊆ X , the following proposition gives sufficient conditions for the existence
of a best approximation.
Theorem 8.2.8 Let V be a finite-dimensional subspace of a normed linear space

(X, · ) and f ∈ X − V . Then there exists v0 ∈ V such that f − v0 = inf{ f −
v : v ∈ V }.
Proof Since V is a finite-dimensional subspace of X , V is complete and hence

closed in X . So inf{ f − v : v ∈ V } = d( f, V ) > 0. If v ∈ V and v > f then
d( f, V ) ≤ f − 0 = f < v . So if f has a best approximation, in V , then the
best approximation lies in B(0; f ) ∩ V , the closed sphere centred in 0 of radius
f lying in V . Now B = B(0; f ) ∩ V = {v ∈ V : v ≤ f } is the closed
sphere centred at 0 and of radius f in the finite-dimensional space V . So B
is compact and so inf{ f − v : v ∈ B} is attained at some v0 . Thus f − v0 =
inf B { f − v } = inf v∈V f − v . So f has a best approximation in V .
Theorem 8.2.9 (Invariants of best approximations) Let X be a normed linear space,

V a finite-dimensional subspace of X and T : X → X a map with a fixed point f such
that for all x, y ∈ V , x − y ≤ d( f, V ) implies T x − T y ≤ x − y , d( f, V )
being inf{d( f, v) : v ∈ V }. If T maps V into itself then f has a best approximation
in V which is a fixed point of T .
Proof Since V is finite-dimensional S f the set of best approximations of f in V is

non-empty by Theorem 8.2.8. As T (V ) ⊂ V and S f ⊆ V , T (S f ) ⊆ V . Since T f =
f , T g − f = T g − T f ≤ g − f for all g ∈ S f . So T maps S f into itself. For
g1 , g2 ∈ S f , f − tg1 + (1 − t)g2 = t f − tg1 + (1 − t) f − (1 − t)g2 ≤ t
f − g1 + (1 − t) f − g2 ≤ d( f, V ). Since S f is a closed subset of B(0, f ) ∩
V and V is finite dimensional, S f is a compact convex subset of V . As T is a non-
expansive self-map on S f , T has a fixed point f 0 in S f . Thus f 0 is a best approximation
of f which is a fixed point of T .
8.2 Non-expansive Maps 173
Corollary 8.2.10 If T : X → X be a non-expansive map with a fixed point f and

leaving a finite dimensional subspace V of E invariant, then f has a best approxi-
mation in V .
Corollary 8.2.11 (Meinardus [14]) Let T : B → B be a continuous map where B

is a compact metric space and C[B] the space of all continuous real (or complex)
functions on B with the supremum norm. Let A : C[B] → C[B] be a non-expansive
map and suppose that
(i) A( f (T (x))) = f (x);
(ii) A(h(T (x))) ∈ V , whenever h(x) ∈ V where V is a finite dimensional subspace
of C[B]. Then there is a best approximation g of f with respect to V such that
A(g(T (x))) = g(x).
This corollary follows from Corollary 8.2.10 upon setting T1 : C[B] → C[B] by
T1 (g(x)) = A(g(T (x))).
Corollary 8.2.12 Let f be an even (odd) function in C[−1, 1] with the supremum
norm. If V is a finite dimensional subspace of C[−1, 1] such that whenever h(x) ∈ V ,
h(−x) also is in V , then f has an even (odd) function as best approximation in V .
8.3 Browder–Gohde–Kirk Fixed Point Theorem
While non-expansive self-maps on bounded closed convex sets in infinite-

dimensional Banach spaces may not have fixed points, fixed points for non-expansive
mappings can be ensured in Banach spaces with nice geometrical features. Browder
[2], Gohde [8] and Kirk [11] independently proved that non-expansive self-maps on
non-empty bounded closed convex subsets of a uniformly convex Banach space have
fixed points. Indeed Kirk [11] using the concept of normal structure proved a more
general fixed point theorem. The study of fixed points of non-expansive mappings
on subsets of Banach spaces is an active area of research employing sophisticated
analytic and geometric concepts. Goebel [7] gave a simpler proof of the Browder-
Gohde-Kirk fixed point theorem in the setting of uniformly convex Banach spaces
and this proof is presented below. The definition of uniform convexity (Definition
1.3.40) is repeated below for the sake of both completeness and convenience.
Definition 8.3.1 A Banach space is uniformly convex if there exists an increasing

positive function δ : I2 = (0, 2] → I1 = (0, 1] such that x , y ≤ r and x −
y ≥ r imply that x+y 2
≤ (1 − δ())r .
Remark 8.3.2 If η is the inverse function of δ, then lim η(y) = 0.

y→0
Theorem 8.3.3 (Browder-Gohde-Kirk) Let K be a non-empty bounded closed con-

vex subset of a uniformly convex Banach space X and T : K → K a non-expansive
map. Then T has a fixed point in K .
Goebel’s proof makes use of the following
Lemma 8.3.4 If u, v, w are elements of a uniformly Banach space B such that

u − w ≤ R, v − w ≤ R and w − u+v 2
≥ r > 0, then u − v ≤ Rη R−r .
R

Proof w − u+v ≥ r means that (w−u)+(w−v) ≥ r = 1 − R−r R and R ≥ r as
2
2 R
R ≥ w−u + w−v ≥ r . So in the contrapositive version of the definition of uniform
2 2
convexity setting = η R−r R
we get u − v ≤ Rη R−r R
.
Proof of Theorem 8.3.3 [7]

Without loss of generality we can assume that 0 ∈ K . Define F = (1 − )F for
∈ I1 = (0, 1]. Since F is a contraction mapping K into itself for each ∈ I1 ,
F has a unique fixed point x in K for each ∈ I1 . Writing d(K ) = diameter
of K , we note that x − F x = F x − F x = F x ≤ d(K ). As → 0,
x − F x → 0. So inf x − F x = 0. Define C = {x ∈ K : x − F x ≤ } and
K
D = {x ∈ C : x ≤ a + } where a = lim a(C ) and a(B) = inf x .
→0 x∈B
We proceed to show that ∩C is non-empty. Otherwise a > 0, since each C is
closed. Let u 1 , u 1 ∈ C . For i = 1, 2,

u i − F u 1 + u 2 ≤ u i − F(u i ) + F(u i ) − F u 1 + u 2
2 2
1
≤ + u1 − u2 (8.3.1)
2
and

u i − u 1 + u 2 = u i − u 2 < + 1 u 1 − u 2 (8.3.2)
2 2 2
As

1 u1 + u2 u1 + u2
u1 − u2 ≤ + u 2 − 1 u 1 + u 2 + F u 1 + u 2 ,
u 1 − 2 2
+F
2 2 2 2
the inequality

1
u i − 1 u 1 + u 2 + F u 1 + u 2 ≥
2 2 2 2 u1 − u2 (8.3.3)
is true for at least one of i = 1, 2. From Lemma 8.3.4 and the inequalities (8.3.1),
(8.3.2) and (8.3.3)
8.3 Browder–Gohde–Kirk Fixed Point Theorem 175

u1 + u2 u1 + u2
− F ≤ + 1 u1 − u2 η

2 2 2 + u 1 −u
2
2

≤ sup (ξ + )η
0<ξ < d(K ) +ξ
2
⎡ ⎤

⎣ ⎦
≤ max sup
√
( + ξ )η , sup ( + ξ )η
0<ξ ≤ − +ξ √
−<ξ ≤ d(K ) +ξ
2

√ d(K ) √
≤ max 2 , + η( ) = φ() (say).
2
So u 1 , u 2 ∈ C implies u 1 +u
2
2
∈ Cφ() . Clearly φ() → 0 as → 0.
For u 1 , u 2 ∈ D , vi ≤ a + for i = 1, 2 and as u 1 +u 2
2
∈ Cφ() , u 1 +u
2
2
≥
a(Cφ() )
Now by Lemma 8.3.4

a + − a(Cφ() )
d(D ) = sup u 1 − u 2 ≤ (a + )η
u 1 ,u 2 ∈D a+
and lim d(D ) = 0. So by Cantor intersection theorem ∩D = φ. So ∩C = φ.

→0
Hence F has a fixed point in K .
As uniformly convex Banach spaces are reflexive and bounded closed convex
subsets of a uniformly convex Banach space are weakly compact, it is natural to
enquire if a weakly compact convex subset of a non-reflexive Banach space has fixed
point property for non-expansive mappings. Alspach [1] has given a counter-example
to this conjecture and it is summarized below.
Example 8.3.5 (Alspach [1]) Let X be the Banach space L 1 [0, 1] and K = { f ∈ X :
1
0 ≤ f ≤ 2, are and 0 f dμ = 1}. Clearly K is convex and weakly closed in X . As
order intervals in X are weakly compact in view of the uniform integrability of its
elements, K is weakly compact. Define T : K → K by

2 f (2t) ∨ 2, 0 ≤ t ≤ 21
T f (t) =
[2 f (2t − 1) − 2] ∨ 0, 21 < t ≤ 1
It can be seen that T is an isometry. If T has a fixed point g, then g = 2χ A for

some subset A of [0, 1] with measure 12 . Now {t : g(t) = 2} = {t : T g(t) = 2} = { 2t :
g(t) = 2} ∪ { 1+t
2
: g(t) = 2} ∪ { 2t : 1 ≤ g(t) < 2}. Here the union of these sets is a
disjoint union. Since μ{t : g(t) = 2} + μ{ 1+t2
: g(t) = 2} = μ{t : g(t) = 2}, μ{t :
1 ≤ g(t) < 2} = 0. Repeated use of this argument shows that
∞
1 1
{t : 0 < g(t) < 2} = t : n ≤ g(t) < n−1
n=0
2 2
is of measure zero, as well.

For g = 2χ A

n
i t
{t : T g(t) = 2} =
n
i
+ n :t ∈ A
i ∈{0,1} i=1
2 2
The above representation can be proved using induction. Since T g = g, A = {t :

T n g(t) = 2} for all natural numbers n and the intersection of A with any interval
with dyadic end points has measure precisely half the measure of the interval. Since
such a measurable set does not exist, T cannot have a fixed point in K .
For other examples Sims [23] may be consulted.
8.4 A Generalization to Metric Spaces
Pasicki [18] obtained an interesting generalization of Browder-Gohde-Kirk Theorem

8.3.3 to special metric spaces. In this section Pasicki’s contributions are highlighted.
Definition 8.4.1 ([18]) Let (X, d) be a metric space and A a non-empty bounded
subset. x ∈ X is called central point for A if
r (a) := inf{t ∈ (0, ∞) : A ⊆ B(z, t) for some z ∈ X }

= inf{t ∈ (0, ∞) : A ⊆ B(x, t)}
The centre c(A) for A is the set of all central points for A and r (A) is the radius of
A.
Proposition 8.4.2 ([18]) For a metric space (X, d) satisfying for each r > 0
and x, y ∈ X with x = y there exists δ > 0, z ∈ X such that B(x, r ) ∩ B(y, r ) ⊆
B(z, r − δ) given a bounded non-void subset A of X , c(A) contains at most one
point.
Proof If x, y ∈ c(A) and x = y, then for r = r (A), A ⊆ B(x, r ) ∩ B(y, r ) ⊆ B
(z, r − δ). This implies r (A) = r ≤ r − δ, a contradiction.
A centre of a set corresponds to Chebyshev centre of a set.
For developing fixed point theory Pasicki [18] introduced the following definition.
Definition 8.4.3 ([18]) Let Y be a non-void bounded subset of a metric space (X, d)
and F : Y → 2Y , a mapping with F(y) = φ for each y ∈ Y . x ∈ X is called a central
point for F if
r (F) := inf{t ∈ (0, ∞) : F n (y) ⊆ B(z, t) for a

z ∈ X and an n ∈ N }
= inf{t ∈ (0, ∞) : F n (Y ) ⊆ B(x, t) for an n ∈ N }
The centre c(F) for F is the set of central points for F and r (F) is the radius of F.
8.4 A Generalization to Metric Spaces 177
It may be noted that if F n (y) ⊆ B(z, t) then F p (Y ) ⊆ F n (Y ) ⊆ B(z, t) for all

p ≥ n.
Theorem 8.4.4 ([18]) Let Y be a non-void bounded subset of a metric space (X, d)
and f : X → X a map such that f (Y ) ⊆ Y and c( f |Y ) = {x}. If d( f (x), f (y)) ≤
d(x, y) for all x, y ∈ Y , then x is a fixed point for f .
Proof For f n−1 (Y ) ⊆ B(x, t), f n (Y ∩ B(x, t)) ⊆ B( f (x), t). As f is non-
expansive f (Y, B(x, t)) ⊆ B( f (x), t) and so f n (Y ) ⊆ B( f (x), t) implying that
f (x) ∈ c( f |Y ). As c( f |Y ) is a singleton, f (x) = x.
The above theorem leads to the formulation of a bead space and a discus space.
Definition 8.4.5 ([18]) A metric space is called a bead space if the following con-
dition holds:
for every r , β > 0, there exists a δ > 0 such that for each pair x, y ∈ X with
d(x, y) ≥ β, there exists a z ∈ X such that B(x, r + δ) ∩ B(y, r + δ) ⊆ B(z, r − δ).
Definition 8.4.6 ([17, 18]) A metric space is called a discus space if there exists a
map ρ : [0, ∞) × (0, ∞) → [0, ∞) such that
ρ(β, r ) < ρ(0, r ) = r, β, r > 0, (8.4.1)
ρ(·, r ) is non-increasing, r > 0, (8.4.2)
ρ(δ, ·) is upper semicontinuous for δ ≥ 0, (8.4.3)
for each pair x, y ∈ X, r, > 0 there exists

z ∈ X such that B(x, r ) ∩ B(y, r ) ⊆ B(z, ρ(d(x, y), r ) + ). (8.4.4)
Proposition 8.4.7 ([18]) Each discus space is a bead space.
Proof Let r > 0, x, y ∈ X and d(x, y) = β > 0 be arbitrary. In view of 8.4.6

(iv) for each , k > 0, there exists z ∈ X such that B(x, r + k) ∩ B(y, r + k) ⊆
B(z, ρ(β, r + k) + ). Writing 2η = r − ρ(β, r ) = ρ(0, r ) − ρ(β, r ) > 0 (by (i)).
For sufficiently small k, , we have ρ(β, r + k) + ≤ ρ(β, r ) + η (by (iii)). So
one gets ρ(β, r + k) + ≤ ρ(β, r ) + η = r − 2η + η = r − η and B(x, r + k) ∩
B(y, r + k) ⊆ B(z, r − η). Then for δ = min{k, η}, B(x, r + δ) ∩ B(η, r + δ) ⊆
B(z, r − δ). If d(x, y) ≥ β, then by (ii) ρ(d(x, y), r + k) ≤ ρ(β, r + k) and the
inclusion follows.
In fact a bead space is also a discus space as noted by Pasicki [19].

Proposition 8.4.8 ([19]) Each bead space is a discus space.

Proof Let (X, d) be a bead space. For β, r > 0 let η(0, r ) = 0 and η(β, r ) = sup{δ ∈
(0, r ) : for each x, y ∈ X with d(x, y) ≥ β there exists z ∈ X with B(x, r + δ)
∩ B(y, r + δ) ⊆ B(z, r − δ)}.
Consider ρ(β, r ) = r − η(β, r ). Clearly ρ maps [0, ∞) × (0, ∞) into [0, ∞).
For β, r > 0, η(β, r ) > 0 and ρ(β, r ) < ρ(0, r ) = r . We now show that η(·, r )
is non-decreasing for r > 0. If for r, β > 0, δ > 0 is such that for d(x, y) ≥ β
B(x, r + δ) ∩ B(y, r + δ) ⊆ B(z, r − δ), then it works for β1 > β so that η(β, r ) ≤
η(β1 , r ). Thus η(·, r ) is non-decreasing and ρ(·, r ) is non-increasing. We now
show that η(β, ·) is lower semicontinuous, β ≥ 0. Now η(0, ·) = 0 and consider
β > 0. Suppose η(β, x0 ) > α > 0. We show that η(β, (r0 − , r0 + )) ⊆ (α, ∞)
for any > 0. Since η(β, r0 ) > r , there exists δ > α such that B(x, r0 + δ) ∩
B(y, r0 + δ) ⊂ B(z, r0 − δ) for a z ∈ X . Let > 0 be such that r0 + δ − 2 > 0
and δ > . Then B(x, r0 − + (δ − )) ∩ B(y, r0 − + (δ − )) ⊂ B(x, r + δ) ∩
B(y, r0 + δ) ⊂ B(z, r0 − δ) = B(z, r0 − − (δ − )). So, for δ > 2 and δ − >
α, η(β, (r − , r0 ]) ⊆ (α, ∞). However for δ > + α, we have
B(x, r0 + + (δ − )) ∩ B(y, x + + (δ − ))

⊆ B(x, r0 + δ) ∩ B(y, r0 + δ) ⊆ B(z, r0 − δ)
= B(z, r0 + − (δ + )) ⊆ B(z, r0 + − (δ − ))
So η(β, [r0 , r0 + )) ⊆ (α, ∞). Finally one has η(β, (r0 − , r0 + )) ⊆ (α, ∞) and
η(β, ·) is lower semicontinuous. Also B(x, r ) ∩ B(y, r ) ⊆ B(x, r + δ) ∩ B(y, r +
δ) ⊂ B(z, r − δ) ⊆ B(z, r − δ + ) and so B(x, r ) ∩ B(y, r ) ⊆ B(z, ρ(d(x, y), r )
+ ). Thus (X, d) is a discus space.
Remark 8.4.9 If (X, d) is a metric space and r > 2 > 0, then B(x, r − 2) ⊆
B(x, r − ) ⊂ B(x, r )
So the definitions of bead and discus spaces can be formulated using closed balls.
We need the following lemma to decode the geometry of normed bead spaces.
Lemma 8.4.10 ([19]) In a normed linear space (X, ) the following conditions
are equivalent:
(i) for every r, β > 0, there exists δ > 0 such that for x, y ∈ X with x − y ≥ β,
there exists z ∈ X with B(x, r + δ) ∩ B(y, r + δ) ⊆ B(z, r − δ);
(ii) for every r, β > 0, there exists δ > 0 such that for x, y ∈ X with x − y ≥ β,
there exists z ∈ X such that B(x, r + δ) ∩ B(y, r + δ) ⊆ B(z, r );
(iii) for every r, β > 0, there exists δ > 0 such that for x ∈ X and 2 x ≥ β imply
B(−x, r ) ∩ B(x, r ) ⊆ B(0, r − δ).
Proof It is enough to prove the equivalence for y = −x. The set C = B(−x, r ) ∩
B(x, r ) is symmetric and therefore C ⊆ B(z, t) implies C ⊆ B(−z, t) ∩ B(z, t) ⊆
B(0, t). Thus we may let z = x+y 2
= 0 in (i), (ii) and (iii). Clearly (i) implies (ii)
and (iii). Choosing r = 1 in (iii) we get for β > 0 there exists δ > 0 such that for
u, x ∈ X with u + x , u − x < 1 and 2 x ≥ β > 0
8.4 A Generalization to Metric Spaces 179
(iv) u < 1 − δ
Suppose u + x , u − x < r + and 2 x ≥ (r + )β hold. Then u+x r +
,
u−x
r+
< 1 and 2 x
≥ β and in view of (iv) u
< 1 − δ or u < (r + )(1 − δ) <
r + r +
r 1 − 2δ for small values of . For y = −x, δ1 = min , r2δ , and 2 x ≥ β1 =
3rβ
2
> (r + )β, we get B(x, r + δ1 ) ∩ B(y, r + δ1 ) ⊆ B(0, r − δ1 ) which is (i).
Thus (i), (iii) and (iv) are equivalent. Now consider (ii). Then (ru+x , u−x < 1 + δ
−) (r −)
and (r2 −)
x
≥ β imply r −
u
< 1. (see (ii)). This relation can be written as u + x , u −
x < (r − )(1 + δ) = r + r δ − (1 + δ) and 2 x ≥ rβ > (r − )β implying
rδ
u < r − . Thus for < r δ − (1 + δ) or < 2+δ , u+x , u−x <r +
and 2 x ≥ rβ implying u < r − , verifying (i).
Theorem 8.4.11 ([19]) A normed linear space is uniformly convex if and only if
for each β > 0 there exists δ > 0 such that for u, x ∈ X , u + x , u − x < 1
and 2 x > β > 0 imply
(v) u < 1 − δ.
Proof From (iv) of Lemma 8.4.10 setting u = y+z 2
, x = y−z
2
one gets
for every β > 0, there exists δ > 0 such that y, z ∈ X , y , z < 1 and y −
z > β > 0 imply y+z 2
< 1 − δ. So X is uniformly convex.
If (v) is satisfied then writing y = u + x, z = u − x, then (iv) of Definition 8.3.1
is satisfied leading to uniformly convexity of X .
We now have
Theorem 8.4.12 ([19]) A normed linear (X, · ) is uniformly convex if and only
if any one of the conditions (i), (ii), (iii), (iv) of Lemma 8.4.10 and (v) of Theorem
8.4.11 are satisfied.
From Propositions 8.4.7, 8.4.8 and Theorem 8.4.11 we have
Theorem 8.4.13 ([19]) For any normed linear space the following conditions are
equivalent:
(i) X is a bead space;
(ii) X is a discus space;
(iii) X is a uniformly convex space.
Remark 8.4.14 ([19]) Each convex subset X of a uniformly convex normed linear
space satisfies the definition of a bead space with z = x+y
2
(see Definition 8.4.5).
We now proceed to provide Pasicki’s generalization of Browder-Gohde-Kirk The-
orem 8.3.3 based on the following lemmata.
Lemma 8.4.15 If (X, d) is a complete discus space, then (iv) of Definition 8.4.6 (of
Discus space) can be replaced by
for each x, y ∈ X and r > 0 there is a z ∈ X such that B(x, r ) ∩ B(y, r ) ⊆
B(z, ρ(d(x, y), r )).
Proof Let α = ρ(d(x, y), r ) and (αn ) ↓ α be such that there exist xn ∈ B(x, r ) ∩
B(y, r ) ⊆ B(xn , αn ). Suppose (xn ) is not Cauchy. Then there is β > 0 with d(xn , xk )
⊂ β for infinitely many k, n with k < n. Set 2η = α − ρ(β, α) = ρ(0, α) − ρ(β, α)
> 0 (by (i) of Definition 8.4.6). So B(x, r ) ∩ B(y, r ) ⊆ B(xn , αn ) ∩ B(xk , αk ) ⊂
B(xn , αk ) ∩ B(xk , αk ) ⊂ B(z n,k , ρ(d(xn , xk ), αk ) + η) for some z n,k ∈ X (by (iv)
of Definition 8.4.6). On the other hand, ρ(d(xn , xk ), αk ) ≤ ρ(β, αk ) (by ii) and
ρ(β, αk ) ≤ ρ(β, α) + η for sufficiently large k (by (iii) of Definition 8.4.6). Now one
gets B(x, r ) ∩ B(y, r ) ⊂ B(z n,k , ρ(β, α) + η) = B(z n,k , α − 2η + η) = B(z n,k ,
α − η) ⊂ B(z n,k , α). Thus the lemma is true. If (xn ) is a Cauchy sequence con-
vergent to z ∈ X , then B(xn , αn ) ⊂ B(z, α + β) for any β > 0 and all large n.
So B(x, r ) ∩ B(y, r ) ⊆ B(z, α + β) for all β > 0 and B(x, r ) ∩ B(y, r ) ⊆ B(z, α).
Since B(x, r ) ∩ B(y, r ) is open, the lemma follows.
Lemma 8.4.16 Let (X, d) be a complete discus space and let A ⊆ X be non-void
and bounded then c(A) is a singleton.
Proof Let (rn ) ↓ r = r (A) while A ⊆ B(xn , rn ). If (xn ) is not Cauchy, then d(xn , xk )
≥ β > 0 for infinitely many k, n with k < n. We have
A ⊆ B(xn , rn ) ∩ B(xk , rk ) ⊂ B(xn , rk ) ∩ B(xk , rk )

⊆ B(z n,k , ρ(d(xn , xk ), rk )) ⊆ B(z n,k , ρ(βk rk ))
(by definition of ρ).
So A ⊆ B(z n,k , r (A) − η (η being 21 (α − ρ(β, α))) as proceed in the course of the
preceding Lemma 8.4.15. This is a contradiction. Let (xn ) converge to x. Then for
any β > 0 B(xn , rn ) ⊆ B(x, r + β) for sufficiently large n implying A ⊆ B(x, r +
β) for all β > 0 and so x ∈ c(A). If x, y ∈ c(A) and d(x, y) ≥ β > 0, then by
Lemma 8.4.15 A ⊆ B(x, r ) ∩ B(y, r ) ⊆ B(z, ρ(β, r )) ⊆ B(z, r − η) for a η > 0,
a contradiction. So c(A) is a singleton.
Lemma 8.4.17 ([19]) Let (X, d) be a complete discus space. If Y = φ ⊆ X is

bounded and F : Y → 2Y is a mapping, then c(F) is a singleton.
Proof Let r = r (F) (refer Definition 8.4.3 for r (F) and c(F)). F n+1 (Y ) ⊆ F n (Y )
and so there exists rn ↓ r and a sequence (xn ) such that F n (y) ⊆ B(xn , rn ) for all n. If
(xn ) is not a Cauchy sequence, for infinitely many n, k with k < n d(xn , xk ) ≥ β > 0.
We have F n (Y ) ⊆ F n (Y ) ∩ F k (Y ) ⊂ B(xn , rn ) ∩ B(xk , rk ) ⊆ B(z n,k , ρ(β, rk )) and
so F n (Y ) ⊆ B(z n,k , r − η) for a η > 0 (as in the preceding proof), a contradiction.
Let (xn ) converge to x. We get F n (Y ) ⊆ B(x, r + β) for any β > 0 and sufficiently
large n. So x ∈ c(F). That c(F) is a singleton follows from Lemma 8.4.16.
Theorem 8.4.4 can be modified as
Theorem 8.4.18 Let (X, d) be a complete bead (discus) metric space and f : X →
X a non-expansive map. If Y is a bounded non-empty subset of X with f (Y ) ⊆ Y ,
then c( f |Y ) is a singleton which is a fixed point of f .
8.5 An Application to a Functional Equation 181
8.5 An Application to a Functional Equation
In this section, Matkowski’s deduction [13] of a fixed point theorem from Pasicki’s
Theorem 8.4.4 is discussed as well as its application to a functional equation. It is in
the setting of a paranormed space.
Definition 8.5.1 Let X be a real linear space over R. A function p : X → R is called
a paranorm if the following conditions are satisfied:
(i) p(x) = 0 if and only if x = 0, p(x) = p(−x) for all x;
(ii) p(x + y) ≤ p(x) + p(y) for all x, y ∈ X ;
(iii) for tn , t ∈ R and xn , x ∈ X with tn → t and p(xn − x) → 0 as n → ∞,
p(tn xn − t x) converges to zero as n → ∞.
Remark 8.5.2 If (X, p) is a paranormed linear space, then d(x, y) = p(x − y)
defines a metric on X .
Definition 8.5.3 A paranormed space (X, p) is called uniformly convex if for
each r > 0 and ∈ (0, 2r ) there exists δ(r, ) ∈ (0, r ) such that for all x, y ∈ X
with p(x), p(y) ≤ r and p(x, y) ≥ , p( x+y2
) ≤ r − δ(r, ) (the function δ : →
(0, ∞) where = {(r, ) : r > 0, 0 < < 2r } is called the modulus of convexity
of (X, p).
The following theorem due to Matkowski [13] provides a class of examples of
paranormed spaces.
Theorem 8.5.4 Let ( , S , μ) be a measure space and S = S( , S , μ) the real
linear space of all μ-integrable simple functions x : → R. Let ϕ : [0, ∞) →
be an increasing bijection with ϕ(0) = 0. Then the functional pϕ (x) =
[0, ∞)
ϕ −1 ϕ(|x|)dμ is well-defined for each x ∈ S. If μ( ) = 1 and if there exists a set
A ∈ S with 0 < μ(A) < 1, then pϕ is a paranorm if and only if F : R+ × R+ → R+
defined by F(r, s) = ϕ(ϕ −1 (r ) + ϕ −1 (s)) r, z ∈ R+ is concave.
If μ( ) ≤ 1 and F is concave, then pϕ is a paranorm on S.
Remark 8.5.5 Let ϕ : R+ → R+ be twice differentiable with ϕ(0) = 0, ϕ (r ), ϕ
(r )
ϕ ϕ (r +s) ϕ (r )
> 0 for r > 0. If ϕ is super additive in (0, ∞) in the sense that ϕ (r +s) ≥ ϕ (r ) +
ϕ (s)
r, s > 0 then F : R+ × R+ → R+ defined by F(r, s) = ϕ(ϕ −1 (r ) + ϕ −1 (s))
ϕ (s)
,
is concave.
For the proofs of the following lemmata Matkowski [13] may be consulted.
Lemma 8.5.6 ([13]) Let ( , S , μ) be a measure space with μ( ) ≤ 1. Let ϕ :
R+ → R+ be an increasing bijection with ϕ(0) = 0 and F : R+ × R+ → R+ be
defined by F(r, s) = ϕ(ϕ −1 (r ) + ϕ −1 (s)), r, s ≥ 0. If F is concave, then pϕ defined
(in Theorem 8.5.4) is a paranorm on S = S( , S , μ). If further ϕ(r + s) + ϕ(|r −
s|) ≥ 2[ϕ(r ) + ϕ(s)] for all r, s ≥ 0 (i.e. ϕ is super quadratic), then for x, y ∈ S
ϕ( pϕ (x + y)) + ϕ( pϕ (x − y)) ≥ 2[ϕ( pϕ (x)) + ϕ( pϕ (y))].

Lemma 8.5.7 ([13]) Let ( , S , μ) be a measure space and ϕ : R+ → R+ an

increasing bijection such that pϕ is a paranorm on S. If ϕ is super-quadratic
(i.e. ϕ(r + s) + ϕ(|r − s|) ≥ 2[ϕ(r ) + ϕ(s)] for r, s ≥ 0), S = S( , S , μ) with
the paranorm is uniformly convex, the modulus of convexity being δ(r, ) = r −
ϕ −1 (ϕ(r ) − ϕ( 2 )), r > 0 and ∈ (0, 2r ).
These lead immediately to
Theorem 8.5.8 ([13]) Let ( , S , μ) be a measure space and μ( ) ≤ 1. If ϕ :

R+ → R+ is an increasing bijection and the map F : R+ × R+ → R+ defined by
F(r, s) = ϕ(ϕ −1 (r ) + ϕ −1 (s)), r, s ≥ 0 is concave then pϕ is a paranorm on S =
S( , S , μ). If ϕ is also superquadratic then S is uniformly convex with the modulus
of continuity being δ(r, ) = r − ϕ 1− (ϕ(r ) − ϕ( 2 ), r > 0 and 0 < < 2r .
Lemma 8.5.9 ([20]) Let X be a set in a linear space such that X +X 2

⊆ X and p :
X − X → R be a map such that d(x, y) = p(x − y), x, y ∈ X defines a metric on
X . If for r , β > 0 there exists δ > 0 such that s, t, s − t ∈ X − X the inequalities
p(s), p(t) < r + δ and p(s − t) > β imply p( s+t 2
) < r − δ, then (X, d) is a bead
space.
Proof Set s = u − x, t = v − y where u, v, x, y ∈ X . Then s+t

2
= u+v
2
− (x+y)
2
∈
X − X as X + X ⊆ 2X . So p( 2 ) is well-defined. Let u, x, y ∈ X be such that
s+t
d(u, x), d(u, y) < r + δ and d(x, y) > β. Then for s = u − x, t = u − y, p(s) <
r + δ, p(t) < r + δ and for s − t = y − x ∈ X − X ⊆ X , p(s − t) = p((u − x) −
(u − y)) = p(y − x) > β. So by assumption of Lemma 8.5.9, p(u − x+y 2
)=
u−x+u−y
p( 2
) = p( 2 ) < r − δ. Thus u ∈ B(z, r − δ) where z = 2 ∈ X . Thus
s+t x+y
(X, d) is a bead space.
Remark 8.5.10 The proof of the above lemma is due to Pasicki [20]. In the above
lemma X can be the whole space.
Remark 8.5.11 From Pasicki’s Theorem 8.4.4 it follows that a non-expansive map-
ping of a non-void bounded closed convex subset C of a complete uniformly convex
paranormed space X into itself has a fixed point, the modulus of convexity of X
being continuous.
Remark 8.5.12 ([13]) The completion of a uniformly convex paranormed linear

space is uniformly convex.
Matkowski deduced an existence theorem for the solution of a functional equation

using Pasicki’s theorem.
Theorem 8.5.13 (Matkowski [13]) Let = [0, 1], S the σ -algebra of Lebesgue
measurable sets and μ the Lebesgue measure. Let ϕ : R+ → R+ be an increasing,
convex, superquadratic function such that F(r, s) = ϕ(ϕ −1 (r ) + ϕ −1 (s)) is concave,
r, s > 0. Let f : I → I be an increasing differentiable function, h : I × R → R
satisfy the Caratheodary conditions:
8.5 An Application to a Functional Equation 183
(a) h(t, .) is continuous for each x;

(b) h(., x) is measurable for almost all x;
(c) |h(t, x)| ≤ m(t) for all (t, x) where m is Lebesgue-integrable on [0, 1].
(d) there exist α, β ∈ Sφ ( , S , μ) with α(t) ≤ h(t, α(t)) ≤ h(t, β(t)) ≤ β(t) for
all t ∈ [0, 1] and for all t ∈ [0, 1] and x ∈ [α(t), β(t)]
h(t, α(t)) ≤ h(t, x) ≤ h(t, β(t))
(e) there exists a Lebesgue-measurable function g : I → R+ such that
|h(t, x) − h(t, y)| ≤ g(t)|x − y|, t ∈ I, x, y ∈ [α(t), β(t)].
If fg ≤ 1, then the functional equation x(t) = h(t, x( f (t))) has a solution in
ϕ
S ( , S , μ), the completion of the paranormed space S = Sϕ ( , S , μ).
Proof Define C = {x ∈ R I : α(t) ≤ x(t) ≤ β(t) for all t ∈ I } ∩ S ϕ ( , S , μ).

Clearly C is a bounded closed convex subset of S ϕ ( , S , μ). Define T : C → R I
by
T (x)(t) = h(t, f (x(t))), t ∈ I
For x ∈ S ϕ ( , S , μ), x ◦ f is measurable and by the Caratheodary conditions (a),

(b) and (c), T (x) is also measurable. From (d), α(t) ≤ T (x(t)) ≤ β(t), t ∈ I . So
T (x) ∈ C. For x, y ∈ C, from the definition of pφ , and conditions (d) and (e) and
the increasing nature of ϕ,
! 1
−1
pφ (T x − T y) = ϕ ϕ(|T (x) − T (y)|)(t)dt
0
! 1
= ϕ −1 ϕ(|h(t, x( f (t))) − h(t, y( f (t)))|)dt
0
! 1
≤ ϕ −1 ϕ(g(t))|x( f (t)) − y( f (t))|dt
0
! 1
g(t)
=ϕ ϕ |x( f (t)) − y( f (t))| f (t)dt
0 f (t)
! 1
−1
≤ϕ ϕ(|x( f (t)) − y( f (t))|) f (t)dt
0
! f (1)
−1
≤ϕ ϕ(|x(t) − y(t)|)dt
f (0)
! 1
≤ ϕ −1 ϕ(|x(t) − y(t)|)dt
0
= pϕ (x − y).
Thus T is non-expansive. As S ϕ is a complete uniformly convex bead space. T has a

fixed point by Pasicki’s theorem. Thus the functional equation x(t) = h(t, x( f (t)))
has a solution in S ϕ .
It may be mentioned that for the choice ϕ(t) = t p , p ≥ 1, the pϕ norms coincide
with L p norms.
8.6 Convergence of Iterates in Normed Spaces
While Banach’s contraction principle and its variants are concerned with the conver-
gence of the iterates of the operator under consideration it is also relevant to discuss
the summability of iterates. For instance, Krasnoselski’s theorem investigates the
behaviour of the sequence xn+1 = 21 (xn + T xn ) (vide section 2.2). Mann [12] gen-
eralized Krasnoselski’s theorem to more general sequences generated by regular
matrices. For discussing Mann’s theorem we need the following.
Definition 8.6.1 Let s = (sn ) be a sequence of elements in a Banach space (X, · )

and (cmn ) be a real or complex matrix with m = 1, 2, . . . and n = 1, 2, . . . . Let T
be the linear transformation represented by (cmn ). The transform of s by T denoted
∞

by t = T (s) is the sequence (tm ) defined by tm = cmn sn for each m ∈ N . The
n=1
matrix (cmn ) (or T ) is said to be regular if tm → s as m → ∞ whenever sm → s as
m → ∞.
A classical theorem in summability called the Silverman–Toeplitz theorem is the

following.
Theorem 8.6.2 (Silverman–Toeplitz, see Hardy [9]) An infinite matrix (cmn ) is reg-
ular if and only if the following are true:
(i) lim cmn = 0 for each n ∈ N ;
m→∞
∞
(ii) lim cmn = 1 and
m→∞
∞
n=1

(iii) sup |cmn | ≤ K < +∞ for some K > 0
m
n=1
Corollary 8.6.3 Let A be the infinite triangular matrix satisfying the following con-
ditions:
(a) ai j ≥ 0 for i, j ∈ N;
(b) ai j = 0 for all j > i
8.6 Convergence of Iterates in Normed Spaces 185

i
(c) ai j = 1 for all i ∈ N.
j=1
Then A is regular.
Proof Clearly (b) implies (i) and (c) and (a) imply both (ii) and (iii).
Theorem 8.6.4 (Mann [12]) Let T : E → E be a continuous map on a compact

convex subset E if a Banach space X . Let (xn ) be the sequence of T -iterates generated
n
by x1 ∈ E. Define the sequence (vn ) inductively by vn = ank xk and vn+1 = T (vn ),
k=1
where A = (ank ) and A is the triangular matrix satisfying (a), (b) and (c) of Corollary
8.6.3.
If either of the sequences (xn ) and (vn ) converges, then the other also converges
to the same point and their common limit is a fixed point of T .

n ∞

Proof Let (xn ) converge to p. Since vn = ank xk , p = ank xk as A is regular,
k=1 k=1
vn converges to p. Now xn+1 = T (vn ) converges to both p and T ( p) (T being
continuous) so that p = T ( p). Thus (xn ) and (vn ) both converge to the same fixed
point of T .
If lim vn = q, then as A is regular and (vn ) = A(xn ), lim xn = q. Hence lim
n→∞ n→∞
xn+1 = q = T (q). Thus both (vn ) and (xn ) converge to the same fixed point of T .
Corollary 8.6.5 Let A = (aik ) where

1
fork = 1, 2, . . . , i, i ∈ N
aik = i
0 if k > i
Then A is a regular matrix and if T : E → E is a continuous map on the compact

n
xk
convex subset E of a Banach space X , then xn+1 = T (vn ) where vn = con-
k=1
n
verges if (vn ) converges and vice-versa. In both the cases the limits are the same
fixed point of T .
Theorem 8.6.6 (Mann [12]) Suppose neither {xn } nor {vn } (defined in Theorem
8.6.4) is convergent. Let X be the set of all limit points of {xn } and V the set of all limit
n
points of {vn }. If A satisfies additionally lim ann = 0 and lim |an+1k − ank | =
n→∞ n→∞
h=1
0, then X and V are closed connected sets.
Proof For a separation of V by two non-void closed sets A1 and A2 , we can find v1 ∈
A1 and v2 ∈ A2 such that d(v1 , v2 ) = inf{d(x, y) : x ∈ A and y ∈ B} = r > 0. For

ni
r
vi , we can find am i k vik ∈ B(vi , ) ∩ Ai , i = 1, 2, leading to a contradiction to the
k=1
3
choice of v1 and v2 with d(v1 , v2 ) = r , in view of the Silverman–Toeplitz conditions
n
and the additional hypotheses lim ann = 0 and lim |an+1k − ank | = 0. So V is
n→∞ n→∞
k=1
connected. Since T is continuous and X = T (V ), X is connected. V being compact,
so is X .
When T has a unique fixed point and E = [a, b] in R, the sequence (vn ) converges
to the unique fixed point.
Theorem 8.6.7 (Mann [12]) Let T : [a, b] → [a, b] be continuous with a unique
1
n
fixed point p. Define xn by xn+1 = T (vn ) where vn = xk where x1 ∈ [a, b].
n k=1
Then xn converges to p.
Proof Define A = (ank ) by

1
for k ≤ n
ank = n
0 for k > n
Clearly A is a regular matrix and vn+1 − vn = T (vn+1n )−vn

. So lim (vn+1 − vn ) = 0.
n→∞
As T is continuous with a unique fixed point p, T x − x > 0 for x p. So for each δ > 0 there exists > 0 with |T x − x| ≥ for |x − p| ≥ δ.
n
T (vk ) − vk
Since vn+1 = v1 + , these constraints imply that lim vn = p. Now by
k=1
k+1
Theorem 8.6.4, lim xn = p as well.
Remark 8.6.8 Bailey’s proof of Krasnoselski’s theorem in R (vide chapter 2) is a
special case of the above result.
8.7 Iterations of a Non-expansive Mapping
In this section, two basic results of Ishikawa [10] on the iterates of a non-expansive
mapping are detailed. These are in the setting of a Banach space. We make use of
the following definitions and lemmata.
Definition 8.7.1 Let D be a subset of a Banach space X , T : D → X a map and
x1 ∈ D. By M(x1 , tn , T ) we denote the sequence (xn ) defined by
xn+1 = (1 − tn )xn + tn T xn , where (tn ) is a real sequence.
x1 ∈ D and the real sequence (tn ) are said to satisfy condition (A)
∞
if 0 ≤ tn ≤ b < 1 for all n ∈ N , tn = +∞ and xn ∈ D for all n ∈ N .
n=1
8.7 Iterations of a Non-expansive Mapping 187
∞

Remark 8.7.2 If tn ∈ [a, b] for all n where 0 < a ≤ b < 1, then tn = +∞ and
n=1
0 ≤ tn ≤ b < 1.
Lemma 8.7.3 Let (sn ) be a real sequence and (u n ) a sequence in a Banach space
X . Then for any natural number M
M−1 M

si (1 − si )u i
i=1 i=1
⎧⎛ ⎞⎛ ⎞ ⎫

M ⎨
M−1
M−1
i ⎬
= 1− si u M − ⎝ s j ⎠ ⎝1 − s j ⎠ (u i+1 − si u i )
⎩ ⎭
i=1 i=1 j=i+1 j=1
(8.7.1)
When X is the real line and u i = 1 for all i, we have the special case
M−1 M

si (1 − si )
i=1 i=1
⎧⎛ ⎞⎛ ⎞ ⎫

M ⎨
M−1
M−1
i ⎬
=1− si − ⎝ s j ⎠ ⎝1 − s j ⎠ (1 − si )
⎩ ⎭
i=1 i=1 j=i+1 j=1

n
n
In this and what follows and are defined as 0 and 1 for n < m.
i=m i=m
Proof The proof is by induction on M. When M = 1, the result is obvious. Suppose

it is true for some M > 1, then
⎧⎛ ⎞⎛ ⎞ ⎫
M ⎨ M i ⎬
⎝ s j ⎠ ⎝1 − s j ⎠ (u i+1 − si u i )
⎩ ⎭
i=1 j=i+1 j=1
⎧⎛ ⎞⎛ ⎞
⎨ M−1
M−1 i
= sM ⎝ s j ⎠ ⎝1 − s j ⎠ (u i+1 − si u i )
⎩
i=1 j=i+1 j=1

M M
−s M 1 − si u M + 1 − si u M−1
i=1 i=1
M

M
M−1
= sM 1− si u M − (1 − si )u i
i=1 i=1 i=1

M
M
−s M 1 − si u M + 1 − si u M−1
i=1 i=1
M

M
M
=− si (1 − si )u i + 1 − si u M−1
i=1 i=1 i=1
Whence we have (the right-hand side of (8.7.1) withM + 1forM)
⎧⎛ ⎞⎛ ⎞ ⎫

M+1 M ⎨ M i ⎬
= 1− si u M+1 − ⎝ s j ⎠ ⎝1 − s j ⎠ (u i+1 − si u i )
⎩ ⎭
i=1 i=1 j=i+1 j=1
M M
M
M
= 1 − s M+1 si u M+1 + si (1 − si )u i − 1 − si u M+1
i=1 i=1 i=1 i=1
M M+1

= si (1 − si )u i
i=1 i=1
So by induction this lemma follows.
Lemma 8.7.4 Let D ⊆ X , a Banach space and T : D → X a non-expansive map. If

x1 ∈ D and (tn ) satisfy condition A and M(x1 , tn , T ) is bounded, then xn − T xn → 0
as n → ∞.
Proof
xn+1 − T xn+1 = (1 − tn )xn + tn T xn − T xn+1

= (1 − tn )(xn − T xn ) + T xn − T xn+1
≤ (1 − tn ) xn − T xn + xn − xn+1
= (1 − tn ) xn − T xn + xn − (1 − tn )xn + tn T xn )
= xn − T xn
Since ( xn − T xn ) is a non-increasing sequence of non-negative numbers, lim

n→∞
xn − T xn exists. Let lim xn − T xn = r > 0. So for > 0 there exists a natural
n→∞
number m such that r ≤ xm+i − T xm+i ≤ (1 + )r for all i ∈ N. As T is non-
expansive
(T xm+i+1 − xm+i+1 ) − (1 − tm+i )(T xm+i − xm+i )

= T ((1 − tm+i )xm+i + tm+i T xm+i )
− ((1 − tm+i )xm+i + tm+i T xm+i ) − (1 − tm+i )(T xm+i − xm+i )
= T ((1 − tm+i )xm+i + tm+i T xm+i ) − T xm+i
≤ tm+i xm+i − T xm+i ≤ tm+i (1 + )r (8.7.2)
∞

Since (xn ) is bounded and tn = +∞, we can find a natural number N such that
n=1
N −1

N
r tm+i ≤ δ(M) + 1 ≤ r tm+i
i=1 i=1
where δ(M) = sup{|xi − x j | : i, j ∈ N}

Set si = 1 − tm+i , vi = T xm+i − xm+i for i ∈ N in (8.7.2) to get
u i+1 − si u i = T xm+i+1 − xm+i+1 − (1 − tm+i )(T xm+i − xm+i )

≤ tm+i (1 + )r = (1 − si )(1 + )r
and

N
xm+N +1 − xm+1 = {((1 − tm+i )xm+i + tm+i T xm+i ) − xm+i }
i=1

N
N
= tm+i (T xm+i − xm+i ) = (1 − si )u i
i=1 i=1
Using Lemma 8.7.3 and the above inequalities we get

N −1 N −1 N

si xm+N +1 − xm+1 = si (1 − si )u i

i=1 i=1 i=1
⎧ ⎛ ⎞ ⎛ ⎞ ⎫
N N −1 ⎨

N −1 i ⎬
≥ 1− si u N − ⎝ s j ⎠ ⎝1 − s j ⎠ u i+1 − si u i
⎩ ⎭
i=1 i=1 j=i+1 j=1
⎧⎛ ⎞⎛ ⎞ ⎫
N N −1 ⎨

N −1 i ⎬
≥ 1− si r − ⎝ s j ⎠ ⎝1 − s j ⎠ (1 − si )(1 − )r
⎩ ⎭
i=1 i=1 j=i+1 j=1
⎡ ⎧ ⎛ ⎞ ⎫⎤
N N −1 ⎨ N
−1 i ⎬
= ⎣1 − si − s j ⎝1 − s j ⎠ (1 − si ) ⎦ r
⎩ ⎭
i=1 i=1 j=i+1 j=1
⎧⎛ ⎞⎛ ⎞ ⎫
N −1 ⎨

N −1 i ⎬
− r ⎝ s j ⎠ ⎝1 − s j ⎠ (1 − si )
⎩ ⎭
i=1 j=i+1 j=1
since si = 1 − tm+i ≥ 1 − b > 0, (8.7.1), the choice of m and Lemma 8.7.3 imply
N −1 −1

N
xm+N +1 − xm+1 ≥r (1 − si ) − r si
i=1 i=1
N −1 N

N
× 1− si − si (1 − si )
i=1 i=1 i=1
N −1 −1

N
≥r (1 − si ) − r si
i=1 i=1

N
N −1
=r tm+i − r (1 − tm+i )−1
i=1 i=1

N −1
≥ δ M + 1 − r (1 − tm+i )−1 (8.7.3)
i=1
As log(1 + y) ≤ y for y > −1, we have

N −1
N −1
(1 − tm+i )−1 = (1 + tm+i (1 − tm+i )−1 )
i=1 i=1
N −1

−1
= exp log(1 + tm+i (1 − tm+i ) )
i=1
N −1

−1
≤ exp tm+i (1 − tm+i )
i=1
N −1

−1
≤ exp (1 − b) tm+i
i=1
≤ exp{(1 − b−1 )(δ(M) + 1)r −1 } (8.7.4)
Now (8.7.3) and (8.7.4) give
δ(M) + 1 − r exp{(1 − b)−1 (δ(M) + 1)r −1 }

≤ xm+N +1 − xm+1 ≤ δ(M).
Since > 0 is arbitrary, proceeding to the limit in the last inequality as → 0, we

get
δ(M) + 1 ≤ δ(M), a contradiction.
So r = 0. Thus lim xn − T xn = 0.
n→∞
Remark 8.7.5 Let T : D → D be a non-expansive map of a convex subset D of a

Banach space with T (D), a bounded subset of D. For 0 < t < 1, let (1 − t)I + t T
be denoted by Tt where I is the identity map. Then M(x1 , t, T ) is bounded since it is
a sequence in the convex hull of T (D) ∪ {x1 }. Since Ttn x1 − Ttn−1 x1 = t (T xn − xn ),

by Lemma 8.7.4 Tt is asymptotically regular in the sense that lim Ttn+1 x − Ttn x =
n→∞
0 for x ∈ D.
The iteration procedure M(x1 , tn , T ) can be used to approximate a fixed point of

a non-expansive map under special conditions.
Theorem 8.7.6 (Ishikawa [10]) Let T : D → X be a non-expansive map, where D

is a closed subset of a Banach space X such that T (D) is a subset of a compact
subset of X . Let x1 ∈ D and {tn } a non-negative sequence such that the condition
(A) of Definition 8.7.1 is satisfied. Then T has a fixed point in D and M(x1 , tn , T )
converges to a fixed point of T .
Proof Let D0 be the closure of the convex hull of T (D) ∪ {x1 }. By Mazur’s theorem
D0 is compact. Now M(x1 , tn , T ) lies in D0 . Condition (A) of Definition 8.7.1 along
with this implies that this sequence which indeed lies in D being in the compact set
D0 has a subsequence (xni ) converging to u ∈ D0 . Since (xni ) is in D and D is closed,
u ∈ D. From the boundedness of D0 and Lemma 8.7.4, lim xni − T xni = 0. As
i→∞
T is non-expansive
u − T u = T u − T xni + T xni − xni + xni − u

≤ 2 u − xni + xni − T xni
Hence u = T u, in view of lim xni − T xni = lim u − xni = 0.

i→∞ i→∞
Also
xn+1 − u = (1 − tn )xn + tn T xn − u
= (1 − tn )(xn − u) + tn (T xn − T u)
≤ xn − u for all n ∈ N.
As lim xni = u and xn − u ≤ xni − u for all n ≥ n i , lim xn = u.

i→∞ n→∞
Corollary 8.7.7 Let D be a closed subset of a Banach space X and T : D → X

a non-expansive map such that T (D) is contained in a compact subset of X . If for
some t ∈ (0, 1), (1 − t)x + t T x ∈ D for all x ∈ D then T has a fixed point in D
and for any x1 ∈ D, M(x1 , t, T ) converges to a fixed point of T .
Corollary 8.7.8 Let D be a closed convex subset of a Banach space X and T , a

non-expansive map from D into a compact subset of D. Then T has a fixed point in
D and M(x1 , 21 , T ) converges to a fixed point of T , where x1 is any element of D.
Remark 8.7.9 Corollary 8.7.8 was proved for uniformly convex Banach spaces by
Krasnoselski.
The assumption on the compactness of T can be dispensed with as in the following.

Theorem 8.7.10 (Ishikawa [10]) Let T : D → X be a non-expansive mapping

where D is a closed subset of a Banach space X with a non-empty set of fixed points F
in D. Suppose there exists f : R+ → R+ (= [0, ∞)) such that f is non-decreasing,
f (0) = 0, f (r ) > 0 for r > 0 with
x − T x ≥ f (d(x, F)) for all x ∈ D, d(x, F) = inf{d(x, y) : y ∈ F}. (8.7.5)
If for some x1 ∈ D and (tn ) condition (A) is satisfied, then M(x1 , tn , T ) converges to
some fixed point of T in F.
Proof If x1 ∈ F, then theorem is obvious. Let x1 ∈

/ F. So for u ∈ F, xn − u ≥
T xn − u . We therefore have
xn+1 − u = (1 − tn )xn + tn T xn − u ≤ xn − u
So d(xn+1 , F) ≤ d(xn , F) for all n ∈ N. Let r = lim d(xn , F). By definition of f ,

n→∞
xn − T xn ≥ f (d(xn , F)) ≥ f (r ).
Clearly M(x1 , tn , T ) is a bounded sequence in D. So by Lemma 8.7.4. xn −

T xn → 0 as n → ∞ and f (r ) = 0. So r = 0 and lim d(xn , F) = 0. So for each
n→∞
i ∈ N we can find Ni ∈ N and u i ∈ F such that x Ni − u i < 2−i with Ni+1 > Ni .
So for n ≥ Ni , xn − u i < 2−i , and for i < j
u i − u j ≤ u i − x Ni+1 + x Ni+1 − u i+1 + · · ·

u j−1 − x N j + x N j − u j
−i
<2 + 2−i−1 · · · + 2− j .
Thus (u i ) is a Cauchy sequence in the closed set F of the Banach space X and hence
it converges to some u ∈ F. Given > 0 we can find i 0 > 0 such that 2−i0 < 2 and
u i0 − u < 2 . So for n > Ni0 ,

xn − u ≤ xn − u i0 + u i0 − u < + = .
2 2
Thus M(x1 , tn , T ) converges to a fixed point of T in D.
Condition (8.7.5) in the above theorem was originally considered by Senter and
Dotson.
8.8 A Generalization of the Contraction Principle Based on Combinatorics 193
8.8 A Generalization of the Contraction Principle Based on

Combinatorics
Merrifield, Rothschild and Stein [15] obtained a generalization of the contraction

principle based on Ramsey’s theorem in combinatorics. Since it involves the study of
certain subsequences of iterates, it is discussed in this section. We need the following.
Theorem 8.8.1 (Ramsey) Let S be an infinite set, n a natural number. Suppose that
every subset S of cardinality n is assigned one of a finite number of colours. Then
there exists an infinite subset T of S such that T is monochromatic (i.e. every subset
T of cardinality n has the same colour).
The following lemma can be derived from the above theorem.
Lemma 8.8.2 Let m and n be natural numbers. Suppose G is a graph whose vertex
set is a disjoint union of countably many blocks, each of size m. Further suppose that
each edge has its two endpoints in distinct blocks and for any n blocks assume that
there is at least one edge having its endpoints in two of those blocks. Then there is
an infinite path in G, visiting no block more than once.
Proof Name the blocks as B0 , B1 , . . ., and number the vertices in each block Bi as
v(i, 1), v(i, 2), . . ., v(i, m). Colour the pairs of natural numbers with m 2 + 1 colours
by assigning as the colour of the pair (i, j) with i < j, either some pair ( p, q) so
that v(i, p) is adjacent to v( j, q) on G or if there is no edge between Bi and B j , the
special colour ‘none’ is assigned. By Ramsey’s Theorem 8.8.1 there is an infinite
set H of natural numbers, every pair of which has the same colour. By hypothesis,
this colour cannot be ‘none’. So let it be ( p, q). That is if i < j are both in H , then
v(i, p) is adjacent to v( j, q). Let h(i) be the ith element of H . The desired infinite
path is v(h(1), p), v(h(3), q), v(h(2), p), v(h(5), q), v(h(4), p), v(h(7), q), . . ..
The following is an elementary consequence of triangle inequality.
Lemma 8.8.3 Let (X, d) be a metric space and 0 < α < 1, J a natural number
satisfying
min{d(T i x, T i y) : i = 1, 2, . . . , J } ≤ αd(x, y)
for all x, y ∈ X . Then for each x, there is a bounded subsequence {T a(n) x : n =

1, 2, . . .} of {T n (x) : n = 1, 2, . . .} such that a(n + 1) − a(n) ≤ J .
Proof Let C = max{d(x, T k x) : k = 1, 2, . . . , J }. We show that we can construct

a sequence a(n) of integers such that d(x, T a(n) x) < 1−α
C
. Let a(1) = 1. Since 0 <
α < 1, 0 < 1 − α < 1, clearly d(x, T x) < 1−α . Proceeding inductively if d(x, T a(n)
C
x) < 1−α
C
, we have for some k < J , d(T k x, T a(n)+k x) < 1−α C
. So by triangular
αC
inequality we have d(x, T a(n)+k
x) ≤ d(x, T x) + d(T x, T
k k a(n)+k
x) ≤ C + 1−α =
C
1−α
. Letting a(n + 1) = a(n) + k completes the proof.
Merrifield, Rothschild and Stein [15] proved the following generalization of the
contraction principle.
Theorem 8.8.4 (Merrifield, Rothschild and Stein [15]) Let T : X → X be a con-

tinuous map. Suppose there is a positive integer J and 0 < α < 1 such that for
x, y ∈ X
min{d(T k x T k y) : 1 ≤ k ≤ J } ≤ αd(x, y). Then T has a fixed point.
Proof Let x ∈ X and i, k = d(T i x, T k x), T i x = x and for a real number r , [r ]
denote the greatest integer less than or equal to r .
By Lemma 8.8.3, there is a sequence {n i : i = 1, 2, . . .} of natural numbers with
0, n i ≤ C and n i+1 − n i ≤ J . Applying the generalized contraction hypothesis to
each pair of points x and T ni x we get sequences {qi j : j = 1, 2, . . .} one sequence
) q *each i such that qi j , n i + qi j ≤ Cα and qi j+1 − qi ≤ J . If q = qi j , then j ≥
j
for
J
≥ q/J − 1, in which case q, n i + q ≤ Cα q/J −1 = C0 Q q where C0 = Cα and
1
Q = α J < 1. Note that if we have q, n i + q ≤ C0 Q q and q, n j + q ≤ C0 Q q for
two different integers i and j by the triangle inequality we have
n i + q, n j + q ≤ n i + q, q + q, n j + q
≤ 2C0 Q q
We say that an integer q is represented if there are infinitely many integers i

for which q, n i + q ≤ C0 Q q . If q is represented and q, n i + q ≤ C0 Q q , we
say that i is a representative of q. If A is a set of integers, let r (A) = {q : q ∈
a, q is represented} and R(A) = {i : for some q in r (A) is a representative of q}
Let A be a set J consecutive integers. The condition qi j+1 − qi j ≤ J together with
the pigeonhole principle shows that at least one member of A has to be represented.
Also for all but finitely many i, there exists q ∈ r (A) such that i is a representative
of q : thus for some integer I0 , i ≥ I0 implies i ∈ R(A).
Again for A, a set of J consecutive integers, there is an integer λ(A) such that
for m ≥ λ(A) some integer of the form n i + q lies in the set {m, m + 1, . . . , m +
2J − 1}, where q ∈ r (A) and i is a representative of q. Since there exists I0 such
that i ≥ I0 implies i ∈ R(A).
Let λ(A) = max{ j : j ∈ A} + n I0 where A is a set of J consecutive integers.
Since each i with i ≥ I0 is a representative of some q ∈ r (A) and if i is a represen-
tative of q ∈ r (A), then n i+1 + q − (n i + q) = (n i+1 − n i ) + (q − q) ≤ 2J .
Finally, if A is a set of J consecutive integers, let N Q(A) = {n i + q : q ∈
r (A) and i is a representative of q}. It may be noted that if A1 , . . . , A2J +1 are dis-
joint sets of J consecutive integers each, then for a = max{λ(Ak ) : k = 1, 2, 2J + 1}
and m ≥ a any set {m, m + 1, . . . , m + 2J − 1} must contain an integer common
to the sets N Q(A j ) and N Q(Ak ) for which j = k. By partitioning the sequence
{a, a + 1, a + 2, . . .} into blocks of length 2J and applying the pigeonhole principle
we see that there are two sets N Q(A j ) and N Q(Ak ) with j = k having infinitely
many elements in common. Invoking the pigeonhole principle, there exist integers
q ∈ A j and q ∈ Ak so that there are infinitely many integers common to N Q(A j )
8.8 A Generalization of the Contraction Principle Based on Combinatorics 195
and N Q(Ak ) that can be expressed both in the form n i + q (an integer in N Q(A j ))
and n p + q (an integer N Q(Ak )).
Regard each integer as a vertex in a graph and partition the integers into a disjoint
union of blocks Bk = {(k − 1)J + 1, (k − 1)J + 2, . . . k J } for k = 1, 2, . . .. We say
that two vertices q and q in distinct blocks B j and Bk respectively are connected by
an edge if there are infinitely many integers that can be expressed in both the forms
n i + q and n p + q . From the above it is clear that for any collection of 2 J + 1
blocks, at least one edge has endpoints in two distinct blocks.
By Lemma 8.8.2 that there is an infinite path through the graph passing through
each block no more than once. Denote the vertices traversed in this path by
{r j : j = 1, 2, . . .}. Choose the sequences of integers {s j : j ∈ N}, and {t j : j ∈ N}
from {n j : j ∈ N} with the following three properties:
(i) if r j ∈ Bk , both r j + s j and r j + t j ∈ N Q(Bk );
(ii) r j + t j = r j+1 + s j+1 ;
(iii) r j + s j < r j+1 + s j+1
Consider the sequence of iterates with exponents r j + s j . Observe that
∞
∞
∞

r j + s j , r j+1 + s j+1 = r j + s j , r j + t j ≤ 2K 0 Q r j
j=1 j=1 j=1
Since the {r j : j ∈ N} are all distinct, the above series converges. So the sequence
of iterates is a Cauchy sequence. As X is complete, the resulting limit of this sequence
will be shown to be a fixed point of T . Let (T m i x) converge to z. We can even choose it
so that m i+1 > m i + J for all i. Since T is continuous, T m i +k → T k z for 1 ≤ k ≤ J .
Define L k = T k z for 0 ≤ k ≤ J . We claim that L j+1 = L j for some j < J . Since
T (L j ) = L j+1 it follows that L j is a fixed point of T .
By the generalized contraction hypothesis applied to x and y = T x, we con-
clude that for each i ∈ N, there exists an integer ji with 0 ≤ ji ≤ J − 1 and
d(T m j + ji x, T m i + ji +1 x) ≤ αri d(x, T x) where ri → ∞ (this can be seen by using
an argument in the second paragraph of the proof of this theorem). By the pigeon-
hole principle, there is an integer k with 0 ≤ k ≤ J − 1 and ji = k for infinitely
many i. For those i with ji = k for infinitely many i we have
d(L k , L k+1 ) ≤ d(L k , T m j +k x) + d(T m i +k x, T m i +k+1 x)

+ d(T m i +k+1 x, L k+1 )
≤ d(L k , T m j +k x) + αri d(x, T x)
+ d(T m i +k+1 x, L k+1 )
As i → ∞, each of the three terms on the right-hand side of the above inequality
tends to zero. Thus L k = L k+1 or L k = T k z is a fixed point of T .
Remark 8.8.5 It is not known if the theorem is true even when T is not continuous
for all J .
References
1. Alspach, D.E.: A fixed point free nonexpansive map. Proc. Am. Math. Soc. 82, 423–424 (1981)
2. Browder, F.E.: Nonexpansive nonlinear operators in a Banach space. Proc. Natl. Acad. Sci.
U.S.A. 54, 1041–1044 (1965)
3. Dieudonne, J.: Foundations of Modern Analysis. Academic, New York (1962)
4. Dinca, G., Mawhin, J.: A constructive fixed point approach to the existence of a triangle with
prescribed angle bisector lengths. Bull. Belg. Math. Soc. Simon Stevin 17, 333–341 (2010)
5. Dotson Jr., W.J.: Fixed point theorems for nonexpansive mappings on star-shaped subsets of
Banach spaces. J. Lond. Math. Soc. 2(4), 408–410 (1972)
6. Edelstein, M.: On fixed and periodic points under contractive mappings. J. Lond. Math. Soc.
37, 74–79 (1962)
7. Goebel, K.: An elementary proof of the fixed point theorem of Browder and Kirk. Mich. Math.
J. 16, 381–383 (1969)
8. Gohde, D.: Uber Fixpunkte bei stetigen Selbst-abbildungen mit kompakten Iterierten. Math.
Nachr. 28, 45–55 (1964)
9. Hardy, G.H.: Divergent Series (Oxford, 1949)
10. Ishikawa, S.: Fixed points and iteration of a nonexpansive mapping in a Banach space. Proc.
Am. Math. Soc. 59, 65–71 (1976)
11. Kirk, W.A.: A fixed point theorem for mappings which do not increase distances. Am. Math.
Mon. 72, 1004–1006 (1965)
12. Mann, W.R.: Mean value methods in iteration. Proc. Am. Math. Soc. 4, 506–510 (1953)
13. Matkowski, J.: Fixed point theorem in a uniformly convex paranormed space and its application.
Topol. Appl. 160, 524–531 (2013)
14. Meinardus, G.: Approximation of Functions: Theory and Numerical Methods. Springer, Berlin
(1964)
15. Merrifield, J., Rothschild, B., Stein Jr., J.D.: An application of Ramsey’s theorem to the Banach
contraction principle. Proc. Am. Math. Soc. 130, 927–933 (2002)
16. Mironescu, P., Panaitopol, L.: The existence of a triangle with prescribed angle bisector lengths.
Am. Math. Mon. 101, 58–60 (1994)
17. Pasicki, L.: A basic fixed point theorem. Bull. Pol. Acad. Sci. Math. 54, 85–88 (2006)
18. Pasicki, L.: Bead spaces and fixed point theorems. Topol. Appl. 156, 1811–1816 (2009)
19. Pasicki, L.: Uniformly convex spaces, bead spaces and equivalence conditions. Czechoslov.
Math. J. 61(2), 383–388 (2011)
20. Pasicki, L.: A comment to Matkowski’s paper. Topol. Appl. 160, 951–952 (2013)
21. Reich, S.: Remarks on fixed points. Att. Acad. Naq. Lincei Rand. Sci. Fis. Mat. Matur. Serie
VIII, LII fasc 5, 689–697 (1972)
22. Rosenholtz, I.: On a fixed point problem of D.R. Smart. Proc. Am. Math. Soc. 55, 252 (1976)
23. Sims, B.: Examples of fixed point free mappings. In: Kirk, W.A., Sims, B. (eds.) Handbook of
Metric Fixed Point Theory. Kluwer Academic Publishers, Dordrecht (2001)
24. Van der Berg, Over dee bepaling van een drichoek, waarvan de deellijmen derdrie supplemen-
taire heoken gegevan zijn Niew Archiev voor Wiskunde, vol. 17, pp. 191–205 (1890)
Chapter 9
Geometric Aspects of Banach Spaces
and Non-expansive Mappings
9.1 Introduction
In this chapter, we outline the proof that a reflexive non-square Banach space has
fixed point property for non-expansive mappings on bounded closed convex sets. To
this end, some definitions are in order.
Definition 9.1.1 For a Banach space (X, · ), the closed unit ball and the unit
sphere in X are denoted by B X and S X , respectively. The Clarkson modulus of
convexity of X is a function δ X : [0, 2] → [0, 1] defined by

x + y
δ X () = inf 1 − : x, y ∈ S X , x − y ≥ .
2
X is called uniformly convex if δ X () > 0 for 0 < < 2.
Remark 9.1.2 δ X is continuous on [0, 2) increasing on [0, 2], strictly increas-

ing on [0 , 2] where 0 = 0 (X ) = sup{ ∈ (0, 2] : δ X () = 0} is called the coef-
0 (X )
ficient of convexity of X . Also δ X () ≤ 2 , lim− δ X () = 1 − and δ X () ≤
→2 2
1 − 1 − 4 = δ2 ().
2
For a real Banach space X with dim X ≥ 2 or a complex Banach space X of

dim X ≥ 1, it can be shown that

x + y
δ X () = inf 1 − : x, y ∈ B X , x − y ≥
2

x + y
= inf 1 − : x, y ∈ S X , x − y =
2

x + y
= inf 1 − : x, y ∈ B X , x − y =
2
https://doi.org/10.1007/978-981-13-3158-9_9
198 9 Geometric Aspects of Banach Spaces and Non-expansive Mappings
See Day [1].
Definition 9.1.3 The modulus of smoothness of a Banach space X is a function

ρ X : R+ → R+ defined by

x + t y + x − t y
ρ X () = sup − 1 : x, y ∈ S X
2
ρ(t)
X is called uniformly smooth if ρ0 (X ) = lim+ = 0.
t→0 t
Remark 9.1.4 ρx is increasing, continuous and convex on R+ and ρ X (0) = 0
with ρ X (t) ≤ t. Also for a real Banach space√X with dim X ≥ 2 (or a complex
Banach space X with dim X ≥ 1), ρ X (t) ≥ 1 + t 2 − 1 = ρ2 (t). Lindenstrauss
[13] proved the important relations

t
ρ (t) = sup
X∗ − δ X (t) : ∈ [0, 2]
2

t
ρ X (t) = sup − δ X (t) : ∈ [0, 2] .
∗
2
So ρ X (t) = ρ X ∗∗ (t) and X is uniformly convex (uniformly smooth) if and only if its
dual X ∗ is uniformly smooth (uniformly convex). Also X is reflexive whenever it is
uniformly convex or uniformly smooth.
Definition 9.1.5 (James [7]) A Banach space X is called uniformly non-square if

there exists δ ∈ (0, 1) such that for any x, y ∈ S X either x+y
2
≤ 1 − δ or x−y
2
≤
1 − δ. The constant J (X ) defined by
J (X ) = sup{min(x + y, x − y) : x, y ∈ S X }
is called the non-square or James constant of X .
Proposition 9.1.6 (Kato [10]) Let X be a real Banach space with dim(X ) ≥ 2 (or
a complex Banach space wth dim(X ) ≥ 1). Then the following are equivalent.
(i) X is uniformly non-square;
(ii) δ X () > 0 for some ∈ (0, 2);
(iii) 0 (X ) < 2;
(iv) J (X ) < 2;
(v) ρ X (t0 ) < t0 for some t0 > 0;
(vi) ρ X (t) < t for all t > 0;
ρ X (t)
(vii) ρ X (0) = lim < 1;
∗
t→0 t
(viii) 0 (X ) < 2;
ρ X ∗ (t)
(ix) ρ X ∗ (0) = lim < 1.
t→0 t
9.1 Introduction 199
We provide only the gist of the main arguments in the

Proof We note that X is uniformly non-square
⇔ for some δ ∈ (0, 1), x, y ∈ S X and
x − y x + y
> 1 − δ implies >1−δ
2 2
⇔ for some δ ∈ (0, 1), x, y ∈ S X and
x − y x + y
> 1 − δ implies 1 − ≥δ
2 2
⇔ for some δ ∈ (0, 1), δ X (2 − 2δ) ≥ δ
(i) ⇒ (ii) If X is uniformly non-square set = 2 − 2δ ∈ (0, 2) in the above equiva-
lences to get δ X () > 1 − 2 (= δ).
(ii) ⇒ (i) If for some 0 ∈ (0, 2), δ X (0 ) ≥ η0 > 0, where η0 ∈ (0, 1), then for 2 −
2δ = ∈ [0 , 2), δ ∈ 0, 1 − 20 and δ X (2 − 2δ) = δ X () = δ X (0 ) ≥ η0 > 0. This
means that for any x, y ∈ S X , with x − y ≥ 2 − 2δ necessarily 1 − x+y 2
≥ η0 .
If δ = min{δ, η0 ) then δ ∈ (0, 1). If x−y
2
≤ 1 − δ
, we are done. If x+y
2
> 1 − δ,
x+y x+y
then 1 − 2 ≥ η0 or 2 ≤ 1 − η0 ≤ 1 − δ . So X is uniformly non-square. (ii)
⇔ (iii) and (i) ⇔ (iv) follow from the definition. (v) ⇔ (vii) follows from the fact
that ρ Xt(t) is increasing. (vii) ⇔ (viii) and (ix) ⇔ (iii) follow from
ρx (t)
0 (X ∗ ) = 2ρ X (0) = 2 lim and
t→0 t
ρx ∗ (t)
0 (X ) = 2ρ X ∗ (0) = 2 lim .
t→0 t
We now show that (v) ⇔ (vi). If ρ X (t0 ) = t0 for some t0 > 0 then ρ X (t) = t for all
t > 0. Since ρ Xt(t) is increasing and ρ X (t) ≤ t, it follows that for t ≥ t0 , 1 = ρ Xt(t0 0 ) ≤
ρ X (t)
t
≤ 1 or ρ X (t) = t for t ≥ t0 . Let 0 < t < t0 and ρ X (t) < t. Since ρ X is convex
for t1 > t0

t0 − t t1 − t0
t0 = ρ X (t0 ) = ρ X t1 + t
t1 − t t1 − t
t0 − t t 1 − t0
≤ ρ X (t1 ) + ρ X (t)
t1 − t t1 − t
t0 − t t1 − t0
< t1 + t = t0 ,
t1 − t t1 − t
a contradiction. So ρ X (t) = t.
Remark 9.1.7 A uniformly non-square Banach space is super-reflexive (James [8])
if it has an equivalent norm ||| · · · ||| in which it is uniformly convex. The converse
is not true in the sense that δ(X,| |) () = 0 for 0 < < 2.
Example 9.1.8 For 1 < p < ∞, consider new norms in p as follows:

⎧ 1p ⎫
⎨ ∞ ⎬
|||x||| = max |x1 | + |x2 |, |xk | p
⎩ ⎭
k=3
⎧ 1p ⎫
⎨ ∞ ⎬
|||x||| = max |x1 |, |x2 |, |xk | p
⎩ ⎭
k=3
||x|| p ≤ 2 p |||x||| ≤ 2 p+1 , |||x||| ≤ 2 p +1 |||x||| ≤ 4||x|| p . Since p is uniformly con-

1 1 1
vex ( p , ||| · |||) and ( p , ||| · ||| ) are super-reflexive but not uniformly non- square
sine δ X () = 0 for all 0 < < 2. To see this take x = (1, 0, 0, . . .) and y =
(0, 1, 0, . . .).
From the above example, one can see that for any real Banach space X with
dim(X ) ≥ 2, there is an equivalent norm in which X is not uniformly non-square.
Garcia-Falset et al. [5] proved that in uniformly non- square Banach spaces, every
bounded closed convex subset has the fixed point property for non-expansive map-
pings. This is achieved by studying the properties of certain coefficients associated
with the geometry of Banach spaces.
9.2 Coefficients of Banach Spaces and Fixed Points
Dominguez-Benavides [3] defined in a Banach space X , the following parameters,

in terms of asymptotic diameter and radius of a sequence.
Definition 9.2.1 For a sequence (xn ) in X
diam a (xn ) = lim sup{||xn − xm : n, m ≥ k}

k
ra (xn ) = inf{lim sup ||xn − y|| : y ∈ (xn )}
n
diam (x )
a n
W C S(X ) = inf : (xn ) is a weakly convergent sequence
ra (xn )

which is not norm convergent
(W C S(X ) is called Bynum’s weakly convergent sequence coefficient of X . If

W C S(X ) > 1, then X has weak normal structure in the sense that every weakly
compact convex subset of X with more than one element is not diametral. X has
normal structure if each bounded convex subset with more than one element has a
non-diametral point.
9.2 Coefficients of Banach Spaces and Fixed Points 201
Definition 9.2.2 (Dominguez Benavides [3]) For a Banach space X and a ≥ 0, the
parameter
R(a, X ) := sup{lim inf ||x + xn || : ||x|| ≤ a and

n→∞
(xn ) weakly null sequence with D[(xn )] ≤ 1}

Here D[(xn )] := lim sup lim sup ||xn − xm || ≤ 1.
n→∞
m→∞
The coefficient M(X ) := sup R(a,X
1+a
)
:a≥0 .
R(X ) := sup{lim||x + xn ||(xn ) weakly null in B X and x ∈ Bx }.
Kirk [11] proved that every closed bounded subset having normal structure in
a reflexive Banach space has the fixed point property for non-expansive self-maps.
This can be deduced from the following lemma due independently to Goebel [6] and
Karlovitz [9], as shown in the proof of Theorem 9.2.5.
Lemma 9.2.3 (Goebel [6], Karlovitz [9]) Lex X be a Banach space, C0 , a weakly
compact convex subset of X and T : C0 → C0 a non-expansive map. Let C0 be a
minimal closed convex set that is invariant under T . (i.e. no proper closed convex
subset of C0 is invariant under T ). For each sequence of xn in C0 with lim ||xn −
n→∞
T xn || = 0 and for each x ∈ C0 lim ||xn − T xn || = diamC0 (such a sequence xn
n→∞
with lim ||xn − T xn || = 0 is called a sequence of approximate fixed points).
n→∞
Proof For y ∈ C0 , let s = lim sup ||y − xn ||. Let D = {x ∈ C0 : lim sup ||x − xn || ≤
n→∞ n→∞
s}. Clearly D is non-void closed and convex. D is invariant under T , since
||T x − xn || ≤ ||T x − T xn || + ||T xn − xn ||

≤ ||x − xn || + ||T xn − xn ||
and ||xn − T xn || → 0. So by the minimality of C0 , D = C0 . We can find a subse-

quence (xn ) of (xn ) with lim ||y − xn || = s . Let z ∈ C0 be such that ||z − xn || → t
for some subsequence {xn } of {xn }. Define E = {x ∈ C0 : lim sup x − xn || ≤
min{t, s }}. By repeating this argument, it follows that E = C0 . So y, z ∈ E and
t = s . Hence, for each x ∈ C0 lim ||x − xn || exists and equals s .
We claim that s = r = diam C0 . Whence it follows that ||xn − y|| → r when-
ever {||y − xn ||} converges and consequently by the boundedness ||y − xn || → r
for the entire sequence. Repetition of this argument by replacing {xn } by {xn } shows
that lim ||x − xn || = diam C0 .
Consider F = {u ∈ C0 : ||u − x|| ≤ s for each x ∈ C0 }. F is non-void as there
is a weakly convergent subsequence say {xn } with limit z. Because ||x − xn || → s
for each x ∈ C0 , ||x − z|| ≤ s for each x ∈ C0 . So z ∈ F. If s < r then F is a proper
subset of C0 , contradicting the minimality of C0 , since F is invariant under T , as
well. As C0 is minimal closed convex subset invariant under T , closed convex hull
m
of T (C0 ) = C0 . For u ∈ C0 , given > 0, we can find v = λk T xk with xk ∈ C0 ,
k=1
λk > 0 and λk = 1 with ||u − v|| ≤ . For w ∈ F, ||T w − u|| ≤ ||T w − v|| +
||u − v|| ≤ λk ||T w − T xk || + ||v − u|| ≤ λk ||w − xk || + ||v − u|| ≤ s + .
So T w ∈ F. Thus F is T -invariant. Hence the lemma.
At this stage we recall the definition of normal structure.
Definition 9.2.4 A convex subset K of of a Banach space X is said to have normal
structure if for each bounded convex subset K 1 of K containing more than one point,
there is a non-diametral point in the sense that for some x0 ∈ K 1 , sup{||x0 − k|| :
k ∈ K 1 } < diam K 1 .
With this we can prove Kirk’s theorem [11].
Theorem 9.2.5 Let K be a non-empty closed convex bounded subset of a reflexive
Banach space X such that K has normal structure. If T : K → K is non-expansive
then T has a fixed point in K .
Proof Since K is a bounded closed convex subset of the reflexive Banach space
X , K is weakly compact. By a standard application of Zorn’s lemma, K has a
minimal non-empty
closed convex subset C0 invariant under T . Let a ∈ C0 . Then
Tn (x) = an + 1 − n1 T xn , n ≥ 2 is a contraction mapping C0 into itself and hence

has a unique fixed point xn . Thus xn = an + 1 − n1 T xn for each n ≥ 2. Further
lim ||xn − T xn || → 0 as n → ∞. So by Goebel-Karlovitz Lemma 9.2.3. lim ||x −
n→∞
xn || = diam C0 for all x ∈ C0 . If C0 contains more than a singleton, it contains a
non-diametral point a such that sup ||a − x|| < diam C0 as C0 ⊆ K is a closed
convex subset of K and K has normal structure. This implies that lim ||a − xn || ≤
n→∞
sup ||a − xn || < diam C0 , contradicting Lemma 9.2.3. So C0 is a singleton which
n
necessarily is a fixed point of T as T C0 ⊆ C0 .
Remark 9.2.6 Lin [12] has generalized Goebel-Karlovitz lemma using certain non-
standard analytic considerations. Let X be a Banach space with norm || · ||, ∞ (X ),
the space of sequences (xn ) in X with norm sup{||xn || : n ∈ N} and C0 (X ) the
subspace of ∞ (X ) with null sequence (xn ) of X and [X ] the quotient space
∞ (X )/C0 (X ) with the norm ||[z n ]|| = lim sup ||z n || where [z n ] is the equivalent
class of {z n } ∈ ∞ . x ∈ X is identified with the class [x, x, . . .] and consequently X
can be considered a subset of [X ]. For a subset K of X the set [K ] = {[z n ] ∈ [X ] :
z n ∈ K for every n ∈ N}. If T : K → K is a map then [T ] : [K ] → [K ] is a map
defined in a natural way by [T ]([xn ]) = [T xn ].
Lemma 9.2.7 (Lin [12]) Let X be a Banach space and K a minimal weakly compact
convex subset of X , invariant under T . If [W ] is a non-empty closed convex subset
of [K ] which is invariant under [T ] then
sup{||[wn ] − x|| : [wn ] ∈ [W ]} = diam(K )

9.2 Coefficients of Banach Spaces and Fixed Points 203
for each x ∈ K .
We leave the details of the proof to the reader. The following theorem due to
Dominguez Benavides makes use of Lin’s lemma.
Theorem 9.2.8 (Dominguez Benavides [3]) If X is a Banach space for which
R(a, X ) < 1 + a for some a ≥ 0, then every non-empty bounded convex weakly
compact subset of X has fixed point property for non-expansive mappings.
Proof Suppose the theorem is false. Then, there is a non-empty weakly compact
convex subset K of X with diam(K ) = 1 and K is minimal invariant for a non-
expansive map T without a fixed point. Further there is a weakly null sequence of
approximate fixed points {xn } of T in X . Define
[W ] := {[z n ] ∈ [K ] : ||[z n ] − [xn ]|| ≤ 1 − t and lim lim||z n − z m || ≤ t}

n m
where t = 1+a 1
.
Clearly [W ] is a a closed convex [T ] invariant set. [W ] is non-empty as it contains
[t xn ]. So by Lin’s Lemma 9.2.7 it follows that
sup{||[wn ] − x|| : [wn ] ∈ [W ]} = 1
for each x ∈ [K ]. For [z n ] ∈ [W ] choose a weakly convergent subsequence {yn } of

{z n } such that limn ||z n || = lim ||yn || and lim ||vm || exists. Now lim ||yn −
n,m;n=m n,m,n=m
ym || = lim lim||yn − ym || ≤ lim lim||z n − z m || ≤ t. Let y be the weak limit of {yn }.
n m n m
For each n ∈ N, ||yn − y|| ≤ lim inf ||yn − ym || ≤ lim sup ||yn − ym ||. So
m m
lim||yn − y|| ≤ lim lim||yn − ym || ≤ t.

n n m
)

We can choose η > 0 such that η R(a, X ) < 1 − R(a,X
1+a
(i.e. R(a, X ) η + 1+a 1
<
n, we have ||yn − y|| ≤ t+ η. Also ||y|| ≤ lim||yn − xn || ≤ 1 −
1. For a large
yn yn −y y
t. So t+η = t+η + t+η ≤ R 1−tt
, X = R(a, X ). So lim||z n || = lim ||yn || ≤
R(a, X )(t + η) < 1, a contradiction to Lemma 9.2.7.
At this stage, we can introduce the concept of Banach–Mazur distance between
isomorphic Banach spaces, a useful concept in fixed point theory of non-expansive
maps in Banach spaces.
Definition 9.2.9 For isomorphic normed spaces X and Y , the Banach–Mazur dis-
tance between X and Y denoted by d(X, Y ) is defined as
d(X, Y ) = inf{||T ||||T −1 || : T : X → Y is an isomorphism}
Remark 9.2.10 For isomorphic normed spaces E, F, G, we have

(i) d(E, F) = d(F, E);

(ii) d(E, F) ≥ 1;
(iii) d(E, F) ≤ d(E, G)d(G, F)
(iv) Let c > 1 and B E and B F be the closed unit balls in E and F. Then, d(E, F) < c
if and only if there exist c1 , c2 > 0 with c1 c2 < c and an invertible T ∈ L(E, F)
such that c11 B(F) ⊆ T B(E) ⊆ c2 B(E) holds.
We skip the proof of
Theorem 9.2.11 (Dominguez Benavides [3]) For isomorphic Banach spaces X and
Y
R(a, Y ) ≤ d(X, Y )R(a, X )
where a ≥ 0.
For a Banach space X , the coefficient M(X ) can be defined as in the following

Definition 9.2.12 If X is a Banach space, M(X ) is defined as sup R(a,X
1+a
)
:a≥0
Theorems 9.2.8 and 9.2.11 lead to
Theorem 9.2.13 (Dominguez Benavides [3]) If X is a Banach space with

M(X ) > 1 then every non-empty convex weakly compact subset of X has fixed point
property for non-expansive mappings. If Y is another Banach space isomorphic to
X and d(X, Y ) < M(X ), then every non-empty convex weakly compact subset of Y
has fixed point property for non-expansive mappings.
Proof Since M(X ) > 1, for some a > 0 R(a,X

1+a
)
> 1. So by Theorem 9.2.8 the first
part of theorem follows. Since M(X ) > 1, for some a > 0, R(a,X
1+a
)
> 1. Since by
Theorem 9.2.11, R(a, Y ) ≤ d(X, Y )R(a, X ) < M(X )R(a, X ) < R(a,X1+a
)
R(a, X ) =
1 + a. So by Theorem 9.2.8, the fixed point property for non-expansive maps of a
non-void convex weakly compact subset of Y follows.
9.3 Nearly Uniformly Smooth Spaces
Definition 9.3.1 Let X be a Banach space with a Schauder basis (en ) (that is each
∞
x ∈ X has a unique representation x = xn en , xn being scalars). X is called nearly
n=1
uniformly smooth (NUS) if for each > 0 there is δ > 0 such that if 0 < t < δ and
(xn ) is a basic sequence in B X there exists k > 1 so that ||x1 + t xk || < 1 + t.
X is called weakly near uniformly smooth (WNUS) if the above definition holds
for some > 0.
Garcia Falset [4] proved the following
9.3 Nearly Uniformly Smooth Spaces 205
Theorem 9.3.2 Let X be a Banach space. The following are equivalent:

(a) there exists > 0 and δ > 0 such for all t ∈ [0, δ) and every weakly null
sequence (xn ) in B X there is k > 1 with ||x1 + t xk || < 1 + t;
(b) there exists c ∈ (0, 1) such that for each weakly null sequence (xn ) in B X there
is k > 1 with ||x1 + xk || ≤ 2 − c;
(c) R(X ) < 2.
Proof (b) ⇒ (c). Let (xn ) be a weakly null sequence in B X and x ∈ B X . Define (yn )
by y1 = x, yn+1 = xn for n ≥ 1. Then (yn ) is a weakly null sequence in B X . By (b),
there exists c ∈ (0, 1) and k1 > 1 with ||x + xk1 || ≤ 2 − c. Define another weakly
null sequence z n ∈ B defined by z 1 = x and z n = xk1 +n , n ∈ N. By (b) there exists
k2 > k1 such that ||x + xk2 || ≤ 2 − c. Thus, proceeding recursively, we can get a
subsequence xkn of weakly null sequence in B X such that ||x + xkn || ≤ 2 − c for all
kn and kn > 1 for all n. So lim||x + xn || ≤ 2 − c, for any weakly null sequence (xn )
in B X . So R(X ) < 2.
(c) ⇒ (b). Let (xn ) be a weakly null sequence in B X . As R(X ) < 2, R(X ) <
2 − c for some c ∈ (0, 1). So lim||x1 + xn || ≤ R(X ) < 2 − c. So for some k > 1,
||x1 + xk || ≤ 2 − c.
(a) ⇒ (b). By assumption, there exist , δ > 0 such that for all t ∈ (0, δ) and
any weakly null sequence (xn ) in B X there is k > 1 with ||x1 + xk || < 1 + t. Let
μ = min{1, δ}. Then for t < δ, ||x1 + xk || ≤ ||x1 + t xk || + (1 − t)||xk || ≤ 1 + t +
1 − t = 2 − t (1 − ) = 2 − c, where c = t (1 + ) ∈ (0, 1).
(b) ⇒ (a). By hypothesis, there exists c ∈ (0, 1) such that for each weakly null
sequence (xn ) in B X , there is k > 1 with ||x1 + xk || ≤ 2 − c. So for all t ∈ (0, 1),
||x1 + t xk || ≤ t||x1 + xk || + (1 − t)||x1 ||

≤ t (2 − c) + 1 − t + 1 + t (1 − c).
Corollary 9.3.3 ([4]) For a Banach space X , the following are equivalent:
(a) X is WNUS;
(b) X is reflexive and R(X ) < 2.
Theorem 9.3.4 (Garcia Falset [4]) Let X be a Banach space such that R(X ) < 2.
Then, every non-empty weakly compact convex subset of X has fixed point property
for non-expansive mappings.
Proof We prove by the method of contradiction. Suppose the theorem is false. Then,
there is a weakly compact convex subset K of X with diam K = 1 which is minimal
for a non-expansive map T : K → K in the sense of Goebel-Karlovitz Lemma 9.2.3.
Let (xn ) be a weakly null sequence of approximate fixed points of T in K .
With the usual notation in Remark 9.2.6, define the subset [W ] of [X ] by [W ] :=
{[z n ] ∈ [K ] : ||[z n ] − [xn ]|| ≤ 21 , D([z n ]) ≤ 21 }. [W ] is seen to be a T -invariant

closed convex subset [X ]. By Lemma 9.2.3 x2n ∈ [W ]. So by Lin’s Lemma 9.2.7
sup{||[wn ] − x|| : [wn ] ∈ [W ]} = 1 for all x ∈ K . Let [z n ] ∈ [W ]. ||[z n ]|| =

lim ||z n || = lim ||z n k || for some subsequence (z n k ) of (z n ), in K . As K is weakly
n→∞ k→∞
compact let (z n k ) converge weakly to y ∈ K without loss of generality. Then, passing
to subsequences and using a diagonal argument, one can assume that for m ∈ N

1 − 1 ||z n − y|| ≤ lim ||z n − y||
m m
m→∞
m
z n k −y
Let ym = 1 − 1
m
. As (ym ) is a weakly null sequence in B X , it
max ||y||, lim ||z n m −y||
m→∞
follows from the definition of R(X ) that

y
R(X ) ≥ lim ym +

m→∞
max lim ||z n m − y||, ||y||
m→∞

So lim ||z n m || ≤ R(X ) max ||y||, lim ||z n m − y|| . Since (z n m − xn m ) converges
m→∞ m→∞
1
weakly to y, ||y|| ≤ lim ||z n m − xn m || ≤ lim ||z n − xn || = ||[z n ] − [xn ]|| ≤
. On
m→∞ m→∞ 2
the other hand, the weak limit of {z n m − y − (z n s − y)} = z n m − y as s → ∞.
So ||z n m − y|| ≤ lim ||z n m − y − (z n s − y)|| or lim ||z n m − y|| ≤ lim lim ||z n m
m→∞ m→∞ m→∞ s→∞
−y − (z n s − y)|| . As D(z n m − y) = D(z n m ) ≤ D(z n ) and D([z n ]) ≤ 21 , we have
1 R(X )
lim ||z n − y|| ≤ D([z n ]) ≤ . So ||[z n ] − 0|| = lim ||z n m || ≤ < 1, as
m→∞ m 2 m→∞ 2
1
max lim ||z n m − y||, ||y|| ≤ . This contradicts Lemma 9.2.7.
m→∞ 2
The following are proved using similar arguments.
Theorem 9.3.5 (Garcia Falset [4]) Let X and Y be isomorphic Banach spaces such
that d(X, Y )R(Y ) < 2. Then, every non-empty convex weakly compact subset of X
has the fixed point property for non-expansive maps, provided it satisfies weak Opial’s
condition. That is, lim inf ||xn || ≤ lim inf ||xn + x|| for each x ∈ X and each weakly
null sequence (xn ).
Proof Suppose false. Then, there is a convex weakly compact subset K of X
with diam(K ) = 1, which is minimal for a non-expansive map T : K → K by
Goebel-Karlovitz lemma. Thus, there is a weakly null sequence of almost fixed
points (xn ) for T in K . As in Remark 9.2.6 consider the subset [W ] of [X ] defined
by
1 1
[W ] = {[z n ] ∈ [K ] : ||[z n ] − [xn ]|| ≤ and for some x ∈ K , ||[z n ] − x|| ≤ }
2 2
9.3 Nearly Uniformly Smooth Spaces 207
It can be verified that [W ] is T -invariant and a closed convex subset of [K ]. Also

[W ] is non-empty as [ x2n ] ∈ [W ] as ||[xn ] − 0|| ≤ 1. Thus, by Lin’s Lemma 9.2.7
sup{||[wn ] − x|| : [wn ] ∈ [W ]} = 1 for all x ∈ K .
Let [z n ] ∈ W with ||[z n ]|| = lim||z n || = lim ||z n k || where ||z n k || is some subse-
n k→∞
quence of (z n ). Since K is weakly compact in X , we can assume that (z n k ) weakly con-
verges to y ∈ K . Let T be an isomorphism of X onto Y with ||T ||||T −1 ||. R(Y ) < 2.
Clearly T (z n k ) converges weakly to T (y). Passing to subsequences we can suppose
that for each k ∈ N

1
1− ||T (z n k ) − T (y)|| ≤ lim ||T z n k − T y||.
k m→∞
−1

So for all k ≥ 1, 1 − k max lim||T (z n k ) − T (y)||, ||T y||
1
(T z n k − T y) and
k
−1
max lim||T (z n k ) − T (y)||, ||T y|| (T y) ∈ B y . So, from the definition of R(Y )
k
we get lim||T z n k || = R(Y ) max lim||T (z n k ) − T (y)||, ||T y|| .
k k
On the other hand since [z n ] ∈ [W ], there exists x0 ∈ K such that ||[z n ] − x0 || ≤ 21
and ||[z n ] − [xn ]|| ≤ 21 . So ||y|| ≤ lim||z n k − xn k || ≤ ||[z n ] − [xn ]|| ≤ 21 . As Y sat-
k
isfies weak Opial’s condition lim||T z n k − T y|| ≤ lim||T z n k − T x0 ||.
k k
So

lim||T z n k || ≤ R(Y )||T || max lim||z n k − x0 ||, ||y||
k k
||T ||
≤ R(Y ) .
2
The above conditions imply that
||[z n ] − 0|| = lim ||z n k || = lim ||T −1 T z n k ||

k k
≤ ||T −1 || lim ||T z n k ||
k
R(Y )
≤ ||T −1 ||||T || < 1.
2
This contradicts Lin’s Lemma 9.2.7.
Corollary 9.3.6 (Garcia Falset [4]) If X is a weakly nearly uniformly smooth

(WNUS) Banach space, then every non-void convex weakly compact subset of X
has fixed point property for non-expansive mappings. In particular, a nearly uni-
formly smooth (NUS) Banach space X also enjoys this property.
Proof As X is WNUS, X satisfies weak Opial condition. As d(X, X ) = 1 and

R(X ) < 2 by Theorem 9.3.5 every non-void convex weakly compact subset of X
has fixed point property for non-expansive maps. Since an NUS Banach space X
has WNUS, every convex weakly compact subset of X has fixed point property for
non-expansive mappings.
9.4 Non-square Banach Spaces
Dominguez Benavides [2] introduced the modulus of nearly uniform smoothness as

follows:
Definition 9.4.1 ([2]) The modulus of nearly uniform smoothness of a Banach space
X is the function X : [0, ∞) → R defined by

||x1 + t xn || + ||x1 − t xn ||
X (t) = sup inf − 1 : n > 1 : (xn )
2

is a basic sequence in B X . i.e., (xn ) is a Schauder basis for X.

Remark 9.4.2 ρ X (t) = sup ||x+t y||+||x−t
2
y||
: x, y ∈ B X is called the modulus of
uniform smoothness. Clearly ρ X (t) ≥ X (t) ≥ 0 for all t ≥ 0. When X is uniformly
ρ X (t) X (t)
smooth, ρ (0) = lim = 0 by definition. In this case lim = X (0) = 0.
t→0 t t→0 t
There is an equivalent characterization of near uniform smoothness for reflexive
Banach spaces, whose proof is available in [2].
Proposition 9.4.3 ([2]) Let X be a reflexive Banach space. Then

||x1 + t xn || + ||x1 − t xn ||
X (t) = sup inf − 1 : n > 1 : (xn )
2

is weakly null in B X
As a result we have
Proposition 9.4.4 ([2]) A Banach space X is nearly uniformly smooth if and only
X (t)
if X is reflexive and lim = 0.
t→0 t
X (t)
Proof If lim = 0, then for > 0 there exists δ > 0 such that X (t) ≤ t for
t→0 t
t ∈ [0, δ). Since X is reflexive, we can find a basic sequence {xn } in B X as suggested
by Prus in §2 of [14] which is not norm convergent for which
9.4 Non-square Banach Spaces 209
1
||x1 + t xn k || ≤ {(1 + c)||x1 − 2t xn k || + ||x1 + 2t xn k ||}
2
(1+3t)
where c > 1, 1 + c < (1+2t) and (xn k ) is a subsequence of {xn }. Then for some k
||x1 + t xn k || ≤ (1 + )(1 + 2t) < 1 + 3t.
Conversely if X is nearly uniformly smooth then X is reflexive. Consider a weakly
null sequence (xn ). For each > 0, there is a δ > 0 with ||x1 + t z n || ≤ 1 + t for all
n > 1, where z n is a subsequence of (xn ) with z 1 = x1 . Since {x1 , −z 2 , −z 3 , . . . , }
is also weakly null ||x1 − t z n || ≤ 1 + t for some 0 < δ < δ for all t ∈ [0, δ). So
for 0 < t < δ
||x1 + t z n || + ||x1 − t z n ||
− 1 ≤ t
2
X (t)
So lim = 0.
t→0 t
For the proof that a non-square Banach space has the fixed point property for non-
expansive mappings on non-empty bounded closed convex subsets, Garcia-Falset et
al. [5] introduced the coefficient RW (X ) and M W (X ) relating them to the coeffi-
cients R(X ) and W (X ).
Definition 9.4.5 ([5]) Let X be a Banach space and a, a positive real number. Then

w
RW (a, X ) := sup min lim ||xn + x||, lim ||xn − x|| : xn ∈ B X , xn → 0, ||x|| ≤ a
n→∞ n→∞

1+a
M W (X ) := sup :a>0 .
RW (a, X )
Remark 9.4.6 For any Banach space X and a > 0, max{a, 1} ≤ RW (a, X ) ≤ 1 +
a. So 1 ≤ M W (X ) ≤ 2.
Lemma 9.4.7 ([5]) Let X be a Banach space. Given x ∈ X and a bounded sequence
(xn ) in X , there is a subsequence (xn k ) of (xn ) such that
lim ||(xn k − xn k+1 ) + x|| ≥ lim lim ||(xn − xm ) + x||

n→∞ n→∞ m→∞
lim ||(xn k − xn k+1 ) − x|| ≥ lim lim ||(xn − xm ) − x||
n→∞ n→∞ m→∞
and
lim ||(xn k − xn k+1 )|| ≤ D[(xn )]
k→∞
Proof Let
a := lim lim ||(xn − xm ) + x||

n→∞ m→∞
b := lim lim ||(xn − xm ) − x||
n→∞ m→∞
By definitions of a, b and D[(xn )], we can find n 1 ≥ 1 such that
1
lim ||(xn 1 − xm ) + x|| > a −
m→∞ 2
1
lim ||(xn 1 − xm ) − x|| > b −
m→∞ 2
and
1
lim ||(xn 1 − xm )|| < D[(xn )] + .
m→∞ 2
Suppose n 1 < n 2 < · · · < n j have been defined such that for each k = 1, 2, . . . , j,
1
lim ||(xn k − xm ) + x|| > a −
m→∞ k+1
1
lim ||(xn k − xm ) − x|| > b −
m→∞ k+1
1
lim ||xn k − xm || < D[(xn )] +
m→∞ k+1
and for each k = 1, 2, . . . , j − 1
1
||(xn k − xn k+1 ) + x|| > a −
k+1
1
||(xn k − xn k+1 ) − x|| > b −
k+1
1
||xn k − xn k+1 || < D[(xn )] + .
k+1
From the above inequalities for k = j and the definition of a, b and D[(xn )], we can
find n j+1 > n j such that
1
||(xn j − xn j+1 ) + x|| > a −
j +1
1
||(xn j − xn j+1 ) − x|| > b −
j +1
1
||xn j − xn j+1 || < D[(xn )] +
j +1
1
lim ||(xn j+1 − xm ) + xm || > a −
m→∞ j +2
1
lim ||(xn j+1 − xm ) − xm || > b −
m→∞ j +2
1
lim ||xn j+1 − xm || < D[(xn )] + .
m→∞ j +2
So by induction, there exists an increasing sequence of natural numbers (n k ) k ≥ 1

such that n k < n k+1 for all k with
lim ||(xn k − xn k+1 ) + x|| ≥ a,

k→∞
lim ||(xn k − xn k+1 ) − x|| ≥ b
k→∞
lim ||xn k − xn k+1 || ≤ D[(xn )]

k→∞
The next theorem makes use of the above lemma.
Theorem 9.4.8 ([5]) Let X be a Banach space for which B X ∗ is w ∗ -sequentially

compact. For each a > 0, R(a, X ) ≤ RW (a, X ). Hence M(X ) ≥ M W (X ).
Proof Let a, η > 0. From the definition of R(a, X ), we can find x ∈ X with ||x|| ≤ a
and a weakly null sequence (xn ) in B X with D[(xn )] ≤ 1 such that
lim ||xn + x|| ≥ R(a, X ) − η

n→∞
For each n ≥ 1, choose f n ∈ S X ∗ such that
f n (xn + x) = ||xn + x||
As B X ∗ is sequentially compact in w ∗ -topology we can without loss of generality

assume that ( f n ) converges in w ∗ topology to some f ∈ B X ∗ . As (xm ) converges
weakly to zero, from the weak lower semicontinuity of the norm, lim ||(xn + xm ) +
n→∞
x|| ≥ ||xn + x|| for all n ≥ 1. So
lim lim ||(xn − xm )|| ≥ lim ||xn + x|| ≥ R(a, X ) − η

n→∞ m→∞ n→∞
w∗
Again for n ≥ 1, as f m → f , lim f m (xn ) = f (xn ), we have
m→∞
lim ||(xn − xm ) − x|| ≥ lim − f m ((xn − xm ) − x)

m→∞ m→∞
= lim f m (xm + x) − f (xn )
m→∞
= lim ||xm + x|| − f (xn )
m→∞
≥ R(a, X ) − η − f (xn ).
So
lim lim ||(xn − xm ) − x|| ≥ R(a, X ) − η,
n→∞ m→∞
W
as xn → 0.
Thus we have shown that
min{ lim lim ||(xn − xm ) + x||, lim lim ||(xn − xm ) − x|| ≥ R(a, X ) − η
n→∞ m→∞ n→∞ m→∞
Further, D[(xn )] ≤ 1. So by Lemma 9.4.7, we can find a subsequence xn k of xn with
min{ lim ||(xn k − xn k+1 ) + x||, lim ||(xn k − xn k+1 ) − x|| ≥ R(a, X ) − η
k→∞ k→∞
and lim ||xn k − xn k+1 || ≤ 1. So we can find k0 such that for all k ≥ k0 , ||xn k −
k→∞
xn k+1 || ≤ 1 + η. Define
xn k0 +k − xn k0 +k+1
yk := , k≥1
1+η
x
y :=
1+η
(yk ) converges weakly to zero in B X and ||y|| ≤ a. So by the definition of RW (a, X )

RW (a, X ) ≥ min lim ||yk + y||, lim ||yk − y||
k→∞ k→∞

1
= lim ||(xn k − xn k+1 ) + x||, lim ||(xn k − xn k+1 ) − x||
1 + η k→∞ k→∞
R(a, X ) − η
≥
1+η
Allowing η to tend to zero we get RW (a, X ) ≥ R(a, X ). So M(X ) ≥ M W (X ).

Corollary 9.4.9 If X is a Banach space with M W (X ) > 1, then every non-expansive
self-map on a convex weakly compact subset of X has a fixed point.
Proof Clearly by Theorem 9.4.8 M(X ) ≥ M W (X ) > 1. We can without loss of
generality assume that X is separable and B X ∗ is w ∗ sequentially compact. Since
M(x) > 1 by Theorems 9.2.13, 9.4.8 implies that on non-void weak compact con-
vex subsets, non-expansive self-maps have fixed points.
9.5 An Equivalent Definition of RW (a, X) and Fixed Points

of Non-expansive Maps in Non-square Banach Spaces
We begin with
Proposition 9.5.1 ([5]) Let X be a Banach space and a > 0. Then
RW (a, X ) = sup{ inf (||ax1 + xn || ∧ ||ax1 − xn ||) : (xn ) a weakly null sequence in B X }
n>1
Here, α ∧ β and α ∨ β represent, respectively, the minimum and maximum of two

numbers α and β.
Proof For a > 0, write

RW (a, X ) = sup{ inf (||ax1 + xn || ∧ ||ax1 − xn ||) : (xn ) a weakly null sequence in B X }.
n>1
Let η > 0, (xn ) a weakly null sequence in B X and x ∈ X with ||x|| ≤ a. Define (yn )
by
x/a, n = 1
yn =
xn , n ≥ 2
(yn ) is a weakly null sequence in B X . So

inf (||ay1 + yn || ∧ ||ay1 − yn ||) ≤ RW (a, X ).
n>1
So there exists n 1 > 1 such that

||x + xn 1 || ∧ ||x − xn 1 || = ||ay1 + yn || ∧ ||ay1 − yn || < RW (a, X ) + η.
Suppose n 1 < n 2 . . . < n k have been found such that for j ∈ {1, 2, . . . , k}

||x + xn j || + ||x − xn j || < RW (a, X ) + η
Define (z n ) by z 1 = ax , z n = xn+n k , n ≥ 2 (z n ) is a weakly null sequence in B X with

inf (||az 1 + z n || ∧ ||az 1 − z n ||) ≤ RW (a, X )
n>1
so that there is k > 1 such that


||az 1 + z k || ∧ ||az 1 − z k || < RW (a, X ) + η.
Define n k+1 = k + n k . We get

||x + xn k+1 || ∧ ||x − xn k+1 || = ||az 1 + z k || ∧ ||az 1 − z k || < RW (a, X ) + η.
So by induction, there exists a subsequence (xn k ) with

||x + xn k+1 || ∧ ||x − xn k+1 || < RW (a, X ) + η
We can get a subsequence (wm ) of (xn k ) for which lim ||x + wm ||, lim ||x − wm ||
m→∞ m→∞
exist.
Now
lim ||x + xn || ∧ lim ||x − xn || ≤ lim ||x + wm || ∧ lim ||x − wm ||

n→∞ n→∞ m→∞ m→∞
= lim (||x + wm || ∧ ||x − wm ||)
m→∞

≤ RW (a, X ) + η
As η > 0 is arbitrary we get

RW (a, X ) ≤ RW (a, X ).
For η > 0 and (xn ) a weakly null sequence in B X , suppose
lim (||ax1 + xn ||) ∧ lim (||ax1 − xn ||) = lim ||ax1 + xn ||

n→∞ n→∞ n→∞
So by definition of RW (a, X ),
lim ||ax1 + xn || ≤ RW (a, X )

n→∞
Thus there exists k > 1 such that
||ax1 + xn || < RW (a, X ) + η
Hence inf (||ax1 + xn || ∧ ||ax1 − xn ||) < RW (a, X ) + η.

n>1
(a,
Soin this case RW X )≤ RW (a, X ), as η > 0.
If lim ||ax1 + xn || ∧ lim ||ax1 − xn || = lim ||ax1 − xn || by a similar argu-
n→∞ n→∞ n→∞

ment RW (a, X ) ≤ RW (a, X ).
9.5 An Equivalent Definition of RW (a, X ) and Fixed … 215
Theorem 9.5.2 ([5]) Let X be a reflexive Banach space. Then

t (RW ( 1t , X ) + 1) − 1
(i) for all t > 0, X (t) ≤ ;
2
1
(ii) for all a > 0, RW (a, X ) ≤ a(1 + X ( )) and in particular M W (X ) ≥
a
1+t
sup{ 1+ X (t)
: t > 0}
Proof (i) Let t, η > 0 and (xn ) be a weakly null sequence in B X . By Proposition 9.5.1
1 1
inf (||x1 + t xn || ∧ ||x1 − t xn ||) = t inf {|| x1 + xn || ∧ || x1 − xn ||}
n>1 n>1 t t
1
≤ t RW ( , X ).
t
So for some k > 1
1
||x1 + t xk || ∧ ||x1 − t xk || < t RW ( , X ) + η
t
As ||x1 + t xk || ∨ ||x1 − t xk || ≤ 1 + t, we have
||x1 + t xk || + ||x1 − t xk || = (||x1 + t xk || ∧ ||x1 − t xk ||) + ||x1 + t xk || ∨ ||x1 − t xk ||

1
< t RW ( , X ) + η + 1 + t
t
So

||x1 + t xn || + ||x1 − t xn || ||x1 + t xk || + ||x1 − t xk || − 1
inf −1 ≤
n>1 2 2
t (RW ( 1t , X ) + 1) − 1 + η
< .
2
t (RW ( 1 ,X )+1)−1+η t (RW ( 1 ,X )+1)−1
So X (t) < t
2
. As η → 0, X (t) ≤ t
2
.
(ii) Let a, η > 0 and (xn ) be a weakly null sequence in B X . So

||x1 + xn
|| + ||x1 − xn
|| 1
inf a a
− 1 ≤ X ( )
n>1 2 a
So there exists k > 1 with

xk xk 1
||x1 + || + ||x1 − || < 2(1 + X ( ) + η)
a a a
So
inf ||ax1 + xn || ∧ ||ax1 − xn || ≤ ||ax1 + xk || ∧ ||ax1 − xk ||

n>1
1
≤ (||ax1 + xk || ∧ ||ax1 − xk ||)
2
a xk xk
= (||x1 + || ∧ ||x1 − ||)
2 a a
1
< a(1 + X ( ) + η)
a
Since η > 0 is arbitrary, we get
1
RW (a, X ) ≤ a(1 + X ( )).
a

Theorem 9.5.3 ([5]) If X is a reflexive Banach space for which X (0) < 1, then
M(X ) > 1. So every non-void bounded closed convex subset of X has fixed point
property for non-expansive maps.
Proof Clearly X (0) < 1 if and only if for some t > 0 sup{ Xs(s) : 0 < s ≤ t} <
1. So for some t > 0 X (t) < t. So by (ii) M W (X ) ≥ sup{ 1+ 1+t
X (t)
: t > 0}. Or
M W (X ) > 1. So by Corollary 9.4.9, and the reflexivity of X every non-void closed
bounded convex subset of X has fixed point property for non-expansive maps.
Corollary 9.5.4 ([5]) If X is a uniformly non-square Banach space, then M(X ) > 1
and so every non-void closed bounded convex subset of X has fixed point property
for non-expansive maps.
Proof Since X is uniformly non-square, so is X ∗ . By Proposition 9.1.6, 0 (X ∗ ) < 2.
So by Lindenstrauss formulae the modulus of smoothness ρ X satisfies ρ X (0) < 1
(again by Proposition 9.1.6). Since X (t) ≤ ρ X (t) for all t > 0, X (0) ≤ ρ X (0) < 1.
As X is reflexive (being non-square) by Theorem 9.5.3 every non-void bounded
closed convex subset of X has fixed point property for non-expansive mappings.
Remark 9.5.5 Garcia Falset et al. [5] have shown that in a reflexive Banach space
X , the following are equivalent:
(i) M W (X ) > 1;
(ii) for some a > 0, RW (a, X ) < 1 + a;
(iii) inf{ RW1+a
(a,X )
: a > 0} > 1;
(iv) X (0) < 1;
(v) for some t > 0, sup{ Xs(s) : 0 ≤ s ≤ t} < 1;
(vi) for some t > 0, X (t) < t.
Remark 9.5.6 If M W (X ) > 1, X may not be non-square although every non-void
closed bounded convex subset of X has fixed point property for non-expansive maps.
This is seen by the following example due to Garcia Falset et al. [5].
9.5 An Equivalent Definition of RW (a, X ) and Fixed … 217
Example 9.5.7 Let X = (R2 , || ||∞ ) and Y = (2 , || · ||2 ) and Z the product space
Z = X × Y with the norm ||(x; y)|| = max{||x||∞ , ||y||2 }, where x ∈ X = R2 and
y ∈ Y = 2 . For z 1 = ((1, 1); 0) and z 2 = (1, −1); 0), ||z 1 || = ||z 2 || = 1, ||z 1 +
z 2 || = ||z 1 − z 2 || = 2. Thus Z is not uniformly non-square. Let z n = (xn ; yn ) be a
W
weakly null sequence in Bz and z ∈ B Z , with z = (x; y). If z n → 0, then xn → 0 and
yn → 0 and lim ||xn + x||∞ = ||x||∞ ≤ ||z|| ≤ 1 and lim ||yn + y||22 =
n→∞
lim ||yn ||22
+ ||y|| ≤ lim ||z n || + ||z|| ≤ 2. Therefore, lim ||z n + z|| = lim
2 2 2
n→∞ n→∞ n→∞ n→∞
√ √
max{||xn + x||, ||yn + y||} ≤ 2. So RW (1, X ) ≤ 2 and M W (X ) > 1. So in X
every non-void closed convex bounded subset has fixed point property for non-
expansive maps.
References
1. Day, M.M.: Normed Linear Spaces. Springer, Berlin (1973)

2. Dominguez Benavides, T.: Modulus of nearly uniform smoothness and Lindenstrauss formulae.
Glasg. Math. J. 37, 143–153 (1995)
3. Dominguez Benavides, T.: A geometrical coefficient implying the fixed point property and
stability results. Houst. J. Math. 22, 835–849 (1996)
4. Garcia-Falset, J.: The fixed point property in Banach spaces with the NUS-property. J. Math.
Anal. Appl. 215, 532–542 (1997)
5. Garcia-Falset, J., Llorens-Fuster, E., Mazcunan-Navrro, E.M.: Uniformly non-square Banach
spaces have the fixed point property for nonexpansive mappings. J. Math. Anal. Appl. 233,
494–514 (2006)
6. Goebel, K.: On the structure of minimal invariant sets for nonexpansive mappings. Ann. Univ.
Marie Curie-Sklowdowska Set. A 29, 73–75 (1975)
7. James, R.C.: Uniformly non-square Banach spaces. Ann. Math. 2nd Ser. 80, 542–550 (1964)
8. James, R.C.: Super-reflexive Banach spaces. Can. J. Math. 24, 896–904 (1972)
9. Karlovitz, L.A.: Existence of fixed points for nonexpansive map in space without normal
structure. Pac. J. Math. 66, 153–159 (1976)
10. Kato, M., Maligranda, L., Takahashi, Y.: On James and Jordan-von Neumann constants and
the normal structure coefficient of Banach spaces. Stud. Math. 144, 275–295 (2001)
11. Kirk, W.A.: A fixed point theorem for mappings which do not increase distances. Am. Math.
Mon. 72, 1004–1006 (1965)
12. Lin, P.K.: Unconditional bases and fixed points of nonexpansive mappings. Pac. J. Math. 116,
69–76 (1985)
13. Lindenstrauss, J., Tzafiri, L.: Classical Banach Spaces I. Springer, Berlin (1977)
14. Prus, S.: Nearly uniformly smooth Banach spaces. Boll. U. M. I. 3(7), 507–522 (1989)
Chapter 10
Brouwer’s Fixed-Point Theorem
10.1 Introduction
It is more than a century since Brouwer [4] proved a fixed- point theorem of great
consequence, in the setting of finite-dimensional Euclidean spaces. It was subse-
quently extended to normed linear spaces by Schauder [25], and later to locally
convex linear topological spaces by Tychonoff [31]. Brouwer’s theorem was gen-
eralized to multifunctions first by Kakutani [12], and later to locally convex linear
topological spaces by Glicksberg [8] and Ky Fan [6]. Brouwer’s theorem admits
of several proofs. Notable among them are those based on Sperner’s lemma [28]
or concepts of homotopy/homology from algebraic topology (see Dugundji [5] or
Munkres [17]) or concepts and results from Real analysis (see Milnor [16], Seki [26],
Rogers [23], Kannai [13], Traynor [30]). However, we provide here only the analytic
proof of Brouwer’s theorem and a proof based on Sperner’s lemma. Needless to state
that Brouwer’s theorem and its generalizations/variants find a wide range of appli-
cations in the solution of nonlinear equations, differential and integral equations,
mathematical biology and mathematical economics.
10.2 Analytic Preliminaries
We collect in this section the basic theorems of analysis needed in the proof of
Brouwer’s theorem.
Theorem 10.2.1 (Weierstrass Approximation Theorem) If f is a continuous real-
valued function defined on a closed bounded subset S of Rn , then for any given
positive number , we can find a polynomial P of n variables x1 , . . . , xn such that
| f (x1 , . . . , xn ) − P (x1 , . . . , xn )| < for all (x1 , x2 , . . . , xn ) ∈ S.
Theorem 10.2.2 (Inverse Function Theorem) Let G be a non-empty open set in Rn
and f = ( f 1 , . . . , f n ) : G ⊆ Rn → Rn be a continuous mapping having continuous
https://doi.org/10.1007/978-981-13-3158-9_10
220 10 Brouwer’s Fixed-Point Theorem
first-order partial derivatives at all points

of G. If for some point P0 ∈ G, the determi-
∂ fi
nant of the Jacobian matrix J (x) = ∂x j (x) is non-zero at P0 , then f is a one-to-one
open map in a neighbourhood of P0 .
Theorem 10.2.3 (Change of variables for multiple integrals) Let g = (g1 , . . . , gn )

be a function defined on an open connected set G ⊆ Rn and taking values in Rn and
having continuous first-order partial derivatives at all points of G. Let g be one-to-
one on G and Det (Jg (x)) = 0 for all x ∈ G, where Jg is the Jacobian matrix of g
and f : g(G) → R be continuous. If X is a Jordan-measurable compact subregion
of G, then

f (x1 , . . . , xn )d x1 d x2 . . . d xn = f (g(t1 , . . . , tn ))|Det Jg (t1 . . . , tn )|dt1 . . . dtn
X g −1 (X )
Theorem 10.2.4 (Mean-value Inequality) If f : G ⊆ Rn → Rn is a continuous

function with continuous first-order partial derivatives which are bounded in the
open connected set G, then for some M > 0 and all x, y ∈ G,
f (x) − f (y) ≤ Mx − y.
For these and other aspects of calculus in finite-dimensional spaces, Apostol [1]
may be consulted.
10.3 Brouwer’s Fixed-Point Theorem
We state first Brouwer’s fixed-point theorem.
Theorem 10.3.1 (Brouwer) Every continuous function f mapping the closed unit
sphere Bn (of Rn ) into itself has a fixed point.
We deduce it from the no-retraction theorem via a lemma, following Rogers [23].
First, we recall the following.
Definition 10.3.2 Let B ⊆ A ⊆ Rn . A continuous map f : A → B is called a

retraction of A onto B if f (A) = B and f is identity on B. B is called a retract
of A if there is a retraction of A onto B.
Lemma 10.3.3 There is no continuously differentiable retraction of B n (the unit

ball in Rn ) onto S n−1 the unit sphere (in Rn ).
Proof Let if possible f : B n → S n−1 be such a retraction. Define g : B n → Rn by

g(x) = f (x) − x, x ∈ B n . Clearly g is continuous and has continuous first-order
∂gi
partial derivatives. Clearly ∂x (x) is continuous for all i, j = 1, 2, . . . , n, where
j
∂gi
g = (g1 , g2 , . . . , gn ) and all x ∈ B n . Since B n is compact ∂x j
(x) is bounded on
10.3 Brouwer’s Fixed-Point Theorem 221
∂gi
B n for all i, j = 1, 2, . . . , n, in view of the continuity of ∂x j
on B n . So by Theorem
10.2.4 for some k > 1,
g(x) − g(y) ≤ kx − y for all x, y ∈ B n .
Define f t : B n → Rn by f t (x) = x + tg(x) = (1 − t)x + t f (x), ∀ t ∈ [0, 1]. Since

f is a retraction of B n onto S n−1 , f t (x) = x for all x ∈ S n−1 for all t ∈ [0, 1]. Now
f t (x) = I + tg (x) for each x ∈ B n , f t (x) being the linear operator represented
by the Jacobian matrix of f t at x. So f t (x) = I + tg (x) ≥ I − tg (x)
for t ∈ [0, 1]. So for t < k1 and x ∈ B n , tg (x) < 1 so that f t is invertible by
Corollary 6.1.7. So by the inverse function Theorem 10.2.2, f t is an open one-to-
one map for t < k. For x ∈ B 0 = interior B n , f t (x) ≤ (1 − t)x + tx < 1,
when t < k1 . Thus, f t maps B 0 into B 0 when t < k1 and in this case, we claim that
.
G t = f t (B 0 ) = B 0 . Let if possible e ∈ B n − G t . Let g ∈ G t . Join e to g and choose
a point b on the line segment [e, g] meeting the boundary of G t . Since f t (B n ) is
compact, being the continuous image of a compact set, b = f t (x) for some x ∈ B n .
Since b, the boundary point of G t is not in the open set G t , x cannot be an interior point
of B n . So x ∈ S n−1 and f t (x) = x. Thus b = x and so e as well as b lies on S n−1 ,
the boundary of B n . Also f t (S n−1 ) = S n−1 , B 0 = G t and f t maps B n bijectively
onto B n .
Let C(A) denote the Jordan content of a subset A of Rn . Clearly Bn has Jordan
content and for t < k1
C(Bn ) = C( f t (Bn ))

= |Det f t (x)|d x by Theorem 10.2.3
B
n
1
= Det ( f t (x))d x for t < ,
Bn k
which is a polynomial in t. However, left-hand side of the above equality has the
constant value C(B n ), and so this polynomial is constant for all t < k1 . But for all t ∈

[0, 1]. It = B n Det ( f t (x))d x is a polynomial in t, which is C(B n ) for 0 < t < k1 .
So It is constant for all t ∈ [0, 1]. Now, f 1 . f 1 = f 1 (x)2 = x2 = 1 for x ∈ B n ,
as f 1 = f is the retraction of B n onto S n−1 . So ∂∂xf1i . f 1 = 0 for 1 ≤ i ≤ n on B n . This

system of linear equations has non-trivial solutions on B n only when Det ∂∂xf11 = 0
where f 1 = ( f 11 , f 12 , . . . , f 1n ). So I (1) = 0 and this contradicts that C(B n ) > 0.
Theorem 10.3.4 (Brouwer’s theorem for differentiable maps) If f : B n → B n is a

continuous map having continuous first-order partial derivatives, then f has a fixed
point in B n .
Proof Suppose f : B n → B n is a continuous function having continuous first-order

partial derivatives without a fixed point. So f (x) = x for all x ∈ B n . Define w :
B n → Rn by
(1 − x2 )
w(x) = x − f (x)
(1− < x, f (x) >)
for x ∈ B n .
Clearly < x, f (x) > < 1 for x ∈ Bn . Otherwise < x, f (x) > ≥ 1 would imply
that 1 ≤ < x, f (x) > ≤ x f (x) ≤ 1 by Cauchy–Schwarz inequality leading
to 1 = < x, f (x) > = x f (x). This would mean that x = c f (x) for some
c = 0 and 1 = < x, f (x) > = c < f (x), f (x) > = c f x2 = 1c x2 with c > 0.
Therefore c = 1 with x = f (x), contradicting that f has no fixed point. Thus
< x, f (x) > < 1 for x ∈ B n .
1−x2 1−x2
Suppose w(x) = 0. Then x = 1−<x, f (x)>
f (x) = c f (x), where c = 1−<x, f (x)>
.
| f (x)
2 2
Since x = c f (x), c = 1−|c1−c f (x)2
, c − c 2 f (x)2 = 1 − |c |2 f (x)2 . So c =
1, contradicting that f has no fixed point. So w(x) = 0 for all x ∈ B n .
w(x)
Define g : B n → Rn by g(x) = w(x) . Clearly g(B n ) ⊆ S n−1 and for x ∈ S n−1 ,
w(x) = x and so g(x) = x. Thus g is a continuously differentiable retraction of B n
onto S n−1 , contradicting Lemma 10.3.3.
Hence every continuously differentiable map of B n into itself has a fixed
point.
We are now in a position to prove Brouwer’s fixed-point theorem 10.3.1.
Proof of Brouwer’s Theorem 10.3.1 Let f : B n → B n be a continuous function.

By Weierstrass Approximation Theorem 10.2.1, for any m1 > 0, m ∈ N there exists
a polynomial Pm such that f (x) − Pm (x) < m1 for all x ∈ B n . So Pm (x) <
f (x) + m1 < 1 + m1 for all x ∈ B n . So by Theorem 10.3.4, (1+
Pm
1
)
mapping B n
m
Pm (xm )
into itself has a fixed point xm in B n . Now f (xm ) − xm ≤ f (xm ) − 1+ m1
≤
Pm (xm )
f (xm ) − Pm (xm ) + Pm (xm ) − 1+ m1
< for each m ∈ N.
2
m
,
Hence inf{x − f (x) : x ∈ B } = 0. As B is compact and x → x − f (x) is
n n
continuous, there exists x ∗ ∈ B n such that x ∗ − f x ∗ = 0. Thus, x ∗ = f (x ∗ ) and

f has a fixed point in B n .
As a corollary we can deduce the no-retraction theorem.
Corollary 10.3.5 There is no retraction of B n onto S n−1 .
Proof If g : B n → S n−1 is a retraction, then (−g) mapping continuously B n into B n

has a fixed point x0 by Brouwer’s theorem. So −g(x0 ) = x0 . Hence g(x0 ) = −x0 . But
g(x0 ) ∈ S n−1 and all the points of S n−1 are invariant under g. So x0 = g(x0 ) = −x0
contradicts that x0 is in S n−1 . Hence there is no retraction of B n onto S n−1 .
Remark 10.3.6 From the no-retraction theorem we can deduce Brouwer’s theorem.
The next theorem points out that all compact convex subsets of Rn have the fixed-
point property for continuous functions and is a precursor to Schauder’s fixed-point
theorem.
10.3 Brouwer’s Fixed-Point Theorem 223
Theorem 10.3.7 If K is a non-empty compact convex subset of Rn , then every con-

tinuous mapping of K into itself has a fixed point.
Proof Let f be a continuous map of K into K and B, closed ball containing K . For
each x ∈ B, let N (x) be the unique point in K nearest to x. (Note that if y1 and y2 are
two points in K equidistant from K , then x1 +x
2
2
∈ K is nearer to x!). Thus x → N (x),
defines a map from B into K . We now note that this map is continuous. Suppose
x ∈ B, xn (∈ B) converges to x and that N (xn ) does not converge to N (x). Since
N (xn ) ∈ K and K is compact we can find (xn(k) ) ∈ K a subsequence of (xn ) such that
N (xn(k) ) converges to y(= N (x)) ∈ K . Now xn(k) − N (xn(k) ) ≤ xn(k) − N (x).
Allowing k to tend to ∞ in the above inequality, we get x − y ≤ x − N (x),
contradicting that y = N (x). Hence x → N (x) is continuous on B.
Now, by Brouwer’s theorem, x → f (N (x)) maps B into itself continuously and
so has a fixed point, x0 say. Since f (N (x)) ∈ K for all x ∈ B and f (N (x0 )) = x0 ,
x0 ∈ K . If x0 ∈ K , N (x0 ) = x0 . Thus f (x0 ) = x0 and f has a fixed point in K .
Remark 10.3.8 It follows that closed bounded intervals such as [a, b]n in Rn have
the fixed-point property.
10.4 A Proof of Brouwer’s Theorem from Sperner’s

Lemma
Sperner [28] proved a combinatorial theorem wherefrom Brouwer’s fixed-point the-

orem can be deduced. In this section Sperner’s lemma is detailed.
Definition 10.4.1 An n-dimensional simplex S in a linear space is a convex lin-
ear combination of n + 1 points in general position. This is for given vertices
n+1

n
v1 , . . . , vn+1 , the simplex S = αi vi : αi ≥ 0 and αi = 1 and the set of
i=1 i=1
vertices {v1 , . . . , vn+1 } is not contained in any (n − 1) dimensional hyperplane of
the linear space. It is written [v1 , . . . , vn+1 ].
Each vi is called a vertex of S and each k-simplex [vi0 , . . . , vik ] is called a face of
S. Thus each vertex is a face as also the whole simplex S.
Definition 10.4.2 Let S be the n-dimensional simplex with the n + 1 vertices
v1 , . . . , vn+1 that are in general position. For y ∈ S, define ψ(y) = i ∈ N : y =

n+1
λi vi ∈ S and λi ≥ 0 . If ψ(y) = {i 0 , . . . , i k } then y is in the face [xi0 , . . . , xik ].
i=1
This face is called the carrier of y.

n+1
For the n-dimensional simplex S described above, if y ∈ S, then y = αi vi
i=0
and in this representation αi are uniquely defined. (α1 , . . . , αn+1 ) are called the
barycentric coordinates of y. Thus the carrier of y is well defined.
Remark 10.4.3 The standard n-simplex is y = (y1 , . . . , yn+1 ) ∈ Rn+1 : yi ≥ 0,

n+1
yi = 1 . It is also conveniently written as n = [e1 , . . . , en+1 ].
i=1
Definition 10.4.4 A simplicial subdivision of an n-dimensional simplex S is a

partition of S into a finite collection {Si : i = 1, 2, . . . , k} of simplexes such that
k
Si = S and Si ∩ S j is either empty or a common face.
i=1
Example 10.4.5
3
1 5 2
In the above 2-simplex [1, 2, 3], the subdivision {[1,3,5], [2,3,4], [2,4,5], [1,3],
[1,5], [2,3], [2,4], [2,5], [3,4], [4,5], 1, 2, 3, [4], [5]} is not simplicial for the inter-
section of the faces [1,3,5] with [2,3,4] is [3,4] and is not a common face. On the
other hand in the following figure:
3
1 6 2
for the simplex [1, 2, 3] {[1,3,4], [1,4,5], [1,5,6], [2,5,6], [2,4,5], [2,3,4], [1,4], {1},
[1,5], [4,5], [5,6], {5}, [2,4], [3,4], {2}, {4}} is a simplicial subdivision.
Definition 10.4.6 Let S = [x1 , . . . , xn+1 ] be an n-dimensional space which is sim-
plicially subdivided, with V being the set of all vertices of all sub-simplexes. A
10.4 A Proof of Brouwer’s Theorem from Sperner’s Lemma 225
function λ : v → {1, . . . , n + 1} satisfying λ(v) ∈ ψ(v) is called a proper labelling

where ψ is the carrier function defined in Definition 10.4.2. A sub-simplex is said to
be completely labelled if λ takes all the values from 1 to n + 1 on the set of all its
vertices.
1
n+1
Remark 10.4.7 For the simplex S = [x1 , . . . , xn+1 ] the centroid xi ,
n + 1 i=1
denoted by b(S) is called the barycentre of S. For simplexes S1 and S2 , we write
S1 ≤ S2 if S1 is a face of S2 . For simplexes Sk ≤ Sk−1 · · · ≤ S1 ≤ S0 ≤ S, the family
of all simplexes b(S0 ), . . . , b(Sk ) is called a barycentric subdivision of S. Given a
simplex S, it is possible to obtain a simplicial subdivision such that the diameter of
each sub-simplex is less than any given positive number. The mesh of a simplicial
subdivision is the diameter of the largest proper sub-simplex.
We are now in a position to state and prove Sperner’s Lemma [28].
Theorem 10.4.8 (Sperner [28]) Let S = [x1 , . . . , xn+1 ] be a simplex which is sim-
plicially subdivided and properly labelled by the function λ. Then there is an odd
number of completely labelled sub-simplexes in the subdivision.
Proof When n = 0, the simplex consists of a single point x1 bearing the label 1 and
thus that is one completely labelled sub-simplex x1 itself.
Suppose the statement is true for n − 1. Given a simplicial subdivision of S, let
A1 be the number of all completely labelled n-simplexes, A2 the number of almost
completely labelled n-simplexes, i.e. those for which the sense of λ is {1, . . . , n}, A3
the number (n − 1) simplexes on the boundary of S that bear all the labels {1, . . . , n}
and A4 the number of all (n − 1) simplexes with labels {1, . . . , n} in the interior of
S.
An (n − 1) simplex lies either on the boundary of S and is the face of a single n-
simplex in the subdivision or it is a common face of two n-simplexes. Each simplex
of type A1 exactly one face labelled {1, 2, . . . , n}. Each sub-simplex of type A2
contributes two faces labelled {1, 2, . . . , n}. However, inside faces appear in two
simplexes while boundary faces appear in one sub-simplex. Thus we get 2 A2 + A1 =
A3 + 2 A4 .
On the boundary, the only (n − 1) dimensional face labelled {1, 2, . . . , n} can be
on the face F ⊆ S whose vertices are labelled {1, 2, . . . , n}. So we can apply the
inductive hypothesis for F which forms a complete labelled (n − 1) dimensional sub-
simplex. By hypothesis F has an odd number of completely labelled sub-simplexes.
Thus by definition of A3 , A3 is odd. Since A1 + 2 A2 = A3 + 2 A4 , A1 must be odd.
So the theorem is true for all n.
Brouwer’s fixed-point theorem can be deduced from Sperner’s lemma. Indeed, it
suffices to prove it for n-dimensional simplexes.
Theorem 10.4.9 (Brouwer) Let S = [v1 , . . . , vn+1 ] be the n-dimensional simplex
in Rn+1 with vertices vi = (0, 0, . . . , 0, 1, 0, 0, . . . , 0) (with 1 in the ith coordinate
and zero elsewhere) for i = 1, 2, . . . , n + 1. If f : S → S is continuous then f has
a fixed point in S.
Proof Let (Sn ) be a sequence of simplicial subdivisions of S such that each Sk

is a subdivision of Sk−1 with mesh(Sk ) → 0 as k → ∞. Let > 0 and Sk be the
division such that mesh(Sk ) < , for k ≥ k(). Let V be the set of vertices of the sub-
division and λ : V → {1, . . . , n + 1} be a labelling function. For v ∈ [vi1 , . . . , vi ]
choose
λ(v) ∈ {i 1 , . . . , i } ∩ {i : f i (v) ≤ λi }

n+1
n+1
where v = λi vi and f (v) = f i (v)vi .
i=1 i=1
This intersection is non-empty, for if f i (v) > λi for all i ∈ {i 1 , . . . , i }, then we
would have

n+1

n+1
1= f i (v) > vi = vi = 1
i=1 j=1 i=1
a contradiction, with the second inequality following from v ∈ [vi1 , . . . , vi ]. It may
be noted that we are using the representation of v, f (v) in barycentric coordi-
nates. Since λ is a labelling function satisfying Sperner’s lemma there exists a
completely labelled sub-simplex [ p1 , . . . , pm+1

] such that f i ( pi ) ≤ ( pi )i . As ↓ 0
there is a subsequence of simplexes such that pi → q as → 0 for each i =
1, 2, . . . , m + 1. Since f is continuous f i (q) ≤ qi , i = 1, . . . , m + 1. If f (q) = q,
f i (q) ≤ qi for all i and f k (q) < q for some k would contradict f k (q) = qk = 1.
So f (q) = q.
Using Sperner’s lemma, one can deduce a classical result due to Knaster Kura-
towski and Mazurkiewicz [14], called Knaster–Kuratowski–Mazurkiewicz Lemma
or simply KKM lemma.
Theorem 10.4.10 (KKM Lemma) Let be the simplex [e1 , . . . , em+1 ] in Rm+1
and F1 , . . . , Fm+1 be a family of non-empty closed subsets of such that for each
A ⊆ {1, 2, . . . , m + 1} the convex hull of {ei : i ∈ A} ⊆ Fi .
i∈A

m+1
Then Fi is non-empty and compact.
i=1

m+1
Proof Clearly Fi is closed and compact. For > 0 given subdivide into sub-
i=1
simplexes with mesh size ≤ . For a vertex v of the subdivision lying on the face
ei1 , . . . , eik+1 , by hypothesis there is an index i ∈ {i 1 , . . . , i k+1 } with v ∈ Fi . Labelling
all the vertices in this way. We observe that it satisfies all the conditions of Sperner’s
Lemma. So there is a completely labelled sub-simplex [ p1 , . . . , pm+1
] with pi ∈

Fi for each i. As ↓ 0, choosing a subsequence pi converging to q and noting
pi ∈ Fi for each i, it follows that z ∈ Fi for each i as Fi is closed for each i. Thus

m+1
z∈ Fi .
i=1
10.4 A Proof of Brouwer’s Theorem from Sperner’s Lemma 227
Now KKM lemma implies Brouwer fixed-point theorem.
Theorem 10.4.11 Theorem 10.4.10 (KKM Lemma) implies Theorem 10.3.1

(Brouwer fixed-point theorem).
Proof Let m = [e1 , . . . , em+1 ] be the m-dimensional simplex in Rm+1 and f :

m → m a continuous map. Define Fi = {x ∈ : f i (x) ≤ xi }. The collections
{e1 , . . . , em+1 } and {Fi , . . . , Fm+1 } satisfy the hypotheses of KKM Lemma. For

m+1
k+1
x ∈ [ei1 , . . . , eik+1 ] f i (x) = xi j . So at least for one xi j , f i j (x) ≤ xi j and
i=1 i=1

m+1
Fi are all closed. So Fi is compact and non-empty. But this set is precisely
i=1
{x ∈ : f (x) ≤ x} and is simply the set of fixed points of f .
We can also prove that Brouwer’s theorem implies the KKM Lemma.
Theorem 10.4.12 Brouwer’s Theorem 10.3.1 implies KKM Lemma 10.4.10.
Proof Let K = convex hull of {ai : i = 1, . . . , m + 1}. K is evidently compact and

m+1
convex. Suppose Fi = φ. Then {Fic : i = 1, . . . , m + 1} is an open cover for
i=1
K . So there exists a partition of unity f 1 , . . . , f m+1 subordinate to this covering

m+1
by Theorem 1.2.20. Let g : K → K be defined by g(x) = f i (x)ai . Clearly g is
i=1
continuous and maps K into itself. By Brouwer’s theorem it has a fixed point p. Let
A = {i : f i ( p) > 0}. Then p ∈ co{ai : i ∈ A} and p ∈ Fi for no i, a contradiction.
Aliter(Peleg [22]) Let F1 , . . . , Fm+1 satisfy the hypotheses of KKM Lemma. Define
gi (x) = d(x, Fi ) and f : → by f i (x) = 1+xi +g i (x)
m+1
g (x)
. Then f is continuous and
j=1 j
so by Brouwer’s theorem f has a fixed point x. So g j (x) = 0 for all j ∈ {1, . . . , m +

m+1
1}. So F j = φ.
j=1
10.5 Scarf’s Algorithm
Scarf [24] provided an algorithm for approximating a fixed point guaranteed by

Brouwer’s fixed-point Theorem 10.3.1. In this section Scarf’s algorithm is outlined.
It may be added that many other algorithms improving on Scarf’s contribution, due
to Eaves, Kuhn and others, have emerged subsequently.
A constructive proof of Sperner’s Lemma 10.4.8 is described below in line with
Scarf.
Let Sn be the standard simplex in Rn . Let {v 1 , v 2 , . . . , v n , v n+1 , v n+k } be a

restricted simplicial subdivision where v 1 , . . . , v n are the unit vectors of the ini-
tial subdivision with no subdivision along the boundaries. We label each vertex
v with (v) ∈ {1, 2, . . . , n} such that (v i ) = i for i = 1, 2, . . . , n. The remaining
vertices in the interior can be labelled arbitrarily from {1, 2, . . . , n}. We construct a
completely labelled simplex in this restricted subdivision as follows:
We construct the unique initial sub-simplex with vertices v2 , . . . , vn by adding
an additional vertex v n+ j such that the algorithm for completely labelled restricted
subdivision halts in (v n+ j ) = 1. If (v n+ j ) = k = 1, then k appears in the labelling
of this sub-simplex twice. We eliminate the vertex whose label coincides with the
label of v n+ j leading to another sub-simplex with a new vertex v n+k( j) . This step can
be repeated again. Thus at each stage of the algorithm, one has a sub-simplex with
n − 1 labels {2, . . . , n} terminating only when we reach a restricted sub-simplex
having all the n labels {1, 2, . . . , n}.
We note that this algorithm leads to a completely labelled sub-simplex in a finite
number of steps which has not been reached previously. Let S be the first simplex
revisited, if possible by this algorithm. If it is not the initial sub-simplex, then it can
be reached in two ways through either one of the adjacent simplices with (n − 1)
distinct labels. As per the algorithm, both these adjoint sub-simplexes had already
been reached during the algorithmic construction contradicting that S is not the first
simplex revisited. This computational scheme clearly encapsulates a constructive
proof of Sperner’s lemma.
Scarf’s algorithm for approximating a fixed point of a continuous self-map on a
standard simplex Sn can now be briefly described.
Definition 10.5.1 Let f : Sn → Sn be a continuous map with a fixed point x ∗ . For

> 0 given, x ∈ Sn is called -almost fixed point if f if x ∗ − x ≤ .
Theorem 10.5.2 (Scarf algorithm) Let f : Sn → Sn be a continuous function. Given

> 0 there exists δ > 0 less then and for G, a simplicial subdivision of Sn such that
each vertex of G is labelled by i = min{ j : f j (x) ≤ x, > 0 where x = (x1 , . . . , xn )}
there is a completely labelled sub-simplex S of G such that for x ∈ S , f (x) − x ≤
2n and f (y) − f (z) < for y − z ≤ δ with y, z ∈ S .
Proof Since f is continuous on Sn , it is uniformly continuous in view of the com-

pactness of Sn . So given > 0 there exists δ > 0 with δ < such that y − z < δ,
y, z ∈ Sn implies f (y) − f (z) < .
j j j
For each vertex v j one can associate an index i such that vi > 0 (v j = (v1 , v2 , . . . ,
j j
vn )) and f i (v j ) ≤ vi , f (v j ) being ( f 1 (v j ), . . . , f n (v j ).
From the constructive proof of the Sperner lemma outlined above, there exists
a restricted completely labelled simplicial subdivision S of G with mesh S < δ
containing a fixed point x ∗ of f . So for x ∈ S , |xi − f i (x)| ≤ x − f (x) ≤ x −
x ∗ + x ∗ − f (x) < δ + < 2. So x − f (x) < 2n.
10.6 More on Brouwer’s Fixed-Point Theorem 229
10.6 More on Brouwer’s Fixed-Point Theorem
Ky Fan [6] proved the following consequence of KKM Lemma.

Theorem 10.6.1 Let X ⊆ Rn and for each x ∈ X , let F(x) be a closed subset of
Rn . Suppose
k
(i) for any subset {x1 , . . . , xk } ⊆ X , c0 {x1 , . . . , xk } ⊆ F(xi ).
i=1
(ii) F(x) is compact for some x ∈ X . Then F(x) is non-empty and compact.
x∈X
non-empty subset S of X , S ∪ {x} is finite and conv{S ∪

Proof Since for each finite
{x}} ⊆ F(s) ∪ F(x) F(s) ∩ F(x) is non-empty and compact by KKM lemma.
s∈S s∈S
Thus {F(s) ∩ F(x) : s ∈ X } has finite intersection property and is a collection of
closed subsets of F(x). So F(s) is non-empty and closed by the compactness of
s∈X
F(x).
We use the following.
Definition 10.6.2 Let U be a binary relation on a set K , i.e. a subset of K × K , such
that U (x) = {(x, y) ∈ K } ⊆ K (this may be viewed as the set of elements y ‘bigger
than’ x). An element x ∈ K is called U -maximal if U (x) = φ. The U -maximal set
of K is simply {x ∈ K : U (x) = φ}. The graph of U is {(x, y) : y ∈ U (x)}.
Theorem 10.6.3 (Sonnenschein [27]) Let K ⊆ Rn be a non-empty compact convex
subset and U be a binary relation on K , satisfying the following:
(i) x ∈
/ convex hull of U (x) for all x ∈ K ;
(ii) for y ∈ U −1 (x), there exists x ∈ K such that y ∈ int U −1 (x ).
Then K has a U -maximal element and the U -maximal set is compact.
Proof Clearly

{x : U (x) = φ} = (K − U −1 (x))
x∈K

= (K − int U −1 (y)) by (ii)
y∈K
For each y, F(y) = K − int U −1 (y) is a closed subset of K and hence is compact.
n n
For y ∈ co{x1 , . . . xn : xi ∈ K }, y ∈ F(xi ) otherwise y ∈
/ F(xi ) implies y ∈
/
i=1 i=1
n
n
(K − int U −1 (xi )) = K − int U −1 (xi ) or y ∈ int U −1 (xi ) ⊆ U −1 (xi ) for all
i=1 i=1
i = 1, 2, . . . , n. So xi ∈ U (y) for all i = 1, 2, . . . n. Since y is in the convex hull of

xi , y ∈ convex hull of U (y), contradicting (i). Thus F satisfies the hypotheses of
Theorem 10.6.1 and so F(x) = φ and compact. Thus the set of all U -maximal
x∈K
elements is non-empty and compact.
Definition 10.6.4 Let S, T be non-void subsets of R and ψ : S → 2 . ψ is called
n T
upper semicontinuous at x0 ∈ S if for xn ∈ S, n ∈ N and (xn ) → x0 yn ∈ ψ(xn ) and

yn → y0 imply that y0 ∈ ψ(x0 ). ψ is said to be lower semicontinuous at x0 if xn ∈ S
for n ∈ N, xn → x0 ∈ S and y0 ∈ ψ(x0 ) imply that there exists yn ∈ ψ(xn ) for n ∈ N
such that yn → y0 . ψ is called continuous if it is both upper and lower semicontinuous
ψ : S → 2T is said to have open lower sections it ψ −1 {y} = {x ∈ S : y ∈ ψ(x)} is
open in S.
Remark 10.6.5 ψ : S → 2R is upper semicontinuous on S if for every open subset
n
V of Rn {x ∈ S : ψ(x) ⊆ V } is open in S.
Following Krasa and Yannelis [15], we can give an alternative proof of Theorem
10.6.3 using Brouwer’s theorem.
Aliter for Theorem 10.6.3 For each x ∈ K , define ϕ(x) = U −1 {x}. ϕ has open
lower sections. Suppose U −1 (x) = φ for all x. By assumption (ii) of Theorem
10.6.3 int U −1 {x} = φ. So {intϕ(x) : x ∈ K } is an open cover for K and as K
is compact it has a finite subcover {int U −1 (y1 ), int U −1 (y2 ), . . . , int U −1 (ym )}
say. Let gi (x) = dist (x, K − int ϕ(yi )} and αi (x) = mgi (x)
g j (x)
. Clearly each αi is
j=1

m
continuous, 0 ≤ αi ≤ 1, αi = 0 on K − int ϕ(yi ) and αi (x) = 1 for x ∈ K .
i=1
Thus {αi } is a partition of unity subordinate to the covering {int ϕ(yi )} for K .
m
The map f : K → K defined by f (x) = αi (x)yi is continuous on the com-
i=1
pact convex set K in Rn and so by Brouwer’s fixed-point theorem it has a fixed

m
point x0 = f (x0 ) = αi (x0 )yi in K . For all the αi (x0 ) = 0, x0 ∈ int ϕ(yi ) and
i=1

m
−1
x0 ∈ U (yi ) or yi ∈ U (x0 ). So x0 = αi (x0 )yi ∈ convex hull of U (x0 ) contra-
i=1
dicting the hypothesis of Theorem 10.6.3. Thus U has a maximal element.
Some consequences of Theorem 10.6.3 can now be deduced.
Theorem 10.6.6 Let U be a binary relation on K ⊆ Rm with values in Rm , K being
a non-empty compact convex subset. Suppose
1. x ∈
/ U (x) for all x ∈ K ;
2. U (x) is convex for all x ∈ K ;
3. {(x, y) : y ∈ U (x)} is open in K × K .
Then U has a maximal element and the set of all such maximal elements of U is
compact.
Theorem 10.6.7 Let K ⊆ Rm be a non-void compact convex set and E ⊆ K × K

be closed. Suppose
1. (x, x) ∈ E for all x ∈ K ,
2. for each y ∈ K , {x ∈ K : (x, y) ∈
/ E} is convex.
Then there exists k0 ∈ K such that K × {k0 } ⊂ K . The set of such k0 is compact.
The above results are due to Ky Fan.

Proof of Theorem 10.6.6 For each x ∈ K , define U (x) = {y : (x, y) ∈ / E}. Clearly
U (x) is convex, open and x ∈
/ U (x). So by Theorem 10.6.3, U has a maximal element
k0 and the set of al such maximal elements of U is compact. Thus U (k0 ) = {k ∈ K :
(k, k0 ) ∈
/ E} = φ. Thus K × {k0 } ⊆ E.
For the proof of Theorem 10.6.7 set E = {(x, y) : (x, y) ∈
/ U }. Clearly (x, x) ∈ E
as x ∈/ U (x). Since {(x, y) : y ∈ U (x)} is open, {(x, y) ∈ E} is closed in K × K .
Further U (x) is convex. So by Theorem 10.6.6 for k0 ∈ K , K × {k0 } ⊆ E or {y :
y ∈ U (k0 )} = φ.
One can deduce the following
Theorem 10.6.8 Let f : K → Rm be a continuous map on a non-void compact

convex subset K . Then there exists k0 ∈ K such that for all p ∈ K , < k0 , f (k0 ) ≥
p, f (k0 ).
Proof Define the binary relation U on K by y ∈ U (x) if and only if
y, f (x) > (x, f (x).
Since f is continuous, {y ∈ K : (x, y) ∈ U (x)} is open in K × K , x ∈ / U (x) and

U (x) is convex, by Theorem 10.6.7, there exists k0 ∈ K such that U (k0 ) = φ. So for
k0 ∈ K , k, f (k0 ) ≤ k0 , f (k0 ), for all k ∈ K .
The above result is referred as Hartman–Stampacchia Lemma.

We can now prove the existence of Walrasian equilibrium in an exchange economy
without production.
Let x = (x1 , . . . , xn ) be an allocation of goods to m consumers (each element in
the above list could itself be a vector of goods). Let w = (w1 , . . . , wn ) be the initial
endowments of the n consumers. Let p = ( p1 , . . . , pn ) be a vector of n prices. In the
competitive economy, a Walrasian equilibrium or competitive equilibrium is repre-
sented by a list (x ∗ , p ∗ ) such that xi∗ is preferred to xi for all xi satisfying the budget
constraint, p ∗ xi ≤ p ∗ wi . Thus all consumers maximize
their utility. We assume that
demand does not exceed supply for each good or (xi∗ − wi ) ≤ 0. Aggregate excess
i
demand function z( p) is defined by z( p) = (xi∗ ( p) − wi ). Further, we assume
i
that excess demand functions are homogeneous of degree 0 whenever z( p ∗ ) ≤ 0,
where (x ∗ , p ∗ ) represents a competitive equilibrium. Also if each consumer has
strictly increasing and strictly convex preferences then z( p) is continuous. Accord-

ing to the Walras law, pz( p) = 0. In practical terms it means that if z 1 = z n−1 = 0
and if pn > 0 then z n must be zero. Thus in calculating competitive equilibrium, we
have to ensure that n − 1 markets are clear for clearing the n th market. The existence
of a Walrasian equilibrium can be proved as follows, using Brouwer’s fixed-point
theorem:
As z is homogeneous of degree zero, z(t p ∗ ) ≤ 0 whenever z( p ∗ ) ≤ 0. We can
normalize the price vector so that the price vector lies in the (n − 1) dimensional

n
unit simplex S = { p ∈ Rn : p = ( p1 , . . . , pn ) with pi ≥ 0 and pi = 1}. The map
i=1
g : S → S defined by g( p) = (g1 ( p), . . . gn ( p)) where gi ( p) = 1+( pi +max{0,z
n
i ( p)})
is
j=1 max{0,z j ( p)}
∗
continuous and maps S into itself. By Brouwer’s theorem g has a fixed point p in S.
p∗ +max{0,z i ( p∗ )}
Thus p ∗ = g( p ∗ ). So for p ∗ = ( p1∗ , . . . , pn∗ ) we have pi∗ = 1+ i n max{0,z ∗ . Thus
j ( p )}
j=1

n
pi∗ max{0, z j ( p ∗ )} = max{0, z i ( p ∗ )}. So n
i=1 pi∗ z( pi∗ ) n
j=1 max{0, z j ( p ∗ )}
j=1
= i=1 n
z i ( p ∗ ) max{0, z i ( p ∗ )}. By the Walras law the right-hand side of the above
equation is zero. Since each of the n-terms in this sum is non-negative, z i ( p ∗ ) ≤ 0
for each i. Since the Walrus law holds at p ∗ , p ∗ is a competitive equilibrium.
Uzawa [32] has proved that Walras equilibrium theorem implies Brouwer’s the-
orem and Uzawa’s theorem is stated and proved below.
Theorem 10.6.9 (Uzawa [32]) Walras’ theorem implies Brouwer fixed-point theo-
rem.
n
Proof Let S be the standard unit simplex {(π1 , . . . , πn ) : πi ≥ 0, i=1 πi = 1} in
Rn and f : S → S be a continuous map. Construct an excess demand function by
x( p) = (x1 ( p), . . . , x2 ( p)) defined by

p
xi ( p) = f i − pi μ( p), i = 1, 2, . . .
λ( p)

n p

n
i=1 pi f i λ( p)
where p = ( p1 , . . . , pn ), λ( p) = pi and μ( p) = n 2
.
i=1 pi
i=1
Clearly f i λ(pp) and pi μ( p) are positive homogeneous of order zero. Thus the
excess demand function x( p) satisfies
1. continuity;
2. homogeneity of order 0; i.e. x(t p) = x( p) for t > 0;

n
3. the Walras law; i.e. pi x( pi ) = 0.
i=1
∗
Soby Walras
equilibrium theorem it has an equilibrium p . So by definition of x( p),
p∗ ∗
f i λ( p∗ ) ≤ pi μ( p ) for i = 1, . . . , n with equality unless pi = 0. Writing π ∗ =
∗
p∗
λ( p∗ )
and β = λ( p ∗ )μ( p ∗ ) we get f i (π ∗ ) ≤ βπi∗ with equality unless πi∗ = 0. Since
S is the standard simplex and π ∗ , f (π ∗ ) ∈ S = {(π1 , . . . πn ) : i=1n
πi = 1, πi ≥ 0}
and so β = 1. So f i (π ∗ ) ≤ πi∗ with equality unless πi∗ = 0. This again implies that
f i (π ∗ ) = πi∗ , i = 1, 2, . . . , n. Thus π ∗ is a fixed point for f .
Complementing von Neumann’s concept of cooperative games, Nash [18] defined

the concept of a non-cooperative game and its equilibrium point. Using Brouwer’s
theorem, he proved the existence of an equilibrium point (or Nash equilibrium) for
an n-person non-cooperative game. In what follows the basic ideas for the existence
of a Nash equilibrium of non-cooperative games are described, following closely
Nash [19].
An n-person games is a set of n-players (or positions), each associated with a
finite set of pure strategies. Corresponding to each player i, a pay-off function pi
mapping the set of all n-tuples of pure strategies into the set of real numbers. By
n-tuple is meant a set of n elements, each element being associated with a different
player. A mixed strategy of player i is a set of non-negative numbers with unit sum
and are in one-one correspondence with his pure strategies. We write si = Ciα πiα
α
with Ciα ≥ 0 Cα = 1 where πiα are pure strategies of the player i (si ) can be
α
regarded as points in a simplex with vertices πiα . Viewing it as a convex set enables
one to get a natural way of linear combination for mixed strategies. Players can be
denoted by suffixes i, j, k while pure strategies are represented by α, β, γ etc. While
si , ti denote mixed strategies, πiα would mean αth pure strategy of the ith player.
The pay-off function pi has a unique extension to the n-tuples of mixed strategies
which is linear in the mixed strategy of each player. This linear extension denoted
by pi , is also written pi (s1 , . . . , sn ). s, t denote the n-tuple of mixed strategies. For
s = (s1 , . . . , sn ), pi (s) = pi (s1 , . . . , sn ) is the corresponding pay-off function. Such
an n-tuple s is a point in the product space of the vector spaces containing the mixed
strategies.
The notation (s, ti ) denotes (s1 , s2 , . . . , si−1 , ti , si+1 , . . . , sn ) where s = (s1 , . . . ,
sn ). The resultant of successive substitutions ((s, ti ); r j ) is denoted by (s; ti ; r j ). An
n-tuple s is an equilibrium point if and only if for each i
pi (s) = max[ p(s; ri )]

ri
In other words, the equilibrium point is an n-tuple such that each player’s mixed
strategy maximizes his pay-off if the strategies of others are held fixed. So each
player’s strategy is optimal against those of the remaining players. A mixed strategy
si is said to use a pure strategy πiα if si = Ciβ πiβ and Ciα > 0. If s = (s1 , . . . , sn )
β
and si uses πiα , we say that s uses πiα . Since pi (s) is linear in si
max[ pi (s; ri )] = max[ pi (s; πiα )]

ri α
Define piα (s) = pi (s; πiα ). Then a trivial necessary and sufficient condition
for s to be an equilibrium point is pi (s) = max piα (s). For s = (s1 , . . . , sn ), si =
α
ciα πiα , then pi (s) = ciα piα (s). So for the validity of pi (s) = max piα (s), we
α
α α
must have ciα = 0 whenever piα (s) < max piβ (s). This simply means that s does
β
not use πiα unless it is an optimal pure strategy. With these preliminaries Nash [19]
proved the following:
Theorem 10.6.10 Every finite game has an equilibrium point.
Proof Let s be an n-tuple of mixed strategies, pi (s) the corresponding pay-off to

player i and piα (s) the pay-off to player i if he changes his αth pure strategy πiα
while the others continue to use their respective mixed strategies from s. Define
for each s = (s1 , . . . , sn ) the map φiα (s) = max(0, piα (s) − pi (s)) and for each
s+ ϕ (s)π
component si of s define si = i 1+ α ϕiαiα (s)iα and write s = (s1 , . . . , sn ). Consider the
α
map T (s) = (s1 , . . . , sn ) = s . Since the pay-off functions are all continuous s →
T (s) is a continuous map on the cell formed by all the strategies. So by Brouwer’s
fixed point theorem T has a fixed point ς = (ς1 , . . . , ςn ). That part of πiα used in ςi
must not be decreased by T . So for all β, ϕiβ (ς) = 0 so that ςi does not exceed 1.
Thus if ς is fixed by T , ϕiβ (ς) = 0 for all β. So no player can improve his pay-off by
moving to a pure strategy πiβ . This is precisely the definition that ς is an equilibrium
point. Conversely if ς is an equilibrium point, then all ϕ’s vanish so that ς is a fixed
point of T .
By a symmetry (or automorphism) of a game is meant a permutation of its strate-
gies such that if two strategies belong to a single player they move to two strategies
belonging to a single player. Thus, a permutation φ of the pure strategies induces
a permutation of the players. Each vector of pure strategies permutes into another
vector of pure strategies. Let ψ be the induced permutation of these vectors. If ξ
is a vector of pure strategies and pi (ξ) the pay-off to player i when the vector ξ is
applied, then symmetry requires if j = i ψ then p j (ξ ψ ) = pi (ξ). This permutation

φ has a unique linear extension to the mixed strategies. Thus for si = Ciα πiα
α
(si )φ = Ciα (πiα )φ . From this extension to mixed strategies, one gets an obvious
α
extension of ψ to n-tuples of mixed strategies. A symmetric n-tuple s of a game is
defined by s ψ = s for all the extensions ψ.
Theorem 10.6.11 (Nash [19]) A finite cooperative game has a symmetric equilib-
rium point.

Proof si = πiα / 1 has the property (si )φ = s j where j = i ψ so that the n-
α α
tuple s = (s1 , . . . , sn ) is fixed under any ψ. So any game has at least one symmetric
n-tuple. Clearly if s and t are symmetric, so is their convex combination. Thus the set
of symmetric n-tuples is convex. It is also closed. Consider the map T taking s to s
defined in the proof of Theorem 10.6.10 and ψ is a symmetry on the game, then for
s = T s, (s )ψ = T (s ψ ). So T maps the closed convex subset of symmetric n-tuples
into itself and by the continuity of T and Brouwer’s theorem T has a fixed point in
the set of symmetric n-tuples. Thus there is a symmetric equilibrium point.
In an earlier paper, Nash [18] outlined the proof of the existence of the equilibrium
point using Kakutani’s fixed-point theorem. This seminal work culminated in the
award of a Nobel prize (jointly in economics) for Nash. For a perspective on the
impact of Nash equilibrium on social sciences Holt and Roth [10] may be referred.
10.7 A Proof of the Fundamental Theorem of Algebra
Arnold [2] and later Niven [21] attempted a proof of the fundamental theorem of
algebra based on Brouwer’s fixed-point theorem. Subsequently, it was noted in [3]
that both the proofs contained errors. Later, Fort [7] salvaged it and we present
his proof first of Brouwer’s fixed-point theorem in the plane and then that of the
fundamental theorem of algebra.
By S we denote the set of all complex numbers z with |z| = 1. For z ∈ S, A(z) is
the set of all real numbers θ for which z = eiθ . Thus A(z) is the set of all arguments
of z. A continuous function defined on a subset X of the plane of complex numbers
and taking values in S is said to have a continuous logarithm on X if there exists a
real-valued continuous function φ of X such that f (z) = eiφ(z) for all z ∈ X . Two
basic properties of complex numbers, used in the sequel, are the following:
(a) for z 1 , z 2 ∈ S, |z 1 − z 2 | < 2 and θ1 ∈ A(z 1 ), then for a unique θ2 ∈ A(z 2 ) with
|θ1 − θ2 | < π.
(b) if θi ∈ A(wi ), i = 1, 2, and |θ1 − θ2 | < π, then |θ1 − θ2 | ≤ π|z 1 − z 2 |.
Theorem 10.7.1 If f : D → S is a continuous mapping, then f has a continuous
logarithm, D being a closed disc in the plane.
Proof Let D be the disc {z : |z − q| ≤ r }. Thus q is the centre of D and r its radius.
From the uniform continuity of f , it follows that there exists δ > 0 such that for
z 1 , z 2 ∈ D with |z 1 − z 2 | < δ, | f (z 1 ) − f (z 2 )| < 13 . Choose n ∈ N such that nr < δ.
Define Dk = {z : |z − q| ≤ rnk } for k ∈ N with 0 ≤ k ≤ n. Define φ on D by defining
it successively on D0 , . . . , Dn . φ(D0 ) is defined as φ(q) = θ ∈ A( f (q)) such that
0 ≤ θ < 2π. If φ is defined on Dk and z ∈ Dk+1 , let z be the nearest point of Dk to
z. Since |z − z | < δ, | f (z) − f (z )| < 13 . So by property (a) stated above, we may
define φ(z) to be the unique number A( f (z)) that differs from φ(z ) by less than π.
Let Sk be the statement: if z 1 , z 2 ∈ Dk and |z 1 − z 2 | < δ, then |φ(z 1 ) − φ(z 2 )| <
π. Clearly S0 is true as D0 contains only one point. Suppose Sk is true. For z 1 , z 2 ∈
Dk+1 with |z 1 − z 2 | < δ. Consider z 1 and z 2 the nearest points of z 1 and z 2 in Dk
respectively. From the definition of φ, it follows that |φ(z i ) − φ(z i )| < π, for i = 1, 2.
It is readily seen that |z 1 − z 2 | < δ and by inductive hypothesis |φ(z 1 ) − φ(z 2 )| < π.
So we have
|φ(z 1 ) − φ(z 2 )| ≤ |φ(z 1 ) − φ(z 1 )| + |φ(z 1 ) − φ(z 2 )| + |φ(z 2 ) − φ(z 2 )|

≤ π{| f (z 1 ) − f (z 1 )| + | f (z 1 )| − f (z 2 )| + | f (z 2 )| − f (z 2 )|}
π π π
< + + =π
3 3 3
So Sk+1 is true. Thus S j is true for all j with 0 ≤ j ≤ n.
Using Sn and (b), we get that for z 1 , z 2 ∈ D with |z 1 − z 2 | < δ, |φ(z 1 ) − φ(z 2 )| ≤
π| f (z 1 ) − f (z 2 )|. This implies that φ is continuous in view of f being continuous.
As φ(z) ∈ A( f (z)) for z ∈ D, f has a continuous logarithm on D.
Proposition 10.7.2 The identity map on S does not have a continuous logarithm.
Proof If the identity map on S has a continuous logarithm, then for z ∈ S, z = eiφ(z) ,
where φ is continuous and real-valued. The map g : [0, 2π] → R defined by g(θ) =
φ(eiθ ) − θ is continuous and an integral multiple of 2π. So g must be a constant
function. Now g(0) = g(1) and g(2π) = g(1) − 2π, a contradiction.
The following no-retraction theorem can be deduced.
Theorem 10.7.3 There does not exist a retraction of the closed unit disc onto its
boundary S.
Proof If f is such a mapping, then by Theorem 10.7.1, there is a continuous function
φ on D such that f (z) = eiφ(z) for all z ∈ D. Since f (z) = z on S, this contradicts
Proposition 10.7.2.
Theorem 10.7.4 (Brouwer’s fixed-point theorem for plane) If g is a continuous
function mapping the closed unit disc into itself, then g has a fixed point in D.
Proof Suppose g has no fixed point in D. For each z ∈ D, let f (z) be the unique
point of S such that z lies on the segment joining f (z) to g(z). It can be shown that
z → f (z) is continuous and maps D onto S with f (z) = z on S. This contradicts
Theorem 10.7.3. Hence g must have a fixed point.
Remark 10.7.5 Since any two closed discs in the plane are homeomorphic it follows
that any closed disc in the plane has the fixed-point property for continuous functions.
For proving the fundamental theorem of algebra using Brouwer’s theorem, we
need the following.
Proposition 10.7.6 If f is a continuous map on a closed disc D into the set of non-
zero complex numbers, then there exist n distinct continuous mappings h 1 , . . . , h n
such that [h k (z)]n = f (z) for z ∈ D.
Proof f (z) = | f (z)| | ff (z)
(z)|
. By Theorem 10.7.1, as z → f (z)
| f (z)|
maps D into S con-
f (z)
tinuously there exists a continuous map φ such that =e
| f (z)|
. Define h k (z) =
iφ(z)
√ i(φ(z)+2kπ)
n
| f (z)|.e n for k = 1, . . . , n. Then these are the required continuous nth roots
of f (z).
10.7 A Proof of the Fundamental Theorem of Algebra 237
Theorem 10.7.7 (Fundamental Theorem of Algebra) If p is a polynomial of degree

n > 0, then for some z ∈ C, p(z) = 0.
Proof Without loss of generality we may assume that p(z) = 0 for all z ∈ C and the
p(z) 1
coefficient of z n in p(z) is 21 . Now lim n = . So for some r > 0, p(z) and z n
z→∞ z 2
have arguments which
√ differ by less than π2 for |z| ≥ r and | p(z)| < |z|n . Now select
R > r such that | p(z)| + |z| < R for |z| ≤ r . Consider D the disc centred at 0 with
n
radius R. By Proposition 10.7.6, as p is non-zero and maps D into non-zero complex

numbers p has continuous nth root f on D. We can select f such that f (r ) and r
π
have arguments differing by less than 3n for all z with |z| ≥ r . (If for some z with
π
|z | ≥ r , f (z ) and z have arguments that differ exactly by 3n , then the arguments
n π
of p(z) and z differ by 3 , an impossibility.)
For |z| ≥ r , | p(z)| < |z n | and so | f (z)| < |z| and z and f (z) both lie in the
π
circular sector of radius |z| and angle 3n . Thus |z − f (z)| ≤ |z| for |z| ≥ r . But
when |z| ≤ r , |z − f (z)| ≤ |z| + | f (z)| < R. So the function z → z − f (z) maps
D into itself continuously and hence has a fixed point, say z 0 . So f (z 0 ) = 0 or
p(z 0 ) = 0, a contradiction.
10.8 A Generalization of Brouwer’s Theorem
Hamilton [9] extended the Brouwer’s fixed-point theorem for peripherally continuous
maps, while Stallings [29] generalized Brouwer’s theorem for connectivity functions.
Whyburn [35] extended an intersection theorem due to Hurewicz and Wallman [11],
whence he deduced both the generalizations of Hamilton and Stallings. In this section,
Whyburn’s approach to these fixed-point theorems is described. See also [34].
Definition 10.8.1 A subset E of a topological space X is said to be quasi-closed or
of external dimension zero, if for each p ∈ X/E every neighbourhood of p contains
an open set having p, whose boundary does not intersect E. A subset G of X is
quasi-open if its complement is quasi-closed.
Definition 10.8.2 A function f : X → Y where X and Y are topological spaces is

said to be peripherallay continuous if for x ∈ X and open sets U in X containing
x and V in Y containing f (x) there exists an open set W containing x such that
W ⊆ U and boundary Fr (W ) is mapped in V by f .
A map f is called a connectivity function if the associated graph function x →
(x, f (x)) maps connected subsets of X onto connected subsets of the graph of f .
Lemma 10.8.3 (Whyburn [36]) A function f : X → Y is peripherally continuous

if and only if the inverse of every closed (open) set in Y is quasi-closed (quasi-open)
in X .
Proof Suppose f is peripherally continuous. If C is closed in Y then for any p ∈

X − f −1 (C) and an open set U containing p for the open set V = Y − C, f ( p) ∈ V .
So by peripheral continuity of f , there exists an open set W in X with p ∈ W ⊆ U

and f [Fr (W )] ⊆ V . So Fr (W ) ∩ f −1 (C) = φ. Thus for p ∈ V there exists an open
set W ⊆ V containing p for which f r (W ) ⊆ V . Thus f −1 (C) is quasi-closed.
Suppose for each closed subset C of Y f −1 (C) is quasi-closed. Let p ∈ X and
P = f ( p) and let U and V be open sets containing p and P respectively. Let C =
Y − V . As p ∈ X − f −1 (C) and C is closed there is an open set W with p ∈ W ⊆ U
and Fr (W ) ∩ f −1 (C) = φ. So f [Fr (W )] ⊆ Y − C = V . Thus f is peripherally
continuous.
Hereinafter, we assume that the topological spaces are regular T1 spaces. The
following definitions are needed in the sequel.
Definition 10.8.4 (Whyburn [36]) A topological space X is called locally cohesive

if it is connected and each open set containing a point p of X contains the closure
of a canonical region about p, i.e. a connected open set R with connected boundary
Fr (R) such that R is unicoherent between x and Fr (R) or equivalently between x
and X − R. Recall that a connected space or set M is unicoherent or cohesive between
disjoint connected subsets (or points) A and B of M if Ha ∩ Hb is connected for every
representation M = Ha ∪ Hb , Ha , Hb being closed connected subsets containing A
and B, respectively, in their interiors relative to M.
Remark 10.8.5 A locally cohesive space is locally connected and has no local cut
point. If W is a canonical region in X about a ∈ X , any set K separating a and
Fr (W ) in W contains the boundary of a canonical region R lying in W . Thus in a
locally cohesive space X , given a closed subset E of X and an open set U containing
a ∈ X − E contains a canonical region R about a with R ⊂ U and E ∩ Fr (R) = φ.
Thus for G quasi-open, any open set U containing a ∈ G has a canonical region R
about a so that R ⊆ U and Fr (R) ⊆ G.
Definition 10.8.6 Two subsets A and B of a connected space X are weakly separated
in X by a set E provided no component of X − E meets both A and B.
Theorem 10.8.7 Let A and B be disjoint nondegenerate closed connected sets in a

locally cohesive space X . Any quasi-closed set L that separates A and B weakly in
X contains a non-void closed set K that separates A − K and B − K in X .
Proof If A (or B) ⊆ L, then K = A (or B). Suppose neither A nor B is contained in

L. Let H be the union of all components in X − L intersecting A and V be the union
of all components of X − (H ∪ A) intersecting B. Define K = Fr (V ) ∪ H ∩ B
and define U = X − (V ∪ K ). U and V are open and disjoint, K is closed and K ⊆
H ∪ (A ∩ L). From these it follows that X − K = U ∪ V and B − K ⊂ B − H ⊂
V . Also since A ∩ V ⊆ K , we have A − K ⊆ X − (V ∪ K ) = U . We claim that
K = Fr (V ) ∪ H ∩ B ⊆ L. Otherwise for some x ∈ K , x ∈ / L. Let R be a canonical
region about x so that R contains neither A nor B intersects A or B only in case x is
in A or B respectively. So Fr (C) of R lies in H . In any case C intersects X − L and
lies in some component Q of X − L in H . If x ∈ A, this is true as C intersects A;
for x ∈/ A, it follows since some component of X − L in H intersects both R and A.
10.8 A Generalization of Brouwer’s Theorem 239
For x ∈ B, this, however is not possible as C ∩ B = φ; for x ∈

/ B this is not possible
as C intersects some component of V . Thus K = Fr (V ) ∪ H ∩ B ⊆ L.
Theorem 10.8.8 ([35]) If X is locally cohesive, any connected set in X lying in the
union of two disjoint quasi-open sets lies entirely in one of them.
Proof Let E be connected and lie in U ∪ V where U and V are two disjoint quasi-
open sets. Let a ∈ E ∩ U and b ∈ E ∩ V . For each x ∈ E, let Q x be a canonical
region containing x whose boundary C x lies in U or V according as x ∈ U or V . E
being connected there exists a simple chain of such regions a ∈ Q 1 , Q 2 , . . . , Q n b
from a to b only the first in the chain containing a and only the last containing b. Let
Ci = Fr (Q i ) for each i. Then for 1 < k < n, both Q k−1 and Q k+1 must intersect
Ck as each meets Q k but is not contained in Q k . So Ck intersects both Ck−1 and Ck+1
because Ck is connected and Q k−1 ∩ Q k+1 = φ. However this implies that Ci ⊆ U
for all i = 1, . . . , n contradicting Cn ⊆ V .
Corollary 10.8.9 ([35]) Any peripherally continuous function f : X → Y of a
locally cohesive space X into a completely normal T1 space Y preserves connected-
ness. It is a connectivity function whenever X × Y is completely normal.
Proof If E is connected in X , and if f (E) were not connected, there would exist
disjoint open sets U1 and V1 in Y intersecting f (E) such that f (E) ⊆ U1 ∪ V1 . But
U = f −1 (U1 ) and V = f −1 (V1 would then be quasi-open sets intersecting E, con-
tradicting Theorem 10.8.8. So f (E) is connected. The proof that f is a connectivity
map is left as an exercise.
The following extension theorem due to Whyburn is an intersection theorem
improving on Hurewicz–Wallman intersection theorem [11].
Theorem 10.8.10 (Whyburn [35]) Given quasi-closed sets C1 , C2 , . . . , Cn in I n =
[0, 1]n such that for each i, 1 ≤ i ≤ n, Ci weakly separates Ai and Bi in I n . Then
n
Ci = φ, Ai and Bi being the faces of I n on which xi = 0 and xi = 1, respectively.
i=1
Proof For each i, by Theorem 10.8.7, Ci contains a closed set K i separating Ai − K i

and Bi − K i in I n with I n − K i = Ui ∪ Vi where Ui and Vi are disjoint and open and
contain Ai − K i and Bi − K i respectively. Define f (x) for x ∈ I n , by letting f (x)
as the terminal end of the position vector x + d(x) in Rn where the ith component
di of the vector d(x) is ±ρ(x, K i ), the sign being + for x ∈ Ui and − for x ∈ Vi .
Then for x ∈ Ui and each i, di = ρ(x, K i ) ≤ 1 − xi so that 0 ≤ xi + di ≤ 1 while
for x ∈ Vi , di = −ρ(x, K i ) ≥ −xi and again 0 ≤ xi + di ≤ xi ≤ 1. Thus ρ maps I n
into itself. Further f is continuous and so by Brouwer’s fixed-point theorem f has

n n
a fixed point x0 ∈ I n . This implies that d(x0 ) = 0 and x0 ∈ Ki ⊂ Ci .
i=1 i=1
Theorem 10.8.11 (Hamilton-Stallings) Any peripherally continuous function of I n

into itself, n ≥ 2 has at least one fixed point. The same is true of any connectivity
function of itself for n ≥ 1.
Proof Let f : I n → I n , n ≥ 2 be a peripherally continuous function. For x =

(x1 , . . . , xn ) ∈ I n , let f (x) = (x1 , x2 , . . . , xn ) ∈ (I )n = I n . Let g : I n → I n × I n
be the graph function of f . Thus g(x) = (x, f (x)). For 1 ≤ i ≤ n, let pi = πi g :
I n → Ii × Ii be the projection of the graph of f into the planar cell Ii × Ii defined by
pi (x) = πi (x, x ) = (xi , xi ) where f (x) = (x1 , x2 , . . . , xn ). Write i = {(xi , xi ) :
xi = xi , xi ∈ Ii = Ii }.
We now show that pi , i = 1, . . . , n is peripherally continuous. If E ⊆ Ii × Ii is
closed, then πi−1 (E) is closed as πi is closed. So g −1 πi−1 (E) = pi−1 (E) is closed as
f is peripherally continuous.
Since i is closed, Ci = pi−1 (i ) is quasi-closed for 1 ≤ i ≤ n as pi is periph-
erally continuous. We claim that Ci weakly separates in I n the faces Ai and Bi of I n
for which xi = 0 and xi = 1 respectively. Otherwise some component Q of I n − Ci
intersects both Ai and Bi and pi (Q) would be connected by Corollary 10.8.9. Then
pi (Q) would intersect i as it contains pi (a) for some a ∈ Ai for which xi ≥ xi
and also b ∈ Bi where xi ≤ xi . So Ci weakly separates in I n the faces Ai and Bi
n n
for 1 ≤ i ≤ n. So by Theorem 10.8.10 Ci = φ. Since pi−1 (i ) = Ci , x ∈ Ci
i=1 i=1
implies x = x or x = f (x).
For n = 1, that every connectivity map of [0, 1] into itself has a fixed point has
already been proved (see Corollary 3.1.7). For n ≥ 2 we note that such a map is a
peripherally continuous map in view of the following Lemma 10.8.19 and so by the
first part of the theorem has a fixed point.
The proof of this lemma is better understood on the basis of the following concepts
and propositions (see Whyburn [33]).
Definition 10.8.12 A collection of sets G in a topological space is called upper

semicontinuous if for G ∈ G and U an open set containing G, there is an open set V
containing G such that any H ∈ G and intersecting V , H ⊆ U .
Definition 10.8.13 Let (X, T ) be a topological space and G a collection of non-

empty subsets of X . The limit superior of G, written lim sup G is the set of all points
p in X such that every neighbourhood of p contains points from infinitely many sets
in G. The set of all points q in X such that every neighbourhood of q has points from
all but a finite number of sets from G is called the limit inferior of G and is written
lim inf G.
For the proofs of the following propositions Whyburn [33] may be consulted.
Proposition 10.8.14 In a compact metric space, a necessary and sufficient condition

for a collection G of closed sets to be upper semicontinuous is that for each sequence
(G n ) in G with lim inf G n ∩ G for some G ∈ G implies lim sup G n ⊆ G.
Proposition 10.8.15 If X is a compact space, the collection G of disjoint closed

sets in the union of three subcollections G1 , G2 and G3 where G1 is the collection of
components of a closed set, G2 is a null sequence (i.e. a sequence of sets such that
10.8 A Generalization of Brouwer’s Theorem 241
for each > 0, it contains at most a finite number of sets with diameter exceeding
) and G3 is a collection of singletons, then G is upper semicontinuous.
Proposition 10.8.16 Proposition 10.8.14 is true when X is locally compact and all
the sets in G are continua.
Definition 10.8.17 A collection G of subsets of a metric space X is called an upper
semicontinous decomposition of X if G = X , each set in G is compact and G is
G∈G
an upper semicontinuous collection.
Remark 10.8.18 If G is an upper semicontinuous decomposition of a metric space,
then a topology on G can be defined by declaring a neighbourhood of G ∈ G as a
subcollection U of G such that ∪{O : O ∈ U} is open in X and contains G. This
topology on G is called the hyperspace topology on G.
Lemma 10.8.19 ([35]) If X is a locally compact, locally cohesive metric space
and Y a regular T1 space, then any connectivity map f : X → Y is peripherally
continuous.
Proof For x ∈ X , let U and V be open sets containing x and f (x), respectively.
Without loss of generality let U be a canonical region with compact closure since X
is T1 , regular and locally compact. So the boundary B of U is connected and U is
unicoherent between x and B. Let U1 and V1 be open sets such that x ∈ U1 ⊆ U 1 ⊆ U
and f (x) ∈ V1 ⊆ V 1 ⊆ V . Let D = U 1 ∩ f −1 (V 1 ). If x is an interior point of A
comprising the component A0 of D containing x together with the union of all
components of U − A0 except the one containing B or x is separated in U from B
by a component H of D, we get an open set W ⊆ U with x ∈ W and Fr (W ) ⊆ A0
or Fr (W ) ⊆ H . In the former case, let W = int A and in the latter case, choose W
= component of U − H . In both the cases f (Fr (W )) ⊆ V .
Suppose x is neither in the interior of A nor is separated in U from B by any
single component D. So the decomposition of U into the sets A, B components of
D not contained in A and single points of U − A − D is upper semicontinuous.
If φ(U ) = M is the natural mapping of this decomposition, then φ is closed and
monotone (i.e. φ−1 (y) is a continuum for each y in the range of φ). So M is a
locally connected continuum. If a = φ(x) and b = φ(B) and N = C(a, b) the cyclic
element taken in M, then no point of φ(D) ∩ N is a cutpoint of N as no such point
can separate a and b in M or N . Since U is unicoherent between x and Fr (U ),
N is unicoherent. As φ(D) is totally disconnected R = N − φ(D) is connected. So
a ⊆ R. So φ−1 (R) = Q is connected as φ is monotone and closed. Also Q ⊇ {x},
as any region S in U1 must intersect Q. If S is not in A and a is not a cutpoint of
M. φ(S) ∩ N is non-degenerate and connected and so is not connected and is not
contained in φ(D). But then Q ∪ {x} is connected while (x, f (x)) is an isolated
point of the graph of f |Q ∪ x, because (q, f (q)) is not in U1 × V1 for q ∈ Q, since
f (q) ∈ Y − V1 for all q ∈ Q. This contradiction implies that x is either in the interior
of A or is separated in U from B by some single component of D. So f is peripherally
continuous.
Remark 10.8.20 Since f : In → In is a connectivity map it is a peripherally con-

tinuous map. So by first part of Theorem 10.8.11. f has a fixed point for n ≥ 2.
For n = 1, a connectivity map, f : I → I has afixed point. For n = 1, let the map
ξ, x ∈ X is rational
f : X = [0, ξ] → [0, ξ] be defined by f (x) = where ξ is
0, x ∈ X is irrational
a positive irrational number less than 1. Clearly f is peripherally continuous and has
no fixed point.
It may be added that Nash [20] raised the question of whether a connectivity map
on I n has a fixed point.
References
1. Apostol, T.A.: Mathematical Analysis, 2nd edn. Narosa Publishing House, New Delhi (1985)
2. Arnold, B.H.: A topological proof of the fundamental theorem of algebra. Am. Math. Mon.
56, 465–466 (1949)
3. Arnold, B.H., Nivan, I.: A correction. Am. Math. Montly 58(104) (1951)
4. Brouwer, L.E.J.: Uber abbildung von mannigfaltigkeiten. Math. Ann. 71, 97–115 (1912)
6. Fan, Ky.: Fixed point and minimax theorems in locally convex topological linear spaces. Proc.
N.A.S. 38, 121–126 (1952)
7. Fort, M.K.: Some properties of continuous functions. Am. Math. Mon. 59, 372–375 (1952)
8. Glicksberg, I.L.: A further generalization of the Kakutani fixed point theorem with application
to Nash equilibrium points. Proc. Am. Math. Soc. 3, 170–174 (1952)
9. Hamilton, O.H.: Fixed points for certain non continuous transformations. Proc. Am. Math.
Soc. 8, 750–756 (1957)
10. Holt, C.A., Roth, A.E.: The Nash equilibrium: a perspective. Proc. Natl. Acad. Sci. (USA) 101,
3999–4002 (2004)
11. Hurewicz, W., Wallman, H.: Dimension Theory. Princeton University Press, Princeton (1941)
12. Kakutani, S.: A generalization of Brouwer’s fixed point theorem. Duke Math. J. 8, 457–459
(1941)
13. Kannai, Y.: An elementary proof of the no-retraction theorem. Am. Math. Mon. 88, 264–268
(1981)
14. Knaster, B., Kuratowski, K., Mazurkiewicz, S.: Ein Beweis des Fixpunktsatzes fur n-
dimensionale simplexe. Fundam. Math. 14, 132–137 (1929)
15. Krasa, S., Yannelis, N.C.: An elementary proof of the Knaster-Kuratowski-Mazurkiewicz-
Shapely theorem. Econ. Theory 4, 467–471 (1994)
16. Milnor, J.: Analytic proofs of the “Hairy Ball thoerem” and the Brouwer fixed point theorem.
Am. Math. Mon. 85, 521–524 (1978)
17. Munkres, J.R.: Topology : A First Course. Prentice-Hall Inc., Englewood Cliffs (1975)
18. Nash, J.: Equilibrium points in N -person games. Ann. Math. 36, 48–49 (1950)
19. Nash, J.: Non-cooperative games. Ann. Math. 54, 286–295 (1951)
20. Nash, J.: Generalized Brouwer theorem. Bull. Am. Math. Soc. 62, 76 (1956)
21. Niven, I.: Extension of a topological proof of the fundamental theorem of algebra. Am. Math.
Mon. 57, 246–248 (1950)
22. Peleg, B.: Equilibrium points for open acyclic relations. Can. J. Math. 19, 366–369 (1967)
23. Rogers, C.A.: A less strange version of Milnor’s proof of Brouwer’s fixed point theorem. Am.
Math. Mon. 87, 525–527 (1980)
References 243
24. Scarf, H.: The approximation of fixed points of a continuous mapping. SIAM J. Appl. Math.
15, 1328–1343 (1967)
25. Schauder, J.: Der Fixpunktsatz in funktional raumen. Stud. Math. 2, 171–180 (1930)
26. Seki, T.: An elementary proof of Brouwer’s fixed point theorem. Tohoku Math. J. 9, 105–109
(1957)
27. Sonnenschein, H.F.: Demand theory without transitive preferences, with applications to the
theory of competitive equilibrium. In: Chapman, J.S., Hurwicz, L., Richter, M.K., Sonnen-
schein, H.F. (eds.) Preferences Utility and Demand: A Minnesota Symposium, pp. 215–233.
Harcourt, Brace, Jovanovich, New York (1971)
28. Sperner, E.: Neuer Beweis fur die invarianz der Dimensionszahl und des gebietes. Abh. Math.
Semin. Univ. Hambg. 6, 265–272 (1928)
29. Stallings, J.: Fixed point theorems for connectivity maps. Fundam. Math. 48, 249–263 (1959)
30. Traynor, T.: An easy analytic proof of Brouwer’s fixed point theorem. Atti Semin. Math. Fis.
Univ. Modena XLIV, 479–483 (1996)
31. Tychonoff, A.: Ein Fixpunktsatz. Math. Ann. 111, 767–776 (1935)
32. Uzawa, H.: Walras’ existence theorem and Brouwer’s fixed point theorem. Econ. Stud. Q. 3,
59–62 (1962)
33. Whyburn, G.T.: Analytic Topology. American Mathematical Society, vol. 28. Colloquium
Publications, AMS, Providence (1942)
34. Whyburn, G.T.: Connectivity of peripherally continuous functions. Proc. Natl. Acad. Sci.
(USA) 55, 1040–41 (1966)
35. Whyburn, G.T.: Quasi-closed sets and fixed points. Proc. Natl. Acad. Sci. (USA) 57, 201–205
(1967)
36. Whyburn, G.T.: Loosely closed sets and partially continuous functions. Mich. Math. J. 14,
193–205 (1967)
Chapter 11
Schauder’s Fixed Point Theorem
and Allied Theorems
11.1 Introduction
Attempts to extend Brouwer’s fixed point theorem to infinite-dimensional spaces

culminated in Schauder’s fixed point theorem [20]. The need for such an extension
arose because existence of solutions to nonlinear equations, especially nonlinear
integral and differential equations can be formulated as fixed point problems in
function-spaces. This chapter discusses Schauder’s and allied fixed point theorems
with their applications including the existence of Haar integral, invariant mean and
Banach limit. In this context, the following simple example shows that Brouwer’s
theorem is not true for closed balls in infinite-dimensional spaces.
Example 11.1.1 B, be the closed unit ball in C0 the space of all null real sequences
x = (xn ) with the norm x = sup |xn | does not have the fixed point property.
n
For example, the map x → T x where T (x) = (1, x1 , x2 , . . . ), x being (x1 , x2 , . . . ,
xn , . . . ) maps B into itself. If T (x) = x, then x = (xn ) with xn ≡ 1 for all n contra-
dicting that xn is a null sequence.
Theorem 11.1.2 (Schauder [20]) If K is a compact convex subset of a normed linear

space, then every continuous function f mapping K into itself has a fixed point.
Proof Given any > 0, by the compactness of K , we can find a finite number of
points x1 , . . . , x N in K such that each x ∈ K lies in an open ball centred at xi for
some i = 1, 2, . . . , N and of radius . Define g j : K → R+ by

− x − x j , if x − x j <
g j (x) =
0, if x − x j ≥ , j = 1, 2, . . . , N .
Clearly, each g j is continuous and so is h j defined by
g j (x)
h j (x) = N .
i=1 gi (x)

https://doi.org/10.1007/978-981-13-3158-9_11
246 11 Schauder’s Fixed Point Theorem and Allied Theorems

N
Further, h j (x) = 1 ∀ x ∈ K and h j (x) = 0 if x − x j ≥ . The map x → V (x)
j=1
defined by

N
V (x) = h j (x)x j
j=1
maps K into the convex hull of {x1 , x2 , . . . , x N }. Also, for x ∈ K

N
x − V (x) = h j (x)(x − x j )
j=1
where the sum is only over those j for which x − x j < has positive contribution.
Thus for x ∈ K
x − V (x) ≤ h j (x)x − x j < .
If we denote by K , the convex hull of x1 , . . . , x N , then the map x → V ( f (x))

maps continuously K , a compact convex set into itself. So, by Theorem 10.3.7,
V ( f x ) = x , for some x .
Set = n1 for each n ∈ N. Thus we have xn ∈ K such that
V f (xn ) = xn .
Now, by compactness of K , there is a subsequence xn(k) converging to some x ∗ . So,

by triangle inequality and continuity of V f, we have
f (x ∗ ) − x ∗ ≤ f (x ∗ ) − f (xn(k) ) + f (xn(k) ) − V f (xn(k) ) + xn(k) − x ∗
Allowing k to tend to ∞ in the above, it follows by continuity of f that f (x ∗ ) = x ∗ .

Schauder’s theorem as mentioned earlier has far-reaching applications. We merely

sketch the well-known Peano’s existence theorem for initial value problem of first-
order ordinary differential equations. For this purpose, we recall the concept of an
equicontinuous family of functions and Arzela–Ascoli theorem.
Definition 11.1.3 A family F of real-valued functions on a subset S of R is said

to be equicontinuous if given any > 0 there exists δ > 0 (independent of f ∈ F)
such that | f (x) − f (y)| < for all f ∈ F whenever |x − y| < δ, x, y ∈ S.
Theorem 11.1.4 (Arzela–Ascoli) A subset K of the space C[0, 1] with supremum

norm is totally bounded if and only if it is uniformly bounded and equicontinuous.
We make use of the above criterion for compactness in the proof of Peano’s
Theorem.
11.1 Introduction 247
Theorem 11.1.5 (Peano’s existence theorem) Let f : J = [t0 − a, t0 + a] × [x0 −

b, x0 + b] → R be continuous (x0 , t0 ∈ R and a, b > 0). Then the initial value prob-
lem
dx
= f (t, x) and x(t0 ) = x0
dt
has a solution in [t0 − h, t0 + h] for some positive h.
Sketch of Proof. This initial value problem is equivalent to solving the integral equa-
tion t
x(t) = x0 + f (s, x(s))ds.
t0
Since f is continuous on the rectangle J , f is bounded on J . Let K = max(t,x)∈J

| f (t, x)|. Set h = min[a, Kb ]. Clearly, h > 0. Consider the space C[t0 − h, t0 + h]
and define the operator A on this space by
t
Ax = x0 + f (s, x(s))ds.
t0
It is easy to see that A is continuous.

Let S be the closed sphere { f ∈ C[t0 − h, t0 + h] : | f (t) − x0 | ≤ b}. We can
show using the definition of h, that A maps S into itself. Further, A(S) is a uniformly
bounded and equicontinuous family (Prove it!). Then, S ∗ , the smallest closed convex
set containing A(S) (viz., the closed convex hull of A(S) is also uniformly bounded
and equicontinuous. Thus, S ∗ is a convex, compact set (in view of Arzela–Ascoli
Theorem 11.1.4) in C[t0 − h, t0 + h]. Clearly, as A(S) ⊆ S, A(S ∗ ) ⊆ A(S) ⊆ S ∗
(by definition of S ∗ ) ⊆ S.
So, by Schauder’s fixed point Theorem 11.1.2, A has a fixed point in S and this
is a solution of the initial-value problem.
Along similar lines one can prove the following existence theorem.
Theorem 11.1.6 Let F : [0, 1] × [0, 1] × R → R be a continuous function and g :

[0, 1] → R be continuous. Then the following integral equation has a continuous
solution f in [0, 1]. 1
g(s) = f (s) + F(s, t, f (t))dt.
0
Even when a continuous function does not map a compact convex set into itself it
may have a fixed point under some additional assumptions. Rothe [16] obtained the
following fixed point theorem belonging to this category. For this purpose, we need
the following.
Definition 11.1.7 A map T : S(⊆ X ) → X is called compact if T (S) is contained

in a compact subset of X . Here S is a subset of a topological space X .
Theorem 11.1.8 (Rothe [16]) Let X be a normed linear space, B its closed unit
ball and S the unit sphere. Let T : B → X be a compact continuous map such that
T (∂ B) ⊆ B. Then T has a fixed point in B.
Proof Define r : X → B by

x, if x ∈ B
r (x) = x
x
, if x ∈
/ B.
Then r , called the radial retraction maps X continuously onto B. For x ∈ B 0 , r x = x,

while for x ∈ / B, r (x) ∈ ∂ B.
Since T : B → X is continuous r ◦ T maps B continuously into B and is compact.
So by Schauder’s theorem r T has a fixed point x0 . If x0 ∈ ∂ B then T x0 ∈ B so that
x0 = r T x0 = T x0 . If x0 ∈ B 0 , then T x0 ∈ B0 so that x0 = r T x0 = T x0 . If T x0 ∈
∂ B, then r T x0 = T x0 ∈ ∂ B ⊆ B. So in this case too x0 = T x0 . Thus T has a fixed
point.
11.2 An Application of Schauder’s Theorem to an Iterative

Functional Equation
11.2.1 Introduction
In what follows an application of Schauder’s fixed point theorem to the solution

of an iterative functional equation with variable coefficients is described. Solution
of iterative functional equations has been studied by Abel, Babbage, Schroder and
other well-known mathematicians. For the history of this problem, Kuczma et al.
[13] may be consulted. A special case called the iterative root problem arises in the
theory of invariant curves and also in the problem of embedding a function into a
flow in dynamical systems. Murugan and Subrahmanyam [15] used Schauder’s fixed
point theorem to solve a special class of iterative functional equations of finding a
continuous function f : I = [a, b] → R such that
∞

λi (x)Hi ( f i (x)) = F(x), x ∈ I
i=1
under suitable assumption on the functions λi , Hi , i = 1, 2, . . . . This section details

the solution described in [15].
11.2 An Application of Schauder’s Theorem to an Iterative Functional Equation 249
11.2.2 A Subset of C(I, R)
For I = [a, b], a < b, a, b ∈ R let C(I, R) be the Banach space of continuous real-
valued functions on I with supremum norm · . For M ≥ 0, we define
Q(M) = { f ∈ C(I, R) : | f (x) − f (y)| ≤ M|x − y|, for x, y ∈ I }
and for δ ≥ 0,
Fδ (M) = { f ∈ C(I, R) : f (a) = a, f (b) = b, δ(x − y) ≤ f (x) − f (y) ≤ M(x − y),

x, y ∈ I and x ≥ y}
we have
Proposition 11.2.1 ([15]) If M < 1 or δ > 1 Fδ (M) = ∅. If M = 1 or δ = 1, then

Fδ (M) is the singleton containing the identity map.
Proof Let f ∈ Fδ (M). If M < 1, then f (x) − f (y) < x − y for x > y. Setting
x > y = a, it follows that f (x) < x as f (a) = a. This implies that f (b) = b < b, a
contradiction. So Fδ (M) = ∅. A similar argument shows that Fδ (M) = ∅ for δ > 1.
For M = 1 and f ∈ Fδ (M) and x > y, f (x) − f (y) ≤ x − y. For y = a we get
f (x) ≤ f (a) for all x. For x = b, y ≤ f (y) for all y, so that f (x) = x on I . A
similar argument for δ = 1 implies that Fδ (M) is a singleton containing only the
identity function.
Proposition 11.2.2 ([15]) Fδ (M) is a compact convex subset of C(I, R).
Proof Clearly, Fδ (M) is a closed and convex subset of C[I, R]. Also | f (x)| ≤
max{|a|, |b|} for x ∈ I . As δ(x − y) ≤ f (x) − f (y) ≤ M(x − y) for x > y, x, y ∈
I , for f ∈ Fδ (M), Fδ (M) is uniformly bounded and equicontinuous. So by Arzela–
Ascoli Theorem Fδ (M) is compact.
Lemma 11.2.3 If f ∈ Fδ (M) where 0 ≤ δ ≤ 1 ≤ M, then f is a self-

homeomorphism on I and f −1 ∈ F M1 1δ .
Proof f ∈ Fδ (M) is strictly increasing self-map onto I = [a, b]. So f −1 exists

on I and f −1 (a) = a and f −1 (b) = b. So for x ≥ y in I , M1 (x − y) ≤ f −1 (x) −
f −1 (y) ≤ 1δ (x − y). Thus f −1 ∈ FM −1 (δ −1 ).

Lemma 11.2.4 Let f 1 , f 2 be self-homeomorphisms on I and for x ≥ y, x, y ∈ I

δ(x − y) ≤ f i (x) − f i (y) ≤ M(x − y) for some δ > 0 and M > 0 where i = 1, 2.
Then δ f 1−1 − f 2−1 ≤ f 1 − f 2 ≤ M f 1−1 − f 2−1 .
For a proof, Zhang and Baker [24] may be consulted. The following proposition
implies that it suffices to solve the functional equation in [0, 1].
Proposition 11.2.5 f is a solution of the functional equation

∞

λi (x)Hi ( f i (x)) = F(x) for x ∈ I = [a, b]
i=1
if and only if g(x)h −1 ( f h(x)) is a solution for

∞

μi (x)Ri (g i (x)) = G(x), x ∈ [0, 1]
i=1
where h(x) = a + x(b − a), μi (x) = λi (h(x)), Ri (x) = h −1 (Hi (h(x))), G(x) =
∞

h −1 (Fh(x)) and λi (x) ≥ 0 with λi (x) = 1.
i=1
Proof Let f be a solution for

∞

λi (x)Hi (ρ i (x)) = F(x), x ∈ I = [a, b].
i=1
∞

Noting that h and h −1 are affine continuous maps and μi (x) = 1 on [0, 1], for
i=1
x ∈ [0, 1]
∞

n
1
μi (x)Ri (g (x)) = lim
i
μi (x)Ri (g i (x))
n→∞
n
i=1
μi (x) i=1
i=1
1
n
= lim μi (x)h −1 (Hi hh −1 f i (h(x)))
n→∞
n
μi (x) i=1
i=1

n
μi (x)
= lim h −1 (Hi f i (h(x))
n→∞
n
i=1
μ j (x)
j=1

n
= h −1 lim μi (x)Hi f i (h(x))
n→∞
i=1
−1
=h F(h(x)) = G(x).
The converse is straightforward.

Lemma 11.2.6 If f, g ∈ Q(M), M > 1 and map I onto itself, then for i ∈ N
Mi − 1
f i − gi ≤ f − g.
M −1
This can be proved using induction.

We state and prove an existence theorem making use of subsequent lemmata.
Theorem 11.2.7 ([15]) Let λi (x) be a sequence of nonnegative continuous functions
on I = [0, 1] such that λi (x) ∈ Q(αi ) and 1 ≤ γi ≤ λi (x) ≤ i for i = 1, 2, . . . and
∞
λi (x) = 1 for x ∈ I . Let li , L i ≥ 0 and Hi ∈ Fli (L i ), for i = 1, 2, . . . . Suppose
i=1
0 < δ < 1, M > 1 and
∞

i L i M < ∞,
i
(i)
i=1
∞
∞
1
(ii) K 0 = γi li δ i−1 − αi > 0.
i=1
δ i=1
Then for any function F in FK 1 δ (K 0 M), the functional equation
∞

λi (x)Hi ( f i (x)) = F(x), x ∈ I
i=1
has a solution f in Fδ (M) where
∞
αi
K1 = + i Li M
i−1
i=1
δ
For f ∈ Fδ (M), define L f : I → I by
∞

L f (x) = λi ( f −1 (x))Hi ( f i−1 (x)), x ∈ I.
i=1
Lemma 11.2.8 Suppose that in addition to the hypotheses of the above theorem
f ∈ F(M). Then L f ∈ FK 0 (K 1 ) where K 0 and K 1 are as defined in that theorem.
Proof It is easy to see that L f (0) = 0 and L f (1) = 1. For x ≥ y, x, y ∈ I
∞
∞

L f (x) − L f (y) = λi ( f −1 (x))Hi ( f i−1 (x)) − λi ( f −1 (y))Hi ( f i−1 (y))
i=1 i=1
∞

= {[λi ( f −1 (x)) − λi ( f −1 (y))]Hi ( f i−1 (x))
i=1
+ λi ( f −1 (y))[Hi ( f i−1 (x)) − Hi ( f i−1 (x))]
From the definitions of λi , Hi and f and by Lemma 11.2.6 one gets

∞

L f (x) − L f (y) ≤ {αi ( f −1 (x) − f −1 (y)) i Li M
i−1
(x − y)}
i=1
∞
αi
≤ (x − y) + i Li M
i−1
(x − y) = K 1 (x − y)
i=1
δ
Similarly
∞
αi
L f (x) − L f (y) ≥ γi li δ i−1 − (x − y) = K 0 (x − y)
i=1
δ
So L f ∈ FK 0 (K 1 ) and by Lemma 11.2.3 L −1

f ∈ F1/K 1 (1/K 0 ).

Lemma 11.2.9 Besides the hypotheses of Theorem 11.2.7, suppose that f ∈ Fδ (M).
Then L f − L g ≤ K 2 f − g and L −1 −1
f − L g ≤ K 0 f − g, K 2 being
K2
∞
αi i L i (M
i−1
)
+ .
i=1
δ M −1
Proof If f, g ∈ Fδ (M) and x ∈ I

∞

−1 −1
|L f (x) − L g (x)| ≤ [λi ( f (x))Hi ( f (x)) − λi (g (x))Hi (g (x))]
i−1 i−1

i=1
∞

≤ {|λi ( f −1 (x)) − λi (g −1 (x))||Hi ( f i−1 (x))|
i=1
+ |λi (g −1 (x))||Hi ( f i−1 (x)) − Hi (g i−1 (x))|}
From Lemmata 11.2.3 and 11.2.6 and the definitions of λi (x) and Hi (x), one obtains
∞

|L f (x) − L g (x)| ≤ αi | f −1 (x) − g −1 (x)| + i Li f i−1 − g i−1
i=1
∞

αi Mi − 1
≤ + i Li f − g
i=1
δ M −1
Thus L f − L g ≤ K 2 f − g. As f ∈ Fδ (M), for x ≥ y, L −1 −1

f (x) − L f (y) ≤
1
K0
(x − y). So by Lemma 11.2.4, L −1 −1
f − L g ≤ K 0 f − g.
K2

The above lemmata lead to the proof of Theorem 11.2.7.

Proof of Theorem 11.2.7. Define T : Fδ (M) → C(I, R) by T f (x) = L −1
f (F(x)) for
x ∈ I . Clearly, T f (0) = 0 and T f (1) = 1. If x ≥ y, x, y ∈ I
1
T f (x) − T f (y) = L −1 −1
f (F(x)) − L f (F(y)) ≤ [F(x) − F(y)]
K0
≤ M(x − y).
Further,
1 1
T f (x) − T f (y) ≥ |F(x) − F(y)| ≥ K 1 δ(x − y) = δ(x − y).
K1 K1
Thus T f ∈ Fδ (M) and maps Fδ (M) into itself. For f, g ∈ Fδ (M) and x ∈ I
|T f (x) − T g(x)| = |L −1 −1
f (F(x)) − L g (F(x))|
≤ L −1 −1
f − L g (x)
Thus T f − T g ≤ KK 20 f − g in view of Lemma 11.2.8. By Proposition 11.2.2,

Fδ (M) is a compact convex subset of C[I, R]. So by Schauder fixed point theorem
T has a fixed point in Fδ (M) which is a solution of the functional equation.
11.3 Measures of Noncompactness and Fixed Point

Theorems
Given a noncompact space, a natural question is to find out if noncompactness can

be measured. In a complete metric space a measure of noncompactness can be for-
mulated in a natural way since compact subsets of such a space are precisely totally
bounded closed subsets. Such a measure, called Kuratowski’s measure of noncom-
pactness is the following.
Definition 11.3.1 Let (M, d) be a complete metric space and X a bounded subset
of M. Kuratowski measure of noncompactness of X denoted by α(X ) is defined by
α(X ) = inf{ > 0 : X can be covered by a finite

number of sets of diameter less than }.
Using the above definition, the following proposition can be proved easily.
Proposition 11.3.2 Let (M, d) be a complete metric space, X, Y bounded subsets

of M and α(·) the Kuratowski measure of noncompactness function. Then
(i) α(X ) = 0 if and only if X is compact;

(ii) α(X ) = α(X );
(iii) X ⊆ Y ⇒ α(X ) ≤ α(Y );
(iv) α(X ∪ Y ) = max{α(X ), α(Y )};
(v) α(X ∩ Y ) ≤ min{α(x), α(Y )}.
Proposition 11.3.3 Let (X n ) be a non-increasing sequence of nonempty bounded
closed subsets of a complete metric space (M, d) such that lim α(X n ) = 0. Then
n→∞
∞

X n is nonempty and compact.
n=1
Proof Define a sequence xn ∈ X n for each n ∈ N and a set Ak = {xn : n ≥ k} for

k ∈ N. Clearly, An ⊆ X n ∀ n and α(A1 ) = α(An ) = α(An ) in view of (i), (ii), (iii)
and (iv). So α(A1 ) = 0. So α(A1 ) = 0. Thus A1 is compact. As every cluster point
of A1 is a cluster point of An and α(An ) = α(An ) = 0, An are all compact. Since
∞

An and hence {X n ⊇ An : n ∈ N} has finite intersection property An and hence
n=1
∞

X n = φ.
n=1
Remark 11.3.4 Proposition 11.3.3 is a generalization of the Cantor intersection the-

orem.
If the metric space (M, d) is a Banach space with the metric d arising from the
norm , then the (Kuratowski) measure of noncompactness satisfies additional
properties. These are collected in
Proposition 11.3.5 If (E, · ) is a Banach space, the Kuratwoski measure of non-
compactness satisfies the additional properties stated below:
(vi) α(X + Y ) ≤ α(X ) + α(Y );
(vii) α(cX ) = |c|α(X ); for c ∈ R;
(viii) α(conv X ) = α(X );
where X, Y are bounded subsets of E.

n
m
Proof We prove only (viii). For x = ti xi and y = s j y j for xi , y j ∈ X
i=1
j=1

n m

n
m
m

and ti , s j ≥ 0 with ti = s j = 1, x − y = ti xi − sj yj
= sj
j=1
i=1
j=1 i=1 j=1

n n m

ti xi − ti yj
≤ s j ti xi − y j ≤ diam X . Consequently α(conv X )
i=1 i=1 j=1 i, j
≤ α(X ). We get α(conv X ) ≤ α(X ). As X ⊆ conv X , α(X ) ≤ α(conv X ). Thus
α(X ) = α(conv X ).
11.3 Measures of Noncompactness and Fixed Point Theorems 255
Remark 11.3.6 When X is compact, the above result leads to Mazur’s theorem that
the closed convex hull of a compact set is compact. The calculation of the measure
of noncompactness is not always easy. In fact, the proof that α(B(0, 1)) = 2 for the
unit open ball B(0, 1) in a Banach space is not obvious.

Remark 11.3.7 α t X = t0 α(X ). This may be deduced from tX ⊆
0≤t≤t0 0≤t≤t0
conv[t0 X ∪ {0}].
Remark 11.3.8 One may alternatively cover a bounded subset X of a metric space
(X, d) by a finite number of open balls of radius smaller than > 0. This leads to the
Hausdorff (or ball) measure of noncompactness of X . This has been used by Gohberg,
Goebel, Nussbaum and others. This measure is closely related to the concept of
Hausdorff metric and is denoted by χ . It can be proved that ψ(X ) = inf{H (X, F) : F
a nonempty compact subset of M and H the Hausdorff distance on the space of all
closed nonempty bounded subsets of M}.
It is, therefore, possible to propose an abstract concept of a measure of noncom-
pactness based on an axiomatic approach.
Definition 11.3.9 A nonempty subfamily P of the family N of nonempty relatively
compact subsets of E is said to be the kernel (of a measure of noncompactness) if
the following conditions are satisfied:
(i) X ∈ P ⇒ X ∈ P;
(ii) X ∈ P, φ = Y ⊆ X ⇒ Y ∈ P;
(iii) X, Y ∈ P ⇒ λX + (1 − λ)Y ∈ P for λ ∈ [0, 1];
(iv) the family of compact sets in P is closed in the family of nonempty compact
sets with Hausdorff metric.
Definition 11.3.10 Let M be the family of non-void bounded sets of E. A function
μ : M → [0, ∞) is said to be a measure of noncompactness with the kernel P if it
satisfies the following:
(i) μ(X ) = 0 ⇔ X ∈ P;
(ii) μ(X ) = μ(X );
(iii) X ⊆ Y ⇒ μ(X ) ≤ μ(Y );
(iv) μ(conv X ) = μ(X );
(v) μ(λX + (1 − λ)Y ) ≤ λμ(X ) + (1 − λ)μ(Y ) for λ ∈ [0, 1];
(vi) for X n ∈ M and X n = X n and X n+1 ⊆ X n for all n ∈ N and lim μ(X n ) = 0
n→∞
∞

imply X n = φ.
n=1
We skip the proof of the following.

Theorem 11.3.11 For any kernel P the function μ(X ) = inf{H (X, P) : P ∈ P} is
a measure of noncompactness with kernel P.
Remark 11.3.12 If P is N, the family of nonempty relatively compact subsets of E

then the kernel is full and the measure is said to be complete. The Kuratowski and
Hausdorff measures are full. Given a nontrivial closed subset V of the Banach space
E, for a nonempty bounded subset X of E, μ(X ) = χ (X ) + d(X, F), where
d(X, F) = inf{ > 0 : X ⊆ N (F; )}
is a measure of noncompactness on the family of non-void compact sets X . X

and diam(X ) are also measures. The kernels are respectively {θ } and the family of
singletons.
Definition 11.3.13 A measure μ is said to have maximum property if μ(X ∪ Y ) =

max(μ(X ), μ(Y )) for all x, y ∈ P. If μ(λX ) = |λ|μ(X ) for all scalars λ, μ is called
homogeneous and if μ(X + Y ) ≤ μ(X ) + μ(Y ), μ is called subadditive. (A homo-
geneous subadditive measure is also called sublinear measure.)
A measure of noncompactness is called regular if it is full, sublinear and has

maximum property. For further details and applications Banas and Goebel [1] may
be referred.
We now state and prove Darbo’s fixed point theorem [4].
Theorem 11.3.14 (Darbo [4]) Let C be a non-void closed convex set which is
bounded and μ be any measure of noncompactness with the kernel of the family
of bounded nonempty convex subsets of E. Let T : C → C be a μ-contraction in the
sense that μ(T (X )) ≤ kμ(X ) for any nonempty convex bounded subset X , where
0 ≤ k < 1. If T is continuous, then T has a fixed point and the set of fixed points of
T in C belongs to the kernel of μ.
Proof Define C1 = C, Cn+1 = conv(T Cn ) for n ≥ 1. Now μ(Cn+1 ) = μ(con

(T Cn )) = μ(T Cn ) ≤ kμ(Cn ) ≤ k n−1 μ(C) for all n ≥ 1. As 0 < k < 1, proceed-
∞

ing to the limit as n → ∞, it follows that C ∗ = Cn = φ is in the kernel of μ and
n=1
μ(C ∗ ) = 0. Further, C ∗ is convex and closed. As C ∗ is precompact, C ∗ is a non-void
compact convex subset of the Banach space E. Also T C ∗ ⊆ T Cn ⊆ (con(T Cn )) =
∞

Cn+1 for all n ≥ 1. So T C ∗ ⊆ Cn+1 ∩ C1 = C ∗ . Thus T maps C ∗ into itself and
n=1
is continuous. So by Schauder’s fixed point theorem T has a fixed point in C. If F
is the set of fixed points of T in C, then μ(F) = μ(T (F)) ≤ kμ(F). So μ(F) = 0.
Since F is closed, F is compact and so belongs to the kernel of μ.
Corollary 11.3.15 Let C be a non-void bounded closed convex subset of a Banach

space and T be a contraction on C with Lipschitz constant k < 1 and S a continuous
map mapping bounded subsets of C into precompact (totally bounded) subsets of C
such that F = T + S maps C into itself. Then F has a fixed point.
11.3 Measures of Noncompactness and Fixed Point Theorems 257
Proof Let α be the Kuratowski measure of noncompactness and B a bounded non-

void subset of C. Then α(F(B)) = α(T + S)(B)) ≤ αT (B) + αS(B) = kα(B) as
α(S(B)) = 0. Thus F is a α-set contraction and so by Darbo’s Theorem 11.3.14 has
a fixed point.
√ t cos(x( 2 ))
s+1
Example 11.3.16 The functional equation x(t) = 34 x( t) + 0 4
ds for t ∈
[0, 1] has a solution in C[0, 1]. Let B be the √ closed unit ball in C[0,
t 1] and T and
S the maps on B defined by T x(t) = 43 x( t) and S(x(t)) = 41 0 cos(x( s+1 2
))ds.
3
Clearly, T is a contraction on B with
Lipschitz constant
4 while S is a compact oper-
t 1
ator since |Sx(t1 ) − Sx(t2 )| ≤ 14 t12 cos(x( s+1
2
))ds ≤ 4 |t1 − t2 | for t1 , t2 ∈ [0, 1].
Further, (T + S)x ∈ B for x ∈ B. So by Corollary 11.3.15. F = T + S has a fixed
point in B ⊆ C[0, 1]. Thus the above functional equation has a solution in C[0, 1].
Sadovskii [19] obtained a generalization of Darbo’s theorem using the concept of
a condensing operator.
Definition 11.3.17 Let λ be a measure of noncompactness with a kernel P. Let C be
a bounded non-void closed convex subset of a Banach space E. A map T : C → E
is called condensing if λ(T (A)) < λ(A) for all A ⊆ C with λ(A) > 0.
Theorem 11.3.18 (Sadovskii [19]) Let C be a nonempty bounded closed convex
subset of a Banach space and T : C → C be a continuous condensing operator.
Then T has a fixed point, provided λ has maximum property.
Proof Let c ∈ C. Denote by F the set of all closed convex subsets K of C such
that c ∈ K and T (K ) ⊆ K . Define B = K and C ∗ = con(T (B) ∪ {c}). Clearly,
K ∈F
F = φ as C ∈ F and B = φ
as c ∈ B by definition of F . Since T (B) ⊆ T (K ) ⊆ K
for each K ∈ F , T (B) ⊆ K = B. Thus T maps B into itself. We claim that
K ∈F
B = C ∗ . As c ⊂ B and T (B) ⊆ B, B ∈ F , C ∗ ⊆ B. So T (C ∗ ) ⊆ T (B) ⊆ C ∗ . So
C ∗ ∈ F and by definition of B, C ∗ ⊇ B. Thus B = C ∗ . By the properties of λ,
the measure of noncompactness, λ(B) = λ(C) = λ{T (B) ∪ {c}} = λ(T (B)). This
implies λ(B) = 0 as T is condensing. So B is compact. Since T is continuous, B is
nonempty compact and convex and T (B) ⊆ B, T has a fixed point by Schauder’s
fixed point theorem.
Remark 11.3.19 Let B be the closed unit ball in the Hilbert space of real sequences
2 . The operator F : B → B defined by

F(x) = F(x1 , . . . , xn , . . . ) = ( 1 − x2 , x1 , x2 , . . . )

= T (x) + S(x) = ( 1 − x2 , 0, 0, . . . ) + (0, x1 , x2 , . . . )
for x = (x1 , x2 , . . . , xn , . . . )
Clearly, S, the shift operator is an isometry and T is a compact operator. Fur-

ther, α(F(B)) = α(T (B) + S(B)) ≤ α(S B) = α(B). So F is 1-set contraction with
respect to α has no fixed point. Thus Darbo’s theorem or Sadovskii’s theorem is not
true for 1-set contractions.
A fixed point theorem for mappings satisfying the so-called Leray-boundary con-
dition involving measures of noncompactness can also be proved.
Theorem 11.3.20 Let μ be a measure with kernel P having maximum property such
that {θ } ∈ P. Suppose C is an open and bounded neighbourhood of θ of a Banach
space E and T : C → E is a continuous k-set contraction with contractive constant
k (< 1) such that for any x ∈ ∂C, T x = λx for λ > 1. Then T has a fixed point in
C and the set of fixed points of T belongs to P.
Proof Let K = {x ∈ C : x = cT x for some c ∈ [0, 1]}. K is nonempty as θ ∈ K
and is obviously closed. As K ⊆ conv(T K ∪ {θ }) we have μ(K ) ≤ μ(T (K ) ∪
{0}) = μ(T (K )) ≤ kμ(K ) and K ∈ P. Clearly, K ∩ ∂C = φ. Since K is compact
and E − C is non-void closed set and disjoint from K , by Urysohn’s Lemma there
is a continuous function g : E → [0, 1] such that g(x) = 1 for x ∈ K and g(x) = 0
for x ∈/ C and 0 < g(x) < 1 for x ∈ C − K . Define the map F : E → E by

g(x)T (x), for x ∈ C
F(x) =
0, for x ∈
/C
Clearly, F maps each ball B(θ ; r ) (r > 0) containing C into itself. For any set X ,
F(X ) ⊆ conv(T (X ∩ C) ∪ {θ }) and so
μ(F(X )) ≤ μconv(T (X ∩ C) ∪ {θ }) = μ(T (X ∩ C))

≤ μ(T (X )) ≤ kμ(X ).
As T is continuous on C. F is continuous on the closed ball B(θ ; r ) containing C

and is a k-set contraction. So by Darbo’s fixed point Theorem 11.3.14. F has a fixed
point x0 in B(θ ; r ). Clearly, x0 = θ and is not in the complement of C. So x0 ∈ C.
If x0 ∈ C − K , then 0 < g(x0 ) < 1 and x0 = g(x0 )T (x0 ) contradicts that x0 ∈ /K
(by the definition of K ). So x0 ∈ K . So g(x0 ) = 1 and F(x0 ) = g(x0 )T (x0 ) = x0
T (x0 ) = x0 . Thus x0 is a fixed point of T . That the set of fixed points of T in C is
compact is left as an exercise.
11.4 Kakutani-Ky Fan–Glicksberg Fixed Point Theorem
Kakutani [10] generalized Brouwer’s fixed point theorem to multivalued functions on

Euclidean spaces. Inspired by this theorem. Glicksberg [8] and Ky Fan [7] extended
it to topological linear spaces. We need the following.
Definition 11.4.1 Let X and Y be topological spaces and 2Y denote the set of all
subsets of Y . Let T be a multivalued function mapping X into 2Y − {φ}. T is said to
11.4 Kakutani-Ky Fan–Glicksberg Fixed Point Theorem 259
be upper semicontinuous if for each x0 ∈ X and any neighbourhood W of T (x0 ) in

Y , there exists a neighbourhood V of x0 such that x ∈ V implies T (x) ⊆ W .
Theorem 11.4.2 (Ky Fan–Glicksberg [7, 8]) Let K be a non-void compact convex
subset of a Hausdorff locally convex linear topological space X . Let T : K → 2 K be
an upper semicontinuous multivalued function such that T (x) is a nonempty closed
convex subset of K for each x in K . Then there exists x0 ∈ K such that x0 ∈ T x0 .
The following proof due to Terkelsen [21] employs a technique of Browder [2].
Proof Let {Ui : i ∈ I } be a neighbourhood base at 0 in X comprising open convex
sets. For each i ∈ I , there exists a finite set {xi j : j ∈ J (i)} ⊆ K such that
circled
K ⊆ (xi j + Ui ). Now there is a continuous partition of unity subordinate to this
j∈J (i)
covering, i.e. for j ∈ J (i), continuous functions
αi, j : K → R exist with αi j (x) ≥ 0
for x ∈ K , αi j (x) = 0 for x ∈/ xi j + Ui and αi j (x) = 1 for x ∈ K . Let yi j ∈
j∈J (i)

T (xi j ) and define f i : K → X by f i (x) = αi j (x)yi j . Let Ci be the convex hull
j∈J (i)
of {yi j : j ∈ J (i)}. Ci is homeomorphic to a closed convex set in a finite-dimensional
euclidean space, where Ci ⊆ K . So by Brouwer’s fixed point theorem f i has a fixed
point xi in Ci .
Let x0 ∈ K be the cluster point of the net {xi : i ∈ I } directed by {Ui , ⊃: i ∈ I }.
Suppose x0 ∈ / T x0 . By separation theorem there exists a closed convex neigh-
bourhood W of T x0 with x0 ∈ / W . As T is upper semicontinuous we can find a
neighbourhood V of x0 with V ∩ W = φ such that x ∈ V ∩ K implies T (x) ⊆ W .
Let m ∈ I be for an open set Um θ such that Um + Um ⊂ V − x0 . We can find
i ∈ I such that Um ⊃ Ui , xi ∈ x0 + Um and xi + Ui ⊂ V . Now for j ∈ J (i) with
αi j (xi ) = 0, xi
∈ xi j + Ui . So xi j ∈ V . This implies that yi j ∈ W . Consequently
xi = f i (xi ) = ai j (xi )yi j ∈ W contradicting that xi ∈ V and V is disjoint from
j∈J (i)
W . Hence x0 ∈ T (x0 ) for some x0 ∈ K .
Corollary 11.4.3 (Kakutani [10]) Let K be a compact convex nonempty subset of
Rn and T : K → 2 K be an upper semicontinuous multivalued function such that
T (x) is a nonempty compact convex subset of K for each x ∈ K . Then T has a fixed
point.
Corollary 11.4.4 (Tychonoff [22]) Let X be a linear topological space which is
both locally convex and Hausdorff and K a nonempty compact convex subset of X .
If T : K → K is continuous then T has a fixed point in K .
Proof The map F : K → 2 K defined by F(x) = {T (x)} satisfies all the hypothesis
of Theorem 11.4.2. So T has a fixed point in K .
Cauty [3] has shown that Tychonoff (Schauder) Theorem is true in arbitrary linear
topological spaces. Tychonoff’s theorem may be used to deduce a common fixed point
theorem for a commuting family of affine continuous mappings, due to Markov [14]
and Kakutani [11].
Definition 11.4.5 A mapping f on a convex set C into a linear space is called affine
if f (αx + (1 − α)y) = α f (x) + (1 − α) f (y) for all x, y ∈ C and α ∈ [0, 1].
Theorem 11.4.6 Let K be a nonempty compact convex subset of a Hausdorff locally

convex topological vector space and F be a commuting family of affine continuous
functions mapping K into itself. Then there is a common fixed point for mappings of
F in K .
Proof By Tychonoff’s theorem (Corollary 11.4.4), F(T ), the set of fixed points, is
nonempty for each T ∈ F . As T is affine, F(T ) is closed and convex as well. If
S ∈ F , then S maps F(T ) into itself by the commutativity of S and T and again
by Tychonoff’s Theorem has a fixed point in F(T ). So F(T ) ∩ F(S) = φ. In fact,
the family {F(T ) : T ∈ F } is a family of nonempty compact convex subsets of K
with finite intersection property. Since K is compact, ∩{F(T ) : T ∈ F } = φ. Any
element in this intersection is a common fixed point for mappings in F .
Kakutani [11] has also given a more elementary proof of this theorem and it is
described below.
Proof ((Aliter) (Kakutani [11])) Given a commuting family A of continuous self-

maps on a non-void compact convex subset of a locally convex linear topological
space, it is clear that for T ∈ A, Tn defined by
1
T(n) (x) = (x + t x + · · · + T n−1 (x))
n
for x ∈ M and n > 1. T k being the kth iterate of T is affine, maps M into
itself and commutes with each U ∈ A. Further, a fixed point of T is also a
fixed point of T(n) for all n > 1. So we may without loss of generality assume
that A is a semigroup containing such convex combinations of the iterates of T .
If S and T are in A then S(T (M)) ⊆ S(M) ⊆ M and S(T (M)) = T (S(M)) ⊆
T (M) ⊆ M. So ST (M) ⊆ S(M) ∩ T (M) ⊆ M. Since ST ∈ A ST (M) = φ. Thus
∩ T (M) = φ. If therefore follows that if A1 is a
S(M) finite subset of A, then
U (M) = φ. Since U (M) is compact for each U ∈ A, T (M) = φ. Let x ∗ ∈
U ∈A1 T ∈A

T (M). Then x ∗ = T(n) y for some y ∈ M. So T x ∗ − x ∗ = n1 (T n y − y) ∈ n1 M1
T ∈A
where M1 = {x − y : x, y ∈ M}. As M is compact, M1 is compact and bounded.
Since n ∈ N is arbitrary, T x ∗ = x ∗ . T being an arbitrary element of A, it follows
that x ∗ is a common fixed point for all T ∈ A.
As an application Hahn–Banach theorem on the extension of linear functionals

on a linear subspace of a locally convex linear topological space can be deduced.
Theorem 11.4.7 Let E be a linear space and f a linear functional defined on a

linear subspace E 0 of E. Let p : E → R be a function such that p(t x) = t p(x) for
t ≥ 0 and p(x + y) ≤ p(x) + p(y) where x, y ∈ E. Suppose f (x) ≤ p(x) for any
x ∈ E 0 . Then there exists a linear functional F on E such that p(F(x)) ≤ p(x) for
x ∈ E and F(x) = f (x) for x ∈ E 0 .
Proof Let B = {F ∈ R E : − p(−x) ≤ F(x + y) − F(y) ≤ p(x) for all x, y ∈ E

and F(x) = f (x) for x ∈ E 0 }. Since B is a closed subset of [− p(x), p(x)] in the
x∈E
product topology of [− p(x), p(x)], B is compact. Note that R E with the product
x∈E
topology is a Hausdorff locally convex linear topological space. The convexity of
B is also clear. Let be the group of all linear transformations on B generated by
St and Ty , t ∈ R, y ∈ E defined by St (F(x)) = F(tt x) , Ty F(x) = F(x + y) − F(y).
Clearly it is abelian. So by Kakutani’s common fixed point theorem it has a fixed
point F which coincides with f on E 0 , F(x) ≤ p(x) for x ∈ E and F(t x) = t F x
as St F = F for all t and F(x + y) − F(y) = F(x) as Ty F = F for all y. Thus F
is a linear extension of f to E.
Werner [23] has proved that Hahn–Banach theorem implies Markov–Kakutani

theorem. To this end, we use the Hahn–Banach theorem in the form of separation
theorem, to prove the following lemma leading to Markov–Kakutani theorem.
Lemma 11.4.8 ([23]) Let T : K → K be a continuous affine map on K , a compact

convex subset of a locally convex Hausdorff linear topological space E. Then T has
a fixed point in K .
Proof Suppose the lemma is false. Then the diagonal = {(x, x) : x ∈ K } and
the graph of T , viz = {(x, T x) : x ∈ K } are disjoint compact convex subsets of
E × E. So by the Hahn–Banach theorem there exist continuous linear functional 1
and 2 on E and real numbers α and β such that
1 (x) + 2 (x) ≤ α < β < 1 (y) + 2 (T (y))
for all x, y ∈ K . So 2 (T (x)) − 2 (x) ≥ β − α for all x ∈ K . Since 2 (T n (x)) −

2 (T n−1 (x)) ≥ β − α for all n > 1, it follows 2 (T n (x)) − 2 (x) ≥ n(β − α) for
all x ∈ K . This implies that 2 (T n (x)) → +∞ contradicting that 2 (K ) is compact.
So ∩ = φ or T has a fixed point in K .
We can now deduce Markov–Kakutani theorem from the preceding lemma.
Proof Let Tλ : K → K be an affine continuous map on a compact convex subset of

a Hausdorff locally convex linear topological space for each λ ∈ . Further, Tλ Tμ =
Tμ Tλ for any λ, μ ∈ . By Lemma 11.4.8 K λ , the set of fixed points of Tλ is non-void
for each λ ∈ . K λ is compact being a closed subset of K and is convex as Tλ is
affine for each λ ∈ . Let k ∈ K μ . Then Tμ (k) = k. Also Tλ k = Tλ Tμ k = Tμ (Tλ k).
So Tλ (k) is a fixed point of Tμ . So Tλ (K μ ) ⊆ K μ and so Tλ has a fixed point K μ . Thus
K λ ∩ K μ = φ. From the principle of finite induction it follows that every

non-void
∗
finite subfamily of {K λ : λ ∈ } has non-void intersection. So K = K λ = φ.
λ∈
Clearly, every point if K ∗ is a fixed point of Tλ , λ ∈ .
Markov–Kakutani theorem also provides other applications, for example the exis-
tence of an invariant mean and the existence of Banach limits. These are described
below.
Definition 11.4.9 Let P(X ) be the set of all Borel probability measures on a compact
Hausdorff space X . Let μ ∈ P(X ). A μ-measurable map f : X → X is called mea-
sure preserving with respect to μ if μ(B) = μ( f −1 (B) for every Borel set B ⊆ X .
In this case μ is called an invariant measure for f .
Remark 11.4.10 For a compact Hausdorff space the dual of C(X ) can be identified
with M(X ), the space of complex regular Borel measures on X , (in view of the Riesz
Representation theorem) with μ = total variation of μ.
If f : X → X is continuous, then f˜μ defined by f˜μ (B) = μ( f −1 (B)) for each
Borel set B defines a probability measure on X .
We have
Lemma 11.4.11 If f : X → X is continuous on a compact Hausdorff space, then

the map f˜ : P(X ) → P(X ) defined by f˜μ (B) = μ( f −1 (B)), B any Borel set is
continuous in the Weak∗ topology.

Proof For g ∈ C(X ), X gd f˜μ = X g ◦ f dμ is well-defined for g and an applica-
tion of monotone convergence theorem to an increasing sequence of nonnegative
simple functions clarifies that it is well-defined for all g ∈ C(X ). If {μλ }λ∈ is a net
converging to μ ∈ P(X ), then for every g ∈ C(X ) it can be seen that

lim gd( f˜μ )λ = lim (g ◦ f )dμλ
λ∈ λ∈
X
X

= (g ◦ f )dμ = gd( f˜μ )
X X
Hence we have
Theorem 11.4.12 For f ∈ C(X ), there exists μ ∈ P(X ) for which f is measure
preserving.
Proof The map f˜ : P(X ) → P(X ) defined by f˜μ (B) = μ( f −1 (B)) for any Borel
subset of X is continuous in the weak∗ topology by Lemma 11.4.11 and P(X ) is a
convex and closed subset of the closed unit sphere of M(X ). M(X ) is compact by
Alaoglu’s theorem. So by Tychonoff’s theorem f˜ has a fixed point μ∗ in P(X ) and
μ∗ (B) = f˜μ∗ (B) = μ∗ ( f −1 (B)) or f is measure preserving with respect to μ∗ .
The next application insures the existence of an invariant mean in a semigroup.

Definition 11.4.13 Let (S, ◦) be a semigroup and B(S) be the real Banach space
of all bounded real functions in S with the supremum norm. For t ∈ S, the left-
translation operator L t : B(S) → B(S) is an operator defined by
(L t f )(s) = f (t ◦ s) for s ∈ S
where ◦ is the associative binary operation on S.

Definition 11.4.14 A left-invariant mean on a semigroup (S, ◦) is a positive linear
functional μ on B(S) satisfying the following conditions:
(i) μ(1) = 1; (ii) μ(L s f ) = μ( f ) ∀ s ∈ S and f ∈ B(S).
If a semigroup has a mean, then the semigroup is called amenable (one can define a
right-invariant mean similarly and in an abelian semigroup both the means coincide).
The next theorem is due to Day [5].
Theorem 11.4.15 Let S be an abelian semigroup. Then S is amenable.
Proof Let K = { ∈ (B(S))∗ : = (1) = 1}. Clearly, K is convex and closed
subset of the closed unit ball in the weak∗ topology of B(S). Since the closed
unit ball in the weak∗ topology is compact by the Banach–Alaoglu theorem. K is
compact and convex. It is nonempty by Hahn–Banach Theorem. Define the fam-
ily of linear operators Ts : B(S)∗ → B(S)∗ by Ts (λ)( f ) = (L s ( f )) for s ∈ S.
If V = { ∈ B(S)∗ : | f i | 0, i =
1, 2, . . . , n, then Ts−1 (V ) = { ∈ B(S)∗ : |Ts ( f i )| < i , i = 1, 2, . . . , n} = { ∈
B(S)∗ : | (L s f i )| < i , i = 1, 2, . . . , n}. So Ts−1 (U ) is a neighbourhood of zero
even as V is a neighbourhood of zero in B(S) with the weak∗ topology.
If s ∈ K , then Ts (1) = L s (1) = 1 = 1 and
TS = sup |TS ( )( f )| = sup | (L S ( f ))|

f ≤1 f ≤1
≤ sup | f | = = 1
f ≤1
as L s f ≤ f . So TS (K ) ⊆ K . Further,
Ts Tt ( ) = Ts ( ◦ L t ) = ( ◦ L t ) ◦ L s
= ◦ L st = ◦ L ts
= Tt Ts ( ) ∀ ∈ B(S)∗ .
Thus {Ts } is a commuting family of linear operators mapping the compact convex
set into itself and so has a fixed point ∗ in K . Clearly, ∗ is left-invariant (and
right-invariant as S is commutative) and ∗ L s (1) = ∗ 1 = 1. We now show that
each element of K is positive. Suppose not. Then there exists f ∈ B(S), f ≥ 0 such
that for some ∈ K , f = β < 0. So for small > 0, we get
1 − f = sup |1 − f (s)| ≤ 1 (as f > 0)

s∈S
So
1 < 1 − β ≤ |1 − β| = | (1 − f )|
≤ 1 − f ≤ 1,
a contradiction. So each element if is positive. Thus ∗ is a left-invariant functional

which is positive with 1 = ∗ (1) = ∗ . Thus ∗ is an invariant mean on S and
S is amenable.
The existence of a generalized limit or Banach limit for bounded sequences can
also be deduced from fixed point theorems.
We can also deduce the existence of a generalized or Banach limit of a bounded
sequence from Tychonoff’s theorem.
Definition 11.4.16 A generalized or Banach limit of a bounded sequence of reals

a = (an ) is a real number L(a) satisfying the following:
(i) L is a linear functional on ∞ , the linear space of bounded sequences;
(ii) L(1, 1, . . . ) = 1;
(iii) L(a) ≥ 0 if a ≥ 0
(iv) L(a1 , a2 , . . . ) = L(a2 , a3 , . . . )
Remark 11.4.17 If a = (an ) ∈ m, then inf an ≤ L(a) ≤ sup an . Let m = inf an and
M = sup an . Clearly, m ≤ an ≤ M for n. So an − m ≥ 0. So L(an − m) ≥ 0, by
(iii). Since L(an − m) = L(an − m(1)) = L(an ) − m L(1, 1, . . . ) ≥ 0. So L(an ) ≥
m L(1) = m by (ii). Similarly L(an ) ≤ L(M) = M L(1, 1, . . . ) = M. Thus inf an ≤
L(a) ≤ sup an . This in turn implies |L(a)| ≤ sup |an | = a. Since L(1) = 1, L =
1. From (iv) and inf(an ) ≤ L(a) ≤ sup(an ) it follows that lim(an ) ≤ L(a) ≤ lim(an )
for a = (an ) ∈ ∞ (v). Also (i) and (v) imply (ii) and (iii).
Theorem 11.4.18 A generalized (or Banach) limit of a bounded sequence always

exists.
Proof Define K = {L ∈ m ∗ : L satisfies (i), (ii), (iii) and (iv) of Definition 11.4.16}.
Clearly, K is non-void as L(a) = a1 for a = (an ) lies in K . Also K is con-
vex. Since K = ∩{L ∈ m ∗ : L(a) ≤ sup an } ∩ {L ∈ m ∗ : L(a) ≥ inf an } and {L :
L(a) ≤ ξ } and {L : L(a) ≥ η} are weak∗ closed in m ∗ , K is weak∗ closed. As
L = 1, K is a weak∗ closed subset of the unit ball in m ∗ ; K is weak∗ compact by
the Banach–Alaoglu theorem. K is thus compact and convex. Define the map T :
K → K by T (a) = T L(a) = L(a2 , a3 , . . . ) for a = (a1 , a2 , . . . ) T ∈ m ∗∗ . For a ∈
m and Q ∈ m ∗∗ , T −1 (N (Q; a)) = {L : T L ∈ N (Q, a)} = {L : |T L(a) − Q(a)| <
1} = {L : |L(a2 , a3 , . . . ) − Q(a1 , . . . , an , . . . )| < 1} = N (Q 1 ; (a2 , a3 , . . . )), Q 1
being in m ∗ for which Q 1 (a2 , a3 , . . . ) = Q(a1 , . . . ).
Thus T is continuous on K with the weak∗ topology. So by Tychonoff’s theorem

T has a fixed point L 0 ∈ K such that T L 0 = L 0 . Thus there is a generalized (Banach)
limit on m, satisfying (i)–(iv).
Kakutani [11] proved another related fixed point theorem and is stated below.
Theorem 11.4.19 Let K be a compact convex subset of a locally convex space and
let F be a group of affine transformations of K into itself. If F is equicontinuous on
K (in the sense that for each neighbourhood N of zero, there is a neighbourhood V
of origin with the property that T x − T y ∈ N for all T ∈ F whenever x − y ∈ V ),
then F has a common fixed point.
The above theorem due to Kakutani follows from a more general theorem of Ryll-
Nardzewski and its proof discussed in the following is based on the ideas of Hahn
[9]. To this end, we need the following.
Definition 11.4.20 A family F of functions mapping a topological space X into
itself is called distal it for every pair x, y of distinct points in X , there is an open cover
{G α : α ∈ } of X such that F(y) ∈ / ∪{G α : F(x) ∈ G α } for each F ∈ F. (In other
words for x = y the set {(F x, F y) : F ∈ F is disjoint from some neighbourhood
∪{Vα × Vα : α ∈ } of the diagonal.)
Theorem 11.4.21 Let K be a non-void compact convex subset of a locally convex
space E and F a semigroup of affine continuous self-maps on K . If F is distal in
each minimal F-invariant closed set in K , then F has a common fixed point.
Proof Let K be the family of all non-void compact convex subsets which are F
invariant. Since K ∈ K , K is nonempty. K is partially ordered by set inclusion and
every chain K α has the lower bound ∩K α ; by Zorn’s lemma there is a minimal element
K 0 ⊆ K in K . Let K0 be the family of all non-void compact subsets of K 0 that are F
invariant. As before an application of Zorn’s lemma insures the existence of a closed
x+y If x, y, x = y are
minimal F invariant subset S0 of K 0 . We claim that S0 is a singleton.
in S0 , then x+y
2
∈ K 0 and K 0 is F invariant. Moreover A = F 2 : F ∈ F ⊆ K0
and A ⊆ K 0 is compact. Also A is F invariant. As each F is affine, conv A ⊂ K 0
is also F invariant and so by the minimal property of K 0 , con A = K 0 . Let z be
an extreme
point of K 0 which exists by the Krein–Milman theorem in A. So z =
x+y
lim Fλ for some net. Fλ (x) and Fλ (y) belong to S0 and S0 is compact.
λ 2
So we may assume without loss of generality Fα x → s ∈ S0 and Fα y → t ∈ S0 so
1 1
that = lim [Fα x + Fα y] = (s + t). As z is an extreme point of K 0 , z = s = t.
2 2
So for each open covering {G α } of S0 any set G α0 containing s will contain all the
Fα x, Fα y eventually. So F is not distal in the closed minimal F invariant set S0 .
This contradiction to the hypothesis implies that S0 must be a singleton F invariant
set. Thus F has a common fixed point.
Corollary 11.4.22 (Hahn [9]) Let K be a compact convex non-void subset of a
locally convex topological vector space and F a semigroup of affine continuous
self-maps on K . If F is distal on K , then F has a common fixed point.
Following Dugundji and Granas [6], we state and prove
Theorem 11.4.23 (Ryll-Nardzewski [18]) Let K be a non-void compact convex

subset of a locally convex topological vector space E and F a semigroup of weakly
continuous self-maps on K . If F is strongly distal on K , then F has a common fixed
point.

Proof To prove that F∈F Fi x F = φ, where Fi x F is the set of fixed points of F in
K , it suffices to show that {Fi x F, F ∈ F} being a family of weakly closed subsets
of the weakly compact set K has finite intersection property.
Let F1 , . . . , Fn be a finite subfamily of F and S be the semigroup generated
by {F1 , . . . , Fn }. S is countable and it suffices to show that S has a common fixed
point. To this end, consider Q the convex closure of {F(k0 ) : F ∈ S} where k0 ∈ K .
Clearly, Q is strongly separable and as each F ∈ S is affine Q is S invariant and is
weakly closed being a closed convex subset of K . K being weakly compact, Q too
is weakly compact. In other words, the proof of the theorem follows from the proof
for the special case of K = Q and F = S.
We now show that S is weakly distal on each weakly closed minimal S invariant set
X in Q. For x, y ∈ X , x = y. Let {Vα : α ∈ } be a strongly open cover satisfying
Definition 11.4.20. Since E is locally convex {Vα } has an open refinement {X ∩
Bβ , β ∈ B} with each Bβ a convex strongly open set in E and for each Bβ , X ∩ B β ⊂
Vα for some α. Due to strong separability, there is a countable subcover {X ∩ Bi : i ∈
Z }. Each B i is strongly closed convex and is weakly closed. Now {X ∩ B i : i ∈ Z } is
a countable weakly closed cover of the weakly compact set X . So by Baire’s theorem,
at least one of these sets must contain a weakly open set U with U ⊆ X ∩ B i ⊂ Vα .
The family {F −1 (U ) : F ∈ S} of weakly open sets satisfies the Definition 11.4.20
for the points x = y. These sets cover X as otherwise X − ∪{F −1 (U ) : F ∈ S}
would be a weakly compact S-invariant proper subset of X contradicting the mini-
mality of X . Also for no G ∈ S, Gx and Gy belong to the same F −1 (U ); otherwise
F Gx and F Gy would be in U ⊆ X ∩ B i ⊆ Vα and as F G ∈ S this contradicts that
{Uα } is a strong open cover. Hence the theorem.
Proof of Theorem 11.4.19. Given a compact convex non-void set K of a locally

convex linear topological space E and F an equicontinuous group of affine maps of
K into itself, F must be distal on K . Otherwise there exist pair of distinct elements
on K , with the property for each neighbourhood U of the origin there is an FU ∈ F
such that FU (x) and FU (y) lie in a common set of the open cover {U + c : c ∈ K }
of K . Let W be a neighbourhood of the origin such that y ∈ / x + W ; then for each
V a neighbourhood of the origin there exists U (a neighbourhood of the origin)
with U = U ⊆ V and FU (x) − FU (y) ∈ V . However for FU−1 ∈ F, FU−1 (FU x) −
FU−1 (FU y) = x − y ∈ / W . This contradicts the equicontinuity of F on K . Hence F
is distal on K . So by Theorem 11.4.21. F has a common fixed point.
We can wind up with some applications.
Definition 11.4.24 Let G be a group and f a bounded function in G. f is called

left-uniformly almost periodic if the set
M f = { f λ : λ ∈ G}
is precompact in the uniform norm, f λ being the function f λ (x) = f (λx), λ, x ∈ G).
f is called left weakly almost periodic if the set M f defined above is precompact
in the weak topology in the linear space of bounded functions on G with uniform
norm.
Theorem 11.4.25 Let G be a group and f a bounded function and M f = { f λ : λ ∈
G, where f λ (x) = f (λx) for all x ∈ G}. Let K = coM f . If
(i) f is left almost periodic, then K contains constant functions;
(ii) if f is almost periodic and M f is precompact in the weak topology, then too K
contains constant functions.
Proof (i) K is a compact convex subset of B(G) the space of bounded functions on
G. The operators Tλ defined by Tλ g = gλ is a group of isometries mapping K into
itself. So by Kakutani’s Theorem 11.4.19. Tλ has a common fixed point f in M.
Thus f (x) = f (λx) ∀ λ ∈ G. Thus f (x) = f (e) for all x ∈ G or f is constant.
If M f is weakly compact, K is weakly compact and convex. Let Tλ be the group
of isometries mapping M into itself as before. Then by Ryll-Nardzewski Theorem
11.4.23. Tλ has a common fixed point which is a constant function again.
Next, following Rudin [17] we deduce the existence of left-invariant Haar measure
on a compact group. Let G be a compact topological group and C(G) the Banach
space of all continuous complex-valued functions with the supremum norm (Recall
that a topological group is a group with a Hausdorff topology such that (a, b) → ab−1
is continuous.
Lemma 11.4.26 Let G be a compact group, f ∈ C(G) and HL ( f ) the closed con-
vex hull of the left translates L s f of f . (Thus HL ( f ) = conv{L s f : L s f (x) =
f (sx), s ∈ G}. Then (a) f is uniformly continuous and (b) HL ( f ) is totally bounded
in C(G).
Proof f : G → C is called uniformly continuous if for each > 0 there is a neigh-
bourhood Ne of e the identity in G such that | f (x) − f (y)| < for x, y ∈ G with
x −1 y ∈ Ne .
As f is continuous on G, for each a ∈ G and > 0, there is a neighbourhood
Na of e such that | f (x) − f (a)| < 2 for all x ∈ a Na . From the definition of the
topology on G, we can find neighbourhoods Ua of e such that Ua Ua−1 ⊆ Na . As G is
k
compact and {aUa : a ∈ G} is an open cover for G, G ⊆ ai Uai for some k ∈ N.
i=1

k
Let U = Uai . Let x −1 y ∈ U . Choose ai , i = 1, 2, . . . , k such that y ∈ ai Uai .
i=1
Then | f (a) − f (ai )| < 2 . Now | f (ai ) − f (x)| < 2 as x ∈ yU −1 ⊆ ai Uai U −1 ⊆
a Na . So | f (x) − f (y)| ≤ | f (x) − f (ai )| + | f (ai ) − f (y)| < 2 + 2 = . Thus f
is uniformly continuous on G.
For x −1 y ∈ U , | f (x) − f (y)| < . But for all s ∈ G, x −1 y = (sx)−1 (sy) and
|L s f (x) − L s f (y)| = | f (sx) − f (sy)| < as (sx)−1 (sy) = x −1 y ∈ U . Thus
HL ( f ) is an equicontinuous subfamily of C(G). Hence the lemma.
Theorem 11.4.27 Given a compact group G, there is a unique
regular Borel prob-
ability measure μ which is left-invariant. That is G f dμ = G (L s f )dμ for s ∈ G
and f ∈ C(G). This μ is also right-invariant and satisfies the relation

f (x)dμ = f (x −1 )dμ
G G
for f ∈ C(G).
Proof The operators L s on C(G) defined by L s f (x) = f (sx) for any given s ∈ G
and f ∈ C(G) for all x form a semigroup and indeed a group of isometries on
G. So it is equicontinuous. If f ∈ C(G), K f the closure of HL ( f ) is compact by
Lemma 11.4.26. Clearly, L s (K f ) = K f . So by Kakutani’s Theorem 11.4.19, there is
a common fixed point φ in K f for all L s . Thus L s φ = φ for all s ∈ G. Thus φ(x) =
φ(sx) for all s ∈ G so that φ(x) = φ(sx) for all s ∈ G so that φ(x) = φ(e) for all
x. Hence φ is constant. Since K f = cl HL ( f ). φ(e) can be uniformly approximated
by functions in HL ( f ). So for each f ∈ C(G), there exists a constant k which can
be uniformly approximated by convex combinations of left translates of f on G.
Similarly there is a constant k that can be uniformly approximated on G by convex
combinations of right translates of f . We will show that k = k. Let > 0 be any
prescribed
number. There exist finite sets A = {ai } and B = {bi } in G with αi , β j > 0
and αi = β j = 1 and
A B

(I) k − A αi f (ai x) < , x ∈ G and
(II) k − B β j f (b j x) < , x ∈ G
Setting x = b j in (I), multiplying (I) by β, and add over j to get

(III) k − i, j αi β j f (ai b j ) <
Similarly setting x = ai in (II), multiplying (II) by αi and add over i to get

(IV) k − i, j αi β j f (ai b j ) <
From (III) and (IV) it follows that k = k .
Thus for each f ∈ C(G) there is a unique number written as I f such that it can
be uniformly approximated by convex combinations of left translates of f (as well
as convex combinations of right translates of f ) with the following properties:
I f ≥ 0 for f ≥ 0,
I1 = 1
Iα f = α I f for any scalar α
I(L s f ) = I f = I(Rs f ) for each s ∈ G
We now show that I f +g = If + Ig . Given > 0 for a finite subset A = {ai } ⊆ G

and αi > 0 with αi = 1, I f − αi f (ai x) < for all x ∈ G. Define h(x) =

A
αi g(ai x), then h ∈ K g . So K h ⊆ K g . Since each of K f and K g contains unique
i
constant functions Ih and Ig , it follows that Ih = Ig . So there is a finite set B =
{b j } ⊆ G with β j > 0 and β j = 1 such that for all x ∈ G

Ig − β j h(b j x) < .

j
From the definition of h we get for all x ∈ G

Ig − αi β j g(ai b j x) < .

i, j
Since for all x ∈ G

I f − α j f (a j x) <

j

replacing x by b j x and multiplying by β j and suming over j we get as βj = 1

I f − αi β j f (ai b j x) <

i, j

So I f + Ig − αi β j ( f + g)(ai b j x) < 2 for all x ∈ G. Since αi β j = 1 and
i, j i, j
αi β j ≥ 0 it follows that I f + Ig = I( f +g) .
Since I f ≥ 0, I f = 1, Iα f = α I f for α ∈ C and I f +g = I f + Ig , it follows from
the Riesz representation theorem that there is a unique regular Borel probability
measure μ such that
If = f dμ for f ∈ C(G).
G
From theconstruction of I f , the left invariance of μ follows.

I f = f (x −1 )dμ is well-defined on C(G) and I f ≥ 0 for f ≥ 0, I1 = 1,
G
I f (α f + βg) = α I f + β Ig for f, g ∈ C(G) and α, β ∈ C. So by the uniqueness
I f = I f .
Kitamura and Kusano [12] proved the existence of oscillatory solutions for a first-
order nonlinear functional differential equation under suitable assumptions using
Tychonoff’s theorem. In what follows this existence result is described.
Definition 11.4.28 A solution of a first-order differential or functional differential

equation defined on [α, ∞) α > 0 is called oscillatory if it has arbitrarily large zeros.
Otherwise it is called nonoscillatory.
Theorem 11.4.29 Consider the differential equation

N
x (t) = qi (t) f i (x(gi (t))), t > a > 0 (11.1)
i=1
Suppose
(a) qi , gi : [a, ∞) → R are continuous functions, with qi (t) ≥ 0 and lim gi (t) =
t→∞
∞ for i = 1, 2, . . . , N ;
(b) f i : R → R is continuous, nondecreasing and t f i (t) > 0 for t = 0 for i =
1, 2, . . . , N ;
N ∞
(c) qi (t)dt < ∞;
i=1 a
Then the differential equation (11.1) has only nonoscillatory solutions.
Proof For an arbitrary positive constant k consider the integral equation
N
t
x(t) = k + qi (s) f i (x(gi (s)))ds
i=1 T
where T > a is chosen such that

N ∞
f i (2k) qi (s)ds < k
i=1 T
in view of (c) and the continuity of fi , i = 1, 2, . . . , N . Define T0 = min {inf gi (t)

i=1,...,N
: t ≥ T } so that T0 > 0 (this can be done by rechoosing T in view of the assumption
lim gi (t) = ∞ for i = 1, 2, . . . , N ). Let M be the locally convex space of all real-
t→∞
valued continuous functions on [T0 , ∞) with the topology of uniform convergence
on compact subintervals of [T0 , ∞). Let X = {x ∈ M : k ≤ x(t) ≤ 2k for t ≥ T0 }.
Define the operator φ : X → M by
N t
k+ qi (s) f i (x(gi (s)))ds
φ(x(t)) = i=1 T
k for T0 ≤ t ≤ T
Clearly, φ(x) ∈ X for each x ∈ X in view of the choice of T . Further, X is a closed

convex subset of M and φ maps X into a compact subset of X . So by Tychonoff’s fixed
point theorem φ has a fixed point which is a solution for the integral equation leading
to the solution of the functional differential equation (11.1). Clearly, the solution is
nonoscillatory.
The next theorem follows similarly and its proof is left as an exercise.
Theorem 11.4.30 Under (a), (b) and (c) of Theorem 11.4.29, the equation

N
x (t) + qi (t) f i (x(gi (t))) = 0 (11.2)
i=1
has nonoscillatory solutions.
Example 11.4.31 (Kitamura and Kusano [12]) For α > 0 and β ≥ 1, the equation
|x(t + sin t)|α sgn x(t + sin t)

x (t) =
t β (log(t + sin t))α
∞
dt
has nonoscillatory solutions for t ≥ 2π as β (log(t + sin t))α
< ∞ and condi-
2π t
tions (a) and (b) of Theorem 11.4.29 are satisfied.
Remark 11.4.32 It has been observed in [12] that under the assumptions (a) and (b)
N ∞

of Theorem 11.4.29, along with qi (t)dt = ∞, for the equation
i=1 a

N
x (t) = qi (t) f i (x(gi (t))) (11.3)
i=1
every solution is oscillatory when gi (t) > t for i = 1, . . . , N ; for the equation

N
x (t) + qi (t) f i (x(gi (t))) = 0 (11.4)
i=1
every solution is oscillatory when gi (t) < t for i = 1, . . . , N .
Next we state and prove an inequality due to Ky Fan.
Theorem 11.4.33 Let K be a nonempty compact convex subset of a linear topolog-

ical space X and g : K × K → R be a map such that
(i) g(·, y) is lower semicontinuous for each y ∈ K (i.e. g −1 (·, y) (α, ∞] is open in
K for each y ∈ K );
(ii) g(x, ·) is concave for each x ∈ K (i.e. −g(x, ·) is a convex function for each
x ∈ K ).
Then there exists x0 ∈ K such that
sup g(x0 , y) ≤ sup g(y, y).

y∈K y∈K
Proof For each > 0, and a given x ∈ K we can find yx ∈ K and a neighbourhood
N x of x such that
g(z, yx ) > sup g(x, y) − for all z ∈ N x

y∈K

Since K ⊆ N x and K is compact we can find a finite number of elements
x∈K

n
x1 , x2 , . . . , xn in K with K ⊆ N xi . Let {ϕ1 , . . . , ϕn } be a partition of unity for K
i=1
subordinate to the covering {N xi : i = 1, . . . , n}. Then the map f defined by

n
f (x) = ϕi (x)yxi
i=1
maps the closed convex hull of {yx1 , . . . , yxn } into itself and is continuous. Since X
is a linear topological space and the subspace topology on the closed convex hull of
{yx1 , . . . , yxn } is euclidean, f has a fixed point x ∗ by Brouwer’s fixed point theorem.
So
sup g(y, y) ≥ g(x ∗ , x ∗ )

y∈K

n
≥ ϕ(x ∗ )(g(x ∗ , yxi ) − )
i=1

n
≥ ϕ(x ∗ )(sup g(xi , y) − )
i=1 y∈K
n
≥ ϕ(x ∗ ) sup g(xi , y) −
i=1 y∈K
≥ inf sup g(x, y) −

x∈K y∈K
= sup g(x0 , y) − for some x0
Allowing to tend to zero we conclude that

sup g(x0 , y) ≤ sup g(y, y)

y∈K y∈K
for some x0 .
Consider a game involving n(≥ 2) players who pursue a strategy depending on the
strategies of other players. Let the strategy set of the ith player be denoted by K I and
K be the set K 1 × K 2 × · · · × K n . An element of K is called a strategy profile. For
n
each player let f i : K → R be the loss function of the ith player. If f k (x) = 0,
k=1
then this game is called zero-sum game.
Definition 11.4.34 A Nash equilibrium is a strategy profile x = (x 1 , . . . , x n ) ∈ K

such that for each i = 1, 2, . . . , n
f i (x) ≤ f i (x 1 , . . . , x i−1 , x, x i , . . . , x n ) for all x ∈ K i
(In other words this strategy profile minimizes the loss for each player.)
The existence of a Nash equilibrium for an n-person game can now be proved in
the setting of a locally convex topological vector space.
Theorem 11.4.35 For i = 1, 2, . . . , n, let K i be a non-void compact convex sub-

set of a locally convex topological vector space X i . Suppose that for each i =
1, 2, . . . , n the loss function f i : K → R is continuous and for each fixed x j ∈ K ,
with j = i, the function f i (x1 , . . . , xi−1 , ·, xi+1 , . . . , xn ) : K i → R is convex. Here
K = K 1 × K 2 × · · · × K n . Then there is a Nash equilibrium in K .
Proof Define g : K × K → R by

n
g(x, y) = f i (x) − f i (x1 , . . . , xi−1 , yi , xi+1 , . . . , xn )
i=1
where x = (x1 , . . . , xn ) and y = (y1 , . . . , yn ). Since g is continuous and g(x, ·)

is concave for each fixed x ∈ K , by Theorem 11.4.33 there exists x ∈ K such
that sup g(x, y) ≤ sup g(y, y) = 0. Setting y = (x 1 , . . . , x i−1 , xi , x i+1 , . . . , x n ) for
y∈K y∈K
each xi ∈ K i we get
g(x, y) ≤ 0 for each xi ∈ K i
This means that for each i = 1, 2, . . . , n
f i (x) = f i (x 1 , . . . , x i−1 , xi , x i+1 , . . . , x n )
for all xi ∈ K i .
In the case of two person zero-sum game the hypotheses can be further weakened
and we get a theorem due to von Neumann.
Clearly, g(x1 , x2 ) = f 1 (x1 , x2 ) + f 2 (x1 , x2 ) = 0 or f 1 (x1 , x2 ) = − f 2 (x1 , x2 ).
Theorem 11.4.36 Let X 1 and X 2 be two locally convex spaces and K i ⊆ X i i = 1, 2

be non-void compact convex subsets of X i . Let ψ : K 1 × K 2 → R be such that
(i) ψ(·, x2 ) is lower semicontinuous and convex for each x2 ∈ K 2 ;
(ii) ψ(x1 , ·) is upper semicontinuous and concave for each x2 ∈ K 2 ;
Then there is a Nash equilibrium (x 1 , x 2 ) ∈ K 1 × K 2 .
For the proof define g : K × K → R by g((x1 , x2 ), (y1 , y2 )) = −ψ(y1 , x2 ) +

ψ(x1 , y2 ) and apply Theorem 11.4.35 to K = K 1 × K 2 . From the theorem, it also
follows that inf sup ψ(x, y) = sup inf ψ(x, y). So this theorem is also called
x∈K 1 y∈K 2 y∈K 2 x∈K 1
minimax theorem.
References
1. Banas, J., Goebel, K.: Measures of Noncompactness in Banach Spaces. Marcel Dekker, Inc.,
New York (1980)
2. Browder, F.E.: The fixed point theory of multivalued mappings in topological vector spaces.
Math. Ann. 177, 283–301 (1968)
3. Cauty, R.: Solution du probleme de point fixe de Schauder. Fundam. Math. 170, 231–246
(2001)
4. Darbo, G.: Punti uniti in transformazioxi a condominio non compatio. Rend. Semin. Mat. Univ.
Padova 24, 84–92 (1955)
5. Day, M.M.: Amenable semigroups. Ill. J. Math. 1, 509–544 (1957)
6. Dugundji, J., Granas, A.: J. Math. Anal. Appl. 97, 301–305 (1983)
7. Fan, Ky.: Fixed point and minimax theorems in locally convex topological linear spaces. Proc.
N.A.S 38, 121–126 (1952)
8. Glicksberg, I.L.: A further generalization of the Kakutani fixed point theorem with application
to Nash equilibrium points. Proc. Am. Math. Soc. 3, 170–174 (1952)
9. Hahn, S.: A fixed point theorem. Math. Syst. Theory 1, 55–57 (1968)
10. Kakutani, S.: Two fixed point theorems concerning bicompact convex sets. Proc. Imp. Acad.
Jpn. 14, 242–245 (1938)
11. Kakutani, S.: A generalization of Brouwer’s fixed point theorem. Duke Math. J. 8, 457–459
(1941)
12. Kitamura, Y., Kusano, T.: Oscillation of first-order nonlinear differential equations with devi-
ating arguments. Proc. Am. Math. Soc. 78, 64–68 (1980)
13. Kuczma, M., Choczewski, B., Ger, R.: Iterative Functional Equations. Encyclopedia of Math-
ematics and Its Application. Cambridge University Press, Cambridge (1990)
14. Markov, A.A.: Quelques theorems sur les ensembles abeliens. C.R. (Dokl.) Acad. Sci. URSS
(N.S) 1, 311–313 (1936)
15. Murugan, V., Subrahmanyam, P.V.: Existence of continuous solutions for an iterative functional
series equation with variable coefficients. Aequ. Math. 78, 167–176 (2009)
16. Rothe, E.: Zur theorie der topologischen Ordnung und der Vektorfelder in Banachschen Rau-
men. Compos. Math. 5, 177–196 (1937)
17. Rudin, W.: Functional Analysis. McGraw Hill International Edition (1991)
References 275
18. Ryll-Nardzewski, C.: On fixed points of semigroups of endomorphisms of linear spaces. In:
Proceedings of the Fifth Berkely Symposium on Statistics and Probability, Berkeley, vol. II
(1966)
19. Sadovski, B.N.: Asymptotically compact and densifying operators. Usp. Math. Nauk. 27, 81–
146 (1972)
20. Schauder, J.: Der Fixpunktsatz in funktional raumen. Stud. Math. 2, 171–180 (1930)
21. Terkelsen, F.: A short proof of Fan’s fixed point theorem. Proc. Am. Math. Soc. 42, 643–644
(1974)
22. Tychonoff, A.: Ein Fixpunktsatz. Math. Ann. 111, 767–776 (1935)
23. Werner, D.: A proof of the Markov-Kakutani fixed point theorem via the Hahn-Banach theorem.
Extr. Math. 8, 37–38 (1993)
24. Zhang, W., Baker, J.A.: Continuous solutions for a polynomial-like iterative equation with
variable coefficients. Ann. Pol. Math. 73, 29–36 (2000)
Chapter 12
Basic Analytic Degree Theory
of a Mapping
12.1 Introduction
The problem of finding the number of solutions of a given equation has engaged a
number of mathematicians. Brouwer, Bohl, Cauchy, Descartes, Gauss, Hadamard,
Hermite, Jacobi, Kronecker, Ostrowski, Picard, Sturm and Sylvester had contributed
to this topic. The Argument principle propounded by Cauchy on the zeros of a func-
tion inside a domain and Sturm’s theorem on the number of zeros of a real polynomial
in a closed bounded interval have evolved into Degree theory of mappings. Even as
the degree of a nonconstant polynomial gives the number of zeros of a polynomial,
the degree of a mapping provides the number of zeros of nonlinear mapping in a
domain. In this chapter, an elementary degree theory of mappings is described from
an analytic point of view proposed by Heinz [3]. For more elaborate treatment, Cronin
[1], Deimling [2], Lloyd [4] Outerelo and Ruiz [6] and Rothe [7] may be referred. It
should be mentioned that Ortega and Rheinboldt [5] had provided a more accessible
version of Heinz’s treatment.
12.2 Heinz’s Elementary Analytic Theory of Mapping

Degree in Finite Dimensional Spaces
Heinz [3] based his approach on some lemmata and relevant definitions. Throughout
we assume that is an open set in Rn , ∂ its boundary and its closure with respect
to the topology generated by the euclidean norm. We consider a map y : → Rn
with y = (y1 , . . . , yn ) where yi = yi (x1 , . . . , xn ) is the ith component function of
y, where i = 1, 2, . . . , n. For y ∈ C 1 (), the Jacobian J [y(x)] is well-defined for
each x ∈ . Let Ai j (x) be the cofactor of ai j (x) in the determinant of J [y(x)] =
det (ai j (x)).

https://doi.org/10.1007/978-981-13-3158-9_12
278 12 Basic Analytic Degree Theory of a Mapping
Lemma 12.2.1 Let y : → Rn be a function in C 2 (). Let Ai j (x) be the cofactor

n
∂
of ai j (x) in the determinant of J [y(x)] = det (ai j (x)). Then Ai j (x) = 0,
j=1
∂x j
i = 1, . . . , n, x ∈ .
Proof Suppose (ai j (x)) is invertible and bi j = (ai j (x))−1 . Then for d = det (ai j (x)).
Clearly, d I = (ai j )(Ai j )T or
∂d
dδi j = nk=1 aik A jk . So ∂a ij
= Ai j
dbi j = A ji .
For y (x0 ) invertible, by the inverse function theorem y maps some open neigh-
bourhood of U of x0 onto an open neighbourhood of y(x0 ) and if z = y −1 , then
y (x0 )z (y(x0 )) = I . So J y(x0 ) = (ai j ). J (z(y(x0 )) = (bi j ), d = det (J y(x0 )) and
e = det J [z(y(x0 )]. Let Bi j be the cofactor of the (i, j)th element of J [z(y(x0 ))] =
∂e n
∂e ∂bi j n
∂bi j n
∂bi j
bi j . Then = = Bi j =e a ji .
∂ yk i, j=1
∂b ij ∂ yk i, j=1
∂ yk i, j=1
∂ yk
But
∂bik ∂ 2 zi ∂ ∂z i ∂ y j n
= =
∂xi ∂xi ∂ yk j=1
∂ yk ∂ y j ∂xi

n
∂bi j
= a ji
j=1
∂ yk
and hence
∂e ∂bik n
=e (12.2.1)
∂ yk i=1
∂ yk
n ∂bik
As de = 1, d ∂∂eyk + e ∂∂dyk = 0, we get from (12.2.1), 0 = ∂∂dyk + d i=1 ∂xi
=
n ∂d ∂bik

i=1 ∂xi bik + e ∂xi = i=1 ∂∂yk (dbik ) = i=1 ∂∂yk Aki , k = 1, . . . n.
n n

If y (x0 ) is singular, then the mapping y = y(x) + x is invertible. So by the first
n
∂
part of the lemma (Aji ) = 0 where Aji is the cofactor of (i, j)th element in
i=1
∂x i
n
∂ n
∂
J [y (x)]. By continuity A ji = lim A = 0. Thus in this case also
i=1
∂xi >0
i=1
∂xi ji
the Lemma is true.
Lemma 12.2.2 Suppose
(i) y : → is a function with y ∈ C 1 (), being an open bounded subset of
Rn , y is continuous on and for some > 0 y(x) > for all x ∈ ∂;
(ii) ϕ : [0, ∞) ∞→ R is continuous, vanishing in [, ∞) and also in a neighbourhood
of 0 and 0 r n−1 ϕ(r )dr = 0.
12.2 Heinz’s Elementary Analytic Theory of Mapping … 279

Then ϕ(y(x))J [y(x)]d x = 0.
Proof Since functions in C() can be uniformly approximated by polynomials in

, it suffices to prove the lemma for y ∈ C 2 ().
Define ψ : [0, ∞) → R by
r
r −n 0 ρn−1 ϕ(ρ)dρ, for r > 0
ψ(r ) =
0, for r = 0
Clearly, ψ ∈ C 1 [0, ∞) and vanishes in a neighbourhood of 0 and in [, ∞) since ϕ

vanishes in a neighbourhood of 0 and in [, ∞). Further r ψ (r ) + nψ(r ) = ϕ(r ) for
r ≥ 0.
For each i = 1, . . . , n, f i (y) = ψ(y)yi is in C 1 (Rn ) and vanishes for y ≥ .
From Lemma 12.2.1 we get
n
∂
n n
∂ fj
Ai j (x) f j (y(x)) = J [y(x)]
i=1
∂xi j=1 j=1
∂yj y=y(x)

= J [(y(x)](r ψ (r ) + nψ(r ))r =y(x)
= ϕ(y(x))J [y(x)]. (12.2.2)
For F : Rn → R, a continuously differentiable function with compact support
div F(x)d x = 0
Rn
So integrating the above equality (12.2.2) over Rn we get
ϕ(y(x))J [y(x)]d x = 0
Rn
Lemma 12.2.3 Let y : ⊆ Rn → Rn be a function in C 1 () where is a bounded

open set and let y be a continuous on . Let z ∈ Rn be such that z = y(x) for x ∈ ∂
and : [0, ∞) → R be continuous function such that
(i) it vanishes in a neighbourhood of 0 and also in [, ∞) where 0 < < min{x ∈
∂
A : |y(x) − z|};
(ii) Rn (x)d x = 1.
Then the number d[y(x); ; z] is uniquely defined by
d[y(x); ; z] = (y(x) − z)J [y(x)]d x

Proof Let D be the linear space of all continuous real-valued functions satisfying
(i). Define L , M, N : D → R by
∞
L = r n−1 (r )dr
0
M = (x)d x
Rn
N = (|y(x) − z|)J [y(x)]d x

Clearly, L , M and N are linear functionals on D. Applying Lemma 12.2.2 to

y = x (for x < 2) and y = y(x) − z (x ∈ ) it follows that L = 0 for ∈ D
implies M = N = 0. For 1 , 2 ∈ D with M1 = M2 = 1, L(L1 .2 −
L2 .1 ) = 0 and so L2 .M1 − L1 .M2 = 0. So L(1 − 2 ) = 0. So
N (1 − 2 ) = 0 and N 1 = N 2 . Thus the definition of N is independent of
in D and d[y(x); , z] is independent of and is uniquely defined.
Definition 12.2.4 Let y : ⊆ Rn → Rn (where is a bounded open set in Rn ) a

continuous function such that y ∈ C 1 (). The number d[y; , z] uniquely obtained
in Lemma 12.2.3 is called the (Brouwer) degree of y.
Lemma 12.2.5 Let yi : ⊆ Rn → Rn be mappings such that yi is in C 1 () and

continuous on for each i = 1, 2. Suppose for z ∈ Rn
(i) yi (x) − z > 7 > 0 for x ∈ ∂ for i = 1, 2 and
(ii) y1 (x) − y2 (x) > for x ∈ .
Then d[y1 (x); , z] = d[y2 (x); , z].
Proof Clearly, d[yi (x); , z] is well-defined and equals d[yi (x) − z; , 0] for
each i = 1, 2. Thus without loss of generality it may be assumed that z = 0.
Let f : [0, ∞) → R be a continuously differentiable function which vanishes in
[3, ∞) and is 1 in [0, 2] with range in [0, 1]. Define y3 : → Rn by y3 (x) =
[1 − f (y1 (x))]y1 (x) + f (y1 (x))y2 (x). Since 0 ∈ / , y3 ∈ C 1 () as f, y1 , y2
are C -functions. Further y3 ∈ C(). Clearly, yi (x) − y j (x)| < for x ∈ , for
1
i, j ∈ {1, 2, 3}. From the definition of f , for y1 (x) > 3, y3 (x) = y1 (x) and for
||y1 (x)|| < 2, y3 (x) = y2 (x).
Let i : [0, ∞) → R be be two continuous functions that vanish in a neighbour-
hood of 0 satisfying
1 (r ) = 0 for r ∈ [0, 4] ∪ [5, ∞)

2 (r ) = 0 in [, ∞) and
i (x)d x = 1 for i = 1, 2.
Rn
12.2 Heinz’s Elementary Analytic Theory of Mapping … 281
From the definitions of i and the choice of y3 it follows that for x ∈
1 (y3 (x))J [y3 (x)] = 1 (y1 (x))J [y1 (x)]
and
2 (y3 (x))J [y3 (x)] = 2 (y2 (x))J [y2 (x)].
Upon integrating we get
d[y3 (x); , 0] = d[y1 (x); , 0]
and
d[y3 (x); , 0] = d[y2 (x); , 0]
whence d[y1 (x); , 0] = d[y2 (x); , 0] as required.
Lemma 12.2.6 Let y : → Rn be continuous, being a nonvoid bounded open

set in Rn . Suppose z ∈ Rn is such that z = y(x) for all x ∈ ∂. If yk : → Rn is
continuous for each k ∈ N and is in C 1 () with yk (x) = z for all x ∈ ∂ and (yk )
converges uniformly in to y then d[y(x); , z) = limk→∞ d[yk (x); , z].
Proof Be Lemma 12.2.5, {d(yk (x); , z)} is a Cauchy sequence of real num-
bers and hence is convergent. (Indeed it is constant after some stage.) Again
as sup{y(x) − yk (x) : x ∈ } can be made less than 10 1
sup{|y(x) − z : x ∈
∂}. d(y(x); , z) = d(yk (x); , z) after some stage. Thus d(y(x); , z) = lim
k→∞
d(yk (x); , z) for z ∈
/ {y(x); x ∈ ∂}.
Remark 12.2.7 For y ∈ C() where is a nonvoid bounded open set and y() ⊆
Rn , for z ∈
/ {y(x); x ∈ ∂}, d[y(x); , z] is well-defined. Indeed by Weierstrass
approximation theorem there is a sequence of polynomials (yk ) on such that
(yk ) converges uniformly on to y and yk (x) = z for all x ∈ ∂ for all k. So
d[y(x); , z) is well-defined for all y ∈ C() with z ∈ / y(∂) and is called the
degree of y.
Remark 12.2.8 For a linear operator A : Rn → Rn , = B(0, 1) the unit open ball.
⎧
⎪
⎨+1, if det of A > 0
d[A, = B(0, 1), 0] = −1, if det of A < 0
⎪
⎩
not defined if det of A = 0
12.3 Properties of the Degree
In this section, we describe the elementary properties of the degree of a mapping.

Theorem 12.3.1 Let 1 and 2 be two disjoint bounded nonempty open sets in Rn .
Let y : 1 ∪ 2 → Rn be a continuous map such that y(x) = z for x in 1 ∪ 2
for some fixed z ∈ Rn . Then
d[y(x); 1 ∪ 2 , z] = d[y(x); 1 , z] + d[y(x); 2 , z]
Proof By definition of the degree
d[y(x); 1 ∪ 2 , z] = lim d[yk (x); 1 ∪ 2 , z]

k→∞
= lim [d[yk (x); 1 , z] + d(yk (x); 2 , z)]
k→∞
= d[y(x); 1 , z] + d[y(x); 2 , z]
where yk ∈ C 1 (1 ∪ 2 ) converges uniformly on 1 ∪ 2 to y. It may be noted that

the linearity of the integral and Lemma 12.2.5 lead to the conclusion.
Corollary 12.3.2 Let y : ⊆ Rn → Rn be continuous, where is a nonempty
m
bounded open set. Suppose = i where each i is a nonempty bounded
i=1
open set in Rn with i ∩ j = φ for i = j. If z ∈
/ y() then d[y(x); , z] =
m
d[y(x); i , z].
i=1
The next theorem generalizes Lemma 12.2.5.

Theorem 12.3.3 Let be a nonvoid bounded open subset of Rn and y : → Rn
a continuous map. If z ∈ Rn is such that min{y(x) − z : x ∈ ∂} > > 0, then
d[y(x); , z] = d[ ŷ(x); , z] for any ŷ : → Rn for which sup{y(x) − ŷ(x) :
x ∈ } < 7 .
Proof We can find sequences yk , ŷ j : 0 → Rn such that (yk ), ( ŷ j ) ∈ C 1 (0 ),
where 0 is a bounded open set containing such that (yk ) and ( ŷ j ) converge
uniformly on to y and ŷ respectively. So there exists m 0 ∈ N such that for all
k, j ≥ m 0 and all x ∈
ŷ j (x) − yk (x) ≤ ŷ j (x) − ŷ(x) + ŷ(x) − y(x) + y(x) − yk (x)

< .
7
(Since sup{y(x) − ŷ(x) : x ∈ } = η < 7 , we have to choose m 0 so that

sup{y(x) − yk (x) : x ∈ } and sup{ ŷ j (x) − ŷ j (x) : x ∈ } < 21 7 − η .) Also
12.3 Properties of the Degree 283
we can choose (yk ) and ( ŷ j ) such that for k, j ≥ m 0 , inf{yk (x) − z : x ∈ ∂} and
inf{ ŷ j (x) − z : x ∈ ∂} > . So by Lemma 12.2.5, d[ ŷ j (x); , z] = d[yk (x);
, z] for all j, k ≥ m 0 . Proceeding to the limit we get d[ ŷ(x); , z] = d[y(x); , z]
by Lemma 12.2.6.
Next theorem establishes the invariance of the degree under homotopy.
Theorem 12.3.4 Let be a nonvoid bounded open set in Rn and H : × [0, 1] →

Rn be a continuous map. Suppose for z ∈ Rn for which H (x, t) = z for all x ∈ ∂
and t ∈ [0, 1]. Then d[H (x, t); , z] is a constant for all t ∈ [0, 1].
Proof As ∂ × [0, 1] is compact (in the product topology), min{H (x, t) − z :

x ∈ ∂, t ∈ [0, 1]} > > 0 for some . As H is uniformly continuous on × [0, 1],
for 7 we can find δ > 0 such that sup{H (x, t) − H (x, s) : x ∈ , s, t ∈ [0, 1]} <

7
for |s − t| < δ. So by Theorem 12.3.3 above d[H (x, t); , z] = d[H (x, s); , z]
for |s − t| < δ, s, t ∈ [0, 1]. Since [0, 1] can be covered by a finite number of sub-
intervals of length δ, it follows that d[H (x, t); , z] is constant for all
t ∈ [0, 1].
The next theorem, also called Poincare–Bohl theorem, is a consequence of the

above result.
Theorem 12.3.5 Let y, ŷ : ⊆ Rn → Rn be two continuous maps on a nonvoid

bounded open set . If z ∈/ {v ∈ Rn : v = t y(x) + (1 − t) ŷ(x), x ∈ ∂, t ∈ [0, 1]}
then d[y(x); , z) = d[ ŷ(x); , z].
Proof Define H : × [0, 1] → Rn by H (x, t) = t y(x) + (1 − t) ŷ(x). By

assumption z ∈
/ {v = t y(x) + (1 − t) ŷ(x), x ∈ ∂, t ∈ [0, 1]}. Further H is a homo-
topy between y(x) and ŷ(x), with H (x, 0) = ŷ(x) and H (x, 1) = y(x) for x ∈ .
So by Theorem 12.3.4 d[y(x); , z] = d[ ŷ(x); , z].
Corollary 12.3.6 Let be a nonvoid bounded open subset of Rn such that y, ŷ :

(⊆ Rn ) → Rn are continuous. Suppose y(x) = ŷ(x) for x ∈ ∂. Then for any
z∈
/ y(∂). d(y(x); , z) = d[ ŷ(x); , z].
Proof H : × [0, 1] × Rn defined by H (x, t) = t y(x) + (1 − t) ŷ(x) x ∈ , t ∈

[0, 1] is a homotopy with H (x, t) = y(x) = z for all x ∈ ∂ and t ∈ [0, 1] since
y(x) = ŷ(x) on ∂. So by Theorem 12.3.5, d(y(x); , z) = d[ ŷ(x); , z].
Remark 12.3.7 The above corollary implies that the degree of a mapping depends
only on its value on the boundary of .
The following theorem shows that the degree remains the same under certain
translates.
Theorem 12.3.8 Let y : ⊆ Rn → Rn be a continuous map on a nonvoid bounded

/ y(∂) and p ∈ Rn . Then d[y(x) − p; , z − p] = d[y(x); , z].
open set . If z ∈
Proof If y : D ⊆ Rn → Rn is continuously differentiable on an open bounded set

D ⊇ , then for ŷ(x) = y(x) − p. Let : [0, ∞) → R be a continuous function
vanishing in a neighbourhood of z as well as in [, ∞) where 0 < < min{y(x) −
z : x ∈ ∂}, such that Rn (x)d x = 1. Then (y(x) − z)det y (x) =
( ŷ(x) − (z − p))det ŷ (x). Consequently d( ŷ(x); , z − p) = d[y(x); , z].
Since there is a sequence (yk ) of functions in C 1 (D) converging uniformly on
D to y, where D is a bounded open set containing and d( ŷk (x); , z − p) =
d[yk (x); , z] by the proceeding argument, and proceeding to the limit as k tends to
infinity, we get d( ŷ(x); , z − p) = d[y(x); , z].
The above theorem in conjunction with the property of homotopy invariance of
the degree leads to
Theorem 12.3.9 Let y : → Rn be a continuous map on a nonvoid bounded open
set in Rn . Let p1 and p2 be two points in Rn such that there is a continuous path
η : [0, 1] → Rn which does not meet y(∂) with η(0) = p1 and η(1) = p2 . Then
d(y(x); , p1 ) = d[y(x); , p2 ].
Proof The map H : × [0, 1] → Rn defined by H (x, t) = y(x) − η(t) is a homo-
topy for which H (x, t) = 0 x ∈ ∂ and t ∈ [0, 1]. So by Theorems 12.3.4 and 12.3.8
d[y(x); , p1 ) = d[H (x, 0), , η(0)) = d[H (x, 1); , η(1)) = d[y(x); , p2 ).
The above results imply that in the computation of the degree of a mapping on ,
certain portions of can be excised. This point is captured in
Theorem 12.3.10 (Excision property) Let y : ⊆ Rn → Rn be a continuous map,
where is a nonvoid bounded open subset of Rn . Let z ∈
/ y(∂) and F any closed
subset of such that z ∈
/ y(F). Then d[y(x); , z] = d[y(x); − F, z]. In partic-
ular if F = , then d[y(x); , z] = 0.
Proof Let yk ∈ C 1 (D) where D is an open set containing such that 0 =
min{yk (x) − z : x ∈ ∂} > 0 and η = min{yk (x) − z : x ∈ F} > 0. Now =
min(0 , η) > 0. We can find : [0,
∞) → R such that vanishes in a neighbour-
hood of zero and on [0 , ∞) with Rn (x)d x = 1 and
d[yk (x); , z] = (yk (x) − z)J (yk (x))d x

= (yk (x) − z)J (yk (x))d x

−F
as (yk (x) − z) = 0 for x ∈ F.
= d[yk (x); − F, z].
Since any y ∈ C() can be uniformly approximated by a sequence yk ∈ C 1 ()

for which limk→∞ d[yk (x); , z] = d[y(x); , z], d[y(x); , z] = lim d[yk (x);
k→∞
, z] = lim [yk (x); − F, z] = d[y(x); − F, z]. In particular when = F,
k→∞
d[yk (x); − F, z] = 0 so that d[y(x); , z] = d[y(x); − F, z] = 0.
Corollary 12.3.11 Let y : ⊆ Rn → Rn be continuous, being a nonvoid

bounded open subset of Rn . If F = {x ∈ : y(x) = z} where z ∈
/ y(∂) then
d[y(x); , z] = d[y(x); − F, z].
The following theorem is the first step in relating the degree of a mapping to the
solution of (non-linear) equations.
Theorem 12.3.12 Let be a nonvoid bounded open set in Rn and y : → Rn be

a continuously differentiable map. Let D be a nonvoid open set contained together
/ y(∂ D), y (x) is non-singular for
with its closure in . Suppose that for a given z ∈
all x ∈ = {x ∈ D : y(x) = z}.
Then contains at most finitely many points such that
⎧ m
⎪
⎨ sgn det J [y(xi )], for = {x1 , . . . , xm }
d[y(x); , z] = i=1
⎪
⎩
0 if = φ.
Proof Since z ∈ / y(x) for x ∈ ∂ by hypothesis the solutions of y(x) = z lie in

. In other words each solution of y(x) = z is an interior point of . Since yx is
non-singular at these solutions, y is a local homeomorphism at each of the solutions
of y(x) = z by the inverse function theorem. Thus for each x with y(x) = z, y
maps a neighbourhood N (x) of x onto a neighbourhood of y injectively. Such a
neighbourhood of x will therefore not contain another solution of y(x) = z. Thus
the solutions of y(x) = z have no cluster points in or . Since is compact, such
isolated solutions of y(x) = z can only be finite. Thus is empty or nonempty and
finite.
If = φ, then we have already shown in Theorem 12.3.10, that d[y(x); , z] = 0.
Suppose = φ and = {x1 , . . . , xm }. Let 0 < < min{y(x) − z : x ∈ ∂}.
We can find : [0, ∞) → R such that is zero in a small neighbourhood of zero as
well as in [, ∞) with Rn φ(x)d x = 1. Further (y(x) − z)J [y(x)]d x = 0.
For each xi ∈ by Inverse function theorem we can find an open neighbourhood Ui
of xi such that the restriction of y on Ui is an open neighbourhood Vi of z and y is a
homeomorphism of Ui onto Vi . Since y is non-singular at each xi , we can choose Ui
such that the sign of J [y[x]] is that of y (xi ) for each i = 1, 2, . . . , m. We can find
0 ∈ (0, ) such that the closed ball B(z, 0 ) = K ⊆ Vi . Define Wi = yi−1 (K ) where
m
yi is the restriction of y to Ui . Now Wi ⊆ and if : [0, ∞) → R is a continuous
i=1
map vanishing in [0, α) and in [α, ∞) for 0 < α < 0 with Rn φ(x)d x = 1, then
by the change of variables theorem for the integral applied to each yi we get

m
d[y(x); , z] = (y(x) − z)J [y(x)]d x
i=1 Wi
m
= φ(yi (x) − z)J [yi (x)]d x
i=1 yi−1 (K )
m
= sgn det J [yi (x)] φ(x)d x
i=1 K
m
= sgn det J [yi (x)]
i=1

as φ(x)d x = Rn φ(x)d x = 1.
K
Thus in this case degree of the mapping y at z with respect to is an integer.
Naturally one would like to know if the degree of the mapping y(x) at a point
remains an integer even if the Jacobian of the mapping in singular at some of the roots
of y(x) = z. Thanks to Sard’s theorem detailed below the answer to this question is
in the affirmative.
Theorem 12.3.13 (Sard’s Theorem [8]) Let y : ⊆ Rn → Rn be continuously dif-

ferentiable on the nonempty bounded open set . Let K be a compact subset of
and C = {x ∈ K .y (x) is singular}. Then y(C) has zero measure.
Proof Since K ⊆ is compact and is open K can be covered by a finite num-

ber of cubes of any prescribed length. So it suffices to assume that K is a cube
Q, of side length . Divide Q into m n subcubes P j of side m . Suppose some
P, a subcube contains u at which y is singular. Given > 0, one can choose m
so large that y(x) − y(u) − yx (x − y)∞ ≤ x − u∞ ≤ m . Let sup{yx : x ∈
Q} = β. Clearly, β > 0. So for x ∈ P,

y(u) − (y(u) + y (x − y))∞ ≤ βx − u∞ ≤ β .
m
As y (u) is singular T (P) lies in a hyperplane of dimension at most n − 1 where

T (x) = y(u) + y (x − u). So y(P) is contained in neighbourhood of T (P). In
m
n−1
other words y(P) is contained in a hyper interval of volume at most 2(β + ) m
2 2 n
m
= m (β + )n−1 . Since > 0 is arbitrarily small, y(K ) is contained in
hyper intervals of total volume at most equal to (2)n (β + )n−1 . So y(K ) is of
measure zero.
In view of Sard’s theorem above, Theorem 12.3.12 can be reformulated as
Theorem 12.3.14 Let y : ⊆ R N → Rn be continuously differentiable where

is a nonvoid bounded open set. Suppose G ⊆ is open such that G ⊆ and
z∈/ y(∂G) ∪ y(S (G)) where S (A) = {x ∈ A : y (x) is singular} for A ⊆ . Then
= {x ∈ G : y(x) = z} is empty and d[y(x); G, z] = 0 or has finitely many points

x1 , . . . , xm and
m
d[y(x); G, z] = sgn det J [y(xi )]
i=1
Theorem 12.3.15 Let y : ⊆ Rn → Rn be continuously differentiable on the

bounded open set and G ⊆ an open set with G ⊆ . If z ∈ / y(∂G), then there
is a sequence z k ∈
/ y(∂G) ∪ y(S (G)) (k = 1, 2, . . . , ) such that lim z k = z and
k→∞
there exists k0 with d[y[x]; G, z] = d[y(x); G, z k ] for all k ≥ k0 .
Proof By Sard’s Theorem 12.3.13 y(S (G)) has measure zero. So there exists a
sequence z k ∈
/ y(S (G)) ∪ y(∂G) with lim z k = z. By assumption there exists >
k→∞
0 such that z k ∈ B(z, ) for all k ≥ k0 and the line segments joining y to yk for k ≥ k0
lie in B(z, ). So these paths do not meet y(∂G). So by Theorem 12.3.9 for k ≥ k0 .
d[y(x); G, z] = d[y(x); G, z k ].
Corollary 12.3.16 Let y : ⊆ Rn → Rn be a continuous map where is a nonvoid

bounded open set. Then for any z ∈
/ y[∂], d[y(x); , z] is an integer.
Proof We can find a sequence of functions yk ∈ C 1 () such that yk converges uni-
formly in to y, with z ∈/ yk (∂). Clearly, d[yk (x); , z] is an integer for each k by
Theorem 12.3.15. Also we can find k0 such that d[yk (x); , z] = d[yk0 (x); , z] for
all k ≥ k0 . Proceeding to the limit as k tends to infinity and noting that d[yk0 (x); , z)
is an integer and lim d[yk (x); , z] = d[y(x); , z] it follows that d[y(x); , z]
k→∞
is an integer.
12.4 Some Consequences
The concept of degree of a mapping at a point with respect to a region can be

used, along with its properties to prove Brouwer’s fixed point theorem as well as
Kronecker’s theorem on the existence of solution to nonlinear equations.
First, we prove Kronecker’s theorem.
Theorem 12.4.1 (Kronecker) Let be a nonempty bounded open set in Rn and

f : ⊆ Rn → Rn be a continuous map. If z ∈ / f (∂) and if d[y(x); , z] = 0,
then the equation f (x) = z has a solution in .
Proof If f (x) = z has no solution in then z ∈

/ f (). So by Excision Theorem
12.3.10. d[ f (x); , z] = 0, a contradiction.
Theorem 12.4.2 (Brouwer) Let f : B(0, 1) → B(0, 1) be a continuous mapping

where B(0, 1) is the closed unit ball in Rn , then f has a fixed point.
Proof Let be the unit open ball {x ∈ Rn : x < 1}. Suppose f (x) = x for
all x ∈ = B(0, 1). For each t ∈ [0, 1] and x ∈ define f (x, t) = x − t f (x).
Now f (x, t) = x − t f (x) ≥ x − t f (x) = 1 − t f (x) ≥ 1 − t > 0 for
x ∈ ∂ and t ∈ [0, 1). By assumption x = f (x) for x ∈ ∂. Thus f (x, t) = 0 for
all x ∈ ∂ and t ∈ [0, 1]. So by Theorem 12.3.4, d[ f (x, t); , 0] is a constant. Thus
d[ f (x, 1); , 0] = d[ f (x0 ); , 0] = 1 by definition of the degree. So by Theorem
12.4.1, the equation f (x) − x = 0 has a solution in , a contradiction. So f has a
fixed poin in .
It may be added by way of conclusion that Heinz had also presented Leray’s
product theorem for the degree of a mapping. This result can be used to deduce
important theorems of topology such as Brouwer’s domain invariance theorem. The
reader is referred to Lloyd [4] for further details.
References
1. Cronin, J.: Fixed Points and Topological Degree in Nonlinear Analysis. Surveys, vol. II. Amer-
ican Mathematical Society, Providence (1964)
2. Deimling, K.: Nichtlineare Gleichungen und Abbildungs grade. Springer, Berlin (1974)
3. Heinz, E.: An elementary analytic theory of the degree of mapping in n-dimensional space. J.
Math. Mech. 8, 231–247 (1959)
4. Lloyd, N.G.: Degree Theory. Cambridge University Press, Cambridge (1978)
5. Ortega, J.M., Rheinboldt, W.C.: Iterative Solution of Nonlinear Equations in Several Variables.
Academic, New York (1970)
6. Outerelo, E., Ruiz, J.M.: Mapping Degree Theory. Graduate Studies in Mathematics, vol. 108.
American Mathematical Society, Providence (2009)
7. Rothe, E.: Introduction to Various Aspects of Degree Theory in Banach Spaces. Mathematical
Surveys and Monographs, vol. 23. American Mathematical Society, Providence (1986)
8. Sard, A.: The measure of the critical values of differentiable maps. Bull. Am. Math. Soc. 48,
883–897 (1942)
Appendix A
A Counterexample on Common Fixed Points
Since every continuous self-map on a bounded closed interval of real numbers has a
fixed point, a natural question is whether two commuting continuous self-maps on a
closed bounded interval of real numbers have a common fixed point. This has been
answered in the negative, independently by Boyce [1] and Huneke [2]. In the sequel,
we highlight the solution provided by Huneke in Part II of his paper [2].
Let h : S ⊆ R → R be a map where S is a subset of the set of real numbers
R. Define h ∗ by h ∗ (x) = 1 − h(1 − x) for each x for which 1 − x ∈ S. Let b ∈
1 1
0, 2 and let s = 3−2b+(6−4b)
2
1−2b
. Clearly, s is well-defined and positive. Define three
continuous functions h 1 , h 2 and h 3 with linear graphs as follows:

(1 − b + sb)
h 1 : I1 = b, → [b, 1] by h 1 (x) = s(x − b) + b
s

(1 − b + sb) (2 − b + sb)
h 2 : I2 = , → [0, 1] by h 2 (x) = 2 − sx + sb − b
s s

(2 − b + sb) (3 − 2b + sb)
h 3 : I3 = , → [0, 1 − b] by h 3 (x) = −2 + sx − sb + b
s s

Define h : b, (3−2b+sb)
s
→ [0, 1] by h(x) = {h j (x), x ∈ I j , j = 1, 2, 3. Clearly, h
is continuous and has a linear graph and h j is invertible for each j = 1, 2, 3.
Definition A.1.1 Let Cb be the set of all continuous self-maps on [0, b] with b as a
fixed point. For each g ∈ Cb define g : [0, 1] → R the unique continuous extension
of g defined by
(1) g(x) = g(x) on [0, b];
(2) g(x) = h(x) on [b, h −1 3 (1 − b)];
(3) g(x) = (h ∗1 )−1 g(h ∗ (x)) on [h −1 ∗ −1 −1
3 (1 − b), (h 2 ) h 2 (0)];
∗ −1 ∗ ∗ −1 −1
(4) g(x) = (h 2 ) g(h (x)) on [(h 2 ) (h 2 (0)), 1 − b];
(5) g(x) = the fixed point of h ∗2 on [1 − b, 1].

https://doi.org/10.1007/978-981-13-3158-9
290 Appendix A: A Counterexample on Common Fixed Points
Remark A.1.2 The verification that g defined above is a unique extension of g is left
as an exercise.
Remark A.1.3 If g ∈ Cb satisfies Lipschitz condition with Lipschitz constant s, then
g too satisfies Lipschitz condition with Lipschitz constant s.
Proof (attributed to David Boyd in [2]) Let g ∈ Cb satisfy Lipschitz condition with
Lipschitz constants and L be the set of all s-Lipschitz self-maps on [0, 1] satisfying
(1), (2) and (5) for g in Definition A.1.1.
Define T : L → L by
⎧
⎪
⎨ f (x) if x ∈ [0, h −13 (1 − b)] ∪ [1 − b, 1]
T f (x) = (h 1 ) f (h (x)), if x ∈ [h 3 (1 − b), (h ∗2 )−1 h −1
∗ −1 ∗ −1
2 (0)]
⎪
⎩ ∗ −1
(h 2 ) f (h ∗ (x)), if x ∈ [(h ∗2 )(h −1
2 (0)), 1 − b]
Since (h ∗1 )−1 and (h ∗2 )−1 have linear graphs and 1s Lipschitzian, (h ∗1 )−1 (0) = (h ∗2 )−1
(s−1)(1−b)
(0), h ∗ is s-Lipschitzian and the fixed points of h ∗ are h −1 3 (1 − b), s+1
and
1 − b. Clearly, L is a complete metric space with the supremum metric on which T
is a contraction with Lipschitz constant 1s < 1. This follows from the fact that (h ∗1 )−1
and (h ∗2 )−1 are contractions with Lipschitz constant 1s . So, there is unique function
g in L with T (g) = g and it can be seen that g satisfies the conditions of Definition
A.1.1.
1
Lemma A.1.4 Let f, g ∈ Cb and x ∈ 2 , 1 . Then
(i) ( f )∗ x = x implies g(x) = x and
(ii) ( f )∗ (g(x)) = g(( f )∗ (x)).
Proof (i) The domain of h ∗ is [(h ∗3 )−1 (b), 1 − b] and (h ∗3 )−1 (b) = − 3+2b+s−sb
s
=
(1−2b)(2b−3) ∗
1 − b + 3−2b+(6−4b)1/2 < 2 as 0 ≤ b < 2 . So, the fixed points of ( f ) are in [1 − b, 1]
1 1
or those of h ∗ . By definition of g on [1 − b, 1], g(x) is the same as as the fixed

point of h ∗2 and this is (1−b)(s−1)
(1+s)
< 1 − b ≤ x for x ∈ [1 − b, 1]. So g and ( f )∗
have no common fixed point in [1 − b, 1]. The only fixed points of h ∗ are those
of h ∗j , j = 1, 2, 3. The fixed point 1 − b of h ∗1 is not a fixed point of g. If x2
is the fixed point of h ∗2 , then x2 = (1−b)(s−1) < b−2+s+ss 2−s b−2s = (h ∗2 )−1 (h −1
2 2
2 (0)).
∗ −1
s+1
(1−b)(s−1)
So g(x2 ) ∈ (h 1 ) [0, 1] = , s
s−1+b−bs s+b−sb
s
. But x2 = s+1
< s−1+b−sb
s
≤
∗ ∗
g(x2 ). So x2 = h 2 (x2 ) = g(x2 ). If x3 = h 3 (x3 ) then x3 = 1−s 3−s+sb−b
= 3−2b+sb
s
=
h −1
3 (1 − b). Thus, g(x 3) = g(h −1
3 (1 − b)) = 1 − b > x 3 . So, ( f ) ∗
and g have no
common fixed point in 21 , 1 .
(ii) If x ∈ [1 − b, 1] then ( f )∗ (g(x)) = ( f )∗ (x2 ) = h ∗2 (x2 ) = x2 and g( f )∗ (x) =
g(1 − f (1 − x)) = g(1 − f (1 − x)) = x2 since f (1 − x) ∈ [0, b]. Thus, ( f )∗ g(x)
= g( f )∗ (x) for x ∈ [1 − b, 1]. For x ∈ [h −1 ∗ −1 −1 ∗
3 (1 − b), (h 2 ) (h 2 (0)))], ( f ) g(x) =
∗
g( f (x)). Now h −1 ∗ ∗ ∗ −1 −1
3 (1 − b) is the fixed point of h 3 . So h 3 [(h 3 ) (b), h 3 (1 − b)] =
−1 ∗
[b, h 3 (1 − b)] and this is the domain of h. Also, h 3 (1 − b) is the fixed point of h 3 .
Appendix A: A Counterexample on Common Fixed Points 291
So h 3 [(h ∗3 )−1 (b), h −1 ∗ −1

3 (1 − b)] = [(h 3 ) b, 1 − b]. This equals the domain of defini-
tion of h . It can be seen that the two piecewise linear functions h ∗ (h|[(h ∗3 )−1 b,h −1
∗
3 (1−b)]
)
∗ ∗ ∗ −1
and h (h|[(h ∗3 )−1 b,h −1
3 (1−b)]
) are the union of three linear functions mapping (h 3 ) (b),
h −1
3 (h ∗ −1
) (1)), h −1
3 ((h ∗ −1
) (0)), h −1
3 (1 − b)to b, 1, 0 and 1 − b, respectively. So
these two functions coincide. So for each x ∈ 21 , h −1 ∗ −1 −1
3 (1 − b) , x ∈ [(h 3 ) b, h 3 (1 −
b)]. Thus, ( f )∗ (g(x)) = ( f )∗ h 3 (x) = h ∗ h(x) = (h ∗ h)∗ (x) = 1 − h ∗ (h(1 − x)) =
h(1 − h(1 − x)) = hh ∗ (x) = g(h ∗3 (x)) = g( f )∗ (x). In other words, ( f )∗ and g
commute on 21 , 1 and are without a common fixed point in 21 , 1 .
Thus, we have
Proposition A.1.5 For any f and g in Cb , f and (g)∗ are commuting functions
without a common fixed point.
Proof For f, g ∈ Cb , by Lemma A.1.4, f and (g)∗ commute without a common

fixed point in 21 , 1 . Also, ( f )∗ and g commute without a common fixed point in
1
2
, 1 . So, (( f )∗ )∗ and (g)∗ commute without a common fixed point in 0, 21 . But
( f )∗∗ = f . Thus, f and g ∗ form a solution to the nonexistence of a common fixed
point for commuting maps on [0, 1].
References
1. Boyce, W.M.: Commuting functions with no common fixed point. Trans. Am. Math. Soc. 137,
77–92 (1969)
2. Huneke, J.P.: On common fixed points of commuting continuous functions on an interval. Trans.
Am. Math. Soc. 139, 371–381 (1969)
Appendix B
A Compact Contractible Space Without Fixed
Point Property
B.1 Introduction
A natural question is whether a compact contractible space has the fixed point prop-
erty for continuous functions. This question was raised in 1932 by Borsuk [1] and
was settled in the negative by Kinoshita [2] in 1953. In the following, Kinoshita’s
counterexample is described.
Recall that a topological space X is called contractible if the identity map on X
is null-homotopic (i.e. homotopic to a constant).
B.2 Kinoshita’s Example
We define the set A = A1 ∪ A2 ∪ A3 in R3 as follows:
A1 = {(r, θ, z) : 0 ≤ r < 1, θ ∈ [0, 2π ], z = 0}

A2 = {(r, θ, z) : r = 1, θ ∈ [0, 2π ], z = [0, 1]}

2
A3 = (r, θ, z) : r = arctan φ, θ = φ mod 2π, φ ≥ 0, z = [0, 1] .
π
A inherits the subspace topology of R3 . Clearly, A is closed and bounded in R3

and so is compact. Define h 1 , h 2 : A → A by h 1 (r, θ, z) = (r, θ, 0) and h 2 : A →
A by h 2 (r, θ, z) = (0, θ, z). From the properties of h 2 ◦ h 1 (r, θ, z) = (0, θ, 0), the
contractibility of A follows.

https://doi.org/10.1007/978-981-13-3158-9
294 Appendix B: A Compact Contractible Space Without Fixed Point Property
Intuitively the continuous map on A constructed by Kinoshita involves twisting

the top of A in counterclockwise direction while rotating the bottom of A in clockwise
direction. More precisely, f : A → A is defined in the following way:
For (r, θ, z) ∈ A1 ,

2
arctan tan πr − π , θ − π, 0 , for r ≥ π2 arctan(π )
f (r, θ, z) = π 2
0, 0, 1 − π1 arctan πr
2
, for 0 ≤ r ≤ π2 arctan(π ).
f maps (0, 0, 0) onto (0, 0, 1), the circle r = π2 arctan(π ) to (0, 0, 0) and the interior
of this circle into the segment r = 0, z ∈ [0, 1]. The annulus of the circle is rotated
π radians while the inner boundary is contracted to (0, 0, 0). f has no fixed points
in A1 .
For (r, θ, z) ∈ A2 , since r = 1

(1, (θ − π + 2π z) mod 2π, z + 2z , z ∈ 0, 21
f (1, θ, z) =
(1, (θ − π + 2π z) mod 2π, 21 + 2z , z ∈ 21 , 1 .
Geometrically, f rotates the top of the cylinder counterclockwise by π radians, the

bottom clockwise by π radians, while the circle r = π2 arctan(π ), z = 1 is pulled
upwards to z = 4z , stretching the bottom half of the cylinder with it and compressing
the top half of the cylinder between 43 ≤ z ≤ 1. No point in A2 off the circle at z = 21
is fixed by f even as no point of the circle is fixed due to the lifting motion of the
map.
On A3 , for 0 ≤ φ ≤ π f is defined by

2
arctan((φ + π )z, (φ + π )z mod 2π, 1 + πφ ( 3z2 − 1) 0 ≤ z ≤ 21
f (r, θ, z) = π2
π
arctan((φ + π )z, (φ + π )z mod 2π, 1 + πφ ( z−12
) 1
2
≤z≤1
On A3 , for π ≤ φ < +∞, f is defined by

2
π arctan((φ − π + 2π z), (φ − π + 2π z)z mod 2π, 3z ) 0 ≤ z ≤ 21
f (r, θ, z) = 2 2
π arctan((φ − π + 2π z), (φ − π + 2π z) mod 2π, 2 ) z ∈ 21 , 1
1+z
Clearly f is well-defined on A1 ∩ A2 , A1 ∩ A3 and continuous not only on each

of A1 , A2 and A3 but also on A1 ∩ A2 , A2 ∩ A3 and A1 ∩ A3 . By construction f has
no fixed point in A (see sketch below for A).
Appendix B: A Compact Contractible Space Without Fixed Point Property 295
References
1. Borsuk, K.: Uber eine Klasse von lokal zusammen hängenden Räumen. Fund. Math. 19, 220–242
(1932)
2. Kinoshita, S.: On some contractible continua without fixed point property. Fund. Math. 40,
96–98 (1953)
Appendix C
Fractals via Fixed Points
C.1 Introduction
Fractals may intuitively be understood as highly irregular non-smooth sets with ‘non-
integral dimension’ arising often from a recursive process of construction displaying
self-similarity. Some of these fractals can be realized as fixed points of set-functions.
(see Hutchinson [2]).
Mandelbrot has pointed out how fractals can be used to model several physical
phenomena. Falconer [1] treats interesting aspects of fractals from a geometric point
of view.
C.2 Hausdorff Measure and Hausdorff Dimension
The following concepts and results are used in the sequel.
Definition C.2.1 Let E be any subset of Rn and s ≥ 0. The Hausdorff s-dimensional

outer measure of E denoted by H s (E) is defined by H s (E) = lim Hδs (E), where
δ→0
∞ ∞

Hδs (E) = inf (diam Ui ) : E ⊆
s
Ui and 0 < diam Ui < δ .
i=1 i=1
Here diam U = sup{ x − y : x, y ∈ U }.
Remark C.2.2 One can show that H s (E) = sup Hδs (E) for any E ⊆ Rv . Further
δ>0
when H s in restricted to the σ -algebra of H s -measurable subsets of Rn , it is a
measure called s-dimensional Hausdorff measure.

https://doi.org/10.1007/978-981-13-3158-9
298 Appendix C: Fractals via Fixed Points
Remark C.2.3 If T : Rn → Rn is a similarity map in the sense that T x − T y =

λ x − y for some λ > 0 for all x, y ∈ Rn , then for a H s -measurable subset E of
Rn . T (E) is H s -measurable and H s (T (E)) = λs H s (E).
Remark C.2.4 For each s ≥ 0 H s is a regular outer measure. Thus all Borel subsets
of Rn are H s -measurable.
Definition C.2.5 Let E ⊆ Rn . The unique real number, dim H E is called the Haus-
dorff dimension of E provided

∞ for 0 ≤ s < dim H (E)
H s (E) =
0 for dim H (E) < s < ∞.
Using the concept of iterated function systems (IFS), a large class of fractal sets
can be constructed with explicit computation of their Hausdorff dimension.
The theorem below stated without proof is used in the existence theorem.
Theorem C.2.6 Let (X, d) be a complete metric space. K (X ) the set of all nonempty
compact subsets of X is a complete metric space under the Hausdorff metric (vide
Definition 5.3.1).
Remark C.2.7 If (An ) ∈ K (X ) is a Cauchy sequence with the Hausdorff metric then
it converges to lim sup(An ) in K (X ).
C.3 Construction of Fractal Sets
Definition C.3.1 Let F be a nonvoid closed subset of Rn . An iterated function

system (IFS for short) is a finite set of contractions {S1 , . . . , Sm }, m ≥ 2 on F. A
m
subset D of F is called an attractor of the IFS {S1 , . . . , Sn } if D = Si (D).
i=1
Theorem C.3.2 Let F ⊆ Rn be a nonempty closed subset of Rn and {S1 , . . . , Sm }

be an IFS on F. Let K (F) be the space of nonvoid compact subsets of F with the

m
Hausdorff metric. For each E ∈ K (F) define S(E) = Si (E). Then there is a
i=1
unique attractor D of S in K (F) and for each E ∈ K (F) with S(E) ⊆ E,
∞

F= S k (E)
k=0
with S 0 (E) = E, S k+1 (E) = S(S k (E)) for k = 1, 2, . . . .

Appendix C: Fractals via Fixed Points 299
Proof Let ci be the contraction constants for Si , for each i = 1, 2, . . . , m. Now,

m

m
H (S A, S B) = H Si (A), Si (B)
i=1 i=1
≤ max H (Si (A), Si (B))
i=1,...,m
≤ max ci H (A, B)
i=1,...,m
for A, B ∈ K (F). Since for each ci , 0 ≤ ci < 1, 0 ≤ c = max{ci : i = 1, 2, . . . , m}

< 1. Thus A → S(A) is a contraction on K (F) with contraction constant c. As
(K (F), H ) is complete, S has a unique fixed point D say. Thus S(D) = D ∈ K (F).
So the IFS {S1 , . . . , Sm } has a unique attractor D. If for E ∈ K (F), S(E) ⊆ E then
S k (E) ∈ K (F) is a sequence of S-iterates which is decreasing and Cauchy in K (F)
and converges to the fixed point D by the contraction principle. But lim S R (E) =
∞
lim sup S k (E). Thus S k (E) = D.
k=0
Remark C.3.3 If the contractions are similarity mappings satisfying an additional
condition called open set condition, then the Hausdorff dimension of the attractor
of the IFS, can be computed. This in turn leads to the construction of a number of
self-similar fractals.
Definition C.3.4 An IFS of similarity contractions {S1 , . . . , Sm } is said to satisfy
the open set condition (or Moran’s condition) if for some nonvoid bounded open set

m
V, V ⊇ Si (V ) with Si (V ) ∩ S j (V ) = φ for i = j.
i=1
The following proposition is left as an exercise.
Proposition C.3.5 Let {Vi } be a family of disjoint nonvoid open sets in Rn , such
that each Vi contains a ball of radius αr and is contained in a ball if radius βr . Then
any ball of radius r intersects at most (1 + 2β)n α −n of the closures of Vi .
Proposition C.3.6 Suppose for some finite positive measurable subset E of Rn there
exist s, k, δ > 0 such that μ(E) ≤ k(diam U )s for all sets U with diamU ≤ δ. Then
H s (E) ≥ μ(E)k
and s ≤ dim H E.
Proof For a cover {Ui } of
E with diamUi ≤ δ. Then from the properties of mea-
sure μ(E) ⊆ μ(UUi ) ≤ μ(Ui ) ≤ k (diam(Ui ))s . So k H s (E) ≥ μ(E) > 0.
i i
As μ(E) > 0. dim H (E) ≥ s.
Theorem C.3.7 Let the IFS {S1 , . . . , Sm } of contracting similarities with con-
traction constant ci satisfy the open set condition and F be the attractor of the
m
IFS guaranteed by Theorem C.3.2, then dim H F = s where cis = 1. Further
i=1
0 < H s (F) < ∞.

m
m
Proof Clearly for the function t → cit , there is a unique s with cis = 1. For
i=1 i=1
Ik = {(i 1 , . . . , i k ) : i k ∈ {1, . . . , m}} and any set B and a given element (i
1 , . . . , ik )
of Ik , define Bi1 ...ik = Si1 ◦ · · · ◦ Sik (B). As F is fixed under the IFS, F = Fi1 ...ik .
Ik
So {Fi1 ...ik } cover F. Since Si ◦ · · · ◦ Sk is a contraction similarity with constant
ci1 . . . cik ,

(diam Fi1 ...ik )s = (ci1 . . . cik )s (diam F)s
Ik

= cis1 . . . cisk (diam F)s

m
= (diam F)s since cis = 1.
i=1
For δ > 0, we can choose k such that (max ci )k diam F ≤ δ. Since diam Fi1 ...ik ≤
(maxi ci )k diam F the sets {F i1 ...ik } cover F with diameter ≤ δ, for sequences
(i 1 , . . . , i k ) in Ik . So H s (F) ≤ Ik (diam(Fi1 . . . Fik ))s .
Consider I the set of all sequences (i n ) such that i k ∈ {1, . . . , m} for all k and let
Ii1 ...ik be the subset of I consisting of only those sequences starting with i 1 , . . . , i k .
∞

Define μ(Ii1 ,...,ik ) = (ci1 . . . cik )s . Define xi1 ,i2 ,... = Fi1 ...ik . For a subset A of F
k=1
define μ(A) = μ{Ii1 ...ik : xi1 ...,ik , · · · ∈ A}. Now μ(F) = μ(I ) = 1 as F contains
every point xi1 ,i2 ,... for each (i 1 , i 2 , . . . ) ∈ I .
Since the IFS satisfies an open set condition, there is an open set V with V ⊃
m m
Si (V ), where Si (V ) ∩ S j (V ) = φ for i = j. So S(V ) = Si (V ) and {S k (V )}
i=1 i=1
converges to F. So F ⊆ V i1 ,...,ik for any sequence {i 1 , . . . , i k }. For a ball B of radius
0 < r < 1, each sequence in I is truncated at the first i k for which

min ci r ≤ ci1 . . . cik ≤ r
1≤i≤m
where ci are the contraction constants of the corresponding similarities. As

V p ∩ Vq = φ for p = q, Vi1 ,...,ik , p and Vi1 ,...,ik ,q are disjoint. Let J be the finite set of
all such truncated sequences in I . Clearly,

F= Fi1 ,i2 ,... ⊆ Fi1 ,...,ik ⊆ V i1 ,...,ik
I J J
Choose α, β such that V contains a ball of radius αr and is contained in a ball of

radius βr . So for all (i 1 , . . . , i k ) ∈ J the set Vi1 ...ik contains a ball of radius ci1 . . . cik αr
and is contained in a ball of radius ci1 . . . cik βr . Now every Vi1 ...ik contains a ball of
Appendix C: Fractals via Fixed Points 301
radius (mini ci r ) α and is contained in a ball of radius rβ. Let J be be the subset
of I consisting of only those sequences (i 1 , . . . , i k ) such that V i1 ...ik intersects B.
By Proposition C.3.5 there exist at most γ = (1 + 2β)n (mini ci α)−n elements in J .
Now
μ(B) = μ(F ∩ B)
= μ({Ii1 ,...,ik : xi1 ,...,ik ... ∈ F ∩ B})

= μ({Ii1 ,...,ik : xi1 ,...,ik ... ∈ ( V i1 ...ik ) ∩ B}).
J
So

μ(B) ≤ μ({Ii1 ,...,ik : xi1 ,...,ik ... ∈ V i1 ...ik })
J

= μ( Ii1 ,...ik )
J

< μ(Ii1 . . . Iik )
J

≤ (ci1 . . . cik )s ≤ r s = qr s .
J J
So for any set U contained in a ball BU of radius diam U , μ(U ) ≤ μ(BU ) ≤

q(diam U )s . So by Proposition C.3.6, H s (F) ≥ μ(F) q
= q1 > 0. Thus dim H F = s.

If the dimension of the set F is non-integral, then F is called a fractal. For instance,
the Cantor set is the unique attractor of the IFS {S1 , S2 } on [0, 1] where S1 (x) = x3 and
S2 (x) = 1 − x3 . It satisfies the open set condition with V = (0, 1). So by the above
theorem the Hausdorff-dimension of the Cantor set is s with 2( 31 )s = 1 and s = log 2
log 3
.
The Cantor set results from throwing off the middle third intervals recursively from
[0, 1] and a small section of the Cantor set is a scaled version of the entire set.
This points to the self-similar nature of the set. The von Koch curve constructed by
throwing off the middle third of a line segment and constructing an equilateral triangle
over the removed middle third segment and continuing recursively this process results
in a fractal set and its dimension is log 4
log 3
.
References
1. Falconer, K.J.: The geometry of fractal sets, Cambridge University Press (1962)
2. Hutchinson, J.E.: Fractals and self-similarity, Indiana University of Mathematics Jl., 30,
713–747 (1981)
Postscript
Apart from the references cited in each chapter, the following books may be referred for further
study.
1. Aksoy, A., Khamsi, M.A.: Nonstandard Methods in Fixed Point Theory. Springer, Berlin (1990).
2. Agarwal, R.P., Meehan, M., Oregan, D.: Fixed Point Theory and Applications. Cambridge
University Press, Cambridge (2001).
3. Bonsall, F.F.: Lectures on some fixed point theorems of functional analysis. Tata Institute,
Bombay (1962).
4. Brown, R.F.: The Letschetz Fixed Point Theorem. Scott Foreman & Co, London (1971).
5. Cronin, J.: Fixed Points and Topological Degree in Nonlinear Analysis. Survey, vol. 11. Amer-
ican Mathematical Society, New York (1964).
6. Deimling, K.: Nichtlineare Gleichungen und Abbildungsgrade. Hochschult Text. Springer,
Berlin (1974).
7. Deimling, K.: Nonlinear Functional Analysis. Springer, Berlin (1985).
8. Dugundji, J., Granas, A.: Fixed Point Theory. PWN-Polish Scientific Publishers, Warsaw
(1982).
9. Goebel, K., Kirk, W.A.: Topics in Metric Fixed Point Theory. Cambridge University Press,
Cambridge (1990).
10. Hadzic, O.: Fixed point theory in topological vector spaces. University of Novisad (1964).
11. Istratescu, V.I.: Fixed point theory. Reidel of Dordrecht (1981).
12. Khamsi, M.A., Kirk, W.A.: An Introduction to Metric Spaces and Fixed Point Theory. Wiley,
New York (2001).
13. Kirk, W., Shahzad, N.: Fixed Point Theory in Distance Spaces. Springer, Berlin (2014).
14. Kirk, W., Sims, B. (eds.) Handbook of Metric Fixed Point Theory. Springer, Netherlands (2001).
15. Rothe, E.H.: Introduction to Various Aspects of Degree Theory in Banach Spaces. Mathematical
Surveys and Monographs, vol. 23. American Mathematical Society, Providence (1986).
16. Smart, D.R.: Fixed Point Theorems. Cambridge University Press, Cambridge (1994).
17. Van der Walt, T.: Fixed and Almost Fixed Points. Mathematical Centrum, Amsterdam (1963).
18. Zeidler, E.: Nonlinear Functional Analysis and Its Applications I: Fixed Point Theorems.
Springer, Berlin (1986).

Elementary Fixed Point Theorems. (PDFDrive)

Uploaded by

Copyright:

Available Formats

Elementary Fixed Point Theorems. (PDFDrive)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Elementary Fixed Point Theorems. (PDFDrive)

Uploaded by

Copyright:

Available Formats

Forum for Interdisciplinary Mathematics

More information about this series at http://www.springer.com/series/13386

Elementary Fixed Point

ISSN 2364-6748 ISSN 2364-6756 (electronic)

Library of Congress Control Number: 2018960278

© Springer Nature Singapore Pte Ltd. 2018

Fixed point theorems constitute an important and interesting aspect of applicable

point property of continuous functions on dendroids. Also highlighted is Klee’s

Chennai, India P. V. Subrahmanyam

4.5 Some Spaces with Fixed Point Property . . . . . . . . . . . . . . . . . . 92

8.7 Iterations of a Non-expansive Mapping . . . . . . . . . . . . . . . . . . 186

Appendix A: A Counterexample on Common Fixed Points . . . . . . . . . . . 289

P. V. Subrahmanyam is Professor Emeritus at the Indian Institute of Technology

1.1 Topological Spaces

Any subset of X belonging to T is called an open set or more precisely T -open

Definition 1.1.10 A binary relation ≤ on a non-empty subset X is called a quasi-

Definition 1.1.13 A net in a topological space X is a pair (S, ≥) where S is a

Proposition 1.1.14 A subset S of a topological space (X, T ) is closed if and only

Definition 1.1.15 Let (X i , Ti ), i = 1, 2 be topological spaces. A map f : X 1 → X 2

The following theorem is well-known.

(v) for each net (S, ≥) converging to x in X 1 , ( f (S), ≥) converges to f (x) in X 2 ;

Theorem 1.1.25 Let {(X λ , Tλ ) : λ ∈ , = φ} be a family of topological spaces

Definition 1.1.26 A topological space X is said to be

Theorem 1.1.27 (Urysohn’s Lemma) If A and B are disjoint closed subsets of a

Theorem 1.1.28 Let X and Y be topological spaces and f : X → Y be a continuous

Corollary 1.1.29 If X is a compact topological space and f : X → R is a continu-

1.2 Metric Spaces

(i) d(x, x) = 0 for all x ∈ X ;

d( f, g) = sup{| f (x) − g(x)| : x ∈ [a, b]}

where f, g ∈ C[a, b]. This metric is called Tschebyshev or uniform metric.

Theorem 1.2.8 (Baire) No complete metric space can be written as a countable

Definition 1.2.9 Let (X i , di ), i = 1, 2 be metric spaces. A map T : X 1 → X 2 is

Theorem 1.2.10 Each metric space (X, d) can be isometrically embedded in a

where x = (xn ) and y = (yn ) are in X . Further, if each (X n , dn ) is complete, then

The following metrization theorem is classical.

Theorem 1.2.16 (Metrization theorems) A topological space is metrizable if and

Another important notion is that of paracompactness formulated below.

Theorem 1.2.18 Every pseudometric space is paracompact and a paracompact T2

Definition 1.2.19 Let X be a topological space. A family { f λ : λ ∈ = φ} of con-

Definition 1.2.21 A subset S of a metric space (X, d) is said to be totally bounded, if

Theorem 1.2.23 Let (X i , di ), i = 1, 2 be metric spaces and f : X 1 → X 2 be a

1.3 Normed Linear Spaces

1. x = 0 if and only if x = 0;

Remark 1.3.6 If (X,  · ) is a normed linear space, then d(x, y) = x − y, x, y ∈

Definition 1.3.7 A normed linear space (X,  · ) is called a Banach space if it is

Definition 1.3.9 Let (X i ,  · i ), i = 1, 2 be normed linear spaces over F = R or C.

Theorem 1.3.10 Let (X i ,  · i ), i = 1, 2 be normed linear spaces over F = R or

inf{K ≥ 0 : x ∈ X 1 and T x2 ≤ K x1 } = sup{T x2 : x1 = 1, x ∈ X 1 }

Theorem 1.3.12 For i = 1, 2, let (X i ,  · i ) be normed linear spaces. B(X 1 , X 2 )

Theorem 1.3.13 (Hahn–Banach) If f is a bounded linear functional from a linear

Theorem 1.3.24 If  · i , i = 1, 2 are two norms on a finite-dimensional normed

K 1 x1 ≤ x2  ≤ K 2 x1 for all x ∈ X.

Consequently a finite-dimensional normed linear space over R or C is equivalent to

for x, y ∈ X, x + y2 + x − y2 = 2(x2 + y2 )

Definition 1.3.31 A subset C of a linear space V over F = R or C is called convex

Theorem 1.3.37(Parseval identity) If {eλ : λ ∈ = φ} is an orthonormal basis of

Definition 1.3.39 A normed linear space (N ,  · ) over F = R or C is said to be

Definition 1.3.40 A normed linear space (N ,  · ) is called uniformly convex if

Hilbert spaces are isometric to their duals, in view of the following.

1. x = 0 if and only if x = 0;

Remark 1.3.6 If (X, · ) is a normed linear space, then d(x, y) = x − y, x, y ∈

Definition 1.3.7 A normed linear space (X, · ) is called a Banach space if it is

Definition 1.3.9 Let (X i , · i ), i = 1, 2 be normed linear spaces over F = R or C.

Theorem 1.3.10 Let (X i , · i ), i = 1, 2 be normed linear spaces over F = R or

inf{K ≥ 0 : x ∈ X 1 and T x2 ≤ K x1 } = sup{T x2 : x1 = 1, x ∈ X 1 }

Theorem 1.3.12 For i = 1, 2, let (X i , · i ) be normed linear spaces. B(X 1 , X 2 )

Theorem 1.3.24 If · i , i = 1, 2 are two norms on a finite-dimensional normed

K 1 x1 ≤ x2 ≤ K 2 x1 for all x ∈ X.

for x, y ∈ X, x + y2 + x − y2 = 2(x2 + y2 )

Theorem 1.3.37(Parseval identity) If {eλ : λ ∈ = φ} is an orthonormal basis of

Definition 1.3.39 A normed linear space (N , · ) over F = R or C is said to be

Definition 1.3.40 A normed linear space (N , · ) is called uniformly convex if

Example 1.4.16 Let be the union of a sequence of compact sets K n ⊆ R m

Example 1.4.17 Let be a non-void open set in Rn . A multi-index α is an

D α f = f . A complex-valued function defined on is said to belong to C ∞ () if